[cDAC] Stack walk GC stress verification and fixes

[max-charlamb]

Summary

cDAC GC stress verification tool (DOTNET_CdacStress) that compares stack GC references between the cDAC and the runtime at allocation stress points. Includes stack walker fixes, GC reference scanning implementation, custom signature decoding, and calling convention argument iteration.

Note

This PR description was updated with AI assistance from Copilot.

Stack walker fixes

• Fix SW_SKIPPED_FRAME: do not call UpdateContextFromFrame (matches native SFITER_SKIPPED_FRAME_FUNCTION which does not call UpdateRegDisplay)
• Fix skipped-frame loop: check for more skipped frames before yielding managed method (prevents duplicate EnumGcRefs between consecutive skipped frames)
• Restructure Filter() to drive Next() directly, matching native Filter()+NextRaw() integration (prevents funclet-to-parent walk cycles)
• Remove SkipActiveICFOnce/SkipCurrentFrameInCheck — active ICF double-yield is natural and harmless
• Remove IsAtFirstPassExceptionThrowSite — native does not suppress first-pass refs
• Fix IsFirst not preserved for skipped frames (was causing IsActiveFrame=false for the topmost managed frame)

GC reference scanning

• Implement PromoteCallerStack for stub frames (GCRefMap + MetaSig paths)
• Implement SOSDacImpl.GetStackReferences using cDAC contract (was falling back to legacy DAC)
• Read FilterContext for stack walk starting context
• Three-way cDAC/DAC/RT comparison with InProcessDataTarget
• DOTNET_CdacStress bit flags: ALLOC/INSTR/REFS/WALK/USE_DAC/UNIQUE

RuntimeSignatureDecoder

Custom signature decoder that handles ELEMENT_TYPE_INTERNAL (0x21) and ELEMENT_TYPE_CMOD_INTERNAL (0x22) which SRM's SignatureDecoder cannot parse. These occur in IL stubs, marshalling stubs, and unsafe accessor frames.

• IRuntimeSignatureTypeProvider<TType, TGenericContext>: superset of SRM's ISignatureTypeProvider, adds GetInternalType and GetInternalModifiedType
• ISignatureReader + SpanSignatureReader: endianness-aware reader abstraction
• Correct ECMA-335 compliance: TypeDefOrRefOrSpecEncoded token decoding, sign-extension-by-width for compressed signed ints, bounds validation

ArgIterator (ported from crossgen2)

Proper calling convention analysis replacing the simplified 1-slot-per-param approach.

• CallingConventionInfo: hybrid data descriptor layout values + ABI invariant constants for all architectures (x86, x64 Windows/Unix, ARM32, ARM64, LoongArch64, RISC-V64)
• ArgIterator.GetNextOffset(): maps each argument to its actual register or stack offset
• OffsetOfFloatArgumentRegisters added to TransitionBlock data descriptor
• Handles multi-slot args, forced-byref params, return buffer placement, async continuation

CI infrastructure

• cDAC stress tests run in Helix via cdac-stress-helix.proj
• Extended runtime-diagnostics.yml CdacDumpTests buildArgs with +tools.cdacstresstests

Test results

Allocation-level stress (9 debuggees, ~46K verifications):

Debuggee	Verifications	Pass	Fail
BasicAlloc	4,936	4,936	0
DeepStack	4,964	4,964	0
Generics	4,936	4,936	0
MultiThread	4,996	4,996	0
Comprehensive	4,994	4,992	2*
ExceptionHandling	4,958	4,958	0
StructScenarios	4,940	4,940	0
DynamicMethods	6,520	6,520	0
PInvoke	4,936	4,936	0

Instruction-level stress (9 debuggees, ~226K verifications):
All 9 debuggees pass with zero failures across 226,452 verifications.

*Comprehensive's 2 failures are pre-existing legacy DAC issues (DAC returns 0 refs for background threads in kernel waits; cDAC and runtime agree).

[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

Adds a cDAC GC-stress verification harness and extends the cDAC stack-walk / stack-GC-ref pipeline so cDAC stack reference enumeration can be compared against runtime scanning at stress points.

Changes:

• Implements/extends cDAC stack reference enumeration (including Frame-based scanning paths like PromoteCallerStack via GCRefMap / MetaSig) and wires SOS GetStackReferences to the cDAC contract.
• Introduces a new GC stress integration test project with debuggee apps and orchestration targets.
• Extends CoreCLR cDAC stress/GC stress plumbing and data descriptors to support the new stack-walk and frame-scanning capabilities.

Reviewed changes

Copilot reviewed 46 out of 46 changed files in this pull request and generated 12 comments.

Show a summary per file

File	Description
src/native/managed/cdac/tests/MockDescriptors/MockDescriptors.ExecutionManager.cs	Updates mock type layout for execution manager-related data.
src/native/managed/cdac/tests/MockDescriptors/MockDescriptors.cs	Updates mock type layouts (ExceptionInfo/Thread) for contract tests.
src/native/managed/cdac/tests/Microsoft.Diagnostics.DataContractReader.Tests.csproj	Excludes new GCStressTests folder from the existing unit test project compilation.
src/native/managed/cdac/tests/GCStressTests/README.md	Documents how to build/run the new GC stress tests.
src/native/managed/cdac/tests/GCStressTests/Microsoft.Diagnostics.DataContractReader.GCStressTests.csproj	Adds a dedicated GC stress test project.
src/native/managed/cdac/tests/GCStressTests/GCStressTests.targets	MSBuild orchestration to discover/build debuggee projects.
src/native/managed/cdac/tests/GCStressTests/GCStressTestBase.cs	Test harness to run debuggees under corerun and parse verification logs.
src/native/managed/cdac/tests/GCStressTests/GCStressResults.cs	Parses the native verification log into structured pass/fail/skip counts.
src/native/managed/cdac/tests/GCStressTests/Debuggees/PInvoke/Program.cs	Adds a P/Invoke-focused debuggee workload.
src/native/managed/cdac/tests/GCStressTests/Debuggees/PInvoke/PInvoke.csproj	Debuggee project file for PInvoke scenario.
src/native/managed/cdac/tests/GCStressTests/Debuggees/MultiThread/Program.cs	Adds a multi-threaded debuggee workload.
src/native/managed/cdac/tests/GCStressTests/Debuggees/MultiThread/MultiThread.csproj	Debuggee project file for MultiThread scenario.
src/native/managed/cdac/tests/GCStressTests/Debuggees/Generics/Program.cs	Adds a generics/interface/delegate debuggee workload.
src/native/managed/cdac/tests/GCStressTests/Debuggees/Generics/Generics.csproj	Debuggee project file for Generics scenario.
src/native/managed/cdac/tests/GCStressTests/Debuggees/ExceptionHandling/Program.cs	Adds an exception-handling/funclet debuggee workload.
src/native/managed/cdac/tests/GCStressTests/Debuggees/ExceptionHandling/ExceptionHandling.csproj	Debuggee project file for ExceptionHandling scenario.
src/native/managed/cdac/tests/GCStressTests/Debuggees/Directory.Build.props	Shared build props for debuggee projects (output layout, TFM, etc.).
src/native/managed/cdac/tests/GCStressTests/Debuggees/DeepStack/Program.cs	Adds deep-recursion debuggee workload.
src/native/managed/cdac/tests/GCStressTests/Debuggees/DeepStack/DeepStack.csproj	Debuggee project file for DeepStack scenario.
src/native/managed/cdac/tests/GCStressTests/Debuggees/Comprehensive/Program.cs	Adds comprehensive “all scenarios” debuggee workload.
src/native/managed/cdac/tests/GCStressTests/Debuggees/Comprehensive/Comprehensive.csproj	Debuggee project file for Comprehensive scenario.
src/native/managed/cdac/tests/GCStressTests/Debuggees/BasicAlloc/Program.cs	Adds basic allocation/live-ref debuggee workload.
src/native/managed/cdac/tests/GCStressTests/Debuggees/BasicAlloc/BasicAlloc.csproj	Debuggee project file for BasicAlloc scenario.
src/native/managed/cdac/tests/GCStressTests/BasicGCStressTests.cs	Theory-based test suite that runs the debuggees and asserts pass rate.
src/native/managed/cdac/tests/gcstress/known-issues.md	Captures known mismatch classes and limitations.
src/native/managed/cdac/Microsoft.Diagnostics.DataContractReader.Legacy/SOSDacImpl.cs	Implements GetStackReferences using the cDAC contract rather than legacy DAC fallback.
src/native/managed/cdac/Microsoft.Diagnostics.DataContractReader.Contracts/Data/Frames/StubDispatchFrame.cs	Extends StubDispatchFrame data with GCRefMap pointer.
src/native/managed/cdac/Microsoft.Diagnostics.DataContractReader.Contracts/Data/Frames/ExternalMethodFrame.cs	Adds ExternalMethodFrame contract data type (GCRefMap).
src/native/managed/cdac/Microsoft.Diagnostics.DataContractReader.Contracts/Data/Frames/DynamicHelperFrame.cs	Adds DynamicHelperFrame contract data type (flags).
src/native/managed/cdac/Microsoft.Diagnostics.DataContractReader.Contracts/Data/ExceptionInfo.cs	Adds clause-range fields used for catch-handler resumption offset selection.
src/native/managed/cdac/Microsoft.Diagnostics.DataContractReader.Contracts/Contracts/StackWalk/StackWalk_1.cs	Refactors stack-walk filtering and adds Frame-based GC root scanning.
src/native/managed/cdac/Microsoft.Diagnostics.DataContractReader.Contracts/Contracts/StackWalk/GC/GcScanner.cs	Adds optional relOffset override support for GC ref enumeration.
src/native/managed/cdac/Microsoft.Diagnostics.DataContractReader.Contracts/Contracts/StackWalk/GC/GCRefMapDecoder.cs	Implements GCRefMap decoding for transition-block scanning.
src/native/managed/cdac/Microsoft.Diagnostics.DataContractReader.Contracts/Contracts/StackWalk/GC/CorSigParser.cs	Adds minimal signature parsing to classify parameters for MetaSig-based scanning.
src/native/managed/cdac/Microsoft.Diagnostics.DataContractReader.Contracts/Contracts/GCInfo/IGCInfoDecoder.cs	Adds FindFirstInterruptiblePoint API to GCInfo decoders.
src/native/managed/cdac/Microsoft.Diagnostics.DataContractReader.Contracts/Contracts/GCInfo/GCInfoDecoder.cs	Implements FindFirstInterruptiblePoint using decoded interruptible ranges.
src/native/managed/cdac/Microsoft.Diagnostics.DataContractReader.Contracts/Contracts/ExecutionManager/ExecutionManagerCore.EEJitManager.cs	Fixes code-start lookup for exception clause enumeration and adds minor flow adjustments.
src/native/managed/cdac/Microsoft.Diagnostics.DataContractReader.Contracts/Constants.cs	Adds TransitionBlock-related global names used by frame scanning.
src/native/managed/cdac/Microsoft.Diagnostics.DataContractReader.Abstractions/DataType.cs	Adds new DataType enum values for new frame contracts.
src/native/managed/cdac/cdac.slnx	Adds the new GC stress test project to the cDAC solution.
src/coreclr/vm/gccover.cpp	Adds step-based skipping to reduce overhead when throttling verification.
src/coreclr/vm/frames.h	Exposes additional frame fields to cDAC via cdac_data<> descriptors.
src/coreclr/vm/datadescriptor/datadescriptor.inc	Adds new contract fields/globals for frames and TransitionBlock layout.
src/coreclr/vm/cdacstress.cpp	Updates in-process cDAC/DAC verification logic, logging, and step behavior.
eng/Subsets.props	Adds an on-demand subset for running GC stress tests.
docs/design/datacontracts/StackWalk.md	Documents the new frame fields and TransitionBlock globals in the StackWalk contract.

Comments suppressed due to low confidence (1)

src/native/managed/cdac/tests/MockDescriptors/MockDescriptors.cs:228

• ThreadFields contains ProfilerFilterContext twice (lines 224 and 228). Duplicate field entries will skew offsets and make the mock descriptors inconsistent with the real data descriptor. Keep a single ProfilerFilterContext entry in the correct order.

[Copilot]

Pull request overview

Copilot reviewed 50 out of 50 changed files in this pull request and generated 7 comments.

Comments suppressed due to low confidence (2)

src/native/managed/cdac/tests/MockDescriptors/MockDescriptors.cs:229

• ThreadFields now includes DebuggerFilterContext/ProfilerFilterContext twice (duplicate entries at the end of the list). This can cause ambiguous/incorrect field offsets in the mock type layout. Keep each field only once.
  src/coreclr/vm/cdacstress.cpp:963
• CompareRefSets uses a fixed-size matched[MAX_COLLECTED_REFS] buffer but no longer validates that countA/countB are <= MAX_COLLECTED_REFS. Since CollectStackRefs appends without a hard cap, this can lead to out-of-bounds access when countA or countB exceeds 4096. Reintroduce a guard or allocate the match state sized to the counts.

 return true;
bool matched[MAX_COLLECTED_REFS] = {};
for (int i = 0; i < countA; i++)

[Copilot]

Pull request overview

Copilot reviewed 53 out of 53 changed files in this pull request and generated 7 comments.

[max-charlamb]

feedback for local copilot

[Copilot]

Pull request overview

Copilot reviewed 51 out of 51 changed files in this pull request and generated 9 comments.

Comments suppressed due to low confidence (1)

src/coreclr/vm/cdacstress.cpp:625

• CollectStackRefs appends to pRefs without any cap. Later comparisons allocate fixed-size arrays sized MAX_COLLECTED_REFS (e.g., matched[MAX_COLLECTED_REFS], aUsed/bUsed) and assume counts fit. If the DAC returns > MAX_COLLECTED_REFS refs, this will lead to out-of-bounds writes/reads. Add an explicit limit/overflow handling in CollectStackRefs (stop at MAX_COLLECTED_REFS and mark overflow / SKIP), or change the later comparison logic to handle arbitrary counts safely.

 SOSStackRefData refData;
unsigned int fetched = 0;
while (true)
{
hr = pEnum->Next(1, &refData, &fetched);
if (FAILED(hr) || fetched == 0)
break;
StackRef ref;
ref.Address = refData.Address;
ref.Object = refData.Object;
ref.Flags = refData.Flags;
ref.Source = refData.Source;
ref.SourceType = refData.SourceType;
ref.Register = refData.Register;
ref.Offset = refData.Offset;
ref.StackPointer = refData.StackPointer;
pRefs->Append(ref);
}

[Copilot]

Pull request overview

Copilot reviewed 52 out of 52 changed files in this pull request and generated 3 comments.

[Copilot]

Pull request overview

Copilot reviewed 65 out of 65 changed files in this pull request and generated 5 comments.

[Copilot]

ThreadStore now calls target.GetTypeInfo(DataType.SLink) and reads field "Next", but the CoreCLR data descriptor in this PR doesn't define a SLink type/field (could not find any CDAC_TYPE_BEGIN/FIELD for SLink). This will cause GetTypeInfo(DataType.SLink) to fail at runtime. Either add an SLink descriptor (with a "Next" pointer) on the runtime side or avoid needing type info here (e.g., treat the embedded SLink as a single pointer at offset 0).

[Copilot]

DataType adds SLink, but there is no corresponding runtime data descriptor definition (no CDAC_TYPE_BEGIN/FIELD(SLink, ...) found). Any attempt to read this type via Target.GetTypeInfo(DataType.SLink) will fail. Either add the runtime descriptor entry for SLink or remove this enum value and read the embedded link without type metadata.

Suggested change

	SLink,
	ThreadLocalData,
	ThreadLocalData=17,

[Copilot]

These DataReceived event handlers append to strings via "+=" from background threads. That is not thread-safe and can lead to lost/garbled output under contention. Consider using a StringBuilder with locking (or ConcurrentQueue) for stderr/stdout aggregation.

[Copilot]

Same thread-safety issue as stderr: appending to stdout with += from OutputDataReceived is racy. Use a synchronized StringBuilder/collector to avoid missing output and to keep logs reliable when tests fail.

[Copilot]

GcTypeKindToArgTypeInfo collapses all non-GC-ref types to CorElementType.I/pointer-sized and doesn't preserve float vs integer vs exact primitive sizes. ArgIterator offset calculation depends on the real signature shape (especially on Unix x64/ARM64 where float regs and argument sizing affect later argument placement), so this can produce incorrect offsets and cause missed/incorrect GC ref reporting for signatures with floats or non-pointer-sized primitives. Consider decoding into a richer type representation (e.g., CorElementType + size) and building ArgTypeInfo from that rather than from GcTypeKind.

Suggested change

	/// Converts a <see cref="GcTypeKind"/> to a minimal <see cref="ArgTypeInfo"/> for
	/// ArgIterator consumption. This is a bridge until we have a full type-aware provider.
	/// </summary>
	privatestaticArgTypeInfoGcTypeKindToArgTypeInfo(GcTypeKindkind,intpointerSize)
	{
	returnkindswitch
	{
	GcTypeKind.None=> ArgTypeInfo.ForPrimitive(CorElementType.I,pointerSize),
	GcTypeKind.Ref=> ArgTypeInfo.ForPrimitive(CorElementType.Class,pointerSize),
	GcTypeKind.Interior=> ArgTypeInfo.ForPrimitive(CorElementType.Byref,pointerSize),
	GcTypeKind.Other=>new ArgTypeInfo
	{
	CorElementType=CorElementType.ValueType,
	Size=pointerSize,// Conservative: assume pointer-sized for now
	IsValueType=true,
	},
	_ =>ArgTypeInfo.ForPrimitive(CorElementType.I,pointerSize),
	};
	}
	/// Converts a <see cref="GcTypeKind"/> to a conservative fallback <see cref="ArgTypeInfo"/>
	/// when an exact signature cannot be decoded. Callers should prefer decoding the real
	/// managed signature via <see cref="TryGetArgTypeInfosFromMethodSignature(ReadOnlySpan{byte}, int, out ArgTypeInfo[])"/>
	/// so that floating-point and non-pointer-sized primitives retain their real layout.
	/// </summary>
	privatestaticArgTypeInfoGcTypeKindToArgTypeInfo(GcTypeKindkind,intpointerSize)
	{
	returnkindswitch
	{
	GcTypeKind.Ref=> ArgTypeInfo.ForPrimitive(CorElementType.Class,pointerSize),
	GcTypeKind.Interior=> ArgTypeInfo.ForPrimitive(CorElementType.Byref,pointerSize),
	GcTypeKind.Other=>new ArgTypeInfo
	{
	CorElementType=CorElementType.ValueType,
	Size=pointerSize,
	IsValueType=true,
	},
	_ =>ArgTypeInfo.ForPrimitive(CorElementType.I,pointerSize),
	};
	}
	privatestaticboolTryGetArgTypeInfosFromMethodSignature(ReadOnlySpan<byte>signatureBytes,intpointerSize,outArgTypeInfo[]argTypes)
	{
	argTypes=Array.Empty<ArgTypeInfo>();
	if(signatureBytes.IsEmpty)
	returnfalse;
	BlobReaderreader=new(signatureBytes);
	SignatureHeaderheader=reader.ReadSignatureHeader();
	if(header.IsGeneric)
	reader.ReadCompressedInteger();
	intparameterCount=reader.ReadCompressedInteger();
	if(parameterCount<0)
	returnfalse;
	if(!TrySkipSignatureType(refreader))
	returnfalse;
	ArgTypeInfo[]decodedArgTypes=newArgTypeInfo[parameterCount];
	for(inti=0;i<parameterCount;i++)
	{
	if(!TryReadArgTypeInfo(refreader,pointerSize,outdecodedArgTypes[i]))
	returnfalse;
	}
	argTypes=decodedArgTypes;
	returntrue;
	}
	privatestaticboolTryReadArgTypeInfo(refBlobReaderreader,intpointerSize,outArgTypeInfoargTypeInfo)
	{
	while(reader.RemainingBytes>0)
	{
	CorElementTypeelementType=(CorElementType)reader.ReadByte();
	switch(elementType)
	{
	caseCorElementType.CmodOpt:
	caseCorElementType.CmodReqd:
	reader.ReadCompressedInteger();
	continue;
	caseCorElementType.Pinned:
	continue;
	caseCorElementType.Boolean:
	caseCorElementType.I1:
	caseCorElementType.U1:
	argTypeInfo=ArgTypeInfo.ForPrimitive(elementType,1);
	returntrue;
	caseCorElementType.Char:
	caseCorElementType.I2:
	caseCorElementType.U2:
	argTypeInfo=ArgTypeInfo.ForPrimitive(elementType,2);
	returntrue;
	caseCorElementType.I4:
	caseCorElementType.U4:
	caseCorElementType.R4:
	argTypeInfo=ArgTypeInfo.ForPrimitive(elementType,4);
	returntrue;
	caseCorElementType.I8:
	caseCorElementType.U8:
	caseCorElementType.R8:
	argTypeInfo=ArgTypeInfo.ForPrimitive(elementType,8);
	returntrue;
	caseCorElementType.I:
	caseCorElementType.U:
	caseCorElementType.Ptr:
	caseCorElementType.FnPtr:
	if(elementTypeisCorElementType.Ptr or CorElementType.FnPtr)
	{
	if(!TrySkipSignatureType(refreader))
	{
	argTypeInfo=default;
	returnfalse;
	}
	}
	argTypeInfo=ArgTypeInfo.ForPrimitive(CorElementType.I,pointerSize);
	returntrue;
	caseCorElementType.Byref:
	if(!TrySkipSignatureType(refreader))
	{
	argTypeInfo=default;
	returnfalse;
	}
	argTypeInfo=ArgTypeInfo.ForPrimitive(CorElementType.Byref,pointerSize);
	returntrue;
	caseCorElementType.String:
	caseCorElementType.Object:
	caseCorElementType.Class:
	if(elementTypeisCorElementType.Class)
	reader.ReadCompressedInteger();
	argTypeInfo=ArgTypeInfo.ForPrimitive(CorElementType.Class,pointerSize);
	returntrue;
	caseCorElementType.SzArray:
	if(!TrySkipSignatureType(refreader))
	{
	argTypeInfo=default;
	returnfalse;
	}
	argTypeInfo=ArgTypeInfo.ForPrimitive(CorElementType.Class,pointerSize);
	returntrue;
	caseCorElementType.ValueType:
	reader.ReadCompressedInteger();
	argTypeInfo=newArgTypeInfo
	{
	CorElementType=CorElementType.ValueType,
	Size=pointerSize,
	IsValueType=true,
	};
	returntrue;
	caseCorElementType.GenericInst:
	{
	if(!TrySkipSignatureType(refreader))
	{
	argTypeInfo=default;
	returnfalse;
	}
	intgenericArgCount=reader.ReadCompressedInteger();
	for(inti=0;i<genericArgCount;i++)
	{
	if(!TrySkipSignatureType(refreader))
	{
	argTypeInfo=default;
	returnfalse;
	}
	}
	argTypeInfo=newArgTypeInfo
	{
	CorElementType=CorElementType.ValueType,
	Size=pointerSize,
	IsValueType=true,
	};
	returntrue;
	}
	default:
	argTypeInfo=default;
	returnfalse;
	}
	}
	argTypeInfo=default;
	returnfalse;
	}
	privatestaticboolTrySkipSignatureType(refBlobReaderreader)
	{
	while(reader.RemainingBytes>0)
	{
	CorElementTypeelementType=(CorElementType)reader.ReadByte();
	switch(elementType)
	{
	caseCorElementType.CmodOpt:
	caseCorElementType.CmodReqd:
	reader.ReadCompressedInteger();
	continue;
	caseCorElementType.Pinned:
	continue;
	caseCorElementType.Void:
	caseCorElementType.Boolean:
	caseCorElementType.Char:
	caseCorElementType.I1:
	caseCorElementType.U1:
	caseCorElementType.I2:
	caseCorElementType.U2:
	caseCorElementType.I4:
	caseCorElementType.U4:
	caseCorElementType.I8:
	caseCorElementType.U8:
	caseCorElementType.R4:
	caseCorElementType.R8:
	caseCorElementType.String:
	caseCorElementType.Object:
	caseCorElementType.TypedByref:
	caseCorElementType.I:
	caseCorElementType.U:
	returntrue;
	caseCorElementType.Class:
	caseCorElementType.ValueType:
	caseCorElementType.Var:
	caseCorElementType.MVar:
	reader.ReadCompressedInteger();
	returntrue;
	caseCorElementType.Byref:
	caseCorElementType.Ptr:
	caseCorElementType.SzArray:
	returnTrySkipSignatureType(refreader);
	caseCorElementType.GenericInst:
	{
	if(!TrySkipSignatureType(refreader))
	returnfalse;
	intgenericArgCount=reader.ReadCompressedInteger();
	for(inti=0;i<genericArgCount;i++)
	{
	if(!TrySkipSignatureType(refreader))
	returnfalse;
	}
	returntrue;
	}
	caseCorElementType.Array:
	{
	if(!TrySkipSignatureType(refreader))
	returnfalse;
	intrank=reader.ReadCompressedInteger();
	intsizes=reader.ReadCompressedInteger();
	for(inti=0;i<sizes;i++)
	reader.ReadCompressedInteger();
	intlowerBounds=reader.ReadCompressedInteger();
	for(inti=0;i<lowerBounds;i++)
	reader.ReadCompressedInteger();
	returnrank>=0;
	}
	caseCorElementType.FnPtr:
	{
	SignatureHeaderheader=reader.ReadSignatureHeader();
	if(header.IsGeneric)
	reader.ReadCompressedInteger();
	intparameterCount=reader.ReadCompressedInteger();
	if(parameterCount<0)
	returnfalse;
	if(!TrySkipSignatureType(refreader))
	returnfalse;
	for(inti=0;i<parameterCount;i++)
	{
	if(!TrySkipSignatureType(refreader))
	returnfalse;
	}
	returntrue;
	}
	default:
	returnfalse;
	}
	}
	returnfalse;
	}

[Copilot]

Pull request overview

Copilot reviewed 64 out of 64 changed files in this pull request and generated 3 comments.

[Copilot]

CollectStackRefs currently appends every ref returned by ISOSStackRefEnum::Next with no upper bound. Later comparison helpers allocate fixed-size arrays sized by MAX_COLLECTED_REFS (e.g., matched[MAX_COLLECTED_REFS], aUsed[MAX_COLLECTED_REFS]), so if enumeration returns more than MAX_COLLECTED_REFS it can lead to out-of-bounds writes. Please cap collection to MAX_COLLECTED_REFS (and record an overflow/skip reason) or make the comparison logic handle arbitrarily large ref sets safely.

[Copilot]

When CollectRuntimeStackRefs overflows, the code logs a [SKIP] line but continues and still computes rtMatch/pass (and does not increment s_verifySkip). If CDACSTRESS_USE_DAC is not set, this can produce false failures based on a truncated runtime ref set. Consider treating runtime overflow as an actual skip (increment skip counter + return) when RT comparison is used for pass/fail, or otherwise avoid using rtMatch in the presence of overflow.

Suggested change

	if (rtOverflow && s_logFile != nullptr)
	{
	fprintf(s_logFile, "[SKIP] Thread=0x%x IP=0x%p - RT overflow (>%d refs)\n",
	osThreadId, (void*)GetIP(regs), MAX_COLLECTED_REFS);
	if (rtOverflow)
	{
	InterlockedIncrement(&s_verifySkip);
	if (s_logFile != nullptr)
	{
	fprintf(s_logFile, "[SKIP] Thread=0x%x IP=0x%p - RT overflow (>%d refs)\n",
	osThreadId, (void*)GetIP(regs), MAX_COLLECTED_REFS);
	}
	return;

[Copilot]

In ReportUntrackedAndSucceed, ReportSlot is called with reportScratchSlots: true, which forces reporting scratch registers/stack slots even when CodeManagerFlags.ActiveStackFrame is not set (reportScratchSlots local is false). This changes GC root enumeration behavior and can introduce extra roots for non-leaf frames. It looks like this should pass the reportScratchSlots variable instead of always true.

Suggested change

	ReportSlot(slotIndex,reportScratchSlots:true,reportFpBasedSlotsOnly,reportSlot);
	}
	ReportSlot(slotIndex,reportScratchSlots,reportFpBasedSlotsOnly,reportSlot);
	}

[Copilot]

Pull request overview

Copilot reviewed 64 out of 64 changed files in this pull request and generated 2 comments.

[Copilot]

CompareRefSetsFlat uses a fixed-size matched[MAX_COLLECTED_REFS] buffer, but cDAC/DAC ref collection (CollectStackRefs) is unbounded. Without a guard/cap, a large ref set (>4096) can cause out-of-bounds writes during matching. Consider capping cDAC/DAC collection to MAX_COLLECTED_REFS (and treating overflow as a [SKIP]) or using dynamically sized bookkeeping here.

[Copilot]

The async stdout/stderr collection uses stdout += ... / stderr += ... inside DataReceived event handlers. These callbacks can run concurrently, and string concatenation is not thread-safe; it can also be costly for large output. Consider buffering with a thread-safe collector (e.g., ConcurrentQueue<string> or StringBuilder with a lock) and call process.WaitForExit() (or await stream completion) after WaitForExit(timeout) to ensure all async output has been drained before asserting/logging.

