Print this page
rev 2691 : [mq]: g1-reference-processing
Split |
Close |
Expand all |
Collapse all |
--- old/src/share/vm/gc_implementation/g1/g1RemSet.cpp
+++ new/src/share/vm/gc_implementation/g1/g1RemSet.cpp
1 1 /*
2 2 * Copyright (c) 2001, 2011, Oracle and/or its affiliates. All rights reserved.
3 3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
4 4 *
5 5 * This code is free software; you can redistribute it and/or modify it
6 6 * under the terms of the GNU General Public License version 2 only, as
7 7 * published by the Free Software Foundation.
8 8 *
9 9 * This code is distributed in the hope that it will be useful, but WITHOUT
10 10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
12 12 * version 2 for more details (a copy is included in the LICENSE file that
13 13 * accompanied this code).
14 14 *
15 15 * You should have received a copy of the GNU General Public License version
16 16 * 2 along with this work; if not, write to the Free Software Foundation,
17 17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
18 18 *
19 19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
20 20 * or visit www.oracle.com if you need additional information or have any
21 21 * questions.
22 22 *
23 23 */
24 24
25 25 #include "precompiled.hpp"
26 26 #include "gc_implementation/g1/bufferingOopClosure.hpp"
27 27 #include "gc_implementation/g1/concurrentG1Refine.hpp"
28 28 #include "gc_implementation/g1/concurrentG1RefineThread.hpp"
29 29 #include "gc_implementation/g1/g1BlockOffsetTable.inline.hpp"
30 30 #include "gc_implementation/g1/g1CollectedHeap.inline.hpp"
31 31 #include "gc_implementation/g1/g1CollectorPolicy.hpp"
32 32 #include "gc_implementation/g1/g1OopClosures.inline.hpp"
33 33 #include "gc_implementation/g1/g1RemSet.inline.hpp"
34 34 #include "gc_implementation/g1/heapRegionSeq.inline.hpp"
35 35 #include "memory/iterator.hpp"
36 36 #include "oops/oop.inline.hpp"
37 37 #include "utilities/intHisto.hpp"
38 38
39 39 #define CARD_REPEAT_HISTO 0
40 40
41 41 #if CARD_REPEAT_HISTO
42 42 static size_t ct_freq_sz;
43 43 static jbyte* ct_freq = NULL;
44 44
45 45 void init_ct_freq_table(size_t heap_sz_bytes) {
46 46 if (ct_freq == NULL) {
47 47 ct_freq_sz = heap_sz_bytes/CardTableModRefBS::card_size;
48 48 ct_freq = new jbyte[ct_freq_sz];
49 49 for (size_t j = 0; j < ct_freq_sz; j++) ct_freq[j] = 0;
50 50 }
51 51 }
52 52
53 53 void ct_freq_note_card(size_t index) {
54 54 assert(0 <= index && index < ct_freq_sz, "Bounds error.");
55 55 if (ct_freq[index] < 100) { ct_freq[index]++; }
56 56 }
57 57
58 58 static IntHistogram card_repeat_count(10, 10);
59 59
60 60 void ct_freq_update_histo_and_reset() {
61 61 for (size_t j = 0; j < ct_freq_sz; j++) {
62 62 card_repeat_count.add_entry(ct_freq[j]);
63 63 ct_freq[j] = 0;
64 64 }
65 65
66 66 }
67 67 #endif
68 68
69 69 G1RemSet::G1RemSet(G1CollectedHeap* g1, CardTableModRefBS* ct_bs)
70 70 : _g1(g1), _conc_refine_cards(0),
71 71 _ct_bs(ct_bs), _g1p(_g1->g1_policy()),
72 72 _cg1r(g1->concurrent_g1_refine()),
73 73 _cset_rs_update_cl(NULL),
74 74 _cards_scanned(NULL), _total_cards_scanned(0)
75 75 {
76 76 _seq_task = new SubTasksDone(NumSeqTasks);
77 77 guarantee(n_workers() > 0, "There should be some workers");
78 78 _cset_rs_update_cl = NEW_C_HEAP_ARRAY(OopsInHeapRegionClosure*, n_workers());
79 79 for (uint i = 0; i < n_workers(); i++) {
80 80 _cset_rs_update_cl[i] = NULL;
81 81 }
82 82 }
83 83
84 84 G1RemSet::~G1RemSet() {
85 85 delete _seq_task;
86 86 for (uint i = 0; i < n_workers(); i++) {
87 87 assert(_cset_rs_update_cl[i] == NULL, "it should be");
88 88 }
89 89 FREE_C_HEAP_ARRAY(OopsInHeapRegionClosure*, _cset_rs_update_cl);
90 90 }
91 91
92 92 void CountNonCleanMemRegionClosure::do_MemRegion(MemRegion mr) {
93 93 if (_g1->is_in_g1_reserved(mr.start())) {
94 94 _n += (int) ((mr.byte_size() / CardTableModRefBS::card_size));
95 95 if (_start_first == NULL) _start_first = mr.start();
96 96 }
97 97 }
98 98
99 99 class ScanRSClosure : public HeapRegionClosure {
100 100 size_t _cards_done, _cards;
101 101 G1CollectedHeap* _g1h;
102 102 OopsInHeapRegionClosure* _oc;
103 103 G1BlockOffsetSharedArray* _bot_shared;
104 104 CardTableModRefBS *_ct_bs;
105 105 int _worker_i;
106 106 int _block_size;
107 107 bool _try_claimed;
108 108 public:
109 109 ScanRSClosure(OopsInHeapRegionClosure* oc, int worker_i) :
110 110 _oc(oc),
111 111 _cards(0),
112 112 _cards_done(0),
113 113 _worker_i(worker_i),
114 114 _try_claimed(false)
115 115 {
116 116 _g1h = G1CollectedHeap::heap();
117 117 _bot_shared = _g1h->bot_shared();
118 118 _ct_bs = (CardTableModRefBS*) (_g1h->barrier_set());
119 119 _block_size = MAX2<int>(G1RSetScanBlockSize, 1);
120 120 }
121 121
122 122 void set_try_claimed() { _try_claimed = true; }
123 123
124 124 void scanCard(size_t index, HeapRegion *r) {
125 125 DirtyCardToOopClosure* cl =
126 126 r->new_dcto_closure(_oc,
127 127 CardTableModRefBS::Precise,
128 128 HeapRegionDCTOC::IntoCSFilterKind);
129 129
130 130 // Set the "from" region in the closure.
