Fri, 18 Feb 2011 10:07:34 -0800
7020042: G1: Partially remove fix for 6994628
Summary: Disable reference discovery and processing during concurrent marking by disabling fix for 6994628.
Reviewed-by: tonyp, ysr
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
70 class IntoCSOopClosure: public OopsInHeapRegionClosure {
71 OopsInHeapRegionClosure* _blk;
72 G1CollectedHeap* _g1;
73 public:
74 IntoCSOopClosure(G1CollectedHeap* g1, OopsInHeapRegionClosure* blk) :
75 _g1(g1), _blk(blk) {}
76 void set_region(HeapRegion* from) {
77 _blk->set_region(from);
78 }
79 virtual void do_oop(narrowOop* p) { do_oop_work(p); }
80 virtual void do_oop( oop* p) { do_oop_work(p); }
81 template <class T> void do_oop_work(T* p) {
82 oop obj = oopDesc::load_decode_heap_oop(p);
83 if (_g1->obj_in_cs(obj)) _blk->do_oop(p);
84 }
85 bool apply_to_weak_ref_discovered_field() { return true; }
86 bool idempotent() { return true; }
87 };
89 class VerifyRSCleanCardOopClosure: public OopClosure {
90 G1CollectedHeap* _g1;
91 public:
92 VerifyRSCleanCardOopClosure(G1CollectedHeap* g1) : _g1(g1) {}
94 virtual void do_oop(narrowOop* p) { do_oop_work(p); }
95 virtual void do_oop( oop* p) { do_oop_work(p); }
96 template <class T> void do_oop_work(T* p) {
97 oop obj = oopDesc::load_decode_heap_oop(p);
98 HeapRegion* to = _g1->heap_region_containing(obj);
99 guarantee(to == NULL || !to->in_collection_set(),
100 "Missed a rem set member.");
101 }
102 };
104 G1RemSet::G1RemSet(G1CollectedHeap* g1, CardTableModRefBS* ct_bs)
105 : _g1(g1), _conc_refine_cards(0),
106 _ct_bs(ct_bs), _g1p(_g1->g1_policy()),
107 _cg1r(g1->concurrent_g1_refine()),
108 _cset_rs_update_cl(NULL),
109 _cards_scanned(NULL), _total_cards_scanned(0)
110 {
111 _seq_task = new SubTasksDone(NumSeqTasks);
112 guarantee(n_workers() > 0, "There should be some workers");
113 _cset_rs_update_cl = NEW_C_HEAP_ARRAY(OopsInHeapRegionClosure*, n_workers());
114 for (uint i = 0; i < n_workers(); i++) {
115 _cset_rs_update_cl[i] = NULL;
116 }
117 }
119 G1RemSet::~G1RemSet() {
120 delete _seq_task;
121 for (uint i = 0; i < n_workers(); i++) {
122 assert(_cset_rs_update_cl[i] == NULL, "it should be");
123 }
124 FREE_C_HEAP_ARRAY(OopsInHeapRegionClosure*, _cset_rs_update_cl);
125 }
127 void CountNonCleanMemRegionClosure::do_MemRegion(MemRegion mr) {
128 if (_g1->is_in_g1_reserved(mr.start())) {
129 _n += (int) ((mr.byte_size() / CardTableModRefBS::card_size));
130 if (_start_first == NULL) _start_first = mr.start();
131 }
132 }
134 class ScanRSClosure : public HeapRegionClosure {
135 size_t _cards_done, _cards;
136 G1CollectedHeap* _g1h;
137 OopsInHeapRegionClosure* _oc;
138 G1BlockOffsetSharedArray* _bot_shared;
139 CardTableModRefBS *_ct_bs;
140 int _worker_i;
141 int _block_size;
142 bool _try_claimed;
143 public:
144 ScanRSClosure(OopsInHeapRegionClosure* oc, int worker_i) :
145 _oc(oc),
146 _cards(0),
147 _cards_done(0),
148 _worker_i(worker_i),
149 _try_claimed(false)
150 {
151 _g1h = G1CollectedHeap::heap();
152 _bot_shared = _g1h->bot_shared();
153 _ct_bs = (CardTableModRefBS*) (_g1h->barrier_set());
154 _block_size = MAX2<int>(G1RSetScanBlockSize, 1);
155 }
157 void set_try_claimed() { _try_claimed = true; }
159 void scanCard(size_t index, HeapRegion *r) {
160 _cards_done++;
161 DirtyCardToOopClosure* cl =
162 r->new_dcto_closure(_oc,
163 CardTableModRefBS::Precise,
164 HeapRegionDCTOC::IntoCSFilterKind);
166 // Set the "from" region in the closure.
167 _oc->set_region(r);
168 HeapWord* card_start = _bot_shared->address_for_index(index);
169 HeapWord* card_end = card_start + G1BlockOffsetSharedArray::N_words;
170 Space *sp = SharedHeap::heap()->space_containing(card_start);
171 MemRegion sm_region;
172 if (ParallelGCThreads > 0) {
173 // first find the used area
174 sm_region = sp->used_region_at_save_marks();
175 } else {
176 // The closure is not idempotent. We shouldn't look at objects
177 // allocated during the GC.
