1 /*
   2  * Copyright (c) 2001, 2011, 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/bufferingOopClosure.hpp"
  27 #include "gc_implementation/g1/concurrentG1Refine.hpp"
  28 #include "gc_implementation/g1/concurrentG1RefineThread.hpp"
  29 #include "gc_implementation/g1/g1BlockOffsetTable.inline.hpp"
  30 #include "gc_implementation/g1/g1CollectedHeap.inline.hpp"
  31 #include "gc_implementation/g1/g1CollectorPolicy.hpp"
  32 #include "gc_implementation/g1/g1OopClosures.inline.hpp"
  33 #include "gc_implementation/g1/g1RemSet.inline.hpp"
  34 #include "gc_implementation/g1/heapRegionSeq.inline.hpp"
  35 #include "memory/iterator.hpp"
  36 #include "oops/oop.inline.hpp"
  37 #include "utilities/intHisto.hpp"
  38 
  39 #define CARD_REPEAT_HISTO 0
  40 
  41 #if CARD_REPEAT_HISTO
  42 static size_t ct_freq_sz;
  43 static jbyte* ct_freq = NULL;
  44 
  45 void init_ct_freq_table(size_t heap_sz_bytes) {
  46   if (ct_freq == NULL) {
  47     ct_freq_sz = heap_sz_bytes/CardTableModRefBS::card_size;
  48     ct_freq = new jbyte[ct_freq_sz];
  49     for (size_t j = 0; j < ct_freq_sz; j++) ct_freq[j] = 0;
  50   }
  51 }
  52 
  53 void ct_freq_note_card(size_t index) {
  54   assert(0 <= index && index < ct_freq_sz, "Bounds error.");
  55   if (ct_freq[index] < 100) { ct_freq[index]++; }
  56 }
  57 
  58 static IntHistogram card_repeat_count(10, 10);
  59 
  60 void ct_freq_update_histo_and_reset() {
  61   for (size_t j = 0; j < ct_freq_sz; j++) {
  62     card_repeat_count.add_entry(ct_freq[j]);
  63     ct_freq[j] = 0;
  64   }
  65 
  66 }
  67 #endif
  68 
  69 G1RemSet::G1RemSet(G1CollectedHeap* g1, CardTableModRefBS* ct_bs)
  70   : _g1(g1), _conc_refine_cards(0),
  71     _ct_bs(ct_bs), _g1p(_g1->g1_policy()),
  72     _cg1r(g1->concurrent_g1_refine()),
  73     _cset_rs_update_cl(NULL),
  74     _cards_scanned(NULL), _total_cards_scanned(0)
  75 {
  76   _seq_task = new SubTasksDone(NumSeqTasks);
  77   guarantee(n_workers() > 0, "There should be some workers");
  78   _cset_rs_update_cl = NEW_C_HEAP_ARRAY(OopsInHeapRegionClosure*, n_workers());
  79   for (uint i = 0; i < n_workers(); i++) {
  80     _cset_rs_update_cl[i] = NULL;
  81   }
  82 }
  83 
  84 G1RemSet::~G1RemSet() {
  85   delete _seq_task;
  86   for (uint i = 0; i < n_workers(); i++) {
  87     assert(_cset_rs_update_cl[i] == NULL, "it should be");
  88   }
  89   FREE_C_HEAP_ARRAY(OopsInHeapRegionClosure*, _cset_rs_update_cl);
  90 }
  91 
  92 void CountNonCleanMemRegionClosure::do_MemRegion(MemRegion mr) {
  93   if (_g1->is_in_g1_reserved(mr.start())) {
  94     _n += (int) ((mr.byte_size() / CardTableModRefBS::card_size));
  95     if (_start_first == NULL) _start_first = mr.start();
  96   }
  97 }
  98 
  99 class ScanRSClosure : public HeapRegionClosure {
 100   size_t _cards_done, _cards;
 101   G1CollectedHeap* _g1h;
 102   OopsInHeapRegionClosure* _oc;
 103   G1BlockOffsetSharedArray* _bot_shared;
 104   CardTableModRefBS *_ct_bs;
 105   int _worker_i;
 106   int _block_size;
 107   bool _try_claimed;
 108 public:
 109   ScanRSClosure(OopsInHeapRegionClosure* oc, int worker_i) :
 110     _oc(oc),
 111     _cards(0),
 112     _cards_done(0),
 113     _worker_i(worker_i),
 114     _try_claimed(false)
 115   {
 116     _g1h = G1CollectedHeap::heap();
 117     _bot_shared = _g1h->bot_shared();
 118     _ct_bs = (CardTableModRefBS*) (_g1h->barrier_set());
 119     _block_size = MAX2<int>(G1RSetScanBlockSize, 1);
 120   }
 121 
 122   void set_try_claimed() { _try_claimed = true; }
 123 
 124   void scanCard(size_t index, HeapRegion *r) {
 125     DirtyCardToOopClosure* cl =
 126       r->new_dcto_closure(_oc,
 127                          CardTableModRefBS::Precise,
 128                          HeapRegionDCTOC::IntoCSFilterKind);
 129 
 130     // Set the "from" region in the closure.
 131     _oc->set_region(r);
 132     HeapWord* card_start = _bot_shared->address_for_index(index);
 133     HeapWord* card_end = card_start + G1BlockOffsetSharedArray::N_words;
 134     Space *sp = SharedHeap::heap()->space_containing(card_start);
 135     MemRegion sm_region = sp->used_region_at_save_marks();
 136     MemRegion mr = sm_region.intersection(MemRegion(card_start,card_end));
 137     if (!mr.is_empty() && !_ct_bs->is_card_claimed(index)) {
 138       // We make the card as "claimed" lazily (so races are possible
 139       // but they're benign), which reduces the number of duplicate
 140       // scans (the rsets of the regions in the cset can intersect).
