1 /*
2 * Copyright (c) 2001, 2012, Oracle and/or its affiliates. All rights reserved.
3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
4 *
5 * This code is free software; you can redistribute it and/or modify it
6 * under the terms of the GNU General Public License version 2 only, as
7 * published by the Free Software Foundation.
8 *
9 * This code is distributed in the hope that it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
12 * version 2 for more details (a copy is included in the LICENSE file that
13 * accompanied this code).
14 *
15 * You should have received a copy of the GNU General Public License version
16 * 2 along with this work; if not, write to the Free Software Foundation,
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
18 *
19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
20 * or visit www.oracle.com if you need additional information or have any
21 * questions.
22 *
23 */
24
25 #include "precompiled.hpp"
26 #include "gc_implementation/g1/bufferingOopClosure.hpp"
27 #include "gc_implementation/g1/concurrentG1Refine.hpp"
28 #include "gc_implementation/g1/concurrentG1RefineThread.hpp"
29 #include "gc_implementation/g1/g1BlockOffsetTable.inline.hpp"
30 #include "gc_implementation/g1/g1CollectedHeap.inline.hpp"
31 #include "gc_implementation/g1/g1CollectorPolicy.hpp"
32 #include "gc_implementation/g1/g1GCPhaseTimes.hpp"
33 #include "gc_implementation/g1/g1OopClosures.inline.hpp"
34 #include "gc_implementation/g1/g1RemSet.inline.hpp"
35 #include "gc_implementation/g1/heapRegionSeq.inline.hpp"
36 #include "memory/iterator.hpp"
37 #include "oops/oop.inline.hpp"
38 #include "utilities/intHisto.hpp"
39
40 #define CARD_REPEAT_HISTO 0
41
42 #if CARD_REPEAT_HISTO
43 static size_t ct_freq_sz;
44 static jbyte* ct_freq = NULL;
45
46 void init_ct_freq_table(size_t heap_sz_bytes) {
47 if (ct_freq == NULL) {
48 ct_freq_sz = heap_sz_bytes/CardTableModRefBS::card_size;
49 ct_freq = new jbyte[ct_freq_sz];
50 for (size_t j = 0; j < ct_freq_sz; j++) ct_freq[j] = 0;
51 }
52 }
53
54 void ct_freq_note_card(size_t index) {
55 assert(0 <= index && index < ct_freq_sz, "Bounds error.");
56 if (ct_freq[index] < 100) { ct_freq[index]++; }
57 }
58
59 static IntHistogram card_repeat_count(10, 10);
60
61 void ct_freq_update_histo_and_reset() {
62 for (size_t j = 0; j < ct_freq_sz; j++) {
63 card_repeat_count.add_entry(ct_freq[j]);
64 ct_freq[j] = 0;
65 }
66
67 }
68 #endif
69
70 G1RemSet::G1RemSet(G1CollectedHeap* g1, CardTableModRefBS* ct_bs)
71 : _g1(g1), _conc_refine_cards(0),
72 _ct_bs(ct_bs), _g1p(_g1->g1_policy()),
73 _cg1r(g1->concurrent_g1_refine()),
74 _cset_rs_update_cl(NULL),
75 _cards_scanned(NULL), _total_cards_scanned(0)
76 {
77 _seq_task = new SubTasksDone(NumSeqTasks);
78 guarantee(n_workers() > 0, "There should be some workers");
79 _cset_rs_update_cl = NEW_C_HEAP_ARRAY(OopsInHeapRegionClosure*, n_workers(), mtGC);
80 for (uint i = 0; i < n_workers(); i++) {
81 _cset_rs_update_cl[i] = NULL;
82 }
83 }
84
85 G1RemSet::~G1RemSet() {
86 delete _seq_task;
87 for (uint i = 0; i < n_workers(); i++) {
88 assert(_cset_rs_update_cl[i] == NULL, "it should be");
89 }
90 FREE_C_HEAP_ARRAY(OopsInHeapRegionClosure*, _cset_rs_update_cl, mtGC);
91 }
92
93 void CountNonCleanMemRegionClosure::do_MemRegion(MemRegion mr) {
94 if (_g1->is_in_g1_reserved(mr.start())) {
95 _n += (int) ((mr.byte_size() / CardTableModRefBS::card_size));
96 if (_start_first == NULL) _start_first = mr.start();
97 }
98 }
99
100 class ScanRSClosure : public HeapRegionClosure {
101 size_t _cards_done, _cards;
102 G1CollectedHeap* _g1h;
103 OopsInHeapRegionClosure* _oc;
104 G1BlockOffsetSharedArray* _bot_shared;
105 CardTableModRefBS *_ct_bs;
106 int _worker_i;
107 int _block_size;
108 bool _try_claimed;
109 public:
110 ScanRSClosure(OopsInHeapRegionClosure* oc, int worker_i) :
111 _oc(oc),
112 _cards(0),
113 _cards_done(0),
114 _worker_i(worker_i),
115 _try_claimed(false)
116 {
117 _g1h = G1CollectedHeap::heap();
118 _bot_shared = _g1h->bot_shared();
119 _ct_bs = (CardTableModRefBS*) (_g1h->barrier_set());
120 _block_size = MAX2<int>(G1RSetScanBlockSize, 1);
121 }
122
123 void set_try_claimed() { _try_claimed = true; }
124
125 void scanCard(size_t index, HeapRegion *r) {
126 // Stack allocate the DirtyCardToOopClosure instance
127 HeapRegionDCTOC cl(_g1h, r, _oc,
128 CardTableModRefBS::Precise,
129 HeapRegionDCTOC::IntoCSFilterKind);
130
131 // Set the "from" region in the closure.
