g1CollectedHeap.hpp 67.6 KB
Newer Older
1
/*
2
 * Copyright (c) 2001, 2010, Oracle and/or its affiliates. All rights reserved.
3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18
 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
 *
 * This code is free software; you can redistribute it and/or modify it
 * under the terms of the GNU General Public License version 2 only, as
 * published by the Free Software Foundation.
 *
 * This code is distributed in the hope that it will be useful, but WITHOUT
 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
 * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
 * version 2 for more details (a copy is included in the LICENSE file that
 * accompanied this code).
 *
 * You should have received a copy of the GNU General Public License version
 * 2 along with this work; if not, write to the Free Software Foundation,
 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
 *
19 20 21
 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
 * or visit www.oracle.com if you need additional information or have any
 * questions.
22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48
 *
 */

// A "G1CollectedHeap" is an implementation of a java heap for HotSpot.
// It uses the "Garbage First" heap organization and algorithm, which
// may combine concurrent marking with parallel, incremental compaction of
// heap subsets that will yield large amounts of garbage.

class HeapRegion;
class HeapRegionSeq;
class PermanentGenerationSpec;
class GenerationSpec;
class OopsInHeapRegionClosure;
class G1ScanHeapEvacClosure;
class ObjectClosure;
class SpaceClosure;
class CompactibleSpaceClosure;
class Space;
class G1CollectorPolicy;
class GenRemSet;
class G1RemSet;
class HeapRegionRemSetIterator;
class ConcurrentMark;
class ConcurrentMarkThread;
class ConcurrentG1Refine;
class ConcurrentZFThread;

49
typedef OverflowTaskQueue<StarTask>         RefToScanQueue;
50
typedef GenericTaskQueueSet<RefToScanQueue> RefToScanQueueSet;
51

52 53 54
typedef int RegionIdx_t;   // needs to hold [ 0..max_regions() )
typedef int CardIdx_t;     // needs to hold [ 0..CardsPerRegion )

55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73
enum G1GCThreadGroups {
  G1CRGroup = 0,
  G1ZFGroup = 1,
  G1CMGroup = 2,
  G1CLGroup = 3
};

enum GCAllocPurpose {
  GCAllocForTenured,
  GCAllocForSurvived,
  GCAllocPurposeCount
};

class YoungList : public CHeapObj {
private:
  G1CollectedHeap* _g1h;

  HeapRegion* _head;

74 75
  HeapRegion* _survivor_head;
  HeapRegion* _survivor_tail;
76 77 78

  HeapRegion* _curr;

79
  size_t      _length;
80 81
  size_t      _survivor_length;

82 83 84 85
  size_t      _last_sampled_rs_lengths;
  size_t      _sampled_rs_lengths;

  void         empty_list(HeapRegion* list);
86 87 88 89

public:
  YoungList(G1CollectedHeap* g1h);

90 91 92 93 94 95 96
  void         push_region(HeapRegion* hr);
  void         add_survivor_region(HeapRegion* hr);

  void         empty_list();
  bool         is_empty() { return _length == 0; }
  size_t       length() { return _length; }
  size_t       survivor_length() { return _survivor_length; }
97 98 99 100 101 102 103 104 105 106 107 108

  void rs_length_sampling_init();
  bool rs_length_sampling_more();
  void rs_length_sampling_next();

  void reset_sampled_info() {
    _last_sampled_rs_lengths =   0;
  }
  size_t sampled_rs_lengths() { return _last_sampled_rs_lengths; }

  // for development purposes
  void reset_auxilary_lists();
109 110 111 112 113 114 115 116
  void clear() { _head = NULL; _length = 0; }

  void clear_survivors() {
    _survivor_head    = NULL;
    _survivor_tail    = NULL;
    _survivor_length  = 0;
  }

117 118
  HeapRegion* first_region() { return _head; }
  HeapRegion* first_survivor_region() { return _survivor_head; }
119
  HeapRegion* last_survivor_region() { return _survivor_tail; }
120 121 122

  // debugging
  bool          check_list_well_formed();
123
  bool          check_list_empty(bool check_sample = true);
124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148
  void          print();
};

class RefineCardTableEntryClosure;
class G1CollectedHeap : public SharedHeap {
  friend class VM_G1CollectForAllocation;
  friend class VM_GenCollectForPermanentAllocation;
  friend class VM_G1CollectFull;
  friend class VM_G1IncCollectionPause;
  friend class VMStructs;

  // Closures used in implementation.
  friend class G1ParCopyHelper;
  friend class G1IsAliveClosure;
  friend class G1EvacuateFollowersClosure;
  friend class G1ParScanThreadState;
  friend class G1ParScanClosureSuper;
  friend class G1ParEvacuateFollowersClosure;
  friend class G1ParTask;
  friend class G1FreeGarbageRegionClosure;
  friend class RefineCardTableEntryClosure;
  friend class G1PrepareCompactClosure;
  friend class RegionSorter;
  friend class CountRCClosure;
  friend class EvacPopObjClosure;
149
  friend class G1ParCleanupCTTask;
150 151 152 153 154 155 156 157

  // Other related classes.
  friend class G1MarkSweep;

private:
  // The one and only G1CollectedHeap, so static functions can find it.
  static G1CollectedHeap* _g1h;

158 159
  static size_t _humongous_object_threshold_in_words;

160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200
  // Storage for the G1 heap (excludes the permanent generation).
  VirtualSpace _g1_storage;
  MemRegion    _g1_reserved;

  // The part of _g1_storage that is currently committed.
  MemRegion _g1_committed;

  // The maximum part of _g1_storage that has ever been committed.
  MemRegion _g1_max_committed;

  // The number of regions that are completely free.
  size_t _free_regions;

  // The number of regions we could create by expansion.
  size_t _expansion_regions;

  // Return the number of free regions in the heap (by direct counting.)
  size_t count_free_regions();
  // Return the number of free regions on the free and unclean lists.
  size_t count_free_regions_list();

  // The block offset table for the G1 heap.
  G1BlockOffsetSharedArray* _bot_shared;

  // Move all of the regions off the free lists, then rebuild those free
  // lists, before and after full GC.
  void tear_down_region_lists();
  void rebuild_region_lists();
  // This sets all non-empty regions to need zero-fill (which they will if
  // they are empty after full collection.)
  void set_used_regions_to_need_zero_fill();

  // The sequence of all heap regions in the heap.
  HeapRegionSeq* _hrs;

  // The region from which normal-sized objects are currently being
  // allocated.  May be NULL.
  HeapRegion* _cur_alloc_region;

  // Postcondition: cur_alloc_region == NULL.
  void abandon_cur_alloc_region();
201
  void abandon_gc_alloc_regions();
202 203 204 205

  // The to-space memory regions into which objects are being copied during
  // a GC.
  HeapRegion* _gc_alloc_regions[GCAllocPurposeCount];
206
  size_t _gc_alloc_region_counts[GCAllocPurposeCount];
207 208 209 210 211 212 213 214
  // These are the regions, one per GCAllocPurpose, that are half-full
  // at the end of a collection and that we want to reuse during the
  // next collection.
  HeapRegion* _retained_gc_alloc_regions[GCAllocPurposeCount];
  // This specifies whether we will keep the last half-full region at
  // the end of a collection so that it can be reused during the next
  // collection (this is specified per GCAllocPurpose)
  bool _retain_gc_alloc_region[GCAllocPurposeCount];
215 216 217 218 219

  // A list of the regions that have been set to be alloc regions in the
  // current collection.
  HeapRegion* _gc_alloc_region_list;

220 221 222
  // Determines PLAB size for a particular allocation purpose.
  static size_t desired_plab_sz(GCAllocPurpose purpose);

223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240
  // When called by par thread, require par_alloc_during_gc_lock() to be held.
  void push_gc_alloc_region(HeapRegion* hr);

  // This should only be called single-threaded.  Undeclares all GC alloc
  // regions.
  void forget_alloc_region_list();

  // Should be used to set an alloc region, because there's other
  // associated bookkeeping.
  void set_gc_alloc_region(int purpose, HeapRegion* r);

  // Check well-formedness of alloc region list.
  bool check_gc_alloc_regions();

  // Outside of GC pauses, the number of bytes used in all regions other
  // than the current allocation region.
  size_t _summary_bytes_used;

241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261
  // This is used for a quick test on whether a reference points into
  // the collection set or not. Basically, we have an array, with one
  // byte per region, and that byte denotes whether the corresponding
  // region is in the collection set or not. The entry corresponding
  // the bottom of the heap, i.e., region 0, is pointed to by
  // _in_cset_fast_test_base.  The _in_cset_fast_test field has been
  // biased so that it actually points to address 0 of the address
  // space, to make the test as fast as possible (we can simply shift
  // the address to address into it, instead of having to subtract the
  // bottom of the heap from the address before shifting it; basically
  // it works in the same way the card table works).
  bool* _in_cset_fast_test;

  // The allocated array used for the fast test on whether a reference
  // points into the collection set or not. This field is also used to
  // free the array.
  bool* _in_cset_fast_test_base;

  // The length of the _in_cset_fast_test_base array.
  size_t _in_cset_fast_test_length;

262
  volatile unsigned _gc_time_stamp;
263 264 265 266 267 268 269

  size_t* _surviving_young_words;

  void setup_surviving_young_words();
  void update_surviving_young_words(size_t* surv_young_words);
  void cleanup_surviving_young_words();

270 271 272 273 274 275 276 277 278 279 280 281
  // It decides whether an explicit GC should start a concurrent cycle
  // instead of doing a STW GC. Currently, a concurrent cycle is
  // explicitly started if:
  // (a) cause == _gc_locker and +GCLockerInvokesConcurrent, or
  // (b) cause == _java_lang_system_gc and +ExplicitGCInvokesConcurrent.
  bool should_do_concurrent_full_gc(GCCause::Cause cause);

