diff --git a/CMakeLists.txt b/CMakeLists.txt index c8677ae2ae9..cb3825ecd15 100644 --- a/CMakeLists.txt +++ b/CMakeLists.txt @@ -439,7 +439,6 @@ option (ENABLE_UNWIND "Enable libunwind (better stacktraces)" ON) include (libs/libdaemon/cmake/find_unwind.cmake) # Need to process before "contrib" dir: -include (libs/libcommon/cmake/find_gperftools.cmake) include (libs/libcommon/cmake/find_jemalloc.cmake) include (libs/libcommon/cmake/find_mimalloc.cmake) include (libs/libcommon/cmake/find_cctz.cmake) diff --git a/NOTICE b/NOTICE index 3fbb523f72b..34da926905b 100644 --- a/NOTICE +++ b/NOTICE @@ -123,11 +123,6 @@ Revision: 21d37dbaa45742c0bd23cc1e5a70b52cbc27f809 Address: https://github.com/ClickHouse/ClickHouse/tree/b13313eecca9455f4fdc923f597fe863df409742/contrib/libsparsehash License: https://github.com/ClickHouse/ClickHouse/blob/b13313eecca9455f4fdc923f597fe863df409742/contrib/libsparsehash/COPYING -------------------------------------- -tcmalloc -Revision: dde32f8bbc95312379f9f5a651799815bb6327c5 -Address: https://github.com/ClickHouse/ClickHouse/tree/b13313eecca9455f4fdc923f597fe863df409742/contrib/libtcmalloc -License: https://github.com/ClickHouse/ClickHouse/blob/b13313eecca9455f4fdc923f597fe863df409742/contrib/libtcmalloc/COPYING --------------------------------------- libunwind Revision: 19a2c01b1e8ac20871ea09d20f596d425ba53aed Address: https://github.com/libunwind/libunwind diff --git a/cmake/Modules/FindGperftools.cmake b/cmake/Modules/FindGperftools.cmake deleted file mode 100644 index dabc13d1611..00000000000 --- a/cmake/Modules/FindGperftools.cmake +++ /dev/null @@ -1,75 +0,0 @@ -# Copyright 2022 PingCAP, Ltd. -# -# Licensed under the Apache License, Version 2.0 (the "License"); -# you may not use this file except in compliance with the License. -# You may obtain a copy of the License at -# -# http://www.apache.org/licenses/LICENSE-2.0 -# -# Unless required by applicable law or agreed to in writing, software -# distributed under the License is distributed on an "AS IS" BASIS, -# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -# See the License for the specific language governing permissions and -# limitations under the License. - -# https://github.com/vast-io/vast/blob/master/cmake/FindGperftools.cmake - -# Tries to find Gperftools. -# -# Usage of this module as follows: -# -# find_package(Gperftools) -# -# Variables used by this module, they can change the default behaviour and need -# to be set before calling find_package: -# -# Gperftools_ROOT_DIR Set this variable to the root installation of -# Gperftools if the module has problems finding -# the proper installation path. -# -# Variables defined by this module: -# -# GPERFTOOLS_FOUND System has Gperftools libs/headers -# GPERFTOOLS_LIBRARIES The Gperftools libraries (tcmalloc & profiler) -# GPERFTOOLS_INCLUDE_DIR The location of Gperftools headers - -find_library(GPERFTOOLS_TCMALLOC - NAMES tcmalloc - HINTS ${Gperftools_ROOT_DIR}/lib) - -find_library(GPERFTOOLS_TCMALLOC_MINIMAL - NAMES tcmalloc_minimal - HINTS ${Gperftools_ROOT_DIR}/lib) - -find_library(GPERFTOOLS_TCMALLOC_MINIMAL_DEBUG - NAMES tcmalloc_minimal_debug - HINTS ${Gperftools_ROOT_DIR}/lib) - -find_library(GPERFTOOLS_PROFILER - NAMES profiler - HINTS ${Gperftools_ROOT_DIR}/lib) - -find_library(GPERFTOOLS_TCMALLOC_AND_PROFILER - NAMES tcmalloc_and_profiler - HINTS ${Gperftools_ROOT_DIR}/lib) - -find_path(GPERFTOOLS_INCLUDE_DIR - NAMES gperftools/heap-profiler.h - HINTS ${Gperftools_ROOT_DIR}/include) - -set(GPERFTOOLS_LIBRARIES ${GPERFTOOLS_TCMALLOC_AND_PROFILER}) - -include(FindPackageHandleStandardArgs) -find_package_handle_standard_args( - Gperftools - DEFAULT_MSG - GPERFTOOLS_LIBRARIES - GPERFTOOLS_INCLUDE_DIR) - -mark_as_advanced( - Gperftools_ROOT_DIR - GPERFTOOLS_TCMALLOC - GPERFTOOLS_PROFILER - GPERFTOOLS_TCMALLOC_AND_PROFILER - GPERFTOOLS_LIBRARIES - GPERFTOOLS_INCLUDE_DIR) diff --git a/contrib/CMakeLists.txt b/contrib/CMakeLists.txt index 4520d1cb176..eca6338104f 100644 --- a/contrib/CMakeLists.txt +++ b/contrib/CMakeLists.txt @@ -123,11 +123,6 @@ if (USE_INTERNAL_TIFLASH_PROXY) add_subdirectory (tiflash-proxy-cmake) endif () -# TODO: remove tcmalloc -if (ENABLE_TCMALLOC AND USE_INTERNAL_GPERFTOOLS_LIBRARY) - add_subdirectory (libtcmalloc) -endif () - if (ENABLE_JEMALLOC AND USE_INTERNAL_JEMALLOC_LIBRARY) add_subdirectory (jemalloc-cmake) endif () @@ -164,7 +159,9 @@ SET (BENCHMARK_ENABLE_GTEST_TESTS OFF CACHE BOOL "Disable google-benchmark testi add_subdirectory(benchmark) set (BUILD_TESTING OFF CACHE BOOL "Disable cpu-features testing" FORCE) -add_subdirectory(cpu_features) +if ((NOT APPLE) AND (NOT ARCH_AARCH64)) + add_subdirectory(cpu_features) +endif() if (ARCH_AARCH64 AND ARCH_LINUX) add_subdirectory(arm-optimized-routines-cmake) diff --git a/contrib/libtcmalloc/AUTHORS b/contrib/libtcmalloc/AUTHORS deleted file mode 100644 index 3995ed4cf57..00000000000 --- a/contrib/libtcmalloc/AUTHORS +++ /dev/null @@ -1,2 +0,0 @@ -google-perftools@googlegroups.com - diff --git a/contrib/libtcmalloc/CMakeLists.txt b/contrib/libtcmalloc/CMakeLists.txt deleted file mode 100644 index 5bf61f98bea..00000000000 --- a/contrib/libtcmalloc/CMakeLists.txt +++ /dev/null @@ -1,81 +0,0 @@ - -message (STATUS "Building: tcmalloc_minimal_internal") - -add_library (tcmalloc_minimal_internal -./src/malloc_hook.cc -./src/base/spinlock_internal.cc -./src/base/spinlock.cc -./src/base/dynamic_annotations.c -./src/base/linuxthreads.cc -./src/base/elf_mem_image.cc -./src/base/vdso_support.cc -./src/base/sysinfo.cc -./src/base/low_level_alloc.cc -./src/base/thread_lister.c -./src/base/logging.cc -./src/base/atomicops-internals-x86.cc -./src/memfs_malloc.cc -./src/tcmalloc.cc -./src/malloc_extension.cc -./src/thread_cache.cc -./src/symbolize.cc -./src/page_heap.cc -./src/maybe_threads.cc -./src/central_freelist.cc -./src/static_vars.cc -./src/sampler.cc -./src/internal_logging.cc -./src/system-alloc.cc -./src/span.cc -./src/common.cc -./src/stacktrace.cc -./src/stack_trace_table.cc -./src/heap-checker.cc -./src/heap-checker-bcad.cc -./src/heap-profile-table.cc -./src/raw_printer.cc -./src/memory_region_map.cc -) - - -target_compile_options (tcmalloc_minimal_internal - PRIVATE - -DNO_TCMALLOC_SAMPLES - -DNDEBUG - -DNO_FRAME_POINTER - -Wwrite-strings - -Wno-sign-compare - -Wno-unused-result - -Wno-deprecated-declarations - -Wno-unused-function - -Wno-unused-private-field - - PUBLIC - -fno-builtin-malloc - -fno-builtin-free - -fno-builtin-realloc - -fno-builtin-calloc - -fno-builtin-cfree - -fno-builtin-memalign - -fno-builtin-posix_memalign - -fno-builtin-valloc - -fno-builtin-pvalloc -) - -if (CMAKE_CXX_COMPILER_ID STREQUAL "Clang" AND CMAKE_CXX_COMPILER_VERSION VERSION_GREATER 3.9.1) - target_compile_options(tcmalloc_minimal_internal PUBLIC -Wno-dynamic-exception-spec ) -endif () - -if (CMAKE_SYSTEM MATCHES "FreeBSD" AND CMAKE_CXX_COMPILER_ID STREQUAL "GNU") - target_compile_options(tcmalloc_minimal_internal PUBLIC -Wno-unused-but-set-variable) -endif () - -if (CMAKE_SYSTEM MATCHES "FreeBSD") - target_compile_definitions(tcmalloc_minimal_internal PUBLIC _GNU_SOURCE) -endif () - -target_include_directories (tcmalloc_minimal_internal PUBLIC include) -target_include_directories (tcmalloc_minimal_internal PRIVATE src) - -find_package (Threads) -target_link_libraries (tcmalloc_minimal_internal ${CMAKE_THREAD_LIBS_INIT}) diff --git a/contrib/libtcmalloc/COPYING b/contrib/libtcmalloc/COPYING deleted file mode 100644 index e4956cfd9fd..00000000000 --- a/contrib/libtcmalloc/COPYING +++ /dev/null @@ -1,28 +0,0 @@ -Copyright (c) 2005, Google Inc. -All rights reserved. - -Redistribution and use in source and binary forms, with or without -modification, are permitted provided that the following conditions are -met: - - * Redistributions of source code must retain the above copyright -notice, this list of conditions and the following disclaimer. - * Redistributions in binary form must reproduce the above -copyright notice, this list of conditions and the following disclaimer -in the documentation and/or other materials provided with the -distribution. - * Neither the name of Google Inc. nor the names of its -contributors may be used to endorse or promote products derived from -this software without specific prior written permission. - -THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS -"AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT -LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR -A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT -OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, -SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT -LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, -DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY -THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT -(INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE -OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. diff --git a/contrib/libtcmalloc/README b/contrib/libtcmalloc/README deleted file mode 100644 index 8f7377c0b3b..00000000000 --- a/contrib/libtcmalloc/README +++ /dev/null @@ -1,8 +0,0 @@ -https://github.com/gperftools/gperftools/commit/dde32f8bbc95312379f9f5a651799815bb6327c5 - -Several modifications: -1. Disabled TCMALLOC_AGGRESSIVE_DECOMMIT by default. It is important. -2. Using only files for tcmalloc_minimal build (./configure --enable-minimal). -3. Using some compiler flags from project. -4. Removed warning about unused variable when build with NDEBUG (by default). -5. Including config.h with relative path. diff --git a/contrib/libtcmalloc/include/gperftools/heap-checker.h b/contrib/libtcmalloc/include/gperftools/heap-checker.h deleted file mode 100644 index 5a87d8da7f7..00000000000 --- a/contrib/libtcmalloc/include/gperftools/heap-checker.h +++ /dev/null @@ -1,422 +0,0 @@ -// -*- Mode: C++; c-basic-offset: 2; indent-tabs-mode: nil -*- -// Copyright (c) 2005, Google Inc. -// All rights reserved. -// -// Redistribution and use in source and binary forms, with or without -// modification, are permitted provided that the following conditions are -// met: -// -// * Redistributions of source code must retain the above copyright -// notice, this list of conditions and the following disclaimer. -// * Redistributions in binary form must reproduce the above -// copyright notice, this list of conditions and the following disclaimer -// in the documentation and/or other materials provided with the -// distribution. -// * Neither the name of Google Inc. nor the names of its -// contributors may be used to endorse or promote products derived from -// this software without specific prior written permission. -// -// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS -// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT -// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR -// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT -// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, -// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT -// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, -// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY -// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT -// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE -// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. - -// --- -// Author: Maxim Lifantsev (with design ideas by Sanjay Ghemawat) -// -// -// Module for detecing heap (memory) leaks. -// -// For full(er) information, see doc/heap_checker.html -// -// This module can be linked into programs with -// no slowdown caused by this unless you activate the leak-checker: -// -// 1. Set the environment variable HEAPCHEK to _type_ before -// running the program. -// -// _type_ is usually "normal" but can also be "minimal", "strict", or -// "draconian". (See the html file for other options, like 'local'.) -// -// After that, just run your binary. If the heap-checker detects -// a memory leak at program-exit, it will print instructions on how -// to track down the leak. - -#ifndef BASE_HEAP_CHECKER_H_ -#define BASE_HEAP_CHECKER_H_ - -#include // for size_t -// I can't #include config.h in this public API file, but I should -// really use configure (and make malloc_extension.h a .in file) to -// figure out if the system has stdint.h or not. But I'm lazy, so -// for now I'm assuming it's a problem only with MSVC. -#ifndef _MSC_VER -#include // for uintptr_t -#endif -#include // for va_list -#include - -// Annoying stuff for windows -- makes sure clients can import these functions -#ifndef PERFTOOLS_DLL_DECL -# ifdef _WIN32 -# define PERFTOOLS_DLL_DECL __declspec(dllimport) -# else -# define PERFTOOLS_DLL_DECL -# endif -#endif - - -// The class is thread-safe with respect to all the provided static methods, -// as well as HeapLeakChecker objects: they can be accessed by multiple threads. -class PERFTOOLS_DLL_DECL HeapLeakChecker { - public: - - // ----------------------------------------------------------------------- // - // Static functions for working with (whole-program) leak checking. - - // If heap leak checking is currently active in some mode - // e.g. if leak checking was started (and is still active now) - // due to HEAPCHECK=... defined in the environment. - // The return value reflects iff HeapLeakChecker objects manually - // constructed right now will be doing leak checking or nothing. - // Note that we can go from active to inactive state during InitGoogle() - // if FLAGS_heap_check gets set to "" by some code before/during InitGoogle(). - static bool IsActive(); - - // Return pointer to the whole-program checker if it has been created - // and NULL otherwise. - // Once GlobalChecker() returns non-NULL that object will not disappear and - // will be returned by all later GlobalChecker calls. - // This is mainly to access BytesLeaked() and ObjectsLeaked() (see below) - // for the whole-program checker after one calls NoGlobalLeaks() - // or similar and gets false. - static HeapLeakChecker* GlobalChecker(); - - // Do whole-program leak check now (if it was activated for this binary); - // return false only if it was activated and has failed. - // The mode of the check is controlled by the command-line flags. - // This method can be called repeatedly. - // Things like GlobalChecker()->SameHeap() can also be called explicitly - // to do the desired flavor of the check. - static bool NoGlobalLeaks(); - - // If whole-program checker if active, - // cancel its automatic execution after main() exits. - // This requires that some leak check (e.g. NoGlobalLeaks()) - // has been called at least once on the whole-program checker. - static void CancelGlobalCheck(); - - // ----------------------------------------------------------------------- // - // Non-static functions for starting and doing leak checking. - - // Start checking and name the leak check performed. - // The name is used in naming dumped profiles - // and needs to be unique only within your binary. - // It must also be a string that can be a part of a file name, - // in particular not contain path expressions. - explicit HeapLeakChecker(const char *name); - - // Destructor (verifies that some *NoLeaks or *SameHeap method - // has been called at least once). - ~HeapLeakChecker(); - - // These used to be different but are all the same now: they return - // true iff all memory allocated since this HeapLeakChecker object - // was constructor is still reachable from global state. - // - // Because we fork to convert addresses to symbol-names, and forking - // is not thread-safe, and we may be called in a threaded context, - // we do not try to symbolize addresses when called manually. - bool NoLeaks() { return DoNoLeaks(DO_NOT_SYMBOLIZE); } - - // These forms are obsolete; use NoLeaks() instead. - // TODO(csilvers): mark as DEPRECATED. - bool QuickNoLeaks() { return NoLeaks(); } - bool BriefNoLeaks() { return NoLeaks(); } - bool SameHeap() { return NoLeaks(); } - bool QuickSameHeap() { return NoLeaks(); } - bool BriefSameHeap() { return NoLeaks(); } - - // Detailed information about the number of leaked bytes and objects - // (both of these can be negative as well). - // These are available only after a *SameHeap or *NoLeaks - // method has been called. - // Note that it's possible for both of these to be zero - // while SameHeap() or NoLeaks() returned false in case - // of a heap state change that is significant - // but preserves the byte and object counts. - ssize_t BytesLeaked() const; - ssize_t ObjectsLeaked() const; - - // ----------------------------------------------------------------------- // - // Static helpers to make us ignore certain leaks. - - // Scoped helper class. Should be allocated on the stack inside a - // block of code. Any heap allocations done in the code block - // covered by the scoped object (including in nested function calls - // done by the code block) will not be reported as leaks. This is - // the recommended replacement for the GetDisableChecksStart() and - // DisableChecksToHereFrom() routines below. - // - // Example: - // void Foo() { - // HeapLeakChecker::Disabler disabler; - // ... code that allocates objects whose leaks should be ignored ... - // } - // - // REQUIRES: Destructor runs in same thread as constructor - class Disabler { - public: - Disabler(); - ~Disabler(); - private: - Disabler(const Disabler&); // disallow copy - void operator=(const Disabler&); // and assign - }; - - // Ignore an object located at 'ptr' (can go at the start or into the object) - // as well as all heap objects (transitively) referenced from it for the - // purposes of heap leak checking. Returns 'ptr' so that one can write - // static T* obj = IgnoreObject(new T(...)); - // - // If 'ptr' does not point to an active allocated object at the time of this - // call, it is ignored; but if it does, the object must not get deleted from - // the heap later on. - // - // See also HiddenPointer, below, if you need to prevent a pointer from - // being traversed by the heap checker but do not wish to transitively - // whitelist objects referenced through it. - template - static T* IgnoreObject(T* ptr) { - DoIgnoreObject(static_cast(const_cast(ptr))); - return ptr; - } - - // Undo what an earlier IgnoreObject() call promised and asked to do. - // At the time of this call 'ptr' must point at or inside of an active - // allocated object which was previously registered with IgnoreObject(). - static void UnIgnoreObject(const void* ptr); - - // ----------------------------------------------------------------------- // - // Internal types defined in .cc - - class Allocator; - struct RangeValue; - - private: - - // ----------------------------------------------------------------------- // - // Various helpers - - // Create the name of the heap profile file. - // Should be deleted via Allocator::Free(). - char* MakeProfileNameLocked(); - - // Helper for constructors - void Create(const char *name, bool make_start_snapshot); - - enum ShouldSymbolize { SYMBOLIZE, DO_NOT_SYMBOLIZE }; - - // Helper for *NoLeaks and *SameHeap - bool DoNoLeaks(ShouldSymbolize should_symbolize); - - // Helper for NoGlobalLeaks, also called by the global destructor. - static bool NoGlobalLeaksMaybeSymbolize(ShouldSymbolize should_symbolize); - - // These used to be public, but they are now deprecated. - // Will remove entirely when all internal uses are fixed. - // In the meantime, use friendship so the unittest can still test them. - static void* GetDisableChecksStart(); - static void DisableChecksToHereFrom(const void* start_address); - static void DisableChecksIn(const char* pattern); - friend void RangeDisabledLeaks(); - friend void NamedTwoDisabledLeaks(); - friend void* RunNamedDisabledLeaks(void*); - friend void TestHeapLeakCheckerNamedDisabling(); - - // Actually implements IgnoreObject(). - static void DoIgnoreObject(const void* ptr); - - // Disable checks based on stack trace entry at a depth <= - // max_depth. Used to hide allocations done inside some special - // libraries. - static void DisableChecksFromToLocked(const void* start_address, - const void* end_address, - int max_depth); - - // Helper for DoNoLeaks to ignore all objects reachable from all live data - static void IgnoreAllLiveObjectsLocked(const void* self_stack_top); - - // Callback we pass to TCMalloc_ListAllProcessThreads (see thread_lister.h) - // that is invoked when all threads of our process are found and stopped. - // The call back does the things needed to ignore live data reachable from - // thread stacks and registers for all our threads - // as well as do other global-live-data ignoring - // (via IgnoreNonThreadLiveObjectsLocked) - // during the quiet state of all threads being stopped. - // For the argument meaning see the comment by TCMalloc_ListAllProcessThreads. - // Here we only use num_threads and thread_pids, that TCMalloc_ListAllProcessThreads - // fills for us with the number and pids of all the threads of our process - // it found and attached to. - static int IgnoreLiveThreadsLocked(void* parameter, - int num_threads, - pid_t* thread_pids, - va_list ap); - - // Helper for IgnoreAllLiveObjectsLocked and IgnoreLiveThreadsLocked - // that we prefer to execute from IgnoreLiveThreadsLocked - // while all threads are stopped. - // This helper does live object discovery and ignoring - // for all objects that are reachable from everything - // not related to thread stacks and registers. - static void IgnoreNonThreadLiveObjectsLocked(); - - // Helper for IgnoreNonThreadLiveObjectsLocked and IgnoreLiveThreadsLocked - // to discover and ignore all heap objects - // reachable from currently considered live objects - // (live_objects static global variable in out .cc file). - // "name", "name2" are two strings that we print one after another - // in a debug message to describe what kind of live object sources - // are being used. - static void IgnoreLiveObjectsLocked(const char* name, const char* name2); - - // Do the overall whole-program heap leak check if needed; - // returns true when did the leak check. - static bool DoMainHeapCheck(); - - // Type of task for UseProcMapsLocked - enum ProcMapsTask { - RECORD_GLOBAL_DATA, - DISABLE_LIBRARY_ALLOCS - }; - - // Success/Error Return codes for UseProcMapsLocked. - enum ProcMapsResult { - PROC_MAPS_USED, - CANT_OPEN_PROC_MAPS, - NO_SHARED_LIBS_IN_PROC_MAPS - }; - - // Read /proc/self/maps, parse it, and do the 'proc_maps_task' for each line. - static ProcMapsResult UseProcMapsLocked(ProcMapsTask proc_maps_task); - - // A ProcMapsTask to disable allocations from 'library' - // that is mapped to [start_address..end_address) - // (only if library is a certain system library). - static void DisableLibraryAllocsLocked(const char* library, - uintptr_t start_address, - uintptr_t end_address); - - // Return true iff "*ptr" points to a heap object - // ("*ptr" can point at the start or inside of a heap object - // so that this works e.g. for pointers to C++ arrays, C++ strings, - // multiple-inherited objects, or pointers to members). - // We also fill *object_size for this object then - // and we move "*ptr" to point to the very start of the heap object. - static inline bool HaveOnHeapLocked(const void** ptr, size_t* object_size); - - // Helper to shutdown heap leak checker when it's not needed - // or can't function properly. - static void TurnItselfOffLocked(); - - // Internally-used c-tor to start whole-executable checking. - HeapLeakChecker(); - - // ----------------------------------------------------------------------- // - // Friends and externally accessed helpers. - - // Helper for VerifyHeapProfileTableStackGet in the unittest - // to get the recorded allocation caller for ptr, - // which must be a heap object. - static const void* GetAllocCaller(void* ptr); - friend void VerifyHeapProfileTableStackGet(); - - // This gets to execute before constructors for all global objects - static void BeforeConstructorsLocked(); - friend void HeapLeakChecker_BeforeConstructors(); - - // This gets to execute after destructors for all global objects - friend void HeapLeakChecker_AfterDestructors(); - - // Full starting of recommended whole-program checking. - friend void HeapLeakChecker_InternalInitStart(); - - // Runs REGISTER_HEAPCHECK_CLEANUP cleanups and potentially - // calls DoMainHeapCheck - friend void HeapLeakChecker_RunHeapCleanups(); - - // ----------------------------------------------------------------------- // - // Member data. - - class SpinLock* lock_; // to make HeapLeakChecker objects thread-safe - const char* name_; // our remembered name (we own it) - // NULL means this leak checker is a noop - - // Snapshot taken when the checker was created. May be NULL - // for the global heap checker object. We use void* instead of - // HeapProfileTable::Snapshot* to avoid including heap-profile-table.h. - void* start_snapshot_; - - bool has_checked_; // if we have done the leak check, so these are ready: - ssize_t inuse_bytes_increase_; // bytes-in-use increase for this checker - ssize_t inuse_allocs_increase_; // allocations-in-use increase - // for this checker - bool keep_profiles_; // iff we should keep the heap profiles we've made - - // ----------------------------------------------------------------------- // - - // Disallow "evil" constructors. - HeapLeakChecker(const HeapLeakChecker&); - void operator=(const HeapLeakChecker&); -}; - - -// Holds a pointer that will not be traversed by the heap checker. -// Contrast with HeapLeakChecker::IgnoreObject(o), in which o and -// all objects reachable from o are ignored by the heap checker. -template -class HiddenPointer { - public: - explicit HiddenPointer(T* t) - : masked_t_(reinterpret_cast(t) ^ kHideMask) { - } - // Returns unhidden pointer. Be careful where you save the result. - T* get() const { return reinterpret_cast(masked_t_ ^ kHideMask); } - - private: - // Arbitrary value, but not such that xor'ing with it is likely - // to map one valid pointer to another valid pointer: - static const uintptr_t kHideMask = - static_cast(0xF03A5F7BF03A5F7Bll); - uintptr_t masked_t_; -}; - -// A class that exists solely to run its destructor. This class should not be -// used directly, but instead by the REGISTER_HEAPCHECK_CLEANUP macro below. -class PERFTOOLS_DLL_DECL HeapCleaner { - public: - typedef void (*void_function)(void); - HeapCleaner(void_function f); - static void RunHeapCleanups(); - private: - static std::vector* heap_cleanups_; -}; - -// A macro to declare module heap check cleanup tasks -// (they run only if we are doing heap leak checking.) -// 'body' should be the cleanup code to run. 'name' doesn't matter, -// but must be unique amongst all REGISTER_HEAPCHECK_CLEANUP calls. -#define REGISTER_HEAPCHECK_CLEANUP(name, body) \ - namespace { \ - void heapcheck_cleanup_##name() { body; } \ - static HeapCleaner heapcheck_cleaner_##name(&heapcheck_cleanup_##name); \ - } - -#endif // BASE_HEAP_CHECKER_H_ diff --git a/contrib/libtcmalloc/include/gperftools/heap-profiler.h b/contrib/libtcmalloc/include/gperftools/heap-profiler.h deleted file mode 100644 index 9b673645747..00000000000 --- a/contrib/libtcmalloc/include/gperftools/heap-profiler.h +++ /dev/null @@ -1,105 +0,0 @@ -// -*- Mode: C++; c-basic-offset: 2; indent-tabs-mode: nil -*- -/* Copyright (c) 2005, Google Inc. - * All rights reserved. - * - * Redistribution and use in source and binary forms, with or without - * modification, are permitted provided that the following conditions are - * met: - * - * * Redistributions of source code must retain the above copyright - * notice, this list of conditions and the following disclaimer. - * * Redistributions in binary form must reproduce the above - * copyright notice, this list of conditions and the following disclaimer - * in the documentation and/or other materials provided with the - * distribution. - * * Neither the name of Google Inc. nor the names of its - * contributors may be used to endorse or promote products derived from - * this software without specific prior written permission. - * - * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS - * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT - * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR - * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT - * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, - * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT - * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, - * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY - * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT - * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE - * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. - * - * --- - * Author: Sanjay Ghemawat - * - * Module for heap-profiling. - * - * For full(er) information, see doc/heapprofile.html - * - * This module can be linked into your program with - * no slowdown caused by this unless you activate the profiler - * using one of the following methods: - * - * 1. Before starting the program, set the environment variable - * "HEAPPROFILE" to be the name of the file to which the profile - * data should be written. - * - * 2. Programmatically, start and stop the profiler using the - * routines "HeapProfilerStart(filename)" and "HeapProfilerStop()". - * - */ - -#ifndef BASE_HEAP_PROFILER_H_ -#define BASE_HEAP_PROFILER_H_ - -#include - -/* Annoying stuff for windows; makes sure clients can import these functions */ -#ifndef PERFTOOLS_DLL_DECL -# ifdef _WIN32 -# define PERFTOOLS_DLL_DECL __declspec(dllimport) -# else -# define PERFTOOLS_DLL_DECL -# endif -#endif - -/* All this code should be usable from within C apps. */ -#ifdef __cplusplus -extern "C" { -#endif - -/* Start profiling and arrange to write profile data to file names - * of the form: "prefix.0000", "prefix.0001", ... - */ -PERFTOOLS_DLL_DECL void HeapProfilerStart(const char* prefix); - -/* Returns non-zero if we are currently profiling the heap. (Returns - * an int rather than a bool so it's usable from C.) This is true - * between calls to HeapProfilerStart() and HeapProfilerStop(), and - * also if the program has been run with HEAPPROFILER, or some other - * way to turn on whole-program profiling. - */ -int IsHeapProfilerRunning(); - -/* Stop heap profiling. Can be restarted again with HeapProfilerStart(), - * but the currently accumulated profiling information will be cleared. - */ -PERFTOOLS_DLL_DECL void HeapProfilerStop(); - -/* Dump a profile now - can be used for dumping at a hopefully - * quiescent state in your program, in order to more easily track down - * memory leaks. Will include the reason in the logged message - */ -PERFTOOLS_DLL_DECL void HeapProfilerDump(const char *reason); - -/* Generate current heap profiling information. - * Returns an empty string when heap profiling is not active. - * The returned pointer is a '\0'-terminated string allocated using malloc() - * and should be free()-ed as soon as the caller does not need it anymore. - */ -PERFTOOLS_DLL_DECL char* GetHeapProfile(); - -#ifdef __cplusplus -} // extern "C" -#endif - -#endif /* BASE_HEAP_PROFILER_H_ */ diff --git a/contrib/libtcmalloc/include/gperftools/malloc_extension.h b/contrib/libtcmalloc/include/gperftools/malloc_extension.h deleted file mode 100644 index 689b5f17cef..00000000000 --- a/contrib/libtcmalloc/include/gperftools/malloc_extension.h +++ /dev/null @@ -1,434 +0,0 @@ -// -*- Mode: C++; c-basic-offset: 2; indent-tabs-mode: nil -*- -// Copyright (c) 2005, Google Inc. -// All rights reserved. -// -// Redistribution and use in source and binary forms, with or without -// modification, are permitted provided that the following conditions are -// met: -// -// * Redistributions of source code must retain the above copyright -// notice, this list of conditions and the following disclaimer. -// * Redistributions in binary form must reproduce the above -// copyright notice, this list of conditions and the following disclaimer -// in the documentation and/or other materials provided with the -// distribution. -// * Neither the name of Google Inc. nor the names of its -// contributors may be used to endorse or promote products derived from -// this software without specific prior written permission. -// -// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS -// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT -// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR -// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT -// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, -// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT -// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, -// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY -// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT -// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE -// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. - -// --- -// Author: Sanjay Ghemawat -// -// Extra extensions exported by some malloc implementations. These -// extensions are accessed through a virtual base class so an -// application can link against a malloc that does not implement these -// extensions, and it will get default versions that do nothing. -// -// NOTE FOR C USERS: If you wish to use this functionality from within -// a C program, see malloc_extension_c.h. - -#ifndef BASE_MALLOC_EXTENSION_H_ -#define BASE_MALLOC_EXTENSION_H_ - -#include -// I can't #include config.h in this public API file, but I should -// really use configure (and make malloc_extension.h a .in file) to -// figure out if the system has stdint.h or not. But I'm lazy, so -// for now I'm assuming it's a problem only with MSVC. -#ifndef _MSC_VER -#include -#endif -#include -#include - -// Annoying stuff for windows -- makes sure clients can import these functions -#ifndef PERFTOOLS_DLL_DECL -# ifdef _WIN32 -# define PERFTOOLS_DLL_DECL __declspec(dllimport) -# else -# define PERFTOOLS_DLL_DECL -# endif -#endif - -static const int kMallocHistogramSize = 64; - -// One day, we could support other types of writers (perhaps for C?) -typedef std::string MallocExtensionWriter; - -namespace base { -struct MallocRange; -} - -// Interface to a pluggable system allocator. -class PERFTOOLS_DLL_DECL SysAllocator { - public: - SysAllocator() { - } - virtual ~SysAllocator(); - - // Allocates "size"-byte of memory from system aligned with "alignment". - // Returns NULL if failed. Otherwise, the returned pointer p up to and - // including (p + actual_size -1) have been allocated. - virtual void* Alloc(size_t size, size_t *actual_size, size_t alignment) = 0; -}; - -// The default implementations of the following routines do nothing. -// All implementations should be thread-safe; the current one -// (TCMallocImplementation) is. -class PERFTOOLS_DLL_DECL MallocExtension { - public: - virtual ~MallocExtension(); - - // Call this very early in the program execution -- say, in a global - // constructor -- to set up parameters and state needed by all - // instrumented malloc implemenatations. One example: this routine - // sets environemnt variables to tell STL to use libc's malloc() - // instead of doing its own memory management. This is safe to call - // multiple times, as long as each time is before threads start up. - static void Initialize(); - - // See "verify_memory.h" to see what these routines do - virtual bool VerifyAllMemory(); - virtual bool VerifyNewMemory(const void* p); - virtual bool VerifyArrayNewMemory(const void* p); - virtual bool VerifyMallocMemory(const void* p); - virtual bool MallocMemoryStats(int* blocks, size_t* total, - int histogram[kMallocHistogramSize]); - - // Get a human readable description of the following malloc data structures. - // - Total inuse memory by application. - // - Free memory(thread, central and page heap), - // - Freelist of central cache, each class. - // - Page heap freelist. - // The state is stored as a null-terminated string - // in a prefix of "buffer[0,buffer_length-1]". - // REQUIRES: buffer_length > 0. - virtual void GetStats(char* buffer, int buffer_length); - - // Outputs to "writer" a sample of live objects and the stack traces - // that allocated these objects. The format of the returned output - // is equivalent to the output of the heap profiler and can - // therefore be passed to "pprof". This function is equivalent to - // ReadStackTraces. The main difference is that this function returns - // serialized data appropriately formatted for use by the pprof tool. - // NOTE: by default, tcmalloc does not do any heap sampling, and this - // function will always return an empty sample. To get useful - // data from GetHeapSample, you must also set the environment - // variable TCMALLOC_SAMPLE_PARAMETER to a value such as 524288. - virtual void GetHeapSample(MallocExtensionWriter* writer); - - // Outputs to "writer" the stack traces that caused growth in the - // address space size. The format of the returned output is - // equivalent to the output of the heap profiler and can therefore - // be passed to "pprof". This function is equivalent to - // ReadHeapGrowthStackTraces. The main difference is that this function - // returns serialized data appropriately formatted for use by the - // pprof tool. (This does not depend on, or require, - // TCMALLOC_SAMPLE_PARAMETER.) - virtual void GetHeapGrowthStacks(MallocExtensionWriter* writer); - - // Invokes func(arg, range) for every controlled memory - // range. *range is filled in with information about the range. - // - // This is a best-effort interface useful only for performance - // analysis. The implementation may not call func at all. - typedef void (RangeFunction)(void*, const base::MallocRange*); - virtual void Ranges(void* arg, RangeFunction func); - - // ------------------------------------------------------------------- - // Control operations for getting and setting malloc implementation - // specific parameters. Some currently useful properties: - // - // generic - // ------- - // "generic.current_allocated_bytes" - // Number of bytes currently allocated by application - // This property is not writable. - // - // "generic.heap_size" - // Number of bytes in the heap == - // current_allocated_bytes + - // fragmentation + - // freed memory regions - // This property is not writable. - // - // tcmalloc - // -------- - // "tcmalloc.max_total_thread_cache_bytes" - // Upper limit on total number of bytes stored across all - // per-thread caches. Default: 16MB. - // - // "tcmalloc.current_total_thread_cache_bytes" - // Number of bytes used across all thread caches. - // This property is not writable. - // - // "tcmalloc.central_cache_free_bytes" - // Number of free bytes in the central cache that have been - // assigned to size classes. They always count towards virtual - // memory usage, and unless the underlying memory is swapped out - // by the OS, they also count towards physical memory usage. - // This property is not writable. - // - // "tcmalloc.transfer_cache_free_bytes" - // Number of free bytes that are waiting to be transfered between - // the central cache and a thread cache. They always count - // towards virtual memory usage, and unless the underlying memory - // is swapped out by the OS, they also count towards physical - // memory usage. This property is not writable. - // - // "tcmalloc.thread_cache_free_bytes" - // Number of free bytes in thread caches. They always count - // towards virtual memory usage, and unless the underlying memory - // is swapped out by the OS, they also count towards physical - // memory usage. This property is not writable. - // - // "tcmalloc.pageheap_free_bytes" - // Number of bytes in free, mapped pages in page heap. These - // bytes can be used to fulfill allocation requests. They - // always count towards virtual memory usage, and unless the - // underlying memory is swapped out by the OS, they also count - // towards physical memory usage. This property is not writable. - // - // "tcmalloc.pageheap_unmapped_bytes" - // Number of bytes in free, unmapped pages in page heap. - // These are bytes that have been released back to the OS, - // possibly by one of the MallocExtension "Release" calls. - // They can be used to fulfill allocation requests, but - // typically incur a page fault. They always count towards - // virtual memory usage, and depending on the OS, typically - // do not count towards physical memory usage. This property - // is not writable. - // ------------------------------------------------------------------- - - // Get the named "property"'s value. Returns true if the property - // is known. Returns false if the property is not a valid property - // name for the current malloc implementation. - // REQUIRES: property != NULL; value != NULL - virtual bool GetNumericProperty(const char* property, size_t* value); - - // Set the named "property"'s value. Returns true if the property - // is known and writable. Returns false if the property is not a - // valid property name for the current malloc implementation, or - // is not writable. - // REQUIRES: property != NULL - virtual bool SetNumericProperty(const char* property, size_t value); - - // Mark the current thread as "idle". This routine may optionally - // be called by threads as a hint to the malloc implementation that - // any thread-specific resources should be released. Note: this may - // be an expensive routine, so it should not be called too often. - // - // Also, if the code that calls this routine will go to sleep for - // a while, it should take care to not allocate anything between - // the call to this routine and the beginning of the sleep. - // - // Most malloc implementations ignore this routine. - virtual void MarkThreadIdle(); - - // Mark the current thread as "busy". This routine should be - // called after MarkThreadIdle() if the thread will now do more - // work. If this method is not called, performance may suffer. - // - // Most malloc implementations ignore this routine. - virtual void MarkThreadBusy(); - - // Gets the system allocator used by the malloc extension instance. Returns - // NULL for malloc implementations that do not support pluggable system - // allocators. - virtual SysAllocator* GetSystemAllocator(); - - // Sets the system allocator to the specified. - // - // Users could register their own system allocators for malloc implementation - // that supports pluggable system allocators, such as TCMalloc, by doing: - // alloc = new MyOwnSysAllocator(); - // MallocExtension::instance()->SetSystemAllocator(alloc); - // It's up to users whether to fall back (recommended) to the default - // system allocator (use GetSystemAllocator() above) or not. The caller is - // responsible to any necessary locking. - // See tcmalloc/system-alloc.h for the interface and - // tcmalloc/memfs_malloc.cc for the examples. - // - // It's a no-op for malloc implementations that do not support pluggable - // system allocators. - virtual void SetSystemAllocator(SysAllocator *a); - - // Try to release num_bytes of free memory back to the operating - // system for reuse. Use this extension with caution -- to get this - // memory back may require faulting pages back in by the OS, and - // that may be slow. (Currently only implemented in tcmalloc.) - virtual void ReleaseToSystem(size_t num_bytes); - - // Same as ReleaseToSystem() but release as much memory as possible. - virtual void ReleaseFreeMemory(); - - // Sets the rate at which we release unused memory to the system. - // Zero means we never release memory back to the system. Increase - // this flag to return memory faster; decrease it to return memory - // slower. Reasonable rates are in the range [0,10]. (Currently - // only implemented in tcmalloc). - virtual void SetMemoryReleaseRate(double rate); - - // Gets the release rate. Returns a value < 0 if unknown. - virtual double GetMemoryReleaseRate(); - - // Returns the estimated number of bytes that will be allocated for - // a request of "size" bytes. This is an estimate: an allocation of - // SIZE bytes may reserve more bytes, but will never reserve less. - // (Currently only implemented in tcmalloc, other implementations - // always return SIZE.) - // This is equivalent to malloc_good_size() in OS X. - virtual size_t GetEstimatedAllocatedSize(size_t size); - - // Returns the actual number N of bytes reserved by tcmalloc for the - // pointer p. The client is allowed to use the range of bytes - // [p, p+N) in any way it wishes (i.e. N is the "usable size" of this - // allocation). This number may be equal to or greater than the number - // of bytes requested when p was allocated. - // p must have been allocated by this malloc implementation, - // must not be an interior pointer -- that is, must be exactly - // the pointer returned to by malloc() et al., not some offset - // from that -- and should not have been freed yet. p may be NULL. - // (Currently only implemented in tcmalloc; other implementations - // will return 0.) - // This is equivalent to malloc_size() in OS X, malloc_usable_size() - // in glibc, and _msize() for windows. - virtual size_t GetAllocatedSize(const void* p); - - // Returns kOwned if this malloc implementation allocated the memory - // pointed to by p, or kNotOwned if some other malloc implementation - // allocated it or p is NULL. May also return kUnknownOwnership if - // the malloc implementation does not keep track of ownership. - // REQUIRES: p must be a value returned from a previous call to - // malloc(), calloc(), realloc(), memalign(), posix_memalign(), - // valloc(), pvalloc(), new, or new[], and must refer to memory that - // is currently allocated (so, for instance, you should not pass in - // a pointer after having called free() on it). - enum Ownership { - // NOTE: Enum values MUST be kept in sync with the version in - // malloc_extension_c.h - kUnknownOwnership = 0, - kOwned, - kNotOwned - }; - virtual Ownership GetOwnership(const void* p); - - // The current malloc implementation. Always non-NULL. - static MallocExtension* instance(); - - // Change the malloc implementation. Typically called by the - // malloc implementation during initialization. - static void Register(MallocExtension* implementation); - - // Returns detailed information about malloc's freelists. For each list, - // return a FreeListInfo: - struct FreeListInfo { - size_t min_object_size; - size_t max_object_size; - size_t total_bytes_free; - const char* type; - }; - // Each item in the vector refers to a different freelist. The lists - // are identified by the range of allocations that objects in the - // list can satisfy ([min_object_size, max_object_size]) and the - // type of freelist (see below). The current size of the list is - // returned in total_bytes_free (which count against a processes - // resident and virtual size). - // - // Currently supported types are: - // - // "tcmalloc.page{_unmapped}" - tcmalloc's page heap. An entry for each size - // class in the page heap is returned. Bytes in "page_unmapped" - // are no longer backed by physical memory and do not count against - // the resident size of a process. - // - // "tcmalloc.large{_unmapped}" - tcmalloc's list of objects larger - // than the largest page heap size class. Only one "large" - // entry is returned. There is no upper-bound on the size - // of objects in the large free list; this call returns - // kint64max for max_object_size. Bytes in - // "large_unmapped" are no longer backed by physical memory - // and do not count against the resident size of a process. - // - // "tcmalloc.central" - tcmalloc's central free-list. One entry per - // size-class is returned. Never unmapped. - // - // "debug.free_queue" - free objects queued by the debug allocator - // and not returned to tcmalloc. - // - // "tcmalloc.thread" - tcmalloc's per-thread caches. Never unmapped. - virtual void GetFreeListSizes(std::vector* v); - - // Get a list of stack traces of sampled allocation points. Returns - // a pointer to a "new[]-ed" result array, and stores the sample - // period in "sample_period". - // - // The state is stored as a sequence of adjacent entries - // in the returned array. Each entry has the following form: - // uintptr_t count; // Number of objects with following trace - // uintptr_t size; // Total size of objects with following trace - // uintptr_t depth; // Number of PC values in stack trace - // void* stack[depth]; // PC values that form the stack trace - // - // The list of entries is terminated by a "count" of 0. - // - // It is the responsibility of the caller to "delete[]" the returned array. - // - // May return NULL to indicate no results. - // - // This is an internal extension. Callers should use the more - // convenient "GetHeapSample(string*)" method defined above. - virtual void** ReadStackTraces(int* sample_period); - - // Like ReadStackTraces(), but returns stack traces that caused growth - // in the address space size. - virtual void** ReadHeapGrowthStackTraces(); - - // Returns the size in bytes of the calling threads cache. - virtual size_t GetThreadCacheSize(); - - // Like MarkThreadIdle, but does not destroy the internal data - // structures of the thread cache. When the thread resumes, it wil - // have an empty cache but will not need to pay to reconstruct the - // cache data structures. - virtual void MarkThreadTemporarilyIdle(); -}; - -namespace base { - -// Information passed per range. More fields may be added later. -struct MallocRange { - enum Type { - INUSE, // Application is using this range - FREE, // Range is currently free - UNMAPPED, // Backing physical memory has been returned to the OS - UNKNOWN - // More enum values may be added in the future - }; - - uintptr_t address; // Address of range - size_t length; // Byte length of range - Type type; // Type of this range - double fraction; // Fraction of range that is being used (0 if !INUSE) - - // Perhaps add the following: - // - stack trace if this range was sampled - // - heap growth stack trace if applicable to this range - // - age when allocated (for inuse) or freed (if not in use) -}; - -} // namespace base - -#endif // BASE_MALLOC_EXTENSION_H_ diff --git a/contrib/libtcmalloc/include/gperftools/malloc_extension_c.h b/contrib/libtcmalloc/include/gperftools/malloc_extension_c.h deleted file mode 100644 index 70ff6868ecf..00000000000 --- a/contrib/libtcmalloc/include/gperftools/malloc_extension_c.h +++ /dev/null @@ -1,101 +0,0 @@ -/* Copyright (c) 2008, Google Inc. - * All rights reserved. - * - * Redistribution and use in source and binary forms, with or without - * modification, are permitted provided that the following conditions are - * met: - * - * * Redistributions of source code must retain the above copyright - * notice, this list of conditions and the following disclaimer. - * * Redistributions in binary form must reproduce the above - * copyright notice, this list of conditions and the following disclaimer - * in the documentation and/or other materials provided with the - * distribution. - * * Neither the name of Google Inc. nor the names of its - * contributors may be used to endorse or promote products derived from - * this software without specific prior written permission. - * - * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS - * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT - * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR - * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT - * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, - * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT - * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, - * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY - * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT - * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE - * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. - * - * -- - * Author: Craig Silverstein - * - * C shims for the C++ malloc_extension.h. See malloc_extension.h for - * details. Note these C shims always work on - * MallocExtension::instance(); it is not possible to have more than - * one MallocExtension object in C applications. - */ - -#ifndef _MALLOC_EXTENSION_C_H_ -#define _MALLOC_EXTENSION_C_H_ - -#include -#include - -/* Annoying stuff for windows -- makes sure clients can import these fns */ -#ifndef PERFTOOLS_DLL_DECL -# ifdef _WIN32 -# define PERFTOOLS_DLL_DECL __declspec(dllimport) -# else -# define PERFTOOLS_DLL_DECL -# endif -#endif - -#ifdef __cplusplus -extern "C" { -#endif - -#define kMallocExtensionHistogramSize 64 - -PERFTOOLS_DLL_DECL int MallocExtension_VerifyAllMemory(void); -PERFTOOLS_DLL_DECL int MallocExtension_VerifyNewMemory(const void* p); -PERFTOOLS_DLL_DECL int MallocExtension_VerifyArrayNewMemory(const void* p); -PERFTOOLS_DLL_DECL int MallocExtension_VerifyMallocMemory(const void* p); -PERFTOOLS_DLL_DECL int MallocExtension_MallocMemoryStats(int* blocks, size_t* total, - int histogram[kMallocExtensionHistogramSize]); -PERFTOOLS_DLL_DECL void MallocExtension_GetStats(char* buffer, int buffer_length); - -/* TODO(csilvers): write a C version of these routines, that perhaps - * takes a function ptr and a void *. - */ -/* void MallocExtension_GetHeapSample(string* result); */ -/* void MallocExtension_GetHeapGrowthStacks(string* result); */ - -PERFTOOLS_DLL_DECL int MallocExtension_GetNumericProperty(const char* property, size_t* value); -PERFTOOLS_DLL_DECL int MallocExtension_SetNumericProperty(const char* property, size_t value); -PERFTOOLS_DLL_DECL void MallocExtension_MarkThreadIdle(void); -PERFTOOLS_DLL_DECL void MallocExtension_MarkThreadBusy(void); -PERFTOOLS_DLL_DECL void MallocExtension_ReleaseToSystem(size_t num_bytes); -PERFTOOLS_DLL_DECL void MallocExtension_ReleaseFreeMemory(void); -PERFTOOLS_DLL_DECL size_t MallocExtension_GetEstimatedAllocatedSize(size_t size); -PERFTOOLS_DLL_DECL size_t MallocExtension_GetAllocatedSize(const void* p); -PERFTOOLS_DLL_DECL size_t MallocExtension_GetThreadCacheSize(void); -PERFTOOLS_DLL_DECL void MallocExtension_MarkThreadTemporarilyIdle(void); - -/* - * NOTE: These enum values MUST be kept in sync with the version in - * malloc_extension.h - */ -typedef enum { - MallocExtension_kUnknownOwnership = 0, - MallocExtension_kOwned, - MallocExtension_kNotOwned -} MallocExtension_Ownership; - -PERFTOOLS_DLL_DECL MallocExtension_Ownership MallocExtension_GetOwnership(const void* p); - -#ifdef __cplusplus -} /* extern "C" */ -#endif - -#endif /* _MALLOC_EXTENSION_C_H_ */ diff --git a/contrib/libtcmalloc/include/gperftools/malloc_hook.h b/contrib/libtcmalloc/include/gperftools/malloc_hook.h deleted file mode 100644 index b76411fb590..00000000000 --- a/contrib/libtcmalloc/include/gperftools/malloc_hook.h +++ /dev/null @@ -1,359 +0,0 @@ -// -*- Mode: C++; c-basic-offset: 2; indent-tabs-mode: nil -*- -// Copyright (c) 2005, Google Inc. -// All rights reserved. -// -// Redistribution and use in source and binary forms, with or without -// modification, are permitted provided that the following conditions are -// met: -// -// * Redistributions of source code must retain the above copyright -// notice, this list of conditions and the following disclaimer. -// * Redistributions in binary form must reproduce the above -// copyright notice, this list of conditions and the following disclaimer -// in the documentation and/or other materials provided with the -// distribution. -// * Neither the name of Google Inc. nor the names of its -// contributors may be used to endorse or promote products derived from -// this software without specific prior written permission. -// -// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS -// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT -// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR -// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT -// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, -// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT -// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, -// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY -// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT -// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE -// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. - -// --- -// Author: Sanjay Ghemawat -// -// Some of our malloc implementations can invoke the following hooks whenever -// memory is allocated or deallocated. MallocHook is thread-safe, and things -// you do before calling AddFooHook(MyHook) are visible to any resulting calls -// to MyHook. Hooks must be thread-safe. If you write: -// -// CHECK(MallocHook::AddNewHook(&MyNewHook)); -// -// MyNewHook will be invoked in subsequent calls in the current thread, but -// there are no guarantees on when it might be invoked in other threads. -// -// There are a limited number of slots available for each hook type. Add*Hook -// will return false if there are no slots available. Remove*Hook will return -// false if the given hook was not already installed. -// -// The order in which individual hooks are called in Invoke*Hook is undefined. -// -// It is safe for a hook to remove itself within Invoke*Hook and add other -// hooks. Any hooks added inside a hook invocation (for the same hook type) -// will not be invoked for the current invocation. -// -// One important user of these hooks is the heap profiler. -// -// CAVEAT: If you add new MallocHook::Invoke* calls then those calls must be -// directly in the code of the (de)allocation function that is provided to the -// user and that function must have an ATTRIBUTE_SECTION(malloc_hook) attribute. -// -// Note: the Invoke*Hook() functions are defined in malloc_hook-inl.h. If you -// need to invoke a hook (which you shouldn't unless you're part of tcmalloc), -// be sure to #include malloc_hook-inl.h in addition to malloc_hook.h. -// -// NOTE FOR C USERS: If you want to use malloc_hook functionality from -// a C program, #include malloc_hook_c.h instead of this file. - -#ifndef _MALLOC_HOOK_H_ -#define _MALLOC_HOOK_H_ - -#include -#include -extern "C" { -#include "malloc_hook_c.h" // a C version of the malloc_hook interface -} - -// Annoying stuff for windows -- makes sure clients can import these functions -#ifndef PERFTOOLS_DLL_DECL -# ifdef _WIN32 -# define PERFTOOLS_DLL_DECL __declspec(dllimport) -# else -# define PERFTOOLS_DLL_DECL -# endif -#endif - -// The C++ methods below call the C version (MallocHook_*), and thus -// convert between an int and a bool. Windows complains about this -// (a "performance warning") which we don't care about, so we suppress. -#ifdef _MSC_VER -#pragma warning(push) -#pragma warning(disable:4800) -#endif - -// Note: malloc_hook_c.h defines MallocHook_*Hook and -// MallocHook_{Add,Remove}*Hook. The version of these inside the MallocHook -// class are defined in terms of the malloc_hook_c version. See malloc_hook_c.h -// for details of these types/functions. - -class PERFTOOLS_DLL_DECL MallocHook { - public: - // The NewHook is invoked whenever an object is allocated. - // It may be passed NULL if the allocator returned NULL. - typedef MallocHook_NewHook NewHook; - inline static bool AddNewHook(NewHook hook) { - return MallocHook_AddNewHook(hook); - } - inline static bool RemoveNewHook(NewHook hook) { - return MallocHook_RemoveNewHook(hook); - } - inline static void InvokeNewHook(const void* p, size_t s); - - // The DeleteHook is invoked whenever an object is deallocated. - // It may be passed NULL if the caller is trying to delete NULL. - typedef MallocHook_DeleteHook DeleteHook; - inline static bool AddDeleteHook(DeleteHook hook) { - return MallocHook_AddDeleteHook(hook); - } - inline static bool RemoveDeleteHook(DeleteHook hook) { - return MallocHook_RemoveDeleteHook(hook); - } - inline static void InvokeDeleteHook(const void* p); - - // The PreMmapHook is invoked with mmap or mmap64 arguments just - // before the call is actually made. Such a hook may be useful - // in memory limited contexts, to catch allocations that will exceed - // a memory limit, and take outside actions to increase that limit. - typedef MallocHook_PreMmapHook PreMmapHook; - inline static bool AddPreMmapHook(PreMmapHook hook) { - return MallocHook_AddPreMmapHook(hook); - } - inline static bool RemovePreMmapHook(PreMmapHook hook) { - return MallocHook_RemovePreMmapHook(hook); - } - inline static void InvokePreMmapHook(const void* start, - size_t size, - int protection, - int flags, - int fd, - off_t offset); - - // The MmapReplacement is invoked after the PreMmapHook but before - // the call is actually made. The MmapReplacement should return true - // if it handled the call, or false if it is still necessary to - // call mmap/mmap64. - // This should be used only by experts, and users must be be - // extremely careful to avoid recursive calls to mmap. The replacement - // should be async signal safe. - // Only one MmapReplacement is supported. After setting an MmapReplacement - // you must call RemoveMmapReplacement before calling SetMmapReplacement - // again. - typedef MallocHook_MmapReplacement MmapReplacement; - inline static bool SetMmapReplacement(MmapReplacement hook) { - return MallocHook_SetMmapReplacement(hook); - } - inline static bool RemoveMmapReplacement(MmapReplacement hook) { - return MallocHook_RemoveMmapReplacement(hook); - } - inline static bool InvokeMmapReplacement(const void* start, - size_t size, - int protection, - int flags, - int fd, - off_t offset, - void** result); - - - // The MmapHook is invoked whenever a region of memory is mapped. - // It may be passed MAP_FAILED if the mmap failed. - typedef MallocHook_MmapHook MmapHook; - inline static bool AddMmapHook(MmapHook hook) { - return MallocHook_AddMmapHook(hook); - } - inline static bool RemoveMmapHook(MmapHook hook) { - return MallocHook_RemoveMmapHook(hook); - } - inline static void InvokeMmapHook(const void* result, - const void* start, - size_t size, - int protection, - int flags, - int fd, - off_t offset); - - // The MunmapReplacement is invoked with munmap arguments just before - // the call is actually made. The MunmapReplacement should return true - // if it handled the call, or false if it is still necessary to - // call munmap. - // This should be used only by experts. The replacement should be - // async signal safe. - // Only one MunmapReplacement is supported. After setting an - // MunmapReplacement you must call RemoveMunmapReplacement before - // calling SetMunmapReplacement again. - typedef MallocHook_MunmapReplacement MunmapReplacement; - inline static bool SetMunmapReplacement(MunmapReplacement hook) { - return MallocHook_SetMunmapReplacement(hook); - } - inline static bool RemoveMunmapReplacement(MunmapReplacement hook) { - return MallocHook_RemoveMunmapReplacement(hook); - } - inline static bool InvokeMunmapReplacement(const void* p, - size_t size, - int* result); - - // The MunmapHook is invoked whenever a region of memory is unmapped. - typedef MallocHook_MunmapHook MunmapHook; - inline static bool AddMunmapHook(MunmapHook hook) { - return MallocHook_AddMunmapHook(hook); - } - inline static bool RemoveMunmapHook(MunmapHook hook) { - return MallocHook_RemoveMunmapHook(hook); - } - inline static void InvokeMunmapHook(const void* p, size_t size); - - // The MremapHook is invoked whenever a region of memory is remapped. - typedef MallocHook_MremapHook MremapHook; - inline static bool AddMremapHook(MremapHook hook) { - return MallocHook_AddMremapHook(hook); - } - inline static bool RemoveMremapHook(MremapHook hook) { - return MallocHook_RemoveMremapHook(hook); - } - inline static void InvokeMremapHook(const void* result, - const void* old_addr, - size_t old_size, - size_t new_size, - int flags, - const void* new_addr); - - // The PreSbrkHook is invoked just before sbrk is called -- except when - // the increment is 0. This is because sbrk(0) is often called - // to get the top of the memory stack, and is not actually a - // memory-allocation call. It may be useful in memory-limited contexts, - // to catch allocations that will exceed the limit and take outside - // actions to increase such a limit. - typedef MallocHook_PreSbrkHook PreSbrkHook; - inline static bool AddPreSbrkHook(PreSbrkHook hook) { - return MallocHook_AddPreSbrkHook(hook); - } - inline static bool RemovePreSbrkHook(PreSbrkHook hook) { - return MallocHook_RemovePreSbrkHook(hook); - } - inline static void InvokePreSbrkHook(ptrdiff_t increment); - - // The SbrkHook is invoked whenever sbrk is called -- except when - // the increment is 0. This is because sbrk(0) is often called - // to get the top of the memory stack, and is not actually a - // memory-allocation call. - typedef MallocHook_SbrkHook SbrkHook; - inline static bool AddSbrkHook(SbrkHook hook) { - return MallocHook_AddSbrkHook(hook); - } - inline static bool RemoveSbrkHook(SbrkHook hook) { - return MallocHook_RemoveSbrkHook(hook); - } - inline static void InvokeSbrkHook(const void* result, ptrdiff_t increment); - - // Get the current stack trace. Try to skip all routines up to and - // and including the caller of MallocHook::Invoke*. - // Use "skip_count" (similarly to GetStackTrace from stacktrace.h) - // as a hint about how many routines to skip if better information - // is not available. - inline static int GetCallerStackTrace(void** result, int max_depth, - int skip_count) { - return MallocHook_GetCallerStackTrace(result, max_depth, skip_count); - } - - // Unhooked versions of mmap() and munmap(). These should be used - // only by experts, since they bypass heapchecking, etc. - // Note: These do not run hooks, but they still use the MmapReplacement - // and MunmapReplacement. - static void* UnhookedMMap(void *start, size_t length, int prot, int flags, - int fd, off_t offset); - static int UnhookedMUnmap(void *start, size_t length); - - // The following are DEPRECATED. - inline static NewHook GetNewHook(); - inline static NewHook SetNewHook(NewHook hook) { - return MallocHook_SetNewHook(hook); - } - - inline static DeleteHook GetDeleteHook(); - inline static DeleteHook SetDeleteHook(DeleteHook hook) { - return MallocHook_SetDeleteHook(hook); - } - - inline static PreMmapHook GetPreMmapHook(); - inline static PreMmapHook SetPreMmapHook(PreMmapHook hook) { - return MallocHook_SetPreMmapHook(hook); - } - - inline static MmapHook GetMmapHook(); - inline static MmapHook SetMmapHook(MmapHook hook) { - return MallocHook_SetMmapHook(hook); - } - - inline static MunmapHook GetMunmapHook(); - inline static MunmapHook SetMunmapHook(MunmapHook hook) { - return MallocHook_SetMunmapHook(hook); - } - - inline static MremapHook GetMremapHook(); - inline static MremapHook SetMremapHook(MremapHook hook) { - return MallocHook_SetMremapHook(hook); - } - - inline static PreSbrkHook GetPreSbrkHook(); - inline static PreSbrkHook SetPreSbrkHook(PreSbrkHook hook) { - return MallocHook_SetPreSbrkHook(hook); - } - - inline static SbrkHook GetSbrkHook(); - inline static SbrkHook SetSbrkHook(SbrkHook hook) { - return MallocHook_SetSbrkHook(hook); - } - // End of DEPRECATED methods. - - private: - // Slow path versions of Invoke*Hook. - static void InvokeNewHookSlow(const void* p, size_t s); - static void InvokeDeleteHookSlow(const void* p); - static void InvokePreMmapHookSlow(const void* start, - size_t size, - int protection, - int flags, - int fd, - off_t offset); - static void InvokeMmapHookSlow(const void* result, - const void* start, - size_t size, - int protection, - int flags, - int fd, - off_t offset); - static bool InvokeMmapReplacementSlow(const void* start, - size_t size, - int protection, - int flags, - int fd, - off_t offset, - void** result); - static void InvokeMunmapHookSlow(const void* p, size_t size); - static bool InvokeMunmapReplacementSlow(const void* p, - size_t size, - int* result); - static void InvokeMremapHookSlow(const void* result, - const void* old_addr, - size_t old_size, - size_t new_size, - int flags, - const void* new_addr); - static void InvokePreSbrkHookSlow(ptrdiff_t increment); - static void InvokeSbrkHookSlow(const void* result, ptrdiff_t increment); -}; - -#ifdef _MSC_VER -#pragma warning(pop) -#endif - - -#endif /* _MALLOC_HOOK_H_ */ diff --git a/contrib/libtcmalloc/include/gperftools/malloc_hook_c.h b/contrib/libtcmalloc/include/gperftools/malloc_hook_c.h deleted file mode 100644 index 56337e15e83..00000000000 --- a/contrib/libtcmalloc/include/gperftools/malloc_hook_c.h +++ /dev/null @@ -1,173 +0,0 @@ -/* Copyright (c) 2008, Google Inc. - * All rights reserved. - * - * Redistribution and use in source and binary forms, with or without - * modification, are permitted provided that the following conditions are - * met: - * - * * Redistributions of source code must retain the above copyright - * notice, this list of conditions and the following disclaimer. - * * Redistributions in binary form must reproduce the above - * copyright notice, this list of conditions and the following disclaimer - * in the documentation and/or other materials provided with the - * distribution. - * * Neither the name of Google Inc. nor the names of its - * contributors may be used to endorse or promote products derived from - * this software without specific prior written permission. - * - * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS - * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT - * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR - * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT - * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, - * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT - * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, - * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY - * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT - * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE - * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. - * - * -- - * Author: Craig Silverstein - * - * C shims for the C++ malloc_hook.h. See malloc_hook.h for details - * on how to use these. - */ - -#ifndef _MALLOC_HOOK_C_H_ -#define _MALLOC_HOOK_C_H_ - -#include -#include - -/* Annoying stuff for windows; makes sure clients can import these functions */ -#ifndef PERFTOOLS_DLL_DECL -# ifdef _WIN32 -# define PERFTOOLS_DLL_DECL __declspec(dllimport) -# else -# define PERFTOOLS_DLL_DECL -# endif -#endif - -#ifdef __cplusplus -extern "C" { -#endif - -/* Get the current stack trace. Try to skip all routines up to and - * and including the caller of MallocHook::Invoke*. - * Use "skip_count" (similarly to GetStackTrace from stacktrace.h) - * as a hint about how many routines to skip if better information - * is not available. - */ -PERFTOOLS_DLL_DECL -int MallocHook_GetCallerStackTrace(void** result, int max_depth, - int skip_count); - -/* The MallocHook_{Add,Remove}*Hook functions return 1 on success and 0 on - * failure. - */ - -typedef void (*MallocHook_NewHook)(const void* ptr, size_t size); -PERFTOOLS_DLL_DECL -int MallocHook_AddNewHook(MallocHook_NewHook hook); -PERFTOOLS_DLL_DECL -int MallocHook_RemoveNewHook(MallocHook_NewHook hook); - -typedef void (*MallocHook_DeleteHook)(const void* ptr); -PERFTOOLS_DLL_DECL -int MallocHook_AddDeleteHook(MallocHook_DeleteHook hook); -PERFTOOLS_DLL_DECL -int MallocHook_RemoveDeleteHook(MallocHook_DeleteHook hook); - -typedef void (*MallocHook_PreMmapHook)(const void *start, - size_t size, - int protection, - int flags, - int fd, - off_t offset); -PERFTOOLS_DLL_DECL -int MallocHook_AddPreMmapHook(MallocHook_PreMmapHook hook); -PERFTOOLS_DLL_DECL -int MallocHook_RemovePreMmapHook(MallocHook_PreMmapHook hook); - -typedef void (*MallocHook_MmapHook)(const void* result, - const void* start, - size_t size, - int protection, - int flags, - int fd, - off_t offset); -PERFTOOLS_DLL_DECL -int MallocHook_AddMmapHook(MallocHook_MmapHook hook); -PERFTOOLS_DLL_DECL -int MallocHook_RemoveMmapHook(MallocHook_MmapHook hook); - -typedef int (*MallocHook_MmapReplacement)(const void* start, - size_t size, - int protection, - int flags, - int fd, - off_t offset, - void** result); -int MallocHook_SetMmapReplacement(MallocHook_MmapReplacement hook); -int MallocHook_RemoveMmapReplacement(MallocHook_MmapReplacement hook); - -typedef void (*MallocHook_MunmapHook)(const void* ptr, size_t size); -PERFTOOLS_DLL_DECL -int MallocHook_AddMunmapHook(MallocHook_MunmapHook hook); -PERFTOOLS_DLL_DECL -int MallocHook_RemoveMunmapHook(MallocHook_MunmapHook hook); - -typedef int (*MallocHook_MunmapReplacement)(const void* ptr, - size_t size, - int* result); -int MallocHook_SetMunmapReplacement(MallocHook_MunmapReplacement hook); -int MallocHook_RemoveMunmapReplacement(MallocHook_MunmapReplacement hook); - -typedef void (*MallocHook_MremapHook)(const void* result, - const void* old_addr, - size_t old_size, - size_t new_size, - int flags, - const void* new_addr); -PERFTOOLS_DLL_DECL -int MallocHook_AddMremapHook(MallocHook_MremapHook hook); -PERFTOOLS_DLL_DECL -int MallocHook_RemoveMremapHook(MallocHook_MremapHook hook); - -typedef void (*MallocHook_PreSbrkHook)(ptrdiff_t increment); -PERFTOOLS_DLL_DECL -int MallocHook_AddPreSbrkHook(MallocHook_PreSbrkHook hook); -PERFTOOLS_DLL_DECL -int MallocHook_RemovePreSbrkHook(MallocHook_PreSbrkHook hook); - -typedef void (*MallocHook_SbrkHook)(const void* result, ptrdiff_t increment); -PERFTOOLS_DLL_DECL -int MallocHook_AddSbrkHook(MallocHook_SbrkHook hook); -PERFTOOLS_DLL_DECL -int MallocHook_RemoveSbrkHook(MallocHook_SbrkHook hook); - -/* The following are DEPRECATED. */ -PERFTOOLS_DLL_DECL -MallocHook_NewHook MallocHook_SetNewHook(MallocHook_NewHook hook); -PERFTOOLS_DLL_DECL -MallocHook_DeleteHook MallocHook_SetDeleteHook(MallocHook_DeleteHook hook); -PERFTOOLS_DLL_DECL -MallocHook_PreMmapHook MallocHook_SetPreMmapHook(MallocHook_PreMmapHook hook); -PERFTOOLS_DLL_DECL -MallocHook_MmapHook MallocHook_SetMmapHook(MallocHook_MmapHook hook); -PERFTOOLS_DLL_DECL -MallocHook_MunmapHook MallocHook_SetMunmapHook(MallocHook_MunmapHook hook); -PERFTOOLS_DLL_DECL -MallocHook_MremapHook MallocHook_SetMremapHook(MallocHook_MremapHook hook); -PERFTOOLS_DLL_DECL -MallocHook_PreSbrkHook MallocHook_SetPreSbrkHook(MallocHook_PreSbrkHook hook); -PERFTOOLS_DLL_DECL -MallocHook_SbrkHook MallocHook_SetSbrkHook(MallocHook_SbrkHook hook); -/* End of DEPRECATED functions. */ - -#ifdef __cplusplus -} // extern "C" -#endif - -#endif /* _MALLOC_HOOK_C_H_ */ diff --git a/contrib/libtcmalloc/include/gperftools/profiler.h b/contrib/libtcmalloc/include/gperftools/profiler.h deleted file mode 100644 index 2d272d616a9..00000000000 --- a/contrib/libtcmalloc/include/gperftools/profiler.h +++ /dev/null @@ -1,169 +0,0 @@ -// -*- Mode: C++; c-basic-offset: 2; indent-tabs-mode: nil -*- -/* Copyright (c) 2005, Google Inc. - * All rights reserved. - * - * Redistribution and use in source and binary forms, with or without - * modification, are permitted provided that the following conditions are - * met: - * - * * Redistributions of source code must retain the above copyright - * notice, this list of conditions and the following disclaimer. - * * Redistributions in binary form must reproduce the above - * copyright notice, this list of conditions and the following disclaimer - * in the documentation and/or other materials provided with the - * distribution. - * * Neither the name of Google Inc. nor the names of its - * contributors may be used to endorse or promote products derived from - * this software without specific prior written permission. - * - * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS - * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT - * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR - * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT - * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, - * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT - * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, - * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY - * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT - * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE - * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. - * - * --- - * Author: Sanjay Ghemawat - * - * Module for CPU profiling based on periodic pc-sampling. - * - * For full(er) information, see doc/cpuprofile.html - * - * This module is linked into your program with - * no slowdown caused by this unless you activate the profiler - * using one of the following methods: - * - * 1. Before starting the program, set the environment variable - * "CPUPROFILE" to be the name of the file to which the profile - * data should be written. - * - * 2. Programmatically, start and stop the profiler using the - * routines "ProfilerStart(filename)" and "ProfilerStop()". - * - * - * (Note: if using linux 2.4 or earlier, only the main thread may be - * profiled.) - * - * Use pprof to view the resulting profile output. - * % pprof - * % pprof --gv - * - * These functions are thread-safe. - */ - -#ifndef BASE_PROFILER_H_ -#define BASE_PROFILER_H_ - -#include /* For time_t */ - -/* Annoying stuff for windows; makes sure clients can import these functions */ -#ifndef PERFTOOLS_DLL_DECL -# ifdef _WIN32 -# define PERFTOOLS_DLL_DECL __declspec(dllimport) -# else -# define PERFTOOLS_DLL_DECL -# endif -#endif - -/* All this code should be usable from within C apps. */ -#ifdef __cplusplus -extern "C" { -#endif - -/* Profiler options, for use with ProfilerStartWithOptions. To use: - * - * struct ProfilerOptions options; - * memset(&options, 0, sizeof options); - * - * then fill in fields as needed. - * - * This structure is intended to be usable from C code, so no constructor - * is provided to initialize it. (Use memset as described above). - */ -struct ProfilerOptions { - /* Filter function and argument. - * - * If filter_in_thread is not NULL, when a profiling tick is delivered - * the profiler will call: - * - * (*filter_in_thread)(filter_in_thread_arg) - * - * If it returns nonzero, the sample will be included in the profile. - * Note that filter_in_thread runs in a signal handler, so must be - * async-signal-safe. - * - * A typical use would be to set up filter results for each thread - * in the system before starting the profiler, then to make - * filter_in_thread be a very simple function which retrieves those - * results in an async-signal-safe way. Retrieval could be done - * using thread-specific data, or using a shared data structure that - * supports async-signal-safe lookups. - */ - int (*filter_in_thread)(void *arg); - void *filter_in_thread_arg; -}; - -/* Start profiling and write profile info into fname, discarding any - * existing profiling data in that file. - * - * This is equivalent to calling ProfilerStartWithOptions(fname, NULL). - */ -PERFTOOLS_DLL_DECL int ProfilerStart(const char* fname); - -/* Start profiling and write profile into fname, discarding any - * existing profiling data in that file. - * - * The profiler is configured using the options given by 'options'. - * Options which are not specified are given default values. - * - * 'options' may be NULL, in which case all are given default values. - * - * Returns nonzero if profiling was started successfully, or zero else. - */ -PERFTOOLS_DLL_DECL int ProfilerStartWithOptions( - const char *fname, const struct ProfilerOptions *options); - -/* Stop profiling. Can be started again with ProfilerStart(), but - * the currently accumulated profiling data will be cleared. - */ -PERFTOOLS_DLL_DECL void ProfilerStop(void); - -/* Flush any currently buffered profiling state to the profile file. - * Has no effect if the profiler has not been started. - */ -PERFTOOLS_DLL_DECL void ProfilerFlush(void); - - -/* DEPRECATED: these functions were used to enable/disable profiling - * in the current thread, but no longer do anything. - */ -PERFTOOLS_DLL_DECL void ProfilerEnable(void); -PERFTOOLS_DLL_DECL void ProfilerDisable(void); - -/* Returns nonzero if profile is currently enabled, zero if it's not. */ -PERFTOOLS_DLL_DECL int ProfilingIsEnabledForAllThreads(void); - -/* Routine for registering new threads with the profiler. - */ -PERFTOOLS_DLL_DECL void ProfilerRegisterThread(void); - -/* Stores state about profiler's current status into "*state". */ -struct ProfilerState { - int enabled; /* Is profiling currently enabled? */ - time_t start_time; /* If enabled, when was profiling started? */ - char profile_name[1024]; /* Name of profile file being written, or '\0' */ - int samples_gathered; /* Number of samples gathered so far (or 0) */ -}; -PERFTOOLS_DLL_DECL void ProfilerGetCurrentState(struct ProfilerState* state); - -#ifdef __cplusplus -} // extern "C" -#endif - -#endif /* BASE_PROFILER_H_ */ diff --git a/contrib/libtcmalloc/include/gperftools/stacktrace.h b/contrib/libtcmalloc/include/gperftools/stacktrace.h deleted file mode 100644 index 2b9c5a13209..00000000000 --- a/contrib/libtcmalloc/include/gperftools/stacktrace.h +++ /dev/null @@ -1,117 +0,0 @@ -// -*- Mode: C++; c-basic-offset: 2; indent-tabs-mode: nil -*- -// Copyright (c) 2005, Google Inc. -// All rights reserved. -// -// Redistribution and use in source and binary forms, with or without -// modification, are permitted provided that the following conditions are -// met: -// -// * Redistributions of source code must retain the above copyright -// notice, this list of conditions and the following disclaimer. -// * Redistributions in binary form must reproduce the above -// copyright notice, this list of conditions and the following disclaimer -// in the documentation and/or other materials provided with the -// distribution. -// * Neither the name of Google Inc. nor the names of its -// contributors may be used to endorse or promote products derived from -// this software without specific prior written permission. -// -// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS -// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT -// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR -// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT -// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, -// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT -// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, -// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY -// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT -// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE -// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. - -// --- -// Author: Sanjay Ghemawat -// -// Routines to extract the current stack trace. These functions are -// thread-safe. - -#ifndef GOOGLE_STACKTRACE_H_ -#define GOOGLE_STACKTRACE_H_ - -// Annoying stuff for windows -- makes sure clients can import these functions -#ifndef PERFTOOLS_DLL_DECL -# ifdef _WIN32 -# define PERFTOOLS_DLL_DECL __declspec(dllimport) -# else -# define PERFTOOLS_DLL_DECL -# endif -#endif - - -// Skips the most recent "skip_count" stack frames (also skips the -// frame generated for the "GetStackFrames" routine itself), and then -// records the pc values for up to the next "max_depth" frames in -// "result", and the corresponding stack frame sizes in "sizes". -// Returns the number of values recorded in "result"/"sizes". -// -// Example: -// main() { foo(); } -// foo() { bar(); } -// bar() { -// void* result[10]; -// int sizes[10]; -// int depth = GetStackFrames(result, sizes, 10, 1); -// } -// -// The GetStackFrames call will skip the frame for "bar". It will -// return 2 and will produce pc values that map to the following -// procedures: -// result[0] foo -// result[1] main -// (Actually, there may be a few more entries after "main" to account for -// startup procedures.) -// And corresponding stack frame sizes will also be recorded: -// sizes[0] 16 -// sizes[1] 16 -// (Stack frame sizes of 16 above are just for illustration purposes.) -// Stack frame sizes of 0 or less indicate that those frame sizes couldn't -// be identified. -// -// This routine may return fewer stack frame entries than are -// available. Also note that "result" and "sizes" must both be non-NULL. -extern PERFTOOLS_DLL_DECL int GetStackFrames(void** result, int* sizes, int max_depth, - int skip_count); - -// Same as above, but to be used from a signal handler. The "uc" parameter -// should be the pointer to ucontext_t which was passed as the 3rd parameter -// to sa_sigaction signal handler. It may help the unwinder to get a -// better stack trace under certain conditions. The "uc" may safely be NULL. -extern PERFTOOLS_DLL_DECL int GetStackFramesWithContext(void** result, int* sizes, int max_depth, - int skip_count, const void *uc); - -// This is similar to the GetStackFrames routine, except that it returns -// the stack trace only, and not the stack frame sizes as well. -// Example: -// main() { foo(); } -// foo() { bar(); } -// bar() { -// void* result[10]; -// int depth = GetStackTrace(result, 10, 1); -// } -// -// This produces: -// result[0] foo -// result[1] main -// .... ... -// -// "result" must not be NULL. -extern PERFTOOLS_DLL_DECL int GetStackTrace(void** result, int max_depth, - int skip_count); - -// Same as above, but to be used from a signal handler. The "uc" parameter -// should be the pointer to ucontext_t which was passed as the 3rd parameter -// to sa_sigaction signal handler. It may help the unwinder to get a -// better stack trace under certain conditions. The "uc" may safely be NULL. -extern PERFTOOLS_DLL_DECL int GetStackTraceWithContext(void** result, int max_depth, - int skip_count, const void *uc); - -#endif /* GOOGLE_STACKTRACE_H_ */ diff --git a/contrib/libtcmalloc/include/gperftools/tcmalloc.h b/contrib/libtcmalloc/include/gperftools/tcmalloc.h deleted file mode 100644 index a5b39dbffbe..00000000000 --- a/contrib/libtcmalloc/include/gperftools/tcmalloc.h +++ /dev/null @@ -1,160 +0,0 @@ -// -*- Mode: C; c-basic-offset: 2; indent-tabs-mode: nil -*- -/* Copyright (c) 2003, Google Inc. - * All rights reserved. - * - * Redistribution and use in source and binary forms, with or without - * modification, are permitted provided that the following conditions are - * met: - * - * * Redistributions of source code must retain the above copyright - * notice, this list of conditions and the following disclaimer. - * * Redistributions in binary form must reproduce the above - * copyright notice, this list of conditions and the following disclaimer - * in the documentation and/or other materials provided with the - * distribution. - * * Neither the name of Google Inc. nor the names of its - * contributors may be used to endorse or promote products derived from - * this software without specific prior written permission. - * - * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS - * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT - * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR - * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT - * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, - * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT - * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, - * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY - * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT - * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE - * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. - * - * --- - * Author: Sanjay Ghemawat - * .h file by Craig Silverstein - */ - -#ifndef TCMALLOC_TCMALLOC_H_ -#define TCMALLOC_TCMALLOC_H_ - -#include /* for size_t */ - -/* Define the version number so folks can check against it */ -#define TC_VERSION_MAJOR 2 -#define TC_VERSION_MINOR 5 -#define TC_VERSION_PATCH "" -#define TC_VERSION_STRING "gperftools 2.5" - -/* For struct mallinfo, if it's defined. */ -#if !defined(__APPLE__) && !defined(__FreeBSD__) -# include -#else -struct mallinfo { - size_t arena; /* non-mmapped space allocated from system */ - size_t ordblks; /* number of free chunks */ - size_t smblks; /* always 0 */ - size_t hblks; /* always 0 */ - size_t hblkhd; /* space in mmapped regions */ - size_t usmblks; /* maximum total allocated space */ - size_t fsmblks; /* always 0 */ - size_t uordblks; /* total allocated space */ - size_t fordblks; /* total free space */ - size_t keepcost; /* releasable (via malloc_trim) space */ -}; -#endif - -#ifdef __cplusplus -#define PERFTOOLS_THROW throw() -#else -# ifdef __GNUC__ -# define PERFTOOLS_THROW __attribute__((__nothrow__)) -# else -# define PERFTOOLS_THROW -# endif -#endif - -#ifndef PERFTOOLS_DLL_DECL -#define PERFTOOLS_DLL_DECL_DEFINED -# ifdef _WIN32 -# define PERFTOOLS_DLL_DECL __declspec(dllimport) -# else -# define PERFTOOLS_DLL_DECL -# endif -#endif - -#ifdef __cplusplus -namespace std { -struct nothrow_t; -} - -extern "C" { -#endif - /* - * Returns a human-readable version string. If major, minor, - * and/or patch are not NULL, they are set to the major version, - * minor version, and patch-code (a string, usually ""). - */ - PERFTOOLS_DLL_DECL const char* tc_version(int* major, int* minor, - const char** patch) PERFTOOLS_THROW; - - PERFTOOLS_DLL_DECL void* tc_malloc(size_t size) PERFTOOLS_THROW; - PERFTOOLS_DLL_DECL void* tc_malloc_skip_new_handler(size_t size) PERFTOOLS_THROW; - PERFTOOLS_DLL_DECL void tc_free(void* ptr) PERFTOOLS_THROW; - PERFTOOLS_DLL_DECL void tc_free_sized(void *ptr, size_t size) PERFTOOLS_THROW; - PERFTOOLS_DLL_DECL void* tc_realloc(void* ptr, size_t size) PERFTOOLS_THROW; - PERFTOOLS_DLL_DECL void* tc_calloc(size_t nmemb, size_t size) PERFTOOLS_THROW; - PERFTOOLS_DLL_DECL void tc_cfree(void* ptr) PERFTOOLS_THROW; - - PERFTOOLS_DLL_DECL void* tc_memalign(size_t __alignment, - size_t __size) PERFTOOLS_THROW; - PERFTOOLS_DLL_DECL int tc_posix_memalign(void** ptr, - size_t align, size_t size) PERFTOOLS_THROW; - PERFTOOLS_DLL_DECL void* tc_valloc(size_t __size) PERFTOOLS_THROW; - PERFTOOLS_DLL_DECL void* tc_pvalloc(size_t __size) PERFTOOLS_THROW; - - PERFTOOLS_DLL_DECL void tc_malloc_stats(void) PERFTOOLS_THROW; - PERFTOOLS_DLL_DECL int tc_mallopt(int cmd, int value) PERFTOOLS_THROW; -#if 1 - PERFTOOLS_DLL_DECL struct mallinfo tc_mallinfo(void) PERFTOOLS_THROW; -#endif - - /* - * This is an alias for MallocExtension::instance()->GetAllocatedSize(). - * It is equivalent to - * OS X: malloc_size() - * glibc: malloc_usable_size() - * Windows: _msize() - */ - PERFTOOLS_DLL_DECL size_t tc_malloc_size(void* ptr) PERFTOOLS_THROW; - -#ifdef __cplusplus - PERFTOOLS_DLL_DECL int tc_set_new_mode(int flag) PERFTOOLS_THROW; - PERFTOOLS_DLL_DECL void* tc_new(size_t size); - PERFTOOLS_DLL_DECL void* tc_new_nothrow(size_t size, - const std::nothrow_t&) PERFTOOLS_THROW; - PERFTOOLS_DLL_DECL void tc_delete(void* p) PERFTOOLS_THROW; - PERFTOOLS_DLL_DECL void tc_delete_sized(void* p, size_t size) throw(); - PERFTOOLS_DLL_DECL void tc_delete_nothrow(void* p, - const std::nothrow_t&) PERFTOOLS_THROW; - PERFTOOLS_DLL_DECL void* tc_newarray(size_t size); - PERFTOOLS_DLL_DECL void* tc_newarray_nothrow(size_t size, - const std::nothrow_t&) PERFTOOLS_THROW; - PERFTOOLS_DLL_DECL void tc_deletearray(void* p) PERFTOOLS_THROW; - PERFTOOLS_DLL_DECL void tc_deletearray_sized(void* p, size_t size) throw(); - PERFTOOLS_DLL_DECL void tc_deletearray_nothrow(void* p, - const std::nothrow_t&) PERFTOOLS_THROW; -} -#endif - -/* We're only un-defining those for public */ -#if !defined(GPERFTOOLS_CONFIG_H_) - -#undef PERFTOOLS_THROW - -#ifdef PERFTOOLS_DLL_DECL_DEFINED -#undef PERFTOOLS_DLL_DECL -#undef PERFTOOLS_DLL_DECL_DEFINED -#endif - -#endif /* GPERFTOOLS_CONFIG_H_ */ - -#endif /* #ifndef TCMALLOC_TCMALLOC_H_ */ diff --git a/contrib/libtcmalloc/src/addressmap-inl.h b/contrib/libtcmalloc/src/addressmap-inl.h deleted file mode 100644 index fd1dc5b6ffe..00000000000 --- a/contrib/libtcmalloc/src/addressmap-inl.h +++ /dev/null @@ -1,422 +0,0 @@ -// -*- Mode: C++; c-basic-offset: 2; indent-tabs-mode: nil -*- -// Copyright (c) 2005, Google Inc. -// All rights reserved. -// -// Redistribution and use in source and binary forms, with or without -// modification, are permitted provided that the following conditions are -// met: -// -// * Redistributions of source code must retain the above copyright -// notice, this list of conditions and the following disclaimer. -// * Redistributions in binary form must reproduce the above -// copyright notice, this list of conditions and the following disclaimer -// in the documentation and/or other materials provided with the -// distribution. -// * Neither the name of Google Inc. nor the names of its -// contributors may be used to endorse or promote products derived from -// this software without specific prior written permission. -// -// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS -// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT -// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR -// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT -// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, -// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT -// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, -// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY -// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT -// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE -// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. - -// --- -// Author: Sanjay Ghemawat -// -// A fast map from addresses to values. Assumes that addresses are -// clustered. The main use is intended to be for heap-profiling. -// May be too memory-hungry for other uses. -// -// We use a user-defined allocator/de-allocator so that we can use -// this data structure during heap-profiling. -// -// IMPLEMENTATION DETAIL: -// -// Some default definitions/parameters: -// * Block -- aligned 128-byte region of the address space -// * Cluster -- aligned 1-MB region of the address space -// * Block-ID -- block-number within a cluster -// * Cluster-ID -- Starting address of cluster divided by cluster size -// -// We use a three-level map to represent the state: -// 1. A hash-table maps from a cluster-ID to the data for that cluster. -// 2. For each non-empty cluster we keep an array indexed by -// block-ID tht points to the first entry in the linked-list -// for the block. -// 3. At the bottom, we keep a singly-linked list of all -// entries in a block (for non-empty blocks). -// -// hash table -// +-------------+ -// | id->cluster |---> ... -// | ... | -// | id->cluster |---> Cluster -// +-------------+ +-------+ Data for one block -// | nil | +------------------------------------+ -// | ----+---|->[addr/value]-->[addr/value]-->... | -// | nil | +------------------------------------+ -// | ----+--> ... -// | nil | -// | ... | -// +-------+ -// -// Note that we require zero-bytes of overhead for completely empty -// clusters. The minimum space requirement for a cluster is the size -// of the hash-table entry plus a pointer value for each block in -// the cluster. Empty blocks impose no extra space requirement. -// -// The cost of a lookup is: -// a. A hash-table lookup to find the cluster -// b. An array access in the cluster structure -// c. A traversal over the linked-list for a block - -#ifndef BASE_ADDRESSMAP_INL_H_ -#define BASE_ADDRESSMAP_INL_H_ - -#include "config.h" -#include -#include -#if defined HAVE_STDINT_H -#include // to get uint16_t (ISO naming madness) -#elif defined HAVE_INTTYPES_H -#include // another place uint16_t might be defined -#else -#include // our last best hope -#endif - -// This class is thread-unsafe -- that is, instances of this class can -// not be accessed concurrently by multiple threads -- because the -// callback function for Iterate() may mutate contained values. If the -// callback functions you pass do not mutate their Value* argument, -// AddressMap can be treated as thread-compatible -- that is, it's -// safe for multiple threads to call "const" methods on this class, -// but not safe for one thread to call const methods on this class -// while another thread is calling non-const methods on the class. -template -class AddressMap { - public: - typedef void* (*Allocator)(size_t size); - typedef void (*DeAllocator)(void* ptr); - typedef const void* Key; - - // Create an AddressMap that uses the specified allocator/deallocator. - // The allocator/deallocator should behave like malloc/free. - // For instance, the allocator does not need to return initialized memory. - AddressMap(Allocator alloc, DeAllocator dealloc); - ~AddressMap(); - - // If the map contains an entry for "key", return it. Else return NULL. - inline const Value* Find(Key key) const; - inline Value* FindMutable(Key key); - - // Insert into the map. Any old value associated - // with key is forgotten. - void Insert(Key key, Value value); - - // Remove any entry for key in the map. If an entry was found - // and removed, stores the associated value in "*removed_value" - // and returns true. Else returns false. - bool FindAndRemove(Key key, Value* removed_value); - - // Similar to Find but we assume that keys are addresses of non-overlapping - // memory ranges whose sizes are given by size_func. - // If the map contains a range into which "key" points - // (at its start or inside of it, but not at the end), - // return the address of the associated value - // and store its key in "*res_key". - // Else return NULL. - // max_size specifies largest range size possibly in existence now. - typedef size_t (*ValueSizeFunc)(const Value& v); - const Value* FindInside(ValueSizeFunc size_func, size_t max_size, - Key key, Key* res_key); - - // Iterate over the address map calling 'callback' - // for all stored key-value pairs and passing 'arg' to it. - // We don't use full Closure/Callback machinery not to add - // unnecessary dependencies to this class with low-level uses. - template - inline void Iterate(void (*callback)(Key, Value*, Type), Type arg) const; - - private: - typedef uintptr_t Number; - - // The implementation assumes that addresses inserted into the map - // will be clustered. We take advantage of this fact by splitting - // up the address-space into blocks and using a linked-list entry - // for each block. - - // Size of each block. There is one linked-list for each block, so - // do not make the block-size too big. Oterwise, a lot of time - // will be spent traversing linked lists. - static const int kBlockBits = 7; - static const int kBlockSize = 1 << kBlockBits; - - // Entry kept in per-block linked-list - struct Entry { - Entry* next; - Key key; - Value value; - }; - - // We further group a sequence of consecutive blocks into a cluster. - // The data for a cluster is represented as a dense array of - // linked-lists, one list per contained block. - static const int kClusterBits = 13; - static const Number kClusterSize = 1 << (kBlockBits + kClusterBits); - static const int kClusterBlocks = 1 << kClusterBits; - - // We use a simple chaining hash-table to represent the clusters. - struct Cluster { - Cluster* next; // Next cluster in hash table chain - Number id; // Cluster ID - Entry* blocks[kClusterBlocks]; // Per-block linked-lists - }; - - // Number of hash-table entries. With the block-size/cluster-size - // defined above, each cluster covers 1 MB, so an 4K entry - // hash-table will give an average hash-chain length of 1 for 4GB of - // in-use memory. - static const int kHashBits = 12; - static const int kHashSize = 1 << 12; - - // Number of entry objects allocated at a time - static const int ALLOC_COUNT = 64; - - Cluster** hashtable_; // The hash-table - Entry* free_; // Free list of unused Entry objects - - // Multiplicative hash function: - // The value "kHashMultiplier" is the bottom 32 bits of - // int((sqrt(5)-1)/2 * 2^32) - // This is a good multiplier as suggested in CLR, Knuth. The hash - // value is taken to be the top "k" bits of the bottom 32 bits - // of the muliplied value. - static const uint32_t kHashMultiplier = 2654435769u; - static int HashInt(Number x) { - // Multiply by a constant and take the top bits of the result. - const uint32_t m = static_cast(x) * kHashMultiplier; - return static_cast(m >> (32 - kHashBits)); - } - - // Find cluster object for specified address. If not found - // and "create" is true, create the object. If not found - // and "create" is false, return NULL. - // - // This method is bitwise-const if create is false. - Cluster* FindCluster(Number address, bool create) { - // Look in hashtable - const Number cluster_id = address >> (kBlockBits + kClusterBits); - const int h = HashInt(cluster_id); - for (Cluster* c = hashtable_[h]; c != NULL; c = c->next) { - if (c->id == cluster_id) { - return c; - } - } - - // Create cluster if necessary - if (create) { - Cluster* c = New(1); - c->id = cluster_id; - c->next = hashtable_[h]; - hashtable_[h] = c; - return c; - } - return NULL; - } - - // Return the block ID for an address within its cluster - static int BlockID(Number address) { - return (address >> kBlockBits) & (kClusterBlocks - 1); - } - - //-------------------------------------------------------------- - // Memory management -- we keep all objects we allocate linked - // together in a singly linked list so we can get rid of them - // when we are all done. Furthermore, we allow the client to - // pass in custom memory allocator/deallocator routines. - //-------------------------------------------------------------- - struct Object { - Object* next; - // The real data starts here - }; - - Allocator alloc_; // The allocator - DeAllocator dealloc_; // The deallocator - Object* allocated_; // List of allocated objects - - // Allocates a zeroed array of T with length "num". Also inserts - // the allocated block into a linked list so it can be deallocated - // when we are all done. - template T* New(int num) { - void* ptr = (*alloc_)(sizeof(Object) + num*sizeof(T)); - memset(ptr, 0, sizeof(Object) + num*sizeof(T)); - Object* obj = reinterpret_cast(ptr); - obj->next = allocated_; - allocated_ = obj; - return reinterpret_cast(reinterpret_cast(ptr) + 1); - } -}; - -// More implementation details follow: - -template -AddressMap::AddressMap(Allocator alloc, DeAllocator dealloc) - : free_(NULL), - alloc_(alloc), - dealloc_(dealloc), - allocated_(NULL) { - hashtable_ = New(kHashSize); -} - -template -AddressMap::~AddressMap() { - // De-allocate all of the objects we allocated - for (Object* obj = allocated_; obj != NULL; /**/) { - Object* next = obj->next; - (*dealloc_)(obj); - obj = next; - } -} - -template -inline const Value* AddressMap::Find(Key key) const { - return const_cast(this)->FindMutable(key); -} - -template -inline Value* AddressMap::FindMutable(Key key) { - const Number num = reinterpret_cast(key); - const Cluster* const c = FindCluster(num, false/*do not create*/); - if (c != NULL) { - for (Entry* e = c->blocks[BlockID(num)]; e != NULL; e = e->next) { - if (e->key == key) { - return &e->value; - } - } - } - return NULL; -} - -template -void AddressMap::Insert(Key key, Value value) { - const Number num = reinterpret_cast(key); - Cluster* const c = FindCluster(num, true/*create*/); - - // Look in linked-list for this block - const int block = BlockID(num); - for (Entry* e = c->blocks[block]; e != NULL; e = e->next) { - if (e->key == key) { - e->value = value; - return; - } - } - - // Create entry - if (free_ == NULL) { - // Allocate a new batch of entries and add to free-list - Entry* array = New(ALLOC_COUNT); - for (int i = 0; i < ALLOC_COUNT-1; i++) { - array[i].next = &array[i+1]; - } - array[ALLOC_COUNT-1].next = free_; - free_ = &array[0]; - } - Entry* e = free_; - free_ = e->next; - e->key = key; - e->value = value; - e->next = c->blocks[block]; - c->blocks[block] = e; -} - -template -bool AddressMap::FindAndRemove(Key key, Value* removed_value) { - const Number num = reinterpret_cast(key); - Cluster* const c = FindCluster(num, false/*do not create*/); - if (c != NULL) { - for (Entry** p = &c->blocks[BlockID(num)]; *p != NULL; p = &(*p)->next) { - Entry* e = *p; - if (e->key == key) { - *removed_value = e->value; - *p = e->next; // Remove e from linked-list - e->next = free_; // Add e to free-list - free_ = e; - return true; - } - } - } - return false; -} - -template -const Value* AddressMap::FindInside(ValueSizeFunc size_func, - size_t max_size, - Key key, - Key* res_key) { - const Number key_num = reinterpret_cast(key); - Number num = key_num; // we'll move this to move back through the clusters - while (1) { - const Cluster* c = FindCluster(num, false/*do not create*/); - if (c != NULL) { - while (1) { - const int block = BlockID(num); - bool had_smaller_key = false; - for (const Entry* e = c->blocks[block]; e != NULL; e = e->next) { - const Number e_num = reinterpret_cast(e->key); - if (e_num <= key_num) { - if (e_num == key_num || // to handle 0-sized ranges - key_num < e_num + (*size_func)(e->value)) { - *res_key = e->key; - return &e->value; - } - had_smaller_key = true; - } - } - if (had_smaller_key) return NULL; // got a range before 'key' - // and it did not contain 'key' - if (block == 0) break; - // try address-wise previous block - num |= kBlockSize - 1; // start at the last addr of prev block - num -= kBlockSize; - if (key_num - num > max_size) return NULL; - } - } - if (num < kClusterSize) return NULL; // first cluster - // go to address-wise previous cluster to try - num |= kClusterSize - 1; // start at the last block of previous cluster - num -= kClusterSize; - if (key_num - num > max_size) return NULL; - // Having max_size to limit the search is crucial: else - // we have to traverse a lot of empty clusters (or blocks). - // We can avoid needing max_size if we put clusters into - // a search tree, but performance suffers considerably - // if we use this approach by using stl::set. - } -} - -template -template -inline void AddressMap::Iterate(void (*callback)(Key, Value*, Type), - Type arg) const { - // We could optimize this by traversing only non-empty clusters and/or blocks - // but it does not speed up heap-checker noticeably. - for (int h = 0; h < kHashSize; ++h) { - for (const Cluster* c = hashtable_[h]; c != NULL; c = c->next) { - for (int b = 0; b < kClusterBlocks; ++b) { - for (Entry* e = c->blocks[b]; e != NULL; e = e->next) { - callback(e->key, &e->value, arg); - } - } - } - } -} - -#endif // BASE_ADDRESSMAP_INL_H_ diff --git a/contrib/libtcmalloc/src/base/arm_instruction_set_select.h b/contrib/libtcmalloc/src/base/arm_instruction_set_select.h deleted file mode 100644 index 6fde685272c..00000000000 --- a/contrib/libtcmalloc/src/base/arm_instruction_set_select.h +++ /dev/null @@ -1,84 +0,0 @@ -// Copyright (c) 2011, Google Inc. -// All rights reserved. -// -// Redistribution and use in source and binary forms, with or without -// modification, are permitted provided that the following conditions are -// met: -// -// * Redistributions of source code must retain the above copyright -// notice, this list of conditions and the following disclaimer. -// * Redistributions in binary form must reproduce the above -// copyright notice, this list of conditions and the following disclaimer -// in the documentation and/or other materials provided with the -// distribution. -// * Neither the name of Google Inc. nor the names of its -// contributors may be used to endorse or promote products derived from -// this software without specific prior written permission. -// -// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS -// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT -// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR -// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT -// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, -// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT -// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, -// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY -// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT -// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE -// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. -// -// Author: Alexander Levitskiy -// -// Generalizes the plethora of ARM flavors available to an easier to manage set -// Defs reference is at https://wiki.edubuntu.org/ARM/Thumb2PortingHowto - -#ifndef ARM_INSTRUCTION_SET_SELECT_H_ -#define ARM_INSTRUCTION_SET_SELECT_H_ - -#if defined(__ARM_ARCH_8A__) -# define ARMV8 1 -#endif - -#if defined(ARMV8) || \ - defined(__ARM_ARCH_7__) || \ - defined(__ARM_ARCH_7R__) || \ - defined(__ARM_ARCH_7A__) -# define ARMV7 1 -#endif - -#if defined(ARMV7) || \ - defined(__ARM_ARCH_6__) || \ - defined(__ARM_ARCH_6J__) || \ - defined(__ARM_ARCH_6K__) || \ - defined(__ARM_ARCH_6Z__) || \ - defined(__ARM_ARCH_6T2__) || \ - defined(__ARM_ARCH_6ZK__) -# define ARMV6 1 -#endif - -#if defined(ARMV6) || \ - defined(__ARM_ARCH_5T__) || \ - defined(__ARM_ARCH_5E__) || \ - defined(__ARM_ARCH_5TE__) || \ - defined(__ARM_ARCH_5TEJ__) -# define ARMV5 1 -#endif - -#if defined(ARMV5) || \ - defined(__ARM_ARCH_4__) || \ - defined(__ARM_ARCH_4T__) -# define ARMV4 1 -#endif - -#if defined(ARMV4) || \ - defined(__ARM_ARCH_3__) || \ - defined(__ARM_ARCH_3M__) -# define ARMV3 1 -#endif - -#if defined(ARMV3) || \ - defined(__ARM_ARCH_2__) -# define ARMV2 1 -#endif - -#endif // ARM_INSTRUCTION_SET_SELECT_H_ diff --git a/contrib/libtcmalloc/src/base/atomicops-internals-arm-generic.h b/contrib/libtcmalloc/src/base/atomicops-internals-arm-generic.h deleted file mode 100644 index d0f941309bb..00000000000 --- a/contrib/libtcmalloc/src/base/atomicops-internals-arm-generic.h +++ /dev/null @@ -1,228 +0,0 @@ -// -*- Mode: C++; c-basic-offset: 2; indent-tabs-mode: nil -*- -// Copyright (c) 2003, Google Inc. -// All rights reserved. -// -// Redistribution and use in source and binary forms, with or without -// modification, are permitted provided that the following conditions are -// met: -// -// * Redistributions of source code must retain the above copyright -// notice, this list of conditions and the following disclaimer. -// * Redistributions in binary form must reproduce the above -// copyright notice, this list of conditions and the following disclaimer -// in the documentation and/or other materials provided with the -// distribution. -// * Neither the name of Google Inc. nor the names of its -// contributors may be used to endorse or promote products derived from -// this software without specific prior written permission. -// -// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS -// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT -// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR -// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT -// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, -// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT -// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, -// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY -// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT -// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE -// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. -// --- -// -// Author: Lei Zhang, Sasha Levitskiy -// -// This file is an internal atomic implementation, use base/atomicops.h instead. -// -// LinuxKernelCmpxchg is from Google Gears. - -#ifndef BASE_ATOMICOPS_INTERNALS_ARM_GENERIC_H_ -#define BASE_ATOMICOPS_INTERNALS_ARM_GENERIC_H_ - -#include -#include -#include "base/basictypes.h" - -typedef int32_t Atomic32; - -namespace base { -namespace subtle { - -typedef int64_t Atomic64; - -// 0xffff0fc0 is the hard coded address of a function provided by -// the kernel which implements an atomic compare-exchange. On older -// ARM architecture revisions (pre-v6) this may be implemented using -// a syscall. This address is stable, and in active use (hard coded) -// by at least glibc-2.7 and the Android C library. -// pLinuxKernelCmpxchg has both acquire and release barrier sematincs. -typedef Atomic32 (*LinuxKernelCmpxchgFunc)(Atomic32 old_value, - Atomic32 new_value, - volatile Atomic32* ptr); -LinuxKernelCmpxchgFunc pLinuxKernelCmpxchg ATTRIBUTE_WEAK = - (LinuxKernelCmpxchgFunc) 0xffff0fc0; - -typedef void (*LinuxKernelMemoryBarrierFunc)(void); -LinuxKernelMemoryBarrierFunc pLinuxKernelMemoryBarrier ATTRIBUTE_WEAK = - (LinuxKernelMemoryBarrierFunc) 0xffff0fa0; - - -inline Atomic32 NoBarrier_CompareAndSwap(volatile Atomic32* ptr, - Atomic32 old_value, - Atomic32 new_value) { - Atomic32 prev_value = *ptr; - do { - if (!pLinuxKernelCmpxchg(old_value, new_value, - const_cast(ptr))) { - return old_value; - } - prev_value = *ptr; - } while (prev_value == old_value); - return prev_value; -} - -inline Atomic32 NoBarrier_AtomicExchange(volatile Atomic32* ptr, - Atomic32 new_value) { - Atomic32 old_value; - do { - old_value = *ptr; - } while (pLinuxKernelCmpxchg(old_value, new_value, - const_cast(ptr))); - return old_value; -} - -inline Atomic32 Acquire_AtomicExchange(volatile Atomic32* ptr, - Atomic32 new_value) { - // pLinuxKernelCmpxchg already has acquire and release barrier semantics. - return NoBarrier_AtomicExchange(ptr, new_value); -} - -inline Atomic32 Release_AtomicExchange(volatile Atomic32* ptr, - Atomic32 new_value) { - // pLinuxKernelCmpxchg already has acquire and release barrier semantics. - return NoBarrier_AtomicExchange(ptr, new_value); -} - -inline Atomic32 Acquire_CompareAndSwap(volatile Atomic32* ptr, - Atomic32 old_value, - Atomic32 new_value) { - return NoBarrier_CompareAndSwap(ptr, old_value, new_value); -} - -inline Atomic32 Release_CompareAndSwap(volatile Atomic32* ptr, - Atomic32 old_value, - Atomic32 new_value) { - return NoBarrier_CompareAndSwap(ptr, old_value, new_value); -} - -inline void NoBarrier_Store(volatile Atomic32* ptr, Atomic32 value) { - *ptr = value; -} - -inline void MemoryBarrier() { - pLinuxKernelMemoryBarrier(); -} - -inline void Acquire_Store(volatile Atomic32* ptr, Atomic32 value) { - *ptr = value; - MemoryBarrier(); -} - -inline void Release_Store(volatile Atomic32* ptr, Atomic32 value) { - MemoryBarrier(); - *ptr = value; -} - -inline Atomic32 NoBarrier_Load(volatile const Atomic32* ptr) { - return *ptr; -} - -inline Atomic32 Acquire_Load(volatile const Atomic32* ptr) { - Atomic32 value = *ptr; - MemoryBarrier(); - return value; -} - -inline Atomic32 Release_Load(volatile const Atomic32* ptr) { - MemoryBarrier(); - return *ptr; -} - - -// 64-bit versions are not implemented yet. - -inline void NotImplementedFatalError(const char *function_name) { - fprintf(stderr, "64-bit %s() not implemented on this platform\n", - function_name); - abort(); -} - -inline Atomic64 NoBarrier_CompareAndSwap(volatile Atomic64* ptr, - Atomic64 old_value, - Atomic64 new_value) { - NotImplementedFatalError("NoBarrier_CompareAndSwap"); - return 0; -} - -inline Atomic64 NoBarrier_AtomicExchange(volatile Atomic64* ptr, - Atomic64 new_value) { - NotImplementedFatalError("NoBarrier_AtomicExchange"); - return 0; -} - -inline Atomic64 Acquire_AtomicExchange(volatile Atomic64* ptr, - Atomic64 new_value) { - // pLinuxKernelCmpxchg already has acquire and release barrier semantics. - return NoBarrier_AtomicExchange(ptr, new_value); -} - -inline Atomic64 Release_AtomicExchange(volatile Atomic64* ptr, - Atomic64 new_value) { - // pLinuxKernelCmpxchg already has acquire and release barrier semantics. - return NoBarrier_AtomicExchange(ptr, new_value); -} - -inline void NoBarrier_Store(volatile Atomic64* ptr, Atomic64 value) { - NotImplementedFatalError("NoBarrier_Store"); -} - -inline void Acquire_Store(volatile Atomic64* ptr, Atomic64 value) { - NotImplementedFatalError("Acquire_Store64"); -} - -inline void Release_Store(volatile Atomic64* ptr, Atomic64 value) { - NotImplementedFatalError("Release_Store"); -} - -inline Atomic64 NoBarrier_Load(volatile const Atomic64* ptr) { - NotImplementedFatalError("NoBarrier_Load"); - return 0; -} - -inline Atomic64 Acquire_Load(volatile const Atomic64* ptr) { - NotImplementedFatalError("Atomic64 Acquire_Load"); - return 0; -} - -inline Atomic64 Release_Load(volatile const Atomic64* ptr) { - NotImplementedFatalError("Atomic64 Release_Load"); - return 0; -} - -inline Atomic64 Acquire_CompareAndSwap(volatile Atomic64* ptr, - Atomic64 old_value, - Atomic64 new_value) { - NotImplementedFatalError("Atomic64 Acquire_CompareAndSwap"); - return 0; -} - -inline Atomic64 Release_CompareAndSwap(volatile Atomic64* ptr, - Atomic64 old_value, - Atomic64 new_value) { - NotImplementedFatalError("Atomic64 Release_CompareAndSwap"); - return 0; -} - -} // namespace base::subtle -} // namespace base - -#endif // BASE_ATOMICOPS_INTERNALS_ARM_GENERIC_H_ diff --git a/contrib/libtcmalloc/src/base/atomicops-internals-arm-v6plus.h b/contrib/libtcmalloc/src/base/atomicops-internals-arm-v6plus.h deleted file mode 100644 index 35f10481b04..00000000000 --- a/contrib/libtcmalloc/src/base/atomicops-internals-arm-v6plus.h +++ /dev/null @@ -1,330 +0,0 @@ -// -*- Mode: C++; c-basic-offset: 2; indent-tabs-mode: nil -*- -// Copyright (c) 2011, Google Inc. -// All rights reserved. -// -// Redistribution and use in source and binary forms, with or without -// modification, are permitted provided that the following conditions are -// met: -// -// * Redistributions of source code must retain the above copyright -// notice, this list of conditions and the following disclaimer. -// * Redistributions in binary form must reproduce the above -// copyright notice, this list of conditions and the following disclaimer -// in the documentation and/or other materials provided with the -// distribution. -// * Neither the name of Google Inc. nor the names of its -// contributors may be used to endorse or promote products derived from -// this software without specific prior written permission. -// -// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS -// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT -// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR -// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT -// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, -// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT -// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, -// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY -// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT -// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE -// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. -// --- -// -// Author: Sasha Levitskiy -// based on atomicops-internals by Sanjay Ghemawat -// -// This file is an internal atomic implementation, use base/atomicops.h instead. -// -// This code implements ARM atomics for architectures V6 and newer. - -#ifndef BASE_ATOMICOPS_INTERNALS_ARM_V6PLUS_H_ -#define BASE_ATOMICOPS_INTERNALS_ARM_V6PLUS_H_ - -#include -#include -#include "base/basictypes.h" // For COMPILE_ASSERT - -// The LDREXD and STREXD instructions in ARM all v7 variants or above. In v6, -// only some variants support it. For simplicity, we only use exclusive -// 64-bit load/store in V7 or above. -#if defined(ARMV7) -# define BASE_ATOMICOPS_HAS_LDREXD_AND_STREXD -#endif - -typedef int32_t Atomic32; - -namespace base { -namespace subtle { - -typedef int64_t Atomic64; - -// 32-bit low-level ops - -inline Atomic32 NoBarrier_CompareAndSwap(volatile Atomic32* ptr, - Atomic32 old_value, - Atomic32 new_value) { - Atomic32 oldval, res; - do { - __asm__ __volatile__( - "ldrex %1, [%3]\n" - "mov %0, #0\n" - "teq %1, %4\n" - // The following IT (if-then) instruction is needed for the subsequent - // conditional instruction STREXEQ when compiling in THUMB mode. - // In ARM mode, the compiler/assembler will not generate any code for it. - "it eq\n" - "strexeq %0, %5, [%3]\n" - : "=&r" (res), "=&r" (oldval), "+Qo" (*ptr) - : "r" (ptr), "Ir" (old_value), "r" (new_value) - : "cc"); - } while (res); - return oldval; -} - -inline Atomic32 NoBarrier_AtomicExchange(volatile Atomic32* ptr, - Atomic32 new_value) { - Atomic32 tmp, old; - __asm__ __volatile__( - "1:\n" - "ldrex %1, [%2]\n" - "strex %0, %3, [%2]\n" - "teq %0, #0\n" - "bne 1b" - : "=&r" (tmp), "=&r" (old) - : "r" (ptr), "r" (new_value) - : "cc", "memory"); - return old; -} - -inline void MemoryBarrier() { -#if !defined(ARMV7) - uint32_t dest = 0; - __asm__ __volatile__("mcr p15,0,%0,c7,c10,5" :"=&r"(dest) : : "memory"); -#else - __asm__ __volatile__("dmb" : : : "memory"); -#endif -} - -inline Atomic32 Acquire_AtomicExchange(volatile Atomic32* ptr, - Atomic32 new_value) { - Atomic32 old_value = NoBarrier_AtomicExchange(ptr, new_value); - MemoryBarrier(); - return old_value; -} - -inline Atomic32 Release_AtomicExchange(volatile Atomic32* ptr, - Atomic32 new_value) { - MemoryBarrier(); - return NoBarrier_AtomicExchange(ptr, new_value); -} - -inline Atomic32 Acquire_CompareAndSwap(volatile Atomic32* ptr, - Atomic32 old_value, - Atomic32 new_value) { - Atomic32 value = NoBarrier_CompareAndSwap(ptr, old_value, new_value); - MemoryBarrier(); - return value; -} - -inline Atomic32 Release_CompareAndSwap(volatile Atomic32* ptr, - Atomic32 old_value, - Atomic32 new_value) { - MemoryBarrier(); - return NoBarrier_CompareAndSwap(ptr, old_value, new_value); -} - -inline void NoBarrier_Store(volatile Atomic32* ptr, Atomic32 value) { - *ptr = value; -} - -inline void Acquire_Store(volatile Atomic32* ptr, Atomic32 value) { - *ptr = value; - MemoryBarrier(); -} - -inline void Release_Store(volatile Atomic32* ptr, Atomic32 value) { - MemoryBarrier(); - *ptr = value; -} - -inline Atomic32 NoBarrier_Load(volatile const Atomic32* ptr) { - return *ptr; -} - -inline Atomic32 Acquire_Load(volatile const Atomic32* ptr) { - Atomic32 value = *ptr; - MemoryBarrier(); - return value; -} - -inline Atomic32 Release_Load(volatile const Atomic32* ptr) { - MemoryBarrier(); - return *ptr; -} - -// 64-bit versions are only available if LDREXD and STREXD instructions -// are available. -#ifdef BASE_ATOMICOPS_HAS_LDREXD_AND_STREXD - -#define BASE_HAS_ATOMIC64 1 - -inline Atomic64 NoBarrier_CompareAndSwap(volatile Atomic64* ptr, - Atomic64 old_value, - Atomic64 new_value) { - Atomic64 oldval, res; - do { - __asm__ __volatile__( - "ldrexd %1, [%3]\n" - "mov %0, #0\n" - "teq %Q1, %Q4\n" - // The following IT (if-then) instructions are needed for the subsequent - // conditional instructions when compiling in THUMB mode. - // In ARM mode, the compiler/assembler will not generate any code for it. - "it eq\n" - "teqeq %R1, %R4\n" - "it eq\n" - "strexdeq %0, %5, [%3]\n" - : "=&r" (res), "=&r" (oldval), "+Q" (*ptr) - : "r" (ptr), "Ir" (old_value), "r" (new_value) - : "cc"); - } while (res); - return oldval; -} - -inline Atomic64 NoBarrier_AtomicExchange(volatile Atomic64* ptr, - Atomic64 new_value) { - int store_failed; - Atomic64 old; - __asm__ __volatile__( - "1:\n" - "ldrexd %1, [%2]\n" - "strexd %0, %3, [%2]\n" - "teq %0, #0\n" - "bne 1b" - : "=&r" (store_failed), "=&r" (old) - : "r" (ptr), "r" (new_value) - : "cc", "memory"); - return old; -} - -inline Atomic64 Acquire_AtomicExchange(volatile Atomic64* ptr, - Atomic64 new_value) { - Atomic64 old_value = NoBarrier_AtomicExchange(ptr, new_value); - MemoryBarrier(); - return old_value; -} - -inline Atomic64 Release_AtomicExchange(volatile Atomic64* ptr, - Atomic64 new_value) { - MemoryBarrier(); - return NoBarrier_AtomicExchange(ptr, new_value); -} - -inline void NoBarrier_Store(volatile Atomic64* ptr, Atomic64 value) { - int store_failed; - Atomic64 dummy; - __asm__ __volatile__( - "1:\n" - // Dummy load to lock cache line. - "ldrexd %1, [%3]\n" - "strexd %0, %2, [%3]\n" - "teq %0, #0\n" - "bne 1b" - : "=&r" (store_failed), "=&r"(dummy) - : "r"(value), "r" (ptr) - : "cc", "memory"); -} - -inline Atomic64 NoBarrier_Load(volatile const Atomic64* ptr) { - Atomic64 res; - __asm__ __volatile__( - "ldrexd %0, [%1]\n" - "clrex\n" - : "=r" (res) - : "r"(ptr), "Q"(*ptr)); - return res; -} - -#else // BASE_ATOMICOPS_HAS_LDREXD_AND_STREXD - -inline void NotImplementedFatalError(const char *function_name) { - fprintf(stderr, "64-bit %s() not implemented on this platform\n", - function_name); - abort(); -} - -inline Atomic64 NoBarrier_CompareAndSwap(volatile Atomic64* ptr, - Atomic64 old_value, - Atomic64 new_value) { - NotImplementedFatalError("NoBarrier_CompareAndSwap"); - return 0; -} - -inline Atomic64 NoBarrier_AtomicExchange(volatile Atomic64* ptr, - Atomic64 new_value) { - NotImplementedFatalError("NoBarrier_AtomicExchange"); - return 0; -} - -inline Atomic64 Acquire_AtomicExchange(volatile Atomic64* ptr, - Atomic64 new_value) { - NotImplementedFatalError("Acquire_AtomicExchange"); - return 0; -} - -inline Atomic64 Release_AtomicExchange(volatile Atomic64* ptr, - Atomic64 new_value) { - NotImplementedFatalError("Release_AtomicExchange"); - return 0; -} - -inline void NoBarrier_Store(volatile Atomic64* ptr, Atomic64 value) { - NotImplementedFatalError("NoBarrier_Store"); -} - -inline Atomic64 NoBarrier_Load(volatile const Atomic64* ptr) { - NotImplementedFatalError("NoBarrier_Load"); - return 0; -} - -#endif // BASE_ATOMICOPS_HAS_LDREXD_AND_STREXD - -inline void Acquire_Store(volatile Atomic64* ptr, Atomic64 value) { - NoBarrier_Store(ptr, value); - MemoryBarrier(); -} - -inline void Release_Store(volatile Atomic64* ptr, Atomic64 value) { - MemoryBarrier(); - NoBarrier_Store(ptr, value); -} - -inline Atomic64 Acquire_Load(volatile const Atomic64* ptr) { - Atomic64 value = NoBarrier_Load(ptr); - MemoryBarrier(); - return value; -} - -inline Atomic64 Release_Load(volatile const Atomic64* ptr) { - MemoryBarrier(); - return NoBarrier_Load(ptr); -} - -inline Atomic64 Acquire_CompareAndSwap(volatile Atomic64* ptr, - Atomic64 old_value, - Atomic64 new_value) { - Atomic64 value = NoBarrier_CompareAndSwap(ptr, old_value, new_value); - MemoryBarrier(); - return value; -} - -inline Atomic64 Release_CompareAndSwap(volatile Atomic64* ptr, - Atomic64 old_value, - Atomic64 new_value) { - MemoryBarrier(); - return NoBarrier_CompareAndSwap(ptr, old_value, new_value); -} - -} // namespace subtle ends -} // namespace base ends - -#endif // BASE_ATOMICOPS_INTERNALS_ARM_V6PLUS_H_ diff --git a/contrib/libtcmalloc/src/base/atomicops-internals-gcc.h b/contrib/libtcmalloc/src/base/atomicops-internals-gcc.h deleted file mode 100644 index f8d27863cb7..00000000000 --- a/contrib/libtcmalloc/src/base/atomicops-internals-gcc.h +++ /dev/null @@ -1,203 +0,0 @@ -// -*- Mode: C++; c-basic-offset: 2; indent-tabs-mode: nil -*- -// Copyright (c) 2014, Linaro -// All rights reserved. -// -// Redistribution and use in source and binary forms, with or without -// modification, are permitted provided that the following conditions are -// met: -// -// * Redistributions of source code must retain the above copyright -// notice, this list of conditions and the following disclaimer. -// * Redistributions in binary form must reproduce the above -// copyright notice, this list of conditions and the following disclaimer -// in the documentation and/or other materials provided with the -// distribution. -// * Neither the name of Google Inc. nor the names of its -// contributors may be used to endorse or promote products derived from -// this software without specific prior written permission. -// -// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS -// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT -// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR -// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT -// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, -// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT -// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, -// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY -// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT -// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE -// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. -// --- -// -// Author: Riku Voipio, riku.voipio@linaro.org -// -// atomic primitives implemented with gcc atomic intrinsics: -// http://gcc.gnu.org/onlinedocs/gcc/_005f_005fatomic-Builtins.html -// - -#ifndef BASE_ATOMICOPS_INTERNALS_GCC_GENERIC_H_ -#define BASE_ATOMICOPS_INTERNALS_GCC_GENERIC_H_ - -#include -#include -#include "base/basictypes.h" - -typedef int32_t Atomic32; - -namespace base { -namespace subtle { - -typedef int64_t Atomic64; - -inline void MemoryBarrier() { - __sync_synchronize(); -} - -inline Atomic32 NoBarrier_CompareAndSwap(volatile Atomic32* ptr, - Atomic32 old_value, - Atomic32 new_value) { - Atomic32 prev_value = old_value; - __atomic_compare_exchange_n(ptr, &prev_value, new_value, - 0, __ATOMIC_RELAXED, __ATOMIC_RELAXED); - return prev_value; -} - -inline Atomic32 NoBarrier_AtomicExchange(volatile Atomic32* ptr, - Atomic32 new_value) { - return __atomic_exchange_n(const_cast(ptr), new_value, __ATOMIC_RELAXED); -} - -inline Atomic32 Acquire_AtomicExchange(volatile Atomic32* ptr, - Atomic32 new_value) { - return __atomic_exchange_n(const_cast(ptr), new_value, __ATOMIC_ACQUIRE); -} - -inline Atomic32 Release_AtomicExchange(volatile Atomic32* ptr, - Atomic32 new_value) { - return __atomic_exchange_n(const_cast(ptr), new_value, __ATOMIC_RELEASE); -} - -inline Atomic32 Acquire_CompareAndSwap(volatile Atomic32* ptr, - Atomic32 old_value, - Atomic32 new_value) { - Atomic32 prev_value = old_value; - __atomic_compare_exchange_n(ptr, &prev_value, new_value, - 0, __ATOMIC_ACQUIRE, __ATOMIC_RELAXED); - return prev_value; -} - -inline Atomic32 Release_CompareAndSwap(volatile Atomic32* ptr, - Atomic32 old_value, - Atomic32 new_value) { - Atomic32 prev_value = old_value; - __atomic_compare_exchange_n(ptr, &prev_value, new_value, - 0, __ATOMIC_RELEASE, __ATOMIC_RELAXED); - return prev_value; -} - -inline void NoBarrier_Store(volatile Atomic32* ptr, Atomic32 value) { - *ptr = value; -} - -inline void Acquire_Store(volatile Atomic32* ptr, Atomic32 value) { - *ptr = value; - MemoryBarrier(); -} - -inline void Release_Store(volatile Atomic32* ptr, Atomic32 value) { - MemoryBarrier(); - *ptr = value; -} - -inline Atomic32 NoBarrier_Load(volatile const Atomic32* ptr) { - return *ptr; -} - -inline Atomic32 Acquire_Load(volatile const Atomic32* ptr) { - Atomic32 value = *ptr; - MemoryBarrier(); - return value; -} - -inline Atomic32 Release_Load(volatile const Atomic32* ptr) { - MemoryBarrier(); - return *ptr; -} - -// 64-bit versions - -inline Atomic64 NoBarrier_CompareAndSwap(volatile Atomic64* ptr, - Atomic64 old_value, - Atomic64 new_value) { - Atomic64 prev_value = old_value; - __atomic_compare_exchange_n(ptr, &prev_value, new_value, - 0, __ATOMIC_RELAXED, __ATOMIC_RELAXED); - return prev_value; -} - -inline Atomic64 NoBarrier_AtomicExchange(volatile Atomic64* ptr, - Atomic64 new_value) { - return __atomic_exchange_n(const_cast(ptr), new_value, __ATOMIC_RELAXED); -} - -inline Atomic64 Acquire_AtomicExchange(volatile Atomic64* ptr, - Atomic64 new_value) { - return __atomic_exchange_n(const_cast(ptr), new_value, __ATOMIC_ACQUIRE); -} - -inline Atomic64 Release_AtomicExchange(volatile Atomic64* ptr, - Atomic64 new_value) { - return __atomic_exchange_n(const_cast(ptr), new_value, __ATOMIC_RELEASE); -} - -inline Atomic64 Acquire_CompareAndSwap(volatile Atomic64* ptr, - Atomic64 old_value, - Atomic64 new_value) { - Atomic64 prev_value = old_value; - __atomic_compare_exchange_n(ptr, &prev_value, new_value, - 0, __ATOMIC_ACQUIRE, __ATOMIC_RELAXED); - return prev_value; -} - -inline Atomic64 Release_CompareAndSwap(volatile Atomic64* ptr, - Atomic64 old_value, - Atomic64 new_value) { - Atomic64 prev_value = old_value; - __atomic_compare_exchange_n(ptr, &prev_value, new_value, - 0, __ATOMIC_RELEASE, __ATOMIC_RELAXED); - return prev_value; -} - -inline void NoBarrier_Store(volatile Atomic64* ptr, Atomic64 value) { - *ptr = value; -} - -inline void Acquire_Store(volatile Atomic64* ptr, Atomic64 value) { - *ptr = value; - MemoryBarrier(); -} - -inline void Release_Store(volatile Atomic64* ptr, Atomic64 value) { - MemoryBarrier(); - *ptr = value; -} - -inline Atomic64 NoBarrier_Load(volatile const Atomic64* ptr) { - return *ptr; -} - -inline Atomic64 Acquire_Load(volatile const Atomic64* ptr) { - Atomic64 value = *ptr; - MemoryBarrier(); - return value; -} - -inline Atomic64 Release_Load(volatile const Atomic64* ptr) { - MemoryBarrier(); - return *ptr; -} - -} // namespace base::subtle -} // namespace base - -#endif // BASE_ATOMICOPS_INTERNALS_GCC_GENERIC_H_ diff --git a/contrib/libtcmalloc/src/base/atomicops-internals-linuxppc.h b/contrib/libtcmalloc/src/base/atomicops-internals-linuxppc.h deleted file mode 100644 index b52fdf0d1ec..00000000000 --- a/contrib/libtcmalloc/src/base/atomicops-internals-linuxppc.h +++ /dev/null @@ -1,437 +0,0 @@ -// -*- Mode: C++; c-basic-offset: 2; indent-tabs-mode: nil -*- -/* Copyright (c) 2008, Google Inc. - * All rights reserved. - * - * Redistribution and use in source and binary forms, with or without - * modification, are permitted provided that the following conditions are - * met: - * - * * Redistributions of source code must retain the above copyright - * notice, this list of conditions and the following disclaimer. - * * Redistributions in binary form must reproduce the above - * copyright notice, this list of conditions and the following disclaimer - * in the documentation and/or other materials provided with the - * distribution. - * * Neither the name of Google Inc. nor the names of its - * contributors may be used to endorse or promote products derived from - * this software without specific prior written permission. - * - * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS - * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT - * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR - * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT - * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, - * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT - * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, - * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY - * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT - * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE - * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. - * - * --- - */ - -// Implementation of atomic operations for ppc-linux. This file should not -// be included directly. Clients should instead include -// "base/atomicops.h". - -#ifndef BASE_ATOMICOPS_INTERNALS_LINUXPPC_H_ -#define BASE_ATOMICOPS_INTERNALS_LINUXPPC_H_ - -typedef int32_t Atomic32; - -#ifdef __PPC64__ -#define BASE_HAS_ATOMIC64 1 -#endif - -namespace base { -namespace subtle { - -static inline void _sync(void) { - __asm__ __volatile__("sync": : : "memory"); -} - -static inline void _lwsync(void) { - // gcc defines __NO_LWSYNC__ when appropriate; see - // http://gcc.gnu.org/ml/gcc-patches/2006-11/msg01238.html -#ifdef __NO_LWSYNC__ - __asm__ __volatile__("msync": : : "memory"); -#else - __asm__ __volatile__("lwsync": : : "memory"); -#endif -} - -static inline void _isync(void) { - __asm__ __volatile__("isync": : : "memory"); -} - -static inline Atomic32 OSAtomicAdd32(Atomic32 amount, Atomic32 *value) { - Atomic32 t; - __asm__ __volatile__( -"1: lwarx %0,0,%3\n\ - add %0,%2,%0\n\ - stwcx. %0,0,%3 \n\ - bne- 1b" - : "=&r" (t), "+m" (*value) - : "r" (amount), "r" (value) - : "cc"); - return t; -} - -static inline Atomic32 OSAtomicAdd32Barrier(Atomic32 amount, Atomic32 *value) { - Atomic32 t; - _lwsync(); - t = OSAtomicAdd32(amount, value); - // This is based on the code snippet in the architecture manual (Vol - // 2, Appendix B). It's a little tricky: correctness depends on the - // fact that the code right before this (in OSAtomicAdd32) has a - // conditional branch with a data dependency on the update. - // Otherwise, we'd have to use sync. - _isync(); - return t; -} - -static inline bool OSAtomicCompareAndSwap32(Atomic32 old_value, - Atomic32 new_value, - Atomic32 *value) { - Atomic32 prev; - __asm__ __volatile__( -"1: lwarx %0,0,%2\n\ - cmpw 0,%0,%3\n\ - bne- 2f\n\ - stwcx. %4,0,%2\n\ - bne- 1b\n\ -2:" - : "=&r" (prev), "+m" (*value) - : "r" (value), "r" (old_value), "r" (new_value) - : "cc"); - return prev == old_value; -} - -static inline Atomic32 OSAtomicCompareAndSwap32Acquire(Atomic32 old_value, - Atomic32 new_value, - Atomic32 *value) { - Atomic32 t; - t = OSAtomicCompareAndSwap32(old_value, new_value, value); - // This is based on the code snippet in the architecture manual (Vol - // 2, Appendix B). It's a little tricky: correctness depends on the - // fact that the code right before this (in - // OSAtomicCompareAndSwap32) has a conditional branch with a data - // dependency on the update. Otherwise, we'd have to use sync. - _isync(); - return t; -} - -static inline Atomic32 OSAtomicCompareAndSwap32Release(Atomic32 old_value, - Atomic32 new_value, - Atomic32 *value) { - _lwsync(); - return OSAtomicCompareAndSwap32(old_value, new_value, value); -} - -typedef int64_t Atomic64; - -inline void MemoryBarrier() { - // This can't be _lwsync(); we need to order the immediately - // preceding stores against any load that may follow, but lwsync - // doesn't guarantee that. - _sync(); -} - -// 32-bit Versions. - -inline Atomic32 NoBarrier_CompareAndSwap(volatile Atomic32 *ptr, - Atomic32 old_value, - Atomic32 new_value) { - Atomic32 prev_value; - do { - if (OSAtomicCompareAndSwap32(old_value, new_value, - const_cast(ptr))) { - return old_value; - } - prev_value = *ptr; - } while (prev_value == old_value); - return prev_value; -} - -inline Atomic32 NoBarrier_AtomicExchange(volatile Atomic32 *ptr, - Atomic32 new_value) { - Atomic32 old_value; - do { - old_value = *ptr; - } while (!OSAtomicCompareAndSwap32(old_value, new_value, - const_cast(ptr))); - return old_value; -} - -inline Atomic32 Acquire_AtomicExchange(volatile Atomic32 *ptr, - Atomic32 new_value) { - Atomic32 old_value; - do { - old_value = *ptr; - } while (!OSAtomicCompareAndSwap32Acquire(old_value, new_value, - const_cast(ptr))); - return old_value; -} - -inline Atomic32 Release_AtomicExchange(volatile Atomic32 *ptr, - Atomic32 new_value) { - Atomic32 old_value; - do { - old_value = *ptr; - } while (!OSAtomicCompareAndSwap32Release(old_value, new_value, - const_cast(ptr))); - return old_value; -} - -inline Atomic32 Acquire_CompareAndSwap(volatile Atomic32 *ptr, - Atomic32 old_value, - Atomic32 new_value) { - Atomic32 prev_value; - do { - if (OSAtomicCompareAndSwap32Acquire(old_value, new_value, - const_cast(ptr))) { - return old_value; - } - prev_value = *ptr; - } while (prev_value == old_value); - return prev_value; -} - -inline Atomic32 Release_CompareAndSwap(volatile Atomic32 *ptr, - Atomic32 old_value, - Atomic32 new_value) { - Atomic32 prev_value; - do { - if (OSAtomicCompareAndSwap32Release(old_value, new_value, - const_cast(ptr))) { - return old_value; - } - prev_value = *ptr; - } while (prev_value == old_value); - return prev_value; -} - -#ifdef __PPC64__ - -// 64-bit Versions. - -static inline Atomic64 OSAtomicAdd64(Atomic64 amount, Atomic64 *value) { - Atomic64 t; - __asm__ __volatile__( -"1: ldarx %0,0,%3\n\ - add %0,%2,%0\n\ - stdcx. %0,0,%3 \n\ - bne- 1b" - : "=&r" (t), "+m" (*value) - : "r" (amount), "r" (value) - : "cc"); - return t; -} - -static inline Atomic64 OSAtomicAdd64Barrier(Atomic64 amount, Atomic64 *value) { - Atomic64 t; - _lwsync(); - t = OSAtomicAdd64(amount, value); - // This is based on the code snippet in the architecture manual (Vol - // 2, Appendix B). It's a little tricky: correctness depends on the - // fact that the code right before this (in OSAtomicAdd64) has a - // conditional branch with a data dependency on the update. - // Otherwise, we'd have to use sync. - _isync(); - return t; -} - -static inline bool OSAtomicCompareAndSwap64(Atomic64 old_value, - Atomic64 new_value, - Atomic64 *value) { - Atomic64 prev; - __asm__ __volatile__( -"1: ldarx %0,0,%2\n\ - cmpd 0,%0,%3\n\ - bne- 2f\n\ - stdcx. %4,0,%2\n\ - bne- 1b\n\ -2:" - : "=&r" (prev), "+m" (*value) - : "r" (value), "r" (old_value), "r" (new_value) - : "cc"); - return prev == old_value; -} - -static inline Atomic64 OSAtomicCompareAndSwap64Acquire(Atomic64 old_value, - Atomic64 new_value, - Atomic64 *value) { - Atomic64 t; - t = OSAtomicCompareAndSwap64(old_value, new_value, value); - // This is based on the code snippet in the architecture manual (Vol - // 2, Appendix B). It's a little tricky: correctness depends on the - // fact that the code right before this (in - // OSAtomicCompareAndSwap64) has a conditional branch with a data - // dependency on the update. Otherwise, we'd have to use sync. - _isync(); - return t; -} - -static inline Atomic64 OSAtomicCompareAndSwap64Release(Atomic64 old_value, - Atomic64 new_value, - Atomic64 *value) { - _lwsync(); - return OSAtomicCompareAndSwap64(old_value, new_value, value); -} - - -inline Atomic64 NoBarrier_CompareAndSwap(volatile Atomic64 *ptr, - Atomic64 old_value, - Atomic64 new_value) { - Atomic64 prev_value; - do { - if (OSAtomicCompareAndSwap64(old_value, new_value, - const_cast(ptr))) { - return old_value; - } - prev_value = *ptr; - } while (prev_value == old_value); - return prev_value; -} - -inline Atomic64 NoBarrier_AtomicExchange(volatile Atomic64 *ptr, - Atomic64 new_value) { - Atomic64 old_value; - do { - old_value = *ptr; - } while (!OSAtomicCompareAndSwap64(old_value, new_value, - const_cast(ptr))); - return old_value; -} - -inline Atomic64 Acquire_AtomicExchange(volatile Atomic64 *ptr, - Atomic64 new_value) { - Atomic64 old_value; - do { - old_value = *ptr; - } while (!OSAtomicCompareAndSwap64Acquire(old_value, new_value, - const_cast(ptr))); - return old_value; -} - -inline Atomic64 Release_AtomicExchange(volatile Atomic64 *ptr, - Atomic64 new_value) { - Atomic64 old_value; - do { - old_value = *ptr; - } while (!OSAtomicCompareAndSwap64Release(old_value, new_value, - const_cast(ptr))); - return old_value; -} - -inline Atomic64 Acquire_CompareAndSwap(volatile Atomic64 *ptr, - Atomic64 old_value, - Atomic64 new_value) { - Atomic64 prev_value; - do { - if (OSAtomicCompareAndSwap64Acquire(old_value, new_value, - const_cast(ptr))) { - return old_value; - } - prev_value = *ptr; - } while (prev_value == old_value); - return prev_value; -} - -inline Atomic64 Release_CompareAndSwap(volatile Atomic64 *ptr, - Atomic64 old_value, - Atomic64 new_value) { - Atomic64 prev_value; - do { - if (OSAtomicCompareAndSwap64Release(old_value, new_value, - const_cast(ptr))) { - return old_value; - } - prev_value = *ptr; - } while (prev_value == old_value); - return prev_value; -} - -#endif - -inline void NoBarrier_Store(volatile Atomic32 *ptr, Atomic32 value) { - *ptr = value; -} - -inline void Acquire_Store(volatile Atomic32 *ptr, Atomic32 value) { - *ptr = value; - // This can't be _lwsync(); we need to order the immediately - // preceding stores against any load that may follow, but lwsync - // doesn't guarantee that. - _sync(); -} - -inline void Release_Store(volatile Atomic32 *ptr, Atomic32 value) { - _lwsync(); - *ptr = value; -} - -inline Atomic32 NoBarrier_Load(volatile const Atomic32 *ptr) { - return *ptr; -} - -inline Atomic32 Acquire_Load(volatile const Atomic32 *ptr) { - Atomic32 value = *ptr; - _lwsync(); - return value; -} - -inline Atomic32 Release_Load(volatile const Atomic32 *ptr) { - // This can't be _lwsync(); we need to order the immediately - // preceding stores against any load that may follow, but lwsync - // doesn't guarantee that. - _sync(); - return *ptr; -} - -#ifdef __PPC64__ - -// 64-bit Versions. - -inline void NoBarrier_Store(volatile Atomic64 *ptr, Atomic64 value) { - *ptr = value; -} - -inline void Acquire_Store(volatile Atomic64 *ptr, Atomic64 value) { - *ptr = value; - // This can't be _lwsync(); we need to order the immediately - // preceding stores against any load that may follow, but lwsync - // doesn't guarantee that. - _sync(); -} - -inline void Release_Store(volatile Atomic64 *ptr, Atomic64 value) { - _lwsync(); - *ptr = value; -} - -inline Atomic64 NoBarrier_Load(volatile const Atomic64 *ptr) { - return *ptr; -} - -inline Atomic64 Acquire_Load(volatile const Atomic64 *ptr) { - Atomic64 value = *ptr; - _lwsync(); - return value; -} - -inline Atomic64 Release_Load(volatile const Atomic64 *ptr) { - // This can't be _lwsync(); we need to order the immediately - // preceding stores against any load that may follow, but lwsync - // doesn't guarantee that. - _sync(); - return *ptr; -} - -#endif - -} // namespace base::subtle -} // namespace base - -#endif // BASE_ATOMICOPS_INTERNALS_LINUXPPC_H_ diff --git a/contrib/libtcmalloc/src/base/atomicops-internals-macosx.h b/contrib/libtcmalloc/src/base/atomicops-internals-macosx.h deleted file mode 100644 index b5130d4f4d7..00000000000 --- a/contrib/libtcmalloc/src/base/atomicops-internals-macosx.h +++ /dev/null @@ -1,370 +0,0 @@ -// -*- Mode: C++; c-basic-offset: 2; indent-tabs-mode: nil -*- -/* Copyright (c) 2006, Google Inc. - * All rights reserved. - * - * Redistribution and use in source and binary forms, with or without - * modification, are permitted provided that the following conditions are - * met: - * - * * Redistributions of source code must retain the above copyright - * notice, this list of conditions and the following disclaimer. - * * Redistributions in binary form must reproduce the above - * copyright notice, this list of conditions and the following disclaimer - * in the documentation and/or other materials provided with the - * distribution. - * * Neither the name of Google Inc. nor the names of its - * contributors may be used to endorse or promote products derived from - * this software without specific prior written permission. - * - * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS - * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT - * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR - * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT - * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, - * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT - * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, - * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY - * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT - * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE - * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. - */ - -// Implementation of atomic operations for Mac OS X. This file should not -// be included directly. Clients should instead include -// "base/atomicops.h". - -#ifndef BASE_ATOMICOPS_INTERNALS_MACOSX_H_ -#define BASE_ATOMICOPS_INTERNALS_MACOSX_H_ - -typedef int32_t Atomic32; - -// MacOS uses long for intptr_t, AtomicWord and Atomic32 are always different -// on the Mac, even when they are the same size. Similarly, on __ppc64__, -// AtomicWord and Atomic64 are always different. Thus, we need explicit -// casting. -#ifdef __LP64__ -#define AtomicWordCastType base::subtle::Atomic64 -#else -#define AtomicWordCastType Atomic32 -#endif - -#if defined(__LP64__) || defined(__i386__) -#define BASE_HAS_ATOMIC64 1 // Use only in tests and base/atomic* -#endif - -#include - -namespace base { -namespace subtle { - -#if !defined(__LP64__) && defined(__ppc__) - -// The Mac 64-bit OSAtomic implementations are not available for 32-bit PowerPC, -// while the underlying assembly instructions are available only some -// implementations of PowerPC. - -// The following inline functions will fail with the error message at compile -// time ONLY IF they are called. So it is safe to use this header if user -// code only calls AtomicWord and Atomic32 operations. -// -// NOTE(vchen): Implementation notes to implement the atomic ops below may -// be found in "PowerPC Virtual Environment Architecture, Book II, -// Version 2.02", January 28, 2005, Appendix B, page 46. Unfortunately, -// extra care must be taken to ensure data are properly 8-byte aligned, and -// that data are returned correctly according to Mac OS X ABI specs. - -inline int64_t OSAtomicCompareAndSwap64( - int64_t oldValue, int64_t newValue, int64_t *theValue) { - __asm__ __volatile__( - "_OSAtomicCompareAndSwap64_not_supported_for_32_bit_ppc\n\t"); - return 0; -} - -inline int64_t OSAtomicAdd64(int64_t theAmount, int64_t *theValue) { - __asm__ __volatile__( - "_OSAtomicAdd64_not_supported_for_32_bit_ppc\n\t"); - return 0; -} - -inline int64_t OSAtomicCompareAndSwap64Barrier( - int64_t oldValue, int64_t newValue, int64_t *theValue) { - int64_t prev = OSAtomicCompareAndSwap64(oldValue, newValue, theValue); - OSMemoryBarrier(); - return prev; -} - -inline int64_t OSAtomicAdd64Barrier( - int64_t theAmount, int64_t *theValue) { - int64_t new_val = OSAtomicAdd64(theAmount, theValue); - OSMemoryBarrier(); - return new_val; -} -#endif - -typedef int64_t Atomic64; - -inline void MemoryBarrier() { - OSMemoryBarrier(); -} - -// 32-bit Versions. - -inline Atomic32 NoBarrier_CompareAndSwap(volatile Atomic32 *ptr, - Atomic32 old_value, - Atomic32 new_value) { - Atomic32 prev_value; - do { - if (OSAtomicCompareAndSwap32(old_value, new_value, - const_cast(ptr))) { - return old_value; - } - prev_value = *ptr; - } while (prev_value == old_value); - return prev_value; -} - -inline Atomic32 NoBarrier_AtomicExchange(volatile Atomic32 *ptr, - Atomic32 new_value) { - Atomic32 old_value; - do { - old_value = *ptr; - } while (!OSAtomicCompareAndSwap32(old_value, new_value, - const_cast(ptr))); - return old_value; -} - -inline Atomic32 Acquire_AtomicExchange(volatile Atomic32 *ptr, - Atomic32 new_value) { - Atomic32 old_value; - do { - old_value = *ptr; - } while (!OSAtomicCompareAndSwap32Barrier(old_value, new_value, - const_cast(ptr))); - return old_value; -} - -inline Atomic32 Release_AtomicExchange(volatile Atomic32 *ptr, - Atomic32 new_value) { - return Acquire_AtomicExchange(ptr, new_value); -} - -inline Atomic32 Acquire_CompareAndSwap(volatile Atomic32 *ptr, - Atomic32 old_value, - Atomic32 new_value) { - Atomic32 prev_value; - do { - if (OSAtomicCompareAndSwap32Barrier(old_value, new_value, - const_cast(ptr))) { - return old_value; - } - prev_value = *ptr; - } while (prev_value == old_value); - return prev_value; -} - -inline Atomic32 Release_CompareAndSwap(volatile Atomic32 *ptr, - Atomic32 old_value, - Atomic32 new_value) { - return Acquire_CompareAndSwap(ptr, old_value, new_value); -} - -inline void NoBarrier_Store(volatile Atomic32* ptr, Atomic32 value) { - *ptr = value; -} - -inline void Acquire_Store(volatile Atomic32 *ptr, Atomic32 value) { - *ptr = value; - MemoryBarrier(); -} - -inline void Release_Store(volatile Atomic32 *ptr, Atomic32 value) { - MemoryBarrier(); - *ptr = value; -} - -inline Atomic32 NoBarrier_Load(volatile const Atomic32* ptr) { - return *ptr; -} - -inline Atomic32 Acquire_Load(volatile const Atomic32 *ptr) { - Atomic32 value = *ptr; - MemoryBarrier(); - return value; -} - -inline Atomic32 Release_Load(volatile const Atomic32 *ptr) { - MemoryBarrier(); - return *ptr; -} - -// 64-bit version - -inline Atomic64 NoBarrier_CompareAndSwap(volatile Atomic64 *ptr, - Atomic64 old_value, - Atomic64 new_value) { - Atomic64 prev_value; - do { - if (OSAtomicCompareAndSwap64(old_value, new_value, - const_cast(ptr))) { - return old_value; - } - prev_value = *ptr; - } while (prev_value == old_value); - return prev_value; -} - -inline Atomic64 NoBarrier_AtomicExchange(volatile Atomic64 *ptr, - Atomic64 new_value) { - Atomic64 old_value; - do { - old_value = *ptr; - } while (!OSAtomicCompareAndSwap64(old_value, new_value, - const_cast(ptr))); - return old_value; -} - -inline Atomic64 Acquire_AtomicExchange(volatile Atomic64 *ptr, - Atomic64 new_value) { - Atomic64 old_value; - do { - old_value = *ptr; - } while (!OSAtomicCompareAndSwap64Barrier(old_value, new_value, - const_cast(ptr))); - return old_value; -} - -inline Atomic64 Release_AtomicExchange(volatile Atomic64 *ptr, - Atomic64 new_value) { - return Acquire_AtomicExchange(ptr, new_value); -} - -inline Atomic64 Acquire_CompareAndSwap(volatile Atomic64 *ptr, - Atomic64 old_value, - Atomic64 new_value) { - Atomic64 prev_value; - do { - if (OSAtomicCompareAndSwap64Barrier(old_value, new_value, - const_cast(ptr))) { - return old_value; - } - prev_value = *ptr; - } while (prev_value == old_value); - return prev_value; -} - -inline Atomic64 Release_CompareAndSwap(volatile Atomic64 *ptr, - Atomic64 old_value, - Atomic64 new_value) { - // The lib kern interface does not distinguish between - // Acquire and Release memory barriers; they are equivalent. - return Acquire_CompareAndSwap(ptr, old_value, new_value); -} - -#ifdef __LP64__ - -// 64-bit implementation on 64-bit platform - -inline void NoBarrier_Store(volatile Atomic64* ptr, Atomic64 value) { - *ptr = value; -} - -inline void Acquire_Store(volatile Atomic64 *ptr, Atomic64 value) { - *ptr = value; - MemoryBarrier(); -} - -inline void Release_Store(volatile Atomic64 *ptr, Atomic64 value) { - MemoryBarrier(); - *ptr = value; -} - -inline Atomic64 NoBarrier_Load(volatile const Atomic64* ptr) { - return *ptr; -} - -inline Atomic64 Acquire_Load(volatile const Atomic64 *ptr) { - Atomic64 value = *ptr; - MemoryBarrier(); - return value; -} - -inline Atomic64 Release_Load(volatile const Atomic64 *ptr) { - MemoryBarrier(); - return *ptr; -} - -#else - -// 64-bit implementation on 32-bit platform - -#if defined(__ppc__) - -inline void NoBarrier_Store(volatile Atomic64* ptr, Atomic64 value) { - __asm__ __volatile__( - "_NoBarrier_Store_not_supported_for_32_bit_ppc\n\t"); -} - -inline Atomic64 NoBarrier_Load(volatile const Atomic64* ptr) { - __asm__ __volatile__( - "_NoBarrier_Load_not_supported_for_32_bit_ppc\n\t"); - return 0; -} - -#elif defined(__i386__) - -inline void NoBarrier_Store(volatile Atomic64* ptr, Atomic64 value) { - __asm__ __volatile__("movq %1, %%mm0\n\t" // Use mmx reg for 64-bit atomic - "movq %%mm0, %0\n\t" // moves (ptr could be read-only) - "emms\n\t" // Reset FP registers - : "=m" (*ptr) - : "m" (value) - : // mark the FP stack and mmx registers as clobbered - "st", "st(1)", "st(2)", "st(3)", "st(4)", - "st(5)", "st(6)", "st(7)", "mm0", "mm1", - "mm2", "mm3", "mm4", "mm5", "mm6", "mm7"); - -} - -inline Atomic64 NoBarrier_Load(volatile const Atomic64* ptr) { - Atomic64 value; - __asm__ __volatile__("movq %1, %%mm0\n\t" // Use mmx reg for 64-bit atomic - "movq %%mm0, %0\n\t" // moves (ptr could be read-only) - "emms\n\t" // Reset FP registers - : "=m" (value) - : "m" (*ptr) - : // mark the FP stack and mmx registers as clobbered - "st", "st(1)", "st(2)", "st(3)", "st(4)", - "st(5)", "st(6)", "st(7)", "mm0", "mm1", - "mm2", "mm3", "mm4", "mm5", "mm6", "mm7"); - - return value; -} -#endif - - -inline void Acquire_Store(volatile Atomic64 *ptr, Atomic64 value) { - NoBarrier_Store(ptr, value); - MemoryBarrier(); -} - -inline void Release_Store(volatile Atomic64 *ptr, Atomic64 value) { - MemoryBarrier(); - NoBarrier_Store(ptr, value); -} - -inline Atomic64 Acquire_Load(volatile const Atomic64 *ptr) { - Atomic64 value = NoBarrier_Load(ptr); - MemoryBarrier(); - return value; -} - -inline Atomic64 Release_Load(volatile const Atomic64 *ptr) { - MemoryBarrier(); - return NoBarrier_Load(ptr); -} -#endif // __LP64__ - -} // namespace base::subtle -} // namespace base - -#endif // BASE_ATOMICOPS_INTERNALS_MACOSX_H_ diff --git a/contrib/libtcmalloc/src/base/atomicops-internals-mips.h b/contrib/libtcmalloc/src/base/atomicops-internals-mips.h deleted file mode 100644 index 4bfd7f6c70d..00000000000 --- a/contrib/libtcmalloc/src/base/atomicops-internals-mips.h +++ /dev/null @@ -1,323 +0,0 @@ -// -*- Mode: C++; c-basic-offset: 2; indent-tabs-mode: nil -*- -/* Copyright (c) 2013, Google Inc. - * All rights reserved. - * - * Redistribution and use in source and binary forms, with or without - * modification, are permitted provided that the following conditions are - * met: - * - * * Redistributions of source code must retain the above copyright - * notice, this list of conditions and the following disclaimer. - * * Redistributions in binary form must reproduce the above - * copyright notice, this list of conditions and the following disclaimer - * in the documentation and/or other materials provided with the - * distribution. - * * Neither the name of Google Inc. nor the names of its - * contributors may be used to endorse or promote products derived from - * this software without specific prior written permission. - * - * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS - * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT - * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR - * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT - * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, - * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT - * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, - * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY - * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT - * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE - * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. - */ - -// Author: Jovan Zelincevic -// based on atomicops-internals by Sanjay Ghemawat - -// This file is an internal atomic implementation, use base/atomicops.h instead. -// -// This code implements MIPS atomics. - -#ifndef BASE_ATOMICOPS_INTERNALS_MIPS_H_ -#define BASE_ATOMICOPS_INTERNALS_MIPS_H_ - -#if (_MIPS_ISA == _MIPS_ISA_MIPS64) -#define BASE_HAS_ATOMIC64 1 -#endif - -typedef int32_t Atomic32; - -namespace base { -namespace subtle { - -// Atomically execute: -// result = *ptr; -// if (*ptr == old_value) -// *ptr = new_value; -// return result; -// -// I.e., replace "*ptr" with "new_value" if "*ptr" used to be "old_value". -// Always return the old value of "*ptr" -// -// This routine implies no memory barriers. -inline Atomic32 NoBarrier_CompareAndSwap(volatile Atomic32* ptr, - Atomic32 old_value, - Atomic32 new_value) -{ - Atomic32 prev, tmp; - __asm__ volatile( - ".set push \n" - ".set noreorder \n" - - "1: \n" - "ll %0, %5 \n" // prev = *ptr - "bne %0, %3, 2f \n" // if (prev != old_value) goto 2 - " move %2, %4 \n" // tmp = new_value - "sc %2, %1 \n" // *ptr = tmp (with atomic check) - "beqz %2, 1b \n" // start again on atomic error - " nop \n" // delay slot nop - "2: \n" - - ".set pop \n" - : "=&r" (prev), "=m" (*ptr), - "=&r" (tmp) - : "Ir" (old_value), "r" (new_value), - "m" (*ptr) - : "memory" - ); - return prev; -} - -// Atomically store new_value into *ptr, returning the previous value held in -// *ptr. This routine implies no memory barriers. -inline Atomic32 NoBarrier_AtomicExchange(volatile Atomic32* ptr, - Atomic32 new_value) -{ - Atomic32 temp, old; - __asm__ volatile( - ".set push \n" - ".set noreorder \n" - - "1: \n" - "ll %1, %2 \n" // old = *ptr - "move %0, %3 \n" // temp = new_value - "sc %0, %2 \n" // *ptr = temp (with atomic check) - "beqz %0, 1b \n" // start again on atomic error - " nop \n" // delay slot nop - - ".set pop \n" - : "=&r" (temp), "=&r" (old), - "=m" (*ptr) - : "r" (new_value), "m" (*ptr) - : "memory" - ); - return old; -} - -inline void MemoryBarrier() -{ - __asm__ volatile("sync" : : : "memory"); -} - -// "Acquire" operations -// ensure that no later memory access can be reordered ahead of the operation. -// "Release" operations ensure that no previous memory access can be reordered -// after the operation. "Barrier" operations have both "Acquire" and "Release" -// semantics. A MemoryBarrier() has "Barrier" semantics, but does no memory -// access. -inline Atomic32 Acquire_CompareAndSwap(volatile Atomic32* ptr, - Atomic32 old_value, - Atomic32 new_value) -{ - Atomic32 res = NoBarrier_CompareAndSwap(ptr, old_value, new_value); - MemoryBarrier(); - return res; -} - -inline Atomic32 Release_CompareAndSwap(volatile Atomic32* ptr, - Atomic32 old_value, - Atomic32 new_value) -{ - MemoryBarrier(); - Atomic32 res = NoBarrier_CompareAndSwap(ptr, old_value, new_value); - return res; -} - -inline void NoBarrier_Store(volatile Atomic32* ptr, Atomic32 value) -{ - *ptr = value; -} - -inline Atomic32 Acquire_AtomicExchange(volatile Atomic32* ptr, - Atomic32 new_value) -{ - Atomic32 old_value = NoBarrier_AtomicExchange(ptr, new_value); - MemoryBarrier(); - return old_value; -} - -inline Atomic32 Release_AtomicExchange(volatile Atomic32* ptr, - Atomic32 new_value) -{ - MemoryBarrier(); - return NoBarrier_AtomicExchange(ptr, new_value); -} - -inline void Acquire_Store(volatile Atomic32* ptr, Atomic32 value) -{ - *ptr = value; - MemoryBarrier(); -} - -inline void Release_Store(volatile Atomic32* ptr, Atomic32 value) -{ - MemoryBarrier(); - *ptr = value; -} - -inline Atomic32 NoBarrier_Load(volatile const Atomic32* ptr) -{ - return *ptr; -} - -inline Atomic32 Acquire_Load(volatile const Atomic32* ptr) -{ - Atomic32 value = *ptr; - MemoryBarrier(); - return value; -} - -inline Atomic32 Release_Load(volatile const Atomic32* ptr) -{ - MemoryBarrier(); - return *ptr; -} - -#if (_MIPS_ISA == _MIPS_ISA_MIPS64) || (_MIPS_SIM == _MIPS_SIM_ABI64) - -typedef int64_t Atomic64; - -inline Atomic64 NoBarrier_CompareAndSwap(volatile Atomic64* ptr, - Atomic64 old_value, - Atomic64 new_value) -{ - Atomic64 prev, tmp; - __asm__ volatile( - ".set push \n" - ".set noreorder \n" - - "1: \n" - "lld %0, %5 \n" // prev = *ptr - "bne %0, %3, 2f \n" // if (prev != old_value) goto 2 - " move %2, %4 \n" // tmp = new_value - "scd %2, %1 \n" // *ptr = tmp (with atomic check) - "beqz %2, 1b \n" // start again on atomic error - " nop \n" // delay slot nop - "2: \n" - - ".set pop \n" - : "=&r" (prev), "=m" (*ptr), - "=&r" (tmp) - : "Ir" (old_value), "r" (new_value), - "m" (*ptr) - : "memory" - ); - return prev; -} - -inline Atomic64 NoBarrier_AtomicExchange(volatile Atomic64* ptr, - Atomic64 new_value) -{ - Atomic64 temp, old; - __asm__ volatile( - ".set push \n" - ".set noreorder \n" - - "1: \n" - "lld %1, %2 \n" // old = *ptr - "move %0, %3 \n" // temp = new_value - "scd %0, %2 \n" // *ptr = temp (with atomic check) - "beqz %0, 1b \n" // start again on atomic error - " nop \n" // delay slot nop - - ".set pop \n" - : "=&r" (temp), "=&r" (old), - "=m" (*ptr) - : "r" (new_value), "m" (*ptr) - : "memory" - ); - return old; -} - -inline Atomic64 Acquire_AtomicExchange(volatile Atomic64* ptr, - Atomic64 new_value) -{ - Atomic64 old_value = NoBarrier_AtomicExchange(ptr, new_value); - MemoryBarrier(); - return old_value; -} - -inline Atomic64 Acquire_CompareAndSwap(volatile Atomic64* ptr, - Atomic64 old_value, - Atomic64 new_value) -{ - Atomic64 res = NoBarrier_CompareAndSwap(ptr, old_value, new_value); - MemoryBarrier(); - return res; -} - -inline Atomic64 Release_CompareAndSwap(volatile Atomic64* ptr, - Atomic64 old_value, - Atomic64 new_value) -{ - MemoryBarrier(); - Atomic64 res = NoBarrier_CompareAndSwap(ptr, old_value, new_value); - return res; -} - -inline void NoBarrier_Store(volatile Atomic64* ptr, Atomic64 value) -{ - *ptr = value; -} - -inline Atomic64 Release_AtomicExchange(volatile Atomic64* ptr, - Atomic64 new_value) -{ - MemoryBarrier(); - return NoBarrier_AtomicExchange(ptr, new_value); -} - -inline void Acquire_Store(volatile Atomic64* ptr, Atomic64 value) -{ - *ptr = value; - MemoryBarrier(); -} - -inline void Release_Store(volatile Atomic64* ptr, Atomic64 value) -{ - MemoryBarrier(); - *ptr = value; -} - -inline Atomic64 NoBarrier_Load(volatile const Atomic64* ptr) -{ - return *ptr; -} - -inline Atomic64 Acquire_Load(volatile const Atomic64* ptr) -{ - Atomic64 value = *ptr; - MemoryBarrier(); - return value; -} - -inline Atomic64 Release_Load(volatile const Atomic64* ptr) -{ - MemoryBarrier(); - return *ptr; -} - -#endif - -} // namespace base::subtle -} // namespace base - -#endif // BASE_ATOMICOPS_INTERNALS_MIPS_H_ diff --git a/contrib/libtcmalloc/src/base/atomicops-internals-windows.h b/contrib/libtcmalloc/src/base/atomicops-internals-windows.h deleted file mode 100644 index 93ced8770d4..00000000000 --- a/contrib/libtcmalloc/src/base/atomicops-internals-windows.h +++ /dev/null @@ -1,457 +0,0 @@ -// -*- Mode: C++; c-basic-offset: 2; indent-tabs-mode: nil -*- -/* Copyright (c) 2006, Google Inc. - * All rights reserved. - * - * Redistribution and use in source and binary forms, with or without - * modification, are permitted provided that the following conditions are - * met: - * - * * Redistributions of source code must retain the above copyright - * notice, this list of conditions and the following disclaimer. - * * Redistributions in binary form must reproduce the above - * copyright notice, this list of conditions and the following disclaimer - * in the documentation and/or other materials provided with the - * distribution. - * * Neither the name of Google Inc. nor the names of its - * contributors may be used to endorse or promote products derived from - * this software without specific prior written permission. - * - * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS - * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT - * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR - * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT - * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, - * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT - * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, - * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY - * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT - * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE - * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. - * - * --- - * Author: Sanjay Ghemawat - */ - -// Implementation of atomic operations using Windows API -// functions. This file should not be included directly. Clients -// should instead include "base/atomicops.h". - -#ifndef BASE_ATOMICOPS_INTERNALS_WINDOWS_H_ -#define BASE_ATOMICOPS_INTERNALS_WINDOWS_H_ - -#include -#include -#include "base/basictypes.h" // For COMPILE_ASSERT - -typedef int32 Atomic32; - -#if defined(_WIN64) -#define BASE_HAS_ATOMIC64 1 // Use only in tests and base/atomic* -#endif - -namespace base { -namespace subtle { - -typedef int64 Atomic64; - -// 32-bit low-level operations on any platform - -extern "C" { -// We use windows intrinsics when we can (they seem to be supported -// well on MSVC 8.0 and above). Unfortunately, in some -// environments, and have conflicting -// declarations of some other intrinsics, breaking compilation: -// http://connect.microsoft.com/VisualStudio/feedback/details/262047 -// Therefore, we simply declare the relevant intrinsics ourself. - -// MinGW has a bug in the header files where it doesn't indicate the -// first argument is volatile -- they're not up to date. See -// http://readlist.com/lists/lists.sourceforge.net/mingw-users/0/3861.html -// We have to const_cast away the volatile to avoid compiler warnings. -// TODO(csilvers): remove this once MinGW has updated MinGW/include/winbase.h -#if defined(__MINGW32__) -inline LONG FastInterlockedCompareExchange(volatile LONG* ptr, - LONG newval, LONG oldval) { - return ::InterlockedCompareExchange(const_cast(ptr), newval, oldval); -} -inline LONG FastInterlockedExchange(volatile LONG* ptr, LONG newval) { - return ::InterlockedExchange(const_cast(ptr), newval); -} -inline LONG FastInterlockedExchangeAdd(volatile LONG* ptr, LONG increment) { - return ::InterlockedExchangeAdd(const_cast(ptr), increment); -} - -#elif _MSC_VER >= 1400 // intrinsics didn't work so well before MSVC 8.0 -// Unfortunately, in some environments, and -// have conflicting declarations of some intrinsics, breaking -// compilation. So we declare the intrinsics we need ourselves. See -// http://connect.microsoft.com/VisualStudio/feedback/details/262047 -LONG _InterlockedCompareExchange(volatile LONG* ptr, LONG newval, LONG oldval); -#pragma intrinsic(_InterlockedCompareExchange) -inline LONG FastInterlockedCompareExchange(volatile LONG* ptr, - LONG newval, LONG oldval) { - return _InterlockedCompareExchange(ptr, newval, oldval); -} - -LONG _InterlockedExchange(volatile LONG* ptr, LONG newval); -#pragma intrinsic(_InterlockedExchange) -inline LONG FastInterlockedExchange(volatile LONG* ptr, LONG newval) { - return _InterlockedExchange(ptr, newval); -} - -LONG _InterlockedExchangeAdd(volatile LONG* ptr, LONG increment); -#pragma intrinsic(_InterlockedExchangeAdd) -inline LONG FastInterlockedExchangeAdd(volatile LONG* ptr, LONG increment) { - return _InterlockedExchangeAdd(ptr, increment); -} - -#else -inline LONG FastInterlockedCompareExchange(volatile LONG* ptr, - LONG newval, LONG oldval) { - return ::InterlockedCompareExchange(ptr, newval, oldval); -} -inline LONG FastInterlockedExchange(volatile LONG* ptr, LONG newval) { - return ::InterlockedExchange(ptr, newval); -} -inline LONG FastInterlockedExchangeAdd(volatile LONG* ptr, LONG increment) { - return ::InterlockedExchangeAdd(ptr, increment); -} - -#endif // ifdef __MINGW32__ -} // extern "C" - -inline Atomic32 NoBarrier_CompareAndSwap(volatile Atomic32* ptr, - Atomic32 old_value, - Atomic32 new_value) { - LONG result = FastInterlockedCompareExchange( - reinterpret_cast(ptr), - static_cast(new_value), - static_cast(old_value)); - return static_cast(result); -} - -inline Atomic32 NoBarrier_AtomicExchange(volatile Atomic32* ptr, - Atomic32 new_value) { - LONG result = FastInterlockedExchange( - reinterpret_cast(ptr), - static_cast(new_value)); - return static_cast(result); -} - -inline Atomic32 Acquire_AtomicExchange(volatile Atomic32* ptr, - Atomic32 new_value) { - // FastInterlockedExchange has both acquire and release memory barriers. - return NoBarrier_AtomicExchange(ptr, new_value); -} - -inline Atomic32 Release_AtomicExchange(volatile Atomic32* ptr, - Atomic32 new_value) { - // FastInterlockedExchange has both acquire and release memory barriers. - return NoBarrier_AtomicExchange(ptr, new_value); -} - -} // namespace base::subtle -} // namespace base - - -// In msvc8/vs2005, winnt.h already contains a definition for -// MemoryBarrier in the global namespace. Add it there for earlier -// versions and forward to it from within the namespace. -#if !(defined(_MSC_VER) && _MSC_VER >= 1400) -inline void MemoryBarrier() { - Atomic32 value = 0; - base::subtle::NoBarrier_AtomicExchange(&value, 0); - // actually acts as a barrier in thisd implementation -} -#endif - -namespace base { -namespace subtle { - -inline void MemoryBarrier() { - ::MemoryBarrier(); -} - -inline Atomic32 Acquire_CompareAndSwap(volatile Atomic32* ptr, - Atomic32 old_value, - Atomic32 new_value) { - return NoBarrier_CompareAndSwap(ptr, old_value, new_value); -} - -inline Atomic32 Release_CompareAndSwap(volatile Atomic32* ptr, - Atomic32 old_value, - Atomic32 new_value) { - return NoBarrier_CompareAndSwap(ptr, old_value, new_value); -} - -inline void NoBarrier_Store(volatile Atomic32* ptr, Atomic32 value) { - *ptr = value; -} - -inline void Acquire_Store(volatile Atomic32* ptr, Atomic32 value) { - Acquire_AtomicExchange(ptr, value); -} - -inline void Release_Store(volatile Atomic32* ptr, Atomic32 value) { - *ptr = value; // works w/o barrier for current Intel chips as of June 2005 - // See comments in Atomic64 version of Release_Store() below. -} - -inline Atomic32 NoBarrier_Load(volatile const Atomic32* ptr) { - return *ptr; -} - -inline Atomic32 Acquire_Load(volatile const Atomic32* ptr) { - Atomic32 value = *ptr; - return value; -} - -inline Atomic32 Release_Load(volatile const Atomic32* ptr) { - MemoryBarrier(); - return *ptr; -} - -// 64-bit operations - -#if defined(_WIN64) || defined(__MINGW64__) - -// 64-bit low-level operations on 64-bit platform. - -COMPILE_ASSERT(sizeof(Atomic64) == sizeof(PVOID), atomic_word_is_atomic); - -// These are the intrinsics needed for 64-bit operations. Similar to the -// 32-bit case above. - -extern "C" { -#if defined(__MINGW64__) -inline PVOID FastInterlockedCompareExchangePointer(volatile PVOID* ptr, - PVOID newval, PVOID oldval) { - return ::InterlockedCompareExchangePointer(const_cast(ptr), - newval, oldval); -} -inline PVOID FastInterlockedExchangePointer(volatile PVOID* ptr, PVOID newval) { - return ::InterlockedExchangePointer(const_cast(ptr), newval); -} -inline LONGLONG FastInterlockedExchangeAdd64(volatile LONGLONG* ptr, - LONGLONG increment) { - return ::InterlockedExchangeAdd64(const_cast(ptr), increment); -} - -#elif _MSC_VER >= 1400 // intrinsics didn't work so well before MSVC 8.0 -// Like above, we need to declare the intrinsics ourselves. -PVOID _InterlockedCompareExchangePointer(volatile PVOID* ptr, - PVOID newval, PVOID oldval); -#pragma intrinsic(_InterlockedCompareExchangePointer) -inline PVOID FastInterlockedCompareExchangePointer(volatile PVOID* ptr, - PVOID newval, PVOID oldval) { - return _InterlockedCompareExchangePointer(const_cast(ptr), - newval, oldval); -} - -PVOID _InterlockedExchangePointer(volatile PVOID* ptr, PVOID newval); -#pragma intrinsic(_InterlockedExchangePointer) -inline PVOID FastInterlockedExchangePointer(volatile PVOID* ptr, PVOID newval) { - return _InterlockedExchangePointer(const_cast(ptr), newval); -} - -LONGLONG _InterlockedExchangeAdd64(volatile LONGLONG* ptr, LONGLONG increment); -#pragma intrinsic(_InterlockedExchangeAdd64) -inline LONGLONG FastInterlockedExchangeAdd64(volatile LONGLONG* ptr, - LONGLONG increment) { - return _InterlockedExchangeAdd64(const_cast(ptr), increment); -} - -#else -inline PVOID FastInterlockedCompareExchangePointer(volatile PVOID* ptr, - PVOID newval, PVOID oldval) { - return ::InterlockedCompareExchangePointer(ptr, newval, oldval); -} -inline PVOID FastInterlockedExchangePointer(volatile PVOID* ptr, PVOID newval) { - return ::InterlockedExchangePointer(ptr, newval); -} -inline LONGLONG FastInterlockedExchangeAdd64(volatile LONGLONG* ptr, - LONGLONG increment) { - return ::InterlockedExchangeAdd64(ptr, increment); -} - -#endif // ifdef __MINGW64__ -} // extern "C" - -inline Atomic64 NoBarrier_CompareAndSwap(volatile Atomic64* ptr, - Atomic64 old_value, - Atomic64 new_value) { - PVOID result = FastInterlockedCompareExchangePointer( - reinterpret_cast(ptr), - reinterpret_cast(new_value), reinterpret_cast(old_value)); - return reinterpret_cast(result); -} - -inline Atomic64 NoBarrier_AtomicExchange(volatile Atomic64* ptr, - Atomic64 new_value) { - PVOID result = FastInterlockedExchangePointer( - reinterpret_cast(ptr), - reinterpret_cast(new_value)); - return reinterpret_cast(result); -} - -inline void NoBarrier_Store(volatile Atomic64* ptr, Atomic64 value) { - *ptr = value; -} - -inline void Acquire_Store(volatile Atomic64* ptr, Atomic64 value) { - NoBarrier_AtomicExchange(ptr, value); - // acts as a barrier in this implementation -} - -inline void Release_Store(volatile Atomic64* ptr, Atomic64 value) { - *ptr = value; // works w/o barrier for current Intel chips as of June 2005 - - // When new chips come out, check: - // IA-32 Intel Architecture Software Developer's Manual, Volume 3: - // System Programming Guide, Chatper 7: Multiple-processor management, - // Section 7.2, Memory Ordering. - // Last seen at: - // http://developer.intel.com/design/pentium4/manuals/index_new.htm -} - -inline Atomic64 NoBarrier_Load(volatile const Atomic64* ptr) { - return *ptr; -} - -inline Atomic64 Acquire_Load(volatile const Atomic64* ptr) { - Atomic64 value = *ptr; - return value; -} - -inline Atomic64 Release_Load(volatile const Atomic64* ptr) { - MemoryBarrier(); - return *ptr; -} - -#else // defined(_WIN64) || defined(__MINGW64__) - -// 64-bit low-level operations on 32-bit platform - -// TODO(vchen): The GNU assembly below must be converted to MSVC inline -// assembly. Then the file should be renamed to ...-x86-msvc.h, probably. - -inline void NotImplementedFatalError(const char *function_name) { - fprintf(stderr, "64-bit %s() not implemented on this platform\n", - function_name); - abort(); -} - -inline Atomic64 NoBarrier_CompareAndSwap(volatile Atomic64* ptr, - Atomic64 old_value, - Atomic64 new_value) { -#if 0 // Not implemented - Atomic64 prev; - __asm__ __volatile__("movl (%3), %%ebx\n\t" // Move 64-bit new_value into - "movl 4(%3), %%ecx\n\t" // ecx:ebx - "lock; cmpxchg8b %1\n\t" // If edx:eax (old_value) same - : "=A" (prev) // as contents of ptr: - : "m" (*ptr), // ecx:ebx => ptr - "0" (old_value), // else: - "r" (&new_value) // old *ptr => edx:eax - : "memory", "%ebx", "%ecx"); - return prev; -#else - NotImplementedFatalError("NoBarrier_CompareAndSwap"); - return 0; -#endif -} - -inline Atomic64 NoBarrier_AtomicExchange(volatile Atomic64* ptr, - Atomic64 new_value) { -#if 0 // Not implemented - __asm__ __volatile__( - "movl (%2), %%ebx\n\t" // Move 64-bit new_value into - "movl 4(%2), %%ecx\n\t" // ecx:ebx - "0:\n\t" - "movl %1, %%eax\n\t" // Read contents of ptr into - "movl 4%1, %%edx\n\t" // edx:eax - "lock; cmpxchg8b %1\n\t" // Attempt cmpxchg; if *ptr - "jnz 0b\n\t" // is no longer edx:eax, loop - : "=A" (new_value) - : "m" (*ptr), - "r" (&new_value) - : "memory", "%ebx", "%ecx"); - return new_value; // Now it's the previous value. -#else - NotImplementedFatalError("NoBarrier_AtomicExchange"); - return 0; -#endif -} - -inline void NoBarrier_Store(volatile Atomic64* ptrValue, Atomic64 value) -{ - __asm { - movq mm0, value; // Use mmx reg for 64-bit atomic moves - mov eax, ptrValue; - movq [eax], mm0; - emms; // Empty mmx state to enable FP registers - } -} - -inline void Acquire_Store(volatile Atomic64* ptr, Atomic64 value) { - NoBarrier_AtomicExchange(ptr, value); - // acts as a barrier in this implementation -} - -inline void Release_Store(volatile Atomic64* ptr, Atomic64 value) { - NoBarrier_Store(ptr, value); -} - -inline Atomic64 NoBarrier_Load(volatile const Atomic64* ptrValue) -{ - Atomic64 value; - __asm { - mov eax, ptrValue; - movq mm0, [eax]; // Use mmx reg for 64-bit atomic moves - movq value, mm0; - emms; // Empty mmx state to enable FP registers - } - return value; -} - -inline Atomic64 Acquire_Load(volatile const Atomic64* ptr) { - Atomic64 value = NoBarrier_Load(ptr); - return value; -} - -inline Atomic64 Release_Load(volatile const Atomic64* ptr) { - MemoryBarrier(); - return NoBarrier_Load(ptr); -} - -#endif // defined(_WIN64) || defined(__MINGW64__) - - -inline Atomic64 Acquire_AtomicExchange(volatile Atomic64* ptr, - Atomic64 new_value) { - // FastInterlockedExchange has both acquire and release memory barriers. - return NoBarrier_AtomicExchange(ptr, new_value); -} - -inline Atomic64 Release_AtomicExchange(volatile Atomic64* ptr, - Atomic64 new_value) { - // FastInterlockedExchange has both acquire and release memory barriers. - return NoBarrier_AtomicExchange(ptr, new_value); -} - -inline Atomic64 Acquire_CompareAndSwap(volatile Atomic64* ptr, - Atomic64 old_value, - Atomic64 new_value) { - return NoBarrier_CompareAndSwap(ptr, old_value, new_value); -} - -inline Atomic64 Release_CompareAndSwap(volatile Atomic64* ptr, - Atomic64 old_value, - Atomic64 new_value) { - return NoBarrier_CompareAndSwap(ptr, old_value, new_value); -} - -} // namespace base::subtle -} // namespace base - -#endif // BASE_ATOMICOPS_INTERNALS_WINDOWS_H_ diff --git a/contrib/libtcmalloc/src/base/atomicops-internals-x86.cc b/contrib/libtcmalloc/src/base/atomicops-internals-x86.cc deleted file mode 100644 index c3391e78234..00000000000 --- a/contrib/libtcmalloc/src/base/atomicops-internals-x86.cc +++ /dev/null @@ -1,112 +0,0 @@ -// -*- Mode: C++; c-basic-offset: 2; indent-tabs-mode: nil -*- -/* Copyright (c) 2007, Google Inc. - * All rights reserved. - * - * Redistribution and use in source and binary forms, with or without - * modification, are permitted provided that the following conditions are - * met: - * - * * Redistributions of source code must retain the above copyright - * notice, this list of conditions and the following disclaimer. - * * Redistributions in binary form must reproduce the above - * copyright notice, this list of conditions and the following disclaimer - * in the documentation and/or other materials provided with the - * distribution. - * * Neither the name of Google Inc. nor the names of its - * contributors may be used to endorse or promote products derived from - * this software without specific prior written permission. - * - * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS - * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT - * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR - * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT - * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, - * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT - * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, - * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY - * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT - * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE - * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. - * - * --- - * This module gets enough CPU information to optimize the - * atomicops module on x86. - */ - -#include "base/atomicops.h" -#include "base/basictypes.h" -#include "base/googleinit.h" -#include "base/logging.h" -#include - -// This file only makes sense with atomicops-internals-x86.h -- it -// depends on structs that are defined in that file. If atomicops.h -// doesn't sub-include that file, then we aren't needed, and shouldn't -// try to do anything. -#ifdef BASE_ATOMICOPS_INTERNALS_X86_H_ - -// Inline cpuid instruction. In PIC compilations, %ebx contains the address -// of the global offset table. To avoid breaking such executables, this code -// must preserve that register's value across cpuid instructions. -#if defined(__i386__) -#define cpuid(a, b, c, d, inp) \ - asm ("mov %%ebx, %%edi\n" \ - "cpuid\n" \ - "xchg %%edi, %%ebx\n" \ - : "=a" (a), "=D" (b), "=c" (c), "=d" (d) : "a" (inp)) -#elif defined (__x86_64__) -#define cpuid(a, b, c, d, inp) \ - asm ("mov %%rbx, %%rdi\n" \ - "cpuid\n" \ - "xchg %%rdi, %%rbx\n" \ - : "=a" (a), "=D" (b), "=c" (c), "=d" (d) : "a" (inp)) -#endif - -#if defined(cpuid) // initialize the struct only on x86 - -// Set the flags so that code will run correctly and conservatively -// until InitGoogle() is called. -struct AtomicOps_x86CPUFeatureStruct AtomicOps_Internalx86CPUFeatures = { - false, // no SSE2 - false // no cmpxchg16b -}; - -// Initialize the AtomicOps_Internalx86CPUFeatures struct. -static void AtomicOps_Internalx86CPUFeaturesInit() { - uint32 eax; - uint32 ebx; - uint32 ecx; - uint32 edx; - - // Get vendor string (issue CPUID with eax = 0) - cpuid(eax, ebx, ecx, edx, 0); - char vendor[13]; - memcpy(vendor, &ebx, 4); - memcpy(vendor + 4, &edx, 4); - memcpy(vendor + 8, &ecx, 4); - vendor[12] = 0; - - // get feature flags in ecx/edx, and family/model in eax - cpuid(eax, ebx, ecx, edx, 1); - - int family = (eax >> 8) & 0xf; // family and model fields - int model = (eax >> 4) & 0xf; - if (family == 0xf) { // use extended family and model fields - family += (eax >> 20) & 0xff; - model += ((eax >> 16) & 0xf) << 4; - } - - // edx bit 26 is SSE2 which we use to tell use whether we can use mfence - AtomicOps_Internalx86CPUFeatures.has_sse2 = ((edx >> 26) & 1); - - // ecx bit 13 indicates whether the cmpxchg16b instruction is supported - AtomicOps_Internalx86CPUFeatures.has_cmpxchg16b = ((ecx >> 13) & 1); -} - -REGISTER_MODULE_INITIALIZER(atomicops_x86, { - AtomicOps_Internalx86CPUFeaturesInit(); -}); - -#endif - -#endif /* ifdef BASE_ATOMICOPS_INTERNALS_X86_H_ */ diff --git a/contrib/libtcmalloc/src/base/atomicops-internals-x86.h b/contrib/libtcmalloc/src/base/atomicops-internals-x86.h deleted file mode 100644 index e441ac7e673..00000000000 --- a/contrib/libtcmalloc/src/base/atomicops-internals-x86.h +++ /dev/null @@ -1,391 +0,0 @@ -// -*- Mode: C++; c-basic-offset: 2; indent-tabs-mode: nil -*- -/* Copyright (c) 2006, Google Inc. - * All rights reserved. - * - * Redistribution and use in source and binary forms, with or without - * modification, are permitted provided that the following conditions are - * met: - * - * * Redistributions of source code must retain the above copyright - * notice, this list of conditions and the following disclaimer. - * * Redistributions in binary form must reproduce the above - * copyright notice, this list of conditions and the following disclaimer - * in the documentation and/or other materials provided with the - * distribution. - * * Neither the name of Google Inc. nor the names of its - * contributors may be used to endorse or promote products derived from - * this software without specific prior written permission. - * - * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS - * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT - * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR - * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT - * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, - * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT - * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, - * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY - * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT - * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE - * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. - * - * --- - * Author: Sanjay Ghemawat - */ - -// Implementation of atomic operations for x86. This file should not -// be included directly. Clients should instead include -// "base/atomicops.h". - -#ifndef BASE_ATOMICOPS_INTERNALS_X86_H_ -#define BASE_ATOMICOPS_INTERNALS_X86_H_ -#include "base/basictypes.h" - -typedef int32_t Atomic32; -#define BASE_HAS_ATOMIC64 1 // Use only in tests and base/atomic* - - -// NOTE(vchen): x86 does not need to define AtomicWordCastType, because it -// already matches Atomic32 or Atomic64, depending on the platform. - - -// This struct is not part of the public API of this module; clients may not -// use it. -// Features of this x86. Values may not be correct before main() is run, -// but are set conservatively. -struct AtomicOps_x86CPUFeatureStruct { - bool has_sse2; // Processor has SSE2. - bool has_cmpxchg16b; // Processor supports cmpxchg16b instruction. -}; - -ATTRIBUTE_VISIBILITY_HIDDEN -extern struct AtomicOps_x86CPUFeatureStruct AtomicOps_Internalx86CPUFeatures; - - -#define ATOMICOPS_COMPILER_BARRIER() __asm__ __volatile__("" : : : "memory") - - -namespace base { -namespace subtle { - -typedef int64_t Atomic64; - -// 32-bit low-level operations on any platform. - -inline Atomic32 NoBarrier_CompareAndSwap(volatile Atomic32* ptr, - Atomic32 old_value, - Atomic32 new_value) { - Atomic32 prev; - __asm__ __volatile__("lock; cmpxchgl %1,%2" - : "=a" (prev) - : "q" (new_value), "m" (*ptr), "0" (old_value) - : "memory"); - return prev; -} - -inline Atomic32 NoBarrier_AtomicExchange(volatile Atomic32* ptr, - Atomic32 new_value) { - __asm__ __volatile__("xchgl %1,%0" // The lock prefix is implicit for xchg. - : "=r" (new_value) - : "m" (*ptr), "0" (new_value) - : "memory"); - return new_value; // Now it's the previous value. -} - -inline Atomic32 Acquire_AtomicExchange(volatile Atomic32* ptr, - Atomic32 new_value) { - Atomic32 old_val = NoBarrier_AtomicExchange(ptr, new_value); - return old_val; -} - -inline Atomic32 Release_AtomicExchange(volatile Atomic32* ptr, - Atomic32 new_value) { - // xchgl already has release memory barrier semantics. - return NoBarrier_AtomicExchange(ptr, new_value); -} - -inline Atomic32 Acquire_CompareAndSwap(volatile Atomic32* ptr, - Atomic32 old_value, - Atomic32 new_value) { - Atomic32 x = NoBarrier_CompareAndSwap(ptr, old_value, new_value); - return x; -} - -inline Atomic32 Release_CompareAndSwap(volatile Atomic32* ptr, - Atomic32 old_value, - Atomic32 new_value) { - return NoBarrier_CompareAndSwap(ptr, old_value, new_value); -} - -inline void NoBarrier_Store(volatile Atomic32* ptr, Atomic32 value) { - *ptr = value; -} - -#if defined(__x86_64__) - -// 64-bit implementations of memory barrier can be simpler, because it -// "mfence" is guaranteed to exist. -inline void MemoryBarrier() { - __asm__ __volatile__("mfence" : : : "memory"); -} - -inline void Acquire_Store(volatile Atomic32* ptr, Atomic32 value) { - *ptr = value; - MemoryBarrier(); -} - -#else - -inline void MemoryBarrier() { - if (AtomicOps_Internalx86CPUFeatures.has_sse2) { - __asm__ __volatile__("mfence" : : : "memory"); - } else { // mfence is faster but not present on PIII - Atomic32 x = 0; - Acquire_AtomicExchange(&x, 0); - } -} - -inline void Acquire_Store(volatile Atomic32* ptr, Atomic32 value) { - if (AtomicOps_Internalx86CPUFeatures.has_sse2) { - *ptr = value; - __asm__ __volatile__("mfence" : : : "memory"); - } else { - Acquire_AtomicExchange(ptr, value); - } -} -#endif - -inline void Release_Store(volatile Atomic32* ptr, Atomic32 value) { - ATOMICOPS_COMPILER_BARRIER(); - *ptr = value; // An x86 store acts as a release barrier. - // See comments in Atomic64 version of Release_Store(), below. -} - -inline Atomic32 NoBarrier_Load(volatile const Atomic32* ptr) { - return *ptr; -} - -inline Atomic32 Acquire_Load(volatile const Atomic32* ptr) { - Atomic32 value = *ptr; // An x86 load acts as a acquire barrier. - // See comments in Atomic64 version of Release_Store(), below. - ATOMICOPS_COMPILER_BARRIER(); - return value; -} - -inline Atomic32 Release_Load(volatile const Atomic32* ptr) { - MemoryBarrier(); - return *ptr; -} - -#if defined(__x86_64__) - -// 64-bit low-level operations on 64-bit platform. - -inline Atomic64 NoBarrier_CompareAndSwap(volatile Atomic64* ptr, - Atomic64 old_value, - Atomic64 new_value) { - Atomic64 prev; - __asm__ __volatile__("lock; cmpxchgq %1,%2" - : "=a" (prev) - : "q" (new_value), "m" (*ptr), "0" (old_value) - : "memory"); - return prev; -} - -inline Atomic64 NoBarrier_AtomicExchange(volatile Atomic64* ptr, - Atomic64 new_value) { - __asm__ __volatile__("xchgq %1,%0" // The lock prefix is implicit for xchg. - : "=r" (new_value) - : "m" (*ptr), "0" (new_value) - : "memory"); - return new_value; // Now it's the previous value. -} - -inline Atomic64 Acquire_AtomicExchange(volatile Atomic64* ptr, - Atomic64 new_value) { - Atomic64 old_val = NoBarrier_AtomicExchange(ptr, new_value); - return old_val; -} - -inline Atomic64 Release_AtomicExchange(volatile Atomic64* ptr, - Atomic64 new_value) { - // xchgq already has release memory barrier semantics. - return NoBarrier_AtomicExchange(ptr, new_value); -} - -inline void NoBarrier_Store(volatile Atomic64* ptr, Atomic64 value) { - *ptr = value; -} - -inline void Acquire_Store(volatile Atomic64* ptr, Atomic64 value) { - *ptr = value; - MemoryBarrier(); -} - -inline void Release_Store(volatile Atomic64* ptr, Atomic64 value) { - ATOMICOPS_COMPILER_BARRIER(); - - *ptr = value; // An x86 store acts as a release barrier - // for current AMD/Intel chips as of Jan 2008. - // See also Acquire_Load(), below. - - // When new chips come out, check: - // IA-32 Intel Architecture Software Developer's Manual, Volume 3: - // System Programming Guide, Chatper 7: Multiple-processor management, - // Section 7.2, Memory Ordering. - // Last seen at: - // http://developer.intel.com/design/pentium4/manuals/index_new.htm - // - // x86 stores/loads fail to act as barriers for a few instructions (clflush - // maskmovdqu maskmovq movntdq movnti movntpd movntps movntq) but these are - // not generated by the compiler, and are rare. Users of these instructions - // need to know about cache behaviour in any case since all of these involve - // either flushing cache lines or non-temporal cache hints. -} - -inline Atomic64 NoBarrier_Load(volatile const Atomic64* ptr) { - return *ptr; -} - -inline Atomic64 Acquire_Load(volatile const Atomic64* ptr) { - Atomic64 value = *ptr; // An x86 load acts as a acquire barrier, - // for current AMD/Intel chips as of Jan 2008. - // See also Release_Store(), above. - ATOMICOPS_COMPILER_BARRIER(); - return value; -} - -inline Atomic64 Release_Load(volatile const Atomic64* ptr) { - MemoryBarrier(); - return *ptr; -} - -#else // defined(__x86_64__) - -// 64-bit low-level operations on 32-bit platform. - -#if !((__GNUC__ > 4) || (__GNUC__ == 4 && __GNUC_MINOR__ >= 1)) -// For compilers older than gcc 4.1, we use inline asm. -// -// Potential pitfalls: -// -// 1. %ebx points to Global offset table (GOT) with -fPIC. -// We need to preserve this register. -// 2. When explicit registers are used in inline asm, the -// compiler may not be aware of it and might try to reuse -// the same register for another argument which has constraints -// that allow it ("r" for example). - -inline Atomic64 __sync_val_compare_and_swap(volatile Atomic64* ptr, - Atomic64 old_value, - Atomic64 new_value) { - Atomic64 prev; - __asm__ __volatile__("push %%ebx\n\t" - "movl (%3), %%ebx\n\t" // Move 64-bit new_value into - "movl 4(%3), %%ecx\n\t" // ecx:ebx - "lock; cmpxchg8b (%1)\n\t"// If edx:eax (old_value) same - "pop %%ebx\n\t" - : "=A" (prev) // as contents of ptr: - : "D" (ptr), // ecx:ebx => ptr - "0" (old_value), // else: - "S" (&new_value) // old *ptr => edx:eax - : "memory", "%ecx"); - return prev; -} -#endif // Compiler < gcc-4.1 - -inline Atomic64 NoBarrier_CompareAndSwap(volatile Atomic64* ptr, - Atomic64 old_val, - Atomic64 new_val) { - return __sync_val_compare_and_swap(ptr, old_val, new_val); -} - -inline Atomic64 NoBarrier_AtomicExchange(volatile Atomic64* ptr, - Atomic64 new_val) { - Atomic64 old_val; - - do { - old_val = *ptr; - } while (__sync_val_compare_and_swap(ptr, old_val, new_val) != old_val); - - return old_val; -} - -inline Atomic64 Acquire_AtomicExchange(volatile Atomic64* ptr, - Atomic64 new_val) { - Atomic64 old_val = NoBarrier_AtomicExchange(ptr, new_val); - return old_val; -} - -inline Atomic64 Release_AtomicExchange(volatile Atomic64* ptr, - Atomic64 new_val) { - return NoBarrier_AtomicExchange(ptr, new_val); -} - -inline void NoBarrier_Store(volatile Atomic64* ptr, Atomic64 value) { - __asm__ __volatile__("movq %1, %%mm0\n\t" // Use mmx reg for 64-bit atomic - "movq %%mm0, %0\n\t" // moves (ptr could be read-only) - "emms\n\t" // Empty mmx state/Reset FP regs - : "=m" (*ptr) - : "m" (value) - : // mark the FP stack and mmx registers as clobbered - "st", "st(1)", "st(2)", "st(3)", "st(4)", - "st(5)", "st(6)", "st(7)", "mm0", "mm1", - "mm2", "mm3", "mm4", "mm5", "mm6", "mm7"); -} - -inline void Acquire_Store(volatile Atomic64* ptr, Atomic64 value) { - NoBarrier_Store(ptr, value); - MemoryBarrier(); -} - -inline void Release_Store(volatile Atomic64* ptr, Atomic64 value) { - ATOMICOPS_COMPILER_BARRIER(); - NoBarrier_Store(ptr, value); -} - -inline Atomic64 NoBarrier_Load(volatile const Atomic64* ptr) { - Atomic64 value; - __asm__ __volatile__("movq %1, %%mm0\n\t" // Use mmx reg for 64-bit atomic - "movq %%mm0, %0\n\t" // moves (ptr could be read-only) - "emms\n\t" // Empty mmx state/Reset FP regs - : "=m" (value) - : "m" (*ptr) - : // mark the FP stack and mmx registers as clobbered - "st", "st(1)", "st(2)", "st(3)", "st(4)", - "st(5)", "st(6)", "st(7)", "mm0", "mm1", - "mm2", "mm3", "mm4", "mm5", "mm6", "mm7"); - return value; -} - -inline Atomic64 Acquire_Load(volatile const Atomic64* ptr) { - Atomic64 value = NoBarrier_Load(ptr); - ATOMICOPS_COMPILER_BARRIER(); - return value; -} - -inline Atomic64 Release_Load(volatile const Atomic64* ptr) { - MemoryBarrier(); - return NoBarrier_Load(ptr); -} - -#endif // defined(__x86_64__) - -inline Atomic64 Acquire_CompareAndSwap(volatile Atomic64* ptr, - Atomic64 old_value, - Atomic64 new_value) { - Atomic64 x = NoBarrier_CompareAndSwap(ptr, old_value, new_value); - return x; -} - -inline Atomic64 Release_CompareAndSwap(volatile Atomic64* ptr, - Atomic64 old_value, - Atomic64 new_value) { - return NoBarrier_CompareAndSwap(ptr, old_value, new_value); -} - -} // namespace base::subtle -} // namespace base - -#undef ATOMICOPS_COMPILER_BARRIER - -#endif // BASE_ATOMICOPS_INTERNALS_X86_H_ diff --git a/contrib/libtcmalloc/src/base/atomicops.h b/contrib/libtcmalloc/src/base/atomicops.h deleted file mode 100644 index 46a4b9bb7a2..00000000000 --- a/contrib/libtcmalloc/src/base/atomicops.h +++ /dev/null @@ -1,399 +0,0 @@ -// -*- Mode: C++; c-basic-offset: 2; indent-tabs-mode: nil -*- -/* Copyright (c) 2006, Google Inc. - * All rights reserved. - * - * Redistribution and use in source and binary forms, with or without - * modification, are permitted provided that the following conditions are - * met: - * - * * Redistributions of source code must retain the above copyright - * notice, this list of conditions and the following disclaimer. - * * Redistributions in binary form must reproduce the above - * copyright notice, this list of conditions and the following disclaimer - * in the documentation and/or other materials provided with the - * distribution. - * * Neither the name of Google Inc. nor the names of its - * contributors may be used to endorse or promote products derived from - * this software without specific prior written permission. - * - * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS - * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT - * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR - * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT - * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, - * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT - * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, - * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY - * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT - * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE - * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. - * - * --- - * Author: Sanjay Ghemawat - */ - -// For atomic operations on statistics counters, see atomic_stats_counter.h. -// For atomic operations on sequence numbers, see atomic_sequence_num.h. -// For atomic operations on reference counts, see atomic_refcount.h. - -// Some fast atomic operations -- typically with machine-dependent -// implementations. This file may need editing as Google code is -// ported to different architectures. - -// The routines exported by this module are subtle. If you use them, even if -// you get the code right, it will depend on careful reasoning about atomicity -// and memory ordering; it will be less readable, and harder to maintain. If -// you plan to use these routines, you should have a good reason, such as solid -// evidence that performance would otherwise suffer, or there being no -// alternative. You should assume only properties explicitly guaranteed by the -// specifications in this file. You are almost certainly _not_ writing code -// just for the x86; if you assume x86 semantics, x86 hardware bugs and -// implementations on other archtectures will cause your code to break. If you -// do not know what you are doing, avoid these routines, and use a Mutex. -// -// These following lower-level operations are typically useful only to people -// implementing higher-level synchronization operations like spinlocks, -// mutexes, and condition-variables. They combine CompareAndSwap(), a load, or -// a store with appropriate memory-ordering instructions. "Acquire" operations -// ensure that no later memory access can be reordered ahead of the operation. -// "Release" operations ensure that no previous memory access can be reordered -// after the operation. "Barrier" operations have both "Acquire" and "Release" -// semantics. A MemoryBarrier() has "Barrier" semantics, but does no memory -// access. -// -// It is incorrect to make direct assignments to/from an atomic variable. -// You should use one of the Load or Store routines. The NoBarrier -// versions are provided when no barriers are needed: -// NoBarrier_Store() -// NoBarrier_Load() -// Although there are currently no compiler enforcement, you are encouraged -// to use these. Moreover, if you choose to use base::subtle::Atomic64 type, -// you MUST use one of the Load or Store routines to get correct behavior -// on 32-bit platforms. -// -// The intent is eventually to put all of these routines in namespace -// base::subtle - -#ifndef THREAD_ATOMICOPS_H_ -#define THREAD_ATOMICOPS_H_ - -#include "../config.h" -#ifdef HAVE_STDINT_H -#include -#endif - -// ------------------------------------------------------------------------ -// Include the platform specific implementations of the types -// and operations listed below. Implementations are to provide Atomic32 -// and Atomic64 operations. If there is a mismatch between intptr_t and -// the Atomic32 or Atomic64 types for a platform, the platform-specific header -// should define the macro, AtomicWordCastType in a clause similar to the -// following: -// #if ...pointers are 64 bits... -// # define AtomicWordCastType base::subtle::Atomic64 -// #else -// # define AtomicWordCastType Atomic32 -// #endif -// TODO(csilvers): figure out ARCH_PIII/ARCH_K8 (perhaps via ./configure?) -// ------------------------------------------------------------------------ - -#include "base/arm_instruction_set_select.h" -#define GCC_VERSION (__GNUC__ * 10000 \ - + __GNUC_MINOR__ * 100 \ - + __GNUC_PATCHLEVEL__) - -#define CLANG_VERSION (__clang_major__ * 10000 \ - + __clang_minor__ * 100 \ - + __clang_patchlevel__) - -#if defined(TCMALLOC_PREFER_GCC_ATOMICS) && defined(__GNUC__) && GCC_VERSION >= 40700 -#include "base/atomicops-internals-gcc.h" -#elif defined(TCMALLOC_PREFER_GCC_ATOMICS) && defined(__clang__) && CLANG_VERSION >= 30400 -#include "base/atomicops-internals-gcc.h" -#elif defined(__MACH__) && defined(__APPLE__) -#include "base/atomicops-internals-macosx.h" -#elif defined(__GNUC__) && defined(ARMV6) -#include "base/atomicops-internals-arm-v6plus.h" -#elif defined(ARMV3) -#include "base/atomicops-internals-arm-generic.h" -#elif defined(__GNUC__) && (defined(__i386) || defined(__x86_64__)) -#include "base/atomicops-internals-x86.h" -#elif defined(_WIN32) -#include "base/atomicops-internals-windows.h" -#elif defined(__linux__) && defined(__PPC__) -#include "base/atomicops-internals-linuxppc.h" -#elif defined(__GNUC__) && defined(__mips__) -#include "base/atomicops-internals-mips.h" -#elif defined(__GNUC__) && GCC_VERSION >= 40700 -#include "base/atomicops-internals-gcc.h" -#elif defined(__clang__) && CLANG_VERSION >= 30400 -#include "base/atomicops-internals-gcc.h" -#else -#error You need to implement atomic operations for this architecture -#endif - -// Signed type that can hold a pointer and supports the atomic ops below, as -// well as atomic loads and stores. Instances must be naturally-aligned. -typedef intptr_t AtomicWord; - -#ifdef AtomicWordCastType -// ------------------------------------------------------------------------ -// This section is needed only when explicit type casting is required to -// cast AtomicWord to one of the basic atomic types (Atomic64 or Atomic32). -// It also serves to document the AtomicWord interface. -// ------------------------------------------------------------------------ - -namespace base { -namespace subtle { - -// Atomically execute: -// result = *ptr; -// if (*ptr == old_value) -// *ptr = new_value; -// return result; -// -// I.e., replace "*ptr" with "new_value" if "*ptr" used to be "old_value". -// Always return the old value of "*ptr" -// -// This routine implies no memory barriers. -inline AtomicWord NoBarrier_CompareAndSwap(volatile AtomicWord* ptr, - AtomicWord old_value, - AtomicWord new_value) { - return NoBarrier_CompareAndSwap( - reinterpret_cast(ptr), - old_value, new_value); -} - -// Atomically store new_value into *ptr, returning the previous value held in -// *ptr. This routine implies no memory barriers. -inline AtomicWord NoBarrier_AtomicExchange(volatile AtomicWord* ptr, - AtomicWord new_value) { - return NoBarrier_AtomicExchange( - reinterpret_cast(ptr), new_value); -} - -inline AtomicWord Acquire_AtomicExchange(volatile AtomicWord* ptr, - AtomicWord new_value) { - return Acquire_AtomicExchange( - reinterpret_cast(ptr), new_value); -} - -inline AtomicWord Release_AtomicExchange(volatile AtomicWord* ptr, - AtomicWord new_value) { - return Release_AtomicExchange( - reinterpret_cast(ptr), new_value); -} - -inline AtomicWord Acquire_CompareAndSwap(volatile AtomicWord* ptr, - AtomicWord old_value, - AtomicWord new_value) { - return base::subtle::Acquire_CompareAndSwap( - reinterpret_cast(ptr), - old_value, new_value); -} - -inline AtomicWord Release_CompareAndSwap(volatile AtomicWord* ptr, - AtomicWord old_value, - AtomicWord new_value) { - return base::subtle::Release_CompareAndSwap( - reinterpret_cast(ptr), - old_value, new_value); -} - -inline void NoBarrier_Store(volatile AtomicWord *ptr, AtomicWord value) { - NoBarrier_Store( - reinterpret_cast(ptr), value); -} - -inline void Acquire_Store(volatile AtomicWord* ptr, AtomicWord value) { - return base::subtle::Acquire_Store( - reinterpret_cast(ptr), value); -} - -inline void Release_Store(volatile AtomicWord* ptr, AtomicWord value) { - return base::subtle::Release_Store( - reinterpret_cast(ptr), value); -} - -inline AtomicWord NoBarrier_Load(volatile const AtomicWord *ptr) { - return NoBarrier_Load( - reinterpret_cast(ptr)); -} - -inline AtomicWord Acquire_Load(volatile const AtomicWord* ptr) { - return base::subtle::Acquire_Load( - reinterpret_cast(ptr)); -} - -inline AtomicWord Release_Load(volatile const AtomicWord* ptr) { - return base::subtle::Release_Load( - reinterpret_cast(ptr)); -} - -} // namespace base::subtle -} // namespace base -#endif // AtomicWordCastType - -// ------------------------------------------------------------------------ -// Commented out type definitions and method declarations for documentation -// of the interface provided by this module. -// ------------------------------------------------------------------------ - -#if 0 - -// Signed 32-bit type that supports the atomic ops below, as well as atomic -// loads and stores. Instances must be naturally aligned. This type differs -// from AtomicWord in 64-bit binaries where AtomicWord is 64-bits. -typedef int32_t Atomic32; - -// Corresponding operations on Atomic32 -namespace base { -namespace subtle { - -// Signed 64-bit type that supports the atomic ops below, as well as atomic -// loads and stores. Instances must be naturally aligned. This type differs -// from AtomicWord in 32-bit binaries where AtomicWord is 32-bits. -typedef int64_t Atomic64; - -Atomic32 NoBarrier_CompareAndSwap(volatile Atomic32* ptr, - Atomic32 old_value, - Atomic32 new_value); -Atomic32 NoBarrier_AtomicExchange(volatile Atomic32* ptr, Atomic32 new_value); -Atomic32 Acquire_AtomicExchange(volatile Atomic32* ptr, Atomic32 new_value); -Atomic32 Release_AtomicExchange(volatile Atomic32* ptr, Atomic32 new_value); -Atomic32 Acquire_CompareAndSwap(volatile Atomic32* ptr, - Atomic32 old_value, - Atomic32 new_value); -Atomic32 Release_CompareAndSwap(volatile Atomic32* ptr, - Atomic32 old_value, - Atomic32 new_value); -void NoBarrier_Store(volatile Atomic32* ptr, Atomic32 value); -void Acquire_Store(volatile Atomic32* ptr, Atomic32 value); -void Release_Store(volatile Atomic32* ptr, Atomic32 value); -Atomic32 NoBarrier_Load(volatile const Atomic32* ptr); -Atomic32 Acquire_Load(volatile const Atomic32* ptr); -Atomic32 Release_Load(volatile const Atomic32* ptr); - -// Corresponding operations on Atomic64 -Atomic64 NoBarrier_CompareAndSwap(volatile Atomic64* ptr, - Atomic64 old_value, - Atomic64 new_value); -Atomic64 NoBarrier_AtomicExchange(volatile Atomic64* ptr, Atomic64 new_value); -Atomic64 Acquire_AtomicExchange(volatile Atomic64* ptr, Atomic64 new_value); -Atomic64 Release_AtomicExchange(volatile Atomic64* ptr, Atomic64 new_value); - -Atomic64 Acquire_CompareAndSwap(volatile Atomic64* ptr, - Atomic64 old_value, - Atomic64 new_value); -Atomic64 Release_CompareAndSwap(volatile Atomic64* ptr, - Atomic64 old_value, - Atomic64 new_value); -void NoBarrier_Store(volatile Atomic64* ptr, Atomic64 value); -void Acquire_Store(volatile Atomic64* ptr, Atomic64 value); -void Release_Store(volatile Atomic64* ptr, Atomic64 value); -Atomic64 NoBarrier_Load(volatile const Atomic64* ptr); -Atomic64 Acquire_Load(volatile const Atomic64* ptr); -Atomic64 Release_Load(volatile const Atomic64* ptr); -} // namespace base::subtle -} // namespace base - -void MemoryBarrier(); - -#endif // 0 - - -// ------------------------------------------------------------------------ -// The following are to be deprecated when all uses have been changed to -// use the base::subtle namespace. -// ------------------------------------------------------------------------ - -#ifdef AtomicWordCastType -// AtomicWord versions to be deprecated -inline AtomicWord Acquire_CompareAndSwap(volatile AtomicWord* ptr, - AtomicWord old_value, - AtomicWord new_value) { - return base::subtle::Acquire_CompareAndSwap(ptr, old_value, new_value); -} - -inline AtomicWord Release_CompareAndSwap(volatile AtomicWord* ptr, - AtomicWord old_value, - AtomicWord new_value) { - return base::subtle::Release_CompareAndSwap(ptr, old_value, new_value); -} - -inline void Acquire_Store(volatile AtomicWord* ptr, AtomicWord value) { - return base::subtle::Acquire_Store(ptr, value); -} - -inline void Release_Store(volatile AtomicWord* ptr, AtomicWord value) { - return base::subtle::Release_Store(ptr, value); -} - -inline AtomicWord Acquire_Load(volatile const AtomicWord* ptr) { - return base::subtle::Acquire_Load(ptr); -} - -inline AtomicWord Release_Load(volatile const AtomicWord* ptr) { - return base::subtle::Release_Load(ptr); -} -#endif // AtomicWordCastType - -// 32-bit Acquire/Release operations to be deprecated. - -inline Atomic32 Acquire_CompareAndSwap(volatile Atomic32* ptr, - Atomic32 old_value, - Atomic32 new_value) { - return base::subtle::Acquire_CompareAndSwap(ptr, old_value, new_value); -} -inline Atomic32 Release_CompareAndSwap(volatile Atomic32* ptr, - Atomic32 old_value, - Atomic32 new_value) { - return base::subtle::Release_CompareAndSwap(ptr, old_value, new_value); -} -inline void Acquire_Store(volatile Atomic32* ptr, Atomic32 value) { - base::subtle::Acquire_Store(ptr, value); -} -inline void Release_Store(volatile Atomic32* ptr, Atomic32 value) { - return base::subtle::Release_Store(ptr, value); -} -inline Atomic32 Acquire_Load(volatile const Atomic32* ptr) { - return base::subtle::Acquire_Load(ptr); -} -inline Atomic32 Release_Load(volatile const Atomic32* ptr) { - return base::subtle::Release_Load(ptr); -} - -#ifdef BASE_HAS_ATOMIC64 - -// 64-bit Acquire/Release operations to be deprecated. - -inline base::subtle::Atomic64 Acquire_CompareAndSwap( - volatile base::subtle::Atomic64* ptr, - base::subtle::Atomic64 old_value, base::subtle::Atomic64 new_value) { - return base::subtle::Acquire_CompareAndSwap(ptr, old_value, new_value); -} -inline base::subtle::Atomic64 Release_CompareAndSwap( - volatile base::subtle::Atomic64* ptr, - base::subtle::Atomic64 old_value, base::subtle::Atomic64 new_value) { - return base::subtle::Release_CompareAndSwap(ptr, old_value, new_value); -} -inline void Acquire_Store( - volatile base::subtle::Atomic64* ptr, base::subtle::Atomic64 value) { - base::subtle::Acquire_Store(ptr, value); -} -inline void Release_Store( - volatile base::subtle::Atomic64* ptr, base::subtle::Atomic64 value) { - return base::subtle::Release_Store(ptr, value); -} -inline base::subtle::Atomic64 Acquire_Load( - volatile const base::subtle::Atomic64* ptr) { - return base::subtle::Acquire_Load(ptr); -} -inline base::subtle::Atomic64 Release_Load( - volatile const base::subtle::Atomic64* ptr) { - return base::subtle::Release_Load(ptr); -} - -#endif // BASE_HAS_ATOMIC64 - -#endif // THREAD_ATOMICOPS_H_ diff --git a/contrib/libtcmalloc/src/base/basictypes.h b/contrib/libtcmalloc/src/base/basictypes.h deleted file mode 100644 index a81d0466c27..00000000000 --- a/contrib/libtcmalloc/src/base/basictypes.h +++ /dev/null @@ -1,408 +0,0 @@ -// -*- Mode: C++; c-basic-offset: 2; indent-tabs-mode: nil -*- -// Copyright (c) 2005, Google Inc. -// All rights reserved. -// -// Redistribution and use in source and binary forms, with or without -// modification, are permitted provided that the following conditions are -// met: -// -// * Redistributions of source code must retain the above copyright -// notice, this list of conditions and the following disclaimer. -// * Redistributions in binary form must reproduce the above -// copyright notice, this list of conditions and the following disclaimer -// in the documentation and/or other materials provided with the -// distribution. -// * Neither the name of Google Inc. nor the names of its -// contributors may be used to endorse or promote products derived from -// this software without specific prior written permission. -// -// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS -// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT -// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR -// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT -// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, -// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT -// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, -// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY -// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT -// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE -// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. - -#ifndef _BASICTYPES_H_ -#define _BASICTYPES_H_ - -#include "../config.h" -#include // for memcpy() -#ifdef HAVE_INTTYPES_H -#include // gets us PRId64, etc -#endif - -// To use this in an autoconf setting, make sure you run the following -// autoconf macros: -// AC_HEADER_STDC /* for stdint_h and inttypes_h */ -// AC_CHECK_TYPES([__int64]) /* defined in some windows platforms */ - -#ifdef HAVE_INTTYPES_H -#include // uint16_t might be here; PRId64 too. -#endif -#ifdef HAVE_STDINT_H -#include // to get uint16_t (ISO naming madness) -#endif -#include // our last best hope for uint16_t - -// Standard typedefs -// All Google code is compiled with -funsigned-char to make "char" -// unsigned. Google code therefore doesn't need a "uchar" type. -// TODO(csilvers): how do we make sure unsigned-char works on non-gcc systems? -typedef signed char schar; -typedef int8_t int8; -typedef int16_t int16; -typedef int32_t int32; -typedef int64_t int64; - -// NOTE: unsigned types are DANGEROUS in loops and other arithmetical -// places. Use the signed types unless your variable represents a bit -// pattern (eg a hash value) or you really need the extra bit. Do NOT -// use 'unsigned' to express "this value should always be positive"; -// use assertions for this. - -typedef uint8_t uint8; -typedef uint16_t uint16; -typedef uint32_t uint32; -typedef uint64_t uint64; - -const uint16 kuint16max = ( (uint16) 0xFFFF); -const uint32 kuint32max = ( (uint32) 0xFFFFFFFF); -const uint64 kuint64max = ( (((uint64) kuint32max) << 32) | kuint32max ); - -const int8 kint8max = ( ( int8) 0x7F); -const int16 kint16max = ( ( int16) 0x7FFF); -const int32 kint32max = ( ( int32) 0x7FFFFFFF); -const int64 kint64max = ( ((( int64) kint32max) << 32) | kuint32max ); - -const int8 kint8min = ( ( int8) 0x80); -const int16 kint16min = ( ( int16) 0x8000); -const int32 kint32min = ( ( int32) 0x80000000); -const int64 kint64min = ( (((uint64) kint32min) << 32) | 0 ); - -// Define the "portable" printf and scanf macros, if they're not -// already there (via the inttypes.h we #included above, hopefully). -// Mostly it's old systems that don't support inttypes.h, so we assume -// they're 32 bit. -#ifndef PRIx64 -#define PRIx64 "llx" -#endif -#ifndef SCNx64 -#define SCNx64 "llx" -#endif -#ifndef PRId64 -#define PRId64 "lld" -#endif -#ifndef SCNd64 -#define SCNd64 "lld" -#endif -#ifndef PRIu64 -#define PRIu64 "llu" -#endif -#ifndef PRIxPTR -#define PRIxPTR "lx" -#endif - -// Also allow for printing of a pthread_t. -#define GPRIuPTHREAD "lu" -#define GPRIxPTHREAD "lx" -#if defined(__CYGWIN__) || defined(__CYGWIN32__) || defined(__APPLE__) || defined(__FreeBSD__) -#define PRINTABLE_PTHREAD(pthreadt) reinterpret_cast(pthreadt) -#else -#define PRINTABLE_PTHREAD(pthreadt) pthreadt -#endif - -// A macro to disallow the evil copy constructor and operator= functions -// This should be used in the private: declarations for a class -#define DISALLOW_EVIL_CONSTRUCTORS(TypeName) \ - TypeName(const TypeName&); \ - void operator=(const TypeName&) - -// An alternate name that leaves out the moral judgment... :-) -#define DISALLOW_COPY_AND_ASSIGN(TypeName) DISALLOW_EVIL_CONSTRUCTORS(TypeName) - -// The COMPILE_ASSERT macro can be used to verify that a compile time -// expression is true. For example, you could use it to verify the -// size of a static array: -// -// COMPILE_ASSERT(sizeof(num_content_type_names) == sizeof(int), -// content_type_names_incorrect_size); -// -// or to make sure a struct is smaller than a certain size: -// -// COMPILE_ASSERT(sizeof(foo) < 128, foo_too_large); -// -// The second argument to the macro is the name of the variable. If -// the expression is false, most compilers will issue a warning/error -// containing the name of the variable. -// -// Implementation details of COMPILE_ASSERT: -// -// - COMPILE_ASSERT works by defining an array type that has -1 -// elements (and thus is invalid) when the expression is false. -// -// - The simpler definition -// -// #define COMPILE_ASSERT(expr, msg) typedef char msg[(expr) ? 