[max-charlamb]

Closing in favor of stacked PR approach

[max-charlamb]

Summary

cDAC GC stress verification tool (DOTNET_CdacStress) that compares stack GC references between the cDAC and the runtime at allocation stress points. Includes stack walker fixes, GC reference scanning implementation, custom signature decoding, and calling convention argument iteration.

Note

This PR description was updated with AI assistance from Copilot.

Stack walker fixes

• Fix SW_SKIPPED_FRAME: do not call UpdateContextFromFrame (matches native SFITER_SKIPPED_FRAME_FUNCTION which does not call UpdateRegDisplay)
• Fix skipped-frame loop: check for more skipped frames before yielding managed method (prevents duplicate EnumGcRefs between consecutive skipped frames)
• Restructure Filter() to drive Next() directly, matching native Filter()+NextRaw() integration (prevents funclet-to-parent walk cycles)
• Remove SkipActiveICFOnce/SkipCurrentFrameInCheck — active ICF double-yield is natural and harmless
• Remove IsAtFirstPassExceptionThrowSite — native does not suppress first-pass refs
• Fix IsFirst not preserved for skipped frames (was causing IsActiveFrame=false for the topmost managed frame)

GC reference scanning

• Implement PromoteCallerStack for stub frames (GCRefMap + MetaSig paths)
• Implement SOSDacImpl.GetStackReferences using cDAC contract (was falling back to legacy DAC)
• Read FilterContext for stack walk starting context
• Three-way cDAC/DAC/RT comparison with InProcessDataTarget
• DOTNET_CdacStress bit flags: ALLOC/INSTR/REFS/WALK/USE_DAC/UNIQUE

RuntimeSignatureDecoder

Custom signature decoder that handles ELEMENT_TYPE_INTERNAL (0x21) and ELEMENT_TYPE_CMOD_INTERNAL (0x22) which SRM's SignatureDecoder cannot parse. These occur in IL stubs, marshalling stubs, and unsafe accessor frames.

• IRuntimeSignatureTypeProvider<TType, TGenericContext>: superset of SRM's ISignatureTypeProvider, adds GetInternalType and GetInternalModifiedType
• ISignatureReader + SpanSignatureReader: endianness-aware reader abstraction
• Correct ECMA-335 compliance: TypeDefOrRefOrSpecEncoded token decoding, sign-extension-by-width for compressed signed ints, bounds validation

ArgIterator (ported from crossgen2)

Proper calling convention analysis replacing the simplified 1-slot-per-param approach.

• CallingConventionInfo: hybrid data descriptor layout values + ABI invariant constants for all architectures (x86, x64 Windows/Unix, ARM32, ARM64, LoongArch64, RISC-V64)
• ArgIterator.GetNextOffset(): maps each argument to its actual register or stack offset
• OffsetOfFloatArgumentRegisters added to TransitionBlock data descriptor
• Handles multi-slot args, forced-byref params, return buffer placement, async continuation

CI infrastructure

• cDAC stress tests run in Helix via cdac-stress-helix.proj
• Extended runtime-diagnostics.yml CdacDumpTests buildArgs with +tools.cdacstresstests

Test results

Allocation-level stress (9 debuggees, ~46K verifications):

Debuggee	Verifications	Pass	Fail
BasicAlloc	4,936	4,936	0
DeepStack	4,964	4,964	0
Generics	4,936	4,936	0
MultiThread	4,996	4,996	0
Comprehensive	4,994	4,992	2*
ExceptionHandling	4,958	4,958	0
StructScenarios	4,940	4,940	0
DynamicMethods	6,520	6,520	0
PInvoke	4,936	4,936	0

Instruction-level stress (9 debuggees, ~226K verifications):
All 9 debuggees pass with zero failures across 226,452 verifications.

*Comprehensive's 2 failures are pre-existing legacy DAC issues (DAC returns 0 refs for background threads in kernel waits; cDAC and runtime agree).

[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

Adds a cDAC GC-stress verification harness and extends the cDAC stack-walk / stack-GC-ref pipeline so cDAC stack reference enumeration can be compared against runtime scanning at stress points.

Changes:

• Implements/extends cDAC stack reference enumeration (including Frame-based scanning paths like PromoteCallerStack via GCRefMap / MetaSig) and wires SOS GetStackReferences to the cDAC contract.
• Introduces a new GC stress integration test project with debuggee apps and orchestration targets.
• Extends CoreCLR cDAC stress/GC stress plumbing and data descriptors to support the new stack-walk and frame-scanning capabilities.

Reviewed changes

Copilot reviewed 46 out of 46 changed files in this pull request and generated 12 comments.

Show a summary per file

File	Description
src/native/managed/cdac/tests/MockDescriptors/MockDescriptors.ExecutionManager.cs	Updates mock type layout for execution manager-related data.
src/native/managed/cdac/tests/MockDescriptors/MockDescriptors.cs	Updates mock type layouts (ExceptionInfo/Thread) for contract tests.
src/native/managed/cdac/tests/Microsoft.Diagnostics.DataContractReader.Tests.csproj	Excludes new GCStressTests folder from the existing unit test project compilation.
src/native/managed/cdac/tests/GCStressTests/README.md	Documents how to build/run the new GC stress tests.
src/native/managed/cdac/tests/GCStressTests/Microsoft.Diagnostics.DataContractReader.GCStressTests.csproj	Adds a dedicated GC stress test project.
src/native/managed/cdac/tests/GCStressTests/GCStressTests.targets	MSBuild orchestration to discover/build debuggee projects.
src/native/managed/cdac/tests/GCStressTests/GCStressTestBase.cs	Test harness to run debuggees under corerun and parse verification logs.
src/native/managed/cdac/tests/GCStressTests/GCStressResults.cs	Parses the native verification log into structured pass/fail/skip counts.
src/native/managed/cdac/tests/GCStressTests/Debuggees/PInvoke/Program.cs	Adds a P/Invoke-focused debuggee workload.
src/native/managed/cdac/tests/GCStressTests/Debuggees/PInvoke/PInvoke.csproj	Debuggee project file for PInvoke scenario.
src/native/managed/cdac/tests/GCStressTests/Debuggees/MultiThread/Program.cs	Adds a multi-threaded debuggee workload.
src/native/managed/cdac/tests/GCStressTests/Debuggees/MultiThread/MultiThread.csproj	Debuggee project file for MultiThread scenario.
src/native/managed/cdac/tests/GCStressTests/Debuggees/Generics/Program.cs	Adds a generics/interface/delegate debuggee workload.
src/native/managed/cdac/tests/GCStressTests/Debuggees/Generics/Generics.csproj	Debuggee project file for Generics scenario.
src/native/managed/cdac/tests/GCStressTests/Debuggees/ExceptionHandling/Program.cs	Adds an exception-handling/funclet debuggee workload.
src/native/managed/cdac/tests/GCStressTests/Debuggees/ExceptionHandling/ExceptionHandling.csproj	Debuggee project file for ExceptionHandling scenario.
src/native/managed/cdac/tests/GCStressTests/Debuggees/Directory.Build.props	Shared build props for debuggee projects (output layout, TFM, etc.).
src/native/managed/cdac/tests/GCStressTests/Debuggees/DeepStack/Program.cs	Adds deep-recursion debuggee workload.
src/native/managed/cdac/tests/GCStressTests/Debuggees/DeepStack/DeepStack.csproj	Debuggee project file for DeepStack scenario.
src/native/managed/cdac/tests/GCStressTests/Debuggees/Comprehensive/Program.cs	Adds comprehensive “all scenarios” debuggee workload.
src/native/managed/cdac/tests/GCStressTests/Debuggees/Comprehensive/Comprehensive.csproj	Debuggee project file for Comprehensive scenario.
src/native/managed/cdac/tests/GCStressTests/Debuggees/BasicAlloc/Program.cs	Adds basic allocation/live-ref debuggee workload.
src/native/managed/cdac/tests/GCStressTests/Debuggees/BasicAlloc/BasicAlloc.csproj	Debuggee project file for BasicAlloc scenario.
src/native/managed/cdac/tests/GCStressTests/BasicGCStressTests.cs	Theory-based test suite that runs the debuggees and asserts pass rate.
src/native/managed/cdac/tests/gcstress/known-issues.md	Captures known mismatch classes and limitations.
src/native/managed/cdac/Microsoft.Diagnostics.DataContractReader.Legacy/SOSDacImpl.cs	Implements GetStackReferences using the cDAC contract rather than legacy DAC fallback.
src/native/managed/cdac/Microsoft.Diagnostics.DataContractReader.Contracts/Data/Frames/StubDispatchFrame.cs	Extends StubDispatchFrame data with GCRefMap pointer.
src/native/managed/cdac/Microsoft.Diagnostics.DataContractReader.Contracts/Data/Frames/ExternalMethodFrame.cs	Adds ExternalMethodFrame contract data type (GCRefMap).
src/native/managed/cdac/Microsoft.Diagnostics.DataContractReader.Contracts/Data/Frames/DynamicHelperFrame.cs	Adds DynamicHelperFrame contract data type (flags).
src/native/managed/cdac/Microsoft.Diagnostics.DataContractReader.Contracts/Data/ExceptionInfo.cs	Adds clause-range fields used for catch-handler resumption offset selection.
src/native/managed/cdac/Microsoft.Diagnostics.DataContractReader.Contracts/Contracts/StackWalk/StackWalk_1.cs	Refactors stack-walk filtering and adds Frame-based GC root scanning.
src/native/managed/cdac/Microsoft.Diagnostics.DataContractReader.Contracts/Contracts/StackWalk/GC/GcScanner.cs	Adds optional relOffset override support for GC ref enumeration.
src/native/managed/cdac/Microsoft.Diagnostics.DataContractReader.Contracts/Contracts/StackWalk/GC/GCRefMapDecoder.cs	Implements GCRefMap decoding for transition-block scanning.
src/native/managed/cdac/Microsoft.Diagnostics.DataContractReader.Contracts/Contracts/StackWalk/GC/CorSigParser.cs	Adds minimal signature parsing to classify parameters for MetaSig-based scanning.
src/native/managed/cdac/Microsoft.Diagnostics.DataContractReader.Contracts/Contracts/GCInfo/IGCInfoDecoder.cs	Adds FindFirstInterruptiblePoint API to GCInfo decoders.
src/native/managed/cdac/Microsoft.Diagnostics.DataContractReader.Contracts/Contracts/GCInfo/GCInfoDecoder.cs	Implements FindFirstInterruptiblePoint using decoded interruptible ranges.
src/native/managed/cdac/Microsoft.Diagnostics.DataContractReader.Contracts/Contracts/ExecutionManager/ExecutionManagerCore.EEJitManager.cs	Fixes code-start lookup for exception clause enumeration and adds minor flow adjustments.
src/native/managed/cdac/Microsoft.Diagnostics.DataContractReader.Contracts/Constants.cs	Adds TransitionBlock-related global names used by frame scanning.
src/native/managed/cdac/Microsoft.Diagnostics.DataContractReader.Abstractions/DataType.cs	Adds new DataType enum values for new frame contracts.
src/native/managed/cdac/cdac.slnx	Adds the new GC stress test project to the cDAC solution.
src/coreclr/vm/gccover.cpp	Adds step-based skipping to reduce overhead when throttling verification.
src/coreclr/vm/frames.h	Exposes additional frame fields to cDAC via cdac_data<> descriptors.
src/coreclr/vm/datadescriptor/datadescriptor.inc	Adds new contract fields/globals for frames and TransitionBlock layout.
src/coreclr/vm/cdacstress.cpp	Updates in-process cDAC/DAC verification logic, logging, and step behavior.
eng/Subsets.props	Adds an on-demand subset for running GC stress tests.
docs/design/datacontracts/StackWalk.md	Documents the new frame fields and TransitionBlock globals in the StackWalk contract.

Comments suppressed due to low confidence (1)

src/native/managed/cdac/tests/MockDescriptors/MockDescriptors.cs:228

• ThreadFields contains ProfilerFilterContext twice (lines 224 and 228). Duplicate field entries will skew offsets and make the mock descriptors inconsistent with the real data descriptor. Keep a single ProfilerFilterContext entry in the correct order.

[Copilot]

Pull request overview

Copilot reviewed 50 out of 50 changed files in this pull request and generated 7 comments.

Comments suppressed due to low confidence (2)

src/native/managed/cdac/tests/MockDescriptors/MockDescriptors.cs:229

• ThreadFields now includes DebuggerFilterContext/ProfilerFilterContext twice (duplicate entries at the end of the list). This can cause ambiguous/incorrect field offsets in the mock type layout. Keep each field only once.
  src/coreclr/vm/cdacstress.cpp:963
• CompareRefSets uses a fixed-size matched[MAX_COLLECTED_REFS] buffer but no longer validates that countA/countB are <= MAX_COLLECTED_REFS. Since CollectStackRefs appends without a hard cap, this can lead to out-of-bounds access when countA or countB exceeds 4096. Reintroduce a guard or allocate the match state sized to the counts.

 return true;
bool matched[MAX_COLLECTED_REFS] = {};
for (int i = 0; i < countA; i++)

[Copilot]

Pull request overview

Copilot reviewed 53 out of 53 changed files in this pull request and generated 7 comments.

[max-charlamb]

feedback for local copilot

[Copilot]

Pull request overview

Copilot reviewed 51 out of 51 changed files in this pull request and generated 9 comments.

Comments suppressed due to low confidence (1)

src/coreclr/vm/cdacstress.cpp:625

• CollectStackRefs appends to pRefs without any cap. Later comparisons allocate fixed-size arrays sized MAX_COLLECTED_REFS (e.g., matched[MAX_COLLECTED_REFS], aUsed/bUsed) and assume counts fit. If the DAC returns > MAX_COLLECTED_REFS refs, this will lead to out-of-bounds writes/reads. Add an explicit limit/overflow handling in CollectStackRefs (stop at MAX_COLLECTED_REFS and mark overflow / SKIP), or change the later comparison logic to handle arbitrary counts safely.

 SOSStackRefData refData;
unsigned int fetched = 0;
while (true)
{
hr = pEnum->Next(1, &refData, &fetched);
if (FAILED(hr) || fetched == 0)
break;
StackRef ref;
ref.Address = refData.Address;
ref.Object = refData.Object;
ref.Flags = refData.Flags;
ref.Source = refData.Source;
ref.SourceType = refData.SourceType;
ref.Register = refData.Register;
ref.Offset = refData.Offset;
ref.StackPointer = refData.StackPointer;
pRefs->Append(ref);
}

[Copilot]

Pull request overview

Copilot reviewed 52 out of 52 changed files in this pull request and generated 3 comments.

[Copilot]

Pull request overview

Copilot reviewed 65 out of 65 changed files in this pull request and generated 5 comments.

[Copilot]

ThreadStore now calls target.GetTypeInfo(DataType.SLink) and reads field "Next", but the CoreCLR data descriptor in this PR doesn't define a SLink type/field (could not find any CDAC_TYPE_BEGIN/FIELD for SLink). This will cause GetTypeInfo(DataType.SLink) to fail at runtime. Either add an SLink descriptor (with a "Next" pointer) on the runtime side or avoid needing type info here (e.g., treat the embedded SLink as a single pointer at offset 0).

[Copilot]

DataType adds SLink, but there is no corresponding runtime data descriptor definition (no CDAC_TYPE_BEGIN/FIELD(SLink, ...) found). Any attempt to read this type via Target.GetTypeInfo(DataType.SLink) will fail. Either add the runtime descriptor entry for SLink or remove this enum value and read the embedded link without type metadata.

Suggested change

	SLink,
	ThreadLocalData,
	ThreadLocalData=17,

[Copilot]

These DataReceived event handlers append to strings via "+=" from background threads. That is not thread-safe and can lead to lost/garbled output under contention. Consider using a StringBuilder with locking (or ConcurrentQueue) for stderr/stdout aggregation.

[Copilot]

Same thread-safety issue as stderr: appending to stdout with += from OutputDataReceived is racy. Use a synchronized StringBuilder/collector to avoid missing output and to keep logs reliable when tests fail.

[Copilot]

GcTypeKindToArgTypeInfo collapses all non-GC-ref types to CorElementType.I/pointer-sized and doesn't preserve float vs integer vs exact primitive sizes. ArgIterator offset calculation depends on the real signature shape (especially on Unix x64/ARM64 where float regs and argument sizing affect later argument placement), so this can produce incorrect offsets and cause missed/incorrect GC ref reporting for signatures with floats or non-pointer-sized primitives. Consider decoding into a richer type representation (e.g., CorElementType + size) and building ArgTypeInfo from that rather than from GcTypeKind.

Suggested change

	/// Converts a <see cref="GcTypeKind"/> to a minimal <see cref="ArgTypeInfo"/> for
	/// ArgIterator consumption. This is a bridge until we have a full type-aware provider.
	/// </summary>
	privatestaticArgTypeInfoGcTypeKindToArgTypeInfo(GcTypeKindkind,intpointerSize)
	{
	returnkindswitch
	{
	GcTypeKind.None=> ArgTypeInfo.ForPrimitive(CorElementType.I,pointerSize),
	GcTypeKind.Ref=> ArgTypeInfo.ForPrimitive(CorElementType.Class,pointerSize),
	GcTypeKind.Interior=> ArgTypeInfo.ForPrimitive(CorElementType.Byref,pointerSize),
	GcTypeKind.Other=>new ArgTypeInfo
	{
	CorElementType=CorElementType.ValueType,
	Size=pointerSize,// Conservative: assume pointer-sized for now
	IsValueType=true,
	},
	_ =>ArgTypeInfo.ForPrimitive(CorElementType.I,pointerSize),
	};
	}
	/// Converts a <see cref="GcTypeKind"/> to a conservative fallback <see cref="ArgTypeInfo"/>
	/// when an exact signature cannot be decoded. Callers should prefer decoding the real
	/// managed signature via <see cref="TryGetArgTypeInfosFromMethodSignature(ReadOnlySpan{byte}, int, out ArgTypeInfo[])"/>
	/// so that floating-point and non-pointer-sized primitives retain their real layout.
	/// </summary>
	privatestaticArgTypeInfoGcTypeKindToArgTypeInfo(GcTypeKindkind,intpointerSize)
	{
	returnkindswitch
	{
	GcTypeKind.Ref=> ArgTypeInfo.ForPrimitive(CorElementType.Class,pointerSize),
	GcTypeKind.Interior=> ArgTypeInfo.ForPrimitive(CorElementType.Byref,pointerSize),
	GcTypeKind.Other=>new ArgTypeInfo
	{
	CorElementType=CorElementType.ValueType,
	Size=pointerSize,
	IsValueType=true,
	},
	_ =>ArgTypeInfo.ForPrimitive(CorElementType.I,pointerSize),
	};
	}
	privatestaticboolTryGetArgTypeInfosFromMethodSignature(ReadOnlySpan<byte>signatureBytes,intpointerSize,outArgTypeInfo[]argTypes)
	{
	argTypes=Array.Empty<ArgTypeInfo>();
	if(signatureBytes.IsEmpty)
	returnfalse;
	BlobReaderreader=new(signatureBytes);
	SignatureHeaderheader=reader.ReadSignatureHeader();
	if(header.IsGeneric)
	reader.ReadCompressedInteger();
	intparameterCount=reader.ReadCompressedInteger();
	if(parameterCount<0)
	returnfalse;
	if(!TrySkipSignatureType(refreader))
	returnfalse;
	ArgTypeInfo[]decodedArgTypes=newArgTypeInfo[parameterCount];
	for(inti=0;i<parameterCount;i++)
	{
	if(!TryReadArgTypeInfo(refreader,pointerSize,outdecodedArgTypes[i]))
	returnfalse;
	}
	argTypes=decodedArgTypes;
	returntrue;
	}
	privatestaticboolTryReadArgTypeInfo(refBlobReaderreader,intpointerSize,outArgTypeInfoargTypeInfo)
	{
	while(reader.RemainingBytes>0)
	{
	CorElementTypeelementType=(CorElementType)reader.ReadByte();
	switch(elementType)
	{
	caseCorElementType.CmodOpt:
	caseCorElementType.CmodReqd:
	reader.ReadCompressedInteger();
	continue;
	caseCorElementType.Pinned:
	continue;
	caseCorElementType.Boolean:
	caseCorElementType.I1:
	caseCorElementType.U1:
	argTypeInfo=ArgTypeInfo.ForPrimitive(elementType,1);
	returntrue;
	caseCorElementType.Char:
	caseCorElementType.I2:
	caseCorElementType.U2:
	argTypeInfo=ArgTypeInfo.ForPrimitive(elementType,2);
	returntrue;
	caseCorElementType.I4:
	caseCorElementType.U4:
	caseCorElementType.R4:
	argTypeInfo=ArgTypeInfo.ForPrimitive(elementType,4);
	returntrue;
	caseCorElementType.I8:
	caseCorElementType.U8:
	caseCorElementType.R8:
	argTypeInfo=ArgTypeInfo.ForPrimitive(elementType,8);
	returntrue;
	caseCorElementType.I:
	caseCorElementType.U:
	caseCorElementType.Ptr:
	caseCorElementType.FnPtr:
	if(elementTypeisCorElementType.Ptr or CorElementType.FnPtr)
	{
	if(!TrySkipSignatureType(refreader))
	{
	argTypeInfo=default;
	returnfalse;
	}
	}
	argTypeInfo=ArgTypeInfo.ForPrimitive(CorElementType.I,pointerSize);
	returntrue;
	caseCorElementType.Byref:
	if(!TrySkipSignatureType(refreader))
	{
	argTypeInfo=default;
	returnfalse;
	}
	argTypeInfo=ArgTypeInfo.ForPrimitive(CorElementType.Byref,pointerSize);
	returntrue;
	caseCorElementType.String:
	caseCorElementType.Object:
	caseCorElementType.Class:
	if(elementTypeisCorElementType.Class)
	reader.ReadCompressedInteger();
	argTypeInfo=ArgTypeInfo.ForPrimitive(CorElementType.Class,pointerSize);
	returntrue;
	caseCorElementType.SzArray:
	if(!TrySkipSignatureType(refreader))
	{
	argTypeInfo=default;
	returnfalse;
	}
	argTypeInfo=ArgTypeInfo.ForPrimitive(CorElementType.Class,pointerSize);
	returntrue;
	caseCorElementType.ValueType:
	reader.ReadCompressedInteger();
	argTypeInfo=newArgTypeInfo
	{
	CorElementType=CorElementType.ValueType,
	Size=pointerSize,
	IsValueType=true,
	};
	returntrue;
	caseCorElementType.GenericInst:
	{
	if(!TrySkipSignatureType(refreader))
	{
	argTypeInfo=default;
	returnfalse;
	}
	intgenericArgCount=reader.ReadCompressedInteger();
	for(inti=0;i<genericArgCount;i++)
	{
	if(!TrySkipSignatureType(refreader))
	{
	argTypeInfo=default;
	returnfalse;
	}
	}
	argTypeInfo=newArgTypeInfo
	{
	CorElementType=CorElementType.ValueType,
	Size=pointerSize,
	IsValueType=true,
	};
	returntrue;
	}
	default:
	argTypeInfo=default;
	returnfalse;
	}
	}
	argTypeInfo=default;
	returnfalse;
	}
	privatestaticboolTrySkipSignatureType(refBlobReaderreader)
	{
	while(reader.RemainingBytes>0)
	{
	CorElementTypeelementType=(CorElementType)reader.ReadByte();
	switch(elementType)
	{
	caseCorElementType.CmodOpt:
	caseCorElementType.CmodReqd:
	reader.ReadCompressedInteger();
	continue;
	caseCorElementType.Pinned:
	continue;
	caseCorElementType.Void:
	caseCorElementType.Boolean:
	caseCorElementType.Char:
	caseCorElementType.I1:
	caseCorElementType.U1:
	caseCorElementType.I2:
	caseCorElementType.U2:
	caseCorElementType.I4:
	caseCorElementType.U4:
	caseCorElementType.I8:
	caseCorElementType.U8:
	caseCorElementType.R4:
	caseCorElementType.R8:
	caseCorElementType.String:
	caseCorElementType.Object:
	caseCorElementType.TypedByref:
	caseCorElementType.I:
	caseCorElementType.U:
	returntrue;
	caseCorElementType.Class:
	caseCorElementType.ValueType:
	caseCorElementType.Var:
	caseCorElementType.MVar:
	reader.ReadCompressedInteger();
	returntrue;
	caseCorElementType.Byref:
	caseCorElementType.Ptr:
	caseCorElementType.SzArray:
	returnTrySkipSignatureType(refreader);
	caseCorElementType.GenericInst:
	{
	if(!TrySkipSignatureType(refreader))
	returnfalse;
	intgenericArgCount=reader.ReadCompressedInteger();
	for(inti=0;i<genericArgCount;i++)
	{
	if(!TrySkipSignatureType(refreader))
	returnfalse;
	}
	returntrue;
	}
	caseCorElementType.Array:
	{
	if(!TrySkipSignatureType(refreader))
	returnfalse;
	intrank=reader.ReadCompressedInteger();
	intsizes=reader.ReadCompressedInteger();
	for(inti=0;i<sizes;i++)
	reader.ReadCompressedInteger();
	intlowerBounds=reader.ReadCompressedInteger();
	for(inti=0;i<lowerBounds;i++)
	reader.ReadCompressedInteger();
	returnrank>=0;
	}
	caseCorElementType.FnPtr:
	{
	SignatureHeaderheader=reader.ReadSignatureHeader();
	if(header.IsGeneric)
	reader.ReadCompressedInteger();
	intparameterCount=reader.ReadCompressedInteger();
	if(parameterCount<0)
	returnfalse;
	if(!TrySkipSignatureType(refreader))
	returnfalse;
	for(inti=0;i<parameterCount;i++)
	{
	if(!TrySkipSignatureType(refreader))
	returnfalse;
	}
	returntrue;
	}
	default:
	returnfalse;
	}
	}
	returnfalse;
	}

[Copilot]

Pull request overview

Copilot reviewed 64 out of 64 changed files in this pull request and generated 3 comments.

[Copilot]

CollectStackRefs currently appends every ref returned by ISOSStackRefEnum::Next with no upper bound. Later comparison helpers allocate fixed-size arrays sized by MAX_COLLECTED_REFS (e.g., matched[MAX_COLLECTED_REFS], aUsed[MAX_COLLECTED_REFS]), so if enumeration returns more than MAX_COLLECTED_REFS it can lead to out-of-bounds writes. Please cap collection to MAX_COLLECTED_REFS (and record an overflow/skip reason) or make the comparison logic handle arbitrarily large ref sets safely.

[Copilot]

When CollectRuntimeStackRefs overflows, the code logs a [SKIP] line but continues and still computes rtMatch/pass (and does not increment s_verifySkip). If CDACSTRESS_USE_DAC is not set, this can produce false failures based on a truncated runtime ref set. Consider treating runtime overflow as an actual skip (increment skip counter + return) when RT comparison is used for pass/fail, or otherwise avoid using rtMatch in the presence of overflow.

Suggested change

	if (rtOverflow && s_logFile != nullptr)
	{
	fprintf(s_logFile, "[SKIP] Thread=0x%x IP=0x%p - RT overflow (>%d refs)\n",
	osThreadId, (void*)GetIP(regs), MAX_COLLECTED_REFS);
	if (rtOverflow)
	{
	InterlockedIncrement(&s_verifySkip);
	if (s_logFile != nullptr)
	{
	fprintf(s_logFile, "[SKIP] Thread=0x%x IP=0x%p - RT overflow (>%d refs)\n",
	osThreadId, (void*)GetIP(regs), MAX_COLLECTED_REFS);
	}
	return;

[Copilot]

In ReportUntrackedAndSucceed, ReportSlot is called with reportScratchSlots: true, which forces reporting scratch registers/stack slots even when CodeManagerFlags.ActiveStackFrame is not set (reportScratchSlots local is false). This changes GC root enumeration behavior and can introduce extra roots for non-leaf frames. It looks like this should pass the reportScratchSlots variable instead of always true.

Suggested change

	ReportSlot(slotIndex,reportScratchSlots:true,reportFpBasedSlotsOnly,reportSlot);
	}
	ReportSlot(slotIndex,reportScratchSlots,reportFpBasedSlotsOnly,reportSlot);
	}

[Copilot]

Pull request overview

Copilot reviewed 64 out of 64 changed files in this pull request and generated 2 comments.

[Copilot]

CompareRefSetsFlat uses a fixed-size matched[MAX_COLLECTED_REFS] buffer, but cDAC/DAC ref collection (CollectStackRefs) is unbounded. Without a guard/cap, a large ref set (>4096) can cause out-of-bounds writes during matching. Consider capping cDAC/DAC collection to MAX_COLLECTED_REFS (and treating overflow as a [SKIP]) or using dynamically sized bookkeeping here.

[Copilot]

The async stdout/stderr collection uses stdout += ... / stderr += ... inside DataReceived event handlers. These callbacks can run concurrently, and string concatenation is not thread-safe; it can also be costly for large output. Consider buffering with a thread-safe collector (e.g., ConcurrentQueue<string> or StringBuilder with a lock) and call process.WaitForExit() (or await stream completion) after WaitForExit(timeout) to ensure all async output has been drained before asserting/logging.

