Unify StubLinker stub memory management with JIT code allocator

[Copilot]

StubLinker-generated stubs used a dedicated LoaderAllocator::m_pStubHeap, separate from the executable memory infrastructure shared by JITed code and other stub kinds. This PR unifies the two paths.

• LoaderAllocator: removed m_pStubHeap and its init/terminate/DAC/size-reporting plumbing.
• StubLinker/Stub: StubLinker::Link and Stub::NewStub now take a LoaderAllocator* instead of a LoaderHeap*, allocating stub memory through ExecutionManager::GetEEJitManager()->AllocCodeFragmentBlock.
• StubCodeBlockKind: added STUB_CODE_BLOCK_STUBLINK and STUB_CODE_BLOCK_SHUFFLE_THUNK to distinguish these stubs in the code-fragment allocator.
• StubLinkStubManager: removed its own range list; RangeSectionStubManager now recognizes the new stub kinds and forwards stub identification/tracing to StubLinkStubManager (mirroring the existing pattern for jump stubs and call-counting stubs).
• DAC/SOS: updated datadescriptor.inc and debug/daccess/request.cpp to drop the removed stub heap field.
• Call sites: updated stubcache.cpp, comdelegate.cpp, prestub.cpp, assembly.cpp, appdomain.hpp/.inl, and class.h to pass LoaderAllocator* instead of a stub LoaderHeap*.

// Before
Stub *Link(LoaderHeap *heap, DWORD flags, const char *stubType);

// After
Stub *Link(LoaderAllocator *pLoaderAllocator, DWORD flags, const char *stubType);

[azure-pipelines]

Azure Pipelines:
Successfully started running 3 pipeline(s).
13 pipeline(s) were filtered out due to trigger conditions.
There may be pipelines that require an authorized user to comment /azp run to run.

[Copilot]

Pull request overview

This PR refactors CoreCLR stub generation so StubLinker-produced stubs (including shuffle thunks) allocate executable memory via the JIT’s code-fragment allocator rather than a dedicated LoaderAllocator::m_pStubHeap, and updates stub identification/tracing plus DAC descriptors accordingly.

Changes:

• Reworks StubLinker::Link / Stub::NewStub to allocate stub memory via EEJitManager::AllocCodeFragmentBlock, keyed by new StubCodeBlockKind values.
• Removes LoaderAllocator’s dedicated stub heap field/plumbing and updates call sites to pass LoaderAllocator*.
• Updates stub manager routing (RangeSectionStubManager ⇄ StubLinkStubManager) and cDAC descriptors for the removed stub heap field.

Reviewed changes

Copilot reviewed 19 out of 19 changed files in this pull request and generated 3 comments.

Show a summary per file

File	Description
src/coreclr/vm/stubmgr.h	Removes StubLinkStubManager’s range list plumbing (moves identification to range-section kind).
src/coreclr/vm/stubmgr.cpp	Routes STUBLINK/SHUFFLE kinds through RangeSectionStubManager and forwards tracing to StubLinkStubManager.
src/coreclr/vm/stublink.h	Changes StubLinker::Link to take LoaderAllocator*; updates Stub::NewStub signature.
src/coreclr/vm/stublink.cpp	Allocates stub memory using AllocCodeFragmentBlock and tags with new stub kinds.
src/coreclr/vm/stubcache.h	Updates stub cache to retain a LoaderAllocator* instead of a LoaderHeap*.
src/coreclr/vm/stubcache.cpp	Passes LoaderAllocator* through to StubLinker::Link.
src/coreclr/vm/prestub.cpp	Updates stub linker call sites to pass LoaderAllocator*.
src/coreclr/vm/loaderallocator.hpp	Removes m_pStubHeap and its accessor; updates cDAC offsets accordingly.
src/coreclr/vm/loaderallocator.cpp	Removes stub heap initialization/termination/enumeration/size accounting; adjusts shuffle thunk cache initialization.
src/coreclr/vm/datadescriptor/datadescriptor.inc	Drops the LoaderAllocator.StubHeap cDAC field.
src/coreclr/vm/comdelegate.h	Updates ShuffleThunkCache to use LoaderAllocator*.
src/coreclr/vm/comdelegate.cpp	Updates global shuffle thunk cache init; changes DelegateEEClass::GetStubHeap to return LoaderAllocator*.
src/coreclr/vm/codeman.h	Adds STUB_CODE_BLOCK_STUBLINK and STUB_CODE_BLOCK_SHUFFLE_THUNK kinds and string mappings.
src/coreclr/vm/class.h	Updates DelegateEEClass::GetStubHeap return type to LoaderAllocator*.
src/coreclr/vm/assembly.hpp	Removes Assembly::GetStubHeap declaration.
src/coreclr/vm/assembly.cpp	Removes Assembly::GetStubHeap implementation.
src/coreclr/vm/appdomain.hpp	Removes AppDomain::GetStubHeap declaration.
src/coreclr/vm/appdomain.inl	Removes AppDomain::GetStubHeap inline implementation.
src/coreclr/debug/daccess/request.cpp	Updates DAC exposure of “stub heap” data and heap-name list for loader allocator heap enumeration.

