[NativeAOT] thread suspension on Unix (#71187)

* missing pieces

* fixes

* pThreadToHijack

* NATIVE_CONTEXT

* at safe point

* enable TestConcurrentIsBackgroundProperty

* couple tweaks

* fixes after rebasing

* temporarily do not handle multireg returns

* support single reg return hijack

* disable return hijacking for ARM64

* enable safe point suspend

* enable multireg

* make gcc happy

* fixws for win-x64

* Do not wait for Unhijack.

* crude check for VirtualUnwind issue in inline suspend and some tweaks

* disable hijacking on ARM64/UNIX

* preserve  xmm1

* IsUnwindable check

* detect not-unwindable code, at least the most common cases.

* IsUnwindable tweaks

* detect epilogs

* hijack in epilogues.

* PR feedback (refactoring)

* Check the trap flag in `HijackCallback`  (optimization)

* Apply suggestions from code review

Co-authored-by: Jan Kotas <jkotas@microsoft.com>

* PR feedback (refactor/renames)

* handle "int 3" as not unwindable, not hijackable.

Co-authored-by: Jan Kotas <jkotas@microsoft.com>

Author: VSadov

src/coreclr/nativeaot/Runtime/CMakeLists.txt

@@ -175,6 +175,7 @@ else()
     unix/cgroup.cpp
     unix/HardwareExceptions.cpp
     unix/UnixContext.cpp
+    unix/UnixSignals.cpp
     unix/UnwindHelpers.cpp
     unix/UnixNativeCodeManager.cpp
     ../libunwind/src/Unwind-EHABI.cpp

src/coreclr/nativeaot/Runtime/ICodeManager.h

@@ -28,7 +28,7 @@ enum GCRefKind : unsigned char
     GCRK_Scalar         = 0x00,
     GCRK_Object         = 0x01,
     GCRK_Byref          = 0x02,
-#ifdef TARGET_ARM64
+#ifdef TARGET_64BIT
     // Composite return kinds for value types returned in two registers (encoded with two bits per register)
     GCRK_Scalar_Obj     = (GCRK_Object << 2) | GCRK_Scalar,
     GCRK_Obj_Obj        = (GCRK_Object << 2) | GCRK_Object,
@@ -66,6 +66,35 @@ inline GCRefKind TransitionFrameFlagsToReturnKind(uint64_t transFrameFlags)
     return returnKind;
 }
 
+#elif defined(TARGET_AMD64) 
+
+// Verify that we can use bitwise shifts to convert from GCRefKind to PInvokeTransitionFrameFlags and back
+C_ASSERT(PTFF_RAX_IS_GCREF == ((uint64_t)GCRK_Object << 16));
+C_ASSERT(PTFF_RAX_IS_BYREF == ((uint64_t)GCRK_Byref << 16));
+C_ASSERT(PTFF_RDX_IS_GCREF == ((uint64_t)GCRK_Scalar_Obj << 16));
+C_ASSERT(PTFF_RDX_IS_BYREF == ((uint64_t)GCRK_Scalar_Byref << 16));
+
+inline uint64_t ReturnKindToTransitionFrameFlags(GCRefKind returnKind)
+{
+    if (returnKind == GCRK_Scalar)
+        return 0;
+
+    return PTFF_SAVE_RAX | PTFF_SAVE_RDX | ((uint64_t)returnKind << 16);
+}
+
+inline GCRefKind TransitionFrameFlagsToReturnKind(uint64_t transFrameFlags)
+{
+    GCRefKind returnKind = (GCRefKind)((transFrameFlags & (PTFF_RAX_IS_GCREF | PTFF_RAX_IS_BYREF | PTFF_RDX_IS_GCREF | PTFF_RDX_IS_BYREF)) >> 16);
+#if defined(TARGET_UNIX)
+    ASSERT((returnKind == GCRK_Scalar) || ((transFrameFlags & PTFF_SAVE_RAX) && (transFrameFlags & PTFF_SAVE_RDX)));
+#else
+    ASSERT((returnKind == GCRK_Scalar) || (transFrameFlags & PTFF_SAVE_RAX));
+#endif
+    return returnKind;
+}
+
+#endif
+
 // Extract individual GCRefKind components from a composite return kind
 inline GCRefKind ExtractReg0ReturnKind(GCRefKind returnKind)
 {
@@ -78,7 +107,6 @@ inline GCRefKind ExtractReg1ReturnKind(GCRefKind returnKind)
     ASSERT(returnKind <= GCRK_LastValid);
     return (GCRefKind)(returnKind >> 2);
 }
-#endif // TARGET_ARM64
 
