Mon, 29 Apr 2013 09:31:59 +0200
Merge
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 */
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"
40 #define CARD_REPEAT_HISTO 0
42 #if CARD_REPEAT_HISTO
43 static size_t ct_freq_sz;
44 static jbyte* ct_freq = NULL;
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 }
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 }
59 static IntHistogram card_repeat_count(10, 10);
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 }
67 }
68 #endif
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 }
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 }
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 }
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 }
123 void set_try_claimed() { _try_claimed = true; }
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);
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 }
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 }
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(hrrs);
173 size_t card_index;
175 // We claim cards in block so as to recude the contention. The block size is determined by
176 // the G1RSetScanBlockSize parameter.
177 size_t jump_to_card = hrrs->iter_claimed_next(_block_size);
178 for (size_t current_card = 0; iter.has_next(card_index); current_card++) {
179 if (current_card >= jump_to_card + _block_size) {
180 jump_to_card = hrrs->iter_claimed_next(_block_size);
181 }
182 if (current_card < jump_to_card) continue;
183 HeapWord* card_start = _g1h->bot_shared()->address_for_index(card_index);
184 #if 0
185 gclog_or_tty->print("Rem set iteration yielded card [" PTR_FORMAT ", " PTR_FORMAT ").\n",
186 card_start, card_start + CardTableModRefBS::card_size_in_words);
187 #endif
189 HeapRegion* card_region = _g1h->heap_region_containing(card_start);
190 assert(card_region != NULL, "Yielding cards not in the heap?");
191 _cards++;
193 if (!card_region->is_on_dirty_cards_region_list()) {
194 _g1h->push_dirty_cards_region(card_region);
195 }
197 // If the card is dirty, then we will scan it during updateRS.
198 if (!card_region->in_collection_set() &&
199 !_ct_bs->is_card_dirty(card_index)) {
200 scanCard(card_index, card_region);
201 }
202 }
203 if (!_try_claimed) {
204 hrrs->set_iter_complete();
205 }
206 return false;
207 }
208 size_t cards_done() { return _cards_done;}
209 size_t cards_looked_up() { return _cards;}
210 };
212 void G1RemSet::scanRS(OopsInHeapRegionClosure* oc, int worker_i) {
213 double rs_time_start = os::elapsedTime();
214 HeapRegion *startRegion = _g1->start_cset_region_for_worker(worker_i);
216 ScanRSClosure scanRScl(oc, worker_i);
218 _g1->collection_set_iterate_from(startRegion, &scanRScl);
219 scanRScl.set_try_claimed();
220 _g1->collection_set_iterate_from(startRegion, &scanRScl);
222 double scan_rs_time_sec = os::elapsedTime() - rs_time_start;
224 assert( _cards_scanned != NULL, "invariant" );
225 _cards_scanned[worker_i] = scanRScl.cards_done();
227 _g1p->phase_times()->record_scan_rs_time(worker_i, scan_rs_time_sec * 1000.0);
228 }
230 // Closure used for updating RSets and recording references that
231 // point into the collection set. Only called during an
232 // evacuation pause.
234 class RefineRecordRefsIntoCSCardTableEntryClosure: public CardTableEntryClosure {
235 G1RemSet* _g1rs;
236 DirtyCardQueue* _into_cset_dcq;
237 public:
238 RefineRecordRefsIntoCSCardTableEntryClosure(G1CollectedHeap* g1h,
239 DirtyCardQueue* into_cset_dcq) :
240 _g1rs(g1h->g1_rem_set()), _into_cset_dcq(into_cset_dcq)
241 {}
242 bool do_card_ptr(jbyte* card_ptr, int worker_i) {
243 // The only time we care about recording cards that
244 // contain references that point into the collection set
245 // is during RSet updating within an evacuation pause.
246 // In this case worker_i should be the id of a GC worker thread.
