Emit task returning thunks in crossgen2 #122651

jtschuster · 2025-12-18T20:58:08Z

Emits Task-returning async method thunks in crossgen2 as MethodSignatures with the AsyncVariant flag set

Adds GetPrimaryMethodDesc to get the most natural "primary" method desc for MethodDesc's that point to the same Ecma/Instantiated method. This is used in the places where we cast to EcmaMethod to get relevant metadata information.

GCRefMap emission also needed to be updated to account for the continuation parameter. Runtime's FakeGcScanRoots (used to verify the ReadyToRun images GCRefMap in debug config) also needed to updated to handle the continuation parameter.

…ithToken

Copilot

Pull request overview

This PR implements support for emitting Task-returning async method thunks in crossgen2 by encoding them as MethodSignatures with the AsyncVariant flag set. The changes introduce a new GetPrimaryMethodDesc helper method to navigate between different MethodDesc variants (async, unboxing, P/Invoke) and their primary metadata representations. The GC reference map emission is updated to account for the additional async continuation parameter.

Key changes:

Adds AsyncVariant flag to ReadyToRunMethodSigFlags enum (requiring uint instead of byte to accommodate 0x100 value)
Implements GetPrimaryMethodDesc extension method to resolve various MethodDesc wrappers to their primary ECMA method
Updates ArgIterator and GCRefMapBuilder to handle async continuation parameter in both managed and native code
Modifies ReadyToRunILProvider to emit Task-returning thunks for async methods

Reviewed changes

Copilot reviewed 16 out of 16 changed files in this pull request and generated 3 comments.

Show a summary per file

File	Description
src/coreclr/vm/frames.cpp	Adds async continuation parameter handling in FakeGcScanRoots for GC scanning
src/coreclr/tools/aot/ILCompiler.Reflection.ReadyToRun/ReadyToRunSignature.cs	Adds display support for ASYNC flag in method signatures
src/coreclr/tools/aot/ILCompiler.ReadyToRun/TypeSystem/MethodDescExtensions.cs	New file introducing GetPrimaryMethodDesc helper to unwrap MethodDesc variants
src/coreclr/tools/aot/ILCompiler.ReadyToRun/JitInterface/CorInfoImpl.ReadyToRun.cs	Updates token resolution to use GetPrimaryMethodDesc and removes version bubble checks
src/coreclr/tools/aot/ILCompiler.ReadyToRun/ILCompiler.ReadyToRun.csproj	Adds new MethodDescExtensions.cs file to project
src/coreclr/tools/aot/ILCompiler.ReadyToRun/IL/ReadyToRunILProvider.cs	Implements Task-returning async thunk emission for async methods
src/coreclr/tools/aot/ILCompiler.ReadyToRun/Compiler/ReadyToRunTableManager.cs	Updates module resolution to use GetPrimaryMethodDesc
src/coreclr/tools/aot/ILCompiler.ReadyToRun/Compiler/ReadyToRunCodegenCompilation.cs	Updates async variant assertion to use new helper method
src/coreclr/tools/aot/ILCompiler.ReadyToRun/Compiler/DependencyAnalysis/ReadyToRunCodegenNodeFactory.cs	Adds IsPrimaryMethodDesc check in debug code and minor formatting fix
src/coreclr/tools/aot/ILCompiler.ReadyToRun/Compiler/DependencyAnalysis/ReadyToRun/SignatureBuilder.cs	Adds AsyncVariant flag emission in method signatures
src/coreclr/tools/aot/ILCompiler.ReadyToRun/Compiler/DependencyAnalysis/ReadyToRun/MethodFixupSignature.cs	Replaces unboxing check with IsPrimaryMethodDesc for fixup optimization
src/coreclr/tools/aot/ILCompiler.ReadyToRun/Compiler/DependencyAnalysis/ReadyToRun/GCRefMapBuilder.cs	Adds async continuation parameter to GC reference map generation
src/coreclr/tools/aot/ILCompiler.ReadyToRun/Compiler/DependencyAnalysis/ReadyToRun/ArgIterator.cs	Implements async continuation argument offset calculation and allocation logic
src/coreclr/tools/Common/TypeSystem/IL/Stubs/AsyncResumptionStub.cs	Exposes TargetMethod property for GetPrimaryMethodDesc navigation
src/coreclr/tools/Common/Internal/Runtime/ReadyToRunConstants.cs	Changes ReadyToRunMethodSigFlags from byte to uint and adds AsyncVariant flag
src/coreclr/tools/Common/Compiler/AsyncMethodVariant.cs	Adds GetAsyncVariant and GetTargetOfAsyncVariant extension methods

src/coreclr/tools/aot/ILCompiler.ReadyToRun/JitInterface/CorInfoImpl.ReadyToRun.cs

