Mon, 28 Jun 2010 14:13:17 -0400
6944166: G1: explicit GCs are not always handled correctly
Summary: G1 was not handling explicit GCs correctly in many ways. It does now. See the CR for the list of improvements contained in this changeset.
Reviewed-by: iveresov, ysr, johnc
1 /*
2 * Copyright (c) 2001, 2009, 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 "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) { do_oop_work(p); }
69 virtual void do_oop( oop* p) { do_oop_work(p); }
70 template <class T> void do_oop_work(T* p) {
71 oop obj = oopDesc::load_decode_heap_oop(p);
72 if (_g1->obj_in_cs(obj)) _blk->do_oop(p);
73 }
74 bool apply_to_weak_ref_discovered_field() { return true; }
75 bool idempotent() { return true; }
76 };
78 class IntoCSRegionClosure: public HeapRegionClosure {
79 IntoCSOopClosure _blk;
80 G1CollectedHeap* _g1;
81 public:
82 IntoCSRegionClosure(G1CollectedHeap* g1, OopsInHeapRegionClosure* blk) :
83 _g1(g1), _blk(g1, blk) {}
84 bool doHeapRegion(HeapRegion* r) {
85 if (!r->in_collection_set()) {
86 _blk.set_region(r);
87 if (r->isHumongous()) {
88 if (r->startsHumongous()) {
89 oop obj = oop(r->bottom());
90 obj->oop_iterate(&_blk);
91 }
92 } else {
93 r->oop_before_save_marks_iterate(&_blk);
94 }
95 }
96 return false;
97 }
98 };
100 void
101 StupidG1RemSet::oops_into_collection_set_do(OopsInHeapRegionClosure* oc,
102 int worker_i) {
103 IntoCSRegionClosure rc(_g1, oc);
104 _g1->heap_region_iterate(&rc);
105 }
107 class VerifyRSCleanCardOopClosure: public OopClosure {
108 G1CollectedHeap* _g1;
109 public:
110 VerifyRSCleanCardOopClosure(G1CollectedHeap* g1) : _g1(g1) {}
112 virtual void do_oop(narrowOop* p) { do_oop_work(p); }
113 virtual void do_oop( oop* p) { do_oop_work(p); }
114 template <class T> void do_oop_work(T* p) {
115 oop obj = oopDesc::load_decode_heap_oop(p);
116 HeapRegion* to = _g1->heap_region_containing(obj);
117 guarantee(to == NULL || !to->in_collection_set(),
118 "Missed a rem set member.");
119 }
120 };
122 HRInto_G1RemSet::HRInto_G1RemSet(G1CollectedHeap* g1, CardTableModRefBS* ct_bs)
123 : G1RemSet(g1), _ct_bs(ct_bs), _g1p(_g1->g1_policy()),
124 _cg1r(g1->concurrent_g1_refine()),
125 _par_traversal_in_progress(false), _new_refs(NULL),
126 _cards_scanned(NULL), _total_cards_scanned(0)
127 {
128 _seq_task = new SubTasksDone(NumSeqTasks);
129 guarantee(n_workers() > 0, "There should be some workers");
130 _new_refs = NEW_C_HEAP_ARRAY(GrowableArray<OopOrNarrowOopStar>*, n_workers());
131 for (uint i = 0; i < n_workers(); i++) {
132 _new_refs[i] = new (ResourceObj::C_HEAP) GrowableArray<OopOrNarrowOopStar>(8192,true);
133 }
134 }
136 HRInto_G1RemSet::~HRInto_G1RemSet() {
137 delete _seq_task;
138 for (uint i = 0; i < n_workers(); i++) {
139 delete _new_refs[i];
140 }
141 FREE_C_HEAP_ARRAY(GrowableArray<OopOrNarrowOopStar>*, _new_refs);
142 }
144 void CountNonCleanMemRegionClosure::do_MemRegion(MemRegion mr) {
145 if (_g1->is_in_g1_reserved(mr.start())) {
146 _n += (int) ((mr.byte_size() / CardTableModRefBS::card_size));
147 if (_start_first == NULL) _start_first = mr.start();
148 }
149 }
151 class ScanRSClosure : public HeapRegionClosure {
152 size_t _cards_done, _cards;
153 G1CollectedHeap* _g1h;
154 OopsInHeapRegionClosure* _oc;
155 G1BlockOffsetSharedArray* _bot_shared;
156 CardTableModRefBS *_ct_bs;
157 int _worker_i;
158 int _block_size;
159 bool _try_claimed;
160 public:
161 ScanRSClosure(OopsInHeapRegionClosure* oc, int worker_i) :
162 _oc(oc),
163 _cards(0),
164 _cards_done(0),
165 _worker_i(worker_i),
166 _try_claimed(false)
167 {
168 _g1h = G1CollectedHeap::heap();
169 _bot_shared = _g1h->bot_shared();
170 _ct_bs = (CardTableModRefBS*) (_g1h->barrier_set());
171 _block_size = MAX2<int>(G1RSetScanBlockSize, 1);
172 }
174 void set_try_claimed() { _try_claimed = true; }
176 void scanCard(size_t index, HeapRegion *r) {
177 _cards_done++;
178 DirtyCardToOopClosure* cl =
179 r->new_dcto_closure(_oc,
180 CardTableModRefBS::Precise,
181 HeapRegionDCTOC::IntoCSFilterKind);
183 // Set the "from" region in the closure.
