BENCH: Remove ExInfo::m_hThrowable — exception handling benchmarks by max-charlamb · Pull Request #127531 · dotnet/runtime

max-charlamb · 2026-04-28T21:24:51Z

Note

This PR was authored with the assistance of GitHub Copilot.

Benchmark-only PR for testing the performance impact of removing ExInfo::m_hThrowable. See #127300 for the main PR and full description.

Do not merge — this exists solely for EgorBot benchmarking.

Replace the GCHandle-based m_hThrowable field in ExInfo with direct use of the existing m_exception OBJECTREF field, matching NativeAOT's approach. Key changes: - Remove OBJECTHANDLE m_hThrowable from ExInfo, saving 8 bytes (64-bit) - Update AsmOffsets constants for the new field layout - Add GC root scanning of ExInfo chain in ScanStackRoots (gcenv.ee.cpp), mirroring NativeAOT's GcScanRootsWorker pattern - Simplify GetThrowable() to return m_exception directly - SetThrowable() no longer creates GC handles for ExInfo - Remove GetThrowableAsHandle() entirely — all callers migrated to use GetThrowable() (OBJECTREF) or m_LastThrownObjectHandle (real handle) - Update StackTraceInfo::AppendElement OBJECTREF overload to preserve foreign-exception semantics and preallocated exception checks - Update Interop propagation callback to take OBJECTREF instead of handle - Update DAC code (request.cpp, task.cpp, dacdbiimpl.cpp) to use m_exception directly - Update debugger code (eedbginterfaceimpl.cpp, debugger.cpp) to use m_LastThrownObjectHandle for handle-based APIs - Update cDAC: ThrownObjectHandle -> ThrownObject (direct pointer) - Update cDAC contracts, data classes, and tests This eliminates ~5 interlocked handle alloc/destroy ops per exception throw, removes OOM fallback paths, and unblocks cDAC unification. Thread::m_LastThrownObjectHandle remains as-is (separate work item). Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

The ExInfo::m_exception field was being reported to the GC twice: once via GCPROTECT_BEGIN and once via ExInfo chain scanning in ScanStackRoots. The CLR code guide (section 2.1.5) explicitly states that reporting the same location twice corrupts the GC's relocation logic. Remove the GCPROTECT_BEGIN/END for m_exception and rely solely on chain scanning (matching NativeAOT's model). Add Thread::ObjectRefProtected calls in checked builds to satisfy the debug OBJECTREF tracking table. Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

max-charlamb · 2026-04-28T21:25:10Z

@EgorBot -amd

using BenchmarkDotNet.Attributes;
using System;
using System.Runtime.CompilerServices;

public class Bench
{
    [Benchmark]
    public Exception ThrowCatch()
    {
        try { throw new Exception(); }
        catch (Exception ex) { return ex; }
        return null;
    }

    [Benchmark]
    public Exception NestedCatchThrowNew()
    {
        try { Nest(10); }
        catch (Exception ex) { return ex; }
        return null;
    }

    [MethodImpl(MethodImplOptions.NoInlining)]
    static void Nest(int depth)
    {
        if (depth == 0) throw new Exception();
        try { Nest(depth - 1); }
        catch (Exception ex) { throw new Exception("wrap", ex); }
    }
}

dotnet-policy-service · 2026-04-28T21:26:36Z

Tagging subscribers to this area: @steveisok, @tommcdon, @dotnet/dotnet-diag
See info in area-owners.md if you want to be subscribed.

Copilot

Pull request overview

Benchmark-focused variant of the main “remove ExInfo::m_hThrowable” change, updating CoreCLR exception tracking and diagnostic contracts (DAC/cDAC) to treat the thrown exception as a direct OBJECTREF (ExInfo::m_exception) and to expose pseudo-handles as “address of an OBJECTREF slot” where needed for debugger/DAC compatibility.

Changes:

Remove m_hThrowable usage across CoreCLR EH paths and switch consumers to m_exception / pseudo-handle semantics.
Teach GC root scanning to walk the ExInfo chain so superseded exception objects remain alive for diagnostics.
Update cDAC descriptors, contracts, and tests to use ThrownObject (direct pointer) and return field-address pseudo-handles.

Reviewed changes

Copilot reviewed 27 out of 27 changed files in this pull request and generated 2 comments.

