1 /*
   2  * Copyright (c) 2001, 2014, Oracle and/or its affiliates. All rights reserved.
   3  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   4  *
   5  * This code is free software; you can redistribute it and/or modify it
   6  * under the terms of the GNU General Public License version 2 only, as
   7  * published by the Free Software Foundation.
   8  *
   9  * This code is distributed in the hope that it will be useful, but WITHOUT
  10  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  11  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
  12  * version 2 for more details (a copy is included in the LICENSE file that
  13  * accompanied this code).
  14  *
  15  * You should have received a copy of the GNU General Public License version
  16  * 2 along with this work; if not, write to the Free Software Foundation,
  17  * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
  18  *
  19  * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
  20  * or visit www.oracle.com if you need additional information or have any
  21  * questions.
  22  *
  23  */
  24 
  25 #include "precompiled.hpp"
  26 #include "gc_implementation/g1/concurrentG1Refine.hpp"
  27 #include "gc_implementation/g1/concurrentG1RefineThread.hpp"
  28 #include "gc_implementation/g1/g1BlockOffsetTable.inline.hpp"
  29 #include "gc_implementation/g1/g1CollectedHeap.inline.hpp"
  30 #include "gc_implementation/g1/g1CollectorPolicy.hpp"
  31 #include "gc_implementation/g1/g1HotCardCache.hpp"
  32 #include "gc_implementation/g1/g1GCPhaseTimes.hpp"
  33 #include "gc_implementation/g1/g1OopClosures.inline.hpp"
  34 #include "gc_implementation/g1/g1RemSet.inline.hpp"
  35 #include "gc_implementation/g1/heapRegionManager.inline.hpp"
  36 #include "gc_implementation/g1/heapRegionRemSet.hpp"
  37 #include "memory/iterator.hpp"
  38 #include "oops/oop.inline.hpp"
  39 #include "utilities/intHisto.hpp"
  40 
  41 PRAGMA_FORMAT_MUTE_WARNINGS_FOR_GCC
  42 
  43 #define CARD_REPEAT_HISTO 0
  44 
  45 #if CARD_REPEAT_HISTO
  46 static size_t ct_freq_sz;
  47 static jbyte* ct_freq = NULL;
  48 
  49 void init_ct_freq_table(size_t heap_sz_bytes) {
  50   if (ct_freq == NULL) {
  51     ct_freq_sz = heap_sz_bytes/CardTableModRefBS::card_size;
  52     ct_freq = new jbyte[ct_freq_sz];
  53     for (size_t j = 0; j < ct_freq_sz; j++) ct_freq[j] = 0;
  54   }
  55 }
  56 
  57 void ct_freq_note_card(size_t index) {
  58   assert(0 <= index && index < ct_freq_sz, "Bounds error.");
  59   if (ct_freq[index] < 100) { ct_freq[index]++; }
  60 }
  61 
  62 static IntHistogram card_repeat_count(10, 10);
  63 
  64 void ct_freq_update_histo_and_reset() {
  65   for (size_t j = 0; j < ct_freq_sz; j++) {
  66     card_repeat_count.add_entry(ct_freq[j]);
  67     ct_freq[j] = 0;
  68   }
  69 
  70 }
  71 #endif
  72 
  73 G1RemSet::G1RemSet(G1CollectedHeap* g1, CardTableModRefBS* ct_bs)
  74   : _g1(g1), _conc_refine_cards(0),
  75     _ct_bs(ct_bs), _g1p(_g1->g1_policy()),
  76     _cg1r(g1->concurrent_g1_refine()),
  77     _cset_rs_update_cl(NULL),
  78     _cards_scanned(NULL), _total_cards_scanned(0),
  79     _prev_period_summary()
  80 {
  81   guarantee(n_workers() > 0, "There should be some workers");
  82   _cset_rs_update_cl = NEW_C_HEAP_ARRAY(G1ParPushHeapRSClosure*, n_workers(), mtGC);
  83   for (uint i = 0; i < n_workers(); i++) {
  84     _cset_rs_update_cl[i] = NULL;
  85   }
  86   if (G1SummarizeRSetStats) {
  87     _prev_period_summary.initialize(this);
  88   }
  89 }
  90 
  91 G1RemSet::~G1RemSet() {