[max-charlamb]

Closing in favor of stacked PR approach

[max-charlamb]

Summary

cDAC GC stress verification tool (DOTNET_CdacStress) that compares stack GC references between the cDAC and the runtime at allocation stress points. Includes stack walker fixes, GC reference scanning implementation, custom signature decoding, and calling convention argument iteration.

Note

This PR description was updated with AI assistance from Copilot.

Stack walker fixes

• Fix SW_SKIPPED_FRAME: do not call UpdateContextFromFrame (matches native SFITER_SKIPPED_FRAME_FUNCTION which does not call UpdateRegDisplay)
• Fix skipped-frame loop: check for more skipped frames before yielding managed method (prevents duplicate EnumGcRefs between consecutive skipped frames)
• Restructure Filter() to drive Next() directly, matching native Filter()+NextRaw() integration (prevents funclet-to-parent walk cycles)
• Remove SkipActiveICFOnce/SkipCurrentFrameInCheck — active ICF double-yield is natural and harmless
• Remove IsAtFirstPassExceptionThrowSite — native does not suppress first-pass refs
• Fix IsFirst not preserved for skipped frames (was causing IsActiveFrame=false for the topmost managed frame)

GC reference scanning

• Implement PromoteCallerStack for stub frames (GCRefMap + MetaSig paths)
• Implement SOSDacImpl.GetStackReferences using cDAC contract (was falling back to legacy DAC)
• Read FilterContext for stack walk starting context
• Three-way cDAC/DAC/RT comparison with InProcessDataTarget
• DOTNET_CdacStress bit flags: ALLOC/INSTR/REFS/WALK/USE_DAC/UNIQUE

RuntimeSignatureDecoder

Custom signature decoder that handles ELEMENT_TYPE_INTERNAL (0x21) and ELEMENT_TYPE_CMOD_INTERNAL (0x22) which SRM's SignatureDecoder cannot parse. These occur in IL stubs, marshalling stubs, and unsafe accessor frames.

• IRuntimeSignatureTypeProvider<TType, TGenericContext>: superset of SRM's ISignatureTypeProvider, adds GetInternalType and GetInternalModifiedType
• ISignatureReader + SpanSignatureReader: endianness-aware reader abstraction
• Correct ECMA-335 compliance: TypeDefOrRefOrSpecEncoded token decoding, sign-extension-by-width for compressed signed ints, bounds validation

ArgIterator (ported from crossgen2)

Proper calling convention analysis replacing the simplified 1-slot-per-param approach.

• CallingConventionInfo: hybrid data descriptor layout values + ABI invariant constants for all architectures (x86, x64 Windows/Unix, ARM32, ARM64, LoongArch64, RISC-V64)
• ArgIterator.GetNextOffset(): maps each argument to its actual register or stack offset
• OffsetOfFloatArgumentRegisters added to TransitionBlock data descriptor
• Handles multi-slot args, forced-byref params, return buffer placement, async continuation

CI infrastructure

• cDAC stress tests run in Helix via cdac-stress-helix.proj
• Extended runtime-diagnostics.yml CdacDumpTests buildArgs with +tools.cdacstresstests

Test results

Allocation-level stress (9 debuggees, ~46K verifications):

Debuggee	Verifications	Pass	Fail
BasicAlloc	4,936	4,936	0
DeepStack	4,964	4,964	0
Generics	4,936	4,936	0
MultiThread	4,996	4,996	0
Comprehensive	4,994	4,992	2*
ExceptionHandling	4,958	4,958	0
StructScenarios	4,940	4,940	0
DynamicMethods	6,520	6,520	0
PInvoke	4,936	4,936	0

Instruction-level stress (9 debuggees, ~226K verifications):
All 9 debuggees pass with zero failures across 226,452 verifications.

*Comprehensive's 2 failures are pre-existing legacy DAC issues (DAC returns 0 refs for background threads in kernel waits; cDAC and runtime agree).

[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

Adds a cDAC GC-stress verification harness and extends the cDAC stack-walk / stack-GC-ref pipeline so cDAC stack reference enumeration can be compared against runtime scanning at stress points.

Changes:

• Implements/extends cDAC stack reference enumeration (including Frame-based scanning paths like PromoteCallerStack via GCRefMap / MetaSig) and wires SOS GetStackReferences to the cDAC contract.
• Introduces a new GC stress integration test project with debuggee apps and orchestration targets.
• Extends CoreCLR cDAC stress/GC stress plumbing and data descriptors to support the new stack-walk and frame-scanning capabilities.

Reviewed changes

Copilot reviewed 46 out of 46 changed files in this pull request and generated 12 comments.

Show a summary per file

File	Description
src/native/managed/cdac/tests/MockDescriptors/MockDescriptors.ExecutionManager.cs	Updates mock type layout for execution manager-related data.
src/native/managed/cdac/tests/MockDescriptors/MockDescriptors.cs	Updates mock type layouts (ExceptionInfo/Thread) for contract tests.
src/native/managed/cdac/tests/Microsoft.Diagnostics.DataContractReader.Tests.csproj	Excludes new GCStressTests folder from the existing unit test project compilation.
src/native/managed/cdac/tests/GCStressTests/README.md	Documents how to build/run the new GC stress tests.
src/native/managed/cdac/tests/GCStressTests/Microsoft.Diagnostics.DataContractReader.GCStressTests.csproj	Adds a dedicated GC stress test project.
src/native/managed/cdac/tests/GCStressTests/GCStressTests.targets	MSBuild orchestration to discover/build debuggee projects.
src/native/managed/cdac/tests/GCStressTests/GCStressTestBase.cs	Test harness to run debuggees under corerun and parse verification logs.
src/native/managed/cdac/tests/GCStressTests/GCStressResults.cs	Parses the native verification log into structured pass/fail/skip counts.
src/native/managed/cdac/tests/GCStressTests/Debuggees/PInvoke/Program.cs	Adds a P/Invoke-focused debuggee workload.
src/native/managed/cdac/tests/GCStressTests/Debuggees/PInvoke/PInvoke.csproj	Debuggee project file for PInvoke scenario.
src/native/managed/cdac/tests/GCStressTests/Debuggees/MultiThread/Program.cs	Adds a multi-threaded debuggee workload.
src/native/managed/cdac/tests/GCStressTests/Debuggees/MultiThread/MultiThread.csproj	Debuggee project file for MultiThread scenario.
src/native/managed/cdac/tests/GCStressTests/Debuggees/Generics/Program.cs	Adds a generics/interface/delegate debuggee workload.
src/native/managed/cdac/tests/GCStressTests/Debuggees/Generics/Generics.csproj	Debuggee project file for Generics scenario.
src/native/managed/cdac/tests/GCStressTests/Debuggees/ExceptionHandling/Program.cs	Adds an exception-handling/funclet debuggee workload.
src/native/managed/cdac/tests/GCStressTests/Debuggees/ExceptionHandling/ExceptionHandling.csproj	Debuggee project file for ExceptionHandling scenario.
src/native/managed/cdac/tests/GCStressTests/Debuggees/Directory.Build.props	Shared build props for debuggee projects (output layout, TFM, etc.).
src/native/managed/cdac/tests/GCStressTests/Debuggees/DeepStack/Program.cs	Adds deep-recursion debuggee workload.
src/native/managed/cdac/tests/GCStressTests/Debuggees/DeepStack/DeepStack.csproj	Debuggee project file for DeepStack scenario.
src/native/managed/cdac/tests/GCStressTests/Debuggees/Comprehensive/Program.cs	Adds comprehensive “all scenarios” debuggee workload.
src/native/managed/cdac/tests/GCStressTests/Debuggees/Comprehensive/Comprehensive.csproj	Debuggee project file for Comprehensive scenario.
src/native/managed/cdac/tests/GCStressTests/Debuggees/BasicAlloc/Program.cs	Adds basic allocation/live-ref debuggee workload.
src/native/managed/cdac/tests/GCStressTests/Debuggees/BasicAlloc/BasicAlloc.csproj	Debuggee project file for BasicAlloc scenario.
src/native/managed/cdac/tests/GCStressTests/BasicGCStressTests.cs	Theory-based test suite that runs the debuggees and asserts pass rate.
src/native/managed/cdac/tests/gcstress/known-issues.md	Captures known mismatch classes and limitations.
src/native/managed/cdac/Microsoft.Diagnostics.DataContractReader.Legacy/SOSDacImpl.cs	Implements GetStackReferences using the cDAC contract rather than legacy DAC fallback.
src/native/managed/cdac/Microsoft.Diagnostics.DataContractReader.Contracts/Data/Frames/StubDispatchFrame.cs	Extends StubDispatchFrame data with GCRefMap pointer.
src/native/managed/cdac/Microsoft.Diagnostics.DataContractReader.Contracts/Data/Frames/ExternalMethodFrame.cs	Adds ExternalMethodFrame contract data type (GCRefMap).
src/native/managed/cdac/Microsoft.Diagnostics.DataContractReader.Contracts/Data/Frames/DynamicHelperFrame.cs	Adds DynamicHelperFrame contract data type (flags).
src/native/managed/cdac/Microsoft.Diagnostics.DataContractReader.Contracts/Data/ExceptionInfo.cs	Adds clause-range fields used for catch-handler resumption offset selection.
src/native/managed/cdac/Microsoft.Diagnostics.DataContractReader.Contracts/Contracts/StackWalk/StackWalk_1.cs	Refactors stack-walk filtering and adds Frame-based GC root scanning.
src/native/managed/cdac/Microsoft.Diagnostics.DataContractReader.Contracts/Contracts/StackWalk/GC/GcScanner.cs	Adds optional relOffset override support for GC ref enumeration.
src/native/managed/cdac/Microsoft.Diagnostics.DataContractReader.Contracts/Contracts/StackWalk/GC/GCRefMapDecoder.cs	Implements GCRefMap decoding for transition-block scanning.
src/native/managed/cdac/Microsoft.Diagnostics.DataContractReader.Contracts/Contracts/StackWalk/GC/CorSigParser.cs	Adds minimal signature parsing to classify parameters for MetaSig-based scanning.
src/native/managed/cdac/Microsoft.Diagnostics.DataContractReader.Contracts/Contracts/GCInfo/IGCInfoDecoder.cs	Adds FindFirstInterruptiblePoint API to GCInfo decoders.
src/native/managed/cdac/Microsoft.Diagnostics.DataContractReader.Contracts/Contracts/GCInfo/GCInfoDecoder.cs	Implements FindFirstInterruptiblePoint using decoded interruptible ranges.
src/native/managed/cdac/Microsoft.Diagnostics.DataContractReader.Contracts/Contracts/ExecutionManager/ExecutionManagerCore.EEJitManager.cs	Fixes code-start lookup for exception clause enumeration and adds minor flow adjustments.
src/native/managed/cdac/Microsoft.Diagnostics.DataContractReader.Contracts/Constants.cs	Adds TransitionBlock-related global names used by frame scanning.
src/native/managed/cdac/Microsoft.Diagnostics.DataContractReader.Abstractions/DataType.cs	Adds new DataType enum values for new frame contracts.
src/native/managed/cdac/cdac.slnx	Adds the new GC stress test project to the cDAC solution.
src/coreclr/vm/gccover.cpp	Adds step-based skipping to reduce overhead when throttling verification.
src/coreclr/vm/frames.h	Exposes additional frame fields to cDAC via cdac_data<> descriptors.
src/coreclr/vm/datadescriptor/datadescriptor.inc	Adds new contract fields/globals for frames and TransitionBlock layout.
src/coreclr/vm/cdacstress.cpp	Updates in-process cDAC/DAC verification logic, logging, and step behavior.
eng/Subsets.props	Adds an on-demand subset for running GC stress tests.
docs/design/datacontracts/StackWalk.md	Documents the new frame fields and TransitionBlock globals in the StackWalk contract.

Comments suppressed due to low confidence (1)

src/native/managed/cdac/tests/MockDescriptors/MockDescriptors.cs:228

• ThreadFields contains ProfilerFilterContext twice (lines 224 and 228). Duplicate field entries will skew offsets and make the mock descriptors inconsistent with the real data descriptor. Keep a single ProfilerFilterContext entry in the correct order.

[Copilot]

Pull request overview

Copilot reviewed 50 out of 50 changed files in this pull request and generated 7 comments.

Comments suppressed due to low confidence (2)

src/native/managed/cdac/tests/MockDescriptors/MockDescriptors.cs:229

• ThreadFields now includes DebuggerFilterContext/ProfilerFilterContext twice (duplicate entries at the end of the list). This can cause ambiguous/incorrect field offsets in the mock type layout. Keep each field only once.
  src/coreclr/vm/cdacstress.cpp:963
• CompareRefSets uses a fixed-size matched[MAX_COLLECTED_REFS] buffer but no longer validates that countA/countB are <= MAX_COLLECTED_REFS. Since CollectStackRefs appends without a hard cap, this can lead to out-of-bounds access when countA or countB exceeds 4096. Reintroduce a guard or allocate the match state sized to the counts.

 return true;
bool matched[MAX_COLLECTED_REFS] = {};
for (int i = 0; i < countA; i++)

[Copilot]

Pull request overview

Copilot reviewed 53 out of 53 changed files in this pull request and generated 7 comments.

[max-charlamb]

feedback for local copilot

[Copilot]

Pull request overview

Copilot reviewed 51 out of 51 changed files in this pull request and generated 9 comments.

Comments suppressed due to low confidence (1)

src/coreclr/vm/cdacstress.cpp:625

• CollectStackRefs appends to pRefs without any cap. Later comparisons allocate fixed-size arrays sized MAX_COLLECTED_REFS (e.g., matched[MAX_COLLECTED_REFS], aUsed/bUsed) and assume counts fit. If the DAC returns > MAX_COLLECTED_REFS refs, this will lead to out-of-bounds writes/reads. Add an explicit limit/overflow handling in CollectStackRefs (stop at MAX_COLLECTED_REFS and mark overflow / SKIP), or change the later comparison logic to handle arbitrary counts safely.

 SOSStackRefData refData;
unsigned int fetched = 0;
while (true)
{
hr = pEnum->Next(1, &refData, &fetched);
if (FAILED(hr) || fetched == 0)
break;
StackRef ref;
ref.Address = refData.Address;
ref.Object = refData.Object;
ref.Flags = refData.Flags;
ref.Source = refData.Source;
ref.SourceType = refData.SourceType;
ref.Register = refData.Register;
ref.Offset = refData.Offset;
ref.StackPointer = refData.StackPointer;
pRefs->Append(ref);
}

[Copilot]

Pull request overview

Copilot reviewed 52 out of 52 changed files in this pull request and generated 3 comments.

[Copilot]

Pull request overview

Copilot reviewed 65 out of 65 changed files in this pull request and generated 5 comments.

[Copilot]

ThreadStore now calls target.GetTypeInfo(DataType.SLink) and reads field "Next", but the CoreCLR data descriptor in this PR doesn't define a SLink type/field (could not find any CDAC_TYPE_BEGIN/FIELD for SLink). This will cause GetTypeInfo(DataType.SLink) to fail at runtime. Either add an SLink descriptor (with a "Next" pointer) on the runtime side or avoid needing type info here (e.g., treat the embedded SLink as a single pointer at offset 0).

[Copilot]

DataType adds SLink, but there is no corresponding runtime data descriptor definition (no CDAC_TYPE_BEGIN/FIELD(SLink, ...) found). Any attempt to read this type via Target.GetTypeInfo(DataType.SLink) will fail. Either add the runtime descriptor entry for SLink or remove this enum value and read the embedded link without type metadata.

Suggested change

	SLink,
	ThreadLocalData,
	ThreadLocalData=17,

[Copilot]

These DataReceived event handlers append to strings via "+=" from background threads. That is not thread-safe and can lead to lost/garbled output under contention. Consider using a StringBuilder with locking (or ConcurrentQueue) for stderr/stdout aggregation.

[Copilot]

Same thread-safety issue as stderr: appending to stdout with += from OutputDataReceived is racy. Use a synchronized StringBuilder/collector to avoid missing output and to keep logs reliable when tests fail.

[Copilot]

GcTypeKindToArgTypeInfo collapses all non-GC-ref types to CorElementType.I/pointer-sized and doesn't preserve float vs integer vs exact primitive sizes. ArgIterator offset calculation depends on the real signature shape (especially on Unix x64/ARM64 where float regs and argument sizing affect later argument placement), so this can produce incorrect offsets and cause missed/incorrect GC ref reporting for signatures with floats or non-pointer-sized primitives. Consider decoding into a richer type representation (e.g., CorElementType + size) and building ArgTypeInfo from that rather than from GcTypeKind.

Suggested change

	/// Converts a <see cref="GcTypeKind"/> to a minimal <see cref="ArgTypeInfo"/> for
	/// ArgIterator consumption. This is a bridge until we have a full type-aware provider.
	/// </summary>
	privatestaticArgTypeInfoGcTypeKindToArgTypeInfo(GcTypeKindkind,intpointerSize)
	{
	returnkindswitch
	{
	GcTypeKind.None=> ArgTypeInfo.ForPrimitive(CorElementType.I,pointerSize),
	GcTypeKind.Ref=> ArgTypeInfo.ForPrimitive(CorElementType.Class,pointerSize),
	GcTypeKind.Interior=> ArgTypeInfo.ForPrimitive(CorElementType.Byref,pointerSize),
	GcTypeKind.Other=>new ArgTypeInfo
	{
	CorElementType=CorElementType.ValueType,
	Size=pointerSize,// Conservative: assume pointer-sized for now
	IsValueType=true,
	},
	_ =>ArgTypeInfo.ForPrimitive(CorElementType.I,pointerSize),
	};
	}
	/// Converts a <see cref="GcTypeKind"/> to a conservative fallback <see cref="ArgTypeInfo"/>
	/// when an exact signature cannot be decoded. Callers should prefer decoding the real
	/// managed signature via <see cref="TryGetArgTypeInfosFromMethodSignature(ReadOnlySpan{byte}, int, out ArgTypeInfo[])"/>
	/// so that floating-point and non-pointer-sized primitives retain their real layout.
	/// </summary>
	privatestaticArgTypeInfoGcTypeKindToArgTypeInfo(GcTypeKindkind,intpointerSize)
	{
	returnkindswitch
	{
	GcTypeKind.Ref=> ArgTypeInfo.ForPrimitive(CorElementType.Class,pointerSize),
	GcTypeKind.Interior=> ArgTypeInfo.ForPrimitive(CorElementType.Byref,pointerSize),
	GcTypeKind.Other=>new ArgTypeInfo
	{
	CorElementType=CorElementType.ValueType,
	Size=pointerSize,
	IsValueType=true,
	},
	_ =>ArgTypeInfo.ForPrimitive(CorElementType.I,pointerSize),
	};
	}
	privatestaticboolTryGetArgTypeInfosFromMethodSignature(ReadOnlySpan<byte>signatureBytes,intpointerSize,outArgTypeInfo[]argTypes)
	{
	argTypes=Array.Empty<ArgTypeInfo>();
	if(signatureBytes.IsEmpty)
	returnfalse;
	BlobReaderreader=new(signatureBytes);
	SignatureHeaderheader=reader.ReadSignatureHeader();
	if(header.IsGeneric)
	reader.ReadCompressedInteger();
	intparameterCount=reader.ReadCompressedInteger();
	if(parameterCount<0)
	returnfalse;
	if(!TrySkipSignatureType(refreader))
	returnfalse;
	ArgTypeInfo[]decodedArgTypes=newArgTypeInfo[parameterCount];
	for(inti=0;i<parameterCount;i++)
	{
	if(!TryReadArgTypeInfo(refreader,pointerSize,outdecodedArgTypes[i]))
	returnfalse;
	}
	argTypes=decodedArgTypes;
	returntrue;
	}
	privatestaticboolTryReadArgTypeInfo(refBlobReaderreader,intpointerSize,outArgTypeInfoargTypeInfo)
	{
	while(reader.RemainingBytes>0)
	{
	CorElementTypeelementType=(CorElementType)reader.ReadByte();
	switch(elementType)
	{
	caseCorElementType.CmodOpt:
	caseCorElementType.CmodReqd:
	reader.ReadCompressedInteger();
	continue;
	caseCorElementType.Pinned:
	continue;
	caseCorElementType.Boolean:
	caseCorElementType.I1:
	caseCorElementType.U1:
	argTypeInfo=ArgTypeInfo.ForPrimitive(elementType,1);
	returntrue;
	caseCorElementType.Char:
	caseCorElementType.I2:
	caseCorElementType.U2:
	argTypeInfo=ArgTypeInfo.ForPrimitive(elementType,2);
	returntrue;
	caseCorElementType.I4:
	caseCorElementType.U4:
	caseCorElementType.R4:
	argTypeInfo=ArgTypeInfo.ForPrimitive(elementType,4);
	returntrue;
	caseCorElementType.I8:
	caseCorElementType.U8:
	caseCorElementType.R8:
	argTypeInfo=ArgTypeInfo.ForPrimitive(elementType,8);
	returntrue;
	caseCorElementType.I:
	caseCorElementType.U:
	caseCorElementType.Ptr:
	caseCorElementType.FnPtr:
	if(elementTypeisCorElementType.Ptr or CorElementType.FnPtr)
	{
	if(!TrySkipSignatureType(refreader))
	{
	argTypeInfo=default;
	returnfalse;
	}
	}
	argTypeInfo=ArgTypeInfo.ForPrimitive(CorElementType.I,pointerSize);
	returntrue;
	caseCorElementType.Byref:
	if(!TrySkipSignatureType(refreader))
	{
	argTypeInfo=default;
	returnfalse;
	}
	argTypeInfo=ArgTypeInfo.ForPrimitive(CorElementType.Byref,pointerSize);
	returntrue;
	caseCorElementType.String:
	caseCorElementType.Object:
	caseCorElementType.Class:
	if(elementTypeisCorElementType.Class)
	reader.ReadCompressedInteger();
	argTypeInfo=ArgTypeInfo.ForPrimitive(CorElementType.Class,pointerSize);
	returntrue;
	caseCorElementType.SzArray:
	if(!TrySkipSignatureType(refreader))
	{
	argTypeInfo=default;
	returnfalse;
	}
	argTypeInfo=ArgTypeInfo.ForPrimitive(CorElementType.Class,pointerSize);
	returntrue;
	caseCorElementType.ValueType:
	reader.ReadCompressedInteger();
	argTypeInfo=newArgTypeInfo
	{
	CorElementType=CorElementType.ValueType,
	Size=pointerSize,
	IsValueType=true,
	};
	returntrue;
	caseCorElementType.GenericInst:
	{
	if(!TrySkipSignatureType(refreader))
	{
	argTypeInfo=default;
	returnfalse;
	}
	intgenericArgCount=reader.ReadCompressedInteger();
	for(inti=0;i<genericArgCount;i++)
	{
	if(!TrySkipSignatureType(refreader))
	{
	argTypeInfo=default;
	returnfalse;
	}
	}
	argTypeInfo=newArgTypeInfo
	{
	CorElementType=CorElementType.ValueType,
	Size=pointerSize,
	IsValueType=true,
	};
	returntrue;
	}
	default:
	argTypeInfo=default;
	returnfalse;
	}
	}
	argTypeInfo=default;
	returnfalse;
	}
	privatestaticboolTrySkipSignatureType(refBlobReaderreader)
	{
	while(reader.RemainingBytes>0)
	{
	CorElementTypeelementType=(CorElementType)reader.ReadByte();
	switch(elementType)
	{
	caseCorElementType.CmodOpt:
	caseCorElementType.CmodReqd:
	reader.ReadCompressedInteger();
	continue;
	caseCorElementType.Pinned:
	continue;
	caseCorElementType.Void:
	caseCorElementType.Boolean:
	caseCorElementType.Char:
	caseCorElementType.I1:
	caseCorElementType.U1:
	caseCorElementType.I2:
	caseCorElementType.U2:
	caseCorElementType.I4:
	caseCorElementType.U4:
	caseCorElementType.I8:
	caseCorElementType.U8:
	caseCorElementType.R4:
	caseCorElementType.R8:
	caseCorElementType.String:
	caseCorElementType.Object:
	caseCorElementType.TypedByref:
	caseCorElementType.I:
	caseCorElementType.U:
	returntrue;
	caseCorElementType.Class:
	caseCorElementType.ValueType:
	caseCorElementType.Var:
	caseCorElementType.MVar:
	reader.ReadCompressedInteger();
	returntrue;
	caseCorElementType.Byref:
	caseCorElementType.Ptr:
	caseCorElementType.SzArray:
	returnTrySkipSignatureType(refreader);
	caseCorElementType.GenericInst:
	{
	if(!TrySkipSignatureType(refreader))
	returnfalse;
	intgenericArgCount=reader.ReadCompressedInteger();
	for(inti=0;i<genericArgCount;i++)
	{
	if(!TrySkipSignatureType(refreader))
	returnfalse;
	}
	returntrue;
	}
	caseCorElementType.Array:
	{
	if(!TrySkipSignatureType(refreader))
	returnfalse;
	intrank=reader.ReadCompressedInteger();
	intsizes=reader.ReadCompressedInteger();
	for(inti=0;i<sizes;i++)
	reader.ReadCompressedInteger();
	intlowerBounds=reader.ReadCompressedInteger();
	for(inti=0;i<lowerBounds;i++)
	reader.ReadCompressedInteger();
	returnrank>=0;
	}
	caseCorElementType.FnPtr:
	{
	SignatureHeaderheader=reader.ReadSignatureHeader();
	if(header.IsGeneric)
	reader.ReadCompressedInteger();
	intparameterCount=reader.ReadCompressedInteger();
	if(parameterCount<0)
	returnfalse;
	if(!TrySkipSignatureType(refreader))
	returnfalse;
	for(inti=0;i<parameterCount;i++)
	{
	if(!TrySkipSignatureType(refreader))
	returnfalse;
	}
	returntrue;
	}
	default:
	returnfalse;
	}
	}
	returnfalse;
	}

[Copilot]

Pull request overview

Copilot reviewed 64 out of 64 changed files in this pull request and generated 3 comments.

[Copilot]

CollectStackRefs currently appends every ref returned by ISOSStackRefEnum::Next with no upper bound. Later comparison helpers allocate fixed-size arrays sized by MAX_COLLECTED_REFS (e.g., matched[MAX_COLLECTED_REFS], aUsed[MAX_COLLECTED_REFS]), so if enumeration returns more than MAX_COLLECTED_REFS it can lead to out-of-bounds writes. Please cap collection to MAX_COLLECTED_REFS (and record an overflow/skip reason) or make the comparison logic handle arbitrarily large ref sets safely.

[Copilot]

When CollectRuntimeStackRefs overflows, the code logs a [SKIP] line but continues and still computes rtMatch/pass (and does not increment s_verifySkip). If CDACSTRESS_USE_DAC is not set, this can produce false failures based on a truncated runtime ref set. Consider treating runtime overflow as an actual skip (increment skip counter + return) when RT comparison is used for pass/fail, or otherwise avoid using rtMatch in the presence of overflow.

Suggested change

	if (rtOverflow && s_logFile != nullptr)
	{
	fprintf(s_logFile, "[SKIP] Thread=0x%x IP=0x%p - RT overflow (>%d refs)\n",
	osThreadId, (void*)GetIP(regs), MAX_COLLECTED_REFS);
	if (rtOverflow)
	{
	InterlockedIncrement(&s_verifySkip);
	if (s_logFile != nullptr)
	{
	fprintf(s_logFile, "[SKIP] Thread=0x%x IP=0x%p - RT overflow (>%d refs)\n",
	osThreadId, (void*)GetIP(regs), MAX_COLLECTED_REFS);
	}
	return;

[Copilot]

In ReportUntrackedAndSucceed, ReportSlot is called with reportScratchSlots: true, which forces reporting scratch registers/stack slots even when CodeManagerFlags.ActiveStackFrame is not set (reportScratchSlots local is false). This changes GC root enumeration behavior and can introduce extra roots for non-leaf frames. It looks like this should pass the reportScratchSlots variable instead of always true.

Suggested change

	ReportSlot(slotIndex,reportScratchSlots:true,reportFpBasedSlotsOnly,reportSlot);
	}
	ReportSlot(slotIndex,reportScratchSlots,reportFpBasedSlotsOnly,reportSlot);
	}

[Copilot]

Pull request overview

Copilot reviewed 64 out of 64 changed files in this pull request and generated 2 comments.

[Copilot]

CompareRefSetsFlat uses a fixed-size matched[MAX_COLLECTED_REFS] buffer, but cDAC/DAC ref collection (CollectStackRefs) is unbounded. Without a guard/cap, a large ref set (>4096) can cause out-of-bounds writes during matching. Consider capping cDAC/DAC collection to MAX_COLLECTED_REFS (and treating overflow as a [SKIP]) or using dynamically sized bookkeeping here.

[Copilot]

The async stdout/stderr collection uses stdout += ... / stderr += ... inside DataReceived event handlers. These callbacks can run concurrently, and string concatenation is not thread-safe; it can also be costly for large output. Consider buffering with a thread-safe collector (e.g., ConcurrentQueue<string> or StringBuilder with a lock) and call process.WaitForExit() (or await stream completion) after WaitForExit(timeout) to ensure all async output has been drained before asserting/logging.

