Mon, 11 May 2009 16:30:56 -0700
6484957: G1: parallel concurrent refinement
6826318: G1: remove traversal-based refinement code
Summary: Removed traversal-based refinement code as it's no longer used. Made the concurrent refinement (queue-based) parallel.
Reviewed-by: tonyp
1 /*
2 * Copyright 2001-2009 Sun Microsystems, Inc. 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 Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
20 * CA 95054 USA or visit www.sun.com if you need additional information or
21 * have any questions.
22 *
23 */
25 #include "incls/_precompiled.incl"
26 #include "incls/_g1RemSet.cpp.incl"
28 #define CARD_REPEAT_HISTO 0
30 #if CARD_REPEAT_HISTO
31 static size_t ct_freq_sz;
32 static jbyte* ct_freq = NULL;
34 void init_ct_freq_table(size_t heap_sz_bytes) {
35 if (ct_freq == NULL) {
36 ct_freq_sz = heap_sz_bytes/CardTableModRefBS::card_size;
37 ct_freq = new jbyte[ct_freq_sz];
38 for (size_t j = 0; j < ct_freq_sz; j++) ct_freq[j] = 0;
39 }
40 }
42 void ct_freq_note_card(size_t index) {
43 assert(0 <= index && index < ct_freq_sz, "Bounds error.");
44 if (ct_freq[index] < 100) { ct_freq[index]++; }
45 }
47 static IntHistogram card_repeat_count(10, 10);
49 void ct_freq_update_histo_and_reset() {
50 for (size_t j = 0; j < ct_freq_sz; j++) {
51 card_repeat_count.add_entry(ct_freq[j]);
52 ct_freq[j] = 0;
53 }
55 }
56 #endif
59 class IntoCSOopClosure: public OopsInHeapRegionClosure {
60 OopsInHeapRegionClosure* _blk;
61 G1CollectedHeap* _g1;
62 public:
63 IntoCSOopClosure(G1CollectedHeap* g1, OopsInHeapRegionClosure* blk) :
64 _g1(g1), _blk(blk) {}
65 void set_region(HeapRegion* from) {
66 _blk->set_region(from);
67 }
68 virtual void do_oop(narrowOop* p) {
69 guarantee(false, "NYI");
70 }
71 virtual void do_oop(oop* p) {
72 oop obj = *p;
73 if (_g1->obj_in_cs(obj)) _blk->do_oop(p);
74 }
75 bool apply_to_weak_ref_discovered_field() { return true; }
76 bool idempotent() { return true; }
77 };
79 class IntoCSRegionClosure: public HeapRegionClosure {
80 IntoCSOopClosure _blk;
81 G1CollectedHeap* _g1;
82 public:
83 IntoCSRegionClosure(G1CollectedHeap* g1, OopsInHeapRegionClosure* blk) :
84 _g1(g1), _blk(g1, blk) {}
85 bool doHeapRegion(HeapRegion* r) {
86 if (!r->in_collection_set()) {
87 _blk.set_region(r);
88 if (r->isHumongous()) {
89 if (r->startsHumongous()) {
90 oop obj = oop(r->bottom());
91 obj->oop_iterate(&_blk);
92 }
93 } else {
94 r->oop_before_save_marks_iterate(&_blk);
95 }
96 }
97 return false;
98 }
99 };
101 void
102 StupidG1RemSet::oops_into_collection_set_do(OopsInHeapRegionClosure* oc,
103 int worker_i) {
104 IntoCSRegionClosure rc(_g1, oc);
105 _g1->heap_region_iterate(&rc);
106 }
108 class VerifyRSCleanCardOopClosure: public OopClosure {
109 G1CollectedHeap* _g1;
110 public:
111 VerifyRSCleanCardOopClosure(G1CollectedHeap* g1) : _g1(g1) {}
113 virtual void do_oop(narrowOop* p) {
114 guarantee(false, "NYI");
115 }
116 virtual void do_oop(oop* p) {
117 oop obj = *p;
118 HeapRegion* to = _g1->heap_region_containing(obj);
119 guarantee(to == NULL || !to->in_collection_set(),
120 "Missed a rem set member.");
121 }
122 };
124 HRInto_G1RemSet::HRInto_G1RemSet(G1CollectedHeap* g1, CardTableModRefBS* ct_bs)
125 : G1RemSet(g1), _ct_bs(ct_bs), _g1p(_g1->g1_policy()),
126 _cg1r(g1->concurrent_g1_refine()),
127 _par_traversal_in_progress(false), _new_refs(NULL),
128 _cards_scanned(NULL), _total_cards_scanned(0)
129 {
130 _seq_task = new SubTasksDone(NumSeqTasks);
131 guarantee(n_workers() > 0, "There should be some workers");
132 _new_refs = NEW_C_HEAP_ARRAY(GrowableArray<oop*>*, n_workers());
133 for (uint i = 0; i < n_workers(); i++) {
134 _new_refs[i] = new (ResourceObj::C_HEAP) GrowableArray<oop*>(8192,true);
135 }
136 }
138 HRInto_G1RemSet::~HRInto_G1RemSet() {