178 sm_region = sp->used_region_at_save_marks();
179 }
180 MemRegion mr = sm_region.intersection(MemRegion(card_start,card_end));
181 if (!mr.is_empty()) {
182 cl->do_MemRegion(mr);
183 }
184 }
186 void printCard(HeapRegion* card_region, size_t card_index,
187 HeapWord* card_start) {
188 gclog_or_tty->print_cr("T %d Region [" PTR_FORMAT ", " PTR_FORMAT ") "
189 "RS names card %p: "
190 "[" PTR_FORMAT ", " PTR_FORMAT ")",
191 _worker_i,
192 card_region->bottom(), card_region->end(),
193 card_index,
194 card_start, card_start + G1BlockOffsetSharedArray::N_words);
195 }
197 bool doHeapRegion(HeapRegion* r) {
198 assert(r->in_collection_set(), "should only be called on elements of CS.");
199 HeapRegionRemSet* hrrs = r->rem_set();
200 if (hrrs->iter_is_complete()) return false; // All done.
201 if (!_try_claimed && !hrrs->claim_iter()) return false;
202 _g1h->push_dirty_cards_region(r);
203 // If we didn't return above, then
204 // _try_claimed || r->claim_iter()
205 // is true: either we're supposed to work on claimed-but-not-complete
206 // regions, or we successfully claimed the region.
207 HeapRegionRemSetIterator* iter = _g1h->rem_set_iterator(_worker_i);
208 hrrs->init_iterator(iter);
209 size_t card_index;
211 // We claim cards in block so as to recude the contention. The block size is determined by
212 // the G1RSetScanBlockSize parameter.
213 size_t jump_to_card = hrrs->iter_claimed_next(_block_size);
214 for (size_t current_card = 0; iter->has_next(card_index); current_card++) {
215 if (current_card >= jump_to_card + _block_size) {
216 jump_to_card = hrrs->iter_claimed_next(_block_size);
217 }
218 if (current_card < jump_to_card) continue;
219 HeapWord* card_start = _g1h->bot_shared()->address_for_index(card_index);
220 #if 0
221 gclog_or_tty->print("Rem set iteration yielded card [" PTR_FORMAT ", " PTR_FORMAT ").\n",
222 card_start, card_start + CardTableModRefBS::card_size_in_words);
223 #endif
225 HeapRegion* card_region = _g1h->heap_region_containing(card_start);
226 assert(card_region != NULL, "Yielding cards not in the heap?");
227 _cards++;
229 if (!card_region->is_on_dirty_cards_region_list()) {
230 _g1h->push_dirty_cards_region(card_region);
231 }
233 // If the card is dirty, then we will scan it during updateRS.
234 if (!card_region->in_collection_set() && !_ct_bs->is_card_dirty(card_index)) {
235 // We make the card as "claimed" lazily (so races are possible but they're benign),
236 // which reduces the number of duplicate scans (the rsets of the regions in the cset
237 // can intersect).
238 if (!_ct_bs->is_card_claimed(card_index)) {
239 _ct_bs->set_card_claimed(card_index);
240 scanCard(card_index, card_region);
241 }
242 }
243 }
244 if (!_try_claimed) {
245 hrrs->set_iter_complete();
246 }
247 return false;
248 }
249 // Set all cards back to clean.
250 void cleanup() {_g1h->cleanUpCardTable();}
251 size_t cards_done() { return _cards_done;}
252 size_t cards_looked_up() { return _cards;}
253 };
255 // We want the parallel threads to start their scanning at
256 // different collection set regions to avoid contention.
257 // If we have:
258 // n collection set regions
259 // p threads
260 // Then thread t will start at region t * floor (n/p)
262 HeapRegion* G1RemSet::calculateStartRegion(int worker_i) {
263 HeapRegion* result = _g1p->collection_set();
264 if (ParallelGCThreads > 0) {
265 size_t cs_size = _g1p->collection_set_size();
266 int n_workers = _g1->workers()->total_workers();
267 size_t cs_spans = cs_size / n_workers;
268 size_t ind = cs_spans * worker_i;
269 for (size_t i = 0; i < ind; i++)
270 result = result->next_in_collection_set();
271 }
272 return result;
273 }
275 void G1RemSet::scanRS(OopsInHeapRegionClosure* oc, int worker_i) {
276 double rs_time_start = os::elapsedTime();
277 HeapRegion *startRegion = calculateStartRegion(worker_i);
279 ScanRSClosure scanRScl(oc, worker_i);
280 _g1->collection_set_iterate_from(startRegion, &scanRScl);
281 scanRScl.set_try_claimed();
282 _g1->collection_set_iterate_from(startRegion, &scanRScl);
284 double scan_rs_time_sec = os::elapsedTime() - rs_time_start;
286 assert( _cards_scanned != NULL, "invariant" );
287 _cards_scanned[worker_i] = scanRScl.cards_done();
289 _g1p->record_scan_rs_time(worker_i, scan_rs_time_sec * 1000.0);
290 }
292 // Closure used for updating RSets and recording references that
293 // point into the collection set. Only called during an
294 // evacuation pause.
296 class RefineRecordRefsIntoCSCardTableEntryClosure: public CardTableEntryClosure {
297 G1RemSet* _g1rs;
298 DirtyCardQueue* _into_cset_dcq;
299 public:
300 RefineRecordRefsIntoCSCardTableEntryClosure(G1CollectedHeap* g1h,
301 DirtyCardQueue* into_cset_dcq) :
302 _g1rs(g1h->g1_rem_set()), _into_cset_dcq(into_cset_dcq)
303 {}
304 bool do_card_ptr(jbyte* card_ptr, int worker_i) {
305 // The only time we care about recording cards that
306 // contain references that point into the collection set
307 // is during RSet updating within an evacuation pause.