[max-charlamb]

Closing in favor of stacked PR approach

[max-charlamb]

Summary

cDAC GC stress verification tool (DOTNET_CdacStress) that compares stack GC references between the cDAC and the runtime at allocation stress points. Includes stack walker fixes, GC reference scanning implementation, custom signature decoding, and calling convention argument iteration.

Note

This PR description was updated with AI assistance from Copilot.

Stack walker fixes

• Fix SW_SKIPPED_FRAME: do not call UpdateContextFromFrame (matches native SFITER_SKIPPED_FRAME_FUNCTION which does not call UpdateRegDisplay)
• Fix skipped-frame loop: check for more skipped frames before yielding managed method (prevents duplicate EnumGcRefs between consecutive skipped frames)
• Restructure Filter() to drive Next() directly, matching native Filter()+NextRaw() integration (prevents funclet-to-parent walk cycles)
• Remove SkipActiveICFOnce/SkipCurrentFrameInCheck — active ICF double-yield is natural and harmless
• Remove IsAtFirstPassExceptionThrowSite — native does not suppress first-pass refs
• Fix IsFirst not preserved for skipped frames (was causing IsActiveFrame=false for the topmost managed frame)

GC reference scanning

• Implement PromoteCallerStack for stub frames (GCRefMap + MetaSig paths)
• Implement SOSDacImpl.GetStackReferences using cDAC contract (was falling back to legacy DAC)
• Read FilterContext for stack walk starting context
• Three-way cDAC/DAC/RT comparison with InProcessDataTarget
• DOTNET_CdacStress bit flags: ALLOC/INSTR/REFS/WALK/USE_DAC/UNIQUE

RuntimeSignatureDecoder

Custom signature decoder that handles ELEMENT_TYPE_INTERNAL (0x21) and ELEMENT_TYPE_CMOD_INTERNAL (0x22) which SRM's SignatureDecoder cannot parse. These occur in IL stubs, marshalling stubs, and unsafe accessor frames.

• IRuntimeSignatureTypeProvider<TType, TGenericContext>: superset of SRM's ISignatureTypeProvider, adds GetInternalType and GetInternalModifiedType
• ISignatureReader + SpanSignatureReader: endianness-aware reader abstraction
• Correct ECMA-335 compliance: TypeDefOrRefOrSpecEncoded token decoding, sign-extension-by-width for compressed signed ints, bounds validation

ArgIterator (ported from crossgen2)

Proper calling convention analysis replacing the simplified 1-slot-per-param approach.

• CallingConventionInfo: hybrid data descriptor layout values + ABI invariant constants for all architectures (x86, x64 Windows/Unix, ARM32, ARM64, LoongArch64, RISC-V64)
• ArgIterator.GetNextOffset(): maps each argument to its actual register or stack offset
• OffsetOfFloatArgumentRegisters added to TransitionBlock data descriptor
• Handles multi-slot args, forced-byref params, return buffer placement, async continuation

CI infrastructure

• cDAC stress tests run in Helix via cdac-stress-helix.proj
• Extended runtime-diagnostics.yml CdacDumpTests buildArgs with +tools.cdacstresstests

Test results

Allocation-level stress (9 debuggees, ~46K verifications):

Debuggee	Verifications	Pass	Fail
BasicAlloc	4,936	4,936	0
DeepStack	4,964	4,964	0
Generics	4,936	4,936	0
MultiThread	4,996	4,996	0
Comprehensive	4,994	4,992	2*
ExceptionHandling	4,958	4,958	0
StructScenarios	4,940	4,940	0
DynamicMethods	6,520	6,520	0
PInvoke	4,936	4,936	0

Instruction-level stress (9 debuggees, ~226K verifications):
All 9 debuggees pass with zero failures across 226,452 verifications.

*Comprehensive's 2 failures are pre-existing legacy DAC issues (DAC returns 0 refs for background threads in kernel waits; cDAC and runtime agree).

[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

Adds a cDAC GC-stress verification harness and extends the cDAC stack-walk / stack-GC-ref pipeline so cDAC stack reference enumeration can be compared against runtime scanning at stress points.

Changes:

• Implements/extends cDAC stack reference enumeration (including Frame-based scanning paths like PromoteCallerStack via GCRefMap / MetaSig) and wires SOS GetStackReferences to the cDAC contract.
• Introduces a new GC stress integration test project with debuggee apps and orchestration targets.
• Extends CoreCLR cDAC stress/GC stress plumbing and data descriptors to support the new stack-walk and frame-scanning capabilities.

Reviewed changes

Copilot reviewed 46 out of 46 changed files in this pull request and generated 12 comments.

Show a summary per file

File	Description
src/native/managed/cdac/tests/MockDescriptors/MockDescriptors.ExecutionManager.cs	Updates mock type layout for execution manager-related data.
src/native/managed/cdac/tests/MockDescriptors/MockDescriptors.cs	Updates mock type layouts (ExceptionInfo/Thread) for contract tests.
src/native/managed/cdac/tests/Microsoft.Diagnostics.DataContractReader.Tests.csproj	Excludes new GCStressTests folder from the existing unit test project compilation.
src/native/managed/cdac/tests/GCStressTests/README.md	Documents how to build/run the new GC stress tests.
src/native/managed/cdac/tests/GCStressTests/Microsoft.Diagnostics.DataContractReader.GCStressTests.csproj	Adds a dedicated GC stress test project.
src/native/managed/cdac/tests/GCStressTests/GCStressTests.targets	MSBuild orchestration to discover/build debuggee projects.
src/native/managed/cdac/tests/GCStressTests/GCStressTestBase.cs	Test harness to run debuggees under corerun and parse verification logs.
src/native/managed/cdac/tests/GCStressTests/GCStressResults.cs	Parses the native verification log into structured pass/fail/skip counts.
src/native/managed/cdac/tests/GCStressTests/Debuggees/PInvoke/Program.cs	Adds a P/Invoke-focused debuggee workload.
src/native/managed/cdac/tests/GCStressTests/Debuggees/PInvoke/PInvoke.csproj	Debuggee project file for PInvoke scenario.
src/native/managed/cdac/tests/GCStressTests/Debuggees/MultiThread/Program.cs	Adds a multi-threaded debuggee workload.
src/native/managed/cdac/tests/GCStressTests/Debuggees/MultiThread/MultiThread.csproj	Debuggee project file for MultiThread scenario.
src/native/managed/cdac/tests/GCStressTests/Debuggees/Generics/Program.cs	Adds a generics/interface/delegate debuggee workload.
src/native/managed/cdac/tests/GCStressTests/Debuggees/Generics/Generics.csproj	Debuggee project file for Generics scenario.
src/native/managed/cdac/tests/GCStressTests/Debuggees/ExceptionHandling/Program.cs	Adds an exception-handling/funclet debuggee workload.
src/native/managed/cdac/tests/GCStressTests/Debuggees/ExceptionHandling/ExceptionHandling.csproj	Debuggee project file for ExceptionHandling scenario.
src/native/managed/cdac/tests/GCStressTests/Debuggees/Directory.Build.props	Shared build props for debuggee projects (output layout, TFM, etc.).
src/native/managed/cdac/tests/GCStressTests/Debuggees/DeepStack/Program.cs	Adds deep-recursion debuggee workload.
src/native/managed/cdac/tests/GCStressTests/Debuggees/DeepStack/DeepStack.csproj	Debuggee project file for DeepStack scenario.
src/native/managed/cdac/tests/GCStressTests/Debuggees/Comprehensive/Program.cs	Adds comprehensive “all scenarios” debuggee workload.
src/native/managed/cdac/tests/GCStressTests/Debuggees/Comprehensive/Comprehensive.csproj	Debuggee project file for Comprehensive scenario.
src/native/managed/cdac/tests/GCStressTests/Debuggees/BasicAlloc/Program.cs	Adds basic allocation/live-ref debuggee workload.
src/native/managed/cdac/tests/GCStressTests/Debuggees/BasicAlloc/BasicAlloc.csproj	Debuggee project file for BasicAlloc scenario.
src/native/managed/cdac/tests/GCStressTests/BasicGCStressTests.cs	Theory-based test suite that runs the debuggees and asserts pass rate.
src/native/managed/cdac/tests/gcstress/known-issues.md	Captures known mismatch classes and limitations.
src/native/managed/cdac/Microsoft.Diagnostics.DataContractReader.Legacy/SOSDacImpl.cs	Implements GetStackReferences using the cDAC contract rather than legacy DAC fallback.
src/native/managed/cdac/Microsoft.Diagnostics.DataContractReader.Contracts/Data/Frames/StubDispatchFrame.cs	Extends StubDispatchFrame data with GCRefMap pointer.
src/native/managed/cdac/Microsoft.Diagnostics.DataContractReader.Contracts/Data/Frames/ExternalMethodFrame.cs	Adds ExternalMethodFrame contract data type (GCRefMap).
src/native/managed/cdac/Microsoft.Diagnostics.DataContractReader.Contracts/Data/Frames/DynamicHelperFrame.cs	Adds DynamicHelperFrame contract data type (flags).
src/native/managed/cdac/Microsoft.Diagnostics.DataContractReader.Contracts/Data/ExceptionInfo.cs	Adds clause-range fields used for catch-handler resumption offset selection.
src/native/managed/cdac/Microsoft.Diagnostics.DataContractReader.Contracts/Contracts/StackWalk/StackWalk_1.cs	Refactors stack-walk filtering and adds Frame-based GC root scanning.
src/native/managed/cdac/Microsoft.Diagnostics.DataContractReader.Contracts/Contracts/StackWalk/GC/GcScanner.cs	Adds optional relOffset override support for GC ref enumeration.
src/native/managed/cdac/Microsoft.Diagnostics.DataContractReader.Contracts/Contracts/StackWalk/GC/GCRefMapDecoder.cs	Implements GCRefMap decoding for transition-block scanning.
src/native/managed/cdac/Microsoft.Diagnostics.DataContractReader.Contracts/Contracts/StackWalk/GC/CorSigParser.cs	Adds minimal signature parsing to classify parameters for MetaSig-based scanning.
src/native/managed/cdac/Microsoft.Diagnostics.DataContractReader.Contracts/Contracts/GCInfo/IGCInfoDecoder.cs	Adds FindFirstInterruptiblePoint API to GCInfo decoders.
src/native/managed/cdac/Microsoft.Diagnostics.DataContractReader.Contracts/Contracts/GCInfo/GCInfoDecoder.cs	Implements FindFirstInterruptiblePoint using decoded interruptible ranges.
src/native/managed/cdac/Microsoft.Diagnostics.DataContractReader.Contracts/Contracts/ExecutionManager/ExecutionManagerCore.EEJitManager.cs	Fixes code-start lookup for exception clause enumeration and adds minor flow adjustments.
src/native/managed/cdac/Microsoft.Diagnostics.DataContractReader.Contracts/Constants.cs	Adds TransitionBlock-related global names used by frame scanning.
src/native/managed/cdac/Microsoft.Diagnostics.DataContractReader.Abstractions/DataType.cs	Adds new DataType enum values for new frame contracts.
src/native/managed/cdac/cdac.slnx	Adds the new GC stress test project to the cDAC solution.
src/coreclr/vm/gccover.cpp	Adds step-based skipping to reduce overhead when throttling verification.
src/coreclr/vm/frames.h	Exposes additional frame fields to cDAC via cdac_data<> descriptors.
src/coreclr/vm/datadescriptor/datadescriptor.inc	Adds new contract fields/globals for frames and TransitionBlock layout.
src/coreclr/vm/cdacstress.cpp	Updates in-process cDAC/DAC verification logic, logging, and step behavior.
eng/Subsets.props	Adds an on-demand subset for running GC stress tests.
docs/design/datacontracts/StackWalk.md	Documents the new frame fields and TransitionBlock globals in the StackWalk contract.

Comments suppressed due to low confidence (1)

src/native/managed/cdac/tests/MockDescriptors/MockDescriptors.cs:228

• ThreadFields contains ProfilerFilterContext twice (lines 224 and 228). Duplicate field entries will skew offsets and make the mock descriptors inconsistent with the real data descriptor. Keep a single ProfilerFilterContext entry in the correct order.

[Copilot]

Pull request overview

Copilot reviewed 50 out of 50 changed files in this pull request and generated 7 comments.

Comments suppressed due to low confidence (2)

src/native/managed/cdac/tests/MockDescriptors/MockDescriptors.cs:229

• ThreadFields now includes DebuggerFilterContext/ProfilerFilterContext twice (duplicate entries at the end of the list). This can cause ambiguous/incorrect field offsets in the mock type layout. Keep each field only once.
  src/coreclr/vm/cdacstress.cpp:963
• CompareRefSets uses a fixed-size matched[MAX_COLLECTED_REFS] buffer but no longer validates that countA/countB are <= MAX_COLLECTED_REFS. Since CollectStackRefs appends without a hard cap, this can lead to out-of-bounds access when countA or countB exceeds 4096. Reintroduce a guard or allocate the match state sized to the counts.

 return true;
bool matched[MAX_COLLECTED_REFS] = {};
for (int i = 0; i < countA; i++)

[Copilot]

Pull request overview

Copilot reviewed 53 out of 53 changed files in this pull request and generated 7 comments.

[max-charlamb]

feedback for local copilot

[Copilot]

Pull request overview

Copilot reviewed 51 out of 51 changed files in this pull request and generated 9 comments.

Comments suppressed due to low confidence (1)

src/coreclr/vm/cdacstress.cpp:625

• CollectStackRefs appends to pRefs without any cap. Later comparisons allocate fixed-size arrays sized MAX_COLLECTED_REFS (e.g., matched[MAX_COLLECTED_REFS], aUsed/bUsed) and assume counts fit. If the DAC returns > MAX_COLLECTED_REFS refs, this will lead to out-of-bounds writes/reads. Add an explicit limit/overflow handling in CollectStackRefs (stop at MAX_COLLECTED_REFS and mark overflow / SKIP), or change the later comparison logic to handle arbitrary counts safely.

 SOSStackRefData refData;
unsigned int fetched = 0;
while (true)
{
hr = pEnum->Next(1, &refData, &fetched);
if (FAILED(hr) || fetched == 0)
break;
StackRef ref;
ref.Address = refData.Address;
ref.Object = refData.Object;
ref.Flags = refData.Flags;
ref.Source = refData.Source;
ref.SourceType = refData.SourceType;
ref.Register = refData.Register;
ref.Offset = refData.Offset;
ref.StackPointer = refData.StackPointer;
pRefs->Append(ref);
}

[Copilot]

Pull request overview

Copilot reviewed 52 out of 52 changed files in this pull request and generated 3 comments.

[Copilot]

Pull request overview

Copilot reviewed 65 out of 65 changed files in this pull request and generated 5 comments.

[Copilot]

ThreadStore now calls target.GetTypeInfo(DataType.SLink) and reads field "Next", but the CoreCLR data descriptor in this PR doesn't define a SLink type/field (could not find any CDAC_TYPE_BEGIN/FIELD for SLink). This will cause GetTypeInfo(DataType.SLink) to fail at runtime. Either add an SLink descriptor (with a "Next" pointer) on the runtime side or avoid needing type info here (e.g., treat the embedded SLink as a single pointer at offset 0).

[Copilot]

DataType adds SLink, but there is no corresponding runtime data descriptor definition (no CDAC_TYPE_BEGIN/FIELD(SLink, ...) found). Any attempt to read this type via Target.GetTypeInfo(DataType.SLink) will fail. Either add the runtime descriptor entry for SLink or remove this enum value and read the embedded link without type metadata.

Suggested change

	SLink,
	ThreadLocalData,
	ThreadLocalData=17,

[Copilot]

These DataReceived event handlers append to strings via "+=" from background threads. That is not thread-safe and can lead to lost/garbled output under contention. Consider using a StringBuilder with locking (or ConcurrentQueue) for stderr/stdout aggregation.

[Copilot]

Same thread-safety issue as stderr: appending to stdout with += from OutputDataReceived is racy. Use a synchronized StringBuilder/collector to avoid missing output and to keep logs reliable when tests fail.

[Copilot]

GcTypeKindToArgTypeInfo collapses all non-GC-ref types to CorElementType.I/pointer-sized and doesn't preserve float vs integer vs exact primitive sizes. ArgIterator offset calculation depends on the real signature shape (especially on Unix x64/ARM64 where float regs and argument sizing affect later argument placement), so this can produce incorrect offsets and cause missed/incorrect GC ref reporting for signatures with floats or non-pointer-sized primitives. Consider decoding into a richer type representation (e.g., CorElementType + size) and building ArgTypeInfo from that rather than from GcTypeKind.

Suggested change

	/// Converts a <see cref="GcTypeKind"/> to a minimal <see cref="ArgTypeInfo"/> for
	/// ArgIterator consumption. This is a bridge until we have a full type-aware provider.
	/// </summary>
	privatestaticArgTypeInfoGcTypeKindToArgTypeInfo(GcTypeKindkind,intpointerSize)
	{
	returnkindswitch
	{
	GcTypeKind.None=> ArgTypeInfo.ForPrimitive(CorElementType.I,pointerSize),
	GcTypeKind.Ref=> ArgTypeInfo.ForPrimitive(CorElementType.Class,pointerSize),
	GcTypeKind.Interior=> ArgTypeInfo.ForPrimitive(CorElementType.Byref,pointerSize),
	GcTypeKind.Other=>new ArgTypeInfo
	{
	CorElementType=CorElementType.ValueType,
	Size=pointerSize,// Conservative: assume pointer-sized for now
	IsValueType=true,
	},
	_ =>ArgTypeInfo.ForPrimitive(CorElementType.I,pointerSize),
	};
	}
	/// Converts a <see cref="GcTypeKind"/> to a conservative fallback <see cref="ArgTypeInfo"/>
	/// when an exact signature cannot be decoded. Callers should prefer decoding the real
	/// managed signature via <see cref="TryGetArgTypeInfosFromMethodSignature(ReadOnlySpan{byte}, int, out ArgTypeInfo[])"/>
	/// so that floating-point and non-pointer-sized primitives retain their real layout.
	/// </summary>
	privatestaticArgTypeInfoGcTypeKindToArgTypeInfo(GcTypeKindkind,intpointerSize)
	{
	returnkindswitch
	{
	GcTypeKind.Ref=> ArgTypeInfo.ForPrimitive(CorElementType.Class,pointerSize),
	GcTypeKind.Interior=> ArgTypeInfo.ForPrimitive(CorElementType.Byref,pointerSize),
	GcTypeKind.Other=>new ArgTypeInfo
	{
	CorElementType=CorElementType.ValueType,
	Size=pointerSize,
	IsValueType=true,
	},
	_ =>ArgTypeInfo.ForPrimitive(CorElementType.I,pointerSize),
	};
	}
	privatestaticboolTryGetArgTypeInfosFromMethodSignature(ReadOnlySpan<byte>signatureBytes,intpointerSize,outArgTypeInfo[]argTypes)
	{
	argTypes=Array.Empty<ArgTypeInfo>();
	if(signatureBytes.IsEmpty)
	returnfalse;
	BlobReaderreader=new(signatureBytes);
	SignatureHeaderheader=reader.ReadSignatureHeader();
	if(header.IsGeneric)
	reader.ReadCompressedInteger();
	intparameterCount=reader.ReadCompressedInteger();
	if(parameterCount<0)
	returnfalse;
	if(!TrySkipSignatureType(refreader))
	returnfalse;
	ArgTypeInfo[]decodedArgTypes=newArgTypeInfo[parameterCount];
	for(inti=0;i<parameterCount;i++)
	{
	if(!TryReadArgTypeInfo(refreader,pointerSize,outdecodedArgTypes[i]))
	returnfalse;
	}
	argTypes=decodedArgTypes;
	returntrue;
	}
	privatestaticboolTryReadArgTypeInfo(refBlobReaderreader,intpointerSize,outArgTypeInfoargTypeInfo)
	{
	while(reader.RemainingBytes>0)
	{
	CorElementTypeelementType=(CorElementType)reader.ReadByte();
	switch(elementType)
	{
	caseCorElementType.CmodOpt:
	caseCorElementType.CmodReqd:
	reader.ReadCompressedInteger();
	continue;
	caseCorElementType.Pinned:
	continue;
	caseCorElementType.Boolean:
	caseCorElementType.I1:
	caseCorElementType.U1:
	argTypeInfo=ArgTypeInfo.ForPrimitive(elementType,1);
	returntrue;
	caseCorElementType.Char:
	caseCorElementType.I2:
	caseCorElementType.U2:
	argTypeInfo=ArgTypeInfo.ForPrimitive(elementType,2);
	returntrue;
	caseCorElementType.I4:
	caseCorElementType.U4:
	caseCorElementType.R4:
	argTypeInfo=ArgTypeInfo.ForPrimitive(elementType,4);
	returntrue;
	caseCorElementType.I8:
	caseCorElementType.U8:
	caseCorElementType.R8:
	argTypeInfo=ArgTypeInfo.ForPrimitive(elementType,8);
	returntrue;
	caseCorElementType.I:
	caseCorElementType.U:
	caseCorElementType.Ptr:
	caseCorElementType.FnPtr:
	if(elementTypeisCorElementType.Ptr or CorElementType.FnPtr)
	{
	if(!TrySkipSignatureType(refreader))
	{
	argTypeInfo=default;
	returnfalse;
	}
	}
	argTypeInfo=ArgTypeInfo.ForPrimitive(CorElementType.I,pointerSize);
	returntrue;
	caseCorElementType.Byref:
	if(!TrySkipSignatureType(refreader))
	{
	argTypeInfo=default;
	returnfalse;
	}
	argTypeInfo=ArgTypeInfo.ForPrimitive(CorElementType.Byref,pointerSize);
	returntrue;
	caseCorElementType.String:
	caseCorElementType.Object:
	caseCorElementType.Class:
	if(elementTypeisCorElementType.Class)
	reader.ReadCompressedInteger();
	argTypeInfo=ArgTypeInfo.ForPrimitive(CorElementType.Class,pointerSize);
	returntrue;
	caseCorElementType.SzArray:
	if(!TrySkipSignatureType(refreader))
	{
	argTypeInfo=default;
	returnfalse;
	}
	argTypeInfo=ArgTypeInfo.ForPrimitive(CorElementType.Class,pointerSize);
	returntrue;
	caseCorElementType.ValueType:
	reader.ReadCompressedInteger();
	argTypeInfo=newArgTypeInfo
	{
	CorElementType=CorElementType.ValueType,
	Size=pointerSize,
	IsValueType=true,
	};
	returntrue;
	caseCorElementType.GenericInst:
	{
	if(!TrySkipSignatureType(refreader))
	{
	argTypeInfo=default;
	returnfalse;
	}
	intgenericArgCount=reader.ReadCompressedInteger();
	for(inti=0;i<genericArgCount;i++)
	{
	if(!TrySkipSignatureType(refreader))
	{
	argTypeInfo=default;
	returnfalse;
	}
	}
	argTypeInfo=newArgTypeInfo
	{
	CorElementType=CorElementType.ValueType,
	Size=pointerSize,
	IsValueType=true,
	};
	returntrue;
	}
	default:
	argTypeInfo=default;
	returnfalse;
	}
	}
	argTypeInfo=default;
	returnfalse;
	}
	privatestaticboolTrySkipSignatureType(refBlobReaderreader)
	{
	while(reader.RemainingBytes>0)
	{
	CorElementTypeelementType=(CorElementType)reader.ReadByte();
	switch(elementType)
	{
	caseCorElementType.CmodOpt:
	caseCorElementType.CmodReqd:
	reader.ReadCompressedInteger();
	continue;
	caseCorElementType.Pinned:
	continue;
	caseCorElementType.Void:
	caseCorElementType.Boolean:
	caseCorElementType.Char:
	caseCorElementType.I1:
	caseCorElementType.U1:
	caseCorElementType.I2:
	caseCorElementType.U2:
	caseCorElementType.I4:
	caseCorElementType.U4:
	caseCorElementType.I8:
	caseCorElementType.U8:
	caseCorElementType.R4:
	caseCorElementType.R8:
	caseCorElementType.String:
	caseCorElementType.Object:
	caseCorElementType.TypedByref:
	caseCorElementType.I:
	caseCorElementType.U:
	returntrue;
	caseCorElementType.Class:
	caseCorElementType.ValueType:
	caseCorElementType.Var:
	caseCorElementType.MVar:
	reader.ReadCompressedInteger();
	returntrue;
	caseCorElementType.Byref:
	caseCorElementType.Ptr:
	caseCorElementType.SzArray:
	returnTrySkipSignatureType(refreader);
	caseCorElementType.GenericInst:
	{
	if(!TrySkipSignatureType(refreader))
	returnfalse;
	intgenericArgCount=reader.ReadCompressedInteger();
	for(inti=0;i<genericArgCount;i++)
	{
	if(!TrySkipSignatureType(refreader))
	returnfalse;
	}
	returntrue;
	}
	caseCorElementType.Array:
	{
	if(!TrySkipSignatureType(refreader))
	returnfalse;
	intrank=reader.ReadCompressedInteger();
	intsizes=reader.ReadCompressedInteger();
	for(inti=0;i<sizes;i++)
	reader.ReadCompressedInteger();
	intlowerBounds=reader.ReadCompressedInteger();
	for(inti=0;i<lowerBounds;i++)
	reader.ReadCompressedInteger();
	returnrank>=0;
	}
	caseCorElementType.FnPtr:
	{
	SignatureHeaderheader=reader.ReadSignatureHeader();
	if(header.IsGeneric)
	reader.ReadCompressedInteger();
	intparameterCount=reader.ReadCompressedInteger();
	if(parameterCount<0)
	returnfalse;
	if(!TrySkipSignatureType(refreader))
	returnfalse;
	for(inti=0;i<parameterCount;i++)
	{
	if(!TrySkipSignatureType(refreader))
	returnfalse;
	}
	returntrue;
	}
	default:
	returnfalse;
	}
	}
	returnfalse;
	}

[Copilot]

Pull request overview

Copilot reviewed 64 out of 64 changed files in this pull request and generated 3 comments.

[Copilot]

CollectStackRefs currently appends every ref returned by ISOSStackRefEnum::Next with no upper bound. Later comparison helpers allocate fixed-size arrays sized by MAX_COLLECTED_REFS (e.g., matched[MAX_COLLECTED_REFS], aUsed[MAX_COLLECTED_REFS]), so if enumeration returns more than MAX_COLLECTED_REFS it can lead to out-of-bounds writes. Please cap collection to MAX_COLLECTED_REFS (and record an overflow/skip reason) or make the comparison logic handle arbitrarily large ref sets safely.

[Copilot]

When CollectRuntimeStackRefs overflows, the code logs a [SKIP] line but continues and still computes rtMatch/pass (and does not increment s_verifySkip). If CDACSTRESS_USE_DAC is not set, this can produce false failures based on a truncated runtime ref set. Consider treating runtime overflow as an actual skip (increment skip counter + return) when RT comparison is used for pass/fail, or otherwise avoid using rtMatch in the presence of overflow.

Suggested change

	if (rtOverflow && s_logFile != nullptr)
	{
	fprintf(s_logFile, "[SKIP] Thread=0x%x IP=0x%p - RT overflow (>%d refs)\n",
	osThreadId, (void*)GetIP(regs), MAX_COLLECTED_REFS);
	if (rtOverflow)
	{
	InterlockedIncrement(&s_verifySkip);
	if (s_logFile != nullptr)
	{
	fprintf(s_logFile, "[SKIP] Thread=0x%x IP=0x%p - RT overflow (>%d refs)\n",
	osThreadId, (void*)GetIP(regs), MAX_COLLECTED_REFS);
	}
	return;

[Copilot]

In ReportUntrackedAndSucceed, ReportSlot is called with reportScratchSlots: true, which forces reporting scratch registers/stack slots even when CodeManagerFlags.ActiveStackFrame is not set (reportScratchSlots local is false). This changes GC root enumeration behavior and can introduce extra roots for non-leaf frames. It looks like this should pass the reportScratchSlots variable instead of always true.

Suggested change

	ReportSlot(slotIndex,reportScratchSlots:true,reportFpBasedSlotsOnly,reportSlot);
	}
	ReportSlot(slotIndex,reportScratchSlots,reportFpBasedSlotsOnly,reportSlot);
	}

[Copilot]

Pull request overview

Copilot reviewed 64 out of 64 changed files in this pull request and generated 2 comments.

[Copilot]

CompareRefSetsFlat uses a fixed-size matched[MAX_COLLECTED_REFS] buffer, but cDAC/DAC ref collection (CollectStackRefs) is unbounded. Without a guard/cap, a large ref set (>4096) can cause out-of-bounds writes during matching. Consider capping cDAC/DAC collection to MAX_COLLECTED_REFS (and treating overflow as a [SKIP]) or using dynamically sized bookkeeping here.

[Copilot]

The async stdout/stderr collection uses stdout += ... / stderr += ... inside DataReceived event handlers. These callbacks can run concurrently, and string concatenation is not thread-safe; it can also be costly for large output. Consider buffering with a thread-safe collector (e.g., ConcurrentQueue<string> or StringBuilder with a lock) and call process.WaitForExit() (or await stream completion) after WaitForExit(timeout) to ensure all async output has been drained before asserting/logging.

