1 /*
  2  * Copyright (c) 2001, 2018, Oracle and/or its affiliates. All rights reserved.
  3  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
  4  *
  5  * This code is free software; you can redistribute it and/or modify it
  6  * under the terms of the GNU General Public License version 2 only, as
  7  * published by the Free Software Foundation.
  8  *
  9  * This code is distributed in the hope that it will be useful, but WITHOUT
 10  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
 11  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
 12  * version 2 for more details (a copy is included in the LICENSE file that
 13  * accompanied this code).
 14  *
 15  * You should have received a copy of the GNU General Public License version
 16  * 2 along with this work; if not, write to the Free Software Foundation,
 17  * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
 18  *
 19  * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
 20  * or visit www.oracle.com if you need additional information or have any
 21  * questions.
 22  *
 23  */
 24 
 25 #include "precompiled.hpp"
 26 #include "gc/g1/dirtyCardQueue.hpp"
 27 #include "gc/g1/g1BlockOffsetTable.inline.hpp"
 28 #include "gc/g1/g1CardTable.inline.hpp"
 29 #include "gc/g1/g1CollectedHeap.inline.hpp"
 30 #include "gc/g1/g1ConcurrentRefine.hpp"
 31 #include "gc/g1/g1FromCardCache.hpp"
 32 #include "gc/g1/g1GCPhaseTimes.hpp"
 33 #include "gc/g1/g1HotCardCache.hpp"
 34 #include "gc/g1/g1OopClosures.inline.hpp"
 35 #include "gc/g1/g1RemSet.hpp"
 36 #include "gc/g1/heapRegion.inline.hpp"
 37 #include "gc/g1/heapRegionManager.inline.hpp"
 38 #include "gc/g1/heapRegionRemSet.hpp"
 39 #include "gc/shared/gcTraceTime.inline.hpp"
 40 #include "gc/shared/suspendibleThreadSet.hpp"
 41 #include "memory/iterator.hpp"
 42 #include "memory/resourceArea.hpp"
 43 #include "oops/oop.inline.hpp"
 44 #include "utilities/align.hpp"
 45 #include "utilities/globalDefinitions.hpp"
 46 #include "utilities/intHisto.hpp"
 47 #include "utilities/stack.inline.hpp"
 48 
 49 // Collects information about the overall remembered set scan progress during an evacuation.
 50 class G1RemSetScanState : public CHeapObj<mtGC> {
 51 private:
 52   class G1ClearCardTableTask : public AbstractGangTask {
 53     G1CollectedHeap* _g1h;
 54     uint* _dirty_region_list;
 55     size_t _num_dirty_regions;
 56     size_t _chunk_length;
 57 
 58     size_t volatile _cur_dirty_regions;
 59   public:
 60     G1ClearCardTableTask(G1CollectedHeap* g1h,
 61                          uint* dirty_region_list,
 62                          size_t num_dirty_regions,
 63                          size_t chunk_length) :
 64       AbstractGangTask("G1 Clear Card Table Task"),
 65       _g1h(g1h),
 66       _dirty_region_list(dirty_region_list),
 67       _num_dirty_regions(num_dirty_regions),
 68       _chunk_length(chunk_length),
 69       _cur_dirty_regions(0) {
 70 
 71       assert(chunk_length > 0, "must be");
 72     }
 73 
 74     static size_t chunk_size() { return M; }
 75 
 76     void work(uint worker_id) {
 77       G1CardTable* ct = _g1h->card_table();
 78 
 79       while (_cur_dirty_regions < _num_dirty_regions) {
 80         size_t next = Atomic::add(_chunk_length, &_cur_dirty_regions) - _chunk_length;
 81         size_t max = MIN2(next + _chunk_length, _num_dirty_regions);
 82 
 83         for (size_t i = next; i < max; i++) {
 84           HeapRegion* r = _g1h->region_at(_dirty_region_list[i]);
 85           if (!r->is_survivor()) {
 86             ct->clear(MemRegion(r->bottom(), r->end()));
 87           }
 88         }
 89       }
 90     }
 91   };
 92 
 93   size_t _max_regions;
 94 
 95   // Scan progress for the remembered set of a single region. Transitions from
 96   // Unclaimed -> Claimed -> Complete.
 97   // At each of the transitions the thread that does the transition needs to perform
 98   // some special action once. This is the reason for the extra "Claimed" state.