184 _oc->set_region(r);
185 HeapWord* card_start = _bot_shared->address_for_index(index);
186 HeapWord* card_end = card_start + G1BlockOffsetSharedArray::N_words;
187 Space *sp = SharedHeap::heap()->space_containing(card_start);
188 MemRegion sm_region;
189 if (ParallelGCThreads > 0) {
190 // first find the used area
191 sm_region = sp->used_region_at_save_marks();
192 } else {
193 // The closure is not idempotent. We shouldn't look at objects
194 // allocated during the GC.
195 sm_region = sp->used_region_at_save_marks();
196 }
197 MemRegion mr = sm_region.intersection(MemRegion(card_start,card_end));
198 if (!mr.is_empty()) {
199 cl->do_MemRegion(mr);
200 }
201 }
203 void printCard(HeapRegion* card_region, size_t card_index,
204 HeapWord* card_start) {
205 gclog_or_tty->print_cr("T %d Region [" PTR_FORMAT ", " PTR_FORMAT ") "
206 "RS names card %p: "
207 "[" PTR_FORMAT ", " PTR_FORMAT ")",
208 _worker_i,
209 card_region->bottom(), card_region->end(),
210 card_index,
211 card_start, card_start + G1BlockOffsetSharedArray::N_words);
212 }
214 bool doHeapRegion(HeapRegion* r) {
215 assert(r->in_collection_set(), "should only be called on elements of CS.");
216 HeapRegionRemSet* hrrs = r->rem_set();
217 if (hrrs->iter_is_complete()) return false; // All done.
218 if (!_try_claimed && !hrrs->claim_iter()) return false;
219 _g1h->push_dirty_cards_region(r);
220 // If we didn't return above, then
221 // _try_claimed || r->claim_iter()
222 // is true: either we're supposed to work on claimed-but-not-complete
223 // regions, or we successfully claimed the region.
224 HeapRegionRemSetIterator* iter = _g1h->rem_set_iterator(_worker_i);
225 hrrs->init_iterator(iter);
226 size_t card_index;
228 // We claim cards in block so as to recude the contention. The block size is determined by
229 // the G1RSetScanBlockSize parameter.
230 size_t jump_to_card = hrrs->iter_claimed_next(_block_size);
231 for (size_t current_card = 0; iter->has_next(card_index); current_card++) {
232 if (current_card >= jump_to_card + _block_size) {
233 jump_to_card = hrrs->iter_claimed_next(_block_size);
234 }
235 if (current_card < jump_to_card) continue;
236 HeapWord* card_start = _g1h->bot_shared()->address_for_index(card_index);
237 #if 0
238 gclog_or_tty->print("Rem set iteration yielded card [" PTR_FORMAT ", " PTR_FORMAT ").\n",
239 card_start, card_start + CardTableModRefBS::card_size_in_words);
240 #endif
242 HeapRegion* card_region = _g1h->heap_region_containing(card_start);
243 assert(card_region != NULL, "Yielding cards not in the heap?");
244 _cards++;
246 if (!card_region->is_on_dirty_cards_region_list()) {
247 _g1h->push_dirty_cards_region(card_region);
248 }
250 // If the card is dirty, then we will scan it during updateRS.
251 if (!card_region->in_collection_set() && !_ct_bs->is_card_dirty(card_index)) {
252 // We make the card as "claimed" lazily (so races are possible but they're benign),
253 // which reduces the number of duplicate scans (the rsets of the regions in the cset
254 // can intersect).
255 if (!_ct_bs->is_card_claimed(card_index)) {
256 _ct_bs->set_card_claimed(card_index);
257 scanCard(card_index, card_region);
258 }
259 }
260 }
261 if (!_try_claimed) {
262 hrrs->set_iter_complete();
263 }
264 return false;
265 }
266 // Set all cards back to clean.
267 void cleanup() {_g1h->cleanUpCardTable();}
268 size_t cards_done() { return _cards_done;}
269 size_t cards_looked_up() { return _cards;}
270 };
272 // We want the parallel threads to start their scanning at
273 // different collection set regions to avoid contention.