[max-charlamb]

Closing in favor of stacked PR approach

[max-charlamb]

Summary

cDAC GC stress verification tool (DOTNET_CdacStress) that compares stack GC references between the cDAC and the runtime at allocation stress points. Includes stack walker fixes, GC reference scanning implementation, custom signature decoding, and calling convention argument iteration.

Note

This PR description was updated with AI assistance from Copilot.

Stack walker fixes

• Fix SW_SKIPPED_FRAME: do not call UpdateContextFromFrame (matches native SFITER_SKIPPED_FRAME_FUNCTION which does not call UpdateRegDisplay)
• Fix skipped-frame loop: check for more skipped frames before yielding managed method (prevents duplicate EnumGcRefs between consecutive skipped frames)
• Restructure Filter() to drive Next() directly, matching native Filter()+NextRaw() integration (prevents funclet-to-parent walk cycles)
• Remove SkipActiveICFOnce/SkipCurrentFrameInCheck — active ICF double-yield is natural and harmless
• Remove IsAtFirstPassExceptionThrowSite — native does not suppress first-pass refs
• Fix IsFirst not preserved for skipped frames (was causing IsActiveFrame=false for the topmost managed frame)

GC reference scanning

• Implement PromoteCallerStack for stub frames (GCRefMap + MetaSig paths)
• Implement SOSDacImpl.GetStackReferences using cDAC contract (was falling back to legacy DAC)
• Read FilterContext for stack walk starting context
• Three-way cDAC/DAC/RT comparison with InProcessDataTarget
• DOTNET_CdacStress bit flags: ALLOC/INSTR/REFS/WALK/USE_DAC/UNIQUE

RuntimeSignatureDecoder

Custom signature decoder that handles ELEMENT_TYPE_INTERNAL (0x21) and ELEMENT_TYPE_CMOD_INTERNAL (0x22) which SRM's SignatureDecoder cannot parse. These occur in IL stubs, marshalling stubs, and unsafe accessor frames.

• IRuntimeSignatureTypeProvider<TType, TGenericContext>: superset of SRM's ISignatureTypeProvider, adds GetInternalType and GetInternalModifiedType
• ISignatureReader + SpanSignatureReader: endianness-aware reader abstraction
• Correct ECMA-335 compliance: TypeDefOrRefOrSpecEncoded token decoding, sign-extension-by-width for compressed signed ints, bounds validation

ArgIterator (ported from crossgen2)

Proper calling convention analysis replacing the simplified 1-slot-per-param approach.

• CallingConventionInfo: hybrid data descriptor layout values + ABI invariant constants for all architectures (x86, x64 Windows/Unix, ARM32, ARM64, LoongArch64, RISC-V64)
• ArgIterator.GetNextOffset(): maps each argument to its actual register or stack offset
• OffsetOfFloatArgumentRegisters added to TransitionBlock data descriptor
• Handles multi-slot args, forced-byref params, return buffer placement, async continuation

CI infrastructure

• cDAC stress tests run in Helix via cdac-stress-helix.proj
• Extended runtime-diagnostics.yml CdacDumpTests buildArgs with +tools.cdacstresstests

Test results

Allocation-level stress (9 debuggees, ~46K verifications):

Debuggee	Verifications	Pass	Fail
BasicAlloc	4,936	4,936	0
DeepStack	4,964	4,964	0
Generics	4,936	4,936	0
MultiThread	4,996	4,996	0
Comprehensive	4,994	4,992	2*
ExceptionHandling	4,958	4,958	0
StructScenarios	4,940	4,940	0
DynamicMethods	6,520	6,520	0
PInvoke	4,936	4,936	0

Instruction-level stress (9 debuggees, ~226K verifications):
All 9 debuggees pass with zero failures across 226,452 verifications.

*Comprehensive's 2 failures are pre-existing legacy DAC issues (DAC returns 0 refs for background threads in kernel waits; cDAC and runtime agree).

[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

Adds a cDAC GC-stress verification harness and extends the cDAC stack-walk / stack-GC-ref pipeline so cDAC stack reference enumeration can be compared against runtime scanning at stress points.

Changes:

• Implements/extends cDAC stack reference enumeration (including Frame-based scanning paths like PromoteCallerStack via GCRefMap / MetaSig) and wires SOS GetStackReferences to the cDAC contract.
• Introduces a new GC stress integration test project with debuggee apps and orchestration targets.
• Extends CoreCLR cDAC stress/GC stress plumbing and data descriptors to support the new stack-walk and frame-scanning capabilities.

Reviewed changes

Copilot reviewed 46 out of 46 changed files in this pull request and generated 12 comments.

Show a summary per file

File	Description
src/native/managed/cdac/tests/MockDescriptors/MockDescriptors.ExecutionManager.cs	Updates mock type layout for execution manager-related data.
src/native/managed/cdac/tests/MockDescriptors/MockDescriptors.cs	Updates mock type layouts (ExceptionInfo/Thread) for contract tests.
src/native/managed/cdac/tests/Microsoft.Diagnostics.DataContractReader.Tests.csproj	Excludes new GCStressTests folder from the existing unit test project compilation.
src/native/managed/cdac/tests/GCStressTests/README.md	Documents how to build/run the new GC stress tests.
src/native/managed/cdac/tests/GCStressTests/Microsoft.Diagnostics.DataContractReader.GCStressTests.csproj	Adds a dedicated GC stress test project.
src/native/managed/cdac/tests/GCStressTests/GCStressTests.targets	MSBuild orchestration to discover/build debuggee projects.
src/native/managed/cdac/tests/GCStressTests/GCStressTestBase.cs	Test harness to run debuggees under corerun and parse verification logs.
src/native/managed/cdac/tests/GCStressTests/GCStressResults.cs	Parses the native verification log into structured pass/fail/skip counts.
src/native/managed/cdac/tests/GCStressTests/Debuggees/PInvoke/Program.cs	Adds a P/Invoke-focused debuggee workload.
src/native/managed/cdac/tests/GCStressTests/Debuggees/PInvoke/PInvoke.csproj	Debuggee project file for PInvoke scenario.
src/native/managed/cdac/tests/GCStressTests/Debuggees/MultiThread/Program.cs	Adds a multi-threaded debuggee workload.
src/native/managed/cdac/tests/GCStressTests/Debuggees/MultiThread/MultiThread.csproj	Debuggee project file for MultiThread scenario.
src/native/managed/cdac/tests/GCStressTests/Debuggees/Generics/Program.cs	Adds a generics/interface/delegate debuggee workload.
src/native/managed/cdac/tests/GCStressTests/Debuggees/Generics/Generics.csproj	Debuggee project file for Generics scenario.
src/native/managed/cdac/tests/GCStressTests/Debuggees/ExceptionHandling/Program.cs	Adds an exception-handling/funclet debuggee workload.
src/native/managed/cdac/tests/GCStressTests/Debuggees/ExceptionHandling/ExceptionHandling.csproj	Debuggee project file for ExceptionHandling scenario.
src/native/managed/cdac/tests/GCStressTests/Debuggees/Directory.Build.props	Shared build props for debuggee projects (output layout, TFM, etc.).
src/native/managed/cdac/tests/GCStressTests/Debuggees/DeepStack/Program.cs	Adds deep-recursion debuggee workload.
src/native/managed/cdac/tests/GCStressTests/Debuggees/DeepStack/DeepStack.csproj	Debuggee project file for DeepStack scenario.
src/native/managed/cdac/tests/GCStressTests/Debuggees/Comprehensive/Program.cs	Adds comprehensive “all scenarios” debuggee workload.
src/native/managed/cdac/tests/GCStressTests/Debuggees/Comprehensive/Comprehensive.csproj	Debuggee project file for Comprehensive scenario.
src/native/managed/cdac/tests/GCStressTests/Debuggees/BasicAlloc/Program.cs	Adds basic allocation/live-ref debuggee workload.
src/native/managed/cdac/tests/GCStressTests/Debuggees/BasicAlloc/BasicAlloc.csproj	Debuggee project file for BasicAlloc scenario.
src/native/managed/cdac/tests/GCStressTests/BasicGCStressTests.cs	Theory-based test suite that runs the debuggees and asserts pass rate.
src/native/managed/cdac/tests/gcstress/known-issues.md	Captures known mismatch classes and limitations.
src/native/managed/cdac/Microsoft.Diagnostics.DataContractReader.Legacy/SOSDacImpl.cs	Implements GetStackReferences using the cDAC contract rather than legacy DAC fallback.
src/native/managed/cdac/Microsoft.Diagnostics.DataContractReader.Contracts/Data/Frames/StubDispatchFrame.cs	Extends StubDispatchFrame data with GCRefMap pointer.
src/native/managed/cdac/Microsoft.Diagnostics.DataContractReader.Contracts/Data/Frames/ExternalMethodFrame.cs	Adds ExternalMethodFrame contract data type (GCRefMap).
src/native/managed/cdac/Microsoft.Diagnostics.DataContractReader.Contracts/Data/Frames/DynamicHelperFrame.cs	Adds DynamicHelperFrame contract data type (flags).
src/native/managed/cdac/Microsoft.Diagnostics.DataContractReader.Contracts/Data/ExceptionInfo.cs	Adds clause-range fields used for catch-handler resumption offset selection.
src/native/managed/cdac/Microsoft.Diagnostics.DataContractReader.Contracts/Contracts/StackWalk/StackWalk_1.cs	Refactors stack-walk filtering and adds Frame-based GC root scanning.
src/native/managed/cdac/Microsoft.Diagnostics.DataContractReader.Contracts/Contracts/StackWalk/GC/GcScanner.cs	Adds optional relOffset override support for GC ref enumeration.
src/native/managed/cdac/Microsoft.Diagnostics.DataContractReader.Contracts/Contracts/StackWalk/GC/GCRefMapDecoder.cs	Implements GCRefMap decoding for transition-block scanning.
src/native/managed/cdac/Microsoft.Diagnostics.DataContractReader.Contracts/Contracts/StackWalk/GC/CorSigParser.cs	Adds minimal signature parsing to classify parameters for MetaSig-based scanning.
src/native/managed/cdac/Microsoft.Diagnostics.DataContractReader.Contracts/Contracts/GCInfo/IGCInfoDecoder.cs	Adds FindFirstInterruptiblePoint API to GCInfo decoders.
src/native/managed/cdac/Microsoft.Diagnostics.DataContractReader.Contracts/Contracts/GCInfo/GCInfoDecoder.cs	Implements FindFirstInterruptiblePoint using decoded interruptible ranges.
src/native/managed/cdac/Microsoft.Diagnostics.DataContractReader.Contracts/Contracts/ExecutionManager/ExecutionManagerCore.EEJitManager.cs	Fixes code-start lookup for exception clause enumeration and adds minor flow adjustments.
src/native/managed/cdac/Microsoft.Diagnostics.DataContractReader.Contracts/Constants.cs	Adds TransitionBlock-related global names used by frame scanning.
src/native/managed/cdac/Microsoft.Diagnostics.DataContractReader.Abstractions/DataType.cs	Adds new DataType enum values for new frame contracts.
src/native/managed/cdac/cdac.slnx	Adds the new GC stress test project to the cDAC solution.
src/coreclr/vm/gccover.cpp	Adds step-based skipping to reduce overhead when throttling verification.
src/coreclr/vm/frames.h	Exposes additional frame fields to cDAC via cdac_data<> descriptors.
src/coreclr/vm/datadescriptor/datadescriptor.inc	Adds new contract fields/globals for frames and TransitionBlock layout.
src/coreclr/vm/cdacstress.cpp	Updates in-process cDAC/DAC verification logic, logging, and step behavior.
eng/Subsets.props	Adds an on-demand subset for running GC stress tests.
docs/design/datacontracts/StackWalk.md	Documents the new frame fields and TransitionBlock globals in the StackWalk contract.

Comments suppressed due to low confidence (1)

src/native/managed/cdac/tests/MockDescriptors/MockDescriptors.cs:228

• ThreadFields contains ProfilerFilterContext twice (lines 224 and 228). Duplicate field entries will skew offsets and make the mock descriptors inconsistent with the real data descriptor. Keep a single ProfilerFilterContext entry in the correct order.

[Copilot]

Pull request overview

Copilot reviewed 50 out of 50 changed files in this pull request and generated 7 comments.

Comments suppressed due to low confidence (2)

src/native/managed/cdac/tests/MockDescriptors/MockDescriptors.cs:229

• ThreadFields now includes DebuggerFilterContext/ProfilerFilterContext twice (duplicate entries at the end of the list). This can cause ambiguous/incorrect field offsets in the mock type layout. Keep each field only once.
  src/coreclr/vm/cdacstress.cpp:963
• CompareRefSets uses a fixed-size matched[MAX_COLLECTED_REFS] buffer but no longer validates that countA/countB are <= MAX_COLLECTED_REFS. Since CollectStackRefs appends without a hard cap, this can lead to out-of-bounds access when countA or countB exceeds 4096. Reintroduce a guard or allocate the match state sized to the counts.

 return true;
bool matched[MAX_COLLECTED_REFS] = {};
for (int i = 0; i < countA; i++)

[Copilot]

Pull request overview

Copilot reviewed 53 out of 53 changed files in this pull request and generated 7 comments.

[max-charlamb]

feedback for local copilot

[Copilot]

Pull request overview

Copilot reviewed 51 out of 51 changed files in this pull request and generated 9 comments.

Comments suppressed due to low confidence (1)

src/coreclr/vm/cdacstress.cpp:625

• CollectStackRefs appends to pRefs without any cap. Later comparisons allocate fixed-size arrays sized MAX_COLLECTED_REFS (e.g., matched[MAX_COLLECTED_REFS], aUsed/bUsed) and assume counts fit. If the DAC returns > MAX_COLLECTED_REFS refs, this will lead to out-of-bounds writes/reads. Add an explicit limit/overflow handling in CollectStackRefs (stop at MAX_COLLECTED_REFS and mark overflow / SKIP), or change the later comparison logic to handle arbitrary counts safely.

 SOSStackRefData refData;
unsigned int fetched = 0;
while (true)
{
hr = pEnum->Next(1, &refData, &fetched);
if (FAILED(hr) || fetched == 0)
break;
StackRef ref;
ref.Address = refData.Address;
ref.Object = refData.Object;
ref.Flags = refData.Flags;
ref.Source = refData.Source;
ref.SourceType = refData.SourceType;
ref.Register = refData.Register;
ref.Offset = refData.Offset;
ref.StackPointer = refData.StackPointer;
pRefs->Append(ref);
}

[Copilot]

Pull request overview

Copilot reviewed 52 out of 52 changed files in this pull request and generated 3 comments.

[Copilot]

Pull request overview

Copilot reviewed 65 out of 65 changed files in this pull request and generated 5 comments.

[Copilot]

ThreadStore now calls target.GetTypeInfo(DataType.SLink) and reads field "Next", but the CoreCLR data descriptor in this PR doesn't define a SLink type/field (could not find any CDAC_TYPE_BEGIN/FIELD for SLink). This will cause GetTypeInfo(DataType.SLink) to fail at runtime. Either add an SLink descriptor (with a "Next" pointer) on the runtime side or avoid needing type info here (e.g., treat the embedded SLink as a single pointer at offset 0).

[Copilot]

DataType adds SLink, but there is no corresponding runtime data descriptor definition (no CDAC_TYPE_BEGIN/FIELD(SLink, ...) found). Any attempt to read this type via Target.GetTypeInfo(DataType.SLink) will fail. Either add the runtime descriptor entry for SLink or remove this enum value and read the embedded link without type metadata.

Suggested change

	SLink,
	ThreadLocalData,
	ThreadLocalData=17,

[Copilot]

These DataReceived event handlers append to strings via "+=" from background threads. That is not thread-safe and can lead to lost/garbled output under contention. Consider using a StringBuilder with locking (or ConcurrentQueue) for stderr/stdout aggregation.

[Copilot]

Same thread-safety issue as stderr: appending to stdout with += from OutputDataReceived is racy. Use a synchronized StringBuilder/collector to avoid missing output and to keep logs reliable when tests fail.

[Copilot]

GcTypeKindToArgTypeInfo collapses all non-GC-ref types to CorElementType.I/pointer-sized and doesn't preserve float vs integer vs exact primitive sizes. ArgIterator offset calculation depends on the real signature shape (especially on Unix x64/ARM64 where float regs and argument sizing affect later argument placement), so this can produce incorrect offsets and cause missed/incorrect GC ref reporting for signatures with floats or non-pointer-sized primitives. Consider decoding into a richer type representation (e.g., CorElementType + size) and building ArgTypeInfo from that rather than from GcTypeKind.

Suggested change

	/// Converts a <see cref="GcTypeKind"/> to a minimal <see cref="ArgTypeInfo"/> for
	/// ArgIterator consumption. This is a bridge until we have a full type-aware provider.
	/// </summary>
	privatestaticArgTypeInfoGcTypeKindToArgTypeInfo(GcTypeKindkind,intpointerSize)
	{
	returnkindswitch
	{
	GcTypeKind.None=> ArgTypeInfo.ForPrimitive(CorElementType.I,pointerSize),
	GcTypeKind.Ref=> ArgTypeInfo.ForPrimitive(CorElementType.Class,pointerSize),
	GcTypeKind.Interior=> ArgTypeInfo.ForPrimitive(CorElementType.Byref,pointerSize),
	GcTypeKind.Other=>new ArgTypeInfo
	{
	CorElementType=CorElementType.ValueType,
	Size=pointerSize,// Conservative: assume pointer-sized for now
	IsValueType=true,
	},
	_ =>ArgTypeInfo.ForPrimitive(CorElementType.I,pointerSize),
	};
	}
	/// Converts a <see cref="GcTypeKind"/> to a conservative fallback <see cref="ArgTypeInfo"/>
	/// when an exact signature cannot be decoded. Callers should prefer decoding the real
	/// managed signature via <see cref="TryGetArgTypeInfosFromMethodSignature(ReadOnlySpan{byte}, int, out ArgTypeInfo[])"/>
	/// so that floating-point and non-pointer-sized primitives retain their real layout.
	/// </summary>
	privatestaticArgTypeInfoGcTypeKindToArgTypeInfo(GcTypeKindkind,intpointerSize)
	{
	returnkindswitch
	{
	GcTypeKind.Ref=> ArgTypeInfo.ForPrimitive(CorElementType.Class,pointerSize),
	GcTypeKind.Interior=> ArgTypeInfo.ForPrimitive(CorElementType.Byref,pointerSize),
	GcTypeKind.Other=>new ArgTypeInfo
	{
	CorElementType=CorElementType.ValueType,
	Size=pointerSize,
	IsValueType=true,
	},
	_ =>ArgTypeInfo.ForPrimitive(CorElementType.I,pointerSize),
	};
	}
	privatestaticboolTryGetArgTypeInfosFromMethodSignature(ReadOnlySpan<byte>signatureBytes,intpointerSize,outArgTypeInfo[]argTypes)
	{
	argTypes=Array.Empty<ArgTypeInfo>();
	if(signatureBytes.IsEmpty)
	returnfalse;
	BlobReaderreader=new(signatureBytes);
	SignatureHeaderheader=reader.ReadSignatureHeader();
	if(header.IsGeneric)
	reader.ReadCompressedInteger();
	intparameterCount=reader.ReadCompressedInteger();
	if(parameterCount<0)
	returnfalse;
	if(!TrySkipSignatureType(refreader))
	returnfalse;
	ArgTypeInfo[]decodedArgTypes=newArgTypeInfo[parameterCount];
	for(inti=0;i<parameterCount;i++)
	{
	if(!TryReadArgTypeInfo(refreader,pointerSize,outdecodedArgTypes[i]))
	returnfalse;
	}
	argTypes=decodedArgTypes;
	returntrue;
	}
	privatestaticboolTryReadArgTypeInfo(refBlobReaderreader,intpointerSize,outArgTypeInfoargTypeInfo)
	{
	while(reader.RemainingBytes>0)
	{
	CorElementTypeelementType=(CorElementType)reader.ReadByte();
	switch(elementType)
	{
	caseCorElementType.CmodOpt:
	caseCorElementType.CmodReqd:
	reader.ReadCompressedInteger();
	continue;
	caseCorElementType.Pinned:
	continue;
	caseCorElementType.Boolean:
	caseCorElementType.I1:
	caseCorElementType.U1:
	argTypeInfo=ArgTypeInfo.ForPrimitive(elementType,1);
	returntrue;
	caseCorElementType.Char:
	caseCorElementType.I2:
	caseCorElementType.U2:
	argTypeInfo=ArgTypeInfo.ForPrimitive(elementType,2);
	returntrue;
	caseCorElementType.I4:
	caseCorElementType.U4:
	caseCorElementType.R4:
	argTypeInfo=ArgTypeInfo.ForPrimitive(elementType,4);
	returntrue;
	caseCorElementType.I8:
	caseCorElementType.U8:
	caseCorElementType.R8:
	argTypeInfo=ArgTypeInfo.ForPrimitive(elementType,8);
	returntrue;
	caseCorElementType.I:
	caseCorElementType.U:
	caseCorElementType.Ptr:
	caseCorElementType.FnPtr:
	if(elementTypeisCorElementType.Ptr or CorElementType.FnPtr)
	{
	if(!TrySkipSignatureType(refreader))
	{
	argTypeInfo=default;
	returnfalse;
	}
	}
	argTypeInfo=ArgTypeInfo.ForPrimitive(CorElementType.I,pointerSize);
	returntrue;
	caseCorElementType.Byref:
	if(!TrySkipSignatureType(refreader))
	{
	argTypeInfo=default;
	returnfalse;
	}
	argTypeInfo=ArgTypeInfo.ForPrimitive(CorElementType.Byref,pointerSize);
	returntrue;
	caseCorElementType.String:
	caseCorElementType.Object:
	caseCorElementType.Class:
	if(elementTypeisCorElementType.Class)
	reader.ReadCompressedInteger();
	argTypeInfo=ArgTypeInfo.ForPrimitive(CorElementType.Class,pointerSize);
	returntrue;
	caseCorElementType.SzArray:
	if(!TrySkipSignatureType(refreader))
	{
	argTypeInfo=default;
	returnfalse;
	}
	argTypeInfo=ArgTypeInfo.ForPrimitive(CorElementType.Class,pointerSize);
	returntrue;
	caseCorElementType.ValueType:
	reader.ReadCompressedInteger();
	argTypeInfo=newArgTypeInfo
	{
	CorElementType=CorElementType.ValueType,
	Size=pointerSize,
	IsValueType=true,
	};
	returntrue;
	caseCorElementType.GenericInst:
	{
	if(!TrySkipSignatureType(refreader))
	{
	argTypeInfo=default;
	returnfalse;
	}
	intgenericArgCount=reader.ReadCompressedInteger();
	for(inti=0;i<genericArgCount;i++)
	{
	if(!TrySkipSignatureType(refreader))
	{
	argTypeInfo=default;
	returnfalse;
	}
	}
	argTypeInfo=newArgTypeInfo
	{
	CorElementType=CorElementType.ValueType,
	Size=pointerSize,
	IsValueType=true,
	};
	returntrue;
	}
	default:
	argTypeInfo=default;
	returnfalse;
	}
	}
	argTypeInfo=default;
	returnfalse;
	}
	privatestaticboolTrySkipSignatureType(refBlobReaderreader)
	{
	while(reader.RemainingBytes>0)
	{
	CorElementTypeelementType=(CorElementType)reader.ReadByte();
	switch(elementType)
	{
	caseCorElementType.CmodOpt:
	caseCorElementType.CmodReqd:
	reader.ReadCompressedInteger();
	continue;
	caseCorElementType.Pinned:
	continue;
	caseCorElementType.Void:
	caseCorElementType.Boolean:
	caseCorElementType.Char:
	caseCorElementType.I1:
	caseCorElementType.U1:
	caseCorElementType.I2:
	caseCorElementType.U2:
	caseCorElementType.I4:
	caseCorElementType.U4:
	caseCorElementType.I8:
	caseCorElementType.U8:
	caseCorElementType.R4:
	caseCorElementType.R8:
	caseCorElementType.String:
	caseCorElementType.Object:
	caseCorElementType.TypedByref:
	caseCorElementType.I:
	caseCorElementType.U:
	returntrue;
	caseCorElementType.Class:
	caseCorElementType.ValueType:
	caseCorElementType.Var:
	caseCorElementType.MVar:
	reader.ReadCompressedInteger();
	returntrue;
	caseCorElementType.Byref:
	caseCorElementType.Ptr:
	caseCorElementType.SzArray:
	returnTrySkipSignatureType(refreader);
	caseCorElementType.GenericInst:
	{
	if(!TrySkipSignatureType(refreader))
	returnfalse;
	intgenericArgCount=reader.ReadCompressedInteger();
	for(inti=0;i<genericArgCount;i++)
	{
	if(!TrySkipSignatureType(refreader))
	returnfalse;
	}
	returntrue;
	}
	caseCorElementType.Array:
	{
	if(!TrySkipSignatureType(refreader))
	returnfalse;
	intrank=reader.ReadCompressedInteger();
	intsizes=reader.ReadCompressedInteger();
	for(inti=0;i<sizes;i++)
	reader.ReadCompressedInteger();
	intlowerBounds=reader.ReadCompressedInteger();
	for(inti=0;i<lowerBounds;i++)
	reader.ReadCompressedInteger();
	returnrank>=0;
	}
	caseCorElementType.FnPtr:
	{
	SignatureHeaderheader=reader.ReadSignatureHeader();
	if(header.IsGeneric)
	reader.ReadCompressedInteger();
	intparameterCount=reader.ReadCompressedInteger();
	if(parameterCount<0)
	returnfalse;
	if(!TrySkipSignatureType(refreader))
	returnfalse;
	for(inti=0;i<parameterCount;i++)
	{
	if(!TrySkipSignatureType(refreader))
	returnfalse;
	}
	returntrue;
	}
	default:
	returnfalse;
	}
	}
	returnfalse;
	}

[Copilot]

Pull request overview

Copilot reviewed 64 out of 64 changed files in this pull request and generated 3 comments.

[Copilot]

CollectStackRefs currently appends every ref returned by ISOSStackRefEnum::Next with no upper bound. Later comparison helpers allocate fixed-size arrays sized by MAX_COLLECTED_REFS (e.g., matched[MAX_COLLECTED_REFS], aUsed[MAX_COLLECTED_REFS]), so if enumeration returns more than MAX_COLLECTED_REFS it can lead to out-of-bounds writes. Please cap collection to MAX_COLLECTED_REFS (and record an overflow/skip reason) or make the comparison logic handle arbitrarily large ref sets safely.

[Copilot]

When CollectRuntimeStackRefs overflows, the code logs a [SKIP] line but continues and still computes rtMatch/pass (and does not increment s_verifySkip). If CDACSTRESS_USE_DAC is not set, this can produce false failures based on a truncated runtime ref set. Consider treating runtime overflow as an actual skip (increment skip counter + return) when RT comparison is used for pass/fail, or otherwise avoid using rtMatch in the presence of overflow.

Suggested change

	if (rtOverflow && s_logFile != nullptr)
	{
	fprintf(s_logFile, "[SKIP] Thread=0x%x IP=0x%p - RT overflow (>%d refs)\n",
	osThreadId, (void*)GetIP(regs), MAX_COLLECTED_REFS);
	if (rtOverflow)
	{
	InterlockedIncrement(&s_verifySkip);
	if (s_logFile != nullptr)
	{
	fprintf(s_logFile, "[SKIP] Thread=0x%x IP=0x%p - RT overflow (>%d refs)\n",
	osThreadId, (void*)GetIP(regs), MAX_COLLECTED_REFS);
	}
	return;

[Copilot]

In ReportUntrackedAndSucceed, ReportSlot is called with reportScratchSlots: true, which forces reporting scratch registers/stack slots even when CodeManagerFlags.ActiveStackFrame is not set (reportScratchSlots local is false). This changes GC root enumeration behavior and can introduce extra roots for non-leaf frames. It looks like this should pass the reportScratchSlots variable instead of always true.

Suggested change

	ReportSlot(slotIndex,reportScratchSlots:true,reportFpBasedSlotsOnly,reportSlot);
	}
	ReportSlot(slotIndex,reportScratchSlots,reportFpBasedSlotsOnly,reportSlot);
	}

[Copilot]

Pull request overview

Copilot reviewed 64 out of 64 changed files in this pull request and generated 2 comments.

[Copilot]

CompareRefSetsFlat uses a fixed-size matched[MAX_COLLECTED_REFS] buffer, but cDAC/DAC ref collection (CollectStackRefs) is unbounded. Without a guard/cap, a large ref set (>4096) can cause out-of-bounds writes during matching. Consider capping cDAC/DAC collection to MAX_COLLECTED_REFS (and treating overflow as a [SKIP]) or using dynamically sized bookkeeping here.

[Copilot]

The async stdout/stderr collection uses stdout += ... / stderr += ... inside DataReceived event handlers. These callbacks can run concurrently, and string concatenation is not thread-safe; it can also be costly for large output. Consider buffering with a thread-safe collector (e.g., ConcurrentQueue<string> or StringBuilder with a lock) and call process.WaitForExit() (or await stream completion) after WaitForExit(timeout) to ensure all async output has been drained before asserting/logging.