 141       _ct_bs->set_card_claimed(index);
 142       _cards_done++;
 143       cl->do_MemRegion(mr);
 144     }
 145   }
 146 
 147   void printCard(HeapRegion* card_region, size_t card_index,
 148                  HeapWord* card_start) {
 149     gclog_or_tty->print_cr("T %d Region [" PTR_FORMAT ", " PTR_FORMAT ") "
 150                            "RS names card %p: "
 151                            "[" PTR_FORMAT ", " PTR_FORMAT ")",
 152                            _worker_i,
 153                            card_region->bottom(), card_region->end(),
 154                            card_index,
 155                            card_start, card_start + G1BlockOffsetSharedArray::N_words);
 156   }
 157 
 158   bool doHeapRegion(HeapRegion* r) {
 159     assert(r->in_collection_set(), "should only be called on elements of CS.");
 160     HeapRegionRemSet* hrrs = r->rem_set();
 161     if (hrrs->iter_is_complete()) return false; // All done.
 162     if (!_try_claimed && !hrrs->claim_iter()) return false;
 163     // If we ever free the collection set concurrently, we should also
 164     // clear the card table concurrently therefore we won't need to
 165     // add regions of the collection set to the dirty cards region.
 166     _g1h->push_dirty_cards_region(r);
 167     // If we didn't return above, then
 168     //   _try_claimed || r->claim_iter()
 169     // is true: either we're supposed to work on claimed-but-not-complete
 170     // regions, or we successfully claimed the region.
 171     HeapRegionRemSetIterator* iter = _g1h->rem_set_iterator(_worker_i);
 172     hrrs->init_iterator(iter);
 173     size_t card_index;
 174 
 175     // We claim cards in block so as to recude the contention. The block size is determined by
 176     // the G1RSetScanBlockSize parameter.
 177     size_t jump_to_card = hrrs->iter_claimed_next(_block_size);
 178     for (size_t current_card = 0; iter->has_next(card_index); current_card++) {
 179       if (current_card >= jump_to_card + _block_size) {
 180         jump_to_card = hrrs->iter_claimed_next(_block_size);
 181       }
 182       if (current_card < jump_to_card) continue;
 183       HeapWord* card_start = _g1h->bot_shared()->address_for_index(card_index);
 184 #if 0
 185       gclog_or_tty->print("Rem set iteration yielded card [" PTR_FORMAT ", " PTR_FORMAT ").\n",
 186                           card_start, card_start + CardTableModRefBS::card_size_in_words);
 187 #endif
 188 
 189       HeapRegion* card_region = _g1h->heap_region_containing(card_start);
 190       assert(card_region != NULL, "Yielding cards not in the heap?");
 191       _cards++;
 192 
 193       if (!card_region->is_on_dirty_cards_region_list()) {
 194         _g1h->push_dirty_cards_region(card_region);
 195       }
 196 
 197       // If the card is dirty, then we will scan it during updateRS.
 198       if (!card_region->in_collection_set() &&
 199           !_ct_bs->is_card_dirty(card_index)) {
 200         scanCard(card_index, card_region);
 201       }
 202     }
 203     if (!_try_claimed) {
 204       hrrs->set_iter_complete();
 205     }
 206     return false;
 207   }
 208   size_t cards_done() { return _cards_done;}
 209   size_t cards_looked_up() { return _cards;}
 210 };
 211 
 212 // We want the parallel threads to start their scanning at
 213 // different collection set regions to avoid contention.
 214 // If we have:
 215 //          n collection set regions
 216 //          p threads
 217 // Then thread t will start at region t * floor (n/p)
 218 
 219 HeapRegion* G1RemSet::calculateStartRegion(int worker_i) {
 220   HeapRegion* result = _g1p->collection_set();
 221   if (ParallelGCThreads > 0) {
 222     size_t cs_size = _g1p->collection_set_size();
 223     int n_workers = _g1->workers()->total_workers();
 224     size_t cs_spans = cs_size / n_workers;
 225     size_t ind      = cs_spans * worker_i;
 226     for (size_t i = 0; i < ind; i++)
 227       result = result->next_in_collection_set();
 228   }
 229   return result;
 230 }
 231 
 232 void G1RemSet::scanRS(OopsInHeapRegionClosure* oc, int worker_i) {
 233   double rs_time_start = os::elapsedTime();
 234   HeapRegion *startRegion = calculateStartRegion(worker_i);
 235 
 236   ScanRSClosure scanRScl(oc, worker_i);
 237   _g1->collection_set_iterate_from(startRegion, &scanRScl);
 238   scanRScl.set_try_claimed();
 239   _g1->collection_set_iterate_from(startRegion, &scanRScl);
 240 
 241   double scan_rs_time_sec = os::elapsedTime() - rs_time_start;
 242 
 243   assert( _cards_scanned != NULL, "invariant" );
 244   _cards_scanned[worker_i] = scanRScl.cards_done();
 245 
 246   _g1p->record_scan_rs_time(worker_i, scan_rs_time_sec * 1000.0);
 247 }
 248 
 249 // Closure used for updating RSets and recording references that
 250 // point into the collection set. Only called during an
 251 // evacuation pause.
