Merge pull request #1670 from FEX-Emu/skmp/processwide-code-invalidat…

…ions Core: context-wide guest code invalidations
FEX-Emu · May 3, 2022 · d810988 · d810988
2 parents 8e36f53 + 09be28a
commit d810988
Show file tree

Hide file tree

Showing 19 changed files with 136 additions and 94 deletions.
diff --git a/External/FEXCore/Source/Interface/Context/Context.h b/External/FEXCore/Source/Interface/Context/Context.h
@@ -155,11 +155,13 @@ namespace FEXCore::Context {
     void RegisterHostSignalHandler(int Signal, HostSignalDelegatorFunction Func, bool Required);
     void RegisterFrontendHostSignalHandler(int Signal, HostSignalDelegatorFunction Func, bool Required);
 
-    static void RemoveCodeEntry(FEXCore::Core::InternalThreadState *Thread, uint64_t GuestRIP);
+    // Must be called from owning thread
+    static void RemoveThreadCodeEntry(FEXCore::Core::InternalThreadState *Thread, uint64_t GuestRIP);
 
     // Wrapper which takes CpuStateFrame instead of InternalThreadState
-    static void RemoveCodeEntryFromJit(FEXCore::Core::CpuStateFrame *Frame, uint64_t GuestRIP) {
-      RemoveCodeEntry(Frame->Thread, GuestRIP);
+    // Must be called from owning thread
+    static void RemoveThreadCodeEntryFromJit(FEXCore::Core::CpuStateFrame *Frame, uint64_t GuestRIP) {
+      RemoveThreadCodeEntry(Frame->Thread, GuestRIP);
     }
 
     // Debugger interface

diff --git a/External/FEXCore/Source/Interface/Core/Core.cpp b/External/FEXCore/Source/Interface/Core/Core.cpp
@@ -447,6 +447,7 @@ namespace FEXCore::Context {
           : nullptr),
         decltype(Entry.DebugData)(new Core::DebugData())
       };
+      std::lock_guard<std::recursive_mutex> lk(Thread->LookupCache->WriteLock);
       Thread->LocalIRCache.insert({Addr, std::move(Entry)});
     };
 
@@ -626,6 +627,8 @@ namespace FEXCore::Context {
   }
 
   void Context::ClearCodeCache(FEXCore::Core::InternalThreadState *Thread, bool AlsoClearIRCache) {
+    std::lock_guard<std::recursive_mutex> lk(Thread->LookupCache->WriteLock);
+
     Thread->LookupCache->ClearCache();
     Thread->CPUBackend->ClearCache();
 
@@ -738,7 +741,7 @@ namespace FEXCore::Context {
           Thread->OpDispatcher->SetTrueJumpTarget(InvalidateCodeCond, CodeWasChangedBlock);
 
           Thread->OpDispatcher->SetCurrentCodeBlock(CodeWasChangedBlock);
-          Thread->OpDispatcher->_RemoveCodeEntry();
+          Thread->OpDispatcher->_RemoveThreadCodeEntry();
           Thread->OpDispatcher->_ExitFunction(Thread->OpDispatcher->_EntrypointOffset(Block.Entry + BlockInstructionsLength - GuestRIP, GPRSize));
 
           auto NextOpBlock = Thread->OpDispatcher->CreateNewCodeBlockAfter(CurrentBlock);
@@ -847,6 +850,7 @@ namespace FEXCore::Context {
     uint64_t StartAddr {};
     uint64_t Length {};
 
+    std::lock_guard<std::recursive_mutex> lk(Thread->LookupCache->WriteLock);
     // Do we already have this in the IR cache?
     auto LocalEntry = Thread->LocalIRCache.find(GuestRIP);
 
@@ -1033,21 +1037,32 @@ namespace FEXCore::Context {
     }
   }
 
-  void FlushCodeRange(FEXCore::Core::InternalThreadState *Thread, uint64_t Start, uint64_t Length) {
+  static void InvalidateGuestThreadCodeRange(FEXCore::Core::InternalThreadState *Thread, uint64_t Start, uint64_t Length) {
+    std::lock_guard<std::recursive_mutex> lk(Thread->LookupCache->WriteLock);
 
