view src/share/vm/gc_implementation/g1/g1RemSet.hpp @ 2034:7e37af9d69ef

7011379: G1: overly long concurrent marking cycles Summary: This changeset introduces filtering of SATB buffers at the point when they are about to be enqueued. If this filtering clears enough entries on each buffer, the buffer can then be re-used and not enqueued. This cuts down the number of SATB buffers that need to be processed by the concurrent marking threads. Reviewed-by: johnc, ysr
author tonyp
date Wed, 19 Jan 2011 09:35:17 -0500
parents 878b57474103
children e8b0b0392037
line wrap: on
line source
 * Copyright (c) 2001, 2010, Oracle and/or its affiliates. All rights reserved.
 * This code is free software; you can redistribute it and/or modify it
 * under the terms of the GNU General Public License version 2 only, as
 * published by the Free Software Foundation.
 * This code is distributed in the hope that it will be useful, but WITHOUT
 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
 * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
 * version 2 for more details (a copy is included in the LICENSE file that
 * accompanied this code).
 * You should have received a copy of the GNU General Public License version
 * 2 along with this work; if not, write to the Free Software Foundation,
 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
 * or visit if you need additional information or have any
 * questions.


// A G1RemSet provides ways of iterating over pointers into a selected
// collection set.

class G1CollectedHeap;
class CardTableModRefBarrierSet;
class ConcurrentG1Refine;

// A G1RemSet in which each heap region has a rem set that records the
// external heap references into it.  Uses a mod ref bs to track updates,
// so that they can be used to update the individual region remsets.

class G1RemSet: public CHeapObj {
  G1CollectedHeap* _g1;
  unsigned _conc_refine_cards;
  size_t n_workers();

  enum SomePrivateConstants {
    UpdateRStoMergeSync  = 0,
    MergeRStoDoDirtySync = 1,
    DoDirtySync          = 2,
    LastSync             = 3,

    SeqTask              = 0,
    NumSeqTasks          = 1

  CardTableModRefBS*             _ct_bs;
  SubTasksDone*                  _seq_task;
  G1CollectorPolicy* _g1p;

  ConcurrentG1Refine* _cg1r;

  size_t*             _cards_scanned;
  size_t              _total_cards_scanned;

  // Used for caching the closure that is responsible for scanning
  // references into the collection set.
  OopsInHeapRegionClosure** _cset_rs_update_cl;

  // The routine that performs the actual work of refining a dirty
  // card.
  // If check_for_refs_into_refs is true then a true result is returned
  // if the card contains oops that have references into the current
  // collection set.
  bool concurrentRefineOneCard_impl(jbyte* card_ptr, int worker_i,
                                    bool check_for_refs_into_cset);

  // This is called to reset dual hash tables after the gc pause
  // is finished and the initial hash table is no longer being
  // scanned.
  void cleanupHRRS();

  G1RemSet(G1CollectedHeap* g1, CardTableModRefBS* ct_bs);

  // Invoke "blk->do_oop" on all pointers into the CS in objects in regions
  // outside the CS (having invoked "blk->set_region" to set the "from"
  // region correctly beforehand.) The "worker_i" param is for the
  // parallel case where the number of the worker thread calling this
  // function can be helpful in partitioning the work to be done. It
  // should be the same as the "i" passed to the calling thread's
  // work(i) function. In the sequential case this param will be ingored.
  void oops_into_collection_set_do(OopsInHeapRegionClosure* blk,
                                   int worker_i);

  // Prepare for and cleanup after an oops_into_collection_set_do
  // call.  Must call each of these once before and after (in sequential
  // code) any threads call oops_into_collection_set_do.  (This offers an
  // opportunity to sequential setup and teardown of structures needed by a
  // parallel iteration over the CS's RS.)
  void prepare_for_oops_into_collection_set_do();
  void cleanup_after_oops_into_collection_set_do();

  void scanRS(OopsInHeapRegionClosure* oc, int worker_i);
  void updateRS(DirtyCardQueue* into_cset_dcq, int worker_i);

  HeapRegion* calculateStartRegion(int i);

  CardTableModRefBS* ct_bs() { return _ct_bs; }
  size_t cardsScanned() { return _total_cards_scanned; }

