1 /*
2 * Copyright (c) 2001, 2017, Oracle and/or its affiliates. All rights reserved.
3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
4 *
5 * This code is free software; you can redistribute it and/or modify it
6 * under the terms of the GNU General Public License version 2 only, as
7 * published by the Free Software Foundation.
8 *
9 * This code is distributed in the hope that it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
12 * version 2 for more details (a copy is included in the LICENSE file that
13 * accompanied this code).
14 *
15 * You should have received a copy of the GNU General Public License version
16 * 2 along with this work; if not, write to the Free Software Foundation,
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
18 *
19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
20 * or visit www.oracle.com if you need additional information or have any
21 * questions.
22 *
23 */
24
25 #include "precompiled.hpp"
26 #include "gc/g1/concurrentG1Refine.hpp"
27 #include "gc/g1/dirtyCardQueue.hpp"
28 #include "gc/g1/g1BlockOffsetTable.inline.hpp"
29 #include "gc/g1/g1CollectedHeap.inline.hpp"
30 #include "gc/g1/g1FromCardCache.hpp"
31 #include "gc/g1/g1GCPhaseTimes.hpp"
32 #include "gc/g1/g1HotCardCache.hpp"
33 #include "gc/g1/g1OopClosures.inline.hpp"
34 #include "gc/g1/g1RemSet.inline.hpp"
35 #include "gc/g1/g1SATBCardTableModRefBS.inline.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 "memory/iterator.hpp"
41 #include "memory/resourceArea.hpp"
42 #include "oops/oop.inline.hpp"
43 #include "utilities/globalDefinitions.hpp"
44 #include "utilities/intHisto.hpp"
45 #include "utilities/stack.inline.hpp"
46
47 // Collects information about the overall remembered set scan progress during an evacuation.
48 class G1RemSetScanState : public CHeapObj<mtGC> {
49 private:
50 class G1ClearCardTableTask : public AbstractGangTask {
51 G1CollectedHeap* _g1h;
52 uint* _dirty_region_list;
53 size_t _num_dirty_regions;
54 size_t _chunk_length;
55
56 size_t volatile _cur_dirty_regions;
57 public:
58 G1ClearCardTableTask(G1CollectedHeap* g1h,
59 uint* dirty_region_list,
60 size_t num_dirty_regions,
61 size_t chunk_length) :
62 AbstractGangTask("G1 Clear Card Table Task"),
63 _g1h(g1h),
64 _dirty_region_list(dirty_region_list),
65 _num_dirty_regions(num_dirty_regions),
66 _chunk_length(chunk_length),
67 _cur_dirty_regions(0) {
68
69 assert(chunk_length > 0, "must be");
70 }
71
72 static size_t chunk_size() { return M; }
73
74 void work(uint worker_id) {
75 G1SATBCardTableModRefBS* ct_bs = _g1h->g1_barrier_set();
76
77 while (_cur_dirty_regions < _num_dirty_regions) {
78 size_t next = Atomic::add(_chunk_length, &_cur_dirty_regions) - _chunk_length;
79 size_t max = MIN2(next + _chunk_length, _num_dirty_regions);
80
81 for (size_t i = next; i < max; i++) {
82 HeapRegion* r = _g1h->region_at(_dirty_region_list[i]);
83 if (!r->is_survivor()) {
84 ct_bs->clear(MemRegion(r->bottom(), r->end()));
85 }
86 }
87 }
88 }
89 };
90
91 size_t _max_regions;
92
93 // Scan progress for the remembered set of a single region. Transitions from
94 // Unclaimed -> Claimed -> Complete.
95 // At each of the transitions the thread that does the transition needs to perform
96 // some special action once. This is the reason for the extra "Claimed" state.
97 typedef jint G1RemsetIterState;
98
99 static const G1RemsetIterState Unclaimed = 0; // The remembered set has not been scanned yet.
100 static const G1RemsetIterState Claimed = 1; // The remembered set is currently being scanned.
101 static const G1RemsetIterState Complete = 2; // The remembered set has been completely scanned.
102
103 G1RemsetIterState volatile* _iter_states;
104 // The current location where the next thread should continue scanning in a region's
105 // remembered set.
106 size_t volatile* _iter_claims;
107
108 // Temporary buffer holding the regions we used to store remembered set scan duplicate
109 // information. These are also called "dirty". Valid entries are from [0.._cur_dirty_region)
110 uint* _dirty_region_buffer;
111
112 typedef jbyte IsDirtyRegionState;
113 static const IsDirtyRegionState Clean = 0;
114 static const IsDirtyRegionState Dirty = 1;
115 // Holds a flag for every region whether it is in the _dirty_region_buffer already
116 // to avoid duplicates. Uses jbyte since there are no atomic instructions for bools.