  // Keeps track of how many "full collections" (i.e., Full GCs or
  // concurrent cycles) we have completed. The number of them we have
  // started is maintained in _total_full_collections in CollectedHeap.
  volatile unsigned int _full_collections_completed;

282 283 284 285 286
protected:

  // Returns "true" iff none of the gc alloc regions have any allocations
  // since the last call to "save_marks".
  bool all_alloc_regions_no_allocs_since_save_marks();
287 288
  // Perform finalization stuff on all allocation regions.
  void retire_all_alloc_regions();
289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356

  // The number of regions allocated to hold humongous objects.
  int         _num_humongous_regions;
  YoungList*  _young_list;

  // The current policy object for the collector.
  G1CollectorPolicy* _g1_policy;

  // Parallel allocation lock to protect the current allocation region.
  Mutex  _par_alloc_during_gc_lock;
  Mutex* par_alloc_during_gc_lock() { return &_par_alloc_during_gc_lock; }

  // If possible/desirable, allocate a new HeapRegion for normal object
  // allocation sufficient for an allocation of the given "word_size".
  // If "do_expand" is true, will attempt to expand the heap if necessary
  // to to satisfy the request.  If "zero_filled" is true, requires a
  // zero-filled region.
  // (Returning NULL will trigger a GC.)
  virtual HeapRegion* newAllocRegion_work(size_t word_size,
                                          bool do_expand,
                                          bool zero_filled);

  virtual HeapRegion* newAllocRegion(size_t word_size,
                                     bool zero_filled = true) {
    return newAllocRegion_work(word_size, false, zero_filled);
  }
  virtual HeapRegion* newAllocRegionWithExpansion(int purpose,
                                                  size_t word_size,
                                                  bool zero_filled = true);

  // Attempt to allocate an object of the given (very large) "word_size".
  // Returns "NULL" on failure.
  virtual HeapWord* humongousObjAllocate(size_t word_size);

  // If possible, allocate a block of the given word_size, else return "NULL".
  // Returning NULL will trigger GC or heap expansion.
  // These two methods have rather awkward pre- and
  // post-conditions. If they are called outside a safepoint, then
  // they assume that the caller is holding the heap lock. Upon return
  // they release the heap lock, if they are returning a non-NULL
  // value. attempt_allocation_slow() also dirties the cards of a
  // newly-allocated young region after it releases the heap
  // lock. This change in interface was the neatest way to achieve
  // this card dirtying without affecting mem_allocate(), which is a
  // more frequently called method. We tried two or three different
  // approaches, but they were even more hacky.
  HeapWord* attempt_allocation(size_t word_size,
                               bool permit_collection_pause = true);

  HeapWord* attempt_allocation_slow(size_t word_size,
                                    bool permit_collection_pause = true);

  // Allocate blocks during garbage collection. Will ensure an
  // allocation region, either by picking one or expanding the
  // heap, and then allocate a block of the given size. The block
  // may not be a humongous - it must fit into a single heap region.
  HeapWord* allocate_during_gc(GCAllocPurpose purpose, size_t word_size);
  HeapWord* par_allocate_during_gc(GCAllocPurpose purpose, size_t word_size);

  HeapWord* allocate_during_gc_slow(GCAllocPurpose purpose,
                                    HeapRegion*    alloc_region,
                                    bool           par,
                                    size_t         word_size);

  // Ensure that no further allocations can happen in "r", bearing in mind
  // that parallel threads might be attempting allocations.
  void par_allocate_remaining_space(HeapRegion* r);

357 358 359 360
  // Retires an allocation region when it is full or at the end of a
  // GC pause.
  void  retire_alloc_region(HeapRegion* alloc_region, bool par);

361 362 363 364 365 366 367 368
  // - if explicit_gc is true, the GC is for a System.gc() or a heap
  // inspection request and should collect the entire heap
  // - if clear_all_soft_refs is true, all soft references are cleared
  // during the GC
  // - if explicit_gc is false, word_size describes the allocation that
  // the GC should attempt (at least) to satisfy
  void do_collection(bool explicit_gc,
                     bool clear_all_soft_refs,
369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399
                     size_t word_size);

  // Callback from VM_G1CollectFull operation.
  // Perform a full collection.
  void do_full_collection(bool clear_all_soft_refs);

  // Resize the heap if necessary after a full collection.  If this is
  // after a collect-for allocation, "word_size" is the allocation size,
  // and will be considered part of the used portion of the heap.
  void resize_if_necessary_after_full_collection(size_t word_size);

  // Callback from VM_G1CollectForAllocation operation.
  // This function does everything necessary/possible to satisfy a
  // failed allocation request (including collection, expansion, etc.)
  HeapWord* satisfy_failed_allocation(size_t word_size);

  // Attempting to expand the heap sufficiently
  // to support an allocation of the given "word_size".  If
  // successful, perform the allocation and return the address of the
  // allocated block, or else "NULL".
  virtual HeapWord* expand_and_allocate(size_t word_size);

public:
  // Expand the garbage-first heap by at least the given size (in bytes!).
  // (Rounds up to a HeapRegion boundary.)
  virtual void expand(size_t expand_bytes);

  // Do anything common to GC's.
  virtual void gc_prologue(bool full);
  virtual void gc_epilogue(bool full);

400 401 402 403 404 405
  // We register a region with the fast "in collection set" test. We
  // simply set to true the array slot corresponding to this region.
  void register_region_with_in_cset_fast_test(HeapRegion* r) {
    assert(_in_cset_fast_test_base != NULL, "sanity");
    assert(r->in_collection_set(), "invariant");
    int index = r->hrs_index();
406
    assert(0 <= index && (size_t) index < _in_cset_fast_test_length, "invariant");
407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430
    assert(!_in_cset_fast_test_base[index], "invariant");
    _in_cset_fast_test_base[index] = true;
  }

  // This is a fast test on whether a reference points into the
  // collection set or not. It does not assume that the reference
  // points into the heap; if it doesn't, it will return false.
  bool in_cset_fast_test(oop obj) {
    assert(_in_cset_fast_test != NULL, "sanity");
    if (_g1_committed.contains((HeapWord*) obj)) {
      // no need to subtract the bottom of the heap from obj,
      // _in_cset_fast_test is biased
      size_t index = ((size_t) obj) >> HeapRegion::LogOfHRGrainBytes;
      bool ret = _in_cset_fast_test[index];
      // let's make sure the result is consistent with what the slower
      // test returns
      assert( ret || !obj_in_cs(obj), "sanity");
      assert(!ret ||  obj_in_cs(obj), "sanity");
      return ret;
    } else {
      return false;
    }
  }

431 432 433 434 435 436
  void clear_cset_fast_test() {
    assert(_in_cset_fast_test_base != NULL, "sanity");
    memset(_in_cset_fast_test_base, false,
        _in_cset_fast_test_length * sizeof(bool));
  }

437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456
  // This is called at the end of either a concurrent cycle or a Full
  // GC to update the number of full collections completed. Those two
  // can happen in a nested fashion, i.e., we start a concurrent
  // cycle, a Full GC happens half-way through it which ends first,
  // and then the cycle notices that a Full GC happened and ends
  // too. The outer parameter is a boolean to help us do a bit tighter
  // consistency checking in the method. If outer is false, the caller
  // is the inner caller in the nesting (i.e., the Full GC). If outer
  // is true, the caller is the outer caller in this nesting (i.e.,
  // the concurrent cycle). Further nesting is not currently
  // supported. The end of the this call also notifies the
  // FullGCCount_lock in case a Java thread is waiting for a full GC
  // to happen (e.g., it called System.gc() with
  // +ExplicitGCInvokesConcurrent).
  void increment_full_collections_completed(bool outer);

  unsigned int full_collections_completed() {
    return _full_collections_completed;
  }

457 458 459 460 461 462 463
protected:

  // Shrink the garbage-first heap by at most the given size (in bytes!).
  // (Rounds down to a HeapRegion boundary.)
  virtual void shrink(size_t expand_bytes);
  void shrink_helper(size_t expand_bytes);

464 465 466 467 468 469
  #if TASKQUEUE_STATS
  static void print_taskqueue_stats_hdr(outputStream* const st = gclog_or_tty);
  void print_taskqueue_stats(outputStream* const st = gclog_or_tty) const;
  void reset_taskqueue_stats();
  #endif // TASKQUEUE_STATS

470 471 472 473 474
  // Do an incremental collection: identify a collection set, and evacuate
  // its live objects elsewhere.
  virtual void do_collection_pause();

  // The guts of the incremental collection pause, executed by the vm
475
  // thread.
476
  virtual void do_collection_pause_at_safepoint(double target_pause_time_ms);
477 478 479 480 481 482 483 484 485 486 487 488 489 490

  // Actually do the work of evacuating the collection set.
  virtual void evacuate_collection_set();

  // If this is an appropriate right time, do a collection pause.
  // The "word_size" argument, if non-zero, indicates the size of an
  // allocation request that is prompting this query.
  void do_collection_pause_if_appropriate(size_t word_size);

  // The g1 remembered set of the heap.
  G1RemSet* _g1_rem_set;
  // And it's mod ref barrier set, used to track updates for the above.
  ModRefBarrierSet* _mr_bs;

491 492 493 494
  // A set of cards that cover the objects for which the Rsets should be updated
  // concurrently after the collection.
  DirtyCardQueueSet _dirty_card_queue_set;

