Consolidate runtime helpers (X86Base, Interlocked, Buffer) across CoreCLR and NativeAOT

src/coreclr/System.Private.CoreLib/System.Private.CoreLib.csproj

@@ -224,7 +224,6 @@
     <Compile Include="$(BclSourcesRoot)\System\Runtime\InteropServices\MemoryMarshal.CoreCLR.cs" />
     <Compile Include="$(BclSourcesRoot)\System\Runtime\InteropServices\NativeLibrary.CoreCLR.cs" />
     <Compile Include="$(BclSourcesRoot)\System\Runtime\InteropServices\Java\JavaMarshal.CoreCLR.cs" Condition="'$(FeatureJavaMarshal)' == 'true'" />
-    <Compile Include="$(BclSourcesRoot)\System\Runtime\Intrinsics\X86\X86Base.CoreCLR.cs" />
     <Compile Include="$(BclSourcesRoot)\System\Runtime\Loader\AssemblyLoadContext.CoreCLR.cs" />
     <Compile Include="$(BclSourcesRoot)\System\RuntimeArgumentHandle.cs" />
     <Compile Include="$(BclSourcesRoot)\System\RuntimeHandles.cs" />

src/coreclr/System.Private.CoreLib/src/System/Buffer.CoreCLR.cs

@@ -9,12 +9,6 @@ namespace System
 {
     public partial class Buffer
     {
-        [LibraryImport(RuntimeHelpers.QCall, EntryPoint = "Buffer_Clear")]
-        private static unsafe partial void ZeroMemoryInternal(void* b, nuint byteLength);
-
-        [LibraryImport(RuntimeHelpers.QCall, EntryPoint = "Buffer_MemMove")]
-        private static unsafe partial void MemmoveInternal(byte* dest, byte* src, nuint len);
-
         [MethodImpl(MethodImplOptions.InternalCall)]
         private static extern void BulkMoveWithWriteBarrierInternal(ref byte destination, ref byte source, nuint byteCount);
 

src/coreclr/System.Private.CoreLib/src/System/Threading/Interlocked.CoreCLR.cs

@@ -248,13 +248,5 @@ private static long ExchangeAdd(ref long location1, long value)
         public static long Read(ref readonly long location) =>
             CompareExchange(ref Unsafe.AsRef(in location), 0, 0);
         #endregion
-
-        #region MemoryBarrierProcessWide
-        [LibraryImport(RuntimeHelpers.QCall, EntryPoint = "Interlocked_MemoryBarrierProcessWide")]
-        private static partial void _MemoryBarrierProcessWide();
-
-        /// <summary>Provides a process-wide memory barrier that ensures that reads and writes from any CPU cannot move across the barrier.</summary>
-        public static void MemoryBarrierProcessWide() => _MemoryBarrierProcessWide();
-        #endregion
     }
 }

src/coreclr/nativeaot/Runtime/CMakeLists.txt

@@ -5,6 +5,7 @@ set(COMMON_RUNTIME_SOURCES
     rhassert.cpp
     ${RUNTIME_DIR}/CachedInterfaceDispatch.cpp
     CachedInterfaceDispatch_Aot.cpp
+    ${RUNTIME_DIR}/MiscNativeHelpers.cpp
     Crst.cpp
     DebugHeader.cpp
     MethodTable.cpp

src/coreclr/nativeaot/Runtime/CachedInterfaceDispatchPal.h

@@ -16,7 +16,7 @@
 #include "Crst.h"
 #include "TargetPtrs.h"
 #include "MethodTable.h"
-#include "Range.h"
+
 #include "allocheap.h"
 #include "rhbinder.h"
 #include "ObjectLayout.h"

src/coreclr/nativeaot/Runtime/CommonMacros.h

@@ -63,7 +63,7 @@ inline bool IS_ALIGNED(T* val, uintptr_t alignment);
 #ifndef DACCESS_COMPILE
 
 #ifndef ZeroMemory
-#define ZeroMemory(_dst, _size) memset((_dst), 0, (_size))
+#define ZeroMemory(Destination,Length) memset((Destination),0,(Length))
 #endif
 
 #endif // !DACCESS_COMPILE
@@ -83,12 +83,6 @@ inline bool IS_ALIGNED(T* val, uintptr_t alignment);
 
 #endif // HOST_64BIT
 
-#ifndef __GCENV_BASE_INCLUDED__
-
-#define OS_PAGE_SIZE    PalOsPageSize()
-
-#endif // __GCENV_BASE_INCLUDED__
-
 #if defined(TARGET_ARM)
 #define THUMB_CODE 1
 #endif
@@ -264,8 +258,6 @@ typedef int32_t FC_BOOL_ARG;
 #define NOT_IN_DAC(x)
 #endif
 