274 // If we have:
275 // n collection set regions
276 // p threads
277 // Then thread t will start at region t * floor (n/p)
279 HeapRegion* HRInto_G1RemSet::calculateStartRegion(int worker_i) {
280 HeapRegion* result = _g1p->collection_set();
281 if (ParallelGCThreads > 0) {
282 size_t cs_size = _g1p->collection_set_size();
283 int n_workers = _g1->workers()->total_workers();
284 size_t cs_spans = cs_size / n_workers;
285 size_t ind = cs_spans * worker_i;
286 for (size_t i = 0; i < ind; i++)
287 result = result->next_in_collection_set();
288 }
289 return result;
290 }
292 void HRInto_G1RemSet::scanRS(OopsInHeapRegionClosure* oc, int worker_i) {
293 double rs_time_start = os::elapsedTime();
294 HeapRegion *startRegion = calculateStartRegion(worker_i);
296 ScanRSClosure scanRScl(oc, worker_i);
297 _g1->collection_set_iterate_from(startRegion, &scanRScl);
298 scanRScl.set_try_claimed();
299 _g1->collection_set_iterate_from(startRegion, &scanRScl);
301 double scan_rs_time_sec = os::elapsedTime() - rs_time_start;
303 assert( _cards_scanned != NULL, "invariant" );
304 _cards_scanned[worker_i] = scanRScl.cards_done();
306 _g1p->record_scan_rs_time(worker_i, scan_rs_time_sec * 1000.0);
307 }
309 void HRInto_G1RemSet::updateRS(int worker_i) {
310 ConcurrentG1Refine* cg1r = _g1->concurrent_g1_refine();
312 double start = os::elapsedTime();
313 // Apply the appropriate closure to all remaining log entries.
314 _g1->iterate_dirty_card_closure(false, worker_i);
315 // Now there should be no dirty cards.
316 if (G1RSLogCheckCardTable) {
317 CountNonCleanMemRegionClosure cl(_g1);
318 _ct_bs->mod_card_iterate(&cl);
319 // XXX This isn't true any more: keeping cards of young regions
320 // marked dirty broke it. Need some reasonable fix.
321 guarantee(cl.n() == 0, "Card table should be clean.");
322 }
324 _g1p->record_update_rs_time(worker_i, (os::elapsedTime() - start) * 1000.0);
325 }
327 #ifndef PRODUCT
328 class PrintRSClosure : public HeapRegionClosure {
329 int _count;
330 public:
331 PrintRSClosure() : _count(0) {}
332 bool doHeapRegion(HeapRegion* r) {
333 HeapRegionRemSet* hrrs = r->rem_set();
334 _count += (int) hrrs->occupied();
335 if (hrrs->occupied() == 0) {
336 gclog_or_tty->print("Heap Region [" PTR_FORMAT ", " PTR_FORMAT ") "
337 "has no remset entries\n",
338 r->bottom(), r->end());
339 } else {
340 gclog_or_tty->print("Printing rem set for heap region [" PTR_FORMAT ", " PTR_FORMAT ")\n",
341 r->bottom(), r->end());
342 r->print();
343 hrrs->print();
344 gclog_or_tty->print("\nDone printing rem set\n");
345 }
346 return false;
347 }
348 int occupied() {return _count;}
349 };
350 #endif
352 class CountRSSizeClosure: public HeapRegionClosure {
353 size_t _n;
354 size_t _tot;
355 size_t _max;
356 HeapRegion* _max_r;
357 enum {
358 N = 20,
359 MIN = 6
360 };
361 int _histo[N];
362 public:
363 CountRSSizeClosure() : _n(0), _tot(0), _max(0), _max_r(NULL) {
364 for (int i = 0; i < N; i++) _histo[i] = 0;
365 }
366 bool doHeapRegion(HeapRegion* r) {
367 if (!r->continuesHumongous()) {
368 size_t occ = r->rem_set()->occupied();
369 _n++;
370 _tot += occ;
371 if (occ > _max) {
372 _max = occ;
373 _max_r = r;
374 }
375 // Fit it into a histo bin.