495 496 497 498 499 500 501 502 503
  // The Heap Region Rem Set Iterator.
  HeapRegionRemSetIterator** _rem_set_iterator;

  // The closure used to refine a single card.
  RefineCardTableEntryClosure* _refine_cte_cl;

  // A function to check the consistency of dirty card logs.
  void check_ct_logs_at_safepoint();

J
johnc 已提交
504 505 506 507 508 509
  // A DirtyCardQueueSet that is used to hold cards that contain
  // references into the current collection set. This is used to
  // update the remembered sets of the regions in the collection
  // set in the event of an evacuation failure.
  DirtyCardQueueSet _into_cset_dirty_card_queue_set;

510 511 512 513
  // After a collection pause, make the regions in the CS into free
  // regions.
  void free_collection_set(HeapRegion* cs_head);

514 515 516 517
  // Abandon the current collection set without recording policy
  // statistics or updating free lists.
  void abandon_collection_set(HeapRegion* cs_head);

518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625
  // Applies "scan_non_heap_roots" to roots outside the heap,
  // "scan_rs" to roots inside the heap (having done "set_region" to
  // indicate the region in which the root resides), and does "scan_perm"
  // (setting the generation to the perm generation.)  If "scan_rs" is
  // NULL, then this step is skipped.  The "worker_i"
  // param is for use with parallel roots processing, and should be
  // the "i" of the calling parallel worker thread's work(i) function.
  // In the sequential case this param will be ignored.
  void g1_process_strong_roots(bool collecting_perm_gen,
                               SharedHeap::ScanningOption so,
                               OopClosure* scan_non_heap_roots,
                               OopsInHeapRegionClosure* scan_rs,
                               OopsInGenClosure* scan_perm,
                               int worker_i);

  // Apply "blk" to all the weak roots of the system.  These include
  // JNI weak roots, the code cache, system dictionary, symbol table,
  // string table, and referents of reachable weak refs.
  void g1_process_weak_roots(OopClosure* root_closure,
                             OopClosure* non_root_closure);

  // Invoke "save_marks" on all heap regions.
  void save_marks();

  // Free a heap region.
  void free_region(HeapRegion* hr);
  // A component of "free_region", exposed for 'batching'.
  // All the params after "hr" are out params: the used bytes of the freed
  // region(s), the number of H regions cleared, the number of regions
  // freed, and pointers to the head and tail of a list of freed contig
  // regions, linked throught the "next_on_unclean_list" field.
  void free_region_work(HeapRegion* hr,
                        size_t& pre_used,
                        size_t& cleared_h,
                        size_t& freed_regions,
                        UncleanRegionList* list,
                        bool par = false);


  // The concurrent marker (and the thread it runs in.)
  ConcurrentMark* _cm;
  ConcurrentMarkThread* _cmThread;
  bool _mark_in_progress;

  // The concurrent refiner.
  ConcurrentG1Refine* _cg1r;

  // The concurrent zero-fill thread.
  ConcurrentZFThread* _czft;

  // The parallel task queues
  RefToScanQueueSet *_task_queues;

  // True iff a evacuation has failed in the current collection.
  bool _evacuation_failed;

  // Set the attribute indicating whether evacuation has failed in the
  // current collection.
  void set_evacuation_failed(bool b) { _evacuation_failed = b; }

  // Failed evacuations cause some logical from-space objects to have
  // forwarding pointers to themselves.  Reset them.
  void remove_self_forwarding_pointers();

  // When one is non-null, so is the other.  Together, they each pair is
  // an object with a preserved mark, and its mark value.
  GrowableArray<oop>*     _objs_with_preserved_marks;
  GrowableArray<markOop>* _preserved_marks_of_objs;

  // Preserve the mark of "obj", if necessary, in preparation for its mark
  // word being overwritten with a self-forwarding-pointer.
  void preserve_mark_if_necessary(oop obj, markOop m);

  // The stack of evac-failure objects left to be scanned.
  GrowableArray<oop>*    _evac_failure_scan_stack;
  // The closure to apply to evac-failure objects.

  OopsInHeapRegionClosure* _evac_failure_closure;
  // Set the field above.
  void
  set_evac_failure_closure(OopsInHeapRegionClosure* evac_failure_closure) {
    _evac_failure_closure = evac_failure_closure;
  }

  // Push "obj" on the scan stack.
  void push_on_evac_failure_scan_stack(oop obj);
  // Process scan stack entries until the stack is empty.
  void drain_evac_failure_scan_stack();
  // True iff an invocation of "drain_scan_stack" is in progress; to
  // prevent unnecessary recursion.
  bool _drain_in_progress;

  // Do any necessary initialization for evacuation-failure handling.
  // "cl" is the closure that will be used to process evac-failure
  // objects.
  void init_for_evac_failure(OopsInHeapRegionClosure* cl);
  // Do any necessary cleanup for evacuation-failure handling data
  // structures.
  void finalize_for_evac_failure();

  // An attempt to evacuate "obj" has failed; take necessary steps.
  void handle_evacuation_failure(oop obj);
  oop handle_evacuation_failure_par(OopsInHeapRegionClosure* cl, oop obj);
  void handle_evacuation_failure_common(oop obj, markOop m);


  // Ensure that the relevant gc_alloc regions are set.
  void get_gc_alloc_regions();
626 627 628 629 630 631 632 633 634 635 636 637 638 639 640
  // We're done with GC alloc regions. We are going to tear down the
  // gc alloc list and remove the gc alloc tag from all the regions on
  // that list. However, we will also retain the last (i.e., the one
  // that is half-full) GC alloc region, per GCAllocPurpose, for
  // possible reuse during the next collection, provided
  // _retain_gc_alloc_region[] indicates that it should be the
  // case. Said regions are kept in the _retained_gc_alloc_regions[]
  // array. If the parameter totally is set, we will not retain any
  // regions, irrespective of what _retain_gc_alloc_region[]
  // indicates.
  void release_gc_alloc_regions(bool totally);
#ifndef PRODUCT
  // Useful for debugging.
  void print_gc_alloc_regions();
#endif // !PRODUCT
641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660

  // ("Weak") Reference processing support
  ReferenceProcessor* _ref_processor;

  enum G1H_process_strong_roots_tasks {
    G1H_PS_mark_stack_oops_do,
    G1H_PS_refProcessor_oops_do,
    // Leave this one last.
    G1H_PS_NumElements
  };

  SubTasksDone* _process_strong_tasks;

  // List of regions which require zero filling.
  UncleanRegionList _unclean_region_list;
  bool _unclean_regions_coming;

public:
  void set_refine_cte_cl_concurrency(bool concurrent);

661
  RefToScanQueue *task_queue(int i) const;
662

663 664 665
  // A set of cards where updates happened during the GC
  DirtyCardQueueSet& dirty_card_queue_set() { return _dirty_card_queue_set; }

J
johnc 已提交
666 667 668 669 670 671 672
  // A DirtyCardQueueSet that is used to hold cards that contain
  // references into the current collection set. This is used to
  // update the remembered sets of the regions in the collection
  // set in the event of an evacuation failure.
  DirtyCardQueueSet& into_cset_dirty_card_queue_set()
        { return _into_cset_dirty_card_queue_set; }

673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718
  // Create a G1CollectedHeap with the specified policy.
  // Must call the initialize method afterwards.
  // May not return if something goes wrong.
  G1CollectedHeap(G1CollectorPolicy* policy);

  // Initialize the G1CollectedHeap to have the initial and
  // maximum sizes, permanent generation, and remembered and barrier sets
  // specified by the policy object.
  jint initialize();

  void ref_processing_init();

  void set_par_threads(int t) {
    SharedHeap::set_par_threads(t);
    _process_strong_tasks->set_par_threads(t);
  }

  virtual CollectedHeap::Name kind() const {
    return CollectedHeap::G1CollectedHeap;
  }

  // The current policy object for the collector.
  G1CollectorPolicy* g1_policy() const { return _g1_policy; }

  // Adaptive size policy.  No such thing for g1.
  virtual AdaptiveSizePolicy* size_policy() { return NULL; }

  // The rem set and barrier set.
  G1RemSet* g1_rem_set() const { return _g1_rem_set; }
  ModRefBarrierSet* mr_bs() const { return _mr_bs; }

  // The rem set iterator.
  HeapRegionRemSetIterator* rem_set_iterator(int i) {
    return _rem_set_iterator[i];
  }

  HeapRegionRemSetIterator* rem_set_iterator() {
    return _rem_set_iterator[0];
  }

  unsigned get_gc_time_stamp() {
    return _gc_time_stamp;
  }

  void reset_gc_time_stamp() {
    _gc_time_stamp = 0;
719 720 721 722 723 724
    OrderAccess::fence();
  }

  void increment_gc_time_stamp() {
    ++_gc_time_stamp;
    OrderAccess::fence();
725 726
  }

J
johnc 已提交
727 728 729
  void iterate_dirty_card_closure(CardTableEntryClosure* cl,
                                  DirtyCardQueue* into_cset_dcq,
                                  bool concurrent, int worker_i);
730 731 732 733 734 735 736 737 738

  // The shared block offset table array.
  G1BlockOffsetSharedArray* bot_shared() const { return _bot_shared; }

  // Reference Processing accessor
  ReferenceProcessor* ref_processor() { return _ref_processor; }

  // Reserved (g1 only; super method includes perm), capacity and the used
  // portion in bytes.
739
  size_t g1_reserved_obj_bytes() const { return _g1_reserved.byte_size(); }
740 741
  virtual size_t capacity() const;
  virtual size_t used() const;
742 743 744
  // This should be called when we're not holding the heap lock. The
  // result might be a bit inaccurate.
  size_t used_unlocked() const;
745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822 823 824 825 826 827 828 829 830 831 832 833 834 835 836 837 838 839 840 841 842 843 844 845 846 847 848 849 850 851 852 853 854 855 856 857 858 859 860 861 862 863 864 865 866 867 868 869 870 871 872 873 874 875 876 877 878 879 880 881 882 883 884 885 886 887 888 889 890 891 892 893 894 895 896 897 898 899 900
  size_t recalculate_used() const;
#ifndef PRODUCT
  size_t recalculate_used_regions() const;
#endif // PRODUCT

  // These virtual functions do the actual allocation.
  virtual HeapWord* mem_allocate(size_t word_size,
                                 bool   is_noref,
                                 bool   is_tlab,
                                 bool* gc_overhead_limit_was_exceeded);

  // Some heaps may offer a contiguous region for shared non-blocking
  // allocation, via inlined code (by exporting the address of the top and
  // end fields defining the extent of the contiguous allocation region.)
  // But G1CollectedHeap doesn't yet support this.

