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/g1SATBCardTableModRefBS.inline.hpp"
37 #include "gc/g1/heapRegion.inline.hpp"
38 #include "gc/g1/heapRegionManager.inline.hpp"
39 #include "gc/g1/heapRegionRemSet.hpp"
40 #include "gc/shared/gcTraceTime.inline.hpp"
41 #include "gc/shared/suspendibleThreadSet.hpp"
42 #include "memory/iterator.hpp"
43 #include "memory/resourceArea.hpp"
44 #include "oops/oop.inline.hpp"
45 #include "utilities/align.hpp"
46 #include "utilities/globalDefinitions.hpp"
47 #include "utilities/intHisto.hpp"
48 #include "utilities/stack.inline.hpp"
49
50 // Collects information about the overall remembered set scan progress during an evacuation.
51 class G1RemSetScanState : public CHeapObj<mtGC> {
52 private:
53 class G1ClearCardTableTask : public AbstractGangTask {
54 G1CollectedHeap* _g1h;
55 uint* _dirty_region_list;
56 size_t _num_dirty_regions;
57 size_t _chunk_length;
58
59 size_t volatile _cur_dirty_regions;
60 public:
61 G1ClearCardTableTask(G1CollectedHeap* g1h,
62 uint* dirty_region_list,
63 size_t num_dirty_regions,
64 size_t chunk_length) :
65 AbstractGangTask("G1 Clear Card Table Task"),
66 _g1h(g1h),
67 _dirty_region_list(dirty_region_list),
68 _num_dirty_regions(num_dirty_regions),
69 _chunk_length(chunk_length),
70 _cur_dirty_regions(0) {
71
72 assert(chunk_length > 0, "must be");
73 }
74
75 static size_t chunk_size() { return M; }
76
77 void work(uint worker_id) {
78 G1CardTable* ct = _g1h->card_table();
79
80 while (_cur_dirty_regions < _num_dirty_regions) {
81 size_t next = Atomic::add(_chunk_length, &_cur_dirty_regions) - _chunk_length;
82 size_t max = MIN2(next + _chunk_length, _num_dirty_regions);
83
84 for (size_t i = next; i < max; i++) {
85 HeapRegion* r = _g1h->region_at(_dirty_region_list[i]);
86 if (!r->is_survivor()) {
87 ct->clear(MemRegion(r->bottom(), r->end()));
88 }
89 }
90 }
91 }
92 };
93
94 size_t _max_regions;
95
96 // Scan progress for the remembered set of a single region. Transitions from
97 // Unclaimed -> Claimed -> Complete.
98 // At each of the transitions the thread that does the transition needs to perform
99 // some special action once. This is the reason for the extra "Claimed" state.
100 typedef jint G1RemsetIterState;
101
102 static const G1RemsetIterState Unclaimed = 0; // The remembered set has not been scanned yet.
103 static const G1RemsetIterState Claimed = 1; // The remembered set is currently being scanned.
104 static const G1RemsetIterState Complete = 2; // The remembered set has been completely scanned.
105
106 G1RemsetIterState volatile* _iter_states;
107 // The current location where the next thread should continue scanning in a region's
108 // remembered set.
109 size_t volatile* _iter_claims;
110
111 // Temporary buffer holding the regions we used to store remembered set scan duplicate
112 // information. These are also called "dirty". Valid entries are from [0.._cur_dirty_region)
113 uint* _dirty_region_buffer;
114
115 typedef jbyte IsDirtyRegionState;
116 static const IsDirtyRegionState Clean = 0;
117 static const IsDirtyRegionState Dirty = 1;
118 // Holds a flag for every region whether it is in the _dirty_region_buffer already
119 // to avoid duplicates. Uses jbyte since there are no atomic instructions for bools.
120 IsDirtyRegionState* _in_dirty_region_buffer;
121 size_t _cur_dirty_region;
122
123 // Creates a snapshot of the current _top values at the start of collection to
124 // filter out card marks that we do not want to scan.
