[cDAC] Implement contract APIs for x86 stack walking

[max-charlamb]

#110758

Implements

• CodeInfo -> Get SavedMethodCode

• CodeInfo -> Get RelativeOffset

• CodeInfo -> Get FuncletStartAddress

  • This one is only partially implemented. On ARM, ARM64, LoongArch64, and RISCV64 the there is a special case to handle runtime function fragments. Currently this is a pain to test and not relevant for x86 stack walking so I'm hoping to make an issue and come back when we implement that. For more info on the runtime implementation see:

    runtime/src/coreclr/vm/codeman.cpp

    Lines 1080 to 1101 in 0e06c21

    	#if defined(EXCEPTION_DATA_SUPPORTS_FUNCTION_FRAGMENTS)
    	// Is the RUNTIME_FUNCTION a fragment? If so, we need to walk backwards until we find the first
    	// non-fragment RUNTIME_FUNCTION, and use that one. This happens when we have very large functions
    	// and multiple RUNTIME_FUNCTION entries per function or funclet. However, all but the first will
    	// have the "F" bit set in the unwind data, indicating a fragment (with phantom prolog unwind codes).
    	for (;;)
    	{
    	if (!IsFunctionFragment(baseAddress, pFunctionEntry))
    	{
    	// This is not a fragment; we're done
    	break;
    	}
    	// We found a fragment. Walk backwards in the RUNTIME_FUNCTION array until we find a non-fragment.
    	// We're guaranteed to find one, because we require that a fragment live in a function or funclet
    	// that has a prolog, which will have non-fragment .xdata.
    	--pFunctionEntry;
    	funcletStartAddress = baseAddress + RUNTIME_FUNCTION__BeginAddress(pFunctionEntry);
    	}
    	#endif // EXCEPTION_DATA_SUPPORTS_FUNCTION_FRAGMENTS

• CodeInfo -> Get IsFunclet

[Copilot]

Pull Request Overview

This PR implements new contract APIs for x86 stack walking focused on updating how unwind information and funclet addresses are retrieved. Key changes include:

• Removing the need for a controlPC parameter in GetUnwindInfo calls.
• Introducing a new GetFuncletStartAddress API in the ExecutionManager implementations.
• Updating the IExecutionManager interface and adding an extension IsFunclet for funclet detection.

Reviewed Changes

Copilot reviewed 6 out of 6 changed files in this pull request and generated no comments.

Show a summary per file

File	Description
Unwinder.cs	Removed the controlPC parameter from the GetUnwindInfo call to align with new API design.
ExecutionManager_2.cs	Updated API calls to reflect removal of the ip parameter and added GetFuncletStartAddress.
ExecutionManager_1.cs	Similar update as in ExecutionManager_2.cs for consistency.
ExecutionManagerCore.cs	Implemented GetFuncletStartAddress and modified GetUnwindInfo calls to use CodeBlockHandle.Address.
IExecutionManager.cs	Revised interface to remove ip parameter and include GetFuncletStartAddress.
IExecutionManagerExtensions.cs	Added a helper extension method for funclet detection.

Comments suppressed due to low confidence (2)

src/native/managed/cdac/Microsoft.Diagnostics.DataContractReader.Contracts/Contracts/StackWalk/Context/Unwinder.cs:151

• Ensure that removing the controlPC parameter from the GetUnwindInfo call is intentional and that the new API behavior fully supports x86 stack walking requirements.

TargetPointer unwindInfo = eman.GetUnwindInfo(cbh);

src/native/managed/cdac/Microsoft.Diagnostics.DataContractReader.Contracts/Contracts/ExecutionManager/ExecutionManagerCore.cs:188

• Verify that replacing the controlPC parameter with codeInfoHandle.Address.Value in the RangeSection.Find call maintains correct unwinding behavior for x86 stack walking.

RangeSection range = RangeSection.Find(_target, _topRangeSectionMap, _rangeSectionMapLookup, codeInfoHandle.Address.Value);

[dotnet-policy-service]

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

[davmason]

Looks good!

[am11]

Unused? Plan is to make funclet available on all platforms once testing is completed #113985 (comment). cc @filipnavara

[filipnavara]

The implementation looks reasonable even for FEATURE_EH_FUNCLETS builds.

[max-charlamb]

These contract APIs are not used yet, but will be used as part of the x86 stack walking. That will probably be in a follow-up PR.

[max-charlamb]

Closing in favor of #116072