Add NUMA and thread affinity support for Unix (#10938)

* Add NUMA and thread affinity support for Unix

This change adds new PAL functions for NUMA and thread affinity support
for Unix and also enables related code in GC and VM for FEATURE_PAL.

It doesn't reflect the limits imposed by CGROUPS on systems with
CGROUPS enables yet.

Author: janvorli

src/dlls/mscordac/mscordac_unixexports.src

@@ -21,6 +21,7 @@ PAL_GetPALDirectoryW
 PAL_GetResourceString
 PAL_get_stdout
 PAL_get_stderr
+PAL_GetCurrentThread
 PAL_GetSymbolModuleBase
 PAL_GetTransportPipeName
 PAL_InitializeDLL

src/gc/gc.cpp

@@ -5151,7 +5151,6 @@ void gc_heap::destroy_thread_support ()
     }
 }
 
-#if !defined(FEATURE_PAL)
 void set_thread_group_affinity_for_heap(int heap_number, GCThreadAffinity* affinity)
 {
     affinity->Group = GCThreadAffinity::None;
@@ -5231,7 +5230,6 @@ void set_thread_affinity_mask_for_heap(int heap_number, GCThreadAffinity* affini
         }
     }
 }
-#endif // !FEATURE_PAL
 
 bool gc_heap::create_gc_thread ()
 {
@@ -5241,7 +5239,6 @@ bool gc_heap::create_gc_thread ()
     affinity.Group = GCThreadAffinity::None;
     affinity.Processor = GCThreadAffinity::None;
 
-#if !defined(FEATURE_PAL)
     if (!gc_thread_no_affinitize_p)
     {
         // We are about to set affinity for GC threads. It is a good place to set up NUMA and
@@ -5252,7 +5249,6 @@ bool gc_heap::create_gc_thread ()
         else
             set_thread_affinity_mask_for_heap(heap_number, &affinity);
     }
-#endif // !FEATURE_PAL
 
     return GCToOSInterface::CreateThread(gc_thread_stub, this, &affinity);
 }

src/inc/utilcode.h

@@ -1380,7 +1380,7 @@ class NumaNodeInfo
     static BOOL CanEnableGCNumaAware();
     static void InitNumaNodeInfo();
 
-#if !defined(FEATURE_REDHAWK)&& !defined(FEATURE_PAL)
+#if !defined(FEATURE_REDHAWK)
 private:	// apis types
 
     //GetNumaHighestNodeNumber()
@@ -1448,7 +1448,7 @@ class CPUGroupInfo
     static DWORD CalculateCurrentProcessorNumber();
     //static void PopulateCPUUsageArray(void * infoBuffer, ULONG infoSize);
 
-#if !defined(FEATURE_REDHAWK) && !defined(FEATURE_PAL)
+#if !defined(FEATURE_REDHAWK)
 private:
     //GetLogicalProcessorInforomationEx()
     typedef BOOL

src/pal/inc/pal.h

@@ -4825,6 +4825,161 @@ RegisterEventSourceW (
 #define RegisterEventSource  RegisterEventSourceA
 #endif // !UNICODE
 