-#define INLINE inline
-
 enum STARTUP_TIMELINE_EVENT_ID
 {
     PROCESS_ATTACH_BEGIN = 0,

src/coreclr/nativeaot/Runtime/MiscHelpers.cpp

@@ -67,15 +67,6 @@ EXTERN_C UInt32_BOOL QCALLTYPE RhYield()
     return PalSwitchToThread();
 }
 
-EXTERN_C void QCALLTYPE RhFlushProcessWriteBuffers()
-{
-    // This must be called via p/invoke -- it's a wait operation and we don't want to block thread suspension on this.
-    ASSERT_MSG(!ThreadStore::GetCurrentThread()->IsCurrentThreadInCooperativeMode(),
-        "You must p/invoke to RhFlushProcessWriteBuffers");
-
-    minipal_memory_barrier_process_wide();
-}
-
 // Get the list of currently loaded NativeAOT modules (as OS HMODULE handles). The caller provides a reference
 // to an array of pointer-sized elements and we return the total number of modules currently loaded (whether
 // that is less than, equal to or greater than the number of elements in the array). If there are more modules
@@ -486,13 +477,6 @@ FCIMPL2(uint32_t, RhGetKnobValues, char *** pResultKeys, char *** pResultValues)
 }
 FCIMPLEND
 
-#if defined(TARGET_X86) || defined(TARGET_AMD64)
-EXTERN_C void QCALLTYPE RhCpuIdEx(int* cpuInfo, int functionId, int subFunctionId)
-{
-    __cpuidex(cpuInfo, functionId, subFunctionId);
-}
-#endif
-
 FCIMPL3(int32_t, RhpLockCmpXchg32, int32_t * location, int32_t value, int32_t comparand)
 {
     return PalInterlockedCompareExchange(location, value, comparand);

src/coreclr/nativeaot/Runtime/Pal.h

@@ -238,7 +238,6 @@ HANDLE PalCreateEventW(_In_opt_ LPSECURITY_ATTRIBUTES pEventAttributes, UInt32_B
 HANDLE PalGetModuleHandleFromPointer(_In_ void* pointer);
 
 #ifdef TARGET_UNIX
-uint32_t PalGetOsPageSize();
 typedef int32_t (*PHARDWARE_EXCEPTION_HANDLER)(uintptr_t faultCode, uintptr_t faultAddress, PAL_LIMITED_CONTEXT* palContext, uintptr_t* arg0Reg, uintptr_t* arg1Reg);
 void PalSetHardwareExceptionHandler(PHARDWARE_EXCEPTION_HANDLER handler);
 #endif

src/coreclr/nativeaot/Runtime/ThunksMapping.cpp

@@ -33,7 +33,7 @@
 static_assert((THUNK_SIZE % 4) == 0, "Thunk stubs size not aligned correctly. This will cause runtime failures.");
 
 // 32 K or OS page
-#define THUNKS_MAP_SIZE (max(0x8000, OS_PAGE_SIZE))
+#define THUNKS_MAP_SIZE (max((size_t)0x8000, OS_PAGE_SIZE))
 
 #ifdef TARGET_ARM
 //*****************************************************************************
@@ -63,13 +63,13 @@ FCIMPL0(int, RhpGetNumThunkBlocksPerMapping)
 {
     ASSERT_MSG((THUNKS_MAP_SIZE % OS_PAGE_SIZE) == 0, "Thunks map size should be in multiples of pages");
 
-    return THUNKS_MAP_SIZE / OS_PAGE_SIZE;
+    return (int)(THUNKS_MAP_SIZE / OS_PAGE_SIZE);
 }
 FCIMPLEND
 
 FCIMPL0(int, RhpGetNumThunksPerBlock)
 {
-    return min(
+    return (int)min(
         OS_PAGE_SIZE / THUNK_SIZE,                              // Number of thunks that can fit in a page
         (OS_PAGE_SIZE - POINTER_SIZE) / (POINTER_SIZE * 2)      // Number of pointer pairs, minus the jump stub cell, that can fit in a page
     );
@@ -96,22 +96,24 @@ FCIMPLEND
 
 FCIMPL0(int, RhpGetThunkBlockSize)
 {
-    return OS_PAGE_SIZE;
+    return (int)OS_PAGE_SIZE;
 }
 FCIMPLEND
 
 EXTERN_C HRESULT QCALLTYPE RhAllocateThunksMapping(void** ppThunksSection)
 {
+    size_t thunksMapSize = THUNKS_MAP_SIZE;
+
 #ifdef WIN32
 
