Migrate CLR to COM stubs to be IL stubs by jkoritzinsky · Pull Request #126330 · dotnet/runtime

jkoritzinsky · 2026-03-30T20:36:48Z

Today, CLR->COM eventing and late binding stubs use a generic stub model, like COM->CLR stubs (being changed in #126002 as well). However, the CLR->COM cases have one important difference: they call into managed methods that exist with known signature shapes. Additionally, there's a decent amount of manually managed code written to support calling back into managed, as well as some diagonstics workarounds to avoid getting confused about a Managed-to-Unmanaged transition frame being on the stack in the same frame as managed code.

This PR converts CLR->COM eventing and late binding stubs to be IL stubs like early-bound COM stubs.

Eventing stubs are extremely simple, so they're implemented on top of the basic IL stub infrastructure.

Late binding stubs don't use marshalling in the IL stub itself, but they do depend on some of the marshalling logic. As a result, the late binding stubs do build on top of the interop IL stub infrastructure with some slight modifications.

A number of features were added to the IL stub infrastructure (namely, support for ldstr, newarr, some stelem instructions) to support calling the ForwardToInvokeMember call from an IL stub.

…ch stub

…ethod.

… diagnostics logic that's specific to it. Collapse StubState into ILStubState as there is no longer any other derived type. Also, move over the late-bound stub tracing step to work on the IL stub path.

dotnet-policy-service · 2026-03-30T20:38:06Z

Tagging subscribers to this area: @dotnet/interop-contrib
See info in area-owners.md if you want to be subscribed.

Copilot

Pull request overview

This PR migrates CLR-to-COM eventing and late-binding stubs away from the legacy “generic stub + CLRToCOMMethodFrame” model to IL stubs, aligning CLR-to-COM behavior with the existing IL stub infrastructure and reducing special-case debugger/stackwalk handling.

Changes:

Convert CLR→COM late-bound and eventing paths to IL stubs, removing CLRToCOMMethodFrame and the architecture-specific GenericCLRToCOMCallStub implementations.
Extend IL stub generation/resolution to support user strings (ldstr) and additional IL instructions (newarr, stelem.i1, stelem.i4) needed by the new stubs.
Simplify debugger/stackwalk logic by removing CLRToCOMMethodFrame-specific stepping/suppression workarounds.

Reviewed changes

Copilot reviewed 35 out of 35 changed files in this pull request and generated 2 comments.