+//
+// NUMA related APIs
+//
+
+typedef enum _PROCESSOR_CACHE_TYPE { 
+  CacheUnified,
+  CacheInstruction,
+  CacheData,
+  CacheTrace
+} PROCESSOR_CACHE_TYPE;
+
+typedef struct _PROCESSOR_NUMBER {
+  WORD Group;
+  BYTE Number;
+  BYTE Reserved;
+} PROCESSOR_NUMBER, *PPROCESSOR_NUMBER;
+
+typedef enum _LOGICAL_PROCESSOR_RELATIONSHIP { 
+  RelationProcessorCore,
+  RelationNumaNode,
+  RelationCache,
+  RelationProcessorPackage,
+  RelationGroup,
+  RelationAll               = 0xffff
+} LOGICAL_PROCESSOR_RELATIONSHIP;
+
+typedef ULONG_PTR KAFFINITY;
+
+#define ANYSIZE_ARRAY 1
+
+typedef struct _GROUP_AFFINITY {
+  KAFFINITY Mask;
+  WORD      Group;
+  WORD      Reserved[3];
+} GROUP_AFFINITY, *PGROUP_AFFINITY;
+
+typedef struct _PROCESSOR_GROUP_INFO {
+  BYTE      MaximumProcessorCount;
+  BYTE      ActiveProcessorCount;
+  BYTE      Reserved[38];
+  KAFFINITY ActiveProcessorMask;
+} PROCESSOR_GROUP_INFO, *PPROCESSOR_GROUP_INFO;
+
+typedef struct _PROCESSOR_RELATIONSHIP {
+  BYTE           Flags;
+  BYTE           EfficiencyClass;
+  BYTE           Reserved[21];
+  WORD           GroupCount;
+  GROUP_AFFINITY GroupMask[ANYSIZE_ARRAY];
+} PROCESSOR_RELATIONSHIP, *PPROCESSOR_RELATIONSHIP;
+
+typedef struct _GROUP_RELATIONSHIP {
+  WORD                 MaximumGroupCount;
+  WORD                 ActiveGroupCount;
+  BYTE                 Reserved[20];
+  PROCESSOR_GROUP_INFO GroupInfo[ANYSIZE_ARRAY];
+} GROUP_RELATIONSHIP, *PGROUP_RELATIONSHIP;
+
+typedef struct _NUMA_NODE_RELATIONSHIP {
+  DWORD          NodeNumber;
+  BYTE           Reserved[20];
+  GROUP_AFFINITY GroupMask;
+} NUMA_NODE_RELATIONSHIP, *PNUMA_NODE_RELATIONSHIP;
+
+typedef struct _CACHE_RELATIONSHIP {
+  BYTE                 Level;
+  BYTE                 Associativity;
+  WORD                 LineSize;
+  DWORD                CacheSize;
+  PROCESSOR_CACHE_TYPE Type;
+  BYTE                 Reserved[20];
+  GROUP_AFFINITY       GroupMask;
+} CACHE_RELATIONSHIP, *PCACHE_RELATIONSHIP;
+
+typedef struct _SYSTEM_LOGICAL_PROCESSOR_INFORMATION_EX {
+  LOGICAL_PROCESSOR_RELATIONSHIP Relationship;
+  DWORD                          Size;
+  union {
+    PROCESSOR_RELATIONSHIP Processor;
+    NUMA_NODE_RELATIONSHIP NumaNode;
+    CACHE_RELATIONSHIP     Cache;
+    GROUP_RELATIONSHIP     Group;
+  };
+} SYSTEM_LOGICAL_PROCESSOR_INFORMATION_EX, *PSYSTEM_LOGICAL_PROCESSOR_INFORMATION_EX;
+
+
+PALIMPORT
+BOOL
+PALAPI
+GetNumaHighestNodeNumber(
+  OUT PULONG HighestNodeNumber
+);
+
+PALIMPORT
+BOOL
+PALAPI
+GetNumaProcessorNodeEx(
+  IN  PPROCESSOR_NUMBER Processor,
+  OUT PUSHORT NodeNumber
+);
+
+PALIMPORT
+LPVOID
+PALAPI
+VirtualAllocExNuma(
+  IN HANDLE hProcess,
+  IN OPTIONAL LPVOID lpAddress,
+  IN SIZE_T dwSize,
+  IN DWORD flAllocationType,
+  IN DWORD flProtect,
+  IN DWORD nndPreferred
+);
+
+PALIMPORT
+BOOL
+PALAPI
+GetLogicalProcessorInformationEx(
+  IN LOGICAL_PROCESSOR_RELATIONSHIP RelationshipType,
+  OUT OPTIONAL PSYSTEM_LOGICAL_PROCESSOR_INFORMATION_EX Buffer,
+  IN OUT PDWORD ReturnedLength
+);
+
+PALIMPORT
+BOOL
+PALAPI
+SetThreadGroupAffinity(
+  IN HANDLE hThread,
+  IN const GROUP_AFFINITY *GroupAffinity,
+  OUT OPTIONAL PGROUP_AFFINITY PreviousGroupAffinity
+);
+
+PALIMPORT
+BOOL
+PALAPI
+GetThreadGroupAffinity(
+  IN HANDLE hThread,
+  OUT PGROUP_AFFINITY GroupAffinity
+);
+
+PALIMPORT
+VOID
+PALAPI
+GetCurrentProcessorNumberEx(
+  OUT PPROCESSOR_NUMBER ProcNumber
+);
+
+PALIMPORT
+BOOL
+PALAPI
+GetProcessAffinityMask(
+  IN HANDLE hProcess,
+  OUT PDWORD_PTR lpProcessAffinityMask,
+  OUT PDWORD_PTR lpSystemAffinityMask
+);
+
 //
 // The types of events that can be logged.
 //