...lr/tools/aot/ILCompiler.ReadyToRun/Compiler/DependencyAnalysis/ReadyToRun/GCRefMapBuilder.cs

Copilot

Copilot encountered an error and was unable to review this pull request. You can try again by re-requesting a review.

Copilot

Pull request overview

Copilot reviewed 16 out of 16 changed files in this pull request and generated no new comments.

jkotas · 2025-12-19T03:23:39Z

/azp run runtime-coreclr crossgen2 outerloop

azure-pipelines · 2025-12-19T03:23:57Z

Azure Pipelines successfully started running 1 pipeline(s).

… token

jtschuster · 2025-12-19T23:29:31Z

/azp run runtime-coreclr crossgen2 outerloop

azure-pipelines · 2025-12-19T23:29:43Z

Azure Pipelines successfully started running 1 pipeline(s).

jtschuster · 2026-01-05T19:08:33Z

/azp run runtime-coreclr crossgen2 outerloop

azure-pipelines · 2026-01-05T19:08:55Z

Azure Pipelines successfully started running 1 pipeline(s).

src/coreclr/vm/frames.cpp

src/coreclr/tools/aot/ILCompiler.ReadyToRun/JitInterface/CorInfoImpl.ReadyToRun.cs

azure-pipelines · 2026-01-11T00:25:35Z

Azure Pipelines successfully started running 1 pipeline(s).

jkotas · 2026-01-11T07:20:09Z

src/coreclr/tools/Common/TypeSystem/IL/Stubs/AsyncThunks.cs

+                    codestream.EndTry(tryFilterRegion);
                }
-                // Catch
+                // Filter - check if the caught object is a System.Exception


Should we make this workaround for R2R encoding limitation specific to R2R to avoid leaving perf on the table in NAOT?

What would it take to fix the encoding limitation in R2R?

Should we make this workaround for R2R encoding limitation specific to R2R to avoid leaving perf on the table in NAOT?

Good idea, I'll add an option to switch between using a filter or typed catch.

What would it take to fix the encoding limitation in R2R?

It looks like crossgen encodes EH info as CORINFO_EH_CLAUSE, which is also in a few places in the runtime.

runtime/src/coreclr/tools/Common/JitInterface/CorInfoTypes.cs

Lines 671 to 684 in cd5f273

public struct CORINFO_EH_CLAUSE

{

public CORINFO_EH_CLAUSE_FLAGS Flags;

public uint TryOffset;

public uint TryLength;

public uint HandlerOffset;

public uint HandlerLength;

public uint ClassTokenOrOffset;

/* union

{

DWORD ClassToken; // use for type-based exception handlers

DWORD FilterOffset; // use for filter-based exception handlers (COR_ILEXCEPTION_FILTER is set)

};*/

}

runtime/src/coreclr/inc/readytorun.h

Lines 482 to 493 in cd5f273

struct READYTORUN_EXCEPTION_CLAUSE

{

CorExceptionFlag Flags;

DWORD TryStartPC;

DWORD TryEndPC;

DWORD HandlerStartPC;

DWORD HandlerEndPC;

union {

mdToken ClassToken;

DWORD FilterOffset;

};

};

runtime/src/coreclr/inc/corcompile.h

Lines 142 to 153 in cd5f273

struct CORCOMPILE_EXCEPTION_CLAUSE

{

CorExceptionFlag Flags;

DWORD TryStartPC;

DWORD TryEndPC;

DWORD HandlerStartPC;

DWORD HandlerEndPC;

union {

mdToken ClassToken;

DWORD FilterOffset;

};

};

runtime/src/coreclr/inc/eexcp.h

Lines 18 to 31 in cd5f273

struct EE_ILEXCEPTION_CLAUSE {

//Flags is not marked as volatile since it is always accessed

// from within a critical section

CorExceptionFlag Flags;

DWORD TryStartPC;

DWORD TryEndPC;

DWORD HandlerStartPC;

DWORD HandlerEndPC;

union {

void* TypeHandle;

mdToken ClassToken;

DWORD FilterOffset;

};

};

We could add a ModuleOverride flag or something similar for the R2R struct. Or we could make a class token of 0 represent "catch anything".

