Mon, 20 Sep 2010 14:38:38 -0700
6984287: Regularize how GC parallel workers are specified.
Summary: Associate number of GC workers with the workgang as opposed to the task.
Reviewed-by: johnc, ysr
1 /*
2 * Copyright (c) 2001, 2010, 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 _traversal_in_progress(false),
126 _cset_rs_update_cl(NULL),
127 _cards_scanned(NULL), _total_cards_scanned(0)
128 {
129 _seq_task = new SubTasksDone(NumSeqTasks);
130 guarantee(n_workers() > 0, "There should be some workers");
131 _cset_rs_update_cl = NEW_C_HEAP_ARRAY(OopsInHeapRegionClosure*, n_workers());
132 for (uint i = 0; i < n_workers(); i++) {
133 _cset_rs_update_cl[i] = NULL;
134 }
135 }
137 HRInto_G1RemSet::~HRInto_G1RemSet() {
138 delete _seq_task;
139 for (uint i = 0; i < n_workers(); i++) {
140 assert(_cset_rs_update_cl[i] == NULL, "it should be");
141 }
142 FREE_C_HEAP_ARRAY(OopsInHeapRegionClosure*, _cset_rs_update_cl);
143 }
145 void CountNonCleanMemRegionClosure::do_MemRegion(MemRegion mr) {
146 if (_g1->is_in_g1_reserved(mr.start())) {
147 _n += (int) ((mr.byte_size() / CardTableModRefBS::card_size));
148 if (_start_first == NULL) _start_first = mr.start();
149 }
150 }
152 class ScanRSClosure : public HeapRegionClosure {
153 size_t _cards_done, _cards;
154 G1CollectedHeap* _g1h;
155 OopsInHeapRegionClosure* _oc;
156 G1BlockOffsetSharedArray* _bot_shared;
157 CardTableModRefBS *_ct_bs;
158 int _worker_i;
159 int _block_size;
160 bool _try_claimed;
161 public:
162 ScanRSClosure(OopsInHeapRegionClosure* oc, int worker_i) :
163 _oc(oc),
164 _cards(0),
165 _cards_done(0),
166 _worker_i(worker_i),
167 _try_claimed(false)
168 {
169 _g1h = G1CollectedHeap::heap();
170 _bot_shared = _g1h->bot_shared();
171 _ct_bs = (CardTableModRefBS*) (_g1h->barrier_set());
172 _block_size = MAX2<int>(G1RSetScanBlockSize, 1);
173 }
175 void set_try_claimed() { _try_claimed = true; }
177 void scanCard(size_t index, HeapRegion *r) {
178 _cards_done++;
179 DirtyCardToOopClosure* cl =
180 r->new_dcto_closure(_oc,
181 CardTableModRefBS::Precise,
182 HeapRegionDCTOC::IntoCSFilterKind);
184 // Set the "from" region in the closure.
185 _oc->set_region(r);
186 HeapWord* card_start = _bot_shared->address_for_index(index);
187 HeapWord* card_end = card_start + G1BlockOffsetSharedArray::N_words;
188 Space *sp = SharedHeap::heap()->space_containing(card_start);
189 MemRegion sm_region;
190 if (ParallelGCThreads > 0) {
191 // first find the used area
192 sm_region = sp->used_region_at_save_marks();
193 } else {
194 // The closure is not idempotent. We shouldn't look at objects
195 // allocated during the GC.
196 sm_region = sp->used_region_at_save_marks();
197 }
198 MemRegion mr = sm_region.intersection(MemRegion(card_start,card_end));
199 if (!mr.is_empty()) {
200 cl->do_MemRegion(mr);
201 }
202 }
204 void printCard(HeapRegion* card_region, size_t card_index,
205 HeapWord* card_start) {
206 gclog_or_tty->print_cr("T %d Region [" PTR_FORMAT ", " PTR_FORMAT ") "
207 "RS names card %p: "
208 "[" PTR_FORMAT ", " PTR_FORMAT ")",
209 _worker_i,
210 card_region->bottom(), card_region->end(),
211 card_index,
212 card_start, card_start + G1BlockOffsetSharedArray::N_words);
213 }
215 bool doHeapRegion(HeapRegion* r) {
216 assert(r->in_collection_set(), "should only be called on elements of CS.");
217 HeapRegionRemSet* hrrs = r->rem_set();
218 if (hrrs->iter_is_complete()) return false; // All done.
219 if (!_try_claimed && !hrrs->claim_iter()) return false;
220 _g1h->push_dirty_cards_region(r);
221 // If we didn't return above, then
222 // _try_claimed || r->claim_iter()
223 // is true: either we're supposed to work on claimed-but-not-complete
224 // regions, or we successfully claimed the region.
225 HeapRegionRemSetIterator* iter = _g1h->rem_set_iterator(_worker_i);
226 hrrs->init_iterator(iter);
227 size_t card_index;
229 // We claim cards in block so as to recude the contention. The block size is determined by
230 // the G1RSetScanBlockSize parameter.