src/pal/src/CMakeLists.txt

@@ -181,6 +181,7 @@ set(SOURCES
   misc/sysinfo.cpp
   misc/time.cpp
   misc/utils.cpp
+  numa/numa.cpp
   objmgr/palobjbase.cpp
   objmgr/shmobject.cpp
   objmgr/shmobjectmanager.cpp
@@ -372,6 +373,12 @@ if(CMAKE_SYSTEM_NAME STREQUAL NetBSD)
   )
 endif(CMAKE_SYSTEM_NAME STREQUAL NetBSD)
 
+if(HAVE_NUMA_H)
+  target_link_libraries(coreclrpal
+    numa
+  )
+endif(HAVE_NUMA_H)
+
 add_subdirectory(examples)
 
 if(FEATURE_EVENT_TRACE)

src/pal/src/config.h.in

@@ -21,6 +21,8 @@
 #cmakedefine01 HAVE_SYS_SYSCTL_H
 #cmakedefine01 HAVE_GNU_LIBNAMES_H
 #cmakedefine01 HAVE_PRCTL_H
+#cmakedefine01 HAVE_NUMA_H
+#cmakedefine01 HAVE_PTHREAD_NP_H
 
 #cmakedefine01 HAVE_KQUEUE
 #cmakedefine01 HAVE_GETPWUID_R
@@ -33,6 +35,8 @@
 #cmakedefine01 HAVE_PTHREAD_GETATTR_NP
 #cmakedefine01 HAVE_PTHREAD_GETCPUCLOCKID
 #cmakedefine01 HAVE_PTHREAD_SIGQUEUE
+#cmakedefine01 HAVE_PTHREAD_GETAFFINITY_NP
+#cmakedefine01 HAVE_CPUSET_T
 #cmakedefine01 HAVE_SIGRETURN
 #cmakedefine01 HAVE__THREAD_SYS_SIGRETURN
 #cmakedefine01 HAVE_COPYSIGN
@@ -57,6 +61,7 @@
 #cmakedefine01 HAS_SYSV_SEMAPHORES
 #cmakedefine01 HAS_PTHREAD_MUTEXES
 #cmakedefine01 HAVE_TTRACE
+#cmakedefine01 HAVE_SCHED_GETAFFINITY
 #cmakedefine HAVE_UNW_GET_SAVE_LOC
 #cmakedefine HAVE_UNW_GET_ACCESSORS
 #cmakedefine01 HAVE_XSWDEV

src/pal/src/configure.cmake

@@ -35,6 +35,8 @@ check_include_files(libunwind.h HAVE_LIBUNWIND_H)
 check_include_files(runetype.h HAVE_RUNETYPE_H)
 check_include_files(semaphore.h HAVE_SEMAPHORE_H)
 check_include_files(sys/prctl.h HAVE_PRCTL_H)
+check_include_files(numa.h HAVE_NUMA_H)
+check_include_files(pthread_np.h HAVE_PTHREAD_NP_H)
 