132 _oc->set_region(r);
133 HeapWord* card_start = _bot_shared->address_for_index(index);
134 HeapWord* card_end = card_start + G1BlockOffsetSharedArray::N_words;
135 Space *sp = SharedHeap::heap()->space_containing(card_start);
136 MemRegion sm_region = sp->used_region_at_save_marks();
137 MemRegion mr = sm_region.intersection(MemRegion(card_start,card_end));
138 if (!mr.is_empty() && !_ct_bs->is_card_claimed(index)) {
139 // We make the card as "claimed" lazily (so races are possible
140 // but they're benign), which reduces the number of duplicate
141 // scans (the rsets of the regions in the cset can intersect).
142 _ct_bs->set_card_claimed(index);
143 _cards_done++;
144 cl.do_MemRegion(mr);
145 }
146 }
147
148 void printCard(HeapRegion* card_region, size_t card_index,
149 HeapWord* card_start) {
150 gclog_or_tty->print_cr("T %d Region [" PTR_FORMAT ", " PTR_FORMAT ") "
151 "RS names card %p: "
152 "[" PTR_FORMAT ", " PTR_FORMAT ")",
153 _worker_i,
154 card_region->bottom(), card_region->end(),
155 card_index,
156 card_start, card_start + G1BlockOffsetSharedArray::N_words);
157 }
158
159 bool doHeapRegion(HeapRegion* r) {
160 assert(r->in_collection_set(), "should only be called on elements of CS.");
161 HeapRegionRemSet* hrrs = r->rem_set();
162 if (hrrs->iter_is_complete()) return false; // All done.
163 if (!_try_claimed && !hrrs->claim_iter()) return false;
164 // If we ever free the collection set concurrently, we should also
165 // clear the card table concurrently therefore we won't need to
166 // add regions of the collection set to the dirty cards region.
167 _g1h->push_dirty_cards_region(r);
168 // If we didn't return above, then
169 // _try_claimed || r->claim_iter()
170 // is true: either we're supposed to work on claimed-but-not-complete
171 // regions, or we successfully claimed the region.
172 HeapRegionRemSetIterator* iter = _g1h->rem_set_iterator(_worker_i);
173 hrrs->init_iterator(iter);
174 size_t card_index;
175
176 // We claim cards in block so as to recude the contention. The block size is determined by
177 // the G1RSetScanBlockSize parameter.
178 size_t jump_to_card = hrrs->iter_claimed_next(_block_size);
179 for (size_t current_card = 0; iter->has_next(card_index); current_card++) {
180 if (current_card >= jump_to_card + _block_size) {
181 jump_to_card = hrrs->iter_claimed_next(_block_size);
182 }
183 if (current_card < jump_to_card) continue;
184 HeapWord* card_start = _g1h->bot_shared()->address_for_index(card_index);
185 #if 0
186 gclog_or_tty->print("Rem set iteration yielded card [" PTR_FORMAT ", " PTR_FORMAT ").\n",
187 card_start, card_start + CardTableModRefBS::card_size_in_words);
188 #endif
189
190 HeapRegion* card_region = _g1h->heap_region_containing(card_start);
191 assert(card_region != NULL, "Yielding cards not in the heap?");
192 _cards++;
193
194 if (!card_region->is_on_dirty_cards_region_list()) {
195 _g1h->push_dirty_cards_region(card_region);
196 }
197
198 // If the card is dirty, then we will scan it during updateRS.
199 if (!card_region->in_collection_set() &&
200 !_ct_bs->is_card_dirty(card_index)) {
201 scanCard(card_index, card_region);
202 }
203 }
204 if (!_try_claimed) {
205 hrrs->set_iter_complete();
206 }
207 return false;
208 }
209 size_t cards_done() { return _cards_done;}
210 size_t cards_looked_up() { return _cards;}
211 };
212
213 void G1RemSet::scanRS(OopsInHeapRegionClosure* oc, int worker_i) {
214 double rs_time_start = os::elapsedTime();
215 HeapRegion *startRegion = _g1->start_cset_region_for_worker(worker_i);
216
217 ScanRSClosure scanRScl(oc, worker_i);
218
219 _g1->collection_set_iterate_from(startRegion, &scanRScl);
220 scanRScl.set_try_claimed();
221 _g1->collection_set_iterate_from(startRegion, &scanRScl);
222
223 double scan_rs_time_sec = os::elapsedTime() - rs_time_start;
224
225 assert( _cards_scanned != NULL, "invariant" );
226 _cards_scanned[worker_i] = scanRScl.cards_done();
227
228 _g1p->phase_times()->record_scan_rs_time(worker_i, scan_rs_time_sec * 1000.0);
229 }
230
231 // Closure used for updating RSets and recording references that
232 // point into the collection set. Only called during an
233 // evacuation pause.