247 assert(SafepointSynchronize::is_at_safepoint(), "not during an evacuation pause");
248 assert(worker_i < (int) (ParallelGCThreads == 0 ? 1 : ParallelGCThreads), "should be a GC worker");
250 if (_g1rs->concurrentRefineOneCard(card_ptr, worker_i, true)) {
251 // 'card_ptr' contains references that point into the collection
252 // set. We need to record the card in the DCQS
253 // (G1CollectedHeap::into_cset_dirty_card_queue_set())
254 // that's used for that purpose.
255 //
256 // Enqueue the card
257 _into_cset_dcq->enqueue(card_ptr);
258 }
259 return true;
260 }
261 };
263 void G1RemSet::updateRS(DirtyCardQueue* into_cset_dcq, int worker_i) {
264 double start = os::elapsedTime();
265 // Apply the given closure to all remaining log entries.
266 RefineRecordRefsIntoCSCardTableEntryClosure into_cset_update_rs_cl(_g1, into_cset_dcq);
268 _g1->iterate_dirty_card_closure(&into_cset_update_rs_cl, into_cset_dcq, false, worker_i);
270 // Now there should be no dirty cards.
271 if (G1RSLogCheckCardTable) {
272 CountNonCleanMemRegionClosure cl(_g1);
273 _ct_bs->mod_card_iterate(&cl);
274 // XXX This isn't true any more: keeping cards of young regions
275 // marked dirty broke it. Need some reasonable fix.
276 guarantee(cl.n() == 0, "Card table should be clean.");
277 }
279 _g1p->phase_times()->record_update_rs_time(worker_i, (os::elapsedTime() - start) * 1000.0);
280 }
282 void G1RemSet::cleanupHRRS() {
283 HeapRegionRemSet::cleanup();
284 }
286 void G1RemSet::oops_into_collection_set_do(OopsInHeapRegionClosure* oc,
287 int worker_i) {
288 #if CARD_REPEAT_HISTO
289 ct_freq_update_histo_and_reset();
290 #endif
291 if (worker_i == 0) {
292 _cg1r->clear_and_record_card_counts();
293 }
295 // We cache the value of 'oc' closure into the appropriate slot in the
296 // _cset_rs_update_cl for this worker
297 assert(worker_i < (int)n_workers(), "sanity");
298 _cset_rs_update_cl[worker_i] = oc;
300 // A DirtyCardQueue that is used to hold cards containing references
301 // that point into the collection set. This DCQ is associated with a
302 // special DirtyCardQueueSet (see g1CollectedHeap.hpp). Under normal
303 // circumstances (i.e. the pause successfully completes), these cards
304 // are just discarded (there's no need to update the RSets of regions
305 // that were in the collection set - after the pause these regions
306 // are wholly 'free' of live objects. In the event of an evacuation
307 // failure the cards/buffers in this queue set are:
308 // * passed to the DirtyCardQueueSet that is used to manage deferred
309 // RSet updates, or
310 // * scanned for references that point into the collection set
311 // and the RSet of the corresponding region in the collection set
312 // is updated immediately.
313 DirtyCardQueue into_cset_dcq(&_g1->into_cset_dirty_card_queue_set());
315 assert((ParallelGCThreads > 0) || worker_i == 0, "invariant");
317 // The two flags below were introduced temporarily to serialize
318 // the updating and scanning of remembered sets. There are some
319 // race conditions when these two operations are done in parallel
320 // and they are causing failures. When we resolve said race
321 // conditions, we'll revert back to parallel remembered set
322 // updating and scanning. See CRs 6677707 and 6677708.