  // Return an estimate of the maximum allocation that could be performed
  // without triggering any collection or expansion activity.  In a
  // generational collector, for example, this is probably the largest
  // allocation that could be supported (without expansion) in the youngest
  // generation.  It is "unsafe" because no locks are taken; the result
  // should be treated as an approximation, not a guarantee, for use in
  // heuristic resizing decisions.
  virtual size_t unsafe_max_alloc();

  virtual bool is_maximal_no_gc() const {
    return _g1_storage.uncommitted_size() == 0;
  }

  // The total number of regions in the heap.
  size_t n_regions();

  // The number of regions that are completely free.
  size_t max_regions();

  // The number of regions that are completely free.
  size_t free_regions();

  // The number of regions that are not completely free.
  size_t used_regions() { return n_regions() - free_regions(); }

  // True iff the ZF thread should run.
  bool should_zf();

  // The number of regions available for "regular" expansion.
  size_t expansion_regions() { return _expansion_regions; }

#ifndef PRODUCT
  bool regions_accounted_for();
  bool print_region_accounting_info();
  void print_region_counts();
#endif

  HeapRegion* alloc_region_from_unclean_list(bool zero_filled);
  HeapRegion* alloc_region_from_unclean_list_locked(bool zero_filled);

  void put_region_on_unclean_list(HeapRegion* r);
  void put_region_on_unclean_list_locked(HeapRegion* r);

  void prepend_region_list_on_unclean_list(UncleanRegionList* list);
  void prepend_region_list_on_unclean_list_locked(UncleanRegionList* list);

  void set_unclean_regions_coming(bool b);
  void set_unclean_regions_coming_locked(bool b);
  // Wait for cleanup to be complete.
  void wait_for_cleanup_complete();
  // Like above, but assumes that the calling thread owns the Heap_lock.
  void wait_for_cleanup_complete_locked();

  // Return the head of the unclean list.
  HeapRegion* peek_unclean_region_list_locked();
  // Remove and return the head of the unclean list.
  HeapRegion* pop_unclean_region_list_locked();

  // List of regions which are zero filled and ready for allocation.
  HeapRegion* _free_region_list;
  // Number of elements on the free list.
  size_t _free_region_list_size;

  // If the head of the unclean list is ZeroFilled, move it to the free
  // list.
  bool move_cleaned_region_to_free_list_locked();
  bool move_cleaned_region_to_free_list();

  void put_free_region_on_list_locked(HeapRegion* r);
  void put_free_region_on_list(HeapRegion* r);

  // Remove and return the head element of the free list.
  HeapRegion* pop_free_region_list_locked();

  // If "zero_filled" is true, we first try the free list, then we try the
  // unclean list, zero-filling the result.  If "zero_filled" is false, we
  // first try the unclean list, then the zero-filled list.
  HeapRegion* alloc_free_region_from_lists(bool zero_filled);

  // Verify the integrity of the region lists.
  void remove_allocated_regions_from_lists();
  bool verify_region_lists();
  bool verify_region_lists_locked();
  size_t unclean_region_list_length();
  size_t free_region_list_length();

  // Perform a collection of the heap; intended for use in implementing
  // "System.gc".  This probably implies as full a collection as the
  // "CollectedHeap" supports.
  virtual void collect(GCCause::Cause cause);

  // The same as above but assume that the caller holds the Heap_lock.
  void collect_locked(GCCause::Cause cause);

  // This interface assumes that it's being called by the
  // vm thread. It collects the heap assuming that the
  // heap lock is already held and that we are executing in
  // the context of the vm thread.
  virtual void collect_as_vm_thread(GCCause::Cause cause);

  // True iff a evacuation has failed in the most-recent collection.
  bool evacuation_failed() { return _evacuation_failed; }

  // Free a region if it is totally full of garbage.  Returns the number of
  // bytes freed (0 ==> didn't free it).
  size_t free_region_if_totally_empty(HeapRegion *hr);
  void free_region_if_totally_empty_work(HeapRegion *hr,
                                         size_t& pre_used,
                                         size_t& cleared_h_regions,
                                         size_t& freed_regions,
                                         UncleanRegionList* list,
                                         bool par = false);

  // If we've done free region work that yields the given changes, update
  // the relevant global variables.
  void finish_free_region_work(size_t pre_used,
                               size_t cleared_h_regions,
                               size_t freed_regions,
                               UncleanRegionList* list);


  // Returns "TRUE" iff "p" points into the allocated area of the heap.
  virtual bool is_in(const void* p) const;

  // Return "TRUE" iff the given object address is within the collection
  // set.
  inline bool obj_in_cs(oop obj);

  // Return "TRUE" iff the given object address is in the reserved
  // region of g1 (excluding the permanent generation).
  bool is_in_g1_reserved(const void* p) const {
    return _g1_reserved.contains(p);
  }

  // Returns a MemRegion that corresponds to the space that  has been
  // committed in the heap
  MemRegion g1_committed() {
    return _g1_committed;
  }

Y
ysr 已提交
901
  NOT_PRODUCT(bool is_in_closed_subset(const void* p) const;)
902 903 904 905 906 907 908 909 910 911 912 913

  // Dirty card table entries covering a list of young regions.
  void dirtyCardsForYoungRegions(CardTableModRefBS* ct_bs, HeapRegion* list);

  // This resets the card table to all zeros.  It is used after
  // a collection pause which used the card table to claim cards.
  void cleanUpCardTable();

  // Iteration functions.

  // Iterate over all the ref-containing fields of all objects, calling
  // "cl.do_oop" on each.
914 915 916 917
  virtual void oop_iterate(OopClosure* cl) {
    oop_iterate(cl, true);
  }
  void oop_iterate(OopClosure* cl, bool do_perm);
918 919

  // Same as above, restricted to a memory region.
920 921 922 923
  virtual void oop_iterate(MemRegion mr, OopClosure* cl) {
    oop_iterate(mr, cl, true);
  }
  void oop_iterate(MemRegion mr, OopClosure* cl, bool do_perm);
924 925

  // Iterate over all objects, calling "cl.do_object" on each.
926 927 928 929 930 931 932
  virtual void object_iterate(ObjectClosure* cl) {
    object_iterate(cl, true);
  }
  virtual void safe_object_iterate(ObjectClosure* cl) {
    object_iterate(cl, true);
  }
  void object_iterate(ObjectClosure* cl, bool do_perm);
933 934 935 936 937 938 939 940 941 942 943 944 945 946 947 948 949 950 951 952 953 954 955 956 957 958 959 960 961 962 963 964 965 966 967 968 969 970 971 972

  // Iterate over all objects allocated since the last collection, calling
  // "cl.do_object" on each.  The heap must have been initialized properly
  // to support this function, or else this call will fail.
  virtual void object_iterate_since_last_GC(ObjectClosure* cl);

  // Iterate over all spaces in use in the heap, in ascending address order.
  virtual void space_iterate(SpaceClosure* cl);

  // Iterate over heap regions, in address order, terminating the
  // iteration early if the "doHeapRegion" method returns "true".
  void heap_region_iterate(HeapRegionClosure* blk);

  // Iterate over heap regions starting with r (or the first region if "r"
  // is NULL), in address order, terminating early if the "doHeapRegion"
  // method returns "true".
  void heap_region_iterate_from(HeapRegion* r, HeapRegionClosure* blk);

  // As above but starting from the region at index idx.
  void heap_region_iterate_from(int idx, HeapRegionClosure* blk);

  HeapRegion* region_at(size_t idx);

  // Divide the heap region sequence into "chunks" of some size (the number
  // of regions divided by the number of parallel threads times some
  // overpartition factor, currently 4).  Assumes that this will be called
  // in parallel by ParallelGCThreads worker threads with discinct worker
  // ids in the range [0..max(ParallelGCThreads-1, 1)], that all parallel
  // calls will use the same "claim_value", and that that claim value is
  // different from the claim_value of any heap region before the start of
  // the iteration.  Applies "blk->doHeapRegion" to each of the regions, by
  // attempting to claim the first region in each chunk, and, if
  // successful, applying the closure to each region in the chunk (and
  // setting the claim value of the second and subsequent regions of the
  // chunk.)  For now requires that "doHeapRegion" always returns "false",
  // i.e., that a closure never attempt to abort a traversal.
  void heap_region_par_iterate_chunked(HeapRegionClosure* blk,
                                       int worker,
                                       jint claim_value);

973 974 975
  // It resets all the region claim values to the default.
  void reset_heap_region_claim_values();

976 977 978 979
#ifdef ASSERT
  bool check_heap_region_claim_values(jint claim_value);
#endif // ASSERT

