1 /*
   2  * Copyright (c) 2001, 2015, 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/g1/concurrentG1Refine.hpp"
  27 #include "gc/g1/concurrentG1RefineThread.hpp"
  28 #include "gc/g1/g1BlockOffsetTable.inline.hpp"
  29 #include "gc/g1/g1CollectedHeap.inline.hpp"
  30 #include "gc/g1/g1CollectorPolicy.hpp"
  31 #include "gc/g1/g1GCPhaseTimes.hpp"
  32 #include "gc/g1/g1HotCardCache.hpp"
  33 #include "gc/g1/g1OopClosures.inline.hpp"
  34 #include "gc/g1/g1RemSet.inline.hpp"
  35 #include "gc/g1/heapRegionManager.inline.hpp"
  36 #include "gc/g1/heapRegionRemSet.hpp"
  37 #include "memory/iterator.hpp"
  38 #include "oops/oop.inline.hpp"
  39 #include "utilities/globalDefinitions.hpp"
  40 #include "utilities/intHisto.hpp"
  41 #include "utilities/stack.inline.hpp"
  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   _cset_rs_update_cl = NEW_C_HEAP_ARRAY(G1ParPushHeapRSClosure*, n_workers(), mtGC);
  82   for (uint i = 0; i < n_workers(); i++) {
  83     _cset_rs_update_cl[i] = NULL;
  84   }
  85   if (G1SummarizeRSetStats) {
  86     _prev_period_summary.initialize(this);
  87   }
  88 }
  89 
  90 G1RemSet::~G1RemSet() {
  91   for (uint i = 0; i < n_workers(); i++) {
  92     assert(_cset_rs_update_cl[i] == NULL, "it should be");
  93   }
  94   FREE_C_HEAP_ARRAY(G1ParPushHeapRSClosure*, _cset_rs_update_cl);
  95 }
  96 
  97 class ScanRSClosure : public HeapRegionClosure {
  98   size_t _cards_done, _cards;
  99   G1CollectedHeap* _g1h;
 100 
 101   G1ParPushHeapRSClosure* _oc;
 102   CodeBlobClosure* _code_root_cl;
 103 
 104   G1BlockOffsetSharedArray* _bot_shared;
 105   G1SATBCardTableModRefBS *_ct_bs;
 106 
 107   double _strong_code_root_scan_time_sec;
 108   uint   _worker_i;
 109   size_t _block_size;
 110   bool   _try_claimed;
 111 
 112 public:
 113   ScanRSClosure(G1ParPushHeapRSClosure* oc,
 114                 CodeBlobClosure* code_root_cl,
 115                 uint worker_i) :
 116     _oc(oc),
 117     _code_root_cl(code_root_cl),
 118     _strong_code_root_scan_time_sec(0.0),
 119     _cards(0),
 120     _cards_done(0),
 121     _worker_i(worker_i),
 122     _try_claimed(false)
 123   {
 124     _g1h = G1CollectedHeap::heap();
 125     _bot_shared = _g1h->bot_shared();
 126     _ct_bs = _g1h->g1_barrier_set();
 127     _block_size = MAX2<size_t>(G1RSetScanBlockSize, 1);
 128   }
 129 
 130   void set_try_claimed() { _try_claimed = true; }
 131 
 132   void scanCard(size_t index, HeapRegion *r) {
 133     // Stack allocate the DirtyCardToOopClosure instance
 134     HeapRegionDCTOC cl(_g1h, r, _oc,
 135                        CardTableModRefBS::Precise);
 136 
 137     // Set the "from" region in the closure.
 138     _oc->set_region(r);
 139     MemRegion card_region(_bot_shared->address_for_index(index), G1BlockOffsetSharedArray::N_words);
 140     MemRegion pre_gc_allocated(r->bottom(), r->scan_top());
 141     MemRegion mr = pre_gc_allocated.intersection(card_region);
 142     if (!mr.is_empty() && !_ct_bs->is_card_claimed(index)) {
 143       // We make the card as "claimed" lazily (so races are possible
 144       // but they're benign), which reduces the number of duplicate
 145       // scans (the rsets of the regions in the cset can intersect).