139 delete _seq_task;
140 for (uint i = 0; i < n_workers(); i++) {
141 delete _new_refs[i];
142 }
143 FREE_C_HEAP_ARRAY(GrowableArray<oop*>*, _new_refs);
144 }
146 void CountNonCleanMemRegionClosure::do_MemRegion(MemRegion mr) {
147 if (_g1->is_in_g1_reserved(mr.start())) {
148 _n += (int) ((mr.byte_size() / CardTableModRefBS::card_size));
149 if (_start_first == NULL) _start_first = mr.start();
150 }
151 }
153 class ScanRSClosure : public HeapRegionClosure {
154 size_t _cards_done, _cards;
155 G1CollectedHeap* _g1h;
156 OopsInHeapRegionClosure* _oc;
157 G1BlockOffsetSharedArray* _bot_shared;
158 CardTableModRefBS *_ct_bs;
159 int _worker_i;
160 bool _try_claimed;
161 size_t _min_skip_distance, _max_skip_distance;
162 public:
163 ScanRSClosure(OopsInHeapRegionClosure* oc, int worker_i) :
164 _oc(oc),
165 _cards(0),
166 _cards_done(0),
167 _worker_i(worker_i),
168 _try_claimed(false)
169 {
170 _g1h = G1CollectedHeap::heap();
171 _bot_shared = _g1h->bot_shared();
172 _ct_bs = (CardTableModRefBS*) (_g1h->barrier_set());
173 _min_skip_distance = 16;
174 _max_skip_distance = 2 * _g1h->n_par_threads() * _min_skip_distance;
175 }
177 void set_try_claimed() { _try_claimed = true; }
179 void scanCard(size_t index, HeapRegion *r) {
180 _cards_done++;
181 DirtyCardToOopClosure* cl =
182 r->new_dcto_closure(_oc,
183 CardTableModRefBS::Precise,
184 HeapRegionDCTOC::IntoCSFilterKind);
186 // Set the "from" region in the closure.
187 _oc->set_region(r);
188 HeapWord* card_start = _bot_shared->address_for_index(index);
189 HeapWord* card_end = card_start + G1BlockOffsetSharedArray::N_words;
190 Space *sp = SharedHeap::heap()->space_containing(card_start);
191 MemRegion sm_region;
192 if (ParallelGCThreads > 0) {
193 // first find the used area
194 sm_region = sp->used_region_at_save_marks();
195 } else {
196 // The closure is not idempotent. We shouldn't look at objects
197 // allocated during the GC.
198 sm_region = sp->used_region_at_save_marks();
199 }
200 MemRegion mr = sm_region.intersection(MemRegion(card_start,card_end));
201 if (!mr.is_empty()) {
202 cl->do_MemRegion(mr);
203 }
204 }
206 void printCard(HeapRegion* card_region, size_t card_index,
207 HeapWord* card_start) {
208 gclog_or_tty->print_cr("T %d Region [" PTR_FORMAT ", " PTR_FORMAT ") "
209 "RS names card %p: "
210 "[" PTR_FORMAT ", " PTR_FORMAT ")",
211 _worker_i,
212 card_region->bottom(), card_region->end(),
213 card_index,
214 card_start, card_start + G1BlockOffsetSharedArray::N_words);
215 }
217 bool doHeapRegion(HeapRegion* r) {
218 assert(r->in_collection_set(), "should only be called on elements of CS.");
219 HeapRegionRemSet* hrrs = r->rem_set();
220 if (hrrs->iter_is_complete()) return false; // All done.
221 if (!_try_claimed && !hrrs->claim_iter()) return false;
222 // If we didn't return above, then
223 // _try_claimed || r->claim_iter()
224 // is true: either we're supposed to work on claimed-but-not-complete
225 // regions, or we successfully claimed the region.
226 HeapRegionRemSetIterator* iter = _g1h->rem_set_iterator(_worker_i);
227 hrrs->init_iterator(iter);
228 size_t card_index;
229 size_t skip_distance = 0, current_card = 0, jump_to_card = 0;
230 while (iter->has_next(card_index)) {
231 if (current_card < jump_to_card) {
232 ++current_card;
233 continue;
234 }
235 HeapWord* card_start = _g1h->bot_shared()->address_for_index(card_index);
236 #if 0
237 gclog_or_tty->print("Rem set iteration yielded card [" PTR_FORMAT ", " PTR_FORMAT ").\n",
238 card_start, card_start + CardTableModRefBS::card_size_in_words);
239 #endif
241 HeapRegion* card_region = _g1h->heap_region_containing(card_start);
242 assert(card_region != NULL, "Yielding cards not in the heap?");
243 _cards++;
245 // If the card is dirty, then we will scan it during updateRS.