Show a summary per file

File	Description
src/native/managed/cdac/tests/ThreadTests.cs	Updates tests for `GetCurrentExceptionHandle` to validate field-address pseudo-handle behavior.
src/native/managed/cdac/tests/MockDescriptors/MockDescriptors.Thread.cs	Renames mock field from `ThrownObjectHandle` to `ThrownObject` and adjusts layout accessors.
src/native/managed/cdac/Microsoft.Diagnostics.DataContractReader.Contracts/Data/ExceptionInfo.cs	Reads `ThrownObject` as a direct pointer instead of `ObjectHandle`.
src/native/managed/cdac/Microsoft.Diagnostics.DataContractReader.Contracts/Contracts/Thread_1.cs	Returns pseudo-handle (field address) for current exception and updates Watson bucket access.
src/native/managed/cdac/Microsoft.Diagnostics.DataContractReader.Contracts/Contracts/Exception_1.cs	Returns nested exception pseudo-handle as `&ThrownObject` and object pointer as `ThrownObject`.
src/coreclr/vm/threads.inl	Removes `Thread::SetThrowable` inline wrapper.
src/coreclr/vm/threads.h	Removes throwable-handle APIs, updates `HasException`/`IsThrowableNull` logic and signature.
src/coreclr/vm/threads.cpp	Refactors `SafeSetThrowables`/`SafeUpdateLastThrownObject` around last-thrown handle only.
src/coreclr/vm/interoplibinterface.h	Changes propagation callback signature from `OBJECTHANDLE` to `OBJECTREF`.
src/coreclr/vm/interoplibinterface_shared.cpp	Updates callback contract/implementation for `OBJECTREF` throwable.
src/coreclr/vm/interoplibinterface_objc.cpp	Removes handle-to-object conversion, expects `OBJECTREF` directly (coop mode).
src/coreclr/vm/gcenv.ee.cpp	Adds GC root scanning of the `ExInfo` chain (`m_exception` slots).
src/coreclr/vm/exstate.h	Removes throwable handle APIs and adds `IsThrowableNull`.
src/coreclr/vm/exstate.cpp	Removes `GetThrowableAsHandle` and implements `GetThrowable` via `m_exception`.
src/coreclr/vm/exinfo.h	Removes `m_hThrowable` field and related APIs; `GetThrowable()` returns `m_exception`.
src/coreclr/vm/exinfo.cpp	Removes handle lifecycle; marks `m_exception` protected for checked OBJECTREF tracking; clears `m_exception` on release.
src/coreclr/vm/exceptionhandling.cpp	Removes `GCPROTECT_BEGIN(exInfo.m_exception)` in dispatch paths; switches stack trace plumbing to pseudo-handle `&m_exception`.
src/coreclr/vm/excep.cpp	Updates throwable-null and preallocated-exception checks for pseudo-handle semantics.
src/coreclr/vm/eepolicy.cpp	Removes debug-only `SetThrowable` error-checking enum usage from SO fatal path.
src/coreclr/vm/eedbginterfaceimpl.cpp	Switches debugger exception retrieval to `m_LastThrownObjectHandle` with debug invariant checks.
src/coreclr/vm/datadescriptor/datadescriptor.inc	cDAC field `ThrownObjectHandle` → `ThrownObject` at `offsetof(ExInfo, m_exception)`.
src/coreclr/System.Private.CoreLib/src/System/Runtime/ExceptionServices/AsmOffsets.cs	Updates `ExInfo` field offsets (validated via `static_assert`).
src/coreclr/debug/ee/debugger.cpp	Uses `m_LastThrownObjectHandle` for catch-handler-found forcing table lookup.
src/coreclr/debug/daccess/task.cpp	Passes pseudo-handle as `&ExInfo::m_exception` to DAC exception state.
src/coreclr/debug/daccess/request.cpp	Reads current throwable via direct `OBJECTREF` instead of a handle.
src/coreclr/debug/daccess/dacimpl.h	Documents that “throwable handle” may be GC handle slot or `&ExInfo::m_exception` slot.
src/coreclr/debug/daccess/dacdbiimpl.cpp	Uses `m_LastThrownObjectHandle` for debugger-facing “current exception handle”.