323 if (G1UseParallelRSetUpdating || (worker_i == 0)) {
324 updateRS(&into_cset_dcq, worker_i);
325 } else {
326 _g1p->phase_times()->record_update_rs_processed_buffers(worker_i, 0);
327 _g1p->phase_times()->record_update_rs_time(worker_i, 0.0);
328 }
329 if (G1UseParallelRSetScanning || (worker_i == 0)) {
330 scanRS(oc, worker_i);
331 } else {
332 _g1p->phase_times()->record_scan_rs_time(worker_i, 0.0);
333 }
335 // We now clear the cached values of _cset_rs_update_cl for this worker
336 _cset_rs_update_cl[worker_i] = NULL;
337 }
339 void G1RemSet::prepare_for_oops_into_collection_set_do() {
340 cleanupHRRS();
341 ConcurrentG1Refine* cg1r = _g1->concurrent_g1_refine();
342 _g1->set_refine_cte_cl_concurrency(false);
343 DirtyCardQueueSet& dcqs = JavaThread::dirty_card_queue_set();
344 dcqs.concatenate_logs();
346 if (G1CollectedHeap::use_parallel_gc_threads()) {
347 // Don't set the number of workers here. It will be set
348 // when the task is run
349 // _seq_task->set_n_termination((int)n_workers());
350 }
351 guarantee( _cards_scanned == NULL, "invariant" );
352 _cards_scanned = NEW_C_HEAP_ARRAY(size_t, n_workers(), mtGC);
353 for (uint i = 0; i < n_workers(); ++i) {
354 _cards_scanned[i] = 0;
355 }
356 _total_cards_scanned = 0;
357 }
360 // This closure, applied to a DirtyCardQueueSet, is used to immediately
361 // update the RSets for the regions in the CSet. For each card it iterates
362 // through the oops which coincide with that card. It scans the reference
363 // fields in each oop; when it finds an oop that points into the collection
364 // set, the RSet for the region containing the referenced object is updated.
365 class UpdateRSetCardTableEntryIntoCSetClosure: public CardTableEntryClosure {
366 G1CollectedHeap* _g1;
367 CardTableModRefBS* _ct_bs;
368 public:
369 UpdateRSetCardTableEntryIntoCSetClosure(G1CollectedHeap* g1,
370 CardTableModRefBS* bs):
371 _g1(g1), _ct_bs(bs)
372 { }
374 bool do_card_ptr(jbyte* card_ptr, int worker_i) {
375 // Construct the region representing the card.
376 HeapWord* start = _ct_bs->addr_for(card_ptr);
377 // And find the region containing it.
378 HeapRegion* r = _g1->heap_region_containing(start);
379 assert(r != NULL, "unexpected null");
381 // Scan oops in the card looking for references into the collection set
382 // Don't use addr_for(card_ptr + 1) which can ask for
383 // a card beyond the heap. This is not safe without a perm
384 // gen.
385 HeapWord* end = start + CardTableModRefBS::card_size_in_words;
386 MemRegion scanRegion(start, end);
388 UpdateRSetImmediate update_rs_cl(_g1->g1_rem_set());
389 FilterIntoCSClosure update_rs_cset_oop_cl(NULL, _g1, &update_rs_cl);
390 FilterOutOfRegionClosure filter_then_update_rs_cset_oop_cl(r, &update_rs_cset_oop_cl);
392 // We can pass false as the "filter_young" parameter here as:
393 // * we should be in a STW pause,
394 // * the DCQS to which this closure is applied is used to hold
395 // references that point into the collection set from the prior
396 // RSet updating,
397 // * the post-write barrier shouldn't be logging updates to young
398 // regions (but there is a situation where this can happen - see
399 // the comment in G1RemSet::concurrentRefineOneCard below -
400 // that should not be applicable here), and
401 // * during actual RSet updating, the filtering of cards in young
402 // regions in HeapRegion::oops_on_card_seq_iterate_careful is
403 // employed.
404 // As a result, when this closure is applied to "refs into cset"
405 // DCQS, we shouldn't see any cards in young regions.
406 update_rs_cl.set_region(r);
407 HeapWord* stop_point =
408 r->oops_on_card_seq_iterate_careful(scanRegion,
409 &filter_then_update_rs_cset_oop_cl,
410 false /* filter_young */,
411 NULL /* card_ptr */);
413 // Since this is performed in the event of an evacuation failure, we
414 // we shouldn't see a non-null stop point
415 assert(stop_point == NULL, "saw an unallocated region");
416 return true;
417 }
418 };
420 void G1RemSet::cleanup_after_oops_into_collection_set_do() {
421 guarantee( _cards_scanned != NULL, "invariant" );
422 _total_cards_scanned = 0;
423 for (uint i = 0; i < n_workers(); ++i) {
424 _total_cards_scanned += _cards_scanned[i];
425 }
426 FREE_C_HEAP_ARRAY(size_t, _cards_scanned, mtGC);
427 _cards_scanned = NULL;
428 // Cleanup after copy
429 _g1->set_refine_cte_cl_concurrency(true);
430 // Set all cards back to clean.