[max-charlamb]

Closing in favor of stacked PR approach

[max-charlamb]

Summary

cDAC GC stress verification tool (DOTNET_CdacStress) that compares stack GC references between the cDAC and the runtime at allocation stress points. Includes stack walker fixes, GC reference scanning implementation, custom signature decoding, and calling convention argument iteration.

Note

This PR description was updated with AI assistance from Copilot.

Stack walker fixes

• Fix SW_SKIPPED_FRAME: do not call UpdateContextFromFrame (matches native SFITER_SKIPPED_FRAME_FUNCTION which does not call UpdateRegDisplay)
• Fix skipped-frame loop: check for more skipped frames before yielding managed method (prevents duplicate EnumGcRefs between consecutive skipped frames)
• Restructure Filter() to drive Next() directly, matching native Filter()+NextRaw() integration (prevents funclet-to-parent walk cycles)
• Remove SkipActiveICFOnce/SkipCurrentFrameInCheck — active ICF double-yield is natural and harmless
• Remove IsAtFirstPassExceptionThrowSite — native does not suppress first-pass refs
• Fix IsFirst not preserved for skipped frames (was causing IsActiveFrame=false for the topmost managed frame)

GC reference scanning

• Implement PromoteCallerStack for stub frames (GCRefMap + MetaSig paths)
• Implement SOSDacImpl.GetStackReferences using cDAC contract (was falling back to legacy DAC)
• Read FilterContext for stack walk starting context
• Three-way cDAC/DAC/RT comparison with InProcessDataTarget
• DOTNET_CdacStress bit flags: ALLOC/INSTR/REFS/WALK/USE_DAC/UNIQUE

RuntimeSignatureDecoder

Custom signature decoder that handles ELEMENT_TYPE_INTERNAL (0x21) and ELEMENT_TYPE_CMOD_INTERNAL (0x22) which SRM's SignatureDecoder cannot parse. These occur in IL stubs, marshalling stubs, and unsafe accessor frames.

• IRuntimeSignatureTypeProvider<TType, TGenericContext>: superset of SRM's ISignatureTypeProvider, adds GetInternalType and GetInternalModifiedType
• ISignatureReader + SpanSignatureReader: endianness-aware reader abstraction
• Correct ECMA-335 compliance: TypeDefOrRefOrSpecEncoded token decoding, sign-extension-by-width for compressed signed ints, bounds validation

ArgIterator (ported from crossgen2)

Proper calling convention analysis replacing the simplified 1-slot-per-param approach.

• CallingConventionInfo: hybrid data descriptor layout values + ABI invariant constants for all architectures (x86, x64 Windows/Unix, ARM32, ARM64, LoongArch64, RISC-V64)
• ArgIterator.GetNextOffset(): maps each argument to its actual register or stack offset
• OffsetOfFloatArgumentRegisters added to TransitionBlock data descriptor
• Handles multi-slot args, forced-byref params, return buffer placement, async continuation

CI infrastructure

• cDAC stress tests run in Helix via cdac-stress-helix.proj
• Extended runtime-diagnostics.yml CdacDumpTests buildArgs with +tools.cdacstresstests

Test results

Allocation-level stress (9 debuggees, ~46K verifications):

Debuggee	Verifications	Pass	Fail
BasicAlloc	4,936	4,936	0
DeepStack	4,964	4,964	0
Generics	4,936	4,936	0
MultiThread	4,996	4,996	0
Comprehensive	4,994	4,992	2*
ExceptionHandling	4,958	4,958	0
StructScenarios	4,940	4,940	0
DynamicMethods	6,520	6,520	0
PInvoke	4,936	4,936	0

Instruction-level stress (9 debuggees, ~226K verifications):
All 9 debuggees pass with zero failures across 226,452 verifications.

*Comprehensive's 2 failures are pre-existing legacy DAC issues (DAC returns 0 refs for background threads in kernel waits; cDAC and runtime agree).

[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

Adds a cDAC GC-stress verification harness and extends the cDAC stack-walk / stack-GC-ref pipeline so cDAC stack reference enumeration can be compared against runtime scanning at stress points.

Changes:

• Implements/extends cDAC stack reference enumeration (including Frame-based scanning paths like PromoteCallerStack via GCRefMap / MetaSig) and wires SOS GetStackReferences to the cDAC contract.
• Introduces a new GC stress integration test project with debuggee apps and orchestration targets.
• Extends CoreCLR cDAC stress/GC stress plumbing and data descriptors to support the new stack-walk and frame-scanning capabilities.

Reviewed changes

Copilot reviewed 46 out of 46 changed files in this pull request and generated 12 comments.

Show a summary per file

File	Description
src/native/managed/cdac/tests/MockDescriptors/MockDescriptors.ExecutionManager.cs	Updates mock type layout for execution manager-related data.
src/native/managed/cdac/tests/MockDescriptors/MockDescriptors.cs	Updates mock type layouts (ExceptionInfo/Thread) for contract tests.
src/native/managed/cdac/tests/Microsoft.Diagnostics.DataContractReader.Tests.csproj	Excludes new GCStressTests folder from the existing unit test project compilation.
src/native/managed/cdac/tests/GCStressTests/README.md	Documents how to build/run the new GC stress tests.
src/native/managed/cdac/tests/GCStressTests/Microsoft.Diagnostics.DataContractReader.GCStressTests.csproj	Adds a dedicated GC stress test project.
src/native/managed/cdac/tests/GCStressTests/GCStressTests.targets	MSBuild orchestration to discover/build debuggee projects.
src/native/managed/cdac/tests/GCStressTests/GCStressTestBase.cs	Test harness to run debuggees under corerun and parse verification logs.
src/native/managed/cdac/tests/GCStressTests/GCStressResults.cs	Parses the native verification log into structured pass/fail/skip counts.
src/native/managed/cdac/tests/GCStressTests/Debuggees/PInvoke/Program.cs	Adds a P/Invoke-focused debuggee workload.
src/native/managed/cdac/tests/GCStressTests/Debuggees/PInvoke/PInvoke.csproj	Debuggee project file for PInvoke scenario.
src/native/managed/cdac/tests/GCStressTests/Debuggees/MultiThread/Program.cs	Adds a multi-threaded debuggee workload.
src/native/managed/cdac/tests/GCStressTests/Debuggees/MultiThread/MultiThread.csproj	Debuggee project file for MultiThread scenario.
src/native/managed/cdac/tests/GCStressTests/Debuggees/Generics/Program.cs	Adds a generics/interface/delegate debuggee workload.
src/native/managed/cdac/tests/GCStressTests/Debuggees/Generics/Generics.csproj	Debuggee project file for Generics scenario.
src/native/managed/cdac/tests/GCStressTests/Debuggees/ExceptionHandling/Program.cs	Adds an exception-handling/funclet debuggee workload.
src/native/managed/cdac/tests/GCStressTests/Debuggees/ExceptionHandling/ExceptionHandling.csproj	Debuggee project file for ExceptionHandling scenario.
src/native/managed/cdac/tests/GCStressTests/Debuggees/Directory.Build.props	Shared build props for debuggee projects (output layout, TFM, etc.).
src/native/managed/cdac/tests/GCStressTests/Debuggees/DeepStack/Program.cs	Adds deep-recursion debuggee workload.
src/native/managed/cdac/tests/GCStressTests/Debuggees/DeepStack/DeepStack.csproj	Debuggee project file for DeepStack scenario.
src/native/managed/cdac/tests/GCStressTests/Debuggees/Comprehensive/Program.cs	Adds comprehensive “all scenarios” debuggee workload.
src/native/managed/cdac/tests/GCStressTests/Debuggees/Comprehensive/Comprehensive.csproj	Debuggee project file for Comprehensive scenario.
src/native/managed/cdac/tests/GCStressTests/Debuggees/BasicAlloc/Program.cs	Adds basic allocation/live-ref debuggee workload.
src/native/managed/cdac/tests/GCStressTests/Debuggees/BasicAlloc/BasicAlloc.csproj	Debuggee project file for BasicAlloc scenario.
src/native/managed/cdac/tests/GCStressTests/BasicGCStressTests.cs	Theory-based test suite that runs the debuggees and asserts pass rate.
src/native/managed/cdac/tests/gcstress/known-issues.md	Captures known mismatch classes and limitations.
src/native/managed/cdac/Microsoft.Diagnostics.DataContractReader.Legacy/SOSDacImpl.cs	Implements GetStackReferences using the cDAC contract rather than legacy DAC fallback.
src/native/managed/cdac/Microsoft.Diagnostics.DataContractReader.Contracts/Data/Frames/StubDispatchFrame.cs	Extends StubDispatchFrame data with GCRefMap pointer.
src/native/managed/cdac/Microsoft.Diagnostics.DataContractReader.Contracts/Data/Frames/ExternalMethodFrame.cs	Adds ExternalMethodFrame contract data type (GCRefMap).
src/native/managed/cdac/Microsoft.Diagnostics.DataContractReader.Contracts/Data/Frames/DynamicHelperFrame.cs	Adds DynamicHelperFrame contract data type (flags).
src/native/managed/cdac/Microsoft.Diagnostics.DataContractReader.Contracts/Data/ExceptionInfo.cs	Adds clause-range fields used for catch-handler resumption offset selection.
src/native/managed/cdac/Microsoft.Diagnostics.DataContractReader.Contracts/Contracts/StackWalk/StackWalk_1.cs	Refactors stack-walk filtering and adds Frame-based GC root scanning.
src/native/managed/cdac/Microsoft.Diagnostics.DataContractReader.Contracts/Contracts/StackWalk/GC/GcScanner.cs	Adds optional relOffset override support for GC ref enumeration.
src/native/managed/cdac/Microsoft.Diagnostics.DataContractReader.Contracts/Contracts/StackWalk/GC/GCRefMapDecoder.cs	Implements GCRefMap decoding for transition-block scanning.
src/native/managed/cdac/Microsoft.Diagnostics.DataContractReader.Contracts/Contracts/StackWalk/GC/CorSigParser.cs	Adds minimal signature parsing to classify parameters for MetaSig-based scanning.
src/native/managed/cdac/Microsoft.Diagnostics.DataContractReader.Contracts/Contracts/GCInfo/IGCInfoDecoder.cs	Adds FindFirstInterruptiblePoint API to GCInfo decoders.
src/native/managed/cdac/Microsoft.Diagnostics.DataContractReader.Contracts/Contracts/GCInfo/GCInfoDecoder.cs	Implements FindFirstInterruptiblePoint using decoded interruptible ranges.
src/native/managed/cdac/Microsoft.Diagnostics.DataContractReader.Contracts/Contracts/ExecutionManager/ExecutionManagerCore.EEJitManager.cs	Fixes code-start lookup for exception clause enumeration and adds minor flow adjustments.
src/native/managed/cdac/Microsoft.Diagnostics.DataContractReader.Contracts/Constants.cs	Adds TransitionBlock-related global names used by frame scanning.
src/native/managed/cdac/Microsoft.Diagnostics.DataContractReader.Abstractions/DataType.cs	Adds new DataType enum values for new frame contracts.
src/native/managed/cdac/cdac.slnx	Adds the new GC stress test project to the cDAC solution.
src/coreclr/vm/gccover.cpp	Adds step-based skipping to reduce overhead when throttling verification.
src/coreclr/vm/frames.h	Exposes additional frame fields to cDAC via cdac_data<> descriptors.
src/coreclr/vm/datadescriptor/datadescriptor.inc	Adds new contract fields/globals for frames and TransitionBlock layout.
src/coreclr/vm/cdacstress.cpp	Updates in-process cDAC/DAC verification logic, logging, and step behavior.
eng/Subsets.props	Adds an on-demand subset for running GC stress tests.
docs/design/datacontracts/StackWalk.md	Documents the new frame fields and TransitionBlock globals in the StackWalk contract.

Comments suppressed due to low confidence (1)

src/native/managed/cdac/tests/MockDescriptors/MockDescriptors.cs:228

• ThreadFields contains ProfilerFilterContext twice (lines 224 and 228). Duplicate field entries will skew offsets and make the mock descriptors inconsistent with the real data descriptor. Keep a single ProfilerFilterContext entry in the correct order.

[Copilot]

Pull request overview

Copilot reviewed 50 out of 50 changed files in this pull request and generated 7 comments.

Comments suppressed due to low confidence (2)

src/native/managed/cdac/tests/MockDescriptors/MockDescriptors.cs:229

• ThreadFields now includes DebuggerFilterContext/ProfilerFilterContext twice (duplicate entries at the end of the list). This can cause ambiguous/incorrect field offsets in the mock type layout. Keep each field only once.
  src/coreclr/vm/cdacstress.cpp:963
• CompareRefSets uses a fixed-size matched[MAX_COLLECTED_REFS] buffer but no longer validates that countA/countB are <= MAX_COLLECTED_REFS. Since CollectStackRefs appends without a hard cap, this can lead to out-of-bounds access when countA or countB exceeds 4096. Reintroduce a guard or allocate the match state sized to the counts.

 return true;
bool matched[MAX_COLLECTED_REFS] = {};
for (int i = 0; i < countA; i++)

[Copilot]

Pull request overview

Copilot reviewed 53 out of 53 changed files in this pull request and generated 7 comments.

[max-charlamb]

feedback for local copilot

[Copilot]

Pull request overview

Copilot reviewed 51 out of 51 changed files in this pull request and generated 9 comments.

Comments suppressed due to low confidence (1)

src/coreclr/vm/cdacstress.cpp:625

• CollectStackRefs appends to pRefs without any cap. Later comparisons allocate fixed-size arrays sized MAX_COLLECTED_REFS (e.g., matched[MAX_COLLECTED_REFS], aUsed/bUsed) and assume counts fit. If the DAC returns > MAX_COLLECTED_REFS refs, this will lead to out-of-bounds writes/reads. Add an explicit limit/overflow handling in CollectStackRefs (stop at MAX_COLLECTED_REFS and mark overflow / SKIP), or change the later comparison logic to handle arbitrary counts safely.

 SOSStackRefData refData;
unsigned int fetched = 0;
while (true)
{
hr = pEnum->Next(1, &refData, &fetched);
if (FAILED(hr) || fetched == 0)
break;
StackRef ref;
ref.Address = refData.Address;
ref.Object = refData.Object;
ref.Flags = refData.Flags;
ref.Source = refData.Source;
ref.SourceType = refData.SourceType;
ref.Register = refData.Register;
ref.Offset = refData.Offset;
ref.StackPointer = refData.StackPointer;
pRefs->Append(ref);
}

[Copilot]

Pull request overview

Copilot reviewed 52 out of 52 changed files in this pull request and generated 3 comments.

[Copilot]

Pull request overview

Copilot reviewed 65 out of 65 changed files in this pull request and generated 5 comments.

[Copilot]

ThreadStore now calls target.GetTypeInfo(DataType.SLink) and reads field "Next", but the CoreCLR data descriptor in this PR doesn't define a SLink type/field (could not find any CDAC_TYPE_BEGIN/FIELD for SLink). This will cause GetTypeInfo(DataType.SLink) to fail at runtime. Either add an SLink descriptor (with a "Next" pointer) on the runtime side or avoid needing type info here (e.g., treat the embedded SLink as a single pointer at offset 0).

[Copilot]

DataType adds SLink, but there is no corresponding runtime data descriptor definition (no CDAC_TYPE_BEGIN/FIELD(SLink, ...) found). Any attempt to read this type via Target.GetTypeInfo(DataType.SLink) will fail. Either add the runtime descriptor entry for SLink or remove this enum value and read the embedded link without type metadata.

Suggested change

	SLink,
	ThreadLocalData,
	ThreadLocalData=17,

[Copilot]

These DataReceived event handlers append to strings via "+=" from background threads. That is not thread-safe and can lead to lost/garbled output under contention. Consider using a StringBuilder with locking (or ConcurrentQueue) for stderr/stdout aggregation.

[Copilot]

Same thread-safety issue as stderr: appending to stdout with += from OutputDataReceived is racy. Use a synchronized StringBuilder/collector to avoid missing output and to keep logs reliable when tests fail.

[Copilot]

GcTypeKindToArgTypeInfo collapses all non-GC-ref types to CorElementType.I/pointer-sized and doesn't preserve float vs integer vs exact primitive sizes. ArgIterator offset calculation depends on the real signature shape (especially on Unix x64/ARM64 where float regs and argument sizing affect later argument placement), so this can produce incorrect offsets and cause missed/incorrect GC ref reporting for signatures with floats or non-pointer-sized primitives. Consider decoding into a richer type representation (e.g., CorElementType + size) and building ArgTypeInfo from that rather than from GcTypeKind.

Suggested change

	/// Converts a <see cref="GcTypeKind"/> to a minimal <see cref="ArgTypeInfo"/> for
	/// ArgIterator consumption. This is a bridge until we have a full type-aware provider.
	/// </summary>
	privatestaticArgTypeInfoGcTypeKindToArgTypeInfo(GcTypeKindkind,intpointerSize)
	{
	returnkindswitch
	{
	GcTypeKind.None=> ArgTypeInfo.ForPrimitive(CorElementType.I,pointerSize),
	GcTypeKind.Ref=> ArgTypeInfo.ForPrimitive(CorElementType.Class,pointerSize),
	GcTypeKind.Interior=> ArgTypeInfo.ForPrimitive(CorElementType.Byref,pointerSize),
	GcTypeKind.Other=>new ArgTypeInfo
	{
	CorElementType=CorElementType.ValueType,
	Size=pointerSize,// Conservative: assume pointer-sized for now
	IsValueType=true,
	},
	_ =>ArgTypeInfo.ForPrimitive(CorElementType.I,pointerSize),
	};
	}
	/// Converts a <see cref="GcTypeKind"/> to a conservative fallback <see cref="ArgTypeInfo"/>
	/// when an exact signature cannot be decoded. Callers should prefer decoding the real
	/// managed signature via <see cref="TryGetArgTypeInfosFromMethodSignature(ReadOnlySpan{byte}, int, out ArgTypeInfo[])"/>
	/// so that floating-point and non-pointer-sized primitives retain their real layout.
	/// </summary>
	privatestaticArgTypeInfoGcTypeKindToArgTypeInfo(GcTypeKindkind,intpointerSize)
	{
	returnkindswitch
	{
	GcTypeKind.Ref=> ArgTypeInfo.ForPrimitive(CorElementType.Class,pointerSize),
	GcTypeKind.Interior=> ArgTypeInfo.ForPrimitive(CorElementType.Byref,pointerSize),
	GcTypeKind.Other=>new ArgTypeInfo
	{
	CorElementType=CorElementType.ValueType,
	Size=pointerSize,
	IsValueType=true,
	},
	_ =>ArgTypeInfo.ForPrimitive(CorElementType.I,pointerSize),
	};
	}
	privatestaticboolTryGetArgTypeInfosFromMethodSignature(ReadOnlySpan<byte>signatureBytes,intpointerSize,outArgTypeInfo[]argTypes)
	{
	argTypes=Array.Empty<ArgTypeInfo>();
	if(signatureBytes.IsEmpty)
	returnfalse;
	BlobReaderreader=new(signatureBytes);
	SignatureHeaderheader=reader.ReadSignatureHeader();
	if(header.IsGeneric)
	reader.ReadCompressedInteger();
	intparameterCount=reader.ReadCompressedInteger();
	if(parameterCount<0)
	returnfalse;
	if(!TrySkipSignatureType(refreader))
	returnfalse;
	ArgTypeInfo[]decodedArgTypes=newArgTypeInfo[parameterCount];
	for(inti=0;i<parameterCount;i++)
	{
	if(!TryReadArgTypeInfo(refreader,pointerSize,outdecodedArgTypes[i]))
	returnfalse;
	}
	argTypes=decodedArgTypes;
	returntrue;
	}
	privatestaticboolTryReadArgTypeInfo(refBlobReaderreader,intpointerSize,outArgTypeInfoargTypeInfo)
	{
	while(reader.RemainingBytes>0)
	{
	CorElementTypeelementType=(CorElementType)reader.ReadByte();
	switch(elementType)
	{
	caseCorElementType.CmodOpt:
	caseCorElementType.CmodReqd:
	reader.ReadCompressedInteger();
	continue;
	caseCorElementType.Pinned:
	continue;
	caseCorElementType.Boolean:
	caseCorElementType.I1:
	caseCorElementType.U1:
	argTypeInfo=ArgTypeInfo.ForPrimitive(elementType,1);
	returntrue;
	caseCorElementType.Char:
	caseCorElementType.I2:
	caseCorElementType.U2:
	argTypeInfo=ArgTypeInfo.ForPrimitive(elementType,2);
	returntrue;
	caseCorElementType.I4:
	caseCorElementType.U4:
	caseCorElementType.R4:
	argTypeInfo=ArgTypeInfo.ForPrimitive(elementType,4);
	returntrue;
	caseCorElementType.I8:
	caseCorElementType.U8:
	caseCorElementType.R8:
	argTypeInfo=ArgTypeInfo.ForPrimitive(elementType,8);
	returntrue;
	caseCorElementType.I:
	caseCorElementType.U:
	caseCorElementType.Ptr:
	caseCorElementType.FnPtr:
	if(elementTypeisCorElementType.Ptr or CorElementType.FnPtr)
	{
	if(!TrySkipSignatureType(refreader))
	{
	argTypeInfo=default;
	returnfalse;
	}
	}
	argTypeInfo=ArgTypeInfo.ForPrimitive(CorElementType.I,pointerSize);
	returntrue;
	caseCorElementType.Byref:
	if(!TrySkipSignatureType(refreader))
	{
	argTypeInfo=default;
	returnfalse;
	}
	argTypeInfo=ArgTypeInfo.ForPrimitive(CorElementType.Byref,pointerSize);
	returntrue;
	caseCorElementType.String:
	caseCorElementType.Object:
	caseCorElementType.Class:
	if(elementTypeisCorElementType.Class)
	reader.ReadCompressedInteger();
	argTypeInfo=ArgTypeInfo.ForPrimitive(CorElementType.Class,pointerSize);
	returntrue;
	caseCorElementType.SzArray:
	if(!TrySkipSignatureType(refreader))
	{
	argTypeInfo=default;
	returnfalse;
	}
	argTypeInfo=ArgTypeInfo.ForPrimitive(CorElementType.Class,pointerSize);
	returntrue;
	caseCorElementType.ValueType:
	reader.ReadCompressedInteger();
	argTypeInfo=newArgTypeInfo
	{
	CorElementType=CorElementType.ValueType,
	Size=pointerSize,
	IsValueType=true,
	};
	returntrue;
	caseCorElementType.GenericInst:
	{
	if(!TrySkipSignatureType(refreader))
	{
	argTypeInfo=default;
	returnfalse;
	}
	intgenericArgCount=reader.ReadCompressedInteger();
	for(inti=0;i<genericArgCount;i++)
	{
	if(!TrySkipSignatureType(refreader))
	{
	argTypeInfo=default;
	returnfalse;
	}
	}
	argTypeInfo=newArgTypeInfo
	{
	CorElementType=CorElementType.ValueType,
	Size=pointerSize,
	IsValueType=true,
	};
	returntrue;
	}
	default:
	argTypeInfo=default;
	returnfalse;
	}
	}
	argTypeInfo=default;
	returnfalse;
	}
	privatestaticboolTrySkipSignatureType(refBlobReaderreader)
	{
	while(reader.RemainingBytes>0)
	{
	CorElementTypeelementType=(CorElementType)reader.ReadByte();
	switch(elementType)
	{
	caseCorElementType.CmodOpt:
	caseCorElementType.CmodReqd:
	reader.ReadCompressedInteger();
	continue;
	caseCorElementType.Pinned:
	continue;
	caseCorElementType.Void:
	caseCorElementType.Boolean:
	caseCorElementType.Char:
	caseCorElementType.I1:
	caseCorElementType.U1:
	caseCorElementType.I2:
	caseCorElementType.U2:
	caseCorElementType.I4:
	caseCorElementType.U4:
	caseCorElementType.I8:
	caseCorElementType.U8:
	caseCorElementType.R4:
	caseCorElementType.R8:
	caseCorElementType.String:
	caseCorElementType.Object:
	caseCorElementType.TypedByref:
	caseCorElementType.I:
	caseCorElementType.U:
	returntrue;
	caseCorElementType.Class:
	caseCorElementType.ValueType:
	caseCorElementType.Var:
	caseCorElementType.MVar:
	reader.ReadCompressedInteger();
	returntrue;
	caseCorElementType.Byref:
	caseCorElementType.Ptr:
	caseCorElementType.SzArray:
	returnTrySkipSignatureType(refreader);
	caseCorElementType.GenericInst:
	{
	if(!TrySkipSignatureType(refreader))
	returnfalse;
	intgenericArgCount=reader.ReadCompressedInteger();
	for(inti=0;i<genericArgCount;i++)
	{
	if(!TrySkipSignatureType(refreader))
	returnfalse;
	}
	returntrue;
	}
	caseCorElementType.Array:
	{
	if(!TrySkipSignatureType(refreader))
	returnfalse;
	intrank=reader.ReadCompressedInteger();
	intsizes=reader.ReadCompressedInteger();
	for(inti=0;i<sizes;i++)
	reader.ReadCompressedInteger();
	intlowerBounds=reader.ReadCompressedInteger();
	for(inti=0;i<lowerBounds;i++)
	reader.ReadCompressedInteger();
	returnrank>=0;
	}
	caseCorElementType.FnPtr:
	{
	SignatureHeaderheader=reader.ReadSignatureHeader();
	if(header.IsGeneric)
	reader.ReadCompressedInteger();
	intparameterCount=reader.ReadCompressedInteger();
	if(parameterCount<0)
	returnfalse;
	if(!TrySkipSignatureType(refreader))
	returnfalse;
	for(inti=0;i<parameterCount;i++)
	{
	if(!TrySkipSignatureType(refreader))
	returnfalse;
	}
	returntrue;
	}
	default:
	returnfalse;
	}
	}
	returnfalse;
	}

[Copilot]

Pull request overview

Copilot reviewed 64 out of 64 changed files in this pull request and generated 3 comments.

[Copilot]

CollectStackRefs currently appends every ref returned by ISOSStackRefEnum::Next with no upper bound. Later comparison helpers allocate fixed-size arrays sized by MAX_COLLECTED_REFS (e.g., matched[MAX_COLLECTED_REFS], aUsed[MAX_COLLECTED_REFS]), so if enumeration returns more than MAX_COLLECTED_REFS it can lead to out-of-bounds writes. Please cap collection to MAX_COLLECTED_REFS (and record an overflow/skip reason) or make the comparison logic handle arbitrarily large ref sets safely.

[Copilot]

When CollectRuntimeStackRefs overflows, the code logs a [SKIP] line but continues and still computes rtMatch/pass (and does not increment s_verifySkip). If CDACSTRESS_USE_DAC is not set, this can produce false failures based on a truncated runtime ref set. Consider treating runtime overflow as an actual skip (increment skip counter + return) when RT comparison is used for pass/fail, or otherwise avoid using rtMatch in the presence of overflow.

Suggested change

	if (rtOverflow && s_logFile != nullptr)
	{
	fprintf(s_logFile, "[SKIP] Thread=0x%x IP=0x%p - RT overflow (>%d refs)\n",
	osThreadId, (void*)GetIP(regs), MAX_COLLECTED_REFS);
	if (rtOverflow)
	{
	InterlockedIncrement(&s_verifySkip);
	if (s_logFile != nullptr)
	{
	fprintf(s_logFile, "[SKIP] Thread=0x%x IP=0x%p - RT overflow (>%d refs)\n",
	osThreadId, (void*)GetIP(regs), MAX_COLLECTED_REFS);
	}
	return;

[Copilot]

In ReportUntrackedAndSucceed, ReportSlot is called with reportScratchSlots: true, which forces reporting scratch registers/stack slots even when CodeManagerFlags.ActiveStackFrame is not set (reportScratchSlots local is false). This changes GC root enumeration behavior and can introduce extra roots for non-leaf frames. It looks like this should pass the reportScratchSlots variable instead of always true.

Suggested change

	ReportSlot(slotIndex,reportScratchSlots:true,reportFpBasedSlotsOnly,reportSlot);
	}
	ReportSlot(slotIndex,reportScratchSlots,reportFpBasedSlotsOnly,reportSlot);
	}

[Copilot]

Pull request overview

Copilot reviewed 64 out of 64 changed files in this pull request and generated 2 comments.

[Copilot]

CompareRefSetsFlat uses a fixed-size matched[MAX_COLLECTED_REFS] buffer, but cDAC/DAC ref collection (CollectStackRefs) is unbounded. Without a guard/cap, a large ref set (>4096) can cause out-of-bounds writes during matching. Consider capping cDAC/DAC collection to MAX_COLLECTED_REFS (and treating overflow as a [SKIP]) or using dynamically sized bookkeeping here.

[Copilot]

The async stdout/stderr collection uses stdout += ... / stderr += ... inside DataReceived event handlers. These callbacks can run concurrently, and string concatenation is not thread-safe; it can also be costly for large output. Consider buffering with a thread-safe collector (e.g., ConcurrentQueue<string> or StringBuilder with a lock) and call process.WaitForExit() (or await stream completion) after WaitForExit(timeout) to ensure all async output has been drained before asserting/logging.