234
235 class RefineRecordRefsIntoCSCardTableEntryClosure: public CardTableEntryClosure {
236 G1RemSet* _g1rs;
237 DirtyCardQueue* _into_cset_dcq;
238 public:
239 RefineRecordRefsIntoCSCardTableEntryClosure(G1CollectedHeap* g1h,
240 DirtyCardQueue* into_cset_dcq) :
241 _g1rs(g1h->g1_rem_set()), _into_cset_dcq(into_cset_dcq)
242 {}
243 bool do_card_ptr(jbyte* card_ptr, int worker_i) {
244 // The only time we care about recording cards that
245 // contain references that point into the collection set
246 // is during RSet updating within an evacuation pause.
247 // In this case worker_i should be the id of a GC worker thread.
248 assert(SafepointSynchronize::is_at_safepoint(), "not during an evacuation pause");
249 assert(worker_i < (int) (ParallelGCThreads == 0 ? 1 : ParallelGCThreads), "should be a GC worker");
250
251 if (_g1rs->concurrentRefineOneCard(card_ptr, worker_i, true)) {
252 // 'card_ptr' contains references that point into the collection
253 // set. We need to record the card in the DCQS
254 // (G1CollectedHeap::into_cset_dirty_card_queue_set())
255 // that's used for that purpose.
256 //
257 // Enqueue the card
258 _into_cset_dcq->enqueue(card_ptr);
259 }
260 return true;
261 }
262 };
263
264 void G1RemSet::updateRS(DirtyCardQueue* into_cset_dcq, int worker_i) {
265 double start = os::elapsedTime();
266 // Apply the given closure to all remaining log entries.
267 RefineRecordRefsIntoCSCardTableEntryClosure into_cset_update_rs_cl(_g1, into_cset_dcq);
268
269 _g1->iterate_dirty_card_closure(&into_cset_update_rs_cl, into_cset_dcq, false, worker_i);
270
271 // Now there should be no dirty cards.
272 if (G1RSLogCheckCardTable) {
273 CountNonCleanMemRegionClosure cl(_g1);
274 _ct_bs->mod_card_iterate(&cl);
275 // XXX This isn't true any more: keeping cards of young regions
276 // marked dirty broke it. Need some reasonable fix.
277 guarantee(cl.n() == 0, "Card table should be clean.");
278 }
279
280 _g1p->phase_times()->record_update_rs_time(worker_i, (os::elapsedTime() - start) * 1000.0);
281 }
282
283 class CountRSSizeClosure: public HeapRegionClosure {
284 size_t _n;
285 size_t _tot;
286 size_t _max;
287 HeapRegion* _max_r;
288 enum {
289 N = 20,
290 MIN = 6
291 };
292 int _histo[N];
293 public:
294 CountRSSizeClosure() : _n(0), _tot(0), _max(0), _max_r(NULL) {
295 for (int i = 0; i < N; i++) _histo[i] = 0;
296 }
297 bool doHeapRegion(HeapRegion* r) {
298 if (!r->continuesHumongous()) {
299 size_t occ = r->rem_set()->occupied();
300 _n++;
301 _tot += occ;
302 if (occ > _max) {
303 _max = occ;
304 _max_r = r;
305 }
306 // Fit it into a histo bin.
307 int s = 1 << MIN;
308 int i = 0;
309 while (occ > (size_t) s && i < (N-1)) {
310 s = s << 1;
311 i++;
312 }
313 _histo[i]++;
314 }
315 return false;
316 }
317 size_t n() { return _n; }
318 size_t tot() { return _tot; }
319 size_t mx() { return _max; }
320 HeapRegion* mxr() { return _max_r; }
321 void print_histo() {
322 int mx = N;
323 while (mx >= 0) {
324 if (_histo[mx-1] > 0) break;
325 mx--;
326 }
327 gclog_or_tty->print_cr("Number of regions with given RS sizes:");
328 gclog_or_tty->print_cr(" <= %8d %8d", 1 << MIN, _histo[0]);
329 for (int i = 1; i < mx-1; i++) {
330 gclog_or_tty->print_cr(" %8d - %8d %8d",
331 (1 << (MIN + i - 1)) + 1,
332 1 << (MIN + i),
333 _histo[i]);
334 }
335 gclog_or_tty->print_cr(" > %8d %8d", (1 << (MIN+mx-2))+1, _histo[mx-1]);
336 }
337 };
338
339 void G1RemSet::cleanupHRRS() {
340 HeapRegionRemSet::cleanup();
341 }
342
343 void G1RemSet::oops_into_collection_set_do(OopsInHeapRegionClosure* oc,
344 int worker_i) {
345 #if CARD_REPEAT_HISTO
346 ct_freq_update_histo_and_reset();
347 #endif
348 if (worker_i == 0) {
349 _cg1r->clear_and_record_card_counts();
350 }
351
352 // Make this into a command-line flag...