  // Record, if necessary, the fact that *p (where "p" is in region "from",
  // which is required to be non-NULL) has changed to a new non-NULL value.
  template <class T> void write_ref(HeapRegion* from, T* p);
  template <class T> void par_write_ref(HeapRegion* from, T* p, int tid);

  // Requires "region_bm" and "card_bm" to be bitmaps with 1 bit per region
  // or card, respectively, such that a region or card with a corresponding
  // 0 bit contains no part of any live object.  Eliminates any remembered
  // set entries that correspond to dead heap ranges.
  void scrub(BitMap* region_bm, BitMap* card_bm);

  // Like the above, but assumes is called in parallel: "worker_num" is the
  // parallel thread id of the current thread, and "claim_val" is the
  // value that should be used to claim heap regions.
  void scrub_par(BitMap* region_bm, BitMap* card_bm,
                 int worker_num, int claim_val);

  // Refine the card corresponding to "card_ptr".  If "sts" is non-NULL,
  // join and leave around parts that must be atomic wrt GC.  (NULL means
  // being done at a safepoint.)
  // If check_for_refs_into_cset is true, a true result is returned
  // if the given card contains oops that have references into the
  // current collection set.
  virtual bool concurrentRefineOneCard(jbyte* card_ptr, int worker_i,
                                       bool check_for_refs_into_cset);

  // Print any relevant summary info.
  virtual void print_summary_info();

  // Prepare remembered set for verification.
  virtual void prepare_for_verify();

#define G1_REM_SET_LOGGING 0

class CountNonCleanMemRegionClosure: public MemRegionClosure {
  G1CollectedHeap* _g1;
  int _n;
  HeapWord* _start_first;
  CountNonCleanMemRegionClosure(G1CollectedHeap* g1) :
    _g1(g1), _n(0), _start_first(NULL)
  void do_MemRegion(MemRegion mr);
  int n() { return _n; };
  HeapWord* start_first() { return _start_first; }

class UpdateRSOopClosure: public OopClosure {
  HeapRegion* _from;
  G1RemSet* _rs;
  int _worker_i;

  template <class T> void do_oop_work(T* p);

  UpdateRSOopClosure(G1RemSet* rs, int worker_i = 0) :
    _from(NULL), _rs(rs), _worker_i(worker_i)

  void set_from(HeapRegion* from) {
    assert(from != NULL, "from region must be non-NULL");
    _from = from;

  virtual void do_oop(narrowOop* p) { do_oop_work(p); }
  virtual void do_oop(oop* p)       { do_oop_work(p); }

  // Override: this closure is idempotent.
  //  bool idempotent() { return true; }
  bool apply_to_weak_ref_discovered_field() { return true; }

class UpdateRSetImmediate: public OopsInHeapRegionClosure {
  G1RemSet* _g1_rem_set;

  template <class T> void do_oop_work(T* p);
  UpdateRSetImmediate(G1RemSet* rs) :
    _g1_rem_set(rs) {}

  virtual void do_oop(narrowOop* p) { do_oop_work(p); }
  virtual void do_oop(      oop* p) { do_oop_work(p); }

class UpdateRSOrPushRefOopClosure: public OopClosure {
  G1CollectedHeap* _g1;
  G1RemSet* _g1_rem_set;
  HeapRegion* _from;
  OopsInHeapRegionClosure* _push_ref_cl;
  bool _record_refs_into_cset;
  int _worker_i;

  template <class T> void do_oop_work(T* p);

  UpdateRSOrPushRefOopClosure(G1CollectedHeap* g1h,
                              G1RemSet* rs,
                              OopsInHeapRegionClosure* push_ref_cl,
                              bool record_refs_into_cset,
                              int worker_i = 0) :
    _worker_i(worker_i) { }

  void set_from(HeapRegion* from) {
    assert(from != NULL, "from region must be non-NULL");
    _from = from;

  bool self_forwarded(oop obj) {
    bool result = (obj->is_forwarded() && (obj->forwardee()== obj));
    return result;

  virtual void do_oop(narrowOop* p) { do_oop_work(p); }
  virtual void do_oop(oop* p)       { do_oop_work(p); }

  bool apply_to_weak_ref_discovered_field() { return true; }