308 // In this case worker_i should be the id of a GC worker thread.
309 assert(SafepointSynchronize::is_at_safepoint(), "not during an evacuation pause");
310 assert(worker_i < (int) (ParallelGCThreads == 0 ? 1 : ParallelGCThreads), "should be a GC worker");
312 if (_g1rs->concurrentRefineOneCard(card_ptr, worker_i, true)) {
313 // 'card_ptr' contains references that point into the collection
314 // set. We need to record the card in the DCQS
315 // (G1CollectedHeap::into_cset_dirty_card_queue_set())
316 // that's used for that purpose.
317 //
318 // Enqueue the card
319 _into_cset_dcq->enqueue(card_ptr);
320 }
321 return true;
322 }
323 };
325 void G1RemSet::updateRS(DirtyCardQueue* into_cset_dcq, int worker_i) {
326 double start = os::elapsedTime();
327 // Apply the given closure to all remaining log entries.
328 RefineRecordRefsIntoCSCardTableEntryClosure into_cset_update_rs_cl(_g1, into_cset_dcq);
329 _g1->iterate_dirty_card_closure(&into_cset_update_rs_cl, into_cset_dcq, false, worker_i);
331 // Now there should be no dirty cards.
332 if (G1RSLogCheckCardTable) {
333 CountNonCleanMemRegionClosure cl(_g1);
334 _ct_bs->mod_card_iterate(&cl);
335 // XXX This isn't true any more: keeping cards of young regions
336 // marked dirty broke it. Need some reasonable fix.
337 guarantee(cl.n() == 0, "Card table should be clean.");
338 }
340 _g1p->record_update_rs_time(worker_i, (os::elapsedTime() - start) * 1000.0);
341 }
343 #ifndef PRODUCT
344 class PrintRSClosure : public HeapRegionClosure {
345 int _count;
346 public:
347 PrintRSClosure() : _count(0) {}
348 bool doHeapRegion(HeapRegion* r) {
349 HeapRegionRemSet* hrrs = r->rem_set();
350 _count += (int) hrrs->occupied();
351 if (hrrs->occupied() == 0) {
352 gclog_or_tty->print("Heap Region [" PTR_FORMAT ", " PTR_FORMAT ") "
353 "has no remset entries\n",
354 r->bottom(), r->end());
355 } else {
356 gclog_or_tty->print("Printing rem set for heap region [" PTR_FORMAT ", " PTR_FORMAT ")\n",
357 r->bottom(), r->end());
358 r->print();
359 hrrs->print();
360 gclog_or_tty->print("\nDone printing rem set\n");
361 }
362 return false;
363 }
364 int occupied() {return _count;}
365 };
366 #endif
368 class CountRSSizeClosure: public HeapRegionClosure {
369 size_t _n;
370 size_t _tot;
371 size_t _max;
372 HeapRegion* _max_r;
373 enum {
374 N = 20,
375 MIN = 6
376 };
377 int _histo[N];
378 public:
379 CountRSSizeClosure() : _n(0), _tot(0), _max(0), _max_r(NULL) {
380 for (int i = 0; i < N; i++) _histo[i] = 0;
381 }
382 bool doHeapRegion(HeapRegion* r) {
383 if (!r->continuesHumongous()) {
384 size_t occ = r->rem_set()->occupied();
385 _n++;
386 _tot += occ;
387 if (occ > _max) {
388 _max = occ;
389 _max_r = r;
390 }
391 // Fit it into a histo bin.
392 int s = 1 << MIN;
393 int i = 0;
394 while (occ > (size_t) s && i < (N-1)) {
395 s = s << 1;
396 i++;
397 }
398 _histo[i]++;
399 }
400 return false;
401 }
402 size_t n() { return _n; }
403 size_t tot() { return _tot; }
404 size_t mx() { return _max; }
405 HeapRegion* mxr() { return _max_r; }
406 void print_histo() {
407 int mx = N;
408 while (mx >= 0) {
409 if (_histo[mx-1] > 0) break;
410 mx--;
411 }
412 gclog_or_tty->print_cr("Number of regions with given RS sizes:");
413 gclog_or_tty->print_cr(" <= %8d %8d", 1 << MIN, _histo[0]);
414 for (int i = 1; i < mx-1; i++) {
415 gclog_or_tty->print_cr(" %8d - %8d %8d",
416 (1 << (MIN + i - 1)) + 1,
417 1 << (MIN + i),
418 _histo[i]);
419 }
420 gclog_or_tty->print_cr(" > %8d %8d", (1 << (MIN+mx-2))+1, _histo[mx-1]);
421 }
422 };
424 void G1RemSet::cleanupHRRS() {
425 HeapRegionRemSet::cleanup();
426 }
428 void G1RemSet::oops_into_collection_set_do(OopsInHeapRegionClosure* oc,
429 int worker_i) {
430 #if CARD_REPEAT_HISTO
431 ct_freq_update_histo_and_reset();
432 #endif
433 if (worker_i == 0) {
434 _cg1r->clear_and_record_card_counts();
435 }
437 // Make this into a command-line flag...