 146       _ct_bs->set_card_claimed(index);
 147       _cards_done++;
 148       cl.do_MemRegion(mr);
 149     }
 150   }
 151 
 152   void printCard(HeapRegion* card_region, size_t card_index,
 153                  HeapWord* card_start) {
 154     gclog_or_tty->print_cr("T %u Region [" PTR_FORMAT ", " PTR_FORMAT ") "
 155                            "RS names card " SIZE_FORMAT_HEX ": "
 156                            "[" PTR_FORMAT ", " PTR_FORMAT ")",
 157                            _worker_i,
 158                            p2i(card_region->bottom()), p2i(card_region->end()),
 159                            card_index,
 160                            p2i(card_start), p2i(card_start + G1BlockOffsetSharedArray::N_words));
 161   }
 162 
 163   void scan_strong_code_roots(HeapRegion* r) {
 164     double scan_start = os::elapsedTime();
 165     r->strong_code_roots_do(_code_root_cl);
 166     _strong_code_root_scan_time_sec += (os::elapsedTime() - scan_start);
 167   }
 168 
 169   bool doHeapRegion(HeapRegion* r) {
 170     assert(r->in_collection_set(), "should only be called on elements of CS.");
 171     HeapRegionRemSet* hrrs = r->rem_set();
 172     if (hrrs->iter_is_complete()) return false; // All done.
 173     if (!_try_claimed && !hrrs->claim_iter()) return false;
 174     // If we ever free the collection set concurrently, we should also
 175     // clear the card table concurrently therefore we won't need to
 176     // add regions of the collection set to the dirty cards region.
 177     _g1h->push_dirty_cards_region(r);
 178     // If we didn't return above, then
 179     //   _try_claimed || r->claim_iter()
 180     // is true: either we're supposed to work on claimed-but-not-complete
 181     // regions, or we successfully claimed the region.
 182 
 183     HeapRegionRemSetIterator iter(hrrs);
 184     size_t card_index;
 185 
 186     // We claim cards in block so as to reduce the contention. The block size is determined by
 187     // the G1RSetScanBlockSize parameter.
 188     size_t jump_to_card = hrrs->iter_claimed_next(_block_size);
 189     for (size_t current_card = 0; iter.has_next(card_index); current_card++) {
 190       if (current_card >= jump_to_card + _block_size) {
 191         jump_to_card = hrrs->iter_claimed_next(_block_size);
 192       }
 193       if (current_card < jump_to_card) continue;
 194       HeapWord* card_start = _g1h->bot_shared()->address_for_index(card_index);
 195 #if 0
 196       gclog_or_tty->print("Rem set iteration yielded card [" PTR_FORMAT ", " PTR_FORMAT ").\n",
 197                           card_start, card_start + CardTableModRefBS::card_size_in_words);
 198 #endif
 199 
 200       HeapRegion* card_region = _g1h->heap_region_containing(card_start);
 201       _cards++;
 202 
 203       if (!card_region->is_on_dirty_cards_region_list()) {
 204         _g1h->push_dirty_cards_region(card_region);
 205       }
 206 
 207       // If the card is dirty, then we will scan it during updateRS.
 208       if (!card_region->in_collection_set() &&
 209           !_ct_bs->is_card_dirty(card_index)) {
 210         scanCard(card_index, card_region);
 211       }
 212     }
 213     if (!_try_claimed) {
 214       // Scan the strong code root list attached to the current region
 215       scan_strong_code_roots(r);
 216 
 217       hrrs->set_iter_complete();
 218     }
 219     return false;
 220   }
 221 
 222   double strong_code_root_scan_time_sec() {
 223     return _strong_code_root_scan_time_sec;
 224   }
 225 
 226   size_t cards_done() { return _cards_done;}
 227   size_t cards_looked_up() { return _cards;}
 228 };
 229 
 230 bool G1RemSet::clean_card(jbyte* &card_ptr,
 231                           uint worker_i,
 232                           MemRegion &dirty_region) {
 233   assert(_g1->is_in_exact(_ct_bs->addr_for(card_ptr)),
 234          err_msg("Card at " PTR_FORMAT " index " SIZE_FORMAT " representing heap at " PTR_FORMAT " (%u) must be in committed heap",
 235                  p2i(card_ptr),
 236                  _ct_bs->index_for(_ct_bs->addr_for(card_ptr)),
 237                  p2i(_ct_bs->addr_for(card_ptr)),
 238                  _g1->addr_to_region(_ct_bs->addr_for(card_ptr))));
 239 
 240   // If the card is no longer dirty, nothing to do.
 241   if (*card_ptr != CardTableModRefBS::dirty_card_val()) {
 242     // No need to return that this card contains refs that point
 243     // into the collection set.
 244     return false;
 245   }
 246 
 247   // Construct the region representing the card.
 248   HeapWord* start = _ct_bs->addr_for(card_ptr);
 249   // And find the region containing it.
 250   HeapRegion* r = _g1->heap_region_containing(start);
 251 
 252   // Why do we have to check here whether a card is on a young region,
 253   // given that we dirty young regions and, as a result, the
 254   // post-barrier is supposed to filter them out and never to enqueue
 255   // them? When we allocate a new region as the "allocation region" we
 256   // actually dirty its cards after we release the lock, since card
 257   // dirtying while holding the lock was a performance bottleneck. So,
 258   // as a result, it is possible for other threads to actually
 259   // allocate objects in the region (after the acquire the lock)
 260   // before all the cards on the region are dirtied. This is unlikely,
 261   // and it doesn't happen often, but it can happen. So, the extra
 262   // check below filters out those cards.
