Thu, 12 Jan 2012 00:06:47 -0800
6484965: G1: piggy-back liveness accounting phase on marking
Summary: Remove the separate counting phase of concurrent marking by tracking the amount of marked bytes and the cards spanned by marked objects in marking task/worker thread local data structures, which are updated as individual objects are marked.
Reviewed-by: brutisso, tonyp
1 /*
2 * Copyright (c) 2001, 2011, 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/g1OopClosures.inline.hpp"
33 #include "gc_implementation/g1/g1RemSet.inline.hpp"
34 #include "gc_implementation/g1/heapRegionSeq.inline.hpp"
35 #include "memory/iterator.hpp"
36 #include "oops/oop.inline.hpp"
37 #include "utilities/intHisto.hpp"
39 #define CARD_REPEAT_HISTO 0
41 #if CARD_REPEAT_HISTO
42 static size_t ct_freq_sz;
43 static jbyte* ct_freq = NULL;
45 void init_ct_freq_table(size_t heap_sz_bytes) {
46 if (ct_freq == NULL) {
47 ct_freq_sz = heap_sz_bytes/CardTableModRefBS::card_size;
48 ct_freq = new jbyte[ct_freq_sz];
49 for (size_t j = 0; j < ct_freq_sz; j++) ct_freq[j] = 0;
50 }
51 }
53 void ct_freq_note_card(size_t index) {
54 assert(0 <= index && index < ct_freq_sz, "Bounds error.");
55 if (ct_freq[index] < 100) { ct_freq[index]++; }
56 }
58 static IntHistogram card_repeat_count(10, 10);
60 void ct_freq_update_histo_and_reset() {
61 for (size_t j = 0; j < ct_freq_sz; j++) {
62 card_repeat_count.add_entry(ct_freq[j]);
63 ct_freq[j] = 0;
64 }
66 }
67 #endif
69 G1RemSet::G1RemSet(G1CollectedHeap* g1, CardTableModRefBS* ct_bs)
70 : _g1(g1), _conc_refine_cards(0),
71 _ct_bs(ct_bs), _g1p(_g1->g1_policy()),
72 _cg1r(g1->concurrent_g1_refine()),
73 _cset_rs_update_cl(NULL),
74 _cards_scanned(NULL), _total_cards_scanned(0)
75 {
76 _seq_task = new SubTasksDone(NumSeqTasks);
77 guarantee(n_workers() > 0, "There should be some workers");
78 _cset_rs_update_cl = NEW_C_HEAP_ARRAY(OopsInHeapRegionClosure*, n_workers());
79 for (uint i = 0; i < n_workers(); i++) {
80 _cset_rs_update_cl[i] = NULL;
81 }
82 }
84 G1RemSet::~G1RemSet() {
85 delete _seq_task;
86 for (uint i = 0; i < n_workers(); i++) {
87 assert(_cset_rs_update_cl[i] == NULL, "it should be");
88 }
89 FREE_C_HEAP_ARRAY(OopsInHeapRegionClosure*, _cset_rs_update_cl);
90 }
92 void CountNonCleanMemRegionClosure::do_MemRegion(MemRegion mr) {
93 if (_g1->is_in_g1_reserved(mr.start())) {
94 _n += (int) ((mr.byte_size() / CardTableModRefBS::card_size));
95 if (_start_first == NULL) _start_first = mr.start();
96 }
97 }
99 class ScanRSClosure : public HeapRegionClosure {
100 size_t _cards_done, _cards;
101 G1CollectedHeap* _g1h;
102 OopsInHeapRegionClosure* _oc;
103 G1BlockOffsetSharedArray* _bot_shared;
104 CardTableModRefBS *_ct_bs;
105 int _worker_i;
106 int _block_size;
107 bool _try_claimed;
108 public:
109 ScanRSClosure(OopsInHeapRegionClosure* oc, int worker_i) :
110 _oc(oc),
111 _cards(0),
112 _cards_done(0),
113 _worker_i(worker_i),
114 _try_claimed(false)
115 {
116 _g1h = G1CollectedHeap::heap();
117 _bot_shared = _g1h->bot_shared();
118 _ct_bs = (CardTableModRefBS*) (_g1h->barrier_set());
119 _block_size = MAX2<int>(G1RSetScanBlockSize, 1);
120 }
122 void set_try_claimed() { _try_claimed = true; }
124 void scanCard(size_t index, HeapRegion *r) {
125 // Stack allocate the DirtyCardToOopClosure instance
126 HeapRegionDCTOC cl(_g1h, r, _oc,
127 CardTableModRefBS::Precise,
128 HeapRegionDCTOC::IntoCSFilterKind);
130 // Set the "from" region in the closure.