980 981 982 983 984 985 986 987 988 989 990 991 992 993 994 995 996 997 998 999 1000 1001 1002 1003 1004 1005 1006 1007 1008 1009 1010 1011 1012 1013 1014 1015 1016 1017 1018 1019 1020 1021 1022 1023 1024 1025 1026 1027 1028 1029 1030 1031 1032 1033 1034 1035 1036 1037 1038
  // Iterate over the regions (if any) in the current collection set.
  void collection_set_iterate(HeapRegionClosure* blk);

  // As above but starting from region r
  void collection_set_iterate_from(HeapRegion* r, HeapRegionClosure *blk);

  // Returns the first (lowest address) compactible space in the heap.
  virtual CompactibleSpace* first_compactible_space();

  // A CollectedHeap will contain some number of spaces.  This finds the
  // space containing a given address, or else returns NULL.
  virtual Space* space_containing(const void* addr) const;

  // A G1CollectedHeap will contain some number of heap regions.  This
  // finds the region containing a given address, or else returns NULL.
  HeapRegion* heap_region_containing(const void* addr) const;

  // Like the above, but requires "addr" to be in the heap (to avoid a
  // null-check), and unlike the above, may return an continuing humongous
  // region.
  HeapRegion* heap_region_containing_raw(const void* addr) const;

  // A CollectedHeap is divided into a dense sequence of "blocks"; that is,
  // each address in the (reserved) heap is a member of exactly
  // one block.  The defining characteristic of a block is that it is
  // possible to find its size, and thus to progress forward to the next
  // block.  (Blocks may be of different sizes.)  Thus, blocks may
  // represent Java objects, or they might be free blocks in a
  // free-list-based heap (or subheap), as long as the two kinds are
  // distinguishable and the size of each is determinable.

  // Returns the address of the start of the "block" that contains the
  // address "addr".  We say "blocks" instead of "object" since some heaps
  // may not pack objects densely; a chunk may either be an object or a
  // non-object.
  virtual HeapWord* block_start(const void* addr) const;

  // Requires "addr" to be the start of a chunk, and returns its size.
  // "addr + size" is required to be the start of a new chunk, or the end
  // of the active area of the heap.
  virtual size_t block_size(const HeapWord* addr) const;

  // Requires "addr" to be the start of a block, and returns "TRUE" iff
  // the block is an object.
  virtual bool block_is_obj(const HeapWord* addr) const;

  // Does this heap support heap inspection? (+PrintClassHistogram)
  virtual bool supports_heap_inspection() const { return true; }

  // Section on thread-local allocation buffers (TLABs)
  // See CollectedHeap for semantics.

  virtual bool supports_tlab_allocation() const;
  virtual size_t tlab_capacity(Thread* thr) const;
  virtual size_t unsafe_max_tlab_alloc(Thread* thr) const;
  virtual HeapWord* allocate_new_tlab(size_t size);

  // Can a compiler initialize a new object without store barriers?
  // This permission only extends from the creation of a new object
1039 1040 1041 1042 1043 1044 1045 1046 1047 1048 1049
  // via a TLAB up to the first subsequent safepoint. If such permission
  // is granted for this heap type, the compiler promises to call
  // defer_store_barrier() below on any slow path allocation of
  // a new object for which such initializing store barriers will
  // have been elided. G1, like CMS, allows this, but should be
  // ready to provide a compensating write barrier as necessary
  // if that storage came out of a non-young region. The efficiency
  // of this implementation depends crucially on being able to
  // answer very efficiently in constant time whether a piece of
  // storage in the heap comes from a young region or not.
  // See ReduceInitialCardMarks.
1050
  virtual bool can_elide_tlab_store_barriers() const {
1051 1052 1053 1054 1055 1056
    // 6920090: Temporarily disabled, because of lingering
    // instabilities related to RICM with G1. In the
    // interim, the option ReduceInitialCardMarksForG1
    // below is left solely as a debugging device at least
    // until 6920109 fixes the instabilities.
    return ReduceInitialCardMarksForG1;
1057 1058
  }

1059 1060 1061 1062
  virtual bool card_mark_must_follow_store() const {
    return true;
  }

1063 1064 1065 1066 1067 1068 1069 1070 1071 1072 1073 1074 1075 1076 1077
  bool is_in_young(oop obj) {
    HeapRegion* hr = heap_region_containing(obj);
    return hr != NULL && hr->is_young();
  }

  // We don't need barriers for initializing stores to objects
  // in the young gen: for the SATB pre-barrier, there is no
  // pre-value that needs to be remembered; for the remembered-set
  // update logging post-barrier, we don't maintain remembered set
  // information for young gen objects. Note that non-generational
  // G1 does not have any "young" objects, should not elide
  // the rs logging barrier and so should always answer false below.
  // However, non-generational G1 (-XX:-G1Gen) appears to have
  // bit-rotted so was not tested below.
  virtual bool can_elide_initializing_store_barrier(oop new_obj) {
1078 1079
    // Re 6920090, 6920109 above.
    assert(ReduceInitialCardMarksForG1, "Else cannot be here");
1080 1081 1082
    assert(G1Gen || !is_in_young(new_obj),
           "Non-generational G1 should never return true below");
    return is_in_young(new_obj);
1083 1084 1085 1086 1087 1088 1089 1090 1091 1092 1093 1094 1095 1096 1097 1098 1099 1100 1101
  }

  // Can a compiler elide a store barrier when it writes
  // a permanent oop into the heap?  Applies when the compiler
  // is storing x to the heap, where x->is_perm() is true.
  virtual bool can_elide_permanent_oop_store_barriers() const {
    // At least until perm gen collection is also G1-ified, at
    // which point this should return false.
    return true;
  }

  virtual bool allocs_are_zero_filled();

  // The boundary between a "large" and "small" array of primitives, in
  // words.
  virtual size_t large_typearray_limit();

  // Returns "true" iff the given word_size is "very large".
  static bool isHumongous(size_t word_size) {
1102 1103 1104 1105 1106 1107
    // Note this has to be strictly greater-than as the TLABs
    // are capped at the humongous thresold and we want to
    // ensure that we don't try to allocate a TLAB as
    // humongous and that we don't allocate a humongous
    // object in a TLAB.
    return word_size > _humongous_object_threshold_in_words;
1108 1109 1110 1111 1112 1113 1114 1115 1116 1117 1118 1119 1120 1121 1122 1123 1124 1125 1126 1127 1128 1129 1130 1131 1132 1133 1134 1135 1136 1137
  }

  // Update mod union table with the set of dirty cards.
  void updateModUnion();

  // Set the mod union bits corresponding to the given memRegion.  Note
  // that this is always a safe operation, since it doesn't clear any
  // bits.
  void markModUnionRange(MemRegion mr);

  // Records the fact that a marking phase is no longer in progress.
  void set_marking_complete() {
    _mark_in_progress = false;
  }
  void set_marking_started() {
    _mark_in_progress = true;
  }
  bool mark_in_progress() {
    return _mark_in_progress;
  }

  // Print the maximum heap capacity.
  virtual size_t max_capacity() const;

  virtual jlong millis_since_last_gc();

  // Perform any cleanup actions necessary before allowing a verification.
  virtual void prepare_for_verify();

  // Perform verification.
1138 1139 1140 1141 1142 1143 1144 1145 1146 1147 1148

  // use_prev_marking == true  -> use "prev" marking information,
  // use_prev_marking == false -> use "next" marking information
  // NOTE: Only the "prev" marking information is guaranteed to be
  // consistent most of the time, so most calls to this should use
  // use_prev_marking == true. Currently, there is only one case where
  // this is called with use_prev_marking == false, which is to verify
  // the "next" marking information at the end of remark.
  void verify(bool allow_dirty, bool silent, bool use_prev_marking);

  // Override; it uses the "prev" marking information
1149
  virtual void verify(bool allow_dirty, bool silent);
1150
  // Default behavior by calling print(tty);
1151
  virtual void print() const;
1152
  // This calls print_on(st, PrintHeapAtGCExtended).
1153
  virtual void print_on(outputStream* st) const;
1154 1155 1156 1157
  // If extended is true, it will print out information for all
  // regions in the heap by calling print_on_extended(st).
  virtual void print_on(outputStream* st, bool extended) const;
  virtual void print_on_extended(outputStream* st) const;
1158 1159 1160 1161 1162 1163 1164 1165 1166 1167 1168 1169 1170 1171 1172 1173 1174 1175 1176 1177 1178 1179

  virtual void print_gc_threads_on(outputStream* st) const;
  virtual void gc_threads_do(ThreadClosure* tc) const;

  // Override
  void print_tracing_info() const;

  // If "addr" is a pointer into the (reserved?) heap, returns a positive
  // number indicating the "arena" within the heap in which "addr" falls.
  // Or else returns 0.
  virtual int addr_to_arena_id(void* addr) const;

  // Convenience function to be used in situations where the heap type can be
  // asserted to be this type.
  static G1CollectedHeap* heap();

  void empty_young_list();
  bool should_set_young_locked();

  void set_region_short_lived_locked(HeapRegion* hr);
  // add appropriate methods for any other surv rate groups

1180
  YoungList* young_list() { return _young_list; }
1181 1182 1183 1184 1185

  // debugging
  bool check_young_list_well_formed() {
    return _young_list->check_list_well_formed();
  }
1186 1187

  bool check_young_list_empty(bool check_heap,
1188 1189 1190 1191 1192 1193 1194 1195 1196 1197 1198 1199 1200 1201 1202 1203 1204 1205 1206 1207 1208
                              bool check_sample = true);

  // *** Stuff related to concurrent marking.  It's not clear to me that so
  // many of these need to be public.