 99   typedef jint G1RemsetIterState;
100 
101   static const G1RemsetIterState Unclaimed = 0; // The remembered set has not been scanned yet.
102   static const G1RemsetIterState Claimed = 1;   // The remembered set is currently being scanned.
103   static const G1RemsetIterState Complete = 2;  // The remembered set has been completely scanned.
104 
105   G1RemsetIterState volatile* _iter_states;
106   // The current location where the next thread should continue scanning in a region's
107   // remembered set.
108   size_t volatile* _iter_claims;
109 
110   // Temporary buffer holding the regions we used to store remembered set scan duplicate
111   // information. These are also called "dirty". Valid entries are from [0.._cur_dirty_region)
112   uint* _dirty_region_buffer;
113 
114   typedef jbyte IsDirtyRegionState;
115   static const IsDirtyRegionState Clean = 0;
116   static const IsDirtyRegionState Dirty = 1;
117   // Holds a flag for every region whether it is in the _dirty_region_buffer already
118   // to avoid duplicates. Uses jbyte since there are no atomic instructions for bools.
119   IsDirtyRegionState* _in_dirty_region_buffer;
120   size_t _cur_dirty_region;
121 
122   // Creates a snapshot of the current _top values at the start of collection to
123   // filter out card marks that we do not want to scan.
124   class G1ResetScanTopClosure : public HeapRegionClosure {
125   private:
126     HeapWord** _scan_top;
127   public:
128     G1ResetScanTopClosure(HeapWord** scan_top) : _scan_top(scan_top) { }
129 
130     virtual bool do_heap_region(HeapRegion* r) {
131       uint hrm_index = r->hrm_index();
132       if (!r->in_collection_set() && r->is_old_or_humongous()) {
133         _scan_top[hrm_index] = r->top();
134       } else {
135         _scan_top[hrm_index] = r->bottom();
136       }
137       return false;
138     }
139   };
140 
141   // For each region, contains the maximum top() value to be used during this garbage
142   // collection. Subsumes common checks like filtering out everything but old and
143   // humongous regions outside the collection set.
144   // This is valid because we are not interested in scanning stray remembered set
145   // entries from free or archive regions.
146   HeapWord** _scan_top;
147 public:
148   G1RemSetScanState() :
149     _max_regions(0),
150     _iter_states(NULL),
151     _iter_claims(NULL),
152     _dirty_region_buffer(NULL),
153     _in_dirty_region_buffer(NULL),
154     _cur_dirty_region(0),
155     _scan_top(NULL) {
156   }
157 
158   ~G1RemSetScanState() {
159     if (_iter_states != NULL) {
160       FREE_C_HEAP_ARRAY(G1RemsetIterState, _iter_states);
161     }
162     if (_iter_claims != NULL) {
163       FREE_C_HEAP_ARRAY(size_t, _iter_claims);
164     }
165     if (_dirty_region_buffer != NULL) {
166       FREE_C_HEAP_ARRAY(uint, _dirty_region_buffer);
167     }
168     if (_in_dirty_region_buffer != NULL) {
169       FREE_C_HEAP_ARRAY(IsDirtyRegionState, _in_dirty_region_buffer);
170     }
171     if (_scan_top != NULL) {
172       FREE_C_HEAP_ARRAY(HeapWord*, _scan_top);
173     }
174   }
175 
176   void initialize(uint max_regions) {
177     assert(_iter_states == NULL, "Must not be initialized twice");
178     assert(_iter_claims == NULL, "Must not be initialized twice");
179     _max_regions = max_regions;
180     _iter_states = NEW_C_HEAP_ARRAY(G1RemsetIterState, max_regions, mtGC);
181     _iter_claims = NEW_C_HEAP_ARRAY(size_t, max_regions, mtGC);
182     _dirty_region_buffer = NEW_C_HEAP_ARRAY(uint, max_regions, mtGC);
183     _in_dirty_region_buffer = NEW_C_HEAP_ARRAY(IsDirtyRegionState, max_regions, mtGC);
184     _scan_top = NEW_C_HEAP_ARRAY(HeapWord*, max_regions, mtGC);
185   }
186 
187   void reset() {
188     for (uint i = 0; i < _max_regions; i++) {
189       _iter_states[i] = Unclaimed;
190     }
191 
192     G1ResetScanTopClosure cl(_scan_top);
193     G1CollectedHeap::heap()->heap_region_iterate(&cl);
194 
195     memset((void*)_iter_claims, 0, _max_regions * sizeof(size_t));
196     memset(_in_dirty_region_buffer, Clean, _max_regions * sizeof(IsDirtyRegionState));
197     _cur_dirty_region = 0;
198   }
199 
200   // Attempt to claim the remembered set of the region for iteration. Returns true
201   // if this call caused the transition from Unclaimed to Claimed.