131 _oc->set_region(r);
132 HeapWord* card_start = _bot_shared->address_for_index(index);
133 HeapWord* card_end = card_start + G1BlockOffsetSharedArray::N_words;
134 Space *sp = SharedHeap::heap()->space_containing(card_start);
135 MemRegion sm_region = sp->used_region_at_save_marks();
136 MemRegion mr = sm_region.intersection(MemRegion(card_start,card_end));
137 if (!mr.is_empty() && !_ct_bs->is_card_claimed(index)) {
138 // We make the card as "claimed" lazily (so races are possible
139 // but they're benign), which reduces the number of duplicate
140 // scans (the rsets of the regions in the cset can intersect).
141 _ct_bs->set_card_claimed(index);
142 _cards_done++;
143 cl.do_MemRegion(mr);
144 }
145 }
147 void printCard(HeapRegion* card_region, size_t card_index,
148 HeapWord* card_start) {
149 gclog_or_tty->print_cr("T %d Region [" PTR_FORMAT ", " PTR_FORMAT ") "
150 "RS names card %p: "
151 "[" PTR_FORMAT ", " PTR_FORMAT ")",
152 _worker_i,
153 card_region->bottom(), card_region->end(),
154 card_index,
155 card_start, card_start + G1BlockOffsetSharedArray::N_words);
156 }
158 bool doHeapRegion(HeapRegion* r) {
159 assert(r->in_collection_set(), "should only be called on elements of CS.");
160 HeapRegionRemSet* hrrs = r->rem_set();
161 if (hrrs->iter_is_complete()) return false; // All done.
162 if (!_try_claimed && !hrrs->claim_iter()) return false;
163 // If we ever free the collection set concurrently, we should also
164 // clear the card table concurrently therefore we won't need to
165 // add regions of the collection set to the dirty cards region.
166 _g1h->push_dirty_cards_region(r);
167 // If we didn't return above, then
168 // _try_claimed || r->claim_iter()
169 // is true: either we're supposed to work on claimed-but-not-complete
170 // regions, or we successfully claimed the region.
171 HeapRegionRemSetIterator* iter = _g1h->rem_set_iterator(_worker_i);
172 hrrs->init_iterator(iter);
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->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->record_update_rs_time(worker_i, (os::elapsedTime() - start) * 1000.0);
280 }
282 class CountRSSizeClosure: public HeapRegionClosure {
283 size_t _n;
284 size_t _tot;
285 size_t _max;
286 HeapRegion* _max_r;
287 enum {
288 N = 20,
289 MIN = 6
290 };
291 int _histo[N];
292 public:
293 CountRSSizeClosure() : _n(0), _tot(0), _max(0), _max_r(NULL) {
294 for (int i = 0; i < N; i++) _histo[i] = 0;
295 }
296 bool doHeapRegion(HeapRegion* r) {
297 if (!r->continuesHumongous()) {
298 size_t occ = r->rem_set()->occupied();
299 _n++;
300 _tot += occ;
301 if (occ > _max) {
302 _max = occ;
303 _max_r = r;
304 }
305 // Fit it into a histo bin.
306 int s = 1 << MIN;
307 int i = 0;
308 while (occ > (size_t) s && i < (N-1)) {
309 s = s << 1;
310 i++;
311 }
312 _histo[i]++;
313 }
314 return false;
315 }
316 size_t n() { return _n; }
317 size_t tot() { return _tot; }
318 size_t mx() { return _max; }
319 HeapRegion* mxr() { return _max_r; }
320 void print_histo() {
321 int mx = N;
322 while (mx >= 0) {
323 if (_histo[mx-1] > 0) break;
324 mx--;
325 }
326 gclog_or_tty->print_cr("Number of regions with given RS sizes:");
327 gclog_or_tty->print_cr(" <= %8d %8d", 1 << MIN, _histo[0]);
328 for (int i = 1; i < mx-1; i++) {
329 gclog_or_tty->print_cr(" %8d - %8d %8d",
330 (1 << (MIN + i - 1)) + 1,
331 1 << (MIN + i),
332 _histo[i]);
333 }
334 gclog_or_tty->print_cr(" > %8d %8d", (1 << (MIN+mx-2))+1, _histo[mx-1]);
335 }
336 };
338 void G1RemSet::cleanupHRRS() {
339 HeapRegionRemSet::cleanup();
340 }
342 void G1RemSet::oops_into_collection_set_do(OopsInHeapRegionClosure* oc,
343 int worker_i) {
344 #if CARD_REPEAT_HISTO
345 ct_freq_update_histo_and_reset();
346 #endif
347 if (worker_i == 0) {
348 _cg1r->clear_and_record_card_counts();
349 }
351 // Make this into a command-line flag...