1 : -1] -// -// does not work, as gcc supports variable-length arrays whose sizes -// are determined at run-time (this is gcc's extension and not part -// of the C++ standard). As a result, gcc fails to reject the -// following code with the simple definition: -// -// int foo; -// COMPILE_ASSERT(foo, msg); // not supposed to compile as foo is -// // not a compile-time constant. -// -// - By using the type CompileAssert<(bool(expr))>, we ensures that -// expr is a compile-time constant. (Template arguments must be -// determined at compile-time.) -// -// - The outter parentheses in CompileAssert<(bool(expr))> are necessary -// to work around a bug in gcc 3.4.4 and 4.0.1. If we had written -// -// CompileAssert -// -// instead, these compilers will refuse to compile -// -// COMPILE_ASSERT(5 > 0, some_message); -// -// (They seem to think the ">" in "5 > 0" marks the end of the -// template argument list.) -// -// - The array size is (bool(expr) ? 1 : -1), instead of simply -// -// ((expr) ? 1 : -1). -// -// This is to avoid running into a bug in MS VC 7.1, which -// causes ((0.0) ? 1 : -1) to incorrectly evaluate to 1. - -template -struct CompileAssert { -}; - -#ifdef HAVE___ATTRIBUTE__ -# define ATTRIBUTE_UNUSED __attribute__((unused)) -#else -# define ATTRIBUTE_UNUSED -#endif - -#if defined(HAVE___ATTRIBUTE__) && defined(HAVE_TLS) -#define ATTR_INITIAL_EXEC __attribute__ ((tls_model ("initial-exec"))) -#else -#define ATTR_INITIAL_EXEC -#endif - -#define COMPILE_ASSERT(expr, msg) \ - typedef CompileAssert<(bool(expr))> msg[bool(expr) ? 1 : -1] ATTRIBUTE_UNUSED - -#define arraysize(a) (sizeof(a) / sizeof(*(a))) - -#define OFFSETOF_MEMBER(strct, field) \ - (reinterpret_cast(&reinterpret_cast(16)->field) - \ - reinterpret_cast(16)) - -// bit_cast implements the equivalent of -// "*reinterpret_cast(&source)". -// -// The reinterpret_cast method would produce undefined behavior -// according to ISO C++ specification section 3.10 -15 -. -// bit_cast<> calls memcpy() which is blessed by the standard, -// especially by the example in section 3.9. -// -// Fortunately memcpy() is very fast. In optimized mode, with a -// constant size, gcc 2.95.3, gcc 4.0.1, and msvc 7.1 produce inline -// code with the minimal amount of data movement. On a 32-bit system, -// memcpy(d,s,4) compiles to one load and one store, and memcpy(d,s,8) -// compiles to two loads and two stores. - -template -inline Dest bit_cast(const Source& source) { - COMPILE_ASSERT(sizeof(Dest) == sizeof(Source), bitcasting_unequal_sizes); - Dest dest; - memcpy(&dest, &source, sizeof(dest)); - return dest; -} - -// bit_store implements the equivalent of -// "dest = *reinterpret_cast(&source)". -// -// This prevents undefined behavior when the dest pointer is unaligned. -template -inline void bit_store(Dest *dest, const Source *source) { - COMPILE_ASSERT(sizeof(Dest) == sizeof(Source), bitcasting_unequal_sizes); - memcpy(dest, source, sizeof(Dest)); -} - -#ifdef HAVE___ATTRIBUTE__ -# define ATTRIBUTE_WEAK __attribute__((weak)) -# define ATTRIBUTE_NOINLINE __attribute__((noinline)) -#else -# define ATTRIBUTE_WEAK -# define ATTRIBUTE_NOINLINE -#endif - -#if defined(HAVE___ATTRIBUTE__) && defined(__ELF__) -# define ATTRIBUTE_VISIBILITY_HIDDEN __attribute__((visibility("hidden"))) -#else -# define ATTRIBUTE_VISIBILITY_HIDDEN -#endif - -// Section attributes are supported for both ELF and Mach-O, but in -// very different ways. Here's the API we provide: -// 1) ATTRIBUTE_SECTION: put this with the declaration of all functions -// you want to be in the same linker section -// 2) DEFINE_ATTRIBUTE_SECTION_VARS: must be called once per unique -// name. You want to make sure this is executed before any -// DECLARE_ATTRIBUTE_SECTION_VARS; the easiest way is to put them -// in the same .cc file. Put this call at the global level. -// 3) INIT_ATTRIBUTE_SECTION_VARS: you can scatter calls to this in -// multiple places to help ensure execution before any -// DECLARE_ATTRIBUTE_SECTION_VARS. You must have at least one -// DEFINE, but you can have many INITs. Put each in its own scope. -// 4) DECLARE_ATTRIBUTE_SECTION_VARS: must be called before using -// ATTRIBUTE_SECTION_START or ATTRIBUTE_SECTION_STOP on a name. -// Put this call at the global level. -// 5) ATTRIBUTE_SECTION_START/ATTRIBUTE_SECTION_STOP: call this to say -// where in memory a given section is. All functions declared with -// ATTRIBUTE_SECTION are guaranteed to be between START and STOP. - -#if defined(HAVE___ATTRIBUTE__) && defined(__ELF__) -# define ATTRIBUTE_SECTION(name) __attribute__ ((section (#name))) - - // Weak section declaration to be used as a global declaration - // for ATTRIBUTE_SECTION_START|STOP(name) to compile and link - // even without functions with ATTRIBUTE_SECTION(name). -# define DECLARE_ATTRIBUTE_SECTION_VARS(name) \ - extern char __start_##name[] ATTRIBUTE_WEAK; \ - extern char __stop_##name[] ATTRIBUTE_WEAK -# define INIT_ATTRIBUTE_SECTION_VARS(name) // no-op for ELF -# define DEFINE_ATTRIBUTE_SECTION_VARS(name) // no-op for ELF - - // Return void* pointers to start/end of a section of code with functions - // having ATTRIBUTE_SECTION(name), or 0 if no such function exists. - // One must DECLARE_ATTRIBUTE_SECTION(name) for this to compile and link. -# define ATTRIBUTE_SECTION_START(name) (reinterpret_cast(__start_##name)) -# define ATTRIBUTE_SECTION_STOP(name) (reinterpret_cast(__stop_##name)) -# define HAVE_ATTRIBUTE_SECTION_START 1 - -#elif defined(HAVE___ATTRIBUTE__) && defined(__MACH__) -# define ATTRIBUTE_SECTION(name) __attribute__ ((section ("__TEXT, " #name))) - -#include -#include -class AssignAttributeStartEnd { - public: - AssignAttributeStartEnd(const char* name, char** pstart, char** pend) { - // Find out what dynamic library name is defined in - if (_dyld_present()) { - for (int i = _dyld_image_count() - 1; i >= 0; --i) { - const mach_header* hdr = _dyld_get_image_header(i); -#ifdef MH_MAGIC_64 - if (hdr->magic == MH_MAGIC_64) { - uint64_t len; - *pstart = getsectdatafromheader_64((mach_header_64*)hdr, - "__TEXT", name, &len); - if (*pstart) { // NULL if not defined in this dynamic library - *pstart += _dyld_get_image_vmaddr_slide(i); // correct for reloc - *pend = *pstart + len; - return; - } - } -#endif - if (hdr->magic == MH_MAGIC) { - uint32_t len; - *pstart = getsectdatafromheader(hdr, "__TEXT", name, &len); - if (*pstart) { // NULL if not defined in this dynamic library - *pstart += _dyld_get_image_vmaddr_slide(i); // correct for reloc - *pend = *pstart + len; - return; - } - } - } - } - // If we get here, not defined in a dll at all. See if defined statically. - unsigned long len; // don't ask me why this type isn't uint32_t too... - *pstart = getsectdata("__TEXT", name, &len); - *pend = *pstart + len; - } -}; - -#define DECLARE_ATTRIBUTE_SECTION_VARS(name) \ - extern char* __start_##name; \ - extern char* __stop_##name - -#define INIT_ATTRIBUTE_SECTION_VARS(name) \ - DECLARE_ATTRIBUTE_SECTION_VARS(name); \ - static const AssignAttributeStartEnd __assign_##name( \ - #name, &__start_##name, &__stop_##name) - -#define DEFINE_ATTRIBUTE_SECTION_VARS(name) \ - char* __start_##name, *__stop_##name; \ - INIT_ATTRIBUTE_SECTION_VARS(name) - -# define ATTRIBUTE_SECTION_START(name) (reinterpret_cast(__start_##name)) -# define ATTRIBUTE_SECTION_STOP(name) (reinterpret_cast(__stop_##name)) -# define HAVE_ATTRIBUTE_SECTION_START 1 - -#else // not HAVE___ATTRIBUTE__ && __ELF__, nor HAVE___ATTRIBUTE__ && __MACH__ -# define ATTRIBUTE_SECTION(name) -# define DECLARE_ATTRIBUTE_SECTION_VARS(name) -# define INIT_ATTRIBUTE_SECTION_VARS(name) -# define DEFINE_ATTRIBUTE_SECTION_VARS(name) -# define ATTRIBUTE_SECTION_START(name) (reinterpret_cast(0)) -# define ATTRIBUTE_SECTION_STOP(name) (reinterpret_cast(0)) - -#endif // HAVE___ATTRIBUTE__ and __ELF__ or __MACH__ - -#if defined(HAVE___ATTRIBUTE__) -# if (defined(__i386__) || defined(__x86_64__)) -# define CACHELINE_ALIGNED __attribute__((aligned(64))) -# elif (defined(__PPC__) || defined(__PPC64__)) -# define CACHELINE_ALIGNED __attribute__((aligned(16))) -# elif (defined(__arm__)) -# define CACHELINE_ALIGNED __attribute__((aligned(64))) - // some ARMs have shorter cache lines (ARM1176JZF-S is 32 bytes for example) but obviously 64-byte aligned implies 32-byte aligned -# elif (defined(__mips__)) -# define CACHELINE_ALIGNED __attribute__((aligned(128))) -# elif (defined(__aarch64__)) -# define CACHELINE_ALIGNED __attribute__((aligned(64))) - // implementation specific, Cortex-A53 and 57 should have 64 bytes -# elif (defined(__s390__)) -# define CACHELINE_ALIGNED __attribute__((aligned(256))) -# else -# error Could not determine cache line length - unknown architecture -# endif -#else -# define CACHELINE_ALIGNED -#endif // defined(HAVE___ATTRIBUTE__) && (__i386__ || __x86_64__) - -// Structure for discovering alignment -union MemoryAligner { - void* p; - double d; - size_t s; -} CACHELINE_ALIGNED; - -// The following enum should be used only as a constructor argument to indicate -// that the variable has static storage class, and that the constructor should -// do nothing to its state. It indicates to the reader that it is legal to -// declare a static nistance of the class, provided the constructor is given -// the base::LINKER_INITIALIZED argument. Normally, it is unsafe to declare a -// static variable that has a constructor or a destructor because invocation -// order is undefined. However, IF the type can be initialized by filling with -// zeroes (which the loader does for static variables), AND the destructor also -// does nothing to the storage, then a constructor declared as -// explicit MyClass(base::LinkerInitialized x) {} -// and invoked as -// static MyClass my_variable_name(base::LINKER_INITIALIZED); -namespace base { -enum LinkerInitialized { LINKER_INITIALIZED }; -} - -#endif // _BASICTYPES_H_ diff --git a/contrib/libtcmalloc/src/base/commandlineflags.h b/contrib/libtcmalloc/src/base/commandlineflags.h deleted file mode 100644 index e940edd3791..00000000000 --- a/contrib/libtcmalloc/src/base/commandlineflags.h +++ /dev/null @@ -1,166 +0,0 @@ -// -*- Mode: C++; c-basic-offset: 2; indent-tabs-mode: nil -*- -// Copyright (c) 2005, Google Inc. -// All rights reserved. -// -// Redistribution and use in source and binary forms, with or without -// modification, are permitted provided that the following conditions are -// met: -// -// * Redistributions of source code must retain the above copyright -// notice, this list of conditions and the following disclaimer. -// * Redistributions in binary form must reproduce the above -// copyright notice, this list of conditions and the following disclaimer -// in the documentation and/or other materials provided with the -// distribution. -// * Neither the name of Google Inc. nor the names of its -// contributors may be used to endorse or promote products derived from -// this software without specific prior written permission. -// -// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS -// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT -// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR -// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT -// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, -// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT -// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, -// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY -// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT -// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE -// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. - -// --- -// This file is a compatibility layer that defines Google's version of -// command line flags that are used for configuration. -// -// We put flags into their own namespace. It is purposefully -// named in an opaque way that people should have trouble typing -// directly. The idea is that DEFINE puts the flag in the weird -// namespace, and DECLARE imports the flag from there into the -// current namespace. The net result is to force people to use -// DECLARE to get access to a flag, rather than saying -// extern bool FLAGS_logtostderr; -// or some such instead. We want this so we can put extra -// functionality (like sanity-checking) in DECLARE if we want, -// and make sure it is picked up everywhere. -// -// We also put the type of the variable in the namespace, so that -// people can't DECLARE_int32 something that they DEFINE_bool'd -// elsewhere. -#ifndef BASE_COMMANDLINEFLAGS_H_ -#define BASE_COMMANDLINEFLAGS_H_ - -#include "../config.h" -#include -#include // for memchr -#include // for getenv -#include "base/basictypes.h" - -#define DECLARE_VARIABLE(type, name) \ - namespace FLAG__namespace_do_not_use_directly_use_DECLARE_##type##_instead { \ - extern PERFTOOLS_DLL_DECL type FLAGS_##name; \ - } \ - using FLAG__namespace_do_not_use_directly_use_DECLARE_##type##_instead::FLAGS_##name - -#define DEFINE_VARIABLE(type, name, value, meaning) \ - namespace FLAG__namespace_do_not_use_directly_use_DECLARE_##type##_instead { \ - PERFTOOLS_DLL_DECL type FLAGS_##name(value); \ - char FLAGS_no##name; \ - } \ - using FLAG__namespace_do_not_use_directly_use_DECLARE_##type##_instead::FLAGS_##name - -// bool specialization -#define DECLARE_bool(name) \ - DECLARE_VARIABLE(bool, name) -#define DEFINE_bool(name, value, meaning) \ - DEFINE_VARIABLE(bool, name, value, meaning) - -// int32 specialization -#define DECLARE_int32(name) \ - DECLARE_VARIABLE(int32, name) -#define DEFINE_int32(name, value, meaning) \ - DEFINE_VARIABLE(int32, name, value, meaning) - -// int64 specialization -#define DECLARE_int64(name) \ - DECLARE_VARIABLE(int64, name) -#define DEFINE_int64(name, value, meaning) \ - DEFINE_VARIABLE(int64, name, value, meaning) - -#define DECLARE_uint64(name) \ - DECLARE_VARIABLE(uint64, name) -#define DEFINE_uint64(name, value, meaning) \ - DEFINE_VARIABLE(uint64, name, value, meaning) - -// double specialization -#define DECLARE_double(name) \ - DECLARE_VARIABLE(double, name) -#define DEFINE_double(name, value, meaning) \ - DEFINE_VARIABLE(double, name, value, meaning) - -// Special case for string, because we have to specify the namespace -// std::string, which doesn't play nicely with our FLAG__namespace hackery. -#define DECLARE_string(name) \ - namespace FLAG__namespace_do_not_use_directly_use_DECLARE_string_instead { \ - extern std::string FLAGS_##name; \ - } \ - using FLAG__namespace_do_not_use_directly_use_DECLARE_string_instead::FLAGS_##name -#define DEFINE_string(name, value, meaning) \ - namespace FLAG__namespace_do_not_use_directly_use_DECLARE_string_instead { \ - std::string FLAGS_##name(value); \ - char FLAGS_no##name; \ - } \ - using FLAG__namespace_do_not_use_directly_use_DECLARE_string_instead::FLAGS_##name - -// implemented in sysinfo.cc -namespace tcmalloc { - namespace commandlineflags { - - inline bool StringToBool(const char *value, bool def) { - if (!value) { - return def; - } - return memchr("tTyY1\0", value[0], 6) != NULL; - } - - inline int StringToInt(const char *value, int def) { - if (!value) { - return def; - } - return strtol(value, NULL, 10); - } - - inline long long StringToLongLong(const char *value, long long def) { - if (!value) { - return def; - } - return strtoll(value, NULL, 10); - } - - inline double StringToDouble(const char *value, double def) { - if (!value) { - return def; - } - return strtod(value, NULL); - } - } -} - -// These macros (could be functions, but I don't want to bother with a .cc -// file), make it easier to initialize flags from the environment. - -#define EnvToString(envname, dflt) \ - (!getenv(envname) ? (dflt) : getenv(envname)) - -#define EnvToBool(envname, dflt) \ - tcmalloc::commandlineflags::StringToBool(getenv(envname), dflt) - -#define EnvToInt(envname, dflt) \ - tcmalloc::commandlineflags::StringToInt(getenv(envname), dflt) - -#define EnvToInt64(envname, dflt) \ - tcmalloc::commandlineflags::StringToLongLong(getenv(envname), dflt) - -#define EnvToDouble(envname, dflt) \ - tcmalloc::commandlineflags::StringToDouble(getenv(envname), dflt) - -#endif // BASE_COMMANDLINEFLAGS_H_ diff --git a/contrib/libtcmalloc/src/base/dynamic_annotations.c b/contrib/libtcmalloc/src/base/dynamic_annotations.c deleted file mode 100644 index 87bd2ecde97..00000000000 --- a/contrib/libtcmalloc/src/base/dynamic_annotations.c +++ /dev/null @@ -1,179 +0,0 @@ -/* Copyright (c) 2008-2009, Google Inc. - * All rights reserved. - * - * Redistribution and use in source and binary forms, with or without - * modification, are permitted provided that the following conditions are - * met: - * - * * Redistributions of source code must retain the above copyright - * notice, this list of conditions and the following disclaimer. - * * Redistributions in binary form must reproduce the above - * copyright notice, this list of conditions and the following disclaimer - * in the documentation and/or other materials provided with the - * distribution. - * * Neither the name of Google Inc. nor the names of its - * contributors may be used to endorse or promote products derived from - * this software without specific prior written permission. - * - * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS - * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT - * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR - * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT - * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, - * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT - * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, - * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY - * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT - * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE - * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. - * - * --- - * Author: Kostya Serebryany - */ - -#ifdef __cplusplus -# error "This file should be built as pure C to avoid name mangling" -#endif - -#include "config.h" -#include -#include - -#include "base/dynamic_annotations.h" -#include "getenv_safe.h" // for TCMallocGetenvSafe - -#ifdef __GNUC__ -/* valgrind.h uses gcc extensions so it won't build with other compilers */ -# ifdef HAVE_VALGRIND_H /* prefer the user's copy if they have it */ -# include -# else /* otherwise just use the copy that we have */ -# include "third_party/valgrind.h" -# endif -#endif - -/* Compiler-based ThreadSanitizer defines - DYNAMIC_ANNOTATIONS_EXTERNAL_IMPL = 1 - and provides its own definitions of the functions. */ - -#ifndef DYNAMIC_ANNOTATIONS_EXTERNAL_IMPL -# define DYNAMIC_ANNOTATIONS_EXTERNAL_IMPL 0 -#endif - -/* Each function is empty and called (via a macro) only in debug mode. - The arguments are captured by dynamic tools at runtime. */ - -#if DYNAMIC_ANNOTATIONS_ENABLED == 1 \ - && DYNAMIC_ANNOTATIONS_EXTERNAL_IMPL == 0 - -void AnnotateRWLockCreate(const char *file, int line, - const volatile void *lock){} -void AnnotateRWLockDestroy(const char *file, int line, - const volatile void *lock){} -void AnnotateRWLockAcquired(const char *file, int line, - const volatile void *lock, long is_w){} -void AnnotateRWLockReleased(const char *file, int line, - const volatile void *lock, long is_w){} -void AnnotateBarrierInit(const char *file, int line, - const volatile void *barrier, long count, - long reinitialization_allowed) {} -void AnnotateBarrierWaitBefore(const char *file, int line, - const volatile void *barrier) {} -void AnnotateBarrierWaitAfter(const char *file, int line, - const volatile void *barrier) {} -void AnnotateBarrierDestroy(const char *file, int line, - const volatile void *barrier) {} - -void AnnotateCondVarWait(const char *file, int line, - const volatile void *cv, - const volatile void *lock){} -void AnnotateCondVarSignal(const char *file, int line, - const volatile void *cv){} -void AnnotateCondVarSignalAll(const char *file, int line, - const volatile void *cv){} -void AnnotatePublishMemoryRange(const char *file, int line, - const volatile void *address, - long size){} -void AnnotateUnpublishMemoryRange(const char *file, int line, - const volatile void *address, - long size){} -void AnnotatePCQCreate(const char *file, int line, - const volatile void *pcq){} -void AnnotatePCQDestroy(const char *file, int line, - const volatile void *pcq){} -void AnnotatePCQPut(const char *file, int line, - const volatile void *pcq){} -void AnnotatePCQGet(const char *file, int line, - const volatile void *pcq){} -void AnnotateNewMemory(const char *file, int line, - const volatile void *mem, - long size){} -void AnnotateExpectRace(const char *file, int line, - const volatile void *mem, - const char *description){} -void AnnotateBenignRace(const char *file, int line, - const volatile void *mem, - const char *description){} -void AnnotateBenignRaceSized(const char *file, int line, - const volatile void *mem, - long size, - const char *description) {} -void AnnotateMutexIsUsedAsCondVar(const char *file, int line, - const volatile void *mu){} -void AnnotateTraceMemory(const char *file, int line, - const volatile void *arg){} -void AnnotateThreadName(const char *file, int line, - const char *name){} -void AnnotateIgnoreReadsBegin(const char *file, int line){} -void AnnotateIgnoreReadsEnd(const char *file, int line){} -void AnnotateIgnoreWritesBegin(const char *file, int line){} -void AnnotateIgnoreWritesEnd(const char *file, int line){} -void AnnotateEnableRaceDetection(const char *file, int line, int enable){} -void AnnotateNoOp(const char *file, int line, - const volatile void *arg){} -void AnnotateFlushState(const char *file, int line){} - -#endif /* DYNAMIC_ANNOTATIONS_ENABLED == 1 - && DYNAMIC_ANNOTATIONS_EXTERNAL_IMPL == 0 */ - -#if DYNAMIC_ANNOTATIONS_EXTERNAL_IMPL == 0 - -static int GetRunningOnValgrind(void) { -#ifdef RUNNING_ON_VALGRIND - if (RUNNING_ON_VALGRIND) return 1; -#endif - const char *running_on_valgrind_str = TCMallocGetenvSafe("RUNNING_ON_VALGRIND"); - if (running_on_valgrind_str) { - return strcmp(running_on_valgrind_str, "0") != 0; - } - return 0; -} - -/* See the comments in dynamic_annotations.h */ -int RunningOnValgrind(void) { - static volatile int running_on_valgrind = -1; - int local_running_on_valgrind = running_on_valgrind; - /* C doesn't have thread-safe initialization of statics, and we - don't want to depend on pthread_once here, so hack it. */ - ANNOTATE_BENIGN_RACE(&running_on_valgrind, "safe hack"); - if (local_running_on_valgrind == -1) - running_on_valgrind = local_running_on_valgrind = GetRunningOnValgrind(); - return local_running_on_valgrind; -} - -#endif /* DYNAMIC_ANNOTATIONS_EXTERNAL_IMPL == 0 */ - -/* See the comments in dynamic_annotations.h */ -double ValgrindSlowdown(void) { - /* Same initialization hack as in RunningOnValgrind(). */ - static volatile double slowdown = 0.0; - double local_slowdown = slowdown; - ANNOTATE_BENIGN_RACE(&slowdown, "safe hack"); - if (RunningOnValgrind() == 0) { - return 1.0; - } - if (local_slowdown == 0.0) { - char *env = getenv("VALGRIND_SLOWDOWN"); - slowdown = local_slowdown = env ? atof(env) : 50.0; - } - return local_slowdown; -} diff --git a/contrib/libtcmalloc/src/base/dynamic_annotations.h b/contrib/libtcmalloc/src/base/dynamic_annotations.h deleted file mode 100644 index 4669315ced3..00000000000 --- a/contrib/libtcmalloc/src/base/dynamic_annotations.h +++ /dev/null @@ -1,627 +0,0 @@ -/* Copyright (c) 2008, Google Inc. - * All rights reserved. - * - * Redistribution and use in source and binary forms, with or without - * modification, are permitted provided that the following conditions are - * met: - * - * * Redistributions of source code must retain the above copyright - * notice, this list of conditions and the following disclaimer. - * * Redistributions in binary form must reproduce the above - * copyright notice, this list of conditions and the following disclaimer - * in the documentation and/or other materials provided with the - * distribution. - * * Neither the name of Google Inc. nor the names of its - * contributors may be used to endorse or promote products derived from - * this software without specific prior written permission. - * - * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS - * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT - * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR - * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT - * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, - * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT - * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, - * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY - * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT - * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE - * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. - * - * --- - * Author: Kostya Serebryany - */ - -/* This file defines dynamic annotations for use with dynamic analysis - tool such as valgrind, PIN, etc. - - Dynamic annotation is a source code annotation that affects - the generated code (that is, the annotation is not a comment). - Each such annotation is attached to a particular - instruction and/or to a particular object (address) in the program. - - The annotations that should be used by users are macros in all upper-case - (e.g., ANNOTATE_NEW_MEMORY). - - Actual implementation of these macros may differ depending on the - dynamic analysis tool being used. - - See http://code.google.com/p/data-race-test/ for more information. - - This file supports the following dynamic analysis tools: - - None (DYNAMIC_ANNOTATIONS_ENABLED is not defined or zero). - Macros are defined empty. - - ThreadSanitizer, Helgrind, DRD (DYNAMIC_ANNOTATIONS_ENABLED is 1). - Macros are defined as calls to non-inlinable empty functions - that are intercepted by Valgrind. */ - -#ifndef BASE_DYNAMIC_ANNOTATIONS_H_ -#define BASE_DYNAMIC_ANNOTATIONS_H_ - -#ifndef DYNAMIC_ANNOTATIONS_ENABLED -# define DYNAMIC_ANNOTATIONS_ENABLED 0 -#endif - -#if DYNAMIC_ANNOTATIONS_ENABLED != 0 - - /* ------------------------------------------------------------- - Annotations useful when implementing condition variables such as CondVar, - using conditional critical sections (Await/LockWhen) and when constructing - user-defined synchronization mechanisms. - - The annotations ANNOTATE_HAPPENS_BEFORE() and ANNOTATE_HAPPENS_AFTER() can - be used to define happens-before arcs in user-defined synchronization - mechanisms: the race detector will infer an arc from the former to the - latter when they share the same argument pointer. - - Example 1 (reference counting): - - void Unref() { - ANNOTATE_HAPPENS_BEFORE(&refcount_); - if (AtomicDecrementByOne(&refcount_) == 0) { - ANNOTATE_HAPPENS_AFTER(&refcount_); - delete this; - } - } - - Example 2 (message queue): - - void MyQueue::Put(Type *e) { - MutexLock lock(&mu_); - ANNOTATE_HAPPENS_BEFORE(e); - PutElementIntoMyQueue(e); - } - - Type *MyQueue::Get() { - MutexLock lock(&mu_); - Type *e = GetElementFromMyQueue(); - ANNOTATE_HAPPENS_AFTER(e); - return e; - } - - Note: when possible, please use the existing reference counting and message - queue implementations instead of inventing new ones. */ - - /* Report that wait on the condition variable at address "cv" has succeeded - and the lock at address "lock" is held. */ - #define ANNOTATE_CONDVAR_LOCK_WAIT(cv, lock) \ - AnnotateCondVarWait(__FILE__, __LINE__, cv, lock) - - /* Report that wait on the condition variable at "cv" has succeeded. Variant - w/o lock. */ - #define ANNOTATE_CONDVAR_WAIT(cv) \ - AnnotateCondVarWait(__FILE__, __LINE__, cv, NULL) - - /* Report that we are about to signal on the condition variable at address - "cv". */ - #define ANNOTATE_CONDVAR_SIGNAL(cv) \ - AnnotateCondVarSignal(__FILE__, __LINE__, cv) - - /* Report that we are about to signal_all on the condition variable at "cv". */ - #define ANNOTATE_CONDVAR_SIGNAL_ALL(cv) \ - AnnotateCondVarSignalAll(__FILE__, __LINE__, cv) - - /* Annotations for user-defined synchronization mechanisms. */ - #define ANNOTATE_HAPPENS_BEFORE(obj) ANNOTATE_CONDVAR_SIGNAL(obj) - #define ANNOTATE_HAPPENS_AFTER(obj) ANNOTATE_CONDVAR_WAIT(obj) - - /* Report that the bytes in the range [pointer, pointer+size) are about - to be published safely. The race checker will create a happens-before - arc from the call ANNOTATE_PUBLISH_MEMORY_RANGE(pointer, size) to - subsequent accesses to this memory. - Note: this annotation may not work properly if the race detector uses - sampling, i.e. does not observe all memory accesses. - */ - #define ANNOTATE_PUBLISH_MEMORY_RANGE(pointer, size) \ - AnnotatePublishMemoryRange(__FILE__, __LINE__, pointer, size) - - /* DEPRECATED. Don't use it. */ - #define ANNOTATE_UNPUBLISH_MEMORY_RANGE(pointer, size) \ - AnnotateUnpublishMemoryRange(__FILE__, __LINE__, pointer, size) - - /* DEPRECATED. Don't use it. */ - #define ANNOTATE_SWAP_MEMORY_RANGE(pointer, size) \ - do { \ - ANNOTATE_UNPUBLISH_MEMORY_RANGE(pointer, size); \ - ANNOTATE_PUBLISH_MEMORY_RANGE(pointer, size); \ - } while (0) - - /* Instruct the tool to create a happens-before arc between mu->Unlock() and - mu->Lock(). This annotation may slow down the race detector and hide real - races. Normally it is used only when it would be difficult to annotate each - of the mutex's critical sections individually using the annotations above. - This annotation makes sense only for hybrid race detectors. For pure - happens-before detectors this is a no-op. For more details see - http://code.google.com/p/data-race-test/wiki/PureHappensBeforeVsHybrid . */ - #define ANNOTATE_PURE_HAPPENS_BEFORE_MUTEX(mu) \ - AnnotateMutexIsUsedAsCondVar(__FILE__, __LINE__, mu) - - /* Deprecated. Use ANNOTATE_PURE_HAPPENS_BEFORE_MUTEX. */ - #define ANNOTATE_MUTEX_IS_USED_AS_CONDVAR(mu) \ - AnnotateMutexIsUsedAsCondVar(__FILE__, __LINE__, mu) - - /* ------------------------------------------------------------- - Annotations useful when defining memory allocators, or when memory that - was protected in one way starts to be protected in another. */ - - /* Report that a new memory at "address" of size "size" has been allocated. - This might be used when the memory has been retrieved from a free list and - is about to be reused, or when a the locking discipline for a variable - changes. */ - #define ANNOTATE_NEW_MEMORY(address, size) \ - AnnotateNewMemory(__FILE__, __LINE__, address, size) - - /* ------------------------------------------------------------- - Annotations useful when defining FIFO queues that transfer data between - threads. */ - - /* Report that the producer-consumer queue (such as ProducerConsumerQueue) at - address "pcq" has been created. The ANNOTATE_PCQ_* annotations - should be used only for FIFO queues. For non-FIFO queues use - ANNOTATE_HAPPENS_BEFORE (for put) and ANNOTATE_HAPPENS_AFTER (for get). */ - #define ANNOTATE_PCQ_CREATE(pcq) \ - AnnotatePCQCreate(__FILE__, __LINE__, pcq) - - /* Report that the queue at address "pcq" is about to be destroyed. */ - #define ANNOTATE_PCQ_DESTROY(pcq) \ - AnnotatePCQDestroy(__FILE__, __LINE__, pcq) - - /* Report that we are about to put an element into a FIFO queue at address - "pcq". */ - #define ANNOTATE_PCQ_PUT(pcq) \ - AnnotatePCQPut(__FILE__, __LINE__, pcq) - - /* Report that we've just got an element from a FIFO queue at address "pcq". */ - #define ANNOTATE_PCQ_GET(pcq) \ - AnnotatePCQGet(__FILE__, __LINE__, pcq) - - /* ------------------------------------------------------------- - Annotations that suppress errors. It is usually better to express the - program's synchronization using the other annotations, but these can - be used when all else fails. */ - - /* Report that we may have a benign race at "pointer", with size - "sizeof(*(pointer))". "pointer" must be a non-void* pointer. Insert at the - point where "pointer" has been allocated, preferably close to the point - where the race happens. See also ANNOTATE_BENIGN_RACE_STATIC. */ - #define ANNOTATE_BENIGN_RACE(pointer, description) \ - AnnotateBenignRaceSized(__FILE__, __LINE__, pointer, \ - sizeof(*(pointer)), description) - - /* Same as ANNOTATE_BENIGN_RACE(address, description), but applies to - the memory range [address, address+size). */ - #define ANNOTATE_BENIGN_RACE_SIZED(address, size, description) \ - AnnotateBenignRaceSized(__FILE__, __LINE__, address, size, description) - - /* Request the analysis tool to ignore all reads in the current thread - until ANNOTATE_IGNORE_READS_END is called. - Useful to ignore intentional racey reads, while still checking - other reads and all writes. - See also ANNOTATE_UNPROTECTED_READ. */ - #define ANNOTATE_IGNORE_READS_BEGIN() \ - AnnotateIgnoreReadsBegin(__FILE__, __LINE__) - - /* Stop ignoring reads. */ - #define ANNOTATE_IGNORE_READS_END() \ - AnnotateIgnoreReadsEnd(__FILE__, __LINE__) - - /* Similar to ANNOTATE_IGNORE_READS_BEGIN, but ignore writes. */ - #define ANNOTATE_IGNORE_WRITES_BEGIN() \ - AnnotateIgnoreWritesBegin(__FILE__, __LINE__) - - /* Stop ignoring writes. */ - #define ANNOTATE_IGNORE_WRITES_END() \ - AnnotateIgnoreWritesEnd(__FILE__, __LINE__) - - /* Start ignoring all memory accesses (reads and writes). */ - #define ANNOTATE_IGNORE_READS_AND_WRITES_BEGIN() \ - do {\ - ANNOTATE_IGNORE_READS_BEGIN();\ - ANNOTATE_IGNORE_WRITES_BEGIN();\ - }while(0)\ - - /* Stop ignoring all memory accesses. */ - #define ANNOTATE_IGNORE_READS_AND_WRITES_END() \ - do {\ - ANNOTATE_IGNORE_WRITES_END();\ - ANNOTATE_IGNORE_READS_END();\ - }while(0)\ - - /* Enable (enable!=0) or disable (enable==0) race detection for all threads. - This annotation could be useful if you want to skip expensive race analysis - during some period of program execution, e.g. during initialization. */ - #define ANNOTATE_ENABLE_RACE_DETECTION(enable) \ - AnnotateEnableRaceDetection(__FILE__, __LINE__, enable) - - /* ------------------------------------------------------------- - Annotations useful for debugging. */ - - /* Request to trace every access to "address". */ - #define ANNOTATE_TRACE_MEMORY(address) \ - AnnotateTraceMemory(__FILE__, __LINE__, address) - - /* Report the current thread name to a race detector. */ - #define ANNOTATE_THREAD_NAME(name) \ - AnnotateThreadName(__FILE__, __LINE__, name) - - /* ------------------------------------------------------------- - Annotations useful when implementing locks. They are not - normally needed by modules that merely use locks. - The "lock" argument is a pointer to the lock object. */ - - /* Report that a lock has been created at address "lock". */ - #define ANNOTATE_RWLOCK_CREATE(lock) \ - AnnotateRWLockCreate(__FILE__, __LINE__, lock) - - /* Report that the lock at address "lock" is about to be destroyed. */ - #define ANNOTATE_RWLOCK_DESTROY(lock) \ - AnnotateRWLockDestroy(__FILE__, __LINE__, lock) - - /* Report that the lock at address "lock" has been acquired. - is_w=1 for writer lock, is_w=0 for reader lock. */ - #define ANNOTATE_RWLOCK_ACQUIRED(lock, is_w) \ - AnnotateRWLockAcquired(__FILE__, __LINE__, lock, is_w) - - /* Report that the lock at address "lock" is about to be released. */ - #define ANNOTATE_RWLOCK_RELEASED(lock, is_w) \ - AnnotateRWLockReleased(__FILE__, __LINE__, lock, is_w) - - /* ------------------------------------------------------------- - Annotations useful when implementing barriers. They are not - normally needed by modules that merely use barriers. - The "barrier" argument is a pointer to the barrier object. */ - - /* Report that the "barrier" has been initialized with initial "count". - If 'reinitialization_allowed' is true, initialization is allowed to happen - multiple times w/o calling barrier_destroy() */ - #define ANNOTATE_BARRIER_INIT(barrier, count, reinitialization_allowed) \ - AnnotateBarrierInit(__FILE__, __LINE__, barrier, count, \ - reinitialization_allowed) - - /* Report that we are about to enter barrier_wait("barrier"). */ - #define ANNOTATE_BARRIER_WAIT_BEFORE(barrier) \ - AnnotateBarrierWaitBefore(__FILE__, __LINE__, barrier) - - /* Report that we just exited barrier_wait("barrier"). */ - #define ANNOTATE_BARRIER_WAIT_AFTER(barrier) \ - AnnotateBarrierWaitAfter(__FILE__, __LINE__, barrier) - - /* Report that the "barrier" has been destroyed. */ - #define ANNOTATE_BARRIER_DESTROY(barrier) \ - AnnotateBarrierDestroy(__FILE__, __LINE__, barrier) - - /* ------------------------------------------------------------- - Annotations useful for testing race detectors. */ - - /* Report that we expect a race on the variable at "address". - Use only in unit tests for a race detector. */ - #define ANNOTATE_EXPECT_RACE(address, description) \ - AnnotateExpectRace(__FILE__, __LINE__, address, description) - - /* A no-op. Insert where you like to test the interceptors. */ - #define ANNOTATE_NO_OP(arg) \ - AnnotateNoOp(__FILE__, __LINE__, arg) - - /* Force the race detector to flush its state. The actual effect depends on - * the implementation of the detector. */ - #define ANNOTATE_FLUSH_STATE() \ - AnnotateFlushState(__FILE__, __LINE__) - - -#else /* DYNAMIC_ANNOTATIONS_ENABLED == 0 */ - - #define ANNOTATE_RWLOCK_CREATE(lock) /* empty */ - #define ANNOTATE_RWLOCK_DESTROY(lock) /* empty */ - #define ANNOTATE_RWLOCK_ACQUIRED(lock, is_w) /* empty */ - #define ANNOTATE_RWLOCK_RELEASED(lock, is_w) /* empty */ - #define ANNOTATE_BARRIER_INIT(barrier, count, reinitialization_allowed) /* */ - #define ANNOTATE_BARRIER_WAIT_BEFORE(barrier) /* empty */ - #define ANNOTATE_BARRIER_WAIT_AFTER(barrier) /* empty */ - #define ANNOTATE_BARRIER_DESTROY(barrier) /* empty */ - #define ANNOTATE_CONDVAR_LOCK_WAIT(cv, lock) /* empty */ - #define ANNOTATE_CONDVAR_WAIT(cv) /* empty */ - #define ANNOTATE_CONDVAR_SIGNAL(cv) /* empty */ - #define ANNOTATE_CONDVAR_SIGNAL_ALL(cv) /* empty */ - #define ANNOTATE_HAPPENS_BEFORE(obj) /* empty */ - #define ANNOTATE_HAPPENS_AFTER(obj) /* empty */ - #define ANNOTATE_PUBLISH_MEMORY_RANGE(address, size) /* empty */ - #define ANNOTATE_UNPUBLISH_MEMORY_RANGE(address, size) /* empty */ - #define ANNOTATE_SWAP_MEMORY_RANGE(address, size) /* empty */ - #define ANNOTATE_PCQ_CREATE(pcq) /* empty */ - #define ANNOTATE_PCQ_DESTROY(pcq) /* empty */ - #define ANNOTATE_PCQ_PUT(pcq) /* empty */ - #define ANNOTATE_PCQ_GET(pcq) /* empty */ - #define ANNOTATE_NEW_MEMORY(address, size) /* empty */ - #define ANNOTATE_EXPECT_RACE(address, description) /* empty */ - #define ANNOTATE_BENIGN_RACE(address, description) /* empty */ - #define ANNOTATE_BENIGN_RACE_SIZED(address, size, description) /* empty */ - #define ANNOTATE_PURE_HAPPENS_BEFORE_MUTEX(mu) /* empty */ - #define ANNOTATE_MUTEX_IS_USED_AS_CONDVAR(mu) /* empty */ - #define ANNOTATE_TRACE_MEMORY(arg) /* empty */ - #define ANNOTATE_THREAD_NAME(name) /* empty */ - #define ANNOTATE_IGNORE_READS_BEGIN() /* empty */ - #define ANNOTATE_IGNORE_READS_END() /* empty */ - #define ANNOTATE_IGNORE_WRITES_BEGIN() /* empty */ - #define ANNOTATE_IGNORE_WRITES_END() /* empty */ - #define ANNOTATE_IGNORE_READS_AND_WRITES_BEGIN() /* empty */ - #define ANNOTATE_IGNORE_READS_AND_WRITES_END() /* empty */ - #define ANNOTATE_ENABLE_RACE_DETECTION(enable) /* empty */ - #define ANNOTATE_NO_OP(arg) /* empty */ - #define ANNOTATE_FLUSH_STATE() /* empty */ - -#endif /* DYNAMIC_ANNOTATIONS_ENABLED */ - -/* Macro definitions for GCC attributes that allow static thread safety - analysis to recognize and use some of the dynamic annotations as - escape hatches. - TODO(lcwu): remove the check for __SUPPORT_DYN_ANNOTATION__ once the - default crosstool/GCC supports these GCC attributes. */ - -#define ANNOTALYSIS_STATIC_INLINE -#define ANNOTALYSIS_SEMICOLON_OR_EMPTY_BODY ; -#define ANNOTALYSIS_IGNORE_READS_BEGIN -#define ANNOTALYSIS_IGNORE_READS_END -#define ANNOTALYSIS_IGNORE_WRITES_BEGIN -#define ANNOTALYSIS_IGNORE_WRITES_END -#define ANNOTALYSIS_UNPROTECTED_READ - -#if defined(__GNUC__) && (!defined(SWIG)) && (!defined(__clang__)) && \ - defined(__SUPPORT_TS_ANNOTATION__) && defined(__SUPPORT_DYN_ANNOTATION__) - -#if DYNAMIC_ANNOTATIONS_ENABLED == 0 -#define ANNOTALYSIS_ONLY 1 -#undef ANNOTALYSIS_STATIC_INLINE -#define ANNOTALYSIS_STATIC_INLINE static inline -#undef ANNOTALYSIS_SEMICOLON_OR_EMPTY_BODY -#define ANNOTALYSIS_SEMICOLON_OR_EMPTY_BODY { (void)file; (void)line; } -#endif - -/* Only emit attributes when annotalysis is enabled. */ -#if defined(__SUPPORT_TS_ANNOTATION__) && defined(__SUPPORT_DYN_ANNOTATION__) -#undef ANNOTALYSIS_IGNORE_READS_BEGIN -#define ANNOTALYSIS_IGNORE_READS_BEGIN __attribute__ ((ignore_reads_begin)) -#undef ANNOTALYSIS_IGNORE_READS_END -#define ANNOTALYSIS_IGNORE_READS_END __attribute__ ((ignore_reads_end)) -#undef ANNOTALYSIS_IGNORE_WRITES_BEGIN -#define ANNOTALYSIS_IGNORE_WRITES_BEGIN __attribute__ ((ignore_writes_begin)) -#undef ANNOTALYSIS_IGNORE_WRITES_END -#define ANNOTALYSIS_IGNORE_WRITES_END __attribute__ ((ignore_writes_end)) -#undef ANNOTALYSIS_UNPROTECTED_READ -#define ANNOTALYSIS_UNPROTECTED_READ __attribute__ ((unprotected_read)) -#endif - -#endif // defined(__GNUC__) && (!defined(SWIG)) && (!defined(__clang__)) - -/* Use the macros above rather than using these functions directly. */ -#ifdef __cplusplus -extern "C" { -#endif -void AnnotateRWLockCreate(const char *file, int line, - const volatile void *lock); -void AnnotateRWLockDestroy(const char *file, int line, - const volatile void *lock); -void AnnotateRWLockAcquired(const char *file, int line, - const volatile void *lock, long is_w); -void AnnotateRWLockReleased(const char *file, int line, - const volatile void *lock, long is_w); -void AnnotateBarrierInit(const char *file, int line, - const volatile void *barrier, long count, - long reinitialization_allowed); -void AnnotateBarrierWaitBefore(const char *file, int line, - const volatile void *barrier); -void AnnotateBarrierWaitAfter(const char *file, int line, - const volatile void *barrier); -void AnnotateBarrierDestroy(const char *file, int line, - const volatile void *barrier); -void AnnotateCondVarWait(const char *file, int line, - const volatile void *cv, - const volatile void *lock); -void AnnotateCondVarSignal(const char *file, int line, - const volatile void *cv); -void AnnotateCondVarSignalAll(const char *file, int line, - const volatile void *cv); -void AnnotatePublishMemoryRange(const char *file, int line, - const volatile void *address, - long size); -void AnnotateUnpublishMemoryRange(const char *file, int line, - const volatile void *address, - long size); -void AnnotatePCQCreate(const char *file, int line, - const volatile void *pcq); -void AnnotatePCQDestroy(const char *file, int line, - const volatile void *pcq); -void AnnotatePCQPut(const char *file, int line, - const volatile void *pcq); -void AnnotatePCQGet(const char *file, int line, - const volatile void *pcq); -void AnnotateNewMemory(const char *file, int line, - const volatile void *address, - long size); -void AnnotateExpectRace(const char *file, int line, - const volatile void *address, - const char *description); -void AnnotateBenignRace(const char *file, int line, - const volatile void *address, - const char *description); -void AnnotateBenignRaceSized(const char *file, int line, - const volatile void *address, - long size, - const char *description); -void AnnotateMutexIsUsedAsCondVar(const char *file, int line, - const volatile void *mu); -void AnnotateTraceMemory(const char *file, int line, - const volatile void *arg); -void AnnotateThreadName(const char *file, int line, - const char *name); -ANNOTALYSIS_STATIC_INLINE -void AnnotateIgnoreReadsBegin(const char *file, int line) - ANNOTALYSIS_IGNORE_READS_BEGIN ANNOTALYSIS_SEMICOLON_OR_EMPTY_BODY -ANNOTALYSIS_STATIC_INLINE -void AnnotateIgnoreReadsEnd(const char *file, int line) - ANNOTALYSIS_IGNORE_READS_END ANNOTALYSIS_SEMICOLON_OR_EMPTY_BODY -ANNOTALYSIS_STATIC_INLINE -void AnnotateIgnoreWritesBegin(const char *file, int line) - ANNOTALYSIS_IGNORE_WRITES_BEGIN ANNOTALYSIS_SEMICOLON_OR_EMPTY_BODY -ANNOTALYSIS_STATIC_INLINE -void AnnotateIgnoreWritesEnd(const char *file, int line) - ANNOTALYSIS_IGNORE_WRITES_END ANNOTALYSIS_SEMICOLON_OR_EMPTY_BODY -void AnnotateEnableRaceDetection(const char *file, int line, int enable); -void AnnotateNoOp(const char *file, int line, - const volatile void *arg); -void AnnotateFlushState(const char *file, int line); - -/* Return non-zero value if running under valgrind. - - If "valgrind.h" is included into dynamic_annotations.c, - the regular valgrind mechanism will be used. - See http://valgrind.org/docs/manual/manual-core-adv.html about - RUNNING_ON_VALGRIND and other valgrind "client requests". - The file "valgrind.h" may be obtained by doing - svn co svn://svn.valgrind.org/valgrind/trunk/include - - If for some reason you can't use "valgrind.h" or want to fake valgrind, - there are two ways to make this function return non-zero: - - Use environment variable: export RUNNING_ON_VALGRIND=1 - - Make your tool intercept the function RunningOnValgrind() and - change its return value. - */ -int RunningOnValgrind(void); - -/* ValgrindSlowdown returns: - * 1.0, if (RunningOnValgrind() == 0) - * 50.0, if (RunningOnValgrind() != 0 && getenv("VALGRIND_SLOWDOWN") == NULL) - * atof(getenv("VALGRIND_SLOWDOWN")) otherwise - This function can be used to scale timeout values: - EXAMPLE: - for (;;) { - DoExpensiveBackgroundTask(); - SleepForSeconds(5 * ValgrindSlowdown()); - } - */ -double ValgrindSlowdown(void); - -#ifdef __cplusplus -} -#endif - -#if DYNAMIC_ANNOTATIONS_ENABLED != 0 && defined(__cplusplus) - - /* ANNOTATE_UNPROTECTED_READ is the preferred way to annotate racey reads. - - Instead of doing - ANNOTATE_IGNORE_READS_BEGIN(); - ... = x; - ANNOTATE_IGNORE_READS_END(); - one can use - ... = ANNOTATE_UNPROTECTED_READ(x); */ - template - inline T ANNOTATE_UNPROTECTED_READ(const volatile T &x) - ANNOTALYSIS_UNPROTECTED_READ { - ANNOTATE_IGNORE_READS_BEGIN(); - T res = x; - ANNOTATE_IGNORE_READS_END(); - return res; - } - /* Apply ANNOTATE_BENIGN_RACE_SIZED to a static variable. */ - #define ANNOTATE_BENIGN_RACE_STATIC(static_var, description) \ - namespace { \ - class static_var ## _annotator { \ - public: \ - static_var ## _annotator() { \ - ANNOTATE_BENIGN_RACE_SIZED(&static_var, \ - sizeof(static_var), \ - # static_var ": " description); \ - } \ - }; \ - static static_var ## _annotator the ## static_var ## _annotator;\ - } -#else /* DYNAMIC_ANNOTATIONS_ENABLED == 0 */ - - #define ANNOTATE_UNPROTECTED_READ(x) (x) - #define ANNOTATE_BENIGN_RACE_STATIC(static_var, description) /* empty */ - -#endif /* DYNAMIC_ANNOTATIONS_ENABLED */ - -/* Annotalysis, a GCC based static analyzer, is able to understand and use - some of the dynamic annotations defined in this file. However, dynamic - annotations are usually disabled in the opt mode (to avoid additional - runtime overheads) while Annotalysis only works in the opt mode. - In order for Annotalysis to use these dynamic annotations when they - are disabled, we re-define these annotations here. Note that unlike the - original macro definitions above, these macros are expanded to calls to - static inline functions so that the compiler will be able to remove the - calls after the analysis. */ - -#ifdef ANNOTALYSIS_ONLY - - #undef ANNOTALYSIS_ONLY - - /* Undefine and re-define the macros that the static analyzer understands. */ - #undef ANNOTATE_IGNORE_READS_BEGIN - #define ANNOTATE_IGNORE_READS_BEGIN() \ - AnnotateIgnoreReadsBegin(__FILE__, __LINE__) - - #undef ANNOTATE_IGNORE_READS_END - #define ANNOTATE_IGNORE_READS_END() \ - AnnotateIgnoreReadsEnd(__FILE__, __LINE__) - - #undef ANNOTATE_IGNORE_WRITES_BEGIN - #define ANNOTATE_IGNORE_WRITES_BEGIN() \ - AnnotateIgnoreWritesBegin(__FILE__, __LINE__) - - #undef ANNOTATE_IGNORE_WRITES_END - #define ANNOTATE_IGNORE_WRITES_END() \ - AnnotateIgnoreWritesEnd(__FILE__, __LINE__) - - #undef ANNOTATE_IGNORE_READS_AND_WRITES_BEGIN - #define ANNOTATE_IGNORE_READS_AND_WRITES_BEGIN() \ - do { \ - ANNOTATE_IGNORE_READS_BEGIN(); \ - ANNOTATE_IGNORE_WRITES_BEGIN(); \ - }while(0) \ - - #undef ANNOTATE_IGNORE_READS_AND_WRITES_END - #define ANNOTATE_IGNORE_READS_AND_WRITES_END() \ - do { \ - ANNOTATE_IGNORE_WRITES_END(); \ - ANNOTATE_IGNORE_READS_END(); \ - }while(0) \ - - #if defined(__cplusplus) - #undef ANNOTATE_UNPROTECTED_READ - template - inline T ANNOTATE_UNPROTECTED_READ(const volatile T &x) - ANNOTALYSIS_UNPROTECTED_READ { - ANNOTATE_IGNORE_READS_BEGIN(); - T res = x; - ANNOTATE_IGNORE_READS_END(); - return res; - } - #endif /* __cplusplus */ - -#endif /* ANNOTALYSIS_ONLY */ - -/* Undefine the macros intended only in this file. */ -#undef ANNOTALYSIS_STATIC_INLINE -#undef ANNOTALYSIS_SEMICOLON_OR_EMPTY_BODY - -#endif /* BASE_DYNAMIC_ANNOTATIONS_H_ */ diff --git a/contrib/libtcmalloc/src/base/elf_mem_image.cc b/contrib/libtcmalloc/src/base/elf_mem_image.cc deleted file mode 100644 index d9605609e3a..00000000000 --- a/contrib/libtcmalloc/src/base/elf_mem_image.cc +++ /dev/null @@ -1,443 +0,0 @@ -// -*- Mode: C++; c-basic-offset: 2; indent-tabs-mode: nil -*- -// Copyright (c) 2008, Google Inc. -// All rights reserved. -// -// Redistribution and use in source and binary forms, with or without -// modification, are permitted provided that the following conditions are -// met: -// -// * Redistributions of source code must retain the above copyright -// notice, this list of conditions and the following disclaimer. -// * Redistributions in binary form must reproduce the above -// copyright notice, this list of conditions and the following disclaimer -// in the documentation and/or other materials provided with the -// distribution. -// * Neither the name of Google Inc. nor the names of its -// contributors may be used to endorse or promote products derived from -// this software without specific prior written permission. -// -// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS -// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT -// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR -// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT -// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, -// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT -// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, -// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY -// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT -// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE -// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. - -// --- -// Author: Paul Pluzhnikov -// -// Allow dynamic symbol lookup in an in-memory Elf image. -// - -#include "base/elf_mem_image.h" - -#ifdef HAVE_ELF_MEM_IMAGE // defined in elf_mem_image.h - -#include // for size_t, ptrdiff_t -#include "base/logging.h" - -// From binutils/include/elf/common.h (this doesn't appear to be documented -// anywhere else). -// -// /* This flag appears in a Versym structure. It means that the symbol -// is hidden, and is only visible with an explicit version number. -// This is a GNU extension. */ -// #define VERSYM_HIDDEN 0x8000 -// -// /* This is the mask for the rest of the Versym information. */ -// #define VERSYM_VERSION 0x7fff - -#define VERSYM_VERSION 0x7fff - -#if __clang__ -#pragma clang diagnostic push -#pragma clang diagnostic ignored "-Wunused-const-variable" -#endif - -namespace base { - -namespace { -template class ElfClass { - public: - static const int kElfClass = -1; - static int ElfBind(const ElfW(Sym) *) { - CHECK(false); // << "Unexpected word size"; - return 0; - } - static int ElfType(const ElfW(Sym) *) { - CHECK(false); // << "Unexpected word size"; - return 0; - } -}; - -template <> class ElfClass<32> { - public: - static const int kElfClass = ELFCLASS32; - static int ElfBind(const ElfW(Sym) *symbol) { - return ELF32_ST_BIND(symbol->st_info); - } - static int ElfType(const ElfW(Sym) *symbol) { - return ELF32_ST_TYPE(symbol->st_info); - } -}; - -template <> class ElfClass<64> { - public: - static const int kElfClass = ELFCLASS64; - static int ElfBind(const ElfW(Sym) *symbol) { - return ELF64_ST_BIND(symbol->st_info); - } - static int ElfType(const ElfW(Sym) *symbol) { - return ELF64_ST_TYPE(symbol->st_info); - } -}; - -typedef ElfClass<__WORDSIZE> CurrentElfClass; - -// Extract an element from one of the ELF tables, cast it to desired type. -// This is just a simple arithmetic and a glorified cast. -// Callers are responsible for bounds checking. -template -const T* GetTableElement(const ElfW(Ehdr) *ehdr, - ElfW(Off) table_offset, - ElfW(Word) element_size, - size_t index) { - return reinterpret_cast(reinterpret_cast(ehdr) - + table_offset - + index * element_size); -} -} // namespace - -const void *const ElfMemImage::kInvalidBase = - reinterpret_cast(~0L); - -ElfMemImage::ElfMemImage(const void *base) { - CHECK(base != kInvalidBase); - Init(base); -} - -int ElfMemImage::GetNumSymbols() const { - if (!hash_) { - return 0; - } - // See http://www.caldera.com/developers/gabi/latest/ch5.dynamic.html#hash - return hash_[1]; -} - -const ElfW(Sym) *ElfMemImage::GetDynsym(int index) const { - CHECK_LT(index, GetNumSymbols()); - return dynsym_ + index; -} - -const ElfW(Versym) *ElfMemImage::GetVersym(int index) const { - CHECK_LT(index, GetNumSymbols()); - return versym_ + index; -} - -const ElfW(Phdr) *ElfMemImage::GetPhdr(int index) const { - CHECK_LT(index, ehdr_->e_phnum); - return GetTableElement(ehdr_, - ehdr_->e_phoff, - ehdr_->e_phentsize, - index); -} - -const char *ElfMemImage::GetDynstr(ElfW(Word) offset) const { - CHECK_LT(offset, strsize_); - return dynstr_ + offset; -} - -const void *ElfMemImage::GetSymAddr(const ElfW(Sym) *sym) const { - if (sym->st_shndx == SHN_UNDEF || sym->st_shndx >= SHN_LORESERVE) { - // Symbol corresponds to "special" (e.g. SHN_ABS) section. - return reinterpret_cast(sym->st_value); - } - CHECK_LT(link_base_, sym->st_value); - return GetTableElement(ehdr_, 0, 1, sym->st_value) - link_base_; -} - -const ElfW(Verdef) *ElfMemImage::GetVerdef(int index) const { - CHECK_LE(index, verdefnum_); - const ElfW(Verdef) *version_definition = verdef_; - while (version_definition->vd_ndx < index && version_definition->vd_next) { - const char *const version_definition_as_char = - reinterpret_cast(version_definition); - version_definition = - reinterpret_cast(version_definition_as_char + - version_definition->vd_next); - } - return version_definition->vd_ndx == index ? version_definition : NULL; -} - -const ElfW(Verdaux) *ElfMemImage::GetVerdefAux( - const ElfW(Verdef) *verdef) const { - return reinterpret_cast(verdef+1); -} - -const char *ElfMemImage::GetVerstr(ElfW(Word) offset) const { - CHECK_LT(offset, strsize_); - return dynstr_ + offset; -} - -void ElfMemImage::Init(const void *base) { - ehdr_ = NULL; - dynsym_ = NULL; - dynstr_ = NULL; - versym_ = NULL; - verdef_ = NULL; - hash_ = NULL; - strsize_ = 0; - verdefnum_ = 0; - link_base_ = ~0L; // Sentinel: PT_LOAD .p_vaddr can't possibly be this. - if (!base) { - return; - } - const intptr_t base_as_uintptr_t = reinterpret_cast(base); - // Fake VDSO has low bit set. - const bool fake_vdso = ((base_as_uintptr_t & 1) != 0); - base = reinterpret_cast(base_as_uintptr_t & ~1); - const char *const base_as_char = reinterpret_cast(base); - if (base_as_char[EI_MAG0] != ELFMAG0 || base_as_char[EI_MAG1] != ELFMAG1 || - base_as_char[EI_MAG2] != ELFMAG2 || base_as_char[EI_MAG3] != ELFMAG3) { - RAW_DCHECK(false, "no ELF magic"); // at %p", base); - return; - } - int elf_class = base_as_char[EI_CLASS]; - if (elf_class != CurrentElfClass::kElfClass) { - DCHECK_EQ(elf_class, CurrentElfClass::kElfClass); - return; - } - switch (base_as_char[EI_DATA]) { - case ELFDATA2LSB: { - if (__LITTLE_ENDIAN != __BYTE_ORDER) { - DCHECK_EQ(__LITTLE_ENDIAN, __BYTE_ORDER); // << ": wrong byte order"; - return; - } - break; - } - case ELFDATA2MSB: { - if (__BIG_ENDIAN != __BYTE_ORDER) { - DCHECK_EQ(__BIG_ENDIAN, __BYTE_ORDER); // << ": wrong byte order"; - return; - } - break; - } - default: { - RAW_DCHECK(false, "unexpected data encoding"); // << base_as_char[EI_DATA]; - return; - } - } - - ehdr_ = reinterpret_cast(base); - const ElfW(Phdr) *dynamic_program_header = NULL; - for (int i = 0; i < ehdr_->e_phnum; ++i) { - const ElfW(Phdr) *const program_header = GetPhdr(i); - switch (program_header->p_type) { - case PT_LOAD: - if (link_base_ == ~0L) { - link_base_ = program_header->p_vaddr; - } - break; - case PT_DYNAMIC: - dynamic_program_header = program_header; - break; - } - } - if (link_base_ == ~0L || !dynamic_program_header) { - RAW_DCHECK(~0L != link_base_, "no PT_LOADs in VDSO"); - RAW_DCHECK(dynamic_program_header, "no PT_DYNAMIC in VDSO"); - // Mark this image as not present. Can not recur infinitely. - Init(0); - return; - } - ptrdiff_t relocation = - base_as_char - reinterpret_cast(link_base_); - ElfW(Dyn) *dynamic_entry = - reinterpret_cast(dynamic_program_header->p_vaddr + - relocation); - for (; dynamic_entry->d_tag != DT_NULL; ++dynamic_entry) { - ElfW(Xword) value = dynamic_entry->d_un.d_val; - if (fake_vdso) { - // A complication: in the real VDSO, dynamic entries are not relocated - // (it wasn't loaded by a dynamic loader). But when testing with a - // "fake" dlopen()ed vdso library, the loader relocates some (but - // not all!) of them before we get here. - if (dynamic_entry->d_tag == DT_VERDEF) { - // The only dynamic entry (of the ones we care about) libc-2.3.6 - // loader doesn't relocate. - value += relocation; - } - } else { - // Real VDSO. Everything needs to be relocated. - value += relocation; - } - switch (dynamic_entry->d_tag) { - case DT_HASH: - hash_ = reinterpret_cast(value); - break; - case DT_SYMTAB: - dynsym_ = reinterpret_cast(value); - break; - case DT_STRTAB: - dynstr_ = reinterpret_cast(value); - break; - case DT_VERSYM: - versym_ = reinterpret_cast(value); - break; - case DT_VERDEF: - verdef_ = reinterpret_cast(value); - break; - case DT_VERDEFNUM: - verdefnum_ = dynamic_entry->d_un.d_val; - break; - case DT_STRSZ: - strsize_ = dynamic_entry->d_un.d_val; - break; - default: - // Unrecognized entries explicitly ignored. - break; - } - } - if (!hash_ || !dynsym_ || !dynstr_ || !versym_ || - !verdef_ || !verdefnum_ || !strsize_) { - RAW_DCHECK(hash_, "invalid VDSO (no DT_HASH)"); - RAW_DCHECK(dynsym_, "invalid VDSO (no DT_SYMTAB)"); - RAW_DCHECK(dynstr_, "invalid VDSO (no DT_STRTAB)"); - RAW_DCHECK(versym_, "invalid VDSO (no DT_VERSYM)"); - RAW_DCHECK(verdef_, "invalid VDSO (no DT_VERDEF)"); - RAW_DCHECK(verdefnum_, "invalid VDSO (no DT_VERDEFNUM)"); - RAW_DCHECK(strsize_, "invalid VDSO (no DT_STRSZ)"); - // Mark this image as not present. Can not recur infinitely. - Init(0); - return; - } -} - -bool ElfMemImage::LookupSymbol(const char *name, - const char *version, - int type, - SymbolInfo *info) const { - for (SymbolIterator it = begin(); it != end(); ++it) { - if (strcmp(it->name, name) == 0 && strcmp(it->version, version) == 0 && - CurrentElfClass::ElfType(it->symbol) == type) { - if (info) { - *info = *it; - } - return true; - } - } - return false; -} - -bool ElfMemImage::LookupSymbolByAddress(const void *address, - SymbolInfo *info_out) const { - for (SymbolIterator it = begin(); it != end(); ++it) { - const char *const symbol_start = - reinterpret_cast(it->address); - const char *const symbol_end = symbol_start + it->symbol->st_size; - if (symbol_start <= address && address < symbol_end) { - if (info_out) { - // Client wants to know details for that symbol (the usual case). - if (CurrentElfClass::ElfBind(it->symbol) == STB_GLOBAL) { - // Strong symbol; just return it. - *info_out = *it; - return true; - } else { - // Weak or local. Record it, but keep looking for a strong one. - *info_out = *it; - } - } else { - // Client only cares if there is an overlapping symbol. - return true; - } - } - } - return false; -} - -ElfMemImage::SymbolIterator::SymbolIterator(const void *const image, int index) - : index_(index), image_(image) { -} - -const ElfMemImage::SymbolInfo *ElfMemImage::SymbolIterator::operator->() const { - return &info_; -} - -const ElfMemImage::SymbolInfo& ElfMemImage::SymbolIterator::operator*() const { - return info_; -} - -bool ElfMemImage::SymbolIterator::operator==(const SymbolIterator &rhs) const { - return this->image_ == rhs.image_ && this->index_ == rhs.index_; -} - -bool ElfMemImage::SymbolIterator::operator!=(const SymbolIterator &rhs) const { - return !(*this == rhs); -} - -ElfMemImage::SymbolIterator &ElfMemImage::SymbolIterator::operator++() { - this->Update(1); - return *this; -} - -ElfMemImage::SymbolIterator ElfMemImage::begin() const { - SymbolIterator it(this, 0); - it.Update(0); - return it; -} - -ElfMemImage::SymbolIterator ElfMemImage::end() const { - return SymbolIterator(this, GetNumSymbols()); -} - -void ElfMemImage::SymbolIterator::Update(int increment) { - const ElfMemImage *image = reinterpret_cast(image_); - CHECK(image->IsPresent() || increment == 0); - if (!image->IsPresent()) { - return; - } - index_ += increment; - if (index_ >= image->GetNumSymbols()) { - index_ = image->GetNumSymbols(); - return; - } - const ElfW(Sym) *symbol = image->GetDynsym(index_); - const ElfW(Versym) *version_symbol = image->GetVersym(index_); - CHECK(symbol && version_symbol); - const char *const symbol_name = image->GetDynstr(symbol->st_name); - const ElfW(Versym) version_index = version_symbol[0] & VERSYM_VERSION; - const ElfW(Verdef) *version_definition = NULL; - const char *version_name = ""; - if (symbol->st_shndx == SHN_UNDEF) { - // Undefined symbols reference DT_VERNEED, not DT_VERDEF, and - // version_index could well be greater than verdefnum_, so calling - // GetVerdef(version_index) may trigger assertion. - } else { - version_definition = image->GetVerdef(version_index); - } - if (version_definition) { - // I am expecting 1 or 2 auxiliary entries: 1 for the version itself, - // optional 2nd if the version has a parent. - CHECK_LE(1, version_definition->vd_cnt); - CHECK_LE(version_definition->vd_cnt, 2); - const ElfW(Verdaux) *version_aux = image->GetVerdefAux(version_definition); - version_name = image->GetVerstr(version_aux->vda_name); - } - info_.name = symbol_name; - info_.version = version_name; - info_.address = image->GetSymAddr(symbol); - info_.symbol = symbol; -} - -} // namespace base - -#if __clang__ -#pragma clang diagnostic pop -#endif - -#endif // HAVE_ELF_MEM_IMAGE diff --git a/contrib/libtcmalloc/src/base/elf_mem_image.h b/contrib/libtcmalloc/src/base/elf_mem_image.h deleted file mode 100644 index df63cf8b4da..00000000000 --- a/contrib/libtcmalloc/src/base/elf_mem_image.h +++ /dev/null @@ -1,135 +0,0 @@ -// -*- Mode: C++; c-basic-offset: 2; indent-tabs-mode: nil -*- -// Copyright (c) 2008, Google Inc. -// All rights reserved. -// -// Redistribution and use in source and binary forms, with or without -// modification, are permitted provided that the following conditions are -// met: -// -// * Redistributions of source code must retain the above copyright -// notice, this list of conditions and the following disclaimer. -// * Redistributions in binary form must reproduce the above -// copyright notice, this list of conditions and the following disclaimer -// in the documentation and/or other materials provided with the -// distribution. -// * Neither the name of Google Inc. nor the names of its -// contributors may be used to endorse or promote products derived from -// this software without specific prior written permission. -// -// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS -// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT -// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR -// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT -// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, -// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT -// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, -// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY -// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT -// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE -// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. - -// --- -// Author: Paul Pluzhnikov -// -// Allow dynamic symbol lookup for in-memory Elf images. - -#ifndef BASE_ELF_MEM_IMAGE_H_ -#define BASE_ELF_MEM_IMAGE_H_ - -#include "../config.h" -#ifdef HAVE_FEATURES_H -#include // for __GLIBC__ -#endif - -// Maybe one day we can rewrite this file not to require the elf -// symbol extensions in glibc, but for right now we need them. -#if defined(__ELF__) && defined(__GLIBC__) && !defined(__native_client__) - -#define HAVE_ELF_MEM_IMAGE 1 - -#include -#include // for ElfW - -namespace base { - -// An in-memory ELF image (may not exist on disk). -class ElfMemImage { - public: - // Sentinel: there could never be an elf image at this address. - static const void *const kInvalidBase; - - // Information about a single vdso symbol. - // All pointers are into .dynsym, .dynstr, or .text of the VDSO. - // Do not free() them or modify through them. - struct SymbolInfo { - const char *name; // E.g. "__vdso_getcpu" - const char *version; // E.g. "LINUX_2.6", could be "" - // for unversioned symbol. - const void *address; // Relocated symbol address. - const ElfW(Sym) *symbol; // Symbol in the dynamic symbol table. - }; - - // Supports iteration over all dynamic symbols. - class SymbolIterator { - public: - friend class ElfMemImage; - const SymbolInfo *operator->() const; - const SymbolInfo &operator*() const; - SymbolIterator& operator++(); - bool operator!=(const SymbolIterator &rhs) const; - bool operator==(const SymbolIterator &rhs) const; - private: - SymbolIterator(const void *const image, int index); - void Update(int incr); - SymbolInfo info_; - int index_; - const void *const image_; - }; - - - explicit ElfMemImage(const void *base); - void Init(const void *base); - bool IsPresent() const { return ehdr_ != NULL; } - const ElfW(Phdr)* GetPhdr(int index) const; - const ElfW(Sym)* GetDynsym(int index) const; - const ElfW(Versym)* GetVersym(int index) const; - const ElfW(Verdef)* GetVerdef(int index) const; - const ElfW(Verdaux)* GetVerdefAux(const ElfW(Verdef) *verdef) const; - const char* GetDynstr(ElfW(Word) offset) const; - const void* GetSymAddr(const ElfW(Sym) *sym) const; - const char* GetVerstr(ElfW(Word) offset) const; - int GetNumSymbols() const; - - SymbolIterator begin() const; - SymbolIterator end() const; - - // Look up versioned dynamic symbol in the image. - // Returns false if image is not present, or doesn't contain given - // symbol/version/type combination. - // If info_out != NULL, additional details are filled in. - bool LookupSymbol(const char *name, const char *version, - int symbol_type, SymbolInfo *info_out) const; - - // Find info about symbol (if any) which overlaps given address. - // Returns true if symbol was found; false if image isn't present - // or doesn't have a symbol overlapping given address. - // If info_out != NULL, additional details are filled in. - bool LookupSymbolByAddress(const void *address, SymbolInfo *info_out) const; - - private: - const ElfW(Ehdr) *ehdr_; - const ElfW(Sym) *dynsym_; - const ElfW(Versym) *versym_; - const ElfW(Verdef) *verdef_; - const ElfW(Word) *hash_; - const char *dynstr_; - size_t strsize_; - size_t verdefnum_; - ElfW(Addr) link_base_; // Link-time base (p_vaddr of first PT_LOAD). -}; - -} // namespace base - -#endif // __ELF__ and __GLIBC__ and !__native_client__ - -#endif // BASE_ELF_MEM_IMAGE_H_ diff --git a/contrib/libtcmalloc/src/base/elfcore.h b/contrib/libtcmalloc/src/base/elfcore.h deleted file mode 100644 index 98fd23b6738..00000000000 --- a/contrib/libtcmalloc/src/base/elfcore.h +++ /dev/null @@ -1,401 +0,0 @@ -// -*- Mode: C++; c-basic-offset: 2; indent-tabs-mode: nil -*- -/* Copyright (c) 2005-2008, Google Inc. - * All rights reserved. - * - * Redistribution and use in source and binary forms, with or without - * modification, are permitted provided that the following conditions are - * met: - * - * * Redistributions of source code must retain the above copyright - * notice, this list of conditions and the following disclaimer. - * * Redistributions in binary form must reproduce the above - * copyright notice, this list of conditions and the following disclaimer - * in the documentation and/or other materials provided with the - * distribution. - * * Neither the name of Google Inc. nor the names of its - * contributors may be used to endorse or promote products derived from - * this software without specific prior written permission. - * - * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS - * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT - * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR - * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT - * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, - * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT - * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, - * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY - * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT - * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE - * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. - * - * --- - * Author: Markus Gutschke, Carl Crous - */ - -#ifndef _ELFCORE_H -#define _ELFCORE_H -#ifdef __cplusplus -extern "C" { -#endif - -/* We currently only support x86-32, x86-64, ARM, MIPS, PPC on Linux. - * Porting to other related platforms should not be difficult. - */ -#if (defined(__i386__) || defined(__x86_64__) || defined(__arm__) || \ - defined(__mips__) || defined(__PPC__)) && defined(__linux) - -#include -#include -#include -#include "../config.h" - - -/* Define the DUMPER symbol to make sure that there is exactly one - * core dumper built into the library. - */ -#define DUMPER "ELF" - -/* By the time that we get a chance to read CPU registers in the - * calling thread, they are already in a not particularly useful - * state. Besides, there will be multiple frames on the stack that are - * just making the core file confusing. To fix this problem, we take a - * snapshot of the frame pointer, stack pointer, and instruction - * pointer at an earlier time, and then insert these values into the - * core file. - */ - -#if defined(__i386__) || defined(__x86_64__) - typedef struct i386_regs { /* Normal (non-FPU) CPU registers */ - #ifdef __x86_64__ - #define BP rbp - #define SP rsp - #define IP rip - uint64_t r15,r14,r13,r12,rbp,rbx,r11,r10; - uint64_t r9,r8,rax,rcx,rdx,rsi,rdi,orig_rax; - uint64_t rip,cs,eflags; - uint64_t rsp,ss; - uint64_t fs_base, gs_base; - uint64_t ds,es,fs,gs; - #else - #define BP ebp - #define SP esp - #define IP eip - uint32_t ebx, ecx, edx, esi, edi, ebp, eax; - uint16_t ds, __ds, es, __es; - uint16_t fs, __fs, gs, __gs; - uint32_t orig_eax, eip; - uint16_t cs, __cs; - uint32_t eflags, esp; - uint16_t ss, __ss; - #endif - } i386_regs; -#elif defined(__arm__) - typedef struct arm_regs { /* General purpose registers */ - #define BP uregs[11] /* Frame pointer */ - #define SP uregs[13] /* Stack pointer */ - #define IP uregs[15] /* Program counter */ - #define LR uregs[14] /* Link register */ - long uregs[18]; - } arm_regs; -#elif defined(__mips__) - typedef struct mips_regs { - unsigned long pad[6]; /* Unused padding to match kernel structures */ - unsigned long uregs[32]; /* General purpose registers. */ - unsigned long hi; /* Used for multiplication and division. */ - unsigned long lo; - unsigned long cp0_epc; /* Program counter. */ - unsigned long cp0_badvaddr; - unsigned long cp0_status; - unsigned long cp0_cause; - unsigned long unused; - } mips_regs; -#elif defined (__PPC__) - typedef struct ppc_regs { - #define SP uregs[1] /* Stack pointer */ - #define IP rip /* Program counter */ - #define LR lr /* Link register */ - unsigned long uregs[32]; /* General Purpose Registers - r0-r31. */ - double fpr[32]; /* Floating-Point Registers - f0-f31. */ - unsigned long rip; /* Program counter. */ - unsigned long msr; - unsigned long ccr; - unsigned long lr; - unsigned long ctr; - unsigned long xeq; - unsigned long mq; - } ppc_regs; -#endif - -#if defined(__i386__) && defined(__GNUC__) - /* On x86 we provide an optimized version of the FRAME() macro, if the - * compiler supports a GCC-style asm() directive. This results in somewhat - * more accurate values for CPU registers. - */ - typedef struct Frame { - struct i386_regs uregs; - int errno_; - pid_t tid; - } Frame; - #define FRAME(f) Frame f; \ - do { \ - f.errno_ = errno; \ - f.tid = sys_gettid(); \ - __asm__ volatile ( \ - "push %%ebp\n" \ - "push %%ebx\n" \ - "mov %%ebx,0(%%eax)\n" \ - "mov %%ecx,4(%%eax)\n" \ - "mov %%edx,8(%%eax)\n" \ - "mov %%esi,12(%%eax)\n" \ - "mov %%edi,16(%%eax)\n" \ - "mov %%ebp,20(%%eax)\n" \ - "mov %%eax,24(%%eax)\n" \ - "mov %%ds,%%ebx\n" \ - "mov %%ebx,28(%%eax)\n" \ - "mov %%es,%%ebx\n" \ - "mov %%ebx,32(%%eax)\n" \ - "mov %%fs,%%ebx\n" \ - "mov %%ebx,36(%%eax)\n" \ - "mov %%gs,%%ebx\n" \ - "mov %%ebx, 40(%%eax)\n" \ - "call 0f\n" \ - "0:pop %%ebx\n" \ - "add $1f-0b,%%ebx\n" \ - "mov %%ebx,48(%%eax)\n" \ - "mov %%cs,%%ebx\n" \ - "mov %%ebx,52(%%eax)\n" \ - "pushf\n" \ - "pop %%ebx\n" \ - "mov %%ebx,56(%%eax)\n" \ - "mov %%esp,%%ebx\n" \ - "add $8,%%ebx\n" \ - "mov %%ebx,60(%%eax)\n" \ - "mov %%ss,%%ebx\n" \ - "mov %%ebx,64(%%eax)\n" \ - "pop %%ebx\n" \ - "pop %%ebp\n" \ - "1:" \ - : : "a" (&f) : "memory"); \ - } while (0) - #define SET_FRAME(f,r) \ - do { \ - errno = (f).errno_; \ - (r) = (f).uregs; \ - } while (0) -#elif defined(__x86_64__) && defined(__GNUC__) - /* The FRAME and SET_FRAME macros for x86_64. */ - typedef struct Frame { - struct i386_regs uregs; - int errno_; - pid_t tid; - } Frame; - #define FRAME(f) Frame f; \ - do { \ - f.errno_ = errno; \ - f.tid = sys_gettid(); \ - __asm__ volatile ( \ - "push %%rbp\n" \ - "push %%rbx\n" \ - "mov %%r15,0(%%rax)\n" \ - "mov %%r14,8(%%rax)\n" \ - "mov %%r13,16(%%rax)\n" \ - "mov %%r12,24(%%rax)\n" \ - "mov %%rbp,32(%%rax)\n" \ - "mov %%rbx,40(%%rax)\n" \ - "mov %%r11,48(%%rax)\n" \ - "mov %%r10,56(%%rax)\n" \ - "mov %%r9,64(%%rax)\n" \ - "mov %%r8,72(%%rax)\n" \ - "mov %%rax,80(%%rax)\n" \ - "mov %%rcx,88(%%rax)\n" \ - "mov %%rdx,96(%%rax)\n" \ - "mov %%rsi,104(%%rax)\n" \ - "mov %%rdi,112(%%rax)\n" \ - "mov %%ds,%%rbx\n" \ - "mov %%rbx,184(%%rax)\n" \ - "mov %%es,%%rbx\n" \ - "mov %%rbx,192(%%rax)\n" \ - "mov %%fs,%%rbx\n" \ - "mov %%rbx,200(%%rax)\n" \ - "mov %%gs,%%rbx\n" \ - "mov %%rbx,208(%%rax)\n" \ - "call 0f\n" \ - "0:pop %%rbx\n" \ - "add $1f-0b,%%rbx\n" \ - "mov %%rbx,128(%%rax)\n" \ - "mov %%cs,%%rbx\n" \ - "mov %%rbx,136(%%rax)\n" \ - "pushf\n" \ - "pop %%rbx\n" \ - "mov %%rbx,144(%%rax)\n" \ - "mov %%rsp,%%rbx\n" \ - "add $16,%%ebx\n" \ - "mov %%rbx,152(%%rax)\n" \ - "mov %%ss,%%rbx\n" \ - "mov %%rbx,160(%%rax)\n" \ - "pop %%rbx\n" \ - "pop %%rbp\n" \ - "1:" \ - : : "a" (&f) : "memory"); \ - } while (0) - #define SET_FRAME(f,r) \ - do { \ - errno = (f).errno_; \ - (f).uregs.fs_base = (r).fs_base; \ - (f).uregs.gs_base = (r).gs_base; \ - (r) = (f).uregs; \ - } while (0) -#elif defined(__arm__) && defined(__GNUC__) - /* ARM calling conventions are a little more tricky. A little assembly - * helps in obtaining an accurate snapshot of all registers. - */ - typedef struct Frame { - struct arm_regs arm; - int errno_; - pid_t tid; - } Frame; - #define FRAME(f) Frame f; \ - do { \ - long cpsr; \ - f.errno_ = errno; \ - f.tid = sys_gettid(); \ - __asm__ volatile( \ - "stmia %0, {r0-r15}\n" /* All integer regs */\ - : : "r"(&f.arm) : "memory"); \ - f.arm.uregs[16] = 0; \ - __asm__ volatile( \ - "mrs %0, cpsr\n" /* Condition code reg */\ - : "=r"(cpsr)); \ - f.arm.uregs[17] = cpsr; \ - } while (0) - #define SET_FRAME(f,r) \ - do { \ - /* Don't override the FPU status register. */\ - /* Use the value obtained from ptrace(). This*/\ - /* works, because our code does not perform */\ - /* any FPU operations, itself. */\ - long fps = (f).arm.uregs[16]; \ - errno = (f).errno_; \ - (r) = (f).arm; \ - (r).uregs[16] = fps; \ - } while (0) -#elif defined(__mips__) && defined(__GNUC__) - typedef struct Frame { - struct mips_regs mips_regs; - int errno_; - pid_t tid; - } Frame; - #define MIPSREG(n) ({ register unsigned long r __asm__("$"#n); r; }) - #define FRAME(f) Frame f = { 0 }; \ - do { \ - unsigned long hi, lo; \ - register unsigned long pc __asm__("$31"); \ - f.mips_regs.uregs[ 0] = MIPSREG( 0); \ - f.mips_regs.uregs[ 1] = MIPSREG( 1); \ - f.mips_regs.uregs[ 2] = MIPSREG( 2); \ - f.mips_regs.uregs[ 3] = MIPSREG( 3); \ - f.mips_regs.uregs[ 4] = MIPSREG( 4); \ - f.mips_regs.uregs[ 5] = MIPSREG( 5); \ - f.mips_regs.uregs[ 6] = MIPSREG( 6); \ - f.mips_regs.uregs[ 7] = MIPSREG( 7); \ - f.mips_regs.uregs[ 8] = MIPSREG( 8); \ - f.mips_regs.uregs[ 9] = MIPSREG( 9); \ - f.mips_regs.uregs[10] = MIPSREG(10); \ - f.mips_regs.uregs[11] = MIPSREG(11); \ - f.mips_regs.uregs[12] = MIPSREG(12); \ - f.mips_regs.uregs[13] = MIPSREG(13); \ - f.mips_regs.uregs[14] = MIPSREG(14); \ - f.mips_regs.uregs[15] = MIPSREG(15); \ - f.mips_regs.uregs[16] = MIPSREG(16); \ - f.mips_regs.uregs[17] = MIPSREG(17); \ - f.mips_regs.uregs[18] = MIPSREG(18); \ - f.mips_regs.uregs[19] = MIPSREG(19); \ - f.mips_regs.uregs[20] = MIPSREG(20); \ - f.mips_regs.uregs[21] = MIPSREG(21); \ - f.mips_regs.uregs[22] = MIPSREG(22); \ - f.mips_regs.uregs[23] = MIPSREG(23); \ - f.mips_regs.uregs[24] = MIPSREG(24); \ - f.mips_regs.uregs[25] = MIPSREG(25); \ - f.mips_regs.uregs[26] = MIPSREG(26); \ - f.mips_regs.uregs[27] = MIPSREG(27); \ - f.mips_regs.uregs[28] = MIPSREG(28); \ - f.mips_regs.uregs[29] = MIPSREG(29); \ - f.mips_regs.uregs[30] = MIPSREG(30); \ - f.mips_regs.uregs[31] = MIPSREG(31); \ - __asm__ volatile ("mfhi %0" : "=r"(hi)); \ - __asm__ volatile ("mflo %0" : "=r"(lo)); \ - __asm__ volatile ("jal 1f; 1:nop" : "=r"(pc)); \ - f.mips_regs.hi = hi; \ - f.mips_regs.lo = lo; \ - f.mips_regs.cp0_epc = pc; \ - f.errno_ = errno; \ - f.tid = sys_gettid(); \ - } while (0) - #define SET_FRAME(f,r) \ - do { \ - errno = (f).errno_; \ - memcpy((r).uregs, (f).mips_regs.uregs, \ - 32*sizeof(unsigned long)); \ - (r).hi = (f).mips_regs.hi; \ - (r).lo = (f).mips_regs.lo; \ - (r).cp0_epc = (f).mips_regs.cp0_epc; \ - } while (0) -#else - /* If we do not have a hand-optimized assembly version of the FRAME() - * macro, we cannot reliably unroll the stack. So, we show a few additional - * stack frames for the coredumper. - */ - typedef struct Frame { - pid_t tid; - } Frame; - #define FRAME(f) Frame f; do { f.tid = sys_gettid(); } while (0) - #define SET_FRAME(f,r) do { } while (0) -#endif - - -/* Internal function for generating a core file. This API can change without - * notice and is only supposed to be used internally by the core dumper. - * - * This function works for both single- and multi-threaded core - * dumps. If called as - * - * FRAME(frame); - * InternalGetCoreDump(&frame, 0, NULL, ap); - * - * it creates a core file that only contains information about the - * calling thread. - * - * Optionally, the caller can provide information about other threads - * by passing their process ids in "thread_pids". The process id of - * the caller should not be included in this array. All of the threads - * must have been attached to with ptrace(), prior to calling this - * function. They will be detached when "InternalGetCoreDump()" returns. - * - * This function either returns a file handle that can be read for obtaining - * a core dump, or "-1" in case of an error. In the latter case, "errno" - * will be set appropriately. - * - * While "InternalGetCoreDump()" is not technically async signal safe, you - * might be tempted to invoke it from a signal handler. The code goes to - * great lengths to make a best effort that this will actually work. But in - * any case, you must make sure that you preserve the value of "errno" - * yourself. It is guaranteed to be clobbered otherwise. - * - * Also, "InternalGetCoreDump" is not strictly speaking re-entrant. Again, - * it makes a best effort to behave reasonably when called in a multi- - * threaded environment, but it is ultimately the caller's responsibility - * to provide locking. - */ -int InternalGetCoreDump(void *frame, int num_threads, pid_t *thread_pids, - va_list ap - /* const struct CoreDumpParameters *params, - const char *file_name, - const char *PATH - */); - -#endif - -#ifdef __cplusplus -} -#endif -#endif /* _ELFCORE_H */ diff --git a/contrib/libtcmalloc/src/base/googleinit.h b/contrib/libtcmalloc/src/base/googleinit.h deleted file mode 100644 index 3ea411a325a..00000000000 --- a/contrib/libtcmalloc/src/base/googleinit.h +++ /dev/null @@ -1,74 +0,0 @@ -// -*- Mode: C++; c-basic-offset: 2; indent-tabs-mode: nil -*- -// Copyright (c) 2005, Google Inc. -// All rights reserved. -// -// Redistribution and use in source and binary forms, with or without -// modification, are permitted provided that the following conditions are -// met: -// -// * Redistributions of source code must retain the above copyright -// notice, this list of conditions and the following disclaimer. -// * Redistributions in binary form must reproduce the above -// copyright notice, this list of conditions and the following disclaimer -// in the documentation and/or other materials provided with the -// distribution. -// * Neither the name of Google Inc. nor the names of its -// contributors may be used to endorse or promote products derived from -// this software without specific prior written permission. -// -// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS -// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT -// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR -// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT -// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, -// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT -// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, -// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY -// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT -// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE -// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. - -// --- -// Author: Jacob Hoffman-Andrews - -#ifndef _GOOGLEINIT_H -#define _GOOGLEINIT_H - -#include "base/logging.h" - -class GoogleInitializer { - public: - typedef void (*VoidFunction)(void); - GoogleInitializer(const char* name, VoidFunction ctor, VoidFunction dtor) - : name_(name), destructor_(dtor) { - RAW_VLOG(10, " constructing: %s\n", name_); - if (ctor) - ctor(); - } - ~GoogleInitializer() { - RAW_VLOG(10, " destroying: %s\n", name_); - if (destructor_) - destructor_(); - } - - private: - const char* const name_; - const VoidFunction destructor_; -}; - -#define REGISTER_MODULE_INITIALIZER(name, body) \ - namespace { \ - static void google_init_module_##name () { body; } \ - GoogleInitializer google_initializer_module_##name(#name, \ - google_init_module_##name, NULL); \ - } - -#define REGISTER_MODULE_DESTRUCTOR(name, body) \ - namespace { \ - static void google_destruct_module_##name () { body; } \ - GoogleInitializer google_destructor_module_##name(#name, \ - NULL, google_destruct_module_##name); \ - } - - -#endif /* _GOOGLEINIT_H */ diff --git a/contrib/libtcmalloc/src/base/linux_syscall_support.h b/contrib/libtcmalloc/src/base/linux_syscall_support.h deleted file mode 100644 index 6a94dc3fc72..00000000000 --- a/contrib/libtcmalloc/src/base/linux_syscall_support.h +++ /dev/null @@ -1,2880 +0,0 @@ -// -*- Mode: C++; c-basic-offset: 2; indent-tabs-mode: nil -*- -/* Copyright (c) 2005-2008, Google Inc. - * All rights reserved. - * - * Redistribution and use in source and binary forms, with or without - * modification, are permitted provided that the following conditions are - * met: - * - * * Redistributions of source code must retain the above copyright - * notice, this list of conditions and the following disclaimer. - * * Redistributions in binary form must reproduce the above - * copyright notice, this list of conditions and the following disclaimer - * in the documentation and/or other materials provided with the - * distribution. - * * Neither the name of Google Inc. nor the names of its - * contributors may be used to endorse or promote products derived from - * this software without specific prior written permission. - * - * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS - * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT - * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR - * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT - * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, - * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT - * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, - * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY - * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT - * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE - * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. - * - * --- - * Author: Markus Gutschke - */ - -/* This file includes Linux-specific support functions common to the - * coredumper and the thread lister; primarily, this is a collection - * of direct system calls, and a couple of symbols missing from - * standard header files. - * There are a few options that the including file can set to control - * the behavior of this file: - * - * SYS_CPLUSPLUS: - * The entire header file will normally be wrapped in 'extern "C" { }", - * making it suitable for compilation as both C and C++ source. If you - * do not want to do this, you can set the SYS_CPLUSPLUS macro to inhibit - * the wrapping. N.B. doing so will suppress inclusion of all prerequisite - * system header files, too. It is the caller's responsibility to provide - * the necessary definitions. - * - * SYS_ERRNO: - * All system calls will update "errno" unless overriden by setting the - * SYS_ERRNO macro prior to including this file. SYS_ERRNO should be - * an l-value. - * - * SYS_INLINE: - * New symbols will be defined "static inline", unless overridden by - * the SYS_INLINE macro. - * - * SYS_LINUX_SYSCALL_SUPPORT_H - * This macro is used to avoid multiple inclusions of this header file. - * If you need to include this file more than once, make sure to - * unset SYS_LINUX_SYSCALL_SUPPORT_H before each inclusion. - * - * SYS_PREFIX: - * New system calls will have a prefix of "sys_" unless overridden by - * the SYS_PREFIX macro. Valid values for this macro are [0..9] which - * results in prefixes "sys[0..9]_". It is also possible to set this - * macro to -1, which avoids all prefixes. - * - * This file defines a few internal symbols that all start with "LSS_". - * Do not access these symbols from outside this file. They are not part - * of the supported API. - * - * NOTE: This is a stripped down version of the official opensource - * version of linux_syscall_support.h, which lives at - * http://code.google.com/p/linux-syscall-support/ - * It includes only the syscalls that are used in perftools, plus a - * few extra. Here's the breakdown: - * 1) Perftools uses these: grep -rho 'sys_[a-z0-9_A-Z]* *(' src | sort -u - * sys__exit( - * sys_clone( - * sys_close( - * sys_fcntl( - * sys_fstat( - * sys_futex( - * sys_getcpu( - * sys_getdents64( - * sys_getppid( - * sys_gettid( - * sys_lseek( - * sys_mmap( - * sys_mremap( - * sys_munmap( - * sys_open( - * sys_pipe( - * sys_prctl( - * sys_ptrace( - * sys_ptrace_detach( - * sys_read( - * sys_sched_yield( - * sys_sigaction( - * sys_sigaltstack( - * sys_sigdelset( - * sys_sigfillset( - * sys_sigprocmask( - * sys_socket( - * sys_stat( - * sys_waitpid( - * 2) These are used as subroutines of the above: - * sys_getpid -- gettid - * sys_kill -- ptrace_detach - * sys_restore -- sigaction - * sys_restore_rt -- sigaction - * sys_socketcall -- socket - * sys_wait4 -- waitpid - * 3) I left these in even though they're not used. They either - * complement the above (write vs read) or are variants (rt_sigaction): - * sys_fstat64 - * sys_llseek - * sys_mmap2 - * sys_openat - * sys_getdents - * sys_rt_sigaction - * sys_rt_sigprocmask - * sys_sigaddset - * sys_sigemptyset - * sys_stat64 - * sys_write - */ -#ifndef SYS_LINUX_SYSCALL_SUPPORT_H -#define SYS_LINUX_SYSCALL_SUPPORT_H - -/* We currently only support x86-32, x86-64, ARM, MIPS, PPC/PPC64, Aarch64, s390 and s390x - * on Linux. - * Porting to other related platforms should not be difficult. - */ -#if (defined(__i386__) || defined(__x86_64__) || defined(__arm__) || \ - defined(__mips__) || defined(__PPC__) || \ - defined(__aarch64__) || defined(__s390__)) \ - && (defined(__linux)) - -#ifndef SYS_CPLUSPLUS -#ifdef __cplusplus -/* Some system header files in older versions of gcc neglect to properly - * handle being included from C++. As it appears to be harmless to have - * multiple nested 'extern "C"' blocks, just add another one here. - */ -extern "C" { -#endif - -#include -#include -#include -#include -#include -#include -#include -#include -#include -#include -#include -#include -#include -#include - -#ifdef __mips__ -/* Include definitions of the ABI currently in use. */ -#include -#endif - -#endif - -/* As glibc often provides subtly incompatible data structures (and implicit - * wrapper functions that convert them), we provide our own kernel data - * structures for use by the system calls. - * These structures have been developed by using Linux 2.6.23 headers for - * reference. Note though, we do not care about exact API compatibility - * with the kernel, and in fact the kernel often does not have a single - * API that works across architectures. Instead, we try to mimic the glibc - * API where reasonable, and only guarantee ABI compatibility with the - * kernel headers. - * Most notably, here are a few changes that were made to the structures - * defined by kernel headers: - * - * - we only define structures, but not symbolic names for kernel data - * types. For the latter, we directly use the native C datatype - * (i.e. "unsigned" instead of "mode_t"). - * - in a few cases, it is possible to define identical structures for - * both 32bit (e.g. i386) and 64bit (e.g. x86-64) platforms by - * standardizing on the 64bit version of the data types. In particular, - * this means that we use "unsigned" where the 32bit headers say - * "unsigned long". - * - overall, we try to minimize the number of cases where we need to - * conditionally define different structures. - * - the "struct kernel_sigaction" class of structures have been - * modified to more closely mimic glibc's API by introducing an - * anonymous union for the function pointer. - * - a small number of field names had to have an underscore appended to - * them, because glibc defines a global macro by the same name. - */ - -/* include/linux/dirent.h */ -struct kernel_dirent64 { - unsigned long long d_ino; - long long d_off; - unsigned short d_reclen; - unsigned char d_type; - char d_name[256]; -}; - -/* include/linux/dirent.h */ -struct kernel_dirent { - long d_ino; - long d_off; - unsigned short d_reclen; - char d_name[256]; -}; - -/* include/linux/time.h */ -struct kernel_timespec { - long tv_sec; - long tv_nsec; -}; - -/* include/linux/time.h */ -struct kernel_timeval { - long tv_sec; - long tv_usec; -}; - -/* include/linux/resource.h */ -struct kernel_rusage { - struct kernel_timeval ru_utime; - struct kernel_timeval ru_stime; - long ru_maxrss; - long ru_ixrss; - long ru_idrss; - long ru_isrss; - long ru_minflt; - long ru_majflt; - long ru_nswap; - long ru_inblock; - long ru_oublock; - long ru_msgsnd; - long ru_msgrcv; - long ru_nsignals; - long ru_nvcsw; - long ru_nivcsw; -}; - -#if defined(__i386__) || defined(__arm__) \ - || defined(__PPC__) || (defined(__s390__) && !defined(__s390x__)) - -/* include/asm-{arm,i386,mips,ppc}/signal.h */ -struct kernel_old_sigaction { - union { - void (*sa_handler_)(int); - void (*sa_sigaction_)(int, siginfo_t *, void *); - }; - unsigned long sa_mask; - unsigned long sa_flags; - void (*sa_restorer)(void); -} __attribute__((packed,aligned(4))); -#elif (defined(__mips__) && _MIPS_SIM == _MIPS_SIM_ABI32) - #define kernel_old_sigaction kernel_sigaction -#elif defined(__aarch64__) - // No kernel_old_sigaction defined for arm64. -#endif - -/* Some kernel functions (e.g. sigaction() in 2.6.23) require that the - * exactly match the size of the signal set, even though the API was - * intended to be extensible. We define our own KERNEL_NSIG to deal with - * this. - * Please note that glibc provides signals [1.._NSIG-1], whereas the - * kernel (and this header) provides the range [1..KERNEL_NSIG]. The - * actual number of signals is obviously the same, but the constants - * differ by one. - */ -#ifdef __mips__ -#define KERNEL_NSIG 128 -#else -#define KERNEL_NSIG 64 -#endif - -/* include/asm-{arm,i386,mips,x86_64}/signal.h */ -struct kernel_sigset_t { - unsigned long sig[(KERNEL_NSIG + 8*sizeof(unsigned long) - 1)/ - (8*sizeof(unsigned long))]; -}; - -/* include/asm-{arm,generic,i386,mips,x86_64,ppc}/signal.h */ -struct kernel_sigaction { -#ifdef __mips__ - unsigned long sa_flags; - union { - void (*sa_handler_)(int); - void (*sa_sigaction_)(int, siginfo_t *, void *); - }; - struct kernel_sigset_t sa_mask; -#else - union { - void (*sa_handler_)(int); - void (*sa_sigaction_)(int, siginfo_t *, void *); - }; - unsigned long sa_flags; - void (*sa_restorer)(void); - struct kernel_sigset_t sa_mask; -#endif -}; - -/* include/asm-{arm,i386,mips,ppc,s390}/stat.h */ -#ifdef __mips__ -#if _MIPS_SIM == _MIPS_SIM_ABI64 -struct kernel_stat { -#else -struct kernel_stat64 { -#endif - unsigned st_dev; - unsigned __pad0[3]; - unsigned long long st_ino; - unsigned st_mode; - unsigned st_nlink; - unsigned st_uid; - unsigned st_gid; - unsigned st_rdev; - unsigned __pad1[3]; - long long st_size; - unsigned st_atime_; - unsigned st_atime_nsec_; - unsigned st_mtime_; - unsigned st_mtime_nsec_; - unsigned st_ctime_; - unsigned st_ctime_nsec_; - unsigned st_blksize; - unsigned __pad2; - unsigned long long st_blocks; -}; -#elif defined __PPC__ -struct kernel_stat64 { - unsigned long long st_dev; - unsigned long long st_ino; - unsigned st_nlink; - unsigned st_mode; - unsigned st_uid; - unsigned st_gid; - int __pad2; - unsigned long long st_rdev; - long long st_size; - long long st_blksize; - long long st_blocks; - kernel_timespec st_atim; - kernel_timespec st_mtim; - kernel_timespec st_ctim; - unsigned long __unused4; - unsigned long __unused5; - unsigned long __unused6; -}; -#else -struct kernel_stat64 { - unsigned long long st_dev; - unsigned char __pad0[4]; - unsigned __st_ino; - unsigned st_mode; - unsigned st_nlink; - unsigned st_uid; - unsigned st_gid; - unsigned long long st_rdev; - unsigned char __pad3[4]; - long long st_size; - unsigned st_blksize; - unsigned long long st_blocks; - unsigned st_atime_; - unsigned st_atime_nsec_; - unsigned st_mtime_; - unsigned st_mtime_nsec_; - unsigned st_ctime_; - unsigned st_ctime_nsec_; - unsigned long long st_ino; -}; -#endif - -/* include/asm-{arm,generic,i386,mips,x86_64,ppc,s390}/stat.h */ -#if defined(__i386__) || defined(__arm__) -struct kernel_stat { - /* The kernel headers suggest that st_dev and st_rdev should be 32bit - * quantities encoding 12bit major and 20bit minor numbers in an interleaved - * format. In reality, we do not see useful data in the top bits. So, - * we'll leave the padding in here, until we find a better solution. - */ - unsigned short st_dev; - short pad1; - unsigned st_ino; - unsigned short st_mode; - unsigned short st_nlink; - unsigned short st_uid; - unsigned short st_gid; - unsigned short st_rdev; - short pad2; - unsigned st_size; - unsigned st_blksize; - unsigned st_blocks; - unsigned st_atime_; - unsigned st_atime_nsec_; - unsigned st_mtime_; - unsigned st_mtime_nsec_; - unsigned st_ctime_; - unsigned st_ctime_nsec_; - unsigned __unused4; - unsigned __unused5; -}; -#elif defined(__x86_64__) -struct kernel_stat { - uint64_t st_dev; - uint64_t st_ino; - uint64_t st_nlink; - unsigned st_mode; - unsigned st_uid; - unsigned st_gid; - unsigned __pad0; - uint64_t st_rdev; - int64_t st_size; - int64_t st_blksize; - int64_t st_blocks; - uint64_t st_atime_; - uint64_t st_atime_nsec_; - uint64_t st_mtime_; - uint64_t st_mtime_nsec_; - uint64_t st_ctime_; - uint64_t st_ctime_nsec_; - int64_t __unused[3]; -}; -#elif defined(__PPC__) -struct kernel_stat { - unsigned long long st_dev; - unsigned long st_ino; - unsigned long st_nlink; - unsigned long st_mode; - unsigned st_uid; - unsigned st_gid; - int __pad2; - unsigned long long st_rdev; - long st_size; - unsigned long st_blksize; - unsigned long st_blocks; - kernel_timespec st_atim; - kernel_timespec st_mtim; - kernel_timespec st_ctim; - unsigned long __unused4; - unsigned long __unused5; - unsigned long __unused6; -}; -#elif (defined(__mips__) && _MIPS_SIM != _MIPS_SIM_ABI64) -struct kernel_stat { - unsigned st_dev; - int st_pad1[3]; - unsigned st_ino; - unsigned st_mode; - unsigned st_nlink; - unsigned st_uid; - unsigned st_gid; - unsigned st_rdev; - int st_pad2[2]; - long st_size; - int st_pad3; - long st_atime_; - long st_atime_nsec_; - long st_mtime_; - long st_mtime_nsec_; - long st_ctime_; - long st_ctime_nsec_; - int st_blksize; - int st_blocks; - int st_pad4[14]; -}; -#elif defined(__aarch64__) -struct kernel_stat { - unsigned long st_dev; - unsigned long st_ino; - unsigned int st_mode; - unsigned int st_nlink; - unsigned int st_uid; - unsigned int st_gid; - unsigned long st_rdev; - unsigned long __pad1; - long st_size; - int st_blksize; - int __pad2; - long st_blocks; - long st_atime_; - unsigned long st_atime_nsec_; - long st_mtime_; - unsigned long st_mtime_nsec_; - long st_ctime_; - unsigned long st_ctime_nsec_; - unsigned int __unused4; - unsigned int __unused5; -}; -#elif defined(__s390x__) -struct kernel_stat { - unsigned long st_dev; - unsigned long st_ino; - unsigned long st_nlink; - unsigned int st_mode; - unsigned int st_uid; - unsigned int st_gid; - unsigned int __pad1; - unsigned long st_rdev; - unsigned long st_size; - unsigned long st_atime_; - unsigned long st_atime_nsec_; - unsigned long st_mtime_; - unsigned long st_mtime_nsec_; - unsigned long st_ctime_; - unsigned long st_ctime_nsec_; - unsigned long st_blksize; - long st_blocks; - unsigned long __unused[3]; -}; -#elif defined(__s390__) -struct kernel_stat { - unsigned short st_dev; - unsigned short __pad1; - unsigned long st_ino; - unsigned short st_mode; - unsigned short st_nlink; - unsigned short st_uid; - unsigned short st_gid; - unsigned short st_rdev; - unsigned short __pad2; - unsigned long st_size; - unsigned long st_blksize; - unsigned long st_blocks; - unsigned long st_atime_; - unsigned long st_atime_nsec_; - unsigned long st_mtime_; - unsigned long st_mtime_nsec_; - unsigned long st_ctime_; - unsigned long st_ctime_nsec_; - unsigned long __unused4; - unsigned long __unused5; -}; -#endif - - -/* Definitions missing from the standard header files */ -#ifndef O_DIRECTORY -#if defined(__arm__) -#define O_DIRECTORY 0040000 -#else -#define O_DIRECTORY 0200000 -#endif -#endif -#ifndef PR_GET_DUMPABLE -#define PR_GET_DUMPABLE 3 -#endif -#ifndef PR_SET_DUMPABLE -#define PR_SET_DUMPABLE 4 -#endif -#ifndef AT_FDCWD -#define AT_FDCWD (-100) -#endif -#ifndef AT_SYMLINK_NOFOLLOW -#define AT_SYMLINK_NOFOLLOW 0x100 -#endif -#ifndef AT_REMOVEDIR -#define AT_REMOVEDIR 0x200 -#endif -#ifndef MREMAP_FIXED -#define MREMAP_FIXED 2 -#endif -#ifndef SA_RESTORER -#define SA_RESTORER 0x04000000 -#endif - -#if defined(__i386__) -#ifndef __NR_rt_sigaction -#define __NR_rt_sigaction 174 -#define __NR_rt_sigprocmask 175 -#endif -#ifndef __NR_stat64 -#define __NR_stat64 195 -#endif -#ifndef __NR_fstat64 -#define __NR_fstat64 197 -#endif -#ifndef __NR_getdents64 -#define __NR_getdents64 220 -#endif -#ifndef __NR_gettid -#define __NR_gettid 224 -#endif -#ifndef __NR_futex -#define __NR_futex 240 -#endif -#ifndef __NR_openat -#define __NR_openat 295 -#endif -#ifndef __NR_getcpu -#define __NR_getcpu 318 -#endif -/* End of i386 definitions */ -#elif defined(__arm__) -#ifndef __syscall -#if defined(__thumb__) || defined(__ARM_EABI__) -#define __SYS_REG(name) register long __sysreg __asm__("r6") = __NR_##name; -#define __SYS_REG_LIST(regs...) [sysreg] "r" (__sysreg) , ##regs -#define __syscall(name) "swi\t0" -#define __syscall_safe(name) \ - "push {r7}\n" \ - "mov r7,%[sysreg]\n" \ - __syscall(name)"\n" \ - "pop {r7}" -#else -#define __SYS_REG(name) -#define __SYS_REG_LIST(regs...) regs -#define __syscall(name) "swi\t" __sys1(__NR_##name) "" -#define __syscall_safe(name) __syscall(name) -#endif -#endif -#ifndef __NR_rt_sigaction -#define __NR_rt_sigaction (__NR_SYSCALL_BASE + 174) -#define __NR_rt_sigprocmask (__NR_SYSCALL_BASE + 175) -#endif -#ifndef __NR_stat64 -#define __NR_stat64 (__NR_SYSCALL_BASE + 195) -#endif -#ifndef __NR_fstat64 -#define __NR_fstat64 (__NR_SYSCALL_BASE + 197) -#endif -#ifndef __NR_getdents64 -#define __NR_getdents64 (__NR_SYSCALL_BASE + 217) -#endif -#ifndef __NR_gettid -#define __NR_gettid (__NR_SYSCALL_BASE + 224) -#endif -#ifndef __NR_futex -#define __NR_futex (__NR_SYSCALL_BASE + 240) -#endif -/* End of ARM definitions */ -#elif defined(__x86_64__) -#ifndef __NR_gettid -#define __NR_gettid 186 -#endif -#ifndef __NR_futex -#define __NR_futex 202 -#endif -#ifndef __NR_getdents64 -#define __NR_getdents64 217 -#endif -#ifndef __NR_openat -#define __NR_openat 257 -#endif -/* End of x86-64 definitions */ -#elif defined(__mips__) -#if _MIPS_SIM == _MIPS_SIM_ABI32 -#ifndef __NR_rt_sigaction -#define __NR_rt_sigaction (__NR_Linux + 194) -#define __NR_rt_sigprocmask (__NR_Linux + 195) -#endif -#ifndef __NR_stat64 -#define __NR_stat64 (__NR_Linux + 213) -#endif -#ifndef __NR_fstat64 -#define __NR_fstat64 (__NR_Linux + 215) -#endif -#ifndef __NR_getdents64 -#define __NR_getdents64 (__NR_Linux + 219) -#endif -#ifndef __NR_gettid -#define __NR_gettid (__NR_Linux + 222) -#endif -#ifndef __NR_futex -#define __NR_futex (__NR_Linux + 238) -#endif -#ifndef __NR_openat -#define __NR_openat (__NR_Linux + 288) -#endif -#ifndef __NR_fstatat -#define __NR_fstatat (__NR_Linux + 293) -#endif -#ifndef __NR_getcpu -#define __NR_getcpu (__NR_Linux + 312) -#endif -/* End of MIPS (old 32bit API) definitions */ -#elif _MIPS_SIM == _MIPS_SIM_ABI64 -#ifndef __NR_gettid -#define __NR_gettid (__NR_Linux + 178) -#endif -#ifndef __NR_futex -#define __NR_futex (__NR_Linux + 194) -#endif -#ifndef __NR_openat -#define __NR_openat (__NR_Linux + 247) -#endif -#ifndef __NR_fstatat -#define __NR_fstatat (__NR_Linux + 252) -#endif -#ifndef __NR_getcpu -#define __NR_getcpu (__NR_Linux + 271) -#endif -/* End of MIPS (64bit API) definitions */ -#else -#ifndef __NR_gettid -#define __NR_gettid (__NR_Linux + 178) -#endif -#ifndef __NR_futex -#define __NR_futex (__NR_Linux + 194) -#endif -#ifndef __NR_openat -#define __NR_openat (__NR_Linux + 251) -#endif -#ifndef __NR_fstatat -#define __NR_fstatat (__NR_Linux + 256) -#endif -#ifndef __NR_getcpu -#define __NR_getcpu (__NR_Linux + 275) -#endif -/* End of MIPS (new 32bit API) definitions */ -#endif -/* End of MIPS definitions */ -#elif defined(__PPC__) -#ifndef __NR_rt_sigaction -#define __NR_rt_sigaction 173 -#define __NR_rt_sigprocmask 174 -#endif -#ifndef __NR_stat64 -#define __NR_stat64 195 -#endif -#ifndef __NR_fstat64 -#define __NR_fstat64 197 -#endif -#ifndef __NR_socket -#define __NR_socket 198 -#endif -#ifndef __NR_getdents64 -#define __NR_getdents64 202 -#endif -#ifndef __NR_gettid -#define __NR_gettid 207 -#endif -#ifndef __NR_futex -#define __NR_futex 221 -#endif -#ifndef __NR_openat -#define __NR_openat 286 -#endif -#ifndef __NR_getcpu -#define __NR_getcpu 302 -#endif -/* End of powerpc defininitions */ -#elif defined(__aarch64__) -#ifndef __NR_fstatat -#define __NR_fstatat 79 -#endif -/* End of aarch64 defininitions */ -#elif defined(__s390__) -#ifndef __NR_quotactl -#define __NR_quotactl 131 -#endif -#ifndef __NR_rt_sigreturn -#define __NR_rt_sigreturn 173 -#endif -#ifndef __NR_rt_sigaction -#define __NR_rt_sigaction 174 -#endif -#ifndef __NR_rt_sigprocmask -#define __NR_rt_sigprocmask 175 -#endif -#ifndef __NR_rt_sigpending -#define __NR_rt_sigpending 176 -#endif -#ifndef __NR_rt_sigsuspend -#define __NR_rt_sigsuspend 179 -#endif -#ifndef __NR_pread64 -#define __NR_pread64 180 -#endif -#ifndef __NR_pwrite64 -#define __NR_pwrite64 181 -#endif -#ifndef __NR_getdents64 -#define __NR_getdents64 220 -#endif -#ifndef __NR_readahead -#define __NR_readahead 222 -#endif -#ifndef __NR_setxattr -#define __NR_setxattr 224 -#endif -#ifndef __NR_lsetxattr -#define __NR_lsetxattr 225 -#endif -#ifndef __NR_getxattr -#define __NR_getxattr 227 -#endif -#ifndef __NR_lgetxattr -#define __NR_lgetxattr 228 -#endif -#ifndef __NR_listxattr -#define __NR_listxattr 230 -#endif -#ifndef __NR_llistxattr -#define __NR_llistxattr 231 -#endif -#ifndef __NR_gettid -#define __NR_gettid 236 -#endif -#ifndef __NR_tkill -#define __NR_tkill 237 -#endif -#ifndef __NR_futex -#define __NR_futex 238 -#endif -#ifndef __NR_sched_setaffinity -#define __NR_sched_setaffinity 239 -#endif -#ifndef __NR_sched_getaffinity -#define __NR_sched_getaffinity 240 -#endif -#ifndef __NR_set_tid_address -#define __NR_set_tid_address 252 -#endif -#ifndef __NR_clock_gettime -#define __NR_clock_gettime 260 -#endif -#ifndef __NR_clock_getres -#define __NR_clock_getres 261 -#endif -#ifndef __NR_statfs64 -#define __NR_statfs64 265 -#endif -#ifndef __NR_fstatfs64 -#define __NR_fstatfs64 266 -#endif -#ifndef __NR_ioprio_set -#define __NR_ioprio_set 282 -#endif -#ifndef __NR_ioprio_get -#define __NR_ioprio_get 283 -#endif -#ifndef __NR_openat -#define __NR_openat 288 -#endif -#ifndef __NR_unlinkat -#define __NR_unlinkat 294 -#endif -#ifndef __NR_move_pages -#define __NR_move_pages 310 -#endif -#ifndef __NR_getcpu -#define __NR_getcpu 311 -#endif -#ifndef __NR_fallocate -#define __NR_fallocate 314 -#endif -/* Some syscalls are named/numbered differently between s390 and s390x. */ -#ifdef __s390x__ -# ifndef __NR_getrlimit -# define __NR_getrlimit 191 -# endif -# ifndef __NR_setresuid -# define __NR_setresuid 208 -# endif -# ifndef __NR_getresuid -# define __NR_getresuid 209 -# endif -# ifndef __NR_setresgid -# define __NR_setresgid 210 -# endif -# ifndef __NR_getresgid -# define __NR_getresgid 211 -# endif -# ifndef __NR_setfsuid -# define __NR_setfsuid 215 -# endif -# ifndef __NR_setfsgid -# define __NR_setfsgid 216 -# endif -# ifndef __NR_fadvise64 -# define __NR_fadvise64 253 -# endif -# ifndef __NR_newfstatat -# define __NR_newfstatat 293 -# endif -#else /* __s390x__ */ -# ifndef __NR_getrlimit -# define __NR_getrlimit 76 -# endif -# ifndef __NR_setfsuid -# define __NR_setfsuid 138 -# endif -# ifndef __NR_setfsgid -# define __NR_setfsgid 139 -# endif -# ifndef __NR_setresuid -# define __NR_setresuid 164 -# endif -# ifndef __NR_getresuid -# define __NR_getresuid 165 -# endif -# ifndef __NR_setresgid -# define __NR_setresgid 170 -# endif -# ifndef __NR_getresgid -# define __NR_getresgid 171 -# endif -# ifndef __NR_ugetrlimit -# define __NR_ugetrlimit 191 -# endif -# ifndef __NR_mmap2 -# define __NR_mmap2 192 -# endif -# ifndef __NR_setresuid32 -# define __NR_setresuid32 208 -# endif -# ifndef __NR_getresuid32 -# define __NR_getresuid32 209 -# endif -# ifndef __NR_setresgid32 -# define __NR_setresgid32 210 -# endif -# ifndef __NR_getresgid32 -# define __NR_getresgid32 211 -# endif -# ifndef __NR_setfsuid32 -# define __NR_setfsuid32 215 -# endif -# ifndef __NR_setfsgid32 -# define __NR_setfsgid32 216 -# endif -# ifndef __NR_fadvise64_64 -# define __NR_fadvise64_64 264 -# endif -# ifndef __NR_fstatat64 -# define __NR_fstatat64 293 -# endif -#endif /* __s390__ */ -/* End of s390/s390x definitions */ -#endif - - -/* After forking, we must make sure to only call system calls. */ -#if __BOUNDED_POINTERS__ - #error "Need to port invocations of syscalls for bounded ptrs" -#else - /* The core dumper and the thread lister get executed after threads - * have been suspended. As a consequence, we cannot call any functions - * that acquire locks. Unfortunately, libc wraps most system calls - * (e.g. in order to implement pthread_atfork, and to make calls - * cancellable), which means we cannot call these functions. Instead, - * we have to call syscall() directly. - */ - #undef LSS_ERRNO - #ifdef SYS_ERRNO - /* Allow the including file to override the location of errno. This can - * be useful when using clone() with the CLONE_VM option. - */ - #define LSS_ERRNO SYS_ERRNO - #else - #define LSS_ERRNO errno - #endif - - #undef LSS_INLINE - #ifdef SYS_INLINE - #define LSS_INLINE SYS_INLINE - #else - #define LSS_INLINE static inline - #endif - - /* Allow the including file to override the prefix used for all new - * system calls. By default, it will be set to "sys_". - */ - #undef LSS_NAME - #ifndef SYS_PREFIX - #define LSS_NAME(name) sys_##name - #elif SYS_PREFIX < 0 - #define LSS_NAME(name) name - #elif SYS_PREFIX == 0 - #define LSS_NAME(name) sys0_##name - #elif SYS_PREFIX == 1 - #define LSS_NAME(name) sys1_##name - #elif SYS_PREFIX == 2 - #define LSS_NAME(name) sys2_##name - #elif SYS_PREFIX == 3 - #define LSS_NAME(name) sys3_##name - #elif SYS_PREFIX == 4 - #define LSS_NAME(name) sys4_##name - #elif SYS_PREFIX == 5 - #define LSS_NAME(name) sys5_##name - #elif SYS_PREFIX == 6 - #define LSS_NAME(name) sys6_##name - #elif SYS_PREFIX == 7 - #define LSS_NAME(name) sys7_##name - #elif SYS_PREFIX == 8 - #define LSS_NAME(name) sys8_##name - #elif SYS_PREFIX == 9 - #define LSS_NAME(name) sys9_##name - #endif - - #undef LSS_RETURN - #if (defined(__i386__) || defined(__x86_64__) || defined(__arm__) || \ - defined(__aarch64__) || defined(__s390__)) - /* Failing system calls return a negative result in the range of - * -1..-4095. These are "errno" values with the sign inverted. - */ - #define LSS_RETURN(type, res) \ - do { \ - if ((unsigned long)(res) >= (unsigned long)(-4095)) { \ - LSS_ERRNO = -(res); \ - res = -1; \ - } \ - return (type) (res); \ - } while (0) - #elif defined(__mips__) - /* On MIPS, failing system calls return -1, and set errno in a - * separate CPU register. - */ - #define LSS_RETURN(type, res, err) \ - do { \ - if (err) { \ - LSS_ERRNO = (res); \ - res = -1; \ - } \ - return (type) (res); \ - } while (0) - #elif defined(__PPC__) - /* On PPC, failing system calls return -1, and set errno in a - * separate CPU register. See linux/unistd.h. - */ - #define LSS_RETURN(type, res, err) \ - do { \ - if (err & 0x10000000 ) { \ - LSS_ERRNO = (res); \ - res = -1; \ - } \ - return (type) (res); \ - } while (0) - #endif - #if defined(__i386__) - #if defined(NO_FRAME_POINTER) && (100 * __GNUC__ + __GNUC_MINOR__ >= 404) - /* This only works for GCC-4.4 and above -- the first version to use - .cfi directives for dwarf unwind info. */ - #define CFI_ADJUST_CFA_OFFSET(adjust) \ - ".cfi_adjust_cfa_offset " #adjust "\n" - #else - #define CFI_ADJUST_CFA_OFFSET(adjust) /**/ - #endif - - /* In PIC mode (e.g. when building shared libraries), gcc for i386 - * reserves ebx. Unfortunately, most distribution ship with implementations - * of _syscallX() which clobber ebx. - * Also, most definitions of _syscallX() neglect to mark "memory" as being - * clobbered. This causes problems with compilers, that do a better job - * at optimizing across __asm__ calls. - * So, we just have to redefine all of the _syscallX() macros. - */ - #undef LSS_BODY - #define LSS_BODY(type,args...) \ - long __res; \ - __asm__ __volatile__("push %%ebx\n" \ - CFI_ADJUST_CFA_OFFSET(4) \ - "movl %2,%%ebx\n" \ - "int $0x80\n" \ - "pop %%ebx\n" \ - CFI_ADJUST_CFA_OFFSET(-4) \ - args \ - : "esp", "memory"); \ - LSS_RETURN(type,__res) - #undef _syscall0 - #define _syscall0(type,name) \ - type LSS_NAME(name)(void) { \ - long __res; \ - __asm__ volatile("int $0x80" \ - : "=a" (__res) \ - : "0" (__NR_##name) \ - : "memory"); \ - LSS_RETURN(type,__res); \ - } - #undef _syscall1 - #define _syscall1(type,name,type1,arg1) \ - type LSS_NAME(name)(type1 arg1) { \ - LSS_BODY(type, \ - : "=a" (__res) \ - : "0" (__NR_##name), "ri" ((long)(arg1))); \ - } - #undef _syscall2 - #define _syscall2(type,name,type1,arg1,type2,arg2) \ - type LSS_NAME(name)(type1 arg1,type2 arg2) { \ - LSS_BODY(type, \ - : "=a" (__res) \ - : "0" (__NR_##name),"ri" ((long)(arg1)), "c" ((long)(arg2))); \ - } - #undef _syscall3 - #define _syscall3(type,name,type1,arg1,type2,arg2,type3,arg3) \ - type LSS_NAME(name)(type1 arg1,type2 arg2,type3 arg3) { \ - LSS_BODY(type, \ - : "=a" (__res) \ - : "0" (__NR_##name), "ri" ((long)(arg1)), "c" ((long)(arg2)), \ - "d" ((long)(arg3))); \ - } - #undef _syscall4 - #define _syscall4(type,name,type1,arg1,type2,arg2,type3,arg3,type4,arg4) \ - type LSS_NAME(name)(type1 arg1, type2 arg2, type3 arg3, type4 arg4) { \ - LSS_BODY(type, \ - : "=a" (__res) \ - : "0" (__NR_##name), "ri" ((long)(arg1)), "c" ((long)(arg2)), \ - "d" ((long)(arg3)),"S" ((long)(arg4))); \ - } - #undef _syscall5 - #define _syscall5(type,name,type1,arg1,type2,arg2,type3,arg3,type4,arg4, \ - type5,arg5) \ - type LSS_NAME(name)(type1 arg1, type2 arg2, type3 arg3, type4 arg4, \ - type5 arg5) { \ - long __res; \ - __asm__ __volatile__("push %%ebx\n" \ - "movl %2,%%ebx\n" \ - "movl %1,%%eax\n" \ - "int $0x80\n" \ - "pop %%ebx" \ - : "=a" (__res) \ - : "i" (__NR_##name), "ri" ((long)(arg1)), \ - "c" ((long)(arg2)), "d" ((long)(arg3)), \ - "S" ((long)(arg4)), "D" ((long)(arg5)) \ - : "esp", "memory"); \ - LSS_RETURN(type,__res); \ - } - #undef _syscall6 - #define _syscall6(type,name,type1,arg1,type2,arg2,type3,arg3,type4,arg4, \ - type5,arg5,type6,arg6) \ - type LSS_NAME(name)(type1 arg1, type2 arg2, type3 arg3, type4 arg4, \ - type5 arg5, type6 arg6) { \ - long __res; \ - struct { long __a1; long __a6; } __s = { (long)arg1, (long) arg6 }; \ - __asm__ __volatile__("push %%ebp\n" \ - "push %%ebx\n" \ - "movl 4(%2),%%ebp\n" \ - "movl 0(%2), %%ebx\n" \ - "movl %1,%%eax\n" \ - "int $0x80\n" \ - "pop %%ebx\n" \ - "pop %%ebp" \ - : "=a" (__res) \ - : "i" (__NR_##name), "0" ((long)(&__s)), \ - "c" ((long)(arg2)), "d" ((long)(arg3)), \ - "S" ((long)(arg4)), "D" ((long)(arg5)) \ - : "esp", "memory"); \ - LSS_RETURN(type,__res); \ - } - LSS_INLINE int LSS_NAME(clone)(int (*fn)(void *), void *child_stack, - int flags, void *arg, int *parent_tidptr, - void *newtls, int *child_tidptr) { - long __res; - __asm__ __volatile__(/* if (fn == NULL) - * return -EINVAL; - */ - "movl %3,%%ecx\n" - "jecxz 1f\n" - - /* if (child_stack == NULL) - * return -EINVAL; - */ - "movl %4,%%ecx\n" - "jecxz 1f\n" - - /* Set up alignment of the child stack: - * child_stack = (child_stack & ~0xF) - 20; - */ - "andl $-16,%%ecx\n" - "subl $20,%%ecx\n" - - /* Push "arg" and "fn" onto the stack that will be - * used by the child. - */ - "movl %6,%%eax\n" - "movl %%eax,4(%%ecx)\n" - "movl %3,%%eax\n" - "movl %%eax,(%%ecx)\n" - - /* %eax = syscall(%eax = __NR_clone, - * %ebx = flags, - * %ecx = child_stack, - * %edx = parent_tidptr, - * %esi = newtls, - * %edi = child_tidptr) - * Also, make sure that %ebx gets preserved as it is - * used in PIC mode. - */ - "movl %8,%%esi\n" - "movl %7,%%edx\n" - "movl %5,%%eax\n" - "movl %9,%%edi\n" - "pushl %%ebx\n" - "movl %%eax,%%ebx\n" - "movl %2,%%eax\n" - "int $0x80\n" - - /* In the parent: restore %ebx - * In the child: move "fn" into %ebx - */ - "popl %%ebx\n" - - /* if (%eax != 0) - * return %eax; - */ - "test %%eax,%%eax\n" - "jnz 1f\n" - - /* In the child, now. Terminate frame pointer chain. - */ - "movl $0,%%ebp\n" - - /* Call "fn". "arg" is already on the stack. - */ - "call *%%ebx\n" - - /* Call _exit(%ebx). Unfortunately older versions - * of gcc restrict the number of arguments that can - * be passed to asm(). So, we need to hard-code the - * system call number. - */ - "movl %%eax,%%ebx\n" - "movl $1,%%eax\n" - "int $0x80\n" - - /* Return to parent. - */ - "1:\n" - : "=a" (__res) - : "0"(-EINVAL), "i"(__NR_clone), - "m"(fn), "m"(child_stack), "m"(flags), "m"(arg), - "m"(parent_tidptr), "m"(newtls), "m"(child_tidptr) - : "esp", "memory", "ecx", "edx", "esi", "edi"); - LSS_RETURN(int, __res); - } - - LSS_INLINE void (*LSS_NAME(restore_rt)(void))(void) { - /* On i386, the kernel does not know how to return from a signal - * handler. Instead, it relies on user space to provide a - * restorer function that calls the {rt_,}sigreturn() system call. - * Unfortunately, we cannot just reference the glibc version of this - * function, as glibc goes out of its way to make it inaccessible. - */ - void (*res)(void); - __asm__ __volatile__("call 2f\n" - "0:.align 16\n" - "1:movl %1,%%eax\n" - "int $0x80\n" - "2:popl %0\n" - "addl $(1b-0b),%0\n" - : "=a" (res) - : "i" (__NR_rt_sigreturn)); - return res; - } - LSS_INLINE void (*LSS_NAME(restore)(void))(void) { - /* On i386, the kernel does not know how to return from a signal - * handler. Instead, it relies on user space to provide a - * restorer function that calls the {rt_,}sigreturn() system call. - * Unfortunately, we cannot just reference the glibc version of this - * function, as glibc goes out of its way to make it inaccessible. - */ - void (*res)(void); - __asm__ __volatile__("call 2f\n" - "0:.align 16\n" - "1:pop %%eax\n" - "movl %1,%%eax\n" - "int $0x80\n" - "2:popl %0\n" - "addl $(1b-0b),%0\n" - : "=a" (res) - : "i" (__NR_sigreturn)); - return res; - } - #elif defined(__x86_64__) - /* There are no known problems with any of the _syscallX() macros - * currently shipping for x86_64, but we still need to be able to define - * our own version so that we can override the location of the errno - * location (e.g. when using the clone() system call with the CLONE_VM - * option). - */ - #undef LSS_ENTRYPOINT - #define LSS_ENTRYPOINT "syscall\n" - - /* The x32 ABI has 32 bit longs, but the syscall interface is 64 bit. - * We need to explicitly cast to an unsigned 64 bit type to avoid implicit - * sign extension. We can't cast pointers directly because those are - * 32 bits, and gcc will dump ugly warnings about casting from a pointer - * to an integer of a different size. - */ - #undef LSS_SYSCALL_ARG - #define LSS_SYSCALL_ARG(a) ((uint64_t)(uintptr_t)(a)) - #undef _LSS_RETURN - #define _LSS_RETURN(type, res, cast) \ - do { \ - if ((uint64_t)(res) >= (uint64_t)(-4095)) { \ - LSS_ERRNO = -(res); \ - res = -1; \ - } \ - return (type)(cast)(res); \ - } while (0) - #undef LSS_RETURN - #define LSS_RETURN(type, res) _LSS_RETURN(type, res, uintptr_t) - - #undef _LSS_BODY - #define _LSS_BODY(nr, type, name, cast, ...) \ - long long __res; \ - __asm__ __volatile__(LSS_BODY_ASM##nr LSS_ENTRYPOINT \ - : "=a" (__res) \ - : "0" (__NR_##name) LSS_BODY_ARG##nr(__VA_ARGS__) \ - : LSS_BODY_CLOBBER##nr "r11", "rcx", "memory"); \ - _LSS_RETURN(type, __res, cast) - #undef LSS_BODY - #define LSS_BODY(nr, type, name, args...) \ - _LSS_BODY(nr, type, name, uintptr_t, ## args) - - #undef LSS_BODY_ASM0 - #undef LSS_BODY_ASM1 - #undef LSS_BODY_ASM2 - #undef LSS_BODY_ASM3 - #undef LSS_BODY_ASM4 - #undef LSS_BODY_ASM5 - #undef LSS_BODY_ASM6 - #define LSS_BODY_ASM0 - #define LSS_BODY_ASM1 LSS_BODY_ASM0 - #define LSS_BODY_ASM2 LSS_BODY_ASM1 - #define LSS_BODY_ASM3 LSS_BODY_ASM2 - #define LSS_BODY_ASM4 LSS_BODY_ASM3 "movq %5,%%r10;" - #define LSS_BODY_ASM5 LSS_BODY_ASM4 "movq %6,%%r8;" - #define LSS_BODY_ASM6 LSS_BODY_ASM5 "movq %7,%%r9;" - - #undef LSS_BODY_CLOBBER0 - #undef LSS_BODY_CLOBBER1 - #undef LSS_BODY_CLOBBER2 - #undef LSS_BODY_CLOBBER3 - #undef LSS_BODY_CLOBBER4 - #undef LSS_BODY_CLOBBER5 - #undef LSS_BODY_CLOBBER6 - #define LSS_BODY_CLOBBER0 - #define LSS_BODY_CLOBBER1 LSS_BODY_CLOBBER0 - #define LSS_BODY_CLOBBER2 LSS_BODY_CLOBBER1 - #define LSS_BODY_CLOBBER3 LSS_BODY_CLOBBER2 - #define LSS_BODY_CLOBBER4 LSS_BODY_CLOBBER3 "r10", - #define LSS_BODY_CLOBBER5 LSS_BODY_CLOBBER4 "r8", - #define LSS_BODY_CLOBBER6 LSS_BODY_CLOBBER5 "r9", - - #undef LSS_BODY_ARG0 - #undef LSS_BODY_ARG1 - #undef LSS_BODY_ARG2 - #undef LSS_BODY_ARG3 - #undef LSS_BODY_ARG4 - #undef LSS_BODY_ARG5 - #undef LSS_BODY_ARG6 - #define LSS_BODY_ARG0() - #define LSS_BODY_ARG1(arg1) \ - LSS_BODY_ARG0(), "D" (arg1) - #define LSS_BODY_ARG2(arg1, arg2) \ - LSS_BODY_ARG1(arg1), "S" (arg2) - #define LSS_BODY_ARG3(arg1, arg2, arg3) \ - LSS_BODY_ARG2(arg1, arg2), "d" (arg3) - #define LSS_BODY_ARG4(arg1, arg2, arg3, arg4) \ - LSS_BODY_ARG3(arg1, arg2, arg3), "r" (arg4) - #define LSS_BODY_ARG5(arg1, arg2, arg3, arg4, arg5) \ - LSS_BODY_ARG4(arg1, arg2, arg3, arg4), "r" (arg5) - #define LSS_BODY_ARG6(arg1, arg2, arg3, arg4, arg5, arg6) \ - LSS_BODY_ARG5(arg1, arg2, arg3, arg4, arg5), "r" (arg6) - - #undef _syscall0 - #define _syscall0(type,name) \ - type LSS_NAME(name)() { \ - LSS_BODY(0, type, name); \ - } - #undef _syscall1 - #define _syscall1(type,name,type1,arg1) \ - type LSS_NAME(name)(type1 arg1) { \ - LSS_BODY(1, type, name, LSS_SYSCALL_ARG(arg1)); \ - } - #undef _syscall2 - #define _syscall2(type,name,type1,arg1,type2,arg2) \ - type LSS_NAME(name)(type1 arg1, type2 arg2) { \ - LSS_BODY(2, type, name, LSS_SYSCALL_ARG(arg1), LSS_SYSCALL_ARG(arg2));\ - } - #undef _syscall3 - #define _syscall3(type,name,type1,arg1,type2,arg2,type3,arg3) \ - type LSS_NAME(name)(type1 arg1, type2 arg2, type3 arg3) { \ - LSS_BODY(3, type, name, LSS_SYSCALL_ARG(arg1), LSS_SYSCALL_ARG(arg2), \ - LSS_SYSCALL_ARG(arg3)); \ - } - #undef _syscall4 - #define _syscall4(type,name,type1,arg1,type2,arg2,type3,arg3,type4,arg4) \ - type LSS_NAME(name)(type1 arg1, type2 arg2, type3 arg3, type4 arg4) { \ - LSS_BODY(4, type, name, LSS_SYSCALL_ARG(arg1), LSS_SYSCALL_ARG(arg2), \ - LSS_SYSCALL_ARG(arg3), LSS_SYSCALL_ARG(arg4));\ - } - #undef _syscall5 - #define _syscall5(type,name,type1,arg1,type2,arg2,type3,arg3,type4,arg4, \ - type5,arg5) \ - type LSS_NAME(name)(type1 arg1, type2 arg2, type3 arg3, type4 arg4, \ - type5 arg5) { \ - LSS_BODY(5, type, name, LSS_SYSCALL_ARG(arg1), LSS_SYSCALL_ARG(arg2), \ - LSS_SYSCALL_ARG(arg3), LSS_SYSCALL_ARG(arg4), \ - LSS_SYSCALL_ARG(arg5)); \ - } - #undef _syscall6 - #define _syscall6(type,name,type1,arg1,type2,arg2,type3,arg3,type4,arg4, \ - type5,arg5,type6,arg6) \ - type LSS_NAME(name)(type1 arg1, type2 arg2, type3 arg3, type4 arg4, \ - type5 arg5, type6 arg6) { \ - LSS_BODY(6, type, name, LSS_SYSCALL_ARG(arg1), LSS_SYSCALL_ARG(arg2), \ - LSS_SYSCALL_ARG(arg3), LSS_SYSCALL_ARG(arg4), \ - LSS_SYSCALL_ARG(arg5), LSS_SYSCALL_ARG(arg6));\ - } - LSS_INLINE int LSS_NAME(clone)(int (*fn)(void *), void *child_stack, - int flags, void *arg, int *parent_tidptr, - void *newtls, int *child_tidptr) { - long long __res; - { - __asm__ __volatile__(/* if (fn == NULL) - * return -EINVAL; - */ - "testq %4,%4\n" - "jz 1f\n" - - /* if (child_stack == NULL) - * return -EINVAL; - */ - "testq %5,%5\n" - "jz 1f\n" - - /* Set up alignment of the child stack: - * child_stack = (child_stack & ~0xF) - 16; - */ - "andq $-16,%5\n" - "subq $16,%5\n" - - /* Push "arg" and "fn" onto the stack that will be - * used by the child. - */ - "movq %7,8(%5)\n" - "movq %4,0(%5)\n" - - /* %rax = syscall(%rax = __NR_clone, - * %rdi = flags, - * %rsi = child_stack, - * %rdx = parent_tidptr, - * %r8 = new_tls, - * %r10 = child_tidptr) - */ - "movq %2,%%rax\n" - "movq %9,%%r8\n" - "movq %10,%%r10\n" - "syscall\n" - - /* if (%rax != 0) - * return; - */ - "testq %%rax,%%rax\n" - "jnz 1f\n" - - /* In the child. Terminate frame pointer chain. - */ - "xorq %%rbp,%%rbp\n" - - /* Call "fn(arg)". - */ - "popq %%rax\n" - "popq %%rdi\n" - "call *%%rax\n" - - /* Call _exit(%ebx). - */ - "movq %%rax,%%rdi\n" - "movq %3,%%rax\n" - "syscall\n" - - /* Return to parent. - */ - "1:\n" - : "=a" (__res) - : "0"(-EINVAL), "i"(__NR_clone), "i"(__NR_exit), - "r"(LSS_SYSCALL_ARG(fn)), - "S"(LSS_SYSCALL_ARG(child_stack)), - "D"(LSS_SYSCALL_ARG(flags)), - "r"(LSS_SYSCALL_ARG(arg)), - "d"(LSS_SYSCALL_ARG(parent_tidptr)), - "r"(LSS_SYSCALL_ARG(newtls)), - "r"(LSS_SYSCALL_ARG(child_tidptr)) - : "rsp", "memory", "r8", "r10", "r11", "rcx"); - } - LSS_RETURN(int, __res); - } - - LSS_INLINE void (*LSS_NAME(restore_rt)(void))(void) { - /* On x86-64, the kernel does not know how to return from - * a signal handler. Instead, it relies on user space to provide a - * restorer function that calls the rt_sigreturn() system call. - * Unfortunately, we cannot just reference the glibc version of this - * function, as glibc goes out of its way to make it inaccessible. - */ - long long res; - __asm__ __volatile__("call 2f\n" - "0:.align 16\n" - "1:movq %1,%%rax\n" - "syscall\n" - "2:popq %0\n" - "addq $(1b-0b),%0\n" - : "=a" (res) - : "i" (__NR_rt_sigreturn)); - return (void (*)(void))(uintptr_t)res; - } - #elif defined(__arm__) - /* Most definitions of _syscallX() neglect to mark "memory" as being - * clobbered. This causes problems with compilers, that do a better job - * at optimizing across __asm__ calls. - * So, we just have to redefine all fo the _syscallX() macros. - */ - #undef LSS_REG - #define LSS_REG(r,a) register long __r##r __asm__("r"#r) = (long)a - - /* r0..r3 are scratch registers and not preserved across function - * calls. We need to first evaluate the first 4 syscall arguments - * and store them on stack. They must be loaded into r0..r3 after - * all function calls to avoid r0..r3 being clobbered. - */ - #undef LSS_SAVE_ARG - #define LSS_SAVE_ARG(r,a) long __tmp##r = (long)a - #undef LSS_LOAD_ARG - #define LSS_LOAD_ARG(r) register long __r##r __asm__("r"#r) = __tmp##r - - #undef LSS_BODY - #define LSS_BODY(type, name, args...) \ - register long __res_r0 __asm__("r0"); \ - long __res; \ - __SYS_REG(name) \ - __asm__ __volatile__ (__syscall_safe(name) \ - : "=r"(__res_r0) \ - : __SYS_REG_LIST(args) \ - : "lr", "memory"); \ - __res = __res_r0; \ - LSS_RETURN(type, __res) - #undef _syscall0 - #define _syscall0(type, name) \ - type LSS_NAME(name)() { \ - LSS_BODY(type, name); \ - } - #undef _syscall1 - #define _syscall1(type, name, type1, arg1) \ - type LSS_NAME(name)(type1 arg1) { \ - /* There is no need for using a volatile temp. */ \ - LSS_REG(0, arg1); \ - LSS_BODY(type, name, "r"(__r0)); \ - } - #undef _syscall2 - #define _syscall2(type, name, type1, arg1, type2, arg2) \ - type LSS_NAME(name)(type1 arg1, type2 arg2) { \ - LSS_SAVE_ARG(0, arg1); \ - LSS_SAVE_ARG(1, arg2); \ - LSS_LOAD_ARG(0); \ - LSS_LOAD_ARG(1); \ - LSS_BODY(type, name, "r"(__r0), "r"(__r1)); \ - } - #undef _syscall3 - #define _syscall3(type, name, type1, arg1, type2, arg2, type3, arg3) \ - type LSS_NAME(name)(type1 arg1, type2 arg2, type3 arg3) { \ - LSS_SAVE_ARG(0, arg1); \ - LSS_SAVE_ARG(1, arg2); \ - LSS_SAVE_ARG(2, arg3); \ - LSS_LOAD_ARG(0); \ - LSS_LOAD_ARG(1); \ - LSS_LOAD_ARG(2); \ - LSS_BODY(type, name, "r"(__r0), "r"(__r1), "r"(__r2)); \ - } - #undef _syscall4 - #define _syscall4(type, name, type1, arg1, type2, arg2, type3, arg3, \ - type4, arg4) \ - type LSS_NAME(name)(type1 arg1, type2 arg2, type3 arg3, type4 arg4) { \ - LSS_SAVE_ARG(0, arg1); \ - LSS_SAVE_ARG(1, arg2); \ - LSS_SAVE_ARG(2, arg3); \ - LSS_SAVE_ARG(3, arg4); \ - LSS_LOAD_ARG(0); \ - LSS_LOAD_ARG(1); \ - LSS_LOAD_ARG(2); \ - LSS_LOAD_ARG(3); \ - LSS_BODY(type, name, "r"(__r0), "r"(__r1), "r"(__r2), "r"(__r3)); \ - } - #undef _syscall5 - #define _syscall5(type, name, type1, arg1, type2, arg2, type3, arg3, \ - type4, arg4, type5, arg5) \ - type LSS_NAME(name)(type1 arg1, type2 arg2, type3 arg3, type4 arg4, \ - type5 arg5) { \ - LSS_SAVE_ARG(0, arg1); \ - LSS_SAVE_ARG(1, arg2); \ - LSS_SAVE_ARG(2, arg3); \ - LSS_SAVE_ARG(3, arg4); \ - LSS_REG(4, arg5); \ - LSS_LOAD_ARG(0); \ - LSS_LOAD_ARG(1); \ - LSS_LOAD_ARG(2); \ - LSS_LOAD_ARG(3); \ - LSS_BODY(type, name, "r"(__r0), "r"(__r1), "r"(__r2), "r"(__r3), \ - "r"(__r4)); \ - } - #undef _syscall6 - #define _syscall6(type, name, type1, arg1, type2, arg2, type3, arg3, \ - type4, arg4, type5, arg5, type6, arg6) \ - type LSS_NAME(name)(type1 arg1, type2 arg2, type3 arg3, type4 arg4, \ - type5 arg5, type6 arg6) { \ - LSS_SAVE_ARG(0, arg1); \ - LSS_SAVE_ARG(1, arg2); \ - LSS_SAVE_ARG(2, arg3); \ - LSS_SAVE_ARG(3, arg4); \ - LSS_REG(4, arg5); \ - LSS_REG(5, arg6); \ - LSS_LOAD_ARG(0); \ - LSS_LOAD_ARG(1); \ - LSS_LOAD_ARG(2); \ - LSS_LOAD_ARG(3); \ - LSS_BODY(type, name, "r"(__r0), "r"(__r1), "r"(__r2), "r"(__r3), \ - "r"(__r4), "r"(__r5)); \ - } - LSS_INLINE int LSS_NAME(clone)(int (*fn)(void *), void *child_stack, - int flags, void *arg, int *parent_tidptr, - void *newtls, int *child_tidptr) { - register long __res __asm__("r5"); - { - if (fn == NULL || child_stack == NULL) { - __res = -EINVAL; - goto clone_exit; - } - - /* stash first 4 arguments on stack first because we can only load - * them after all function calls. - */ - int tmp_flags = flags; - int * tmp_stack = (int*) child_stack; - void * tmp_ptid = parent_tidptr; - void * tmp_tls = newtls; - - register int *__ctid __asm__("r4") = child_tidptr; - - /* Push "arg" and "fn" onto the stack that will be - * used by the child. - */ - *(--tmp_stack) = (int) arg; - *(--tmp_stack) = (int) fn; - - /* We must load r0..r3 last after all possible function calls. */ - register int __flags __asm__("r0") = tmp_flags; - register void *__stack __asm__("r1") = tmp_stack; - register void *__ptid __asm__("r2") = tmp_ptid; - register void *__tls __asm__("r3") = tmp_tls; - - /* %r0 = syscall(%r0 = flags, - * %r1 = child_stack, - * %r2 = parent_tidptr, - * %r3 = newtls, - * %r4 = child_tidptr) - */ - __SYS_REG(clone) - __asm__ __volatile__(/* %r0 = syscall(%r0 = flags, - * %r1 = child_stack, - * %r2 = parent_tidptr, - * %r3 = newtls, - * %r4 = child_tidptr) - */ - "push {r7}\n" - "mov r7,%1\n" - __syscall(clone)"\n" - - /* if (%r0 != 0) - * return %r0; - */ - "movs %0,r0\n" - "bne 1f\n" - - /* In the child, now. Call "fn(arg)". - */ - "ldr r0,[sp, #4]\n" - "mov lr,pc\n" - "ldr pc,[sp]\n" - - /* Call _exit(%r0), which never returns. We only - * need to set r7 for EABI syscall ABI but we do - * this always to simplify code sharing between - * old and new syscall ABIs. - */ - "mov r7,%2\n" - __syscall(exit)"\n" - - /* Pop r7 from the stack only in the parent. - */ - "1: pop {r7}\n" - : "=r" (__res) - : "r"(__sysreg), - "i"(__NR_exit), "r"(__stack), "r"(__flags), - "r"(__ptid), "r"(__tls), "r"(__ctid) - : "cc", "lr", "memory"); - } - clone_exit: - LSS_RETURN(int, __res); - } - #elif defined(__mips__) - #undef LSS_REG - #define LSS_REG(r,a) register unsigned long __r##r __asm__("$"#r) = \ - (unsigned long)(a) - - #if _MIPS_SIM == _MIPS_SIM_ABI32 - // See http://sources.redhat.com/ml/libc-alpha/2004-10/msg00050.html - // or http://www.linux-mips.org/archives/linux-mips/2004-10/msg00142.html - #define MIPS_SYSCALL_CLOBBERS "$1", "$3", "$8", "$9", "$10", "$11", "$12",\ - "$13", "$14", "$15", "$24", "$25", "memory" - #else - #define MIPS_SYSCALL_CLOBBERS "$1", "$3", "$10", "$11", "$12", "$13", \ - "$14", "$15", "$24", "$25", "memory" - #endif - - #undef LSS_BODY - #define LSS_BODY(type,name,r7,...) \ - register unsigned long __v0 __asm__("$2") = __NR_##name; \ - __asm__ __volatile__ ("syscall\n" \ - : "=&r"(__v0), r7 (__r7) \ - : "0"(__v0), ##__VA_ARGS__ \ - : MIPS_SYSCALL_CLOBBERS); \ - LSS_RETURN(type, __v0, __r7) - #undef _syscall0 - #define _syscall0(type, name) \ - type LSS_NAME(name)() { \ - register unsigned long __r7 __asm__("$7"); \ - LSS_BODY(type, name, "=r"); \ - } - #undef _syscall1 - #define _syscall1(type, name, type1, arg1) \ - type LSS_NAME(name)(type1 arg1) { \ - register unsigned long __r7 __asm__("$7"); \ - LSS_REG(4, arg1); LSS_BODY(type, name, "=r", "r"(__r4)); \ - } - #undef _syscall2 - #define _syscall2(type, name, type1, arg1, type2, arg2) \ - type LSS_NAME(name)(type1 arg1, type2 arg2) { \ - register unsigned long __r7 __asm__("$7"); \ - LSS_REG(4, arg1); LSS_REG(5, arg2); \ - LSS_BODY(type, name, "=r", "r"(__r4), "r"(__r5)); \ - } - #undef _syscall3 - #define _syscall3(type, name, type1, arg1, type2, arg2, type3, arg3) \ - type LSS_NAME(name)(type1 arg1, type2 arg2, type3 arg3) { \ - register unsigned long __r7 __asm__("$7"); \ - LSS_REG(4, arg1); LSS_REG(5, arg2); LSS_REG(6, arg3); \ - LSS_BODY(type, name, "=r", "r"(__r4), "r"(__r5), "r"(__r6)); \ - } - #undef _syscall4 - #define _syscall4(type,name,type1,arg1,type2,arg2,type3,arg3,type4,arg4) \ - type LSS_NAME(name)(type1 arg1, type2 arg2, type3 arg3, type4 arg4) { \ - LSS_REG(4, arg1); LSS_REG(5, arg2); LSS_REG(6, arg3); \ - LSS_REG(7, arg4); \ - LSS_BODY(type, name, "+r", "r"(__r4), "r"(__r5), "r"(__r6)); \ - } - #undef _syscall5 - #if _MIPS_SIM == _MIPS_SIM_ABI32 - /* The old 32bit MIPS system call API passes the fifth and sixth argument - * on the stack, whereas the new APIs use registers "r8" and "r9". - */ - #define _syscall5(type,name,type1,arg1,type2,arg2,type3,arg3,type4,arg4, \ - type5,arg5) \ - type LSS_NAME(name)(type1 arg1, type2 arg2, type3 arg3, type4 arg4, \ - type5 arg5) { \ - LSS_REG(4, arg1); LSS_REG(5, arg2); LSS_REG(6, arg3); \ - LSS_REG(7, arg4); \ - register unsigned long __v0 __asm__("$2"); \ - __asm__ __volatile__ (".set noreorder\n" \ - "lw $2, %6\n" \ - "subu $29, 32\n" \ - "sw $2, 16($29)\n" \ - "li $2, %2\n" \ - "syscall\n" \ - "addiu $29, 32\n" \ - ".set reorder\n" \ - : "=&r"(__v0), "+r" (__r7) \ - : "i" (__NR_##name), "r"(__r4), "r"(__r5), \ - "r"(__r6), "m" ((unsigned long)arg5) \ - : MIPS_SYSCALL_CLOBBERS); \ - LSS_RETURN(type, __v0, __r7); \ - } - #else - #define _syscall5(type,name,type1,arg1,type2,arg2,type3,arg3,type4,arg4, \ - type5,arg5) \ - type LSS_NAME(name)(type1 arg1, type2 arg2, type3 arg3, type4 arg4, \ - type5 arg5) { \ - LSS_REG(4, arg1); LSS_REG(5, arg2); LSS_REG(6, arg3); \ - LSS_REG(7, arg4); LSS_REG(8, arg5); \ - LSS_BODY(type, name, "+r", "r"(__r4), "r"(__r5), "r"(__r6), \ - "r"(__r8)); \ - } - #endif - #undef _syscall6 - #if _MIPS_SIM == _MIPS_SIM_ABI32 - /* The old 32bit MIPS system call API passes the fifth and sixth argument - * on the stack, whereas the new APIs use registers "r8" and "r9". - */ - #define _syscall6(type,name,type1,arg1,type2,arg2,type3,arg3,type4,arg4, \ - type5,arg5,type6,arg6) \ - type LSS_NAME(name)(type1 arg1, type2 arg2, type3 arg3, type4 arg4, \ - type5 arg5, type6 arg6) { \ - LSS_REG(4, arg1); LSS_REG(5, arg2); LSS_REG(6, arg3); \ - LSS_REG(7, arg4); \ - register unsigned long __v0 __asm__("$2"); \ - __asm__ __volatile__ (".set noreorder\n" \ - "lw $2, %6\n" \ - "lw $8, %7\n" \ - "subu $29, 32\n" \ - "sw $2, 16($29)\n" \ - "sw $8, 20($29)\n" \ - "li $2, %2\n" \ - "syscall\n" \ - "addiu $29, 32\n" \ - ".set reorder\n" \ - : "=&r"(__v0), "+r" (__r7) \ - : "i" (__NR_##name), "r"(__r4), "r"(__r5), \ - "r"(__r6), "m" ((unsigned long)arg5), \ - "m" ((unsigned long)arg6) \ - : MIPS_SYSCALL_CLOBBERS); \ - LSS_RETURN(type, __v0, __r7); \ - } - #else - #define _syscall6(type,name,type1,arg1,type2,arg2,type3,arg3,type4,arg4, \ - type5,arg5,type6,arg6) \ - type LSS_NAME(name)(type1 arg1, type2 arg2, type3 arg3, type4 arg4, \ - type5 arg5,type6 arg6) { \ - LSS_REG(4, arg1); LSS_REG(5, arg2); LSS_REG(6, arg3); \ - LSS_REG(7, arg4); LSS_REG(8, arg5); LSS_REG(9, arg6); \ - LSS_BODY(type, name, "+r", "r"(__r4), "r"(__r5), "r"(__r6), \ - "r"(__r8), "r"(__r9)); \ - } - #endif - LSS_INLINE int LSS_NAME(clone)(int (*fn)(void *), void *child_stack, - int flags, void *arg, int *parent_tidptr, - void *newtls, int *child_tidptr) { - register unsigned long __v0 __asm__("$2"); - register unsigned long __r7 __asm__("$7") = (unsigned long)newtls; - { - register int __flags __asm__("$4") = flags; - register void *__stack __asm__("$5") = child_stack; - register void *__ptid __asm__("$6") = parent_tidptr; - register int *__ctid __asm__("$8") = child_tidptr; - __asm__ __volatile__( - #if _MIPS_SIM == _MIPS_SIM_ABI32 && _MIPS_SZPTR == 32 - "subu $29,24\n" - #elif _MIPS_SIM == _MIPS_SIM_NABI32 - "sub $29,16\n" - #else - "dsubu $29,16\n" - #endif - - /* if (fn == NULL || child_stack == NULL) - * return -EINVAL; - */ - "li %0,%2\n" - "beqz %5,1f\n" - "beqz %6,1f\n" - - /* Push "arg" and "fn" onto the stack that will be - * used by the child. - */ - #if _MIPS_SIM == _MIPS_SIM_ABI32 && _MIPS_SZPTR == 32 - "subu %6,32\n" - "sw %5,0(%6)\n" - "sw %8,4(%6)\n" - #elif _MIPS_SIM == _MIPS_SIM_NABI32 - "sub %6,32\n" - "sw %5,0(%6)\n" - "sw %8,8(%6)\n" - #else - "dsubu %6,32\n" - "sd %5,0(%6)\n" - "sd %8,8(%6)\n" - #endif - - /* $7 = syscall($4 = flags, - * $5 = child_stack, - * $6 = parent_tidptr, - * $7 = newtls, - * $8 = child_tidptr) - */ - "li $2,%3\n" - "syscall\n" - - /* if ($7 != 0) - * return $2; - */ - "bnez $7,1f\n" - "bnez $2,1f\n" - - /* In the child, now. Call "fn(arg)". - */ - #if _MIPS_SIM == _MIPS_SIM_ABI32 && _MIPS_SZPTR == 32 - "lw $25,0($29)\n" - "lw $4,4($29)\n" - #elif _MIPS_SIM == _MIPS_SIM_NABI32 - "lw $25,0($29)\n" - "lw $4,8($29)\n" - #else - "ld $25,0($29)\n" - "ld $4,8($29)\n" - #endif - "jalr $25\n" - - /* Call _exit($2) - */ - "move $4,$2\n" - "li $2,%4\n" - "syscall\n" - - "1:\n" - #if _MIPS_SIM == _MIPS_SIM_ABI32 && _MIPS_SZPTR == 32 - "addu $29, 24\n" - #elif _MIPS_SIM == _MIPS_SIM_NABI32 - "add $29, 16\n" - #else - "daddu $29,16\n" - #endif - : "=&r" (__v0), "=r" (__r7) - : "i"(-EINVAL), "i"(__NR_clone), "i"(__NR_exit), - "r"(fn), "r"(__stack), "r"(__flags), "r"(arg), - "r"(__ptid), "r"(__r7), "r"(__ctid) - : "$9", "$10", "$11", "$12", "$13", "$14", "$15", - "$24", "memory"); - } - LSS_RETURN(int, __v0, __r7); - } - #elif defined (__PPC__) - #undef LSS_LOADARGS_0 - #define LSS_LOADARGS_0(name, dummy...) \ - __sc_0 = __NR_##name - #undef LSS_LOADARGS_1 - #define LSS_LOADARGS_1(name, arg1) \ - LSS_LOADARGS_0(name); \ - __sc_3 = (unsigned long) (arg1) - #undef LSS_LOADARGS_2 - #define LSS_LOADARGS_2(name, arg1, arg2) \ - LSS_LOADARGS_1(name, arg1); \ - __sc_4 = (unsigned long) (arg2) - #undef LSS_LOADARGS_3 - #define LSS_LOADARGS_3(name, arg1, arg2, arg3) \ - LSS_LOADARGS_2(name, arg1, arg2); \ - __sc_5 = (unsigned long) (arg3) - #undef LSS_LOADARGS_4 - #define LSS_LOADARGS_4(name, arg1, arg2, arg3, arg4) \ - LSS_LOADARGS_3(name, arg1, arg2, arg3); \ - __sc_6 = (unsigned long) (arg4) - #undef LSS_LOADARGS_5 - #define LSS_LOADARGS_5(name, arg1, arg2, arg3, arg4, arg5) \ - LSS_LOADARGS_4(name, arg1, arg2, arg3, arg4); \ - __sc_7 = (unsigned long) (arg5) - #undef LSS_LOADARGS_6 - #define LSS_LOADARGS_6(name, arg1, arg2, arg3, arg4, arg5, arg6) \ - LSS_LOADARGS_5(name, arg1, arg2, arg3, arg4, arg5); \ - __sc_8 = (unsigned long) (arg6) - #undef LSS_ASMINPUT_0 - #define LSS_ASMINPUT_0 "0" (__sc_0) - #undef LSS_ASMINPUT_1 - #define LSS_ASMINPUT_1 LSS_ASMINPUT_0, "1" (__sc_3) - #undef LSS_ASMINPUT_2 - #define LSS_ASMINPUT_2 LSS_ASMINPUT_1, "2" (__sc_4) - #undef LSS_ASMINPUT_3 - #define LSS_ASMINPUT_3 LSS_ASMINPUT_2, "3" (__sc_5) - #undef LSS_ASMINPUT_4 - #define LSS_ASMINPUT_4 LSS_ASMINPUT_3, "4" (__sc_6) - #undef LSS_ASMINPUT_5 - #define LSS_ASMINPUT_5 LSS_ASMINPUT_4, "5" (__sc_7) - #undef LSS_ASMINPUT_6 - #define LSS_ASMINPUT_6 LSS_ASMINPUT_5, "6" (__sc_8) - #undef LSS_BODY - #define LSS_BODY(nr, type, name, args...) \ - long __sc_ret, __sc_err; \ - { \ - register unsigned long __sc_0 __asm__ ("r0"); \ - register unsigned long __sc_3 __asm__ ("r3"); \ - register unsigned long __sc_4 __asm__ ("r4"); \ - register unsigned long __sc_5 __asm__ ("r5"); \ - register unsigned long __sc_6 __asm__ ("r6"); \ - register unsigned long __sc_7 __asm__ ("r7"); \ - register unsigned long __sc_8 __asm__ ("r8"); \ - \ - LSS_LOADARGS_##nr(name, args); \ - __asm__ __volatile__ \ - ("sc\n\t" \ - "mfcr %0" \ - : "=&r" (__sc_0), \ - "=&r" (__sc_3), "=&r" (__sc_4), \ - "=&r" (__sc_5), "=&r" (__sc_6), \ - "=&r" (__sc_7), "=&r" (__sc_8) \ - : LSS_ASMINPUT_##nr \ - : "cr0", "ctr", "memory", \ - "r9", "r10", "r11", "r12"); \ - __sc_ret = __sc_3; \ - __sc_err = __sc_0; \ - } \ - LSS_RETURN(type, __sc_ret, __sc_err) - #undef _syscall0 - #define _syscall0(type, name) \ - type LSS_NAME(name)(void) { \ - LSS_BODY(0, type, name); \ - } - #undef _syscall1 - #define _syscall1(type, name, type1, arg1) \ - type LSS_NAME(name)(type1 arg1) { \ - LSS_BODY(1, type, name, arg1); \ - } - #undef _syscall2 - #define _syscall2(type, name, type1, arg1, type2, arg2) \ - type LSS_NAME(name)(type1 arg1, type2 arg2) { \ - LSS_BODY(2, type, name, arg1, arg2); \ - } - #undef _syscall3 - #define _syscall3(type, name, type1, arg1, type2, arg2, type3, arg3) \ - type LSS_NAME(name)(type1 arg1, type2 arg2, type3 arg3) { \ - LSS_BODY(3, type, name, arg1, arg2, arg3); \ - } - #undef _syscall4 - #define _syscall4(type, name, type1, arg1, type2, arg2, type3, arg3, \ - type4, arg4) \ - type LSS_NAME(name)(type1 arg1, type2 arg2, type3 arg3, type4 arg4) { \ - LSS_BODY(4, type, name, arg1, arg2, arg3, arg4); \ - } - #undef _syscall5 - #define _syscall5(type, name, type1, arg1, type2, arg2, type3, arg3, \ - type4, arg4, type5, arg5) \ - type LSS_NAME(name)(type1 arg1, type2 arg2, type3 arg3, type4 arg4, \ - type5 arg5) { \ - LSS_BODY(5, type, name, arg1, arg2, arg3, arg4, arg5); \ - } - #undef _syscall6 - #define _syscall6(type, name, type1, arg1, type2, arg2, type3, arg3, \ - type4, arg4, type5, arg5, type6, arg6) \ - type LSS_NAME(name)(type1 arg1, type2 arg2, type3 arg3, type4 arg4, \ - type5 arg5, type6 arg6) { \ - LSS_BODY(6, type, name, arg1, arg2, arg3, arg4, arg5, arg6); \ - } - /* clone function adapted from glibc 2.18 clone.S */ - LSS_INLINE int LSS_NAME(clone)(int (*fn)(void *), void *child_stack, - int flags, void *arg, int *parent_tidptr, - void *newtls, int *child_tidptr) { - long __ret, __err; - { -#if defined(__PPC64__) - -/* Stack frame offsets. */ -#if _CALL_ELF != 2 -#define FRAME_MIN_SIZE 112 -#define FRAME_TOC_SAVE 40 -#else -#define FRAME_MIN_SIZE 32 -#define FRAME_TOC_SAVE 24 -#endif - - - register int (*__fn)(void *) __asm__ ("r3") = fn; - register void *__cstack __asm__ ("r4") = child_stack; - register int __flags __asm__ ("r5") = flags; - register void * __arg __asm__ ("r6") = arg; - register int * __ptidptr __asm__ ("r7") = parent_tidptr; - register void * __newtls __asm__ ("r8") = newtls; - register int * __ctidptr __asm__ ("r9") = child_tidptr; - __asm__ __volatile__( - /* check for fn == NULL - * and child_stack == NULL - */ - "cmpdi cr0, %6, 0\n\t" - "cmpdi cr1, %7, 0\n\t" - "cror cr0*4+eq, cr1*4+eq, cr0*4+eq\n\t" - "beq- cr0, 1f\n\t" - - /* set up stack frame for child */ - "clrrdi %7, %7, 4\n\t" - "li 0, 0\n\t" - "stdu 0, -%13(%7)\n\t" - - /* fn, arg, child_stack are saved acrVoss the syscall */ - "mr 28, %6\n\t" - "mr 29, %7\n\t" - "mr 27, %9\n\t" - - /* syscall - r3 == flags - r4 == child_stack - r5 == parent_tidptr - r6 == newtls - r7 == child_tidptr */ - "mr 3, %8\n\t" - "mr 5, %10\n\t" - "mr 6, %11\n\t" - "mr 7, %12\n\t" - "li 0, %4\n\t" - "sc\n\t" - - /* Test if syscall was successful */ - "cmpdi cr1, 3, 0\n\t" - "crandc cr1*4+eq, cr1*4+eq, cr0*4+so\n\t" - "bne- cr1, 1f\n\t" - - /* Do the function call */ - "std 2, %14(1)\n\t" -#if _CALL_ELF != 2 - "ld 0, 0(28)\n\t" - "ld 2, 8(28)\n\t" - "mtctr 0\n\t" -#else - "mr 12, 28\n\t" - "mtctr 12\n\t" -#endif - "mr 3, 27\n\t" - "bctrl\n\t" - "ld 2, %14(1)\n\t" - - /* Call _exit(r3) */ - "li 0, %5\n\t" - "sc\n\t" - - /* Return to parent */ - "1:\n\t" - "mr %0, 3\n\t" - : "=r" (__ret), "=r" (__err) - : "0" (-1), "i" (EINVAL), - "i" (__NR_clone), "i" (__NR_exit), - "r" (__fn), "r" (__cstack), "r" (__flags), - "r" (__arg), "r" (__ptidptr), "r" (__newtls), - "r" (__ctidptr), "i" (FRAME_MIN_SIZE), "i" (FRAME_TOC_SAVE) - : "cr0", "cr1", "memory", "ctr", - "r0", "r29", "r27", "r28"); -#else - register int (*__fn)(void *) __asm__ ("r8") = fn; - register void *__cstack __asm__ ("r4") = child_stack; - register int __flags __asm__ ("r3") = flags; - register void * __arg __asm__ ("r9") = arg; - register int * __ptidptr __asm__ ("r5") = parent_tidptr; - register void * __newtls __asm__ ("r6") = newtls; - register int * __ctidptr __asm__ ("r7") = child_tidptr; - __asm__ __volatile__( - /* check for fn == NULL - * and child_stack == NULL - */ - "cmpwi cr0, %6, 0\n\t" - "cmpwi cr1, %7, 0\n\t" - "cror cr0*4+eq, cr1*4+eq, cr0*4+eq\n\t" - "beq- cr0, 1f\n\t" - - /* set up stack frame for child */ - "clrrwi %7, %7, 4\n\t" - "li 0, 0\n\t" - "stwu 0, -16(%7)\n\t" - - /* fn, arg, child_stack are saved across the syscall: r28-30 */ - "mr 28, %6\n\t" - "mr 29, %7\n\t" - "mr 27, %9\n\t" - - /* syscall */ - "li 0, %4\n\t" - /* flags already in r3 - * child_stack already in r4 - * ptidptr already in r5 - * newtls already in r6 - * ctidptr already in r7 - */ - "sc\n\t" - - /* Test if syscall was successful */ - "cmpwi cr1, 3, 0\n\t" - "crandc cr1*4+eq, cr1*4+eq, cr0*4+so\n\t" - "bne- cr1, 1f\n\t" - - /* Do the function call */ - "mtctr 28\n\t" - "mr 3, 27\n\t" - "bctrl\n\t" - - /* Call _exit(r3) */ - "li 0, %5\n\t" - "sc\n\t" - - /* Return to parent */ - "1:\n" - "mfcr %1\n\t" - "mr %0, 3\n\t" - : "=r" (__ret), "=r" (__err) - : "0" (-1), "1" (EINVAL), - "i" (__NR_clone), "i" (__NR_exit), - "r" (__fn), "r" (__cstack), "r" (__flags), - "r" (__arg), "r" (__ptidptr), "r" (__newtls), - "r" (__ctidptr) - : "cr0", "cr1", "memory", "ctr", - "r0", "r29", "r27", "r28"); - -#endif - } - LSS_RETURN(int, __ret, __err); - } - #elif defined(__aarch64__) - #undef LSS_REG - #define LSS_REG(r,a) register long __x##r __asm__("x"#r) = (long)a - #undef LSS_BODY - #define LSS_BODY(type,name,args...) \ - register long __res_x0 __asm__("x0"); \ - long __res; \ - __asm__ __volatile__ ("mov x8, %1\n" \ - "svc 0x0\n" \ - : "=r"(__res_x0) \ - : "i"(__NR_##name) , ## args \ - : "memory"); \ - __res = __res_x0; \ - LSS_RETURN(type, __res) - #undef _syscall0 - #define _syscall0(type, name) \ - type LSS_NAME(name)(void) { \ - LSS_BODY(type, name); \ - } - #undef _syscall1 - #define _syscall1(type, name, type1, arg1) \ - type LSS_NAME(name)(type1 arg1) { \ - LSS_REG(0, arg1); LSS_BODY(type, name, "r"(__x0)); \ - } - #undef _syscall2 - #define _syscall2(type, name, type1, arg1, type2, arg2) \ - type LSS_NAME(name)(type1 arg1, type2 arg2) { \ - LSS_REG(0, arg1); LSS_REG(1, arg2); \ - LSS_BODY(type, name, "r"(__x0), "r"(__x1)); \ - } - #undef _syscall3 - #define _syscall3(type, name, type1, arg1, type2, arg2, type3, arg3) \ - type LSS_NAME(name)(type1 arg1, type2 arg2, type3 arg3) { \ - LSS_REG(0, arg1); LSS_REG(1, arg2); LSS_REG(2, arg3); \ - LSS_BODY(type, name, "r"(__x0), "r"(__x1), "r"(__x2)); \ - } - #undef _syscall4 - #define _syscall4(type,name,type1,arg1,type2,arg2,type3,arg3,type4,arg4) \ - type LSS_NAME(name)(type1 arg1, type2 arg2, type3 arg3, type4 arg4) { \ - LSS_REG(0, arg1); LSS_REG(1, arg2); LSS_REG(2, arg3); \ - LSS_REG(3, arg4); \ - LSS_BODY(type, name, "r"(__x0), "r"(__x1), "r"(__x2), "r"(__x3)); \ - } - #undef _syscall5 - #define _syscall5(type,name,type1,arg1,type2,arg2,type3,arg3,type4,arg4, \ - type5,arg5) \ - type LSS_NAME(name)(type1 arg1, type2 arg2, type3 arg3, type4 arg4, \ - type5 arg5) { \ - LSS_REG(0, arg1); LSS_REG(1, arg2); LSS_REG(2, arg3); \ - LSS_REG(3, arg4); LSS_REG(4, arg5); \ - LSS_BODY(type, name, "r"(__x0), "r"(__x1), "r"(__x2), "r"(__x3), \ - "r"(__x4)); \ - } - #undef _syscall6 - #define _syscall6(type,name,type1,arg1,type2,arg2,type3,arg3,type4,arg4, \ - type5,arg5,type6,arg6) \ - type LSS_NAME(name)(type1 arg1, type2 arg2, type3 arg3, type4 arg4, \ - type5 arg5, type6 arg6) { \ - LSS_REG(0, arg1); LSS_REG(1, arg2); LSS_REG(2, arg3); \ - LSS_REG(3, arg4); LSS_REG(4, arg5); LSS_REG(5, arg6); \ - LSS_BODY(type, name, "r"(__x0), "r"(__x1), "x"(__x2), "r"(__x3), \ - "r"(__x4), "r"(__x5)); \ - } - /* clone function adapted from glibc 2.18 clone.S */ - LSS_INLINE int LSS_NAME(clone)(int (*fn)(void *), void *child_stack, - int flags, void *arg, int *parent_tidptr, - void *newtls, int *child_tidptr) { - long __res; - { - register int (*__fn)(void *) __asm__("x0") = fn; - register void *__stack __asm__("x1") = child_stack; - register int __flags __asm__("x2") = flags; - register void *__arg __asm__("x3") = arg; - register int *__ptid __asm__("x4") = parent_tidptr; - register void *__tls __asm__("x5") = newtls; - register int *__ctid __asm__("x6") = child_tidptr; - __asm__ __volatile__(/* if (fn == NULL || child_stack == NULL) - * return -EINVAL; - */ - "cbz x0,1f\n" - "cbz x1,1f\n" - - /* Push "arg" and "fn" onto the stack that will be - * used by the child. - */ - "stp x0,x3, [x1, #-16]!\n" - - "mov x0,x2\n" /* flags */ - "mov x2,x4\n" /* ptid */ - "mov x3,x5\n" /* tls */ - "mov x4,x6\n" /* ctid */ - "mov x8,%9\n" /* clone */ - - "svc 0x0\n" - - /* if (%r0 != 0) - * return %r0; - */ - "cmp x0, #0\n" - "bne 2f\n" - - /* In the child, now. Call "fn(arg)". - */ - "ldp x1, x0, [sp], #16\n" - "blr x1\n" - - /* Call _exit(%r0). - */ - "mov x8, %10\n" - "svc 0x0\n" - "1:\n" - "mov x8, %1\n" - "2:\n" - : "=r" (__res) - : "i"(-EINVAL), - "r"(__fn), "r"(__stack), "r"(__flags), "r"(__arg), - "r"(__ptid), "r"(__tls), "r"(__ctid), - "i"(__NR_clone), "i"(__NR_exit) - : "x30", "memory"); - } - LSS_RETURN(int, __res); - } - #elif defined(__s390__) - #undef LSS_REG - #define LSS_REG(r, a) register unsigned long __r##r __asm__("r"#r) = (unsigned long) a - #undef LSS_BODY - #define LSS_BODY(type, name, args...) \ - register unsigned long __nr __asm__("r1") \ - = (unsigned long)(__NR_##name); \ - register long __res_r2 __asm__("r2"); \ - long __res; \ - __asm__ __volatile__ \ - ("svc 0\n\t" \ - : "=d"(__res_r2) \ - : "d"(__nr), ## args \ - : "memory"); \ - __res = __res_r2; \ - LSS_RETURN(type, __res) - #undef _syscall0 - #define _syscall0(type, name) \ - type LSS_NAME(name)(void) { \ - LSS_BODY(type, name); \ - } - #undef _syscall1 - #define _syscall1(type, name, type1, arg1) \ - type LSS_NAME(name)(type1 arg1) { \ - LSS_REG(2, arg1); \ - LSS_BODY(type, name, "0"(__r2)); \ - } - #undef _syscall2 - #define _syscall2(type, name, type1, arg1, type2, arg2) \ - type LSS_NAME(name)(type1 arg1, type2 arg2) { \ - LSS_REG(2, arg1); LSS_REG(3, arg2); \ - LSS_BODY(type, name, "0"(__r2), "d"(__r3)); \ - } - #undef _syscall3 - #define _syscall3(type, name, type1, arg1, type2, arg2, type3, arg3) \ - type LSS_NAME(name)(type1 arg1, type2 arg2, type3 arg3) { \ - LSS_REG(2, arg1); LSS_REG(3, arg2); LSS_REG(4, arg3); \ - LSS_BODY(type, name, "0"(__r2), "d"(__r3), "d"(__r4)); \ - } - #undef _syscall4 - #define _syscall4(type, name, type1, arg1, type2, arg2, type3, arg3, \ - type4, arg4) \ - type LSS_NAME(name)(type1 arg1, type2 arg2, type3 arg3, \ - type4 arg4) { \ - LSS_REG(2, arg1); LSS_REG(3, arg2); LSS_REG(4, arg3); \ - LSS_REG(5, arg4); \ - LSS_BODY(type, name, "0"(__r2), "d"(__r3), "d"(__r4), \ - "d"(__r5)); \ - } - #undef _syscall5 - #define _syscall5(type, name, type1, arg1, type2, arg2, type3, arg3, \ - type4, arg4, type5, arg5) \ - type LSS_NAME(name)(type1 arg1, type2 arg2, type3 arg3, \ - type4 arg4, type5 arg5) { \ - LSS_REG(2, arg1); LSS_REG(3, arg2); LSS_REG(4, arg3); \ - LSS_REG(5, arg4); LSS_REG(6, arg5); \ - LSS_BODY(type, name, "0"(__r2), "d"(__r3), "d"(__r4), \ - "d"(__r5), "d"(__r6)); \ - } - #undef _syscall6 - #define _syscall6(type, name, type1, arg1, type2, arg2, type3, arg3, \ - type4, arg4, type5, arg5, type6, arg6) \ - type LSS_NAME(name)(type1 arg1, type2 arg2, type3 arg3, \ - type4 arg4, type5 arg5, type6 arg6) { \ - LSS_REG(2, arg1); LSS_REG(3, arg2); LSS_REG(4, arg3); \ - LSS_REG(5, arg4); LSS_REG(6, arg5); LSS_REG(7, arg6); \ - LSS_BODY(type, name, "0"(__r2), "d"(__r3), "d"(__r4), \ - "d"(__r5), "d"(__r6), "d"(__r7)); \ - } - LSS_INLINE int LSS_NAME(clone)(int (*fn)(void *), void *child_stack, - int flags, void *arg, int *parent_tidptr, - void *newtls, int *child_tidptr) { - long __ret; - { - register int (*__fn)(void *) __asm__ ("r1") = fn; - register void *__cstack __asm__ ("r2") = child_stack; - register int __flags __asm__ ("r3") = flags; - register void *__arg __asm__ ("r0") = arg; - register int *__ptidptr __asm__ ("r4") = parent_tidptr; - register void *__newtls __asm__ ("r6") = newtls; - register int *__ctidptr __asm__ ("r5") = child_tidptr; - __asm__ __volatile__ ( - #ifndef __s390x__ - /* arg already in r0 */ - "ltr %4, %4\n\t" /* check fn, which is already in r1 */ - "jz 1f\n\t" /* NULL function pointer, return -EINVAL */ - "ltr %5, %5\n\t" /* check child_stack, which is already in r2 */ - "jz 1f\n\t" /* NULL stack pointer, return -EINVAL */ - /* flags already in r3 */ - /* parent_tidptr already in r4 */ - /* child_tidptr already in r5 */ - /* newtls already in r6 */ - "svc %2\n\t" /* invoke clone syscall */ - "ltr %0,%%r2\n\t" /* load return code into __ret and test */ - "jnz 1f\n\t" /* return to parent if non-zero */ - /* start child thread */ - "lr %%r2, %7\n\t" /* set first parameter to void *arg */ - "ahi %%r15, -96\n\t" /* make room on the stack for the save area */ - "xc 0(4,%%r15), 0(%%r15)\n\t" - "basr %%r14, %4\n\t" /* jump to fn */ - "svc %3\n" /* invoke exit syscall */ - "1:\n" - #else - /* arg already in r0 */ - "ltgr %4, %4\n\t" /* check fn, which is already in r1 */ - "jz 1f\n\t" /* NULL function pointer, return -EINVAL */ - "ltgr %5, %5\n\t" /* check child_stack, which is already in r2 */ - "jz 1f\n\t" /* NULL stack pointer, return -EINVAL */ - /* flags already in r3 */ - /* parent_tidptr already in r4 */ - /* child_tidptr already in r5 */ - /* newtls already in r6 */ - "svc %2\n\t" /* invoke clone syscall */ - "ltgr %0, %%r2\n\t" /* load return code into __ret and test */ - "jnz 1f\n\t" /* return to parent if non-zero */ - /* start child thread */ - "lgr %%r2, %7\n\t" /* set first parameter to void *arg */ - "aghi %%r15, -160\n\t" /* make room on the stack for the save area */ - "xc 0(8,%%r15), 0(%%r15)\n\t" - "basr %%r14, %4\n\t" /* jump to fn */ - "svc %3\n" /* invoke exit syscall */ - "1:\n" - #endif - : "=r" (__ret) - : "0" (-EINVAL), "i" (__NR_clone), "i" (__NR_exit), - "d" (__fn), "d" (__cstack), "d" (__flags), "d" (__arg), - "d" (__ptidptr), "d" (__newtls), "d" (__ctidptr) - : "cc", "r14", "memory" - ); - } - LSS_RETURN(int, __ret); - } - #endif - #define __NR__exit __NR_exit - #define __NR__gettid __NR_gettid - #define __NR__mremap __NR_mremap - LSS_INLINE _syscall1(int, close, int, f) - LSS_INLINE _syscall1(int, _exit, int, e) - LSS_INLINE _syscall3(int, fcntl, int, f, - int, c, long, a) - LSS_INLINE _syscall2(int, fstat, int, f, - struct kernel_stat*, b) - LSS_INLINE _syscall6(int, futex, int*, a, - int, o, int, v, - struct kernel_timespec*, t, - int*, a2, - int, v3) -#ifdef __NR_getdents64 - LSS_INLINE _syscall3(int, getdents64, int, f, - struct kernel_dirent64*, d, int, c) -#define KERNEL_DIRENT kernel_dirent64 -#define GETDENTS sys_getdents64 -#else - LSS_INLINE _syscall3(int, getdents, int, f, - struct kernel_dirent*, d, int, c) -#define KERNEL_DIRENT kernel_dirent -#define GETDENTS sys_getdents -#endif - LSS_INLINE _syscall0(pid_t, getpid) - LSS_INLINE _syscall0(pid_t, getppid) - LSS_INLINE _syscall0(pid_t, _gettid) - LSS_INLINE _syscall2(int, kill, pid_t, p, - int, s) - #if defined(__x86_64__) - /* Need to make sure off_t isn't truncated to 32-bits under x32. */ - LSS_INLINE off_t LSS_NAME(lseek)(int f, off_t o, int w) { - _LSS_BODY(3, off_t, lseek, off_t, LSS_SYSCALL_ARG(f), (uint64_t)(o), - LSS_SYSCALL_ARG(w)); - } - #else - LSS_INLINE _syscall3(off_t, lseek, int, f, - off_t, o, int, w) - #endif - LSS_INLINE _syscall2(int, munmap, void*, s, - size_t, l) - LSS_INLINE _syscall5(void*, _mremap, void*, o, - size_t, os, size_t, ns, - unsigned long, f, void *, a) - LSS_INLINE _syscall2(int, prctl, int, o, - long, a) - LSS_INLINE _syscall4(long, ptrace, int, r, - pid_t, p, void *, a, void *, d) - LSS_INLINE _syscall3(ssize_t, read, int, f, - void *, b, size_t, c) - LSS_INLINE _syscall4(int, rt_sigaction, int, s, - const struct kernel_sigaction*, a, - struct kernel_sigaction*, o, size_t, c) - LSS_INLINE _syscall4(int, rt_sigprocmask, int, h, - const struct kernel_sigset_t*, s, - struct kernel_sigset_t*, o, size_t, c); - LSS_INLINE _syscall0(int, sched_yield) - LSS_INLINE _syscall2(int, sigaltstack, const stack_t*, s, - const stack_t*, o) - #if defined(__NR_fstatat) - LSS_INLINE _syscall4(int, fstatat, int, d, const char *, p, - struct kernel_stat*, b, int, flags) - LSS_INLINE int LSS_NAME(stat)(const char* p, struct kernel_stat* b) { - return LSS_NAME(fstatat)(AT_FDCWD,p,b,0); - } - #else - LSS_INLINE _syscall2(int, stat, const char*, f, - struct kernel_stat*, b) - #endif - LSS_INLINE _syscall3(ssize_t, write, int, f, - const void *, b, size_t, c) - #if defined(__NR_getcpu) - LSS_INLINE _syscall3(long, getcpu, unsigned *, cpu, - unsigned *, node, void *, unused); - #endif - #if defined(__x86_64__) || defined(__aarch64__) || \ - (defined(__mips__) && _MIPS_SIM != _MIPS_SIM_ABI32) - LSS_INLINE _syscall3(int, socket, int, d, - int, t, int, p) - #endif - #if defined(__x86_64__) || defined(__s390x__) - LSS_INLINE int LSS_NAME(sigaction)(int signum, - const struct kernel_sigaction *act, - struct kernel_sigaction *oldact) { - #if defined(__x86_64__) - /* On x86_64, the kernel requires us to always set our own - * SA_RESTORER in order to be able to return from a signal handler. - * This function must have a "magic" signature that the "gdb" - * (and maybe the kernel?) can recognize. - */ - if (act != NULL && !(act->sa_flags & SA_RESTORER)) { - struct kernel_sigaction a = *act; - a.sa_flags |= SA_RESTORER; - a.sa_restorer = LSS_NAME(restore_rt)(); - return LSS_NAME(rt_sigaction)(signum, &a, oldact, - (KERNEL_NSIG+7)/8); - } else - #endif - return LSS_NAME(rt_sigaction)(signum, act, oldact, - (KERNEL_NSIG+7)/8); - } - - LSS_INLINE int LSS_NAME(sigprocmask)(int how, - const struct kernel_sigset_t *set, - struct kernel_sigset_t *oldset) { - return LSS_NAME(rt_sigprocmask)(how, set, oldset, (KERNEL_NSIG+7)/8); - } - #endif - #if (defined(__aarch64__)) || \ - (defined(__mips__) && (_MIPS_ISA == _MIPS_ISA_MIPS64)) - LSS_INLINE int LSS_NAME(sigaction)(int signum, - const struct kernel_sigaction *act, - struct kernel_sigaction *oldact) { - return LSS_NAME(rt_sigaction)(signum, act, oldact, (KERNEL_NSIG+7)/8); - - } - LSS_INLINE int LSS_NAME(sigprocmask)(int how, - const struct kernel_sigset_t *set, - struct kernel_sigset_t *oldset) { - return LSS_NAME(rt_sigprocmask)(how, set, oldset, (KERNEL_NSIG+7)/8); - } - #endif - #ifdef __NR_wait4 - LSS_INLINE _syscall4(pid_t, wait4, pid_t, p, - int*, s, int, o, - struct kernel_rusage*, r) - LSS_INLINE pid_t LSS_NAME(waitpid)(pid_t pid, int *status, int options){ - return LSS_NAME(wait4)(pid, status, options, 0); - } - #else - LSS_INLINE _syscall3(pid_t, waitpid, pid_t, p, - int*, s, int, o) - #endif - #ifdef __NR_openat - LSS_INLINE _syscall4(int, openat, int, d, const char *, p, int, f, int, m) - LSS_INLINE int LSS_NAME(open)(const char* p, int f, int m) { - return LSS_NAME(openat)(AT_FDCWD,p,f,m ); - } - #else - LSS_INLINE _syscall3(int, open, const char*, p, - int, f, int, m) - #endif - LSS_INLINE int LSS_NAME(sigemptyset)(struct kernel_sigset_t *set) { - memset(&set->sig, 0, sizeof(set->sig)); - return 0; - } - - LSS_INLINE int LSS_NAME(sigfillset)(struct kernel_sigset_t *set) { - memset(&set->sig, -1, sizeof(set->sig)); - return 0; - } - - LSS_INLINE int LSS_NAME(sigaddset)(struct kernel_sigset_t *set, - int signum) { - if (signum < 1 || signum > (int)(8*sizeof(set->sig))) { - LSS_ERRNO = EINVAL; - return -1; - } else { - set->sig[(signum - 1)/(8*sizeof(set->sig[0]))] - |= 1UL << ((signum - 1) % (8*sizeof(set->sig[0]))); - return 0; - } - } - - LSS_INLINE int LSS_NAME(sigdelset)(struct kernel_sigset_t *set, - int signum) { - if (signum < 1 || signum > (int)(8*sizeof(set->sig))) { - LSS_ERRNO = EINVAL; - return -1; - } else { - set->sig[(signum - 1)/(8*sizeof(set->sig[0]))] - &= ~(1UL << ((signum - 1) % (8*sizeof(set->sig[0])))); - return 0; - } - } - - #if defined(__i386__) || \ - defined(__arm__) || \ - (defined(__mips__) && _MIPS_SIM == _MIPS_SIM_ABI32) || \ - defined(__PPC__) || \ - (defined(__s390__) && !defined(__s390x__)) - #define __NR__sigaction __NR_sigaction - #define __NR__sigprocmask __NR_sigprocmask - LSS_INLINE _syscall2(int, fstat64, int, f, - struct kernel_stat64 *, b) - LSS_INLINE _syscall5(int, _llseek, uint, fd, ulong, hi, ulong, lo, - loff_t *, res, uint, wh) -#if defined(__s390__) && !defined(__s390x__) - /* On s390, mmap2() arguments are passed in memory. */ - LSS_INLINE void* LSS_NAME(_mmap2)(void *s, size_t l, int p, int f, int d, - off_t o) { - unsigned long buf[6] = { (unsigned long) s, (unsigned long) l, - (unsigned long) p, (unsigned long) f, - (unsigned long) d, (unsigned long) o }; - LSS_REG(2, buf); - LSS_BODY(void*, mmap2, "0"(__r2)); - } -#elif !defined(__PPC64__) - #define __NR__mmap2 __NR_mmap2 - LSS_INLINE _syscall6(void*, _mmap2, void*, s, - size_t, l, int, p, - int, f, int, d, - off_t, o) -#endif - LSS_INLINE _syscall3(int, _sigaction, int, s, - const struct kernel_old_sigaction*, a, - struct kernel_old_sigaction*, o) - LSS_INLINE _syscall3(int, _sigprocmask, int, h, - const unsigned long*, s, - unsigned long*, o) - LSS_INLINE _syscall2(int, stat64, const char *, p, - struct kernel_stat64 *, b) - - LSS_INLINE int LSS_NAME(sigaction)(int signum, - const struct kernel_sigaction *act, - struct kernel_sigaction *oldact) { - int old_errno = LSS_ERRNO; - int rc; - struct kernel_sigaction a; - if (act != NULL) { - a = *act; - #ifdef __i386__ - /* On i386, the kernel requires us to always set our own - * SA_RESTORER when using realtime signals. Otherwise, it does not - * know how to return from a signal handler. This function must have - * a "magic" signature that the "gdb" (and maybe the kernel?) can - * recognize. - * Apparently, a SA_RESTORER is implicitly set by the kernel, when - * using non-realtime signals. - * - * TODO: Test whether ARM needs a restorer - */ - if (!(a.sa_flags & SA_RESTORER)) { - a.sa_flags |= SA_RESTORER; - a.sa_restorer = (a.sa_flags & SA_SIGINFO) - ? LSS_NAME(restore_rt)() : LSS_NAME(restore)(); - } - #endif - } - rc = LSS_NAME(rt_sigaction)(signum, act ? &a : act, oldact, - (KERNEL_NSIG+7)/8); - if (rc < 0 && LSS_ERRNO == ENOSYS) { - struct kernel_old_sigaction oa, ooa, *ptr_a = &oa, *ptr_oa = &ooa; - if (!act) { - ptr_a = NULL; - } else { - oa.sa_handler_ = act->sa_handler_; - memcpy(&oa.sa_mask, &act->sa_mask, sizeof(oa.sa_mask)); - #ifndef __mips__ - oa.sa_restorer = act->sa_restorer; - #endif - oa.sa_flags = act->sa_flags; - } - if (!oldact) { - ptr_oa = NULL; - } - LSS_ERRNO = old_errno; - rc = LSS_NAME(_sigaction)(signum, ptr_a, ptr_oa); - if (rc == 0 && oldact) { - if (act) { - memcpy(oldact, act, sizeof(*act)); - } else { - memset(oldact, 0, sizeof(*oldact)); - } - oldact->sa_handler_ = ptr_oa->sa_handler_; - oldact->sa_flags = ptr_oa->sa_flags; - memcpy(&oldact->sa_mask, &ptr_oa->sa_mask, sizeof(ptr_oa->sa_mask)); - #ifndef __mips__ - oldact->sa_restorer = ptr_oa->sa_restorer; - #endif - } - } - return rc; - } - - LSS_INLINE int LSS_NAME(sigprocmask)(int how, - const struct kernel_sigset_t *set, - struct kernel_sigset_t *oldset) { - int olderrno = LSS_ERRNO; - int rc = LSS_NAME(rt_sigprocmask)(how, set, oldset, (KERNEL_NSIG+7)/8); - if (rc < 0 && LSS_ERRNO == ENOSYS) { - LSS_ERRNO = olderrno; - if (oldset) { - LSS_NAME(sigemptyset)(oldset); - } - rc = LSS_NAME(_sigprocmask)(how, - set ? &set->sig[0] : NULL, - oldset ? &oldset->sig[0] : NULL); - } - return rc; - } - #endif - #if defined(__i386__) || \ - defined(__ARM_ARCH_3__) || defined(__ARM_EABI__) || \ - (defined(__mips__) && _MIPS_SIM == _MIPS_SIM_ABI32) || \ - (defined(__PPC__) && !defined(__PPC64__)) || \ - (defined(__s390__) && !defined(__s390x__)) - /* On these architectures, implement mmap() with mmap2(). */ - LSS_INLINE void* LSS_NAME(mmap)(void *s, size_t l, int p, int f, int d, - int64_t o) { - if (o % 4096) { - LSS_ERRNO = EINVAL; - return (void *) -1; - } - return LSS_NAME(_mmap2)(s, l, p, f, d, (o / 4096)); - } - #elif defined(__s390x__) - /* On s390x, mmap() arguments are passed in memory. */ - LSS_INLINE void* LSS_NAME(mmap)(void *s, size_t l, int p, int f, int d, - int64_t o) { - unsigned long buf[6] = { (unsigned long) s, (unsigned long) l, - (unsigned long) p, (unsigned long) f, - (unsigned long) d, (unsigned long) o }; - LSS_REG(2, buf); - LSS_BODY(void*, mmap, "0"(__r2)); - } - #elif defined(__x86_64__) - /* Need to make sure __off64_t isn't truncated to 32-bits under x32. */ - LSS_INLINE void* LSS_NAME(mmap)(void *s, size_t l, int p, int f, int d, - int64_t o) { - LSS_BODY(6, void*, mmap, LSS_SYSCALL_ARG(s), LSS_SYSCALL_ARG(l), - LSS_SYSCALL_ARG(p), LSS_SYSCALL_ARG(f), - LSS_SYSCALL_ARG(d), (uint64_t)(o)); - } - #else - /* Remaining 64-bit architectures. */ - LSS_INLINE _syscall6(void*, mmap, void*, addr, size_t, length, int, prot, - int, flags, int, fd, int64_t, offset) - #endif - #if defined(__i386__) || \ - defined(__PPC__) || \ - (defined(__arm__) && !defined(__ARM_EABI__)) || \ - (defined(__mips__) && _MIPS_SIM == _MIPS_SIM_ABI32) || \ - defined(__s390__) - - /* See sys_socketcall in net/socket.c in kernel source. - * It de-multiplexes on its first arg and unpacks the arglist - * array in its second arg. - */ - LSS_INLINE _syscall2(int, socketcall, int, c, unsigned long*, a) - - LSS_INLINE int LSS_NAME(socket)(int domain, int type, int protocol) { - unsigned long args[3] = { - (unsigned long) domain, - (unsigned long) type, - (unsigned long) protocol - }; - return LSS_NAME(socketcall)(1, args); - } - #elif defined(__ARM_EABI__) - LSS_INLINE _syscall3(int, socket, int, d, - int, t, int, p) - #endif - #if defined(__mips__) - /* sys_pipe() on MIPS has non-standard calling conventions, as it returns - * both file handles through CPU registers. - */ - LSS_INLINE int LSS_NAME(pipe)(int *p) { - register unsigned long __v0 __asm__("$2") = __NR_pipe; - register unsigned long __v1 __asm__("$3"); - register unsigned long __r7 __asm__("$7"); - __asm__ __volatile__ ("syscall\n" - : "=&r"(__v0), "=&r"(__v1), "+r" (__r7) - : "0"(__v0) - : "$8", "$9", "$10", "$11", "$12", - "$13", "$14", "$15", "$24", "memory"); - if (__r7) { - LSS_ERRNO = __v0; - return -1; - } else { - p[0] = __v0; - p[1] = __v1; - return 0; - } - } - #elif defined(__NR_pipe2) - LSS_INLINE _syscall2(int, pipe2, int *, p, - int, f ) - LSS_INLINE int LSS_NAME(pipe)( int * p) { - return LSS_NAME(pipe2)(p, 0); - } - #else - LSS_INLINE _syscall1(int, pipe, int *, p) - #endif - - LSS_INLINE pid_t LSS_NAME(gettid)() { - pid_t tid = LSS_NAME(_gettid)(); - if (tid != -1) { - return tid; - } - return LSS_NAME(getpid)(); - } - - LSS_INLINE void *LSS_NAME(mremap)(void *old_address, size_t old_size, - size_t new_size, int flags, ...) { - va_list ap; - void *new_address, *rc; - va_start(ap, flags); - new_address = va_arg(ap, void *); - rc = LSS_NAME(_mremap)(old_address, old_size, new_size, - flags, new_address); - va_end(ap); - return rc; - } - - LSS_INLINE int LSS_NAME(ptrace_detach)(pid_t pid) { - /* PTRACE_DETACH can sometimes forget to wake up the tracee and it - * then sends job control signals to the real parent, rather than to - * the tracer. We reduce the risk of this happening by starting a - * whole new time slice, and then quickly sending a SIGCONT signal - * right after detaching from the tracee. - */ - int rc, err; - LSS_NAME(sched_yield)(); - rc = LSS_NAME(ptrace)(PTRACE_DETACH, pid, (void *)0, (void *)0); - err = LSS_ERRNO; - LSS_NAME(kill)(pid, SIGCONT); - LSS_ERRNO = err; - return rc; - } -#endif - -#if defined(__cplusplus) && !defined(SYS_CPLUSPLUS) -} -#endif - -#endif -#endif diff --git a/contrib/libtcmalloc/src/base/linuxthreads.cc b/contrib/libtcmalloc/src/base/linuxthreads.cc deleted file mode 100644 index 891e70c88c4..00000000000 --- a/contrib/libtcmalloc/src/base/linuxthreads.cc +++ /dev/null @@ -1,707 +0,0 @@ -// -*- Mode: C++; c-basic-offset: 2; indent-tabs-mode: nil -*- -/* Copyright (c) 2005-2007, Google Inc. - * All rights reserved. - * - * Redistribution and use in source and binary forms, with or without - * modification, are permitted provided that the following conditions are - * met: - * - * * Redistributions of source code must retain the above copyright - * notice, this list of conditions and the following disclaimer. - * * Redistributions in binary form must reproduce the above - * copyright notice, this list of conditions and the following disclaimer - * in the documentation and/or other materials provided with the - * distribution. - * * Neither the name of Google Inc. nor the names of its - * contributors may be used to endorse or promote products derived from - * this software without specific prior written permission. - * - * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS - * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT - * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR - * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT - * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, - * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT - * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, - * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY - * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT - * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE - * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. - * - * --- - * Author: Markus Gutschke - */ - -#include "base/linuxthreads.h" - -#ifdef THREADS -#ifdef __cplusplus -extern "C" { -#endif - -#include -#include -#include -#include -#include -#include -#include -#include -#include - -#include "base/linux_syscall_support.h" -#include "base/thread_lister.h" - -#ifndef CLONE_UNTRACED -#define CLONE_UNTRACED 0x00800000 -#endif - - -/* Synchronous signals that should not be blocked while in the lister thread. - */ -static const int sync_signals[] = { SIGABRT, SIGILL, SIGFPE, SIGSEGV, SIGBUS, - SIGXCPU, SIGXFSZ }; - -/* itoa() is not a standard function, and we cannot safely call printf() - * after suspending threads. So, we just implement our own copy. A - * recursive approach is the easiest here. - */ -static char *local_itoa(char *buf, int i) { - if (i < 0) { - *buf++ = '-'; - return local_itoa(buf, -i); - } else { - if (i >= 10) - buf = local_itoa(buf, i/10); - *buf++ = (i%10) + '0'; - *buf = '\000'; - return buf; - } -} - - -/* Wrapper around clone() that runs "fn" on the same stack as the - * caller! Unlike fork(), the cloned thread shares the same address space. - * The caller must be careful to use only minimal amounts of stack until - * the cloned thread has returned. - * There is a good chance that the cloned thread and the caller will share - * the same copy of errno! - */ -#ifdef __GNUC__ -#if __GNUC__ == 3 && __GNUC_MINOR__ >= 1 || __GNUC__ > 3 -/* Try to force this function into a separate stack frame, and make sure - * that arguments are passed on the stack. - */ -static int local_clone (int (*fn)(void *), void *arg, ...) - __attribute__ ((noinline)); -#endif -#endif - -/* To avoid the gap cross page boundaries, increase by the large parge - * size mostly PowerPC system uses. */ -#ifdef __PPC64__ -#define CLONE_STACK_SIZE 65536 -#else -#define CLONE_STACK_SIZE 4096 -#endif - -static int local_clone (int (*fn)(void *), void *arg, ...) { - /* Leave 4kB of gap between the callers stack and the new clone. This - * should be more than sufficient for the caller to call waitpid() until - * the cloned thread terminates. - * - * It is important that we set the CLONE_UNTRACED flag, because newer - * versions of "gdb" otherwise attempt to attach to our thread, and will - * attempt to reap its status codes. This subsequently results in the - * caller hanging indefinitely in waitpid(), waiting for a change in - * status that will never happen. By setting the CLONE_UNTRACED flag, we - * prevent "gdb" from stealing events, but we still expect the thread - * lister to fail, because it cannot PTRACE_ATTACH to the process that - * is being debugged. This is OK and the error code will be reported - * correctly. - */ - return sys_clone(fn, (char *)&arg - CLONE_STACK_SIZE, - CLONE_VM|CLONE_FS|CLONE_FILES|CLONE_UNTRACED, arg, 0, 0, 0); -} - - -/* Local substitute for the atoi() function, which is not necessarily safe - * to call once threads are suspended (depending on whether libc looks up - * locale information, when executing atoi()). - */ -static int local_atoi(const char *s) { - int n = 0; - int neg = *s == '-'; - if (neg) - s++; - while (*s >= '0' && *s <= '9') - n = 10*n + (*s++ - '0'); - return neg ? -n : n; -} - - -/* Re-runs fn until it doesn't cause EINTR - */ -#define NO_INTR(fn) do {} while ((fn) < 0 && errno == EINTR) - - -/* Wrap a class around system calls, in order to give us access to - * a private copy of errno. This only works in C++, but it has the - * advantage of not needing nested functions, which are a non-standard - * language extension. - */ -#ifdef __cplusplus -namespace { - class SysCalls { - public: - #define SYS_CPLUSPLUS - #define SYS_ERRNO my_errno - #define SYS_INLINE inline - #define SYS_PREFIX -1 - #undef SYS_LINUX_SYSCALL_SUPPORT_H - #include "linux_syscall_support.h" - SysCalls() : my_errno(0) { } - int my_errno; - }; -} -#define ERRNO sys.my_errno -#else -#define ERRNO my_errno -#endif - - -/* Wrapper for open() which is guaranteed to never return EINTR. - */ -static int c_open(const char *fname, int flags, int mode) { - ssize_t rc; - NO_INTR(rc = sys_open(fname, flags, mode)); - return rc; -} - - -/* abort() is not safely reentrant, and changes it's behavior each time - * it is called. This means, if the main application ever called abort() - * we cannot safely call it again. This would happen if we were called - * from a SIGABRT signal handler in the main application. So, document - * that calling SIGABRT from the thread lister makes it not signal safe - * (and vice-versa). - * Also, since we share address space with the main application, we - * cannot call abort() from the callback and expect the main application - * to behave correctly afterwards. In fact, the only thing we can do, is - * to terminate the main application with extreme prejudice (aka - * PTRACE_KILL). - * We set up our own SIGABRT handler to do this. - * In order to find the main application from the signal handler, we - * need to store information about it in global variables. This is - * safe, because the main application should be suspended at this - * time. If the callback ever called TCMalloc_ResumeAllProcessThreads(), then - * we are running a higher risk, though. So, try to avoid calling - * abort() after calling TCMalloc_ResumeAllProcessThreads. - */ -static volatile int *sig_pids, sig_num_threads, sig_proc, sig_marker; - - -/* Signal handler to help us recover from dying while we are attached to - * other threads. - */ -static void SignalHandler(int signum, siginfo_t *si, void *data) { - if (sig_pids != NULL) { - if (signum == SIGABRT) { - while (sig_num_threads-- > 0) { - /* Not sure if sched_yield is really necessary here, but it does not */ - /* hurt, and it might be necessary for the same reasons that we have */ - /* to do so in sys_ptrace_detach(). */ - sys_sched_yield(); - sys_ptrace(PTRACE_KILL, sig_pids[sig_num_threads], 0, 0); - } - } else if (sig_num_threads > 0) { - TCMalloc_ResumeAllProcessThreads(sig_num_threads, (int *)sig_pids); - } - } - sig_pids = NULL; - if (sig_marker >= 0) - NO_INTR(sys_close(sig_marker)); - sig_marker = -1; - if (sig_proc >= 0) - NO_INTR(sys_close(sig_proc)); - sig_proc = -1; - - sys__exit(signum == SIGABRT ? 1 : 2); -} - - -/* Try to dirty the stack, and hope that the compiler is not smart enough - * to optimize this function away. Or worse, the compiler could inline the - * function and permanently allocate the data on the stack. - */ -static void DirtyStack(size_t amount) { - char buf[amount]; - memset(buf, 0, amount); - sys_read(-1, buf, amount); -} - - -/* Data structure for passing arguments to the lister thread. - */ -#define ALT_STACKSIZE (MINSIGSTKSZ + 4096) - -struct ListerParams { - int result, err; - char *altstack_mem; - ListAllProcessThreadsCallBack callback; - void *parameter; - va_list ap; - sem_t *lock; -}; - - -static void ListerThread(struct ListerParams *args) { - int found_parent = 0; - pid_t clone_pid = sys_gettid(), ppid = sys_getppid(); - char proc_self_task[80], marker_name[48], *marker_path; - const char *proc_paths[3]; - const char *const *proc_path = proc_paths; - int proc = -1, marker = -1, num_threads = 0; - int max_threads = 0, sig; - struct kernel_stat marker_sb, proc_sb; - stack_t altstack; - - /* Wait for parent thread to set appropriate permissions - * to allow ptrace activity - */ - if (sem_wait(args->lock) < 0) { - goto failure; - } - - /* Create "marker" that we can use to detect threads sharing the same - * address space and the same file handles. By setting the FD_CLOEXEC flag - * we minimize the risk of misidentifying child processes as threads; - * and since there is still a race condition, we will filter those out - * later, anyway. - */ - if ((marker = sys_socket(PF_LOCAL, SOCK_DGRAM, 0)) < 0 || - sys_fcntl(marker, F_SETFD, FD_CLOEXEC) < 0) { - failure: - args->result = -1; - args->err = errno; - if (marker >= 0) - NO_INTR(sys_close(marker)); - sig_marker = marker = -1; - if (proc >= 0) - NO_INTR(sys_close(proc)); - sig_proc = proc = -1; - sys__exit(1); - } - - /* Compute search paths for finding thread directories in /proc */ - local_itoa(strrchr(strcpy(proc_self_task, "/proc/"), '\000'), ppid); - strcpy(marker_name, proc_self_task); - marker_path = marker_name + strlen(marker_name); - strcat(proc_self_task, "/task/"); - proc_paths[0] = proc_self_task; /* /proc/$$/task/ */ - proc_paths[1] = "/proc/"; /* /proc/ */ - proc_paths[2] = NULL; - - /* Compute path for marker socket in /proc */ - local_itoa(strcpy(marker_path, "/fd/") + 4, marker); - if (sys_stat(marker_name, &marker_sb) < 0) { - goto failure; - } - - /* Catch signals on an alternate pre-allocated stack. This way, we can - * safely execute the signal handler even if we ran out of memory. - */ - memset(&altstack, 0, sizeof(altstack)); - altstack.ss_sp = args->altstack_mem; - altstack.ss_flags = 0; - altstack.ss_size = ALT_STACKSIZE; - sys_sigaltstack(&altstack, (const stack_t *)NULL); - - /* Some kernels forget to wake up traced processes, when the - * tracer dies. So, intercept synchronous signals and make sure - * that we wake up our tracees before dying. It is the caller's - * responsibility to ensure that asynchronous signals do not - * interfere with this function. - */ - sig_marker = marker; - sig_proc = -1; - for (sig = 0; sig < sizeof(sync_signals)/sizeof(*sync_signals); sig++) { - struct kernel_sigaction sa; - memset(&sa, 0, sizeof(sa)); - sa.sa_sigaction_ = SignalHandler; - sys_sigfillset(&sa.sa_mask); - sa.sa_flags = SA_ONSTACK|SA_SIGINFO|SA_RESETHAND; - sys_sigaction(sync_signals[sig], &sa, (struct kernel_sigaction *)NULL); - } - - /* Read process directories in /proc/... */ - for (;;) { - /* Some kernels know about threads, and hide them in "/proc" - * (although they are still there, if you know the process - * id). Threads are moved into a separate "task" directory. We - * check there first, and then fall back on the older naming - * convention if necessary. - */ - if ((sig_proc = proc = c_open(*proc_path, O_RDONLY|O_DIRECTORY, 0)) < 0) { - if (*++proc_path != NULL) - continue; - goto failure; - } - if (sys_fstat(proc, &proc_sb) < 0) - goto failure; - - /* Since we are suspending threads, we cannot call any libc - * functions that might acquire locks. Most notably, we cannot - * call malloc(). So, we have to allocate memory on the stack, - * instead. Since we do not know how much memory we need, we - * make a best guess. And if we guessed incorrectly we retry on - * a second iteration (by jumping to "detach_threads"). - * - * Unless the number of threads is increasing very rapidly, we - * should never need to do so, though, as our guestimate is very - * conservative. - */ - if (max_threads < proc_sb.st_nlink + 100) - max_threads = proc_sb.st_nlink + 100; - - /* scope */ { - pid_t pids[max_threads]; - int added_entries = 0; - sig_num_threads = num_threads; - sig_pids = pids; - for (;;) { - struct KERNEL_DIRENT *entry; - char buf[4096]; - ssize_t nbytes = GETDENTS(proc, (struct KERNEL_DIRENT *)buf, - sizeof(buf)); - if (nbytes < 0) - goto failure; - else if (nbytes == 0) { - if (added_entries) { - /* Need to keep iterating over "/proc" in multiple - * passes until we no longer find any more threads. This - * algorithm eventually completes, when all threads have - * been suspended. - */ - added_entries = 0; - sys_lseek(proc, 0, SEEK_SET); - continue; - } - break; - } - for (entry = (struct KERNEL_DIRENT *)buf; - entry < (struct KERNEL_DIRENT *)&buf[nbytes]; - entry = (struct KERNEL_DIRENT *)((char *)entry+entry->d_reclen)) { - if (entry->d_ino != 0) { - const char *ptr = entry->d_name; - pid_t pid; - - /* Some kernels hide threads by preceding the pid with a '.' */ - if (*ptr == '.') - ptr++; - - /* If the directory is not numeric, it cannot be a - * process/thread - */ - if (*ptr < '0' || *ptr > '9') - continue; - pid = local_atoi(ptr); - - /* Attach (and suspend) all threads */ - if (pid && pid != clone_pid) { - struct kernel_stat tmp_sb; - char fname[entry->d_reclen + 48]; - strcat(strcat(strcpy(fname, "/proc/"), - entry->d_name), marker_path); - - /* Check if the marker is identical to the one we created */ - if (sys_stat(fname, &tmp_sb) >= 0 && - marker_sb.st_ino == tmp_sb.st_ino) { - long i, j; - - /* Found one of our threads, make sure it is no duplicate */ - for (i = 0; i < num_threads; i++) { - /* Linear search is slow, but should not matter much for - * the typically small number of threads. - */ - if (pids[i] == pid) { - /* Found a duplicate; most likely on second pass */ - goto next_entry; - } - } - - /* Check whether data structure needs growing */ - if (num_threads >= max_threads) { - /* Back to square one, this time with more memory */ - NO_INTR(sys_close(proc)); - goto detach_threads; - } - - /* Attaching to thread suspends it */ - pids[num_threads++] = pid; - sig_num_threads = num_threads; - if (sys_ptrace(PTRACE_ATTACH, pid, (void *)0, - (void *)0) < 0) { - /* If operation failed, ignore thread. Maybe it - * just died? There might also be a race - * condition with a concurrent core dumper or - * with a debugger. In that case, we will just - * make a best effort, rather than failing - * entirely. - */ - num_threads--; - sig_num_threads = num_threads; - goto next_entry; - } - while (sys_waitpid(pid, (int *)0, __WALL) < 0) { - if (errno != EINTR) { - sys_ptrace_detach(pid); - num_threads--; - sig_num_threads = num_threads; - goto next_entry; - } - } - - if (sys_ptrace(PTRACE_PEEKDATA, pid, &i, &j) || i++ != j || - sys_ptrace(PTRACE_PEEKDATA, pid, &i, &j) || i != j) { - /* Address spaces are distinct, even though both - * processes show the "marker". This is probably - * a forked child process rather than a thread. - */ - sys_ptrace_detach(pid); - num_threads--; - sig_num_threads = num_threads; - } else { - found_parent |= pid == ppid; - added_entries++; - } - } - } - } - next_entry:; - } - } - NO_INTR(sys_close(proc)); - sig_proc = proc = -1; - - /* If we failed to find any threads, try looking somewhere else in - * /proc. Maybe, threads are reported differently on this system. - */ - if (num_threads > 1 || !*++proc_path) { - NO_INTR(sys_close(marker)); - sig_marker = marker = -1; - - /* If we never found the parent process, something is very wrong. - * Most likely, we are running in debugger. Any attempt to operate - * on the threads would be very incomplete. Let's just report an - * error to the caller. - */ - if (!found_parent) { - TCMalloc_ResumeAllProcessThreads(num_threads, pids); - sys__exit(3); - } - - /* Now we are ready to call the callback, - * which takes care of resuming the threads for us. - */ - args->result = args->callback(args->parameter, num_threads, - pids, args->ap); - args->err = errno; - - /* Callback should have resumed threads, but better safe than sorry */ - if (TCMalloc_ResumeAllProcessThreads(num_threads, pids)) { - /* Callback forgot to resume at least one thread, report error */ - args->err = EINVAL; - args->result = -1; - } - - sys__exit(0); - } - detach_threads: - /* Resume all threads prior to retrying the operation */ - TCMalloc_ResumeAllProcessThreads(num_threads, pids); - sig_pids = NULL; - num_threads = 0; - sig_num_threads = num_threads; - max_threads += 100; - } - } -} - - -/* This function gets the list of all linux threads of the current process - * passes them to the 'callback' along with the 'parameter' pointer; at the - * call back call time all the threads are paused via - * PTRACE_ATTACH. - * The callback is executed from a separate thread which shares only the - * address space, the filesystem, and the filehandles with the caller. Most - * notably, it does not share the same pid and ppid; and if it terminates, - * the rest of the application is still there. 'callback' is supposed to do - * or arrange for TCMalloc_ResumeAllProcessThreads. This happens automatically, if - * the thread raises a synchronous signal (e.g. SIGSEGV); asynchronous - * signals are blocked. If the 'callback' decides to unblock them, it must - * ensure that they cannot terminate the application, or that - * TCMalloc_ResumeAllProcessThreads will get called. - * It is an error for the 'callback' to make any library calls that could - * acquire locks. Most notably, this means that most system calls have to - * avoid going through libc. Also, this means that it is not legal to call - * exit() or abort(). - * We return -1 on error and the return value of 'callback' on success. - */ -int TCMalloc_ListAllProcessThreads(void *parameter, - ListAllProcessThreadsCallBack callback, ...) { - char altstack_mem[ALT_STACKSIZE]; - struct ListerParams args; - pid_t clone_pid; - int dumpable = 1, sig; - struct kernel_sigset_t sig_blocked, sig_old; - sem_t lock; - - va_start(args.ap, callback); - - /* If we are short on virtual memory, initializing the alternate stack - * might trigger a SIGSEGV. Let's do this early, before it could get us - * into more trouble (i.e. before signal handlers try to use the alternate - * stack, and before we attach to other threads). - */ - memset(altstack_mem, 0, sizeof(altstack_mem)); - - /* Some of our cleanup functions could conceivable use more stack space. - * Try to touch the stack right now. This could be defeated by the compiler - * being too smart for it's own good, so try really hard. - */ - DirtyStack(32768); - - /* Make this process "dumpable". This is necessary in order to ptrace() - * after having called setuid(). - */ - dumpable = sys_prctl(PR_GET_DUMPABLE, 0); - if (!dumpable) - sys_prctl(PR_SET_DUMPABLE, 1); - - /* Fill in argument block for dumper thread */ - args.result = -1; - args.err = 0; - args.altstack_mem = altstack_mem; - args.parameter = parameter; - args.callback = callback; - args.lock = &lock; - - /* Before cloning the thread lister, block all asynchronous signals, as we */ - /* are not prepared to handle them. */ - sys_sigfillset(&sig_blocked); - for (sig = 0; sig < sizeof(sync_signals)/sizeof(*sync_signals); sig++) { - sys_sigdelset(&sig_blocked, sync_signals[sig]); - } - if (sys_sigprocmask(SIG_BLOCK, &sig_blocked, &sig_old)) { - args.err = errno; - args.result = -1; - goto failed; - } - - /* scope */ { - /* After cloning, both the parent and the child share the same instance - * of errno. We must make sure that at least one of these processes - * (in our case, the parent) uses modified syscall macros that update - * a local copy of errno, instead. - */ - #ifdef __cplusplus - #define sys0_sigprocmask sys.sigprocmask - #define sys0_waitpid sys.waitpid - SysCalls sys; - #else - int my_errno; - #define SYS_ERRNO my_errno - #define SYS_INLINE inline - #define SYS_PREFIX 0 - #undef SYS_LINUX_SYSCALL_SUPPORT_H - #include "linux_syscall_support.h" - #endif - - /* Lock before clone so that parent can set - * ptrace permissions (if necessary) prior - * to ListerThread actually executing - */ - if (sem_init(&lock, 0, 0) == 0) { - - int clone_errno; - clone_pid = local_clone((int (*)(void *))ListerThread, &args); - clone_errno = errno; - - sys_sigprocmask(SIG_SETMASK, &sig_old, &sig_old); - - if (clone_pid >= 0) { -#ifdef PR_SET_PTRACER - /* In newer versions of glibc permission must explicitly - * be given to allow for ptrace. - */ - prctl(PR_SET_PTRACER, clone_pid, 0, 0, 0); -#endif - /* Releasing the lock here allows the - * ListerThread to execute and ptrace us. - */ - sem_post(&lock); - int status, rc; - while ((rc = sys0_waitpid(clone_pid, &status, __WALL)) < 0 && - ERRNO == EINTR) { - /* Keep waiting */ - } - if (rc < 0) { - args.err = ERRNO; - args.result = -1; - } else if (WIFEXITED(status)) { - switch (WEXITSTATUS(status)) { - case 0: break; /* Normal process termination */ - case 2: args.err = EFAULT; /* Some fault (e.g. SIGSEGV) detected */ - args.result = -1; - break; - case 3: args.err = EPERM; /* Process is already being traced */ - args.result = -1; - break; - default:args.err = ECHILD; /* Child died unexpectedly */ - args.result = -1; - break; - } - } else if (!WIFEXITED(status)) { - args.err = EFAULT; /* Terminated due to an unhandled signal*/ - args.result = -1; - } - sem_destroy(&lock); - } else { - args.result = -1; - args.err = clone_errno; - } - } else { - args.result = -1; - args.err = errno; - } - } - - /* Restore the "dumpable" state of the process */ -failed: - if (!dumpable) - sys_prctl(PR_SET_DUMPABLE, dumpable); - - va_end(args.ap); - - errno = args.err; - return args.result; -} - -/* This function resumes the list of all linux threads that - * TCMalloc_ListAllProcessThreads pauses before giving to its callback. - * The function returns non-zero if at least one thread was - * suspended and has now been resumed. - */ -int TCMalloc_ResumeAllProcessThreads(int num_threads, pid_t *thread_pids) { - int detached_at_least_one = 0; - while (num_threads-- > 0) { - detached_at_least_one |= sys_ptrace_detach(thread_pids[num_threads]) >= 0; - } - return detached_at_least_one; -} - -#ifdef __cplusplus -} -#endif -#endif diff --git a/contrib/libtcmalloc/src/base/linuxthreads.h b/contrib/libtcmalloc/src/base/linuxthreads.h deleted file mode 100644 index 09ce45fc13f..00000000000 --- a/contrib/libtcmalloc/src/base/linuxthreads.h +++ /dev/null @@ -1,54 +0,0 @@ -/* Copyright (c) 2005-2007, Google Inc. - * All rights reserved. - * - * Redistribution and use in source and binary forms, with or without - * modification, are permitted provided that the following conditions are - * met: - * - * * Redistributions of source code must retain the above copyright - * notice, this list of conditions and the following disclaimer. - * * Redistributions in binary form must reproduce the above - * copyright notice, this list of conditions and the following disclaimer - * in the documentation and/or other materials provided with the - * distribution. - * * Neither the name of Google Inc. nor the names of its - * contributors may be used to endorse or promote products derived from - * this software without specific prior written permission. - * - * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS - * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT - * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR - * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT - * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, - * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT - * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, - * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY - * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT - * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE - * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. - * - * --- - * Author: Markus Gutschke - */ - -#ifndef _LINUXTHREADS_H -#define _LINUXTHREADS_H - -/* Include thread_lister.h to get the interface that we implement for linux. - */ - -/* We currently only support certain platforms on Linux. Porting to other - * related platforms should not be difficult. - */ -#if (defined(__i386__) || defined(__x86_64__) || defined(__arm__) || \ - defined(__mips__) || defined(__PPC__) || defined(__aarch64__) || \ - defined(__s390__)) && defined(__linux) - -/* Define the THREADS symbol to make sure that there is exactly one core dumper - * built into the library. - */ -#define THREADS "Linux /proc" - -#endif - -#endif /* _LINUXTHREADS_H */ diff --git a/contrib/libtcmalloc/src/base/logging.cc b/contrib/libtcmalloc/src/base/logging.cc deleted file mode 100644 index 2b0adcb8945..00000000000 --- a/contrib/libtcmalloc/src/base/logging.cc +++ /dev/null @@ -1,108 +0,0 @@ -// -*- Mode: C++; c-basic-offset: 2; indent-tabs-mode: nil -*- -// Copyright (c) 2007, Google Inc. -// All rights reserved. -// -// Redistribution and use in source and binary forms, with or without -// modification, are permitted provided that the following conditions are -// met: -// -// * Redistributions of source code must retain the above copyright -// notice, this list of conditions and the following disclaimer. -// * Redistributions in binary form must reproduce the above -// copyright notice, this list of conditions and the following disclaimer -// in the documentation and/or other materials provided with the -// distribution. -// * Neither the name of Google Inc. nor the names of its -// contributors may be used to endorse or promote products derived from -// this software without specific prior written permission. -// -// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS -// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT -// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR -// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT -// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, -// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT -// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, -// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY -// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT -// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE -// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. - -// --- -// This file just provides storage for FLAGS_verbose. - -#include "../config.h" -#include "base/logging.h" -#include "base/commandlineflags.h" - -DEFINE_int32(verbose, EnvToInt("PERFTOOLS_VERBOSE", 0), - "Set to numbers >0 for more verbose output, or <0 for less. " - "--verbose == -4 means we log fatal errors only."); - - -#if defined(_WIN32) || defined(__CYGWIN__) || defined(__CYGWIN32__) - -// While windows does have a POSIX-compatible API -// (_open/_write/_close), it acquires memory. Using this lower-level -// windows API is the closest we can get to being "raw". -RawFD RawOpenForWriting(const char* filename) { - // CreateFile allocates memory if file_name isn't absolute, so if - // that ever becomes a problem then we ought to compute the absolute - // path on its behalf (perhaps the ntdll/kernel function isn't aware - // of the working directory?) - RawFD fd = CreateFileA(filename, GENERIC_WRITE, 0, NULL, - CREATE_ALWAYS, 0, NULL); - if (fd != kIllegalRawFD && GetLastError() == ERROR_ALREADY_EXISTS) - SetEndOfFile(fd); // truncate the existing file - return fd; -} - -void RawWrite(RawFD handle, const char* buf, size_t len) { - while (len > 0) { - DWORD wrote; - BOOL ok = WriteFile(handle, buf, len, &wrote, NULL); - // We do not use an asynchronous file handle, so ok==false means an error - if (!ok) break; - buf += wrote; - len -= wrote; - } -} - -void RawClose(RawFD handle) { - CloseHandle(handle); -} - -#else // _WIN32 || __CYGWIN__ || __CYGWIN32__ - -#ifdef HAVE_SYS_TYPES_H -#include -#endif -#ifdef HAVE_UNISTD_H -#include -#endif -#ifdef HAVE_FCNTL_H -#include -#endif - -// Re-run fn until it doesn't cause EINTR. -#define NO_INTR(fn) do {} while ((fn) < 0 && errno == EINTR) - -RawFD RawOpenForWriting(const char* filename) { - return open(filename, O_WRONLY|O_CREAT|O_TRUNC, 0664); -} - -void RawWrite(RawFD fd, const char* buf, size_t len) { - while (len > 0) { - ssize_t r; - NO_INTR(r = write(fd, buf, len)); - if (r <= 0) break; - buf += r; - len -= r; - } -} - -void RawClose(RawFD fd) { - NO_INTR(close(fd)); -} - -#endif // _WIN32 || __CYGWIN__ || __CYGWIN32__ diff --git a/contrib/libtcmalloc/src/base/logging.h b/contrib/libtcmalloc/src/base/logging.h deleted file mode 100644 index fa22489bea3..00000000000 --- a/contrib/libtcmalloc/src/base/logging.h +++ /dev/null @@ -1,259 +0,0 @@ -// -*- Mode: C++; c-basic-offset: 2; indent-tabs-mode: nil -*- -// Copyright (c) 2005, Google Inc. -// All rights reserved. -// -// Redistribution and use in source and binary forms, with or without -// modification, are permitted provided that the following conditions are -// met: -// -// * Redistributions of source code must retain the above copyright -// notice, this list of conditions and the following disclaimer. -// * Redistributions in binary form must reproduce the above -// copyright notice, this list of conditions and the following disclaimer -// in the documentation and/or other materials provided with the -// distribution. -// * Neither the name of Google Inc. nor the names of its -// contributors may be used to endorse or promote products derived from -// this software without specific prior written permission. -// -// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS -// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT -// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR -// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT -// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, -// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT -// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, -// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY -// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT -// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE -// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. - -// --- -// This file contains #include information about logging-related stuff. -// Pretty much everybody needs to #include this file so that they can -// log various happenings. -// -#ifndef _LOGGING_H_ -#define _LOGGING_H_ - -#include "../config.h" -#include -#include -#include -#ifdef HAVE_UNISTD_H -#include // for write() -#endif -#include // for strlen(), strcmp() -#include -#include // for errno -#include "base/commandlineflags.h" - -// On some systems (like freebsd), we can't call write() at all in a -// global constructor, perhaps because errno hasn't been set up. -// (In windows, we can't call it because it might call malloc.) -// Calling the write syscall is safer (it doesn't set errno), so we -// prefer that. Note we don't care about errno for logging: we just -// do logging on a best-effort basis. -#if defined(_MSC_VER) -#define WRITE_TO_STDERR(buf, len) WriteToStderr(buf, len); // in port.cc -#elif defined(HAVE_SYS_SYSCALL_H) -#include -#define WRITE_TO_STDERR(buf, len) syscall(SYS_write, STDERR_FILENO, buf, len) -#else -#define WRITE_TO_STDERR(buf, len) write(STDERR_FILENO, buf, len) -#endif - -// MSVC and mingw define their own, safe version of vnsprintf (the -// windows one in broken) in port.cc. Everyone else can use the -// version here. We had to give it a unique name for windows. -#ifndef _WIN32 -# define perftools_vsnprintf vsnprintf -#endif - - -// We log all messages at this log-level and below. -// INFO == -1, WARNING == -2, ERROR == -3, FATAL == -4 -DECLARE_int32(verbose); - -// CHECK dies with a fatal error if condition is not true. It is *not* -// controlled by NDEBUG, so the check will be executed regardless of -// compilation mode. Therefore, it is safe to do things like: -// CHECK(fp->Write(x) == 4) -// Note we use write instead of printf/puts to avoid the risk we'll -// call malloc(). -#define CHECK(condition) \ - do { \ - if (!