  // The functions below are helper functions that a subclass of
  // "CollectedHeap" can use in the implementation of its virtual
  // functions.
  // This performs a concurrent marking of the live objects in a
  // bitmap off to the side.
  void doConcurrentMark();

  // This is called from the marksweep collector which then does
  // a concurrent mark and verifies that the results agree with
  // the stop the world marking.
  void checkConcurrentMark();
  void do_sync_mark();

  bool isMarkedPrev(oop obj) const;
  bool isMarkedNext(oop obj) const;

1209 1210 1211 1212 1213 1214 1215 1216 1217 1218 1219 1220
  // use_prev_marking == true  -> use "prev" marking information,
  // use_prev_marking == false -> use "next" marking information
  bool is_obj_dead_cond(const oop obj,
                        const HeapRegion* hr,
                        const bool use_prev_marking) const {
    if (use_prev_marking) {
      return is_obj_dead(obj, hr);
    } else {
      return is_obj_ill(obj, hr);
    }
  }

1221 1222 1223 1224 1225 1226 1227 1228 1229 1230 1231 1232 1233 1234 1235 1236 1237 1238 1239 1240 1241 1242 1243 1244 1245 1246 1247 1248 1249 1250 1251 1252 1253 1254 1255 1256 1257
  // Determine if an object is dead, given the object and also
  // the region to which the object belongs. An object is dead
  // iff a) it was not allocated since the last mark and b) it
  // is not marked.

  bool is_obj_dead(const oop obj, const HeapRegion* hr) const {
    return
      !hr->obj_allocated_since_prev_marking(obj) &&
      !isMarkedPrev(obj);
  }

  // This is used when copying an object to survivor space.
  // If the object is marked live, then we mark the copy live.
  // If the object is allocated since the start of this mark
  // cycle, then we mark the copy live.
  // If the object has been around since the previous mark
  // phase, and hasn't been marked yet during this phase,
  // then we don't mark it, we just wait for the
  // current marking cycle to get to it.

  // This function returns true when an object has been
  // around since the previous marking and hasn't yet
  // been marked during this marking.

  bool is_obj_ill(const oop obj, const HeapRegion* hr) const {
    return
      !hr->obj_allocated_since_next_marking(obj) &&
      !isMarkedNext(obj);
  }

  // Determine if an object is dead, given only the object itself.
  // This will find the region to which the object belongs and
  // then call the region version of the same function.

  // Added if it is in permanent gen it isn't dead.
  // Added if it is NULL it isn't dead.

1258 1259 1260 1261 1262 1263 1264 1265 1266 1267 1268 1269 1270
  // use_prev_marking == true  -> use "prev" marking information,
  // use_prev_marking == false -> use "next" marking information
  bool is_obj_dead_cond(const oop obj,
                        const bool use_prev_marking) {
    if (use_prev_marking) {
      return is_obj_dead(obj);
    } else {
      return is_obj_ill(obj);
    }
  }

  bool is_obj_dead(const oop obj) {
    const HeapRegion* hr = heap_region_containing(obj);
1271 1272 1273 1274 1275 1276 1277 1278 1279
    if (hr == NULL) {
      if (Universe::heap()->is_in_permanent(obj))
        return false;
      else if (obj == NULL) return false;
      else return true;
    }
    else return is_obj_dead(obj, hr);
  }

1280 1281
  bool is_obj_ill(const oop obj) {
    const HeapRegion* hr = heap_region_containing(obj);
1282 1283 1284 1285 1286 1287 1288 1289 1290 1291 1292 1293 1294 1295 1296 1297 1298 1299 1300 1301
    if (hr == NULL) {
      if (Universe::heap()->is_in_permanent(obj))
        return false;
      else if (obj == NULL) return false;
      else return true;
    }
    else return is_obj_ill(obj, hr);
  }

  // The following is just to alert the verification code
  // that a full collection has occurred and that the
  // remembered sets are no longer up to date.
  bool _full_collection;
  void set_full_collection() { _full_collection = true;}
  void clear_full_collection() {_full_collection = false;}
  bool full_collection() {return _full_collection;}

  ConcurrentMark* concurrent_mark() const { return _cm; }
  ConcurrentG1Refine* concurrent_g1_refine() const { return _cg1r; }

1302 1303 1304 1305 1306 1307 1308 1309 1310 1311
  // The dirty cards region list is used to record a subset of regions
  // whose cards need clearing. The list if populated during the
  // remembered set scanning and drained during the card table
  // cleanup. Although the methods are reentrant, population/draining
  // phases must not overlap. For synchronization purposes the last
  // element on the list points to itself.
  HeapRegion* _dirty_cards_region_list;
  void push_dirty_cards_region(HeapRegion* hr);
  HeapRegion* pop_dirty_cards_region();

1312 1313 1314 1315 1316 1317 1318 1319 1320 1321 1322 1323 1324 1325 1326 1327 1328 1329 1330 1331 1332 1333 1334 1335 1336 1337
public:
  void stop_conc_gc_threads();

  // <NEW PREDICTION>

  double predict_region_elapsed_time_ms(HeapRegion* hr, bool young);
  void check_if_region_is_too_expensive(double predicted_time_ms);
  size_t pending_card_num();
  size_t max_pending_card_num();
  size_t cards_scanned();

  // </NEW PREDICTION>

protected:
  size_t _max_heap_capacity;

//  debug_only(static void check_for_valid_allocation_state();)

public:
  // Temporary: call to mark things unimplemented for the G1 heap (e.g.,
  // MemoryService).  In productization, we can make this assert false
  // to catch such places (as well as searching for calls to this...)
  static void g1_unimplemented();

};

1338 1339 1340 1341 1342 1343 1344 1345 1346 1347 1348 1349 1350 1351 1352 1353 1354 1355 1356 1357 1358 1359 1360 1361 1362 1363 1364 1365 1366 1367 1368 1369 1370 1371 1372 1373 1374 1375 1376 1377 1378 1379 1380 1381 1382 1383 1384 1385 1386 1387 1388 1389 1390 1391
#define use_local_bitmaps         1
#define verify_local_bitmaps      0
#define oop_buffer_length       256

#ifndef PRODUCT
class GCLabBitMap;
class GCLabBitMapClosure: public BitMapClosure {
private:
  ConcurrentMark* _cm;
  GCLabBitMap*    _bitmap;

public:
  GCLabBitMapClosure(ConcurrentMark* cm,
                     GCLabBitMap* bitmap) {
    _cm     = cm;
    _bitmap = bitmap;
  }

  virtual bool do_bit(size_t offset);
};
#endif // !PRODUCT

class GCLabBitMap: public BitMap {
private:
  ConcurrentMark* _cm;

  int       _shifter;
  size_t    _bitmap_word_covers_words;

  // beginning of the heap
  HeapWord* _heap_start;

  // this is the actual start of the GCLab
  HeapWord* _real_start_word;

  // this is the actual end of the GCLab
  HeapWord* _real_end_word;

  // this is the first word, possibly located before the actual start
  // of the GCLab, that corresponds to the first bit of the bitmap
  HeapWord* _start_word;

  // size of a GCLab in words
  size_t _gclab_word_size;

  static int shifter() {
    return MinObjAlignment - 1;
  }

  // how many heap words does a single bitmap word corresponds to?
  static size_t bitmap_word_covers_words() {
    return BitsPerWord << shifter();
  }

1392 1393 1394 1395 1396 1397 1398 1399
  size_t gclab_word_size() const {
    return _gclab_word_size;
  }

  // Calculates actual GCLab size in words
  size_t gclab_real_word_size() const {
    return bitmap_size_in_bits(pointer_delta(_real_end_word, _start_word))
           / BitsPerWord;
1400 1401
  }

1402 1403
  static size_t bitmap_size_in_bits(size_t gclab_word_size) {
    size_t bits_in_bitmap = gclab_word_size >> shifter();
1404 1405 1406 1407 1408 1409 1410 1411 1412
    // We are going to ensure that the beginning of a word in this
    // bitmap also corresponds to the beginning of a word in the
    // global marking bitmap. To handle the case where a GCLab
    // starts from the middle of the bitmap, we need to add enough
    // space (i.e. up to a bitmap word) to ensure that we have
    // enough bits in the bitmap.
    return bits_in_bitmap + BitsPerWord - 1;
  }
public:
1413 1414
  GCLabBitMap(HeapWord* heap_start, size_t gclab_word_size)
    : BitMap(bitmap_size_in_bits(gclab_word_size)),
1415 1416 1417 1418
      _cm(G1CollectedHeap::heap()->concurrent_mark()),
      _shifter(shifter()),
      _bitmap_word_covers_words(bitmap_word_covers_words()),
      _heap_start(heap_start),
1419
      _gclab_word_size(gclab_word_size),
1420 1421 1422 1423 1424 1425 1426 1427 1428 1429 1430 1431 1432 1433 1434 1435 1436 1437 1438 1439 1440 1441 1442 1443 1444 1445 1446 1447 1448 1449 1450 1451 1452 1453 1454 1455 1456 1457 1458 1459 1460 1461 1462 1463 1464 1465 1466 1467 1468 1469 1470 1471 1472 1473 1474 1475 1476 1477 1478 1479 1480 1481 1482 1483 1484 1485 1486 1487 1488 1489 1490 1491 1492 1493 1494 1495 1496 1497 1498 1499 1500 1501 1502 1503 1504 1505 1506 1507 1508 1509 1510 1511 1512 1513
      _real_start_word(NULL),
      _real_end_word(NULL),
      _start_word(NULL)
  {
    guarantee( size_in_words() >= bitmap_size_in_words(),
               "just making sure");
  }

  inline unsigned heapWordToOffset(HeapWord* addr) {
    unsigned offset = (unsigned) pointer_delta(addr, _start_word) >> _shifter;
    assert(offset < size(), "offset should be within bounds");
    return offset;
  }

  inline HeapWord* offsetToHeapWord(size_t offset) {
    HeapWord* addr =  _start_word + (offset << _shifter);
    assert(_real_start_word <= addr && addr < _real_end_word, "invariant");
    return addr;
  }

  bool fields_well_formed() {
    bool ret1 = (_real_start_word == NULL) &&
                (_real_end_word == NULL) &&
                (_start_word == NULL);
    if (ret1)
      return true;

    bool ret2 = _real_start_word >= _start_word &&
      _start_word < _real_end_word &&
      (_real_start_word + _gclab_word_size) == _real_end_word &&
      (_start_word + _gclab_word_size + _bitmap_word_covers_words)
                                                              > _real_end_word;
    return ret2;
  }

  inline bool mark(HeapWord* addr) {
    guarantee(use_local_bitmaps, "invariant");
    assert(fields_well_formed(), "invariant");

    if (addr >= _real_start_word && addr < _real_end_word) {
      assert(!isMarked(addr), "should not have already been marked");