352 if (G1RSCountHisto && (ParallelGCThreads == 0 || worker_i == 0)) {
353 CountRSSizeClosure count_cl;
354 _g1->heap_region_iterate(&count_cl);
355 gclog_or_tty->print_cr("Avg of %d RS counts is %f, max is %d, "
356 "max region is " PTR_FORMAT,
357 count_cl.n(), (float)count_cl.tot()/(float)count_cl.n(),
358 count_cl.mx(), count_cl.mxr());
359 count_cl.print_histo();
360 }
362 // We cache the value of 'oc' closure into the appropriate slot in the
363 // _cset_rs_update_cl for this worker
364 assert(worker_i < (int)n_workers(), "sanity");
365 _cset_rs_update_cl[worker_i] = oc;
367 // A DirtyCardQueue that is used to hold cards containing references
368 // that point into the collection set. This DCQ is associated with a
369 // special DirtyCardQueueSet (see g1CollectedHeap.hpp). Under normal
370 // circumstances (i.e. the pause successfully completes), these cards
371 // are just discarded (there's no need to update the RSets of regions
372 // that were in the collection set - after the pause these regions
373 // are wholly 'free' of live objects. In the event of an evacuation
374 // failure the cards/buffers in this queue set are:
375 // * passed to the DirtyCardQueueSet that is used to manage deferred
376 // RSet updates, or
377 // * scanned for references that point into the collection set
378 // and the RSet of the corresponding region in the collection set
379 // is updated immediately.
380 DirtyCardQueue into_cset_dcq(&_g1->into_cset_dirty_card_queue_set());
382 assert((ParallelGCThreads > 0) || worker_i == 0, "invariant");
384 // The two flags below were introduced temporarily to serialize
385 // the updating and scanning of remembered sets. There are some
386 // race conditions when these two operations are done in parallel
387 // and they are causing failures. When we resolve said race
388 // conditions, we'll revert back to parallel remembered set
389 // updating and scanning. See CRs 6677707 and 6677708.
390 if (G1UseParallelRSetUpdating || (worker_i == 0)) {
391 updateRS(&into_cset_dcq, worker_i);
392 } else {
393 _g1p->record_update_rs_processed_buffers(worker_i, 0.0);
394 _g1p->record_update_rs_time(worker_i, 0.0);
395 }
396 if (G1UseParallelRSetScanning || (worker_i == 0)) {
397 scanRS(oc, worker_i);
398 } else {
399 _g1p->record_scan_rs_time(worker_i, 0.0);
400 }
402 // We now clear the cached values of _cset_rs_update_cl for this worker
403 _cset_rs_update_cl[worker_i] = NULL;
404 }
406 void G1RemSet::prepare_for_oops_into_collection_set_do() {
407 cleanupHRRS();
408 ConcurrentG1Refine* cg1r = _g1->concurrent_g1_refine();
409 _g1->set_refine_cte_cl_concurrency(false);
410 DirtyCardQueueSet& dcqs = JavaThread::dirty_card_queue_set();
411 dcqs.concatenate_logs();
413 if (G1CollectedHeap::use_parallel_gc_threads()) {
414 // Don't set the number of workers here. It will be set
415 // when the task is run
416 // _seq_task->set_n_termination((int)n_workers());
417 }
418 guarantee( _cards_scanned == NULL, "invariant" );
419 _cards_scanned = NEW_C_HEAP_ARRAY(size_t, n_workers());
420 for (uint i = 0; i < n_workers(); ++i) {
421 _cards_scanned[i] = 0;
422 }
423 _total_cards_scanned = 0;
424 }
427 // This closure, applied to a DirtyCardQueueSet, is used to immediately
428 // update the RSets for the regions in the CSet. For each card it iterates
429 // through the oops which coincide with that card. It scans the reference
430 // fields in each oop; when it finds an oop that points into the collection
431 // set, the RSet for the region containing the referenced object is updated.
432 class UpdateRSetCardTableEntryIntoCSetClosure: public CardTableEntryClosure {
433 G1CollectedHeap* _g1;
434 CardTableModRefBS* _ct_bs;
435 public:
436 UpdateRSetCardTableEntryIntoCSetClosure(G1CollectedHeap* g1,
437 CardTableModRefBS* bs):
438 _g1(g1), _ct_bs(bs)
439 { }
441 bool do_card_ptr(jbyte* card_ptr, int worker_i) {
442 // Construct the region representing the card.
