8025305: Cleanup CardTableModRefBS usage in G1

Summary: Move some G1 specific code from CardTableModRefBS to G1SATBCardTableModRefBS.
Reviewed-by: brutisso, tschatzl, ehelin

src/share/vm/gc_implementation/g1/concurrentMark.inline.hpp

@@ -81,7 +81,7 @@ inline void ConcurrentMark::count_region(MemRegion mr, HeapRegion* hr,
                                          size_t* marked_bytes_array,
                                          BitMap* task_card_bm) {
   G1CollectedHeap* g1h = _g1h;
-  CardTableModRefBS* ct_bs = (CardTableModRefBS*) (g1h->barrier_set());
+  CardTableModRefBS* ct_bs = g1h->g1_barrier_set();
 
   HeapWord* start = mr.start();
   HeapWord* end = mr.end();

src/share/vm/gc_implementation/g1/g1CardCounts.cpp

@@ -65,9 +65,7 @@ void G1CardCounts::initialize() {
     // threshold limit is no more than this.
     guarantee(G1ConcRSHotCardLimit <= max_jubyte, "sanity");
 
-    ModRefBarrierSet* bs = _g1h->mr_bs();
-    guarantee(bs->is_a(BarrierSet::CardTableModRef), "Precondition");
-    _ct_bs = (CardTableModRefBS*)bs;
+    _ct_bs = _g1h->g1_barrier_set();
     _ct_bot = _ct_bs->byte_for_const(_g1h->reserved_region().start());
 
     // Allocate/Reserve the counts table

src/share/vm/gc_implementation/g1/g1CollectedHeap.cpp

@@ -125,10 +125,8 @@ class ClearLoggedCardTableEntryClosure: public CardTableEntryClosure {
   int _histo[256];
 public:
   ClearLoggedCardTableEntryClosure() :
-    _calls(0)
+    _calls(0), _g1h(G1CollectedHeap::heap()), _ctbs(_g1h->g1_barrier_set())
   {
-    _g1h = G1CollectedHeap::heap();
-    _ctbs = (CardTableModRefBS*)_g1h->barrier_set();
     for (int i = 0; i < 256; i++) _histo[i] = 0;
   }
   bool do_card_ptr(jbyte* card_ptr, int worker_i) {
@@ -158,11 +156,8 @@ class RedirtyLoggedCardTableEntryClosure: public CardTableEntryClosure {
   CardTableModRefBS* _ctbs;
 public:
   RedirtyLoggedCardTableEntryClosure() :
-    _calls(0)
-  {
-    _g1h = G1CollectedHeap::heap();
-    _ctbs = (CardTableModRefBS*)_g1h->barrier_set();
-  }
+    _calls(0), _g1h(G1CollectedHeap::heap()), _ctbs(_g1h->g1_barrier_set()) {}
+
   bool do_card_ptr(jbyte* card_ptr, int worker_i) {
     if (_g1h->is_in_reserved(_ctbs->addr_for(card_ptr))) {
       _calls++;
@@ -478,7 +473,7 @@ bool G1CollectedHeap::is_scavengable(const void* p) {
 
 void G1CollectedHeap::check_ct_logs_at_safepoint() {
   DirtyCardQueueSet& dcqs = JavaThread::dirty_card_queue_set();
-  CardTableModRefBS* ct_bs = (CardTableModRefBS*)barrier_set();
+  CardTableModRefBS* ct_bs = g1_barrier_set();
 
   // Count the dirty cards at the start.
   CountNonCleanMemRegionClosure count1(this);
@@ -1205,7 +1200,7 @@ public:
 };
 
 void G1CollectedHeap::clear_rsets_post_compaction() {
-  PostMCRemSetClearClosure rs_clear(this, mr_bs());
+  PostMCRemSetClearClosure rs_clear(this, g1_barrier_set());
   heap_region_iterate(&rs_clear);
 }
 
@@ -2045,20 +2040,13 @@ jint G1CollectedHeap::initialize() {
   // Create the gen rem set (and barrier set) for the entire reserved region.
   _rem_set = collector_policy()->create_rem_set(_reserved, 2);
   set_barrier_set(rem_set()->bs());
-  if (barrier_set()->is_a(BarrierSet::ModRef)) {
-    _mr_bs = (ModRefBarrierSet*)_barrier_set;
-  } else {
-    vm_exit_during_initialization("G1 requires a mod ref bs.");
+  if (!barrier_set()->is_a(BarrierSet::G1SATBCTLogging)) {
+    vm_exit_during_initialization("G1 requires a G1SATBLoggingCardTableModRefBS");
     return JNI_ENOMEM;
   }
 