The token in EH clause from R2R image is resolved here:

runtime/src/coreclr/vm/codeman.cpp

Lines 6494 to 6500 in 3a11eaf

Module* pModule = pMD->GetModule();

_ASSERTE(pModule != NULL);

SigTypeContext typeContext(pMD);

mdToken typeTok = pEHClause->ClassToken;

return ClassLoader::LoadTypeDefOrRefOrSpecThrowing(pModule, typeTok, &typeContext,

ClassLoader::ReturnNullIfNotFound);

We just need to ensure that the encoding produced by crossgen2 can be resolved in this method.

Or we could make a class token of 0 represent "catch anything".

Yes, it would be an option. There is an interesting detail in how to handle exceptions that do not inherit from System.Exception. These exceptions may or may not get wrapped as RuntimeWrappedException depending on per-assembly settings. What is the behavior we want for these stubs? Do we want them to be wrapped unconditionally to compat with async v1?

I see that you have added code in the filter to specifically handle this case. Do we have any test for it?

It doesn't look like we have tests for throwing non-exceptions.

It looks like async state machines catch System.Exception, not object.

In the stubs, the catch block passes the exception to TaskFromException(), so I'd expect we should rethrow / not catch anything that isn't a System.Exception.

We also could consider a new CorExceptionFlag enum value for "catch System.Exception".

The goal here is to match async v1 behavior. This implies that we need to test/validate the behavior of traditional async with regards to the RuntimeWrappedException behavior and match up to whatever it is (for cases where the async method is defined in an assemblies with both of the relevant options. Sounds like a test hole we need to fill.

I've opened #123194 to track adding tests for this

It looks like async state machines catch System.Exception, not object.

Do you mean async v1 state machines? They catch System.Exception in corelib. Corelib is marked with RuntimeCompatibilityAttribute(WrapNonExceptionThrows = true). It means they will catch non-exceptions wrapped in RuntimeWrappedException.

I don't think I fully understand all the implications of WrapNonExceptionThrows on async v1 vs async v2. Do we want a passing test suite as a prereq for merging this PR? And do you have a preference for adding a CorExceptionFlag or treating ClassToken == 0 as System.Exception, or finding another solution? I lean towards using a new CorExceptionFlag for R2R.

Do we want a passing test suite as a prereq for merging this PR?

David opened #123194 on it. It should not block this PR. I assume we may have problems in the existing JIT version as well.

I lean towards using a new CorExceptionFlag for R2R.

Fine with me.

…kReturningThunks

jtschuster · 2026-01-13T00:52:24Z

/azp run runtime-coreclr crossgen2 outerloop

azure-pipelines · 2026-01-13T00:52:42Z

Azure Pipelines successfully started running 1 pipeline(s).

The issue I noted is gone

With these changes, we no longer need separate IL for nativeaot and readytorun

jtschuster · 2026-01-15T01:43:21Z

/azp run runtime-coreclr crossgen2 outerloop

azure-pipelines · 2026-01-15T01:43:33Z

Azure Pipelines successfully started running 1 pipeline(s).

jtschuster · 2026-01-15T18:13:46Z

/azp run runtime-coreclr crossgen2 outerloop

azure-pipelines · 2026-01-15T18:14:08Z

Azure Pipelines successfully started running 1 pipeline(s).

* Rework async thunks generation to cross module generics infra - Add ability for code generating an ILStub to mark that the tokens might be generated - Funnel through various details to ensure that we trigger creation of the ManifestModuleWrappedMethodIL as needed - Adjust method fixup signature generation to allow for VersionsWithMethodBody code which is in the MutableModule - Add a few todos about work we might do to allow cross module compilation of runtime async code. - Disable this for now though, since its much to complex to test without a working end-to-end scenario (See code in CrossModuleInlineableUncached * Revert logic which attempts to treat System.Exception in EH clause specially

jtschuster · 2026-01-15T23:58:52Z

/azp run runtime-coreclr crossgen2 outerloop

azure-pipelines · 2026-01-15T23:59:03Z

Azure Pipelines successfully started running 1 pipeline(s).