131 131 _oc->set_region(r);
132 132 HeapWord* card_start = _bot_shared->address_for_index(index);
133 133 HeapWord* card_end = card_start + G1BlockOffsetSharedArray::N_words;
134 134 Space *sp = SharedHeap::heap()->space_containing(card_start);
135 135 MemRegion sm_region = sp->used_region_at_save_marks();
136 136 MemRegion mr = sm_region.intersection(MemRegion(card_start,card_end));
137 137 if (!mr.is_empty() && !_ct_bs->is_card_claimed(index)) {
138 138 // We make the card as "claimed" lazily (so races are possible
139 139 // but they're benign), which reduces the number of duplicate
140 140 // scans (the rsets of the regions in the cset can intersect).
141 141 _ct_bs->set_card_claimed(index);
142 142 _cards_done++;
143 143 cl->do_MemRegion(mr);
144 144 }
145 145 }
146 146
147 147 void printCard(HeapRegion* card_region, size_t card_index,
148 148 HeapWord* card_start) {
149 149 gclog_or_tty->print_cr("T %d Region [" PTR_FORMAT ", " PTR_FORMAT ") "
150 150 "RS names card %p: "
151 151 "[" PTR_FORMAT ", " PTR_FORMAT ")",
152 152 _worker_i,
153 153 card_region->bottom(), card_region->end(),
154 154 card_index,
155 155 card_start, card_start + G1BlockOffsetSharedArray::N_words);
156 156 }
157 157
158 158 bool doHeapRegion(HeapRegion* r) {
159 159 assert(r->in_collection_set(), "should only be called on elements of CS.");
160 160 HeapRegionRemSet* hrrs = r->rem_set();
161 161 if (hrrs->iter_is_complete()) return false; // All done.
162 162 if (!_try_claimed && !hrrs->claim_iter()) return false;
163 163 // If we ever free the collection set concurrently, we should also
164 164 // clear the card table concurrently therefore we won't need to
165 165 // add regions of the collection set to the dirty cards region.
166 166 _g1h->push_dirty_cards_region(r);
167 167 // If we didn't return above, then
168 168 // _try_claimed || r->claim_iter()
169 169 // is true: either we're supposed to work on claimed-but-not-complete
170 170 // regions, or we successfully claimed the region.
171 171 HeapRegionRemSetIterator* iter = _g1h->rem_set_iterator(_worker_i);
172 172 hrrs->init_iterator(iter);
173 173 size_t card_index;
174 174
175 175 // We claim cards in block so as to recude the contention. The block size is determined by
176 176 // the G1RSetScanBlockSize parameter.
177 177 size_t jump_to_card = hrrs->iter_claimed_next(_block_size);
178 178 for (size_t current_card = 0; iter->has_next(card_index); current_card++) {
179 179 if (current_card >= jump_to_card + _block_size) {
180 180 jump_to_card = hrrs->iter_claimed_next(_block_size);
181 181 }
182 182 if (current_card < jump_to_card) continue;
183 183 HeapWord* card_start = _g1h->bot_shared()->address_for_index(card_index);
184 184 #if 0
185 185 gclog_or_tty->print("Rem set iteration yielded card [" PTR_FORMAT ", " PTR_FORMAT ").\n",
186 186 card_start, card_start + CardTableModRefBS::card_size_in_words);
187 187 #endif
188 188
189 189 HeapRegion* card_region = _g1h->heap_region_containing(card_start);
190 190 assert(card_region != NULL, "Yielding cards not in the heap?");
191 191 _cards++;
192 192
193 193 if (!card_region->is_on_dirty_cards_region_list()) {
194 194 _g1h->push_dirty_cards_region(card_region);
195 195 }
196 196
197 197 // If the card is dirty, then we will scan it during updateRS.
198 198 if (!card_region->in_collection_set() &&
199 199 !_ct_bs->is_card_dirty(card_index)) {
200 200 scanCard(card_index, card_region);
201 201 }
202 202 }
203 203 if (!_try_claimed) {
204 204 hrrs->set_iter_complete();
205 205 }
206 206 return false;
207 207 }
208 208 size_t cards_done() { return _cards_done;}
209 209 size_t cards_looked_up() { return _cards;}
210 210 };
211 211
212 212 // We want the parallel threads to start their scanning at
213 213 // different collection set regions to avoid contention.
214 214 // If we have:
215 215 // n collection set regions
216 216 // p threads
217 217 // Then thread t will start at region t * floor (n/p)
218 218
219 219 HeapRegion* G1RemSet::calculateStartRegion(int worker_i) {
220 220 HeapRegion* result = _g1p->collection_set();
221 221 if (ParallelGCThreads > 0) {
222 222 size_t cs_size = _g1p->collection_set_size();
223 223 int n_workers = _g1->workers()->total_workers();
224 224 size_t cs_spans = cs_size / n_workers;
225 225 size_t ind = cs_spans * worker_i;
226 226 for (size_t i = 0; i < ind; i++)
↓ open down ↓ |
226 lines elided |
↑ open up ↑ |
227 227 result = result->next_in_collection_set();
228 228 }
229 229 return result;
230 230 }
231 231
232 232 void G1RemSet::scanRS(OopsInHeapRegionClosure* oc, int worker_i) {
233 233 double rs_time_start = os::elapsedTime();
234 234 HeapRegion *startRegion = calculateStartRegion(worker_i);
235 235
236 236 ScanRSClosure scanRScl(oc, worker_i);
237 +
237 238 _g1->collection_set_iterate_from(startRegion, &scanRScl);
238 239 scanRScl.set_try_claimed();
239 240 _g1->collection_set_iterate_from(startRegion, &scanRScl);
240 241
241 242 double scan_rs_time_sec = os::elapsedTime() - rs_time_start;
242 243
243 244 assert( _cards_scanned != NULL, "invariant" );
244 245 _cards_scanned[worker_i] = scanRScl.cards_done();
245 246
246 247 _g1p->record_scan_rs_time(worker_i, scan_rs_time_sec * 1000.0);
247 248 }
248 249
249 250 // Closure used for updating RSets and recording references that
250 251 // point into the collection set. Only called during an
251 252 // evacuation pause.