202   inline bool claim_iter(uint region) {
203     assert(region < _max_regions, "Tried to access invalid region %u", region);
204     if (_iter_states[region] != Unclaimed) {
205       return false;
206     }
207     G1RemsetIterState res = Atomic::cmpxchg(Claimed, &_iter_states[region], Unclaimed);
208     return (res == Unclaimed);
209   }
210 
211   // Try to atomically sets the iteration state to "complete". Returns true for the
212   // thread that caused the transition.
213   inline bool set_iter_complete(uint region) {
214     if (iter_is_complete(region)) {
215       return false;
216     }
217     G1RemsetIterState res = Atomic::cmpxchg(Complete, &_iter_states[region], Claimed);
218     return (res == Claimed);
219   }
220 
221   // Returns true if the region's iteration is complete.
222   inline bool iter_is_complete(uint region) const {
223     assert(region < _max_regions, "Tried to access invalid region %u", region);
224     return _iter_states[region] == Complete;
225   }
226 
227   // The current position within the remembered set of the given region.
228   inline size_t iter_claimed(uint region) const {
229     assert(region < _max_regions, "Tried to access invalid region %u", region);
230     return _iter_claims[region];
231   }
232 
233   // Claim the next block of cards within the remembered set of the region with
234   // step size.
235   inline size_t iter_claimed_next(uint region, size_t step) {
236     return Atomic::add(step, &_iter_claims[region]) - step;
237   }
238 
239   void add_dirty_region(uint region) {
240     if (_in_dirty_region_buffer[region] == Dirty) {
241       return;
242     }
243 
244     bool marked_as_dirty = Atomic::cmpxchg(Dirty, &_in_dirty_region_buffer[region], Clean) == Clean;
245     if (marked_as_dirty) {
246       size_t allocated = Atomic::add(1u, &_cur_dirty_region) - 1;
247       _dirty_region_buffer[allocated] = region;
248     }
249   }
250 
251   HeapWord* scan_top(uint region_idx) const {
252     return _scan_top[region_idx];
253   }
254 
255   // Clear the card table of "dirty" regions.
256   void clear_card_table(WorkGang* workers) {
257     if (_cur_dirty_region == 0) {
258       return;
259     }
260 
261     size_t const num_chunks = align_up(_cur_dirty_region * HeapRegion::CardsPerRegion, G1ClearCardTableTask::chunk_size()) / G1ClearCardTableTask::chunk_size();
262     uint const num_workers = (uint)MIN2(num_chunks, (size_t)workers->active_workers());
263     size_t const chunk_length = G1ClearCardTableTask::chunk_size() / HeapRegion::CardsPerRegion;
264 
265     // Iterate over the dirty cards region list.
266     G1ClearCardTableTask cl(G1CollectedHeap::heap(), _dirty_region_buffer, _cur_dirty_region, chunk_length);
267 
268     log_debug(gc, ergo)("Running %s using %u workers for " SIZE_FORMAT " "
269                         "units of work for " SIZE_FORMAT " regions.",
270                         cl.name(), num_workers, num_chunks, _cur_dirty_region);
271     workers->run_task(&cl, num_workers);
272 
273 #ifndef PRODUCT
274     // Need to synchronize with concurrent cleanup since it needs to
275     // finish its card table clearing before we can verify.