117 IsDirtyRegionState* _in_dirty_region_buffer;
118 size_t _cur_dirty_region;
119
120 // Creates a snapshot of the current _top values at the start of collection to
121 // filter out card marks that we do not want to scan.
122 class G1ResetScanTopClosure : public HeapRegionClosure {
123 private:
124 HeapWord** _scan_top;
125 public:
126 G1ResetScanTopClosure(HeapWord** scan_top) : _scan_top(scan_top) { }
127
128 virtual bool doHeapRegion(HeapRegion* r) {
129 uint hrm_index = r->hrm_index();
130 if (!r->in_collection_set() && r->is_old_or_humongous()) {
131 _scan_top[hrm_index] = r->top();
132 } else {
133 _scan_top[hrm_index] = r->bottom();
134 }
135 return false;
136 }
137 };
138
139 // For each region, contains the maximum top() value to be used during this garbage
140 // collection. Subsumes common checks like filtering out everything but old and
141 // humongous regions outside the collection set.
142 // This is valid because we are not interested in scanning stray remembered set
143 // entries from free or archive regions.
144 HeapWord** _scan_top;
145 public:
146 G1RemSetScanState() :
147 _max_regions(0),
148 _iter_states(NULL),
149 _iter_claims(NULL),
150 _dirty_region_buffer(NULL),
151 _in_dirty_region_buffer(NULL),
152 _cur_dirty_region(0),
153 _scan_top(NULL) {
154 }
155
156 ~G1RemSetScanState() {
157 if (_iter_states != NULL) {
158 FREE_C_HEAP_ARRAY(G1RemsetIterState, _iter_states);
159 }
160 if (_iter_claims != NULL) {
161 FREE_C_HEAP_ARRAY(size_t, _iter_claims);
162 }
163 if (_dirty_region_buffer != NULL) {
164 FREE_C_HEAP_ARRAY(uint, _dirty_region_buffer);
165 }
166 if (_in_dirty_region_buffer != NULL) {
167 FREE_C_HEAP_ARRAY(IsDirtyRegionState, _in_dirty_region_buffer);
168 }
169 if (_scan_top != NULL) {
170 FREE_C_HEAP_ARRAY(HeapWord*, _scan_top);
171 }
172 }
173
174 void initialize(uint max_regions) {
175 assert(_iter_states == NULL, "Must not be initialized twice");
176 assert(_iter_claims == NULL, "Must not be initialized twice");
177 _max_regions = max_regions;
178 _iter_states = NEW_C_HEAP_ARRAY(G1RemsetIterState, max_regions, mtGC);
179 _iter_claims = NEW_C_HEAP_ARRAY(size_t, max_regions, mtGC);
180 _dirty_region_buffer = NEW_C_HEAP_ARRAY(uint, max_regions, mtGC);
181 _in_dirty_region_buffer = NEW_C_HEAP_ARRAY(IsDirtyRegionState, max_regions, mtGC);
182 _scan_top = NEW_C_HEAP_ARRAY(HeapWord*, max_regions, mtGC);
183 }
184
185 void reset() {
186 for (uint i = 0; i < _max_regions; i++) {
187 _iter_states[i] = Unclaimed;
188 }
189
190 memset(_scan_top, 0, _max_regions * sizeof(HeapWord*));
191 G1ResetScanTopClosure cl(_scan_top);
192 G1CollectedHeap::heap()->heap_region_iterate(&cl);
193
194 memset((void*)_iter_claims, 0, _max_regions * sizeof(size_t));
195 memset(_in_dirty_region_buffer, Clean, _max_regions * sizeof(IsDirtyRegionState));
196 _cur_dirty_region = 0;
197 }
198
199 // Attempt to claim the remembered set of the region for iteration. Returns true
200 // if this call caused the transition from Unclaimed to Claimed.
201 inline bool claim_iter(uint region) {
202 assert(region < _max_regions, "Tried to access invalid region %u", region);
203 if (_iter_states[region] != Unclaimed) {
204 return false;
205 }
206 jint res = Atomic::cmpxchg(Claimed, (jint*)(&_iter_states[region]), Unclaimed);
207 return (res == Unclaimed);
208 }
209
210 // Try to atomically sets the iteration state to "complete". Returns true for the
211 // thread that caused the transition.
212 inline bool set_iter_complete(uint region) {
213 if (iter_is_complete(region)) {
214 return false;
215 }
216 jint res = Atomic::cmpxchg(Complete, (jint*)(&_iter_states[region]), Claimed);
217 return (res == Claimed);
218 }
219
220 // Returns true if the region's iteration is complete.
221 inline bool iter_is_complete(uint region) const {
222 assert(region < _max_regions, "Tried to access invalid region %u", region);
223 return _iter_states[region] == Complete;
224 }
225
226 // The current position within the remembered set of the given region.