246 if (!card_region->in_collection_set() && !_ct_bs->is_card_dirty(card_index)) {
247 if (!_ct_bs->is_card_claimed(card_index) && _ct_bs->claim_card(card_index)) {
248 scanCard(card_index, card_region);
249 } else if (_try_claimed) {
250 if (jump_to_card == 0 || jump_to_card != current_card) {
251 // We did some useful work in the previous iteration.
252 // Decrease the distance.
253 skip_distance = MAX2(skip_distance >> 1, _min_skip_distance);
254 } else {
255 // Previous iteration resulted in a claim failure.
256 // Increase the distance.
257 skip_distance = MIN2(skip_distance << 1, _max_skip_distance);
258 }
259 jump_to_card = current_card + skip_distance;
260 }
261 }
262 ++current_card;
263 }
264 if (!_try_claimed) {
265 hrrs->set_iter_complete();
266 }
267 return false;
268 }
269 // Set all cards back to clean.
270 void cleanup() {_g1h->cleanUpCardTable();}
271 size_t cards_done() { return _cards_done;}
272 size_t cards_looked_up() { return _cards;}
273 };
275 // We want the parallel threads to start their scanning at
276 // different collection set regions to avoid contention.
277 // If we have:
278 // n collection set regions
279 // p threads
280 // Then thread t will start at region t * floor (n/p)
282 HeapRegion* HRInto_G1RemSet::calculateStartRegion(int worker_i) {
283 HeapRegion* result = _g1p->collection_set();
284 if (ParallelGCThreads > 0) {
285 size_t cs_size = _g1p->collection_set_size();
286 int n_workers = _g1->workers()->total_workers();
287 size_t cs_spans = cs_size / n_workers;
288 size_t ind = cs_spans * worker_i;
289 for (size_t i = 0; i < ind; i++)
290 result = result->next_in_collection_set();
291 }
292 return result;
293 }
295 void HRInto_G1RemSet::scanRS(OopsInHeapRegionClosure* oc, int worker_i) {
296 double rs_time_start = os::elapsedTime();
297 HeapRegion *startRegion = calculateStartRegion(worker_i);
299 BufferingOopsInHeapRegionClosure boc(oc);
300 ScanRSClosure scanRScl(&boc, worker_i);
301 _g1->collection_set_iterate_from(startRegion, &scanRScl);
302 scanRScl.set_try_claimed();
303 _g1->collection_set_iterate_from(startRegion, &scanRScl);
305 boc.done();
306 double closure_app_time_sec = boc.closure_app_seconds();
307 double scan_rs_time_sec = (os::elapsedTime() - rs_time_start) -
308 closure_app_time_sec;
309 double closure_app_time_ms = closure_app_time_sec * 1000.0;
311 assert( _cards_scanned != NULL, "invariant" );
312 _cards_scanned[worker_i] = scanRScl.cards_done();
314 _g1p->record_scan_rs_start_time(worker_i, rs_time_start * 1000.0);
315 _g1p->record_scan_rs_time(worker_i, scan_rs_time_sec * 1000.0);
317 double scan_new_refs_time_ms = _g1p->get_scan_new_refs_time(worker_i);
318 if (scan_new_refs_time_ms > 0.0) {
319 closure_app_time_ms += scan_new_refs_time_ms;
320 }
322 _g1p->record_obj_copy_time(worker_i, closure_app_time_ms);
323 }
325 void HRInto_G1RemSet::updateRS(int worker_i) {
326 ConcurrentG1Refine* cg1r = _g1->concurrent_g1_refine();
328 double start = os::elapsedTime();
329 _g1p->record_update_rs_start_time(worker_i, start * 1000.0);
331 // Apply the appropriate closure to all remaining log entries.
332 _g1->iterate_dirty_card_closure(false, worker_i);
333 // Now there should be no dirty cards.
334 if (G1RSLogCheckCardTable) {
335 CountNonCleanMemRegionClosure cl(_g1);
336 _ct_bs->mod_card_iterate(&cl);
337 // XXX This isn't true any more: keeping cards of young regions
338 // marked dirty broke it. Need some reasonable fix.