252 253
253 254 class RefineRecordRefsIntoCSCardTableEntryClosure: public CardTableEntryClosure {
254 255 G1RemSet* _g1rs;
255 256 DirtyCardQueue* _into_cset_dcq;
256 257 public:
257 258 RefineRecordRefsIntoCSCardTableEntryClosure(G1CollectedHeap* g1h,
258 259 DirtyCardQueue* into_cset_dcq) :
259 260 _g1rs(g1h->g1_rem_set()), _into_cset_dcq(into_cset_dcq)
260 261 {}
261 262 bool do_card_ptr(jbyte* card_ptr, int worker_i) {
262 263 // The only time we care about recording cards that
263 264 // contain references that point into the collection set
264 265 // is during RSet updating within an evacuation pause.
265 266 // In this case worker_i should be the id of a GC worker thread.
266 267 assert(SafepointSynchronize::is_at_safepoint(), "not during an evacuation pause");
267 268 assert(worker_i < (int) (ParallelGCThreads == 0 ? 1 : ParallelGCThreads), "should be a GC worker");
268 269
269 270 if (_g1rs->concurrentRefineOneCard(card_ptr, worker_i, true)) {
270 271 // 'card_ptr' contains references that point into the collection
271 272 // set. We need to record the card in the DCQS
272 273 // (G1CollectedHeap::into_cset_dirty_card_queue_set())
273 274 // that's used for that purpose.
274 275 //
275 276 // Enqueue the card
↓ open down ↓ |
29 lines elided |
↑ open up ↑ |
276 277 _into_cset_dcq->enqueue(card_ptr);
277 278 }
278 279 return true;
279 280 }
280 281 };
281 282
282 283 void G1RemSet::updateRS(DirtyCardQueue* into_cset_dcq, int worker_i) {
283 284 double start = os::elapsedTime();
284 285 // Apply the given closure to all remaining log entries.
285 286 RefineRecordRefsIntoCSCardTableEntryClosure into_cset_update_rs_cl(_g1, into_cset_dcq);
287 +
286 288 _g1->iterate_dirty_card_closure(&into_cset_update_rs_cl, into_cset_dcq, false, worker_i);
287 289
288 290 // Now there should be no dirty cards.
289 291 if (G1RSLogCheckCardTable) {
290 292 CountNonCleanMemRegionClosure cl(_g1);
291 293 _ct_bs->mod_card_iterate(&cl);
292 294 // XXX This isn't true any more: keeping cards of young regions
293 295 // marked dirty broke it. Need some reasonable fix.
294 296 guarantee(cl.n() == 0, "Card table should be clean.");
295 297 }
296 298
297 299 _g1p->record_update_rs_time(worker_i, (os::elapsedTime() - start) * 1000.0);
298 300 }
299 301
300 302 class CountRSSizeClosure: public HeapRegionClosure {
301 303 size_t _n;
302 304 size_t _tot;
303 305 size_t _max;
304 306 HeapRegion* _max_r;
305 307 enum {
306 308 N = 20,
307 309 MIN = 6
308 310 };
309 311 int _histo[N];
310 312 public:
311 313 CountRSSizeClosure() : _n(0), _tot(0), _max(0), _max_r(NULL) {
312 314 for (int i = 0; i < N; i++) _histo[i] = 0;
313 315 }
314 316 bool doHeapRegion(HeapRegion* r) {
315 317 if (!r->continuesHumongous()) {
316 318 size_t occ = r->rem_set()->occupied();
317 319 _n++;
318 320 _tot += occ;
319 321 if (occ > _max) {
320 322 _max = occ;
321 323 _max_r = r;
322 324 }
323 325 // Fit it into a histo bin.
324 326 int s = 1 << MIN;
325 327 int i = 0;
326 328 while (occ > (size_t) s && i < (N-1)) {
327 329 s = s << 1;
328 330 i++;
329 331 }
330 332 _histo[i]++;
331 333 }
332 334 return false;
333 335 }
334 336 size_t n() { return _n; }
335 337 size_t tot() { return _tot; }
336 338 size_t mx() { return _max; }
337 339 HeapRegion* mxr() { return _max_r; }
338 340 void print_histo() {
339 341 int mx = N;
340 342 while (mx >= 0) {
341 343 if (_histo[mx-1] > 0) break;
342 344 mx--;
343 345 }
344 346 gclog_or_tty->print_cr("Number of regions with given RS sizes:");
345 347 gclog_or_tty->print_cr(" <= %8d %8d", 1 << MIN, _histo[0]);
346 348 for (int i = 1; i < mx-1; i++) {
347 349 gclog_or_tty->print_cr(" %8d - %8d %8d",
348 350 (1 << (MIN + i - 1)) + 1,
349 351 1 << (MIN + i),
350 352 _histo[i]);
351 353 }
352 354 gclog_or_tty->print_cr(" > %8d %8d", (1 << (MIN+mx-2))+1, _histo[mx-1]);
353 355 }
354 356 };
355 357
356 358 void G1RemSet::cleanupHRRS() {
357 359 HeapRegionRemSet::cleanup();
358 360 }
359 361
360 362 void G1RemSet::oops_into_collection_set_do(OopsInHeapRegionClosure* oc,
361 363 int worker_i) {
362 364 #if CARD_REPEAT_HISTO
363 365 ct_freq_update_histo_and_reset();
364 366 #endif
365 367 if (worker_i == 0) {
366 368 _cg1r->clear_and_record_card_counts();
367 369 }
368 370
369 371 // Make this into a command-line flag...