  92   for (uint i = 0; i < n_workers(); i++) {
  93     assert(_cset_rs_update_cl[i] == NULL, "it should be");
  94   }
  95   FREE_C_HEAP_ARRAY(G1ParPushHeapRSClosure*, _cset_rs_update_cl, mtGC);
  96 }
  97 
  98 void CountNonCleanMemRegionClosure::do_MemRegion(MemRegion mr) {
  99   if (_g1->is_in_g1_reserved(mr.start())) {
 100     _n += (int) ((mr.byte_size() / CardTableModRefBS::card_size));
 101     if (_start_first == NULL) _start_first = mr.start();
 102   }
 103 }
 104 
 105 class ScanRSClosure : public HeapRegionClosure {
 106   size_t _cards_done, _cards;
 107   G1CollectedHeap* _g1h;
 108 
 109   G1ParPushHeapRSClosure* _oc;
 110   CodeBlobClosure* _code_root_cl;
 111 
 112   G1BlockOffsetSharedArray* _bot_shared;
 113   G1SATBCardTableModRefBS *_ct_bs;
 114 
 115   double _strong_code_root_scan_time_sec;
 116   uint   _worker_i;
 117   int    _block_size;
 118   bool   _try_claimed;
 119 
 120 public:
 121   ScanRSClosure(G1ParPushHeapRSClosure* oc,
 122                 CodeBlobClosure* code_root_cl,
 123                 uint worker_i) :
 124     _oc(oc),
 125     _code_root_cl(code_root_cl),
 126     _strong_code_root_scan_time_sec(0.0),
 127     _cards(0),
 128     _cards_done(0),
 129     _worker_i(worker_i),
 130     _try_claimed(false)
 131   {
 132     _g1h = G1CollectedHeap::heap();
 133     _bot_shared = _g1h->bot_shared();
 134     _ct_bs = _g1h->g1_barrier_set();
 135     _block_size = MAX2<int>(G1RSetScanBlockSize, 1);
 136   }
 137 
 138   void set_try_claimed() { _try_claimed = true; }
 139 
 140   void scanCard(size_t index, HeapRegion *r) {
 141     // Stack allocate the DirtyCardToOopClosure instance
 142     HeapRegionDCTOC cl(_g1h, r, _oc,
 143                        CardTableModRefBS::Precise);
 144 
 145     // Set the "from" region in the closure.
 146     _oc->set_region(r);
 147     MemRegion card_region(_bot_shared->address_for_index(index), G1BlockOffsetSharedArray::N_words);
 148     MemRegion pre_gc_allocated(r->bottom(), r->scan_top());
 149     MemRegion mr = pre_gc_allocated.intersection(card_region);
 150     if (!mr.is_empty() && !_ct_bs->is_card_claimed(index)) {
 151       // We make the card as "claimed" lazily (so races are possible
 152       // but they're benign), which reduces the number of duplicate
 153       // scans (the rsets of the regions in the cset can intersect).
 154       _ct_bs->set_card_claimed(index);
 155       _cards_done++;
 156       cl.do_MemRegion(mr);
 157     }
 158   }
 159 
 160   void printCard(HeapRegion* card_region, size_t card_index,
 161                  HeapWord* card_start) {
 162     gclog_or_tty->print_cr("T " UINT32_FORMAT " Region [" PTR_FORMAT ", " PTR_FORMAT ") "
 163                            "RS names card %p: "
 164                            "[" PTR_FORMAT ", " PTR_FORMAT ")",
 165                            _worker_i,
 166                            card_region->bottom(), card_region->end(),
 167                            card_index,
 168                            card_start, card_start + G1BlockOffsetSharedArray::N_words);
 169   }
 170 
 171   void scan_strong_code_roots(HeapRegion* r) {
 172     double scan_start = os::elapsedTime();
 173     r->strong_code_roots_do(_code_root_cl);
 174     _strong_code_root_scan_time_sec += (os::elapsedTime() - scan_start);
 175   }
 176 
 177   bool doHeapRegion(HeapRegion* r) {
 178     assert(r->in_collection_set(), "should only be called on elements of CS.");
 179     HeapRegionRemSet* hrrs = r->rem_set();
 180     if (hrrs->iter_is_complete()) return false; // All done.