376 int s = 1 << MIN;
377 int i = 0;
378 while (occ > (size_t) s && i < (N-1)) {
379 s = s << 1;
380 i++;
381 }
382 _histo[i]++;
383 }
384 return false;
385 }
386 size_t n() { return _n; }
387 size_t tot() { return _tot; }
388 size_t mx() { return _max; }
389 HeapRegion* mxr() { return _max_r; }
390 void print_histo() {
391 int mx = N;
392 while (mx >= 0) {
393 if (_histo[mx-1] > 0) break;
394 mx--;
395 }
396 gclog_or_tty->print_cr("Number of regions with given RS sizes:");
397 gclog_or_tty->print_cr(" <= %8d %8d", 1 << MIN, _histo[0]);
398 for (int i = 1; i < mx-1; i++) {
399 gclog_or_tty->print_cr(" %8d - %8d %8d",
400 (1 << (MIN + i - 1)) + 1,
401 1 << (MIN + i),
402 _histo[i]);
403 }
404 gclog_or_tty->print_cr(" > %8d %8d", (1 << (MIN+mx-2))+1, _histo[mx-1]);
405 }
406 };
408 template <class T> void
409 HRInto_G1RemSet::scanNewRefsRS_work(OopsInHeapRegionClosure* oc,
410 int worker_i) {
411 double scan_new_refs_start_sec = os::elapsedTime();
412 G1CollectedHeap* g1h = G1CollectedHeap::heap();
413 CardTableModRefBS* ct_bs = (CardTableModRefBS*) (g1h->barrier_set());
414 for (int i = 0; i < _new_refs[worker_i]->length(); i++) {
415 T* p = (T*) _new_refs[worker_i]->at(i);
416 oop obj = oopDesc::load_decode_heap_oop(p);
417 // *p was in the collection set when p was pushed on "_new_refs", but
418 // another thread may have processed this location from an RS, so it
419 // might not point into the CS any longer. If so, it's obviously been
420 // processed, and we don't need to do anything further.
421 if (g1h->obj_in_cs(obj)) {
422 HeapRegion* r = g1h->heap_region_containing(p);
424 DEBUG_ONLY(HeapRegion* to = g1h->heap_region_containing(obj));
425 oc->set_region(r);
426 // If "p" has already been processed concurrently, this is
427 // idempotent.
428 oc->do_oop(p);
429 }
430 }
431 double scan_new_refs_time_ms = (os::elapsedTime() - scan_new_refs_start_sec) * 1000.0;
432 _g1p->record_scan_new_refs_time(worker_i, scan_new_refs_time_ms);
433 }
435 void HRInto_G1RemSet::cleanupHRRS() {
436 HeapRegionRemSet::cleanup();
437 }
439 void
440 HRInto_G1RemSet::oops_into_collection_set_do(OopsInHeapRegionClosure* oc,
441 int worker_i) {
442 #if CARD_REPEAT_HISTO
443 ct_freq_update_histo_and_reset();
444 #endif
445 if (worker_i == 0) {
446 _cg1r->clear_and_record_card_counts();
447 }
449 // Make this into a command-line flag...
450 if (G1RSCountHisto && (ParallelGCThreads == 0 || worker_i == 0)) {
451 CountRSSizeClosure count_cl;
452 _g1->heap_region_iterate(&count_cl);
453 gclog_or_tty->print_cr("Avg of %d RS counts is %f, max is %d, "
454 "max region is " PTR_FORMAT,
455 count_cl.n(), (float)count_cl.tot()/(float)count_cl.n(),
456 count_cl.mx(), count_cl.mxr());
457 count_cl.print_histo();
458 }
460 if (ParallelGCThreads > 0) {
461 // The two flags below were introduced temporarily to serialize
462 // the updating and scanning of remembered sets. There are some
463 // race conditions when these two operations are done in parallel
464 // and they are causing failures. When we resolve said race
465 // conditions, we'll revert back to parallel remembered set
466 // updating and scanning. See CRs 6677707 and 6677708.
467 if (G1UseParallelRSetUpdating || (worker_i == 0)) {
468 updateRS(worker_i);
469 scanNewRefsRS(oc, worker_i);
470 } else {
471 _g1p->record_update_rs_processed_buffers(worker_i, 0.0);
472 _g1p->record_update_rs_time(worker_i, 0.0);
473 _g1p->record_scan_new_refs_time(worker_i, 0.0);
474 }
475 if (G1UseParallelRSetScanning || (worker_i == 0)) {
476 scanRS(oc, worker_i);
477 } else {