(condition)) { \ - WRITE_TO_STDERR("Check failed: " #condition "\n", \ - sizeof("Check failed: " #condition "\n")-1); \ - abort(); \ - } \ - } while (0) - -// This takes a message to print. The name is historical. -#define RAW_CHECK(condition, message) \ - do { \ - if (!(condition)) { \ - WRITE_TO_STDERR("Check failed: " #condition ": " message "\n", \ - sizeof("Check failed: " #condition ": " message "\n")-1);\ - abort(); \ - } \ - } while (0) - -// This is like RAW_CHECK, but only in debug-mode -#ifdef NDEBUG -enum { DEBUG_MODE = 0 }; -#define RAW_DCHECK(condition, message) -#else -enum { DEBUG_MODE = 1 }; -#define RAW_DCHECK(condition, message) RAW_CHECK(condition, message) -#endif - -// This prints errno as well. Note we use write instead of printf/puts to -// avoid the risk we'll call malloc(). -#define PCHECK(condition) \ - do { \ - if (!(condition)) { \ - const int err_no = errno; \ - WRITE_TO_STDERR("Check failed: " #condition ": ", \ - sizeof("Check failed: " #condition ": ")-1); \ - WRITE_TO_STDERR(strerror(err_no), strlen(strerror(err_no))); \ - WRITE_TO_STDERR("\n", sizeof("\n")-1); \ - abort(); \ - } \ - } while (0) - -// Helper macro for binary operators; prints the two values on error -// Don't use this macro directly in your code, use CHECK_EQ et al below - -// WARNING: These don't compile correctly if one of the arguments is a pointer -// and the other is NULL. To work around this, simply static_cast NULL to the -// type of the desired pointer. - -// TODO(jandrews): Also print the values in case of failure. Requires some -// sort of type-sensitive ToString() function. -#define CHECK_OP(op, val1, val2) \ - do { \ - if (!((val1) op (val2))) { \ - fprintf(stderr, "Check failed: %s %s %s\n", #val1, #op, #val2); \ - abort(); \ - } \ - } while (0) - -#define CHECK_EQ(val1, val2) CHECK_OP(==, val1, val2) -#define CHECK_NE(val1, val2) CHECK_OP(!=, val1, val2) -#define CHECK_LE(val1, val2) CHECK_OP(<=, val1, val2) -#define CHECK_LT(val1, val2) CHECK_OP(< , val1, val2) -#define CHECK_GE(val1, val2) CHECK_OP(>=, val1, val2) -#define CHECK_GT(val1, val2) CHECK_OP(> , val1, val2) - -// Synonyms for CHECK_* that are used in some unittests. -#define EXPECT_EQ(val1, val2) CHECK_EQ(val1, val2) -#define EXPECT_NE(val1, val2) CHECK_NE(val1, val2) -#define EXPECT_LE(val1, val2) CHECK_LE(val1, val2) -#define EXPECT_LT(val1, val2) CHECK_LT(val1, val2) -#define EXPECT_GE(val1, val2) CHECK_GE(val1, val2) -#define EXPECT_GT(val1, val2) CHECK_GT(val1, val2) -#define ASSERT_EQ(val1, val2) EXPECT_EQ(val1, val2) -#define ASSERT_NE(val1, val2) EXPECT_NE(val1, val2) -#define ASSERT_LE(val1, val2) EXPECT_LE(val1, val2) -#define ASSERT_LT(val1, val2) EXPECT_LT(val1, val2) -#define ASSERT_GE(val1, val2) EXPECT_GE(val1, val2) -#define ASSERT_GT(val1, val2) EXPECT_GT(val1, val2) -// As are these variants. -#define EXPECT_TRUE(cond) CHECK(cond) -#define EXPECT_FALSE(cond) CHECK(!(cond)) -#define EXPECT_STREQ(a, b) CHECK(strcmp(a, b) == 0) -#define ASSERT_TRUE(cond) EXPECT_TRUE(cond) -#define ASSERT_FALSE(cond) EXPECT_FALSE(cond) -#define ASSERT_STREQ(a, b) EXPECT_STREQ(a, b) - -// Used for (libc) functions that return -1 and set errno -#define CHECK_ERR(invocation) PCHECK((invocation) != -1) - -// A few more checks that only happen in debug mode -#ifdef NDEBUG -#define DCHECK_EQ(val1, val2) -#define DCHECK_NE(val1, val2) -#define DCHECK_LE(val1, val2) -#define DCHECK_LT(val1, val2) -#define DCHECK_GE(val1, val2) -#define DCHECK_GT(val1, val2) -#else -#define DCHECK_EQ(val1, val2) CHECK_EQ(val1, val2) -#define DCHECK_NE(val1, val2) CHECK_NE(val1, val2) -#define DCHECK_LE(val1, val2) CHECK_LE(val1, val2) -#define DCHECK_LT(val1, val2) CHECK_LT(val1, val2) -#define DCHECK_GE(val1, val2) CHECK_GE(val1, val2) -#define DCHECK_GT(val1, val2) CHECK_GT(val1, val2) -#endif - - -#ifdef ERROR -#undef ERROR // may conflict with ERROR macro on windows -#endif -enum LogSeverity {INFO = -1, WARNING = -2, ERROR = -3, FATAL = -4}; - -// NOTE: we add a newline to the end of the output if it's not there already -inline void LogPrintf(int severity, const char* pat, va_list ap) { - // We write directly to the stderr file descriptor and avoid FILE - // buffering because that may invoke malloc() - char buf[600]; - perftools_vsnprintf(buf, sizeof(buf)-1, pat, ap); - if (buf[0] != '\0' && buf[strlen(buf)-1] != '\n') { - assert(strlen(buf)+1 < sizeof(buf)); - strcat(buf, "\n"); - } - WRITE_TO_STDERR(buf, strlen(buf)); - if ((severity) == FATAL) - abort(); // LOG(FATAL) indicates a big problem, so don't run atexit() calls -} - -// Note that since the order of global constructors is unspecified, -// global code that calls RAW_LOG may execute before FLAGS_verbose is set. -// Such code will run with verbosity == 0 no matter what. -#define VLOG_IS_ON(severity) (FLAGS_verbose >= severity) - -// In a better world, we'd use __VA_ARGS__, but VC++ 7 doesn't support it. -#define LOG_PRINTF(severity, pat) do { \ - if (VLOG_IS_ON(severity)) { \ - va_list ap; \ - va_start(ap, pat); \ - LogPrintf(severity, pat, ap); \ - va_end(ap); \ - } \ -} while (0) - -// RAW_LOG is the main function; some synonyms are used in unittests. -inline void RAW_LOG(int lvl, const char* pat, ...) { LOG_PRINTF(lvl, pat); } -inline void RAW_VLOG(int lvl, const char* pat, ...) { LOG_PRINTF(lvl, pat); } -inline void LOG(int lvl, const char* pat, ...) { LOG_PRINTF(lvl, pat); } -inline void VLOG(int lvl, const char* pat, ...) { LOG_PRINTF(lvl, pat); } -inline void LOG_IF(int lvl, bool cond, const char* pat, ...) { - if (cond) LOG_PRINTF(lvl, pat); -} - -// This isn't technically logging, but it's also IO and also is an -// attempt to be "raw" -- that is, to not use any higher-level libc -// routines that might allocate memory or (ideally) try to allocate -// locks. We use an opaque file handle (not necessarily an int) -// to allow even more low-level stuff in the future. -// Like other "raw" routines, these functions are best effort, and -// thus don't return error codes (except RawOpenForWriting()). -#if defined(_WIN32) || defined(__CYGWIN__) || defined(__CYGWIN32__) -#ifndef NOMINMAX -#define NOMINMAX // @#!$& windows -#endif -#include -typedef HANDLE RawFD; -const RawFD kIllegalRawFD = INVALID_HANDLE_VALUE; -#else -typedef int RawFD; -const RawFD kIllegalRawFD = -1; // what open returns if it fails -#endif // defined(_WIN32) || defined(__CYGWIN__) || defined(__CYGWIN32__) - -RawFD RawOpenForWriting(const char* filename); // uses default permissions -void RawWrite(RawFD fd, const char* buf, size_t len); -void RawClose(RawFD fd); - -#endif // _LOGGING_H_ diff --git a/contrib/libtcmalloc/src/base/low_level_alloc.cc b/contrib/libtcmalloc/src/base/low_level_alloc.cc deleted file mode 100644 index 6b467cff123..00000000000 --- a/contrib/libtcmalloc/src/base/low_level_alloc.cc +++ /dev/null @@ -1,582 +0,0 @@ -// -*- Mode: C++; c-basic-offset: 2; indent-tabs-mode: nil -*- -/* Copyright (c) 2006, Google Inc. - * All rights reserved. - * - * Redistribution and use in source and binary forms, with or without - * modification, are permitted provided that the following conditions are - * met: - * - * * Redistributions of source code must retain the above copyright - * notice, this list of conditions and the following disclaimer. - * * Redistributions in binary form must reproduce the above - * copyright notice, this list of conditions and the following disclaimer - * in the documentation and/or other materials provided with the - * distribution. - * * Neither the name of Google Inc. nor the names of its - * contributors may be used to endorse or promote products derived from - * this software without specific prior written permission. - * - * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS - * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT - * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR - * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT - * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, - * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT - * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, - * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY - * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT - * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE - * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. - */ - -// A low-level allocator that can be used by other low-level -// modules without introducing dependency cycles. -// This allocator is slow and wasteful of memory; -// it should not be used when performance is key. - -#include "base/low_level_alloc.h" -#include "base/dynamic_annotations.h" -#include "base/spinlock.h" -#include "base/logging.h" -#include "malloc_hook-inl.h" -#include -#include -#ifdef HAVE_UNISTD_H -#include -#endif -#ifdef HAVE_MMAP -#include -#endif -#include // for placement-new - -// On systems (like freebsd) that don't define MAP_ANONYMOUS, use the old -// form of the name instead. -#ifndef MAP_ANONYMOUS -# define MAP_ANONYMOUS MAP_ANON -#endif - -// A first-fit allocator with amortized logarithmic free() time. - -LowLevelAlloc::PagesAllocator::~PagesAllocator() { -} - -// --------------------------------------------------------------------------- -static const int kMaxLevel = 30; - -// We put this class-only struct in a namespace to avoid polluting the -// global namespace with this struct name (thus risking an ODR violation). -namespace low_level_alloc_internal { - // This struct describes one allocated block, or one free block. - struct AllocList { - struct Header { - intptr_t size; // size of entire region, including this field. Must be - // first. Valid in both allocated and unallocated blocks - intptr_t magic; // kMagicAllocated or kMagicUnallocated xor this - LowLevelAlloc::Arena *arena; // pointer to parent arena - void *dummy_for_alignment; // aligns regions to 0 mod 2*sizeof(void*) - } header; - - // Next two fields: in unallocated blocks: freelist skiplist data - // in allocated blocks: overlaps with client data - int levels; // levels in skiplist used - AllocList *next[kMaxLevel]; // actually has levels elements. - // The AllocList node may not have room for - // all kMaxLevel entries. See max_fit in - // LLA_SkiplistLevels() - }; -} -using low_level_alloc_internal::AllocList; - - -// --------------------------------------------------------------------------- -// A trivial skiplist implementation. This is used to keep the freelist -// in address order while taking only logarithmic time per insert and delete. - -// An integer approximation of log2(size/base) -// Requires size >= base. -static int IntLog2(size_t size, size_t base) { - int result = 0; - for (size_t i = size; i > base; i >>= 1) { // i == floor(size/2**result) - result++; - } - // floor(size / 2**result) <= base < floor(size / 2**(result-1)) - // => log2(size/(base+1)) <= result < 1+log2(size/base) - // => result ~= log2(size/base) - return result; -} - -// Return a random integer n: p(n)=1/(2**n) if 1 <= n; p(n)=0 if n < 1. -static int Random() { - static uint32 r = 1; // no locking---it's not critical - ANNOTATE_BENIGN_RACE(&r, "benign race, not critical."); - int result = 1; - while ((((r = r*1103515245 + 12345) >> 30) & 1) == 0) { - result++; - } - return result; -} - -// Return a number of skiplist levels for a node of size bytes, where -// base is the minimum node size. Compute level=log2(size / base)+n -// where n is 1 if random is false and otherwise a random number generated with -// the standard distribution for a skiplist: See Random() above. -// Bigger nodes tend to have more skiplist levels due to the log2(size / base) -// term, so first-fit searches touch fewer nodes. "level" is clipped so -// level max_fit) level = max_fit; - if (level > kMaxLevel-1) level = kMaxLevel - 1; - RAW_CHECK(level >= 1, "block not big enough for even one level"); - return level; -} - -// Return "atleast", the first element of AllocList *head s.t. *atleast >= *e. -// For 0 <= i < head->levels, set prev[i] to "no_greater", where no_greater -// points to the last element at level i in the AllocList less than *e, or is -// head if no such element exists. -static AllocList *LLA_SkiplistSearch(AllocList *head, - AllocList *e, AllocList **prev) { - AllocList *p = head; - for (int level = head->levels - 1; level >= 0; level--) { - for (AllocList *n; (n = p->next[level]) != 0 && n < e; p = n) { - } - prev[level] = p; - } - return (head->levels == 0) ? 0 : prev[0]->next[0]; -} - -// Insert element *e into AllocList *head. Set prev[] as LLA_SkiplistSearch. -// Requires that e->levels be previously set by the caller (using -// LLA_SkiplistLevels()) -static void LLA_SkiplistInsert(AllocList *head, AllocList *e, - AllocList **prev) { - LLA_SkiplistSearch(head, e, prev); - for (; head->levels < e->levels; head->levels++) { // extend prev pointers - prev[head->levels] = head; // to all *e's levels - } - for (int i = 0; i != e->levels; i++) { // add element to list - e->next[i] = prev[i]->next[i]; - prev[i]->next[i] = e; - } -} - -// Remove element *e from AllocList *head. Set prev[] as LLA_SkiplistSearch(). -// Requires that e->levels be previous set by the caller (using -// LLA_SkiplistLevels()) -static void LLA_SkiplistDelete(AllocList *head, AllocList *e, - AllocList **prev) { - AllocList *found = LLA_SkiplistSearch(head, e, prev); - RAW_CHECK(e == found, "element not in freelist"); - for (int i = 0; i != e->levels && prev[i]->next[i] == e; i++) { - prev[i]->next[i] = e->next[i]; - } - while (head->levels > 0 && head->next[head->levels - 1] == 0) { - head->levels--; // reduce head->levels if level unused - } -} - -// --------------------------------------------------------------------------- -// Arena implementation - -struct LowLevelAlloc::Arena { - Arena() : mu(SpinLock::LINKER_INITIALIZED) {} // does nothing; for static init - explicit Arena(int) : pagesize(0) {} // set pagesize to zero explicitly - // for non-static init - - SpinLock mu; // protects freelist, allocation_count, - // pagesize, roundup, min_size - AllocList freelist; // head of free list; sorted by addr (under mu) - int32 allocation_count; // count of allocated blocks (under mu) - int32 flags; // flags passed to NewArena (ro after init) - size_t pagesize; // ==getpagesize() (init under mu, then ro) - size_t roundup; // lowest power of 2 >= max(16,sizeof (AllocList)) - // (init under mu, then ro) - size_t min_size; // smallest allocation block size - // (init under mu, then ro) - PagesAllocator *allocator; -}; - -// The default arena, which is used when 0 is passed instead of an Arena -// pointer. -static struct LowLevelAlloc::Arena default_arena; - -// Non-malloc-hooked arenas: used only to allocate metadata for arenas that -// do not want malloc hook reporting, so that for them there's no malloc hook -// reporting even during arena creation. -static struct LowLevelAlloc::Arena unhooked_arena; -static struct LowLevelAlloc::Arena unhooked_async_sig_safe_arena; - -namespace { - - class DefaultPagesAllocator : public LowLevelAlloc::PagesAllocator { - public: - virtual ~DefaultPagesAllocator() {}; - virtual void *MapPages(int32 flags, size_t size); - virtual void UnMapPages(int32 flags, void *addr, size_t size); - }; - -} - -// magic numbers to identify allocated and unallocated blocks -static const intptr_t kMagicAllocated = 0x4c833e95; -static const intptr_t kMagicUnallocated = ~kMagicAllocated; - -namespace { - class SCOPED_LOCKABLE ArenaLock { - public: - explicit ArenaLock(LowLevelAlloc::Arena *arena) - EXCLUSIVE_LOCK_FUNCTION(arena->mu) - : left_(false), mask_valid_(false), arena_(arena) { - if ((arena->flags & LowLevelAlloc::kAsyncSignalSafe) != 0) { - // We've decided not to support async-signal-safe arena use until - // there a demonstrated need. Here's how one could do it though - // (would need to be made more portable). -#if 0 - sigset_t all; - sigfillset(&all); - this->mask_valid_ = - (pthread_sigmask(SIG_BLOCK, &all, &this->mask_) == 0); -#else - RAW_CHECK(false, "We do not yet support async-signal-safe arena."); -#endif - } - this->arena_->mu.Lock(); - } - ~ArenaLock() { RAW_CHECK(this->left_, "haven't left Arena region"); } - void Leave() /*UNLOCK_FUNCTION()*/ { - this->arena_->mu.Unlock(); -#if 0 - if (this->mask_valid_) { - pthread_sigmask(SIG_SETMASK, &this->mask_, 0); - } -#endif - this->left_ = true; - } - private: - bool left_; // whether left region - bool mask_valid_; -#if 0 - sigset_t mask_; // old mask of blocked signals -#endif - LowLevelAlloc::Arena *arena_; - DISALLOW_COPY_AND_ASSIGN(ArenaLock); - }; -} // anonymous namespace - -// create an appropriate magic number for an object at "ptr" -// "magic" should be kMagicAllocated or kMagicUnallocated -inline static intptr_t Magic(intptr_t magic, AllocList::Header *ptr) { - return magic ^ reinterpret_cast(ptr); -} - -// Initialize the fields of an Arena -static void ArenaInit(LowLevelAlloc::Arena *arena) { - if (arena->pagesize == 0) { - arena->pagesize = getpagesize(); - // Round up block sizes to a power of two close to the header size. - arena->roundup = 16; - while (arena->roundup < sizeof (arena->freelist.header)) { - arena->roundup += arena->roundup; - } - // Don't allocate blocks less than twice the roundup size to avoid tiny - // free blocks. - arena->min_size = 2 * arena->roundup; - arena->freelist.header.size = 0; - arena->freelist.header.magic = - Magic(kMagicUnallocated, &arena->freelist.header); - arena->freelist.header.arena = arena; - arena->freelist.levels = 0; - memset(arena->freelist.next, 0, sizeof (arena->freelist.next)); - arena->allocation_count = 0; - if (arena == &default_arena) { - // Default arena should be hooked, e.g. for heap-checker to trace - // pointer chains through objects in the default arena. - arena->flags = LowLevelAlloc::kCallMallocHook; - } else if (arena == &unhooked_async_sig_safe_arena) { - arena->flags = LowLevelAlloc::kAsyncSignalSafe; - } else { - arena->flags = 0; // other arenas' flags may be overridden by client, - // but unhooked_arena will have 0 in 'flags'. - } - arena->allocator = LowLevelAlloc::GetDefaultPagesAllocator(); - } -} - -// L < meta_data_arena->mu -LowLevelAlloc::Arena *LowLevelAlloc::NewArena(int32 flags, - Arena *meta_data_arena) { - return NewArenaWithCustomAlloc(flags, meta_data_arena, NULL); -} - -// L < meta_data_arena->mu -LowLevelAlloc::Arena *LowLevelAlloc::NewArenaWithCustomAlloc(int32 flags, - Arena *meta_data_arena, - PagesAllocator *allocator) { - RAW_CHECK(meta_data_arena != 0, "must pass a valid arena"); - if (meta_data_arena == &default_arena) { - if ((flags & LowLevelAlloc::kAsyncSignalSafe) != 0) { - meta_data_arena = &unhooked_async_sig_safe_arena; - } else if ((flags & LowLevelAlloc::kCallMallocHook) == 0) { - meta_data_arena = &unhooked_arena; - } - } - // Arena(0) uses the constructor for non-static contexts - Arena *result = - new (AllocWithArena(sizeof (*result), meta_data_arena)) Arena(0); - ArenaInit(result); - result->flags = flags; - if (allocator) { - result->allocator = allocator; - } - return result; -} - -// L < arena->mu, L < arena->arena->mu -bool LowLevelAlloc::DeleteArena(Arena *arena) { - RAW_CHECK(arena != 0 && arena != &default_arena && arena != &unhooked_arena, - "may not delete default arena"); - ArenaLock section(arena); - bool empty = (arena->allocation_count == 0); - section.Leave(); - if (empty) { - while (arena->freelist.next[0] != 0) { - AllocList *region = arena->freelist.next[0]; - size_t size = region->header.size; - arena->freelist.next[0] = region->next[0]; - RAW_CHECK(region->header.magic == - Magic(kMagicUnallocated, ®ion->header), - "bad magic number in DeleteArena()"); - RAW_CHECK(region->header.arena == arena, - "bad arena pointer in DeleteArena()"); - RAW_CHECK(size % arena->pagesize == 0, - "empty arena has non-page-aligned block size"); - RAW_CHECK(reinterpret_cast(region) % arena->pagesize == 0, - "empty arena has non-page-aligned block"); - int munmap_result; - if ((arena->flags & LowLevelAlloc::kAsyncSignalSafe) == 0) { - munmap_result = munmap(region, size); - } else { - munmap_result = MallocHook::UnhookedMUnmap(region, size); - } - RAW_CHECK(munmap_result == 0, - "LowLevelAlloc::DeleteArena: munmap failed address"); - } - Free(arena); - } - return empty; -} - -// --------------------------------------------------------------------------- - -// Return value rounded up to next multiple of align. -// align must be a power of two. -static intptr_t RoundUp(intptr_t addr, intptr_t align) { - return (addr + align - 1) & ~(align - 1); -} - -// Equivalent to "return prev->next[i]" but with sanity checking -// that the freelist is in the correct order, that it -// consists of regions marked "unallocated", and that no two regions -// are adjacent in memory (they should have been coalesced). -// L < arena->mu -static AllocList *Next(int i, AllocList *prev, LowLevelAlloc::Arena *arena) { - RAW_CHECK(i < prev->levels, "too few levels in Next()"); - AllocList *next = prev->next[i]; - if (next != 0) { - RAW_CHECK(next->header.magic == Magic(kMagicUnallocated, &next->header), - "bad magic number in Next()"); - RAW_CHECK(next->header.arena == arena, - "bad arena pointer in Next()"); - if (prev != &arena->freelist) { - RAW_CHECK(prev < next, "unordered freelist"); - RAW_CHECK(reinterpret_cast(prev) + prev->header.size < - reinterpret_cast(next), "malformed freelist"); - } - } - return next; -} - -// Coalesce list item "a" with its successor if they are adjacent. -static void Coalesce(AllocList *a) { - AllocList *n = a->next[0]; - if (n != 0 && reinterpret_cast(a) + a->header.size == - reinterpret_cast(n)) { - LowLevelAlloc::Arena *arena = a->header.arena; - a->header.size += n->header.size; - n->header.magic = 0; - n->header.arena = 0; - AllocList *prev[kMaxLevel]; - LLA_SkiplistDelete(&arena->freelist, n, prev); - LLA_SkiplistDelete(&arena->freelist, a, prev); - a->levels = LLA_SkiplistLevels(a->header.size, arena->min_size, true); - LLA_SkiplistInsert(&arena->freelist, a, prev); - } -} - -// Adds block at location "v" to the free list -// L >= arena->mu -static void AddToFreelist(void *v, LowLevelAlloc::Arena *arena) { - AllocList *f = reinterpret_cast( - reinterpret_cast(v) - sizeof (f->header)); - RAW_CHECK(f->header.magic == Magic(kMagicAllocated, &f->header), - "bad magic number in AddToFreelist()"); - RAW_CHECK(f->header.arena == arena, - "bad arena pointer in AddToFreelist()"); - f->levels = LLA_SkiplistLevels(f->header.size, arena->min_size, true); - AllocList *prev[kMaxLevel]; - LLA_SkiplistInsert(&arena->freelist, f, prev); - f->header.magic = Magic(kMagicUnallocated, &f->header); - Coalesce(f); // maybe coalesce with successor - Coalesce(prev[0]); // maybe coalesce with predecessor -} - -// Frees storage allocated by LowLevelAlloc::Alloc(). -// L < arena->mu -void LowLevelAlloc::Free(void *v) { - if (v != 0) { - AllocList *f = reinterpret_cast( - reinterpret_cast(v) - sizeof (f->header)); - RAW_CHECK(f->header.magic == Magic(kMagicAllocated, &f->header), - "bad magic number in Free()"); - LowLevelAlloc::Arena *arena = f->header.arena; - if ((arena->flags & kCallMallocHook) != 0) { - MallocHook::InvokeDeleteHook(v); - } - ArenaLock section(arena); - AddToFreelist(v, arena); - RAW_CHECK(arena->allocation_count > 0, "nothing in arena to free"); - arena->allocation_count--; - section.Leave(); - } -} - -// allocates and returns a block of size bytes, to be freed with Free() -// L < arena->mu -static void *DoAllocWithArena(size_t request, LowLevelAlloc::Arena *arena) { - void *result = 0; - if (request != 0) { - AllocList *s; // will point to region that satisfies request - ArenaLock section(arena); - ArenaInit(arena); - // round up with header - size_t req_rnd = RoundUp(request + sizeof (s->header), arena->roundup); - for (;;) { // loop until we find a suitable region - // find the minimum levels that a block of this size must have - int i = LLA_SkiplistLevels(req_rnd, arena->min_size, false) - 1; - if (i < arena->freelist.levels) { // potential blocks exist - AllocList *before = &arena->freelist; // predecessor of s - while ((s = Next(i, before, arena)) != 0 && s->header.size < req_rnd) { - before = s; - } - if (s != 0) { // we found a region - break; - } - } - // we unlock before mmap() both because mmap() may call a callback hook, - // and because it may be slow. - arena->mu.Unlock(); - // mmap generous 64K chunks to decrease - // the chances/impact of fragmentation: - size_t new_pages_size = RoundUp(req_rnd, arena->pagesize * 16); - void *new_pages = arena->allocator->MapPages(arena->flags, new_pages_size); - arena->mu.Lock(); - s = reinterpret_cast(new_pages); - s->header.size = new_pages_size; - // Pretend the block is allocated; call AddToFreelist() to free it. - s->header.magic = Magic(kMagicAllocated, &s->header); - s->header.arena = arena; - AddToFreelist(&s->levels, arena); // insert new region into free list - } - AllocList *prev[kMaxLevel]; - LLA_SkiplistDelete(&arena->freelist, s, prev); // remove from free list - // s points to the first free region that's big enough - if (req_rnd + arena->min_size <= s->header.size) { // big enough to split - AllocList *n = reinterpret_cast - (req_rnd + reinterpret_cast(s)); - n->header.size = s->header.size - req_rnd; - n->header.magic = Magic(kMagicAllocated, &n->header); - n->header.arena = arena; - s->header.size = req_rnd; - AddToFreelist(&n->levels, arena); - } - s->header.magic = Magic(kMagicAllocated, &s->header); - RAW_CHECK(s->header.arena == arena, ""); - arena->allocation_count++; - section.Leave(); - result = &s->levels; - } - ANNOTATE_NEW_MEMORY(result, request); - return result; -} - -void *LowLevelAlloc::Alloc(size_t request) { - void *result = DoAllocWithArena(request, &default_arena); - if ((default_arena.flags & kCallMallocHook) != 0) { - // this call must be directly in the user-called allocator function - // for MallocHook::GetCallerStackTrace to work properly - MallocHook::InvokeNewHook(result, request); - } - return result; -} - -void *LowLevelAlloc::AllocWithArena(size_t request, Arena *arena) { - RAW_CHECK(arena != 0, "must pass a valid arena"); - void *result = DoAllocWithArena(request, arena); - if ((arena->flags & kCallMallocHook) != 0) { - // this call must be directly in the user-called allocator function - // for MallocHook::GetCallerStackTrace to work properly - MallocHook::InvokeNewHook(result, request); - } - return result; -} - -LowLevelAlloc::Arena *LowLevelAlloc::DefaultArena() { - return &default_arena; -} - -static DefaultPagesAllocator *default_pages_allocator; -static union { - char chars[sizeof(DefaultPagesAllocator)]; - void *ptr; -} debug_pages_allocator_space; - -LowLevelAlloc::PagesAllocator *LowLevelAlloc::GetDefaultPagesAllocator(void) { - if (default_pages_allocator) { - return default_pages_allocator; - } - default_pages_allocator = new (debug_pages_allocator_space.chars) DefaultPagesAllocator(); - return default_pages_allocator; -} - -void *DefaultPagesAllocator::MapPages(int32 flags, size_t size) { - void *new_pages; - if ((flags & LowLevelAlloc::kAsyncSignalSafe) != 0) { - new_pages = MallocHook::UnhookedMMap(0, size, - PROT_WRITE|PROT_READ, - MAP_ANONYMOUS|MAP_PRIVATE, -1, 0); - } else { - new_pages = mmap(0, size, - PROT_WRITE|PROT_READ, - MAP_ANONYMOUS|MAP_PRIVATE, -1, 0); - } - RAW_CHECK(new_pages != MAP_FAILED, "mmap error"); - - return new_pages; -} - -void DefaultPagesAllocator::UnMapPages(int32 flags, void *region, size_t size) { - int munmap_result; - if ((flags & LowLevelAlloc::kAsyncSignalSafe) == 0) { - munmap_result = munmap(region, size); - } else { - munmap_result = MallocHook::UnhookedMUnmap(region, size); - } - RAW_CHECK(munmap_result == 0, - "LowLevelAlloc::DeleteArena: munmap failed address"); -} diff --git a/contrib/libtcmalloc/src/base/low_level_alloc.h b/contrib/libtcmalloc/src/base/low_level_alloc.h deleted file mode 100644 index 8a20dd8b870..00000000000 --- a/contrib/libtcmalloc/src/base/low_level_alloc.h +++ /dev/null @@ -1,120 +0,0 @@ -// -*- Mode: C++; c-basic-offset: 2; indent-tabs-mode: nil -*- -/* Copyright (c) 2006, Google Inc. - * All rights reserved. - * - * Redistribution and use in source and binary forms, with or without - * modification, are permitted provided that the following conditions are - * met: - * - * * Redistributions of source code must retain the above copyright - * notice, this list of conditions and the following disclaimer. - * * Redistributions in binary form must reproduce the above - * copyright notice, this list of conditions and the following disclaimer - * in the documentation and/or other materials provided with the - * distribution. - * * Neither the name of Google Inc. nor the names of its - * contributors may be used to endorse or promote products derived from - * this software without specific prior written permission. - * - * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS - * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT - * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR - * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT - * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, - * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT - * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, - * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY - * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT - * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE - * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. - */ - -#if !defined(_BASE_LOW_LEVEL_ALLOC_H_) -#define _BASE_LOW_LEVEL_ALLOC_H_ - -// A simple thread-safe memory allocator that does not depend on -// mutexes or thread-specific data. It is intended to be used -// sparingly, and only when malloc() would introduce an unwanted -// dependency, such as inside the heap-checker. - -#include "../config.h" -#include // for size_t -#include "base/basictypes.h" - -class LowLevelAlloc { - public: - class PagesAllocator { - public: - virtual ~PagesAllocator(); - virtual void *MapPages(int32 flags, size_t size) = 0; - virtual void UnMapPages(int32 flags, void *addr, size_t size) = 0; - }; - - static PagesAllocator *GetDefaultPagesAllocator(void); - - struct Arena; // an arena from which memory may be allocated - - // Returns a pointer to a block of at least "request" bytes - // that have been newly allocated from the specific arena. - // for Alloc() call the DefaultArena() is used. - // Returns 0 if passed request==0. - // Does not return 0 under other circumstances; it crashes if memory - // is not available. - static void *Alloc(size_t request) - ATTRIBUTE_SECTION(malloc_hook); - static void *AllocWithArena(size_t request, Arena *arena) - ATTRIBUTE_SECTION(malloc_hook); - - // Deallocates a region of memory that was previously allocated with - // Alloc(). Does nothing if passed 0. "s" must be either 0, - // or must have been returned from a call to Alloc() and not yet passed to - // Free() since that call to Alloc(). The space is returned to the arena - // from which it was allocated. - static void Free(void *s) ATTRIBUTE_SECTION(malloc_hook); - - // ATTRIBUTE_SECTION(malloc_hook) for Alloc* and Free - // are to put all callers of MallocHook::Invoke* in this module - // into special section, - // so that MallocHook::GetCallerStackTrace can function accurately. - - // Create a new arena. - // The root metadata for the new arena is allocated in the - // meta_data_arena; the DefaultArena() can be passed for meta_data_arena. - // These values may be ored into flags: - enum { - // Report calls to Alloc() and Free() via the MallocHook interface. - // Set in the DefaultArena. - kCallMallocHook = 0x0001, - - // Make calls to Alloc(), Free() be async-signal-safe. Not set in - // DefaultArena(). - kAsyncSignalSafe = 0x0002, - - // When used with DefaultArena(), the NewArena() and DeleteArena() calls - // obey the flags given explicitly in the NewArena() call, even if those - // flags differ from the settings in DefaultArena(). So the call - // NewArena(kAsyncSignalSafe, DefaultArena()) is itself async-signal-safe, - // as well as generatating an arena that provides async-signal-safe - // Alloc/Free. - }; - static Arena *NewArena(int32 flags, Arena *meta_data_arena); - - // note: pages allocator will never be destroyed and allocated pages will never be freed - // When allocator is NULL, it's same as NewArena - static Arena *NewArenaWithCustomAlloc(int32 flags, Arena *meta_data_arena, PagesAllocator *allocator); - - // Destroys an arena allocated by NewArena and returns true, - // provided no allocated blocks remain in the arena. - // If allocated blocks remain in the arena, does nothing and - // returns false. - // It is illegal to attempt to destroy the DefaultArena(). - static bool DeleteArena(Arena *arena); - - // The default arena that always exists. - static Arena *DefaultArena(); - - private: - LowLevelAlloc(); // no instances -}; - -#endif diff --git a/contrib/libtcmalloc/src/base/simple_mutex.h b/contrib/libtcmalloc/src/base/simple_mutex.h deleted file mode 100644 index e57c1079283..00000000000 --- a/contrib/libtcmalloc/src/base/simple_mutex.h +++ /dev/null @@ -1,332 +0,0 @@ -// -*- Mode: C++; c-basic-offset: 2; indent-tabs-mode: nil -*- -// Copyright (c) 2007, Google Inc. -// All rights reserved. -// -// Redistribution and use in source and binary forms, with or without -// modification, are permitted provided that the following conditions are -// met: -// -// * Redistributions of source code must retain the above copyright -// notice, this list of conditions and the following disclaimer. -// * Redistributions in binary form must reproduce the above -// copyright notice, this list of conditions and the following disclaimer -// in the documentation and/or other materials provided with the -// distribution. -// * Neither the name of Google Inc. nor the names of its -// contributors may be used to endorse or promote products derived from -// this software without specific prior written permission. -// -// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS -// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT -// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR -// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT -// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, -// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT -// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, -// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY -// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT -// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE -// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. -// -// --- -// Author: Craig Silverstein. -// -// A simple mutex wrapper, supporting locks and read-write locks. -// You should assume the locks are *not* re-entrant. -// -// To use: you should define the following macros in your configure.ac: -// ACX_PTHREAD -// AC_RWLOCK -// The latter is defined in ../autoconf. -// -// This class is meant to be internal-only and should be wrapped by an -// internal namespace. Before you use this module, please give the -// name of your internal namespace for this module. Or, if you want -// to expose it, you'll want to move it to the Google namespace. We -// cannot put this class in global namespace because there can be some -// problems when we have multiple versions of Mutex in each shared object. -// -// NOTE: TryLock() is broken for NO_THREADS mode, at least in NDEBUG -// mode. -// -// CYGWIN NOTE: Cygwin support for rwlock seems to be buggy: -// http://www.cygwin.com/ml/cygwin/2008-12/msg00017.html -// Because of that, we might as well use windows locks for -// cygwin. They seem to be more reliable than the cygwin pthreads layer. -// -// TRICKY IMPLEMENTATION NOTE: -// This class is designed to be safe to use during -// dynamic-initialization -- that is, by global constructors that are -// run before main() starts. The issue in this case is that -// dynamic-initialization happens in an unpredictable order, and it -// could be that someone else's dynamic initializer could call a -// function that tries to acquire this mutex -- but that all happens -// before this mutex's constructor has run. (This can happen even if -// the mutex and the function that uses the mutex are in the same .cc -// file.) Basically, because Mutex does non-trivial work in its -// constructor, it's not, in the naive implementation, safe to use -// before dynamic initialization has run on it. -// -// The solution used here is to pair the actual mutex primitive with a -// bool that is set to true when the mutex is dynamically initialized. -// (Before that it's false.) Then we modify all mutex routines to -// look at the bool, and not try to lock/unlock until the bool makes -// it to true (which happens after the Mutex constructor has run.) -// -// This works because before main() starts -- particularly, during -// dynamic initialization -- there are no threads, so a) it's ok that -// the mutex operations are a no-op, since we don't need locking then -// anyway; and b) we can be quite confident our bool won't change -// state between a call to Lock() and a call to Unlock() (that would -// require a global constructor in one translation unit to call Lock() -// and another global constructor in another translation unit to call -// Unlock() later, which is pretty perverse). -// -// That said, it's tricky, and can conceivably fail; it's safest to -// avoid trying to acquire a mutex in a global constructor, if you -// can. One way it can fail is that a really smart compiler might -// initialize the bool to true at static-initialization time (too -// early) rather than at dynamic-initialization time. To discourage -// that, we set is_safe_ to true in code (not the constructor -// colon-initializer) and set it to true via a function that always -// evaluates to true, but that the compiler can't know always -// evaluates to true. This should be good enough. -// -// A related issue is code that could try to access the mutex -// after it's been destroyed in the global destructors (because -// the Mutex global destructor runs before some other global -// destructor, that tries to acquire the mutex). The way we -// deal with this is by taking a constructor arg that global -// mutexes should pass in, that causes the destructor to do no -// work. We still depend on the compiler not doing anything -// weird to a Mutex's memory after it is destroyed, but for a -// static global variable, that's pretty safe. - -#ifndef GOOGLE_MUTEX_H_ -#define GOOGLE_MUTEX_H_ - -#include "../config.h" - -#if defined(NO_THREADS) - typedef int MutexType; // to keep a lock-count -#elif defined(_WIN32) || defined(__CYGWIN__) || defined(__CYGWIN32__) -# ifndef WIN32_LEAN_AND_MEAN -# define WIN32_LEAN_AND_MEAN // We only need minimal includes -# endif - // We need Windows NT or later for TryEnterCriticalSection(). If you - // don't need that functionality, you can remove these _WIN32_WINNT - // lines, and change TryLock() to assert(0) or something. -# ifndef _WIN32_WINNT -# define _WIN32_WINNT 0x0400 -# endif -# include - typedef CRITICAL_SECTION MutexType; -#elif defined(HAVE_PTHREAD) && defined(HAVE_RWLOCK) - // Needed for pthread_rwlock_*. If it causes problems, you could take it - // out, but then you'd have to unset HAVE_RWLOCK (at least on linux -- it - // *does* cause problems for FreeBSD, or MacOSX, but isn't needed - // for locking there.) -# ifdef __linux__ -# define _XOPEN_SOURCE 500 // may be needed to get the rwlock calls -# endif -# include - typedef pthread_rwlock_t MutexType; -#elif defined(HAVE_PTHREAD) -# include - typedef pthread_mutex_t MutexType; -#else -# error Need to implement mutex.h for your architecture, or #define NO_THREADS -#endif - -#include -#include // for abort() - -#define MUTEX_NAMESPACE perftools_mutex_namespace - -namespace MUTEX_NAMESPACE { - -class Mutex { - public: - // This is used for the single-arg constructor - enum LinkerInitialized { LINKER_INITIALIZED }; - - // Create a Mutex that is not held by anybody. This constructor is - // typically used for Mutexes allocated on the heap or the stack. - inline Mutex(); - // This constructor should be used for global, static Mutex objects. - // It inhibits work being done by the destructor, which makes it - // safer for code that tries to acqiure this mutex in their global - // destructor. - inline Mutex(LinkerInitialized); - - // Destructor - inline ~Mutex(); - - inline void Lock(); // Block if needed until free then acquire exclusively - inline void Unlock(); // Release a lock acquired via Lock() - inline bool TryLock(); // If free, Lock() and return true, else return false - // Note that on systems that don't support read-write locks, these may - // be implemented as synonyms to Lock() and Unlock(). So you can use - // these for efficiency, but don't use them anyplace where being able - // to do shared reads is necessary to avoid deadlock. - inline void ReaderLock(); // Block until free or shared then acquire a share - inline void ReaderUnlock(); // Release a read share of this Mutex - inline void WriterLock() { Lock(); } // Acquire an exclusive lock - inline void WriterUnlock() { Unlock(); } // Release a lock from WriterLock() - - private: - MutexType mutex_; - // We want to make sure that the compiler sets is_safe_ to true only - // when we tell it to, and never makes assumptions is_safe_ is - // always true. volatile is the most reliable way to do that. - volatile bool is_safe_; - // This indicates which constructor was called. - bool destroy_; - - inline void SetIsSafe() { is_safe_ = true; } - - // Catch the error of writing Mutex when intending MutexLock. - Mutex(Mutex* /*ignored*/) {} - // Disallow "evil" constructors - Mutex(const Mutex&); - void operator=(const Mutex&); -}; - -// Now the implementation of Mutex for various systems -#if defined(NO_THREADS) - -// When we don't have threads, we can be either reading or writing, -// but not both. We can have lots of readers at once (in no-threads -// mode, that's most likely to happen in recursive function calls), -// but only one writer. We represent this by having mutex_ be -1 when -// writing and a number > 0 when reading (and 0 when no lock is held). -// -// In debug mode, we assert these invariants, while in non-debug mode -// we do nothing, for efficiency. That's why everything is in an -// assert. - -Mutex::Mutex() : mutex_(0) { } -Mutex::Mutex(Mutex::LinkerInitialized) : mutex_(0) { } -Mutex::~Mutex() { assert(mutex_ == 0); } -void Mutex::Lock() { assert(--mutex_ == -1); } -void Mutex::Unlock() { assert(mutex_++ == -1); } -bool Mutex::TryLock() { if (mutex_) return false; Lock(); return true; } -void Mutex::ReaderLock() { assert(++mutex_ > 0); } -void Mutex::ReaderUnlock() { assert(mutex_-- > 0); } - -#elif defined(_WIN32) || defined(__CYGWIN__) || defined(__CYGWIN32__) - -Mutex::Mutex() : destroy_(true) { - InitializeCriticalSection(&mutex_); - SetIsSafe(); -} -Mutex::Mutex(LinkerInitialized) : destroy_(false) { - InitializeCriticalSection(&mutex_); - SetIsSafe(); -} -Mutex::~Mutex() { if (destroy_) DeleteCriticalSection(&mutex_); } -void Mutex::Lock() { if (is_safe_) EnterCriticalSection(&mutex_); } -void Mutex::Unlock() { if (is_safe_) LeaveCriticalSection(&mutex_); } -bool Mutex::TryLock() { return is_safe_ ? - TryEnterCriticalSection(&mutex_) != 0 : true; } -void Mutex::ReaderLock() { Lock(); } // we don't have read-write locks -void Mutex::ReaderUnlock() { Unlock(); } - -#elif defined(HAVE_PTHREAD) && defined(HAVE_RWLOCK) - -#define SAFE_PTHREAD(fncall) do { /* run fncall if is_safe_ is true */ \ - if (is_safe_ && fncall(&mutex_) != 0) abort(); \ -} while (0) - -Mutex::Mutex() : destroy_(true) { - SetIsSafe(); - if (is_safe_ && pthread_rwlock_init(&mutex_, NULL) != 0) abort(); -} -Mutex::Mutex(Mutex::LinkerInitialized) : destroy_(false) { - SetIsSafe(); - if (is_safe_ && pthread_rwlock_init(&mutex_, NULL) != 0) abort(); -} -Mutex::~Mutex() { if (destroy_) SAFE_PTHREAD(pthread_rwlock_destroy); } -void Mutex::Lock() { SAFE_PTHREAD(pthread_rwlock_wrlock); } -void Mutex::Unlock() { SAFE_PTHREAD(pthread_rwlock_unlock); } -bool Mutex::TryLock() { return is_safe_ ? - pthread_rwlock_trywrlock(&mutex_) == 0 : true; } -void Mutex::ReaderLock() { SAFE_PTHREAD(pthread_rwlock_rdlock); } -void Mutex::ReaderUnlock() { SAFE_PTHREAD(pthread_rwlock_unlock); } -#undef SAFE_PTHREAD - -#elif defined(HAVE_PTHREAD) - -#define SAFE_PTHREAD(fncall) do { /* run fncall if is_safe_ is true */ \ - if (is_safe_ && fncall(&mutex_) != 0) abort(); \ -} while (0) - -Mutex::Mutex() : destroy_(true) { - SetIsSafe(); - if (is_safe_ && pthread_mutex_init(&mutex_, NULL) != 0) abort(); -} -Mutex::Mutex(Mutex::LinkerInitialized) : destroy_(false) { - SetIsSafe(); - if (is_safe_ && pthread_mutex_init(&mutex_, NULL) != 0) abort(); -} -Mutex::~Mutex() { if (destroy_) SAFE_PTHREAD(pthread_mutex_destroy); } -void Mutex::Lock() { SAFE_PTHREAD(pthread_mutex_lock); } -void Mutex::Unlock() { SAFE_PTHREAD(pthread_mutex_unlock); } -bool Mutex::TryLock() { return is_safe_ ? - pthread_mutex_trylock(&mutex_) == 0 : true; } -void Mutex::ReaderLock() { Lock(); } -void Mutex::ReaderUnlock() { Unlock(); } -#undef SAFE_PTHREAD - -#endif - -// -------------------------------------------------------------------------- -// Some helper classes - -// MutexLock(mu) acquires mu when constructed and releases it when destroyed. -class MutexLock { - public: - explicit MutexLock(Mutex *mu) : mu_(mu) { mu_->Lock(); } - ~MutexLock() { mu_->Unlock(); } - private: - Mutex * const mu_; - // Disallow "evil" constructors - MutexLock(const MutexLock&); - void operator=(const MutexLock&); -}; - -// ReaderMutexLock and WriterMutexLock do the same, for rwlocks -class ReaderMutexLock { - public: - explicit ReaderMutexLock(Mutex *mu) : mu_(mu) { mu_->ReaderLock(); } - ~ReaderMutexLock() { mu_->ReaderUnlock(); } - private: - Mutex * const mu_; - // Disallow "evil" constructors - ReaderMutexLock(const ReaderMutexLock&); - void operator=(const ReaderMutexLock&); -}; - -class WriterMutexLock { - public: - explicit WriterMutexLock(Mutex *mu) : mu_(mu) { mu_->WriterLock(); } - ~WriterMutexLock() { mu_->WriterUnlock(); } - private: - Mutex * const mu_; - // Disallow "evil" constructors - WriterMutexLock(const WriterMutexLock&); - void operator=(const WriterMutexLock&); -}; - -// Catch bug where variable name is omitted, e.g. MutexLock (&mu); -#define MutexLock(x) COMPILE_ASSERT(0, mutex_lock_decl_missing_var_name) -#define ReaderMutexLock(x) COMPILE_ASSERT(0, rmutex_lock_decl_missing_var_name) -#define WriterMutexLock(x) COMPILE_ASSERT(0, wmutex_lock_decl_missing_var_name) - -} // namespace MUTEX_NAMESPACE - -using namespace MUTEX_NAMESPACE; - -#undef MUTEX_NAMESPACE - -#endif /* #define GOOGLE_SIMPLE_MUTEX_H_ */ diff --git a/contrib/libtcmalloc/src/base/spinlock.cc b/contrib/libtcmalloc/src/base/spinlock.cc deleted file mode 100644 index 48bb163d1de..00000000000 --- a/contrib/libtcmalloc/src/base/spinlock.cc +++ /dev/null @@ -1,129 +0,0 @@ -// -*- Mode: C++; c-basic-offset: 2; indent-tabs-mode: nil -*- -/* Copyright (c) 2006, Google Inc. - * All rights reserved. - * - * Redistribution and use in source and binary forms, with or without - * modification, are permitted provided that the following conditions are - * met: - * - * * Redistributions of source code must retain the above copyright - * notice, this list of conditions and the following disclaimer. - * * Redistributions in binary form must reproduce the above - * copyright notice, this list of conditions and the following disclaimer - * in the documentation and/or other materials provided with the - * distribution. - * * Neither the name of Google Inc. nor the names of its - * contributors may be used to endorse or promote products derived from - * this software without specific prior written permission. - * - * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS - * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT - * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR - * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT - * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, - * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT - * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, - * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY - * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT - * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE - * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. - * - * --- - * Author: Sanjay Ghemawat - */ - -#include "../config.h" -#include "base/spinlock.h" -#include "base/spinlock_internal.h" -#include "base/sysinfo.h" /* for GetSystemCPUsCount() */ - -// NOTE on the Lock-state values: -// -// kSpinLockFree represents the unlocked state -// kSpinLockHeld represents the locked state with no waiters -// kSpinLockSleeper represents the locked state with waiters - -static int adaptive_spin_count = 0; - -const base::LinkerInitialized SpinLock::LINKER_INITIALIZED = - base::LINKER_INITIALIZED; - -namespace { -struct SpinLock_InitHelper { - SpinLock_InitHelper() { - // On multi-cpu machines, spin for longer before yielding - // the processor or sleeping. Reduces idle time significantly. - if (GetSystemCPUsCount() > 1) { - adaptive_spin_count = 1000; - } - } -}; - -// Hook into global constructor execution: -// We do not do adaptive spinning before that, -// but nothing lock-intensive should be going on at that time. -static SpinLock_InitHelper init_helper; - -inline void SpinlockPause(void) { -#if defined(__GNUC__) && (defined(__i386__) || defined(__x86_64__)) - __asm__ __volatile__("rep; nop" : : ); -#endif -} - -} // unnamed namespace - -// Monitor the lock to see if its value changes within some time -// period (adaptive_spin_count loop iterations). The last value read -// from the lock is returned from the method. -Atomic32 SpinLock::SpinLoop() { - int c = adaptive_spin_count; - while (base::subtle::NoBarrier_Load(&lockword_) != kSpinLockFree && --c > 0) { - SpinlockPause(); - } - return base::subtle::Acquire_CompareAndSwap(&lockword_, kSpinLockFree, - kSpinLockSleeper); -} - -void SpinLock::SlowLock() { - Atomic32 lock_value = SpinLoop(); - - int lock_wait_call_count = 0; - while (lock_value != kSpinLockFree) { - // If the lock is currently held, but not marked as having a sleeper, mark - // it as having a sleeper. - if (lock_value == kSpinLockHeld) { - // Here, just "mark" that the thread is going to sleep. Don't store the - // lock wait time in the lock as that will cause the current lock - // owner to think it experienced contention. - lock_value = base::subtle::Acquire_CompareAndSwap(&lockword_, - kSpinLockHeld, - kSpinLockSleeper); - if (lock_value == kSpinLockHeld) { - // Successfully transitioned to kSpinLockSleeper. Pass - // kSpinLockSleeper to the SpinLockDelay routine to properly indicate - // the last lock_value observed. - lock_value = kSpinLockSleeper; - } else if (lock_value == kSpinLockFree) { - // Lock is free again, so try and acquire it before sleeping. The - // new lock state will be the number of cycles this thread waited if - // this thread obtains the lock. - lock_value = base::subtle::Acquire_CompareAndSwap(&lockword_, - kSpinLockFree, - kSpinLockSleeper); - continue; // skip the delay at the end of the loop - } - } - - // Wait for an OS specific delay. - base::internal::SpinLockDelay(&lockword_, lock_value, - ++lock_wait_call_count); - // Spin again after returning from the wait routine to give this thread - // some chance of obtaining the lock. - lock_value = SpinLoop(); - } -} - -void SpinLock::SlowUnlock() { - // wake waiter if necessary - base::internal::SpinLockWake(&lockword_, false); -} diff --git a/contrib/libtcmalloc/src/base/spinlock.h b/contrib/libtcmalloc/src/base/spinlock.h deleted file mode 100644 index 42a4eb906a0..00000000000 --- a/contrib/libtcmalloc/src/base/spinlock.h +++ /dev/null @@ -1,143 +0,0 @@ -// -*- Mode: C++; c-basic-offset: 2; indent-tabs-mode: nil -*- -/* Copyright (c) 2006, Google Inc. - * All rights reserved. - * - * Redistribution and use in source and binary forms, with or without - * modification, are permitted provided that the following conditions are - * met: - * - * * Redistributions of source code must retain the above copyright - * notice, this list of conditions and the following disclaimer. - * * Redistributions in binary form must reproduce the above - * copyright notice, this list of conditions and the following disclaimer - * in the documentation and/or other materials provided with the - * distribution. - * * Neither the name of Google Inc. nor the names of its - * contributors may be used to endorse or promote products derived from - * this software without specific prior written permission. - * - * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS - * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT - * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR - * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT - * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, - * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT - * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, - * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY - * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT - * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE - * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. - * - * --- - * Author: Sanjay Ghemawat - */ - -// SpinLock is async signal safe. -// If used within a signal handler, all lock holders -// should block the signal even outside the signal handler. - -#ifndef BASE_SPINLOCK_H_ -#define BASE_SPINLOCK_H_ - -#include "../config.h" -#include "base/atomicops.h" -#include "base/basictypes.h" -#include "base/dynamic_annotations.h" -#include "base/thread_annotations.h" - -class LOCKABLE SpinLock { - public: - SpinLock() : lockword_(kSpinLockFree) { } - - // Special constructor for use with static SpinLock objects. E.g., - // - // static SpinLock lock(base::LINKER_INITIALIZED); - // - // When intialized using this constructor, we depend on the fact - // that the linker has already initialized the memory appropriately. - // A SpinLock constructed like this can be freely used from global - // initializers without worrying about the order in which global - // initializers run. - explicit SpinLock(base::LinkerInitialized /*x*/) { - // Does nothing; lockword_ is already initialized - } - - // Acquire this SpinLock. - // TODO(csilvers): uncomment the annotation when we figure out how to - // support this macro with 0 args (see thread_annotations.h) - inline void Lock() /*EXCLUSIVE_LOCK_FUNCTION()*/ { - if (base::subtle::Acquire_CompareAndSwap(&lockword_, kSpinLockFree, - kSpinLockHeld) != kSpinLockFree) { - SlowLock(); - } - ANNOTATE_RWLOCK_ACQUIRED(this, 1); - } - - // Try to acquire this SpinLock without blocking and return true if the - // acquisition was successful. If the lock was not acquired, false is - // returned. If this SpinLock is free at the time of the call, TryLock - // will return true with high probability. - inline bool TryLock() EXCLUSIVE_TRYLOCK_FUNCTION(true) { - bool res = - (base::subtle::Acquire_CompareAndSwap(&lockword_, kSpinLockFree, - kSpinLockHeld) == kSpinLockFree); - if (res) { - ANNOTATE_RWLOCK_ACQUIRED(this, 1); - } - return res; - } - - // Release this SpinLock, which must be held by the calling thread. - // TODO(csilvers): uncomment the annotation when we figure out how to - // support this macro with 0 args (see thread_annotations.h) - inline void Unlock() /*UNLOCK_FUNCTION()*/ { - ANNOTATE_RWLOCK_RELEASED(this, 1); - uint64 prev_value = static_cast( - base::subtle::Release_AtomicExchange(&lockword_, kSpinLockFree)); - if (prev_value != kSpinLockHeld) { - // Speed the wakeup of any waiter. - SlowUnlock(); - } - } - - // Determine if the lock is held. When the lock is held by the invoking - // thread, true will always be returned. Intended to be used as - // CHECK(lock.IsHeld()). - inline bool IsHeld() const { - return base::subtle::NoBarrier_Load(&lockword_) != kSpinLockFree; - } - - static const base::LinkerInitialized LINKER_INITIALIZED; // backwards compat - private: - enum { kSpinLockFree = 0 }; - enum { kSpinLockHeld = 1 }; - enum { kSpinLockSleeper = 2 }; - - volatile Atomic32 lockword_; - - void SlowLock(); - void SlowUnlock(); - Atomic32 SpinLoop(); - - DISALLOW_COPY_AND_ASSIGN(SpinLock); -}; - -// Corresponding locker object that arranges to acquire a spinlock for -// the duration of a C++ scope. -class SCOPED_LOCKABLE SpinLockHolder { - private: - SpinLock* lock_; - public: - inline explicit SpinLockHolder(SpinLock* l) EXCLUSIVE_LOCK_FUNCTION(l) - : lock_(l) { - l->Lock(); - } - // TODO(csilvers): uncomment the annotation when we figure out how to - // support this macro with 0 args (see thread_annotations.h) - inline ~SpinLockHolder() /*UNLOCK_FUNCTION()*/ { lock_->Unlock(); } -}; -// Catch bug where variable name is omitted, e.g. SpinLockHolder (&lock); -#define SpinLockHolder(x) COMPILE_ASSERT(0, spin_lock_decl_missing_var_name) - - -#endif // BASE_SPINLOCK_H_ diff --git a/contrib/libtcmalloc/src/base/spinlock_internal.cc b/contrib/libtcmalloc/src/base/spinlock_internal.cc deleted file mode 100644 index d9629717be1..00000000000 --- a/contrib/libtcmalloc/src/base/spinlock_internal.cc +++ /dev/null @@ -1,102 +0,0 @@ -// -*- Mode: C++; c-basic-offset: 2; indent-tabs-mode: nil -*- -/* Copyright (c) 2010, Google Inc. - * All rights reserved. - * - * Redistribution and use in source and binary forms, with or without - * modification, are permitted provided that the following conditions are - * met: - * - * * Redistributions of source code must retain the above copyright - * notice, this list of conditions and the following disclaimer. - * * Redistributions in binary form must reproduce the above - * copyright notice, this list of conditions and the following disclaimer - * in the documentation and/or other materials provided with the - * distribution. - * * Neither the name of Google Inc. nor the names of its - * contributors may be used to endorse or promote products derived from - * this software without specific prior written permission. - * - * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS - * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT - * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR - * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT - * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, - * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT - * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, - * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY - * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT - * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE - * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. - */ - -// The OS-specific header included below must provide two calls: -// base::internal::SpinLockDelay() and base::internal::SpinLockWake(). -// See spinlock_internal.h for the spec of SpinLockWake(). - -// void SpinLockDelay(volatile Atomic32 *w, int32 value, int loop) -// SpinLockDelay() generates an apprproate spin delay on iteration "loop" of a -// spin loop on location *w, whose previously observed value was "value". -// SpinLockDelay() may do nothing, may yield the CPU, may sleep a clock tick, -// or may wait for a delay that can be truncated by a call to SpinlockWake(w). -// In all cases, it must return in bounded time even if SpinlockWake() is not -// called. - -#include "base/spinlock_internal.h" - -// forward declaration for use by spinlock_*-inl.h -namespace base { namespace internal { static int SuggestedDelayNS(int loop); }} - -#if defined(_WIN32) -#include "base/spinlock_win32-inl.h" -#elif defined(__linux__) -#include "base/spinlock_linux-inl.h" -#else -#include "base/spinlock_posix-inl.h" -#endif - -namespace base { -namespace internal { - -// Return a suggested delay in nanoseconds for iteration number "loop" -static int SuggestedDelayNS(int loop) { - // Weak pseudo-random number generator to get some spread between threads - // when many are spinning. -#ifdef BASE_HAS_ATOMIC64 - static base::subtle::Atomic64 rand; - uint64 r = base::subtle::NoBarrier_Load(&rand); - r = 0x5deece66dLL * r + 0xb; // numbers from nrand48() - base::subtle::NoBarrier_Store(&rand, r); - - r <<= 16; // 48-bit random number now in top 48-bits. - if (loop < 0 || loop > 32) { // limit loop to 0..32 - loop = 32; - } - // loop>>3 cannot exceed 4 because loop cannot exceed 32. - // Select top 20..24 bits of lower 48 bits, - // giving approximately 0ms to 16ms. - // Mean is exponential in loop for first 32 iterations, then 8ms. - // The futex path multiplies this by 16, since we expect explicit wakeups - // almost always on that path. - return r >> (44 - (loop >> 3)); -#else - static Atomic32 rand; - uint32 r = base::subtle::NoBarrier_Load(&rand); - r = 0x343fd * r + 0x269ec3; // numbers from MSVC++ - base::subtle::NoBarrier_Store(&rand, r); - - r <<= 1; // 31-bit random number now in top 31-bits. - if (loop < 0 || loop > 32) { // limit loop to 0..32 - loop = 32; - } - // loop>>3 cannot exceed 4 because loop cannot exceed 32. - // Select top 20..24 bits of lower 31 bits, - // giving approximately 0ms to 16ms. - // Mean is exponential in loop for first 32 iterations, then 8ms. - // The futex path multiplies this by 16, since we expect explicit wakeups - // almost always on that path. - return r >> (12 - (loop >> 3)); -#endif -} - -} // namespace internal -} // namespace base diff --git a/contrib/libtcmalloc/src/base/spinlock_internal.h b/contrib/libtcmalloc/src/base/spinlock_internal.h deleted file mode 100644 index 636885cd6e5..00000000000 --- a/contrib/libtcmalloc/src/base/spinlock_internal.h +++ /dev/null @@ -1,51 +0,0 @@ -// -*- Mode: C++; c-basic-offset: 2; indent-tabs-mode: nil -*- -/* Copyright (c) 2010, Google Inc. - * All rights reserved. - * - * Redistribution and use in source and binary forms, with or without - * modification, are permitted provided that the following conditions are - * met: - * - * * Redistributions of source code must retain the above copyright - * notice, this list of conditions and the following disclaimer. - * * Redistributions in binary form must reproduce the above - * copyright notice, this list of conditions and the following disclaimer - * in the documentation and/or other materials provided with the - * distribution. - * * Neither the name of Google Inc. nor the names of its - * contributors may be used to endorse or promote products derived from - * this software without specific prior written permission. - * - * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS - * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT - * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR - * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT - * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, - * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT - * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, - * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY - * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT - * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE - * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. - * - * --- - * This file is an internal part spinlock.cc and once.cc - * It may not be used directly by code outside of //base. - */ - -#ifndef BASE_SPINLOCK_INTERNAL_H_ -#define BASE_SPINLOCK_INTERNAL_H_ - -#include "../config.h" -#include "base/basictypes.h" -#include "base/atomicops.h" - -namespace base { -namespace internal { - -void SpinLockWake(volatile Atomic32 *w, bool all); -void SpinLockDelay(volatile Atomic32 *w, int32 value, int loop); - -} // namespace internal -} // namespace base -#endif diff --git a/contrib/libtcmalloc/src/base/spinlock_linux-inl.h b/contrib/libtcmalloc/src/base/spinlock_linux-inl.h deleted file mode 100644 index aadf62a4b67..00000000000 --- a/contrib/libtcmalloc/src/base/spinlock_linux-inl.h +++ /dev/null @@ -1,101 +0,0 @@ -// -*- Mode: C++; c-basic-offset: 2; indent-tabs-mode: nil -*- -/* Copyright (c) 2009, Google Inc. - * All rights reserved. - * - * Redistribution and use in source and binary forms, with or without - * modification, are permitted provided that the following conditions are - * met: - * - * * Redistributions of source code must retain the above copyright - * notice, this list of conditions and the following disclaimer. - * * Redistributions in binary form must reproduce the above - * copyright notice, this list of conditions and the following disclaimer - * in the documentation and/or other materials provided with the - * distribution. - * * Neither the name of Google Inc. nor the names of its - * contributors may be used to endorse or promote products derived from - * this software without specific prior written permission. - * - * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS - * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT - * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR - * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT - * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, - * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT - * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, - * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY - * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT - * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE - * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. - * - * --- - * This file is a Linux-specific part of spinlock_internal.cc - */ - -#include -#include -#include -#include -#include "base/linux_syscall_support.h" - -#define FUTEX_WAIT 0 -#define FUTEX_WAKE 1 -#define FUTEX_PRIVATE_FLAG 128 - -static bool have_futex; -static int futex_private_flag = FUTEX_PRIVATE_FLAG; - -namespace { -static struct InitModule { - InitModule() { - int x = 0; - // futexes are ints, so we can use them only when - // that's the same size as the lockword_ in SpinLock. - have_futex = (sizeof (Atomic32) == sizeof (int) && - sys_futex(&x, FUTEX_WAKE, 1, NULL, NULL, 0) >= 0); - if (have_futex && - sys_futex(&x, FUTEX_WAKE | futex_private_flag, 1, NULL, NULL, 0) < 0) { - futex_private_flag = 0; - } - } -} init_module; - -} // anonymous namespace - - -namespace base { -namespace internal { - -void SpinLockDelay(volatile Atomic32 *w, int32 value, int loop) { - if (loop != 0) { - int save_errno = errno; - struct timespec tm; - tm.tv_sec = 0; - if (have_futex) { - tm.tv_nsec = base::internal::SuggestedDelayNS(loop); - } else { - tm.tv_nsec = 2000001; // above 2ms so linux 2.4 doesn't spin - } - if (have_futex) { - tm.tv_nsec *= 16; // increase the delay; we expect explicit wakeups - sys_futex(reinterpret_cast(const_cast(w)), - FUTEX_WAIT | futex_private_flag, - value, reinterpret_cast(&tm), - NULL, 0); - } else { - nanosleep(&tm, NULL); - } - errno = save_errno; - } -} - -void SpinLockWake(volatile Atomic32 *w, bool all) { - if (have_futex) { - sys_futex(reinterpret_cast(const_cast(w)), - FUTEX_WAKE | futex_private_flag, all? INT_MAX : 1, - NULL, NULL, 0); - } -} - -} // namespace internal -} // namespace base diff --git a/contrib/libtcmalloc/src/base/spinlock_posix-inl.h b/contrib/libtcmalloc/src/base/spinlock_posix-inl.h deleted file mode 100644 index 2695b7b1bb9..00000000000 --- a/contrib/libtcmalloc/src/base/spinlock_posix-inl.h +++ /dev/null @@ -1,63 +0,0 @@ -// -*- Mode: C++; c-basic-offset: 2; indent-tabs-mode: nil -*- -/* Copyright (c) 2009, Google Inc. - * All rights reserved. - * - * Redistribution and use in source and binary forms, with or without - * modification, are permitted provided that the following conditions are - * met: - * - * * Redistributions of source code must retain the above copyright - * notice, this list of conditions and the following disclaimer. - * * Redistributions in binary form must reproduce the above - * copyright notice, this list of conditions and the following disclaimer - * in the documentation and/or other materials provided with the - * distribution. - * * Neither the name of Google Inc. nor the names of its - * contributors may be used to endorse or promote products derived from - * this software without specific prior written permission. - * - * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS - * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT - * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR - * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT - * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, - * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT - * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, - * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY - * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT - * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE - * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. - * - * --- - * This file is a Posix-specific part of spinlock_internal.cc - */ - -#include "../config.h" -#include -#ifdef HAVE_SCHED_H -#include /* For sched_yield() */ -#endif -#include /* For nanosleep() */ - -namespace base { -namespace internal { - -void SpinLockDelay(volatile Atomic32 *w, int32 value, int loop) { - int save_errno = errno; - if (loop == 0) { - } else if (loop == 1) { - sched_yield(); - } else { - struct timespec tm; - tm.tv_sec = 0; - tm.tv_nsec = base::internal::SuggestedDelayNS(loop); - nanosleep(&tm, NULL); - } - errno = save_errno; -} - -void SpinLockWake(volatile Atomic32 *w, bool all) { -} - -} // namespace internal -} // namespace base diff --git a/contrib/libtcmalloc/src/base/spinlock_win32-inl.h b/contrib/libtcmalloc/src/base/spinlock_win32-inl.h deleted file mode 100644 index 956b9653e6d..00000000000 --- a/contrib/libtcmalloc/src/base/spinlock_win32-inl.h +++ /dev/null @@ -1,54 +0,0 @@ -// -*- Mode: C++; c-basic-offset: 2; indent-tabs-mode: nil -*- -/* Copyright (c) 2009, Google Inc. - * All rights reserved. - * - * Redistribution and use in source and binary forms, with or without - * modification, are permitted provided that the following conditions are - * met: - * - * * Redistributions of source code must retain the above copyright - * notice, this list of conditions and the following disclaimer. - * * Redistributions in binary form must reproduce the above - * copyright notice, this list of conditions and the following disclaimer - * in the documentation and/or other materials provided with the - * distribution. - * * Neither the name of Google Inc. nor the names of its - * contributors may be used to endorse or promote products derived from - * this software without specific prior written permission. - * - * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS - * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT - * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR - * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT - * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, - * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT - * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, - * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY - * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT - * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE - * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. - * - * --- - * This file is a Win32-specific part of spinlock_internal.cc - */ - - -#include - -namespace base { -namespace internal { - -void SpinLockDelay(volatile Atomic32 *w, int32 value, int loop) { - if (loop == 0) { - } else if (loop == 1) { - Sleep(0); - } else { - Sleep(base::internal::SuggestedDelayNS(loop) / 1000000); - } -} - -void SpinLockWake(volatile Atomic32 *w, bool all) { -} - -} // namespace internal -} // namespace base diff --git a/contrib/libtcmalloc/src/base/stl_allocator.h b/contrib/libtcmalloc/src/base/stl_allocator.h deleted file mode 100644 index 4520713622f..00000000000 --- a/contrib/libtcmalloc/src/base/stl_allocator.h +++ /dev/null @@ -1,98 +0,0 @@ -// -*- Mode: C++; c-basic-offset: 2; indent-tabs-mode: nil -*- -/* Copyright (c) 2006, Google Inc. - * All rights reserved. - * - * Redistribution and use in source and binary forms, with or without - * modification, are permitted provided that the following conditions are - * met: - * - * * Redistributions of source code must retain the above copyright - * notice, this list of conditions and the following disclaimer. - * * Redistributions in binary form must reproduce the above - * copyright notice, this list of conditions and the following disclaimer - * in the documentation and/or other materials provided with the - * distribution. - * * Neither the name of Google Inc. nor the names of its - * contributors may be used to endorse or promote products derived from - * this software without specific prior written permission. - * - * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS - * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT - * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR - * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT - * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, - * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT - * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, - * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY - * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT - * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE - * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. - * - * --- - * Author: Maxim Lifantsev - */ - - -#ifndef BASE_STL_ALLOCATOR_H_ -#define BASE_STL_ALLOCATOR_H_ - -#include "../config.h" - -#include // for ptrdiff_t -#include - -#include "base/logging.h" - -// Generic allocator class for STL objects -// that uses a given type-less allocator Alloc, which must provide: -// static void* Alloc::Allocate(size_t size); -// static void Alloc::Free(void* ptr, size_t size); -// -// STL_Allocator provides the same thread-safety -// guarantees as MyAlloc. -// -// Usage example: -// set, STL_Allocator > my_set; -// CAVEAT: Parts of the code below are probably specific -// to the STL version(s) we are using. -// The code is simply lifted from what std::allocator<> provides. -template -class STL_Allocator { - public: - typedef size_t size_type; - typedef ptrdiff_t difference_type; - typedef T* pointer; - typedef const T* const_pointer; - typedef T& reference; - typedef const T& const_reference; - typedef T value_type; - - template struct rebind { - typedef STL_Allocator other; - }; - - STL_Allocator() { } - STL_Allocator(const STL_Allocator&) { } - template STL_Allocator(const STL_Allocator&) { } - ~STL_Allocator() { } - - pointer address(reference x) const { return &x; } - const_pointer address(const_reference x) const { return &x; } - - pointer allocate(size_type n, const void* = 0) { - RAW_DCHECK((n * sizeof(T)) / sizeof(T) == n, "n is too big to allocate"); - return static_cast(Alloc::Allocate(n * sizeof(T))); - } - void deallocate(pointer p, size_type n) { Alloc::Free(p, n * sizeof(T)); } - - size_type max_size() const { return size_t(-1) / sizeof(T); } - - void construct(pointer p, const T& val) { ::new(p) T(val); } - void construct(pointer p) { ::new(p) T(); } - void destroy(pointer p) { p->~T(); } - - // There's no state, so these allocators are always equal - bool operator==(const STL_Allocator&) const { return true; } -}; - -#endif // BASE_STL_ALLOCATOR_H_ diff --git a/contrib/libtcmalloc/src/base/sysinfo.cc b/contrib/libtcmalloc/src/base/sysinfo.cc deleted file mode 100644 index 75217e6795a..00000000000 --- a/contrib/libtcmalloc/src/base/sysinfo.cc +++ /dev/null @@ -1,860 +0,0 @@ -// -*- Mode: C++; c-basic-offset: 2; indent-tabs-mode: nil -*- -// Copyright (c) 2006, Google Inc. -// All rights reserved. -// -// Redistribution and use in source and binary forms, with or without -// modification, are permitted provided that the following conditions are -// met: -// -// * Redistributions of source code must retain the above copyright -// notice, this list of conditions and the following disclaimer. -// * Redistributions in binary form must reproduce the above -// copyright notice, this list of conditions and the following disclaimer -// in the documentation and/or other materials provided with the -// distribution. -// * Neither the name of Google Inc. nor the names of its -// contributors may be used to endorse or promote products derived from -// this software without specific prior written permission. -// -// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS -// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT -// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR -// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT -// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, -// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT -// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, -// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY -// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT -// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE -// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. - -#include "../config.h" -#if (defined(_WIN32) || defined(__MINGW32__)) && !defined(__CYGWIN__) && !defined(__CYGWIN32) -# define PLATFORM_WINDOWS 1 -#endif - -#include // for isspace() -#include // for getenv() -#include // for snprintf(), sscanf() -#include // for memmove(), memchr(), etc. -#include // for open() -#include // for errno -#ifdef HAVE_UNISTD_H -#include // for read() -#endif -#if defined __MACH__ // Mac OS X, almost certainly -#include // for iterating over dll's in ProcMapsIter -#include // for iterating over dll's in ProcMapsIter -#include -#include // how we figure out numcpu's on OS X -#elif defined __FreeBSD__ -#include -#elif defined __sun__ // Solaris -#include // for, e.g., prmap_t -#elif defined(PLATFORM_WINDOWS) -#include // for getpid() (actually, _getpid()) -#include // for SHGetValueA() -#include // for Module32First() -#endif -#include "base/sysinfo.h" -#include "base/commandlineflags.h" -#include "base/dynamic_annotations.h" // for RunningOnValgrind -#include "base/logging.h" - -#ifdef PLATFORM_WINDOWS -#ifdef MODULEENTRY32 -// In a change from the usual W-A pattern, there is no A variant of -// MODULEENTRY32. Tlhelp32.h #defines the W variant, but not the A. -// In unicode mode, tlhelp32.h #defines MODULEENTRY32 to be -// MODULEENTRY32W. These #undefs are the only way I see to get back -// access to the original, ascii struct (and related functions). -#undef MODULEENTRY32 -#undef Module32First -#undef Module32Next -#undef PMODULEENTRY32 -#undef LPMODULEENTRY32 -#endif /* MODULEENTRY32 */ -// MinGW doesn't seem to define this, perhaps some windowsen don't either. -#ifndef TH32CS_SNAPMODULE32 -#define TH32CS_SNAPMODULE32 0 -#endif /* TH32CS_SNAPMODULE32 */ -#endif /* PLATFORM_WINDOWS */ - -// Re-run fn until it doesn't cause EINTR. -#define NO_INTR(fn) do {} while ((fn) < 0 && errno == EINTR) - -// open/read/close can set errno, which may be illegal at this -// time, so prefer making the syscalls directly if we can. -#ifdef HAVE_SYS_SYSCALL_H -# include -#endif -#ifdef SYS_open // solaris 11, at least sometimes, only defines SYS_openat -# define safeopen(filename, mode) syscall(SYS_open, filename, mode) -#else -# define safeopen(filename, mode) open(filename, mode) -#endif -#ifdef SYS_read -# define saferead(fd, buffer, size) syscall(SYS_read, fd, buffer, size) -#else -# define saferead(fd, buffer, size) read(fd, buffer, size) -#endif -#ifdef SYS_close -# define safeclose(fd) syscall(SYS_close, fd) -#else -# define safeclose(fd) close(fd) -#endif - -// ---------------------------------------------------------------------- -// GetenvBeforeMain() -// GetUniquePathFromEnv() -// Some non-trivial getenv-related functions. -// ---------------------------------------------------------------------- - -// It's not safe to call getenv() in the malloc hooks, because they -// might be called extremely early, before libc is done setting up -// correctly. In particular, the thread library may not be done -// setting up errno. So instead, we use the built-in __environ array -// if it exists, and otherwise read /proc/self/environ directly, using -// system calls to read the file, and thus avoid setting errno. -// /proc/self/environ has a limit of how much data it exports (around -// 8K), so it's not an ideal solution. -const char* GetenvBeforeMain(const char* name) { -#if defined(HAVE___ENVIRON) // if we have it, it's declared in unistd.h - if (__environ) { // can exist but be NULL, if statically linked - const int namelen = strlen(name); - for (char** p = __environ; *p; p++) { - if (strlen(*p) < namelen) { - continue; - } - if (!memcmp(*p, name, namelen) && (*p)[namelen] == '=') // it's a match - return *p + namelen+1; // point after = - } - return NULL; - } -#endif -#if defined(PLATFORM_WINDOWS) - // TODO(mbelshe) - repeated calls to this function will overwrite the - // contents of the static buffer. - static char envvar_buf[1024]; // enough to hold any envvar we care about - if (!GetEnvironmentVariableA(name, envvar_buf, sizeof(envvar_buf)-1)) - return NULL; - return envvar_buf; -#endif - // static is ok because this function should only be called before - // main(), when we're single-threaded. - static char envbuf[16<<10]; - if (*envbuf == '\0') { // haven't read the environ yet - int fd = safeopen("/proc/self/environ", O_RDONLY); - // The -2 below guarantees the last two bytes of the buffer will be \0\0 - if (fd == -1 || // unable to open the file, fall back onto libc - saferead(fd, envbuf, sizeof(envbuf) - 2) < 0) { // error reading file - RAW_VLOG(1, "Unable to open /proc/self/environ, falling back " - "on getenv(\"%s\"), which may not work", name); - if (fd != -1) safeclose(fd); - return getenv(name); - } - safeclose(fd); - } - const int namelen = strlen(name); - const char* p = envbuf; - while (*p != '\0') { // will happen at the \0\0 that terminates the buffer - // proc file has the format NAME=value\0NAME=value\0NAME=value\0... - const char* endp = (char*)memchr(p, '\0', sizeof(envbuf) - (p - envbuf)); - if (endp == NULL) // this entry isn't NUL terminated - return NULL; - else if (!memcmp(p, name, namelen) && p[namelen] == '=') // it's a match - return p + namelen+1; // point after = - p = endp + 1; - } - return NULL; // env var never found -} - -extern "C" { - const char* TCMallocGetenvSafe(const char* name) { - return GetenvBeforeMain(name); - } -} - -// This takes as an argument an environment-variable name (like -// CPUPROFILE) whose value is supposed to be a file-path, and sets -// path to that path, and returns true. If the env var doesn't exist, -// or is the empty string, leave path unchanged and returns false. -// The reason this is non-trivial is that this function handles munged -// pathnames. Here's why: -// -// If we're a child process of the 'main' process, we can't just use -// getenv("CPUPROFILE") -- the parent process will be using that path. -// Instead we append our pid to the pathname. How do we tell if we're a -// child process? Ideally we'd set an environment variable that all -// our children would inherit. But -- and this is seemingly a bug in -// gcc -- if you do a setenv() in a shared libarary in a global -// constructor, the environment setting is lost by the time main() is -// called. The only safe thing we can do in such a situation is to -// modify the existing envvar. So we do a hack: in the parent, we set -// the high bit of the 1st char of CPUPROFILE. In the child, we -// notice the high bit is set and append the pid(). This works -// assuming cpuprofile filenames don't normally have the high bit set -// in their first character! If that assumption is violated, we'll -// still get a profile, but one with an unexpected name. -// TODO(csilvers): set an envvar instead when we can do it reliably. -bool GetUniquePathFromEnv(const char* env_name, char* path) { - char* envval = getenv(env_name); - if (envval == NULL || *envval == '\0') - return false; - if (envval[0] & 128) { // high bit is set - snprintf(path, PATH_MAX, "%c%s_%u", // add pid and clear high bit - envval[0] & 127, envval+1, (unsigned int)(getpid())); - } else { - snprintf(path, PATH_MAX, "%s", envval); - envval[0] |= 128; // set high bit for kids to see - } - return true; -} - -void SleepForMilliseconds(int milliseconds) { -#ifdef PLATFORM_WINDOWS - _sleep(milliseconds); // Windows's _sleep takes milliseconds argument -#else - // Sleep for a few milliseconds - struct timespec sleep_time; - sleep_time.tv_sec = milliseconds / 1000; - sleep_time.tv_nsec = (milliseconds % 1000) * 1000000; - while (nanosleep(&sleep_time, &sleep_time) != 0 && errno == EINTR) - ; // Ignore signals and wait for the full interval to elapse. -#endif -} - -int GetSystemCPUsCount() -{ -#if defined(PLATFORM_WINDOWS) - // Get the number of processors. - SYSTEM_INFO info; - GetSystemInfo(&info); - return info.dwNumberOfProcessors; -#else - long rv = sysconf(_SC_NPROCESSORS_ONLN); - if (rv < 0) { - return 1; - } - return static_cast(rv); -#endif -} - -// ---------------------------------------------------------------------- - -#if defined __linux__ || defined __FreeBSD__ || defined __sun__ || defined __CYGWIN__ || defined __CYGWIN32__ -static void ConstructFilename(const char* spec, pid_t pid, - char* buf, int buf_size) { - CHECK_LT(snprintf(buf, buf_size, - spec, - static_cast(pid ? pid : getpid())), buf_size); -} -#endif - -// A templatized helper function instantiated for Mach (OS X) only. -// It can handle finding info for both 32 bits and 64 bits. -// Returns true if it successfully handled the hdr, false else. -#ifdef __MACH__ // Mac OS X, almost certainly -template -static bool NextExtMachHelper(const mach_header* hdr, - int current_image, int current_load_cmd, - uint64 *start, uint64 *end, char **flags, - uint64 *offset, int64 *inode, char **filename, - uint64 *file_mapping, uint64 *file_pages, - uint64 *anon_mapping, uint64 *anon_pages, - dev_t *dev) { - static char kDefaultPerms[5] = "r-xp"; - if (hdr->magic != kMagic) - return false; - const char* lc = (const char *)hdr + sizeof(MachHeader); - // TODO(csilvers): make this not-quadradic (increment and hold state) - for (int j = 0; j < current_load_cmd; j++) // advance to *our* load_cmd - lc += ((const load_command *)lc)->cmdsize; - if (((const load_command *)lc)->cmd == kLCSegment) { - const intptr_t dlloff = _dyld_get_image_vmaddr_slide(current_image); - const SegmentCommand* sc = (const SegmentCommand *)lc; - if (start) *start = sc->vmaddr + dlloff; - if (end) *end = sc->vmaddr + sc->vmsize + dlloff; - if (flags) *flags = kDefaultPerms; // can we do better? - if (offset) *offset = sc->fileoff; - if (inode) *inode = 0; - if (filename) - *filename = const_cast(_dyld_get_image_name(current_image)); - if (file_mapping) *file_mapping = 0; - if (file_pages) *file_pages = 0; // could we use sc->filesize? - if (anon_mapping) *anon_mapping = 0; - if (anon_pages) *anon_pages = 0; - if (dev) *dev = 0; - return true; - } - - return false; -} -#endif - -// Finds |c| in |text|, and assign '\0' at the found position. -// The original character at the modified position should be |c|. -// A pointer to the modified position is stored in |endptr|. -// |endptr| should not be NULL. -static bool ExtractUntilChar(char *text, int c, char **endptr) { - CHECK_NE(text, NULL); - CHECK_NE(endptr, NULL); - char *found; - found = strchr(text, c); - if (found == NULL) { - *endptr = NULL; - return false; - } - - *endptr = found; - *found = '\0'; - return true; -} - -// Increments |*text_pointer| while it points a whitespace character. -// It is to follow sscanf's whilespace handling. -static void SkipWhileWhitespace(char **text_pointer, int c) { - if (isspace(c)) { - while (isspace(**text_pointer) && isspace(*((*text_pointer) + 1))) { - ++(*text_pointer); - } - } -} - -template -static T StringToInteger(char *text, char **endptr, int base) { - assert(false); - return T(); -} - -template<> -int StringToInteger(char *text, char **endptr, int base) { - return strtol(text, endptr, base); -} - -template<> -int64 StringToInteger(char *text, char **endptr, int base) { - return strtoll(text, endptr, base); -} - -template<> -uint64 StringToInteger(char *text, char **endptr, int base) { - return strtoull(text, endptr, base); -} - -template -static T StringToIntegerUntilChar( - char *text, int base, int c, char **endptr_result) { - CHECK_NE(endptr_result, NULL); - *endptr_result = NULL; - - char *endptr_extract; - if (!ExtractUntilChar(text, c, &endptr_extract)) - return 0; - - T result; - char *endptr_strto; - result = StringToInteger(text, &endptr_strto, base); - *endptr_extract = c; - - if (endptr_extract != endptr_strto) - return 0; - - *endptr_result = endptr_extract; - SkipWhileWhitespace(endptr_result, c); - - return result; -} - -static char *CopyStringUntilChar( - char *text, unsigned out_len, int c, char *out) { - char *endptr; - if (!ExtractUntilChar(text, c, &endptr)) - return NULL; - - strncpy(out, text, out_len); - out[out_len-1] = '\0'; - *endptr = c; - - SkipWhileWhitespace(&endptr, c); - return endptr; -} - -template -static bool StringToIntegerUntilCharWithCheck( - T *outptr, char *text, int base, int c, char **endptr) { - *outptr = StringToIntegerUntilChar(*endptr, base, c, endptr); - if (*endptr == NULL || **endptr == '\0') return false; - ++(*endptr); - return true; -} - -static bool ParseProcMapsLine(char *text, uint64 *start, uint64 *end, - char *flags, uint64 *offset, - int *major, int *minor, int64 *inode, - unsigned *filename_offset) { -#if defined(__linux__) - /* - * It's similar to: - * sscanf(text, "%"SCNx64"-%"SCNx64" %4s %"SCNx64" %x:%x %"SCNd64" %n", - * start, end, flags, offset, major, minor, inode, filename_offset) - */ - char *endptr = text; - if (endptr == NULL || *endptr == '\0') return false; - - if (!StringToIntegerUntilCharWithCheck(start, endptr, 16, '-', &endptr)) - return false; - - if (!StringToIntegerUntilCharWithCheck(end, endptr, 16, ' ', &endptr)) - return false; - - endptr = CopyStringUntilChar(endptr, 5, ' ', flags); - if (endptr == NULL || *endptr == '\0') return false; - ++endptr; - - if (!StringToIntegerUntilCharWithCheck(offset, endptr, 16, ' ', &endptr)) - return false; - - if (!StringToIntegerUntilCharWithCheck(major, endptr, 16, ':', &endptr)) - return false; - - if (!StringToIntegerUntilCharWithCheck(minor, endptr, 16, ' ', &endptr)) - return false; - - if (!StringToIntegerUntilCharWithCheck(inode, endptr, 10, ' ', &endptr)) - return false; - - *filename_offset = (endptr - text); - return true; -#else - return false; -#endif -} - -ProcMapsIterator::ProcMapsIterator(pid_t pid) { - Init(pid, NULL, false); -} - -ProcMapsIterator::ProcMapsIterator(pid_t pid, Buffer *buffer) { - Init(pid, buffer, false); -} - -ProcMapsIterator::ProcMapsIterator(pid_t pid, Buffer *buffer, - bool use_maps_backing) { - Init(pid, buffer, use_maps_backing); -} - -void ProcMapsIterator::Init(pid_t pid, Buffer *buffer, - bool use_maps_backing) { - pid_ = pid; - using_maps_backing_ = use_maps_backing; - dynamic_buffer_ = NULL; - if (!buffer) { - // If the user didn't pass in any buffer storage, allocate it - // now. This is the normal case; the signal handler passes in a - // static buffer. - buffer = dynamic_buffer_ = new Buffer; - } else { - dynamic_buffer_ = NULL; - } - - ibuf_ = buffer->buf_; - - stext_ = etext_ = nextline_ = ibuf_; - ebuf_ = ibuf_ + Buffer::kBufSize - 1; - nextline_ = ibuf_; - -#if defined(__linux__) || defined(__CYGWIN__) || defined(__CYGWIN32__) - if (use_maps_backing) { // don't bother with clever "self" stuff in this case - ConstructFilename("/proc/%d/maps_backing", pid, ibuf_, Buffer::kBufSize); - } else if (pid == 0) { - // We have to kludge a bit to deal with the args ConstructFilename - // expects. The 1 is never used -- it's only impt. that it's not 0. - ConstructFilename("/proc/self/maps", 1, ibuf_, Buffer::kBufSize); - } else { - ConstructFilename("/proc/%d/maps", pid, ibuf_, Buffer::kBufSize); - } - // No error logging since this can be called from the crash dump - // handler at awkward moments. Users should call Valid() before - // using. - NO_INTR(fd_ = open(ibuf_, O_RDONLY)); -#elif defined(__FreeBSD__) - // We don't support maps_backing on freebsd - if (pid == 0) { - ConstructFilename("/proc/curproc/map", 1, ibuf_, Buffer::kBufSize); - } else { - ConstructFilename("/proc/%d/map", pid, ibuf_, Buffer::kBufSize); - } - NO_INTR(fd_ = open(ibuf_, O_RDONLY)); -#elif defined(__sun__) - if (pid == 0) { - ConstructFilename("/proc/self/map", 1, ibuf_, Buffer::kBufSize); - } else { - ConstructFilename("/proc/%d/map", pid, ibuf_, Buffer::kBufSize); - } - NO_INTR(fd_ = open(ibuf_, O_RDONLY)); -#elif defined(__MACH__) - current_image_ = _dyld_image_count(); // count down from the top - current_load_cmd_ = -1; -#elif defined(PLATFORM_WINDOWS) - snapshot_ = CreateToolhelp32Snapshot(TH32CS_SNAPMODULE | - TH32CS_SNAPMODULE32, - GetCurrentProcessId()); - memset(&module_, 0, sizeof(module_)); -#else - fd_ = -1; // so Valid() is always false -#endif - -} - -ProcMapsIterator::~ProcMapsIterator() { -#if defined(PLATFORM_WINDOWS) - if (snapshot_ != INVALID_HANDLE_VALUE) CloseHandle(snapshot_); -#elif defined(__MACH__) - // no cleanup necessary! -#else - if (fd_ >= 0) NO_INTR(close(fd_)); -#endif - delete dynamic_buffer_; -} - -bool ProcMapsIterator::Valid() const { -#if defined(PLATFORM_WINDOWS) - return snapshot_ != INVALID_HANDLE_VALUE; -#elif defined(__MACH__) - return 1; -#else - return fd_ != -1; -#endif -} - -bool ProcMapsIterator::Next(uint64 *start, uint64 *end, char **flags, - uint64 *offset, int64 *inode, char **filename) { - return NextExt(start, end, flags, offset, inode, filename, NULL, NULL, - NULL, NULL, NULL); -} - -// This has too many arguments. It should really be building -// a map object and returning it. The problem is that this is called -// when the memory allocator state is undefined, hence the arguments. -bool ProcMapsIterator::NextExt(uint64 *start, uint64 *end, char **flags, - uint64 *offset, int64 *inode, char **filename, - uint64 *file_mapping, uint64 *file_pages, - uint64 *anon_mapping, uint64 *anon_pages, - dev_t *dev) { - -#if defined(__linux__) || defined(__FreeBSD__) || defined(__CYGWIN__) || defined(__CYGWIN32__) - do { - // Advance to the start of the next line - stext_ = nextline_; - - // See if we have a complete line in the buffer already - nextline_ = static_cast(memchr (stext_, '\n', etext_ - stext_)); - if (!nextline_) { - // Shift/fill the buffer so we do have a line - int count = etext_ - stext_; - - // Move the current text to the start of the buffer - memmove(ibuf_, stext_, count); - stext_ = ibuf_; - etext_ = ibuf_ + count; - - int nread = 0; // fill up buffer with text - while (etext_ < ebuf_) { - NO_INTR(nread = read(fd_, etext_, ebuf_ - etext_)); - if (nread > 0) - etext_ += nread; - else - break; - } - - // Zero out remaining characters in buffer at EOF to avoid returning - // garbage from subsequent calls. - if (etext_ != ebuf_ && nread == 0) { - memset(etext_, 0, ebuf_ - etext_); - } - *etext_ = '\n'; // sentinel; safe because ibuf extends 1 char beyond ebuf - nextline_ = static_cast(memchr (stext_, '\n', etext_ + 1 - stext_)); - } - *nextline_ = 0; // turn newline into nul - nextline_ += ((nextline_ < etext_)? 1 : 0); // skip nul if not end of text - // stext_ now points at a nul-terminated line - uint64 tmpstart, tmpend, tmpoffset; - int64 tmpinode; - int major, minor; - unsigned filename_offset = 0; -#if defined(__linux__) - // for now, assume all linuxes have the same format - if (!ParseProcMapsLine( - stext_, - start ? start : &tmpstart, - end ? end : &tmpend, - flags_, - offset ? offset : &tmpoffset, - &major, &minor, - inode ? inode : &tmpinode, &filename_offset)) continue; -#elif defined(__CYGWIN__) || defined(__CYGWIN32__) - // cygwin is like linux, except the third field is the "entry point" - // rather than the offset (see format_process_maps at - // http://cygwin.com/cgi-bin/cvsweb.cgi/src/winsup/cygwin/fhandler_process.cc?rev=1.89&content-type=text/x-cvsweb-markup&cvsroot=src - // Offset is always be 0 on cygwin: cygwin implements an mmap - // by loading the whole file and then calling NtMapViewOfSection. - // Cygwin also seems to set its flags kinda randomly; use windows default. - char tmpflags[5]; - if (offset) - *offset = 0; - strcpy(flags_, "r-xp"); - if (sscanf(stext_, "%llx-%llx %4s %llx %x:%x %lld %n", - start ? start : &tmpstart, - end ? end : &tmpend, - tmpflags, - &tmpoffset, - &major, &minor, - inode ? inode : &tmpinode, &filename_offset) != 7) continue; -#elif defined(__FreeBSD__) - // For the format, see http://www.freebsd.org/cgi/cvsweb.cgi/src/sys/fs/procfs/procfs_map.c?rev=1.31&content-type=text/x-cvsweb-markup - tmpstart = tmpend = tmpoffset = 0; - tmpinode = 0; - major = minor = 0; // can't get this info in freebsd - if (inode) - *inode = 0; // nor this - if (offset) - *offset = 0; // seems like this should be in there, but maybe not - // start end resident privateresident obj(?) prot refcnt shadowcnt - // flags copy_on_write needs_copy type filename: - // 0x8048000 0x804a000 2 0 0xc104ce70 r-x 1 0 0x0 COW NC vnode /bin/cat - if (sscanf(stext_, "0x%" SCNx64 " 0x%" SCNx64 " %*d %*d %*p %3s %*d %*d 0x%*x %*s %*s %*s %n", - start ? start : &tmpstart, - end ? end : &tmpend, - flags_, - &filename_offset) != 3) continue; -#endif - - // Depending on the Linux kernel being used, there may or may not be a space - // after the inode if there is no filename. sscanf will in such situations - // nondeterministically either fill in filename_offset or not (the results - // differ on multiple calls in the same run even with identical arguments). - // We don't want to wander off somewhere beyond the end of the string. - size_t stext_length = strlen(stext_); - if (filename_offset == 0 || filename_offset > stext_length) - filename_offset = stext_length; - - // We found an entry - if (flags) *flags = flags_; - if (filename) *filename = stext_ + filename_offset; - if (dev) *dev = minor | (major << 8); - - if (using_maps_backing_) { - // Extract and parse physical page backing info. - char *backing_ptr = stext_ + filename_offset + - strlen(stext_+filename_offset); - - // find the second '(' - int paren_count = 0; - while (--backing_ptr > stext_) { - if (*backing_ptr == '(') { - ++paren_count; - if (paren_count >= 2) { - uint64 tmp_file_mapping; - uint64 tmp_file_pages; - uint64 tmp_anon_mapping; - uint64 tmp_anon_pages; - - sscanf(backing_ptr+1, "F %" SCNx64 " %" SCNd64 ") (A %" SCNx64 " %" SCNd64 ")", - file_mapping ? file_mapping : &tmp_file_mapping, - file_pages ? file_pages : &tmp_file_pages, - anon_mapping ? anon_mapping : &tmp_anon_mapping, - anon_pages ? anon_pages : &tmp_anon_pages); - // null terminate the file name (there is a space - // before the first (. - backing_ptr[-1] = 0; - break; - } - } - } - } - - return true; - } while (etext_ > ibuf_); -#elif defined(__sun__) - // This is based on MA_READ == 4, MA_WRITE == 2, MA_EXEC == 1 - static char kPerms[8][4] = { "---", "--x", "-w-", "-wx", - "r--", "r-x", "rw-", "rwx" }; - COMPILE_ASSERT(MA_READ == 4, solaris_ma_read_must_equal_4); - COMPILE_ASSERT(MA_WRITE == 2, solaris_ma_write_must_equal_2); - COMPILE_ASSERT(MA_EXEC == 1, solaris_ma_exec_must_equal_1); - Buffer object_path; - int nread = 0; // fill up buffer with text - NO_INTR(nread = read(fd_, ibuf_, sizeof(prmap_t))); - if (nread == sizeof(prmap_t)) { - long inode_from_mapname = 0; - prmap_t* mapinfo = reinterpret_cast(ibuf_); - // Best-effort attempt to get the inode from the filename. I think the - // two middle ints are major and minor device numbers, but I'm not sure. - sscanf(mapinfo->pr_mapname, "ufs.%*d.%*d.%ld", &inode_from_mapname); - - if (pid_ == 0) { - CHECK_LT(snprintf(object_path.buf_, Buffer::kBufSize, - "/proc/self/path/%s", mapinfo->pr_mapname), - Buffer::kBufSize); - } else { - CHECK_LT(snprintf(object_path.buf_, Buffer::kBufSize, - "/proc/%d/path/%s", - static_cast(pid_), mapinfo->pr_mapname), - Buffer::kBufSize); - } - ssize_t len = readlink(object_path.buf_, current_filename_, PATH_MAX); - CHECK_LT(len, PATH_MAX); - if (len < 0) - len = 0; - current_filename_[len] = '\0'; - - if (start) *start = mapinfo->pr_vaddr; - if (end) *end = mapinfo->pr_vaddr + mapinfo->pr_size; - if (flags) *flags = kPerms[mapinfo->pr_mflags & 7]; - if (offset) *offset = mapinfo->pr_offset; - if (inode) *inode = inode_from_mapname; - if (filename) *filename = current_filename_; - if (file_mapping) *file_mapping = 0; - if (file_pages) *file_pages = 0; - if (anon_mapping) *anon_mapping = 0; - if (anon_pages) *anon_pages = 0; - if (dev) *dev = 0; - return true; - } -#elif defined(__MACH__) - // We return a separate entry for each segment in the DLL. (TODO(csilvers): - // can we do better?) A DLL ("image") has load-commands, some of which - // talk about segment boundaries. - // cf image_for_address from http://svn.digium.com/view/asterisk/team/oej/minivoicemail/dlfcn.c?revision=53912 - for (; current_image_ >= 0; current_image_--) { - const mach_header* hdr = _dyld_get_image_header(current_image_); - if (!hdr) continue; - if (current_load_cmd_ < 0) // set up for this image - current_load_cmd_ = hdr->ncmds; // again, go from the top down - - // We start with the next load command (we've already looked at this one). - for (current_load_cmd_--; current_load_cmd_ >= 0; current_load_cmd_--) { -#ifdef MH_MAGIC_64 - if (NextExtMachHelper( - hdr, current_image_, current_load_cmd_, - start, end, flags, offset, inode, filename, - file_mapping, file_pages, anon_mapping, - anon_pages, dev)) { - return true; - } -#endif - if (NextExtMachHelper( - hdr, current_image_, current_load_cmd_, - start, end, flags, offset, inode, filename, - file_mapping, file_pages, anon_mapping, - anon_pages, dev)) { - return true; - } - } - // If we get here, no more load_cmd's in this image talk about - // segments. Go on to the next image. - } -#elif defined(PLATFORM_WINDOWS) - static char kDefaultPerms[5] = "r-xp"; - BOOL ok; - if (module_.dwSize == 0) { // only possible before first call - module_.dwSize = sizeof(module_); - ok = Module32First(snapshot_, &module_); - } else { - ok = Module32Next(snapshot_, &module_); - } - if (ok) { - uint64 base_addr = reinterpret_cast(module_.modBaseAddr); - if (start) *start = base_addr; - if (end) *end = base_addr + module_.modBaseSize; - if (flags) *flags = kDefaultPerms; - if (offset) *offset = 0; - if (inode) *inode = 0; - if (filename) *filename = module_.szExePath; - if (file_mapping) *file_mapping = 0; - if (file_pages) *file_pages = 0; - if (anon_mapping) *anon_mapping = 0; - if (anon_pages) *anon_pages = 0; - if (dev) *dev = 0; - return true; - } -#endif - - // We didn't find anything - return false; -} - -int ProcMapsIterator::FormatLine(char* buffer, int bufsize, - uint64 start, uint64 end, const char *flags, - uint64 offset, int64 inode, - const char *filename, dev_t dev) { - // We assume 'flags' looks like 'rwxp' or 'rwx'. - char r = (flags && flags[0] == 'r') ? 'r' : '-'; - char w = (flags && flags[0] && flags[1] == 'w') ? 'w' : '-'; - char x = (flags && flags[0] && flags[1] && flags[2] == 'x') ? 'x' : '-'; - // p always seems set on linux, so we set the default to 'p', not '-' - char p = (flags && flags[0] && flags[1] && flags[2] && flags[3] != 'p') - ? '-' : 'p'; - - const int rc = snprintf(buffer, bufsize, - "%08" PRIx64 "-%08" PRIx64 " %c%c%c%c %08" PRIx64 " %02x:%02x %-11" PRId64 " %s\n", - start, end, r,w,x,p, offset, - static_cast(dev/256), static_cast(dev%256), - inode, filename); - return (rc < 0 || rc >= bufsize) ? 0 : rc; -} - -namespace tcmalloc { - -// Helper to add the list of mapped shared libraries to a profile. -// Fill formatted "/proc/self/maps" contents into buffer 'buf' of size 'size' -// and return the actual size occupied in 'buf'. We fill wrote_all to true -// if we successfully wrote all proc lines to buf, false else. -// We do not provision for 0-terminating 'buf'. -int FillProcSelfMaps(char buf[], int size, bool* wrote_all) { - ProcMapsIterator::Buffer iterbuf; - ProcMapsIterator it(0, &iterbuf); // 0 means "current pid" - - uint64 start, end, offset; - int64 inode; - char *flags, *filename; - int bytes_written = 0; - *wrote_all = true; - while (it.Next(&start, &end, &flags, &offset, &inode, &filename)) { - const int line_length = it.FormatLine(buf + bytes_written, - size - bytes_written, - start, end, flags, offset, - inode, filename, 0); - if (line_length == 0) - *wrote_all = false; // failed to write this line out - else - bytes_written += line_length; - - } - return bytes_written; -} - -// Dump the same data as FillProcSelfMaps reads to fd. -// It seems easier to repeat parts of FillProcSelfMaps here than to -// reuse it via a call. -void DumpProcSelfMaps(RawFD fd) { - ProcMapsIterator::Buffer iterbuf; - ProcMapsIterator it(0, &iterbuf); // 0 means "current pid" - - uint64 start, end, offset; - int64 inode; - char *flags, *filename; - ProcMapsIterator::Buffer linebuf; - while (it.Next(&start, &end, &flags, &offset, &inode, &filename)) { - int written = it.FormatLine(linebuf.buf_, sizeof(linebuf.buf_), - start, end, flags, offset, inode, filename, - 0); - RawWrite(fd, linebuf.buf_, written); - } -} - -} // namespace tcmalloc diff --git a/contrib/libtcmalloc/src/base/sysinfo.h b/contrib/libtcmalloc/src/base/sysinfo.h deleted file mode 100644 index 75b101376c5..00000000000 --- a/contrib/libtcmalloc/src/base/sysinfo.h +++ /dev/null @@ -1,232 +0,0 @@ -// -*- Mode: C++; c-basic-offset: 2; indent-tabs-mode: nil -*- -// Copyright (c) 2006, Google Inc. -// All rights reserved. -// -// Redistribution and use in source and binary forms, with or without -// modification, are permitted provided that the following conditions are -// met: -// -// * Redistributions of source code must retain the above copyright -// notice, this list of conditions and the following disclaimer. -// * Redistributions in binary form must reproduce the above -// copyright notice, this list of conditions and the following disclaimer -// in the documentation and/or other materials provided with the -// distribution. -// * Neither the name of Google Inc. nor the names of its -// contributors may be used to endorse or promote products derived from -// this software without specific prior written permission. -// -// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS -// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT -// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR -// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT -// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, -// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT -// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, -// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY -// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT -// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE -// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. - -// All functions here are thread-hostile due to file caching unless -// commented otherwise. - -#ifndef _SYSINFO_H_ -#define _SYSINFO_H_ - -#include "../config.h" - -#include -#if (defined(_WIN32) || defined(__MINGW32__)) && (!defined(__CYGWIN__) && !defined(__CYGWIN32__)) -#include // for DWORD -#include // for CreateToolhelp32Snapshot -#endif -#ifdef HAVE_UNISTD_H -#include // for pid_t -#endif -#include // for size_t -#include // for PATH_MAX -#include "base/basictypes.h" -#include "base/logging.h" // for RawFD - -// This getenv function is safe to call before the C runtime is initialized. -// On Windows, it utilizes GetEnvironmentVariable() and on unix it uses -// /proc/self/environ instead calling getenv(). It's intended to be used in -// routines that run before main(), when the state required for getenv() may -// not be set up yet. In particular, errno isn't set up until relatively late -// (after the pthreads library has a chance to make it threadsafe), and -// getenv() doesn't work until then. -// On some platforms, this call will utilize the same, static buffer for -// repeated GetenvBeforeMain() calls. Callers should not expect pointers from -// this routine to be long lived. -// Note that on unix, /proc only has the environment at the time the -// application was started, so this routine ignores setenv() calls/etc. Also -// note it only reads the first 16K of the environment. -extern const char* GetenvBeforeMain(const char* name); - -// This takes as an argument an environment-variable name (like -// CPUPROFILE) whose value is supposed to be a file-path, and sets -// path to that path, and returns true. Non-trivial for surprising -// reasons, as documented in sysinfo.cc. path must have space PATH_MAX. -extern bool GetUniquePathFromEnv(const char* env_name, char* path); - -extern int GetSystemCPUsCount(); - -void SleepForMilliseconds(int milliseconds); - -// Return true if we're running POSIX (e.g., NPTL on Linux) threads, -// as opposed to a non-POSIX thread library. The thing that we care -// about is whether a thread's pid is the same as the thread that -// spawned it. If so, this function returns true. -// Thread-safe. -// Note: We consider false negatives to be OK. -bool HasPosixThreads(); - -#ifndef SWIG // SWIG doesn't like struct Buffer and variable arguments. - -// A ProcMapsIterator abstracts access to /proc/maps for a given -// process. Needs to be stack-allocatable and avoid using stdio/malloc -// so it can be used in the google stack dumper, heap-profiler, etc. -// -// On Windows and Mac OS X, this iterator iterates *only* over DLLs -// mapped into this process space. For Linux, FreeBSD, and Solaris, -// it iterates over *all* mapped memory regions, including anonymous -// mmaps. For other O/Ss, it is unlikely to work at all, and Valid() -// will always return false. Also note: this routine only works on -// FreeBSD if procfs is mounted: make sure this is in your /etc/fstab: -// proc /proc procfs rw 0 0 -class ProcMapsIterator { - public: - struct Buffer { -#ifdef __FreeBSD__ - // FreeBSD requires us to read all of the maps file at once, so - // we have to make a buffer that's "always" big enough - static const size_t kBufSize = 102400; -#else // a one-line buffer is good enough - static const size_t kBufSize = PATH_MAX + 1024; -#endif - char buf_[kBufSize]; - }; - - - // Create a new iterator for the specified pid. pid can be 0 for "self". - explicit ProcMapsIterator(pid_t pid); - - // Create an iterator with specified storage (for use in signal - // handler). "buffer" should point to a ProcMapsIterator::Buffer - // buffer can be NULL in which case a bufer will be allocated. - ProcMapsIterator(pid_t pid, Buffer *buffer); - - // Iterate through maps_backing instead of maps if use_maps_backing - // is true. Otherwise the same as above. buffer can be NULL and - // it will allocate a buffer itself. - ProcMapsIterator(pid_t pid, Buffer *buffer, - bool use_maps_backing); - - // Returns true if the iterator successfully initialized; - bool Valid() const; - - // Returns a pointer to the most recently parsed line. Only valid - // after Next() returns true, and until the iterator is destroyed or - // Next() is called again. This may give strange results on non-Linux - // systems. Prefer FormatLine() if that may be a concern. - const char *CurrentLine() const { return stext_; } - - // Writes the "canonical" form of the /proc/xxx/maps info for a single - // line to the passed-in buffer. Returns the number of bytes written, - // or 0 if it was not able to write the complete line. (To guarantee - // success, buffer should have size at least Buffer::kBufSize.) - // Takes as arguments values set via a call to Next(). The - // "canonical" form of the line (taken from linux's /proc/xxx/maps): - // - + - // : Note: the - // eg - // 08048000-0804c000 r-xp 00000000 03:01 3793678 /bin/cat - // If you don't have the dev_t (dev), feel free to pass in 0. - // (Next() doesn't return a dev_t, though NextExt does.) - // - // Note: if filename and flags were obtained via a call to Next(), - // then the output of this function is only valid if Next() returned - // true, and only until the iterator is destroyed or Next() is - // called again. (Since filename, at least, points into CurrentLine.) - static int FormatLine(char* buffer, int bufsize, - uint64 start, uint64 end, const char *flags, - uint64 offset, int64 inode, const char *filename, - dev_t dev); - - // Find the next entry in /proc/maps; return true if found or false - // if at the end of the file. - // - // Any of the result pointers can be NULL if you're not interested - // in those values. - // - // If "flags" and "filename" are passed, they end up pointing to - // storage within the ProcMapsIterator that is valid only until the - // iterator is destroyed or Next() is called again. The caller may - // modify the contents of these strings (up as far as the first NUL, - // and only until the subsequent call to Next()) if desired. - - // The offsets are all uint64 in order to handle the case of a - // 32-bit process running on a 64-bit kernel - // - // IMPORTANT NOTE: see top-of-class notes for details about what - // mapped regions Next() iterates over, depending on O/S. - // TODO(csilvers): make flags and filename const. - bool Next(uint64 *start, uint64 *end, char **flags, - uint64 *offset, int64 *inode, char **filename); - - bool NextExt(uint64 *start, uint64 *end, char **flags, - uint64 *offset, int64 *inode, char **filename, - uint64 *file_mapping, uint64 *file_pages, - uint64 *anon_mapping, uint64 *anon_pages, - dev_t *dev); - - ~ProcMapsIterator(); - - private: - void Init(pid_t pid, Buffer *buffer, bool use_maps_backing); - - char *ibuf_; // input buffer - char *stext_; // start of text - char *etext_; // end of text - char *nextline_; // start of next line - char *ebuf_; // end of buffer (1 char for a nul) -#if (defined(_WIN32) || defined(__MINGW32__)) && (!defined(__CYGWIN__) && !defined(__CYGWIN32__)) - HANDLE snapshot_; // filehandle on dll info - // In a change from the usual W-A pattern, there is no A variant of - // MODULEENTRY32. Tlhelp32.h #defines the W variant, but not the A. - // We want the original A variants, and this #undef is the only - // way I see to get them. Redefining it when we're done prevents us - // from affecting other .cc files. -# ifdef MODULEENTRY32 // Alias of W -# undef MODULEENTRY32 - MODULEENTRY32 module_; // info about current dll (and dll iterator) -# define MODULEENTRY32 MODULEENTRY32W -# else // It's the ascii, the one we want. - MODULEENTRY32 module_; // info about current dll (and dll iterator) -# endif -#elif defined(__MACH__) - int current_image_; // dll's are called "images" in macos parlance - int current_load_cmd_; // the segment of this dll we're examining -#elif defined(__sun__) // Solaris - int fd_; - char current_filename_[PATH_MAX]; -#else - int fd_; // filehandle on /proc/*/maps -#endif - pid_t pid_; - char flags_[10]; - Buffer* dynamic_buffer_; // dynamically-allocated Buffer - bool using_maps_backing_; // true if we are looking at maps_backing instead of maps. -}; - -#endif /* #ifndef SWIG */ - -// Helper routines - -namespace tcmalloc { -int FillProcSelfMaps(char buf[], int size, bool* wrote_all); -void DumpProcSelfMaps(RawFD fd); -} - -#endif /* #ifndef _SYSINFO_H_ */ diff --git a/contrib/libtcmalloc/src/base/thread_annotations.h b/contrib/libtcmalloc/src/base/thread_annotations.h deleted file mode 100644 index f57b2999ee7..00000000000 --- a/contrib/libtcmalloc/src/base/thread_annotations.h +++ /dev/null @@ -1,134 +0,0 @@ -// Copyright (c) 2008, Google Inc. -// All rights reserved. -// -// Redistribution and use in source and binary forms, with or without -// modification, are permitted provided that the following conditions are -// met: -// -// * Redistributions of source code must retain the above copyright -// notice, this list of conditions and the following disclaimer. -// * Redistributions in binary form must reproduce the above -// copyright notice, this list of conditions and the following disclaimer -// in the documentation and/or other materials provided with the -// distribution. -// * Neither the name of Google Inc. nor the names of its -// contributors may be used to endorse or promote products derived from -// this software without specific prior written permission. -// -// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS -// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT -// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR -// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT -// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, -// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT -// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, -// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY -// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT -// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE -// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. - -// --- -// Author: Le-Chun Wu -// -// This header file contains the macro definitions for thread safety -// annotations that allow the developers to document the locking policies -// of their multi-threaded code. The annotations can also help program -// analysis tools to identify potential thread safety issues. -// -// The annotations are implemented using GCC's "attributes" extension. -// Using the macros defined here instead of the raw GCC attributes allows -// for portability and future compatibility. -// -// This functionality is not yet fully implemented in perftools, -// but may be one day. - -#ifndef BASE_THREAD_ANNOTATIONS_H_ -#define BASE_THREAD_ANNOTATIONS_H_ - - -#if defined(__GNUC__) \ - && (__GNUC__ > 4 || (__GNUC__ == 4 && __GNUC_MINOR__ >= 4)) \ - && defined(__SUPPORT_TS_ANNOTATION__) && (!defined(SWIG)) -#define THREAD_ANNOTATION_ATTRIBUTE__(x) __attribute__((x)) -#else -#define THREAD_ANNOTATION_ATTRIBUTE__(x) // no-op -#endif - - -// Document if a shared variable/field needs to be protected by a lock. -// GUARDED_BY allows the user to specify a particular lock that should be -// held when accessing the annotated variable, while GUARDED_VAR only -// indicates a shared variable should be guarded (by any lock). GUARDED_VAR -// is primarily used when the client cannot express the name of the lock. -#define GUARDED_BY(x) THREAD_ANNOTATION_ATTRIBUTE__(guarded_by(x)) -#define GUARDED_VAR THREAD_ANNOTATION_ATTRIBUTE__(guarded) - -// Document if the memory location pointed to by a pointer should be guarded -// by a lock when dereferencing the pointer. Similar to GUARDED_VAR, -// PT_GUARDED_VAR is primarily used when the client cannot express the name -// of the lock. Note that a pointer variable to a shared memory location -// could itself be a shared variable. For example, if a shared global pointer -// q, which is guarded by mu1, points to a shared memory location that is -// guarded by mu2, q should be annotated as follows: -// int *q GUARDED_BY(mu1) PT_GUARDED_BY(mu2); -#define PT_GUARDED_BY(x) \ - THREAD_ANNOTATION_ATTRIBUTE__(point_to_guarded_by(x)) -#define PT_GUARDED_VAR \ - THREAD_ANNOTATION_ATTRIBUTE__(point_to_guarded) - -// Document the acquisition order between locks that can be held -// simultaneously by a thread. For any two locks that need to be annotated -// to establish an acquisition order, only one of them needs the annotation. -// (i.e. You don't have to annotate both locks with both ACQUIRED_AFTER -// and ACQUIRED_BEFORE.) -#define ACQUIRED_AFTER(x) \ - THREAD_ANNOTATION_ATTRIBUTE__(acquired_after(x)) -#define ACQUIRED_BEFORE(x) \ - THREAD_ANNOTATION_ATTRIBUTE__(acquired_before(x)) - -// The following three annotations document the lock requirements for -// functions/methods. - -// Document if a function expects certain locks to be held before it is called -#define EXCLUSIVE_LOCKS_REQUIRED(x) \ - THREAD_ANNOTATION_ATTRIBUTE__(exclusive_locks_required(x)) - -#define SHARED_LOCKS_REQUIRED(x) \ - THREAD_ANNOTATION_ATTRIBUTE__(shared_locks_required(x)) - -// Document the locks acquired in the body of the function. These locks -// cannot be held when calling this function (as google3's Mutex locks are -// non-reentrant). -#define LOCKS_EXCLUDED(x) \ - THREAD_ANNOTATION_ATTRIBUTE__(locks_excluded(x)) - -// Document the lock the annotated function returns without acquiring it. -#define LOCK_RETURNED(x) THREAD_ANNOTATION_ATTRIBUTE__(lock_returned(x)) - -// Document if a class/type is a lockable type (such as the Mutex class). -#define LOCKABLE THREAD_ANNOTATION_ATTRIBUTE__(lockable) - -// Document if a class is a scoped lockable type (such as the MutexLock class). -#define SCOPED_LOCKABLE THREAD_ANNOTATION_ATTRIBUTE__(scoped_lockable) - -// The following annotations specify lock and unlock primitives. -#define EXCLUSIVE_LOCK_FUNCTION(x) \ - THREAD_ANNOTATION_ATTRIBUTE__(exclusive_lock(x)) - -#define SHARED_LOCK_FUNCTION(x) \ - THREAD_ANNOTATION_ATTRIBUTE__(shared_lock(x)) - -#define EXCLUSIVE_TRYLOCK_FUNCTION(x) \ - THREAD_ANNOTATION_ATTRIBUTE__(exclusive_trylock(x)) - -#define SHARED_TRYLOCK_FUNCTION(x) \ - THREAD_ANNOTATION_ATTRIBUTE__(shared_trylock(x)) - -#define UNLOCK_FUNCTION(x) \ - THREAD_ANNOTATION_ATTRIBUTE__(unlock(x)) - -// An escape hatch for thread safety analysis to ignore the annotated function. -#define NO_THREAD_SAFETY_ANALYSIS \ - THREAD_ANNOTATION_ATTRIBUTE__(no_thread_safety_analysis) - -#endif // BASE_THREAD_ANNOTATIONS_H_ diff --git a/contrib/libtcmalloc/src/base/thread_lister.c b/contrib/libtcmalloc/src/base/thread_lister.c deleted file mode 100644 index 9dc8d721892..00000000000 --- a/contrib/libtcmalloc/src/base/thread_lister.c +++ /dev/null @@ -1,83 +0,0 @@ -/* Copyright (c) 2005-2007, Google Inc. - * All rights reserved. - * - * Redistribution and use in source and binary forms, with or without - * modification, are permitted provided that the following conditions are - * met: - * - * * Redistributions of source code must retain the above copyright - * notice, this list of conditions and the following disclaimer. - * * Redistributions in binary form must reproduce the above - * copyright notice, this list of conditions and the following disclaimer - * in the documentation and/or other materials provided with the - * distribution. - * * Neither the name of Google Inc. nor the names of its - * contributors may be used to endorse or promote products derived from - * this software without specific prior written permission. - * - * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS - * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT - * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR - * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT - * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, - * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT - * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, - * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY - * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT - * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE - * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. - * - * --- - * Author: Markus Gutschke - */ - -#include "config.h" - -#include "base/thread_lister.h" - -#include /* needed for NULL on some powerpc platforms (?!) */ -#include -#include /* for getpid */ - -#ifdef HAVE_SYS_PRCTL -# include -#endif - -#include "base/linuxthreads.h" -/* Include other thread listers here that define THREADS macro - * only when they can provide a good implementation. - */ - -#ifndef THREADS - -/* Default trivial thread lister for single-threaded applications, - * or if the multi-threading code has not been ported, yet. - */ - -int TCMalloc_ListAllProcessThreads(void *parameter, - ListAllProcessThreadsCallBack callback, ...) { - int rc; - va_list ap; - pid_t pid; - -#ifdef HAVE_SYS_PRCTL - int dumpable = prctl(PR_GET_DUMPABLE, 0); - if (!dumpable) - prctl(PR_SET_DUMPABLE, 1); -#endif - va_start(ap, callback); - pid = getpid(); - rc = callback(parameter, 1, &pid, ap); - va_end(ap); -#ifdef HAVE_SYS_PRCTL - if (!dumpable) - prctl(PR_SET_DUMPABLE, 0); -#endif - return rc; -} - -int TCMalloc_ResumeAllProcessThreads(int num_threads, pid_t *thread_pids) { - return 1; -} - -#endif /* ifndef THREADS */ diff --git a/contrib/libtcmalloc/src/base/thread_lister.h b/contrib/libtcmalloc/src/base/thread_lister.h deleted file mode 100644 index 6e70b89fef5..00000000000 --- a/contrib/libtcmalloc/src/base/thread_lister.h +++ /dev/null @@ -1,83 +0,0 @@ -/* -*- Mode: c; c-basic-offset: 2; indent-tabs-mode: nil -*- */ -/* Copyright (c) 2005-2007, Google Inc. - * All rights reserved. - * - * Redistribution and use in source and binary forms, with or without - * modification, are permitted provided that the following conditions are - * met: - * - * * Redistributions of source code must retain the above copyright - * notice, this list of conditions and the following disclaimer. - * * Redistributions in binary form must reproduce the above - * copyright notice, this list of conditions and the following disclaimer - * in the documentation and/or other materials provided with the - * distribution. - * * Neither the name of Google Inc. nor the names of its - * contributors may be used to endorse or promote products derived from - * this software without specific prior written permission. - * - * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS - * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT - * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR - * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT - * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, - * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT - * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, - * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY - * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT - * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE - * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. - * - * --- - * Author: Markus Gutschke - */ - -#ifndef _THREAD_LISTER_H -#define _THREAD_LISTER_H - -#include -#include - -#ifdef __cplusplus -extern "C" { -#endif - -typedef int (*ListAllProcessThreadsCallBack)(void *parameter, - int num_threads, - pid_t *thread_pids, - va_list ap); - -/* This function gets the list of all linux threads of the current process - * passes them to the 'callback' along with the 'parameter' pointer; at the - * call back call time all the threads are paused via - * PTRACE_ATTACH. - * The callback is executed from a separate thread which shares only the - * address space, the filesystem, and the filehandles with the caller. Most - * notably, it does not share the same pid and ppid; and if it terminates, - * the rest of the application is still there. 'callback' is supposed to do - * or arrange for TCMalloc_ResumeAllProcessThreads. This happens automatically, if - * the thread raises a synchronous signal (e.g. SIGSEGV); asynchronous - * signals are blocked. If the 'callback' decides to unblock them, it must - * ensure that they cannot terminate the application, or that - * TCMalloc_ResumeAllProcessThreads will get called. - * It is an error for the 'callback' to make any library calls that could - * acquire locks. Most notably, this means that most system calls have to - * avoid going through libc. Also, this means that it is not legal to call - * exit() or abort(). - * We return -1 on error and the return value of 'callback' on success. - */ -int TCMalloc_ListAllProcessThreads(void *parameter, - ListAllProcessThreadsCallBack callback, ...); - -/* This function resumes the list of all linux threads that - * TCMalloc_ListAllProcessThreads pauses before giving to its - * callback. The function returns non-zero if at least one thread was - * suspended and has now been resumed. - */ -int TCMalloc_ResumeAllProcessThreads(int num_threads, pid_t *thread_pids); - -#ifdef __cplusplus -} -#endif - -#endif /* _THREAD_LISTER_H */ diff --git a/contrib/libtcmalloc/src/base/vdso_support.cc b/contrib/libtcmalloc/src/base/vdso_support.cc deleted file mode 100644 index 730df3011d6..00000000000 --- a/contrib/libtcmalloc/src/base/vdso_support.cc +++ /dev/null @@ -1,143 +0,0 @@ -// Copyright (c) 2008, Google Inc. -// All rights reserved. -// -// Redistribution and use in source and binary forms, with or without -// modification, are permitted provided that the following conditions are -// met: -// -// * Redistributions of source code must retain the above copyright -// notice, this list of conditions and the following disclaimer. -// * Redistributions in binary form must reproduce the above -// copyright notice, this list of conditions and the following disclaimer -// in the documentation and/or other materials provided with the -// distribution. -// * Neither the name of Google Inc. nor the names of its -// contributors may be used to endorse or promote products derived from -// this software without specific prior written permission. -// -// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS -// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT -// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR -// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT -// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, -// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT -// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, -// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY -// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT -// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE -// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. - -// --- -// Author: Paul Pluzhnikov -// -// Allow dynamic symbol lookup in the kernel VDSO page. -// -// VDSOSupport -- a class representing kernel VDSO (if present). -// - -#include "base/vdso_support.h" - -#ifdef HAVE_VDSO_SUPPORT // defined in vdso_support.h - -#include -#include // for ptrdiff_t - -#include "base/atomicops.h" // for MemoryBarrier -#include "base/linux_syscall_support.h" -#include "base/logging.h" -#include "base/dynamic_annotations.h" -#include "base/basictypes.h" // for COMPILE_ASSERT - -using base::subtle::MemoryBarrier; - -#ifndef AT_SYSINFO_EHDR -#define AT_SYSINFO_EHDR 33 -#endif - -namespace base { - -const void *VDSOSupport::vdso_base_ = ElfMemImage::kInvalidBase; -VDSOSupport::VDSOSupport() - // If vdso_base_ is still set to kInvalidBase, we got here - // before VDSOSupport::Init has been called. Call it now. - : image_(vdso_base_ == ElfMemImage::kInvalidBase ? Init() : vdso_base_) { -} - -// NOTE: we can't use GoogleOnceInit() below, because we can be -// called by tcmalloc, and none of the *once* stuff may be functional yet. -// -// In addition, we hope that the VDSOSupportHelper constructor -// causes this code to run before there are any threads, and before -// InitGoogle() has executed any chroot or setuid calls. -// -// Finally, even if there is a race here, it is harmless, because -// the operation should be idempotent. -const void *VDSOSupport::Init() { - if (vdso_base_ == ElfMemImage::kInvalidBase) { - // Valgrind zaps AT_SYSINFO_EHDR and friends from the auxv[] - // on stack, and so glibc works as if VDSO was not present. - // But going directly to kernel via /proc/self/auxv below bypasses - // Valgrind zapping. So we check for Valgrind separately. - if (RunningOnValgrind()) { - vdso_base_ = NULL; - return NULL; - } - int fd = open("/proc/self/auxv", O_RDONLY); - if (fd == -1) { - // Kernel too old to have a VDSO. - vdso_base_ = NULL; - return NULL; - } - ElfW(auxv_t) aux; - while (read(fd, &aux, sizeof(aux)) == sizeof(aux)) { - if (aux.a_type == AT_SYSINFO_EHDR) { - COMPILE_ASSERT(sizeof(vdso_base_) == sizeof(aux.a_un.a_val), - unexpected_sizeof_pointer_NE_sizeof_a_val); - vdso_base_ = reinterpret_cast(aux.a_un.a_val); - break; - } - } - close(fd); - if (vdso_base_ == ElfMemImage::kInvalidBase) { - // Didn't find AT_SYSINFO_EHDR in auxv[]. - vdso_base_ = NULL; - } - } - return vdso_base_; -} - -const void *VDSOSupport::SetBase(const void *base) { - CHECK(base != ElfMemImage::kInvalidBase); - const void *old_base = vdso_base_; - vdso_base_ = base; - image_.Init(base); - return old_base; -} - -bool VDSOSupport::LookupSymbol(const char *name, - const char *version, - int type, - SymbolInfo *info) const { - return image_.LookupSymbol(name, version, type, info); -} - -bool VDSOSupport::LookupSymbolByAddress(const void *address, - SymbolInfo *info_out) const { - return image_.LookupSymbolByAddress(address, info_out); -} - -// We need to make sure VDSOSupport::Init() is called before -// the main() runs, since it might do something like setuid or -// chroot. If VDSOSupport -// is used in any global constructor, this will happen, since -// VDSOSupport's constructor calls Init. But if not, we need to -// ensure it here, with a global constructor of our own. This -// is an allowed exception to the normal rule against non-trivial -// global constructors. -static class VDSOInitHelper { - public: - VDSOInitHelper() { VDSOSupport::Init(); } -} vdso_init_helper; -} - -#endif // HAVE_VDSO_SUPPORT diff --git a/contrib/libtcmalloc/src/base/vdso_support.h b/contrib/libtcmalloc/src/base/vdso_support.h deleted file mode 100644 index 0c2213c7c06..00000000000 --- a/contrib/libtcmalloc/src/base/vdso_support.h +++ /dev/null @@ -1,132 +0,0 @@ -// Copyright (c) 2008, Google Inc. -// All rights reserved. -// -// Redistribution and use in source and binary forms, with or without -// modification, are permitted provided that the following conditions are -// met: -// -// * Redistributions of source code must retain the above copyright -// notice, this list of conditions and the following disclaimer. -// * Redistributions in binary form must reproduce the above -// copyright notice, this list of conditions and the following disclaimer -// in the documentation and/or other materials provided with the -// distribution. -// * Neither the name of Google Inc. nor the names of its -// contributors may be used to endorse or promote products derived from -// this software without specific prior written permission. -// -// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS -// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT -// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR -// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT -// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, -// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT -// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, -// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY -// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT -// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE -// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. - -// --- -// Author: Paul Pluzhnikov -// -// Allow dynamic symbol lookup in the kernel VDSO page. -// -// VDSO stands for "Virtual Dynamic Shared Object" -- a page of -// executable code, which looks like a shared library, but doesn't -// necessarily exist anywhere on disk, and which gets mmap()ed into -// every process by kernels which support VDSO, such as 2.6.x for 32-bit -// executables, and 2.6.24 and above for 64-bit executables. -// -// More details could be found here: -// http://www.trilithium.com/johan/2005/08/linux-gate/ -// -// VDSOSupport -- a class representing kernel VDSO (if present). -// -// Example usage: -// VDSOSupport vdso; -// VDSOSupport::SymbolInfo info; -// typedef (*FN)(unsigned *, void *, void *); -// FN fn = NULL; -// if (vdso.LookupSymbol("__vdso_getcpu", "LINUX_2.6", STT_FUNC, &info)) { -// fn = reinterpret_cast(info.address); -// } - -#ifndef BASE_VDSO_SUPPORT_H_ -#define BASE_VDSO_SUPPORT_H_ - -#include "../config.h" -#include "base/basictypes.h" -#include "base/elf_mem_image.h" - -#ifdef HAVE_ELF_MEM_IMAGE - -#define HAVE_VDSO_SUPPORT 1 - -#include // for NULL - -namespace base { - -// NOTE: this class may be used from within tcmalloc, and can not -// use any memory allocation routines. -class VDSOSupport { - public: - VDSOSupport(); - - typedef ElfMemImage::SymbolInfo SymbolInfo; - typedef ElfMemImage::SymbolIterator SymbolIterator; - - // Answers whether we have a vdso at all. - bool IsPresent() const { return image_.IsPresent(); } - - // Allow to iterate over all VDSO symbols. - SymbolIterator begin() const { return image_.begin(); } - SymbolIterator end() const { return image_.end(); } - - // Look up versioned dynamic symbol in the kernel VDSO. - // Returns false if VDSO is not present, or doesn't contain given - // symbol/version/type combination. - // If info_out != NULL, additional details are filled in. - bool LookupSymbol(const char *name, const char *version, - int symbol_type, SymbolInfo *info_out) const; - - // Find info about symbol (if any) which overlaps given address. - // Returns true if symbol was found; false if VDSO isn't present - // or doesn't have a symbol overlapping given address. - // If info_out != NULL, additional details are filled in. - bool LookupSymbolByAddress(const void *address, SymbolInfo *info_out) const; - - // Used only for testing. Replace real VDSO base with a mock. - // Returns previous value of vdso_base_. After you are done testing, - // you are expected to call SetBase() with previous value, in order to - // reset state to the way it was. - const void *SetBase(const void *s); - - // Computes vdso_base_ and returns it. Should be called as early as - // possible; before any thread creation, chroot or setuid. - static const void *Init(); - - private: - // image_ represents VDSO ELF image in memory. - // image_.ehdr_ == NULL implies there is no VDSO. - ElfMemImage image_; - - // Cached value of auxv AT_SYSINFO_EHDR, computed once. - // This is a tri-state: - // kInvalidBase => value hasn't been determined yet. - // 0 => there is no VDSO. - // else => vma of VDSO Elf{32,64}_Ehdr. - // - // When testing with mock VDSO, low bit is set. - // The low bit is always available because vdso_base_ is - // page-aligned. - static const void *vdso_base_; - - DISALLOW_COPY_AND_ASSIGN(VDSOSupport); -}; - -} // namespace base - -#endif // HAVE_ELF_MEM_IMAGE - -#endif // BASE_VDSO_SUPPORT_H_ diff --git a/contrib/libtcmalloc/src/central_freelist.cc b/contrib/libtcmalloc/src/central_freelist.cc deleted file mode 100644 index 11b190dcfee..00000000000 --- a/contrib/libtcmalloc/src/central_freelist.cc +++ /dev/null @@ -1,387 +0,0 @@ -// -*- Mode: C++; c-basic-offset: 2; indent-tabs-mode: nil -*- -// Copyright (c) 2008, Google Inc. -// All rights reserved. -// -// Redistribution and use in source and binary forms, with or without -// modification, are permitted provided that the following conditions are -// met: -// -// * Redistributions of source code must retain the above copyright -// notice, this list of conditions and the following disclaimer. -// * Redistributions in binary form must reproduce the above -// copyright notice, this list of conditions and the following disclaimer -// in the documentation and/or other materials provided with the -// distribution. -// * Neither the name of Google Inc. nor the names of its -// contributors may be used to endorse or promote products derived from -// this software without specific prior written permission. -// -// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS -// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT -// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR -// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT -// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, -// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT -// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, -// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY -// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT -// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE -// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. - -// --- -// Author: Sanjay Ghemawat - -#include "config.h" -#include -#include "central_freelist.h" -#include "internal_logging.h" // for ASSERT, MESSAGE -#include "linked_list.h" // for SLL_Next, SLL_Push, etc -#include "page_heap.h" // for PageHeap -#include "static_vars.h" // for Static - -using std::min; -using std::max; - -namespace tcmalloc { - -void CentralFreeList::Init(size_t cl) { - size_class_ = cl; - tcmalloc::DLL_Init(&empty_); - tcmalloc::DLL_Init(&nonempty_); - num_spans_ = 0; - counter_ = 0; - - max_cache_size_ = kMaxNumTransferEntries; -#ifdef TCMALLOC_SMALL_BUT_SLOW - // Disable the transfer cache for the small footprint case. - cache_size_ = 0; -#else - cache_size_ = 16; -#endif - if (cl > 0) { - // Limit the maximum size of the cache based on the size class. If this - // is not done, large size class objects will consume a lot of memory if - // they just sit in the transfer cache. - int32_t bytes = Static::sizemap()->ByteSizeForClass(cl); - int32_t objs_to_move = Static::sizemap()->num_objects_to_move(cl); - - ASSERT(objs_to_move > 0 && bytes > 0); - // Limit each size class cache to at most 1MB of objects or one entry, - // whichever is greater. Total transfer cache memory used across all - // size classes then can't be greater than approximately - // 1MB * kMaxNumTransferEntries. - // min and max are in parens to avoid macro-expansion on windows. - max_cache_size_ = (min)(max_cache_size_, - (max)(1, (1024 * 1024) / (bytes * objs_to_move))); - cache_size_ = (min)(cache_size_, max_cache_size_); - } - used_slots_ = 0; - ASSERT(cache_size_ <= max_cache_size_); -} - -void CentralFreeList::ReleaseListToSpans(void* start) { - while (start) { - void *next = SLL_Next(start); - ReleaseToSpans(start); - start = next; - } -} - -// MapObjectToSpan should logically be part of ReleaseToSpans. But -// this triggers an optimization bug in gcc 4.5.0. Moving to a -// separate function, and making sure that function isn't inlined, -// seems to fix the problem. It also should be fixed for gcc 4.5.1. -static -#if __GNUC__ == 4 && __GNUC_MINOR__ == 5 && __GNUC_PATCHLEVEL__ == 0 -__attribute__ ((noinline)) -#endif -Span* MapObjectToSpan(void* object) { - const PageID p = reinterpret_cast(object) >> kPageShift; - Span* span = Static::pageheap()->GetDescriptor(p); - return span; -} - -void CentralFreeList::ReleaseToSpans(void* object) { - Span* span = MapObjectToSpan(object); - ASSERT(span != NULL); - ASSERT(span->refcount > 0); - - // If span is empty, move it to non-empty list - if (span->objects == NULL) { - tcmalloc::DLL_Remove(span); - tcmalloc::DLL_Prepend(&nonempty_, span); - Event(span, 'N', 0); - } - - // The following check is expensive, so it is disabled by default - if (false) { - // Check that object does not occur in list - int got = 0; - for (void* p = span->objects; p != NULL; p = *((void**) p)) { - ASSERT(p != object); - got++; - } - ASSERT(got + span->refcount == - (span->length<ByteSizeForClass(span->sizeclass)); - } - - counter_++; - span->refcount--; - if (span->refcount == 0) { - Event(span, '#', 0); - counter_ -= ((span->length<ByteSizeForClass(span->sizeclass)); - tcmalloc::DLL_Remove(span); - --num_spans_; - - // Release central list lock while operating on pageheap - lock_.Unlock(); - { - SpinLockHolder h(Static::pageheap_lock()); - Static::pageheap()->Delete(span); - } - lock_.Lock(); - } else { - *(reinterpret_cast(object)) = span->objects; - span->objects = object; - } -} - -bool CentralFreeList::EvictRandomSizeClass( - int locked_size_class, bool force) { - static int race_counter = 0; - int t = race_counter++; // Updated without a lock, but who cares. - if (t >= kNumClasses) { - while (t >= kNumClasses) { - t -= kNumClasses; - } - race_counter = t; - } - ASSERT(t >= 0); - ASSERT(t < kNumClasses); - if (t == locked_size_class) return false; - return Static::central_cache()[t].ShrinkCache(locked_size_class, force); -} - -bool CentralFreeList::MakeCacheSpace() { - // Is there room in the cache? - if (used_slots_ < cache_size_) return true; - // Check if we can expand this cache? - if (cache_size_ == max_cache_size_) return false; - // Ok, we'll try to grab an entry from some other size class. - if (EvictRandomSizeClass(size_class_, false) || - EvictRandomSizeClass(size_class_, true)) { - // Succeeded in evicting, we're going to make our cache larger. - // However, we may have dropped and re-acquired the lock in - // EvictRandomSizeClass (via ShrinkCache and the LockInverter), so the - // cache_size may have changed. Therefore, check and verify that it is - // still OK to increase the cache_size. - if (cache_size_ < max_cache_size_) { - cache_size_++; - return true; - } - } - return false; -} - - -namespace { -class LockInverter { - private: - SpinLock *held_, *temp_; - public: - inline explicit LockInverter(SpinLock* held, SpinLock *temp) - : held_(held), temp_(temp) { held_->Unlock(); temp_->Lock(); } - inline ~LockInverter() { temp_->Unlock(); held_->Lock(); } -}; -} - -// This function is marked as NO_THREAD_SAFETY_ANALYSIS because it uses -// LockInverter to release one lock and acquire another in scoped-lock -// style, which our current annotation/analysis does not support. -bool CentralFreeList::ShrinkCache(int locked_size_class, bool force) - NO_THREAD_SAFETY_ANALYSIS { - // Start with a quick check without taking a lock. - if (cache_size_ == 0) return false; - // We don't evict from a full cache unless we are 'forcing'. - if (force == false && used_slots_ == cache_size_) return false; - - // Grab lock, but first release the other lock held by this thread. We use - // the lock inverter to ensure that we never hold two size class locks - // concurrently. That can create a deadlock because there is no well - // defined nesting order. - LockInverter li(&Static::central_cache()[locked_size_class].lock_, &lock_); - ASSERT(used_slots_ <= cache_size_); - ASSERT(0 <= cache_size_); - if (cache_size_ == 0) return false; - if (used_slots_ == cache_size_) { - if (force == false) return false; - // ReleaseListToSpans releases the lock, so we have to make all the - // updates to the central list before calling it. - cache_size_--; - used_slots_--; - ReleaseListToSpans(tc_slots_[used_slots_].head); - return true; - } - cache_size_--; - return true; -} - -void CentralFreeList::InsertRange(void *start, void *end, int N) { - SpinLockHolder h(&lock_); - if (N == Static::sizemap()->num_objects_to_move(size_class_) && - MakeCacheSpace()) { - int slot = used_slots_++; - ASSERT(slot >=0); - ASSERT(slot < max_cache_size_); - TCEntry *entry = &tc_slots_[slot]; - entry->head = start; - entry->tail = end; - return; - } - ReleaseListToSpans(start); -} - -int CentralFreeList::RemoveRange(void **start, void **end, int N) { - ASSERT(N > 0); - lock_.Lock(); - if (N == Static::sizemap()->num_objects_to_move(size_class_) && - used_slots_ > 0) { - int slot = --used_slots_; - ASSERT(slot >= 0); - TCEntry *entry = &tc_slots_[slot]; - *start = entry->head; - *end = entry->tail; - lock_.Unlock(); - return N; - } - - int result = 0; - *start = NULL; - *end = NULL; - // TODO: Prefetch multiple TCEntries? - result = FetchFromOneSpansSafe(N, start, end); - if (result != 0) { - while (result < N) { - int n; - void* head = NULL; - void* tail = NULL; - n = FetchFromOneSpans(N - result, &head, &tail); - if (!n) break; - result += n; - SLL_PushRange(start, head, tail); - } - } - lock_.Unlock(); - return result; -} - - -int CentralFreeList::FetchFromOneSpansSafe(int N, void **start, void **end) { - int result = FetchFromOneSpans(N, start, end); - if (!result) { - Populate(); - result = FetchFromOneSpans(N, start, end); - } - return result; -} - -int CentralFreeList::FetchFromOneSpans(int N, void **start, void **end) { - if (tcmalloc::DLL_IsEmpty(&nonempty_)) return 0; - Span* span = nonempty_.next; - - ASSERT(span->objects != NULL); - - int result = 0; - void *prev, *curr; - curr = span->objects; - do { - prev = curr; - curr = *(reinterpret_cast(curr)); - } while (++result < N && curr != NULL); - - if (curr == NULL) { - // Move to empty list - tcmalloc::DLL_Remove(span); - tcmalloc::DLL_Prepend(&empty_, span); - Event(span, 'E', 0); - } - - *start = span->objects; - *end = prev; - span->objects = curr; - SLL_SetNext(*end, NULL); - span->refcount += result; - counter_ -= result; - return result; -} - -// Fetch memory from the system and add to the central cache freelist. -void CentralFreeList::Populate() { - // Release central list lock while operating on pageheap - lock_.Unlock(); - const size_t npages = Static::sizemap()->class_to_pages(size_class_); - - Span* span; - { - SpinLockHolder h(Static::pageheap_lock()); - span = Static::pageheap()->New(npages); - if (span) Static::pageheap()->RegisterSizeClass(span, size_class_); - } - if (span == NULL) { - Log(kLog, __FILE__, __LINE__, - "tcmalloc: allocation failed", npages << kPageShift); - lock_.Lock(); - return; - } - ASSERT(span->length == npages); - // Cache sizeclass info eagerly. Locking is not necessary. - // (Instead of being eager, we could just replace any stale info - // about this span, but that seems to be no better in practice.) - for (int i = 0; i < npages; i++) { - Static::pageheap()->CacheSizeClass(span->start + i, size_class_); - } - - // Split the block into pieces and add to the free-list - // TODO: coloring of objects to avoid cache conflicts? - void** tail = &span->objects; - char* ptr = reinterpret_cast(span->start << kPageShift); - char* limit = ptr + (npages << kPageShift); - const size_t size = Static::sizemap()->ByteSizeForClass(size_class_); - int num = 0; - while (ptr + size <= limit) { - *tail = ptr; - tail = reinterpret_cast(ptr); - ptr += size; - num++; - } - ASSERT(ptr <= limit); - *tail = NULL; - span->refcount = 0; // No sub-object in use yet - - // Add span to list of non-empty spans - lock_.Lock(); - tcmalloc::DLL_Prepend(&nonempty_, span); - ++num_spans_; - counter_ += num; -} - -int CentralFreeList::tc_length() { - SpinLockHolder h(&lock_); - return used_slots_ * Static::sizemap()->num_objects_to_move(size_class_); -} - -size_t CentralFreeList::OverheadBytes() { - SpinLockHolder h(&lock_); - if (size_class_ == 0) { // 0 holds the 0-sized allocations - return 0; - } - const size_t pages_per_span = Static::sizemap()->class_to_pages(size_class_); - const size_t object_size = Static::sizemap()->class_to_size(size_class_); - ASSERT(object_size > 0); - const size_t overhead_per_span = (pages_per_span * kPageSize) % object_size; - return num_spans_ * overhead_per_span; -} - -} // namespace tcmalloc diff --git a/contrib/libtcmalloc/src/central_freelist.h b/contrib/libtcmalloc/src/central_freelist.h deleted file mode 100644 index 4148680d20a..00000000000 --- a/contrib/libtcmalloc/src/central_freelist.h +++ /dev/null @@ -1,211 +0,0 @@ -// -*- Mode: C++; c-basic-offset: 2; indent-tabs-mode: nil -*- -// Copyright (c) 2008, Google Inc. -// All rights reserved. -// -// Redistribution and use in source and binary forms, with or without -// modification, are permitted provided that the following conditions are -// met: -// -// * Redistributions of source code must retain the above copyright -// notice, this list of conditions and the following disclaimer. -// * Redistributions in binary form must reproduce the above -// copyright notice, this list of conditions and the following disclaimer -// in the documentation and/or other materials provided with the -// distribution. -// * Neither the name of Google Inc. nor the names of its -// contributors may be used to endorse or promote products derived from -// this software without specific prior written permission. -// -// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS -// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT -// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR -// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT -// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, -// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT -// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, -// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY -// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT -// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE -// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. - -// --- -// Author: Sanjay Ghemawat - -#ifndef TCMALLOC_CENTRAL_FREELIST_H_ -#define TCMALLOC_CENTRAL_FREELIST_H_ - -#include "config.h" -#include // for size_t -#ifdef HAVE_STDINT_H -#include // for int32_t -#endif -#include "base/spinlock.h" -#include "base/thread_annotations.h" -#include "common.h" -#include "span.h" - -namespace tcmalloc { - -// Data kept per size-class in central cache. -class CentralFreeList { - public: - // A CentralFreeList may be used before its constructor runs. - // So we prevent lock_'s constructor from doing anything to the - // lock_ state. - CentralFreeList() : lock_(base::LINKER_INITIALIZED) { } - - void Init(size_t cl); - - // These methods all do internal locking. - - // Insert the specified range into the central freelist. N is the number of - // elements in the range. RemoveRange() is the opposite operation. - void InsertRange(void *start, void *end, int N); - - // Returns the actual number of fetched elements and sets *start and *end. - int RemoveRange(void **start, void **end, int N); - - // Returns the number of free objects in cache. - int length() { - SpinLockHolder h(&lock_); - return counter_; - } - - // Returns the number of free objects in the transfer cache. - int tc_length(); - - // Returns the memory overhead (internal fragmentation) attributable - // to the freelist. This is memory lost when the size of elements - // in a freelist doesn't exactly divide the page-size (an 8192-byte - // page full of 5-byte objects would have 2 bytes memory overhead). - size_t OverheadBytes(); - - // Lock/Unlock the internal SpinLock. Used on the pthread_atfork call - // to set the lock in a consistent state before the fork. - void Lock() { - lock_.Lock(); - } - - void Unlock() { - lock_.Unlock(); - } - - private: - // TransferCache is used to cache transfers of - // sizemap.num_objects_to_move(size_class) back and forth between - // thread caches and the central cache for a given size class. - struct TCEntry { - void *head; // Head of chain of objects. - void *tail; // Tail of chain of objects. - }; - - // A central cache freelist can have anywhere from 0 to kMaxNumTransferEntries - // slots to put link list chains into. -#ifdef TCMALLOC_SMALL_BUT_SLOW - // For the small memory model, the transfer cache is not used. - static const int kMaxNumTransferEntries = 0; -#else - // Starting point for the the maximum number of entries in the transfer cache. - // This actual maximum for a given size class may be lower than this - // maximum value. - static const int kMaxNumTransferEntries = 64; -#endif - - // REQUIRES: lock_ is held - // Remove object from cache and return. - // Return NULL if no free entries in cache. - int FetchFromOneSpans(int N, void **start, void **end) EXCLUSIVE_LOCKS_REQUIRED(lock_); - - // REQUIRES: lock_ is held - // Remove object from cache and return. Fetches - // from pageheap if cache is empty. Only returns - // NULL on allocation failure. - int FetchFromOneSpansSafe(int N, void **start, void **end) EXCLUSIVE_LOCKS_REQUIRED(lock_); - - // REQUIRES: lock_ is held - // Release a linked list of objects to spans. - // May temporarily release lock_. - void ReleaseListToSpans(void *start) EXCLUSIVE_LOCKS_REQUIRED(lock_); - - // REQUIRES: lock_ is held - // Release an object to spans. - // May temporarily release lock_. - void ReleaseToSpans(void* object) EXCLUSIVE_LOCKS_REQUIRED(lock_); - - // REQUIRES: lock_ is held - // Populate cache by fetching from the page heap. - // May temporarily release lock_. - void Populate() EXCLUSIVE_LOCKS_REQUIRED(lock_); - - // REQUIRES: lock is held. - // Tries to make room for a TCEntry. If the cache is full it will try to - // expand it at the cost of some other cache size. Return false if there is - // no space. - bool MakeCacheSpace() EXCLUSIVE_LOCKS_REQUIRED(lock_); - - // REQUIRES: lock_ for locked_size_class is held. - // Picks a "random" size class to steal TCEntry slot from. In reality it - // just iterates over the sizeclasses but does so without taking a lock. - // Returns true on success. - // May temporarily lock a "random" size class. - static bool EvictRandomSizeClass(int locked_size_class, bool force); - - // REQUIRES: lock_ is *not* held. - // Tries to shrink the Cache. If force is true it will relase objects to - // spans if it allows it to shrink the cache. Return false if it failed to - // shrink the cache. Decrements cache_size_ on succeess. - // May temporarily take lock_. If it takes lock_, the locked_size_class - // lock is released to keep the thread from holding two size class locks - // concurrently which could lead to a deadlock. - bool ShrinkCache(int locked_size_class, bool force) LOCKS_EXCLUDED(lock_); - - // This lock protects all the data members. cached_entries and cache_size_ - // may be looked at without holding the lock. - SpinLock lock_; - - // We keep linked lists of empty and non-empty spans. - size_t size_class_; // My size class - Span empty_; // Dummy header for list of empty spans - Span nonempty_; // Dummy header for list of non-empty spans - size_t num_spans_; // Number of spans in empty_ plus nonempty_ - size_t counter_; // Number of free objects in cache entry - - // Here we reserve space for TCEntry cache slots. Space is preallocated - // for the largest possible number of entries than any one size class may - // accumulate. Not all size classes are allowed to accumulate - // kMaxNumTransferEntries, so there is some wasted space for those size - // classes. - TCEntry tc_slots_[kMaxNumTransferEntries]; - - // Number of currently used cached entries in tc_slots_. This variable is - // updated under a lock but can be read without one. - int32_t used_slots_; - // The current number of slots for this size class. This is an - // adaptive value that is increased if there is lots of traffic - // on a given size class. - int32_t cache_size_; - // Maximum size of the cache for a given size class. - int32_t max_cache_size_; -}; - -// Pads each CentralCache object to multiple of 64 bytes. Since some -// compilers (such as MSVC) don't like it when the padding is 0, I use -// template specialization to remove the padding entirely when -// sizeof(CentralFreeList) is a multiple of 64. -template -class CentralFreeListPaddedTo : public CentralFreeList { - private: - char pad_[64 - kFreeListSizeMod64]; -}; - -template<> -class CentralFreeListPaddedTo<0> : public CentralFreeList { -}; - -class CentralFreeListPadded : public CentralFreeListPaddedTo< - sizeof(CentralFreeList) % 64> { -}; - -} // namespace tcmalloc - -#endif // TCMALLOC_CENTRAL_FREELIST_H_ diff --git a/contrib/libtcmalloc/src/common.cc b/contrib/libtcmalloc/src/common.cc deleted file mode 100644 index 313848c37b6..00000000000 --- a/contrib/libtcmalloc/src/common.cc +++ /dev/null @@ -1,275 +0,0 @@ -// -*- Mode: C++; c-basic-offset: 2; indent-tabs-mode: nil -*- -// Copyright (c) 2008, Google Inc. -// All rights reserved. -// -// Redistribution and use in source and binary forms, with or without -// modification, are permitted provided that the following conditions are -// met: -// -// * Redistributions of source code must retain the above copyright -// notice, this list of conditions and the following disclaimer. -// * Redistributions in binary form must reproduce the above -// copyright notice, this list of conditions and the following disclaimer -// in the documentation and/or other materials provided with the -// distribution. -// * Neither the name of Google Inc. nor the names of its -// contributors may be used to endorse or promote products derived from -// this software without specific prior written permission. -// -// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS -// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT -// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR -// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT -// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, -// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT -// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, -// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY -// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT -// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE -// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. - -// --- -// Author: Sanjay Ghemawat - -#include // for getenv and strtol -#include "config.h" -#include "common.h" -#include "system-alloc.h" -#include "base/spinlock.h" -#include "getenv_safe.h" // TCMallocGetenvSafe - -namespace tcmalloc { - -// Define the maximum number of object per classe type to transfer between -// thread and central caches. -static int32 FLAGS_tcmalloc_transfer_num_objects; - -static const int32 kDefaultTransferNumObjecs = 512; - -// The init function is provided to explicit initialize the variable value -// from the env. var to avoid C++ global construction that might defer its -// initialization after a malloc/new call. -static inline void InitTCMallocTransferNumObjects() -{ - if (UNLIKELY(FLAGS_tcmalloc_transfer_num_objects == 0)) { - const char *envval = TCMallocGetenvSafe("TCMALLOC_TRANSFER_NUM_OBJ"); - FLAGS_tcmalloc_transfer_num_objects = !envval ? kDefaultTransferNumObjecs : - strtol(envval, NULL, 10); - } -} - -// Note: the following only works for "n"s that fit in 32-bits, but -// that is fine since we only use it for small sizes. -static inline int LgFloor(size_t n) { - int log = 0; - for (int i = 4; i >= 0; --i) { - int shift = (1 << i); - size_t x = n >> shift; - if (x != 0) { - n = x; - log += shift; - } - } - ASSERT(n == 1); - return log; -} - -int AlignmentForSize(size_t size) { - int alignment = kAlignment; - if (size > kMaxSize) { - // Cap alignment at kPageSize for large sizes. - alignment = kPageSize; - } else if (size >= 128) { - // Space wasted due to alignment is at most 1/8, i.e., 12.5%. - alignment = (1 << LgFloor(size)) / 8; - } else if (size >= kMinAlign) { - // We need an alignment of at least 16 bytes to satisfy - // requirements for some SSE types. - alignment = kMinAlign; - } - // Maximum alignment allowed is page size alignment. - if (alignment > kPageSize) { - alignment = kPageSize; - } - CHECK_CONDITION(size < kMinAlign || alignment >= kMinAlign); - CHECK_CONDITION((alignment & (alignment - 1)) == 0); - return alignment; -} - -int SizeMap::NumMoveSize(size_t size) { - if (size == 0) return 0; - // Use approx 64k transfers between thread and central caches. - int num = static_cast(64.0 * 1024.0 / size); - if (num < 2) num = 2; - - // Avoid bringing too many objects into small object free lists. - // If this value is too large: - // - We waste memory with extra objects sitting in the thread caches. - // - The central freelist holds its lock for too long while - // building a linked list of objects, slowing down the allocations - // of other threads. - // If this value is too small: - // - We go to the central freelist too often and we have to acquire - // its lock each time. - // This value strikes a balance between the constraints above. - if (num > FLAGS_tcmalloc_transfer_num_objects) - num = FLAGS_tcmalloc_transfer_num_objects; - - return num; -} - -// Initialize the mapping arrays -void SizeMap::Init() { - InitTCMallocTransferNumObjects(); - - // Do some sanity checking on add_amount[]/shift_amount[]/class_array[] - if (ClassIndex(0) != 0) { - Log(kCrash, __FILE__, __LINE__, - "Invalid class index for size 0", ClassIndex(0)); - } - if (ClassIndex(kMaxSize) >= sizeof(class_array_)) { - Log(kCrash, __FILE__, __LINE__, - "Invalid class index for kMaxSize", ClassIndex(kMaxSize)); - } - - // Compute the size classes we want to use - int sc = 1; // Next size class to assign - int alignment = kAlignment; - CHECK_CONDITION(kAlignment <= kMinAlign); - for (size_t size = kAlignment; size <= kMaxSize; size += alignment) { - alignment = AlignmentForSize(size); - CHECK_CONDITION((size % alignment) == 0); - - int blocks_to_move = NumMoveSize(size) / 4; - size_t psize = 0; - do { - psize += kPageSize; - // Allocate enough pages so leftover is less than 1/8 of total. - // This bounds wasted space to at most 12.5%. - while ((psize % size) > (psize >> 3)) { - psize += kPageSize; - } - // Continue to add pages until there are at least as many objects in - // the span as are needed when moving objects from the central - // freelists and spans to the thread caches. - } while ((psize / size) < (blocks_to_move)); - const size_t my_pages = psize >> kPageShift; - - if (sc > 1 && my_pages == class_to_pages_[sc-1]) { - // See if we can merge this into the previous class without - // increasing the fragmentation of the previous class. - const size_t my_objects = (my_pages << kPageShift) / size; - const size_t prev_objects = (class_to_pages_[sc-1] << kPageShift) - / class_to_size_[sc-1]; - if (my_objects == prev_objects) { - // Adjust last class to include this size - class_to_size_[sc-1] = size; - continue; - } - } - - // Add new class - class_to_pages_[sc] = my_pages; - class_to_size_[sc] = size; - sc++; - } - if (sc != kNumClasses) { - Log(kCrash, __FILE__, __LINE__, - "wrong number of size classes: (found vs. expected )", sc, kNumClasses); - } - - // Initialize the mapping arrays - int next_size = 0; - for (int c = 1; c < kNumClasses; c++) { - const int max_size_in_class = class_to_size_[c]; - for (int s = next_size; s <= max_size_in_class; s += kAlignment) { - class_array_[ClassIndex(s)] = c; - } - next_size = max_size_in_class + kAlignment; - } - - // Double-check sizes just to be safe - for (size_t size = 0; size <= kMaxSize;) { - const int sc = SizeClass(size); - if (sc <= 0 || sc >= kNumClasses) { - Log(kCrash, __FILE__, __LINE__, - "Bad size class (class, size)", sc, size); - } - if (sc > 1 && size <= class_to_size_[sc-1]) { - Log(kCrash, __FILE__, __LINE__, - "Allocating unnecessarily large class (class, size)", sc, size); - } - const size_t s = class_to_size_[sc]; - if (size > s || s == 0) { - Log(kCrash, __FILE__, __LINE__, - "Bad (class, size, requested)", sc, s, size); - } - if (size <= kMaxSmallSize) { - size += 8; - } else { - size += 128; - } - } - - // Initialize the num_objects_to_move array. - for (size_t cl = 1; cl < kNumClasses; ++cl) { - num_objects_to_move_[cl] = NumMoveSize(ByteSizeForClass(cl)); - } -} - -// Metadata allocator -- keeps stats about how many bytes allocated. -static uint64_t metadata_system_bytes_ = 0; -static const size_t kMetadataAllocChunkSize = 8*1024*1024; -// As ThreadCache objects are allocated with MetaDataAlloc, and also -// CACHELINE_ALIGNED, we must use the same alignment as TCMalloc_SystemAlloc. -static const size_t kMetadataAllignment = sizeof(MemoryAligner); - -static char *metadata_chunk_alloc_; -static size_t metadata_chunk_avail_; - -static SpinLock metadata_alloc_lock(SpinLock::LINKER_INITIALIZED); - -void* MetaDataAlloc(size_t bytes) { - if (bytes >= kMetadataAllocChunkSize) { - void *rv = TCMalloc_SystemAlloc(bytes, - NULL, kMetadataAllignment); - if (rv != NULL) { - metadata_system_bytes_ += bytes; - } - return rv; - } - - SpinLockHolder h(&metadata_alloc_lock); - - // the following works by essentially turning address to integer of - // log_2 kMetadataAllignment size and negating it. I.e. negated - // value + original value gets 0 and that's what we want modulo - // kMetadataAllignment. Note, we negate before masking higher bits - // off, otherwise we'd have to mask them off after negation anyways. - intptr_t alignment = -reinterpret_cast(metadata_chunk_alloc_) & (kMetadataAllignment-1); - - if (metadata_chunk_avail_ < bytes + alignment) { - size_t real_size; - void *ptr = TCMalloc_SystemAlloc(kMetadataAllocChunkSize, - &real_size, kMetadataAllignment); - if (ptr == NULL) { - return NULL; - } - - metadata_chunk_alloc_ = static_cast(ptr); - metadata_chunk_avail_ = real_size; - - alignment = 0; - } - - void *rv = static_cast(metadata_chunk_alloc_ + alignment); - bytes += alignment; - metadata_chunk_alloc_ += bytes; - metadata_chunk_avail_ -= bytes; - metadata_system_bytes_ += bytes; - return rv; -} - -uint64_t metadata_system_bytes() { return metadata_system_bytes_; } - -} // namespace tcmalloc diff --git a/contrib/libtcmalloc/src/common.h b/contrib/libtcmalloc/src/common.h deleted file mode 100644 index e8a1ba6972c..00000000000 --- a/contrib/libtcmalloc/src/common.h +++ /dev/null @@ -1,295 +0,0 @@ -// -*- Mode: C++; c-basic-offset: 2; indent-tabs-mode: nil -*- -// Copyright (c) 2008, Google Inc. -// All rights reserved. -// -// Redistribution and use in source and binary forms, with or without -// modification, are permitted provided that the following conditions are -// met: -// -// * Redistributions of source code must retain the above copyright -// notice, this list of conditions and the following disclaimer. -// * Redistributions in binary form must reproduce the above -// copyright notice, this list of conditions and the following disclaimer -// in the documentation and/or other materials provided with the -// distribution. -// * Neither the name of Google Inc. nor the names of its -// contributors may be used to endorse or promote products derived from -// this software without specific prior written permission. -// -// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS -// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT -// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR -// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT -// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, -// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT -// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, -// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY -// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT -// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE -// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. - -// --- -// Author: Sanjay Ghemawat -// -// Common definitions for tcmalloc code. - -#ifndef TCMALLOC_COMMON_H_ -#define TCMALLOC_COMMON_H_ - -#include "config.h" -#include // for size_t -#ifdef HAVE_STDINT_H -#include // for uintptr_t, uint64_t -#endif -#include "internal_logging.h" // for ASSERT, etc -#include "base/basictypes.h" // for LIKELY, etc - -#ifdef HAVE_BUILTIN_EXPECT -#define LIKELY(x) __builtin_expect(!!(x), 1) -#define UNLIKELY(x) __builtin_expect(!!(x), 0) -#else -#define LIKELY(x) (x) -#define UNLIKELY(x) (x) -#endif - -// Type that can hold a page number -typedef uintptr_t PageID; - -// Type that can hold the length of a run of pages -typedef uintptr_t Length; - -//------------------------------------------------------------------- -// Configuration -//------------------------------------------------------------------- - -#if defined(TCMALLOC_ALIGN_8BYTES) -// Unless we force to use 8 bytes alignment we use an alignment of -// at least 16 bytes to statisfy requirements for some SSE types. -// Keep in mind when using the 16 bytes alignment you can have a space -// waste due alignment of 25%. (eg malloc of 24 bytes will get 32 bytes) -static const size_t kMinAlign = 8; -// Number of classes created until reach page size 128. -static const size_t kBaseClasses = 16; -#else -static const size_t kMinAlign = 16; -static const size_t kBaseClasses = 9; -#endif - -// Using large pages speeds up the execution at a cost of larger memory use. -// Deallocation may speed up by a factor as the page map gets 8x smaller, so -// lookups in the page map result in fewer L2 cache misses, which translates to -// speedup for application/platform combinations with high L2 cache pressure. -// As the number of size classes increases with large pages, we increase -// the thread cache allowance to avoid passing more free ranges to and from -// central lists. Also, larger pages are less likely to get freed. -// These two factors cause a bounded increase in memory use. -#if defined(TCMALLOC_32K_PAGES) -static const size_t kPageShift = 15; -static const size_t kNumClasses = kBaseClasses + 69; -#elif defined(TCMALLOC_64K_PAGES) -static const size_t kPageShift = 16; -static const size_t kNumClasses = kBaseClasses + 73; -#else -static const size_t kPageShift = 13; -static const size_t kNumClasses = kBaseClasses + 79; -#endif - -static const size_t kMaxThreadCacheSize = 4 << 20; - -static const size_t kPageSize = 1 << kPageShift; -static const size_t kMaxSize = 256 * 1024; -static const size_t kAlignment = 8; -static const size_t kLargeSizeClass = 0; -// For all span-lengths < kMaxPages we keep an exact-size list. -static const size_t kMaxPages = 1 << (20 - kPageShift); - -// Default bound on the total amount of thread caches. -#ifdef TCMALLOC_SMALL_BUT_SLOW -// Make the overall thread cache no bigger than that of a single thread -// for the small memory footprint case. -static const size_t kDefaultOverallThreadCacheSize = kMaxThreadCacheSize; -#else -static const size_t kDefaultOverallThreadCacheSize = 8u * kMaxThreadCacheSize; -#endif - -// Lower bound on the per-thread cache sizes -static const size_t kMinThreadCacheSize = kMaxSize * 2; - -// The number of bytes one ThreadCache will steal from another when -// the first ThreadCache is forced to Scavenge(), delaying the -// next call to Scavenge for this thread. -static const size_t kStealAmount = 1 << 16; - -// The number of times that a deallocation can cause a freelist to -// go over its max_length() before shrinking max_length(). -static const int kMaxOverages = 3; - -// Maximum length we allow a per-thread free-list to have before we -// move objects from it into the corresponding central free-list. We -// want this big to avoid locking the central free-list too often. It -// should not hurt to make this list somewhat big because the -// scavenging code will shrink it down when its contents are not in use. -static const int kMaxDynamicFreeListLength = 8192; - -static const Length kMaxValidPages = (~static_cast(0)) >> kPageShift; - -#if defined __x86_64__ -// All current and planned x86_64 processors only look at the lower 48 bits -// in virtual to physical address translation. The top 16 are thus unused. -// TODO(rus): Under what operating systems can we increase it safely to 17? -// This lets us use smaller page maps. On first allocation, a 36-bit page map -// uses only 96 KB instead of the 4.5 MB used by a 52-bit page map. -static const int kAddressBits = (sizeof(void*) < 8 ? (8 * sizeof(void*)) : 48); -#else -static const int kAddressBits = 8 * sizeof(void*); -#endif - -namespace tcmalloc { - -// Convert byte size into pages. This won't overflow, but may return -// an unreasonably large value if bytes is huge enough. -inline Length pages(size_t bytes) { - return (bytes >> kPageShift) + - ((bytes & (kPageSize - 1)) > 0 ? 1 : 0); -} - -// For larger allocation sizes, we use larger memory alignments to -// reduce the number of size classes. -int AlignmentForSize(size_t size); - -// Size-class information + mapping -class SizeMap { - private: - // Number of objects to move between a per-thread list and a central - // list in one shot. We want this to be not too small so we can - // amortize the lock overhead for accessing the central list. Making - // it too big may temporarily cause unnecessary memory wastage in the - // per-thread free list until the scavenger cleans up the list. - int num_objects_to_move_[kNumClasses]; - - //------------------------------------------------------------------- - // Mapping from size to size_class and vice versa - //------------------------------------------------------------------- - - // Sizes <= 1024 have an alignment >= 8. So for such sizes we have an - // array indexed by ceil(size/8). Sizes > 1024 have an alignment >= 128. - // So for these larger sizes we have an array indexed by ceil(size/128). - // - // We flatten both logical arrays into one physical array and use - // arithmetic to compute an appropriate index. The constants used by - // ClassIndex() were selected to make the flattening work. - // - // Examples: - // Size Expression Index - // ------------------------------------------------------- - // 0 (0 + 7) / 8 0 - // 1 (1 + 7) / 8 1 - // ... - // 1024 (1024 + 7) / 8 128 - // 1025 (1025 + 127 + (120<<7)) / 128 129 - // ... - // 32768 (32768 + 127 + (120<<7)) / 128 376 - static const int kMaxSmallSize = 1024; - static const size_t kClassArraySize = - ((kMaxSize + 127 + (120 << 7)) >> 7) + 1; - unsigned char class_array_[kClassArraySize]; - - static inline size_t SmallSizeClass(size_t s) { - return (static_cast(s) + 7) >> 3; - } - - static inline size_t LargeSizeClass(size_t s) { - return (static_cast(s) + 127 + (120 << 7)) >> 7; - } - - // Compute index of the class_array[] entry for a given size - static inline size_t ClassIndex(size_t s) { - // Use unsigned arithmetic to avoid unnecessary sign extensions. - ASSERT(0 <= s); - ASSERT(s <= kMaxSize); - if (LIKELY(s <= kMaxSmallSize)) { - return SmallSizeClass(s); - } else { - return LargeSizeClass(s); - } - } - - int NumMoveSize(size_t size); - - // Mapping from size class to max size storable in that class - size_t class_to_size_[kNumClasses]; - - // Mapping from size class to number of pages to allocate at a time - size_t class_to_pages_[kNumClasses]; - - public: - // Constructor should do nothing since we rely on explicit Init() - // call, which may or may not be called before the constructor runs. - SizeMap() { } - - // Initialize the mapping arrays - void Init(); - - inline int SizeClass(size_t size) { - return class_array_[ClassIndex(size)]; - } - - inline bool MaybeSizeClass(size_t size, size_t *size_class) { - size_t class_idx; - if (LIKELY(size <= kMaxSmallSize)) { - class_idx = SmallSizeClass(size); - } else if (size <= kMaxSize) { - class_idx = LargeSizeClass(size); - } else { - return false; - } - *size_class = class_array_[class_idx]; - return true; - } - - // Get the byte-size for a specified class - inline size_t ByteSizeForClass(size_t cl) { - return class_to_size_[cl]; - } - - // Mapping from size class to max size storable in that class - inline size_t class_to_size(size_t cl) { - return class_to_size_[cl]; - } - - // Mapping from size class to number of pages to allocate at a time - inline size_t class_to_pages(size_t cl) { - return class_to_pages_[cl]; - } - - // Number of objects to move between a per-thread list and a central - // list in one shot. We want this to be not too small so we can - // amortize the lock overhead for accessing the central list. Making - // it too big may temporarily cause unnecessary memory wastage in the - // per-thread free list until the scavenger cleans up the list. - inline int num_objects_to_move(size_t cl) { - return num_objects_to_move_[cl]; - } -}; - -// Allocates "bytes" worth of memory and returns it. Increments -// metadata_system_bytes appropriately. May return NULL if allocation -// fails. Requires pageheap_lock is held. -void* MetaDataAlloc(size_t bytes); - -// Returns the total number of bytes allocated from the system. -// Requires pageheap_lock is held. -uint64_t metadata_system_bytes(); - -// size/depth are made the same size as a pointer so that some generic -// code below can conveniently cast them back and forth to void*. -static const int kMaxStackDepth = 31; -struct StackTrace { - uintptr_t size; // Size of object - uintptr_t depth; // Number of PC values stored in array below - void* stack[kMaxStackDepth]; -}; - -} // namespace tcmalloc - -#endif // TCMALLOC_COMMON_H_ diff --git a/contrib/libtcmalloc/src/config.h b/contrib/libtcmalloc/src/config.h deleted file mode 100644 index 9f9a7a259e5..00000000000 --- a/contrib/libtcmalloc/src/config.h +++ /dev/null @@ -1,323 +0,0 @@ -/* src/config.h. Generated from config.h.in by configure. */ -/* src/config.h.in. Generated from configure.ac by autoheader. */ - - -#ifndef GPERFTOOLS_CONFIG_H_ -#define GPERFTOOLS_CONFIG_H_ - - -/* Build runtime detection for sized delete */ -/* #undef ENABLE_DYNAMIC_SIZED_DELETE */ - -/* Build sized deletion operators */ -/* #undef ENABLE_SIZED_DELETE */ - -/* Define to 1 if compiler supports __builtin_expect */ -#if _MSC_VER -#define HAVE_BUILTIN_EXPECT 0 -#else -#define HAVE_BUILTIN_EXPECT 1 -#endif - -/* Define to 1 if compiler supports __builtin_stack_pointer */ -/* #undef HAVE_BUILTIN_STACK_POINTER */ - -/* Define to 1 if you have the header file. */ -/* #undef HAVE_CONFLICT_SIGNAL_H */ - -/* Define to 1 if you have the header file. */ -/* #undef HAVE_CYGWIN_SIGNAL_H */ - -/* Define to 1 if you have the declaration of `backtrace', and to 0 if you - don't. */ -/* #undef HAVE_DECL_BACKTRACE */ - -/* Define to 1 if you have the declaration of `cfree', and to 0 if you don't. - */ -#define HAVE_DECL_CFREE 1 - -/* Define to 1 if you have the declaration of `memalign', and to 0 if you - don't. */ -#define HAVE_DECL_MEMALIGN 1 - -/* Define to 1 if you have the declaration of `nanosleep', and to 0 if you - don't. */ -/* #undef HAVE_DECL_NANOSLEEP */ - -/* Define to 1 if you have the declaration of `posix_memalign', and to 0 if - you don't. */ -#define HAVE_DECL_POSIX_MEMALIGN 1 - -/* Define to 1 if you have the declaration of `pvalloc', and to 0 if you - don't. */ -#define HAVE_DECL_PVALLOC 1 - -/* Define to 1 if you have the declaration of `sleep', and to 0 if you don't. - */ -/* #undef HAVE_DECL_SLEEP */ - -/* Define to 1 if you have the declaration of `uname', and to 0 if you don't. - */ -#define HAVE_DECL_UNAME 1 - -/* Define to 1 if you have the declaration of `valloc', and to 0 if you don't. - */ -#define HAVE_DECL_VALLOC 1 - -/* Define to 1 if you have the header file. */ -#define HAVE_DLFCN_H 1 - -/* Define to 1 if the system has the type `Elf32_Versym'. */ -#define HAVE_ELF32_VERSYM 1 - -/* Define to 1 if you have the header file. */ -#define HAVE_EXECINFO_H 1 - -/* Define to 1 if you have the header file. */ -#define HAVE_FCNTL_H 1 - -/* Define to 1 if you have the header file. */ -#if !defined(__APPLE__) && !defined(__FreeBSD__) -#define HAVE_FEATURES_H 1 -#endif - -/* Define to 1 if you have the `fork' function. */ -#define HAVE_FORK 1 - -/* Define to 1 if you have the `geteuid' function. */ -#define HAVE_GETEUID 1 - -/* Define to 1 if you have the `getpagesize' function. */ -#define HAVE_GETPAGESIZE 1 - -/* Define to 1 if you have the header file. */ -#define HAVE_GLOB_H 1 - -/* Define to 1 if you have the header file. */ -#define HAVE_GRP_H 1 - -/* Define to 1 if you have the header file. */ -#define HAVE_INTTYPES_H 1 - -/* Define to 1 if you have the header file. */ -//#define HAVE_LIBUNWIND_H 1 - -/* Define to 1 if you have the header file. */ -#if !defined(__APPLE__) && !defined(__FreeBSD__) -#define HAVE_LINUX_PTRACE_H 1 -#endif - -/* Define if this is Linux that has SIGEV_THREAD_ID */ -#define HAVE_LINUX_SIGEV_THREAD_ID 1 - -/* Define to 1 if you have the header file. */ -#if !defined(__FreeBSD__) -#define HAVE_MALLOC_H 1 -#endif - -/* Define to 1 if you have the header file. */ -#define HAVE_MEMORY_H 1 - -/* Define to 1 if you have a working `mmap' system call. */ -#define HAVE_MMAP 1 - -/* define if the compiler implements namespaces */ -#define HAVE_NAMESPACES 1 - -/* Define to 1 if you have the header file. */ -#define HAVE_POLL_H 1 - -/* define if libc has program_invocation_name */ -#if !defined(__APPLE__) && !defined(__FreeBSD__) -#define HAVE_PROGRAM_INVOCATION_NAME 1 -#endif - -/* Define if you have POSIX threads libraries and header files. */ -#define HAVE_PTHREAD 1 - -/* defined to 1 if pthread symbols are exposed even without include pthread.h - */ -/* #undef HAVE_PTHREAD_DESPITE_ASKING_FOR */ - -/* Define to 1 if you have the header file. */ -#define HAVE_PWD_H 1 - -/* Define to 1 if you have the `sbrk' function. */ -#define HAVE_SBRK 1 - -/* Define to 1 if you have the header file. */ -#define HAVE_SCHED_H 1 - -/* Define to 1 if you have the header file. */ -#define HAVE_STDINT_H 1 - -/* Define to 1 if you have the header file. */ -#define HAVE_STDLIB_H 1 - -/* Define to 1 if you have the header file. */ -#define HAVE_STRINGS_H 1 - -/* Define to 1 if you have the header file. */ -#define HAVE_STRING_H 1 - -/* Define to 1 if the system has the type `struct mallinfo'. */ -//#if !defined(__APPLE__) && !defined(__FreeBSD__) -#if !defined(__APPLE__) -#define HAVE_STRUCT_MALLINFO 1 -#endif - -/* Define to 1 if you have the header file. */ -#define HAVE_SYS_CDEFS_H 1 - -/* Define to 1 if you have the header file. */ -#define HAVE_SYS_PARAM_H 1 - -/* Define to 1 if you have the header file. */ -#define HAVE_SYS_PRCTL_H 1 - -/* Define to 1 if you have the header file. */ -#define HAVE_SYS_RESOURCE_H 1 - -/* Define to 1 if you have the header file. */ -#define HAVE_SYS_SOCKET_H 1 - -/* Define to 1 if you have the header file. */ -#define HAVE_SYS_STAT_H 1 - -/* Define to 1 if you have the header file. */ -#define HAVE_SYS_SYSCALL_H 1 - -/* Define to 1 if you have the header file. */ -#define HAVE_SYS_TYPES_H 1 - -/* Define to 1 if you have the header file. */ -/* #undef HAVE_SYS_UCONTEXT_H */ - -/* Define to 1 if you have the header file. */ -#define HAVE_SYS_WAIT_H 1 - -/* Define to 1 if compiler supports __thread */ -#define HAVE_TLS 1 - -/* Define to 1 if you have the header file. */ -/* #undef HAVE_UCONTEXT_H */ - -/* Define to 1 if you have the header file. */ -#define HAVE_UNISTD_H 1 - -/* Whether contains _Unwind_Backtrace */ -#define HAVE_UNWIND_BACKTRACE 1 - -/* Define to 1 if you have the header file. */ -#define HAVE_UNWIND_H 1 - -/* Define to 1 if you have the header file. */ -/* #undef HAVE_VALGRIND_H */ - -/* define if your compiler has __attribute__ */ -#define HAVE___ATTRIBUTE__ 1 - -/* Define to 1 if compiler supports __environ */ -#if !defined(__APPLE__) && !defined(__FreeBSD__) -#define HAVE___ENVIRON 1 -#endif - -/* Define to 1 if the system has the type `__int64'. */ -/* #undef HAVE___INT64 */ - -/* prefix where we look for installed files */ -#define INSTALL_PREFIX "/usr/local" - -/* Define to 1 if int32_t is equivalent to intptr_t */ -/* #undef INT32_EQUALS_INTPTR */ - -/* Define to the sub-directory in which libtool stores uninstalled libraries. - */ -#define LT_OBJDIR ".libs/" - -/* Name of package */ -#define PACKAGE "gperftools" - -/* Define to the address where bug reports for this package should be sent. */ -#define PACKAGE_BUGREPORT "gperftools@googlegroups.com" - -/* Define to the full name of this package. */ -#define PACKAGE_NAME "gperftools" - -/* Define to the full name and version of this package. */ -#define PACKAGE_STRING "gperftools 2.5" - -/* Define to the one symbol short name of this package. */ -#define PACKAGE_TARNAME "gperftools" - -/* Define to the home page for this package. */ -#define PACKAGE_URL "" - -/* Define to the version of this package. */ -#define PACKAGE_VERSION "2.5" - -/* How to access the PC from a struct ucontext */ -/* #undef PC_FROM_UCONTEXT */ - -/* Always the empty-string on non-windows systems. On windows, should be - "__declspec(dllexport)". This way, when we compile the dll, we export our - functions/classes. It's safe to define this here because config.h is only - used internally, to compile the DLL, and every DLL source file #includes - "config.h" before anything else. */ -#define PERFTOOLS_DLL_DECL /**/ - -/* printf format code for printing a size_t and ssize_t */ -#define PRIdS "ld" - -/* printf format code for printing a size_t and ssize_t */ -#define PRIuS "lu" - -/* printf format code for printing a size_t and ssize_t */ -#define PRIxS "lx" - -/* Mark the systems where we know it's bad if pthreads runs too - early before main (before threads are initialized, presumably). */ -#ifdef __FreeBSD__ -#define PTHREADS_CRASHES_IF_RUN_TOO_EARLY 1 -#endif - -/* Define to necessary symbol if this constant uses a non-standard name on - your system. */ -/* #undef PTHREAD_CREATE_JOINABLE */ - -/* Define to 1 if you have the ANSI C header files. */ -#define STDC_HEADERS 1 - -/* the namespace where STL code like vector<> is defined */ -#define STL_NAMESPACE std - -/* Define 32K of internal pages size for tcmalloc */ -/* #undef TCMALLOC_32K_PAGES */ - -/* Define 64K of internal pages size for tcmalloc */ -/* #undef TCMALLOC_64K_PAGES */ - -/* Define 8 bytes of allocation alignment for tcmalloc */ -/* #undef TCMALLOC_ALIGN_8BYTES */ - -/* Version number of package */ -#define VERSION "2.5" - -/* C99 says: define this to get the PRI... macros from stdint.h */ -#ifndef __STDC_FORMAT_MACROS -# define __STDC_FORMAT_MACROS 1 -#endif - -/* Define to `__inline__' or `__inline' if that's what the C compiler - calls it, or to nothing if 'inline' is not supported under any name. */ -#ifndef __cplusplus -/* #undef inline */ -#endif - - -#ifdef __MINGW32__ -#include "windows/mingw.h" -#endif - -#endif /* #ifndef GPERFTOOLS_CONFIG_H_ */ - diff --git a/contrib/libtcmalloc/src/debugallocation.cc b/contrib/libtcmalloc/src/debugallocation.cc deleted file mode 100644 index 178809bc8a3..00000000000 --- a/contrib/libtcmalloc/src/debugallocation.cc +++ /dev/null @@ -1,1500 +0,0 @@ -// -*- Mode: C++; c-basic-offset: 2; indent-tabs-mode: nil -*- -// Copyright (c) 2000, Google Inc. -// All rights reserved. -// -// Redistribution and use in source and binary forms, with or without -// modification, are permitted provided that the following conditions are -// met: -// -// * Redistributions of source code must retain the above copyright -// notice, this list of conditions and the following disclaimer. -// * Redistributions in binary form must reproduce the above -// copyright notice, this list of conditions and the following disclaimer -// in the documentation and/or other materials provided with the -// distribution. -// * Neither the name of Google Inc. nor the names of its -// contributors may be used to endorse or promote products derived from -// this software without specific prior written permission. -// -// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS -// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT -// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR -// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT -// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, -// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT -// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, -// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY -// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT -// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE -// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. - -// --- -// Author: Urs Holzle - -#include "config.h" -#include -#ifdef HAVE_FCNTL_H -#include -#endif -#ifdef HAVE_INTTYPES_H -#include -#endif -// We only need malloc.h for struct mallinfo. -#ifdef HAVE_STRUCT_MALLINFO -// Malloc can be in several places on older versions of OS X. -# if defined(HAVE_MALLOC_H) -# include -# elif defined(HAVE_MALLOC_MALLOC_H) -# include -# elif defined(HAVE_SYS_MALLOC_H) -# include -# endif -#endif -#ifdef HAVE_PTHREAD -#include -#endif -#include -#include -#include -#ifdef HAVE_MMAP -#include -#endif -#include -#include -#ifdef HAVE_UNISTD_H -#include -#endif - -#include -#include -#include -#include "addressmap-inl.h" -#include "base/commandlineflags.h" -#include "base/googleinit.h" -#include "base/logging.h" -#include "base/spinlock.h" -#include "malloc_hook-inl.h" -#include "symbolize.h" - -// NOTE: due to #define below, tcmalloc.cc will omit tc_XXX -// definitions. So that debug implementations can be defined -// instead. We're going to use do_malloc, do_free and other do_XXX -// functions that are defined in tcmalloc.cc for actual memory -// management -#define TCMALLOC_USING_DEBUGALLOCATION -#include "tcmalloc.cc" - -// __THROW is defined in glibc systems. It means, counter-intuitively, -// "This function will never throw an exception." It's an optional -// optimization tool, but we may need to use it to match glibc prototypes. -#ifndef __THROW // I guess we're not on a glibc system -# define __THROW // __THROW is just an optimization, so ok to make it "" -#endif - -// On systems (like freebsd) that don't define MAP_ANONYMOUS, use the old -// form of the name instead. -#ifndef MAP_ANONYMOUS -# define MAP_ANONYMOUS MAP_ANON -#endif - -#pragma GCC diagnostic push -#ifdef __clang__ -#pragma GCC diagnostic ignored "-Wunused-private-field" -#pragma GCC diagnostic ignored "-Wgnu-alignof-expression" -#endif - -// ========================================================================= // - -DEFINE_bool(malloctrace, - EnvToBool("TCMALLOC_TRACE", false), - "Enables memory (de)allocation tracing to /tmp/google.alloc."); -#ifdef HAVE_MMAP -DEFINE_bool(malloc_page_fence, - EnvToBool("TCMALLOC_PAGE_FENCE", false), - "Enables putting of memory allocations at page boundaries " - "with a guard page following the allocation (to catch buffer " - "overruns right when they happen)."); -DEFINE_bool(malloc_page_fence_never_reclaim, - EnvToBool("TCMALLOC_PAGE_FRANCE_NEVER_RECLAIM", false), - "Enables making the virtual address space inaccessible " - "upon a deallocation instead of returning it and reusing later."); -#else -DEFINE_bool(malloc_page_fence, false, "Not usable (requires mmap)"); -DEFINE_bool(malloc_page_fence_never_reclaim, false, "Not usable (required mmap)"); -#endif -DEFINE_bool(malloc_reclaim_memory, - EnvToBool("TCMALLOC_RECLAIM_MEMORY", true), - "If set to false, we never return memory to malloc " - "when an object is deallocated. This ensures that all " - "heap object addresses are unique."); -DEFINE_int32(max_free_queue_size, - EnvToInt("TCMALLOC_MAX_FREE_QUEUE_SIZE", 10*1024*1024), - "If greater than 0, keep freed blocks in a queue instead of " - "releasing them to the allocator immediately. Release them when " - "the total size of all blocks in the queue would otherwise exceed " - "this limit."); - -DEFINE_bool(symbolize_stacktrace, - EnvToBool("TCMALLOC_SYMBOLIZE_STACKTRACE", true), - "Symbolize the stack trace when provided (on some error exits)"); - -// If we are LD_PRELOAD-ed against a non-pthreads app, then -// pthread_once won't be defined. We declare it here, for that -// case (with weak linkage) which will cause the non-definition to -// resolve to NULL. We can then check for NULL or not in Instance. -extern "C" int pthread_once(pthread_once_t *, void (*)(void)) - ATTRIBUTE_WEAK; - -// ========================================================================= // - -// A safe version of printf() that does not do any allocation and -// uses very little stack space. -static void TracePrintf(int fd, const char *fmt, ...) - __attribute__ ((__format__ (__printf__, 2, 3))); - -// Round "value" up to next "alignment" boundary. -// Requires that "alignment" be a power of two. -static intptr_t RoundUp(intptr_t value, intptr_t alignment) { - return (value + alignment - 1) & ~(alignment - 1); -} - -// ========================================================================= // - -class MallocBlock; - -// A circular buffer to hold freed blocks of memory. MallocBlock::Deallocate -// (below) pushes blocks into this queue instead of returning them to the -// underlying allocator immediately. See MallocBlock::Deallocate for more -// information. -// -// We can't use an STL class for this because we need to be careful not to -// perform any heap de-allocations in any of the code in this class, since the -// code in MallocBlock::Deallocate is not re-entrant. -template -class FreeQueue { - public: - FreeQueue() : q_front_(0), q_back_(0) {} - - bool Full() { - return (q_front_ + 1) % kFreeQueueSize == q_back_; - } - - void Push(const QueueEntry& block) { - q_[q_front_] = block; - q_front_ = (q_front_ + 1) % kFreeQueueSize; - } - - QueueEntry Pop() { - RAW_CHECK(q_back_ != q_front_, "Queue is empty"); - const QueueEntry& ret = q_[q_back_]; - q_back_ = (q_back_ + 1) % kFreeQueueSize; - return ret; - } - - size_t size() const { - return (q_front_ - q_back_ + kFreeQueueSize) % kFreeQueueSize; - } - - private: - // Maximum number of blocks kept in the free queue before being freed. - static const int kFreeQueueSize = 1024; - - QueueEntry q_[kFreeQueueSize]; - int q_front_; - int q_back_; -}; - -struct MallocBlockQueueEntry { - MallocBlockQueueEntry() : block(NULL), size(0), - num_deleter_pcs(0), deleter_threadid(0) {} - MallocBlockQueueEntry(MallocBlock* b, size_t s) : block(b), size(s) { - if (FLAGS_max_free_queue_size != 0 && b != NULL) { - // Adjust the number of frames to skip (4) if you change the - // location of this call. - num_deleter_pcs = - GetStackTrace(deleter_pcs, - sizeof(deleter_pcs) / sizeof(deleter_pcs[0]), - 4); - deleter_threadid = pthread_self(); - } else { - num_deleter_pcs = 0; - // Zero is an illegal pthread id by my reading of the pthread - // implementation: - deleter_threadid = 0; - } - } - - MallocBlock* block; - size_t size; - - // When deleted and put in the free queue, we (flag-controlled) - // record the stack so that if corruption is later found, we can - // print the deleter's stack. (These three vars add 144 bytes of - // overhead under the LP64 data model.) - void* deleter_pcs[16]; - int num_deleter_pcs; - pthread_t deleter_threadid; -}; - -class MallocBlock { - public: // allocation type constants - - // Different allocation types we distinguish. - // Note: The lower 4 bits are not random: we index kAllocName array - // by these values masked with kAllocTypeMask; - // the rest are "random" magic bits to help catch memory corruption. - static const int kMallocType = 0xEFCDAB90; - static const int kNewType = 0xFEBADC81; - static const int kArrayNewType = 0xBCEADF72; - - private: // constants - - // A mask used on alloc types above to get to 0, 1, 2 - static const int kAllocTypeMask = 0x3; - // An additional bit to set in AllocType constants - // to mark now deallocated regions. - static const int kDeallocatedTypeBit = 0x4; - - // For better memory debugging, we initialize all storage to known - // values, and overwrite the storage when it's deallocated: - // Byte that fills uninitialized storage. - static const int kMagicUninitializedByte = 0xAB; - // Byte that fills deallocated storage. - // NOTE: tcmalloc.cc depends on the value of kMagicDeletedByte - // to work around a bug in the pthread library. - static const int kMagicDeletedByte = 0xCD; - // A size_t (type of alloc_type_ below) in a deallocated storage - // filled with kMagicDeletedByte. - static const size_t kMagicDeletedSizeT = - 0xCDCDCDCD | (((size_t)0xCDCDCDCD << 16) << 16); - // Initializer works for 32 and 64 bit size_ts; - // "<< 16 << 16" is to fool gcc from issuing a warning - // when size_ts are 32 bits. - - // NOTE: on Linux, you can enable malloc debugging support in libc by - // setting the environment variable MALLOC_CHECK_ to 1 before you - // start the program (see man malloc). - - // We use either do_malloc or mmap to make the actual allocation. In - // order to remember which one of the two was used for any block, we store an - // appropriate magic word next to the block. - static const size_t kMagicMalloc = 0xDEADBEEF; - static const size_t kMagicMMap = 0xABCDEFAB; - - // This array will be filled with 0xCD, for use with memcmp. - static unsigned char kMagicDeletedBuffer[1024]; - static pthread_once_t deleted_buffer_initialized_; - static bool deleted_buffer_initialized_no_pthreads_; - - private: // data layout - - // The four fields size1_,offset_,magic1_,alloc_type_ - // should together occupy a multiple of 16 bytes. (At the - // moment, sizeof(size_t) == 4 or 8 depending on piii vs - // k8, and 4 of those sum to 16 or 32 bytes). - // This, combined with do_malloc's alignment guarantees, - // ensures that SSE types can be stored into the returned - // block, at &size2_. - size_t size1_; - size_t offset_; // normally 0 unless memaligned memory - // see comments in memalign() and FromRawPointer(). - size_t magic1_; - size_t alloc_type_; - // here comes the actual data (variable length) - // ... - // then come the size2_ and magic2_, or a full page of mprotect-ed memory - // if the malloc_page_fence feature is enabled. - size_t size2_; - size_t magic2_; - - private: // static data and helpers - - // Allocation map: stores the allocation type for each allocated object, - // or the type or'ed with kDeallocatedTypeBit - // for each formerly allocated object. - typedef AddressMap AllocMap; - static AllocMap* alloc_map_; - // This protects alloc_map_ and consistent state of metadata - // for each still-allocated object in it. - // We use spin locks instead of pthread_mutex_t locks - // to prevent crashes via calls to pthread_mutex_(un)lock - // for the (de)allocations coming from pthreads initialization itself. - static SpinLock alloc_map_lock_; - - // A queue of freed blocks. Instead of releasing blocks to the allocator - // immediately, we put them in a queue, freeing them only when necessary - // to keep the total size of all the freed blocks below the limit set by - // FLAGS_max_free_queue_size. - static FreeQueue* free_queue_; - - static size_t free_queue_size_; // total size of blocks in free_queue_ - // protects free_queue_ and free_queue_size_ - static SpinLock free_queue_lock_; - - // Names of allocation types (kMallocType, kNewType, kArrayNewType) - static const char* const kAllocName[]; - // Names of corresponding deallocation types - static const char* const kDeallocName[]; - - static const char* AllocName(int type) { - return kAllocName[type & kAllocTypeMask]; - } - - static const char* DeallocName(int type) { - return kDeallocName[type & kAllocTypeMask]; - } - - private: // helper accessors - - bool IsMMapped() const { return kMagicMMap == magic1_; } - - bool IsValidMagicValue(size_t value) const { - return kMagicMMap == value || kMagicMalloc == value; - } - - static size_t real_malloced_size(size_t size) { - return size + sizeof(MallocBlock); - } - - /* - * Here we assume size of page is kMinAlign aligned, - * so if size is MALLOC_ALIGNMENT aligned too, then we could - * guarantee return address is also kMinAlign aligned, because - * mmap return address at nearby page boundary on Linux. - */ - static size_t real_mmapped_size(size_t size) { - size_t tmp = size + MallocBlock::data_offset(); - tmp = RoundUp(tmp, kMinAlign); - return tmp; - } - - size_t real_size() { - return IsMMapped() ? real_mmapped_size(size1_) : real_malloced_size(size1_); - } - - // NOTE: if the block is mmapped (that is, we're using the - // malloc_page_fence option) then there's no size2 or magic2 - // (instead, the guard page begins where size2 would be). - - size_t* size2_addr() { return (size_t*)((char*)&size2_ + size1_); } - const size_t* size2_addr() const { - return (const size_t*)((char*)&size2_ + size1_); - } - - size_t* magic2_addr() { return (size_t*)(size2_addr() + 1); } - const size_t* magic2_addr() const { return (const size_t*)(size2_addr() + 1); } - - private: // other helpers - - void Initialize(size_t size, int type) { - RAW_CHECK(IsValidMagicValue(magic1_), ""); - // record us as allocated in the map - alloc_map_lock_.Lock(); - if (!alloc_map_) { - void* p = do_malloc(sizeof(AllocMap)); - alloc_map_ = new(p) AllocMap(do_malloc, do_free); - } - alloc_map_->Insert(data_addr(), type); - // initialize us - size1_ = size; - offset_ = 0; - alloc_type_ = type; - if (!IsMMapped()) { - bit_store(magic2_addr(), &magic1_); - bit_store(size2_addr(), &size); - } - alloc_map_lock_.Unlock(); - memset(data_addr(), kMagicUninitializedByte, size); - if (!IsMMapped()) { - RAW_CHECK(memcmp(&size1_, size2_addr(), sizeof(size1_)) == 0, "should hold"); - RAW_CHECK(memcmp(&magic1_, magic2_addr(), sizeof(magic1_)) == 0, "should hold"); - } - } - - size_t CheckAndClear(int type, size_t given_size) { - alloc_map_lock_.Lock(); - CheckLocked(type); - if (!IsMMapped()) { - RAW_CHECK(memcmp(&size1_, size2_addr(), sizeof(size1_)) == 0, "should hold"); - } - // record us as deallocated in the map - alloc_map_->Insert(data_addr(), type | kDeallocatedTypeBit); - alloc_map_lock_.Unlock(); - // clear us - const size_t size = real_size(); - RAW_CHECK(!given_size || given_size == size1_, - "right size must be passed to sized delete"); - memset(this, kMagicDeletedByte, size); - return size; - } - - void CheckLocked(int type) const { - int map_type = 0; - const int* found_type = - alloc_map_ != NULL ? alloc_map_->Find(data_addr()) : NULL; - if (found_type == NULL) { - RAW_LOG(FATAL, "memory allocation bug: object at %p " - "has never been allocated", data_addr()); - } else { - map_type = *found_type; - } - if ((map_type & kDeallocatedTypeBit) != 0) { - RAW_LOG(FATAL, "memory allocation bug: object at %p " - "has been already deallocated (it was allocated with %s)", - data_addr(), AllocName(map_type & ~kDeallocatedTypeBit)); - } - if (alloc_type_ == kMagicDeletedSizeT) { - RAW_LOG(FATAL, "memory stomping bug: a word before object at %p " - "has been corrupted; or else the object has been already " - "deallocated and our memory map has been corrupted", - data_addr()); - } - if (!IsValidMagicValue(magic1_)) { - RAW_LOG(FATAL, "memory stomping bug: a word before object at %p " - "has been corrupted; " - "or else our memory map has been corrupted and this is a " - "deallocation for not (currently) heap-allocated object", - data_addr()); - } - if (!IsMMapped()) { - if (memcmp(&size1_, size2_addr(), sizeof(size1_))) { - RAW_LOG(FATAL, "memory stomping bug: a word after object at %p " - "has been corrupted", data_addr()); - } - size_t addr; - bit_store(&addr, magic2_addr()); - if (!IsValidMagicValue(addr)) { - RAW_LOG(FATAL, "memory stomping bug: a word after object at %p " - "has been corrupted", data_addr()); - } - } - if (alloc_type_ != type) { - if ((alloc_type_ != MallocBlock::kMallocType) && - (alloc_type_ != MallocBlock::kNewType) && - (alloc_type_ != MallocBlock::kArrayNewType)) { - RAW_LOG(FATAL, "memory stomping bug: a word before object at %p " - "has been corrupted", data_addr()); - } - RAW_LOG(FATAL, "memory allocation/deallocation mismatch at %p: " - "allocated with %s being deallocated with %s", - data_addr(), AllocName(alloc_type_), DeallocName(type)); - } - if (alloc_type_ != map_type) { - RAW_LOG(FATAL, "memory stomping bug: our memory map has been corrupted : " - "allocation at %p made with %s " - "is recorded in the map to be made with %s", - data_addr(), AllocName(alloc_type_), AllocName(map_type)); - } - } - - public: // public accessors - - void* data_addr() { return (void*)&size2_; } - const void* data_addr() const { return (const void*)&size2_; } - - static size_t data_offset() { return OFFSETOF_MEMBER(MallocBlock, size2_); } - - size_t data_size() const { return size1_; } - - void set_offset(int offset) { this->offset_ = offset; } - - public: // our main interface - - static MallocBlock* Allocate(size_t size, int type) { - // Prevent an integer overflow / crash with large allocation sizes. - // TODO - Note that for a e.g. 64-bit size_t, max_size_t may not actually - // be the maximum value, depending on how the compiler treats ~0. The worst - // practical effect is that allocations are limited to 4Gb or so, even if - // the address space could take more. - static size_t max_size_t = ~0; - if (size > max_size_t - sizeof(MallocBlock)) { - RAW_LOG(ERROR, "Massive size passed to malloc: %" PRIuS "", size); - return NULL; - } - MallocBlock* b = NULL; - const bool use_malloc_page_fence = FLAGS_malloc_page_fence; -#ifdef HAVE_MMAP - if (use_malloc_page_fence) { - // Put the block towards the end of the page and make the next page - // inaccessible. This will catch buffer overrun right when it happens. - size_t sz = real_mmapped_size(size); - int pagesize = getpagesize(); - int num_pages = (sz + pagesize - 1) / pagesize + 1; - char* p = (char*) mmap(NULL, num_pages * pagesize, PROT_READ|PROT_WRITE, - MAP_PRIVATE|MAP_ANONYMOUS, -1, 0); - if (p == MAP_FAILED) { - // If the allocation fails, abort rather than returning NULL to - // malloc. This is because in most cases, the program will run out - // of memory in this mode due to tremendous amount of wastage. There - // is no point in propagating the error elsewhere. - RAW_LOG(FATAL, "Out of memory: possibly due to page fence overhead: %s", - strerror(errno)); - } - // Mark the page after the block inaccessible - if (mprotect(p + (num_pages - 1) * pagesize, pagesize, PROT_NONE)) { - RAW_LOG(FATAL, "Guard page setup failed: %s", strerror(errno)); - } - b = (MallocBlock*) (p + (num_pages - 1) * pagesize - sz); - } else { - b = (MallocBlock*) do_malloc(real_malloced_size(size)); - } -#else - b = (MallocBlock*) do_malloc(real_malloced_size(size)); -#endif - - // It would be nice to output a diagnostic on allocation failure - // here, but logging (other than FATAL) requires allocating - // memory, which could trigger a nasty recursion. Instead, preserve - // malloc semantics and return NULL on failure. - if (b != NULL) { - b->magic1_ = use_malloc_page_fence ? kMagicMMap : kMagicMalloc; - b->Initialize(size, type); - } - return b; - } - - void Deallocate(int type, size_t given_size) { - if (IsMMapped()) { // have to do this before CheckAndClear -#ifdef HAVE_MMAP - int size = CheckAndClear(type, given_size); - int pagesize = getpagesize(); - int num_pages = (size + pagesize - 1) / pagesize + 1; - char* p = (char*) this; - if (FLAGS_malloc_page_fence_never_reclaim || - !FLAGS_malloc_reclaim_memory) { - mprotect(p - (num_pages - 1) * pagesize + size, - num_pages * pagesize, PROT_NONE); - } else { - munmap(p - (num_pages - 1) * pagesize + size, num_pages * pagesize); - } -#endif - } else { - const size_t size = CheckAndClear(type, given_size); - if (FLAGS_malloc_reclaim_memory) { - // Instead of freeing the block immediately, push it onto a queue of - // recently freed blocks. Free only enough blocks to keep from - // exceeding the capacity of the queue or causing the total amount of - // un-released memory in the queue from exceeding - // FLAGS_max_free_queue_size. - ProcessFreeQueue(this, size, FLAGS_max_free_queue_size); - } - } - } - - static size_t FreeQueueSize() { - SpinLockHolder l(&free_queue_lock_); - return free_queue_size_; - } - - static void ProcessFreeQueue(MallocBlock* b, size_t size, - int max_free_queue_size) { - // MallocBlockQueueEntry are about 144 in size, so we can only - // use a small array of them on the stack. - MallocBlockQueueEntry entries[4]; - int num_entries = 0; - MallocBlockQueueEntry new_entry(b, size); - free_queue_lock_.Lock(); - if (free_queue_ == NULL) - free_queue_ = new FreeQueue; - RAW_CHECK(!free_queue_->Full(), "Free queue mustn't be full!"); - - if (b != NULL) { - free_queue_size_ += size + sizeof(MallocBlockQueueEntry); - free_queue_->Push(new_entry); - } - - // Free blocks until the total size of unfreed blocks no longer exceeds - // max_free_queue_size, and the free queue has at least one free - // space in it. - while (free_queue_size_ > max_free_queue_size || free_queue_->Full()) { - RAW_CHECK(num_entries < arraysize(entries), "entries array overflow"); - entries[num_entries] = free_queue_->Pop(); - free_queue_size_ -= - entries[num_entries].size + sizeof(MallocBlockQueueEntry); - num_entries++; - if (num_entries == arraysize(entries)) { - // The queue will not be full at this point, so it is ok to - // release the lock. The queue may still contain more than - // max_free_queue_size, but this is not a strict invariant. - free_queue_lock_.Unlock(); - for (int i = 0; i < num_entries; i++) { - CheckForDanglingWrites(entries[i]); - do_free(entries[i].block); - } - num_entries = 0; - free_queue_lock_.Lock(); - } - } - RAW_CHECK(free_queue_size_ >= 0, "Free queue size went negative!"); - free_queue_lock_.Unlock(); - for (int i = 0; i < num_entries; i++) { - CheckForDanglingWrites(entries[i]); - do_free(entries[i].block); - } - } - - static void InitDeletedBuffer() { - memset(kMagicDeletedBuffer, kMagicDeletedByte, sizeof(kMagicDeletedBuffer)); - deleted_buffer_initialized_no_pthreads_ = true; - } - - static void CheckForDanglingWrites(const MallocBlockQueueEntry& queue_entry) { - // Initialize the buffer if necessary. - if (pthread_once) - pthread_once(&deleted_buffer_initialized_, &InitDeletedBuffer); - if (!deleted_buffer_initialized_no_pthreads_) { - // This will be the case on systems that don't link in pthreads, - // including on FreeBSD where pthread_once has a non-zero address - // (but doesn't do anything) even when pthreads isn't linked in. - InitDeletedBuffer(); - } - - const unsigned char* p = - reinterpret_cast(queue_entry.block); - - static const size_t size_of_buffer = sizeof(kMagicDeletedBuffer); - const size_t size = queue_entry.size; - const size_t buffers = size / size_of_buffer; - const size_t remainder = size % size_of_buffer; - size_t buffer_idx; - for (buffer_idx = 0; buffer_idx < buffers; ++buffer_idx) { - CheckForCorruptedBuffer(queue_entry, buffer_idx, p, size_of_buffer); - p += size_of_buffer; - } - CheckForCorruptedBuffer(queue_entry, buffer_idx, p, remainder); - } - - static void CheckForCorruptedBuffer(const MallocBlockQueueEntry& queue_entry, - size_t buffer_idx, - const unsigned char* buffer, - size_t size_of_buffer) { - if (memcmp(buffer, kMagicDeletedBuffer, size_of_buffer) == 0) { - return; - } - - RAW_LOG(ERROR, - "Found a corrupted memory buffer in MallocBlock (may be offset " - "from user ptr): buffer index: %zd, buffer ptr: %p, size of " - "buffer: %zd", buffer_idx, buffer, size_of_buffer); - - // The magic deleted buffer should only be 1024 bytes, but in case - // this changes, let's put an upper limit on the number of debug - // lines we'll output: - if (size_of_buffer <= 1024) { - for (int i = 0; i < size_of_buffer; ++i) { - if (buffer[i] != kMagicDeletedByte) { - RAW_LOG(ERROR, "Buffer byte %d is 0x%02x (should be 0x%02x).", - i, buffer[i], kMagicDeletedByte); - } - } - } else { - RAW_LOG(ERROR, "Buffer too large to print corruption."); - } - - const MallocBlock* b = queue_entry.block; - const size_t size = queue_entry.size; - if (queue_entry.num_deleter_pcs > 0) { - TracePrintf(STDERR_FILENO, "Deleted by thread %p\n", - reinterpret_cast( - PRINTABLE_PTHREAD(queue_entry.deleter_threadid))); - - // We don't want to allocate or deallocate memory here, so we use - // placement-new. It's ok that we don't destroy this, since we're - // just going to error-exit below anyway. Union is for alignment. - union { void* alignment; char buf[sizeof(SymbolTable)]; } tablebuf; - SymbolTable* symbolization_table = new (tablebuf.buf) SymbolTable; - for (int i = 0; i < queue_entry.num_deleter_pcs; i++) { - // Symbolizes the previous address of pc because pc may be in the - // next function. This may happen when the function ends with - // a call to a function annotated noreturn (e.g. CHECK). - char *pc = reinterpret_cast(queue_entry.deleter_pcs[i]); - symbolization_table->Add(pc - 1); - } - if (FLAGS_symbolize_stacktrace) - symbolization_table->Symbolize(); - for (int i = 0; i < queue_entry.num_deleter_pcs; i++) { - char *pc = reinterpret_cast(queue_entry.deleter_pcs[i]); - TracePrintf(STDERR_FILENO, " @ %p %s\n", - pc, symbolization_table->GetSymbol(pc - 1)); - } - } else { - RAW_LOG(ERROR, - "Skipping the printing of the deleter's stack! Its stack was " - "not found; either the corruption occurred too early in " - "execution to obtain a stack trace or --max_free_queue_size was " - "set to 0."); - } - - RAW_LOG(FATAL, - "Memory was written to after being freed. MallocBlock: %p, user " - "ptr: %p, size: %zd. If you can't find the source of the error, " - "try using ASan (http://code.google.com/p/address-sanitizer/), " - "Valgrind, or Purify, or study the " - "output of the deleter's stack printed above.", - b, b->data_addr(), size); - } - - static MallocBlock* FromRawPointer(void* p) { - const size_t data_offset = MallocBlock::data_offset(); - // Find the header just before client's memory. - MallocBlock *mb = reinterpret_cast( - reinterpret_cast(p) - data_offset); - // If mb->alloc_type_ is kMagicDeletedSizeT, we're not an ok pointer. - if (mb->alloc_type_ == kMagicDeletedSizeT) { - RAW_LOG(FATAL, "memory allocation bug: object at %p has been already" - " deallocated; or else a word before the object has been" - " corrupted (memory stomping bug)", p); - } - // If mb->offset_ is zero (common case), mb is the real header. - // If mb->offset_ is non-zero, this block was allocated by debug - // memallign implementation, and mb->offset_ is the distance - // backwards to the real header from mb, which is a fake header. - if (mb->offset_ == 0) { - return mb; - } - - MallocBlock *main_block = reinterpret_cast( - reinterpret_cast(mb) - mb->offset_); - - if (main_block->offset_ != 0) { - RAW_LOG(FATAL, "memory corruption bug: offset_ field is corrupted." - " Need 0 but got %x", - (unsigned)(main_block->offset_)); - } - if (main_block >= p) { - RAW_LOG(FATAL, "memory corruption bug: offset_ field is corrupted." - " Detected main_block address overflow: %x", - (unsigned)(mb->offset_)); - } - if (main_block->size2_addr() < p) { - RAW_LOG(FATAL, "memory corruption bug: offset_ field is corrupted." - " It points below it's own main_block: %x", - (unsigned)(mb->offset_)); - } - - return main_block; - } - - static const MallocBlock* FromRawPointer(const void* p) { - // const-safe version: we just cast about - return FromRawPointer(const_cast(p)); - } - - void Check(int type) const { - alloc_map_lock_.Lock(); - CheckLocked(type); - alloc_map_lock_.Unlock(); - } - - static bool CheckEverything() { - alloc_map_lock_.Lock(); - if (alloc_map_ != NULL) alloc_map_->Iterate(CheckCallback, 0); - alloc_map_lock_.Unlock(); - return true; // if we get here, we're okay - } - - static bool MemoryStats(int* blocks, size_t* total, - int histogram[kMallocHistogramSize]) { - memset(histogram, 0, kMallocHistogramSize * sizeof(int)); - alloc_map_lock_.Lock(); - stats_blocks_ = 0; - stats_total_ = 0; - stats_histogram_ = histogram; - if (alloc_map_ != NULL) alloc_map_->Iterate(StatsCallback, 0); - *blocks = stats_blocks_; - *total = stats_total_; - alloc_map_lock_.Unlock(); - return true; - } - - private: // helpers for CheckEverything and MemoryStats - - static void CheckCallback(const void* ptr, int* type, int dummy) { - if ((*type & kDeallocatedTypeBit) == 0) { - FromRawPointer(ptr)->CheckLocked(*type); - } - } - - // Accumulation variables for StatsCallback protected by alloc_map_lock_ - static int stats_blocks_; - static size_t stats_total_; - static int* stats_histogram_; - - static void StatsCallback(const void* ptr, int* type, int dummy) { - if ((*type & kDeallocatedTypeBit) == 0) { - const MallocBlock* b = FromRawPointer(ptr); - b->CheckLocked(*type); - ++stats_blocks_; - size_t mysize = b->size1_; - int entry = 0; - stats_total_ += mysize; - while (mysize) { - ++entry; - mysize >>= 1; - } - RAW_CHECK(entry < kMallocHistogramSize, - "kMallocHistogramSize should be at least as large as log2 " - "of the maximum process memory size"); - stats_histogram_[entry] += 1; - } - } -}; - -void DanglingWriteChecker() { - // Clear out the remaining free queue to check for dangling writes. - MallocBlock::ProcessFreeQueue(NULL, 0, 0); -} - -// ========================================================================= // - -const size_t MallocBlock::kMagicMalloc; -const size_t MallocBlock::kMagicMMap; - -MallocBlock::AllocMap* MallocBlock::alloc_map_ = NULL; -SpinLock MallocBlock::alloc_map_lock_(SpinLock::LINKER_INITIALIZED); - -FreeQueue* MallocBlock::free_queue_ = NULL; -size_t MallocBlock::free_queue_size_ = 0; -SpinLock MallocBlock::free_queue_lock_(SpinLock::LINKER_INITIALIZED); - -unsigned char MallocBlock::kMagicDeletedBuffer[1024]; -pthread_once_t MallocBlock::deleted_buffer_initialized_ = PTHREAD_ONCE_INIT; -bool MallocBlock::deleted_buffer_initialized_no_pthreads_ = false; - -const char* const MallocBlock::kAllocName[] = { - "malloc", - "new", - "new []", - NULL, -}; - -const char* const MallocBlock::kDeallocName[] = { - "free", - "delete", - "delete []", - NULL, -}; - -int MallocBlock::stats_blocks_; -size_t MallocBlock::stats_total_; -int* MallocBlock::stats_histogram_; - -// ========================================================================= // - -// The following cut-down version of printf() avoids -// using stdio or ostreams. -// This is to guarantee no recursive calls into -// the allocator and to bound the stack space consumed. (The pthread -// manager thread in linuxthreads has a very small stack, -// so fprintf can't be called.) -static void TracePrintf(int fd, const char *fmt, ...) { - char buf[64]; - int i = 0; - va_list ap; - va_start(ap, fmt); - const char *p = fmt; - char numbuf[25]; - if (fd < 0) { - return; - } - numbuf[sizeof(numbuf)-1] = 0; - while (*p != '\0') { // until end of format string - char *s = &numbuf[sizeof(numbuf)-1]; - if (p[0] == '%' && p[1] != 0) { // handle % formats - int64 l = 0; - unsigned long base = 0; - if (*++p == 's') { // %s - s = va_arg(ap, char *); - } else if (*p == 'l' && p[1] == 'd') { // %ld - l = va_arg(ap, long); - base = 10; - p++; - } else if (*p == 'l' && p[1] == 'u') { // %lu - l = va_arg(ap, unsigned long); - base = 10; - p++; - } else if (*p == 'z' && p[1] == 'u') { // %zu - l = va_arg(ap, size_t); - base = 10; - p++; - } else if (*p == 'u') { // %u - l = va_arg(ap, unsigned int); - base = 10; - } else if (*p == 'd') { // %d - l = va_arg(ap, int); - base = 10; - } else if (*p == 'p') { // %p - l = va_arg(ap, intptr_t); - base = 16; - } else { - write(STDERR_FILENO, "Unimplemented TracePrintf format\n", 33); - write(STDERR_FILENO, p, 2); - write(STDERR_FILENO, "\n", 1); - abort(); - } - p++; - if (base != 0) { - bool minus = (l < 0 && base == 10); - uint64 ul = minus? -l : l; - do { - *--s = "0123456789abcdef"[ul % base]; - ul /= base; - } while (ul != 0); - if (base == 16) { - *--s = 'x'; - *--s = '0'; - } else if (minus) { - *--s = '-'; - } - } - } else { // handle normal characters - *--s = *p++; - } - while (*s != 0) { - if (i == sizeof(buf)) { - write(fd, buf, i); - i = 0; - } - buf[i++] = *s++; - } - } - if (i != 0) { - write(fd, buf, i); - } - va_end(ap); -} - -// Return the file descriptor we're writing a log to -static int TraceFd() { - static int trace_fd = -1; - if (trace_fd == -1) { // Open the trace file on the first call - const char *val = getenv("TCMALLOC_TRACE_FILE"); - bool fallback_to_stderr = false; - if (!val) { - val = "/tmp/google.alloc"; - fallback_to_stderr = true; - } - trace_fd = open(val, O_CREAT|O_TRUNC|O_WRONLY, 0666); - if (trace_fd == -1) { - if (fallback_to_stderr) { - trace_fd = 2; - TracePrintf(trace_fd, "Can't open %s. Logging to stderr.