 181     if (!_try_claimed && !hrrs->claim_iter()) return false;
 182     // If we ever free the collection set concurrently, we should also
 183     // clear the card table concurrently therefore we won't need to
 184     // add regions of the collection set to the dirty cards region.
 185     _g1h->push_dirty_cards_region(r);
 186     // If we didn't return above, then
 187     //   _try_claimed || r->claim_iter()
 188     // is true: either we're supposed to work on claimed-but-not-complete
 189     // regions, or we successfully claimed the region.
 190 
 191     HeapRegionRemSetIterator iter(hrrs);
 192     size_t card_index;
 193 
 194     // We claim cards in block so as to recude the contention. The block size is determined by
 195     // the G1RSetScanBlockSize parameter.
 196     size_t jump_to_card = hrrs->iter_claimed_next(_block_size);
 197     for (size_t current_card = 0; iter.has_next(card_index); current_card++) {
 198       if (current_card >= jump_to_card + _block_size) {
 199         jump_to_card = hrrs->iter_claimed_next(_block_size);
 200       }
 201       if (current_card < jump_to_card) continue;
 202       HeapWord* card_start = _g1h->bot_shared()->address_for_index(card_index);
 203 #if 0
 204       gclog_or_tty->print("Rem set iteration yielded card [" PTR_FORMAT ", " PTR_FORMAT ").\n",
 205                           card_start, card_start + CardTableModRefBS::card_size_in_words);
 206 #endif
 207 
 208       HeapRegion* card_region = _g1h->heap_region_containing(card_start);
 209       _cards++;
 210 
 211       if (!card_region->is_on_dirty_cards_region_list()) {
 212         _g1h->push_dirty_cards_region(card_region);
 213       }
 214 
 215       // If the card is dirty, then we will scan it during updateRS.
 216       if (!card_region->in_collection_set() &&
 217           !_ct_bs->is_card_dirty(card_index)) {
 218         scanCard(card_index, card_region);
 219       }
 220     }
 221     if (!_try_claimed) {
 222       // Scan the strong code root list attached to the current region
 223       scan_strong_code_roots(r);
 224 
 225       hrrs->set_iter_complete();
 226     }
 227     return false;
 228   }
 229 
 230   double strong_code_root_scan_time_sec() {
 231     return _strong_code_root_scan_time_sec;
 232   }
 233 
 234   size_t cards_done() { return _cards_done;}
 235   size_t cards_looked_up() { return _cards;}
 236 };
 237 
 238 void G1RemSet::scanRS(G1ParPushHeapRSClosure* oc,
 239                       CodeBlobClosure* code_root_cl,
 240                       uint worker_i) {
 241   double rs_time_start = os::elapsedTime();
 242   HeapRegion *startRegion = _g1->start_cset_region_for_worker(worker_i);
 243 
 244   ScanRSClosure scanRScl(oc, code_root_cl, worker_i);
 245 
 246   _g1->collection_set_iterate_from(startRegion, &scanRScl);
 247   scanRScl.set_try_claimed();
 248   _g1->collection_set_iterate_from(startRegion, &scanRScl);
 249 
 250   double scan_rs_time_sec = (os::elapsedTime() - rs_time_start)
 251                             - scanRScl.strong_code_root_scan_time_sec();
 252 
 253   assert(_cards_scanned != NULL, "invariant");
 254   _cards_scanned[worker_i] = scanRScl.cards_done();
 255 
 256   _g1p->phase_times()->record_time_secs(G1GCPhaseTimes::ScanRS, worker_i, scan_rs_time_sec);
 257   _g1p->phase_times()->record_time_secs(G1GCPhaseTimes::CodeRoots, worker_i, scanRScl.strong_code_root_scan_time_sec());
 258 }
 259 
 260 // Closure used for updating RSets and recording references that
 261 // point into the collection set. Only called during an
 262 // evacuation pause.
 263 
 264 class RefineRecordRefsIntoCSCardTableEntryClosure: public CardTableEntryClosure {
 265   G1RemSet* _g1rs;
 266   DirtyCardQueue* _into_cset_dcq;
 267 public:
 268   RefineRecordRefsIntoCSCardTableEntryClosure(G1CollectedHeap* g1h,
 269                                               DirtyCardQueue* into_cset_dcq) :
 270     _g1rs(g1h->g1_rem_set()), _into_cset_dcq(into_cset_dcq)
 271   {}
 272   bool do_card_ptr(jbyte* card_ptr, uint worker_i) {
 273     // The only time we care about recording cards that
 274     // contain references that point into the collection set
 275     // is during RSet updating within an evacuation pause.