-    if (Thread->CTX->Config.SMCChecks == FEXCore::Config::CONFIG_SMC_MMAN) {
-      auto lower = Thread->LookupCache->CodePages.lower_bound(Start >> 12);
-      auto upper = Thread->LookupCache->CodePages.upper_bound((Start + Length) >> 12);
+    auto lower = Thread->LookupCache->CodePages.lower_bound(Start >> 12);
+    auto upper = Thread->LookupCache->CodePages.upper_bound((Start + Length - 1) >> 12);
 
-      for (auto it = lower; it != upper; it++) {
-        for (auto Address: it->second)
-          Context::RemoveCodeEntry(Thread, Address);
-        it->second.clear();
+    for (auto it = lower; it != upper; it++) {
+      for (auto Address: it->second) {
+        Context::RemoveThreadCodeEntry(Thread, Address);
       }
+      it->second.clear();
     }
   }
 
-  void Context::RemoveCodeEntry(FEXCore::Core::InternalThreadState *Thread, uint64_t GuestRIP) {
+  void InvalidateGuestCodeRange(FEXCore::Context::Context *CTX, uint64_t Start, uint64_t Length) {
+    std::lock_guard<std::mutex> lk(CTX->ThreadCreationMutex);
+    for (auto &Thread : CTX->Threads) {
+      if (Thread->RunningEvents.Running.load()) {
+        InvalidateGuestThreadCodeRange(Thread, Start, Length);
+      }
+    }
+  }
+
+  void Context::RemoveThreadCodeEntry(FEXCore::Core::InternalThreadState *Thread, uint64_t GuestRIP) {
+    std::lock_guard<std::recursive_mutex> lk(Thread->LookupCache->WriteLock);
+
     Thread->LocalIRCache.erase(GuestRIP);
     Thread->LookupCache->Erase(GuestRIP);
   }
@@ -1058,7 +1073,7 @@ namespace FEXCore::Context {
     Thread->CurrentFrame->State.rip = RIP;
 
     // Erase the RIP from all the storage backings if it exists
-    RemoveCodeEntry(Thread, RIP);
+    RemoveThreadCodeEntry(Thread, RIP);
 
     // We don't care if compilation passes or not
     CompileBlock(Thread->CurrentFrame, RIP);
@@ -1075,6 +1090,7 @@ namespace FEXCore::Context {
   }
 
   bool Context::GetDebugDataForRIP(uint64_t RIP, FEXCore::Core::DebugData *Data) {
+    std::lock_guard<std::recursive_mutex> lk(ParentThread->LookupCache->WriteLock);
     auto it = ParentThread->LocalIRCache.find(RIP);
     if (it == ParentThread->LocalIRCache.end()) {
       return false;

diff --git a/External/FEXCore/Source/Interface/Core/Interpreter/BranchOps.cpp b/External/FEXCore/Source/Interface/Core/Interpreter/BranchOps.cpp
@@ -147,8 +147,8 @@ DEF_OP(ValidateCode) {
   }
 }
 
-DEF_OP(RemoveCodeEntry) {
-  Data->State->CTX->RemoveCodeEntry(Data->State, Data->CurrentEntry);
+DEF_OP(RemoveThreadCodeEntry) {
+  Data->State->CTX->RemoveThreadCodeEntry(Data->State, Data->CurrentEntry);
 }
 
 DEF_OP(CPUID) {

diff --git a/External/FEXCore/Source/Interface/Core/Interpreter/InterpreterOps.cpp b/External/FEXCore/Source/Interface/Core/Interpreter/InterpreterOps.cpp
@@ -123,7 +123,7 @@ constexpr OpHandlerArray InterpreterOpHandlers = [] {
   REGISTER_OP(INLINESYSCALL,          InlineSyscall);
   REGISTER_OP(THUNK,                  Thunk);
   REGISTER_OP(VALIDATECODE,           ValidateCode);
-  REGISTER_OP(REMOVECODEENTRY,        RemoveCodeEntry);
+  REGISTER_OP(REMOVETHREADCODEENTRY,        RemoveThreadCodeEntry);
   REGISTER_OP(CPUID,                  CPUID);
 
