Mercurial > hg > graal-compiler
comparison src/share/vm/gc_implementation/g1/g1RemSet.cpp @ 342:37f87013dfd8
6711316: Open source the Garbage-First garbage collector
Summary: First mercurial integration of the code for the Garbage-First garbage collector.
Reviewed-by: apetrusenko, iveresov, jmasa, sgoldman, tonyp, ysr
author | ysr |
---|---|
date | Thu, 05 Jun 2008 15:57:56 -0700 |
parents | |
children | 58054a18d735 |
comparison
equal
deleted
inserted
replaced
189:0b27f3512f9e | 342:37f87013dfd8 |
---|---|
1 /* | |
2 * Copyright 2001-2007 Sun Microsystems, Inc. All Rights Reserved. | |
3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. | |
4 * | |
5 * This code is free software; you can redistribute it and/or modify it | |
6 * under the terms of the GNU General Public License version 2 only, as | |
7 * published by the Free Software Foundation. | |
8 * | |
9 * This code is distributed in the hope that it will be useful, but WITHOUT | |
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or | |
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License | |
12 * version 2 for more details (a copy is included in the LICENSE file that | |
13 * accompanied this code). | |
14 * | |
15 * You should have received a copy of the GNU General Public License version | |
16 * 2 along with this work; if not, write to the Free Software Foundation, | |
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA. | |
18 * | |
19 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara, | |
20 * CA 95054 USA or visit www.sun.com if you need additional information or | |
21 * have any questions. | |
22 * | |
23 */ | |
24 | |
25 #include "incls/_precompiled.incl" | |
26 #include "incls/_g1RemSet.cpp.incl" | |
27 | |
28 #define CARD_REPEAT_HISTO 0 | |
29 | |
30 #if CARD_REPEAT_HISTO | |
31 static size_t ct_freq_sz; | |
32 static jbyte* ct_freq = NULL; | |
33 | |
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 } | |
41 | |
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 } | |
46 | |
47 static IntHistogram card_repeat_count(10, 10); | |
48 | |
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 } | |
54 | |
55 } | |
56 #endif | |
57 | |
58 | |
59 class IntoCSOopClosure: public OopsInHeapRegionClosure { | |
60 OopsInHeapRegionClosure* _blk; | |
61 G1CollectedHeap* _g1; | |
62 public: | |
63 IntoCSOopClosure(G1CollectedHeap* g1, OopsInHeapRegionClosure* blk) : | |
64 _g1(g1), _blk(blk) {} | |
65 void set_region(HeapRegion* from) { | |
66 _blk->set_region(from); | |
67 } | |
68 virtual void do_oop(narrowOop* p) { | |
69 guarantee(false, "NYI"); | |
70 } | |
71 virtual void do_oop(oop* p) { | |
72 oop obj = *p; | |
73 if (_g1->obj_in_cs(obj)) _blk->do_oop(p); | |
74 } | |
75 bool apply_to_weak_ref_discovered_field() { return true; } | |
76 bool idempotent() { return true; } | |
77 }; | |
78 | |
79 class IntoCSRegionClosure: public HeapRegionClosure { | |
80 IntoCSOopClosure _blk; | |
81 G1CollectedHeap* _g1; | |
82 public: | |
83 IntoCSRegionClosure(G1CollectedHeap* g1, OopsInHeapRegionClosure* blk) : | |
84 _g1(g1), _blk(g1, blk) {} | |
85 bool doHeapRegion(HeapRegion* r) { | |
86 if (!r->in_collection_set()) { | |
87 _blk.set_region(r); | |
88 if (r->isHumongous()) { | |
89 if (r->startsHumongous()) { | |
90 oop obj = oop(r->bottom()); | |
91 obj->oop_iterate(&_blk); | |
92 } | |
93 } else { | |
94 r->oop_before_save_marks_iterate(&_blk); | |
95 } | |
96 } | |
97 return false; | |
98 } | |
99 }; | |
100 | |
101 void | |
102 StupidG1RemSet::oops_into_collection_set_do(OopsInHeapRegionClosure* oc, | |