276     G1CollectedHeap::heap()->wait_while_free_regions_coming();
277     G1CollectedHeap::heap()->verifier()->verify_card_table_cleanup();
278 #endif
279   }
280 };
281 
282 G1RemSet::G1RemSet(G1CollectedHeap* g1,
283                    G1CardTable* ct,
284                    G1HotCardCache* hot_card_cache) :
285   _g1(g1),
286   _scan_state(new G1RemSetScanState()),
287   _num_conc_refined_cards(0),
288   _ct(ct),
289   _g1p(_g1->g1_policy()),
290   _hot_card_cache(hot_card_cache),
291   _prev_period_summary() {
292 }
293 
294 G1RemSet::~G1RemSet() {
295   if (_scan_state != NULL) {
296     delete _scan_state;
297   }
298 }
299 
300 uint G1RemSet::num_par_rem_sets() {
301   return MAX2(DirtyCardQueueSet::num_par_ids() + G1ConcurrentRefine::max_num_threads(), ParallelGCThreads);
302 }
303 
304 void G1RemSet::initialize(size_t capacity, uint max_regions) {
305   G1FromCardCache::initialize(num_par_rem_sets(), max_regions);
306   _scan_state->initialize(max_regions);
307   {
308     GCTraceTime(Debug, gc, marking)("Initialize Card Live Data");
309     _card_live_data.initialize(capacity, max_regions);
310   }
311   if (G1PretouchAuxiliaryMemory) {
312     GCTraceTime(Debug, gc, marking)("Pre-Touch Card Live Data");
313     _card_live_data.pretouch();
314   }
315 }
316 
317 G1ScanRSForRegionClosure::G1ScanRSForRegionClosure(G1RemSetScanState* scan_state,
318                                                    G1ScanObjsDuringScanRSClosure* scan_obj_on_card,
319                                                    CodeBlobClosure* code_root_cl,
320                                                    uint worker_i) :
321   _scan_state(scan_state),
322   _scan_objs_on_card_cl(scan_obj_on_card),
323   _code_root_cl(code_root_cl),
324   _strong_code_root_scan_time_sec(0.0),
325   _cards_claimed(0),
326   _cards_scanned(0),
327   _cards_skipped(0),
328   _worker_i(worker_i) {
329   _g1h = G1CollectedHeap::heap();
330   _bot = _g1h->bot();
331   _ct = _g1h->card_table();
332 }
333 
334 void G1ScanRSForRegionClosure::scan_card(MemRegion mr, uint region_idx_for_card) {
335   HeapRegion* const card_region = _g1h->region_at(region_idx_for_card);
336   _scan_objs_on_card_cl->set_region(card_region);
337   card_region->oops_on_card_seq_iterate_careful<true>(mr, _scan_objs_on_card_cl);
338   _cards_scanned++;
339 }
340 
341 void G1ScanRSForRegionClosure::scan_strong_code_roots(HeapRegion* r) {
342   double scan_start = os::elapsedTime();
343   r->strong_code_roots_do(_code_root_cl);
344   _strong_code_root_scan_time_sec += (os::elapsedTime() - scan_start);
345 }
346 
347 void G1ScanRSForRegionClosure::claim_card(size_t card_index, const uint region_idx_for_card){
348   _ct->set_card_claimed(card_index);
349   _scan_state->add_dirty_region(region_idx_for_card);
350 }
351 
352 bool G1ScanRSForRegionClosure::do_heap_region(HeapRegion* r) {
353   assert(r->in_collection_set(), "should only be called on elements of CS.");
354   uint region_idx = r->hrm_index();
355 
356   if (_scan_state->iter_is_complete(region_idx)) {
357     return false;
358   }
359   if (_scan_state->claim_iter(region_idx)) {
360     // If we ever free the collection set concurrently, we should also
361     // clear the card table concurrently therefore we won't need to
362     // add regions of the collection set to the dirty cards region.
363     _scan_state->add_dirty_region(region_idx);
364   }
365 
366   // We claim cards in blocks so as to reduce the contention.
367   size_t const block_size = G1RSetScanBlockSize;
368 
369   HeapRegionRemSetIterator iter(r->rem_set());
370   size_t card_index;
371 
372   size_t claimed_card_block = _scan_state->iter_claimed_next(region_idx, block_size);
373   for (size_t current_card = 0; iter.has_next(card_index); current_card++) {
374     if (current_card >= claimed_card_block + block_size) {
375       claimed_card_block = _scan_state->iter_claimed_next(region_idx, block_size);
376     }
377     if (current_card < claimed_card_block) {
378       _cards_skipped++;
379       continue;
380     }
381     _cards_claimed++;
382 
383     // If the card is dirty, then G1 will scan it during Update RS.
384     if (_ct->is_card_claimed(card_index) || _ct->is_card_dirty(card_index)) {
385       continue;
386     }
387 
388     HeapWord* const card_start = _g1h->bot()->address_for_index(card_index);
389     uint const region_idx_for_card = _g1h->addr_to_region(card_start);
390 
391     assert(_g1h->region_at(region_idx_for_card)->is_in_reserved(card_start),
392            "Card start " PTR_FORMAT " to scan outside of region %u", p2i(card_start), _g1h->region_at(region_idx_for_card)->hrm_index());
393     HeapWord* const top = _scan_state->scan_top(region_idx_for_card);
394     if (card_start >= top) {
395       continue;
396     }
397 
398     // We claim lazily (so races are possible but they're benign), which reduces the
399     // number of duplicate scans (the rsets of the regions in the cset can intersect).