227 inline size_t iter_claimed(uint region) const {
228 assert(region < _max_regions, "Tried to access invalid region %u", region);
229 return _iter_claims[region];
230 }
231
232 // Claim the next block of cards within the remembered set of the region with
233 // step size.
234 inline size_t iter_claimed_next(uint region, size_t step) {
235 return Atomic::add(step, &_iter_claims[region]) - step;
236 }
237
238 void add_dirty_region(uint region) {
239 if (_in_dirty_region_buffer[region] == Dirty) {
240 return;
241 }
242
243 bool marked_as_dirty = Atomic::cmpxchg(Dirty, &_in_dirty_region_buffer[region], Clean) == Clean;
244 if (marked_as_dirty) {
245 size_t allocated = Atomic::add(1, &_cur_dirty_region) - 1;
246 _dirty_region_buffer[allocated] = region;
247 }
248 }
249
250 HeapWord* scan_top(uint region_idx) const {
251 return _scan_top[region_idx];
252 }
253
254 // Clear the card table of "dirty" regions.
255 void clear_card_table(WorkGang* workers) {
256 if (_cur_dirty_region == 0) {
257 return;
258 }
259
260 size_t const num_chunks = align_size_up(_cur_dirty_region * HeapRegion::CardsPerRegion, G1ClearCardTableTask::chunk_size()) / G1ClearCardTableTask::chunk_size();
261 uint const num_workers = (uint)MIN2(num_chunks, (size_t)workers->active_workers());
262 size_t const chunk_length = G1ClearCardTableTask::chunk_size() / HeapRegion::CardsPerRegion;
263
264 // Iterate over the dirty cards region list.
265 G1ClearCardTableTask cl(G1CollectedHeap::heap(), _dirty_region_buffer, _cur_dirty_region, chunk_length);
266
267 log_debug(gc, ergo)("Running %s using %u workers for " SIZE_FORMAT " "
268 "units of work for " SIZE_FORMAT " regions.",
269 cl.name(), num_workers, num_chunks, _cur_dirty_region);
270 workers->run_task(&cl, num_workers);
271
272 #ifndef PRODUCT
273 // Need to synchronize with concurrent cleanup since it needs to
274 // finish its card table clearing before we can verify.
275 G1CollectedHeap::heap()->wait_while_free_regions_coming();
276 G1CollectedHeap::heap()->verifier()->verify_card_table_cleanup();
277 #endif
278 }
279 };
280
281 G1RemSet::G1RemSet(G1CollectedHeap* g1,
282 CardTableModRefBS* ct_bs,
283 G1HotCardCache* hot_card_cache) :
284 _g1(g1),
285 _scan_state(new G1RemSetScanState()),
286 _conc_refine_cards(0),
287 _ct_bs(ct_bs),
288 _g1p(_g1->g1_policy()),
289 _hot_card_cache(hot_card_cache),
290 _prev_period_summary(),
291 _into_cset_dirty_card_queue_set(false)
292 {
293 if (log_is_enabled(Trace, gc, remset)) {
294 _prev_period_summary.initialize(this);
295 }
296 // Initialize the card queue set used to hold cards containing
297 // references into the collection set.
298 _into_cset_dirty_card_queue_set.initialize(NULL, // Should never be called by the Java code
299 DirtyCardQ_CBL_mon,
300 DirtyCardQ_FL_lock,
301 -1, // never trigger processing
302 -1, // no limit on length
303 Shared_DirtyCardQ_lock,
304 &JavaThread::dirty_card_queue_set());
305 }
306
307 G1RemSet::~G1RemSet() {
308 if (_scan_state != NULL) {
309 delete _scan_state;
310 }
311 }
312
313 uint G1RemSet::num_par_rem_sets() {
314 return MAX2(DirtyCardQueueSet::num_par_ids() + ConcurrentG1Refine::thread_num(), ParallelGCThreads);
315 }
316
317 void G1RemSet::initialize(size_t capacity, uint max_regions) {
318 G1FromCardCache::initialize(num_par_rem_sets(), max_regions);
319 _scan_state->initialize(max_regions);
320 {
321 GCTraceTime(Debug, gc, marking)("Initialize Card Live Data");
322 _card_live_data.initialize(capacity, max_regions);
323 }
324 if (G1PretouchAuxiliaryMemory) {
325 GCTraceTime(Debug, gc, marking)("Pre-Touch Card Live Data");
326 _card_live_data.pretouch();
327 }
328 }
329
330 G1ScanRSClosure::G1ScanRSClosure(G1RemSetScanState* scan_state,
331 G1ParPushHeapRSClosure* push_heap_cl,
332 CodeBlobClosure* code_root_cl,
333 uint worker_i) :
334 _scan_state(scan_state),
335 _push_heap_cl(push_heap_cl),
336 _code_root_cl(code_root_cl),
337 _strong_code_root_scan_time_sec(0.0),
338 _cards_claimed(0),
339 _cards_scanned(0),
340 _cards_skipped(0),
341 _worker_i(worker_i) {
342 _g1h = G1CollectedHeap::heap();
343 _bot = _g1h->bot();
344 _ct_bs = _g1h->g1_barrier_set();
345 _block_size = MAX2<size_t>(G1RSetScanBlockSize, 1);
346 }
347
348 void G1ScanRSClosure::scan_card(size_t index, HeapWord* card_start, HeapRegion *r) {
349 MemRegion card_region(card_start, BOTConstants::N_words);
350 MemRegion pre_gc_allocated(r->bottom(), _scan_state->scan_top(r->hrm_index()));
351 MemRegion mr = pre_gc_allocated.intersection(card_region);
352 if (!mr.is_empty() && !_ct_bs->is_card_claimed(index)) {
353 // We make the card as "claimed" lazily (so races are possible
354 // but they're benign), which reduces the number of duplicate
355 // scans (the rsets of the regions in the cset can intersect).