\n", val); - } else { - TracePrintf(2, "Can't open %s. Logging disabled.\n", val); - } - } - // Add a header to the log. - TracePrintf(trace_fd, "Trace started: %lu\n", - static_cast(time(NULL))); - TracePrintf(trace_fd, - "func\tsize\tptr\tthread_id\tstack pcs for tools/symbolize\n"); - } - return trace_fd; -} - -// Print the hex stack dump on a single line. PCs are separated by tabs. -static void TraceStack(void) { - void *pcs[16]; - int n = GetStackTrace(pcs, sizeof(pcs)/sizeof(pcs[0]), 0); - for (int i = 0; i != n; i++) { - TracePrintf(TraceFd(), "\t%p", pcs[i]); - } -} - -// This protects MALLOC_TRACE, to make sure its info is atomically written. -static SpinLock malloc_trace_lock(SpinLock::LINKER_INITIALIZED); - -#define MALLOC_TRACE(name, size, addr) \ - do { \ - if (FLAGS_malloctrace) { \ - SpinLockHolder l(&malloc_trace_lock); \ - TracePrintf(TraceFd(), "%s\t%" PRIuS "\t%p\t%" GPRIuPTHREAD, \ - name, size, addr, PRINTABLE_PTHREAD(pthread_self())); \ - TraceStack(); \ - TracePrintf(TraceFd(), "\n"); \ - } \ - } while (0) - -// ========================================================================= // - -// Write the characters buf[0, ..., size-1] to -// the malloc trace buffer. -// This function is intended for debugging, -// and is not declared in any header file. -// You must insert a declaration of it by hand when you need -// to use it. -void __malloctrace_write(const char *buf, size_t size) { - if (FLAGS_malloctrace) { - write(TraceFd(), buf, size); - } -} - -// ========================================================================= // - -// General debug allocation/deallocation - -static inline void* DebugAllocate(size_t size, int type) { - MallocBlock* ptr = MallocBlock::Allocate(size, type); - if (ptr == NULL) return NULL; - MALLOC_TRACE("malloc", size, ptr->data_addr()); - return ptr->data_addr(); -} - -static inline void DebugDeallocate(void* ptr, int type, size_t given_size) { - MALLOC_TRACE("free", - (ptr != 0 ? MallocBlock::FromRawPointer(ptr)->data_size() : 0), - ptr); - if (ptr) MallocBlock::FromRawPointer(ptr)->Deallocate(type, given_size); -} - -// ========================================================================= // - -// The following functions may be called via MallocExtension::instance() -// for memory verification and statistics. -class DebugMallocImplementation : public TCMallocImplementation { - public: - virtual bool GetNumericProperty(const char* name, size_t* value) { - bool result = TCMallocImplementation::GetNumericProperty(name, value); - if (result && (strcmp(name, "generic.current_allocated_bytes") == 0)) { - // Subtract bytes kept in the free queue - size_t qsize = MallocBlock::FreeQueueSize(); - if (*value >= qsize) { - *value -= qsize; - } - } - return result; - } - - virtual bool VerifyNewMemory(const void* p) { - if (p) MallocBlock::FromRawPointer(p)->Check(MallocBlock::kNewType); - return true; - } - - virtual bool VerifyArrayNewMemory(const void* p) { - if (p) MallocBlock::FromRawPointer(p)->Check(MallocBlock::kArrayNewType); - return true; - } - - virtual bool VerifyMallocMemory(const void* p) { - if (p) MallocBlock::FromRawPointer(p)->Check(MallocBlock::kMallocType); - return true; - } - - virtual bool VerifyAllMemory() { - return MallocBlock::CheckEverything(); - } - - virtual bool MallocMemoryStats(int* blocks, size_t* total, - int histogram[kMallocHistogramSize]) { - return MallocBlock::MemoryStats(blocks, total, histogram); - } - - virtual size_t GetEstimatedAllocatedSize(size_t size) { - return size; - } - - virtual size_t GetAllocatedSize(const void* p) { - if (p) { - RAW_CHECK(GetOwnership(p) != MallocExtension::kNotOwned, - "ptr not allocated by tcmalloc"); - return MallocBlock::FromRawPointer(p)->data_size(); - } - return 0; - } - - virtual MallocExtension::Ownership GetOwnership(const void* p) { - if (!p) { - // nobody owns NULL - return MallocExtension::kNotOwned; - } - - // FIXME: note that correct GetOwnership should not touch memory - // that is not owned by tcmalloc. Main implementation is using - // pagemap to discover if page in question is owned by us or - // not. But pagemap only has marks for first and last page of - // spans. Note that if p was returned out of our memalign with - // big alignment, then it will point outside of marked pages. Also - // note that FromRawPointer call below requires touching memory - // before pointer in order to handle memalign-ed chunks - // (offset_). This leaves us with two options: - // - // * do FromRawPointer first and have possibility of crashing if - // we're given not owned pointer - // - // * return incorrect ownership for those large memalign chunks - // - // I've decided to choose later, which appears to happen rarer and - // therefore is arguably a lesser evil - - MallocExtension::Ownership rv = TCMallocImplementation::GetOwnership(p); - if (rv != MallocExtension::kOwned) { - return rv; - } - - const MallocBlock* mb = MallocBlock::FromRawPointer(p); - return TCMallocImplementation::GetOwnership(mb); - } - - virtual void GetFreeListSizes(vector* v) { - static const char* kDebugFreeQueue = "debug.free_queue"; - - TCMallocImplementation::GetFreeListSizes(v); - - MallocExtension::FreeListInfo i; - i.type = kDebugFreeQueue; - i.min_object_size = 0; - i.max_object_size = numeric_limits::max(); - i.total_bytes_free = MallocBlock::FreeQueueSize(); - v->push_back(i); - } - - }; - -static union { - char chars[sizeof(DebugMallocImplementation)]; - void *ptr; -} debug_malloc_implementation_space; - -REGISTER_MODULE_INITIALIZER(debugallocation, { -#if (__cplusplus >= 201103L) - COMPILE_ASSERT(alignof(debug_malloc_implementation_space) >= alignof(DebugMallocImplementation), - debug_malloc_implementation_space_is_not_properly_aligned); -#endif - // Either we or valgrind will control memory management. We - // register our extension if we're the winner. Otherwise let - // Valgrind use its own malloc (so don't register our extension). - if (!RunningOnValgrind()) { - DebugMallocImplementation *impl = new (debug_malloc_implementation_space.chars) DebugMallocImplementation(); - MallocExtension::Register(impl); - } -}); - -REGISTER_MODULE_DESTRUCTOR(debugallocation, { - if (!RunningOnValgrind()) { - // When the program exits, check all blocks still in the free - // queue for corruption. - DanglingWriteChecker(); - } -}); - -// ========================================================================= // - -struct debug_alloc_retry_data { - size_t size; - int new_type; -}; - -static void *retry_debug_allocate(void *arg) { - debug_alloc_retry_data *data = static_cast(arg); - return DebugAllocate(data->size, data->new_type); -} - -// This is mostly the same a cpp_alloc in tcmalloc.cc. -// TODO(csilvers): change Allocate() above to call cpp_alloc, so we -// don't have to reproduce the logic here. To make tc_new_mode work -// properly, I think we'll need to separate out the logic of throwing -// from the logic of calling the new-handler. -inline void* debug_cpp_alloc(size_t size, int new_type, bool nothrow) { - void* p = DebugAllocate(size, new_type); - if (p != NULL) { - return p; - } - struct debug_alloc_retry_data data; - data.size = size; - data.new_type = new_type; - return handle_oom(retry_debug_allocate, &data, - true, nothrow); -} - -inline void* do_debug_malloc_or_debug_cpp_alloc(size_t size) { - void* p = DebugAllocate(size, MallocBlock::kMallocType); - if (p != NULL) { - return p; - } - struct debug_alloc_retry_data data; - data.size = size; - data.new_type = MallocBlock::kMallocType; - return handle_oom(retry_debug_allocate, &data, - false, true); -} - -// Exported routines - -extern "C" PERFTOOLS_DLL_DECL void* tc_malloc(size_t size) PERFTOOLS_THROW { - if (ThreadCache::IsUseEmergencyMalloc()) { - return tcmalloc::EmergencyMalloc(size); - } - void* ptr = do_debug_malloc_or_debug_cpp_alloc(size); - MallocHook::InvokeNewHook(ptr, size); - return ptr; -} - -extern "C" PERFTOOLS_DLL_DECL void tc_free(void* ptr) PERFTOOLS_THROW { - if (tcmalloc::IsEmergencyPtr(ptr)) { - return tcmalloc::EmergencyFree(ptr); - } - MallocHook::InvokeDeleteHook(ptr); - DebugDeallocate(ptr, MallocBlock::kMallocType, 0); -} - -extern "C" PERFTOOLS_DLL_DECL void tc_free_sized(void *ptr, size_t size) PERFTOOLS_THROW { - MallocHook::InvokeDeleteHook(ptr); - DebugDeallocate(ptr, MallocBlock::kMallocType, size); -} - -extern "C" PERFTOOLS_DLL_DECL void* tc_calloc(size_t count, size_t size) PERFTOOLS_THROW { - if (ThreadCache::IsUseEmergencyMalloc()) { - return tcmalloc::EmergencyCalloc(count, size); - } - // Overflow check - const size_t total_size = count * size; - if (size != 0 && total_size / size != count) return NULL; - - void* block = do_debug_malloc_or_debug_cpp_alloc(total_size); - MallocHook::InvokeNewHook(block, total_size); - if (block) memset(block, 0, total_size); - return block; -} - -extern "C" PERFTOOLS_DLL_DECL void tc_cfree(void* ptr) PERFTOOLS_THROW { - if (tcmalloc::IsEmergencyPtr(ptr)) { - return tcmalloc::EmergencyFree(ptr); - } - MallocHook::InvokeDeleteHook(ptr); - DebugDeallocate(ptr, MallocBlock::kMallocType, 0); -} - -extern "C" PERFTOOLS_DLL_DECL void* tc_realloc(void* ptr, size_t size) PERFTOOLS_THROW { - if (tcmalloc::IsEmergencyPtr(ptr)) { - return tcmalloc::EmergencyRealloc(ptr, size); - } - if (ptr == NULL) { - ptr = do_debug_malloc_or_debug_cpp_alloc(size); - MallocHook::InvokeNewHook(ptr, size); - return ptr; - } - if (size == 0) { - MallocHook::InvokeDeleteHook(ptr); - DebugDeallocate(ptr, MallocBlock::kMallocType, 0); - return NULL; - } - MallocBlock* old = MallocBlock::FromRawPointer(ptr); - old->Check(MallocBlock::kMallocType); - MallocBlock* p = MallocBlock::Allocate(size, MallocBlock::kMallocType); - - // If realloc fails we are to leave the old block untouched and - // return null - if (p == NULL) return NULL; - - // if ptr was allocated via memalign, then old->data_size() is not - // start of user data. So we must be careful to copy only user-data - char *old_begin = (char *)old->data_addr(); - char *old_end = old_begin + old->data_size(); - - ssize_t old_ssize = old_end - (char *)ptr; - CHECK_CONDITION(old_ssize >= 0); - - size_t old_size = (size_t)old_ssize; - CHECK_CONDITION(old_size <= old->data_size()); - - memcpy(p->data_addr(), ptr, (old_size < size) ? old_size : size); - MallocHook::InvokeDeleteHook(ptr); - MallocHook::InvokeNewHook(p->data_addr(), size); - DebugDeallocate(ptr, MallocBlock::kMallocType, 0); - MALLOC_TRACE("realloc", p->data_size(), p->data_addr()); - return p->data_addr(); -} - -extern "C" PERFTOOLS_DLL_DECL void* tc_new(size_t size) { - void* ptr = debug_cpp_alloc(size, MallocBlock::kNewType, false); - MallocHook::InvokeNewHook(ptr, size); - if (ptr == NULL) { - RAW_LOG(FATAL, "Unable to allocate %" PRIuS " bytes: new failed.", size); - } - return ptr; -} - -extern "C" PERFTOOLS_DLL_DECL void* tc_new_nothrow(size_t size, const std::nothrow_t&) PERFTOOLS_THROW { - void* ptr = debug_cpp_alloc(size, MallocBlock::kNewType, true); - MallocHook::InvokeNewHook(ptr, size); - return ptr; -} - -extern "C" PERFTOOLS_DLL_DECL void tc_delete(void* p) PERFTOOLS_THROW { - MallocHook::InvokeDeleteHook(p); - DebugDeallocate(p, MallocBlock::kNewType, 0); -} - -extern "C" PERFTOOLS_DLL_DECL void tc_delete_sized(void* p, size_t size) throw() { - MallocHook::InvokeDeleteHook(p); - DebugDeallocate(p, MallocBlock::kNewType, size); -} - -// Some STL implementations explicitly invoke this. -// It is completely equivalent to a normal delete (delete never throws). -extern "C" PERFTOOLS_DLL_DECL void tc_delete_nothrow(void* p, const std::nothrow_t&) PERFTOOLS_THROW { - MallocHook::InvokeDeleteHook(p); - DebugDeallocate(p, MallocBlock::kNewType, 0); -} - -extern "C" PERFTOOLS_DLL_DECL void* tc_newarray(size_t size) { - void* ptr = debug_cpp_alloc(size, MallocBlock::kArrayNewType, false); - MallocHook::InvokeNewHook(ptr, size); - if (ptr == NULL) { - RAW_LOG(FATAL, "Unable to allocate %" PRIuS " bytes: new[] failed.", size); - } - return ptr; -} - -extern "C" PERFTOOLS_DLL_DECL void* tc_newarray_nothrow(size_t size, const std::nothrow_t&) - PERFTOOLS_THROW { - void* ptr = debug_cpp_alloc(size, MallocBlock::kArrayNewType, true); - MallocHook::InvokeNewHook(ptr, size); - return ptr; -} - -extern "C" PERFTOOLS_DLL_DECL void tc_deletearray(void* p) PERFTOOLS_THROW { - MallocHook::InvokeDeleteHook(p); - DebugDeallocate(p, MallocBlock::kArrayNewType, 0); -} - -extern "C" PERFTOOLS_DLL_DECL void tc_deletearray_sized(void* p, size_t size) throw() { - MallocHook::InvokeDeleteHook(p); - DebugDeallocate(p, MallocBlock::kArrayNewType, size); -} - -// Some STL implementations explicitly invoke this. -// It is completely equivalent to a normal delete (delete never throws). -extern "C" PERFTOOLS_DLL_DECL void tc_deletearray_nothrow(void* p, const std::nothrow_t&) PERFTOOLS_THROW { - MallocHook::InvokeDeleteHook(p); - DebugDeallocate(p, MallocBlock::kArrayNewType, 0); -} - -// This is mostly the same as do_memalign in tcmalloc.cc. -static void *do_debug_memalign(size_t alignment, size_t size) { - // Allocate >= size bytes aligned on "alignment" boundary - // "alignment" is a power of two. - void *p = 0; - RAW_CHECK((alignment & (alignment-1)) == 0, "must be power of two"); - const size_t data_offset = MallocBlock::data_offset(); - // Allocate "alignment-1" extra bytes to ensure alignment is possible, and - // a further data_offset bytes for an additional fake header. - size_t extra_bytes = data_offset + alignment - 1; - if (size + extra_bytes < size) return NULL; // Overflow - p = DebugAllocate(size + extra_bytes, MallocBlock::kMallocType); - if (p != 0) { - intptr_t orig_p = reinterpret_cast(p); - // Leave data_offset bytes for fake header, and round up to meet - // alignment. - p = reinterpret_cast(RoundUp(orig_p + data_offset, alignment)); - // Create a fake header block with an offset_ that points back to the - // real header. FromRawPointer uses this value. - MallocBlock *fake_hdr = reinterpret_cast( - reinterpret_cast(p) - data_offset); - // offset_ is distance between real and fake headers. - // p is now end of fake header (beginning of client area), - // and orig_p is the end of the real header, so offset_ - // is their difference. - // - // Note that other fields of fake_hdr are initialized with - // kMagicUninitializedByte - fake_hdr->set_offset(reinterpret_cast(p) - orig_p); - } - return p; -} - -struct memalign_retry_data { - size_t align; - size_t size; -}; - -static void *retry_debug_memalign(void *arg) { - memalign_retry_data *data = static_cast(arg); - return do_debug_memalign(data->align, data->size); -} - -inline void* do_debug_memalign_or_debug_cpp_memalign(size_t align, - size_t size) { - void* p = do_debug_memalign(align, size); - if (p != NULL) { - return p; - } - - struct memalign_retry_data data; - data.align = align; - data.size = size; - return handle_oom(retry_debug_memalign, &data, - false, true); -} - -extern "C" PERFTOOLS_DLL_DECL void* tc_memalign(size_t align, size_t size) PERFTOOLS_THROW { - void *p = do_debug_memalign_or_debug_cpp_memalign(align, size); - MallocHook::InvokeNewHook(p, size); - return p; -} - -// Implementation taken from tcmalloc/tcmalloc.cc -extern "C" PERFTOOLS_DLL_DECL int tc_posix_memalign(void** result_ptr, size_t align, size_t size) - PERFTOOLS_THROW { - if (((align % sizeof(void*)) != 0) || - ((align & (align - 1)) != 0) || - (align == 0)) { - return EINVAL; - } - - void* result = do_debug_memalign_or_debug_cpp_memalign(align, size); - MallocHook::InvokeNewHook(result, size); - if (result == NULL) { - return ENOMEM; - } else { - *result_ptr = result; - return 0; - } -} - -extern "C" PERFTOOLS_DLL_DECL void* tc_valloc(size_t size) PERFTOOLS_THROW { - // Allocate >= size bytes starting on a page boundary - void *p = do_debug_memalign_or_debug_cpp_memalign(getpagesize(), size); - MallocHook::InvokeNewHook(p, size); - return p; -} - -extern "C" PERFTOOLS_DLL_DECL void* tc_pvalloc(size_t size) PERFTOOLS_THROW { - // Round size up to a multiple of pages - // then allocate memory on a page boundary - int pagesize = getpagesize(); - size = RoundUp(size, pagesize); - if (size == 0) { // pvalloc(0) should allocate one page, according to - size = pagesize; // http://man.free4web.biz/man3/libmpatrol.3.html - } - void *p = do_debug_memalign_or_debug_cpp_memalign(pagesize, size); - MallocHook::InvokeNewHook(p, size); - return p; -} - -// malloc_stats just falls through to the base implementation. -extern "C" PERFTOOLS_DLL_DECL void tc_malloc_stats(void) PERFTOOLS_THROW { - do_malloc_stats(); -} - -extern "C" PERFTOOLS_DLL_DECL int tc_mallopt(int cmd, int value) PERFTOOLS_THROW { - return do_mallopt(cmd, value); -} - -#ifdef HAVE_STRUCT_MALLINFO -extern "C" PERFTOOLS_DLL_DECL struct mallinfo tc_mallinfo(void) PERFTOOLS_THROW { - return do_mallinfo(); -} -#endif - -extern "C" PERFTOOLS_DLL_DECL size_t tc_malloc_size(void* ptr) PERFTOOLS_THROW { - return MallocExtension::instance()->GetAllocatedSize(ptr); -} - -extern "C" PERFTOOLS_DLL_DECL void* tc_malloc_skip_new_handler(size_t size) PERFTOOLS_THROW { - void* result = DebugAllocate(size, MallocBlock::kMallocType); - MallocHook::InvokeNewHook(result, size); - return result; -} - -#pragma GCC diagnostic pop diff --git a/contrib/libtcmalloc/src/getenv_safe.h b/contrib/libtcmalloc/src/getenv_safe.h deleted file mode 100644 index 3b9f4dbbcb2..00000000000 --- a/contrib/libtcmalloc/src/getenv_safe.h +++ /dev/null @@ -1,63 +0,0 @@ -/* -*- Mode: C++; c-basic-offset: 2; indent-tabs-mode: nil -*- - * Copyright (c) 2014, gperftools Contributors - * All rights reserved. - * - * Redistribution and use in source and binary forms, with or without - * modification, are permitted provided that the following conditions are - * met: - * - * * Redistributions of source code must retain the above copyright - * notice, this list of conditions and the following disclaimer. - * * Redistributions in binary form must reproduce the above - * copyright notice, this list of conditions and the following disclaimer - * in the documentation and/or other materials provided with the - * distribution. - * * Neither the name of Google Inc. nor the names of its - * contributors may be used to endorse or promote products derived from - * this software without specific prior written permission. - * - * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS - * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT - * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR - * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT - * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, - * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT - * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, - * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY - * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT - * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE - * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. - */ - -#ifndef GETENV_SAFE_H -#define GETENV_SAFE_H - -#ifdef __cplusplus -extern "C" { -#endif - -/* - * This getenv function is safe to call before the C runtime is initialized. - * On Windows, it utilizes GetEnvironmentVariable() and on unix it uses - * /proc/self/environ instead calling getenv(). It's intended to be used in - * routines that run before main(), when the state required for getenv() may - * not be set up yet. In particular, errno isn't set up until relatively late - * (after the pthreads library has a chance to make it threadsafe), and - * getenv() doesn't work until then. - * On some platforms, this call will utilize the same, static buffer for - * repeated GetenvBeforeMain() calls. Callers should not expect pointers from - * this routine to be long lived. - * Note that on unix, /proc only has the environment at the time the - * application was started, so this routine ignores setenv() calls/etc. Also - * note it only reads the first 16K of the environment. - * - * NOTE: this is version of GetenvBeforeMain that's usable from - * C. Implementation is in sysinfo.cc - */ -const char* TCMallocGetenvSafe(const char* name); - -#ifdef __cplusplus -} -#endif - -#endif diff --git a/contrib/libtcmalloc/src/getpc.h b/contrib/libtcmalloc/src/getpc.h deleted file mode 100644 index 163873eabc6..00000000000 --- a/contrib/libtcmalloc/src/getpc.h +++ /dev/null @@ -1,192 +0,0 @@ -// -*- Mode: C++; c-basic-offset: 2; indent-tabs-mode: nil -*- -// Copyright (c) 2005, Google Inc. -// All rights reserved. -// -// Redistribution and use in source and binary forms, with or without -// modification, are permitted provided that the following conditions are -// met: -// -// * Redistributions of source code must retain the above copyright -// notice, this list of conditions and the following disclaimer. -// * Redistributions in binary form must reproduce the above -// copyright notice, this list of conditions and the following disclaimer -// in the documentation and/or other materials provided with the -// distribution. -// * Neither the name of Google Inc. nor the names of its -// contributors may be used to endorse or promote products derived from -// this software without specific prior written permission. -// -// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS -// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT -// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR -// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT -// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, -// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT -// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, -// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY -// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT -// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE -// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. - -// --- -// Author: Craig Silverstein -// -// This is an internal header file used by profiler.cc. It defines -// the single (inline) function GetPC. GetPC is used in a signal -// handler to figure out the instruction that was being executed when -// the signal-handler was triggered. -// -// To get this, we use the ucontext_t argument to the signal-handler -// callback, which holds the full context of what was going on when -// the signal triggered. How to get from a ucontext_t to a Program -// Counter is OS-dependent. - -#ifndef BASE_GETPC_H_ -#define BASE_GETPC_H_ - -#include "config.h" - -// On many linux systems, we may need _GNU_SOURCE to get access to -// the defined constants that define the register we want to see (eg -// REG_EIP). Note this #define must come first! -#define _GNU_SOURCE 1 -// If #define _GNU_SOURCE causes problems, this might work instead. -// It will cause problems for FreeBSD though!, because it turns off -// the needed __BSD_VISIBLE. -//#define _XOPEN_SOURCE 500 - -#include // for memcmp -#if defined(HAVE_SYS_UCONTEXT_H) -#include -#elif defined(HAVE_UCONTEXT_H) -#include // for ucontext_t (and also mcontext_t) -#elif defined(HAVE_CYGWIN_SIGNAL_H) -#include -typedef ucontext ucontext_t; -#endif - - -// Take the example where function Foo() calls function Bar(). For -// many architectures, Bar() is responsible for setting up and tearing -// down its own stack frame. In that case, it's possible for the -// interrupt to happen when execution is in Bar(), but the stack frame -// is not properly set up (either before it's done being set up, or -// after it's been torn down but before Bar() returns). In those -// cases, the stack trace cannot see the caller function anymore. -// -// GetPC can try to identify this situation, on architectures where it -// might occur, and unwind the current function call in that case to -// avoid false edges in the profile graph (that is, edges that appear -// to show a call skipping over a function). To do this, we hard-code -// in the asm instructions we might see when setting up or tearing -// down a stack frame. -// -// This is difficult to get right: the instructions depend on the -// processor, the compiler ABI, and even the optimization level. This -// is a best effort patch -- if we fail to detect such a situation, or -// mess up the PC, nothing happens; the returned PC is not used for -// any further processing. -struct CallUnrollInfo { - // Offset from (e)ip register where this instruction sequence - // should be matched. Interpreted as bytes. Offset 0 is the next - // instruction to execute. Be extra careful with negative offsets in - // architectures of variable instruction length (like x86) - it is - // not that easy as taking an offset to step one instruction back! - int pc_offset; - // The actual instruction bytes. Feel free to make it larger if you - // need a longer sequence. - unsigned char ins[16]; - // How many bytes to match from ins array? - int ins_size; - // The offset from the stack pointer (e)sp where to look for the - // call return address. Interpreted as bytes. - int return_sp_offset; -}; - - -// The dereferences needed to get the PC from a struct ucontext were -// determined at configure time, and stored in the macro -// PC_FROM_UCONTEXT in config.h. The only thing we need to do here, -// then, is to do the magic call-unrolling for systems that support it. - -// -- Special case 1: linux x86, for which we have CallUnrollInfo -#if defined(__linux) && defined(__i386) && defined(__GNUC__) -static const CallUnrollInfo callunrollinfo[] = { - // Entry to a function: push %ebp; mov %esp,%ebp - // Top-of-stack contains the caller IP. - { 0, - {0x55, 0x89, 0xe5}, 3, - 0 - }, - // Entry to a function, second instruction: push %ebp; mov %esp,%ebp - // Top-of-stack contains the old frame, caller IP is +4. - { -1, - {0x55, 0x89, 0xe5}, 3, - 4 - }, - // Return from a function: RET. - // Top-of-stack contains the caller IP. - { 0, - {0xc3}, 1, - 0 - } -}; - -inline void* GetPC(const ucontext_t& signal_ucontext) { - // See comment above struct CallUnrollInfo. Only try instruction - // flow matching if both eip and esp looks reasonable. - const int eip = signal_ucontext.uc_mcontext.gregs[REG_EIP]; - const int esp = signal_ucontext.uc_mcontext.gregs[REG_ESP]; - if ((eip & 0xffff0000) != 0 && (~eip & 0xffff0000) != 0 && - (esp & 0xffff0000) != 0) { - char* eip_char = reinterpret_cast(eip); - for (int i = 0; i < sizeof(callunrollinfo)/sizeof(*callunrollinfo); ++i) { - if (!memcmp(eip_char + callunrollinfo[i].pc_offset, - callunrollinfo[i].ins, callunrollinfo[i].ins_size)) { - // We have a match. - void **retaddr = (void**)(esp + callunrollinfo[i].return_sp_offset); - return *retaddr; - } - } - } - return (void*)eip; -} - -// Special case #2: Windows, which has to do something totally different. -#elif defined(_WIN32) || defined(__CYGWIN__) || defined(__CYGWIN32__) || defined(__MINGW32__) -// If this is ever implemented, probably the way to do it is to have -// profiler.cc use a high-precision timer via timeSetEvent: -// http://msdn2.microsoft.com/en-us/library/ms712713.aspx -// We'd use it in mode TIME_CALLBACK_FUNCTION/TIME_PERIODIC. -// The callback function would be something like prof_handler, but -// alas the arguments are different: no ucontext_t! I don't know -// how we'd get the PC (using StackWalk64?) -// http://msdn2.microsoft.com/en-us/library/ms680650.aspx - -#include "base/logging.h" // for RAW_LOG -#ifndef HAVE_CYGWIN_SIGNAL_H -typedef int ucontext_t; -#endif - -inline void* GetPC(const struct ucontext_t& signal_ucontext) { - RAW_LOG(ERROR, "GetPC is not yet implemented on Windows\n"); - return NULL; -} - -// Normal cases. If this doesn't compile, it's probably because -// PC_FROM_UCONTEXT is the empty string. You need to figure out -// the right value for your system, and add it to the list in -// configure.ac (or set it manually in your config.h). -#else -inline void* GetPC(const ucontext_t& signal_ucontext) { -#if defined(__s390__) && !defined(__s390x__) - // Mask out the AMODE31 bit from the PC recorded in the context. - return (void*)((unsigned long)signal_ucontext.PC_FROM_UCONTEXT & 0x7fffffffUL); -#else - return (void*)signal_ucontext.PC_FROM_UCONTEXT; // defined in config.h -#endif -} - -#endif - -#endif // BASE_GETPC_H_ diff --git a/contrib/libtcmalloc/src/heap-checker-bcad.cc b/contrib/libtcmalloc/src/heap-checker-bcad.cc deleted file mode 100644 index 00efdb7cfd4..00000000000 --- a/contrib/libtcmalloc/src/heap-checker-bcad.cc +++ /dev/null @@ -1,93 +0,0 @@ -// -*- Mode: C++; c-basic-offset: 2; indent-tabs-mode: nil -*- -// Copyright (c) 2005, Google Inc. -// All rights reserved. -// -// Redistribution and use in source and binary forms, with or without -// modification, are permitted provided that the following conditions are -// met: -// -// * Redistributions of source code must retain the above copyright -// notice, this list of conditions and the following disclaimer. -// * Redistributions in binary form must reproduce the above -// copyright notice, this list of conditions and the following disclaimer -// in the documentation and/or other materials provided with the -// distribution. -// * Neither the name of Google Inc. nor the names of its -// contributors may be used to endorse or promote products derived from -// this software without specific prior written permission. -// -// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS -// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT -// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR -// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT -// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, -// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT -// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, -// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY -// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT -// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE -// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. - -// --- -// All Rights Reserved. -// -// Author: Maxim Lifantsev -// -// A file to ensure that components of heap leak checker run before -// all global object constructors and after all global object -// destructors. -// -// This file must be the last library any binary links against. -// Otherwise, the heap checker may not be able to run early enough to -// catalog all the global objects in your program. If this happens, -// and later in the program you allocate memory and have one of these -// "uncataloged" global objects point to it, the heap checker will -// consider that allocation to be a leak, even though it's not (since -// the allocated object is reachable from global data and hence "live"). - -#include // for abort() -#include - -// A dummy variable to refer from heap-checker.cc. This is to make -// sure this file is not optimized out by the linker. -bool heap_leak_checker_bcad_variable; - -extern void HeapLeakChecker_AfterDestructors(); // in heap-checker.cc - -// A helper class to ensure that some components of heap leak checking -// can happen before construction and after destruction -// of all global/static objects. -class HeapLeakCheckerGlobalPrePost { - public: - HeapLeakCheckerGlobalPrePost() { - if (count_ == 0) { - // The 'new int' will ensure that we have run an initial malloc - // hook, which will set up the heap checker via - // MallocHook_InitAtFirstAllocation_HeapLeakChecker. See malloc_hook.cc. - // This is done in this roundabout fashion in order to avoid self-deadlock - // if we directly called HeapLeakChecker_BeforeConstructors here. - delete new int; - // This needs to be called before the first allocation of an STL - // object, but after libc is done setting up threads (because it - // calls setenv, which requires a thread-aware errno). By - // putting it here, we hope it's the first bit of code executed - // after the libc global-constructor code. - MallocExtension::Initialize(); - } - ++count_; - } - ~HeapLeakCheckerGlobalPrePost() { - if (count_ <= 0) abort(); - --count_; - if (count_ == 0) HeapLeakChecker_AfterDestructors(); - } - private: - // Counter of constructions/destructions of objects of this class - // (just in case there are more than one of them). - static int count_; -}; - -int HeapLeakCheckerGlobalPrePost::count_ = 0; - -// The early-construction/late-destruction global object. -static const HeapLeakCheckerGlobalPrePost heap_leak_checker_global_pre_post; diff --git a/contrib/libtcmalloc/src/heap-checker.cc b/contrib/libtcmalloc/src/heap-checker.cc deleted file mode 100644 index 9c82dea08e4..00000000000 --- a/contrib/libtcmalloc/src/heap-checker.cc +++ /dev/null @@ -1,2388 +0,0 @@ -// -*- Mode: C++; c-basic-offset: 2; indent-tabs-mode: nil -*- -// Copyright (c) 2005, Google Inc. -// All rights reserved. -// -// Redistribution and use in source and binary forms, with or without -// modification, are permitted provided that the following conditions are -// met: -// -// * Redistributions of source code must retain the above copyright -// notice, this list of conditions and the following disclaimer. -// * Redistributions in binary form must reproduce the above -// copyright notice, this list of conditions and the following disclaimer -// in the documentation and/or other materials provided with the -// distribution. -// * Neither the name of Google Inc. nor the names of its -// contributors may be used to endorse or promote products derived from -// this software without specific prior written permission. -// -// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS -// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT -// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR -// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT -// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, -// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT -// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, -// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY -// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT -// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE -// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. - -// --- -// All Rights Reserved. -// -// Author: Maxim Lifantsev -// - -#include "config.h" - -#include // for O_RDONLY (we use syscall to do actual reads) -#include -#include -#ifdef HAVE_UNISTD_H -#include -#endif -#ifdef HAVE_MMAP -#include -#endif -#ifdef HAVE_PTHREAD -#include -#endif -#include -#include -#include -#include - -#if defined(HAVE_LINUX_PTRACE_H) -#include -#endif -#ifdef HAVE_SYS_SYSCALL_H -#include -#endif -#if defined(_WIN32) || defined(__CYGWIN__) || defined(__CYGWIN32__) || defined(__MINGW32__) -#include -#include -#undef ERROR // windows defines these as macros, which can cause trouble -#undef max -#undef min -#endif - -#include -#include -#include -#include -#include -#include - -#include - -#include "base/basictypes.h" -#include "base/googleinit.h" -#include "base/logging.h" -#include -#include "base/commandlineflags.h" -#include "base/elfcore.h" // for i386_regs -#include "base/thread_lister.h" -#include "heap-profile-table.h" -#include "base/low_level_alloc.h" -#include "malloc_hook-inl.h" -#include -#include -#include "maybe_threads.h" -#include "memory_region_map.h" -#include "base/spinlock.h" -#include "base/sysinfo.h" -#include "base/stl_allocator.h" - -using std::string; -using std::basic_string; -using std::pair; -using std::map; -using std::set; -using std::vector; -using std::swap; -using std::make_pair; -using std::min; -using std::max; -using std::less; -using std::char_traits; - -// If current process is being ptrace()d, 'TracerPid' in /proc/self/status -// will be non-zero. -static bool IsDebuggerAttached(void) { // only works under linux, probably - char buf[256]; // TracerPid comes relatively earlier in status output - int fd = open("/proc/self/status", O_RDONLY); - if (fd == -1) { - return false; // Can't tell for sure. - } - const int len = read(fd, buf, sizeof(buf)); - bool rc = false; - if (len > 0) { - const char *const kTracerPid = "TracerPid:\t"; - buf[len - 1] = '\0'; - const char *p = strstr(buf, kTracerPid); - if (p != NULL) { - rc = (strncmp(p + strlen(kTracerPid), "0\n", 2) != 0); - } - } - close(fd); - return rc; -} - -// This is the default if you don't link in -lprofiler -extern "C" { -ATTRIBUTE_WEAK PERFTOOLS_DLL_DECL bool ProfilingIsEnabledForAllThreads(); -bool ProfilingIsEnabledForAllThreads() { return false; } -} - -//---------------------------------------------------------------------- -// Flags that control heap-checking -//---------------------------------------------------------------------- - -DEFINE_string(heap_check, - EnvToString("HEAPCHECK", ""), - "The heap leak checking to be done over the whole executable: " - "\"minimal\", \"normal\", \"strict\", " - "\"draconian\", \"as-is\", and \"local\" " - " or the empty string are the supported choices. " - "(See HeapLeakChecker_InternalInitStart for details.)"); - -DEFINE_bool(heap_check_report, true, "Obsolete"); - -DEFINE_bool(heap_check_before_constructors, - true, - "deprecated; pretty much always true now"); - -DEFINE_bool(heap_check_after_destructors, - EnvToBool("HEAP_CHECK_AFTER_DESTRUCTORS", false), - "If overall heap check is to end after global destructors " - "or right after all REGISTER_HEAPCHECK_CLEANUP's"); - -DEFINE_bool(heap_check_strict_check, true, "Obsolete"); - -DEFINE_bool(heap_check_ignore_global_live, - EnvToBool("HEAP_CHECK_IGNORE_GLOBAL_LIVE", true), - "If overall heap check is to ignore heap objects reachable " - "from the global data"); - -DEFINE_bool(heap_check_identify_leaks, - EnvToBool("HEAP_CHECK_IDENTIFY_LEAKS", false), - "If heap check should generate the addresses of the leaked " - "objects in the memory leak profiles. This may be useful " - "in tracking down leaks where only a small fraction of " - "objects allocated at the same stack trace are leaked."); - -DEFINE_bool(heap_check_ignore_thread_live, - EnvToBool("HEAP_CHECK_IGNORE_THREAD_LIVE", true), - "If set to true, objects reachable from thread stacks " - "and registers are not reported as leaks"); - -DEFINE_bool(heap_check_test_pointer_alignment, - EnvToBool("HEAP_CHECK_TEST_POINTER_ALIGNMENT", false), - "Set to true to check if the found leak can be due to " - "use of unaligned pointers"); - -// Alignment at which all pointers in memory are supposed to be located; -// use 1 if any alignment is ok. -// heap_check_test_pointer_alignment flag guides if we try the value of 1. -// The larger it can be, the lesser is the chance of missing real leaks. -static const size_t kPointerSourceAlignment = sizeof(void*); -DEFINE_int32(heap_check_pointer_source_alignment, - EnvToInt("HEAP_CHECK_POINTER_SOURCE_ALIGNMENT", - kPointerSourceAlignment), - "Alignment at which all pointers in memory are supposed to be " - "located. Use 1 if any alignment is ok."); - -// A reasonable default to handle pointers inside of typical class objects: -// Too low and we won't be able to traverse pointers to normally-used -// nested objects and base parts of multiple-inherited objects. -// Too high and it will both slow down leak checking (FindInsideAlloc -// in HaveOnHeapLocked will get slower when there are large on-heap objects) -// and make it probabilistically more likely to miss leaks -// of large-sized objects. -static const int64 kHeapCheckMaxPointerOffset = 1024; -DEFINE_int64(heap_check_max_pointer_offset, - EnvToInt("HEAP_CHECK_MAX_POINTER_OFFSET", - kHeapCheckMaxPointerOffset), - "Largest pointer offset for which we traverse " - "pointers going inside of heap allocated objects. " - "Set to -1 to use the actual largest heap object size."); - -DEFINE_bool(heap_check_run_under_gdb, - EnvToBool("HEAP_CHECK_RUN_UNDER_GDB", false), - "If false, turns off heap-checking library when running under gdb " - "(normally, set to 'true' only when debugging the heap-checker)"); - -DEFINE_int32(heap_check_delay_seconds, 0, - "Number of seconds to delay on-exit heap checking." - " If you set this flag," - " you may also want to set exit_timeout_seconds in order to" - " avoid exit timeouts.\n" - "NOTE: This flag is to be used only to help diagnose issues" - " where it is suspected that the heap checker is reporting" - " false leaks that will disappear if the heap checker delays" - " its checks. Report any such issues to the heap-checker" - " maintainer(s)."); - -//---------------------------------------------------------------------- - -DEFINE_string(heap_profile_pprof, - EnvToString("PPROF_PATH", "pprof"), - "OBSOLETE; not used"); - -DEFINE_string(heap_check_dump_directory, - EnvToString("HEAP_CHECK_DUMP_DIRECTORY", "/tmp"), - "Directory to put heap-checker leak dump information"); - - -//---------------------------------------------------------------------- -// HeapLeakChecker global data -//---------------------------------------------------------------------- - -// Global lock for all the global data of this module. -static SpinLock heap_checker_lock(SpinLock::LINKER_INITIALIZED); - -//---------------------------------------------------------------------- - -// Heap profile prefix for leak checking profiles. -// Gets assigned once when leak checking is turned on, then never modified. -static const string* profile_name_prefix = NULL; - -// Whole-program heap leak checker. -// Gets assigned once when leak checking is turned on, -// then main_heap_checker is never deleted. -static HeapLeakChecker* main_heap_checker = NULL; - -// Whether we will use main_heap_checker to do a check at program exit -// automatically. In any case user can ask for more checks on main_heap_checker -// via GlobalChecker(). -static bool do_main_heap_check = false; - -// The heap profile we use to collect info about the heap. -// This is created in HeapLeakChecker::BeforeConstructorsLocked -// together with setting heap_checker_on (below) to true -// and registering our new/delete malloc hooks; -// similarly all are unset in HeapLeakChecker::TurnItselfOffLocked. -static HeapProfileTable* heap_profile = NULL; - -// If we are doing (or going to do) any kind of heap-checking. -static bool heap_checker_on = false; - -// pid of the process that does whole-program heap leak checking -static pid_t heap_checker_pid = 0; - -// If we did heap profiling during global constructors execution -static bool constructor_heap_profiling = false; - -// RAW_VLOG level we dump key INFO messages at. If you want to turn -// off these messages, set the environment variable PERFTOOLS_VERBOSE=-1. -static const int heap_checker_info_level = 0; - -//---------------------------------------------------------------------- -// HeapLeakChecker's own memory allocator that is -// independent of the normal program allocator. -//---------------------------------------------------------------------- - -// Wrapper of LowLevelAlloc for STL_Allocator and direct use. -// We always access this class under held heap_checker_lock, -// this allows us to in particular protect the period when threads are stopped -// at random spots with TCMalloc_ListAllProcessThreads by heap_checker_lock, -// w/o worrying about the lock in LowLevelAlloc::Arena. -// We rely on the fact that we use an own arena with an own lock here. -class HeapLeakChecker::Allocator { - public: - static void Init() { - RAW_DCHECK(heap_checker_lock.IsHeld(), ""); - RAW_DCHECK(arena_ == NULL, ""); - arena_ = LowLevelAlloc::NewArena(0, LowLevelAlloc::DefaultArena()); - } - static void Shutdown() { - RAW_DCHECK(heap_checker_lock.IsHeld(), ""); - if (!LowLevelAlloc::DeleteArena(arena_) || alloc_count_ != 0) { - RAW_LOG(FATAL, "Internal heap checker leak of %d objects", alloc_count_); - } - } - static int alloc_count() { - RAW_DCHECK(heap_checker_lock.IsHeld(), ""); - return alloc_count_; - } - static void* Allocate(size_t n) { - RAW_DCHECK(arena_ && heap_checker_lock.IsHeld(), ""); - void* p = LowLevelAlloc::AllocWithArena(n, arena_); - if (p) alloc_count_ += 1; - return p; - } - static void Free(void* p) { - RAW_DCHECK(heap_checker_lock.IsHeld(), ""); - if (p) alloc_count_ -= 1; - LowLevelAlloc::Free(p); - } - static void Free(void* p, size_t /* n */) { - Free(p); - } - // destruct, free, and make *p to be NULL - template static void DeleteAndNull(T** p) { - (*p)->~T(); - Free(*p); - *p = NULL; - } - template static void DeleteAndNullIfNot(T** p) { - if (*p != NULL) DeleteAndNull(p); - } - private: - static LowLevelAlloc::Arena* arena_; - static int alloc_count_; -}; - -LowLevelAlloc::Arena* HeapLeakChecker::Allocator::arena_ = NULL; -int HeapLeakChecker::Allocator::alloc_count_ = 0; - -//---------------------------------------------------------------------- -// HeapLeakChecker live object tracking components -//---------------------------------------------------------------------- - -// Cases of live object placement we distinguish -enum ObjectPlacement { - MUST_BE_ON_HEAP, // Must point to a live object of the matching size in the - // heap_profile map of the heap when we get to it - IGNORED_ON_HEAP, // Is a live (ignored) object on heap - MAYBE_LIVE, // Is a piece of writable memory from /proc/self/maps - IN_GLOBAL_DATA, // Is part of global data region of the executable - THREAD_DATA, // Part of a thread stack and a thread descriptor with TLS - THREAD_REGISTERS, // Values in registers of some thread -}; - -// Information about an allocated object -struct AllocObject { - const void* ptr; // the object - uintptr_t size; // its size - ObjectPlacement place; // where ptr points to - - AllocObject(const void* p, size_t s, ObjectPlacement l) - : ptr(p), size(s), place(l) { } -}; - -// All objects (memory ranges) ignored via HeapLeakChecker::IgnoreObject -// Key is the object's address; value is its size. -typedef map, - STL_Allocator, - HeapLeakChecker::Allocator> - > IgnoredObjectsMap; -static IgnoredObjectsMap* ignored_objects = NULL; - -// All objects (memory ranges) that we consider to be the sources of pointers -// to live (not leaked) objects. -// At different times this holds (what can be reached from) global data regions -// and the objects we've been told to ignore. -// For any AllocObject::ptr "live_objects" is supposed to contain at most one -// record at any time. We maintain this by checking with the heap_profile map -// of the heap and removing the live heap objects we've handled from it. -// This vector is maintained as a stack and the frontier of reachable -// live heap objects in our flood traversal of them. -typedef vector - > LiveObjectsStack; -static LiveObjectsStack* live_objects = NULL; - -// A special string type that uses my allocator -typedef basic_string, - STL_Allocator - > HCL_string; - -// A placeholder to fill-in the starting values for live_objects -// for each library so we can keep the library-name association for logging. -typedef map, - STL_Allocator, - HeapLeakChecker::Allocator> - > LibraryLiveObjectsStacks; -static LibraryLiveObjectsStacks* library_live_objects = NULL; - -// Value stored in the map of disabled address ranges; -// its key is the end of the address range. -// We'll ignore allocations with a return address in a disabled range -// if the address occurs at 'max_depth' or less in the stack trace. -struct HeapLeakChecker::RangeValue { - uintptr_t start_address; // the start of the range - int max_depth; // the maximal stack depth to disable at -}; -typedef map, - STL_Allocator, - HeapLeakChecker::Allocator> - > DisabledRangeMap; -// The disabled program counter address ranges for profile dumping -// that are registered with HeapLeakChecker::DisableChecksFromToLocked. -static DisabledRangeMap* disabled_ranges = NULL; - -// Set of stack tops. -// These are used to consider live only appropriate chunks of the memory areas -// that are used for stacks (and maybe thread-specific data as well) -// so that we do not treat pointers from outdated stack frames as live. -typedef set, - STL_Allocator - > StackTopSet; -static StackTopSet* stack_tops = NULL; - -// A map of ranges of code addresses for the system libraries -// that can mmap/mremap/sbrk-allocate memory regions for stacks -// and thread-local storage that we want to consider as live global data. -// Maps from the end address to the start address. -typedef map, - STL_Allocator, - HeapLeakChecker::Allocator> - > GlobalRegionCallerRangeMap; -static GlobalRegionCallerRangeMap* global_region_caller_ranges = NULL; - -// TODO(maxim): make our big data structs into own modules - -// Disabler is implemented by keeping track of a per-thread count -// of active Disabler objects. Any objects allocated while the -// count > 0 are not reported. - -#ifdef HAVE_TLS - -static __thread int thread_disable_counter -// The "inital exec" model is faster than the default TLS model, at -// the cost you can't dlopen this library. But dlopen on heap-checker -// doesn't work anyway -- it must run before main -- so this is a good -// trade-off. -# ifdef HAVE___ATTRIBUTE__ - __attribute__ ((tls_model ("initial-exec"))) -# endif - ; -inline int get_thread_disable_counter() { - return thread_disable_counter; -} -inline void set_thread_disable_counter(int value) { - thread_disable_counter = value; -} - -#else // #ifdef HAVE_TLS - -static pthread_key_t thread_disable_counter_key; -static int main_thread_counter; // storage for use before main() -static bool use_main_thread_counter = true; - -// TODO(csilvers): this is called from NewHook, in the middle of malloc(). -// If perftools_pthread_getspecific calls malloc, that will lead to an -// infinite loop. I don't know how to fix that, so I hope it never happens! -inline int get_thread_disable_counter() { - if (use_main_thread_counter) // means we're running really early - return main_thread_counter; - void* p = perftools_pthread_getspecific(thread_disable_counter_key); - return (intptr_t)p; // kinda evil: store the counter directly in the void* -} - -inline void set_thread_disable_counter(int value) { - if (use_main_thread_counter) { // means we're running really early - main_thread_counter = value; - return; - } - intptr_t pointer_sized_value = value; - // kinda evil: store the counter directly in the void* - void* p = (void*)pointer_sized_value; - // NOTE: this may call malloc, which will call NewHook which will call - // get_thread_disable_counter() which will call pthread_getspecific(). I - // don't know if anything bad can happen if we call getspecific() in the - // middle of a setspecific() call. It seems to work ok in practice... - perftools_pthread_setspecific(thread_disable_counter_key, p); -} - -// The idea here is that this initializer will run pretty late: after -// pthreads have been totally set up. At this point we can call -// pthreads routines, so we set those up. -class InitThreadDisableCounter { - public: - InitThreadDisableCounter() { - perftools_pthread_key_create(&thread_disable_counter_key, NULL); - // Set up the main thread's value, which we have a special variable for. - void* p = (void*)main_thread_counter; // store the counter directly - perftools_pthread_setspecific(thread_disable_counter_key, p); - use_main_thread_counter = false; - } -}; -InitThreadDisableCounter init_thread_disable_counter; - -#endif // #ifdef HAVE_TLS - -HeapLeakChecker::Disabler::Disabler() { - // It is faster to unconditionally increment the thread-local - // counter than to check whether or not heap-checking is on - // in a thread-safe manner. - int counter = get_thread_disable_counter(); - set_thread_disable_counter(counter + 1); - RAW_VLOG(10, "Increasing thread disable counter to %d", counter + 1); -} - -HeapLeakChecker::Disabler::~Disabler() { - int counter = get_thread_disable_counter(); - RAW_DCHECK(counter > 0, ""); - if (counter > 0) { - set_thread_disable_counter(counter - 1); - RAW_VLOG(10, "Decreasing thread disable counter to %d", counter); - } else { - RAW_VLOG(0, "Thread disable counter underflow : %d", counter); - } -} - -//---------------------------------------------------------------------- - -// The size of the largest heap object allocated so far. -static size_t max_heap_object_size = 0; -// The possible range of addresses that can point -// into one of the elements of heap_objects. -static uintptr_t min_heap_address = uintptr_t(-1LL); -static uintptr_t max_heap_address = 0; - -//---------------------------------------------------------------------- - -// Simple casting helpers for uintptr_t and void*: -template -inline static const void* AsPtr(T addr) { - return reinterpret_cast(addr); -} -inline static uintptr_t AsInt(const void* ptr) { - return reinterpret_cast(ptr); -} - -//---------------------------------------------------------------------- - -// We've seen reports that strstr causes heap-checker crashes in some -// libc's (?): -// http://code.google.com/p/gperftools/issues/detail?id=263 -// It's simple enough to use our own. This is not in time-critical code. -static const char* hc_strstr(const char* s1, const char* s2) { - const size_t len = strlen(s2); - RAW_CHECK(len > 0, "Unexpected empty string passed to strstr()"); - for (const char* p = strchr(s1, *s2); p != NULL; p = strchr(p+1, *s2)) { - if (strncmp(p, s2, len) == 0) { - return p; - } - } - return NULL; -} - -//---------------------------------------------------------------------- - -// Our hooks for MallocHook -static void NewHook(const void* ptr, size_t size) { - if (ptr != NULL) { - const int counter = get_thread_disable_counter(); - const bool ignore = (counter > 0); - RAW_VLOG(16, "Recording Alloc: %p of %" PRIuS "; %d", ptr, size, - int(counter)); - - // Fetch the caller's stack trace before acquiring heap_checker_lock. - void* stack[HeapProfileTable::kMaxStackDepth]; - int depth = HeapProfileTable::GetCallerStackTrace(0, stack); - - { SpinLockHolder l(&heap_checker_lock); - if (size > max_heap_object_size) max_heap_object_size = size; - uintptr_t addr = AsInt(ptr); - if (addr < min_heap_address) min_heap_address = addr; - addr += size; - if (addr > max_heap_address) max_heap_address = addr; - if (heap_checker_on) { - heap_profile->RecordAlloc(ptr, size, depth, stack); - if (ignore) { - heap_profile->MarkAsIgnored(ptr); - } - } - } - RAW_VLOG(17, "Alloc Recorded: %p of %" PRIuS "", ptr, size); - } -} - -static void DeleteHook(const void* ptr) { - if (ptr != NULL) { - RAW_VLOG(16, "Recording Free %p", ptr); - { SpinLockHolder l(&heap_checker_lock); - if (heap_checker_on) heap_profile->RecordFree(ptr); - } - RAW_VLOG(17, "Free Recorded: %p", ptr); - } -} - -//---------------------------------------------------------------------- - -enum StackDirection { - GROWS_TOWARDS_HIGH_ADDRESSES, - GROWS_TOWARDS_LOW_ADDRESSES, - UNKNOWN_DIRECTION -}; - -// Determine which way the stack grows: - -static StackDirection ATTRIBUTE_NOINLINE GetStackDirection( - const uintptr_t *const ptr) { - uintptr_t x; - if (&x < ptr) - return GROWS_TOWARDS_LOW_ADDRESSES; - if (ptr < &x) - return GROWS_TOWARDS_HIGH_ADDRESSES; - - RAW_CHECK(0, ""); // Couldn't determine the stack direction. - - return UNKNOWN_DIRECTION; -} - -// Direction of stack growth (will initialize via GetStackDirection()) -static StackDirection stack_direction = UNKNOWN_DIRECTION; - -// This routine is called for every thread stack we know about to register it. -static void RegisterStackLocked(const void* top_ptr) { - RAW_DCHECK(heap_checker_lock.IsHeld(), ""); - RAW_DCHECK(MemoryRegionMap::LockIsHeld(), ""); - RAW_VLOG(10, "Thread stack at %p", top_ptr); - uintptr_t top = AsInt(top_ptr); - stack_tops->insert(top); // add for later use - - // make sure stack_direction is initialized - if (stack_direction == UNKNOWN_DIRECTION) { - stack_direction = GetStackDirection(&top); - } - - // Find memory region with this stack - MemoryRegionMap::Region region; - if (MemoryRegionMap::FindAndMarkStackRegion(top, ®ion)) { - // Make the proper portion of the stack live: - if (stack_direction == GROWS_TOWARDS_LOW_ADDRESSES) { - RAW_VLOG(11, "Live stack at %p of %" PRIuPTR " bytes", - top_ptr, region.end_addr - top); - live_objects->push_back(AllocObject(top_ptr, region.end_addr - top, - THREAD_DATA)); - } else { // GROWS_TOWARDS_HIGH_ADDRESSES - RAW_VLOG(11, "Live stack at %p of %" PRIuPTR " bytes", - AsPtr(region.start_addr), - top - region.start_addr); - live_objects->push_back(AllocObject(AsPtr(region.start_addr), - top - region.start_addr, - THREAD_DATA)); - } - // not in MemoryRegionMap, look in library_live_objects: - } else if (FLAGS_heap_check_ignore_global_live) { - for (LibraryLiveObjectsStacks::iterator lib = library_live_objects->begin(); - lib != library_live_objects->end(); ++lib) { - for (LiveObjectsStack::iterator span = lib->second.begin(); - span != lib->second.end(); ++span) { - uintptr_t start = AsInt(span->ptr); - uintptr_t end = start + span->size; - if (start <= top && top < end) { - RAW_VLOG(11, "Stack at %p is inside /proc/self/maps chunk %p..%p", - top_ptr, AsPtr(start), AsPtr(end)); - // Shrink start..end region by chopping away the memory regions in - // MemoryRegionMap that land in it to undo merging of regions - // in /proc/self/maps, so that we correctly identify what portion - // of start..end is actually the stack region. - uintptr_t stack_start = start; - uintptr_t stack_end = end; - // can optimize-away this loop, but it does not run often - RAW_DCHECK(MemoryRegionMap::LockIsHeld(), ""); - for (MemoryRegionMap::RegionIterator r = - MemoryRegionMap::BeginRegionLocked(); - r != MemoryRegionMap::EndRegionLocked(); ++r) { - if (top < r->start_addr && r->start_addr < stack_end) { - stack_end = r->start_addr; - } - if (stack_start < r->end_addr && r->end_addr <= top) { - stack_start = r->end_addr; - } - } - if (stack_start != start || stack_end != end) { - RAW_VLOG(11, "Stack at %p is actually inside memory chunk %p..%p", - top_ptr, AsPtr(stack_start), AsPtr(stack_end)); - } - // Make the proper portion of the stack live: - if (stack_direction == GROWS_TOWARDS_LOW_ADDRESSES) { - RAW_VLOG(11, "Live stack at %p of %" PRIuPTR " bytes", - top_ptr, stack_end - top); - live_objects->push_back( - AllocObject(top_ptr, stack_end - top, THREAD_DATA)); - } else { // GROWS_TOWARDS_HIGH_ADDRESSES - RAW_VLOG(11, "Live stack at %p of %" PRIuPTR " bytes", - AsPtr(stack_start), top - stack_start); - live_objects->push_back( - AllocObject(AsPtr(stack_start), top - stack_start, THREAD_DATA)); - } - lib->second.erase(span); // kill the rest of the region - // Put the non-stack part(s) of the region back: - if (stack_start != start) { - lib->second.push_back(AllocObject(AsPtr(start), stack_start - start, - MAYBE_LIVE)); - } - if (stack_end != end) { - lib->second.push_back(AllocObject(AsPtr(stack_end), end - stack_end, - MAYBE_LIVE)); - } - return; - } - } - } - RAW_LOG(ERROR, "Memory region for stack at %p not found. " - "Will likely report false leak positives.", top_ptr); - } -} - -// Iterator for heap allocation map data to make ignored objects "live" -// (i.e., treated as roots for the mark-and-sweep phase) -static void MakeIgnoredObjectsLiveCallbackLocked( - const void* ptr, const HeapProfileTable::AllocInfo& info) { - RAW_DCHECK(heap_checker_lock.IsHeld(), ""); - if (info.ignored) { - live_objects->push_back(AllocObject(ptr, info.object_size, - MUST_BE_ON_HEAP)); - } -} - -// Iterator for heap allocation map data to make objects allocated from -// disabled regions of code to be live. -static void MakeDisabledLiveCallbackLocked( - const void* ptr, const HeapProfileTable::AllocInfo& info) { - RAW_DCHECK(heap_checker_lock.IsHeld(), ""); - bool stack_disable = false; - bool range_disable = false; - for (int depth = 0; depth < info.stack_depth; depth++) { - uintptr_t addr = AsInt(info.call_stack[depth]); - if (disabled_ranges) { - DisabledRangeMap::const_iterator iter - = disabled_ranges->upper_bound(addr); - if (iter != disabled_ranges->end()) { - RAW_DCHECK(iter->first > addr, ""); - if (iter->second.start_address < addr && - iter->second.max_depth > depth) { - range_disable = true; // in range; dropping - break; - } - } - } - } - if (stack_disable || range_disable) { - uintptr_t start_address = AsInt(ptr); - uintptr_t end_address = start_address + info.object_size; - StackTopSet::const_iterator iter - = stack_tops->lower_bound(start_address); - if (iter != stack_tops->end()) { - RAW_DCHECK(*iter >= start_address, ""); - if (*iter < end_address) { - // We do not disable (treat as live) whole allocated regions - // if they are used to hold thread call stacks - // (i.e. when we find a stack inside). - // The reason is that we'll treat as live the currently used - // stack portions anyway (see RegisterStackLocked), - // and the rest of the region where the stack lives can well - // contain outdated stack variables which are not live anymore, - // hence should not be treated as such. - RAW_VLOG(11, "Not %s-disabling %" PRIuS " bytes at %p" - ": have stack inside: %p", - (stack_disable ? "stack" : "range"), - info.object_size, ptr, AsPtr(*iter)); - return; - } - } - RAW_VLOG(11, "%s-disabling %" PRIuS " bytes at %p", - (stack_disable ? "Stack" : "Range"), info.object_size, ptr); - live_objects->push_back(AllocObject(ptr, info.object_size, - MUST_BE_ON_HEAP)); - } -} - -static const char kUnnamedProcSelfMapEntry[] = "UNNAMED"; - -// This function takes some fields from a /proc/self/maps line: -// -// start_address start address of a memory region. -// end_address end address of a memory region -// permissions rwx + private/shared bit -// filename filename of the mapped file -// -// If the region is not writeable, then it cannot have any heap -// pointers in it, otherwise we record it as a candidate live region -// to get filtered later. -static void RecordGlobalDataLocked(uintptr_t start_address, - uintptr_t end_address, - const char* permissions, - const char* filename) { - RAW_DCHECK(heap_checker_lock.IsHeld(), ""); - // Ignore non-writeable regions. - if (strchr(permissions, 'w') == NULL) return; - if (filename == NULL || *filename == '\0') { - filename = kUnnamedProcSelfMapEntry; - } - RAW_VLOG(11, "Looking into %s: 0x%" PRIxPTR "..0x%" PRIxPTR, - filename, start_address, end_address); - (*library_live_objects)[filename]. - push_back(AllocObject(AsPtr(start_address), - end_address - start_address, - MAYBE_LIVE)); -} - -// See if 'library' from /proc/self/maps has base name 'library_base' -// i.e. contains it and has '.' or '-' after it. -static bool IsLibraryNamed(const char* library, const char* library_base) { - const char* p = hc_strstr(library, library_base); - size_t sz = strlen(library_base); - return p != NULL && (p[sz] == '.' || p[sz] == '-'); -} - -// static -void HeapLeakChecker::DisableLibraryAllocsLocked(const char* library, - uintptr_t start_address, - uintptr_t end_address) { - RAW_DCHECK(heap_checker_lock.IsHeld(), ""); - int depth = 0; - // TODO(maxim): maybe this should be extended to also use objdump - // and pick the text portion of the library more precisely. - if (IsLibraryNamed(library, "/libpthread") || - // libpthread has a lot of small "system" leaks we don't care about. - // In particular it allocates memory to store data supplied via - // pthread_setspecific (which can be the only pointer to a heap object). - IsLibraryNamed(library, "/libdl") || - // library loaders leak some "system" heap that we don't care about - IsLibraryNamed(library, "/libcrypto") || - // Sometimes libcrypto of OpenSSH is compiled with -fomit-frame-pointer - // (any library can be, of course, but this one often is because speed - // is so important for making crypto usable). We ignore all its - // allocations because we can't see the call stacks. We'd prefer - // to ignore allocations done in files/symbols that match - // "default_malloc_ex|default_realloc_ex" - // but that doesn't work when the end-result binary is stripped. - IsLibraryNamed(library, "/libjvm") || - // JVM has a lot of leaks we don't care about. - IsLibraryNamed(library, "/libzip") - // The JVM leaks java.util.zip.Inflater after loading classes. - ) { - depth = 1; // only disable allocation calls directly from the library code - } else if (IsLibraryNamed(library, "/ld") - // library loader leaks some "system" heap - // (e.g. thread-local storage) that we don't care about - ) { - depth = 2; // disable allocation calls directly from the library code - // and at depth 2 from it. - // We need depth 2 here solely because of a libc bug that - // forces us to jump through __memalign_hook and MemalignOverride hoops - // in tcmalloc.cc. - // Those buggy __libc_memalign() calls are in ld-linux.so and happen for - // thread-local storage allocations that we want to ignore here. - // We go with the depth-2 hack as a workaround for this libc bug: - // otherwise we'd need to extend MallocHook interface - // so that correct stack depth adjustment can be propagated from - // the exceptional case of MemalignOverride. - // Using depth 2 here should not mask real leaks because ld-linux.so - // does not call user code. - } - if (depth) { - RAW_VLOG(10, "Disabling allocations from %s at depth %d:", library, depth); - DisableChecksFromToLocked(AsPtr(start_address), AsPtr(end_address), depth); - if (IsLibraryNamed(library, "/libpthread") || - IsLibraryNamed(library, "/libdl") || - IsLibraryNamed(library, "/ld")) { - RAW_VLOG(10, "Global memory regions made by %s will be live data", - library); - if (global_region_caller_ranges == NULL) { - global_region_caller_ranges = - new(Allocator::Allocate(sizeof(GlobalRegionCallerRangeMap))) - GlobalRegionCallerRangeMap; - } - global_region_caller_ranges - ->insert(make_pair(end_address, start_address)); - } - } -} - -// static -HeapLeakChecker::ProcMapsResult HeapLeakChecker::UseProcMapsLocked( - ProcMapsTask proc_maps_task) { - RAW_DCHECK(heap_checker_lock.IsHeld(), ""); - // Need to provide own scratch memory to ProcMapsIterator: - ProcMapsIterator::Buffer buffer; - ProcMapsIterator it(0, &buffer); - if (!it.Valid()) { - int errsv = errno; - RAW_LOG(ERROR, "Could not open /proc/self/maps: errno=%d. " - "Libraries will not be handled correctly.", errsv); - return CANT_OPEN_PROC_MAPS; - } - uint64 start_address, end_address, file_offset; - int64 inode; - char *permissions, *filename; - bool saw_shared_lib = false; - bool saw_nonzero_inode = false; - bool saw_shared_lib_with_nonzero_inode = false; - while (it.Next(&start_address, &end_address, &permissions, - &file_offset, &inode, &filename)) { - if (start_address >= end_address) { - // Warn if a line we can be interested in is ill-formed: - if (inode != 0) { - RAW_LOG(ERROR, "Errors reading /proc/self/maps. " - "Some global memory regions will not " - "be handled correctly."); - } - // Silently skip other ill-formed lines: some are possible - // probably due to the interplay of how /proc/self/maps is updated - // while we read it in chunks in ProcMapsIterator and - // do things in this loop. - continue; - } - // Determine if any shared libraries are present (this is the same - // list of extensions as is found in pprof). We want to ignore - // 'fake' libraries with inode 0 when determining. However, some - // systems don't share inodes via /proc, so we turn off this check - // if we don't see any evidence that we're getting inode info. - if (inode != 0) { - saw_nonzero_inode = true; - } - if ((hc_strstr(filename, "lib") && hc_strstr(filename, ".so")) || - hc_strstr(filename, ".dll") || - // not all .dylib filenames start with lib. .dylib is big enough - // that we are unlikely to get false matches just checking that. - hc_strstr(filename, ".dylib") || hc_strstr(filename, ".bundle")) { - saw_shared_lib = true; - if (inode != 0) { - saw_shared_lib_with_nonzero_inode = true; - } - } - - switch (proc_maps_task) { - case DISABLE_LIBRARY_ALLOCS: - // All lines starting like - // "401dc000-4030f000 r??p 00132000 03:01 13991972 lib/bin" - // identify a data and code sections of a shared library or our binary - if (inode != 0 && strncmp(permissions, "r-xp", 4) == 0) { - DisableLibraryAllocsLocked(filename, start_address, end_address); - } - break; - case RECORD_GLOBAL_DATA: - RecordGlobalDataLocked(start_address, end_address, - permissions, filename); - break; - default: - RAW_CHECK(0, ""); - } - } - // If /proc/self/maps is reporting inodes properly (we saw a - // non-zero inode), then we only say we saw a shared lib if we saw a - // 'real' one, with a non-zero inode. - if (saw_nonzero_inode) { - saw_shared_lib = saw_shared_lib_with_nonzero_inode; - } - if (!saw_shared_lib) { - RAW_LOG(ERROR, "No shared libs detected. Will likely report false leak " - "positives for statically linked executables."); - return NO_SHARED_LIBS_IN_PROC_MAPS; - } - return PROC_MAPS_USED; -} - -// Total number and size of live objects dropped from the profile; -// (re)initialized in IgnoreAllLiveObjectsLocked. -static int64 live_objects_total; -static int64 live_bytes_total; - -// pid of the thread that is doing the current leak check -// (protected by our lock; IgnoreAllLiveObjectsLocked sets it) -static pid_t self_thread_pid = 0; - -// Status of our thread listing callback execution -// (protected by our lock; used from within IgnoreAllLiveObjectsLocked) -static enum { - CALLBACK_NOT_STARTED, - CALLBACK_STARTED, - CALLBACK_COMPLETED, -} thread_listing_status = CALLBACK_NOT_STARTED; - -// Ideally to avoid deadlocks this function should not result in any libc -// or other function calls that might need to lock a mutex: -// It is called when all threads of a process are stopped -// at arbitrary points thus potentially holding those locks. -// -// In practice we are calling some simple i/o and sprintf-type library functions -// for logging messages, but use only our own LowLevelAlloc::Arena allocator. -// -// This is known to be buggy: the library i/o function calls are able to cause -// deadlocks when they request a lock that a stopped thread happens to hold. -// This issue as far as we know have so far not resulted in any deadlocks -// in practice, so for now we are taking our chance that the deadlocks -// have insignificant frequency. -// -// If such deadlocks become a problem we should make the i/o calls -// into appropriately direct system calls (or eliminate them), -// in particular write() is not safe and vsnprintf() is potentially dangerous -// due to reliance on locale functions (these are called through RAW_LOG -// and in other ways). -// - -#if defined(HAVE_LINUX_PTRACE_H) && defined(HAVE_SYS_SYSCALL_H) && defined(DUMPER) -# if (defined(__i386__) || defined(__x86_64)) -# define THREAD_REGS i386_regs -# elif defined(__PPC__) -# define THREAD_REGS ppc_regs -# endif -#endif - -/*static*/ int HeapLeakChecker::IgnoreLiveThreadsLocked(void* parameter, - int num_threads, - pid_t* thread_pids, - va_list /*ap*/) { - RAW_DCHECK(heap_checker_lock.IsHeld(), ""); - thread_listing_status = CALLBACK_STARTED; - RAW_VLOG(11, "Found %d threads (from pid %d)", num_threads, getpid()); - - if (FLAGS_heap_check_ignore_global_live) { - UseProcMapsLocked(RECORD_GLOBAL_DATA); - } - - // We put the registers from other threads here - // to make pointers stored in them live. - vector > thread_registers; - - int failures = 0; - for (int i = 0; i < num_threads; ++i) { - // the leak checking thread itself is handled - // specially via self_thread_stack, not here: - if (thread_pids[i] == self_thread_pid) continue; - RAW_VLOG(11, "Handling thread with pid %d", thread_pids[i]); -#ifdef THREAD_REGS - THREAD_REGS thread_regs; -#define sys_ptrace(r, p, a, d) syscall(SYS_ptrace, (r), (p), (a), (d)) - // We use sys_ptrace to avoid thread locking - // because this is called from TCMalloc_ListAllProcessThreads - // when all but this thread are suspended. - if (sys_ptrace(PTRACE_GETREGS, thread_pids[i], NULL, &thread_regs) == 0) { - // Need to use SP to get all the data from the very last stack frame: - COMPILE_ASSERT(sizeof(thread_regs.SP) == sizeof(void*), - SP_register_does_not_look_like_a_pointer); - RegisterStackLocked(reinterpret_cast(thread_regs.SP)); - // Make registers live (just in case PTRACE_ATTACH resulted in some - // register pointers still being in the registers and not on the stack): - for (void** p = reinterpret_cast(&thread_regs); - p < reinterpret_cast(&thread_regs + 1); ++p) { - RAW_VLOG(12, "Thread register %p", *p); - thread_registers.push_back(*p); - } - } else { - failures += 1; - } -#else - failures += 1; -#endif - } - // Use all the collected thread (stack) liveness sources: - IgnoreLiveObjectsLocked("threads stack data", ""); - if (thread_registers.size()) { - // Make thread registers be live heap data sources. - // we rely here on the fact that vector is in one memory chunk: - RAW_VLOG(11, "Live registers at %p of %" PRIuS " bytes", - &thread_registers[0], thread_registers.size() * sizeof(void*)); - live_objects->push_back(AllocObject(&thread_registers[0], - thread_registers.size() * sizeof(void*), - THREAD_REGISTERS)); - IgnoreLiveObjectsLocked("threads register data", ""); - } - // Do all other liveness walking while all threads are stopped: - IgnoreNonThreadLiveObjectsLocked(); - // Can now resume the threads: - TCMalloc_ResumeAllProcessThreads(num_threads, thread_pids); - thread_listing_status = CALLBACK_COMPLETED; - return failures; -} - -// Stack top of the thread that is doing the current leak check -// (protected by our lock; IgnoreAllLiveObjectsLocked sets it) -static const void* self_thread_stack_top; - -// static -void HeapLeakChecker::IgnoreNonThreadLiveObjectsLocked() { - RAW_DCHECK(heap_checker_lock.IsHeld(), ""); - RAW_DCHECK(MemoryRegionMap::LockIsHeld(), ""); - RAW_VLOG(11, "Handling self thread with pid %d", self_thread_pid); - // Register our own stack: - - // Important that all stack ranges (including the one here) - // are known before we start looking at them - // in MakeDisabledLiveCallbackLocked: - RegisterStackLocked(self_thread_stack_top); - IgnoreLiveObjectsLocked("stack data", ""); - - // Make objects we were told to ignore live: - if (ignored_objects) { - for (IgnoredObjectsMap::const_iterator object = ignored_objects->begin(); - object != ignored_objects->end(); ++object) { - const void* ptr = AsPtr(object->first); - RAW_VLOG(11, "Ignored live object at %p of %" PRIuS " bytes", - ptr, object->second); - live_objects-> - push_back(AllocObject(ptr, object->second, MUST_BE_ON_HEAP)); - // we do this liveness check for ignored_objects before doing any - // live heap walking to make sure it does not fail needlessly: - size_t object_size; - if (!(heap_profile->FindAlloc(ptr, &object_size) && - object->second == object_size)) { - RAW_LOG(FATAL, "Object at %p of %" PRIuS " bytes from an" - " IgnoreObject() has disappeared", ptr, object->second); - } - } - IgnoreLiveObjectsLocked("ignored objects", ""); - } - - // Treat objects that were allocated when a Disabler was live as - // roots. I.e., if X was allocated while a Disabler was active, - // and Y is reachable from X, arrange that neither X nor Y are - // treated as leaks. - heap_profile->IterateAllocs(MakeIgnoredObjectsLiveCallbackLocked); - IgnoreLiveObjectsLocked("disabled objects", ""); - - // Make code-address-disabled objects live and ignored: - // This in particular makes all thread-specific data live - // because the basic data structure to hold pointers to thread-specific data - // is allocated from libpthreads and we have range-disabled that - // library code with UseProcMapsLocked(DISABLE_LIBRARY_ALLOCS); - // so now we declare all thread-specific data reachable from there as live. - heap_profile->IterateAllocs(MakeDisabledLiveCallbackLocked); - IgnoreLiveObjectsLocked("disabled code", ""); - - // Actually make global data live: - if (FLAGS_heap_check_ignore_global_live) { - bool have_null_region_callers = false; - for (LibraryLiveObjectsStacks::iterator l = library_live_objects->begin(); - l != library_live_objects->end(); ++l) { - RAW_CHECK(live_objects->empty(), ""); - // Process library_live_objects in l->second - // filtering them by MemoryRegionMap: - // It's safe to iterate over MemoryRegionMap - // w/o locks here as we are inside MemoryRegionMap::Lock(): - RAW_DCHECK(MemoryRegionMap::LockIsHeld(), ""); - // The only change to MemoryRegionMap possible in this loop - // is region addition as a result of allocating more memory - // for live_objects. This won't invalidate the RegionIterator - // or the intent of the loop. - // --see the comment by MemoryRegionMap::BeginRegionLocked(). - for (MemoryRegionMap::RegionIterator region = - MemoryRegionMap::BeginRegionLocked(); - region != MemoryRegionMap::EndRegionLocked(); ++region) { - // "region" from MemoryRegionMap is to be subtracted from - // (tentatively live) regions in l->second - // if it has a stack inside or it was allocated by - // a non-special caller (not one covered by a range - // in global_region_caller_ranges). - // This will in particular exclude all memory chunks used - // by the heap itself as well as what's been allocated with - // any allocator on top of mmap. - bool subtract = true; - if (!region->is_stack && global_region_caller_ranges) { - if (region->caller() == static_cast(NULL)) { - have_null_region_callers = true; - } else { - GlobalRegionCallerRangeMap::const_iterator iter - = global_region_caller_ranges->upper_bound(region->caller()); - if (iter != global_region_caller_ranges->end()) { - RAW_DCHECK(iter->first > region->caller(), ""); - if (iter->second < region->caller()) { // in special region - subtract = false; - } - } - } - } - if (subtract) { - // The loop puts the result of filtering l->second into live_objects: - for (LiveObjectsStack::const_iterator i = l->second.begin(); - i != l->second.end(); ++i) { - // subtract *region from *i - uintptr_t start = AsInt(i->ptr); - uintptr_t end = start + i->size; - if (region->start_addr <= start && end <= region->end_addr) { - // full deletion due to subsumption - } else if (start < region->start_addr && - region->end_addr < end) { // cutting-out split - live_objects->push_back(AllocObject(i->ptr, - region->start_addr - start, - IN_GLOBAL_DATA)); - live_objects->push_back(AllocObject(AsPtr(region->end_addr), - end - region->end_addr, - IN_GLOBAL_DATA)); - } else if (region->end_addr > start && - region->start_addr <= start) { // cut from start - live_objects->push_back(AllocObject(AsPtr(region->end_addr), - end - region->end_addr, - IN_GLOBAL_DATA)); - } else if (region->start_addr > start && - region->start_addr < end) { // cut from end - live_objects->push_back(AllocObject(i->ptr, - region->start_addr - start, - IN_GLOBAL_DATA)); - } else { // pass: no intersection - live_objects->push_back(AllocObject(i->ptr, i->size, - IN_GLOBAL_DATA)); - } - } - // Move live_objects back into l->second - // for filtering by the next region. - live_objects->swap(l->second); - live_objects->clear(); - } - } - // Now get and use live_objects from the final version of l->second: - if (VLOG_IS_ON(11)) { - for (LiveObjectsStack::const_iterator i = l->second.begin(); - i != l->second.end(); ++i) { - RAW_VLOG(11, "Library live region at %p of %" PRIuPTR " bytes", - i->ptr, i->size); - } - } - live_objects->swap(l->second); - IgnoreLiveObjectsLocked("in globals of\n ", l->first.c_str()); - } - if (have_null_region_callers) { - RAW_LOG(ERROR, "Have memory regions w/o callers: " - "might report false leaks"); - } - Allocator::DeleteAndNull(&library_live_objects); - } -} - -// Callback for TCMalloc_ListAllProcessThreads in IgnoreAllLiveObjectsLocked below -// to test/verify that we have just the one main thread, in which case -// we can do everything in that main thread, -// so that CPU profiler can collect all its samples. -// Returns the number of threads in the process. -static int IsOneThread(void* parameter, int num_threads, - pid_t* thread_pids, va_list ap) { - if (num_threads != 1) { - RAW_LOG(WARNING, "Have threads: Won't CPU-profile the bulk of leak " - "checking work happening in IgnoreLiveThreadsLocked!"); - } - TCMalloc_ResumeAllProcessThreads(num_threads, thread_pids); - return num_threads; -} - -// Dummy for IgnoreAllLiveObjectsLocked below. -// Making it global helps with compiler warnings. -static va_list dummy_ap; - -// static -void HeapLeakChecker::IgnoreAllLiveObjectsLocked(const void* self_stack_top) { - RAW_DCHECK(heap_checker_lock.IsHeld(), ""); - RAW_CHECK(live_objects == NULL, ""); - live_objects = new(Allocator::Allocate(sizeof(LiveObjectsStack))) - LiveObjectsStack; - stack_tops = new(Allocator::Allocate(sizeof(StackTopSet))) StackTopSet; - // reset the counts - live_objects_total = 0; - live_bytes_total = 0; - // Reduce max_heap_object_size to FLAGS_heap_check_max_pointer_offset - // for the time of leak check. - // FLAGS_heap_check_max_pointer_offset caps max_heap_object_size - // to manage reasonably low chances of random bytes - // appearing to be pointing into large actually leaked heap objects. - const size_t old_max_heap_object_size = max_heap_object_size; - max_heap_object_size = ( - FLAGS_heap_check_max_pointer_offset != -1 - ? min(size_t(FLAGS_heap_check_max_pointer_offset), max_heap_object_size) - : max_heap_object_size); - // Record global data as live: - if (FLAGS_heap_check_ignore_global_live) { - library_live_objects = - new(Allocator::Allocate(sizeof(LibraryLiveObjectsStacks))) - LibraryLiveObjectsStacks; - } - // Ignore all thread stacks: - thread_listing_status = CALLBACK_NOT_STARTED; - bool need_to_ignore_non_thread_objects = true; - self_thread_pid = getpid(); - self_thread_stack_top = self_stack_top; - if (FLAGS_heap_check_ignore_thread_live) { - // In case we are doing CPU profiling we'd like to do all the work - // in the main thread, not in the special thread created by - // TCMalloc_ListAllProcessThreads, so that CPU profiler can - // collect all its samples. The machinery of - // TCMalloc_ListAllProcessThreads conflicts with the CPU profiler - // by also relying on signals and ::sigaction. We can do this - // (run everything in the main thread) safely only if there's just - // the main thread itself in our process. This variable reflects - // these two conditions: - bool want_and_can_run_in_main_thread = - ProfilingIsEnabledForAllThreads() && - TCMalloc_ListAllProcessThreads(NULL, IsOneThread) == 1; - // When the normal path of TCMalloc_ListAllProcessThreads below is taken, - // we fully suspend the threads right here before any liveness checking - // and keep them suspended for the whole time of liveness checking - // inside of the IgnoreLiveThreadsLocked callback. - // (The threads can't (de)allocate due to lock on the delete hook but - // if not suspended they could still mess with the pointer - // graph while we walk it). - int r = want_and_can_run_in_main_thread - ? IgnoreLiveThreadsLocked(NULL, 1, &self_thread_pid, dummy_ap) - : TCMalloc_ListAllProcessThreads(NULL, IgnoreLiveThreadsLocked); - need_to_ignore_non_thread_objects = r < 0; - if (r < 0) { - RAW_LOG(WARNING, "Thread finding failed with %d errno=%d", r, errno); - if (thread_listing_status == CALLBACK_COMPLETED) { - RAW_LOG(INFO, "Thread finding callback " - "finished ok; hopefully everything is fine"); - need_to_ignore_non_thread_objects = false; - } else if (thread_listing_status == CALLBACK_STARTED) { - RAW_LOG(FATAL, "Thread finding callback was " - "interrupted or crashed; can't fix this"); - } else { // CALLBACK_NOT_STARTED - RAW_LOG(ERROR, "Could not find thread stacks. " - "Will likely report false leak positives."); - } - } else if (r != 0) { - RAW_LOG(ERROR, "Thread stacks not found for %d threads. " - "Will likely report false leak positives.", r); - } else { - RAW_VLOG(11, "Thread stacks appear to be found for all threads"); - } - } else { - RAW_LOG(WARNING, "Not looking for thread stacks; " - "objects reachable only from there " - "will be reported as leaks"); - } - // Do all other live data ignoring here if we did not do it - // within thread listing callback with all threads stopped. - if (need_to_ignore_non_thread_objects) { - if (FLAGS_heap_check_ignore_global_live) { - UseProcMapsLocked(RECORD_GLOBAL_DATA); - } - IgnoreNonThreadLiveObjectsLocked(); - } - if (live_objects_total) { - RAW_VLOG(10, "Ignoring %" PRId64 " reachable objects of %" PRId64 " bytes", - live_objects_total, live_bytes_total); - } - // Free these: we made them here and heap_profile never saw them - Allocator::DeleteAndNull(&live_objects); - Allocator::DeleteAndNull(&stack_tops); - max_heap_object_size = old_max_heap_object_size; // reset this var -} - -// Alignment at which we should consider pointer positions -// in IgnoreLiveObjectsLocked. Will normally use the value of -// FLAGS_heap_check_pointer_source_alignment. -static size_t pointer_source_alignment = kPointerSourceAlignment; -// Global lock for HeapLeakChecker::DoNoLeaks -// to protect pointer_source_alignment. -static SpinLock alignment_checker_lock(SpinLock::LINKER_INITIALIZED); - -// This function changes the live bits in the heap_profile-table's state: -// we only record the live objects to be skipped. -// -// When checking if a byte sequence points to a heap object we use -// HeapProfileTable::FindInsideAlloc to handle both pointers to -// the start and inside of heap-allocated objects. -// The "inside" case needs to be checked to support -// at least the following relatively common cases: -// - C++ arrays allocated with new FooClass[size] for classes -// with destructors have their size recorded in a sizeof(int) field -// before the place normal pointers point to. -// - basic_string<>-s for e.g. the C++ library of gcc 3.4 -// have the meta-info in basic_string<...>::_Rep recorded -// before the place normal pointers point to. -// - Multiple-inherited objects have their pointers when cast to -// different base classes pointing inside of the actually -// allocated object. -// - Sometimes reachability pointers point to member objects of heap objects, -// and then those member objects point to the full heap object. -// - Third party UnicodeString: it stores a 32-bit refcount -// (in both 32-bit and 64-bit binaries) as the first uint32 -// in the allocated memory and a normal pointer points at -// the second uint32 behind the refcount. -// By finding these additional objects here -// we slightly increase the chance to mistake random memory bytes -// for a pointer and miss a leak in a particular run of a binary. -// -/*static*/ void HeapLeakChecker::IgnoreLiveObjectsLocked(const char* name, - const char* name2) { - RAW_DCHECK(heap_checker_lock.IsHeld(), ""); - int64 live_object_count = 0; - int64 live_byte_count = 0; - while (!live_objects->empty()) { - const char* object = - reinterpret_cast(live_objects->back().ptr); - size_t size = live_objects->back().size; - const ObjectPlacement place = live_objects->back().place; - live_objects->pop_back(); - if (place == MUST_BE_ON_HEAP && heap_profile->MarkAsLive(object)) { - live_object_count += 1; - live_byte_count += size; - } - RAW_VLOG(13, "Looking for heap pointers in %p of %" PRIuS " bytes", - object, size); - const char* const whole_object = object; - size_t const whole_size = size; - // Try interpretting any byte sequence in object,size as a heap pointer: - const size_t remainder = AsInt(object) % pointer_source_alignment; - if (remainder) { - object += pointer_source_alignment - remainder; - if (size >= pointer_source_alignment - remainder) { - size -= pointer_source_alignment - remainder; - } else { - size = 0; - } - } - if (size < sizeof(void*)) continue; - -#ifdef NO_FRAME_POINTER - // Frame pointer omission requires us to use libunwind, which uses direct - // mmap and munmap system calls, and that needs special handling. - if (name2 == kUnnamedProcSelfMapEntry) { - static const uintptr_t page_mask = ~(getpagesize() - 1); - const uintptr_t addr = reinterpret_cast(object); - if ((addr & page_mask) == 0 && (size & page_mask) == 0) { - // This is an object we slurped from /proc/self/maps. - // It may or may not be readable at this point. - // - // In case all the above conditions made a mistake, and the object is - // not related to libunwind, we also verify that it's not readable - // before ignoring it. - if (msync(const_cast(object), size, MS_ASYNC) != 0) { - // Skip unreadable object, so we don't crash trying to sweep it. - RAW_VLOG(0, "Ignoring inaccessible object [%p, %p) " - "(msync error %d (%s))", - object, object + size, errno, strerror(errno)); - continue; - } - } - } -#endif - - const char* const max_object = object + size - sizeof(void*); - while (object <= max_object) { - // potentially unaligned load: - const uintptr_t addr = *reinterpret_cast(object); - // Do fast check before the more expensive HaveOnHeapLocked lookup: - // this code runs for all memory words that are potentially pointers: - const bool can_be_on_heap = - // Order tests by the likelyhood of the test failing in 64/32 bit modes. - // Yes, this matters: we either lose 5..6% speed in 32 bit mode - // (which is already slower) or by a factor of 1.5..1.91 in 64 bit mode. - // After the alignment test got dropped the above performance figures - // must have changed; might need to revisit this. -#if defined(__x86_64__) - addr <= max_heap_address && // <= is for 0-sized object with max addr - min_heap_address <= addr; -#else - min_heap_address <= addr && - addr <= max_heap_address; // <= is for 0-sized object with max addr -#endif - if (can_be_on_heap) { - const void* ptr = reinterpret_cast(addr); - // Too expensive (inner loop): manually uncomment when debugging: - // RAW_VLOG(17, "Trying pointer to %p at %p", ptr, object); - size_t object_size; - if (HaveOnHeapLocked(&ptr, &object_size) && - heap_profile->MarkAsLive(ptr)) { - // We take the (hopefully low) risk here of encountering by accident - // a byte sequence in memory that matches an address of - // a heap object which is in fact leaked. - // I.e. in very rare and probably not repeatable/lasting cases - // we might miss some real heap memory leaks. - RAW_VLOG(14, "Found pointer to %p of %" PRIuS " bytes at %p " - "inside %p of size %" PRIuS "", - ptr, object_size, object, whole_object, whole_size); - if (VLOG_IS_ON(15)) { - // log call stacks to help debug how come something is not a leak - HeapProfileTable::AllocInfo alloc; - if (!heap_profile->FindAllocDetails(ptr, &alloc)) { - RAW_LOG(FATAL, "FindAllocDetails failed on ptr %p", ptr); - } - RAW_LOG(INFO, "New live %p object's alloc stack:", ptr); - for (int i = 0; i < alloc.stack_depth; ++i) { - RAW_LOG(INFO, " @ %p", alloc.call_stack[i]); - } - } - live_object_count += 1; - live_byte_count += object_size; - live_objects->push_back(AllocObject(ptr, object_size, - IGNORED_ON_HEAP)); - } - } - object += pointer_source_alignment; - } - } - live_objects_total += live_object_count; - live_bytes_total += live_byte_count; - if (live_object_count) { - RAW_VLOG(10, "Removed %" PRId64 " live heap objects of %" PRId64 " bytes: %s%s", - live_object_count, live_byte_count, name, name2); - } -} - -//---------------------------------------------------------------------- -// HeapLeakChecker leak check disabling components -//---------------------------------------------------------------------- - -// static -void HeapLeakChecker::DisableChecksIn(const char* pattern) { - RAW_LOG(WARNING, "DisableChecksIn(%s) is ignored", pattern); -} - -// static -void HeapLeakChecker::DoIgnoreObject(const void* ptr) { - SpinLockHolder l(&heap_checker_lock); - if (!heap_checker_on) return; - size_t object_size; - if (!HaveOnHeapLocked(&ptr, &object_size)) { - RAW_LOG(ERROR, "No live heap object at %p to ignore", ptr); - } else { - RAW_VLOG(10, "Going to ignore live object at %p of %" PRIuS " bytes", - ptr, object_size); - if (ignored_objects == NULL) { - ignored_objects = new(Allocator::Allocate(sizeof(IgnoredObjectsMap))) - IgnoredObjectsMap; - } - if (!ignored_objects->insert(make_pair(AsInt(ptr), object_size)).second) { - RAW_LOG(WARNING, "Object at %p is already being ignored", ptr); - } - } -} - -// static -void HeapLeakChecker::UnIgnoreObject(const void* ptr) { - SpinLockHolder l(&heap_checker_lock); - if (!heap_checker_on) return; - size_t object_size; - if (!HaveOnHeapLocked(&ptr, &object_size)) { - RAW_LOG(FATAL, "No live heap object at %p to un-ignore", ptr); - } else { - bool found = false; - if (ignored_objects) { - IgnoredObjectsMap::iterator object = ignored_objects->find(AsInt(ptr)); - if (object != ignored_objects->end() && object_size == object->second) { - ignored_objects->erase(object); - found = true; - RAW_VLOG(10, "Now not going to ignore live object " - "at %p of %" PRIuS " bytes", ptr, object_size); - } - } - if (!found) RAW_LOG(FATAL, "Object at %p has not been ignored", ptr); - } -} - -//---------------------------------------------------------------------- -// HeapLeakChecker non-static functions -//---------------------------------------------------------------------- - -char* HeapLeakChecker::MakeProfileNameLocked() { - RAW_DCHECK(lock_->IsHeld(), ""); - RAW_DCHECK(heap_checker_lock.IsHeld(), ""); - const int len = profile_name_prefix->size() + strlen(name_) + 5 + - strlen(HeapProfileTable::kFileExt) + 1; - char* file_name = reinterpret_cast(Allocator::Allocate(len)); - snprintf(file_name, len, "%s.%s-end%s", - profile_name_prefix->c_str(), name_, - HeapProfileTable::kFileExt); - return file_name; -} - -void HeapLeakChecker::Create(const char *name, bool make_start_snapshot) { - SpinLockHolder l(lock_); - name_ = NULL; // checker is inactive - start_snapshot_ = NULL; - has_checked_ = false; - inuse_bytes_increase_ = 0; - inuse_allocs_increase_ = 0; - keep_profiles_ = false; - char* n = new char[strlen(name) + 1]; // do this before we lock - IgnoreObject(n); // otherwise it might be treated as live due to our stack - { // Heap activity in other threads is paused for this whole scope. - SpinLockHolder al(&alignment_checker_lock); - SpinLockHolder hl(&heap_checker_lock); - MemoryRegionMap::LockHolder ml; - if (heap_checker_on && profile_name_prefix != NULL) { - RAW_DCHECK(strchr(name, '/') == NULL, "must be a simple name"); - memcpy(n, name, strlen(name) + 1); - name_ = n; // checker is active - if (make_start_snapshot) { - start_snapshot_ = heap_profile->TakeSnapshot(); - } - - const HeapProfileTable::Stats& t = heap_profile->total(); - const size_t start_inuse_bytes = t.alloc_size - t.free_size; - const size_t start_inuse_allocs = t.allocs - t.frees; - RAW_VLOG(10, "Start check \"%s\" profile: %" PRIuS " bytes " - "in %" PRIuS " objects", - name_, start_inuse_bytes, start_inuse_allocs); - } else { - RAW_LOG(WARNING, "Heap checker is not active, " - "hence checker \"%s\" will do nothing!", name); - RAW_LOG(WARNING, "To activate set the HEAPCHECK environment variable.\n"); - } - } - if (name_ == NULL) { - UnIgnoreObject(n); - delete[] n; // must be done after we unlock - } -} - -HeapLeakChecker::HeapLeakChecker(const char *name) : lock_(new SpinLock) { - RAW_DCHECK(strcmp(name, "_main_") != 0, "_main_ is reserved"); - Create(name, true/*create start_snapshot_*/); -} - -HeapLeakChecker::HeapLeakChecker() : lock_(new SpinLock) { - if (FLAGS_heap_check_before_constructors) { - // We want to check for leaks of objects allocated during global - // constructors (i.e., objects allocated already). So we do not - // create a baseline snapshot and hence check for leaks of objects - // that may have already been created. - Create("_main_", false); - } else { - // We want to ignore leaks of objects allocated during global - // constructors (i.e., objects allocated already). So we snapshot - // the current heap contents and use them as a baseline that is - // not reported by the leak checker. - Create("_main_", true); - } -} - -ssize_t HeapLeakChecker::BytesLeaked() const { - SpinLockHolder l(lock_); - if (!has_checked_) { - RAW_LOG(FATAL, "*NoLeaks|SameHeap must execute before this call"); - } - return inuse_bytes_increase_; -} - -ssize_t HeapLeakChecker::ObjectsLeaked() const { - SpinLockHolder l(lock_); - if (!has_checked_) { - RAW_LOG(FATAL, "*NoLeaks|SameHeap must execute before this call"); - } - return inuse_allocs_increase_; -} - -// Save pid of main thread for using in naming dump files -static int32 main_thread_pid = getpid(); -#ifdef HAVE_PROGRAM_INVOCATION_NAME -#ifdef __UCLIBC__ -extern const char* program_invocation_name; -extern const char* program_invocation_short_name; -#else -extern char* program_invocation_name; -extern char* program_invocation_short_name; -#endif -static const char* invocation_name() { return program_invocation_short_name; } -static string invocation_path() { return program_invocation_name; } -#else -static const char* invocation_name() { return ""; } -static string invocation_path() { return ""; } -#endif - -// Prints commands that users can run to get more information -// about the reported leaks. -static void SuggestPprofCommand(const char* pprof_file_arg) { - // Extra help information to print for the user when the test is - // being run in a way where the straightforward pprof command will - // not suffice. - string extra_help; - - // Common header info to print for remote runs - const string remote_header = - "This program is being executed remotely and therefore the pprof\n" - "command printed above will not work. Either run this program\n" - "locally, or adjust the pprof command as follows to allow it to\n" - "work on your local machine:\n"; - - // Extra command for fetching remote data - string fetch_cmd; - - RAW_LOG(WARNING, - "\n\n" - "If the preceding stack traces are not enough to find " - "the leaks, try running THIS shell command:\n\n" - "%s%s %s \"%s\" --inuse_objects --lines --heapcheck " - " --edgefraction=1e-10 --nodefraction=1e-10 --gv\n" - "\n" - "%s" - "If you are still puzzled about why the leaks are " - "there, try rerunning this program with " - "HEAP_CHECK_TEST_POINTER_ALIGNMENT=1 and/or with " - "HEAP_CHECK_MAX_POINTER_OFFSET=-1\n" - "If the leak report occurs in a small fraction of runs, " - "try running with TCMALLOC_MAX_FREE_QUEUE_SIZE of few hundred MB " - "or with TCMALLOC_RECLAIM_MEMORY=false, " // only works for debugalloc - "it might help find leaks more repeatably\n", - fetch_cmd.c_str(), - "pprof", // works as long as pprof is on your path - invocation_path().c_str(), - pprof_file_arg, - extra_help.c_str() - ); -} - -bool HeapLeakChecker::DoNoLeaks(ShouldSymbolize should_symbolize) { - SpinLockHolder l(lock_); - // The locking also helps us keep the messages - // for the two checks close together. - SpinLockHolder al(&alignment_checker_lock); - - // thread-safe: protected by alignment_checker_lock - static bool have_disabled_hooks_for_symbolize = false; - // Once we've checked for leaks and symbolized the results once, it's - // not safe to do it again. This is because in order to symbolize - // safely, we had to disable all the malloc hooks here, so we no - // longer can be confident we've collected all the data we need. - if (have_disabled_hooks_for_symbolize) { - RAW_LOG(FATAL, "Must not call heap leak checker manually after " - " program-exit's automatic check."); - } - - HeapProfileTable::Snapshot* leaks = NULL; - char* pprof_file = NULL; - - { - // Heap activity in other threads is paused during this function - // (i.e. until we got all profile difference info). - SpinLockHolder hl(&heap_checker_lock); - if (heap_checker_on == false) { - if (name_ != NULL) { // leak checking enabled when created the checker - RAW_LOG(WARNING, "Heap leak checker got turned off after checker " - "\"%s\" has been created, no leak check is being done for it!", - name_); - } - return true; - } - - // Update global_region_caller_ranges. They may need to change since - // e.g. initialization because shared libraries might have been loaded or - // unloaded. - Allocator::DeleteAndNullIfNot(&global_region_caller_ranges); - ProcMapsResult pm_result = UseProcMapsLocked(DISABLE_LIBRARY_ALLOCS); - RAW_CHECK(pm_result == PROC_MAPS_USED, ""); - - // Keep track of number of internally allocated objects so we - // can detect leaks in the heap-leak-checket itself - const int initial_allocs = Allocator::alloc_count(); - - if (name_ == NULL) { - RAW_LOG(FATAL, "Heap leak checker must not be turned on " - "after construction of a HeapLeakChecker"); - } - - MemoryRegionMap::LockHolder ml; - int a_local_var; // Use our stack ptr to make stack data live: - - // Make the heap profile, other threads are locked out. - HeapProfileTable::Snapshot* base = - reinterpret_cast(start_snapshot_); - RAW_DCHECK(FLAGS_heap_check_pointer_source_alignment > 0, ""); - pointer_source_alignment = FLAGS_heap_check_pointer_source_alignment; - IgnoreAllLiveObjectsLocked(&a_local_var); - leaks = heap_profile->NonLiveSnapshot(base); - - inuse_bytes_increase_ = static_cast(leaks->total().alloc_size); - inuse_allocs_increase_ = static_cast(leaks->total().allocs); - if (leaks->Empty()) { - heap_profile->ReleaseSnapshot(leaks); - leaks = NULL; - - // We can only check for internal leaks along the no-user-leak - // path since in the leak path we temporarily release - // heap_checker_lock and another thread can come in and disturb - // allocation counts. - if (Allocator::alloc_count() != initial_allocs) { - RAW_LOG(FATAL, "Internal HeapChecker leak of %d objects ; %d -> %d", - Allocator::alloc_count() - initial_allocs, - initial_allocs, Allocator::alloc_count()); - } - } else if (FLAGS_heap_check_test_pointer_alignment) { - if (pointer_source_alignment == 1) { - RAW_LOG(WARNING, "--heap_check_test_pointer_alignment has no effect: " - "--heap_check_pointer_source_alignment was already set to 1"); - } else { - // Try with reduced pointer aligment - pointer_source_alignment = 1; - IgnoreAllLiveObjectsLocked(&a_local_var); - HeapProfileTable::Snapshot* leaks_wo_align = - heap_profile->NonLiveSnapshot(base); - pointer_source_alignment = FLAGS_heap_check_pointer_source_alignment; - if (leaks_wo_align->Empty()) { - RAW_LOG(WARNING, "Found no leaks without pointer alignment: " - "something might be placing pointers at " - "unaligned addresses! This needs to be fixed."); - } else { - RAW_LOG(INFO, "Found leaks without pointer alignment as well: " - "unaligned pointers must not be the cause of leaks."); - RAW_LOG(INFO, "--heap_check_test_pointer_alignment did not help " - "to diagnose the leaks."); - } - heap_profile->ReleaseSnapshot(leaks_wo_align); - } - } - - if (leaks != NULL) { - pprof_file = MakeProfileNameLocked(); - } - } - - has_checked_ = true; - if (leaks == NULL) { - if (FLAGS_heap_check_max_pointer_offset == -1) { - RAW_LOG(WARNING, - "Found no leaks without max_pointer_offset restriction: " - "it's possible that the default value of " - "heap_check_max_pointer_offset flag is too low. " - "Do you use pointers with larger than that offsets " - "pointing in the middle of heap-allocated objects?"); - } - const HeapProfileTable::Stats& stats = heap_profile->total(); - RAW_VLOG(heap_checker_info_level, - "No leaks found for check \"%s\" " - "(but no 100%% guarantee that there aren't any): " - "found %" PRId64 " reachable heap objects of %" PRId64 " bytes", - name_, - int64(stats.allocs - stats.frees), - int64(stats.alloc_size - stats.free_size)); - } else { - if (should_symbolize == SYMBOLIZE) { - // To turn addresses into symbols, we need to fork, which is a - // problem if both parent and child end up trying to call the - // same malloc-hooks we've set up, at the same time. To avoid - // trouble, we turn off the hooks before symbolizing. Note that - // this makes it unsafe to ever leak-report again! Luckily, we - // typically only want to report once in a program's run, at the - // very end. - if (MallocHook::GetNewHook() == NewHook) - MallocHook::SetNewHook(NULL); - if (MallocHook::GetDeleteHook() == DeleteHook) - MallocHook::SetDeleteHook(NULL); - MemoryRegionMap::Shutdown(); - // Make sure all the hooks really got unset: - RAW_CHECK(MallocHook::GetNewHook() == NULL, ""); - RAW_CHECK(MallocHook::GetDeleteHook() == NULL, ""); - RAW_CHECK(MallocHook::GetMmapHook() == NULL, ""); - RAW_CHECK(MallocHook::GetSbrkHook() == NULL, ""); - have_disabled_hooks_for_symbolize = true; - leaks->ReportLeaks(name_, pprof_file, true); // true = should_symbolize - } else { - leaks->ReportLeaks(name_, pprof_file, false); - } - if (FLAGS_heap_check_identify_leaks) { - leaks->ReportIndividualObjects(); - } - - SuggestPprofCommand(pprof_file); - - { - SpinLockHolder hl(&heap_checker_lock); - heap_profile->ReleaseSnapshot(leaks); - Allocator::Free(pprof_file); - } - } - - return (leaks == NULL); -} - -HeapLeakChecker::~HeapLeakChecker() { - if (name_ != NULL) { // had leak checking enabled when created the checker - if (!has_checked_) { - RAW_LOG(FATAL, "Some *NoLeaks|SameHeap method" - " must be called on any created HeapLeakChecker"); - } - - // Deallocate any snapshot taken at start - if (start_snapshot_ != NULL) { - SpinLockHolder l(&heap_checker_lock); - heap_profile->ReleaseSnapshot( - reinterpret_cast(start_snapshot_)); - } - - UnIgnoreObject(name_); - delete[] name_; - name_ = NULL; - } - delete lock_; -} - -//---------------------------------------------------------------------- -// HeapLeakChecker overall heap check components -//---------------------------------------------------------------------- - -// static -bool HeapLeakChecker::IsActive() { - SpinLockHolder l(&heap_checker_lock); - return heap_checker_on; -} - -vector* HeapCleaner::heap_cleanups_ = NULL; - -// When a HeapCleaner object is intialized, add its function to the static list -// of cleaners to be run before leaks checking. -HeapCleaner::HeapCleaner(void_function f) { - if (heap_cleanups_ == NULL) - heap_cleanups_ = new vector; - heap_cleanups_->push_back(f); -} - -// Run all of the cleanup functions and delete the vector. -void HeapCleaner::RunHeapCleanups() { - if (!heap_cleanups_) - return; - for (int i = 0; i < heap_cleanups_->size(); i++) { - void (*f)(void) = (*heap_cleanups_)[i]; - f(); - } - delete heap_cleanups_; - heap_cleanups_ = NULL; -} - -// Program exit heap cleanup registered as a module object destructor. -// Will not get executed when we crash on a signal. -// -void HeapLeakChecker_RunHeapCleanups() { - if (FLAGS_heap_check == "local") // don't check heap in this mode - return; - { SpinLockHolder l(&heap_checker_lock); - // can get here (via forks?) with other pids - if (heap_checker_pid != getpid()) return; - } - HeapCleaner::RunHeapCleanups(); - if (!FLAGS_heap_check_after_destructors) HeapLeakChecker::DoMainHeapCheck(); -} - -static bool internal_init_start_has_run = false; - -// Called exactly once, before main() (but hopefully just before). -// This picks a good unique name for the dumped leak checking heap profiles. -// -// Because we crash when InternalInitStart is called more than once, -// it's fine that we hold heap_checker_lock only around pieces of -// this function: this is still enough for thread-safety w.r.t. other functions -// of this module. -// We can't hold heap_checker_lock throughout because it would deadlock -// on a memory allocation since our new/delete hooks can be on. -// -void HeapLeakChecker_InternalInitStart() { - { SpinLockHolder l(&heap_checker_lock); - RAW_CHECK(!internal_init_start_has_run, - "Heap-check constructor called twice. Perhaps you both linked" - " in the heap checker, and also used LD_PRELOAD to load it?"); - internal_init_start_has_run = true; - -#ifdef ADDRESS_SANITIZER - // AddressSanitizer's custom malloc conflicts with HeapChecker. - FLAGS_heap_check = ""; -#endif - - if (FLAGS_heap_check.empty()) { - // turns out we do not need checking in the end; can stop profiling - HeapLeakChecker::TurnItselfOffLocked(); - return; - } else if (RunningOnValgrind()) { - // There is no point in trying -- we'll just fail. - RAW_LOG(WARNING, "Can't run under Valgrind; will turn itself off"); - HeapLeakChecker::TurnItselfOffLocked(); - return; - } - } - - // Changing this to false can be useful when debugging heap-checker itself: - if (!FLAGS_heap_check_run_under_gdb && IsDebuggerAttached()) { - RAW_LOG(WARNING, "Someone is ptrace()ing us; will turn itself off"); - SpinLockHolder l(&heap_checker_lock); - HeapLeakChecker::TurnItselfOffLocked(); - return; - } - - { SpinLockHolder l(&heap_checker_lock); - if (!constructor_heap_profiling) { - RAW_LOG(FATAL, "Can not start so late. You have to enable heap checking " - "with HEAPCHECK=."); - } - } - - // Set all flags - RAW_DCHECK(FLAGS_heap_check_pointer_source_alignment > 0, ""); - if (FLAGS_heap_check == "minimal") { - // The least we can check. - FLAGS_heap_check_before_constructors = false; // from after main - // (ignore more) - FLAGS_heap_check_after_destructors = false; // to after cleanup - // (most data is live) - FLAGS_heap_check_ignore_thread_live = true; // ignore all live - FLAGS_heap_check_ignore_global_live = true; // ignore all live - } else if (FLAGS_heap_check == "normal") { - // Faster than 'minimal' and not much stricter. - FLAGS_heap_check_before_constructors = true; // from no profile (fast) - FLAGS_heap_check_after_destructors = false; // to after cleanup - // (most data is live) - FLAGS_heap_check_ignore_thread_live = true; // ignore all live - FLAGS_heap_check_ignore_global_live = true; // ignore all live - } else if (FLAGS_heap_check == "strict") { - // A bit stricter than 'normal': global destructors must fully clean up - // after themselves if they are present. - FLAGS_heap_check_before_constructors = true; // from no profile (fast) - FLAGS_heap_check_after_destructors = true; // to after destructors - // (less data live) - FLAGS_heap_check_ignore_thread_live = true; // ignore all live - FLAGS_heap_check_ignore_global_live = true; // ignore all live - } else if (FLAGS_heap_check == "draconian") { - // Drop not very portable and not very exact live heap flooding. - FLAGS_heap_check_before_constructors = true; // from no profile (fast) - FLAGS_heap_check_after_destructors = true; // to after destructors - // (need them) - FLAGS_heap_check_ignore_thread_live = false; // no live flood (stricter) - FLAGS_heap_check_ignore_global_live = false; // no live flood (stricter) - } else if (FLAGS_heap_check == "as-is") { - // do nothing: use other flags as is - } else if (FLAGS_heap_check == "local") { - // do nothing - } else { - RAW_LOG(FATAL, "Unsupported heap_check flag: %s", - FLAGS_heap_check.c_str()); - } - // FreeBSD doesn't seem to honor atexit execution order: - // http://code.google.com/p/gperftools/issues/detail?id=375 - // Since heap-checking before destructors depends on atexit running - // at the right time, on FreeBSD we always check after, even in the - // less strict modes. This just means FreeBSD is always a bit - // stricter in its checking than other OSes. - // This now appears to be the case in other OSes as well; - // so always check afterwards. - FLAGS_heap_check_after_destructors = true; - - { SpinLockHolder l(&heap_checker_lock); - RAW_DCHECK(heap_checker_pid == getpid(), ""); - heap_checker_on = true; - RAW_DCHECK(heap_profile, ""); - HeapLeakChecker::ProcMapsResult pm_result = HeapLeakChecker::UseProcMapsLocked(HeapLeakChecker::DISABLE_LIBRARY_ALLOCS); - // might neeed to do this more than once - // if one later dynamically loads libraries that we want disabled - if (pm_result != HeapLeakChecker::PROC_MAPS_USED) { // can't function - HeapLeakChecker::TurnItselfOffLocked(); - return; - } - } - - // make a good place and name for heap profile leak dumps - string* profile_prefix = - new string(FLAGS_heap_check_dump_directory + "/" + invocation_name()); - - // Finalize prefix for dumping leak checking profiles. - const int32 our_pid = getpid(); // safest to call getpid() outside lock - { SpinLockHolder l(&heap_checker_lock); - // main_thread_pid might still be 0 if this function is being called before - // global constructors. In that case, our pid *is* the main pid. - if (main_thread_pid == 0) - main_thread_pid = our_pid; - } - char pid_buf[15]; - snprintf(pid_buf, sizeof(pid_buf), ".%d", main_thread_pid); - *profile_prefix += pid_buf; - { SpinLockHolder l(&heap_checker_lock); - RAW_DCHECK(profile_name_prefix == NULL, ""); - profile_name_prefix = profile_prefix; - } - - // Make sure new/delete hooks are installed properly - // and heap profiler is indeed able to keep track - // of the objects being allocated. - // We test this to make sure we are indeed checking for leaks. - char* test_str = new char[5]; - size_t size; - { SpinLockHolder l(&heap_checker_lock); - RAW_CHECK(heap_profile->FindAlloc(test_str, &size), - "our own new/delete not linked?"); - } - delete[] test_str; - { SpinLockHolder l(&heap_checker_lock); - // This check can fail when it should not if another thread allocates - // into this same spot right this moment, - // which is unlikely since this code runs in InitGoogle. - RAW_CHECK(!heap_profile->FindAlloc(test_str, &size), - "our own new/delete not linked?"); - } - // If we crash in the above code, it probably means that - // "nm | grep new" will show that tcmalloc's new/delete - // implementation did not get linked-in into this binary - // (i.e. nm will list __builtin_new and __builtin_vec_new as undefined). - // If this happens, it is a BUILD bug to be fixed. - - RAW_VLOG(heap_checker_info_level, - "WARNING: Perftools heap leak checker is active " - "-- Performance may suffer"); - - if (FLAGS_heap_check != "local") { - HeapLeakChecker* main_hc = new HeapLeakChecker(); - SpinLockHolder l(&heap_checker_lock); - RAW_DCHECK(main_heap_checker == NULL, - "Repeated creation of main_heap_checker"); - main_heap_checker = main_hc; - do_main_heap_check = true; - } - - { SpinLockHolder l(&heap_checker_lock); - RAW_CHECK(heap_checker_on && constructor_heap_profiling, - "Leak checking is expected to be fully turned on now"); - } - - // For binaries built in debug mode, this will set release queue of - // debugallocation.cc to 100M to make it less likely for real leaks to - // be hidden due to reuse of heap memory object addresses. - // Running a test with --malloc_reclaim_memory=0 would help find leaks even - // better, but the test might run out of memory as a result. - // The scenario is that a heap object at address X is allocated and freed, - // but some other data-structure still retains a pointer to X. - // Then the same heap memory is used for another object, which is leaked, - // but the leak is not noticed due to the pointer to the original object at X. - // TODO(csilvers): support this in some manner. -#if 0 - SetCommandLineOptionWithMode("max_free_queue_size", "104857600", // 100M - SET_FLAG_IF_DEFAULT); -#endif -} - -// We want this to run early as well, but not so early as -// ::BeforeConstructors (we want flag assignments to have already -// happened, for instance). Initializer-registration does the trick. -REGISTER_MODULE_INITIALIZER(init_start, HeapLeakChecker_InternalInitStart()); -REGISTER_MODULE_DESTRUCTOR(init_start, HeapLeakChecker_RunHeapCleanups()); - -// static -bool HeapLeakChecker::NoGlobalLeaksMaybeSymbolize( - ShouldSymbolize should_symbolize) { - // we never delete or change main_heap_checker once it's set: - HeapLeakChecker* main_hc = GlobalChecker(); - if (main_hc) { - RAW_VLOG(10, "Checking for whole-program memory leaks"); - return main_hc->DoNoLeaks(should_symbolize); - } - return true; -} - -// static -bool HeapLeakChecker::DoMainHeapCheck() { - if (FLAGS_heap_check_delay_seconds > 0) { - sleep(FLAGS_heap_check_delay_seconds); - } - { SpinLockHolder l(&heap_checker_lock); - if (!do_main_heap_check) return false; - RAW_DCHECK(heap_checker_pid == getpid(), ""); - do_main_heap_check = false; // will do it now; no need to do it more - } - - // The program is over, so it's safe to symbolize addresses (which - // requires a fork) because no serious work is expected to be done - // after this. Symbolizing is really useful -- knowing what - // function has a leak is better than knowing just an address -- - // and while we can only safely symbolize once in a program run, - // now is the time (after all, there's no "later" that would be better). - if (!NoGlobalLeaksMaybeSymbolize(SYMBOLIZE)) { - if (FLAGS_heap_check_identify_leaks) { - RAW_LOG(FATAL, "Whole-program memory leaks found."); - } - RAW_LOG(ERROR, "Exiting with error code (instead of crashing) " - "because of whole-program memory leaks"); - _exit(1); // we don't want to call atexit() routines! - } - return true; -} - -// static -HeapLeakChecker* HeapLeakChecker::GlobalChecker() { - SpinLockHolder l(&heap_checker_lock); - return main_heap_checker; -} - -// static -bool HeapLeakChecker::NoGlobalLeaks() { - // symbolizing requires a fork, which isn't safe to do in general. - return NoGlobalLeaksMaybeSymbolize(DO_NOT_SYMBOLIZE); -} - -// static -void HeapLeakChecker::CancelGlobalCheck() { - SpinLockHolder l(&heap_checker_lock); - if (do_main_heap_check) { - RAW_VLOG(heap_checker_info_level, - "Canceling the automatic at-exit whole-program memory leak check"); - do_main_heap_check = false; - } -} - -// static -void HeapLeakChecker::BeforeConstructorsLocked() { - RAW_DCHECK(heap_checker_lock.IsHeld(), ""); - RAW_CHECK(!constructor_heap_profiling, - "BeforeConstructorsLocked called multiple times"); -#ifdef ADDRESS_SANITIZER - // AddressSanitizer's custom malloc conflicts with HeapChecker. - return; -#endif - // Set hooks early to crash if 'new' gets called before we make heap_profile, - // and make sure no other hooks existed: - RAW_CHECK(MallocHook::AddNewHook(&NewHook), ""); - RAW_CHECK(MallocHook::AddDeleteHook(&DeleteHook), ""); - constructor_heap_profiling = true; - MemoryRegionMap::Init(1, /* use_buckets */ false); - // Set up MemoryRegionMap with (at least) one caller stack frame to record - // (important that it's done before HeapProfileTable creation below). - Allocator::Init(); - RAW_CHECK(heap_profile == NULL, ""); - heap_profile = new(Allocator::Allocate(sizeof(HeapProfileTable))) - HeapProfileTable(&Allocator::Allocate, &Allocator::Free, - /* profile_mmap */ false); - RAW_VLOG(10, "Starting tracking the heap"); - heap_checker_on = true; -} - -// static -void HeapLeakChecker::TurnItselfOffLocked() { - RAW_DCHECK(heap_checker_lock.IsHeld(), ""); - // Set FLAGS_heap_check to "", for users who test for it - if (!FLAGS_heap_check.empty()) // be a noop in the common case - FLAGS_heap_check.clear(); // because clear() could allocate memory - if (constructor_heap_profiling) { - RAW_CHECK(heap_checker_on, ""); - RAW_VLOG(heap_checker_info_level, "Turning perftools heap leak checking off"); - heap_checker_on = false; - // Unset our hooks checking they were set: - RAW_CHECK(MallocHook::RemoveNewHook(&NewHook), ""); - RAW_CHECK(MallocHook::RemoveDeleteHook(&DeleteHook), ""); - Allocator::DeleteAndNull(&heap_profile); - // free our optional global data: - Allocator::DeleteAndNullIfNot(&ignored_objects); - Allocator::DeleteAndNullIfNot(&disabled_ranges); - Allocator::DeleteAndNullIfNot(&global_region_caller_ranges); - Allocator::Shutdown(); - MemoryRegionMap::Shutdown(); - } - RAW_CHECK(!heap_checker_on, ""); -} - -extern bool heap_leak_checker_bcad_variable; // in heap-checker-bcad.cc - -static bool has_called_before_constructors = false; - -// TODO(maxim): inline this function with -// MallocHook_InitAtFirstAllocation_HeapLeakChecker, and also rename -// HeapLeakChecker::BeforeConstructorsLocked. -void HeapLeakChecker_BeforeConstructors() { - SpinLockHolder l(&heap_checker_lock); - // We can be called from several places: the first mmap/sbrk/alloc call - // or the first global c-tor from heap-checker-bcad.cc: - // Do not re-execute initialization: - if (has_called_before_constructors) return; - has_called_before_constructors = true; - - heap_checker_pid = getpid(); // set it always - heap_leak_checker_bcad_variable = true; - // just to reference it, so that heap-checker-bcad.o is linked in - - // This function can be called *very* early, before the normal - // global-constructor that sets FLAGS_verbose. Set it manually now, - // so the RAW_LOG messages here are controllable. - const char* verbose_str = GetenvBeforeMain("PERFTOOLS_VERBOSE"); - if (verbose_str && atoi(verbose_str)) { // different than the default of 0? - FLAGS_verbose = atoi(verbose_str); - } - - bool need_heap_check = true; - // The user indicates a desire for heap-checking via the HEAPCHECK - // environment variable. If it's not set, there's no way to do - // heap-checking. - if (!GetenvBeforeMain("HEAPCHECK")) { - need_heap_check = false; - } -#ifdef HAVE_GETEUID - if (need_heap_check && getuid() != geteuid()) { - // heap-checker writes out files. Thus, for security reasons, we don't - // recognize the env. var. to turn on heap-checking if we're setuid. - RAW_LOG(WARNING, ("HeapChecker: ignoring HEAPCHECK because " - "program seems to be setuid\n")); - need_heap_check = false; - } -#endif - if (need_heap_check) { - HeapLeakChecker::BeforeConstructorsLocked(); - } -} - -// This function overrides the weak function defined in malloc_hook.cc and -// called by one of the initial malloc hooks (malloc_hook.cc) when the very -// first memory allocation or an mmap/sbrk happens. This ensures that -// HeapLeakChecker is initialized and installs all its hooks early enough to -// track absolutely all memory allocations and all memory region acquisitions -// via mmap and sbrk. -extern "C" void MallocHook_InitAtFirstAllocation_HeapLeakChecker() { - HeapLeakChecker_BeforeConstructors(); -} - -// This function is executed after all global object destructors run. -void HeapLeakChecker_AfterDestructors() { - { SpinLockHolder l(&heap_checker_lock); - // can get here (via forks?) with other pids - if (heap_checker_pid != getpid()) return; - } - if (FLAGS_heap_check_after_destructors) { - if (HeapLeakChecker::DoMainHeapCheck()) { - const struct timespec sleep_time = { 0, 500000000 }; // 500 ms - nanosleep(&sleep_time, NULL); - // Need this hack to wait for other pthreads to exit. - // Otherwise tcmalloc find errors - // on a free() call from pthreads. - } - } - SpinLockHolder l(&heap_checker_lock); - RAW_CHECK(!do_main_heap_check, "should have done it"); -} - -//---------------------------------------------------------------------- -// HeapLeakChecker disabling helpers -//---------------------------------------------------------------------- - -// These functions are at the end of the file to prevent their inlining: - -// static -void HeapLeakChecker::DisableChecksFromToLocked(const void* start_address, - const void* end_address, - int max_depth) { - RAW_DCHECK(heap_checker_lock.IsHeld(), ""); - RAW_DCHECK(start_address < end_address, ""); - if (disabled_ranges == NULL) { - disabled_ranges = new(Allocator::Allocate(sizeof(DisabledRangeMap))) - DisabledRangeMap; - } - RangeValue value; - value.start_address = AsInt(start_address); - value.max_depth = max_depth; - if (disabled_ranges->insert(make_pair(AsInt(end_address), value)).second) { - RAW_VLOG(10, "Disabling leak checking in stack traces " - "under frame addresses between %p..%p", - start_address, end_address); - } else { // check that this is just a verbatim repetition - RangeValue const& val = disabled_ranges->find(AsInt(end_address))->second; - if (val.max_depth != value.max_depth || - val.start_address != value.start_address) { - RAW_LOG(FATAL, "Two DisableChecksToHereFrom calls conflict: " - "(%p, %p, %d) vs. (%p, %p, %d)", - AsPtr(val.start_address), end_address, val.max_depth, - start_address, end_address, max_depth); - } - } -} - -// static -inline bool HeapLeakChecker::HaveOnHeapLocked(const void** ptr, - size_t* object_size) { - // Commented-out because HaveOnHeapLocked is very performance-critical: - // RAW_DCHECK(heap_checker_lock.IsHeld(), ""); - const uintptr_t addr = AsInt(*ptr); - if (heap_profile->FindInsideAlloc( - *ptr, max_heap_object_size, ptr, object_size)) { - RAW_VLOG(16, "Got pointer into %p at +%" PRIuPTR " offset", - *ptr, addr - AsInt(*ptr)); - return true; - } - return false; -} - -// static -const void* HeapLeakChecker::GetAllocCaller(void* ptr) { - // this is used only in the unittest, so the heavy checks are fine - HeapProfileTable::AllocInfo info; - { SpinLockHolder l(&heap_checker_lock); - RAW_CHECK(heap_profile->FindAllocDetails(ptr, &info), ""); - } - RAW_CHECK(info.stack_depth >= 1, ""); - return info.call_stack[0]; -} diff --git a/contrib/libtcmalloc/src/heap-profile-stats.h b/contrib/libtcmalloc/src/heap-profile-stats.h deleted file mode 100644 index ae45d5883fa..00000000000 --- a/contrib/libtcmalloc/src/heap-profile-stats.h +++ /dev/null @@ -1,78 +0,0 @@ -// -*- Mode: C++; c-basic-offset: 2; indent-tabs-mode: nil -*- -// Copyright (c) 2013, Google Inc. -// All rights reserved. -// -// Redistribution and use in source and binary forms, with or without -// modification, are permitted provided that the following conditions are -// met: -// -// * Redistributions of source code must retain the above copyright -// notice, this list of conditions and the following disclaimer. -// * Redistributions in binary form must reproduce the above -// copyright notice, this list of conditions and the following disclaimer -// in the documentation and/or other materials provided with the -// distribution. -// * Neither the name of Google Inc. nor the names of its -// contributors may be used to endorse or promote products derived from -// this software without specific prior written permission. -// -// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS -// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT -// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR -// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT -// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, -// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT -// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, -// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY -// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT -// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE -// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. - -// This file defines structs to accumulate memory allocation and deallocation -// counts. These structs are commonly used for malloc (in HeapProfileTable) -// and mmap (in MemoryRegionMap). - -// A bucket is data structure for heap profiling to store a pair of a stack -// trace and counts of (de)allocation. Buckets are stored in a hash table -// which is declared as "HeapProfileBucket**". -// -// A hash value is computed from a stack trace. Collision in the hash table -// is resolved by separate chaining with linked lists. The links in the list -// are implemented with the member "HeapProfileBucket* next". -// -// A structure of a hash table HeapProfileBucket** bucket_table would be like: -// bucket_table[0] => NULL -// bucket_table[1] => HeapProfileBucket() => HeapProfileBucket() => NULL -// ... -// bucket_table[i] => HeapProfileBucket() => NULL -// ... -// bucket_table[n] => HeapProfileBucket() => NULL - -#ifndef HEAP_PROFILE_STATS_H_ -#define HEAP_PROFILE_STATS_H_ - -struct HeapProfileStats { - // Returns true if the two HeapProfileStats are semantically equal. - bool Equivalent(const HeapProfileStats& other) const { - return allocs - frees == other.allocs - other.frees && - alloc_size - free_size == other.alloc_size - other.free_size; - } - - int32 allocs; // Number of allocation calls. - int32 frees; // Number of free calls. - int64 alloc_size; // Total size of all allocated objects so far. - int64 free_size; // Total size of all freed objects so far. -}; - -// Allocation and deallocation statistics per each stack trace. -struct HeapProfileBucket : public HeapProfileStats { - // Longest stack trace we record. - static const int kMaxStackDepth = 32; - - uintptr_t hash; // Hash value of the stack trace. - int depth; // Depth of stack trace. - const void** stack; // Stack trace. - HeapProfileBucket* next; // Next entry in hash-table. -}; - -#endif // HEAP_PROFILE_STATS_H_ diff --git a/contrib/libtcmalloc/src/heap-profile-table.cc b/contrib/libtcmalloc/src/heap-profile-table.cc deleted file mode 100644 index 7486468c056..00000000000 --- a/contrib/libtcmalloc/src/heap-profile-table.cc +++ /dev/null @@ -1,631 +0,0 @@ -// -*- Mode: C++; c-basic-offset: 2; indent-tabs-mode: nil -*- -// Copyright (c) 2006, Google Inc. -// All rights reserved. -// -// Redistribution and use in source and binary forms, with or without -// modification, are permitted provided that the following conditions are -// met: -// -// * Redistributions of source code must retain the above copyright -// notice, this list of conditions and the following disclaimer. -// * Redistributions in binary form must reproduce the above -// copyright notice, this list of conditions and the following disclaimer -// in the documentation and/or other materials provided with the -// distribution. -// * Neither the name of Google Inc. nor the names of its -// contributors may be used to endorse or promote products derived from -// this software without specific prior written permission. -// -// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS -// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT -// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR -// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT -// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, -// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT -// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, -// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY -// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT -// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE -// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. - -// --- -// Author: Sanjay Ghemawat -// Maxim Lifantsev (refactoring) -// - -#include - -#ifdef HAVE_UNISTD_H -#include // for write() -#endif -#include // for open() -#ifdef HAVE_GLOB_H -#include -#ifndef GLOB_NOMATCH // true on some old cygwins -# define GLOB_NOMATCH 0 -#endif -#endif -#ifdef HAVE_INTTYPES_H -#include // for PRIxPTR -#endif -#ifdef HAVE_POLL_H -#include -#endif -#include -#include -#include -#include -#include // for sort(), equal(), and copy() - -#include "heap-profile-table.h" - -#include "base/logging.h" -#include "raw_printer.h" -#include "symbolize.h" -#include -#include -#include "memory_region_map.h" -#include "base/commandlineflags.h" -#include "base/logging.h" // for the RawFD I/O commands -#include "base/sysinfo.h" - -using std::sort; -using std::equal; -using std::copy; -using std::string; -using std::map; - -using tcmalloc::FillProcSelfMaps; // from sysinfo.h -using tcmalloc::DumpProcSelfMaps; // from sysinfo.h - -//---------------------------------------------------------------------- - -DEFINE_bool(cleanup_old_heap_profiles, - EnvToBool("HEAP_PROFILE_CLEANUP", true), - "At initialization time, delete old heap profiles."); - -DEFINE_int32(heap_check_max_leaks, - EnvToInt("HEAP_CHECK_MAX_LEAKS", 20), - "The maximum number of leak reports to print."); - -//---------------------------------------------------------------------- - -// header of the dumped heap profile -static const char kProfileHeader[] = "heap profile: "; -static const char kProcSelfMapsHeader[] = "\nMAPPED_LIBRARIES:\n"; - -//---------------------------------------------------------------------- - -const char HeapProfileTable::kFileExt[] = ".heap"; - -//---------------------------------------------------------------------- - -static const int kHashTableSize = 179999; // Size for bucket_table_. -/*static*/ const int HeapProfileTable::kMaxStackDepth; - -//---------------------------------------------------------------------- - -// We strip out different number of stack frames in debug mode -// because less inlining happens in that case -#ifdef NDEBUG -static const int kStripFrames = 2; -#else -static const int kStripFrames = 3; -#endif - -// For sorting Stats or Buckets by in-use space -static bool ByAllocatedSpace(HeapProfileTable::Stats* a, - HeapProfileTable::Stats* b) { - // Return true iff "a" has more allocated space than "b" - return (a->alloc_size - a->free_size) > (b->alloc_size - b->free_size); -} - -//---------------------------------------------------------------------- - -HeapProfileTable::HeapProfileTable(Allocator alloc, - DeAllocator dealloc, - bool profile_mmap) - : alloc_(alloc), - dealloc_(dealloc), - profile_mmap_(profile_mmap), - bucket_table_(NULL), - num_buckets_(0), - address_map_(NULL) { - // Make a hash table for buckets. - const int table_bytes = kHashTableSize * sizeof(*bucket_table_); - bucket_table_ = static_cast(alloc_(table_bytes)); - memset(bucket_table_, 0, table_bytes); - - // Make an allocation map. - address_map_ = - new(alloc_(sizeof(AllocationMap))) AllocationMap(alloc_, dealloc_); - - // Initialize. - memset(&total_, 0, sizeof(total_)); - num_buckets_ = 0; -} - -HeapProfileTable::~HeapProfileTable() { - // Free the allocation map. - address_map_->~AllocationMap(); - dealloc_(address_map_); - address_map_ = NULL; - - // Free the hash table. - for (int i = 0; i < kHashTableSize; i++) { - for (Bucket* curr = bucket_table_[i]; curr != 0; /**/) { - Bucket* bucket = curr; - curr = curr->next; - dealloc_(bucket->stack); - dealloc_(bucket); - } - } - dealloc_(bucket_table_); - bucket_table_ = NULL; -} - -HeapProfileTable::Bucket* HeapProfileTable::GetBucket(int depth, - const void* const key[]) { - // Make hash-value - uintptr_t h = 0; - for (int i = 0; i < depth; i++) { - h += reinterpret_cast(key[i]); - h += h << 10; - h ^= h >> 6; - } - h += h << 3; - h ^= h >> 11; - - // Lookup stack trace in table - unsigned int buck = ((unsigned int) h) % kHashTableSize; - for (Bucket* b = bucket_table_[buck]; b != 0; b = b->next) { - if ((b->hash == h) && - (b->depth == depth) && - equal(key, key + depth, b->stack)) { - return b; - } - } - - // Create new bucket - const size_t key_size = sizeof(key[0]) * depth; - const void** kcopy = reinterpret_cast(alloc_(key_size)); - copy(key, key + depth, kcopy); - Bucket* b = reinterpret_cast(alloc_(sizeof(Bucket))); - memset(b, 0, sizeof(*b)); - b->hash = h; - b->depth = depth; - b->stack = kcopy; - b->next = bucket_table_[buck]; - bucket_table_[buck] = b; - num_buckets_++; - return b; -} - -int HeapProfileTable::GetCallerStackTrace( - int skip_count, void* stack[kMaxStackDepth]) { - return MallocHook::GetCallerStackTrace( - stack, kMaxStackDepth, kStripFrames + skip_count + 1); -} - -void HeapProfileTable::RecordAlloc( - const void* ptr, size_t bytes, int stack_depth, - const void* const call_stack[]) { - Bucket* b = GetBucket(stack_depth, call_stack); - b->allocs++; - b->alloc_size += bytes; - total_.allocs++; - total_.alloc_size += bytes; - - AllocValue v; - v.set_bucket(b); // also did set_live(false); set_ignore(false) - v.bytes = bytes; - address_map_->Insert(ptr, v); -} - -void HeapProfileTable::RecordFree(const void* ptr) { - AllocValue v; - if (address_map_->FindAndRemove(ptr, &v)) { - Bucket* b = v.bucket(); - b->frees++; - b->free_size += v.bytes; - total_.frees++; - total_.free_size += v.bytes; - } -} - -bool HeapProfileTable::FindAlloc(const void* ptr, size_t* object_size) const { - const AllocValue* alloc_value = address_map_->Find(ptr); - if (alloc_value != NULL) *object_size = alloc_value->bytes; - return alloc_value != NULL; -} - -bool HeapProfileTable::FindAllocDetails(const void* ptr, - AllocInfo* info) const { - const AllocValue* alloc_value = address_map_->Find(ptr); - if (alloc_value != NULL) { - info->object_size = alloc_value->bytes; - info->call_stack = alloc_value->bucket()->stack; - info->stack_depth = alloc_value->bucket()->depth; - } - return alloc_value != NULL; -} - -bool HeapProfileTable::FindInsideAlloc(const void* ptr, - size_t max_size, - const void** object_ptr, - size_t* object_size) const { - const AllocValue* alloc_value = - address_map_->FindInside(&AllocValueSize, max_size, ptr, object_ptr); - if (alloc_value != NULL) *object_size = alloc_value->bytes; - return alloc_value != NULL; -} - -bool HeapProfileTable::MarkAsLive(const void* ptr) { - AllocValue* alloc = address_map_->FindMutable(ptr); - if (alloc && !alloc->live()) { - alloc->set_live(true); - return true; - } - return false; -} - -void HeapProfileTable::MarkAsIgnored(const void* ptr) { - AllocValue* alloc = address_map_->FindMutable(ptr); - if (alloc) { - alloc->set_ignore(true); - } -} - -// We'd be happier using snprintfer, but we don't to reduce dependencies. -int HeapProfileTable::UnparseBucket(const Bucket& b, - char* buf, int buflen, int bufsize, - const char* extra, - Stats* profile_stats) { - if (profile_stats != NULL) { - profile_stats->allocs += b.allocs; - profile_stats->alloc_size += b.alloc_size; - profile_stats->frees += b.frees; - profile_stats->free_size += b.free_size; - } - int printed = - snprintf(buf + buflen, bufsize - buflen, "%6d: %8" PRId64 " [%6d: %8" PRId64 "] @%s", - b.allocs - b.frees, - b.alloc_size - b.free_size, - b.allocs, - b.alloc_size, - extra); - // If it looks like the snprintf failed, ignore the fact we printed anything - if (printed < 0 || printed >= bufsize - buflen) return buflen; - buflen += printed; - for (int d = 0; d < b.depth; d++) { - printed = snprintf(buf + buflen, bufsize - buflen, " 0x%08" PRIxPTR, - reinterpret_cast(b.stack[d])); - if (printed < 0 || printed >= bufsize - buflen) return buflen; - buflen += printed; - } - printed = snprintf(buf + buflen, bufsize - buflen, "\n"); - if (printed < 0 || printed >= bufsize - buflen) return buflen; - buflen += printed; - return buflen; -} - -HeapProfileTable::Bucket** -HeapProfileTable::MakeSortedBucketList() const { - Bucket** list = static_cast(alloc_(sizeof(Bucket) * num_buckets_)); - - int bucket_count = 0; - for (int i = 0; i < kHashTableSize; i++) { - for (Bucket* curr = bucket_table_[i]; curr != 0; curr = curr->next) { - list[bucket_count++] = curr; - } - } - RAW_DCHECK(bucket_count == num_buckets_, ""); - - sort(list, list + num_buckets_, ByAllocatedSpace); - - return list; -} - -void HeapProfileTable::IterateOrderedAllocContexts( - AllocContextIterator callback) const { - Bucket** list = MakeSortedBucketList(); - AllocContextInfo info; - for (int i = 0; i < num_buckets_; ++i) { - *static_cast(&info) = *static_cast(list[i]); - info.stack_depth = list[i]->depth; - info.call_stack = list[i]->stack; - callback(info); - } - dealloc_(list); -} - -int HeapProfileTable::FillOrderedProfile(char buf[], int size) const { - Bucket** list = MakeSortedBucketList(); - - // Our file format is "bucket, bucket, ..., bucket, proc_self_maps_info". - // In the cases buf is too small, we'd rather leave out the last - // buckets than leave out the /proc/self/maps info. To ensure that, - // we actually print the /proc/self/maps info first, then move it to - // the end of the buffer, then write the bucket info into whatever - // is remaining, and then move the maps info one last time to close - // any gaps. Whew! - int map_length = snprintf(buf, size, "%s", kProcSelfMapsHeader); - if (map_length < 0 || map_length >= size) { - dealloc_(list); - return 0; - } - bool dummy; // "wrote_all" -- did /proc/self/maps fit in its entirety? - map_length += FillProcSelfMaps(buf + map_length, size - map_length, &dummy); - RAW_DCHECK(map_length <= size, ""); - char* const map_start = buf + size - map_length; // move to end - memmove(map_start, buf, map_length); - size -= map_length; - - Stats stats; - memset(&stats, 0, sizeof(stats)); - int bucket_length = snprintf(buf, size, "%s", kProfileHeader); - if (bucket_length < 0 || bucket_length >= size) { - dealloc_(list); - return 0; - } - bucket_length = UnparseBucket(total_, buf, bucket_length, size, - " heapprofile", &stats); - - // Dump the mmap list first. - if (profile_mmap_) { - BufferArgs buffer(buf, bucket_length, size); - MemoryRegionMap::IterateBuckets(DumpBucketIterator, &buffer); - bucket_length = buffer.buflen; - } - - for (int i = 0; i < num_buckets_; i++) { - bucket_length = UnparseBucket(*list[i], buf, bucket_length, size, "", - &stats); - } - RAW_DCHECK(bucket_length < size, ""); - - dealloc_(list); - - RAW_DCHECK(buf + bucket_length <= map_start, ""); - memmove(buf + bucket_length, map_start, map_length); // close the gap - - return bucket_length + map_length; -} - -// static -void HeapProfileTable::DumpBucketIterator(const Bucket* bucket, - BufferArgs* args) { - args->buflen = UnparseBucket(*bucket, args->buf, args->buflen, args->bufsize, - "", NULL); -} - -inline -void HeapProfileTable::DumpNonLiveIterator(const void* ptr, AllocValue* v, - const DumpArgs& args) { - if (v->live()) { - v->set_live(false); - return; - } - if (v->ignore()) { - return; - } - Bucket b; - memset(&b, 0, sizeof(b)); - b.allocs = 1; - b.alloc_size = v->bytes; - b.depth = v->bucket()->depth; - b.stack = v->bucket()->stack; - char buf[1024]; - int len = UnparseBucket(b, buf, 0, sizeof(buf), "", args.profile_stats); - RawWrite(args.fd, buf, len); -} - -// Callback from NonLiveSnapshot; adds entry to arg->dest -// if not the entry is not live and is not present in arg->base. -void HeapProfileTable::AddIfNonLive(const void* ptr, AllocValue* v, - AddNonLiveArgs* arg) { - if (v->live()) { - v->set_live(false); - } else { - if (arg->base != NULL && arg->base->map_.Find(ptr) != NULL) { - // Present in arg->base, so do not save - } else { - arg->dest->Add(ptr, *v); - } - } -} - -bool HeapProfileTable::WriteProfile(const char* file_name, - const Bucket& total, - AllocationMap* allocations) { - RAW_VLOG(1, "Dumping non-live heap profile to %s", file_name); - RawFD fd = RawOpenForWriting(file_name); - if (fd != kIllegalRawFD) { - RawWrite(fd, kProfileHeader, strlen(kProfileHeader)); - char buf[512]; - int len = UnparseBucket(total, buf, 0, sizeof(buf), " heapprofile", - NULL); - RawWrite(fd, buf, len); - const DumpArgs args(fd, NULL); - allocations->Iterate(DumpNonLiveIterator, args); - RawWrite(fd, kProcSelfMapsHeader, strlen(kProcSelfMapsHeader)); - DumpProcSelfMaps(fd); - RawClose(fd); - return true; - } else { - RAW_LOG(ERROR, "Failed dumping filtered heap profile to %s", file_name); - return false; - } -} - -void HeapProfileTable::CleanupOldProfiles(const char* prefix) { - if (!FLAGS_cleanup_old_heap_profiles) - return; - string pattern = string(prefix) + ".*" + kFileExt; -#if defined(HAVE_GLOB_H) - glob_t g; - const int r = glob(pattern.c_str(), GLOB_ERR, NULL, &g); - if (r == 0 || r == GLOB_NOMATCH) { - const int prefix_length = strlen(prefix); - for (int i = 0; i < g.gl_pathc; i++) { - const char* fname = g.gl_pathv[i]; - if ((strlen(fname) >= prefix_length) && - (memcmp(fname, prefix, prefix_length) == 0)) { - RAW_VLOG(1, "Removing old heap profile %s", fname); - unlink(fname); - } - } - } - globfree(&g); -#else /* HAVE_GLOB_H */ - RAW_LOG(WARNING, "Unable to remove old heap profiles (can't run glob())"); -#endif -} - -HeapProfileTable::Snapshot* HeapProfileTable::TakeSnapshot() { - Snapshot* s = new (alloc_(sizeof(Snapshot))) Snapshot(alloc_, dealloc_); - address_map_->Iterate(AddToSnapshot, s); - return s; -} - -void HeapProfileTable::ReleaseSnapshot(Snapshot* s) { - s->~Snapshot(); - dealloc_(s); -} - -// Callback from TakeSnapshot; adds a single entry to snapshot -void HeapProfileTable::AddToSnapshot(const void* ptr, AllocValue* v, - Snapshot* snapshot) { - snapshot->Add(ptr, *v); -} - -HeapProfileTable::Snapshot* HeapProfileTable::NonLiveSnapshot( - Snapshot* base) { - RAW_VLOG(2, "NonLiveSnapshot input: %d %d\n", - int(total_.allocs - total_.frees), - int(total_.alloc_size - total_.free_size)); - - Snapshot* s = new (alloc_(sizeof(Snapshot))) Snapshot(alloc_, dealloc_); - AddNonLiveArgs args; - args.dest = s; - args.base = base; - address_map_->Iterate(AddIfNonLive, &args); - RAW_VLOG(2, "NonLiveSnapshot output: %d %d\n", - int(s->total_.allocs - s->total_.frees), - int(s->total_.alloc_size - s->total_.free_size)); - return s; -} - -// Information kept per unique bucket seen -struct HeapProfileTable::Snapshot::Entry { - int count; - int bytes; - Bucket* bucket; - Entry() : count(0), bytes(0) { } - - // Order by decreasing bytes - bool operator<(const Entry& x) const { - return this->bytes > x.bytes; - } -}; - -// State used to generate leak report. We keep a mapping from Bucket pointer -// the collected stats for that bucket. -struct HeapProfileTable::Snapshot::ReportState { - map buckets_; -}; - -// Callback from ReportLeaks; updates ReportState. -void HeapProfileTable::Snapshot::ReportCallback(const void* ptr, - AllocValue* v, - ReportState* state) { - Entry* e = &state->buckets_[v->bucket()]; // Creates empty Entry first time - e->bucket = v->bucket(); - e->count++; - e->bytes += v->bytes; -} - -void HeapProfileTable::Snapshot::ReportLeaks(const char* checker_name, - const char* filename, - bool should_symbolize) { - // This is only used by the heap leak checker, but is intimately - // tied to the allocation map that belongs in this module and is - // therefore placed here. - RAW_LOG(ERROR, "Leak check %s detected leaks of %" PRIuS " bytes " - "in %" PRIuS " objects", - checker_name, - size_t(total_.alloc_size), - size_t(total_.allocs)); - - // Group objects by Bucket - ReportState state; - map_.Iterate(&ReportCallback, &state); - - // Sort buckets by decreasing leaked size - const int n = state.buckets_.size(); - Entry* entries = new Entry[n]; - int dst = 0; - for (map::const_iterator iter = state.buckets_.begin(); - iter != state.buckets_.end(); - ++iter) { - entries[dst++] = iter->second; - } - sort(entries, entries + n); - - // Report a bounded number of leaks to keep the leak report from - // growing too long. - const int to_report = - (FLAGS_heap_check_max_leaks > 0 && - n > FLAGS_heap_check_max_leaks) ? FLAGS_heap_check_max_leaks : n; - RAW_LOG(ERROR, "The %d largest leaks:", to_report); - - // Print - SymbolTable symbolization_table; - for (int i = 0; i < to_report; i++) { - const Entry& e = entries[i]; - for (int j = 0; j < e.bucket->depth; j++) { - symbolization_table.Add(e.bucket->stack[j]); - } - } - static const int kBufSize = 2<<10; - char buffer[kBufSize]; - if (should_symbolize) - symbolization_table.Symbolize(); - for (int i = 0; i < to_report; i++) { - const Entry& e = entries[i]; - base::RawPrinter printer(buffer, kBufSize); - printer.Printf("Leak of %d bytes in %d objects allocated from:\n", - e.bytes, e.count); - for (int j = 0; j < e.bucket->depth; j++) { - const void* pc = e.bucket->stack[j]; - printer.Printf("\t@ %" PRIxPTR " %s\n", - reinterpret_cast(pc), symbolization_table.GetSymbol(pc)); - } - RAW_LOG(ERROR, "%s", buffer); - } - - if (to_report < n) { - RAW_LOG(ERROR, "Skipping leaks numbered %d..%d", - to_report, n-1); - } - delete[] entries; - - // TODO: Dump the sorted Entry list instead of dumping raw data? - // (should be much shorter) - if (!HeapProfileTable::WriteProfile(filename, total_, &map_)) { - RAW_LOG(ERROR, "Could not write pprof profile to %s", filename); - } -} - -void HeapProfileTable::Snapshot::ReportObject(const void* ptr, - AllocValue* v, - char* unused) { - // Perhaps also log the allocation stack trace (unsymbolized) - // on this line in case somebody finds it useful. - RAW_LOG(ERROR, "leaked %" PRIuS " byte object %p", v->bytes, ptr); -} - -void HeapProfileTable::Snapshot::ReportIndividualObjects() { - char unused; - map_.Iterate(ReportObject, &unused); -} diff --git a/contrib/libtcmalloc/src/heap-profile-table.h b/contrib/libtcmalloc/src/heap-profile-table.h deleted file mode 100644 index 3c6284741af..00000000000 --- a/contrib/libtcmalloc/src/heap-profile-table.h +++ /dev/null @@ -1,399 +0,0 @@ -// -*- Mode: C++; c-basic-offset: 2; indent-tabs-mode: nil -*- -// Copyright (c) 2006, Google Inc. -// All rights reserved. -// -// Redistribution and use in source and binary forms, with or without -// modification, are permitted provided that the following conditions are -// met: -// -// * Redistributions of source code must retain the above copyright -// notice, this list of conditions and the following disclaimer. -// * Redistributions in binary form must reproduce the above -// copyright notice, this list of conditions and the following disclaimer -// in the documentation and/or other materials provided with the -// distribution. -// * Neither the name of Google Inc. nor the names of its -// contributors may be used to endorse or promote products derived from -// this software without specific prior written permission. -// -// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS -// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT -// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR -// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT -// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, -// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT -// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, -// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY -// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT -// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE -// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. - -// --- -// Author: Sanjay Ghemawat -// Maxim Lifantsev (refactoring) -// - -#ifndef BASE_HEAP_PROFILE_TABLE_H_ -#define BASE_HEAP_PROFILE_TABLE_H_ - -#include "addressmap-inl.h" -#include "base/basictypes.h" -#include "base/logging.h" // for RawFD -#include "heap-profile-stats.h" - -// Table to maintain a heap profile data inside, -// i.e. the set of currently active heap memory allocations. -// thread-unsafe and non-reentrant code: -// each instance object must be used by one thread -// at a time w/o self-recursion. -// -// TODO(maxim): add a unittest for this class. -class HeapProfileTable { - public: - - // Extension to be used for heap pforile files. - static const char kFileExt[]; - - // Longest stack trace we record. - static const int kMaxStackDepth = 32; - - // data types ---------------------------- - - // Profile stats. - typedef HeapProfileStats Stats; - - // Info we can return about an allocation. - struct AllocInfo { - size_t object_size; // size of the allocation - const void* const* call_stack; // call stack that made the allocation call - int stack_depth; // depth of call_stack - bool live; - bool ignored; - }; - - // Info we return about an allocation context. - // An allocation context is a unique caller stack trace - // of an allocation operation. - struct AllocContextInfo : public Stats { - int stack_depth; // Depth of stack trace - const void* const* call_stack; // Stack trace - }; - - // Memory (de)allocator interface we'll use. - typedef void* (*Allocator)(size_t size); - typedef void (*DeAllocator)(void* ptr); - - // interface --------------------------- - - HeapProfileTable(Allocator alloc, DeAllocator dealloc, bool profile_mmap); - ~HeapProfileTable(); - - // Collect the stack trace for the function that asked to do the - // allocation for passing to RecordAlloc() below. - // - // The stack trace is stored in 'stack'. The stack depth is returned. - // - // 'skip_count' gives the number of stack frames between this call - // and the memory allocation function. - static int GetCallerStackTrace(int skip_count, void* stack[kMaxStackDepth]); - - // Record an allocation at 'ptr' of 'bytes' bytes. 'stack_depth' - // and 'call_stack' identifying the function that requested the - // allocation. They can be generated using GetCallerStackTrace() above. - void RecordAlloc(const void* ptr, size_t bytes, - int stack_depth, const void* const call_stack[]); - - // Record the deallocation of memory at 'ptr'. - void RecordFree(const void* ptr); - - // Return true iff we have recorded an allocation at 'ptr'. - // If yes, fill *object_size with the allocation byte size. - bool FindAlloc(const void* ptr, size_t* object_size) const; - // Same as FindAlloc, but fills all of *info. - bool FindAllocDetails(const void* ptr, AllocInfo* info) const; - - // Return true iff "ptr" points into a recorded allocation - // If yes, fill *object_ptr with the actual allocation address - // and *object_size with the allocation byte size. - // max_size specifies largest currently possible allocation size. - bool FindInsideAlloc(const void* ptr, size_t max_size, - const void** object_ptr, size_t* object_size) const; - - // If "ptr" points to a recorded allocation and it's not marked as live - // mark it as live and return true. Else return false. - // All allocations start as non-live. - bool MarkAsLive(const void* ptr); - - // If "ptr" points to a recorded allocation, mark it as "ignored". - // Ignored objects are treated like other objects, except that they - // are skipped in heap checking reports. - void MarkAsIgnored(const void* ptr); - - // Return current total (de)allocation statistics. It doesn't contain - // mmap'ed regions. - const Stats& total() const { return total_; } - - // Allocation data iteration callback: gets passed object pointer and - // fully-filled AllocInfo. - typedef void (*AllocIterator)(const void* ptr, const AllocInfo& info); - - // Iterate over the allocation profile data calling "callback" - // for every allocation. - void IterateAllocs(AllocIterator callback) const { - address_map_->Iterate(MapArgsAllocIterator, callback); - } - - // Allocation context profile data iteration callback - typedef void (*AllocContextIterator)(const AllocContextInfo& info); - - // Iterate over the allocation context profile data calling "callback" - // for every allocation context. Allocation contexts are ordered by the - // size of allocated space. - void IterateOrderedAllocContexts(AllocContextIterator callback) const; - - // Fill profile data into buffer 'buf' of size 'size' - // and return the actual size occupied by the dump in 'buf'. - // The profile buckets are dumped in the decreasing order - // of currently allocated bytes. - // We do not provision for 0-terminating 'buf'. - int FillOrderedProfile(char buf[], int size) const; - - // Cleanup any old profile files matching prefix + ".*" + kFileExt. - static void CleanupOldProfiles(const char* prefix); - - // Return a snapshot of the current contents of *this. - // Caller must call ReleaseSnapshot() on result when no longer needed. - // The result is only valid while this exists and until - // the snapshot is discarded by calling ReleaseSnapshot(). - class Snapshot; - Snapshot* TakeSnapshot(); - - // Release a previously taken snapshot. snapshot must not - // be used after this call. - void ReleaseSnapshot(Snapshot* snapshot); - - // Return a snapshot of every non-live, non-ignored object in *this. - // If "base" is non-NULL, skip any objects present in "base". - // As a side-effect, clears the "live" bit on every live object in *this. - // Caller must call ReleaseSnapshot() on result when no longer needed. - Snapshot* NonLiveSnapshot(Snapshot* base); - - private: - - // data types ---------------------------- - - // Hash table bucket to hold (de)allocation stats - // for a given allocation call stack trace. - typedef HeapProfileBucket Bucket; - - // Info stored in the address map - struct AllocValue { - // Access to the stack-trace bucket - Bucket* bucket() const { - return reinterpret_cast(bucket_rep & ~uintptr_t(kMask)); - } - // This also does set_live(false). - void set_bucket(Bucket* b) { bucket_rep = reinterpret_cast(b); } - size_t bytes; // Number of bytes in this allocation - - // Access to the allocation liveness flag (for leak checking) - bool live() const { return bucket_rep & kLive; } - void set_live(bool l) { - bucket_rep = (bucket_rep & ~uintptr_t(kLive)) | (l ? kLive : 0); - } - - // Should this allocation be ignored if it looks like a leak? - bool ignore() const { return bucket_rep & kIgnore; } - void set_ignore(bool r) { - bucket_rep = (bucket_rep & ~uintptr_t(kIgnore)) | (r ? kIgnore : 0); - } - - private: - // We store a few bits in the bottom bits of bucket_rep. - // (Alignment is at least four, so we have at least two bits.) - static const int kLive = 1; - static const int kIgnore = 2; - static const int kMask = kLive | kIgnore; - - uintptr_t bucket_rep; - }; - - // helper for FindInsideAlloc - static size_t AllocValueSize(const AllocValue& v) { return v.bytes; } - - typedef AddressMap AllocationMap; - - // Arguments that need to be passed DumpBucketIterator callback below. - struct BufferArgs { - BufferArgs(char* buf_arg, int buflen_arg, int bufsize_arg) - : buf(buf_arg), - buflen(buflen_arg), - bufsize(bufsize_arg) { - } - - char* buf; - int buflen; - int bufsize; - - DISALLOW_COPY_AND_ASSIGN(BufferArgs); - }; - - // Arguments that need to be passed DumpNonLiveIterator callback below. - struct DumpArgs { - DumpArgs(RawFD fd_arg, Stats* profile_stats_arg) - : fd(fd_arg), - profile_stats(profile_stats_arg) { - } - - RawFD fd; // file to write to - Stats* profile_stats; // stats to update (may be NULL) - }; - - // helpers ---------------------------- - - // Unparse bucket b and print its portion of profile dump into buf. - // We return the amount of space in buf that we use. We start printing - // at buf + buflen, and promise not to go beyond buf + bufsize. - // We do not provision for 0-terminating 'buf'. - // - // If profile_stats is non-NULL, we update *profile_stats by - // counting bucket b. - // - // "extra" is appended to the unparsed bucket. Typically it is empty, - // but may be set to something like " heapprofile" for the total - // bucket to indicate the type of the profile. - static int UnparseBucket(const Bucket& b, - char* buf, int buflen, int bufsize, - const char* extra, - Stats* profile_stats); - - // Get the bucket for the caller stack trace 'key' of depth 'depth' - // creating the bucket if needed. - Bucket* GetBucket(int depth, const void* const key[]); - - // Helper for IterateAllocs to do callback signature conversion - // from AllocationMap::Iterate to AllocIterator. - static void MapArgsAllocIterator(const void* ptr, AllocValue* v, - AllocIterator callback) { - AllocInfo info; - info.object_size = v->bytes; - info.call_stack = v->bucket()->stack; - info.stack_depth = v->bucket()->depth; - info.live = v->live(); - info.ignored = v->ignore(); - callback(ptr, info); - } - - // Helper to dump a bucket. - inline static void DumpBucketIterator(const Bucket* bucket, - BufferArgs* args); - - // Helper for DumpNonLiveProfile to do object-granularity - // heap profile dumping. It gets passed to AllocationMap::Iterate. - inline static void DumpNonLiveIterator(const void* ptr, AllocValue* v, - const DumpArgs& args); - - // Helper for IterateOrderedAllocContexts and FillOrderedProfile. - // Creates a sorted list of Buckets whose length is num_buckets_. - // The caller is responsible for deallocating the returned list. - Bucket** MakeSortedBucketList() const; - - // Helper for TakeSnapshot. Saves object to snapshot. - static void AddToSnapshot(const void* ptr, AllocValue* v, Snapshot* s); - - // Arguments passed to AddIfNonLive - struct AddNonLiveArgs { - Snapshot* dest; - Snapshot* base; - }; - - // Helper for NonLiveSnapshot. Adds the object to the destination - // snapshot if it is non-live. - static void AddIfNonLive(const void* ptr, AllocValue* v, - AddNonLiveArgs* arg); - - // Write contents of "*allocations" as a heap profile to - // "file_name". "total" must contain the total of all entries in - // "*allocations". - static bool WriteProfile(const char* file_name, - const Bucket& total, - AllocationMap* allocations); - - // data ---------------------------- - - // Memory (de)allocator that we use. - Allocator alloc_; - DeAllocator dealloc_; - - // Overall profile stats; we use only the Stats part, - // but make it a Bucket to pass to UnparseBucket. - Bucket total_; - - bool profile_mmap_; - - // Bucket hash table for malloc. - // We hand-craft one instead of using one of the pre-written - // ones because we do not want to use malloc when operating on the table. - // It is only few lines of code, so no big deal. - Bucket** bucket_table_; - int num_buckets_; - - // Map of all currently allocated objects and mapped regions we know about. - AllocationMap* address_map_; - - DISALLOW_COPY_AND_ASSIGN(HeapProfileTable); -}; - -class HeapProfileTable::Snapshot { - public: - const Stats& total() const { return total_; } - - // Report anything in this snapshot as a leak. - // May use new/delete for temporary storage. - // If should_symbolize is true, will fork (which is not threadsafe) - // to turn addresses into symbol names. Set to false for maximum safety. - // Also writes a heap profile to "filename" that contains - // all of the objects in this snapshot. - void ReportLeaks(const char* checker_name, const char* filename, - bool should_symbolize); - - // Report the addresses of all leaked objects. - // May use new/delete for temporary storage. - void ReportIndividualObjects(); - - bool Empty() const { - return (total_.allocs == 0) && (total_.alloc_size == 0); - } - - private: - friend class HeapProfileTable; - - // Total count/size are stored in a Bucket so we can reuse UnparseBucket - Bucket total_; - - // We share the Buckets managed by the parent table, but have our - // own object->bucket map. - AllocationMap map_; - - Snapshot(Allocator alloc, DeAllocator dealloc) : map_(alloc, dealloc) { - memset(&total_, 0, sizeof(total_)); - } - - // Callback used to populate a Snapshot object with entries found - // in another allocation map. - inline void Add(const void* ptr, const AllocValue& v) { - map_.Insert(ptr, v); - total_.allocs++; - total_.alloc_size += v.bytes; - } - - // Helpers for sorting and generating leak reports - struct Entry; - struct ReportState; - static void ReportCallback(const void* ptr, AllocValue* v, ReportState*); - static void ReportObject(const void* ptr, AllocValue* v, char*); - - DISALLOW_COPY_AND_ASSIGN(Snapshot); -}; - -#endif // BASE_HEAP_PROFILE_TABLE_H_ diff --git a/contrib/libtcmalloc/src/internal_logging.cc b/contrib/libtcmalloc/src/internal_logging.cc deleted file mode 100644 index 3b2d0cb80f9..00000000000 --- a/contrib/libtcmalloc/src/internal_logging.cc +++ /dev/null @@ -1,192 +0,0 @@ -// -*- Mode: C++; c-basic-offset: 2; indent-tabs-mode: nil -*- -// Copyright (c) 2005, Google Inc. -// All rights reserved. -// -// Redistribution and use in source and binary forms, with or without -// modification, are permitted provided that the following conditions are -// met: -// -// * Redistributions of source code must retain the above copyright -// notice, this list of conditions and the following disclaimer. -// * Redistributions in binary form must reproduce the above -// copyright notice, this list of conditions and the following disclaimer -// in the documentation and/or other materials provided with the -// distribution. -// * Neither the name of Google Inc. nor the names of its -// contributors may be used to endorse or promote products derived from -// this software without specific prior written permission. -// -// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS -// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT -// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR -// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT -// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, -// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT -// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, -// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY -// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT -// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE -// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. - -// --- -// Sanjay Ghemawat - -#include "config.h" -#include "internal_logging.h" -#include // for va_end, va_start -#include // for vsnprintf, va_list, etc -#include // for abort -#include // for strlen, memcpy -#ifdef HAVE_UNISTD_H -#include // for write() -#endif - -#include -#include "base/logging.h" // for perftools_vsnprintf -#include "base/spinlock.h" // for SpinLockHolder, SpinLock - -// Variables for storing crash output. Allocated statically since we -// may not be able to heap-allocate while crashing. -static SpinLock crash_lock(base::LINKER_INITIALIZED); -static bool crashed = false; -static const int kStatsBufferSize = 16 << 10; -static char stats_buffer[kStatsBufferSize] = { 0 }; - -namespace tcmalloc { - -static void WriteMessage(const char* msg, int length) { - write(STDERR_FILENO, msg, length); -} - -void (*log_message_writer)(const char* msg, int length) = WriteMessage; - - -class Logger { - public: - bool Add(const LogItem& item); - bool AddStr(const char* str, int n); - bool AddNum(uint64_t num, int base); // base must be 10 or 16. - - static const int kBufSize = 200; - char* p_; - char* end_; - char buf_[kBufSize]; -}; - -void Log(LogMode mode, const char* filename, int line, - LogItem a, LogItem b, LogItem c, LogItem d) { - Logger state; - state.p_ = state.buf_; - state.end_ = state.buf_ + sizeof(state.buf_); - state.AddStr(filename, strlen(filename)) - && state.AddStr(":", 1) - && state.AddNum(line, 10) - && state.AddStr("]", 1) - && state.Add(a) - && state.Add(b) - && state.Add(c) - && state.Add(d); - - // Teminate with newline - if (state.p_ >= state.end_) { - state.p_ = state.end_ - 1; - } - *state.p_ = '\n'; - state.p_++; - - int msglen = state.p_ - state.buf_; - if (mode == kLog) { - (*log_message_writer)(state.buf_, msglen); - return; - } - - bool first_crash = false; - { - SpinLockHolder l(&crash_lock); - if (!crashed) { - crashed = true; - first_crash = true; - } - } - - (*log_message_writer)(state.buf_, msglen); - if (first_crash && mode == kCrashWithStats) { - MallocExtension::instance()->GetStats(stats_buffer, kStatsBufferSize); - (*log_message_writer)(stats_buffer, strlen(stats_buffer)); - } - - abort(); -} - -bool Logger::Add(const LogItem& item) { - // Separate items with spaces - if (p_ < end_) { - *p_ = ' '; - p_++; - } - - switch (item.tag_) { - case LogItem::kStr: - return AddStr(item.u_.str, strlen(item.u_.str)); - case LogItem::kUnsigned: - return AddNum(item.u_.unum, 10); - case LogItem::kSigned: - if (item.u_.snum < 0) { - // The cast to uint64_t is intentionally before the negation - // so that we do not attempt to negate -2^63. - return AddStr("-", 1) - && AddNum(- static_cast(item.u_.snum), 10); - } else { - return AddNum(static_cast(item.u_.snum), 10); - } - case LogItem::kPtr: - return AddStr("0x", 2) - && AddNum(reinterpret_cast(item.u_.ptr), 16); - default: - return false; - } -} - -bool Logger::AddStr(const char* str, int n) { - if (end_ - p_ < n) { - return false; - } else { - memcpy(p_, str, n); - p_ += n; - return true; - } -} - -bool Logger::AddNum(uint64_t num, int base) { - static const char kDigits[] = "0123456789abcdef"; - char space[22]; // more than enough for 2^64 in smallest supported base (10) - char* end = space + sizeof(space); - char* pos = end; - do { - pos--; - *pos = kDigits[num % base]; - num /= base; - } while (num > 0 && pos > space); - return AddStr(pos, end - pos); -} - -} // end tcmalloc namespace - -void TCMalloc_Printer::printf(const char* format, ...) { - if (left_ > 0) { - va_list ap; - va_start(ap, format); - const int r = perftools_vsnprintf(buf_, left_, format, ap); - va_end(ap); - if (r < 0) { - // Perhaps an old glibc that returns -1 on truncation? - left_ = 0; - } else if (r > left_) { - // Truncation - left_ = 0; - } else { - left_ -= r; - buf_ += r; - } - } -} diff --git a/contrib/libtcmalloc/src/internal_logging.h b/contrib/libtcmalloc/src/internal_logging.h deleted file mode 100644 index c6363894911..00000000000 --- a/contrib/libtcmalloc/src/internal_logging.h +++ /dev/null @@ -1,144 +0,0 @@ -// -*- Mode: C++; c-basic-offset: 2; indent-tabs-mode: nil -*- -// Copyright (c) 2005, Google Inc. -// All rights reserved. -// -// Redistribution and use in source and binary forms, with or without -// modification, are permitted provided that the following conditions are -// met: -// -// * Redistributions of source code must retain the above copyright -// notice, this list of conditions and the following disclaimer. -// * Redistributions in binary form must reproduce the above -// copyright notice, this list of conditions and the following disclaimer -// in the documentation and/or other materials provided with the -// distribution. -// * Neither the name of Google Inc. nor the names of its -// contributors may be used to endorse or promote products derived from -// this software without specific prior written permission. -// -// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS -// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT -// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR -// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT -// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, -// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT -// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, -// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY -// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT -// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE -// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. - -// --- -// Author: Sanjay Ghemawat -// -// Internal logging and related utility routines. - -#ifndef TCMALLOC_INTERNAL_LOGGING_H_ -#define TCMALLOC_INTERNAL_LOGGING_H_ - -#include "config.h" -#include // for size_t -#if defined HAVE_STDINT_H -#include -#elif defined HAVE_INTTYPES_H -#include -#else -#include -#endif - -//------------------------------------------------------------------- -// Utility routines -//------------------------------------------------------------------- - -// Safe logging helper: we write directly to the stderr file -// descriptor and avoid FILE buffering because that may invoke -// malloc(). -// -// Example: -// Log(kLog, __FILE__, __LINE__, "error", bytes); - -namespace tcmalloc { -enum LogMode { - kLog, // Just print the message - kCrash, // Print the message and crash - kCrashWithStats // Print the message, some stats, and crash -}; - -class Logger; - -// A LogItem holds any of the argument types that can be passed to Log() -class LogItem { - public: - LogItem() : tag_(kEnd) { } - LogItem(const char* v) : tag_(kStr) { u_.str = v; } - LogItem(int v) : tag_(kSigned) { u_.snum = v; } - LogItem(long v) : tag_(kSigned) { u_.snum = v; } - LogItem(long long v) : tag_(kSigned) { u_.snum = v; } - LogItem(unsigned int v) : tag_(kUnsigned) { u_.unum = v; } - LogItem(unsigned long v) : tag_(kUnsigned) { u_.unum = v; } - LogItem(unsigned long long v) : tag_(kUnsigned) { u_.unum = v; } - LogItem(const void* v) : tag_(kPtr) { u_.ptr = v; } - private: - friend class Logger; - enum Tag { - kStr, - kSigned, - kUnsigned, - kPtr, - kEnd - }; - Tag tag_; - union { - const char* str; - const void* ptr; - int64_t snum; - uint64_t unum; - } u_; -}; - -extern PERFTOOLS_DLL_DECL void Log(LogMode mode, const char* filename, int line, - LogItem a, LogItem b = LogItem(), - LogItem c = LogItem(), LogItem d = LogItem()); - -// Tests can override this function to collect logging messages. -extern PERFTOOLS_DLL_DECL void (*log_message_writer)(const char* msg, int length); - -} // end tcmalloc namespace - -// Like assert(), but executed even in NDEBUG mode -#undef CHECK_CONDITION -#define CHECK_CONDITION(cond) \ -do { \ - if (!