431 _g1->cleanUpCardTable();
433 DirtyCardQueueSet& into_cset_dcqs = _g1->into_cset_dirty_card_queue_set();
434 int into_cset_n_buffers = into_cset_dcqs.completed_buffers_num();
436 if (_g1->evacuation_failed()) {
437 // Restore remembered sets for the regions pointing into the collection set.
439 if (G1DeferredRSUpdate) {
440 // If deferred RS updates are enabled then we just need to transfer
441 // the completed buffers from (a) the DirtyCardQueueSet used to hold
442 // cards that contain references that point into the collection set
443 // to (b) the DCQS used to hold the deferred RS updates
444 _g1->dirty_card_queue_set().merge_bufferlists(&into_cset_dcqs);
445 } else {
447 CardTableModRefBS* bs = (CardTableModRefBS*)_g1->barrier_set();
448 UpdateRSetCardTableEntryIntoCSetClosure update_rs_cset_immediate(_g1, bs);
450 int n_completed_buffers = 0;
451 while (into_cset_dcqs.apply_closure_to_completed_buffer(&update_rs_cset_immediate,
452 0, 0, true)) {
453 n_completed_buffers++;
454 }
455 assert(n_completed_buffers == into_cset_n_buffers, "missed some buffers");
456 }
457 }
459 // Free any completed buffers in the DirtyCardQueueSet used to hold cards
460 // which contain references that point into the collection.
461 _g1->into_cset_dirty_card_queue_set().clear();
462 assert(_g1->into_cset_dirty_card_queue_set().completed_buffers_num() == 0,
463 "all buffers should be freed");
464 _g1->into_cset_dirty_card_queue_set().clear_n_completed_buffers();
465 }
467 class ScrubRSClosure: public HeapRegionClosure {
468 G1CollectedHeap* _g1h;
469 BitMap* _region_bm;
470 BitMap* _card_bm;
471 CardTableModRefBS* _ctbs;
472 public:
473 ScrubRSClosure(BitMap* region_bm, BitMap* card_bm) :
474 _g1h(G1CollectedHeap::heap()),
475 _region_bm(region_bm), _card_bm(card_bm),
476 _ctbs(NULL)
477 {
478 ModRefBarrierSet* bs = _g1h->mr_bs();
479 guarantee(bs->is_a(BarrierSet::CardTableModRef), "Precondition");
480 _ctbs = (CardTableModRefBS*)bs;
481 }
483 bool doHeapRegion(HeapRegion* r) {
484 if (!r->continuesHumongous()) {
485 r->rem_set()->scrub(_ctbs, _region_bm, _card_bm);
486 }
487 return false;
488 }
489 };
491 void G1RemSet::scrub(BitMap* region_bm, BitMap* card_bm) {
492 ScrubRSClosure scrub_cl(region_bm, card_bm);
493 _g1->heap_region_iterate(&scrub_cl);
494 }
496 void G1RemSet::scrub_par(BitMap* region_bm, BitMap* card_bm,
497 uint worker_num, int claim_val) {
498 ScrubRSClosure scrub_cl(region_bm, card_bm);
499 _g1->heap_region_par_iterate_chunked(&scrub_cl,
500 worker_num,
501 n_workers(),
502 claim_val);
503 }
507 G1TriggerClosure::G1TriggerClosure() :
508 _triggered(false) { }
510 G1InvokeIfNotTriggeredClosure::G1InvokeIfNotTriggeredClosure(G1TriggerClosure* t_cl,
511 OopClosure* oop_cl) :
512 _trigger_cl(t_cl), _oop_cl(oop_cl) { }
514 G1Mux2Closure::G1Mux2Closure(OopClosure *c1, OopClosure *c2) :
515 _c1(c1), _c2(c2) { }
517 G1UpdateRSOrPushRefOopClosure::
518 G1UpdateRSOrPushRefOopClosure(G1CollectedHeap* g1h,
519 G1RemSet* rs,
520 OopsInHeapRegionClosure* push_ref_cl,
521 bool record_refs_into_cset,
522 int worker_i) :
523 _g1(g1h), _g1_rem_set(rs), _from(NULL),
524 _record_refs_into_cset(record_refs_into_cset),
525 _push_ref_cl(push_ref_cl), _worker_i(worker_i) { }
527 bool G1RemSet::concurrentRefineOneCard_impl(jbyte* card_ptr, int worker_i,
528 bool check_for_refs_into_cset) {
529 // Construct the region representing the card.