356 _ct_bs->set_card_claimed(index);
357 _push_heap_cl->set_region(r);
358 r->oops_on_card_seq_iterate_careful<true>(mr, _push_heap_cl);
359 _cards_scanned++;
360 }
361 }
362
363 void G1ScanRSClosure::scan_strong_code_roots(HeapRegion* r) {
364 double scan_start = os::elapsedTime();
365 r->strong_code_roots_do(_code_root_cl);
366 _strong_code_root_scan_time_sec += (os::elapsedTime() - scan_start);
367 }
368
369 bool G1ScanRSClosure::doHeapRegion(HeapRegion* r) {
370 assert(r->in_collection_set(), "should only be called on elements of CS.");
371 uint region_idx = r->hrm_index();
372
373 if (_scan_state->iter_is_complete(region_idx)) {
374 return false;
375 }
376 if (_scan_state->claim_iter(region_idx)) {
377 // If we ever free the collection set concurrently, we should also
378 // clear the card table concurrently therefore we won't need to
379 // add regions of the collection set to the dirty cards region.
380 _scan_state->add_dirty_region(region_idx);
381 }
382
383 HeapRegionRemSetIterator iter(r->rem_set());
384 size_t card_index;
385
386 // We claim cards in block so as to reduce the contention. The block size is determined by
387 // the G1RSetScanBlockSize parameter.
388 size_t claimed_card_block = _scan_state->iter_claimed_next(region_idx, _block_size);
389 for (size_t current_card = 0; iter.has_next(card_index); current_card++) {
390 if (current_card >= claimed_card_block + _block_size) {
391 claimed_card_block = _scan_state->iter_claimed_next(region_idx, _block_size);
392 }
393 if (current_card < claimed_card_block) {
394 _cards_skipped++;
395 continue;
396 }
397 HeapWord* card_start = _g1h->bot()->address_for_index(card_index);
398
399 HeapRegion* card_region = _g1h->heap_region_containing(card_start);
400 _cards_claimed++;
401
402 _scan_state->add_dirty_region(card_region->hrm_index());
403
404 // If the card is dirty, then we will scan it during updateRS.
405 if (!card_region->in_collection_set() &&
406 !_ct_bs->is_card_dirty(card_index)) {
407 scan_card(card_index, card_start, card_region);
408 }
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(G1ParPushHeapRSClosure* oops_in_heap_closure,
418 CodeBlobClosure* heap_region_codeblobs,
419 uint worker_i) {
420 double rs_time_start = os::elapsedTime();
421
422 G1ScanRSClosure cl(_scan_state, oops_in_heap_closure, heap_region_codeblobs, worker_i);
423 _g1->collection_set_iterate_from(&cl, worker_i);
424
425 double scan_rs_time_sec = (os::elapsedTime() - rs_time_start) -
426 cl.strong_code_root_scan_time_sec();
427
428 G1GCPhaseTimes* p = _g1p->phase_times();
429
430 p->record_time_secs(G1GCPhaseTimes::ScanRS, worker_i, scan_rs_time_sec);
431 p->record_thread_work_item(G1GCPhaseTimes::ScanRS, worker_i, cl.cards_scanned(), G1GCPhaseTimes::ScannedCards);
432 p->record_thread_work_item(G1GCPhaseTimes::ScanRS, worker_i, cl.cards_claimed(), G1GCPhaseTimes::ClaimedCards);
433 p->record_thread_work_item(G1GCPhaseTimes::ScanRS, worker_i, cl.cards_skipped(), G1GCPhaseTimes::SkippedCards);
434
435 p->record_time_secs(G1GCPhaseTimes::CodeRoots, worker_i, cl.strong_code_root_scan_time_sec());
436 }
437
438 // Closure used for updating RSets and recording references that
439 // point into the collection set. Only called during an
440 // evacuation pause.