 //
 // MethodInfo is placeholder type used to allocate space for MethodInfo. Maximum size
@@ -162,6 +190,8 @@ class ICodeManager
     virtual uintptr_t GetConservativeUpperBoundForOutgoingArgs(MethodInfo *   pMethodInfo,
                                                                 REGDISPLAY *   pRegisterSet) = 0;
 
+    virtual bool IsUnwindable(PTR_VOID pvAddress) = 0;
+
     virtual bool GetReturnAddressHijackInfo(MethodInfo *    pMethodInfo,
                                             REGDISPLAY *    pRegisterSet,           // in
                                             PTR_PTR_VOID *  ppvRetAddrLocation,     // out

src/coreclr/nativeaot/Runtime/PalRedhawk.h

@@ -596,8 +596,6 @@ REDHAWK_PALIMPORT bool REDHAWK_PALAPI PalGetCompleteThreadContext(HANDLE hThread
 REDHAWK_PALIMPORT bool REDHAWK_PALAPI PalSetThreadContext(HANDLE hThread, _Out_ CONTEXT * pCtx);
 REDHAWK_PALIMPORT void REDHAWK_PALAPI PalRestoreContext(CONTEXT * pCtx);
 
-REDHAWK_PALIMPORT bool REDHAWK_PALAPI PalGetThreadContext(HANDLE hThread, _Out_ PAL_LIMITED_CONTEXT * pCtx);
-
 REDHAWK_PALIMPORT int32_t REDHAWK_PALAPI PalGetProcessCpuCount();
 
 // Retrieves the entire range of memory dedicated to the calling thread's stack.  This does
@@ -683,6 +681,13 @@ REDHAWK_PALIMPORT uint64_t REDHAWK_PALAPI PalGetTickCount64();
 REDHAWK_PALIMPORT void REDHAWK_PALAPI PalTerminateCurrentProcess(uint32_t exitCode);
 REDHAWK_PALIMPORT HANDLE REDHAWK_PALAPI PalGetModuleHandleFromPointer(_In_ void* pointer);
 
+#ifdef TARGET_UNIX
+struct UNIX_CONTEXT;
+#define NATIVE_CONTEXT UNIX_CONTEXT
+#else
+#define NATIVE_CONTEXT CONTEXT
+#endif
+
 #ifdef TARGET_UNIX
 REDHAWK_PALIMPORT void REDHAWK_PALAPI PalSetHardwareExceptionHandler(PHARDWARE_EXCEPTION_HANDLER handler);
 #else
@@ -693,8 +698,9 @@ typedef uint32_t (__stdcall *BackgroundCallback)(_In_opt_ void* pCallbackContext
 REDHAWK_PALIMPORT bool REDHAWK_PALAPI PalStartBackgroundGCThread(_In_ BackgroundCallback callback, _In_opt_ void* pCallbackContext);
 REDHAWK_PALIMPORT bool REDHAWK_PALAPI PalStartFinalizerThread(_In_ BackgroundCallback callback, _In_opt_ void* pCallbackContext);
 
-typedef UInt32_BOOL (*PalHijackCallback)(HANDLE hThread, _In_ PAL_LIMITED_CONTEXT* pThreadContext, _In_opt_ void* pCallbackContext);
-REDHAWK_PALIMPORT uint32_t REDHAWK_PALAPI PalHijack(HANDLE hThread, _In_ PalHijackCallback callback, _In_opt_ void* pCallbackContext);
+typedef void (*PalHijackCallback)(_In_ NATIVE_CONTEXT* pThreadContext, _In_opt_ void* pThreadToHijack);
+REDHAWK_PALIMPORT void REDHAWK_PALAPI PalHijack(HANDLE hThread, _In_opt_ void* pThreadToHijack);
+REDHAWK_PALIMPORT UInt32_BOOL REDHAWK_PALAPI PalRegisterHijackCallback(_In_ PalHijackCallback callback);
 