Show a summary per file

File	Description
src/native/managed/cdac/Microsoft.Diagnostics.DataContractReader.Contracts/Contracts/StackWalk/FrameHandling/FrameIterator.cs	Removes `CLRToCOMMethodFrame` handling from CDAC stackwalking frame classification.
src/coreclr/vm/stubmgr.h	Updates comments to reflect removal of `GenericCLRToCOMCallStub` recognition.
src/coreclr/vm/stubmgr.cpp	Routes CLR→COM trace logic through IL-stub-based path and adds late-bound COM target resolution helper.
src/coreclr/vm/stubgen.h	Adds user-string token support and new IL emit helpers (`ldstr`, `newarr`, `stelem.*`).
src/coreclr/vm/stubgen.cpp	Implements `EmitLDSTR`, `EmitNEWARR`, `EmitSTELEM_I1/I4`, and user-string token plumbing.
src/coreclr/vm/prestub.cpp	Updates comment wording to “CLR-to-COM”.
src/coreclr/vm/method.hpp	Extends IL stub type enum and adjusts CLR→COM call info layout to support both IL stub + event provider MD.
src/coreclr/vm/ilstubresolver.h	Changes string-literal construction API to return `STRINGREF*`.
src/coreclr/vm/ilstubresolver.cpp	Implements IL-stub string literal support backed by new user-string token map.
src/coreclr/vm/ilstubcache.cpp	Tags COM event call stubs with a dedicated IL stub type for debugger behavior.
src/coreclr/vm/i386/asmhelpers.asm	Removes x86 `GenericCLRToCOMCallStub` implementation.
src/coreclr/vm/FrameTypes.h	Removes `CLRToCOMMethodFrame` from frame-type list.
src/coreclr/vm/frames.h	Removes `CLRToCOMMethodFrame` documentation and implementation declarations.
src/coreclr/vm/frames.cpp	Removes `CLRToCOMMethodFrame` implementation and related GC promotion logic.
src/coreclr/vm/dynamicmethod.h	Updates `DynamicResolver` interface to return `STRINGREF*` for string literals.
src/coreclr/vm/dynamicmethod.cpp	Updates `LCGMethodResolver` string literal construction to return `STRINGREF*`.
src/coreclr/vm/dllimport.cpp	Introduces `LateBoundCLRToCOM_ILStubState` and refactors IL-stub state abstraction; late-bound stubs now build argument arrays/types and call into managed reflection helper.
src/coreclr/vm/corelib.h	Switches COM interop method binder entries to `NoSig` for direct managed calls (non-UCO).
src/coreclr/vm/CMakeLists.txt	Removes obsolete sources/headers and ensures `clrtocomcall.*` remains correctly included for COM interop builds.
src/coreclr/vm/clrtocomcall.h	Removes obsolete debugger helper declarations tied to `CLRToCOMMethodFrame`.
src/coreclr/vm/clrtocomcall.cpp	Generates eventing stubs as IL stubs and unifies CLR→COM stub generation under IL stubs.
src/coreclr/vm/cgensys.h	Removes exports for `CLRToCOMWorker`/`GenericCLRToCOMCallStub`.
src/coreclr/vm/callsiteinspect.h	Deletes callsite inspection helper header (no longer needed).
src/coreclr/vm/callsiteinspect.cpp	Deletes callsite inspection helper implementation (no longer needed).
src/coreclr/vm/arm64/asmhelpers.asm	Removes ARM64 `GenericCLRToCOMCallStub`.
src/coreclr/vm/amd64/GenericCLRToCOMCallStubs.asm	Deletes AMD64 `GenericCLRToCOMCallStub` file.
src/coreclr/vm/amd64/cgencpu.h	Removes `COM_STUBS_SEPARATE_FP_LOCATIONS` define tied to old CLR→COM frame/stub model.
src/coreclr/System.Private.CoreLib/src/System/RuntimeType.CoreCLR.cs	Removes unmanaged-entry wrapper for `ForwardCallToInvokeMember` now that IL stubs call managed method directly.
src/coreclr/System.Private.CoreLib/src/System/__ComObject.cs	Removes unmanaged-entry wrapper for `GetEventProvider` now that IL stubs call managed method directly.
src/coreclr/debug/ee/frameinfo.h	Removes CLRToCOMMethodFrame-specific stepping suppression flag.
src/coreclr/debug/ee/frameinfo.cpp	Removes initialization and propagation of CLRToCOMMethodFrame-specific stepping suppression flag.
src/coreclr/debug/ee/controller.h	Removes `suppressUMChainFromCLRToCOMMethodFrameGeneric` parameter/state.
src/coreclr/debug/ee/controller.cpp	Removes CLRToCOMMethodFrame-specific stackwalk suppression logic and related comments.
src/coreclr/debug/daccess/dacdbiimpl.cpp	Removes `CLRToCOMMethodFrame` from illustrative frame list.
docs/design/datacontracts/StackWalk.md	Updates StackWalk contract docs to remove `CLRToCOMMethodFrame`.

Copilot · 2026-03-30T20:48:09Z

src/coreclr/vm/dllimport.cpp

+        STANDARD_VM_CONTRACT;
+        m_wrapperTypes.ReSizeThrows(m_signature.NumFixedArgs());
+


LateBoundCLRToCOM_ILStubState reads m_wrapperTypesNeeded in FinishEmit/EmitInvokeTarget, but the field is never initialized in the constructor. If a late-bound call has no wrapper types (or has 0 args), this can lead to nondeterministic IL generation (e.g., trying to load m_wrapperArrayLocalNum without creating the local). Initialize m_wrapperTypesNeeded (and ideally m_wrapperArrayLocalNum) to a known default (e.g., FALSE / LOCAL_NUM_UNUSED) in the ctor or member initializer list.