478 _g1p->record_scan_rs_time(worker_i, 0.0);
479 }
480 } else {
481 assert(worker_i == 0, "invariant");
482 updateRS(0);
483 scanNewRefsRS(oc, 0);
484 scanRS(oc, 0);
485 }
486 }
488 void HRInto_G1RemSet::
489 prepare_for_oops_into_collection_set_do() {
490 #if G1_REM_SET_LOGGING
491 PrintRSClosure cl;
492 _g1->collection_set_iterate(&cl);
493 #endif
494 cleanupHRRS();
495 ConcurrentG1Refine* cg1r = _g1->concurrent_g1_refine();
496 _g1->set_refine_cte_cl_concurrency(false);
497 DirtyCardQueueSet& dcqs = JavaThread::dirty_card_queue_set();
498 dcqs.concatenate_logs();
500 assert(!_par_traversal_in_progress, "Invariant between iterations.");
501 if (ParallelGCThreads > 0) {
502 set_par_traversal(true);
503 _seq_task->set_par_threads((int)n_workers());
504 }
505 guarantee( _cards_scanned == NULL, "invariant" );
506 _cards_scanned = NEW_C_HEAP_ARRAY(size_t, n_workers());
507 for (uint i = 0; i < n_workers(); ++i) {
508 _cards_scanned[i] = 0;
509 }
510 _total_cards_scanned = 0;
511 }
514 class cleanUpIteratorsClosure : public HeapRegionClosure {
515 bool doHeapRegion(HeapRegion *r) {
516 HeapRegionRemSet* hrrs = r->rem_set();
517 hrrs->init_for_par_iteration();
518 return false;
519 }
520 };
522 class UpdateRSetOopsIntoCSImmediate : public OopClosure {
523 G1CollectedHeap* _g1;
524 public:
525 UpdateRSetOopsIntoCSImmediate(G1CollectedHeap* g1) : _g1(g1) { }
526 virtual void do_oop(narrowOop* p) { do_oop_work(p); }
527 virtual void do_oop( oop* p) { do_oop_work(p); }
528 template <class T> void do_oop_work(T* p) {
529 HeapRegion* to = _g1->heap_region_containing(oopDesc::load_decode_heap_oop(p));
530 if (to->in_collection_set()) {
531 to->rem_set()->add_reference(p, 0);
532 }
533 }
534 };
536 class UpdateRSetOopsIntoCSDeferred : public OopClosure {
537 G1CollectedHeap* _g1;
538 CardTableModRefBS* _ct_bs;
539 DirtyCardQueue* _dcq;
540 public:
541 UpdateRSetOopsIntoCSDeferred(G1CollectedHeap* g1, DirtyCardQueue* dcq) :
542 _g1(g1), _ct_bs((CardTableModRefBS*)_g1->barrier_set()), _dcq(dcq) { }
543 virtual void do_oop(narrowOop* p) { do_oop_work(p); }
544 virtual void do_oop( oop* p) { do_oop_work(p); }
545 template <class T> void do_oop_work(T* p) {
546 oop obj = oopDesc::load_decode_heap_oop(p);
547 if (_g1->obj_in_cs(obj)) {
548 size_t card_index = _ct_bs->index_for(p);
549 if (_ct_bs->mark_card_deferred(card_index)) {
550 _dcq->enqueue((jbyte*)_ct_bs->byte_for_index(card_index));
551 }
552 }
553 }
554 };
556 template <class T> void HRInto_G1RemSet::new_refs_iterate_work(OopClosure* cl) {
557 for (size_t i = 0; i < n_workers(); i++) {
558 for (int j = 0; j < _new_refs[i]->length(); j++) {
559 T* p = (T*) _new_refs[i]->at(j);
560 cl->do_oop(p);
561 }
562 }
563 }
565 void HRInto_G1RemSet::cleanup_after_oops_into_collection_set_do() {
566 guarantee( _cards_scanned != NULL, "invariant" );
567 _total_cards_scanned = 0;
568 for (uint i = 0; i < n_workers(); ++i)
569 _total_cards_scanned += _cards_scanned[i];
570 FREE_C_HEAP_ARRAY(size_t, _cards_scanned);
571 _cards_scanned = NULL;
572 // Cleanup after copy
573 #if G1_REM_SET_LOGGING
574 PrintRSClosure cl;
575 _g1->heap_region_iterate(&cl);
576 #endif
577 _g1->set_refine_cte_cl_concurrency(true);
578 cleanUpIteratorsClosure iterClosure;
579 _g1->collection_set_iterate(&iterClosure);
580 // Set all cards back to clean.
581 _g1->cleanUpCardTable();
583 if (ParallelGCThreads > 0) {
584 set_par_traversal(false);
585 }
587 if (_g1->evacuation_failed()) {
588 // Restore remembered sets for the regions pointing into
589 // the collection set.