 276     // In this case worker_i should be the id of a GC worker thread.
 277     assert(SafepointSynchronize::is_at_safepoint(), "not during an evacuation pause");
 278     assert(worker_i < (ParallelGCThreads == 0 ? 1 : ParallelGCThreads), "should be a GC worker");
 279 
 280     if (_g1rs->refine_card(card_ptr, worker_i, true)) {
 281       // 'card_ptr' contains references that point into the collection
 282       // set. We need to record the card in the DCQS
 283       // (G1CollectedHeap::into_cset_dirty_card_queue_set())
 284       // that's used for that purpose.
 285       //
 286       // Enqueue the card
 287       _into_cset_dcq->enqueue(card_ptr);
 288     }
 289     return true;
 290   }
 291 };
 292 
 293 void G1RemSet::updateRS(DirtyCardQueue* into_cset_dcq, uint worker_i) {
 294   G1GCParPhaseTimesTracker x(_g1p->phase_times(), G1GCPhaseTimes::UpdateRS, worker_i);
 295   // Apply the given closure to all remaining log entries.
 296   RefineRecordRefsIntoCSCardTableEntryClosure into_cset_update_rs_cl(_g1, into_cset_dcq);
 297 
 298   _g1->iterate_dirty_card_closure(&into_cset_update_rs_cl, into_cset_dcq, false, worker_i);
 299 }
 300 
 301 void G1RemSet::cleanupHRRS() {
 302   HeapRegionRemSet::cleanup();
 303 }
 304 
 305 void G1RemSet::oops_into_collection_set_do(G1ParPushHeapRSClosure* oc,
 306                                            CodeBlobClosure* code_root_cl,
 307                                            uint worker_i) {
 308 #if CARD_REPEAT_HISTO
 309   ct_freq_update_histo_and_reset();
 310 #endif
 311 
 312   // We cache the value of 'oc' closure into the appropriate slot in the
 313   // _cset_rs_update_cl for this worker
 314   assert(worker_i < n_workers(), "sanity");
 315   _cset_rs_update_cl[worker_i] = oc;
 316 
 317   // A DirtyCardQueue that is used to hold cards containing references
 318   // that point into the collection set. This DCQ is associated with a
 319   // special DirtyCardQueueSet (see g1CollectedHeap.hpp).  Under normal
 320   // circumstances (i.e. the pause successfully completes), these cards
 321   // are just discarded (there's no need to update the RSets of regions
 322   // that were in the collection set - after the pause these regions
 323   // are wholly 'free' of live objects. In the event of an evacuation
 324   // failure the cards/buffers in this queue set are passed to the
 325   // DirtyCardQueueSet that is used to manage RSet updates
 326   DirtyCardQueue into_cset_dcq(&_g1->into_cset_dirty_card_queue_set());
 327 
 328   assert((ParallelGCThreads > 0) || worker_i == 0, "invariant");
 329 
 330   updateRS(&into_cset_dcq, worker_i);
 331   scanRS(oc, code_root_cl, worker_i);
 332 
 333   // We now clear the cached values of _cset_rs_update_cl for this worker
 334   _cset_rs_update_cl[worker_i] = NULL;
 335 }
 336 
 337 void G1RemSet::prepare_for_oops_into_collection_set_do() {
 338   _g1->set_refine_cte_cl_concurrency(false);
 339   DirtyCardQueueSet& dcqs = JavaThread::dirty_card_queue_set();
 340   dcqs.concatenate_logs();
 341 
 342   guarantee( _cards_scanned == NULL, "invariant" );
 343   _cards_scanned = NEW_C_HEAP_ARRAY(size_t, n_workers(), mtGC);
 344   for (uint i = 0; i < n_workers(); ++i) {
 345     _cards_scanned[i] = 0;
 346   }
 347   _total_cards_scanned = 0;
 348 }
 349 
 350 void G1RemSet::cleanup_after_oops_into_collection_set_do() {
 351   guarantee( _cards_scanned != NULL, "invariant" );
 352   _total_cards_scanned = 0;
 353   for (uint i = 0; i < n_workers(); ++i) {
 354     _total_cards_scanned += _cards_scanned[i];
 355   }
 356   FREE_C_HEAP_ARRAY(size_t, _cards_scanned, mtGC);
 357   _cards_scanned = NULL;
 358   // Cleanup after copy
 359   _g1->set_refine_cte_cl_concurrency(true);
 360   // Set all cards back to clean.