438 if (G1RSCountHisto && (ParallelGCThreads == 0 || worker_i == 0)) {
439 CountRSSizeClosure count_cl;
440 _g1->heap_region_iterate(&count_cl);
441 gclog_or_tty->print_cr("Avg of %d RS counts is %f, max is %d, "
442 "max region is " PTR_FORMAT,
443 count_cl.n(), (float)count_cl.tot()/(float)count_cl.n(),
444 count_cl.mx(), count_cl.mxr());
445 count_cl.print_histo();
446 }
448 // We cache the value of 'oc' closure into the appropriate slot in the
449 // _cset_rs_update_cl for this worker
450 assert(worker_i < (int)n_workers(), "sanity");
451 _cset_rs_update_cl[worker_i] = oc;
453 // A DirtyCardQueue that is used to hold cards containing references
454 // that point into the collection set. This DCQ is associated with a
455 // special DirtyCardQueueSet (see g1CollectedHeap.hpp). Under normal
456 // circumstances (i.e. the pause successfully completes), these cards
457 // are just discarded (there's no need to update the RSets of regions
458 // that were in the collection set - after the pause these regions
459 // are wholly 'free' of live objects. In the event of an evacuation
460 // failure the cards/buffers in this queue set are:
461 // * passed to the DirtyCardQueueSet that is used to manage deferred
462 // RSet updates, or
463 // * scanned for references that point into the collection set
464 // and the RSet of the corresponding region in the collection set
465 // is updated immediately.
466 DirtyCardQueue into_cset_dcq(&_g1->into_cset_dirty_card_queue_set());
468 assert((ParallelGCThreads > 0) || worker_i == 0, "invariant");
470 // The two flags below were introduced temporarily to serialize
471 // the updating and scanning of remembered sets. There are some
472 // race conditions when these two operations are done in parallel
473 // and they are causing failures. When we resolve said race
474 // conditions, we'll revert back to parallel remembered set
475 // updating and scanning. See CRs 6677707 and 6677708.
476 if (G1UseParallelRSetUpdating || (worker_i == 0)) {
477 updateRS(&into_cset_dcq, worker_i);
478 } else {
479 _g1p->record_update_rs_processed_buffers(worker_i, 0.0);
480 _g1p->record_update_rs_time(worker_i, 0.0);
481 }
482 if (G1UseParallelRSetScanning || (worker_i == 0)) {
483 scanRS(oc, worker_i);
484 } else {
485 _g1p->record_scan_rs_time(worker_i, 0.0);
486 }
488 // We now clear the cached values of _cset_rs_update_cl for this worker
489 _cset_rs_update_cl[worker_i] = NULL;
490 }
492 void G1RemSet::prepare_for_oops_into_collection_set_do() {
493 #if G1_REM_SET_LOGGING
494 PrintRSClosure cl;
495 _g1->collection_set_iterate(&cl);
496 #endif
497 cleanupHRRS();
498 ConcurrentG1Refine* cg1r = _g1->concurrent_g1_refine();
499 _g1->set_refine_cte_cl_concurrency(false);
500 DirtyCardQueueSet& dcqs = JavaThread::dirty_card_queue_set();
501 dcqs.concatenate_logs();
503 if (ParallelGCThreads > 0) {
504 _seq_task->set_n_threads((int)n_workers());
505 }
506 guarantee( _cards_scanned == NULL, "invariant" );
507 _cards_scanned = NEW_C_HEAP_ARRAY(size_t, n_workers());
508 for (uint i = 0; i < n_workers(); ++i) {
509 _cards_scanned[i] = 0;
510 }
511 _total_cards_scanned = 0;
512 }
515 class cleanUpIteratorsClosure : public HeapRegionClosure {
516 bool doHeapRegion(HeapRegion *r) {
517 HeapRegionRemSet* hrrs = r->rem_set();
518 hrrs->init_for_par_iteration();
519 return false;
520 }
521 };
523 // This closure, applied to a DirtyCardQueueSet, is used to immediately
524 // update the RSets for the regions in the CSet. For each card it iterates
525 // through the oops which coincide with that card. It scans the reference
526 // fields in each oop; when it finds an oop that points into the collection
527 // set, the RSet for the region containing the referenced object is updated.
528 class UpdateRSetCardTableEntryIntoCSetClosure: public CardTableEntryClosure {
529 G1CollectedHeap* _g1;
530 CardTableModRefBS* _ct_bs;
531 public:
532 UpdateRSetCardTableEntryIntoCSetClosure(G1CollectedHeap* g1,
533 CardTableModRefBS* bs):
534 _g1(g1), _ct_bs(bs)
535 { }
537 bool do_card_ptr(jbyte* card_ptr, int worker_i) {
538 // Construct the region representing the card.
539 HeapWord* start = _ct_bs->addr_for(card_ptr);
540 // And find the region containing it.
541 HeapRegion* r = _g1->heap_region_containing(start);
542 assert(r != NULL, "unexpected null");
544 // Scan oops in the card looking for references into the collection set
545 HeapWord* end = _ct_bs->addr_for(card_ptr + 1);
546 MemRegion scanRegion(start, end);
548 UpdateRSetImmediate update_rs_cl(_g1->g1_rem_set());
549 FilterIntoCSClosure update_rs_cset_oop_cl(NULL, _g1, &update_rs_cl);
550 FilterOutOfRegionClosure filter_then_update_rs_cset_oop_cl(r, &update_rs_cset_oop_cl);
552 // We can pass false as the "filter_young" parameter here as:
553 // * we should be in a STW pause,
554 // * the DCQS to which this closure is applied is used to hold
555 // references that point into the collection set from the prior
556 // RSet updating,
557 // * the post-write barrier shouldn't be logging updates to young
558 // regions (but there is a situation where this can happen - see
559 // the comment in G1RemSet::concurrentRefineOneCard below -
560 // that should not be applicable here), and
561 // * during actual RSet updating, the filtering of cards in young
562 // regions in HeapRegion::oops_on_card_seq_iterate_careful is
563 // employed.