 252 
 253 class RefineRecordRefsIntoCSCardTableEntryClosure: public CardTableEntryClosure {
 254   G1RemSet* _g1rs;
 255   DirtyCardQueue* _into_cset_dcq;
 256 public:
 257   RefineRecordRefsIntoCSCardTableEntryClosure(G1CollectedHeap* g1h,
 258                                               DirtyCardQueue* into_cset_dcq) :
 259     _g1rs(g1h->g1_rem_set()), _into_cset_dcq(into_cset_dcq)
 260   {}
 261   bool do_card_ptr(jbyte* card_ptr, int worker_i) {
 262     // The only time we care about recording cards that
 263     // contain references that point into the collection set
 264     // is during RSet updating within an evacuation pause.
 265     // In this case worker_i should be the id of a GC worker thread.
 266     assert(SafepointSynchronize::is_at_safepoint(), "not during an evacuation pause");
 267     assert(worker_i < (int) (ParallelGCThreads == 0 ? 1 : ParallelGCThreads), "should be a GC worker");
 268 
 269     if (_g1rs->concurrentRefineOneCard(card_ptr, worker_i, true)) {
 270       // 'card_ptr' contains references that point into the collection
 271       // set. We need to record the card in the DCQS
 272       // (G1CollectedHeap::into_cset_dirty_card_queue_set())
 273       // that's used for that purpose.
 274       //
 275       // Enqueue the card
 276       _into_cset_dcq->enqueue(card_ptr);
 277     }
 278     return true;
 279   }
 280 };
 281 
 282 void G1RemSet::updateRS(DirtyCardQueue* into_cset_dcq, int worker_i) {
 283   double start = os::elapsedTime();
 284   // Apply the given closure to all remaining log entries.
 285   RefineRecordRefsIntoCSCardTableEntryClosure into_cset_update_rs_cl(_g1, into_cset_dcq);
 286   _g1->iterate_dirty_card_closure(&into_cset_update_rs_cl, into_cset_dcq, false, worker_i);
 287 
 288   // Now there should be no dirty cards.
 289   if (G1RSLogCheckCardTable) {
 290     CountNonCleanMemRegionClosure cl(_g1);
 291     _ct_bs->mod_card_iterate(&cl);
 292     // XXX This isn't true any more: keeping cards of young regions
 293     // marked dirty broke it.  Need some reasonable fix.
 294     guarantee(cl.n() == 0, "Card table should be clean.");
 295   }
 296 
 297   _g1p->record_update_rs_time(worker_i, (os::elapsedTime() - start) * 1000.0);
 298 }
 299 
 300 class CountRSSizeClosure: public HeapRegionClosure {
 301   size_t _n;
 302   size_t _tot;
 303   size_t _max;
 304   HeapRegion* _max_r;
 305   enum {
 306     N = 20,
 307     MIN = 6
 308   };
 309   int _histo[N];
 310 public:
 311   CountRSSizeClosure() : _n(0), _tot(0), _max(0), _max_r(NULL) {
 312     for (int i = 0; i < N; i++) _histo[i] = 0;
 313   }
 314   bool doHeapRegion(HeapRegion* r) {
 315     if (!r->continuesHumongous()) {
 316       size_t occ = r->rem_set()->occupied();
 317       _n++;
 318       _tot += occ;
 319       if (occ > _max) {
 320         _max = occ;
 321         _max_r = r;
 322       }
 323       // Fit it into a histo bin.
 324       int s = 1 << MIN;
 325       int i = 0;
 326       while (occ > (size_t) s && i < (N-1)) {
 327         s = s << 1;
 328         i++;
 329       }
 330       _histo[i]++;
 331     }
 332     return false;
 333   }
 334   size_t n() { return _n; }
 335   size_t tot() { return _tot; }
 336   size_t mx() { return _max; }
 337   HeapRegion* mxr() { return _max_r; }
 338   void print_histo() {
 339     int mx = N;
 340     while (mx >= 0) {
 341       if (_histo[mx-1] > 0) break;
 342       mx--;
 343     }
 344     gclog_or_tty->print_cr("Number of regions with given RS sizes:");
 345     gclog_or_tty->print_cr("           <= %8d   %8d", 1 << MIN, _histo[0]);
 346     for (int i = 1; i < mx-1; i++) {
 347       gclog_or_tty->print_cr("  %8d  - %8d   %8d",
 348                     (1 << (MIN + i - 1)) + 1,
 349                     1 << (MIN + i),
 350                     _histo[i]);
 351     }
 352     gclog_or_tty->print_cr("            > %8d   %8d", (1 << (MIN+mx-2))+1, _histo[mx-1]);
 353   }
 354 };
 355 
 356 void G1RemSet::cleanupHRRS() {
 357   HeapRegionRemSet::cleanup();
 358 }
 359 
 360 void G1RemSet::oops_into_collection_set_do(OopsInHeapRegionClosure* oc,
 361                                              int worker_i) {
 362 #if CARD_REPEAT_HISTO
 363   ct_freq_update_histo_and_reset();
 364 #endif
 365   if (worker_i == 0) {
 366     _cg1r->clear_and_record_card_counts();
 367   }
 368 
 369   // Make this into a command-line flag...
