Merge pull request #5408 from Sonicadvance1/129

FEXCore: Use WritePriorityMutex for Code Invalidation Mutex

FEXCore/Source/Interface/Context/Context.h

@@ -209,7 +209,7 @@ public:
   void ClearCodeCache(FEXCore::Core::InternalThreadState* Thread, bool NewCodeBuffer = true) override;
   void InvalidateCodeBuffersCodeRange(uint64_t Start, uint64_t Length) override;
   void InvalidateThreadCachedCodeRange(FEXCore::Core::InternalThreadState* Thread, uint64_t Start, uint64_t Length) override;
-  FEXCore::ForkableSharedMutex& GetCodeInvalidationMutex() override {
+  FEXCore::Utils::WritePriorityMutex::Mutex& GetCodeInvalidationMutex() override {
     return CodeInvalidationMutex;
   }
 
@@ -263,7 +263,7 @@ public:
     FEX_CONFIG_OPT(MonoHacks, MONOHACKS);
   } Config;
 
-  FEXCore::ForkableSharedMutex CodeInvalidationMutex;
+  FEXCore::Utils::WritePriorityMutex::Mutex CodeInvalidationMutex {};
 
   uint32_t StrictSplitLockMutex {};
 

FEXCore/Source/Interface/Core/CodeCache.cpp

@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-#include "Utils/SpinWaitLock.h"
+#include <FEXCore/Utils/SpinWaitLock.h>
 
 #include <Interface/Context/Context.h>
 #include <Interface/Core/ArchHelpers/Arm64Emitter.h>

FEXCore/Source/Interface/Core/Core.cpp

@@ -30,7 +30,7 @@ $end_info$
 #include "Interface/IR/RegisterAllocationData.h"
 #include "Utils/Allocator.h"
 #include "Utils/Allocator/HostAllocator.h"
-#include "Utils/SpinWaitLock.h"
+#include <FEXCore/Utils/SpinWaitLock.h>
 #include "Utils/variable_length_integer.h"
 
 #include <FEXCore/Config/Config.h>

FEXCore/Source/Interface/Core/LookupCache.h

@@ -3,7 +3,7 @@
 #include "Interface/Context/Context.h"
 #include <FEXCore/Utils/LogManager.h>
 #include <FEXCore/Utils/SHMStats.h>
-#include "Utils/WritePriorityMutex.h"
+#include <FEXCore/Utils/WritePriorityMutex.h>
 
 #include <FEXCore/fextl/map.h>
 #include <FEXCore/fextl/memory_resource.h>

FEXCore/Source/Utils/ArchHelpers/Arm64.cpp

@@ -2,7 +2,7 @@
 
 #include "Interface/Core/CPUBackend.h"
 #include "Interface/Context/Context.h"
-#include "Utils/SpinWaitLock.h"
+#include <FEXCore/Utils/SpinWaitLock.h>
 
 #include <FEXCore/Debug/InternalThreadState.h>
 #include <FEXCore/Utils/EnumUtils.h>

FEXCore/Source/Utils/SpinWaitLock.cpp

@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-#include "Utils/SpinWaitLock.h"
+#include <FEXCore/Utils/SpinWaitLock.h>
 
 namespace FEXCore::Utils::SpinWaitLock {
 #ifdef ARCHITECTURE_arm64

FEXCore/include/FEXCore/Core/Context.h

@@ -13,10 +13,10 @@
 #include <FEXCore/fextl/set.h>
 #include <FEXCore/fextl/string.h>
 #include <FEXCore/fextl/vector.h>
+#include <FEXCore/Utils/WritePriorityMutex.h>
 
 namespace FEXCore {
 struct HostFeatures;
-class ForkableSharedMutex;
 class ThunkHandler;
 } // namespace FEXCore
 
@@ -144,7 +144,7 @@ public:
   FEX_DEFAULT_VISIBILITY virtual void InvalidateCodeBuffersCodeRange(uint64_t Start, uint64_t Length) = 0;
   FEX_DEFAULT_VISIBILITY virtual void
   InvalidateThreadCachedCodeRange(FEXCore::Core::InternalThreadState* Thread, uint64_t Start, uint64_t Length) = 0;
-  FEX_DEFAULT_VISIBILITY virtual FEXCore::ForkableSharedMutex& GetCodeInvalidationMutex() = 0;
+  FEX_DEFAULT_VISIBILITY virtual FEXCore::Utils::WritePriorityMutex::Mutex& GetCodeInvalidationMutex() = 0;
 
   FEX_DEFAULT_VISIBILITY virtual void
   ConfigureAOTGen(FEXCore::Core::InternalThreadState* Thread, fextl::set<uint64_t>* ExternalBranches, uint64_t SectionMaxAddress) = 0;

FEXCore/include/FEXCore/Utils/SpinWaitLock.h

@@ -50,8 +50,9 @@ namespace FEXCore::Utils::SpinWaitLock {
 #define SPINLOOP_32BIT SPINLOOP_BODY(ldar, w)
 #define SPINLOOP_64BIT SPINLOOP_BODY(ldar, x)
 
-extern uint64_t CycleCounterFrequency;
-extern uint64_t CyclesPerNanosecond;
+FEX_DEFAULT_VISIBILITY extern uint64_t CycleCounterFrequency;
+
+FEX_DEFAULT_VISIBILITY extern uint64_t CyclesPerNanosecond;
 
 ///< Get the raw cycle counter which is synchronizing.
 /// `CNTVCTSS_EL0` also does the same thing, but requires the FEAT_ECV feature.