125 class G1ResetScanTopClosure : public HeapRegionClosure {
126 private:
127 HeapWord** _scan_top;
128 public:
129 G1ResetScanTopClosure(HeapWord** scan_top) : _scan_top(scan_top) { }
130
131 virtual bool do_heap_region(HeapRegion* r) {
132 uint hrm_index = r->hrm_index();
133 if (!r->in_collection_set() && r->is_old_or_humongous()) {
134 _scan_top[hrm_index] = r->top();
135 } else {
136 _scan_top[hrm_index] = r->bottom();
137 }
138 return false;
139 }
140 };
141
142 // For each region, contains the maximum top() value to be used during this garbage
143 // collection. Subsumes common checks like filtering out everything but old and
144 // humongous regions outside the collection set.
145 // This is valid because we are not interested in scanning stray remembered set
146 // entries from free or archive regions.
147 HeapWord** _scan_top;
148 public:
149 G1RemSetScanState() :
150 _max_regions(0),
151 _iter_states(NULL),
152 _iter_claims(NULL),
153 _dirty_region_buffer(NULL),
154 _in_dirty_region_buffer(NULL),
155 _cur_dirty_region(0),
156 _scan_top(NULL) {
157 }
158
159 ~G1RemSetScanState() {
160 if (_iter_states != NULL) {
161 FREE_C_HEAP_ARRAY(G1RemsetIterState, _iter_states);
162 }
163 if (_iter_claims != NULL) {
164 FREE_C_HEAP_ARRAY(size_t, _iter_claims);
165 }
166 if (_dirty_region_buffer != NULL) {
167 FREE_C_HEAP_ARRAY(uint, _dirty_region_buffer);
168 }
169 if (_in_dirty_region_buffer != NULL) {
170 FREE_C_HEAP_ARRAY(IsDirtyRegionState, _in_dirty_region_buffer);
171 }
172 if (_scan_top != NULL) {
173 FREE_C_HEAP_ARRAY(HeapWord*, _scan_top);
174 }
175 }
176
177 void initialize(uint max_regions) {
178 assert(_iter_states == NULL, "Must not be initialized twice");
179 assert(_iter_claims == NULL, "Must not be initialized twice");
180 _max_regions = max_regions;
181 _iter_states = NEW_C_HEAP_ARRAY(G1RemsetIterState, max_regions, mtGC);
182 _iter_claims = NEW_C_HEAP_ARRAY(size_t, max_regions, mtGC);
183 _dirty_region_buffer = NEW_C_HEAP_ARRAY(uint, max_regions, mtGC);
184 _in_dirty_region_buffer = NEW_C_HEAP_ARRAY(IsDirtyRegionState, max_regions, mtGC);
185 _scan_top = NEW_C_HEAP_ARRAY(HeapWord*, max_regions, mtGC);
186 }
187
188 void reset() {
189 for (uint i = 0; i < _max_regions; i++) {
190 _iter_states[i] = Unclaimed;
191 }
192
193 G1ResetScanTopClosure cl(_scan_top);
194 G1CollectedHeap::heap()->heap_region_iterate(&cl);
195
196 memset((void*)_iter_claims, 0, _max_regions * sizeof(size_t));
197 memset(_in_dirty_region_buffer, Clean, _max_regions * sizeof(IsDirtyRegionState));
198 _cur_dirty_region = 0;
199 }
200
201 // Attempt to claim the remembered set of the region for iteration. Returns true
202 // if this call caused the transition from Unclaimed to Claimed.
203 inline bool claim_iter(uint region) {
204 assert(region < _max_regions, "Tried to access invalid region %u", region);
205 if (_iter_states[region] != Unclaimed) {
206 return false;
207 }
208 G1RemsetIterState res = Atomic::cmpxchg(Claimed, &_iter_states[region], Unclaimed);
209 return (res == Unclaimed);
210 }
211
212 // Try to atomically sets the iteration state to "complete". Returns true for the
213 // thread that caused the transition.
214 inline bool set_iter_complete(uint region) {
215 if (iter_is_complete(region)) {
216 return false;
217 }
218 G1RemsetIterState res = Atomic::cmpxchg(Complete, &_iter_states[region], Claimed);
219 return (res == Claimed);
220 }
221
222 // Returns true if the region's iteration is complete.