 370   if (G1RSCountHisto && (ParallelGCThreads == 0 || worker_i == 0)) {
 371     CountRSSizeClosure count_cl;
 372     _g1->heap_region_iterate(&count_cl);
 373     gclog_or_tty->print_cr("Avg of %d RS counts is %f, max is %d, "
 374                   "max region is " PTR_FORMAT,
 375                   count_cl.n(), (float)count_cl.tot()/(float)count_cl.n(),
 376                   count_cl.mx(), count_cl.mxr());
 377     count_cl.print_histo();
 378   }
 379 
 380   // We cache the value of 'oc' closure into the appropriate slot in the
 381   // _cset_rs_update_cl for this worker
 382   assert(worker_i < (int)n_workers(), "sanity");
 383   _cset_rs_update_cl[worker_i] = oc;
 384 
 385   // A DirtyCardQueue that is used to hold cards containing references
 386   // that point into the collection set. This DCQ is associated with a
 387   // special DirtyCardQueueSet (see g1CollectedHeap.hpp).  Under normal
 388   // circumstances (i.e. the pause successfully completes), these cards
 389   // are just discarded (there's no need to update the RSets of regions
 390   // that were in the collection set - after the pause these regions
 391   // are wholly 'free' of live objects. In the event of an evacuation
 392   // failure the cards/buffers in this queue set are:
 393   // * passed to the DirtyCardQueueSet that is used to manage deferred
 394   //   RSet updates, or
 395   // * scanned for references that point into the collection set
 396   //   and the RSet of the corresponding region in the collection set
 397   //   is updated immediately.
 398   DirtyCardQueue into_cset_dcq(&_g1->into_cset_dirty_card_queue_set());
 399 
 400   assert((ParallelGCThreads > 0) || worker_i == 0, "invariant");
 401 
 402   // The two flags below were introduced temporarily to serialize
 403   // the updating and scanning of remembered sets. There are some
 404   // race conditions when these two operations are done in parallel
 405   // and they are causing failures. When we resolve said race
 406   // conditions, we'll revert back to parallel remembered set
 407   // updating and scanning. See CRs 6677707 and 6677708.
 408   if (G1UseParallelRSetUpdating || (worker_i == 0)) {
 409     updateRS(&into_cset_dcq, worker_i);
 410   } else {
 411     _g1p->record_update_rs_processed_buffers(worker_i, 0.0);
 412     _g1p->record_update_rs_time(worker_i, 0.0);
 413   }
 414   if (G1UseParallelRSetScanning || (worker_i == 0)) {
 415     scanRS(oc, worker_i);
 416   } else {
 417     _g1p->record_scan_rs_time(worker_i, 0.0);
 418   }
 419 
 420   // We now clear the cached values of _cset_rs_update_cl for this worker
 421   _cset_rs_update_cl[worker_i] = NULL;
 422 }
 423 
 424 void G1RemSet::prepare_for_oops_into_collection_set_do() {
 425   cleanupHRRS();
 426   ConcurrentG1Refine* cg1r = _g1->concurrent_g1_refine();
 427   _g1->set_refine_cte_cl_concurrency(false);
 428   DirtyCardQueueSet& dcqs = JavaThread::dirty_card_queue_set();
 429   dcqs.concatenate_logs();
 430 
 431   if (ParallelGCThreads > 0) {
 432     _seq_task->set_n_threads((int)n_workers());
 433   }
 434   guarantee( _cards_scanned == NULL, "invariant" );
 435   _cards_scanned = NEW_C_HEAP_ARRAY(size_t, n_workers());
 436   for (uint i = 0; i < n_workers(); ++i) {
 437     _cards_scanned[i] = 0;
 438   }
 439   _total_cards_scanned = 0;
 440 }
 441 
 442 
 443 // This closure, applied to a DirtyCardQueueSet, is used to immediately
 444 // update the RSets for the regions in the CSet. For each card it iterates
 445 // through the oops which coincide with that card. It scans the reference
 446 // fields in each oop; when it finds an oop that points into the collection
 447 // set, the RSet for the region containing the referenced object is updated.
 448 class UpdateRSetCardTableEntryIntoCSetClosure: public CardTableEntryClosure {
 449   G1CollectedHeap* _g1;
 450   CardTableModRefBS* _ct_bs;
 451 public:
 452   UpdateRSetCardTableEntryIntoCSetClosure(G1CollectedHeap* g1,
 453                                           CardTableModRefBS* bs):
 454     _g1(g1), _ct_bs(bs)
 455   { }
 456 
 457   bool do_card_ptr(jbyte* card_ptr, int worker_i) {
 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     assert(r != NULL, "unexpected null");
 463 
 464     // Scan oops in the card looking for references into the collection set
 465     HeapWord* end   = _ct_bs->addr_for(card_ptr + 1);
 466     MemRegion scanRegion(start, end);
 467 
 468     UpdateRSetImmediate update_rs_cl(_g1->g1_rem_set());
 469     FilterIntoCSClosure update_rs_cset_oop_cl(NULL, _g1, &update_rs_cl);
 470     FilterOutOfRegionClosure filter_then_update_rs_cset_oop_cl(r, &update_rs_cset_oop_cl);
 471 
 472     // We can pass false as the "filter_young" parameter here as:
 473     // * we should be in a STW pause,
 474     // * the DCQS to which this closure is applied is used to hold
 475     //   references that point into the collection set from the prior
 476     //   RSet updating,
 477     // * the post-write barrier shouldn't be logging updates to young
 478     //   regions (but there is a situation where this can happen - see
 479     //   the comment in G1RemSet::concurrentRefineOneCard below -
 480     //   that should not be applicable here), and
 481     // * during actual RSet updating, the filtering of cards in young
 482     //   regions in HeapRegion::oops_on_card_seq_iterate_careful is
 483     //   employed.
 484     // As a result, when this closure is applied to "refs into cset"
 485     // DCQS, we shouldn't see any cards in young regions.
