Fix x86 runtime async frame pointer mismatch in GetSpForDiagnosticReporting

[tommcdon]

For certain runtime async frames this resulted in the ICorDebugManagedCallback2::Exception to return a null ICorDebugFrame for DEBUG_EXCEPTION_CATCH_HANDLER_FOUND notifications. The fix addresses this by adjusting GetSpForDiagnosticReporting to account for runtime async variant method stack layout differences on x86.

[dotnet-policy-service]

Tagging subscribers to this area: @steveisok, @tommcdon, @dotnet/dotnet-diag
See info in area-owners.md if you want to be subscribed.

[Copilot]

Pull request overview

Adjusts the stack pointer reported to the debugger on x86 for runtime “async call” frames so that ICorDebugManagedCallback2::Exception callbacks (notably DEBUG_EXCEPTION_CATCH_HANDLER_FOUND) can resolve a non-null ICorDebugFrame.

Changes:

• Extends GetSpForDiagnosticReporting to optionally accept a MethodDesc* and apply an extra x86 adjustment for runtime async methods (IsAsyncMethod()).
• Updates exception/debugger callback sites to pass the current MethodDesc* into GetSpForDiagnosticReporting.

Comments suppressed due to low confidence (1)

src/coreclr/vm/exceptionhandling.cpp:2968

• pMD is only referenced under ESTABLISHER_FRAME_ADDRESS_IS_CALLER_SP + TARGET_X86. In other builds (e.g., x64 Unix where -Wall is enabled, often with -Werror), this new parameter can become unused and may trigger an unused-parameter warning. Consider adding UNREFERENCED_PARAMETER(pMD); in the #else path and/or in the non-x86 path to keep all configurations warning-free.

static TADDR GetSpForDiagnosticReporting(REGDISPLAY *pRD, MethodDesc *pMD = NULL)
{
#ifdef ESTABLISHER_FRAME_ADDRESS_IS_CALLER_SP
TADDR sp = CallerStackFrame::FromRegDisplay(pRD).SP;
#if defined(TARGET_X86)
sp -= sizeof(TADDR);
// On x86, runtime async methods have stack parameters that cause CallerSP
// to sit above the parameter area. The DBI uses PCTAddr as the frame
// pointer, which is at the return address (below the parameters).
// Subtract an extra sizeof(TADDR) to account for the stack parameter.
if (pMD != NULL && pMD->IsAsyncMethod())
{
sp -= sizeof(TADDR);
}
#endif
return sp;
#else
return GetSP(pRD->pCurrentContext);
#endif

[janvorli]

@tommcdon I wonder - was this always a problem? I am asking since there was a change in the ifdef in this code in January (#122833) from

#if defined(FEATURE_EH_FUNCLETS) && defined(TARGET_X86)
sp -= sizeof(TADDR); // For X86 with funclets we want the address 1 pointer into the callee.
#endif // defined(FEATURE_EH_FUNCLETS) && defined(TARGET_X86)

I wonder if the ifdef really meant to be for Linux x86 only (which was previously the only one with defined(FEATURE_EH_FUNCLETS)) and was unrelated to the funclets per se.

[tommcdon]

@tommcdon I wonder - was this always a problem?

Yes I think you are correct. A previous fix for this was on #115630.

[tommcdon]

For certain runtime async frames this resulted in the ICorDebugManagedCallback2::Exception to return a null ICorDebugFrame for DEBUG_EXCEPTION_CATCH_HANDLER_FOUND notifications. The fix addresses this by adjusting GetSpForDiagnosticReporting to account for runtime async variant method stack layout differences on x86.

[dotnet-policy-service]

Tagging subscribers to this area: @steveisok, @tommcdon, @dotnet/dotnet-diag
See info in area-owners.md if you want to be subscribed.

[Copilot]

Pull request overview

Adjusts the stack pointer reported to the debugger on x86 for runtime “async call” frames so that ICorDebugManagedCallback2::Exception callbacks (notably DEBUG_EXCEPTION_CATCH_HANDLER_FOUND) can resolve a non-null ICorDebugFrame.

Changes:

• Extends GetSpForDiagnosticReporting to optionally accept a MethodDesc* and apply an extra x86 adjustment for runtime async methods (IsAsyncMethod()).
• Updates exception/debugger callback sites to pass the current MethodDesc* into GetSpForDiagnosticReporting.