231 size_t jump_to_card = hrrs->iter_claimed_next(_block_size);
232 for (size_t current_card = 0; iter->has_next(card_index); current_card++) {
233 if (current_card >= jump_to_card + _block_size) {
234 jump_to_card = hrrs->iter_claimed_next(_block_size);
235 }
236 if (current_card < jump_to_card) continue;
237 HeapWord* card_start = _g1h->bot_shared()->address_for_index(card_index);
238 #if 0
239 gclog_or_tty->print("Rem set iteration yielded card [" PTR_FORMAT ", " PTR_FORMAT ").\n",
240 card_start, card_start + CardTableModRefBS::card_size_in_words);
241 #endif
243 HeapRegion* card_region = _g1h->heap_region_containing(card_start);
244 assert(card_region != NULL, "Yielding cards not in the heap?");
245 _cards++;
247 if (!card_region->is_on_dirty_cards_region_list()) {
248 _g1h->push_dirty_cards_region(card_region);
249 }
251 // If the card is dirty, then we will scan it during updateRS.
252 if (!card_region->in_collection_set() && !_ct_bs->is_card_dirty(card_index)) {
253 // We make the card as "claimed" lazily (so races are possible but they're benign),
254 // which reduces the number of duplicate scans (the rsets of the regions in the cset
255 // can intersect).
256 if (!_ct_bs->is_card_claimed(card_index)) {
257 _ct_bs->set_card_claimed(card_index);
258 scanCard(card_index, card_region);
259 }
260 }
261 }
262 if (!_try_claimed) {
263 hrrs->set_iter_complete();
264 }
265 return false;
266 }
267 // Set all cards back to clean.
268 void cleanup() {_g1h->cleanUpCardTable();}
269 size_t cards_done() { return _cards_done;}
270 size_t cards_looked_up() { return _cards;}
271 };
273 // We want the parallel threads to start their scanning at
274 // different collection set regions to avoid contention.
275 // If we have:
276 // n collection set regions
277 // p threads
278 // Then thread t will start at region t * floor (n/p)
280 HeapRegion* HRInto_G1RemSet::calculateStartRegion(int worker_i) {
281 HeapRegion* result = _g1p->collection_set();
282 if (ParallelGCThreads > 0) {
283 size_t cs_size = _g1p->collection_set_size();
284 int n_workers = _g1->workers()->total_workers();
285 size_t cs_spans = cs_size / n_workers;
286 size_t ind = cs_spans * worker_i;
287 for (size_t i = 0; i < ind; i++)
288 result = result->next_in_collection_set();
289 }
290 return result;
291 }
293 void HRInto_G1RemSet::scanRS(OopsInHeapRegionClosure* oc, int worker_i) {
294 double rs_time_start = os::elapsedTime();
295 HeapRegion *startRegion = calculateStartRegion(worker_i);
297 ScanRSClosure scanRScl(oc, worker_i);
298 _g1->collection_set_iterate_from(startRegion, &scanRScl);
299 scanRScl.set_try_claimed();
300 _g1->collection_set_iterate_from(startRegion, &scanRScl);
302 double scan_rs_time_sec = os::elapsedTime() - rs_time_start;
304 assert( _cards_scanned != NULL, "invariant" );
305 _cards_scanned[worker_i] = scanRScl.cards_done();
307 _g1p->record_scan_rs_time(worker_i, scan_rs_time_sec * 1000.0);
308 }
310 // Closure used for updating RSets and recording references that
311 // point into the collection set. Only called during an
312 // evacuation pause.
314 class RefineRecordRefsIntoCSCardTableEntryClosure: public CardTableEntryClosure {
315 G1RemSet* _g1rs;
316 DirtyCardQueue* _into_cset_dcq;
317 public:
318 RefineRecordRefsIntoCSCardTableEntryClosure(G1CollectedHeap* g1h,
319 DirtyCardQueue* into_cset_dcq) :
320 _g1rs(g1h->g1_rem_set()), _into_cset_dcq(into_cset_dcq)
321 {}
322 bool do_card_ptr(jbyte* card_ptr, int worker_i) {
323 // The only time we care about recording cards that
324 // contain references that point into the collection set
325 // is during RSet updating within an evacuation pause.
326 // In this case worker_i should be the id of a GC worker thread.
327 assert(SafepointSynchronize::is_at_safepoint(), "not during an evacuation pause");
328 assert(worker_i < (int) DirtyCardQueueSet::num_par_ids(), "should be a GC worker");
330 if (_g1rs->concurrentRefineOneCard(card_ptr, worker_i, true)) {
331 // 'card_ptr' contains references that point into the collection
332 // set. We need to record the card in the DCQS
333 // (G1CollectedHeap::into_cset_dirty_card_queue_set())
334 // that's used for that purpose.
335 //
336 // Enqueue the card
337 _into_cset_dcq->enqueue(card_ptr);
338 }
339 return true;
340 }
341 };
343 void HRInto_G1RemSet::updateRS(DirtyCardQueue* into_cset_dcq, int worker_i) {
344 double start = os::elapsedTime();
345 // Apply the given closure to all remaining log entries.
346 RefineRecordRefsIntoCSCardTableEntryClosure into_cset_update_rs_cl(_g1, into_cset_dcq);
347 _g1->iterate_dirty_card_closure(&into_cset_update_rs_cl, into_cset_dcq, false, worker_i);
349 // Now there should be no dirty cards.
350 if (G1RSLogCheckCardTable) {
351 CountNonCleanMemRegionClosure cl(_g1);
352 _ct_bs->mod_card_iterate(&cl);
353 // XXX This isn't true any more: keeping cards of young regions
354 // marked dirty broke it. Need some reasonable fix.