370 372 if (G1RSCountHisto && (ParallelGCThreads == 0 || worker_i == 0)) {
371 373 CountRSSizeClosure count_cl;
372 374 _g1->heap_region_iterate(&count_cl);
373 375 gclog_or_tty->print_cr("Avg of %d RS counts is %f, max is %d, "
374 376 "max region is " PTR_FORMAT,
375 377 count_cl.n(), (float)count_cl.tot()/(float)count_cl.n(),
376 378 count_cl.mx(), count_cl.mxr());
377 379 count_cl.print_histo();
378 380 }
379 381
380 382 // We cache the value of 'oc' closure into the appropriate slot in the
381 383 // _cset_rs_update_cl for this worker
382 384 assert(worker_i < (int)n_workers(), "sanity");
383 385 _cset_rs_update_cl[worker_i] = oc;
384 386
385 387 // A DirtyCardQueue that is used to hold cards containing references
386 388 // that point into the collection set. This DCQ is associated with a
387 389 // special DirtyCardQueueSet (see g1CollectedHeap.hpp). Under normal
388 390 // circumstances (i.e. the pause successfully completes), these cards
389 391 // are just discarded (there's no need to update the RSets of regions
390 392 // that were in the collection set - after the pause these regions
391 393 // are wholly 'free' of live objects. In the event of an evacuation
392 394 // failure the cards/buffers in this queue set are:
393 395 // * passed to the DirtyCardQueueSet that is used to manage deferred
394 396 // RSet updates, or
395 397 // * scanned for references that point into the collection set
396 398 // and the RSet of the corresponding region in the collection set
397 399 // is updated immediately.
398 400 DirtyCardQueue into_cset_dcq(&_g1->into_cset_dirty_card_queue_set());
399 401
400 402 assert((ParallelGCThreads > 0) || worker_i == 0, "invariant");
401 403
402 404 // The two flags below were introduced temporarily to serialize
403 405 // the updating and scanning of remembered sets. There are some
404 406 // race conditions when these two operations are done in parallel
405 407 // and they are causing failures. When we resolve said race
406 408 // conditions, we'll revert back to parallel remembered set
407 409 // updating and scanning. See CRs 6677707 and 6677708.
408 410 if (G1UseParallelRSetUpdating || (worker_i == 0)) {
409 411 updateRS(&into_cset_dcq, worker_i);
410 412 } else {
411 413 _g1p->record_update_rs_processed_buffers(worker_i, 0.0);
412 414 _g1p->record_update_rs_time(worker_i, 0.0);
413 415 }
414 416 if (G1UseParallelRSetScanning || (worker_i == 0)) {
415 417 scanRS(oc, worker_i);
416 418 } else {
417 419 _g1p->record_scan_rs_time(worker_i, 0.0);
418 420 }
419 421
420 422 // We now clear the cached values of _cset_rs_update_cl for this worker
421 423 _cset_rs_update_cl[worker_i] = NULL;
422 424 }
423 425
424 426 void G1RemSet::prepare_for_oops_into_collection_set_do() {
425 427 cleanupHRRS();
426 428 ConcurrentG1Refine* cg1r = _g1->concurrent_g1_refine();
427 429 _g1->set_refine_cte_cl_concurrency(false);
428 430 DirtyCardQueueSet& dcqs = JavaThread::dirty_card_queue_set();
429 431 dcqs.concatenate_logs();
430 432
431 433 if (ParallelGCThreads > 0) {
432 434 _seq_task->set_n_threads((int)n_workers());
433 435 }
434 436 guarantee( _cards_scanned == NULL, "invariant" );
435 437 _cards_scanned = NEW_C_HEAP_ARRAY(size_t, n_workers());
436 438 for (uint i = 0; i < n_workers(); ++i) {
437 439 _cards_scanned[i] = 0;
438 440 }
439 441 _total_cards_scanned = 0;
440 442 }
441 443
442 444
443 445 // This closure, applied to a DirtyCardQueueSet, is used to immediately
444 446 // update the RSets for the regions in the CSet. For each card it iterates
445 447 // through the oops which coincide with that card. It scans the reference
446 448 // fields in each oop; when it finds an oop that points into the collection
447 449 // set, the RSet for the region containing the referenced object is updated.
448 450 class UpdateRSetCardTableEntryIntoCSetClosure: public CardTableEntryClosure {
449 451 G1CollectedHeap* _g1;
450 452 CardTableModRefBS* _ct_bs;
451 453 public:
452 454 UpdateRSetCardTableEntryIntoCSetClosure(G1CollectedHeap* g1,
453 455 CardTableModRefBS* bs):
454 456 _g1(g1), _ct_bs(bs)
455 457 { }
456 458
457 459 bool do_card_ptr(jbyte* card_ptr, int worker_i) {
458 460 // Construct the region representing the card.
↓ open down ↓ |
163 lines elided |
↑ open up ↑ |
459 461 HeapWord* start = _ct_bs->addr_for(card_ptr);
460 462 // And find the region containing it.
461 463 HeapRegion* r = _g1->heap_region_containing(start);
462 464 assert(r != NULL, "unexpected null");
463 465
464 466 // Scan oops in the card looking for references into the collection set
465 467 HeapWord* end = _ct_bs->addr_for(card_ptr + 1);
466 468 MemRegion scanRegion(start, end);
467 469
468 470 UpdateRSetImmediate update_rs_cl(_g1->g1_rem_set());
469 - FilterIntoCSClosure update_rs_cset_oop_cl(NULL, _g1, &update_rs_cl);
471 + FilterIntoCSClosure update_rs_cset_oop_cl(NULL, _g1, &update_rs_cl, NULL /* rp */);
470 472 FilterOutOfRegionClosure filter_then_update_rs_cset_oop_cl(r, &update_rs_cset_oop_cl);
471 473
472 474 // We can pass false as the "filter_young" parameter here as:
473 475 // * we should be in a STW pause,
474 476 // * the DCQS to which this closure is applied is used to hold
475 477 // references that point into the collection set from the prior
476 478 // RSet updating,
477 479 // * the post-write barrier shouldn't be logging updates to young
478 480 // regions (but there is a situation where this can happen - see
479 481 // the comment in G1RemSet::concurrentRefineOneCard below -
480 482 // that should not be applicable here), and
481 483 // * during actual RSet updating, the filtering of cards in young
482 484 // regions in HeapRegion::oops_on_card_seq_iterate_careful is
483 485 // employed.