223 inline bool iter_is_complete(uint region) const {
224 assert(region < _max_regions, "Tried to access invalid region %u", region);
225 return _iter_states[region] == Complete;
226 }
227
228 // The current position within the remembered set of the given region.
229 inline size_t iter_claimed(uint region) const {
230 assert(region < _max_regions, "Tried to access invalid region %u", region);
231 return _iter_claims[region];
232 }
233
234 // Claim the next block of cards within the remembered set of the region with
235 // step size.
236 inline size_t iter_claimed_next(uint region, size_t step) {
237 return Atomic::add(step, &_iter_claims[region]) - step;
238 }
239
240 void add_dirty_region(uint region) {
241 if (_in_dirty_region_buffer[region] == Dirty) {
242 return;
243 }
244
245 bool marked_as_dirty = Atomic::cmpxchg(Dirty, &_in_dirty_region_buffer[region], Clean) == Clean;
246 if (marked_as_dirty) {
247 size_t allocated = Atomic::add(1u, &_cur_dirty_region) - 1;
248 _dirty_region_buffer[allocated] = region;
249 }
250 }
251
252 HeapWord* scan_top(uint region_idx) const {
253 return _scan_top[region_idx];
254 }
255
256 // Clear the card table of "dirty" regions.
257 void clear_card_table(WorkGang* workers) {
258 if (_cur_dirty_region == 0) {
259 return;
260 }
261
262 size_t const num_chunks = align_up(_cur_dirty_region * HeapRegion::CardsPerRegion, G1ClearCardTableTask::chunk_size()) / G1ClearCardTableTask::chunk_size();
263 uint const num_workers = (uint)MIN2(num_chunks, (size_t)workers->active_workers());
264 size_t const chunk_length = G1ClearCardTableTask::chunk_size() / HeapRegion::CardsPerRegion;
265
266 // Iterate over the dirty cards region list.
267 G1ClearCardTableTask cl(G1CollectedHeap::heap(), _dirty_region_buffer, _cur_dirty_region, chunk_length);
268
269 log_debug(gc, ergo)("Running %s using %u workers for " SIZE_FORMAT " "
270 "units of work for " SIZE_FORMAT " regions.",
271 cl.name(), num_workers, num_chunks, _cur_dirty_region);
272 workers->run_task(&cl, num_workers);
273
274 #ifndef PRODUCT
275 // Need to synchronize with concurrent cleanup since it needs to
276 // finish its card table clearing before we can verify.
277 G1CollectedHeap::heap()->wait_while_free_regions_coming();
278 G1CollectedHeap::heap()->verifier()->verify_card_table_cleanup();
279 #endif
280 }
281 };
282
283 G1RemSet::G1RemSet(G1CollectedHeap* g1,
284 G1CardTable* ct,
285 G1HotCardCache* hot_card_cache) :
286 _g1(g1),
287 _scan_state(new G1RemSetScanState()),
288 _num_conc_refined_cards(0),
289 _ct(ct),
290 _g1p(_g1->g1_policy()),
291 _hot_card_cache(hot_card_cache),
292 _prev_period_summary() {
293 }
294
295 G1RemSet::~G1RemSet() {
296 if (_scan_state != NULL) {
297 delete _scan_state;
298 }
299 }
300
301 uint G1RemSet::num_par_rem_sets() {
302 return MAX2(DirtyCardQueueSet::num_par_ids() + G1ConcurrentRefine::max_num_threads(), ParallelGCThreads);
303 }
304
305 void G1RemSet::initialize(size_t capacity, uint max_regions) {
306 G1FromCardCache::initialize(num_par_rem_sets(), max_regions);
307 _scan_state->initialize(max_regions);
308 {
309 GCTraceTime(Debug, gc, marking)("Initialize Card Live Data");
310 _card_live_data.initialize(capacity, max_regions);
311 }
312 if (G1PretouchAuxiliaryMemory) {
313 GCTraceTime(Debug, gc, marking)("Pre-Touch Card Live Data");
314 _card_live_data.pretouch();
315 }
316 }
317
318 G1ScanRSForRegionClosure::G1ScanRSForRegionClosure(G1RemSetScanState* scan_state,
319 G1ScanObjsDuringScanRSClosure* scan_obj_on_card,