530 HeapWord* start = _ct_bs->addr_for(card_ptr);
531 // And find the region containing it.
532 HeapRegion* r = _g1->heap_region_containing(start);
533 assert(r != NULL, "unexpected null");
535 // Don't use addr_for(card_ptr + 1) which can ask for
536 // a card beyond the heap. This is not safe without a perm
537 // gen at the upper end of the heap.
538 HeapWord* end = start + CardTableModRefBS::card_size_in_words;
539 MemRegion dirtyRegion(start, end);
541 #if CARD_REPEAT_HISTO
542 init_ct_freq_table(_g1->max_capacity());
543 ct_freq_note_card(_ct_bs->index_for(start));
544 #endif
546 OopsInHeapRegionClosure* oops_in_heap_closure = NULL;
547 if (check_for_refs_into_cset) {
548 // ConcurrentG1RefineThreads have worker numbers larger than what
549 // _cset_rs_update_cl[] is set up to handle. But those threads should
550 // only be active outside of a collection which means that when they
551 // reach here they should have check_for_refs_into_cset == false.
552 assert((size_t)worker_i < n_workers(), "index of worker larger than _cset_rs_update_cl[].length");
553 oops_in_heap_closure = _cset_rs_update_cl[worker_i];
554 }
555 G1UpdateRSOrPushRefOopClosure update_rs_oop_cl(_g1,
556 _g1->g1_rem_set(),
557 oops_in_heap_closure,
558 check_for_refs_into_cset,
559 worker_i);
560 update_rs_oop_cl.set_from(r);
562 G1TriggerClosure trigger_cl;
563 FilterIntoCSClosure into_cs_cl(NULL, _g1, &trigger_cl);
564 G1InvokeIfNotTriggeredClosure invoke_cl(&trigger_cl, &into_cs_cl);
565 G1Mux2Closure mux(&invoke_cl, &update_rs_oop_cl);
567 FilterOutOfRegionClosure filter_then_update_rs_oop_cl(r,
568 (check_for_refs_into_cset ?
569 (OopClosure*)&mux :
570 (OopClosure*)&update_rs_oop_cl));
572 // The region for the current card may be a young region. The
573 // current card may have been a card that was evicted from the
574 // card cache. When the card was inserted into the cache, we had
575 // determined that its region was non-young. While in the cache,
576 // the region may have been freed during a cleanup pause, reallocated
577 // and tagged as young.
578 //
579 // We wish to filter out cards for such a region but the current
580 // thread, if we're running concurrently, may "see" the young type
581 // change at any time (so an earlier "is_young" check may pass or
582 // fail arbitrarily). We tell the iteration code to perform this
583 // filtering when it has been determined that there has been an actual
584 // allocation in this region and making it safe to check the young type.
585 bool filter_young = true;
587 HeapWord* stop_point =
588 r->oops_on_card_seq_iterate_careful(dirtyRegion,
589 &filter_then_update_rs_oop_cl,
590 filter_young,
591 card_ptr);
593 // If stop_point is non-null, then we encountered an unallocated region
594 // (perhaps the unfilled portion of a TLAB.) For now, we'll dirty the
595 // card and re-enqueue: if we put off the card until a GC pause, then the
596 // unallocated portion will be filled in. Alternatively, we might try
597 // the full complexity of the technique used in "regular" precleaning.