103 int worker_i) { | |
104 IntoCSRegionClosure rc(_g1, oc); | |
105 _g1->heap_region_iterate(&rc); | |
106 } | |
107 | |
108 class UpdateRSOopClosure: public OopClosure { | |
109 HeapRegion* _from; | |
110 HRInto_G1RemSet* _rs; | |
111 int _worker_i; | |
112 public: | |
113 UpdateRSOopClosure(HRInto_G1RemSet* rs, int worker_i = 0) : | |
114 _from(NULL), _rs(rs), _worker_i(worker_i) { | |
115 guarantee(_rs != NULL, "Requires an HRIntoG1RemSet"); | |
116 } | |
117 | |
118 void set_from(HeapRegion* from) { | |
119 assert(from != NULL, "from region must be non-NULL"); | |
120 _from = from; | |
121 } | |
122 | |
123 virtual void do_oop(narrowOop* p) { | |
124 guarantee(false, "NYI"); | |
125 } | |
126 virtual void do_oop(oop* p) { | |
127 assert(_from != NULL, "from region must be non-NULL"); | |
128 _rs->par_write_ref(_from, p, _worker_i); | |
129 } | |
130 // Override: this closure is idempotent. | |
131 // bool idempotent() { return true; } | |
132 bool apply_to_weak_ref_discovered_field() { return true; } | |
133 }; | |
134 | |
135 class UpdateRSOutOfRegionClosure: public HeapRegionClosure { | |
136 G1CollectedHeap* _g1h; | |
137 ModRefBarrierSet* _mr_bs; | |
138 UpdateRSOopClosure _cl; | |
139 int _worker_i; | |
140 public: | |
141 UpdateRSOutOfRegionClosure(G1CollectedHeap* g1, int worker_i = 0) : | |
142 _cl(g1->g1_rem_set()->as_HRInto_G1RemSet(), worker_i), | |
143 _mr_bs(g1->mr_bs()), | |
144 _worker_i(worker_i), | |
145 _g1h(g1) | |
146 {} | |
147 bool doHeapRegion(HeapRegion* r) { | |
148 if (!r->in_collection_set() && !r->continuesHumongous()) { | |
149 _cl.set_from(r); | |
150 r->set_next_filter_kind(HeapRegionDCTOC::OutOfRegionFilterKind); | |
151 _mr_bs->mod_oop_in_space_iterate(r, &_cl, true, true); | |
152 } | |
153 return false; | |
154 } | |
155 }; | |
156 | |
157 class VerifyRSCleanCardOopClosure: public OopClosure { | |
158 G1CollectedHeap* _g1; | |
159 public: | |
160 VerifyRSCleanCardOopClosure(G1CollectedHeap* g1) : _g1(g1) {} | |
161 | |
162 virtual void do_oop(narrowOop* p) { | |
163 guarantee(false, "NYI"); | |
164 } | |
165 virtual void do_oop(oop* p) { | |
166 oop obj = *p; | |
167 HeapRegion* to = _g1->heap_region_containing(obj); | |
168 guarantee(to == NULL || !to->in_collection_set(), | |
169 "Missed a rem set member."); | |
170 } | |
171 }; | |
172 | |
173 HRInto_G1RemSet::HRInto_G1RemSet(G1CollectedHeap* g1, CardTableModRefBS* ct_bs) | |
174 : G1RemSet(g1), _ct_bs(ct_bs), _g1p(_g1->g1_policy()), | |
175 _cg1r(g1->concurrent_g1_refine()), | |
176 _par_traversal_in_progress(false), _new_refs(NULL), | |
177 _cards_scanned(NULL), _total_cards_scanned(0) | |
178 { | |
179 _seq_task = new SubTasksDone(NumSeqTasks); | |
180 _new_refs = NEW_C_HEAP_ARRAY(GrowableArray<oop*>*, ParallelGCThreads); | |
181 } | |
182 | |
183 HRInto_G1RemSet::~HRInto_G1RemSet() { | |
184 delete _seq_task; | |
185 } | |
186 | |
187 void CountNonCleanMemRegionClosure::do_MemRegion(MemRegion mr) { | |
188 if (_g1->is_in_g1_reserved(mr.start())) { | |
189 _n += (int) ((mr.byte_size() / CardTableModRefBS::card_size)); | |
190 if (_start_first == NULL) _start_first = mr.start(); | |
191 } | |
192 } | |
193 | |
194 class ScanRSClosure : public HeapRegionClosure { | |
195 size_t _cards_done, _cards; | |
196 G1CollectedHeap* _g1h; | |
197 OopsInHeapRegionClosure* _oc; | |
198 G1BlockOffsetSharedArray* _bot_shared; | |
199 CardTableModRefBS *_ct_bs; | |
200 int _worker_i; | |
201 bool _try_claimed; | |
202 public: | |
203 ScanRSClosure(OopsInHeapRegionClosure* oc, int worker_i) : | |