320 CodeBlobClosure* code_root_cl,
321 uint worker_i) :
322 _scan_state(scan_state),
323 _scan_objs_on_card_cl(scan_obj_on_card),
324 _code_root_cl(code_root_cl),
325 _strong_code_root_scan_time_sec(0.0),
326 _cards_claimed(0),
327 _cards_scanned(0),
328 _cards_skipped(0),
329 _worker_i(worker_i) {
330 _g1h = G1CollectedHeap::heap();
331 _bot = _g1h->bot();
332 _ct = _g1h->card_table();
333 }
334
335 void G1ScanRSForRegionClosure::scan_card(MemRegion mr, uint region_idx_for_card) {
336 HeapRegion* const card_region = _g1h->region_at(region_idx_for_card);
337 _scan_objs_on_card_cl->set_region(card_region);
338 card_region->oops_on_card_seq_iterate_careful<true>(mr, _scan_objs_on_card_cl);
339 _cards_scanned++;
340 }
341
342 void G1ScanRSForRegionClosure::scan_strong_code_roots(HeapRegion* r) {
343 double scan_start = os::elapsedTime();
344 r->strong_code_roots_do(_code_root_cl);
345 _strong_code_root_scan_time_sec += (os::elapsedTime() - scan_start);
346 }
347
348 void G1ScanRSForRegionClosure::claim_card(size_t card_index, const uint region_idx_for_card){
349 _ct->set_card_claimed(card_index);
350 _scan_state->add_dirty_region(region_idx_for_card);
351 }
352
353 bool G1ScanRSForRegionClosure::do_heap_region(HeapRegion* r) {
354 assert(r->in_collection_set(), "should only be called on elements of CS.");
355 uint region_idx = r->hrm_index();
356
357 if (_scan_state->iter_is_complete(region_idx)) {
358 return false;
359 }
360 if (_scan_state->claim_iter(region_idx)) {
361 // If we ever free the collection set concurrently, we should also
362 // clear the card table concurrently therefore we won't need to
363 // add regions of the collection set to the dirty cards region.
364 _scan_state->add_dirty_region(region_idx);
365 }
366
367 // We claim cards in blocks so as to reduce the contention.
368 size_t const block_size = G1RSetScanBlockSize;
369
370 HeapRegionRemSetIterator iter(r->rem_set());
371 size_t card_index;
372
373 size_t claimed_card_block = _scan_state->iter_claimed_next(region_idx, block_size);
374 for (size_t current_card = 0; iter.has_next(card_index); current_card++) {
375 if (current_card >= claimed_card_block + block_size) {
376 claimed_card_block = _scan_state->iter_claimed_next(region_idx, block_size);
377 }
378 if (current_card < claimed_card_block) {
379 _cards_skipped++;
380 continue;
381 }
382 _cards_claimed++;
383
384 // If the card is dirty, then G1 will scan it during Update RS.
385 if (_ct->is_card_claimed(card_index) || _ct->is_card_dirty(card_index)) {
386 continue;
387 }
388
389 HeapWord* const card_start = _g1h->bot()->address_for_index(card_index);
390 uint const region_idx_for_card = _g1h->addr_to_region(card_start);
391
392 assert(_g1h->region_at(region_idx_for_card)->is_in_reserved(card_start),
393 "Card start " PTR_FORMAT " to scan outside of region %u", p2i(card_start), _g1h->region_at(region_idx_for_card)->hrm_index());
394 HeapWord* const top = _scan_state->scan_top(region_idx_for_card);
395 if (card_start >= top) {
396 continue;
397 }
398
399 // We claim lazily (so races are possible but they're benign), which reduces the
400 // number of duplicate scans (the rsets of the regions in the cset can intersect).
401 // Claim the card after checking bounds above: the remembered set may contain
402 // random cards into current survivor, and we would then have an incorrectly
403 // claimed card in survivor space. Card table clear does not reset the card table
404 // of survivor space regions.