443 HeapWord* start = _ct_bs->addr_for(card_ptr);
444 // And find the region containing it.
445 HeapRegion* r = _g1->heap_region_containing(start);
446 assert(r != NULL, "unexpected null");
448 // Scan oops in the card looking for references into the collection set
449 HeapWord* end = _ct_bs->addr_for(card_ptr + 1);
450 MemRegion scanRegion(start, end);
452 UpdateRSetImmediate update_rs_cl(_g1->g1_rem_set());
453 FilterIntoCSClosure update_rs_cset_oop_cl(NULL, _g1, &update_rs_cl);
454 FilterOutOfRegionClosure filter_then_update_rs_cset_oop_cl(r, &update_rs_cset_oop_cl);
456 // We can pass false as the "filter_young" parameter here as:
457 // * we should be in a STW pause,
458 // * the DCQS to which this closure is applied is used to hold
459 // references that point into the collection set from the prior
460 // RSet updating,
461 // * the post-write barrier shouldn't be logging updates to young
462 // regions (but there is a situation where this can happen - see
463 // the comment in G1RemSet::concurrentRefineOneCard below -
464 // that should not be applicable here), and
465 // * during actual RSet updating, the filtering of cards in young
466 // regions in HeapRegion::oops_on_card_seq_iterate_careful is
467 // employed.
468 // As a result, when this closure is applied to "refs into cset"
469 // DCQS, we shouldn't see any cards in young regions.
470 update_rs_cl.set_region(r);
471 HeapWord* stop_point =
472 r->oops_on_card_seq_iterate_careful(scanRegion,
473 &filter_then_update_rs_cset_oop_cl,
474 false /* filter_young */,
475 NULL /* card_ptr */);
477 // Since this is performed in the event of an evacuation failure, we
478 // we shouldn't see a non-null stop point
479 assert(stop_point == NULL, "saw an unallocated region");
480 return true;
481 }
482 };
484 void G1RemSet::cleanup_after_oops_into_collection_set_do() {
485 guarantee( _cards_scanned != NULL, "invariant" );
486 _total_cards_scanned = 0;
487 for (uint i = 0; i < n_workers(); ++i) {
488 _total_cards_scanned += _cards_scanned[i];
489 }
490 FREE_C_HEAP_ARRAY(size_t, _cards_scanned);
491 _cards_scanned = NULL;
492 // Cleanup after copy
493 _g1->set_refine_cte_cl_concurrency(true);
494 // Set all cards back to clean.
495 _g1->cleanUpCardTable();
497 DirtyCardQueueSet& into_cset_dcqs = _g1->into_cset_dirty_card_queue_set();
498 int into_cset_n_buffers = into_cset_dcqs.completed_buffers_num();
500 if (_g1->evacuation_failed()) {
501 // Restore remembered sets for the regions pointing into the collection set.
503 if (G1DeferredRSUpdate) {
504 // If deferred RS updates are enabled then we just need to transfer
505 // the completed buffers from (a) the DirtyCardQueueSet used to hold
506 // cards that contain references that point into the collection set
507 // to (b) the DCQS used to hold the deferred RS updates
508 _g1->dirty_card_queue_set().merge_bufferlists(&into_cset_dcqs);
509 } else {
511 CardTableModRefBS* bs = (CardTableModRefBS*)_g1->barrier_set();
512 UpdateRSetCardTableEntryIntoCSetClosure update_rs_cset_immediate(_g1, bs);
514 int n_completed_buffers = 0;
515 while (into_cset_dcqs.apply_closure_to_completed_buffer(&update_rs_cset_immediate,
516 0, 0, true)) {
517 n_completed_buffers++;
518 }
519 assert(n_completed_buffers == into_cset_n_buffers, "missed some buffers");
520 }
521 }
523 // Free any completed buffers in the DirtyCardQueueSet used to hold cards
524 // which contain references that point into the collection.