 486     update_rs_cl.set_region(r);
 487     HeapWord* stop_point =
 488       r->oops_on_card_seq_iterate_careful(scanRegion,
 489                                           &filter_then_update_rs_cset_oop_cl,
 490                                           false /* filter_young */,
 491                                           NULL  /* card_ptr */);
 492 
 493     // Since this is performed in the event of an evacuation failure, we
 494     // we shouldn't see a non-null stop point
 495     assert(stop_point == NULL, "saw an unallocated region");
 496     return true;
 497   }
 498 };
 499 
 500 void G1RemSet::cleanup_after_oops_into_collection_set_do() {
 501   guarantee( _cards_scanned != NULL, "invariant" );
 502   _total_cards_scanned = 0;
 503   for (uint i = 0; i < n_workers(); ++i) {
 504     _total_cards_scanned += _cards_scanned[i];
 505   }
 506   FREE_C_HEAP_ARRAY(size_t, _cards_scanned);
 507   _cards_scanned = NULL;
 508   // Cleanup after copy
 509   _g1->set_refine_cte_cl_concurrency(true);
 510   // Set all cards back to clean.
 511   _g1->cleanUpCardTable();
 512 
 513   DirtyCardQueueSet& into_cset_dcqs = _g1->into_cset_dirty_card_queue_set();
 514   int into_cset_n_buffers = into_cset_dcqs.completed_buffers_num();
 515 
 516   if (_g1->evacuation_failed()) {
 517     // Restore remembered sets for the regions pointing into the collection set.
 518 
 519     if (G1DeferredRSUpdate) {
 520       // If deferred RS updates are enabled then we just need to transfer
 521       // the completed buffers from (a) the DirtyCardQueueSet used to hold
 522       // cards that contain references that point into the collection set
 523       // to (b) the DCQS used to hold the deferred RS updates
 524       _g1->dirty_card_queue_set().merge_bufferlists(&into_cset_dcqs);
 525     } else {
 526 
 527       CardTableModRefBS* bs = (CardTableModRefBS*)_g1->barrier_set();
 528       UpdateRSetCardTableEntryIntoCSetClosure update_rs_cset_immediate(_g1, bs);
 529 
 530       int n_completed_buffers = 0;
 531       while (into_cset_dcqs.apply_closure_to_completed_buffer(&update_rs_cset_immediate,
 532                                                     0, 0, true)) {
 533         n_completed_buffers++;
 534       }
 535       assert(n_completed_buffers == into_cset_n_buffers, "missed some buffers");
 536     }
 537   }
 538 
 539   // Free any completed buffers in the DirtyCardQueueSet used to hold cards
 540   // which contain references that point into the collection.
 541   _g1->into_cset_dirty_card_queue_set().clear();
 542   assert(_g1->into_cset_dirty_card_queue_set().completed_buffers_num() == 0,
 543          "all buffers should be freed");
 544   _g1->into_cset_dirty_card_queue_set().clear_n_completed_buffers();
 545 }
 546 
 547 class ScrubRSClosure: public HeapRegionClosure {
 548   G1CollectedHeap* _g1h;
 549   BitMap* _region_bm;
 550   BitMap* _card_bm;
 551   CardTableModRefBS* _ctbs;
 552 public:
 553   ScrubRSClosure(BitMap* region_bm, BitMap* card_bm) :
 554     _g1h(G1CollectedHeap::heap()),
 555     _region_bm(region_bm), _card_bm(card_bm),
 556     _ctbs(NULL)
 557   {
 558     ModRefBarrierSet* bs = _g1h->mr_bs();
 559     guarantee(bs->is_a(BarrierSet::CardTableModRef), "Precondition");
 560     _ctbs = (CardTableModRefBS*)bs;
 561   }
 562 
 563   bool doHeapRegion(HeapRegion* r) {
 564     if (!r->continuesHumongous()) {
 565       r->rem_set()->scrub(_ctbs, _region_bm, _card_bm);
 566     }
 567     return false;
 568   }
 569 };
 570 
 571 void G1RemSet::scrub(BitMap* region_bm, BitMap* card_bm) {
 572   ScrubRSClosure scrub_cl(region_bm, card_bm);
 573   _g1->heap_region_iterate(&scrub_cl);
 574 }
 575 
 576 void G1RemSet::scrub_par(BitMap* region_bm, BitMap* card_bm,
 577                                 int worker_num, int claim_val) {
 578   ScrubRSClosure scrub_cl(region_bm, card_bm);
 579   _g1->heap_region_par_iterate_chunked(&scrub_cl, worker_num, claim_val);
 580 }
 581 
 582 
 583 static IntHistogram out_of_histo(50, 50);
 584 
 585 class TriggerClosure : public OopClosure {
 586   bool _trigger;
 587 public:
 588   TriggerClosure() : _trigger(false) { }
 589   bool value() const { return _trigger; }
 590   template <class T> void do_oop_nv(T* p) { _trigger = true; }
 591   virtual void do_oop(oop* p)        { do_oop_nv(p); }
 592   virtual void do_oop(narrowOop* p)  { do_oop_nv(p); }
 593 };
 594 
 595 class InvokeIfNotTriggeredClosure: public OopClosure {
 596   TriggerClosure* _t;
 597   OopClosure* _oc;
 598 public:
 599   InvokeIfNotTriggeredClosure(TriggerClosure* t, OopClosure* oc):
 600     _t(t), _oc(oc) { }
 601   template <class T> void do_oop_nv(T* p) {
 602     if (!_t->value()) _oc->do_oop(p);
 603   }
 604   virtual void do_oop(oop* p)        { do_oop_nv(p); }
 605   virtual void do_oop(narrowOop* p)  { do_oop_nv(p); }
 606 };
 607 
 608 class Mux2Closure : public OopClosure {
 609   OopClosure* _c1;
 610   OopClosure* _c2;
 611 public:
 612   Mux2Closure(OopClosure *c1, OopClosure *c2) : _c1(c1), _c2(c2) { }
 613   template <class T> void do_oop_nv(T* p) {
 614     _c1->do_oop(p); _c2->do_oop(p);
 615   }
 616   virtual void do_oop(oop* p)        { do_oop_nv(p); }
 617   virtual void do_oop(narrowOop* p)  { do_oop_nv(p); }
 618 };
 619 
 620 bool G1RemSet::concurrentRefineOneCard_impl(jbyte* card_ptr, int worker_i,
 621                                                    bool check_for_refs_into_cset) {
 622   // Construct the region representing the card.