590 if (G1DeferredRSUpdate) {
591 DirtyCardQueue dcq(&_g1->dirty_card_queue_set());
592 UpdateRSetOopsIntoCSDeferred deferred_update(_g1, &dcq);
593 new_refs_iterate(&deferred_update);
594 } else {
595 UpdateRSetOopsIntoCSImmediate immediate_update(_g1);
596 new_refs_iterate(&immediate_update);
597 }
598 }
599 for (uint i = 0; i < n_workers(); i++) {
600 _new_refs[i]->clear();
601 }
603 assert(!_par_traversal_in_progress, "Invariant between iterations.");
604 }
606 class UpdateRSObjectClosure: public ObjectClosure {
607 UpdateRSOopClosure* _update_rs_oop_cl;
608 public:
609 UpdateRSObjectClosure(UpdateRSOopClosure* update_rs_oop_cl) :
610 _update_rs_oop_cl(update_rs_oop_cl) {}
611 void do_object(oop obj) {
612 obj->oop_iterate(_update_rs_oop_cl);
613 }
615 };
617 class ScrubRSClosure: public HeapRegionClosure {
618 G1CollectedHeap* _g1h;
619 BitMap* _region_bm;
620 BitMap* _card_bm;
621 CardTableModRefBS* _ctbs;
622 public:
623 ScrubRSClosure(BitMap* region_bm, BitMap* card_bm) :
624 _g1h(G1CollectedHeap::heap()),
625 _region_bm(region_bm), _card_bm(card_bm),
626 _ctbs(NULL)
627 {
628 ModRefBarrierSet* bs = _g1h->mr_bs();
629 guarantee(bs->is_a(BarrierSet::CardTableModRef), "Precondition");
630 _ctbs = (CardTableModRefBS*)bs;
631 }
633 bool doHeapRegion(HeapRegion* r) {
634 if (!r->continuesHumongous()) {
635 r->rem_set()->scrub(_ctbs, _region_bm, _card_bm);
636 }
637 return false;
638 }
639 };
641 void HRInto_G1RemSet::scrub(BitMap* region_bm, BitMap* card_bm) {
642 ScrubRSClosure scrub_cl(region_bm, card_bm);
643 _g1->heap_region_iterate(&scrub_cl);
644 }
646 void HRInto_G1RemSet::scrub_par(BitMap* region_bm, BitMap* card_bm,
647 int worker_num, int claim_val) {
648 ScrubRSClosure scrub_cl(region_bm, card_bm);
649 _g1->heap_region_par_iterate_chunked(&scrub_cl, worker_num, claim_val);
650 }
653 static IntHistogram out_of_histo(50, 50);
655 void HRInto_G1RemSet::concurrentRefineOneCard_impl(jbyte* card_ptr, int worker_i) {
656 // Construct the region representing the card.
657 HeapWord* start = _ct_bs->addr_for(card_ptr);
658 // And find the region containing it.
659 HeapRegion* r = _g1->heap_region_containing(start);
660 assert(r != NULL, "unexpected null");
662 HeapWord* end = _ct_bs->addr_for(card_ptr + 1);
663 MemRegion dirtyRegion(start, end);
665 #if CARD_REPEAT_HISTO
666 init_ct_freq_table(_g1->g1_reserved_obj_bytes());
667 ct_freq_note_card(_ct_bs->index_for(start));
668 #endif
670 UpdateRSOopClosure update_rs_oop_cl(this, worker_i);
671 update_rs_oop_cl.set_from(r);
672 FilterOutOfRegionClosure filter_then_update_rs_oop_cl(r, &update_rs_oop_cl);
674 // Undirty the card.
675 *card_ptr = CardTableModRefBS::clean_card_val();
676 // We must complete this write before we do any of the reads below.
677 OrderAccess::storeload();
678 // And process it, being careful of unallocated portions of TLAB's.
679 HeapWord* stop_point =
680 r->oops_on_card_seq_iterate_careful(dirtyRegion,
681 &filter_then_update_rs_oop_cl);
682 // If stop_point is non-null, then we encountered an unallocated region
683 // (perhaps the unfilled portion of a TLAB.) For now, we'll dirty the
684 // card and re-enqueue: if we put off the card until a GC pause, then the
685 // unallocated portion will be filled in. Alternatively, we might try
686 // the full complexity of the technique used in "regular" precleaning.
687 if (stop_point != NULL) {
688 // The card might have gotten re-dirtied and re-enqueued while we
689 // worked. (In fact, it's pretty likely.)
690 if (*card_ptr != CardTableModRefBS::dirty_card_val()) {
691 *card_ptr = CardTableModRefBS::dirty_card_val();
692 MutexLockerEx x(Shared_DirtyCardQ_lock,
693 Mutex::_no_safepoint_check_flag);
694 DirtyCardQueue* sdcq =
695 JavaThread::dirty_card_queue_set().shared_dirty_card_queue();
696 sdcq->enqueue(card_ptr);
697 }
698 } else {
699 out_of_histo.add_entry(filter_then_update_rs_oop_cl.out_of_region());
700 _conc_refine_cards++;
701 }
702 }
704 void HRInto_G1RemSet::concurrentRefineOneCard(jbyte* card_ptr, int worker_i) {
705 // If the card is no longer dirty, nothing to do.
706 if (*card_ptr != CardTableModRefBS::dirty_card_val()) return;
708 // Construct the region representing the card.
709 HeapWord* start = _ct_bs->addr_for(card_ptr);
710 // And find the region containing it.
711 HeapRegion* r = _g1->heap_region_containing(start);
712 if (r == NULL) {
713 guarantee(_g1->is_in_permanent(start), "Or else where?");
714 return; // Not in the G1 heap (might be in perm, for example.)