   // Conversion ops

diff --git a/External/FEXCore/Source/Interface/Core/Interpreter/InterpreterOps.h b/External/FEXCore/Source/Interface/Core/Interpreter/InterpreterOps.h
@@ -153,7 +153,7 @@ namespace FEXCore::CPU {
   DEF_OP(InlineSyscall);
   DEF_OP(Thunk);
   DEF_OP(ValidateCode);
-  DEF_OP(RemoveCodeEntry);
+  DEF_OP(RemoveThreadCodeEntry);
   DEF_OP(CPUID);
 
   ///< Conversion ops

diff --git a/External/FEXCore/Source/Interface/Core/JIT/Arm64/BranchOps.cpp b/External/FEXCore/Source/Interface/Core/JIT/Arm64/BranchOps.cpp
@@ -442,7 +442,7 @@ DEF_OP(ValidateCode) {
   }
 }
 
-DEF_OP(RemoveCodeEntry) {
+DEF_OP(RemoveThreadCodeEntry) {
   // Arguments are passed as follows:
   // X0: Thread
   // X1: RIP
@@ -452,7 +452,7 @@ DEF_OP(RemoveCodeEntry) {
   mov(x0, STATE);
   LoadConstant(x1, Entry);
 
-  ldr(x2, MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.AArch64.RemoveCodeEntryFromJIT)));
+  ldr(x2, MemOperand(STATE, offsetof(FEXCore::Core::CpuStateFrame, Pointers.AArch64.RemoveThreadCodeEntryFromJIT)));
   SpillStaticRegs();
   blr(x2);
   FillStaticRegs();
@@ -500,7 +500,7 @@ void Arm64JITCore::RegisterBranchHandlers() {
   REGISTER_OP(INLINESYSCALL,     InlineSyscall);
   REGISTER_OP(THUNK,             Thunk);
   REGISTER_OP(VALIDATECODE,      ValidateCode);
-  REGISTER_OP(REMOVECODEENTRY,   RemoveCodeEntry);
+  REGISTER_OP(REMOVETHREADCODEENTRY,   RemoveThreadCodeEntry);
   REGISTER_OP(CPUID,             CPUID);
 #undef REGISTER_OP
 }

diff --git a/External/FEXCore/Source/Interface/Core/JIT/Arm64/JIT.cpp b/External/FEXCore/Source/Interface/Core/JIT/Arm64/JIT.cpp
@@ -453,7 +453,7 @@ Arm64JITCore::Arm64JITCore(FEXCore::Context::Context *ctx, FEXCore::Core::Intern
     Pointers.LREM = reinterpret_cast<uint64_t>(LREM);
     Pointers.PrintValue = reinterpret_cast<uint64_t>(PrintValue);
     Pointers.PrintVectorValue = reinterpret_cast<uint64_t>(PrintVectorValue);
-    Pointers.RemoveCodeEntryFromJIT = reinterpret_cast<uintptr_t>(&Context::Context::RemoveCodeEntryFromJit);
+    Pointers.RemoveThreadCodeEntryFromJIT = reinterpret_cast<uintptr_t>(&Context::Context::RemoveThreadCodeEntryFromJit);
     Pointers.CPUIDObj = reinterpret_cast<uint64_t>(&CTX->CPUID);
 
     {

diff --git a/External/FEXCore/Source/Interface/Core/JIT/Arm64/JITClass.h b/External/FEXCore/Source/Interface/Core/JIT/Arm64/JITClass.h
@@ -278,7 +278,7 @@ class Arm64JITCore final : public CPUBackend, public Arm64Emitter  {
   DEF_OP(InlineSyscall);
   DEF_OP(Thunk);
   DEF_OP(ValidateCode);
-  DEF_OP(RemoveCodeEntry);
+  DEF_OP(RemoveThreadCodeEntry);
   DEF_OP(CPUID);
 
   ///< Conversion ops

diff --git a/External/FEXCore/Source/Interface/Core/JIT/x86_64/BranchOps.cpp b/External/FEXCore/Source/Interface/Core/JIT/x86_64/BranchOps.cpp
@@ -259,7 +259,7 @@ DEF_OP(ValidateCode) {
   }
 }
 