353 if (G1RSCountHisto && (ParallelGCThreads == 0 || worker_i == 0)) {
354 CountRSSizeClosure count_cl;
355 _g1->heap_region_iterate(&count_cl);
356 gclog_or_tty->print_cr("Avg of %d RS counts is %f, max is %d, "
357 "max region is " PTR_FORMAT,
358 count_cl.n(), (float)count_cl.tot()/(float)count_cl.n(),
359 count_cl.mx(), count_cl.mxr());
360 count_cl.print_histo();
361 }
362
363 // We cache the value of 'oc' closure into the appropriate slot in the
364 // _cset_rs_update_cl for this worker
365 assert(worker_i < (int)n_workers(), "sanity");
366 _cset_rs_update_cl[worker_i] = oc;
367
368 // A DirtyCardQueue that is used to hold cards containing references
369 // that point into the collection set. This DCQ is associated with a
370 // special DirtyCardQueueSet (see g1CollectedHeap.hpp). Under normal
371 // circumstances (i.e. the pause successfully completes), these cards
372 // are just discarded (there's no need to update the RSets of regions
373 // that were in the collection set - after the pause these regions
374 // are wholly 'free' of live objects. In the event of an evacuation
375 // failure the cards/buffers in this queue set are:
376 // * passed to the DirtyCardQueueSet that is used to manage deferred
377 // RSet updates, or
378 // * scanned for references that point into the collection set
379 // and the RSet of the corresponding region in the collection set
380 // is updated immediately.
381 DirtyCardQueue into_cset_dcq(&_g1->into_cset_dirty_card_queue_set());
382
383 assert((ParallelGCThreads > 0) || worker_i == 0, "invariant");
384
385 // The two flags below were introduced temporarily to serialize
386 // the updating and scanning of remembered sets. There are some
387 // race conditions when these two operations are done in parallel
388 // and they are causing failures. When we resolve said race
389 // conditions, we'll revert back to parallel remembered set
390 // updating and scanning. See CRs 6677707 and 6677708.
391 if (G1UseParallelRSetUpdating || (worker_i == 0)) {
392 updateRS(&into_cset_dcq, worker_i);
393 } else {
394 _g1p->phase_times()->record_update_rs_processed_buffers(worker_i, 0);
395 _g1p->phase_times()->record_update_rs_time(worker_i, 0.0);
396 }
397 if (G1UseParallelRSetScanning || (worker_i == 0)) {
398 scanRS(oc, worker_i);
399 } else {
400 _g1p->phase_times()->record_scan_rs_time(worker_i, 0.0);
401 }
402
403 // We now clear the cached values of _cset_rs_update_cl for this worker
404 _cset_rs_update_cl[worker_i] = NULL;
405 }
406
407 void G1RemSet::prepare_for_oops_into_collection_set_do() {
408 cleanupHRRS();
409 ConcurrentG1Refine* cg1r = _g1->concurrent_g1_refine();
410 _g1->set_refine_cte_cl_concurrency(false);
411 DirtyCardQueueSet& dcqs = JavaThread::dirty_card_queue_set();
412 dcqs.concatenate_logs();
413
414 if (G1CollectedHeap::use_parallel_gc_threads()) {
415 // Don't set the number of workers here. It will be set
416 // when the task is run
417 // _seq_task->set_n_termination((int)n_workers());
418 }
419 guarantee( _cards_scanned == NULL, "invariant" );
420 _cards_scanned = NEW_C_HEAP_ARRAY(size_t, n_workers(), mtGC);
421 for (uint i = 0; i < n_workers(); ++i) {
422 _cards_scanned[i] = 0;
423 }
424 _total_cards_scanned = 0;
425 }
426
427
428 // This closure, applied to a DirtyCardQueueSet, is used to immediately
429 // update the RSets for the regions in the CSet. For each card it iterates
430 // through the oops which coincide with that card. It scans the reference
431 // fields in each oop; when it finds an oop that points into the collection
432 // set, the RSet for the region containing the referenced object is updated.
433 class UpdateRSetCardTableEntryIntoCSetClosure: public CardTableEntryClosure {
434 G1CollectedHeap* _g1;
435 CardTableModRefBS* _ct_bs;
436 public:
437 UpdateRSetCardTableEntryIntoCSetClosure(G1CollectedHeap* g1,
438 CardTableModRefBS* bs):
439 _g1(g1), _ct_bs(bs)
440 { }
441
442 bool do_card_ptr(jbyte* card_ptr, int worker_i) {
443 // Construct the region representing the card.
444 HeapWord* start = _ct_bs->addr_for(card_ptr);
445 // And find the region containing it.
446 HeapRegion* r = _g1->heap_region_containing(start);
447 assert(r != NULL, "unexpected null");
448
449 // Scan oops in the card looking for references into the collection set
450 HeapWord* end = _ct_bs->addr_for(card_ptr + 1);
451 MemRegion scanRegion(start, end);
452
453 UpdateRSetImmediate update_rs_cl(_g1->g1_rem_set());
454 FilterIntoCSClosure update_rs_cset_oop_cl(NULL, _g1, &update_rs_cl);
455 FilterOutOfRegionClosure filter_then_update_rs_cset_oop_cl(r, &update_rs_cset_oop_cl);
456
457 // We can pass false as the "filter_young" parameter here as:
458 // * we should be in a STW pause,
459 // * the DCQS to which this closure is applied is used to hold
460 // references that point into the collection set from the prior
461 // RSet updating,
462 // * the post-write barrier shouldn't be logging updates to young
463 // regions (but there is a situation where this can happen - see
464 // the comment in G1RemSet::concurrentRefineOneCard below -
465 // that should not be applicable here), and
466 // * during actual RSet updating, the filtering of cards in young
467 // regions in HeapRegion::oops_on_card_seq_iterate_careful is
468 // employed.
469 // As a result, when this closure is applied to "refs into cset"
470 // DCQS, we shouldn't see any cards in young regions.