 361   _g1->cleanUpCardTable();
 362 
 363   DirtyCardQueueSet& into_cset_dcqs = _g1->into_cset_dirty_card_queue_set();
 364   int into_cset_n_buffers = into_cset_dcqs.completed_buffers_num();
 365 
 366   if (_g1->evacuation_failed()) {
 367     double restore_remembered_set_start = os::elapsedTime();
 368 
 369     // Restore remembered sets for the regions pointing into the collection set.
 370     // We just need to transfer the completed buffers from the DirtyCardQueueSet
 371     // used to hold cards that contain references that point into the collection set
 372     // to the DCQS used to hold the deferred RS updates.
 373     _g1->dirty_card_queue_set().merge_bufferlists(&into_cset_dcqs);
 374     _g1->g1_policy()->phase_times()->record_evac_fail_restore_remsets((os::elapsedTime() - restore_remembered_set_start) * 1000.0);
 375   }
 376 
 377   // Free any completed buffers in the DirtyCardQueueSet used to hold cards
 378   // which contain references that point into the collection.
 379   _g1->into_cset_dirty_card_queue_set().clear();
 380   assert(_g1->into_cset_dirty_card_queue_set().completed_buffers_num() == 0,
 381          "all buffers should be freed");
 382   _g1->into_cset_dirty_card_queue_set().clear_n_completed_buffers();
 383 }
 384 
 385 class ScrubRSClosure: public HeapRegionClosure {
 386   G1CollectedHeap* _g1h;
 387   BitMap* _region_bm;
 388   BitMap* _card_bm;
 389   CardTableModRefBS* _ctbs;
 390 public:
 391   ScrubRSClosure(BitMap* region_bm, BitMap* card_bm) :
 392     _g1h(G1CollectedHeap::heap()),
 393     _region_bm(region_bm), _card_bm(card_bm),
 394     _ctbs(_g1h->g1_barrier_set()) {}
 395 
 396   bool doHeapRegion(HeapRegion* r) {
 397     if (!r->continuesHumongous()) {
 398       r->rem_set()->scrub(_ctbs, _region_bm, _card_bm);
 399     }
 400     return false;
 401   }
 402 };
 403 
 404 void G1RemSet::scrub(BitMap* region_bm, BitMap* card_bm) {
 405   ScrubRSClosure scrub_cl(region_bm, card_bm);
 406   _g1->heap_region_iterate(&scrub_cl);
 407 }
 408 
 409 void G1RemSet::scrub_par(BitMap* region_bm, BitMap* card_bm,
 410                                 uint worker_num, int claim_val) {
 411   ScrubRSClosure scrub_cl(region_bm, card_bm);
 412   _g1->heap_region_par_iterate_chunked(&scrub_cl,
 413                                        worker_num,
 414                                        n_workers(),
 415                                        claim_val);
 416 }
 417 
 418 G1TriggerClosure::G1TriggerClosure() :
 419   _triggered(false) { }
 420 
 421 G1InvokeIfNotTriggeredClosure::G1InvokeIfNotTriggeredClosure(G1TriggerClosure* t_cl,
 422                                                              OopClosure* oop_cl)  :
 423   _trigger_cl(t_cl), _oop_cl(oop_cl) { }
 424 
 425 G1Mux2Closure::G1Mux2Closure(OopClosure *c1, OopClosure *c2) :
 426   _c1(c1), _c2(c2) { }
 427 
 428 G1UpdateRSOrPushRefOopClosure::
 429 G1UpdateRSOrPushRefOopClosure(G1CollectedHeap* g1h,
 430                               G1RemSet* rs,
 431                               G1ParPushHeapRSClosure* push_ref_cl,
 432                               bool record_refs_into_cset,
 433                               uint worker_i) :
 434   _g1(g1h), _g1_rem_set(rs), _from(NULL),
 435   _record_refs_into_cset(record_refs_into_cset),
 436   _push_ref_cl(push_ref_cl), _worker_i(worker_i) { }
 437 
 438 // Returns true if the given card contains references that point
 439 // into the collection set, if we're checking for such references;
 440 // false otherwise.