204 _oc(oc), | |
205 _cards(0), | |
206 _cards_done(0), | |
207 _worker_i(worker_i), | |
208 _try_claimed(false) | |
209 { | |
210 _g1h = G1CollectedHeap::heap(); | |
211 _bot_shared = _g1h->bot_shared(); | |
212 _ct_bs = (CardTableModRefBS*) (_g1h->barrier_set()); | |
213 } | |
214 | |
215 void set_try_claimed() { _try_claimed = true; } | |
216 | |
217 void scanCard(size_t index, HeapRegion *r) { | |
218 _cards_done++; | |
219 DirtyCardToOopClosure* cl = | |
220 r->new_dcto_closure(_oc, | |
221 CardTableModRefBS::Precise, | |
222 HeapRegionDCTOC::IntoCSFilterKind); | |
223 | |
224 // Set the "from" region in the closure. | |
225 _oc->set_region(r); | |
226 HeapWord* card_start = _bot_shared->address_for_index(index); | |
227 HeapWord* card_end = card_start + G1BlockOffsetSharedArray::N_words; | |
228 Space *sp = SharedHeap::heap()->space_containing(card_start); | |
229 MemRegion sm_region; | |
230 if (ParallelGCThreads > 0) { | |
231 // first find the used area | |
232 sm_region = sp->used_region_at_save_marks(); | |
233 } else { | |
234 // The closure is not idempotent. We shouldn't look at objects | |
235 // allocated during the GC. | |
236 sm_region = sp->used_region_at_save_marks(); | |
237 } | |
238 MemRegion mr = sm_region.intersection(MemRegion(card_start,card_end)); | |
239 if (!mr.is_empty()) { | |
240 cl->do_MemRegion(mr); | |
241 } | |
242 } | |
243 | |
244 void printCard(HeapRegion* card_region, size_t card_index, | |
245 HeapWord* card_start) { | |
246 gclog_or_tty->print_cr("T %d Region [" PTR_FORMAT ", " PTR_FORMAT ") " | |
247 "RS names card %p: " | |
248 "[" PTR_FORMAT ", " PTR_FORMAT ")", | |
249 _worker_i, | |
250 card_region->bottom(), card_region->end(), | |
251 card_index, | |
252 card_start, card_start + G1BlockOffsetSharedArray::N_words); | |
253 } | |
254 | |
255 bool doHeapRegion(HeapRegion* r) { | |
256 assert(r->in_collection_set(), "should only be called on elements of CS."); | |
257 HeapRegionRemSet* hrrs = r->rem_set(); | |
258 if (hrrs->iter_is_complete()) return false; // All done. | |
259 if (!_try_claimed && !hrrs->claim_iter()) return false; | |
260 // If we didn't return above, then | |
261 // _try_claimed || r->claim_iter() | |
262 // is true: either we're supposed to work on claimed-but-not-complete | |
263 // regions, or we successfully claimed the region. | |
264 HeapRegionRemSetIterator* iter = _g1h->rem_set_iterator(_worker_i); | |
265 hrrs->init_iterator(iter); | |
266 size_t card_index; | |
267 while (iter->has_next(card_index)) { | |
268 HeapWord* card_start = _g1h->bot_shared()->address_for_index(card_index); | |
269 | |
270 #if 0 | |
271 gclog_or_tty->print("Rem set iteration yielded card [" PTR_FORMAT ", " PTR_FORMAT ").\n", | |
272 card_start, card_start + CardTableModRefBS::card_size_in_words); | |
273 #endif | |
274 | |
275 HeapRegion* card_region = _g1h->heap_region_containing(card_start); | |
276 assert(card_region != NULL, "Yielding cards not in the heap?"); | |
277 _cards++; | |
278 | |
279 if (!card_region->in_collection_set()) { | |
280 // If the card is dirty, then we will scan it during updateRS. | |
281 if (!_ct_bs->is_card_claimed(card_index) && | |
282 !_ct_bs->is_card_dirty(card_index)) { | |
283 assert(_ct_bs->is_card_clean(card_index) || | |
284 _ct_bs->is_card_claimed(card_index), | |
285 "Card is either dirty, clean, or claimed"); | |
286 if (_ct_bs->claim_card(card_index)) | |
287 scanCard(card_index, card_region); | |
288 } | |
289 } | |
290 } | |
291 hrrs->set_iter_complete(); | |
292 return false; | |
293 } | |