598 if (stop_point != NULL) {
599 // The card might have gotten re-dirtied and re-enqueued while we
600 // worked. (In fact, it's pretty likely.)
601 if (*card_ptr != CardTableModRefBS::dirty_card_val()) {
602 *card_ptr = CardTableModRefBS::dirty_card_val();
603 MutexLockerEx x(Shared_DirtyCardQ_lock,
604 Mutex::_no_safepoint_check_flag);
605 DirtyCardQueue* sdcq =
606 JavaThread::dirty_card_queue_set().shared_dirty_card_queue();
607 sdcq->enqueue(card_ptr);
608 }
609 } else {
610 _conc_refine_cards++;
611 }
613 return trigger_cl.triggered();
614 }
616 bool G1RemSet::concurrentRefineOneCard(jbyte* card_ptr, int worker_i,
617 bool check_for_refs_into_cset) {
618 // If the card is no longer dirty, nothing to do.
619 if (*card_ptr != CardTableModRefBS::dirty_card_val()) {
620 // No need to return that this card contains refs that point
621 // into the collection set.
622 return false;
623 }
625 // Construct the region representing the card.
626 HeapWord* start = _ct_bs->addr_for(card_ptr);
627 // And find the region containing it.
628 HeapRegion* r = _g1->heap_region_containing(start);
629 if (r == NULL) {
630 // Again no need to return that this card contains refs that
631 // point into the collection set.
632 return false; // Not in the G1 heap (might be in perm, for example.)
633 }
634 // Why do we have to check here whether a card is on a young region,
635 // given that we dirty young regions and, as a result, the
636 // post-barrier is supposed to filter them out and never to enqueue
637 // them? When we allocate a new region as the "allocation region" we
638 // actually dirty its cards after we release the lock, since card
639 // dirtying while holding the lock was a performance bottleneck. So,
640 // as a result, it is possible for other threads to actually
641 // allocate objects in the region (after the acquire the lock)
642 // before all the cards on the region are dirtied. This is unlikely,
643 // and it doesn't happen often, but it can happen. So, the extra
644 // check below filters out those cards.
645 if (r->is_young()) {
646 return false;
647 }
648 // While we are processing RSet buffers during the collection, we
649 // actually don't want to scan any cards on the collection set,
650 // since we don't want to update remebered sets with entries that
651 // point into the collection set, given that live objects from the
652 // collection set are about to move and such entries will be stale
653 // very soon. This change also deals with a reliability issue which
654 // involves scanning a card in the collection set and coming across
655 // an array that was being chunked and looking malformed. Note,
656 // however, that if evacuation fails, we have to scan any objects
657 // that were not moved and create any missing entries.
658 if (r->in_collection_set()) {
659 return false;
660 }
662 // Should we defer processing the card?
663 //
664 // Previously the result from the insert_cache call would be
665 // either card_ptr (implying that card_ptr was currently "cold"),
666 // null (meaning we had inserted the card ptr into the "hot"
667 // cache, which had some headroom), or a "hot" card ptr
668 // extracted from the "hot" cache.
669 //
670 // Now that the _card_counts cache in the ConcurrentG1Refine
671 // instance is an evicting hash table, the result we get back
672 // could be from evicting the card ptr in an already occupied
673 // bucket (in which case we have replaced the card ptr in the
674 // bucket with card_ptr and "defer" is set to false). To avoid
675 // having a data structure (updates to which would need a lock)
676 // to hold these unprocessed dirty cards, we need to immediately
677 // process card_ptr. The actions needed to be taken on return
678 // from cache_insert are summarized in the following table:
679 //
680 // res defer action
681 // --------------------------------------------------------------
682 // null false card evicted from _card_counts & replaced with
683 // card_ptr; evicted ptr added to hot cache.