471 update_rs_cl.set_region(r);
472 HeapWord* stop_point =
473 r->oops_on_card_seq_iterate_careful(scanRegion,
474 &filter_then_update_rs_cset_oop_cl,
475 false /* filter_young */,
476 NULL /* card_ptr */);
477
478 // Since this is performed in the event of an evacuation failure, we
479 // we shouldn't see a non-null stop point
480 assert(stop_point == NULL, "saw an unallocated region");
481 return true;
482 }
483 };
484
485 void G1RemSet::cleanup_after_oops_into_collection_set_do() {
486 guarantee( _cards_scanned != NULL, "invariant" );
487 _total_cards_scanned = 0;
488 for (uint i = 0; i < n_workers(); ++i) {
489 _total_cards_scanned += _cards_scanned[i];
490 }
491 FREE_C_HEAP_ARRAY(size_t, _cards_scanned, mtGC);
492 _cards_scanned = NULL;
493 // Cleanup after copy
494 _g1->set_refine_cte_cl_concurrency(true);
495 // Set all cards back to clean.
496 _g1->cleanUpCardTable();
497
498 DirtyCardQueueSet& into_cset_dcqs = _g1->into_cset_dirty_card_queue_set();
499 int into_cset_n_buffers = into_cset_dcqs.completed_buffers_num();
500
501 if (_g1->evacuation_failed()) {
502 // Restore remembered sets for the regions pointing into the collection set.
503
504 if (G1DeferredRSUpdate) {
505 // If deferred RS updates are enabled then we just need to transfer
506 // the completed buffers from (a) the DirtyCardQueueSet used to hold
507 // cards that contain references that point into the collection set
508 // to (b) the DCQS used to hold the deferred RS updates
509 _g1->dirty_card_queue_set().merge_bufferlists(&into_cset_dcqs);
510 } else {
511
512 CardTableModRefBS* bs = (CardTableModRefBS*)_g1->barrier_set();
513 UpdateRSetCardTableEntryIntoCSetClosure update_rs_cset_immediate(_g1, bs);
514
515 int n_completed_buffers = 0;
516 while (into_cset_dcqs.apply_closure_to_completed_buffer(&update_rs_cset_immediate,
517 0, 0, true)) {
518 n_completed_buffers++;
519 }
520 assert(n_completed_buffers == into_cset_n_buffers, "missed some buffers");
521 }
522 }
523
524 // Free any completed buffers in the DirtyCardQueueSet used to hold cards
525 // which contain references that point into the collection.
526 _g1->into_cset_dirty_card_queue_set().clear();
527 assert(_g1->into_cset_dirty_card_queue_set().completed_buffers_num() == 0,
528 "all buffers should be freed");
529 _g1->into_cset_dirty_card_queue_set().clear_n_completed_buffers();
530 }
531
532 class ScrubRSClosure: public HeapRegionClosure {
533 G1CollectedHeap* _g1h;
534 BitMap* _region_bm;
535 BitMap* _card_bm;
536 CardTableModRefBS* _ctbs;
537 public:
538 ScrubRSClosure(BitMap* region_bm, BitMap* card_bm) :
539 _g1h(G1CollectedHeap::heap()),
540 _region_bm(region_bm), _card_bm(card_bm),
541 _ctbs(NULL)
542 {
543 ModRefBarrierSet* bs = _g1h->mr_bs();
544 guarantee(bs->is_a(BarrierSet::CardTableModRef), "Precondition");
545 _ctbs = (CardTableModRefBS*)bs;
546 }
547
548 bool doHeapRegion(HeapRegion* r) {
549 if (!r->continuesHumongous()) {
550 r->rem_set()->scrub(_ctbs, _region_bm, _card_bm);
551 }
552 return false;
553 }
554 };
555
556 void G1RemSet::scrub(BitMap* region_bm, BitMap* card_bm) {
557 ScrubRSClosure scrub_cl(region_bm, card_bm);
558 _g1->heap_region_iterate(&scrub_cl);
559 }
560
561 void G1RemSet::scrub_par(BitMap* region_bm, BitMap* card_bm,
562 uint worker_num, int claim_val) {
563 ScrubRSClosure scrub_cl(region_bm, card_bm);
564 _g1->heap_region_par_iterate_chunked(&scrub_cl,
565 worker_num,
566 n_workers(),
567 claim_val);
568 }
569
570
571 static IntHistogram out_of_histo(50, 50);
572
573
574 G1TriggerClosure::G1TriggerClosure() :
575 _triggered(false) { }
576
577 G1InvokeIfNotTriggeredClosure::G1InvokeIfNotTriggeredClosure(G1TriggerClosure* t_cl,
578 OopClosure* oop_cl) :
579 _trigger_cl(t_cl), _oop_cl(oop_cl) { }
580
581 G1Mux2Closure::G1Mux2Closure(OopClosure *c1, OopClosure *c2) :
582 _c1(c1), _c2(c2) { }
583
584 G1UpdateRSOrPushRefOopClosure::
585 G1UpdateRSOrPushRefOopClosure(G1CollectedHeap* g1h,
586 G1RemSet* rs,
587 OopsInHeapRegionClosure* push_ref_cl,
588 bool record_refs_into_cset,
589 int worker_i) :
590 _g1(g1h), _g1_rem_set(rs), _from(NULL),
591 _record_refs_into_cset(record_refs_into_cset),
592 _push_ref_cl(push_ref_cl), _worker_i(worker_i) { }
593
594 bool G1RemSet::concurrentRefineOneCard_impl(jbyte* card_ptr, int worker_i,
595 bool check_for_refs_into_cset) {
596 // Construct the region representing the card.