400     // Claim the card after checking bounds above: the remembered set may contain
401     // random cards into current survivor, and we would then have an incorrectly
402     // claimed card in survivor space. Card table clear does not reset the card table
403     // of survivor space regions.
404     claim_card(card_index, region_idx_for_card);
405 
406     MemRegion const mr(card_start, MIN2(card_start + BOTConstants::N_words, top));
407 
408     scan_card(mr, region_idx_for_card);
409   }
410   if (_scan_state->set_iter_complete(region_idx)) {
411     // Scan the strong code root list attached to the current region
412     scan_strong_code_roots(r);
413   }
414   return false;
415 }
416 
417 void G1RemSet::scan_rem_set(G1ParScanThreadState* pss,
418                             CodeBlobClosure* heap_region_codeblobs,
419                             uint worker_i) {
420   double rs_time_start = os::elapsedTime();
421 
422   G1ScanObjsDuringScanRSClosure scan_cl(_g1, pss);
423   G1ScanRSForRegionClosure cl(_scan_state, &scan_cl, heap_region_codeblobs, worker_i);
424   _g1->collection_set_iterate_from(&cl, worker_i);
425 
426   double scan_rs_time_sec = (os::elapsedTime() - rs_time_start) -
427                              cl.strong_code_root_scan_time_sec();
428 
429   G1GCPhaseTimes* p = _g1p->phase_times();
430 
431   p->record_time_secs(G1GCPhaseTimes::ScanRS, worker_i, scan_rs_time_sec);
432   p->record_thread_work_item(G1GCPhaseTimes::ScanRS, worker_i, cl.cards_scanned(), G1GCPhaseTimes::ScanRSScannedCards);
433   p->record_thread_work_item(G1GCPhaseTimes::ScanRS, worker_i, cl.cards_claimed(), G1GCPhaseTimes::ScanRSClaimedCards);
434   p->record_thread_work_item(G1GCPhaseTimes::ScanRS, worker_i, cl.cards_skipped(), G1GCPhaseTimes::ScanRSSkippedCards);
435 
436   p->record_time_secs(G1GCPhaseTimes::CodeRoots, worker_i, cl.strong_code_root_scan_time_sec());
437 }
438 
439 // Closure used for updating rem sets. Only called during an evacuation pause.
440 class G1RefineCardClosure: public CardTableEntryClosure {
441   G1RemSet* _g1rs;
442   G1ScanObjsDuringUpdateRSClosure* _update_rs_cl;
443 
444   size_t _cards_scanned;
445   size_t _cards_skipped;
446 public:
447   G1RefineCardClosure(G1CollectedHeap* g1h, G1ScanObjsDuringUpdateRSClosure* update_rs_cl) :
448     _g1rs(g1h->g1_rem_set()), _update_rs_cl(update_rs_cl), _cards_scanned(0), _cards_skipped(0)
449   {}
450 
451   bool do_card_ptr(jbyte* card_ptr, uint worker_i) {
452     // The only time we care about recording cards that
453     // contain references that point into the collection set
454     // is during RSet updating within an evacuation pause.
455     // In this case worker_i should be the id of a GC worker thread.
456     assert(SafepointSynchronize::is_at_safepoint(), "not during an evacuation pause");
457 
458     bool card_scanned = _g1rs->refine_card_during_gc(card_ptr, _update_rs_cl);
459 
460     if (card_scanned) {
461       _cards_scanned++;
462     } else {
463       _cards_skipped++;
464     }
465     return true;
466   }
467 
468   size_t cards_scanned() const { return _cards_scanned; }
469   size_t cards_skipped() const { return _cards_skipped; }
470 };
471 
472 void G1RemSet::update_rem_set(G1ParScanThreadState* pss, uint worker_i) {
473   G1ScanObjsDuringUpdateRSClosure update_rs_cl(_g1, pss, worker_i);
474   G1RefineCardClosure refine_card_cl(_g1, &update_rs_cl);
475 
476   G1GCParPhaseTimesTracker x(_g1p->phase_times(), G1GCPhaseTimes::UpdateRS, worker_i);
477   if (G1HotCardCache::default_use_cache()) {
478     // Apply the closure to the entries of the hot card cache.