525 _g1->into_cset_dirty_card_queue_set().clear();
526 assert(_g1->into_cset_dirty_card_queue_set().completed_buffers_num() == 0,
527 "all buffers should be freed");
528 _g1->into_cset_dirty_card_queue_set().clear_n_completed_buffers();
529 }
531 class ScrubRSClosure: public HeapRegionClosure {
532 G1CollectedHeap* _g1h;
533 BitMap* _region_bm;
534 BitMap* _card_bm;
535 CardTableModRefBS* _ctbs;
536 public:
537 ScrubRSClosure(BitMap* region_bm, BitMap* card_bm) :
538 _g1h(G1CollectedHeap::heap()),
539 _region_bm(region_bm), _card_bm(card_bm),
540 _ctbs(NULL)
541 {
542 ModRefBarrierSet* bs = _g1h->mr_bs();
543 guarantee(bs->is_a(BarrierSet::CardTableModRef), "Precondition");
544 _ctbs = (CardTableModRefBS*)bs;
545 }
547 bool doHeapRegion(HeapRegion* r) {
548 if (!r->continuesHumongous()) {
549 r->rem_set()->scrub(_ctbs, _region_bm, _card_bm);
550 }
551 return false;
552 }
553 };
555 void G1RemSet::scrub(BitMap* region_bm, BitMap* card_bm) {
556 ScrubRSClosure scrub_cl(region_bm, card_bm);
557 _g1->heap_region_iterate(&scrub_cl);
558 }
560 void G1RemSet::scrub_par(BitMap* region_bm, BitMap* card_bm,
561 uint worker_num, int claim_val) {
562 ScrubRSClosure scrub_cl(region_bm, card_bm);
563 _g1->heap_region_par_iterate_chunked(&scrub_cl,
564 worker_num,
565 n_workers(),
566 claim_val);
567 }
570 static IntHistogram out_of_histo(50, 50);
572 class TriggerClosure : public OopClosure {
573 bool _trigger;
574 public:
575 TriggerClosure() : _trigger(false) { }
576 bool value() const { return _trigger; }
577 template <class T> void do_oop_nv(T* p) { _trigger = true; }
578 virtual void do_oop(oop* p) { do_oop_nv(p); }
579 virtual void do_oop(narrowOop* p) { do_oop_nv(p); }
580 };
582 class InvokeIfNotTriggeredClosure: public OopClosure {
583 TriggerClosure* _t;
584 OopClosure* _oc;
585 public:
586 InvokeIfNotTriggeredClosure(TriggerClosure* t, OopClosure* oc):
587 _t(t), _oc(oc) { }
588 template <class T> void do_oop_nv(T* p) {
589 if (!_t->value()) _oc->do_oop(p);
590 }
591 virtual void do_oop(oop* p) { do_oop_nv(p); }
592 virtual void do_oop(narrowOop* p) { do_oop_nv(p); }
593 };
595 class Mux2Closure : public OopClosure {
596 OopClosure* _c1;
597 OopClosure* _c2;
598 public:
599 Mux2Closure(OopClosure *c1, OopClosure *c2) : _c1(c1), _c2(c2) { }
600 template <class T> void do_oop_nv(T* p) {
601 _c1->do_oop(p); _c2->do_oop(p);
602 }
603 virtual void do_oop(oop* p) { do_oop_nv(p); }
604 virtual void do_oop(narrowOop* p) { do_oop_nv(p); }
605 };
607 bool G1RemSet::concurrentRefineOneCard_impl(jbyte* card_ptr, int worker_i,
608 bool check_for_refs_into_cset) {
609 // Construct the region representing the card.
610 HeapWord* start = _ct_bs->addr_for(card_ptr);
611 // And find the region containing it.
612 HeapRegion* r = _g1->heap_region_containing(start);
613 assert(r != NULL, "unexpected null");
615 HeapWord* end = _ct_bs->addr_for(card_ptr + 1);
616 MemRegion dirtyRegion(start, end);
618 #if CARD_REPEAT_HISTO
619 init_ct_freq_table(_g1->max_capacity());
620 ct_freq_note_card(_ct_bs->index_for(start));
621 #endif
623 OopsInHeapRegionClosure* oops_in_heap_closure = NULL;
624 if (check_for_refs_into_cset) {
625 // ConcurrentG1RefineThreads have worker numbers larger than what
626 // _cset_rs_update_cl[] is set up to handle. But those threads should
627 // only be active outside of a collection which means that when they
628 // reach here they should have check_for_refs_into_cset == false.
629 assert((size_t)worker_i < n_workers(), "index of worker larger than _cset_rs_update_cl[].length");
630 oops_in_heap_closure = _cset_rs_update_cl[worker_i];
631 }
632 UpdateRSOrPushRefOopClosure update_rs_oop_cl(_g1,
633 _g1->g1_rem_set(),
634 oops_in_heap_closure,
635 check_for_refs_into_cset,
636 worker_i);
637 update_rs_oop_cl.set_from(r);
639 TriggerClosure trigger_cl;
640 FilterIntoCSClosure into_cs_cl(NULL, _g1, &trigger_cl);
641 InvokeIfNotTriggeredClosure invoke_cl(&trigger_cl, &into_cs_cl);
642 Mux2Closure mux(&invoke_cl, &update_rs_oop_cl);
644 FilterOutOfRegionClosure filter_then_update_rs_oop_cl(r,
645 (check_for_refs_into_cset ?