597 HeapWord* start = _ct_bs->addr_for(card_ptr);
598 // And find the region containing it.
599 HeapRegion* r = _g1->heap_region_containing(start);
600 assert(r != NULL, "unexpected null");
601
602 HeapWord* end = _ct_bs->addr_for(card_ptr + 1);
603 MemRegion dirtyRegion(start, end);
604
605 #if CARD_REPEAT_HISTO
606 init_ct_freq_table(_g1->max_capacity());
607 ct_freq_note_card(_ct_bs->index_for(start));
608 #endif
609
610 OopsInHeapRegionClosure* oops_in_heap_closure = NULL;
611 if (check_for_refs_into_cset) {
612 // ConcurrentG1RefineThreads have worker numbers larger than what
613 // _cset_rs_update_cl[] is set up to handle. But those threads should
614 // only be active outside of a collection which means that when they
615 // reach here they should have check_for_refs_into_cset == false.
616 assert((size_t)worker_i < n_workers(), "index of worker larger than _cset_rs_update_cl[].length");
617 oops_in_heap_closure = _cset_rs_update_cl[worker_i];
618 }
619 G1UpdateRSOrPushRefOopClosure update_rs_oop_cl(_g1,
620 _g1->g1_rem_set(),
621 oops_in_heap_closure,
622 check_for_refs_into_cset,
623 worker_i);
624 update_rs_oop_cl.set_from(r);
625
626 G1TriggerClosure trigger_cl;
627 FilterIntoCSClosure into_cs_cl(NULL, _g1, &trigger_cl);
628 G1InvokeIfNotTriggeredClosure invoke_cl(&trigger_cl, &into_cs_cl);
629 G1Mux2Closure mux(&invoke_cl, &update_rs_oop_cl);
630
631 FilterOutOfRegionClosure filter_then_update_rs_oop_cl(r,
632 (check_for_refs_into_cset ?
633 (OopClosure*)&mux :
634 (OopClosure*)&update_rs_oop_cl));
635
636 // The region for the current card may be a young region. The
637 // current card may have been a card that was evicted from the
638 // card cache. When the card was inserted into the cache, we had
639 // determined that its region was non-young. While in the cache,
640 // the region may have been freed during a cleanup pause, reallocated
641 // and tagged as young.
642 //
643 // We wish to filter out cards for such a region but the current
644 // thread, if we're running concurrently, may "see" the young type
645 // change at any time (so an earlier "is_young" check may pass or
646 // fail arbitrarily). We tell the iteration code to perform this
647 // filtering when it has been determined that there has been an actual
648 // allocation in this region and making it safe to check the young type.
649 bool filter_young = true;
650
651 HeapWord* stop_point =
652 r->oops_on_card_seq_iterate_careful(dirtyRegion,
653 &filter_then_update_rs_oop_cl,
654 filter_young,
655 card_ptr);
656
657 // If stop_point is non-null, then we encountered an unallocated region
658 // (perhaps the unfilled portion of a TLAB.) For now, we'll dirty the
659 // card and re-enqueue: if we put off the card until a GC pause, then the
660 // unallocated portion will be filled in. Alternatively, we might try
661 // the full complexity of the technique used in "regular" precleaning.
662 if (stop_point != NULL) {
663 // The card might have gotten re-dirtied and re-enqueued while we
664 // worked. (In fact, it's pretty likely.)
665 if (*card_ptr != CardTableModRefBS::dirty_card_val()) {
666 *card_ptr = CardTableModRefBS::dirty_card_val();
667 MutexLockerEx x(Shared_DirtyCardQ_lock,
668 Mutex::_no_safepoint_check_flag);
669 DirtyCardQueue* sdcq =
670 JavaThread::dirty_card_queue_set().shared_dirty_card_queue();
671 sdcq->enqueue(card_ptr);
672 }
673 } else {
674 out_of_histo.add_entry(filter_then_update_rs_oop_cl.out_of_region());
675 _conc_refine_cards++;
676 }
677
678 return trigger_cl.triggered();
679 }
680
681 bool G1RemSet::concurrentRefineOneCard(jbyte* card_ptr, int worker_i,
682 bool check_for_refs_into_cset) {
683 // If the card is no longer dirty, nothing to do.
684 if (*card_ptr != CardTableModRefBS::dirty_card_val()) {
685 // No need to return that this card contains refs that point
686 // into the collection set.
687 return false;
688 }
689
690 // Construct the region representing the card.
691 HeapWord* start = _ct_bs->addr_for(card_ptr);
692 // And find the region containing it.
693 HeapRegion* r = _g1->heap_region_containing(start);
694 if (r == NULL) {
695 guarantee(_g1->is_in_permanent(start), "Or else where?");
696 // Again no need to return that this card contains refs that
697 // point into the collection set.
698 return false; // Not in the G1 heap (might be in perm, for example.)