564 // As a result, when this closure is applied to "refs into cset"
565 // DCQS, we shouldn't see any cards in young regions.
566 update_rs_cl.set_region(r);
567 HeapWord* stop_point =
568 r->oops_on_card_seq_iterate_careful(scanRegion,
569 &filter_then_update_rs_cset_oop_cl,
570 false /* filter_young */);
572 // Since this is performed in the event of an evacuation failure, we
573 // we shouldn't see a non-null stop point
574 assert(stop_point == NULL, "saw an unallocated region");
575 return true;
576 }
577 };
579 void G1RemSet::cleanup_after_oops_into_collection_set_do() {
580 guarantee( _cards_scanned != NULL, "invariant" );
581 _total_cards_scanned = 0;
582 for (uint i = 0; i < n_workers(); ++i)
583 _total_cards_scanned += _cards_scanned[i];
584 FREE_C_HEAP_ARRAY(size_t, _cards_scanned);
585 _cards_scanned = NULL;
586 // Cleanup after copy
587 #if G1_REM_SET_LOGGING
588 PrintRSClosure cl;
589 _g1->heap_region_iterate(&cl);
590 #endif
591 _g1->set_refine_cte_cl_concurrency(true);
592 cleanUpIteratorsClosure iterClosure;
593 _g1->collection_set_iterate(&iterClosure);
594 // Set all cards back to clean.
595 _g1->cleanUpCardTable();
597 DirtyCardQueueSet& into_cset_dcqs = _g1->into_cset_dirty_card_queue_set();
598 int into_cset_n_buffers = into_cset_dcqs.completed_buffers_num();
600 if (_g1->evacuation_failed()) {
601 // Restore remembered sets for the regions pointing into the collection set.
603 if (G1DeferredRSUpdate) {
604 // If deferred RS updates are enabled then we just need to transfer
605 // the completed buffers from (a) the DirtyCardQueueSet used to hold
606 // cards that contain references that point into the collection set
607 // to (b) the DCQS used to hold the deferred RS updates
608 _g1->dirty_card_queue_set().merge_bufferlists(&into_cset_dcqs);
609 } else {
611 CardTableModRefBS* bs = (CardTableModRefBS*)_g1->barrier_set();
612 UpdateRSetCardTableEntryIntoCSetClosure update_rs_cset_immediate(_g1, bs);
614 int n_completed_buffers = 0;
615 while (into_cset_dcqs.apply_closure_to_completed_buffer(&update_rs_cset_immediate,
616 0, 0, true)) {
617 n_completed_buffers++;
618 }
619 assert(n_completed_buffers == into_cset_n_buffers, "missed some buffers");
620 }
621 }
623 // Free any completed buffers in the DirtyCardQueueSet used to hold cards
624 // which contain references that point into the collection.
625 _g1->into_cset_dirty_card_queue_set().clear();
626 assert(_g1->into_cset_dirty_card_queue_set().completed_buffers_num() == 0,
627 "all buffers should be freed");
628 _g1->into_cset_dirty_card_queue_set().clear_n_completed_buffers();
629 }
631 class ScrubRSClosure: public HeapRegionClosure {
632 G1CollectedHeap* _g1h;
633 BitMap* _region_bm;
634 BitMap* _card_bm;
635 CardTableModRefBS* _ctbs;
636 public:
637 ScrubRSClosure(BitMap* region_bm, BitMap* card_bm) :
638 _g1h(G1CollectedHeap::heap()),
639 _region_bm(region_bm), _card_bm(card_bm),
640 _ctbs(NULL)
641 {
642 ModRefBarrierSet* bs = _g1h->mr_bs();
643 guarantee(bs->is_a(BarrierSet::CardTableModRef), "Precondition");
644 _ctbs = (CardTableModRefBS*)bs;
645 }
647 bool doHeapRegion(HeapRegion* r) {
648 if (!r->continuesHumongous()) {
649 r->rem_set()->scrub(_ctbs, _region_bm, _card_bm);
650 }
651 return false;
652 }
653 };
655 void G1RemSet::scrub(BitMap* region_bm, BitMap* card_bm) {
656 ScrubRSClosure scrub_cl(region_bm, card_bm);
657 _g1->heap_region_iterate(&scrub_cl);
658 }
660 void G1RemSet::scrub_par(BitMap* region_bm, BitMap* card_bm,
661 int worker_num, int claim_val) {
662 ScrubRSClosure scrub_cl(region_bm, card_bm);
663 _g1->heap_region_par_iterate_chunked(&scrub_cl, worker_num, claim_val);
664 }
667 static IntHistogram out_of_histo(50, 50);
669 class TriggerClosure : public OopClosure {
670 bool _trigger;
671 public:
672 TriggerClosure() : _trigger(false) { }
673 bool value() const { return _trigger; }
674 template <class T> void do_oop_nv(T* p) { _trigger = true; }
675 virtual void do_oop(oop* p) { do_oop_nv(p); }
676 virtual void do_oop(narrowOop* p) { do_oop_nv(p); }
677 };
679 class InvokeIfNotTriggeredClosure: public OopClosure {
680 TriggerClosure* _t;
681 OopClosure* _oc;
682 public:
683 InvokeIfNotTriggeredClosure(TriggerClosure* t, OopClosure* oc):
684 _t(t), _oc(oc) { }
685 template <class T> void do_oop_nv(T* p) {
686 if (!_t->value()) _oc->do_oop(p);
687 }
688 virtual void do_oop(oop* p) { do_oop_nv(p); }
689 virtual void do_oop(narrowOop* p) { do_oop_nv(p); }
690 };
692 class Mux2Closure : public OopClosure {
693 OopClosure* _c1;
694 OopClosure* _c2;
695 public:
696 Mux2Closure(OopClosure *c1, OopClosure *c2) : _c1(c1), _c2(c2) { }
697 template <class T> void do_oop_nv(T* p) {
698 _c1->do_oop(p); _c2->do_oop(p);
699 }
700 virtual void do_oop(oop* p) { do_oop_nv(p); }
701 virtual void do_oop(narrowOop* p) { do_oop_nv(p); }
702 };
704 bool G1RemSet::concurrentRefineOneCard_impl(jbyte* card_ptr, int worker_i,
705 bool check_for_refs_into_cset) {
706 // Construct the region representing the card.