479     G1GCParPhaseTimesTracker y(_g1p->phase_times(), G1GCPhaseTimes::ScanHCC, worker_i);
480     _g1->iterate_hcc_closure(&refine_card_cl, worker_i);
481   }
482   // Apply the closure to all remaining log entries.
483   _g1->iterate_dirty_card_closure(&refine_card_cl, worker_i);
484 
485   G1GCPhaseTimes* p = _g1p->phase_times();
486   p->record_thread_work_item(G1GCPhaseTimes::UpdateRS, worker_i, refine_card_cl.cards_scanned(), G1GCPhaseTimes::UpdateRSScannedCards);
487   p->record_thread_work_item(G1GCPhaseTimes::UpdateRS, worker_i, refine_card_cl.cards_skipped(), G1GCPhaseTimes::UpdateRSSkippedCards);
488 }
489 
490 void G1RemSet::cleanupHRRS() {
491   HeapRegionRemSet::cleanup();
492 }
493 
494 void G1RemSet::oops_into_collection_set_do(G1ParScanThreadState* pss,
495                                            CodeBlobClosure* heap_region_codeblobs,
496                                            uint worker_i) {
497   update_rem_set(pss, worker_i);
498   scan_rem_set(pss, heap_region_codeblobs, worker_i);;
499 }
500 
501 void G1RemSet::prepare_for_oops_into_collection_set_do() {
502   DirtyCardQueueSet& dcqs = JavaThread::dirty_card_queue_set();
503   dcqs.concatenate_logs();
504 
505   _scan_state->reset();
506 }
507 
508 void G1RemSet::cleanup_after_oops_into_collection_set_do() {
509   G1GCPhaseTimes* phase_times = _g1->g1_policy()->phase_times();
510 
511   // Set all cards back to clean.
512   double start = os::elapsedTime();
513   _scan_state->clear_card_table(_g1->workers());
514   phase_times->record_clear_ct_time((os::elapsedTime() - start) * 1000.0);
515 }
516 
517 class G1ScrubRSClosure: public HeapRegionClosure {
518   G1CollectedHeap* _g1h;
519   G1CardLiveData* _live_data;
520 public:
521   G1ScrubRSClosure(G1CardLiveData* live_data) :
522     _g1h(G1CollectedHeap::heap()),
523     _live_data(live_data) { }
524 
525   bool do_heap_region(HeapRegion* r) {
526     if (!r->is_continues_humongous()) {
527       r->rem_set()->scrub(_live_data);
528     }
529     return false;
530   }
531 };
532 
533 void G1RemSet::scrub(uint worker_num, HeapRegionClaimer *hrclaimer) {
534   G1ScrubRSClosure scrub_cl(&_card_live_data);
535   _g1->heap_region_par_iterate_from_worker_offset(&scrub_cl, hrclaimer, worker_num);
536 }
537 
538 inline void check_card_ptr(jbyte* card_ptr, G1CardTable* ct) {
539 #ifdef ASSERT
540   G1CollectedHeap* g1 = G1CollectedHeap::heap();
541   assert(g1->is_in_exact(ct->addr_for(card_ptr)),
542          "Card at " PTR_FORMAT " index " SIZE_FORMAT " representing heap at " PTR_FORMAT " (%u) must be in committed heap",
543          p2i(card_ptr),
544          ct->index_for(ct->addr_for(card_ptr)),
545          p2i(ct->addr_for(card_ptr)),
546          g1->addr_to_region(ct->addr_for(card_ptr)));
547 #endif
548 }
549 
550 void G1RemSet::refine_card_concurrently(jbyte* card_ptr,
551                                         uint worker_i) {
552   assert(!_g1->is_gc_active(), "Only call concurrently");
553 
554   check_card_ptr(card_ptr, _ct);
555 
556   // If the card is no longer dirty, nothing to do.
557   if (*card_ptr != G1CardTable::dirty_card_val()) {
558     return;
559   }
560 
561   // Construct the region representing the card.
562   HeapWord* start = _ct->addr_for(card_ptr);
563   // And find the region containing it.
564   HeapRegion* r = _g1->heap_region_containing(start);
565 
566   // This check is needed for some uncommon cases where we should
567   // ignore the card.