405 claim_card(card_index, region_idx_for_card);
406
407 MemRegion const mr(card_start, MIN2(card_start + BOTConstants::N_words, top));
408
409 scan_card(mr, region_idx_for_card);
410 }
411 if (_scan_state->set_iter_complete(region_idx)) {
412 // Scan the strong code root list attached to the current region
413 scan_strong_code_roots(r);
414 }
415 return false;
416 }
417
418 void G1RemSet::scan_rem_set(G1ParScanThreadState* pss,
419 CodeBlobClosure* heap_region_codeblobs,
420 uint worker_i) {
421 double rs_time_start = os::elapsedTime();
422
423 G1ScanObjsDuringScanRSClosure scan_cl(_g1, pss);
424 G1ScanRSForRegionClosure cl(_scan_state, &scan_cl, heap_region_codeblobs, worker_i);
425 _g1->collection_set_iterate_from(&cl, worker_i);
426
427 double scan_rs_time_sec = (os::elapsedTime() - rs_time_start) -
428 cl.strong_code_root_scan_time_sec();
429
430 G1GCPhaseTimes* p = _g1p->phase_times();
431
432 p->record_time_secs(G1GCPhaseTimes::ScanRS, worker_i, scan_rs_time_sec);
433 p->record_thread_work_item(G1GCPhaseTimes::ScanRS, worker_i, cl.cards_scanned(), G1GCPhaseTimes::ScanRSScannedCards);
434 p->record_thread_work_item(G1GCPhaseTimes::ScanRS, worker_i, cl.cards_claimed(), G1GCPhaseTimes::ScanRSClaimedCards);
435 p->record_thread_work_item(G1GCPhaseTimes::ScanRS, worker_i, cl.cards_skipped(), G1GCPhaseTimes::ScanRSSkippedCards);
436
437 p->record_time_secs(G1GCPhaseTimes::CodeRoots, worker_i, cl.strong_code_root_scan_time_sec());
438 }
439
440 // Closure used for updating rem sets. Only called during an evacuation pause.
441 class G1RefineCardClosure: public CardTableEntryClosure {
442 G1RemSet* _g1rs;
443 G1ScanObjsDuringUpdateRSClosure* _update_rs_cl;
444
445 size_t _cards_scanned;
446 size_t _cards_skipped;
447 public:
448 G1RefineCardClosure(G1CollectedHeap* g1h, G1ScanObjsDuringUpdateRSClosure* update_rs_cl) :
449 _g1rs(g1h->g1_rem_set()), _update_rs_cl(update_rs_cl), _cards_scanned(0), _cards_skipped(0)
450 {}
451
452 bool do_card_ptr(jbyte* card_ptr, uint worker_i) {
453 // The only time we care about recording cards that
454 // contain references that point into the collection set
455 // is during RSet updating within an evacuation pause.
456 // In this case worker_i should be the id of a GC worker thread.
457 assert(SafepointSynchronize::is_at_safepoint(), "not during an evacuation pause");
458
459 bool card_scanned = _g1rs->refine_card_during_gc(card_ptr, _update_rs_cl);
460
461 if (card_scanned) {
462 _cards_scanned++;
463 } else {
464 _cards_skipped++;
465 }
466 return true;
467 }
468
469 size_t cards_scanned() const { return _cards_scanned; }
470 size_t cards_skipped() const { return _cards_skipped; }
471 };
472
473 void G1RemSet::update_rem_set(G1ParScanThreadState* pss, uint worker_i) {
474 G1ScanObjsDuringUpdateRSClosure update_rs_cl(_g1, pss, worker_i);
475 G1RefineCardClosure refine_card_cl(_g1, &update_rs_cl);
476
477 G1GCParPhaseTimesTracker x(_g1p->phase_times(), G1GCPhaseTimes::UpdateRS, worker_i);
478 if (G1HotCardCache::default_use_cache()) {
479 // Apply the closure to the entries of the hot card cache.
480 G1GCParPhaseTimesTracker y(_g1p->phase_times(), G1GCPhaseTimes::ScanHCC, worker_i);
481 _g1->iterate_hcc_closure(&refine_card_cl, worker_i);
482 }
483 // Apply the closure to all remaining log entries.