Comments suppressed due to low confidence (1)

src/coreclr/vm/exceptionhandling.cpp:2968

• pMD is only referenced under ESTABLISHER_FRAME_ADDRESS_IS_CALLER_SP + TARGET_X86. In other builds (e.g., x64 Unix where -Wall is enabled, often with -Werror), this new parameter can become unused and may trigger an unused-parameter warning. Consider adding UNREFERENCED_PARAMETER(pMD); in the #else path and/or in the non-x86 path to keep all configurations warning-free.

static TADDR GetSpForDiagnosticReporting(REGDISPLAY *pRD, MethodDesc *pMD = NULL)
{
#ifdef ESTABLISHER_FRAME_ADDRESS_IS_CALLER_SP
TADDR sp = CallerStackFrame::FromRegDisplay(pRD).SP;
#if defined(TARGET_X86)
sp -= sizeof(TADDR);
// On x86, runtime async methods have stack parameters that cause CallerSP
// to sit above the parameter area. The DBI uses PCTAddr as the frame
// pointer, which is at the return address (below the parameters).
// Subtract an extra sizeof(TADDR) to account for the stack parameter.
if (pMD != NULL && pMD->IsAsyncMethod())
{
sp -= sizeof(TADDR);
}
#endif
return sp;
#else
return GetSP(pRD->pCurrentContext);
#endif

[janvorli]

@tommcdon I wonder - was this always a problem? I am asking since there was a change in the ifdef in this code in January (#122833) from

#if defined(FEATURE_EH_FUNCLETS) && defined(TARGET_X86)
sp -= sizeof(TADDR); // For X86 with funclets we want the address 1 pointer into the callee.
#endif // defined(FEATURE_EH_FUNCLETS) && defined(TARGET_X86)

I wonder if the ifdef really meant to be for Linux x86 only (which was previously the only one with defined(FEATURE_EH_FUNCLETS)) and was unrelated to the funclets per se.

[tommcdon]

@tommcdon I wonder - was this always a problem?

Yes I think you are correct. A previous fix for this was on #115630.

[tommcdon]

For certain runtime async frames this resulted in the ICorDebugManagedCallback2::Exception to return a null ICorDebugFrame for DEBUG_EXCEPTION_CATCH_HANDLER_FOUND notifications. The fix addresses this by adjusting GetSpForDiagnosticReporting to account for runtime async variant method stack layout differences on x86.

[dotnet-policy-service]

Tagging subscribers to this area: @steveisok, @tommcdon, @dotnet/dotnet-diag
See info in area-owners.md if you want to be subscribed.

[Copilot]

Pull request overview

Adjusts the stack pointer reported to the debugger on x86 for runtime “async call” frames so that ICorDebugManagedCallback2::Exception callbacks (notably DEBUG_EXCEPTION_CATCH_HANDLER_FOUND) can resolve a non-null ICorDebugFrame.

Changes:

• Extends GetSpForDiagnosticReporting to optionally accept a MethodDesc* and apply an extra x86 adjustment for runtime async methods (IsAsyncMethod()).
• Updates exception/debugger callback sites to pass the current MethodDesc* into GetSpForDiagnosticReporting.

Comments suppressed due to low confidence (1)

src/coreclr/vm/exceptionhandling.cpp:2968

• pMD is only referenced under ESTABLISHER_FRAME_ADDRESS_IS_CALLER_SP + TARGET_X86. In other builds (e.g., x64 Unix where -Wall is enabled, often with -Werror), this new parameter can become unused and may trigger an unused-parameter warning. Consider adding UNREFERENCED_PARAMETER(pMD); in the #else path and/or in the non-x86 path to keep all configurations warning-free.

static TADDR GetSpForDiagnosticReporting(REGDISPLAY *pRD, MethodDesc *pMD = NULL)
{
#ifdef ESTABLISHER_FRAME_ADDRESS_IS_CALLER_SP
TADDR sp = CallerStackFrame::FromRegDisplay(pRD).SP;
#if defined(TARGET_X86)
sp -= sizeof(TADDR);
// On x86, runtime async methods have stack parameters that cause CallerSP
// to sit above the parameter area. The DBI uses PCTAddr as the frame
// pointer, which is at the return address (below the parameters).
// Subtract an extra sizeof(TADDR) to account for the stack parameter.
if (pMD != NULL && pMD->IsAsyncMethod())
{
sp -= sizeof(TADDR);
}
#endif
return sp;
#else
return GetSP(pRD->pCurrentContext);
#endif