 441 
 442 bool G1RemSet::refine_card(jbyte* card_ptr, uint worker_i,
 443                            bool check_for_refs_into_cset) {
 444   assert(_g1->is_in_exact(_ct_bs->addr_for(card_ptr)),
 445          err_msg("Card at " PTR_FORMAT " index " SIZE_FORMAT " representing heap at " PTR_FORMAT " (%u) must be in committed heap",
 446                  p2i(card_ptr),
 447                  _ct_bs->index_for(_ct_bs->addr_for(card_ptr)),
 448                  _ct_bs->addr_for(card_ptr),
 449                  _g1->addr_to_region(_ct_bs->addr_for(card_ptr))));
 450 
 451   // If the card is no longer dirty, nothing to do.
 452   if (*card_ptr != CardTableModRefBS::dirty_card_val()) {
 453     // No need to return that this card contains refs that point
 454     // into the collection set.
 455     return false;
 456   }
 457 
 458   // Construct the region representing the card.
 459   HeapWord* start = _ct_bs->addr_for(card_ptr);
 460   // And find the region containing it.
 461   HeapRegion* r = _g1->heap_region_containing(start);
 462 
 463   // This check is needed for some uncommon cases where we should
 464   // ignore the card.
 465   //
 466   // The region could be young.  Cards for young regions are
 467   // distinctly marked (set to g1_young_gen), so the post-barrier will
 468   // filter them out.  However, that marking is performed
 469   // concurrently.  A write to a young object could occur before the
 470   // card has been marked young, slipping past the filter.
 471   //
 472   // The card could be stale, because the region has been freed since
 473   // the card was recorded. In this case the region type could be
 474   // anything.  If (still) free or (reallocated) young, just ignore
 475   // it.  If (reallocated) old or humongous, the later card trimming
 476   // and additional checks in iteration may detect staleness.  At
 477   // worst, we end up processing a stale card unnecessarily.
 478   //
 479   // In the normal (non-stale) case, the synchronization between the
 480   // enqueueing of the card and processing it here will have ensured
 481   // we see the up-to-date region type here.
 482   if (!r->is_old_or_humongous()) {
 483     return false;
 484   }
 485 
 486   // While we are processing RSet buffers during the collection, we
 487   // actually don't want to scan any cards on the collection set,
 488   // since we don't want to update remebered sets with entries that
 489   // point into the collection set, given that live objects from the
 490   // collection set are about to move and such entries will be stale
 491   // very soon. This change also deals with a reliability issue which
 492   // involves scanning a card in the collection set and coming across
 493   // an array that was being chunked and looking malformed. Note,
 494   // however, that if evacuation fails, we have to scan any objects
 495   // that were not moved and create any missing entries.
 496   if (r->in_collection_set()) {
 497     return false;
 498   }
 499 
 500   // The result from the hot card cache insert call is either:
 501   //   * pointer to the current card
 502   //     (implying that the current card is not 'hot'),
 503   //   * null
 504   //     (meaning we had inserted the card ptr into the "hot" card cache,
 505   //     which had some headroom),
 506   //   * a pointer to a "hot" card that was evicted from the "hot" cache.
 507   //
 508 
 509   G1HotCardCache* hot_card_cache = _cg1r->hot_card_cache();
 510   if (hot_card_cache->use_cache()) {
 511     assert(!check_for_refs_into_cset, "sanity");
 512     assert(!SafepointSynchronize::is_at_safepoint(), "sanity");
 513 
 514     const jbyte* orig_card_ptr = card_ptr;
 515     card_ptr = hot_card_cache->insert(card_ptr);
 516     if (card_ptr == NULL) {
 517       // There was no eviction. Nothing to do.
 518       return false;
 519     } else if (card_ptr != orig_card_ptr) {
 520       // Original card was inserted and an old card was evicted.