src/coreclr/tools/aot/ILCompiler.ReadyToRun/JitInterface/CorInfoImpl.ReadyToRun.cs

…foImpl.ReadyToRun.cs Co-authored-by: Jan Kotas <jkotas@microsoft.com>

…Token == 0

…kReturningThunks

jtschuster · 2026-01-16T21:29:05Z

/azp run runtime-coreclr crossgen2 outerloop

azure-pipelines · 2026-01-16T21:29:25Z

Azure Pipelines successfully started running 1 pipeline(s).

jtschuster added 5 commits December 18, 2025 10:03

Handle continuation parameter in R2R GCRefMaps

0f2cdb5

Add helper extension methods

5be88cb

Emit Task-returning thunks in crossgen2

d048346

Merge branch 'main' of https://github.com/dotnet/runtime into MethodW…

e947ed4

…ithToken

Undo additional changes

b779849

jtschuster requested review from davidwrighton, jakobbotsch and jkotas December 18, 2025 20:58

jtschuster requested a review from MichalStrehovsky as a code owner December 18, 2025 20:58

Copilot AI review requested due to automatic review settings December 18, 2025 20:58

github-actions bot added the area-crossgen2-coreclr label Dec 18, 2025

Copilot started reviewing on behalf of jtschuster December 18, 2025 20:58 View session

dotnet-policy-service bot assigned jtschuster Dec 18, 2025

Copilot AI reviewed Dec 18, 2025

View reviewed changes

src/coreclr/tools/aot/ILCompiler.ReadyToRun/JitInterface/CorInfoImpl.ReadyToRun.cs Outdated Show resolved Hide resolved

...lr/tools/aot/ILCompiler.ReadyToRun/Compiler/DependencyAnalysis/ReadyToRun/GCRefMapBuilder.cs Show resolved Hide resolved

Copilot AI reviewed Dec 18, 2025

View reviewed changes

am11 added the runtime-async label Dec 18, 2025

jtschuster force-pushed the EmitTaskReturningThunks branch from 54e7d22 to b779849 Compare December 18, 2025 21:48

jtschuster requested a review from Copilot December 18, 2025 21:48

Copilot started reviewing on behalf of jtschuster December 18, 2025 21:49 View session

Copilot AI reviewed Dec 18, 2025

View reviewed changes

Ensure type doesn't version with module group before finding external…

3f87ad6

… token

jakobbotsch reviewed Jan 6, 2026

View reviewed changes

src/coreclr/vm/frames.cpp Show resolved Hide resolved

davidwrighton previously requested changes Jan 6, 2026

View reviewed changes

src/coreclr/tools/aot/ILCompiler.ReadyToRun/JitInterface/CorInfoImpl.ReadyToRun.cs Outdated Show resolved Hide resolved

Punt on generic async thunks, skip all async-variant methods

7df757a

build-analysis bot mentioned this pull request Jan 11, 2026

System.Transactions.Tests.OleTxTests.Recovery timeout #122621

Open

jkotas reviewed Jan 11, 2026

View reviewed changes

jtschuster added 2 commits January 11, 2026 09:00

Use class token EHInfo in NAOT, filter in R2R

03ace75

Merge branch 'main' of https://github.com/dotnet/runtime into EmitTas…

31ea783

…kReturningThunks

davidwrighton self-requested a review January 14, 2026 15:39

jtschuster mentioned this pull request Jan 14, 2026

MethodWithToken doesn't resolve the correct OwningType for methods on generic types. #123190

Open

davidwrighton mentioned this pull request Jan 14, 2026

Runtime async needs tests for catching non-exceptions which are thrown #123194

Open

Emit null token to represent catching a System.Exception

c87316a

With these changes, we no longer need separate IL for nativeaot and readytorun

jtschuster added 2 commits January 14, 2026 18:02

Undo extra ClassToken == 0 check

3ee2e57

Add the emission logic in crossgen

04fcdac

jkotas mentioned this pull request Jan 15, 2026

Composite-r2r with interpreter fallback leads to significant IL_STUB interpretation if FEATURE_JIT is disabled #123120