 263   if (r->is_young()) {
 264     return false;
 265   }
 266 
 267   // While we are processing RSet buffers during the collection, we
 268   // actually don't want to scan any cards on the collection set,
 269   // since we don't want to update remembered sets with entries that
 270   // point into the collection set, given that live objects from the
 271   // collection set are about to move and such entries will be stale
 272   // very soon. This change also deals with a reliability issue which
 273   // involves scanning a card in the collection set and coming across
 274   // an array that was being chunked and looking malformed. Note,
 275   // however, that if evacuation fails, we have to scan any objects
 276   // that were not moved and create any missing entries.
 277   if (r->in_collection_set()) {
 278     return false;
 279   }
 280 
 281   // The result from the hot card cache insert call is either:
 282   //   * pointer to the current card
 283   //     (implying that the current card is not 'hot'),
 284   //   * null
 285   //     (meaning we had inserted the card ptr into the "hot" card cache,
 286   //     which had some headroom),
 287   //   * a pointer to a "hot" card that was evicted from the "hot" cache.
 288   //
 289 
 290   G1HotCardCache* hot_card_cache = _cg1r->hot_card_cache();
 291   if (hot_card_cache->use_cache()) {
 292     assert(!SafepointSynchronize::is_at_safepoint(), "sanity");
 293 
 294     card_ptr = hot_card_cache->insert(card_ptr);
 295     if (card_ptr == NULL) {
 296       // There was no eviction. Nothing to do.
 297       return false;
 298     }
 299 
 300     start = _ct_bs->addr_for(card_ptr);
 301     r = _g1->heap_region_containing(start);
 302 
 303     // Checking whether the region we got back from the cache
 304     // is young here is inappropriate. The region could have been
 305     // freed, reallocated and tagged as young while in the cache.
 306     // Hence we could see its young type change at any time.
 307   }
 308 
 309   // Don't use addr_for(card_ptr + 1) which can ask for
 310   // a card beyond the heap.  This is not safe without a perm
 311   // gen at the upper end of the heap.
 312   HeapWord* end   = start + CardTableModRefBS::card_size_in_words;
 313   dirty_region = MemRegion(start, end);
 314 
 315 #if CARD_REPEAT_HISTO
 316   init_ct_freq_table(_g1->max_capacity());
 317   ct_freq_note_card(_ct_bs->index_for(start));
 318 #endif
 319 
 320   return r->clean_card(dirty_region, /*filter young*/ true, card_ptr);
 321 }
 322 
 323 bool G1RemSet::refine_card_buffered(jbyte* card_ptr,
 324                                     uint worker_i,
 325                                     bool check_for_refs_into_cset,
 326                                     MemRegion dirty_region) {
 327   // And find the region containing it.
 328   HeapRegion* r = _g1->heap_region_containing(dirty_region.start());
 329 
 330   G1ParPushHeapRSClosure* oops_in_heap_closure = NULL;
 331   if (check_for_refs_into_cset) {
 332     // ConcurrentG1RefineThreads have worker numbers larger than what
 333     // _cset_rs_update_cl[] is set up to handle. But those threads should
 334     // only be active outside of a collection which means that when they
 335     // reach here they should have check_for_refs_into_cset == false.
 336     assert((size_t)worker_i < n_workers(), "index of worker larger than _cset_rs_update_cl[].length");
 337     oops_in_heap_closure = _cset_rs_update_cl[worker_i];
 338   }
 339 
 340   G1UpdateRSOrPushRefOopClosure update_rs_oop_cl(_g1,
 341                                                  _g1->g1_rem_set(),
 342                                                  oops_in_heap_closure,
 343                                                  check_for_refs_into_cset,
 344                                                  worker_i);
 345   update_rs_oop_cl.set_from(r);
 346 
 347   G1TriggerClosure trigger_cl;
 348   FilterIntoCSClosure into_cs_cl(NULL, _g1, &trigger_cl);
 349   G1InvokeIfNotTriggeredClosure invoke_cl(&trigger_cl, &into_cs_cl);
 350   G1Mux2Closure mux(&invoke_cl, &update_rs_oop_cl);
 351 
 352   FilterOutOfRegionClosure filter_then_update_rs_oop_cl(r,
 353                         (check_for_refs_into_cset ?