 521       start = _ct_bs->addr_for(card_ptr);
 522       r = _g1->heap_region_containing(start);
 523 
 524       // Check whether the region formerly in the cache should be
 525       // ignored, as discussed earlier for the original card.  The
 526       // region could have been freed while in the cache.  The cset is
 527       // not relevant here, since we're in concurrent phase.
 528       if (!r->is_old_or_humongous()) {
 529         return false;
 530       }
 531     } // Else we still have the original card.
 532   }
 533 
 534   // Trim the region designated by the card to what's been allocated
 535   // in the region.  The card could be stale, or the card could cover
 536   // (part of) an object at the end of the allocated space and extend
 537   // beyond the end of allocation.
 538   HeapWord* scan_limit;
 539   if (_g1->is_gc_active()) {
 540     // If we're in a STW GC, then a card might be in a GC alloc region
 541     // and extend onto a GC LAB, which may not be parsable.  Stop such
 542     // at the "scan_top" of the region.
 543     scan_limit = r->scan_top();
 544   } else {
 545     // Non-humongous objects are only allocated in the old-gen during
 546     // GC, so if region is old then top is stable.  Humongous object
 547     // allocation sets top last; if top has not yet been set, this is
 548     // a stale card and we'll end up with an empty intersection.  If
 549     // this is not a stale card, the synchronization between the
 550     // enqueuing of the card and processing it here will have ensured
 551     // we see the up-to-date top here.
 552     scan_limit = r->top();
 553   }
 554   if (scan_limit <= start) {
 555     // If the trimmed region is empty, the card must be stale.
 556     return false;
 557   }
 558 
 559   // Okay to clean and process the card now.  There are still some
 560   // stale card cases that may be detected by iteration and dealt with
 561   // as iteration failure.
 562   *const_cast<volatile jbyte*>(card_ptr) = CardTableModRefBS::clean_card_val();
 563 
 564   // This fence serves two purposes.  First, the card must be cleaned
 565   // before processing the contents.  Second, we can't proceed with
 566   // processing until after the read of top, for synchronization with
 567   // possibly concurrent humongous object allocation.  It's okay that
 568   // reading top and reading type were racy wrto each other.  We need
 569   // both set, in any order, to proceed.
 570   OrderAccess::fence();
 571 
 572   // Don't use addr_for(card_ptr + 1) which can ask for
 573   // a card beyond the heap.
 574   HeapWord* end = start + CardTableModRefBS::card_size_in_words;
 575   MemRegion dirty_region(start, MIN2(scan_limit, end));
 576   assert(!dirty_region.is_empty(), "sanity");
 577 
 578 #if CARD_REPEAT_HISTO
 579   init_ct_freq_table(_g1->max_capacity());
 580   ct_freq_note_card(_ct_bs->index_for(start));
 581 #endif
 582 
 583   G1ParPushHeapRSClosure* oops_in_heap_closure = NULL;
 584   if (check_for_refs_into_cset) {
 585     // ConcurrentG1RefineThreads have worker numbers larger than what
 586     // _cset_rs_update_cl[] is set up to handle. But those threads should
 587     // only be active outside of a collection which means that when they
 588     // reach here they should have check_for_refs_into_cset == false.
 589     assert((size_t)worker_i < n_workers(), "index of worker larger than _cset_rs_update_cl[].length");
 590     oops_in_heap_closure = _cset_rs_update_cl[worker_i];
 591   }
 592   G1UpdateRSOrPushRefOopClosure update_rs_oop_cl(_g1,
 593                                                  _g1->g1_rem_set(),
 594                                                  oops_in_heap_closure,
 595                                                  check_for_refs_into_cset,
 596                                                  worker_i);
 597   update_rs_oop_cl.set_from(r);
 598 
 599   G1TriggerClosure trigger_cl;
 600   FilterIntoCSClosure into_cs_cl(NULL, _g1, &trigger_cl);
 601   G1InvokeIfNotTriggeredClosure invoke_cl(&trigger_cl, &into_cs_cl);
 602   G1Mux2Closure mux(&invoke_cl, &update_rs_oop_cl);
 603 
 604   FilterOutOfRegionClosure filter_then_update_rs_oop_cl(r,
 605                         (check_for_refs_into_cset ?