684 // No need to process res; immediately process card_ptr
685 //
686 // null true card not evicted from _card_counts; card_ptr added
687 // to hot cache.
688 // Nothing to do.
689 //
690 // non-null false card evicted from _card_counts & replaced with
691 // card_ptr; evicted ptr is currently "cold" or
692 // caused an eviction from the hot cache.
693 // Immediately process res; process card_ptr.
694 //
695 // non-null true card not evicted from _card_counts; card_ptr is
696 // currently cold, or caused an eviction from hot
697 // cache.
698 // Immediately process res; no need to process card_ptr.
701 jbyte* res = card_ptr;
702 bool defer = false;
704 // This gets set to true if the card being refined has references
705 // that point into the collection set.
706 bool oops_into_cset = false;
708 if (_cg1r->use_cache()) {
709 jbyte* res = _cg1r->cache_insert(card_ptr, &defer);
710 if (res != NULL && (res != card_ptr || defer)) {
711 start = _ct_bs->addr_for(res);
712 r = _g1->heap_region_containing(start);
713 if (r != NULL) {
714 // Checking whether the region we got back from the cache
715 // is young here is inappropriate. The region could have been
716 // freed, reallocated and tagged as young while in the cache.
717 // Hence we could see its young type change at any time.
718 //
719 // Process card pointer we get back from the hot card cache. This
720 // will check whether the region containing the card is young
721 // _after_ checking that the region has been allocated from.
722 oops_into_cset = concurrentRefineOneCard_impl(res, worker_i,
723 false /* check_for_refs_into_cset */);
724 // The above call to concurrentRefineOneCard_impl is only
725 // performed if the hot card cache is enabled. This cache is
726 // disabled during an evacuation pause - which is the only
727 // time when we need know if the card contains references
728 // that point into the collection set. Also when the hot card
729 // cache is enabled, this code is executed by the concurrent
730 // refine threads - rather than the GC worker threads - and
731 // concurrentRefineOneCard_impl will return false.
732 assert(!oops_into_cset, "should not see true here");
733 }
734 }
735 }
737 if (!defer) {
738 oops_into_cset =
739 concurrentRefineOneCard_impl(card_ptr, worker_i, check_for_refs_into_cset);
740 // We should only be detecting that the card contains references
741 // that point into the collection set if the current thread is
742 // a GC worker thread.
743 assert(!oops_into_cset || SafepointSynchronize::is_at_safepoint(),
744 "invalid result at non safepoint");
745 }
746 return oops_into_cset;
747 }
749 class HRRSStatsIter: public HeapRegionClosure {
750 size_t _occupied;
751 size_t _total_mem_sz;
752 size_t _max_mem_sz;
753 HeapRegion* _max_mem_sz_region;
754 public:
755 HRRSStatsIter() :
756 _occupied(0),
757 _total_mem_sz(0),
758 _max_mem_sz(0),
759 _max_mem_sz_region(NULL)
760 {}
762 bool doHeapRegion(HeapRegion* r) {
763 if (r->continuesHumongous()) return false;
764 size_t mem_sz = r->rem_set()->mem_size();
765 if (mem_sz > _max_mem_sz) {
766 _max_mem_sz = mem_sz;
767 _max_mem_sz_region = r;
768 }
769 _total_mem_sz += mem_sz;
770 size_t occ = r->rem_set()->occupied();
771 _occupied += occ;
772 return false;
773 }
774 size_t total_mem_sz() { return _total_mem_sz; }
775 size_t max_mem_sz() { return _max_mem_sz; }