484 486 // As a result, when this closure is applied to "refs into cset"
485 487 // DCQS, we shouldn't see any cards in young regions.
486 488 update_rs_cl.set_region(r);
487 489 HeapWord* stop_point =
488 490 r->oops_on_card_seq_iterate_careful(scanRegion,
489 491 &filter_then_update_rs_cset_oop_cl,
490 492 false /* filter_young */,
491 493 NULL /* card_ptr */);
492 494
493 495 // Since this is performed in the event of an evacuation failure, we
494 496 // we shouldn't see a non-null stop point
495 497 assert(stop_point == NULL, "saw an unallocated region");
496 498 return true;
497 499 }
498 500 };
499 501
500 502 void G1RemSet::cleanup_after_oops_into_collection_set_do() {
501 503 guarantee( _cards_scanned != NULL, "invariant" );
502 504 _total_cards_scanned = 0;
503 505 for (uint i = 0; i < n_workers(); ++i) {
504 506 _total_cards_scanned += _cards_scanned[i];
505 507 }
506 508 FREE_C_HEAP_ARRAY(size_t, _cards_scanned);
507 509 _cards_scanned = NULL;
508 510 // Cleanup after copy
509 511 _g1->set_refine_cte_cl_concurrency(true);
510 512 // Set all cards back to clean.
511 513 _g1->cleanUpCardTable();
512 514
513 515 DirtyCardQueueSet& into_cset_dcqs = _g1->into_cset_dirty_card_queue_set();
514 516 int into_cset_n_buffers = into_cset_dcqs.completed_buffers_num();
515 517
516 518 if (_g1->evacuation_failed()) {
517 519 // Restore remembered sets for the regions pointing into the collection set.
518 520
519 521 if (G1DeferredRSUpdate) {
520 522 // If deferred RS updates are enabled then we just need to transfer
521 523 // the completed buffers from (a) the DirtyCardQueueSet used to hold
522 524 // cards that contain references that point into the collection set
523 525 // to (b) the DCQS used to hold the deferred RS updates
524 526 _g1->dirty_card_queue_set().merge_bufferlists(&into_cset_dcqs);
525 527 } else {
526 528
527 529 CardTableModRefBS* bs = (CardTableModRefBS*)_g1->barrier_set();
528 530 UpdateRSetCardTableEntryIntoCSetClosure update_rs_cset_immediate(_g1, bs);
529 531
530 532 int n_completed_buffers = 0;
531 533 while (into_cset_dcqs.apply_closure_to_completed_buffer(&update_rs_cset_immediate,
532 534 0, 0, true)) {
533 535 n_completed_buffers++;
534 536 }
535 537 assert(n_completed_buffers == into_cset_n_buffers, "missed some buffers");
536 538 }
537 539 }
538 540
539 541 // Free any completed buffers in the DirtyCardQueueSet used to hold cards
540 542 // which contain references that point into the collection.
541 543 _g1->into_cset_dirty_card_queue_set().clear();
542 544 assert(_g1->into_cset_dirty_card_queue_set().completed_buffers_num() == 0,
543 545 "all buffers should be freed");
544 546 _g1->into_cset_dirty_card_queue_set().clear_n_completed_buffers();
545 547 }
546 548
547 549 class ScrubRSClosure: public HeapRegionClosure {
548 550 G1CollectedHeap* _g1h;
549 551 BitMap* _region_bm;
550 552 BitMap* _card_bm;
551 553 CardTableModRefBS* _ctbs;
552 554 public:
553 555 ScrubRSClosure(BitMap* region_bm, BitMap* card_bm) :
554 556 _g1h(G1CollectedHeap::heap()),
555 557 _region_bm(region_bm), _card_bm(card_bm),
556 558 _ctbs(NULL)
557 559 {
558 560 ModRefBarrierSet* bs = _g1h->mr_bs();
559 561 guarantee(bs->is_a(BarrierSet::CardTableModRef), "Precondition");
560 562 _ctbs = (CardTableModRefBS*)bs;
561 563 }
562 564
563 565 bool doHeapRegion(HeapRegion* r) {
564 566 if (!r->continuesHumongous()) {
565 567 r->rem_set()->scrub(_ctbs, _region_bm, _card_bm);
566 568 }
567 569 return false;
568 570 }
569 571 };
570 572
571 573 void G1RemSet::scrub(BitMap* region_bm, BitMap* card_bm) {
572 574 ScrubRSClosure scrub_cl(region_bm, card_bm);
573 575 _g1->heap_region_iterate(&scrub_cl);
574 576 }
575 577
576 578 void G1RemSet::scrub_par(BitMap* region_bm, BitMap* card_bm,
577 579 int worker_num, int claim_val) {
578 580 ScrubRSClosure scrub_cl(region_bm, card_bm);
579 581 _g1->heap_region_par_iterate_chunked(&scrub_cl, worker_num, claim_val);
580 582 }
581 583
582 584
583 585 static IntHistogram out_of_histo(50, 50);
584 586
585 587 class TriggerClosure : public OopClosure {
586 588 bool _trigger;
587 589 public:
588 590 TriggerClosure() : _trigger(false) { }
589 591 bool value() const { return _trigger; }
590 592 template <class T> void do_oop_nv(T* p) { _trigger = true; }
591 593 virtual void do_oop(oop* p) { do_oop_nv(p); }
592 594 virtual void do_oop(narrowOop* p) { do_oop_nv(p); }
593 595 };
594 596
595 597 class InvokeIfNotTriggeredClosure: public OopClosure {
596 598 TriggerClosure* _t;
597 599 OopClosure* _oc;
598 600 public:
599 601 InvokeIfNotTriggeredClosure(TriggerClosure* t, OopClosure* oc):
600 602 _t(t), _oc(oc) { }
601 603 template <class T> void do_oop_nv(T* p) {
602 604 if (!_t->value()) _oc->do_oop(p);
603 605 }
604 606 virtual void do_oop(oop* p) { do_oop_nv(p); }
605 607 virtual void do_oop(narrowOop* p) { do_oop_nv(p); }
606 608 };
607 609
608 610 class Mux2Closure : public OopClosure {
609 611 OopClosure* _c1;
610 612 OopClosure* _c2;
611 613 public:
612 614 Mux2Closure(OopClosure *c1, OopClosure *c2) : _c1(c1), _c2(c2) { }
613 615 template <class T> void do_oop_nv(T* p) {
614 616 _c1->do_oop(p); _c2->do_oop(p);
615 617 }
616 618 virtual void do_oop(oop* p) { do_oop_nv(p); }
617 619 virtual void do_oop(narrowOop* p) { do_oop_nv(p); }
618 620 };
619 621
620 622 bool G1RemSet::concurrentRefineOneCard_impl(jbyte* card_ptr, int worker_i,
621 623 bool check_for_refs_into_cset) {
622 624 // Construct the region representing the card.