294 // Set all cards back to clean. | |
295 void cleanup() {_g1h->cleanUpCardTable();} | |
296 size_t cards_done() { return _cards_done;} | |
297 size_t cards_looked_up() { return _cards;} | |
298 }; | |
299 | |
300 // We want the parallel threads to start their scanning at | |
301 // different collection set regions to avoid contention. | |
302 // If we have: | |
303 // n collection set regions | |
304 // p threads | |
305 // Then thread t will start at region t * floor (n/p) | |
306 | |
307 HeapRegion* HRInto_G1RemSet::calculateStartRegion(int worker_i) { | |
308 HeapRegion* result = _g1p->collection_set(); | |
309 if (ParallelGCThreads > 0) { | |
310 size_t cs_size = _g1p->collection_set_size(); | |
311 int n_workers = _g1->workers()->total_workers(); | |
312 size_t cs_spans = cs_size / n_workers; | |
313 size_t ind = cs_spans * worker_i; | |
314 for (size_t i = 0; i < ind; i++) | |
315 result = result->next_in_collection_set(); | |
316 } | |
317 return result; | |
318 } | |
319 | |
320 void HRInto_G1RemSet::scanRS(OopsInHeapRegionClosure* oc, int worker_i) { | |
321 double rs_time_start = os::elapsedTime(); | |
322 HeapRegion *startRegion = calculateStartRegion(worker_i); | |
323 | |
324 BufferingOopsInHeapRegionClosure boc(oc); | |
325 ScanRSClosure scanRScl(&boc, worker_i); | |
326 _g1->collection_set_iterate_from(startRegion, &scanRScl); | |
327 scanRScl.set_try_claimed(); | |
328 _g1->collection_set_iterate_from(startRegion, &scanRScl); | |
329 | |
330 boc.done(); | |
331 double closure_app_time_sec = boc.closure_app_seconds(); | |
332 double scan_rs_time_sec = (os::elapsedTime() - rs_time_start) - | |
333 closure_app_time_sec; | |
334 double closure_app_time_ms = closure_app_time_sec * 1000.0; | |
335 | |
336 assert( _cards_scanned != NULL, "invariant" ); | |
337 _cards_scanned[worker_i] = scanRScl.cards_done(); | |
338 | |
339 _g1p->record_scan_rs_start_time(worker_i, rs_time_start * 1000.0); | |
340 _g1p->record_scan_rs_time(worker_i, scan_rs_time_sec * 1000.0); | |
341 if (ParallelGCThreads > 0) { | |
342 // In this case, we called scanNewRefsRS and recorded the corresponding | |
343 // time. | |
344 double scan_new_refs_time_ms = _g1p->get_scan_new_refs_time(worker_i); | |
345 if (scan_new_refs_time_ms > 0.0) { | |
346 closure_app_time_ms += scan_new_refs_time_ms; | |
347 } | |
348 } | |
349 _g1p->record_obj_copy_time(worker_i, closure_app_time_ms); | |
350 } | |
351 | |
352 void HRInto_G1RemSet::updateRS(int worker_i) { | |
353 ConcurrentG1Refine* cg1r = _g1->concurrent_g1_refine(); | |
354 | |
355 double start = os::elapsedTime(); | |
356 _g1p->record_update_rs_start_time(worker_i, start * 1000.0); | |
357 | |
358 if (G1RSBarrierUseQueue && !cg1r->do_traversal()) { | |
359 // Apply the appropriate closure to all remaining log entries. | |
360 _g1->iterate_dirty_card_closure(false, worker_i); | |
361 // Now there should be no dirty cards. | |
362 if (G1RSLogCheckCardTable) { | |
363 CountNonCleanMemRegionClosure cl(_g1); | |
364 _ct_bs->mod_card_iterate(&cl); | |
365 // XXX This isn't true any more: keeping cards of young regions | |
366 // marked dirty broke it. Need some reasonable fix. | |
367 guarantee(cl.n() == 0, "Card table should be clean."); | |
368 } | |
369 } else { | |
370 UpdateRSOutOfRegionClosure update_rs(_g1, worker_i); | |
371 _g1->heap_region_iterate(&update_rs); | |
372 // We did a traversal; no further one is necessary. | |
373 if (G1RSBarrierUseQueue) { | |
374 assert(cg1r->do_traversal(), "Or we shouldn't have gotten here."); | |
375 cg1r->set_pya_cancel(); | |