 623   HeapWord* start = _ct_bs->addr_for(card_ptr);
 624   // And find the region containing it.
 625   HeapRegion* r = _g1->heap_region_containing(start);
 626   assert(r != NULL, "unexpected null");
 627 
 628   HeapWord* end   = _ct_bs->addr_for(card_ptr + 1);
 629   MemRegion dirtyRegion(start, end);
 630 
 631 #if CARD_REPEAT_HISTO
 632   init_ct_freq_table(_g1->max_capacity());
 633   ct_freq_note_card(_ct_bs->index_for(start));
 634 #endif
 635 
 636   assert(!check_for_refs_into_cset || _cset_rs_update_cl[worker_i] != NULL, "sanity");
 637   UpdateRSOrPushRefOopClosure update_rs_oop_cl(_g1,
 638                                                _g1->g1_rem_set(),
 639                                                _cset_rs_update_cl[worker_i],
 640                                                check_for_refs_into_cset,
 641                                                worker_i);
 642   update_rs_oop_cl.set_from(r);
 643 
 644   TriggerClosure trigger_cl;
 645   FilterIntoCSClosure into_cs_cl(NULL, _g1, &trigger_cl);
 646   InvokeIfNotTriggeredClosure invoke_cl(&trigger_cl, &into_cs_cl);
 647   Mux2Closure mux(&invoke_cl, &update_rs_oop_cl);
 648 
 649   FilterOutOfRegionClosure filter_then_update_rs_oop_cl(r,
 650                         (check_for_refs_into_cset ?
 651                                 (OopClosure*)&mux :
 652                                 (OopClosure*)&update_rs_oop_cl));
 653 
 654   // The region for the current card may be a young region. The
 655   // current card may have been a card that was evicted from the
 656   // card cache. When the card was inserted into the cache, we had
 657   // determined that its region was non-young. While in the cache,
 658   // the region may have been freed during a cleanup pause, reallocated
 659   // and tagged as young.
 660   //
 661   // We wish to filter out cards for such a region but the current
 662   // thread, if we're running concurrently, may "see" the young type
 663   // change at any time (so an earlier "is_young" check may pass or
 664   // fail arbitrarily). We tell the iteration code to perform this
 665   // filtering when it has been determined that there has been an actual
 666   // allocation in this region and making it safe to check the young type.
 667   bool filter_young = true;
 668 
 669   HeapWord* stop_point =
 670     r->oops_on_card_seq_iterate_careful(dirtyRegion,
 671                                         &filter_then_update_rs_oop_cl,
 672                                         filter_young,
 673                                         card_ptr);
 674 
 675   // If stop_point is non-null, then we encountered an unallocated region
 676   // (perhaps the unfilled portion of a TLAB.)  For now, we'll dirty the
 677   // card and re-enqueue: if we put off the card until a GC pause, then the
 678   // unallocated portion will be filled in.  Alternatively, we might try
 679   // the full complexity of the technique used in "regular" precleaning.
 680   if (stop_point != NULL) {
 681     // The card might have gotten re-dirtied and re-enqueued while we
 682     // worked.  (In fact, it's pretty likely.)
 683     if (*card_ptr != CardTableModRefBS::dirty_card_val()) {
 684       *card_ptr = CardTableModRefBS::dirty_card_val();
 685       MutexLockerEx x(Shared_DirtyCardQ_lock,
 686                       Mutex::_no_safepoint_check_flag);
 687       DirtyCardQueue* sdcq =
 688         JavaThread::dirty_card_queue_set().shared_dirty_card_queue();
 689       sdcq->enqueue(card_ptr);
 690     }
 691   } else {
 692     out_of_histo.add_entry(filter_then_update_rs_oop_cl.out_of_region());
 693     _conc_refine_cards++;
 694   }
 695 
 696   return trigger_cl.value();
 697 }
 698 
 699 bool G1RemSet::concurrentRefineOneCard(jbyte* card_ptr, int worker_i,
 700                                               bool check_for_refs_into_cset) {
 701   // If the card is no longer dirty, nothing to do.
 702   if (*card_ptr != CardTableModRefBS::dirty_card_val()) {
 703     // No need to return that this card contains refs that point
 704     // into the collection set.
 705     return false;
 706   }
 707 
 708   // Construct the region representing the card.
 709   HeapWord* start = _ct_bs->addr_for(card_ptr);
 710   // And find the region containing it.
 711   HeapRegion* r = _g1->heap_region_containing(start);
 712   if (r == NULL) {
 713     guarantee(_g1->is_in_permanent(start), "Or else where?");
 714     // Again no need to return that this card contains refs that
 715     // point into the collection set.
 716     return false;  // Not in the G1 heap (might be in perm, for example.)
