[release/7.0] Flush instruction cache after thunk pool allocation (#75403)

* Flush instruction cache after thunk pool allocation

Fixes #74710

* More precise ifdef

Co-authored-by: Jan Kotas <jkotas@microsoft.com>

5 files changed, 39 insertions, 5 deletions

diff --git a/src/coreclr/nativeaot/Runtime/PalRedhawk.h b/src/coreclr/nativeaot/Runtime/PalRedhawk.h
index 5e8751f6945..007b3914415 100644
--- a/src/coreclr/nativeaot/Runtime/PalRedhawk.h
+++ b/src/coreclr/nativeaot/Runtime/PalRedhawk.h
@@ -674,6 +674,7 @@ EXTERN_C void * __cdecl _alloca(size_t);
 REDHAWK_PALIMPORT _Ret_maybenull_ _Post_writable_byte_size_(size) void* REDHAWK_PALAPI PalVirtualAlloc(_In_opt_ void* pAddress, uintptr_t size, uint32_t allocationType, uint32_t protect);
 REDHAWK_PALIMPORT UInt32_BOOL REDHAWK_PALAPI PalVirtualFree(_In_ void* pAddress, uintptr_t size, uint32_t freeType);
 REDHAWK_PALIMPORT UInt32_BOOL REDHAWK_PALAPI PalVirtualProtect(_In_ void* pAddress, uintptr_t size, uint32_t protect);
+REDHAWK_PALIMPORT void PalFlushInstructionCache(_In_ void* pAddress, size_t size);
 REDHAWK_PALIMPORT void REDHAWK_PALAPI PalSleep(uint32_t milliseconds);
 REDHAWK_PALIMPORT UInt32_BOOL REDHAWK_PALAPI PalSwitchToThread();
 REDHAWK_PALIMPORT HANDLE REDHAWK_PALAPI PalCreateEventW(_In_opt_ LPSECURITY_ATTRIBUTES pEventAttributes, UInt32_BOOL manualReset, UInt32_BOOL initialState, _In_opt_z_ LPCWSTR pName);
diff --git a/src/coreclr/nativeaot/Runtime/ThunksMapping.cpp b/src/coreclr/nativeaot/Runtime/ThunksMapping.cpp
index 6fe0e768f15..760dbc3b0ad 100644
--- a/src/coreclr/nativeaot/Runtime/ThunksMapping.cpp
+++ b/src/coreclr/nativeaot/Runtime/ThunksMapping.cpp
@@ -229,6 +229,8 @@ EXTERN_C NATIVEAOT_API void* __cdecl RhAllocateThunksMapping()
         return NULL;
     }
 
+    PalFlushInstructionCache(pThunksSection, THUNKS_MAP_SIZE);
+
     return pThunksSection;
 }
 
diff --git a/src/coreclr/nativeaot/Runtime/unix/PalRedhawkUnix.cpp b/src/coreclr/nativeaot/Runtime/unix/PalRedhawkUnix.cpp
index f8be025c062..55ad31c82d8 100644
--- a/src/coreclr/nativeaot/Runtime/unix/PalRedhawkUnix.cpp
+++ b/src/coreclr/nativeaot/Runtime/unix/PalRedhawkUnix.cpp
@@ -832,6 +832,33 @@ REDHAWK_PALEXPORT UInt32_BOOL REDHAWK_PALAPI PalVirtualProtect(_In_ void* pAddre
     return mprotect(pPageStart, memSize, unixProtect) == 0;
 }
 
+REDHAWK_PALEXPORT void PalFlushInstructionCache(_In_ void* pAddress, size_t size)
+{
+#if defined(__linux__) && defined(HOST_ARM)
+    // On Linux/arm (at least on 3.10) we found that there is a problem with __do_cache_op (arch/arm/kernel/traps.c)
+    // implementing cacheflush syscall. cacheflush flushes only the first page in range [pAddress, pAddress + size)
+    // and leaves other pages in undefined state which causes random tests failures (often due to SIGSEGV) with no particular pattern.
+    //
+    // As a workaround, we call __builtin___clear_cache on each page separately.
+
+    const size_t pageSize = getpagesize();
+    uint8_t* begin = (uint8_t*)pAddress;
+    uint8_t* end = begin + size;
+
+    while (begin < end)
+    {
+        uint8_t* endOrNextPageBegin = ALIGN_UP(begin + 1, pageSize);
+        if (endOrNextPageBegin > end)
+            endOrNextPageBegin = end;
+
+        __builtin___clear_cache((char *)begin, (char *)endOrNextPageBegin);
+        begin = endOrNextPageBegin;
+    }
+#else
+    __builtin___clear_cache((char *)pAddress, (char *)pAddress + size);
+#endif
+}
+
 REDHAWK_PALEXPORT _Ret_maybenull_ void* REDHAWK_PALAPI PalSetWerDataBuffer(_In_ void* pNewBuffer)
 {
     static void* pBuffer;
diff --git a/src/coreclr/nativeaot/Runtime/windows/PalRedhawkMinWin.cpp b/src/coreclr/nativeaot/Runtime/windows/PalRedhawkMinWin.cpp
index 39f8ffd00a0..20238e87501 100644
--- a/src/coreclr/nativeaot/Runtime/windows/PalRedhawkMinWin.cpp
+++ b/src/coreclr/nativeaot/Runtime/windows/PalRedhawkMinWin.cpp
@@ -599,6 +599,11 @@ REDHAWK_PALEXPORT UInt32_BOOL REDHAWK_PALAPI PalVirtualProtect(_In_ void* pAddre
     return VirtualProtect(pAddress, size, protect, &oldProtect);
 }
 
+REDHAWK_PALEXPORT void PalFlushInstructionCache(_In_ void* pAddress, size_t size)
+{
+    FlushInstructionCache(GetCurrentProcess(), pAddress, size);
+}
+
 REDHAWK_PALEXPORT _Ret_maybenull_ void* REDHAWK_PALAPI PalSetWerDataBuffer(_In_ void* pNewBuffer)
 {
     static void* pBuffer;
diff --git a/src/coreclr/pal/src/thread/context.cpp b/src/coreclr/pal/src/thread/context.cpp
index 184116239ca..a9a00bb04a0 100644
--- a/src/coreclr/pal/src/thread/context.cpp
+++ b/src/coreclr/pal/src/thread/context.cpp
@@ -1637,10 +1637,7 @@ DBG_FlushInstructionCache(
                           IN LPCVOID lpBaseAddress,
                           IN SIZE_T dwSize)
 {
-#ifndef HOST_ARM
-    // Intrinsic should do the right thing across all platforms (except Linux arm)
-    __builtin___clear_cache((char *)lpBaseAddress, (char *)((INT_PTR)lpBaseAddress + dwSize));
-#else // HOST_ARM
+#if defined(__linux__) && defined(HOST_ARM)
     // On Linux/arm (at least on 3.10) we found that there is a problem with __do_cache_op (arch/arm/kernel/traps.c)
     // implementing cacheflush syscall. cacheflush flushes only the first page in range [lpBaseAddress, lpBaseAddress + dwSize)
     // and leaves other pages in undefined state which causes random tests failures (often due to SIGSEGV) with no particular pattern.
@@ -1660,6 +1657,8 @@ DBG_FlushInstructionCache(
         __builtin___clear_cache((char *)begin, (char *)endOrNextPageBegin);
         begin = endOrNextPageBegin;
     }
-#endif // HOST_ARM
+#else
+    __builtin___clear_cache((char *)lpBaseAddress, (char *)((INT_PTR)lpBaseAddress + dwSize));
+#endif
     return TRUE;
 }