623 625 HeapWord* start = _ct_bs->addr_for(card_ptr);
624 626 // And find the region containing it.
625 627 HeapRegion* r = _g1->heap_region_containing(start);
626 628 assert(r != NULL, "unexpected null");
627 629
628 630 HeapWord* end = _ct_bs->addr_for(card_ptr + 1);
629 631 MemRegion dirtyRegion(start, end);
630 632
631 633 #if CARD_REPEAT_HISTO
632 634 init_ct_freq_table(_g1->max_capacity());
633 635 ct_freq_note_card(_ct_bs->index_for(start));
634 636 #endif
↓ open down ↓ |
155 lines elided |
↑ open up ↑ |
635 637
636 638 assert(!check_for_refs_into_cset || _cset_rs_update_cl[worker_i] != NULL, "sanity");
637 639 UpdateRSOrPushRefOopClosure update_rs_oop_cl(_g1,
638 640 _g1->g1_rem_set(),
639 641 _cset_rs_update_cl[worker_i],
640 642 check_for_refs_into_cset,
641 643 worker_i);
642 644 update_rs_oop_cl.set_from(r);
643 645
644 646 TriggerClosure trigger_cl;
645 - FilterIntoCSClosure into_cs_cl(NULL, _g1, &trigger_cl);
647 + FilterIntoCSClosure into_cs_cl(NULL, _g1, &trigger_cl, NULL /* rp */);
646 648 InvokeIfNotTriggeredClosure invoke_cl(&trigger_cl, &into_cs_cl);
647 649 Mux2Closure mux(&invoke_cl, &update_rs_oop_cl);
648 650
649 651 FilterOutOfRegionClosure filter_then_update_rs_oop_cl(r,
650 652 (check_for_refs_into_cset ?
651 653 (OopClosure*)&mux :
652 654 (OopClosure*)&update_rs_oop_cl));
653 655
654 656 // The region for the current card may be a young region. The
655 657 // current card may have been a card that was evicted from the
656 658 // card cache. When the card was inserted into the cache, we had
657 659 // determined that its region was non-young. While in the cache,
658 660 // the region may have been freed during a cleanup pause, reallocated
659 661 // and tagged as young.
660 662 //
661 663 // We wish to filter out cards for such a region but the current
662 664 // thread, if we're running concurrently, may "see" the young type
663 665 // change at any time (so an earlier "is_young" check may pass or
664 666 // fail arbitrarily). We tell the iteration code to perform this
665 667 // filtering when it has been determined that there has been an actual
666 668 // allocation in this region and making it safe to check the young type.
667 669 bool filter_young = true;
668 670
669 671 HeapWord* stop_point =
670 672 r->oops_on_card_seq_iterate_careful(dirtyRegion,
671 673 &filter_then_update_rs_oop_cl,
672 674 filter_young,
673 675 card_ptr);
674 676
675 677 // If stop_point is non-null, then we encountered an unallocated region
676 678 // (perhaps the unfilled portion of a TLAB.) For now, we'll dirty the
677 679 // card and re-enqueue: if we put off the card until a GC pause, then the
678 680 // unallocated portion will be filled in. Alternatively, we might try
679 681 // the full complexity of the technique used in "regular" precleaning.
680 682 if (stop_point != NULL) {
681 683 // The card might have gotten re-dirtied and re-enqueued while we
682 684 // worked. (In fact, it's pretty likely.)
683 685 if (*card_ptr != CardTableModRefBS::dirty_card_val()) {
684 686 *card_ptr = CardTableModRefBS::dirty_card_val();
685 687 MutexLockerEx x(Shared_DirtyCardQ_lock,
686 688 Mutex::_no_safepoint_check_flag);
687 689 DirtyCardQueue* sdcq =
688 690 JavaThread::dirty_card_queue_set().shared_dirty_card_queue();
689 691 sdcq->enqueue(card_ptr);
690 692 }
691 693 } else {
692 694 out_of_histo.add_entry(filter_then_update_rs_oop_cl.out_of_region());
693 695 _conc_refine_cards++;
694 696 }
695 697
696 698 return trigger_cl.value();
697 699 }
698 700
699 701 bool G1RemSet::concurrentRefineOneCard(jbyte* card_ptr, int worker_i,
700 702 bool check_for_refs_into_cset) {
701 703 // If the card is no longer dirty, nothing to do.
702 704 if (*card_ptr != CardTableModRefBS::dirty_card_val()) {
703 705 // No need to return that this card contains refs that point
704 706 // into the collection set.
705 707 return false;
706 708 }
707 709
708 710 // Construct the region representing the card.
709 711 HeapWord* start = _ct_bs->addr_for(card_ptr);
710 712 // And find the region containing it.
711 713 HeapRegion* r = _g1->heap_region_containing(start);
712 714 if (r == NULL) {
713 715 guarantee(_g1->is_in_permanent(start), "Or else where?");
714 716 // Again no need to return that this card contains refs that
715 717 // point into the collection set.
