From 41241d7500419ac03c82641e560fc6afe9d8b8f9 Mon Sep 17 00:00:00 2001
From: Ryan Houdek <Sonicadvance1@gmail.com>
Date: Mon, 6 Jul 2026 15:47:05 -0700
Subject: [PATCH] Wow64: Spin loop on atomic with WFE

Instead of burning roughly a million watts, put this spinloop on a WFE.
This tends to occur on a crash during shutdown that isn't fully able to
be avoided. The least we can do is not consume all the power in the
world.
---
 FEXCore/include/FEXCore/Utils/SpinWaitLock.h | 25 ++++++++++++++++++++
 Source/Windows/WOW64/Module.cpp              |  7 ++++--
 2 files changed, 30 insertions(+), 2 deletions(-)

diff --git a/FEXCore/include/FEXCore/Utils/SpinWaitLock.h b/FEXCore/include/FEXCore/Utils/SpinWaitLock.h
index 1fa0c03561..af59a6efb3 100644
--- a/FEXCore/include/FEXCore/Utils/SpinWaitLock.h
+++ b/FEXCore/include/FEXCore/Utils/SpinWaitLock.h
@@ -144,6 +144,21 @@ static inline void WaitPred(T* Futex, T ComparisonValue) {
   }
 }
 
+template<typename T, typename Pred>
+static inline void WaitBitMaskPred(T* Futex, T BitMask, T ComparisonValue, Pred Predicate) {
+  auto AtomicFutex = std::atomic_ref<T>(*Futex);
+  T Result = AtomicFutex.load();
+
+  while (!Predicate(Result & BitMask, ComparisonValue)) {
+    Result = LoadExclusive(Futex);
+    if (Predicate(Result & BitMask, ComparisonValue)) {
+      return;
+    }
+
+    Result = WFELoadAtomic(Futex);
+  }
+}
+
 template<typename T, typename TT>
 static inline bool Wait(T* Futex, TT ExpectedValue, const std::chrono::nanoseconds& Timeout) {
   auto AtomicFutex = std::atomic_ref<T>(*Futex);
@@ -213,6 +228,16 @@ static inline void WaitPred(T* Futex, T ComparisonValue) {
   }
 }
 
+template<typename T, typename Pred>
+static inline void WaitBitMaskPred(T* Futex, T BitMask, T ComparisonValue, Pred Predicate) {
+  auto AtomicFutex = std::atomic_ref<T>(*Futex);
+  T Result = AtomicFutex.load();
+
+  while (!Predicate(Result & BitMask, ComparisonValue)) {
+    Result = AtomicFutex.load();
+  }
+}
+
 template<typename T, typename TT>
 static inline bool Wait(T* Futex, TT ExpectedValue, const std::chrono::nanoseconds& Timeout) {
   auto AtomicFutex = std::atomic_ref<T>(*Futex);
diff --git a/Source/Windows/WOW64/Module.cpp b/Source/Windows/WOW64/Module.cpp
index bff9ae2da5..f623023154 100644
--- a/Source/Windows/WOW64/Module.cpp
+++ b/Source/Windows/WOW64/Module.cpp
@@ -95,6 +95,10 @@ struct TLS {
     return reinterpret_cast<std::atomic<uint32_t>&>(TEB->TlsSlots[FEXCore::ToUnderlying(Slot::CONTROL_WORD)]);
   }
 
+  uint32_t* ControlWordAddress() const {
+    return reinterpret_cast<uint32_t*>(&TEB->TlsSlots[FEXCore::ToUnderlying(Slot::CONTROL_WORD)]);
+  }
+
   CONTEXT*& EntryContext() const {
     return reinterpret_cast<CONTEXT*&>(TEB->TlsSlots[FEXCore::ToUnderlying(Slot::ENTRY_CONTEXT)]);
   }
@@ -822,8 +826,7 @@ NTSTATUS BTCpuSuspendLocalThread(HANDLE Thread, ULONG* Count) {
   }
 
   // Spin until the JIT is interrupted
-  while (TLS.ControlWord().load() & ControlBits::IN_JIT)
-    ;
+  FEXCore::Utils::SpinWaitLock::WaitBitMaskPred(TLS.ControlWordAddress(), ControlBits::IN_JIT, 0U, std::equal_to<>());
 
   // The JIT has now been interrupted and the context stored in the thread's CPU area is up-to-date
   if (Err = NtSuspendThread(*ThreadDup, Count); Err) {