441
442 class RefineRecordRefsIntoCSCardTableEntryClosure: public CardTableEntryClosure {
443 G1RemSet* _g1rs;
444 DirtyCardQueue* _into_cset_dcq;
445 G1ParPushHeapRSClosure* _cl;
446 public:
447 RefineRecordRefsIntoCSCardTableEntryClosure(G1CollectedHeap* g1h,
448 DirtyCardQueue* into_cset_dcq,
449 G1ParPushHeapRSClosure* cl) :
450 _g1rs(g1h->g1_rem_set()), _into_cset_dcq(into_cset_dcq), _cl(cl)
451 {}
452
453 bool do_card_ptr(jbyte* card_ptr, uint worker_i) {
454 // The only time we care about recording cards that
455 // contain references that point into the collection set
456 // is during RSet updating within an evacuation pause.
457 // In this case worker_i should be the id of a GC worker thread.
458 assert(SafepointSynchronize::is_at_safepoint(), "not during an evacuation pause");
459 assert(worker_i < ParallelGCThreads, "should be a GC worker");
460
461 if (_g1rs->refine_card_during_gc(card_ptr, worker_i, _cl)) {
462 // 'card_ptr' contains references that point into the collection
463 // set. We need to record the card in the DCQS
464 // (_into_cset_dirty_card_queue_set)
465 // that's used for that purpose.
466 //
467 // Enqueue the card
468 _into_cset_dcq->enqueue(card_ptr);
469 }
470 return true;
471 }
472 };
473
474 void G1RemSet::update_rem_set(DirtyCardQueue* into_cset_dcq,
475 G1ParPushHeapRSClosure* oops_in_heap_closure,
476 uint worker_i) {
477 RefineRecordRefsIntoCSCardTableEntryClosure into_cset_update_rs_cl(_g1, into_cset_dcq, oops_in_heap_closure);
478
479 G1GCParPhaseTimesTracker x(_g1p->phase_times(), G1GCPhaseTimes::UpdateRS, worker_i);
480 if (G1HotCardCache::default_use_cache()) {
481 // Apply the closure to the entries of the hot card cache.
482 G1GCParPhaseTimesTracker y(_g1p->phase_times(), G1GCPhaseTimes::ScanHCC, worker_i);
483 _g1->iterate_hcc_closure(&into_cset_update_rs_cl, worker_i);
484 }
485 // Apply the closure to all remaining log entries.
486 _g1->iterate_dirty_card_closure(&into_cset_update_rs_cl, worker_i);
487 }
488
489 void G1RemSet::cleanupHRRS() {
490 HeapRegionRemSet::cleanup();
491 }
492
493 void G1RemSet::oops_into_collection_set_do(G1ParPushHeapRSClosure* cl,
494 CodeBlobClosure* heap_region_codeblobs,
495 uint worker_i) {
496 // A DirtyCardQueue that is used to hold cards containing references
497 // that point into the collection set. This DCQ is associated with a
498 // special DirtyCardQueueSet (see g1CollectedHeap.hpp). Under normal
499 // circumstances (i.e. the pause successfully completes), these cards
500 // are just discarded (there's no need to update the RSets of regions
501 // that were in the collection set - after the pause these regions
502 // are wholly 'free' of live objects. In the event of an evacuation
503 // failure the cards/buffers in this queue set are passed to the
504 // DirtyCardQueueSet that is used to manage RSet updates
505 DirtyCardQueue into_cset_dcq(&_into_cset_dirty_card_queue_set);
506
507 update_rem_set(&into_cset_dcq, cl, worker_i);
508 scan_rem_set(cl, heap_region_codeblobs, worker_i);;
509 }
510
511 void G1RemSet::prepare_for_oops_into_collection_set_do() {
512 _g1->set_refine_cte_cl_concurrency(false);
513 DirtyCardQueueSet& dcqs = JavaThread::dirty_card_queue_set();
514 dcqs.concatenate_logs();
515
516 _scan_state->reset();
517 }
518
519 void G1RemSet::cleanup_after_oops_into_collection_set_do() {
520 G1GCPhaseTimes* phase_times = _g1->g1_policy()->phase_times();
521 // Cleanup after copy
522 _g1->set_refine_cte_cl_concurrency(true);
523
524 // Set all cards back to clean.
525 double start = os::elapsedTime();
526 _scan_state->clear_card_table(_g1->workers());
527 phase_times->record_clear_ct_time((os::elapsedTime() - start) * 1000.0);
528
529 DirtyCardQueueSet& into_cset_dcqs = _into_cset_dirty_card_queue_set;
530
531 if (_g1->evacuation_failed()) {
532 double restore_remembered_set_start = os::elapsedTime();
533
534 // Restore remembered sets for the regions pointing into the collection set.