646 (OopClosure*)&mux :
647 (OopClosure*)&update_rs_oop_cl));
649 // The region for the current card may be a young region. The
650 // current card may have been a card that was evicted from the
651 // card cache. When the card was inserted into the cache, we had
652 // determined that its region was non-young. While in the cache,
653 // the region may have been freed during a cleanup pause, reallocated
654 // and tagged as young.
655 //
656 // We wish to filter out cards for such a region but the current
657 // thread, if we're running concurrently, may "see" the young type
658 // change at any time (so an earlier "is_young" check may pass or
659 // fail arbitrarily). We tell the iteration code to perform this
660 // filtering when it has been determined that there has been an actual
661 // allocation in this region and making it safe to check the young type.
662 bool filter_young = true;
664 HeapWord* stop_point =
665 r->oops_on_card_seq_iterate_careful(dirtyRegion,
666 &filter_then_update_rs_oop_cl,
667 filter_young,
668 card_ptr);
670 // If stop_point is non-null, then we encountered an unallocated region
671 // (perhaps the unfilled portion of a TLAB.) For now, we'll dirty the
672 // card and re-enqueue: if we put off the card until a GC pause, then the
673 // unallocated portion will be filled in. Alternatively, we might try
674 // the full complexity of the technique used in "regular" precleaning.
675 if (stop_point != NULL) {
676 // The card might have gotten re-dirtied and re-enqueued while we
677 // worked. (In fact, it's pretty likely.)
678 if (*card_ptr != CardTableModRefBS::dirty_card_val()) {
679 *card_ptr = CardTableModRefBS::dirty_card_val();
680 MutexLockerEx x(Shared_DirtyCardQ_lock,
681 Mutex::_no_safepoint_check_flag);
682 DirtyCardQueue* sdcq =
683 JavaThread::dirty_card_queue_set().shared_dirty_card_queue();
684 sdcq->enqueue(card_ptr);
685 }
686 } else {
687 out_of_histo.add_entry(filter_then_update_rs_oop_cl.out_of_region());
688 _conc_refine_cards++;
689 }
691 return trigger_cl.value();
692 }
694 bool G1RemSet::concurrentRefineOneCard(jbyte* card_ptr, int worker_i,
695 bool check_for_refs_into_cset) {
696 // If the card is no longer dirty, nothing to do.
697 if (*card_ptr != CardTableModRefBS::dirty_card_val()) {
698 // No need to return that this card contains refs that point
699 // into the collection set.
700 return false;
701 }
703 // Construct the region representing the card.
704 HeapWord* start = _ct_bs->addr_for(card_ptr);
705 // And find the region containing it.
706 HeapRegion* r = _g1->heap_region_containing(start);
707 if (r == NULL) {
708 guarantee(_g1->is_in_permanent(start), "Or else where?");
709 // Again no need to return that this card contains refs that
710 // point into the collection set.
711 return false; // Not in the G1 heap (might be in perm, for example.)
712 }
713 // Why do we have to check here whether a card is on a young region,
714 // given that we dirty young regions and, as a result, the
715 // post-barrier is supposed to filter them out and never to enqueue
716 // them? When we allocate a new region as the "allocation region" we
717 // actually dirty its cards after we release the lock, since card
718 // dirtying while holding the lock was a performance bottleneck. So,
719 // as a result, it is possible for other threads to actually
720 // allocate objects in the region (after the acquire the lock)
721 // before all the cards on the region are dirtied. This is unlikely,
722 // and it doesn't happen often, but it can happen. So, the extra
723 // check below filters out those cards.
724 if (r->is_young()) {
725 return false;
726 }
727 // While we are processing RSet buffers during the collection, we
728 // actually don't want to scan any cards on the collection set,
729 // since we don't want to update remebered sets with entries that
730 // point into the collection set, given that live objects from the
731 // collection set are about to move and such entries will be stale
732 // very soon. This change also deals with a reliability issue which
733 // involves scanning a card in the collection set and coming across
734 // an array that was being chunked and looking malformed. Note,
735 // however, that if evacuation fails, we have to scan any objects
736 // that were not moved and create any missing entries.
737 if (r->in_collection_set()) {
738 return false;
739 }
741 // Should we defer processing the card?
742 //
743 // Previously the result from the insert_cache call would be
744 // either card_ptr (implying that card_ptr was currently "cold"),
745 // null (meaning we had inserted the card ptr into the "hot"
746 // cache, which had some headroom), or a "hot" card ptr
747 // extracted from the "hot" cache.