355 guarantee(cl.n() == 0, "Card table should be clean.");
356 }
358 _g1p->record_update_rs_time(worker_i, (os::elapsedTime() - start) * 1000.0);
359 }
361 #ifndef PRODUCT
362 class PrintRSClosure : public HeapRegionClosure {
363 int _count;
364 public:
365 PrintRSClosure() : _count(0) {}
366 bool doHeapRegion(HeapRegion* r) {
367 HeapRegionRemSet* hrrs = r->rem_set();
368 _count += (int) hrrs->occupied();
369 if (hrrs->occupied() == 0) {
370 gclog_or_tty->print("Heap Region [" PTR_FORMAT ", " PTR_FORMAT ") "
371 "has no remset entries\n",
372 r->bottom(), r->end());
373 } else {
374 gclog_or_tty->print("Printing rem set for heap region [" PTR_FORMAT ", " PTR_FORMAT ")\n",
375 r->bottom(), r->end());
376 r->print();
377 hrrs->print();
378 gclog_or_tty->print("\nDone printing rem set\n");
379 }
380 return false;
381 }
382 int occupied() {return _count;}
383 };
384 #endif
386 class CountRSSizeClosure: public HeapRegionClosure {
387 size_t _n;
388 size_t _tot;
389 size_t _max;
390 HeapRegion* _max_r;
391 enum {
392 N = 20,
393 MIN = 6
394 };
395 int _histo[N];
396 public:
397 CountRSSizeClosure() : _n(0), _tot(0), _max(0), _max_r(NULL) {
398 for (int i = 0; i < N; i++) _histo[i] = 0;
399 }
400 bool doHeapRegion(HeapRegion* r) {
401 if (!r->continuesHumongous()) {
402 size_t occ = r->rem_set()->occupied();
403 _n++;
404 _tot += occ;
405 if (occ > _max) {
406 _max = occ;
407 _max_r = r;
408 }
409 // Fit it into a histo bin.
410 int s = 1 << MIN;
411 int i = 0;
412 while (occ > (size_t) s && i < (N-1)) {
413 s = s << 1;
414 i++;
415 }
416 _histo[i]++;
417 }
418 return false;
419 }
420 size_t n() { return _n; }
421 size_t tot() { return _tot; }
422 size_t mx() { return _max; }
423 HeapRegion* mxr() { return _max_r; }
424 void print_histo() {
425 int mx = N;
426 while (mx >= 0) {
427 if (_histo[mx-1] > 0) break;
428 mx--;
429 }
430 gclog_or_tty->print_cr("Number of regions with given RS sizes:");
431 gclog_or_tty->print_cr(" <= %8d %8d", 1 << MIN, _histo[0]);
432 for (int i = 1; i < mx-1; i++) {
433 gclog_or_tty->print_cr(" %8d - %8d %8d",
434 (1 << (MIN + i - 1)) + 1,
435 1 << (MIN + i),
436 _histo[i]);
437 }
438 gclog_or_tty->print_cr(" > %8d %8d", (1 << (MIN+mx-2))+1, _histo[mx-1]);
439 }
440 };
442 void HRInto_G1RemSet::cleanupHRRS() {
443 HeapRegionRemSet::cleanup();
444 }
446 void
447 HRInto_G1RemSet::oops_into_collection_set_do(OopsInHeapRegionClosure* oc,
448 int worker_i) {
449 #if CARD_REPEAT_HISTO
450 ct_freq_update_histo_and_reset();
451 #endif
452 if (worker_i == 0) {
453 _cg1r->clear_and_record_card_counts();
454 }
456 // Make this into a command-line flag...
457 if (G1RSCountHisto && (ParallelGCThreads == 0 || worker_i == 0)) {
458 CountRSSizeClosure count_cl;
459 _g1->heap_region_iterate(&count_cl);
460 gclog_or_tty->print_cr("Avg of %d RS counts is %f, max is %d, "
461 "max region is " PTR_FORMAT,
462 count_cl.n(), (float)count_cl.tot()/(float)count_cl.n(),
463 count_cl.mx(), count_cl.mxr());
464 count_cl.print_histo();
465 }
467 // We cache the value of 'oc' closure into the appropriate slot in the
468 // _cset_rs_update_cl for this worker
469 assert(worker_i < (int)n_workers(), "sanity");
470 _cset_rs_update_cl[worker_i] = oc;
472 // A DirtyCardQueue that is used to hold cards containing references
473 // that point into the collection set. This DCQ is associated with a
474 // special DirtyCardQueueSet (see g1CollectedHeap.hpp). Under normal
475 // circumstances (i.e. the pause successfully completes), these cards
476 // are just discarded (there's no need to update the RSets of regions
477 // that were in the collection set - after the pause these regions
478 // are wholly 'free' of live objects. In the event of an evacuation
479 // failure the cards/buffers in this queue set are:
480 // * passed to the DirtyCardQueueSet that is used to manage deferred
481 // RSet updates, or
482 // * scanned for references that point into the collection set
483 // and the RSet of the corresponding region in the collection set
484 // is updated immediately.
485 DirtyCardQueue into_cset_dcq(&_g1->into_cset_dirty_card_queue_set());
487 assert((ParallelGCThreads > 0) || worker_i == 0, "invariant");
489 // The two flags below were introduced temporarily to serialize
490 // the updating and scanning of remembered sets. There are some
491 // race conditions when these two operations are done in parallel
492 // and they are causing failures. When we resolve said race
493 // conditions, we'll revert back to parallel remembered set
494 // updating and scanning. See CRs 6677707 and 6677708.