      // first mark it on the bitmap
      at_put(heapWordToOffset(addr), true);

      return true;
    } else {
      return false;
    }
  }

  inline bool isMarked(HeapWord* addr) {
    guarantee(use_local_bitmaps, "invariant");
    assert(fields_well_formed(), "invariant");

    return at(heapWordToOffset(addr));
  }

  void set_buffer(HeapWord* start) {
    guarantee(use_local_bitmaps, "invariant");
    clear();

    assert(start != NULL, "invariant");
    _real_start_word = start;
    _real_end_word   = start + _gclab_word_size;

    size_t diff =
      pointer_delta(start, _heap_start) % _bitmap_word_covers_words;
    _start_word = start - diff;

    assert(fields_well_formed(), "invariant");
  }

#ifndef PRODUCT
  void verify() {
    // verify that the marks have been propagated
    GCLabBitMapClosure cl(_cm, this);
    iterate(&cl);
  }
#endif // PRODUCT

  void retire() {
    guarantee(use_local_bitmaps, "invariant");
    assert(fields_well_formed(), "invariant");

    if (_start_word != NULL) {
      CMBitMap*       mark_bitmap = _cm->nextMarkBitMap();

      // this means that the bitmap was set up for the GCLab
      assert(_real_start_word != NULL && _real_end_word != NULL, "invariant");

      mark_bitmap->mostly_disjoint_range_union(this,
                                0, // always start from the start of the bitmap
                                _start_word,
1514
                                gclab_real_word_size());
1515 1516 1517 1518 1519 1520 1521 1522 1523 1524 1525
      _cm->grayRegionIfNecessary(MemRegion(_real_start_word, _real_end_word));

#ifndef PRODUCT
      if (use_local_bitmaps && verify_local_bitmaps)
        verify();
#endif // PRODUCT
    } else {
      assert(_real_start_word == NULL && _real_end_word == NULL, "invariant");
    }
  }

1526 1527
  size_t bitmap_size_in_words() const {
    return (bitmap_size_in_bits(gclab_word_size()) + BitsPerWord - 1) / BitsPerWord;
1528
  }
1529

1530 1531 1532 1533 1534 1535 1536 1537 1538
};

class G1ParGCAllocBuffer: public ParGCAllocBuffer {
private:
  bool        _retired;
  bool        _during_marking;
  GCLabBitMap _bitmap;

public:
1539 1540
  G1ParGCAllocBuffer(size_t gclab_word_size) :
    ParGCAllocBuffer(gclab_word_size),
1541
    _during_marking(G1CollectedHeap::heap()->mark_in_progress()),
1542
    _bitmap(G1CollectedHeap::heap()->reserved_region().start(), gclab_word_size),
1543 1544 1545 1546 1547 1548 1549 1550 1551 1552 1553 1554 1555 1556 1557 1558 1559 1560 1561 1562 1563 1564 1565 1566 1567 1568 1569 1570 1571 1572 1573 1574 1575 1576 1577
    _retired(false)
  { }

  inline bool mark(HeapWord* addr) {
    guarantee(use_local_bitmaps, "invariant");
    assert(_during_marking, "invariant");
    return _bitmap.mark(addr);
  }

  inline void set_buf(HeapWord* buf) {
    if (use_local_bitmaps && _during_marking)
      _bitmap.set_buffer(buf);
    ParGCAllocBuffer::set_buf(buf);
    _retired = false;
  }

  inline void retire(bool end_of_gc, bool retain) {
    if (_retired)
      return;
    if (use_local_bitmaps && _during_marking) {
      _bitmap.retire();
    }
    ParGCAllocBuffer::retire(end_of_gc, retain);
    _retired = true;
  }
};

class G1ParScanThreadState : public StackObj {
protected:
  G1CollectedHeap* _g1h;
  RefToScanQueue*  _refs;
  DirtyCardQueue   _dcq;
  CardTableModRefBS* _ct_bs;
  G1RemSet* _g1_rem;

1578 1579 1580 1581
  G1ParGCAllocBuffer  _surviving_alloc_buffer;
  G1ParGCAllocBuffer  _tenured_alloc_buffer;
  G1ParGCAllocBuffer* _alloc_buffers[GCAllocPurposeCount];
  ageTable            _age_table;
1582 1583 1584 1585 1586 1587 1588 1589 1590 1591 1592

  size_t           _alloc_buffer_waste;
  size_t           _undo_waste;

  OopsInHeapRegionClosure*      _evac_failure_cl;
  G1ParScanHeapEvacClosure*     _evac_cl;
  G1ParScanPartialArrayClosure* _partial_scan_cl;

  int _hash_seed;
  int _queue_num;

1593
  size_t _term_attempts;
1594 1595 1596 1597 1598 1599 1600 1601 1602 1603 1604 1605 1606

  double _start;
  double _start_strong_roots;
  double _strong_roots_time;
  double _start_term;
  double _term_time;

  // Map from young-age-index (0 == not young, 1 is youngest) to
  // surviving words. base is what we get back from the malloc call
  size_t* _surviving_young_words_base;
  // this points into the array, as we use the first few entries for padding
  size_t* _surviving_young_words;

1607
#define PADDING_ELEM_NUM (DEFAULT_CACHE_LINE_SIZE / sizeof(size_t))
1608 1609 1610 1611 1612 1613 1614 1615 1616 1617 1618 1619 1620 1621 1622 1623 1624 1625 1626 1627 1628 1629 1630 1631 1632 1633 1634 1635 1636 1637 1638 1639 1640 1641 1642 1643 1644

  void   add_to_alloc_buffer_waste(size_t waste) { _alloc_buffer_waste += waste; }

  void   add_to_undo_waste(size_t waste)         { _undo_waste += waste; }

  DirtyCardQueue& dirty_card_queue()             { return _dcq;  }
  CardTableModRefBS* ctbs()                      { return _ct_bs; }

  template <class T> void immediate_rs_update(HeapRegion* from, T* p, int tid) {
    if (!from->is_survivor()) {
      _g1_rem->par_write_ref(from, p, tid);
    }
  }

  template <class T> void deferred_rs_update(HeapRegion* from, T* p, int tid) {
    // If the new value of the field points to the same region or
    // is the to-space, we don't need to include it in the Rset updates.
    if (!from->is_in_reserved(oopDesc::load_decode_heap_oop(p)) && !from->is_survivor()) {
      size_t card_index = ctbs()->index_for(p);
      // If the card hasn't been added to the buffer, do it.
      if (ctbs()->mark_card_deferred(card_index)) {
        dirty_card_queue().enqueue((jbyte*)ctbs()->byte_for_index(card_index));
      }
    }
  }

public:
  G1ParScanThreadState(G1CollectedHeap* g1h, int queue_num);

  ~G1ParScanThreadState() {
    FREE_C_HEAP_ARRAY(size_t, _surviving_young_words_base);
  }

  RefToScanQueue*   refs()            { return _refs;             }
  ageTable*         age_table()       { return &_age_table;       }

  G1ParGCAllocBuffer* alloc_buffer(GCAllocPurpose purpose) {
1645
    return _alloc_buffers[purpose];
1646 1647
  }

1648 1649
  size_t alloc_buffer_waste() const              { return _alloc_buffer_waste; }
  size_t undo_waste() const                      { return _undo_waste; }
1650 1651 1652 1653

  template <class T> void push_on_queue(T* ref) {
    assert(ref != NULL, "invariant");
    assert(has_partial_array_mask(ref) ||
1654
           _g1h->is_in_g1_reserved(oopDesc::load_decode_heap_oop(ref)), "invariant");
1655 1656 1657 1658 1659 1660 1661
#ifdef ASSERT
    if (has_partial_array_mask(ref)) {
      oop p = clear_partial_array_mask(ref);
      // Verify that we point into the CS
      assert(_g1h->obj_in_cs(p), "Should be in CS");
    }
#endif
1662
    refs()->push(ref);
1663 1664 1665 1666 1667 1668 1669
  }

  void pop_from_queue(StarTask& ref) {
    if (refs()->pop_local(ref)) {
      assert((oop*)ref != NULL, "pop_local() returned true");
      assert(UseCompressedOops || !ref.is_narrow(), "Error");
      assert(has_partial_array_mask((oop*)ref) ||
1670 1671 1672
             _g1h->is_in_g1_reserved(ref.is_narrow() ? oopDesc::load_decode_heap_oop((narrowOop*)ref)
                                                     : oopDesc::load_decode_heap_oop((oop*)ref)),
              "invariant");
1673 1674 1675 1676 1677 1678 1679
    } else {
      StarTask null_task;
      ref = null_task;
    }
  }

  void pop_from_overflow_queue(StarTask& ref) {
1680 1681
    StarTask new_ref;
    refs()->pop_overflow(new_ref);
1682 1683 1684
    assert((oop*)new_ref != NULL, "pop() from a local non-empty stack");
    assert(UseCompressedOops || !new_ref.is_narrow(), "Error");
    assert(has_partial_array_mask((oop*)new_ref) ||
1685 1686 1687
           _g1h->is_in_g1_reserved(new_ref.is_narrow() ? oopDesc::load_decode_heap_oop((narrowOop*)new_ref)
                                                       : oopDesc::load_decode_heap_oop((oop*)new_ref)),
           "invariant");
1688 1689 1690
    ref = new_ref;
  }