339 guarantee(cl.n() == 0, "Card table should be clean.");
340 }
342 _g1p->record_update_rs_time(worker_i, (os::elapsedTime() - start) * 1000.0);
343 }
345 #ifndef PRODUCT
346 class PrintRSClosure : public HeapRegionClosure {
347 int _count;
348 public:
349 PrintRSClosure() : _count(0) {}
350 bool doHeapRegion(HeapRegion* r) {
351 HeapRegionRemSet* hrrs = r->rem_set();
352 _count += (int) hrrs->occupied();
353 if (hrrs->occupied() == 0) {
354 gclog_or_tty->print("Heap Region [" PTR_FORMAT ", " PTR_FORMAT ") "
355 "has no remset entries\n",
356 r->bottom(), r->end());
357 } else {
358 gclog_or_tty->print("Printing rem set for heap region [" PTR_FORMAT ", " PTR_FORMAT ")\n",
359 r->bottom(), r->end());
360 r->print();
361 hrrs->print();
362 gclog_or_tty->print("\nDone printing rem set\n");
363 }
364 return false;
365 }
366 int occupied() {return _count;}
367 };
368 #endif
370 class CountRSSizeClosure: public HeapRegionClosure {
371 size_t _n;
372 size_t _tot;
373 size_t _max;
374 HeapRegion* _max_r;
375 enum {
376 N = 20,
377 MIN = 6
378 };
379 int _histo[N];
380 public:
381 CountRSSizeClosure() : _n(0), _tot(0), _max(0), _max_r(NULL) {
382 for (int i = 0; i < N; i++) _histo[i] = 0;
383 }
384 bool doHeapRegion(HeapRegion* r) {
385 if (!r->continuesHumongous()) {
386 size_t occ = r->rem_set()->occupied();
387 _n++;
388 _tot += occ;
389 if (occ > _max) {
390 _max = occ;
391 _max_r = r;
392 }
393 // Fit it into a histo bin.
394 int s = 1 << MIN;
395 int i = 0;
396 while (occ > (size_t) s && i < (N-1)) {
397 s = s << 1;
398 i++;
399 }
400 _histo[i]++;
401 }
402 return false;
403 }
404 size_t n() { return _n; }
405 size_t tot() { return _tot; }
406 size_t mx() { return _max; }
407 HeapRegion* mxr() { return _max_r; }
408 void print_histo() {
409 int mx = N;
410 while (mx >= 0) {
411 if (_histo[mx-1] > 0) break;
412 mx--;
413 }
414 gclog_or_tty->print_cr("Number of regions with given RS sizes:");
415 gclog_or_tty->print_cr(" <= %8d %8d", 1 << MIN, _histo[0]);
416 for (int i = 1; i < mx-1; i++) {
417 gclog_or_tty->print_cr(" %8d - %8d %8d",
418 (1 << (MIN + i - 1)) + 1,
419 1 << (MIN + i),
420 _histo[i]);
421 }
422 gclog_or_tty->print_cr(" > %8d %8d", (1 << (MIN+mx-2))+1, _histo[mx-1]);
423 }
424 };
426 void
427 HRInto_G1RemSet::scanNewRefsRS(OopsInHeapRegionClosure* oc,
428 int worker_i) {
429 double scan_new_refs_start_sec = os::elapsedTime();
430 G1CollectedHeap* g1h = G1CollectedHeap::heap();
431 CardTableModRefBS* ct_bs = (CardTableModRefBS*) (g1h->barrier_set());
432 for (int i = 0; i < _new_refs[worker_i]->length(); i++) {
433 oop* p = _new_refs[worker_i]->at(i);
434 oop obj = *p;
435 // *p was in the collection set when p was pushed on "_new_refs", but
436 // another thread may have processed this location from an RS, so it
437 // might not point into the CS any longer. If so, it's obviously been
438 // processed, and we don't need to do anything further.
439 if (g1h->obj_in_cs(obj)) {
440 HeapRegion* r = g1h->heap_region_containing(p);
442 DEBUG_ONLY(HeapRegion* to = g1h->heap_region_containing(obj));
443 oc->set_region(r);
444 // If "p" has already been processed concurrently, this is
445 // idempotent.
446 oc->do_oop(p);
447 }
448 }
449 _g1p->record_scan_new_refs_time(worker_i,
450 (os::elapsedTime() - scan_new_refs_start_sec)
451 * 1000.0);
452 }
454 void HRInto_G1RemSet::cleanupHRRS() {
455 HeapRegionRemSet::cleanup();
456 }
458 void
459 HRInto_G1RemSet::oops_into_collection_set_do(OopsInHeapRegionClosure* oc,
460 int worker_i) {
461 #if CARD_REPEAT_HISTO
462 ct_freq_update_histo_and_reset();
463 #endif
464 if (worker_i == 0) {
465 _cg1r->clear_and_record_card_counts();
466 }
468 // Make this into a command-line flag...
469 if (G1RSCountHisto && (ParallelGCThreads == 0 || worker_i == 0)) {
470 CountRSSizeClosure count_cl;
471 _g1->heap_region_iterate(&count_cl);
472 gclog_or_tty->print_cr("Avg of %d RS counts is %f, max is %d, "
473 "max region is " PTR_FORMAT,
474 count_cl.n(), (float)count_cl.tot()/(float)count_cl.n(),
475 count_cl.mx(), count_cl.mxr());
476 count_cl.print_histo();
477 }
479 if (ParallelGCThreads > 0) {
480 // The two flags below were introduced temporarily to serialize
481 // the updating and scanning of remembered sets. There are some
482 // race conditions when these two operations are done in parallel
483 // and they are causing failures. When we resolve said race
484 // conditions, we'll revert back to parallel remembered set
485 // updating and scanning. See CRs 6677707 and 6677708.
486 if (G1ParallelRSetUpdatingEnabled || (worker_i == 0)) {
487 updateRS(worker_i);
488 scanNewRefsRS(oc, worker_i);
489 } else {
490 _g1p->record_update_rs_start_time(worker_i, os::elapsedTime() * 1000.0);
491 _g1p->record_update_rs_processed_buffers(worker_i, 0.0);
492 _g1p->record_update_rs_time(worker_i, 0.0);
493 _g1p->record_scan_new_refs_time(worker_i, 0.0);
494 }
495 if (G1ParallelRSetScanningEnabled || (worker_i == 0)) {
496 scanRS(oc, worker_i);
497 } else {
498 _g1p->record_scan_rs_start_time(worker_i, os::elapsedTime() * 1000.0);
499 _g1p->record_scan_rs_time(worker_i, 0.0);
500 }
501 } else {
502 assert(worker_i == 0, "invariant");
503 updateRS(0);
504 scanNewRefsRS(oc, 0);
505 scanRS(oc, 0);
506 }
507 }
509 void HRInto_G1RemSet::
510 prepare_for_oops_into_collection_set_do() {
511 #if G1_REM_SET_LOGGING
512 PrintRSClosure cl;
513 _g1->collection_set_iterate(&cl);
514 #endif
515 cleanupHRRS();
516 ConcurrentG1Refine* cg1r = _g1->concurrent_g1_refine();
517 _g1->set_refine_cte_cl_concurrency(false);
518 DirtyCardQueueSet& dcqs = JavaThread::dirty_card_queue_set();
519 dcqs.concatenate_logs();
521 assert(!_par_traversal_in_progress, "Invariant between iterations.");
522 if (ParallelGCThreads > 0) {
523 set_par_traversal(true);
524 _seq_task->set_par_threads((int)n_workers());
525 }
526 guarantee( _cards_scanned == NULL, "invariant" );
527 _cards_scanned = NEW_C_HEAP_ARRAY(size_t, n_workers());
528 for (uint i = 0; i < n_workers(); ++i) {
529 _cards_scanned[i] = 0;
530 }
531 _total_cards_scanned = 0;
532 }
535 class cleanUpIteratorsClosure : public HeapRegionClosure {
536 bool doHeapRegion(HeapRegion *r) {
537 HeapRegionRemSet* hrrs = r->rem_set();
538 hrrs->init_for_par_iteration();
539 return false;
540 }
541 };
543 class UpdateRSetOopsIntoCSImmediate : public OopClosure {
544 G1CollectedHeap* _g1;
545 public:
546 UpdateRSetOopsIntoCSImmediate(G1CollectedHeap* g1) : _g1(g1) { }
547 virtual void do_oop(narrowOop* p) {
548 guarantee(false, "NYI");
549 }
550 virtual void do_oop(oop* p) {
551 HeapRegion* to = _g1->heap_region_containing(*p);
552 if (to->in_collection_set()) {
553 to->rem_set()->add_reference(p, 0);
554 }
555 }
556 };
558 class UpdateRSetOopsIntoCSDeferred : public OopClosure {
559 G1CollectedHeap* _g1;
560 CardTableModRefBS* _ct_bs;
561 DirtyCardQueue* _dcq;
562 public:
563 UpdateRSetOopsIntoCSDeferred(G1CollectedHeap* g1, DirtyCardQueue* dcq) :
564 _g1(g1), _ct_bs((CardTableModRefBS*)_g1->barrier_set()), _dcq(dcq) { }
565 virtual void do_oop(narrowOop* p) {
566 guarantee(false, "NYI");
567 }
568 virtual void do_oop(oop* p) {
569 oop obj = *p;
570 if (_g1->obj_in_cs(obj)) {
571 size_t card_index = _ct_bs->index_for(p);
572 if (_ct_bs->mark_card_deferred(card_index)) {
573 _dcq->enqueue((jbyte*)_ct_bs->byte_for_index(card_index));
574 }
575 }
576 }
577 };
579 void HRInto_G1RemSet::new_refs_iterate(OopClosure* cl) {
580 for (size_t i = 0; i < n_workers(); i++) {
581 for (int j = 0; j < _new_refs[i]->length(); j++) {
582 oop* p = _new_refs[i]->at(j);
583 cl->do_oop(p);
584 }
585 }
586 }
588 void HRInto_G1RemSet::cleanup_after_oops_into_collection_set_do() {
589 guarantee( _cards_scanned != NULL, "invariant" );
590 _total_cards_scanned = 0;
591 for (uint i = 0; i < n_workers(); ++i)
592 _total_cards_scanned += _cards_scanned[i];
593 FREE_C_HEAP_ARRAY(size_t, _cards_scanned);
594 _cards_scanned = NULL;
595 // Cleanup after copy
596 #if G1_REM_SET_LOGGING
597 PrintRSClosure cl;
598 _g1->heap_region_iterate(&cl);
599 #endif
600 _g1->set_refine_cte_cl_concurrency(true);
601 cleanUpIteratorsClosure iterClosure;
602 _g1->collection_set_iterate(&iterClosure);
603 // Set all cards back to clean.
604 _g1->cleanUpCardTable();
606 if (ParallelGCThreads > 0) {
607 set_par_traversal(false);
608 }
610 if (_g1->evacuation_failed()) {
611 // Restore remembered sets for the regions pointing into
612 // the collection set.
613 if (G1DeferredRSUpdate) {
614 DirtyCardQueue dcq(&_g1->dirty_card_queue_set());
615 UpdateRSetOopsIntoCSDeferred deferred_update(_g1, &dcq);
616 new_refs_iterate(&deferred_update);
617 } else {
618 UpdateRSetOopsIntoCSImmediate immediate_update(_g1);
619 new_refs_iterate(&immediate_update);
620 }
621 }
622 for (uint i = 0; i < n_workers(); i++) {
623 _new_refs[i]->clear();
624 }
626 assert(!_par_traversal_in_progress, "Invariant between iterations.");
627 }
629 class UpdateRSObjectClosure: public ObjectClosure {
630 UpdateRSOopClosure* _update_rs_oop_cl;
631 public:
632 UpdateRSObjectClosure(UpdateRSOopClosure* update_rs_oop_cl) :
633 _update_rs_oop_cl(update_rs_oop_cl) {}
634 void do_object(oop obj) {
635 obj->oop_iterate(_update_rs_oop_cl);
636 }
638 };
640 class ScrubRSClosure: public HeapRegionClosure {
641 G1CollectedHeap* _g1h;
642 BitMap* _region_bm;
643 BitMap* _card_bm;
644 CardTableModRefBS* _ctbs;
645 public:
646 ScrubRSClosure(BitMap* region_bm, BitMap* card_bm) :
647 _g1h(G1CollectedHeap::heap()),
648 _region_bm(region_bm), _card_bm(card_bm),
649 _ctbs(NULL)
650 {
651 ModRefBarrierSet* bs = _g1h->mr_bs();
652 guarantee(bs->is_a(BarrierSet::CardTableModRef), "Precondition");
653 _ctbs = (CardTableModRefBS*)bs;
654 }
656 bool doHeapRegion(HeapRegion* r) {
657 if (!r->continuesHumongous()) {
658 r->rem_set()->scrub(_ctbs, _region_bm, _card_bm);
659 }
660 return false;
661 }
662 };
664 void HRInto_G1RemSet::scrub(BitMap* region_bm, BitMap* card_bm) {
665 ScrubRSClosure scrub_cl(region_bm, card_bm);
666 _g1->heap_region_iterate(&scrub_cl);
667 }
669 void HRInto_G1RemSet::scrub_par(BitMap* region_bm, BitMap* card_bm,
670 int worker_num, int claim_val) {
671 ScrubRSClosure scrub_cl(region_bm, card_bm);
672 _g1->heap_region_par_iterate_chunked(&scrub_cl, worker_num, claim_val);
673 }
676 static IntHistogram out_of_histo(50, 50);
678 void HRInto_G1RemSet::concurrentRefineOneCard(jbyte* card_ptr, int worker_i) {
679 // If the card is no longer dirty, nothing to do.
680 if (*card_ptr != CardTableModRefBS::dirty_card_val()) return;
682 // Construct the region representing the card.
683 HeapWord* start = _ct_bs->addr_for(card_ptr);
684 // And find the region containing it.
685 HeapRegion* r = _g1->heap_region_containing(start);
686 if (r == NULL) {
687 guarantee(_g1->is_in_permanent(start), "Or else where?");
688 return; // Not in the G1 heap (might be in perm, for example.)
689 }
690 // Why do we have to check here whether a card is on a young region,
691 // given that we dirty young regions and, as a result, the
692 // post-barrier is supposed to filter them out and never to enqueue
693 // them? When we allocate a new region as the "allocation region" we
694 // actually dirty its cards after we release the lock, since card
695 // dirtying while holding the lock was a performance bottleneck. So,
696 // as a result, it is possible for other threads to actually
697 // allocate objects in the region (after the acquire the lock)
698 // before all the cards on the region are dirtied. This is unlikely,
699 // and it doesn't happen often, but it can happen. So, the extra
700 // check below filters out those cards.
701 if (r->is_young()) {
702 return;
703 }
704 // While we are processing RSet buffers during the collection, we
705 // actually don't want to scan any cards on the collection set,
706 // since we don't want to update remebered sets with entries that
707 // point into the collection set, given that live objects from the
708 // collection set are about to move and such entries will be stale
709 // very soon. This change also deals with a reliability issue which
710 // involves scanning a card in the collection set and coming across
711 // an array that was being chunked and looking malformed. Note,
712 // however, that if evacuation fails, we have to scan any objects
713 // that were not moved and create any missing entries.
714 if (r->in_collection_set()) {
715 return;
716 }
718 // Should we defer it?
719 if (_cg1r->use_cache()) {
720 card_ptr = _cg1r->cache_insert(card_ptr);
721 // If it was not an eviction, nothing to do.
722 if (card_ptr == NULL) return;
724 // OK, we have to reset the card start, region, etc.
725 start = _ct_bs->addr_for(card_ptr);
726 r = _g1->heap_region_containing(start);
727 if (r == NULL) {
728 guarantee(_g1->is_in_permanent(start), "Or else where?");
729 return; // Not in the G1 heap (might be in perm, for example.)
730 }
731 guarantee(!r->is_young(), "It was evicted in the current minor cycle.");
732 }
734 HeapWord* end = _ct_bs->addr_for(card_ptr + 1);
735 MemRegion dirtyRegion(start, end);
737 #if CARD_REPEAT_HISTO
738 init_ct_freq_table(_g1->g1_reserved_obj_bytes());
739 ct_freq_note_card(_ct_bs->index_for(start));
740 #endif
742 UpdateRSOopClosure update_rs_oop_cl(this, worker_i);
743 update_rs_oop_cl.set_from(r);
744 FilterOutOfRegionClosure filter_then_update_rs_oop_cl(r, &update_rs_oop_cl);
746 // Undirty the card.
747 *card_ptr = CardTableModRefBS::clean_card_val();
748 // We must complete this write before we do any of the reads below.
749 OrderAccess::storeload();
750 // And process it, being careful of unallocated portions of TLAB's.
751 HeapWord* stop_point =
752 r->oops_on_card_seq_iterate_careful(dirtyRegion,
753 &filter_then_update_rs_oop_cl);
754 // If stop_point is non-null, then we encountered an unallocated region
755 // (perhaps the unfilled portion of a TLAB.) For now, we'll dirty the
756 // card and re-enqueue: if we put off the card until a GC pause, then the
757 // unallocated portion will be filled in. Alternatively, we might try
758 // the full complexity of the technique used in "regular" precleaning.
759 if (stop_point != NULL) {
760 // The card might have gotten re-dirtied and re-enqueued while we
761 // worked. (In fact, it's pretty likely.)
762 if (*card_ptr != CardTableModRefBS::dirty_card_val()) {
763 *card_ptr = CardTableModRefBS::dirty_card_val();
764 MutexLockerEx x(Shared_DirtyCardQ_lock,
765 Mutex::_no_safepoint_check_flag);
766 DirtyCardQueue* sdcq =
767 JavaThread::dirty_card_queue_set().shared_dirty_card_queue();
768 sdcq->enqueue(card_ptr);
769 }
770 } else {
771 out_of_histo.add_entry(filter_then_update_rs_oop_cl.out_of_region());
772 _conc_refine_cards++;
773 }
774 }
776 class HRRSStatsIter: public HeapRegionClosure {
777 size_t _occupied;
778 size_t _total_mem_sz;
779 size_t _max_mem_sz;
780 HeapRegion* _max_mem_sz_region;
781 public:
