Windows: Enable downstream kernel-side unaligned atomic handling

Source/Windows/ARM64EC/Module.cpp

@@ -640,7 +640,6 @@ NTSTATUS ProcessInit() {
 
   SignalDelegator = fextl::make_unique<FEX::DummyHandlers::DummySignalDelegator>();
   SyscallHandler = fextl::make_unique<Exception::ECSyscallHandler>();
-  Exception::HandlerConfig.emplace();
 
   const auto NtDll = GetModuleHandle("ntdll.dll");
   const bool IsWine = !!GetProcAddress(NtDll, "wine_get_version");
@@ -654,7 +653,7 @@ NTSTATUS ProcessInit() {
   CTX->SetSignalDelegator(SignalDelegator.get());
   CTX->SetSyscallHandler(SyscallHandler.get());
   CTX->InitCore();
-
+  Exception::HandlerConfig.emplace(*CTX);
   InvalidationTracker.emplace(*CTX, Threads);
 
   HandleImageMap(NtDllBase);

Source/Windows/Common/TSOHandlerConfig.h

@@ -1,18 +1,32 @@
 // SPDX-License-Identifier: MIT
 #pragma once
 
+#include <FEXCore/Core/Context.h>
 #include <FEXCore/Config/Config.h>
 #include <FEXCore/Utils/ArchHelpers/Arm64.h>
 
 namespace FEX::Windows {
 class TSOHandlerConfig final {
 public:
-  TSOHandlerConfig() {
+  TSOHandlerConfig(FEXCore::Context::Context& CTX) {
     if (HalfBarrierTSOEnabled()) {
       UnalignedHandlerType = FEXCore::ArchHelpers::Arm64::UnalignedHandlerType::HalfBarrier;
     } else {
       UnalignedHandlerType = FEXCore::ArchHelpers::Arm64::UnalignedHandlerType::NonAtomic;
     }
+
+    if (TSOEnabled()) {
+      BOOL Enable = TRUE;
+      NTSTATUS Status = NtSetInformationProcess(NtCurrentProcess(), ProcessFexHardwareTso, &Enable, sizeof(Enable));
+      if (Status == STATUS_SUCCESS) {
+        CTX.SetHardwareTSOSupport(true);
+      }
+    }
+
+    uint64_t Flags =
+      (StrictInProcessSplitLocks() ? FEX_UNALIGN_ATOMIC_STRICT_SPLIT_LOCKS : 0) | FEX_UNALIGN_ATOMIC_BACKPATCH | FEX_UNALIGN_ATOMIC_EMULATE;
+
+    NtSetInformationProcess(NtCurrentProcess(), ProcessFexUnalignAtomic, &Flags, sizeof(Flags));
   }
 
   FEXCore::ArchHelpers::Arm64::UnalignedHandlerType GetUnalignedHandlerType() const {
@@ -20,7 +34,9 @@ public:
   }
 
 private:
+  FEX_CONFIG_OPT(TSOEnabled, TSOENABLED);
   FEX_CONFIG_OPT(HalfBarrierTSOEnabled, HALFBARRIERTSOENABLED);
+  FEX_CONFIG_OPT(StrictInProcessSplitLocks, STRICTINPROCESSSPLITLOCKS);
 
   FEXCore::ArchHelpers::Arm64::UnalignedHandlerType UnalignedHandlerType {FEXCore::ArchHelpers::Arm64::UnalignedHandlerType::HalfBarrier};
 };

Source/Windows/WOW64/Module.cpp

@@ -523,7 +523,6 @@ void BTCpuProcessInit() {
 
   SignalDelegator = fextl::make_unique<FEX::DummyHandlers::DummySignalDelegator>();
   SyscallHandler = fextl::make_unique<WowSyscallHandler>();
-  Context::HandlerConfig.emplace();
   const auto NtDll = GetModuleHandle("ntdll.dll");
   const bool IsWine = !!GetProcAddress(NtDll, "wine_get_version");
   OvercommitTracker.emplace(IsWine);
@@ -538,7 +537,7 @@ void BTCpuProcessInit() {
   CTX->SetSignalDelegator(SignalDelegator.get());
   CTX->SetSyscallHandler(SyscallHandler.get());
   CTX->InitCore();
-
+  Context::HandlerConfig.emplace(*CTX);
   InvalidationTracker.emplace(*CTX, Threads);
 
   auto NtDllX86 = reinterpret_cast<SYSTEM_DLL_INIT_BLOCK*>(GetProcAddress(NtDll, "LdrSystemDllInitBlock"))->ntdll_handle;

Source/Windows/include/winternl.h

@@ -455,6 +455,12 @@ typedef enum _MEMORY_INFORMATION_CLASS {
 #define SystemEmulationBasicInformation (SYSTEM_INFORMATION_CLASS)62
 
 #define ProcessFexHardwareTso (PROCESSINFOCLASS)2000
+#define ProcessFexUnalignAtomic (PROCESSINFOCLASS)2001
+
+// These match the prctl flag values
+#define FEX_UNALIGN_ATOMIC_EMULATE (1ULL << 0)
+#define FEX_UNALIGN_ATOMIC_BACKPATCH (1ULL << 1)
+#define FEX_UNALIGN_ATOMIC_STRICT_SPLIT_LOCKS (1ULL << 2)
 
 typedef enum _KEY_VALUE_INFORMATION_CLASS {
   KeyValueBasicInformation,