Copilot · 2026-03-30T20:48:09Z

src/coreclr/vm/ilstubresolver.cpp

-    _ASSERTE(FALSE);
-    return FALSE;
+
+    return TypeFromToken(metaTok) == mdtString && metaTok <= m_pCompileTimeState->m_tokenLookupMap.GetMaxUserStringToken();


ILStubResolver::IsValidStringRef treats any mdtString token <= GetMaxUserStringToken() as valid, but doesn’t reject RID 0 (token == mdtString), which is not a valid user-string token. This can make later code (e.g., LookupUserString preconditions) fail for edge tokens. Consider also requiring RidFromToken(metaTok) != 0 (and optionally metaTok >= TokenFromRid(1, mdtString)).

Suggested change

return TypeFromToken(metaTok) == mdtString && metaTok <= m_pCompileTimeState->m_tokenLookupMap.GetMaxUserStringToken();

if (TypeFromToken(metaTok) != mdtString)

return FALSE;

if (RidFromToken(metaTok) == 0)

return FALSE;

mdToken maxUserStringToken = m_pCompileTimeState->m_tokenLookupMap.GetMaxUserStringToken();

return metaTok >= TokenFromRid(1, mdtString) && metaTok <= maxUserStringToken;

AaronRobinsonMSFT · 2026-03-30T20:43:37Z

src/coreclr/vm/clrtocomcall.cpp

-        STANDARD_VM_CHECK;
+        THROWS;
+        GC_NOTRIGGER;
+        MODE_ANY;


Why is this multiple modes now? Is the no trigger to narrow impact for some reason?

This can be called from the StubManager to trace the target of the IL stub for the debugger. That's in (technically) COOP right now (something @hoyosjs is looking into).

GC_NOTRIGGER is to help with the problems he's seeing (deadlocks between stack walking and GC).

AaronRobinsonMSFT · 2026-03-30T20:44:18Z

src/coreclr/vm/clrtocomcall.cpp

+        );

-        if (strcmp(pMD->GetName(), "BeginInvoke") == 0)
+    #if defined(FEATURE_DYNAMIC_METHOD_HAS_NATIVE_STACK_ARG_SIZE)


Suggested change

#if defined(FEATURE_DYNAMIC_METHOD_HAS_NATIVE_STACK_ARG_SIZE)

#if defined(FEATURE_DYNAMIC_METHOD_HAS_NATIVE_STACK_ARG_SIZE)

AaronRobinsonMSFT · 2026-03-30T20:44:31Z

src/coreclr/vm/clrtocomcall.cpp

-    // Retrieve the method table and the method desc of the call.
-    MethodTable *pItfMT = pMD->GetInterfaceMethodTable();
-    CLRToCOMCallMethodDesc *pItfMD = pMD;
+    #endif


Suggested change

#endif

#endif // FEATURE_DYNAMIC_METHOD_HAS_NATIVE_STACK_ARG_SIZE

AaronRobinsonMSFT · 2026-03-30T20:46:21Z

src/coreclr/vm/clrtocomcall.cpp

-    LOG((LF_CORDB, LL_INFO10000,
-         "CLRToCOMMethodFrame::TraceFrame: ip=0x%p\n", ip));
+    PCODE pCode = JitILStub(*ppStubMD);
+    InterlockedCompareExchangeT<PCODE>(pComInfo->GetAddrOfILStubField(), pCode, NULL);


Do we ned to do any backout here if this fails to set?

AaronRobinsonMSFT · 2026-03-30T21:05:02Z

src/coreclr/vm/ilstubresolver.cpp


 STRINGREF ILStubResolver::GetStringLiteral(mdToken metaTok)
 {
    LIMITED_METHOD_CONTRACT;


This need to be marked GCX_COOP mode.

AaronRobinsonMSFT · 2026-03-30T21:07:45Z

src/coreclr/vm/stubmgr.cpp

+
+

Suggested change

AaronRobinsonMSFT · 2026-03-30T21:08:25Z

src/coreclr/vm/stubmgr.cpp

+    pUnk = ComObject::GetComIPFromRCWThrowing(&oref, pCLRToCOMCallInfo->m_pInterfaceMT);
+    GCPROTECT_END();
+
+    LPVOID *lpVtbl = *(LPVOID **)(IUnknown *)pUnk;


I'd expect an assert that shows a positive QI for IDispatch.