495 if (G1UseParallelRSetUpdating || (worker_i == 0)) {
496 updateRS(&into_cset_dcq, worker_i);
497 } else {
498 _g1p->record_update_rs_processed_buffers(worker_i, 0.0);
499 _g1p->record_update_rs_time(worker_i, 0.0);
500 }
501 if (G1UseParallelRSetScanning || (worker_i == 0)) {
502 scanRS(oc, worker_i);
503 } else {
504 _g1p->record_scan_rs_time(worker_i, 0.0);
505 }
507 // We now clear the cached values of _cset_rs_update_cl for this worker
508 _cset_rs_update_cl[worker_i] = NULL;
509 }
511 void HRInto_G1RemSet::
512 prepare_for_oops_into_collection_set_do() {
513 #if G1_REM_SET_LOGGING
514 PrintRSClosure cl;
515 _g1->collection_set_iterate(&cl);
516 #endif
517 cleanupHRRS();
518 ConcurrentG1Refine* cg1r = _g1->concurrent_g1_refine();
519 _g1->set_refine_cte_cl_concurrency(false);
520 DirtyCardQueueSet& dcqs = JavaThread::dirty_card_queue_set();
521 dcqs.concatenate_logs();
523 assert(!_traversal_in_progress, "Invariant between iterations.");
524 set_traversal(true);
525 if (ParallelGCThreads > 0) {
526 _seq_task->set_n_threads((int)n_workers());
527 }
528 guarantee( _cards_scanned == NULL, "invariant" );
529 _cards_scanned = NEW_C_HEAP_ARRAY(size_t, n_workers());
530 for (uint i = 0; i < n_workers(); ++i) {
531 _cards_scanned[i] = 0;
532 }
533 _total_cards_scanned = 0;
534 }
537 class cleanUpIteratorsClosure : public HeapRegionClosure {
538 bool doHeapRegion(HeapRegion *r) {
539 HeapRegionRemSet* hrrs = r->rem_set();
540 hrrs->init_for_par_iteration();
541 return false;
542 }
543 };
545 // This closure, applied to a DirtyCardQueueSet, is used to immediately
546 // update the RSets for the regions in the CSet. For each card it iterates
547 // through the oops which coincide with that card. It scans the reference
548 // fields in each oop; when it finds an oop that points into the collection
549 // set, the RSet for the region containing the referenced object is updated.
550 // Note: _par_traversal_in_progress in the G1RemSet must be FALSE; otherwise
551 // the UpdateRSetImmediate closure will cause cards to be enqueued on to
552 // the DCQS that we're iterating over, causing an infinite loop.
553 class UpdateRSetCardTableEntryIntoCSetClosure: public CardTableEntryClosure {
554 G1CollectedHeap* _g1;
555 CardTableModRefBS* _ct_bs;
556 public:
557 UpdateRSetCardTableEntryIntoCSetClosure(G1CollectedHeap* g1,
558 CardTableModRefBS* bs):
559 _g1(g1), _ct_bs(bs)
560 { }
562 bool do_card_ptr(jbyte* card_ptr, int worker_i) {
563 // Construct the region representing the card.
564 HeapWord* start = _ct_bs->addr_for(card_ptr);
565 // And find the region containing it.
566 HeapRegion* r = _g1->heap_region_containing(start);
567 assert(r != NULL, "unexpected null");
569 // Scan oops in the card looking for references into the collection set
570 HeapWord* end = _ct_bs->addr_for(card_ptr + 1);
571 MemRegion scanRegion(start, end);
573 UpdateRSetImmediate update_rs_cl(_g1->g1_rem_set());
574 FilterIntoCSClosure update_rs_cset_oop_cl(NULL, _g1, &update_rs_cl);
575 FilterOutOfRegionClosure filter_then_update_rs_cset_oop_cl(r, &update_rs_cset_oop_cl);
577 // We can pass false as the "filter_young" parameter here as:
578 // * we should be in a STW pause,
579 // * the DCQS to which this closure is applied is used to hold
580 // references that point into the collection set from the prior
581 // RSet updating,
582 // * the post-write barrier shouldn't be logging updates to young
583 // regions (but there is a situation where this can happen - see
584 // the comment in HRInto_G1RemSet::concurrentRefineOneCard below -
585 // that should not be applicable here), and
586 // * during actual RSet updating, the filtering of cards in young
587 // regions in HeapRegion::oops_on_card_seq_iterate_careful is
588 // employed.
589 // As a result, when this closure is applied to "refs into cset"
590 // DCQS, we shouldn't see any cards in young regions.
591 update_rs_cl.set_region(r);
592 HeapWord* stop_point =
593 r->oops_on_card_seq_iterate_careful(scanRegion,
594 &filter_then_update_rs_cset_oop_cl,
595 false /* filter_young */);
597 // Since this is performed in the event of an evacuation failure, we
598 // we shouldn't see a non-null stop point
599 assert(stop_point == NULL, "saw an unallocated region");
600 return true;
601 }
602 };
604 void HRInto_G1RemSet::cleanup_after_oops_into_collection_set_do() {
605 guarantee( _cards_scanned != NULL, "invariant" );
606 _total_cards_scanned = 0;
607 for (uint i = 0; i < n_workers(); ++i)
608 _total_cards_scanned += _cards_scanned[i];
609 FREE_C_HEAP_ARRAY(size_t, _cards_scanned);
610 _cards_scanned = NULL;
611 // Cleanup after copy
612 #if G1_REM_SET_LOGGING
613 PrintRSClosure cl;
614 _g1->heap_region_iterate(&cl);
615 #endif
616 _g1->set_refine_cte_cl_concurrency(true);
617 cleanUpIteratorsClosure iterClosure;
618 _g1->collection_set_iterate(&iterClosure);
619 // Set all cards back to clean.
