Test failure Interop\\COM\\Activator\\ActivatorBuiltInComDisabled\\ActivatorBuiltInComDisabled.cmd

[VincentBu]

Run: runtime-coreclr gcstress0x3-gcstress0xc 20211124.1

Failed test:

CoreCLR windows x86 Checked gcstress0x3 @ Windows.10.Amd64.Open
- Interop\\COM\\Activator\\ActivatorBuiltInComDisabled\\ActivatorBuiltInComDisabled.cmd
- Interop\\COM\\Activator\\Activator\\Activator.cmd

CoreCLR windows x64 Checked gcstress0x3 @ Windows.10.Amd64.Open
- Interop\\COM\\Activator\\ActivatorBuiltInComDisabled\\ActivatorBuiltInComDisabled.cmd
- Interop\\COM\\Activator\\Activator\\Activator.cmd

CoreCLR windows x86 Checked gcstress0xc @ Windows.10.Amd64.Open
- Interop\\COM\\Activator\\ActivatorBuiltInComDisabled\\ActivatorBuiltInComDisabled.cmd
- Interop\\COM\\Activator\\Activator\\Activator.cmd

CoreCLR windows x64 Checked gcstress0xc @ Windows.10.Amd64.Open
- Interop\\COM\\Activator\\ActivatorBuiltInComDisabled\\ActivatorBuiltInComDisabled.cmd
- Interop\\COM\\Activator\\Activator\\Activator.cmd

Error message:

Assert failure(PID 3468 [0x00000d8c], Thread: 4896 [0x1320]): !th.IsNull()

CORECLR! SigPointer::GetTypeHandleThrowing + 0x620 (0x70dce98d)
CORECLR! SigPointer::PeekElemTypeNormalized + 0x1D7 (0x70dd2f32)
CORECLR! MetaSig::PeekArgNormalized + 0x20 (0x70dd2aae)
CORECLR! ArgIteratorTemplate<ArgIteratorBase>::ForceSigWalk + 0x189 (0x70c1b603)
CORECLR! ArgIteratorTemplate<ArgIteratorBase>::SizeOfArgStack + 0x10 (0x70c372cb)
CORECLR! ArgIteratorTemplate<ArgIteratorBase>::GetNextOffset + 0x66 (0x70c36338)
CORECLR! TransitionFrame::PromoteCallerStackHelper + 0x113 (0x70d0ba35)
CORECLR! TransitionFrame::PromoteCallerStack + 0xEE (0x70d0b91b)
CORECLR! GcStackCrawlCallBack + 0x1EA (0x70ec3c1a)
CORECLR! Thread::MakeStackwalkerCallback + 0x48 (0x70dd5e83)
File: D:\a_work\1\s\src\coreclr\vm\siginfo.cpp Line: 1112
Image: C:\h\w\A93B08DB\p\corerun.exe

Return code:      1
Raw output file:      C:\h\w\A93B08DB\w\A06008C3\uploads\Reports\Interop.COM\Activator\ActivatorBuiltInComDisabled\ActivatorBuiltInComDisabled.output.txt
Raw output:
BEGIN EXECUTION
"C:\h\w\A93B08DB\p\corerun.exe" -p "System.Runtime.InteropServices.BuiltInComInterop.IsSupported=false" -p "System.Reflection.Metadata.MetadataUpdater.IsSupported=false"  ActivatorBuiltInComDisabled.dll
Built-in COM Disabled?: True
Running InvalidInterfaceRequest...
Expected: 100
Actual: -1073740286
END EXECUTION - FAILED
FAILED
Test Harness Exitcode is : 1
To run the test:

set CORE_ROOT=C:\h\w\A93B08DB\p
C:\h\w\A93B08DB\w\A06008C3\e\Interop\COM\Activator\ActivatorBuiltInComDisabled\ActivatorBuiltInComDisabled.cmd
Expected: True
Actual:   False


Stack trace
   at Interop_COM._Activator_ActivatorBuiltInComDisabled_ActivatorBuiltInComDisabled_._Activator_ActivatorBuiltInComDisabled_ActivatorBuiltInComDisabled_cmd()

[AaronRobinsonMSFT]

This appears to have uncovered an issue with the changes in #61754. It seems the type system is failing to preload the type, but honestly I don't fully understand this part of the codebase.

runtime/src/coreclr/vm/siginfo.cpp

Lines 1095 to 1114 in 39b4f26

	#ifdef _DEBUG_IMPL
	// This verifies that we won't try and load a type
	// if FORBIDGC_LOADER_USE_ENABLED is true.
	//
	// The FORBIDGC_LOADER_USE is limited to very specific scenarios that need to retrieve
	// GC_OTHER typehandles for size and gcroot information. This assert attempts to prevent
	// this abuse from proliferating.
	//
	if (FORBIDGC_LOADER_USE_ENABLED() && (fLoadTypes == ClassLoader::LoadTypes))
	{
	TypeHandle th = GetTypeHandleThrowing(pModule,
	pTypeContext,
	ClassLoader::DontLoadTypes,
	level,
	dropGenericArgumentLevel,
	pSubst,
	pZapSigContext);
	_ASSERTE(!th.IsNull());
	}
	#endif

@jkotas or @davidwrighton Any thoughts on why the above would trigger for GCStress when during the JIT of GetClassFactoryForType?

runtime/src/tests/Interop/COM/Activator/Program.cs

Lines 55 to 61 in 39b4f26

	private static delegate*<ComActivationContext, object> GetClassFactoryForTypeMethod = (delegate*<ComActivationContext, object>)typeof(object).Assembly.GetType("Internal.Runtime.InteropServices.ComActivator", throwOnError: true).GetMethod("GetClassFactoryForType", BindingFlags.NonPublic | BindingFlags.Static).MethodHandle.GetFunctionPointer();
	private static delegate*<ComActivationContext, bool, void> ClassRegistrationScenarioForType = (delegate*<ComActivationContext, bool, void>)typeof(object).Assembly.GetType("Internal.Runtime.InteropServices.ComActivator", throwOnError: true).GetMethod("ClassRegistrationScenarioForType", BindingFlags.NonPublic | BindingFlags.Static).MethodHandle.GetFunctionPointer();
	private static ClassFactoryWrapper GetClassFactoryForType(ComActivationContext context)
	{
	return new ClassFactoryWrapper(GetClassFactoryForTypeMethod(context));
	}

[jkotas]

It looks like a bug in RuntimeMethodHandle_GetFunctionPointer and potentially other places that hand back function pointers to unsafe code. The runtime requires that all types in method signatures are loaded before the method is called. This invariant is violated here.

Can you try to add pMethod->PrepareForUseAsADependencyOfANativeImage() to RuntimeMethodHandle_GetFunctionPointer to see whether it fixes the problem?

If it does, we should introduce a new method (e.g. called PrepareForUseAsAFunctionPointer) that is just like PrepareForUseAsADependencyOfANativeImage except that it does not catch exceptions, and call it there.

[AaronRobinsonMSFT]

That doesn't seem to do anything for this case. From what I can tell all it does it resolve to the correct TypeDef Token. Looking at other uses, Type Equivalence, I need to also load the type so I am trying that now.