 354                                 (OopClosure*)&mux :
 355                                 (OopClosure*)&update_rs_oop_cl));
 356 
 357   // The region for the current card may be a young region. The
 358   // current card may have been a card that was evicted from the
 359   // card cache. When the card was inserted into the cache, we had
 360   // determined that its region was non-young. While in the cache,
 361   // the region may have been freed during a cleanup pause, reallocated
 362   // and tagged as young.
 363   //
 364   // We wish to filter out cards for such a region but the current
 365   // thread, if we're running concurrently, may "see" the young type
 366   // change at any time (so an earlier "is_young" check may pass or
 367   // fail arbitrarily). We tell the iteration code to perform this
 368   // filtering when it has been determined that there has been an actual
 369   // allocation in this region and making it safe to check the young type.
 370   bool filter_young = true;
 371 
 372   HeapWord* stop_point =
 373     r->process_oops_on_card(dirty_region, &filter_then_update_rs_oop_cl, card_ptr);
 374 
 375   // If stop_point is non-null, then we encountered an unallocated region
 376   // (perhaps the unfilled portion of a TLAB.)  For now, we'll dirty the
 377   // card and re-enqueue: if we put off the card until a GC pause, then the
 378   // unallocated portion will be filled in.  Alternatively, we might try
 379   // the full complexity of the technique used in "regular" precleaning.
 380   if (stop_point != NULL) {
 381     // The card might have gotten re-dirtied and re-enqueued while we
 382     // worked.  (In fact, it's pretty likely.)
 383     card_ptr = G1CollectedHeap::heap()->g1_barrier_set()->byte_for(stop_point);
 384 
 385     if (*card_ptr != CardTableModRefBS::dirty_card_val()) {
 386       *card_ptr = CardTableModRefBS::dirty_card_val();
 387       MutexLockerEx x(Shared_DirtyCardQ_lock,
 388                       Mutex::_no_safepoint_check_flag);
 389       DirtyCardQueue* sdcq =
 390         JavaThread::dirty_card_queue_set().shared_dirty_card_queue();
 391       sdcq->enqueue(card_ptr);
 392     }
 393   } else {
 394     _conc_refine_cards++;
 395   }
 396 
 397   // This gets set to true if the card being refined has
 398   // references that point into the collection set.
 399   bool has_refs_into_cset = trigger_cl.triggered();
 400 
 401   // We should only be detecting that the card contains references
 402   // that point into the collection set if the current thread is
 403   // a GC worker thread.
 404   assert(!has_refs_into_cset || SafepointSynchronize::is_at_safepoint(),
 405            "invalid result at non safepoint");
 406 
 407   return has_refs_into_cset;
 408 }
 409 
 410 void G1RemSet::scanRS(G1ParPushHeapRSClosure* oc,
 411                       CodeBlobClosure* code_root_cl,
 412                       uint worker_i) {
 413   double rs_time_start = os::elapsedTime();
 414   HeapRegion *startRegion = _g1->start_cset_region_for_worker(worker_i);
 415 
 416   ScanRSClosure scanRScl(oc, code_root_cl, worker_i);
 417 
 418   _g1->collection_set_iterate_from(startRegion, &scanRScl);
 419   scanRScl.set_try_claimed();
 420   _g1->collection_set_iterate_from(startRegion, &scanRScl);
 421 
 422   double scan_rs_time_sec = (os::elapsedTime() - rs_time_start)
 423                             - scanRScl.strong_code_root_scan_time_sec();
 424 
 425   assert(_cards_scanned != NULL, "invariant");
 426   _cards_scanned[worker_i] = scanRScl.cards_done();
 427 
 428   _g1p->phase_times()->record_time_secs(G1GCPhaseTimes::ScanRS, worker_i, scan_rs_time_sec);
 429   _g1p->phase_times()->record_time_secs(G1GCPhaseTimes::CodeRoots, worker_i, scanRScl.strong_code_root_scan_time_sec());
 430 }
 431 
 432 // Closure used for updating RSets and recording references that
 433 // point into the collection set. Only called during an
 434 // evacuation pause.
 435 
 436 class RefineRecordRefsIntoCSCardTableEntryClosure: public CardTableEntryClosure {
 437   G1RemSet* _g1rs;
 438   DirtyCardQueue* _into_cset_dcq;
 439 public:
 440   RefineRecordRefsIntoCSCardTableEntryClosure(G1CollectedHeap* g1h,
 441                                               DirtyCardQueue* into_cset_dcq) :
 442     _g1rs(g1h->g1_rem_set()), _into_cset_dcq(into_cset_dcq)
 443   {}
 444   bool do_card_ptr(jbyte* card_ptr, uint worker_i) {
 445     // The only time we care about recording cards that
 446     // contain references that point into the collection set
 447     // is during RSet updating within an evacuation pause.
 448     // In this case worker_i should be the id of a GC worker thread.