FEXCore/include/FEXCore/Utils/WritePriorityMutex.h

@@ -9,11 +9,15 @@
 #include <unistd.h>
 #else
 #include <synchapi.h>
+// Don't pull in all WIN32 headers for INFINITE. Causes too many problems.
+#ifndef INFINITE
+#define INFINITE 0xffffffff
+#endif
 #endif
 
 #include <FEXCore/Utils/LogManager.h>
 
-#include "Utils/SpinWaitLock.h"
+#include <FEXCore/Utils/SpinWaitLock.h>
 
 namespace FEXCore::Utils::WritePriorityMutex {
 

FEXCore/unittests/APITests/FutexSpinTest.cpp

@@ -1,5 +1,5 @@
 // SPDX-License-Identifier: MIT
-#include "Utils/SpinWaitLock.h"
+#include <FEXCore/Utils/SpinWaitLock.h>
 #include <catch2/catch_test_macros.hpp>
 #include <chrono>
 #include <thread>

Source/Tools/LinuxEmulation/LinuxSyscalls/SyscallsSMCTracking.cpp

@@ -295,7 +295,7 @@ void* SyscallHandler::GuestMmap(bool Is64Bit, FEXCore::Core::InternalThreadState
   InvalidateCodeRangeIfNecessary(Thread, Result, Size);
 
   if (LateMetadata) {
-    auto CodeInvalidationlk = GuardSignalDeferringSectionWithFallback(CTX->GetCodeInvalidationMutex(), Thread);
+    auto CodeInvalidationlk = FEXCore::GuardSignalDeferringSectionWithFallback(CTX->GetCodeInvalidationMutex(), Thread);
     CTX->AddForceTSOInformation(LateMetadata->VolatileValidRanges, std::move(LateMetadata->VolatileInstructions));
   }
 
@@ -335,7 +335,7 @@ uint64_t SyscallHandler::GuestMunmap(bool Is64Bit, FEXCore::Core::InternalThread
   InvalidateCodeRangeIfNecessary(Thread, reinterpret_cast<uint64_t>(addr), Size);
 
   if (length) {
-    auto CodeInvalidationlk = GuardSignalDeferringSectionWithFallback(CTX->GetCodeInvalidationMutex(), Thread);
+    auto CodeInvalidationlk = FEXCore::GuardSignalDeferringSectionWithFallback(CTX->GetCodeInvalidationMutex(), Thread);
     CTX->RemoveForceTSOInformation(reinterpret_cast<uint64_t>(addr), length);
   }
 

Source/Tools/LinuxEmulation/LinuxSyscalls/ThreadManager.h

@@ -221,7 +221,7 @@ public:
     // Potential deferred since Thread might not be valid.
     // Thread object isn't valid very early in frontend's initialization.
     // To be more optimal the frontend should provide this code with a valid Thread object earlier.
-    auto CodeInvalidationlk = GuardSignalDeferringSectionWithFallback(CTX->GetCodeInvalidationMutex(), CallingThread);
+    auto CodeInvalidationlk = FEXCore::GuardSignalDeferringSectionWithFallback(CTX->GetCodeInvalidationMutex(), CallingThread);
     CTX->InvalidateCodeBuffersCodeRange(Start, Length);
     for (auto& Thread : Threads) {
       CTX->InvalidateThreadCachedCodeRange(Thread->Thread, Start, Length);
@@ -235,7 +235,7 @@ public:
     // Potential deferred since Thread might not be valid.
     // Thread object isn't valid very early in frontend's initialization.
     // To be more optimal the frontend should provide this code with a valid Thread object earlier.
-    auto CodeInvalidationlk = GuardSignalDeferringSectionWithFallback(CTX->GetCodeInvalidationMutex(), CallingThread);
+    auto CodeInvalidationlk = FEXCore::GuardSignalDeferringSectionWithFallback(CTX->GetCodeInvalidationMutex(), CallingThread);
     CTX->InvalidateCodeBuffersCodeRange(Start, Length);
     for (auto& Thread : Threads) {
       CTX->InvalidateThreadCachedCodeRange(Thread->Thread, Start, Length);