715 }
716 // Why do we have to check here whether a card is on a young region,
717 // given that we dirty young regions and, as a result, the
718 // post-barrier is supposed to filter them out and never to enqueue
719 // them? When we allocate a new region as the "allocation region" we
720 // actually dirty its cards after we release the lock, since card
721 // dirtying while holding the lock was a performance bottleneck. So,
722 // as a result, it is possible for other threads to actually
723 // allocate objects in the region (after the acquire the lock)
724 // before all the cards on the region are dirtied. This is unlikely,
725 // and it doesn't happen often, but it can happen. So, the extra
726 // check below filters out those cards.
727 if (r->is_young()) {
728 return;
729 }
730 // While we are processing RSet buffers during the collection, we
731 // actually don't want to scan any cards on the collection set,
732 // since we don't want to update remebered sets with entries that
733 // point into the collection set, given that live objects from the
734 // collection set are about to move and such entries will be stale
735 // very soon. This change also deals with a reliability issue which
736 // involves scanning a card in the collection set and coming across
737 // an array that was being chunked and looking malformed. Note,
738 // however, that if evacuation fails, we have to scan any objects
739 // that were not moved and create any missing entries.
740 if (r->in_collection_set()) {
741 return;
742 }
744 // Should we defer processing the card?
745 //
746 // Previously the result from the insert_cache call would be
747 // either card_ptr (implying that card_ptr was currently "cold"),
748 // null (meaning we had inserted the card ptr into the "hot"
749 // cache, which had some headroom), or a "hot" card ptr
750 // extracted from the "hot" cache.
751 //
752 // Now that the _card_counts cache in the ConcurrentG1Refine
753 // instance is an evicting hash table, the result we get back
754 // could be from evicting the card ptr in an already occupied
755 // bucket (in which case we have replaced the card ptr in the
756 // bucket with card_ptr and "defer" is set to false). To avoid
757 // having a data structure (updates to which would need a lock)
758 // to hold these unprocessed dirty cards, we need to immediately
759 // process card_ptr. The actions needed to be taken on return
760 // from cache_insert are summarized in the following table:
761 //
762 // res defer action
763 // --------------------------------------------------------------
764 // null false card evicted from _card_counts & replaced with
765 // card_ptr; evicted ptr added to hot cache.
766 // No need to process res; immediately process card_ptr
767 //
768 // null true card not evicted from _card_counts; card_ptr added
769 // to hot cache.
770 // Nothing to do.
771 //
772 // non-null false card evicted from _card_counts & replaced with
773 // card_ptr; evicted ptr is currently "cold" or
774 // caused an eviction from the hot cache.
775 // Immediately process res; process card_ptr.
776 //
777 // non-null true card not evicted from _card_counts; card_ptr is
778 // currently cold, or caused an eviction from hot
779 // cache.
780 // Immediately process res; no need to process card_ptr.
782 jbyte* res = card_ptr;
783 bool defer = false;
784 if (_cg1r->use_cache()) {
785 jbyte* res = _cg1r->cache_insert(card_ptr, &defer);
786 if (res != NULL && (res != card_ptr || defer)) {
787 start = _ct_bs->addr_for(res);
788 r = _g1->heap_region_containing(start);
789 if (r == NULL) {
790 assert(_g1->is_in_permanent(start), "Or else where?");
791 } else {
792 guarantee(!r->is_young(), "It was evicted in the current minor cycle.");
793 // Process card pointer we get back from the hot card cache
794 concurrentRefineOneCard_impl(res, worker_i);
795 }
796 }
797 }
799 if (!defer) {
800 concurrentRefineOneCard_impl(card_ptr, worker_i);
801 }
802 }
804 class HRRSStatsIter: public HeapRegionClosure {
805 size_t _occupied;
806 size_t _total_mem_sz;