 #ifdef FEATURE_ETW
 REDHAWK_PALIMPORT bool REDHAWK_PALAPI PalEventEnabled(REGHANDLE regHandle, _In_ const EVENT_DESCRIPTOR* eventDescriptor);

src/coreclr/nativeaot/Runtime/StackFrameIterator.cpp

@@ -27,6 +27,10 @@
 #include "RuntimeInstance.h"
 #include "rhbinder.h"
 
+#ifdef TARGET_UNIX
+#include "UnixContext.h"
+#endif
+
 // warning C4061: enumerator '{blah}' in switch of enum '{blarg}' is not explicitly handled by a case label
 #pragma warning(disable:4061)
 
@@ -93,13 +97,11 @@ StackFrameIterator::StackFrameIterator(Thread * pThreadToWalk, PInvokeTransition
     STRESS_LOG0(LF_STACKWALK, LL_INFO10000, "----Init---- [ GC ]\n");
     ASSERT(!pThreadToWalk->DangerousCrossThreadIsHijacked());
 
-#ifdef FEATURE_SUSPEND_REDIRECTION
-    if (pInitialTransitionFrame == REDIRECTED_THREAD_MARKER)
+    if (pInitialTransitionFrame == INTERRUPTED_THREAD_MARKER)
     {
-        InternalInit(pThreadToWalk, pThreadToWalk->GetRedirectionContext(), GcStackWalkFlags | ActiveStackFrame);
+        InternalInit(pThreadToWalk, pThreadToWalk->GetInterruptedContext(), GcStackWalkFlags | ActiveStackFrame);
     }
     else
-#endif
     {
         InternalInit(pThreadToWalk, pInitialTransitionFrame, GcStackWalkFlags);
     }
@@ -109,7 +111,14 @@ StackFrameIterator::StackFrameIterator(Thread * pThreadToWalk, PInvokeTransition
 
 StackFrameIterator::StackFrameIterator(Thread * pThreadToWalk, PTR_PAL_LIMITED_CONTEXT pCtx)
 {
-    STRESS_LOG0(LF_STACKWALK, LL_INFO10000, "----Init---- [ hijack ]\n");
+    STRESS_LOG0(LF_STACKWALK, LL_INFO10000, "----Init with limited ctx---- [ hijack ]\n");
+    InternalInit(pThreadToWalk, pCtx, 0);
+    PrepareToYieldFrame();
+}
+
+StackFrameIterator::StackFrameIterator(Thread* pThreadToWalk, NATIVE_CONTEXT* pCtx)
+{
+    STRESS_LOG0(LF_STACKWALK, LL_INFO10000, "----Init with native ctx---- [ hijack ]\n");
     InternalInit(pThreadToWalk, pCtx, 0);
     PrepareToYieldFrame();
 }
@@ -283,15 +292,11 @@ void StackFrameIterator::InternalInit(Thread * pThreadToWalk, PInvokeTransitionF
     if (pFrame->m_Flags & PTFF_SAVE_R11)  { m_RegDisplay.pR11 = pPreservedRegsCursor++; }
 #endif // TARGET_AMD64
 
-    if (pFrame->m_Flags & PTFF_RAX_IS_GCREF)
-    {
-        m_pHijackedReturnValue = (PTR_RtuObjectRef) m_RegDisplay.pRax;
-        m_HijackedReturnValueKind = GCRK_Object;
-    }
-    if (pFrame->m_Flags & PTFF_RAX_IS_BYREF)
+    GCRefKind retValueKind = TransitionFrameFlagsToReturnKind(pFrame->m_Flags);
+    if (retValueKind != GCRK_Scalar)
     {
-        m_pHijackedReturnValue = (PTR_RtuObjectRef) m_RegDisplay.pRax;
-        m_HijackedReturnValueKind = GCRK_Byref;
+        m_pHijackedReturnValue = (PTR_RtuObjectRef)m_RegDisplay.pRax;
+        m_HijackedReturnValueKind = retValueKind;
     }
 
 #endif // TARGET_ARM
@@ -508,10 +513,9 @@ void StackFrameIterator::InternalInit(Thread * pThreadToWalk, PTR_PAL_LIMITED_CO
 #endif // TARGET_ARM
 }
 