1691 1692
  int refs_to_scan()            { return refs()->size(); }
  int overflowed_refs_to_scan() { return refs()->overflow_stack()->length(); }
1693 1694 1695 1696 1697 1698 1699 1700 1701 1702 1703 1704

  template <class T> void update_rs(HeapRegion* from, T* p, int tid) {
    if (G1DeferredRSUpdate) {
      deferred_rs_update(from, p, tid);
    } else {
      immediate_rs_update(from, p, tid);
    }
  }

  HeapWord* allocate_slow(GCAllocPurpose purpose, size_t word_sz) {

    HeapWord* obj = NULL;
1705 1706
    size_t gclab_word_size = _g1h->desired_plab_sz(purpose);
    if (word_sz * 100 < gclab_word_size * ParallelGCBufferWastePct) {
1707
      G1ParGCAllocBuffer* alloc_buf = alloc_buffer(purpose);
1708 1709
      assert(gclab_word_size == alloc_buf->word_sz(),
             "dynamic resizing is not supported");
1710 1711 1712
      add_to_alloc_buffer_waste(alloc_buf->words_remaining());
      alloc_buf->retire(false, false);

1713
      HeapWord* buf = _g1h->par_allocate_during_gc(purpose, gclab_word_size);
1714 1715 1716 1717 1718 1719 1720 1721 1722 1723 1724 1725 1726 1727 1728 1729 1730 1731 1732 1733 1734 1735 1736 1737 1738 1739 1740 1741 1742 1743 1744 1745 1746 1747 1748 1749 1750 1751 1752 1753 1754 1755 1756 1757 1758 1759 1760
      if (buf == NULL) return NULL; // Let caller handle allocation failure.
      // Otherwise.
      alloc_buf->set_buf(buf);

      obj = alloc_buf->allocate(word_sz);
      assert(obj != NULL, "buffer was definitely big enough...");
    } else {
      obj = _g1h->par_allocate_during_gc(purpose, word_sz);
    }
    return obj;
  }

  HeapWord* allocate(GCAllocPurpose purpose, size_t word_sz) {
    HeapWord* obj = alloc_buffer(purpose)->allocate(word_sz);
    if (obj != NULL) return obj;
    return allocate_slow(purpose, word_sz);
  }

  void undo_allocation(GCAllocPurpose purpose, HeapWord* obj, size_t word_sz) {
    if (alloc_buffer(purpose)->contains(obj)) {
      assert(alloc_buffer(purpose)->contains(obj + word_sz - 1),
             "should contain whole object");
      alloc_buffer(purpose)->undo_allocation(obj, word_sz);
    } else {
      CollectedHeap::fill_with_object(obj, word_sz);
      add_to_undo_waste(word_sz);
    }
  }

  void set_evac_failure_closure(OopsInHeapRegionClosure* evac_failure_cl) {
    _evac_failure_cl = evac_failure_cl;
  }
  OopsInHeapRegionClosure* evac_failure_closure() {
    return _evac_failure_cl;
  }

  void set_evac_closure(G1ParScanHeapEvacClosure* evac_cl) {
    _evac_cl = evac_cl;
  }

  void set_partial_scan_closure(G1ParScanPartialArrayClosure* partial_scan_cl) {
    _partial_scan_cl = partial_scan_cl;
  }

  int* hash_seed() { return &_hash_seed; }
  int  queue_num() { return _queue_num; }

1761
  size_t term_attempts() const  { return _term_attempts; }
1762
  void note_term_attempt() { _term_attempts++; }
1763 1764 1765 1766 1767 1768 1769

  void start_strong_roots() {
    _start_strong_roots = os::elapsedTime();
  }
  void end_strong_roots() {
    _strong_roots_time += (os::elapsedTime() - _start_strong_roots);
  }
1770
  double strong_roots_time() const { return _strong_roots_time; }
1771 1772 1773 1774 1775 1776 1777 1778

  void start_term_time() {
    note_term_attempt();
    _start_term = os::elapsedTime();
  }
  void end_term_time() {
    _term_time += (os::elapsedTime() - _start_term);
  }
1779
  double term_time() const { return _term_time; }
1780

1781
  double elapsed_time() const {
1782 1783 1784
    return os::elapsedTime() - _start;
  }

1785 1786 1787 1788 1789
  static void
    print_termination_stats_hdr(outputStream* const st = gclog_or_tty);
  void
    print_termination_stats(int i, outputStream* const st = gclog_or_tty) const;

1790 1791 1792 1793 1794 1795 1796 1797
  size_t* surviving_young_words() {
    // We add on to hide entry 0 which accumulates surviving words for
    // age -1 regions (i.e. non-young ones)
    return _surviving_young_words;
  }

  void retire_alloc_buffers() {
    for (int ap = 0; ap < GCAllocPurposeCount; ++ap) {
1798
      size_t waste = _alloc_buffers[ap]->words_remaining();
1799
      add_to_alloc_buffer_waste(waste);
1800
      _alloc_buffers[ap]->retire(true, false);
1801 1802 1803 1804 1805 1806 1807 1808 1809 1810 1811 1812 1813 1814 1815 1816 1817 1818 1819 1820 1821 1822 1823 1824 1825 1826 1827 1828 1829 1830 1831 1832 1833 1834 1835 1836 1837 1838 1839 1840 1841
    }
  }

private:
  template <class T> void deal_with_reference(T* ref_to_scan) {
    if (has_partial_array_mask(ref_to_scan)) {
      _partial_scan_cl->do_oop_nv(ref_to_scan);
    } else {
      // Note: we can use "raw" versions of "region_containing" because
      // "obj_to_scan" is definitely in the heap, and is not in a
      // humongous region.
      HeapRegion* r = _g1h->heap_region_containing_raw(ref_to_scan);
      _evac_cl->set_region(r);
      _evac_cl->do_oop_nv(ref_to_scan);
    }
  }

public:
  void trim_queue() {
    // I've replicated the loop twice, first to drain the overflow
    // queue, second to drain the task queue. This is better than
    // having a single loop, which checks both conditions and, inside
    // it, either pops the overflow queue or the task queue, as each
    // loop is tighter. Also, the decision to drain the overflow queue
    // first is not arbitrary, as the overflow queue is not visible
    // to the other workers, whereas the task queue is. So, we want to
    // drain the "invisible" entries first, while allowing the other
    // workers to potentially steal the "visible" entries.

    while (refs_to_scan() > 0 || overflowed_refs_to_scan() > 0) {
      while (overflowed_refs_to_scan() > 0) {
        StarTask ref_to_scan;
        assert((oop*)ref_to_scan == NULL, "Constructed above");
        pop_from_overflow_queue(ref_to_scan);
        // We shouldn't have pushed it on the queue if it was not
        // pointing into the CSet.
        assert((oop*)ref_to_scan != NULL, "Follows from inner loop invariant");
        if (ref_to_scan.is_narrow()) {
          assert(UseCompressedOops, "Error");
          narrowOop* p = (narrowOop*)ref_to_scan;
          assert(!has_partial_array_mask(p) &&
1842
                 _g1h->is_in_g1_reserved(oopDesc::load_decode_heap_oop(p)), "sanity");
1843 1844 1845
          deal_with_reference(p);
        } else {
          oop* p = (oop*)ref_to_scan;
1846 1847
          assert((has_partial_array_mask(p) && _g1h->is_in_g1_reserved(clear_partial_array_mask(p))) ||
                 _g1h->is_in_g1_reserved(oopDesc::load_decode_heap_oop(p)), "sanity");
1848 1849 1850 1851 1852 1853 1854 1855 1856 1857 1858 1859 1860
          deal_with_reference(p);
        }
      }

      while (refs_to_scan() > 0) {
        StarTask ref_to_scan;
        assert((oop*)ref_to_scan == NULL, "Constructed above");
        pop_from_queue(ref_to_scan);
        if ((oop*)ref_to_scan != NULL) {
          if (ref_to_scan.is_narrow()) {
            assert(UseCompressedOops, "Error");
            narrowOop* p = (narrowOop*)ref_to_scan;
            assert(!has_partial_array_mask(p) &&
1861
                    _g1h->is_in_g1_reserved(oopDesc::load_decode_heap_oop(p)), "sanity");
1862 1863 1864 1865
            deal_with_reference(p);
          } else {
            oop* p = (oop*)ref_to_scan;
            assert((has_partial_array_mask(p) && _g1h->obj_in_cs(clear_partial_array_mask(p))) ||
1866
                   _g1h->is_in_g1_reserved(oopDesc::load_decode_heap_oop(p)), "sanity");
1867 1868 1869 1870 1871 1872 1873
            deal_with_reference(p);
          }
        }
      }
    }
  }
};