620 _g1->cleanUpCardTable();
622 set_traversal(false);
624 DirtyCardQueueSet& into_cset_dcqs = _g1->into_cset_dirty_card_queue_set();
625 int into_cset_n_buffers = into_cset_dcqs.completed_buffers_num();
627 if (_g1->evacuation_failed()) {
628 // Restore remembered sets for the regions pointing into the collection set.
630 if (G1DeferredRSUpdate) {
631 // If deferred RS updates are enabled then we just need to transfer
632 // the completed buffers from (a) the DirtyCardQueueSet used to hold
633 // cards that contain references that point into the collection set
634 // to (b) the DCQS used to hold the deferred RS updates
635 _g1->dirty_card_queue_set().merge_bufferlists(&into_cset_dcqs);
636 } else {
638 CardTableModRefBS* bs = (CardTableModRefBS*)_g1->barrier_set();
639 UpdateRSetCardTableEntryIntoCSetClosure update_rs_cset_immediate(_g1, bs);
641 int n_completed_buffers = 0;
642 while (into_cset_dcqs.apply_closure_to_completed_buffer(&update_rs_cset_immediate,
643 0, 0, true)) {
644 n_completed_buffers++;
645 }
646 assert(n_completed_buffers == into_cset_n_buffers, "missed some buffers");
647 }
648 }
650 // Free any completed buffers in the DirtyCardQueueSet used to hold cards
651 // which contain references that point into the collection.
652 _g1->into_cset_dirty_card_queue_set().clear();
653 assert(_g1->into_cset_dirty_card_queue_set().completed_buffers_num() == 0,
654 "all buffers should be freed");
655 _g1->into_cset_dirty_card_queue_set().clear_n_completed_buffers();
657 assert(!_traversal_in_progress, "Invariant between iterations.");
658 }
660 class UpdateRSObjectClosure: public ObjectClosure {
661 UpdateRSOopClosure* _update_rs_oop_cl;
662 public:
663 UpdateRSObjectClosure(UpdateRSOopClosure* update_rs_oop_cl) :
664 _update_rs_oop_cl(update_rs_oop_cl) {}
665 void do_object(oop obj) {
666 obj->oop_iterate(_update_rs_oop_cl);
667 }
669 };
671 class ScrubRSClosure: public HeapRegionClosure {
672 G1CollectedHeap* _g1h;
673 BitMap* _region_bm;
674 BitMap* _card_bm;
675 CardTableModRefBS* _ctbs;
676 public:
677 ScrubRSClosure(BitMap* region_bm, BitMap* card_bm) :
678 _g1h(G1CollectedHeap::heap()),
679 _region_bm(region_bm), _card_bm(card_bm),
680 _ctbs(NULL)
681 {
682 ModRefBarrierSet* bs = _g1h->mr_bs();
683 guarantee(bs->is_a(BarrierSet::CardTableModRef), "Precondition");
684 _ctbs = (CardTableModRefBS*)bs;
685 }
687 bool doHeapRegion(HeapRegion* r) {
688 if (!r->continuesHumongous()) {
689 r->rem_set()->scrub(_ctbs, _region_bm, _card_bm);
690 }
691 return false;
692 }
693 };
695 void HRInto_G1RemSet::scrub(BitMap* region_bm, BitMap* card_bm) {
696 ScrubRSClosure scrub_cl(region_bm, card_bm);
697 _g1->heap_region_iterate(&scrub_cl);
698 }
700 void HRInto_G1RemSet::scrub_par(BitMap* region_bm, BitMap* card_bm,
701 int worker_num, int claim_val) {
702 ScrubRSClosure scrub_cl(region_bm, card_bm);
703 _g1->heap_region_par_iterate_chunked(&scrub_cl, worker_num, claim_val);
704 }
707 static IntHistogram out_of_histo(50, 50);
709 class TriggerClosure : public OopClosure {
710 bool _trigger;
711 public:
712 TriggerClosure() : _trigger(false) { }
713 bool value() const { return _trigger; }
714 template <class T> void do_oop_nv(T* p) { _trigger = true; }
715 virtual void do_oop(oop* p) { do_oop_nv(p); }
716 virtual void do_oop(narrowOop* p) { do_oop_nv(p); }
717 };
719 class InvokeIfNotTriggeredClosure: public OopClosure {
720 TriggerClosure* _t;
721 OopClosure* _oc;
722 public:
723 InvokeIfNotTriggeredClosure(TriggerClosure* t, OopClosure* oc):
724 _t(t), _oc(oc) { }
725 template <class T> void do_oop_nv(T* p) {
726 if (!_t->value()) _oc->do_oop(p);
727 }
728 virtual void do_oop(oop* p) { do_oop_nv(p); }
729 virtual void do_oop(narrowOop* p) { do_oop_nv(p); }
730 };
732 class Mux2Closure : public OopClosure {
733 OopClosure* _c1;
734 OopClosure* _c2;
735 public:
736 Mux2Closure(OopClosure *c1, OopClosure *c2) : _c1(c1), _c2(c2) { }
737 template <class T> void do_oop_nv(T* p) {
738 _c1->do_oop(p); _c2->do_oop(p);
739 }
740 virtual void do_oop(oop* p) { do_oop_nv(p); }
741 virtual void do_oop(narrowOop* p) { do_oop_nv(p); }
742 };
744 bool HRInto_G1RemSet::concurrentRefineOneCard_impl(jbyte* card_ptr, int worker_i,
745 bool check_for_refs_into_cset) {
746 // Construct the region representing the card.