Copilot · 2026-04-28T21:33:46Z

+    // Scan the ExInfo chain for exception objects held by direct pointer.
+    // Superseded ExInfo objects may live in logically dead parts of the stack
+    // that the normal GC stackwalk skips (e.g., when one exception dispatch
+    // supersedes a previous one). We keep them alive for post-mortem debugging
+    // and SOS. This mirrors NativeAOT's GcScanRootsWorker (thread.cpp:569-573).
+    PTR_ExInfo pExInfo = pThread->GetExceptionState()->GetCurrentExceptionTracker();
+    while (pExInfo != NULL)
+    {
+        PTR_PTR_Object pRef = dac_cast<PTR_PTR_Object>(&pExInfo->m_exception);
+        fn(pRef, sc, 0);
+        pExInfo = pExInfo->GetPreviousExceptionTracker();
+    }


The new ExInfo-chain scanning reports the address of ExInfo::m_exception as a GC root. There is still at least one code path that also GCPROTECTs the same slot (e.g., HandleManagedFault in src/coreclr/vm/excep.cpp uses GCPROTECT_BEGIN(exInfo.m_exception)), which will cause the same location to be reported twice. If the “same slot reported twice corrupts relocation” invariant applies, this needs to be fixed either by removing the remaining GCPROTECT(s) or by ensuring ScanStackRoots only reports slots that are not already covered by GCFrames.

Copilot · 2026-04-28T21:33:46Z

        EX_CATCH
        {
-            // If we can't create a duplicate handle, we set both throwables to the preallocated OOM exception.
+            // If we can't create a handle, we set both throwables to the preallocated OOM exception.


The EX_CATCH comment says we “set both throwables” to the preallocated OOM exception, but SafeSetThrowables was changed to only update the last-thrown handle/state (the current exception lives in ExInfo::m_exception). This comment is now misleading and should be updated to reflect the new behavior/invariant.

Suggested change

// If we can't create a handle, we set both throwables to the preallocated OOM exception.

// If we can't create a handle, update the last-thrown object/state to the

// preallocated OOM exception. The current exception remains in ExInfo::m_exception.

max-charlamb and others added 3 commits April 23, 2026 21:37

Update SafeSetThrowables comment to reflect current behavior

1192b46

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

Copilot AI review requested due to automatic review settings April 28, 2026 21:24

Copilot started reviewing on behalf of max-charlamb April 28, 2026 21:25 View session

github-actions Bot added the area-Diagnostics-coreclr label Apr 28, 2026

dotnet-policy-service Bot assigned max-charlamb Apr 28, 2026

max-charlamb added NO-MERGE The PR is not ready for merge yet (see discussion for detailed reasons) NO-REVIEW Experimental/testing PR, do NOT review it labels Apr 28, 2026

EgorBot mentioned this pull request Apr 28, 2026

Benchmarks for dotnet/runtime#127531 (for @max-charlamb) EgorBot/Benchmarks#170

Open

Copilot AI reviewed Apr 28, 2026

View reviewed changes

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

BENCH: Remove ExInfo::m_hThrowable — exception handling benchmarks#127531

BENCH: Remove ExInfo::m_hThrowable — exception handling benchmarks#127531
max-charlamb wants to merge 3 commits intomainfrom
dev/max-charlamb/exception-direct-pointer-bench

max-charlamb commented Apr 28, 2026 •

edited

Loading

Uh oh!

max-charlamb commented Apr 28, 2026 •

edited

Loading

Uh oh!

dotnet-policy-service Bot commented Apr 28, 2026

Uh oh!

Copilot AI left a comment

Uh oh!

Copilot AI Apr 28, 2026

Uh oh!

Copilot AI Apr 28, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

	// If we can't create a handle, we set both throwables to the preallocated OOM exception.
	// If we can't create a handle, update the last-thrown object/state to the
	// preallocated OOM exception. The current exception remains in ExInfo::m_exception.

Conversation

max-charlamb commented Apr 28, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

max-charlamb commented Apr 28, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

dotnet-policy-service Bot commented Apr 28, 2026

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Pull request overview

Reviewed changes

Uh oh!

Copilot AI Apr 28, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Apr 28, 2026

Choose a reason for hiding this comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

max-charlamb commented Apr 28, 2026 •

edited

Loading

max-charlamb commented Apr 28, 2026 •

edited

Loading