[janvorli]

@tommcdon I wonder - was this always a problem? I am asking since there was a change in the ifdef in this code in January (#122833) from

#if defined(FEATURE_EH_FUNCLETS) && defined(TARGET_X86)
sp -= sizeof(TADDR); // For X86 with funclets we want the address 1 pointer into the callee.
#endif // defined(FEATURE_EH_FUNCLETS) && defined(TARGET_X86)

I wonder if the ifdef really meant to be for Linux x86 only (which was previously the only one with defined(FEATURE_EH_FUNCLETS)) and was unrelated to the funclets per se.

[tommcdon]

@tommcdon I wonder - was this always a problem?

Yes I think you are correct. A previous fix for this was on #115630.

[tommcdon]

For certain runtime async frames this resulted in the ICorDebugManagedCallback2::Exception to return a null ICorDebugFrame for DEBUG_EXCEPTION_CATCH_HANDLER_FOUND notifications. The fix addresses this by adjusting GetSpForDiagnosticReporting to account for runtime async variant method stack layout differences on x86.

[dotnet-policy-service]

Tagging subscribers to this area: @steveisok, @tommcdon, @dotnet/dotnet-diag
See info in area-owners.md if you want to be subscribed.

[Copilot]

Pull request overview

Adjusts the stack pointer reported to the debugger on x86 for runtime “async call” frames so that ICorDebugManagedCallback2::Exception callbacks (notably DEBUG_EXCEPTION_CATCH_HANDLER_FOUND) can resolve a non-null ICorDebugFrame.

Changes:

• Extends GetSpForDiagnosticReporting to optionally accept a MethodDesc* and apply an extra x86 adjustment for runtime async methods (IsAsyncMethod()).
• Updates exception/debugger callback sites to pass the current MethodDesc* into GetSpForDiagnosticReporting.

Comments suppressed due to low confidence (1)

src/coreclr/vm/exceptionhandling.cpp:2968

• pMD is only referenced under ESTABLISHER_FRAME_ADDRESS_IS_CALLER_SP + TARGET_X86. In other builds (e.g., x64 Unix where -Wall is enabled, often with -Werror), this new parameter can become unused and may trigger an unused-parameter warning. Consider adding UNREFERENCED_PARAMETER(pMD); in the #else path and/or in the non-x86 path to keep all configurations warning-free.

static TADDR GetSpForDiagnosticReporting(REGDISPLAY *pRD, MethodDesc *pMD = NULL)
{
#ifdef ESTABLISHER_FRAME_ADDRESS_IS_CALLER_SP
TADDR sp = CallerStackFrame::FromRegDisplay(pRD).SP;
#if defined(TARGET_X86)
sp -= sizeof(TADDR);
// On x86, runtime async methods have stack parameters that cause CallerSP
// to sit above the parameter area. The DBI uses PCTAddr as the frame
// pointer, which is at the return address (below the parameters).
// Subtract an extra sizeof(TADDR) to account for the stack parameter.
if (pMD != NULL && pMD->IsAsyncMethod())
{
sp -= sizeof(TADDR);
}
#endif
return sp;
#else
return GetSP(pRD->pCurrentContext);
#endif

[janvorli]

@tommcdon I wonder - was this always a problem? I am asking since there was a change in the ifdef in this code in January (#122833) from

#if defined(FEATURE_EH_FUNCLETS) && defined(TARGET_X86)
sp -= sizeof(TADDR); // For X86 with funclets we want the address 1 pointer into the callee.
#endif // defined(FEATURE_EH_FUNCLETS) && defined(TARGET_X86)

I wonder if the ifdef really meant to be for Linux x86 only (which was previously the only one with defined(FEATURE_EH_FUNCLETS)) and was unrelated to the funclets per se.

[tommcdon]

@tommcdon I wonder - was this always a problem?