-DEF_OP(RemoveCodeEntry) {
+DEF_OP(RemoveThreadCodeEntry) {
   auto NumPush = RA64.size();
 
   for (auto &Reg : RA64)
@@ -272,7 +272,7 @@ DEF_OP(RemoveCodeEntry) {
   mov(rax, Entry); // imm64 move
   mov(rsi, rax);
 
-  call(qword [STATE + offsetof(FEXCore::Core::CpuStateFrame, Pointers.X86.RemoveCodeEntryFromJIT)]);
+  call(qword [STATE + offsetof(FEXCore::Core::CpuStateFrame, Pointers.X86.RemoveThreadCodeEntryFromJIT)]);
 
   if (NumPush & 1)
     add(rsp, 8); // Align
@@ -330,7 +330,7 @@ void X86JITCore::RegisterBranchHandlers() {
   REGISTER_OP(SYSCALL,           Syscall);
   REGISTER_OP(THUNK,             Thunk);
   REGISTER_OP(VALIDATECODE,      ValidateCode);
-  REGISTER_OP(REMOVECODEENTRY,   RemoveCodeEntry);
+  REGISTER_OP(REMOVETHREADCODEENTRY,   RemoveThreadCodeEntry);
   REGISTER_OP(CPUID,             CPUID);
 #undef REGISTER_OP
 }

diff --git a/External/FEXCore/Source/Interface/Core/JIT/x86_64/JIT.cpp b/External/FEXCore/Source/Interface/Core/JIT/x86_64/JIT.cpp
@@ -371,7 +371,7 @@ X86JITCore::X86JITCore(FEXCore::Context::Context *ctx, FEXCore::Core::InternalTh
     // Process specific
     Pointers.PrintValue = reinterpret_cast<uint64_t>(PrintValue);
     Pointers.PrintVectorValue = reinterpret_cast<uint64_t>(PrintVectorValue);
-    Pointers.RemoveCodeEntryFromJIT = reinterpret_cast<uintptr_t>(&Context::Context::RemoveCodeEntryFromJit);
+    Pointers.RemoveThreadCodeEntryFromJIT = reinterpret_cast<uintptr_t>(&Context::Context::RemoveThreadCodeEntryFromJit);
     Pointers.CPUIDObj = reinterpret_cast<uint64_t>(&CTX->CPUID);
 
     {

diff --git a/External/FEXCore/Source/Interface/Core/JIT/x86_64/JITClass.h b/External/FEXCore/Source/Interface/Core/JIT/x86_64/JITClass.h
@@ -276,7 +276,7 @@ class X86JITCore final : public CPUBackend, public Xbyak::CodeGenerator {
   DEF_OP(Syscall);
   DEF_OP(Thunk);
   DEF_OP(ValidateCode);
-  DEF_OP(RemoveCodeEntry);
+  DEF_OP(RemoveThreadCodeEntry);
   DEF_OP(CPUID);
 
   ///< Conversion ops

diff --git a/External/FEXCore/Source/Interface/Core/LookupCache.cpp b/External/FEXCore/Source/Interface/Core/LookupCache.cpp
@@ -61,13 +61,16 @@ void LookupCache::HintUsedRange(uint64_t Address, uint64_t Size) {
 }
 
 void LookupCache::ClearL2Cache() {
+  std::lock_guard<std::recursive_mutex> lk(WriteLock);
   // Clear out the page memory
   madvise(reinterpret_cast<void*>(PagePointer), ctx->Config.VirtualMemSize / 4096 * 8, MADV_DONTNEED);
   madvise(reinterpret_cast<void*>(PageMemory), CODE_SIZE, MADV_DONTNEED);
   AllocateOffset = 0;
 }
 
 void LookupCache::ClearCache() {
+  std::lock_guard<std::recursive_mutex> lk(WriteLock);
+
   // Clear L1
   madvise(reinterpret_cast<void*>(L1Pointer), L1_SIZE, MADV_DONTNEED);
   // Clear L2