716 718 return false; // Not in the G1 heap (might be in perm, for example.)
717 719 }
718 720 // Why do we have to check here whether a card is on a young region,
719 721 // given that we dirty young regions and, as a result, the
720 722 // post-barrier is supposed to filter them out and never to enqueue
721 723 // them? When we allocate a new region as the "allocation region" we
722 724 // actually dirty its cards after we release the lock, since card
723 725 // dirtying while holding the lock was a performance bottleneck. So,
724 726 // as a result, it is possible for other threads to actually
725 727 // allocate objects in the region (after the acquire the lock)
726 728 // before all the cards on the region are dirtied. This is unlikely,
727 729 // and it doesn't happen often, but it can happen. So, the extra
728 730 // check below filters out those cards.
729 731 if (r->is_young()) {
730 732 return false;
731 733 }
732 734 // While we are processing RSet buffers during the collection, we
733 735 // actually don't want to scan any cards on the collection set,
734 736 // since we don't want to update remebered sets with entries that
735 737 // point into the collection set, given that live objects from the
736 738 // collection set are about to move and such entries will be stale
737 739 // very soon. This change also deals with a reliability issue which
738 740 // involves scanning a card in the collection set and coming across
739 741 // an array that was being chunked and looking malformed. Note,
740 742 // however, that if evacuation fails, we have to scan any objects
741 743 // that were not moved and create any missing entries.
742 744 if (r->in_collection_set()) {
743 745 return false;
744 746 }
745 747
746 748 // Should we defer processing the card?
747 749 //
748 750 // Previously the result from the insert_cache call would be
749 751 // either card_ptr (implying that card_ptr was currently "cold"),
750 752 // null (meaning we had inserted the card ptr into the "hot"
751 753 // cache, which had some headroom), or a "hot" card ptr
752 754 // extracted from the "hot" cache.
753 755 //
754 756 // Now that the _card_counts cache in the ConcurrentG1Refine
755 757 // instance is an evicting hash table, the result we get back
756 758 // could be from evicting the card ptr in an already occupied
757 759 // bucket (in which case we have replaced the card ptr in the
758 760 // bucket with card_ptr and "defer" is set to false). To avoid
759 761 // having a data structure (updates to which would need a lock)
760 762 // to hold these unprocessed dirty cards, we need to immediately
761 763 // process card_ptr. The actions needed to be taken on return
762 764 // from cache_insert are summarized in the following table:
763 765 //
764 766 // res defer action
765 767 // --------------------------------------------------------------
766 768 // null false card evicted from _card_counts & replaced with
767 769 // card_ptr; evicted ptr added to hot cache.
768 770 // No need to process res; immediately process card_ptr
769 771 //
770 772 // null true card not evicted from _card_counts; card_ptr added
771 773 // to hot cache.
772 774 // Nothing to do.
773 775 //
774 776 // non-null false card evicted from _card_counts & replaced with
775 777 // card_ptr; evicted ptr is currently "cold" or
776 778 // caused an eviction from the hot cache.
777 779 // Immediately process res; process card_ptr.
778 780 //
779 781 // non-null true card not evicted from _card_counts; card_ptr is
780 782 // currently cold, or caused an eviction from hot
781 783 // cache.
782 784 // Immediately process res; no need to process card_ptr.
783 785
784 786
785 787 jbyte* res = card_ptr;
786 788 bool defer = false;
787 789
788 790 // This gets set to true if the card being refined has references
789 791 // that point into the collection set.
790 792 bool oops_into_cset = false;
791 793
792 794 if (_cg1r->use_cache()) {
793 795 jbyte* res = _cg1r->cache_insert(card_ptr, &defer);
794 796 if (res != NULL && (res != card_ptr || defer)) {
795 797 start = _ct_bs->addr_for(res);
796 798 r = _g1->heap_region_containing(start);
797 799 if (r == NULL) {
798 800 assert(_g1->is_in_permanent(start), "Or else where?");
799 801 } else {
800 802 // Checking whether the region we got back from the cache
801 803 // is young here is inappropriate. The region could have been
802 804 // freed, reallocated and tagged as young while in the cache.
803 805 // Hence we could see its young type change at any time.
804 806 //
805 807 // Process card pointer we get back from the hot card cache. This
806 808 // will check whether the region containing the card is young
807 809 // _after_ checking that the region has been allocated from.
808 810 oops_into_cset = concurrentRefineOneCard_impl(res, worker_i,
809 811 false /* check_for_refs_into_cset */);
810 812 // The above call to concurrentRefineOneCard_impl is only
811 813 // performed if the hot card cache is enabled. This cache is
812 814 // disabled during an evacuation pause - which is the only
813 815 // time when we need know if the card contains references
814 816 // that point into the collection set. Also when the hot card
815 817 // cache is enabled, this code is executed by the concurrent
816 818 // refine threads - rather than the GC worker threads - and
817 819 // concurrentRefineOneCard_impl will return false.
818 820 assert(!oops_into_cset, "should not see true here");
819 821 }
820 822 }
821 823 }
822 824
823 825 if (!defer) {
824 826 oops_into_cset =
825 827 concurrentRefineOneCard_impl(card_ptr, worker_i, check_for_refs_into_cset);
826 828 // We should only be detecting that the card contains references
827 829 // that point into the collection set if the current thread is
828 830 // a GC worker thread.