 717   }
 718   // Why do we have to check here whether a card is on a young region,
 719   // given that we dirty young regions and, as a result, the
 720   // post-barrier is supposed to filter them out and never to enqueue
 721   // them? When we allocate a new region as the "allocation region" we
 722   // actually dirty its cards after we release the lock, since card
 723   // dirtying while holding the lock was a performance bottleneck. So,
 724   // as a result, it is possible for other threads to actually
 725   // allocate objects in the region (after the acquire the lock)
 726   // before all the cards on the region are dirtied. This is unlikely,
 727   // and it doesn't happen often, but it can happen. So, the extra
 728   // check below filters out those cards.
 729   if (r->is_young()) {
 730     return false;
 731   }
 732   // While we are processing RSet buffers during the collection, we
 733   // actually don't want to scan any cards on the collection set,
 734   // since we don't want to update remebered sets with entries that
 735   // point into the collection set, given that live objects from the
 736   // collection set are about to move and such entries will be stale
 737   // very soon. This change also deals with a reliability issue which
 738   // involves scanning a card in the collection set and coming across
 739   // an array that was being chunked and looking malformed. Note,
 740   // however, that if evacuation fails, we have to scan any objects
 741   // that were not moved and create any missing entries.
 742   if (r->in_collection_set()) {
 743     return false;
 744   }
 745 
 746   // Should we defer processing the card?
 747   //
 748   // Previously the result from the insert_cache call would be
 749   // either card_ptr (implying that card_ptr was currently "cold"),
 750   // null (meaning we had inserted the card ptr into the "hot"
 751   // cache, which had some headroom), or a "hot" card ptr
 752   // extracted from the "hot" cache.
 753   //
 754   // Now that the _card_counts cache in the ConcurrentG1Refine
 755   // instance is an evicting hash table, the result we get back
 756   // could be from evicting the card ptr in an already occupied
 757   // bucket (in which case we have replaced the card ptr in the
 758   // bucket with card_ptr and "defer" is set to false). To avoid
 759   // having a data structure (updates to which would need a lock)
 760   // to hold these unprocessed dirty cards, we need to immediately
 761   // process card_ptr. The actions needed to be taken on return
 762   // from cache_insert are summarized in the following table:
 763   //
 764   // res      defer   action
 765   // --------------------------------------------------------------
 766   // null     false   card evicted from _card_counts & replaced with
 767   //                  card_ptr; evicted ptr added to hot cache.
 768   //                  No need to process res; immediately process card_ptr
 769   //
 770   // null     true    card not evicted from _card_counts; card_ptr added
 771   //                  to hot cache.
 772   //                  Nothing to do.
 773   //
 774   // non-null false   card evicted from _card_counts & replaced with
 775   //                  card_ptr; evicted ptr is currently "cold" or
 776   //                  caused an eviction from the hot cache.
 777   //                  Immediately process res; process card_ptr.
 778   //
 779   // non-null true    card not evicted from _card_counts; card_ptr is
 780   //                  currently cold, or caused an eviction from hot
 781   //                  cache.
 782   //                  Immediately process res; no need to process card_ptr.
 783 
 784 
 785   jbyte* res = card_ptr;
 786   bool defer = false;
 787 
 788   // This gets set to true if the card being refined has references
 789   // that point into the collection set.
 790   bool oops_into_cset = false;
 791 
 792   if (_cg1r->use_cache()) {
 793     jbyte* res = _cg1r->cache_insert(card_ptr, &defer);
 794     if (res != NULL && (res != card_ptr || defer)) {
 795       start = _ct_bs->addr_for(res);
 796       r = _g1->heap_region_containing(start);
 797       if (r == NULL) {
 798         assert(_g1->is_in_permanent(start), "Or else where?");
 799       } else {
 800         // Checking whether the region we got back from the cache
 801         // is young here is inappropriate. The region could have been
 802         // freed, reallocated and tagged as young while in the cache.
 803         // Hence we could see its young type change at any time.
 804         //
 805         // Process card pointer we get back from the hot card cache. This
 806         // will check whether the region containing the card is young
 807         // _after_ checking that the region has been allocated from.
 808         oops_into_cset = concurrentRefineOneCard_impl(res, worker_i,
 809                                                       false /* check_for_refs_into_cset */);
 810         // The above call to concurrentRefineOneCard_impl is only
 811         // performed if the hot card cache is enabled. This cache is
 812         // disabled during an evacuation pause - which is the only
 813         // time when we need know if the card contains references
 814         // that point into the collection set. Also when the hot card
 815         // cache is enabled, this code is executed by the concurrent
 816         // refine threads - rather than the GC worker threads - and
 817         // concurrentRefineOneCard_impl will return false.
 818         assert(!oops_into_cset, "should not see true here");
 819       }
 820     }
 821   }
 822 
 823   if (!defer) {
 824     oops_into_cset =
 825       concurrentRefineOneCard_impl(card_ptr, worker_i, check_for_refs_into_cset);
 826     // We should only be detecting that the card contains references
 827     // that point into the collection set if the current thread is
 828     // a GC worker thread.