707 HeapWord* start = _ct_bs->addr_for(card_ptr);
708 // And find the region containing it.
709 HeapRegion* r = _g1->heap_region_containing(start);
710 assert(r != NULL, "unexpected null");
712 HeapWord* end = _ct_bs->addr_for(card_ptr + 1);
713 MemRegion dirtyRegion(start, end);
715 #if CARD_REPEAT_HISTO
716 init_ct_freq_table(_g1->max_capacity());
717 ct_freq_note_card(_ct_bs->index_for(start));
718 #endif
720 assert(!check_for_refs_into_cset || _cset_rs_update_cl[worker_i] != NULL, "sanity");
721 UpdateRSOrPushRefOopClosure update_rs_oop_cl(_g1,
722 _g1->g1_rem_set(),
723 _cset_rs_update_cl[worker_i],
724 check_for_refs_into_cset,
725 worker_i);
726 update_rs_oop_cl.set_from(r);
728 TriggerClosure trigger_cl;
729 FilterIntoCSClosure into_cs_cl(NULL, _g1, &trigger_cl);
730 InvokeIfNotTriggeredClosure invoke_cl(&trigger_cl, &into_cs_cl);
731 Mux2Closure mux(&invoke_cl, &update_rs_oop_cl);
733 FilterOutOfRegionClosure filter_then_update_rs_oop_cl(r,
734 (check_for_refs_into_cset ?
735 (OopClosure*)&mux :
736 (OopClosure*)&update_rs_oop_cl));
738 // Undirty the card.
739 *card_ptr = CardTableModRefBS::clean_card_val();
740 // We must complete this write before we do any of the reads below.
741 OrderAccess::storeload();
742 // And process it, being careful of unallocated portions of TLAB's.
744 // The region for the current card may be a young region. The
745 // current card may have been a card that was evicted from the
746 // card cache. When the card was inserted into the cache, we had
747 // determined that its region was non-young. While in the cache,
748 // the region may have been freed during a cleanup pause, reallocated
749 // and tagged as young.
750 //
751 // We wish to filter out cards for such a region but the current
752 // thread, if we're running conucrrently, may "see" the young type
753 // change at any time (so an earlier "is_young" check may pass or
754 // fail arbitrarily). We tell the iteration code to perform this
755 // filtering when it has been determined that there has been an actual
756 // allocation in this region and making it safe to check the young type.
757 bool filter_young = true;
759 HeapWord* stop_point =
760 r->oops_on_card_seq_iterate_careful(dirtyRegion,
761 &filter_then_update_rs_oop_cl,
762 filter_young);
764 // If stop_point is non-null, then we encountered an unallocated region
765 // (perhaps the unfilled portion of a TLAB.) For now, we'll dirty the
766 // card and re-enqueue: if we put off the card until a GC pause, then the
767 // unallocated portion will be filled in. Alternatively, we might try
768 // the full complexity of the technique used in "regular" precleaning.
769 if (stop_point != NULL) {
770 // The card might have gotten re-dirtied and re-enqueued while we
771 // worked. (In fact, it's pretty likely.)
772 if (*card_ptr != CardTableModRefBS::dirty_card_val()) {
773 *card_ptr = CardTableModRefBS::dirty_card_val();
774 MutexLockerEx x(Shared_DirtyCardQ_lock,
775 Mutex::_no_safepoint_check_flag);
776 DirtyCardQueue* sdcq =
777 JavaThread::dirty_card_queue_set().shared_dirty_card_queue();
778 sdcq->enqueue(card_ptr);
779 }
780 } else {
781 out_of_histo.add_entry(filter_then_update_rs_oop_cl.out_of_region());
782 _conc_refine_cards++;
783 }
785 return trigger_cl.value();
786 }
788 bool G1RemSet::concurrentRefineOneCard(jbyte* card_ptr, int worker_i,
789 bool check_for_refs_into_cset) {
790 // If the card is no longer dirty, nothing to do.
791 if (*card_ptr != CardTableModRefBS::dirty_card_val()) {
792 // No need to return that this card contains refs that point
793 // into the collection set.
794 return false;
795 }
797 // Construct the region representing the card.
798 HeapWord* start = _ct_bs->addr_for(card_ptr);
799 // And find the region containing it.
800 HeapRegion* r = _g1->heap_region_containing(start);
801 if (r == NULL) {
802 guarantee(_g1->is_in_permanent(start), "Or else where?");
803 // Again no need to return that this card contains refs that
804 // point into the collection set.