747 HeapWord* start = _ct_bs->addr_for(card_ptr);
748 // And find the region containing it.
749 HeapRegion* r = _g1->heap_region_containing(start);
750 assert(r != NULL, "unexpected null");
752 HeapWord* end = _ct_bs->addr_for(card_ptr + 1);
753 MemRegion dirtyRegion(start, end);
755 #if CARD_REPEAT_HISTO
756 init_ct_freq_table(_g1->g1_reserved_obj_bytes());
757 ct_freq_note_card(_ct_bs->index_for(start));
758 #endif
760 UpdateRSOopClosure update_rs_oop_cl(this, worker_i);
761 update_rs_oop_cl.set_from(r);
763 TriggerClosure trigger_cl;
764 FilterIntoCSClosure into_cs_cl(NULL, _g1, &trigger_cl);
765 InvokeIfNotTriggeredClosure invoke_cl(&trigger_cl, &into_cs_cl);
766 Mux2Closure mux(&invoke_cl, &update_rs_oop_cl);
768 FilterOutOfRegionClosure filter_then_update_rs_oop_cl(r,
769 (check_for_refs_into_cset ?
770 (OopClosure*)&mux :
771 (OopClosure*)&update_rs_oop_cl));
773 // Undirty the card.
774 *card_ptr = CardTableModRefBS::clean_card_val();
775 // We must complete this write before we do any of the reads below.
776 OrderAccess::storeload();
777 // And process it, being careful of unallocated portions of TLAB's.
779 // The region for the current card may be a young region. The
780 // current card may have been a card that was evicted from the
781 // card cache. When the card was inserted into the cache, we had
782 // determined that its region was non-young. While in the cache,
783 // the region may have been freed during a cleanup pause, reallocated
784 // and tagged as young.
785 //
786 // We wish to filter out cards for such a region but the current
787 // thread, if we're running conucrrently, may "see" the young type
788 // change at any time (so an earlier "is_young" check may pass or
789 // fail arbitrarily). We tell the iteration code to perform this
790 // filtering when it has been determined that there has been an actual
791 // allocation in this region and making it safe to check the young type.
792 bool filter_young = true;
794 HeapWord* stop_point =
795 r->oops_on_card_seq_iterate_careful(dirtyRegion,
796 &filter_then_update_rs_oop_cl,
797 filter_young);
799 // If stop_point is non-null, then we encountered an unallocated region
800 // (perhaps the unfilled portion of a TLAB.) For now, we'll dirty the
801 // card and re-enqueue: if we put off the card until a GC pause, then the
802 // unallocated portion will be filled in. Alternatively, we might try
803 // the full complexity of the technique used in "regular" precleaning.
804 if (stop_point != NULL) {
805 // The card might have gotten re-dirtied and re-enqueued while we
806 // worked. (In fact, it's pretty likely.)
807 if (*card_ptr != CardTableModRefBS::dirty_card_val()) {
808 *card_ptr = CardTableModRefBS::dirty_card_val();
809 MutexLockerEx x(Shared_DirtyCardQ_lock,
810 Mutex::_no_safepoint_check_flag);
811 DirtyCardQueue* sdcq =
812 JavaThread::dirty_card_queue_set().shared_dirty_card_queue();
813 sdcq->enqueue(card_ptr);
814 }
815 } else {
816 out_of_histo.add_entry(filter_then_update_rs_oop_cl.out_of_region());
817 _conc_refine_cards++;
818 }
820 return trigger_cl.value();
821 }
823 bool HRInto_G1RemSet::concurrentRefineOneCard(jbyte* card_ptr, int worker_i,
824 bool check_for_refs_into_cset) {
825 // If the card is no longer dirty, nothing to do.
826 if (*card_ptr != CardTableModRefBS::dirty_card_val()) {
827 // No need to return that this card contains refs that point
828 // into the collection set.
829 return false;
830 }
832 // Construct the region representing the card.
833 HeapWord* start = _ct_bs->addr_for(card_ptr);
834 // And find the region containing it.
835 HeapRegion* r = _g1->heap_region_containing(start);
836 if (r == NULL) {
837 guarantee(_g1->is_in_permanent(start), "Or else where?");
838 // Again no need to return that this card contains refs that
839 // point into the collection set.
840 return false; // Not in the G1 heap (might be in perm, for example.)