484 _g1->iterate_dirty_card_closure(&refine_card_cl, worker_i);
485
486 G1GCPhaseTimes* p = _g1p->phase_times();
487 p->record_thread_work_item(G1GCPhaseTimes::UpdateRS, worker_i, refine_card_cl.cards_scanned(), G1GCPhaseTimes::UpdateRSScannedCards);
488 p->record_thread_work_item(G1GCPhaseTimes::UpdateRS, worker_i, refine_card_cl.cards_skipped(), G1GCPhaseTimes::UpdateRSSkippedCards);
489 }
490
491 void G1RemSet::cleanupHRRS() {
492 HeapRegionRemSet::cleanup();
493 }
494
495 void G1RemSet::oops_into_collection_set_do(G1ParScanThreadState* pss,
496 CodeBlobClosure* heap_region_codeblobs,
497 uint worker_i) {
498 update_rem_set(pss, worker_i);
499 scan_rem_set(pss, heap_region_codeblobs, worker_i);;
500 }
501
502 void G1RemSet::prepare_for_oops_into_collection_set_do() {
503 DirtyCardQueueSet& dcqs = JavaThread::dirty_card_queue_set();
504 dcqs.concatenate_logs();
505
506 _scan_state->reset();
507 }
508
509 void G1RemSet::cleanup_after_oops_into_collection_set_do() {
510 G1GCPhaseTimes* phase_times = _g1->g1_policy()->phase_times();
511
512 // Set all cards back to clean.
513 double start = os::elapsedTime();
514 _scan_state->clear_card_table(_g1->workers());
515 phase_times->record_clear_ct_time((os::elapsedTime() - start) * 1000.0);
516 }
517
518 class G1ScrubRSClosure: public HeapRegionClosure {
519 G1CollectedHeap* _g1h;
520 G1CardLiveData* _live_data;
521 public:
522 G1ScrubRSClosure(G1CardLiveData* live_data) :
523 _g1h(G1CollectedHeap::heap()),
524 _live_data(live_data) { }
525
526 bool do_heap_region(HeapRegion* r) {
527 if (!r->is_continues_humongous()) {
528 r->rem_set()->scrub(_live_data);
529 }
530 return false;
531 }
532 };
533
534 void G1RemSet::scrub(uint worker_num, HeapRegionClaimer *hrclaimer) {
535 G1ScrubRSClosure scrub_cl(&_card_live_data);
536 _g1->heap_region_par_iterate_from_worker_offset(&scrub_cl, hrclaimer, worker_num);
537 }
538
539 inline void check_card_ptr(jbyte* card_ptr, G1CardTable* ct) {
540 #ifdef ASSERT
541 G1CollectedHeap* g1 = G1CollectedHeap::heap();
542 assert(g1->is_in_exact(ct->addr_for(card_ptr)),
543 "Card at " PTR_FORMAT " index " SIZE_FORMAT " representing heap at " PTR_FORMAT " (%u) must be in committed heap",
544 p2i(card_ptr),
545 ct->index_for(ct->addr_for(card_ptr)),
546 p2i(ct->addr_for(card_ptr)),
547 g1->addr_to_region(ct->addr_for(card_ptr)));
548 #endif
549 }
550
551 void G1RemSet::refine_card_concurrently(jbyte* card_ptr,
552 uint worker_i) {
553 assert(!_g1->is_gc_active(), "Only call concurrently");
554
555 check_card_ptr(card_ptr, _ct);
556
557 // If the card is no longer dirty, nothing to do.
558 if (*card_ptr != G1CardTable::dirty_card_val()) {
559 return;
560 }
561
562 // Construct the region representing the card.
563 HeapWord* start = _ct->addr_for(card_ptr);
564 // And find the region containing it.
565 HeapRegion* r = _g1->heap_region_containing(start);
566
567 // This check is needed for some uncommon cases where we should
568 // ignore the card.
569 //
570 // The region could be young. Cards for young regions are
571 // distinctly marked (set to g1_young_gen), so the post-barrier will
572 // filter them out. However, that marking is performed
573 // concurrently. A write to a young object could occur before the
574 // card has been marked young, slipping past the filter.