Open

3 tasks

jkotas reviewed Jan 16, 2026

View reviewed changes

src/coreclr/tools/aot/ILCompiler.ReadyToRun/JitInterface/CorInfoImpl.ReadyToRun.cs Outdated Show resolved Hide resolved

jtschuster and others added 4 commits January 16, 2026 08:17

Update src/coreclr/tools/aot/ILCompiler.ReadyToRun/JitInterface/CorIn…

ba0405f

…foImpl.ReadyToRun.cs Co-authored-by: Jan Kotas <jkotas@microsoft.com>

Encode/decode ClassToken of 0 as System.Exception for async thunks

5586607

Use COR_ILEXCEPTION_CLAUSE_R2R_SYSTEM_EXCEPTION flag instead of Class…

7551267

…Token == 0

Merge branch 'main' of https://github.com/dotnet/runtime into EmitTas…

20801c9

…kReturningThunks

build-analysis bot mentioned this pull request Jan 16, 2026

BinderTracingTest.ResolutionFlow times out #104670

Open

	public struct CORINFO_EH_CLAUSE
	{
	public CORINFO_EH_CLAUSE_FLAGS Flags;
	public uint TryOffset;
	public uint TryLength;
	public uint HandlerOffset;
	public uint HandlerLength;
	public uint ClassTokenOrOffset;
	/* union
	{
	DWORD ClassToken; // use for type-based exception handlers
	DWORD FilterOffset; // use for filter-based exception handlers (COR_ILEXCEPTION_FILTER is set)
	};*/
	}

	struct READYTORUN_EXCEPTION_CLAUSE
	{
	CorExceptionFlag Flags;
	DWORD TryStartPC;
	DWORD TryEndPC;
	DWORD HandlerStartPC;
	DWORD HandlerEndPC;
	union {
	mdToken ClassToken;
	DWORD FilterOffset;
	};
	};

	struct CORCOMPILE_EXCEPTION_CLAUSE
	{
	CorExceptionFlag Flags;
	DWORD TryStartPC;
	DWORD TryEndPC;
	DWORD HandlerStartPC;
	DWORD HandlerEndPC;
	union {
	mdToken ClassToken;
	DWORD FilterOffset;
	};
	};

	struct EE_ILEXCEPTION_CLAUSE {
	//Flags is not marked as volatile since it is always accessed
	// from within a critical section
	CorExceptionFlag Flags;
	DWORD TryStartPC;
	DWORD TryEndPC;
	DWORD HandlerStartPC;
	DWORD HandlerEndPC;
	union {
	void* TypeHandle;
	mdToken ClassToken;
	DWORD FilterOffset;
	};
	};

	Module* pModule = pMD->GetModule();
	_ASSERTE(pModule != NULL);

	SigTypeContext typeContext(pMD);
	mdToken typeTok = pEHClause->ClassToken;
	return ClassLoader::LoadTypeDefOrRefOrSpecThrowing(pModule, typeTok, &typeContext,
	ClassLoader::ReturnNullIfNotFound);

Emit task returning thunks in crossgen2 #122651

Are you sure you want to change the base?

Emit task returning thunks in crossgen2 #122651

Uh oh!

Conversation

jtschuster commented Dec 18, 2025

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Reviewed changes

Uh oh!

Uh oh!

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Uh oh!

jkotas commented Dec 19, 2025

Uh oh!

azure-pipelines bot commented Dec 19, 2025

Uh oh!

jtschuster commented Dec 19, 2025

Uh oh!

azure-pipelines bot commented Dec 19, 2025

Uh oh!

jtschuster commented Jan 5, 2026

Uh oh!

azure-pipelines bot commented Jan 5, 2026

Uh oh!

Uh oh!

Uh oh!

azure-pipelines bot commented Jan 11, 2026

Uh oh!

Choose a reason for hiding this comment

Uh oh!

jtschuster Jan 11, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

jtschuster commented Jan 13, 2026

Uh oh!

azure-pipelines bot commented Jan 13, 2026

Uh oh!

jtschuster commented Jan 15, 2026

Uh oh!

azure-pipelines bot commented Jan 15, 2026

Uh oh!

jtschuster commented Jan 15, 2026

Uh oh!

azure-pipelines bot commented Jan 15, 2026

Uh oh!

jtschuster commented Jan 15, 2026

Uh oh!

azure-pipelines bot commented Jan 15, 2026

Uh oh!

Uh oh!

jtschuster commented Jan 16, 2026

Uh oh!

azure-pipelines bot commented Jan 16, 2026

Uh oh!

Reviewers

Assignees

jtschuster Jan 11, 2026 •

edited

Loading