841 }
842 // Why do we have to check here whether a card is on a young region,
843 // given that we dirty young regions and, as a result, the
844 // post-barrier is supposed to filter them out and never to enqueue
845 // them? When we allocate a new region as the "allocation region" we
846 // actually dirty its cards after we release the lock, since card
847 // dirtying while holding the lock was a performance bottleneck. So,
848 // as a result, it is possible for other threads to actually
849 // allocate objects in the region (after the acquire the lock)
850 // before all the cards on the region are dirtied. This is unlikely,
851 // and it doesn't happen often, but it can happen. So, the extra
852 // check below filters out those cards.
853 if (r->is_young()) {
854 return false;
855 }
856 // While we are processing RSet buffers during the collection, we
857 // actually don't want to scan any cards on the collection set,
858 // since we don't want to update remebered sets with entries that
859 // point into the collection set, given that live objects from the
860 // collection set are about to move and such entries will be stale
861 // very soon. This change also deals with a reliability issue which
862 // involves scanning a card in the collection set and coming across
863 // an array that was being chunked and looking malformed. Note,
864 // however, that if evacuation fails, we have to scan any objects
865 // that were not moved and create any missing entries.
866 if (r->in_collection_set()) {
867 return false;
868 }
870 // Should we defer processing the card?
871 //
872 // Previously the result from the insert_cache call would be
873 // either card_ptr (implying that card_ptr was currently "cold"),
874 // null (meaning we had inserted the card ptr into the "hot"
875 // cache, which had some headroom), or a "hot" card ptr
876 // extracted from the "hot" cache.
877 //
878 // Now that the _card_counts cache in the ConcurrentG1Refine
879 // instance is an evicting hash table, the result we get back
880 // could be from evicting the card ptr in an already occupied
881 // bucket (in which case we have replaced the card ptr in the
882 // bucket with card_ptr and "defer" is set to false). To avoid
883 // having a data structure (updates to which would need a lock)
884 // to hold these unprocessed dirty cards, we need to immediately
885 // process card_ptr. The actions needed to be taken on return
886 // from cache_insert are summarized in the following table:
887 //
888 // res defer action
889 // --------------------------------------------------------------
890 // null false card evicted from _card_counts & replaced with
891 // card_ptr; evicted ptr added to hot cache.
892 // No need to process res; immediately process card_ptr
893 //
894 // null true card not evicted from _card_counts; card_ptr added
895 // to hot cache.
896 // Nothing to do.
897 //
898 // non-null false card evicted from _card_counts & replaced with
899 // card_ptr; evicted ptr is currently "cold" or
900 // caused an eviction from the hot cache.
901 // Immediately process res; process card_ptr.
902 //
903 // non-null true card not evicted from _card_counts; card_ptr is
904 // currently cold, or caused an eviction from hot
905 // cache.
906 // Immediately process res; no need to process card_ptr.
909 jbyte* res = card_ptr;
910 bool defer = false;
912 // This gets set to true if the card being refined has references
913 // that point into the collection set.
914 bool oops_into_cset = false;
916 if (_cg1r->use_cache()) {
917 jbyte* res = _cg1r->cache_insert(card_ptr, &defer);
918 if (res != NULL && (res != card_ptr || defer)) {
919 start = _ct_bs->addr_for(res);
920 r = _g1->heap_region_containing(start);
921 if (r == NULL) {
922 assert(_g1->is_in_permanent(start), "Or else where?");
923 } else {
924 // Checking whether the region we got back from the cache
925 // is young here is inappropriate. The region could have been
926 // freed, reallocated and tagged as young while in the cache.
927 // Hence we could see its young type change at any time.
928 //
929 // Process card pointer we get back from the hot card cache. This
930 // will check whether the region containing the card is young
931 // _after_ checking that the region has been allocated from.
932 oops_into_cset = concurrentRefineOneCard_impl(res, worker_i,
933 false /* check_for_refs_into_cset */);
934 // The above call to concurrentRefineOneCard_impl is only
935 // performed if the hot card cache is enabled. This cache is
936 // disabled during an evacuation pause - which is the only
937 // time when we need know if the card contains references
938 // that point into the collection set. Also when the hot card
939 // cache is enabled, this code is executed by the concurrent
940 // refine threads - rather than the GC worker threads - and
941 // concurrentRefineOneCard_impl will return false.
942 assert(!oops_into_cset, "should not see true here");
943 }
944 }
945 }
947 if (!defer) {
948 oops_into_cset =
949 concurrentRefineOneCard_impl(card_ptr, worker_i, check_for_refs_into_cset);
950 // We should only be detecting that the card contains references
951 // that point into the collection set if the current thread is
952 // a GC worker thread.