   // Also create a G1 rem set.
-  if (mr_bs()->is_a(BarrierSet::CardTableModRef)) {
-    _g1_rem_set = new G1RemSet(this, (CardTableModRefBS*)mr_bs());
-  } else {
-    vm_exit_during_initialization("G1 requires a cardtable mod ref bs.");
-    return JNI_ENOMEM;
-  }
+  _g1_rem_set = new G1RemSet(this, g1_barrier_set());
 
   // Carve out the G1 part of the heap.
 
@@ -4555,7 +4543,7 @@ G1ParScanThreadState::G1ParScanThreadState(G1CollectedHeap* g1h, uint queue_num)
   : _g1h(g1h),
     _refs(g1h->task_queue(queue_num)),
     _dcq(&g1h->dirty_card_queue_set()),
-    _ct_bs((CardTableModRefBS*)_g1h->barrier_set()),
+    _ct_bs(g1h->g1_barrier_set()),
     _g1_rem(g1h->g1_rem_set()),
     _hash_seed(17), _queue_num(queue_num),
     _term_attempts(0),
@@ -5984,11 +5972,11 @@ void G1CollectedHeap::update_sets_after_freeing_regions(size_t pre_used,
 }
 
 class G1ParCleanupCTTask : public AbstractGangTask {
-  CardTableModRefBS* _ct_bs;
+  G1SATBCardTableModRefBS* _ct_bs;
   G1CollectedHeap* _g1h;
   HeapRegion* volatile _su_head;
 public:
-  G1ParCleanupCTTask(CardTableModRefBS* ct_bs,
+  G1ParCleanupCTTask(G1SATBCardTableModRefBS* ct_bs,
                      G1CollectedHeap* g1h) :
     AbstractGangTask("G1 Par Cleanup CT Task"),
     _ct_bs(ct_bs), _g1h(g1h) { }
@@ -6011,9 +5999,9 @@ public:
 #ifndef PRODUCT
 class G1VerifyCardTableCleanup: public HeapRegionClosure {
   G1CollectedHeap* _g1h;
-  CardTableModRefBS* _ct_bs;
+  G1SATBCardTableModRefBS* _ct_bs;
 public:
-  G1VerifyCardTableCleanup(G1CollectedHeap* g1h, CardTableModRefBS* ct_bs)
+  G1VerifyCardTableCleanup(G1CollectedHeap* g1h, G1SATBCardTableModRefBS* ct_bs)
     : _g1h(g1h), _ct_bs(ct_bs) { }
   virtual bool doHeapRegion(HeapRegion* r) {
     if (r->is_survivor()) {
@@ -6027,7 +6015,7 @@ public:
 
 void G1CollectedHeap::verify_not_dirty_region(HeapRegion* hr) {
   // All of the region should be clean.
-  CardTableModRefBS* ct_bs = (CardTableModRefBS*)barrier_set();
+  G1SATBCardTableModRefBS* ct_bs = g1_barrier_set();
   MemRegion mr(hr->bottom(), hr->end());
   ct_bs->verify_not_dirty_region(mr);
 }
@@ -6040,13 +6028,13 @@ void G1CollectedHeap::verify_dirty_region(HeapRegion* hr) {
   // not dirty that area (one less thing to have to do while holding
   // a lock). So we can only verify that [bottom(),pre_dummy_top()]
   // is dirty.
-  CardTableModRefBS* ct_bs = (CardTableModRefBS*) barrier_set();
+  G1SATBCardTableModRefBS* ct_bs = g1_barrier_set();
   MemRegion mr(hr->bottom(), hr->pre_dummy_top());
   ct_bs->verify_dirty_region(mr);
 }
 
 void G1CollectedHeap::verify_dirty_young_list(HeapRegion* head) {
-  CardTableModRefBS* ct_bs = (CardTableModRefBS*) barrier_set();
+  G1SATBCardTableModRefBS* ct_bs = g1_barrier_set();
   for (HeapRegion* hr = head; hr != NULL; hr = hr->get_next_young_region()) {
     verify_dirty_region(hr);
   }
@@ -6058,7 +6046,7 @@ void G1CollectedHeap::verify_dirty_young_regions() {
 #endif
 
 void G1CollectedHeap::cleanUpCardTable() {
-  CardTableModRefBS* ct_bs = (CardTableModRefBS*) (barrier_set());
+  G1SATBCardTableModRefBS* ct_bs = g1_barrier_set();
   double start = os::elapsedTime();
 
