NativeAOT app crashes when run on a machine with an ETW session with GCHeapDump and Type keywords enabled

[elinor-fung]

1. Start an ETW session, enabling the .NET runtime provider with GCHeapDump and Type keywords
   • For example, collect with PerfView and set additional providers to Microsoft-Windows-DotNETRuntime:GCHeapDumpKeyword|TypeKeyword:Verbose
2. Run a nativeaot app that has <EventSourceSupport>true</EventSourceSupport> and does a GC (explicit call or otherwise)
   • The ETW session is for the machine, so regardless of whether whoever started the session cares about the nativeaot app, the app is affected.

App crashes trying to get the type manager from a method table. The runtime is trying to use the GCStaticEEType method table, which is intentionally not fully populated.

> helloworld.exe!TypeManagerHandle::AsTypeManager() Line 105
  helloworld.exe!BulkTypeEventLogger::LogSingleType(MethodTable * pEEType) Line 252
  helloworld.exe!BulkTypeEventLogger::LogTypeAndParameters(unsigned __int64 thAsAddr, ETW::TypeSystemLog::TypeLogBehavior typeLogBehavior) Line 54
  helloworld.exe!ETW::TypeSystemLog::LogTypeAndParametersIfNecessary(BulkTypeEventLogger * pLogger, unsigned __int64 thAsAddr, ETW::TypeSystemLog::TypeLogBehavior typeLogBehavior) Line 357
  helloworld.exe!ETW::GCLog::ObjectReference(ProfilerWalkHeapContext * profilerWalkHeapContext, Object * pObjReferenceSource, unsigned __int64 typeID, unsigned __int64 cRefs, Object * * rgObjReferenceTargets) Line 1046
  helloworld.exe!HeapWalkHelper(Object * pBO, void * pvContext) Line 188
  helloworld.exe!WKS::gc_heap::walk_heap_per_heap(bool(*)(Object *, void *) fn, void * context, int gen_number, int walk_large_object_heap_p) Line 51537
  helloworld.exe!WKS::gc_heap::walk_heap(bool(*)(Object *, void *) fn, void * context, int gen_number, int walk_large_object_heap_p) Line 51563
  helloworld.exe!WKS::GCHeap::DiagWalkHeap(bool(*)(Object *, void *) fn, void * context, int gen_number, bool walk_large_object_heap_p) Line 51617
  helloworld.exe!GCProfileWalkHeapWorker(int fShouldWalkHeapRootsForEtw, int fShouldWalkHeapObjectsForEtw) Line 897
  helloworld.exe!GCProfileWalkHeap() Line 927
  helloworld.exe!GCToEEInterface::DiagGCEnd(unsigned __int64 index, int gen, int reason, bool fConcurrent) Line 955
  helloworld.exe!WKS::gc_heap::do_post_gc() Line 49754

cc @LakshanF

Tagging subscribers to this area: @agocke, @MichalStrehovsky, @jkotas
See info in area-owners.md if you want to be subscribed.

Issue Details

---

1. Start an ETW session, enabling the .NET runtime provider with GCHeapDump and Type keywords
   • For example, collect with PerfView and set additional providers to Microsoft-Windows-DotNETRuntime:GCHeapDumpKeyword|TypeKeyword:Verbose
2. Run a nativeaot app that does a GC (explicit call or otherwise)
   • The ETW session is for the machine, so regardless of whether whoever started the session cares about the nativeaot app, the app is affected.

App crashes trying to get the type manager from a method table. The runtime is trying to use the GCStaticEEType method table, which is intentionally not fully populated.

> helloworld.exe!TypeManagerHandle::AsTypeManager() Line 105
  helloworld.exe!BulkTypeEventLogger::LogSingleType(MethodTable * pEEType) Line 252
  helloworld.exe!BulkTypeEventLogger::LogTypeAndParameters(unsigned __int64 thAsAddr, ETW::TypeSystemLog::TypeLogBehavior typeLogBehavior) Line 54
  helloworld.exe!ETW::TypeSystemLog::LogTypeAndParametersIfNecessary(BulkTypeEventLogger * pLogger, unsigned __int64 thAsAddr, ETW::TypeSystemLog::TypeLogBehavior typeLogBehavior) Line 357
  helloworld.exe!ETW::GCLog::ObjectReference(ProfilerWalkHeapContext * profilerWalkHeapContext, Object * pObjReferenceSource, unsigned __int64 typeID, unsigned __int64 cRefs, Object * * rgObjReferenceTargets) Line 1046
  helloworld.exe!HeapWalkHelper(Object * pBO, void * pvContext) Line 188
  helloworld.exe!WKS::gc_heap::walk_heap_per_heap(bool(*)(Object *, void *) fn, void * context, int gen_number, int walk_large_object_heap_p) Line 51537
  helloworld.exe!WKS::gc_heap::walk_heap(bool(*)(Object *, void *) fn, void * context, int gen_number, int walk_large_object_heap_p) Line 51563
  helloworld.exe!WKS::GCHeap::DiagWalkHeap(bool(*)(Object *, void *) fn, void * context, int gen_number, bool walk_large_object_heap_p) Line 51617
  helloworld.exe!GCProfileWalkHeapWorker(int fShouldWalkHeapRootsForEtw, int fShouldWalkHeapObjectsForEtw) Line 897
  helloworld.exe!GCProfileWalkHeap() Line 927
  helloworld.exe!GCToEEInterface::DiagGCEnd(unsigned __int64 index, int gen, int reason, bool fConcurrent) Line 955
  helloworld.exe!WKS::gc_heap::do_post_gc() Line 49754

cc @LakshanF

Author:	elinor-fung
Assignees:	-
Labels:	area-NativeAOT-coreclr
Milestone:	-

[agocke]

@VSadov could you look into this one?

[elinor-fung]

Note that after #88800, you also need to specify <EventSourceSupport>true</EventSourceSupport> in the app (updated the issue description) because firing ETW events was moved to no longer be enabled by default.

[LakshanF]

I think it would be OK to return false on these complex events for this release, given our position only to support basic events for now.

[agocke]

Looks like the crash is fixed. Work to enable is tracked in #88162

[elinor-fung]

I'll leave it to @LakshanF to decide how he'd like to track it, but I think it is worth noting that #88162 will be fixed naturally by switching to the generated scripts (generated Fire* functions replacing dotnetruntime.cpp will use both EventPipe and ETW - also generated - on Windows). However, the crash here was in trying to determine the information to include in the events - so there'll still be different work to sort that out.