535 // We just need to transfer the completed buffers from the DirtyCardQueueSet
536 // used to hold cards that contain references that point into the collection set
537 // to the DCQS used to hold the deferred RS updates.
538 _g1->dirty_card_queue_set().merge_bufferlists(&into_cset_dcqs);
539 phase_times->record_evac_fail_restore_remsets((os::elapsedTime() - restore_remembered_set_start) * 1000.0);
540 }
541
542 // Free any completed buffers in the DirtyCardQueueSet used to hold cards
543 // which contain references that point into the collection.
544 _into_cset_dirty_card_queue_set.clear();
545 assert(_into_cset_dirty_card_queue_set.completed_buffers_num() == 0,
546 "all buffers should be freed");
547 _into_cset_dirty_card_queue_set.clear_n_completed_buffers();
548 }
549
550 class G1ScrubRSClosure: public HeapRegionClosure {
551 G1CollectedHeap* _g1h;
552 G1CardLiveData* _live_data;
553 public:
554 G1ScrubRSClosure(G1CardLiveData* live_data) :
555 _g1h(G1CollectedHeap::heap()),
556 _live_data(live_data) { }
557
558 bool doHeapRegion(HeapRegion* r) {
559 if (!r->is_continues_humongous()) {
560 r->rem_set()->scrub(_live_data);
561 }
562 return false;
563 }
564 };
565
566 void G1RemSet::scrub(uint worker_num, HeapRegionClaimer *hrclaimer) {
567 G1ScrubRSClosure scrub_cl(&_card_live_data);
568 _g1->heap_region_par_iterate(&scrub_cl, worker_num, hrclaimer);
569 }
570
571 inline void check_card_ptr(jbyte* card_ptr, CardTableModRefBS* ct_bs) {
572 #ifdef ASSERT
573 G1CollectedHeap* g1 = G1CollectedHeap::heap();
574 assert(g1->is_in_exact(ct_bs->addr_for(card_ptr)),
575 "Card at " PTR_FORMAT " index " SIZE_FORMAT " representing heap at " PTR_FORMAT " (%u) must be in committed heap",
576 p2i(card_ptr),
577 ct_bs->index_for(ct_bs->addr_for(card_ptr)),
578 p2i(ct_bs->addr_for(card_ptr)),
579 g1->addr_to_region(ct_bs->addr_for(card_ptr)));
580 #endif
581 }
582
583 G1UpdateRSOrPushRefOopClosure::G1UpdateRSOrPushRefOopClosure(G1CollectedHeap* g1h,
584 G1ParPushHeapRSClosure* push_ref_cl,
585 bool record_refs_into_cset,
586 uint worker_i) :
587 _g1(g1h),
588 _from(NULL),
589 _record_refs_into_cset(record_refs_into_cset),
590 _has_refs_into_cset(false),
591 _push_ref_cl(push_ref_cl),
592 _worker_i(worker_i) { }
593
594 void G1RemSet::refine_card_concurrently(jbyte* card_ptr,
595 uint worker_i) {
596 assert(!_g1->is_gc_active(), "Only call concurrently");
597
598 check_card_ptr(card_ptr, _ct_bs);
599
600 // If the card is no longer dirty, nothing to do.
601 if (*card_ptr != CardTableModRefBS::dirty_card_val()) {
602 return;
603 }
604
605 // Construct the region representing the card.
606 HeapWord* start = _ct_bs->addr_for(card_ptr);
607 // And find the region containing it.
608 HeapRegion* r = _g1->heap_region_containing(start);
609
610 // This check is needed for some uncommon cases where we should
611 // ignore the card.
612 //
613 // The region could be young. Cards for young regions are
614 // distinctly marked (set to g1_young_gen), so the post-barrier will
615 // filter them out. However, that marking is performed
616 // concurrently. A write to a young object could occur before the
617 // card has been marked young, slipping past the filter.
618 //
619 // The card could be stale, because the region has been freed since
620 // the card was recorded. In this case the region type could be
621 // anything. If (still) free or (reallocated) young, just ignore
622 // it. If (reallocated) old or humongous, the later card trimming
623 // and additional checks in iteration may detect staleness. At
624 // worst, we end up processing a stale card unnecessarily.
625 //
626 // In the normal (non-stale) case, the synchronization between the
627 // enqueueing of the card and processing it here will have ensured
628 // we see the up-to-date region type here.
629 if (!r->is_old_or_humongous()) {
630 return;
631 }
632
633 // While we are processing RSet buffers during the collection, we
634 // actually don't want to scan any cards on the collection set,
635 // since we don't want to update remembered sets with entries that
636 // point into the collection set, given that live objects from the
637 // collection set are about to move and such entries will be stale
638 // very soon. This change also deals with a reliability issue which
639 // involves scanning a card in the collection set and coming across
640 // an array that was being chunked and looking malformed. Note,
641 // however, that if evacuation fails, we have to scan any objects
642 // that were not moved and create any missing entries.