   {

src/share/vm/gc_implementation/g1/g1CollectedHeap.hpp

@@ -31,6 +31,7 @@
 #include "gc_implementation/g1/g1HRPrinter.hpp"
 #include "gc_implementation/g1/g1MonitoringSupport.hpp"
 #include "gc_implementation/g1/g1RemSet.hpp"
+#include "gc_implementation/g1/g1SATBCardTableModRefBS.hpp"
 #include "gc_implementation/g1/g1YCTypes.hpp"
 #include "gc_implementation/g1/heapRegionSeq.hpp"
 #include "gc_implementation/g1/heapRegionSets.hpp"
@@ -791,8 +792,6 @@ protected:
 
   // The g1 remembered set of the heap.
   G1RemSet* _g1_rem_set;
-  // And it's mod ref barrier set, used to track updates for the above.
-  ModRefBarrierSet* _mr_bs;
 
   // A set of cards that cover the objects for which the Rsets should be updated
   // concurrently after the collection.
@@ -1127,7 +1126,6 @@ public:
 
   // The rem set and barrier set.
   G1RemSet* g1_rem_set() const { return _g1_rem_set; }
-  ModRefBarrierSet* mr_bs() const { return _mr_bs; }
 
   unsigned get_gc_time_stamp() {
     return _gc_time_stamp;
@@ -1346,6 +1344,10 @@ public:
 
   virtual bool is_in_closed_subset(const void* p) const;
 
+  G1SATBCardTableModRefBS* g1_barrier_set() {
+    return (G1SATBCardTableModRefBS*) barrier_set();
+  }
+
   // This resets the card table to all zeros.  It is used after
   // a collection pause which used the card table to claim cards.
   void cleanUpCardTable();
@@ -1875,7 +1877,7 @@ protected:
   G1CollectedHeap* _g1h;
   RefToScanQueue*  _refs;
   DirtyCardQueue   _dcq;
-  CardTableModRefBS* _ct_bs;
+  G1SATBCardTableModRefBS* _ct_bs;
   G1RemSet* _g1_rem;
 
   G1ParGCAllocBufferContainer  _surviving_alloc_buffer;
@@ -1914,7 +1916,7 @@ protected:
   void   add_to_undo_waste(size_t waste)         { _undo_waste += waste; }
 
   DirtyCardQueue& dirty_card_queue()             { return _dcq;  }
-  CardTableModRefBS* ctbs()                      { return _ct_bs; }
+  G1SATBCardTableModRefBS* ctbs()                { return _ct_bs; }
 
   template <class T> void immediate_rs_update(HeapRegion* from, T* p, int tid) {
     if (!from->is_survivor()) {

src/share/vm/gc_implementation/g1/g1CollectedHeap.inline.hpp

@@ -134,7 +134,7 @@ G1CollectedHeap::dirty_young_block(HeapWord* start, size_t word_size) {
   assert(containing_hr->is_in(end - 1), "it should also contain end - 1");
 
   MemRegion mr(start, end);
-  ((CardTableModRefBS*)_g1h->barrier_set())->dirty(mr);
+  g1_barrier_set()->dirty(mr);
 }
 
 inline RefToScanQueue* G1CollectedHeap::task_queue(int i) const {

src/share/vm/gc_implementation/g1/g1EvacFailure.hpp

@@ -41,11 +41,11 @@ class UpdateRSetDeferred : public OopsInHeapRegionClosure {
 private:
   G1CollectedHeap* _g1;
   DirtyCardQueue *_dcq;
-  CardTableModRefBS* _ct_bs;
+  G1SATBCardTableModRefBS* _ct_bs;
 
 public:
   UpdateRSetDeferred(G1CollectedHeap* g1, DirtyCardQueue* dcq) :
-    _g1(g1), _ct_bs((CardTableModRefBS*)_g1->barrier_set()), _dcq(dcq) {}
+    _g1(g1), _ct_bs(_g1->g1_barrier_set()), _dcq(dcq) {}
 
   virtual void do_oop(narrowOop* p) { do_oop_work(p); }
   virtual void do_oop(      oop* p) { do_oop_work(p); }

src/share/vm/gc_implementation/g1/g1MarkSweep.cpp

@@ -220,7 +220,7 @@ class G1PrepareCompactClosure: public HeapRegionClosure {
 public:
   G1PrepareCompactClosure(CompactibleSpace* cs)
   : _g1h(G1CollectedHeap::heap()),
-    _mrbs(G1CollectedHeap::heap()->mr_bs()),
+    _mrbs(_g1h->g1_barrier_set()),
     _cp(NULL, cs, cs->initialize_threshold()),
     _humongous_proxy_set("G1MarkSweep Humongous Proxy Set") { }
 

src/share/vm/gc_implementation/g1/g1RemSet.cpp

@@ -109,7 +109,7 @@ class ScanRSClosure : public HeapRegionClosure {
   CodeBlobToOopClosure* _code_root_cl;
 
   G1BlockOffsetSharedArray* _bot_shared;
-  CardTableModRefBS *_ct_bs;
+  G1SATBCardTableModRefBS *_ct_bs;
 
   double _strong_code_root_scan_time_sec;
   int    _worker_i;
@@ -130,7 +130,7 @@ public:
   {
     _g1h = G1CollectedHeap::heap();
     _bot_shared = _g1h->bot_shared();
-    _ct_bs = (CardTableModRefBS*) (_g1h->barrier_set());
+    _ct_bs = _g1h->g1_barrier_set();
     _block_size = MAX2<int>(G1RSetScanBlockSize, 1);
   }
 
@@ -505,12 +505,7 @@ public:
   ScrubRSClosure(BitMap* region_bm, BitMap* card_bm) :
     _g1h(G1CollectedHeap::heap()),
     _region_bm(region_bm), _card_bm(card_bm),
-    _ctbs(NULL)
-  {
-    ModRefBarrierSet* bs = _g1h->mr_bs();
-    guarantee(bs->is_a(BarrierSet::CardTableModRef), "Precondition");
-    _ctbs = (CardTableModRefBS*)bs;
-  }
+    _ctbs(_g1h->g1_barrier_set()) {}
 
   bool doHeapRegion(HeapRegion* r) {
     if (!r->continuesHumongous()) {

src/share/vm/gc_implementation/g1/g1SATBCardTableModRefBS.cpp

@@ -64,6 +64,27 @@ G1SATBCardTableModRefBS::write_ref_array_pre_work(T* dst, int count) {
   }
 }
 
+bool G1SATBCardTableModRefBS::mark_card_deferred(size_t card_index) {
+  jbyte val = _byte_map[card_index];
+  // It's already processed
+  if ((val & (clean_card_mask_val() | deferred_card_val())) == deferred_card_val()) {
+    return false;
+  }
+  // Cached bit can be installed either on a clean card or on a claimed card.
+  jbyte new_val = val;
+  if (val == clean_card_val()) {
+    new_val = (jbyte)deferred_card_val();
+  } else {
+    if (val & claimed_card_val()) {
+      new_val = val | (jbyte)deferred_card_val();
+    }
+  }
+  if (new_val != val) {
+    Atomic::cmpxchg(new_val, &_byte_map[card_index], val);
+  }
+  return true;
+}
+
 G1SATBCardTableLoggingModRefBS::
 G1SATBCardTableLoggingModRefBS(MemRegion whole_heap,
                                int max_covered_regions) :

src/share/vm/gc_implementation/g1/g1SATBCardTableModRefBS.hpp

@@ -89,6 +89,42 @@ public:
       write_ref_array_pre_work(dst, count);
     }
   }
+
+/*
+   Claimed and deferred bits are used together in G1 during the evacuation
+   pause. These bits can have the following state transitions:
+   1. The claimed bit can be put over any other card state. Except that
+      the "dirty -> dirty and claimed" transition is checked for in
+      G1 code and is not used.
+   2. Deferred bit can be set only if the previous state of the card
+      was either clean or claimed. mark_card_deferred() is wait-free.
+      We do not care if the operation is be successful because if
+      it does not it will only result in duplicate entry in the update
+      buffer because of the "cache-miss". So it's not worth spinning.
+ */
+
+  bool is_card_claimed(size_t card_index) {
+    jbyte val = _byte_map[card_index];
+    return (val & (clean_card_mask_val() | claimed_card_val())) == claimed_card_val();
+  }
+
+  void set_card_claimed(size_t card_index) {
+      jbyte val = _byte_map[card_index];
+      if (val == clean_card_val()) {
+        val = (jbyte)claimed_card_val();
+      } else {
+        val |= (jbyte)claimed_card_val();
+      }
+      _byte_map[card_index] = val;
+  }
+
+  bool mark_card_deferred(size_t card_index);
+
+  bool is_card_deferred(size_t card_index) {
+    jbyte val = _byte_map[card_index];
+    return (val & (clean_card_mask_val() | deferred_card_val())) == deferred_card_val();
+  }
+
 };
 