-    void * pNewMapping = PalVirtualAlloc(THUNKS_MAP_SIZE * 2, PAGE_READWRITE);
+    void * pNewMapping = PalVirtualAlloc(thunksMapSize * 2, PAGE_READWRITE);
     if (pNewMapping == NULL)
     {
         return E_OUTOFMEMORY;
     }
 
     void * pThunksSection = pNewMapping;
-    void * pDataSection = (uint8_t*)pNewMapping + THUNKS_MAP_SIZE;
+    void * pDataSection = (uint8_t*)pNewMapping + thunksMapSize;
 
 #else
 
@@ -120,17 +122,17 @@ EXTERN_C HRESULT QCALLTYPE RhAllocateThunksMapping(void** ppThunksSection)
     // reduce it to RW for the data section. For the stubs section we need to increase to RWX to generate the stubs
     // instructions. After this we go back to RX for the stubs section before the stubs are used and should not be
     // changed anymore.
-    void * pNewMapping = PalVirtualAlloc(THUNKS_MAP_SIZE * 2, PAGE_EXECUTE_READ);
+    void * pNewMapping = PalVirtualAlloc(thunksMapSize * 2, PAGE_EXECUTE_READ);
     if (pNewMapping == NULL)
     {
         return E_OUTOFMEMORY;
     }
 
     void * pThunksSection = pNewMapping;
-    void * pDataSection = (uint8_t*)pNewMapping + THUNKS_MAP_SIZE;
+    void * pDataSection = (uint8_t*)pNewMapping + thunksMapSize;
 
-    if (!PalVirtualProtect(pDataSection, THUNKS_MAP_SIZE, PAGE_READWRITE) ||
-        !PalVirtualProtect(pThunksSection, THUNKS_MAP_SIZE, PAGE_EXECUTE_READWRITE))
+    if (!PalVirtualProtect(pDataSection, thunksMapSize, PAGE_READWRITE) ||
+        !PalVirtualProtect(pThunksSection, thunksMapSize, PAGE_EXECUTE_READWRITE))
     {
         PalVirtualFree(pNewMapping, THUNKS_MAP_SIZE * 2);
         return E_FAIL;
@@ -172,7 +174,7 @@ EXTERN_C HRESULT QCALLTYPE RhAllocateThunksMapping(void** ppThunksSection)
 
             *((uint32_t*)pCurrentThunkAddress) = 0x00a2ff41;
             pCurrentThunkAddress += 3;
-            *((uint32_t*)pCurrentThunkAddress) = OS_PAGE_SIZE - POINTER_SIZE - (i * POINTER_SIZE * 2);
+            *((uint32_t*)pCurrentThunkAddress) = (uint32_t)(OS_PAGE_SIZE - POINTER_SIZE - (i * POINTER_SIZE * 2));
             pCurrentThunkAddress += 4;
 
             // nops for alignment
@@ -231,7 +233,7 @@ EXTERN_C HRESULT QCALLTYPE RhAllocateThunksMapping(void** ppThunksSection)
             *((uint32_t*)pCurrentThunkAddress) = 0x10000010 | (((delta & 0x03) << 29) | (((delta & 0x1FFFFC) >> 2) << 5));
             pCurrentThunkAddress += 4;
 
-            *((uint32_t*)pCurrentThunkAddress) = 0xF9400211 | (((OS_PAGE_SIZE - POINTER_SIZE - (i * POINTER_SIZE * 2)) / 8) << 10);
+            *((uint32_t*)pCurrentThunkAddress) = 0xF9400211 | (((uint32_t)((OS_PAGE_SIZE - POINTER_SIZE - (i * POINTER_SIZE * 2)) / 8) << 10));
             pCurrentThunkAddress += 4;
 
             *((uint32_t*)pCurrentThunkAddress) = 0xD61F0220;
@@ -307,14 +309,14 @@ EXTERN_C HRESULT QCALLTYPE RhAllocateThunksMapping(void** ppThunksSection)
     #error "Unknown OS"
 #endif
 #else
-    if (!PalVirtualProtect(pThunksSection, THUNKS_MAP_SIZE, PAGE_EXECUTE_READ))
+    if (!PalVirtualProtect(pThunksSection, thunksMapSize, PAGE_EXECUTE_READ))
     {
-        PalVirtualFree(pNewMapping, THUNKS_MAP_SIZE * 2);
+        PalVirtualFree(pNewMapping, thunksMapSize * 2);
         return E_FAIL;
     }
 #endif
 
-    PalFlushInstructionCache(pThunksSection, THUNKS_MAP_SIZE);
+    PalFlushInstructionCache(pThunksSection, thunksMapSize);
 
     *ppThunksSection = pThunksSection;
     return S_OK;