-// Prepare to start a stack walk from the context listed in the supplied CONTEXT.
-// The supplied context can describe a location in either managed or unmanaged code.  In the
-// latter case the iterator is left in an invalid state when this function returns.
-void StackFrameIterator::InternalInit(Thread * pThreadToWalk, CONTEXT* pCtx, uint32_t dwFlags)
+// Prepare to start a stack walk from the context listed in the supplied NATIVE_CONTEXT.
+// The supplied context can describe a location in managed code.
+void StackFrameIterator::InternalInit(Thread * pThreadToWalk, NATIVE_CONTEXT* pCtx, uint32_t dwFlags)
 {
     ASSERT((dwFlags & MethodStateCalculated) == 0);
 
@@ -524,10 +528,8 @@ void StackFrameIterator::InternalInit(Thread * pThreadToWalk, CONTEXT* pCtx, uin
     // properly walk it in parallel.
     ResetNextExInfoForSP(pCtx->GetSp());
 
-    // This codepath is used by the hijack stackwalk and we can get arbitrary ControlPCs from there.  If this
-    // context has a non-managed control PC, then we're done.
-    if (!m_pInstance->IsManaged(dac_cast<PTR_VOID>(pCtx->GetIp())))
-        return;
+    // This codepath is used by the hijack stackwalk. The IP must be in managed code.
+    ASSERT(m_pInstance->IsManaged(dac_cast<PTR_VOID>(pCtx->GetIp())));
 
     //
     // control state
@@ -536,82 +538,90 @@ void StackFrameIterator::InternalInit(Thread * pThreadToWalk, CONTEXT* pCtx, uin
     m_RegDisplay.SP   = pCtx->GetSp();
     m_RegDisplay.IP   = pCtx->GetIp();
 
+#ifdef TARGET_UNIX
+#define PTR_TO_REG(ptr, reg) (&((ptr)->reg()))
+#else
+#define PTR_TO_REG(ptr, reg) (&((ptr)->reg))
+#endif
+
 #ifdef TARGET_ARM64
 
-    m_RegDisplay.pIP  = PTR_TO_MEMBER(CONTEXT, pCtx, Pc);
+    m_RegDisplay.pIP  = (PTR_PCODE)PTR_TO_REG(pCtx, Pc);
 
     //
     // preserved regs
     //
-    m_RegDisplay.pX19 = PTR_TO_MEMBER(CONTEXT, pCtx, X19);
-    m_RegDisplay.pX20 = PTR_TO_MEMBER(CONTEXT, pCtx, X20);
-    m_RegDisplay.pX21 = PTR_TO_MEMBER(CONTEXT, pCtx, X21);
-    m_RegDisplay.pX22 = PTR_TO_MEMBER(CONTEXT, pCtx, X22);
-    m_RegDisplay.pX23 = PTR_TO_MEMBER(CONTEXT, pCtx, X23);
-    m_RegDisplay.pX24 = PTR_TO_MEMBER(CONTEXT, pCtx, X24);
-    m_RegDisplay.pX25 = PTR_TO_MEMBER(CONTEXT, pCtx, X25);
-    m_RegDisplay.pX26 = PTR_TO_MEMBER(CONTEXT, pCtx, X26);
-    m_RegDisplay.pX27 = PTR_TO_MEMBER(CONTEXT, pCtx, X27);
-    m_RegDisplay.pX28 = PTR_TO_MEMBER(CONTEXT, pCtx, X28);
-    m_RegDisplay.pFP = PTR_TO_MEMBER(CONTEXT, pCtx, Fp);
-    m_RegDisplay.pLR = PTR_TO_MEMBER(CONTEXT, pCtx, Lr);
+    m_RegDisplay.pX19 = (PTR_UIntNative)PTR_TO_REG(pCtx, X19);
+    m_RegDisplay.pX20 = (PTR_UIntNative)PTR_TO_REG(pCtx, X20);
+    m_RegDisplay.pX21 = (PTR_UIntNative)PTR_TO_REG(pCtx, X21);
+    m_RegDisplay.pX22 = (PTR_UIntNative)PTR_TO_REG(pCtx, X22);
+    m_RegDisplay.pX23 = (PTR_UIntNative)PTR_TO_REG(pCtx, X23);
+    m_RegDisplay.pX24 = (PTR_UIntNative)PTR_TO_REG(pCtx, X24);
+    m_RegDisplay.pX25 = (PTR_UIntNative)PTR_TO_REG(pCtx, X25);
+    m_RegDisplay.pX26 = (PTR_UIntNative)PTR_TO_REG(pCtx, X26);
+    m_RegDisplay.pX27 = (PTR_UIntNative)PTR_TO_REG(pCtx, X27);
+    m_RegDisplay.pX28 = (PTR_UIntNative)PTR_TO_REG(pCtx, X28);
+    m_RegDisplay.pFP = (PTR_UIntNative)PTR_TO_REG(pCtx, Fp);
+    m_RegDisplay.pLR = (PTR_UIntNative)PTR_TO_REG(pCtx, Lr);
 
