Call ReferenceQueue add outside of active GC block
We were calling ReferenceQueue.add within the runtime GC active block.
This caused java code to be run and could (potentially) cause
deadlocks with JVMTI and debuggers.
To fix this we collect the cleared references during the GC and only
enqueue them after FinishGC.
Test: ./test.py --host
Test: atest CtsJdwpTunnelHostTestCases # with goldfish emulator
Test: ./art/tools/run-libjdwp-tests.sh --mode=host
Bug: 132460313
Change-Id: I276870096fb60a06afba7f850325d06709227b8e
diff --git a/runtime/gc/reference_processor.h b/runtime/gc/reference_processor.h
index c1c9a3c..54de5cc 100644
--- a/runtime/gc/reference_processor.h
+++ b/runtime/gc/reference_processor.h
@@ -61,7 +61,9 @@
// Decode the referent, may block if references are being processed.
ObjPtr<mirror::Object> GetReferent(Thread* self, ObjPtr<mirror::Reference> reference)
REQUIRES_SHARED(Locks::mutator_lock_) REQUIRES(!Locks::reference_processor_lock_);
- void EnqueueClearedReferences(Thread* self) REQUIRES(!Locks::mutator_lock_);
+ // Collects the cleared references and returns a task, to be executed after FinishGC, that will
+ // enqueue all of them.
+ SelfDeletingTask* CollectClearedReferences(Thread* self) REQUIRES(!Locks::mutator_lock_);
void DelayReferenceReferent(ObjPtr<mirror::Class> klass,
ObjPtr<mirror::Reference> ref,
collector::GarbageCollector* collector)