Yes I think you are correct. A previous fix for this was on #115630.

[tommcdon]

For certain runtime async frames this resulted in the ICorDebugManagedCallback2::Exception to return a null ICorDebugFrame for DEBUG_EXCEPTION_CATCH_HANDLER_FOUND notifications. The fix addresses this by adjusting GetSpForDiagnosticReporting to account for runtime async variant method stack layout differences on x86.

[dotnet-policy-service]

Tagging subscribers to this area: @steveisok, @tommcdon, @dotnet/dotnet-diag
See info in area-owners.md if you want to be subscribed.

[Copilot]

Pull request overview

Adjusts the stack pointer reported to the debugger on x86 for runtime “async call” frames so that ICorDebugManagedCallback2::Exception callbacks (notably DEBUG_EXCEPTION_CATCH_HANDLER_FOUND) can resolve a non-null ICorDebugFrame.

Changes:

• Extends GetSpForDiagnosticReporting to optionally accept a MethodDesc* and apply an extra x86 adjustment for runtime async methods (IsAsyncMethod()).
• Updates exception/debugger callback sites to pass the current MethodDesc* into GetSpForDiagnosticReporting.

Comments suppressed due to low confidence (1)

src/coreclr/vm/exceptionhandling.cpp:2968

• pMD is only referenced under ESTABLISHER_FRAME_ADDRESS_IS_CALLER_SP + TARGET_X86. In other builds (e.g., x64 Unix where -Wall is enabled, often with -Werror), this new parameter can become unused and may trigger an unused-parameter warning. Consider adding UNREFERENCED_PARAMETER(pMD); in the #else path and/or in the non-x86 path to keep all configurations warning-free.

static TADDR GetSpForDiagnosticReporting(REGDISPLAY *pRD, MethodDesc *pMD = NULL)
{
#ifdef ESTABLISHER_FRAME_ADDRESS_IS_CALLER_SP
TADDR sp = CallerStackFrame::FromRegDisplay(pRD).SP;
#if defined(TARGET_X86)
sp -= sizeof(TADDR);
// On x86, runtime async methods have stack parameters that cause CallerSP
// to sit above the parameter area. The DBI uses PCTAddr as the frame
// pointer, which is at the return address (below the parameters).
// Subtract an extra sizeof(TADDR) to account for the stack parameter.
if (pMD != NULL && pMD->IsAsyncMethod())
{
sp -= sizeof(TADDR);
}
#endif
return sp;
#else
return GetSP(pRD->pCurrentContext);
#endif

[janvorli]

@tommcdon I wonder - was this always a problem? I am asking since there was a change in the ifdef in this code in January (#122833) from

#if defined(FEATURE_EH_FUNCLETS) && defined(TARGET_X86)
sp -= sizeof(TADDR); // For X86 with funclets we want the address 1 pointer into the callee.
#endif // defined(FEATURE_EH_FUNCLETS) && defined(TARGET_X86)

I wonder if the ifdef really meant to be for Linux x86 only (which was previously the only one with defined(FEATURE_EH_FUNCLETS)) and was unrelated to the funclets per se.

[tommcdon]

@tommcdon I wonder - was this always a problem?

Yes I think you are correct. A previous fix for this was on #115630.

[tommcdon]

For certain runtime async frames this resulted in the ICorDebugManagedCallback2::Exception to return a null ICorDebugFrame for DEBUG_EXCEPTION_CATCH_HANDLER_FOUND notifications. The fix addresses this by adjusting GetSpForDiagnosticReporting to account for runtime async variant method stack layout differences on x86.

[dotnet-policy-service]

Tagging subscribers to this area: @steveisok, @tommcdon, @dotnet/dotnet-diag
See info in area-owners.md if you want to be subscribed.

[Copilot]

Pull request overview

Adjusts the stack pointer reported to the debugger on x86 for runtime “async call” frames so that ICorDebugManagedCallback2::Exception callbacks (notably DEBUG_EXCEPTION_CATCH_HANDLER_FOUND) can resolve a non-null ICorDebugFrame.

Changes:

• Extends GetSpForDiagnosticReporting to optionally accept a MethodDesc* and apply an extra x86 adjustment for runtime async methods (IsAsyncMethod()).
• Updates exception/debugger callback sites to pass the current MethodDesc* into GetSpForDiagnosticReporting.