 449     assert(SafepointSynchronize::is_at_safepoint(), "not during an evacuation pause");
 450     assert(worker_i < ParallelGCThreads, "should be a GC worker");
 451 
 452     if (_g1rs->refine_card(card_ptr, worker_i, true)) {
 453       // 'card_ptr' contains references that point into the collection
 454       // set. We need to record the card in the DCQS
 455       // (G1CollectedHeap::into_cset_dirty_card_queue_set())
 456       // that's used for that purpose.
 457       //
 458       // Enqueue the card
 459       _into_cset_dcq->enqueue(card_ptr);
 460     }
 461     return true;
 462   }
 463 };
 464 
 465 void G1RemSet::updateRS(DirtyCardQueue* into_cset_dcq, uint worker_i) {
 466   G1GCParPhaseTimesTracker x(_g1p->phase_times(), G1GCPhaseTimes::UpdateRS, worker_i);
 467   // Apply the given closure to all remaining log entries.
 468   RefineRecordRefsIntoCSCardTableEntryClosure into_cset_update_rs_cl(_g1, into_cset_dcq);
 469 
 470   _g1->iterate_dirty_card_closure(&into_cset_update_rs_cl, into_cset_dcq, false, worker_i);
 471 }
 472 
 473 void G1RemSet::cleanupHRRS() {
 474   HeapRegionRemSet::cleanup();
 475 }
 476 
 477 void G1RemSet::oops_into_collection_set_do(G1ParPushHeapRSClosure* oc,
 478                                            CodeBlobClosure* code_root_cl,
 479                                            uint worker_i) {
 480 #if CARD_REPEAT_HISTO
 481   ct_freq_update_histo_and_reset();
 482 #endif
 483 
 484   // We cache the value of 'oc' closure into the appropriate slot in the
 485   // _cset_rs_update_cl for this worker
 486   assert(worker_i < n_workers(), "sanity");
 487   _cset_rs_update_cl[worker_i] = oc;
 488 
 489   // A DirtyCardQueue that is used to hold cards containing references
 490   // that point into the collection set. This DCQ is associated with a
 491   // special DirtyCardQueueSet (see g1CollectedHeap.hpp).  Under normal
 492   // circumstances (i.e. the pause successfully completes), these cards
 493   // are just discarded (there's no need to update the RSets of regions
 494   // that were in the collection set - after the pause these regions
 495   // are wholly 'free' of live objects. In the event of an evacuation
 496   // failure the cards/buffers in this queue set are passed to the
 497   // DirtyCardQueueSet that is used to manage RSet updates
 498   DirtyCardQueue into_cset_dcq(&_g1->into_cset_dirty_card_queue_set());
 499 
 500   updateRS(&into_cset_dcq, worker_i);
 501   scanRS(oc, code_root_cl, worker_i);
 502 
 503   // We now clear the cached values of _cset_rs_update_cl for this worker
 504   _cset_rs_update_cl[worker_i] = NULL;
 505 }
 506 
 507 void G1RemSet::prepare_for_oops_into_collection_set_do() {
 508   cleanupHRRS();
 509   _g1->set_refine_cte_cl_concurrency(false);
 510   DirtyCardQueueSet& dcqs = JavaThread::dirty_card_queue_set();
 511   dcqs.concatenate_logs();
 512 
 513   guarantee( _cards_scanned == NULL, "invariant" );
 514   _cards_scanned = NEW_C_HEAP_ARRAY(size_t, n_workers(), mtGC);
 515   for (uint i = 0; i < n_workers(); ++i) {
 516     _cards_scanned[i] = 0;
 517   }
 518   _total_cards_scanned = 0;
 519 }
 520 
 521 void G1RemSet::cleanup_after_oops_into_collection_set_do() {
 522   guarantee( _cards_scanned != NULL, "invariant" );
 523   _total_cards_scanned = 0;
 524   for (uint i = 0; i < n_workers(); ++i) {
 525     _total_cards_scanned += _cards_scanned[i];
 526   }
 527   FREE_C_HEAP_ARRAY(size_t, _cards_scanned);
 528   _cards_scanned = NULL;
 529   // Cleanup after copy
 530   _g1->set_refine_cte_cl_concurrency(true);
 531   // Set all cards back to clean.
 532   _g1->cleanUpCardTable();
 533 
 534   DirtyCardQueueSet& into_cset_dcqs = _g1->into_cset_dirty_card_queue_set();
 535   int into_cset_n_buffers = into_cset_dcqs.completed_buffers_num();
 536 
 537   if (_g1->evacuation_failed()) {
 538     double restore_remembered_set_start = os::elapsedTime();
 539 
 540     // Restore remembered sets for the regions pointing into the collection set.