805 return false; // Not in the G1 heap (might be in perm, for example.)
806 }
807 // Why do we have to check here whether a card is on a young region,
808 // given that we dirty young regions and, as a result, the
809 // post-barrier is supposed to filter them out and never to enqueue
810 // them? When we allocate a new region as the "allocation region" we
811 // actually dirty its cards after we release the lock, since card
812 // dirtying while holding the lock was a performance bottleneck. So,
813 // as a result, it is possible for other threads to actually
814 // allocate objects in the region (after the acquire the lock)
815 // before all the cards on the region are dirtied. This is unlikely,
816 // and it doesn't happen often, but it can happen. So, the extra
817 // check below filters out those cards.
818 if (r->is_young()) {
819 return false;
820 }
821 // While we are processing RSet buffers during the collection, we
822 // actually don't want to scan any cards on the collection set,
823 // since we don't want to update remebered sets with entries that
824 // point into the collection set, given that live objects from the
825 // collection set are about to move and such entries will be stale
826 // very soon. This change also deals with a reliability issue which
827 // involves scanning a card in the collection set and coming across
828 // an array that was being chunked and looking malformed. Note,
829 // however, that if evacuation fails, we have to scan any objects
830 // that were not moved and create any missing entries.
831 if (r->in_collection_set()) {
832 return false;
833 }
835 // Should we defer processing the card?
836 //
837 // Previously the result from the insert_cache call would be
838 // either card_ptr (implying that card_ptr was currently "cold"),
839 // null (meaning we had inserted the card ptr into the "hot"
840 // cache, which had some headroom), or a "hot" card ptr
841 // extracted from the "hot" cache.
842 //
843 // Now that the _card_counts cache in the ConcurrentG1Refine
844 // instance is an evicting hash table, the result we get back
845 // could be from evicting the card ptr in an already occupied
846 // bucket (in which case we have replaced the card ptr in the
847 // bucket with card_ptr and "defer" is set to false). To avoid
848 // having a data structure (updates to which would need a lock)
849 // to hold these unprocessed dirty cards, we need to immediately
850 // process card_ptr. The actions needed to be taken on return
851 // from cache_insert are summarized in the following table:
852 //
853 // res defer action
854 // --------------------------------------------------------------
855 // null false card evicted from _card_counts & replaced with
856 // card_ptr; evicted ptr added to hot cache.
857 // No need to process res; immediately process card_ptr
858 //
859 // null true card not evicted from _card_counts; card_ptr added
860 // to hot cache.
861 // Nothing to do.
862 //
863 // non-null false card evicted from _card_counts & replaced with
864 // card_ptr; evicted ptr is currently "cold" or
865 // caused an eviction from the hot cache.
866 // Immediately process res; process card_ptr.
867 //
868 // non-null true card not evicted from _card_counts; card_ptr is
869 // currently cold, or caused an eviction from hot
870 // cache.
871 // Immediately process res; no need to process card_ptr.
874 jbyte* res = card_ptr;
875 bool defer = false;
877 // This gets set to true if the card being refined has references
878 // that point into the collection set.
879 bool oops_into_cset = false;
881 if (_cg1r->use_cache()) {
882 jbyte* res = _cg1r->cache_insert(card_ptr, &defer);
883 if (res != NULL && (res != card_ptr || defer)) {
884 start = _ct_bs->addr_for(res);
885 r = _g1->heap_region_containing(start);
886 if (r == NULL) {
887 assert(_g1->is_in_permanent(start), "Or else where?");
888 } else {
889 // Checking whether the region we got back from the cache
890 // is young here is inappropriate. The region could have been
891 // freed, reallocated and tagged as young while in the cache.
892 // Hence we could see its young type change at any time.
893 //
894 // Process card pointer we get back from the hot card cache. This
895 // will check whether the region containing the card is young
896 // _after_ checking that the region has been allocated from.
897 oops_into_cset = concurrentRefineOneCard_impl(res, worker_i,
898 false /* check_for_refs_into_cset */);
899 // The above call to concurrentRefineOneCard_impl is only
900 // performed if the hot card cache is enabled. This cache is
901 // disabled during an evacuation pause - which is the only
902 // time when we need know if the card contains references
903 // that point into the collection set. Also when the hot card
904 // cache is enabled, this code is executed by the concurrent
905 // refine threads - rather than the GC worker threads - and
906 // concurrentRefineOneCard_impl will return false.
907 assert(!oops_into_cset, "should not see true here");
908 }
909 }
910 }
912 if (!defer) {
913 oops_into_cset =
914 concurrentRefineOneCard_impl(card_ptr, worker_i, check_for_refs_into_cset);
915 // We should only be detecting that the card contains references
916 // that point into the collection set if the current thread is
917 // a GC worker thread.