 // Adds card-table logging to the post-barrier.

src/share/vm/memory/cardTableModRefBS.cpp

@@ -423,60 +423,6 @@ void CardTableModRefBS::write_ref_field_work(void* field, oop newVal) {
   inline_write_ref_field(field, newVal);
 }
 
-/*
-   Claimed and deferred bits are used together in G1 during the evacuation
-   pause. These bits can have the following state transitions:
-   1. The claimed bit can be put over any other card state. Except that
-      the "dirty -> dirty and claimed" transition is checked for in
-      G1 code and is not used.
-   2. Deferred bit can be set only if the previous state of the card
-      was either clean or claimed. mark_card_deferred() is wait-free.
-      We do not care if the operation is be successful because if
-      it does not it will only result in duplicate entry in the update
-      buffer because of the "cache-miss". So it's not worth spinning.
- */
-
-
-bool CardTableModRefBS::claim_card(size_t card_index) {
-  jbyte val = _byte_map[card_index];
-  assert(val != dirty_card_val(), "Shouldn't claim a dirty card");
-  while (val == clean_card_val() ||
-         (val & (clean_card_mask_val() | claimed_card_val())) != claimed_card_val()) {
-    jbyte new_val = val;
-    if (val == clean_card_val()) {
-      new_val = (jbyte)claimed_card_val();
-    } else {
-      new_val = val | (jbyte)claimed_card_val();
-    }
-    jbyte res = Atomic::cmpxchg(new_val, &_byte_map[card_index], val);
-    if (res == val) {
-      return true;
-    }
-    val = res;
-  }
-  return false;
-}
-
-bool CardTableModRefBS::mark_card_deferred(size_t card_index) {
-  jbyte val = _byte_map[card_index];
-  // It's already processed
-  if ((val & (clean_card_mask_val() | deferred_card_val())) == deferred_card_val()) {
-    return false;
-  }
-  // Cached bit can be installed either on a clean card or on a claimed card.
-  jbyte new_val = val;
-  if (val == clean_card_val()) {
-    new_val = (jbyte)deferred_card_val();
-  } else {
-    if (val & claimed_card_val()) {
-      new_val = val | (jbyte)deferred_card_val();
-    }
-  }
-  if (new_val != val) {
-    Atomic::cmpxchg(new_val, &_byte_map[card_index], val);
-  }
-  return true;
-}
 
 void CardTableModRefBS::non_clean_card_iterate_possibly_parallel(Space* sp,
                                                                  MemRegion mr,

src/share/vm/memory/cardTableModRefBS.hpp

@@ -339,34 +339,10 @@ public:
     _byte_map[card_index] = dirty_card_val();
   }
 
-  bool is_card_claimed(size_t card_index) {
-    jbyte val = _byte_map[card_index];
-    return (val & (clean_card_mask_val() | claimed_card_val())) == claimed_card_val();
-  }
-
-  void set_card_claimed(size_t card_index) {
-      jbyte val = _byte_map[card_index];
-      if (val == clean_card_val()) {
-        val = (jbyte)claimed_card_val();
-      } else {
-        val |= (jbyte)claimed_card_val();
-      }
-      _byte_map[card_index] = val;
-  }
-
-  bool claim_card(size_t card_index);
-
   bool is_card_clean(size_t card_index) {
     return _byte_map[card_index] == clean_card_val();
   }
 
-  bool is_card_deferred(size_t card_index) {
-    jbyte val = _byte_map[card_index];
-    return (val & (clean_card_mask_val() | deferred_card_val())) == deferred_card_val();
-  }
-
-  bool mark_card_deferred(size_t card_index);
-
   // Card marking array base (adjusted for heap low boundary)
   // This would be the 0th element of _byte_map, if the heap started at 0x0.
   // But since the heap starts at some higher address, this points to somewhere