776 size_t occupied() { return _occupied; }
777 HeapRegion* max_mem_sz_region() { return _max_mem_sz_region; }
778 };
780 class PrintRSThreadVTimeClosure : public ThreadClosure {
781 public:
782 virtual void do_thread(Thread *t) {
783 ConcurrentG1RefineThread* crt = (ConcurrentG1RefineThread*) t;
784 gclog_or_tty->print(" %5.2f", crt->vtime_accum());
785 }
786 };
788 void G1RemSet::print_summary_info() {
789 G1CollectedHeap* g1 = G1CollectedHeap::heap();
791 #if CARD_REPEAT_HISTO
792 gclog_or_tty->print_cr("\nG1 card_repeat count histogram: ");
793 gclog_or_tty->print_cr(" # of repeats --> # of cards with that number.");
794 card_repeat_count.print_on(gclog_or_tty);
795 #endif
797 gclog_or_tty->print_cr("\n Concurrent RS processed %d cards",
798 _conc_refine_cards);
799 DirtyCardQueueSet& dcqs = JavaThread::dirty_card_queue_set();
800 jint tot_processed_buffers =
801 dcqs.processed_buffers_mut() + dcqs.processed_buffers_rs_thread();
802 gclog_or_tty->print_cr(" Of %d completed buffers:", tot_processed_buffers);
803 gclog_or_tty->print_cr(" %8d (%5.1f%%) by conc RS threads.",
804 dcqs.processed_buffers_rs_thread(),
805 100.0*(float)dcqs.processed_buffers_rs_thread()/
806 (float)tot_processed_buffers);
807 gclog_or_tty->print_cr(" %8d (%5.1f%%) by mutator threads.",
808 dcqs.processed_buffers_mut(),
809 100.0*(float)dcqs.processed_buffers_mut()/
810 (float)tot_processed_buffers);
811 gclog_or_tty->print_cr(" Conc RS threads times(s)");
812 PrintRSThreadVTimeClosure p;
813 gclog_or_tty->print(" ");
814 g1->concurrent_g1_refine()->threads_do(&p);
815 gclog_or_tty->print_cr("");
817 HRRSStatsIter blk;
818 g1->heap_region_iterate(&blk);
819 gclog_or_tty->print_cr(" Total heap region rem set sizes = "SIZE_FORMAT"K."
820 " Max = "SIZE_FORMAT"K.",
821 blk.total_mem_sz()/K, blk.max_mem_sz()/K);
822 gclog_or_tty->print_cr(" Static structures = "SIZE_FORMAT"K,"
823 " free_lists = "SIZE_FORMAT"K.",
824 HeapRegionRemSet::static_mem_size() / K,
825 HeapRegionRemSet::fl_mem_size() / K);
826 gclog_or_tty->print_cr(" "SIZE_FORMAT" occupied cards represented.",
827 blk.occupied());
828 HeapRegion* max_mem_sz_region = blk.max_mem_sz_region();
829 HeapRegionRemSet* rem_set = max_mem_sz_region->rem_set();
830 gclog_or_tty->print_cr(" Max size region = "HR_FORMAT", "
831 "size = "SIZE_FORMAT "K, occupied = "SIZE_FORMAT"K.",
832 HR_FORMAT_PARAMS(max_mem_sz_region),
833 (rem_set->mem_size() + K - 1)/K,
834 (rem_set->occupied() + K - 1)/K);
835 gclog_or_tty->print_cr(" Did %d coarsenings.",
836 HeapRegionRemSet::n_coarsenings());
837 }
839 void G1RemSet::prepare_for_verify() {
840 if (G1HRRSFlushLogBuffersOnVerify &&
841 (VerifyBeforeGC || VerifyAfterGC)
842 && !_g1->full_collection()) {
843 cleanupHRRS();
844 _g1->set_refine_cte_cl_concurrency(false);
845 if (SafepointSynchronize::is_at_safepoint()) {
846 DirtyCardQueueSet& dcqs = JavaThread::dirty_card_queue_set();
847 dcqs.concatenate_logs();
848 }
849 bool cg1r_use_cache = _cg1r->use_cache();
850 _cg1r->set_use_cache(false);
851 DirtyCardQueue into_cset_dcq(&_g1->into_cset_dirty_card_queue_set());
852 updateRS(&into_cset_dcq, 0);
853 _g1->into_cset_dirty_card_queue_set().clear();
854 _cg1r->set_use_cache(cg1r_use_cache);
856 assert(JavaThread::dirty_card_queue_set().completed_buffers_num() == 0, "All should be consumed");
857 }
858 }