(cond)) { \ - ::tcmalloc::Log(::tcmalloc::kCrash, __FILE__, __LINE__, #cond); \ - } \ -} while (0) - -// Our own version of assert() so we can avoid hanging by trying to do -// all kinds of goofy printing while holding the malloc lock. -#ifndef NDEBUG -#define ASSERT(cond) CHECK_CONDITION(cond) -#else -#define ASSERT(cond) ((void) 0) -#endif - -// Print into buffer -class TCMalloc_Printer { - private: - char* buf_; // Where should we write next - int left_; // Space left in buffer (including space for \0) - - public: - // REQUIRES: "length > 0" - TCMalloc_Printer(char* buf, int length) : buf_(buf), left_(length) { - buf[0] = '\0'; - } - - void printf(const char* format, ...) -#ifdef HAVE___ATTRIBUTE__ - __attribute__ ((__format__ (__printf__, 2, 3))) -#endif -; -}; - -#endif // TCMALLOC_INTERNAL_LOGGING_H_ diff --git a/contrib/libtcmalloc/src/libc_override.h b/contrib/libtcmalloc/src/libc_override.h deleted file mode 100644 index 0dbabb2d169..00000000000 --- a/contrib/libtcmalloc/src/libc_override.h +++ /dev/null @@ -1,91 +0,0 @@ -// -*- Mode: C++; c-basic-offset: 2; indent-tabs-mode: nil -*- -// Copyright (c) 2011, Google Inc. -// All rights reserved. -// -// Redistribution and use in source and binary forms, with or without -// modification, are permitted provided that the following conditions are -// met: -// -// * Redistributions of source code must retain the above copyright -// notice, this list of conditions and the following disclaimer. -// * Redistributions in binary form must reproduce the above -// copyright notice, this list of conditions and the following disclaimer -// in the documentation and/or other materials provided with the -// distribution. -// * Neither the name of Google Inc. nor the names of its -// contributors may be used to endorse or promote products derived from -// this software without specific prior written permission. -// -// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS -// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT -// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR -// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT -// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, -// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT -// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, -// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY -// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT -// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE -// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. - -// --- -// Author: Craig Silverstein -// -// This .h file imports the code that causes tcmalloc to override libc -// versions of malloc/free/new/delete/etc. That is, it provides the -// logic that makes it so calls to malloc(10) go through tcmalloc, -// rather than the default (libc) malloc. -// -// This file also provides a method: ReplaceSystemAlloc(), that every -// libc_override_*.h file it #includes is required to provide. This -// is called when first setting up tcmalloc -- that is, when a global -// constructor in tcmalloc.cc is executed -- to do any initialization -// work that may be required for this OS. (Note we cannot entirely -// control when tcmalloc is initialized, and the system may do some -// mallocs and frees before this routine is called.) It may be a -// noop. -// -// Every libc has its own way of doing this, and sometimes the compiler -// matters too, so we have a different file for each libc, and often -// for different compilers and OS's. - -#ifndef TCMALLOC_LIBC_OVERRIDE_INL_H_ -#define TCMALLOC_LIBC_OVERRIDE_INL_H_ - -#include "config.h" -#ifdef HAVE_FEATURES_H -#include // for __GLIBC__ -#endif -#include - -static void ReplaceSystemAlloc(); // defined in the .h files below - -// For windows, there are two ways to get tcmalloc. If we're -// patching, then src/windows/patch_function.cc will do the necessary -// overriding here. Otherwise, we doing the 'redefine' trick, where -// we remove malloc/new/etc from mscvcrt.dll, and just need to define -// them now. -#if defined(_WIN32) && defined(WIN32_DO_PATCHING) -void PatchWindowsFunctions(); // in src/windows/patch_function.cc -static void ReplaceSystemAlloc() { PatchWindowsFunctions(); } - -#elif defined(_WIN32) && !defined(WIN32_DO_PATCHING) -#include "libc_override_redefine.h" - -#elif defined(__APPLE__) -#include "libc_override_osx.h" - -#elif defined(__GLIBC__) -#include "libc_override_glibc.h" - -// Not all gcc systems necessarily support weak symbols, but all the -// ones I know of do, so for now just assume they all do. -#elif defined(__GNUC__) -#include "libc_override_gcc_and_weak.h" - -#else -#error Need to add support for your libc/OS here - -#endif - -#endif // TCMALLOC_LIBC_OVERRIDE_INL_H_ diff --git a/contrib/libtcmalloc/src/libc_override_gcc_and_weak.h b/contrib/libtcmalloc/src/libc_override_gcc_and_weak.h deleted file mode 100644 index 323d615d0ac..00000000000 --- a/contrib/libtcmalloc/src/libc_override_gcc_and_weak.h +++ /dev/null @@ -1,172 +0,0 @@ -// -*- Mode: C++; c-basic-offset: 2; indent-tabs-mode: nil -*- -// Copyright (c) 2011, Google Inc. -// All rights reserved. -// -// Redistribution and use in source and binary forms, with or without -// modification, are permitted provided that the following conditions are -// met: -// -// * Redistributions of source code must retain the above copyright -// notice, this list of conditions and the following disclaimer. -// * Redistributions in binary form must reproduce the above -// copyright notice, this list of conditions and the following disclaimer -// in the documentation and/or other materials provided with the -// distribution. -// * Neither the name of Google Inc. nor the names of its -// contributors may be used to endorse or promote products derived from -// this software without specific prior written permission. -// -// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS -// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT -// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR -// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT -// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, -// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT -// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, -// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY -// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT -// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE -// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. - -// --- -// Author: Craig Silverstein -// -// Used to override malloc routines on systems that define the -// memory allocation routines to be weak symbols in their libc -// (almost all unix-based systems are like this), on gcc, which -// suppports the 'alias' attribute. - -#ifndef TCMALLOC_LIBC_OVERRIDE_GCC_AND_WEAK_INL_H_ -#define TCMALLOC_LIBC_OVERRIDE_GCC_AND_WEAK_INL_H_ - -#ifdef HAVE_SYS_CDEFS_H -#include // for __THROW -#endif -#include - -#include "getenv_safe.h" // TCMallocGetenvSafe -#include "base/commandlineflags.h" - -#ifndef __THROW // I guess we're not on a glibc-like system -# define __THROW // __THROW is just an optimization, so ok to make it "" -#endif - -#ifndef __GNUC__ -# error libc_override_gcc_and_weak.h is for gcc distributions only. -#endif - -#define ALIAS(tc_fn) __attribute__ ((alias (#tc_fn), used)) - -void* operator new(size_t size) - ALIAS(tc_new); -void operator delete(void* p) noexcept - ALIAS(tc_delete); -void* operator new[](size_t size) - ALIAS(tc_newarray); -void operator delete[](void* p) noexcept - ALIAS(tc_deletearray); -void* operator new(size_t size, const std::nothrow_t& nt) noexcept - ALIAS(tc_new_nothrow); -void* operator new[](size_t size, const std::nothrow_t& nt) noexcept - ALIAS(tc_newarray_nothrow); -void operator delete(void* p, const std::nothrow_t& nt) noexcept - ALIAS(tc_delete_nothrow); -void operator delete[](void* p, const std::nothrow_t& nt) noexcept - ALIAS(tc_deletearray_nothrow); - -#if defined(ENABLE_SIZED_DELETE) - -void operator delete(void *p, size_t size) throw() - ALIAS(tc_delete_sized); -void operator delete[](void *p, size_t size) throw() - ALIAS(tc_deletearray_sized); - -#elif defined(ENABLE_DYNAMIC_SIZED_DELETE) && \ - (__GNUC__ * 100 + __GNUC_MINOR__) >= 405 - -static void delegate_sized_delete(void *p, size_t s) throw() { - (operator delete)(p); -} - -static void delegate_sized_deletearray(void *p, size_t s) throw() { - (operator delete[])(p); -} - -extern "C" __attribute__((weak)) -int tcmalloc_sized_delete_enabled(void); - -static bool sized_delete_enabled(void) { - if (tcmalloc_sized_delete_enabled != 0) { - return !!tcmalloc_sized_delete_enabled(); - } - - const char *flag = TCMallocGetenvSafe("TCMALLOC_ENABLE_SIZED_DELETE"); - return tcmalloc::commandlineflags::StringToBool(flag, false); -} - -extern "C" { - -static void *resolve_delete_sized(void) { - if (sized_delete_enabled()) { - return reinterpret_cast(tc_delete_sized); - } - return reinterpret_cast(delegate_sized_delete); -} - -static void *resolve_deletearray_sized(void) { - if (sized_delete_enabled()) { - return reinterpret_cast(tc_deletearray_sized); - } - return reinterpret_cast(delegate_sized_deletearray); -} - -} - -void operator delete(void *p, size_t size) throw() - __attribute__((ifunc("resolve_delete_sized"))); -void operator delete[](void *p, size_t size) throw() - __attribute__((ifunc("resolve_deletearray_sized"))); - -#else /* !ENABLE_SIZED_DELETE && !ENABLE_DYN_SIZED_DELETE */ - -void operator delete(void *p, size_t size) throw() - ALIAS(tc_delete); -void operator delete[](void *p, size_t size) throw() - ALIAS(tc_deletearray); - -#endif /* !ENABLE_SIZED_DELETE && !ENABLE_DYN_SIZED_DELETE */ - -extern "C" { - void* malloc(size_t size) __THROW ALIAS(tc_malloc); - void free(void* ptr) __THROW ALIAS(tc_free); - void* realloc(void* ptr, size_t size) __THROW ALIAS(tc_realloc); - void* calloc(size_t n, size_t size) __THROW ALIAS(tc_calloc); - void cfree(void* ptr) __THROW ALIAS(tc_cfree); - void* memalign(size_t align, size_t s) __THROW ALIAS(tc_memalign); - void* valloc(size_t size) __THROW ALIAS(tc_valloc); - void* pvalloc(size_t size) __THROW ALIAS(tc_pvalloc); - int posix_memalign(void** r, size_t a, size_t s) __THROW - ALIAS(tc_posix_memalign); -#ifndef __UCLIBC__ - void malloc_stats(void) __THROW ALIAS(tc_malloc_stats); -#endif - int mallopt(int cmd, int value) __THROW ALIAS(tc_mallopt); -#ifdef HAVE_STRUCT_MALLINFO - struct mallinfo mallinfo(void) __THROW ALIAS(tc_mallinfo); -#endif - size_t malloc_size(void* p) __THROW ALIAS(tc_malloc_size); -#if defined(__ANDROID__) - size_t malloc_usable_size(const void* p) __THROW - ALIAS(tc_malloc_size); -#else - size_t malloc_usable_size(void* p) __THROW ALIAS(tc_malloc_size); -#endif -} // extern "C" - -#undef ALIAS - -// No need to do anything at tcmalloc-registration time: we do it all -// via overriding weak symbols (at link time). -static void ReplaceSystemAlloc() { } - -#endif // TCMALLOC_LIBC_OVERRIDE_GCC_AND_WEAK_INL_H_ diff --git a/contrib/libtcmalloc/src/libc_override_glibc.h b/contrib/libtcmalloc/src/libc_override_glibc.h deleted file mode 100644 index cc17df315c0..00000000000 --- a/contrib/libtcmalloc/src/libc_override_glibc.h +++ /dev/null @@ -1,92 +0,0 @@ -// -*- Mode: C++; c-basic-offset: 2; indent-tabs-mode: nil -*- -// Copyright (c) 2011, Google Inc. -// All rights reserved. -// -// Redistribution and use in source and binary forms, with or without -// modification, are permitted provided that the following conditions are -// met: -// -// * Redistributions of source code must retain the above copyright -// notice, this list of conditions and the following disclaimer. -// * Redistributions in binary form must reproduce the above -// copyright notice, this list of conditions and the following disclaimer -// in the documentation and/or other materials provided with the -// distribution. -// * Neither the name of Google Inc. nor the names of its -// contributors may be used to endorse or promote products derived from -// this software without specific prior written permission. -// -// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS -// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT -// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR -// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT -// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, -// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT -// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, -// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY -// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT -// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE -// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. - -// --- -// Author: Craig Silverstein -// -// Used to override malloc routines on systems that are using glibc. - -#ifndef TCMALLOC_LIBC_OVERRIDE_GLIBC_INL_H_ -#define TCMALLOC_LIBC_OVERRIDE_GLIBC_INL_H_ - -#include "config.h" -#include // for __GLIBC__ -#include - -#ifndef __GLIBC__ -# error libc_override_glibc.h is for glibc distributions only. -#endif - -// In glibc, the memory-allocation methods are weak symbols, so we can -// just override them with our own. If we're using gcc, we can use -// __attribute__((alias)) to do the overriding easily (exception: -// Mach-O, which doesn't support aliases). Otherwise we have to use a -// function call. -#if !defined(__GNUC__) || defined(__MACH__) - -// This also defines ReplaceSystemAlloc(). -# include "libc_override_redefine.h" // defines functions malloc()/etc - -#else // #if !defined(__GNUC__) || defined(__MACH__) - -// If we get here, we're a gcc system, so do all the overriding we do -// with gcc. This does the overriding of all the 'normal' memory -// allocation. This also defines ReplaceSystemAlloc(). -# include "libc_override_gcc_and_weak.h" - -// We also have to do some glibc-specific overriding. Some library -// routines on RedHat 9 allocate memory using malloc() and free it -// using __libc_free() (or vice-versa). Since we provide our own -// implementations of malloc/free, we need to make sure that the -// __libc_XXX variants (defined as part of glibc) also point to the -// same implementations. Since it only matters for redhat, we -// do it inside the gcc #ifdef, since redhat uses gcc. -// TODO(csilvers): only do this if we detect we're an old enough glibc? - -#define ALIAS(tc_fn) __attribute__ ((alias (#tc_fn))) -extern "C" { - void* __libc_malloc(size_t size) ALIAS(tc_malloc); - void __libc_free(void* ptr) ALIAS(tc_free); - void* __libc_realloc(void* ptr, size_t size) ALIAS(tc_realloc); - void* __libc_calloc(size_t n, size_t size) ALIAS(tc_calloc); - void __libc_cfree(void* ptr) ALIAS(tc_cfree); - void* __libc_memalign(size_t align, size_t s) ALIAS(tc_memalign); - void* __libc_valloc(size_t size) ALIAS(tc_valloc); - void* __libc_pvalloc(size_t size) ALIAS(tc_pvalloc); - int __posix_memalign(void** r, size_t a, size_t s) ALIAS(tc_posix_memalign); -} // extern "C" -#undef ALIAS - -#endif // #if defined(__GNUC__) && !defined(__MACH__) - -// No need to write ReplaceSystemAlloc(); one of the #includes above -// did it for us. - -#endif // TCMALLOC_LIBC_OVERRIDE_GLIBC_INL_H_ diff --git a/contrib/libtcmalloc/src/libc_override_osx.h b/contrib/libtcmalloc/src/libc_override_osx.h deleted file mode 100644 index afd57d1560a..00000000000 --- a/contrib/libtcmalloc/src/libc_override_osx.h +++ /dev/null @@ -1,308 +0,0 @@ -// -*- Mode: C++; c-basic-offset: 2; indent-tabs-mode: nil -*- -// Copyright (c) 2011, Google Inc. -// All rights reserved. -// -// Redistribution and use in source and binary forms, with or without -// modification, are permitted provided that the following conditions are -// met: -// -// * Redistributions of source code must retain the above copyright -// notice, this list of conditions and the following disclaimer. -// * Redistributions in binary form must reproduce the above -// copyright notice, this list of conditions and the following disclaimer -// in the documentation and/or other materials provided with the -// distribution. -// * Neither the name of Google Inc. nor the names of its -// contributors may be used to endorse or promote products derived from -// this software without specific prior written permission. -// -// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS -// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT -// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR -// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT -// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, -// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT -// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, -// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY -// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT -// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE -// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. - -// --- -// Author: Craig Silverstein -// -// Used to override malloc routines on OS X systems. We use the -// malloc-zone functionality built into OS X to register our malloc -// routine. -// -// 1) We used to use the normal 'override weak libc malloc/etc' -// technique for OS X. This is not optimal because mach does not -// support the 'alias' attribute, so we had to have forwarding -// functions. It also does not work very well with OS X shared -// libraries (dylibs) -- in general, the shared libs don't use -// tcmalloc unless run with the DYLD_FORCE_FLAT_NAMESPACE envvar. -// -// 2) Another approach would be to use an interposition array: -// static const interpose_t interposers[] __attribute__((section("__DATA, __interpose"))) = { -// { (void *)tc_malloc, (void *)malloc }, -// { (void *)tc_free, (void *)free }, -// }; -// This requires the user to set the DYLD_INSERT_LIBRARIES envvar, so -// is not much better. -// -// 3) Registering a new malloc zone avoids all these issues: -// http://www.opensource.apple.com/source/Libc/Libc-583/include/malloc/malloc.h -// http://www.opensource.apple.com/source/Libc/Libc-583/gen/malloc.c -// If we make tcmalloc the default malloc zone (undocumented but -// possible) then all new allocs use it, even those in shared -// libraries. Allocs done before tcmalloc was installed, or in libs -// that aren't using tcmalloc for some reason, will correctly go -// through the malloc-zone interface when free-ing, and will pick up -// the libc free rather than tcmalloc free. So it should "never" -// cause a crash (famous last words). -// -// 4) The routines one must define for one's own malloc have changed -// between OS X versions. This requires some hoops on our part, but -// is only really annoying when it comes to posix_memalign. The right -// behavior there depends on what OS version tcmalloc was compiled on, -// but also what OS version the program is running on. For now, we -// punt and don't implement our own posix_memalign. Apps that really -// care can use tc_posix_memalign directly. - -#ifndef TCMALLOC_LIBC_OVERRIDE_OSX_INL_H_ -#define TCMALLOC_LIBC_OVERRIDE_OSX_INL_H_ - -#include "config.h" -#ifdef HAVE_FEATURES_H -#include -#endif -#include - -#if !defined(__APPLE__) -# error libc_override_glibc-osx.h is for OS X distributions only. -#endif - -#include -#include - -namespace tcmalloc { - void CentralCacheLockAll(); - void CentralCacheUnlockAll(); -} - -// from AvailabilityMacros.h -#if defined(MAC_OS_X_VERSION_10_6) && \ - MAC_OS_X_VERSION_MAX_ALLOWED >= MAC_OS_X_VERSION_10_6 -extern "C" { - // This function is only available on 10.6 (and later) but the - // LibSystem headers do not use AvailabilityMacros.h to handle weak - // importing automatically. This prototype is a copy of the one in - // with the WEAK_IMPORT_ATTRBIUTE added. - extern malloc_zone_t *malloc_default_purgeable_zone(void) - WEAK_IMPORT_ATTRIBUTE; -} -#endif - -// We need to provide wrappers around all the libc functions. -namespace { -size_t mz_size(malloc_zone_t* zone, const void* ptr) { - if (MallocExtension::instance()->GetOwnership(ptr) != MallocExtension::kOwned) - return 0; // malloc_zone semantics: return 0 if we don't own the memory - - // TODO(csilvers): change this method to take a const void*, one day. - return MallocExtension::instance()->GetAllocatedSize(const_cast(ptr)); -} - -void* mz_malloc(malloc_zone_t* zone, size_t size) { - return tc_malloc(size); -} - -void* mz_calloc(malloc_zone_t* zone, size_t num_items, size_t size) { - return tc_calloc(num_items, size); -} - -void* mz_valloc(malloc_zone_t* zone, size_t size) { - return tc_valloc(size); -} - -void mz_free(malloc_zone_t* zone, void* ptr) { - return tc_free(ptr); -} - -void* mz_realloc(malloc_zone_t* zone, void* ptr, size_t size) { - return tc_realloc(ptr, size); -} - -void* mz_memalign(malloc_zone_t* zone, size_t align, size_t size) { - return tc_memalign(align, size); -} - -void mz_destroy(malloc_zone_t* zone) { - // A no-op -- we will not be destroyed! -} - -// malloc_introspection callbacks. I'm not clear on what all of these do. -kern_return_t mi_enumerator(task_t task, void *, - unsigned type_mask, vm_address_t zone_address, - memory_reader_t reader, - vm_range_recorder_t recorder) { - // Should enumerate all the pointers we have. Seems like a lot of work. - return KERN_FAILURE; -} - -size_t mi_good_size(malloc_zone_t *zone, size_t size) { - // I think it's always safe to return size, but we maybe could do better. - return size; -} - -boolean_t mi_check(malloc_zone_t *zone) { - return MallocExtension::instance()->VerifyAllMemory(); -} - -void mi_print(malloc_zone_t *zone, boolean_t verbose) { - int bufsize = 8192; - if (verbose) - bufsize = 102400; // I picked this size arbitrarily - char* buffer = new char[bufsize]; - MallocExtension::instance()->GetStats(buffer, bufsize); - fprintf(stdout, "%s", buffer); - delete[] buffer; -} - -void mi_log(malloc_zone_t *zone, void *address) { - // I don't think we support anything like this -} - -void mi_force_lock(malloc_zone_t *zone) { - tcmalloc::CentralCacheLockAll(); -} - -void mi_force_unlock(malloc_zone_t *zone) { - tcmalloc::CentralCacheUnlockAll(); -} - -void mi_statistics(malloc_zone_t *zone, malloc_statistics_t *stats) { - // TODO(csilvers): figure out how to fill these out - stats->blocks_in_use = 0; - stats->size_in_use = 0; - stats->max_size_in_use = 0; - stats->size_allocated = 0; -} - -boolean_t mi_zone_locked(malloc_zone_t *zone) { - return false; // Hopefully unneeded by us! -} - -} // unnamed namespace - -// OS X doesn't have pvalloc, cfree, malloc_statc, etc, so we can just -// define our own. :-) OS X supplies posix_memalign in some versions -// but not others, either strongly or weakly linked, in a way that's -// difficult enough to code to correctly, that I just don't try to -// support either memalign() or posix_memalign(). If you need them -// and are willing to code to tcmalloc, you can use tc_posix_memalign(). -extern "C" { - void cfree(void* p) { tc_cfree(p); } - void* pvalloc(size_t s) { return tc_pvalloc(s); } - void malloc_stats(void) { tc_malloc_stats(); } - int mallopt(int cmd, int v) { return tc_mallopt(cmd, v); } - // No struct mallinfo on OS X, so don't define mallinfo(). - // An alias for malloc_size(), which OS X defines. - size_t malloc_usable_size(void* p) { return tc_malloc_size(p); } -} // extern "C" - -static malloc_zone_t *get_default_zone() { - malloc_zone_t **zones = NULL; - unsigned int num_zones = 0; - - /* - * On OSX 10.12, malloc_default_zone returns a special zone that is not - * present in the list of registered zones. That zone uses a "lite zone" - * if one is present (apparently enabled when malloc stack logging is - * enabled), or the first registered zone otherwise. In practice this - * means unless malloc stack logging is enabled, the first registered - * zone is the default. - * So get the list of zones to get the first one, instead of relying on - * malloc_default_zone. - */ - if (KERN_SUCCESS != malloc_get_all_zones(0, NULL, (vm_address_t**) &zones, - &num_zones)) { - /* Reset the value in case the failure happened after it was set. */ - num_zones = 0; - } - - if (num_zones) - return zones[0]; - - return malloc_default_zone(); -} - - -static void ReplaceSystemAlloc() { - static malloc_introspection_t tcmalloc_introspection; - memset(&tcmalloc_introspection, 0, sizeof(tcmalloc_introspection)); - - tcmalloc_introspection.enumerator = &mi_enumerator; - tcmalloc_introspection.good_size = &mi_good_size; - tcmalloc_introspection.check = &mi_check; - tcmalloc_introspection.print = &mi_print; - tcmalloc_introspection.log = &mi_log; - tcmalloc_introspection.force_lock = &mi_force_lock; - tcmalloc_introspection.force_unlock = &mi_force_unlock; - - static malloc_zone_t tcmalloc_zone; - memset(&tcmalloc_zone, 0, sizeof(malloc_zone_t)); - - // Start with a version 4 zone which is used for OS X 10.4 and 10.5. - tcmalloc_zone.version = 4; - tcmalloc_zone.zone_name = "tcmalloc"; - tcmalloc_zone.size = &mz_size; - tcmalloc_zone.malloc = &mz_malloc; - tcmalloc_zone.calloc = &mz_calloc; - tcmalloc_zone.valloc = &mz_valloc; - tcmalloc_zone.free = &mz_free; - tcmalloc_zone.realloc = &mz_realloc; - tcmalloc_zone.destroy = &mz_destroy; - tcmalloc_zone.batch_malloc = NULL; - tcmalloc_zone.batch_free = NULL; - tcmalloc_zone.introspect = &tcmalloc_introspection; - - // from AvailabilityMacros.h -#if defined(MAC_OS_X_VERSION_10_6) && \ - MAC_OS_X_VERSION_MAX_ALLOWED >= MAC_OS_X_VERSION_10_6 - // Switch to version 6 on OSX 10.6 to support memalign. - tcmalloc_zone.version = 6; - tcmalloc_zone.free_definite_size = NULL; - tcmalloc_zone.memalign = &mz_memalign; - tcmalloc_introspection.zone_locked = &mi_zone_locked; - - // Request the default purgable zone to force its creation. The - // current default zone is registered with the purgable zone for - // doing tiny and small allocs. Sadly, it assumes that the default - // zone is the szone implementation from OS X and will crash if it - // isn't. By creating the zone now, this will be true and changing - // the default zone won't cause a problem. This only needs to - // happen when actually running on OS X 10.6 and higher (note the - // ifdef above only checks if we were *compiled* with 10.6 or - // higher; at runtime we have to check if this symbol is defined.) - if (malloc_default_purgeable_zone) { - malloc_default_purgeable_zone(); - } -#endif - - // Register the tcmalloc zone. At this point, it will not be the - // default zone. - malloc_zone_register(&tcmalloc_zone); - - // Unregister and reregister the default zone. Unregistering swaps - // the specified zone with the last one registered which for the - // default zone makes the more recently registered zone the default - // zone. The default zone is then re-registered to ensure that - // allocations made from it earlier will be handled correctly. - // Things are not guaranteed to work that way, but it's how they work now. - malloc_zone_t *default_zone = get_default_zone(); - malloc_zone_unregister(default_zone); - malloc_zone_register(default_zone); -} - -#endif // TCMALLOC_LIBC_OVERRIDE_OSX_INL_H_ diff --git a/contrib/libtcmalloc/src/libc_override_redefine.h b/contrib/libtcmalloc/src/libc_override_redefine.h deleted file mode 100644 index 72679ef38b8..00000000000 --- a/contrib/libtcmalloc/src/libc_override_redefine.h +++ /dev/null @@ -1,92 +0,0 @@ -// -*- Mode: C++; c-basic-offset: 2; indent-tabs-mode: nil -*- -// Copyright (c) 2011, Google Inc. -// All rights reserved. -// -// Redistribution and use in source and binary forms, with or without -// modification, are permitted provided that the following conditions are -// met: -// -// * Redistributions of source code must retain the above copyright -// notice, this list of conditions and the following disclaimer. -// * Redistributions in binary form must reproduce the above -// copyright notice, this list of conditions and the following disclaimer -// in the documentation and/or other materials provided with the -// distribution. -// * Neither the name of Google Inc. nor the names of its -// contributors may be used to endorse or promote products derived from -// this software without specific prior written permission. -// -// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS -// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT -// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR -// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT -// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, -// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT -// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, -// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY -// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT -// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE -// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. - -// --- -// Author: Craig Silverstein -// -// Used on systems that don't have their own definition of -// malloc/new/etc. (Typically this will be a windows msvcrt.dll that -// has been edited to remove the definitions.) We can just define our -// own as normal functions. -// -// This should also work on systems were all the malloc routines are -// defined as weak symbols, and there's no support for aliasing. - -#ifndef TCMALLOC_LIBC_OVERRIDE_REDEFINE_H_ -#define TCMALLOC_LIBC_OVERRIDE_REDEFINE_H_ - -void* operator new(size_t size) { return tc_new(size); } -void operator delete(void* p) throw() { tc_delete(p); } -void* operator new[](size_t size) { return tc_newarray(size); } -void operator delete[](void* p) throw() { tc_deletearray(p); } -void* operator new(size_t size, const std::nothrow_t& nt) throw() { - return tc_new_nothrow(size, nt); -} -void* operator new[](size_t size, const std::nothrow_t& nt) throw() { - return tc_newarray_nothrow(size, nt); -} -void operator delete(void* ptr, const std::nothrow_t& nt) throw() { - return tc_delete_nothrow(ptr, nt); -} -void operator delete[](void* ptr, const std::nothrow_t& nt) throw() { - return tc_deletearray_nothrow(ptr, nt); -} - -#ifdef ENABLE_SIZED_DELETE -void operator delete(void* p, size_t s) throw() { tc_delete_sized(p, s); } -void operator delete[](void* p, size_t s) throw(){ tc_deletearray_sized(p); } -#endif - -extern "C" { - void* malloc(size_t s) { return tc_malloc(s); } - void free(void* p) { tc_free(p); } - void* realloc(void* p, size_t s) { return tc_realloc(p, s); } - void* calloc(size_t n, size_t s) { return tc_calloc(n, s); } - void cfree(void* p) { tc_cfree(p); } - void* memalign(size_t a, size_t s) { return tc_memalign(a, s); } - void* valloc(size_t s) { return tc_valloc(s); } - void* pvalloc(size_t s) { return tc_pvalloc(s); } - int posix_memalign(void** r, size_t a, size_t s) { - return tc_posix_memalign(r, a, s); - } - void malloc_stats(void) { tc_malloc_stats(); } - int mallopt(int cmd, int v) { return tc_mallopt(cmd, v); } -#ifdef HAVE_STRUCT_MALLINFO - struct mallinfo mallinfo(void) { return tc_mallinfo(); } -#endif - size_t malloc_size(void* p) { return tc_malloc_size(p); } - size_t malloc_usable_size(void* p) { return tc_malloc_size(p); } -} // extern "C" - -// No need to do anything at tcmalloc-registration time: we do it all -// via overriding weak symbols (at link time). -static void ReplaceSystemAlloc() { } - -#endif // TCMALLOC_LIBC_OVERRIDE_REDEFINE_H_ diff --git a/contrib/libtcmalloc/src/linked_list.h b/contrib/libtcmalloc/src/linked_list.h deleted file mode 100644 index 66a07410760..00000000000 --- a/contrib/libtcmalloc/src/linked_list.h +++ /dev/null @@ -1,103 +0,0 @@ -// -*- Mode: C++; c-basic-offset: 2; indent-tabs-mode: nil -*- -// Copyright (c) 2008, Google Inc. -// All rights reserved. -// -// Redistribution and use in source and binary forms, with or without -// modification, are permitted provided that the following conditions are -// met: -// -// * Redistributions of source code must retain the above copyright -// notice, this list of conditions and the following disclaimer. -// * Redistributions in binary form must reproduce the above -// copyright notice, this list of conditions and the following disclaimer -// in the documentation and/or other materials provided with the -// distribution. -// * Neither the name of Google Inc. nor the names of its -// contributors may be used to endorse or promote products derived from -// this software without specific prior written permission. -// -// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS -// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT -// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR -// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT -// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, -// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT -// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, -// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY -// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT -// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE -// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. - -// --- -// Author: Sanjay Ghemawat -// -// Some very basic linked list functions for dealing with using void * as -// storage. - -#ifndef TCMALLOC_LINKED_LIST_H_ -#define TCMALLOC_LINKED_LIST_H_ - -#include - -namespace tcmalloc { - -inline void *SLL_Next(void *t) { - return *(reinterpret_cast(t)); -} - -inline void SLL_SetNext(void *t, void *n) { - *(reinterpret_cast(t)) = n; -} - -inline void SLL_Push(void **list, void *element) { - SLL_SetNext(element, *list); - *list = element; -} - -inline void *SLL_Pop(void **list) { - void *result = *list; - *list = SLL_Next(*list); - return result; -} - -// Remove N elements from a linked list to which head points. head will be -// modified to point to the new head. start and end will point to the first -// and last nodes of the range. Note that end will point to NULL after this -// function is called. -inline void SLL_PopRange(void **head, int N, void **start, void **end) { - if (N == 0) { - *start = NULL; - *end = NULL; - return; - } - - void *tmp = *head; - for (int i = 1; i < N; ++i) { - tmp = SLL_Next(tmp); - } - - *start = *head; - *end = tmp; - *head = SLL_Next(tmp); - // Unlink range from list. - SLL_SetNext(tmp, NULL); -} - -inline void SLL_PushRange(void **head, void *start, void *end) { - if (!start) return; - SLL_SetNext(end, *head); - *head = start; -} - -inline size_t SLL_Size(void *head) { - int count = 0; - while (head) { - count++; - head = SLL_Next(head); - } - return count; -} - -} // namespace tcmalloc - -#endif // TCMALLOC_LINKED_LIST_H_ diff --git a/contrib/libtcmalloc/src/malloc_extension.cc b/contrib/libtcmalloc/src/malloc_extension.cc deleted file mode 100644 index 13a06f4dbe8..00000000000 --- a/contrib/libtcmalloc/src/malloc_extension.cc +++ /dev/null @@ -1,388 +0,0 @@ -// -*- Mode: C++; c-basic-offset: 2; indent-tabs-mode: nil -*- -// Copyright (c) 2005, Google Inc. -// All rights reserved. -// -// Redistribution and use in source and binary forms, with or without -// modification, are permitted provided that the following conditions are -// met: -// -// * Redistributions of source code must retain the above copyright -// notice, this list of conditions and the following disclaimer. -// * Redistributions in binary form must reproduce the above -// copyright notice, this list of conditions and the following disclaimer -// in the documentation and/or other materials provided with the -// distribution. -// * Neither the name of Google Inc. nor the names of its -// contributors may be used to endorse or promote products derived from -// this software without specific prior written permission. -// -// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS -// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT -// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR -// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT -// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, -// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT -// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, -// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY -// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT -// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE -// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. - -// --- -// Author: Sanjay Ghemawat - -#include "config.h" -#include -#include -#include -#if defined HAVE_STDINT_H -#include -#elif defined HAVE_INTTYPES_H -#include -#else -#include -#endif -#include -#include "base/dynamic_annotations.h" -#include "base/sysinfo.h" // for FillProcSelfMaps -#ifndef NO_HEAP_CHECK -#include "gperftools/heap-checker.h" -#endif -#include "gperftools/malloc_extension.h" -#include "gperftools/malloc_extension_c.h" -#include "maybe_threads.h" -#include "base/googleinit.h" - -using STL_NAMESPACE::string; -using STL_NAMESPACE::vector; - -static void DumpAddressMap(string* result) { - *result += "\nMAPPED_LIBRARIES:\n"; - // We keep doubling until we get a fit - const size_t old_resultlen = result->size(); - for (int amap_size = 10240; amap_size < 10000000; amap_size *= 2) { - result->resize(old_resultlen + amap_size); - bool wrote_all = false; - const int bytes_written = - tcmalloc::FillProcSelfMaps(&((*result)[old_resultlen]), amap_size, - &wrote_all); - if (wrote_all) { // we fit! - (*result)[old_resultlen + bytes_written] = '\0'; - result->resize(old_resultlen + bytes_written); - return; - } - } - result->reserve(old_resultlen); // just don't print anything -} - -// Note: this routine is meant to be called before threads are spawned. -void MallocExtension::Initialize() { - static bool initialize_called = false; - - if (initialize_called) return; - initialize_called = true; - -#ifdef __GLIBC__ - // GNU libc++ versions 3.3 and 3.4 obey the environment variables - // GLIBCPP_FORCE_NEW and GLIBCXX_FORCE_NEW respectively. Setting - // one of these variables forces the STL default allocator to call - // new() or delete() for each allocation or deletion. Otherwise - // the STL allocator tries to avoid the high cost of doing - // allocations by pooling memory internally. However, tcmalloc - // does allocations really fast, especially for the types of small - // items one sees in STL, so it's better off just using us. - // TODO: control whether we do this via an environment variable? - setenv("GLIBCPP_FORCE_NEW", "1", false /* no overwrite*/); - setenv("GLIBCXX_FORCE_NEW", "1", false /* no overwrite*/); - - // Now we need to make the setenv 'stick', which it may not do since - // the env is flakey before main() is called. But luckily stl only - // looks at this env var the first time it tries to do an alloc, and - // caches what it finds. So we just cause an stl alloc here. - string dummy("I need to be allocated"); - dummy += "!"; // so the definition of dummy isn't optimized out -#endif /* __GLIBC__ */ -} - -// SysAllocator implementation -SysAllocator::~SysAllocator() {} - -// Default implementation -- does nothing -MallocExtension::~MallocExtension() { } -bool MallocExtension::VerifyAllMemory() { return true; } -bool MallocExtension::VerifyNewMemory(const void* p) { return true; } -bool MallocExtension::VerifyArrayNewMemory(const void* p) { return true; } -bool MallocExtension::VerifyMallocMemory(const void* p) { return true; } - -bool MallocExtension::GetNumericProperty(const char* property, size_t* value) { - return false; -} - -bool MallocExtension::SetNumericProperty(const char* property, size_t value) { - return false; -} - -void MallocExtension::GetStats(char* buffer, int length) { - assert(length > 0); - buffer[0] = '\0'; -} - -bool MallocExtension::MallocMemoryStats(int* blocks, size_t* total, - int histogram[kMallocHistogramSize]) { - *blocks = 0; - *total = 0; - memset(histogram, 0, sizeof(*histogram) * kMallocHistogramSize); - return true; -} - -void** MallocExtension::ReadStackTraces(int* sample_period) { - return NULL; -} - -void** MallocExtension::ReadHeapGrowthStackTraces() { - return NULL; -} - -void MallocExtension::MarkThreadIdle() { - // Default implementation does nothing -} - -void MallocExtension::MarkThreadBusy() { - // Default implementation does nothing -} - -SysAllocator* MallocExtension::GetSystemAllocator() { - return NULL; -} - -void MallocExtension::SetSystemAllocator(SysAllocator *a) { - // Default implementation does nothing -} - -void MallocExtension::ReleaseToSystem(size_t num_bytes) { - // Default implementation does nothing -} - -void MallocExtension::ReleaseFreeMemory() { - ReleaseToSystem(static_cast(-1)); // SIZE_T_MAX -} - -void MallocExtension::SetMemoryReleaseRate(double rate) { - // Default implementation does nothing -} - -double MallocExtension::GetMemoryReleaseRate() { - return -1.0; -} - -size_t MallocExtension::GetEstimatedAllocatedSize(size_t size) { - return size; -} - -size_t MallocExtension::GetAllocatedSize(const void* p) { - assert(GetOwnership(p) != kNotOwned); - return 0; -} - -MallocExtension::Ownership MallocExtension::GetOwnership(const void* p) { - return kUnknownOwnership; -} - -void MallocExtension::GetFreeListSizes( - vector* v) { - v->clear(); -} - -size_t MallocExtension::GetThreadCacheSize() { - return 0; -} - -void MallocExtension::MarkThreadTemporarilyIdle() { - // Default implementation does nothing -} - -// The current malloc extension object. - -static MallocExtension* current_instance; - -static void InitModule() { - if (current_instance != NULL) { - return; - } - current_instance = new MallocExtension; -#ifndef NO_HEAP_CHECK - HeapLeakChecker::IgnoreObject(current_instance); -#endif -} - -REGISTER_MODULE_INITIALIZER(malloc_extension_init, InitModule()) - -MallocExtension* MallocExtension::instance() { - InitModule(); - return current_instance; -} - -void MallocExtension::Register(MallocExtension* implementation) { - InitModule(); - // When running under valgrind, our custom malloc is replaced with - // valgrind's one and malloc extensions will not work. (Note: - // callers should be responsible for checking that they are the - // malloc that is really being run, before calling Register. This - // is just here as an extra sanity check.) - if (!RunningOnValgrind()) { - current_instance = implementation; - } -} - -// ----------------------------------------------------------------------- -// Heap sampling support -// ----------------------------------------------------------------------- - -namespace { - -// Accessors -uintptr_t Count(void** entry) { - return reinterpret_cast(entry[0]); -} -uintptr_t Size(void** entry) { - return reinterpret_cast(entry[1]); -} -uintptr_t Depth(void** entry) { - return reinterpret_cast(entry[2]); -} -void* PC(void** entry, int i) { - return entry[3+i]; -} - -void PrintCountAndSize(MallocExtensionWriter* writer, - uintptr_t count, uintptr_t size) { - char buf[100]; - snprintf(buf, sizeof(buf), - "%6" PRIu64 ": %8" PRIu64 " [%6" PRIu64 ": %8" PRIu64 "] @", - static_cast(count), - static_cast(size), - static_cast(count), - static_cast(size)); - writer->append(buf, strlen(buf)); -} - -void PrintHeader(MallocExtensionWriter* writer, - const char* label, void** entries) { - // Compute the total count and total size - uintptr_t total_count = 0; - uintptr_t total_size = 0; - for (void** entry = entries; Count(entry) != 0; entry += 3 + Depth(entry)) { - total_count += Count(entry); - total_size += Size(entry); - } - - const char* const kTitle = "heap profile: "; - writer->append(kTitle, strlen(kTitle)); - PrintCountAndSize(writer, total_count, total_size); - writer->append(" ", 1); - writer->append(label, strlen(label)); - writer->append("\n", 1); -} - -void PrintStackEntry(MallocExtensionWriter* writer, void** entry) { - PrintCountAndSize(writer, Count(entry), Size(entry)); - - for (int i = 0; i < Depth(entry); i++) { - char buf[32]; - snprintf(buf, sizeof(buf), " %p", PC(entry, i)); - writer->append(buf, strlen(buf)); - } - writer->append("\n", 1); -} - -} - -void MallocExtension::GetHeapSample(MallocExtensionWriter* writer) { - int sample_period = 0; - void** entries = ReadStackTraces(&sample_period); - if (entries == NULL) { - const char* const kErrorMsg = - "This malloc implementation does not support sampling.\n" - "As of 2005/01/26, only tcmalloc supports sampling, and\n" - "you are probably running a binary that does not use\n" - "tcmalloc.\n"; - writer->append(kErrorMsg, strlen(kErrorMsg)); - return; - } - - char label[32]; - sprintf(label, "heap_v2/%d", sample_period); - PrintHeader(writer, label, entries); - for (void** entry = entries; Count(entry) != 0; entry += 3 + Depth(entry)) { - PrintStackEntry(writer, entry); - } - delete[] entries; - - DumpAddressMap(writer); -} - -void MallocExtension::GetHeapGrowthStacks(MallocExtensionWriter* writer) { - void** entries = ReadHeapGrowthStackTraces(); - if (entries == NULL) { - const char* const kErrorMsg = - "This malloc implementation does not support " - "ReadHeapGrowthStackTraces().\n" - "As of 2005/09/27, only tcmalloc supports this, and you\n" - "are probably running a binary that does not use tcmalloc.\n"; - writer->append(kErrorMsg, strlen(kErrorMsg)); - return; - } - - // Do not canonicalize the stack entries, so that we get a - // time-ordered list of stack traces, which may be useful if the - // client wants to focus on the latest stack traces. - PrintHeader(writer, "growth", entries); - for (void** entry = entries; Count(entry) != 0; entry += 3 + Depth(entry)) { - PrintStackEntry(writer, entry); - } - delete[] entries; - - DumpAddressMap(writer); -} - -void MallocExtension::Ranges(void* arg, RangeFunction func) { - // No callbacks by default -} - -// These are C shims that work on the current instance. - -#define C_SHIM(fn, retval, paramlist, arglist) \ - extern "C" PERFTOOLS_DLL_DECL retval MallocExtension_##fn paramlist { \ - return MallocExtension::instance()->fn arglist; \ - } - -C_SHIM(VerifyAllMemory, int, (void), ()); -C_SHIM(VerifyNewMemory, int, (const void* p), (p)); -C_SHIM(VerifyArrayNewMemory, int, (const void* p), (p)); -C_SHIM(VerifyMallocMemory, int, (const void* p), (p)); -C_SHIM(MallocMemoryStats, int, - (int* blocks, size_t* total, int histogram[kMallocHistogramSize]), - (blocks, total, histogram)); - -C_SHIM(GetStats, void, - (char* buffer, int buffer_length), (buffer, buffer_length)); -C_SHIM(GetNumericProperty, int, - (const char* property, size_t* value), (property, value)); -C_SHIM(SetNumericProperty, int, - (const char* property, size_t value), (property, value)); - -C_SHIM(MarkThreadIdle, void, (void), ()); -C_SHIM(MarkThreadBusy, void, (void), ()); -C_SHIM(ReleaseFreeMemory, void, (void), ()); -C_SHIM(ReleaseToSystem, void, (size_t num_bytes), (num_bytes)); -C_SHIM(GetEstimatedAllocatedSize, size_t, (size_t size), (size)); -C_SHIM(GetAllocatedSize, size_t, (const void* p), (p)); -C_SHIM(GetThreadCacheSize, size_t, (void), ()); -C_SHIM(MarkThreadTemporarilyIdle, void, (void), ()); - -// Can't use the shim here because of the need to translate the enums. -extern "C" -MallocExtension_Ownership MallocExtension_GetOwnership(const void* p) { - return static_cast( - MallocExtension::instance()->GetOwnership(p)); -} diff --git a/contrib/libtcmalloc/src/malloc_hook-inl.h b/contrib/libtcmalloc/src/malloc_hook-inl.h deleted file mode 100644 index dbf4d46ed47..00000000000 --- a/contrib/libtcmalloc/src/malloc_hook-inl.h +++ /dev/null @@ -1,249 +0,0 @@ -// -*- Mode: C++; c-basic-offset: 2; indent-tabs-mode: nil -*- -// Copyright (c) 2005, Google Inc. -// All rights reserved. -// -// Redistribution and use in source and binary forms, with or without -// modification, are permitted provided that the following conditions are -// met: -// -// * Redistributions of source code must retain the above copyright -// notice, this list of conditions and the following disclaimer. -// * Redistributions in binary form must reproduce the above -// copyright notice, this list of conditions and the following disclaimer -// in the documentation and/or other materials provided with the -// distribution. -// * Neither the name of Google Inc. nor the names of its -// contributors may be used to endorse or promote products derived from -// this software without specific prior written permission. -// -// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS -// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT -// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR -// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT -// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, -// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT -// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, -// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY -// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT -// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE -// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. - -// --- -// Author: Sanjay Ghemawat -// -// This has the implementation details of malloc_hook that are needed -// to use malloc-hook inside the tcmalloc system. It does not hold -// any of the client-facing calls that are used to add new hooks. - -#ifndef _MALLOC_HOOK_INL_H_ -#define _MALLOC_HOOK_INL_H_ - -#include -#include -#include "base/atomicops.h" -#include "base/basictypes.h" -#include - -#include "common.h" // for UNLIKELY - -namespace base { namespace internal { - -// Capacity of 8 means that HookList is 9 words. -static const int kHookListCapacity = 8; -// last entry is reserved for deprecated "singular" hooks. So we have -// 7 "normal" hooks per list -static const int kHookListMaxValues = 7; -static const int kHookListSingularIdx = 7; - -// HookList: a class that provides synchronized insertions and removals and -// lockless traversal. Most of the implementation is in malloc_hook.cc. -template -struct PERFTOOLS_DLL_DECL HookList { - COMPILE_ASSERT(sizeof(T) <= sizeof(AtomicWord), T_should_fit_in_AtomicWord); - - // Adds value to the list. Note that duplicates are allowed. Thread-safe and - // blocking (acquires hooklist_spinlock). Returns true on success; false - // otherwise (failures include invalid value and no space left). - bool Add(T value); - - void FixupPrivEndLocked(); - - // Removes the first entry matching value from the list. Thread-safe and - // blocking (acquires hooklist_spinlock). Returns true on success; false - // otherwise (failures include invalid value and no value found). - bool Remove(T value); - - // Store up to n values of the list in output_array, and return the number of - // elements stored. Thread-safe and non-blocking. This is fast (one memory - // access) if the list is empty. - int Traverse(T* output_array, int n) const; - - // Fast inline implementation for fast path of Invoke*Hook. - bool empty() const { - return base::subtle::NoBarrier_Load(&priv_end) == 0; - } - - // Used purely to handle deprecated singular hooks - T GetSingular() const { - const AtomicWord *place = &priv_data[kHookListSingularIdx]; - return bit_cast(base::subtle::NoBarrier_Load(place)); - } - - T ExchangeSingular(T new_val); - - // This internal data is not private so that the class is an aggregate and can - // be initialized by the linker. Don't access this directly. Use the - // INIT_HOOK_LIST macro in malloc_hook.cc. - - // One more than the index of the last valid element in priv_data. During - // 'Remove' this may be past the last valid element in priv_data, but - // subsequent values will be 0. - // - // Index kHookListCapacity-1 is reserved as 'deprecated' single hook pointer - AtomicWord priv_end; - AtomicWord priv_data[kHookListCapacity]; -}; - -ATTRIBUTE_VISIBILITY_HIDDEN extern HookList new_hooks_; -ATTRIBUTE_VISIBILITY_HIDDEN extern HookList delete_hooks_; -ATTRIBUTE_VISIBILITY_HIDDEN extern HookList premmap_hooks_; -ATTRIBUTE_VISIBILITY_HIDDEN extern HookList mmap_hooks_; -ATTRIBUTE_VISIBILITY_HIDDEN extern HookList mmap_replacement_; -ATTRIBUTE_VISIBILITY_HIDDEN extern HookList munmap_hooks_; -ATTRIBUTE_VISIBILITY_HIDDEN extern HookList munmap_replacement_; -ATTRIBUTE_VISIBILITY_HIDDEN extern HookList mremap_hooks_; -ATTRIBUTE_VISIBILITY_HIDDEN extern HookList presbrk_hooks_; -ATTRIBUTE_VISIBILITY_HIDDEN extern HookList sbrk_hooks_; - -} } // namespace base::internal - -// The following method is DEPRECATED -inline MallocHook::NewHook MallocHook::GetNewHook() { - return base::internal::new_hooks_.GetSingular(); -} - -inline void MallocHook::InvokeNewHook(const void* p, size_t s) { - if (UNLIKELY(!base::internal::new_hooks_.empty())) { - InvokeNewHookSlow(p, s); - } -} - -// The following method is DEPRECATED -inline MallocHook::DeleteHook MallocHook::GetDeleteHook() { - return base::internal::delete_hooks_.GetSingular(); -} - -inline void MallocHook::InvokeDeleteHook(const void* p) { - if (UNLIKELY(!base::internal::delete_hooks_.empty())) { - InvokeDeleteHookSlow(p); - } -} - -// The following method is DEPRECATED -inline MallocHook::PreMmapHook MallocHook::GetPreMmapHook() { - return base::internal::premmap_hooks_.GetSingular(); -} - -inline void MallocHook::InvokePreMmapHook(const void* start, - size_t size, - int protection, - int flags, - int fd, - off_t offset) { - if (!base::internal::premmap_hooks_.empty()) { - InvokePreMmapHookSlow(start, size, protection, flags, fd, offset); - } -} - -// The following method is DEPRECATED -inline MallocHook::MmapHook MallocHook::GetMmapHook() { - return base::internal::mmap_hooks_.GetSingular(); -} - -inline void MallocHook::InvokeMmapHook(const void* result, - const void* start, - size_t size, - int protection, - int flags, - int fd, - off_t offset) { - if (!base::internal::mmap_hooks_.empty()) { - InvokeMmapHookSlow(result, start, size, protection, flags, fd, offset); - } -} - -inline bool MallocHook::InvokeMmapReplacement(const void* start, - size_t size, - int protection, - int flags, - int fd, - off_t offset, - void** result) { - if (!base::internal::mmap_replacement_.empty()) { - return InvokeMmapReplacementSlow(start, size, - protection, flags, - fd, offset, - result); - } - return false; -} - -// The following method is DEPRECATED -inline MallocHook::MunmapHook MallocHook::GetMunmapHook() { - return base::internal::munmap_hooks_.GetSingular(); -} - -inline void MallocHook::InvokeMunmapHook(const void* p, size_t size) { - if (!base::internal::munmap_hooks_.empty()) { - InvokeMunmapHookSlow(p, size); - } -} - -inline bool MallocHook::InvokeMunmapReplacement( - const void* p, size_t size, int* result) { - if (!base::internal::mmap_replacement_.empty()) { - return InvokeMunmapReplacementSlow(p, size, result); - } - return false; -} - -// The following method is DEPRECATED -inline MallocHook::MremapHook MallocHook::GetMremapHook() { - return base::internal::mremap_hooks_.GetSingular(); -} - -inline void MallocHook::InvokeMremapHook(const void* result, - const void* old_addr, - size_t old_size, - size_t new_size, - int flags, - const void* new_addr) { - if (!base::internal::mremap_hooks_.empty()) { - InvokeMremapHookSlow(result, old_addr, old_size, new_size, flags, new_addr); - } -} - -// The following method is DEPRECATED -inline MallocHook::PreSbrkHook MallocHook::GetPreSbrkHook() { - return base::internal::presbrk_hooks_.GetSingular(); -} - -inline void MallocHook::InvokePreSbrkHook(ptrdiff_t increment) { - if (!base::internal::presbrk_hooks_.empty() && increment != 0) { - InvokePreSbrkHookSlow(increment); - } -} - -// The following method is DEPRECATED -inline MallocHook::SbrkHook MallocHook::GetSbrkHook() { - return base::internal::sbrk_hooks_.GetSingular(); -} - -inline void MallocHook::InvokeSbrkHook(const void* result, - ptrdiff_t increment) { - if (!base::internal::sbrk_hooks_.empty() && increment != 0) { - InvokeSbrkHookSlow(result, increment); - } -} - -#endif /* _MALLOC_HOOK_INL_H_ */ diff --git a/contrib/libtcmalloc/src/malloc_hook.cc b/contrib/libtcmalloc/src/malloc_hook.cc deleted file mode 100644 index f87da8abbeb..00000000000 --- a/contrib/libtcmalloc/src/malloc_hook.cc +++ /dev/null @@ -1,703 +0,0 @@ -// -*- Mode: C++; c-basic-offset: 2; indent-tabs-mode: nil -*- -// Copyright (c) 2005, Google Inc. -// All rights reserved. -// -// Redistribution and use in source and binary forms, with or without -// modification, are permitted provided that the following conditions are -// met: -// -// * Redistributions of source code must retain the above copyright -// notice, this list of conditions and the following disclaimer. -// * Redistributions in binary form must reproduce the above -// copyright notice, this list of conditions and the following disclaimer -// in the documentation and/or other materials provided with the -// distribution. -// * Neither the name of Google Inc. nor the names of its -// contributors may be used to endorse or promote products derived from -// this software without specific prior written permission. -// -// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS -// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT -// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR -// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT -// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, -// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT -// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, -// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY -// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT -// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE -// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. - -// --- -// Author: Sanjay Ghemawat - -#include "config.h" - -// Disable the glibc prototype of mremap(), as older versions of the -// system headers define this function with only four arguments, -// whereas newer versions allow an optional fifth argument: -#ifdef HAVE_MMAP -# define mremap glibc_mremap -# include -# undef mremap -#endif - -#include -#ifdef HAVE_STDINT_H -#include -#endif -#include -#include "base/logging.h" -#include "base/spinlock.h" -#include "maybe_emergency_malloc.h" -#include "maybe_threads.h" -#include "malloc_hook-inl.h" -#include - -// This #ifdef should almost never be set. Set NO_TCMALLOC_SAMPLES if -// you're porting to a system where you really can't get a stacktrace. -#ifdef NO_TCMALLOC_SAMPLES - // We use #define so code compiles even if you #include stacktrace.h somehow. -# define GetStackTrace(stack, depth, skip) (0) -#else -# include -#endif - -// __THROW is defined in glibc systems. It means, counter-intuitively, -// "This function will never throw an exception." It's an optional -// optimization tool, but we may need to use it to match glibc prototypes. -#ifndef __THROW // I guess we're not on a glibc system -# define __THROW // __THROW is just an optimization, so ok to make it "" -#endif - -using std::copy; - - -// Declaration of default weak initialization function, that can be overridden -// by linking-in a strong definition (as heap-checker.cc does). This is -// extern "C" so that it doesn't trigger gold's --detect-odr-violations warning, -// which only looks at C++ symbols. -// -// This function is declared here as weak, and defined later, rather than a more -// straightforward simple weak definition, as a workround for an icc compiler -// issue ((Intel reference 290819). This issue causes icc to resolve weak -// symbols too early, at compile rather than link time. By declaring it (weak) -// here, then defining it below after its use, we can avoid the problem. -extern "C" { -ATTRIBUTE_WEAK void MallocHook_InitAtFirstAllocation_HeapLeakChecker(); -} - -namespace { - -void RemoveInitialHooksAndCallInitializers(); // below. - -pthread_once_t once = PTHREAD_ONCE_INIT; - -// These hooks are installed in MallocHook as the only initial hooks. The first -// hook that is called will run RemoveInitialHooksAndCallInitializers (see the -// definition below) and then redispatch to any malloc hooks installed by -// RemoveInitialHooksAndCallInitializers. -// -// Note(llib): there is a possibility of a race in the event that there are -// multiple threads running before the first allocation. This is pretty -// difficult to achieve, but if it is then multiple threads may concurrently do -// allocations. The first caller will call -// RemoveInitialHooksAndCallInitializers via one of the initial hooks. A -// concurrent allocation may, depending on timing either: -// * still have its initial malloc hook installed, run that and block on waiting -// for the first caller to finish its call to -// RemoveInitialHooksAndCallInitializers, and proceed normally. -// * occur some time during the RemoveInitialHooksAndCallInitializers call, at -// which point there could be no initial hooks and the subsequent hooks that -// are about to be set up by RemoveInitialHooksAndCallInitializers haven't -// been installed yet. I think the worst we can get is that some allocations -// will not get reported to some hooks set by the initializers called from -// RemoveInitialHooksAndCallInitializers. - -void InitialNewHook(const void* ptr, size_t size) { - perftools_pthread_once(&once, &RemoveInitialHooksAndCallInitializers); - MallocHook::InvokeNewHook(ptr, size); -} - -void InitialPreMMapHook(const void* start, - size_t size, - int protection, - int flags, - int fd, - off_t offset) { - perftools_pthread_once(&once, &RemoveInitialHooksAndCallInitializers); - MallocHook::InvokePreMmapHook(start, size, protection, flags, fd, offset); -} - -void InitialPreSbrkHook(ptrdiff_t increment) { - perftools_pthread_once(&once, &RemoveInitialHooksAndCallInitializers); - MallocHook::InvokePreSbrkHook(increment); -} - -// This function is called at most once by one of the above initial malloc -// hooks. It removes all initial hooks and initializes all other clients that -// want to get control at the very first memory allocation. The initializers -// may assume that the initial malloc hooks have been removed. The initializers -// may set up malloc hooks and allocate memory. -void RemoveInitialHooksAndCallInitializers() { - RAW_CHECK(MallocHook::RemoveNewHook(&InitialNewHook), ""); - RAW_CHECK(MallocHook::RemovePreMmapHook(&InitialPreMMapHook), ""); - RAW_CHECK(MallocHook::RemovePreSbrkHook(&InitialPreSbrkHook), ""); - - // HeapLeakChecker is currently the only module that needs to get control on - // the first memory allocation, but one can add other modules by following the - // same weak/strong function pattern. - MallocHook_InitAtFirstAllocation_HeapLeakChecker(); -} - -} // namespace - -// Weak default initialization function that must go after its use. -extern "C" void MallocHook_InitAtFirstAllocation_HeapLeakChecker() { - // Do nothing. -} - -namespace base { namespace internal { - -// This lock is shared between all implementations of HookList::Add & Remove. -// The potential for contention is very small. This needs to be a SpinLock and -// not a Mutex since it's possible for Mutex locking to allocate memory (e.g., -// per-thread allocation in debug builds), which could cause infinite recursion. -static SpinLock hooklist_spinlock(base::LINKER_INITIALIZED); - -template -bool HookList::Add(T value_as_t) { - AtomicWord value = bit_cast(value_as_t); - if (value == 0) { - return false; - } - SpinLockHolder l(&hooklist_spinlock); - // Find the first slot in data that is 0. - int index = 0; - while ((index < kHookListMaxValues) && - (base::subtle::NoBarrier_Load(&priv_data[index]) != 0)) { - ++index; - } - if (index == kHookListMaxValues) { - return false; - } - AtomicWord prev_num_hooks = base::subtle::Acquire_Load(&priv_end); - base::subtle::NoBarrier_Store(&priv_data[index], value); - if (prev_num_hooks <= index) { - base::subtle::NoBarrier_Store(&priv_end, index + 1); - } - return true; -} - -template -void HookList::FixupPrivEndLocked() { - AtomicWord hooks_end = base::subtle::NoBarrier_Load(&priv_end); - while ((hooks_end > 0) && - (base::subtle::NoBarrier_Load(&priv_data[hooks_end - 1]) == 0)) { - --hooks_end; - } - base::subtle::NoBarrier_Store(&priv_end, hooks_end); -} - -template -bool HookList::Remove(T value_as_t) { - if (value_as_t == 0) { - return false; - } - SpinLockHolder l(&hooklist_spinlock); - AtomicWord hooks_end = base::subtle::NoBarrier_Load(&priv_end); - int index = 0; - while (index < hooks_end && value_as_t != bit_cast( - base::subtle::NoBarrier_Load(&priv_data[index]))) { - ++index; - } - if (index == hooks_end) { - return false; - } - base::subtle::NoBarrier_Store(&priv_data[index], 0); - FixupPrivEndLocked(); - return true; -} - -template -int HookList::Traverse(T* output_array, int n) const { - AtomicWord hooks_end = base::subtle::Acquire_Load(&priv_end); - int actual_hooks_end = 0; - for (int i = 0; i < hooks_end && n > 0; ++i) { - AtomicWord data = base::subtle::Acquire_Load(&priv_data[i]); - if (data != 0) { - *output_array++ = bit_cast(data); - ++actual_hooks_end; - --n; - } - } - return actual_hooks_end; -} - -template -T HookList::ExchangeSingular(T value_as_t) { - AtomicWord value = bit_cast(value_as_t); - AtomicWord old_value; - SpinLockHolder l(&hooklist_spinlock); - old_value = base::subtle::NoBarrier_Load(&priv_data[kHookListSingularIdx]); - base::subtle::NoBarrier_Store(&priv_data[kHookListSingularIdx], value); - if (value != 0) { - base::subtle::NoBarrier_Store(&priv_end, kHookListSingularIdx + 1); - } else { - FixupPrivEndLocked(); - } - return bit_cast(old_value); -} - -// Initialize a HookList (optionally with the given initial_value in index 0). -#define INIT_HOOK_LIST { 0 } -#define INIT_HOOK_LIST_WITH_VALUE(initial_value) \ - { 1, { reinterpret_cast(initial_value) } } - -// Explicit instantiation for malloc_hook_test.cc. This ensures all the methods -// are instantiated. -template struct HookList; - -HookList new_hooks_ = - INIT_HOOK_LIST_WITH_VALUE(&InitialNewHook); -HookList delete_hooks_ = INIT_HOOK_LIST; -HookList premmap_hooks_ = - INIT_HOOK_LIST_WITH_VALUE(&InitialPreMMapHook); -HookList mmap_hooks_ = INIT_HOOK_LIST; -HookList munmap_hooks_ = INIT_HOOK_LIST; -HookList mremap_hooks_ = INIT_HOOK_LIST; -HookList presbrk_hooks_ = - INIT_HOOK_LIST_WITH_VALUE(InitialPreSbrkHook); -HookList sbrk_hooks_ = INIT_HOOK_LIST; - -// These lists contain either 0 or 1 hooks. -HookList mmap_replacement_ = { 0 }; -HookList munmap_replacement_ = { 0 }; - -#undef INIT_HOOK_LIST_WITH_VALUE -#undef INIT_HOOK_LIST - -} } // namespace base::internal - -using base::internal::kHookListMaxValues; -using base::internal::new_hooks_; -using base::internal::delete_hooks_; -using base::internal::premmap_hooks_; -using base::internal::mmap_hooks_; -using base::internal::mmap_replacement_; -using base::internal::munmap_hooks_; -using base::internal::munmap_replacement_; -using base::internal::mremap_hooks_; -using base::internal::presbrk_hooks_; -using base::internal::sbrk_hooks_; - -// These are available as C bindings as well as C++, hence their -// definition outside the MallocHook class. -extern "C" -int MallocHook_AddNewHook(MallocHook_NewHook hook) { - RAW_VLOG(10, "AddNewHook(%p)", hook); - return new_hooks_.Add(hook); -} - -extern "C" -int MallocHook_RemoveNewHook(MallocHook_NewHook hook) { - RAW_VLOG(10, "RemoveNewHook(%p)", hook); - return new_hooks_.Remove(hook); -} - -extern "C" -int MallocHook_AddDeleteHook(MallocHook_DeleteHook hook) { - RAW_VLOG(10, "AddDeleteHook(%p)", hook); - return delete_hooks_.Add(hook); -} - -extern "C" -int MallocHook_RemoveDeleteHook(MallocHook_DeleteHook hook) { - RAW_VLOG(10, "RemoveDeleteHook(%p)", hook); - return delete_hooks_.Remove(hook); -} - -extern "C" -int MallocHook_AddPreMmapHook(MallocHook_PreMmapHook hook) { - RAW_VLOG(10, "AddPreMmapHook(%p)", hook); - return premmap_hooks_.Add(hook); -} - -extern "C" -int MallocHook_RemovePreMmapHook(MallocHook_PreMmapHook hook) { - RAW_VLOG(10, "RemovePreMmapHook(%p)", hook); - return premmap_hooks_.Remove(hook); -} - -extern "C" -int MallocHook_SetMmapReplacement(MallocHook_MmapReplacement hook) { - RAW_VLOG(10, "SetMmapReplacement(%p)", hook); - // NOTE this is a best effort CHECK. Concurrent sets could succeed since - // this test is outside of the Add spin lock. - RAW_CHECK(mmap_replacement_.empty(), "Only one MMapReplacement is allowed."); - return mmap_replacement_.Add(hook); -} - -extern "C" -int MallocHook_RemoveMmapReplacement(MallocHook_MmapReplacement hook) { - RAW_VLOG(10, "RemoveMmapReplacement(%p)", hook); - return mmap_replacement_.Remove(hook); -} - -extern "C" -int MallocHook_AddMmapHook(MallocHook_MmapHook hook) { - RAW_VLOG(10, "AddMmapHook(%p)", hook); - return mmap_hooks_.Add(hook); -} - -extern "C" -int MallocHook_RemoveMmapHook(MallocHook_MmapHook hook) { - RAW_VLOG(10, "RemoveMmapHook(%p)", hook); - return mmap_hooks_.Remove(hook); -} - -extern "C" -int MallocHook_AddMunmapHook(MallocHook_MunmapHook hook) { - RAW_VLOG(10, "AddMunmapHook(%p)", hook); - return munmap_hooks_.Add(hook); -} - -extern "C" -int MallocHook_RemoveMunmapHook(MallocHook_MunmapHook hook) { - RAW_VLOG(10, "RemoveMunmapHook(%p)", hook); - return munmap_hooks_.Remove(hook); -} - -extern "C" -int MallocHook_SetMunmapReplacement(MallocHook_MunmapReplacement hook) { - RAW_VLOG(10, "SetMunmapReplacement(%p)", hook); - // NOTE this is a best effort CHECK. Concurrent sets could succeed since - // this test is outside of the Add spin lock. - RAW_CHECK(munmap_replacement_.empty(), - "Only one MunmapReplacement is allowed."); - return munmap_replacement_.Add(hook); -} - -extern "C" -int MallocHook_RemoveMunmapReplacement(MallocHook_MunmapReplacement hook) { - RAW_VLOG(10, "RemoveMunmapReplacement(%p)", hook); - return munmap_replacement_.Remove(hook); -} - -extern "C" -int MallocHook_AddMremapHook(MallocHook_MremapHook hook) { - RAW_VLOG(10, "AddMremapHook(%p)", hook); - return mremap_hooks_.Add(hook); -} - -extern "C" -int MallocHook_RemoveMremapHook(MallocHook_MremapHook hook) { - RAW_VLOG(10, "RemoveMremapHook(%p)", hook); - return mremap_hooks_.Remove(hook); -} - -extern "C" -int MallocHook_AddPreSbrkHook(MallocHook_PreSbrkHook hook) { - RAW_VLOG(10, "AddPreSbrkHook(%p)", hook); - return presbrk_hooks_.Add(hook); -} - -extern "C" -int MallocHook_RemovePreSbrkHook(MallocHook_PreSbrkHook hook) { - RAW_VLOG(10, "RemovePreSbrkHook(%p)", hook); - return presbrk_hooks_.Remove(hook); -} - -extern "C" -int MallocHook_AddSbrkHook(MallocHook_SbrkHook hook) { - RAW_VLOG(10, "AddSbrkHook(%p)", hook); - return sbrk_hooks_.Add(hook); -} - -extern "C" -int MallocHook_RemoveSbrkHook(MallocHook_SbrkHook hook) { - RAW_VLOG(10, "RemoveSbrkHook(%p)", hook); - return sbrk_hooks_.Remove(hook); -} - -// The code below is DEPRECATED. -extern "C" -MallocHook_NewHook MallocHook_SetNewHook(MallocHook_NewHook hook) { - RAW_VLOG(10, "SetNewHook(%p)", hook); - return new_hooks_.ExchangeSingular(hook); -} - -extern "C" -MallocHook_DeleteHook MallocHook_SetDeleteHook(MallocHook_DeleteHook hook) { - RAW_VLOG(10, "SetDeleteHook(%p)", hook); - return delete_hooks_.ExchangeSingular(hook); -} - -extern "C" -MallocHook_PreMmapHook MallocHook_SetPreMmapHook(MallocHook_PreMmapHook hook) { - RAW_VLOG(10, "SetPreMmapHook(%p)", hook); - return premmap_hooks_.ExchangeSingular(hook); -} - -extern "C" -MallocHook_MmapHook MallocHook_SetMmapHook(MallocHook_MmapHook hook) { - RAW_VLOG(10, "SetMmapHook(%p)", hook); - return mmap_hooks_.ExchangeSingular(hook); -} - -extern "C" -MallocHook_MunmapHook MallocHook_SetMunmapHook(MallocHook_MunmapHook hook) { - RAW_VLOG(10, "SetMunmapHook(%p)", hook); - return munmap_hooks_.ExchangeSingular(hook); -} - -extern "C" -MallocHook_MremapHook MallocHook_SetMremapHook(MallocHook_MremapHook hook) { - RAW_VLOG(10, "SetMremapHook(%p)", hook); - return mremap_hooks_.ExchangeSingular(hook); -} - -extern "C" -MallocHook_PreSbrkHook MallocHook_SetPreSbrkHook(MallocHook_PreSbrkHook hook) { - RAW_VLOG(10, "SetPreSbrkHook(%p)", hook); - return presbrk_hooks_.ExchangeSingular(hook); -} - -extern "C" -MallocHook_SbrkHook MallocHook_SetSbrkHook(MallocHook_SbrkHook hook) { - RAW_VLOG(10, "SetSbrkHook(%p)", hook); - return sbrk_hooks_.ExchangeSingular(hook); -} -// End of DEPRECATED code section. - -// Note: embedding the function calls inside the traversal of HookList would be -// very confusing, as it is legal for a hook to remove itself and add other -// hooks. Doing traversal first, and then calling the hooks ensures we only -// call the hooks registered at the start. -#define INVOKE_HOOKS(HookType, hook_list, args) do { \ - HookType hooks[kHookListMaxValues]; \ - int num_hooks = hook_list.Traverse(hooks, kHookListMaxValues); \ - for (int i = 0; i < num_hooks; ++i) { \ - (*hooks[i])args; \ - } \ - } while (0) - -// There should only be one replacement. Return the result of the first -// one, or false if there is none. -#define INVOKE_REPLACEMENT(HookType, hook_list, args) do { \ - HookType hooks[kHookListMaxValues]; \ - int num_hooks = hook_list.Traverse(hooks, kHookListMaxValues); \ - return (num_hooks > 0 && (*hooks[0])args); \ - } while (0) - - -void MallocHook::InvokeNewHookSlow(const void* p, size_t s) { - if (tcmalloc::IsEmergencyPtr(p)) { - return; - } - INVOKE_HOOKS(NewHook, new_hooks_, (p, s)); -} - -void MallocHook::InvokeDeleteHookSlow(const void* p) { - if (tcmalloc::IsEmergencyPtr(p)) { - return; - } - INVOKE_HOOKS(DeleteHook, delete_hooks_, (p)); -} - -void MallocHook::InvokePreMmapHookSlow(const void* start, - size_t size, - int protection, - int flags, - int fd, - off_t offset) { - INVOKE_HOOKS(PreMmapHook, premmap_hooks_, (start, size, protection, flags, fd, - offset)); -} - -void MallocHook::InvokeMmapHookSlow(const void* result, - const void* start, - size_t size, - int protection, - int flags, - int fd, - off_t offset) { - INVOKE_HOOKS(MmapHook, mmap_hooks_, (result, start, size, protection, flags, - fd, offset)); -} - -bool MallocHook::InvokeMmapReplacementSlow(const void* start, - size_t size, - int protection, - int flags, - int fd, - off_t offset, - void** result) { - INVOKE_REPLACEMENT(MmapReplacement, mmap_replacement_, - (start, size, protection, flags, fd, offset, result)); -} - -void MallocHook::InvokeMunmapHookSlow(const void* p, size_t s) { - INVOKE_HOOKS(MunmapHook, munmap_hooks_, (p, s)); -} - -bool MallocHook::InvokeMunmapReplacementSlow(const void* p, - size_t s, - int* result) { - INVOKE_REPLACEMENT(MunmapReplacement, munmap_replacement_, (p, s, result)); -} - -void MallocHook::InvokeMremapHookSlow(const void* result, - const void* old_addr, - size_t old_size, - size_t new_size, - int flags, - const void* new_addr) { - INVOKE_HOOKS(MremapHook, mremap_hooks_, (result, old_addr, old_size, new_size, - flags, new_addr)); -} - -void MallocHook::InvokePreSbrkHookSlow(ptrdiff_t increment) { - INVOKE_HOOKS(PreSbrkHook, presbrk_hooks_, (increment)); -} - -void MallocHook::InvokeSbrkHookSlow(const void* result, ptrdiff_t increment) { - INVOKE_HOOKS(SbrkHook, sbrk_hooks_, (result, increment)); -} - -#undef INVOKE_HOOKS - -#ifndef NO_TCMALLOC_SAMPLES - -DEFINE_ATTRIBUTE_SECTION_VARS(google_malloc); -DECLARE_ATTRIBUTE_SECTION_VARS(google_malloc); - // actual functions are in debugallocation.cc or tcmalloc.cc -DEFINE_ATTRIBUTE_SECTION_VARS(malloc_hook); -DECLARE_ATTRIBUTE_SECTION_VARS(malloc_hook); - // actual functions are in this file, malloc_hook.cc, and low_level_alloc.cc - -#define ADDR_IN_ATTRIBUTE_SECTION(addr, name) \ - (reinterpret_cast(ATTRIBUTE_SECTION_START(name)) <= \ - reinterpret_cast(addr) && \ - reinterpret_cast(addr) < \ - reinterpret_cast(ATTRIBUTE_SECTION_STOP(name))) - -// Return true iff 'caller' is a return address within a function -// that calls one of our hooks via MallocHook:Invoke*. -// A helper for GetCallerStackTrace. -static inline bool InHookCaller(const void* caller) { - return ADDR_IN_ATTRIBUTE_SECTION(caller, google_malloc) || - ADDR_IN_ATTRIBUTE_SECTION(caller, malloc_hook); - // We can use one section for everything except tcmalloc_or_debug - // due to its special linkage mode, which prevents merging of the sections. -} - -#undef ADDR_IN_ATTRIBUTE_SECTION - -static bool checked_sections = false; - -static inline void CheckInHookCaller() { - if (!checked_sections) { - INIT_ATTRIBUTE_SECTION_VARS(google_malloc); - if (ATTRIBUTE_SECTION_START(google_malloc) == - ATTRIBUTE_SECTION_STOP(google_malloc)) { - RAW_LOG(ERROR, "google_malloc section is missing, " - "thus InHookCaller is broken!"); - } - INIT_ATTRIBUTE_SECTION_VARS(malloc_hook); - if (ATTRIBUTE_SECTION_START(malloc_hook) == - ATTRIBUTE_SECTION_STOP(malloc_hook)) { - RAW_LOG(ERROR, "malloc_hook section is missing, " - "thus InHookCaller is broken!"); - } - checked_sections = true; - } -} - -#endif // !NO_TCMALLOC_SAMPLES - -// We can improve behavior/compactness of this function -// if we pass a generic test function (with a generic arg) -// into the implementations for GetStackTrace instead of the skip_count. -extern "C" int MallocHook_GetCallerStackTrace(void** result, int max_depth, - int skip_count) { -#if defined(NO_TCMALLOC_SAMPLES) - return 0; -#elif !defined(HAVE_ATTRIBUTE_SECTION_START) - // Fall back to GetStackTrace and good old but fragile frame skip counts. - // Note: this path is inaccurate when a hook is not called directly by an - // allocation function but is daisy-chained through another hook, - // search for MallocHook::(Get|Set|Invoke)* to find such cases. - return GetStackTrace(result, max_depth, skip_count + int(DEBUG_MODE)); - // due to -foptimize-sibling-calls in opt mode - // there's no need for extra frame skip here then -#else - CheckInHookCaller(); - // MallocHook caller determination via InHookCaller works, use it: - static const int kMaxSkip = 32 + 6 + 3; - // Constant tuned to do just one GetStackTrace call below in practice - // and not get many frames that we don't actually need: - // currently max passsed max_depth is 32, - // max passed/needed skip_count is 6 - // and 3 is to account for some hook daisy chaining. - static const int kStackSize = kMaxSkip + 1; - void* stack[kStackSize]; - int depth = GetStackTrace(stack, kStackSize, 1); // skip this function frame - if (depth == 0) // silenty propagate cases when GetStackTrace does not work - return 0; - for (int i = 0; i < depth; ++i) { // stack[0] is our immediate caller - if (InHookCaller(stack[i])) { - RAW_VLOG(10, "Found hooked allocator at %d: %p <- %p", - i, stack[i], stack[i+1]); - i += 1; // skip hook caller frame - depth -= i; // correct depth - if (depth > max_depth) depth = max_depth; - copy(stack + i, stack + i + depth, result); - if (depth < max_depth && depth + i == kStackSize) { - // get frames for the missing depth - depth += - GetStackTrace(result + depth, max_depth - depth, 1 + kStackSize); - } - return depth; - } - } - RAW_LOG(WARNING, "Hooked allocator frame not found, returning empty trace"); - // If this happens try increasing kMaxSkip - // or else something must be wrong with InHookCaller, - // e.g. for every section used in InHookCaller - // all functions in that section must be inside the same library. - return 0; -#endif -} - -// On systems where we know how, we override mmap/munmap/mremap/sbrk -// to provide support for calling the related hooks (in addition, -// of course, to doing what these functions normally do). - -#if defined(__linux) -# include "malloc_hook_mmap_linux.h" - -#elif defined(__FreeBSD__) -# include "malloc_hook_mmap_freebsd.h" - -#else - -/*static*/void* MallocHook::UnhookedMMap(void *start, size_t length, int prot, - int flags, int fd, off_t offset) { - void* result; - if (!MallocHook::InvokeMmapReplacement( - start, length, prot, flags, fd, offset, &result)) { - result = mmap(start, length, prot, flags, fd, offset); - } - return result; -} - -/*static*/int MallocHook::UnhookedMUnmap(void *start, size_t length) { - int result; - if (!MallocHook::InvokeMunmapReplacement(start, length, &result)) { - result = munmap(start, length); - } - return result; -} - -#endif diff --git a/contrib/libtcmalloc/src/malloc_hook_mmap_freebsd.h b/contrib/libtcmalloc/src/malloc_hook_mmap_freebsd.h deleted file mode 100644 index 8575dcc7c08..00000000000 --- a/contrib/libtcmalloc/src/malloc_hook_mmap_freebsd.h +++ /dev/null @@ -1,135 +0,0 @@ -// -*- Mode: C++; c-basic-offset: 2; indent-tabs-mode: nil -*- -// Copyright (c) 2011, Google Inc. -// All rights reserved. -// -// Redistribution and use in source and binary forms, with or without -// modification, are permitted provided that the following conditions are -// met: -// -// * Redistributions of source code must retain the above copyright -// notice, this list of conditions and the following disclaimer. -// * Redistributions in binary form must reproduce the above -// copyright notice, this list of conditions and the following disclaimer -// in the documentation and/or other materials provided with the -// distribution. -// * Neither the name of Google Inc. nor the names of its -// contributors may be used to endorse or promote products derived from -// this software without specific prior written permission. -// -// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS -// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT -// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR -// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT -// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, -// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT -// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, -// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY -// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT -// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE -// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. - -// Override mmap/munmap/mremap/sbrk to provide support for calling the -// related hooks (in addition, of course, to doing what these -// functions normally do). - -#ifndef __FreeBSD__ -# error Should only be including malloc_hook_mmap_freebsd.h on FreeBSD systems. -#endif - -#include -#include -#include -#include -#include - -// Make sure mmap doesn't get #define'd away by -#undef mmap - -// According to the FreeBSD documentation, use syscall if you do not -// need 64-bit alignment otherwise use __syscall. Indeed, syscall -// doesn't work correctly in most situations on 64-bit. It's return -// type is 'int' so for things like SYS_mmap, it actually truncates -// the returned address to 32-bits. -#if defined(__amd64__) || defined(__x86_64__) -# define MALLOC_HOOK_SYSCALL __syscall -#else -# define MALLOC_HOOK_SYSCALL syscall -#endif - - -extern "C" { - void* mmap(void *start, size_t length,int prot, int flags, - int fd, off_t offset) __THROW - ATTRIBUTE_SECTION(malloc_hook); - int munmap(void* start, size_t length) __THROW - ATTRIBUTE_SECTION(malloc_hook); - void* sbrk(intptr_t increment) __THROW - ATTRIBUTE_SECTION(malloc_hook); -} - -static inline void* do_mmap(void *start, size_t length, - int prot, int flags, - int fd, off_t offset) __THROW { - return (void *)MALLOC_HOOK_SYSCALL(SYS_mmap, - start, length, prot, flags, fd, offset); -} - -static inline void* do_sbrk(intptr_t increment) { - static void *(*libc_sbrk)(intptr_t); - if (libc_sbrk == NULL) - libc_sbrk = (void *(*)(intptr_t))dlsym(RTLD_NEXT, "sbrk"); - - return libc_sbrk(increment); -} - - -extern "C" void* mmap(void *start, size_t length, int prot, int flags, - int fd, off_t offset) __THROW { - MallocHook::InvokePreMmapHook(start, length, prot, flags, fd, offset); - void *result; - if (!MallocHook::InvokeMmapReplacement( - start, length, prot, flags, fd, offset, &result)) { - result = do_mmap(start, length, prot, flags, fd, - static_cast(offset)); // avoid sign extension - } - MallocHook::InvokeMmapHook(result, start, length, prot, flags, fd, offset); - return result; -} - -extern "C" int munmap(void* start, size_t length) __THROW { - MallocHook::InvokeMunmapHook(start, length); - int result; - if (!MallocHook::InvokeMunmapReplacement(start, length, &result)) { - result = MALLOC_HOOK_SYSCALL(SYS_munmap, start, length); - } - - return result; -} - -extern "C" void* sbrk(intptr_t increment) __THROW { - MallocHook::InvokePreSbrkHook(increment); - void *result = do_sbrk(increment); - MallocHook::InvokeSbrkHook(result, increment); - return result; -} - -/*static*/void* MallocHook::UnhookedMMap(void *start, size_t length, int prot, - int flags, int fd, off_t offset) { - void* result; - if (!MallocHook::InvokeMmapReplacement( - start, length, prot, flags, fd, offset, &result)) { - result = do_mmap(start, length, prot, flags, fd, offset); - } - - return result; -} - -/*static*/int MallocHook::UnhookedMUnmap(void *start, size_t length) { - int result; - if (!MallocHook::InvokeMunmapReplacement(start, length, &result)) { - result = MALLOC_HOOK_SYSCALL(SYS_munmap, start, length); - } - return result; -} - -#undef MALLOC_HOOK_SYSCALL diff --git a/contrib/libtcmalloc/src/malloc_hook_mmap_linux.h b/contrib/libtcmalloc/src/malloc_hook_mmap_linux.h deleted file mode 100644 index 4b1386185bc..00000000000 --- a/contrib/libtcmalloc/src/malloc_hook_mmap_linux.h +++ /dev/null @@ -1,238 +0,0 @@ -// -*- Mode: C++; c-basic-offset: 2; indent-tabs-mode: nil -*- -// Copyright (c) 2005, Google Inc. -// All rights reserved. -// -// Redistribution and use in source and binary forms, with or without -// modification, are permitted provided that the following conditions are -// met: -// -// * Redistributions of source code must retain the above copyright -// notice, this list of conditions and the following disclaimer. -// * Redistributions in binary form must reproduce the above -// copyright notice, this list of conditions and the following disclaimer -// in the documentation and/or other materials provided with the -// distribution. -// * Neither the name of Google Inc. nor the names of its -// contributors may be used to endorse or promote products derived from -// this software without specific prior written permission. -// -// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS -// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT -// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR -// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT -// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, -// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT -// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, -// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY -// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT -// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE -// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. - -// --- -// Author: Sanjay Ghemawat - -// We define mmap() and mmap64(), which somewhat reimplements libc's mmap -// syscall stubs. Unfortunately libc only exports the stubs via weak symbols -// (which we're overriding with our mmap64() and mmap() wrappers) so we can't -// just call through to them. - -#ifndef __linux -# error Should only be including malloc_hook_mmap_linux.h on linux systems. -#endif - -#include -#include -#include -#include -#include "base/linux_syscall_support.h" - -// The x86-32 case and the x86-64 case differ: -// 32b has a mmap2() syscall, 64b does not. -// 64b and 32b have different calling conventions for mmap(). - -// I test for 64-bit first so I don't have to do things like -// '#if (defined(__mips__) && !defined(__MIPS64__))' as a mips32 check. -#if defined(__x86_64__) || defined(__PPC64__) || defined(__aarch64__) || (defined(_MIPS_SIM) && _MIPS_SIM == _ABI64) || defined(__s390__) - -static inline void* do_mmap64(void *start, size_t length, - int prot, int flags, - int fd, __off64_t offset) __THROW { - return sys_mmap(start, length, prot, flags, fd, offset); -} - -#define MALLOC_HOOK_HAVE_DO_MMAP64 1 - -#elif defined(__i386__) || defined(__PPC__) || defined(__mips__) || \ - defined(__arm__) - -static inline void* do_mmap64(void *start, size_t length, - int prot, int flags, - int fd, __off64_t offset) __THROW { - void *result; - - // Try mmap2() unless it's not supported - static bool have_mmap2 = true; - if (have_mmap2) { - static int pagesize = 0; - if (!pagesize) pagesize = getpagesize(); - - // Check that the offset is page aligned - if (offset & (pagesize - 1)) { - result = MAP_FAILED; - errno = EINVAL; - goto out; - } - - result = (void *)syscall(SYS_mmap2, - start, length, prot, flags, fd, - (off_t) (offset / pagesize)); - if (result != MAP_FAILED || errno != ENOSYS) goto out; - - // We don't have mmap2() after all - don't bother trying it in future - have_mmap2 = false; - } - - if (((off_t)offset) != offset) { - // If we're trying to map a 64-bit offset, fail now since we don't - // have 64-bit mmap() support. - result = MAP_FAILED; - errno = EINVAL; - goto out; - } - -#ifdef __NR_mmap - { - // Fall back to old 32-bit offset mmap() call - // Old syscall interface cannot handle six args, so pass in an array - int32 args[6] = { (int32) start, (int32) length, prot, flags, fd, - (int32)(off_t) offset }; - result = (void *)syscall(SYS_mmap, args); - } -#else - // Some Linux ports like ARM EABI Linux has no mmap, just mmap2. - result = MAP_FAILED; -#endif - - out: - return result; -} - -#define MALLOC_HOOK_HAVE_DO_MMAP64 1 - -#endif // #if defined(__x86_64__) - - -#ifdef MALLOC_HOOK_HAVE_DO_MMAP64 - -// We use do_mmap64 abstraction to put MallocHook::InvokeMmapHook -// calls right into mmap and mmap64, so that the stack frames in the caller's -// stack are at the same offsets for all the calls of memory allocating -// functions. - -// Put all callers of MallocHook::Invoke* in this module into -// malloc_hook section, -// so that MallocHook::GetCallerStackTrace can function accurately: - -// Make sure mmap doesn't get #define'd away by -# undef mmap - -extern "C" { - void* mmap64(void *start, size_t length, int prot, int flags, - int fd, __off64_t offset ) __THROW - ATTRIBUTE_SECTION(malloc_hook); - void* mmap(void *start, size_t length,int prot, int flags, - int fd, off_t offset) __THROW - ATTRIBUTE_SECTION(malloc_hook); - int munmap(void* start, size_t length) __THROW - ATTRIBUTE_SECTION(malloc_hook); - void* mremap(void* old_addr, size_t old_size, size_t new_size, - int flags, ...) __THROW - ATTRIBUTE_SECTION(malloc_hook); - void* sbrk(ptrdiff_t increment) __THROW - ATTRIBUTE_SECTION(malloc_hook); -} - -extern "C" void* mmap64(void *start, size_t length, int prot, int flags, - int fd, __off64_t offset) __THROW { - MallocHook::InvokePreMmapHook(start, length, prot, flags, fd, offset); - void *result; - if (!MallocHook::InvokeMmapReplacement( - start, length, prot, flags, fd, offset, &result)) { - result = do_mmap64(start, length, prot, flags, fd, offset); - } - MallocHook::InvokeMmapHook(result, start, length, prot, flags, fd, offset); - return result; -} - -# if !defined(__USE_FILE_OFFSET64) || !defined(__REDIRECT_NTH) - -extern "C" void* mmap(void *start, size_t length, int prot, int flags, - int fd, off_t offset) __THROW { - MallocHook::InvokePreMmapHook(start, length, prot, flags, fd, offset); - void *result; - if (!MallocHook::InvokeMmapReplacement( - start, length, prot, flags, fd, offset, &result)) { - result = do_mmap64(start, length, prot, flags, fd, - static_cast(offset)); // avoid sign extension - } - MallocHook::InvokeMmapHook(result, start, length, prot, flags, fd, offset); - return result; -} - -# endif // !defined(__USE_FILE_OFFSET64) || !defined(__REDIRECT_NTH) - -extern "C" int munmap(void* start, size_t length) __THROW { - MallocHook::InvokeMunmapHook(start, length); - int result; - if (!MallocHook::InvokeMunmapReplacement(start, length, &result)) { - result = sys_munmap(start, length); - } - return result; -} - -extern "C" void* mremap(void* old_addr, size_t old_size, size_t new_size, - int flags, ...) __THROW { - va_list ap; - va_start(ap, flags); - void *new_address = va_arg(ap, void *); - va_end(ap); - void* result = sys_mremap(old_addr, old_size, new_size, flags, new_address); - MallocHook::InvokeMremapHook(result, old_addr, old_size, new_size, flags, - new_address); - return result; -} - -#ifndef __UCLIBC__ -// libc's version: -extern "C" void* __sbrk(ptrdiff_t increment); - -extern "C" void* sbrk(ptrdiff_t increment) __THROW { - MallocHook::InvokePreSbrkHook(increment); - void *result = __sbrk(increment); - MallocHook::InvokeSbrkHook(result, increment); - return result; -} - -#endif - -/*static*/void* MallocHook::UnhookedMMap(void *start, size_t length, int prot, - int flags, int fd, off_t offset) { - void* result; - if (!MallocHook::InvokeMmapReplacement( - start, length, prot, flags, fd, offset, &result)) { - result = do_mmap64(start, length, prot, flags, fd, offset); - } - return result; -} - -/*static*/int MallocHook::UnhookedMUnmap(void *start, size_t length) { - int result; - if (!MallocHook::InvokeMunmapReplacement(start, length, &result)) { - result = syscall(SYS_munmap, start, length); - } - return result; -} - -#undef MALLOC_HOOK_HAVE_DO_MMAP64 - -#endif // #ifdef MALLOC_HOOK_HAVE_DO_MMAP64 diff --git a/contrib/libtcmalloc/src/maybe_emergency_malloc.h b/contrib/libtcmalloc/src/maybe_emergency_malloc.h deleted file mode 100644 index 250ecf01a3f..00000000000 --- a/contrib/libtcmalloc/src/maybe_emergency_malloc.h +++ /dev/null @@ -1,55 +0,0 @@ -// -*- Mode: C++; c-basic-offset: 2; indent-tabs-mode: nil -*- -// Copyright (c) 2014, gperftools Contributors -// All rights reserved. -// -// Redistribution and use in source and binary forms, with or without -// modification, are permitted provided that the following conditions are -// met: -// -// * Redistributions of source code must retain the above copyright -// notice, this list of conditions and the following disclaimer. -// * Redistributions in binary form must reproduce the above -// copyright notice, this list of conditions and the following disclaimer -// in the documentation and/or other materials provided with the -// distribution. -// * Neither the name of Google Inc. nor the names of its -// contributors may be used to endorse or promote products derived from -// this software without specific prior written permission. -// -// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS -// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT -// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR -// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT -// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, -// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT -// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, -// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY -// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT -// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE -// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. - -#ifndef MAYBE_EMERGENCY_MALLOC_H -#define MAYBE_EMERGENCY_MALLOC_H - -#include "config.h" - -#ifdef ENABLE_EMERGENCY_MALLOC - -#include "emergency_malloc.h" - -#else - -namespace tcmalloc { - static inline void *EmergencyMalloc(size_t size) {return NULL;} - static inline void EmergencyFree(void *p) {} - static inline void *EmergencyCalloc(size_t n, size_t elem_size) {return NULL;} - static inline void *EmergencyRealloc(void *old_ptr, size_t new_size) {return NULL;} - - static inline bool IsEmergencyPtr(const void *_ptr) { - return false; - } -} - -#endif // ENABLE_EMERGENCY_MALLOC - -#endif diff --git a/contrib/libtcmalloc/src/maybe_threads.cc b/contrib/libtcmalloc/src/maybe_threads.cc deleted file mode 100644 index acfc99a5ae5..00000000000 --- a/contrib/libtcmalloc/src/maybe_threads.cc +++ /dev/null @@ -1,171 +0,0 @@ -// -*- Mode: C++; c-basic-offset: 2; indent-tabs-mode: nil -*- -// Copyright (c) 2005, Google Inc. -// All rights reserved. -// -// Redistribution and use in source and binary forms, with or without -// modification, are permitted provided that the following conditions are -// met: -// -// * Redistributions of source code must retain the above copyright -// notice, this list of conditions and the following disclaimer. -// * Redistributions in binary form must reproduce the above -// copyright notice, this list of conditions and the following disclaimer -// in the documentation and/or other materials provided with the -// distribution. -// * Neither the name of Google Inc. nor the names of its -// contributors may be used to endorse or promote products derived from -// this software without specific prior written permission. -// -// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS -// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT -// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR -// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT -// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, -// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT -// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, -// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY -// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT -// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE -// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. - -// --- -// Author: Paul Menage -// -// Some wrappers for pthread functions so that we can be LD_PRELOADed -// against non-pthreads apps. -// -// This module will behave very strangely if some pthreads functions -// exist and others don't. - -#include "config.h" -#include -#include // for memcmp -#include // for __isthreaded on FreeBSD -// We don't actually need strings. But including this header seems to -// stop the compiler trying to short-circuit our pthreads existence -// tests and claiming that the address of a function is always -// non-zero. I have no idea why ... -#include -#include "maybe_threads.h" -#include "base/basictypes.h" -#include "base/logging.h" - -// __THROW is defined in glibc systems. It means, counter-intuitively, -// "This function will never throw an exception." It's an optional -// optimization tool, but we may need to use it to match glibc prototypes. -#ifndef __THROW // I guess we're not on a glibc system -# define __THROW // __THROW is just an optimization, so ok to make it "" -#endif - -// These are the methods we're going to conditionally include. -extern "C" { - int pthread_key_create (pthread_key_t*, void (*)(void*)) - __THROW ATTRIBUTE_WEAK; - int pthread_key_delete (pthread_key_t) - __THROW ATTRIBUTE_WEAK; - void *pthread_getspecific(pthread_key_t) - __THROW ATTRIBUTE_WEAK; - int pthread_setspecific(pthread_key_t, const void*) - __THROW ATTRIBUTE_WEAK; - int pthread_once(pthread_once_t *, void (*)(void)) - ATTRIBUTE_WEAK; - int pthread_atfork(void (*__prepare) (void), - void (*__parent) (void), - void (*__child) (void)) - __THROW ATTRIBUTE_WEAK; -} - -#define MAX_PERTHREAD_VALS 16 -static void *perftools_pthread_specific_vals[MAX_PERTHREAD_VALS]; -static int next_key; - -// NOTE: it's similar to bitcast defined in basic_types.h with -// exception of ignoring sizes mismatch -template -static T2 memcpy_cast(const T1 &input) { - T2 output; - size_t s = sizeof(input); - if (sizeof(output) < s) { - s = sizeof(output); - } - memcpy(&output, &input, s); - return output; -} - -int perftools_pthread_key_create(pthread_key_t *key, - void (*destr_function) (void *)) { - if (pthread_key_create) { - return pthread_key_create(key, destr_function); - } else { - assert(next_key < MAX_PERTHREAD_VALS); - *key = memcpy_cast(next_key++); - return 0; - } -} - -int perftools_pthread_key_delete(pthread_key_t key) { - if (pthread_key_delete) { - return pthread_key_delete(key); - } else { - return 0; - } -} - -void *perftools_pthread_getspecific(pthread_key_t key) { - if (pthread_getspecific) { - return pthread_getspecific(key); - } else { - return perftools_pthread_specific_vals[memcpy_cast(key)]; - } -} - -int perftools_pthread_setspecific(pthread_key_t key, void *val) { - if (pthread_setspecific) { - return pthread_setspecific(key, val); - } else { - perftools_pthread_specific_vals[memcpy_cast(key)] = val; - return 0; - } -} - - -static pthread_once_t pthread_once_init = PTHREAD_ONCE_INIT; -int perftools_pthread_once(pthread_once_t *ctl, - void (*init_routine) (void)) { -#ifdef __FreeBSD__ - // On __FreeBSD__, calling pthread_once on a system that is not - // linked with -pthread is silently a noop. :-( Luckily, we have a - // workaround: FreeBSD exposes __isthreaded in , which is - // set to 1 when the first thread is spawned. So on those systems, - // we can use our own separate pthreads-once mechanism, which is - // used until __isthreaded is 1 (which will never be true if the app - // is not linked with -pthread). - static bool pthread_once_ran_before_threads = false; - if (pthread_once_ran_before_threads) { - return 0; - } - if (!__isthreaded) { - init_routine(); - pthread_once_ran_before_threads = true; - return 0; - } -#endif - if (pthread_once) { - return pthread_once(ctl, init_routine); - } else { - if (memcmp(ctl, &pthread_once_init, sizeof(*ctl)) == 0) { - init_routine(); - ++*(char*)(ctl); // make it so it's no longer equal to init - } - return 0; - } -} - -void perftools_pthread_atfork(void (*before)(), - void (*parent_after)(), - void (*child_after)()) { - if (pthread_atfork) { - int rv = pthread_atfork(before, parent_after, child_after); - CHECK(rv == 0); - } -} diff --git a/contrib/libtcmalloc/src/maybe_threads.h b/contrib/libtcmalloc/src/maybe_threads.h deleted file mode 100644 index c6cfdf7d158..00000000000 --- a/contrib/libtcmalloc/src/maybe_threads.h +++ /dev/null @@ -1,61 +0,0 @@ -// -*- Mode: C++; c-basic-offset: 2; indent-tabs-mode: nil -*- -// Copyright (c) 2005, Google Inc. -// All rights reserved. -// -// Redistribution and use in source and binary forms, with or without -// modification, are permitted provided that the following conditions are -// met: -// -// * Redistributions of source code must retain the above copyright -// notice, this list of conditions and the following disclaimer. -// * Redistributions in binary form must reproduce the above -// copyright notice, this list of conditions and the following disclaimer -// in the documentation and/or other materials provided with the -// distribution. -// * Neither the name of Google Inc. nor the names of its -// contributors may be used to endorse or promote products derived from -// this software without specific prior written permission. -// -// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS -// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT -// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR -// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT -// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, -// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT -// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, -// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY -// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT -// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE -// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. - -// --- -// Author: Paul Menage - -//------------------------------------------------------------------- -// Some wrappers for pthread functions so that we can be LD_PRELOADed -// against non-pthreads apps. -//------------------------------------------------------------------- - -#ifndef GOOGLE_MAYBE_THREADS_H_ -#define GOOGLE_MAYBE_THREADS_H_ - -#ifdef HAVE_PTHREAD -#include -#endif - -int perftools_pthread_key_create(pthread_key_t *key, - void (*destr_function) (void *)); -int perftools_pthread_key_delete(pthread_key_t key); -void *perftools_pthread_getspecific(pthread_key_t key); -int perftools_pthread_setspecific(pthread_key_t key, void *val); -int perftools_pthread_once(pthread_once_t *ctl, - void (*init_routine) (void)); - -// Our wrapper for pthread_atfork. Does _nothing_ when there are no -// threads. See static_vars.cc:SetupAtForkLocksHandler for only user -// of this. -void perftools_pthread_atfork(void (*before)(), - void (*parent_after)(), - void (*child_after)()); - -#endif /* GOOGLE_MAYBE_THREADS_H_ */ diff --git a/contrib/libtcmalloc/src/memfs_malloc.cc b/contrib/libtcmalloc/src/memfs_malloc.cc deleted file mode 100644 index 419ef24e43b..00000000000 --- a/contrib/libtcmalloc/src/memfs_malloc.cc +++ /dev/null @@ -1,272 +0,0 @@ -// -*- Mode: C++; c-basic-offset: 2; indent-tabs-mode: nil -*- -// Copyright (c) 2007, Google Inc. -// All rights reserved. -// -// Redistribution and use in source and binary forms, with or without -// modification, are permitted provided that the following conditions are -// met: -// -// * Redistributions of source code must retain the above copyright -// notice, this list of conditions and the following disclaimer. -// * Redistributions in binary form must reproduce the above -// copyright notice, this list of conditions and the following disclaimer -// in the documentation and/or other materials provided with the -// distribution. -// * Neither the name of Google Inc. nor the names of its -// contributors may be used to endorse or promote products derived from -// this software without specific prior written permission. -// -// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS -// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT -// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR -// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT -// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, -// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT -// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, -// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY -// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT -// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE -// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. - -// --- -// Author: Arun Sharma -// -// A tcmalloc system allocator that uses a memory based filesystem such as -// tmpfs or hugetlbfs -// -// Since these only exist on linux, we only register this allocator there. - -#ifdef __linux - -#include "config.h" -#include // for errno, EINVAL -#include // for PRId64 -#include // for PATH_MAX -#include // for size_t, NULL -#ifdef HAVE_STDINT_H -#include // for int64_t, uintptr_t -#endif -#include // for snprintf -#include // for mkstemp -#include // for strerror -#include // for mmap, MAP_FAILED, etc -#include // for fstatfs, statfs -#include // for ftruncate, off_t, unlink -#include // for operator new -#include - -#include -#include "base/basictypes.h" -#include "base/googleinit.h" -#include "base/sysinfo.h" -#include "internal_logging.h" - -// TODO(sanjay): Move the code below into the tcmalloc namespace -using tcmalloc::kLog; -using tcmalloc::kCrash; -using tcmalloc::Log; -using std::string; - -DEFINE_string(memfs_malloc_path, EnvToString("TCMALLOC_MEMFS_MALLOC_PATH", ""), - "Path where hugetlbfs or tmpfs is mounted. The caller is " - "responsible for ensuring that the path is unique and does " - "not conflict with another process"); -DEFINE_int64(memfs_malloc_limit_mb, - EnvToInt("TCMALLOC_MEMFS_LIMIT_MB", 0), - "Limit total allocation size to the " - "specified number of MiB. 0 == no limit."); -DEFINE_bool(memfs_malloc_abort_on_fail, - EnvToBool("TCMALLOC_MEMFS_ABORT_ON_FAIL", false), - "abort() whenever memfs_malloc fails to satisfy an allocation " - "for any reason."); -DEFINE_bool(memfs_malloc_ignore_mmap_fail, - EnvToBool("TCMALLOC_MEMFS_IGNORE_MMAP_FAIL", false), - "Ignore failures from mmap"); -DEFINE_bool(memfs_malloc_map_private, - EnvToBool("TCMALLOC_MEMFS_MAP_PRIVATE", false), - "Use MAP_PRIVATE with mmap"); - -// Hugetlbfs based allocator for tcmalloc -class HugetlbSysAllocator: public SysAllocator { -public: - explicit HugetlbSysAllocator(SysAllocator* fallback) - : failed_(true), // To disable allocator until Initialize() is called. - big_page_size_(0), - hugetlb_fd_(-1), - hugetlb_base_(0), - fallback_(fallback) { - } - - void* Alloc(size_t size, size_t *actual_size, size_t alignment); - bool Initialize(); - - bool failed_; // Whether failed to allocate memory. - -private: - void* AllocInternal(size_t size, size_t *actual_size, size_t alignment); - - int64 big_page_size_; - int hugetlb_fd_; // file descriptor for hugetlb - off_t hugetlb_base_; - - SysAllocator* fallback_; // Default system allocator to fall back to. -}; -static union { - char buf[sizeof(HugetlbSysAllocator)]; - void *ptr; -} hugetlb_space; - -// No locking needed here since we assume that tcmalloc calls -// us with an internal lock held (see tcmalloc/system-alloc.cc). -void* HugetlbSysAllocator::Alloc(size_t size, size_t *actual_size, - size_t alignment) { - if (failed_) { - return fallback_->Alloc(size, actual_size, alignment); - } - - // We don't respond to allocation requests smaller than big_page_size_ unless - // the caller is ok to take more than they asked for. Used by MetaDataAlloc. - if (actual_size == NULL && size < big_page_size_) { - return fallback_->Alloc(size, actual_size, alignment); - } - - // Enforce huge page alignment. Be careful to deal with overflow. - size_t new_alignment = alignment; - if (new_alignment < big_page_size_) new_alignment = big_page_size_; - size_t aligned_size = ((size + new_alignment - 1) / - new_alignment) * new_alignment; - if (aligned_size < size) { - return fallback_->Alloc(size, actual_size, alignment); - } - - void* result = AllocInternal(aligned_size, actual_size, new_alignment); - if (result != NULL) { - return result; - } - Log(kLog, __FILE__, __LINE__, - "HugetlbSysAllocator: (failed, allocated)", failed_, hugetlb_base_); - if (FLAGS_memfs_malloc_abort_on_fail) { - Log(kCrash, __FILE__, __LINE__, - "memfs_malloc_abort_on_fail is set"); - } - return fallback_->Alloc(size, actual_size, alignment); -} - -void* HugetlbSysAllocator::AllocInternal(size_t size, size_t* actual_size, - size_t alignment) { - // Ask for extra memory if alignment > pagesize - size_t extra = 0; - if (alignment > big_page_size_) { - extra = alignment - big_page_size_; - } - - // Test if this allocation would put us over the limit. - off_t limit = FLAGS_memfs_malloc_limit_mb*1024*1024; - if (limit > 0 && hugetlb_base_ + size + extra > limit) { - // Disable the allocator when there's less than one page left. - if (limit - hugetlb_base_ < big_page_size_) { - Log(kLog, __FILE__, __LINE__, "reached memfs_malloc_limit_mb"); - failed_ = true; - } - else { - Log(kLog, __FILE__, __LINE__, - "alloc too large (size, bytes left)", size, limit-hugetlb_base_); - } - return NULL; - } - - // This is not needed for hugetlbfs, but needed for tmpfs. Annoyingly - // hugetlbfs returns EINVAL for ftruncate. - int ret = ftruncate(hugetlb_fd_, hugetlb_base_ + size + extra); - if (ret != 0 && errno != EINVAL) { - Log(kLog, __FILE__, __LINE__, - "ftruncate failed", strerror(errno)); - failed_ = true; - return NULL; - } - - // Note: size + extra does not overflow since: - // size + alignment < (1<(MAP_FAILED)) { - if (!FLAGS_memfs_malloc_ignore_mmap_fail) { - Log(kLog, __FILE__, __LINE__, - "mmap failed (size, error)", size + extra, strerror(errno)); - failed_ = true; - } - return NULL; - } - uintptr_t ptr = reinterpret_cast(result); - - // Adjust the return memory so it is aligned - size_t adjust = 0; - if ((ptr & (alignment - 1)) != 0) { - adjust = alignment - (ptr & (alignment - 1)); - } - ptr += adjust; - hugetlb_base_ += (size + extra); - - if (actual_size) { - *actual_size = size + extra - adjust; - } - - return reinterpret_cast(ptr); -} - -bool HugetlbSysAllocator::Initialize() { - char path[PATH_MAX]; - const int pathlen = FLAGS_memfs_malloc_path.size(); - if (pathlen + 8 > sizeof(path)) { - Log(kCrash, __FILE__, __LINE__, "XX fatal: memfs_malloc_path too long"); - return false; - } - memcpy(path, FLAGS_memfs_malloc_path.data(), pathlen); - memcpy(path + pathlen, ".XXXXXX", 8); // Also copies terminating \0 - - int hugetlb_fd = mkstemp(path); - if (hugetlb_fd == -1) { - Log(kLog, __FILE__, __LINE__, - "warning: unable to create memfs_malloc_path", - path, strerror(errno)); - return false; - } - - // Cleanup memory on process exit - if (unlink(path) == -1) { - Log(kCrash, __FILE__, __LINE__, - "fatal: error unlinking memfs_malloc_path", path, strerror(errno)); - return false; - } - - // Use fstatfs to figure out the default page size for memfs - struct statfs sfs; - if (fstatfs(hugetlb_fd, &sfs) == -1) { - Log(kCrash, __FILE__, __LINE__, - "fatal: error fstatfs of memfs_malloc_path", strerror(errno)); - return false; - } - int64 page_size = sfs.f_bsize; - - hugetlb_fd_ = hugetlb_fd; - big_page_size_ = page_size; - failed_ = false; - return true; -} - -REGISTER_MODULE_INITIALIZER(memfs_malloc, { - if (FLAGS_memfs_malloc_path.length()) { - SysAllocator* alloc = MallocExtension::instance()->GetSystemAllocator(); - HugetlbSysAllocator* hp = - new (hugetlb_space.buf) HugetlbSysAllocator(alloc); - if (hp->Initialize()) { - MallocExtension::instance()->SetSystemAllocator(hp); - } - } -}); - -#endif /* ifdef __linux */ diff --git a/contrib/libtcmalloc/src/memory_region_map.cc b/contrib/libtcmalloc/src/memory_region_map.cc deleted file mode 100644 index 841d6f3cf85..00000000000 --- a/contrib/libtcmalloc/src/memory_region_map.cc +++ /dev/null @@ -1,831 +0,0 @@ -// -*- Mode: C++; c-basic-offset: 2; indent-tabs-mode: nil -*- -/* Copyright (c) 2006, Google Inc. - * All rights reserved. - * - * Redistribution and use in source and binary forms, with or without - * modification, are permitted provided that the following conditions are - * met: - * - * * Redistributions of source code must retain the above copyright - * notice, this list of conditions and the following disclaimer. - * * Redistributions in binary form must reproduce the above - * copyright notice, this list of conditions and the following disclaimer - * in the documentation and/or other materials provided with the - * distribution. - * * Neither the name of Google Inc. nor the names of its - * contributors may be used to endorse or promote products derived from - * this software without specific prior written permission. - * - * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS - * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT - * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR - * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT - * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, - * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT - * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, - * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY - * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT - * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE - * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. - * - * --- - * Author: Maxim Lifantsev - */ - -// -// Background and key design points of MemoryRegionMap. -// -// MemoryRegionMap is a low-level module with quite atypical requirements that -// result in some degree of non-triviality of the implementation and design. -// -// MemoryRegionMap collects info about *all* memory regions created with -// mmap, munmap, mremap, sbrk. -// They key word above is 'all': all that are happening in a process -// during its lifetime frequently starting even before global object -// constructor execution. -// -// This is needed by the primary client of MemoryRegionMap: -// HeapLeakChecker uses the regions and the associated stack traces -// to figure out what part of the memory is the heap: -// if MemoryRegionMap were to miss some (early) regions, leak checking would -// stop working correctly. -// -// To accomplish the goal of functioning before/during global object -// constructor execution MemoryRegionMap is done as a singleton service -// that relies on own on-demand initialized static constructor-less data, -// and only relies on other low-level modules that can also function properly -// even before global object constructors run. -// -// Accomplishing the goal of collecting data about all mmap, munmap, mremap, -// sbrk occurrences is a more involved: conceptually to do this one needs to -// record some bits of data in particular about any mmap or sbrk call, -// but to do that one needs to allocate memory for that data at some point, -// but all memory allocations in the end themselves come from an mmap -// or sbrk call (that's how the address space of the process grows). -// -// Also note that we need to do all the above recording from -// within an mmap/sbrk hook which is sometimes/frequently is made by a memory -// allocator, including the allocator MemoryRegionMap itself must rely on. -// In the case of heap-checker usage this includes even the very first -// mmap/sbrk call happening in the program: heap-checker gets activated due to -// a link-time installed mmap/sbrk hook and it initializes MemoryRegionMap -// and asks it to record info about this very first call right from that -// very first hook invocation. -// -// MemoryRegionMap is doing its memory allocations via LowLevelAlloc: -// unlike more complex standard memory allocator, LowLevelAlloc cooperates with -// MemoryRegionMap by not holding any of its own locks while it calls mmap -// to get memory, thus we are able to call LowLevelAlloc from -// our mmap/sbrk hooks without causing a deadlock in it. -// For the same reason of deadlock prevention the locking in MemoryRegionMap -// itself is write-recursive which is an exception to Google's mutex usage. -// -// We still need to break the infinite cycle of mmap calling our hook, -// which asks LowLevelAlloc for memory to record this mmap, -// which (sometimes) causes mmap, which calls our hook, and so on. -// We do this as follows: on a recursive call of MemoryRegionMap's -// mmap/sbrk/mremap hook we record the data about the allocation in a -// static fixed-sized stack (saved_regions and saved_buckets), when the -// recursion unwinds but before returning from the outer hook call we unwind -// this stack and move the data from saved_regions and saved_buckets to its -// permanent place in the RegionSet and "bucket_table" respectively, -// which can cause more allocations and mmap-s and recursion and unwinding, -// but the whole process ends eventually due to the fact that for the small -// allocations we are doing LowLevelAlloc reuses one mmap call and parcels out -// the memory it created to satisfy several of our allocation requests. -// - -// ========================================================================= // - -#include - -#ifdef HAVE_UNISTD_H -#include -#endif -#ifdef HAVE_INTTYPES_H -#include -#endif -#ifdef HAVE_MMAP -#include -#elif !defined(MAP_FAILED) -#define MAP_FAILED -1 // the only thing we need from mman.h -#endif -#ifdef HAVE_PTHREAD -#include // for pthread_t, pthread_self() -#endif -#include - -#include -#include - -#include "memory_region_map.h" - -#include "base/googleinit.h" -#include "base/logging.h" -#include "base/low_level_alloc.h" -#include "malloc_hook-inl.h" - -#include -#include - -// MREMAP_FIXED is a linux extension. How it's used in this file, -// setting it to 0 is equivalent to saying, "This feature isn't -// supported", which is right. -#ifndef MREMAP_FIXED -# define MREMAP_FIXED 0 -#endif - -using std::max; - -// ========================================================================= // - -int MemoryRegionMap::client_count_ = 0; -int MemoryRegionMap::max_stack_depth_ = 0; -MemoryRegionMap::RegionSet* MemoryRegionMap::regions_ = NULL; -LowLevelAlloc::Arena* MemoryRegionMap::arena_ = NULL; -SpinLock MemoryRegionMap::lock_(SpinLock::LINKER_INITIALIZED); -SpinLock MemoryRegionMap::owner_lock_( // ACQUIRED_AFTER(lock_) - SpinLock::LINKER_INITIALIZED); -int MemoryRegionMap::recursion_count_ = 0; // GUARDED_BY(owner_lock_) -pthread_t MemoryRegionMap::lock_owner_tid_; // GUARDED_BY(owner_lock_) -int64 MemoryRegionMap::map_size_ = 0; -int64 MemoryRegionMap::unmap_size_ = 0; -HeapProfileBucket** MemoryRegionMap::bucket_table_ = NULL; // GUARDED_BY(lock_) -int MemoryRegionMap::num_buckets_ = 0; // GUARDED_BY(lock_) -int MemoryRegionMap::saved_buckets_count_ = 0; // GUARDED_BY(lock_) -HeapProfileBucket MemoryRegionMap::saved_buckets_[20]; // GUARDED_BY(lock_) - -// GUARDED_BY(lock_) -const void* MemoryRegionMap::saved_buckets_keys_[20][kMaxStackDepth]; - -// ========================================================================= // - -// Simple hook into execution of global object constructors, -// so that we do not call pthread_self() when it does not yet work. -static bool libpthread_initialized = false; -REGISTER_MODULE_INITIALIZER(libpthread_initialized_setter, - libpthread_initialized = true); - -static inline bool current_thread_is(pthread_t should_be) { - // Before main() runs, there's only one thread, so we're always that thread - if (!libpthread_initialized) return true; - // this starts working only sometime well into global constructor execution: - return pthread_equal(pthread_self(), should_be); -} - -// ========================================================================= // - -// Constructor-less place-holder to store a RegionSet in. -union MemoryRegionMap::RegionSetRep { - char rep[sizeof(RegionSet)]; - void* align_it; // do not need a better alignment for 'rep' than this - RegionSet* region_set() { return reinterpret_cast(rep); } -}; - -// The bytes where MemoryRegionMap::regions_ will point to. -// We use RegionSetRep with noop c-tor so that global construction -// does not interfere. -static MemoryRegionMap::RegionSetRep regions_rep; - -// ========================================================================= // - -// Has InsertRegionLocked been called recursively -// (or rather should we *not* use regions_ to record a hooked mmap). -static bool recursive_insert = false; - -void MemoryRegionMap::Init(int max_stack_depth, bool use_buckets) { - RAW_VLOG(10, "MemoryRegionMap Init"); - RAW_CHECK(max_stack_depth >= 0, ""); - // Make sure we don't overflow the memory in region stacks: - RAW_CHECK(max_stack_depth <= kMaxStackDepth, - "need to increase kMaxStackDepth?"); - Lock(); - client_count_ += 1; - max_stack_depth_ = max(max_stack_depth_, max_stack_depth); - if (client_count_ > 1) { - // not first client: already did initialization-proper - Unlock(); - RAW_VLOG(10, "MemoryRegionMap Init increment done"); - return; - } - // Set our hooks and make sure they were installed: - RAW_CHECK(MallocHook::AddMmapHook(&MmapHook), ""); - RAW_CHECK(MallocHook::AddMremapHook(&MremapHook), ""); - RAW_CHECK(MallocHook::AddSbrkHook(&SbrkHook), ""); - RAW_CHECK(MallocHook::AddMunmapHook(&MunmapHook), ""); - // We need to set recursive_insert since the NewArena call itself - // will already do some allocations with mmap which our hooks will catch - // recursive_insert allows us to buffer info about these mmap calls. - // Note that Init() can be (and is) sometimes called - // already from within an mmap/sbrk hook. - recursive_insert = true; - arena_ = LowLevelAlloc::NewArena(0, LowLevelAlloc::DefaultArena()); - recursive_insert = false; - HandleSavedRegionsLocked(&InsertRegionLocked); // flush the buffered ones - // Can't instead use HandleSavedRegionsLocked(&DoInsertRegionLocked) before - // recursive_insert = false; as InsertRegionLocked will also construct - // regions_ on demand for us. - if (use_buckets) { - const int table_bytes = kHashTableSize * sizeof(*bucket_table_); - recursive_insert = true; - bucket_table_ = static_cast( - MyAllocator::Allocate(table_bytes)); - recursive_insert = false; - memset(bucket_table_, 0, table_bytes); - num_buckets_ = 0; - } - Unlock(); - RAW_VLOG(10, "MemoryRegionMap Init done"); -} - -bool MemoryRegionMap::Shutdown() { - RAW_VLOG(10, "MemoryRegionMap Shutdown"); - Lock(); - RAW_CHECK(client_count_ > 0, ""); - client_count_ -= 1; - if (client_count_ != 0) { // not last client; need not really shutdown - Unlock(); - RAW_VLOG(10, "MemoryRegionMap Shutdown decrement done"); - return true; - } - if (bucket_table_ != NULL) { - for (int i = 0; i < kHashTableSize; i++) { - for (HeapProfileBucket* curr = bucket_table_[i]; curr != 0; /**/) { - HeapProfileBucket* bucket = curr; - curr = curr->next; - MyAllocator::Free(bucket->stack, 0); - MyAllocator::Free(bucket, 0); - } - } - MyAllocator::Free(bucket_table_, 0); - num_buckets_ = 0; - bucket_table_ = NULL; - } - RAW_CHECK(MallocHook::RemoveMmapHook(&MmapHook), ""); - RAW_CHECK(MallocHook::RemoveMremapHook(&MremapHook), ""); - RAW_CHECK(MallocHook::RemoveSbrkHook(&SbrkHook), ""); - RAW_CHECK(MallocHook::RemoveMunmapHook(&MunmapHook), ""); - if (regions_) regions_->~RegionSet(); - regions_ = NULL; - bool deleted_arena = LowLevelAlloc::DeleteArena(arena_); - if (deleted_arena) { - arena_ = 0; - } else { - RAW_LOG(WARNING, "Can't delete LowLevelAlloc arena: it's being used"); - } - Unlock(); - RAW_VLOG(10, "MemoryRegionMap Shutdown done"); - return deleted_arena; -} - -bool MemoryRegionMap::IsRecordingLocked() { - RAW_CHECK(LockIsHeld(), "should be held (by this thread)"); - return client_count_ > 0; -} - -// Invariants (once libpthread_initialized is true): -// * While lock_ is not held, recursion_count_ is 0 (and -// lock_owner_tid_ is the previous owner, but we don't rely on -// that). -// * recursion_count_ and lock_owner_tid_ are only written while -// both lock_ and owner_lock_ are held. They may be read under -// just owner_lock_. -// * At entry and exit of Lock() and Unlock(), the current thread -// owns lock_ iff pthread_equal(lock_owner_tid_, pthread_self()) -// && recursion_count_ > 0. -void MemoryRegionMap::Lock() { - { - SpinLockHolder l(&owner_lock_); - if (recursion_count_ > 0 && current_thread_is(lock_owner_tid_)) { - RAW_CHECK(lock_.IsHeld(), "Invariants violated"); - recursion_count_++; - RAW_CHECK(recursion_count_ <= 5, - "recursive lock nesting unexpectedly deep"); - return; - } - } - lock_.Lock(); - { - SpinLockHolder l(&owner_lock_); - RAW_CHECK(recursion_count_ == 0, - "Last Unlock didn't reset recursion_count_"); - if (libpthread_initialized) - lock_owner_tid_ = pthread_self(); - recursion_count_ = 1; - } -} - -void MemoryRegionMap::Unlock() { - SpinLockHolder l(&owner_lock_); - RAW_CHECK(recursion_count_ > 0, "unlock when not held"); - RAW_CHECK(lock_.IsHeld(), - "unlock when not held, and recursion_count_ is wrong"); - RAW_CHECK(current_thread_is(lock_owner_tid_), "unlock by non-holder"); - recursion_count_--; - if (recursion_count_ == 0) { - lock_.Unlock(); - } -} - -bool MemoryRegionMap::LockIsHeld() { - SpinLockHolder l(&owner_lock_); - return lock_.IsHeld() && current_thread_is(lock_owner_tid_); -} - -const MemoryRegionMap::Region* -MemoryRegionMap::DoFindRegionLocked(uintptr_t addr) { - RAW_CHECK(LockIsHeld(), "should be held (by this thread)"); - if (regions_ != NULL) { - Region sample; - sample.SetRegionSetKey(addr); - RegionSet::iterator region = regions_->lower_bound(sample); - if (region != regions_->end()) { - RAW_CHECK(addr <= region->end_addr, ""); - if (region->start_addr <= addr && addr < region->end_addr) { - return &(*region); - } - } - } - return NULL; -} - -bool MemoryRegionMap::FindRegion(uintptr_t addr, Region* result) { - Lock(); - const Region* region = DoFindRegionLocked(addr); - if (region != NULL) *result = *region; // create it as an independent copy - Unlock(); - return region != NULL; -} - -bool MemoryRegionMap::FindAndMarkStackRegion(uintptr_t stack_top, - Region* result) { - Lock(); - const Region* region = DoFindRegionLocked(stack_top); - if (region != NULL) { - RAW_VLOG(10, "Stack at %p is inside region %p..%p", - reinterpret_cast(stack_top), - reinterpret_cast(region->start_addr), - reinterpret_cast(region->end_addr)); - const_cast(region)->set_is_stack(); // now we know - // cast is safe (set_is_stack does not change the set ordering key) - *result = *region; // create *result as an independent copy - } - Unlock(); - return region != NULL; -} - -HeapProfileBucket* MemoryRegionMap::GetBucket(int depth, - const void* const key[]) { - RAW_CHECK(LockIsHeld(), "should be held (by this thread)"); - // Make hash-value - uintptr_t hash = 0; - for (int i = 0; i < depth; i++) { - hash += reinterpret_cast(key[i]); - hash += hash << 10; - hash ^= hash >> 6; - } - hash += hash << 3; - hash ^= hash >> 11; - - // Lookup stack trace in table - unsigned int hash_index = (static_cast(hash)) % kHashTableSize; - for (HeapProfileBucket* bucket = bucket_table_[hash_index]; - bucket != 0; - bucket = bucket->next) { - if ((bucket->hash == hash) && (bucket->depth == depth) && - std::equal(key, key + depth, bucket->stack)) { - return bucket; - } - } - - // Create new bucket - const size_t key_size = sizeof(key[0]) * depth; - HeapProfileBucket* bucket; - if (recursive_insert) { // recursion: save in saved_buckets_ - const void** key_copy = saved_buckets_keys_[saved_buckets_count_]; - std::copy(key, key + depth, key_copy); - bucket = &saved_buckets_[saved_buckets_count_]; - memset(bucket, 0, sizeof(*bucket)); - ++saved_buckets_count_; - bucket->stack = key_copy; - bucket->next = NULL; - } else { - recursive_insert = true; - const void** key_copy = static_cast( - MyAllocator::Allocate(key_size)); - recursive_insert = false; - std::copy(key, key + depth, key_copy); - recursive_insert = true; - bucket = static_cast( - MyAllocator::Allocate(sizeof(HeapProfileBucket))); - recursive_insert = false; - memset(bucket, 0, sizeof(*bucket)); - bucket->stack = key_copy; - bucket->next = bucket_table_[hash_index]; - } - bucket->hash = hash; - bucket->depth = depth; - bucket_table_[hash_index] = bucket; - ++num_buckets_; - return bucket; -} - -MemoryRegionMap::RegionIterator MemoryRegionMap::BeginRegionLocked() { - RAW_CHECK(LockIsHeld(), "should be held (by this thread)"); - RAW_CHECK(regions_ != NULL, ""); - return regions_->begin(); -} - -MemoryRegionMap::RegionIterator MemoryRegionMap::EndRegionLocked() { - RAW_CHECK(LockIsHeld(), "should be held (by this thread)"); - RAW_CHECK(regions_ != NULL, ""); - return regions_->end(); -} - -inline void MemoryRegionMap::DoInsertRegionLocked(const Region& region) { - RAW_VLOG(12, "Inserting region %p..%p from %p", - reinterpret_cast(region.start_addr), - reinterpret_cast(region.end_addr), - reinterpret_cast(region.caller())); - RegionSet::const_iterator i = regions_->lower_bound(region); - if (i != regions_->end() && i->start_addr <= region.start_addr) { - RAW_DCHECK(region.end_addr <= i->end_addr, ""); // lower_bound ensures this - return; // 'region' is a subset of an already recorded region; do nothing - // We can be stricter and allow this only when *i has been created via - // an mmap with MAP_NORESERVE flag set. - } - if (DEBUG_MODE) { - RAW_CHECK(i == regions_->end() || !region.Overlaps(*i), - "Wow, overlapping memory regions"); - Region sample; - sample.SetRegionSetKey(region.start_addr); - i = regions_->lower_bound(sample); - RAW_CHECK(i == regions_->end() || !region.Overlaps(*i), - "Wow, overlapping memory regions"); - } - region.AssertIsConsistent(); // just making sure - // This inserts and allocates permanent storage for region - // and its call stack data: it's safe to do it now: - regions_->insert(region); - RAW_VLOG(12, "Inserted region %p..%p :", - reinterpret_cast(region.start_addr), - reinterpret_cast(region.end_addr)); - if (VLOG_IS_ON(12)) LogAllLocked(); -} - -// These variables are local to MemoryRegionMap::InsertRegionLocked() -// and MemoryRegionMap::HandleSavedRegionsLocked() -// and are file-level to ensure that they are initialized at load time. - -// Number of unprocessed region inserts. -static int saved_regions_count = 0; - -// Unprocessed inserts (must be big enough to hold all allocations that can -// be caused by a InsertRegionLocked call). -// Region has no constructor, so that c-tor execution does not interfere -// with the any-time use of the static memory behind saved_regions. -static MemoryRegionMap::Region saved_regions[20]; - -inline void MemoryRegionMap::HandleSavedRegionsLocked( - void (*insert_func)(const Region& region)) { - while (saved_regions_count > 0) { - // Making a local-var copy of the region argument to insert_func - // including its stack (w/o doing any memory allocations) is important: - // in many cases the memory in saved_regions - // will get written-to during the (*insert_func)(r) call below. - Region r = saved_regions[--saved_regions_count]; - (*insert_func)(r); - } -} - -void MemoryRegionMap::RestoreSavedBucketsLocked() { - RAW_CHECK(LockIsHeld(), "should be held (by this thread)"); - while (saved_buckets_count_ > 0) { - HeapProfileBucket bucket = saved_buckets_[--saved_buckets_count_]; - unsigned int hash_index = - static_cast(bucket.hash) % kHashTableSize; - bool is_found = false; - for (HeapProfileBucket* curr = bucket_table_[hash_index]; - curr != 0; - curr = curr->next) { - if ((curr->hash == bucket.hash) && (curr->depth == bucket.depth) && - std::equal(bucket.stack, bucket.stack + bucket.depth, curr->stack)) { - curr->allocs += bucket.allocs; - curr->alloc_size += bucket.alloc_size; - curr->frees += bucket.frees; - curr->free_size += bucket.free_size; - is_found = true; - break; - } - } - if (is_found) continue; - - const size_t key_size = sizeof(bucket.stack[0]) * bucket.depth; - const void** key_copy = static_cast( - MyAllocator::Allocate(key_size)); - std::copy(bucket.stack, bucket.stack + bucket.depth, key_copy); - HeapProfileBucket* new_bucket = static_cast( - MyAllocator::Allocate(sizeof(HeapProfileBucket))); - memset(new_bucket, 0, sizeof(*new_bucket)); - new_bucket->hash = bucket.hash; - new_bucket->depth = bucket.depth; - new_bucket->stack = key_copy; - new_bucket->next = bucket_table_[hash_index]; - bucket_table_[hash_index] = new_bucket; - ++num_buckets_; - } -} - -inline void MemoryRegionMap::InsertRegionLocked(const Region& region) { - RAW_CHECK(LockIsHeld(), "should be held (by this thread)"); - // We can be called recursively, because RegionSet constructor - // and DoInsertRegionLocked() (called below) can call the allocator. - // recursive_insert tells us if that's the case. When this happens, - // region insertion information is recorded in saved_regions[], - // and taken into account when the recursion unwinds. - // Do the insert: - if (recursive_insert) { // recursion: save in saved_regions - RAW_VLOG(12, "Saving recursive insert of region %p..%p from %p", - reinterpret_cast(region.start_addr), - reinterpret_cast(region.end_addr), - reinterpret_cast(region.caller())); - RAW_CHECK(saved_regions_count < arraysize(saved_regions), ""); - // Copy 'region' to saved_regions[saved_regions_count] - // together with the contents of its call_stack, - // then increment saved_regions_count. - saved_regions[saved_regions_count++] = region; - } else { // not a recusrive call - if (regions_ == NULL) { // init regions_ - RAW_VLOG(12, "Initializing region set"); - regions_ = regions_rep.region_set(); - recursive_insert = true; - new(regions_) RegionSet(); - HandleSavedRegionsLocked(&DoInsertRegionLocked); - recursive_insert = false; - } - recursive_insert = true; - // Do the actual insertion work to put new regions into regions_: - DoInsertRegionLocked(region); - HandleSavedRegionsLocked(&DoInsertRegionLocked); - recursive_insert = false; - } -} - -// We strip out different number of stack frames in debug mode -// because less inlining happens in that case -#ifdef NDEBUG -static const int kStripFrames = 1; -#else -static const int kStripFrames = 3; -#endif - -void MemoryRegionMap::RecordRegionAddition(const void* start, size_t size) { - // Record start/end info about this memory acquisition call in a new region: - Region region; - region.Create(start, size); - // First get the call stack info into the local varible 'region': - int depth = 0; - // NOTE: libunwind also does mmap and very much likely while holding - // it's own lock(s). So some threads may first take libunwind lock, - // and then take region map lock (necessary to record mmap done from - // inside libunwind). On the other hand other thread(s) may do - // normal mmap. Which would call this method to record it. Which - // would then proceed with installing that record to region map - // while holding region map lock. That may cause mmap from our own - // internal allocators, so attempt to unwind in this case may cause - // reverse order of taking libuwind and region map locks. Which is - // obvious deadlock. - // - // Thankfully, we can easily detect if we're holding region map lock - // and avoid recording backtrace in this (rare and largely - // irrelevant) case. By doing this we "declare" that thread needing - // both locks must take region map lock last. In other words we do - // not allow taking libuwind lock when we already have region map - // lock. Note, this is generally impossible when somebody tries to - // mix cpu profiling and heap checking/profiling, because cpu - // profiler grabs backtraces at arbitrary places. But at least such - // combination is rarer and less relevant. - if (max_stack_depth_ > 0 && !LockIsHeld()) { - depth = MallocHook::GetCallerStackTrace(const_cast(region.call_stack), - max_stack_depth_, kStripFrames + 1); - } - region.set_call_stack_depth(depth); // record stack info fully - RAW_VLOG(10, "New global region %p..%p from %p", - reinterpret_cast(region.start_addr), - reinterpret_cast(region.end_addr), - reinterpret_cast(region.caller())); - // Note: none of the above allocates memory. - Lock(); // recursively lock - map_size_ += size; - InsertRegionLocked(region); - // This will (eventually) allocate storage for and copy over the stack data - // from region.call_stack_data_ that is pointed by region.call_stack(). - if (bucket_table_ != NULL) { - HeapProfileBucket* b = GetBucket(depth, region.call_stack); - ++b->allocs; - b->alloc_size += size; - if (!recursive_insert) { - recursive_insert = true; - RestoreSavedBucketsLocked(); - recursive_insert = false; - } - } - Unlock(); -} - -void MemoryRegionMap::RecordRegionRemoval(const void* start, size_t size) { - Lock(); - if (recursive_insert) { - // First remove the removed region from saved_regions, if it's - // there, to prevent overrunning saved_regions in recursive - // map/unmap call sequences, and also from later inserting regions - // which have already been unmapped. - uintptr_t start_addr = reinterpret_cast(start); - uintptr_t end_addr = start_addr + size; - int put_pos = 0; - int old_count = saved_regions_count; - for (int i = 0; i < old_count; ++i, ++put_pos) { - Region& r = saved_regions[i]; - if (r.start_addr == start_addr && r.end_addr == end_addr) { - // An exact match, so it's safe to remove. - RecordRegionRemovalInBucket(r.call_stack_depth, r.call_stack, size); - --saved_regions_count; - --put_pos; - RAW_VLOG(10, ("Insta-Removing saved region %p..%p; " - "now have %d saved regions"), - reinterpret_cast(start_addr), - reinterpret_cast(end_addr), - saved_regions_count); - } else { - if (put_pos < i) { - saved_regions[put_pos] = saved_regions[i]; - } - } - } - } - if (regions_ == NULL) { // We must have just unset the hooks, - // but this thread was already inside the hook. - Unlock(); - return; - } - if (!recursive_insert) { - HandleSavedRegionsLocked(&InsertRegionLocked); - } - // first handle adding saved regions if any - uintptr_t start_addr = reinterpret_cast(start); - uintptr_t end_addr = start_addr + size; - // subtract start_addr, end_addr from all the regions - RAW_VLOG(10, "Removing global region %p..%p; have %" PRIuS " regions", - reinterpret_cast(start_addr), - reinterpret_cast(end_addr), - regions_->size()); - Region sample; - sample.SetRegionSetKey(start_addr); - // Only iterate over the regions that might overlap start_addr..end_addr: - for (RegionSet::iterator region = regions_->lower_bound(sample); - region != regions_->end() && region->start_addr < end_addr; - /*noop*/) { - RAW_VLOG(13, "Looking at region %p..%p", - reinterpret_cast(region->start_addr), - reinterpret_cast(region->end_addr)); - if (start_addr <= region->start_addr && - region->end_addr <= end_addr) { // full deletion - RAW_VLOG(12, "Deleting region %p..%p", - reinterpret_cast(region->start_addr), - reinterpret_cast(region->end_addr)); - RecordRegionRemovalInBucket(region->call_stack_depth, region->call_stack, - region->end_addr - region->start_addr); - RegionSet::iterator d = region; - ++region; - regions_->erase(d); - continue; - } else if (region->start_addr < start_addr && - end_addr < region->end_addr) { // cutting-out split - RAW_VLOG(12, "Splitting region %p..%p in two", - reinterpret_cast(region->start_addr), - reinterpret_cast(region->end_addr)); - RecordRegionRemovalInBucket(region->call_stack_depth, region->call_stack, - end_addr - start_addr); - // Make another region for the start portion: - // The new region has to be the start portion because we can't - // just modify region->end_addr as it's the sorting key. - Region r = *region; - r.set_end_addr(start_addr); - InsertRegionLocked(r); - // cut *region from start: - const_cast(*region).set_start_addr(end_addr); - } else if (end_addr > region->start_addr && - start_addr <= region->start_addr) { // cut from start - RAW_VLOG(12, "Start-chopping region %p..%p", - reinterpret_cast(region->start_addr), - reinterpret_cast(region->end_addr)); - RecordRegionRemovalInBucket(region->call_stack_depth, region->call_stack, - end_addr - region->start_addr); - const_cast(*region).set_start_addr(end_addr); - } else if (start_addr > region->start_addr && - start_addr < region->end_addr) { // cut from end - RAW_VLOG(12, "End-chopping region %p..%p", - reinterpret_cast(region->start_addr), - reinterpret_cast(region->end_addr)); - RecordRegionRemovalInBucket(region->call_stack_depth, region->call_stack, - region->end_addr - start_addr); - // Can't just modify region->end_addr (it's the sorting key): - Region r = *region; - r.set_end_addr(start_addr); - RegionSet::iterator d = region; - ++region; - // It's safe to erase before inserting since r is independent of *d: - // r contains an own copy of the call stack: - regions_->erase(d); - InsertRegionLocked(r); - continue; - } - ++region; - } - RAW_VLOG(12, "Removed region %p..%p; have %" PRIuS " regions", - reinterpret_cast(start_addr), - reinterpret_cast(end_addr), - regions_->size()); - if (VLOG_IS_ON(12)) LogAllLocked(); - unmap_size_ += size; - Unlock(); -} - -void MemoryRegionMap::RecordRegionRemovalInBucket(int depth, - const void* const stack[], - size_t size) { - RAW_CHECK(LockIsHeld(), "should be held (by this thread)"); - if (bucket_table_ == NULL) return; - HeapProfileBucket* b = GetBucket(depth, stack); - ++b->frees; - b->free_size += size; -} - -void MemoryRegionMap::MmapHook(const void* result, - const void* start, size_t size, - int prot, int flags, - int fd, off_t offset) { - // TODO(maxim): replace all 0x%" PRIxS " by %p when RAW_VLOG uses a safe - // snprintf reimplementation that does not malloc to pretty-print NULL - RAW_VLOG(10, "MMap = 0x%" PRIxPTR " of %" PRIuS " at %" PRIu64 " " - "prot %d flags %d fd %d offs %" PRId64, - reinterpret_cast(result), size, - reinterpret_cast(start), prot, flags, fd, - static_cast(offset)); - if (result != reinterpret_cast(MAP_FAILED) && size != 0) { - RecordRegionAddition(result, size); - } -} - -void MemoryRegionMap::MunmapHook(const void* ptr, size_t size) { - RAW_VLOG(10, "MUnmap of %p %" PRIuS "", ptr, size); - if (size != 0) { - RecordRegionRemoval(ptr, size); - } -} - -void MemoryRegionMap::MremapHook(const void* result, - const void* old_addr, size_t old_size, - size_t new_size, int flags, - const void* new_addr) { - RAW_VLOG(10, "MRemap = 0x%" PRIxPTR " of 0x%" PRIxPTR " %" PRIuS " " - "to %" PRIuS " flags %d new_addr=0x%" PRIxPTR, - (uintptr_t)result, (uintptr_t)old_addr, - old_size, new_size, flags, - flags & MREMAP_FIXED ? (uintptr_t)new_addr : 0); - if (result != reinterpret_cast(-1)) { - RecordRegionRemoval(old_addr, old_size); - RecordRegionAddition(result, new_size); - } -} - -void MemoryRegionMap::SbrkHook(const void* result, ptrdiff_t increment) { - RAW_VLOG(10, "Sbrk = 0x%" PRIxPTR " of %" PRIdS "", (uintptr_t)result, increment); - if (result != reinterpret_cast(-1)) { - if (increment > 0) { - void* new_end = sbrk(0); - RecordRegionAddition(result, reinterpret_cast(new_end) - - reinterpret_cast(result)); - } else if (increment < 0) { - void* new_end = sbrk(0); - RecordRegionRemoval(new_end, reinterpret_cast(result) - - reinterpret_cast(new_end)); - } - } -} - -void MemoryRegionMap::LogAllLocked() { - RAW_CHECK(LockIsHeld(), "should be held (by this thread)"); - RAW_LOG(INFO, "List of regions:"); - uintptr_t previous = 0; - for (RegionSet::const_iterator r = regions_->begin(); - r != regions_->end(); ++r) { - RAW_LOG(INFO, "Memory region 0x%" PRIxPTR "..0x%" PRIxPTR " " - "from 0x%" PRIxPTR " stack=%d", - r->start_addr, r->end_addr, r->caller(), r->is_stack); - RAW_CHECK(previous < r->end_addr, "wow, we messed up the set order"); - // this must be caused by uncontrolled recursive operations on regions_ - previous = r->end_addr; - } - RAW_LOG(INFO, "End of regions list"); -} diff --git a/contrib/libtcmalloc/src/memory_region_map.h b/contrib/libtcmalloc/src/memory_region_map.h deleted file mode 100644 index ec388e1cc54..00000000000 --- a/contrib/libtcmalloc/src/memory_region_map.h +++ /dev/null @@ -1,413 +0,0 @@ -// -*- Mode: C++; c-basic-offset: 2; indent-tabs-mode: nil -*- -/* Copyright (c) 2006, Google Inc. - * All rights reserved. - * - * Redistribution and use in source and binary forms, with or without - * modification, are permitted provided that the following conditions are - * met: - * - * * Redistributions of source code must retain the above copyright - * notice, this list of conditions and the following disclaimer. - * * Redistributions in binary form must reproduce the above - * copyright notice, this list of conditions and the following disclaimer - * in the documentation and/or other materials provided with the - * distribution. - * * Neither the name of Google Inc. nor the names of its - * contributors may be used to endorse or promote products derived from - * this software without specific prior written permission. - * - * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS - * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT - * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR - * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT - * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, - * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT - * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, - * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY - * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT - * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE - * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. - * - * --- - * Author: Maxim Lifantsev - */ - -#ifndef BASE_MEMORY_REGION_MAP_H_ -#define BASE_MEMORY_REGION_MAP_H_ - -#include - -#ifdef HAVE_PTHREAD -#include -#endif -#include -#include -#include "base/stl_allocator.h" -#include "base/spinlock.h" -#include "base/thread_annotations.h" -#include "base/low_level_alloc.h" -#include "heap-profile-stats.h" - -// TODO(maxim): add a unittest: -// execute a bunch of mmaps and compare memory map what strace logs -// execute a bunch of mmap/munmup and compare memory map with -// own accounting of what those mmaps generated - -// Thread-safe class to collect and query the map of all memory regions -// in a process that have been created with mmap, munmap, mremap, sbrk. -// For each memory region, we keep track of (and provide to users) -// the stack trace that allocated that memory region. -// The recorded stack trace depth is bounded by -// a user-supplied max_stack_depth parameter of Init(). -// After initialization with Init() -// (which can happened even before global object constructor execution) -// we collect the map by installing and monitoring MallocHook-s -// to mmap, munmap, mremap, sbrk. -// At any time one can query this map via provided interface. -// For more details on the design of MemoryRegionMap -// see the comment at the top of our .cc file. -class MemoryRegionMap { - private: - // Max call stack recording depth supported by Init(). Set it to be - // high enough for all our clients. Note: we do not define storage - // for this (doing that requires special handling in windows), so - // don't take the address of it! - static const int kMaxStackDepth = 32; - - // Size of the hash table of buckets. A structure of the bucket table is - // described in heap-profile-stats.h. - static const int kHashTableSize = 179999; - - public: - // interface ================================================================ - - // Every client of MemoryRegionMap must call Init() before first use, - // and Shutdown() after last use. This allows us to reference count - // this (singleton) class properly. MemoryRegionMap assumes it's the - // only client of MallocHooks, so a client can only register other - // MallocHooks after calling Init() and must unregister them before - // calling Shutdown(). - - // Initialize this module to record memory allocation stack traces. - // Stack traces that have more than "max_stack_depth" frames - // are automatically shrunk to "max_stack_depth" when they are recorded. - // Init() can be called more than once w/o harm, largest max_stack_depth - // will be the effective one. - // When "use_buckets" is true, then counts of mmap and munmap sizes will be - // recorded with each stack trace. If Init() is called more than once, then - // counting will be effective after any call contained "use_buckets" of true. - // It will install mmap, munmap, mremap, sbrk hooks - // and initialize arena_ and our hook and locks, hence one can use - // MemoryRegionMap::Lock()/Unlock() to manage the locks. - // Uses Lock/Unlock inside. - static void Init(int max_stack_depth, bool use_buckets); - - // Try to shutdown this module undoing what Init() did. - // Returns true iff could do full shutdown (or it was not attempted). - // Full shutdown is attempted when the number of Shutdown() calls equals - // the number of Init() calls. - static bool Shutdown(); - - // Return true if MemoryRegionMap is initialized and recording, i.e. when - // then number of Init() calls are more than the number of Shutdown() calls. - static bool IsRecordingLocked(); - - // Locks to protect our internal data structures. - // These also protect use of arena_ if our Init() has been done. - // The lock is recursive. - static void Lock() EXCLUSIVE_LOCK_FUNCTION(lock_); - static void Unlock() UNLOCK_FUNCTION(lock_); - - // Returns true when the lock is held by this thread (for use in RAW_CHECK-s). - static bool LockIsHeld(); - - // Locker object that acquires the MemoryRegionMap::Lock - // for the duration of its lifetime (a C++ scope). - class LockHolder { - public: - LockHolder() { Lock(); } - ~LockHolder() { Unlock(); } - private: - DISALLOW_COPY_AND_ASSIGN(LockHolder); - }; - - // A memory region that we know about through malloc_hook-s. - // This is essentially an interface through which MemoryRegionMap - // exports the collected data to its clients. Thread-compatible. - struct Region { - uintptr_t start_addr; // region start address - uintptr_t end_addr; // region end address - int call_stack_depth; // number of caller stack frames that we saved - const void* call_stack[kMaxStackDepth]; // caller address stack array - // filled to call_stack_depth size - bool is_stack; // does this region contain a thread's stack: - // a user of MemoryRegionMap supplies this info - - // Convenience accessor for call_stack[0], - // i.e. (the program counter of) the immediate caller - // of this region's allocation function, - // but it also returns NULL when call_stack_depth is 0, - // i.e whe we weren't able to get the call stack. - // This usually happens in recursive calls, when the stack-unwinder - // calls mmap() which in turn calls the stack-unwinder. - uintptr_t caller() const { - return reinterpret_cast(call_stack_depth >= 1 - ? call_stack[0] : NULL); - } - - // Return true iff this region overlaps region x. - bool Overlaps(const Region& x) const { - return start_addr < x.end_addr && end_addr > x.start_addr; - } - - private: // helpers for MemoryRegionMap - friend class MemoryRegionMap; - - // The ways we create Region-s: - void Create(const void* start, size_t size) { - start_addr = reinterpret_cast(start); - end_addr = start_addr + size; - is_stack = false; // not a stack till marked such - call_stack_depth = 0; - AssertIsConsistent(); - } - void set_call_stack_depth(int depth) { - RAW_DCHECK(call_stack_depth == 0, ""); // only one such set is allowed - call_stack_depth = depth; - AssertIsConsistent(); - } - - // The ways we modify Region-s: - void set_is_stack() { is_stack = true; } - void set_start_addr(uintptr_t addr) { - start_addr = addr; - AssertIsConsistent(); - } - void set_end_addr(uintptr_t addr) { - end_addr = addr; - AssertIsConsistent(); - } - - // Verifies that *this contains consistent data, crashes if not the case. - void AssertIsConsistent() const { - RAW_DCHECK(start_addr < end_addr, ""); - RAW_DCHECK(call_stack_depth >= 0 && - call_stack_depth <= kMaxStackDepth, ""); - } - - // Post-default construction helper to make a Region suitable - // for searching in RegionSet regions_. - void SetRegionSetKey(uintptr_t addr) { - // make sure *this has no usable data: - if (DEBUG_MODE) memset(this, 0xFF, sizeof(*this)); - end_addr = addr; - } - - // Note: call_stack[kMaxStackDepth] as a member lets us make Region - // a simple self-contained struct with correctly behaving bit-vise copying. - // This simplifies the code of this module but wastes some memory: - // in most-often use case of this module (leak checking) - // only one call_stack element out of kMaxStackDepth is actually needed. - // Making the storage for call_stack variable-sized, - // substantially complicates memory management for the Region-s: - // as they need to be created and manipulated for some time - // w/o any memory allocations, yet are also given out to the users. - }; - - // Find the region that covers addr and write its data into *result if found, - // in which case *result gets filled so that it stays fully functional - // even when the underlying region gets removed from MemoryRegionMap. - // Returns success. Uses Lock/Unlock inside. - static bool FindRegion(uintptr_t addr, Region* result); - - // Find the region that contains stack_top, mark that region as - // a stack region, and write its data into *result if found, - // in which case *result gets filled so that it stays fully functional - // even when the underlying region gets removed from MemoryRegionMap. - // Returns success. Uses Lock/Unlock inside. - static bool FindAndMarkStackRegion(uintptr_t stack_top, Region* result); - - // Iterate over the buckets which store mmap and munmap counts per stack - // trace. It calls "callback" for each bucket, and passes "arg" to it. - template - static void IterateBuckets(void (*callback)(const HeapProfileBucket*, Type), - Type arg); - - // Get the bucket whose caller stack trace is "key". The stack trace is - // used to a depth of "depth" at most. The requested bucket is created if - // needed. - // The bucket table is described in heap-profile-stats.h. - static HeapProfileBucket* GetBucket(int depth, const void* const key[]); - - private: // our internal types ============================================== - - // Region comparator for sorting with STL - struct RegionCmp { - bool operator()(const Region& x, const Region& y) const { - return x.end_addr < y.end_addr; - } - }; - - // We allocate STL objects in our own arena. - struct MyAllocator { - static void *Allocate(size_t n) { - return LowLevelAlloc::AllocWithArena(n, arena_); - } - static void Free(const void *p, size_t /* n */) { - LowLevelAlloc::Free(const_cast(p)); - } - }; - - // Set of the memory regions - typedef std::set > RegionSet; - - public: // more in-depth interface ========================================== - - // STL iterator with values of Region - typedef RegionSet::const_iterator RegionIterator; - - // Return the begin/end iterators to all the regions. - // These need Lock/Unlock protection around their whole usage (loop). - // Even when the same thread causes modifications during such a loop - // (which are permitted due to recursive locking) - // the loop iterator will still be valid as long as its region - // has not been deleted, but EndRegionLocked should be - // re-evaluated whenever the set of regions has changed. - static RegionIterator BeginRegionLocked(); - static RegionIterator EndRegionLocked(); - - // Return the accumulated sizes of mapped and unmapped regions. - static int64 MapSize() { return map_size_; } - static int64 UnmapSize() { return unmap_size_; } - - // Effectively private type from our .cc ================================= - // public to let us declare global objects: - union RegionSetRep; - - private: - // representation =========================================================== - - // Counter of clients of this module that have called Init(). - static int client_count_; - - // Maximal number of caller stack frames to save (>= 0). - static int max_stack_depth_; - - // Arena used for our allocations in regions_. - static LowLevelAlloc::Arena* arena_; - - // Set of the mmap/sbrk/mremap-ed memory regions - // To be accessed *only* when Lock() is held. - // Hence we protect the non-recursive lock used inside of arena_ - // with our recursive Lock(). This lets a user prevent deadlocks - // when threads are stopped by TCMalloc_ListAllProcessThreads at random spots - // simply by acquiring our recursive Lock() before that. - static RegionSet* regions_; - - // Lock to protect regions_ and buckets_ variables and the data behind. - static SpinLock lock_; - // Lock to protect the recursive lock itself. - static SpinLock owner_lock_; - - // Recursion count for the recursive lock. - static int recursion_count_; - // The thread id of the thread that's inside the recursive lock. - static pthread_t lock_owner_tid_; - - // Total size of all mapped pages so far - static int64 map_size_; - // Total size of all unmapped pages so far - static int64 unmap_size_; - - // Bucket hash table which is described in heap-profile-stats.h. - static HeapProfileBucket** bucket_table_ GUARDED_BY(lock_); - static int num_buckets_ GUARDED_BY(lock_); - - // The following members are local to MemoryRegionMap::GetBucket() - // and MemoryRegionMap::HandleSavedBucketsLocked() - // and are file-level to ensure that they are initialized at load time. - // - // These are used as temporary storage to break the infinite cycle of mmap - // calling our hook which (sometimes) causes mmap. It must be a static - // fixed-size array. The size 20 is just an expected value for safety. - // The details are described in memory_region_map.cc. - - // Number of unprocessed bucket inserts. - static int saved_buckets_count_ GUARDED_BY(lock_); - - // Unprocessed inserts (must be big enough to hold all mmaps that can be - // caused by a GetBucket call). - // Bucket has no constructor, so that c-tor execution does not interfere - // with the any-time use of the static memory behind saved_buckets. - static HeapProfileBucket saved_buckets_[20] GUARDED_BY(lock_); - - static const void* saved_buckets_keys_[20][kMaxStackDepth] GUARDED_BY(lock_); - - // helpers ================================================================== - - // Helper for FindRegion and FindAndMarkStackRegion: - // returns the region covering 'addr' or NULL; assumes our lock_ is held. - static const Region* DoFindRegionLocked(uintptr_t addr); - - // Verifying wrapper around regions_->insert(region) - // To be called to do InsertRegionLocked's work only! - inline static void DoInsertRegionLocked(const Region& region); - // Handle regions saved by InsertRegionLocked into a tmp static array - // by calling insert_func on them. - inline static void HandleSavedRegionsLocked( - void (*insert_func)(const Region& region)); - - // Restore buckets saved in a tmp static array by GetBucket to the bucket - // table where all buckets eventually should be. - static void RestoreSavedBucketsLocked(); - - // Wrapper around DoInsertRegionLocked - // that handles the case of recursive allocator calls. - inline static void InsertRegionLocked(const Region& region); - - // Record addition of a memory region at address "start" of size "size" - // (called from our mmap/mremap/sbrk hooks). - static void RecordRegionAddition(const void* start, size_t size); - // Record deletion of a memory region at address "start" of size "size" - // (called from our munmap/mremap/sbrk hooks). - static void RecordRegionRemoval(const void* start, size_t size); - - // Record deletion of a memory region of size "size" in a bucket whose - // caller stack trace is "key". The stack trace is used to a depth of - // "depth" at most. - static void RecordRegionRemovalInBucket(int depth, - const void* const key[], - size_t size); - - // Hooks for MallocHook - static void MmapHook(const void* result, - const void* start, size_t size, - int prot, int flags, - int fd, off_t offset); - static void MunmapHook(const void* ptr, size_t size); - static void MremapHook(const void* result, const void* old_addr, - size_t old_size, size_t new_size, int flags, - const void* new_addr); - static void SbrkHook(const void* result, ptrdiff_t increment); - - // Log all memory regions; Useful for debugging only. - // Assumes Lock() is held - static void LogAllLocked(); - - DISALLOW_COPY_AND_ASSIGN(MemoryRegionMap); -}; - -template -void MemoryRegionMap::IterateBuckets( - void (*callback)(const HeapProfileBucket*, Type), Type callback_arg) { - for (int index = 0; index < kHashTableSize; index++) { - for (HeapProfileBucket* bucket = bucket_table_[index]; - bucket != NULL; - bucket = bucket->next) { - callback(bucket, callback_arg); - } - } -} - -#endif // BASE_MEMORY_REGION_MAP_H_ diff --git a/contrib/libtcmalloc/src/packed-cache-inl.h b/contrib/libtcmalloc/src/packed-cache-inl.h deleted file mode 100644 index 09462608ece..00000000000 --- a/contrib/libtcmalloc/src/packed-cache-inl.h +++ /dev/null @@ -1,239 +0,0 @@ -// -*- Mode: C++; c-basic-offset: 2; indent-tabs-mode: nil -*- -// Copyright (c) 2007, Google Inc. -// All rights reserved. -// -// Redistribution and use in source and binary forms, with or without -// modification, are permitted provided that the following conditions are -// met: -// -// * Redistributions of source code must retain the above copyright -// notice, this list of conditions and the following disclaimer. -// * Redistributions in binary form must reproduce the above -// copyright notice, this list of conditions and the following disclaimer -// in the documentation and/or other materials provided with the -// distribution. -// * Neither the name of Google Inc. nor the names of its -// contributors may be used to endorse or promote products derived from -// this software without specific prior written permission. -// -// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS -// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT -// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR -// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT -// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, -// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT -// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, -// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY -// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT -// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE -// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. - -// --- -// Author: Geoff Pike -// -// This file provides a minimal cache that can hold a pair -// with little if any wasted space. The types of the key and value -// must be unsigned integral types or at least have unsigned semantics -// for >>, casting, and similar operations. -// -// Synchronization is not provided. However, the cache is implemented -// as an array of cache entries whose type is chosen at compile time. -// If a[i] is atomic on your hardware for the chosen array type then -// raciness will not necessarily lead to bugginess. The cache entries -// must be large enough to hold a partial key and a value packed -// together. The partial keys are bit strings of length -// kKeybits - kHashbits, and the values are bit strings of length kValuebits. -// -// In an effort to use minimal space, every cache entry represents -// some pair; the class provides no way to mark a cache -// entry as empty or uninitialized. In practice, you may want to have -// reserved keys or values to get around this limitation. For example, in -// tcmalloc's PageID-to-sizeclass cache, a value of 0 is used as -// "unknown sizeclass." -// -// Usage Considerations -// -------------------- -// -// kHashbits controls the size of the cache. The best value for -// kHashbits will of course depend on the application. Perhaps try -// tuning the value of kHashbits by measuring different values on your -// favorite benchmark. Also remember not to be a pig; other -// programs that need resources may suffer if you are. -// -// The main uses for this class will be when performance is -// critical and there's a convenient type to hold the cache's -// entries. As described above, the number of bits required -// for a cache entry is (kKeybits - kHashbits) + kValuebits. Suppose -// kKeybits + kValuebits is 43. Then it probably makes sense to -// chose kHashbits >= 11 so that cache entries fit in a uint32. -// -// On the other hand, suppose kKeybits = kValuebits = 64. Then -// using this class may be less worthwhile. You'll probably -// be using 128 bits for each entry anyway, so maybe just pick -// a hash function, H, and use an array indexed by H(key): -// void Put(K key, V value) { a_[H(key)] = pair(key, value); } -// V GetOrDefault(K key, V default) { const pair &p = a_[H(key)]; ... } -// etc. -// -// Further Details -// --------------- -// -// For caches used only by one thread, the following is true: -// 1. For a cache c, -// (c.Put(key, value), c.GetOrDefault(key, 0)) == value -// and -// (c.Put(key, value), <...>, c.GetOrDefault(key, 0)) == value -// if the elided code contains no c.Put calls. -// -// 2. Has(key) will return false if no pair with that key -// has ever been Put. However, a newly initialized cache will have -// some pairs already present. When you create a new -// cache, you must specify an "initial value." The initialization -// procedure is equivalent to Clear(initial_value), which is -// equivalent to Put(k, initial_value) for all keys k from 0 to -// 2^kHashbits - 1. -// -// 3. If key and key' differ then the only way Put(key, value) may -// cause Has(key') to change is that Has(key') may change from true to -// false. Furthermore, a Put() call that doesn't change Has(key') -// doesn't change GetOrDefault(key', ...) either. -// -// Implementation details: -// -// This is a direct-mapped cache with 2^kHashbits entries; the hash -// function simply takes the low bits of the key. We store whole keys -// if a whole key plus a whole value fits in an entry. Otherwise, an -// entry is the high bits of a key and a value, packed together. -// E.g., a 20 bit key and a 7 bit value only require a uint16 for each -// entry if kHashbits >= 11. -// -// Alternatives to this scheme will be added as needed. - -#ifndef TCMALLOC_PACKED_CACHE_INL_H_ -#define TCMALLOC_PACKED_CACHE_INL_H_ - -#include "config.h" -#include // for size_t -#ifdef HAVE_STDINT_H -#include // for uintptr_t -#endif -#include "base/basictypes.h" -#include "internal_logging.h" - -// A safe way of doing "(1 << n) - 1" -- without worrying about overflow -// Note this will all be resolved to a constant expression at compile-time -#define N_ONES_(IntType, N) \ - ( (N) == 0 ? 0 : ((static_cast(1) << ((N)-1))-1 + \ - (static_cast(1) << ((N)-1))) ) - -// The types K and V provide upper bounds on the number of valid keys -// and values, but we explicitly require the keys to be less than -// 2^kKeybits and the values to be less than 2^kValuebits. The size of -// the table is controlled by kHashbits, and the type of each entry in -// the cache is T. See also the big comment at the top of the file. -template -class PackedCache { - public: - typedef uintptr_t K; - typedef size_t V; -#ifdef TCMALLOC_SMALL_BUT_SLOW - // Decrease the size map cache if running in the small memory mode. - static const int kHashbits = 12; -#else - static const int kHashbits = 16; -#endif - static const int kValuebits = 7; - static const bool kUseWholeKeys = kKeybits + kValuebits <= 8 * sizeof(T); - - explicit PackedCache(V initial_value) { - COMPILE_ASSERT(kKeybits <= sizeof(K) * 8, key_size); - COMPILE_ASSERT(kValuebits <= sizeof(V) * 8, value_size); - COMPILE_ASSERT(kHashbits <= kKeybits, hash_function); - COMPILE_ASSERT(kKeybits - kHashbits + kValuebits <= kTbits, - entry_size_must_be_big_enough); - Clear(initial_value); - } - - void Put(K key, V value) { - ASSERT(key == (key & kKeyMask)); - ASSERT(value == (value & kValueMask)); - array_[Hash(key)] = KeyToUpper(key) | value; - } - - bool Has(K key) const { - ASSERT(key == (key & kKeyMask)); - return KeyMatch(array_[Hash(key)], key); - } - - V GetOrDefault(K key, V default_value) const { - // As with other code in this class, we touch array_ as few times - // as we can. Assuming entries are read atomically (e.g., their - // type is uintptr_t on most hardware) then certain races are - // harmless. - ASSERT(key == (key & kKeyMask)); - T entry = array_[Hash(key)]; - return KeyMatch(entry, key) ? EntryToValue(entry) : default_value; - } - - void Clear(V value) { - ASSERT(value == (value & kValueMask)); - for (int i = 0; i < 1 << kHashbits; i++) { - ASSERT(kUseWholeKeys || KeyToUpper(i) == 0); - array_[i] = kUseWholeKeys ? (value | KeyToUpper(i)) : value; - } - } - - private: - // We are going to pack a value and the upper part of a key (or a - // whole key) into an entry of type T. The UPPER type is for the - // upper part of a key, after the key has been masked and shifted - // for inclusion in an entry. - typedef T UPPER; - - static V EntryToValue(T t) { return t & kValueMask; } - - // If we have space for a whole key, we just shift it left. - // Otherwise kHashbits determines where in a K to find the upper - // part of the key, and kValuebits determines where in the entry to - // put it. - static UPPER KeyToUpper(K k) { - if (kUseWholeKeys) { - return static_cast(k) << kValuebits; - } else { - const int shift = kHashbits - kValuebits; - // Assume kHashbits >= kValuebits. It'd be easy to lift this assumption. - return static_cast(k >> shift) & kUpperMask; - } - } - - static size_t Hash(K key) { - return static_cast(key) & N_ONES_(size_t, kHashbits); - } - - // Does the entry match the relevant part of the given key? - static bool KeyMatch(T entry, K key) { - return kUseWholeKeys ? - (entry >> kValuebits == key) : - ((KeyToUpper(key) ^ entry) & kUpperMask) == 0; - } - - static const int kTbits = 8 * sizeof(T); - static const int kUpperbits = kUseWholeKeys ? kKeybits : kKeybits - kHashbits; - - // For masking a K. - static const K kKeyMask = N_ONES_(K, kKeybits); - - // For masking a T. - static const T kUpperMask = N_ONES_(T, kUpperbits) << kValuebits; - - // For masking a V or a T. - static const V kValueMask = N_ONES_(V, kValuebits); - - // array_ is the cache. Its elements are volatile because any - // thread can write any array element at any time. - volatile T array_[1 << kHashbits]; -}; - -#undef N_ONES_ - -#endif // TCMALLOC_PACKED_CACHE_INL_H_ diff --git a/contrib/libtcmalloc/src/page_heap.cc b/contrib/libtcmalloc/src/page_heap.cc deleted file mode 100644 index f1915623308..00000000000 --- a/contrib/libtcmalloc/src/page_heap.cc +++ /dev/null @@ -1,682 +0,0 @@ -// -*- Mode: C++; c-basic-offset: 2; indent-tabs-mode: nil -*- -// Copyright (c) 2008, Google Inc. -// All rights reserved. -// -// Redistribution and use in source and binary forms, with or without -// modification, are permitted provided that the following conditions are -// met: -// -// * Redistributions of source code must retain the above copyright -// notice, this list of conditions and the following disclaimer. -// * Redistributions in binary form must reproduce the above -// copyright notice, this list of conditions and the following disclaimer -// in the documentation and/or other materials provided with the -// distribution. -// * Neither the name of Google Inc. nor the names of its -// contributors may be used to endorse or promote products derived from -// this software without specific prior written permission. -// -// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS -// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT -// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR -// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT -// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, -// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT -// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, -// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY -// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT -// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE -// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. - -// --- -// Author: Sanjay Ghemawat - -#include "config.h" -#ifdef HAVE_INTTYPES_H -#include // for PRIuPTR -#endif -#include // for ENOMEM, errno -#include // for MallocRange, etc -#include "base/basictypes.h" -#include "base/commandlineflags.h" -#include "internal_logging.h" // for ASSERT, TCMalloc_Printer, etc -#include "page_heap_allocator.h" // for PageHeapAllocator -#include "static_vars.h" // for Static -#include "system-alloc.h" // for TCMalloc_SystemAlloc, etc - -DEFINE_double(tcmalloc_release_rate, - EnvToDouble("TCMALLOC_RELEASE_RATE", 1.0), - "Rate at which we release unused memory to the system. " - "Zero means we never release memory back to the system. " - "Increase this flag to return memory faster; decrease it " - "to return memory slower. Reasonable rates are in the " - "range [0,10]"); - -DEFINE_int64(tcmalloc_heap_limit_mb, - EnvToInt("TCMALLOC_HEAP_LIMIT_MB", 0), - "Limit total size of the process heap to the " - "specified number of MiB. " - "When we approach the limit the memory is released " - "to the system more aggressively (more minor page faults). " - "Zero means to allocate as long as system allows."); - -namespace tcmalloc { - -PageHeap::PageHeap() - : pagemap_(MetaDataAlloc), - pagemap_cache_(0), - scavenge_counter_(0), - // Start scavenging at kMaxPages list - release_index_(kMaxPages), - aggressive_decommit_(false) { - COMPILE_ASSERT(kNumClasses <= (1 << PageMapCache::kValuebits), valuebits); - DLL_Init(&large_.normal); - DLL_Init(&large_.returned); - for (int i = 0; i < kMaxPages; i++) { - DLL_Init(&free_[i].normal); - DLL_Init(&free_[i].returned); - } -} - -Span* PageHeap::SearchFreeAndLargeLists(Length n) { - ASSERT(Check()); - ASSERT(n > 0); - - // Find first size >= n that has a non-empty list - for (Length s = n; s < kMaxPages; s++) { - Span* ll = &free_[s].normal; - // If we're lucky, ll is non-empty, meaning it has a suitable span. - if (!DLL_IsEmpty(ll)) { - ASSERT(ll->next->location == Span::ON_NORMAL_FREELIST); - return Carve(ll->next, n); - } - // Alternatively, maybe there's a usable returned span. - ll = &free_[s].returned; - if (!DLL_IsEmpty(ll)) { - // We did not call EnsureLimit before, to avoid releasing the span - // that will be taken immediately back. - // Calling EnsureLimit here is not very expensive, as it fails only if - // there is no more normal spans (and it fails efficiently) - // or SystemRelease does not work (there is probably no returned spans). - if (EnsureLimit(n)) { - // ll may have became empty due to coalescing - if (!DLL_IsEmpty(ll)) { - ASSERT(ll->next->location == Span::ON_RETURNED_FREELIST); - return Carve(ll->next, n); - } - } - } - } - // No luck in free lists, our last chance is in a larger class. - return AllocLarge(n); // May be NULL -} - -static const size_t kForcedCoalesceInterval = 128*1024*1024; - -Span* PageHeap::New(Length n) { - ASSERT(Check()); - ASSERT(n > 0); - - Span* result = SearchFreeAndLargeLists(n); - if (result != NULL) - return result; - - if (stats_.free_bytes != 0 && stats_.unmapped_bytes != 0 - && stats_.free_bytes + stats_.unmapped_bytes >= stats_.system_bytes / 4 - && (stats_.system_bytes / kForcedCoalesceInterval - != (stats_.system_bytes + (n << kPageShift)) / kForcedCoalesceInterval)) { - // We're about to grow heap, but there are lots of free pages. - // tcmalloc's design decision to keep unmapped and free spans - // separately and never coalesce them means that sometimes there - // can be free pages span of sufficient size, but it consists of - // "segments" of different type so page heap search cannot find - // it. In order to prevent growing heap and wasting memory in such - // case we're going to unmap all free pages. So that all free - // spans are maximally coalesced. - // - // We're also limiting 'rate' of going into this path to be at - // most once per 128 megs of heap growth. Otherwise programs that - // grow heap frequently (and that means by small amount) could be - // penalized with higher count of minor page faults. - // - // See also large_heap_fragmentation_unittest.cc and - // https://code.google.com/p/gperftools/issues/detail?id=368 - ReleaseAtLeastNPages(static_cast(0x7fffffff)); - - // then try again. If we are forced to grow heap because of large - // spans fragmentation and not because of problem described above, - // then at the very least we've just unmapped free but - // insufficiently big large spans back to OS. So in case of really - // unlucky memory fragmentation we'll be consuming virtual address - // space, but not real memory - result = SearchFreeAndLargeLists(n); - if (result != NULL) return result; - } - - // Grow the heap and try again. - if (!GrowHeap(n)) { - ASSERT(stats_.unmapped_bytes+ stats_.committed_bytes==stats_.system_bytes); - ASSERT(Check()); - // underlying SysAllocator likely set ENOMEM but we can get here - // due to EnsureLimit so we set it here too. - // - // Setting errno to ENOMEM here allows us to avoid dealing with it - // in fast-path. - errno = ENOMEM; - return NULL; - } - return SearchFreeAndLargeLists(n); -} - -Span* PageHeap::AllocLarge(Length n) { - // find the best span (closest to n in size). - // The following loops implements address-ordered best-fit. - Span *best = NULL; - - // Search through normal list - for (Span* span = large_.normal.next; - span != &large_.normal; - span = span->next) { - if (span->length >= n) { - if ((best == NULL) - || (span->length < best->length) - || ((span->length == best->length) && (span->start < best->start))) { - best = span; - ASSERT(best->location == Span::ON_NORMAL_FREELIST); - } - } - } - - Span *bestNormal = best; - - // Search through released list in case it has a better fit - for (Span* span = large_.returned.next; - span != &large_.returned; - span = span->next) { - if (span->length >= n) { - if ((best == NULL) - || (span->length < best->length) - || ((span->length == best->length) && (span->start < best->start))) { - best = span; - ASSERT(best->location == Span::ON_RETURNED_FREELIST); - } - } - } - - if (best == bestNormal) { - return best == NULL ? NULL : Carve(best, n); - } - - // best comes from returned list. - - if (EnsureLimit(n, false)) { - return Carve(best, n); - } - - if (EnsureLimit(n, true)) { - // best could have been destroyed by coalescing. - // bestNormal is not a best-fit, and it could be destroyed as well. - // We retry, the limit is already ensured: - return AllocLarge(n); - } - - // If bestNormal existed, EnsureLimit would succeeded: - ASSERT(bestNormal == NULL); - // We are not allowed to take best from returned list. - return NULL; -} - -Span* PageHeap::Split(Span* span, Length n) { - ASSERT(0 < n); - ASSERT(n < span->length); - ASSERT(span->location == Span::IN_USE); - ASSERT(span->sizeclass == 0); - Event(span, 'T', n); - - const int extra = span->length - n; - Span* leftover = NewSpan(span->start + n, extra); - ASSERT(leftover->location == Span::IN_USE); - Event(leftover, 'U', extra); - RecordSpan(leftover); - pagemap_.set(span->start + n - 1, span); // Update map from pageid to span - span->length = n; - - return leftover; -} - -void PageHeap::CommitSpan(Span* span) { - TCMalloc_SystemCommit(reinterpret_cast(span->start << kPageShift), - static_cast(span->length << kPageShift)); - stats_.committed_bytes += span->length << kPageShift; -} - -bool PageHeap::DecommitSpan(Span* span) { - bool rv = TCMalloc_SystemRelease(reinterpret_cast(span->start << kPageShift), - static_cast(span->length << kPageShift)); - if (rv) { - stats_.committed_bytes -= span->length << kPageShift; - } - - return rv; -} - -Span* PageHeap::Carve(Span* span, Length n) { - ASSERT(n > 0); - ASSERT(span->location != Span::IN_USE); - const int old_location = span->location; - RemoveFromFreeList(span); - span->location = Span::IN_USE; - Event(span, 'A', n); - - const int extra = span->length - n; - ASSERT(extra >= 0); - if (extra > 0) { - Span* leftover = NewSpan(span->start + n, extra); - leftover->location = old_location; - Event(leftover, 'S', extra); - RecordSpan(leftover); - - // The previous span of |leftover| was just splitted -- no need to - // coalesce them. The next span of |leftover| was not previously coalesced - // with |span|, i.e. is NULL or has got location other than |old_location|. -#ifndef NDEBUG - const PageID p = leftover->start; - const Length len = leftover->length; - Span* next = GetDescriptor(p+len); - ASSERT (next == NULL || - next->location == Span::IN_USE || - next->location != leftover->location); -#endif - - PrependToFreeList(leftover); // Skip coalescing - no candidates possible - span->length = n; - pagemap_.set(span->start + n - 1, span); - } - ASSERT(Check()); - if (old_location == Span::ON_RETURNED_FREELIST) { - // We need to recommit this address space. - CommitSpan(span); - } - ASSERT(span->location == Span::IN_USE); - ASSERT(span->length == n); - ASSERT(stats_.unmapped_bytes+ stats_.committed_bytes==stats_.system_bytes); - return span; -} - -void PageHeap::Delete(Span* span) { - ASSERT(Check()); - ASSERT(span->location == Span::IN_USE); - ASSERT(span->length > 0); - ASSERT(GetDescriptor(span->start) == span); - ASSERT(GetDescriptor(span->start + span->length - 1) == span); - const Length n = span->length; - span->sizeclass = 0; - span->sample = 0; - span->location = Span::ON_NORMAL_FREELIST; - Event(span, 'D', span->length); - MergeIntoFreeList(span); // Coalesces if possible - IncrementalScavenge(n); - ASSERT(stats_.unmapped_bytes+ stats_.committed_bytes==stats_.system_bytes); - ASSERT(Check()); -} - -bool PageHeap::MayMergeSpans(Span *span, Span *other) { - if (aggressive_decommit_) { - return other->location != Span::IN_USE; - } - return span->location == other->location; -} - -void PageHeap::MergeIntoFreeList(Span* span) { - ASSERT(span->location != Span::IN_USE); - - // Coalesce -- we guarantee that "p" != 0, so no bounds checking - // necessary. We do not bother resetting the stale pagemap - // entries for the pieces we are merging together because we only - // care about the pagemap entries for the boundaries. - // - // Note: depending on aggressive_decommit_ mode we allow only - // similar spans to be coalesced. - // - // The following applies if aggressive_decommit_ is enabled: - // - // Note that the adjacent spans we merge into "span" may come out of a - // "normal" (committed) list, and cleanly merge with our IN_USE span, which - // is implicitly committed. If the adjacents spans are on the "returned" - // (decommitted) list, then we must get both spans into the same state before - // or after we coalesce them. The current code always decomits. This is - // achieved by blindly decommitting the entire coalesced region, which may - // include any combination of committed and decommitted spans, at the end of - // the method. - - // TODO(jar): "Always decommit" causes some extra calls to commit when we are - // called in GrowHeap() during an allocation :-/. We need to eval the cost of - // that oscillation, and possibly do something to reduce it. - - // TODO(jar): We need a better strategy for deciding to commit, or decommit, - // based on memory usage and free heap sizes. - - uint64_t temp_committed = 0; - - const PageID p = span->start; - const Length n = span->length; - Span* prev = GetDescriptor(p-1); - if (prev != NULL && MayMergeSpans(span, prev)) { - // Merge preceding span into this span - ASSERT(prev->start + prev->length == p); - const Length len = prev->length; - if (aggressive_decommit_ && prev->location == Span::ON_RETURNED_FREELIST) { - // We're about to put the merge span into the returned freelist and call - // DecommitSpan() on it, which will mark the entire span including this - // one as released and decrease stats_.committed_bytes by the size of the - // merged span. To make the math work out we temporarily increase the - // stats_.committed_bytes amount. - temp_committed = prev->length << kPageShift; - } - RemoveFromFreeList(prev); - DeleteSpan(prev); - span->start -= len; - span->length += len; - pagemap_.set(span->start, span); - Event(span, 'L', len); - } - Span* next = GetDescriptor(p+n); - if (next != NULL && MayMergeSpans(span, next)) { - // Merge next span into this span - ASSERT(next->start == p+n); - const Length len = next->length; - if (aggressive_decommit_ && next->location == Span::ON_RETURNED_FREELIST) { - // See the comment below 'if (prev->location ...' for explanation. - temp_committed += next->length << kPageShift; - } - RemoveFromFreeList(next); - DeleteSpan(next); - span->length += len; - pagemap_.set(span->start + span->length - 1, span); - Event(span, 'R', len); - } - - if (aggressive_decommit_) { - if (DecommitSpan(span)) { - span->location = Span::ON_RETURNED_FREELIST; - stats_.committed_bytes += temp_committed; - } else { - ASSERT(temp_committed == 0); - } - } - PrependToFreeList(span); -} - -void PageHeap::PrependToFreeList(Span* span) { - ASSERT(span->location != Span::IN_USE); - SpanList* list = (span->length < kMaxPages) ? &free_[span->length] : &large_; - if (span->location == Span::ON_NORMAL_FREELIST) { - stats_.free_bytes += (span->length << kPageShift); - DLL_Prepend(&list->normal, span); - } else { - stats_.unmapped_bytes += (span->length << kPageShift); - DLL_Prepend(&list->returned, span); - } -} - -void PageHeap::RemoveFromFreeList(Span* span) { - ASSERT(span->location != Span::IN_USE); - if (span->location == Span::ON_NORMAL_FREELIST) { - stats_.free_bytes -= (span->length << kPageShift); - } else { - stats_.unmapped_bytes -= (span->length << kPageShift); - } - DLL_Remove(span); -} - -void PageHeap::IncrementalScavenge(Length n) { - // Fast path; not yet time to release memory - scavenge_counter_ -= n; - if (scavenge_counter_ >= 0) return; // Not yet time to scavenge - - const double rate = FLAGS_tcmalloc_release_rate; - if (rate <= 1e-6) { - // Tiny release rate means that releasing is disabled. - scavenge_counter_ = kDefaultReleaseDelay; - return; - } - - Length released_pages = ReleaseAtLeastNPages(1); - - if (released_pages == 0) { - // Nothing to scavenge, delay for a while. - scavenge_counter_ = kDefaultReleaseDelay; - } else { - // Compute how long to wait until we return memory. - // FLAGS_tcmalloc_release_rate==1 means wait for 1000 pages - // after releasing one page. - const double mult = 1000.0 / rate; - double wait = mult * static_cast(released_pages); - if (wait > kMaxReleaseDelay) { - // Avoid overflow and bound to reasonable range. - wait = kMaxReleaseDelay; - } - scavenge_counter_ = static_cast(wait); - } -} - -Length PageHeap::ReleaseLastNormalSpan(SpanList* slist) { - Span* s = slist->normal.prev; - ASSERT(s->location == Span::ON_NORMAL_FREELIST); - - if (DecommitSpan(s)) { - RemoveFromFreeList(s); - const Length n = s->length; - s->location = Span::ON_RETURNED_FREELIST; - MergeIntoFreeList(s); // Coalesces if possible. - return n; - } - - return 0; -} - -Length PageHeap::ReleaseAtLeastNPages(Length num_pages) { - Length released_pages = 0; - - // Round robin through the lists of free spans, releasing the last - // span in each list. Stop after releasing at least num_pages - // or when there is nothing more to release. - while (released_pages < num_pages && stats_.free_bytes > 0) { - for (int i = 0; i < kMaxPages+1 && released_pages < num_pages; - i++, release_index_++) { - if (release_index_ > kMaxPages) release_index_ = 0; - SpanList* slist = (release_index_ == kMaxPages) ? - &large_ : &free_[release_index_]; - if (!DLL_IsEmpty(&slist->normal)) { - Length released_len = ReleaseLastNormalSpan(slist); - // Some systems do not support release - if (released_len == 0) return released_pages; - released_pages += released_len; - } - } - } - return released_pages; -} - -bool PageHeap::EnsureLimit(Length n, bool withRelease) -{ - Length limit = (FLAGS_tcmalloc_heap_limit_mb*1024*1024) >> kPageShift; - if (limit == 0) return true; //there is no limit - - // We do not use stats_.system_bytes because it does not take - // MetaDataAllocs into account. - Length takenPages = TCMalloc_SystemTaken >> kPageShift; - //XXX takenPages may be slightly bigger than limit for two reasons: - //* MetaDataAllocs ignore the limit (it is not easy to handle - // out of memory there) - //* sys_alloc may round allocation up to huge page size, - // although smaller limit was ensured - - ASSERT(takenPages >= stats_.unmapped_bytes >> kPageShift); - takenPages -= stats_.unmapped_bytes >> kPageShift; - - if (takenPages + n > limit && withRelease) { - takenPages -= ReleaseAtLeastNPages(takenPages + n - limit); - } - - return takenPages + n <= limit; -} - -void PageHeap::RegisterSizeClass(Span* span, size_t sc) { - // Associate span object with all interior pages as well - ASSERT(span->location == Span::IN_USE); - ASSERT(GetDescriptor(span->start) == span); - ASSERT(GetDescriptor(span->start+span->length-1) == span); - Event(span, 'C', sc); - span->sizeclass = sc; - for (Length i = 1; i < span->length-1; i++) { - pagemap_.set(span->start+i, span); - } -} - -void PageHeap::GetSmallSpanStats(SmallSpanStats* result) { - for (int s = 0; s < kMaxPages; s++) { - result->normal_length[s] = DLL_Length(&free_[s].normal); - result->returned_length[s] = DLL_Length(&free_[s].returned); - } -} - -void PageHeap::GetLargeSpanStats(LargeSpanStats* result) { - result->spans = 0; - result->normal_pages = 0; - result->returned_pages = 0; - for (Span* s = large_.normal.next; s != &large_.normal; s = s->next) { - result->normal_pages += s->length;; - result->spans++; - } - for (Span* s = large_.returned.next; s != &large_.returned; s = s->next) { - result->returned_pages += s->length; - result->spans++; - } -} - -bool PageHeap::GetNextRange(PageID start, base::MallocRange* r) { - Span* span = reinterpret_cast(pagemap_.Next(start)); - if (span == NULL) { - return false; - } - r->address = span->start << kPageShift; - r->length = span->length << kPageShift; - r->fraction = 0; - switch (span->location) { - case Span::IN_USE: - r->type = base::MallocRange::INUSE; - r->fraction = 1; - if (span->sizeclass > 0) { - // Only some of the objects in this span may be in use. - const size_t osize = Static::sizemap()->class_to_size(span->sizeclass); - r->fraction = (1.0 * osize * span->refcount) / r->length; - } - break; - case Span::ON_NORMAL_FREELIST: - r->type = base::MallocRange::FREE; - break; - case Span::ON_RETURNED_FREELIST: - r->type = base::MallocRange::UNMAPPED; - break; - default: - r->type = base::MallocRange::UNKNOWN; - break; - } - return true; -} - -static void RecordGrowth(size_t growth) { - StackTrace* t = Static::stacktrace_allocator()->New(); - t->depth = GetStackTrace(t->stack, kMaxStackDepth-1, 3); - t->size = growth; - t->stack[kMaxStackDepth-1] = reinterpret_cast(Static::growth_stacks()); - Static::set_growth_stacks(t); -} - -bool PageHeap::GrowHeap(Length n) { - ASSERT(kMaxPages >= kMinSystemAlloc); - if (n > kMaxValidPages) return false; - Length ask = (n>kMinSystemAlloc) ? n : static_cast(kMinSystemAlloc); - size_t actual_size; - void* ptr = NULL; - if (EnsureLimit(ask)) { - ptr = TCMalloc_SystemAlloc(ask << kPageShift, &actual_size, kPageSize); - } - if (ptr == NULL) { - if (n < ask) { - // Try growing just "n" pages - ask = n; - if (EnsureLimit(ask)) { - ptr = TCMalloc_SystemAlloc(ask << kPageShift, &actual_size, kPageSize); - } - } - if (ptr == NULL) return false; - } - ask = actual_size >> kPageShift; - RecordGrowth(ask << kPageShift); - - uint64_t old_system_bytes = stats_.system_bytes; - stats_.system_bytes += (ask << kPageShift); - stats_.committed_bytes += (ask << kPageShift); - const PageID p = reinterpret_cast(ptr) >> kPageShift; - ASSERT(p > 0); - - // If we have already a lot of pages allocated, just pre allocate a bunch of - // memory for the page map. This prevents fragmentation by pagemap metadata - // when a program keeps allocating and freeing large blocks. - - if (old_system_bytes < kPageMapBigAllocationThreshold - && stats_.system_bytes >= kPageMapBigAllocationThreshold) { - pagemap_.PreallocateMoreMemory(); - } - - // Make sure pagemap_ has entries for all of the new pages. - // Plus ensure one before and one after so coalescing code - // does not need bounds-checking. - if (pagemap_.Ensure(p-1, ask+2)) { - // Pretend the new area is allocated and then Delete() it to cause - // any necessary coalescing to occur. - Span* span = NewSpan(p, ask); - RecordSpan(span); - Delete(span); - ASSERT(stats_.unmapped_bytes+ stats_.committed_bytes==stats_.system_bytes); - ASSERT(Check()); - return true; - } else { - // We could not allocate memory within "pagemap_" - // TODO: Once we can return memory to the system, return the new span - return false; - } -} - -bool PageHeap::Check() { - ASSERT(free_[0].normal.next == &free_[0].normal); - ASSERT(free_[0].returned.next == &free_[0].returned); - return true; -} - -bool PageHeap::CheckExpensive() { - bool result = Check(); - CheckList(&large_.normal, kMaxPages, 1000000000, Span::ON_NORMAL_FREELIST); - CheckList(&large_.returned, kMaxPages, 1000000000, Span::ON_RETURNED_FREELIST); - for (Length s = 1; s < kMaxPages; s++) { - CheckList(&free_[s].normal, s, s, Span::ON_NORMAL_FREELIST); - CheckList(&free_[s].returned, s, s, Span::ON_RETURNED_FREELIST); - } - return result; -} - -bool PageHeap::CheckList(Span* list, Length min_pages, Length max_pages, - int freelist) { - for (Span* s = list->next; s != list; s = s->next) { - CHECK_CONDITION(s->location == freelist); // NORMAL or RETURNED - CHECK_CONDITION(s->length >= min_pages); - CHECK_CONDITION(s->length <= max_pages); - CHECK_CONDITION(GetDescriptor(s->start) == s); - CHECK_CONDITION(GetDescriptor(s->start+s->length-1) == s); - } - return true; -} - -} // namespace tcmalloc diff --git a/contrib/libtcmalloc/src/page_heap.h b/contrib/libtcmalloc/src/page_heap.h deleted file mode 100644 index 89fab81da69..00000000000 --- a/contrib/libtcmalloc/src/page_heap.h +++ /dev/null @@ -1,316 +0,0 @@ -// -*- Mode: C++; c-basic-offset: 2; indent-tabs-mode: nil -*- -// Copyright (c) 2008, Google Inc. -// All rights reserved. -// -// Redistribution and use in source and binary forms, with or without -// modification, are permitted provided that the following conditions are -// met: -// -// * Redistributions of source code must retain the above copyright -// notice, this list of conditions and the following disclaimer. -// * Redistributions in binary form must reproduce the above -// copyright notice, this list of conditions and the following disclaimer -// in the documentation and/or other materials provided with the -// distribution. -// * Neither the name of Google Inc. nor the names of its -// contributors may be used to endorse or promote products derived from -// this software without specific prior written permission. -// -// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS -// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT -// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR -// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT -// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, -// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT -// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, -// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY -// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT -// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE -// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. - -// --- -// Author: Sanjay Ghemawat - -#ifndef TCMALLOC_PAGE_HEAP_H_ -#define TCMALLOC_PAGE_HEAP_H_ - -#include "config.h" -#include // for size_t -#ifdef HAVE_STDINT_H -#include // for uint64_t, int64_t, uint16_t -#endif -#include -#include "base/basictypes.h" -#include "common.h" -#include "packed-cache-inl.h" -#include "pagemap.h" -#include "span.h" - -// We need to dllexport PageHeap just for the unittest. MSVC complains -// that we don't dllexport the PageHeap members, but we don't need to -// test those, so I just suppress this warning. -#ifdef _MSC_VER -#pragma warning(push) -#pragma warning(disable:4251) -#endif - -// This #ifdef should almost never be set. Set NO_TCMALLOC_SAMPLES if -// you're porting to a system where you really can't get a stacktrace. -// Because we control the definition of GetStackTrace, all clients of -// GetStackTrace should #include us rather than stacktrace.h. -#ifdef NO_TCMALLOC_SAMPLES - // We use #define so code compiles even if you #include stacktrace.h somehow. -# define GetStackTrace(stack, depth, skip) (0) -#else -# include -#endif - -namespace base { -struct MallocRange; -} - -namespace tcmalloc { - -// ------------------------------------------------------------------------- -// Map from page-id to per-page data -// ------------------------------------------------------------------------- - -// We use PageMap2<> for 32-bit and PageMap3<> for 64-bit machines. -// We also use a simple one-level cache for hot PageID-to-sizeclass mappings, -// because sometimes the sizeclass is all the information we need. - -// Selector class -- general selector uses 3-level map -template class MapSelector { - public: - typedef TCMalloc_PageMap3 Type; - typedef PackedCache CacheType; -}; - -// A two-level map for 32-bit machines -template <> class MapSelector<32> { - public: - typedef TCMalloc_PageMap2<32-kPageShift> Type; - typedef PackedCache<32-kPageShift, uint16_t> CacheType; -}; - -// ------------------------------------------------------------------------- -// Page-level allocator -// * Eager coalescing -// -// Heap for page-level allocation. We allow allocating and freeing a -// contiguous runs of pages (called a "span"). -// ------------------------------------------------------------------------- - -class PERFTOOLS_DLL_DECL PageHeap { - public: - PageHeap(); - - // Allocate a run of "n" pages. Returns zero if out of memory. - // Caller should not pass "n == 0" -- instead, n should have - // been rounded up already. - Span* New(Length n); - - // Delete the span "[p, p+n-1]". - // REQUIRES: span was returned by earlier call to New() and - // has not yet been deleted. - void Delete(Span* span); - - // Mark an allocated span as being used for small objects of the - // specified size-class. - // REQUIRES: span was returned by an earlier call to New() - // and has not yet been deleted. - void RegisterSizeClass(Span* span, size_t sc); - - // Split an allocated span into two spans: one of length "n" pages - // followed by another span of length "span->length - n" pages. - // Modifies "*span" to point to the first span of length "n" pages. - // Returns a pointer to the second span. - // - // REQUIRES: "0 < n < span->length" - // REQUIRES: span->location == IN_USE - // REQUIRES: span->sizeclass == 0 - Span* Split(Span* span, Length n); - - // Return the descriptor for the specified page. Returns NULL if - // this PageID was not allocated previously. - inline Span* GetDescriptor(PageID p) const { - return reinterpret_cast(pagemap_.get(p)); - } - - // If this page heap is managing a range with starting page # >= start, - // store info about the range in *r and return true. Else return false. - bool GetNextRange(PageID start, base::MallocRange* r); - - // Page heap statistics - struct Stats { - Stats() : system_bytes(0), free_bytes(0), unmapped_bytes(0), committed_bytes(0) {} - uint64_t system_bytes; // Total bytes allocated from system - uint64_t free_bytes; // Total bytes on normal freelists - uint64_t unmapped_bytes; // Total bytes on returned freelists - uint64_t committed_bytes; // Bytes committed, always <= system_bytes_. - - }; - inline Stats stats() const { return stats_; } - - struct SmallSpanStats { - // For each free list of small spans, the length (in spans) of the - // normal and returned free lists for that size. - int64 normal_length[kMaxPages]; - int64 returned_length[kMaxPages]; - }; - void GetSmallSpanStats(SmallSpanStats* result); - - // Stats for free large spans (i.e., spans with more than kMaxPages pages). - struct LargeSpanStats { - int64 spans; // Number of such spans - int64 normal_pages; // Combined page length of normal large spans - int64 returned_pages; // Combined page length of unmapped spans - }; - void GetLargeSpanStats(LargeSpanStats* result); - - bool Check(); - // Like Check() but does some more comprehensive checking. - bool CheckExpensive(); - bool CheckList(Span* list, Length min_pages, Length max_pages, - int freelist); // ON_NORMAL_FREELIST or ON_RETURNED_FREELIST - - // Try to release at least num_pages for reuse by the OS. Returns - // the actual number of pages released, which may be less than - // num_pages if there weren't enough pages to release. The result - // may also be larger than num_pages since page_heap might decide to - // release one large range instead of fragmenting it into two - // smaller released and unreleased ranges. - Length ReleaseAtLeastNPages(Length num_pages); - - // Return 0 if we have no information, or else the correct sizeclass for p. - // Reads and writes to pagemap_cache_ do not require locking. - // The entries are 64 bits on 64-bit hardware and 16 bits on - // 32-bit hardware, and we don't mind raciness as long as each read of - // an entry yields a valid entry, not a partially updated entry. - size_t GetSizeClassIfCached(PageID p) const { - return pagemap_cache_.GetOrDefault(p, 0); - } - void CacheSizeClass(PageID p, size_t cl) const { pagemap_cache_.Put(p, cl); } - - bool GetAggressiveDecommit(void) {return aggressive_decommit_;} - void SetAggressiveDecommit(bool aggressive_decommit) { - aggressive_decommit_ = aggressive_decommit; - } - - private: - // Allocates a big block of memory for the pagemap once we reach more than - // 128MB - static const size_t kPageMapBigAllocationThreshold = 128 << 20; - - // Minimum number of pages to fetch from system at a time. Must be - // significantly bigger than kBlockSize to amortize system-call - // overhead, and also to reduce external fragementation. Also, we - // should keep this value big because various incarnations of Linux - // have small limits on the number of mmap() regions per - // address-space. - // REQUIRED: kMinSystemAlloc <= kMaxPages; - static const int kMinSystemAlloc = kMaxPages; - - // Never delay scavenging for more than the following number of - // deallocated pages. With 4K pages, this comes to 4GB of - // deallocation. - static const int kMaxReleaseDelay = 1 << 20; - - // If there is nothing to release, wait for so many pages before - // scavenging again. With 4K pages, this comes to 1GB of memory. - static const int kDefaultReleaseDelay = 1 << 18; - - // Pick the appropriate map and cache types based on pointer size - typedef MapSelector::Type PageMap; - typedef MapSelector::CacheType PageMapCache; - PageMap pagemap_; - mutable PageMapCache pagemap_cache_; - - // We segregate spans of a given size into two circular linked - // lists: one for normal spans, and one for spans whose memory - // has been returned to the system. - struct SpanList { - Span normal; - Span returned; - }; - - // List of free spans of length >= kMaxPages - SpanList large_; - - // Array mapping from span length to a doubly linked list of free spans - SpanList free_[kMaxPages]; - - // Statistics on system, free, and unmapped bytes - Stats stats_; - - Span* SearchFreeAndLargeLists(Length n); - - bool GrowHeap(Length n); - - // REQUIRES: span->length >= n - // REQUIRES: span->location != IN_USE - // Remove span from its free list, and move any leftover part of - // span into appropriate free lists. Also update "span" to have - // length exactly "n" and mark it as non-free so it can be returned - // to the client. After all that, decrease free_pages_ by n and - // return span. - Span* Carve(Span* span, Length n); - - void RecordSpan(Span* span) { - pagemap_.set(span->start, span); - if (span->length > 1) { - pagemap_.set(span->start + span->length - 1, span); - } - } - - // Allocate a large span of length == n. If successful, returns a - // span of exactly the specified length. Else, returns NULL. - Span* AllocLarge(Length n); - - // Coalesce span with neighboring spans if possible, prepend to - // appropriate free list, and adjust stats. - void MergeIntoFreeList(Span* span); - - // Commit the span. - void CommitSpan(Span* span); - - // Decommit the span. - bool DecommitSpan(Span* span); - - // Prepends span to appropriate free list, and adjusts stats. - void PrependToFreeList(Span* span); - - // Removes span from its free list, and adjust stats. - void RemoveFromFreeList(Span* span); - - // Incrementally release some memory to the system. - // IncrementalScavenge(n) is called whenever n pages are freed. - void IncrementalScavenge(Length n); - - // Release the last span on the normal portion of this list. - // Return the length of that span or zero if release failed. - Length ReleaseLastNormalSpan(SpanList* slist); - - // Checks if we are allowed to take more memory from the system. - // If limit is reached and allowRelease is true, tries to release - // some unused spans. - bool EnsureLimit(Length n, bool allowRelease = true); - - bool MayMergeSpans(Span *span, Span *other); - - // Number of pages to deallocate before doing more scavenging - int64_t scavenge_counter_; - - // Index of last free list where we released memory to the OS. - int release_index_; - - bool aggressive_decommit_; -}; - -} // namespace tcmalloc - -#ifdef _MSC_VER -#pragma warning(pop) -#endif - -#endif // TCMALLOC_PAGE_HEAP_H_ diff --git a/contrib/libtcmalloc/src/page_heap_allocator.h b/contrib/libtcmalloc/src/page_heap_allocator.h deleted file mode 100644 index 892d1c1abe3..00000000000 --- a/contrib/libtcmalloc/src/page_heap_allocator.h +++ /dev/null @@ -1,114 +0,0 @@ -// -*- Mode: C++; c-basic-offset: 2; indent-tabs-mode: nil -*- -// Copyright (c) 2008, Google Inc. -// All rights reserved. -// -// Redistribution and use in source and binary forms, with or without -// modification, are permitted provided that the following conditions are -// met: -// -// * Redistributions of source code must retain the above copyright -// notice, this list of conditions and the following disclaimer. -// * Redistributions in binary form must reproduce the above -// copyright notice, this list of conditions and the following disclaimer -// in the documentation and/or other materials provided with the -// distribution. -// * Neither the name of Google Inc. nor the names of its -// contributors may be used to endorse or promote products derived from -// this software without specific prior written permission. -// -// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS -// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT -// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR -// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT -// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, -// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT -// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, -// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY -// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT -// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE -// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. - -// --- -// Author: Sanjay Ghemawat - -#ifndef TCMALLOC_PAGE_HEAP_ALLOCATOR_H_ -#define TCMALLOC_PAGE_HEAP_ALLOCATOR_H_ - -#include // for NULL, size_t - -#include "common.h" // for MetaDataAlloc -#include "internal_logging.h" // for ASSERT - -namespace tcmalloc { - -// Simple allocator for objects of a specified type. External locking -// is required before accessing one of these objects. -template -class PageHeapAllocator { - public: - // We use an explicit Init function because these variables are statically - // allocated and their constructors might not have run by the time some - // other static variable tries to allocate memory. - void Init() { - ASSERT(sizeof(T) <= kAllocIncrement); - inuse_ = 0; - free_area_ = NULL; - free_avail_ = 0; - free_list_ = NULL; - // Reserve some space at the beginning to avoid fragmentation. - Delete(New()); - } - - T* New() { - // Consult free list - void* result; - if (free_list_ != NULL) { - result = free_list_; - free_list_ = *(reinterpret_cast(result)); - } else { - if (free_avail_ < sizeof(T)) { - // Need more room. We assume that MetaDataAlloc returns - // suitably aligned memory. - free_area_ = reinterpret_cast(MetaDataAlloc(kAllocIncrement)); - if (free_area_ == NULL) { - Log(kCrash, __FILE__, __LINE__, - "FATAL ERROR: Out of memory trying to allocate internal " - "tcmalloc data (bytes, object-size)", - kAllocIncrement, sizeof(T)); - } - free_avail_ = kAllocIncrement; - } - result = free_area_; - free_area_ += sizeof(T); - free_avail_ -= sizeof(T); - } - inuse_++; - return reinterpret_cast(result); - } - - void Delete(T* p) { - *(reinterpret_cast(p)) = free_list_; - free_list_ = p; - inuse_--; - } - - int inuse() const { return inuse_; } - - private: - // How much to allocate from system at a time - static const int kAllocIncrement = 128 << 10; - - // Free area from which to carve new objects - char* free_area_; - size_t free_avail_; - - // Free list of already carved objects - void* free_list_; - - // Number of allocated but unfreed objects - int inuse_; -}; - -} // namespace tcmalloc - -#endif // TCMALLOC_PAGE_HEAP_ALLOCATOR_H_ diff --git a/contrib/libtcmalloc/src/pagemap.h b/contrib/libtcmalloc/src/pagemap.h deleted file mode 100644 index dd9442313af..00000000000 --- a/contrib/libtcmalloc/src/pagemap.h +++ /dev/null @@ -1,324 +0,0 @@ -// -*- Mode: C++; c-basic-offset: 2; indent-tabs-mode: nil -*- -// Copyright (c) 2005, Google Inc. -// All rights reserved. -// -// Redistribution and use in source and binary forms, with or without -// modification, are permitted provided that the following conditions are -// met: -// -// * Redistributions of source code must retain the above copyright -// notice, this list of conditions and the following disclaimer. -// * Redistributions in binary form must reproduce the above -// copyright notice, this list of conditions and the following disclaimer -// in the documentation and/or other materials provided with the -// distribution. -// * Neither the name of Google Inc. nor the names of its -// contributors may be used to endorse or promote products derived from -// this software without specific prior written permission. -// -// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS -// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT -// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR -// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT -// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, -// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT -// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, -// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY -// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT -// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE -// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. - -// --- -// Author: Sanjay Ghemawat -// -// A data structure used by the caching malloc. It maps from page# to -// a pointer that contains info about that page. We use two -// representations: one for 32-bit addresses, and another for 64 bit -// addresses. Both representations provide the same interface. The -// first representation is implemented as a flat array, the seconds as -// a three-level radix tree that strips away approximately 1/3rd of -// the bits every time. -// -// The BITS parameter should be the number of bits required to hold -// a page number. E.g., with 32 bit pointers and 4K pages (i.e., -// page offset fits in lower 12 bits), BITS == 20. - -#ifndef TCMALLOC_PAGEMAP_H_ -#define TCMALLOC_PAGEMAP_H_ - -#include "config.h" - -#include // for NULL, size_t -#include // for memset -#if defined HAVE_STDINT_H -#include -#elif defined HAVE_INTTYPES_H -#include -#else -#include -#endif -#include "internal_logging.h" // for ASSERT - -// Single-level array -template -class TCMalloc_PageMap1 { - private: - static const int LENGTH = 1 << BITS; - - void** array_; - - public: - typedef uintptr_t Number; - - explicit TCMalloc_PageMap1(void* (*allocator)(size_t)) { - array_ = reinterpret_cast((*allocator)(sizeof(void*) << BITS)); - memset(array_, 0, sizeof(void*) << BITS); - } - - // Ensure that the map contains initialized entries "x .. x+n-1". - // Returns true if successful, false if we could not allocate memory. - bool Ensure(Number x, size_t n) { - // Nothing to do since flat array was allocated at start. All - // that's left is to check for overflow (that is, we don't want to - // ensure a number y where array_[y] would be an out-of-bounds - // access). - return n <= LENGTH - x; // an overflow-free way to do "x + n <= LENGTH" - } - - void PreallocateMoreMemory() {} - - // Return the current value for KEY. Returns NULL if not yet set, - // or if k is out of range. - void* get(Number k) const { - if ((k >> BITS) > 0) { - return NULL; - } - return array_[k]; - } - - // REQUIRES "k" is in range "[0,2^BITS-1]". - // REQUIRES "k" has been ensured before. - // - // Sets the value 'v' for key 'k'. - void set(Number k, void* v) { - array_[k] = v; - } - - // Return the first non-NULL pointer found in this map for - // a page number >= k. Returns NULL if no such number is found. - void* Next(Number k) const { - while (k < (1 << BITS)) { - if (array_[k] != NULL) return array_[k]; - k++; - } - return NULL; - } -}; - -// Two-level radix tree -template -class TCMalloc_PageMap2 { - private: - // Put 32 entries in the root and (2^BITS)/32 entries in each leaf. - static const int ROOT_BITS = 5; - static const int ROOT_LENGTH = 1 << ROOT_BITS; - - static const int LEAF_BITS = BITS - ROOT_BITS; - static const int LEAF_LENGTH = 1 << LEAF_BITS; - - // Leaf node - struct Leaf { - void* values[LEAF_LENGTH]; - }; - - Leaf* root_[ROOT_LENGTH]; // Pointers to 32 child nodes - void* (*allocator_)(size_t); // Memory allocator - - public: - typedef uintptr_t Number; - - explicit TCMalloc_PageMap2(void* (*allocator)(size_t)) { - allocator_ = allocator; - memset(root_, 0, sizeof(root_)); - } - - void* get(Number k) const { - const Number i1 = k >> LEAF_BITS; - const Number i2 = k & (LEAF_LENGTH-1); - if ((k >> BITS) > 0 || root_[i1] == NULL) { - return NULL; - } - return root_[i1]->values[i2]; - } - - void set(Number k, void* v) { - const Number i1 = k >> LEAF_BITS; - const Number i2 = k & (LEAF_LENGTH-1); - ASSERT(i1 < ROOT_LENGTH); - root_[i1]->values[i2] = v; - } - - bool Ensure(Number start, size_t n) { - for (Number key = start; key <= start + n - 1; ) { - const Number i1 = key >> LEAF_BITS; - - // Check for overflow - if (i1 >= ROOT_LENGTH) - return false; - - // Make 2nd level node if necessary - if (root_[i1] == NULL) { - Leaf* leaf = reinterpret_cast((*allocator_)(sizeof(Leaf))); - if (leaf == NULL) return false; - memset(leaf, 0, sizeof(*leaf)); - root_[i1] = leaf; - } - - // Advance key past whatever is covered by this leaf node - key = ((key >> LEAF_BITS) + 1) << LEAF_BITS; - } - return true; - } - - void PreallocateMoreMemory() { - // Allocate enough to keep track of all possible pages - Ensure(0, 1 << BITS); - } - - void* Next(Number k) const { - while (k < (1 << BITS)) { - const Number i1 = k >> LEAF_BITS; - Leaf* leaf = root_[i1]; - if (leaf != NULL) { - // Scan forward in leaf - for (Number i2 = k & (LEAF_LENGTH - 1); i2 < LEAF_LENGTH; i2++) { - if (leaf->values[i2] != NULL) { - return leaf->values[i2]; - } - } - } - // Skip to next top-level entry - k = (i1 + 1) << LEAF_BITS; - } - return NULL; - } -}; - -// Three-level radix tree -template -class TCMalloc_PageMap3 { - private: - // How many bits should we consume at each interior level - static const int INTERIOR_BITS = (BITS + 2) / 3; // Round-up - static const int INTERIOR_LENGTH = 1 << INTERIOR_BITS; - - // How many bits should we consume at leaf level - static const int LEAF_BITS = BITS - 2*INTERIOR_BITS; - static const int LEAF_LENGTH = 1 << LEAF_BITS; - - // Interior node - struct Node { - Node* ptrs[INTERIOR_LENGTH]; - }; - - // Leaf node - struct Leaf { - void* values[LEAF_LENGTH]; - }; - - Node* root_; // Root of radix tree - void* (*allocator_)(size_t); // Memory allocator - - Node* NewNode() { - Node* result = reinterpret_cast((*allocator_)(sizeof(Node))); - if (result != NULL) { - memset(result, 0, sizeof(*result)); - } - return result; - } - - public: - typedef uintptr_t Number; - - explicit TCMalloc_PageMap3(void* (*allocator)(size_t)) { - allocator_ = allocator; - root_ = NewNode(); - } - - void* get(Number k) const { - const Number i1 = k >> (LEAF_BITS + INTERIOR_BITS); - const Number i2 = (k >> LEAF_BITS) & (INTERIOR_LENGTH-1); - const Number i3 = k & (LEAF_LENGTH-1); - if ((k >> BITS) > 0 || - root_->ptrs[i1] == NULL || root_->ptrs[i1]->ptrs[i2] == NULL) { - return NULL; - } - return reinterpret_cast(root_->ptrs[i1]->ptrs[i2])->values[i3]; - } - - void set(Number k, void* v) { - ASSERT(k >> BITS == 0); - const Number i1 = k >> (LEAF_BITS + INTERIOR_BITS); - const Number i2 = (k >> LEAF_BITS) & (INTERIOR_LENGTH-1); - const Number i3 = k & (LEAF_LENGTH-1); - reinterpret_cast(root_->ptrs[i1]->ptrs[i2])->values[i3] = v; - } - - bool Ensure(Number start, size_t n) { - for (Number key = start; key <= start + n - 1; ) { - const Number i1 = key >> (LEAF_BITS + INTERIOR_BITS); - const Number i2 = (key >> LEAF_BITS) & (INTERIOR_LENGTH-1); - - // Check for overflow - if (i1 >= INTERIOR_LENGTH || i2 >= INTERIOR_LENGTH) - return false; - - // Make 2nd level node if necessary - if (root_->ptrs[i1] == NULL) { - Node* n = NewNode(); - if (n == NULL) return false; - root_->ptrs[i1] = n; - } - - // Make leaf node if necessary - if (root_->ptrs[i1]->ptrs[i2] == NULL) { - Leaf* leaf = reinterpret_cast((*allocator_)(sizeof(Leaf))); - if (leaf == NULL) return false; - memset(leaf, 0, sizeof(*leaf)); - root_->ptrs[i1]->ptrs[i2] = reinterpret_cast(leaf); - } - - // Advance key past whatever is covered by this leaf node - key = ((key >> LEAF_BITS) + 1) << LEAF_BITS; - } - return true; - } - - void PreallocateMoreMemory() { - } - - void* Next(Number k) const { - while (k < (Number(1) << BITS)) { - const Number i1 = k >> (LEAF_BITS + INTERIOR_BITS); - const Number i2 = (k >> LEAF_BITS) & (INTERIOR_LENGTH-1); - if (root_->ptrs[i1] == NULL) { - // Advance to next top-level entry - k = (i1 + 1) << (LEAF_BITS + INTERIOR_BITS); - } else { - Leaf* leaf = reinterpret_cast(root_->ptrs[i1]->ptrs[i2]); - if (leaf != NULL) { - for (Number i3 = (k & (LEAF_LENGTH-1)); i3 < LEAF_LENGTH; i3++) { - if (leaf->values[i3] != NULL) { - return leaf->values[i3]; - } - } - } - // Advance to next interior entry - k = ((k >> LEAF_BITS) + 1) << LEAF_BITS; - } - } - return NULL; - } -}; - -#endif // TCMALLOC_PAGEMAP_H_ diff --git a/contrib/libtcmalloc/src/raw_printer.cc b/contrib/libtcmalloc/src/raw_printer.cc deleted file mode 100644 index 3cf028eeae0..00000000000 --- a/contrib/libtcmalloc/src/raw_printer.cc +++ /dev/null @@ -1,72 +0,0 @@ -// -*- Mode: C++; c-basic-offset: 2; indent-tabs-mode: nil -*- -// Copyright (c) 2008, Google Inc. -// All rights reserved. -// -// Redistribution and use in source and binary forms, with or without -// modification, are permitted provided that the following conditions are -// met: -// -// * Redistributions of source code must retain the above copyright -// notice, this list of conditions and the following disclaimer. -// * Redistributions in binary form must reproduce the above -// copyright notice, this list of conditions and the following disclaimer -// in the documentation and/or other materials provided with the -// distribution. -// * Neither the name of Google Inc. nor the names of its -// contributors may be used to endorse or promote products derived from -// this software without specific prior written permission. -// -// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS -// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT -// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR -// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT -// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, -// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT -// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, -// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY -// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT -// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE -// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. - -// --- -// Author: sanjay@google.com (Sanjay Ghemawat) - -#include -#include -#include -#include "raw_printer.h" -#include "base/logging.h" - -namespace base { - -RawPrinter::RawPrinter(char* buf, int length) - : base_(buf), - ptr_(buf), - limit_(buf + length - 1) { - RAW_DCHECK(length > 0, ""); - *ptr_ = '\0'; - *limit_ = '\0'; -} - -void RawPrinter::Printf(const char* format, ...) { - if (limit_ > ptr_) { - va_list ap; - va_start(ap, format); - int avail = limit_ - ptr_; - // We pass avail+1 to vsnprintf() since that routine needs room - // to store the trailing \0. - const int r = perftools_vsnprintf(ptr_, avail+1, format, ap); - va_end(ap); - if (r < 0) { - // Perhaps an old glibc that returns -1 on truncation? - ptr_ = limit_; - } else if (r > avail) { - // Truncation - ptr_ = limit_; - } else { - ptr_ += r; - } - } -} - -} diff --git a/contrib/libtcmalloc/src/raw_printer.h b/contrib/libtcmalloc/src/raw_printer.h deleted file mode 100644 index 9288bb5eeaa..00000000000 --- a/contrib/libtcmalloc/src/raw_printer.h +++ /dev/null @@ -1,90 +0,0 @@ -// -*- Mode: C++; c-basic-offset: 2; indent-tabs-mode: nil -*- -// Copyright (c) 2008, Google Inc. -// All rights reserved. -// -// Redistribution and use in source and binary forms, with or without -// modification, are permitted provided that the following conditions are -// met: -// -// * Redistributions of source code must retain the above copyright -// notice, this list of conditions and the following disclaimer. -// * Redistributions in binary form must reproduce the above -// copyright notice, this list of conditions and the following disclaimer -// in the documentation and/or other materials provided with the -// distribution. -// * Neither the name of Google Inc. nor the names of its -// contributors may be used to endorse or promote products derived from -// this software without specific prior written permission. -// -// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS -// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT -// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR -// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT -// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, -// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT -// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, -// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY -// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT -// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE -// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. - -// --- -// Author: Sanjay Ghemawat -// -// A printf() wrapper that writes into a fixed length buffer. -// Useful in low-level code that does not want to use allocating -// routines like StringPrintf(). -// -// The implementation currently uses vsnprintf(). This seems to -// be fine for use in many low-level contexts, but we may need to -// rethink this decision if we hit a problem with it calling -// down into malloc() etc. - -#ifndef BASE_RAW_PRINTER_H_ -#define BASE_RAW_PRINTER_H_ - -#include -#include "base/basictypes.h" - -namespace base { - -class RawPrinter { - public: - // REQUIRES: "length > 0" - // Will printf any data added to this into "buf[0,length-1]" and - // will arrange to always keep buf[] null-terminated. - RawPrinter(char* buf, int length); - - // Return the number of bytes that have been appended to the string - // so far. Does not count any bytes that were dropped due to overflow. - int length() const { return (ptr_ - base_); } - - // Return the number of bytes that can be added to this. - int space_left() const { return (limit_ - ptr_); } - - // Format the supplied arguments according to the "format" string - // and append to this. Will silently truncate the output if it does - // not fit. - void Printf(const char* format, ...) -#ifdef HAVE___ATTRIBUTE__ - __attribute__ ((__format__ (__printf__, 2, 3))) -#endif -; - - private: - // We can write into [ptr_ .. limit_-1]. - // *limit_ is also writable, but reserved for a terminating \0 - // in case we overflow. - // - // Invariants: *ptr_ == \0 - // Invariants: *limit_ == \0 - char* base_; // Initial pointer - char* ptr_; // Where should we write next - char* limit_; // One past last non-\0 char we can write - - DISALLOW_COPY_AND_ASSIGN(RawPrinter); -}; - -} - -#endif // BASE_RAW_PRINTER_H_ diff --git a/contrib/libtcmalloc/src/sampler.cc b/contrib/libtcmalloc/src/sampler.cc deleted file mode 100644 index cc711123340..00000000000 --- a/contrib/libtcmalloc/src/sampler.cc +++ /dev/null @@ -1,131 +0,0 @@ -// -*- Mode: C++; c-basic-offset: 2; indent-tabs-mode: nil -*- -// Copyright (c) 2008, Google Inc. -// All rights reserved. -// -// Redistribution and use in source and binary forms, with or without -// modification, are permitted provided that the following conditions are -// met: -// -// * Redistributions of source code must retain the above copyright -// notice, this list of conditions and the following disclaimer. -// * Redistributions in binary form must reproduce the above -// copyright notice, this list of conditions and the following disclaimer -// in the documentation and/or other materials provided with the -// distribution. -// * Neither the name of Google Inc. nor the names of its -// contributors may be used to endorse or promote products derived from -// this software without specific prior written permission. -// -// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS -// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT -// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR -// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT -// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, -// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT -// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, -// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY -// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT -// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE -// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. - -// --- -// All Rights Reserved. -// -// Author: Daniel Ford - -#include "sampler.h" - -#include // For min() -#include -#include "base/commandlineflags.h" - -using std::min; - -// The approximate gap in bytes between sampling actions. -// I.e., we take one sample approximately once every -// tcmalloc_sample_parameter bytes of allocation -// i.e. about once every 512KB if value is 1<<19. -#ifdef NO_TCMALLOC_SAMPLES -DEFINE_int64(tcmalloc_sample_parameter, 0, - "Unused: code is compiled with NO_TCMALLOC_SAMPLES"); -#else -DEFINE_int64(tcmalloc_sample_parameter, - EnvToInt64("TCMALLOC_SAMPLE_PARAMETER", 0), - "The approximate gap in bytes between sampling actions. " - "This must be between 1 and 2^58."); -#endif - -namespace tcmalloc { - -// Statics for Sampler -double Sampler::log_table_[1<(i+0.5)/(1<(reinterpret_cast(this)); - if (rnd_ == 0) { - rnd_ = 1; - } - } - // Step it forward 20 times for good measure - for (int i = 0; i < 20; i++) { - rnd_ = NextRandom(rnd_); - } - // Initialize counter - bytes_until_sample_ = PickNextSamplingPoint(); -} - -// Initialize the Statics for the Sampler class -void Sampler::InitStatics() { - PopulateFastLog2Table(); -} - -// Generates a geometric variable with the specified mean (512K by default). -// This is done by generating a random number between 0 and 1 and applying -// the inverse cumulative distribution function for an exponential. -// Specifically: Let m be the inverse of the sample period, then -// the probability distribution function is m*exp(-mx) so the CDF is -// p = 1 - exp(-mx), so -// q = 1 - p = exp(-mx) -// log_e(q) = -mx -// -log_e(q)/m = x -// log_2(q) * (-log_e(2) * 1/m) = x -// In the code, q is actually in the range 1 to 2**26, hence the -26 below -size_t Sampler::PickNextSamplingPoint() { - rnd_ = NextRandom(rnd_); - // Take the top 26 bits as the random number - // (This plus the 1<<58 sampling bound give a max possible step of - // 5194297183973780480 bytes.) - const uint64_t prng_mod_power = 48; // Number of bits in prng - // The uint32_t cast is to prevent a (hard-to-reproduce) NAN - // under piii debug for some binaries. - double q = static_cast(rnd_ >> (prng_mod_power - 26)) + 1.0; - // Put the computed p-value through the CDF of a geometric. - // For faster performance (save ~1/20th exec time), replace - // min(0.0, FastLog2(q) - 26) by (Fastlog2(q) - 26.000705) - // The value 26.000705 is used rather than 26 to compensate - // for inaccuracies in FastLog2 which otherwise result in a - // negative answer. - return static_cast(min(0.0, (FastLog2(q) - 26)) * (-log(2.0) - * FLAGS_tcmalloc_sample_parameter) + 1); -} - -} // namespace tcmalloc diff --git a/contrib/libtcmalloc/src/sampler.h b/contrib/libtcmalloc/src/sampler.h deleted file mode 100644 index eb316d7493d..00000000000 --- a/contrib/libtcmalloc/src/sampler.h +++ /dev/null @@ -1,180 +0,0 @@ -// -*- Mode: C++; c-basic-offset: 2; indent-tabs-mode: nil -*- -// Copyright (c) 2008, Google Inc. -// All rights reserved. -// -// Redistribution and use in source and binary forms, with or without -// modification, are permitted provided that the following conditions are -// met: -// -// * Redistributions of source code must retain the above copyright -// notice, this list of conditions and the following disclaimer. -// * Redistributions in binary form must reproduce the above -// copyright notice, this list of conditions and the following disclaimer -// in the documentation and/or other materials provided with the -// distribution. -// * Neither the name of Google Inc. nor the names of its -// contributors may be used to endorse or promote products derived from -// this software without specific prior written permission. -// -// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS -// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT -// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR -// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT -// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, -// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT -// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, -// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY -// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT -// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE -// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. - -// --- -// All Rights Reserved. -// -// Author: Daniel Ford - -#ifndef TCMALLOC_SAMPLER_H_ -#define TCMALLOC_SAMPLER_H_ - -#include "config.h" -#include // for size_t -#ifdef HAVE_STDINT_H -#include // for uint64_t, uint32_t, int32_t -#endif -#include // for memcpy -#include "base/basictypes.h" // for ASSERT -#include "internal_logging.h" // for ASSERT - -namespace tcmalloc { - -//------------------------------------------------------------------- -// Sampler to decide when to create a sample trace for an allocation -// Not thread safe: Each thread should have it's own sampler object. -// Caller must use external synchronization if used -// from multiple threads. -// -// With 512K average sample step (the default): -// the probability of sampling a 4K allocation is about 0.00778 -// the probability of sampling a 1MB allocation is about 0.865 -// the probability of sampling a 1GB allocation is about 1.00000 -// In general, the probablity of sampling is an allocation of size X -// given a flag value of Y (default 1M) is: -// 1 - e^(-X/Y) -// -// With 128K average sample step: -// the probability of sampling a 1MB allocation is about 0.99966 -// the probability of sampling a 1GB allocation is about 1.0 -// (about 1 - 2**(-26)) -// With 1M average sample step: -// the probability of sampling a 4K allocation is about 0.00390 -// the probability of sampling a 1MB allocation is about 0.632 -// the probability of sampling a 1GB allocation is about 1.0 -// -// The sampler works by representing memory as a long stream from -// which allocations are taken. Some of the bytes in this stream are -// marked and if an allocation includes a marked byte then it is -// sampled. Bytes are marked according to a Poisson point process -// with each byte being marked independently with probability -// p = 1/tcmalloc_sample_parameter. This makes the probability -// of sampling an allocation of X bytes equal to the CDF of -// a geometric with mean tcmalloc_sample_parameter. (ie. the -// probability that at least one byte in the range is marked). This -// is accurately given by the CDF of the corresponding exponential -// distribution : 1 - e^(X/tcmalloc_sample_parameter_) -// Independence of the byte marking ensures independence of -// the sampling of each allocation. -// -// This scheme is implemented by noting that, starting from any -// fixed place, the number of bytes until the next marked byte -// is geometrically distributed. This number is recorded as -// bytes_until_sample_. Every allocation subtracts from this -// number until it is less than 0. When this happens the current -// allocation is sampled. -// -// When an allocation occurs, bytes_until_sample_ is reset to -// a new independtly sampled geometric number of bytes. The -// memoryless property of the point process means that this may -// be taken as the number of bytes after the end of the current -// allocation until the next marked byte. This ensures that -// very large allocations which would intersect many marked bytes -// only result in a single call to PickNextSamplingPoint. -//------------------------------------------------------------------- - -class PERFTOOLS_DLL_DECL Sampler { - public: - // Initialize this sampler. - // Passing a seed of 0 gives a non-deterministic - // seed value given by casting the object ("this") - void Init(uint32_t seed); - void Cleanup(); - - // Record allocation of "k" bytes. Return true iff allocation - // should be sampled - bool SampleAllocation(size_t k); - - // Generate a geometric with mean 512K (or FLAG_tcmalloc_sample_parameter) - size_t PickNextSamplingPoint(); - - // Initialize the statics for the Sampler class - static void InitStatics(); - - // Returns the current sample period - int GetSamplePeriod(); - - // The following are public for the purposes of testing - static uint64_t NextRandom(uint64_t rnd_); // Returns the next prng value - static double FastLog2(const double & d); // Computes Log2(x) quickly - static void PopulateFastLog2Table(); // Populate the lookup table - - private: - size_t bytes_until_sample_; // Bytes until we sample next - uint64_t rnd_; // Cheap random number generator - - // Statics for the fast log - // Note that this code may not depend on anything in //util - // hence the duplication of functionality here - static const int kFastlogNumBits = 10; - static const int kFastlogMask = (1 << kFastlogNumBits) - 1; - static double log_table_[1<(0)) << prng_mod_power); - return (prng_mult * rnd + prng_add) & prng_mod_mask; -} - -// Adapted from //util/math/fastmath.