 829     assert(!oops_into_cset || SafepointSynchronize::is_at_safepoint(),
 830            "invalid result at non safepoint");
 831   }
 832   return oops_into_cset;
 833 }
 834 
 835 class HRRSStatsIter: public HeapRegionClosure {
 836   size_t _occupied;
 837   size_t _total_mem_sz;
 838   size_t _max_mem_sz;
 839   HeapRegion* _max_mem_sz_region;
 840 public:
 841   HRRSStatsIter() :
 842     _occupied(0),
 843     _total_mem_sz(0),
 844     _max_mem_sz(0),
 845     _max_mem_sz_region(NULL)
 846   {}
 847 
 848   bool doHeapRegion(HeapRegion* r) {
 849     if (r->continuesHumongous()) return false;
 850     size_t mem_sz = r->rem_set()->mem_size();
 851     if (mem_sz > _max_mem_sz) {
 852       _max_mem_sz = mem_sz;
 853       _max_mem_sz_region = r;
 854     }
 855     _total_mem_sz += mem_sz;
 856     size_t occ = r->rem_set()->occupied();
 857     _occupied += occ;
 858     return false;
 859   }
 860   size_t total_mem_sz() { return _total_mem_sz; }
 861   size_t max_mem_sz() { return _max_mem_sz; }
 862   size_t occupied() { return _occupied; }
 863   HeapRegion* max_mem_sz_region() { return _max_mem_sz_region; }
 864 };
 865 
 866 class PrintRSThreadVTimeClosure : public ThreadClosure {
 867 public:
 868   virtual void do_thread(Thread *t) {
 869     ConcurrentG1RefineThread* crt = (ConcurrentG1RefineThread*) t;
 870     gclog_or_tty->print("    %5.2f", crt->vtime_accum());
 871   }
 872 };
 873 
 874 void G1RemSet::print_summary_info() {
 875   G1CollectedHeap* g1 = G1CollectedHeap::heap();
 876 
 877 #if CARD_REPEAT_HISTO
 878   gclog_or_tty->print_cr("\nG1 card_repeat count histogram: ");
 879   gclog_or_tty->print_cr("  # of repeats --> # of cards with that number.");
 880   card_repeat_count.print_on(gclog_or_tty);
 881 #endif
 882 
 883   if (FILTEROUTOFREGIONCLOSURE_DOHISTOGRAMCOUNT) {
 884     gclog_or_tty->print_cr("\nG1 rem-set out-of-region histogram: ");
 885     gclog_or_tty->print_cr("  # of CS ptrs --> # of cards with that number.");
 886     out_of_histo.print_on(gclog_or_tty);
 887   }
 888   gclog_or_tty->print_cr("\n Concurrent RS processed %d cards",
 889                          _conc_refine_cards);
 890   DirtyCardQueueSet& dcqs = JavaThread::dirty_card_queue_set();
 891   jint tot_processed_buffers =
 892     dcqs.processed_buffers_mut() + dcqs.processed_buffers_rs_thread();
 893   gclog_or_tty->print_cr("  Of %d completed buffers:", tot_processed_buffers);
 894   gclog_or_tty->print_cr("     %8d (%5.1f%%) by conc RS threads.",
 895                 dcqs.processed_buffers_rs_thread(),
 896                 100.0*(float)dcqs.processed_buffers_rs_thread()/
 897                 (float)tot_processed_buffers);
 898   gclog_or_tty->print_cr("     %8d (%5.1f%%) by mutator threads.",
 899                 dcqs.processed_buffers_mut(),
 900                 100.0*(float)dcqs.processed_buffers_mut()/
 901                 (float)tot_processed_buffers);
 902   gclog_or_tty->print_cr("  Conc RS threads times(s)");
 903   PrintRSThreadVTimeClosure p;
 904   gclog_or_tty->print("     ");
 905   g1->concurrent_g1_refine()->threads_do(&p);
 906   gclog_or_tty->print_cr("");
 907 
 908   HRRSStatsIter blk;
 909   g1->heap_region_iterate(&blk);
 910   gclog_or_tty->print_cr("  Total heap region rem set sizes = " SIZE_FORMAT "K."
 911                          "  Max = " SIZE_FORMAT "K.",
 912                          blk.total_mem_sz()/K, blk.max_mem_sz()/K);
 913   gclog_or_tty->print_cr("  Static structures = " SIZE_FORMAT "K,"
 914                          " free_lists = " SIZE_FORMAT "K.",
 915                          HeapRegionRemSet::static_mem_size()/K,
 916                          HeapRegionRemSet::fl_mem_size()/K);
 917   gclog_or_tty->print_cr("    %d occupied cards represented.",
 918                          blk.occupied());
 919   gclog_or_tty->print_cr("    Max sz region = [" PTR_FORMAT ", " PTR_FORMAT " )"
 920                          ", cap = " SIZE_FORMAT "K, occ = " SIZE_FORMAT "K.",
 921                          blk.max_mem_sz_region()->bottom(), blk.max_mem_sz_region()->end(),
 922                          (blk.max_mem_sz_region()->rem_set()->mem_size() + K - 1)/K,
 923                          (blk.max_mem_sz_region()->rem_set()->occupied() + K - 1)/K);
 924   gclog_or_tty->print_cr("    Did %d coarsenings.", HeapRegionRemSet::n_coarsenings());
 925 }
 926 
 927 void G1RemSet::prepare_for_verify() {
 928   if (G1HRRSFlushLogBuffersOnVerify &&
 929       (VerifyBeforeGC || VerifyAfterGC)
 930       &&  !_g1->full_collection()) {
 931     cleanupHRRS();
 932     _g1->set_refine_cte_cl_concurrency(false);
 933     if (SafepointSynchronize::is_at_safepoint()) {
 934       DirtyCardQueueSet& dcqs = JavaThread::dirty_card_queue_set();
 935       dcqs.concatenate_logs();
 936     }
 937     bool cg1r_use_cache = _cg1r->use_cache();
 938     _cg1r->set_use_cache(false);
 939     DirtyCardQueue into_cset_dcq(&_g1->into_cset_dirty_card_queue_set());
 940     updateRS(&into_cset_dcq, 0);
 941     _g1->into_cset_dirty_card_queue_set().clear();
 942     _cg1r->set_use_cache(cg1r_use_cache);
 943 
 944     assert(JavaThread::dirty_card_queue_set().completed_buffers_num() == 0, "All should be consumed");
 945   }
 946 }