568   //
569   // The region could be young.  Cards for young regions are
570   // distinctly marked (set to g1_young_gen), so the post-barrier will
571   // filter them out.  However, that marking is performed
572   // concurrently.  A write to a young object could occur before the
573   // card has been marked young, slipping past the filter.
574   //
575   // The card could be stale, because the region has been freed since
576   // the card was recorded. In this case the region type could be
577   // anything.  If (still) free or (reallocated) young, just ignore
578   // it.  If (reallocated) old or humongous, the later card trimming
579   // and additional checks in iteration may detect staleness.  At
580   // worst, we end up processing a stale card unnecessarily.
581   //
582   // In the normal (non-stale) case, the synchronization between the
583   // enqueueing of the card and processing it here will have ensured
584   // we see the up-to-date region type here.
585   if (!r->is_old_or_humongous()) {
586     return;
587   }
588 
589   // The result from the hot card cache insert call is either:
590   //   * pointer to the current card
591   //     (implying that the current card is not 'hot'),
592   //   * null
593   //     (meaning we had inserted the card ptr into the "hot" card cache,
594   //     which had some headroom),
595   //   * a pointer to a "hot" card that was evicted from the "hot" cache.
596   //
597 
598   if (_hot_card_cache->use_cache()) {
599     assert(!SafepointSynchronize::is_at_safepoint(), "sanity");
600 
601     const jbyte* orig_card_ptr = card_ptr;
602     card_ptr = _hot_card_cache->insert(card_ptr);
603     if (card_ptr == NULL) {
604       // There was no eviction. Nothing to do.
605       return;
606     } else if (card_ptr != orig_card_ptr) {
607       // Original card was inserted and an old card was evicted.
608       start = _ct->addr_for(card_ptr);
609       r = _g1->heap_region_containing(start);
610 
611       // Check whether the region formerly in the cache should be
612       // ignored, as discussed earlier for the original card.  The
613       // region could have been freed while in the cache.
614       if (!r->is_old_or_humongous()) {
615         return;
616       }
617     } // Else we still have the original card.
618   }
619 
620   // Trim the region designated by the card to what's been allocated
621   // in the region.  The card could be stale, or the card could cover
622   // (part of) an object at the end of the allocated space and extend
623   // beyond the end of allocation.
624 
625   // Non-humongous objects are only allocated in the old-gen during
626   // GC, so if region is old then top is stable.  Humongous object
627   // allocation sets top last; if top has not yet been set, this is
628   // a stale card and we'll end up with an empty intersection.  If
629   // this is not a stale card, the synchronization between the
630   // enqueuing of the card and processing it here will have ensured
631   // we see the up-to-date top here.
632   HeapWord* scan_limit = r->top();
633 
634   if (scan_limit <= start) {
635     // If the trimmed region is empty, the card must be stale.
636     return;
637   }
638 
639   // Okay to clean and process the card now.  There are still some
640   // stale card cases that may be detected by iteration and dealt with
641   // as iteration failure.
642   *const_cast<volatile jbyte*>(card_ptr) = G1CardTable::clean_card_val();
643 
644   // This fence serves two purposes.  First, the card must be cleaned
645   // before processing the contents.  Second, we can't proceed with
646   // processing until after the read of top, for synchronization with
647   // possibly concurrent humongous object allocation.  It's okay that
648   // reading top and reading type were racy wrto each other.  We need
649   // both set, in any order, to proceed.
650   OrderAccess::fence();
651 
652   // Don't use addr_for(card_ptr + 1) which can ask for
653   // a card beyond the heap.
654   HeapWord* end = start + G1CardTable::card_size_in_words;
655   MemRegion dirty_region(start, MIN2(scan_limit, end));
656   assert(!dirty_region.is_empty(), "sanity");
657 
658   G1ConcurrentRefineOopClosure conc_refine_cl(_g1, worker_i);
659 
660   bool card_processed =
661     r->oops_on_card_seq_iterate_careful<false>(dirty_region, &conc_refine_cl);
662 
663   // If unable to process the card then we encountered an unparsable
664   // part of the heap (e.g. a partially allocated object) while
665   // processing a stale card.  Despite the card being stale, redirty
666   // and re-enqueue, because we've already cleaned the card.  Without
667   // this we could incorrectly discard a non-stale card.
668   if (!card_processed) {
669     // The card might have gotten re-dirtied and re-enqueued while we
670     // worked.  (In fact, it's pretty likely.)