699 }
700 // Why do we have to check here whether a card is on a young region,
701 // given that we dirty young regions and, as a result, the
702 // post-barrier is supposed to filter them out and never to enqueue
703 // them? When we allocate a new region as the "allocation region" we
704 // actually dirty its cards after we release the lock, since card
705 // dirtying while holding the lock was a performance bottleneck. So,
706 // as a result, it is possible for other threads to actually
707 // allocate objects in the region (after the acquire the lock)
708 // before all the cards on the region are dirtied. This is unlikely,
709 // and it doesn't happen often, but it can happen. So, the extra
710 // check below filters out those cards.
711 if (r->is_young()) {
712 return false;
713 }
714 // While we are processing RSet buffers during the collection, we
715 // actually don't want to scan any cards on the collection set,
716 // since we don't want to update remebered sets with entries that
717 // point into the collection set, given that live objects from the
718 // collection set are about to move and such entries will be stale
719 // very soon. This change also deals with a reliability issue which
720 // involves scanning a card in the collection set and coming across
721 // an array that was being chunked and looking malformed. Note,
722 // however, that if evacuation fails, we have to scan any objects
723 // that were not moved and create any missing entries.
724 if (r->in_collection_set()) {
725 return false;
726 }
727
728 // Should we defer processing the card?
729 //
730 // Previously the result from the insert_cache call would be
731 // either card_ptr (implying that card_ptr was currently "cold"),
732 // null (meaning we had inserted the card ptr into the "hot"
733 // cache, which had some headroom), or a "hot" card ptr
734 // extracted from the "hot" cache.
735 //
736 // Now that the _card_counts cache in the ConcurrentG1Refine
737 // instance is an evicting hash table, the result we get back
738 // could be from evicting the card ptr in an already occupied
739 // bucket (in which case we have replaced the card ptr in the
740 // bucket with card_ptr and "defer" is set to false). To avoid
741 // having a data structure (updates to which would need a lock)
742 // to hold these unprocessed dirty cards, we need to immediately
743 // process card_ptr. The actions needed to be taken on return
744 // from cache_insert are summarized in the following table:
745 //
746 // res defer action
747 // --------------------------------------------------------------
748 // null false card evicted from _card_counts & replaced with
749 // card_ptr; evicted ptr added to hot cache.
750 // No need to process res; immediately process card_ptr
751 //
752 // null true card not evicted from _card_counts; card_ptr added
753 // to hot cache.
754 // Nothing to do.
755 //
756 // non-null false card evicted from _card_counts & replaced with
757 // card_ptr; evicted ptr is currently "cold" or
758 // caused an eviction from the hot cache.
759 // Immediately process res; process card_ptr.
760 //
761 // non-null true card not evicted from _card_counts; card_ptr is
762 // currently cold, or caused an eviction from hot
763 // cache.
764 // Immediately process res; no need to process card_ptr.
765
766
767 jbyte* res = card_ptr;
768 bool defer = false;
769
770 // This gets set to true if the card being refined has references
771 // that point into the collection set.
772 bool oops_into_cset = false;
773
774 if (_cg1r->use_cache()) {
775 jbyte* res = _cg1r->cache_insert(card_ptr, &defer);
776 if (res != NULL && (res != card_ptr || defer)) {
777 start = _ct_bs->addr_for(res);
778 r = _g1->heap_region_containing(start);
779 if (r == NULL) {
780 assert(_g1->is_in_permanent(start), "Or else where?");
781 } else {
782 // Checking whether the region we got back from the cache
783 // is young here is inappropriate. The region could have been
784 // freed, reallocated and tagged as young while in the cache.
785 // Hence we could see its young type change at any time.
786 //
787 // Process card pointer we get back from the hot card cache. This
788 // will check whether the region containing the card is young
789 // _after_ checking that the region has been allocated from.
790 oops_into_cset = concurrentRefineOneCard_impl(res, worker_i,
791 false /* check_for_refs_into_cset */);
792 // The above call to concurrentRefineOneCard_impl is only
793 // performed if the hot card cache is enabled. This cache is
794 // disabled during an evacuation pause - which is the only
795 // time when we need know if the card contains references
796 // that point into the collection set. Also when the hot card
797 // cache is enabled, this code is executed by the concurrent
798 // refine threads - rather than the GC worker threads - and
799 // concurrentRefineOneCard_impl will return false.
800 assert(!oops_into_cset, "should not see true here");
801 }
802 }
803 }
804
805 if (!defer) {
806 oops_into_cset =
807 concurrentRefineOneCard_impl(card_ptr, worker_i, check_for_refs_into_cset);
808 // We should only be detecting that the card contains references
809 // that point into the collection set if the current thread is
810 // a GC worker thread.