829 831 assert(!oops_into_cset || SafepointSynchronize::is_at_safepoint(),
830 832 "invalid result at non safepoint");
831 833 }
832 834 return oops_into_cset;
833 835 }
834 836
835 837 class HRRSStatsIter: public HeapRegionClosure {
836 838 size_t _occupied;
837 839 size_t _total_mem_sz;
838 840 size_t _max_mem_sz;
839 841 HeapRegion* _max_mem_sz_region;
840 842 public:
841 843 HRRSStatsIter() :
842 844 _occupied(0),
843 845 _total_mem_sz(0),
844 846 _max_mem_sz(0),
845 847 _max_mem_sz_region(NULL)
846 848 {}
847 849
848 850 bool doHeapRegion(HeapRegion* r) {
849 851 if (r->continuesHumongous()) return false;
850 852 size_t mem_sz = r->rem_set()->mem_size();
851 853 if (mem_sz > _max_mem_sz) {
852 854 _max_mem_sz = mem_sz;
853 855 _max_mem_sz_region = r;
854 856 }
855 857 _total_mem_sz += mem_sz;
856 858 size_t occ = r->rem_set()->occupied();
857 859 _occupied += occ;
858 860 return false;
859 861 }
860 862 size_t total_mem_sz() { return _total_mem_sz; }
861 863 size_t max_mem_sz() { return _max_mem_sz; }
862 864 size_t occupied() { return _occupied; }
863 865 HeapRegion* max_mem_sz_region() { return _max_mem_sz_region; }
864 866 };
865 867
866 868 class PrintRSThreadVTimeClosure : public ThreadClosure {
867 869 public:
868 870 virtual void do_thread(Thread *t) {
869 871 ConcurrentG1RefineThread* crt = (ConcurrentG1RefineThread*) t;
870 872 gclog_or_tty->print(" %5.2f", crt->vtime_accum());
871 873 }
872 874 };
873 875
874 876 void G1RemSet::print_summary_info() {
875 877 G1CollectedHeap* g1 = G1CollectedHeap::heap();
876 878
877 879 #if CARD_REPEAT_HISTO
878 880 gclog_or_tty->print_cr("\nG1 card_repeat count histogram: ");
879 881 gclog_or_tty->print_cr(" # of repeats --> # of cards with that number.");
880 882 card_repeat_count.print_on(gclog_or_tty);
881 883 #endif
882 884
883 885 if (FILTEROUTOFREGIONCLOSURE_DOHISTOGRAMCOUNT) {
884 886 gclog_or_tty->print_cr("\nG1 rem-set out-of-region histogram: ");
885 887 gclog_or_tty->print_cr(" # of CS ptrs --> # of cards with that number.");
886 888 out_of_histo.print_on(gclog_or_tty);
887 889 }
888 890 gclog_or_tty->print_cr("\n Concurrent RS processed %d cards",
889 891 _conc_refine_cards);
890 892 DirtyCardQueueSet& dcqs = JavaThread::dirty_card_queue_set();
891 893 jint tot_processed_buffers =
892 894 dcqs.processed_buffers_mut() + dcqs.processed_buffers_rs_thread();
893 895 gclog_or_tty->print_cr(" Of %d completed buffers:", tot_processed_buffers);
894 896 gclog_or_tty->print_cr(" %8d (%5.1f%%) by conc RS threads.",
895 897 dcqs.processed_buffers_rs_thread(),
896 898 100.0*(float)dcqs.processed_buffers_rs_thread()/
897 899 (float)tot_processed_buffers);
898 900 gclog_or_tty->print_cr(" %8d (%5.1f%%) by mutator threads.",
899 901 dcqs.processed_buffers_mut(),
900 902 100.0*(float)dcqs.processed_buffers_mut()/
901 903 (float)tot_processed_buffers);
902 904 gclog_or_tty->print_cr(" Conc RS threads times(s)");
903 905 PrintRSThreadVTimeClosure p;
904 906 gclog_or_tty->print(" ");
905 907 g1->concurrent_g1_refine()->threads_do(&p);
906 908 gclog_or_tty->print_cr("");
907 909
908 910 HRRSStatsIter blk;
909 911 g1->heap_region_iterate(&blk);
910 912 gclog_or_tty->print_cr(" Total heap region rem set sizes = " SIZE_FORMAT "K."
911 913 " Max = " SIZE_FORMAT "K.",
912 914 blk.total_mem_sz()/K, blk.max_mem_sz()/K);
913 915 gclog_or_tty->print_cr(" Static structures = " SIZE_FORMAT "K,"
914 916 " free_lists = " SIZE_FORMAT "K.",
915 917 HeapRegionRemSet::static_mem_size()/K,
916 918 HeapRegionRemSet::fl_mem_size()/K);
917 919 gclog_or_tty->print_cr(" %d occupied cards represented.",
918 920 blk.occupied());
919 921 gclog_or_tty->print_cr(" Max sz region = [" PTR_FORMAT ", " PTR_FORMAT " )"
920 922 ", cap = " SIZE_FORMAT "K, occ = " SIZE_FORMAT "K.",
921 923 blk.max_mem_sz_region()->bottom(), blk.max_mem_sz_region()->end(),
922 924 (blk.max_mem_sz_region()->rem_set()->mem_size() + K - 1)/K,
923 925 (blk.max_mem_sz_region()->rem_set()->occupied() + K - 1)/K);
924 926 gclog_or_tty->print_cr(" Did %d coarsenings.", HeapRegionRemSet::n_coarsenings());
925 927 }
926 928
927 929 void G1RemSet::prepare_for_verify() {
928 930 if (G1HRRSFlushLogBuffersOnVerify &&
929 931 (VerifyBeforeGC || VerifyAfterGC)
930 932 && !_g1->full_collection()) {
931 933 cleanupHRRS();
932 934 _g1->set_refine_cte_cl_concurrency(false);
933 935 if (SafepointSynchronize::is_at_safepoint()) {
934 936 DirtyCardQueueSet& dcqs = JavaThread::dirty_card_queue_set();
935 937 dcqs.concatenate_logs();
936 938 }
937 939 bool cg1r_use_cache = _cg1r->use_cache();
938 940 _cg1r->set_use_cache(false);
939 941 DirtyCardQueue into_cset_dcq(&_g1->into_cset_dirty_card_queue_set());
940 942 updateRS(&into_cset_dcq, 0);
941 943 _g1->into_cset_dirty_card_queue_set().clear();
942 944 _cg1r->set_use_cache(cg1r_use_cache);
943 945
944 946 assert(JavaThread::dirty_card_queue_set().completed_buffers_num() == 0, "All should be consumed");
945 947 }
946 948 }
↓ open down ↓ |
291 lines elided |
↑ open up ↑ |
XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX