--- a/src/share/vm/gc_implementation/g1/concurrentMark.cpp	Tue Aug 24 17:24:33 2010 -0400
+++ b/src/share/vm/gc_implementation/g1/concurrentMark.cpp	Wed Dec 01 17:34:02 2010 -0800
@@ -1051,6 +1051,7 @@
   void work(int worker_i) {
     assert(Thread::current()->is_ConcurrentGC_thread(),
            "this should only be done by a conc GC thread");
+    ResourceMark rm;
 
     double start_vtime = os::elapsedVTime();
 
@@ -1888,6 +1889,9 @@
   G1CollectedHeap* g1h   = G1CollectedHeap::heap();
   ReferenceProcessor* rp = g1h->ref_processor();
 
+  // See the comment in G1CollectedHeap::ref_processing_init()
+  // about how reference processing currently works in G1.
+
   // Process weak references.
   rp->setup_policy(clear_all_soft_refs);
   assert(_markStack.isEmpty(), "mark stack should be empty");
@@ -2918,7 +2922,11 @@
   CMOopClosure(G1CollectedHeap* g1h,
                ConcurrentMark* cm,
                CMTask* task)
-    : _g1h(g1h), _cm(cm), _task(task) { }
+    : _g1h(g1h), _cm(cm), _task(task)
+  {
+    _ref_processor = g1h->ref_processor();
+    assert(_ref_processor != NULL, "should not be NULL");
+  }
 };
 
 void CMTask::setup_for_region(HeapRegion* hr) {

--- a/src/share/vm/gc_implementation/g1/g1CollectedHeap.cpp	Tue Aug 24 17:24:33 2010 -0400
+++ b/src/share/vm/gc_implementation/g1/g1CollectedHeap.cpp	Wed Dec 01 17:34:02 2010 -0800
@@ -1251,6 +1251,9 @@
       g1_policy()->set_full_young_gcs(true);
     }
 
+    // See the comment in G1CollectedHeap::ref_processing_init() about
+    // how reference processing currently works in G1.
+
     // Temporarily make reference _discovery_ single threaded (non-MT).
     ReferenceProcessorMTMutator rp_disc_ser(ref_processor(), false);
 
@@ -2012,6 +2015,24 @@
 }
 
 void G1CollectedHeap::ref_processing_init() {
+  // Reference processing in G1 currently works as follows:
+  //
+  // * There is only one reference processor instance that
+  //   'spans' the entire heap. It is created by the code
+  //   below.
+  // * Reference discovery is not enabled during an incremental
+  //   pause (see 6484982).
+  // * Discoverered refs are not enqueued nor are they processed
+  //   during an incremental pause (see 6484982).
+  // * Reference discovery is enabled at initial marking.
+  // * Reference discovery is disabled and the discovered
+  //   references processed etc during remarking.
+  // * Reference discovery is MT (see below).
+  // * Reference discovery requires a barrier (see below).
+  // * Reference processing is currently not MT (see 6608385).
+  // * A full GC enables (non-MT) reference discovery and
+  //   processes any discovered references.
+
   SharedHeap::ref_processing_init();
   MemRegion mr = reserved_region();
   _ref_processor = ReferenceProcessor::create_ref_processor(
@@ -3231,6 +3252,9 @@
 
       COMPILER2_PRESENT(DerivedPointerTable::clear());
 
+      // Please see comment in G1CollectedHeap::ref_processing_init()
+      // to see how reference processing currently works in G1.
+      //
       // We want to turn off ref discovery, if necessary, and turn it back on
       // on again later if we do. XXX Dubious: why is discovery disabled?
       bool was_enabled = ref_processor()->discovery_enabled();
@@ -3660,6 +3684,7 @@
   // untag the GC alloc regions and tear down the GC alloc region
   // list. It's desirable that no regions are tagged as GC alloc
   // outside GCs.
+
   forget_alloc_region_list();
 
   // The current alloc regions contain objs that have survived
@@ -4665,6 +4690,10 @@
   }
   // Finish with the ref_processor roots.
   if (!_process_strong_tasks->is_task_claimed(G1H_PS_refProcessor_oops_do)) {
+    // We need to treat the discovered reference lists as roots and
+    // keep entries (which are added by the marking threads) on them
+    // live until they can be processed at the end of marking.
+    ref_processor()->weak_oops_do(scan_non_heap_roots);
     ref_processor()->oops_do(scan_non_heap_roots);
   }
   g1_policy()->record_collection_pause_end_G1_strong_roots();
@@ -4730,6 +4759,11 @@
   // on individual heap regions when we allocate from
   // them in parallel, so this seems like the correct place for this.
   retire_all_alloc_regions();
+
+  // Weak root processing.
+  // Note: when JSR 292 is enabled and code blobs can contain
+  // non-perm oops then we will need to process the code blobs
+  // here too.
   {
     G1IsAliveClosure is_alive(this);
     G1KeepAliveClosure keep_alive(this);

--- a/src/share/vm/memory/referenceProcessor.cpp	Tue Aug 24 17:24:33 2010 -0400
+++ b/src/share/vm/memory/referenceProcessor.cpp	Wed Dec 01 17:34:02 2010 -0800
@@ -770,9 +770,8 @@
   // loop over the lists
   for (int i = 0; i < _max_num_q * subclasses_of_ref; i++) {
     if (TraceReferenceGC && PrintGCDetails && ((i % _max_num_q) == 0)) {
-      gclog_or_tty->print_cr(
-        "\nAbandoning %s discovered list",
-        list_name(i));
+      gclog_or_tty->print_cr("\nAbandoning %s discovered list",
+                             list_name(i));
     }
     abandon_partial_discovered_list(_discoveredSoftRefs[i]);
   }
@@ -1059,9 +1058,7 @@
     // During a multi-threaded discovery phase,
     // each thread saves to its "own" list.
     Thread* thr = Thread::current();
-    assert(thr->is_GC_task_thread(),
-           "Dubious cast from Thread* to WorkerThread*?");
-    id = ((WorkerThread*)thr)->id();
+    id = thr->as_Worker_thread()->id();
   } else {
     // single-threaded discovery, we save in round-robin
     // fashion to each of the lists.
@@ -1095,8 +1092,7 @@
       ShouldNotReachHere();
   }
   if (TraceReferenceGC && PrintGCDetails) {
-    gclog_or_tty->print_cr("Thread %d gets list " INTPTR_FORMAT,
-      id, list);
+    gclog_or_tty->print_cr("Thread %d gets list " INTPTR_FORMAT, id, list);
   }
   return list;
 }
@@ -1135,6 +1131,11 @@
     if (_discovered_list_needs_barrier) {
       _bs->write_ref_field((void*)discovered_addr, current_head);
     }
+
+    if (TraceReferenceGC) {
+      gclog_or_tty->print_cr("Enqueued reference (mt) (" INTPTR_FORMAT ": %s)",
+                             obj, obj->blueprint()->internal_name());
+    }
   } else {
     // If retest was non NULL, another thread beat us to it:
     // The reference has already been discovered...
@@ -1239,8 +1240,8 @@
       // Check assumption that an object is not potentially
       // discovered twice except by concurrent collectors that potentially
       // trace the same Reference object twice.
-      assert(UseConcMarkSweepGC,
-             "Only possible with an incremental-update concurrent collector");
+      assert(UseConcMarkSweepGC || UseG1GC,
+             "Only possible with a concurrent marking collector");
       return true;
     }
   }
@@ -1293,26 +1294,14 @@
     }
     list->set_head(obj);
     list->inc_length(1);
-  }
 
-  // In the MT discovery case, it is currently possible to see
-  // the following message multiple times if several threads
-  // discover a reference about the same time. Only one will
-  // however have actually added it to the disocvered queue.
-  // One could let add_to_discovered_list_mt() return an
-  // indication for success in queueing (by 1 thread) or
-  // failure (by all other threads), but I decided the extra
-  // code was not worth the effort for something that is
-  // only used for debugging support.
-  if (TraceReferenceGC) {
-    oop referent = java_lang_ref_Reference::referent(obj);
-    if (PrintGCDetails) {
+    if (TraceReferenceGC) {
       gclog_or_tty->print_cr("Enqueued reference (" INTPTR_FORMAT ": %s)",
-                             obj, obj->blueprint()->internal_name());
+                                obj, obj->blueprint()->internal_name());
     }
-    assert(referent->is_oop(), "Enqueued a bad referent");
   }
   assert(obj->is_oop(), "Enqueued a bad reference");
+  assert(java_lang_ref_Reference::referent(obj)->is_oop(), "Enqueued a bad referent");
   return true;
 }
 

--- a/src/share/vm/runtime/thread.hpp	Tue Aug 24 17:24:33 2010 -0400
+++ b/src/share/vm/runtime/thread.hpp	Wed Dec 01 17:34:02 2010 -0800
@@ -78,6 +78,8 @@
 class ThreadClosure;
 class IdealGraphPrinter;
 
+class WorkerThread;
+
 // Class hierarchy
 // - Thread
 //   - NamedThread
@@ -289,6 +291,10 @@
   virtual bool is_Watcher_thread() const             { return false; }
   virtual bool is_ConcurrentGC_thread() const        { return false; }
   virtual bool is_Named_thread() const               { return false; }
+  virtual bool is_Worker_thread() const              { return false; }
+
+  // Casts
+  virtual WorkerThread* as_Worker_thread() const     { return NULL; }
 
   virtual char* name() const { return (char*)"Unknown thread"; }
 
@@ -628,9 +634,16 @@
 private:
   uint _id;
 public:
-  WorkerThread() : _id(0) { }
-  void set_id(uint work_id) { _id = work_id; }
-  uint id() const { return _id; }
+  WorkerThread() : _id(0)               { }
+  virtual bool is_Worker_thread() const { return true; }
+
+  virtual WorkerThread* as_Worker_thread() const {
+    assert(is_Worker_thread(), "Dubious cast to WorkerThread*?");
+    return (WorkerThread*) this;
+  }
+
+  void set_id(uint work_id)             { _id = work_id; }
+  uint id() const                       { return _id; }
 };
 
 // A single WatcherThread is used for simulating timer interrupts.