[h|cc] by Noam Shazeer -// This mimics the VeryFastLog2 code in those files -inline double Sampler::FastLog2(const double & d) { - ASSERT(d>0); - COMPILE_ASSERT(sizeof(d) == sizeof(uint64_t), DoubleMustBe64Bits); - uint64_t x; - memcpy(&x, &d, sizeof(x)); // we depend on the compiler inlining this - const uint32_t x_high = x >> 32; - const uint32_t y = x_high >> (20 - kFastlogNumBits) & kFastlogMask; - const int32_t exponent = ((x_high >> 20) & 0x7FF) - 1023; - return exponent + log_table_[y]; -} - -} // namespace tcmalloc - -#endif // TCMALLOC_SAMPLER_H_ diff --git a/contrib/libtcmalloc/src/span.cc b/contrib/libtcmalloc/src/span.cc deleted file mode 100644 index 5f7ae436086..00000000000 --- a/contrib/libtcmalloc/src/span.cc +++ /dev/null @@ -1,102 +0,0 @@ -// -*- Mode: C++; c-basic-offset: 2; indent-tabs-mode: nil -*- -// Copyright (c) 2008, Google Inc. -// All rights reserved. -// -// Redistribution and use in source and binary forms, with or without -// modification, are permitted provided that the following conditions are -// met: -// -// * Redistributions of source code must retain the above copyright -// notice, this list of conditions and the following disclaimer. -// * Redistributions in binary form must reproduce the above -// copyright notice, this list of conditions and the following disclaimer -// in the documentation and/or other materials provided with the -// distribution. -// * Neither the name of Google Inc. nor the names of its -// contributors may be used to endorse or promote products derived from -// this software without specific prior written permission. -// -// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS -// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT -// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR -// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT -// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, -// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT -// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, -// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY -// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT -// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE -// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. - -// --- -// Author: Sanjay Ghemawat - -#include "config.h" -#include "span.h" - -#include // for NULL, memset - -#include "internal_logging.h" // for ASSERT -#include "page_heap_allocator.h" // for PageHeapAllocator -#include "static_vars.h" // for Static - -namespace tcmalloc { - -#ifdef SPAN_HISTORY -void Event(Span* span, char op, int v = 0) { - span->history[span->nexthistory] = op; - span->value[span->nexthistory] = v; - span->nexthistory++; - if (span->nexthistory == sizeof(span->history)) span->nexthistory = 0; -} -#endif - -Span* NewSpan(PageID p, Length len) { - Span* result = Static::span_allocator()->New(); - memset(result, 0, sizeof(*result)); - result->start = p; - result->length = len; -#ifdef SPAN_HISTORY - result->nexthistory = 0; -#endif - return result; -} - -void DeleteSpan(Span* span) { -#ifndef NDEBUG - // In debug mode, trash the contents of deleted Spans - memset(span, 0x3f, sizeof(*span)); -#endif - Static::span_allocator()->Delete(span); -} - -void DLL_Init(Span* list) { - list->next = list; - list->prev = list; -} - -void DLL_Remove(Span* span) { - span->prev->next = span->next; - span->next->prev = span->prev; - span->prev = NULL; - span->next = NULL; -} - -int DLL_Length(const Span* list) { - int result = 0; - for (Span* s = list->next; s != list; s = s->next) { - result++; - } - return result; -} - -void DLL_Prepend(Span* list, Span* span) { - ASSERT(span->next == NULL); - ASSERT(span->prev == NULL); - span->next = list->next; - span->prev = list; - list->next->prev = span; - list->next = span; -} - -} // namespace tcmalloc diff --git a/contrib/libtcmalloc/src/span.h b/contrib/libtcmalloc/src/span.h deleted file mode 100644 index 3fe30ba33d0..00000000000 --- a/contrib/libtcmalloc/src/span.h +++ /dev/null @@ -1,102 +0,0 @@ -// -*- Mode: C++; c-basic-offset: 2; indent-tabs-mode: nil -*- -// Copyright (c) 2008, Google Inc. -// All rights reserved. -// -// Redistribution and use in source and binary forms, with or without -// modification, are permitted provided that the following conditions are -// met: -// -// * Redistributions of source code must retain the above copyright -// notice, this list of conditions and the following disclaimer. -// * Redistributions in binary form must reproduce the above -// copyright notice, this list of conditions and the following disclaimer -// in the documentation and/or other materials provided with the -// distribution. -// * Neither the name of Google Inc. nor the names of its -// contributors may be used to endorse or promote products derived from -// this software without specific prior written permission. -// -// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS -// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT -// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR -// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT -// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, -// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT -// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, -// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY -// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT -// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE -// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. - -// --- -// Author: Sanjay Ghemawat -// -// A Span is a contiguous run of pages. - -#ifndef TCMALLOC_SPAN_H_ -#define TCMALLOC_SPAN_H_ - -#include "config.h" -#include "common.h" - -namespace tcmalloc { - -// Information kept for a span (a contiguous run of pages). -struct Span { - PageID start; // Starting page number - Length length; // Number of pages in span - Span* next; // Used when in link list - Span* prev; // Used when in link list - void* objects; // Linked list of free objects - unsigned int refcount : 16; // Number of non-free objects - unsigned int sizeclass : 8; // Size-class for small objects (or 0) - unsigned int location : 2; // Is the span on a freelist, and if so, which? - unsigned int sample : 1; // Sampled object? - -#undef SPAN_HISTORY -#ifdef SPAN_HISTORY - // For debugging, we can keep a log events per span - int nexthistory; - char history[64]; - int value[64]; -#endif - - // What freelist the span is on: IN_USE if on none, or normal or returned - enum { IN_USE, ON_NORMAL_FREELIST, ON_RETURNED_FREELIST }; -}; - -#ifdef SPAN_HISTORY -void Event(Span* span, char op, int v = 0); -#else -#define Event(s,o,v) ((void) 0) -#endif - -// Allocator/deallocator for spans -Span* NewSpan(PageID p, Length len); -void DeleteSpan(Span* span); - -// ------------------------------------------------------------------------- -// Doubly linked list of spans. -// ------------------------------------------------------------------------- - -// Initialize *list to an empty list. -void DLL_Init(Span* list); - -// Remove 'span' from the linked list in which it resides, updating the -// pointers of adjacent Spans and setting span's next and prev to NULL. -void DLL_Remove(Span* span); - -// Return true iff "list" is empty. -inline bool DLL_IsEmpty(const Span* list) { - return list->next == list; -} - -// Add span to the front of list. -void DLL_Prepend(Span* list, Span* span); - -// Return the length of the linked list. O(n) -int DLL_Length(const Span* list); - -} // namespace tcmalloc - -#endif // TCMALLOC_SPAN_H_ diff --git a/contrib/libtcmalloc/src/stack_trace_table.cc b/contrib/libtcmalloc/src/stack_trace_table.cc deleted file mode 100644 index 049cca524b5..00000000000 --- a/contrib/libtcmalloc/src/stack_trace_table.cc +++ /dev/null @@ -1,160 +0,0 @@ -// -*- Mode: C++; c-basic-offset: 2; indent-tabs-mode: nil -*- -// Copyright (c) 2009, Google Inc. -// All rights reserved. -// -// Redistribution and use in source and binary forms, with or without -// modification, are permitted provided that the following conditions are -// met: -// -// * Redistributions of source code must retain the above copyright -// notice, this list of conditions and the following disclaimer. -// * Redistributions in binary form must reproduce the above -// copyright notice, this list of conditions and the following disclaimer -// in the documentation and/or other materials provided with the -// distribution. -// * Neither the name of Google Inc. nor the names of its -// contributors may be used to endorse or promote products derived from -// this software without specific prior written permission. -// -// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS -// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT -// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR -// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT -// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, -// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT -// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, -// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY -// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT -// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE -// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. - -// --- -// Author: Andrew Fikes - -#include "config.h" -#include "stack_trace_table.h" -#include // for NULL, memset -#include "base/spinlock.h" // for SpinLockHolder -#include "common.h" // for StackTrace -#include "internal_logging.h" // for ASSERT, Log -#include "page_heap_allocator.h" // for PageHeapAllocator -#include "static_vars.h" // for Static - -namespace tcmalloc { - -bool StackTraceTable::Bucket::KeyEqual(uintptr_t h, - const StackTrace& t) const { - const bool eq = (this->hash == h && this->trace.depth == t.depth); - for (int i = 0; eq && i < t.depth; ++i) { - if (this->trace.stack[i] != t.stack[i]) { - return false; - } - } - return eq; -} - -StackTraceTable::StackTraceTable() - : error_(false), - depth_total_(0), - bucket_total_(0), - table_(new Bucket*[kHashTableSize]()) { - memset(table_, 0, kHashTableSize * sizeof(Bucket*)); -} - -StackTraceTable::~StackTraceTable() { - delete[] table_; -} - -void StackTraceTable::AddTrace(const StackTrace& t) { - if (error_) { - return; - } - - // Hash function borrowed from base/heap-profile-table.cc - uintptr_t h = 0; - for (int i = 0; i < t.depth; ++i) { - h += reinterpret_cast(t.stack[i]); - h += h << 10; - h ^= h >> 6; - } - h += h << 3; - h ^= h >> 11; - - const int idx = h % kHashTableSize; - - Bucket* b = table_[idx]; - while (b != NULL && !b->KeyEqual(h, t)) { - b = b->next; - } - if (b != NULL) { - b->count++; - b->trace.size += t.size; // keep cumulative size - } else { - depth_total_ += t.depth; - bucket_total_++; - b = Static::bucket_allocator()->New(); - if (b == NULL) { - Log(kLog, __FILE__, __LINE__, - "tcmalloc: could not allocate bucket", sizeof(*b)); - error_ = true; - } else { - b->hash = h; - b->trace = t; - b->count = 1; - b->next = table_[idx]; - table_[idx] = b; - } - } -} - -void** StackTraceTable::ReadStackTracesAndClear() { - if (error_) { - return NULL; - } - - // Allocate output array - const int out_len = bucket_total_ * 3 + depth_total_ + 1; - void** out = new void*[out_len]; - if (out == NULL) { - Log(kLog, __FILE__, __LINE__, - "tcmalloc: allocation failed for stack traces", - out_len * sizeof(*out)); - return NULL; - } - - // Fill output array - int idx = 0; - for (int i = 0; i < kHashTableSize; ++i) { - Bucket* b = table_[i]; - while (b != NULL) { - out[idx++] = reinterpret_cast(static_cast(b->count)); - out[idx++] = reinterpret_cast(b->trace.size); // cumulative size - out[idx++] = reinterpret_cast(b->trace.depth); - for (int d = 0; d < b->trace.depth; ++d) { - out[idx++] = b->trace.stack[d]; - } - b = b->next; - } - } - out[idx++] = NULL; - ASSERT(idx == out_len); - - // Clear state - error_ = false; - depth_total_ = 0; - bucket_total_ = 0; - SpinLockHolder h(Static::pageheap_lock()); - for (int i = 0; i < kHashTableSize; ++i) { - Bucket* b = table_[i]; - while (b != NULL) { - Bucket* next = b->next; - Static::bucket_allocator()->Delete(b); - b = next; - } - table_[i] = NULL; - } - - return out; -} - -} // namespace tcmalloc diff --git a/contrib/libtcmalloc/src/stack_trace_table.h b/contrib/libtcmalloc/src/stack_trace_table.h deleted file mode 100644 index 66ed5d92822..00000000000 --- a/contrib/libtcmalloc/src/stack_trace_table.h +++ /dev/null @@ -1,92 +0,0 @@ -// -*- Mode: C++; c-basic-offset: 2; indent-tabs-mode: nil -*- -// Copyright (c) 2009, Google Inc. -// All rights reserved. -// -// Redistribution and use in source and binary forms, with or without -// modification, are permitted provided that the following conditions are -// met: -// -// * Redistributions of source code must retain the above copyright -// notice, this list of conditions and the following disclaimer. -// * Redistributions in binary form must reproduce the above -// copyright notice, this list of conditions and the following disclaimer -// in the documentation and/or other materials provided with the -// distribution. -// * Neither the name of Google Inc. nor the names of its -// contributors may be used to endorse or promote products derived from -// this software without specific prior written permission. -// -// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS -// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT -// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR -// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT -// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, -// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT -// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, -// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY -// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT -// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE -// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. - -// --- -// Author: Andrew Fikes -// -// Utility class for coalescing sampled stack traces. Not thread-safe. - -#ifndef TCMALLOC_STACK_TRACE_TABLE_H_ -#define TCMALLOC_STACK_TRACE_TABLE_H_ - -#include "config.h" -#ifdef HAVE_STDINT_H -#include // for uintptr_t -#endif -#include "common.h" - -namespace tcmalloc { - -class PERFTOOLS_DLL_DECL StackTraceTable { - public: - // REQUIRES: L < pageheap_lock - StackTraceTable(); - ~StackTraceTable(); - - // Adds stack trace "t" to table. - // - // REQUIRES: L >= pageheap_lock - void AddTrace(const StackTrace& t); - - // Returns stack traces formatted per MallocExtension guidelines. - // May return NULL on error. Clears state before returning. - // - // REQUIRES: L < pageheap_lock - void** ReadStackTracesAndClear(); - - // Exposed for PageHeapAllocator - struct Bucket { - // Key - uintptr_t hash; - StackTrace trace; - - // Payload - int count; - Bucket* next; - - bool KeyEqual(uintptr_t h, const StackTrace& t) const; - }; - - // For testing - int depth_total() const { return depth_total_; } - int bucket_total() const { return bucket_total_; } - - private: - static const int kHashTableSize = 1 << 14; // => table_ is 128k - - bool error_; - int depth_total_; - int bucket_total_; - Bucket** table_; -}; - -} // namespace tcmalloc - -#endif // TCMALLOC_STACK_TRACE_TABLE_H_ diff --git a/contrib/libtcmalloc/src/stacktrace.cc b/contrib/libtcmalloc/src/stacktrace.cc deleted file mode 100644 index 88c8b15946d..00000000000 --- a/contrib/libtcmalloc/src/stacktrace.cc +++ /dev/null @@ -1,339 +0,0 @@ -// -*- Mode: C++; c-basic-offset: 2; indent-tabs-mode: nil -*- -// Copyright (c) 2005, Google Inc. -// All rights reserved. -// -// Redistribution and use in source and binary forms, with or without -// modification, are permitted provided that the following conditions are -// met: -// -// * Redistributions of source code must retain the above copyright -// notice, this list of conditions and the following disclaimer. -// * Redistributions in binary form must reproduce the above -// copyright notice, this list of conditions and the following disclaimer -// in the documentation and/or other materials provided with the -// distribution. -// * Neither the name of Google Inc. nor the names of its -// contributors may be used to endorse or promote products derived from -// this software without specific prior written permission. -// -// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS -// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT -// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR -// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT -// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, -// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT -// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, -// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY -// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT -// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE -// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. - -// --- -// Author: Sanjay Ghemawat -// -// Produce stack trace. -// -// There are three different ways we can try to get the stack trace: -// -// 1) Our hand-coded stack-unwinder. This depends on a certain stack -// layout, which is used by gcc (and those systems using a -// gcc-compatible ABI) on x86 systems, at least since gcc 2.95. -// It uses the frame pointer to do its work. -// -// 2) The libunwind library. This is still in development, and as a -// separate library adds a new dependency, abut doesn't need a frame -// pointer. It also doesn't call malloc. -// -// 3) The gdb unwinder -- also the one used by the c++ exception code. -// It's obviously well-tested, but has a fatal flaw: it can call -// malloc() from the unwinder. This is a problem because we're -// trying to use the unwinder to instrument malloc(). -// -// Note: if you add a new implementation here, make sure it works -// correctly when GetStackTrace() is called with max_depth == 0. -// Some code may do that. - -#include "config.h" -#include // for getenv -#include // for strcmp -#include // for fprintf -#include "gperftools/stacktrace.h" -#include "base/commandlineflags.h" -#include "base/googleinit.h" - - -// we're using plain struct and not class to avoid any possible issues -// during initialization. Struct of pointers is easy to init at -// link-time. -struct GetStackImplementation { - int (*GetStackFramesPtr)(void** result, int* sizes, int max_depth, - int skip_count); - - int (*GetStackFramesWithContextPtr)(void** result, int* sizes, int max_depth, - int skip_count, const void *uc); - - int (*GetStackTracePtr)(void** result, int max_depth, - int skip_count); - - int (*GetStackTraceWithContextPtr)(void** result, int max_depth, - int skip_count, const void *uc); - - const char *name; -}; - -#if HAVE_DECL_BACKTRACE -#define STACKTRACE_INL_HEADER "stacktrace_generic-inl.h" -#define GST_SUFFIX generic -#include "stacktrace_impl_setup-inl.h" -#undef GST_SUFFIX -#undef STACKTRACE_INL_HEADER -#define HAVE_GST_generic -#endif - -#ifdef HAVE_UNWIND_BACKTRACE -#define STACKTRACE_INL_HEADER "stacktrace_libgcc-inl.h" -#define GST_SUFFIX libgcc -#include "stacktrace_impl_setup-inl.h" -#undef GST_SUFFIX -#undef STACKTRACE_INL_HEADER -#define HAVE_GST_libgcc -#endif - -// libunwind uses __thread so we check for both libunwind.h and -// __thread support -#if defined(HAVE_LIBUNWIND_H) && defined(HAVE_TLS) -#define STACKTRACE_INL_HEADER "stacktrace_libunwind-inl.h" -#define GST_SUFFIX libunwind -#include "stacktrace_impl_setup-inl.h" -#undef GST_SUFFIX -#undef STACKTRACE_INL_HEADER -#define HAVE_GST_libunwind -#endif // HAVE_LIBUNWIND_H - -#if defined(__i386__) || defined(__x86_64__) -#define STACKTRACE_INL_HEADER "stacktrace_x86-inl.h" -#define GST_SUFFIX x86 -#include "stacktrace_impl_setup-inl.h" -#undef GST_SUFFIX -#undef STACKTRACE_INL_HEADER -#define HAVE_GST_x86 -#endif // i386 || x86_64 - -#if defined(__ppc__) || defined(__PPC__) -#if defined(__linux__) -#define STACKTRACE_INL_HEADER "stacktrace_powerpc-linux-inl.h" -#else -#define STACKTRACE_INL_HEADER "stacktrace_powerpc-darwin-inl.h" -#endif -#define GST_SUFFIX ppc -#include "stacktrace_impl_setup-inl.h" -#undef GST_SUFFIX -#undef STACKTRACE_INL_HEADER -#define HAVE_GST_ppc -#endif - -#if defined(__arm__) -#define STACKTRACE_INL_HEADER "stacktrace_arm-inl.h" -#define GST_SUFFIX arm -#include "stacktrace_impl_setup-inl.h" -#undef GST_SUFFIX -#undef STACKTRACE_INL_HEADER -#define HAVE_GST_arm -#endif - -#ifdef TCMALLOC_ENABLE_INSTRUMENT_STACKTRACE -#define STACKTRACE_INL_HEADER "stacktrace_instrument-inl.h" -#define GST_SUFFIX instrument -#include "stacktrace_impl_setup-inl.h" -#undef GST_SUFFIX -#undef STACKTRACE_INL_HEADER -#define HAVE_GST_instrument -#endif - -// The Windows case -- probably cygwin and mingw will use one of the -// x86-includes above, but if not, we can fall back to windows intrinsics. -#if defined(_WIN32) || defined(__CYGWIN__) || defined(__CYGWIN32__) || defined(__MINGW32__) -#define STACKTRACE_INL_HEADER "stacktrace_win32-inl.h" -#define GST_SUFFIX win32 -#include "stacktrace_impl_setup-inl.h" -#undef GST_SUFFIX -#undef STACKTRACE_INL_HEADER -#define HAVE_GST_win32 -#endif - -static GetStackImplementation *all_impls[] = { -#ifdef HAVE_GST_libgcc - &impl__libgcc, -#endif -#ifdef HAVE_GST_generic - &impl__generic, -#endif -#ifdef HAVE_GST_libunwind - &impl__libunwind, -#endif -#ifdef HAVE_GST_x86 - &impl__x86, -#endif -#ifdef HAVE_GST_arm - &impl__arm, -#endif -#ifdef HAVE_GST_ppc - &impl__ppc, -#endif -#ifdef HAVE_GST_instrument - &impl__instrument, -#endif -#ifdef HAVE_GST_win32 - &impl__win32, -#endif - NULL -}; - -// ppc and i386 implementations prefer arch-specific asm implementations. -// arm's asm implementation is broken -#if defined(__i386__) || defined(__x86_64__) || defined(__ppc__) || defined(__PPC__) -#if !defined(NO_FRAME_POINTER) -#define TCMALLOC_DONT_PREFER_LIBUNWIND -#endif -#endif - -static bool get_stack_impl_inited; - -#if defined(HAVE_GST_instrument) -static GetStackImplementation *get_stack_impl = &impl__instrument; -#elif defined(HAVE_GST_win32) -static GetStackImplementation *get_stack_impl = &impl__win32; -#elif defined(HAVE_GST_x86) && defined(TCMALLOC_DONT_PREFER_LIBUNWIND) -static GetStackImplementation *get_stack_impl = &impl__x86; -#elif defined(HAVE_GST_ppc) && defined(TCMALLOC_DONT_PREFER_LIBUNWIND) -static GetStackImplementation *get_stack_impl = &impl__ppc; -#elif defined(HAVE_GST_libunwind) -static GetStackImplementation *get_stack_impl = &impl__libunwind; -#elif defined(HAVE_GST_libgcc) -static GetStackImplementation *get_stack_impl = &impl__libgcc; -#elif defined(HAVE_GST_generic) -static GetStackImplementation *get_stack_impl = &impl__generic; -#elif defined(HAVE_GST_arm) -static GetStackImplementation *get_stack_impl = &impl__arm; -#elif 0 -// This is for the benefit of code analysis tools that may have -// trouble with the computed #include above. -# include "stacktrace_x86-inl.h" -# include "stacktrace_libunwind-inl.h" -# include "stacktrace_generic-inl.h" -# include "stacktrace_powerpc-inl.h" -# include "stacktrace_win32-inl.h" -# include "stacktrace_arm-inl.h" -# include "stacktrace_instrument-inl.h" -#else -#error Cannot calculate stack trace: will need to write for your environment -#endif - -static int ATTRIBUTE_NOINLINE frame_forcer(int rv) { - return rv; -} - -static void init_default_stack_impl_inner(void); - -namespace tcmalloc { - bool EnterStacktraceScope(void); - void LeaveStacktraceScope(void); -} - -namespace { - using tcmalloc::EnterStacktraceScope; - using tcmalloc::LeaveStacktraceScope; - - class StacktraceScope { - bool stacktrace_allowed; - public: - StacktraceScope() { - stacktrace_allowed = true; - stacktrace_allowed = EnterStacktraceScope(); - } - bool IsStacktraceAllowed() { - return stacktrace_allowed; - } - ~StacktraceScope() { - if (stacktrace_allowed) { - LeaveStacktraceScope(); - } - } - }; -} - -PERFTOOLS_DLL_DECL int GetStackFrames(void** result, int* sizes, int max_depth, - int skip_count) { - StacktraceScope scope; - if (!scope.IsStacktraceAllowed()) { - return 0; - } - init_default_stack_impl_inner(); - return frame_forcer(get_stack_impl->GetStackFramesPtr(result, sizes, max_depth, skip_count)); -} - -PERFTOOLS_DLL_DECL int GetStackFramesWithContext(void** result, int* sizes, int max_depth, - int skip_count, const void *uc) { - StacktraceScope scope; - if (!scope.IsStacktraceAllowed()) { - return 0; - } - init_default_stack_impl_inner(); - return frame_forcer(get_stack_impl->GetStackFramesWithContextPtr( - result, sizes, max_depth, - skip_count, uc)); -} - -PERFTOOLS_DLL_DECL int GetStackTrace(void** result, int max_depth, - int skip_count) { - StacktraceScope scope; - if (!scope.IsStacktraceAllowed()) { - return 0; - } - init_default_stack_impl_inner(); - return frame_forcer(get_stack_impl->GetStackTracePtr(result, max_depth, skip_count)); -} - -PERFTOOLS_DLL_DECL int GetStackTraceWithContext(void** result, int max_depth, - int skip_count, const void *uc) { - StacktraceScope scope; - if (!scope.IsStacktraceAllowed()) { - return 0; - } - init_default_stack_impl_inner(); - return frame_forcer(get_stack_impl->GetStackTraceWithContextPtr( - result, max_depth, skip_count, uc)); -} - -static void init_default_stack_impl_inner(void) { - if (get_stack_impl_inited) { - return; - } - get_stack_impl_inited = true; - char *val = getenv("TCMALLOC_STACKTRACE_METHOD"); - if (!val || !*val) { - return; - } - for (GetStackImplementation **p = all_impls; *p; p++) { - GetStackImplementation *c = *p; - if (strcmp(c->name, val) == 0) { - get_stack_impl = c; - return; - } - } - fprintf(stderr, "Unknown or unsupported stacktrace method requested: %s. Ignoring it\n", val); -} - -static void init_default_stack_impl(void) { - init_default_stack_impl_inner(); - if (EnvToBool("TCMALLOC_STACKTRACE_METHOD_VERBOSE", false)) { - fprintf(stderr, "Chosen stacktrace method is %s\nSupported methods:\n", get_stack_impl->name); - for (GetStackImplementation **p = all_impls; *p; p++) { - GetStackImplementation *c = *p; - fprintf(stderr, "* %s\n", c->name); - } - fputs("\n", stderr); - } -} - -REGISTER_MODULE_INITIALIZER(stacktrace_init_default_stack_impl, init_default_stack_impl()); diff --git a/contrib/libtcmalloc/src/stacktrace_arm-inl.h b/contrib/libtcmalloc/src/stacktrace_arm-inl.h deleted file mode 100644 index 1586b8fec62..00000000000 --- a/contrib/libtcmalloc/src/stacktrace_arm-inl.h +++ /dev/null @@ -1,148 +0,0 @@ -// -*- Mode: C++; c-basic-offset: 2; indent-tabs-mode: nil -*- -// Copyright (c) 2011, Google Inc. -// All rights reserved. -// -// Redistribution and use in source and binary forms, with or without -// modification, are permitted provided that the following conditions are -// met: -// -// * Redistributions of source code must retain the above copyright -// notice, this list of conditions and the following disclaimer. -// * Redistributions in binary form must reproduce the above -// copyright notice, this list of conditions and the following disclaimer -// in the documentation and/or other materials provided with the -// distribution. -// * Neither the name of Google Inc. nor the names of its -// contributors may be used to endorse or promote products derived from -// this software without specific prior written permission. -// -// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS -// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT -// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR -// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT -// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, -// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT -// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, -// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY -// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT -// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE -// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. - -// --- -// Author: Doug Kwan -// This is inspired by Craig Silverstein's PowerPC stacktrace code. -// - -#ifndef BASE_STACKTRACE_ARM_INL_H_ -#define BASE_STACKTRACE_ARM_INL_H_ -// Note: this file is included into stacktrace.cc more than once. -// Anything that should only be defined once should be here: - -#include // for uintptr_t -#include "base/basictypes.h" // for NULL -#include - -// WARNING: -// This only works if all your code is in either ARM or THUMB mode. With -// interworking, the frame pointer of the caller can either be in r11 (ARM -// mode) or r7 (THUMB mode). A callee only saves the frame pointer of its -// mode in a fixed location on its stack frame. If the caller is a different -// mode, there is no easy way to find the frame pointer. It can either be -// still in the designated register or saved on stack along with other callee -// saved registers. - -// Given a pointer to a stack frame, locate and return the calling -// stackframe, or return NULL if no stackframe can be found. Perform sanity -// checks (the strictness of which is controlled by the boolean parameter -// "STRICT_UNWINDING") to reduce the chance that a bad pointer is returned. -template -static void **NextStackFrame(void **old_sp) { - void **new_sp = (void**) old_sp[-1]; - - // Check that the transition from frame pointer old_sp to frame - // pointer new_sp isn't clearly bogus - if (STRICT_UNWINDING) { - // With the stack growing downwards, older stack frame must be - // at a greater address that the current one. - if (new_sp <= old_sp) return NULL; - // Assume stack frames larger than 100,000 bytes are bogus. - if ((uintptr_t)new_sp - (uintptr_t)old_sp > 100000) return NULL; - } else { - // In the non-strict mode, allow discontiguous stack frames. - // (alternate-signal-stacks for example). - if (new_sp == old_sp) return NULL; - // And allow frames upto about 1MB. - if ((new_sp > old_sp) - && ((uintptr_t)new_sp - (uintptr_t)old_sp > 1000000)) return NULL; - } - if ((uintptr_t)new_sp & (sizeof(void *) - 1)) return NULL; - return new_sp; -} - -// This ensures that GetStackTrace stes up the Link Register properly. -#ifdef __GNUC__ -void StacktraceArmDummyFunction() __attribute__((noinline)); -void StacktraceArmDummyFunction() { __asm__ volatile(""); } -#else -# error StacktraceArmDummyFunction() needs to be ported to this platform. -#endif -#endif // BASE_STACKTRACE_ARM_INL_H_ - -// Note: this part of the file is included several times. -// Do not put globals below. - -// The following 4 functions are generated from the code below: -// GetStack{Trace,Frames}() -// GetStack{Trace,Frames}WithContext() -// -// These functions take the following args: -// void** result: the stack-trace, as an array -// int* sizes: the size of each stack frame, as an array -// (GetStackFrames* only) -// int max_depth: the size of the result (and sizes) array(s) -// int skip_count: how many stack pointers to skip before storing in result -// void* ucp: a ucontext_t* (GetStack{Trace,Frames}WithContext only) -static int GET_STACK_TRACE_OR_FRAMES { -#ifdef __GNUC__ - void **sp = reinterpret_cast(__builtin_frame_address(0)); -#else -# error reading stack point not yet supported on this platform. -#endif - - // On ARM, the return address is stored in the link register (r14). - // This is not saved on the stack frame of a leaf function. To - // simplify code that reads return addresses, we call a dummy - // function so that the return address of this function is also - // stored in the stack frame. This works at least for gcc. - StacktraceArmDummyFunction(); - - skip_count++; // skip parent frame due to indirection in stacktrace.cc - - int n = 0; - while (sp && n < max_depth) { - // The GetStackFrames routine is called when we are in some - // informational context (the failure signal handler for example). - // Use the non-strict unwinding rules to produce a stack trace - // that is as complete as possible (even if it contains a few bogus - // entries in some rare cases). - void **next_sp = NextStackFrame(sp); - - if (skip_count > 0) { - skip_count--; - } else { - result[n] = *sp; - -#if IS_STACK_FRAMES - if (next_sp > sp) { - sizes[n] = (uintptr_t)next_sp - (uintptr_t)sp; - } else { - // A frame-size of 0 is used to indicate unknown frame size. - sizes[n] = 0; - } -#endif - n++; - } - sp = next_sp; - } - return n; -} diff --git a/contrib/libtcmalloc/src/stacktrace_generic-inl.h b/contrib/libtcmalloc/src/stacktrace_generic-inl.h deleted file mode 100644 index 7d7c22d9e45..00000000000 --- a/contrib/libtcmalloc/src/stacktrace_generic-inl.h +++ /dev/null @@ -1,84 +0,0 @@ -// -*- Mode: C++; c-basic-offset: 2; indent-tabs-mode: nil -*- -// Copyright (c) 2005, Google Inc. -// All rights reserved. -// -// Redistribution and use in source and binary forms, with or without -// modification, are permitted provided that the following conditions are -// met: -// -// * Redistributions of source code must retain the above copyright -// notice, this list of conditions and the following disclaimer. -// * Redistributions in binary form must reproduce the above -// copyright notice, this list of conditions and the following disclaimer -// in the documentation and/or other materials provided with the -// distribution. -// * Neither the name of Google Inc. nor the names of its -// contributors may be used to endorse or promote products derived from -// this software without specific prior written permission. -// -// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS -// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT -// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR -// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT -// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, -// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT -// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, -// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY -// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT -// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE -// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. - -// --- -// Author: Sanjay Ghemawat -// -// Portable implementation - just use glibc -// -// Note: The glibc implementation may cause a call to malloc. -// This can cause a deadlock in HeapProfiler. - -#ifndef BASE_STACKTRACE_GENERIC_INL_H_ -#define BASE_STACKTRACE_GENERIC_INL_H_ -// Note: this file is included into stacktrace.cc more than once. -// Anything that should only be defined once should be here: - -#include -#include -#include "gperftools/stacktrace.h" -#endif // BASE_STACKTRACE_GENERIC_INL_H_ - -// Note: this part of the file is included several times. -// Do not put globals below. - -// The following 4 functions are generated from the code below: -// GetStack{Trace,Frames}() -// GetStack{Trace,Frames}WithContext() -// -// These functions take the following args: -// void** result: the stack-trace, as an array -// int* sizes: the size of each stack frame, as an array -// (GetStackFrames* only) -// int max_depth: the size of the result (and sizes) array(s) -// int skip_count: how many stack pointers to skip before storing in result -// void* ucp: a ucontext_t* (GetStack{Trace,Frames}WithContext only) -static int GET_STACK_TRACE_OR_FRAMES { - static const int kStackLength = 64; - void * stack[kStackLength]; - int size; - - size = backtrace(stack, kStackLength); - skip_count += 2; // we want to skip the current and it's parent frame as well - int result_count = size - skip_count; - if (result_count < 0) - result_count = 0; - if (result_count > max_depth) - result_count = max_depth; - for (int i = 0; i < result_count; i++) - result[i] = stack[i + skip_count]; - -#if IS_STACK_FRAMES - // No implementation for finding out the stack frame sizes yet. - memset(sizes, 0, sizeof(*sizes) * result_count); -#endif - - return result_count; -} diff --git a/contrib/libtcmalloc/src/stacktrace_impl_setup-inl.h b/contrib/libtcmalloc/src/stacktrace_impl_setup-inl.h deleted file mode 100644 index 698c5b38196..00000000000 --- a/contrib/libtcmalloc/src/stacktrace_impl_setup-inl.h +++ /dev/null @@ -1,94 +0,0 @@ -// -*- Mode: C++; c-basic-offset: 2; indent-tabs-mode: nil -*- -// NOTE: this is NOT to be #include-d normally. It's internal -// implementation detail of stacktrace.cc -// - -// Copyright (c) 2014, gperftools Contributors. -// All rights reserved. -// -// Redistribution and use in source and binary forms, with or without -// modification, are permitted provided that the following conditions are -// met: -// -// * Redistributions of source code must retain the above copyright -// notice, this list of conditions and the following disclaimer. -// * Redistributions in binary form must reproduce the above -// copyright notice, this list of conditions and the following disclaimer -// in the documentation and/or other materials provided with the -// distribution. -// * Neither the name of Google Inc. nor the names of its -// contributors may be used to endorse or promote products derived from -// this software without specific prior written permission. -// -// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS -// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT -// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR -// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT -// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, -// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT -// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, -// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY -// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT -// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE -// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. - -// --- -// Author: Aliaksey Kandratsenka -// -// based on stacktrace.cc and stacktrace_config.h by Sanjay Ghemawat -// and Paul Pluzhnikov from Google Inc - -#define SIS_CONCAT2(a, b) a##b -#define SIS_CONCAT(a, b) SIS_CONCAT2(a,b) - -#define SIS_STRINGIFY(a) SIS_STRINGIFY2(a) -#define SIS_STRINGIFY2(a) #a - -#define IS_STACK_FRAMES 0 -#define IS_WITH_CONTEXT 0 -#define GET_STACK_TRACE_OR_FRAMES \ - SIS_CONCAT(GetStackTrace_, GST_SUFFIX)(void **result, int max_depth, int skip_count) -#include STACKTRACE_INL_HEADER -#undef IS_STACK_FRAMES -#undef IS_WITH_CONTEXT -#undef GET_STACK_TRACE_OR_FRAMES - -#define IS_STACK_FRAMES 1 -#define IS_WITH_CONTEXT 0 -#define GET_STACK_TRACE_OR_FRAMES \ - SIS_CONCAT(GetStackFrames_, GST_SUFFIX)(void **result, int *sizes, int max_depth, int skip_count) -#include STACKTRACE_INL_HEADER -#undef IS_STACK_FRAMES -#undef IS_WITH_CONTEXT -#undef GET_STACK_TRACE_OR_FRAMES - -#define IS_STACK_FRAMES 0 -#define IS_WITH_CONTEXT 1 -#define GET_STACK_TRACE_OR_FRAMES \ - SIS_CONCAT(GetStackTraceWithContext_, GST_SUFFIX)(void **result, int max_depth, \ - int skip_count, const void *ucp) -#include STACKTRACE_INL_HEADER -#undef IS_STACK_FRAMES -#undef IS_WITH_CONTEXT -#undef GET_STACK_TRACE_OR_FRAMES - -#define IS_STACK_FRAMES 1 -#define IS_WITH_CONTEXT 1 -#define GET_STACK_TRACE_OR_FRAMES \ - SIS_CONCAT(GetStackFramesWithContext_, GST_SUFFIX)(void **result, int *sizes, int max_depth, \ - int skip_count, const void *ucp) -#include STACKTRACE_INL_HEADER -#undef IS_STACK_FRAMES -#undef IS_WITH_CONTEXT -#undef GET_STACK_TRACE_OR_FRAMES - -static GetStackImplementation SIS_CONCAT(impl__,GST_SUFFIX) = { - SIS_CONCAT(GetStackFrames_, GST_SUFFIX), - SIS_CONCAT(GetStackFramesWithContext_, GST_SUFFIX), - SIS_CONCAT(GetStackTrace_, GST_SUFFIX), - SIS_CONCAT(GetStackTraceWithContext_, GST_SUFFIX), - SIS_STRINGIFY(GST_SUFFIX) -}; - -#undef SIS_CONCAT2 -#undef SIS_CONCAT diff --git a/contrib/libtcmalloc/src/stacktrace_instrument-inl.h b/contrib/libtcmalloc/src/stacktrace_instrument-inl.h deleted file mode 100644 index c631765c8a2..00000000000 --- a/contrib/libtcmalloc/src/stacktrace_instrument-inl.h +++ /dev/null @@ -1,155 +0,0 @@ -// -*- Mode: C++; c-basic-offset: 2; indent-tabs-mode: nil -*- -// Copyright (c) 2013, Google Inc. -// All rights reserved. -// -// Redistribution and use in source and binary forms, with or without -// modification, are permitted provided that the following conditions are -// met: -// -// * Redistributions of source code must retain the above copyright -// notice, this list of conditions and the following disclaimer. -// * Redistributions in binary form must reproduce the above -// copyright notice, this list of conditions and the following disclaimer -// in the documentation and/or other materials provided with the -// distribution. -// * Neither the name of Google Inc. nor the names of its -// contributors may be used to endorse or promote products derived from -// this software without specific prior written permission. -// -// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS -// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT -// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR -// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT -// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, -// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT -// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, -// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY -// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT -// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE -// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. - -// --- -// Author: Jean Lee -// based on gcc Code-Gen-Options "-finstrument-functions" listed in -// http://gcc.gnu.org/onlinedocs/gcc/Code-Gen-Options.html . -// Should run configure with CXXFLAGS="-finstrument-functions". - -// This file is a backtrace implementation for systems : -// * The glibc implementation of backtrace() may cause a call to malloc, -// and cause a deadlock in HeapProfiler. -// * The libunwind implementation prints no backtrace. - -// The backtrace arrays are stored in "thread_back_trace" variable. -// Maybe to use thread local storage is better and should save memorys. - -#ifndef BASE_STACKTRACE_INSTRUMENT_INL_H_ -#define BASE_STACKTRACE_INSTRUMENT_INL_H_ -// Note: this file is included into stacktrace.cc more than once. -// Anything that should only be defined once should be here: - -#include -#include -#include -#include -#include "gperftools/stacktrace.h" - -#define gettid() syscall(__NR_gettid) -#ifndef __x86_64__ -#define MAX_THREAD (32768) -#else -#define MAX_THREAD (65536) -#endif -#define MAX_DEPTH (30) -#define ATTRIBUTE_NOINSTRUMENT __attribute__ ((no_instrument_function)) - -typedef struct { - int stack_depth; - void* frame[MAX_DEPTH]; -}BACK_TRACE; - -static BACK_TRACE thread_back_trace[MAX_THREAD]; -extern "C" { -void __cyg_profile_func_enter(void *func_address, - void *call_site) ATTRIBUTE_NOINSTRUMENT; -void __cyg_profile_func_enter(void *func_address, void *call_site) { - (void)func_address; - - BACK_TRACE* backtrace = thread_back_trace + gettid(); - int stack_depth = backtrace->stack_depth; - backtrace->stack_depth = stack_depth + 1; - if ( stack_depth >= MAX_DEPTH ) { - return; - } - backtrace->frame[stack_depth] = call_site; -} - -void __cyg_profile_func_exit(void *func_address, - void *call_site) ATTRIBUTE_NOINSTRUMENT; -void __cyg_profile_func_exit(void *func_address, void *call_site) { - (void)func_address; - (void)call_site; - - BACK_TRACE* backtrace = thread_back_trace + gettid(); - int stack_depth = backtrace->stack_depth; - backtrace->stack_depth = stack_depth - 1; - if ( stack_depth >= MAX_DEPTH ) { - return; - } - backtrace->frame[stack_depth] = 0; -} -} // extern "C" - -static int cyg_backtrace(void **buffer, int size) { - BACK_TRACE* backtrace = thread_back_trace + gettid(); - int stack_depth = backtrace->stack_depth; - if ( stack_depth >= MAX_DEPTH ) { - stack_depth = MAX_DEPTH; - } - int nSize = (size > stack_depth) ? stack_depth : size; - for (int i = 0; i < nSize; i++) { - buffer[i] = backtrace->frame[nSize - i - 1]; - } - - return nSize; -} - -#endif // BASE_STACKTRACE_INSTRUMENT_INL_H_ - - -// Note: this part of the file is included several times. -// Do not put globals below. - -// The following 4 functions are generated from the code below: -// GetStack{Trace,Frames}() -// GetStack{Trace,Frames}WithContext() -// -// These functions take the following args: -// void** result: the stack-trace, as an array -// int* sizes: the size of each stack frame, as an array -// (GetStackFrames* only) -// int max_depth: the size of the result (and sizes) array(s) -// int skip_count: how many stack pointers to skip before storing in result -// void* ucp: a ucontext_t* (GetStack{Trace,Frames}WithContext only) -static int GET_STACK_TRACE_OR_FRAMES { - static const int kStackLength = 64; - void * stack[kStackLength]; - int size; - memset(stack, 0, sizeof(stack)); - - size = cyg_backtrace(stack, kStackLength); - skip_count += 2; // we want to skip the current and parent frame as well - int result_count = size - skip_count; - if (result_count < 0) - result_count = 0; - if (result_count > max_depth) - result_count = max_depth; - for (int i = 0; i < result_count; i++) - result[i] = stack[i + skip_count]; - -#if IS_STACK_FRAMES - // No implementation for finding out the stack frame sizes yet. - memset(sizes, 0, sizeof(*sizes) * result_count); -#endif - - return result_count; -} diff --git a/contrib/libtcmalloc/src/stacktrace_libgcc-inl.h b/contrib/libtcmalloc/src/stacktrace_libgcc-inl.h deleted file mode 100644 index ce9cf5196ad..00000000000 --- a/contrib/libtcmalloc/src/stacktrace_libgcc-inl.h +++ /dev/null @@ -1,111 +0,0 @@ -// -*- Mode: C++; c-basic-offset: 2; indent-tabs-mode: nil -*- -// Copyright (c) 2016, gperftools Contributors -// All rights reserved. -// -// Redistribution and use in source and binary forms, with or without -// modification, are permitted provided that the following conditions are -// met: -// -// * Redistributions of source code must retain the above copyright -// notice, this list of conditions and the following disclaimer. -// * Redistributions in binary form must reproduce the above -// copyright notice, this list of conditions and the following disclaimer -// in the documentation and/or other materials provided with the -// distribution. -// * Neither the name of Google Inc. nor the names of its -// contributors may be used to endorse or promote products derived from -// this software without specific prior written permission. -// -// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS -// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT -// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR -// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT -// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, -// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT -// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, -// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY -// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT -// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE -// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. - -// This file implements backtrace capturing via libgcc's -// _Unwind_Backtrace. This generally works almost always. It will fail -// sometimes when we're trying to capture backtrace from signal -// handler (i.e. in cpu profiler) while some C++ code is throwing -// exception. - -#ifndef BASE_STACKTRACE_LIBGCC_INL_H_ -#define BASE_STACKTRACE_LIBGCC_INL_H_ -// Note: this file is included into stacktrace.cc more than once. -// Anything that should only be defined once should be here: - -extern "C" { -#include -#include // for memset() -} - -#include - -#include "gperftools/stacktrace.h" - -struct libgcc_backtrace_data { - void **array; - int skip; - int pos; - int limit; -}; - -static _Unwind_Reason_Code libgcc_backtrace_helper(struct _Unwind_Context *ctx, - void *_data) { - libgcc_backtrace_data *data = - reinterpret_cast(_data); - - if (data->skip > 0) { - data->skip--; - return _URC_NO_REASON; - } - - if (data->pos < data->limit) { - void *ip = reinterpret_cast(_Unwind_GetIP(ctx));; - data->array[data->pos++] = ip; - } - - return _URC_NO_REASON; -} - -#endif // BASE_STACKTRACE_LIBGCC_INL_H_ - -// Note: this part of the file is included several times. -// Do not put globals below. - -// The following 4 functions are generated from the code below: -// GetStack{Trace,Frames}() -// GetStack{Trace,Frames}WithContext() -// -// These functions take the following args: -// void** result: the stack-trace, as an array -// int* sizes: the size of each stack frame, as an array -// (GetStackFrames* only) -// int max_depth: the size of the result (and sizes) array(s) -// int skip_count: how many stack pointers to skip before storing in result -// void* ucp: a ucontext_t* (GetStack{Trace,Frames}WithContext only) -static int GET_STACK_TRACE_OR_FRAMES { - libgcc_backtrace_data data; - data.array = result; - // we're also skipping current and parent's frame - data.skip = skip_count + 2; - data.pos = 0; - data.limit = max_depth; - - _Unwind_Backtrace(libgcc_backtrace_helper, &data); - - if (data.pos > 1 && data.array[data.pos - 1] == NULL) - --data.pos; - -#if IS_STACK_FRAMES - // No implementation for finding out the stack frame sizes. - memset(sizes, 0, sizeof(*sizes) * data.pos); -#endif - - return data.pos; -} diff --git a/contrib/libtcmalloc/src/stacktrace_libunwind-inl.h b/contrib/libtcmalloc/src/stacktrace_libunwind-inl.h deleted file mode 100644 index e8257af6c2c..00000000000 --- a/contrib/libtcmalloc/src/stacktrace_libunwind-inl.h +++ /dev/null @@ -1,152 +0,0 @@ -// -*- Mode: C++; c-basic-offset: 2; indent-tabs-mode: nil -*- -// Copyright (c) 2005, Google Inc. -// All rights reserved. -// -// Redistribution and use in source and binary forms, with or without -// modification, are permitted provided that the following conditions are -// met: -// -// * Redistributions of source code must retain the above copyright -// notice, this list of conditions and the following disclaimer. -// * Redistributions in binary form must reproduce the above -// copyright notice, this list of conditions and the following disclaimer -// in the documentation and/or other materials provided with the -// distribution. -// * Neither the name of Google Inc. nor the names of its -// contributors may be used to endorse or promote products derived from -// this software without specific prior written permission. -// -// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS -// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT -// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR -// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT -// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, -// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT -// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, -// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY -// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT -// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE -// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. - -// --- -// Author: Arun Sharma -// -// Produce stack trace using libunwind - -#ifndef BASE_STACKTRACE_LIBINWIND_INL_H_ -#define BASE_STACKTRACE_LIBINWIND_INL_H_ -// Note: this file is included into stacktrace.cc more than once. -// Anything that should only be defined once should be here: - -// We only need local unwinder. -#define UNW_LOCAL_ONLY - -extern "C" { -#include -#include // for memset() -#include -} -#include "gperftools/stacktrace.h" - -#include "base/basictypes.h" -#include "base/logging.h" - -// Sometimes, we can try to get a stack trace from within a stack -// trace, because libunwind can call mmap (maybe indirectly via an -// internal mmap based memory allocator), and that mmap gets trapped -// and causes a stack-trace request. If were to try to honor that -// recursive request, we'd end up with infinite recursion or deadlock. -// Luckily, it's safe to ignore those subsequent traces. In such -// cases, we return 0 to indicate the situation. -static __thread int recursive ATTR_INITIAL_EXEC; - -#if defined(TCMALLOC_ENABLE_UNWIND_FROM_UCONTEXT) && (defined(__i386__) || defined(__x86_64__)) && defined(__GNU_LIBRARY__) -#define BASE_STACKTRACE_UNW_CONTEXT_IS_UCONTEXT 1 -#endif - -#endif // BASE_STACKTRACE_LIBINWIND_INL_H_ - -// Note: this part of the file is included several times. -// Do not put globals below. - -// The following 4 functions are generated from the code below: -// GetStack{Trace,Frames}() -// GetStack{Trace,Frames}WithContext() -// -// These functions take the following args: -// void** result: the stack-trace, as an array -// int* sizes: the size of each stack frame, as an array -// (GetStackFrames* only) -// int max_depth: the size of the result (and sizes) array(s) -// int skip_count: how many stack pointers to skip before storing in result -// void* ucp: a ucontext_t* (GetStack{Trace,Frames}WithContext only) -static int GET_STACK_TRACE_OR_FRAMES { - void *ip; - int n = 0; - unw_cursor_t cursor; - unw_context_t uc; -#if IS_STACK_FRAMES - unw_word_t sp = 0, next_sp = 0; -#endif - - if (recursive) { - return 0; - } - ++recursive; - -#if (IS_WITH_CONTEXT && defined(BASE_STACKTRACE_UNW_CONTEXT_IS_UCONTEXT)) - if (ucp) { - uc = *(static_cast(const_cast(ucp))); - /* this is a bit weird. profiler.cc calls us with signal's ucontext - * yet passing us 2 as skip_count and essentially assuming we won't - * use ucontext. */ - /* In order to fix that I'm going to assume that if ucp is - * non-null we're asked to ignore skip_count in case we're - * able to use ucp */ - skip_count = 0; - } else { - unw_getcontext(&uc); - skip_count += 2; // Do not include current and parent frame - } -#else - unw_getcontext(&uc); - skip_count += 2; // Do not include current and parent frame -#endif - - /*int ret =*/ unw_init_local(&cursor, &uc); - //assert(ret >= 0); - - while (skip_count--) { - if (unw_step(&cursor) <= 0) { - goto out; - } -#if IS_STACK_FRAMES - if (unw_get_reg(&cursor, UNW_REG_SP, &next_sp)) { - goto out; - } -#endif - } - - while (n < max_depth) { - if (unw_get_reg(&cursor, UNW_REG_IP, (unw_word_t *) &ip) < 0) { - break; - } -#if IS_STACK_FRAMES - sizes[n] = 0; -#endif - result[n++] = ip; - if (unw_step(&cursor) <= 0) { - break; - } -#if IS_STACK_FRAMES - sp = next_sp; - if (unw_get_reg(&cursor, UNW_REG_SP, &next_sp) , 0) { - break; - } - sizes[n - 1] = next_sp - sp; -#endif - } -out: - --recursive; - return n; -} diff --git a/contrib/libtcmalloc/src/stacktrace_x86-inl.h b/contrib/libtcmalloc/src/stacktrace_x86-inl.h deleted file mode 100644 index 46eb5d82d71..00000000000 --- a/contrib/libtcmalloc/src/stacktrace_x86-inl.h +++ /dev/null @@ -1,354 +0,0 @@ -// -*- Mode: C++; c-basic-offset: 2; indent-tabs-mode: nil -*- -// Copyright (c) 2005, Google Inc. -// All rights reserved. -// -// Redistribution and use in source and binary forms, with or without -// modification, are permitted provided that the following conditions are -// met: -// -// * Redistributions of source code must retain the above copyright -// notice, this list of conditions and the following disclaimer. -// * Redistributions in binary form must reproduce the above -// copyright notice, this list of conditions and the following disclaimer -// in the documentation and/or other materials provided with the -// distribution. -// * Neither the name of Google Inc. nor the names of its -// contributors may be used to endorse or promote products derived from -// this software without specific prior written permission. -// -// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS -// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT -// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR -// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT -// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, -// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT -// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, -// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY -// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT -// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE -// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. - -// --- -// Author: Sanjay Ghemawat -// -// Produce stack trace - -#ifndef BASE_STACKTRACE_X86_INL_H_ -#define BASE_STACKTRACE_X86_INL_H_ -// Note: this file is included into stacktrace.cc more than once. -// Anything that should only be defined once should be here: - -#include "config.h" -#include // for NULL -#include -#if defined(HAVE_SYS_UCONTEXT_H) -#include -#elif defined(HAVE_UCONTEXT_H) -#include // for ucontext_t -#elif defined(HAVE_CYGWIN_SIGNAL_H) -// cygwin/signal.h has a buglet where it uses pthread_attr_t without -// #including itself. So we have to do it. -# ifdef HAVE_PTHREAD -# include -# endif -#include -typedef ucontext ucontext_t; -#endif -#ifdef HAVE_STDINT_H -#include // for uintptr_t -#endif -#ifdef HAVE_UNISTD_H -#include -#endif -#ifdef HAVE_MMAP -#include // for msync -#include "base/vdso_support.h" -#endif - -#include "gperftools/stacktrace.h" - -#if defined(__linux__) && defined(__i386__) && defined(__ELF__) && defined(HAVE_MMAP) -// Count "push %reg" instructions in VDSO __kernel_vsyscall(), -// preceding "syscall" or "sysenter". -// If __kernel_vsyscall uses frame pointer, answer 0. -// -// kMaxBytes tells how many instruction bytes of __kernel_vsyscall -// to analyze before giving up. Up to kMaxBytes+1 bytes of -// instructions could be accessed. -// -// Here are known __kernel_vsyscall instruction sequences: -// -// SYSENTER (linux-2.6.26/arch/x86/vdso/vdso32/sysenter.S). -// Used on Intel. -// 0xffffe400 <__kernel_vsyscall+0>: push %ecx -// 0xffffe401 <__kernel_vsyscall+1>: push %edx -// 0xffffe402 <__kernel_vsyscall+2>: push %ebp -// 0xffffe403 <__kernel_vsyscall+3>: mov %esp,%ebp -// 0xffffe405 <__kernel_vsyscall+5>: sysenter -// -// SYSCALL (see linux-2.6.26/arch/x86/vdso/vdso32/syscall.S). -// Used on AMD. -// 0xffffe400 <__kernel_vsyscall+0>: push %ebp -// 0xffffe401 <__kernel_vsyscall+1>: mov %ecx,%ebp -// 0xffffe403 <__kernel_vsyscall+3>: syscall -// -// i386 (see linux-2.6.26/arch/x86/vdso/vdso32/int80.S) -// 0xffffe400 <__kernel_vsyscall+0>: int $0x80 -// 0xffffe401 <__kernel_vsyscall+1>: ret -// -static const int kMaxBytes = 10; - -// We use assert()s instead of DCHECK()s -- this is too low level -// for DCHECK(). - -static int CountPushInstructions(const unsigned char *const addr) { - int result = 0; - for (int i = 0; i < kMaxBytes; ++i) { - if (addr[i] == 0x89) { - // "mov reg,reg" - if (addr[i + 1] == 0xE5) { - // Found "mov %esp,%ebp". - return 0; - } - ++i; // Skip register encoding byte. - } else if (addr[i] == 0x0F && - (addr[i + 1] == 0x34 || addr[i + 1] == 0x05)) { - // Found "sysenter" or "syscall". - return result; - } else if ((addr[i] & 0xF0) == 0x50) { - // Found "push %reg". - ++result; - } else if (addr[i] == 0xCD && addr[i + 1] == 0x80) { - // Found "int $0x80" - assert(result == 0); - return 0; - } else { - // Unexpected instruction. - assert(0 == "unexpected instruction in __kernel_vsyscall"); - return 0; - } - } - // Unexpected: didn't find SYSENTER or SYSCALL in - // [__kernel_vsyscall, __kernel_vsyscall + kMaxBytes) interval. - assert(0 == "did not find SYSENTER or SYSCALL in __kernel_vsyscall"); - return 0; -} -#endif - -// Given a pointer to a stack frame, locate and return the calling -// stackframe, or return NULL if no stackframe can be found. Perform sanity -// checks (the strictness of which is controlled by the boolean parameter -// "STRICT_UNWINDING") to reduce the chance that a bad pointer is returned. -template -static void **NextStackFrame(void **old_sp, const void *uc) { - void **new_sp = (void **) *old_sp; - -#if defined(__linux__) && defined(__i386__) && defined(HAVE_VDSO_SUPPORT) - if (WITH_CONTEXT && uc != NULL) { - // How many "push %reg" instructions are there at __kernel_vsyscall? - // This is constant for a given kernel and processor, so compute - // it only once. - static int num_push_instructions = -1; // Sentinel: not computed yet. - // Initialize with sentinel value: __kernel_rt_sigreturn can not possibly - // be there. - static const unsigned char *kernel_rt_sigreturn_address = NULL; - static const unsigned char *kernel_vsyscall_address = NULL; - if (num_push_instructions == -1) { - base::VDSOSupport vdso; - if (vdso.IsPresent()) { - base::VDSOSupport::SymbolInfo rt_sigreturn_symbol_info; - base::VDSOSupport::SymbolInfo vsyscall_symbol_info; - if (!vdso.LookupSymbol("__kernel_rt_sigreturn", "LINUX_2.5", - STT_FUNC, &rt_sigreturn_symbol_info) || - !vdso.LookupSymbol("__kernel_vsyscall", "LINUX_2.5", - STT_FUNC, &vsyscall_symbol_info) || - rt_sigreturn_symbol_info.address == NULL || - vsyscall_symbol_info.address == NULL) { - // Unexpected: 32-bit VDSO is present, yet one of the expected - // symbols is missing or NULL. - assert(0 == "VDSO is present, but doesn't have expected symbols"); - num_push_instructions = 0; - } else { - kernel_rt_sigreturn_address = - reinterpret_cast( - rt_sigreturn_symbol_info.address); - kernel_vsyscall_address = - reinterpret_cast( - vsyscall_symbol_info.address); - num_push_instructions = - CountPushInstructions(kernel_vsyscall_address); - } - } else { - num_push_instructions = 0; - } - } - if (num_push_instructions != 0 && kernel_rt_sigreturn_address != NULL && - old_sp[1] == kernel_rt_sigreturn_address) { - const ucontext_t *ucv = static_cast(uc); - // This kernel does not use frame pointer in its VDSO code, - // and so %ebp is not suitable for unwinding. - void **const reg_ebp = - reinterpret_cast(ucv->uc_mcontext.gregs[REG_EBP]); - const unsigned char *const reg_eip = - reinterpret_cast(ucv->uc_mcontext.gregs[REG_EIP]); - if (new_sp == reg_ebp && - kernel_vsyscall_address <= reg_eip && - reg_eip - kernel_vsyscall_address < kMaxBytes) { - // We "stepped up" to __kernel_vsyscall, but %ebp is not usable. - // Restore from 'ucv' instead. - void **const reg_esp = - reinterpret_cast(ucv->uc_mcontext.gregs[REG_ESP]); - // Check that alleged %esp is not NULL and is reasonably aligned. - if (reg_esp && - ((uintptr_t)reg_esp & (sizeof(reg_esp) - 1)) == 0) { - // Check that alleged %esp is actually readable. This is to prevent - // "double fault" in case we hit the first fault due to e.g. stack - // corruption. - // - // page_size is linker-initalized to avoid async-unsafe locking - // that GCC would otherwise insert (__cxa_guard_acquire etc). - static int page_size; - if (page_size == 0) { - // First time through. - page_size = getpagesize(); - } - void *const reg_esp_aligned = - reinterpret_cast( - (uintptr_t)(reg_esp + num_push_instructions - 1) & - ~(page_size - 1)); - if (msync(reg_esp_aligned, page_size, MS_ASYNC) == 0) { - // Alleged %esp is readable, use it for further unwinding. - new_sp = reinterpret_cast( - reg_esp[num_push_instructions - 1]); - } - } - } - } - } -#endif - - // Check that the transition from frame pointer old_sp to frame - // pointer new_sp isn't clearly bogus - if (STRICT_UNWINDING) { - // With the stack growing downwards, older stack frame must be - // at a greater address that the current one. - if (new_sp <= old_sp) return NULL; - // Assume stack frames larger than 100,000 bytes are bogus. - if ((uintptr_t)new_sp - (uintptr_t)old_sp > 100000) return NULL; - } else { - // In the non-strict mode, allow discontiguous stack frames. - // (alternate-signal-stacks for example). - if (new_sp == old_sp) return NULL; - if (new_sp > old_sp) { - // And allow frames upto about 1MB. - const uintptr_t delta = (uintptr_t)new_sp - (uintptr_t)old_sp; - const uintptr_t acceptable_delta = 1000000; - if (delta > acceptable_delta) { - return NULL; - } - } - } - if ((uintptr_t)new_sp & (sizeof(void *) - 1)) return NULL; -#ifdef __i386__ - // On 64-bit machines, the stack pointer can be very close to - // 0xffffffff, so we explicitly check for a pointer into the - // last two pages in the address space - if ((uintptr_t)new_sp >= 0xffffe000) return NULL; -#endif -#ifdef HAVE_MMAP - if (!STRICT_UNWINDING) { - // Lax sanity checks cause a crash on AMD-based machines with - // VDSO-enabled kernels. - // Make an extra sanity check to insure new_sp is readable. - // Note: NextStackFrame() is only called while the program - // is already on its last leg, so it's ok to be slow here. - static int page_size = getpagesize(); - void *new_sp_aligned = (void *)((uintptr_t)new_sp & ~(page_size - 1)); - if (msync(new_sp_aligned, page_size, MS_ASYNC) == -1) - return NULL; - } -#endif - return new_sp; -} - -#endif // BASE_STACKTRACE_X86_INL_H_ - -// Note: this part of the file is included several times. -// Do not put globals below. - -// The following 4 functions are generated from the code below: -// GetStack{Trace,Frames}() -// GetStack{Trace,Frames}WithContext() -// -// These functions take the following args: -// void** result: the stack-trace, as an array -// int* sizes: the size of each stack frame, as an array -// (GetStackFrames* only) -// int max_depth: the size of the result (and sizes) array(s) -// int skip_count: how many stack pointers to skip before storing in result -// void* ucp: a ucontext_t* (GetStack{Trace,Frames}WithContext only) - -static int GET_STACK_TRACE_OR_FRAMES { - void **sp; -#if (__GNUC__ > 4) || (__GNUC__ == 4 && __GNUC_MINOR__ >= 2) || __llvm__ - // __builtin_frame_address(0) can return the wrong address on gcc-4.1.0-k8. - // It's always correct on llvm, and the techniques below aren't (in - // particular, llvm-gcc will make a copy of pcs, so it's not in sp[2]), - // so we also prefer __builtin_frame_address when running under llvm. - sp = reinterpret_cast(__builtin_frame_address(0)); -#elif defined(__i386__) - // Stack frame format: - // sp[0] pointer to previous frame - // sp[1] caller address - // sp[2] first argument - // ... - // NOTE: This will break under llvm, since result is a copy and not in sp[2] - sp = (void **)&result - 2; -#elif defined(__x86_64__) - unsigned long rbp; - // Move the value of the register %rbp into the local variable rbp. - // We need 'volatile' to prevent this instruction from getting moved - // around during optimization to before function prologue is done. - // An alternative way to achieve this - // would be (before this __asm__ instruction) to call Noop() defined as - // static void Noop() __attribute__ ((noinline)); // prevent inlining - // static void Noop() { asm(""); } // prevent optimizing-away - __asm__ volatile ("mov %%rbp, %0" : "=r" (rbp)); - // Arguments are passed in registers on x86-64, so we can't just - // offset from &result - sp = (void **) rbp; -#else -# error Using stacktrace_x86-inl.h on a non x86 architecture! -#endif - - skip_count++; // skip parent's frame due to indirection in stacktrace.cc - - int n = 0; - while (sp && n < max_depth) { - if (*(sp+1) == reinterpret_cast(0)) { - // In 64-bit code, we often see a frame that - // points to itself and has a return address of 0. - break; - } -#if !IS_WITH_CONTEXT - const void *const ucp = NULL; -#endif - void **next_sp = NextStackFrame(sp, ucp); - if (skip_count > 0) { - skip_count--; - } else { - result[n] = *(sp+1); -#if IS_STACK_FRAMES - if (next_sp > sp) { - sizes[n] = (uintptr_t)next_sp - (uintptr_t)sp; - } else { - // A frame-size of 0 is used to indicate unknown frame size. - sizes[n] = 0; - } -#endif - n++; - } - sp = next_sp; - } - return n; -} diff --git a/contrib/libtcmalloc/src/static_vars.cc b/contrib/libtcmalloc/src/static_vars.cc deleted file mode 100644 index 1e29d339996..00000000000 --- a/contrib/libtcmalloc/src/static_vars.cc +++ /dev/null @@ -1,125 +0,0 @@ -// -*- Mode: C++; c-basic-offset: 2; indent-tabs-mode: nil -*- -// Copyright (c) 2008, Google Inc. -// All rights reserved. -// -// Redistribution and use in source and binary forms, with or without -// modification, are permitted provided that the following conditions are -// met: -// -// * Redistributions of source code must retain the above copyright -// notice, this list of conditions and the following disclaimer. -// * Redistributions in binary form must reproduce the above -// copyright notice, this list of conditions and the following disclaimer -// in the documentation and/or other materials provided with the -// distribution. -// * Neither the name of Google Inc. nor the names of its -// contributors may be used to endorse or promote products derived from -// this software without specific prior written permission. -// -// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS -// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT -// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR -// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT -// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, -// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT -// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, -// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY -// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT -// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE -// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. - -// --- -// Author: Ken Ashcraft - -#include "config.h" -#include "static_vars.h" -#include // for NULL -#include // for operator new -#ifdef HAVE_PTHREAD -#include // for pthread_atfork -#endif -#include "internal_logging.h" // for CHECK_CONDITION -#include "common.h" -#include "sampler.h" // for Sampler -#include "getenv_safe.h" // TCMallocGetenvSafe -#include "base/googleinit.h" -#include "maybe_threads.h" - -namespace tcmalloc { - -#if defined(HAVE_FORK) && defined(HAVE_PTHREAD) -// These following two functions are registered via pthread_atfork to make -// sure the central_cache locks remain in a consisten state in the forked -// version of the thread. - -void CentralCacheLockAll() -{ - Static::pageheap_lock()->Lock(); - for (int i = 0; i < kNumClasses; ++i) - Static::central_cache()[i].Lock(); -} - -void CentralCacheUnlockAll() -{ - for (int i = 0; i < kNumClasses; ++i) - Static::central_cache()[i].Unlock(); - Static::pageheap_lock()->Unlock(); -} -#endif - -SpinLock Static::pageheap_lock_(SpinLock::LINKER_INITIALIZED); -SizeMap Static::sizemap_; -CentralFreeListPadded Static::central_cache_[kNumClasses]; -PageHeapAllocator Static::span_allocator_; -PageHeapAllocator Static::stacktrace_allocator_; -Span Static::sampled_objects_; -PageHeapAllocator Static::bucket_allocator_; -StackTrace* Static::growth_stacks_ = NULL; -PageHeap* Static::pageheap_ = NULL; - - -void Static::InitStaticVars() { - sizemap_.Init(); - span_allocator_.Init(); - span_allocator_.New(); // Reduce cache conflicts - span_allocator_.New(); // Reduce cache conflicts - stacktrace_allocator_.Init(); - bucket_allocator_.Init(); - // Do a bit of sanitizing: make sure central_cache is aligned properly - CHECK_CONDITION((sizeof(central_cache_[0]) % 64) == 0); - for (int i = 0; i < kNumClasses; ++i) { - central_cache_[i].Init(i); - } - - // It's important to have PageHeap allocated, not in static storage, - // so that HeapLeakChecker does not consider all the byte patterns stored - // in is caches as pointers that are sources of heap object liveness, - // which leads to it missing some memory leaks. - pageheap_ = new (MetaDataAlloc(sizeof(PageHeap))) PageHeap; - - bool aggressive_decommit = - tcmalloc::commandlineflags::StringToBool( - TCMallocGetenvSafe("TCMALLOC_AGGRESSIVE_DECOMMIT"), true); - - pageheap_->SetAggressiveDecommit(aggressive_decommit); - - DLL_Init(&sampled_objects_); - Sampler::InitStatics(); -} - - -#if defined(HAVE_FORK) && defined(HAVE_PTHREAD) && !defined(__APPLE__) - -static inline -void SetupAtForkLocksHandler() -{ - perftools_pthread_atfork( - CentralCacheLockAll, // parent calls before fork - CentralCacheUnlockAll, // parent calls after fork - CentralCacheUnlockAll); // child calls after fork -} -REGISTER_MODULE_INITIALIZER(tcmalloc_fork_handler, SetupAtForkLocksHandler()); - -#endif - -} // namespace tcmalloc diff --git a/contrib/libtcmalloc/src/static_vars.h b/contrib/libtcmalloc/src/static_vars.h deleted file mode 100644 index d6dfa334ab2..00000000000 --- a/contrib/libtcmalloc/src/static_vars.h +++ /dev/null @@ -1,115 +0,0 @@ -// -*- Mode: C++; c-basic-offset: 2; indent-tabs-mode: nil -*- -// Copyright (c) 2008, Google Inc. -// All rights reserved. -// -// Redistribution and use in source and binary forms, with or without -// modification, are permitted provided that the following conditions are -// met: -// -// * Redistributions of source code must retain the above copyright -// notice, this list of conditions and the following disclaimer. -// * Redistributions in binary form must reproduce the above -// copyright notice, this list of conditions and the following disclaimer -// in the documentation and/or other materials provided with the -// distribution. -// * Neither the name of Google Inc. nor the names of its -// contributors may be used to endorse or promote products derived from -// this software without specific prior written permission. -// -// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS -// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT -// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR -// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT -// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, -// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT -// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, -// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY -// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT -// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE -// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. - -// --- -// Author: Ken Ashcraft -// -// Static variables shared by multiple classes. - -#ifndef TCMALLOC_STATIC_VARS_H_ -#define TCMALLOC_STATIC_VARS_H_ - -#include "config.h" -#include "base/spinlock.h" -#include "central_freelist.h" -#include "common.h" -#include "page_heap.h" -#include "page_heap_allocator.h" -#include "span.h" -#include "stack_trace_table.h" - -namespace tcmalloc { - -class Static { - public: - // Linker initialized, so this lock can be accessed at any time. - static SpinLock* pageheap_lock() { return &pageheap_lock_; } - - // Must be called before calling any of the accessors below. - static void InitStaticVars(); - - // Central cache -- an array of free-lists, one per size-class. - // We have a separate lock per free-list to reduce contention. - static CentralFreeListPadded* central_cache() { return central_cache_; } - - static SizeMap* sizemap() { return &sizemap_; } - - ////////////////////////////////////////////////////////////////////// - // In addition to the explicit initialization comment, the variables below - // must be protected by pageheap_lock. - - // Page-level allocator. - static PageHeap* pageheap() { return pageheap_; } - - static PageHeapAllocator* span_allocator() { return &span_allocator_; } - - static PageHeapAllocator* stacktrace_allocator() { - return &stacktrace_allocator_; - } - - static StackTrace* growth_stacks() { return growth_stacks_; } - static void set_growth_stacks(StackTrace* s) { growth_stacks_ = s; } - - // State kept for sampled allocations (/pprof/heap support) - static Span* sampled_objects() { return &sampled_objects_; } - static PageHeapAllocator* bucket_allocator() { - return &bucket_allocator_; - } - - // Check if InitStaticVars() has been run. - static bool IsInited() { return pageheap() != NULL; } - - private: - static SpinLock pageheap_lock_; - - // These static variables require explicit initialization. We cannot - // count on their constructors to do any initialization because other - // static variables may try to allocate memory before these variables - // can run their constructors. - - static SizeMap sizemap_; - static CentralFreeListPadded central_cache_[kNumClasses]; - static PageHeapAllocator span_allocator_; - static PageHeapAllocator stacktrace_allocator_; - static Span sampled_objects_; - static PageHeapAllocator bucket_allocator_; - - // Linked list of stack traces recorded every time we allocated memory - // from the system. Useful for finding allocation sites that cause - // increase in the footprint of the system. The linked list pointer - // is stored in trace->stack[kMaxStackDepth-1]. - static StackTrace* growth_stacks_; - - static PageHeap* pageheap_; -}; - -} // namespace tcmalloc - -#endif // TCMALLOC_STATIC_VARS_H_ diff --git a/contrib/libtcmalloc/src/symbolize.cc b/contrib/libtcmalloc/src/symbolize.cc deleted file mode 100644 index a27106e8bce..00000000000 --- a/contrib/libtcmalloc/src/symbolize.cc +++ /dev/null @@ -1,285 +0,0 @@ -// -*- Mode: C++; c-basic-offset: 2; indent-tabs-mode: nil -*- -// Copyright (c) 2009, Google Inc. -// All rights reserved. -// -// Redistribution and use in source and binary forms, with or without -// modification, are permitted provided that the following conditions are -// met: -// -// * Redistributions of source code must retain the above copyright -// notice, this list of conditions and the following disclaimer. -// * Redistributions in binary form must reproduce the above -// copyright notice, this list of conditions and the following disclaimer -// in the documentation and/or other materials provided with the -// distribution. -// * Neither the name of Google Inc. nor the names of its -// contributors may be used to endorse or promote products derived from -// this software without specific prior written permission. -// -// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS -// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT -// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR -// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT -// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, -// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT -// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, -// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY -// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT -// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE -// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. - -// --- -// Author: Craig Silverstein -// -// This forks out to pprof to do the actual symbolizing. We might -// be better off writing our own in C++. - -#include "config.h" -#include "symbolize.h" -#include -#ifdef HAVE_UNISTD_H -#include // for write() -#endif -#ifdef HAVE_SYS_SOCKET_H -#include // for socketpair() -- needed by Symbolize -#endif -#ifdef HAVE_SYS_WAIT_H -#include // for wait() -- needed by Symbolize -#endif -#ifdef HAVE_POLL_H -#include -#endif -#ifdef __MACH__ -#include // for GetProgramInvocationName() -#include // for PATH_MAX -#endif -#if defined(__CYGWIN__) || defined(__CYGWIN32__) -#include // for get_osfhandle() -#endif -#include -#include "base/commandlineflags.h" -#include "base/logging.h" -#include "base/sysinfo.h" - -using std::string; -using tcmalloc::DumpProcSelfMaps; // from sysinfo.h - - -DEFINE_string(symbolize_pprof, - EnvToString("PPROF_PATH", "pprof"), - "Path to pprof to call for reporting function names."); - -// heap_profile_table_pprof may be referenced after destructors are -// called (since that's when leak-checking is done), so we make -// a more-permanent copy that won't ever get destroyed. -static string* g_pprof_path = new string(FLAGS_symbolize_pprof); - -// Returns NULL if we're on an OS where we can't get the invocation name. -// Using a static var is ok because we're not called from a thread. -static const char* GetProgramInvocationName() { -#if defined(HAVE_PROGRAM_INVOCATION_NAME) -#ifdef __UCLIBC__ - extern const char* program_invocation_name; // uclibc provides this -#else - extern char* program_invocation_name; // gcc provides this -#endif - return program_invocation_name; -#elif defined(__MACH__) - // We don't want to allocate memory for this since we may be - // calculating it when memory is corrupted. - static char program_invocation_name[PATH_MAX]; - if (program_invocation_name[0] == '\0') { // first time calculating - uint32_t length = sizeof(program_invocation_name); - if (_NSGetExecutablePath(program_invocation_name, &length)) - return NULL; - } - return program_invocation_name; -#else - return NULL; // figure out a way to get argv[0] -#endif -} - -// Prints an error message when you can't run Symbolize(). -static void PrintError(const char* reason) { - RAW_LOG(ERROR, - "*** WARNING: Cannot convert addresses to symbols in output below.\n" - "*** Reason: %s\n" - "*** If you cannot fix this, try running pprof directly.\n", - reason); -} - -void SymbolTable::Add(const void* addr) { - symbolization_table_[addr] = ""; -} - -const char* SymbolTable::GetSymbol(const void* addr) { - return symbolization_table_[addr]; -} - -// Updates symbolization_table with the pointers to symbol names corresponding -// to its keys. The symbol names are stored in out, which is allocated and -// freed by the caller of this routine. -// Note that the forking/etc is not thread-safe or re-entrant. That's -// ok for the purpose we need -- reporting leaks detected by heap-checker -// -- but be careful if you decide to use this routine for other purposes. -// Returns number of symbols read on error. If can't symbolize, returns 0 -// and emits an error message about why. -int SymbolTable::Symbolize() { -#if !defined(HAVE_UNISTD_H) || !defined(HAVE_SYS_SOCKET_H) || !defined(HAVE_SYS_WAIT_H) - PrintError("Perftools does not know how to call a sub-process on this O/S"); - return 0; -#else - const char* argv0 = GetProgramInvocationName(); - if (argv0 == NULL) { // can't call symbolize if we can't figure out our name - PrintError("Cannot figure out the name of this executable (argv0)"); - return 0; - } - if (access(g_pprof_path->c_str(), R_OK) != 0) { - PrintError("Cannot find 'pprof' (is PPROF_PATH set correctly?)"); - return 0; - } - - // All this work is to do two-way communication. ugh. - int *child_in = NULL; // file descriptors - int *child_out = NULL; // for now, we don't worry about child_err - int child_fds[5][2]; // socketpair may be called up to five times below - - // The client program may close its stdin and/or stdout and/or stderr - // thus allowing socketpair to reuse file descriptors 0, 1 or 2. - // In this case the communication between the forked processes may be broken - // if either the parent or the child tries to close or duplicate these - // descriptors. The loop below produces two pairs of file descriptors, each - // greater than 2 (stderr). - for (int i = 0; i < 5; i++) { - if (socketpair(AF_UNIX, SOCK_STREAM, 0, child_fds[i]) == -1) { - for (int j = 0; j < i; j++) { - close(child_fds[j][0]); - close(child_fds[j][1]); - PrintError("Cannot create a socket pair"); - } - return 0; - } else { - if ((child_fds[i][0] > 2) && (child_fds[i][1] > 2)) { - if (child_in == NULL) { - child_in = child_fds[i]; - } else { - child_out = child_fds[i]; - for (int j = 0; j < i; j++) { - if (child_fds[j] == child_in) continue; - close(child_fds[j][0]); - close(child_fds[j][1]); - } - break; - } - } - } - } - - switch (fork()) { - case -1: { // error - close(child_in[0]); - close(child_in[1]); - close(child_out[0]); - close(child_out[1]); - PrintError("Unknown error calling fork()"); - return 0; - } - case 0: { // child - close(child_in[1]); // child uses the 0's, parent uses the 1's - close(child_out[1]); // child uses the 0's, parent uses the 1's - close(0); - close(1); - if (dup2(child_in[0], 0) == -1) _exit(1); - if (dup2(child_out[0], 1) == -1) _exit(2); - // Unset vars that might cause trouble when we fork - unsetenv("CPUPROFILE"); - unsetenv("HEAPPROFILE"); - unsetenv("HEAPCHECK"); - unsetenv("PERFTOOLS_VERBOSE"); - execlp(g_pprof_path->c_str(), g_pprof_path->c_str(), - "--symbols", argv0, NULL); - _exit(3); // if execvp fails, it's bad news for us - } - default: { // parent - close(child_in[0]); // child uses the 0's, parent uses the 1's - close(child_out[0]); // child uses the 0's, parent uses the 1's -#ifdef HAVE_POLL_H - // Waiting for 1ms seems to give the OS time to notice any errors. - poll(0, 0, 1); - // For maximum safety, we check to make sure the execlp - // succeeded before trying to write. (Otherwise we'll get a - // SIGPIPE.) For systems without poll.h, we'll just skip this - // check, and trust that the user set PPROF_PATH correctly! - struct pollfd pfd = { child_in[1], POLLOUT, 0 }; - if (!poll(&pfd, 1, 0) || !(pfd.revents & POLLOUT) || - (pfd.revents & (POLLHUP|POLLERR))) { - PrintError("Cannot run 'pprof' (is PPROF_PATH set correctly?)"); - return 0; - } -#endif -#if defined(__CYGWIN__) || defined(__CYGWIN32__) - // On cygwin, DumpProcSelfMaps() takes a HANDLE, not an fd. Convert. - const HANDLE symbols_handle = (HANDLE) get_osfhandle(child_in[1]); - DumpProcSelfMaps(symbols_handle); -#else - DumpProcSelfMaps(child_in[1]); // what pprof expects on stdin -#endif - - // Allocate 24 bytes = ("0x" + 8 bytes + "\n" + overhead) for each - // address to feed to pprof. - const int kOutBufSize = 24 * symbolization_table_.size(); - char *pprof_buffer = new char[kOutBufSize]; - int written = 0; - for (SymbolMap::const_iterator iter = symbolization_table_.begin(); - iter != symbolization_table_.end(); ++iter) { - written += snprintf(pprof_buffer + written, kOutBufSize - written, - // pprof expects format to be 0xXXXXXX - "0x%" PRIxPTR "\n", reinterpret_cast(iter->first)); - } - write(child_in[1], pprof_buffer, strlen(pprof_buffer)); - close(child_in[1]); // that's all we need to write - - const int kSymbolBufferSize = kSymbolSize * symbolization_table_.size(); - int total_bytes_read = 0; - delete[] symbol_buffer_; - symbol_buffer_ = new char[kSymbolBufferSize]; - memset(symbol_buffer_, '\0', kSymbolBufferSize); - while (1) { - int bytes_read = read(child_out[1], symbol_buffer_ + total_bytes_read, - kSymbolBufferSize - total_bytes_read); - if (bytes_read < 0) { - close(child_out[1]); - PrintError("Cannot read data from pprof"); - return 0; - } else if (bytes_read == 0) { - close(child_out[1]); - wait(NULL); - break; - } else { - total_bytes_read += bytes_read; - } - } - // We have successfully read the output of pprof into out. Make sure - // the last symbol is full (we can tell because it ends with a \n). - if (total_bytes_read == 0 || symbol_buffer_[total_bytes_read - 1] != '\n') - return 0; - // make the symbolization_table_ values point to the output vector - SymbolMap::iterator fill = symbolization_table_.begin(); - int num_symbols = 0; - const char *current_name = symbol_buffer_; - for (int i = 0; i < total_bytes_read; i++) { - if (symbol_buffer_[i] == '\n') { - fill->second = current_name; - symbol_buffer_[i] = '\0'; - current_name = symbol_buffer_ + i + 1; - fill++; - num_symbols++; - } - } - return num_symbols; - } - } - PrintError("Unkown error (should never occur!)"); - return 0; // shouldn't be reachable -#endif -} diff --git a/contrib/libtcmalloc/src/symbolize.h b/contrib/libtcmalloc/src/symbolize.h deleted file mode 100644 index 728d073308a..00000000000 --- a/contrib/libtcmalloc/src/symbolize.h +++ /dev/null @@ -1,84 +0,0 @@ -// -*- Mode: C++; c-basic-offset: 2; indent-tabs-mode: nil -*- -// Copyright (c) 2009, Google Inc. -// All rights reserved. -// -// Redistribution and use in source and binary forms, with or without -// modification, are permitted provided that the following conditions are -// met: -// -// * Redistributions of source code must retain the above copyright -// notice, this list of conditions and the following disclaimer. -// * Redistributions in binary form must reproduce the above -// copyright notice, this list of conditions and the following disclaimer -// in the documentation and/or other materials provided with the -// distribution. -// * Neither the name of Google Inc. nor the names of its -// contributors may be used to endorse or promote products derived from -// this software without specific prior written permission. -// -// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS -// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT -// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR -// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT -// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, -// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT -// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, -// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY -// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT -// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE -// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. - -// --- -// Author: Craig Silverstein - -#ifndef TCMALLOC_SYMBOLIZE_H_ -#define TCMALLOC_SYMBOLIZE_H_ - -#include "config.h" -#ifdef HAVE_STDINT_H -#include // for uintptr_t -#endif -#include // for NULL -#include - -using std::map; - -// SymbolTable encapsulates the address operations necessary for stack trace -// symbolization. A common use-case is to Add() the addresses from one or -// several stack traces to a table, call Symbolize() once and use GetSymbol() -// to get the symbol names for pretty-printing the stack traces. -class SymbolTable { - public: - SymbolTable() - : symbol_buffer_(NULL) {} - ~SymbolTable() { - delete[] symbol_buffer_; - } - - // Adds an address to the table. This may overwrite a currently known symbol - // name, so Add() should not generally be called after Symbolize(). - void Add(const void* addr); - - // Returns the symbol name for addr, if the given address was added before - // the last successful call to Symbolize(). Otherwise may return an empty - // c-string. - const char* GetSymbol(const void* addr); - - // Obtains the symbol names for the addresses stored in the table and returns - // the number of addresses actually symbolized. - int Symbolize(); - - private: - typedef map SymbolMap; - - // An average size of memory allocated for a stack trace symbol. - static const int kSymbolSize = 1024; - - // Map from addresses to symbol names. - SymbolMap symbolization_table_; - - // Pointer to the buffer that stores the symbol names. - char *symbol_buffer_; -}; - -#endif // TCMALLOC_SYMBOLIZE_H_ diff --git a/contrib/libtcmalloc/src/system-alloc.cc b/contrib/libtcmalloc/src/system-alloc.cc deleted file mode 100644 index 1356513c1ce..00000000000 --- a/contrib/libtcmalloc/src/system-alloc.cc +++ /dev/null @@ -1,567 +0,0 @@ -// -*- Mode: C++; c-basic-offset: 2; indent-tabs-mode: nil -*- -// Copyright (c) 2005, Google Inc. -// All rights reserved. -// -// Redistribution and use in source and binary forms, with or without -// modification, are permitted provided that the following conditions are -// met: -// -// * Redistributions of source code must retain the above copyright -// notice, this list of conditions and the following disclaimer. -// * Redistributions in binary form must reproduce the above -// copyright notice, this list of conditions and the following disclaimer -// in the documentation and/or other materials provided with the -// distribution. -// * Neither the name of Google Inc. nor the names of its -// contributors may be used to endorse or promote products derived from -// this software without specific prior written permission. -// -// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS -// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT -// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR -// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT -// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, -// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT -// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, -// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY -// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT -// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE -// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. - -// --- -// Author: Sanjay Ghemawat - -#include "config.h" -#include // for EAGAIN, errno -#include // for open, O_RDWR -#include // for size_t, NULL, ptrdiff_t -#if defined HAVE_STDINT_H -#include // for uintptr_t, intptr_t -#elif defined HAVE_INTTYPES_H -#include -#else -#include -#endif -#ifdef HAVE_MMAP -#include // for munmap, mmap, MADV_DONTNEED, etc -#endif -#ifdef HAVE_UNISTD_H -#include // for sbrk, getpagesize, off_t -#endif -#include // for operator new -#include -#include "base/basictypes.h" -#include "base/commandlineflags.h" -#include "base/spinlock.h" // for SpinLockHolder, SpinLock, etc -#include "common.h" -#include "internal_logging.h" - -#pragma GCC diagnostic push -#pragma GCC diagnostic ignored "-Wunused-function" - -// On systems (like freebsd) that don't define MAP_ANONYMOUS, use the old -// form of the name instead. -#ifndef MAP_ANONYMOUS -# define MAP_ANONYMOUS MAP_ANON -#endif - -// Linux added support for MADV_FREE in 4.5 but we aren't ready to use it -// yet. Among other things, using compile-time detection leads to poor -// results when compiling on a system with MADV_FREE and running on a -// system without it. See https://github.com/gperftools/gperftools/issues/780. -#if defined(__linux__) && defined(MADV_FREE) && !defined(TCMALLOC_USE_MADV_FREE) -# undef MADV_FREE -#endif - -// MADV_FREE is specifically designed for use by malloc(), but only -// FreeBSD supports it; in linux we fall back to the somewhat inferior -// MADV_DONTNEED. -#if !defined(MADV_FREE) && defined(MADV_DONTNEED) -# define MADV_FREE MADV_DONTNEED -#endif - -// Solaris has a bug where it doesn't declare madvise() for C++. -// http://www.opensolaris.org/jive/thread.jspa?threadID=21035&tstart=0 -#if defined(__sun) && defined(__SVR4) -# include // for caddr_t - extern "C" { extern int madvise(caddr_t, size_t, int); } -#endif - -// Set kDebugMode mode so that we can have use C++ conditionals -// instead of preprocessor conditionals. -#ifdef NDEBUG -static const bool kDebugMode = false; -#else -static const bool kDebugMode = true; -#endif - -// TODO(sanjay): Move the code below into the tcmalloc namespace -using tcmalloc::kLog; -using tcmalloc::Log; - -// Anonymous namespace to avoid name conflicts on "CheckAddressBits". -namespace { - -// Check that no bit is set at position ADDRESS_BITS or higher. -template bool CheckAddressBits(uintptr_t ptr) { - return (ptr >> ADDRESS_BITS) == 0; -} - -// Specialize for the bit width of a pointer to avoid undefined shift. -template <> bool CheckAddressBits<8 * sizeof(void*)>(uintptr_t ptr) { - return true; -} - -} // Anonymous namespace to avoid name conflicts on "CheckAddressBits". - -COMPILE_ASSERT(kAddressBits <= 8 * sizeof(void*), - address_bits_larger_than_pointer_size); - -static SpinLock spinlock(SpinLock::LINKER_INITIALIZED); - -#if defined(HAVE_MMAP) || defined(MADV_FREE) -// Page size is initialized on demand (only needed for mmap-based allocators) -static size_t pagesize = 0; -#endif - -// The current system allocator -SysAllocator* sys_alloc = NULL; - -// Number of bytes taken from system. -size_t TCMalloc_SystemTaken = 0; - -// Configuration parameters. -DEFINE_int32(malloc_devmem_start, - EnvToInt("TCMALLOC_DEVMEM_START", 0), - "Physical memory starting location in MB for /dev/mem allocation." - " Setting this to 0 disables /dev/mem allocation"); -DEFINE_int32(malloc_devmem_limit, - EnvToInt("TCMALLOC_DEVMEM_LIMIT", 0), - "Physical memory limit location in MB for /dev/mem allocation." - " Setting this to 0 means no limit."); -DEFINE_bool(malloc_skip_sbrk, - EnvToBool("TCMALLOC_SKIP_SBRK", false), - "Whether sbrk can be used to obtain memory."); -DEFINE_bool(malloc_skip_mmap, - EnvToBool("TCMALLOC_SKIP_MMAP", false), - "Whether mmap can be used to obtain memory."); -DEFINE_bool(malloc_disable_memory_release, - EnvToBool("TCMALLOC_DISABLE_MEMORY_RELEASE", false), - "Whether MADV_FREE/MADV_DONTNEED should be used" - " to return unused memory to the system."); - -// static allocators -class SbrkSysAllocator : public SysAllocator { -public: - SbrkSysAllocator() : SysAllocator() { - } - void* Alloc(size_t size, size_t *actual_size, size_t alignment); -}; -static union { - char buf[sizeof(SbrkSysAllocator)]; - void *ptr; -} sbrk_space; - -class MmapSysAllocator : public SysAllocator { -public: - MmapSysAllocator() : SysAllocator() { - } - void* Alloc(size_t size, size_t *actual_size, size_t alignment); -}; -static union { - char buf[sizeof(MmapSysAllocator)]; - void *ptr; -} mmap_space; - -class DevMemSysAllocator : public SysAllocator { -public: - DevMemSysAllocator() : SysAllocator() { - } - void* Alloc(size_t size, size_t *actual_size, size_t alignment); -}; - -class DefaultSysAllocator : public SysAllocator { - public: - DefaultSysAllocator() : SysAllocator() { - for (int i = 0; i < kMaxAllocators; i++) { - failed_[i] = true; - allocs_[i] = NULL; - names_[i] = NULL; - } - } - void SetChildAllocator(SysAllocator* alloc, unsigned int index, - const char* name) { - if (index < kMaxAllocators && alloc != NULL) { - allocs_[index] = alloc; - failed_[index] = false; - names_[index] = name; - } - } - void* Alloc(size_t size, size_t *actual_size, size_t alignment); - - private: - static const int kMaxAllocators = 2; - bool failed_[kMaxAllocators]; - SysAllocator* allocs_[kMaxAllocators]; - const char* names_[kMaxAllocators]; -}; -static union { - char buf[sizeof(DefaultSysAllocator)]; - void *ptr; -} default_space; -static const char sbrk_name[] = "SbrkSysAllocator"; -static const char mmap_name[] = "MmapSysAllocator"; - - -void* SbrkSysAllocator::Alloc(size_t size, size_t *actual_size, - size_t alignment) { -#if !defined(HAVE_SBRK) || defined(__UCLIBC__) - return NULL; -#else - // Check if we should use sbrk allocation. - // FLAGS_malloc_skip_sbrk starts out as false (its uninitialized - // state) and eventually gets initialized to the specified value. Note - // that this code runs for a while before the flags are initialized. - // That means that even if this flag is set to true, some (initial) - // memory will be allocated with sbrk before the flag takes effect. - if (FLAGS_malloc_skip_sbrk) { - return NULL; - } - - // sbrk will release memory if passed a negative number, so we do - // a strict check here - if (static_cast(size + alignment) < 0) return NULL; - - // This doesn't overflow because TCMalloc_SystemAlloc has already - // tested for overflow at the alignment boundary. - size = ((size + alignment - 1) / alignment) * alignment; - - // "actual_size" indicates that the bytes from the returned pointer - // p up to and including (p + actual_size - 1) have been allocated. - if (actual_size) { - *actual_size = size; - } - - // Check that we we're not asking for so much more memory that we'd - // wrap around the end of the virtual address space. (This seems - // like something sbrk() should check for us, and indeed opensolaris - // does, but glibc does not: - // http://src.opensolaris.org/source/xref/onnv/onnv-gate/usr/src/lib/libc/port/sys/sbrk.c?a=true - // http://sourceware.org/cgi-bin/cvsweb.cgi/~checkout~/libc/misc/sbrk.c?rev=1.1.2.1&content-type=text/plain&cvsroot=glibc - // Without this check, sbrk may succeed when it ought to fail.) - if (reinterpret_cast(sbrk(0)) + size < size) { - return NULL; - } - - void* result = sbrk(size); - if (result == reinterpret_cast(-1)) { - return NULL; - } - - // Is it aligned? - uintptr_t ptr = reinterpret_cast(result); - if ((ptr & (alignment-1)) == 0) return result; - - // Try to get more memory for alignment - size_t extra = alignment - (ptr & (alignment-1)); - void* r2 = sbrk(extra); - if (reinterpret_cast(r2) == (ptr + size)) { - // Contiguous with previous result - return reinterpret_cast(ptr + extra); - } - - // Give up and ask for "size + alignment - 1" bytes so - // that we can find an aligned region within it. - result = sbrk(size + alignment - 1); - if (result == reinterpret_cast(-1)) { - return NULL; - } - ptr = reinterpret_cast(result); - if ((ptr & (alignment-1)) != 0) { - ptr += alignment - (ptr & (alignment-1)); - } - return reinterpret_cast(ptr); -#endif // HAVE_SBRK -} - -void* MmapSysAllocator::Alloc(size_t size, size_t *actual_size, - size_t alignment) { -#ifndef HAVE_MMAP - return NULL; -#else - // Check if we should use mmap allocation. - // FLAGS_malloc_skip_mmap starts out as false (its uninitialized - // state) and eventually gets initialized to the specified value. Note - // that this code runs for a while before the flags are initialized. - // Chances are we never get here before the flags are initialized since - // sbrk is used until the heap is exhausted (before mmap is used). - if (FLAGS_malloc_skip_mmap) { - return NULL; - } - - // Enforce page alignment - if (pagesize == 0) pagesize = getpagesize(); - if (alignment < pagesize) alignment = pagesize; - size_t aligned_size = ((size + alignment - 1) / alignment) * alignment; - if (aligned_size < size) { - return NULL; - } - size = aligned_size; - - // "actual_size" indicates that the bytes from the returned pointer - // p up to and including (p + actual_size - 1) have been allocated. - if (actual_size) { - *actual_size = size; - } - - // Ask for extra memory if alignment > pagesize - size_t extra = 0; - if (alignment > pagesize) { - extra = alignment - pagesize; - } - - // Note: size + extra does not overflow since: - // size + alignment < (1<(MAP_FAILED)) { - return NULL; - } - - // Adjust the return memory so it is aligned - uintptr_t ptr = reinterpret_cast(result); - size_t adjust = 0; - if ((ptr & (alignment - 1)) != 0) { - adjust = alignment - (ptr & (alignment - 1)); - } - - // Return the unused memory to the system - if (adjust > 0) { - munmap(reinterpret_cast(ptr), adjust); - } - if (adjust < extra) { - munmap(reinterpret_cast(ptr + adjust + size), extra - adjust); - } - - ptr += adjust; - return reinterpret_cast(ptr); -#endif // HAVE_MMAP -} - -void* DevMemSysAllocator::Alloc(size_t size, size_t *actual_size, - size_t alignment) { -#ifndef HAVE_MMAP - return NULL; -#else - static bool initialized = false; - static off_t physmem_base; // next physical memory address to allocate - static off_t physmem_limit; // maximum physical address allowed - static int physmem_fd; // file descriptor for /dev/mem - - // Check if we should use /dev/mem allocation. Note that it may take - // a while to get this flag initialized, so meanwhile we fall back to - // the next allocator. (It looks like 7MB gets allocated before - // this flag gets initialized -khr.) - if (FLAGS_malloc_devmem_start == 0) { - // NOTE: not a devmem_failure - we'd like TCMalloc_SystemAlloc to - // try us again next time. - return NULL; - } - - if (!initialized) { - physmem_fd = open("/dev/mem", O_RDWR); - if (physmem_fd < 0) { - return NULL; - } - physmem_base = FLAGS_malloc_devmem_start*1024LL*1024LL; - physmem_limit = FLAGS_malloc_devmem_limit*1024LL*1024LL; - initialized = true; - } - - // Enforce page alignment - if (pagesize == 0) pagesize = getpagesize(); - if (alignment < pagesize) alignment = pagesize; - size_t aligned_size = ((size + alignment - 1) / alignment) * alignment; - if (aligned_size < size) { - return NULL; - } - size = aligned_size; - - // "actual_size" indicates that the bytes from the returned pointer - // p up to and including (p + actual_size - 1) have been allocated. - if (actual_size) { - *actual_size = size; - } - - // Ask for extra memory if alignment > pagesize - size_t extra = 0; - if (alignment > pagesize) { - extra = alignment - pagesize; - } - - // check to see if we have any memory left - if (physmem_limit != 0 && - ((size + extra) > (physmem_limit - physmem_base))) { - return NULL; - } - - // Note: size + extra does not overflow since: - // size + alignment < (1<(MAP_FAILED)) { - return NULL; - } - uintptr_t ptr = reinterpret_cast(result); - - // Adjust the return memory so it is aligned - size_t adjust = 0; - if ((ptr & (alignment - 1)) != 0) { - adjust = alignment - (ptr & (alignment - 1)); - } - - // Return the unused virtual memory to the system - if (adjust > 0) { - munmap(reinterpret_cast(ptr), adjust); - } - if (adjust < extra) { - munmap(reinterpret_cast(ptr + adjust + size), extra - adjust); - } - - ptr += adjust; - physmem_base += adjust + size; - - return reinterpret_cast(ptr); -#endif // HAVE_MMAP -} - -void* DefaultSysAllocator::Alloc(size_t size, size_t *actual_size, - size_t alignment) { - for (int i = 0; i < kMaxAllocators; i++) { - if (!failed_[i] && allocs_[i] != NULL) { - void* result = allocs_[i]->Alloc(size, actual_size, alignment); - if (result != NULL) { - return result; - } - failed_[i] = true; - } - } - // After both failed, reset "failed_" to false so that a single failed - // allocation won't make the allocator never work again. - for (int i = 0; i < kMaxAllocators; i++) { - failed_[i] = false; - } - return NULL; -} - -ATTRIBUTE_WEAK ATTRIBUTE_NOINLINE -SysAllocator *tc_get_sysalloc_override(SysAllocator *def) -{ - return def; -} - -static bool system_alloc_inited = false; -void InitSystemAllocators(void) { - MmapSysAllocator *mmap = new (mmap_space.buf) MmapSysAllocator(); - SbrkSysAllocator *sbrk = new (sbrk_space.buf) SbrkSysAllocator(); - - // In 64-bit debug mode, place the mmap allocator first since it - // allocates pointers that do not fit in 32 bits and therefore gives - // us better testing of code's 64-bit correctness. It also leads to - // less false negatives in heap-checking code. (Numbers are less - // likely to look like pointers and therefore the conservative gc in - // the heap-checker is less likely to misinterpret a number as a - // pointer). - DefaultSysAllocator *sdef = new (default_space.buf) DefaultSysAllocator(); - if (kDebugMode && sizeof(void*) > 4) { - sdef->SetChildAllocator(mmap, 0, mmap_name); - sdef->SetChildAllocator(sbrk, 1, sbrk_name); - } else { - sdef->SetChildAllocator(sbrk, 0, sbrk_name); - sdef->SetChildAllocator(mmap, 1, mmap_name); - } - - sys_alloc = tc_get_sysalloc_override(sdef); -} - -void* TCMalloc_SystemAlloc(size_t size, size_t *actual_size, - size_t alignment) { - // Discard requests that overflow - if (size + alignment < size) return NULL; - - SpinLockHolder lock_holder(&spinlock); - - if (!system_alloc_inited) { - InitSystemAllocators(); - system_alloc_inited = true; - } - - // Enforce minimum alignment - if (alignment < sizeof(MemoryAligner)) alignment = sizeof(MemoryAligner); - - size_t actual_size_storage; - if (actual_size == NULL) { - actual_size = &actual_size_storage; - } - - void* result = sys_alloc->Alloc(size, actual_size, alignment); - if (result != NULL) { - CHECK_CONDITION( - CheckAddressBits( - reinterpret_cast(result) + *actual_size - 1)); - TCMalloc_SystemTaken += *actual_size; - } - return result; -} - -bool TCMalloc_SystemRelease(void* start, size_t length) { -#ifdef MADV_FREE - if (FLAGS_malloc_devmem_start) { - // It's not safe to use MADV_FREE/MADV_DONTNEED if we've been - // mapping /dev/mem for heap memory. - return false; - } - if (FLAGS_malloc_disable_memory_release) return false; - if (pagesize == 0) pagesize = getpagesize(); - const size_t pagemask = pagesize - 1; - - size_t new_start = reinterpret_cast(start); - size_t end = new_start + length; - size_t new_end = end; - - // Round up the starting address and round down the ending address - // to be page aligned: - new_start = (new_start + pagesize - 1) & ~pagemask; - new_end = new_end & ~pagemask; - - ASSERT((new_start & pagemask) == 0); - ASSERT((new_end & pagemask) == 0); - ASSERT(new_start >= reinterpret_cast(start)); - ASSERT(new_end <= end); - - if (new_end > new_start) { - int result; - do { - result = madvise(reinterpret_cast(new_start), - new_end - new_start, MADV_FREE); - } while (result == -1 && errno == EAGAIN); - - return result != -1; - } -#endif - return false; -} - -void TCMalloc_SystemCommit(void* start, size_t length) { - // Nothing to do here. TCMalloc_SystemRelease does not alter pages - // such that they need to be re-committed before they can be used by the - // application. -} - -#pragma GCC diagnostic pop diff --git a/contrib/libtcmalloc/src/system-alloc.h b/contrib/libtcmalloc/src/system-alloc.h deleted file mode 100644 index 2c06c183d63..00000000000 --- a/contrib/libtcmalloc/src/system-alloc.h +++ /dev/null @@ -1,92 +0,0 @@ -// -*- Mode: C++; c-basic-offset: 2; indent-tabs-mode: nil -*- -// Copyright (c) 2005, Google Inc. -// All rights reserved. -// -// Redistribution and use in source and binary forms, with or without -// modification, are permitted provided that the following conditions are -// met: -// -// * Redistributions of source code must retain the above copyright -// notice, this list of conditions and the following disclaimer. -// * Redistributions in binary form must reproduce the above -// copyright notice, this list of conditions and the following disclaimer -// in the documentation and/or other materials provided with the -// distribution. -// * Neither the name of Google Inc. nor the names of its -// contributors may be used to endorse or promote products derived from -// this software without specific prior written permission. -// -// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS -// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT -// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR -// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT -// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, -// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT -// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, -// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY -// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT -// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE -// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. - -// --- -// Author: Sanjay Ghemawat -// -// Routine that uses sbrk/mmap to allocate memory from the system. -// Useful for implementing malloc. - -#ifndef TCMALLOC_SYSTEM_ALLOC_H_ -#define TCMALLOC_SYSTEM_ALLOC_H_ - -#include "config.h" -#include // for size_t - -class SysAllocator; - -// REQUIRES: "alignment" is a power of two or "0" to indicate default alignment -// -// Allocate and return "N" bytes of zeroed memory. -// -// If actual_bytes is NULL then the returned memory is exactly the -// requested size. If actual bytes is non-NULL then the allocator -// may optionally return more bytes than asked for (i.e. return an -// entire "huge" page if a huge page allocator is in use). -// -// The returned pointer is a multiple of "alignment" if non-zero. The -// returned pointer will always be aligned suitably for holding a -// void*, double, or size_t. In addition, if this platform defines -// CACHELINE_ALIGNED, the return pointer will always be cacheline -// aligned. -// -// Returns NULL when out of memory. -extern PERFTOOLS_DLL_DECL -void* TCMalloc_SystemAlloc(size_t bytes, size_t *actual_bytes, - size_t alignment = 0); - -// This call is a hint to the operating system that the pages -// contained in the specified range of memory will not be used for a -// while, and can be released for use by other processes or the OS. -// Pages which are released in this way may be destroyed (zeroed) by -// the OS. The benefit of this function is that it frees memory for -// use by the system, the cost is that the pages are faulted back into -// the address space next time they are touched, which can impact -// performance. (Only pages fully covered by the memory region will -// be released, partial pages will not.) -// -// Returns false if release failed or not supported. -extern PERFTOOLS_DLL_DECL -bool TCMalloc_SystemRelease(void* start, size_t length); - -// Called to ressurect memory which has been previously released -// to the system via TCMalloc_SystemRelease. An attempt to -// commit a page that is already committed does not cause this -// function to fail. -extern PERFTOOLS_DLL_DECL -void TCMalloc_SystemCommit(void* start, size_t length); - -// The current system allocator. -extern PERFTOOLS_DLL_DECL SysAllocator* sys_alloc; - -// Number of bytes taken from system. -extern PERFTOOLS_DLL_DECL size_t TCMalloc_SystemTaken; - -#endif /* TCMALLOC_SYSTEM_ALLOC_H_ */ diff --git a/contrib/libtcmalloc/src/tcmalloc.cc b/contrib/libtcmalloc/src/tcmalloc.cc deleted file mode 100644 index b52524b1361..00000000000 --- a/contrib/libtcmalloc/src/tcmalloc.cc +++ /dev/null @@ -1,1842 +0,0 @@ -// -*- Mode: C++; c-basic-offset: 2; indent-tabs-mode: nil -*- -// Copyright (c) 2005, Google Inc. -// All rights reserved. -// -// Redistribution and use in source and binary forms, with or without -// modification, are permitted provided that the following conditions are -// met: -// -// * Redistributions of source code must retain the above copyright -// notice, this list of conditions and the following disclaimer. -// * Redistributions in binary form must reproduce the above -// copyright notice, this list of conditions and the following disclaimer -// in the documentation and/or other materials provided with the -// distribution. -// * Neither the name of Google Inc. nor the names of its -// contributors may be used to endorse or promote products derived from -// this software without specific prior written permission. -// -// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS -// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT -// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR -// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT -// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, -// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT -// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, -// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY -// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT -// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE -// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. - -// --- -// Author: Sanjay Ghemawat -// -// A malloc that uses a per-thread cache to satisfy small malloc requests. -// (The time for malloc/free of a small object drops from 300 ns to 50 ns.) -// -// See doc/tcmalloc.html for a high-level -// description of how this malloc works. -// -// SYNCHRONIZATION -// 1. The thread-specific lists are accessed without acquiring any locks. -// This is safe because each such list is only accessed by one thread. -// 2. We have a lock per central free-list, and hold it while manipulating -// the central free list for a particular size. -// 3. The central page allocator is protected by "pageheap_lock". -// 4. The pagemap (which maps from page-number to descriptor), -// can be read without holding any locks, and written while holding -// the "pageheap_lock". -// 5. To improve performance, a subset of the information one can get -// from the pagemap is cached in a data structure, pagemap_cache_, -// that atomically reads and writes its entries. This cache can be -// read and written without locking. -// -// This multi-threaded access to the pagemap is safe for fairly -// subtle reasons. We basically assume that when an object X is -// allocated by thread A and deallocated by thread B, there must -// have been appropriate synchronization in the handoff of object -// X from thread A to thread B. The same logic applies to pagemap_cache_. -// -// THE PAGEID-TO-SIZECLASS CACHE -// Hot PageID-to-sizeclass mappings are held by pagemap_cache_. If this cache -// returns 0 for a particular PageID then that means "no information," not that -// the sizeclass is 0. The cache may have stale information for pages that do -// not hold the beginning of any free()'able object. Staleness is eliminated -// in Populate() for pages with sizeclass > 0 objects, and in do_malloc() and -// do_memalign() for all other relevant pages. -// -// PAGEMAP -// ------- -// Page map contains a mapping from page id to Span. -// -// If Span s occupies pages [p..q], -// pagemap[p] == s -// pagemap[q] == s -// pagemap[p+1..q-1] are undefined -// pagemap[p-1] and pagemap[q+1] are defined: -// NULL if the corresponding page is not yet in the address space. -// Otherwise it points to a Span. This span may be free -// or allocated. If free, it is in one of pageheap's freelist. -// -// TODO: Bias reclamation to larger addresses -// TODO: implement mallinfo/mallopt -// TODO: Better testing -// -// 9/28/2003 (new page-level allocator replaces ptmalloc2): -// * malloc/free of small objects goes from ~300 ns to ~50 ns. -// * allocation of a reasonably complicated struct -// goes from about 1100 ns to about 300 ns. - -#include "config.h" -#include - -#include // for ENOMEM, EINVAL, errno -#if defined HAVE_STDINT_H -#include -#elif defined HAVE_INTTYPES_H -#include -#else -#include -#endif -#include // for size_t, NULL -#include // for getenv -#include // for strcmp, memset, strlen, etc -#ifdef HAVE_UNISTD_H -#include // for getpagesize, write, etc -#endif -#include // for max, min -#include // for numeric_limits -#include // for nothrow_t (ptr only), etc -#include // for vector - -#include -#include // for MallocHook -#include "base/basictypes.h" // for int64 -#include "base/commandlineflags.h" // for RegisterFlagValidator, etc -#include "base/dynamic_annotations.h" // for RunningOnValgrind -#include "base/spinlock.h" // for SpinLockHolder -#include "central_freelist.h" // for CentralFreeListPadded -#include "common.h" // for StackTrace, kPageShift, etc -#include "internal_logging.h" // for ASSERT, TCMalloc_Printer, etc -#include "linked_list.h" // for SLL_SetNext -#include "malloc_hook-inl.h" // for MallocHook::InvokeNewHook, etc -#include "page_heap.h" // for PageHeap, PageHeap::Stats -#include "page_heap_allocator.h" // for PageHeapAllocator -#include "span.h" // for Span, DLL_Prepend, etc -#include "stack_trace_table.h" // for StackTraceTable -#include "static_vars.h" // for Static -#include "system-alloc.h" // for DumpSystemAllocatorStats, etc -#include "tcmalloc_guard.h" // for TCMallocGuard -#include "thread_cache.h" // for ThreadCache - -#ifdef __clang__ -// clang's apparent focus on code size somehow causes it to ignore -// normal inline directives even for few functions which inlining is -// key for performance. In order to get performance of clang's -// generated code closer to normal, we're forcing inlining via -// attribute. -#define ALWAYS_INLINE inline __attribute__((always_inline)) -#else -#define ALWAYS_INLINE inline -#endif - -#include "maybe_emergency_malloc.h" - -#if (defined(_WIN32) && !defined(__CYGWIN__) && !defined(__CYGWIN32__)) && !defined(WIN32_OVERRIDE_ALLOCATORS) -# define WIN32_DO_PATCHING 1 -#endif - -// Some windows file somewhere (at least on cygwin) #define's small (!) -#undef small - -using STL_NAMESPACE::max; -using STL_NAMESPACE::numeric_limits; -using STL_NAMESPACE::vector; - -#include "libc_override.h" - -#pragma GCC diagnostic push -#pragma GCC diagnostic ignored "-Wunused-function" - -using tcmalloc::AlignmentForSize; -using tcmalloc::kLog; -using tcmalloc::kCrash; -using tcmalloc::kCrashWithStats; -using tcmalloc::Log; -using tcmalloc::PageHeap; -using tcmalloc::PageHeapAllocator; -using tcmalloc::SizeMap; -using tcmalloc::Span; -using tcmalloc::StackTrace; -using tcmalloc::Static; -using tcmalloc::ThreadCache; - -DECLARE_double(tcmalloc_release_rate); - -// For windows, the printf we use to report large allocs is -// potentially dangerous: it could cause a malloc that would cause an -// infinite loop. So by default we set the threshold to a huge number -// on windows, so this bad situation will never trigger. You can -// always set TCMALLOC_LARGE_ALLOC_REPORT_THRESHOLD manually if you -// want this functionality. -#ifdef _WIN32 -const int64 kDefaultLargeAllocReportThreshold = static_cast(1) << 62; -#else -const int64 kDefaultLargeAllocReportThreshold = static_cast(1) << 30; -#endif -DEFINE_int64(tcmalloc_large_alloc_report_threshold, - EnvToInt64("TCMALLOC_LARGE_ALLOC_REPORT_THRESHOLD", - kDefaultLargeAllocReportThreshold), - "Allocations larger than this value cause a stack " - "trace to be dumped to stderr. The threshold for " - "dumping stack traces is increased by a factor of 1.125 " - "every time we print a message so that the threshold " - "automatically goes up by a factor of ~1000 every 60 " - "messages. This bounds the amount of extra logging " - "generated by this flag. Default value of this flag " - "is very large and therefore you should see no extra " - "logging unless the flag is overridden. Set to 0 to " - "disable reporting entirely."); - - -// We already declared these functions in tcmalloc.h, but we have to -// declare them again to give them an ATTRIBUTE_SECTION: we want to -// put all callers of MallocHook::Invoke* in this module into -// ATTRIBUTE_SECTION(google_malloc) section, so that -// MallocHook::GetCallerStackTrace can function accurately. -#ifndef _WIN32 // windows doesn't have attribute_section, so don't bother -extern "C" { - void* tc_malloc(size_t size) PERFTOOLS_THROW - ATTRIBUTE_SECTION(google_malloc); - void tc_free(void* ptr) PERFTOOLS_THROW - ATTRIBUTE_SECTION(google_malloc); - void* tc_realloc(void* ptr, size_t size) PERFTOOLS_THROW - ATTRIBUTE_SECTION(google_malloc); - void* tc_calloc(size_t nmemb, size_t size) PERFTOOLS_THROW - ATTRIBUTE_SECTION(google_malloc); - void tc_cfree(void* ptr) PERFTOOLS_THROW - ATTRIBUTE_SECTION(google_malloc); - - void* tc_memalign(size_t __alignment, size_t __size) PERFTOOLS_THROW - ATTRIBUTE_SECTION(google_malloc); - int tc_posix_memalign(void** ptr, size_t align, size_t size) PERFTOOLS_THROW - ATTRIBUTE_SECTION(google_malloc); - void* tc_valloc(size_t __size) PERFTOOLS_THROW - ATTRIBUTE_SECTION(google_malloc); - void* tc_pvalloc(size_t __size) PERFTOOLS_THROW - ATTRIBUTE_SECTION(google_malloc); - - void tc_malloc_stats(void) PERFTOOLS_THROW - ATTRIBUTE_SECTION(google_malloc); - int tc_mallopt(int cmd, int value) PERFTOOLS_THROW - ATTRIBUTE_SECTION(google_malloc); -#ifdef HAVE_STRUCT_MALLINFO - struct mallinfo tc_mallinfo(void) PERFTOOLS_THROW - ATTRIBUTE_SECTION(google_malloc); -#endif - - void* tc_new(size_t size) - ATTRIBUTE_SECTION(google_malloc); - void tc_delete(void* p) PERFTOOLS_THROW - ATTRIBUTE_SECTION(google_malloc); - void* tc_newarray(size_t size) - ATTRIBUTE_SECTION(google_malloc); - void tc_deletearray(void* p) PERFTOOLS_THROW - ATTRIBUTE_SECTION(google_malloc); - - // And the nothrow variants of these: - void* tc_new_nothrow(size_t size, const std::nothrow_t&) PERFTOOLS_THROW - ATTRIBUTE_SECTION(google_malloc); - void* tc_newarray_nothrow(size_t size, const std::nothrow_t&) PERFTOOLS_THROW - ATTRIBUTE_SECTION(google_malloc); - // Surprisingly, standard C++ library implementations use a - // nothrow-delete internally. See, eg: - // http://www.dinkumware.com/manuals/?manual=compleat&page=new.html - void tc_delete_nothrow(void* ptr, const std::nothrow_t&) PERFTOOLS_THROW - ATTRIBUTE_SECTION(google_malloc); - void tc_deletearray_nothrow(void* ptr, const std::nothrow_t&) PERFTOOLS_THROW - ATTRIBUTE_SECTION(google_malloc); - - // Some non-standard extensions that we support. - - // This is equivalent to - // OS X: malloc_size() - // glibc: malloc_usable_size() - // Windows: _msize() - size_t tc_malloc_size(void* p) PERFTOOLS_THROW - ATTRIBUTE_SECTION(google_malloc); -} // extern "C" -#endif // #ifndef _WIN32 - -// ----------------------- IMPLEMENTATION ------------------------------- - -static int tc_new_mode = 0; // See tc_set_new_mode(). - -// Routines such as free() and realloc() catch some erroneous pointers -// passed to them, and invoke the below when they do. (An erroneous pointer -// won't be caught if it's within a valid span or a stale span for which -// the pagemap cache has a non-zero sizeclass.) This is a cheap (source-editing -// required) kind of exception handling for these routines. -namespace { -void InvalidFree(void* ptr) { - if (tcmalloc::IsEmergencyPtr(ptr)) { - tcmalloc::EmergencyFree(ptr); - return; - } - Log(kCrash, __FILE__, __LINE__, "Attempt to free invalid pointer", ptr); -} - -size_t InvalidGetSizeForRealloc(const void* old_ptr) { - Log(kCrash, __FILE__, __LINE__, - "Attempt to realloc invalid pointer", old_ptr); - return 0; -} - -size_t InvalidGetAllocatedSize(const void* ptr) { - Log(kCrash, __FILE__, __LINE__, - "Attempt to get the size of an invalid pointer", ptr); - return 0; -} -} // unnamed namespace - -// Extract interesting stats -struct TCMallocStats { - uint64_t thread_bytes; // Bytes in thread caches - uint64_t central_bytes; // Bytes in central cache - uint64_t transfer_bytes; // Bytes in central transfer cache - uint64_t metadata_bytes; // Bytes alloced for metadata - PageHeap::Stats pageheap; // Stats from page heap -}; - -// Get stats into "r". Also, if class_count != NULL, class_count[k] -// will be set to the total number of objects of size class k in the -// central cache, transfer cache, and per-thread caches. If small_spans -// is non-NULL, it is filled. Same for large_spans. -static void ExtractStats(TCMallocStats* r, uint64_t* class_count, - PageHeap::SmallSpanStats* small_spans, - PageHeap::LargeSpanStats* large_spans) { - r->central_bytes = 0; - r->transfer_bytes = 0; - for (int cl = 0; cl < kNumClasses; ++cl) { - const int length = Static::central_cache()[cl].length(); - const int tc_length = Static::central_cache()[cl].tc_length(); - const size_t cache_overhead = Static::central_cache()[cl].OverheadBytes(); - const size_t size = static_cast( - Static::sizemap()->ByteSizeForClass(cl)); - r->central_bytes += (size * length) + cache_overhead; - r->transfer_bytes += (size * tc_length); - if (class_count) { - // Sum the lengths of all per-class freelists, except the per-thread - // freelists, which get counted when we call GetThreadStats(), below. - class_count[cl] = length + tc_length; - } - - } - - // Add stats from per-thread heaps - r->thread_bytes = 0; - { // scope - SpinLockHolder h(Static::pageheap_lock()); - ThreadCache::GetThreadStats(&r->thread_bytes, class_count); - r->metadata_bytes = tcmalloc::metadata_system_bytes(); - r->pageheap = Static::pageheap()->stats(); - if (small_spans != NULL) { - Static::pageheap()->GetSmallSpanStats(small_spans); - } - if (large_spans != NULL) { - Static::pageheap()->GetLargeSpanStats(large_spans); - } - } -} - -static double PagesToMiB(uint64_t pages) { - return (pages << kPageShift) / 1048576.0; -} - -// WRITE stats to "out" -static void DumpStats(TCMalloc_Printer* out, int level) { - TCMallocStats stats; - uint64_t class_count[kNumClasses]; - PageHeap::SmallSpanStats small; - PageHeap::LargeSpanStats large; - if (level >= 2) { - ExtractStats(&stats, class_count, &small, &large); - } else { - ExtractStats(&stats, NULL, NULL, NULL); - } - - static const double MiB = 1048576.0; - - const uint64_t virtual_memory_used = (stats.pageheap.system_bytes - + stats.metadata_bytes); - const uint64_t physical_memory_used = (virtual_memory_used - - stats.pageheap.unmapped_bytes); - const uint64_t bytes_in_use_by_app = (physical_memory_used - - stats.metadata_bytes - - stats.pageheap.free_bytes - - stats.central_bytes - - stats.transfer_bytes - - stats.thread_bytes); - -#ifdef TCMALLOC_SMALL_BUT_SLOW - out->printf( - "NOTE: SMALL MEMORY MODEL IS IN USE, PERFORMANCE MAY SUFFER.\n"); -#endif - out->printf( - "------------------------------------------------\n" - "MALLOC: %12" PRIu64 " (%7.1f MiB) Bytes in use by application\n" - "MALLOC: + %12" PRIu64 " (%7.1f MiB) Bytes in page heap freelist\n" - "MALLOC: + %12" PRIu64 " (%7.1f MiB) Bytes in central cache freelist\n" - "MALLOC: + %12" PRIu64 " (%7.1f MiB) Bytes in transfer cache freelist\n" - "MALLOC: + %12" PRIu64 " (%7.1f MiB) Bytes in thread cache freelists\n" - "MALLOC: + %12" PRIu64 " (%7.1f MiB) Bytes in malloc metadata\n" - "MALLOC: ------------\n" - "MALLOC: = %12" PRIu64 " (%7.1f MiB) Actual memory used (physical + swap)\n" - "MALLOC: + %12" PRIu64 " (%7.1f MiB) Bytes released to OS (aka unmapped)\n" - "MALLOC: ------------\n" - "MALLOC: = %12" PRIu64 " (%7.1f MiB) Virtual address space used\n" - "MALLOC:\n" - "MALLOC: %12" PRIu64 " Spans in use\n" - "MALLOC: %12" PRIu64 " Thread heaps in use\n" - "MALLOC: %12" PRIu64 " Tcmalloc page size\n" - "------------------------------------------------\n" - "Call ReleaseFreeMemory() to release freelist memory to the OS" - " (via madvise()).\n" - "Bytes released to the OS take up virtual address space" - " but no physical memory.\n", - bytes_in_use_by_app, bytes_in_use_by_app / MiB, - stats.pageheap.free_bytes, stats.pageheap.free_bytes / MiB, - stats.central_bytes, stats.central_bytes / MiB, - stats.transfer_bytes, stats.transfer_bytes / MiB, - stats.thread_bytes, stats.thread_bytes / MiB, - stats.metadata_bytes, stats.metadata_bytes / MiB, - physical_memory_used, physical_memory_used / MiB, - stats.pageheap.unmapped_bytes, stats.pageheap.unmapped_bytes / MiB, - virtual_memory_used, virtual_memory_used / MiB, - uint64_t(Static::span_allocator()->inuse()), - uint64_t(ThreadCache::HeapsInUse()), - uint64_t(kPageSize)); - - if (level >= 2) { - out->printf("------------------------------------------------\n"); - out->printf("Total size of freelists for per-thread caches,\n"); - out->printf("transfer cache, and central cache, by size class\n"); - out->printf("------------------------------------------------\n"); - uint64_t cumulative = 0; - for (int cl = 0; cl < kNumClasses; ++cl) { - if (class_count[cl] > 0) { - uint64_t class_bytes = - class_count[cl] * Static::sizemap()->ByteSizeForClass(cl); - cumulative += class_bytes; - out->printf("class %3d [ %8" PRIuS " bytes ] : " - "%8" PRIu64 " objs; %5.1f MiB; %5.1f cum MiB\n", - cl, Static::sizemap()->ByteSizeForClass(cl), - class_count[cl], - class_bytes / MiB, - cumulative / MiB); - } - } - - // append page heap info - int nonempty_sizes = 0; - for (int s = 0; s < kMaxPages; s++) { - if (small.normal_length[s] + small.returned_length[s] > 0) { - nonempty_sizes++; - } - } - out->printf("------------------------------------------------\n"); - out->printf("PageHeap: %d sizes; %6.1f MiB free; %6.1f MiB unmapped\n", - nonempty_sizes, stats.pageheap.free_bytes / MiB, - stats.pageheap.unmapped_bytes / MiB); - out->printf("------------------------------------------------\n"); - uint64_t total_normal = 0; - uint64_t total_returned = 0; - for (int s = 0; s < kMaxPages; s++) { - const int n_length = small.normal_length[s]; - const int r_length = small.returned_length[s]; - if (n_length + r_length > 0) { - uint64_t n_pages = s * n_length; - uint64_t r_pages = s * r_length; - total_normal += n_pages; - total_returned += r_pages; - out->printf("%6u pages * %6u spans ~ %6.1f MiB; %6.1f MiB cum" - "; unmapped: %6.1f MiB; %6.1f MiB cum\n", - s, - (n_length + r_length), - PagesToMiB(n_pages + r_pages), - PagesToMiB(total_normal + total_returned), - PagesToMiB(r_pages), - PagesToMiB(total_returned)); - } - } - - total_normal += large.normal_pages; - total_returned += large.returned_pages; - out->printf(">255 large * %6u spans ~ %6.1f MiB; %6.1f MiB cum" - "; unmapped: %6.1f MiB; %6.1f MiB cum\n", - static_cast(large.spans), - PagesToMiB(large.normal_pages + large.returned_pages), - PagesToMiB(total_normal + total_returned), - PagesToMiB(large.returned_pages), - PagesToMiB(total_returned)); - } -} - -static void PrintStats(int level) { - const int kBufferSize = 16 << 10; - char* buffer = new char[kBufferSize]; - TCMalloc_Printer printer(buffer, kBufferSize); - DumpStats(&printer, level); - write(STDERR_FILENO, buffer, strlen(buffer)); - delete[] buffer; -} - -static void** DumpHeapGrowthStackTraces() { - // Count how much space we need - int needed_slots = 0; - { - SpinLockHolder h(Static::pageheap_lock()); - for (StackTrace* t = Static::growth_stacks(); - t != NULL; - t = reinterpret_cast( - t->stack[tcmalloc::kMaxStackDepth-1])) { - needed_slots += 3 + t->depth; - } - needed_slots += 100; // Slop in case list grows - needed_slots += needed_slots/8; // An extra 12.5% slop - } - - void** result = new void*[needed_slots]; - if (result == NULL) { - Log(kLog, __FILE__, __LINE__, - "tcmalloc: allocation failed for stack trace slots", - needed_slots * sizeof(*result)); - return NULL; - } - - SpinLockHolder h(Static::pageheap_lock()); - int used_slots = 0; - for (StackTrace* t = Static::growth_stacks(); - t != NULL; - t = reinterpret_cast( - t->stack[tcmalloc::kMaxStackDepth-1])) { - ASSERT(used_slots < needed_slots); // Need to leave room for terminator - if (used_slots + 3 + t->depth >= needed_slots) { - // No more room - break; - } - - result[used_slots+0] = reinterpret_cast(static_cast(1)); - result[used_slots+1] = reinterpret_cast(t->size); - result[used_slots+2] = reinterpret_cast(t->depth); - for (int d = 0; d < t->depth; d++) { - result[used_slots+3+d] = t->stack[d]; - } - used_slots += 3 + t->depth; - } - result[used_slots] = reinterpret_cast(static_cast(0)); - return result; -} - -static void IterateOverRanges(void* arg, MallocExtension::RangeFunction func) { - PageID page = 1; // Some code may assume that page==0 is never used - bool done = false; - while (!done) { - // Accumulate a small number of ranges in a local buffer - static const int kNumRanges = 16; - static base::MallocRange ranges[kNumRanges]; - int n = 0; - { - SpinLockHolder h(Static::pageheap_lock()); - while (n < kNumRanges) { - if (!Static::pageheap()->GetNextRange(page, &ranges[n])) { - done = true; - break; - } else { - uintptr_t limit = ranges[n].address + ranges[n].length; - page = (limit + kPageSize - 1) >> kPageShift; - n++; - } - } - } - - for (int i = 0; i < n; i++) { - (*func)(arg, &ranges[i]); - } - } -} - -// TCMalloc's support for extra malloc interfaces -class TCMallocImplementation : public MallocExtension { - private: - // ReleaseToSystem() might release more than the requested bytes because - // the page heap releases at the span granularity, and spans are of wildly - // different sizes. This member keeps track of the extra bytes bytes - // released so that the app can periodically call ReleaseToSystem() to - // release memory at a constant rate. - // NOTE: Protected by Static::pageheap_lock(). - size_t extra_bytes_released_; - - public: - TCMallocImplementation() - : extra_bytes_released_(0) { - } - - virtual void GetStats(char* buffer, int buffer_length) { - ASSERT(buffer_length > 0); - TCMalloc_Printer printer(buffer, buffer_length); - - // Print level one stats unless lots of space is available - if (buffer_length < 10000) { - DumpStats(&printer, 1); - } else { - DumpStats(&printer, 2); - } - } - - // We may print an extra, tcmalloc-specific warning message here. - virtual void GetHeapSample(MallocExtensionWriter* writer) { - if (FLAGS_tcmalloc_sample_parameter == 0) { - const char* const kWarningMsg = - "%warn\n" - "%warn This heap profile does not have any data in it, because\n" - "%warn the application was run with heap sampling turned off.\n" - "%warn To get useful data from GetHeapSample(), you must\n" - "%warn set the environment variable TCMALLOC_SAMPLE_PARAMETER to\n" - "%warn a positive sampling period, such as 524288.\n" - "%warn\n"; - writer->append(kWarningMsg, strlen(kWarningMsg)); - } - MallocExtension::GetHeapSample(writer); - } - - virtual void** ReadStackTraces(int* sample_period) { - tcmalloc::StackTraceTable table; - { - SpinLockHolder h(Static::pageheap_lock()); - Span* sampled = Static::sampled_objects(); - for (Span* s = sampled->next; s != sampled; s = s->next) { - table.AddTrace(*reinterpret_cast(s->objects)); - } - } - *sample_period = ThreadCache::GetCache()->GetSamplePeriod(); - return table.ReadStackTracesAndClear(); // grabs and releases pageheap_lock - } - - virtual void** ReadHeapGrowthStackTraces() { - return DumpHeapGrowthStackTraces(); - } - - virtual size_t GetThreadCacheSize() { - ThreadCache* tc = ThreadCache::GetCacheIfPresent(); - if (!tc) - return 0; - return tc->Size(); - } - - virtual void MarkThreadTemporarilyIdle() { - ThreadCache::BecomeTemporarilyIdle(); - } - - virtual void Ranges(void* arg, RangeFunction func) { - IterateOverRanges(arg, func); - } - - virtual bool GetNumericProperty(const char* name, size_t* value) { - ASSERT(name != NULL); - - if (strcmp(name, "generic.current_allocated_bytes") == 0) { - TCMallocStats stats; - ExtractStats(&stats, NULL, NULL, NULL); - *value = stats.pageheap.system_bytes - - stats.thread_bytes - - stats.central_bytes - - stats.transfer_bytes - - stats.pageheap.free_bytes - - stats.pageheap.unmapped_bytes; - return true; - } - - if (strcmp(name, "generic.heap_size") == 0) { - TCMallocStats stats; - ExtractStats(&stats, NULL, NULL, NULL); - *value = stats.pageheap.system_bytes; - return true; - } - - if (strcmp(name, "tcmalloc.slack_bytes") == 0) { - // Kept for backwards compatibility. Now defined externally as: - // pageheap_free_bytes + pageheap_unmapped_bytes. - SpinLockHolder l(Static::pageheap_lock()); - PageHeap::Stats stats = Static::pageheap()->stats(); - *value = stats.free_bytes + stats.unmapped_bytes; - return true; - } - - if (strcmp(name, "tcmalloc.central_cache_free_bytes") == 0) { - TCMallocStats stats; - ExtractStats(&stats, NULL, NULL, NULL); - *value = stats.central_bytes; - return true; - } - - if (strcmp(name, "tcmalloc.transfer_cache_free_bytes") == 0) { - TCMallocStats stats; - ExtractStats(&stats, NULL, NULL, NULL); - *value = stats.transfer_bytes; - return true; - } - - if (strcmp(name, "tcmalloc.thread_cache_free_bytes") == 0) { - TCMallocStats stats; - ExtractStats(&stats, NULL, NULL, NULL); - *value = stats.thread_bytes; - return true; - } - - if (strcmp(name, "tcmalloc.pageheap_free_bytes") == 0) { - SpinLockHolder l(Static::pageheap_lock()); - *value = Static::pageheap()->stats().free_bytes; - return true; - } - - if (strcmp(name, "tcmalloc.pageheap_unmapped_bytes") == 0) { - SpinLockHolder l(Static::pageheap_lock()); - *value = Static::pageheap()->stats().unmapped_bytes; - return true; - } - - if (strcmp(name, "tcmalloc.max_total_thread_cache_bytes") == 0) { - SpinLockHolder l(Static::pageheap_lock()); - *value = ThreadCache::overall_thread_cache_size(); - return true; - } - - if (strcmp(name, "tcmalloc.current_total_thread_cache_bytes") == 0) { - TCMallocStats stats; - ExtractStats(&stats, NULL, NULL, NULL); - *value = stats.thread_bytes; - return true; - } - - if (strcmp(name, "tcmalloc.aggressive_memory_decommit") == 0) { - *value = size_t(Static::pageheap()->GetAggressiveDecommit()); - return true; - } - - return false; - } - - virtual bool SetNumericProperty(const char* name, size_t value) { - ASSERT(name != NULL); - - if (strcmp(name, "tcmalloc.max_total_thread_cache_bytes") == 0) { - SpinLockHolder l(Static::pageheap_lock()); - ThreadCache::set_overall_thread_cache_size(value); - return true; - } - - if (strcmp(name, "tcmalloc.aggressive_memory_decommit") == 0) { - Static::pageheap()->SetAggressiveDecommit(value != 0); - return true; - } - - return false; - } - - virtual void MarkThreadIdle() { - ThreadCache::BecomeIdle(); - } - - virtual void MarkThreadBusy(); // Implemented below - - virtual SysAllocator* GetSystemAllocator() { - SpinLockHolder h(Static::pageheap_lock()); - return sys_alloc; - } - - virtual void SetSystemAllocator(SysAllocator* alloc) { - SpinLockHolder h(Static::pageheap_lock()); - sys_alloc = alloc; - } - - virtual void ReleaseToSystem(size_t num_bytes) { - SpinLockHolder h(Static::pageheap_lock()); - if (num_bytes <= extra_bytes_released_) { - // We released too much on a prior call, so don't release any - // more this time. - extra_bytes_released_ = extra_bytes_released_ - num_bytes; - return; - } - num_bytes = num_bytes - extra_bytes_released_; - // num_bytes might be less than one page. If we pass zero to - // ReleaseAtLeastNPages, it won't do anything, so we release a whole - // page now and let extra_bytes_released_ smooth it out over time. - Length num_pages = max(num_bytes >> kPageShift, 1); - size_t bytes_released = Static::pageheap()->ReleaseAtLeastNPages( - num_pages) << kPageShift; - if (bytes_released > num_bytes) { - extra_bytes_released_ = bytes_released - num_bytes; - } else { - // The PageHeap wasn't able to release num_bytes. Don't try to - // compensate with a big release next time. Specifically, - // ReleaseFreeMemory() calls ReleaseToSystem(LONG_MAX). - extra_bytes_released_ = 0; - } - } - - virtual void SetMemoryReleaseRate(double rate) { - FLAGS_tcmalloc_release_rate = rate; - } - - virtual double GetMemoryReleaseRate() { - return FLAGS_tcmalloc_release_rate; - } - virtual size_t GetEstimatedAllocatedSize(size_t size) { - if (size <= kMaxSize) { - const size_t cl = Static::sizemap()->SizeClass(size); - const size_t alloc_size = Static::sizemap()->ByteSizeForClass(cl); - return alloc_size; - } else { - return tcmalloc::pages(size) << kPageShift; - } - } - - // This just calls GetSizeWithCallback, but because that's in an - // unnamed namespace, we need to move the definition below it in the - // file. - virtual size_t GetAllocatedSize(const void* ptr); - - // This duplicates some of the logic in GetSizeWithCallback, but is - // faster. This is important on OS X, where this function is called - // on every allocation operation. - virtual Ownership GetOwnership(const void* ptr) { - const PageID p = reinterpret_cast(ptr) >> kPageShift; - // The rest of tcmalloc assumes that all allocated pointers use at - // most kAddressBits bits. If ptr doesn't, then it definitely - // wasn't alloacted by tcmalloc. - if ((p >> (kAddressBits - kPageShift)) > 0) { - return kNotOwned; - } - size_t cl = Static::pageheap()->GetSizeClassIfCached(p); - if (cl != 0) { - return kOwned; - } - const Span *span = Static::pageheap()->GetDescriptor(p); - return span ? kOwned : kNotOwned; - } - - virtual void GetFreeListSizes(vector* v) { - static const char* kCentralCacheType = "tcmalloc.central"; - static const char* kTransferCacheType = "tcmalloc.transfer"; - static const char* kThreadCacheType = "tcmalloc.thread"; - static const char* kPageHeapType = "tcmalloc.page"; - static const char* kPageHeapUnmappedType = "tcmalloc.page_unmapped"; - static const char* kLargeSpanType = "tcmalloc.large"; - static const char* kLargeUnmappedSpanType = "tcmalloc.large_unmapped"; - - v->clear(); - - // central class information - int64 prev_class_size = 0; - for (int cl = 1; cl < kNumClasses; ++cl) { - size_t class_size = Static::sizemap()->ByteSizeForClass(cl); - MallocExtension::FreeListInfo i; - i.min_object_size = prev_class_size + 1; - i.max_object_size = class_size; - i.total_bytes_free = - Static::central_cache()[cl].length() * class_size; - i.type = kCentralCacheType; - v->push_back(i); - - // transfer cache - i.total_bytes_free = - Static::central_cache()[cl].tc_length() * class_size; - i.type = kTransferCacheType; - v->push_back(i); - - prev_class_size = Static::sizemap()->ByteSizeForClass(cl); - } - - // Add stats from per-thread heaps - uint64_t class_count[kNumClasses]; - memset(class_count, 0, sizeof(class_count)); - { - SpinLockHolder h(Static::pageheap_lock()); - uint64_t thread_bytes = 0; - ThreadCache::GetThreadStats(&thread_bytes, class_count); - } - - prev_class_size = 0; - for (int cl = 1; cl < kNumClasses; ++cl) { - MallocExtension::FreeListInfo i; - i.min_object_size = prev_class_size + 1; - i.max_object_size = Static::sizemap()->ByteSizeForClass(cl); - i.total_bytes_free = - class_count[cl] * Static::sizemap()->ByteSizeForClass(cl); - i.type = kThreadCacheType; - v->push_back(i); - } - - // append page heap info - PageHeap::SmallSpanStats small; - PageHeap::LargeSpanStats large; - { - SpinLockHolder h(Static::pageheap_lock()); - Static::pageheap()->GetSmallSpanStats(&small); - Static::pageheap()->GetLargeSpanStats(&large); - } - - // large spans: mapped - MallocExtension::FreeListInfo span_info; - span_info.type = kLargeSpanType; - span_info.max_object_size = (numeric_limits::max)(); - span_info.min_object_size = kMaxPages << kPageShift; - span_info.total_bytes_free = large.normal_pages << kPageShift; - v->push_back(span_info); - - // large spans: unmapped - span_info.type = kLargeUnmappedSpanType; - span_info.total_bytes_free = large.returned_pages << kPageShift; - v->push_back(span_info); - - // small spans - for (int s = 1; s < kMaxPages; s++) { - MallocExtension::FreeListInfo i; - i.max_object_size = (s << kPageShift); - i.min_object_size = ((s - 1) << kPageShift); - - i.type = kPageHeapType; - i.total_bytes_free = (s << kPageShift) * small.normal_length[s]; - v->push_back(i); - - i.type = kPageHeapUnmappedType; - i.total_bytes_free = (s << kPageShift) * small.returned_length[s]; - v->push_back(i); - } - } -}; - -// The constructor allocates an object to ensure that initialization -// runs before main(), and therefore we do not have a chance to become -// multi-threaded before initialization. We also create the TSD key -// here. Presumably by the time this constructor runs, glibc is in -// good enough shape to handle pthread_key_create(). -// -// The constructor also takes the opportunity to tell STL to use -// tcmalloc. We want to do this early, before construct time, so -// all user STL allocations go through tcmalloc (which works really -// well for STL). -// -// The destructor prints stats when the program exits. -static int tcmallocguard_refcount = 0; // no lock needed: runs before main() -TCMallocGuard::TCMallocGuard() { - if (tcmallocguard_refcount++ == 0) { - ReplaceSystemAlloc(); // defined in libc_override_*.h - tc_free(tc_malloc(1)); - ThreadCache::InitTSD(); - tc_free(tc_malloc(1)); - // Either we, or debugallocation.cc, or valgrind will control memory - // management. We register our extension if we're the winner. -#ifdef TCMALLOC_USING_DEBUGALLOCATION - // Let debugallocation register its extension. -#else - if (RunningOnValgrind()) { - // Let Valgrind uses its own malloc (so don't register our extension). - } else { - MallocExtension::Register(new TCMallocImplementation); - } -#endif - } -} - -TCMallocGuard::~TCMallocGuard() { - if (--tcmallocguard_refcount == 0) { - const char* env = NULL; - if (!RunningOnValgrind()) { - // Valgrind uses it's own malloc so we cannot do MALLOCSTATS - env = getenv("MALLOCSTATS"); - } - if (env != NULL) { - int level = atoi(env); - if (level < 1) level = 1; - PrintStats(level); - } - } -} -#ifndef WIN32_OVERRIDE_ALLOCATORS -static TCMallocGuard module_enter_exit_hook; -#endif - -//------------------------------------------------------------------- -// Helpers for the exported routines below -//------------------------------------------------------------------- - -static inline bool CheckCachedSizeClass(void *ptr) { - PageID p = reinterpret_cast(ptr) >> kPageShift; - size_t cached_value = Static::pageheap()->GetSizeClassIfCached(p); - return cached_value == 0 || - cached_value == Static::pageheap()->GetDescriptor(p)->sizeclass; -} - -static inline void* CheckedMallocResult(void *result) { - ASSERT(result == NULL || CheckCachedSizeClass(result)); - return result; -} - -static inline void* SpanToMallocResult(Span *span) { - Static::pageheap()->CacheSizeClass(span->start, 0); - return - CheckedMallocResult(reinterpret_cast(span->start << kPageShift)); -} - -static void* DoSampledAllocation(size_t size) { -#ifndef NO_TCMALLOC_SAMPLES - // Grab the stack trace outside the heap lock - StackTrace tmp; - tmp.depth = GetStackTrace(tmp.stack, tcmalloc::kMaxStackDepth, 1); - tmp.size = size; - - SpinLockHolder h(Static::pageheap_lock()); - // Allocate span - Span *span = Static::pageheap()->New(tcmalloc::pages(size == 0 ? 1 : size)); - if (UNLIKELY(span == NULL)) { - return NULL; - } - - // Allocate stack trace - StackTrace *stack = Static::stacktrace_allocator()->New(); - if (UNLIKELY(stack == NULL)) { - // Sampling failed because of lack of memory - return span; - } - *stack = tmp; - span->sample = 1; - span->objects = stack; - tcmalloc::DLL_Prepend(Static::sampled_objects(), span); - - return SpanToMallocResult(span); -#else - abort(); -#endif -} - -namespace { - -typedef void* (*malloc_fn)(void *arg); - -SpinLock set_new_handler_lock(SpinLock::LINKER_INITIALIZED); - -void* handle_oom(malloc_fn retry_fn, - void* retry_arg, - bool from_operator, - bool nothrow) { - if (!from_operator && !tc_new_mode) { - // we're out of memory in C library function (malloc etc) and no - // "new mode" forced on us. Just return NULL - return NULL; - } - // we're OOM in operator new or "new mode" is set. We might have to - // call new_handle and maybe retry allocation. - - for (;;) { - // Get the current new handler. NB: this function is not - // thread-safe. We make a feeble stab at making it so here, but - // this lock only protects against tcmalloc interfering with - // itself, not with other libraries calling set_new_handler. - std::new_handler nh; - { - SpinLockHolder h(&set_new_handler_lock); - nh = std::set_new_handler(0); - (void) std::set_new_handler(nh); - } -#if (defined(__GNUC__) && !defined(__EXCEPTIONS)) || (defined(_HAS_EXCEPTIONS) && !_HAS_EXCEPTIONS) - if (!nh) { - return NULL; - } - // Since exceptions are disabled, we don't really know if new_handler - // failed. Assume it will abort if it fails. - (*nh)(); -#else - // If no new_handler is established, the allocation failed. - if (!nh) { - if (nothrow) { - return NULL; - } - throw std::bad_alloc(); - } - // Otherwise, try the new_handler. If it returns, retry the - // allocation. If it throws std::bad_alloc, fail the allocation. - // if it throws something else, don't interfere. - try { - (*nh)(); - } catch (const std::bad_alloc&) { - if (!nothrow) throw; - return NULL; - } -#endif // (defined(__GNUC__) && !defined(__EXCEPTIONS)) || (defined(_HAS_EXCEPTIONS) && !_HAS_EXCEPTIONS) - - // we get here if new_handler returns successfully. So we retry - // allocation. - void* rv = retry_fn(retry_arg); - if (rv != NULL) { - return rv; - } - - // if allocation failed again we go to next loop iteration - } -} - -// Copy of FLAGS_tcmalloc_large_alloc_report_threshold with -// automatic increases factored in. -static int64_t large_alloc_threshold = - (kPageSize > FLAGS_tcmalloc_large_alloc_report_threshold - ? kPageSize : FLAGS_tcmalloc_large_alloc_report_threshold); - -static void ReportLargeAlloc(Length num_pages, void* result) { - StackTrace stack; - stack.depth = GetStackTrace(stack.stack, tcmalloc::kMaxStackDepth, 1); - - static const int N = 1000; - char buffer[N]; - TCMalloc_Printer printer(buffer, N); - printer.printf("tcmalloc: large alloc %" PRIu64 " bytes == %p @ ", - static_cast(num_pages) << kPageShift, - result); - for (int i = 0; i < stack.depth; i++) { - printer.printf(" %p", stack.stack[i]); - } - printer.printf("\n"); - write(STDERR_FILENO, buffer, strlen(buffer)); -} - -void* do_memalign(size_t align, size_t size); - -struct retry_memaligh_data { - size_t align; - size_t size; -}; - -static void *retry_do_memalign(void *arg) { - retry_memaligh_data *data = static_cast(arg); - return do_memalign(data->align, data->size); -} - -static void *maybe_do_cpp_memalign_slow(size_t align, size_t size) { - retry_memaligh_data data; - data.align = align; - data.size = size; - return handle_oom(retry_do_memalign, &data, - false, true); -} - -inline void* do_memalign_or_cpp_memalign(size_t align, size_t size) { - void *rv = do_memalign(align, size); - if (LIKELY(rv != NULL)) { - return rv; - } - return maybe_do_cpp_memalign_slow(align, size); -} - -// Must be called with the page lock held. -inline bool should_report_large(Length num_pages) { - const int64 threshold = large_alloc_threshold; - if (threshold > 0 && num_pages >= (threshold >> kPageShift)) { - // Increase the threshold by 1/8 every time we generate a report. - // We cap the threshold at 8GiB to avoid overflow problems. - large_alloc_threshold = (threshold + threshold/8 < 8ll<<30 - ? threshold + threshold/8 : 8ll<<30); - return true; - } - return false; -} - -// Helper for do_malloc(). -inline void* do_malloc_pages(ThreadCache* heap, size_t size) { - void* result; - bool report_large; - - Length num_pages = tcmalloc::pages(size); - - // NOTE: we're passing original size here as opposed to rounded-up - // size as we do in do_malloc_small. The difference is small here - // (at most 4k out of at least 256k). And not rounding up saves us - // from possibility of overflow, which rounding up could produce. - // - // See https://github.com/gperftools/gperftools/issues/723 - if (heap->SampleAllocation(size)) { - result = DoSampledAllocation(size); - - SpinLockHolder h(Static::pageheap_lock()); - report_large = should_report_large(num_pages); - } else { - SpinLockHolder h(Static::pageheap_lock()); - Span* span = Static::pageheap()->New(num_pages); - result = (UNLIKELY(span == NULL) ? NULL : SpanToMallocResult(span)); - report_large = should_report_large(num_pages); - } - - if (report_large) { - ReportLargeAlloc(num_pages, result); - } - return result; -} - -ALWAYS_INLINE void* do_malloc_small(ThreadCache* heap, size_t size) { - ASSERT(Static::IsInited()); - ASSERT(heap != NULL); - size_t cl = Static::sizemap()->SizeClass(size); - size = Static::sizemap()->class_to_size(cl); - - if (UNLIKELY(heap->SampleAllocation(size))) { - return DoSampledAllocation(size); - } else { - // The common case, and also the simplest. This just pops the - // size-appropriate freelist, after replenishing it if it's empty. - return CheckedMallocResult(heap->Allocate(size, cl)); - } -} - -ALWAYS_INLINE void* do_malloc(size_t size) { - if (ThreadCache::have_tls) { - if (LIKELY(size < ThreadCache::MinSizeForSlowPath())) { - return do_malloc_small(ThreadCache::GetCacheWhichMustBePresent(), size); - } - if (UNLIKELY(ThreadCache::IsUseEmergencyMalloc())) { - return tcmalloc::EmergencyMalloc(size); - } - } - - if (size <= kMaxSize) { - return do_malloc_small(ThreadCache::GetCache(), size); - } else { - return do_malloc_pages(ThreadCache::GetCache(), size); - } -} - -static void *retry_malloc(void* size) { - return do_malloc(reinterpret_cast(size)); -} - -ALWAYS_INLINE void* do_malloc_or_cpp_alloc(size_t size) { - void *rv = do_malloc(size); - if (LIKELY(rv != NULL)) { - return rv; - } - return handle_oom(retry_malloc, reinterpret_cast(size), - false, true); -} - -ALWAYS_INLINE void* do_calloc(size_t n, size_t elem_size) { - // Overflow check - const size_t size = n * elem_size; - if (elem_size != 0 && size / elem_size != n) return NULL; - - void* result = do_malloc_or_cpp_alloc(size); - if (result != NULL) { - memset(result, 0, size); - } - return result; -} - -// If ptr is NULL, do nothing. Otherwise invoke the given function. -inline void free_null_or_invalid(void* ptr, void (*invalid_free_fn)(void*)) { - if (ptr != NULL) { - (*invalid_free_fn)(ptr); - } -} - -// Helper for do_free_with_callback(), below. Inputs: -// ptr is object to be freed -// invalid_free_fn is a function that gets invoked on certain "bad frees" -// heap is the ThreadCache for this thread, or NULL if it isn't known -// heap_must_be_valid is whether heap is known to be non-NULL -// -// This function may only be used after Static::IsInited() is true. -// -// We can usually detect the case where ptr is not pointing to a page that -// tcmalloc is using, and in those cases we invoke invalid_free_fn. -// -// To maximize speed in the common case, we usually get here with -// heap_must_be_valid being a manifest constant equal to true. -ALWAYS_INLINE void do_free_helper(void* ptr, - void (*invalid_free_fn)(void*), - ThreadCache* heap, - bool heap_must_be_valid, - bool use_hint, - size_t size_hint) { - ASSERT((Static::IsInited() && heap != NULL) || !heap_must_be_valid); - if (!heap_must_be_valid && !Static::IsInited()) { - // We called free() before malloc(). This can occur if the - // (system) malloc() is called before tcmalloc is loaded, and then - // free() is called after tcmalloc is loaded (and tc_free has - // replaced free), but before the global constructor has run that - // sets up the tcmalloc data structures. - free_null_or_invalid(ptr, invalid_free_fn); - return; - } - Span* span = NULL; - const PageID p = reinterpret_cast(ptr) >> kPageShift; - size_t cl; - if (use_hint && Static::sizemap()->MaybeSizeClass(size_hint, &cl)) { - goto non_zero; - } - - cl = Static::pageheap()->GetSizeClassIfCached(p); - if (UNLIKELY(cl == 0)) { - span = Static::pageheap()->GetDescriptor(p); - if (UNLIKELY(!span)) { - // span can be NULL because the pointer passed in is NULL or invalid - // (not something returned by malloc or friends), or because the - // pointer was allocated with some other allocator besides - // tcmalloc. The latter can happen if tcmalloc is linked in via - // a dynamic library, but is not listed last on the link line. - // In that case, libraries after it on the link line will - // allocate with libc malloc, but free with tcmalloc's free. - free_null_or_invalid(ptr, invalid_free_fn); - return; - } - cl = span->sizeclass; - Static::pageheap()->CacheSizeClass(p, cl); - } - - ASSERT(ptr != NULL); - if (LIKELY(cl != 0)) { - non_zero: - ASSERT(!Static::pageheap()->GetDescriptor(p)->sample); - if (heap_must_be_valid || heap != NULL) { - heap->Deallocate(ptr, cl); - } else { - // Delete directly into central cache - tcmalloc::SLL_SetNext(ptr, NULL); - Static::central_cache()[cl].InsertRange(ptr, ptr, 1); - } - } else { - SpinLockHolder h(Static::pageheap_lock()); - ASSERT(reinterpret_cast(ptr) % kPageSize == 0); - ASSERT(span != NULL && span->start == p); - if (span->sample) { - StackTrace* st = reinterpret_cast(span->objects); - tcmalloc::DLL_Remove(span); - Static::stacktrace_allocator()->Delete(st); - span->objects = NULL; - } - Static::pageheap()->Delete(span); - } -} - -// Helper for the object deletion (free, delete, etc.). Inputs: -// ptr is object to be freed -// invalid_free_fn is a function that gets invoked on certain "bad frees" -// -// We can usually detect the case where ptr is not pointing to a page that -// tcmalloc is using, and in those cases we invoke invalid_free_fn. -ALWAYS_INLINE void do_free_with_callback(void* ptr, - void (*invalid_free_fn)(void*), - bool use_hint, size_t size_hint) { - ThreadCache* heap = NULL; - heap = ThreadCache::GetCacheIfPresent(); - if (LIKELY(heap)) { - do_free_helper(ptr, invalid_free_fn, heap, true, use_hint, size_hint); - } else { - do_free_helper(ptr, invalid_free_fn, heap, false, use_hint, size_hint); - } -} - -// The default "do_free" that uses the default callback. -ALWAYS_INLINE void do_free(void* ptr) { - return do_free_with_callback(ptr, &InvalidFree, false, 0); -} - -// NOTE: some logic here is duplicated in GetOwnership (above), for -// speed. If you change this function, look at that one too. -inline size_t GetSizeWithCallback(const void* ptr, - size_t (*invalid_getsize_fn)(const void*)) { - if (ptr == NULL) - return 0; - const PageID p = reinterpret_cast(ptr) >> kPageShift; - size_t cl = Static::pageheap()->GetSizeClassIfCached(p); - if (cl != 0) { - return Static::sizemap()->ByteSizeForClass(cl); - } else { - const Span *span = Static::pageheap()->GetDescriptor(p); - if (UNLIKELY(span == NULL)) { // means we do not own this memory - return (*invalid_getsize_fn)(ptr); - } else if (span->sizeclass != 0) { - Static::pageheap()->CacheSizeClass(p, span->sizeclass); - return Static::sizemap()->ByteSizeForClass(span->sizeclass); - } else { - return span->length << kPageShift; - } - } -} - -// This lets you call back to a given function pointer if ptr is invalid. -// It is used primarily by windows code which wants a specialized callback. -ALWAYS_INLINE void* do_realloc_with_callback( - void* old_ptr, size_t new_size, - void (*invalid_free_fn)(void*), - size_t (*invalid_get_size_fn)(const void*)) { - // Get the size of the old entry - const size_t old_size = GetSizeWithCallback(old_ptr, invalid_get_size_fn); - - // Reallocate if the new size is larger than the old size, - // or if the new size is significantly smaller than the old size. - // We do hysteresis to avoid resizing ping-pongs: - // . If we need to grow, grow to max(new_size, old_size * 1.X) - // . Don't shrink unless new_size < old_size * 0.Y - // X and Y trade-off time for wasted space. For now we do 1.25 and 0.5. - const size_t lower_bound_to_grow = old_size + old_size / 4ul; - const size_t upper_bound_to_shrink = old_size / 2ul; - if ((new_size > old_size) || (new_size < upper_bound_to_shrink)) { - // Need to reallocate. - void* new_ptr = NULL; - - if (new_size > old_size && new_size < lower_bound_to_grow) { - new_ptr = do_malloc_or_cpp_alloc(lower_bound_to_grow); - } - if (new_ptr == NULL) { - // Either new_size is not a tiny increment, or last do_malloc failed. - new_ptr = do_malloc_or_cpp_alloc(new_size); - } - if (UNLIKELY(new_ptr == NULL)) { - return NULL; - } - MallocHook::InvokeNewHook(new_ptr, new_size); - memcpy(new_ptr, old_ptr, ((old_size < new_size) ? old_size : new_size)); - MallocHook::InvokeDeleteHook(old_ptr); - // We could use a variant of do_free() that leverages the fact - // that we already know the sizeclass of old_ptr. The benefit - // would be small, so don't bother. - do_free_with_callback(old_ptr, invalid_free_fn, false, 0); - return new_ptr; - } else { - // We still need to call hooks to report the updated size: - MallocHook::InvokeDeleteHook(old_ptr); - MallocHook::InvokeNewHook(old_ptr, new_size); - return old_ptr; - } -} - -ALWAYS_INLINE void* do_realloc(void* old_ptr, size_t new_size) { - return do_realloc_with_callback(old_ptr, new_size, - &InvalidFree, &InvalidGetSizeForRealloc); -} - -// For use by exported routines below that want specific alignments -// -// Note: this code can be slow for alignments > 16, and can -// significantly fragment memory. The expectation is that -// memalign/posix_memalign/valloc/pvalloc will not be invoked very -// often. This requirement simplifies our implementation and allows -// us to tune for expected allocation patterns. -void* do_memalign(size_t align, size_t size) { - ASSERT((align & (align - 1)) == 0); - ASSERT(align > 0); - if (size + align < size) return NULL; // Overflow - - // Fall back to malloc if we would already align this memory access properly. - if (align <= AlignmentForSize(size)) { - void* p = do_malloc(size); - ASSERT((reinterpret_cast(p) % align) == 0); - return p; - } - - if (UNLIKELY(Static::pageheap() == NULL)) ThreadCache::InitModule(); - - // Allocate at least one byte to avoid boundary conditions below - if (size == 0) size = 1; - - if (size <= kMaxSize && align < kPageSize) { - // Search through acceptable size classes looking for one with - // enough alignment. This depends on the fact that - // InitSizeClasses() currently produces several size classes that - // are aligned at powers of two. We will waste time and space if - // we miss in the size class array, but that is deemed acceptable - // since memalign() should be used rarely. - int cl = Static::sizemap()->SizeClass(size); - while (cl < kNumClasses && - ((Static::sizemap()->class_to_size(cl) & (align - 1)) != 0)) { - cl++; - } - if (cl < kNumClasses) { - ThreadCache* heap = ThreadCache::GetCache(); - size = Static::sizemap()->class_to_size(cl); - return CheckedMallocResult(heap->Allocate(size, cl)); - } - } - - // We will allocate directly from the page heap - SpinLockHolder h(Static::pageheap_lock()); - - if (align <= kPageSize) { - // Any page-level allocation will be fine - // TODO: We could put the rest of this page in the appropriate - // TODO: cache but it does not seem worth it. - Span* span = Static::pageheap()->New(tcmalloc::pages(size)); - return UNLIKELY(span == NULL) ? NULL : SpanToMallocResult(span); - } - - // Allocate extra pages and carve off an aligned portion - const Length alloc = tcmalloc::pages(size + align); - Span* span = Static::pageheap()->New(alloc); - if (UNLIKELY(span == NULL)) return NULL; - - // Skip starting portion so that we end up aligned - Length skip = 0; - while ((((span->start+skip) << kPageShift) & (align - 1)) != 0) { - skip++; - } - ASSERT(skip < alloc); - if (skip > 0) { - Span* rest = Static::pageheap()->Split(span, skip); - Static::pageheap()->Delete(span); - span = rest; - } - - // Skip trailing portion that we do not need to return - const Length needed = tcmalloc::pages(size); - ASSERT(span->length >= needed); - if (span->length > needed) { - Span* trailer = Static::pageheap()->Split(span, needed); - Static::pageheap()->Delete(trailer); - } - return SpanToMallocResult(span); -} - -// Helpers for use by exported routines below: - -inline void do_malloc_stats() { - PrintStats(1); -} - -inline int do_mallopt(int cmd, int value) { - return 1; // Indicates error -} - -#ifdef HAVE_STRUCT_MALLINFO -inline struct mallinfo do_mallinfo() { - TCMallocStats stats; - ExtractStats(&stats, NULL, NULL, NULL); - - // Just some of the fields are filled in. - struct mallinfo info; - memset(&info, 0, sizeof(info)); - - // Unfortunately, the struct contains "int" field, so some of the - // size values will be truncated. - info.arena = static_cast(stats.pageheap.system_bytes); - info.fsmblks = static_cast(stats.thread_bytes - + stats.central_bytes - + stats.transfer_bytes); - info.fordblks = static_cast(stats.pageheap.free_bytes + - stats.pageheap.unmapped_bytes); - info.uordblks = static_cast(stats.pageheap.system_bytes - - stats.thread_bytes - - stats.central_bytes - - stats.transfer_bytes - - stats.pageheap.free_bytes - - stats.pageheap.unmapped_bytes); - - return info; -} -#endif // HAVE_STRUCT_MALLINFO - -inline void* cpp_alloc(size_t size, bool nothrow) { - void* p = do_malloc(size); - if (LIKELY(p)) { - return p; - } - return handle_oom(retry_malloc, reinterpret_cast(size), - true, nothrow); -} - -} // end unnamed namespace - -// As promised, the definition of this function, declared above. -size_t TCMallocImplementation::GetAllocatedSize(const void* ptr) { - if (ptr == NULL) - return 0; - ASSERT(TCMallocImplementation::GetOwnership(ptr) - != TCMallocImplementation::kNotOwned); - return GetSizeWithCallback(ptr, &InvalidGetAllocatedSize); -} - -void TCMallocImplementation::MarkThreadBusy() { - // Allocate to force the creation of a thread cache, but avoid - // invoking any hooks. - do_free(do_malloc(0)); -} - -//------------------------------------------------------------------- -// Exported routines -//------------------------------------------------------------------- - -extern "C" PERFTOOLS_DLL_DECL const char* tc_version( - int* major, int* minor, const char** patch) PERFTOOLS_THROW { - if (major) *major = TC_VERSION_MAJOR; - if (minor) *minor = TC_VERSION_MINOR; - if (patch) *patch = TC_VERSION_PATCH; - return TC_VERSION_STRING; -} - -// This function behaves similarly to MSVC's _set_new_mode. -// If flag is 0 (default), calls to malloc will behave normally. -// If flag is 1, calls to malloc will behave like calls to new, -// and the std_new_handler will be invoked on failure. -// Returns the previous mode. -extern "C" PERFTOOLS_DLL_DECL int tc_set_new_mode(int flag) PERFTOOLS_THROW { - int old_mode = tc_new_mode; - tc_new_mode = flag; - return old_mode; -} - -#ifndef TCMALLOC_USING_DEBUGALLOCATION // debugallocation.cc defines its own - -#if defined(__GNUC__) && defined(__ELF__) && !defined(TCMALLOC_NO_ALIASES) -#define TC_ALIAS(name) __attribute__((alias(#name))) -#endif - -// CAVEAT: The code structure below ensures that MallocHook methods are always -// called from the stack frame of the invoked allocation function. -// heap-checker.cc depends on this to start a stack trace from -// the call to the (de)allocation function. - -extern "C" PERFTOOLS_DLL_DECL void* tc_malloc(size_t size) PERFTOOLS_THROW { - void* result = do_malloc_or_cpp_alloc(size); - MallocHook::InvokeNewHook(result, size); - return result; -} - -extern "C" PERFTOOLS_DLL_DECL void tc_free(void* ptr) PERFTOOLS_THROW { - MallocHook::InvokeDeleteHook(ptr); - do_free(ptr); -} - -extern "C" PERFTOOLS_DLL_DECL void tc_free_sized(void *ptr, size_t size) PERFTOOLS_THROW { - if ((reinterpret_cast(ptr) & (kPageSize-1)) == 0) { - tc_free(ptr); - return; - } - MallocHook::InvokeDeleteHook(ptr); - do_free_with_callback(ptr, &InvalidFree, true, size); -} - -#ifdef TC_ALIAS - -extern "C" PERFTOOLS_DLL_DECL void tc_delete_sized(void *p, size_t size) throw() - TC_ALIAS(tc_free_sized); -extern "C" PERFTOOLS_DLL_DECL void tc_deletearray_sized(void *p, size_t size) throw() - TC_ALIAS(tc_free_sized); - -#else - -extern "C" PERFTOOLS_DLL_DECL void tc_delete_sized(void *p, size_t size) throw() { - tc_free_sized(p, size); -} -extern "C" PERFTOOLS_DLL_DECL void tc_deletearray_sized(void *p, size_t size) throw() { - tc_free_sized(p, size); -} - -#endif - -extern "C" PERFTOOLS_DLL_DECL void* tc_calloc(size_t n, - size_t elem_size) PERFTOOLS_THROW { - if (ThreadCache::IsUseEmergencyMalloc()) { - return tcmalloc::EmergencyCalloc(n, elem_size); - } - void* result = do_calloc(n, elem_size); - MallocHook::InvokeNewHook(result, n * elem_size); - return result; -} - -extern "C" PERFTOOLS_DLL_DECL void tc_cfree(void* ptr) PERFTOOLS_THROW -#ifdef TC_ALIAS -TC_ALIAS(tc_free); -#else -{ - MallocHook::InvokeDeleteHook(ptr); - do_free(ptr); -} -#endif - -extern "C" PERFTOOLS_DLL_DECL void* tc_realloc(void* old_ptr, - size_t new_size) PERFTOOLS_THROW { - if (old_ptr == NULL) { - void* result = do_malloc_or_cpp_alloc(new_size); - MallocHook::InvokeNewHook(result, new_size); - return result; - } - if (new_size == 0) { - MallocHook::InvokeDeleteHook(old_ptr); - do_free(old_ptr); - return NULL; - } - if (UNLIKELY(tcmalloc::IsEmergencyPtr(old_ptr))) { - return tcmalloc::EmergencyRealloc(old_ptr, new_size); - } - return do_realloc(old_ptr, new_size); -} - -extern "C" PERFTOOLS_DLL_DECL void* tc_new(size_t size) { - void* p = cpp_alloc(size, false); - // We keep this next instruction out of cpp_alloc for a reason: when - // it's in, and new just calls cpp_alloc, the optimizer may fold the - // new call into cpp_alloc, which messes up our whole section-based - // stacktracing (see ATTRIBUTE_SECTION, above). This ensures cpp_alloc - // isn't the last thing this fn calls, and prevents the folding. - MallocHook::InvokeNewHook(p, size); - return p; -} - -extern "C" PERFTOOLS_DLL_DECL void* tc_new_nothrow(size_t size, const std::nothrow_t&) PERFTOOLS_THROW { - void* p = cpp_alloc(size, true); - MallocHook::InvokeNewHook(p, size); - return p; -} - -extern "C" PERFTOOLS_DLL_DECL void tc_delete(void* p) PERFTOOLS_THROW -#ifdef TC_ALIAS -TC_ALIAS(tc_free); -#else -{ - MallocHook::InvokeDeleteHook(p); - do_free(p); -} -#endif - -// Standard C++ library implementations define and use this -// (via ::operator delete(ptr, nothrow)). -// But it's really the same as normal delete, so we just do the same thing. -extern "C" PERFTOOLS_DLL_DECL void tc_delete_nothrow(void* p, const std::nothrow_t&) PERFTOOLS_THROW -#ifdef TC_ALIAS -TC_ALIAS(tc_free); -#else -{ - MallocHook::InvokeDeleteHook(p); - do_free(p); -} -#endif - -extern "C" PERFTOOLS_DLL_DECL void* tc_newarray(size_t size) -#ifdef TC_ALIAS -TC_ALIAS(tc_new); -#else -{ - void* p = cpp_alloc(size, false); - // We keep this next instruction out of cpp_alloc for a reason: when - // it's in, and new just calls cpp_alloc, the optimizer may fold the - // new call into cpp_alloc, which messes up our whole section-based - // stacktracing (see ATTRIBUTE_SECTION, above). This ensures cpp_alloc - // isn't the last thing this fn calls, and prevents the folding. - MallocHook::InvokeNewHook(p, size); - return p; -} -#endif - -extern "C" PERFTOOLS_DLL_DECL void* tc_newarray_nothrow(size_t size, const std::nothrow_t&) - PERFTOOLS_THROW -#ifdef TC_ALIAS -TC_ALIAS(tc_new_nothrow); -#else -{ - void* p = cpp_alloc(size, true); - MallocHook::InvokeNewHook(p, size); - return p; -} -#endif - -extern "C" PERFTOOLS_DLL_DECL void tc_deletearray(void* p) PERFTOOLS_THROW -#ifdef TC_ALIAS -TC_ALIAS(tc_free); -#else -{ - MallocHook::InvokeDeleteHook(p); - do_free(p); -} -#endif - -extern "C" PERFTOOLS_DLL_DECL void tc_deletearray_nothrow(void* p, const std::nothrow_t&) PERFTOOLS_THROW -#ifdef TC_ALIAS -TC_ALIAS(tc_free); -#else -{ - MallocHook::InvokeDeleteHook(p); - do_free(p); -} -#endif - -extern "C" PERFTOOLS_DLL_DECL void* tc_memalign(size_t align, - size_t size) PERFTOOLS_THROW { - void* result = do_memalign_or_cpp_memalign(align, size); - MallocHook::InvokeNewHook(result, size); - return result; -} - -extern "C" PERFTOOLS_DLL_DECL int tc_posix_memalign( - void** result_ptr, size_t align, size_t size) PERFTOOLS_THROW { - if (((align % sizeof(void*)) != 0) || - ((align & (align - 1)) != 0) || - (align == 0)) { - return EINVAL; - } - - void* result = do_memalign_or_cpp_memalign(align, size); - MallocHook::InvokeNewHook(result, size); - if (UNLIKELY(result == NULL)) { - return ENOMEM; - } else { - *result_ptr = result; - return 0; - } -} - -static size_t pagesize = 0; - -extern "C" PERFTOOLS_DLL_DECL void* tc_valloc(size_t size) PERFTOOLS_THROW { - // Allocate page-aligned object of length >= size bytes - if (pagesize == 0) pagesize = getpagesize(); - void* result = do_memalign_or_cpp_memalign(pagesize, size); - MallocHook::InvokeNewHook(result, size); - return result; -} - -extern "C" PERFTOOLS_DLL_DECL void* tc_pvalloc(size_t size) PERFTOOLS_THROW { - // Round up size to a multiple of pagesize - if (pagesize == 0) pagesize = getpagesize(); - if (size == 0) { // pvalloc(0) should allocate one page, according to - size = pagesize; // http://man.free4web.biz/man3/libmpatrol.3.html - } - size = (size + pagesize - 1) & ~(pagesize - 1); - void* result = do_memalign_or_cpp_memalign(pagesize, size); - MallocHook::InvokeNewHook(result, size); - return result; -} - -extern "C" PERFTOOLS_DLL_DECL void tc_malloc_stats(void) PERFTOOLS_THROW { - do_malloc_stats(); -} - -extern "C" PERFTOOLS_DLL_DECL int tc_mallopt(int cmd, int value) PERFTOOLS_THROW { - return do_mallopt(cmd, value); -} - -#ifdef HAVE_STRUCT_MALLINFO -extern "C" PERFTOOLS_DLL_DECL struct mallinfo tc_mallinfo(void) PERFTOOLS_THROW { - return do_mallinfo(); -} -#endif - -extern "C" PERFTOOLS_DLL_DECL size_t tc_malloc_size(void* ptr) PERFTOOLS_THROW { - return MallocExtension::instance()->GetAllocatedSize(ptr); -} - -extern "C" PERFTOOLS_DLL_DECL void* tc_malloc_skip_new_handler(size_t size) PERFTOOLS_THROW { - void* result = do_malloc(size); - MallocHook::InvokeNewHook(result, size); - return result; -} - -#pragma GCC diagnostic pop - -#endif // TCMALLOC_USING_DEBUGALLOCATION diff --git a/contrib/libtcmalloc/src/tcmalloc.h b/contrib/libtcmalloc/src/tcmalloc.h deleted file mode 100644 index 70d567268c2..00000000000 --- a/contrib/libtcmalloc/src/tcmalloc.h +++ /dev/null @@ -1,70 +0,0 @@ -// -*- Mode: C++; c-basic-offset: 2; indent-tabs-mode: nil -*- -// Copyright (c) 2007, Google Inc. -// All rights reserved. -// -// Redistribution and use in source and binary forms, with or without -// modification, are permitted provided that the following conditions are -// met: -// -// * Redistributions of source code must retain the above copyright -// notice, this list of conditions and the following disclaimer. -// * Redistributions in binary form must reproduce the above -// copyright notice, this list of conditions and the following disclaimer -// in the documentation and/or other materials provided with the -// distribution. -// * Neither the name of Google Inc. nor the names of its -// contributors may be used to endorse or promote products derived from -// this software without specific prior written permission. -// -// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS -// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT -// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR -// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT -// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, -// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT -// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, -// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY -// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT -// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE -// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. - -// --- -// Author: Craig Silverstein -// -// Some obscure memory-allocation routines may not be declared on all -// systems. In those cases, we'll just declare them ourselves. -// This file is meant to be used only internally, for unittests. - -#include "config.h" - -#ifndef _XOPEN_SOURCE -# define _XOPEN_SOURCE 600 // for posix_memalign -#endif -#include // for posix_memalign -// FreeBSD has malloc.h, but complains if you use it -#if defined(HAVE_MALLOC_H) && !defined(__FreeBSD__) -#include // for memalign, valloc, pvalloc -#endif - -// __THROW is defined in glibc systems. It means, counter-intuitively, -// "This function will never throw an exception." It's an optional -// optimization tool, but we may need to use it to match glibc prototypes. -#ifndef __THROW // I guess we're not on a glibc system -# define __THROW // __THROW is just an optimization, so ok to make it "" -#endif - -#if !HAVE_CFREE_SYMBOL -extern "C" void cfree(void* ptr) __THROW; -#endif -#if !HAVE_POSIX_MEMALIGN_SYMBOL -extern "C" int posix_memalign(void** ptr, size_t align, size_t size) __THROW; -#endif -#if !HAVE_MEMALIGN_SYMBOL -extern "C" void* memalign(size_t __alignment, size_t __size) __THROW; -#endif -#if !HAVE_VALLOC_SYMBOL -extern "C" void* valloc(size_t __size) __THROW; -#endif -#if !HAVE_PVALLOC_SYMBOL -extern "C" void* pvalloc(size_t __size) __THROW; -#endif diff --git a/contrib/libtcmalloc/src/tcmalloc_guard.h b/contrib/libtcmalloc/src/tcmalloc_guard.h deleted file mode 100644 index 84952bac2ea..00000000000 --- a/contrib/libtcmalloc/src/tcmalloc_guard.h +++ /dev/null @@ -1,49 +0,0 @@ -// -*- Mode: C++; c-basic-offset: 2; indent-tabs-mode: nil -*- -// Copyright (c) 2005, Google Inc. -// All rights reserved. -// -// Redistribution and use in source and binary forms, with or without -// modification, are permitted provided that the following conditions are -// met: -// -// * Redistributions of source code must retain the above copyright -// notice, this list of conditions and the following disclaimer. -// * Redistributions in binary form must reproduce the above -// copyright notice, this list of conditions and the following disclaimer -// in the documentation and/or other materials provided with the -// distribution. -// * Neither the name of Google Inc. nor the names of its -// contributors may be used to endorse or promote products derived from -// this software without specific prior written permission. -// -// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS -// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT -// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR -// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT -// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, -// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT -// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, -// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY -// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT -// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE -// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. - -// --- -// Author: Craig Silverstein -// -// We expose the TCMallocGuard class -- which initializes the tcmalloc -// allocator -- so classes that need to be sure tcmalloc is loaded -// before they do stuff -- notably heap-profiler -- can. To use this -// create a static TCMallocGuard instance at the top of a file where -// you need tcmalloc to be initialized before global constructors run. - -#ifndef TCMALLOC_TCMALLOC_GUARD_H_ -#define TCMALLOC_TCMALLOC_GUARD_H_ - -class TCMallocGuard { - public: - TCMallocGuard(); - ~TCMallocGuard(); -}; - -#endif // TCMALLOC_TCMALLOC_GUARD_H_ diff --git a/contrib/libtcmalloc/src/third_party/valgrind.h b/contrib/libtcmalloc/src/third_party/valgrind.h deleted file mode 100644 index 577c59ab0cd..00000000000 --- a/contrib/libtcmalloc/src/third_party/valgrind.h +++ /dev/null @@ -1,3924 +0,0 @@ -/* -*- c -*- - ---------------------------------------------------------------- - - Notice that the following BSD-style license applies to this one - file (valgrind.h) only. The rest of Valgrind is licensed under the - terms of the GNU General Public License, version 2, unless - otherwise indicated. See the COPYING file in the source - distribution for details. - - ---------------------------------------------------------------- - - This file is part of Valgrind, a dynamic binary instrumentation - framework. - - Copyright (C) 2000-2008 Julian Seward. All rights reserved. - - Redistribution and use in source and binary forms, with or without - modification, are permitted provided that the following conditions - are met: - - 1. Redistributions of source code must retain the above copyright - notice, this list of conditions and the following disclaimer. - - 2. The origin of this software must not be misrepresented; you must - not claim that you wrote the original software. If you use this - software in a product, an acknowledgment in the product - documentation would be appreciated but is not required. - - 3. Altered source versions must be plainly marked as such, and must - not be misrepresented as being the original software. - - 4. The name of the author may not be used to endorse or promote - products derived from this software without specific prior written - permission. - - THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS - OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED - WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE - ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY - DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL - DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE - GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS - INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, - WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING - NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS - SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. - - ---------------------------------------------------------------- - - Notice that the above BSD-style license applies to this one file - (valgrind.h) only. The entire rest of Valgrind is licensed under - the terms of the GNU General Public License, version 2. See the - COPYING file in the source distribution for details. - - ---------------------------------------------------------------- -*/ - - -/* This file is for inclusion into client (your!) code. - - You can use these macros to manipulate and query Valgrind's - execution inside your own programs. - - The resulting executables will still run without Valgrind, just a - little bit more slowly than they otherwise would, but otherwise - unchanged. When not running on valgrind, each client request - consumes very few (eg. 7) instructions, so the resulting performance - loss is negligible unless you plan to execute client requests - millions of times per second. Nevertheless, if that is still a - problem, you can compile with the NVALGRIND symbol defined (gcc - -DNVALGRIND) so that client requests are not even compiled in. */ - -#ifndef __VALGRIND_H -#define __VALGRIND_H - -#include - -/* Nb: this file might be included in a file compiled with -ansi. So - we can't use C++ style "//" comments nor the "asm" keyword (instead - use "__asm__"). */ - -/* Derive some tags indicating what the target platform is. Note - that in this file we're using the compiler's CPP symbols for - identifying architectures, which are different to the ones we use - within the rest of Valgrind. Note, __powerpc__ is active for both - 32 and 64-bit PPC, whereas __powerpc64__ is only active for the - latter (on Linux, that is). */ -#undef PLAT_x86_linux -#undef PLAT_amd64_linux -#undef PLAT_ppc32_linux -#undef PLAT_ppc64_linux -#undef PLAT_ppc32_aix5 -#undef PLAT_ppc64_aix5 - -#if !defined(_AIX) && defined(__i386__) -# define PLAT_x86_linux 1 -#elif !defined(_AIX) && defined(__x86_64__) -# define PLAT_amd64_linux 1 -#elif !defined(_AIX) && defined(__powerpc__) && !defined(__powerpc64__) -# define PLAT_ppc32_linux 1 -#elif !defined(_AIX) && defined(__powerpc__) && defined(__powerpc64__) -# define PLAT_ppc64_linux 1 -#elif defined(_AIX) && defined(__64BIT__) -# define PLAT_ppc64_aix5 1 -#elif defined(_AIX) && !defined(__64BIT__) -# define PLAT_ppc32_aix5 1 -#endif - - -/* If we're not compiling for our target platform, don't generate - any inline asms. */ -#if !defined(PLAT_x86_linux) && !defined(PLAT_amd64_linux) \ - && !defined(PLAT_ppc32_linux) && !defined(PLAT_ppc64_linux) \ - && !defined(PLAT_ppc32_aix5) && !defined(PLAT_ppc64_aix5) -# if !defined(NVALGRIND) -# define NVALGRIND 1 -# endif -#endif - - -/* ------------------------------------------------------------------ */ -/* ARCHITECTURE SPECIFICS for SPECIAL INSTRUCTIONS. There is nothing */ -/* in here of use to end-users -- skip to the next section. */ -/* ------------------------------------------------------------------ */ - -#if defined(NVALGRIND) - -/* Define NVALGRIND to completely remove the Valgrind magic sequence - from the compiled code (analogous to NDEBUG's effects on - assert()) */ -#define VALGRIND_DO_CLIENT_REQUEST( \ - _zzq_rlval, _zzq_default, _zzq_request, \ - _zzq_arg1, _zzq_arg2, _zzq_arg3, _zzq_arg4, _zzq_arg5) \ - { \ - (_zzq_rlval) = (_zzq_default); \ - } - -#else /* ! NVALGRIND */ - -/* The following defines the magic code sequences which the JITter - spots and handles magically. Don't look too closely at them as - they will rot your brain. - - The assembly code sequences for all architectures is in this one - file. This is because this file must be stand-alone, and we don't - want to have multiple files. - - For VALGRIND_DO_CLIENT_REQUEST, we must ensure that the default - value gets put in the return slot, so that everything works when - this is executed not under Valgrind. Args are passed in a memory - block, and so there's no intrinsic limit to the number that could - be passed, but it's currently five. - - The macro args are: - _zzq_rlval result lvalue - _zzq_default default value (result returned when running on real CPU) - _zzq_request request code - _zzq_arg1..5 request params - - The other two macros are used to support function wrapping, and are - a lot simpler. VALGRIND_GET_NR_CONTEXT returns the value of the - guest's NRADDR pseudo-register and whatever other information is - needed to safely run the call original from the wrapper: on - ppc64-linux, the R2 value at the divert point is also needed. This - information is abstracted into a user-visible type, OrigFn. - - VALGRIND_CALL_NOREDIR_* behaves the same as the following on the - guest, but guarantees that the branch instruction will not be - redirected: x86: call *%eax, amd64: call *%rax, ppc32/ppc64: - branch-and-link-to-r11. VALGRIND_CALL_NOREDIR is just text, not a - complete inline asm, since it needs to be combined with more magic - inline asm stuff to be useful. -*/ - -/* ------------------------- x86-linux ------------------------- */ - -#if defined(PLAT_x86_linux) - -typedef - struct { - unsigned int nraddr; /* where's the code? */ - } - OrigFn; - -#define __SPECIAL_INSTRUCTION_PREAMBLE \ - "roll $3, %%edi ; roll $13, %%edi\n\t" \ - "roll $29, %%edi ; roll $19, %%edi\n\t" - -#define VALGRIND_DO_CLIENT_REQUEST( \ - _zzq_rlval, _zzq_default, _zzq_request, \ - _zzq_arg1, _zzq_arg2, _zzq_arg3, _zzq_arg4, _zzq_arg5) \ - { volatile unsigned int _zzq_args[6]; \ - volatile unsigned int _zzq_result; \ - _zzq_args[0] = (unsigned int)(_zzq_request); \ - _zzq_args[1] = (unsigned int)(_zzq_arg1); \ - _zzq_args[2] = (unsigned int)(_zzq_arg2); \ - _zzq_args[3] = (unsigned int)(_zzq_arg3); \ - _zzq_args[4] = (unsigned int)(_zzq_arg4); \ - _zzq_args[5] = (unsigned int)(_zzq_arg5); \ - __asm__ volatile(__SPECIAL_INSTRUCTION_PREAMBLE \ - /* %EDX = client_request ( %EAX ) */ \ - "xchgl %%ebx,%%ebx" \ - : "=d" (_zzq_result) \ - : "a" (&_zzq_args[0]), "0" (_zzq_default) \ - : "cc", "memory" \ - ); \ - _zzq_rlval = _zzq_result; \ - } - -#define VALGRIND_GET_NR_CONTEXT(_zzq_rlval) \ - { volatile OrigFn* _zzq_orig = &(_zzq_rlval); \ - volatile unsigned int __addr; \ - __asm__ volatile(__SPECIAL_INSTRUCTION_PREAMBLE \ - /* %EAX = guest_NRADDR */ \ - "xchgl %%ecx,%%ecx" \ - : "=a" (__addr) \ - : \ - : "cc", "memory" \ - ); \ - _zzq_orig->nraddr = __addr; \ - } - -#define VALGRIND_CALL_NOREDIR_EAX \ - __SPECIAL_INSTRUCTION_PREAMBLE \ - /* call-noredir *%EAX */ \ - "xchgl %%edx,%%edx\n\t" -#endif /* PLAT_x86_linux */ - -/* ------------------------ amd64-linux ------------------------ */ - -#if defined(PLAT_amd64_linux) - -typedef - struct { - unsigned long long int nraddr; /* where's the code? */ - } - OrigFn; - -#define __SPECIAL_INSTRUCTION_PREAMBLE \ - "rolq $3, %%rdi ; rolq $13, %%rdi\n\t" \ - "rolq $61, %%rdi ; rolq $51, %%rdi\n\t" - -#define VALGRIND_DO_CLIENT_REQUEST( \ - _zzq_rlval, _zzq_default, _zzq_request, \ - _zzq_arg1, _zzq_arg2, _zzq_arg3, _zzq_arg4, _zzq_arg5) \ - { volatile unsigned long long int _zzq_args[6]; \ - volatile unsigned long long int _zzq_result; \ - _zzq_args[0] = (unsigned long long int)(_zzq_request); \ - _zzq_args[1] = (unsigned long long int)(_zzq_arg1); \ - _zzq_args[2] = (unsigned long long int)(_zzq_arg2); \ - _zzq_args[3] = (unsigned long long int)(_zzq_arg3); \ - _zzq_args[4] = (unsigned long long int)(_zzq_arg4); \ - _zzq_args[5] = (unsigned long long int)(_zzq_arg5); \ - __asm__ volatile(__SPECIAL_INSTRUCTION_PREAMBLE \ - /* %RDX = client_request ( %RAX ) */ \ - "xchgq %%rbx,%%rbx" \ - : "=d" (_zzq_result) \ - : "a" (&_zzq_args[0]), "0" (_zzq_default) \ - : "cc", "memory" \ - ); \ - _zzq_rlval = _zzq_result; \ - } - -#define VALGRIND_GET_NR_CONTEXT(_zzq_rlval) \ - { volatile OrigFn* _zzq_orig = &(_zzq_rlval); \ - volatile unsigned long long int __addr; \ - __asm__ volatile(__SPECIAL_INSTRUCTION_PREAMBLE \ - /* %RAX = guest_NRADDR */ \ - "xchgq %%rcx,%%rcx" \ - : "=a" (__addr) \ - : \ - : "cc", "memory" \ - ); \ - _zzq_orig->nraddr = __addr; \ - } - -#define VALGRIND_CALL_NOREDIR_RAX \ - __SPECIAL_INSTRUCTION_PREAMBLE \ - /* call-noredir *%RAX */ \ - "xchgq %%rdx,%%rdx\n\t" -#endif /* PLAT_amd64_linux */ - -/* ------------------------ ppc32-linux ------------------------ */ - -#if defined(PLAT_ppc32_linux) - -typedef - struct { - unsigned int nraddr; /* where's the code? */ - } - OrigFn; - -#define __SPECIAL_INSTRUCTION_PREAMBLE \ - "rlwinm 0,0,3,0,0 ; rlwinm 0,0,13,0,0\n\t" \ - "rlwinm 0,0,29,0,0 ; rlwinm 0,0,19,0,0\n\t" - -#define VALGRIND_DO_CLIENT_REQUEST( \ - _zzq_rlval, _zzq_default, _zzq_request, \ - _zzq_arg1, _zzq_arg2, _zzq_arg3, _zzq_arg4, _zzq_arg5) \ - \ - { unsigned int _zzq_args[6]; \ - unsigned int _zzq_result; \ - unsigned int* _zzq_ptr; \ - _zzq_args[0] = (unsigned int)(_zzq_request); \ - _zzq_args[1] = (unsigned int)(_zzq_arg1); \ - _zzq_args[2] = (unsigned int)(_zzq_arg2); \ - _zzq_args[3] = (unsigned int)(_zzq_arg3); \ - _zzq_args[4] = (unsigned int)(_zzq_arg4); \ - _zzq_args[5] = (unsigned int)(_zzq_arg5); \ - _zzq_ptr = _zzq_args; \ - __asm__ volatile("mr 3,%1\n\t" /*default*/ \ - "mr 4,%2\n\t" /*ptr*/ \ - __SPECIAL_INSTRUCTION_PREAMBLE \ - /* %R3 = client_request ( %R4 ) */ \ - "or 1,1,1\n\t" \ - "mr %0,3" /*result*/ \ - : "=b" (_zzq_result) \ - : "b" (_zzq_default), "b" (_zzq_ptr) \ - : "cc", "memory", "r3", "r4"); \ - _zzq_rlval = _zzq_result; \ - } - -#define VALGRIND_GET_NR_CONTEXT(_zzq_rlval) \ - { volatile OrigFn* _zzq_orig = &(_zzq_rlval); \ - unsigned int __addr; \ - __asm__ volatile(__SPECIAL_INSTRUCTION_PREAMBLE \ - /* %R3 = guest_NRADDR */ \ - "or 2,2,2\n\t" \ - "mr %0,3" \ - : "=b" (__addr) \ - : \ - : "cc", "memory", "r3" \ - ); \ - _zzq_orig->nraddr = __addr; \ - } - -#define VALGRIND_BRANCH_AND_LINK_TO_NOREDIR_R11 \ - __SPECIAL_INSTRUCTION_PREAMBLE \ - /* branch-and-link-to-noredir *%R11 */ \ - "or 3,3,3\n\t" -#endif /* PLAT_ppc32_linux */ - -/* ------------------------ ppc64-linux ------------------------ */ - -#if defined(PLAT_ppc64_linux) - -typedef - struct { - unsigned long long int nraddr; /* where's the code? */ - unsigned long long int r2; /* what tocptr do we need? */ - } - OrigFn; - -#define __SPECIAL_INSTRUCTION_PREAMBLE \ - "rotldi 0,0,3 ; rotldi 0,0,13\n\t" \ - "rotldi 0,0,61 ; rotldi 0,0,51\n\t" - -#define VALGRIND_DO_CLIENT_REQUEST( \ - _zzq_rlval, _zzq_default, _zzq_request, \ - _zzq_arg1, _zzq_arg2, _zzq_arg3, _zzq_arg4, _zzq_arg5) \ - \ - { unsigned long long int _zzq_args[6]; \ - register unsigned long long int _zzq_result __asm__("r3"); \ - register unsigned long long int* _zzq_ptr __asm__("r4"); \ - _zzq_args[0] = (unsigned long long int)(_zzq_request); \ - _zzq_args[1] = (unsigned long long int)(_zzq_arg1); \ - _zzq_args[2] = (unsigned long long int)(_zzq_arg2); \ - _zzq_args[3] = (unsigned long long int)(_zzq_arg3); \ - _zzq_args[4] = (unsigned long long int)(_zzq_arg4); \ - _zzq_args[5] = (unsigned long long int)(_zzq_arg5); \ - _zzq_ptr = _zzq_args; \ - __asm__ volatile(__SPECIAL_INSTRUCTION_PREAMBLE \ - /* %R3 = client_request ( %R4 ) */ \ - "or 1,1,1" \ - : "=r" (_zzq_result) \ - : "0" (_zzq_default), "r" (_zzq_ptr) \ - : "cc", "memory"); \ - _zzq_rlval = _zzq_result; \ - } - -#define VALGRIND_GET_NR_CONTEXT(_zzq_rlval) \ - { volatile OrigFn* _zzq_orig = &(_zzq_rlval); \ - register unsigned long long int __addr __asm__("r3"); \ - __asm__ volatile(__SPECIAL_INSTRUCTION_PREAMBLE \ - /* %R3 = guest_NRADDR */ \ - "or 2,2,2" \ - : "=r" (__addr) \ - : \ - : "cc", "memory" \ - ); \ - _zzq_orig->nraddr = __addr; \ - __asm__ volatile(__SPECIAL_INSTRUCTION_PREAMBLE \ - /* %R3 = guest_NRADDR_GPR2 */ \ - "or 4,4,4" \ - : "=r" (__addr) \ - : \ - : "cc", "memory" \ - ); \ - _zzq_orig->r2 = __addr; \ - } - -#define VALGRIND_BRANCH_AND_LINK_TO_NOREDIR_R11 \ - __SPECIAL_INSTRUCTION_PREAMBLE \ - /* branch-and-link-to-noredir *%R11 */ \ - "or 3,3,3\n\t" - -#endif /* PLAT_ppc64_linux */ - -/* ------------------------ ppc32-aix5 ------------------------- */ - -#if defined(PLAT_ppc32_aix5) - -typedef - struct { - unsigned int nraddr; /* where's the code? */ - unsigned int r2; /* what tocptr do we need? */ - } - OrigFn; - -#define __SPECIAL_INSTRUCTION_PREAMBLE \ - "rlwinm 0,0,3,0,0 ; rlwinm 0,0,13,0,0\n\t" \ - "rlwinm 0,0,29,0,0 ; rlwinm 0,0,19,0,0\n\t" - -#define VALGRIND_DO_CLIENT_REQUEST( \ - _zzq_rlval, _zzq_default, _zzq_request, \ - _zzq_arg1, _zzq_arg2, _zzq_arg3, _zzq_arg4, _zzq_arg5) \ - \ - { unsigned int _zzq_args[7]; \ - register unsigned int _zzq_result; \ - register unsigned int* _zzq_ptr; \ - _zzq_args[0] = (unsigned int)(_zzq_request); \ - _zzq_args[1] = (unsigned int)(_zzq_arg1); \ - _zzq_args[2] = (unsigned int)(_zzq_arg2); \ - _zzq_args[3] = (unsigned int)(_zzq_arg3); \ - _zzq_args[4] = (unsigned int)(_zzq_arg4); \ - _zzq_args[5] = (unsigned int)(_zzq_arg5); \ - _zzq_args[6] = (unsigned int)(_zzq_default); \ - _zzq_ptr = _zzq_args; \ - __asm__ volatile("mr 4,%1\n\t" \ - "lwz 3, 24(4)\n\t" \ - __SPECIAL_INSTRUCTION_PREAMBLE \ - /* %R3 = client_request ( %R4 ) */ \ - "or 1,1,1\n\t" \ - "mr %0,3" \ - : "=b" (_zzq_result) \ - : "b" (_zzq_ptr) \ - : "r3", "r4", "cc", "memory"); \ - _zzq_rlval = _zzq_result; \ - } - -#define VALGRIND_GET_NR_CONTEXT(_zzq_rlval) \ - { volatile OrigFn* _zzq_orig = &(_zzq_rlval); \ - register unsigned int __addr; \ - __asm__ volatile(__SPECIAL_INSTRUCTION_PREAMBLE \ - /* %R3 = guest_NRADDR */ \ - "or 2,2,2\n\t" \ - "mr %0,3" \ - : "=b" (__addr) \ - : \ - : "r3", "cc", "memory" \ - ); \ - _zzq_orig->nraddr = __addr; \ - __asm__ volatile(__SPECIAL_INSTRUCTION_PREAMBLE \ - /* %R3 = guest_NRADDR_GPR2 */ \ - "or 4,4,4\n\t" \ - "mr %0,3" \ - : "=b" (__addr) \ - : \ - : "r3", "cc", "memory" \ - ); \ - _zzq_orig->r2 = __addr; \ - } - -#define VALGRIND_BRANCH_AND_LINK_TO_NOREDIR_R11 \ - __SPECIAL_INSTRUCTION_PREAMBLE \ - /* branch-and-link-to-noredir *%R11 */ \ - "or 3,3,3\n\t" - -#endif /* PLAT_ppc32_aix5 */ - -/* ------------------------ ppc64-aix5 ------------------------- */ - -#if defined(PLAT_ppc64_aix5) - -typedef - struct { - unsigned long long int nraddr; /* where's the code? */ - unsigned long long int r2; /* what tocptr do we need? */ - } - OrigFn; - -#define __SPECIAL_INSTRUCTION_PREAMBLE \ - "rotldi 0,0,3 ; rotldi 0,0,13\n\t" \ - "rotldi 0,0,61 ; rotldi 0,0,51\n\t" - -#define VALGRIND_DO_CLIENT_REQUEST( \ - _zzq_rlval, _zzq_default, _zzq_request, \ - _zzq_arg1, _zzq_arg2, _zzq_arg3, _zzq_arg4, _zzq_arg5) \ - \ - { unsigned long long int _zzq_args[7]; \ - register unsigned long long int _zzq_result; \ - register unsigned long long int* _zzq_ptr; \ - _zzq_args[0] = (unsigned int long long)(_zzq_request); \ - _zzq_args[1] = (unsigned int long long)(_zzq_arg1); \ - _zzq_args[2] = (unsigned int long long)(_zzq_arg2); \ - _zzq_args[3] = (unsigned int long long)(_zzq_arg3); \ - _zzq_args[4] = (unsigned int long long)(_zzq_arg4); \ - _zzq_args[5] = (unsigned int long long)(_zzq_arg5); \ - _zzq_args[6] = (unsigned int long long)(_zzq_default); \ - _zzq_ptr = _zzq_args; \ - __asm__ volatile("mr 4,%1\n\t" \ - "ld 3, 