[max-charlamb]

Closing in favor of stacked PR approach

[max-charlamb]

Summary

cDAC GC stress verification tool (DOTNET_CdacStress) that compares stack GC references between the cDAC and the runtime at allocation stress points. Includes stack walker fixes, GC reference scanning implementation, custom signature decoding, and calling convention argument iteration.

Note

This PR description was updated with AI assistance from Copilot.

Stack walker fixes

• Fix SW_SKIPPED_FRAME: do not call UpdateContextFromFrame (matches native SFITER_SKIPPED_FRAME_FUNCTION which does not call UpdateRegDisplay)
• Fix skipped-frame loop: check for more skipped frames before yielding managed method (prevents duplicate EnumGcRefs between consecutive skipped frames)
• Restructure Filter() to drive Next() directly, matching native Filter()+NextRaw() integration (prevents funclet-to-parent walk cycles)
• Remove SkipActiveICFOnce/SkipCurrentFrameInCheck — active ICF double-yield is natural and harmless
• Remove IsAtFirstPassExceptionThrowSite — native does not suppress first-pass refs
• Fix IsFirst not preserved for skipped frames (was causing IsActiveFrame=false for the topmost managed frame)

GC reference scanning

• Implement PromoteCallerStack for stub frames (GCRefMap + MetaSig paths)
• Implement SOSDacImpl.GetStackReferences using cDAC contract (was falling back to legacy DAC)
• Read FilterContext for stack walk starting context
• Three-way cDAC/DAC/RT comparison with InProcessDataTarget
• DOTNET_CdacStress bit flags: ALLOC/INSTR/REFS/WALK/USE_DAC/UNIQUE

RuntimeSignatureDecoder

Custom signature decoder that handles ELEMENT_TYPE_INTERNAL (0x21) and ELEMENT_TYPE_CMOD_INTERNAL (0x22) which SRM's SignatureDecoder cannot parse. These occur in IL stubs, marshalling stubs, and unsafe accessor frames.

• IRuntimeSignatureTypeProvider<TType, TGenericContext>: superset of SRM's ISignatureTypeProvider, adds GetInternalType and GetInternalModifiedType
• ISignatureReader + SpanSignatureReader: endianness-aware reader abstraction
• Correct ECMA-335 compliance: TypeDefOrRefOrSpecEncoded token decoding, sign-extension-by-width for compressed signed ints, bounds validation

ArgIterator (ported from crossgen2)

Proper calling convention analysis replacing the simplified 1-slot-per-param approach.

• CallingConventionInfo: hybrid data descriptor layout values + ABI invariant constants for all architectures (x86, x64 Windows/Unix, ARM32, ARM64, LoongArch64, RISC-V64)
• ArgIterator.GetNextOffset(): maps each argument to its actual register or stack offset
• OffsetOfFloatArgumentRegisters added to TransitionBlock data descriptor
• Handles multi-slot args, forced-byref params, return buffer placement, async continuation

CI infrastructure

• cDAC stress tests run in Helix via cdac-stress-helix.proj
• Extended runtime-diagnostics.yml CdacDumpTests buildArgs with +tools.cdacstresstests

Test results

Allocation-level stress (9 debuggees, ~46K verifications):

Debuggee	Verifications	Pass	Fail
BasicAlloc	4,936	4,936	0
DeepStack	4,964	4,964	0
Generics	4,936	4,936	0
MultiThread	4,996	4,996	0
Comprehensive	4,994	4,992	2*
ExceptionHandling	4,958	4,958	0
StructScenarios	4,940	4,940	0
DynamicMethods	6,520	6,520	0
PInvoke	4,936	4,936	0

Instruction-level stress (9 debuggees, ~226K verifications):
All 9 debuggees pass with zero failures across 226,452 verifications.

*Comprehensive's 2 failures are pre-existing legacy DAC issues (DAC returns 0 refs for background threads in kernel waits; cDAC and runtime agree).

[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

Adds a cDAC GC-stress verification harness and extends the cDAC stack-walk / stack-GC-ref pipeline so cDAC stack reference enumeration can be compared against runtime scanning at stress points.

Changes:

• Implements/extends cDAC stack reference enumeration (including Frame-based scanning paths like PromoteCallerStack via GCRefMap / MetaSig) and wires SOS GetStackReferences to the cDAC contract.
• Introduces a new GC stress integration test project with debuggee apps and orchestration targets.
• Extends CoreCLR cDAC stress/GC stress plumbing and data descriptors to support the new stack-walk and frame-scanning capabilities.

Reviewed changes

Copilot reviewed 46 out of 46 changed files in this pull request and generated 12 comments.

Show a summary per file

File	Description
src/native/managed/cdac/tests/MockDescriptors/MockDescriptors.ExecutionManager.cs	Updates mock type layout for execution manager-related data.
src/native/managed/cdac/tests/MockDescriptors/MockDescriptors.cs	Updates mock type layouts (ExceptionInfo/Thread) for contract tests.
src/native/managed/cdac/tests/Microsoft.Diagnostics.DataContractReader.Tests.csproj	Excludes new GCStressTests folder from the existing unit test project compilation.
src/native/managed/cdac/tests/GCStressTests/README.md	Documents how to build/run the new GC stress tests.
src/native/managed/cdac/tests/GCStressTests/Microsoft.Diagnostics.DataContractReader.GCStressTests.csproj	Adds a dedicated GC stress test project.
src/native/managed/cdac/tests/GCStressTests/GCStressTests.targets	MSBuild orchestration to discover/build debuggee projects.
src/native/managed/cdac/tests/GCStressTests/GCStressTestBase.cs	Test harness to run debuggees under corerun and parse verification logs.
src/native/managed/cdac/tests/GCStressTests/GCStressResults.cs	Parses the native verification log into structured pass/fail/skip counts.
src/native/managed/cdac/tests/GCStressTests/Debuggees/PInvoke/Program.cs	Adds a P/Invoke-focused debuggee workload.
src/native/managed/cdac/tests/GCStressTests/Debuggees/PInvoke/PInvoke.csproj	Debuggee project file for PInvoke scenario.
src/native/managed/cdac/tests/GCStressTests/Debuggees/MultiThread/Program.cs	Adds a multi-threaded debuggee workload.
src/native/managed/cdac/tests/GCStressTests/Debuggees/MultiThread/MultiThread.csproj	Debuggee project file for MultiThread scenario.
src/native/managed/cdac/tests/GCStressTests/Debuggees/Generics/Program.cs	Adds a generics/interface/delegate debuggee workload.
src/native/managed/cdac/tests/GCStressTests/Debuggees/Generics/Generics.csproj	Debuggee project file for Generics scenario.
src/native/managed/cdac/tests/GCStressTests/Debuggees/ExceptionHandling/Program.cs	Adds an exception-handling/funclet debuggee workload.
src/native/managed/cdac/tests/GCStressTests/Debuggees/ExceptionHandling/ExceptionHandling.csproj	Debuggee project file for ExceptionHandling scenario.
src/native/managed/cdac/tests/GCStressTests/Debuggees/Directory.Build.props	Shared build props for debuggee projects (output layout, TFM, etc.).
src/native/managed/cdac/tests/GCStressTests/Debuggees/DeepStack/Program.cs	Adds deep-recursion debuggee workload.
src/native/managed/cdac/tests/GCStressTests/Debuggees/DeepStack/DeepStack.csproj	Debuggee project file for DeepStack scenario.
src/native/managed/cdac/tests/GCStressTests/Debuggees/Comprehensive/Program.cs	Adds comprehensive “all scenarios” debuggee workload.
src/native/managed/cdac/tests/GCStressTests/Debuggees/Comprehensive/Comprehensive.csproj	Debuggee project file for Comprehensive scenario.
src/native/managed/cdac/tests/GCStressTests/Debuggees/BasicAlloc/Program.cs	Adds basic allocation/live-ref debuggee workload.
src/native/managed/cdac/tests/GCStressTests/Debuggees/BasicAlloc/BasicAlloc.csproj	Debuggee project file for BasicAlloc scenario.
src/native/managed/cdac/tests/GCStressTests/BasicGCStressTests.cs	Theory-based test suite that runs the debuggees and asserts pass rate.
src/native/managed/cdac/tests/gcstress/known-issues.md	Captures known mismatch classes and limitations.
src/native/managed/cdac/Microsoft.Diagnostics.DataContractReader.Legacy/SOSDacImpl.cs	Implements GetStackReferences using the cDAC contract rather than legacy DAC fallback.
src/native/managed/cdac/Microsoft.Diagnostics.DataContractReader.Contracts/Data/Frames/StubDispatchFrame.cs	Extends StubDispatchFrame data with GCRefMap pointer.
src/native/managed/cdac/Microsoft.Diagnostics.DataContractReader.Contracts/Data/Frames/ExternalMethodFrame.cs	Adds ExternalMethodFrame contract data type (GCRefMap).
src/native/managed/cdac/Microsoft.Diagnostics.DataContractReader.Contracts/Data/Frames/DynamicHelperFrame.cs	Adds DynamicHelperFrame contract data type (flags).
src/native/managed/cdac/Microsoft.Diagnostics.DataContractReader.Contracts/Data/ExceptionInfo.cs	Adds clause-range fields used for catch-handler resumption offset selection.
src/native/managed/cdac/Microsoft.Diagnostics.DataContractReader.Contracts/Contracts/StackWalk/StackWalk_1.cs	Refactors stack-walk filtering and adds Frame-based GC root scanning.
src/native/managed/cdac/Microsoft.Diagnostics.DataContractReader.Contracts/Contracts/StackWalk/GC/GcScanner.cs	Adds optional relOffset override support for GC ref enumeration.
src/native/managed/cdac/Microsoft.Diagnostics.DataContractReader.Contracts/Contracts/StackWalk/GC/GCRefMapDecoder.cs	Implements GCRefMap decoding for transition-block scanning.
src/native/managed/cdac/Microsoft.Diagnostics.DataContractReader.Contracts/Contracts/StackWalk/GC/CorSigParser.cs	Adds minimal signature parsing to classify parameters for MetaSig-based scanning.
src/native/managed/cdac/Microsoft.Diagnostics.DataContractReader.Contracts/Contracts/GCInfo/IGCInfoDecoder.cs	Adds FindFirstInterruptiblePoint API to GCInfo decoders.
src/native/managed/cdac/Microsoft.Diagnostics.DataContractReader.Contracts/Contracts/GCInfo/GCInfoDecoder.cs	Implements FindFirstInterruptiblePoint using decoded interruptible ranges.
src/native/managed/cdac/Microsoft.Diagnostics.DataContractReader.Contracts/Contracts/ExecutionManager/ExecutionManagerCore.EEJitManager.cs	Fixes code-start lookup for exception clause enumeration and adds minor flow adjustments.
src/native/managed/cdac/Microsoft.Diagnostics.DataContractReader.Contracts/Constants.cs	Adds TransitionBlock-related global names used by frame scanning.
src/native/managed/cdac/Microsoft.Diagnostics.DataContractReader.Abstractions/DataType.cs	Adds new DataType enum values for new frame contracts.
src/native/managed/cdac/cdac.slnx	Adds the new GC stress test project to the cDAC solution.
src/coreclr/vm/gccover.cpp	Adds step-based skipping to reduce overhead when throttling verification.
src/coreclr/vm/frames.h	Exposes additional frame fields to cDAC via cdac_data<> descriptors.
src/coreclr/vm/datadescriptor/datadescriptor.inc	Adds new contract fields/globals for frames and TransitionBlock layout.
src/coreclr/vm/cdacstress.cpp	Updates in-process cDAC/DAC verification logic, logging, and step behavior.
eng/Subsets.props	Adds an on-demand subset for running GC stress tests.
docs/design/datacontracts/StackWalk.md	Documents the new frame fields and TransitionBlock globals in the StackWalk contract.

Comments suppressed due to low confidence (1)

src/native/managed/cdac/tests/MockDescriptors/MockDescriptors.cs:228

• ThreadFields contains ProfilerFilterContext twice (lines 224 and 228). Duplicate field entries will skew offsets and make the mock descriptors inconsistent with the real data descriptor. Keep a single ProfilerFilterContext entry in the correct order.

[Copilot]

Pull request overview

Copilot reviewed 50 out of 50 changed files in this pull request and generated 7 comments.

Comments suppressed due to low confidence (2)

src/native/managed/cdac/tests/MockDescriptors/MockDescriptors.cs:229

• ThreadFields now includes DebuggerFilterContext/ProfilerFilterContext twice (duplicate entries at the end of the list). This can cause ambiguous/incorrect field offsets in the mock type layout. Keep each field only once.
  src/coreclr/vm/cdacstress.cpp:963
• CompareRefSets uses a fixed-size matched[MAX_COLLECTED_REFS] buffer but no longer validates that countA/countB are <= MAX_COLLECTED_REFS. Since CollectStackRefs appends without a hard cap, this can lead to out-of-bounds access when countA or countB exceeds 4096. Reintroduce a guard or allocate the match state sized to the counts.

 return true;
bool matched[MAX_COLLECTED_REFS] = {};
for (int i = 0; i < countA; i++)

[Copilot]

Pull request overview

Copilot reviewed 53 out of 53 changed files in this pull request and generated 7 comments.

[max-charlamb]

feedback for local copilot

[Copilot]

Pull request overview

Copilot reviewed 51 out of 51 changed files in this pull request and generated 9 comments.

Comments suppressed due to low confidence (1)

src/coreclr/vm/cdacstress.cpp:625

• CollectStackRefs appends to pRefs without any cap. Later comparisons allocate fixed-size arrays sized MAX_COLLECTED_REFS (e.g., matched[MAX_COLLECTED_REFS], aUsed/bUsed) and assume counts fit. If the DAC returns > MAX_COLLECTED_REFS refs, this will lead to out-of-bounds writes/reads. Add an explicit limit/overflow handling in CollectStackRefs (stop at MAX_COLLECTED_REFS and mark overflow / SKIP), or change the later comparison logic to handle arbitrary counts safely.

 SOSStackRefData refData;
unsigned int fetched = 0;
while (true)
{
hr = pEnum->Next(1, &refData, &fetched);
if (FAILED(hr) || fetched == 0)
break;
StackRef ref;
ref.Address = refData.Address;
ref.Object = refData.Object;
ref.Flags = refData.Flags;
ref.Source = refData.Source;
ref.SourceType = refData.SourceType;
ref.Register = refData.Register;
ref.Offset = refData.Offset;
ref.StackPointer = refData.StackPointer;
pRefs->Append(ref);
}

[Copilot]

Pull request overview

Copilot reviewed 52 out of 52 changed files in this pull request and generated 3 comments.

[Copilot]

Pull request overview

Copilot reviewed 65 out of 65 changed files in this pull request and generated 5 comments.

[Copilot]

ThreadStore now calls target.GetTypeInfo(DataType.SLink) and reads field "Next", but the CoreCLR data descriptor in this PR doesn't define a SLink type/field (could not find any CDAC_TYPE_BEGIN/FIELD for SLink). This will cause GetTypeInfo(DataType.SLink) to fail at runtime. Either add an SLink descriptor (with a "Next" pointer) on the runtime side or avoid needing type info here (e.g., treat the embedded SLink as a single pointer at offset 0).

[Copilot]

DataType adds SLink, but there is no corresponding runtime data descriptor definition (no CDAC_TYPE_BEGIN/FIELD(SLink, ...) found). Any attempt to read this type via Target.GetTypeInfo(DataType.SLink) will fail. Either add the runtime descriptor entry for SLink or remove this enum value and read the embedded link without type metadata.

Suggested change

	SLink,
	ThreadLocalData,
	ThreadLocalData=17,

[Copilot]

These DataReceived event handlers append to strings via "+=" from background threads. That is not thread-safe and can lead to lost/garbled output under contention. Consider using a StringBuilder with locking (or ConcurrentQueue) for stderr/stdout aggregation.

[Copilot]

Same thread-safety issue as stderr: appending to stdout with += from OutputDataReceived is racy. Use a synchronized StringBuilder/collector to avoid missing output and to keep logs reliable when tests fail.

[Copilot]

GcTypeKindToArgTypeInfo collapses all non-GC-ref types to CorElementType.I/pointer-sized and doesn't preserve float vs integer vs exact primitive sizes. ArgIterator offset calculation depends on the real signature shape (especially on Unix x64/ARM64 where float regs and argument sizing affect later argument placement), so this can produce incorrect offsets and cause missed/incorrect GC ref reporting for signatures with floats or non-pointer-sized primitives. Consider decoding into a richer type representation (e.g., CorElementType + size) and building ArgTypeInfo from that rather than from GcTypeKind.

Suggested change

	/// Converts a <see cref="GcTypeKind"/> to a minimal <see cref="ArgTypeInfo"/> for
	/// ArgIterator consumption. This is a bridge until we have a full type-aware provider.
	/// </summary>
	privatestaticArgTypeInfoGcTypeKindToArgTypeInfo(GcTypeKindkind,intpointerSize)
	{
	returnkindswitch
	{
	GcTypeKind.None=> ArgTypeInfo.ForPrimitive(CorElementType.I,pointerSize),
	GcTypeKind.Ref=> ArgTypeInfo.ForPrimitive(CorElementType.Class,pointerSize),
	GcTypeKind.Interior=> ArgTypeInfo.ForPrimitive(CorElementType.Byref,pointerSize),
	GcTypeKind.Other=>new ArgTypeInfo
	{
	CorElementType=CorElementType.ValueType,
	Size=pointerSize,// Conservative: assume pointer-sized for now
	IsValueType=true,
	},
	_ =>ArgTypeInfo.ForPrimitive(CorElementType.I,pointerSize),
	};
	}
	/// Converts a <see cref="GcTypeKind"/> to a conservative fallback <see cref="ArgTypeInfo"/>
	/// when an exact signature cannot be decoded. Callers should prefer decoding the real
	/// managed signature via <see cref="TryGetArgTypeInfosFromMethodSignature(ReadOnlySpan{byte}, int, out ArgTypeInfo[])"/>
	/// so that floating-point and non-pointer-sized primitives retain their real layout.
	/// </summary>
	privatestaticArgTypeInfoGcTypeKindToArgTypeInfo(GcTypeKindkind,intpointerSize)
	{
	returnkindswitch
	{
	GcTypeKind.Ref=> ArgTypeInfo.ForPrimitive(CorElementType.Class,pointerSize),
	GcTypeKind.Interior=> ArgTypeInfo.ForPrimitive(CorElementType.Byref,pointerSize),
	GcTypeKind.Other=>new ArgTypeInfo
	{
	CorElementType=CorElementType.ValueType,
	Size=pointerSize,
	IsValueType=true,
	},
	_ =>ArgTypeInfo.ForPrimitive(CorElementType.I,pointerSize),
	};
	}
	privatestaticboolTryGetArgTypeInfosFromMethodSignature(ReadOnlySpan<byte>signatureBytes,intpointerSize,outArgTypeInfo[]argTypes)
	{
	argTypes=Array.Empty<ArgTypeInfo>();
	if(signatureBytes.IsEmpty)
	returnfalse;
	BlobReaderreader=new(signatureBytes);
	SignatureHeaderheader=reader.ReadSignatureHeader();
	if(header.IsGeneric)
	reader.ReadCompressedInteger();
	intparameterCount=reader.ReadCompressedInteger();
	if(parameterCount<0)
	returnfalse;
	if(!TrySkipSignatureType(refreader))
	returnfalse;
	ArgTypeInfo[]decodedArgTypes=newArgTypeInfo[parameterCount];
	for(inti=0;i<parameterCount;i++)
	{
	if(!TryReadArgTypeInfo(refreader,pointerSize,outdecodedArgTypes[i]))
	returnfalse;
	}
	argTypes=decodedArgTypes;
	returntrue;
	}
	privatestaticboolTryReadArgTypeInfo(refBlobReaderreader,intpointerSize,outArgTypeInfoargTypeInfo)
	{
	while(reader.RemainingBytes>0)
	{
	CorElementTypeelementType=(CorElementType)reader.ReadByte();
	switch(elementType)
	{
	caseCorElementType.CmodOpt:
	caseCorElementType.CmodReqd:
	reader.ReadCompressedInteger();
	continue;
	caseCorElementType.Pinned:
	continue;
	caseCorElementType.Boolean:
	caseCorElementType.I1:
	caseCorElementType.U1:
	argTypeInfo=ArgTypeInfo.ForPrimitive(elementType,1);
	returntrue;
	caseCorElementType.Char:
	caseCorElementType.I2:
	caseCorElementType.U2:
	argTypeInfo=ArgTypeInfo.ForPrimitive(elementType,2);
	returntrue;
	caseCorElementType.I4:
	caseCorElementType.U4:
	caseCorElementType.R4:
	argTypeInfo=ArgTypeInfo.ForPrimitive(elementType,4);
	returntrue;
	caseCorElementType.I8:
	caseCorElementType.U8:
	caseCorElementType.R8:
	argTypeInfo=ArgTypeInfo.ForPrimitive(elementType,8);
	returntrue;
	caseCorElementType.I:
	caseCorElementType.U:
	caseCorElementType.Ptr:
	caseCorElementType.FnPtr:
	if(elementTypeisCorElementType.Ptr or CorElementType.FnPtr)
	{
	if(!TrySkipSignatureType(refreader))
	{
	argTypeInfo=default;
	returnfalse;
	}
	}
	argTypeInfo=ArgTypeInfo.ForPrimitive(CorElementType.I,pointerSize);
	returntrue;
	caseCorElementType.Byref:
	if(!TrySkipSignatureType(refreader))
	{
	argTypeInfo=default;
	returnfalse;
	}
	argTypeInfo=ArgTypeInfo.ForPrimitive(CorElementType.Byref,pointerSize);
	returntrue;
	caseCorElementType.String:
	caseCorElementType.Object:
	caseCorElementType.Class:
	if(elementTypeisCorElementType.Class)
	reader.ReadCompressedInteger();
	argTypeInfo=ArgTypeInfo.ForPrimitive(CorElementType.Class,pointerSize);
	returntrue;
	caseCorElementType.SzArray:
	if(!TrySkipSignatureType(refreader))
	{
	argTypeInfo=default;
	returnfalse;
	}
	argTypeInfo=ArgTypeInfo.ForPrimitive(CorElementType.Class,pointerSize);
	returntrue;
	caseCorElementType.ValueType:
	reader.ReadCompressedInteger();
	argTypeInfo=newArgTypeInfo
	{
	CorElementType=CorElementType.ValueType,
	Size=pointerSize,
	IsValueType=true,
	};
	returntrue;
	caseCorElementType.GenericInst:
	{
	if(!TrySkipSignatureType(refreader))
	{
	argTypeInfo=default;
	returnfalse;
	}
	intgenericArgCount=reader.ReadCompressedInteger();
	for(inti=0;i<genericArgCount;i++)
	{
	if(!TrySkipSignatureType(refreader))
	{
	argTypeInfo=default;
	returnfalse;
	}
	}
	argTypeInfo=newArgTypeInfo
	{
	CorElementType=CorElementType.ValueType,
	Size=pointerSize,
	IsValueType=true,
	};
	returntrue;
	}
	default:
	argTypeInfo=default;
	returnfalse;
	}
	}
	argTypeInfo=default;
	returnfalse;
	}
	privatestaticboolTrySkipSignatureType(refBlobReaderreader)
	{
	while(reader.RemainingBytes>0)
	{
	CorElementTypeelementType=(CorElementType)reader.ReadByte();
	switch(elementType)
	{
	caseCorElementType.CmodOpt:
	caseCorElementType.CmodReqd:
	reader.ReadCompressedInteger();
	continue;
	caseCorElementType.Pinned:
	continue;
	caseCorElementType.Void:
	caseCorElementType.Boolean:
	caseCorElementType.Char:
	caseCorElementType.I1:
	caseCorElementType.U1:
	caseCorElementType.I2:
	caseCorElementType.U2:
	caseCorElementType.I4:
	caseCorElementType.U4:
	caseCorElementType.I8:
	caseCorElementType.U8:
	caseCorElementType.R4:
	caseCorElementType.R8:
	caseCorElementType.String:
	caseCorElementType.Object:
	caseCorElementType.TypedByref:
	caseCorElementType.I:
	caseCorElementType.U:
	returntrue;
	caseCorElementType.Class:
	caseCorElementType.ValueType:
	caseCorElementType.Var:
	caseCorElementType.MVar:
	reader.ReadCompressedInteger();
	returntrue;
	caseCorElementType.Byref:
	caseCorElementType.Ptr:
	caseCorElementType.SzArray:
	returnTrySkipSignatureType(refreader);
	caseCorElementType.GenericInst:
	{
	if(!TrySkipSignatureType(refreader))
	returnfalse;
	intgenericArgCount=reader.ReadCompressedInteger();
	for(inti=0;i<genericArgCount;i++)
	{
	if(!TrySkipSignatureType(refreader))
	returnfalse;
	}
	returntrue;
	}
	caseCorElementType.Array:
	{
	if(!TrySkipSignatureType(refreader))
	returnfalse;
	intrank=reader.ReadCompressedInteger();
	intsizes=reader.ReadCompressedInteger();
	for(inti=0;i<sizes;i++)
	reader.ReadCompressedInteger();
	intlowerBounds=reader.ReadCompressedInteger();
	for(inti=0;i<lowerBounds;i++)
	reader.ReadCompressedInteger();
	returnrank>=0;
	}
	caseCorElementType.FnPtr:
	{
	SignatureHeaderheader=reader.ReadSignatureHeader();
	if(header.IsGeneric)
	reader.ReadCompressedInteger();
	intparameterCount=reader.ReadCompressedInteger();
	if(parameterCount<0)
	returnfalse;
	if(!TrySkipSignatureType(refreader))
	returnfalse;
	for(inti=0;i<parameterCount;i++)
	{
	if(!TrySkipSignatureType(refreader))
	returnfalse;
	}
	returntrue;
	}
	default:
	returnfalse;
	}
	}
	returnfalse;
	}

[Copilot]

Pull request overview

Copilot reviewed 64 out of 64 changed files in this pull request and generated 3 comments.

[Copilot]

CollectStackRefs currently appends every ref returned by ISOSStackRefEnum::Next with no upper bound. Later comparison helpers allocate fixed-size arrays sized by MAX_COLLECTED_REFS (e.g., matched[MAX_COLLECTED_REFS], aUsed[MAX_COLLECTED_REFS]), so if enumeration returns more than MAX_COLLECTED_REFS it can lead to out-of-bounds writes. Please cap collection to MAX_COLLECTED_REFS (and record an overflow/skip reason) or make the comparison logic handle arbitrarily large ref sets safely.

[Copilot]

When CollectRuntimeStackRefs overflows, the code logs a [SKIP] line but continues and still computes rtMatch/pass (and does not increment s_verifySkip). If CDACSTRESS_USE_DAC is not set, this can produce false failures based on a truncated runtime ref set. Consider treating runtime overflow as an actual skip (increment skip counter + return) when RT comparison is used for pass/fail, or otherwise avoid using rtMatch in the presence of overflow.

Suggested change

	if (rtOverflow && s_logFile != nullptr)
	{
	fprintf(s_logFile, "[SKIP] Thread=0x%x IP=0x%p - RT overflow (>%d refs)\n",
	osThreadId, (void*)GetIP(regs), MAX_COLLECTED_REFS);
	if (rtOverflow)
	{
	InterlockedIncrement(&s_verifySkip);
	if (s_logFile != nullptr)
	{
	fprintf(s_logFile, "[SKIP] Thread=0x%x IP=0x%p - RT overflow (>%d refs)\n",
	osThreadId, (void*)GetIP(regs), MAX_COLLECTED_REFS);
	}
	return;

[Copilot]

In ReportUntrackedAndSucceed, ReportSlot is called with reportScratchSlots: true, which forces reporting scratch registers/stack slots even when CodeManagerFlags.ActiveStackFrame is not set (reportScratchSlots local is false). This changes GC root enumeration behavior and can introduce extra roots for non-leaf frames. It looks like this should pass the reportScratchSlots variable instead of always true.

Suggested change

	ReportSlot(slotIndex,reportScratchSlots:true,reportFpBasedSlotsOnly,reportSlot);
	}
	ReportSlot(slotIndex,reportScratchSlots,reportFpBasedSlotsOnly,reportSlot);
	}

[Copilot]

Pull request overview

Copilot reviewed 64 out of 64 changed files in this pull request and generated 2 comments.

[Copilot]

CompareRefSetsFlat uses a fixed-size matched[MAX_COLLECTED_REFS] buffer, but cDAC/DAC ref collection (CollectStackRefs) is unbounded. Without a guard/cap, a large ref set (>4096) can cause out-of-bounds writes during matching. Consider capping cDAC/DAC collection to MAX_COLLECTED_REFS (and treating overflow as a [SKIP]) or using dynamically sized bookkeeping here.

[Copilot]

The async stdout/stderr collection uses stdout += ... / stderr += ... inside DataReceived event handlers. These callbacks can run concurrently, and string concatenation is not thread-safe; it can also be costly for large output. Consider buffering with a thread-safe collector (e.g., ConcurrentQueue<string> or StringBuilder with a lock) and call process.WaitForExit() (or await stream completion) after WaitForExit(timeout) to ensure all async output has been drained before asserting/logging.

