From d67b1645c979acd621143079b72cbeb71ffc9ea7 Mon Sep 17 00:00:00 2001
From: Billy Laws <blaws05@gmail.com>
Date: Mon, 14 Jul 2025 19:19:40 +0100
Subject: [PATCH] FEXCore: Hold a frontend allocation for the call-ret stack

This can't be handled fully within FEXCore due to the frontend-specific
handling of guard pages. Frontends can populate this at init time and
are expected to handle setting the CPUState field and register as approriate.
---
 FEXCore/Source/Interface/Core/Core.cpp              | 11 ++++++++++-
 FEXCore/Source/Interface/Core/JIT/JIT.cpp           |  1 +
 FEXCore/include/FEXCore/Debug/InternalThreadState.h |  5 +++++
 3 files changed, 16 insertions(+), 1 deletion(-)

diff --git a/FEXCore/Source/Interface/Core/Core.cpp b/FEXCore/Source/Interface/Core/Core.cpp
index c30bbdf47..2ed90ba03 100644
--- a/FEXCore/Source/Interface/Core/Core.cpp
+++ b/FEXCore/Source/Interface/Core/Core.cpp
@@ -516,6 +516,7 @@ void ContextImpl::ClearCodeCache(FEXCore::Core::InternalThreadState* Thread, boo
     // Clear L1+L2 cache of this thread, and clear L3 cache across any threads using it
     Thread->LookupCache->ClearCache();
   }
+  Allocator::VirtualDontNeed(Thread->CallRetStackBase, FEXCore::Core::InternalThreadState::CALLRET_STACK_SIZE);
 }
 
 static void IRDumper(FEXCore::Core::InternalThreadState* Thread, IR::IREmitter* IREmitter, uint64_t GuestRIP) {
@@ -940,11 +941,19 @@ static void InvalidateGuestThreadCodeRange(FEXCore::Core::InternalThreadState* T
     Accumulator.emplace_back(std::move(it->second));
   }
 
+  bool InvalidatedAnyEntries = false;
   for (const auto& PageEntries : Accumulator) {
     for (const auto& Entry : PageEntries) {
-      ContextImpl::ThreadRemoveCodeEntry(Thread, Entry);
+      if (ContextImpl::ThreadRemoveCodeEntry(Thread, Entry)) {
+        InvalidatedAnyEntries = true;
+      }
     }
   }
+
+  if (InvalidatedAnyEntries) {
+    // This may cause access violations in the thread on Windows as zeroing is not atomic, this is handled by the frontend
+    Allocator::VirtualDontNeed(Thread->CallRetStackBase, FEXCore::Core::InternalThreadState::CALLRET_STACK_SIZE);
+  }
 }
 
 void ContextImpl::InvalidateGuestCodeRange(FEXCore::Core::InternalThreadState* Thread, InvalidatedEntryAccumulator& Accumulator,
diff --git a/FEXCore/Source/Interface/Core/JIT/JIT.cpp b/FEXCore/Source/Interface/Core/JIT/JIT.cpp
index 2a46f0788..bc93171f5 100644
--- a/FEXCore/Source/Interface/Core/JIT/JIT.cpp
+++ b/FEXCore/Source/Interface/Core/JIT/JIT.cpp
@@ -957,6 +957,7 @@ CPUBackend::CompiledCode Arm64JITCore::CompileCode(uint64_t Entry, uint64_t Size
       LOGMAN_THROW_A_FMT(CurrentCodeBuffer->LookupCache.get() == ThreadState->LookupCache->Shared, "INVARIANT VIOLATED: SharedLookupCache "
                                                                                                    "doesn't match up!\n");
       if (auto Prev = CheckCodeBufferUpdate()) {
+        Allocator::VirtualDontNeed(ThreadState->CallRetStackBase, FEXCore::Core::InternalThreadState::CALLRET_STACK_SIZE);
         ThreadState->LookupCache->ChangeGuestToHostMapping(*Prev, *CurrentCodeBuffer->LookupCache);
       }
 
diff --git a/FEXCore/include/FEXCore/Debug/InternalThreadState.h b/FEXCore/include/FEXCore/Debug/InternalThreadState.h
index 5c8bc424f..fb881ee54 100644
--- a/FEXCore/include/FEXCore/Debug/InternalThreadState.h
+++ b/FEXCore/include/FEXCore/Debug/InternalThreadState.h
@@ -105,6 +105,11 @@ struct InternalThreadState : public FEXCore::Allocator::FEXAllocOperators {
   ///< Data pointer for exclusive use by the frontend
   void* FrontendPtr;
 
+  static constexpr size_t CALLRET_STACK_SIZE {0x400000};
+
+  // The low address of the call-ret stack allocation (not including guard pages)
+  void* CallRetStackBase {};
+
   // BaseFrameState should always be at the end, directly before the interrupt fault page
   alignas(16) FEXCore::Core::CpuStateFrame BaseFrameState {};