748 //
749 // Now that the _card_counts cache in the ConcurrentG1Refine
750 // instance is an evicting hash table, the result we get back
751 // could be from evicting the card ptr in an already occupied
752 // bucket (in which case we have replaced the card ptr in the
753 // bucket with card_ptr and "defer" is set to false). To avoid
754 // having a data structure (updates to which would need a lock)
755 // to hold these unprocessed dirty cards, we need to immediately
756 // process card_ptr. The actions needed to be taken on return
757 // from cache_insert are summarized in the following table:
758 //
759 // res defer action
760 // --------------------------------------------------------------
761 // null false card evicted from _card_counts & replaced with
762 // card_ptr; evicted ptr added to hot cache.
763 // No need to process res; immediately process card_ptr
764 //
765 // null true card not evicted from _card_counts; card_ptr added
766 // to hot cache.
767 // Nothing to do.
768 //
769 // non-null false card evicted from _card_counts & replaced with
770 // card_ptr; evicted ptr is currently "cold" or
771 // caused an eviction from the hot cache.
772 // Immediately process res; process card_ptr.
773 //
774 // non-null true card not evicted from _card_counts; card_ptr is
775 // currently cold, or caused an eviction from hot
776 // cache.
777 // Immediately process res; no need to process card_ptr.
780 jbyte* res = card_ptr;
781 bool defer = false;
783 // This gets set to true if the card being refined has references
784 // that point into the collection set.
785 bool oops_into_cset = false;
787 if (_cg1r->use_cache()) {
788 jbyte* res = _cg1r->cache_insert(card_ptr, &defer);
789 if (res != NULL && (res != card_ptr || defer)) {
790 start = _ct_bs->addr_for(res);
791 r = _g1->heap_region_containing(start);
792 if (r == NULL) {
793 assert(_g1->is_in_permanent(start), "Or else where?");
794 } else {
795 // Checking whether the region we got back from the cache
796 // is young here is inappropriate. The region could have been
797 // freed, reallocated and tagged as young while in the cache.
798 // Hence we could see its young type change at any time.
799 //
800 // Process card pointer we get back from the hot card cache. This
801 // will check whether the region containing the card is young
802 // _after_ checking that the region has been allocated from.
803 oops_into_cset = concurrentRefineOneCard_impl(res, worker_i,
804 false /* check_for_refs_into_cset */);
805 // The above call to concurrentRefineOneCard_impl is only
806 // performed if the hot card cache is enabled. This cache is
807 // disabled during an evacuation pause - which is the only
808 // time when we need know if the card contains references
809 // that point into the collection set. Also when the hot card
810 // cache is enabled, this code is executed by the concurrent
811 // refine threads - rather than the GC worker threads - and
812 // concurrentRefineOneCard_impl will return false.
813 assert(!oops_into_cset, "should not see true here");
814 }
815 }
816 }
818 if (!defer) {
819 oops_into_cset =
820 concurrentRefineOneCard_impl(card_ptr, worker_i, check_for_refs_into_cset);
821 // We should only be detecting that the card contains references
822 // that point into the collection set if the current thread is
823 // a GC worker thread.
824 assert(!oops_into_cset || SafepointSynchronize::is_at_safepoint(),
825 "invalid result at non safepoint");
826 }
827 return oops_into_cset;
828 }
830 class HRRSStatsIter: public HeapRegionClosure {
831 size_t _occupied;
832 size_t _total_mem_sz;
833 size_t _max_mem_sz;
834 HeapRegion* _max_mem_sz_region;
835 public:
836 HRRSStatsIter() :
837 _occupied(0),
838 _total_mem_sz(0),
839 _max_mem_sz(0),
840 _max_mem_sz_region(NULL)
841 {}
843 bool doHeapRegion(HeapRegion* r) {
844 if (r->continuesHumongous()) return false;