Comments suppressed due to low confidence (1)

src/coreclr/vm/exceptionhandling.cpp:2968

• pMD is only referenced under ESTABLISHER_FRAME_ADDRESS_IS_CALLER_SP + TARGET_X86. In other builds (e.g., x64 Unix where -Wall is enabled, often with -Werror), this new parameter can become unused and may trigger an unused-parameter warning. Consider adding UNREFERENCED_PARAMETER(pMD); in the #else path and/or in the non-x86 path to keep all configurations warning-free.

static TADDR GetSpForDiagnosticReporting(REGDISPLAY *pRD, MethodDesc *pMD = NULL)
{
#ifdef ESTABLISHER_FRAME_ADDRESS_IS_CALLER_SP
TADDR sp = CallerStackFrame::FromRegDisplay(pRD).SP;
#if defined(TARGET_X86)
sp -= sizeof(TADDR);
// On x86, runtime async methods have stack parameters that cause CallerSP
// to sit above the parameter area. The DBI uses PCTAddr as the frame
// pointer, which is at the return address (below the parameters).
// Subtract an extra sizeof(TADDR) to account for the stack parameter.
if (pMD != NULL && pMD->IsAsyncMethod())
{
sp -= sizeof(TADDR);
}
#endif
return sp;
#else
return GetSP(pRD->pCurrentContext);
#endif

[janvorli]

@tommcdon I wonder - was this always a problem? I am asking since there was a change in the ifdef in this code in January (#122833) from

#if defined(FEATURE_EH_FUNCLETS) && defined(TARGET_X86)
sp -= sizeof(TADDR); // For X86 with funclets we want the address 1 pointer into the callee.
#endif // defined(FEATURE_EH_FUNCLETS) && defined(TARGET_X86)

I wonder if the ifdef really meant to be for Linux x86 only (which was previously the only one with defined(FEATURE_EH_FUNCLETS)) and was unrelated to the funclets per se.

[tommcdon]

@tommcdon I wonder - was this always a problem?

Yes I think you are correct. A previous fix for this was on #115630.

[tommcdon]

For certain runtime async frames this resulted in the ICorDebugManagedCallback2::Exception to return a null ICorDebugFrame for DEBUG_EXCEPTION_CATCH_HANDLER_FOUND notifications. The fix addresses this by adjusting GetSpForDiagnosticReporting to account for runtime async variant method stack layout differences on x86.

[dotnet-policy-service]

Tagging subscribers to this area: @steveisok, @tommcdon, @dotnet/dotnet-diag
See info in area-owners.md if you want to be subscribed.

[Copilot]

Pull request overview

Adjusts the stack pointer reported to the debugger on x86 for runtime “async call” frames so that ICorDebugManagedCallback2::Exception callbacks (notably DEBUG_EXCEPTION_CATCH_HANDLER_FOUND) can resolve a non-null ICorDebugFrame.

Changes:

• Extends GetSpForDiagnosticReporting to optionally accept a MethodDesc* and apply an extra x86 adjustment for runtime async methods (IsAsyncMethod()).
• Updates exception/debugger callback sites to pass the current MethodDesc* into GetSpForDiagnosticReporting.

Comments suppressed due to low confidence (1)

src/coreclr/vm/exceptionhandling.cpp:2968

• pMD is only referenced under ESTABLISHER_FRAME_ADDRESS_IS_CALLER_SP + TARGET_X86. In other builds (e.g., x64 Unix where -Wall is enabled, often with -Werror), this new parameter can become unused and may trigger an unused-parameter warning. Consider adding UNREFERENCED_PARAMETER(pMD); in the #else path and/or in the non-x86 path to keep all configurations warning-free.

static TADDR GetSpForDiagnosticReporting(REGDISPLAY *pRD, MethodDesc *pMD = NULL)
{
#ifdef ESTABLISHER_FRAME_ADDRESS_IS_CALLER_SP
TADDR sp = CallerStackFrame::FromRegDisplay(pRD).SP;
#if defined(TARGET_X86)
sp -= sizeof(TADDR);
// On x86, runtime async methods have stack parameters that cause CallerSP
// to sit above the parameter area. The DBI uses PCTAddr as the frame
// pointer, which is at the return address (below the parameters).
// Subtract an extra sizeof(TADDR) to account for the stack parameter.
if (pMD != NULL && pMD->IsAsyncMethod())
{
sp -= sizeof(TADDR);
}
#endif
return sp;
#else
return GetSP(pRD->pCurrentContext);
#endif