671     if (*card_ptr != G1CardTable::dirty_card_val()) {
672       *card_ptr = G1CardTable::dirty_card_val();
673       MutexLockerEx x(Shared_DirtyCardQ_lock,
674                       Mutex::_no_safepoint_check_flag);
675       DirtyCardQueue* sdcq =
676         JavaThread::dirty_card_queue_set().shared_dirty_card_queue();
677       sdcq->enqueue(card_ptr);
678     }
679   } else {
680     _num_conc_refined_cards++; // Unsynchronized update, only used for logging.
681   }
682 }
683 
684 bool G1RemSet::refine_card_during_gc(jbyte* card_ptr,
685                                      G1ScanObjsDuringUpdateRSClosure* update_rs_cl) {
686   assert(_g1->is_gc_active(), "Only call during GC");
687 
688   check_card_ptr(card_ptr, _ct);
689 
690   // If the card is no longer dirty, nothing to do. This covers cards that were already
691   // scanned as parts of the remembered sets.
692   if (*card_ptr != G1CardTable::dirty_card_val()) {
693     return false;
694   }
695 
696   // We claim lazily (so races are possible but they're benign), which reduces the
697   // number of potential duplicate scans (multiple threads may enqueue the same card twice).
698   *card_ptr = G1CardTable::clean_card_val() | G1CardTable::claimed_card_val();
699 
700   // Construct the region representing the card.
701   HeapWord* card_start = _ct->addr_for(card_ptr);
702   // And find the region containing it.
703   uint const card_region_idx = _g1->addr_to_region(card_start);
704 
705   _scan_state->add_dirty_region(card_region_idx);
706   HeapWord* scan_limit = _scan_state->scan_top(card_region_idx);
707   if (scan_limit <= card_start) {
708     // If the card starts above the area in the region containing objects to scan, skip it.
709     return false;
710   }
711 
712   // Don't use addr_for(card_ptr + 1) which can ask for
713   // a card beyond the heap.
714   HeapWord* card_end = card_start + G1CardTable::card_size_in_words;
715   MemRegion dirty_region(card_start, MIN2(scan_limit, card_end));
716   assert(!dirty_region.is_empty(), "sanity");
717 
718   HeapRegion* const card_region = _g1->region_at(card_region_idx);
719   update_rs_cl->set_region(card_region);
720   bool card_processed = card_region->oops_on_card_seq_iterate_careful<true>(dirty_region, update_rs_cl);
721   assert(card_processed, "must be");
722   return true;
723 }
724 
725 void G1RemSet::print_periodic_summary_info(const char* header, uint period_count) {
726   if ((G1SummarizeRSetStatsPeriod > 0) && log_is_enabled(Trace, gc, remset) &&
727       (period_count % G1SummarizeRSetStatsPeriod == 0)) {
728 
729     G1RemSetSummary current(this);
730     _prev_period_summary.subtract_from(&current);
731 
732     Log(gc, remset) log;
733     log.trace("%s", header);
734     ResourceMark rm;
735     LogStream ls(log.trace());
736     _prev_period_summary.print_on(&ls);
737 
738     _prev_period_summary.set(&current);
739   }
740 }
741 
742 void G1RemSet::print_summary_info() {
743   Log(gc, remset, exit) log;
744   if (log.is_trace()) {
745     log.trace(" Cumulative RS summary");
746     G1RemSetSummary current(this);
747     ResourceMark rm;
748     LogStream ls(log.trace());
749     current.print_on(&ls);
750   }
751 }
752 
753 void G1RemSet::create_card_live_data(WorkGang* workers, G1CMBitMap* mark_bitmap) {
754   _card_live_data.create(workers, mark_bitmap);
755 }
756 
757 void G1RemSet::finalize_card_live_data(WorkGang* workers, G1CMBitMap* mark_bitmap) {
758   _card_live_data.finalize(workers, mark_bitmap);
759 }
760 
761 void G1RemSet::verify_card_live_data(WorkGang* workers, G1CMBitMap* bitmap) {
762   _card_live_data.verify(workers, bitmap);
763 }
764 
765 void G1RemSet::clear_card_live_data(WorkGang* workers) {
766   _card_live_data.clear(workers);
767 }
768 
769 #ifdef ASSERT
770 void G1RemSet::verify_card_live_data_is_clear() {
771   _card_live_data.verify_is_clear();
772 }
773 #endif