 541     // We just need to transfer the completed buffers from the DirtyCardQueueSet
 542     // used to hold cards that contain references that point into the collection set
 543     // to the DCQS used to hold the deferred RS updates.
 544     _g1->dirty_card_queue_set().merge_bufferlists(&into_cset_dcqs);
 545     _g1->g1_policy()->phase_times()->record_evac_fail_restore_remsets((os::elapsedTime() - restore_remembered_set_start) * 1000.0);
 546   }
 547 
 548   // Free any completed buffers in the DirtyCardQueueSet used to hold cards
 549   // which contain references that point into the collection.
 550   _g1->into_cset_dirty_card_queue_set().clear();
 551   assert(_g1->into_cset_dirty_card_queue_set().completed_buffers_num() == 0,
 552          "all buffers should be freed");
 553   _g1->into_cset_dirty_card_queue_set().clear_n_completed_buffers();
 554 }
 555 
 556 class ScrubRSClosure: public HeapRegionClosure {
 557   G1CollectedHeap* _g1h;
 558   BitMap* _region_bm;
 559   BitMap* _card_bm;
 560   CardTableModRefBS* _ctbs;
 561 public:
 562   ScrubRSClosure(BitMap* region_bm, BitMap* card_bm) :
 563     _g1h(G1CollectedHeap::heap()),
 564     _region_bm(region_bm), _card_bm(card_bm),
 565     _ctbs(_g1h->g1_barrier_set()) {}
 566 
 567   bool doHeapRegion(HeapRegion* r) {
 568     if (!r->is_continues_humongous()) {
 569       r->rem_set()->scrub(_ctbs, _region_bm, _card_bm);
 570     }
 571     return false;
 572   }
 573 };
 574 
 575 void G1RemSet::scrub(BitMap* region_bm, BitMap* card_bm, uint worker_num, HeapRegionClaimer *hrclaimer) {
 576   ScrubRSClosure scrub_cl(region_bm, card_bm);
 577   _g1->heap_region_par_iterate(&scrub_cl, worker_num, hrclaimer);
 578 }
 579 
 580 G1TriggerClosure::G1TriggerClosure() :
 581   _triggered(false) { }
 582 
 583 G1InvokeIfNotTriggeredClosure::G1InvokeIfNotTriggeredClosure(G1TriggerClosure* t_cl,
 584                                                              OopClosure* oop_cl)  :
 585   _trigger_cl(t_cl), _oop_cl(oop_cl) { }
 586 
 587 G1Mux2Closure::G1Mux2Closure(OopClosure *c1, OopClosure *c2) :
 588   _c1(c1), _c2(c2) { }
 589 
 590 G1UpdateRSOrPushRefOopClosure::
 591 G1UpdateRSOrPushRefOopClosure(G1CollectedHeap* g1h,
 592                               G1RemSet* rs,
 593                               G1ParPushHeapRSClosure* push_ref_cl,
 594                               bool record_refs_into_cset,
 595                               uint worker_i) :
 596   _g1(g1h), _g1_rem_set(rs), _from(NULL),
 597   _record_refs_into_cset(record_refs_into_cset),
 598   _push_ref_cl(push_ref_cl), _worker_i(worker_i) { }
 599 
 600 // Returns true if the given card contains references that point
 601 // into the collection set, if we're checking for such references;
 602 // false otherwise.
 603 
 604 bool G1RemSet::refine_card(jbyte* card_ptr, uint worker_i,
 605                            bool check_for_refs_into_cset) {
 606   assert(_g1->is_in_exact(_ct_bs->addr_for(card_ptr)),
 607          err_msg("Card at " PTR_FORMAT " index " SIZE_FORMAT " representing heap at " PTR_FORMAT " (%u) must be in committed heap",
 608                  p2i(card_ptr),
 609                  _ct_bs->index_for(_ct_bs->addr_for(card_ptr)),
 610                  p2i(_ct_bs->addr_for(card_ptr)),
 611                  _g1->addr_to_region(_ct_bs->addr_for(card_ptr))));
 612 
 613   // If the card is no longer dirty, nothing to do.
 614   if (*card_ptr != CardTableModRefBS::dirty_card_val()) {
 615     // No need to return that this card contains refs that point
 616     // into the collection set.
 617     return false;
 618   }
 619 
 620   // Construct the region representing the card.
 621   HeapWord* start = _ct_bs->addr_for(card_ptr);
 622   // And find the region containing it.