643 if (r->in_collection_set()) {
644 return;
645 }
646
647 // The result from the hot card cache insert call is either:
648 // * pointer to the current card
649 // (implying that the current card is not 'hot'),
650 // * null
651 // (meaning we had inserted the card ptr into the "hot" card cache,
652 // which had some headroom),
653 // * a pointer to a "hot" card that was evicted from the "hot" cache.
654 //
655
656 if (_hot_card_cache->use_cache()) {
657 assert(!SafepointSynchronize::is_at_safepoint(), "sanity");
658
659 const jbyte* orig_card_ptr = card_ptr;
660 card_ptr = _hot_card_cache->insert(card_ptr);
661 if (card_ptr == NULL) {
662 // There was no eviction. Nothing to do.
663 return;
664 } else if (card_ptr != orig_card_ptr) {
665 // Original card was inserted and an old card was evicted.
666 start = _ct_bs->addr_for(card_ptr);
667 r = _g1->heap_region_containing(start);
668
669 // Check whether the region formerly in the cache should be
670 // ignored, as discussed earlier for the original card. The
671 // region could have been freed while in the cache. The cset is
672 // not relevant here, since we're in concurrent phase.
673 if (!r->is_old_or_humongous()) {
674 return;
675 }
676 } // Else we still have the original card.
677 }
678
679 // Trim the region designated by the card to what's been allocated
680 // in the region. The card could be stale, or the card could cover
681 // (part of) an object at the end of the allocated space and extend
682 // beyond the end of allocation.
683
684 // Non-humongous objects are only allocated in the old-gen during
685 // GC, so if region is old then top is stable. Humongous object
686 // allocation sets top last; if top has not yet been set, this is
687 // a stale card and we'll end up with an empty intersection. If
688 // this is not a stale card, the synchronization between the
689 // enqueuing of the card and processing it here will have ensured
690 // we see the up-to-date top here.
691 HeapWord* scan_limit = r->top();
692
693 if (scan_limit <= start) {
694 // If the trimmed region is empty, the card must be stale.
695 return;
696 }
697
698 // Okay to clean and process the card now. There are still some
699 // stale card cases that may be detected by iteration and dealt with
700 // as iteration failure.
701 *const_cast<volatile jbyte*>(card_ptr) = CardTableModRefBS::clean_card_val();
702
703 // This fence serves two purposes. First, the card must be cleaned
704 // before processing the contents. Second, we can't proceed with
705 // processing until after the read of top, for synchronization with
706 // possibly concurrent humongous object allocation. It's okay that
707 // reading top and reading type were racy wrto each other. We need
708 // both set, in any order, to proceed.
709 OrderAccess::fence();
710
711 // Don't use addr_for(card_ptr + 1) which can ask for
712 // a card beyond the heap.
713 HeapWord* end = start + CardTableModRefBS::card_size_in_words;
714 MemRegion dirty_region(start, MIN2(scan_limit, end));
715 assert(!dirty_region.is_empty(), "sanity");
716
717 G1ConcurrentRefineOopClosure conc_refine_cl(_g1, worker_i);
718
719 bool card_processed =
720 r->oops_on_card_seq_iterate_careful<false>(dirty_region, &conc_refine_cl);
721
722 // If unable to process the card then we encountered an unparsable
723 // part of the heap (e.g. a partially allocated object) while
724 // processing a stale card. Despite the card being stale, redirty
725 // and re-enqueue, because we've already cleaned the card. Without
726 // this we could incorrectly discard a non-stale card.
727 if (!card_processed) {
728 // The card might have gotten re-dirtied and re-enqueued while we
729 // worked. (In fact, it's pretty likely.)
730 if (*card_ptr != CardTableModRefBS::dirty_card_val()) {
731 *card_ptr = CardTableModRefBS::dirty_card_val();
732 MutexLockerEx x(Shared_DirtyCardQ_lock,
733 Mutex::_no_safepoint_check_flag);
734 DirtyCardQueue* sdcq =
735 JavaThread::dirty_card_queue_set().shared_dirty_card_queue();
736 sdcq->enqueue(card_ptr);
737 }
738 } else {
739 _conc_refine_cards++;
740 }
741 }
742
743 bool G1RemSet::refine_card_during_gc(jbyte* card_ptr,
744 uint worker_i,
745 G1ParPushHeapRSClosure* oops_in_heap_closure) {
746 assert(_g1->is_gc_active(), "Only call during GC");
747
748 check_card_ptr(card_ptr, _ct_bs);
749
750 // If the card is no longer dirty, nothing to do. This covers cards that were already
751 // scanned as parts of the remembered sets.
752 if (*card_ptr != CardTableModRefBS::dirty_card_val()) {
753 // No need to return that this card contains refs that point
754 // into the collection set.
755 return false;
756 }
757
758 // Construct the region representing the card.
759 HeapWord* start = _ct_bs->addr_for(card_ptr);
760 // And find the region containing it.
761 HeapRegion* r = _g1->heap_region_containing(start);
762
763 HeapWord* scan_limit = _scan_state->scan_top(r->hrm_index());
764 if (scan_limit <= start) {
765 // If the card starts above the area in the region containing objects to scan, skip it.
766 return false;
767 }
768
769 // Okay to clean and process the card now. There are still some
770 // stale card cases that may be detected by iteration and dealt with
771 // as iteration failure.
772 *const_cast<volatile jbyte*>(card_ptr) = CardTableModRefBS::clean_card_val();
773
774 // Don't use addr_for(card_ptr + 1) which can ask for
775 // a card beyond the heap.
776 HeapWord* end = start + CardTableModRefBS::card_size_in_words;
777 MemRegion dirty_region(start, MIN2(scan_limit, end));
778 assert(!dirty_region.is_empty(), "sanity");
779
780 G1UpdateRSOrPushRefOopClosure update_rs_oop_cl(_g1,
781 oops_in_heap_closure,
782 true,
783 worker_i);
784 update_rs_oop_cl.set_from(r);
785
786 bool card_processed =
787 r->oops_on_card_seq_iterate_careful<true>(dirty_region,
788 &update_rs_oop_cl);
789 assert(card_processed, "must be");
790 _conc_refine_cards++;
791
792 return update_rs_oop_cl.has_refs_into_cset();
793 }
794
795 void G1RemSet::print_periodic_summary_info(const char* header, uint period_count) {
796 if ((G1SummarizeRSetStatsPeriod > 0) && log_is_enabled(Trace, gc, remset) &&
797 (period_count % G1SummarizeRSetStatsPeriod == 0)) {
798
799 if (!_prev_period_summary.initialized()) {
800 _prev_period_summary.initialize(this);
801 }
802
803 G1RemSetSummary current;
804 current.initialize(this);
805 _prev_period_summary.subtract_from(¤t);
806
807 Log(gc, remset) log;
808 log.trace("%s", header);
809 ResourceMark rm;
810 _prev_period_summary.print_on(log.trace_stream());
811
812 _prev_period_summary.set(¤t);
813 }
814 }
815
816 void G1RemSet::print_summary_info() {
817 Log(gc, remset, exit) log;
818 if (log.is_trace()) {
819 log.trace(" Cumulative RS summary");
820 G1RemSetSummary current;
821 current.initialize(this);
822 ResourceMark rm;
823 current.print_on(log.trace_stream());
824 }
825 }
826
827 void G1RemSet::prepare_for_verify() {
828 if (G1HRRSFlushLogBuffersOnVerify &&
829 (VerifyBeforeGC || VerifyAfterGC)
830 && (!_g1->collector_state()->full_collection() || G1VerifyRSetsDuringFullGC)) {
831 cleanupHRRS();
832 _g1->set_refine_cte_cl_concurrency(false);
833 if (SafepointSynchronize::is_at_safepoint()) {
834 DirtyCardQueueSet& dcqs = JavaThread::dirty_card_queue_set();
835 dcqs.concatenate_logs();
836 }
837
838 bool use_hot_card_cache = _hot_card_cache->use_cache();
839 _hot_card_cache->set_use_cache(false);
840
841 DirtyCardQueue into_cset_dcq(&_into_cset_dirty_card_queue_set);
842 update_rem_set(&into_cset_dcq, NULL, 0);
843 _into_cset_dirty_card_queue_set.clear();
844
845 _hot_card_cache->set_use_cache(use_hot_card_cache);
846 assert(JavaThread::dirty_card_queue_set().completed_buffers_num() == 0, "All should be consumed");
847 }
848 }
849
850 void G1RemSet::create_card_live_data(WorkGang* workers, G1CMBitMap* mark_bitmap) {
851 _card_live_data.create(workers, mark_bitmap);
852 }
853
854 void G1RemSet::finalize_card_live_data(WorkGang* workers, G1CMBitMap* mark_bitmap) {
855 _card_live_data.finalize(workers, mark_bitmap);
856 }
857
858 void G1RemSet::verify_card_live_data(WorkGang* workers, G1CMBitMap* bitmap) {
859 _card_live_data.verify(workers, bitmap);
860 }
861
862 void G1RemSet::clear_card_live_data(WorkGang* workers) {
863 _card_live_data.clear(workers);
864 }
865
866 #ifdef ASSERT
867 void G1RemSet::verify_card_live_data_is_clear() {
868 _card_live_data.verify_is_clear();
869 }
870 #endif