953 assert(!oops_into_cset || SafepointSynchronize::is_at_safepoint(),
954 "invalid result at non safepoint");
955 }
956 return oops_into_cset;
957 }
959 class HRRSStatsIter: public HeapRegionClosure {
960 size_t _occupied;
961 size_t _total_mem_sz;
962 size_t _max_mem_sz;
963 HeapRegion* _max_mem_sz_region;
964 public:
965 HRRSStatsIter() :
966 _occupied(0),
967 _total_mem_sz(0),
968 _max_mem_sz(0),
969 _max_mem_sz_region(NULL)
970 {}
972 bool doHeapRegion(HeapRegion* r) {
973 if (r->continuesHumongous()) return false;
974 size_t mem_sz = r->rem_set()->mem_size();
975 if (mem_sz > _max_mem_sz) {
976 _max_mem_sz = mem_sz;
977 _max_mem_sz_region = r;
978 }
979 _total_mem_sz += mem_sz;
980 size_t occ = r->rem_set()->occupied();
981 _occupied += occ;
982 return false;
983 }
984 size_t total_mem_sz() { return _total_mem_sz; }
985 size_t max_mem_sz() { return _max_mem_sz; }
986 size_t occupied() { return _occupied; }
987 HeapRegion* max_mem_sz_region() { return _max_mem_sz_region; }
988 };
990 class PrintRSThreadVTimeClosure : public ThreadClosure {
991 public:
992 virtual void do_thread(Thread *t) {
993 ConcurrentG1RefineThread* crt = (ConcurrentG1RefineThread*) t;
994 gclog_or_tty->print(" %5.2f", crt->vtime_accum());
995 }
996 };
998 void HRInto_G1RemSet::print_summary_info() {
999 G1CollectedHeap* g1 = G1CollectedHeap::heap();
1001 #if CARD_REPEAT_HISTO
1002 gclog_or_tty->print_cr("\nG1 card_repeat count histogram: ");
1003 gclog_or_tty->print_cr(" # of repeats --> # of cards with that number.");
1004 card_repeat_count.print_on(gclog_or_tty);
1005 #endif
1007 if (FILTEROUTOFREGIONCLOSURE_DOHISTOGRAMCOUNT) {
1008 gclog_or_tty->print_cr("\nG1 rem-set out-of-region histogram: ");
1009 gclog_or_tty->print_cr(" # of CS ptrs --> # of cards with that number.");
1010 out_of_histo.print_on(gclog_or_tty);
1011 }
1012 gclog_or_tty->print_cr("\n Concurrent RS processed %d cards",
1013 _conc_refine_cards);
1014 DirtyCardQueueSet& dcqs = JavaThread::dirty_card_queue_set();
1015 jint tot_processed_buffers =
1016 dcqs.processed_buffers_mut() + dcqs.processed_buffers_rs_thread();
1017 gclog_or_tty->print_cr(" Of %d completed buffers:", tot_processed_buffers);
1018 gclog_or_tty->print_cr(" %8d (%5.1f%%) by conc RS threads.",
1019 dcqs.processed_buffers_rs_thread(),
1020 100.0*(float)dcqs.processed_buffers_rs_thread()/
1021 (float)tot_processed_buffers);
1022 gclog_or_tty->print_cr(" %8d (%5.1f%%) by mutator threads.",
1023 dcqs.processed_buffers_mut(),
1024 100.0*(float)dcqs.processed_buffers_mut()/
1025 (float)tot_processed_buffers);
1026 gclog_or_tty->print_cr(" Conc RS threads times(s)");
1027 PrintRSThreadVTimeClosure p;
1028 gclog_or_tty->print(" ");
1029 g1->concurrent_g1_refine()->threads_do(&p);
1030 gclog_or_tty->print_cr("");
1032 if (G1UseHRIntoRS) {
1033 HRRSStatsIter blk;
1034 g1->heap_region_iterate(&blk);
1035 gclog_or_tty->print_cr(" Total heap region rem set sizes = " SIZE_FORMAT "K."
1036 " Max = " SIZE_FORMAT "K.",
1037 blk.total_mem_sz()/K, blk.max_mem_sz()/K);
1038 gclog_or_tty->print_cr(" Static structures = " SIZE_FORMAT "K,"
1039 " free_lists = " SIZE_FORMAT "K.",
1040 HeapRegionRemSet::static_mem_size()/K,
1041 HeapRegionRemSet::fl_mem_size()/K);
1042 gclog_or_tty->print_cr(" %d occupied cards represented.",
1043 blk.occupied());
1044 gclog_or_tty->print_cr(" Max sz region = [" PTR_FORMAT ", " PTR_FORMAT " )"
1045 ", cap = " SIZE_FORMAT "K, occ = " SIZE_FORMAT "K.",
1046 blk.max_mem_sz_region()->bottom(), blk.max_mem_sz_region()->end(),
1047 (blk.max_mem_sz_region()->rem_set()->mem_size() + K - 1)/K,
1048 (blk.max_mem_sz_region()->rem_set()->occupied() + K - 1)/K);
1049 gclog_or_tty->print_cr(" Did %d coarsenings.",
1050 HeapRegionRemSet::n_coarsenings());
1052 }
1053 }
1055 void HRInto_G1RemSet::prepare_for_verify() {
1056 if (G1HRRSFlushLogBuffersOnVerify &&
1057 (VerifyBeforeGC || VerifyAfterGC)
1058 && !_g1->full_collection()) {
1059 cleanupHRRS();
1060 _g1->set_refine_cte_cl_concurrency(false);
1061 if (SafepointSynchronize::is_at_safepoint()) {
1062 DirtyCardQueueSet& dcqs = JavaThread::dirty_card_queue_set();
1063 dcqs.concatenate_logs();
1064 }
1065 bool cg1r_use_cache = _cg1r->use_cache();
1066 _cg1r->set_use_cache(false);
1067 DirtyCardQueue into_cset_dcq(&_g1->into_cset_dirty_card_queue_set());
1068 updateRS(&into_cset_dcq, 0);
1069 _g1->into_cset_dirty_card_queue_set().clear();
1070 _cg1r->set_use_cache(cg1r_use_cache);
1072 assert(JavaThread::dirty_card_queue_set().completed_buffers_num() == 0, "All should be consumed");
1073 }
1074 }