 606                                 (OopClosure*)&mux :
 607                                 (OopClosure*)&update_rs_oop_cl));
 608 
 609   bool card_processed =
 610     r->oops_on_card_seq_iterate_careful(dirty_region,
 611                                         &filter_then_update_rs_oop_cl);
 612 
 613   // If unable to process the card then we encountered an unparsable
 614   // part of the heap (e.g. a partially allocated object) while
 615   // processing a stale card.  Despite the card being stale, redirty
 616   // and re-enqueue, because we've already cleaned the card.  Without
 617   // this we could incorrectly discard a non-stale card.
 618   if (!card_processed) {
 619     assert(!_g1->is_gc_active(), "Unparsable heap during GC");
 620     // The card might have gotten re-dirtied and re-enqueued while we
 621     // worked.  (In fact, it's pretty likely.)
 622     if (*card_ptr != CardTableModRefBS::dirty_card_val()) {
 623       *card_ptr = CardTableModRefBS::dirty_card_val();
 624       MutexLockerEx x(Shared_DirtyCardQ_lock,
 625                       Mutex::_no_safepoint_check_flag);
 626       DirtyCardQueue* sdcq =
 627         JavaThread::dirty_card_queue_set().shared_dirty_card_queue();
 628       sdcq->enqueue(card_ptr);
 629     }
 630   } else {
 631     _conc_refine_cards++;
 632   }
 633 
 634   // This gets set to true if the card being refined has
 635   // references that point into the collection set.
 636   bool has_refs_into_cset = trigger_cl.triggered();
 637 
 638   // We should only be detecting that the card contains references
 639   // that point into the collection set if the current thread is
 640   // a GC worker thread.
 641   assert(!has_refs_into_cset || SafepointSynchronize::is_at_safepoint(),
 642            "invalid result at non safepoint");
 643 
 644   return has_refs_into_cset;
 645 }
 646 
 647 void G1RemSet::print_periodic_summary_info(const char* header) {
 648   G1RemSetSummary current;
 649   current.initialize(this);
 650 
 651   _prev_period_summary.subtract_from(&current);
 652   print_summary_info(&_prev_period_summary, header);
 653 
 654   _prev_period_summary.set(&current);
 655 }
 656 
 657 void G1RemSet::print_summary_info() {
 658   G1RemSetSummary current;
 659   current.initialize(this);
 660 
 661   print_summary_info(&current, " Cumulative RS summary");
 662 }
 663 
 664 void G1RemSet::print_summary_info(G1RemSetSummary * summary, const char * header) {
 665   assert(summary != NULL, "just checking");
 666 
 667   if (header != NULL) {
 668     gclog_or_tty->print_cr("%s", header);
 669   }
 670 
 671 #if CARD_REPEAT_HISTO
 672   gclog_or_tty->print_cr("\nG1 card_repeat count histogram: ");
 673   gclog_or_tty->print_cr("  # of repeats --> # of cards with that number.");
 674   card_repeat_count.print_on(gclog_or_tty);
 675 #endif
 676 
 677   summary->print_on(gclog_or_tty);
 678 }
 679 
 680 void G1RemSet::prepare_for_verify() {
 681   if (G1HRRSFlushLogBuffersOnVerify &&
 682       (VerifyBeforeGC || VerifyAfterGC)
 683       &&  (!_g1->full_collection() || G1VerifyRSetsDuringFullGC)) {
 684     cleanupHRRS();
 685     _g1->set_refine_cte_cl_concurrency(false);
 686     if (SafepointSynchronize::is_at_safepoint()) {
 687       DirtyCardQueueSet& dcqs = JavaThread::dirty_card_queue_set();
 688       dcqs.concatenate_logs();
 689     }
 690 
 691     G1HotCardCache* hot_card_cache = _cg1r->hot_card_cache();
 692     bool use_hot_card_cache = hot_card_cache->use_cache();
 693     hot_card_cache->set_use_cache(false);
 694 
 695     DirtyCardQueue into_cset_dcq(&_g1->into_cset_dirty_card_queue_set());
 696     updateRS(&into_cset_dcq, 0);
 697     _g1->into_cset_dirty_card_queue_set().clear();
 698 
 699     hot_card_cache->set_use_cache(use_hot_card_cache);
 700     assert(JavaThread::dirty_card_queue_set().completed_buffers_num() == 0, "All should be consumed");
 701   }
 702 }