AaronRobinsonMSFT · 2026-03-30T21:09:25Z

src/coreclr/vm/dllimport.cpp

    FieldDesc *m_pFD;
 };
+
+class LateBoundCLRToCOM_ILStubState : public ILStubState


Suggested change

class LateBoundCLRToCOM_ILStubState : public ILStubState

class LateBoundCLRToCOM_ILStubState final : public ILStubState

AaronRobinsonMSFT · 2026-03-30T21:09:30Z

src/coreclr/vm/dllimport.cpp

+class LateBoundCLRToCOM_ILStubState : public ILStubState
+{
+public:
+


Suggested change

AaronRobinsonMSFT · 2026-03-30T21:17:58Z

src/coreclr/vm/ilstubresolver.cpp

+    SString& str = m_pCompileTimeState->m_tokenLookupMap.LookupUserString(metaTok);
+
+    LPCWSTR unicodeStr = str.GetUnicode();


If this is the contract, I would do this slightly different. I don't think exposing a reference to an SString is ever a good idea. In particular, the current approach means if another callers GetUTF8(), we have race and will have a problem. I assume that since callers are all working with tokens it means they are expecting UTF0-8, which means I would ensure the cached SString is properly in that encoding and then update the LookupUserString() API to return a LPCWSTR.

If the length is really important, make the length an out.

jkoritzinsky added 6 commits March 26, 2026 15:29

Migrate COM event stubs over to IL stubs from the generic late dispat…

8b47160

…ch stub

Treat CLRToCOMEvent stubs as step-through stubs to the target event m…

a617f5d

…ethod.

Add support for strings in IL stubs

2d5062b

Add various additional instructions needed for late bound stubs

f844e3b

Convert CLR->COM stubs to use an IL stub

0e184d2

Remove generic late-bound fallback, all code only used by it, and any…

990a016

… diagnostics logic that's specific to it. Collapse StubState into ILStubState as there is no longer any other derived type. Also, move over the late-bound stub tracing step to work on the IL stub path.

jkoritzinsky requested review from AaronRobinsonMSFT, Copilot and davidwrighton March 30, 2026 20:36

jkoritzinsky added os-windows area-Interop-coreclr labels Mar 30, 2026

dotnet-policy-service bot assigned jkoritzinsky Mar 30, 2026

github-project-automation bot added this to AppModel Mar 30, 2026

Copilot started reviewing on behalf of jkoritzinsky March 30, 2026 20:38 View session

Copilot AI reviewed Mar 30, 2026

View reviewed changes

AaronRobinsonMSFT reviewed Mar 30, 2026

View reviewed changes

build-analysis bot mentioned this pull request Mar 31, 2026

XHarness package install failure on iOS due to devicectl NSPOSIXErrorDomain error 49 #123796

Open

		STANDARD_VM_CONTRACT;
		m_wrapperTypes.ReSizeThrows(m_signature.NumFixedArgs());

-    return TypeFromToken(metaTok) == mdtString && metaTok <= m_pCompileTimeState->m_tokenLookupMap.GetMaxUserStringToken();
+    if (TypeFromToken(metaTok) != mdtString)
+        return FALSE;
+    if (RidFromToken(metaTok) == 0)
+        return FALSE;
+    mdToken maxUserStringToken = m_pCompileTimeState->m_tokenLookupMap.GetMaxUserStringToken();
+    return metaTok >= TokenFromRid(1, mdtString) && metaTok <= maxUserStringToken;

	#if defined(FEATURE_DYNAMIC_METHOD_HAS_NATIVE_STACK_ARG_SIZE)
	#if defined(FEATURE_DYNAMIC_METHOD_HAS_NATIVE_STACK_ARG_SIZE)

	#endif
	#endif // FEATURE_DYNAMIC_METHOD_HAS_NATIVE_STACK_ARG_SIZE

	class LateBoundCLRToCOM_ILStubState : public ILStubState
	class LateBoundCLRToCOM_ILStubState final : public ILStubState

		SString& str = m_pCompileTimeState->m_tokenLookupMap.LookupUserString(metaTok);

		LPCWSTR unicodeStr = str.GetUnicode();

Conversation

jkoritzinsky commented Mar 30, 2026

Uh oh!

dotnet-policy-service bot commented Mar 30, 2026

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Reviewed changes

Uh oh!

Copilot AI Mar 30, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Mar 30, 2026

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!

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!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants