Tue, 05 May 2009 22:15:35 -0700
6833576: G1: assert illegal index, growableArray.hpp:186
Summary: The code that calculates the heap region index for an object address incorrectly used signed arithmetic.
Reviewed-by: jcoomes, ysr
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 UpdateRSOutOfRegionClosure: public HeapRegionClosure {
109 G1CollectedHeap* _g1h;
110 ModRefBarrierSet* _mr_bs;
111 UpdateRSOopClosure _cl;
112 int _worker_i;
113 public:
114 UpdateRSOutOfRegionClosure(G1CollectedHeap* g1, int worker_i = 0) :
115 _cl(g1->g1_rem_set()->as_HRInto_G1RemSet(), worker_i),
116 _mr_bs(g1->mr_bs()),
117 _worker_i(worker_i),
118 _g1h(g1)
119 {}
120 bool doHeapRegion(HeapRegion* r) {
121 if (!r->in_collection_set() && !r->continuesHumongous()) {
122 _cl.set_from(r);
123 r->set_next_filter_kind(HeapRegionDCTOC::OutOfRegionFilterKind);
124 _mr_bs->mod_oop_in_space_iterate(r, &_cl, true, true);
125 }
126 return false;
127 }
128 };
130 class VerifyRSCleanCardOopClosure: public OopClosure {
131 G1CollectedHeap* _g1;
132 public:
133 VerifyRSCleanCardOopClosure(G1CollectedHeap* g1) : _g1(g1) {}
135 virtual void do_oop(narrowOop* p) {
136 guarantee(false, "NYI");
137 }
138 virtual void do_oop(oop* p) {
139 oop obj = *p;
140 HeapRegion* to = _g1->heap_region_containing(obj);
141 guarantee(to == NULL || !to->in_collection_set(),
142 "Missed a rem set member.");
143 }
144 };
146 HRInto_G1RemSet::HRInto_G1RemSet(G1CollectedHeap* g1, CardTableModRefBS* ct_bs)
147 : G1RemSet(g1), _ct_bs(ct_bs), _g1p(_g1->g1_policy()),
148 _cg1r(g1->concurrent_g1_refine()),
149 _par_traversal_in_progress(false), _new_refs(NULL),
150 _cards_scanned(NULL), _total_cards_scanned(0)
151 {
152 _seq_task = new SubTasksDone(NumSeqTasks);
153 guarantee(n_workers() > 0, "There should be some workers");
154 _new_refs = NEW_C_HEAP_ARRAY(GrowableArray<oop*>*, n_workers());
155 for (uint i = 0; i < n_workers(); i++) {
156 _new_refs[i] = new (ResourceObj::C_HEAP) GrowableArray<oop*>(8192,true);
157 }
158 }
160 HRInto_G1RemSet::~HRInto_G1RemSet() {
161 delete _seq_task;
162 for (uint i = 0; i < n_workers(); i++) {
163 delete _new_refs[i];
164 }
165 FREE_C_HEAP_ARRAY(GrowableArray<oop*>*, _new_refs);
166 }
168 void CountNonCleanMemRegionClosure::do_MemRegion(MemRegion mr) {
169 if (_g1->is_in_g1_reserved(mr.start())) {
170 _n += (int) ((mr.byte_size() / CardTableModRefBS::card_size));
171 if (_start_first == NULL) _start_first = mr.start();
172 }
173 }
175 class ScanRSClosure : public HeapRegionClosure {
176 size_t _cards_done, _cards;
177 G1CollectedHeap* _g1h;
178 OopsInHeapRegionClosure* _oc;
179 G1BlockOffsetSharedArray* _bot_shared;
180 CardTableModRefBS *_ct_bs;
181 int _worker_i;
182 bool _try_claimed;
183 size_t _min_skip_distance, _max_skip_distance;
184 public:
185 ScanRSClosure(OopsInHeapRegionClosure* oc, int worker_i) :
186 _oc(oc),
187 _cards(0),
188 _cards_done(0),
189 _worker_i(worker_i),
190 _try_claimed(false)
191 {
192 _g1h = G1CollectedHeap::heap();
193 _bot_shared = _g1h->bot_shared();
194 _ct_bs = (CardTableModRefBS*) (_g1h->barrier_set());
195 _min_skip_distance = 16;
196 _max_skip_distance = 2 * _g1h->n_par_threads() * _min_skip_distance;
197 }
199 void set_try_claimed() { _try_claimed = true; }
201 void scanCard(size_t index, HeapRegion *r) {
202 _cards_done++;
203 DirtyCardToOopClosure* cl =
204 r->new_dcto_closure(_oc,
205 CardTableModRefBS::Precise,
206 HeapRegionDCTOC::IntoCSFilterKind);
208 // Set the "from" region in the closure.
209 _oc->set_region(r);
210 HeapWord* card_start = _bot_shared->address_for_index(index);
211 HeapWord* card_end = card_start + G1BlockOffsetSharedArray::N_words;
212 Space *sp = SharedHeap::heap()->space_containing(card_start);
213 MemRegion sm_region;
214 if (ParallelGCThreads > 0) {
215 // first find the used area
216 sm_region = sp->used_region_at_save_marks();
217 } else {
218 // The closure is not idempotent. We shouldn't look at objects
219 // allocated during the GC.
220 sm_region = sp->used_region_at_save_marks();
221 }
222 MemRegion mr = sm_region.intersection(MemRegion(card_start,card_end));
223 if (!mr.is_empty()) {
224 cl->do_MemRegion(mr);
225 }
226 }
228 void printCard(HeapRegion* card_region, size_t card_index,
229 HeapWord* card_start) {
230 gclog_or_tty->print_cr("T %d Region [" PTR_FORMAT ", " PTR_FORMAT ") "
231 "RS names card %p: "
232 "[" PTR_FORMAT ", " PTR_FORMAT ")",
233 _worker_i,
234 card_region->bottom(), card_region->end(),
235 card_index,
236 card_start, card_start + G1BlockOffsetSharedArray::N_words);
237 }
239 bool doHeapRegion(HeapRegion* r) {
240 assert(r->in_collection_set(), "should only be called on elements of CS.");
241 HeapRegionRemSet* hrrs = r->rem_set();
242 if (hrrs->iter_is_complete()) return false; // All done.
243 if (!_try_claimed && !hrrs->claim_iter()) return false;
244 // If we didn't return above, then
245 // _try_claimed || r->claim_iter()
246 // is true: either we're supposed to work on claimed-but-not-complete
247 // regions, or we successfully claimed the region.
248 HeapRegionRemSetIterator* iter = _g1h->rem_set_iterator(_worker_i);
249 hrrs->init_iterator(iter);
250 size_t card_index;
251 size_t skip_distance = 0, current_card = 0, jump_to_card = 0;
252 while (iter->has_next(card_index)) {
253 if (current_card < jump_to_card) {
254 ++current_card;
255 continue;
256 }
257 HeapWord* card_start = _g1h->bot_shared()->address_for_index(card_index);
258 #if 0
259 gclog_or_tty->print("Rem set iteration yielded card [" PTR_FORMAT ", " PTR_FORMAT ").\n",
260 card_start, card_start + CardTableModRefBS::card_size_in_words);
261 #endif
263 HeapRegion* card_region = _g1h->heap_region_containing(card_start);
264 assert(card_region != NULL, "Yielding cards not in the heap?");
265 _cards++;
267 // If the card is dirty, then we will scan it during updateRS.
268 if (!card_region->in_collection_set() && !_ct_bs->is_card_dirty(card_index)) {
269 if (!_ct_bs->is_card_claimed(card_index) && _ct_bs->claim_card(card_index)) {
270 scanCard(card_index, card_region);
271 } else if (_try_claimed) {
272 if (jump_to_card == 0 || jump_to_card != current_card) {
273 // We did some useful work in the previous iteration.
274 // Decrease the distance.
275 skip_distance = MAX2(skip_distance >> 1, _min_skip_distance);
276 } else {
277 // Previous iteration resulted in a claim failure.
278 // Increase the distance.
279 skip_distance = MIN2(skip_distance << 1, _max_skip_distance);
280 }
281 jump_to_card = current_card + skip_distance;
282 }
283 }
284 ++current_card;
285 }
286 if (!_try_claimed) {
287 hrrs->set_iter_complete();
288 }
289 return false;
290 }
291 // Set all cards back to clean.
292 void cleanup() {_g1h->cleanUpCardTable();}
293 size_t cards_done() { return _cards_done;}
294 size_t cards_looked_up() { return _cards;}
295 };
297 // We want the parallel threads to start their scanning at
298 // different collection set regions to avoid contention.
299 // If we have:
300 // n collection set regions
301 // p threads
302 // Then thread t will start at region t * floor (n/p)
304 HeapRegion* HRInto_G1RemSet::calculateStartRegion(int worker_i) {
305 HeapRegion* result = _g1p->collection_set();
306 if (ParallelGCThreads > 0) {
307 size_t cs_size = _g1p->collection_set_size();
308 int n_workers = _g1->workers()->total_workers();
309 size_t cs_spans = cs_size / n_workers;
310 size_t ind = cs_spans * worker_i;
311 for (size_t i = 0; i < ind; i++)
312 result = result->next_in_collection_set();
313 }
314 return result;
315 }
317 void HRInto_G1RemSet::scanRS(OopsInHeapRegionClosure* oc, int worker_i) {
318 double rs_time_start = os::elapsedTime();
319 HeapRegion *startRegion = calculateStartRegion(worker_i);
321 BufferingOopsInHeapRegionClosure boc(oc);
322 ScanRSClosure scanRScl(&boc, worker_i);
323 _g1->collection_set_iterate_from(startRegion, &scanRScl);
324 scanRScl.set_try_claimed();
325 _g1->collection_set_iterate_from(startRegion, &scanRScl);
327 boc.done();
328 double closure_app_time_sec = boc.closure_app_seconds();
329 double scan_rs_time_sec = (os::elapsedTime() - rs_time_start) -
330 closure_app_time_sec;
331 double closure_app_time_ms = closure_app_time_sec * 1000.0;
333 assert( _cards_scanned != NULL, "invariant" );
334 _cards_scanned[worker_i] = scanRScl.cards_done();
336 _g1p->record_scan_rs_start_time(worker_i, rs_time_start * 1000.0);
337 _g1p->record_scan_rs_time(worker_i, scan_rs_time_sec * 1000.0);
339 double scan_new_refs_time_ms = _g1p->get_scan_new_refs_time(worker_i);
340 if (scan_new_refs_time_ms > 0.0) {
341 closure_app_time_ms += scan_new_refs_time_ms;
342 }
344 _g1p->record_obj_copy_time(worker_i, closure_app_time_ms);
345 }
347 void HRInto_G1RemSet::updateRS(int worker_i) {
348 ConcurrentG1Refine* cg1r = _g1->concurrent_g1_refine();
350 double start = os::elapsedTime();
351 _g1p->record_update_rs_start_time(worker_i, start * 1000.0);
353 if (G1RSBarrierUseQueue && !cg1r->do_traversal()) {
354 // Apply the appropriate closure to all remaining log entries.
355 _g1->iterate_dirty_card_closure(false, worker_i);
356 // Now there should be no dirty cards.
357 if (G1RSLogCheckCardTable) {
358 CountNonCleanMemRegionClosure cl(_g1);
359 _ct_bs->mod_card_iterate(&cl);
360 // XXX This isn't true any more: keeping cards of young regions
361 // marked dirty broke it. Need some reasonable fix.
362 guarantee(cl.n() == 0, "Card table should be clean.");
363 }
364 } else {
365 UpdateRSOutOfRegionClosure update_rs(_g1, worker_i);
366 _g1->heap_region_iterate(&update_rs);
367 // We did a traversal; no further one is necessary.
368 if (G1RSBarrierUseQueue) {
369 assert(cg1r->do_traversal(), "Or we shouldn't have gotten here.");
370 cg1r->set_pya_cancel();
371 }
372 if (_cg1r->use_cache()) {
373 _cg1r->clear_and_record_card_counts();
374 _cg1r->clear_hot_cache();
375 }
376 }
377 _g1p->record_update_rs_time(worker_i, (os::elapsedTime() - start) * 1000.0);
378 }
380 #ifndef PRODUCT
381 class PrintRSClosure : public HeapRegionClosure {
382 int _count;
383 public:
384 PrintRSClosure() : _count(0) {}
385 bool doHeapRegion(HeapRegion* r) {
386 HeapRegionRemSet* hrrs = r->rem_set();
387 _count += (int) hrrs->occupied();
388 if (hrrs->occupied() == 0) {
389 gclog_or_tty->print("Heap Region [" PTR_FORMAT ", " PTR_FORMAT ") "
390 "has no remset entries\n",
391 r->bottom(), r->end());
392 } else {
393 gclog_or_tty->print("Printing rem set for heap region [" PTR_FORMAT ", " PTR_FORMAT ")\n",
394 r->bottom(), r->end());
395 r->print();
396 hrrs->print();
397 gclog_or_tty->print("\nDone printing rem set\n");
398 }
399 return false;
400 }
401 int occupied() {return _count;}
402 };
403 #endif
405 class CountRSSizeClosure: public HeapRegionClosure {
406 size_t _n;
407 size_t _tot;
408 size_t _max;
409 HeapRegion* _max_r;
410 enum {
411 N = 20,
412 MIN = 6
413 };
414 int _histo[N];
415 public:
416 CountRSSizeClosure() : _n(0), _tot(0), _max(0), _max_r(NULL) {
417 for (int i = 0; i < N; i++) _histo[i] = 0;
418 }
419 bool doHeapRegion(HeapRegion* r) {
420 if (!r->continuesHumongous()) {
421 size_t occ = r->rem_set()->occupied();
422 _n++;
423 _tot += occ;
424 if (occ > _max) {
425 _max = occ;
426 _max_r = r;
427 }
428 // Fit it into a histo bin.
429 int s = 1 << MIN;
430 int i = 0;
431 while (occ > (size_t) s && i < (N-1)) {
432 s = s << 1;
433 i++;
434 }
435 _histo[i]++;
436 }
437 return false;
438 }
439 size_t n() { return _n; }
440 size_t tot() { return _tot; }
441 size_t mx() { return _max; }
442 HeapRegion* mxr() { return _max_r; }
443 void print_histo() {
444 int mx = N;
445 while (mx >= 0) {
446 if (_histo[mx-1] > 0) break;
447 mx--;
448 }
449 gclog_or_tty->print_cr("Number of regions with given RS sizes:");
450 gclog_or_tty->print_cr(" <= %8d %8d", 1 << MIN, _histo[0]);
451 for (int i = 1; i < mx-1; i++) {
452 gclog_or_tty->print_cr(" %8d - %8d %8d",
453 (1 << (MIN + i - 1)) + 1,
454 1 << (MIN + i),
455 _histo[i]);
456 }
457 gclog_or_tty->print_cr(" > %8d %8d", (1 << (MIN+mx-2))+1, _histo[mx-1]);
458 }
459 };
461 void
462 HRInto_G1RemSet::scanNewRefsRS(OopsInHeapRegionClosure* oc,
463 int worker_i) {
464 double scan_new_refs_start_sec = os::elapsedTime();
465 G1CollectedHeap* g1h = G1CollectedHeap::heap();
466 CardTableModRefBS* ct_bs = (CardTableModRefBS*) (g1h->barrier_set());
467 for (int i = 0; i < _new_refs[worker_i]->length(); i++) {
468 oop* p = _new_refs[worker_i]->at(i);
469 oop obj = *p;
470 // *p was in the collection set when p was pushed on "_new_refs", but
471 // another thread may have processed this location from an RS, so it
472 // might not point into the CS any longer. If so, it's obviously been
473 // processed, and we don't need to do anything further.
474 if (g1h->obj_in_cs(obj)) {
475 HeapRegion* r = g1h->heap_region_containing(p);
477 DEBUG_ONLY(HeapRegion* to = g1h->heap_region_containing(obj));
478 oc->set_region(r);
479 // If "p" has already been processed concurrently, this is
480 // idempotent.
481 oc->do_oop(p);
482 }
483 }
484 _g1p->record_scan_new_refs_time(worker_i,
485 (os::elapsedTime() - scan_new_refs_start_sec)
486 * 1000.0);
487 }
489 void HRInto_G1RemSet::set_par_traversal(bool b) {
490 _par_traversal_in_progress = b;
491 HeapRegionRemSet::set_par_traversal(b);
492 }
494 void HRInto_G1RemSet::cleanupHRRS() {
495 HeapRegionRemSet::cleanup();
496 }
498 void
499 HRInto_G1RemSet::oops_into_collection_set_do(OopsInHeapRegionClosure* oc,
500 int worker_i) {
501 #if CARD_REPEAT_HISTO
502 ct_freq_update_histo_and_reset();
503 #endif
504 if (worker_i == 0) {
505 _cg1r->clear_and_record_card_counts();
506 }
508 // Make this into a command-line flag...
509 if (G1RSCountHisto && (ParallelGCThreads == 0 || worker_i == 0)) {
510 CountRSSizeClosure count_cl;
511 _g1->heap_region_iterate(&count_cl);
512 gclog_or_tty->print_cr("Avg of %d RS counts is %f, max is %d, "
513 "max region is " PTR_FORMAT,
514 count_cl.n(), (float)count_cl.tot()/(float)count_cl.n(),
515 count_cl.mx(), count_cl.mxr());
516 count_cl.print_histo();
517 }
519 if (ParallelGCThreads > 0) {
520 // The two flags below were introduced temporarily to serialize
521 // the updating and scanning of remembered sets. There are some
522 // race conditions when these two operations are done in parallel
523 // and they are causing failures. When we resolve said race
524 // conditions, we'll revert back to parallel remembered set
525 // updating and scanning. See CRs 6677707 and 6677708.
526 if (G1ParallelRSetUpdatingEnabled || (worker_i == 0)) {
527 updateRS(worker_i);
528 scanNewRefsRS(oc, worker_i);
529 } else {
530 _g1p->record_update_rs_start_time(worker_i, os::elapsedTime());
531 _g1p->record_update_rs_processed_buffers(worker_i, 0.0);
532 _g1p->record_update_rs_time(worker_i, 0.0);
533 _g1p->record_scan_new_refs_time(worker_i, 0.0);
534 }
535 if (G1ParallelRSetScanningEnabled || (worker_i == 0)) {
536 scanRS(oc, worker_i);
537 } else {
538 _g1p->record_scan_rs_start_time(worker_i, os::elapsedTime());
539 _g1p->record_scan_rs_time(worker_i, 0.0);
540 }
541 } else {
542 assert(worker_i == 0, "invariant");
543 updateRS(0);
544 scanNewRefsRS(oc, 0);
545 scanRS(oc, 0);
546 }
547 }
549 void HRInto_G1RemSet::
550 prepare_for_oops_into_collection_set_do() {
551 #if G1_REM_SET_LOGGING
552 PrintRSClosure cl;
553 _g1->collection_set_iterate(&cl);
554 #endif
555 cleanupHRRS();
556 ConcurrentG1Refine* cg1r = _g1->concurrent_g1_refine();
557 _g1->set_refine_cte_cl_concurrency(false);
558 DirtyCardQueueSet& dcqs = JavaThread::dirty_card_queue_set();
559 dcqs.concatenate_logs();
561 assert(!_par_traversal_in_progress, "Invariant between iterations.");
562 if (ParallelGCThreads > 0) {
563 set_par_traversal(true);
564 _seq_task->set_par_threads((int)n_workers());
565 if (cg1r->do_traversal()) {
566 updateRS(0);
567 // Have to do this again after updaters
568 cleanupHRRS();
569 }
570 }
571 guarantee( _cards_scanned == NULL, "invariant" );
572 _cards_scanned = NEW_C_HEAP_ARRAY(size_t, n_workers());
573 for (uint i = 0; i < n_workers(); ++i) {
574 _cards_scanned[i] = 0;
575 }
576 _total_cards_scanned = 0;
577 }
580 class cleanUpIteratorsClosure : public HeapRegionClosure {
581 bool doHeapRegion(HeapRegion *r) {
582 HeapRegionRemSet* hrrs = r->rem_set();
583 hrrs->init_for_par_iteration();
584 return false;
585 }
586 };
588 class UpdateRSetOopsIntoCSImmediate : public OopClosure {
589 G1CollectedHeap* _g1;
590 public:
591 UpdateRSetOopsIntoCSImmediate(G1CollectedHeap* g1) : _g1(g1) { }
592 virtual void do_oop(narrowOop* p) {
593 guarantee(false, "NYI");
594 }
595 virtual void do_oop(oop* p) {
596 HeapRegion* to = _g1->heap_region_containing(*p);
597 if (to->in_collection_set()) {
598 to->rem_set()->add_reference(p, 0);
599 }
600 }
601 };
603 class UpdateRSetOopsIntoCSDeferred : public OopClosure {
604 G1CollectedHeap* _g1;
605 CardTableModRefBS* _ct_bs;
606 DirtyCardQueue* _dcq;
607 public:
608 UpdateRSetOopsIntoCSDeferred(G1CollectedHeap* g1, DirtyCardQueue* dcq) :
609 _g1(g1), _ct_bs((CardTableModRefBS*)_g1->barrier_set()), _dcq(dcq) { }
610 virtual void do_oop(narrowOop* p) {
611 guarantee(false, "NYI");
612 }
613 virtual void do_oop(oop* p) {
614 oop obj = *p;
615 if (_g1->obj_in_cs(obj)) {
616 size_t card_index = _ct_bs->index_for(p);
617 if (_ct_bs->mark_card_deferred(card_index)) {
618 _dcq->enqueue((jbyte*)_ct_bs->byte_for_index(card_index));
619 }
620 }
621 }
622 };
624 void HRInto_G1RemSet::new_refs_iterate(OopClosure* cl) {
625 for (size_t i = 0; i < n_workers(); i++) {
626 for (int j = 0; j < _new_refs[i]->length(); j++) {
627 oop* p = _new_refs[i]->at(j);
628 cl->do_oop(p);
629 }
630 }
631 }
633 void HRInto_G1RemSet::cleanup_after_oops_into_collection_set_do() {
634 guarantee( _cards_scanned != NULL, "invariant" );
635 _total_cards_scanned = 0;
636 for (uint i = 0; i < n_workers(); ++i)
637 _total_cards_scanned += _cards_scanned[i];
638 FREE_C_HEAP_ARRAY(size_t, _cards_scanned);
639 _cards_scanned = NULL;
640 // Cleanup after copy
641 #if G1_REM_SET_LOGGING
642 PrintRSClosure cl;
643 _g1->heap_region_iterate(&cl);
644 #endif
645 _g1->set_refine_cte_cl_concurrency(true);
646 cleanUpIteratorsClosure iterClosure;
647 _g1->collection_set_iterate(&iterClosure);
648 // Set all cards back to clean.
649 _g1->cleanUpCardTable();
650 if (ParallelGCThreads > 0) {
651 ConcurrentG1Refine* cg1r = _g1->concurrent_g1_refine();
652 if (cg1r->do_traversal()) {
653 cg1r->cg1rThread()->set_do_traversal(false);
654 }
655 set_par_traversal(false);
656 }
658 if (_g1->evacuation_failed()) {
659 // Restore remembered sets for the regions pointing into
660 // the collection set.
661 if (G1DeferredRSUpdate) {
662 DirtyCardQueue dcq(&_g1->dirty_card_queue_set());
663 UpdateRSetOopsIntoCSDeferred deferred_update(_g1, &dcq);
664 new_refs_iterate(&deferred_update);
665 } else {
666 UpdateRSetOopsIntoCSImmediate immediate_update(_g1);
667 new_refs_iterate(&immediate_update);
668 }
669 }
670 for (uint i = 0; i < n_workers(); i++) {
671 _new_refs[i]->clear();
672 }
674 assert(!_par_traversal_in_progress, "Invariant between iterations.");
675 }
677 class UpdateRSObjectClosure: public ObjectClosure {
678 UpdateRSOopClosure* _update_rs_oop_cl;
679 public:
680 UpdateRSObjectClosure(UpdateRSOopClosure* update_rs_oop_cl) :
681 _update_rs_oop_cl(update_rs_oop_cl) {}
682 void do_object(oop obj) {
683 obj->oop_iterate(_update_rs_oop_cl);
684 }
686 };
688 class ScrubRSClosure: public HeapRegionClosure {
689 G1CollectedHeap* _g1h;
690 BitMap* _region_bm;
691 BitMap* _card_bm;
692 CardTableModRefBS* _ctbs;
693 public:
694 ScrubRSClosure(BitMap* region_bm, BitMap* card_bm) :
695 _g1h(G1CollectedHeap::heap()),
696 _region_bm(region_bm), _card_bm(card_bm),
697 _ctbs(NULL)
698 {
699 ModRefBarrierSet* bs = _g1h->mr_bs();
700 guarantee(bs->is_a(BarrierSet::CardTableModRef), "Precondition");
701 _ctbs = (CardTableModRefBS*)bs;
702 }
704 bool doHeapRegion(HeapRegion* r) {
705 if (!r->continuesHumongous()) {
706 r->rem_set()->scrub(_ctbs, _region_bm, _card_bm);
707 }
708 return false;
709 }
710 };
712 void HRInto_G1RemSet::scrub(BitMap* region_bm, BitMap* card_bm) {
713 ScrubRSClosure scrub_cl(region_bm, card_bm);
714 _g1->heap_region_iterate(&scrub_cl);
715 }
717 void HRInto_G1RemSet::scrub_par(BitMap* region_bm, BitMap* card_bm,
718 int worker_num, int claim_val) {
719 ScrubRSClosure scrub_cl(region_bm, card_bm);
720 _g1->heap_region_par_iterate_chunked(&scrub_cl, worker_num, claim_val);
721 }
724 class ConcRefineRegionClosure: public HeapRegionClosure {
725 G1CollectedHeap* _g1h;
726 CardTableModRefBS* _ctbs;
727 ConcurrentGCThread* _cgc_thrd;
728 ConcurrentG1Refine* _cg1r;
729 unsigned _cards_processed;
730 UpdateRSOopClosure _update_rs_oop_cl;
731 public:
732 ConcRefineRegionClosure(CardTableModRefBS* ctbs,
733 ConcurrentG1Refine* cg1r,
734 HRInto_G1RemSet* g1rs) :
735 _ctbs(ctbs), _cg1r(cg1r), _cgc_thrd(cg1r->cg1rThread()),
736 _update_rs_oop_cl(g1rs), _cards_processed(0),
737 _g1h(G1CollectedHeap::heap())
738 {}
740 bool doHeapRegion(HeapRegion* r) {
741 if (!r->in_collection_set() &&
742 !r->continuesHumongous() &&
743 !r->is_young()) {
744 _update_rs_oop_cl.set_from(r);
745 UpdateRSObjectClosure update_rs_obj_cl(&_update_rs_oop_cl);
747 // For each run of dirty card in the region:
748 // 1) Clear the cards.
749 // 2) Process the range corresponding to the run, adding any
750 // necessary RS entries.
751 // 1 must precede 2, so that a concurrent modification redirties the
752 // card. If a processing attempt does not succeed, because it runs
753 // into an unparseable region, we will do binary search to find the
754 // beginning of the next parseable region.
755 HeapWord* startAddr = r->bottom();
756 HeapWord* endAddr = r->used_region().end();
757 HeapWord* lastAddr;
758 HeapWord* nextAddr;
760 for (nextAddr = lastAddr = startAddr;
761 nextAddr < endAddr;
762 nextAddr = lastAddr) {
763 MemRegion dirtyRegion;
765 // Get and clear dirty region from card table
766 MemRegion next_mr(nextAddr, endAddr);
767 dirtyRegion =
768 _ctbs->dirty_card_range_after_reset(
769 next_mr,
770 true, CardTableModRefBS::clean_card_val());
771 assert(dirtyRegion.start() >= nextAddr,
772 "returned region inconsistent?");
774 if (!dirtyRegion.is_empty()) {
775 HeapWord* stop_point =
776 r->object_iterate_mem_careful(dirtyRegion,
777 &update_rs_obj_cl);
778 if (stop_point == NULL) {
779 lastAddr = dirtyRegion.end();
780 _cards_processed +=
781 (int) (dirtyRegion.word_size() / CardTableModRefBS::card_size_in_words);
782 } else {
783 // We're going to skip one or more cards that we can't parse.
784 HeapWord* next_parseable_card =
785 r->next_block_start_careful(stop_point);
786 // Round this up to a card boundary.
787 next_parseable_card =
788 _ctbs->addr_for(_ctbs->byte_after_const(next_parseable_card));
789 // Now we invalidate the intervening cards so we'll see them
790 // again.
791 MemRegion remaining_dirty =
792 MemRegion(stop_point, dirtyRegion.end());
793 MemRegion skipped =
794 MemRegion(stop_point, next_parseable_card);
795 _ctbs->invalidate(skipped.intersection(remaining_dirty));
797 // Now start up again where we can parse.
798 lastAddr = next_parseable_card;
800 // Count how many we did completely.
801 _cards_processed +=
802 (stop_point - dirtyRegion.start()) /
803 CardTableModRefBS::card_size_in_words;
804 }
805 // Allow interruption at regular intervals.
806 // (Might need to make them more regular, if we get big
807 // dirty regions.)
808 if (_cgc_thrd != NULL) {
809 if (_cgc_thrd->should_yield()) {
810 _cgc_thrd->yield();
811 switch (_cg1r->get_pya()) {
812 case PYA_continue:
813 // This may have changed: re-read.
814 endAddr = r->used_region().end();
815 continue;
816 case PYA_restart: case PYA_cancel:
817 return true;
818 }
819 }
820 }
821 } else {
822 break;
823 }
824 }
825 }
826 // A good yield opportunity.
827 if (_cgc_thrd != NULL) {
828 if (_cgc_thrd->should_yield()) {
829 _cgc_thrd->yield();
830 switch (_cg1r->get_pya()) {
831 case PYA_restart: case PYA_cancel:
832 return true;
833 default:
834 break;
835 }
837 }
838 }
839 return false;
840 }
842 unsigned cards_processed() { return _cards_processed; }
843 };
846 void HRInto_G1RemSet::concurrentRefinementPass(ConcurrentG1Refine* cg1r) {
847 ConcRefineRegionClosure cr_cl(ct_bs(), cg1r, this);
848 _g1->heap_region_iterate(&cr_cl);
849 _conc_refine_traversals++;
850 _conc_refine_cards += cr_cl.cards_processed();
851 }
853 static IntHistogram out_of_histo(50, 50);
857 void HRInto_G1RemSet::concurrentRefineOneCard(jbyte* card_ptr, int worker_i) {
858 // If the card is no longer dirty, nothing to do.
859 if (*card_ptr != CardTableModRefBS::dirty_card_val()) return;
861 // Construct the region representing the card.
862 HeapWord* start = _ct_bs->addr_for(card_ptr);
863 // And find the region containing it.
864 HeapRegion* r = _g1->heap_region_containing(start);
865 if (r == NULL) {
866 guarantee(_g1->is_in_permanent(start), "Or else where?");
867 return; // Not in the G1 heap (might be in perm, for example.)
868 }
869 // Why do we have to check here whether a card is on a young region,
870 // given that we dirty young regions and, as a result, the
871 // post-barrier is supposed to filter them out and never to enqueue
872 // them? When we allocate a new region as the "allocation region" we
873 // actually dirty its cards after we release the lock, since card
874 // dirtying while holding the lock was a performance bottleneck. So,
875 // as a result, it is possible for other threads to actually
876 // allocate objects in the region (after the acquire the lock)
877 // before all the cards on the region are dirtied. This is unlikely,
878 // and it doesn't happen often, but it can happen. So, the extra
879 // check below filters out those cards.
880 if (r->is_young()) {
881 return;
882 }
883 // While we are processing RSet buffers during the collection, we
884 // actually don't want to scan any cards on the collection set,
885 // since we don't want to update remebered sets with entries that
886 // point into the collection set, given that live objects from the
887 // collection set are about to move and such entries will be stale
888 // very soon. This change also deals with a reliability issue which
889 // involves scanning a card in the collection set and coming across
890 // an array that was being chunked and looking malformed. Note,
891 // however, that if evacuation fails, we have to scan any objects
892 // that were not moved and create any missing entries.
893 if (r->in_collection_set()) {
894 return;
895 }
897 // Should we defer it?
898 if (_cg1r->use_cache()) {
899 card_ptr = _cg1r->cache_insert(card_ptr);
900 // If it was not an eviction, nothing to do.
901 if (card_ptr == NULL) return;
903 // OK, we have to reset the card start, region, etc.
904 start = _ct_bs->addr_for(card_ptr);
905 r = _g1->heap_region_containing(start);
906 if (r == NULL) {
907 guarantee(_g1->is_in_permanent(start), "Or else where?");
908 return; // Not in the G1 heap (might be in perm, for example.)
909 }
910 guarantee(!r->is_young(), "It was evicted in the current minor cycle.");
911 }
913 HeapWord* end = _ct_bs->addr_for(card_ptr + 1);
914 MemRegion dirtyRegion(start, end);
916 #if CARD_REPEAT_HISTO
917 init_ct_freq_table(_g1->g1_reserved_obj_bytes());
918 ct_freq_note_card(_ct_bs->index_for(start));
919 #endif
921 UpdateRSOopClosure update_rs_oop_cl(this, worker_i);
922 update_rs_oop_cl.set_from(r);
923 FilterOutOfRegionClosure filter_then_update_rs_oop_cl(r, &update_rs_oop_cl);
925 // Undirty the card.
926 *card_ptr = CardTableModRefBS::clean_card_val();
927 // We must complete this write before we do any of the reads below.
928 OrderAccess::storeload();
929 // And process it, being careful of unallocated portions of TLAB's.
930 HeapWord* stop_point =
931 r->oops_on_card_seq_iterate_careful(dirtyRegion,
932 &filter_then_update_rs_oop_cl);
933 // If stop_point is non-null, then we encountered an unallocated region
934 // (perhaps the unfilled portion of a TLAB.) For now, we'll dirty the
935 // card and re-enqueue: if we put off the card until a GC pause, then the
936 // unallocated portion will be filled in. Alternatively, we might try
937 // the full complexity of the technique used in "regular" precleaning.
938 if (stop_point != NULL) {
939 // The card might have gotten re-dirtied and re-enqueued while we
940 // worked. (In fact, it's pretty likely.)