575 //
576 // The card could be stale, because the region has been freed since
577 // the card was recorded. In this case the region type could be
578 // anything. If (still) free or (reallocated) young, just ignore
579 // it. If (reallocated) old or humongous, the later card trimming
580 // and additional checks in iteration may detect staleness. At
581 // worst, we end up processing a stale card unnecessarily.
582 //
583 // In the normal (non-stale) case, the synchronization between the
584 // enqueueing of the card and processing it here will have ensured
585 // we see the up-to-date region type here.
586 if (!r->is_old_or_humongous()) {
587 return;
588 }
589
590 // The result from the hot card cache insert call is either:
591 // * pointer to the current card
592 // (implying that the current card is not 'hot'),
593 // * null
594 // (meaning we had inserted the card ptr into the "hot" card cache,
595 // which had some headroom),
596 // * a pointer to a "hot" card that was evicted from the "hot" cache.
597 //
598
599 if (_hot_card_cache->use_cache()) {
600 assert(!SafepointSynchronize::is_at_safepoint(), "sanity");
601
602 const jbyte* orig_card_ptr = card_ptr;
603 card_ptr = _hot_card_cache->insert(card_ptr);
604 if (card_ptr == NULL) {
605 // There was no eviction. Nothing to do.
606 return;
607 } else if (card_ptr != orig_card_ptr) {
608 // Original card was inserted and an old card was evicted.
609 start = _ct->addr_for(card_ptr);
610 r = _g1->heap_region_containing(start);
611
612 // Check whether the region formerly in the cache should be
613 // ignored, as discussed earlier for the original card. The
614 // region could have been freed while in the cache.
615 if (!r->is_old_or_humongous()) {
616 return;
617 }
618 } // Else we still have the original card.
619 }
620
621 // Trim the region designated by the card to what's been allocated
622 // in the region. The card could be stale, or the card could cover
623 // (part of) an object at the end of the allocated space and extend
624 // beyond the end of allocation.
625
626 // Non-humongous objects are only allocated in the old-gen during
627 // GC, so if region is old then top is stable. Humongous object
628 // allocation sets top last; if top has not yet been set, this is
629 // a stale card and we'll end up with an empty intersection. If
630 // this is not a stale card, the synchronization between the
631 // enqueuing of the card and processing it here will have ensured
632 // we see the up-to-date top here.
633 HeapWord* scan_limit = r->top();
634
635 if (scan_limit <= start) {
636 // If the trimmed region is empty, the card must be stale.
637 return;
638 }
639
640 // Okay to clean and process the card now. There are still some
641 // stale card cases that may be detected by iteration and dealt with
642 // as iteration failure.
643 *const_cast<volatile jbyte*>(card_ptr) = G1CardTable::clean_card_val();
644
645 // This fence serves two purposes. First, the card must be cleaned
646 // before processing the contents. Second, we can't proceed with
647 // processing until after the read of top, for synchronization with
648 // possibly concurrent humongous object allocation. It's okay that
649 // reading top and reading type were racy wrto each other. We need
650 // both set, in any order, to proceed.
651 OrderAccess::fence();
652
653 // Don't use addr_for(card_ptr + 1) which can ask for
654 // a card beyond the heap.
655 HeapWord* end = start + G1CardTable::card_size_in_words;
656 MemRegion dirty_region(start, MIN2(scan_limit, end));
657 assert(!dirty_region.is_empty(), "sanity");
658
659 G1ConcurrentRefineOopClosure conc_refine_cl(_g1, worker_i);
660
661 bool card_processed =
662 r->oops_on_card_seq_iterate_careful<false>(dirty_region, &conc_refine_cl);
663
664 // If unable to process the card then we encountered an unparsable
665 // part of the heap (e.g. a partially allocated object) while
666 // processing a stale card. Despite the card being stale, redirty
667 // and re-enqueue, because we've already cleaned the card. Without
668 // this we could incorrectly discard a non-stale card.
669 if (!card_processed) {
670 // The card might have gotten re-dirtied and re-enqueued while we
671 // worked. (In fact, it's pretty likely.)