     //
     // scratch regs
     //
-    m_RegDisplay.pX0 = PTR_TO_MEMBER(CONTEXT, pCtx, X0);
-    m_RegDisplay.pX1 = PTR_TO_MEMBER(CONTEXT, pCtx, X1);
-    m_RegDisplay.pX2 = PTR_TO_MEMBER(CONTEXT, pCtx, X2);
-    m_RegDisplay.pX3 = PTR_TO_MEMBER(CONTEXT, pCtx, X3);
-    m_RegDisplay.pX4 = PTR_TO_MEMBER(CONTEXT, pCtx, X4);
-    m_RegDisplay.pX5 = PTR_TO_MEMBER(CONTEXT, pCtx, X5);
-    m_RegDisplay.pX6 = PTR_TO_MEMBER(CONTEXT, pCtx, X6);
-    m_RegDisplay.pX7 = PTR_TO_MEMBER(CONTEXT, pCtx, X7);
-    m_RegDisplay.pX8 = PTR_TO_MEMBER(CONTEXT, pCtx, X8);
-    m_RegDisplay.pX9 = PTR_TO_MEMBER(CONTEXT, pCtx, X9);
-    m_RegDisplay.pX10 = PTR_TO_MEMBER(CONTEXT, pCtx, X10);
-    m_RegDisplay.pX11 = PTR_TO_MEMBER(CONTEXT, pCtx, X11);
-    m_RegDisplay.pX12 = PTR_TO_MEMBER(CONTEXT, pCtx, X12);
-    m_RegDisplay.pX13 = PTR_TO_MEMBER(CONTEXT, pCtx, X13);
-    m_RegDisplay.pX14 = PTR_TO_MEMBER(CONTEXT, pCtx, X14);
-    m_RegDisplay.pX15 = PTR_TO_MEMBER(CONTEXT, pCtx, X15);
-    m_RegDisplay.pX16 = PTR_TO_MEMBER(CONTEXT, pCtx, X16);
-    m_RegDisplay.pX17 = PTR_TO_MEMBER(CONTEXT, pCtx, X17);
-    m_RegDisplay.pX18 = PTR_TO_MEMBER(CONTEXT, pCtx, X18);
+    m_RegDisplay.pX0 = (PTR_UIntNative)PTR_TO_REG(pCtx, X0);
+    m_RegDisplay.pX1 = (PTR_UIntNative)PTR_TO_REG(pCtx, X1);
+    m_RegDisplay.pX2 = (PTR_UIntNative)PTR_TO_REG(pCtx, X2);
+    m_RegDisplay.pX3 = (PTR_UIntNative)PTR_TO_REG(pCtx, X3);
+    m_RegDisplay.pX4 = (PTR_UIntNative)PTR_TO_REG(pCtx, X4);
+    m_RegDisplay.pX5 = (PTR_UIntNative)PTR_TO_REG(pCtx, X5);
+    m_RegDisplay.pX6 = (PTR_UIntNative)PTR_TO_REG(pCtx, X6);
+    m_RegDisplay.pX7 = (PTR_UIntNative)PTR_TO_REG(pCtx, X7);
+    m_RegDisplay.pX8 = (PTR_UIntNative)PTR_TO_REG(pCtx, X8);
+    m_RegDisplay.pX9 = (PTR_UIntNative)PTR_TO_REG(pCtx, X9);
+    m_RegDisplay.pX10 = (PTR_UIntNative)PTR_TO_REG(pCtx, X10);
+    m_RegDisplay.pX11 = (PTR_UIntNative)PTR_TO_REG(pCtx, X11);
+    m_RegDisplay.pX12 = (PTR_UIntNative)PTR_TO_REG(pCtx, X12);
+    m_RegDisplay.pX13 = (PTR_UIntNative)PTR_TO_REG(pCtx, X13);
+    m_RegDisplay.pX14 = (PTR_UIntNative)PTR_TO_REG(pCtx, X14);
+    m_RegDisplay.pX15 = (PTR_UIntNative)PTR_TO_REG(pCtx, X15);
+    m_RegDisplay.pX16 = (PTR_UIntNative)PTR_TO_REG(pCtx, X16);
+    m_RegDisplay.pX17 = (PTR_UIntNative)PTR_TO_REG(pCtx, X17);
+    m_RegDisplay.pX18 = (PTR_UIntNative)PTR_TO_REG(pCtx, X18);
 