 623   HeapRegion* r = _g1->heap_region_containing(start);
 624 
 625   // Why do we have to check here whether a card is on a young region,
 626   // given that we dirty young regions and, as a result, the
 627   // post-barrier is supposed to filter them out and never to enqueue
 628   // them? When we allocate a new region as the "allocation region" we
 629   // actually dirty its cards after we release the lock, since card
 630   // dirtying while holding the lock was a performance bottleneck. So,
 631   // as a result, it is possible for other threads to actually
 632   // allocate objects in the region (after the acquire the lock)
 633   // before all the cards on the region are dirtied. This is unlikely,
 634   // and it doesn't happen often, but it can happen. So, the extra
 635   // check below filters out those cards.
 636   if (r->is_young()) {
 637     return false;
 638   }
 639 
 640   // While we are processing RSet buffers during the collection, we
 641   // actually don't want to scan any cards on the collection set,
 642   // since we don't want to update remembered sets with entries that
 643   // point into the collection set, given that live objects from the
 644   // collection set are about to move and such entries will be stale
 645   // very soon. This change also deals with a reliability issue which
 646   // involves scanning a card in the collection set and coming across
 647   // an array that was being chunked and looking malformed. Note,
 648   // however, that if evacuation fails, we have to scan any objects
 649   // that were not moved and create any missing entries.
 650   if (r->in_collection_set()) {
 651     return false;
 652   }
 653 
 654   // The result from the hot card cache insert call is either:
 655   //   * pointer to the current card
 656   //     (implying that the current card is not 'hot'),
 657   //   * null
 658   //     (meaning we had inserted the card ptr into the "hot" card cache,
 659   //     which had some headroom),
 660   //   * a pointer to a "hot" card that was evicted from the "hot" cache.
 661   //
 662 
 663   G1HotCardCache* hot_card_cache = _cg1r->hot_card_cache();
 664   if (hot_card_cache->use_cache()) {
 665     assert(!check_for_refs_into_cset, "sanity");
 666     assert(!SafepointSynchronize::is_at_safepoint(), "sanity");
 667 
 668     card_ptr = hot_card_cache->insert(card_ptr);
 669     if (card_ptr == NULL) {
 670       // There was no eviction. Nothing to do.
 671       return false;
 672     }
 673 
 674     start = _ct_bs->addr_for(card_ptr);
 675     r = _g1->heap_region_containing(start);
 676 
 677     // Checking whether the region we got back from the cache
 678     // is young here is inappropriate. The region could have been
 679     // freed, reallocated and tagged as young while in the cache.
 680     // Hence we could see its young type change at any time.
 681   }
 682 
 683   // Don't use addr_for(card_ptr + 1) which can ask for
 684   // a card beyond the heap.  This is not safe without a perm
 685   // gen at the upper end of the heap.
 686   HeapWord* end   = start + CardTableModRefBS::card_size_in_words;
 687   MemRegion dirtyRegion(start, end);
 688 
 689 #if CARD_REPEAT_HISTO
 690   init_ct_freq_table(_g1->max_capacity());
 691   ct_freq_note_card(_ct_bs->index_for(start));
 692 #endif
 693 
 694   G1ParPushHeapRSClosure* oops_in_heap_closure = NULL;
 695   if (check_for_refs_into_cset) {
 696     // ConcurrentG1RefineThreads have worker numbers larger than what
 697     // _cset_rs_update_cl[] is set up to handle. But those threads should
 698     // only be active outside of a collection which means that when they
 699     // reach here they should have check_for_refs_into_cset == false.
 700     assert((size_t)worker_i < n_workers(), "index of worker larger than _cset_rs_update_cl[].length");
 701     oops_in_heap_closure = _cset_rs_update_cl[worker_i];
 702   }
 703   G1UpdateRSOrPushRefOopClosure update_rs_oop_cl(_g1,
 704                                                  _g1->g1_rem_set(),
 705                                                  oops_in_heap_closure,
 706                                                  check_for_refs_into_cset,
 707                                                  worker_i);
 708   update_rs_oop_cl.set_from(r);
 709 
 710   G1TriggerClosure trigger_cl;
 711   FilterIntoCSClosure into_cs_cl(NULL, _g1, &trigger_cl);
 712   G1InvokeIfNotTriggeredClosure invoke_cl(&trigger_cl, &into_cs_cl);
 713   G1Mux2Closure mux(&invoke_cl, &update_rs_oop_cl);
 714 
 715   FilterOutOfRegionClosure filter_then_update_rs_oop_cl(r,
 716                         (check_for_refs_into_cset ?
 717                                 (OopClosure*)&mux :
 718                                 (OopClosure*)&update_rs_oop_cl));
 719 
 720   // The region for the current card may be a young region. The
 721   // current card may have been a card that was evicted from the
 722   // card cache. When the card was inserted into the cache, we had
 723   // determined that its region was non-young. While in the cache,
 724   // the region may have been freed during a cleanup pause, reallocated
 725   // and tagged as young.