782 HRRSStatsIter() :
783 _occupied(0),
784 _total_mem_sz(0),
785 _max_mem_sz(0),
786 _max_mem_sz_region(NULL)
787 {}
789 bool doHeapRegion(HeapRegion* r) {
790 if (r->continuesHumongous()) return false;
791 size_t mem_sz = r->rem_set()->mem_size();
792 if (mem_sz > _max_mem_sz) {
793 _max_mem_sz = mem_sz;
794 _max_mem_sz_region = r;
795 }
796 _total_mem_sz += mem_sz;
797 size_t occ = r->rem_set()->occupied();
798 _occupied += occ;
799 return false;
800 }
801 size_t total_mem_sz() { return _total_mem_sz; }
802 size_t max_mem_sz() { return _max_mem_sz; }
803 size_t occupied() { return _occupied; }
804 HeapRegion* max_mem_sz_region() { return _max_mem_sz_region; }
805 };
807 class PrintRSThreadVTimeClosure : public ThreadClosure {
808 public:
809 virtual void do_thread(Thread *t) {
810 ConcurrentG1RefineThread* crt = (ConcurrentG1RefineThread*) t;
811 gclog_or_tty->print(" %5.2f", crt->vtime_accum());
812 }
813 };
815 void HRInto_G1RemSet::print_summary_info() {
816 G1CollectedHeap* g1 = G1CollectedHeap::heap();
818 #if CARD_REPEAT_HISTO
819 gclog_or_tty->print_cr("\nG1 card_repeat count histogram: ");
820 gclog_or_tty->print_cr(" # of repeats --> # of cards with that number.");
821 card_repeat_count.print_on(gclog_or_tty);
822 #endif
824 if (FILTEROUTOFREGIONCLOSURE_DOHISTOGRAMCOUNT) {
825 gclog_or_tty->print_cr("\nG1 rem-set out-of-region histogram: ");
826 gclog_or_tty->print_cr(" # of CS ptrs --> # of cards with that number.");
827 out_of_histo.print_on(gclog_or_tty);
828 }
829 gclog_or_tty->print_cr("\n Concurrent RS processed %d cards",
830 _conc_refine_cards);
831 DirtyCardQueueSet& dcqs = JavaThread::dirty_card_queue_set();
832 jint tot_processed_buffers =
833 dcqs.processed_buffers_mut() + dcqs.processed_buffers_rs_thread();
834 gclog_or_tty->print_cr(" Of %d completed buffers:", tot_processed_buffers);
835 gclog_or_tty->print_cr(" %8d (%5.1f%%) by conc RS threads.",
836 dcqs.processed_buffers_rs_thread(),
837 100.0*(float)dcqs.processed_buffers_rs_thread()/
838 (float)tot_processed_buffers);
839 gclog_or_tty->print_cr(" %8d (%5.1f%%) by mutator threads.",
840 dcqs.processed_buffers_mut(),
841 100.0*(float)dcqs.processed_buffers_mut()/
842 (float)tot_processed_buffers);
843 gclog_or_tty->print_cr(" Conc RS threads times(s)");
844 PrintRSThreadVTimeClosure p;
845 gclog_or_tty->print(" ");
846 g1->concurrent_g1_refine()->threads_do(&p);
847 gclog_or_tty->print_cr("");
849 if (G1UseHRIntoRS) {
850 HRRSStatsIter blk;
851 g1->heap_region_iterate(&blk);
852 gclog_or_tty->print_cr(" Total heap region rem set sizes = " SIZE_FORMAT "K."
853 " Max = " SIZE_FORMAT "K.",
854 blk.total_mem_sz()/K, blk.max_mem_sz()/K);
855 gclog_or_tty->print_cr(" Static structures = " SIZE_FORMAT "K,"
856 " free_lists = " SIZE_FORMAT "K.",
857 HeapRegionRemSet::static_mem_size()/K,
858 HeapRegionRemSet::fl_mem_size()/K);
859 gclog_or_tty->print_cr(" %d occupied cards represented.",
860 blk.occupied());
861 gclog_or_tty->print_cr(" Max sz region = [" PTR_FORMAT ", " PTR_FORMAT " )"
862 ", cap = " SIZE_FORMAT "K, occ = " SIZE_FORMAT "K.",
863 blk.max_mem_sz_region()->bottom(), blk.max_mem_sz_region()->end(),
864 (blk.max_mem_sz_region()->rem_set()->mem_size() + K - 1)/K,
865 (blk.max_mem_sz_region()->rem_set()->occupied() + K - 1)/K);
866 gclog_or_tty->print_cr(" Did %d coarsenings.",
867 HeapRegionRemSet::n_coarsenings());
869 }
870 }
871 void HRInto_G1RemSet::prepare_for_verify() {
872 if (G1HRRSFlushLogBuffersOnVerify &&
873 (VerifyBeforeGC || VerifyAfterGC)
874 && !_g1->full_collection()) {
875 cleanupHRRS();
876 _g1->set_refine_cte_cl_concurrency(false);
877 if (SafepointSynchronize::is_at_safepoint()) {
878 DirtyCardQueueSet& dcqs = JavaThread::dirty_card_queue_set();
879 dcqs.concatenate_logs();
880 }
881 bool cg1r_use_cache = _cg1r->use_cache();
882 _cg1r->set_use_cache(false);
883 updateRS(0);
884 _cg1r->set_use_cache(cg1r_use_cache);
886 assert(JavaThread::dirty_card_queue_set().completed_buffers_num() == 0, "All should be consumed");
887 }
888 }