807 size_t _max_mem_sz;
808 HeapRegion* _max_mem_sz_region;
809 public:
810 HRRSStatsIter() :
811 _occupied(0),
812 _total_mem_sz(0),
813 _max_mem_sz(0),
814 _max_mem_sz_region(NULL)
815 {}
817 bool doHeapRegion(HeapRegion* r) {
818 if (r->continuesHumongous()) return false;
819 size_t mem_sz = r->rem_set()->mem_size();
820 if (mem_sz > _max_mem_sz) {
821 _max_mem_sz = mem_sz;
822 _max_mem_sz_region = r;
823 }
824 _total_mem_sz += mem_sz;
825 size_t occ = r->rem_set()->occupied();
826 _occupied += occ;
827 return false;
828 }
829 size_t total_mem_sz() { return _total_mem_sz; }
830 size_t max_mem_sz() { return _max_mem_sz; }
831 size_t occupied() { return _occupied; }
832 HeapRegion* max_mem_sz_region() { return _max_mem_sz_region; }
833 };
835 class PrintRSThreadVTimeClosure : public ThreadClosure {
836 public:
837 virtual void do_thread(Thread *t) {
838 ConcurrentG1RefineThread* crt = (ConcurrentG1RefineThread*) t;
839 gclog_or_tty->print(" %5.2f", crt->vtime_accum());
840 }
841 };
843 void HRInto_G1RemSet::print_summary_info() {
844 G1CollectedHeap* g1 = G1CollectedHeap::heap();
846 #if CARD_REPEAT_HISTO
847 gclog_or_tty->print_cr("\nG1 card_repeat count histogram: ");
848 gclog_or_tty->print_cr(" # of repeats --> # of cards with that number.");
849 card_repeat_count.print_on(gclog_or_tty);
850 #endif
852 if (FILTEROUTOFREGIONCLOSURE_DOHISTOGRAMCOUNT) {
853 gclog_or_tty->print_cr("\nG1 rem-set out-of-region histogram: ");
854 gclog_or_tty->print_cr(" # of CS ptrs --> # of cards with that number.");
855 out_of_histo.print_on(gclog_or_tty);
856 }
857 gclog_or_tty->print_cr("\n Concurrent RS processed %d cards",
858 _conc_refine_cards);
859 DirtyCardQueueSet& dcqs = JavaThread::dirty_card_queue_set();
860 jint tot_processed_buffers =
861 dcqs.processed_buffers_mut() + dcqs.processed_buffers_rs_thread();
862 gclog_or_tty->print_cr(" Of %d completed buffers:", tot_processed_buffers);
863 gclog_or_tty->print_cr(" %8d (%5.1f%%) by conc RS threads.",
864 dcqs.processed_buffers_rs_thread(),
865 100.0*(float)dcqs.processed_buffers_rs_thread()/
866 (float)tot_processed_buffers);
867 gclog_or_tty->print_cr(" %8d (%5.1f%%) by mutator threads.",
868 dcqs.processed_buffers_mut(),
869 100.0*(float)dcqs.processed_buffers_mut()/
870 (float)tot_processed_buffers);
871 gclog_or_tty->print_cr(" Conc RS threads times(s)");
872 PrintRSThreadVTimeClosure p;
873 gclog_or_tty->print(" ");
874 g1->concurrent_g1_refine()->threads_do(&p);
875 gclog_or_tty->print_cr("");
877 if (G1UseHRIntoRS) {
878 HRRSStatsIter blk;
879 g1->heap_region_iterate(&blk);
880 gclog_or_tty->print_cr(" Total heap region rem set sizes = " SIZE_FORMAT "K."
881 " Max = " SIZE_FORMAT "K.",
882 blk.total_mem_sz()/K, blk.max_mem_sz()/K);
883 gclog_or_tty->print_cr(" Static structures = " SIZE_FORMAT "K,"
884 " free_lists = " SIZE_FORMAT "K.",
885 HeapRegionRemSet::static_mem_size()/K,
886 HeapRegionRemSet::fl_mem_size()/K);
887 gclog_or_tty->print_cr(" %d occupied cards represented.",
888 blk.occupied());
889 gclog_or_tty->print_cr(" Max sz region = [" PTR_FORMAT ", " PTR_FORMAT " )"
890 ", cap = " SIZE_FORMAT "K, occ = " SIZE_FORMAT "K.",
891 blk.max_mem_sz_region()->bottom(), blk.max_mem_sz_region()->end(),
892 (blk.max_mem_sz_region()->rem_set()->mem_size() + K - 1)/K,
893 (blk.max_mem_sz_region()->rem_set()->occupied() + K - 1)/K);
894 gclog_or_tty->print_cr(" Did %d coarsenings.",
895 HeapRegionRemSet::n_coarsenings());
897 }
898 }
899 void HRInto_G1RemSet::prepare_for_verify() {
900 if (G1HRRSFlushLogBuffersOnVerify &&
901 (VerifyBeforeGC || VerifyAfterGC)
902 && !_g1->full_collection()) {
903 cleanupHRRS();
904 _g1->set_refine_cte_cl_concurrency(false);
905 if (SafepointSynchronize::is_at_safepoint()) {
906 DirtyCardQueueSet& dcqs = JavaThread::dirty_card_queue_set();
907 dcqs.concatenate_logs();
908 }
909 bool cg1r_use_cache = _cg1r->use_cache();
910 _cg1r->set_use_cache(false);
911 updateRS(0);
912 _cg1r->set_use_cache(cg1r_use_cache);
914 assert(JavaThread::dirty_card_queue_set().completed_buffers_num() == 0, "All should be consumed");
915 }
916 }