[NativeAOT] Fix stack trace iteration over exceptions

[filipnavara]

Fixes #101364

The test b08944b on x86 in Release causes the bomb() method to be inlined. The stack trace iterator then incorrectly adjusts the pointer to the faulting instruction by -1 which causes the try block not to be matched.

On x64 the inlined generated code for the bomb() method ended up being

00007ff7`3ecda705 33c9           xor     ecx, ecx
00007ff7`3ecda707 833900         cmp     dword ptr [rcx], 0

On x86 it is

004478ef 833d0000000000     cmp     dword ptr ds:[0], 0

Hence, for x64 the previous instruction to the faulting instruction was still in the try region. On x86 that's not the case.

[filipnavara]

Alternatively I can change to code to remove the ApplyReturnAddressAdjustment flag in the code path that sets ExCollide. Let's wait for the CI to see what breaks...

That would not work. ApplyReturnAddressAdjustment applies to all frames in the iterator, ExCollide applies only for a single frame and it's reset on each iteration.

[MichalStrehovsky]

/azp run runtime-nativeaot-outerloop

[azure-pipelines]

Azure Pipelines successfully started running 1 pipeline(s).

[filipnavara]

Hmm, the test failures do look suspicious. I'll check it in detail when all the pipelines finish running.

Cursory look at the source code suggests that we may need to remove this:

runtime/src/coreclr/nativeaot/Runtime/StackFrameIterator.cpp

Lines 814 to 817 in 3e572e9

	// In case m_ControlPC is pre-adjusted, counteract here, since the caller of this routine
	// will apply the adjustment again once we return. If the m_ControlPC is not pre-adjusted,
	// this is simply an no-op.
	m_ControlPC = m_OriginalControlPC;

[tannergooding]

/azp run runtime-nativeaot-outerloop

[azure-pipelines]

Azure Pipelines successfully started running 1 pipeline(s).

[filipnavara]

There are 6 test failures:

• 3x JIT\Regression\JitBlue\Runtime_105821\Runtime_105821\Runtime_105821 which was meanwhile fixed by Fix Runtime_105821 ISA check #105889
• AV in System.Formats.Asn1.Tests on linux-arm
• SIGKILL in System.IO.Tests on linux-musl-x64
• Timeout in System.Threading.Channels.Tests on linux-arm64

The AV in System.Formats.Asn1.Tests has interesting stack trace:

Running assembly:System.Formats.Asn1.Tests, Version=9.0.0.0, Culture=neutral, PublicKeyToken=cc7b13ffcd2ddd51
Process terminated. Access Violation: Attempted to read or write protected memory. This is often an indication that other memory is corrupt. The application will be terminated since this platform does not support throwing an AccessViolationException.
   at System.RuntimeExceptionHelpers.FailFast(String, Exception, String, RhFailFastReason, IntPtr, IntPtr) + 0x1fd
   at System.RuntimeExceptionHelpers.GetRuntimeException(ExceptionIDs) + 0x167
   at System.Runtime.EH.GetClasslibException(ExceptionIDs, IntPtr) + 0x3d
   at System.Runtime.EH.RhThrowHwEx(UInt32, EH.ExInfo&) + 0x69
   at Test.Cryptography.ByteUtils.ByteArrayToHex(ReadOnlySpan`1 bytes) + 0x5d
   at System.Formats.Asn1.Tests.Writer.WriteNamedBitList.VerifyWriteNamedBitList_KeyUsage_OneByte(AsnEncodingRules ruleSet) + 0x10d
   at System.Formats.Asn1!<BaseAddress>+0x96a10a

There's nothing in ByteArrayToHex that would obviously cause an AV exception, so it looks like something was collected by GC that shouldn't have been. That could possibly be triggered by a change in the stack walking logic but there's no exception on the given code path so I don't see how the change in this PR could have effect in this specific instance.

[jkotas]

/azp run runtime-nativeaot-outerloop

[azure-pipelines]

Azure Pipelines successfully started running 1 pipeline(s).

[jkotas]

LGTM. @janvorli @VSadov Could you please review this one as well?

[JulieLeeMSFT]

LGTM. @janvorli @VSadov Could you please review this one as well?

Ping @janvorli @VSadov for code review.
cc @dotnet/jit-contrib

[janvorli]

LGTM modulo a nit regarding the comment.