[janvorli]

@tommcdon I wonder - was this always a problem? I am asking since there was a change in the ifdef in this code in January (#122833) from

#if defined(FEATURE_EH_FUNCLETS) && defined(TARGET_X86)
sp -= sizeof(TADDR); // For X86 with funclets we want the address 1 pointer into the callee.
#endif // defined(FEATURE_EH_FUNCLETS) && defined(TARGET_X86)

I wonder if the ifdef really meant to be for Linux x86 only (which was previously the only one with defined(FEATURE_EH_FUNCLETS)) and was unrelated to the funclets per se.

[tommcdon]

@tommcdon I wonder - was this always a problem?

Yes I think you are correct. A previous fix for this was on #115630.

[tommcdon]

For certain runtime async frames this resulted in the ICorDebugManagedCallback2::Exception to return a null ICorDebugFrame for DEBUG_EXCEPTION_CATCH_HANDLER_FOUND notifications. The fix addresses this by adjusting GetSpForDiagnosticReporting to account for runtime async variant method stack layout differences on x86.

[dotnet-policy-service]

Tagging subscribers to this area: @steveisok, @tommcdon, @dotnet/dotnet-diag
See info in area-owners.md if you want to be subscribed.

[Copilot]

Pull request overview

Adjusts the stack pointer reported to the debugger on x86 for runtime “async call” frames so that ICorDebugManagedCallback2::Exception callbacks (notably DEBUG_EXCEPTION_CATCH_HANDLER_FOUND) can resolve a non-null ICorDebugFrame.

Changes:

• Extends GetSpForDiagnosticReporting to optionally accept a MethodDesc* and apply an extra x86 adjustment for runtime async methods (IsAsyncMethod()).
• Updates exception/debugger callback sites to pass the current MethodDesc* into GetSpForDiagnosticReporting.

Comments suppressed due to low confidence (1)

src/coreclr/vm/exceptionhandling.cpp:2968

• pMD is only referenced under ESTABLISHER_FRAME_ADDRESS_IS_CALLER_SP + TARGET_X86. In other builds (e.g., x64 Unix where -Wall is enabled, often with -Werror), this new parameter can become unused and may trigger an unused-parameter warning. Consider adding UNREFERENCED_PARAMETER(pMD); in the #else path and/or in the non-x86 path to keep all configurations warning-free.

static TADDR GetSpForDiagnosticReporting(REGDISPLAY *pRD, MethodDesc *pMD = NULL)
{
#ifdef ESTABLISHER_FRAME_ADDRESS_IS_CALLER_SP
TADDR sp = CallerStackFrame::FromRegDisplay(pRD).SP;
#if defined(TARGET_X86)
sp -= sizeof(TADDR);
// On x86, runtime async methods have stack parameters that cause CallerSP
// to sit above the parameter area. The DBI uses PCTAddr as the frame
// pointer, which is at the return address (below the parameters).
// Subtract an extra sizeof(TADDR) to account for the stack parameter.
if (pMD != NULL && pMD->IsAsyncMethod())
{
sp -= sizeof(TADDR);
}
#endif
return sp;
#else
return GetSP(pRD->pCurrentContext);
#endif

[janvorli]

@tommcdon I wonder - was this always a problem? I am asking since there was a change in the ifdef in this code in January (#122833) from

#if defined(FEATURE_EH_FUNCLETS) && defined(TARGET_X86)
sp -= sizeof(TADDR); // For X86 with funclets we want the address 1 pointer into the callee.
#endif // defined(FEATURE_EH_FUNCLETS) && defined(TARGET_X86)

I wonder if the ifdef really meant to be for Linux x86 only (which was previously the only one with defined(FEATURE_EH_FUNCLETS)) and was unrelated to the funclets per se.

[tommcdon]

@tommcdon I wonder - was this always a problem?

Yes I think you are correct. A previous fix for this was on #115630.