[dotnet-policy-service]

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

[Copilot]

Pull request overview

Copilot reviewed 19 out of 19 changed files in this pull request and generated 1 comment.

Suppressed comments (3)

src/coreclr/vm/comdelegate.cpp:824

• The comment says this method returns a LoaderHeap, but it now returns a LoaderAllocator*. Updating the comment avoids misleading future readers about what lifetime guarantee is being relied on here.

// We need a LoaderHeap that lives at least as long as the DelegateEEClass, but ideally no longer
LoaderAllocator *DelegateEEClass::GetStubHeap()

src/coreclr/vm/class.h:1925

• This comment still refers to returning a LoaderHeap, but the method now returns a LoaderAllocator*. Keeping the comment accurate is important since this is the declaration most readers will see.

    // We need a LoaderHeap that lives at least as long as the DelegateEEClass, but ideally no longer
    LoaderAllocator *GetStubHeap();

src/coreclr/debug/daccess/request.cpp:3686

• LoaderAllocatorLoaderHeapNames is out of sync with GetLoaderAllocatorHeaps(): the names array includes both "ExecutableStubHeap" and "ExecutableHeap", but GetLoaderAllocatorHeaps() now only returns one executable heap pointer at that slot. This makes GetLoaderAllocatorHeapNames() report a different count/order than the returned heap addresses, which can confuse SOS/DAC consumers.

    "ExecutableStubHeap",
    "ExecutableHeap",
    "FixupPrecodeHeap",

[Copilot]

Pull request overview

Copilot reviewed 41 out of 41 changed files in this pull request and generated no new comments.

Suppressed comments (1)

src/coreclr/vm/eventtrace.cpp:3627

• StubInitialized now reports every helper as size 1, which makes ETW helper MethodLoad events effectively unusable (tools won’t be able to map IPs within the helper body to the helper name). We should report a realistic size (e.g., the write-barrier group size on x86, and at least a page-sized upper bound elsewhere) rather than a constant.

            DWORD dwHelperSize = 1;
            ETW::MethodLog::SendHelperEvent(ullHelperStartAddress, dwHelperSize, pHelperName);

[azure-pipelines]

Azure Pipelines:
Successfully started running 5 pipeline(s).
11 pipeline(s) were filtered out due to trigger conditions.
There may be pipelines that require an authorized user to comment /azp run to run.

[Copilot]

Pull request overview

Copilot reviewed 42 out of 42 changed files in this pull request and generated no new comments.

Suppressed comments (1)

src/coreclr/vm/eventtrace.cpp:3626

• StubInitialized now reports a hard-coded helper size of 1 byte, which looks like a placeholder after removing Stub::RecoverStubAndSize. This will produce incorrect ETW MethodLoad events (size is part of the payload) for write-barrier helpers and any other helpers that call into this path.

Consider restoring a real size calculation (e.g., change StubInitialized to take a size parameter and pass the known helper size from the call sites), or otherwise derive the helper size from the allocation metadata instead of a constant.