48(4)\n\t" \ - __SPECIAL_INSTRUCTION_PREAMBLE \ - /* %R3 = client_request ( %R4 ) */ \ - "or 1,1,1\n\t" \ - "mr %0,3" \ - : "=b" (_zzq_result) \ - : "b" (_zzq_ptr) \ - : "r3", "r4", "cc", "memory"); \ - _zzq_rlval = _zzq_result; \ - } - -#define VALGRIND_GET_NR_CONTEXT(_zzq_rlval) \ - { volatile OrigFn* _zzq_orig = &(_zzq_rlval); \ - register unsigned long long int __addr; \ - __asm__ volatile(__SPECIAL_INSTRUCTION_PREAMBLE \ - /* %R3 = guest_NRADDR */ \ - "or 2,2,2\n\t" \ - "mr %0,3" \ - : "=b" (__addr) \ - : \ - : "r3", "cc", "memory" \ - ); \ - _zzq_orig->nraddr = __addr; \ - __asm__ volatile(__SPECIAL_INSTRUCTION_PREAMBLE \ - /* %R3 = guest_NRADDR_GPR2 */ \ - "or 4,4,4\n\t" \ - "mr %0,3" \ - : "=b" (__addr) \ - : \ - : "r3", "cc", "memory" \ - ); \ - _zzq_orig->r2 = __addr; \ - } - -#define VALGRIND_BRANCH_AND_LINK_TO_NOREDIR_R11 \ - __SPECIAL_INSTRUCTION_PREAMBLE \ - /* branch-and-link-to-noredir *%R11 */ \ - "or 3,3,3\n\t" - -#endif /* PLAT_ppc64_aix5 */ - -/* Insert assembly code for other platforms here... */ - -#endif /* NVALGRIND */ - - -/* ------------------------------------------------------------------ */ -/* PLATFORM SPECIFICS for FUNCTION WRAPPING. This is all very */ -/* ugly. It's the least-worst tradeoff I can think of. */ -/* ------------------------------------------------------------------ */ - -/* This section defines magic (a.k.a appalling-hack) macros for doing - guaranteed-no-redirection macros, so as to get from function - wrappers to the functions they are wrapping. The whole point is to - construct standard call sequences, but to do the call itself with a - special no-redirect call pseudo-instruction that the JIT - understands and handles specially. This section is long and - repetitious, and I can't see a way to make it shorter. - - The naming scheme is as follows: - - CALL_FN_{W,v}_{v,W,WW,WWW,WWWW,5W,6W,7W,etc} - - 'W' stands for "word" and 'v' for "void". Hence there are - different macros for calling arity 0, 1, 2, 3, 4, etc, functions, - and for each, the possibility of returning a word-typed result, or - no result. -*/ - -/* Use these to write the name of your wrapper. NOTE: duplicates - VG_WRAP_FUNCTION_Z{U,Z} in pub_tool_redir.h. */ - -#define I_WRAP_SONAME_FNNAME_ZU(soname,fnname) \ - _vgwZU_##soname##_##fnname - -#define I_WRAP_SONAME_FNNAME_ZZ(soname,fnname) \ - _vgwZZ_##soname##_##fnname - -/* Use this macro from within a wrapper function to collect the - context (address and possibly other info) of the original function. - Once you have that you can then use it in one of the CALL_FN_ - macros. The type of the argument _lval is OrigFn. */ -#define VALGRIND_GET_ORIG_FN(_lval) VALGRIND_GET_NR_CONTEXT(_lval) - -/* Derivatives of the main macros below, for calling functions - returning void. */ - -#define CALL_FN_v_v(fnptr) \ - do { volatile unsigned long _junk; \ - CALL_FN_W_v(_junk,fnptr); } while (0) - -#define CALL_FN_v_W(fnptr, arg1) \ - do { volatile unsigned long _junk; \ - CALL_FN_W_W(_junk,fnptr,arg1); } while (0) - -#define CALL_FN_v_WW(fnptr, arg1,arg2) \ - do { volatile unsigned long _junk; \ - CALL_FN_W_WW(_junk,fnptr,arg1,arg2); } while (0) - -#define CALL_FN_v_WWW(fnptr, arg1,arg2,arg3) \ - do { volatile unsigned long _junk; \ - CALL_FN_W_WWW(_junk,fnptr,arg1,arg2,arg3); } while (0) - -/* ------------------------- x86-linux ------------------------- */ - -#if defined(PLAT_x86_linux) - -/* These regs are trashed by the hidden call. No need to mention eax - as gcc can already see that, plus causes gcc to bomb. */ -#define __CALLER_SAVED_REGS /*"eax"*/ "ecx", "edx" - -/* These CALL_FN_ macros assume that on x86-linux, sizeof(unsigned - long) == 4. */ - -#define CALL_FN_W_v(lval, orig) \ - do { \ - volatile OrigFn _orig = (orig); \ - volatile unsigned long _argvec[1]; \ - volatile unsigned long _res; \ - _argvec[0] = (unsigned long)_orig.nraddr; \ - __asm__ volatile( \ - "movl (%%eax), %%eax\n\t" /* target->%eax */ \ - VALGRIND_CALL_NOREDIR_EAX \ - : /*out*/ "=a" (_res) \ - : /*in*/ "a" (&_argvec[0]) \ - : /*trash*/ "cc", "memory", __CALLER_SAVED_REGS \ - ); \ - lval = (__typeof__(lval)) _res; \ - } while (0) - -#define CALL_FN_W_W(lval, orig, arg1) \ - do { \ - volatile OrigFn _orig = (orig); \ - volatile unsigned long _argvec[2]; \ - volatile unsigned long _res; \ - _argvec[0] = (unsigned long)_orig.nraddr; \ - _argvec[1] = (unsigned long)(arg1); \ - __asm__ volatile( \ - "pushl 4(%%eax)\n\t" \ - "movl (%%eax), %%eax\n\t" /* target->%eax */ \ - VALGRIND_CALL_NOREDIR_EAX \ - "addl $4, %%esp\n" \ - : /*out*/ "=a" (_res) \ - : /*in*/ "a" (&_argvec[0]) \ - : /*trash*/ "cc", "memory", __CALLER_SAVED_REGS \ - ); \ - lval = (__typeof__(lval)) _res; \ - } while (0) - -#define CALL_FN_W_WW(lval, orig, arg1,arg2) \ - do { \ - volatile OrigFn _orig = (orig); \ - volatile unsigned long _argvec[3]; \ - volatile unsigned long _res; \ - _argvec[0] = (unsigned long)_orig.nraddr; \ - _argvec[1] = (unsigned long)(arg1); \ - _argvec[2] = (unsigned long)(arg2); \ - __asm__ volatile( \ - "pushl 8(%%eax)\n\t" \ - "pushl 4(%%eax)\n\t" \ - "movl (%%eax), %%eax\n\t" /* target->%eax */ \ - VALGRIND_CALL_NOREDIR_EAX \ - "addl $8, %%esp\n" \ - : /*out*/ "=a" (_res) \ - : /*in*/ "a" (&_argvec[0]) \ - : /*trash*/ "cc", "memory", __CALLER_SAVED_REGS \ - ); \ - lval = (__typeof__(lval)) _res; \ - } while (0) - -#define CALL_FN_W_WWW(lval, orig, arg1,arg2,arg3) \ - do { \ - volatile OrigFn _orig = (orig); \ - volatile unsigned long _argvec[4]; \ - volatile unsigned long _res; \ - _argvec[0] = (unsigned long)_orig.nraddr; \ - _argvec[1] = (unsigned long)(arg1); \ - _argvec[2] = (unsigned long)(arg2); \ - _argvec[3] = (unsigned long)(arg3); \ - __asm__ volatile( \ - "pushl 12(%%eax)\n\t" \ - "pushl 8(%%eax)\n\t" \ - "pushl 4(%%eax)\n\t" \ - "movl (%%eax), %%eax\n\t" /* target->%eax */ \ - VALGRIND_CALL_NOREDIR_EAX \ - "addl $12, %%esp\n" \ - : /*out*/ "=a" (_res) \ - : /*in*/ "a" (&_argvec[0]) \ - : /*trash*/ "cc", "memory", __CALLER_SAVED_REGS \ - ); \ - lval = (__typeof__(lval)) _res; \ - } while (0) - -#define CALL_FN_W_WWWW(lval, orig, arg1,arg2,arg3,arg4) \ - do { \ - volatile OrigFn _orig = (orig); \ - volatile unsigned long _argvec[5]; \ - volatile unsigned long _res; \ - _argvec[0] = (unsigned long)_orig.nraddr; \ - _argvec[1] = (unsigned long)(arg1); \ - _argvec[2] = (unsigned long)(arg2); \ - _argvec[3] = (unsigned long)(arg3); \ - _argvec[4] = (unsigned long)(arg4); \ - __asm__ volatile( \ - "pushl 16(%%eax)\n\t" \ - "pushl 12(%%eax)\n\t" \ - "pushl 8(%%eax)\n\t" \ - "pushl 4(%%eax)\n\t" \ - "movl (%%eax), %%eax\n\t" /* target->%eax */ \ - VALGRIND_CALL_NOREDIR_EAX \ - "addl $16, %%esp\n" \ - : /*out*/ "=a" (_res) \ - : /*in*/ "a" (&_argvec[0]) \ - : /*trash*/ "cc", "memory", __CALLER_SAVED_REGS \ - ); \ - lval = (__typeof__(lval)) _res; \ - } while (0) - -#define CALL_FN_W_5W(lval, orig, arg1,arg2,arg3,arg4,arg5) \ - do { \ - volatile OrigFn _orig = (orig); \ - volatile unsigned long _argvec[6]; \ - volatile unsigned long _res; \ - _argvec[0] = (unsigned long)_orig.nraddr; \ - _argvec[1] = (unsigned long)(arg1); \ - _argvec[2] = (unsigned long)(arg2); \ - _argvec[3] = (unsigned long)(arg3); \ - _argvec[4] = (unsigned long)(arg4); \ - _argvec[5] = (unsigned long)(arg5); \ - __asm__ volatile( \ - "pushl 20(%%eax)\n\t" \ - "pushl 16(%%eax)\n\t" \ - "pushl 12(%%eax)\n\t" \ - "pushl 8(%%eax)\n\t" \ - "pushl 4(%%eax)\n\t" \ - "movl (%%eax), %%eax\n\t" /* target->%eax */ \ - VALGRIND_CALL_NOREDIR_EAX \ - "addl $20, %%esp\n" \ - : /*out*/ "=a" (_res) \ - : /*in*/ "a" (&_argvec[0]) \ - : /*trash*/ "cc", "memory", __CALLER_SAVED_REGS \ - ); \ - lval = (__typeof__(lval)) _res; \ - } while (0) - -#define CALL_FN_W_6W(lval, orig, arg1,arg2,arg3,arg4,arg5,arg6) \ - do { \ - volatile OrigFn _orig = (orig); \ - volatile unsigned long _argvec[7]; \ - volatile unsigned long _res; \ - _argvec[0] = (unsigned long)_orig.nraddr; \ - _argvec[1] = (unsigned long)(arg1); \ - _argvec[2] = (unsigned long)(arg2); \ - _argvec[3] = (unsigned long)(arg3); \ - _argvec[4] = (unsigned long)(arg4); \ - _argvec[5] = (unsigned long)(arg5); \ - _argvec[6] = (unsigned long)(arg6); \ - __asm__ volatile( \ - "pushl 24(%%eax)\n\t" \ - "pushl 20(%%eax)\n\t" \ - "pushl 16(%%eax)\n\t" \ - "pushl 12(%%eax)\n\t" \ - "pushl 8(%%eax)\n\t" \ - "pushl 4(%%eax)\n\t" \ - "movl (%%eax), %%eax\n\t" /* target->%eax */ \ - VALGRIND_CALL_NOREDIR_EAX \ - "addl $24, %%esp\n" \ - : /*out*/ "=a" (_res) \ - : /*in*/ "a" (&_argvec[0]) \ - : /*trash*/ "cc", "memory", __CALLER_SAVED_REGS \ - ); \ - lval = (__typeof__(lval)) _res; \ - } while (0) - -#define CALL_FN_W_7W(lval, orig, arg1,arg2,arg3,arg4,arg5,arg6, \ - arg7) \ - do { \ - volatile OrigFn _orig = (orig); \ - volatile unsigned long _argvec[8]; \ - volatile unsigned long _res; \ - _argvec[0] = (unsigned long)_orig.nraddr; \ - _argvec[1] = (unsigned long)(arg1); \ - _argvec[2] = (unsigned long)(arg2); \ - _argvec[3] = (unsigned long)(arg3); \ - _argvec[4] = (unsigned long)(arg4); \ - _argvec[5] = (unsigned long)(arg5); \ - _argvec[6] = (unsigned long)(arg6); \ - _argvec[7] = (unsigned long)(arg7); \ - __asm__ volatile( \ - "pushl 28(%%eax)\n\t" \ - "pushl 24(%%eax)\n\t" \ - "pushl 20(%%eax)\n\t" \ - "pushl 16(%%eax)\n\t" \ - "pushl 12(%%eax)\n\t" \ - "pushl 8(%%eax)\n\t" \ - "pushl 4(%%eax)\n\t" \ - "movl (%%eax), %%eax\n\t" /* target->%eax */ \ - VALGRIND_CALL_NOREDIR_EAX \ - "addl $28, %%esp\n" \ - : /*out*/ "=a" (_res) \ - : /*in*/ "a" (&_argvec[0]) \ - : /*trash*/ "cc", "memory", __CALLER_SAVED_REGS \ - ); \ - lval = (__typeof__(lval)) _res; \ - } while (0) - -#define CALL_FN_W_8W(lval, orig, arg1,arg2,arg3,arg4,arg5,arg6, \ - arg7,arg8) \ - do { \ - volatile OrigFn _orig = (orig); \ - volatile unsigned long _argvec[9]; \ - volatile unsigned long _res; \ - _argvec[0] = (unsigned long)_orig.nraddr; \ - _argvec[1] = (unsigned long)(arg1); \ - _argvec[2] = (unsigned long)(arg2); \ - _argvec[3] = (unsigned long)(arg3); \ - _argvec[4] = (unsigned long)(arg4); \ - _argvec[5] = (unsigned long)(arg5); \ - _argvec[6] = (unsigned long)(arg6); \ - _argvec[7] = (unsigned long)(arg7); \ - _argvec[8] = (unsigned long)(arg8); \ - __asm__ volatile( \ - "pushl 32(%%eax)\n\t" \ - "pushl 28(%%eax)\n\t" \ - "pushl 24(%%eax)\n\t" \ - "pushl 20(%%eax)\n\t" \ - "pushl 16(%%eax)\n\t" \ - "pushl 12(%%eax)\n\t" \ - "pushl 8(%%eax)\n\t" \ - "pushl 4(%%eax)\n\t" \ - "movl (%%eax), %%eax\n\t" /* target->%eax */ \ - VALGRIND_CALL_NOREDIR_EAX \ - "addl $32, %%esp\n" \ - : /*out*/ "=a" (_res) \ - : /*in*/ "a" (&_argvec[0]) \ - : /*trash*/ "cc", "memory", __CALLER_SAVED_REGS \ - ); \ - lval = (__typeof__(lval)) _res; \ - } while (0) - -#define CALL_FN_W_9W(lval, orig, arg1,arg2,arg3,arg4,arg5,arg6, \ - arg7,arg8,arg9) \ - do { \ - volatile OrigFn _orig = (orig); \ - volatile unsigned long _argvec[10]; \ - volatile unsigned long _res; \ - _argvec[0] = (unsigned long)_orig.nraddr; \ - _argvec[1] = (unsigned long)(arg1); \ - _argvec[2] = (unsigned long)(arg2); \ - _argvec[3] = (unsigned long)(arg3); \ - _argvec[4] = (unsigned long)(arg4); \ - _argvec[5] = (unsigned long)(arg5); \ - _argvec[6] = (unsigned long)(arg6); \ - _argvec[7] = (unsigned long)(arg7); \ - _argvec[8] = (unsigned long)(arg8); \ - _argvec[9] = (unsigned long)(arg9); \ - __asm__ volatile( \ - "pushl 36(%%eax)\n\t" \ - "pushl 32(%%eax)\n\t" \ - "pushl 28(%%eax)\n\t" \ - "pushl 24(%%eax)\n\t" \ - "pushl 20(%%eax)\n\t" \ - "pushl 16(%%eax)\n\t" \ - "pushl 12(%%eax)\n\t" \ - "pushl 8(%%eax)\n\t" \ - "pushl 4(%%eax)\n\t" \ - "movl (%%eax), %%eax\n\t" /* target->%eax */ \ - VALGRIND_CALL_NOREDIR_EAX \ - "addl $36, %%esp\n" \ - : /*out*/ "=a" (_res) \ - : /*in*/ "a" (&_argvec[0]) \ - : /*trash*/ "cc", "memory", __CALLER_SAVED_REGS \ - ); \ - lval = (__typeof__(lval)) _res; \ - } while (0) - -#define CALL_FN_W_10W(lval, orig, arg1,arg2,arg3,arg4,arg5,arg6, \ - arg7,arg8,arg9,arg10) \ - do { \ - volatile OrigFn _orig = (orig); \ - volatile unsigned long _argvec[11]; \ - volatile unsigned long _res; \ - _argvec[0] = (unsigned long)_orig.nraddr; \ - _argvec[1] = (unsigned long)(arg1); \ - _argvec[2] = (unsigned long)(arg2); \ - _argvec[3] = (unsigned long)(arg3); \ - _argvec[4] = (unsigned long)(arg4); \ - _argvec[5] = (unsigned long)(arg5); \ - _argvec[6] = (unsigned long)(arg6); \ - _argvec[7] = (unsigned long)(arg7); \ - _argvec[8] = (unsigned long)(arg8); \ - _argvec[9] = (unsigned long)(arg9); \ - _argvec[10] = (unsigned long)(arg10); \ - __asm__ volatile( \ - "pushl 40(%%eax)\n\t" \ - "pushl 36(%%eax)\n\t" \ - "pushl 32(%%eax)\n\t" \ - "pushl 28(%%eax)\n\t" \ - "pushl 24(%%eax)\n\t" \ - "pushl 20(%%eax)\n\t" \ - "pushl 16(%%eax)\n\t" \ - "pushl 12(%%eax)\n\t" \ - "pushl 8(%%eax)\n\t" \ - "pushl 4(%%eax)\n\t" \ - "movl (%%eax), %%eax\n\t" /* target->%eax */ \ - VALGRIND_CALL_NOREDIR_EAX \ - "addl $40, %%esp\n" \ - : /*out*/ "=a" (_res) \ - : /*in*/ "a" (&_argvec[0]) \ - : /*trash*/ "cc", "memory", __CALLER_SAVED_REGS \ - ); \ - lval = (__typeof__(lval)) _res; \ - } while (0) - -#define CALL_FN_W_11W(lval, orig, arg1,arg2,arg3,arg4,arg5, \ - arg6,arg7,arg8,arg9,arg10, \ - arg11) \ - do { \ - volatile OrigFn _orig = (orig); \ - volatile unsigned long _argvec[12]; \ - volatile unsigned long _res; \ - _argvec[0] = (unsigned long)_orig.nraddr; \ - _argvec[1] = (unsigned long)(arg1); \ - _argvec[2] = (unsigned long)(arg2); \ - _argvec[3] = (unsigned long)(arg3); \ - _argvec[4] = (unsigned long)(arg4); \ - _argvec[5] = (unsigned long)(arg5); \ - _argvec[6] = (unsigned long)(arg6); \ - _argvec[7] = (unsigned long)(arg7); \ - _argvec[8] = (unsigned long)(arg8); \ - _argvec[9] = (unsigned long)(arg9); \ - _argvec[10] = (unsigned long)(arg10); \ - _argvec[11] = (unsigned long)(arg11); \ - __asm__ volatile( \ - "pushl 44(%%eax)\n\t" \ - "pushl 40(%%eax)\n\t" \ - "pushl 36(%%eax)\n\t" \ - "pushl 32(%%eax)\n\t" \ - "pushl 28(%%eax)\n\t" \ - "pushl 24(%%eax)\n\t" \ - "pushl 20(%%eax)\n\t" \ - "pushl 16(%%eax)\n\t" \ - "pushl 12(%%eax)\n\t" \ - "pushl 8(%%eax)\n\t" \ - "pushl 4(%%eax)\n\t" \ - "movl (%%eax), %%eax\n\t" /* target->%eax */ \ - VALGRIND_CALL_NOREDIR_EAX \ - "addl $44, %%esp\n" \ - : /*out*/ "=a" (_res) \ - : /*in*/ "a" (&_argvec[0]) \ - : /*trash*/ "cc", "memory", __CALLER_SAVED_REGS \ - ); \ - lval = (__typeof__(lval)) _res; \ - } while (0) - -#define CALL_FN_W_12W(lval, orig, arg1,arg2,arg3,arg4,arg5, \ - arg6,arg7,arg8,arg9,arg10, \ - arg11,arg12) \ - do { \ - volatile OrigFn _orig = (orig); \ - volatile unsigned long _argvec[13]; \ - volatile unsigned long _res; \ - _argvec[0] = (unsigned long)_orig.nraddr; \ - _argvec[1] = (unsigned long)(arg1); \ - _argvec[2] = (unsigned long)(arg2); \ - _argvec[3] = (unsigned long)(arg3); \ - _argvec[4] = (unsigned long)(arg4); \ - _argvec[5] = (unsigned long)(arg5); \ - _argvec[6] = (unsigned long)(arg6); \ - _argvec[7] = (unsigned long)(arg7); \ - _argvec[8] = (unsigned long)(arg8); \ - _argvec[9] = (unsigned long)(arg9); \ - _argvec[10] = (unsigned long)(arg10); \ - _argvec[11] = (unsigned long)(arg11); \ - _argvec[12] = (unsigned long)(arg12); \ - __asm__ volatile( \ - "pushl 48(%%eax)\n\t" \ - "pushl 44(%%eax)\n\t" \ - "pushl 40(%%eax)\n\t" \ - "pushl 36(%%eax)\n\t" \ - "pushl 32(%%eax)\n\t" \ - "pushl 28(%%eax)\n\t" \ - "pushl 24(%%eax)\n\t" \ - "pushl 20(%%eax)\n\t" \ - "pushl 16(%%eax)\n\t" \ - "pushl 12(%%eax)\n\t" \ - "pushl 8(%%eax)\n\t" \ - "pushl 4(%%eax)\n\t" \ - "movl (%%eax), %%eax\n\t" /* target->%eax */ \ - VALGRIND_CALL_NOREDIR_EAX \ - "addl $48, %%esp\n" \ - : /*out*/ "=a" (_res) \ - : /*in*/ "a" (&_argvec[0]) \ - : /*trash*/ "cc", "memory", __CALLER_SAVED_REGS \ - ); \ - lval = (__typeof__(lval)) _res; \ - } while (0) - -#endif /* PLAT_x86_linux */ - -/* ------------------------ amd64-linux ------------------------ */ - -#if defined(PLAT_amd64_linux) - -/* ARGREGS: rdi rsi rdx rcx r8 r9 (the rest on stack in R-to-L order) */ - -/* These regs are trashed by the hidden call. */ -#define __CALLER_SAVED_REGS /*"rax",*/ "rcx", "rdx", "rsi", \ - "rdi", "r8", "r9", "r10", "r11" - -/* These CALL_FN_ macros assume that on amd64-linux, sizeof(unsigned - long) == 8. */ - -/* NB 9 Sept 07. There is a nasty kludge here in all these CALL_FN_ - macros. In order not to trash the stack redzone, we need to drop - %rsp by 128 before the hidden call, and restore afterwards. The - nastyness is that it is only by luck that the stack still appears - to be unwindable during the hidden call - since then the behaviour - of any routine using this macro does not match what the CFI data - says. Sigh. - - Why is this important? Imagine that a wrapper has a stack - allocated local, and passes to the hidden call, a pointer to it. - Because gcc does not know about the hidden call, it may allocate - that local in the redzone. Unfortunately the hidden call may then - trash it before it comes to use it. So we must step clear of the - redzone, for the duration of the hidden call, to make it safe. - - Probably the same problem afflicts the other redzone-style ABIs too - (ppc64-linux, ppc32-aix5, ppc64-aix5); but for those, the stack is - self describing (none of this CFI nonsense) so at least messing - with the stack pointer doesn't give a danger of non-unwindable - stack. */ - -#define CALL_FN_W_v(lval, orig) \ - do { \ - volatile OrigFn _orig = (orig); \ - volatile unsigned long _argvec[1]; \ - volatile unsigned long _res; \ - _argvec[0] = (unsigned long)_orig.nraddr; \ - __asm__ volatile( \ - "subq $128,%%rsp\n\t" \ - "movq (%%rax), %%rax\n\t" /* target->%rax */ \ - VALGRIND_CALL_NOREDIR_RAX \ - "addq $128,%%rsp\n\t" \ - : /*out*/ "=a" (_res) \ - : /*in*/ "a" (&_argvec[0]) \ - : /*trash*/ "cc", "memory", __CALLER_SAVED_REGS \ - ); \ - lval = (__typeof__(lval)) _res; \ - } while (0) - -#define CALL_FN_W_W(lval, orig, arg1) \ - do { \ - volatile OrigFn _orig = (orig); \ - volatile unsigned long _argvec[2]; \ - volatile unsigned long _res; \ - _argvec[0] = (unsigned long)_orig.nraddr; \ - _argvec[1] = (unsigned long)(arg1); \ - __asm__ volatile( \ - "subq $128,%%rsp\n\t" \ - "movq 8(%%rax), %%rdi\n\t" \ - "movq (%%rax), %%rax\n\t" /* target->%rax */ \ - VALGRIND_CALL_NOREDIR_RAX \ - "addq $128,%%rsp\n\t" \ - : /*out*/ "=a" (_res) \ - : /*in*/ "a" (&_argvec[0]) \ - : /*trash*/ "cc", "memory", __CALLER_SAVED_REGS \ - ); \ - lval = (__typeof__(lval)) _res; \ - } while (0) - -#define CALL_FN_W_WW(lval, orig, arg1,arg2) \ - do { \ - volatile OrigFn _orig = (orig); \ - volatile unsigned long _argvec[3]; \ - volatile unsigned long _res; \ - _argvec[0] = (unsigned long)_orig.nraddr; \ - _argvec[1] = (unsigned long)(arg1); \ - _argvec[2] = (unsigned long)(arg2); \ - __asm__ volatile( \ - "subq $128,%%rsp\n\t" \ - "movq 16(%%rax), %%rsi\n\t" \ - "movq 8(%%rax), %%rdi\n\t" \ - "movq (%%rax), %%rax\n\t" /* target->%rax */ \ - VALGRIND_CALL_NOREDIR_RAX \ - "addq $128,%%rsp\n\t" \ - : /*out*/ "=a" (_res) \ - : /*in*/ "a" (&_argvec[0]) \ - : /*trash*/ "cc", "memory", __CALLER_SAVED_REGS \ - ); \ - lval = (__typeof__(lval)) _res; \ - } while (0) - -#define CALL_FN_W_WWW(lval, orig, arg1,arg2,arg3) \ - do { \ - volatile OrigFn _orig = (orig); \ - volatile unsigned long _argvec[4]; \ - volatile unsigned long _res; \ - _argvec[0] = (unsigned long)_orig.nraddr; \ - _argvec[1] = (unsigned long)(arg1); \ - _argvec[2] = (unsigned long)(arg2); \ - _argvec[3] = (unsigned long)(arg3); \ - __asm__ volatile( \ - "subq $128,%%rsp\n\t" \ - "movq 24(%%rax), %%rdx\n\t" \ - "movq 16(%%rax), %%rsi\n\t" \ - "movq 8(%%rax), %%rdi\n\t" \ - "movq (%%rax), %%rax\n\t" /* target->%rax */ \ - VALGRIND_CALL_NOREDIR_RAX \ - "addq $128,%%rsp\n\t" \ - : /*out*/ "=a" (_res) \ - : /*in*/ "a" (&_argvec[0]) \ - : /*trash*/ "cc", "memory", __CALLER_SAVED_REGS \ - ); \ - lval = (__typeof__(lval)) _res; \ - } while (0) - -#define CALL_FN_W_WWWW(lval, orig, arg1,arg2,arg3,arg4) \ - do { \ - volatile OrigFn _orig = (orig); \ - volatile unsigned long _argvec[5]; \ - volatile unsigned long _res; \ - _argvec[0] = (unsigned long)_orig.nraddr; \ - _argvec[1] = (unsigned long)(arg1); \ - _argvec[2] = (unsigned long)(arg2); \ - _argvec[3] = (unsigned long)(arg3); \ - _argvec[4] = (unsigned long)(arg4); \ - __asm__ volatile( \ - "subq $128,%%rsp\n\t" \ - "movq 32(%%rax), %%rcx\n\t" \ - "movq 24(%%rax), %%rdx\n\t" \ - "movq 16(%%rax), %%rsi\n\t" \ - "movq 8(%%rax), %%rdi\n\t" \ - "movq (%%rax), %%rax\n\t" /* target->%rax */ \ - VALGRIND_CALL_NOREDIR_RAX \ - "addq $128,%%rsp\n\t" \ - : /*out*/ "=a" (_res) \ - : /*in*/ "a" (&_argvec[0]) \ - : /*trash*/ "cc", "memory", __CALLER_SAVED_REGS \ - ); \ - lval = (__typeof__(lval)) _res; \ - } while (0) - -#define CALL_FN_W_5W(lval, orig, arg1,arg2,arg3,arg4,arg5) \ - do { \ - volatile OrigFn _orig = (orig); \ - volatile unsigned long _argvec[6]; \ - volatile unsigned long _res; \ - _argvec[0] = (unsigned long)_orig.nraddr; \ - _argvec[1] = (unsigned long)(arg1); \ - _argvec[2] = (unsigned long)(arg2); \ - _argvec[3] = (unsigned long)(arg3); \ - _argvec[4] = (unsigned long)(arg4); \ - _argvec[5] = (unsigned long)(arg5); \ - __asm__ volatile( \ - "subq $128,%%rsp\n\t" \ - "movq 40(%%rax), %%r8\n\t" \ - "movq 32(%%rax), %%rcx\n\t" \ - "movq 24(%%rax), %%rdx\n\t" \ - "movq 16(%%rax), %%rsi\n\t" \ - "movq 8(%%rax), %%rdi\n\t" \ - "movq (%%rax), %%rax\n\t" /* target->%rax */ \ - VALGRIND_CALL_NOREDIR_RAX \ - "addq $128,%%rsp\n\t" \ - : /*out*/ "=a" (_res) \ - : /*in*/ "a" (&_argvec[0]) \ - : /*trash*/ "cc", "memory", __CALLER_SAVED_REGS \ - ); \ - lval = (__typeof__(lval)) _res; \ - } while (0) - -#define CALL_FN_W_6W(lval, orig, arg1,arg2,arg3,arg4,arg5,arg6) \ - do { \ - volatile OrigFn _orig = (orig); \ - volatile unsigned long _argvec[7]; \ - volatile unsigned long _res; \ - _argvec[0] = (unsigned long)_orig.nraddr; \ - _argvec[1] = (unsigned long)(arg1); \ - _argvec[2] = (unsigned long)(arg2); \ - _argvec[3] = (unsigned long)(arg3); \ - _argvec[4] = (unsigned long)(arg4); \ - _argvec[5] = (unsigned long)(arg5); \ - _argvec[6] = (unsigned long)(arg6); \ - __asm__ volatile( \ - "subq $128,%%rsp\n\t" \ - "movq 48(%%rax), %%r9\n\t" \ - "movq 40(%%rax), %%r8\n\t" \ - "movq 32(%%rax), %%rcx\n\t" \ - "movq 24(%%rax), %%rdx\n\t" \ - "movq 16(%%rax), %%rsi\n\t" \ - "movq 8(%%rax), %%rdi\n\t" \ - "movq (%%rax), %%rax\n\t" /* target->%rax */ \ - "addq $128,%%rsp\n\t" \ - VALGRIND_CALL_NOREDIR_RAX \ - : /*out*/ "=a" (_res) \ - : /*in*/ "a" (&_argvec[0]) \ - : /*trash*/ "cc", "memory", __CALLER_SAVED_REGS \ - ); \ - lval = (__typeof__(lval)) _res; \ - } while (0) - -#define CALL_FN_W_7W(lval, orig, arg1,arg2,arg3,arg4,arg5,arg6, \ - arg7) \ - do { \ - volatile OrigFn _orig = (orig); \ - volatile unsigned long _argvec[8]; \ - volatile unsigned long _res; \ - _argvec[0] = (unsigned long)_orig.nraddr; \ - _argvec[1] = (unsigned long)(arg1); \ - _argvec[2] = (unsigned long)(arg2); \ - _argvec[3] = (unsigned long)(arg3); \ - _argvec[4] = (unsigned long)(arg4); \ - _argvec[5] = (unsigned long)(arg5); \ - _argvec[6] = (unsigned long)(arg6); \ - _argvec[7] = (unsigned long)(arg7); \ - __asm__ volatile( \ - "subq $128,%%rsp\n\t" \ - "pushq 56(%%rax)\n\t" \ - "movq 48(%%rax), %%r9\n\t" \ - "movq 40(%%rax), %%r8\n\t" \ - "movq 32(%%rax), %%rcx\n\t" \ - "movq 24(%%rax), %%rdx\n\t" \ - "movq 16(%%rax), %%rsi\n\t" \ - "movq 8(%%rax), %%rdi\n\t" \ - "movq (%%rax), %%rax\n\t" /* target->%rax */ \ - VALGRIND_CALL_NOREDIR_RAX \ - "addq $8, %%rsp\n" \ - "addq $128,%%rsp\n\t" \ - : /*out*/ "=a" (_res) \ - : /*in*/ "a" (&_argvec[0]) \ - : /*trash*/ "cc", "memory", __CALLER_SAVED_REGS \ - ); \ - lval = (__typeof__(lval)) _res; \ - } while (0) - -#define CALL_FN_W_8W(lval, orig, arg1,arg2,arg3,arg4,arg5,arg6, \ - arg7,arg8) \ - do { \ - volatile OrigFn _orig = (orig); \ - volatile unsigned long _argvec[9]; \ - volatile unsigned long _res; \ - _argvec[0] = (unsigned long)_orig.nraddr; \ - _argvec[1] = (unsigned long)(arg1); \ - _argvec[2] = (unsigned long)(arg2); \ - _argvec[3] = (unsigned long)(arg3); \ - _argvec[4] = (unsigned long)(arg4); \ - _argvec[5] = (unsigned long)(arg5); \ - _argvec[6] = (unsigned long)(arg6); \ - _argvec[7] = (unsigned long)(arg7); \ - _argvec[8] = (unsigned long)(arg8); \ - __asm__ volatile( \ - "subq $128,%%rsp\n\t" \ - "pushq 64(%%rax)\n\t" \ - "pushq 56(%%rax)\n\t" \ - "movq 48(%%rax), %%r9\n\t" \ - "movq 40(%%rax), %%r8\n\t" \ - "movq 32(%%rax), %%rcx\n\t" \ - "movq 24(%%rax), %%rdx\n\t" \ - "movq 16(%%rax), %%rsi\n\t" \ - "movq 8(%%rax), %%rdi\n\t" \ - "movq (%%rax), %%rax\n\t" /* target->%rax */ \ - VALGRIND_CALL_NOREDIR_RAX \ - "addq $16, %%rsp\n" \ - "addq $128,%%rsp\n\t" \ - : /*out*/ "=a" (_res) \ - : /*in*/ "a" (&_argvec[0]) \ - : /*trash*/ "cc", "memory", __CALLER_SAVED_REGS \ - ); \ - lval = (__typeof__(lval)) _res; \ - } while (0) - -#define CALL_FN_W_9W(lval, orig, arg1,arg2,arg3,arg4,arg5,arg6, \ - arg7,arg8,arg9) \ - do { \ - volatile OrigFn _orig = (orig); \ - volatile unsigned long _argvec[10]; \ - volatile unsigned long _res; \ - _argvec[0] = (unsigned long)_orig.nraddr; \ - _argvec[1] = (unsigned long)(arg1); \ - _argvec[2] = (unsigned long)(arg2); \ - _argvec[3] = (unsigned long)(arg3); \ - _argvec[4] = (unsigned long)(arg4); \ - _argvec[5] = (unsigned long)(arg5); \ - _argvec[6] = (unsigned long)(arg6); \ - _argvec[7] = (unsigned long)(arg7); \ - _argvec[8] = (unsigned long)(arg8); \ - _argvec[9] = (unsigned long)(arg9); \ - __asm__ volatile( \ - "subq $128,%%rsp\n\t" \ - "pushq 72(%%rax)\n\t" \ - "pushq 64(%%rax)\n\t" \ - "pushq 56(%%rax)\n\t" \ - "movq 48(%%rax), %%r9\n\t" \ - "movq 40(%%rax), %%r8\n\t" \ - "movq 32(%%rax), %%rcx\n\t" \ - "movq 24(%%rax), %%rdx\n\t" \ - "movq 16(%%rax), %%rsi\n\t" \ - "movq 8(%%rax), %%rdi\n\t" \ - "movq (%%rax), %%rax\n\t" /* target->%rax */ \ - VALGRIND_CALL_NOREDIR_RAX \ - "addq $24, %%rsp\n" \ - "addq $128,%%rsp\n\t" \ - : /*out*/ "=a" (_res) \ - : /*in*/ "a" (&_argvec[0]) \ - : /*trash*/ "cc", "memory", __CALLER_SAVED_REGS \ - ); \ - lval = (__typeof__(lval)) _res; \ - } while (0) - -#define CALL_FN_W_10W(lval, orig, arg1,arg2,arg3,arg4,arg5,arg6, \ - arg7,arg8,arg9,arg10) \ - do { \ - volatile OrigFn _orig = (orig); \ - volatile unsigned long _argvec[11]; \ - volatile unsigned long _res; \ - _argvec[0] = (unsigned long)_orig.nraddr; \ - _argvec[1] = (unsigned long)(arg1); \ - _argvec[2] = (unsigned long)(arg2); \ - _argvec[3] = (unsigned long)(arg3); \ - _argvec[4] = (unsigned long)(arg4); \ - _argvec[5] = (unsigned long)(arg5); \ - _argvec[6] = (unsigned long)(arg6); \ - _argvec[7] = (unsigned long)(arg7); \ - _argvec[8] = (unsigned long)(arg8); \ - _argvec[9] = (unsigned long)(arg9); \ - _argvec[10] = (unsigned long)(arg10); \ - __asm__ volatile( \ - "subq $128,%%rsp\n\t" \ - "pushq 80(%%rax)\n\t" \ - "pushq 72(%%rax)\n\t" \ - "pushq 64(%%rax)\n\t" \ - "pushq 56(%%rax)\n\t" \ - "movq 48(%%rax), %%r9\n\t" \ - "movq 40(%%rax), %%r8\n\t" \ - "movq 32(%%rax), %%rcx\n\t" \ - "movq 24(%%rax), %%rdx\n\t" \ - "movq 16(%%rax), %%rsi\n\t" \ - "movq 8(%%rax), %%rdi\n\t" \ - "movq (%%rax), %%rax\n\t" /* target->%rax */ \ - VALGRIND_CALL_NOREDIR_RAX \ - "addq $32, %%rsp\n" \ - "addq $128,%%rsp\n\t" \ - : /*out*/ "=a" (_res) \ - : /*in*/ "a" (&_argvec[0]) \ - : /*trash*/ "cc", "memory", __CALLER_SAVED_REGS \ - ); \ - lval = (__typeof__(lval)) _res; \ - } while (0) - -#define CALL_FN_W_11W(lval, orig, arg1,arg2,arg3,arg4,arg5,arg6, \ - arg7,arg8,arg9,arg10,arg11) \ - do { \ - volatile OrigFn _orig = (orig); \ - volatile unsigned long _argvec[12]; \ - volatile unsigned long _res; \ - _argvec[0] = (unsigned long)_orig.nraddr; \ - _argvec[1] = (unsigned long)(arg1); \ - _argvec[2] = (unsigned long)(arg2); \ - _argvec[3] = (unsigned long)(arg3); \ - _argvec[4] = (unsigned long)(arg4); \ - _argvec[5] = (unsigned long)(arg5); \ - _argvec[6] = (unsigned long)(arg6); \ - _argvec[7] = (unsigned long)(arg7); \ - _argvec[8] = (unsigned long)(arg8); \ - _argvec[9] = (unsigned long)(arg9); \ - _argvec[10] = (unsigned long)(arg10); \ - _argvec[11] = (unsigned long)(arg11); \ - __asm__ volatile( \ - "subq $128,%%rsp\n\t" \ - "pushq 88(%%rax)\n\t" \ - "pushq 80(%%rax)\n\t" \ - "pushq 72(%%rax)\n\t" \ - "pushq 64(%%rax)\n\t" \ - "pushq 56(%%rax)\n\t" \ - "movq 48(%%rax), %%r9\n\t" \ - "movq 40(%%rax), %%r8\n\t" \ - "movq 32(%%rax), %%rcx\n\t" \ - "movq 24(%%rax), %%rdx\n\t" \ - "movq 16(%%rax), %%rsi\n\t" \ - "movq 8(%%rax), %%rdi\n\t" \ - "movq (%%rax), %%rax\n\t" /* target->%rax */ \ - VALGRIND_CALL_NOREDIR_RAX \ - "addq $40, %%rsp\n" \ - "addq $128,%%rsp\n\t" \ - : /*out*/ "=a" (_res) \ - : /*in*/ "a" (&_argvec[0]) \ - : /*trash*/ "cc", "memory", __CALLER_SAVED_REGS \ - ); \ - lval = (__typeof__(lval)) _res; \ - } while (0) - -#define CALL_FN_W_12W(lval, orig, arg1,arg2,arg3,arg4,arg5,arg6, \ - arg7,arg8,arg9,arg10,arg11,arg12) \ - do { \ - volatile OrigFn _orig = (orig); \ - volatile unsigned long _argvec[13]; \ - volatile unsigned long _res; \ - _argvec[0] = (unsigned long)_orig.nraddr; \ - _argvec[1] = (unsigned long)(arg1); \ - _argvec[2] = (unsigned long)(arg2); \ - _argvec[3] = (unsigned long)(arg3); \ - _argvec[4] = (unsigned long)(arg4); \ - _argvec[5] = (unsigned long)(arg5); \ - _argvec[6] = (unsigned long)(arg6); \ - _argvec[7] = (unsigned long)(arg7); \ - _argvec[8] = (unsigned long)(arg8); \ - _argvec[9] = (unsigned long)(arg9); \ - _argvec[10] = (unsigned long)(arg10); \ - _argvec[11] = (unsigned long)(arg11); \ - _argvec[12] = (unsigned long)(arg12); \ - __asm__ volatile( \ - "subq $128,%%rsp\n\t" \ - "pushq 96(%%rax)\n\t" \ - "pushq 88(%%rax)\n\t" \ - "pushq 80(%%rax)\n\t" \ - "pushq 72(%%rax)\n\t" \ - "pushq 64(%%rax)\n\t" \ - "pushq 56(%%rax)\n\t" \ - "movq 48(%%rax), %%r9\n\t" \ - "movq 40(%%rax), %%r8\n\t" \ - "movq 32(%%rax), %%rcx\n\t" \ - "movq 24(%%rax), %%rdx\n\t" \ - "movq 16(%%rax), %%rsi\n\t" \ - "movq 8(%%rax), %%rdi\n\t" \ - "movq (%%rax), %%rax\n\t" /* target->%rax */ \ - VALGRIND_CALL_NOREDIR_RAX \ - "addq $48, %%rsp\n" \ - "addq $128,%%rsp\n\t" \ - : /*out*/ "=a" (_res) \ - : /*in*/ "a" (&_argvec[0]) \ - : /*trash*/ "cc", "memory", __CALLER_SAVED_REGS \ - ); \ - lval = (__typeof__(lval)) _res; \ - } while (0) - -#endif /* PLAT_amd64_linux */ - -/* ------------------------ ppc32-linux ------------------------ */ - -#if defined(PLAT_ppc32_linux) - -/* This is useful for finding out about the on-stack stuff: - - extern int f9 ( int,int,int,int,int,int,int,int,int ); - extern int f10 ( int,int,int,int,int,int,int,int,int,int ); - extern int f11 ( int,int,int,int,int,int,int,int,int,int,int ); - extern int f12 ( int,int,int,int,int,int,int,int,int,int,int,int ); - - int g9 ( void ) { - return f9(11,22,33,44,55,66,77,88,99); - } - int g10 ( void ) { - return f10(11,22,33,44,55,66,77,88,99,110); - } - int g11 ( void ) { - return f11(11,22,33,44,55,66,77,88,99,110,121); - } - int g12 ( void ) { - return f12(11,22,33,44,55,66,77,88,99,110,121,132); - } -*/ - -/* ARGREGS: r3 r4 r5 r6 r7 r8 r9 r10 (the rest on stack somewhere) */ - -/* These regs are trashed by the hidden call. */ -#define __CALLER_SAVED_REGS \ - "lr", "ctr", "xer", \ - "cr0", "cr1", "cr2", "cr3", "cr4", "cr5", "cr6", "cr7", \ - "r0", "r2", "r3", "r4", "r5", "r6", "r7", "r8", "r9", "r10", \ - "r11", "r12", "r13" - -/* These CALL_FN_ macros assume that on ppc32-linux, - sizeof(unsigned long) == 4. */ - -#define CALL_FN_W_v(lval, orig) \ - do { \ - volatile OrigFn _orig = (orig); \ - volatile unsigned long _argvec[1]; \ - volatile unsigned long _res; \ - _argvec[0] = (unsigned long)_orig.nraddr; \ - __asm__ volatile( \ - "mr 11,%1\n\t" \ - "lwz 11,0(11)\n\t" /* target->r11 */ \ - VALGRIND_BRANCH_AND_LINK_TO_NOREDIR_R11 \ - "mr %0,3" \ - : /*out*/ "=r" (_res) \ - : /*in*/ "r" (&_argvec[0]) \ - : /*trash*/ "cc", "memory", __CALLER_SAVED_REGS \ - ); \ - lval = (__typeof__(lval)) _res; \ - } while (0) - -#define CALL_FN_W_W(lval, orig, arg1) \ - do { \ - volatile OrigFn _orig = (orig); \ - volatile unsigned long _argvec[2]; \ - volatile unsigned long _res; \ - _argvec[0] = (unsigned long)_orig.nraddr; \ - _argvec[1] = (unsigned long)arg1; \ - __asm__ volatile( \ - "mr 11,%1\n\t" \ - "lwz 3,4(11)\n\t" /* arg1->r3 */ \ - "lwz 11,0(11)\n\t" /* target->r11 */ \ - VALGRIND_BRANCH_AND_LINK_TO_NOREDIR_R11 \ - "mr %0,3" \ - : /*out*/ "=r" (_res) \ - : /*in*/ "r" (&_argvec[0]) \ - : /*trash*/ "cc", "memory", __CALLER_SAVED_REGS \ - ); \ - lval = (__typeof__(lval)) _res; \ - } while (0) - -#define CALL_FN_W_WW(lval, orig, arg1,arg2) \ - do { \ - volatile OrigFn _orig = (orig); \ - volatile unsigned long _argvec[3]; \ - volatile unsigned long _res; \ - _argvec[0] = (unsigned long)_orig.nraddr; \ - _argvec[1] = (unsigned long)arg1; \ - _argvec[2] = (unsigned long)arg2; \ - __asm__ volatile( \ - "mr 11,%1\n\t" \ - "lwz 3,4(11)\n\t" /* arg1->r3 */ \ - "lwz 4,8(11)\n\t" \ - "lwz 11,0(11)\n\t" /* target->r11 */ \ - VALGRIND_BRANCH_AND_LINK_TO_NOREDIR_R11 \ - "mr %0,3" \ - : /*out*/ "=r" (_res) \ - : /*in*/ "r" (&_argvec[0]) \ - : /*trash*/ "cc", "memory", __CALLER_SAVED_REGS \ - ); \ - lval = (__typeof__(lval)) _res; \ - } while (0) - -#define CALL_FN_W_WWW(lval, orig, arg1,arg2,arg3) \ - do { \ - volatile OrigFn _orig = (orig); \ - volatile unsigned long _argvec[4]; \ - volatile unsigned long _res; \ - _argvec[0] = (unsigned long)_orig.nraddr; \ - _argvec[1] = (unsigned long)arg1; \ - _argvec[2] = (unsigned long)arg2; \ - _argvec[3] = (unsigned long)arg3; \ - __asm__ volatile( \ - "mr 11,%1\n\t" \ - "lwz 3,4(11)\n\t" /* arg1->r3 */ \ - "lwz 4,8(11)\n\t" \ - "lwz 5,12(11)\n\t" \ - "lwz 11,0(11)\n\t" /* target->r11 */ \ - VALGRIND_BRANCH_AND_LINK_TO_NOREDIR_R11 \ - "mr %0,3" \ - : /*out*/ "=r" (_res) \ - : /*in*/ "r" (&_argvec[0]) \ - : /*trash*/ "cc", "memory", __CALLER_SAVED_REGS \ - ); \ - lval = (__typeof__(lval)) _res; \ - } while (0) - -#define CALL_FN_W_WWWW(lval, orig, arg1,arg2,arg3,arg4) \ - do { \ - volatile OrigFn _orig = (orig); \ - volatile unsigned long _argvec[5]; \ - volatile unsigned long _res; \ - _argvec[0] = (unsigned long)_orig.nraddr; \ - _argvec[1] = (unsigned long)arg1; \ - _argvec[2] = (unsigned long)arg2; \ - _argvec[3] = (unsigned long)arg3; \ - _argvec[4] = (unsigned long)arg4; \ - __asm__ volatile( \ - "mr 11,%1\n\t" \ - "lwz 3,4(11)\n\t" /* arg1->r3 */ \ - "lwz 4,8(11)\n\t" \ - "lwz 5,12(11)\n\t" \ - "lwz 6,16(11)\n\t" /* arg4->r6 */ \ - "lwz 11,0(11)\n\t" /* target->r11 */ \ - VALGRIND_BRANCH_AND_LINK_TO_NOREDIR_R11 \ - "mr %0,3" \ - : /*out*/ "=r" (_res) \ - : /*in*/ "r" (&_argvec[0]) \ - : /*trash*/ "cc", "memory", __CALLER_SAVED_REGS \ - ); \ - lval = (__typeof__(lval)) _res; \ - } while (0) - -#define CALL_FN_W_5W(lval, orig, arg1,arg2,arg3,arg4,arg5) \ - do { \ - volatile OrigFn _orig = (orig); \ - volatile unsigned long _argvec[6]; \ - volatile unsigned long _res; \ - _argvec[0] = (unsigned long)_orig.nraddr; \ - _argvec[1] = (unsigned long)arg1; \ - _argvec[2] = (unsigned long)arg2; \ - _argvec[3] = (unsigned long)arg3; \ - _argvec[4] = (unsigned long)arg4; \ - _argvec[5] = (unsigned long)arg5; \ - __asm__ volatile( \ - "mr 11,%1\n\t" \ - "lwz 3,4(11)\n\t" /* arg1->r3 */ \ - "lwz 4,8(11)\n\t" \ - "lwz 5,12(11)\n\t" \ - "lwz 6,16(11)\n\t" /* arg4->r6 */ \ - "lwz 7,20(11)\n\t" \ - "lwz 11,0(11)\n\t" /* target->r11 */ \ - VALGRIND_BRANCH_AND_LINK_TO_NOREDIR_R11 \ - "mr %0,3" \ - : /*out*/ "=r" (_res) \ - : /*in*/ "r" (&_argvec[0]) \ - : /*trash*/ "cc", "memory", __CALLER_SAVED_REGS \ - ); \ - lval = (__typeof__(lval)) _res; \ - } while (0) - -#define CALL_FN_W_6W(lval, orig, arg1,arg2,arg3,arg4,arg5,arg6) \ - do { \ - volatile OrigFn _orig = (orig); \ - volatile unsigned long _argvec[7]; \ - volatile unsigned long _res; \ - _argvec[0] = (unsigned long)_orig.nraddr; \ - _argvec[1] = (unsigned long)arg1; \ - _argvec[2] = (unsigned long)arg2; \ - _argvec[3] = (unsigned long)arg3; \ - _argvec[4] = (unsigned long)arg4; \ - _argvec[5] = (unsigned long)arg5; \ - _argvec[6] = (unsigned long)arg6; \ - __asm__ volatile( \ - "mr 11,%1\n\t" \ - "lwz 3,4(11)\n\t" /* arg1->r3 */ \ - "lwz 4,8(11)\n\t" \ - "lwz 5,12(11)\n\t" \ - "lwz 6,16(11)\n\t" /* arg4->r6 */ \ - "lwz 7,20(11)\n\t" \ - "lwz 8,24(11)\n\t" \ - "lwz 11,0(11)\n\t" /* target->r11 */ \ - VALGRIND_BRANCH_AND_LINK_TO_NOREDIR_R11 \ - "mr %0,3" \ - : /*out*/ "=r" (_res) \ - : /*in*/ "r" (&_argvec[0]) \ - : /*trash*/ "cc", "memory", __CALLER_SAVED_REGS \ - ); \ - lval = (__typeof__(lval)) _res; \ - } while (0) - -#define CALL_FN_W_7W(lval, orig, arg1,arg2,arg3,arg4,arg5,arg6, \ - arg7) \ - do { \ - volatile OrigFn _orig = (orig); \ - volatile unsigned long _argvec[8]; \ - volatile unsigned long _res; \ - _argvec[0] = (unsigned long)_orig.nraddr; \ - _argvec[1] = (unsigned long)arg1; \ - _argvec[2] = (unsigned long)arg2; \ - _argvec[3] = (unsigned long)arg3; \ - _argvec[4] = (unsigned long)arg4; \ - _argvec[5] = (unsigned long)arg5; \ - _argvec[6] = (unsigned long)arg6; \ - _argvec[7] = (unsigned long)arg7; \ - __asm__ volatile( \ - "mr 11,%1\n\t" \ - "lwz 3,4(11)\n\t" /* arg1->r3 */ \ - "lwz 4,8(11)\n\t" \ - "lwz 5,12(11)\n\t" \ - "lwz 6,16(11)\n\t" /* arg4->r6 */ \ - "lwz 7,20(11)\n\t" \ - "lwz 8,24(11)\n\t" \ - "lwz 9,28(11)\n\t" \ - "lwz 11,0(11)\n\t" /* target->r11 */ \ - VALGRIND_BRANCH_AND_LINK_TO_NOREDIR_R11 \ - "mr %0,3" \ - : /*out*/ "=r" (_res) \ - : /*in*/ "r" (&_argvec[0]) \ - : /*trash*/ "cc", "memory", __CALLER_SAVED_REGS \ - ); \ - lval = (__typeof__(lval)) _res; \ - } while (0) - -#define CALL_FN_W_8W(lval, orig, arg1,arg2,arg3,arg4,arg5,arg6, \ - arg7,arg8) \ - do { \ - volatile OrigFn _orig = (orig); \ - volatile unsigned long _argvec[9]; \ - volatile unsigned long _res; \ - _argvec[0] = (unsigned long)_orig.nraddr; \ - _argvec[1] = (unsigned long)arg1; \ - _argvec[2] = (unsigned long)arg2; \ - _argvec[3] = (unsigned long)arg3; \ - _argvec[4] = (unsigned long)arg4; \ - _argvec[5] = (unsigned long)arg5; \ - _argvec[6] = (unsigned long)arg6; \ - _argvec[7] = (unsigned long)arg7; \ - _argvec[8] = (unsigned long)arg8; \ - __asm__ volatile( \ - "mr 11,%1\n\t" \ - "lwz 3,4(11)\n\t" /* arg1->r3 */ \ - "lwz 4,8(11)\n\t" \ - "lwz 5,12(11)\n\t" \ - "lwz 6,16(11)\n\t" /* arg4->r6 */ \ - "lwz 7,20(11)\n\t" \ - "lwz 8,24(11)\n\t" \ - "lwz 9,28(11)\n\t" \ - "lwz 10,32(11)\n\t" /* arg8->r10 */ \ - "lwz 11,0(11)\n\t" /* target->r11 */ \ - VALGRIND_BRANCH_AND_LINK_TO_NOREDIR_R11 \ - "mr %0,3" \ - : /*out*/ "=r" (_res) \ - : /*in*/ "r" (&_argvec[0]) \ - : /*trash*/ "cc", "memory", __CALLER_SAVED_REGS \ - ); \ - lval = (__typeof__(lval)) _res; \ - } while (0) - -#define CALL_FN_W_9W(lval, orig, arg1,arg2,arg3,arg4,arg5,arg6, \ - arg7,arg8,arg9) \ - do { \ - volatile OrigFn _orig = (orig); \ - volatile unsigned long _argvec[10]; \ - volatile unsigned long _res; \ - _argvec[0] = (unsigned long)_orig.nraddr; \ - _argvec[1] = (unsigned long)arg1; \ - _argvec[2] = (unsigned long)arg2; \ - _argvec[3] = (unsigned long)arg3; \ - _argvec[4] = (unsigned long)arg4; \ - _argvec[5] = (unsigned long)arg5; \ - _argvec[6] = (unsigned long)arg6; \ - _argvec[7] = (unsigned long)arg7; \ - _argvec[8] = (unsigned long)arg8; \ - _argvec[9] = (unsigned long)arg9; \ - __asm__ volatile( \ - "mr 11,%1\n\t" \ - "addi 1,1,-16\n\t" \ - /* arg9 */ \ - "lwz 3,36(11)\n\t" \ - "stw 3,8(1)\n\t" \ - /* args1-8 */ \ - "lwz 3,4(11)\n\t" /* arg1->r3 */ \ - "lwz 4,8(11)\n\t" \ - "lwz 5,12(11)\n\t" \ - "lwz 6,16(11)\n\t" /* arg4->r6 */ \ - "lwz 7,20(11)\n\t" \ - "lwz 8,24(11)\n\t" \ - "lwz 9,28(11)\n\t" \ - "lwz 10,32(11)\n\t" /* arg8->r10 */ \ - "lwz 11,0(11)\n\t" /* target->r11 */ \ - VALGRIND_BRANCH_AND_LINK_TO_NOREDIR_R11 \ - "addi 1,1,16\n\t" \ - "mr %0,3" \ - : /*out*/ "=r" (_res) \ - : /*in*/ "r" (&_argvec[0]) \ - : /*trash*/ "cc", "memory", __CALLER_SAVED_REGS \ - ); \ - lval = (__typeof__(lval)) _res; \ - } while (0) - -#define CALL_FN_W_10W(lval, orig, arg1,arg2,arg3,arg4,arg5,arg6, \ - arg7,arg8,arg9,arg10) \ - do { \ - volatile OrigFn _orig = (orig); \ - volatile unsigned long _argvec[11]; \ - volatile unsigned long _res; \ - _argvec[0] = (unsigned long)_orig.nraddr; \ - _argvec[1] = (unsigned long)arg1; \ - _argvec[2] = (unsigned long)arg2; \ - _argvec[3] = (unsigned long)arg3; \ - _argvec[4] = (unsigned long)arg4; \ - _argvec[5] = (unsigned long)arg5; \ - _argvec[6] = (unsigned long)arg6; \ - _argvec[7] = (unsigned long)arg7; \ - _argvec[8] = (unsigned long)arg8; \ - _argvec[9] = (unsigned long)arg9; \ - _argvec[10] = (unsigned long)arg10; \ - __asm__ volatile( \ - "mr 11,%1\n\t" \ - "addi 1,1,-16\n\t" \ - /* arg10 */ \ - "lwz 3,40(11)\n\t" \ - "stw 3,12(1)\n\t" \ - /* arg9 */ \ - "lwz 3,36(11)\n\t" \ - "stw 3,8(1)\n\t" \ - /* args1-8 */ \ - "lwz 3,4(11)\n\t" /* arg1->r3 */ \ - "lwz 4,8(11)\n\t" \ - "lwz 5,12(11)\n\t" \ - "lwz 6,16(11)\n\t" /* arg4->r6 */ \ - "lwz 7,20(11)\n\t" \ - "lwz 8,24(11)\n\t" \ - "lwz 9,28(11)\n\t" \ - "lwz 10,32(11)\n\t" /* arg8->r10 */ \ - "lwz 11,0(11)\n\t" /* target->r11 */ \ - VALGRIND_BRANCH_AND_LINK_TO_NOREDIR_R11 \ - "addi 1,1,16\n\t" \ - "mr %0,3" \ - : /*out*/ "=r" (_res) \ - : /*in*/ "r" (&_argvec[0]) \ - : /*trash*/ "cc", "memory", __CALLER_SAVED_REGS \ - ); \ - lval = (__typeof__(lval)) _res; \ - } while (0) - -#define CALL_FN_W_11W(lval, orig, arg1,arg2,arg3,arg4,arg5,arg6, \ - arg7,arg8,arg9,arg10,arg11) \ - do { \ - volatile OrigFn _orig = (orig); \ - volatile unsigned long _argvec[12]; \ - volatile unsigned long _res; \ - _argvec[0] = (unsigned long)_orig.nraddr; \ - _argvec[1] = (unsigned long)arg1; \ - _argvec[2] = (unsigned long)arg2; \ - _argvec[3] = (unsigned long)arg3; \ - _argvec[4] = (unsigned long)arg4; \ - _argvec[5] = (unsigned long)arg5; \ - _argvec[6] = (unsigned long)arg6; \ - _argvec[7] = (unsigned long)arg7; \ - _argvec[8] = (unsigned long)arg8; \ - _argvec[9] = (unsigned long)arg9; \ - _argvec[10] = (unsigned long)arg10; \ - _argvec[11] = (unsigned long)arg11; \ - __asm__ volatile( \ - "mr 11,%1\n\t" \ - "addi 1,1,-32\n\t" \ - /* arg11 */ \ - "lwz 3,44(11)\n\t" \ - "stw 3,16(1)\n\t" \ - /* arg10 */ \ - "lwz 3,40(11)\n\t" \ - "stw 3,12(1)\n\t" \ - /* arg9 */ \ - "lwz 3,36(11)\n\t" \ - "stw 3,8(1)\n\t" \ - /* args1-8 */ \ - "lwz 3,4(11)\n\t" /* arg1->r3 */ \ - "lwz 4,8(11)\n\t" \ - "lwz 5,12(11)\n\t" \ - "lwz 6,16(11)\n\t" /* arg4->r6 */ \ - "lwz 7,20(11)\n\t" \ - "lwz 8,24(11)\n\t" \ - "lwz 9,28(11)\n\t" \ - "lwz 10,32(11)\n\t" /* arg8->r10 */ \ - "lwz 11,0(11)\n\t" /* target->r11 */ \ - VALGRIND_BRANCH_AND_LINK_TO_NOREDIR_R11 \ - "addi 1,1,32\n\t" \ - "mr %0,3" \ - : /*out*/ "=r" (_res) \ - : /*in*/ "r" (&_argvec[0]) \ - : /*trash*/ "cc", "memory", __CALLER_SAVED_REGS \ - ); \ - lval = (__typeof__(lval)) _res; \ - } while (0) - -#define CALL_FN_W_12W(lval, orig, arg1,arg2,arg3,arg4,arg5,arg6, \ - arg7,arg8,arg9,arg10,arg11,arg12) \ - do { \ - volatile OrigFn _orig = (orig); \ - volatile unsigned long _argvec[13]; \ - volatile unsigned long _res; \ - _argvec[0] = (unsigned long)_orig.nraddr; \ - _argvec[1] = (unsigned long)arg1; \ - _argvec[2] = (unsigned long)arg2; \ - _argvec[3] = (unsigned long)arg3; \ - _argvec[4] = (unsigned long)arg4; \ - _argvec[5] = (unsigned long)arg5; \ - _argvec[6] = (unsigned long)arg6; \ - _argvec[7] = (unsigned long)arg7; \ - _argvec[8] = (unsigned long)arg8; \ - _argvec[9] = (unsigned long)arg9; \ - _argvec[10] = (unsigned long)arg10; \ - _argvec[11] = (unsigned long)arg11; \ - _argvec[12] = (unsigned long)arg12; \ - __asm__ volatile( \ - "mr 11,%1\n\t" \ - "addi 1,1,-32\n\t" \ - /* arg12 */ \ - "lwz 3,48(11)\n\t" \ - "stw 3,20(1)\n\t" \ - /* arg11 */ \ - "lwz 3,44(11)\n\t" \ - "stw 3,16(1)\n\t" \ - /* arg10 */ \ - "lwz 3,40(11)\n\t" \ - "stw 3,12(1)\n\t" \ - /* arg9 */ \ - "lwz 3,36(11)\n\t" \ - "stw 3,8(1)\n\t" \ - /* args1-8 */ \ - "lwz 3,4(11)\n\t" /* arg1->r3 */ \ - "lwz 4,8(11)\n\t" \ - "lwz 5,12(11)\n\t" \ - "lwz 6,16(11)\n\t" /* arg4->r6 */ \ - "lwz 7,20(11)\n\t" \ - "lwz 8,24(11)\n\t" \ - "lwz 9,28(11)\n\t" \ - "lwz 10,32(11)\n\t" /* arg8->r10 */ \ - "lwz 11,0(11)\n\t" /* target->r11 */ \ - VALGRIND_BRANCH_AND_LINK_TO_NOREDIR_R11 \ - "addi 1,1,32\n\t" \ - "mr %0,3" \ - : /*out*/ "=r" (_res) \ - : /*in*/ "r" (&_argvec[0]) \ - : /*trash*/ "cc", "memory", __CALLER_SAVED_REGS \ - ); \ - lval = (__typeof__(lval)) _res; \ - } while (0) - -#endif /* PLAT_ppc32_linux */ - -/* ------------------------ ppc64-linux ------------------------ */ - -#if defined(PLAT_ppc64_linux) - -/* ARGREGS: r3 r4 r5 r6 r7 r8 r9 r10 (the rest on stack somewhere) */ - -/* These regs are trashed by the hidden call. */ -#define __CALLER_SAVED_REGS \ - "lr", "ctr", "xer", \ - "cr0", "cr1", "cr2", "cr3", "cr4", "cr5", "cr6", "cr7", \ - "r0", "r2", "r3", "r4", "r5", "r6", "r7", "r8", "r9", "r10", \ - "r11", "r12", "r13" - -/* These CALL_FN_ macros assume that on ppc64-linux, sizeof(unsigned - long) == 8. */ - -#define CALL_FN_W_v(lval, orig) \ - do { \ - volatile OrigFn _orig = (orig); \ - volatile unsigned long _argvec[3+0]; \ - volatile unsigned long _res; \ - /* _argvec[0] holds current r2 across the call */ \ - _argvec[1] = (unsigned long)_orig.r2; \ - _argvec[2] = (unsigned long)_orig.nraddr; \ - __asm__ volatile( \ - "mr 11,%1\n\t" \ - "std 2,-16(11)\n\t" /* save tocptr */ \ - "ld 2,-8(11)\n\t" /* use nraddr's tocptr */ \ - "ld 11, 0(11)\n\t" /* target->r11 */ \ - VALGRIND_BRANCH_AND_LINK_TO_NOREDIR_R11 \ - "mr 11,%1\n\t" \ - "mr %0,3\n\t" \ - "ld 2,-16(11)" /* restore tocptr */ \ - : /*out*/ "=r" (_res) \ - : /*in*/ "r" (&_argvec[2]) \ - : /*trash*/ "cc", "memory", __CALLER_SAVED_REGS \ - ); \ - lval = (__typeof__(lval)) _res; \ - } while (0) - -#define CALL_FN_W_W(lval, orig, arg1) \ - do { \ - volatile OrigFn _orig = (orig); \ - volatile unsigned long _argvec[3+1]; \ - volatile unsigned long _res; \ - /* _argvec[0] holds current r2 across the call */ \ - _argvec[1] = (unsigned long)_orig.r2; \ - _argvec[2] = (unsigned long)_orig.nraddr; \ - _argvec[2+1] = (unsigned long)arg1; \ - __asm__ volatile( \ - "mr 11,%1\n\t" \ - "std 2,-16(11)\n\t" /* save tocptr */ \ - "ld 2,-8(11)\n\t" /* use nraddr's tocptr */ \ - "ld 3, 8(11)\n\t" /* arg1->r3 */ \ - "ld 11, 0(11)\n\t" /* target->r11 */ \ - VALGRIND_BRANCH_AND_LINK_TO_NOREDIR_R11 \ - "mr 11,%1\n\t" \ - "mr %0,3\n\t" \ - "ld 2,-16(11)" /* restore tocptr */ \ - : /*out*/ "=r" (_res) \ - : /*in*/ "r" (&_argvec[2]) \ - : /*trash*/ "cc", "memory", __CALLER_SAVED_REGS \ - ); \ - lval = (__typeof__(lval)) _res; \ - } while (0) - -#define CALL_FN_W_WW(lval, orig, arg1,arg2) \ - do { \ - volatile OrigFn _orig = (orig); \ - volatile unsigned long _argvec[3+2]; \ - volatile unsigned long _res; \ - /* _argvec[0] holds current r2 across the call */ \ - _argvec[1] = (unsigned long)_orig.r2; \ - _argvec[2] = (unsigned long)_orig.nraddr; \ - _argvec[2+1] = (unsigned long)arg1; \ - _argvec[2+2] = (unsigned long)arg2; \ - __asm__ volatile( \ - "mr 11,%1\n\t" \ - "std 2,-16(11)\n\t" /* save tocptr */ \ - "ld 2,-8(11)\n\t" /* use nraddr's tocptr */ \ - "ld 3, 8(11)\n\t" /* arg1->r3 */ \ - "ld 4, 16(11)\n\t" /* arg2->r4 */ \ - "ld 11, 0(11)\n\t" /* target->r11 */ \ - VALGRIND_BRANCH_AND_LINK_TO_NOREDIR_R11 \ - "mr 11,%1\n\t" \ - "mr %0,3\n\t" \ - "ld 2,-16(11)" /* restore tocptr */ \ - : /*out*/ "=r" (_res) \ - : /*in*/ "r" (&_argvec[2]) \ - : /*trash*/ "cc", "memory", __CALLER_SAVED_REGS \ - ); \ - lval = (__typeof__(lval)) _res; \ - } while (0) - -#define CALL_FN_W_WWW(lval, orig, arg1,arg2,arg3) \ - do { \ - volatile OrigFn _orig = (orig); \ - volatile unsigned long _argvec[3+3]; \ - volatile unsigned long _res; \ - /* _argvec[0] holds current r2 across the call */ \ - _argvec[1] = (unsigned long)_orig.r2; \ - _argvec[2] = (unsigned long)_orig.nraddr; \ - _argvec[2+1] = (unsigned long)arg1; \ - _argvec[2+2] = (unsigned long)arg2; \ - _argvec[2+3] = (unsigned long)arg3; \ - __asm__ volatile( \ - "mr 11,%1\n\t" \ - "std 2,-16(11)\n\t" /* save tocptr */ \ - "ld 2,-8(11)\n\t" /* use nraddr's tocptr */ \ - "ld 3, 8(11)\n\t" /* arg1->r3 */ \ - "ld 4, 16(11)\n\t" /* arg2->r4 */ \ - "ld 5, 24(11)\n\t" /* arg3->r5 */ \ - "ld 11, 0(11)\n\t" /* target->r11 */ \ - VALGRIND_BRANCH_AND_LINK_TO_NOREDIR_R11 \ - "mr 11,%1\n\t" \ - "mr %0,3\n\t" \ - "ld 2,-16(11)" /* restore tocptr */ \ - : /*out*/ "=r" (_res) \ - : /*in*/ "r" (&_argvec[2]) \ - : /*trash*/ "cc", "memory", __CALLER_SAVED_REGS \ - ); \ - lval = (__typeof__(lval)) _res; \ - } while (0) - -#define CALL_FN_W_WWWW(lval, orig, arg1,arg2,arg3,arg4) \ - do { \ - volatile OrigFn _orig = (orig); \ - volatile unsigned long _argvec[3+4]; \ - volatile unsigned long _res; \ - /* _argvec[0] holds current r2 across the call */ \ - _argvec[1] = (unsigned long)_orig.r2; \ - _argvec[2] = (unsigned long)_orig.nraddr; \ - _argvec[2+1] = (unsigned long)arg1; \ - _argvec[2+2] = (unsigned long)arg2; \ - _argvec[2+3] = (unsigned long)arg3; \ - _argvec[2+4] = (unsigned long)arg4; \ - __asm__ volatile( \ - "mr 11,%1\n\t" \ - "std 2,-16(11)\n\t" /* save tocptr */ \ - "ld 2,-8(11)\n\t" /* use nraddr's tocptr */ \ - "ld 3, 8(11)\n\t" /* arg1->r3 */ \ - "ld 4, 16(11)\n\t" /* arg2->r4 */ \ - "ld 5, 24(11)\n\t" /* arg3->r5 */ \ - "ld 6, 32(11)\n\t" /* arg4->r6 */ \ - "ld 11, 0(11)\n\t" /* target->r11 */ \ - VALGRIND_BRANCH_AND_LINK_TO_NOREDIR_R11 \ - "mr 11,%1\n\t" \ - "mr %0,3\n\t" \ - "ld 2,-16(11)" /* restore tocptr */ \ - : /*out*/ "=r" (_res) \ - : /*in*/ "r" (&_argvec[2]) \ - : /*trash*/ "cc", "memory", __CALLER_SAVED_REGS \ - ); \ - lval = (__typeof__(lval)) _res; \ - } while (0) - -#define CALL_FN_W_5W(lval, orig, arg1,arg2,arg3,arg4,arg5) \ - do { \ - volatile OrigFn _orig = (orig); \ - volatile unsigned long _argvec[3+5]; \ - volatile unsigned long _res; \ - /* _argvec[0] holds current r2 across the call */ \ - _argvec[1] = (unsigned long)_orig.r2; \ - _argvec[2] = (unsigned long)_orig.nraddr; \ - _argvec[2+1] = (unsigned long)arg1; \ - _argvec[2+2] = (unsigned long)arg2; \ - _argvec[2+3] = (unsigned long)arg3; \ - _argvec[2+4] = (unsigned long)arg4; \ - _argvec[2+5] = (unsigned long)arg5; \ - __asm__ volatile( \ - "mr 11,%1\n\t" \ - "std 2,-16(11)\n\t" /* save tocptr */ \ - "ld 2,-8(11)\n\t" /* use nraddr's tocptr */ \ - "ld 3, 8(11)\n\t" /* arg1->r3 */ \ - "ld 4, 16(11)\n\t" /* arg2->r4 */ \ - "ld 5, 24(11)\n\t" /* arg3->r5 */ \ - "ld 6, 32(11)\n\t" /* arg4->r6 */ \ - "ld 7, 40(11)\n\t" /* arg5->r7 */ \ - "ld 11, 0(11)\n\t" /* target->r11 */ \ - VALGRIND_BRANCH_AND_LINK_TO_NOREDIR_R11 \ - "mr 11,%1\n\t" \ - "mr %0,3\n\t" \ - "ld 2,-16(11)" /* restore tocptr */ \ - : /*out*/ "=r" (_res) \ - : /*in*/ "r" (&_argvec[2]) \ - : /*trash*/ "cc", "memory", __CALLER_SAVED_REGS \ - ); \ - lval = (__typeof__(lval)) _res; \ - } while (0) - -#define CALL_FN_W_6W(lval, orig, arg1,arg2,arg3,arg4,arg5,arg6) \ - do { \ - volatile OrigFn _orig = (orig); \ - volatile unsigned long _argvec[3+6]; \ - volatile unsigned long _res; \ - /* _argvec[0] holds current r2 across the call */ \ - _argvec[1] = (unsigned long)_orig.r2; \ - _argvec[2] = (unsigned long)_orig.nraddr; \ - _argvec[2+1] = (unsigned long)arg1; \ - _argvec[2+2] = (unsigned long)arg2; \ - _argvec[2+3] = (unsigned long)arg3; \ - _argvec[2+4] = (unsigned long)arg4; \ - _argvec[2+5] = (unsigned long)arg5; \ - _argvec[2+6] = (unsigned long)arg6; \ - __asm__ volatile( \ - "mr 11,%1\n\t" \ - "std 2,-16(11)\n\t" /* save tocptr */ \ - "ld 2,-8(11)\n\t" /* use nraddr's tocptr */ \ - "ld 3, 8(11)\n\t" /* arg1->r3 */ \ - "ld 4, 16(11)\n\t" /* arg2->r4 */ \ - "ld 5, 24(11)\n\t" /* arg3->r5 */ \ - "ld 6, 32(11)\n\t" /* arg4->r6 */ \ - "ld 7, 40(11)\n\t" /* arg5->r7 */ \ - "ld 8, 48(11)\n\t" /* arg6->r8 */ \ - "ld 11, 0(11)\n\t" /* target->r11 */ \ - VALGRIND_BRANCH_AND_LINK_TO_NOREDIR_R11 \ - "mr 11,%1\n\t" \ - "mr %0,3\n\t" \ - "ld 2,-16(11)" /* restore tocptr */ \ - : /*out*/ "=r" (_res) \ - : /*in*/ "r" (&_argvec[2]) \ - : /*trash*/ "cc", "memory", __CALLER_SAVED_REGS \ - ); \ - lval = (__typeof__(lval)) _res; \ - } while (0) - -#define CALL_FN_W_7W(lval, orig, arg1,arg2,arg3,arg4,arg5,arg6, \ - arg7) \ - do { \ - volatile OrigFn _orig = (orig); \ - volatile unsigned long _argvec[3+7]; \ - volatile unsigned long _res; \ - /* _argvec[0] holds current r2 across the call */ \ - _argvec[1] = (unsigned long)_orig.r2; \ - _argvec[2] = (unsigned long)_orig.nraddr; \ - _argvec[2+1] = (unsigned long)arg1; \ - _argvec[2+2] = (unsigned long)arg2; \ - _argvec[2+3] = (unsigned long)arg3; \ - _argvec[2+4] = (unsigned long)arg4; \ - _argvec[2+5] = (unsigned long)arg5; \ - _argvec[2+6] = (unsigned long)arg6; \ - _argvec[2+7] = (unsigned long)arg7; \ - __asm__ volatile( \ - "mr 11,%1\n\t" \ - "std 2,-16(11)\n\t" /* save tocptr */ \ - "ld 2,-8(11)\n\t" /* use nraddr's tocptr */ \ - "ld 3, 8(11)\n\t" /* arg1->r3 */ \ - "ld 4, 16(11)\n\t" /* arg2->r4 */ \ - "ld 5, 24(11)\n\t" /* arg3->r5 */ \ - "ld 6, 32(11)\n\t" /* arg4->r6 */ \ - "ld 7, 40(11)\n\t" /* arg5->r7 */ \ - "ld 8, 48(11)\n\t" /* arg6->r8 */ \ - "ld 9, 56(11)\n\t" /* arg7->r9 */ \ - "ld 11, 0(11)\n\t" /* target->r11 */ \ - VALGRIND_BRANCH_AND_LINK_TO_NOREDIR_R11 \ - "mr 11,%1\n\t" \ - "mr %0,3\n\t" \ - "ld 2,-16(11)" /* restore tocptr */ \ - : /*out*/ "=r" (_res) \ - : /*in*/ "r" (&_argvec[2]) \ - : /*trash*/ "cc", "memory", __CALLER_SAVED_REGS \ - ); \ - lval = (__typeof__(lval)) _res; \ - } while (0) - -#define CALL_FN_W_8W(lval, orig, arg1,arg2,arg3,arg4,arg5,arg6, \ - arg7,arg8) \ - do { \ - volatile OrigFn _orig = (orig); \ - volatile unsigned long _argvec[3+8]; \ - volatile unsigned long _res; \ - /* _argvec[0] holds current r2 across the call */ \ - _argvec[1] = (unsigned long)_orig.r2; \ - _argvec[2] = (unsigned long)_orig.nraddr; \ - _argvec[2+1] = (unsigned long)arg1; \ - _argvec[2+2] = (unsigned long)arg2; \ - _argvec[2+3] = (unsigned long)arg3; \ - _argvec[2+4] = (unsigned long)arg4; \ - _argvec[2+5] = (unsigned long)arg5; \ - _argvec[2+6] = (unsigned long)arg6; \ - _argvec[2+7] = (unsigned long)arg7; \ - _argvec[2+8] = (unsigned long)arg8; \ - __asm__ volatile( \ - "mr 11,%1\n\t" \ - "std 2,-16(11)\n\t" /* save tocptr */ \ - "ld 2,-8(11)\n\t" /* use nraddr's tocptr */ \ - "ld 3, 8(11)\n\t" /* arg1->r3 */ \ - "ld 4, 16(11)\n\t" /* arg2->r4 */ \ - "ld 5, 24(11)\n\t" /* arg3->r5 */ \ - "ld 6, 32(11)\n\t" /* arg4->r6 */ \ - "ld 7, 40(11)\n\t" /* arg5->r7 */ \ - "ld 8, 48(11)\n\t" /* arg6->r8 */ \ - "ld 9, 56(11)\n\t" /* arg7->r9 */ \ - "ld 10, 64(11)\n\t" /* arg8->r10 */ \ - "ld 11, 0(11)\n\t" /* target->r11 */ \ - VALGRIND_BRANCH_AND_LINK_TO_NOREDIR_R11 \ - "mr 11,%1\n\t" \ - "mr %0,3\n\t" \ - "ld 2,-16(11)" /* restore tocptr */ \ - : /*out*/ "=r" (_res) \ - : /*in*/ "r" (&_argvec[2]) \ - : /*trash*/ "cc", "memory", __CALLER_SAVED_REGS \ - ); \ - lval = (__typeof__(lval)) _res; \ - } while (0) - -#define CALL_FN_W_9W(lval, orig, arg1,arg2,arg3,arg4,arg5,arg6, \ - arg7,arg8,arg9) \ - do { \ - volatile OrigFn _orig = (orig); \ - volatile unsigned long _argvec[3+9]; \ - volatile unsigned long _res; \ - /* _argvec[0] holds current r2 across the call */ \ - _argvec[1] = (unsigned long)_orig.r2; \ - _argvec[2] = (unsigned long)_orig.nraddr; \ - _argvec[2+1] = (unsigned long)arg1; \ - _argvec[2+2] = (unsigned long)arg2; \ - _argvec[2+3] = (unsigned long)arg3; \ - _argvec[2+4] = (unsigned long)arg4; \ - _argvec[2+5] = (unsigned long)arg5; \ - _argvec[2+6] = (unsigned long)arg6; \ - _argvec[2+7] = (unsigned long)arg7; \ - _argvec[2+8] = (unsigned long)arg8; \ - _argvec[2+9] = (unsigned long)arg9; \ - __asm__ volatile( \ - "mr 11,%1\n\t" \ - "std 2,-16(11)\n\t" /* save tocptr */ \ - "ld 2,-8(11)\n\t" /* use nraddr's tocptr */ \ - "addi 1,1,-128\n\t" /* expand stack frame */ \ - /* arg9 */ \ - "ld 3,72(11)\n\t" \ - "std 3,112(1)\n\t" \ - /* args1-8 */ \ - "ld 3, 8(11)\n\t" /* arg1->r3 */ \ - "ld 4, 16(11)\n\t" /* arg2->r4 */ \ - "ld 5, 24(11)\n\t" /* arg3->r5 */ \ - "ld 6, 32(11)\n\t" /* arg4->r6 */ \ - "ld 7, 40(11)\n\t" /* arg5->r7 */ \ - "ld 8, 48(11)\n\t" /* arg6->r8 */ \ - "ld 9, 56(11)\n\t" /* arg7->r9 */ \ - "ld 10, 64(11)\n\t" /* arg8->r10 */ \ - "ld 11, 0(11)\n\t" /* target->r11 */ \ - VALGRIND_BRANCH_AND_LINK_TO_NOREDIR_R11 \ - "mr 11,%1\n\t" \ - "mr %0,3\n\t" \ - "ld 2,-16(11)\n\t" /* restore tocptr */ \ - "addi 1,1,128" /* restore frame */ \ - : /*out*/ "=r" (_res) \ - : /*in*/ "r" (&_argvec[2]) \ - : /*trash*/ "cc", "memory", __CALLER_SAVED_REGS \ - ); \ - lval = (__typeof__(lval)) _res; \ - } while (0) - -#define CALL_FN_W_10W(lval, orig, arg1,arg2,arg3,arg4,arg5,arg6, \ - arg7,arg8,arg9,arg10) \ - do { \ - volatile OrigFn _orig = (orig); \ - volatile unsigned long _argvec[3+10]; \ - volatile unsigned long _res; \ - /* _argvec[0] holds current r2 across the call */ \ - _argvec[1] = (unsigned long)_orig.r2; \ - _argvec[2] = (unsigned long)_orig.nraddr; \ - _argvec[2+1] = (unsigned long)arg1; \ - _argvec[2+2] = (unsigned long)arg2; \ - _argvec[2+3] = (unsigned long)arg3; \ - _argvec[2+4] = (unsigned long)arg4; \ - _argvec[2+5] = (unsigned long)arg5; \ - _argvec[2+6] = (unsigned long)arg6; \ - _argvec[2+7] = (unsigned long)arg7; \ - _argvec[2+8] = (unsigned long)arg8; \ - _argvec[2+9] = (unsigned long)arg9; \ - _argvec[2+10] = (unsigned long)arg10; \ - __asm__ volatile( \ - "mr 11,%1\n\t" \ - "std 2,-16(11)\n\t" /* save tocptr */ \ - "ld 2,-8(11)\n\t" /* use nraddr's tocptr */ \ - "addi 1,1,-128\n\t" /* expand stack frame */ \ - /* arg10 */ \ - "ld 3,80(11)\n\t" \ - "std 3,120(1)\n\t" \ - /* arg9 */ \ - "ld 3,72(11)\n\t" \ - "std 3,112(1)\n\t" \ - /* args1-8 */ \ - "ld 3, 8(11)\n\t" /* arg1->r3 */ \ - "ld 4, 16(11)\n\t" /* arg2->r4 */ \ - "ld 5, 24(11)\n\t" /* arg3->r5 */ \ - "ld 6, 32(11)\n\t" /* arg4->r6 */ \ - "ld 7, 40(11)\n\t" /* arg5->r7 */ \ - "ld 8, 48(11)\n\t" /* arg6->r8 */ \ - "ld 9, 56(11)\n\t" /* arg7->r9 */ \ - "ld 10, 64(11)\n\t" /* arg8->r10 */ \ - "ld 11, 0(11)\n\t" /* target->r11 */ \ - VALGRIND_BRANCH_AND_LINK_TO_NOREDIR_R11 \ - "mr 11,%1\n\t" \ - "mr %0,3\n\t" \ - "ld 2,-16(11)\n\t" /* restore tocptr */ \ - "addi 1,1,128" /* restore frame */ \ - : /*out*/ "=r" (_res) \ - : /*in*/ "r" (&_argvec[2]) \ - : /*trash*/ "cc", "memory", __CALLER_SAVED_REGS \ - ); \ - lval = (__typeof__(lval)) _res; \ - } while (0) - -#define CALL_FN_W_11W(lval, orig, arg1,arg2,arg3,arg4,arg5,arg6, \ - arg7,arg8,arg9,arg10,arg11) \ - do { \ - volatile OrigFn _orig = (orig); \ - volatile unsigned long _argvec[3+11]; \ - volatile unsigned long _res; \ - /* _argvec[0] holds current r2 across the call */ \ - _argvec[1] = (unsigned long)_orig.r2; \ - _argvec[2] = (unsigned long)_orig.nraddr; \ - _argvec[2+1] = (unsigned long)arg1; \ - _argvec[2+2] = (unsigned long)arg2; \ - _argvec[2+3] = (unsigned long)arg3; \ - _argvec[2+4] = (unsigned long)arg4; \ - _argvec[2+5] = (unsigned long)arg5; \ - _argvec[2+6] = (unsigned long)arg6; \ - _argvec[2+7] = (unsigned long)arg7; \ - _argvec[2+8] = (unsigned long)arg8; \ - _argvec[2+9] = (unsigned long)arg9; \ - _argvec[2+10] = (unsigned long)arg10; \ - _argvec[2+11] = (unsigned long)arg11; \ - __asm__ volatile( \ - "mr 11,%1\n\t" \ - "std 2,-16(11)\n\t" /* save tocptr */ \ - "ld 2,-8(11)\n\t" /* use nraddr's tocptr */ \ - "addi 1,1,-144\n\t" /* expand stack frame */ \ - /* arg11 */ \ - "ld 3,88(11)\n\t" \ - "std 3,128(1)\n\t" \ - /* arg10 */ \ - "ld 3,80(11)\n\t" \ - "std 3,120(1)\n\t" \ - /* arg9 */ \ - "ld 3,72(11)\n\t" \ - "std 3,112(1)\n\t" \ - /* args1-8 */ \ - "ld 3, 8(11)\n\t" /* arg1->r3 */ \ - "ld 4, 16(11)\n\t" /* arg2->r4 */ \ - "ld 5, 24(11)\n\t" /* arg3->r5 */ \ - "ld 6, 32(11)\n\t" /* arg4->r6 */ \ - "ld 7, 40(11)\n\t" /* arg5->r7 */ \ - "ld 8, 48(11)\n\t" /* arg6->r8 */ \ - "ld 9, 56(11)\n\t" /* arg7->r9 */ \ - "ld 10, 64(11)\n\t" /* arg8->r10 */ \ - "ld 11, 0(11)\n\t" /* target->r11 */ \ - VALGRIND_BRANCH_AND_LINK_TO_NOREDIR_R11 \ - "mr 11,%1\n\t" \ - "mr %0,3\n\t" \ - "ld 2,-16(11)\n\t" /* restore tocptr */ \ - "addi 1,1,144" /* restore frame */ \ - : /*out*/ "=r" (_res) \ - : /*in*/ "r" (&_argvec[2]) \ - : /*trash*/ "cc", "memory", __CALLER_SAVED_REGS \ - ); \ - lval = (__typeof__(lval)) _res; \ - } while (0) - -#define CALL_FN_W_12W(lval, orig, arg1,arg2,arg3,arg4,arg5,arg6, \ - arg7,arg8,arg9,arg10,arg11,arg12) \ - do { \ - volatile OrigFn _orig = (orig); \ - volatile unsigned long _argvec[3+12]; \ - volatile unsigned long _res; \ - /* _argvec[0] holds current r2 across the call */ \ - _argvec[1] = (unsigned long)_orig.r2; \ - _argvec[2] = (unsigned long)_orig.nraddr; \ - _argvec[2+1] = (unsigned long)arg1; \ - _argvec[2+2] = (unsigned long)arg2; \ - _argvec[2+3] = (unsigned long)arg3; \ - _argvec[2+4] = (unsigned long)arg4; \ - _argvec[2+5] = (unsigned long)arg5; \ - _argvec[2+6] = (unsigned long)arg6; \ - _argvec[2+7] = (unsigned long)arg7; \ - _argvec[2+8] = (unsigned long)arg8; \ - _argvec[2+9] = (unsigned long)arg9; \ - _argvec[2+10] = (unsigned long)arg10; \ - _argvec[2+11] = (unsigned long)arg11; \ - _argvec[2+12] = (unsigned long)arg12; \ - __asm__ volatile( \ - "mr 11,%1\n\t" \ - "std 2,-16(11)\n\t" /* save tocptr */ \ - "ld 2,-8(11)\n\t" /* use nraddr's tocptr */ \ - "addi 1,1,-144\n\t" /* expand stack frame */ \ - /* arg12 */ \ - "ld 3,96(11)\n\t" \ - "std 3,136(1)\n\t" \ - /* arg11 */ \ - "ld 3,88(11)\n\t" \ - "std 3,128(1)\n\t" \ - /* arg10 */ \ - "ld 3,80(11)\n\t" \ - "std 3,120(1)\n\t" \ - /* arg9 */ \ - "ld 3,72(11)\n\t" \ - "std 3,112(1)\n\t" \ - /* args1-8 */ \ - "ld 3, 8(11)\n\t" /* arg1->r3 */ \ - "ld 4, 16(11)\n\t" /* arg2->r4 */ \ - "ld 5, 24(11)\n\t" /* arg3->r5 */ \ - "ld 6, 32(11)\n\t" /* arg4->r6 */ \ - "ld 7, 40(11)\n\t" /* arg5->r7 */ \ - "ld 8, 48(11)\n\t" /* arg6->r8 */ \ - "ld 9, 56(11)\n\t" /* arg7->r9 */ \ - "ld 10, 64(11)\n\t" /* arg8->r10 */ \ - "ld 11, 0(11)\n\t" /* target->r11 */ \ - VALGRIND_BRANCH_AND_LINK_TO_NOREDIR_R11 \ - "mr 11,%1\n\t" \ - "mr %0,3\n\t" \ - "ld 2,-16(11)\n\t" /* restore tocptr */ \ - "addi 1,1,144" /* restore frame */ \ - : /*out*/ "=r" (_res) \ - : /*in*/ "r" (&_argvec[2]) \ - : /*trash*/ "cc", "memory", __CALLER_SAVED_REGS \ - ); \ - lval = (__typeof__(lval)) _res; \ - } while (0) - -#endif /* PLAT_ppc64_linux */ - -/* ------------------------ ppc32-aix5 ------------------------- */ - -#if defined(PLAT_ppc32_aix5) - -/* ARGREGS: r3 r4 r5 r6 r7 r8 r9 r10 (the rest on stack somewhere) */ - -/* These regs are trashed by the hidden call. */ -#define __CALLER_SAVED_REGS \ - "lr", "ctr", "xer", \ - "cr0", "cr1", "cr2", "cr3", "cr4", "cr5", "cr6", "cr7", \ - "r0", "r2", "r3", "r4", "r5", "r6", "r7", "r8", "r9", "r10", \ - "r11", "r12", "r13" - -/* Expand the stack frame, copying enough info that unwinding - still works. Trashes r3. */ - -#define VG_EXPAND_FRAME_BY_trashes_r3(_n_fr) \ - "addi 1,1,-" #_n_fr "\n\t" \ - "lwz 3," #_n_fr "(1)\n\t" \ - "stw 3,0(1)\n\t" - -#define VG_CONTRACT_FRAME_BY(_n_fr) \ - "addi 1,1," #_n_fr "\n\t" - -/* These CALL_FN_ macros assume that on ppc32-aix5, sizeof(unsigned - long) == 4. */ - -#define CALL_FN_W_v(lval, orig) \ - do { \ - volatile OrigFn _orig = (orig); \ - volatile unsigned long _argvec[3+0]; \ - volatile unsigned long _res; \ - /* _argvec[0] holds current r2 across the call */ \ - _argvec[1] = (unsigned long)_orig.r2; \ - _argvec[2] = (unsigned long)_orig.nraddr; \ - __asm__ volatile( \ - "mr 11,%1\n\t" \ - VG_EXPAND_FRAME_BY_trashes_r3(512) \ - "stw 2,-8(11)\n\t" /* save tocptr */ \ - "lwz 2,-4(11)\n\t" /* use nraddr's tocptr */ \ - "lwz 11, 0(11)\n\t" /* target->r11 */ \ - VALGRIND_BRANCH_AND_LINK_TO_NOREDIR_R11 \ - "mr 11,%1\n\t" \ - "mr %0,3\n\t" \ - "lwz 2,-8(11)\n\t" /* restore tocptr */ \ - VG_CONTRACT_FRAME_BY(512) \ - : /*out*/ "=r" (_res) \ - : /*in*/ "r" (&_argvec[2]) \ - : /*trash*/ "cc", "memory", __CALLER_SAVED_REGS \ - ); \ - lval = (__typeof__(lval)) _res; \ - } while (0) - -#define CALL_FN_W_W(lval, orig, arg1) \ - do { \ - volatile OrigFn _orig = (orig); \ - volatile unsigned long _argvec[3+1]; \ - volatile unsigned long _res; \ - /* _argvec[0] holds current r2 across the call */ \ - _argvec[1] = (unsigned long)_orig.r2; \ - _argvec[2] = (unsigned long)_orig.nraddr; \ - _argvec[2+1] = (unsigned long)arg1; \ - __asm__ volatile( \ - "mr 11,%1\n\t" \ - VG_EXPAND_FRAME_BY_trashes_r3(512) \ - "stw 2,-8(11)\n\t" /* save tocptr */ \ - "lwz 2,-4(11)\n\t" /* use nraddr's tocptr */ \ - "lwz 3, 4(11)\n\t" /* arg1->r3 */ \ - "lwz 11, 0(11)\n\t" /* target->r11 */ \ - VALGRIND_BRANCH_AND_LINK_TO_NOREDIR_R11 \ - "mr 11,%1\n\t" \ - "mr %0,3\n\t" \ - "lwz 2,-8(11)\n\t" /* restore tocptr */ \ - VG_CONTRACT_FRAME_BY(512) \ - : /*out*/ "=r" (_res) \ - : /*in*/ "r" (&_argvec[2]) \ - : /*trash*/ "cc", "memory", __CALLER_SAVED_REGS \ - ); \ - lval = (__typeof__(lval)) _res; \ - } while (0) - -#define CALL_FN_W_WW(lval, orig, arg1,arg2) \ - do { \ - volatile OrigFn _orig = (orig); \ - volatile unsigned long _argvec[3+2]; \ - volatile unsigned long _res; \ - /* _argvec[0] holds current r2 across the call */ \ - _argvec[1] = (unsigned long)_orig.r2; \ - _argvec[2] = (unsigned long)_orig.nraddr; \ - _argvec[2+1] = (unsigned long)arg1; \ - _argvec[2+2] = (unsigned long)arg2; \ - __asm__ volatile( \ - "mr 11,%1\n\t" \ - VG_EXPAND_FRAME_BY_trashes_r3(512) \ - "stw 2,-8(11)\n\t" /* save tocptr */ \ - "lwz 2,-4(11)\n\t" /* use nraddr's tocptr */ \ - "lwz 3, 4(11)\n\t" /* arg1->r3 */ \ - "lwz 4, 8(11)\n\t" /* arg2->r4 */ \ - "lwz 11, 0(11)\n\t" /* target->r11 */ \ - VALGRIND_BRANCH_AND_LINK_TO_NOREDIR_R11 \ - "mr 11,%1\n\t" \ - "mr %0,3\n\t" \ - "lwz 2,-8(11)\n\t" /* restore tocptr */ \ - VG_CONTRACT_FRAME_BY(512) \ - : /*out*/ "=r" (_res) \ - : /*in*/ "r" (&_argvec[2]) \ - : /*trash*/ "cc", "memory", __CALLER_SAVED_REGS \ - ); \ - lval = (__typeof__(lval)) _res; \ - } while (0) - -#define CALL_FN_W_WWW(lval, orig, arg1,arg2,arg3) \ - do { \ - volatile OrigFn _orig = (orig); \ - volatile unsigned long _argvec[3+3]; \ - volatile unsigned long _res; \ - /* _argvec[0] holds current r2 across the call */ \ - _argvec[1] = (unsigned long)_orig.r2; \ - _argvec[2] = (unsigned long)_orig.nraddr; \ - _argvec[2+1] = (unsigned long)arg1; \ - _argvec[2+2] = (unsigned long)arg2; \ - _argvec[2+3] = (unsigned long)arg3; \ - __asm__ volatile( \ - "mr 11,%1\n\t" \ - VG_EXPAND_FRAME_BY_trashes_r3(512) \ - "stw 2,-8(11)\n\t" /* save tocptr */ \ - "lwz 2,-4(11)\n\t" /* use nraddr's tocptr */ \ - "lwz 3, 4(11)\n\t" /* arg1->r3 */ \ - "lwz 4, 8(11)\n\t" /* arg2->r4 */ \ - "lwz 5, 12(11)\n\t" /* arg3->r5 */ \ - "lwz 11, 0(11)\n\t" /* target->r11 */ \ - VALGRIND_BRANCH_AND_LINK_TO_NOREDIR_R11 \ - "mr 11,%1\n\t" \ - "mr %0,3\n\t" \ - "lwz 2,-8(11)\n\t" /* restore tocptr */ \ - VG_CONTRACT_FRAME_BY(512) \ - : /*out*/ "=r" (_res) \ - : /*in*/ "r" (&_argvec[2]) \ - : /*trash*/ "cc", "memory", __CALLER_SAVED_REGS \ - ); \ - lval = (__typeof__(lval)) _res; \ - } while (0) - -#define CALL_FN_W_WWWW(lval, orig, arg1,arg2,arg3,arg4) \ - do { \ - volatile OrigFn _orig = (orig); \ - volatile unsigned long _argvec[3+4]; \ - volatile unsigned long _res; \ - /* _argvec[0] holds current r2 across the call */ \ - _argvec[1] = (unsigned long)_orig.r2; \ - _argvec[2] = (unsigned long)_orig.nraddr; \ - _argvec[2+1] = (unsigned long)arg1; \ - _argvec[2+2] = (unsigned long)arg2; \ - _argvec[2+3] = (unsigned long)arg3; \ - _argvec[2+4] = (unsigned long)arg4; \ - __asm__ volatile( \ - "mr 11,%1\n\t" \ - VG_EXPAND_FRAME_BY_trashes_r3(512) \ - "stw 2,-8(11)\n\t" /* save tocptr */ \ - "lwz 2,-4(11)\n\t" /* use nraddr's tocptr */ \ - "lwz 3, 4(11)\n\t" /* arg1->r3 */ \ - "lwz 4, 8(11)\n\t" /* arg2->r4 */ \ - "lwz 5, 12(11)\n\t" /* arg3->r5 */ \ - "lwz 6, 16(11)\n\t" /* arg4->r6 */ \ - "lwz 11, 0(11)\n\t" /* target->r11 */ \ - VALGRIND_BRANCH_AND_LINK_TO_NOREDIR_R11 \ - "mr 11,%1\n\t" \ - "mr %0,3\n\t" \ - "lwz 2,-8(11)\n\t" /* restore tocptr */ \ - VG_CONTRACT_FRAME_BY(512) \ - : /*out*/ "=r" (_res) \ - : /*in*/ "r" (&_argvec[2]) \ - : /*trash*/ "cc", "memory", __CALLER_SAVED_REGS \ - ); \ - lval = (__typeof__(lval)) _res; \ - } while (0) - -#define CALL_FN_W_5W(lval, orig, arg1,arg2,arg3,arg4,arg5) \ - do { \ - volatile OrigFn _orig = (orig); \ - volatile unsigned long _argvec[3+5]; \ - volatile unsigned long _res; \ - /* _argvec[0] holds current r2 across the call */ \ - _argvec[1] = (unsigned long)_orig.r2; \ - _argvec[2] = (unsigned long)_orig.nraddr; \ - _argvec[2+1] = (unsigned long)arg1; \ - _argvec[2+2] = (unsigned long)arg2; \ - _argvec[2+3] = (unsigned long)arg3; \ - _argvec[2+4] = (unsigned long)arg4; \ - _argvec[2+5] = (unsigned long)arg5; \ - __asm__ volatile( \ - "mr 11,%1\n\t" \ - VG_EXPAND_FRAME_BY_trashes_r3(512) \ - "stw 2,-8(11)\n\t" /* save tocptr */ \ - "lwz 2,-4(11)\n\t" /* use nraddr's tocptr */ \ - "lwz 3, 4(11)\n\t" /* arg1->r3 */ \ - "lwz 4, 8(11)\n\t" /* arg2->r4 */ \ - "lwz 5, 12(11)\n\t" /* arg3->r5 */ \ - "lwz 6, 16(11)\n\t" /* arg4->r6 */ \ - "lwz 7, 20(11)\n\t" /* arg5->r7 */ \ - "lwz 11, 0(11)\n\t" /* target->r11 */ \ - VALGRIND_BRANCH_AND_LINK_TO_NOREDIR_R11 \ - "mr 11,%1\n\t" \ - "mr %0,3\n\t" \ - "lwz 2,-8(11)\n\t" /* restore tocptr */ \ - VG_CONTRACT_FRAME_BY(512) \ - : /*out*/ "=r" (_res) \ - : /*in*/ "r" (&_argvec[2]) \ - : /*trash*/ "cc", "memory", __CALLER_SAVED_REGS \ - ); \ - lval = (__typeof__(lval)) _res; \ - } while (0) - -#define CALL_FN_W_6W(lval, orig, arg1,arg2,arg3,arg4,arg5,arg6) \ - do { \ - volatile OrigFn _orig = (orig); \ - volatile unsigned long _argvec[3+6]; \ - volatile unsigned long _res; \ - /* _argvec[0] holds current r2 across the call */ \ - _argvec[1] = (unsigned long)_orig.r2; \ - _argvec[2] = (unsigned long)_orig.nraddr; \ - _argvec[2+1] = (unsigned long)arg1; \ - _argvec[2+2] = (unsigned long)arg2; \ - _argvec[2+3] = (unsigned long)arg3; \ - _argvec[2+4] = (unsigned long)arg4; \ - _argvec[2+5] = (unsigned long)arg5; \ - _argvec[2+6] = (unsigned long)arg6; \ - __asm__ volatile( \ - "mr 11,%1\n\t" \ - VG_EXPAND_FRAME_BY_trashes_r3(512) \ - "stw 2,-8(11)\n\t" /* save tocptr */ \ - "lwz 2,-4(11)\n\t" /* use nraddr's tocptr */ \ - "lwz 3, 4(11)\n\t" /* arg1->r3 */ \ - "lwz 4, 8(11)\n\t" /* arg2->r4 */ \ - "lwz 5, 12(11)\n\t" /* arg3->r5 */ \ - "lwz 6, 16(11)\n\t" /* arg4->r6 */ \ - "lwz 7, 20(11)\n\t" /* arg5->r7 */ \ - "lwz 8, 24(11)\n\t" /* arg6->r8 */ \ - "lwz 11, 0(11)\n\t" /* target->r11 */ \ - VALGRIND_BRANCH_AND_LINK_TO_NOREDIR_R11 \ - "mr 11,%1\n\t" \ - "mr %0,3\n\t" \ - "lwz 2,-8(11)\n\t" /* restore tocptr */ \ - VG_CONTRACT_FRAME_BY(512) \ - : /*out*/ "=r" (_res) \ - : /*in*/ "r" (&_argvec[2]) \ - : /*trash*/ "cc", "memory", __CALLER_SAVED_REGS \ - ); \ - lval = (__typeof__(lval)) _res; \ - } while (0) - -#define CALL_FN_W_7W(lval, orig, arg1,arg2,arg3,arg4,arg5,arg6, \ - arg7) \ - do { \ - volatile OrigFn _orig = (orig); \ - volatile unsigned long _argvec[3+7]; \ - volatile unsigned long _res; \ - /* _argvec[0] holds current r2 across the call */ \ - _argvec[1] = (unsigned long)_orig.r2; \ - _argvec[2] = (unsigned long)_orig.nraddr; \ - _argvec[2+1] = (unsigned long)arg1; \ - _argvec[2+2] = (unsigned long)arg2; \ - _argvec[2+3] = (unsigned long)arg3; \ - _argvec[2+4] = (unsigned long)arg4; \ - _argvec[2+5] = (unsigned long)arg5; \ - _argvec[2+6] = (unsigned long)arg6; \ - _argvec[2+7] = (unsigned long)arg7; \ - __asm__ volatile( \ - "mr 11,%1\n\t" \ - VG_EXPAND_FRAME_BY_trashes_r3(512) \ - "stw 2,-8(11)\n\t" /* save tocptr */ \ - "lwz 2,-4(11)\n\t" /* use nraddr's tocptr */ \ - "lwz 3, 4(11)\n\t" /* arg1->r3 */ \ - "lwz 4, 8(11)\n\t" /* arg2->r4 */ \ - "lwz 5, 12(11)\n\t" /* arg3->r5 */ \ - "lwz 6, 16(11)\n\t" /* arg4->r6 */ \ - "lwz 7, 20(11)\n\t" /* arg5->r7 */ \ - "lwz 8, 24(11)\n\t" /* arg6->r8 */ \ - "lwz 9, 28(11)\n\t" /* arg7->r9 */ \ - "lwz 11, 0(11)\n\t" /* target->r11 */ \ - VALGRIND_BRANCH_AND_LINK_TO_NOREDIR_R11 \ - "mr 11,%1\n\t" \ - "mr %0,3\n\t" \ - "lwz 2,-8(11)\n\t" /* restore tocptr */ \ - VG_CONTRACT_FRAME_BY(512) \ - : /*out*/ "=r" (_res) \ - : /*in*/ "r" (&_argvec[2]) \ - : /*trash*/ "cc", "memory", __CALLER_SAVED_REGS \ - ); \ - lval = (__typeof__(lval)) _res; \ - } while (0) - -#define CALL_FN_W_8W(lval, orig, arg1,arg2,arg3,arg4,arg5,arg6, \ - arg7,arg8) \ - do { \ - volatile OrigFn _orig = (orig); \ - volatile unsigned long _argvec[3+8]; \ - volatile unsigned long _res; \ - /* _argvec[0] holds current r2 across the call */ \ - _argvec[1] = (unsigned long)_orig.r2; \ - _argvec[2] = (unsigned long)_orig.nraddr; \ - _argvec[2+1] = (unsigned long)arg1; \ - _argvec[2+2] = (unsigned long)arg2; \ - _argvec[2+3] = (unsigned long)arg3; \ - _argvec[2+4] = (unsigned long)arg4; \ - _argvec[2+5] = (unsigned long)arg5; \ - _argvec[2+6] = (unsigned long)arg6; \ - _argvec[2+7] = (unsigned long)arg7; \ - _argvec[2+8] = (unsigned long)arg8; \ - __asm__ volatile( \ - "mr 11,%1\n\t" \ - VG_EXPAND_FRAME_BY_trashes_r3(512) \ - "stw 2,-8(11)\n\t" /* save tocptr */ \ - "lwz 2,-4(11)\n\t" /* use nraddr's tocptr */ \ - "lwz 3, 4(11)\n\t" /* arg1->r3 */ \ - "lwz 4, 8(11)\n\t" /* arg2->r4 */ \ - "lwz 5, 12(11)\n\t" /* arg3->r5 */ \ - "lwz 6, 16(11)\n\t" /* arg4->r6 */ \ - "lwz 7, 20(11)\n\t" /* arg5->r7 */ \ - "lwz 8, 24(11)\n\t" /* arg6->r8 */ \ - "lwz 9, 28(11)\n\t" /* arg7->r9 */ \ - "lwz 10, 32(11)\n\t" /* arg8->r10 */ \ - "lwz 11, 0(11)\n\t" /* target->r11 */ \ - VALGRIND_BRANCH_AND_LINK_TO_NOREDIR_R11 \ - "mr 11,%1\n\t" \ - "mr %0,3\n\t" \ - "lwz 2,-8(11)\n\t" /* restore tocptr */ \ - VG_CONTRACT_FRAME_BY(512) \ - : /*out*/ "=r" (_res) \ - : /*in*/ "r" (&_argvec[2]) \ - : /*trash*/ "cc", "memory", __CALLER_SAVED_REGS \ - ); \ - lval = (__typeof__(lval)) _res; \ - } while (0) - -#define CALL_FN_W_9W(lval, orig, arg1,arg2,arg3,arg4,arg5,arg6, \ - arg7,arg8,arg9) \ - do { \ - volatile OrigFn _orig = (orig); \ - volatile unsigned long _argvec[3+9]; \ - volatile unsigned long _res; \ - /* _argvec[0] holds current r2 across the call */ \ - _argvec[1] = (unsigned long)_orig.r2; \ - _argvec[2] = (unsigned long)_orig.nraddr; \ - _argvec[2+1] = (unsigned long)arg1; \ - _argvec[2+2] = (unsigned long)arg2; \ - _argvec[2+3] = (unsigned long)arg3; \ - _argvec[2+4] = (unsigned long)arg4; \ - _argvec[2+5] = (unsigned long)arg5; \ - _argvec[2+6] = (unsigned long)arg6; \ - _argvec[2+7] = (unsigned long)arg7; \ - _argvec[2+8] = (unsigned long)arg8; \ - _argvec[2+9] = (unsigned long)arg9; \ - __asm__ volatile( \ - "mr 11,%1\n\t" \ - VG_EXPAND_FRAME_BY_trashes_r3(512) \ - "stw 2,-8(11)\n\t" /* save tocptr */ \ - "lwz 2,-4(11)\n\t" /* use nraddr's tocptr */ \ - VG_EXPAND_FRAME_BY_trashes_r3(64) \ - /* arg9 */ \ - "lwz 3,36(11)\n\t" \ - "stw 3,56(1)\n\t" \ - /* args1-8 */ \ - "lwz 3, 4(11)\n\t" /* arg1->r3 */ \ - "lwz 4, 8(11)\n\t" /* arg2->r4 */ \ - "lwz 5, 12(11)\n\t" /* arg3->r5 */ \ - "lwz 6, 16(11)\n\t" /* arg4->r6 */ \ - "lwz 7, 20(11)\n\t" /* arg5->r7 */ \ - "lwz 8, 24(11)\n\t" /* arg6->r8 */ \ - "lwz 9, 28(11)\n\t" /* arg7->r9 */ \ - "lwz 10, 32(11)\n\t" /* arg8->r10 */ \ - "lwz 11, 0(11)\n\t" /* target->r11 */ \ - VALGRIND_BRANCH_AND_LINK_TO_NOREDIR_R11 \ - "mr 11,%1\n\t" \ - "mr %0,3\n\t" \ - "lwz 2,-8(11)\n\t" /* restore tocptr */ \ - VG_CONTRACT_FRAME_BY(64) \ - VG_CONTRACT_FRAME_BY(512) \ - : /*out*/ "=r" (_res) \ - : /*in*/ "r" (&_argvec[2]) \ - : /*trash*/ "cc", "memory", __CALLER_SAVED_REGS \ - ); \ - lval = (__typeof__(lval)) _res; \ - } while (0) - -#define CALL_FN_W_10W(lval, orig, arg1,arg2,arg3,arg4,arg5,arg6, \ - arg7,arg8,arg9,arg10) \ - do { \ - volatile OrigFn _orig = (orig); \ - volatile unsigned long _argvec[3+10]; \ - volatile unsigned long _res; \ - /* _argvec[0] holds current r2 across the call */ \ - _argvec[1] = (unsigned long)_orig.r2; \ - _argvec[2] = (unsigned long)_orig.nraddr; \ - _argvec[2+1] = (unsigned long)arg1; \ - _argvec[2+2] = (unsigned long)arg2; \ - _argvec[2+3] = (unsigned long)arg3; \ - _argvec[2+4] = (unsigned long)arg4; \ - _argvec[2+5] = (unsigned long)arg5; \ - _argvec[2+6] = (unsigned long)arg6; \ - _argvec[2+7] = (unsigned long)arg7; \ - _argvec[2+8] = (unsigned long)arg8; \ - _argvec[2+9] = (unsigned long)arg9; \ - _argvec[2+10] = (unsigned long)arg10; \ - __asm__ volatile( \ - "mr 11,%1\n\t" \ - VG_EXPAND_FRAME_BY_trashes_r3(512) \ - "stw 2,-8(11)\n\t" /* save tocptr */ \ - "lwz 2,-4(11)\n\t" /* use nraddr's tocptr */ \ - VG_EXPAND_FRAME_BY_trashes_r3(64) \ - /* arg10 */ \ - "lwz 3,40(11)\n\t" \ - "stw 3,60(1)\n\t" \ - /* arg9 */ \ - "lwz 3,36(11)\n\t" \ - "stw 3,56(1)\n\t" \ - /* args1-8 */ \ - "lwz 3, 4(11)\n\t" /* arg1->r3 */ \ - "lwz 4, 8(11)\n\t" /* arg2->r4 */ \ - "lwz 5, 12(11)\n\t" /* arg3->r5 */ \ - "lwz 6, 16(11)\n\t" /* arg4->r6 */ \ - "lwz 7, 20(11)\n\t" /* arg5->r7 */ \ - "lwz 8, 24(11)\n\t" /* arg6->r8 */ \ - "lwz 9, 28(11)\n\t" /* arg7->r9 */ \ - "lwz 10, 32(11)\n\t" /* arg8->r10 */ \ - "lwz 11, 0(11)\n\t" /* target->r11 */ \ - VALGRIND_BRANCH_AND_LINK_TO_NOREDIR_R11 \ - "mr 11,%1\n\t" \ - "mr %0,3\n\t" \ - "lwz 2,-8(11)\n\t" /* restore tocptr */ \ - VG_CONTRACT_FRAME_BY(64) \ - VG_CONTRACT_FRAME_BY(512) \ - : /*out*/ "=r" (_res) \ - : /*in*/ "r" (&_argvec[2]) \ - : /*trash*/ "cc", "memory", __CALLER_SAVED_REGS \ - ); \ - lval = (__typeof__(lval)) _res; \ - } while (0) - -#define CALL_FN_W_11W(lval, orig, arg1,arg2,arg3,arg4,arg5,arg6, \ - arg7,arg8,arg9,arg10,arg11) \ - do { \ - volatile OrigFn _orig = (orig); \ - volatile unsigned long _argvec[3+11]; \ - volatile unsigned long _res; \ - /* _argvec[0] holds current r2 across the call */ \ - _argvec[1] = (unsigned long)_orig.r2; \ - _argvec[2] = (unsigned long)_orig.nraddr; \ - _argvec[2+1] = (unsigned long)arg1; \ - _argvec[2+2] = (unsigned long)arg2; \ - _argvec[2+3] = (unsigned long)arg3; \ - _argvec[2+4] = (unsigned long)arg4; \ - _argvec[2+5] = (unsigned long)arg5; \ - _argvec[2+6] = (unsigned long)arg6; \ - _argvec[2+7] = (unsigned long)arg7; \ - _argvec[2+8] = (unsigned long)arg8; \ - _argvec[2+9] = (unsigned long)arg9; \ - _argvec[2+10] = (unsigned long)arg10; \ - _argvec[2+11] = (unsigned long)arg11; \ - __asm__ volatile( \ - "mr 11,%1\n\t" \ - VG_EXPAND_FRAME_BY_trashes_r3(512) \ - "stw 2,-8(11)\n\t" /* save tocptr */ \ - "lwz 2,-4(11)\n\t" /* use nraddr's tocptr */ \ - VG_EXPAND_FRAME_BY_trashes_r3(72) \ - /* arg11 */ \ - "lwz 3,44(11)\n\t" \ - "stw 3,64(1)\n\t" \ - /* arg10 */ \ - "lwz 3,40(11)\n\t" \ - "stw 3,60(1)\n\t" \ - /* arg9 */ \ - "lwz 3,36(11)\n\t" \ - "stw 3,56(1)\n\t" \ - /* args1-8 */ \ - "lwz 3, 4(11)\n\t" /* arg1->r3 */ \ - "lwz 4, 8(11)\n\t" /* arg2->r4 */ \ - "lwz 5, 12(11)\n\t" /* arg3->r5 */ \ - "lwz 6, 16(11)\n\t" /* arg4->r6 */ \ - "lwz 7, 20(11)\n\t" /* arg5->r7 */ \ - "lwz 8, 24(11)\n\t" /* arg6->r8 */ \ - "lwz 9, 28(11)\n\t" /* arg7->r9 */ \ - "lwz 10, 32(11)\n\t" /* arg8->r10 */ \ - "lwz 11, 0(11)\n\t" /* target->r11 */ \ - VALGRIND_BRANCH_AND_LINK_TO_NOREDIR_R11 \ - "mr 11,%1\n\t" \ - "mr %0,3\n\t" \ - "lwz 2,-8(11)\n\t" /* restore tocptr */ \ - VG_CONTRACT_FRAME_BY(72) \ - VG_CONTRACT_FRAME_BY(512) \ - : /*out*/ "=r" (_res) \ - : /*in*/ "r" (&_argvec[2]) \ - : /*trash*/ "cc", "memory", __CALLER_SAVED_REGS \ - ); \ - lval = (__typeof__(lval)) _res; \ - } while (0) - -#define CALL_FN_W_12W(lval, orig, arg1,arg2,arg3,arg4,arg5,arg6, \ - arg7,arg8,arg9,arg10,arg11,arg12) \ - do { \ - volatile OrigFn _orig = (orig); \ - volatile unsigned long _argvec[3+12]; \ - volatile unsigned long _res; \ - /* _argvec[0] holds current r2 across the call */ \ - _argvec[1] = (unsigned long)_orig.r2; \ - _argvec[2] = (unsigned long)_orig.nraddr; \ - _argvec[2+1] = (unsigned long)arg1; \ - _argvec[2+2] = (unsigned long)arg2; \ - _argvec[2+3] = (unsigned long)arg3; \ - _argvec[2+4] = (unsigned long)arg4; \ - _argvec[2+5] = (unsigned long)arg5; \ - _argvec[2+6] = (unsigned long)arg6; \ - _argvec[2+7] = (unsigned long)arg7; \ - _argvec[2+8] = (unsigned long)arg8; \ - _argvec[2+9] = (unsigned long)arg9; \ - _argvec[2+10] = (unsigned long)arg10; \ - _argvec[2+11] = (unsigned long)arg11; \ - _argvec[2+12] = (unsigned long)arg12; \ - __asm__ volatile( \ - "mr 11,%1\n\t" \ - VG_EXPAND_FRAME_BY_trashes_r3(512) \ - "stw 2,-8(11)\n\t" /* save tocptr */ \ - "lwz 2,-4(11)\n\t" /* use nraddr's tocptr */ \ - VG_EXPAND_FRAME_BY_trashes_r3(72) \ - /* arg12 */ \ - "lwz 3,48(11)\n\t" \ - "stw 3,68(1)\n\t" \ - /* arg11 */ \ - "lwz 3,44(11)\n\t" \ - "stw 3,64(1)\n\t" \ - /* arg10 */ \ - "lwz 3,40(11)\n\t" \ - "stw 3,60(1)\n\t" \ - /* arg9 */ \ - "lwz 3,36(11)\n\t" \ - "stw 3,56(1)\n\t" \ - /* args1-8 */ \ - "lwz 3, 4(11)\n\t" /* arg1->r3 */ \ - "lwz 4, 8(11)\n\t" /* arg2->r4 */ \ - "lwz 5, 12(11)\n\t" /* arg3->r5 */ \ - "lwz 6, 16(11)\n\t" /* arg4->r6 */ \ - "lwz 7, 20(11)\n\t" /* arg5->r7 */ \ - "lwz 8, 24(11)\n\t" /* arg6->r8 */ \ - "lwz 9, 28(11)\n\t" /* arg7->r9 */ \ - "lwz 10, 32(11)\n\t" /* arg8->r10 */ \ - "lwz 11, 0(11)\n\t" /* target->r11 */ \ - VALGRIND_BRANCH_AND_LINK_TO_NOREDIR_R11 \ - "mr 11,%1\n\t" \ - "mr %0,3\n\t" \ - "lwz 2,-8(11)\n\t" /* restore tocptr */ \ - VG_CONTRACT_FRAME_BY(72) \ - VG_CONTRACT_FRAME_BY(512) \ - : /*out*/ "=r" (_res) \ - : /*in*/ "r" (&_argvec[2]) \ - : /*trash*/ "cc", "memory", __CALLER_SAVED_REGS \ - ); \ - lval = (__typeof__(lval)) _res; \ - } while (0) - -#endif /* PLAT_ppc32_aix5 */ - -/* ------------------------ ppc64-aix5 ------------------------- */ - -#if defined(PLAT_ppc64_aix5) - -/* ARGREGS: r3 r4 r5 r6 r7 r8 r9 r10 (the rest on stack somewhere) */ - -/* These regs are trashed by the hidden call. */ -#define __CALLER_SAVED_REGS \ - "lr", "ctr", "xer", \ - "cr0", "cr1", "cr2", "cr3", "cr4", "cr5", "cr6", "cr7", \ - "r0", "r2", "r3", "r4", "r5", "r6", "r7", "r8", "r9", "r10", \ - "r11", "r12", "r13" - -/* Expand the stack frame, copying enough info that unwinding - still works. Trashes r3. */ - -#define VG_EXPAND_FRAME_BY_trashes_r3(_n_fr) \ - "addi 1,1,-" #_n_fr "\n\t" \ - "ld 3," #_n_fr "(1)\n\t" \ - "std 3,0(1)\n\t" - -#define VG_CONTRACT_FRAME_BY(_n_fr) \ - "addi 1,1," #_n_fr "\n\t" - -/* These CALL_FN_ macros assume that on ppc64-aix5, sizeof(unsigned - long) == 8. */ - -#define CALL_FN_W_v(lval, orig) \ - do { \ - volatile OrigFn _orig = (orig); \ - volatile unsigned long _argvec[3+0]; \ - volatile unsigned long _res; \ - /* _argvec[0] holds current r2 across the call */ \ - _argvec[1] = (unsigned long)_orig.r2; \ - _argvec[2] = (unsigned long)_orig.nraddr; \ - __asm__ volatile( \ - "mr 11,%1\n\t" \ - VG_EXPAND_FRAME_BY_trashes_r3(512) \ - "std 2,-16(11)\n\t" /* save tocptr */ \ - "ld 2,-8(11)\n\t" /* use nraddr's tocptr */ \ - "ld 11, 0(11)\n\t" /* target->r11 */ \ - VALGRIND_BRANCH_AND_LINK_TO_NOREDIR_R11 \ - "mr 11,%1\n\t" \ - "mr %0,3\n\t" \ - "ld 2,-16(11)\n\t" /* restore tocptr */ \ - VG_CONTRACT_FRAME_BY(512) \ - : /*out*/ "=r" (_res) \ - : /*in*/ "r" (&_argvec[2]) \ - : /*trash*/ "cc", "memory", __CALLER_SAVED_REGS \ - ); \ - lval = (__typeof__(lval)) _res; \ - } while (0) - -#define CALL_FN_W_W(lval, orig, arg1) \ - do { \ - volatile OrigFn _orig = (orig); \ - volatile unsigned long _argvec[3+1]; \ - volatile unsigned long _res; \ - /* _argvec[0] holds current r2 across the call */ \ - _argvec[1] = (unsigned long)_orig.r2; \ - _argvec[2] = (unsigned long)_orig.nraddr; \ - _argvec[2+1] = (unsigned long)arg1; \ - __asm__ volatile( \ - "mr 11,%1\n\t" \ - VG_EXPAND_FRAME_BY_trashes_r3(512) \ - "std 2,-16(11)\n\t" /* save tocptr */ \ - "ld 2,-8(11)\n\t" /* use nraddr's tocptr */ \ - "ld 3, 8(11)\n\t" /* arg1->r3 */ \ - "ld 11, 0(11)\n\t" /* target->r11 */ \ - VALGRIND_BRANCH_AND_LINK_TO_NOREDIR_R11 \ - "mr 11,%1\n\t" \ - "mr %0,3\n\t" \ - "ld 2,-16(11)\n\t" /* restore tocptr */ \ - VG_CONTRACT_FRAME_BY(512) \ - : /*out*/ "=r" (_res) \ - : /*in*/ "r" (&_argvec[2]) \ - : /*trash*/ "cc", "memory", __CALLER_SAVED_REGS \ - ); \ - lval = (__typeof__(lval)) _res; \ - } while (0) - -#define CALL_FN_W_WW(lval, orig, arg1,arg2) \ - do { \ - volatile OrigFn _orig = (orig); \ - volatile unsigned long _argvec[3+2]; \ - volatile unsigned long _res; \ - /* _argvec[0] holds current r2 across the call */ \ - _argvec[1] = (unsigned long)_orig.r2; \ - _argvec[2] = (unsigned long)_orig.nraddr; \ - _argvec[2+1] = (unsigned long)arg1; \ - _argvec[2+2] = (unsigned long)arg2; \ - __asm__ volatile( \ - "mr 11,%1\n\t" \ - VG_EXPAND_FRAME_BY_trashes_r3(512) \ - "std 2,-16(11)\n\t" /* save tocptr */ \ - "ld 2,-8(11)\n\t" /* use nraddr's tocptr */ \ - "ld 3, 8(11)\n\t" /* arg1->r3 */ \ - "ld 4, 16(11)\n\t" /* arg2->r4 */ \ - "ld 11, 0(11)\n\t" /* target->r11 */ \ - VALGRIND_BRANCH_AND_LINK_TO_NOREDIR_R11 \ - "mr 11,%1\n\t" \ - "mr %0,3\n\t" \ - "ld 2,-16(11)\n\t" /* restore tocptr */ \ - VG_CONTRACT_FRAME_BY(512) \ - : /*out*/ "=r" (_res) \ - : /*in*/ "r" (&_argvec[2]) \ - : /*trash*/ "cc", "memory", __CALLER_SAVED_REGS \ - ); \ - lval = (__typeof__(lval)) _res; \ - } while (0) - -#define CALL_FN_W_WWW(lval, orig, arg1,arg2,arg3) \ - do { \ - volatile OrigFn _orig = (orig); \ - volatile unsigned long _argvec[3+3]; \ - volatile unsigned long _res; \ - /* _argvec[0] holds current r2 across the call */ \ - _argvec[1] = (unsigned long)_orig.r2; \ - _argvec[2] = (unsigned long)_orig.nraddr; \ - _argvec[2+1] = (unsigned long)arg1; \ - _argvec[2+2] = (unsigned long)arg2; \ - _argvec[2+3] = (unsigned long)arg3; \ - __asm__ volatile( \ - "mr 11,%1\n\t" \ - VG_EXPAND_FRAME_BY_trashes_r3(512) \ - "std 2,-16(11)\n\t" /* save tocptr */ \ - "ld 2,-8(11)\n\t" /* use nraddr's tocptr */ \ - "ld 3, 8(11)\n\t" /* arg1->r3 */ \ - "ld 4, 16(11)\n\t" /* arg2->r4 */ \ - "ld 5, 24(11)\n\t" /* arg3->r5 */ \ - "ld 11, 0(11)\n\t" /* target->r11 */ \ - VALGRIND_BRANCH_AND_LINK_TO_NOREDIR_R11 \ - "mr 11,%1\n\t" \ - "mr %0,3\n\t" \ - "ld 2,-16(11)\n\t" /* restore tocptr */ \ - VG_CONTRACT_FRAME_BY(512) \ - : /*out*/ "=r" (_res) \ - : /*in*/ "r" (&_argvec[2]) \ - : /*trash*/ "cc", "memory", __CALLER_SAVED_REGS \ - ); \ - lval = (__typeof__(lval)) _res; \ - } while (0) - -#define CALL_FN_W_WWWW(lval, orig, arg1,arg2,arg3,arg4) \ - do { \ - volatile OrigFn _orig = (orig); \ - volatile unsigned long _argvec[3+4]; \ - volatile unsigned long _res; \ - /* _argvec[0] holds current r2 across the call */ \ - _argvec[1] = (unsigned long)_orig.r2; \ - _argvec[2] = (unsigned long)_orig.nraddr; \ - _argvec[2+1] = (unsigned long)arg1; \ - _argvec[2+2] = (unsigned long)arg2; \ - _argvec[2+3] = (unsigned long)arg3; \ - _argvec[2+4] = (unsigned long)arg4; \ - __asm__ volatile( \ - "mr 11,%1\n\t" \ - VG_EXPAND_FRAME_BY_trashes_r3(512) \ - "std 2,-16(11)\n\t" /* save tocptr */ \ - "ld 2,-8(11)\n\t" /* use nraddr's tocptr */ \ - "ld 3, 8(11)\n\t" /* arg1->r3 */ \ - "ld 4, 16(11)\n\t" /* arg2->r4 */ \ - "ld 5, 24(11)\n\t" /* arg3->r5 */ \ - "ld 6, 32(11)\n\t" /* arg4->r6 */ \ - "ld 11, 0(11)\n\t" /* target->r11 */ \ - VALGRIND_BRANCH_AND_LINK_TO_NOREDIR_R11 \ - "mr 11,%1\n\t" \ - "mr %0,3\n\t" \ - "ld 2,-16(11)\n\t" /* restore tocptr */ \ - VG_CONTRACT_FRAME_BY(512) \ - : /*out*/ "=r" (_res) \ - : /*in*/ "r" (&_argvec[2]) \ - : /*trash*/ "cc", "memory", __CALLER_SAVED_REGS \ - ); \ - lval = (__typeof__(lval)) _res; \ - } while (0) - -#define CALL_FN_W_5W(lval, orig, arg1,arg2,arg3,arg4,arg5) \ - do { \ - volatile OrigFn _orig = (orig); \ - volatile unsigned long _argvec[3+5]; \ - volatile unsigned long _res; \ - /* _argvec[0] holds current r2 across the call */ \ - _argvec[1] = (unsigned long)_orig.r2; \ - _argvec[2] = (unsigned long)_orig.nraddr; \ - _argvec[2+1] = (unsigned long)arg1; \ - _argvec[2+2] = (unsigned long)arg2; \ - _argvec[2+3] = (unsigned long)arg3; \ - _argvec[2+4] = (unsigned long)arg4; \ - _argvec[2+5] = (unsigned long)arg5; \ - __asm__ volatile( \ - "mr 11,%1\n\t" \ - VG_EXPAND_FRAME_BY_trashes_r3(512) \ - "std 2,-16(11)\n\t" /* save tocptr */ \ - "ld 2,-8(11)\n\t" /* use nraddr's tocptr */ \ - "ld 3, 8(11)\n\t" /* arg1->r3 */ \ - "ld 4, 16(11)\n\t" /* arg2->r4 */ \ - "ld 5, 24(11)\n\t" /* arg3->r5 */ \ - "ld 6, 32(11)\n\t" /* arg4->r6 */ \ - "ld 7, 40(11)\n\t" /* arg5->r7 */ \ - "ld 11, 0(11)\n\t" /* target->r11 */ \ - VALGRIND_BRANCH_AND_LINK_TO_NOREDIR_R11 \ - "mr 11,%1\n\t" \ - "mr %0,3\n\t" \ - "ld 2,-16(11)\n\t" /* restore tocptr */ \ - VG_CONTRACT_FRAME_BY(512) \ - : /*out*/ "=r" (_res) \ - : /*in*/ "r" (&_argvec[2]) \ - : /*trash*/ "cc", "memory", __CALLER_SAVED_REGS \ - ); \ - lval = (__typeof__(lval)) _res; \ - } while (0) - -#define CALL_FN_W_6W(lval, orig, arg1,arg2,arg3,arg4,arg5,arg6) \ - do { \ - volatile OrigFn _orig = (orig); \ - volatile unsigned long _argvec[3+6]; \ - volatile unsigned long _res; \ - /* _argvec[0] holds current r2 across the call */ \ - _argvec[1] = (unsigned long)_orig.r2; \ - _argvec[2] = (unsigned long)_orig.nraddr; \ - _argvec[2+1] = (unsigned long)arg1; \ - _argvec[2+2] = (unsigned long)arg2; \ - _argvec[2+3] = (unsigned long)arg3; \ - _argvec[2+4] = (unsigned long)arg4; \ - _argvec[2+5] = (unsigned long)arg5; \ - _argvec[2+6] = (unsigned long)arg6; \ - __asm__ volatile( \ - "mr 11,%1\n\t" \ - VG_EXPAND_FRAME_BY_trashes_r3(512) \ - "std 2,-16(11)\n\t" /* save tocptr */ \ - "ld 2,-8(11)\n\t" /* use nraddr's tocptr */ \ - "ld 3, 8(11)\n\t" /* arg1->r3 */ \ - "ld 4, 16(11)\n\t" /* arg2->r4 */ \ - "ld 5, 24(11)\n\t" /* arg3->r5 */ \ - "ld 6, 32(11)\n\t" /* arg4->r6 */ \ - "ld 7, 40(11)\n\t" /* arg5->r7 */ \ - "ld 8, 48(11)\n\t" /* arg6->r8 */ \ - "ld 11, 0(11)\n\t" /* target->r11 */ \ - VALGRIND_BRANCH_AND_LINK_TO_NOREDIR_R11 \ - "mr 11,%1\n\t" \ - "mr %0,3\n\t" \ - "ld 2,-16(11)\n\t" /* restore tocptr */ \ - VG_CONTRACT_FRAME_BY(512) \ - : /*out*/ "=r" (_res) \ - : /*in*/ "r" (&_argvec[2]) \ - : /*trash*/ "cc", "memory", __CALLER_SAVED_REGS \ - ); \ - lval = (__typeof__(lval)) _res; \ - } while (0) - -#define CALL_FN_W_7W(lval, orig, arg1,arg2,arg3,arg4,arg5,arg6, \ - arg7) \ - do { \ - volatile OrigFn _orig = (orig); \ - volatile unsigned long _argvec[3+7]; \ - volatile unsigned long _res; \ - /* _argvec[0] holds current r2 across the call */ \ - _argvec[1] = (unsigned long)_orig.r2; \ - _argvec[2] = (unsigned long)_orig.nraddr; \ - _argvec[2+1] = (unsigned long)arg1; \ - _argvec[2+2] = (unsigned long)arg2; \ - _argvec[2+3] = (unsigned long)arg3; \ - _argvec[2+4] = (unsigned long)arg4; \ - _argvec[2+5] = (unsigned long)arg5; \ - _argvec[2+6] = (unsigned long)arg6; \ - _argvec[2+7] = (unsigned long)arg7; \ - __asm__ volatile( \ - "mr 11,%1\n\t" \ - VG_EXPAND_FRAME_BY_trashes_r3(512) \ - "std 2,-16(11)\n\t" /* save tocptr */ \ - "ld 2,-8(11)\n\t" /* use nraddr's tocptr */ \ - "ld 3, 8(11)\n\t" /* arg1->r3 */ \ - "ld 4, 16(11)\n\t" /* arg2->r4 */ \ - "ld 5, 24(11)\n\t" /* arg3->r5 */ \ - "ld 6, 32(11)\n\t" /* arg4->r6 */ \ - "ld 7, 40(11)\n\t" /* arg5->r7 */ \ - "ld 8, 48(11)\n\t" /* arg6->r8 */ \ - "ld 9, 56(11)\n\t" /* arg7->r9 */ \ - "ld 11, 0(11)\n\t" /* target->r11 */ \ - VALGRIND_BRANCH_AND_LINK_TO_NOREDIR_R11 \ - "mr 11,%1\n\t" \ - "mr %0,3\n\t" \ - "ld 2,-16(11)\n\t" /* restore tocptr */ \ - VG_CONTRACT_FRAME_BY(512) \ - : /*out*/ "=r" (_res) \ - : /*in*/ "r" (&_argvec[2]) \ - : /*trash*/ "cc", "memory", __CALLER_SAVED_REGS \ - ); \ - lval = (__typeof__(lval)) _res; \ - } while (0) - -#define CALL_FN_W_8W(lval, orig, arg1,arg2,arg3,arg4,arg5,arg6, \ - arg7,arg8) \ - do { \ - volatile OrigFn _orig = (orig); \ - volatile unsigned long _argvec[3+8]; \ - volatile unsigned long _res; \ - /* _argvec[0] holds current r2 across the call */ \ - _argvec[1] = (unsigned long)_orig.r2; \ - _argvec[2] = (unsigned long)_orig.nraddr; \ - _argvec[2+1] = (unsigned long)arg1; \ - _argvec[2+2] = (unsigned long)arg2; \ - _argvec[2+3] = (unsigned long)arg3; \ - _argvec[2+4] = (unsigned long)arg4; \ - _argvec[2+5] = (unsigned long)arg5; \ - _argvec[2+6] = (unsigned long)arg6; \ - _argvec[2+7] = (unsigned long)arg7; \ - _argvec[2+8] = (unsigned long)arg8; \ - __asm__ volatile( \ - "mr 11,%1\n\t" \ - VG_EXPAND_FRAME_BY_trashes_r3(512) \ - "std 2,-16(11)\n\t" /* save tocptr */ \ - "ld 2,-8(11)\n\t" /* use nraddr's tocptr */ \ - "ld 3, 8(11)\n\t" /* arg1->r3 */ \ - "ld 4, 16(11)\n\t" /* arg2->r4 */ \ - "ld 5, 24(11)\n\t" /* arg3->r5 */ \ - "ld 6, 32(11)\n\t" /* arg4->r6 */ \ - "ld 7, 40(11)\n\t" /* arg5->r7 */ \ - "ld 8, 48(11)\n\t" /* arg6->r8 */ \ - "ld 9, 56(11)\n\t" /* arg7->r9 */ \ - "ld 10, 64(11)\n\t" /* arg8->r10 */ \ - "ld 11, 0(11)\n\t" /* target->r11 */ \ - VALGRIND_BRANCH_AND_LINK_TO_NOREDIR_R11 \ - "mr 11,%1\n\t" \ - "mr %0,3\n\t" \ - "ld 2,-16(11)\n\t" /* restore tocptr */ \ - VG_CONTRACT_FRAME_BY(512) \ - : /*out*/ "=r" (_res) \ - : /*in*/ "r" (&_argvec[2]) \ - : /*trash*/ "cc", "memory", __CALLER_SAVED_REGS \ - ); \ - lval = (__typeof__(lval)) _res; \ - } while (0) - -#define CALL_FN_W_9W(lval, orig, arg1,arg2,arg3,arg4,arg5,arg6, \ - arg7,arg8,arg9) \ - do { \ - volatile OrigFn _orig = (orig); \ - volatile unsigned long _argvec[3+9]; \ - volatile unsigned long _res; \ - /* _argvec[0] holds current r2 across the call */ \ - _argvec[1] = (unsigned long)_orig.r2; \ - _argvec[2] = (unsigned long)_orig.nraddr; \ - _argvec[2+1] = (unsigned long)arg1; \ - _argvec[2+2] = (unsigned long)arg2; \ - _argvec[2+3] = (unsigned long)arg3; \ - _argvec[2+4] = (unsigned long)arg4; \ - _argvec[2+5] = (unsigned long)arg5; \ - _argvec[2+6] = (unsigned long)arg6; \ - _argvec[2+7] = (unsigned long)arg7; \ - _argvec[2+8] = (unsigned long)arg8; \ - _argvec[2+9] = (unsigned long)arg9; \ - __asm__ volatile( \ - "mr 11,%1\n\t" \ - VG_EXPAND_FRAME_BY_trashes_r3(512) \ - "std 2,-16(11)\n\t" /* save tocptr */ \ - "ld 2,-8(11)\n\t" /* use nraddr's tocptr */ \ - VG_EXPAND_FRAME_BY_trashes_r3(128) \ - /* arg9 */ \ - "ld 3,72(11)\n\t" \ - "std 3,112(1)\n\t" \ - /* args1-8 */ \ - "ld 3, 8(11)\n\t" /* arg1->r3 */ \ - "ld 4, 16(11)\n\t" /* arg2->r4 */ \ - "ld 5, 24(11)\n\t" /* arg3->r5 */ \ - "ld 6, 32(11)\n\t" /* arg4->r6 */ \ - "ld 7, 40(11)\n\t" /* arg5->r7 */ \ - "ld 8, 48(11)\n\t" /* arg6->r8 */ \ - "ld 9, 56(11)\n\t" /* arg7->r9 */ \ - "ld 10, 64(11)\n\t" /* arg8->r10 */ \ - "ld 11, 0(11)\n\t" /* target->r11 */ \ - VALGRIND_BRANCH_AND_LINK_TO_NOREDIR_R11 \ - "mr 11,%1\n\t" \ - "mr %0,3\n\t" \ - "ld 2,-16(11)\n\t" /* restore tocptr */ \ - VG_CONTRACT_FRAME_BY(128) \ - VG_CONTRACT_FRAME_BY(512) \ - : /*out*/ "=r" (_res) \ - : /*in*/ "r" (&_argvec[2]) \ - : /*trash*/ "cc", "memory", __CALLER_SAVED_REGS \ - ); \ - lval = (__typeof__(lval)) _res; \ - } while (0) - -#define CALL_FN_W_10W(lval, orig, arg1,arg2,arg3,arg4,arg5,arg6, \ - arg7,arg8,arg9,arg10) \ - do { \ - volatile OrigFn _orig = (orig); \ - volatile unsigned long _argvec[3+10]; \ - volatile unsigned long _res; \ - /* _argvec[0] holds current r2 across the call */ \ - _argvec[1] = (unsigned long)_orig.r2; \ - _argvec[2] = (unsigned long)_orig.nraddr; \ - _argvec[2+1] = (unsigned long)arg1; \ - _argvec[2+2] = (unsigned long)arg2; \ - _argvec[2+3] = (unsigned long)arg3; \ - _argvec[2+4] = (unsigned long)arg4; \ - _argvec[2+5] = (unsigned long)arg5; \ - _argvec[2+6] = (unsigned long)arg6; \ - _argvec[2+7] = (unsigned long)arg7; \ - _argvec[2+8] = (unsigned long)arg8; \ - _argvec[2+9] = (unsigned long)arg9; \ - _argvec[2+10] = (unsigned long)arg10; \ - __asm__ volatile( \ - "mr 11,%1\n\t" \ - VG_EXPAND_FRAME_BY_trashes_r3(512) \ - "std 2,-16(11)\n\t" /* save tocptr */ \ - "ld 2,-8(11)\n\t" /* use nraddr's tocptr */ \ - VG_EXPAND_FRAME_BY_trashes_r3(128) \ - /* arg10 */ \ - "ld 3,80(11)\n\t" \ - "std 3,120(1)\n\t" \ - /* arg9 */ \ - "ld 3,72(11)\n\t" \ - "std 3,112(1)\n\t" \ - /* args1-8 */ \ - "ld 3, 8(11)\n\t" /* arg1->r3 */ \ - "ld 4, 16(11)\n\t" /* arg2->r4 */ \ - "ld 5, 24(11)\n\t" /* arg3->r5 */ \ - "ld 6, 32(11)\n\t" /* arg4->r6 */ \ - "ld 7, 40(11)\n\t" /* arg5->r7 */ \ - "ld 8, 48(11)\n\t" /* arg6->r8 */ \ - "ld 9, 56(11)\n\t" /* arg7->r9 */ \ - "ld 10, 64(11)\n\t" /* arg8->r10 */ \ - "ld 11, 0(11)\n\t" /* target->r11 */ \ - VALGRIND_BRANCH_AND_LINK_TO_NOREDIR_R11 \ - "mr 11,%1\n\t" \ - "mr %0,3\n\t" \ - "ld 2,-16(11)\n\t" /* restore tocptr */ \ - VG_CONTRACT_FRAME_BY(128) \ - VG_CONTRACT_FRAME_BY(512) \ - : /*out*/ "=r" (_res) \ - : /*in*/ "r" (&_argvec[2]) \ - : /*trash*/ "cc", "memory", __CALLER_SAVED_REGS \ - ); \ - lval = (__typeof__(lval)) _res; \ - } while (0) - -#define CALL_FN_W_11W(lval, orig, arg1,arg2,arg3,arg4,arg5,arg6, \ - arg7,arg8,arg9,arg10,arg11) \ - do { \ - volatile OrigFn _orig = (orig); \ - volatile unsigned long _argvec[3+11]; \ - volatile unsigned long _res; \ - /* _argvec[0] holds current r2 across the call */ \ - _argvec[1] = (unsigned long)_orig.r2; \ - _argvec[2] = (unsigned long)_orig.nraddr; \ - _argvec[2+1] = (unsigned long)arg1; \ - _argvec[2+2] = (unsigned long)arg2; \ - _argvec[2+3] = (unsigned long)arg3; \ - _argvec[2+4] = (unsigned long)arg4; \ - _argvec[2+5] = (unsigned long)arg5; \ - _argvec[2+6] = (unsigned long)arg6; \ - _argvec[2+7] = (unsigned long)arg7; \ - _argvec[2+8] = (unsigned long)arg8; \ - _argvec[2+9] = (unsigned long)arg9; \ - _argvec[2+10] = (unsigned long)arg10; \ - _argvec[2+11] = (unsigned long)arg11; \ - __asm__ volatile( \ - "mr 11,%1\n\t" \ - VG_EXPAND_FRAME_BY_trashes_r3(512) \ - "std 2,-16(11)\n\t" /* save tocptr */ \ - "ld 2,-8(11)\n\t" /* use nraddr's tocptr */ \ - VG_EXPAND_FRAME_BY_trashes_r3(144) \ - /* arg11 */ \ - "ld 3,88(11)\n\t" \ - "std 3,128(1)\n\t" \ - /* arg10 */ \ - "ld 3,80(11)\n\t" \ - "std 3,120(1)\n\t" \ - /* arg9 */ \ - "ld 3,72(11)\n\t" \ - "std 3,112(1)\n\t" \ - /* args1-8 */ \ - "ld 3, 8(11)\n\t" /* arg1->r3 */ \ - "ld 4, 16(11)\n\t" /* arg2->r4 */ \ - "ld 5, 24(11)\n\t" /* arg3->r5 */ \ - "ld 6, 32(11)\n\t" /* arg4->r6 */ \ - "ld 7, 40(11)\n\t" /* arg5->r7 */ \ - "ld 8, 48(11)\n\t" /* arg6->r8 */ \ - "ld 9, 56(11)\n\t" /* arg7->r9 */ \ - "ld 10, 64(11)\n\t" /* arg8->r10 */ \ - "ld 11, 0(11)\n\t" /* target->r11 */ \ - VALGRIND_BRANCH_AND_LINK_TO_NOREDIR_R11 \ - "mr 11,%1\n\t" \ - "mr %0,3\n\t" \ - "ld 2,-16(11)\n\t" /* restore tocptr */ \ - VG_CONTRACT_FRAME_BY(144) \ - VG_CONTRACT_FRAME_BY(512) \ - : /*out*/ "=r" (_res) \ - : /*in*/ "r" (&_argvec[2]) \ - : /*trash*/ "cc", "memory", __CALLER_SAVED_REGS \ - ); \ - lval = (__typeof__(lval)) _res; \ - } while (0) - -#define CALL_FN_W_12W(lval, orig, arg1,arg2,arg3,arg4,arg5,arg6, \ - arg7,arg8,arg9,arg10,arg11,arg12) \ - do { \ - volatile OrigFn _orig = (orig); \ - volatile unsigned long _argvec[3+12]; \ - volatile unsigned long _res; \ - /* _argvec[0] holds current r2 across the call */ \ - _argvec[1] = (unsigned long)_orig.r2; \ - _argvec[2] = (unsigned long)_orig.nraddr; \ - _argvec[2+1] = (unsigned long)arg1; \ - _argvec[2+2] = (unsigned long)arg2; \ - _argvec[2+3] = (unsigned long)arg3; \ - _argvec[2+4] = (unsigned long)arg4; \ - _argvec[2+5] = (unsigned long)arg5; \ - _argvec[2+6] = (unsigned long)arg6; \ - _argvec[2+7] = (unsigned long)arg7; \ - _argvec[2+8] = (unsigned long)arg8; \ - _argvec[2+9] = (unsigned long)arg9; \ - _argvec[2+10] = (unsigned long)arg10; \ - _argvec[2+11] = (unsigned long)arg11; \ - _argvec[2+12] = (unsigned long)arg12; \ - __asm__ volatile( \ - "mr 11,%1\n\t" \ - VG_EXPAND_FRAME_BY_trashes_r3(512) \ - "std 2,-16(11)\n\t" /* save tocptr */ \ - "ld 2,-8(11)\n\t" /* use nraddr's tocptr */ \ - VG_EXPAND_FRAME_BY_trashes_r3(144) \ - /* arg12 */ \ - "ld 3,96(11)\n\t" \ - "std 3,136(1)\n\t" \ - /* arg11 */ \ - "ld 3,88(11)\n\t" \ - "std 3,128(1)\n\t" \ - /* arg10 */ \ - "ld 3,80(11)\n\t" \ - "std 3,120(1)\n\t" \ - /* arg9 */ \ - "ld 3,72(11)\n\t" \ - "std 3,112(1)\n\t" \ - /* args1-8 */ \ - "ld 3, 8(11)\n\t" /* arg1->r3 */ \ - "ld 4, 16(11)\n\t" /* arg2->r4 */ \ - "ld 5, 24(11)\n\t" /* arg3->r5 */ \ - "ld 6, 32(11)\n\t" /* arg4->r6 */ \ - "ld 7, 40(11)\n\t" /* arg5->r7 */ \ - "ld 8, 48(11)\n\t" /* arg6->r8 */ \ - "ld 9, 56(11)\n\t" /* arg7->r9 */ \ - "ld 10, 64(11)\n\t" /* arg8->r10 */ \ - "ld 11, 0(11)\n\t" /* target->r11 */ \ - VALGRIND_BRANCH_AND_LINK_TO_NOREDIR_R11 \ - "mr 11,%1\n\t" \ - "mr %0,3\n\t" \ - "ld 2,-16(11)\n\t" /* restore tocptr */ \ - VG_CONTRACT_FRAME_BY(144) \ - VG_CONTRACT_FRAME_BY(512) \ - : /*out*/ "=r" (_res) \ - : /*in*/ "r" (&_argvec[2]) \ - : /*trash*/ "cc", "memory", __CALLER_SAVED_REGS \ - ); \ - lval = (__typeof__(lval)) _res; \ - } while (0) - -#endif /* PLAT_ppc64_aix5 */ - - -/* ------------------------------------------------------------------ */ -/* ARCHITECTURE INDEPENDENT MACROS for CLIENT REQUESTS. */ -/* */ -/* ------------------------------------------------------------------ */ - -/* Some request codes. There are many more of these, but most are not - exposed to end-user view. These are the public ones, all of the - form 0x1000 + small_number. - - Core ones are in the range 0x00000000--0x0000ffff. The non-public - ones start at 0x2000. -*/ - -/* These macros are used by tools -- they must be public, but don't - embed them into other programs. */ -#define VG_USERREQ_TOOL_BASE(a,b) \ - ((unsigned int)(((a)&0xff) << 24 | ((b)&0xff) << 16)) -#define VG_IS_TOOL_USERREQ(a, b, v) \ - (VG_USERREQ_TOOL_BASE(a,b) == ((v) & 0xffff0000)) - -/* !! ABIWARNING !! ABIWARNING !! ABIWARNING !! ABIWARNING !! - This enum comprises an ABI exported by Valgrind to programs - which use client requests. DO NOT CHANGE THE ORDER OF THESE - ENTRIES, NOR DELETE ANY -- add new ones at the end. */ -typedef - enum { VG_USERREQ__RUNNING_ON_VALGRIND = 0x1001, - VG_USERREQ__DISCARD_TRANSLATIONS = 0x1002, - - /* These allow any function to be called from the simulated - CPU but run on the real CPU. Nb: the first arg passed to - the function is always the ThreadId of the running - thread! So CLIENT_CALL0 actually requires a 1 arg - function, etc. */ - VG_USERREQ__CLIENT_CALL0 = 0x1101, - VG_USERREQ__CLIENT_CALL1 = 0x1102, - VG_USERREQ__CLIENT_CALL2 = 0x1103, - VG_USERREQ__CLIENT_CALL3 = 0x1104, - - /* Can be useful in regression testing suites -- eg. can - send Valgrind's output to /dev/null and still count - errors. */ - VG_USERREQ__COUNT_ERRORS = 0x1201, - - /* These are useful and can be interpreted by any tool that - tracks malloc() et al, by using vg_replace_malloc.c. */ - VG_USERREQ__MALLOCLIKE_BLOCK = 0x1301, - VG_USERREQ__FREELIKE_BLOCK = 0x1302, - /* Memory pool support. */ - VG_USERREQ__CREATE_MEMPOOL = 0x1303, - VG_USERREQ__DESTROY_MEMPOOL = 0x1304, - VG_USERREQ__MEMPOOL_ALLOC = 0x1305, - VG_USERREQ__MEMPOOL_FREE = 0x1306, - VG_USERREQ__MEMPOOL_TRIM = 0x1307, - VG_USERREQ__MOVE_MEMPOOL = 0x1308, - VG_USERREQ__MEMPOOL_CHANGE = 0x1309, - VG_USERREQ__MEMPOOL_EXISTS = 0x130a, - - /* Allow printfs to valgrind log. */ - VG_USERREQ__PRINTF = 0x1401, - VG_USERREQ__PRINTF_BACKTRACE = 0x1402, - - /* Stack support. */ - VG_USERREQ__STACK_REGISTER = 0x1501, - VG_USERREQ__STACK_DEREGISTER = 0x1502, - VG_USERREQ__STACK_CHANGE = 0x1503 - } Vg_ClientRequest; - -#if !defined(__GNUC__) -# define __extension__ /* */ -#endif - -/* Returns the number of Valgrinds this code is running under. That - is, 0 if running natively, 1 if running under Valgrind, 2 if - running under Valgrind which is running under another Valgrind, - etc. */ -#define RUNNING_ON_VALGRIND __extension__ \ - ({unsigned int _qzz_res; \ - VALGRIND_DO_CLIENT_REQUEST(_qzz_res, 0 /* if not */, \ - VG_USERREQ__RUNNING_ON_VALGRIND, \ - 0, 0, 0, 0, 0); \ - _qzz_res; \ - }) - - -/* Discard translation of code in the range [_qzz_addr .. _qzz_addr + - _qzz_len - 1]. Useful if you are debugging a JITter or some such, - since it provides a way to make sure valgrind will retranslate the - invalidated area. Returns no value. */ -#define VALGRIND_DISCARD_TRANSLATIONS(_qzz_addr,_qzz_len) \ - {unsigned int _qzz_res; \ - VALGRIND_DO_CLIENT_REQUEST(_qzz_res, 0, \ - VG_USERREQ__DISCARD_TRANSLATIONS, \ - _qzz_addr, _qzz_len, 0, 0, 0); \ - } - - -/* These requests are for getting Valgrind itself to print something. - Possibly with a backtrace. This is a really ugly hack. */ - -#if defined(NVALGRIND) - -# define VALGRIND_PRINTF(...) -# define VALGRIND_PRINTF_BACKTRACE(...) - -#else /* NVALGRIND */ - -/* Modern GCC will optimize the static routine out if unused, - and unused attribute will shut down warnings about it. */ -static int VALGRIND_PRINTF(const char *format, ...) - __attribute__((format(__printf__, 1, 2), __unused__)); -static int -VALGRIND_PRINTF(const char *format, ...) -{ - unsigned long _qzz_res; - va_list vargs; - va_start(vargs, format); - VALGRIND_DO_CLIENT_REQUEST(_qzz_res, 0, VG_USERREQ__PRINTF, - (unsigned long)format, (unsigned long)vargs, - 0, 0, 0); - va_end(vargs); - return (int)_qzz_res; -} - -static int VALGRIND_PRINTF_BACKTRACE(const char *format, ...) - __attribute__((format(__printf__, 1, 2), __unused__)); -static int -VALGRIND_PRINTF_BACKTRACE(const char *format, ...) -{ - unsigned long _qzz_res; - va_list vargs; - va_start(vargs, format); - VALGRIND_DO_CLIENT_REQUEST(_qzz_res, 0, VG_USERREQ__PRINTF_BACKTRACE, - (unsigned long)format, (unsigned long)vargs, - 0, 0, 0); - va_end(vargs); - return (int)_qzz_res; -} - -#endif /* NVALGRIND */ - - -/* These requests allow control to move from the simulated CPU to the - real CPU, calling an arbitary function. - - Note that the current ThreadId is inserted as the first argument. - So this call: - - VALGRIND_NON_SIMD_CALL2(f, arg1, arg2) - - requires f to have this signature: - - Word f(Word tid, Word arg1, Word arg2) - - where "Word" is a word-sized type. - - Note that these client requests are not entirely reliable. For example, - if you call a function with them that subsequently calls printf(), - there's a high chance Valgrind will crash. Generally, your prospects of - these working are made higher if the called function does not refer to - any global variables, and does not refer to any libc or other functions - (printf et al). Any kind of entanglement with libc or dynamic linking is - likely to have a bad outcome, for tricky reasons which we've grappled - with a lot in the past. -*/ -#define VALGRIND_NON_SIMD_CALL0(_qyy_fn) \ - __extension__ \ - ({unsigned long _qyy_res; \ - VALGRIND_DO_CLIENT_REQUEST(_qyy_res, 0 /* default return */, \ - VG_USERREQ__CLIENT_CALL0, \ - _qyy_fn, \ - 0, 0, 0, 0); \ - _qyy_res; \ - }) - -#define VALGRIND_NON_SIMD_CALL1(_qyy_fn, _qyy_arg1) \ - __extension__ \ - ({unsigned long _qyy_res; \ - VALGRIND_DO_CLIENT_REQUEST(_qyy_res, 0 /* default return */, \ - VG_USERREQ__CLIENT_CALL1, \ - _qyy_fn, \ - _qyy_arg1, 0, 0, 0); \ - _qyy_res; \ - }) - -#define VALGRIND_NON_SIMD_CALL2(_qyy_fn, _qyy_arg1, _qyy_arg2) \ - __extension__ \ - ({unsigned long _qyy_res; \ - VALGRIND_DO_CLIENT_REQUEST(_qyy_res, 0 /* default return */, \ - VG_USERREQ__CLIENT_CALL2, \ - _qyy_fn, \ - _qyy_arg1, _qyy_arg2, 0, 0); \ - _qyy_res; \ - }) - -#define VALGRIND_NON_SIMD_CALL3(_qyy_fn, _qyy_arg1, _qyy_arg2, _qyy_arg3) \ - __extension__ \ - ({unsigned long _qyy_res; \ - VALGRIND_DO_CLIENT_REQUEST(_qyy_res, 0 /* default return */, \ - VG_USERREQ__CLIENT_CALL3, \ - _qyy_fn, \ - _qyy_arg1, _qyy_arg2, \ - _qyy_arg3, 0); \ - _qyy_res; \ - }) - - -/* Counts the number of errors that have been recorded by a tool. Nb: - the tool must record the errors with VG_(maybe_record_error)() or - VG_(unique_error)() for them to be counted. */ -#define VALGRIND_COUNT_ERRORS \ - __extension__ \ - ({unsigned int _qyy_res; \ - VALGRIND_DO_CLIENT_REQUEST(_qyy_res, 0 /* default return */, \ - VG_USERREQ__COUNT_ERRORS, \ - 0, 0, 0, 0, 0); \ - _qyy_res; \ - }) - -/* Mark a block of memory as having been allocated by a malloc()-like - function. `addr' is the start of the usable block (ie. after any - redzone) `rzB' is redzone size if the allocator can apply redzones; - use '0' if not. Adding redzones makes it more likely Valgrind will spot - block overruns. `is_zeroed' indicates if the memory is zeroed, as it is - for calloc(). Put it immediately after the point where a block is - allocated. - - If you're using Memcheck: If you're allocating memory via superblocks, - and then handing out small chunks of each superblock, if you don't have - redzones on your small blocks, it's worth marking the superblock with - VALGRIND_MAKE_MEM_NOACCESS when it's created, so that block overruns are - detected. But if you can put redzones on, it's probably better to not do - this, so that messages for small overruns are described in terms of the - small block rather than the superblock (but if you have a big overrun - that skips over a redzone, you could miss an error this way). See - memcheck/tests/custom_alloc.c for an example. - - WARNING: if your allocator uses malloc() or 'new' to allocate - superblocks, rather than mmap() or brk(), this will not work properly -- - you'll likely get assertion failures during leak detection. This is - because Valgrind doesn't like seeing overlapping heap blocks. Sorry. - - Nb: block must be freed via a free()-like function specified - with VALGRIND_FREELIKE_BLOCK or mismatch errors will occur. */ -#define VALGRIND_MALLOCLIKE_BLOCK(addr, sizeB, rzB, is_zeroed) \ - {unsigned int _qzz_res; \ - VALGRIND_DO_CLIENT_REQUEST(_qzz_res, 0, \ - VG_USERREQ__MALLOCLIKE_BLOCK, \ - addr, sizeB, rzB, is_zeroed, 0); \ - } - -/* Mark a block of memory as having been freed by a free()-like function. - `rzB' is redzone size; it must match that given to - VALGRIND_MALLOCLIKE_BLOCK. Memory not freed will be detected by the leak - checker. Put it immediately after the point where the block is freed. */ -#define VALGRIND_FREELIKE_BLOCK(addr, rzB) \ - {unsigned int _qzz_res; \ - VALGRIND_DO_CLIENT_REQUEST(_qzz_res, 0, \ - VG_USERREQ__FREELIKE_BLOCK, \ - addr, rzB, 0, 0, 0); \ - } - -/* Create a memory pool. */ -#define VALGRIND_CREATE_MEMPOOL(pool, rzB, is_zeroed) \ - {unsigned int _qzz_res; \ - VALGRIND_DO_CLIENT_REQUEST(_qzz_res, 0, \ - VG_USERREQ__CREATE_MEMPOOL, \ - pool, rzB, is_zeroed, 0, 0); \ - } - -/* Destroy a memory pool. */ -#define VALGRIND_DESTROY_MEMPOOL(pool) \ - {unsigned int _qzz_res; \ - VALGRIND_DO_CLIENT_REQUEST(_qzz_res, 0, \ - VG_USERREQ__DESTROY_MEMPOOL, \ - pool, 0, 0, 0, 0); \ - } - -/* Associate a piece of memory with a memory pool. */ -#define VALGRIND_MEMPOOL_ALLOC(pool, addr, size) \ - {unsigned int _qzz_res; \ - VALGRIND_DO_CLIENT_REQUEST(_qzz_res, 0, \ - VG_USERREQ__MEMPOOL_ALLOC, \ - pool, addr, size, 0, 0); \ - } - -/* Disassociate a piece of memory from a memory pool. */ -#define VALGRIND_MEMPOOL_FREE(pool, addr) \ - {unsigned int _qzz_res; \ - VALGRIND_DO_CLIENT_REQUEST(_qzz_res, 0, \ - VG_USERREQ__MEMPOOL_FREE, \ - pool, addr, 0, 0, 0); \ - } - -/* Disassociate any pieces outside a particular range. */ -#define VALGRIND_MEMPOOL_TRIM(pool, addr, size) \ - {unsigned int _qzz_res; \ - VALGRIND_DO_CLIENT_REQUEST(_qzz_res, 0, \ - VG_USERREQ__MEMPOOL_TRIM, \ - pool, addr, size, 0, 0); \ - } - -/* Resize and/or move a piece associated with a memory pool. */ -#define VALGRIND_MOVE_MEMPOOL(poolA, poolB) \ - {unsigned int _qzz_res; \ - VALGRIND_DO_CLIENT_REQUEST(_qzz_res, 0, \ - VG_USERREQ__MOVE_MEMPOOL, \ - poolA, poolB, 0, 0, 0); \ - } - -/* Resize and/or move a piece associated with a memory pool. */ -#define VALGRIND_MEMPOOL_CHANGE(pool, addrA, addrB, size) \ - {unsigned int _qzz_res; \ - VALGRIND_DO_CLIENT_REQUEST(_qzz_res, 0, \ - VG_USERREQ__MEMPOOL_CHANGE, \ - pool, addrA, addrB, size, 0); \ - } - -/* Return 1 if a mempool exists, else 0. */ -#define VALGRIND_MEMPOOL_EXISTS(pool) \ - ({unsigned int _qzz_res; \ - VALGRIND_DO_CLIENT_REQUEST(_qzz_res, 0, \ - VG_USERREQ__MEMPOOL_EXISTS, \ - pool, 0, 0, 0, 0); \ - _qzz_res; \ - }) - -/* Mark a piece of memory as being a stack. Returns a stack id. */ -#define VALGRIND_STACK_REGISTER(start, end) \ - ({unsigned int _qzz_res; \ - VALGRIND_DO_CLIENT_REQUEST(_qzz_res, 0, \ - VG_USERREQ__STACK_REGISTER, \ - start, end, 0, 0, 0); \ - _qzz_res; \ - }) - -/* Unmark the piece of memory associated with a stack id as being a - stack. */ -#define VALGRIND_STACK_DEREGISTER(id) \ - {unsigned int _qzz_res; \ - VALGRIND_DO_CLIENT_REQUEST(_qzz_res, 0, \ - VG_USERREQ__STACK_DEREGISTER, \ - id, 0, 0, 0, 0); \ - } - -/* Change the start and end address of the stack id. */ -#define VALGRIND_STACK_CHANGE(id, start, end) \ - {unsigned int _qzz_res; \ - VALGRIND_DO_CLIENT_REQUEST(_qzz_res, 0, \ - VG_USERREQ__STACK_CHANGE, \ - id, start, end, 0, 0); \ - } - - -#undef PLAT_x86_linux -#undef PLAT_amd64_linux -#undef PLAT_ppc32_linux -#undef PLAT_ppc64_linux -#undef PLAT_ppc32_aix5 -#undef PLAT_ppc64_aix5 - -#endif /* __VALGRIND_H */ diff --git a/contrib/libtcmalloc/src/thread_cache.cc b/contrib/libtcmalloc/src/thread_cache.cc deleted file mode 100644 index 81b3694d563..00000000000 --- a/contrib/libtcmalloc/src/thread_cache.cc +++ /dev/null @@ -1,479 +0,0 @@ -// -*- Mode: C++; c-basic-offset: 2; indent-tabs-mode: nil -*- -// Copyright (c) 2008, Google Inc. -// All rights reserved. -// -// Redistribution and use in source and binary forms, with or without -// modification, are permitted provided that the following conditions are -// met: -// -// * Redistributions of source code must retain the above copyright -// notice, this list of conditions and the following disclaimer. -// * Redistributions in binary form must reproduce the above -// copyright notice, this list of conditions and the following disclaimer -// in the documentation and/or other materials provided with the -// distribution. -// * Neither the name of Google Inc. nor the names of its -// contributors may be used to endorse or promote products derived from -// this software without specific prior written permission. -// -// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS -// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT -// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR -// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT -// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, -// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT -// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, -// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY -// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT -// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE -// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. - -// --- -// Author: Ken Ashcraft - -#include "config.h" -#include "thread_cache.h" -#include -#include // for memcpy -#include // for max, min -#include "base/commandlineflags.h" // for SpinLockHolder -#include "base/spinlock.h" // for SpinLockHolder -#include "getenv_safe.h" // for TCMallocGetenvSafe -#include "central_freelist.h" // for CentralFreeListPadded -#include "maybe_threads.h" - -using std::min; -using std::max; - -// Note: this is initialized manually in InitModule to ensure that -// it's configured at right time -// -// DEFINE_int64(tcmalloc_max_total_thread_cache_bytes, -// EnvToInt64("TCMALLOC_MAX_TOTAL_THREAD_CACHE_BYTES", -// kDefaultOverallThreadCacheSize), -// "Bound on the total amount of bytes allocated to " -// "thread caches. This bound is not strict, so it is possible " -// "for the cache to go over this bound in certain circumstances. " -// "Maximum value of this flag is capped to 1 GB."); - - -namespace tcmalloc { - -static bool phinited = false; - -volatile size_t ThreadCache::per_thread_cache_size_ = kMaxThreadCacheSize; -size_t ThreadCache::overall_thread_cache_size_ = kDefaultOverallThreadCacheSize; -ssize_t ThreadCache::unclaimed_cache_space_ = kDefaultOverallThreadCacheSize; -PageHeapAllocator threadcache_allocator; -ThreadCache* ThreadCache::thread_heaps_ = NULL; -int ThreadCache::thread_heap_count_ = 0; -ThreadCache* ThreadCache::next_memory_steal_ = NULL; -#ifdef HAVE_TLS -__thread ThreadCache::ThreadLocalData ThreadCache::threadlocal_data_ - ATTR_INITIAL_EXEC - = {0, 0}; -#endif -bool ThreadCache::tsd_inited_ = false; -pthread_key_t ThreadCache::heap_key_; - -void ThreadCache::Init(pthread_t tid) { - size_ = 0; - - max_size_ = 0; - IncreaseCacheLimitLocked(); - if (max_size_ == 0) { - // There isn't enough memory to go around. Just give the minimum to - // this thread. - max_size_ = kMinThreadCacheSize; - - // Take unclaimed_cache_space_ negative. - unclaimed_cache_space_ -= kMinThreadCacheSize; - ASSERT(unclaimed_cache_space_ < 0); - } - - next_ = NULL; - prev_ = NULL; - tid_ = tid; - in_setspecific_ = false; - for (size_t cl = 0; cl < kNumClasses; ++cl) { - list_[cl].Init(); - } - - uint32_t sampler_seed; - memcpy(&sampler_seed, &tid, sizeof(sampler_seed)); - sampler_.Init(sampler_seed); -} - -void ThreadCache::Cleanup() { - // Put unused memory back into central cache - for (int cl = 0; cl < kNumClasses; ++cl) { - if (list_[cl].length() > 0) { - ReleaseToCentralCache(&list_[cl], cl, list_[cl].length()); - } - } -} - -// Remove some objects of class "cl" from central cache and add to thread heap. -// On success, return the first object for immediate use; otherwise return NULL. -void* ThreadCache::FetchFromCentralCache(size_t cl, size_t byte_size) { - FreeList* list = &list_[cl]; - ASSERT(list->empty()); - const int batch_size = Static::sizemap()->num_objects_to_move(cl); - - const int num_to_move = min(list->max_length(), batch_size); - void *start, *end; - int fetch_count = Static::central_cache()[cl].RemoveRange( - &start, &end, num_to_move); - - ASSERT((start == NULL) == (fetch_count == 0)); - if (--fetch_count >= 0) { - size_ += byte_size * fetch_count; - list->PushRange(fetch_count, SLL_Next(start), end); - } - - // Increase max length slowly up to batch_size. After that, - // increase by batch_size in one shot so that the length is a - // multiple of batch_size. - if (list->max_length() < batch_size) { - list->set_max_length(list->max_length() + 1); - } else { - // Don't let the list get too long. In 32 bit builds, the length - // is represented by a 16 bit int, so we need to watch out for - // integer overflow. - int new_length = min(list->max_length() + batch_size, - kMaxDynamicFreeListLength); - // The list's max_length must always be a multiple of batch_size, - // and kMaxDynamicFreeListLength is not necessarily a multiple - // of batch_size. - new_length -= new_length % batch_size; - ASSERT(new_length % batch_size == 0); - list->set_max_length(new_length); - } - return start; -} - -void ThreadCache::ListTooLong(FreeList* list, size_t cl) { - const int batch_size = Static::sizemap()->num_objects_to_move(cl); - ReleaseToCentralCache(list, cl, batch_size); - - // If the list is too long, we need to transfer some number of - // objects to the central cache. Ideally, we would transfer - // num_objects_to_move, so the code below tries to make max_length - // converge on num_objects_to_move. - - if (list->max_length() < batch_size) { - // Slow start the max_length so we don't overreserve. - list->set_max_length(list->max_length() + 1); - } else if (list->max_length() > batch_size) { - // If we consistently go over max_length, shrink max_length. If we don't - // shrink it, some amount of memory will always stay in this freelist. - list->set_length_overages(list->length_overages() + 1); - if (list->length_overages() > kMaxOverages) { - ASSERT(list->max_length() > batch_size); - list->set_max_length(list->max_length() - batch_size); - list->set_length_overages(0); - } - } -} - -// Remove some objects of class "cl" from thread heap and add to central cache -void ThreadCache::ReleaseToCentralCache(FreeList* src, size_t cl, int N) { - ASSERT(src == &list_[cl]); - if (N > src->length()) N = src->length(); - size_t delta_bytes = N * Static::sizemap()->ByteSizeForClass(cl); - - // We return prepackaged chains of the correct size to the central cache. - // TODO: Use the same format internally in the thread caches? - int batch_size = Static::sizemap()->num_objects_to_move(cl); - while (N > batch_size) { - void *tail, *head; - src->PopRange(batch_size, &head, &tail); - Static::central_cache()[cl].InsertRange(head, tail, batch_size); - N -= batch_size; - } - void *tail, *head; - src->PopRange(N, &head, &tail); - Static::central_cache()[cl].InsertRange(head, tail, N); - size_ -= delta_bytes; -} - -// Release idle memory to the central cache -void ThreadCache::Scavenge() { - // If the low-water mark for the free list is L, it means we would - // not have had to allocate anything from the central cache even if - // we had reduced the free list size by L. We aim to get closer to - // that situation by dropping L/2 nodes from the free list. This - // may not release much memory, but if so we will call scavenge again - // pretty soon and the low-water marks will be high on that call. - for (int cl = 0; cl < kNumClasses; cl++) { - FreeList* list = &list_[cl]; - const int lowmark = list->lowwatermark(); - if (lowmark > 0) { - const int drop = (lowmark > 1) ? lowmark/2 : 1; - ReleaseToCentralCache(list, cl, drop); - - // Shrink the max length if it isn't used. Only shrink down to - // batch_size -- if the thread was active enough to get the max_length - // above batch_size, it will likely be that active again. If - // max_length shinks below batch_size, the thread will have to - // go through the slow-start behavior again. The slow-start is useful - // mainly for threads that stay relatively idle for their entire - // lifetime. - const int batch_size = Static::sizemap()->num_objects_to_move(cl); - if (list->max_length() > batch_size) { - list->set_max_length( - max(list->max_length() - batch_size, batch_size)); - } - } - list->clear_lowwatermark(); - } - - IncreaseCacheLimit(); -} - -void ThreadCache::IncreaseCacheLimit() { - SpinLockHolder h(Static::pageheap_lock()); - IncreaseCacheLimitLocked(); -} - -void ThreadCache::IncreaseCacheLimitLocked() { - if (unclaimed_cache_space_ > 0) { - // Possibly make unclaimed_cache_space_ negative. - unclaimed_cache_space_ -= kStealAmount; - max_size_ += kStealAmount; - return; - } - // Don't hold pageheap_lock too long. Try to steal from 10 other - // threads before giving up. The i < 10 condition also prevents an - // infinite loop in case none of the existing thread heaps are - // suitable places to steal from. - for (int i = 0; i < 10; - ++i, next_memory_steal_ = next_memory_steal_->next_) { - // Reached the end of the linked list. Start at the beginning. - if (next_memory_steal_ == NULL) { - ASSERT(thread_heaps_ != NULL); - next_memory_steal_ = thread_heaps_; - } - if (next_memory_steal_ == this || - next_memory_steal_->max_size_ <= kMinThreadCacheSize) { - continue; - } - next_memory_steal_->max_size_ -= kStealAmount; - max_size_ += kStealAmount; - - next_memory_steal_ = next_memory_steal_->next_; - return; - } -} - -int ThreadCache::GetSamplePeriod() { - return sampler_.GetSamplePeriod(); -} - -void ThreadCache::InitModule() { - SpinLockHolder h(Static::pageheap_lock()); - if (!phinited) { - const char *tcb = TCMallocGetenvSafe("TCMALLOC_MAX_TOTAL_THREAD_CACHE_BYTES"); - if (tcb) { - set_overall_thread_cache_size(strtoll(tcb, NULL, 10)); - } - Static::InitStaticVars(); - threadcache_allocator.Init(); - phinited = 1; - } -} - -void ThreadCache::InitTSD() { - ASSERT(!tsd_inited_); - perftools_pthread_key_create(&heap_key_, DestroyThreadCache); - tsd_inited_ = true; - -#ifdef PTHREADS_CRASHES_IF_RUN_TOO_EARLY - // We may have used a fake pthread_t for the main thread. Fix it. - pthread_t zero; - memset(&zero, 0, sizeof(zero)); - SpinLockHolder h(Static::pageheap_lock()); - for (ThreadCache* h = thread_heaps_; h != NULL; h = h->next_) { - if (h->tid_ == zero) { - h->tid_ = pthread_self(); - } - } -#endif -} - -ThreadCache* ThreadCache::CreateCacheIfNecessary() { - // Initialize per-thread data if necessary - ThreadCache* heap = NULL; - { - SpinLockHolder h(Static::pageheap_lock()); - // On some old glibc's, and on freebsd's libc (as of freebsd 8.1), - // calling pthread routines (even pthread_self) too early could - // cause a segfault. Since we can call pthreads quite early, we - // have to protect against that in such situations by making a - // 'fake' pthread. This is not ideal since it doesn't work well - // when linking tcmalloc statically with apps that create threads - // before main, so we only do it if we have to. -#ifdef PTHREADS_CRASHES_IF_RUN_TOO_EARLY - pthread_t me; - if (!tsd_inited_) { - memset(&me, 0, sizeof(me)); - } else { - me = pthread_self(); - } -#else - const pthread_t me = pthread_self(); -#endif - - // This may be a recursive malloc call from pthread_setspecific() - // In that case, the heap for this thread has already been created - // and added to the linked list. So we search for that first. - for (ThreadCache* h = thread_heaps_; h != NULL; h = h->next_) { - if (h->tid_ == me) { - heap = h; - break; - } - } - - if (heap == NULL) heap = NewHeap(me); - } - - // We call pthread_setspecific() outside the lock because it may - // call malloc() recursively. We check for the recursive call using - // the "in_setspecific_" flag so that we can avoid calling - // pthread_setspecific() if we are already inside pthread_setspecific(). - if (!heap->in_setspecific_ && tsd_inited_) { - heap->in_setspecific_ = true; - perftools_pthread_setspecific(heap_key_, heap); -#ifdef HAVE_TLS - // Also keep a copy in __thread for faster retrieval - threadlocal_data_.heap = heap; - SetMinSizeForSlowPath(kMaxSize + 1); -#endif - heap->in_setspecific_ = false; - } - return heap; -} - -ThreadCache* ThreadCache::NewHeap(pthread_t tid) { - // Create the heap and add it to the linked list - ThreadCache *heap = threadcache_allocator.New(); - heap->Init(tid); - heap->next_ = thread_heaps_; - heap->prev_ = NULL; - if (thread_heaps_ != NULL) { - thread_heaps_->prev_ = heap; - } else { - // This is the only thread heap at the momment. - ASSERT(next_memory_steal_ == NULL); - next_memory_steal_ = heap; - } - thread_heaps_ = heap; - thread_heap_count_++; - return heap; -} - -void ThreadCache::BecomeIdle() { - if (!tsd_inited_) return; // No caches yet - ThreadCache* heap = GetThreadHeap(); - if (heap == NULL) return; // No thread cache to remove - if (heap->in_setspecific_) return; // Do not disturb the active caller - - heap->in_setspecific_ = true; - perftools_pthread_setspecific(heap_key_, NULL); -#ifdef HAVE_TLS - // Also update the copy in __thread - threadlocal_data_.heap = NULL; - SetMinSizeForSlowPath(0); -#endif - heap->in_setspecific_ = false; - if (GetThreadHeap() == heap) { - // Somehow heap got reinstated by a recursive call to malloc - // from pthread_setspecific. We give up in this case. - return; - } - - // We can now get rid of the heap - DeleteCache(heap); -} - -void ThreadCache::BecomeTemporarilyIdle() { - ThreadCache* heap = GetCacheIfPresent(); - if (heap) - heap->Cleanup(); -} - -void ThreadCache::DestroyThreadCache(void* ptr) { - // Note that "ptr" cannot be NULL since pthread promises not - // to invoke the destructor on NULL values, but for safety, - // we check anyway. - if (ptr == NULL) return; -#ifdef HAVE_TLS - // Prevent fast path of GetThreadHeap() from returning heap. - threadlocal_data_.heap = NULL; - SetMinSizeForSlowPath(0); -#endif - DeleteCache(reinterpret_cast(ptr)); -} - -void ThreadCache::DeleteCache(ThreadCache* heap) { - // Remove all memory from heap - heap->Cleanup(); - - // Remove from linked list - SpinLockHolder h(Static::pageheap_lock()); - if (heap->next_ != NULL) heap->next_->prev_ = heap->prev_; - if (heap->prev_ != NULL) heap->prev_->next_ = heap->next_; - if (thread_heaps_ == heap) thread_heaps_ = heap->next_; - thread_heap_count_--; - - if (next_memory_steal_ == heap) next_memory_steal_ = heap->next_; - if (next_memory_steal_ == NULL) next_memory_steal_ = thread_heaps_; - unclaimed_cache_space_ += heap->max_size_; - - threadcache_allocator.Delete(heap); -} - -void ThreadCache::RecomputePerThreadCacheSize() { - // Divide available space across threads - int n = thread_heap_count_ > 0 ? thread_heap_count_ : 1; - size_t space = overall_thread_cache_size_ / n; - - // Limit to allowed range - if (space < kMinThreadCacheSize) space = kMinThreadCacheSize; - if (space > kMaxThreadCacheSize) space = kMaxThreadCacheSize; - - double ratio = space / max(1, per_thread_cache_size_); - size_t claimed = 0; - for (ThreadCache* h = thread_heaps_; h != NULL; h = h->next_) { - // Increasing the total cache size should not circumvent the - // slow-start growth of max_size_. - if (ratio < 1.0) { - h->max_size_ = static_cast(h->max_size_ * ratio); - } - claimed += h->max_size_; - } - unclaimed_cache_space_ = overall_thread_cache_size_ - claimed; - per_thread_cache_size_ = space; -} - -void ThreadCache::GetThreadStats(uint64_t* total_bytes, uint64_t* class_count) { - for (ThreadCache* h = thread_heaps_; h != NULL; h = h->next_) { - *total_bytes += h->Size(); - if (class_count) { - for (int cl = 0; cl < kNumClasses; ++cl) { - class_count[cl] += h->freelist_length(cl); - } - } - } -} - -void ThreadCache::set_overall_thread_cache_size(size_t new_size) { - // Clip the value to a reasonable range - if (new_size < kMinThreadCacheSize) new_size = kMinThreadCacheSize; - if (new_size > (1<<30)) new_size = (1<<30); // Limit to 1GB - overall_thread_cache_size_ = new_size; - - RecomputePerThreadCacheSize(); -} - -} // namespace tcmalloc diff --git a/contrib/libtcmalloc/src/thread_cache.h b/contrib/libtcmalloc/src/thread_cache.h deleted file mode 100644 index ff7ab1ae77d..00000000000 --- a/contrib/libtcmalloc/src/thread_cache.h +++ /dev/null @@ -1,474 +0,0 @@ -// -*- Mode: C++; c-basic-offset: 2; indent-tabs-mode: nil -*- -// Copyright (c) 2008, Google Inc. -// All rights reserved. -// -// Redistribution and use in source and binary forms, with or without -// modification, are permitted provided that the following conditions are -// met: -// -// * Redistributions of source code must retain the above copyright -// notice, this list of conditions and the following disclaimer. -// * Redistributions in binary form must reproduce the above -// copyright notice, this list of conditions and the following disclaimer -// in the documentation and/or other materials provided with the -// distribution. -// * Neither the name of Google Inc. nor the names of its -// contributors may be used to endorse or promote products derived from -// this software without specific prior written permission. -// -// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS -// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT -// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR -// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT -// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, -// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT -// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, -// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY -// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT -// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE -// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. - -// --- -// Author: Sanjay Ghemawat - -#ifndef TCMALLOC_THREAD_CACHE_H_ -#define TCMALLOC_THREAD_CACHE_H_ - -#include "config.h" -#ifdef HAVE_PTHREAD -#include // for pthread_t, pthread_key_t -#endif -#include // for size_t, NULL -#ifdef HAVE_STDINT_H -#include // for uint32_t, uint64_t -#endif -#include // for ssize_t -#include "base/commandlineflags.h" -#include "common.h" -#include "linked_list.h" -#include "maybe_threads.h" -#include "page_heap_allocator.h" -#include "sampler.h" -#include "static_vars.h" - -#include "common.h" // for SizeMap, kMaxSize, etc -#include "internal_logging.h" // for ASSERT, etc -#include "linked_list.h" // for SLL_Pop, SLL_PopRange, etc -#include "page_heap_allocator.h" // for PageHeapAllocator -#include "sampler.h" // for Sampler -#include "static_vars.h" // for Static - -DECLARE_int64(tcmalloc_sample_parameter); - -namespace tcmalloc { - -//------------------------------------------------------------------- -// Data kept per thread -//------------------------------------------------------------------- - -class ThreadCache { - public: -#ifdef HAVE_TLS - enum { have_tls = true }; -#else - enum { have_tls = false }; -#endif - - // All ThreadCache objects are kept in a linked list (for stats collection) - ThreadCache* next_; - ThreadCache* prev_; - - void Init(pthread_t tid); - void Cleanup(); - - // Accessors (mostly just for printing stats) - int freelist_length(size_t cl) const { return list_[cl].length(); } - - // Total byte size in cache - size_t Size() const { return size_; } - - // Allocate an object of the given size and class. The size given - // must be the same as the size of the class in the size map. - void* Allocate(size_t size, size_t cl); - void Deallocate(void* ptr, size_t size_class); - - void Scavenge(); - - int GetSamplePeriod(); - - // Record allocation of "k" bytes. Return true iff allocation - // should be sampled - bool SampleAllocation(size_t k); - - static void InitModule(); - static void InitTSD(); - static ThreadCache* GetThreadHeap(); - static ThreadCache* GetCache(); - static ThreadCache* GetCacheIfPresent(); - static ThreadCache* GetCacheWhichMustBePresent(); - static ThreadCache* CreateCacheIfNecessary(); - static void BecomeIdle(); - static void BecomeTemporarilyIdle(); - static size_t MinSizeForSlowPath(); - static void SetMinSizeForSlowPath(size_t size); - static void SetUseEmergencyMalloc(); - static void ResetUseEmergencyMalloc(); - static bool IsUseEmergencyMalloc(); - - static bool IsFastPathAllowed() { return MinSizeForSlowPath() != 0; } - - // Return the number of thread heaps in use. - static inline int HeapsInUse(); - - // Adds to *total_bytes the total number of bytes used by all thread heaps. - // Also, if class_count is not NULL, it must be an array of size kNumClasses, - // and this function will increment each element of class_count by the number - // of items in all thread-local freelists of the corresponding size class. - // REQUIRES: Static::pageheap_lock is held. - static void GetThreadStats(uint64_t* total_bytes, uint64_t* class_count); - - // Sets the total thread cache size to new_size, recomputing the - // individual thread cache sizes as necessary. - // REQUIRES: Static::pageheap lock is held. - static void set_overall_thread_cache_size(size_t new_size); - static size_t overall_thread_cache_size() { - return overall_thread_cache_size_; - } - - private: - class FreeList { - private: - void* list_; // Linked list of nodes - -#ifdef _LP64 - // On 64-bit hardware, manipulating 16-bit values may be slightly slow. - uint32_t length_; // Current length. - uint32_t lowater_; // Low water mark for list length. - uint32_t max_length_; // Dynamic max list length based on usage. - // Tracks the number of times a deallocation has caused - // length_ > max_length_. After the kMaxOverages'th time, max_length_ - // shrinks and length_overages_ is reset to zero. - uint32_t length_overages_; -#else - // If we aren't using 64-bit pointers then pack these into less space. - uint16_t length_; - uint16_t lowater_; - uint16_t max_length_; - uint16_t length_overages_; -#endif - - public: - void Init() { - list_ = NULL; - length_ = 0; - lowater_ = 0; - max_length_ = 1; - length_overages_ = 0; - } - - // Return current length of list - size_t length() const { - return length_; - } - - // Return the maximum length of the list. - size_t max_length() const { - return max_length_; - } - - // Set the maximum length of the list. If 'new_max' > length(), the - // client is responsible for removing objects from the list. - void set_max_length(size_t new_max) { - max_length_ = new_max; - } - - // Return the number of times that length() has gone over max_length(). - size_t length_overages() const { - return length_overages_; - } - - void set_length_overages(size_t new_count) { - length_overages_ = new_count; - } - - // Is list empty? - bool empty() const { - return list_ == NULL; - } - - // Low-water mark management - int lowwatermark() const { return lowater_; } - void clear_lowwatermark() { lowater_ = length_; } - - void Push(void* ptr) { - SLL_Push(&list_, ptr); - length_++; - } - - void* Pop() { - ASSERT(list_ != NULL); - length_--; - if (length_ < lowater_) lowater_ = length_; - return SLL_Pop(&list_); - } - - void* Next() { - return SLL_Next(&list_); - } - - void PushRange(int N, void *start, void *end) { - SLL_PushRange(&list_, start, end); - length_ += N; - } - - void PopRange(int N, void **start, void **end) { - SLL_PopRange(&list_, N, start, end); - ASSERT(length_ >= N); - length_ -= N; - if (length_ < lowater_) lowater_ = length_; - } - }; - - // Gets and returns an object from the central cache, and, if possible, - // also adds some objects of that size class to this thread cache. - void* FetchFromCentralCache(size_t cl, size_t byte_size); - - // Releases some number of items from src. Adjusts the list's max_length - // to eventually converge on num_objects_to_move(cl). - void ListTooLong(FreeList* src, size_t cl); - - // Releases N items from this thread cache. - void ReleaseToCentralCache(FreeList* src, size_t cl, int N); - - // Increase max_size_ by reducing unclaimed_cache_space_ or by - // reducing the max_size_ of some other thread. In both cases, - // the delta is kStealAmount. - void IncreaseCacheLimit(); - // Same as above but requires Static::pageheap_lock() is held. - void IncreaseCacheLimitLocked(); - - // If TLS is available, we also store a copy of the per-thread object - // in a __thread variable since __thread variables are faster to read - // than pthread_getspecific(). We still need pthread_setspecific() - // because __thread variables provide no way to run cleanup code when - // a thread is destroyed. - // We also give a hint to the compiler to use the "initial exec" TLS - // model. This is faster than the default TLS model, at the cost that - // you cannot dlopen this library. (To see the difference, look at - // the CPU use of __tls_get_addr with and without this attribute.) - // Since we don't really use dlopen in google code -- and using dlopen - // on a malloc replacement is asking for trouble in any case -- that's - // a good tradeoff for us. -#ifdef HAVE_TLS - struct ThreadLocalData { - ThreadCache* heap; - // min_size_for_slow_path is 0 if heap is NULL or kMaxSize + 1 otherwise. - // The latter is the common case and allows allocation to be faster - // than it would be otherwise: typically a single branch will - // determine that the requested allocation is no more than kMaxSize - // and we can then proceed, knowing that global and thread-local tcmalloc - // state is initialized. - size_t min_size_for_slow_path; - - bool use_emergency_malloc; - size_t old_min_size_for_slow_path; - }; - static __thread ThreadLocalData threadlocal_data_ ATTR_INITIAL_EXEC; -#endif - - // Thread-specific key. Initialization here is somewhat tricky - // because some Linux startup code invokes malloc() before it - // is in a good enough state to handle pthread_keycreate(). - // Therefore, we use TSD keys only after tsd_inited is set to true. - // Until then, we use a slow path to get the heap object. - static bool tsd_inited_; - static pthread_key_t heap_key_; - - // Linked list of heap objects. Protected by Static::pageheap_lock. - static ThreadCache* thread_heaps_; - static int thread_heap_count_; - - // A pointer to one of the objects in thread_heaps_. Represents - // the next ThreadCache from which a thread over its max_size_ should - // steal memory limit. Round-robin through all of the objects in - // thread_heaps_. Protected by Static::pageheap_lock. - static ThreadCache* next_memory_steal_; - - // Overall thread cache size. Protected by Static::pageheap_lock. - static size_t overall_thread_cache_size_; - - // Global per-thread cache size. Writes are protected by - // Static::pageheap_lock. Reads are done without any locking, which should be - // fine as long as size_t can be written atomically and we don't place - // invariants between this variable and other pieces of state. - static volatile size_t per_thread_cache_size_; - - // Represents overall_thread_cache_size_ minus the sum of max_size_ - // across all ThreadCaches. Protected by Static::pageheap_lock. - static ssize_t unclaimed_cache_space_; - - // This class is laid out with the most frequently used fields - // first so that hot elements are placed on the same cache line. - - size_t size_; // Combined size of data - size_t max_size_; // size_ > max_size_ --> Scavenge() - - // We sample allocations, biased by the size of the allocation - Sampler sampler_; // A sampler - - FreeList list_[kNumClasses]; // Array indexed by size-class - - pthread_t tid_; // Which thread owns it - bool in_setspecific_; // In call to pthread_setspecific? - - // Allocate a new heap. REQUIRES: Static::pageheap_lock is held. - static ThreadCache* NewHeap(pthread_t tid); - - // Use only as pthread thread-specific destructor function. - static void DestroyThreadCache(void* ptr); - - static void DeleteCache(ThreadCache* heap); - static void RecomputePerThreadCacheSize(); - - // Ensure that this class is cacheline-aligned. This is critical for - // performance, as false sharing would negate many of the benefits - // of a per-thread cache. -} CACHELINE_ALIGNED; - -// Allocator for thread heaps -// This is logically part of the ThreadCache class, but MSVC, at -// least, does not like using ThreadCache as a template argument -// before the class is fully defined. So we put it outside the class. -extern PageHeapAllocator threadcache_allocator; - -inline int ThreadCache::HeapsInUse() { - return threadcache_allocator.inuse(); -} - -inline bool ThreadCache::SampleAllocation(size_t k) { -#ifndef NO_TCMALLOC_SAMPLES - return UNLIKELY(FLAGS_tcmalloc_sample_parameter > 0) && sampler_.SampleAllocation(k); -#else - return false; -#endif -} - -inline void* ThreadCache::Allocate(size_t size, size_t cl) { - ASSERT(size <= kMaxSize); - ASSERT(size == Static::sizemap()->ByteSizeForClass(cl)); - - FreeList* list = &list_[cl]; - if (UNLIKELY(list->empty())) { - return FetchFromCentralCache(cl, size); - } - size_ -= size; - return list->Pop(); -} - -inline void ThreadCache::Deallocate(void* ptr, size_t cl) { - FreeList* list = &list_[cl]; - size_ += Static::sizemap()->ByteSizeForClass(cl); - ssize_t size_headroom = max_size_ - size_ - 1; - - // This catches back-to-back frees of allocs in the same size - // class. A more comprehensive (and expensive) test would be to walk - // the entire freelist. But this might be enough to find some bugs. - ASSERT(ptr != list->Next()); - - list->Push(ptr); - ssize_t list_headroom = - static_cast(list->max_length()) - list->length(); - - // There are two relatively uncommon things that require further work. - // In the common case we're done, and in that case we need a single branch - // because of the bitwise-or trick that follows. - if (UNLIKELY((list_headroom | size_headroom) < 0)) { - if (list_headroom < 0) { - ListTooLong(list, cl); - } - if (size_ >= max_size_) Scavenge(); - } -} - -inline ThreadCache* ThreadCache::GetThreadHeap() { -#ifdef HAVE_TLS - return threadlocal_data_.heap; -#else - return reinterpret_cast( - perftools_pthread_getspecific(heap_key_)); -#endif -} - -inline ThreadCache* ThreadCache::GetCacheWhichMustBePresent() { -#ifdef HAVE_TLS - ASSERT(threadlocal_data_.heap); - return threadlocal_data_.heap; -#else - ASSERT(perftools_pthread_getspecific(heap_key_)); - return reinterpret_cast( - perftools_pthread_getspecific(heap_key_)); -#endif -} - -inline ThreadCache* ThreadCache::GetCache() { - ThreadCache* ptr = NULL; - if (!tsd_inited_) { - InitModule(); - } else { - ptr = GetThreadHeap(); - } - if (ptr == NULL) ptr = CreateCacheIfNecessary(); - return ptr; -} - -// In deletion paths, we do not try to create a thread-cache. This is -// because we may be in the thread destruction code and may have -// already cleaned up the cache for this thread. -inline ThreadCache* ThreadCache::GetCacheIfPresent() { -#ifndef HAVE_TLS - if (!tsd_inited_) return NULL; -#endif - return GetThreadHeap(); -} - -inline size_t ThreadCache::MinSizeForSlowPath() { -#ifdef HAVE_TLS - return threadlocal_data_.min_size_for_slow_path; -#else - return 0; -#endif -} - -inline void ThreadCache::SetMinSizeForSlowPath(size_t size) { -#ifdef HAVE_TLS - threadlocal_data_.min_size_for_slow_path = size; -#endif -} - -inline void ThreadCache::SetUseEmergencyMalloc() { -#ifdef HAVE_TLS - threadlocal_data_.old_min_size_for_slow_path = threadlocal_data_.min_size_for_slow_path; - threadlocal_data_.min_size_for_slow_path = 0; - threadlocal_data_.use_emergency_malloc = true; -#endif -} - -inline void ThreadCache::ResetUseEmergencyMalloc() { -#ifdef HAVE_TLS - threadlocal_data_.min_size_for_slow_path = threadlocal_data_.old_min_size_for_slow_path; - threadlocal_data_.use_emergency_malloc = false; -#endif -} - -inline bool ThreadCache::IsUseEmergencyMalloc() { -#if defined(HAVE_TLS) && defined(ENABLE_EMERGENCY_MALLOC) - return UNLIKELY(threadlocal_data_.use_emergency_malloc); -#else - return false; -#endif -} - - -} // namespace tcmalloc - -#endif // TCMALLOC_THREAD_CACHE_H_ diff --git a/contrib/tiflash-proxy b/contrib/tiflash-proxy index fbd8e6f2099..3bd934fa8e1 160000 --- a/contrib/tiflash-proxy +++ b/contrib/tiflash-proxy @@ -1 +1 @@ -Subproject commit fbd8e6f209952d5bb79897d8bce17230acf23495 +Subproject commit 3bd934fa8e1868ffbc29e1c92cda37e0c2049bf7 diff --git a/contrib/tiflash-proxy-cmake/CMakeLists.txt b/contrib/tiflash-proxy-cmake/CMakeLists.txt index e3e2df379a1..cdbeaa6f11d 100644 --- a/contrib/tiflash-proxy-cmake/CMakeLists.txt +++ b/contrib/tiflash-proxy-cmake/CMakeLists.txt @@ -27,6 +27,8 @@ if(TIFLASH_LLVM_TOOLCHAIN AND USE_LIBCXX) endif() set(TIFLASH_RUST_ENV CC=${CMAKE_C_COMPILER} CXX=${CMAKE_CXX_COMPILER} CXXSTDLIB=c++ ${TIFLASH_RUST_ENV} RUSTFLAGS=${TIFLASH_RUSTFLAGS} PROTOC=${Protobuf_PROTOC_EXECUTABLE} PROTOC_INCLUDE=${Protobuf_INCLUDE_DIR}) message(STATUS "Enforce LLVM toolchain for rust") +else() + set(TIFLASH_RUST_ENV PROTOC=${Protobuf_PROTOC_EXECUTABLE} PROTOC_INCLUDE=${Protobuf_INCLUDE_DIR}) endif() message(STATUS "Using rust env for tiflash-proxy: ${TIFLASH_RUST_ENV}") diff --git a/contrib/tipb b/contrib/tipb index 0f4f873beca..3e2483c20a9 160000 --- a/contrib/tipb +++ b/contrib/tipb @@ -1 +1 @@ -Subproject commit 0f4f873beca8d5078dde0a23d15ad5ce3188ed0d +Subproject commit 3e2483c20a9ec3f9ea587c49808c4b66c9716a2c diff --git a/dbms/CMakeLists.txt b/dbms/CMakeLists.txt index 0df79f89a84..3e38c2a9fdb 100644 --- a/dbms/CMakeLists.txt +++ b/dbms/CMakeLists.txt @@ -80,6 +80,7 @@ add_headers_and_sources(dbms src/Storages/DeltaMerge/FilterParser) add_headers_and_sources(dbms src/Storages/DeltaMerge/File) add_headers_and_sources(dbms src/Storages/DeltaMerge/ColumnFile) add_headers_and_sources(dbms src/Storages/DeltaMerge/Delta) +add_headers_and_sources(dbms src/Storages/DeltaMerge/ReadThread) add_headers_and_sources(dbms src/Storages/Distributed) add_headers_and_sources(dbms src/Storages/Transaction) add_headers_and_sources(dbms src/Storages/Page/V1) diff --git a/dbms/src/Columns/ColumnString.cpp b/dbms/src/Columns/ColumnString.cpp index 54d4238616f..424f7e72052 100644 --- a/dbms/src/Columns/ColumnString.cpp +++ b/dbms/src/Columns/ColumnString.cpp @@ -17,8 +17,10 @@ #include #include #include +#include #include + /// Used in the `reserve` method, when the number of rows is known, but sizes of elements are not. #define APPROX_STRING_SIZE 64 @@ -315,58 +317,100 @@ int ColumnString::compareAtWithCollationImpl(size_t n, size_t m, const IColumn & return collator.compare( reinterpret_cast(&chars[offsetAt(n)]), - sizeAt(n), + sizeAt(n) - 1, // Skip last zero byte. reinterpret_cast(&rhs.chars[rhs.offsetAt(m)]), - rhs.sizeAt(m)); + rhs.sizeAt(m) - 1 // Skip last zero byte. + ); } - -template -struct ColumnString::lessWithCollation +// Derived must implement function `int compare(const char *, size_t, const char *, size_t)`. +template +struct ColumnString::LessWithCollation { const ColumnString & parent; - const ICollator & collator; + const Derived & inner; - lessWithCollation(const ColumnString & parent_, const ICollator & collator_) + LessWithCollation(const ColumnString & parent_, const Derived & inner_) : parent(parent_) - , collator(collator_) + , inner(inner_) {} - bool operator()(size_t lhs, size_t rhs) const + FLATTEN_INLINE_PURE inline bool operator()(size_t lhs, size_t rhs) const { - int res = collator.compare( + int res = inner.compare( reinterpret_cast(&parent.chars[parent.offsetAt(lhs)]), - parent.sizeAt(lhs), + parent.sizeAt(lhs) - 1, // Skip last zero byte. reinterpret_cast(&parent.chars[parent.offsetAt(rhs)]), - parent.sizeAt(rhs)); + parent.sizeAt(rhs) - 1 // Skip last zero byte. + ); - return positive ? (res < 0) : (res > 0); + if constexpr (positive) + { + return (res < 0); + } + else + { + return (res > 0); + } } }; -void ColumnString::getPermutationWithCollationImpl(const ICollator & collator, bool reverse, size_t limit, Permutation & res) const +struct Utf8MB4BinCmp { - size_t s = offsets.size(); - res.resize(s); - for (size_t i = 0; i < s; ++i) - res[i] = i; - - if (limit >= s) - limit = 0; + static FLATTEN_INLINE_PURE inline int compare(const char * s1, size_t length1, const char * s2, size_t length2) + { + return DB::BinCollatorCompare(s1, length1, s2, length2); + } +}; - if (limit) +// common util functions +template <> +struct ColumnString::LessWithCollation +{ + // `CollationCmpImpl` must implement function `int compare(const char *, size_t, const char *, size_t)`. + template + static void getPermutationWithCollationImpl(const ColumnString & src, const CollationCmpImpl & collator_cmp_impl, bool reverse, size_t limit, Permutation & res) { - if (reverse) - std::partial_sort(res.begin(), res.begin() + limit, res.end(), lessWithCollation(*this, collator)); + size_t s = src.offsets.size(); + res.resize(s); + for (size_t i = 0; i < s; ++i) + res[i] = i; + + if (limit >= s) + limit = 0; + + if (limit) + { + if (reverse) + std::partial_sort(res.begin(), res.begin() + limit, res.end(), LessWithCollation(src, collator_cmp_impl)); + else + std::partial_sort(res.begin(), res.begin() + limit, res.end(), LessWithCollation(src, collator_cmp_impl)); + } else - std::partial_sort(res.begin(), res.begin() + limit, res.end(), lessWithCollation(*this, collator)); + { + if (reverse) + std::sort(res.begin(), res.end(), LessWithCollation(src, collator_cmp_impl)); + else + std::sort(res.begin(), res.end(), LessWithCollation(src, collator_cmp_impl)); + } } - else +}; + +void ColumnString::getPermutationWithCollationImpl(const ICollator & collator, bool reverse, size_t limit, Permutation & res) const +{ + using PermutationWithCollationUtils = ColumnString::LessWithCollation; + + // optimize path for default collator `UTF8MB4_BIN` + if (TiDB::ITiDBCollator::getCollator(TiDB::ITiDBCollator::UTF8MB4_BIN) == &collator) { - if (reverse) - std::sort(res.begin(), res.end(), lessWithCollation(*this, collator)); - else - std::sort(res.begin(), res.end(), lessWithCollation(*this, collator)); + Utf8MB4BinCmp cmp_impl; + PermutationWithCollationUtils::getPermutationWithCollationImpl(*this, cmp_impl, reverse, limit, res); + /// + return; + } + + { + PermutationWithCollationUtils::getPermutationWithCollationImpl(*this, collator, reverse, limit, res); } } @@ -377,35 +421,73 @@ void ColumnString::updateWeakHash32(WeakHash32 & hash, const TiDB::TiDBCollatorP if (hash.getData().size() != s) throw Exception(fmt::format("Size of WeakHash32 does not match size of column: column size is {}, hash size is {}", s, hash.getData().size()), ErrorCodes::LOGICAL_ERROR); - const UInt8 * pos = chars.data(); UInt32 * hash_data = hash.getData().data(); - Offset prev_offset = 0; if (collator != nullptr) { - for (const auto & offset : offsets) + if (collator->getCollatorId() == TiDB::ITiDBCollator::UTF8MB4_BIN) { - auto str_size = offset - prev_offset; - /// Skip last zero byte. - auto sort_key = collator->sortKey(reinterpret_cast(pos), str_size - 1, sort_key_container); - *hash_data = ::updateWeakHash32(reinterpret_cast(sort_key.data), sort_key.size, *hash_data); - - pos += str_size; - prev_offset = offset; + // Skip last zero byte. + LoopOneColumn(chars, offsets, offsets.size(), [&](const std::string_view & view, size_t) { + auto sort_key = BinCollatorSortKey(view.data(), view.size()); + *hash_data = ::updateWeakHash32(reinterpret_cast(sort_key.data), sort_key.size, *hash_data); + ++hash_data; + }); + } + else + { + // Skip last zero byte. + LoopOneColumn(chars, offsets, offsets.size(), [&](const std::string_view & view, size_t) { + auto sort_key = collator->sortKey(view.data(), view.size(), sort_key_container); + *hash_data = ::updateWeakHash32(reinterpret_cast(sort_key.data), sort_key.size, *hash_data); + ++hash_data; + }); + } + } + else + { + // Skip last zero byte. + LoopOneColumn(chars, offsets, offsets.size(), [&](const std::string_view & view, size_t) { + *hash_data = ::updateWeakHash32(reinterpret_cast(view.data()), view.size(), *hash_data); ++hash_data; + }); + } +} + +void ColumnString::updateHashWithValues(IColumn::HashValues & hash_values, const TiDB::TiDBCollatorPtr & collator, String & sort_key_container) const +{ + if (collator != nullptr) + { + if (collator->getCollatorId() == TiDB::ITiDBCollator::UTF8MB4_BIN) + { + // Skip last zero byte. + LoopOneColumn(chars, offsets, offsets.size(), [&hash_values](const std::string_view & view, size_t i) { + auto sort_key = BinCollatorSortKey(view.data(), view.size()); + size_t string_size = sort_key.size; + hash_values[i].update(reinterpret_cast(&string_size), sizeof(string_size)); + hash_values[i].update(sort_key.data, sort_key.size); + }); + } + else + { + // Skip last zero byte. + LoopOneColumn(chars, offsets, offsets.size(), [&](const std::string_view & view, size_t i) { + auto sort_key = collator->sortKey(view.data(), view.size(), sort_key_container); + size_t string_size = sort_key.size; + hash_values[i].update(reinterpret_cast(&string_size), sizeof(string_size)); + hash_values[i].update(sort_key.data, sort_key.size); + }); } } else { - for (const auto & offset : offsets) + for (size_t i = 0; i < offsets.size(); ++i) { - auto str_size = offset - prev_offset; - /// Skip last zero byte. - *hash_data = ::updateWeakHash32(pos, str_size - 1, *hash_data); + size_t string_size = sizeAt(i); + size_t offset = offsetAt(i); - pos += str_size; - prev_offset = offset; - ++hash_data; + hash_values[i].update(reinterpret_cast(&string_size), sizeof(string_size)); + hash_values[i].update(reinterpret_cast(&chars[offset]), string_size); } } } diff --git a/dbms/src/Columns/ColumnString.h b/dbms/src/Columns/ColumnString.h index 48b02388a6c..2204319e090 100644 --- a/dbms/src/Columns/ColumnString.h +++ b/dbms/src/Columns/ColumnString.h @@ -52,8 +52,8 @@ class ColumnString final : public COWPtrHelper template struct less; - template - struct lessWithCollation; + template + struct LessWithCollation; ColumnString() = default; @@ -118,7 +118,7 @@ class ColumnString final : public COWPtrHelper void insert(const Field & x) override { - const String & s = DB::get(x); + const auto & s = DB::get(x); const size_t old_size = chars.size(); const size_t size_to_append = s.size() + 1; const size_t new_size = old_size + size_to_append; @@ -134,7 +134,7 @@ class ColumnString final : public COWPtrHelper void insertFrom(const IColumn & src_, size_t n) override { - const ColumnString & src = static_cast(src_); + const auto & src = static_cast(src_); if (n != 0) { @@ -213,7 +213,7 @@ class ColumnString final : public COWPtrHelper if (collator != nullptr) { - /// Skip last zero byte. + // Skip last zero byte. auto sort_key = collator->sortKey(reinterpret_cast(src), string_size - 1, sort_key_container); string_size = sort_key.size; src = sort_key.data; @@ -246,7 +246,8 @@ class ColumnString final : public COWPtrHelper size_t offset = offsetAt(n); if (collator != nullptr) { - auto sort_key = collator->sortKey(reinterpret_cast(&chars[offset]), string_size, sort_key_container); + // Skip last zero byte. + auto sort_key = collator->sortKey(reinterpret_cast(&chars[offset]), string_size - 1, sort_key_container); string_size = sort_key.size; hash.update(reinterpret_cast(&string_size), sizeof(string_size)); hash.update(sort_key.data, sort_key.size); @@ -258,33 +259,7 @@ class ColumnString final : public COWPtrHelper } } - void updateHashWithValues(IColumn::HashValues & hash_values, const TiDB::TiDBCollatorPtr & collator, String & sort_key_container) const override - { - if (collator != nullptr) - { - for (size_t i = 0; i < offsets.size(); ++i) - { - size_t string_size = sizeAt(i); - size_t offset = offsetAt(i); - - auto sort_key = collator->sortKey(reinterpret_cast(&chars[offset]), string_size, sort_key_container); - string_size = sort_key.size; - hash_values[i].update(reinterpret_cast(&string_size), sizeof(string_size)); - hash_values[i].update(sort_key.data, sort_key.size); - } - } - else - { - for (size_t i = 0; i < offsets.size(); ++i) - { - size_t string_size = sizeAt(i); - size_t offset = offsetAt(i); - - hash_values[i].update(reinterpret_cast(&string_size), sizeof(string_size)); - hash_values[i].update(reinterpret_cast(&chars[offset]), string_size); - } - } - } + void updateHashWithValues(IColumn::HashValues & hash_values, const TiDB::TiDBCollatorPtr & collator, String & sort_key_container) const override; void updateWeakHash32(WeakHash32 & hash, const TiDB::TiDBCollatorPtr &, String &) const override; @@ -302,7 +277,7 @@ class ColumnString final : public COWPtrHelper int compareAt(size_t n, size_t m, const IColumn & rhs_, int /*nan_direction_hint*/) const override { - const ColumnString & rhs = static_cast(rhs_); + const auto & rhs = static_cast(rhs_); const size_t size = sizeAt(n); const size_t rhs_size = rhs.sizeAt(m); diff --git a/dbms/src/Common/TiFlashBuildInfo.cpp b/dbms/src/Common/TiFlashBuildInfo.cpp index f2353256b9b..6841528c065 100644 --- a/dbms/src/Common/TiFlashBuildInfo.cpp +++ b/dbms/src/Common/TiFlashBuildInfo.cpp @@ -60,8 +60,6 @@ std::string getEnabledFeatures() "jemalloc", #elif USE_MIMALLOC "mimalloc", -#elif USE_TCMALLOC - "tcmalloc", #endif // mem-profiling diff --git a/dbms/src/Common/TiFlashMetrics.h b/dbms/src/Common/TiFlashMetrics.h index c0ce60af01e..5b627783544 100644 --- a/dbms/src/Common/TiFlashMetrics.h +++ b/dbms/src/Common/TiFlashMetrics.h @@ -205,8 +205,23 @@ namespace DB F(type_thread_hard_limit, {"type", "thread_hard_limit"}), \ F(type_hard_limit_exceeded_count, {"type", "hard_limit_exceeded_count"})) \ M(tiflash_task_scheduler_waiting_duration_seconds, "Bucketed histogram of task waiting for scheduling duration", Histogram, \ - F(type_task_scheduler_waiting_duration, {{"type", "task_waiting_duration"}}, ExpBuckets{0.001, 2, 20})) - + F(type_task_scheduler_waiting_duration, {{"type", "task_waiting_duration"}}, ExpBuckets{0.001, 2, 20})) \ + M(tiflash_storage_read_thread_counter, "The counter of storage read thread", Counter, \ + F(type_sche_no_pool, {"type", "sche_no_pool"}), \ + F(type_sche_no_slot, {"type", "sche_no_slot"}), \ + F(type_sche_no_segment, {"type", "sche_no_segment"}), \ + F(type_sche_from_cache, {"type", "sche_from_cache"}), \ + F(type_sche_new_task, {"type", "sche_new_task"}), \ + F(type_add_cache_succ, {"type", "add_cache_succ"}), \ + F(type_add_cache_stale, {"type", "add_cache_stale"}), \ + F(type_get_cache_miss, {"type", "get_cache_miss"}), \ + F(type_get_cache_part, {"type", "get_cache_part"}), \ + F(type_get_cache_hit, {"type", "get_cache_hit"}), \ + F(type_get_cache_copy, {"type", "add_cache_copy"})) \ + M(tiflash_storage_read_thread_gauge, "The gauge of storage read thread", Gauge, \ + F(type_merged_task, {"type", "merged_task"})) \ + M(tiflash_storage_read_thread_seconds, "Bucketed histogram of read thread", Histogram, \ + F(type_merged_task, {{"type", "merged_task"}}, ExpBuckets{0.001, 2, 20})) // clang-format on struct ExpBuckets diff --git a/dbms/src/Common/config_build.cpp.in b/dbms/src/Common/config_build.cpp.in index a9fddd4723a..0ceab9a6d71 100644 --- a/dbms/src/Common/config_build.cpp.in +++ b/dbms/src/Common/config_build.cpp.in @@ -47,8 +47,6 @@ const char * auto_config_build[]{ "@GLIBC_COMPATIBILITY@", "USE_JEMALLOC", "@USE_JEMALLOC@", - "USE_TCMALLOC", - "@USE_TCMALLOC@", "USE_MIMALLOC", "@USE_MIMALLOC@", "USE_UNWIND", diff --git a/dbms/src/Common/tests/allocator_perf.cpp b/dbms/src/Common/tests/allocator_perf.cpp index 2298cc63089..016967f6776 100644 --- a/dbms/src/Common/tests/allocator_perf.cpp +++ b/dbms/src/Common/tests/allocator_perf.cpp @@ -94,13 +94,11 @@ void dump_profile(int type) { fmt::print( R"raw(USE_JEMALLOC={};USE_JEMALLOC_PROF={} -USE_TCMALLOC={} USE_MIMALLOC={} type={} )raw", USE_JEMALLOC, USE_JEMALLOC_PROF, - USE_TCMALLOC, USE_MIMALLOC, type); } diff --git a/dbms/src/DataStreams/CreatingSetsBlockInputStream.cpp b/dbms/src/DataStreams/CreatingSetsBlockInputStream.cpp index 22545327edd..134645e56e3 100644 --- a/dbms/src/DataStreams/CreatingSetsBlockInputStream.cpp +++ b/dbms/src/DataStreams/CreatingSetsBlockInputStream.cpp @@ -257,6 +257,8 @@ void CreatingSetsBlockInputStream::createOne(SubqueryForSet & subquery) if (subquery.join) subquery.join->setTotals(profiling_in->getTotals()); } + if (subquery.join) + head_rows = subquery.join->getTotalBuildInputRows(); if (head_rows != 0) { diff --git a/dbms/src/DataStreams/ParallelAggregatingBlockInputStream.cpp b/dbms/src/DataStreams/ParallelAggregatingBlockInputStream.cpp index cd9d6235f52..4d85abcb3b2 100644 --- a/dbms/src/DataStreams/ParallelAggregatingBlockInputStream.cpp +++ b/dbms/src/DataStreams/ParallelAggregatingBlockInputStream.cpp @@ -195,10 +195,9 @@ void ParallelAggregatingBlockInputStream::Handler::onException(std::exception_pt Int32 old_value = -1; parent.first_exception_index.compare_exchange_strong(old_value, static_cast(thread_num), std::memory_order_seq_cst, std::memory_order_relaxed); - /// can not cancel parent inputStream or the exception might be lost if (!parent.executed) /// use cancel instead of kill to avoid too many useless error message - parent.processor.cancel(false); + parent.cancel(false); } diff --git a/dbms/src/Flash/Coprocessor/DAGContext.h b/dbms/src/Flash/Coprocessor/DAGContext.h index 7bfc67afcad..1e9d2d3628e 100644 --- a/dbms/src/Flash/Coprocessor/DAGContext.h +++ b/dbms/src/Flash/Coprocessor/DAGContext.h @@ -52,6 +52,7 @@ struct JoinExecuteInfo String build_side_root_executor_id; JoinPtr join_ptr; BlockInputStreams non_joined_streams; + BlockInputStreams join_build_streams; }; using MPPTunnelSetPtr = std::shared_ptr; @@ -133,7 +134,6 @@ class DAGContext , collect_execution_summaries(dag_request->has_collect_execution_summaries() && dag_request->collect_execution_summaries()) , is_mpp_task(false) , is_root_mpp_task(false) - , tunnel_set(nullptr) , flags(dag_request->flags()) , sql_mode(dag_request->sql_mode()) , max_recorded_error_count(getMaxErrorCount(*dag_request)) @@ -153,7 +153,6 @@ class DAGContext , return_executor_id(true) , is_mpp_task(true) , is_root_mpp_task(is_root_mpp_task_) - , tunnel_set(nullptr) , flags(dag_request->flags()) , sql_mode(dag_request->sql_mode()) , mpp_task_meta(meta_) @@ -175,7 +174,6 @@ class DAGContext , collect_execution_summaries(false) , is_mpp_task(false) , is_root_mpp_task(false) - , tunnel_set(nullptr) , flags(0) , sql_mode(0) , max_recorded_error_count(max_error_count_) @@ -190,7 +188,6 @@ class DAGContext , initialize_concurrency(concurrency) , is_mpp_task(true) , is_root_mpp_task(false) - , tunnel_set(nullptr) , log(Logger::get(log_identifier)) , flags(dag_request->flags()) , sql_mode(dag_request->sql_mode()) @@ -343,6 +340,7 @@ class DAGContext bool is_mpp_task = false; bool is_root_mpp_task = false; bool is_batch_cop = false; + // `tunnel_set` is always set by `MPPTask` and is intended to be used for `DAGQueryBlockInterpreter`. MPPTunnelSetPtr tunnel_set; TablesRegionsInfo tables_regions_info; // part of regions_for_local_read + regions_for_remote_read, only used for batch-cop diff --git a/dbms/src/Flash/Coprocessor/DAGQueryBlockInterpreter.cpp b/dbms/src/Flash/Coprocessor/DAGQueryBlockInterpreter.cpp index d67a8f6ec52..7fa32c316a1 100644 --- a/dbms/src/Flash/Coprocessor/DAGQueryBlockInterpreter.cpp +++ b/dbms/src/Flash/Coprocessor/DAGQueryBlockInterpreter.cpp @@ -264,6 +264,8 @@ void DAGQueryBlockInterpreter::handleJoin(const tipb::Join & join, DAGPipeline & recordJoinExecuteInfo(tiflash_join.build_side_index, join_ptr); + auto & join_execute_info = dagContext().getJoinExecuteInfoMap()[query_block.source_name]; + size_t join_build_concurrency = settings.join_concurrent_build ? std::min(max_streams, build_pipeline.streams.size()) : 1; /// build side streams @@ -274,6 +276,7 @@ void DAGQueryBlockInterpreter::handleJoin(const tipb::Join & join, DAGPipeline & stream = std::make_shared(stream, join_ptr, get_concurrency_build_index(), log->identifier()); stream->setExtraInfo( fmt::format("join build, build_side_root_executor_id = {}", dagContext().getJoinExecuteInfoMap()[query_block.source_name].build_side_root_executor_id)); + join_execute_info.join_build_streams.push_back(stream); }); // for test, join executor need the return blocks to output. executeUnion(build_pipeline, max_streams, log, /*ignore_block=*/!dagContext().isTest(), "for join"); @@ -294,7 +297,6 @@ void DAGQueryBlockInterpreter::handleJoin(const tipb::Join & join, DAGPipeline & /// add join input stream if (is_tiflash_right_join) { - auto & join_execute_info = dagContext().getJoinExecuteInfoMap()[query_block.source_name]; size_t not_joined_concurrency = join_ptr->getNotJoinedStreamConcurrency(); for (size_t i = 0; i < not_joined_concurrency; ++i) { diff --git a/dbms/src/Flash/Coprocessor/DAGResponseWriter.cpp b/dbms/src/Flash/Coprocessor/DAGResponseWriter.cpp index 33f6d99f9d8..fe4e9c23bca 100644 --- a/dbms/src/Flash/Coprocessor/DAGResponseWriter.cpp +++ b/dbms/src/Flash/Coprocessor/DAGResponseWriter.cpp @@ -120,17 +120,13 @@ void DAGResponseWriter::addExecuteSummaries(tipb::SelectResponse & response, boo auto it = dag_context.getJoinExecuteInfoMap().find(join_executor_id); if (it != dag_context.getJoinExecuteInfoMap().end()) { - auto build_side_it = dag_context.getProfileStreamsMap().find(it->second.build_side_root_executor_id); - if (build_side_it != dag_context.getProfileStreamsMap().end()) + UInt64 process_time_for_build = 0; + for (const auto & join_build_stream : it->second.join_build_streams) { - UInt64 process_time_for_build = 0; - for (const auto & join_build_stream : build_side_it->second) - { - if (auto * p_stream = dynamic_cast(join_build_stream.get())) - process_time_for_build = std::max(process_time_for_build, p_stream->getProfileInfo().execution_time); - } - current.time_processed_ns += process_time_for_build; + if (auto * p_stream = dynamic_cast(join_build_stream.get()); p_stream) + process_time_for_build = std::max(process_time_for_build, p_stream->getProfileInfo().execution_time); } + current.time_processed_ns += process_time_for_build; } } } diff --git a/dbms/src/Flash/Coprocessor/DAGStorageInterpreter.cpp b/dbms/src/Flash/Coprocessor/DAGStorageInterpreter.cpp index 390ce7b9948..fb4a03f0999 100644 --- a/dbms/src/Flash/Coprocessor/DAGStorageInterpreter.cpp +++ b/dbms/src/Flash/Coprocessor/DAGStorageInterpreter.cpp @@ -599,6 +599,7 @@ std::unordered_map DAGStorageInterpreter::generateSele analyzer->getCurrentInputColumns(), context.getTimezoneInfo()); query_info.req_id = fmt::format("{} Table<{}>", log->identifier(), table_id); + query_info.keep_order = table_scan.keepOrder(); return query_info; }; if (table_scan.isPartitionTableScan()) diff --git a/dbms/src/Flash/Coprocessor/TiDBTableScan.cpp b/dbms/src/Flash/Coprocessor/TiDBTableScan.cpp index 7d7ad2f2b57..dca738a2e6c 100644 --- a/dbms/src/Flash/Coprocessor/TiDBTableScan.cpp +++ b/dbms/src/Flash/Coprocessor/TiDBTableScan.cpp @@ -24,6 +24,9 @@ TiDBTableScan::TiDBTableScan( , executor_id(executor_id_) , is_partition_table_scan(table_scan->tp() == tipb::TypePartitionTableScan) , columns(is_partition_table_scan ? table_scan->partition_table_scan().columns() : table_scan->tbl_scan().columns()) + // Only No-partition table need keep order when tablescan executor required keep order. + // If keep_order is not set, keep order for safety. + , keep_order(!is_partition_table_scan && (table_scan->tbl_scan().keep_order() || !table_scan->tbl_scan().has_keep_order())) { if (is_partition_table_scan) { diff --git a/dbms/src/Flash/Coprocessor/TiDBTableScan.h b/dbms/src/Flash/Coprocessor/TiDBTableScan.h index 6ac07d326f6..a5a463a8ff2 100644 --- a/dbms/src/Flash/Coprocessor/TiDBTableScan.h +++ b/dbms/src/Flash/Coprocessor/TiDBTableScan.h @@ -51,6 +51,10 @@ class TiDBTableScan { return executor_id; } + bool keepOrder() const + { + return keep_order; + } private: const tipb::Executor * table_scan; @@ -66,6 +70,7 @@ class TiDBTableScan /// physical_table_ids contains the table ids of its partitions std::vector physical_table_ids; Int64 logical_table_id; + bool keep_order; }; } // namespace DB diff --git a/dbms/src/Flash/EstablishCall.cpp b/dbms/src/Flash/EstablishCall.cpp index 6dece74d007..32b2a803035 100644 --- a/dbms/src/Flash/EstablishCall.cpp +++ b/dbms/src/Flash/EstablishCall.cpp @@ -86,7 +86,7 @@ void EstablishCallData::initRpc() } if (eptr) { - state = FINISH; + setFinishState("initRpc called"); grpc::Status status(static_cast(GRPC_STATUS_UNKNOWN), getExceptionMessage(eptr, false)); responderFinish(status); } @@ -112,9 +112,20 @@ void EstablishCallData::writeErr(const mpp::MPPDataPacket & packet) err_status = grpc::Status(grpc::StatusCode::UNKNOWN, "Write error message failed for unknown reason."); } -void EstablishCallData::writeDone(const ::grpc::Status & status) +void EstablishCallData::setFinishState(const String & msg) { state = FINISH; + if (async_tunnel_sender && !async_tunnel_sender->isConsumerFinished()) + { + async_tunnel_sender->consumerFinish(fmt::format("{}: {}", + async_tunnel_sender->getTunnelId(), + msg)); //trigger mpp tunnel finish work + } +} + +void EstablishCallData::writeDone(const ::grpc::Status & status) +{ + setFinishState("writeDone called"); if (stopwatch) { LOG_FMT_INFO(async_tunnel_sender->getLogger(), "connection for {} cost {} ms.", async_tunnel_sender->getTunnelId(), stopwatch->elapsedMilliseconds()); @@ -140,12 +151,7 @@ void EstablishCallData::cancel() void EstablishCallData::finishTunnelAndResponder() { - state = FINISH; - if (async_tunnel_sender) - { - async_tunnel_sender->consumerFinish(fmt::format("{}: finishTunnelAndResponder called.", - async_tunnel_sender->getTunnelId())); //trigger mpp tunnel finish work - } + setFinishState("finishTunnelAndResponder called"); grpc::Status status(static_cast(GRPC_STATUS_UNKNOWN), "Consumer exits unexpected, grpc writes failed."); responder.Finish(status, this); } @@ -174,7 +180,6 @@ void EstablishCallData::proceed() } else if (state == ERR_HANDLE) { - state = FINISH; writeDone(err_status); } else diff --git a/dbms/src/Flash/EstablishCall.h b/dbms/src/Flash/EstablishCall.h index 4a62e5f0c6b..1faa988a49c 100644 --- a/dbms/src/Flash/EstablishCall.h +++ b/dbms/src/Flash/EstablishCall.h @@ -84,6 +84,9 @@ class EstablishCallData : public PacketWriter void finishTunnelAndResponder(); + // will try to call async_sender's consumerFinish if needed + void setFinishState(const String & msg); + void responderFinish(const grpc::Status & status); std::mutex mu; diff --git a/dbms/src/Flash/Mpp/MPPTunnel.cpp b/dbms/src/Flash/Mpp/MPPTunnel.cpp index 582fe3c2796..716d3c0207a 100644 --- a/dbms/src/Flash/Mpp/MPPTunnel.cpp +++ b/dbms/src/Flash/Mpp/MPPTunnel.cpp @@ -349,11 +349,18 @@ void SyncTunnelSender::startSendThread() void AsyncTunnelSender::tryFlushOne() { + // When consumer finished, sending work is done already, just return + if (consumer_state.msgHasSet()) + return; writer->tryFlushOne(); } void AsyncTunnelSender::sendOne() { + // When consumer finished, sending work is done already, just return + if (consumer_state.msgHasSet()) + return; + String err_msg; bool queue_empty_flag = false; try diff --git a/dbms/src/Flash/Statistics/ExecutorStatistics.h b/dbms/src/Flash/Statistics/ExecutorStatistics.h index 4ed5bb02aaa..6b5fb88a971 100644 --- a/dbms/src/Flash/Statistics/ExecutorStatistics.h +++ b/dbms/src/Flash/Statistics/ExecutorStatistics.h @@ -84,10 +84,11 @@ class ExecutorStatistics : public ExecutorStatisticsBase { for (const auto & input_stream : it->second) { - auto * p_stream = dynamic_cast(input_stream.get()); - assert(p_stream); - const auto & profile_info = p_stream->getProfileInfo(); - base.append(profile_info); + if (auto * p_stream = dynamic_cast(input_stream.get()); p_stream) + { + const auto & profile_info = p_stream->getProfileInfo(); + base.append(profile_info); + } } } if constexpr (ExecutorImpl::has_extra_info) diff --git a/dbms/src/Flash/Statistics/JoinImpl.cpp b/dbms/src/Flash/Statistics/JoinImpl.cpp index 35f8c5b40bd..9faf3da16eb 100644 --- a/dbms/src/Flash/Statistics/JoinImpl.cpp +++ b/dbms/src/Flash/Statistics/JoinImpl.cpp @@ -21,13 +21,18 @@ void JoinStatistics::appendExtraJson(FmtBuffer & fmt_buffer) const { fmt_buffer.fmtAppend( R"("hash_table_bytes":{},"build_side_child":"{}",)" - R"("non_joined_outbound_rows":{},"non_joined_outbound_blocks":{},"non_joined_outbound_bytes":{},"non_joined_execution_time_ns":{})", + R"("non_joined_outbound_rows":{},"non_joined_outbound_blocks":{},"non_joined_outbound_bytes":{},"non_joined_execution_time_ns":{},)" + R"("join_build_inbound_rows":{},"join_build_inbound_blocks":{},"join_build_inbound_bytes":{},"join_build_execution_time_ns":{})", hash_table_bytes, build_side_child, non_joined_base.rows, non_joined_base.blocks, non_joined_base.bytes, - non_joined_base.execution_time_ns); + non_joined_base.execution_time_ns, + join_build_base.rows, + join_build_base.blocks, + join_build_base.bytes, + join_build_base.execution_time_ns); } void JoinStatistics::collectExtraRuntimeDetail() @@ -41,10 +46,19 @@ void JoinStatistics::collectExtraRuntimeDetail() build_side_child = join_execute_info.build_side_root_executor_id; for (const auto & non_joined_stream : join_execute_info.non_joined_streams) { - auto * p_stream = dynamic_cast(non_joined_stream.get()); - assert(p_stream); - const auto & profile_info = p_stream->getProfileInfo(); - non_joined_base.append(profile_info); + if (auto * p_stream = dynamic_cast(non_joined_stream.get()); p_stream) + { + const auto & profile_info = p_stream->getProfileInfo(); + non_joined_base.append(profile_info); + } + } + for (const auto & join_build_stream : join_execute_info.join_build_streams) + { + if (auto * p_stream = dynamic_cast(join_build_stream.get()); p_stream) + { + const auto & profile_info = p_stream->getProfileInfo(); + join_build_base.append(profile_info); + } } } } diff --git a/dbms/src/Flash/Statistics/JoinImpl.h b/dbms/src/Flash/Statistics/JoinImpl.h index a9e0ef8418e..3fedb77f2ff 100644 --- a/dbms/src/Flash/Statistics/JoinImpl.h +++ b/dbms/src/Flash/Statistics/JoinImpl.h @@ -44,6 +44,8 @@ class JoinStatistics : public JoinStatisticsBase BaseRuntimeStatistics non_joined_base; + BaseRuntimeStatistics join_build_base; + protected: void appendExtraJson(FmtBuffer &) const override; void collectExtraRuntimeDetail() override; diff --git a/dbms/src/Flash/tests/gtest_aggregation_executor.cpp b/dbms/src/Flash/tests/gtest_aggregation_executor.cpp new file mode 100644 index 00000000000..0cf865b5d20 --- /dev/null +++ b/dbms/src/Flash/tests/gtest_aggregation_executor.cpp @@ -0,0 +1,347 @@ +// Copyright 2022 PingCAP, Ltd. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +#include +#include + +namespace DB +{ +namespace tests +{ + +#define DT DecimalField +#define COL_GROUP2(a, b) \ + { \ + col(types_col_name[a]), col(types_col_name[b]) \ + } +#define COL_PROJ2(a, b) \ + { \ + types_col_name[a], types_col_name[b] \ + } + +class ExecutorAggTestRunner : public DB::tests::ExecutorTest +{ +public: + using ColStringNullableType = std::optional::FieldType>; + using ColInt8NullableType = std::optional::FieldType>; + using ColInt16NullableType = std::optional::FieldType>; + using ColInt32NullableType = std::optional::FieldType>; + using ColInt64NullableType = std::optional::FieldType>; + using ColFloat32NullableType = std::optional::FieldType>; + using ColFloat64NullableType = std::optional::FieldType>; + using ColMyDateNullableType = std::optional::FieldType>; + using ColMyDateTimeNullableType = std::optional::FieldType>; + using ColDecimalNullableType = std::optional::FieldType>; + using ColUInt64Type = typename TypeTraits::FieldType; + + using ColumnWithNullableString = std::vector; + using ColumnWithNullableInt8 = std::vector; + using ColumnWithNullableInt16 = std::vector; + using ColumnWithNullableInt32 = std::vector; + using ColumnWithNullableInt64 = std::vector; + using ColumnWithNullableFloat32 = std::vector; + using ColumnWithNullableFloat64 = std::vector; + using ColumnWithNullableMyDate = std::vector; + using ColumnWithNullableMyDateTime = std::vector; + using ColumnWithNullableDecimal = std::vector; + using ColumnWithUInt64 = std::vector; + + void initializeContext() override + { + ExecutorTest::initializeContext(); + + /// Create table for tests of group by + context.addMockTable(/* name= */ {db_name, table_types}, + /* columnInfos= */ + {{types_col_name[0], TiDB::TP::TypeLong}, + {types_col_name[1], TiDB::TP::TypeDecimal}, + {types_col_name[2], TiDB::TP::TypeTiny}, + {types_col_name[3], TiDB::TP::TypeShort}, + {types_col_name[4], TiDB::TP::TypeLong}, + {types_col_name[5], TiDB::TP::TypeLongLong}, + {types_col_name[6], TiDB::TP::TypeFloat}, + {types_col_name[7], TiDB::TP::TypeDouble}, + {types_col_name[8], TiDB::TP::TypeDate}, + {types_col_name[9], TiDB::TP::TypeDatetime}, + {types_col_name[10], TiDB::TP::TypeString}}, + /* columns= */ + {toNullableVec(types_col_name[0], col_id), + toNullableVec(types_col_name[1], col_decimal), + toNullableVec(types_col_name[2], col_tinyint), + toNullableVec(types_col_name[3], col_smallint), + toNullableVec(types_col_name[4], col_int), + toNullableVec(types_col_name[5], col_bigint), + toNullableVec(types_col_name[6], col_float), + toNullableVec(types_col_name[7], col_double), + toNullableVec(types_col_name[8], col_mydate), + toNullableVec(types_col_name[9], col_mydatetime), + toNullableVec(types_col_name[10], col_string)}); + + /// Create table for tests of aggregation functions + context.addMockTable(/* name= */ {db_name, table_name}, + /* columnInfos= */ + {{col_name[0], TiDB::TP::TypeLong}, + {col_name[1], TiDB::TP::TypeString}, + {col_name[2], TiDB::TP::TypeString}, + {col_name[3], TiDB::TP::TypeDouble}}, + /* columns= */ + {toNullableVec(col_name[0], col_age), + toNullableVec(col_name[1], col_gender), + toNullableVec(col_name[2], col_country), + toNullableVec(col_name[3], col_salary)}); + + context.addMockTable({"aggnull_test", "t1"}, + {{"s1", TiDB::TP::TypeString}, {"s2", TiDB::TP::TypeString}}, + {toNullableVec("s1", {"banana", {}, "banana"}), + toNullableVec("s2", {"apple", {}, "banana"})}); + } + + std::shared_ptr buildDAGRequest(std::pair src, MockAstVec agg_funcs, MockAstVec group_by_exprs, MockColumnNameVec proj) + { + /// We can filter the group by column with project operator. + /// project is applied to get partial aggregation output, so that we can remove redundant outputs and compare results with less handwriting codes. + return context.scan(src.first, src.second).aggregation(agg_funcs, group_by_exprs).project(proj).build(context); + } + + void executeWithConcurrency(const std::shared_ptr & request, const ColumnsWithTypeAndName & expect_columns) + { + for (size_t i = 1; i <= max_concurrency; i += step) + ASSERT_COLUMNS_EQ_UR(expect_columns, executeStreams(request, i)); + } + + static const size_t max_concurrency = 10; + static const size_t step = 2; + + const String db_name{"test_db"}; + + /// Prepare some data and names for tests of group by + const String table_types{"types"}; + const std::vector types_col_name{"id", "decimal_", "tinyint_", "smallint_", "int_", "bigint_", "float_", "double_", "date_", "datetime_", "string_"}; + ColumnWithNullableInt32 col_id{1, 2, 3, 4, 5, 6, 7, 8, 9}; + ColumnWithNullableDecimal col_decimal{DT(55, 1), {}, DT(-24, 1), DT(40, 1), DT(-40, 1), DT(40, 1), {}, DT(55, 1), DT(0, 1)}; + ColumnWithNullableInt8 col_tinyint{1, 2, 3, {}, {}, 0, 0, -1, -2}; + ColumnWithNullableInt16 col_smallint{2, 3, {}, {}, 0, -1, -2, 4, 0}; + ColumnWithNullableInt32 col_int{4, {}, {}, 0, 123, -1, -1, 123, 4}; + ColumnWithNullableInt64 col_bigint{2, 2, {}, 0, -1, {}, -1, 0, 123}; + ColumnWithNullableFloat32 col_float{3.3, {}, 0, 4.0, 3.3, 5.6, -0.1, -0.1, {}}; + ColumnWithNullableFloat64 col_double{0.1, 0, 1.1, 1.1, 1.2, {}, {}, -1.2, -1.2}; + ColumnWithNullableMyDate col_mydate{1000000, 2000000, {}, 300000, 1000000, {}, 0, 2000000, {}}; + ColumnWithNullableMyDateTime col_mydatetime{2000000, 0, {}, 3000000, 1000000, {}, 0, 2000000, 1000000}; + ColumnWithNullableString col_string{{}, "pingcap", "PingCAP", {}, "PINGCAP", "PingCAP", {}, "Shanghai", "Shanghai"}; + + /// Prepare some data and names for aggregation functions + const String table_name{"clerk"}; + const std::vector col_name{"age", "gender", "country", "salary"}; + ColumnWithNullableInt32 col_age{30, {}, 27, 32, 25, 36, {}, 22, 34}; + ColumnWithNullableString col_gender{ + "male", + "female", + "female", + "male", + "female", + "female", + "male", + "female", + "male", + }; + ColumnWithNullableString col_country{"russia", "korea", "usa", "usa", "usa", "china", "china", "china", "china"}; + ColumnWithNullableFloat64 col_salary{1000.1, 1300.2, 0.3, {}, -200.4, 900.5, -999.6, 2000.7, -300.8}; +}; + +/// Guarantee the correctness of group by +TEST_F(ExecutorAggTestRunner, GroupBy) +try +{ + WRAP_FOR_DIS_ENABLE_PLANNER_BEGIN + std::shared_ptr request; + std::vector group_by_exprs; + std::vector projections; + std::vector expect_cols; + size_t test_num; + + { + /// group by single column + group_by_exprs = {{col(types_col_name[2])}, {col(types_col_name[3])}, {col(types_col_name[4])}, {col(types_col_name[5])}, {col(types_col_name[6])}, {col(types_col_name[7])}, {col(types_col_name[8])}, {col(types_col_name[9])}, {col(types_col_name[10])}}; + projections = {{types_col_name[2]}, {types_col_name[3]}, {types_col_name[4]}, {types_col_name[5]}, {types_col_name[6]}, {types_col_name[7]}, {types_col_name[8]}, {types_col_name[9]}, {types_col_name[10]}}; + expect_cols = { + {toNullableVec(types_col_name[2], ColumnWithNullableInt8{-1, 2, {}, 0, 1, 3, -2})}, /// select tinyint_ from test_db.types group by tinyint_; + {toNullableVec(types_col_name[3], ColumnWithNullableInt16{-1, 2, -2, {}, 0, 4, 3})}, /// select smallint_ from test_db.types group by smallint_; + {toNullableVec(types_col_name[4], ColumnWithNullableInt32{-1, {}, 4, 0, 123})}, /// select int_ from test_db.types group by int_; + {toNullableVec(types_col_name[5], ColumnWithNullableInt64{2, -1, 0, 123, {}})}, /// select bigint_ from test_db.types group by bigint_; + {toNullableVec(types_col_name[6], ColumnWithNullableFloat32{0, 4, 3.3, {}, 5.6, -0.1})}, /// select float_ from test_db.types group by float_; + {toNullableVec(types_col_name[7], ColumnWithNullableFloat64{0, {}, -1.2, 1.1, 1.2, 0.1})}, /// select double_ from test_db.types group by double_; + {toNullableVec(types_col_name[8], ColumnWithNullableMyDate{{}, 0, 300000, 1000000, 2000000})}, /// select date_ from test_db.types group by date_; + {toNullableVec(types_col_name[9], ColumnWithNullableMyDateTime{{}, 0, 1000000, 2000000, 3000000})}, /// select datetime_ from test_db.types group by datetime_; + {toNullableVec(types_col_name[10], ColumnWithNullableString{{}, "pingcap", "PingCAP", "PINGCAP", "Shanghai"})}}; /// select string_ from test_db.types group by string_; + test_num = expect_cols.size(); + ASSERT_EQ(group_by_exprs.size(), test_num); + ASSERT_EQ(projections.size(), test_num); + + for (size_t i = 0; i < test_num; ++i) + { + request = buildDAGRequest(std::make_pair(db_name, table_types), {}, group_by_exprs[i], projections[i]); + executeWithConcurrency(request, expect_cols[i]); + } + } + + { + /// group by two columns + group_by_exprs = {COL_GROUP2(2, 6), COL_GROUP2(3, 9), COL_GROUP2(4, 7), COL_GROUP2(5, 10), COL_GROUP2(8, 9), COL_GROUP2(9, 10)}; + projections = {COL_PROJ2(2, 6), COL_PROJ2(3, 9), COL_PROJ2(4, 7), COL_PROJ2(5, 10), COL_PROJ2(8, 9), COL_PROJ2(9, 10)}; + expect_cols = {/// select tinyint_, float_ from test_db.types group by tinyint_, float_; + {toNullableVec(types_col_name[2], ColumnWithNullableInt8{1, 2, {}, 3, 0, 0, -1, {}, -2}), + toNullableVec(types_col_name[6], ColumnWithNullableFloat32{3.3, {}, 4, 0, -0.1, 5.6, -0.1, 3.3, {}})}, + /// select smallint_, datetime_ from test_db.types group by smallint_, datetime_; + {toNullableVec(types_col_name[3], ColumnWithNullableInt16{2, 3, {}, {}, 0, -1, -2, 4}), + toNullableVec(types_col_name[9], ColumnWithNullableMyDateTime{2000000, 0, {}, 3000000, 1000000, {}, 0, 2000000})}, + /// select int_, double_ from test_db.types group by int_, double_; + {toNullableVec(types_col_name[4], ColumnWithNullableInt32{{}, 123, -1, 0, {}, 4, 4, 123}), + toNullableVec(types_col_name[7], ColumnWithNullableFloat64{0, -1.2, {}, 1.1, 1.1, -1.2, 0.1, 1.2})}, + /// select bigint_, string_ from test_db.types group by bigint_, string_; + {toNullableVec(types_col_name[5], ColumnWithNullableInt64{-1, 0, 0, 123, 2, {}, -1, 2}), + toNullableVec(types_col_name[10], ColumnWithNullableString{{}, {}, "Shanghai", "Shanghai", {}, "PingCAP", "PINGCAP", "pingcap"})}, + /// select date_, datetime_ from test_db.types group by date_, datetime_; + {toNullableVec(types_col_name[8], ColumnWithNullableMyDate{1000000, 2000000, {}, 300000, 1000000, 0, 2000000, {}}), + toNullableVec(types_col_name[9], ColumnWithNullableMyDateTime{2000000, 0, {}, 3000000, 1000000, 0, 2000000, 1000000})}, + /// select datetime_, string_ from test_db.types group by datetime_, string_; + {toNullableVec(types_col_name[9], ColumnWithNullableMyDateTime{2000000, 0, {}, 3000000, 1000000, 0, 2000000, 1000000}), + toNullableVec(types_col_name[10], ColumnWithNullableString{{}, "pingcap", "PingCAP", {}, "PINGCAP", {}, "Shanghai", "Shanghai"})}}; + test_num = expect_cols.size(); + ASSERT_EQ(group_by_exprs.size(), test_num); + ASSERT_EQ(projections.size(), test_num); + + for (size_t i = 0; i < test_num; ++i) + { + request = buildDAGRequest(std::make_pair(db_name, table_types), {}, group_by_exprs[i], projections[i]); + executeWithConcurrency(request, expect_cols[i]); + } + } + + /// TODO type: decimal, enum and unsigned numbers + WRAP_FOR_DIS_ENABLE_PLANNER_END +} +CATCH + +TEST_F(ExecutorAggTestRunner, AggregationMaxAndMin) +try +{ + WRAP_FOR_DIS_ENABLE_PLANNER_BEGIN + std::shared_ptr request; + auto agg_func0 = Max(col(col_name[0])); /// select max(age) from clerk group by country; + auto agg_func1 = Max(col(col_name[3])); /// select max(salary) from clerk group by country, gender; + + auto group_by_expr0 = col(col_name[2]); + auto group_by_expr10 = col(col_name[2]); + auto group_by_expr11 = col(col_name[1]); + + /// Prepare some data for max function test + std::vector expect_cols{ + {toNullableVec("max(age)", ColumnWithNullableInt32{36, 32, 30, {}})}, + {toNullableVec("max(salary)", ColumnWithNullableFloat64{2000.7, 1300.2, 1000.1, 0.3, -300.8, {}})}}; + std::vector group_by_exprs{{group_by_expr0}, {group_by_expr10, group_by_expr11}}; + std::vector projections{{"max(age)"}, {"max(salary)"}}; + std::vector agg_funcs{{agg_func0}, {agg_func1}}; + size_t test_num = expect_cols.size(); + + /// Start to test max function + for (size_t i = 0; i < test_num; ++i) + { + request = buildDAGRequest(std::make_pair(db_name, table_name), agg_funcs[i], group_by_exprs[i], projections[i]); + executeWithConcurrency(request, expect_cols[i]); + } + + /// Min function tests + + agg_func0 = Min(col(col_name[0])); /// select min(age) from clerk group by country; + agg_func1 = Min(col(col_name[3])); /// select min(salary) from clerk group by country, gender; + + expect_cols = { + {toNullableVec("min(age)", ColumnWithNullableInt32{30, 25, 22, {}})}, + {toNullableVec("min(salary)", ColumnWithNullableFloat64{1300.2, 1000.1, 900.5, -200.4, -999.6, {}})}}; + projections = {{"min(age)"}, {"min(salary)"}}; + agg_funcs = {{agg_func0}, {agg_func1}}; + test_num = expect_cols.size(); + + /// Start to test min function + for (size_t i = 0; i < test_num; ++i) + { + request = buildDAGRequest(std::make_pair(db_name, table_name), agg_funcs[i], group_by_exprs[i], projections[i]); + executeWithConcurrency(request, expect_cols[i]); + } + WRAP_FOR_DIS_ENABLE_PLANNER_END +} +CATCH + +TEST_F(ExecutorAggTestRunner, AggregationCount) +try +{ + WRAP_FOR_DIS_ENABLE_PLANNER_BEGIN + /// Prepare some data + std::shared_ptr request; + auto agg_func0 = Count(col(col_name[0])); /// select count(age) from clerk group by country; + auto agg_func1 = Count(col(col_name[1])); /// select count(gender) from clerk group by country, gender; + std::vector agg_funcs = {{agg_func0}, {agg_func1}}; + + auto group_by_expr0 = col(col_name[2]); + auto group_by_expr10 = col(col_name[2]); + auto group_by_expr11 = col(col_name[1]); + + std::vector expect_cols{ + {toVec("count(age)", ColumnWithUInt64{3, 3, 1, 0})}, + {toVec("count(gender)", ColumnWithUInt64{2, 2, 2, 1, 1, 1})}}; + std::vector group_by_exprs{{group_by_expr0}, {group_by_expr10, group_by_expr11}}; + std::vector projections{{"count(age)"}, {"count(gender)"}}; + size_t test_num = expect_cols.size(); + + /// Start to test + for (size_t i = 0; i < test_num; ++i) + { + request = buildDAGRequest(std::make_pair(db_name, table_name), {agg_funcs[i]}, group_by_exprs[i], projections[i]); + executeWithConcurrency(request, expect_cols[i]); + } + WRAP_FOR_DIS_ENABLE_PLANNER_END +} +CATCH + +TEST_F(ExecutorAggTestRunner, AggNull) +try +{ + WRAP_FOR_DIS_ENABLE_PLANNER_BEGIN + auto request = context + .scan("aggnull_test", "t1") + .aggregation({Max(col("s1"))}, {}) + .build(context); + { + ASSERT_COLUMNS_EQ_R(executeStreams(request), + createColumns({toNullableVec({"banana"})})); + } + + request = context + .scan("aggnull_test", "t1") + .aggregation({}, {col("s1")}) + .build(context); + { + ASSERT_COLUMNS_EQ_R(executeStreams(request), + createColumns({toNullableVec("s1", {{}, "banana"})})); + } + WRAP_FOR_DIS_ENABLE_PLANNER_END +} +CATCH + + +// TODO support more type of min, max, count. +// support more aggregation functions: sum, forst_row, group_concat + +} // namespace tests +} // namespace DB diff --git a/dbms/src/Flash/tests/gtest_executor.cpp b/dbms/src/Flash/tests/gtest_executor.cpp index ee35af0c03d..b8fac961832 100644 --- a/dbms/src/Flash/tests/gtest_executor.cpp +++ b/dbms/src/Flash/tests/gtest_executor.cpp @@ -33,196 +33,32 @@ class ExecutorTestRunner : public DB::tests::ExecutorTest {{"s1", TiDB::TP::TypeString}, {"s2", TiDB::TP::TypeString}}, {toNullableVec("s1", {"banana", {}, "banana"}), toNullableVec("s2", {"apple", {}, "banana"})}); - - context.addExchangeReceiver("exchange_r_table", - {{"s1", TiDB::TP::TypeString}, {"join_c", TiDB::TP::TypeString}}, - {toNullableVec("s", {"banana", "banana"}), - toNullableVec("join_c", {"apple", "banana"})}); - - context.addExchangeReceiver("exchange_l_table", - {{"s1", TiDB::TP::TypeString}, {"join_c", TiDB::TP::TypeString}}, - {toNullableVec("s", {"banana", "banana"}), - toNullableVec("join_c", {"apple", "banana"})}); - - context.addMockTable({"test_db", "r_table"}, - {{"s", TiDB::TP::TypeString}, {"join_c", TiDB::TP::TypeString}}, - {toVec("s", {"banana", "banana"}), - toVec("join_c", {"apple", "banana"})}); - - context.addMockTable({"test_db", "r_table_2"}, - {{"s", TiDB::TP::TypeString}, {"join_c", TiDB::TP::TypeString}}, - {toVec("s", {"banana", "banana", "banana"}), - toVec("join_c", {"apple", "apple", "apple"})}); - - context.addMockTable({"test_db", "l_table"}, - {{"s", TiDB::TP::TypeString}, {"join_c", TiDB::TP::TypeString}}, - {toVec("s", {"banana", "banana"}), - toVec("join_c", {"apple", "banana"})}); } }; TEST_F(ExecutorTestRunner, Filter) try { - wrapForDisEnablePlanner([&]() { - auto request = context - .scan("test_db", "test_table") - .filter(eq(col("s1"), col("s2"))) - .build(context); - { - ASSERT_COLUMNS_EQ_R(executeStreams(request), - createColumns({toNullableVec({"banana"}), - toNullableVec({"banana"})})); - } - - request = context.receive("exchange1") - .filter(eq(col("s1"), col("s2"))) - .build(context); - { - ASSERT_COLUMNS_EQ_R(executeStreams(request), - createColumns({toNullableVec({"banana"}), - toNullableVec({"banana"})})); - } - }); -} -CATCH - -TEST_F(ExecutorTestRunner, JoinWithTableScan) -try -{ - wrapForDisEnablePlanner([&]() { - auto request = context - .scan("test_db", "l_table") - .join(context.scan("test_db", "r_table"), {col("join_c")}, ASTTableJoin::Kind::Left) - .topN("join_c", false, 2) - .build(context); - { - String expected = "topn_3 | order_by: {(<1, String>, desc: false)}, limit: 2\n" - " Join_2 | LeftOuterJoin, HashJoin. left_join_keys: {<0, String>}, right_join_keys: {<0, String>}\n" - " table_scan_0 | {<0, String>, <1, String>}\n" - " table_scan_1 | {<0, String>, <1, String>}\n"; - ASSERT_DAGREQUEST_EQAUL(expected, request); - ASSERT_COLUMNS_EQ_R(executeStreams(request, 2), - createColumns({toNullableVec({"banana", "banana"}), - toNullableVec({"apple", "banana"}), - toNullableVec({"banana", "banana"}), - toNullableVec({"apple", "banana"})})); - - ASSERT_COLUMNS_EQ_R(executeStreams(request, 5), - createColumns({toNullableVec({"banana", "banana"}), - toNullableVec({"apple", "banana"}), - toNullableVec({"banana", "banana"}), - toNullableVec({"apple", "banana"})})); - - ASSERT_COLUMNS_EQ_R(executeStreams(request), - createColumns({toNullableVec({"banana", "banana"}), - toNullableVec({"apple", "banana"}), - toNullableVec({"banana", "banana"}), - toNullableVec({"apple", "banana"})})); - } - request = context - .scan("test_db", "l_table") - .join(context.scan("test_db", "r_table"), {col("join_c")}, ASTTableJoin::Kind::Left) - .project({"s", "join_c"}) - .topN("join_c", false, 2) - .build(context); - { - String expected = "topn_4 | order_by: {(<1, String>, desc: false)}, limit: 2\n" - " project_3 | {<0, String>, <1, String>}\n" - " Join_2 | LeftOuterJoin, HashJoin. left_join_keys: {<0, String>}, right_join_keys: {<0, String>}\n" - " table_scan_0 | {<0, String>, <1, String>}\n" - " table_scan_1 | {<0, String>, <1, String>}\n"; - ASSERT_DAGREQUEST_EQAUL(expected, request); - ASSERT_COLUMNS_EQ_R(executeStreams(request, 2), - createColumns({toNullableVec({"banana", "banana"}), - toNullableVec({"apple", "banana"})})); - } - - request = context - .scan("test_db", "l_table") - .join(context.scan("test_db", "r_table_2"), {col("join_c")}, ASTTableJoin::Kind::Left) - .topN("join_c", false, 4) - .build(context); - { - String expected = "topn_3 | order_by: {(<1, String>, desc: false)}, limit: 4\n" - " Join_2 | LeftOuterJoin, HashJoin. left_join_keys: {<0, String>}, right_join_keys: {<0, String>}\n" - " table_scan_0 | {<0, String>, <1, String>}\n" - " table_scan_1 | {<0, String>, <1, String>}\n"; - ASSERT_DAGREQUEST_EQAUL(expected, request); - ASSERT_COLUMNS_EQ_R(executeStreams(request, 2), - createColumns({toNullableVec({"banana", "banana", "banana", "banana"}), - toNullableVec({"apple", "apple", "apple", "banana"}), - toNullableVec({"banana", "banana", "banana", {}}), - toNullableVec({"apple", "apple", "apple", {}})})); - ASSERT_COLUMNS_EQ_R(executeStreams(request, 3), - createColumns({toNullableVec({"banana", "banana", "banana", "banana"}), - toNullableVec({"apple", "apple", "apple", "banana"}), - toNullableVec({"banana", "banana", "banana", {}}), - toNullableVec({"apple", "apple", "apple", {}})})); - } - }); -} -CATCH - -TEST_F(ExecutorTestRunner, JoinWithExchangeReceiver) -try -{ - wrapForDisEnablePlanner([&]() { - auto request = context - .receive("exchange_l_table") - .join(context.receive("exchange_r_table"), {col("join_c")}, ASTTableJoin::Kind::Left) - .topN("join_c", false, 2) - .build(context); - { - String expected = "topn_3 | order_by: {(<1, String>, desc: false)}, limit: 2\n" - " Join_2 | LeftOuterJoin, HashJoin. left_join_keys: {<0, String>}, right_join_keys: {<0, String>}\n" - " exchange_receiver_0 | type:PassThrough, {<0, String>, <1, String>}\n" - " exchange_receiver_1 | type:PassThrough, {<0, String>, <1, String>}\n"; - ASSERT_DAGREQUEST_EQAUL(expected, request); - ASSERT_COLUMNS_EQ_R(executeStreams(request, 2), - createColumns({toNullableVec({"banana", "banana"}), - toNullableVec({"apple", "banana"}), - toNullableVec({"banana", "banana"}), - toNullableVec({"apple", "banana"})})); - - ASSERT_COLUMNS_EQ_R(executeStreams(request, 5), - createColumns({toNullableVec({"banana", "banana"}), - toNullableVec({"apple", "banana"}), - toNullableVec({"banana", "banana"}), - toNullableVec({"apple", "banana"})})); - - ASSERT_COLUMNS_EQ_R(executeStreams(request), - createColumns({toNullableVec({"banana", "banana"}), - toNullableVec({"apple", "banana"}), - toNullableVec({"banana", "banana"}), - toNullableVec({"apple", "banana"})})); - } - }); -} -CATCH + WRAP_FOR_DIS_ENABLE_PLANNER_BEGIN + auto request = context + .scan("test_db", "test_table") + .filter(eq(col("s1"), col("s2"))) + .build(context); + { + ASSERT_COLUMNS_EQ_R(executeStreams(request), + createColumns({toNullableVec({"banana"}), + toNullableVec({"banana"})})); + } -TEST_F(ExecutorTestRunner, JoinWithTableScanAndReceiver) -try -{ - wrapForDisEnablePlanner([&]() { - auto request = context - .scan("test_db", "l_table") - .join(context.receive("exchange_r_table"), {col("join_c")}, ASTTableJoin::Kind::Left) - .topN("join_c", false, 2) - .build(context); - { - String expected = "topn_3 | order_by: {(<1, String>, desc: false)}, limit: 2\n" - " Join_2 | LeftOuterJoin, HashJoin. left_join_keys: {<0, String>}, right_join_keys: {<0, String>}\n" - " table_scan_0 | {<0, String>, <1, String>}\n" - " exchange_receiver_1 | type:PassThrough, {<0, String>, <1, String>}\n"; - ASSERT_DAGREQUEST_EQAUL(expected, request); - ASSERT_COLUMNS_EQ_R(executeStreams(request, 2), - createColumns({toNullableVec({"banana", "banana"}), - toNullableVec({"apple", "banana"}), - toNullableVec({"banana", "banana"}), - toNullableVec({"apple", "banana"})})); - } - }); + request = context.receive("exchange1") + .filter(eq(col("s1"), col("s2"))) + .build(context); + { + ASSERT_COLUMNS_EQ_R(executeStreams(request), + createColumns({toNullableVec({"banana"}), + toNullableVec({"banana"})})); + } + WRAP_FOR_DIS_ENABLE_PLANNER_END } CATCH diff --git a/dbms/src/Flash/tests/gtest_join_executor.cpp b/dbms/src/Flash/tests/gtest_join_executor.cpp new file mode 100644 index 00000000000..736f33d15b2 --- /dev/null +++ b/dbms/src/Flash/tests/gtest_join_executor.cpp @@ -0,0 +1,477 @@ +// Copyright 2022 PingCAP, Ltd. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +#include + +#include + +namespace DB +{ +namespace tests +{ +class JoinExecutorTestRunner : public DB::tests::ExecutorTest +{ + static const size_t max_concurrency_level = 10; + +public: + void initializeContext() override + { + ExecutorTest::initializeContext(); + context.addMockTable({"test_db", "test_table"}, + {{"s1", TiDB::TP::TypeString}, {"s2", TiDB::TP::TypeString}}, + {toNullableVec("s1", {"banana", {}, "banana"}), + toNullableVec("s2", {"apple", {}, "banana"})}); + + context.addMockTable({"test_db", "r_table"}, + {{"s", TiDB::TP::TypeString}, {"join_c", TiDB::TP::TypeString}}, + {toVec("s", {"banana", "banana"}), + toVec("join_c", {"apple", "banana"})}); + + context.addMockTable({"test_db", "r_table_2"}, + {{"s", TiDB::TP::TypeString}, {"join_c", TiDB::TP::TypeString}}, + {toVec("s", {"banana", "banana", "banana"}), + toVec("join_c", {"apple", "apple", "apple"})}); + + context.addMockTable({"test_db", "l_table"}, + {{"s", TiDB::TP::TypeString}, {"join_c", TiDB::TP::TypeString}}, + {toVec("s", {"banana", "banana"}), + toVec("join_c", {"apple", "banana"})}); + + context.addMockTable("simple_test", "t1", {{"a", TiDB::TP::TypeString}, {"b", TiDB::TP::TypeString}}, {toNullableVec("a", {"1", "2", {}, "1", {}}), toNullableVec("b", {"3", "4", "3", {}, {}})}); + + context.addMockTable("simple_test", "t2", {{"a", TiDB::TP::TypeString}, {"b", TiDB::TP::TypeString}}, {toNullableVec("a", {"1", "3", {}, "1", {}}), toNullableVec("b", {"3", "4", "3", {}, {}})}); + + context.addMockTable("multi_test", "t1", {{"a", TiDB::TP::TypeLong}, {"b", TiDB::TP::TypeLong}, {"c", TiDB::TP::TypeLong}}, {toVec("a", {1, 3, 0}), toVec("b", {2, 2, 0}), toVec("c", {3, 2, 0})}); + + context.addMockTable("multi_test", "t2", {{"a", TiDB::TP::TypeLong}, {"b", TiDB::TP::TypeLong}, {"c", TiDB::TP::TypeLong}}, {toVec("a", {3, 3, 0}), toVec("b", {4, 2, 0}), toVec("c", {5, 3, 0})}); + + context.addMockTable("multi_test", "t3", {{"a", TiDB::TP::TypeLong}, {"b", TiDB::TP::TypeLong}}, {toVec("a", {1, 2, 0}), toVec("b", {2, 2, 0})}); + + context.addMockTable("multi_test", "t4", {{"a", TiDB::TP::TypeLong}, {"b", TiDB::TP::TypeLong}}, {toVec("a", {3, 2, 0}), toVec("b", {4, 2, 0})}); + + context.addMockTable("join_agg", "t1", {{"a", TiDB::TP::TypeLong}, {"b", TiDB::TP::TypeLong}}, {toVec("a", {1, 1, 3, 4}), toVec("b", {1, 1, 4, 1})}); + + context.addMockTable("join_agg", "t2", {{"a", TiDB::TP::TypeLong}, {"b", TiDB::TP::TypeLong}}, {toVec("a", {1, 4, 2}), toVec("b", {2, 6, 2})}); + + context.addExchangeReceiver("exchange_r_table", + {{"s1", TiDB::TP::TypeString}, {"join_c", TiDB::TP::TypeString}}, + {toNullableVec("s", {"banana", "banana"}), + toNullableVec("join_c", {"apple", "banana"})}); + + context.addExchangeReceiver("exchange_l_table", + {{"s1", TiDB::TP::TypeString}, {"join_c", TiDB::TP::TypeString}}, + {toNullableVec("s", {"banana", "banana"}), + toNullableVec("join_c", {"apple", "banana"})}); + } + + std::tuple multiTestScan() + { + return {context.scan("multi_test", "t1"), context.scan("multi_test", "t2"), context.scan("multi_test", "t3"), context.scan("multi_test", "t4")}; + } + + void executeWithConcurrency(const std::shared_ptr & request, const ColumnsWithTypeAndName & expect_columns) + { + for (size_t i = 1; i <= max_concurrency_level; ++i) + { + ASSERT_COLUMNS_EQ_UR(expect_columns, executeStreams(request, i)); + } + } +}; + +TEST_F(JoinExecutorTestRunner, SimpleInnerJoin) +try +{ + WRAP_FOR_DIS_ENABLE_PLANNER_BEGIN + auto request = context.scan("simple_test", "t1") + .join(context.scan("simple_test", "t2"), {col("a")}, ASTTableJoin::Kind::Inner) + .build(context); + { + executeWithConcurrency(request, {toNullableVec({"1", "1", "1", "1"}), toNullableVec({{}, "3", {}, "3"}), toNullableVec({"1", "1", "1", "1"}), toNullableVec({"3", "3", {}, {}})}); + } + + request = context.scan("simple_test", "t2") + .join(context.scan("simple_test", "t1"), {col("a")}, ASTTableJoin::Kind::Inner) + .build(context); + { + executeWithConcurrency(request, {toNullableVec({"1", "1", "1", "1"}), toNullableVec({{}, "3", {}, "3"}), toNullableVec({"1", "1", "1", "1"}), toNullableVec({"3", "3", {}, {}})}); + } + + request = context.scan("simple_test", "t1") + .join(context.scan("simple_test", "t2"), {col("b")}, ASTTableJoin::Kind::Inner) + .build(context); + { + executeWithConcurrency(request, {toNullableVec({{}, "1", "2", {}, "1"}), toNullableVec({"3", "3", "4", "3", "3"}), toNullableVec({"1", "1", "3", {}, {}}), toNullableVec({"3", "3", "4", "3", "3"})}); + } + + request = context.scan("simple_test", "t2") + .join(context.scan("simple_test", "t1"), {col("b")}, ASTTableJoin::Kind::Inner) + .build(context); + { + executeWithConcurrency(request, {toNullableVec({{}, "1", "3", {}, "1"}), toNullableVec({"3", "3", "4", "3", "3"}), toNullableVec({"1", "1", "2", {}, {}}), toNullableVec({"3", "3", "4", "3", "3"})}); + } + WRAP_FOR_DIS_ENABLE_PLANNER_END +} +CATCH + +TEST_F(JoinExecutorTestRunner, SimpleLeftJoin) +try +{ + WRAP_FOR_DIS_ENABLE_PLANNER_BEGIN + auto request = context.scan("simple_test", "t1") + .join(context.scan("simple_test", "t2"), {col("a")}, ASTTableJoin::Kind::Left) + .build(context); + { + executeWithConcurrency(request, {toNullableVec({"1", "1", "2", {}, "1", "1", {}}), toNullableVec({"3", "3", "4", "3", {}, {}, {}}), toNullableVec({"1", "1", {}, {}, "1", "1", {}}), toNullableVec({{}, "3", {}, {}, {}, "3", {}})}); + } + + request = context.scan("simple_test", "t2") + .join(context.scan("simple_test", "t1"), {col("a")}, ASTTableJoin::Kind::Left) + .build(context); + { + executeWithConcurrency(request, {toNullableVec({"1", "1", "3", {}, "1", "1", {}}), toNullableVec({"3", "3", "4", "3", {}, {}, {}}), toNullableVec({"1", "1", {}, {}, "1", "1", {}}), toNullableVec({{}, "3", {}, {}, {}, "3", {}})}); + } + + request = context.scan("simple_test", "t1") + .join(context.scan("simple_test", "t2"), {col("b")}, ASTTableJoin::Kind::Left) + .build(context); + { + executeWithConcurrency(request, {toNullableVec({"1", "1", "2", {}, {}, "1", {}}), toNullableVec({"3", "3", "4", "3", "3", {}, {}}), toNullableVec({{}, "1", "3", {}, "1", {}, {}}), toNullableVec({"3", "3", "4", "3", "3", {}, {}})}); + } + + request = context.scan("simple_test", "t2") + .join(context.scan("simple_test", "t1"), {col("b")}, ASTTableJoin::Kind::Left) + .build(context); + { + executeWithConcurrency(request, {toNullableVec({"1", "1", "3", {}, {}, "1", {}}), toNullableVec({"3", "3", "4", "3", "3", {}, {}}), toNullableVec({{}, "1", "2", {}, "1", {}, {}}), toNullableVec({"3", "3", "4", "3", "3", {}, {}})}); + } + WRAP_FOR_DIS_ENABLE_PLANNER_END +} +CATCH + +TEST_F(JoinExecutorTestRunner, SimpleRightJoin) +try +{ + WRAP_FOR_DIS_ENABLE_PLANNER_BEGIN + auto request = context + .scan("simple_test", "t1") + .join(context.scan("simple_test", "t2"), {col("a")}, ASTTableJoin::Kind::Right) + .build(context); + { + executeWithConcurrency(request, {toNullableVec({"1", "1", {}, {}, "1", "1", {}}), toNullableVec({{}, "3", {}, {}, {}, "3", {}}), toNullableVec({"1", "1", "3", {}, "1", "1", {}}), toNullableVec({"3", "3", "4", "3", {}, {}, {}})}); + } + + request = context + .scan("simple_test", "t2") + .join(context.scan("simple_test", "t1"), {col("a")}, ASTTableJoin::Kind::Right) + .build(context); + { + executeWithConcurrency(request, {toNullableVec({"1", "1", {}, {}, "1", "1", {}}), toNullableVec({{}, "3", {}, {}, {}, "3", {}}), toNullableVec({"1", "1", "2", {}, "1", "1", {}}), toNullableVec({"3", "3", "4", "3", {}, {}, {}})}); + } + + request = context + .scan("simple_test", "t1") + .join(context.scan("simple_test", "t2"), {col("b")}, ASTTableJoin::Kind::Right) + .build(context); + { + executeWithConcurrency(request, {toNullableVec({{}, "1", "2", {}, "1", {}, {}}), toNullableVec({"3", "3", "4", "3", "3", {}, {}}), toNullableVec({"1", "1", "3", {}, {}, "1", {}}), toNullableVec({"3", "3", "4", "3", "3", {}, {}})}); + } + + request = context + .scan("simple_test", "t2") + .join(context.scan("simple_test", "t1"), {col("b")}, ASTTableJoin::Kind::Right) + .build(context); + { + executeWithConcurrency(request, {toNullableVec({{}, "1", "3", {}, "1", {}, {}}), toNullableVec({"3", "3", "4", "3", "3", {}, {}}), toNullableVec({"1", "1", "2", {}, {}, "1", {}}), toNullableVec({"3", "3", "4", "3", "3", {}, {}})}); + } + WRAP_FOR_DIS_ENABLE_PLANNER_END +} +CATCH + +TEST_F(JoinExecutorTestRunner, MultiInnerLeftJoin) +try +{ + WRAP_FOR_DIS_ENABLE_PLANNER_BEGIN + auto [t1, t2, t3, t4] = multiTestScan(); + auto request = t1.join(t2, {col("a")}, ASTTableJoin::Kind::Inner) + .join(t3.join(t4, {col("a")}, ASTTableJoin::Kind::Inner), + {col("b")}, + ASTTableJoin::Kind::Left) + .build(context); + + executeWithConcurrency(request, {toNullableVec({3, 3, 0}), toNullableVec({2, 2, 0}), toNullableVec({2, 2, 0}), toNullableVec({3, 3, 0}), toNullableVec({4, 2, 0}), toNullableVec({5, 3, 0}), toNullableVec({2, 2, 0}), toNullableVec({2, 2, 0}), toNullableVec({2, 2, 0}), toNullableVec({2, 2, 0})}); + WRAP_FOR_DIS_ENABLE_PLANNER_END +} +CATCH + +TEST_F(JoinExecutorTestRunner, MultiInnerRightJoin) +try +{ + WRAP_FOR_DIS_ENABLE_PLANNER_BEGIN + auto [t1, t2, t3, t4] = multiTestScan(); + auto request = t1.join(t2, {col("a")}, ASTTableJoin::Kind::Inner) + .join(t3.join(t4, {col("a")}, ASTTableJoin::Kind::Inner), + {col("b")}, + ASTTableJoin::Kind::Right) + .build(context); + + executeWithConcurrency(request, {toNullableVec({3, 3, 0}), toNullableVec({2, 2, 0}), toNullableVec({2, 2, 0}), toNullableVec({3, 3, 0}), toNullableVec({4, 2, 0}), toNullableVec({5, 3, 0}), toNullableVec({2, 2, 0}), toNullableVec({2, 2, 0}), toNullableVec({2, 2, 0}), toNullableVec({2, 2, 0})}); + WRAP_FOR_DIS_ENABLE_PLANNER_END +} +CATCH + +TEST_F(JoinExecutorTestRunner, MultiLeftInnerJoin) +try +{ + WRAP_FOR_DIS_ENABLE_PLANNER_BEGIN + auto [t1, t2, t3, t4] = multiTestScan(); + auto request = t1.join(t2, {col("a")}, ASTTableJoin::Kind::Left) + .join(t3.join(t4, {col("a")}, ASTTableJoin::Kind::Left), + {col("b")}, + ASTTableJoin::Kind::Inner) + .build(context); + + executeWithConcurrency(request, {toNullableVec({1, 1, 3, 3, 3, 3, 0}), toNullableVec({2, 2, 2, 2, 2, 2, 0}), toNullableVec({3, 3, 2, 2, 2, 2, 0}), toNullableVec({{}, {}, 3, 3, 3, 3, 0}), toNullableVec({{}, {}, 4, 4, 2, 2, 0}), toNullableVec({{}, {}, 5, 5, 3, 3, 0}), toNullableVec({1, 2, 1, 2, 1, 2, 0}), toNullableVec({2, 2, 2, 2, 2, 2, 0}), toNullableVec({{}, 2, {}, 2, {}, 2, 0}), toNullableVec({{}, 2, {}, 2, {}, 2, 0})}); + WRAP_FOR_DIS_ENABLE_PLANNER_END +} +CATCH + +TEST_F(JoinExecutorTestRunner, MultiLeftRightJoin) +try +{ + WRAP_FOR_DIS_ENABLE_PLANNER_BEGIN + auto [t1, t2, t3, t4] = multiTestScan(); + auto request = t1.join(t2, {col("a")}, ASTTableJoin::Kind::Left) + .join(t3.join(t4, {col("a")}, ASTTableJoin::Kind::Left), + {col("b")}, + ASTTableJoin::Kind::Right) + .build(context); + + executeWithConcurrency(request, {toNullableVec({1, 3, 3, 1, 3, 3, 0}), toNullableVec({2, 2, 2, 2, 2, 2, 0}), toNullableVec({3, 2, 2, 3, 2, 2, 0}), toNullableVec({{}, 3, 3, {}, 3, 3, 0}), toNullableVec({{}, 4, 2, {}, 4, 2, 0}), toNullableVec({{}, 5, 3, {}, 5, 3, 0}), toNullableVec({1, 1, 1, 2, 2, 2, 0}), toNullableVec({2, 2, 2, 2, 2, 2, 0}), toNullableVec({{}, {}, {}, 2, 2, 2, 0}), toNullableVec({{}, {}, {}, 2, 2, 2, 0})}); + WRAP_FOR_DIS_ENABLE_PLANNER_END +} +CATCH + +TEST_F(JoinExecutorTestRunner, MultiRightInnerJoin) +try +{ + WRAP_FOR_DIS_ENABLE_PLANNER_BEGIN + auto [t1, t2, t3, t4] = multiTestScan(); + auto request = t1.join(t2, {col("a")}, ASTTableJoin::Kind::Right) + .join(t3.join(t4, {col("a")}, ASTTableJoin::Kind::Right), + {col("b")}, + ASTTableJoin::Kind::Inner) + .build(context); + + executeWithConcurrency(request, {toNullableVec({3, 3, 0}), toNullableVec({2, 2, 0}), toNullableVec({2, 2, 0}), toNullableVec({3, 3, 0}), toNullableVec({4, 2, 0}), toNullableVec({5, 3, 0}), toNullableVec({2, 2, 0}), toNullableVec({2, 2, 0}), toNullableVec({2, 2, 0}), toNullableVec({2, 2, 0})}); + WRAP_FOR_DIS_ENABLE_PLANNER_END +} +CATCH + +TEST_F(JoinExecutorTestRunner, MultiRightLeftJoin) +try +{ + WRAP_FOR_DIS_ENABLE_PLANNER_BEGIN + auto [t1, t2, t3, t4] = multiTestScan(); + auto request = t1.join(t2, {col("a")}, ASTTableJoin::Kind::Right) + .join(t3.join(t4, {col("a")}, ASTTableJoin::Kind::Right), + {col("b")}, + ASTTableJoin::Kind::Left) + .build(context); + + executeWithConcurrency(request, {toNullableVec({3, 3, 0}), toNullableVec({2, 2, 0}), toNullableVec({2, 2, 0}), toNullableVec({3, 3, 0}), toNullableVec({4, 2, 0}), toNullableVec({5, 3, 0}), toNullableVec({2, 2, 0}), toNullableVec({2, 2, 0}), toNullableVec({2, 2, 0}), toNullableVec({2, 2, 0})}); + WRAP_FOR_DIS_ENABLE_PLANNER_END +} +CATCH + +TEST_F(JoinExecutorTestRunner, JoinCast) +try +{ + WRAP_FOR_DIS_ENABLE_PLANNER_BEGIN + auto cast_request = [&]() { + return context.scan("cast", "t1") + .join(context.scan("cast", "t2"), {col("a")}, ASTTableJoin::Kind::Inner) + .build(context); + }; + + /// int(1) == float(1.0) + context.addMockTable("cast", "t1", {{"a", TiDB::TP::TypeLong}}, {toVec("a", {1})}); + + context.addMockTable("cast", "t2", {{"a", TiDB::TP::TypeFloat}}, {toVec("a", {1.0})}); + + executeWithConcurrency(cast_request(), {toNullableVec({1}), toNullableVec({1.0})}); + + /// int(1) == double(1.0) + context.addMockTable("cast", "t1", {{"a", TiDB::TP::TypeLong}}, {toVec("a", {1})}); + + context.addMockTable("cast", "t2", {{"a", TiDB::TP::TypeDouble}}, {toVec("a", {1.0})}); + + executeWithConcurrency(cast_request(), {toNullableVec({1}), toNullableVec({1.0})}); + + /// float(1) == double(1.0) + context.addMockTable("cast", "t1", {{"a", TiDB::TP::TypeFloat}}, {toVec("a", {1})}); + + context.addMockTable("cast", "t2", {{"a", TiDB::TP::TypeDouble}}, {toVec("a", {1})}); + + executeWithConcurrency(cast_request(), {toNullableVec({1}), toNullableVec({1})}); + + /// varchar('x') == char('x') + context.addMockTable("cast", "t1", {{"a", TiDB::TP::TypeString}}, {toVec("a", {"x"})}); + + context.addMockTable("cast", "t2", {{"a", TiDB::TP::TypeVarchar}}, {toVec("a", {"x"})}); + + executeWithConcurrency(cast_request(), {toNullableVec({"x"}), toNullableVec({"x"})}); + + /// tinyblob('x') == varchar('x') + context.addMockTable("cast", "t1", {{"a", TiDB::TP::TypeTinyBlob}}, {toVec("a", {"x"})}); + + context.addMockTable("cast", "t2", {{"a", TiDB::TP::TypeVarchar}}, {toVec("a", {"x"})}); + + executeWithConcurrency(cast_request(), {toNullableVec({"x"}), toNullableVec({"x"})}); + + /// mediumBlob('x') == varchar('x') + context.addMockTable("cast", "t1", {{"a", TiDB::TP::TypeMediumBlob}}, {toVec("a", {"x"})}); + + context.addMockTable("cast", "t2", {{"a", TiDB::TP::TypeVarchar}}, {toVec("a", {"x"})}); + + executeWithConcurrency(cast_request(), {toNullableVec({"x"}), toNullableVec({"x"})}); + + /// blob('x') == varchar('x') + context.addMockTable("cast", "t1", {{"a", TiDB::TP::TypeBlob}}, {toVec("a", {"x"})}); + + context.addMockTable("cast", "t2", {{"a", TiDB::TP::TypeVarchar}}, {toVec("a", {"x"})}); + + executeWithConcurrency(cast_request(), {toNullableVec({"x"}), toNullableVec({"x"})}); + + /// longBlob('x') == varchar('x') + context.addMockTable("cast", "t1", {{"a", TiDB::TP::TypeLongBlob}}, {toVec("a", {"x"})}); + + context.addMockTable("cast", "t2", {{"a", TiDB::TP::TypeVarchar}}, {toVec("a", {"x"})}); + + executeWithConcurrency(cast_request(), {toNullableVec({"x"}), toNullableVec({"x"})}); + + /// decimal with different scale + context.addMockTable("cast", "t1", {{"a", TiDB::TP::TypeNewDecimal}}, {createColumn(std::make_tuple(9, 4), {"0.12"}, "a")}); + + context.addMockTable("cast", "t2", {{"a", TiDB::TP::TypeNewDecimal}}, {createColumn(std::make_tuple(9, 3), {"0.12"}, "a")}); + + executeWithConcurrency(cast_request(), {createNullableColumn(std::make_tuple(65, 0), {"0.12"}, {0}), createNullableColumn(std::make_tuple(65, 0), {"0.12"}, {0})}); + + /// datetime(1970-01-01 00:00:01) == timestamp(1970-01-01 00:00:01) + context.addMockTable("cast", "t1", {{"a", TiDB::TP::TypeDatetime}}, {createDateTimeColumn({{{1970, 1, 1, 0, 0, 1, 0}}}, 6)}); + + context.addMockTable("cast", "t2", {{"a", TiDB::TP::TypeTimestamp}}, {createDateTimeColumn({{{1970, 1, 1, 0, 0, 1, 0}}}, 6)}); + + executeWithConcurrency(cast_request(), {createDateTimeColumn({{{1970, 1, 1, 0, 0, 1, 0}}}, 0), createDateTimeColumn({{{1970, 1, 1, 0, 0, 1, 0}}}, 0)}); + WRAP_FOR_DIS_ENABLE_PLANNER_END +} +CATCH + +TEST_F(JoinExecutorTestRunner, JoinAgg) +try +{ + WRAP_FOR_DIS_ENABLE_PLANNER_BEGIN + auto request = context.scan("join_agg", "t1") + .join(context.scan("join_agg", "t2"), {col("a")}, ASTTableJoin::Kind::Inner) + .aggregation({Max(col("a")), Min(col("a")), Count(col("a"))}, {col("b")}) + .build(context); + + { + executeWithConcurrency(request, {toNullableVec({4}), toNullableVec({1}), toVec({3}), toNullableVec({1})}); + } + + request = context.scan("join_agg", "t1") + .join(context.scan("join_agg", "t2"), {col("a")}, ASTTableJoin::Kind::Left) + .aggregation({Max(col("a")), Min(col("a")), Count(col("a"))}, {col("b")}) + .build(context); + + { + executeWithConcurrency(request, {toNullableVec({4, 3}), toNullableVec({1, 3}), toVec({3, 1}), toNullableVec({1, 4})}); + } + + request = context.scan("join_agg", "t1") + .join(context.scan("join_agg", "t2"), {col("a")}, ASTTableJoin::Kind::Right) + .aggregation({Max(col("a")), Min(col("a")), Count(col("a"))}, {col("b")}) + .build(context); + + { + executeWithConcurrency(request, {toNullableVec({4, {}}), toNullableVec({1, {}}), toVec({3, 0}), toNullableVec({1, {}})}); + } + WRAP_FOR_DIS_ENABLE_PLANNER_END +} +CATCH + +TEST_F(JoinExecutorTestRunner, JoinWithTableScan) +try +{ + WRAP_FOR_DIS_ENABLE_PLANNER_BEGIN + auto request = context + .scan("test_db", "l_table") + .join(context.scan("test_db", "r_table"), {col("join_c")}, ASTTableJoin::Kind::Left) + .build(context); + { + executeWithConcurrency(request, {toNullableVec({"banana", "banana"}), toNullableVec({"apple", "banana"}), toNullableVec({"banana", "banana"}), toNullableVec({"apple", "banana"})}); + } + + request = context + .scan("test_db", "l_table") + .join(context.scan("test_db", "r_table"), {col("join_c")}, ASTTableJoin::Kind::Left) + .project({"s", "join_c"}) + .build(context); + { + executeWithConcurrency(request, {toNullableVec({"banana", "banana"}), toNullableVec({"apple", "banana"})}); + } + + request = context + .scan("test_db", "l_table") + .join(context.scan("test_db", "r_table_2"), {col("join_c")}, ASTTableJoin::Kind::Left) + .build(context); + { + executeWithConcurrency(request, {toNullableVec({"banana", "banana", "banana", "banana"}), toNullableVec({"apple", "apple", "apple", "banana"}), toNullableVec({"banana", "banana", "banana", {}}), toNullableVec({"apple", "apple", "apple", {}})}); + } + WRAP_FOR_DIS_ENABLE_PLANNER_END +} +CATCH + +TEST_F(JoinExecutorTestRunner, JoinWithExchangeReceiver) +try +{ + WRAP_FOR_DIS_ENABLE_PLANNER_BEGIN + auto request = context + .receive("exchange_l_table") + .join(context.receive("exchange_r_table"), {col("join_c")}, ASTTableJoin::Kind::Left) + .build(context); + { + executeWithConcurrency(request, {toNullableVec({"banana", "banana"}), toNullableVec({"apple", "banana"}), toNullableVec({"banana", "banana"}), toNullableVec({"apple", "banana"})}); + } + WRAP_FOR_DIS_ENABLE_PLANNER_END +} +CATCH + +TEST_F(JoinExecutorTestRunner, JoinWithTableScanAndReceiver) +try +{ + WRAP_FOR_DIS_ENABLE_PLANNER_BEGIN + auto request = context + .scan("test_db", "l_table") + .join(context.receive("exchange_r_table"), {col("join_c")}, ASTTableJoin::Kind::Left) + .build(context); + { + executeWithConcurrency(request, {toNullableVec({"banana", "banana"}), toNullableVec({"apple", "banana"}), toNullableVec({"banana", "banana"}), toNullableVec({"apple", "banana"})}); + } + WRAP_FOR_DIS_ENABLE_PLANNER_END +} +CATCH + +} // namespace tests +} // namespace DB diff --git a/dbms/src/Flash/tests/gtest_limit_executor.cpp b/dbms/src/Flash/tests/gtest_limit_executor.cpp index a3ddf341525..b4363c2db2a 100644 --- a/dbms/src/Flash/tests/gtest_limit_executor.cpp +++ b/dbms/src/Flash/tests/gtest_limit_executor.cpp @@ -50,28 +50,28 @@ class ExecutorLimitTestRunner : public DB::tests::ExecutorTest TEST_F(ExecutorLimitTestRunner, Limit) try { - wrapForDisEnablePlanner([&]() { - std::shared_ptr request; - ColumnsWithTypeAndName expect_cols; + WRAP_FOR_DIS_ENABLE_PLANNER_BEGIN + std::shared_ptr request; + ColumnsWithTypeAndName expect_cols; - /// Check limit result with various parameters - const size_t col_data_num = col0.size(); - for (size_t limit_num = 0; limit_num <= col_data_num + 3; ++limit_num) - { - if (limit_num == col_data_num + 3) - limit_num = INT_MAX; - request = buildDAGRequest(limit_num); + /// Check limit result with various parameters + const size_t col_data_num = col0.size(); + for (size_t limit_num = 0; limit_num <= col_data_num + 3; ++limit_num) + { + if (limit_num == col_data_num + 3) + limit_num = INT_MAX; + request = buildDAGRequest(limit_num); - if (limit_num == 0) - expect_cols = {}; - else if (limit_num > col_data_num) - expect_cols = {toNullableVec(col_name, ColumnWithData(col0.begin(), col0.end()))}; - else - expect_cols = {toNullableVec(col_name, ColumnWithData(col0.begin(), col0.begin() + limit_num))}; + if (limit_num == 0) + expect_cols = {}; + else if (limit_num > col_data_num) + expect_cols = {toNullableVec(col_name, ColumnWithData(col0.begin(), col0.end()))}; + else + expect_cols = {toNullableVec(col_name, ColumnWithData(col0.begin(), col0.begin() + limit_num))}; - ASSERT_COLUMNS_EQ_R(executeStreams(request), expect_cols); - } - }); + ASSERT_COLUMNS_EQ_R(executeStreams(request), expect_cols); + } + WRAP_FOR_DIS_ENABLE_PLANNER_END } CATCH diff --git a/dbms/src/Flash/tests/gtest_projection_executor.cpp b/dbms/src/Flash/tests/gtest_projection_executor.cpp index 65b75bad9ab..9381a1e73dc 100644 --- a/dbms/src/Flash/tests/gtest_projection_executor.cpp +++ b/dbms/src/Flash/tests/gtest_projection_executor.cpp @@ -85,142 +85,142 @@ class ExecutorProjectionTestRunner : public DB::tests::ExecutorTest TEST_F(ExecutorProjectionTestRunner, Projection) try { - wrapForDisEnablePlanner([&]() { - /// Check single column - auto request = buildDAGRequest({col_names[4]}); - executeWithConcurrency(request, {toNullableVec(col_names[4], col4_sorted_asc)}); - - /// Check multi columns - request = buildDAGRequest({col_names[0], col_names[4]}); - executeWithConcurrency(request, - { - toNullableVec(col_names[0], col0_sorted_asc), - toNullableVec(col_names[4], col4_sorted_asc), - }); - - /// Check multi columns - request = buildDAGRequest({col_names[0], col_names[1], col_names[4]}); - executeWithConcurrency(request, - {toNullableVec(col_names[0], col0_sorted_asc), - toNullableVec(col_names[1], col1_sorted_asc), - toNullableVec(col_names[4], col4_sorted_asc)}); - - /// Check duplicate columns - request = buildDAGRequest({col_names[4], col_names[4], col_names[4]}); - executeWithConcurrency(request, - {toNullableVec(col_names[4], col4_sorted_asc), - toNullableVec(col_names[4], col4_sorted_asc), - toNullableVec(col_names[4], col4_sorted_asc)}); + WRAP_FOR_DIS_ENABLE_PLANNER_BEGIN + /// Check single column + auto request = buildDAGRequest({col_names[4]}); + executeWithConcurrency(request, {toNullableVec(col_names[4], col4_sorted_asc)}); + + /// Check multi columns + request = buildDAGRequest({col_names[0], col_names[4]}); + executeWithConcurrency(request, + { + toNullableVec(col_names[0], col0_sorted_asc), + toNullableVec(col_names[4], col4_sorted_asc), + }); + + /// Check multi columns + request = buildDAGRequest({col_names[0], col_names[1], col_names[4]}); + executeWithConcurrency(request, + {toNullableVec(col_names[0], col0_sorted_asc), + toNullableVec(col_names[1], col1_sorted_asc), + toNullableVec(col_names[4], col4_sorted_asc)}); + + /// Check duplicate columns + request = buildDAGRequest({col_names[4], col_names[4], col_names[4]}); + executeWithConcurrency(request, + {toNullableVec(col_names[4], col4_sorted_asc), + toNullableVec(col_names[4], col4_sorted_asc), + toNullableVec(col_names[4], col4_sorted_asc)}); + { + /// Check large number of columns + const size_t col_num = 100; + MockColumnNameVec projection_input; + ColumnsWithTypeAndName columns; + auto expect_column = toNullableVec(col_names[4], col4_sorted_asc); + + for (size_t i = 0; i < col_num; ++i) { - /// Check large number of columns - const size_t col_num = 100; - MockColumnNameVec projection_input; - ColumnsWithTypeAndName columns; - auto expect_column = toNullableVec(col_names[4], col4_sorted_asc); - - for (size_t i = 0; i < col_num; ++i) - { - projection_input.push_back(col_names[4]); - columns.push_back(expect_column); - } - - request = buildDAGRequest(projection_input); - executeWithConcurrency(request, columns); + projection_input.push_back(col_names[4]); + columns.push_back(expect_column); } - }); + + request = buildDAGRequest(projection_input); + executeWithConcurrency(request, columns); + } + WRAP_FOR_DIS_ENABLE_PLANNER_END } CATCH TEST_F(ExecutorProjectionTestRunner, ProjectionFunction) try { - wrapForDisEnablePlanner([&]() { - std::shared_ptr request; - - /// Test "equal" function - - /// Data type: TypeString - request = buildDAGRequest({eq(col(col_names[0]), col(col_names[0])), col(col_names[4])}); - executeWithConcurrency(request, - {toNullableVec({{}, 1, 1, 1, 1, 1, 1}), - toNullableVec(col_names[4], col4_sorted_asc)}); - - request = buildDAGRequest({eq(col(col_names[0]), col(col_names[1])), col(col_names[4])}); - executeWithConcurrency(request, - {toNullableVec({{}, 0, 1, 0, {}, 0, 0}), - toNullableVec(col_names[4], col4_sorted_asc)}); - - /// Data type: TypeLong - request = buildDAGRequest({eq(col(col_names[3]), col(col_names[4])), col(col_names[4])}); - executeWithConcurrency(request, - {toNullableVec({{}, 0, 0, 0, {}, 1, 0}), - toNullableVec(col_names[4], col4_sorted_asc)}); - - - /// Test "greater" function - - /// Data type: TypeString - request = buildDAGRequest({gt(col(col_names[0]), col(col_names[1])), col(col_names[4])}); - executeWithConcurrency(request, - {toNullableVec({{}, 0, 0, 0, {}, 0, 0}), - toNullableVec(col_names[4], col4_sorted_asc)}); - - request = buildDAGRequest({gt(col(col_names[1]), col(col_names[0])), col(col_names[4])}); - executeWithConcurrency(request, - {toNullableVec({{}, 1, 0, 1, {}, 1, 1}), - toNullableVec(col_names[4], col4_sorted_asc)}); - - /// Data type: TypeLong - request = buildDAGRequest({gt(col(col_names[3]), col(col_names[4])), col(col_names[4])}); - executeWithConcurrency(request, - {toNullableVec({{}, 0, 1, 1, {}, 0, 0}), - toNullableVec(col_names[4], col4_sorted_asc)}); - - request = buildDAGRequest({gt(col(col_names[4]), col(col_names[3])), col(col_names[4])}); - executeWithConcurrency(request, - {toNullableVec({{}, 1, 0, 0, {}, 0, 1}), - toNullableVec(col_names[4], col4_sorted_asc)}); - - - /// Test "and" function - - /// Data type: TypeString - request = buildDAGRequest({And(col(col_names[0]), col(col_names[0])), col(col_names[4])}); - executeWithConcurrency(request, - {toNullableVec({{}, 0, 0, 0, 0, 0, 0}), - toNullableVec(col_names[4], col4_sorted_asc)}); - - request = buildDAGRequest({And(col(col_names[0]), col(col_names[1])), col(col_names[4])}); - executeWithConcurrency(request, - {toNullableVec({0, 0, 0, 0, 0, 0, 0}), - toNullableVec(col_names[4], col4_sorted_asc)}); - - /// Data type: TypeLong - request = buildDAGRequest({And(col(col_names[3]), col(col_names[4])), col(col_names[4])}); - executeWithConcurrency(request, - {toNullableVec({{}, 1, 0, 0, {}, 1, 0}), - toNullableVec(col_names[4], col4_sorted_asc)}); - - /// Test "not" function - - /// Data type: TypeString - request = buildDAGRequest({NOT(col(col_names[0])), NOT(col(col_names[1])), NOT(col(col_names[2])), col(col_names[4])}); - executeWithConcurrency(request, - {toNullableVec({{}, 1, 1, 1, 1, 1, 1}), - toNullableVec({1, 1, 1, 1, {}, 1, 1}), - toNullableVec({1, {}, 1, 1, 1, 1, {}}), - toNullableVec(col_names[4], col4_sorted_asc)}); - - /// Data type: TypeLong - request = buildDAGRequest({NOT(col(col_names[3])), NOT(col(col_names[4])), col(col_names[4])}); - executeWithConcurrency(request, - {toNullableVec({{}, 0, 1, 0, {}, 0, 1}), - toNullableVec({{}, 0, 0, 1, 0, 0, 0}), - toNullableVec(col_names[4], col4_sorted_asc)}); - - /// TODO more functions... - }); + WRAP_FOR_DIS_ENABLE_PLANNER_BEGIN + std::shared_ptr request; + + /// Test "equal" function + + /// Data type: TypeString + request = buildDAGRequest({eq(col(col_names[0]), col(col_names[0])), col(col_names[4])}); + executeWithConcurrency(request, + {toNullableVec({{}, 1, 1, 1, 1, 1, 1}), + toNullableVec(col_names[4], col4_sorted_asc)}); + + request = buildDAGRequest({eq(col(col_names[0]), col(col_names[1])), col(col_names[4])}); + executeWithConcurrency(request, + {toNullableVec({{}, 0, 1, 0, {}, 0, 0}), + toNullableVec(col_names[4], col4_sorted_asc)}); + + /// Data type: TypeLong + request = buildDAGRequest({eq(col(col_names[3]), col(col_names[4])), col(col_names[4])}); + executeWithConcurrency(request, + {toNullableVec({{}, 0, 0, 0, {}, 1, 0}), + toNullableVec(col_names[4], col4_sorted_asc)}); + + + /// Test "greater" function + + /// Data type: TypeString + request = buildDAGRequest({gt(col(col_names[0]), col(col_names[1])), col(col_names[4])}); + executeWithConcurrency(request, + {toNullableVec({{}, 0, 0, 0, {}, 0, 0}), + toNullableVec(col_names[4], col4_sorted_asc)}); + + request = buildDAGRequest({gt(col(col_names[1]), col(col_names[0])), col(col_names[4])}); + executeWithConcurrency(request, + {toNullableVec({{}, 1, 0, 1, {}, 1, 1}), + toNullableVec(col_names[4], col4_sorted_asc)}); + + /// Data type: TypeLong + request = buildDAGRequest({gt(col(col_names[3]), col(col_names[4])), col(col_names[4])}); + executeWithConcurrency(request, + {toNullableVec({{}, 0, 1, 1, {}, 0, 0}), + toNullableVec(col_names[4], col4_sorted_asc)}); + + request = buildDAGRequest({gt(col(col_names[4]), col(col_names[3])), col(col_names[4])}); + executeWithConcurrency(request, + {toNullableVec({{}, 1, 0, 0, {}, 0, 1}), + toNullableVec(col_names[4], col4_sorted_asc)}); + + + /// Test "and" function + + /// Data type: TypeString + request = buildDAGRequest({And(col(col_names[0]), col(col_names[0])), col(col_names[4])}); + executeWithConcurrency(request, + {toNullableVec({{}, 0, 0, 0, 0, 0, 0}), + toNullableVec(col_names[4], col4_sorted_asc)}); + + request = buildDAGRequest({And(col(col_names[0]), col(col_names[1])), col(col_names[4])}); + executeWithConcurrency(request, + {toNullableVec({0, 0, 0, 0, 0, 0, 0}), + toNullableVec(col_names[4], col4_sorted_asc)}); + + /// Data type: TypeLong + request = buildDAGRequest({And(col(col_names[3]), col(col_names[4])), col(col_names[4])}); + executeWithConcurrency(request, + {toNullableVec({{}, 1, 0, 0, {}, 1, 0}), + toNullableVec(col_names[4], col4_sorted_asc)}); + + /// Test "not" function + + /// Data type: TypeString + request = buildDAGRequest({NOT(col(col_names[0])), NOT(col(col_names[1])), NOT(col(col_names[2])), col(col_names[4])}); + executeWithConcurrency(request, + {toNullableVec({{}, 1, 1, 1, 1, 1, 1}), + toNullableVec({1, 1, 1, 1, {}, 1, 1}), + toNullableVec({1, {}, 1, 1, 1, 1, {}}), + toNullableVec(col_names[4], col4_sorted_asc)}); + + /// Data type: TypeLong + request = buildDAGRequest({NOT(col(col_names[3])), NOT(col(col_names[4])), col(col_names[4])}); + executeWithConcurrency(request, + {toNullableVec({{}, 0, 1, 0, {}, 0, 1}), + toNullableVec({{}, 0, 0, 1, 0, 0, 0}), + toNullableVec(col_names[4], col4_sorted_asc)}); + + /// TODO more functions... + WRAP_FOR_DIS_ENABLE_PLANNER_END } CATCH diff --git a/dbms/src/Flash/tests/gtest_topn_executor.cpp b/dbms/src/Flash/tests/gtest_topn_executor.cpp index a6ba3183118..d2d138e01aa 100644 --- a/dbms/src/Flash/tests/gtest_topn_executor.cpp +++ b/dbms/src/Flash/tests/gtest_topn_executor.cpp @@ -44,7 +44,7 @@ class ExecutorTopNTestRunner : public DB::tests::ExecutorTest {toNullableVec(col_name[0], col_age), toNullableVec(col_name[1], col_gender), toNullableVec(col_name[2], col_country), - toNullableVec(col_name[3], c0l_salary)}); + toNullableVec(col_name[3], col_salary)}); } std::shared_ptr buildDAGRequest(const String & table_name, const String & col_name, bool is_desc, int limit_num) @@ -72,152 +72,152 @@ class ExecutorTopNTestRunner : public DB::tests::ExecutorTest ColumnWithInt32 col_age{{}, 27, 32, 36, {}, 34}; ColumnWithString col_gender{"female", "female", "male", "female", "male", "male"}; ColumnWithString col_country{"korea", "usa", "usa", "china", "china", "china"}; - ColumnWithInt32 c0l_salary{1300, 0, {}, 900, {}, -300}; + ColumnWithInt32 col_salary{1300, 0, {}, 900, {}, -300}; }; TEST_F(ExecutorTopNTestRunner, TopN) try { - wrapForDisEnablePlanner([&]() { - std::shared_ptr request; - std::vector expect_cols; + WRAP_FOR_DIS_ENABLE_PLANNER_BEGIN + std::shared_ptr request; + std::vector expect_cols; + { + /// Test single column + size_t col_data_num = col0.size(); + for (size_t i = 1; i <= 1; ++i) { - /// Test single column - size_t col_data_num = col0.size(); - for (size_t i = 1; i <= 1; ++i) + bool is_desc; + is_desc = static_cast(i); /// Set descent or ascent + if (is_desc) + sort(col0.begin(), col0.end(), std::greater()); /// Sort col0 for the following comparison + else + sort(col0.begin(), col0.end()); + + for (size_t limit_num = 0; limit_num <= col_data_num + 5; ++limit_num) { - bool is_desc; - is_desc = static_cast(i); /// Set descent or ascent - if (is_desc) - sort(col0.begin(), col0.end(), std::greater()); /// Sort col0 for the following comparison + request = buildDAGRequest(table_single_name, single_col_name, is_desc, limit_num); + + expect_cols.clear(); + if (limit_num == 0 || limit_num > col_data_num) + expect_cols.push_back({toNullableVec(single_col_name, ColumnWithString(col0.begin(), col0.end()))}); else - sort(col0.begin(), col0.end()); - - for (size_t limit_num = 0; limit_num <= col_data_num + 5; ++limit_num) - { - request = buildDAGRequest(table_single_name, single_col_name, is_desc, limit_num); - - expect_cols.clear(); - if (limit_num == 0 || limit_num > col_data_num) - expect_cols.push_back({toNullableVec(single_col_name, ColumnWithString(col0.begin(), col0.end()))}); - else - expect_cols.push_back({toNullableVec(single_col_name, ColumnWithString(col0.begin(), col0.begin() + limit_num))}); - - ASSERT_COLUMNS_EQ_R(executeStreams(request), expect_cols[0]); - ASSERT_COLUMNS_EQ_R(executeStreams(request, 2), expect_cols[0]); - ASSERT_COLUMNS_EQ_R(executeStreams(request, 4), expect_cols[0]); - ASSERT_COLUMNS_EQ_R(executeStreams(request, 8), expect_cols[0]); - } + expect_cols.push_back({toNullableVec(single_col_name, ColumnWithString(col0.begin(), col0.begin() + limit_num))}); + + ASSERT_COLUMNS_EQ_R(executeStreams(request), expect_cols[0]); + ASSERT_COLUMNS_EQ_R(executeStreams(request, 2), expect_cols[0]); + ASSERT_COLUMNS_EQ_R(executeStreams(request, 4), expect_cols[0]); + ASSERT_COLUMNS_EQ_R(executeStreams(request, 8), expect_cols[0]); } } + } + { + /// Test multi-columns + expect_cols = {{toNullableVec(col_name[0], ColumnWithInt32{36, 34, 32, 27, {}, {}}), + toNullableVec(col_name[1], ColumnWithString{"female", "male", "male", "female", "male", "female"}), + toNullableVec(col_name[2], ColumnWithString{"china", "china", "usa", "usa", "china", "korea"}), + toNullableVec(col_name[3], ColumnWithInt32{900, -300, {}, 0, {}, 1300})}, + {toNullableVec(col_name[0], ColumnWithInt32{32, {}, 34, 27, 36, {}}), + toNullableVec(col_name[1], ColumnWithString{"male", "male", "male", "female", "female", "female"}), + toNullableVec(col_name[2], ColumnWithString{"usa", "china", "china", "usa", "china", "korea"}), + toNullableVec(col_name[3], ColumnWithInt32{{}, {}, -300, 0, 900, 1300})}, + {toNullableVec(col_name[0], ColumnWithInt32{34, {}, 32, 36, {}, 27}), + toNullableVec(col_name[1], ColumnWithString{"male", "male", "male", "female", "female", "female"}), + toNullableVec(col_name[2], ColumnWithString{"china", "china", "usa", "china", "korea", "usa"}), + toNullableVec(col_name[3], ColumnWithInt32{-300, {}, {}, 900, 1300, 0})}}; + + std::vector order_by_items{ + /// select * from clerk order by age DESC, gender DESC; + {MockOrderByItem(col_name[0], true), MockOrderByItem(col_name[1], true)}, + /// select * from clerk order by gender DESC, salary ASC; + {MockOrderByItem(col_name[1], true), MockOrderByItem(col_name[3], false)}, + /// select * from clerk order by gender DESC, country ASC, salary DESC; + {MockOrderByItem(col_name[1], true), MockOrderByItem(col_name[2], false), MockOrderByItem(col_name[3], true)}}; + + size_t test_num = expect_cols.size(); + + for (size_t i = 0; i < test_num; ++i) { - /// Test multi-columns - expect_cols = {{toNullableVec(col_name[0], ColumnWithInt32{36, 34, 32, 27, {}, {}}), - toNullableVec(col_name[1], ColumnWithString{"female", "male", "male", "female", "male", "female"}), - toNullableVec(col_name[2], ColumnWithString{"china", "china", "usa", "usa", "china", "korea"}), - toNullableVec(col_name[3], ColumnWithInt32{900, -300, {}, 0, {}, 1300})}, - {toNullableVec(col_name[0], ColumnWithInt32{32, {}, 34, 27, 36, {}}), - toNullableVec(col_name[1], ColumnWithString{"male", "male", "male", "female", "female", "female"}), - toNullableVec(col_name[2], ColumnWithString{"usa", "china", "china", "usa", "china", "korea"}), - toNullableVec(col_name[3], ColumnWithInt32{{}, {}, -300, 0, 900, 1300})}, - {toNullableVec(col_name[0], ColumnWithInt32{34, {}, 32, 36, {}, 27}), - toNullableVec(col_name[1], ColumnWithString{"male", "male", "male", "female", "female", "female"}), - toNullableVec(col_name[2], ColumnWithString{"china", "china", "usa", "china", "korea", "usa"}), - toNullableVec(col_name[3], ColumnWithInt32{-300, {}, {}, 900, 1300, 0})}}; - - std::vector order_by_items{ - /// select * from clerk order by age DESC, gender DESC; - {MockOrderByItem(col_name[0], true), MockOrderByItem(col_name[1], true)}, - /// select * from clerk order by gender DESC, salary ASC; - {MockOrderByItem(col_name[1], true), MockOrderByItem(col_name[3], false)}, - /// select * from clerk order by gender DESC, country ASC, salary DESC; - {MockOrderByItem(col_name[1], true), MockOrderByItem(col_name[2], false), MockOrderByItem(col_name[3], true)}}; - - size_t test_num = expect_cols.size(); - - for (size_t i = 0; i < test_num; ++i) - { - request = buildDAGRequest(table_name, order_by_items[i], 100); - ASSERT_COLUMNS_EQ_R(executeStreams(request), expect_cols[i]); - } + request = buildDAGRequest(table_name, order_by_items[i], 100); + ASSERT_COLUMNS_EQ_R(executeStreams(request), expect_cols[i]); } - }); + } + WRAP_FOR_DIS_ENABLE_PLANNER_END } CATCH TEST_F(ExecutorTopNTestRunner, TopNFunction) try { - wrapForDisEnablePlanner([&]() { - std::shared_ptr request; - std::vector expect_cols; - MockColumnNameVec output_projection{col_name[0], col_name[1], col_name[2], col_name[3]}; - MockAstVec func_projection; // Do function operation for topn - MockOrderByItemVec order_by_items; - ASTPtr col0_ast = col(col_name[0]); - ASTPtr col1_ast = col(col_name[1]); - ASTPtr col2_ast = col(col_name[2]); - ASTPtr col3_ast = col(col_name[3]); - ASTPtr func_ast; + WRAP_FOR_DIS_ENABLE_PLANNER_BEGIN + std::shared_ptr request; + std::vector expect_cols; + MockColumnNameVec output_projection{col_name[0], col_name[1], col_name[2], col_name[3]}; + MockAstVec func_projection; // Do function operation for topn + MockOrderByItemVec order_by_items; + ASTPtr col0_ast = col(col_name[0]); + ASTPtr col1_ast = col(col_name[1]); + ASTPtr col2_ast = col(col_name[2]); + ASTPtr col3_ast = col(col_name[3]); + ASTPtr func_ast; - { - /// "and" function - expect_cols = {{toNullableVec(col_name[0], ColumnWithInt32{{}, {}, 32, 27, 36, 34}), - toNullableVec(col_name[1], ColumnWithString{"female", "male", "male", "female", "female", "male"}), - toNullableVec(col_name[2], ColumnWithString{"korea", "china", "usa", "usa", "china", "china"}), - toNullableVec(col_name[3], ColumnWithInt32{1300, {}, {}, 0, 900, -300})}}; + { + /// "and" function + expect_cols = {{toNullableVec(col_name[0], ColumnWithInt32{{}, {}, 32, 27, 36, 34}), + toNullableVec(col_name[1], ColumnWithString{"female", "male", "male", "female", "female", "male"}), + toNullableVec(col_name[2], ColumnWithString{"korea", "china", "usa", "usa", "china", "china"}), + toNullableVec(col_name[3], ColumnWithInt32{1300, {}, {}, 0, 900, -300})}}; - { - /// select * from clerk order by age and salary ASC limit 100; - order_by_items = {MockOrderByItem("and(age, salary)", false)}; - func_ast = And(col(col_name[0]), col(col_name[3])); - func_projection = {col0_ast, col1_ast, col2_ast, col3_ast, func_ast}; + { + /// select * from clerk order by age and salary ASC limit 100; + order_by_items = {MockOrderByItem("and(age, salary)", false)}; + func_ast = And(col(col_name[0]), col(col_name[3])); + func_projection = {col0_ast, col1_ast, col2_ast, col3_ast, func_ast}; - request = buildDAGRequest(table_name, order_by_items, 100, func_projection, output_projection); - ASSERT_COLUMNS_EQ_R(executeStreams(request), expect_cols[0]); - } + request = buildDAGRequest(table_name, order_by_items, 100, func_projection, output_projection); + ASSERT_COLUMNS_EQ_R(executeStreams(request), expect_cols[0]); } + } - { - /// "equal" function - expect_cols = {{toNullableVec(col_name[0], ColumnWithInt32{27, 36, 34, 32, {}, {}}), - toNullableVec(col_name[1], ColumnWithString{"female", "female", "male", "male", "female", "male"}), - toNullableVec(col_name[2], ColumnWithString{"usa", "china", "china", "usa", "korea", "china"}), - toNullableVec(col_name[3], ColumnWithInt32{0, 900, -300, {}, 1300, {}})}}; + { + /// "equal" function + expect_cols = {{toNullableVec(col_name[0], ColumnWithInt32{27, 36, 34, 32, {}, {}}), + toNullableVec(col_name[1], ColumnWithString{"female", "female", "male", "male", "female", "male"}), + toNullableVec(col_name[2], ColumnWithString{"usa", "china", "china", "usa", "korea", "china"}), + toNullableVec(col_name[3], ColumnWithInt32{0, 900, -300, {}, 1300, {}})}}; - { - /// select age, salary from clerk order by age = salary DESC limit 100; - order_by_items = {MockOrderByItem("equals(age, salary)", true)}; - func_ast = eq(col(col_name[0]), col(col_name[3])); - func_projection = {col0_ast, col1_ast, col2_ast, col3_ast, func_ast}; + { + /// select age, salary from clerk order by age = salary DESC limit 100; + order_by_items = {MockOrderByItem("equals(age, salary)", true)}; + func_ast = eq(col(col_name[0]), col(col_name[3])); + func_projection = {col0_ast, col1_ast, col2_ast, col3_ast, func_ast}; - request = buildDAGRequest(table_name, order_by_items, 100, func_projection, output_projection); - ASSERT_COLUMNS_EQ_R(executeStreams(request), expect_cols[0]); - } + request = buildDAGRequest(table_name, order_by_items, 100, func_projection, output_projection); + ASSERT_COLUMNS_EQ_R(executeStreams(request), expect_cols[0]); } + } - { - /// "greater" function - expect_cols = {{toNullableVec(col_name[0], ColumnWithInt32{{}, 32, {}, 36, 27, 34}), - toNullableVec(col_name[1], ColumnWithString{"female", "male", "male", "female", "female", "male"}), - toNullableVec(col_name[2], ColumnWithString{"korea", "usa", "china", "china", "usa", "china"}), - toNullableVec(col_name[3], ColumnWithInt32{1300, {}, {}, 900, 0, -300})}}; + { + /// "greater" function + expect_cols = {{toNullableVec(col_name[0], ColumnWithInt32{{}, 32, {}, 36, 27, 34}), + toNullableVec(col_name[1], ColumnWithString{"female", "male", "male", "female", "female", "male"}), + toNullableVec(col_name[2], ColumnWithString{"korea", "usa", "china", "china", "usa", "china"}), + toNullableVec(col_name[3], ColumnWithInt32{1300, {}, {}, 900, 0, -300})}}; - { - /// select age, gender, country, salary from clerk order by age > salary ASC limit 100; - order_by_items = {MockOrderByItem("greater(age, salary)", false)}; - func_ast = gt(col(col_name[0]), col(col_name[3])); - func_projection = {col0_ast, col1_ast, col2_ast, col3_ast, func_ast}; + { + /// select age, gender, country, salary from clerk order by age > salary ASC limit 100; + order_by_items = {MockOrderByItem("greater(age, salary)", false)}; + func_ast = gt(col(col_name[0]), col(col_name[3])); + func_projection = {col0_ast, col1_ast, col2_ast, col3_ast, func_ast}; - request = buildDAGRequest(table_name, order_by_items, 100, func_projection, output_projection); - ASSERT_COLUMNS_EQ_R(executeStreams(request), expect_cols[0]); - } + request = buildDAGRequest(table_name, order_by_items, 100, func_projection, output_projection); + ASSERT_COLUMNS_EQ_R(executeStreams(request), expect_cols[0]); } + } - /// TODO more functions... - }); + /// TODO more functions... + WRAP_FOR_DIS_ENABLE_PLANNER_END } CATCH diff --git a/dbms/src/Functions/CollationOperatorOptimized.h b/dbms/src/Functions/CollationOperatorOptimized.h index 395ecc5b9eb..8276a41fa17 100644 --- a/dbms/src/Functions/CollationOperatorOptimized.h +++ b/dbms/src/Functions/CollationOperatorOptimized.h @@ -17,6 +17,7 @@ #include #include #include +#include #include #include @@ -27,42 +28,6 @@ namespace DB { -template -ALWAYS_INLINE inline int signum(T val) -{ - return (0 < val) - (val < 0); -} - -// Check equality is much faster than other comparison. -// - check size first -// - return 0 if equal else 1 -__attribute__((flatten, always_inline, pure)) inline uint8_t RawStrEqualCompare(const std::string_view & lhs, const std::string_view & rhs) -{ - return StringRef(lhs) == StringRef(rhs) ? 0 : 1; -} - -// Compare str view by memcmp -__attribute__((flatten, always_inline, pure)) inline int RawStrCompare(const std::string_view & v1, const std::string_view & v2) -{ - return signum(v1.compare(v2)); -} - -constexpr char SPACE = ' '; - -// Remove tail space -__attribute__((flatten, always_inline, pure)) inline std::string_view RightTrim(const std::string_view & v) -{ - if (likely(v.empty() || v.back() != SPACE)) - return v; - size_t end = v.find_last_not_of(SPACE); - return end == std::string_view::npos ? std::string_view{} : std::string_view(v.data(), end + 1); -} - -__attribute__((flatten, always_inline, pure)) inline int RtrimStrCompare(const std::string_view & va, const std::string_view & vb) -{ - return RawStrCompare(RightTrim(va), RightTrim(vb)); -} - // If true, only need to check equal or not. template struct IsEqualRelated @@ -83,6 +48,7 @@ struct IsEqualRelated> }; // Loop columns and invoke callback for each pair. +// Remove last zero byte. template __attribute__((flatten, always_inline)) inline void LoopTwoColumns( const ColumnString::Chars_t & a_data, @@ -92,18 +58,29 @@ __attribute__((flatten, always_inline)) inline void LoopTwoColumns( size_t size, F && func) { + ColumnString::Offset a_prev_offset = 0; + ColumnString::Offset b_prev_offset = 0; + const auto * a_ptr = reinterpret_cast(a_data.data()); + const auto * b_ptr = reinterpret_cast(b_data.data()); + for (size_t i = 0; i < size; ++i) { - size_t a_size = StringUtil::sizeAt(a_offsets, i) - 1; - size_t b_size = StringUtil::sizeAt(b_offsets, i) - 1; - const auto * a_ptr = reinterpret_cast(&a_data[StringUtil::offsetAt(a_offsets, i)]); - const auto * b_ptr = reinterpret_cast(&b_data[StringUtil::offsetAt(b_offsets, i)]); + auto a_size = a_offsets[i] - a_prev_offset; + auto b_size = b_offsets[i] - b_prev_offset; + + // Remove last zero byte. + func({a_ptr, a_size - 1}, {b_ptr, b_size - 1}, i); + + a_ptr += a_size; + b_ptr += b_size; - func({a_ptr, a_size}, {b_ptr, b_size}, i); + a_prev_offset = a_offsets[i]; + b_prev_offset = b_offsets[i]; } } // Loop one column and invoke callback for each pair. +// Remove last zero byte. template __attribute__((flatten, always_inline)) inline void LoopOneColumn( const ColumnString::Chars_t & a_data, @@ -111,12 +88,18 @@ __attribute__((flatten, always_inline)) inline void LoopOneColumn( size_t size, F && func) { + ColumnString::Offset a_prev_offset = 0; + const auto * a_ptr = reinterpret_cast(a_data.data()); + for (size_t i = 0; i < size; ++i) { - size_t a_size = StringUtil::sizeAt(a_offsets, i) - 1; - const auto * a_ptr = reinterpret_cast(&a_data[StringUtil::offsetAt(a_offsets, i)]); + auto a_size = a_offsets[i] - a_prev_offset; + + // Remove last zero byte. + func({a_ptr, a_size - 1}, i); - func({a_ptr, a_size}, i); + a_ptr += a_size; + a_prev_offset = a_offsets[i]; } } diff --git a/dbms/src/Interpreters/AggregationCommon.h b/dbms/src/Interpreters/AggregationCommon.h index 9f390133088..a043760b9af 100644 --- a/dbms/src/Interpreters/AggregationCommon.h +++ b/dbms/src/Interpreters/AggregationCommon.h @@ -171,7 +171,7 @@ static inline T ALWAYS_INLINE packFixed( return key; } - +/* /// Hash a set of keys into a UInt128 value. static inline UInt128 ALWAYS_INLINE hash128( size_t i, @@ -202,7 +202,7 @@ static inline UInt128 ALWAYS_INLINE hash128( return key; } - +*/ /// Almost the same as above but it doesn't return any reference to key data. static inline UInt128 ALWAYS_INLINE hash128( @@ -212,7 +212,7 @@ static inline UInt128 ALWAYS_INLINE hash128( const TiDB::TiDBCollators & collators, std::vector & sort_key_containers) { - UInt128 key; + UInt128 key{}; SipHash hash; if (collators.empty()) @@ -252,7 +252,7 @@ static inline StringRef * ALWAYS_INLINE placeKeysInPool( return reinterpret_cast(res); } - +/* /// Copy keys to the pool. Then put into pool StringRefs to them and return the pointer to the first. static inline StringRef * ALWAYS_INLINE extractKeysAndPlaceInPool( size_t i, @@ -326,7 +326,7 @@ inline StringRef ALWAYS_INLINE extractKeysAndPlaceInPoolContiguous( return {res, sum_keys_size}; } - +*/ /** Serialize keys into a continuous chunk of memory. */ diff --git a/dbms/src/Interpreters/AsynchronousMetrics.cpp b/dbms/src/Interpreters/AsynchronousMetrics.cpp index 8095fbb0e59..3f6ac940ab4 100644 --- a/dbms/src/Interpreters/AsynchronousMetrics.cpp +++ b/dbms/src/Interpreters/AsynchronousMetrics.cpp @@ -31,16 +31,6 @@ #include -#if USE_TCMALLOC -#include - -/// Initializing malloc extension in global constructor as required. -struct MallocExtensionInitializer -{ - MallocExtensionInitializer() { MallocExtension::Initialize(); } -} malloc_extension_initializer; -#endif - #if USE_JEMALLOC #include #endif @@ -209,32 +199,6 @@ void AsynchronousMetrics::update() set("BlobValidBytes", usage.total_valid_size); } -#if USE_TCMALLOC - { - /// tcmalloc related metrics. Remove if you switch to different allocator. - - MallocExtension & malloc_extension = *MallocExtension::instance(); - - auto malloc_metrics = { - "generic.current_allocated_bytes", - "generic.heap_size", - "tcmalloc.current_total_thread_cache_bytes", - "tcmalloc.central_cache_free_bytes", - "tcmalloc.transfer_cache_free_bytes", - "tcmalloc.thread_cache_free_bytes", - "tcmalloc.pageheap_free_bytes", - "tcmalloc.pageheap_unmapped_bytes", - }; - - for (auto malloc_metric : malloc_metrics) - { - size_t value = 0; - if (malloc_extension.GetNumericProperty(malloc_metric, &value)) - set(malloc_metric, value); - } - } -#endif - #if USE_MIMALLOC #define MI_STATS_SET(X) set("mimalloc." #X, X) diff --git a/dbms/src/Interpreters/Join.cpp b/dbms/src/Interpreters/Join.cpp index 181ebcaaa64..9c9c9eb92ab 100644 --- a/dbms/src/Interpreters/Join.cpp +++ b/dbms/src/Interpreters/Join.cpp @@ -788,6 +788,7 @@ bool Join::insertFromBlock(const Block & block) std::unique_lock lock(rwlock); if (unlikely(!initialized)) throw Exception("Logical error: Join was not initialized", ErrorCodes::LOGICAL_ERROR); + total_input_build_rows += block.rows(); blocks.push_back(block); Block * stored_block = &blocks.back(); return insertFromBlockInternal(stored_block, 0); @@ -805,6 +806,7 @@ void Join::insertFromBlock(const Block & block, size_t stream_index) Block * stored_block = nullptr; { std::lock_guard lk(blocks_lock); + total_input_build_rows += block.rows(); blocks.push_back(block); stored_block = &blocks.back(); original_blocks.push_back(block); diff --git a/dbms/src/Interpreters/Join.h b/dbms/src/Interpreters/Join.h index 01916aa1dcc..f30479b6f46 100644 --- a/dbms/src/Interpreters/Join.h +++ b/dbms/src/Interpreters/Join.h @@ -141,6 +141,8 @@ class Join /// Sum size in bytes of all buffers, used for JOIN maps and for all memory pools. size_t getTotalByteCount() const; + size_t getTotalBuildInputRows() const { return total_input_build_rows; } + ASTTableJoin::Kind getKind() const { return kind; } bool useNulls() const { return use_nulls; } @@ -337,6 +339,7 @@ class Join SizeLimits limits; Block totals; + std::atomic total_input_build_rows{0}; /** Protect state for concurrent use in insertFromBlock and joinBlock. * Note that these methods could be called simultaneously only while use of StorageJoin, * and StorageJoin only calls these two methods. diff --git a/dbms/src/Interpreters/SetVariants.h b/dbms/src/Interpreters/SetVariants.h index 08ebb5d5100..58f102555b6 100644 --- a/dbms/src/Interpreters/SetVariants.h +++ b/dbms/src/Interpreters/SetVariants.h @@ -21,7 +21,7 @@ #include #include #include -#include + namespace DB { diff --git a/dbms/src/Interpreters/Settings.h b/dbms/src/Interpreters/Settings.h index 663a2957511..ecee2ab0671 100644 --- a/dbms/src/Interpreters/Settings.h +++ b/dbms/src/Interpreters/Settings.h @@ -317,6 +317,9 @@ struct Settings \ M(SettingDouble, dt_storage_blob_heavy_gc_valid_rate, 0.2, "Max valid rate of deciding a blob can be compact") \ M(SettingDouble, dt_storage_blob_block_alignment_bytes, 0, "Blob IO alignment size") \ + M(SettingBool, dt_enable_read_thread, false, "Enable storage read thread or not") \ + M(SettingDouble, dt_block_slots_scale, 1.0, "Block slots limit of a read request") \ + M(SettingDouble, dt_active_segments_scale, 1.0, "Acitve segments limit of a read request") \ \ M(SettingChecksumAlgorithm, dt_checksum_algorithm, ChecksumAlgo::XXH3, "Checksum algorithm for delta tree stable storage") \ M(SettingCompressionMethod, dt_compression_method, CompressionMethod::LZ4, "The method of data compression when writing.") \ diff --git a/dbms/src/Server/Server.cpp b/dbms/src/Server/Server.cpp index aabca11cf9c..fa34400d5e3 100644 --- a/dbms/src/Server/Server.cpp +++ b/dbms/src/Server/Server.cpp @@ -64,6 +64,9 @@ #include #include #include +#include +#include +#include #include #include #include @@ -1335,6 +1338,12 @@ int Server::main(const std::vector & /*args*/) global_context->getTMTContext().reloadConfig(config()); } + // Initialize the thread pool of storage before the storage engine is initialized. + LOG_FMT_INFO(log, "dt_enable_read_thread {}", global_context->getSettingsRef().dt_enable_read_thread); + DM::SegmentReaderPoolManager::instance().init(server_info); + DM::SegmentReadTaskScheduler::instance(); + DM::DMFileReaderPool::instance(); + { // Note that this must do before initialize schema sync service. do diff --git a/dbms/src/Server/main.cpp b/dbms/src/Server/main.cpp index dbcaa4f38fc..83629eda1ae 100644 --- a/dbms/src/Server/main.cpp +++ b/dbms/src/Server/main.cpp @@ -25,10 +25,6 @@ #include /// pair #include -#if USE_TCMALLOC -#include -#endif - #if ENABLE_CLICKHOUSE_SERVER #include "Server.h" #endif @@ -154,14 +150,6 @@ bool isClickhouseApp(const std::string & app_suffix, std::vector & argv) int main(int argc_, char ** argv_) { -#if USE_TCMALLOC - /** Without this option, tcmalloc returns memory to OS too frequently for medium-sized memory allocations - * (like IO buffers, column vectors, hash tables, etc.), - * that lead to page faults and significantly hurts performance. - */ - MallocExtension::instance()->SetNumericProperty("tcmalloc.aggressive_memory_decommit", false); -#endif - std::vector argv(argv_, argv_ + argc_); /// Print a basic help if nothing was matched diff --git a/dbms/src/Storages/DeltaMerge/ColumnFile/ColumnFileBig.cpp b/dbms/src/Storages/DeltaMerge/ColumnFile/ColumnFileBig.cpp index 3328f60b15d..da4372670be 100644 --- a/dbms/src/Storages/DeltaMerge/ColumnFile/ColumnFileBig.cpp +++ b/dbms/src/Storages/DeltaMerge/ColumnFile/ColumnFileBig.cpp @@ -91,6 +91,7 @@ void ColumnFileBigReader::initStream() .setTracingID(context.tracing_id) .build(column_file.getFile(), *col_defs, RowKeyRanges{column_file.segment_range}); + header = file_stream->getHeader(); // If we only need to read pk and version columns, then cache columns data in memory. if (pk_ver_only) { @@ -223,7 +224,20 @@ Block ColumnFileBigReader::readNextBlock() { initStream(); - return file_stream->read(); + if (pk_ver_only) + { + if (next_block_index_in_cache >= cached_pk_ver_columns.size()) + { + return {}; + } + auto & columns = cached_pk_ver_columns[next_block_index_in_cache]; + next_block_index_in_cache += 1; + return header.cloneWithColumns(std::move(columns)); + } + else + { + return file_stream->read(); + } } ColumnFileReaderPtr ColumnFileBigReader::createNewReader(const ColumnDefinesPtr & new_col_defs) diff --git a/dbms/src/Storages/DeltaMerge/ColumnFile/ColumnFileBig.h b/dbms/src/Storages/DeltaMerge/ColumnFile/ColumnFileBig.h index dba7eca7247..44302399960 100644 --- a/dbms/src/Storages/DeltaMerge/ColumnFile/ColumnFileBig.h +++ b/dbms/src/Storages/DeltaMerge/ColumnFile/ColumnFileBig.h @@ -104,6 +104,8 @@ class ColumnFileBigReader : public ColumnFileReader const ColumnFileBig & column_file; const ColumnDefinesPtr col_defs; + Block header; + bool pk_ver_only; DMFileBlockInputStreamPtr file_stream; @@ -112,6 +114,7 @@ class ColumnFileBigReader : public ColumnFileReader // we cache them to minimize the cost. std::vector cached_pk_ver_columns; std::vector cached_block_rows_end; + size_t next_block_index_in_cache = 0; // The data members for reading all columns, but can only read once. size_t rows_before_cur_block = 0; @@ -131,7 +134,24 @@ class ColumnFileBigReader : public ColumnFileReader , column_file(column_file_) , col_defs(col_defs_) { - pk_ver_only = col_defs->size() <= 2; + if (col_defs_->size() == 1) + { + if ((*col_defs)[0].id == EXTRA_HANDLE_COLUMN_ID) + { + pk_ver_only = true; + } + } + else if (col_defs_->size() == 2) + { + if ((*col_defs)[0].id == EXTRA_HANDLE_COLUMN_ID && (*col_defs)[1].id == VERSION_COLUMN_ID) + { + pk_ver_only = true; + } + } + else + { + pk_ver_only = false; + } } size_t readRows(MutableColumns & output_cols, size_t rows_offset, size_t rows_limit, const RowKeyRange * range) override; diff --git a/dbms/src/Storages/DeltaMerge/DeltaMergeStore.cpp b/dbms/src/Storages/DeltaMerge/DeltaMergeStore.cpp index d34856ec32f..27062c2c108 100644 --- a/dbms/src/Storages/DeltaMerge/DeltaMergeStore.cpp +++ b/dbms/src/Storages/DeltaMerge/DeltaMergeStore.cpp @@ -26,6 +26,8 @@ #include #include #include +#include +#include #include #include #include @@ -39,10 +41,6 @@ #include #include -#if USE_TCMALLOC -#include -#endif - namespace ProfileEvents { extern const Event DMWriteBlock; @@ -469,11 +467,6 @@ void DeltaMergeStore::drop() // Drop data in storage path pool path_pool.drop(/*recursive=*/true, /*must_success=*/false); LOG_FMT_INFO(log, "Drop DeltaMerge done [{}.{}]", db_name, table_name); - -#if USE_TCMALLOC - // Reclaim memory. - MallocExtension::instance()->ReleaseFreeMemory(); -#endif } void DeltaMergeStore::shutdown() @@ -982,6 +975,8 @@ void DeltaMergeStore::deleteRange(const Context & db_context, const DB::Settings bool DeltaMergeStore::flushCache(const DMContextPtr & dm_context, const RowKeyRange & range, bool try_until_succeed) { + size_t sleep_ms = 5; + RowKeyRange cur_range = range; while (!cur_range.none()) { @@ -1005,7 +1000,6 @@ bool DeltaMergeStore::flushCache(const DMContextPtr & dm_context, const RowKeyRa } segment_range = segment->getRowKeyRange(); - // Flush could fail. if (segment->flushCache(*dm_context)) { break; @@ -1014,6 +1008,13 @@ bool DeltaMergeStore::flushCache(const DMContextPtr & dm_context, const RowKeyRa { return false; } + + // Flush could fail. Typical cases: + // #1. The segment is abandoned (due to an update is finished) + // #2. There is another flush in progress, for example, triggered in background + // Let's sleep 5ms ~ 100ms and then retry flush again. + std::this_thread::sleep_for(std::chrono::milliseconds(sleep_ms)); + sleep_ms = std::min(sleep_ms * 2, 100); } cur_range.setStart(segment_range.end); @@ -1047,6 +1048,8 @@ std::optional DeltaMergeStore::mergeDeltaBySegment(const Contex auto dm_context = newDMContext(context, context.getSettingsRef(), /*tracing_id*/ fmt::format("mergeDeltaBySegment_{}", latest_gc_safe_point.load(std::memory_order_relaxed))); + size_t sleep_ms = 50; + while (true) { SegmentPtr segment; @@ -1060,25 +1063,33 @@ std::optional DeltaMergeStore::mergeDeltaBySegment(const Contex segment = segment_it->second; } - if (!segment->flushCache(*dm_context)) - { - // If the flush failed, it means there are parallel updates to the segment in the background. - // In this case, we try again. - continue; - } - - const auto new_segment = segmentMergeDelta(*dm_context, segment, run_thread); - if (new_segment) + if (segment->flushCache(*dm_context)) { - const auto segment_end = new_segment->getRowKeyRange().end; - if (unlikely(*segment_end.value <= *start_key.value)) + const auto new_segment = segmentMergeDelta(*dm_context, segment, run_thread); + if (new_segment) { - // The next start key must be > current start key - LOG_FMT_ERROR(log, "Assert new_segment.end {} > start {} failed", segment_end.toDebugString(), start_key.toDebugString()); - throw Exception("Assert segment range failed", ErrorCodes::LOGICAL_ERROR); - } - return new_segment->getRowKeyRange(); - } + const auto segment_end = new_segment->getRowKeyRange().end; + if (unlikely(*segment_end.value <= *start_key.value)) + { + // The next start key must be > current start key + LOG_FMT_ERROR(log, "Assert new_segment.end {} > start {} failed", segment_end.toDebugString(), start_key.toDebugString()); + throw Exception("Assert segment range failed", ErrorCodes::LOGICAL_ERROR); + } + return new_segment->getRowKeyRange(); + } // else: sleep and retry + } // else: sleep and retry + + // Typical cases: + // #1. flushCache failed + // - The segment is abandoned (due to segment updated) + // - There is another flush in progress (e.g. triggered in background) + // #2. segmentMergeDelta failed + // - The segment is abandoned (due to segment updated) + // - The segment is updating (e.g. a split-preparation is working, which occupies a for-write snapshot). + // It could be possible to take seconds to finish the segment updating, so let's sleep for a short time + // (50ms ~ 1000ms) and then retry. + std::this_thread::sleep_for(std::chrono::milliseconds(sleep_ms)); + sleep_ms = std::min(sleep_ms * 2, 1000); } } @@ -1126,13 +1137,15 @@ BlockInputStreams DeltaMergeStore::readRaw(const Context & db_context, const DB::Settings & db_settings, const ColumnDefines & columns_to_read, size_t num_streams, + bool keep_order, const SegmentIdSet & read_segments, size_t extra_table_id_index) { SegmentReadTasks tasks; auto dm_context = newDMContext(db_context, db_settings, fmt::format("read_raw_{}", db_context.getCurrentQueryId())); - + // If keep order is required, disable read thread. + auto enable_read_thread = db_context.getSettingsRef().dt_enable_read_thread && !keep_order; { std::shared_lock lock(read_write_mutex); @@ -1164,7 +1177,17 @@ BlockInputStreams DeltaMergeStore::readRaw(const Context & db_context, this->checkSegmentUpdate(dm_context_, segment_, ThreadType::Read); }; size_t final_num_stream = std::min(num_streams, tasks.size()); - auto read_task_pool = std::make_shared(std::move(tasks)); + auto read_task_pool = std::make_shared( + physical_table_id, + dm_context, + columns_to_read, + EMPTY_FILTER, + std::numeric_limits::max(), + DEFAULT_BLOCK_SIZE, + /* is_raw = */ true, + /* do_delete_mark_filter_for_raw = */ false, + std::move(tasks), + after_segment_read); String req_info; if (db_context.getDAGContext() != nullptr && db_context.getDAGContext()->isMPPTask()) @@ -1172,21 +1195,38 @@ BlockInputStreams DeltaMergeStore::readRaw(const Context & db_context, BlockInputStreams res; for (size_t i = 0; i < final_num_stream; ++i) { - BlockInputStreamPtr stream = std::make_shared( - dm_context, - read_task_pool, - after_segment_read, - columns_to_read, - EMPTY_FILTER, - std::numeric_limits::max(), - DEFAULT_BLOCK_SIZE, - /* is_raw_ */ true, - /* do_delete_mark_filter_for_raw_ */ false, // don't do filter based on del_mark = 1 - extra_table_id_index, - physical_table_id, - req_info); + BlockInputStreamPtr stream; + if (enable_read_thread) + { + stream = std::make_shared( + read_task_pool, + columns_to_read, + extra_table_id_index, + physical_table_id, + req_info); + } + else + { + stream = std::make_shared( + dm_context, + read_task_pool, + after_segment_read, + columns_to_read, + EMPTY_FILTER, + std::numeric_limits::max(), + DEFAULT_BLOCK_SIZE, + /* is_raw_ */ true, + /* do_delete_mark_filter_for_raw_ */ false, // don't do filter based on del_mark = 1 + extra_table_id_index, + physical_table_id, + req_info); + } res.push_back(stream); } + if (enable_read_thread) + { + SegmentReadTaskScheduler::instance().add(read_task_pool); + } return res; } @@ -1198,6 +1238,7 @@ BlockInputStreams DeltaMergeStore::read(const Context & db_context, UInt64 max_version, const RSOperatorPtr & filter, const String & tracing_id, + bool keep_order, bool is_fast_mode, size_t expected_block_size, const SegmentIdSet & read_segments, @@ -1205,11 +1246,19 @@ BlockInputStreams DeltaMergeStore::read(const Context & db_context, { // Use the id from MPP/Coprocessor level as tracing_id auto dm_context = newDMContext(db_context, db_settings, tracing_id); - - SegmentReadTasks tasks = getReadTasksByRanges(*dm_context, sorted_ranges, num_streams, read_segments); + // If keep order is required, disable read thread. + auto enable_read_thread = db_context.getSettingsRef().dt_enable_read_thread && !keep_order; + // SegmentReadTaskScheduler and SegmentReadTaskPool use table_id + segment id as unique ID when read thread is enabled. + // 'try_split_task' can result in several read tasks with the same id that can cause some trouble. + // Also, too many read tasks of a segment with different samll ranges is not good for data sharing cache. + SegmentReadTasks tasks = getReadTasksByRanges(*dm_context, sorted_ranges, num_streams, read_segments, /*try_split_task =*/!enable_read_thread); auto tracing_logger = Logger::get(log->name(), dm_context->tracing_id); - LOG_FMT_DEBUG(tracing_logger, "Read create segment snapshot done"); + LOG_FMT_DEBUG(tracing_logger, + "Read create segment snapshot done keep_order {} dt_enable_read_thread {} => enable_read_thread {}", + keep_order, + db_context.getSettingsRef().dt_enable_read_thread, + enable_read_thread); auto after_segment_read = [&](const DMContextPtr & dm_context_, const SegmentPtr & segment_) { // TODO: Update the tracing_id before checkSegmentUpdate? @@ -1218,7 +1267,17 @@ BlockInputStreams DeltaMergeStore::read(const Context & db_context, GET_METRIC(tiflash_storage_read_tasks_count).Increment(tasks.size()); size_t final_num_stream = std::max(1, std::min(num_streams, tasks.size())); - auto read_task_pool = std::make_shared(std::move(tasks)); + auto read_task_pool = std::make_shared( + physical_table_id, + dm_context, + columns_to_read, + filter, + max_version, + expected_block_size, + /* is_raw = */ is_fast_mode, + /* do_delete_mark_filter_for_raw = */ is_fast_mode, + std::move(tasks), + after_segment_read); String req_info; if (db_context.getDAGContext() != nullptr && db_context.getDAGContext()->isMPPTask()) @@ -1226,22 +1285,38 @@ BlockInputStreams DeltaMergeStore::read(const Context & db_context, BlockInputStreams res; for (size_t i = 0; i < final_num_stream; ++i) { - BlockInputStreamPtr stream = std::make_shared( - dm_context, - read_task_pool, - after_segment_read, - columns_to_read, - filter, - max_version, - expected_block_size, - /* is_raw_ */ is_fast_mode, - /* do_delete_mark_filter_for_raw_ */ is_fast_mode, - extra_table_id_index, - physical_table_id, - req_info); + BlockInputStreamPtr stream; + if (enable_read_thread) + { + stream = std::make_shared( + read_task_pool, + columns_to_read, + extra_table_id_index, + physical_table_id, + req_info); + } + else + { + stream = std::make_shared( + dm_context, + read_task_pool, + after_segment_read, + columns_to_read, + filter, + max_version, + expected_block_size, + /* is_raw_ */ is_fast_mode, + /* do_delete_mark_filter_for_raw_ */ is_fast_mode, + extra_table_id_index, + physical_table_id, + req_info); + } res.push_back(stream); } - + if (enable_read_thread) + { + SegmentReadTaskScheduler::instance().add(read_task_pool); + } LOG_FMT_DEBUG(tracing_logger, "Read create stream done"); return res; @@ -2623,7 +2698,8 @@ SegmentReadTasks DeltaMergeStore::getReadTasksByRanges( DMContext & dm_context, const RowKeyRanges & sorted_ranges, size_t expected_tasks_count, - const SegmentIdSet & read_segments) + const SegmentIdSet & read_segments, + bool try_split_task) { SegmentReadTasks tasks; @@ -2678,12 +2754,15 @@ SegmentReadTasks DeltaMergeStore::getReadTasksByRanges( ++seg_it; } } - - /// Try to make task number larger or equal to expected_tasks_count. - auto result_tasks = SegmentReadTask::trySplitReadTasks(tasks, expected_tasks_count); + auto tasks_before_split = tasks.size(); + if (try_split_task) + { + /// Try to make task number larger or equal to expected_tasks_count. + tasks = SegmentReadTask::trySplitReadTasks(tasks, expected_tasks_count); + } size_t total_ranges = 0; - for (auto & task : result_tasks) + for (auto & task : tasks) { /// Merge continuously ranges. task->mergeRanges(); @@ -2695,11 +2774,11 @@ SegmentReadTasks DeltaMergeStore::getReadTasksByRanges( tracing_logger, "[sorted_ranges: {}] [tasks before split: {}] [tasks final: {}] [ranges final: {}]", sorted_ranges.size(), + tasks_before_split, tasks.size(), - result_tasks.size(), total_ranges); - return result_tasks; + return tasks; } } // namespace DM diff --git a/dbms/src/Storages/DeltaMerge/DeltaMergeStore.h b/dbms/src/Storages/DeltaMerge/DeltaMergeStore.h index bd7a5bb542a..62d28e4743b 100644 --- a/dbms/src/Storages/DeltaMerge/DeltaMergeStore.h +++ b/dbms/src/Storages/DeltaMerge/DeltaMergeStore.h @@ -350,6 +350,7 @@ class DeltaMergeStore : private boost::noncopyable const DB::Settings & db_settings, const ColumnDefines & columns_to_read, size_t num_streams, + bool keep_order, const SegmentIdSet & read_segments = {}, size_t extra_table_id_index = InvalidColumnID); @@ -366,6 +367,7 @@ class DeltaMergeStore : private boost::noncopyable UInt64 max_version, const RSOperatorPtr & filter, const String & tracing_id, + bool keep_order, bool is_fast_mode = false, // set true when read in fast mode size_t expected_block_size = DEFAULT_BLOCK_SIZE, const SegmentIdSet & read_segments = {}, @@ -493,7 +495,8 @@ class DeltaMergeStore : private boost::noncopyable SegmentReadTasks getReadTasksByRanges(DMContext & dm_context, const RowKeyRanges & sorted_ranges, size_t expected_tasks_count = 1, - const SegmentIdSet & read_segments = {}); + const SegmentIdSet & read_segments = {}, + bool try_split_task = true); private: void dropAllSegments(bool keep_first_segment); diff --git a/dbms/src/Storages/DeltaMerge/File/DMFileBlockInputStream.cpp b/dbms/src/Storages/DeltaMerge/File/DMFileBlockInputStream.cpp index 56b11b975a1..196a300688b 100644 --- a/dbms/src/Storages/DeltaMerge/File/DMFileBlockInputStream.cpp +++ b/dbms/src/Storages/DeltaMerge/File/DMFileBlockInputStream.cpp @@ -56,6 +56,8 @@ DMFileBlockInputStreamPtr DMFileBlockInputStreamBuilder::build(const DMFilePtr & read_limiter, tracing_id); + bool enable_read_thread = SegmentReaderPoolManager::instance().isSegmentReader(); + DMFileReader reader( dmfile, read_columns, @@ -73,8 +75,9 @@ DMFileBlockInputStreamPtr DMFileBlockInputStreamBuilder::build(const DMFilePtr & read_limiter, rows_threshold_per_read, read_one_pack_every_time, - tracing_id); + tracing_id, + enable_read_thread); - return std::make_shared(std::move(reader)); + return std::make_shared(std::move(reader), enable_read_thread); } } // namespace DB::DM diff --git a/dbms/src/Storages/DeltaMerge/File/DMFileBlockInputStream.h b/dbms/src/Storages/DeltaMerge/File/DMFileBlockInputStream.h index a7f2fe9d556..e1b5fa835c3 100644 --- a/dbms/src/Storages/DeltaMerge/File/DMFileBlockInputStream.h +++ b/dbms/src/Storages/DeltaMerge/File/DMFileBlockInputStream.h @@ -16,9 +16,9 @@ #include #include +#include #include #include - namespace DB { class Context; @@ -29,11 +29,23 @@ namespace DM class DMFileBlockInputStream : public SkippableBlockInputStream { public: - explicit DMFileBlockInputStream(DMFileReader && reader_) + explicit DMFileBlockInputStream(DMFileReader && reader_, bool enable_read_thread_) : reader(std::move(reader_)) - {} + , enable_read_thread(enable_read_thread_) + { + if (enable_read_thread) + { + DMFileReaderPool::instance().add(reader); + } + } - ~DMFileBlockInputStream() = default; + ~DMFileBlockInputStream() + { + if (enable_read_thread) + { + DMFileReaderPool::instance().del(reader); + } + } String getName() const override { return "DMFile"; } @@ -45,6 +57,7 @@ class DMFileBlockInputStream : public SkippableBlockInputStream private: DMFileReader reader; + bool enable_read_thread; }; using DMFileBlockInputStreamPtr = std::shared_ptr; @@ -129,6 +142,7 @@ class DMFileBlockInputStreamBuilder enable_column_cache = settings.dt_enable_stable_column_cache; aio_threshold = settings.min_bytes_to_use_direct_io; max_read_buffer_size = settings.max_read_buffer_size; + enable_read_thread = settings.dt_enable_read_thread; return *this; } DMFileBlockInputStreamBuilder & setCaches(const MarkCachePtr & mark_cache_, const MinMaxIndexCachePtr & index_cache_) @@ -159,7 +173,7 @@ class DMFileBlockInputStreamBuilder size_t max_read_buffer_size{}; size_t rows_threshold_per_read = DMFILE_READ_ROWS_THRESHOLD; bool read_one_pack_every_time = false; - + bool enable_read_thread = false; String tracing_id; }; diff --git a/dbms/src/Storages/DeltaMerge/File/DMFileReader.cpp b/dbms/src/Storages/DeltaMerge/File/DMFileReader.cpp index 779de36da63..b00e3cc5c10 100644 --- a/dbms/src/Storages/DeltaMerge/File/DMFileReader.cpp +++ b/dbms/src/Storages/DeltaMerge/File/DMFileReader.cpp @@ -226,7 +226,8 @@ DMFileReader::DMFileReader( const ReadLimiterPtr & read_limiter, size_t rows_threshold_per_read_, bool read_one_pack_every_time_, - const String & tracing_id_) + const String & tracing_id_, + bool enable_col_sharing_cache) : dmfile(dmfile_) , read_columns(read_columns_) , is_common_handle(is_common_handle_) @@ -266,6 +267,14 @@ DMFileReader::DMFileReader( const auto data_type = dmfile->getColumnStat(cd.id).type; data_type->enumerateStreams(callback, {}); } + if (enable_col_sharing_cache) + { + col_data_cache = std::make_unique(path(), read_columns, log); + for (const auto & cd : read_columns) + { + last_read_from_cache[cd.id] = false; + } + } } bool DMFileReader::shouldSeek(size_t pack_id) @@ -305,7 +314,6 @@ Block DMFileReader::read() const auto & use_packs = pack_filter.getUsePacks(); if (next_pack_id >= use_packs.size()) return {}; - // Find max continuing rows we can read. size_t start_pack_id = next_pack_id; // When single_file_mode is true, or read_one_pack_every_time is true, we can just read one pack every time. @@ -440,7 +448,9 @@ Block DMFileReader::read() { rows_count += pack_stats[cursor].rows; } - readFromDisk(cd, column, range.first, rows_count, skip_packs_by_column[i], single_file_mode); + ColumnPtr col; + readColumn(cd, col, range.first, range.second - range.first, rows_count, skip_packs_by_column[i], single_file_mode); + column->insertRangeFrom(*col, 0, col->size()); skip_packs_by_column[i] = 0; } else @@ -462,8 +472,8 @@ Block DMFileReader::read() else { auto data_type = dmfile->getColumnStat(cd.id).type; - auto column = data_type->createColumn(); - readFromDisk(cd, column, start_pack_id, read_rows, skip_packs_by_column[i], single_file_mode); + ColumnPtr column; + readColumn(cd, column, start_pack_id, read_packs, read_rows, skip_packs_by_column[i], single_file_mode); auto converted_column = convertColumnByColumnDefineIfNeed(data_type, std::move(column), cd); res.insert(ColumnWithTypeAndName{std::move(converted_column), cd.type, cd.name, cd.id}); @@ -534,5 +544,58 @@ void DMFileReader::readFromDisk( } } +void DMFileReader::readColumn(ColumnDefine & column_define, + ColumnPtr & column, + size_t start_pack_id, + size_t pack_count, + size_t read_rows, + size_t skip_packs, + bool force_seek) +{ + if (!getCachedPacks(column_define.id, start_pack_id, pack_count, read_rows, column)) + { + auto data_type = dmfile->getColumnStat(column_define.id).type; + auto col = data_type->createColumn(); + readFromDisk(column_define, col, start_pack_id, read_rows, skip_packs, force_seek || last_read_from_cache[column_define.id]); + column = std::move(col); + last_read_from_cache[column_define.id] = false; + } + else + { + last_read_from_cache[column_define.id] = true; + } + + if (col_data_cache != nullptr) + { + DMFileReaderPool::instance().set(*this, column_define.id, start_pack_id, pack_count, column); + } +} + +void DMFileReader::addCachedPacks(ColId col_id, size_t start_pack_id, size_t pack_count, ColumnPtr & col) +{ + if (col_data_cache == nullptr) + { + return; + } + if (next_pack_id >= start_pack_id + pack_count) + { + col_data_cache->addStale(); + } + else + { + col_data_cache->add(col_id, start_pack_id, pack_count, col); + } +} + +bool DMFileReader::getCachedPacks(ColId col_id, size_t start_pack_id, size_t pack_count, size_t read_rows, ColumnPtr & col) +{ + if (col_data_cache == nullptr) + { + return false; + } + auto found = col_data_cache->get(col_id, start_pack_id, pack_count, read_rows, col, dmfile->getColumnStat(col_id).type); + col_data_cache->del(col_id, next_pack_id); + return found; +} } // namespace DM } // namespace DB diff --git a/dbms/src/Storages/DeltaMerge/File/DMFileReader.h b/dbms/src/Storages/DeltaMerge/File/DMFileReader.h index c04c93871a7..71f6df84bf9 100644 --- a/dbms/src/Storages/DeltaMerge/File/DMFileReader.h +++ b/dbms/src/Storages/DeltaMerge/File/DMFileReader.h @@ -21,6 +21,7 @@ #include #include #include +#include #include #include @@ -91,7 +92,8 @@ class DMFileReader const ReadLimiterPtr & read_limiter, size_t rows_threshold_per_read_, bool read_one_pack_every_time_, - const String & tracing_id_); + const String & tracing_id_, + bool enable_col_sharing_cache); Block getHeader() const { return toEmptyBlock(read_columns); } @@ -99,6 +101,15 @@ class DMFileReader /// Return false if it is the end of stream. bool getSkippedRows(size_t & skip_rows); Block read(); + UInt64 fileId() const + { + return dmfile->fileId(); + } + std::string path() const + { + return dmfile->path(); + } + void addCachedPacks(ColId col_id, size_t start_pack_id, size_t pack_count, ColumnPtr & col); private: bool shouldSeek(size_t pack_id); @@ -109,6 +120,14 @@ class DMFileReader size_t read_rows, size_t skip_packs, bool force_seek); + void readColumn(ColumnDefine & column_define, + ColumnPtr & column, + size_t start_pack_id, + size_t pack_count, + size_t read_rows, + size_t skip_packs, + bool force_seek); + bool getCachedPacks(ColId col_id, size_t start_pack_id, size_t pack_count, size_t read_rows, ColumnPtr & col); private: DMFilePtr dmfile; @@ -146,6 +165,9 @@ class DMFileReader FileProviderPtr file_provider; LoggerPtr log; + + std::unique_ptr col_data_cache; + std::unordered_map last_read_from_cache; }; } // namespace DM diff --git a/dbms/src/Storages/DeltaMerge/ReadThread/CPU.cpp b/dbms/src/Storages/DeltaMerge/ReadThread/CPU.cpp new file mode 100644 index 00000000000..fbecd79c6d8 --- /dev/null +++ b/dbms/src/Storages/DeltaMerge/ReadThread/CPU.cpp @@ -0,0 +1,125 @@ +// Copyright 2022 PingCAP, Ltd. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. +#include +#include +#include +#include +#include +#include + +#include +#include + +namespace DB::DM +{ +// In Linux a numa node is represented by a device directory, such as '/sys/devices/system/node/node0', '/sys/devices/system/node/node01'. +static inline bool isNodeDir(const std::string & name) +{ + return name.size() > 4 && name.substr(0, 4) == "node" && std::all_of(name.begin() + 4, name.end(), [](unsigned char c) { return std::isdigit(c); }); +} + +// Under a numa node directory is CPU cores and memory, such as '/sys/devices/system/node/node0/cpu0' and '/sys/devices/system/node/node0/memory0'. +static inline bool isCPU(const std::string & name) +{ + return name.size() > 3 && name.substr(0, 3) == "cpu" && std::all_of(name.begin() + 3, name.end(), [](unsigned char c) { return std::isdigit(c); }); +} + +static inline int parseCPUNumber(const std::string & name) +{ + return std::stoi(name.substr(3)); +} + +// Scan the numa node directory and parse the CPU numbers. +static inline std::vector getCPUs(const std::string & dir_name) +{ + std::vector cpus; + Poco::File dir(dir_name); + Poco::DirectoryIterator end; + for (auto iter = Poco::DirectoryIterator(dir); iter != end; ++iter) + { + if (isCPU(iter.name())) + { + cpus.push_back(parseCPUNumber(iter.name())); + } + } + return cpus; +} + +// TODO: What if the process running in the container and the CPU is limited. + +// Scan the device directory and parse the CPU information. +std::vector> getLinuxNumaNodes() +{ + static const std::string nodes_dir_name{"/sys/devices/system/node"}; + static const std::string cpus_dir_name{"/sys/devices/system/cpu"}; + + std::vector> numa_nodes; + Poco::File nodes(nodes_dir_name); + if (!nodes.exists() || !nodes.isDirectory()) + { + auto cpus = getCPUs(cpus_dir_name); + if (cpus.empty()) + { + throw Exception("Not recognize CPU: " + cpus_dir_name); + } + numa_nodes.push_back(std::move(cpus)); + } + else + { + Poco::DirectoryIterator end; + for (Poco::DirectoryIterator iter(nodes); iter != end; ++iter) + { + if (!isNodeDir(iter.name())) + { + continue; + } + auto dir_name = nodes_dir_name + "/" + iter.name(); + auto cpus = getCPUs(dir_name); + if (cpus.empty()) + { + throw Exception("Not recognize CPU: " + nodes_dir_name); + } + numa_nodes.push_back(std::move(cpus)); + } + } + if (numa_nodes.empty()) + { + throw Exception("Not recognize CPU"); + } + return numa_nodes; +} + +std::vector> getNumaNodes(Poco::Logger * log) +{ + try + { + return getLinuxNumaNodes(); + } + catch (Exception & e) + { + LOG_FMT_WARNING(log, "{}", e.message()); + } + catch (std::exception & e) + { + LOG_FMT_WARNING(log, "{}", e.what()); + } + catch (...) + { + LOG_FMT_WARNING(log, "Unknow Error"); + } + LOG_FMT_WARNING(log, "Cannot recognize the CPU NUMA infomation, use the CPU as 'one numa node'"); + std::vector> numa_nodes(1); // "One numa node" + return numa_nodes; +} +} // namespace DB::DM \ No newline at end of file diff --git a/dbms/src/Storages/DeltaMerge/ReadThread/CPU.h b/dbms/src/Storages/DeltaMerge/ReadThread/CPU.h new file mode 100644 index 00000000000..2d1f564a43a --- /dev/null +++ b/dbms/src/Storages/DeltaMerge/ReadThread/CPU.h @@ -0,0 +1,24 @@ +// Copyright 2022 PingCAP, Ltd. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. +#pragma once + +#include + +#include + +namespace DB::DM +{ +// `getNumaNodes` returns cpus of each Numa node. +std::vector> getNumaNodes(Poco::Logger * log); +} // namespace DB::DM \ No newline at end of file diff --git a/dbms/src/Storages/DeltaMerge/ReadThread/CircularScanList.h b/dbms/src/Storages/DeltaMerge/ReadThread/CircularScanList.h new file mode 100644 index 00000000000..a9a976f2f30 --- /dev/null +++ b/dbms/src/Storages/DeltaMerge/ReadThread/CircularScanList.h @@ -0,0 +1,104 @@ +// Copyright 2022 PingCAP, Ltd. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. +#pragma once + +#include +#include + +namespace DB::DM +{ +// CircularScanList is a special circular list. +// It remembers the location of the last iteration and will check whether the object is expired. +template +class CircularScanList +{ +public: + using Value = std::shared_ptr; + + CircularScanList() + : last_itr(l.end()) + {} + + void add(const Value & ptr) + { + l.push_back(ptr); + } + + Value next() + { + last_itr = nextItr(last_itr); + while (!l.empty()) + { + auto ptr = *last_itr; + if (ptr->valid()) + { + return ptr; + } + else + { + last_itr = l.erase(last_itr); + if (last_itr == l.end()) + { + last_itr = l.begin(); + } + } + } + return nullptr; + } + + std::pair count(int64_t table_id) const + { + int64_t valid = 0; + int64_t invalid = 0; + for (const auto & p : l) + { + if (table_id == 0 || p->tableId() == table_id) + { + p->valid() ? valid++ : invalid++; + } + } + return {valid, invalid}; + } + + Value get(uint64_t pool_id) const + { + for (const auto & p : l) + { + if (p->poolId() == pool_id) + { + return p; + } + } + return nullptr; + } + +private: + using Iter = typename std::list::iterator; + Iter nextItr(Iter itr) + { + if (itr == l.end() || std::next(itr) == l.end()) + { + return l.begin(); + } + else + { + return std::next(itr); + } + } + + std::list l; + Iter last_itr; +}; + +} // namespace DB::DM \ No newline at end of file diff --git a/dbms/src/Storages/DeltaMerge/ReadThread/ColumnSharingCache.cpp b/dbms/src/Storages/DeltaMerge/ReadThread/ColumnSharingCache.cpp new file mode 100644 index 00000000000..13b187167b6 --- /dev/null +++ b/dbms/src/Storages/DeltaMerge/ReadThread/ColumnSharingCache.cpp @@ -0,0 +1,63 @@ +// Copyright 2022 PingCAP, Ltd. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. +#include +#include + +namespace DB::DM +{ +DMFileReaderPool & DMFileReaderPool::instance() +{ + static DMFileReaderPool reader_pool; + return reader_pool; +} + +void DMFileReaderPool::add(DMFileReader & reader) +{ + std::lock_guard lock(mtx); + readers[reader.path()].insert(&reader); +} + +void DMFileReaderPool::del(DMFileReader & reader) +{ + std::lock_guard lock(mtx); + auto itr = readers.find(reader.path()); + if (itr == readers.end()) + { + return; + } + itr->second.erase(&reader); + if (itr->second.empty()) + { + readers.erase(itr); + } +} + +void DMFileReaderPool::set(DMFileReader & from_reader, int64_t col_id, size_t start, size_t count, ColumnPtr & col) +{ + std::lock_guard lock(mtx); + auto itr = readers.find(from_reader.path()); + if (itr == readers.end()) + { + return; + } + for (auto * r : itr->second) + { + if (&from_reader == r) + { + continue; + } + r->addCachedPacks(col_id, start, count, col); + } +} +} // namespace DB::DM \ No newline at end of file diff --git a/dbms/src/Storages/DeltaMerge/ReadThread/ColumnSharingCache.h b/dbms/src/Storages/DeltaMerge/ReadThread/ColumnSharingCache.h new file mode 100644 index 00000000000..446e6777338 --- /dev/null +++ b/dbms/src/Storages/DeltaMerge/ReadThread/ColumnSharingCache.h @@ -0,0 +1,234 @@ +// Copyright 2022 PingCAP, Ltd. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. +#pragma once + +#include +#include +#include +#include +#include + +#include + +namespace DB::DM +{ +using PackId = size_t; + +enum class ColumnCacheStatus +{ + ADD_COUNT = 0, + ADD_STALE, + GET_MISS, + GET_PART, + GET_HIT, + GET_COPY, + + _TOTAL_COUNT, +}; + +class ColumnSharingCache +{ +public: + struct ColumnData + { + size_t pack_count{0}; + ColumnPtr col_data; + }; + + void add(size_t start_pack_id, size_t pack_count, ColumnPtr & col_data) + { + GET_METRIC(tiflash_storage_read_thread_counter, type_add_cache_succ).Increment(); + std::lock_guard lock(mtx); + auto & value = packs[start_pack_id]; + if (value.pack_count < pack_count) + { + value.pack_count = pack_count; + value.col_data = col_data; + } + } + + ColumnCacheStatus get(size_t start_pack_id, size_t pack_count, size_t read_rows, ColumnPtr & col_data, DataTypePtr data_type) + { + ColumnCacheStatus status; + std::lock_guard lock(mtx); + auto target = packs.find(start_pack_id); + if (target == packs.end()) + { + GET_METRIC(tiflash_storage_read_thread_counter, type_get_cache_miss).Increment(); + status = ColumnCacheStatus::GET_MISS; + } + else if (target->second.pack_count < pack_count) + { + GET_METRIC(tiflash_storage_read_thread_counter, type_get_cache_part).Increment(); + status = ColumnCacheStatus::GET_PART; + } + else if (target->second.pack_count == pack_count) + { + GET_METRIC(tiflash_storage_read_thread_counter, type_get_cache_hit).Increment(); + status = ColumnCacheStatus::GET_HIT; + col_data = target->second.col_data; + } + else + { + GET_METRIC(tiflash_storage_read_thread_counter, type_get_cache_copy).Increment(); + status = ColumnCacheStatus::GET_COPY; + auto column = data_type->createColumn(); + column->insertRangeFrom(*(target->second.col_data), 0, read_rows); + col_data = std::move(column); + } + return status; + } + + void del(size_t upper_start_pack_id) + { + std::lock_guard lock(mtx); + for (auto itr = packs.begin(); itr != packs.end();) + { + if (itr->first + itr->second.pack_count <= upper_start_pack_id) + { + itr = packs.erase(itr); + } + else + { + break; + } + } + } + +private: + std::mutex mtx; + // start_pack_id -> + std::map packs; +}; + +// `ColumnSharingCacheMap` is a per DMFileReader cache. +// It store a ColumnSharingCache(std::map) for each column. +// When read threads are enable, each DMFileReader will be add to DMFileReaderPool, +// so we can find DMFileReaders of the same DMFile easily. +// Each DMFileReader will add the block that they read to other DMFileReaders' cache if the start_pack_id of the block +// is greater or equal than the next_pack_id of the DMFileReader —— This means that the DMFileReader maybe read the block later. +class ColumnSharingCacheMap +{ +public: + ColumnSharingCacheMap(const std::string & dmfile_name_, const ColumnDefines & cds, LoggerPtr & log_) + : dmfile_name(dmfile_name_) + , stats(static_cast(ColumnCacheStatus::_TOTAL_COUNT)) + , log(log_) + { + for (const auto & cd : cds) + { + addColumn(cd.id); + } + } + + ~ColumnSharingCacheMap() + { + LOG_FMT_DEBUG(log, "dmfile {} stat {}", dmfile_name, statString()); + } + + // `addStale` just do some statistics. + void addStale() + { + GET_METRIC(tiflash_storage_read_thread_counter, type_add_cache_stale).Increment(); + stats[static_cast(ColumnCacheStatus::ADD_STALE)].fetch_add(1, std::memory_order_relaxed); + } + + void add(int64_t col_id, size_t start_pack_id, size_t pack_count, ColumnPtr & col_data) + { + stats[static_cast(ColumnCacheStatus::ADD_COUNT)].fetch_add(1, std::memory_order_relaxed); + auto itr = cols.find(col_id); + if (itr == cols.end()) + { + return; + } + itr->second.add(start_pack_id, pack_count, col_data); + } + + bool get(int64_t col_id, size_t start_pack_id, size_t pack_count, size_t read_rows, ColumnPtr & col_data, DataTypePtr data_type) + { + auto status = ColumnCacheStatus::GET_MISS; + auto itr = cols.find(col_id); + if (itr != cols.end()) + { + status = itr->second.get(start_pack_id, pack_count, read_rows, col_data, data_type); + } + stats[static_cast(status)].fetch_add(1, std::memory_order_relaxed); + return status == ColumnCacheStatus::GET_HIT || status == ColumnCacheStatus::GET_COPY; + } + + // Each read operator of DMFileReader will advance the next_pack_id. + // So we can delete packs if their pack_id is less than next_pack_id to save memory. + void del(int64_t col_id, size_t upper_start_pack_id) + { + auto itr = cols.find(col_id); + if (itr != cols.end()) + { + itr->second.del(upper_start_pack_id); + } + } + +private: + void addColumn(int64_t col_id) + { + cols[col_id]; + } + std::string statString() const + { + auto add_count = stats[static_cast(ColumnCacheStatus::ADD_COUNT)].load(std::memory_order_relaxed); + auto add_stale = stats[static_cast(ColumnCacheStatus::ADD_STALE)].load(std::memory_order_relaxed); + auto get_miss = stats[static_cast(ColumnCacheStatus::GET_MISS)].load(std::memory_order_relaxed); + auto get_part = stats[static_cast(ColumnCacheStatus::GET_PART)].load(std::memory_order_relaxed); + auto get_hit = stats[static_cast(ColumnCacheStatus::GET_HIT)].load(std::memory_order_relaxed); + auto get_copy = stats[static_cast(ColumnCacheStatus::GET_COPY)].load(std::memory_order_relaxed); + auto add_total = add_count + add_stale; + auto get_cached = get_hit + get_copy; + auto get_total = get_miss + get_part + get_hit + get_copy; + return fmt::format("add_count {} add_stale {} add_ratio {} get_miss {} get_part {} get_hit {} get_copy {} cached_ratio {}", + add_count, + add_stale, + add_total > 0 ? add_count * 1.0 / add_total : 0, + get_miss, + get_part, + get_hit, + get_copy, + get_total > 0 ? get_cached * 1.0 / get_total : 0); + } + std::string dmfile_name; + std::unordered_map cols; + std::vector> stats; + LoggerPtr log; +}; + +class DMFileReader; + +// DMFileReaderPool holds all the DMFileReader objects, so we can easily find DMFileReader objects with the same DMFile ID. +// When a DMFileReader object successfully reads a column's packs, it will try to put these packs into other DMFileReader objects' cache. +class DMFileReaderPool +{ +public: + static DMFileReaderPool & instance(); + DMFileReaderPool() = default; + ~DMFileReaderPool() = default; + DISALLOW_COPY_AND_MOVE(DMFileReaderPool); + + void add(DMFileReader & reader); + void del(DMFileReader & reader); + void set(DMFileReader & from_reader, int64_t col_id, size_t start, size_t count, ColumnPtr & col); + +private: + std::mutex mtx; + std::unordered_map> readers; +}; + +} // namespace DB::DM \ No newline at end of file diff --git a/dbms/src/Storages/DeltaMerge/ReadThread/MergedTask.cpp b/dbms/src/Storages/DeltaMerge/ReadThread/MergedTask.cpp new file mode 100644 index 00000000000..c81fcfe8ccd --- /dev/null +++ b/dbms/src/Storages/DeltaMerge/ReadThread/MergedTask.cpp @@ -0,0 +1,123 @@ +// Copyright 2022 PingCAP, Ltd. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. +#include + +namespace DB::DM +{ +int MergedTask::readBlock() +{ + initOnce(); + return readOneBlock(); +} + +void MergedTask::initOnce() +{ + if (inited) + { + return; + } + + for (cur_idx = 0; cur_idx < static_cast(units.size()); cur_idx++) + { + auto & [pool, task, stream] = units[cur_idx]; + if (!pool->valid()) + { + setStreamFinished(cur_idx); + continue; + } + stream = pool->buildInputStream(task); + } + + inited = true; +} + +int MergedTask::readOneBlock() +{ + int read_block_count = 0; + for (cur_idx = 0; cur_idx < static_cast(units.size()); cur_idx++) + { + if (isStreamFinished(cur_idx)) + { + continue; + } + + auto & [pool, task, stream] = units[cur_idx]; + + if (!pool->valid()) + { + setStreamFinished(cur_idx); + continue; + } + + if (pool->getFreeBlockSlots() <= 0) + { + continue; + } + + if (pool->readOneBlock(stream, task->segment)) + { + read_block_count++; + } + else + { + setStreamFinished(cur_idx); + } + } + return read_block_count; +} + +void MergedTask::setException(const DB::Exception & e) +{ + if (cur_idx >= 0 && cur_idx < static_cast(units.size())) + { + auto & pool = units[cur_idx].pool; + if (pool != nullptr) + { + pool->setException(e); + } + } + else + { + for (auto & unit : units) + { + if (unit.pool != nullptr) + { + unit.pool->setException(e); + } + } + } +} + +MergedTaskPtr MergedTaskPool::pop(uint64_t pool_id) +{ + std::lock_guard lock(mtx); + MergedTaskPtr target; + for (auto itr = merged_task_pool.begin(); itr != merged_task_pool.end(); ++itr) + { + if ((*itr)->containPool(pool_id)) + { + target = *itr; + merged_task_pool.erase(itr); + break; + } + } + return target; +} + +void MergedTaskPool::push(const MergedTaskPtr & t) +{ + std::lock_guard lock(mtx); + merged_task_pool.push_back(t); +} +} // namespace DB::DM \ No newline at end of file diff --git a/dbms/src/Storages/DeltaMerge/ReadThread/MergedTask.h b/dbms/src/Storages/DeltaMerge/ReadThread/MergedTask.h new file mode 100644 index 00000000000..8233cd2e431 --- /dev/null +++ b/dbms/src/Storages/DeltaMerge/ReadThread/MergedTask.h @@ -0,0 +1,156 @@ +// Copyright 2022 PingCAP, Ltd. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. +#pragma once +#include +#include +#include + +namespace DB::DM +{ +struct MergedUnit +{ + MergedUnit(const SegmentReadTaskPoolPtr & pool_, const SegmentReadTaskPtr & task_) + : pool(pool_) + , task(task_) + {} + + SegmentReadTaskPoolPtr pool; // The information of a read request. + SegmentReadTaskPtr task; // The information of a segment that want to read. + BlockInputStreamPtr stream; // BlockInputStream of a segment, will be created by read threads. +}; + +// MergedTask merges the same segment of different SegmentReadTaskPools. +// Read segment input streams of different SegmentReadTaskPools sequentially to improve cache sharing. +// MergedTask is NOT thread-safe. +class MergedTask +{ +public: + static int64_t getPassiveMergedSegments() + { + return passive_merged_segments.load(std::memory_order_relaxed); + } + + MergedTask(uint64_t seg_id_, std::vector && units_) + : seg_id(seg_id_) + , units(std::move(units_)) + , inited(false) + , cur_idx(-1) + , finished_count(0) + , log(&Poco::Logger::get("MergedTask")) + { + passive_merged_segments.fetch_add(units.size() - 1, std::memory_order_relaxed); + GET_METRIC(tiflash_storage_read_thread_gauge, type_merged_task).Increment(); + } + ~MergedTask() + { + passive_merged_segments.fetch_sub(units.size() - 1, std::memory_order_relaxed); + GET_METRIC(tiflash_storage_read_thread_gauge, type_merged_task).Decrement(); + GET_METRIC(tiflash_storage_read_thread_seconds, type_merged_task).Observe(sw.elapsedSeconds()); + } + + int readBlock(); + + bool allStreamsFinished() const + { + return finished_count >= units.size(); + } + + uint64_t getSegmentId() const + { + return seg_id; + } + + size_t getPoolCount() const + { + return units.size(); + } + + std::vector getPoolIds() const + { + std::vector ids; + ids.reserve(units.size()); + for (const auto & unit : units) + { + if (unit.pool != nullptr) + { + ids.push_back(unit.pool->poolId()); + } + } + return ids; + } + + bool containPool(uint64_t pool_id) const + { + for (const auto & unit : units) + { + if (unit.pool != nullptr && unit.pool->poolId() == pool_id) + { + return true; + } + } + return false; + } + void setException(const DB::Exception & e); + +private: + void initOnce(); + int readOneBlock(); + + bool isStreamFinished(size_t i) const + { + return units[i].pool == nullptr && units[i].task == nullptr && units[i].stream == nullptr; + } + + void setStreamFinished(size_t i) + { + if (!isStreamFinished(i)) + { + units[i].pool = nullptr; + units[i].task = nullptr; + units[i].stream = nullptr; + finished_count++; + } + } + + uint64_t seg_id; + std::vector units; + bool inited; + int cur_idx; + size_t finished_count; + Poco::Logger * log; + Stopwatch sw; + inline static std::atomic passive_merged_segments{0}; +}; + +using MergedTaskPtr = std::shared_ptr; + +// MergedTaskPool is a MergedTask list. +// When SegmentReadTaskPool's block queue reaching limit, read thread will push MergedTask into it. +// The scheduler thread will try to pop a MergedTask of a related pool_id before build a new MergedTask object. +class MergedTaskPool +{ +public: + MergedTaskPool() + : log(&Poco::Logger::get("MergedTaskPool")) + {} + + MergedTaskPtr pop(uint64_t pool_id); + void push(const MergedTaskPtr & t); + +private: + std::mutex mtx; + std::list merged_task_pool; + Poco::Logger * log; +}; +} // namespace DB::DM \ No newline at end of file diff --git a/dbms/src/Storages/DeltaMerge/ReadThread/SegmentReadTaskScheduler.cpp b/dbms/src/Storages/DeltaMerge/ReadThread/SegmentReadTaskScheduler.cpp new file mode 100644 index 00000000000..9bec9e18afa --- /dev/null +++ b/dbms/src/Storages/DeltaMerge/ReadThread/SegmentReadTaskScheduler.cpp @@ -0,0 +1,211 @@ +// Copyright 2022 PingCAP, Ltd. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. +#include +#include +#include + +namespace DB::DM +{ +SegmentReadTaskScheduler::SegmentReadTaskScheduler() + : stop(false) + , log(&Poco::Logger::get("SegmentReadTaskScheduler")) +{ + sched_thread = std::thread(&SegmentReadTaskScheduler::schedThread, this); +} + +SegmentReadTaskScheduler::~SegmentReadTaskScheduler() +{ + setStop(); + sched_thread.join(); +} + +void SegmentReadTaskScheduler::add(const SegmentReadTaskPoolPtr & pool) +{ + std::lock_guard lock(mtx); + + read_pools.add(pool); + + std::unordered_set seg_ids; + for (const auto & task : pool->getTasks()) + { + auto seg_id = task->segment->segmentId(); + merging_segments[pool->tableId()][seg_id].push_back(pool->poolId()); + if (!seg_ids.insert(seg_id).second) + { + throw DB::Exception(fmt::format("Not support split segment task. seg_ids {} => seg_id {} already exist.", seg_ids, seg_id)); + } + } + auto block_slots = pool->getFreeBlockSlots(); + auto [unexpired, expired] = read_pools.count(pool->tableId()); + LOG_FMT_DEBUG(log, "add pool {} table {} block_slots {} segment count {} segments {} unexpired pool {} expired pool {}", // + pool->poolId(), + pool->tableId(), + block_slots, + seg_ids.size(), + seg_ids, + unexpired, + expired); +} + +std::pair SegmentReadTaskScheduler::scheduleMergedTask() +{ + std::lock_guard lock(mtx); + auto pool = scheduleSegmentReadTaskPoolUnlock(); + if (pool == nullptr) + { + // No SegmentReadTaskPool to schedule. Maybe no read request or + // block queue of each SegmentReadTaskPool reaching the limit. + return {nullptr, false}; + } + + auto merged_task = merged_task_pool.pop(pool->poolId()); + if (merged_task != nullptr) + { + GET_METRIC(tiflash_storage_read_thread_counter, type_sche_from_cache).Increment(); + return {merged_task, true}; + } + + auto segment = scheduleSegmentUnlock(pool); + if (!segment) + { + // The number of active segments reaches the limit. + GET_METRIC(tiflash_storage_read_thread_counter, type_sche_no_segment).Increment(); + return {nullptr, true}; + } + auto pools = getPoolsUnlock(segment->second); + if (pools.empty()) + { + // Maybe SegmentReadTaskPools are expired because of upper threads finish the request. + return {nullptr, true}; + } + + std::vector units; + units.reserve(pools.size()); + for (auto & pool : pools) + { + units.emplace_back(pool, pool->getTask(segment->first)); + } + GET_METRIC(tiflash_storage_read_thread_counter, type_sche_new_task).Increment(); + + return {std::make_shared(segment->first, std::move(units)), true}; +} + +SegmentReadTaskPools SegmentReadTaskScheduler::getPoolsUnlock(const std::vector & pool_ids) +{ + SegmentReadTaskPools pools; + pools.reserve(pool_ids.size()); + for (uint64_t id : pool_ids) + { + auto p = read_pools.get(id); + if (p != nullptr) + { + pools.push_back(p); + } + } + return pools; +} + +SegmentReadTaskPoolPtr SegmentReadTaskScheduler::scheduleSegmentReadTaskPoolUnlock() +{ + auto [unexpired, expired] = read_pools.count(0); + for (int64_t i = 0; i < unexpired; i++) + { + auto pool = read_pools.next(); + if (pool != nullptr && pool->getFreeBlockSlots() > 0) + { + return pool; + } + } + if (unexpired == 0) + { + GET_METRIC(tiflash_storage_read_thread_counter, type_sche_no_pool).Increment(); + } + else + { + GET_METRIC(tiflash_storage_read_thread_counter, type_sche_no_slot).Increment(); + } + return nullptr; +} + +std::optional>> SegmentReadTaskScheduler::scheduleSegmentUnlock(const SegmentReadTaskPoolPtr & pool) +{ + auto [unexpired, expired] = read_pools.count(pool->tableId()); + auto expected_merge_seg_count = std::min(unexpired, 2); + auto itr = merging_segments.find(pool->tableId()); + if (itr == merging_segments.end()) + { + // No segment of tableId left. + return std::nullopt; + } + std::optional>> result; + auto & segments = itr->second; + auto target = pool->scheduleSegment(segments, expected_merge_seg_count); + if (target != segments.end()) + { + if (MergedTask::getPassiveMergedSegments() < 100 || target->second.size() == 1) + { + result = *target; + segments.erase(target); + if (segments.empty()) + { + merging_segments.erase(itr); + } + } + else + { + result = std::pair{target->first, std::vector(1, pool->poolId())}; + auto mutable_target = segments.find(target->first); + auto itr = std::find(mutable_target->second.begin(), mutable_target->second.end(), pool->poolId()); + *itr = mutable_target->second.back(); // SegmentReadTaskPool::scheduleSegment ensures `pool->poolId` must exists in `target`. + mutable_target->second.resize(mutable_target->second.size() - 1); + } + } + return result; +} + +void SegmentReadTaskScheduler::setStop() +{ + stop.store(true, std::memory_order_relaxed); +} + +bool SegmentReadTaskScheduler::isStop() const +{ + return stop.load(std::memory_order_relaxed); +} + +bool SegmentReadTaskScheduler::schedule() +{ + Stopwatch sw; + auto [merged_task, run_sche] = scheduleMergedTask(); + if (merged_task != nullptr) + { + LOG_FMT_DEBUG(log, "scheduleMergedTask seg_id {} pools {} => {} ms", merged_task->getSegmentId(), merged_task->getPoolIds(), sw.elapsedMilliseconds()); + SegmentReaderPoolManager::instance().addTask(std::move(merged_task)); + } + return run_sche; +} + +void SegmentReadTaskScheduler::schedThread() +{ + while (!isStop()) + { + if (!schedule()) + { + using namespace std::chrono_literals; + std::this_thread::sleep_for(2ms); + } + } +} + +} // namespace DB::DM \ No newline at end of file diff --git a/dbms/src/Storages/DeltaMerge/ReadThread/SegmentReadTaskScheduler.h b/dbms/src/Storages/DeltaMerge/ReadThread/SegmentReadTaskScheduler.h new file mode 100644 index 00000000000..51aef4d176c --- /dev/null +++ b/dbms/src/Storages/DeltaMerge/ReadThread/SegmentReadTaskScheduler.h @@ -0,0 +1,83 @@ +// Copyright 2022 PingCAP, Ltd. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. +#pragma once + +#include +#include +#include + +#include +namespace DB::DM +{ +using SegmentReadTaskPoolList = CircularScanList; + +// SegmentReadTaskScheduler is a global singleton. +// All SegmentReadTaskPool will be added to it and be scheduled by it. + +// 1. DeltaMergeStore::read/readRaw will call SegmentReadTaskScheduler::add to add a SegmentReadTaskPool object to the `read_pools` list and +// index segments information into `merging_segments`. +// 2. A schedule-thread will scheduling read tasks: +// a. It scans the read_pools list and choosing a SegmentReadTaskPool. +// b. Chooses a segment of the SegmentReadTaskPool and build a MergedTask. +// c. Sends the MergedTask to read threads(SegmentReader). +class SegmentReadTaskScheduler +{ +public: + static SegmentReadTaskScheduler & instance() + { + static SegmentReadTaskScheduler scheduler; + return scheduler; + } + + ~SegmentReadTaskScheduler(); + DISALLOW_COPY_AND_MOVE(SegmentReadTaskScheduler); + + // Add SegmentReadTaskPool to `read_pools` and index segments into merging_segments. + void add(const SegmentReadTaskPoolPtr & pool); + + void pushMergedTask(const MergedTaskPtr & p) + { + merged_task_pool.push(p); + } + +private: + SegmentReadTaskScheduler(); + + // Choose segment to read. + // Returns + std::pair scheduleMergedTask(); + + void setStop(); + bool isStop() const; + bool schedule(); + void schedThread(); + + SegmentReadTaskPools getPoolsUnlock(const std::vector & pool_ids); + // + std::optional>> scheduleSegmentUnlock(const SegmentReadTaskPoolPtr & pool); + SegmentReadTaskPoolPtr scheduleSegmentReadTaskPoolUnlock(); + + std::mutex mtx; + SegmentReadTaskPoolList read_pools; + // table_id -> {seg_id -> pool_ids, seg_id -> pool_ids, ...} + std::unordered_map>> merging_segments; + + MergedTaskPool merged_task_pool; + + std::atomic stop; + std::thread sched_thread; + + Poco::Logger * log; +}; +} // namespace DB::DM \ No newline at end of file diff --git a/dbms/src/Storages/DeltaMerge/ReadThread/SegmentReader.cpp b/dbms/src/Storages/DeltaMerge/ReadThread/SegmentReader.cpp new file mode 100644 index 00000000000..77806fdef72 --- /dev/null +++ b/dbms/src/Storages/DeltaMerge/ReadThread/SegmentReader.cpp @@ -0,0 +1,233 @@ +// Copyright 2022 PingCAP, Ltd. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. +#include +#include +#include +#include +#include + +#include + +namespace DB::DM +{ +class SegmentReader +{ + inline static const std::string name{"SegmentReader"}; + +public: + SegmentReader(WorkQueue & task_queue_, const std::vector & cpus_) + : task_queue(task_queue_) + , stop(false) + , log(&Poco::Logger::get(name)) + , cpus(cpus_) + { + t = std::thread(&SegmentReader::run, this); + } + + void setStop() + { + stop.store(true, std::memory_order_relaxed); + } + + ~SegmentReader() + { + LOG_FMT_DEBUG(log, "SegmentReader stop begin"); + t.join(); + LOG_FMT_DEBUG(log, "SegmentReader stop end"); + } + + std::thread::id getId() const + { + return t.get_id(); + } + +private: + void setCPUAffinity() + { + if (cpus.empty()) + { + return; + } +#ifdef __linux__ + cpu_set_t cpu_set; + CPU_ZERO(&cpu_set); + for (int i : cpus) + { + CPU_SET(i, &cpu_set); + } + int ret = sched_setaffinity(0, sizeof(cpu_set), &cpu_set); + if (ret != 0) + { + LOG_FMT_ERROR(log, "sched_setaffinity fail: {}", std::strerror(errno)); + throw Exception(fmt::format("sched_setaffinity fail: {}", std::strerror(errno))); + } + LOG_FMT_DEBUG(log, "sched_setaffinity cpus {} succ", cpus); +#endif + } + + bool isStop() + { + return stop.load(std::memory_order_relaxed); + } + + void readSegments() + { + MergedTaskPtr merged_task; + try + { + if (!task_queue.pop(merged_task)) + { + LOG_FMT_INFO(log, "pop fail, stop {}", isStop()); + return; + } + + SCOPE_EXIT({ + if (!merged_task->allStreamsFinished()) + { + SegmentReadTaskScheduler::instance().pushMergedTask(merged_task); + } + }); + + int read_count = 0; + while (!merged_task->allStreamsFinished() && !isStop()) + { + auto c = merged_task->readBlock(); + read_count += c; + if (c <= 0) + { + break; + } + } + if (read_count <= 0) + { + LOG_FMT_DEBUG(log, "pool {} seg_id {} read_count {}", merged_task->getPoolIds(), merged_task->getSegmentId(), read_count); + } + } + catch (DB::Exception & e) + { + LOG_FMT_ERROR(log, "ErrMsg: {} StackTrace {}", e.message(), e.getStackTrace().toString()); + if (merged_task != nullptr) + { + merged_task->setException(e); + } + } + catch (std::exception & e) + { + LOG_FMT_ERROR(log, "ErrMsg: {}", e.what()); + if (merged_task != nullptr) + { + merged_task->setException(DB::Exception(e.what())); + } + } + catch (...) + { + tryLogCurrentException("exception thrown in SegmentReader"); + if (merged_task != nullptr) + { + merged_task->setException(DB::Exception("unknown exception thrown in SegmentReader")); + } + } + } + + void run() + { + setCPUAffinity(); + setThreadName(name.c_str()); + while (!isStop()) + { + readSegments(); + } + } + + WorkQueue & task_queue; + std::atomic stop; + Poco::Logger * log; + std::thread t; + std::vector cpus; +}; + +void SegmentReaderPool::addTask(MergedTaskPtr && task) +{ + if (!task_queue.push(std::forward(task), nullptr)) + { + throw Exception("addTask fail"); + } +} + +SegmentReaderPool::SegmentReaderPool(int thread_count, const std::vector & cpus) + : log(&Poco::Logger::get("SegmentReaderPool")) +{ + LOG_FMT_INFO(log, "Create SegmentReaderPool thread_count {} cpus {} start", thread_count, cpus); + for (int i = 0; i < thread_count; i++) + { + readers.push_back(std::make_unique(task_queue, cpus)); + } + LOG_FMT_INFO(log, "Create SegmentReaderPool thread_count {} cpus {} end", thread_count, cpus); +} + +SegmentReaderPool::~SegmentReaderPool() +{ + for (auto & reader : readers) + { + reader->setStop(); + } + task_queue.finish(); +} + +std::vector SegmentReaderPool::getReaderIds() const +{ + std::vector ids; + for (const auto & r : readers) + { + ids.push_back(r->getId()); + } + return ids; +} + +SegmentReaderPoolManager::SegmentReaderPoolManager() + : log(&Poco::Logger::get("SegmentReaderPoolManager")) +{} + +SegmentReaderPoolManager::~SegmentReaderPoolManager() = default; + +void SegmentReaderPoolManager::init(const ServerInfo & server_info) +{ + auto numa_nodes = getNumaNodes(log); + LOG_FMT_INFO(log, "numa_nodes {} => {}", numa_nodes.size(), numa_nodes); + for (const auto & node : numa_nodes) + { + int thread_count = node.empty() ? server_info.cpu_info.logical_cores : node.size(); + reader_pools.push_back(std::make_unique(thread_count, node)); + auto ids = reader_pools.back()->getReaderIds(); + reader_ids.insert(ids.begin(), ids.end()); + } + LOG_FMT_INFO(log, "readers count {}", reader_ids.size()); +} + +void SegmentReaderPoolManager::addTask(MergedTaskPtr && task) +{ + static std::hash hash_func; + auto idx = hash_func(task->getSegmentId()) % reader_pools.size(); + reader_pools[idx]->addTask(std::move(task)); +} + +// `isSegmentReader` checks whether this thread is a `SegmentReader`. +// Use this function in DMFileBlockInputSteam to check whether enable read thread of this read request, +// Maybe we can pass the argument from DeltaMerge -> SegmentReadTaskPool -> ... -> DMFileBlockInputSteam. +// But this is a long code path and can affect a lot a code. +bool SegmentReaderPoolManager::isSegmentReader() const +{ + return reader_ids.find(std::this_thread::get_id()) != reader_ids.end(); +} +} // namespace DB::DM \ No newline at end of file diff --git a/dbms/src/Storages/DeltaMerge/ReadThread/SegmentReader.h b/dbms/src/Storages/DeltaMerge/ReadThread/SegmentReader.h new file mode 100644 index 00000000000..5408dcb4fd6 --- /dev/null +++ b/dbms/src/Storages/DeltaMerge/ReadThread/SegmentReader.h @@ -0,0 +1,77 @@ +// Copyright 2022 PingCAP, Ltd. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. +#pragma once + +#include +#include +#include +#include + +namespace DB::DM +{ +class MergedTask; +using MergedTaskPtr = std::shared_ptr; + +class SegmentReader; +using SegmentReaderUPtr = std::unique_ptr; + +class SegmentReaderPool +{ +public: + SegmentReaderPool(int thread_count, const std::vector & cpus); + ~SegmentReaderPool(); + SegmentReaderPool(const SegmentReaderPool &) = delete; + SegmentReaderPool & operator=(const SegmentReaderPool &) = delete; + SegmentReaderPool(SegmentReaderPool &&) = delete; + SegmentReaderPool & operator=(SegmentReaderPool &&) = delete; + + void addTask(MergedTaskPtr && task); + std::vector getReaderIds() const; + +private: + void init(int thread_count, const std::vector & cpus); + + WorkQueue task_queue; + std::vector readers; + Poco::Logger * log; +}; + +// SegmentReaderPoolManager is a NUMA-aware singleton that manages several SegmentReaderPool objects. +// The number of SegmentReadPool object is the same as the number of CPU NUMA node. +// Thread number of a SegmentReadPool object is the same as the number of CPU logical core of a CPU NUMA node. +// Function `addTask` dispatches MergedTask to SegmentReadPool by their segment id, so a segment read task +// wouldn't be processed across NUMA nodes. +class SegmentReaderPoolManager +{ +public: + static SegmentReaderPoolManager & instance() + { + static SegmentReaderPoolManager pool_manager; + return pool_manager; + } + void init(const ServerInfo & server_info); + ~SegmentReaderPoolManager(); + DISALLOW_COPY_AND_MOVE(SegmentReaderPoolManager); + + void addTask(MergedTaskPtr && task); + bool isSegmentReader() const; + +private: + SegmentReaderPoolManager(); + std::vector> reader_pools; + std::unordered_set reader_ids; + Poco::Logger * log; +}; + +} // namespace DB::DM \ No newline at end of file diff --git a/dbms/src/Storages/DeltaMerge/ReadThread/UnorderedInputStream.h b/dbms/src/Storages/DeltaMerge/ReadThread/UnorderedInputStream.h new file mode 100644 index 00000000000..5bdbc7ed59d --- /dev/null +++ b/dbms/src/Storages/DeltaMerge/ReadThread/UnorderedInputStream.h @@ -0,0 +1,131 @@ +// Copyright 2022 PingCAP, Ltd. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +#pragma once + +#include +#include +#include + +namespace DB::FailPoints +{ +extern const char pause_when_reading_from_dt_stream[]; +} + +namespace DB::DM +{ +class UnorderedInputStream : public IProfilingBlockInputStream +{ + static constexpr auto NAME = "UnorderedInputStream"; + +public: + UnorderedInputStream( + const SegmentReadTaskPoolPtr & task_pool_, + const ColumnDefines & columns_to_read_, + const int extra_table_id_index, + const TableID physical_table_id, + const String & req_id) + : task_pool(task_pool_) + , header(toEmptyBlock(columns_to_read_)) + , extra_table_id_index(extra_table_id_index) + , physical_table_id(physical_table_id) + , log(Logger::get(NAME, req_id)) + , ref_no(0) + { + if (extra_table_id_index != InvalidColumnID) + { + ColumnDefine extra_table_id_col_define = getExtraTableIDColumnDefine(); + ColumnWithTypeAndName col{extra_table_id_col_define.type->createColumn(), extra_table_id_col_define.type, extra_table_id_col_define.name, extra_table_id_col_define.id, extra_table_id_col_define.default_value}; + header.insert(extra_table_id_index, col); + } + ref_no = task_pool->increaseUnorderedInputStreamRefCount(); + LOG_FMT_DEBUG(log, "pool {} ref {} created", task_pool->poolId(), ref_no); + } + + ~UnorderedInputStream() + { + task_pool->decreaseUnorderedInputStreamRefCount(); + LOG_FMT_DEBUG(log, "pool {} ref {} destroy", task_pool->poolId(), ref_no); + } + + String getName() const override { return NAME; } + + Block getHeader() const override { return header; } + +protected: + Block readImpl() override + { + FilterPtr filter_ignored; + return readImpl(filter_ignored, false); + } + + // Currently, res_fiter and return_filter is unused. + Block readImpl(FilterPtr & /*res_filter*/, bool /*return_filter*/) override + { + if (done) + { + return {}; + } + while (true) + { + FAIL_POINT_PAUSE(FailPoints::pause_when_reading_from_dt_stream); + + Block res; + task_pool->popBlock(res); + if (res) + { + if (extra_table_id_index != InvalidColumnID) + { + ColumnDefine extra_table_id_col_define = getExtraTableIDColumnDefine(); + ColumnWithTypeAndName col{{}, extra_table_id_col_define.type, extra_table_id_col_define.name, extra_table_id_col_define.id}; + size_t row_number = res.rows(); + auto col_data = col.type->createColumnConst(row_number, Field(physical_table_id)); + col.column = std::move(col_data); + res.insert(extra_table_id_index, std::move(col)); + } + if (!res.rows()) + { + continue; + } + else + { + total_rows += res.rows(); + return res; + } + } + else + { + done = true; + return {}; + } + } + } + + void readSuffixImpl() override + { + LOG_FMT_DEBUG(log, "pool {} ref {} finish read {} rows from storage", task_pool->poolId(), ref_no, total_rows); + } + +private: + SegmentReadTaskPoolPtr task_pool; + Block header; + // position of the ExtraPhysTblID column in column_names parameter in the StorageDeltaMerge::read function. + const int extra_table_id_index; + bool done = false; + TableID physical_table_id; + LoggerPtr log; + int64_t ref_no; + size_t total_rows = 0; +}; +} // namespace DB::DM \ No newline at end of file diff --git a/dbms/src/Storages/DeltaMerge/ReadThread/WorkQueue.h b/dbms/src/Storages/DeltaMerge/ReadThread/WorkQueue.h new file mode 100644 index 00000000000..7726617ebef --- /dev/null +++ b/dbms/src/Storages/DeltaMerge/ReadThread/WorkQueue.h @@ -0,0 +1,174 @@ +// Copyright 2022 PingCAP, Ltd. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. +#pragma once + +#include + +#include +#include +#include + +namespace DB::DM +{ +template +class WorkQueue +{ + // Protects all member variable access + std::mutex mutex_; + std::condition_variable readerCv_; + std::condition_variable writerCv_; + std::condition_variable finishCv_; + std::queue queue_; + bool done_; + std::size_t maxSize_; + + std::size_t peak_queue_size; + int64_t pop_times; + int64_t pop_empty_times; + // Must have lock to call this function + bool full() const + { + if (maxSize_ == 0) + { + return false; + } + return queue_.size() >= maxSize_; + } + +public: + /** + * Constructs an empty work queue with an optional max size. + * If `maxSize == 0` the queue size is unbounded. + * + * @param maxSize The maximum allowed size of the work queue. + */ + WorkQueue(std::size_t maxSize = 0) + : done_(false) + , maxSize_(maxSize) + , peak_queue_size(0) + , pop_times(0) + , pop_empty_times(0) + {} + /** + * Push an item onto the work queue. Notify a single thread that work is + * available. If `finish()` has been called, do nothing and return false. + * If `push()` returns false, then `item` has not been copied from. + * + * @param item Item to push onto the queue. + * @returns True upon success, false if `finish()` has been called. An + * item was pushed iff `push()` returns true. + */ + template + bool push(U && item, size_t * size) + { + { + std::unique_lock lock(mutex_); + while (full() && !done_) + { + writerCv_.wait(lock); + } + if (done_) + { + return false; + } + queue_.push(std::forward(item)); + peak_queue_size = std::max(queue_.size(), peak_queue_size); + if (size != nullptr) + { + *size = queue_.size(); + } + } + readerCv_.notify_one(); + return true; + } + /** + * Attempts to pop an item off the work queue. It will block until data is + * available or `finish()` has been called. + * + * @param[out] item If `pop` returns `true`, it contains the popped item. + * If `pop` returns `false`, it is unmodified. + * @returns True upon success. False if the queue is empty and + * `finish()` has been called. + */ + bool pop(T & item) + { + { + std::unique_lock lock(mutex_); + pop_times++; + while (queue_.empty() && !done_) + { + pop_empty_times++; + readerCv_.wait(lock); + } + if (queue_.empty()) + { + assert(done_); + return false; + } + item = std::move(queue_.front()); + queue_.pop(); + } + writerCv_.notify_one(); + return true; + } + /** + * Sets the maximum queue size. If `maxSize == 0` then it is unbounded. + * + * @param maxSize The new maximum queue size. + */ + void setMaxSize(std::size_t maxSize) + { + { + std::lock_guard lock(mutex_); + maxSize_ = maxSize; + } + writerCv_.notify_all(); + } + /** + * Promise that `push()` won't be called again, so once the queue is empty + * there will never any more work. + */ + void finish() + { + { + std::lock_guard lock(mutex_); + assert(!done_); + done_ = true; + } + readerCv_.notify_all(); + writerCv_.notify_all(); + finishCv_.notify_all(); + } + /// Blocks until `finish()` has been called (but the queue may not be empty). + void waitUntilFinished() + { + std::unique_lock lock(mutex_); + while (!done_) + { + finishCv_.wait(lock); + } + } + + size_t size() + { + std::lock_guard lock(mutex_); + return queue_.size(); + } + + std::tuple getStat() const + { + return std::tuple{pop_times, pop_empty_times, peak_queue_size}; + } +}; +} // namespace DB::DM \ No newline at end of file diff --git a/dbms/src/Storages/DeltaMerge/SegmentReadTaskPool.cpp b/dbms/src/Storages/DeltaMerge/SegmentReadTaskPool.cpp index 0e61a01b01a..80ff8487c21 100644 --- a/dbms/src/Storages/DeltaMerge/SegmentReadTaskPool.cpp +++ b/dbms/src/Storages/DeltaMerge/SegmentReadTaskPool.cpp @@ -61,7 +61,7 @@ SegmentReadTasks SegmentReadTask::trySplitReadTasks(const SegmentReadTasks & tas return a->ranges.size() < b->ranges.size(); }; std::priority_queue, decltype(cmp)> largest_ranges_first(cmp); - for (auto & task : tasks) + for (const auto & task : tasks) largest_ranges_first.push(task); // Split the top task. @@ -95,4 +95,171 @@ SegmentReadTasks SegmentReadTask::trySplitReadTasks(const SegmentReadTasks & tas return result_tasks; } +BlockInputStreamPtr SegmentReadTaskPool::buildInputStream(SegmentReadTaskPtr & t) +{ + MemoryTrackerSetter setter(true, mem_tracker); + auto seg = t->segment; + BlockInputStreamPtr stream; + if (is_raw) + { + stream = seg->getInputStreamRaw(*dm_context, columns_to_read, t->read_snapshot, t->ranges, filter, do_range_filter_for_raw); + } + else + { + auto block_size = std::max(expected_block_size, static_cast(dm_context->db_context.getSettingsRef().dt_segment_stable_pack_rows)); + stream = seg->getInputStream(*dm_context, columns_to_read, t->read_snapshot, t->ranges, filter, max_version, block_size); + } + LOG_FMT_DEBUG(log, "getInputStream pool {} seg_id {} succ", pool_id, seg->segmentId()); + return stream; +} + +void SegmentReadTaskPool::finishSegment(const SegmentPtr & seg) +{ + after_segment_read(dm_context, seg); + bool pool_finished = false; + { + std::lock_guard lock(mutex); + active_segment_ids.erase(seg->segmentId()); + pool_finished = active_segment_ids.empty() && tasks.empty(); + } + LOG_FMT_DEBUG(log, "finishSegment pool {} seg_id {} pool_finished {}", pool_id, seg->segmentId(), pool_finished); + if (pool_finished) + { + q.finish(); + } +} + +SegmentReadTaskPtr SegmentReadTaskPool::getTask(uint64_t seg_id) +{ + std::lock_guard lock(mutex); + // TODO(jinhelin): use unordered_map + auto itr = std::find_if(tasks.begin(), tasks.end(), [seg_id](const SegmentReadTaskPtr & task) { return task->segment->segmentId() == seg_id; }); + if (itr == tasks.end()) + { + throw Exception(fmt::format("pool {} seg_id {} not found", pool_id, seg_id)); + } + auto t = *(itr); + tasks.erase(itr); + active_segment_ids.insert(seg_id); + return t; +} + +// Choose a segment to read. +// Returns . +std::unordered_map>::const_iterator SegmentReadTaskPool::scheduleSegment(const std::unordered_map> & segments, uint64_t expected_merge_count) +{ + auto target = segments.end(); + std::lock_guard lock(mutex); + if (getFreeActiveSegmentCountUnlock() <= 0) + { + return target; + } + for (const auto & task : tasks) + { + auto itr = segments.find(task->segment->segmentId()); + if (itr == segments.end()) + { + throw DB::Exception(fmt::format("seg_id {} not found from merging segments", task->segment->segmentId())); + } + if (std::find(itr->second.begin(), itr->second.end(), poolId()) == itr->second.end()) + { + throw DB::Exception(fmt::format("pool {} not found from merging segment {}=>{}", poolId(), itr->first, itr->second)); + } + if (target == segments.end() || itr->second.size() > target->second.size()) + { + target = itr; + } + if (target->second.size() >= expected_merge_count) + { + break; + } + } + return target; +} + +bool SegmentReadTaskPool::readOneBlock(BlockInputStreamPtr & stream, const SegmentPtr & seg) +{ + MemoryTrackerSetter setter(true, mem_tracker); + auto block = stream->read(); + if (block) + { + pushBlock(std::move(block)); + return true; + } + else + { + finishSegment(seg); + return false; + } +} + +void SegmentReadTaskPool::popBlock(Block & block) +{ + q.pop(block); + blk_stat.pop(block); + global_blk_stat.pop(block); + if (exceptionHappened()) + { + throw exception; + } +} + +void SegmentReadTaskPool::pushBlock(Block && block) +{ + blk_stat.push(block); + global_blk_stat.push(block); + q.push(std::move(block), nullptr); +} + +int64_t SegmentReadTaskPool::increaseUnorderedInputStreamRefCount() +{ + return unordered_input_stream_ref_count.fetch_add(1, std::memory_order_relaxed); +} +int64_t SegmentReadTaskPool::decreaseUnorderedInputStreamRefCount() +{ + return unordered_input_stream_ref_count.fetch_sub(1, std::memory_order_relaxed); +} + +int64_t SegmentReadTaskPool::getFreeBlockSlots() const +{ + double block_slots_scale = dm_context->db_context.getSettingsRef().dt_block_slots_scale; + auto block_slots = static_cast(std::ceil(unordered_input_stream_ref_count.load(std::memory_order_relaxed) * block_slots_scale)); + if (block_slots < 3) + { + block_slots = 3; + } + return block_slots - blk_stat.pendingCount(); +} + +int64_t SegmentReadTaskPool::getFreeActiveSegmentCountUnlock() +{ + double active_segments_scale = dm_context->db_context.getSettingsRef().dt_active_segments_scale; + auto active_segment_limit = static_cast(std::ceil(unordered_input_stream_ref_count.load(std::memory_order_relaxed) * active_segments_scale)); + if (active_segment_limit < 2) + { + active_segment_limit = 2; + } + return active_segment_limit - static_cast(active_segment_ids.size()); +} + +bool SegmentReadTaskPool::exceptionHappened() const +{ + return exception_happened.load(std::memory_order_relaxed); +} + +bool SegmentReadTaskPool::valid() const +{ + return !exceptionHappened() && unordered_input_stream_ref_count.load(std::memory_order_relaxed) > 0; +} +void SegmentReadTaskPool::setException(const DB::Exception & e) +{ + std::lock_guard lock(mutex); + if (!exceptionHappened()) + { + exception = e; + exception_happened.store(true, std::memory_order_relaxed); + q.finish(); + } +} + } // namespace DB::DM diff --git a/dbms/src/Storages/DeltaMerge/SegmentReadTaskPool.h b/dbms/src/Storages/DeltaMerge/SegmentReadTaskPool.h index dc14716af80..e87099a22b4 100644 --- a/dbms/src/Storages/DeltaMerge/SegmentReadTaskPool.h +++ b/dbms/src/Storages/DeltaMerge/SegmentReadTaskPool.h @@ -13,23 +13,22 @@ // limitations under the License. #pragma once - +#include +#include +#include +#include #include -#include - namespace DB { namespace DM { -struct DMContext; struct SegmentReadTask; class Segment; using SegmentPtr = std::shared_ptr; struct SegmentSnapshot; using SegmentSnapshotPtr = std::shared_ptr; -using DMContextPtr = std::shared_ptr; using SegmentReadTaskPtr = std::shared_ptr; using SegmentReadTasks = std::list; using AfterSegmentRead = std::function; @@ -57,13 +56,113 @@ struct SegmentReadTask static SegmentReadTasks trySplitReadTasks(const SegmentReadTasks & tasks, size_t expected_size); }; +class BlockStat +{ +public: + BlockStat() + : pending_count(0) + , pending_bytes(0) + , total_count(0) + , total_bytes(0) + {} + + void push(const Block & blk) + { + pending_count.fetch_add(1, std::memory_order_relaxed); + total_count.fetch_add(1, std::memory_order_relaxed); + + auto b = blk.bytes(); + pending_bytes.fetch_add(b, std::memory_order_relaxed); + total_bytes.fetch_add(b, std::memory_order_relaxed); + } + + void pop(const Block & blk) + { + if (likely(blk)) + { + pending_count.fetch_sub(1, std::memory_order_relaxed); + pending_bytes.fetch_sub(blk.bytes(), std::memory_order_relaxed); + } + } + + int64_t pendingCount() const + { + return pending_count.load(std::memory_order_relaxed); + } + + int64_t pendingBytes() const + { + return pending_bytes.load(std::memory_order_relaxed); + } + + int64_t totalCount() const + { + return total_count.load(std::memory_order_relaxed); + } + int64_t totalBytes() const + { + return total_bytes.load(std::memory_order_relaxed); + } + +private: + std::atomic pending_count; + std::atomic pending_bytes; + std::atomic total_count; + std::atomic total_bytes; +}; + class SegmentReadTaskPool : private boost::noncopyable { public: - explicit SegmentReadTaskPool(SegmentReadTasks && tasks_) - : tasks(std::move(tasks_)) + explicit SegmentReadTaskPool( + int64_t table_id_, + const DMContextPtr & dm_context_, + const ColumnDefines & columns_to_read_, + const RSOperatorPtr & filter_, + uint64_t max_version_, + size_t expected_block_size_, + bool is_raw_, + bool do_range_filter_for_raw_, + SegmentReadTasks && tasks_, + AfterSegmentRead after_segment_read_) + : pool_id(nextPoolId()) + , table_id(table_id_) + , dm_context(dm_context_) + , columns_to_read(columns_to_read_) + , filter(filter_) + , max_version(max_version_) + , expected_block_size(expected_block_size_) + , is_raw(is_raw_) + , do_range_filter_for_raw(do_range_filter_for_raw_) + , tasks(std::move(tasks_)) + , after_segment_read(after_segment_read_) + , log(&Poco::Logger::get("SegmentReadTaskPool")) + , unordered_input_stream_ref_count(0) + , exception_happened(false) + , mem_tracker(current_memory_tracker) {} + ~SegmentReadTaskPool() + { + auto [pop_times, pop_empty_times, max_queue_size] = q.getStat(); + auto pop_empty_ratio = pop_times > 0 ? pop_empty_times * 1.0 / pop_times : 0.0; + auto total_count = blk_stat.totalCount(); + auto total_bytes = blk_stat.totalBytes(); + auto blk_avg_bytes = total_count > 0 ? total_bytes / total_count : 0; + auto approximate_max_pending_block_bytes = blk_avg_bytes * max_queue_size; + LOG_FMT_DEBUG(log, "pool {} table {} pop {} pop_empty {} pop_empty_ratio {} max_queue_size {} blk_avg_bytes {} approximate_max_pending_block_bytes {} MB total_count {} total_bytes {} MB", // + pool_id, + table_id, + pop_times, + pop_empty_times, + pop_empty_ratio, + max_queue_size, + blk_avg_bytes, + approximate_max_pending_block_bytes / 1024.0 / 1024.0, + total_count, + total_bytes / 1024.0 / 1024.0); + } + SegmentReadTaskPtr nextTask() { std::lock_guard lock(mutex); @@ -74,13 +173,68 @@ class SegmentReadTaskPool : private boost::noncopyable return task; } + uint64_t poolId() const { return pool_id; } + + int64_t tableId() const { return table_id; } + + const SegmentReadTasks & getTasks() const { return tasks; } + + BlockInputStreamPtr buildInputStream(SegmentReadTaskPtr & t); + + bool readOneBlock(BlockInputStreamPtr & stream, const SegmentPtr & seg); + void popBlock(Block & block); + + std::unordered_map>::const_iterator scheduleSegment( + const std::unordered_map> & segments, + uint64_t expected_merge_count); + + int64_t increaseUnorderedInputStreamRefCount(); + int64_t decreaseUnorderedInputStreamRefCount(); + int64_t getFreeBlockSlots() const; + bool valid() const; + void setException(const DB::Exception & e); + SegmentReadTaskPtr getTask(uint64_t seg_id); + private: + int64_t getFreeActiveSegmentCountUnlock(); + bool exceptionHappened() const; + void finishSegment(const SegmentPtr & seg); + void pushBlock(Block && block); + + const uint64_t pool_id; + const int64_t table_id; + DMContextPtr dm_context; + ColumnDefines columns_to_read; + RSOperatorPtr filter; + const uint64_t max_version; + const size_t expected_block_size; + const bool is_raw; + const bool do_range_filter_for_raw; SegmentReadTasks tasks; - + AfterSegmentRead after_segment_read; std::mutex mutex; + std::unordered_set active_segment_ids; + WorkQueue q; + BlockStat blk_stat; + Poco::Logger * log; + + std::atomic unordered_input_stream_ref_count; + + std::atomic exception_happened; + DB::Exception exception; + + MemoryTracker * mem_tracker; + + inline static std::atomic pool_id_gen{1}; + inline static BlockStat global_blk_stat; + static uint64_t nextPoolId() + { + return pool_id_gen.fetch_add(1, std::memory_order_relaxed); + } }; using SegmentReadTaskPoolPtr = std::shared_ptr; +using SegmentReadTaskPools = std::vector; } // namespace DM } // namespace DB diff --git a/dbms/src/Storages/DeltaMerge/StoragePool.cpp b/dbms/src/Storages/DeltaMerge/StoragePool.cpp index 2791a74e9e3..523856578a0 100644 --- a/dbms/src/Storages/DeltaMerge/StoragePool.cpp +++ b/dbms/src/Storages/DeltaMerge/StoragePool.cpp @@ -163,10 +163,11 @@ bool GlobalStoragePool::gc(const Settings & settings, bool immediately, const Se return done_anything; } -StoragePool::StoragePool(Context & global_ctx, NamespaceId ns_id_, StoragePathPool & storage_path_pool, const String & name) +StoragePool::StoragePool(Context & global_ctx, NamespaceId ns_id_, StoragePathPool & storage_path_pool_, const String & name) : logger(Logger::get("StoragePool", !name.empty() ? name : DB::toString(ns_id_))) , run_mode(global_ctx.getPageStorageRunMode()) , ns_id(ns_id_) + , storage_path_pool(storage_path_pool_) , global_context(global_ctx) , storage_pool_metrics(CurrentMetrics::StoragePoolV3Only, 0) { @@ -219,18 +220,42 @@ StoragePool::StoragePool(Context & global_ctx, NamespaceId ns_id_, StoragePathPo data_storage_v3 = global_storage_pool->data_storage; meta_storage_v3 = global_storage_pool->meta_storage; - log_storage_v2 = PageStorage::create(name + ".log", - storage_path_pool.getPSDiskDelegatorMulti("log"), - extractConfig(global_context.getSettingsRef(), StorageType::Log), - global_context.getFileProvider()); - data_storage_v2 = PageStorage::create(name + ".data", - storage_path_pool.getPSDiskDelegatorMulti("data"), - extractConfig(global_context.getSettingsRef(), StorageType::Data), - global_ctx.getFileProvider()); - meta_storage_v2 = PageStorage::create(name + ".meta", - storage_path_pool.getPSDiskDelegatorMulti("meta"), - extractConfig(global_context.getSettingsRef(), StorageType::Meta), - global_ctx.getFileProvider()); + if (storage_path_pool.isPSV2Deleted()) + { + LOG_FMT_INFO(logger, "PageStorage V2 is already mark deleted. Current pagestorage change from {} to {} [ns_id={}]", // + static_cast(PageStorageRunMode::MIX_MODE), // + static_cast(PageStorageRunMode::ONLY_V3), // + ns_id); + log_storage_v2 = nullptr; + data_storage_v2 = nullptr; + meta_storage_v2 = nullptr; + run_mode = PageStorageRunMode::ONLY_V3; + storage_path_pool.clearPSV2ObsoleteData(); + } + else + { + // Although there is no more write to ps v2 in mixed mode, the ps instances will keep running if there is some data in log storage when restart, + // so we keep its original config here. + // And we rely on the mechanism that writing file will be rotated if no valid pages in non writing files to reduce the disk space usage of these ps instances. + log_storage_v2 = PageStorage::create(name + ".log", + storage_path_pool.getPSDiskDelegatorMulti("log"), + extractConfig(global_context.getSettingsRef(), StorageType::Log), + global_context.getFileProvider(), + /* use_v3 */ false, + /* no_more_write_to_v2 */ true); + data_storage_v2 = PageStorage::create(name + ".data", + storage_path_pool.getPSDiskDelegatorMulti("data"), + extractConfig(global_context.getSettingsRef(), StorageType::Data), + global_ctx.getFileProvider(), + /* use_v3 */ false, + /* no_more_write_to_v2 */ true); + meta_storage_v2 = PageStorage::create(name + ".meta", + storage_path_pool.getPSDiskDelegatorMulti("meta"), + extractConfig(global_context.getSettingsRef(), StorageType::Meta), + global_ctx.getFileProvider(), + /* use_v3 */ false, + /* no_more_write_to_v2 */ true); + } log_storage_reader = std::make_shared(run_mode, ns_id, log_storage_v2, log_storage_v3, nullptr); data_storage_reader = std::make_shared(run_mode, ns_id, data_storage_v2, data_storage_v3, nullptr); @@ -272,7 +297,7 @@ void StoragePool::forceTransformMetaV2toV3() const auto & page_transform_entry = meta_transform_storage_reader->getPageEntry(page_transform.page_id); if (!page_transform_entry.field_offsets.empty()) { - throw Exception(fmt::format("Can't transfrom meta from V2 to V3, [page_id={}] {}", // + throw Exception(fmt::format("Can't transform meta from V2 to V3, [page_id={}] {}", // page_transform.page_id, page_transform_entry.toDebugString()), ErrorCodes::LOGICAL_ERROR); @@ -431,18 +456,23 @@ PageStorageRunMode StoragePool::restore() } else { - LOG_FMT_INFO(logger, "Current pool.data translate already done before restored [ns_id={}] ", ns_id); + LOG_FMT_INFO(logger, "Current pool.data translate already done before restored [ns_id={}]", ns_id); } // Check number of valid pages in v2 // If V2 already have no any data in disk, Then change run_mode to ONLY_V3 if (log_storage_v2->getNumberOfPages() == 0 && data_storage_v2->getNumberOfPages() == 0 && meta_storage_v2->getNumberOfPages() == 0) { - // TODO: Need drop V2 in disk and check. - LOG_FMT_INFO(logger, "Current pagestorage change from {} to {}", // + LOG_FMT_INFO(logger, "Current pagestorage change from {} to {} [ns_id={}]", // static_cast(PageStorageRunMode::MIX_MODE), - static_cast(PageStorageRunMode::ONLY_V3)); - + static_cast(PageStorageRunMode::ONLY_V3), + ns_id); + if (storage_path_pool.createPSV2DeleteMarkFile()) + { + log_storage_v2->drop(); + data_storage_v2->drop(); + meta_storage_v2->drop(); + } log_storage_v2 = nullptr; data_storage_v2 = nullptr; meta_storage_v2 = nullptr; diff --git a/dbms/src/Storages/DeltaMerge/StoragePool.h b/dbms/src/Storages/DeltaMerge/StoragePool.h index 77684ea46cb..2ab07a53ad9 100644 --- a/dbms/src/Storages/DeltaMerge/StoragePool.h +++ b/dbms/src/Storages/DeltaMerge/StoragePool.h @@ -79,7 +79,7 @@ class StoragePool : private boost::noncopyable using Timepoint = Clock::time_point; using Seconds = std::chrono::seconds; - StoragePool(Context & global_ctx, NamespaceId ns_id_, StoragePathPool & path_pool, const String & name = ""); + StoragePool(Context & global_ctx, NamespaceId ns_id_, StoragePathPool & storage_path_pool_, const String & name = ""); PageStorageRunMode restore(); @@ -181,6 +181,8 @@ class StoragePool : private boost::noncopyable // whether the three storage instance is owned by this StoragePool const NamespaceId ns_id; + StoragePathPool & storage_path_pool; + PageStoragePtr log_storage_v2; PageStoragePtr data_storage_v2; PageStoragePtr meta_storage_v2; diff --git a/dbms/src/Storages/DeltaMerge/tests/CMakeLists.txt b/dbms/src/Storages/DeltaMerge/tests/CMakeLists.txt index 14ecdfc5e79..44ca6ccb367 100644 --- a/dbms/src/Storages/DeltaMerge/tests/CMakeLists.txt +++ b/dbms/src/Storages/DeltaMerge/tests/CMakeLists.txt @@ -63,6 +63,9 @@ target_link_libraries(dm_test_delta_index_manager dbms gtest_main clickhouse_fun add_executable(dm_test_delta_value_space EXCLUDE_FROM_ALL gtest_dm_delta_value_space.cpp) target_link_libraries(dm_test_delta_value_space dbms gtest_main clickhouse_functions) +add_executable(dm_test_column_file EXCLUDE_FROM_ALL gtest_dm_column_file.cpp) +target_link_libraries(dm_test_column_file dbms gtest_main clickhouse_functions) + add_subdirectory (bank EXCLUDE_FROM_ALL) add_subdirectory (stress EXCLUDE_FROM_ALL) diff --git a/dbms/src/Storages/DeltaMerge/tests/gtest_circular_scan_list.cpp b/dbms/src/Storages/DeltaMerge/tests/gtest_circular_scan_list.cpp new file mode 100644 index 00000000000..ab44b3b2ec2 --- /dev/null +++ b/dbms/src/Storages/DeltaMerge/tests/gtest_circular_scan_list.cpp @@ -0,0 +1,136 @@ +// Copyright 2022 PingCAP, Ltd. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. +#include +#include + +#include + +namespace DB::DM::tests +{ +class Node +{ +public: + explicit Node(uint64_t id_) + : id(id_) + , table_id(1) + , v(true) + {} + + bool valid() const + { + return v; + } + uint64_t poolId() const + { + return id; + } + int64_t tableId() const + { + return table_id; + } + void setInvalid() + { + v = false; + } + +private: + uint64_t id; + int64_t table_id; + bool v; +}; + +TEST(CircularScanListTest, Normal) +{ + CircularScanList lst; + + { + ASSERT_EQ(lst.next(), nullptr); + auto [valid, invalid] = lst.count(0); + ASSERT_EQ(valid, 0); + ASSERT_EQ(invalid, 0); + ASSERT_EQ(lst.get(1), nullptr); + } + + for (uint64_t i = 0; i < 10; i++) + { + lst.add(std::make_shared(i)); + } + + { + auto [valid, invalid] = lst.count(0); + ASSERT_EQ(valid, 10); + ASSERT_EQ(invalid, 0); + } + + for (uint64_t i = 0; i < 20; i++) + { + auto sp = lst.next(); + ASSERT_EQ(sp->poolId(), i % 10); + } + + lst.get(1)->setInvalid(); + lst.get(3)->setInvalid(); + lst.get(5)->setInvalid(); + + { + auto [valid, invalid] = lst.count(0); + ASSERT_EQ(valid, 7); + ASSERT_EQ(invalid, 3); + } + + const std::vector valid_ids = {0, 2, 4, 6, 7, 8, 9}; + for (uint64_t i = 0; i < 20; i++) + { + auto sp = lst.next(); + ASSERT_EQ(sp->poolId(), valid_ids[i % valid_ids.size()]); + } + + { + auto [valid, invalid] = lst.count(0); + ASSERT_EQ(valid, 7); + ASSERT_EQ(invalid, 0); + } + + for (uint64_t id : valid_ids) + { + lst.get(id)->setInvalid(); + } + + { + auto [valid, invalid] = lst.count(0); + ASSERT_EQ(valid, 0); + ASSERT_EQ(invalid, 7); + } + + ASSERT_EQ(lst.next(), nullptr); +} + +TEST(CircularScanListTest, valid) +{ + CircularScanList l; + l.add(std::make_shared(1)); + + ASSERT_EQ(l.next()->poolId(), 1); + ASSERT_EQ(l.next()->poolId(), 1); + + l.next()->setInvalid(); + + ASSERT_EQ(l.next(), nullptr); + ASSERT_EQ(l.next(), nullptr); + l.add(std::make_shared(2)); + + ASSERT_EQ(l.next()->poolId(), 2); + ASSERT_EQ(l.next()->poolId(), 2); +} +} // namespace DB::DM::tests \ No newline at end of file diff --git a/dbms/src/Storages/DeltaMerge/tests/gtest_column_sharing_cache.cpp b/dbms/src/Storages/DeltaMerge/tests/gtest_column_sharing_cache.cpp new file mode 100644 index 00000000000..82ac6bdecf2 --- /dev/null +++ b/dbms/src/Storages/DeltaMerge/tests/gtest_column_sharing_cache.cpp @@ -0,0 +1,122 @@ +// Copyright 2022 PingCAP, Ltd. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. +#include +#include +#include +#include + +#include "gtest/gtest.h" + +namespace DB::DM::tests +{ + +constexpr int TEST_PACK_ROWS = 1024; +DataTypePtr TEST_DATA_TYPE = std::make_shared(); + +ColumnPtr createColumn(int packs) +{ + int rows = packs * TEST_PACK_ROWS; + + auto mut_col = TEST_DATA_TYPE->createColumn(); + for (int i = 0; i < rows; i++) + { + Field f = std::to_string(packs * 100000 + i); + mut_col->insert(f); + } + return std::move(mut_col); +} + +void compareColumn(ColumnPtr & col1, ColumnPtr & col2, int rows) +{ + for (int i = 0; i < rows; i++) + { + ASSERT_EQ((*col1)[i].toString(), (*col2)[i].toString()); + } +} + +TEST(ColumnSharingCacheTest, AddAndGet) +{ + ColumnSharingCache cache; + + auto col = createColumn(8); + cache.add(1, 8, col); + + ColumnPtr col1; + auto st = cache.get(1, 8, 8 * TEST_PACK_ROWS, col1, TEST_DATA_TYPE); + ASSERT_EQ(st, ColumnCacheStatus::GET_HIT); + ASSERT_EQ(col1->size(), 8 * TEST_PACK_ROWS); + compareColumn(col1, col, col1->size()); + + ColumnPtr col2; + st = cache.get(1, 7, 7 * TEST_PACK_ROWS, col2, TEST_DATA_TYPE); + ASSERT_EQ(st, ColumnCacheStatus::GET_COPY); + ASSERT_EQ(col2->size(), 7 * TEST_PACK_ROWS); + compareColumn(col2, col, col2->size()); + + ColumnPtr col3; + st = cache.get(1, 9, 9 * TEST_PACK_ROWS, col3, TEST_DATA_TYPE); + ASSERT_EQ(st, ColumnCacheStatus::GET_PART); + ASSERT_EQ(col3, nullptr); + + ColumnPtr col4; + st = cache.get(2, 8, 8 * TEST_PACK_ROWS, col4, TEST_DATA_TYPE); + ASSERT_EQ(st, ColumnCacheStatus::GET_MISS); + ASSERT_EQ(col4, nullptr); + + auto col5 = createColumn(7); + cache.add(1, 7, col5); + ColumnPtr col6; + st = cache.get(1, 8, 8 * TEST_PACK_ROWS, col6, TEST_DATA_TYPE); + ASSERT_EQ(st, ColumnCacheStatus::GET_HIT); + ASSERT_EQ(col6->size(), 8 * TEST_PACK_ROWS); + compareColumn(col6, col, col6->size()); + + auto col7 = createColumn(9); + cache.add(1, 9, col7); + ColumnPtr col8; + st = cache.get(1, 8, 8 * TEST_PACK_ROWS, col8, TEST_DATA_TYPE); + ASSERT_EQ(st, ColumnCacheStatus::GET_COPY); + ASSERT_EQ(col8->size(), 8 * TEST_PACK_ROWS); + compareColumn(col8, col7, col8->size()); +} + +TEST(ColumnSharingCacheTest, Del) +{ + ColumnSharingCache cache; + + auto col1 = createColumn(8); + cache.add(1, 8, col1); + + auto col2 = createColumn(8); + cache.add(9, 8, col2); + + auto col3 = createColumn(8); + cache.add(17, 8, col3); + + cache.del(10); + + ColumnPtr col4; + auto st = cache.get(9, 8, 8 * TEST_PACK_ROWS, col4, TEST_DATA_TYPE); + ASSERT_EQ(st, ColumnCacheStatus::GET_HIT); + ASSERT_EQ(col4->size(), 8 * TEST_PACK_ROWS); + compareColumn(col4, col2, col4->size()); + + ColumnPtr col5; + st = cache.get(17, 6, 6 * TEST_PACK_ROWS, col5, TEST_DATA_TYPE); + ASSERT_EQ(st, ColumnCacheStatus::GET_COPY); + ASSERT_EQ(col5->size(), 6 * TEST_PACK_ROWS); + compareColumn(col5, col2, col5->size()); +} + +} // namespace DB::DM::tests \ No newline at end of file diff --git a/dbms/src/Storages/DeltaMerge/tests/gtest_dm_column_file.cpp b/dbms/src/Storages/DeltaMerge/tests/gtest_dm_column_file.cpp new file mode 100644 index 00000000000..e8744f6fe22 --- /dev/null +++ b/dbms/src/Storages/DeltaMerge/tests/gtest_dm_column_file.cpp @@ -0,0 +1,167 @@ +// Copyright 2022 PingCAP, Ltd. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include + +#include + +namespace DB +{ +namespace DM +{ +namespace tests +{ +class ColumnFileTest + : public DB::base::TiFlashStorageTestBasic +{ +public: + ColumnFileTest() = default; + + static void SetUpTestCase() {} + + void SetUp() override + { + TiFlashStorageTestBasic::SetUp(); + + parent_path = TiFlashStorageTestBasic::getTemporaryPath(); + path_pool = std::make_unique(db_context->getPathPool().withTable("test", "DMFile_Test", false)); + storage_pool = std::make_unique(*db_context, /*ns_id*/ 100, *path_pool, "test.t1"); + column_cache = std::make_shared(); + dm_context = std::make_unique( // + *db_context, + *path_pool, + *storage_pool, + /*hash_salt*/ 0, + 0, + settings.not_compress_columns, + false, + 1, + db_context->getSettingsRef()); + } + + DMContext & dmContext() { return *dm_context; } + + Context & dbContext() { return *db_context; } + +private: + std::unique_ptr dm_context; + /// all these var live as ref in dm_context + std::unique_ptr path_pool; + std::unique_ptr storage_pool; + DeltaMergeStore::Settings settings; + +protected: + String parent_path; + ColumnCachePtr column_cache; +}; + +TEST_F(ColumnFileTest, ColumnFileBigRead) +try +{ + auto table_columns = DMTestEnv::getDefaultColumns(); + auto dm_file = DMFile::create(1, parent_path, false, std::make_optional()); + const size_t num_rows_write_per_batch = 8192; + const size_t batch_num = 3; + const UInt64 tso_value = 100; + { + auto stream = std::make_shared(dbContext(), dm_file, *table_columns); + stream->writePrefix(); + for (size_t i = 0; i < batch_num; i += 1) + { + Block block = DMTestEnv::prepareSimpleWriteBlock(num_rows_write_per_batch * i, num_rows_write_per_batch * (i + 1), false, 100); + stream->write(block, {}); + } + stream->writeSuffix(); + } + + // test read + ColumnFileBig column_file_big(dmContext(), dm_file, RowKeyRange::newAll(false, 1)); + ColumnDefinesPtr column_defines = std::make_shared(); + column_defines->emplace_back(getExtraHandleColumnDefine(/*is_common_handle=*/false)); + column_defines->emplace_back(getVersionColumnDefine()); + auto column_file_big_reader = column_file_big.getReader(dmContext(), /*storage_snap*/ nullptr, column_defines); + { + size_t num_rows_read = 0; + while (Block in = column_file_big_reader->readNextBlock()) + { + ASSERT_EQ(in.columns(), column_defines->size()); + ASSERT_TRUE(in.has(EXTRA_HANDLE_COLUMN_NAME)); + ASSERT_TRUE(in.has(VERSION_COLUMN_NAME)); + auto & pk_column = in.getByName(EXTRA_HANDLE_COLUMN_NAME).column; + for (size_t i = 0; i < pk_column->size(); i++) + { + ASSERT_EQ(pk_column->getInt(i), num_rows_read + i); + } + auto & version_column = in.getByName(VERSION_COLUMN_NAME).column; + for (size_t i = 0; i < version_column->size(); i++) + { + ASSERT_EQ(version_column->getInt(i), tso_value); + } + num_rows_read += in.rows(); + } + ASSERT_EQ(num_rows_read, num_rows_write_per_batch * batch_num); + } + + { + ColumnDefinesPtr column_defines_pk_and_del = std::make_shared(); + column_defines_pk_and_del->emplace_back(getExtraHandleColumnDefine(/*is_common_handle=*/false)); + column_defines_pk_and_del->emplace_back(getTagColumnDefine()); + auto column_file_big_reader2 = column_file_big_reader->createNewReader(column_defines_pk_and_del); + size_t num_rows_read = 0; + while (Block in = column_file_big_reader2->readNextBlock()) + { + ASSERT_EQ(in.columns(), column_defines_pk_and_del->size()); + ASSERT_TRUE(in.has(EXTRA_HANDLE_COLUMN_NAME)); + ASSERT_TRUE(in.has(TAG_COLUMN_NAME)); + auto & pk_column = in.getByName(EXTRA_HANDLE_COLUMN_NAME).column; + for (size_t i = 0; i < pk_column->size(); i++) + { + ASSERT_EQ(pk_column->getInt(i), num_rows_read + i); + } + auto & del_column = in.getByName(TAG_COLUMN_NAME).column; + for (size_t i = 0; i < del_column->size(); i++) + { + ASSERT_EQ(del_column->getInt(i), 0); + } + num_rows_read += in.rows(); + } + ASSERT_EQ(num_rows_read, num_rows_write_per_batch * batch_num); + } + + { + auto column_file_big_reader3 = column_file_big_reader->createNewReader(table_columns); + size_t num_rows_read = 0; + while (Block in = column_file_big_reader3->readNextBlock()) + { + ASSERT_EQ(in.columns(), table_columns->size()); + num_rows_read += in.rows(); + } + ASSERT_EQ(num_rows_read, num_rows_write_per_batch * batch_num); + } +} +CATCH + +} // namespace tests +} // namespace DM +} // namespace DB diff --git a/dbms/src/Storages/DeltaMerge/tests/gtest_dm_delta_merge_store.cpp b/dbms/src/Storages/DeltaMerge/tests/gtest_dm_delta_merge_store.cpp index 79038e1a349..89fc676dfee 100644 --- a/dbms/src/Storages/DeltaMerge/tests/gtest_dm_delta_merge_store.cpp +++ b/dbms/src/Storages/DeltaMerge/tests/gtest_dm_delta_merge_store.cpp @@ -262,6 +262,7 @@ try /* max_version= */ std::numeric_limits::max(), EMPTY_FILTER, TRACING_NAME, + /* keep_order= */ false, /* is_fast_mode= */ false, /* expected_block_size= */ 1024)[0]; @@ -301,7 +302,7 @@ try { // test readRaw const auto & columns = store->getTableColumns(); - BlockInputStreamPtr in = store->readRaw(*db_context, db_context->getSettingsRef(), columns, 1)[0]; + BlockInputStreamPtr in = store->readRaw(*db_context, db_context->getSettingsRef(), columns, 1, /* keep_order= */ false)[0]; size_t num_rows_read = 0; in->readPrefix(); @@ -405,6 +406,7 @@ try /* max_version= */ std::numeric_limits::max(), EMPTY_FILTER, TRACING_NAME, + /* keep_order= */ false, /* is_fast_mode= */ false, /* expected_block_size= */ 1024)[0]; @@ -491,6 +493,7 @@ try /* max_version= */ std::numeric_limits::max(), EMPTY_FILTER, TRACING_NAME, + /* keep_order= */ false, /* is_fast_mode= */ false, /* expected_block_size= */ 1024)[0]; @@ -563,6 +566,7 @@ try /* max_version= */ std::numeric_limits::max(), EMPTY_FILTER, TRACING_NAME, + /* keep_order= */ false, /* is_fast_mode= */ false, /* expected_block_size= */ 1024)[0]; size_t num_rows_read = 0; @@ -601,6 +605,7 @@ try /* max_version= */ std::numeric_limits::max(), EMPTY_FILTER, TRACING_NAME, + /* keep_order= */ false, /* is_fast_mode= */ false, /* expected_block_size= */ 1024)[0]; size_t num_rows_read = 0; @@ -687,6 +692,7 @@ try /* max_version= */ std::numeric_limits::max(), EMPTY_FILTER, TRACING_NAME, + /* keep_order= */ false, /* is_fast_mode= */ false, /* expected_block_size= */ 1024)[0]; size_t num_rows_read = 0; @@ -773,6 +779,7 @@ try /* max_version= */ std::numeric_limits::max(), EMPTY_FILTER, TRACING_NAME, + /* keep_order= */ false, /* is_fast_mode= */ false, /* expected_block_size= */ 1024)[0]; size_t num_rows_read = 0; @@ -805,6 +812,7 @@ try /* max_version= */ UInt64(1), EMPTY_FILTER, TRACING_NAME, + /* keep_order= */ false, /* is_fast_mode= */ false, /* expected_block_size= */ 1024)[0]; size_t num_rows_read = 0; @@ -861,6 +869,7 @@ try /* max_version= */ std::numeric_limits::max(), EMPTY_FILTER, TRACING_NAME, + /* keep_order= */ false, /* is_fast_mode= */ false, /* expected_block_size= */ 1024)[0]; size_t num_rows_read = 0; @@ -901,6 +910,7 @@ try /* max_version= */ std::numeric_limits::max(), EMPTY_FILTER, TRACING_NAME, + /* keep_order= */ false, /* is_fast_mode= */ false, /* expected_block_size= */ 1024)[0]; size_t num_rows_read = 0; @@ -961,6 +971,7 @@ try /* max_version= */ std::numeric_limits::max(), EMPTY_FILTER, TRACING_NAME, + /* keep_order= */ false, /* is_fast_mode= */ false, /* expected_block_size= */ 1024); ASSERT_EQ(ins.size(), 1UL); @@ -985,6 +996,7 @@ try /* max_version= */ tso2, EMPTY_FILTER, TRACING_NAME, + /* keep_order= */ false, /* is_fast_mode= */ false, /* expected_block_size= */ 1024); ASSERT_EQ(ins.size(), 1UL); @@ -1009,6 +1021,7 @@ try /* max_version= */ tso1, EMPTY_FILTER, TRACING_NAME, + /* keep_order= */ false, /* is_fast_mode= */ false, /* expected_block_size= */ 1024); ASSERT_EQ(ins.size(), 1UL); @@ -1033,6 +1046,7 @@ try /* max_version= */ tso1 - 1, EMPTY_FILTER, TRACING_NAME, + /* keep_order= */ false, /* is_fast_mode= */ false, /* expected_block_size= */ 1024); ASSERT_EQ(ins.size(), 1UL); @@ -1095,6 +1109,7 @@ try /* max_version= */ tso1, EMPTY_FILTER, TRACING_NAME, + /* keep_order= */ false, /* is_fast_mode= */ false, /* expected_block_size= */ 1024); ASSERT_EQ(ins.size(), 1UL); @@ -1133,6 +1148,7 @@ try /* max_version= */ tso2 - 1, EMPTY_FILTER, TRACING_NAME, + /* keep_order= */ false, /* is_fast_mode= */ false, /* expected_block_size= */ 1024); ASSERT_EQ(ins.size(), 1UL); @@ -1172,6 +1188,7 @@ try /* max_version= */ tso3 - 1, EMPTY_FILTER, TRACING_NAME, + /* keep_order= */ false, /* is_fast_mode= */ false, /* expected_block_size= */ 1024); ASSERT_EQ(ins.size(), 1UL); @@ -1198,6 +1215,7 @@ try /* max_version= */ std::numeric_limits::max(), EMPTY_FILTER, TRACING_NAME, + /* keep_order= */ false, /* is_fast_mode= */ false, /* expected_block_size= */ 1024); ASSERT_EQ(ins.size(), 1UL); @@ -1224,6 +1242,7 @@ try /* max_version= */ std::numeric_limits::max(), EMPTY_FILTER, TRACING_NAME, + /* keep_order= */ false, /* is_fast_mode= */ false, /* expected_block_size= */ 1024); ASSERT_EQ(ins.size(), 1UL); @@ -1286,6 +1305,7 @@ try /* max_version= */ tso1, EMPTY_FILTER, TRACING_NAME, + /* keep_order= */ false, /* is_fast_mode= */ false, /* expected_block_size= */ 1024); ASSERT_EQ(ins.size(), 1); @@ -1324,6 +1344,7 @@ try /* max_version= */ tso2 - 1, EMPTY_FILTER, TRACING_NAME, + /* keep_order= */ false, /* is_fast_mode= */ false, /* expected_block_size= */ 1024); ASSERT_EQ(ins.size(), 1); @@ -1363,6 +1384,7 @@ try /* max_version= */ std::numeric_limits::max(), EMPTY_FILTER, TRACING_NAME, + /* keep_order= */ false, /* is_fast_mode= */ false, /* expected_block_size= */ 1024); ASSERT_EQ(ins.size(), 1); @@ -1420,6 +1442,7 @@ try /* max_version= */ tso1, EMPTY_FILTER, TRACING_NAME, + /* keep_order= */ false, /* is_fast_mode= */ false, /* expected_block_size= */ 1024); ASSERT_EQ(ins.size(), 1); @@ -1458,6 +1481,7 @@ try /* max_version= */ std::numeric_limits::max(), EMPTY_FILTER, TRACING_NAME, + /* keep_order= */ false, /* is_fast_mode= */ false, /* expected_block_size= */ 1024); ASSERT_EQ(ins.size(), 1); @@ -1545,6 +1569,7 @@ try /* max_version= */ std::numeric_limits::max(), EMPTY_FILTER, TRACING_NAME, + /* keep_order= */ false, /* is_fast_mode= */ false, /* expected_block_size= */ 1024); ASSERT_EQ(ins.size(), 1UL); @@ -1654,6 +1679,7 @@ try /* max_version= */ std::numeric_limits::max(), EMPTY_FILTER, TRACING_NAME, + /* keep_order= */ false, /* is_fast_mode= */ false, /* expected_block_size= */ 1024); ASSERT_EQ(ins.size(), 1UL); @@ -1762,6 +1788,7 @@ try /* max_version= */ std::numeric_limits::max(), EMPTY_FILTER, TRACING_NAME, + /* keep_order= */ false, /* is_fast_mode= */ false, /* expected_block_size= */ 1024); ASSERT_EQ(ins.size(), 1UL); @@ -1851,6 +1878,7 @@ try /* max_version= */ std::numeric_limits::max(), EMPTY_FILTER, TRACING_NAME, + /* keep_order= */ false, /* is_fast_mode= */ false, /* expected_block_size= */ 1024); ASSERT_EQ(ins.size(), 1UL); @@ -1955,6 +1983,7 @@ try /* max_version= */ std::numeric_limits::max(), EMPTY_FILTER, TRACING_NAME, + /* keep_order= */ false, /* is_fast_mode= */ false, /* expected_block_size= */ 1024)[0]; @@ -2032,6 +2061,7 @@ try /* max_version= */ std::numeric_limits::max(), EMPTY_FILTER, TRACING_NAME, + /* keep_order= */ false, /* is_fast_mode= */ false, /* expected_block_size= */ 1024)[0]; @@ -2109,6 +2139,7 @@ try /* max_version= */ std::numeric_limits::max(), EMPTY_FILTER, TRACING_NAME, + /* keep_order= */ false, /* is_fast_mode= */ false, /* expected_block_size= */ 1024)[0]; @@ -2186,6 +2217,7 @@ try /* max_version= */ std::numeric_limits::max(), EMPTY_FILTER, TRACING_NAME, + /* keep_order= */ false, /* is_fast_mode= */ false, /* expected_block_size= */ 1024)[0]; @@ -2263,6 +2295,7 @@ try /* max_version= */ std::numeric_limits::max(), EMPTY_FILTER, TRACING_NAME, + /* keep_order= */ false, /* is_fast_mode= */ false, /* expected_block_size= */ 1024)[0]; @@ -2338,6 +2371,7 @@ try /* max_version= */ std::numeric_limits::max(), EMPTY_FILTER, TRACING_NAME, + /* keep_order= */ false, /* is_fast_mode= */ false, /* expected_block_size= */ 1024)[0]; @@ -2412,6 +2446,7 @@ try /* max_version= */ std::numeric_limits::max(), EMPTY_FILTER, TRACING_NAME, + /* keep_order= */ false, /* is_fast_mode= */ false, /* expected_block_size= */ 1024)[0]; @@ -2503,6 +2538,7 @@ try /* max_version= */ std::numeric_limits::max(), EMPTY_FILTER, TRACING_NAME, + /* keep_order= */ false, /* is_fast_mode= */ false, /* expected_block_size= */ 1024); ASSERT_EQ(ins.size(), 1UL); @@ -2638,6 +2674,7 @@ try /* max_version= */ std::numeric_limits::max(), EMPTY_FILTER, TRACING_NAME, + /* keep_order= */ false, /* is_fast_mode= */ false, /* expected_block_size= */ 1024); ASSERT_EQ(ins.size(), 1UL); @@ -2700,6 +2737,7 @@ try /* max_version= */ std::numeric_limits::max(), EMPTY_FILTER, TRACING_NAME, + /* keep_order= */ false, /* is_fast_mode= */ false, /* expected_block_size= */ 1024); ASSERT_EQ(ins.size(), 1UL); @@ -2800,6 +2838,7 @@ try /* max_version= */ std::numeric_limits::max(), EMPTY_FILTER, TRACING_NAME, + /* keep_order= */ false, /* is_fast_mode= */ false, /* expected_block_size= */ 1024)[0]; @@ -2854,6 +2893,7 @@ try /* max_version= */ std::numeric_limits::max(), EMPTY_FILTER, TRACING_NAME, + /* keep_order= */ false, /* is_fast_mode= */ false, /* expected_block_size= */ 1024)[0]; @@ -2971,6 +3011,7 @@ try /* max_version= */ std::numeric_limits::max(), EMPTY_FILTER, TRACING_NAME, + /* keep_order= */ false, /* is_fast_mode= */ false, /* expected_block_size= */ 1024)[0]; @@ -3009,7 +3050,7 @@ try { // test readRaw const auto & columns = store->getTableColumns(); - BlockInputStreamPtr in = store->readRaw(*db_context, db_context->getSettingsRef(), columns, 1)[0]; + BlockInputStreamPtr in = store->readRaw(*db_context, db_context->getSettingsRef(), columns, 1, /* keep_order= */ false)[0]; size_t num_rows_read = 0; in->readPrefix(); @@ -3101,6 +3142,7 @@ try /* max_version= */ std::numeric_limits::max(), EMPTY_FILTER, TRACING_NAME, + /* keep_order= */ false, /* is_fast_mode= */ false, /* expected_block_size= */ 1024)[0]; size_t num_rows_read = 0; @@ -3176,6 +3218,7 @@ try /* max_version= */ std::numeric_limits::max(), EMPTY_FILTER, TRACING_NAME, + /* keep_order= */ false, /* is_fast_mode= */ false, /* expected_block_size= */ 1024)[0]; size_t num_rows_read = 0; @@ -3208,6 +3251,7 @@ try /* max_version= */ UInt64(1), EMPTY_FILTER, TRACING_NAME, + /* keep_order= */ false, /* is_fast_mode= */ false, /* expected_block_size= */ 1024)[0]; size_t num_rows_read = 0; @@ -3267,6 +3311,7 @@ try /* max_version= */ std::numeric_limits::max(), EMPTY_FILTER, TRACING_NAME, + /* keep_order= */ false, /* is_fast_mode= */ false, /* expected_block_size= */ 1024)[0]; size_t num_rows_read = 0; @@ -3318,6 +3363,7 @@ try /* max_version= */ std::numeric_limits::max(), EMPTY_FILTER, TRACING_NAME, + /* keep_order= */ false, /* is_fast_mode= */ false, /* expected_block_size= */ 1024)[0]; size_t num_rows_read = 0; @@ -3392,6 +3438,7 @@ try /* max_version= */ std::numeric_limits::max(), EMPTY_FILTER, TRACING_NAME, + /* keep_order= */ false, /* is_fast_mode= */ false, /* expected_block_size= */ 1024); ASSERT_EQ(ins.size(), 1UL); diff --git a/dbms/src/Storages/DeltaMerge/tests/gtest_dm_delta_merge_store_for_fast_mode.cpp b/dbms/src/Storages/DeltaMerge/tests/gtest_dm_delta_merge_store_for_fast_mode.cpp index 360f3e96315..8cfa7c07642 100644 --- a/dbms/src/Storages/DeltaMerge/tests/gtest_dm_delta_merge_store_for_fast_mode.cpp +++ b/dbms/src/Storages/DeltaMerge/tests/gtest_dm_delta_merge_store_for_fast_mode.cpp @@ -89,6 +89,7 @@ TEST_P(DeltaMergeStoreRWTest, TestFastModeWithOnlyInsertWithoutRangeFilter) /* max_version= */ std::numeric_limits::max(), EMPTY_FILTER, TRACING_NAME, + /* keep_order= */ false, /* is_raw_read= */ true, /* expected_block_size= */ 1024)[0]; @@ -192,6 +193,7 @@ TEST_P(DeltaMergeStoreRWTest, TestFastModeWithOnlyInsertWithRangeFilter) /* max_version= */ std::numeric_limits::max(), EMPTY_FILTER, TRACING_NAME, + /* keep_order= */ false, /* is_raw_read= */ true, /* expected_block_size= */ 1024)[0]; @@ -287,6 +289,7 @@ try /* max_version= */ std::numeric_limits::max(), EMPTY_FILTER, TRACING_NAME, + /* keep_order= */ false, /* is_raw_read= */ true, /* expected_block_size= */ 1024)[0]; size_t num_rows_read = 0; @@ -433,6 +436,7 @@ try /* max_version= */ std::numeric_limits::max(), EMPTY_FILTER, TRACING_NAME, + /* keep_order= */ false, /* is_raw_read= */ true, /* expected_block_size= */ 1024)[0]; size_t num_rows_read = 0; @@ -580,6 +584,7 @@ try /* max_version= */ std::numeric_limits::max(), EMPTY_FILTER, TRACING_NAME, + /* keep_order= */ false, /* is_raw_read= */ true, /* expected_block_size= */ 1024)[0]; size_t num_rows_read = 0; @@ -731,6 +736,7 @@ try /* max_version= */ std::numeric_limits::max(), EMPTY_FILTER, TRACING_NAME, + /* keep_order= */ false, /* is_raw_read= */ true, /* expected_block_size= */ 1024)[0]; size_t num_rows_read = 0; @@ -826,6 +832,7 @@ try /* max_version= */ std::numeric_limits::max(), EMPTY_FILTER, TRACING_NAME, + /* keep_order= */ false, /* is_raw_read= */ true, /* expected_block_size= */ 1024)[0]; size_t num_rows_read = 0; @@ -1011,6 +1018,7 @@ try /* max_version= */ std::numeric_limits::max(), EMPTY_FILTER, TRACING_NAME, + /* keep_order= */ false, /* is_raw_read= */ true, /* expected_block_size= */ 1024)[0]; size_t num_rows_read = 0; @@ -1052,6 +1060,7 @@ try /* max_version= */ std::numeric_limits::max(), EMPTY_FILTER, TRACING_NAME, + /* keep_order= */ false, /* is_raw_read= */ true, /* expected_block_size= */ 1024)[0]; size_t num_rows_read = 0; @@ -1114,6 +1123,7 @@ try /* max_version= */ std::numeric_limits::max(), EMPTY_FILTER, TRACING_NAME, + /* keep_order= */ false, /* is_raw_read= */ true, /* expected_block_size= */ 1024)[0]; size_t num_rows_read = 0; @@ -1153,6 +1163,7 @@ try /* max_version= */ std::numeric_limits::max(), EMPTY_FILTER, TRACING_NAME, + /* keep_order = */ false, /* is_raw_read= */ true, /* expected_block_size= */ 1024)[0]; size_t num_rows_read = 0; @@ -1229,6 +1240,7 @@ try /* max_version= */ std::numeric_limits::max(), EMPTY_FILTER, TRACING_NAME, + /* keep_order= */ false, /* is_raw_read= */ true, /* expected_block_size= */ 1024)[0]; size_t num_rows_read = 0; @@ -1331,6 +1343,7 @@ try /* max_version= */ std::numeric_limits::max(), EMPTY_FILTER, TRACING_NAME, + /* keep_order= */ false, /* is_raw_read= */ true, /* expected_block_size= */ 1024)[0]; size_t num_rows_read = 0; @@ -1445,6 +1458,7 @@ try /* max_version= */ UInt64(1), EMPTY_FILTER, TRACING_NAME, + /* keep_order= */ false, /* is_raw_read= */ false, /* expected_block_size= */ 1024)[0]; size_t num_rows_read = 0; @@ -1513,6 +1527,7 @@ try /* max_version= */ std::numeric_limits::max(), EMPTY_FILTER, TRACING_NAME, + /* keep_order= */ false, /* is_raw_read= */ true, /* expected_block_size= */ 1024)[0]; size_t num_rows_read = 0; @@ -1571,6 +1586,7 @@ try /* max_version= */ std::numeric_limits::max(), EMPTY_FILTER, TRACING_NAME, + /* keep_order= */ false, /* is_raw_read= */ true, /* expected_block_size= */ 1024)[0]; size_t num_rows_read = 0; diff --git a/dbms/src/Storages/DeltaMerge/tests/gtest_dm_minmax_index.cpp b/dbms/src/Storages/DeltaMerge/tests/gtest_dm_minmax_index.cpp index a575d022476..f066237d8e2 100644 --- a/dbms/src/Storages/DeltaMerge/tests/gtest_dm_minmax_index.cpp +++ b/dbms/src/Storages/DeltaMerge/tests/gtest_dm_minmax_index.cpp @@ -124,7 +124,7 @@ bool checkMatch( store->mergeDeltaAll(context); const ColumnDefine & col_to_read = check_pk ? getExtraHandleColumnDefine(is_common_handle) : cd; - auto streams = store->read(context, context.getSettingsRef(), {col_to_read}, {all_range}, 1, std::numeric_limits::max(), filter, name); + auto streams = store->read(context, context.getSettingsRef(), {col_to_read}, {all_range}, 1, std::numeric_limits::max(), filter, name, false); streams[0]->readPrefix(); auto rows = streams[0]->read().rows(); streams[0]->readSuffix(); diff --git a/dbms/src/Storages/DeltaMerge/workload/DTWorkload.cpp b/dbms/src/Storages/DeltaMerge/workload/DTWorkload.cpp index faa0100e9e2..93877181e55 100644 --- a/dbms/src/Storages/DeltaMerge/workload/DTWorkload.cpp +++ b/dbms/src/Storages/DeltaMerge/workload/DTWorkload.cpp @@ -190,7 +190,7 @@ void DTWorkload::read(const ColumnDefines & columns, int stream_count, T func) auto filter = EMPTY_FILTER; int excepted_block_size = 1024; uint64_t read_ts = ts_gen->get(); - auto streams = store->read(*context, context->getSettingsRef(), columns, ranges, stream_count, read_ts, filter, "DTWorkload", opts->is_fast_mode, excepted_block_size); + auto streams = store->read(*context, context->getSettingsRef(), columns, ranges, stream_count, read_ts, filter, "DTWorkload", false, opts->is_fast_mode, excepted_block_size); std::vector threads; threads.reserve(streams.size()); for (auto & stream : streams) diff --git a/dbms/src/Storages/Page/PageDefines.h b/dbms/src/Storages/Page/PageDefines.h index 7feea494eb4..330ed7776a3 100644 --- a/dbms/src/Storages/Page/PageDefines.h +++ b/dbms/src/Storages/Page/PageDefines.h @@ -41,7 +41,7 @@ static_assert(PAGE_SIZE_STEP >= ((1 << 10) * 16), "PAGE_SIZE_STEP should be at l static_assert((PAGE_SIZE_STEP & (PAGE_SIZE_STEP - 1)) == 0, "PAGE_SIZE_STEP should be power of 2"); // PageStorage V3 define -static constexpr UInt64 BLOBFILE_LIMIT_SIZE = 512 * MB; +static constexpr UInt64 BLOBFILE_LIMIT_SIZE = 256 * MB; static constexpr UInt64 BLOBSTORE_CACHED_FD_SIZE = 100; static constexpr UInt64 PAGE_META_ROLL_SIZE = 2 * MB; static constexpr UInt64 MAX_PERSISTED_LOG_FILES = 4; diff --git a/dbms/src/Storages/Page/PageStorage.cpp b/dbms/src/Storages/Page/PageStorage.cpp index d8b767a6c15..c56d47d16b3 100644 --- a/dbms/src/Storages/Page/PageStorage.cpp +++ b/dbms/src/Storages/Page/PageStorage.cpp @@ -23,12 +23,13 @@ PageStoragePtr PageStorage::create( PSDiskDelegatorPtr delegator, const PageStorage::Config & config, const FileProviderPtr & file_provider, - bool use_v3) + bool use_v3, + bool no_more_insert_to_v2) { if (use_v3) return std::make_shared(name, delegator, config, file_provider); else - return std::make_shared(name, delegator, config, file_provider); + return std::make_shared(name, delegator, config, file_provider, no_more_insert_to_v2); } /*************************** diff --git a/dbms/src/Storages/Page/PageStorage.h b/dbms/src/Storages/Page/PageStorage.h index 0059c0570c1..759f16f53e5 100644 --- a/dbms/src/Storages/Page/PageStorage.h +++ b/dbms/src/Storages/Page/PageStorage.h @@ -138,7 +138,7 @@ class PageStorage : private boost::noncopyable SettingUInt64 blob_file_limit_size = BLOBFILE_LIMIT_SIZE; SettingUInt64 blob_spacemap_type = 2; SettingUInt64 blob_cached_fd_size = BLOBSTORE_CACHED_FD_SIZE; - SettingDouble blob_heavy_gc_valid_rate = 0.2; + SettingDouble blob_heavy_gc_valid_rate = 0.5; SettingUInt64 blob_block_alignment_bytes = 0; SettingUInt64 wal_roll_size = PAGE_META_ROLL_SIZE; @@ -208,6 +208,14 @@ class PageStorage : private boost::noncopyable void reloadSettings(const Config & new_config) { config.reload(new_config); }; Config getSettings() const { return config; } + // Use a more easy gc config for v2 when all of its data will be transformed to v3. + static Config getEasyGCConfig() + { + Config gc_config; + gc_config.file_roll_size = PAGE_FILE_SMALL_SIZE; + return gc_config; + } + public: static PageStoragePtr create( @@ -215,7 +223,8 @@ class PageStorage : private boost::noncopyable PSDiskDelegatorPtr delegator, const PageStorage::Config & config, const FileProviderPtr & file_provider, - bool use_v3 = false); + bool use_v3 = false, + bool no_more_insert_to_v2 = false); PageStorage( String name, diff --git a/dbms/src/Storages/Page/V2/PageStorage.cpp b/dbms/src/Storages/Page/V2/PageStorage.cpp index 662e9edd686..ba78cf6a75b 100644 --- a/dbms/src/Storages/Page/V2/PageStorage.cpp +++ b/dbms/src/Storages/Page/V2/PageStorage.cpp @@ -160,12 +160,14 @@ PageFileSet PageStorage::listAllPageFiles(const FileProviderPtr & file_provider, PageStorage::PageStorage(String name, PSDiskDelegatorPtr delegator_, // const Config & config_, - const FileProviderPtr & file_provider_) + const FileProviderPtr & file_provider_, + bool no_more_insert_) : DB::PageStorage(name, delegator_, config_, file_provider_) , write_files(std::max(1UL, config_.num_write_slots.get())) , page_file_log(&Poco::Logger::get("PageFile")) , log(&Poco::Logger::get("PageStorage")) , versioned_page_entries(storage_name, config.version_set_config, log) + , no_more_insert(no_more_insert_) { // at least 1 write slots config.num_write_slots = std::max(1UL, config.num_write_slots.get()); @@ -405,18 +407,20 @@ DB::PageEntry PageStorage::getEntryImpl(NamespaceId /*ns_id*/, PageId page_id, S // - Writable, reuse `old_writer` if it is not a nullptr, otherwise, create a new writer from `page_file` // - Not writable, renew the `page_file` and its belonging writer. // The of the new `page_file` is of all `write_files` +// If `force` is true, always create new page file for writing. PageStorage::WriterPtr PageStorage::checkAndRenewWriter( // WritingPageFile & writing_file, PageFileIdAndLevel max_page_file_id_lvl_hint, const String & parent_path_hint, PageStorage::WriterPtr && old_writer, - const String & logging_msg) + const String & logging_msg, + bool force) { WriterPtr write_file_writer; PageFile & page_file = writing_file.file; - bool is_writable = page_file.isValid() && page_file.getType() == PageFile::Type::Formal // - && isPageFileSizeFitsWritable(page_file, config); + bool is_writable = (!force && page_file.isValid() && page_file.getType() == PageFile::Type::Formal // + && isPageFileSizeFitsWritable(page_file, config)); if (is_writable) { if (old_writer) @@ -451,6 +455,7 @@ PageStorage::WriterPtr PageStorage::checkAndRenewWriter( // PageFileIdAndLevel max_writing_id_lvl{max_page_file_id_lvl_hint}; for (const auto & wf : write_files) max_writing_id_lvl = std::max(max_writing_id_lvl, wf.file.fileIdLevel()); + delegator->addPageFileUsedSize( // PageFileIdAndLevel(max_writing_id_lvl.first + 1, 0), 0, @@ -1086,13 +1091,55 @@ bool PageStorage::gcImpl(bool not_skip, const WriteLimiterPtr & write_limiter, c { external_pages_remover(external_pages, live_normal_pages); } + + // This instance will accept no more write, and we check whether there is any valid page in non writing page file, + // If not, it's very possible that all data have been moved to v3, so we try to roll all writing files to make them can be gced + if (no_more_insert) + { + bool has_normal_non_writing_files = false; + bool has_non_empty_write_file = false; + for (const auto & page_file : page_files) + { + if (page_file.fileIdLevel() < min_writing_file_id_level) + { + if (page_file.getType() == PageFile::Type::Formal) + { + has_normal_non_writing_files = true; + } + } + else + { + // writing files + if (page_file.getDiskSize() > 0) + { + has_non_empty_write_file = true; + } + } + if (has_normal_non_writing_files && has_non_empty_write_file) + break; + } + if (!has_normal_non_writing_files && has_non_empty_write_file) + { + std::unique_lock lock(write_mutex); + LOG_FMT_DEBUG(log, "{} No valid pages in non writing page files and the writing page files are not all empty. Try to roll all {} writing page files", storage_name, write_files.size()); + idle_writers.clear(); + for (auto & write_file : write_files) + { + auto writer = checkAndRenewWriter(write_file, {0, 0}, /*parent_path_hint=*/"", nullptr, "", /*force*/ true); + idle_writers.emplace_back(std::move(writer)); + } + } + } }; GCType gc_type = GCType::Normal; // Ignore page files that maybe writing to. do { - if (not_skip) // For page_storage_ctl, don't skip the GC + // Don't skip the GC under the case: + // 1. For page_storage_ctl + // 2. After transform all data from v2 to v3 + if (not_skip) { gc_type = GCType::Normal; break; @@ -1103,6 +1150,10 @@ bool PageStorage::gcImpl(bool not_skip, const WriteLimiterPtr & write_limiter, c // If only few page files, running gc is useless. gc_type = GCType::Skip; } + else if (no_more_insert) + { + gc_type = GCType::Normal; + } else if (last_gc_statistics.equals(statistics_snapshot)) { // No write since last gc. Give it a chance for running GC, ensure that we are able to diff --git a/dbms/src/Storages/Page/V2/PageStorage.h b/dbms/src/Storages/Page/V2/PageStorage.h index b85bc2ff355..1fcb76885bb 100644 --- a/dbms/src/Storages/Page/V2/PageStorage.h +++ b/dbms/src/Storages/Page/V2/PageStorage.h @@ -88,7 +88,8 @@ class PageStorage : public DB::PageStorage PageStorage(String name, PSDiskDelegatorPtr delegator, // const Config & config_, - const FileProviderPtr & file_provider_); + const FileProviderPtr & file_provider_, + bool no_more_insert_ = false); ~PageStorage() = default; void restore() override; @@ -272,13 +273,17 @@ class PageStorage : public DB::PageStorage StatisticsInfo last_gc_statistics; + // true means this instance runs under mix mode + bool no_more_insert = false; + private: WriterPtr checkAndRenewWriter( WritingPageFile & writing_file, PageFileIdAndLevel max_page_file_id_lvl_hint, const String & parent_path_hint, WriterPtr && old_writer = nullptr, - const String & logging_msg = ""); + const String & logging_msg = "", + bool force = false); }; } // namespace PS::V2 diff --git a/dbms/src/Storages/PathPool.cpp b/dbms/src/Storages/PathPool.cpp index 0ae42ec7313..2e7edd7435b 100644 --- a/dbms/src/Storages/PathPool.cpp +++ b/dbms/src/Storages/PathPool.cpp @@ -182,6 +182,52 @@ StoragePathPool & StoragePathPool::operator=(const StoragePathPool & rhs) return *this; } +bool StoragePathPool::createPSV2DeleteMarkFile() +{ + try + { + return Poco::File(getPSV2DeleteMarkFilePath()).createFile(); + } + catch (...) + { + tryLogCurrentException(log); + return false; + } +} + +bool StoragePathPool::isPSV2Deleted() const +{ + return Poco::File(getPSV2DeleteMarkFilePath()).exists(); +} + +void StoragePathPool::clearPSV2ObsoleteData() +{ + std::lock_guard lock{mutex}; + auto drop_instance_data = [&](const String & prefix) { + for (auto & path_info : latest_path_infos) + { + try + { + auto path = fmt::format("{}/{}", path_info.path, prefix); + if (Poco::File dir(path); dir.exists()) + { + // This function is used to clean obsolete data in ps v2 instance at restart, + // so no need to update global_capacity here. + LOG_FMT_INFO(log, "Begin to drop obsolete data[dir={}]", path); + file_provider->deleteDirectory(dir.path(), false, true); + } + } + catch (...) + { + tryLogCurrentException(log); + } + } + }; + drop_instance_data("meta"); + drop_instance_data("log"); + drop_instance_data("data"); +} + void StoragePathPool::rename(const String & new_database, const String & new_table, bool clean_rename) { if (unlikely(new_database.empty() || new_table.empty())) @@ -311,6 +357,11 @@ void StoragePathPool::renamePath(const String & old_path, const String & new_pat LOG_FMT_WARNING(log, "Path \"{}\" is missed.", old_path); } +String StoragePathPool::getPSV2DeleteMarkFilePath() const +{ + return fmt::format("{}/{}", latest_path_infos[0].path, "PS_V2_DELETED"); +} + //========================================================================================== // Generic functions //========================================================================================== diff --git a/dbms/src/Storages/PathPool.h b/dbms/src/Storages/PathPool.h index 40a3cb9a636..4f16511feff 100644 --- a/dbms/src/Storages/PathPool.h +++ b/dbms/src/Storages/PathPool.h @@ -406,6 +406,12 @@ class StoragePathPool // Those paths are generated from the first path of `latest_path_infos` and `prefix` PSDiskDelegatorPtr getPSDiskDelegatorSingle(const String & prefix) { return std::make_shared(*this, prefix); } + bool createPSV2DeleteMarkFile(); + + bool isPSV2Deleted() const; + + void clearPSV2ObsoleteData(); + void rename(const String & new_database, const String & new_table, bool clean_rename); void drop(bool recursive, bool must_success = true); @@ -415,6 +421,8 @@ class StoragePathPool void renamePath(const String & old_path, const String & new_path); + String getPSV2DeleteMarkFilePath() const; + private: using DMFilePathMap = std::unordered_map; struct MainPathInfo diff --git a/dbms/src/Storages/SelectQueryInfo.cpp b/dbms/src/Storages/SelectQueryInfo.cpp index 075c0ad0631..9b5dbf0d30f 100644 --- a/dbms/src/Storages/SelectQueryInfo.cpp +++ b/dbms/src/Storages/SelectQueryInfo.cpp @@ -29,6 +29,7 @@ SelectQueryInfo::SelectQueryInfo(const SelectQueryInfo & rhs) , mvcc_query_info(rhs.mvcc_query_info != nullptr ? std::make_unique(*rhs.mvcc_query_info) : nullptr) , dag_query(rhs.dag_query != nullptr ? std::make_unique(*rhs.dag_query) : nullptr) , req_id(rhs.req_id) + , keep_order(rhs.keep_order) {} SelectQueryInfo::SelectQueryInfo(SelectQueryInfo && rhs) noexcept @@ -37,6 +38,7 @@ SelectQueryInfo::SelectQueryInfo(SelectQueryInfo && rhs) noexcept , mvcc_query_info(std::move(rhs.mvcc_query_info)) , dag_query(std::move(rhs.dag_query)) , req_id(std::move(rhs.req_id)) + , keep_order(rhs.keep_order) {} } // namespace DB diff --git a/dbms/src/Storages/SelectQueryInfo.h b/dbms/src/Storages/SelectQueryInfo.h index d49d4c831d1..0807647aa26 100644 --- a/dbms/src/Storages/SelectQueryInfo.h +++ b/dbms/src/Storages/SelectQueryInfo.h @@ -51,7 +51,7 @@ struct SelectQueryInfo std::unique_ptr dag_query; std::string req_id; - + bool keep_order = true; SelectQueryInfo(); ~SelectQueryInfo(); diff --git a/dbms/src/Storages/StorageDeltaMerge.cpp b/dbms/src/Storages/StorageDeltaMerge.cpp index c39a882d53c..1bf3e7c1b73 100644 --- a/dbms/src/Storages/StorageDeltaMerge.cpp +++ b/dbms/src/Storages/StorageDeltaMerge.cpp @@ -632,6 +632,7 @@ BlockInputStreams StorageDeltaMerge::read( context.getSettingsRef(), columns_to_read, num_streams, + query_info.keep_order, parseSegmentSet(select_query.segment_expression_list), extra_table_id_index); } @@ -756,6 +757,7 @@ BlockInputStreams StorageDeltaMerge::read( /*max_version=*/mvcc_query_info.read_tso, rs_operator, query_info.req_id, + query_info.keep_order, /* is_fast_mode */ tidb_table_info.tiflash_mode == TiDB::TiFlashMode::Fast, // read in normal mode or read in fast mode max_block_size, parseSegmentSet(select_query.segment_expression_list), @@ -838,7 +840,8 @@ size_t getRows(DM::DeltaMergeStorePtr & store, const Context & context, const DM 1, std::numeric_limits::max(), EMPTY_FILTER, - /*tracing_id*/ "getRows")[0]; + /*tracing_id*/ "getRows", + /*keep_order*/ false)[0]; stream->readPrefix(); Block block; while ((block = stream->read())) @@ -863,7 +866,8 @@ DM::RowKeyRange getRange(DM::DeltaMergeStorePtr & store, const Context & context 1, std::numeric_limits::max(), EMPTY_FILTER, - /*tracing_id*/ "getRange")[0]; + /*tracing_id*/ "getRange", + /*keep_order*/ false)[0]; stream->readPrefix(); Block block; size_t index = 0; diff --git a/dbms/src/Storages/Transaction/Collator.cpp b/dbms/src/Storages/Transaction/Collator.cpp index 1b0221a6829..fc40701e3c5 100644 --- a/dbms/src/Storages/Transaction/Collator.cpp +++ b/dbms/src/Storages/Transaction/Collator.cpp @@ -13,9 +13,9 @@ // limitations under the License. #include -#include #include #include +#include #include @@ -92,15 +92,12 @@ class Pattern : public ITiDBCollator::IPattern tp = MatchType::Match; if (offset < pattern.length()) { - auto old_offset = offset; - c = Collator::decodeChar(pattern.data(), old_offset); - if (c == escape || c == '_' || c == '%') - offset = old_offset; - else - { - assert(escape >= 0); - c = static_cast(escape); // NOLINT(bugprone-signed-char-misuse) - } + // use next char to match + c = Collator::decodeChar(pattern.data(), offset); + } + else + { + // use `escape` to match } } else if (c == '_') @@ -186,22 +183,12 @@ class BinCollator final : public ITiDBCollator int compare(const char * s1, size_t length1, const char * s2, size_t length2) const override { - if constexpr (padding) - return DB::RtrimStrCompare({s1, length1}, {s2, length2}); - else - return DB::RawStrCompare({s1, length1}, {s2, length2}); + return DB::BinCollatorCompare(s1, length1, s2, length2); } StringRef sortKey(const char * s, size_t length, std::string &) const override { - if constexpr (padding) - { - return StringRef(rtrim(s, length)); - } - else - { - return StringRef(s, length); - } + return DB::BinCollatorSortKey(s, length); } std::unique_ptr pattern() const override { return std::make_unique>>(); } @@ -593,6 +580,9 @@ TiDBCollatorPtr ITiDBCollator::getCollator(int32_t id) { switch (id) { + case ITiDBCollator::UTF8MB4_BIN: + static const auto utf8mb4_collator = UTF8MB4_BIN_TYPE(UTF8MB4_BIN); + return &utf8mb4_collator; case ITiDBCollator::BINARY: static const auto binary_collator = BinCollator(BINARY); return &binary_collator; @@ -602,9 +592,6 @@ TiDBCollatorPtr ITiDBCollator::getCollator(int32_t id) case ITiDBCollator::LATIN1_BIN: static const auto latin1_collator = BinCollator(LATIN1_BIN); return &latin1_collator; - case ITiDBCollator::UTF8MB4_BIN: - static const auto utf8mb4_collator = UTF8MB4_BIN_TYPE(UTF8MB4_BIN); - return &utf8mb4_collator; case ITiDBCollator::UTF8_BIN: static const auto utf8_collator = UTF8MB4_BIN_TYPE(UTF8_BIN); return &utf8_collator; diff --git a/dbms/src/Storages/Transaction/CollatorUtils.h b/dbms/src/Storages/Transaction/CollatorUtils.h new file mode 100644 index 00000000000..3f318a5b700 --- /dev/null +++ b/dbms/src/Storages/Transaction/CollatorUtils.h @@ -0,0 +1,96 @@ +// Copyright 2022 PingCAP, Ltd. +// +// Licensed under the Apache License, Version 2.0 (the "License"); +// you may not use this file except in compliance with the License. +// You may obtain a copy of the License at +// +// http://www.apache.org/licenses/LICENSE-2.0 +// +// Unless required by applicable law or agreed to in writing, software +// distributed under the License is distributed on an "AS IS" BASIS, +// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +// See the License for the specific language governing permissions and +// limitations under the License. + +#pragma once + +#include +#include + +#include + +#define FLATTEN_INLINE_PURE __attribute__((flatten, always_inline, pure)) + +namespace DB +{ + +template +ALWAYS_INLINE inline int signum(T val) +{ + return (0 < val) - (val < 0); +} + +// Check equality is much faster than other comparison. +// - check size first +// - return 0 if equal else 1 +FLATTEN_INLINE_PURE inline int RawStrEqualCompare(const std::string_view & lhs, const std::string_view & rhs) +{ + return StringRef(lhs) == StringRef(rhs) ? 0 : 1; +} + +// Compare str view by memcmp +FLATTEN_INLINE_PURE inline int RawStrCompare(const std::string_view & v1, const std::string_view & v2) +{ + return v1.compare(v2); +} + +constexpr char SPACE = ' '; + +FLATTEN_INLINE_PURE inline std::string_view RightTrimRaw(const std::string_view & v) +{ + size_t end = v.find_last_not_of(SPACE); + return end == std::string_view::npos ? std::string_view{} : std::string_view(v.data(), end + 1); +} + +// Remove tail space +FLATTEN_INLINE_PURE inline std::string_view RightTrim(const std::string_view & v) +{ + if (likely(v.empty() || v.back() != SPACE)) + return v; + return RightTrimRaw(v); +} + +FLATTEN_INLINE_PURE inline std::string_view RightTrimNoEmpty(const std::string_view & v) +{ + if (likely(v.back() != SPACE)) + return v; + return RightTrimRaw(v); +} + +FLATTEN_INLINE_PURE inline int RtrimStrCompare(const std::string_view & va, const std::string_view & vb) +{ + return RawStrCompare(RightTrim(va), RightTrim(vb)); +} + +template +FLATTEN_INLINE_PURE inline int BinCollatorCompare(const char * s1, size_t length1, const char * s2, size_t length2) +{ + if constexpr (padding) + return DB::RtrimStrCompare({s1, length1}, {s2, length2}); + else + return DB::RawStrCompare({s1, length1}, {s2, length2}); +} + +template +FLATTEN_INLINE_PURE inline StringRef BinCollatorSortKey(const char * s, size_t length) +{ + if constexpr (padding) + { + return StringRef(RightTrim({s, length})); + } + else + { + return StringRef(s, length); + } +} +} // namespace DB diff --git a/dbms/src/Storages/Transaction/KVStore.cpp b/dbms/src/Storages/Transaction/KVStore.cpp index fb31e4476bb..c1394746e1a 100644 --- a/dbms/src/Storages/Transaction/KVStore.cpp +++ b/dbms/src/Storages/Transaction/KVStore.cpp @@ -410,20 +410,12 @@ EngineStoreApplyRes KVStore::handleUselessAdminRaftCmd( curr_region.handleWriteRaftCmd({}, index, term, tmt); - const auto check_sync_log = [&]() { - if (cmd_type != raft_cmdpb::AdminCmdType::CompactLog) - { - // ignore ComputeHash, VerifyHash or other useless cmd. - return false; - } - else - { - return canFlushRegionDataImpl(curr_region_ptr, true, /* try_until_succeed */ false, tmt, region_task_lock); - } - }; - - if (check_sync_log()) + if (cmd_type == raft_cmdpb::AdminCmdType::CompactLog) { + // Before CompactLog, we ought to make sure all data of this region are persisted. + // So proxy will firstly call an FFI `fn_try_flush_data` to trigger a attempt to flush data on TiFlash's side. + // If the attempt fails, Proxy will filter execution of this CompactLog, which means every CompactLog observed by TiFlash can ALWAYS succeed now. + // ref. https://github.com/pingcap/tidb-engine-ext/blob/e83a37d2d8d8ae1778fe279c5f06a851f8c9e56a/components/raftstore/src/engine_store_ffi/observer.rs#L175 return EngineStoreApplyRes::Persist; } return EngineStoreApplyRes::None; diff --git a/dbms/src/Storages/Transaction/RegionPersister.cpp b/dbms/src/Storages/Transaction/RegionPersister.cpp index 7ce52c6caa1..8aae893efdc 100644 --- a/dbms/src/Storages/Transaction/RegionPersister.cpp +++ b/dbms/src/Storages/Transaction/RegionPersister.cpp @@ -189,7 +189,7 @@ void RegionPersister::forceTransformKVStoreV2toV3() const auto & page_transform_entry = page_reader->getPageEntry(page.page_id); if (!page_transform_entry.field_offsets.empty()) { - throw Exception(fmt::format("Can't transfrom kvstore from V2 to V3, [page_id={}] {}", + throw Exception(fmt::format("Can't transform kvstore from V2 to V3, [page_id={}] {}", page.page_id, page_transform_entry.toDebugString()), ErrorCodes::LOGICAL_ERROR); @@ -276,10 +276,12 @@ RegionMap RegionPersister::restore(const TiFlashRaftProxyHelper * proxy_helper, } case PageStorageRunMode::MIX_MODE: { + // The ps v2 instance will be destroyed soon after transform its data to v3, + // so we can safely use some aggressive gc config for it. auto page_storage_v2 = std::make_shared( "RegionPersister", delegator, - config, + PageStorage::getEasyGCConfig(), provider); // V3 should not used getPSDiskDelegatorRaft // Because V2 will delete all invalid(unrecognized) file when it restore @@ -300,7 +302,7 @@ RegionMap RegionPersister::restore(const TiFlashRaftProxyHelper * proxy_helper, LOG_FMT_INFO(log, "Current kvstore transform to V3 begin [pages_before_transform={}]", kvstore_remain_pages); forceTransformKVStoreV2toV3(); const auto & kvstore_remain_pages_after_transform = page_storage_v2->getNumberOfPages(); - LOG_FMT_INFO(log, "Current kvstore transfrom to V3 finished. [ns_id={}] [done={}] [pages_before_transform={}] [pages_after_transform={}]", // + LOG_FMT_INFO(log, "Current kvstore transform to V3 finished. [ns_id={}] [done={}] [pages_before_transform={}] [pages_after_transform={}]", // ns_id, kvstore_remain_pages_after_transform == 0, kvstore_remain_pages, @@ -313,8 +315,10 @@ RegionMap RegionPersister::restore(const TiFlashRaftProxyHelper * proxy_helper, } else // no need do transform { - LOG_FMT_INFO(log, "Current kvstore translate already done before restored."); + LOG_FMT_INFO(log, "Current kvstore transform already done before restored."); } + // running gc on v2 to decrease its disk space usage + page_storage_v2->gcImpl(/*not_skip=*/true, nullptr, nullptr); // change run_mode to ONLY_V3 page_storage_v2 = nullptr; diff --git a/dbms/src/Storages/Transaction/tests/gtest_kvstore.cpp b/dbms/src/Storages/Transaction/tests/gtest_kvstore.cpp index 77aab06f6cf..e157d711f1d 100644 --- a/dbms/src/Storages/Transaction/tests/gtest_kvstore.cpp +++ b/dbms/src/Storages/Transaction/tests/gtest_kvstore.cpp @@ -46,6 +46,7 @@ class RegionKVStoreTest : public ::testing::Test static void testKVStore(); static void testRegion(); static void testReadIndex(); + static void testNewProxy(); private: static void testRaftSplit(KVStore & kvs, TMTContext & tmt); @@ -54,6 +55,66 @@ class RegionKVStoreTest : public ::testing::Test static void testRaftMergeRollback(KVStore & kvs, TMTContext & tmt); }; +void RegionKVStoreTest::testNewProxy() +{ + std::string path = TiFlashTestEnv::getTemporaryPath("/region_kvs_tmp") + "/basic"; + + Poco::File file(path); + if (file.exists()) + file.remove(true); + file.createDirectories(); + + auto ctx = TiFlashTestEnv::getContext( + DB::Settings(), + Strings{ + path, + }); + KVStore & kvs = *ctx.getTMTContext().getKVStore(); + MockRaftStoreProxy proxy_instance; + TiFlashRaftProxyHelper proxy_helper; + { + proxy_helper = MockRaftStoreProxy::SetRaftStoreProxyFFIHelper(RaftStoreProxyPtr{&proxy_instance}); + proxy_instance.init(100); + } + kvs.restore(&proxy_helper); + { + auto store = metapb::Store{}; + store.set_id(1234); + kvs.setStore(store); + ASSERT_EQ(kvs.getStoreID(), store.id()); + } + { + ASSERT_EQ(kvs.getRegion(0), nullptr); + auto task_lock = kvs.genTaskLock(); + auto lock = kvs.genRegionWriteLock(task_lock); + { + auto region = makeRegion(1, RecordKVFormat::genKey(1, 0), RecordKVFormat::genKey(1, 10)); + lock.regions.emplace(1, region); + lock.index.add(region); + } + } + { + kvs.tryPersist(1); + kvs.gcRegionPersistedCache(Seconds{0}); + } + { + // test CompactLog + raft_cmdpb::AdminRequest request; + raft_cmdpb::AdminResponse response; + auto region = kvs.getRegion(1); + region->markCompactLog(); + kvs.setRegionCompactLogConfig(100000, 1000, 1000); + request.mutable_compact_log(); + request.set_cmd_type(::raft_cmdpb::AdminCmdType::CompactLog); + // CompactLog always returns true now, even if we can't do a flush. + // We use a tryFlushData to pre-filter. + ASSERT_EQ(kvs.handleAdminRaftCmd(std::move(request), std::move(response), 1, 5, 1, ctx.getTMTContext()), EngineStoreApplyRes::Persist); + + // Filter + ASSERT_EQ(kvs.tryFlushRegionData(1, false, ctx.getTMTContext()), false); + } +} + void RegionKVStoreTest::testReadIndex() { std::string path = TiFlashTestEnv::getTemporaryPath("/region_kvs_tmp") + "/basic"; @@ -968,20 +1029,17 @@ void RegionKVStoreTest::testKVStore() request.mutable_compact_log(); request.set_cmd_type(::raft_cmdpb::AdminCmdType::CompactLog); - - raft_cmdpb::AdminRequest first_request = request; raft_cmdpb::AdminResponse first_response = response; - - ASSERT_EQ(kvs.handleAdminRaftCmd(std::move(first_request), std::move(first_response), 7, 22, 6, ctx.getTMTContext()), EngineStoreApplyRes::Persist); + ASSERT_EQ(kvs.handleAdminRaftCmd(raft_cmdpb::AdminRequest{request}, std::move(first_response), 7, 22, 6, ctx.getTMTContext()), EngineStoreApplyRes::Persist); raft_cmdpb::AdminResponse second_response = response; - ASSERT_EQ(kvs.handleAdminRaftCmd(raft_cmdpb::AdminRequest{request}, std::move(second_response), 7, 23, 6, ctx.getTMTContext()), EngineStoreApplyRes::None); - request.set_cmd_type(::raft_cmdpb::AdminCmdType::ComputeHash); + ASSERT_EQ(kvs.handleAdminRaftCmd(raft_cmdpb::AdminRequest{request}, std::move(second_response), 7, 23, 6, ctx.getTMTContext()), EngineStoreApplyRes::Persist); + request.set_cmd_type(::raft_cmdpb::AdminCmdType::ComputeHash); raft_cmdpb::AdminResponse third_response = response; ASSERT_EQ(kvs.handleAdminRaftCmd(raft_cmdpb::AdminRequest{request}, std::move(third_response), 7, 24, 6, ctx.getTMTContext()), EngineStoreApplyRes::None); - request.set_cmd_type(::raft_cmdpb::AdminCmdType::VerifyHash); + request.set_cmd_type(::raft_cmdpb::AdminCmdType::VerifyHash); raft_cmdpb::AdminResponse fourth_response = response; ASSERT_EQ(kvs.handleAdminRaftCmd(raft_cmdpb::AdminRequest{request}, std::move(fourth_response), 7, 25, 6, ctx.getTMTContext()), EngineStoreApplyRes::None); @@ -1422,5 +1480,12 @@ try } CATCH +TEST_F(RegionKVStoreTest, NewProxy) +try +{ + testNewProxy(); +} +CATCH + } // namespace tests } // namespace DB diff --git a/dbms/src/Storages/Transaction/tests/gtest_tidb_collator.cpp b/dbms/src/Storages/Transaction/tests/gtest_tidb_collator.cpp index 9a6dae3db08..13c51dba2db 100644 --- a/dbms/src/Storages/Transaction/tests/gtest_tidb_collator.cpp +++ b/dbms/src/Storages/Transaction/tests/gtest_tidb_collator.cpp @@ -13,6 +13,7 @@ // limitations under the License. #include +#include #include namespace DB::tests @@ -66,59 +67,45 @@ const typename CollatorCases::SortKeyCase CollatorCases::sk_cases[] = { {"a", {PREVENT_TRUNC("\x61"), PREVENT_TRUNC("\x61"), PREVENT_TRUNC("\x00\x41"), PREVENT_TRUNC("\x61"), PREVENT_TRUNC("\x0e\x33")}}, {"A", {PREVENT_TRUNC("\x41"), PREVENT_TRUNC("\x41"), PREVENT_TRUNC("\x00\x41"), PREVENT_TRUNC("\x41"), PREVENT_TRUNC("\x0e\x33")}}, {"😃", - {PREVENT_TRUNC("\xf0\x9f\x98\x83"), PREVENT_TRUNC("\xf0\x9f\x98\x83"), PREVENT_TRUNC("\xff\xfd"), PREVENT_TRUNC("\xf0\x9f\x98\x83"), - PREVENT_TRUNC("\xff\xfd")}}, + {PREVENT_TRUNC("\xf0\x9f\x98\x83"), PREVENT_TRUNC("\xf0\x9f\x98\x83"), PREVENT_TRUNC("\xff\xfd"), PREVENT_TRUNC("\xf0\x9f\x98\x83"), PREVENT_TRUNC("\xff\xfd")}}, {"Foo © bar 𝌆 baz ☃ qux", - {PREVENT_TRUNC("\x46\x6f\x6f\x20\xc2\xa9\x20\x62\x61\x72\x20\xf0\x9d\x8c\x86\x20\x62\x61\x7a\x20\xe2\x98\x83\x20\x71\x75\x78"), - PREVENT_TRUNC("\x46\x6f\x6f\x20\xc2\xa9\x20\x62\x61\x72\x20\xf0\x9d\x8c\x86\x20\x62\x61\x7a\x20\xe2\x98\x83\x20\x71\x75\x78"), - PREVENT_TRUNC("\x00\x46\x00\x4f\x00\x4f\x00\x20\x00\xa9\x00\x20\x00\x42\x00\x41\x00\x52\x00\x20\xff\xfd\x00\x20\x00\x42\x00\x41" - "\x00\x5a\x00\x20\x26\x03\x00\x20\x00\x51\x00\x55\x00\x58"), - PREVENT_TRUNC("\x46\x6f\x6f\x20\xc2\xa9\x20\x62\x61\x72\x20\xf0\x9d\x8c\x86\x20\x62\x61\x7a\x20\xe2\x98\x83\x20\x71\x75\x78"), - PREVENT_TRUNC("\x0E\xB9\x0F\x82\x0F\x82\x02\x09\x02\xC5\x02\x09\x0E\x4A\x0E\x33\x0F\xC0\x02\x09\xFF\xFD\x02\x09\x0E\x4A\x0E\x33" - "\x10\x6A\x02\x09\x06\xFF\x02\x09\x0F\xB4\x10\x1F\x10\x5A")}}, + {PREVENT_TRUNC("\x46\x6f\x6f\x20\xc2\xa9\x20\x62\x61\x72\x20\xf0\x9d\x8c\x86\x20\x62\x61\x7a\x20\xe2\x98\x83\x20\x71\x75\x78"), + PREVENT_TRUNC("\x46\x6f\x6f\x20\xc2\xa9\x20\x62\x61\x72\x20\xf0\x9d\x8c\x86\x20\x62\x61\x7a\x20\xe2\x98\x83\x20\x71\x75\x78"), + PREVENT_TRUNC("\x00\x46\x00\x4f\x00\x4f\x00\x20\x00\xa9\x00\x20\x00\x42\x00\x41\x00\x52\x00\x20\xff\xfd\x00\x20\x00\x42\x00\x41" + "\x00\x5a\x00\x20\x26\x03\x00\x20\x00\x51\x00\x55\x00\x58"), + PREVENT_TRUNC("\x46\x6f\x6f\x20\xc2\xa9\x20\x62\x61\x72\x20\xf0\x9d\x8c\x86\x20\x62\x61\x7a\x20\xe2\x98\x83\x20\x71\x75\x78"), + PREVENT_TRUNC("\x0E\xB9\x0F\x82\x0F\x82\x02\x09\x02\xC5\x02\x09\x0E\x4A\x0E\x33\x0F\xC0\x02\x09\xFF\xFD\x02\x09\x0E\x4A\x0E\x33" + "\x10\x6A\x02\x09\x06\xFF\x02\x09\x0F\xB4\x10\x1F\x10\x5A")}}, {"a ", {PREVENT_TRUNC("\x61\x20"), PREVENT_TRUNC("\x61"), PREVENT_TRUNC("\x00\x41"), PREVENT_TRUNC("\x61"), PREVENT_TRUNC("\x0e\x33")}}, {"", {PREVENT_TRUNC(""), PREVENT_TRUNC(""), PREVENT_TRUNC(""), PREVENT_TRUNC(""), PREVENT_TRUNC("")}}, {"ß", - {PREVENT_TRUNC("\xc3\x9f"), PREVENT_TRUNC("\xc3\x9f"), PREVENT_TRUNC("\x00\x53"), PREVENT_TRUNC("\xc3\x9f"), - PREVENT_TRUNC("\x0F\xEA\x0F\xEA")}}, + {PREVENT_TRUNC("\xc3\x9f"), PREVENT_TRUNC("\xc3\x9f"), PREVENT_TRUNC("\x00\x53"), PREVENT_TRUNC("\xc3\x9f"), PREVENT_TRUNC("\x0F\xEA\x0F\xEA")}}, }; const typename CollatorCases::PatternCase CollatorCases::pattern_cases[] = { {"A", - {{"a", {false, false, true, false, true}}, {"A", {true, true, true, true, true}}, {"À", {false, false, true, false, true}}, - {"", {false, false, false, false, false}}}}, + {{"a", {false, false, true, false, true}}, {"A", {true, true, true, true, true}}, {"À", {false, false, true, false, true}}, {"", {false, false, false, false, false}}}}, {"_A", - {{"aA", {true, true, true, true, true}}, {"ÀA", {false, false, true, true, true}}, {"ÀÀ", {false, false, true, false, true}}, - {"", {false, false, false, false, false}}}}, + {{"aA", {true, true, true, true, true}}, {"ÀA", {false, false, true, true, true}}, {"ÀÀ", {false, false, true, false, true}}, {"", {false, false, false, false, false}}}}, {"%A", - {{"a", {false, false, true, false, true}}, {"ÀA", {true, true, true, true, true}}, {"À", {false, false, true, false, true}}, - {"", {false, false, false, false, false}}}}, + {{"a", {false, false, true, false, true}}, {"ÀA", {true, true, true, true, true}}, {"À", {false, false, true, false, true}}, {"", {false, false, false, false, false}}}}, {"À", - {{"a", {false, false, true, false, true}}, {"A", {false, false, true, false, true}}, {"À", {true, true, true, true, true}}, - {"", {false, false, false, false, false}}}}, + {{"a", {false, false, true, false, true}}, {"A", {false, false, true, false, true}}, {"À", {true, true, true, true, true}}, {"", {false, false, false, false, false}}}}, {"_À", - {{" À", {true, true, true, true, true}}, {"ÀA", {false, false, true, false, true}}, {"ÀÀ", {false, false, true, true, true}}, - {"", {false, false, false, false, false}}}}, + {{" À", {true, true, true, true, true}}, {"ÀA", {false, false, true, false, true}}, {"ÀÀ", {false, false, true, true, true}}, {"", {false, false, false, false, false}}}}, {"%À", - {{"À", {true, true, true, true, true}}, {"ÀÀÀ", {true, true, true, true, true}}, {"ÀA", {false, false, true, false, true}}, - {"", {false, false, false, false, false}}}}, + {{"À", {true, true, true, true, true}}, {"ÀÀÀ", {true, true, true, true, true}}, {"ÀA", {false, false, true, false, true}}, {"", {false, false, false, false, false}}}}, {"À_", - {{"À ", {true, true, true, true, true}}, {"ÀAA", {false, false, false, false, false}}, {"À", {false, false, false, false, false}}, - {"", {false, false, false, false, false}}}}, + {{"À ", {true, true, true, true, true}}, {"ÀAA", {false, false, false, false, false}}, {"À", {false, false, false, false, false}}, {"", {false, false, false, false, false}}}}, {"À%", - {{"À", {true, true, true, true, true}}, {"ÀÀÀ", {true, true, true, true, true}}, {"AÀ", {false, false, true, false, true}}, - {"", {false, false, false, false, false}}}}, + {{"À", {true, true, true, true, true}}, {"ÀÀÀ", {true, true, true, true, true}}, {"AÀ", {false, false, true, false, true}}, {"", {false, false, false, false, false}}}}, {"", - {{"À", {false, false, false, false, false}}, {"ÀÀÀ", {false, false, false, false, false}}, - {"AÀ", {false, false, false, false, false}}, {"", {true, true, true, true, true}}}}, + {{"À", {false, false, false, false, false}}, {"ÀÀÀ", {false, false, false, false, false}}, {"AÀ", {false, false, false, false, false}}, {"", {true, true, true, true, true}}}}, {"%", - {{"À", {true, true, true, true, true}}, {"ÀÀÀ", {true, true, true, true, true}}, {"AÀ", {true, true, true, true, true}}, - {"", {true, true, true, true, true}}}}, + {{"À", {true, true, true, true, true}}, {"ÀÀÀ", {true, true, true, true, true}}, {"AÀ", {true, true, true, true, true}}, {"", {true, true, true, true, true}}}}, {"a_%À", - {{"ÀÀ", {false, false, false, false, false}}, {"aÀÀ", {true, true, true, true, true}}, {"ÀÀÀÀ", {false, false, true, false, true}}, - {"ÀÀÀa", {false, false, true, false, true}}}}, + {{"ÀÀ", {false, false, false, false, false}}, {"aÀÀ", {true, true, true, true, true}}, {"ÀÀÀÀ", {false, false, true, false, true}}, {"ÀÀÀa", {false, false, true, false, true}}}}, {"À%_a", - {{"ÀÀ", {false, false, false, false, false}}, {"aÀÀ", {false, false, true, false, true}}, {"ÀÀÀa", {true, true, true, true, true}}, - {"aÀÀÀ", {false, false, true, false, true}}}}, + {{"ÀÀ", {false, false, false, false, false}}, {"aÀÀ", {false, false, true, false, true}}, {"ÀÀÀa", {true, true, true, true, true}}, {"aÀÀÀ", {false, false, true, false, true}}}}, {"___a", {{"中a", {true, true, false, false, false}}, {"中文字a", {false, false, true, true, true}}}}, {"𐐭", {{"𐐨", {false, false, true, false, false}}}}, }; @@ -133,7 +120,7 @@ void testCollator() const std::string & s2 = std::get<1>(c); int ans = std::get(std::get<2>(c)); std::cout << "Compare case (" << s1 << ", " << s2 << ", " << ans << ")" << std::endl; - ASSERT_EQ(collator->compare(s1.data(), s1.length(), s2.data(), s2.length()), ans); + ASSERT_EQ(signum((collator->compare(s1.data(), s1.length(), s2.data(), s2.length()))), ans); } for (const auto & c : CollatorCases::sk_cases) { @@ -189,14 +176,29 @@ struct UnicodeCICollator static constexpr auto collation_case = CollatorCases::UnicodeCI; }; -TEST(CollatorSuite, BinCollator) { testCollator(); } +TEST(CollatorSuite, BinCollator) +{ + testCollator(); +} -TEST(CollatorSuite, BinPaddingCollator) { testCollator(); } +TEST(CollatorSuite, BinPaddingCollator) +{ + testCollator(); +} -TEST(CollatorSuite, Utf8BinPaddingCollator) { testCollator(); } +TEST(CollatorSuite, Utf8BinPaddingCollator) +{ + testCollator(); +} -TEST(CollatorSuite, GeneralCICollator) { testCollator(); } +TEST(CollatorSuite, GeneralCICollator) +{ + testCollator(); +} -TEST(CollatorSuite, UnicodeCICollator) { testCollator(); } +TEST(CollatorSuite, UnicodeCICollator) +{ + testCollator(); +} } // namespace DB::tests diff --git a/dbms/src/TestUtils/ExecutorTestUtils.h b/dbms/src/TestUtils/ExecutorTestUtils.h index 54fed31f88d..01c923f3d7e 100644 --- a/dbms/src/TestUtils/ExecutorTestUtils.h +++ b/dbms/src/TestUtils/ExecutorTestUtils.h @@ -30,6 +30,14 @@ ::testing::AssertionResult check_columns_equality(const ColumnsWithTypeAndName & DB::ColumnsWithTypeAndName readBlock(BlockInputStreamPtr stream); +#define WRAP_FOR_DIS_ENABLE_PLANNER_BEGIN \ + std::vector bools{false, true}; \ + for (auto flag : bools) \ + { \ + enablePlanner(flag); + +#define WRAP_FOR_DIS_ENABLE_PLANNER_END } + class ExecutorTest : public ::testing::Test { protected: @@ -51,16 +59,7 @@ class ExecutorTest : public ::testing::Test DAGContext & getDAGContext(); - /// for planner void enablePlanner(bool is_enable); - template - void wrapForDisEnablePlanner(FF && ff) - { - enablePlanner(false); - ff(); - enablePlanner(true); - ff(); - } static void dagRequestEqual(const String & expected_string, const std::shared_ptr & actual); diff --git a/dbms/src/TestUtils/gtests_dbms_main.cpp b/dbms/src/TestUtils/gtests_dbms_main.cpp index 26c456b5b31..9c53ccb9084 100644 --- a/dbms/src/TestUtils/gtests_dbms_main.cpp +++ b/dbms/src/TestUtils/gtests_dbms_main.cpp @@ -13,6 +13,9 @@ // limitations under the License. #include +#include +#include +#include #include namespace DB::FailPoints @@ -25,6 +28,10 @@ int main(int argc, char ** argv) { DB::tests::TiFlashTestEnv::setupLogger(); DB::tests::TiFlashTestEnv::initializeGlobalContext(); + DB::ServerInfo server_info; + DB::DM::SegmentReaderPoolManager::instance().init(server_info); + DB::DM::SegmentReadTaskScheduler::instance(); + DB::DM::DMFileReaderPool::instance(); #ifdef FIU_ENABLE fiu_init(0); // init failpoint diff --git a/dbms/src/TestUtils/mockExecutor.cpp b/dbms/src/TestUtils/mockExecutor.cpp index cc8160761e6..c5c25d1adff 100644 --- a/dbms/src/TestUtils/mockExecutor.cpp +++ b/dbms/src/TestUtils/mockExecutor.cpp @@ -270,8 +270,10 @@ DAGRequestBuilder & DAGRequestBuilder::aggregation(ASTPtr agg_func, ASTPtr group { auto agg_funcs = std::make_shared(); auto group_by_exprs = std::make_shared(); - agg_funcs->children.push_back(agg_func); - group_by_exprs->children.push_back(group_by_expr); + if (agg_func) + agg_funcs->children.push_back(agg_func); + if (group_by_expr) + group_by_exprs->children.push_back(group_by_expr); return buildAggregation(agg_funcs, group_by_exprs); } diff --git a/dbms/src/TestUtils/mockExecutor.h b/dbms/src/TestUtils/mockExecutor.h index db0dbef929e..317b2e362fb 100644 --- a/dbms/src/TestUtils/mockExecutor.h +++ b/dbms/src/TestUtils/mockExecutor.h @@ -176,8 +176,13 @@ MockWindowFrame buildDefaultRowsFrame(); #define And(expr1, expr2) makeASTFunction("and", (expr1), (expr2)) #define Or(expr1, expr2) makeASTFunction("or", (expr1), (expr2)) #define NOT(expr) makeASTFunction("not", (expr)) + +// Aggregation functions #define Max(expr) makeASTFunction("max", (expr)) +#define Min(expr) makeASTFunction("min", (expr)) +#define Count(expr) makeASTFunction("count", (expr)) #define Sum(expr) makeASTFunction("sum", (expr)) + /// Window functions #define RowNumber() makeASTFunction("RowNumber") #define Rank() makeASTFunction("Rank") diff --git a/dbms/src/TiDB/Schema/SchemaBuilder.cpp b/dbms/src/TiDB/Schema/SchemaBuilder.cpp index 93a36d8e3ee..3486895d7f2 100644 --- a/dbms/src/TiDB/Schema/SchemaBuilder.cpp +++ b/dbms/src/TiDB/Schema/SchemaBuilder.cpp @@ -1323,6 +1323,7 @@ void SchemaBuilder::applySetTiFlashModeOnPhysicalTable( TiDB::TableInfo table_info = storage->getTableInfo(); table_info.tiflash_mode = latest_table_info->tiflash_mode; + table_info.schema_version = target_version; AlterCommands commands; LOG_FMT_INFO(log, "Updating tiflash mode for {} to {}", name_mapper.debugCanonicalName(*db_info, table_info), TiFlashModeToString(table_info.tiflash_mode)); diff --git a/libs/libcommon/CMakeLists.txt b/libs/libcommon/CMakeLists.txt index 2bedb312d07..536cc97081f 100644 --- a/libs/libcommon/CMakeLists.txt +++ b/libs/libcommon/CMakeLists.txt @@ -100,19 +100,6 @@ add_library (common ${SPLIT_SHARED} if (USE_JEMALLOC) message (STATUS "Link jemalloc: ${JEMALLOC_LIBRARIES}") set (MALLOC_LIBRARIES ${JEMALLOC_LIBRARIES}) -elseif (USE_TCMALLOC) - - if (DEBUG_TCMALLOC AND NOT GPERFTOOLS_TCMALLOC_MINIMAL_DEBUG) - message (WARNING "Requested DEBUG_TCMALLOC but debug lib not found. try install libgoogle-perftools-dev") - endif () - - if (DEBUG_TCMALLOC AND GPERFTOOLS_TCMALLOC_MINIMAL_DEBUG) - message (STATUS "Link libtcmalloc_minimal_debug for testing: ${GPERFTOOLS_TCMALLOC_MINIMAL_DEBUG}") - set (MALLOC_LIBRARIES ${GPERFTOOLS_TCMALLOC_MINIMAL_DEBUG}) - else () - message (STATUS "Link libtcmalloc_minimal: ${GPERFTOOLS_TCMALLOC_MINIMAL}") - set (MALLOC_LIBRARIES ${GPERFTOOLS_TCMALLOC_MINIMAL}) - endif () elseif(USE_MIMALLOC) message (STATUS "Link mimalloc: ${MIMALLOC_LIBRARIES}") set (MALLOC_LIBRARIES ${MIMALLOC_LIBRARIES}) @@ -136,6 +123,12 @@ target_include_directories (common PUBLIC ${COMMON_INCLUDE_DIR}) target_include_directories (common BEFORE PUBLIC ${Boost_INCLUDE_DIRS}) +if(APPLE AND ARCH_AARCH64) + set(CPU_FEATURES_LIBRARY) +else() + set(CPU_FEATURES_LIBRARY cpu_features) +endif() + # `libcommon` provides a bridge for many other libs. # We expose the linkage to public because we want headers to be inherited. target_link_libraries ( @@ -151,7 +144,7 @@ target_link_libraries ( ${GLIBC_COMPATIBILITY_LIBRARIES} ${MEMCPY_LIBRARIES} fmt - cpu_features + ${CPU_FEATURES_LIBRARY} ) if (RT_LIBRARY) diff --git a/libs/libcommon/cmake/find_gperftools.cmake b/libs/libcommon/cmake/find_gperftools.cmake deleted file mode 100644 index 98016f0e818..00000000000 --- a/libs/libcommon/cmake/find_gperftools.cmake +++ /dev/null @@ -1,50 +0,0 @@ -# Copyright 2022 PingCAP, Ltd. -# -# Licensed under the Apache License, Version 2.0 (the "License"); -# you may not use this file except in compliance with the License. -# You may obtain a copy of the License at -# -# http://www.apache.org/licenses/LICENSE-2.0 -# -# Unless required by applicable law or agreed to in writing, software -# distributed under the License is distributed on an "AS IS" BASIS, -# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. -# See the License for the specific language governing permissions and -# limitations under the License. - -if (ARCH_FREEBSD OR ARCH_32) - option (USE_INTERNAL_GPERFTOOLS_LIBRARY "Set to FALSE to use system gperftools (tcmalloc) library instead of bundled" OFF) -else () - option (USE_INTERNAL_GPERFTOOLS_LIBRARY "Set to FALSE to use system gperftools (tcmalloc) library instead of bundled" ${NOT_UNBUNDLED}) -endif () - -if(${CMAKE_SYSTEM_NAME} MATCHES "Linux") - set(ENABLE_TCMALLOC_DEFAULT 0) -elseif(${CMAKE_SYSTEM_NAME} MATCHES "FreeBSD") - set(ENABLE_TCMALLOC_DEFAULT 1) -elseif(${CMAKE_SYSTEM_NAME} MATCHES "Darwin") - set(ENABLE_TCMALLOC_DEFAULT 1) -endif() - -option (ENABLE_TCMALLOC "Set to TRUE to enable tcmalloc" ${ENABLE_TCMALLOC_DEFAULT}) - -option (DEBUG_TCMALLOC "Set to TRUE to use debug version of libtcmalloc" OFF) - -if (ENABLE_TCMALLOC) - #contrib/libtcmalloc doesnt build debug version, try find in system - if (DEBUG_TCMALLOC OR NOT USE_INTERNAL_GPERFTOOLS_LIBRARY) - find_package (Gperftools) - endif () - - if (NOT (GPERFTOOLS_FOUND AND GPERFTOOLS_INCLUDE_DIR AND GPERFTOOLS_TCMALLOC_MINIMAL) AND NOT (ARCH_FREEBSD OR ARCH_32)) - set (USE_INTERNAL_GPERFTOOLS_LIBRARY 1) - set (GPERFTOOLS_INCLUDE_DIR "${TiFlash_SOURCE_DIR}/contrib/libtcmalloc/include") - set (GPERFTOOLS_TCMALLOC_MINIMAL tcmalloc_minimal_internal) - endif () - - if (GPERFTOOLS_FOUND OR USE_INTERNAL_GPERFTOOLS_LIBRARY) - set (USE_TCMALLOC 1) - endif () - - message (STATUS "Using tcmalloc=${USE_TCMALLOC}: ${GPERFTOOLS_INCLUDE_DIR} : ${GPERFTOOLS_TCMALLOC_MINIMAL}") -endif () diff --git a/libs/libcommon/cmake/find_mimalloc.cmake b/libs/libcommon/cmake/find_mimalloc.cmake index 92d50a199f0..1468361a912 100644 --- a/libs/libcommon/cmake/find_mimalloc.cmake +++ b/libs/libcommon/cmake/find_mimalloc.cmake @@ -19,7 +19,7 @@ if (ENABLE_MIMALLOC) set (USE_MIMALLOC 1) message (STATUS "Using mimalloc=${USE_MIMALLOC}: ${MIMALLOC_LIBRARIES}") - if (ENABLE_JEMALLOC OR ENABLE_TCMALLOC) + if (ENABLE_JEMALLOC) message(FATAL_ERROR "multiple global allocator detected!") endif() endif () diff --git a/libs/libcommon/include/common/config_common.h.in b/libs/libcommon/include/common/config_common.h.in index 0ec263216f4..6f6dc90168c 100644 --- a/libs/libcommon/include/common/config_common.h.in +++ b/libs/libcommon/include/common/config_common.h.in @@ -3,7 +3,6 @@ // .h autogenerated by cmake ! #cmakedefine01 APPLE_SIERRA_OR_NEWER -#cmakedefine01 USE_TCMALLOC #cmakedefine01 USE_JEMALLOC #cmakedefine01 USE_JEMALLOC_PROF #cmakedefine01 USE_MIMALLOC diff --git a/libs/libcommon/include/common/detect_features.h b/libs/libcommon/include/common/detect_features.h index 0291015ead4..c8d0c238209 100644 --- a/libs/libcommon/include/common/detect_features.h +++ b/libs/libcommon/include/common/detect_features.h @@ -15,6 +15,7 @@ #pragma once #if defined(__aarch64__) || defined(__arm64__) || defined(__arm64) || defined(__ARM64) || defined(__AARCH64__) +#ifndef __APPLE__ #include namespace common { @@ -24,6 +25,42 @@ using CPUInfo = cpu_features::Aarch64Info; extern const CPUInfo cpu_info; static inline const CPUFeatures & cpu_feature_flags = cpu_info.features; } // namespace common +#else +#include +#include +namespace common +{ +static inline int detect(const char * name) +{ + int enabled; + size_t enabled_len = sizeof(enabled); + const int failure = ::sysctlbyname(name, &enabled, &enabled_len, NULL, 0); + return failure ? 0 : enabled; +} +static inline const struct CPUFeatures +{ + bool asimd = detect("hw.optional.neon"); + bool sve = false; // currently not supported + bool sve2 = false; // currently not supported + + bool atomics = detect("hw.optional.armv8_1_atomics"); + bool hpfp = detect("hw.optional.neon_hpfp"); + bool fp16 = detect("hw.optional.neon_fp16"); + bool fhm = detect("hw.optional.armv8_2_fhm"); + bool crc32 = detect("hw.optional.armv8_crc32"); + bool sha512 = detect("hw.optional.armv8_2_sha512"); + bool sha3 = detect("hw.optional.armv8_2_sha3"); + + // the followings are assumed to be true. + // see https://github.com/golang/sys/pull/114/files + bool cpuid = true; + bool aes = true; + bool pmull = true; + bool sha1 = true; + bool sha2 = true; +} cpu_feature_flags; +} // namespace common +#endif #endif #if defined(__x86_64__) || defined(__x86_64) || defined(__amd64) || defined(__amd64__) diff --git a/tests/fullstack-test/expr/like.test b/tests/fullstack-test/expr/like.test index af76095d157..5f13c7a8159 100644 --- a/tests/fullstack-test/expr/like.test +++ b/tests/fullstack-test/expr/like.test @@ -38,4 +38,29 @@ mysql> set @@tidb_isolation_read_engines='tiflash'; select * from test.t where ' | a | b | +------+------+ | aaaa | %a% | -+------+------+ \ No newline at end of file ++------+------+ + +mysql> insert into test.t values ('1234', ''); + +mysql> set @@tidb_isolation_read_engines='tiflash'; select a from test.t where a like '1234' escape '4'; ++------+ +| a | ++------+ +| 1234 | ++------+ + +mysql> set @@tidb_isolation_read_engines='tiflash'; select a from test.t where a like '1234' escape '2'; + +mysql> set @@tidb_isolation_read_engines='tiflash'; select a from test.t where a like '15234' escape '5'; ++------+ +| a | ++------+ +| 1234 | ++------+ + +mysql> set @@tidb_isolation_read_engines='tiflash'; select a from test.t where a like '_223_' escape '2'; ++------+ +| a | ++------+ +| 1234 | ++------+ diff --git a/tests/fullstack-test2/ddl/alter_table_tiflash_replica_and_mode.test b/tests/fullstack-test2/ddl/alter_table_tiflash_replica_and_mode.test index 684948145ba..0d22a8a242e 100644 --- a/tests/fullstack-test2/ddl/alter_table_tiflash_replica_and_mode.test +++ b/tests/fullstack-test2/ddl/alter_table_tiflash_replica_and_mode.test @@ -142,4 +142,27 @@ mysql> set session tidb_isolation_read_engines='tiflash'; select * from test.t; | 1 | 2| | 2 | 4| | 5 | 7| -+------+-----+ \ No newline at end of file ++------+-----+ + +## test the case across 'do query' and 'set tiflash mode' + +=> DBGInvoke __enable_schema_sync_service('false') + +## TODO: case1 can not simulation in this script +# case 1 : start query (set failpoint) --> set tiflash mode and refresh schema --> continue query, and due to query_version < local_schema, query fail + +# case 2 : set tiflash mode first, then do query --> local_schema < query_schema_version --> sync schema and do query in the new mode +mysql> alter table test.t set tiflash mode normal; + +# query in normal mode +mysql> set session tidb_isolation_read_engines='tiflash'; select * from test.t; ++------+-----+ +| a | b | ++------+-----+ +| 2 | 4| +| 5 | 7| ++------+-----+ + + +=> DBGInvoke __enable_schema_sync_service('true') + diff --git a/tests/tidb-ci/new_collation_fullstack/expr.test b/tests/tidb-ci/new_collation_fullstack/expr.test index 1e2135c4f2d..c88db20ba2f 100644 --- a/tests/tidb-ci/new_collation_fullstack/expr.test +++ b/tests/tidb-ci/new_collation_fullstack/expr.test @@ -88,4 +88,20 @@ mysql> set session tidb_isolation_read_engines='tiflash'; select /*+ read_from_s | min(value) | max(value) | min(value1) | max(value1) | +------------+------------+-------------+-------------+ | abc | def | abc | def | -+------------+------------+-------------+-------------+ \ No newline at end of file ++------------+------------+-------------+-------------+ + +mysql> insert into test.t values (4, '', 'def\n'), (5, '', 'def '); + +mysql> select /*+ read_from_storage(tiflash[t]) */ hex(max(value1)) from test.t; ++------------------+ +| hex(max(value1)) | ++------------------+ +| 6465660A | ++------------------+ + +mysql> select /*+ read_from_storage(tiflash[t]) */ hex(min(value1)) from test.t; ++------------------+ +| hex(min(value1)) | ++------------------+ +| 61626320 | ++------------------+