845 size_t mem_sz = r->rem_set()->mem_size();
846 if (mem_sz > _max_mem_sz) {
847 _max_mem_sz = mem_sz;
848 _max_mem_sz_region = r;
849 }
850 _total_mem_sz += mem_sz;
851 size_t occ = r->rem_set()->occupied();
852 _occupied += occ;
853 return false;
854 }
855 size_t total_mem_sz() { return _total_mem_sz; }
856 size_t max_mem_sz() { return _max_mem_sz; }
857 size_t occupied() { return _occupied; }
858 HeapRegion* max_mem_sz_region() { return _max_mem_sz_region; }
859 };
861 class PrintRSThreadVTimeClosure : public ThreadClosure {
862 public:
863 virtual void do_thread(Thread *t) {
864 ConcurrentG1RefineThread* crt = (ConcurrentG1RefineThread*) t;
865 gclog_or_tty->print(" %5.2f", crt->vtime_accum());
866 }
867 };
869 void G1RemSet::print_summary_info() {
870 G1CollectedHeap* g1 = G1CollectedHeap::heap();
872 #if CARD_REPEAT_HISTO
873 gclog_or_tty->print_cr("\nG1 card_repeat count histogram: ");
874 gclog_or_tty->print_cr(" # of repeats --> # of cards with that number.");
875 card_repeat_count.print_on(gclog_or_tty);
876 #endif
878 if (FILTEROUTOFREGIONCLOSURE_DOHISTOGRAMCOUNT) {
879 gclog_or_tty->print_cr("\nG1 rem-set out-of-region histogram: ");
880 gclog_or_tty->print_cr(" # of CS ptrs --> # of cards with that number.");
881 out_of_histo.print_on(gclog_or_tty);
882 }
883 gclog_or_tty->print_cr("\n Concurrent RS processed %d cards",
884 _conc_refine_cards);
885 DirtyCardQueueSet& dcqs = JavaThread::dirty_card_queue_set();
886 jint tot_processed_buffers =
887 dcqs.processed_buffers_mut() + dcqs.processed_buffers_rs_thread();
888 gclog_or_tty->print_cr(" Of %d completed buffers:", tot_processed_buffers);
889 gclog_or_tty->print_cr(" %8d (%5.1f%%) by conc RS threads.",
890 dcqs.processed_buffers_rs_thread(),
891 100.0*(float)dcqs.processed_buffers_rs_thread()/
892 (float)tot_processed_buffers);
893 gclog_or_tty->print_cr(" %8d (%5.1f%%) by mutator threads.",
894 dcqs.processed_buffers_mut(),
895 100.0*(float)dcqs.processed_buffers_mut()/
896 (float)tot_processed_buffers);
897 gclog_or_tty->print_cr(" Conc RS threads times(s)");
898 PrintRSThreadVTimeClosure p;
899 gclog_or_tty->print(" ");
900 g1->concurrent_g1_refine()->threads_do(&p);
901 gclog_or_tty->print_cr("");
903 HRRSStatsIter blk;
904 g1->heap_region_iterate(&blk);
905 gclog_or_tty->print_cr(" Total heap region rem set sizes = " SIZE_FORMAT "K."
906 " Max = " SIZE_FORMAT "K.",
907 blk.total_mem_sz()/K, blk.max_mem_sz()/K);
908 gclog_or_tty->print_cr(" Static structures = " SIZE_FORMAT "K,"
909 " free_lists = " SIZE_FORMAT "K.",
910 HeapRegionRemSet::static_mem_size()/K,
911 HeapRegionRemSet::fl_mem_size()/K);
912 gclog_or_tty->print_cr(" %d occupied cards represented.",
913 blk.occupied());
914 gclog_or_tty->print_cr(" Max sz region = [" PTR_FORMAT ", " PTR_FORMAT " )"
915 ", cap = " SIZE_FORMAT "K, occ = " SIZE_FORMAT "K.",
916 blk.max_mem_sz_region()->bottom(), blk.max_mem_sz_region()->end(),
917 (blk.max_mem_sz_region()->rem_set()->mem_size() + K - 1)/K,
918 (blk.max_mem_sz_region()->rem_set()->occupied() + K - 1)/K);
919 gclog_or_tty->print_cr(" Did %d coarsenings.", HeapRegionRemSet::n_coarsenings());
920 }
922 void G1RemSet::prepare_for_verify() {
923 if (G1HRRSFlushLogBuffersOnVerify &&
924 (VerifyBeforeGC || VerifyAfterGC)
925 && !_g1->full_collection()) {
926 cleanupHRRS();
927 _g1->set_refine_cte_cl_concurrency(false);
928 if (SafepointSynchronize::is_at_safepoint()) {
929 DirtyCardQueueSet& dcqs = JavaThread::dirty_card_queue_set();
930 dcqs.concatenate_logs();
931 }
932 bool cg1r_use_cache = _cg1r->use_cache();
933 _cg1r->set_use_cache(false);
934 DirtyCardQueue into_cset_dcq(&_g1->into_cset_dirty_card_queue_set());
935 updateRS(&into_cset_dcq, 0);
936 _g1->into_cset_dirty_card_queue_set().clear();
937 _cg1r->set_use_cache(cg1r_use_cache);
939 assert(JavaThread::dirty_card_queue_set().completed_buffers_num() == 0, "All should be consumed");
940 }
941 }