            DWORD dwHelperSize = 1;

[Copilot]

Pull request overview

Copilot reviewed 42 out of 42 changed files in this pull request and generated no new comments.

Suppressed comments (2)

src/coreclr/vm/stubcache.h:47

• StubCacheBase allows a default NULL loader allocator, but StubLinker::Link ultimately calls EEJitManager::AllocCodeFragmentBlock, which has a PRECONDITION(pLoaderAllocator != NULL). This makes the API foot-gun: any future (or accidental) default construction would crash/assert. Make the requirement explicit by removing the default argument (and optionally assert non-null in the constructor).

    StubCacheBase(LoaderAllocator *pLoaderAllocator = NULL);

src/coreclr/vm/eventtrace.cpp:3627

• dwHelperSize is now hardcoded to 1, so ETW MethodLoadVerbose will report every helper stub as a 1-byte method. That’s very likely misleading for diagnostics tooling that expects size/range information. If the size is no longer recoverable from a Stub header, consider plumbing the size from the call site (e.g., write-barrier helper generation knows sizes) or adding a helper to derive the real stub/code-fragment size from allocation metadata; if neither is possible, please add an explicit comment explaining why reporting size=1 is acceptable.

            DWORD dwHelperSize = 1;
            ETW::MethodLog::SendHelperEvent(ullHelperStartAddress, dwHelperSize, pHelperName);

[Copilot]

Pull request overview

Copilot reviewed 42 out of 42 changed files in this pull request and generated no new comments.

Suppressed comments (3)

src/coreclr/vm/codeman.h:131

• GetStubCodeBlockKindString no longer returns accurate names for several stub kinds: STUB_CODE_BLOCK_DYNAMICHELPER and STUB_CODE_BLOCK_FIXUPPRECODE both return "MethodCallThunk", and STUB_CODE_BLOCK_STUBPRECODE is not handled at all (falls through to "Unknown"). This breaks perfmap/event naming for these stubs and makes it harder to diagnose stub allocations.

    case STUB_CODE_BLOCK_JUMPSTUB:
        return "JumpStub";
    case STUB_CODE_BLOCK_DYNAMICHELPER:
        return "MethodCallThunk";
    case STUB_CODE_BLOCK_FIXUPPRECODE:
        return "MethodCallThunk";
#ifdef FEATURE_VIRTUAL_STUB_DISPATCH

src/coreclr/vm/eventtrace.cpp:3628

• StubInitialized now reports a hardcoded helper size of 1 byte. This value is emitted into FireEtwMethodLoadVerbose_V1 and will produce incorrect ETW method-load metadata for helper stubs (notably the write barrier helpers reported from threads.cpp). The size should either be computed from the actual stub allocation or passed in by the caller that knows the exact size.

        if(ETW_TRACING_CATEGORY_ENABLED(MICROSOFT_WINDOWS_DOTNETRUNTIME_PROVIDER_DOTNET_Context,
                                        TRACE_LEVEL_INFORMATION,
                                        CLR_JIT_KEYWORD))
        {
            DWORD dwHelperSize = 1;
            ETW::MethodLog::SendHelperEvent(ullHelperStartAddress, dwHelperSize, pHelperName);
        }

src/native/managed/cdac/Microsoft.Diagnostics.DataContractReader.Contracts/Contracts/ExecutionManager/ExecutionManagerCore.cs:107

• StubKind was extended with WrapperStub and ShuffleThunk, but GetStubKind(StubKind stubKind) (later in this file) does not map these new values to any CodeKind, so these stubs will be reported as CodeKind.Unknown via GetCodeKind when walking range-list stubs. Either map them to an existing CodeKind (e.g., MethodCallThunk) or extend CodeKind if tools need to distinguish them.

[rcj1]

Pass in actual write barrier size

[jkotas]

@copilot

[Copilot]

Implemented in 6b1d87d: StubInitialized now takes a helper size parameter, and the write barrier callers pass the computed write barrier/check write barrier code size.

Note

This comment was generated by GitHub Copilot.

[rcj1]

Extend GetStubKind with these new kinds

[jkotas]

@copilot