672 if (*card_ptr != G1CardTable::dirty_card_val()) {
673 *card_ptr = G1CardTable::dirty_card_val();
674 MutexLockerEx x(Shared_DirtyCardQ_lock,
675 Mutex::_no_safepoint_check_flag);
676 DirtyCardQueue* sdcq =
677 JavaThread::dirty_card_queue_set().shared_dirty_card_queue();
678 sdcq->enqueue(card_ptr);
679 }
680 } else {
681 _num_conc_refined_cards++; // Unsynchronized update, only used for logging.
682 }
683 }
684
685 bool G1RemSet::refine_card_during_gc(jbyte* card_ptr,
686 G1ScanObjsDuringUpdateRSClosure* update_rs_cl) {
687 assert(_g1->is_gc_active(), "Only call during GC");
688
689 check_card_ptr(card_ptr, _ct);
690
691 // If the card is no longer dirty, nothing to do. This covers cards that were already
692 // scanned as parts of the remembered sets.
693 if (*card_ptr != G1CardTable::dirty_card_val()) {
694 return false;
695 }
696
697 // We claim lazily (so races are possible but they're benign), which reduces the
698 // number of potential duplicate scans (multiple threads may enqueue the same card twice).
699 *card_ptr = G1CardTable::clean_card_val() | G1CardTable::claimed_card_val();
700
701 // Construct the region representing the card.
702 HeapWord* card_start = _ct->addr_for(card_ptr);
703 // And find the region containing it.
704 uint const card_region_idx = _g1->addr_to_region(card_start);
705
706 _scan_state->add_dirty_region(card_region_idx);
707 HeapWord* scan_limit = _scan_state->scan_top(card_region_idx);
708 if (scan_limit <= card_start) {
709 // If the card starts above the area in the region containing objects to scan, skip it.
710 return false;
711 }
712
713 // Don't use addr_for(card_ptr + 1) which can ask for
714 // a card beyond the heap.
715 HeapWord* card_end = card_start + G1CardTable::card_size_in_words;
716 MemRegion dirty_region(card_start, MIN2(scan_limit, card_end));
717 assert(!dirty_region.is_empty(), "sanity");
718
719 HeapRegion* const card_region = _g1->region_at(card_region_idx);
720 update_rs_cl->set_region(card_region);
721 bool card_processed = card_region->oops_on_card_seq_iterate_careful<true>(dirty_region, update_rs_cl);
722 assert(card_processed, "must be");
723 return true;
724 }
725
726 void G1RemSet::print_periodic_summary_info(const char* header, uint period_count) {
727 if ((G1SummarizeRSetStatsPeriod > 0) && log_is_enabled(Trace, gc, remset) &&
728 (period_count % G1SummarizeRSetStatsPeriod == 0)) {
729
730 G1RemSetSummary current(this);
731 _prev_period_summary.subtract_from(¤t);
732
733 Log(gc, remset) log;
734 log.trace("%s", header);
735 ResourceMark rm;
736 LogStream ls(log.trace());
737 _prev_period_summary.print_on(&ls);
738
739 _prev_period_summary.set(¤t);
740 }
741 }
742
743 void G1RemSet::print_summary_info() {
744 Log(gc, remset, exit) log;
745 if (log.is_trace()) {
746 log.trace(" Cumulative RS summary");
747 G1RemSetSummary current(this);
748 ResourceMark rm;
749 LogStream ls(log.trace());
750 current.print_on(&ls);
751 }
752 }
753
754 void G1RemSet::create_card_live_data(WorkGang* workers, G1CMBitMap* mark_bitmap) {
755 _card_live_data.create(workers, mark_bitmap);
756 }
757
758 void G1RemSet::finalize_card_live_data(WorkGang* workers, G1CMBitMap* mark_bitmap) {
759 _card_live_data.finalize(workers, mark_bitmap);
760 }
761
762 void G1RemSet::verify_card_live_data(WorkGang* workers, G1CMBitMap* bitmap) {
763 _card_live_data.verify(workers, bitmap);
764 }
765
766 void G1RemSet::clear_card_live_data(WorkGang* workers) {
767 _card_live_data.clear(workers);
768 }
769
770 #ifdef ASSERT
771 void G1RemSet::verify_card_live_data_is_clear() {
772 _card_live_data.verify_is_clear();
773 }
774 #endif