 #elif defined(TARGET_X86) || defined(TARGET_AMD64)
 
-    m_RegDisplay.pIP  = (PTR_PCODE)PTR_TO_MEMBER(CONTEXT, pCtx, Rip);
+    m_RegDisplay.pIP  = (PTR_PCODE)PTR_TO_REG(pCtx, Rip);
 
     //
     // preserved regs
     //
-    m_RegDisplay.pRbp = (PTR_UIntNative)PTR_TO_MEMBER(CONTEXT, pCtx, Rbp);
-    m_RegDisplay.pRsi = (PTR_UIntNative)PTR_TO_MEMBER(CONTEXT, pCtx, Rsi);
-    m_RegDisplay.pRdi = (PTR_UIntNative)PTR_TO_MEMBER(CONTEXT, pCtx, Rdi);
-    m_RegDisplay.pRbx = (PTR_UIntNative)PTR_TO_MEMBER(CONTEXT, pCtx, Rbx);
+    m_RegDisplay.pRbp = (PTR_UIntNative)PTR_TO_REG(pCtx, Rbp);
+    m_RegDisplay.pRsi = (PTR_UIntNative)PTR_TO_REG(pCtx, Rsi);
+    m_RegDisplay.pRdi = (PTR_UIntNative)PTR_TO_REG(pCtx, Rdi);
+    m_RegDisplay.pRbx = (PTR_UIntNative)PTR_TO_REG(pCtx, Rbx);
 #ifdef TARGET_AMD64     
-    m_RegDisplay.pR12 = (PTR_UIntNative)PTR_TO_MEMBER(CONTEXT, pCtx, R12);
-    m_RegDisplay.pR13 = (PTR_UIntNative)PTR_TO_MEMBER(CONTEXT, pCtx, R13);
-    m_RegDisplay.pR14 = (PTR_UIntNative)PTR_TO_MEMBER(CONTEXT, pCtx, R14);
-    m_RegDisplay.pR15 = (PTR_UIntNative)PTR_TO_MEMBER(CONTEXT, pCtx, R15);
+    m_RegDisplay.pR12 = (PTR_UIntNative)PTR_TO_REG(pCtx, R12);
+    m_RegDisplay.pR13 = (PTR_UIntNative)PTR_TO_REG(pCtx, R13);
+    m_RegDisplay.pR14 = (PTR_UIntNative)PTR_TO_REG(pCtx, R14);
+    m_RegDisplay.pR15 = (PTR_UIntNative)PTR_TO_REG(pCtx, R15);
 #endif // TARGET_AMD64  
 
     //                  
     // scratch regs     
     //                  
-    m_RegDisplay.pRax = (PTR_UIntNative)PTR_TO_MEMBER(CONTEXT, pCtx, Rax);
-    m_RegDisplay.pRcx = (PTR_UIntNative)PTR_TO_MEMBER(CONTEXT, pCtx, Rcx);
-    m_RegDisplay.pRdx = (PTR_UIntNative)PTR_TO_MEMBER(CONTEXT, pCtx, Rdx);
+    m_RegDisplay.pRax = (PTR_UIntNative)PTR_TO_REG(pCtx, Rax);
+    m_RegDisplay.pRcx = (PTR_UIntNative)PTR_TO_REG(pCtx, Rcx);
+    m_RegDisplay.pRdx = (PTR_UIntNative)PTR_TO_REG(pCtx, Rdx);
 #ifdef TARGET_AMD64     
-    m_RegDisplay.pR8  = (PTR_UIntNative)PTR_TO_MEMBER(CONTEXT, pCtx, R8);
-    m_RegDisplay.pR9  = (PTR_UIntNative)PTR_TO_MEMBER(CONTEXT, pCtx, R9);
-    m_RegDisplay.pR10 = (PTR_UIntNative)PTR_TO_MEMBER(CONTEXT, pCtx, R10);
-    m_RegDisplay.pR11 = (PTR_UIntNative)PTR_TO_MEMBER(CONTEXT, pCtx, R11);
+    m_RegDisplay.pR8  = (PTR_UIntNative)PTR_TO_REG(pCtx, R8);
+    m_RegDisplay.pR9  = (PTR_UIntNative)PTR_TO_REG(pCtx, R9);
+    m_RegDisplay.pR10 = (PTR_UIntNative)PTR_TO_REG(pCtx, R10);
+    m_RegDisplay.pR11 = (PTR_UIntNative)PTR_TO_REG(pCtx, R11);
 #endif // TARGET_AMD64
 #else
     PORTABILITY_ASSERT("StackFrameIterator::InternalInit");
 #endif // TARGET_ARM
+
+#undef PTR_TO_REG
 }
 
 PTR_VOID StackFrameIterator::HandleExCollide(PTR_ExInfo pExInfo)

src/coreclr/nativeaot/Runtime/StackFrameIterator.h

@@ -34,6 +34,7 @@ class StackFrameIterator
 public:
     StackFrameIterator() {}
     StackFrameIterator(Thread * pThreadToWalk, PInvokeTransitionFrame* pInitialTransitionFrame);
+    StackFrameIterator(Thread* pThreadToWalk, NATIVE_CONTEXT* pCtx);
     StackFrameIterator(Thread * pThreadToWalk, PTR_PAL_LIMITED_CONTEXT pCtx);
 
     bool             IsValid();
@@ -82,7 +83,7 @@ class StackFrameIterator
 
     void InternalInit(Thread * pThreadToWalk, PTR_PInvokeTransitionFrame pFrame, uint32_t dwFlags); // GC stackwalk
     void InternalInit(Thread * pThreadToWalk, PTR_PAL_LIMITED_CONTEXT pCtx, uint32_t dwFlags);  // EH and hijack stackwalk, and collided unwind
-    void InternalInit(Thread * pThreadToWalk, CONTEXT* pCtx, uint32_t dwFlags);  // GC stackwalk of redirected thread
+    void InternalInit(Thread * pThreadToWalk, NATIVE_CONTEXT* pCtx, uint32_t dwFlags);  // GC stackwalk of redirected thread
 
     void InternalInitForEH(Thread * pThreadToWalk, PAL_LIMITED_CONTEXT * pCtx, bool instructionFault); // EH stackwalk
     void InternalInitForStackTrace();  // Environment.StackTrace

src/coreclr/nativeaot/Runtime/amd64/AsmMacros.inc

@@ -365,11 +365,11 @@ PTFF_SAVE_R14           equ 00000040h
 PTFF_SAVE_R15           equ 00000080h
 PTFF_SAVE_ALL_PRESERVED equ 000000F7h   ;; NOTE: RBP is not included in this set!
 PTFF_SAVE_RSP           equ 00008000h
-PTFF_SAVE_RAX           equ 00000100h   ;; RAX is saved if it contains a GC ref and we're in hijack handler
+PTFF_SAVE_RAX           equ 00000100h   ;; RAX is saved in hijack handler - in case it contains a GC ref
 PTFF_SAVE_ALL_SCRATCH   equ 00007F00h
 PTFF_RAX_IS_GCREF       equ 00010000h   ;; iff PTFF_SAVE_RAX: set -> eax is Object, clear -> eax is scalar
 PTFF_RAX_IS_BYREF       equ 00020000h   ;; iff PTFF_SAVE_RAX: set -> eax is ByRef, clear -> eax is Object or scalar
-PTFF_THREAD_ABORT       equ 00040000h   ;; indicates that ThreadAbortException should be thrown when returning from the transition
+PTFF_THREAD_ABORT       equ 00100000h   ;; indicates that ThreadAbortException should be thrown when returning from the transition
 
 ;; These must match the TrapThreadsFlags enum
 TrapThreadsFlags_None            equ 0