 if(NOT CMAKE_SYSTEM_NAME STREQUAL FreeBSD AND NOT CMAKE_SYSTEM_NAME STREQUAL NetBSD)
   set(CMAKE_REQUIRED_FLAGS "-ldl")
@@ -69,6 +71,7 @@ check_library_exists(${PTHREAD_LIBRARY} pthread_attr_get_np "" HAVE_PTHREAD_ATTR
 check_library_exists(${PTHREAD_LIBRARY} pthread_getattr_np "" HAVE_PTHREAD_GETATTR_NP)
 check_library_exists(${PTHREAD_LIBRARY} pthread_getcpuclockid "" HAVE_PTHREAD_GETCPUCLOCKID)
 check_library_exists(${PTHREAD_LIBRARY} pthread_sigqueue "" HAVE_PTHREAD_SIGQUEUE)
+check_library_exists(${PTHREAD_LIBRARY} pthread_getaffinity_np "" HAVE_PTHREAD_GETAFFINITY_NP)
 
 check_function_exists(sigreturn HAVE_SIGRETURN)
 check_function_exists(_thread_sys_sigreturn HAVE__THREAD_SYS_SIGRETURN)
@@ -119,6 +122,14 @@ int main(int argc, char **argv) {
 }" HAVE_UNW_GET_ACCESSORS)
 set(CMAKE_REQUIRED_LIBRARIES)
 
+check_cxx_source_compiles("
+#include <pthread_np.h>
+int main(int argc, char **argv) {
+  cpuset_t cpuSet;
+
+  return 0;
+}" HAVE_CPUSET_T)
+
 check_struct_has_member ("struct stat" st_atimespec "sys/types.h;sys/stat.h" HAVE_STAT_TIMESPEC)
 check_struct_has_member ("struct stat" st_atimensec "sys/types.h;sys/stat.h" HAVE_STAT_NSEC)
 check_struct_has_member ("struct tm" tm_gmtoff time.h HAVE_TM_GMTOFF)

src/pal/src/include/pal/dbgmsg.h

@@ -194,7 +194,7 @@ typedef enum
 #ifdef FEATURE_PAL_SXS
     DCI_SXS,
 #endif // FEATURE_PAL_SXS
-
+    DCI_NUMA,
     DCI_LAST
 } DBG_CHANNEL_ID;
 

src/pal/src/include/pal/numa.h

@@ -0,0 +1,39 @@
+// Licensed to the .NET Foundation under one or more agreements.
+// The .NET Foundation licenses this file to you under the MIT license.
+// See the LICENSE file in the project root for more information.
+
+/*++
+
+
+
+Module Name:
+
+    include/pal/numa.h
+
+Abstract:
+    
+    Header file for the NUMA functions.
+    
+
+
+--*/
+
+#ifndef _PAL_NUMA_H_
+#define _PAL_NUMA_H_
+
+#ifdef __cplusplus
+extern "C"
+{
+#endif // __cplusplus
+
+BOOL
+NUMASupportInitialize();
+
+VOID
+NUMASupportCleanup();
+
+#ifdef __cplusplus
+}
+#endif // __cplusplus
+
+#endif /* _PAL_CRITSECT_H_ */

src/pal/src/init/pal.cpp

@@ -42,6 +42,7 @@ SET_DEFAULT_DEBUG_CHANNEL(PAL); // some headers have code with asserts, so do th
 #include "pal/debug.h"
 #include "pal/locale.h"
 #include "pal/init.h"
+#include "pal/numa.h"
 #include "pal/stackstring.hpp"
 
 #if HAVE_MACH_EXCEPTIONS
@@ -523,6 +524,12 @@ Initialize(
             goto CLEANUP15;
         }
 
+        if (FALSE == NUMASupportInitialize())
+        {
+            ERROR("Unable to initialize NUMA support\n");
+            goto CLEANUP15;
+        }
+
         TRACE("First-time PAL initialization complete.\n");
         init_count++;        
 
@@ -548,6 +555,7 @@ Initialize(
     }
     goto done;
 
+    NUMASupportCleanup();
     /* No cleanup required for CRTInitStdStreams */ 
 CLEANUP15:
     FILECleanupStdHandles();