811 assert(!oops_into_cset || SafepointSynchronize::is_at_safepoint(),
812 "invalid result at non safepoint");
813 }
814 return oops_into_cset;
815 }
816
817 class HRRSStatsIter: public HeapRegionClosure {
818 size_t _occupied;
819 size_t _total_mem_sz;
820 size_t _max_mem_sz;
821 HeapRegion* _max_mem_sz_region;
822 public:
823 HRRSStatsIter() :
824 _occupied(0),
825 _total_mem_sz(0),
826 _max_mem_sz(0),
827 _max_mem_sz_region(NULL)
828 {}
829
830 bool doHeapRegion(HeapRegion* r) {
831 if (r->continuesHumongous()) return false;
832 size_t mem_sz = r->rem_set()->mem_size();
833 if (mem_sz > _max_mem_sz) {
834 _max_mem_sz = mem_sz;
835 _max_mem_sz_region = r;
836 }
837 _total_mem_sz += mem_sz;
838 size_t occ = r->rem_set()->occupied();
839 _occupied += occ;
840 return false;
841 }
842 size_t total_mem_sz() { return _total_mem_sz; }
843 size_t max_mem_sz() { return _max_mem_sz; }
844 size_t occupied() { return _occupied; }
845 HeapRegion* max_mem_sz_region() { return _max_mem_sz_region; }
846 };
847
848 class PrintRSThreadVTimeClosure : public ThreadClosure {
849 public:
850 virtual void do_thread(Thread *t) {
851 ConcurrentG1RefineThread* crt = (ConcurrentG1RefineThread*) t;
852 gclog_or_tty->print(" %5.2f", crt->vtime_accum());
853 }
854 };
855
856 void G1RemSet::print_summary_info() {
857 G1CollectedHeap* g1 = G1CollectedHeap::heap();
858
859 #if CARD_REPEAT_HISTO
860 gclog_or_tty->print_cr("\nG1 card_repeat count histogram: ");
861 gclog_or_tty->print_cr(" # of repeats --> # of cards with that number.");
862 card_repeat_count.print_on(gclog_or_tty);
863 #endif
864
865 if (FILTEROUTOFREGIONCLOSURE_DOHISTOGRAMCOUNT) {
866 gclog_or_tty->print_cr("\nG1 rem-set out-of-region histogram: ");
867 gclog_or_tty->print_cr(" # of CS ptrs --> # of cards with that number.");
868 out_of_histo.print_on(gclog_or_tty);
869 }
870 gclog_or_tty->print_cr("\n Concurrent RS processed %d cards",
871 _conc_refine_cards);
872 DirtyCardQueueSet& dcqs = JavaThread::dirty_card_queue_set();
873 jint tot_processed_buffers =
874 dcqs.processed_buffers_mut() + dcqs.processed_buffers_rs_thread();
875 gclog_or_tty->print_cr(" Of %d completed buffers:", tot_processed_buffers);
876 gclog_or_tty->print_cr(" %8d (%5.1f%%) by conc RS threads.",
877 dcqs.processed_buffers_rs_thread(),
878 100.0*(float)dcqs.processed_buffers_rs_thread()/
879 (float)tot_processed_buffers);
880 gclog_or_tty->print_cr(" %8d (%5.1f%%) by mutator threads.",
881 dcqs.processed_buffers_mut(),
882 100.0*(float)dcqs.processed_buffers_mut()/
883 (float)tot_processed_buffers);
884 gclog_or_tty->print_cr(" Conc RS threads times(s)");
885 PrintRSThreadVTimeClosure p;
886 gclog_or_tty->print(" ");
887 g1->concurrent_g1_refine()->threads_do(&p);
888 gclog_or_tty->print_cr("");
889
890 HRRSStatsIter blk;
891 g1->heap_region_iterate(&blk);
892 gclog_or_tty->print_cr(" Total heap region rem set sizes = " SIZE_FORMAT "K."
893 " Max = " SIZE_FORMAT "K.",
894 blk.total_mem_sz()/K, blk.max_mem_sz()/K);
895 gclog_or_tty->print_cr(" Static structures = " SIZE_FORMAT "K,"
896 " free_lists = " SIZE_FORMAT "K.",
897 HeapRegionRemSet::static_mem_size()/K,
898 HeapRegionRemSet::fl_mem_size()/K);
899 gclog_or_tty->print_cr(" %d occupied cards represented.",
900 blk.occupied());
901 gclog_or_tty->print_cr(" Max sz region = [" PTR_FORMAT ", " PTR_FORMAT " )"
902 ", cap = " SIZE_FORMAT "K, occ = " SIZE_FORMAT "K.",
903 blk.max_mem_sz_region()->bottom(), blk.max_mem_sz_region()->end(),
904 (blk.max_mem_sz_region()->rem_set()->mem_size() + K - 1)/K,
905 (blk.max_mem_sz_region()->rem_set()->occupied() + K - 1)/K);
906 gclog_or_tty->print_cr(" Did %d coarsenings.", HeapRegionRemSet::n_coarsenings());
907 }
908
909 void G1RemSet::prepare_for_verify() {
910 if (G1HRRSFlushLogBuffersOnVerify &&
911 (VerifyBeforeGC || VerifyAfterGC)
912 && !_g1->full_collection()) {
913 cleanupHRRS();
914 _g1->set_refine_cte_cl_concurrency(false);
915 if (SafepointSynchronize::is_at_safepoint()) {
916 DirtyCardQueueSet& dcqs = JavaThread::dirty_card_queue_set();
917 dcqs.concatenate_logs();
918 }
919 bool cg1r_use_cache = _cg1r->use_cache();
920 _cg1r->set_use_cache(false);
921 DirtyCardQueue into_cset_dcq(&_g1->into_cset_dirty_card_queue_set());
922 updateRS(&into_cset_dcq, 0);
923 _g1->into_cset_dirty_card_queue_set().clear();
924 _cg1r->set_use_cache(cg1r_use_cache);
925
926 assert(JavaThread::dirty_card_queue_set().completed_buffers_num() == 0, "All should be consumed");
927 }
928 }