[max-charlamb]

Closing in favor of stacked PR approach

[max-charlamb]

Summary

cDAC GC stress verification tool (DOTNET_CdacStress) that compares stack GC references between the cDAC and the runtime at allocation stress points. Includes stack walker fixes, GC reference scanning implementation, custom signature decoding, and calling convention argument iteration.

Note

This PR description was updated with AI assistance from Copilot.

Stack walker fixes

• Fix SW_SKIPPED_FRAME: do not call UpdateContextFromFrame (matches native SFITER_SKIPPED_FRAME_FUNCTION which does not call UpdateRegDisplay)
• Fix skipped-frame loop: check for more skipped frames before yielding managed method (prevents duplicate EnumGcRefs between consecutive skipped frames)
• Restructure Filter() to drive Next() directly, matching native Filter()+NextRaw() integration (prevents funclet-to-parent walk cycles)
• Remove SkipActiveICFOnce/SkipCurrentFrameInCheck — active ICF double-yield is natural and harmless
• Remove IsAtFirstPassExceptionThrowSite — native does not suppress first-pass refs
• Fix IsFirst not preserved for skipped frames (was causing IsActiveFrame=false for the topmost managed frame)

GC reference scanning

• Implement PromoteCallerStack for stub frames (GCRefMap + MetaSig paths)
• Implement SOSDacImpl.GetStackReferences using cDAC contract (was falling back to legacy DAC)
• Read FilterContext for stack walk starting context
• Three-way cDAC/DAC/RT comparison with InProcessDataTarget
• DOTNET_CdacStress bit flags: ALLOC/INSTR/REFS/WALK/USE_DAC/UNIQUE

RuntimeSignatureDecoder

Custom signature decoder that handles ELEMENT_TYPE_INTERNAL (0x21) and ELEMENT_TYPE_CMOD_INTERNAL (0x22) which SRM's SignatureDecoder cannot parse. These occur in IL stubs, marshalling stubs, and unsafe accessor frames.

• IRuntimeSignatureTypeProvider<TType, TGenericContext>: superset of SRM's ISignatureTypeProvider, adds GetInternalType and GetInternalModifiedType
• ISignatureReader + SpanSignatureReader: endianness-aware reader abstraction
• Correct ECMA-335 compliance: TypeDefOrRefOrSpecEncoded token decoding, sign-extension-by-width for compressed signed ints, bounds validation

ArgIterator (ported from crossgen2)

Proper calling convention analysis replacing the simplified 1-slot-per-param approach.

• CallingConventionInfo: hybrid data descriptor layout values + ABI invariant constants for all architectures (x86, x64 Windows/Unix, ARM32, ARM64, LoongArch64, RISC-V64)
• ArgIterator.GetNextOffset(): maps each argument to its actual register or stack offset
• OffsetOfFloatArgumentRegisters added to TransitionBlock data descriptor
• Handles multi-slot args, forced-byref params, return buffer placement, async continuation

CI infrastructure

• cDAC stress tests run in Helix via cdac-stress-helix.proj
• Extended runtime-diagnostics.yml CdacDumpTests buildArgs with +tools.cdacstresstests

Test results

Allocation-level stress (9 debuggees, ~46K verifications):

Debuggee	Verifications	Pass	Fail
BasicAlloc	4,936	4,936	0
DeepStack	4,964	4,964	0
Generics	4,936	4,936	0
MultiThread	4,996	4,996	0
Comprehensive	4,994	4,992	2*
ExceptionHandling	4,958	4,958	0
StructScenarios	4,940	4,940	0
DynamicMethods	6,520	6,520	0
PInvoke	4,936	4,936	0

Instruction-level stress (9 debuggees, ~226K verifications):
All 9 debuggees pass with zero failures across 226,452 verifications.

*Comprehensive's 2 failures are pre-existing legacy DAC issues (DAC returns 0 refs for background threads in kernel waits; cDAC and runtime agree).

[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

Adds a cDAC GC-stress verification harness and extends the cDAC stack-walk / stack-GC-ref pipeline so cDAC stack reference enumeration can be compared against runtime scanning at stress points.

Changes:

• Implements/extends cDAC stack reference enumeration (including Frame-based scanning paths like PromoteCallerStack via GCRefMap / MetaSig) and wires SOS GetStackReferences to the cDAC contract.
• Introduces a new GC stress integration test project with debuggee apps and orchestration targets.
• Extends CoreCLR cDAC stress/GC stress plumbing and data descriptors to support the new stack-walk and frame-scanning capabilities.

Reviewed changes

Copilot reviewed 46 out of 46 changed files in this pull request and generated 12 comments.

Show a summary per file

File	Description
src/native/managed/cdac/tests/MockDescriptors/MockDescriptors.ExecutionManager.cs	Updates mock type layout for execution manager-related data.
src/native/managed/cdac/tests/MockDescriptors/MockDescriptors.cs	Updates mock type layouts (ExceptionInfo/Thread) for contract tests.
src/native/managed/cdac/tests/Microsoft.Diagnostics.DataContractReader.Tests.csproj	Excludes new GCStressTests folder from the existing unit test project compilation.
src/native/managed/cdac/tests/GCStressTests/README.md	Documents how to build/run the new GC stress tests.
src/native/managed/cdac/tests/GCStressTests/Microsoft.Diagnostics.DataContractReader.GCStressTests.csproj	Adds a dedicated GC stress test project.
src/native/managed/cdac/tests/GCStressTests/GCStressTests.targets	MSBuild orchestration to discover/build debuggee projects.
src/native/managed/cdac/tests/GCStressTests/GCStressTestBase.cs	Test harness to run debuggees under corerun and parse verification logs.
src/native/managed/cdac/tests/GCStressTests/GCStressResults.cs	Parses the native verification log into structured pass/fail/skip counts.
src/native/managed/cdac/tests/GCStressTests/Debuggees/PInvoke/Program.cs	Adds a P/Invoke-focused debuggee workload.
src/native/managed/cdac/tests/GCStressTests/Debuggees/PInvoke/PInvoke.csproj	Debuggee project file for PInvoke scenario.
src/native/managed/cdac/tests/GCStressTests/Debuggees/MultiThread/Program.cs	Adds a multi-threaded debuggee workload.
src/native/managed/cdac/tests/GCStressTests/Debuggees/MultiThread/MultiThread.csproj	Debuggee project file for MultiThread scenario.
src/native/managed/cdac/tests/GCStressTests/Debuggees/Generics/Program.cs	Adds a generics/interface/delegate debuggee workload.
src/native/managed/cdac/tests/GCStressTests/Debuggees/Generics/Generics.csproj	Debuggee project file for Generics scenario.
src/native/managed/cdac/tests/GCStressTests/Debuggees/ExceptionHandling/Program.cs	Adds an exception-handling/funclet debuggee workload.
src/native/managed/cdac/tests/GCStressTests/Debuggees/ExceptionHandling/ExceptionHandling.csproj	Debuggee project file for ExceptionHandling scenario.
src/native/managed/cdac/tests/GCStressTests/Debuggees/Directory.Build.props	Shared build props for debuggee projects (output layout, TFM, etc.).
src/native/managed/cdac/tests/GCStressTests/Debuggees/DeepStack/Program.cs	Adds deep-recursion debuggee workload.
src/native/managed/cdac/tests/GCStressTests/Debuggees/DeepStack/DeepStack.csproj	Debuggee project file for DeepStack scenario.
src/native/managed/cdac/tests/GCStressTests/Debuggees/Comprehensive/Program.cs	Adds comprehensive “all scenarios” debuggee workload.
src/native/managed/cdac/tests/GCStressTests/Debuggees/Comprehensive/Comprehensive.csproj	Debuggee project file for Comprehensive scenario.
src/native/managed/cdac/tests/GCStressTests/Debuggees/BasicAlloc/Program.cs	Adds basic allocation/live-ref debuggee workload.
src/native/managed/cdac/tests/GCStressTests/Debuggees/BasicAlloc/BasicAlloc.csproj	Debuggee project file for BasicAlloc scenario.
src/native/managed/cdac/tests/GCStressTests/BasicGCStressTests.cs	Theory-based test suite that runs the debuggees and asserts pass rate.
src/native/managed/cdac/tests/gcstress/known-issues.md	Captures known mismatch classes and limitations.
src/native/managed/cdac/Microsoft.Diagnostics.DataContractReader.Legacy/SOSDacImpl.cs	Implements GetStackReferences using the cDAC contract rather than legacy DAC fallback.
src/native/managed/cdac/Microsoft.Diagnostics.DataContractReader.Contracts/Data/Frames/StubDispatchFrame.cs	Extends StubDispatchFrame data with GCRefMap pointer.
src/native/managed/cdac/Microsoft.Diagnostics.DataContractReader.Contracts/Data/Frames/ExternalMethodFrame.cs	Adds ExternalMethodFrame contract data type (GCRefMap).
src/native/managed/cdac/Microsoft.Diagnostics.DataContractReader.Contracts/Data/Frames/DynamicHelperFrame.cs	Adds DynamicHelperFrame contract data type (flags).
src/native/managed/cdac/Microsoft.Diagnostics.DataContractReader.Contracts/Data/ExceptionInfo.cs	Adds clause-range fields used for catch-handler resumption offset selection.
src/native/managed/cdac/Microsoft.Diagnostics.DataContractReader.Contracts/Contracts/StackWalk/StackWalk_1.cs	Refactors stack-walk filtering and adds Frame-based GC root scanning.
src/native/managed/cdac/Microsoft.Diagnostics.DataContractReader.Contracts/Contracts/StackWalk/GC/GcScanner.cs	Adds optional relOffset override support for GC ref enumeration.
src/native/managed/cdac/Microsoft.Diagnostics.DataContractReader.Contracts/Contracts/StackWalk/GC/GCRefMapDecoder.cs	Implements GCRefMap decoding for transition-block scanning.
src/native/managed/cdac/Microsoft.Diagnostics.DataContractReader.Contracts/Contracts/StackWalk/GC/CorSigParser.cs	Adds minimal signature parsing to classify parameters for MetaSig-based scanning.
src/native/managed/cdac/Microsoft.Diagnostics.DataContractReader.Contracts/Contracts/GCInfo/IGCInfoDecoder.cs	Adds FindFirstInterruptiblePoint API to GCInfo decoders.
src/native/managed/cdac/Microsoft.Diagnostics.DataContractReader.Contracts/Contracts/GCInfo/GCInfoDecoder.cs	Implements FindFirstInterruptiblePoint using decoded interruptible ranges.
src/native/managed/cdac/Microsoft.Diagnostics.DataContractReader.Contracts/Contracts/ExecutionManager/ExecutionManagerCore.EEJitManager.cs	Fixes code-start lookup for exception clause enumeration and adds minor flow adjustments.
src/native/managed/cdac/Microsoft.Diagnostics.DataContractReader.Contracts/Constants.cs	Adds TransitionBlock-related global names used by frame scanning.
src/native/managed/cdac/Microsoft.Diagnostics.DataContractReader.Abstractions/DataType.cs	Adds new DataType enum values for new frame contracts.
src/native/managed/cdac/cdac.slnx	Adds the new GC stress test project to the cDAC solution.
src/coreclr/vm/gccover.cpp	Adds step-based skipping to reduce overhead when throttling verification.
src/coreclr/vm/frames.h	Exposes additional frame fields to cDAC via cdac_data<> descriptors.
src/coreclr/vm/datadescriptor/datadescriptor.inc	Adds new contract fields/globals for frames and TransitionBlock layout.
src/coreclr/vm/cdacstress.cpp	Updates in-process cDAC/DAC verification logic, logging, and step behavior.
eng/Subsets.props	Adds an on-demand subset for running GC stress tests.
docs/design/datacontracts/StackWalk.md	Documents the new frame fields and TransitionBlock globals in the StackWalk contract.

Comments suppressed due to low confidence (1)

src/native/managed/cdac/tests/MockDescriptors/MockDescriptors.cs:228

• ThreadFields contains ProfilerFilterContext twice (lines 224 and 228). Duplicate field entries will skew offsets and make the mock descriptors inconsistent with the real data descriptor. Keep a single ProfilerFilterContext entry in the correct order.

[Copilot]

Pull request overview

Copilot reviewed 50 out of 50 changed files in this pull request and generated 7 comments.

Comments suppressed due to low confidence (2)

src/native/managed/cdac/tests/MockDescriptors/MockDescriptors.cs:229

• ThreadFields now includes DebuggerFilterContext/ProfilerFilterContext twice (duplicate entries at the end of the list). This can cause ambiguous/incorrect field offsets in the mock type layout. Keep each field only once.
  src/coreclr/vm/cdacstress.cpp:963
• CompareRefSets uses a fixed-size matched[MAX_COLLECTED_REFS] buffer but no longer validates that countA/countB are <= MAX_COLLECTED_REFS. Since CollectStackRefs appends without a hard cap, this can lead to out-of-bounds access when countA or countB exceeds 4096. Reintroduce a guard or allocate the match state sized to the counts.

 return true;
bool matched[MAX_COLLECTED_REFS] = {};
for (int i = 0; i < countA; i++)

[Copilot]

Pull request overview

Copilot reviewed 53 out of 53 changed files in this pull request and generated 7 comments.

[max-charlamb]

feedback for local copilot

[Copilot]

Pull request overview

Copilot reviewed 51 out of 51 changed files in this pull request and generated 9 comments.

Comments suppressed due to low confidence (1)

src/coreclr/vm/cdacstress.cpp:625

• CollectStackRefs appends to pRefs without any cap. Later comparisons allocate fixed-size arrays sized MAX_COLLECTED_REFS (e.g., matched[MAX_COLLECTED_REFS], aUsed/bUsed) and assume counts fit. If the DAC returns > MAX_COLLECTED_REFS refs, this will lead to out-of-bounds writes/reads. Add an explicit limit/overflow handling in CollectStackRefs (stop at MAX_COLLECTED_REFS and mark overflow / SKIP), or change the later comparison logic to handle arbitrary counts safely.

 SOSStackRefData refData;
unsigned int fetched = 0;
while (true)
{
hr = pEnum->Next(1, &refData, &fetched);
if (FAILED(hr) || fetched == 0)
break;
StackRef ref;
ref.Address = refData.Address;
ref.Object = refData.Object;
ref.Flags = refData.Flags;
ref.Source = refData.Source;
ref.SourceType = refData.SourceType;
ref.Register = refData.Register;
ref.Offset = refData.Offset;
ref.StackPointer = refData.StackPointer;
pRefs->Append(ref);
}

[Copilot]

Pull request overview

Copilot reviewed 52 out of 52 changed files in this pull request and generated 3 comments.

[Copilot]

Pull request overview

Copilot reviewed 65 out of 65 changed files in this pull request and generated 5 comments.

[Copilot]

ThreadStore now calls target.GetTypeInfo(DataType.SLink) and reads field "Next", but the CoreCLR data descriptor in this PR doesn't define a SLink type/field (could not find any CDAC_TYPE_BEGIN/FIELD for SLink). This will cause GetTypeInfo(DataType.SLink) to fail at runtime. Either add an SLink descriptor (with a "Next" pointer) on the runtime side or avoid needing type info here (e.g., treat the embedded SLink as a single pointer at offset 0).

[Copilot]

DataType adds SLink, but there is no corresponding runtime data descriptor definition (no CDAC_TYPE_BEGIN/FIELD(SLink, ...) found). Any attempt to read this type via Target.GetTypeInfo(DataType.SLink) will fail. Either add the runtime descriptor entry for SLink or remove this enum value and read the embedded link without type metadata.

Suggested change

	SLink,
	ThreadLocalData,
	ThreadLocalData=17,

[Copilot]

These DataReceived event handlers append to strings via "+=" from background threads. That is not thread-safe and can lead to lost/garbled output under contention. Consider using a StringBuilder with locking (or ConcurrentQueue) for stderr/stdout aggregation.

[Copilot]

Same thread-safety issue as stderr: appending to stdout with += from OutputDataReceived is racy. Use a synchronized StringBuilder/collector to avoid missing output and to keep logs reliable when tests fail.

[Copilot]

GcTypeKindToArgTypeInfo collapses all non-GC-ref types to CorElementType.I/pointer-sized and doesn't preserve float vs integer vs exact primitive sizes. ArgIterator offset calculation depends on the real signature shape (especially on Unix x64/ARM64 where float regs and argument sizing affect later argument placement), so this can produce incorrect offsets and cause missed/incorrect GC ref reporting for signatures with floats or non-pointer-sized primitives. Consider decoding into a richer type representation (e.g., CorElementType + size) and building ArgTypeInfo from that rather than from GcTypeKind.

Suggested change

	/// Converts a <see cref="GcTypeKind"/> to a minimal <see cref="ArgTypeInfo"/> for
	/// ArgIterator consumption. This is a bridge until we have a full type-aware provider.
	/// </summary>
	privatestaticArgTypeInfoGcTypeKindToArgTypeInfo(GcTypeKindkind,intpointerSize)
	{
	returnkindswitch
	{
	GcTypeKind.None=> ArgTypeInfo.ForPrimitive(CorElementType.I,pointerSize),
	GcTypeKind.Ref=> ArgTypeInfo.ForPrimitive(CorElementType.Class,pointerSize),
	GcTypeKind.Interior=> ArgTypeInfo.ForPrimitive(CorElementType.Byref,pointerSize),
	GcTypeKind.Other=>new ArgTypeInfo
	{
	CorElementType=CorElementType.ValueType,
	Size=pointerSize,// Conservative: assume pointer-sized for now
	IsValueType=true,
	},
	_ =>ArgTypeInfo.ForPrimitive(CorElementType.I,pointerSize),
	};
	}
	/// Converts a <see cref="GcTypeKind"/> to a conservative fallback <see cref="ArgTypeInfo"/>
	/// when an exact signature cannot be decoded. Callers should prefer decoding the real
	/// managed signature via <see cref="TryGetArgTypeInfosFromMethodSignature(ReadOnlySpan{byte}, int, out ArgTypeInfo[])"/>
	/// so that floating-point and non-pointer-sized primitives retain their real layout.
	/// </summary>
	privatestaticArgTypeInfoGcTypeKindToArgTypeInfo(GcTypeKindkind,intpointerSize)
	{
	returnkindswitch
	{
	GcTypeKind.Ref=> ArgTypeInfo.ForPrimitive(CorElementType.Class,pointerSize),
	GcTypeKind.Interior=> ArgTypeInfo.ForPrimitive(CorElementType.Byref,pointerSize),
	GcTypeKind.Other=>new ArgTypeInfo
	{
	CorElementType=CorElementType.ValueType,
	Size=pointerSize,
	IsValueType=true,
	},
	_ =>ArgTypeInfo.ForPrimitive(CorElementType.I,pointerSize),
	};
	}
	privatestaticboolTryGetArgTypeInfosFromMethodSignature(ReadOnlySpan<byte>signatureBytes,intpointerSize,outArgTypeInfo[]argTypes)
	{
	argTypes=Array.Empty<ArgTypeInfo>();
	if(signatureBytes.IsEmpty)
	returnfalse;
	BlobReaderreader=new(signatureBytes);
	SignatureHeaderheader=reader.ReadSignatureHeader();
	if(header.IsGeneric)
	reader.ReadCompressedInteger();
	intparameterCount=reader.ReadCompressedInteger();
	if(parameterCount<0)
	returnfalse;
	if(!TrySkipSignatureType(refreader))
	returnfalse;
	ArgTypeInfo[]decodedArgTypes=newArgTypeInfo[parameterCount];
	for(inti=0;i<parameterCount;i++)
	{
	if(!TryReadArgTypeInfo(refreader,pointerSize,outdecodedArgTypes[i]))
	returnfalse;
	}
	argTypes=decodedArgTypes;
	returntrue;
	}
	privatestaticboolTryReadArgTypeInfo(refBlobReaderreader,intpointerSize,outArgTypeInfoargTypeInfo)
	{
	while(reader.RemainingBytes>0)
	{
	CorElementTypeelementType=(CorElementType)reader.ReadByte();
	switch(elementType)
	{
	caseCorElementType.CmodOpt:
	caseCorElementType.CmodReqd:
	reader.ReadCompressedInteger();
	continue;
	caseCorElementType.Pinned:
	continue;
	caseCorElementType.Boolean:
	caseCorElementType.I1:
	caseCorElementType.U1:
	argTypeInfo=ArgTypeInfo.ForPrimitive(elementType,1);
	returntrue;
	caseCorElementType.Char:
	caseCorElementType.I2:
	caseCorElementType.U2:
	argTypeInfo=ArgTypeInfo.ForPrimitive(elementType,2);
	returntrue;
	caseCorElementType.I4:
	caseCorElementType.U4:
	caseCorElementType.R4:
	argTypeInfo=ArgTypeInfo.ForPrimitive(elementType,4);
	returntrue;
	caseCorElementType.I8:
	caseCorElementType.U8:
	caseCorElementType.R8:
	argTypeInfo=ArgTypeInfo.ForPrimitive(elementType,8);
	returntrue;
	caseCorElementType.I:
	caseCorElementType.U:
	caseCorElementType.Ptr:
	caseCorElementType.FnPtr:
	if(elementTypeisCorElementType.Ptr or CorElementType.FnPtr)
	{
	if(!TrySkipSignatureType(refreader))
	{
	argTypeInfo=default;
	returnfalse;
	}
	}
	argTypeInfo=ArgTypeInfo.ForPrimitive(CorElementType.I,pointerSize);
	returntrue;
	caseCorElementType.Byref:
	if(!TrySkipSignatureType(refreader))
	{
	argTypeInfo=default;
	returnfalse;
	}
	argTypeInfo=ArgTypeInfo.ForPrimitive(CorElementType.Byref,pointerSize);
	returntrue;
	caseCorElementType.String:
	caseCorElementType.Object:
	caseCorElementType.Class:
	if(elementTypeisCorElementType.Class)
	reader.ReadCompressedInteger();
	argTypeInfo=ArgTypeInfo.ForPrimitive(CorElementType.Class,pointerSize);
	returntrue;
	caseCorElementType.SzArray:
	if(!TrySkipSignatureType(refreader))
	{
	argTypeInfo=default;
	returnfalse;
	}
	argTypeInfo=ArgTypeInfo.ForPrimitive(CorElementType.Class,pointerSize);
	returntrue;
	caseCorElementType.ValueType:
	reader.ReadCompressedInteger();
	argTypeInfo=newArgTypeInfo
	{
	CorElementType=CorElementType.ValueType,
	Size=pointerSize,
	IsValueType=true,
	};
	returntrue;
	caseCorElementType.GenericInst:
	{
	if(!TrySkipSignatureType(refreader))
	{
	argTypeInfo=default;
	returnfalse;
	}
	intgenericArgCount=reader.ReadCompressedInteger();
	for(inti=0;i<genericArgCount;i++)
	{
	if(!TrySkipSignatureType(refreader))
	{
	argTypeInfo=default;
	returnfalse;
	}
	}
	argTypeInfo=newArgTypeInfo
	{
	CorElementType=CorElementType.ValueType,
	Size=pointerSize,
	IsValueType=true,
	};
	returntrue;
	}
	default:
	argTypeInfo=default;
	returnfalse;
	}
	}
	argTypeInfo=default;
	returnfalse;
	}
	privatestaticboolTrySkipSignatureType(refBlobReaderreader)
	{
	while(reader.RemainingBytes>0)
	{
	CorElementTypeelementType=(CorElementType)reader.ReadByte();
	switch(elementType)
	{
	caseCorElementType.CmodOpt:
	caseCorElementType.CmodReqd:
	reader.ReadCompressedInteger();
	continue;
	caseCorElementType.Pinned:
	continue;
	caseCorElementType.Void:
	caseCorElementType.Boolean:
	caseCorElementType.Char:
	caseCorElementType.I1:
	caseCorElementType.U1:
	caseCorElementType.I2:
	caseCorElementType.U2:
	caseCorElementType.I4:
	caseCorElementType.U4:
	caseCorElementType.I8:
	caseCorElementType.U8:
	caseCorElementType.R4:
	caseCorElementType.R8:
	caseCorElementType.String:
	caseCorElementType.Object:
	caseCorElementType.TypedByref:
	caseCorElementType.I:
	caseCorElementType.U:
	returntrue;
	caseCorElementType.Class:
	caseCorElementType.ValueType:
	caseCorElementType.Var:
	caseCorElementType.MVar:
	reader.ReadCompressedInteger();
	returntrue;
	caseCorElementType.Byref:
	caseCorElementType.Ptr:
	caseCorElementType.SzArray:
	returnTrySkipSignatureType(refreader);
	caseCorElementType.GenericInst:
	{
	if(!TrySkipSignatureType(refreader))
	returnfalse;
	intgenericArgCount=reader.ReadCompressedInteger();
	for(inti=0;i<genericArgCount;i++)
	{
	if(!TrySkipSignatureType(refreader))
	returnfalse;
	}
	returntrue;
	}
	caseCorElementType.Array:
	{
	if(!TrySkipSignatureType(refreader))
	returnfalse;
	intrank=reader.ReadCompressedInteger();
	intsizes=reader.ReadCompressedInteger();
	for(inti=0;i<sizes;i++)
	reader.ReadCompressedInteger();
	intlowerBounds=reader.ReadCompressedInteger();
	for(inti=0;i<lowerBounds;i++)
	reader.ReadCompressedInteger();
	returnrank>=0;
	}
	caseCorElementType.FnPtr:
	{
	SignatureHeaderheader=reader.ReadSignatureHeader();
	if(header.IsGeneric)
	reader.ReadCompressedInteger();
	intparameterCount=reader.ReadCompressedInteger();
	if(parameterCount<0)
	returnfalse;
	if(!TrySkipSignatureType(refreader))
	returnfalse;
	for(inti=0;i<parameterCount;i++)
	{
	if(!TrySkipSignatureType(refreader))
	returnfalse;
	}
	returntrue;
	}
	default:
	returnfalse;
	}
	}
	returnfalse;
	}

[Copilot]

Pull request overview

Copilot reviewed 64 out of 64 changed files in this pull request and generated 3 comments.

[Copilot]

CollectStackRefs currently appends every ref returned by ISOSStackRefEnum::Next with no upper bound. Later comparison helpers allocate fixed-size arrays sized by MAX_COLLECTED_REFS (e.g., matched[MAX_COLLECTED_REFS], aUsed[MAX_COLLECTED_REFS]), so if enumeration returns more than MAX_COLLECTED_REFS it can lead to out-of-bounds writes. Please cap collection to MAX_COLLECTED_REFS (and record an overflow/skip reason) or make the comparison logic handle arbitrarily large ref sets safely.

[Copilot]

When CollectRuntimeStackRefs overflows, the code logs a [SKIP] line but continues and still computes rtMatch/pass (and does not increment s_verifySkip). If CDACSTRESS_USE_DAC is not set, this can produce false failures based on a truncated runtime ref set. Consider treating runtime overflow as an actual skip (increment skip counter + return) when RT comparison is used for pass/fail, or otherwise avoid using rtMatch in the presence of overflow.

Suggested change

	if (rtOverflow && s_logFile != nullptr)
	{
	fprintf(s_logFile, "[SKIP] Thread=0x%x IP=0x%p - RT overflow (>%d refs)\n",
	osThreadId, (void*)GetIP(regs), MAX_COLLECTED_REFS);
	if (rtOverflow)
	{
	InterlockedIncrement(&s_verifySkip);
	if (s_logFile != nullptr)
	{
	fprintf(s_logFile, "[SKIP] Thread=0x%x IP=0x%p - RT overflow (>%d refs)\n",
	osThreadId, (void*)GetIP(regs), MAX_COLLECTED_REFS);
	}
	return;

[Copilot]

In ReportUntrackedAndSucceed, ReportSlot is called with reportScratchSlots: true, which forces reporting scratch registers/stack slots even when CodeManagerFlags.ActiveStackFrame is not set (reportScratchSlots local is false). This changes GC root enumeration behavior and can introduce extra roots for non-leaf frames. It looks like this should pass the reportScratchSlots variable instead of always true.

Suggested change

	ReportSlot(slotIndex,reportScratchSlots:true,reportFpBasedSlotsOnly,reportSlot);
	}
	ReportSlot(slotIndex,reportScratchSlots,reportFpBasedSlotsOnly,reportSlot);
	}

[Copilot]

Pull request overview

Copilot reviewed 64 out of 64 changed files in this pull request and generated 2 comments.

[Copilot]

CompareRefSetsFlat uses a fixed-size matched[MAX_COLLECTED_REFS] buffer, but cDAC/DAC ref collection (CollectStackRefs) is unbounded. Without a guard/cap, a large ref set (>4096) can cause out-of-bounds writes during matching. Consider capping cDAC/DAC collection to MAX_COLLECTED_REFS (and treating overflow as a [SKIP]) or using dynamically sized bookkeeping here.

[Copilot]

The async stdout/stderr collection uses stdout += ... / stderr += ... inside DataReceived event handlers. These callbacks can run concurrently, and string concatenation is not thread-safe; it can also be costly for large output. Consider buffering with a thread-safe collector (e.g., ConcurrentQueue<string> or StringBuilder with a lock) and call process.WaitForExit() (or await stream completion) after WaitForExit(timeout) to ensure all async output has been drained before asserting/logging.

[max-charlamb]

Closing in favor of stacked PR approach