Open Bug 2062114 Opened 3 months ago Updated 3 months ago

Crash in [@ js::gc::ChunkBase::isNurseryChunk]

Categories

(Core :: JavaScript: GC, defect, P3)

defect

Tracking

()

People

(Reporter: calixte, Unassigned)

References

(Blocks 2 open bugs)

Details

(Keywords: crash)

Crash Data

Crash report: https://crash-stats.mozilla.org/report/index/91710ff6-daae-4752-acb3-560710260810

Crash Reason:

EXCEPTION_ACCESS_VIOLATION_READ at 0x000031b78e200010

Top 10 frames:

0  xul.dll  js::GCMarker::markCurrentColorInParallel(js::gc::ParallelMarkTask*, JS::SliceBud...  js/src/gc/Marking.cpp:1534
1  xul.dll  js::gc::ParallelMarkTask::run(js::AutoLockHelperThreadState&)  js/src/gc/ParallelMarking.cpp:165
2  xul.dll  js::GCParallelTask::runTask(JS::GCContext*, js::AutoLockHelperThreadState&)  js/src/gc/GCParallelTask.cpp:219
3  xul.dll  js::gc::ParallelMarker::markOneColor(js::gc::GCRuntime*, js::gc::MarkColor, JS::...  js/src/gc/ParallelMarking.cpp:53
4  xul.dll  js::gc::GCRuntime::markSynchronously(JS::SliceBudget&, js::gc::GCRuntime::Parall...  js/src/gc/GC.cpp:3638
5  xul.dll  js::gc::GCRuntime::gcCycle(bool, JS::SliceBudget const&, JS::GCReason)  js/src/gc/GC.cpp:5100
6  xul.dll  js::gc::GCRuntime::collect(bool, JS::SliceBudget const&, JS::GCReason)  js/src/gc/GC.cpp:5290
7  xul.dll  GarbageCollectImpl(JS::GCReason, nsJSContext::IsShrinking, JS::SliceBudget const...  dom/base/nsJSEnvironment.cpp:902
8  xul.dll  std::_Func_impl_no_alloc<`lambda at .\..\..\..\..\checkouts\gecko\dom\base\CCGCS...
9  xul.dll  mozilla::IdleTaskRunner::Run()  xpcom/threads/IdleTaskRunner.cpp:124

There are 2 crashes (from 2 installations) in nightly 155 with buildid 20260806095421.

Clouseau analysis (confidence probable): js::gc::BufferAllocator::TraceEdge loads a GCBuffer pointer atomically but then dereferences the pointed-to chunk's header (ChunkBase::kind, offset 0x10, matching the fault address 0x…200010) with no synchronization; js::BaseScript::traceChildren reaches this exact path for PrivateScriptData because 72aa2b26aa44 (bug 2042863) wired it through GCBuffer<PrivateScriptData*>/TraceEdgeAndBuffer.

The js/src/gc/BufferAllocator.cpp/ParallelMarking.cpp subsystem is new (introduced ~2026-05) and has a long, ongoing series of correctness fixes by the same author for concurrent-marking-vs-chunk-lifetime hazards (bug 2044578, 2051881, 2045637, 2049401, 2052490) through the exact build window, none of which are in the given seed list — this is an off-stack widening of the candidate set, not a seed-list hit.

Suspected regressor: 72aa2b26aa44 (gh) (bug 2042863) by Jon Coppeard.

Code references:

:jonco, can you have a look please?

Flags: needinfo?(jcoppeard)
Blocks: clouseau, 2042863

(In reply to Calixte Denizet (:calixte) from comment #0)

Clouseau analysis (confidence probable): js::gc::BufferAllocator::TraceEdge loads a GCBuffer pointer atomically but then dereferences the pointed-to chunk's header (ChunkBase::kind, offset 0x10, matching the fault address 0x…200010) with no synchronization

No synchronization is necessary here because chunks are never freed until after marking has finished.

The js/src/gc/BufferAllocator.cpp/ParallelMarking.cpp subsystem is new (introduced ~2026-05) and has a long, ongoing series of correctness fixes by the same author for concurrent-marking-vs-chunk-lifetime hazards

Concurrent marking is under development and is not present in any release builds so is not relevant to these crashes.

This is probably a random memory corruption bug.

Flags: needinfo?(jcoppeard)

:calixte, since this bug is a regression, could you fill (if possible) the regressed_by field?
For more information, please visit BugBot documentation.

Flags: needinfo?(cdenizet)
Blocks: GCCrashes
Flags: needinfo?(cdenizet)
Keywords: regression
Severity: -- → S3
Priority: -- → P3
You need to log in before you can comment on or make changes to this bug.