 726   //
 727   // We wish to filter out cards for such a region but the current
 728   // thread, if we're running concurrently, may "see" the young type
 729   // change at any time (so an earlier "is_young" check may pass or
 730   // fail arbitrarily). We tell the iteration code to perform this
 731   // filtering when it has been determined that there has been an actual
 732   // allocation in this region and making it safe to check the young type.
 733   bool filter_young = true;
 734 
 735   HeapWord* stop_point =
 736     r->oops_on_card_seq_iterate_careful(dirtyRegion,
 737                                         &filter_then_update_rs_oop_cl,
 738                                         filter_young,
 739                                         card_ptr);
 740 
 741   // If stop_point is non-null, then we encountered an unallocated region
 742   // (perhaps the unfilled portion of a TLAB.)  For now, we'll dirty the
 743   // card and re-enqueue: if we put off the card until a GC pause, then the
 744   // unallocated portion will be filled in.  Alternatively, we might try
 745   // the full complexity of the technique used in "regular" precleaning.
 746   if (stop_point != NULL) {
 747     // The card might have gotten re-dirtied and re-enqueued while we
 748     // worked.  (In fact, it's pretty likely.)
 749     if (*card_ptr != CardTableModRefBS::dirty_card_val()) {
 750       *card_ptr = CardTableModRefBS::dirty_card_val();
 751       MutexLockerEx x(Shared_DirtyCardQ_lock,
 752                       Mutex::_no_safepoint_check_flag);
 753       DirtyCardQueue* sdcq =
 754         JavaThread::dirty_card_queue_set().shared_dirty_card_queue();
 755       sdcq->enqueue(card_ptr);
 756     }
 757   } else {
 758     _conc_refine_cards++;
 759   }
 760 
 761   // This gets set to true if the card being refined has
 762   // references that point into the collection set.
 763   bool has_refs_into_cset = trigger_cl.triggered();
 764 
 765   // We should only be detecting that the card contains references
 766   // that point into the collection set if the current thread is
 767   // a GC worker thread.
 768   assert(!has_refs_into_cset || SafepointSynchronize::is_at_safepoint(),
 769            "invalid result at non safepoint");
 770 
 771   return has_refs_into_cset;
 772 }
 773 
 774 void G1RemSet::print_periodic_summary_info(const char* header) {
 775   G1RemSetSummary current;
 776   current.initialize(this);
 777 
 778   _prev_period_summary.subtract_from(&current);
 779   print_summary_info(&_prev_period_summary, header);
 780 
 781   _prev_period_summary.set(&current);
 782 }
 783 
 784 void G1RemSet::print_summary_info() {
 785   G1RemSetSummary current;
 786   current.initialize(this);
 787 
 788   print_summary_info(&current, " Cumulative RS summary");
 789 }
 790 
 791 void G1RemSet::print_summary_info(G1RemSetSummary * summary, const char * header) {
 792   assert(summary != NULL, "just checking");
 793 
 794   if (header != NULL) {
 795     gclog_or_tty->print_cr("%s", header);
 796   }
 797 
 798 #if CARD_REPEAT_HISTO
 799   gclog_or_tty->print_cr("\nG1 card_repeat count histogram: ");
 800   gclog_or_tty->print_cr("  # of repeats --> # of cards with that number.");
 801   card_repeat_count.print_on(gclog_or_tty);
 802 #endif
 803 
 804   summary->print_on(gclog_or_tty);
 805 }
 806 
 807 void G1RemSet::prepare_for_verify() {
 808   if (G1HRRSFlushLogBuffersOnVerify &&
 809       (VerifyBeforeGC || VerifyAfterGC)
 810       &&  (!_g1->collector_state()->full_collection() || G1VerifyRSetsDuringFullGC)) {
 811     cleanupHRRS();
 812     _g1->set_refine_cte_cl_concurrency(false);
 813     if (SafepointSynchronize::is_at_safepoint()) {
 814       DirtyCardQueueSet& dcqs = JavaThread::dirty_card_queue_set();
 815       dcqs.concatenate_logs();
 816     }
 817 
 818     G1HotCardCache* hot_card_cache = _cg1r->hot_card_cache();
 819     bool use_hot_card_cache = hot_card_cache->use_cache();
 820     hot_card_cache->set_use_cache(false);
 821 
 822     DirtyCardQueue into_cset_dcq(&_g1->into_cset_dirty_card_queue_set());
 823     updateRS(&into_cset_dcq, 0);
 824     _g1->into_cset_dirty_card_queue_set().clear();
 825 
 826     hot_card_cache->set_use_cache(use_hot_card_cache);
 827     assert(JavaThread::dirty_card_queue_set().completed_buffers_num() == 0, "All should be consumed");
 828   }
 829 }