941 if (*card_ptr != CardTableModRefBS::dirty_card_val()) {
942 *card_ptr = CardTableModRefBS::dirty_card_val();
943 MutexLockerEx x(Shared_DirtyCardQ_lock,
944 Mutex::_no_safepoint_check_flag);
945 DirtyCardQueue* sdcq =
946 JavaThread::dirty_card_queue_set().shared_dirty_card_queue();
947 sdcq->enqueue(card_ptr);
948 }
949 } else {
950 out_of_histo.add_entry(filter_then_update_rs_oop_cl.out_of_region());
951 _conc_refine_cards++;
952 }
953 }
955 class HRRSStatsIter: public HeapRegionClosure {
956 size_t _occupied;
957 size_t _total_mem_sz;
958 size_t _max_mem_sz;
959 HeapRegion* _max_mem_sz_region;
960 public:
961 HRRSStatsIter() :
962 _occupied(0),
963 _total_mem_sz(0),
964 _max_mem_sz(0),
965 _max_mem_sz_region(NULL)
966 {}
968 bool doHeapRegion(HeapRegion* r) {
969 if (r->continuesHumongous()) return false;
970 size_t mem_sz = r->rem_set()->mem_size();
971 if (mem_sz > _max_mem_sz) {
972 _max_mem_sz = mem_sz;
973 _max_mem_sz_region = r;
974 }
975 _total_mem_sz += mem_sz;
976 size_t occ = r->rem_set()->occupied();
977 _occupied += occ;
978 return false;
979 }
980 size_t total_mem_sz() { return _total_mem_sz; }
981 size_t max_mem_sz() { return _max_mem_sz; }
982 size_t occupied() { return _occupied; }
983 HeapRegion* max_mem_sz_region() { return _max_mem_sz_region; }
984 };
986 void HRInto_G1RemSet::print_summary_info() {
987 G1CollectedHeap* g1 = G1CollectedHeap::heap();
988 ConcurrentG1RefineThread* cg1r_thrd =
989 g1->concurrent_g1_refine()->cg1rThread();
991 #if CARD_REPEAT_HISTO
992 gclog_or_tty->print_cr("\nG1 card_repeat count histogram: ");
993 gclog_or_tty->print_cr(" # of repeats --> # of cards with that number.");
994 card_repeat_count.print_on(gclog_or_tty);
995 #endif
997 if (FILTEROUTOFREGIONCLOSURE_DOHISTOGRAMCOUNT) {
998 gclog_or_tty->print_cr("\nG1 rem-set out-of-region histogram: ");
999 gclog_or_tty->print_cr(" # of CS ptrs --> # of cards with that number.");
1000 out_of_histo.print_on(gclog_or_tty);
1001 }
1002 gclog_or_tty->print_cr("\n Concurrent RS processed %d cards in "
1003 "%5.2fs.",
1004 _conc_refine_cards, cg1r_thrd->vtime_accum());
1006 DirtyCardQueueSet& dcqs = JavaThread::dirty_card_queue_set();
1007 jint tot_processed_buffers =
1008 dcqs.processed_buffers_mut() + dcqs.processed_buffers_rs_thread();
1009 gclog_or_tty->print_cr(" Of %d completed buffers:", tot_processed_buffers);
1010 gclog_or_tty->print_cr(" %8d (%5.1f%%) by conc RS thread.",
1011 dcqs.processed_buffers_rs_thread(),
1012 100.0*(float)dcqs.processed_buffers_rs_thread()/
1013 (float)tot_processed_buffers);
1014 gclog_or_tty->print_cr(" %8d (%5.1f%%) by mutator threads.",
1015 dcqs.processed_buffers_mut(),
1016 100.0*(float)dcqs.processed_buffers_mut()/
1017 (float)tot_processed_buffers);
1018 gclog_or_tty->print_cr(" Did %d concurrent refinement traversals.",
1019 _conc_refine_traversals);
1020 if (!G1RSBarrierUseQueue) {
1021 gclog_or_tty->print_cr(" Scanned %8.2f cards/traversal.",
1022 _conc_refine_traversals > 0 ?
1023 (float)_conc_refine_cards/(float)_conc_refine_traversals :
1024 0);
1025 }
1026 gclog_or_tty->print_cr("");
1027 if (G1UseHRIntoRS) {
1028 HRRSStatsIter blk;
1029 g1->heap_region_iterate(&blk);
1030 gclog_or_tty->print_cr(" Total heap region rem set sizes = " SIZE_FORMAT "K."
1031 " Max = " SIZE_FORMAT "K.",
1032 blk.total_mem_sz()/K, blk.max_mem_sz()/K);
1033 gclog_or_tty->print_cr(" Static structures = " SIZE_FORMAT "K,"
1034 " free_lists = " SIZE_FORMAT "K.",
1035 HeapRegionRemSet::static_mem_size()/K,
1036 HeapRegionRemSet::fl_mem_size()/K);
1037 gclog_or_tty->print_cr(" %d occupied cards represented.",
1038 blk.occupied());
1039 gclog_or_tty->print_cr(" Max sz region = [" PTR_FORMAT ", " PTR_FORMAT " )"
1040 ", cap = " SIZE_FORMAT "K, occ = " SIZE_FORMAT "K.",
1041 blk.max_mem_sz_region()->bottom(), blk.max_mem_sz_region()->end(),
1042 (blk.max_mem_sz_region()->rem_set()->mem_size() + K - 1)/K,
1043 (blk.max_mem_sz_region()->rem_set()->occupied() + K - 1)/K);
1044 gclog_or_tty->print_cr(" Did %d coarsenings.",
1045 HeapRegionRemSet::n_coarsenings());
1047 }
1048 }
1049 void HRInto_G1RemSet::prepare_for_verify() {
1050 if (G1HRRSFlushLogBuffersOnVerify &&
1051 (VerifyBeforeGC || VerifyAfterGC)
1052 && !_g1->full_collection()) {
1053 cleanupHRRS();
1054 _g1->set_refine_cte_cl_concurrency(false);
1055 if (SafepointSynchronize::is_at_safepoint()) {
1056 DirtyCardQueueSet& dcqs = JavaThread::dirty_card_queue_set();
1057 dcqs.concatenate_logs();
1058 }
1059 bool cg1r_use_cache = _cg1r->use_cache();
1060 _cg1r->set_use_cache(false);
1061 updateRS(0);
1062 _cg1r->set_use_cache(cg1r_use_cache);
1064 assert(JavaThread::dirty_card_queue_set().completed_buffers_num() == 0, "All should be consumed");
1065 }
1066 }