918 assert(!oops_into_cset || SafepointSynchronize::is_at_safepoint(),
919 "invalid result at non safepoint");
920 }
921 return oops_into_cset;
922 }
924 class HRRSStatsIter: public HeapRegionClosure {
925 size_t _occupied;
926 size_t _total_mem_sz;
927 size_t _max_mem_sz;
928 HeapRegion* _max_mem_sz_region;
929 public:
930 HRRSStatsIter() :
931 _occupied(0),
932 _total_mem_sz(0),
933 _max_mem_sz(0),
934 _max_mem_sz_region(NULL)
935 {}
937 bool doHeapRegion(HeapRegion* r) {
938 if (r->continuesHumongous()) return false;
939 size_t mem_sz = r->rem_set()->mem_size();
940 if (mem_sz > _max_mem_sz) {
941 _max_mem_sz = mem_sz;
942 _max_mem_sz_region = r;
943 }
944 _total_mem_sz += mem_sz;
945 size_t occ = r->rem_set()->occupied();
946 _occupied += occ;
947 return false;
948 }
949 size_t total_mem_sz() { return _total_mem_sz; }
950 size_t max_mem_sz() { return _max_mem_sz; }
951 size_t occupied() { return _occupied; }
952 HeapRegion* max_mem_sz_region() { return _max_mem_sz_region; }
953 };
955 class PrintRSThreadVTimeClosure : public ThreadClosure {
956 public:
957 virtual void do_thread(Thread *t) {
958 ConcurrentG1RefineThread* crt = (ConcurrentG1RefineThread*) t;
959 gclog_or_tty->print(" %5.2f", crt->vtime_accum());
960 }
961 };
963 void G1RemSet::print_summary_info() {
964 G1CollectedHeap* g1 = G1CollectedHeap::heap();
966 #if CARD_REPEAT_HISTO
967 gclog_or_tty->print_cr("\nG1 card_repeat count histogram: ");
968 gclog_or_tty->print_cr(" # of repeats --> # of cards with that number.");
969 card_repeat_count.print_on(gclog_or_tty);
970 #endif
972 if (FILTEROUTOFREGIONCLOSURE_DOHISTOGRAMCOUNT) {
973 gclog_or_tty->print_cr("\nG1 rem-set out-of-region histogram: ");
974 gclog_or_tty->print_cr(" # of CS ptrs --> # of cards with that number.");
975 out_of_histo.print_on(gclog_or_tty);
976 }
977 gclog_or_tty->print_cr("\n Concurrent RS processed %d cards",
978 _conc_refine_cards);
979 DirtyCardQueueSet& dcqs = JavaThread::dirty_card_queue_set();
980 jint tot_processed_buffers =
981 dcqs.processed_buffers_mut() + dcqs.processed_buffers_rs_thread();
982 gclog_or_tty->print_cr(" Of %d completed buffers:", tot_processed_buffers);
983 gclog_or_tty->print_cr(" %8d (%5.1f%%) by conc RS threads.",
984 dcqs.processed_buffers_rs_thread(),
985 100.0*(float)dcqs.processed_buffers_rs_thread()/
986 (float)tot_processed_buffers);
987 gclog_or_tty->print_cr(" %8d (%5.1f%%) by mutator threads.",
988 dcqs.processed_buffers_mut(),
989 100.0*(float)dcqs.processed_buffers_mut()/
990 (float)tot_processed_buffers);
991 gclog_or_tty->print_cr(" Conc RS threads times(s)");
992 PrintRSThreadVTimeClosure p;
993 gclog_or_tty->print(" ");
994 g1->concurrent_g1_refine()->threads_do(&p);
995 gclog_or_tty->print_cr("");
997 HRRSStatsIter blk;
998 g1->heap_region_iterate(&blk);
999 gclog_or_tty->print_cr(" Total heap region rem set sizes = " SIZE_FORMAT "K."
1000 " Max = " SIZE_FORMAT "K.",
1001 blk.total_mem_sz()/K, blk.max_mem_sz()/K);
1002 gclog_or_tty->print_cr(" Static structures = " SIZE_FORMAT "K,"
1003 " free_lists = " SIZE_FORMAT "K.",
1004 HeapRegionRemSet::static_mem_size()/K,
1005 HeapRegionRemSet::fl_mem_size()/K);
1006 gclog_or_tty->print_cr(" %d occupied cards represented.",
1007 blk.occupied());
1008 gclog_or_tty->print_cr(" Max sz region = [" PTR_FORMAT ", " PTR_FORMAT " )"
1009 ", cap = " SIZE_FORMAT "K, occ = " SIZE_FORMAT "K.",
1010 blk.max_mem_sz_region()->bottom(), blk.max_mem_sz_region()->end(),
1011 (blk.max_mem_sz_region()->rem_set()->mem_size() + K - 1)/K,
1012 (blk.max_mem_sz_region()->rem_set()->occupied() + K - 1)/K);
1013 gclog_or_tty->print_cr(" Did %d coarsenings.", HeapRegionRemSet::n_coarsenings());
1014 }
1016 void G1RemSet::prepare_for_verify() {
1017 if (G1HRRSFlushLogBuffersOnVerify &&
1018 (VerifyBeforeGC || VerifyAfterGC)
1019 && !_g1->full_collection()) {
1020 cleanupHRRS();
1021 _g1->set_refine_cte_cl_concurrency(false);
1022 if (SafepointSynchronize::is_at_safepoint()) {
1023 DirtyCardQueueSet& dcqs = JavaThread::dirty_card_queue_set();
1024 dcqs.concatenate_logs();
1025 }
1026 bool cg1r_use_cache = _cg1r->use_cache();
1027 _cg1r->set_use_cache(false);
1028 DirtyCardQueue into_cset_dcq(&_g1->into_cset_dirty_card_queue_set());
1029 updateRS(&into_cset_dcq, 0);
1030 _g1->into_cset_dirty_card_queue_set().clear();
1031 _cg1r->set_use_cache(cg1r_use_cache);
1033 assert(JavaThread::dirty_card_queue_set().completed_buffers_num() == 0, "All should be consumed");
1034 }
1035 }