376 } | |
377 if (_cg1r->use_cache()) { | |
378 _cg1r->clear_and_record_card_counts(); | |
379 _cg1r->clear_hot_cache(); | |
380 } | |
381 } | |
382 _g1p->record_update_rs_time(worker_i, (os::elapsedTime() - start) * 1000.0); | |
383 } | |
384 | |
385 #ifndef PRODUCT | |
386 class PrintRSClosure : public HeapRegionClosure { | |
387 int _count; | |
388 public: | |
389 PrintRSClosure() : _count(0) {} | |
390 bool doHeapRegion(HeapRegion* r) { | |
391 HeapRegionRemSet* hrrs = r->rem_set(); | |
392 _count += (int) hrrs->occupied(); | |
393 if (hrrs->occupied() == 0) { | |
394 gclog_or_tty->print("Heap Region [" PTR_FORMAT ", " PTR_FORMAT ") " | |
395 "has no remset entries\n", | |
396 r->bottom(), r->end()); | |
397 } else { | |
398 gclog_or_tty->print("Printing rem set for heap region [" PTR_FORMAT ", " PTR_FORMAT ")\n", | |
399 r->bottom(), r->end()); | |
400 r->print(); | |
401 hrrs->print(); | |
402 gclog_or_tty->print("\nDone printing rem set\n"); | |
403 } | |
404 return false; | |
405 } | |
406 int occupied() {return _count;} | |
407 }; | |
408 #endif | |
409 | |
410 class CountRSSizeClosure: public HeapRegionClosure { | |
411 size_t _n; | |
412 size_t _tot; | |
413 size_t _max; | |
414 HeapRegion* _max_r; | |
415 enum { | |
416 N = 20, | |
417 MIN = 6 | |
418 }; | |
419 int _histo[N]; | |
420 public: | |
421 CountRSSizeClosure() : _n(0), _tot(0), _max(0), _max_r(NULL) { | |
422 for (int i = 0; i < N; i++) _histo[i] = 0; | |
423 } | |
424 bool doHeapRegion(HeapRegion* r) { | |
425 if (!r->continuesHumongous()) { | |
426 size_t occ = r->rem_set()->occupied(); | |
427 _n++; | |
428 _tot += occ; | |
429 if (occ > _max) { | |
430 _max = occ; | |
431 _max_r = r; | |
432 } | |
433 // Fit it into a histo bin. | |
434 int s = 1 << MIN; | |
435 int i = 0; | |
436 while (occ > (size_t) s && i < (N-1)) { | |
437 s = s << 1; | |
438 i++; | |
439 } | |
440 _histo[i]++; | |
441 } | |
442 return false; | |
443 } | |
444 size_t n() { return _n; } | |
445 size_t tot() { return _tot; } | |
446 size_t mx() { return _max; } | |
447 HeapRegion* mxr() { return _max_r; } | |
448 void print_histo() { | |
449 int mx = N; | |
450 while (mx >= 0) { | |
451 if (_histo[mx-1] > 0) break; | |
452 mx--; | |
453 } | |
454 gclog_or_tty->print_cr("Number of regions with given RS sizes:"); | |
455 gclog_or_tty->print_cr(" <= %8d %8d", 1 << MIN, _histo[0]); | |
456 for (int i = 1; i < mx-1; i++) { | |
457 gclog_or_tty->print_cr(" %8d - %8d %8d", | |
458 (1 << (MIN + i - 1)) + 1, | |
459 1 << (MIN + i), | |
460 _histo[i]); | |
461 } | |
462 gclog_or_tty->print_cr(" > %8d %8d", (1 << (MIN+mx-2))+1, _histo[mx-1]); | |
463 } | |
464 }; | |
465 | |
466 void | |
467 HRInto_G1RemSet::scanNewRefsRS(OopsInHeapRegionClosure* oc, | |
468 int worker_i) { | |
469 double scan_new_refs_start_sec = os::elapsedTime(); | |
470 G1CollectedHeap* g1h = G1CollectedHeap::heap(); | |
471 CardTableModRefBS* ct_bs = (CardTableModRefBS*) (g1h->barrier_set()); | |
472 while (_new_refs[worker_i]->is_nonempty()) { | |
473 oop* p = _new_refs[worker_i]->pop(); | |
474 oop obj = *p; | |
475 // *p was in the collection set when p was pushed on "_new_refs", but | |
476 // another thread may have processed this location from an RS, so it | |
477 // might not point into the CS any longer. If so, it's obviously been | |
478 // processed, and we don't need to do anything further. | |
479 if (g1h->obj_in_cs(obj)) { | |
480 HeapRegion* r = g1h->heap_region_containing(p); | |
481 | |
482 DEBUG_ONLY(HeapRegion* to = g1h->heap_region_containing(obj)); | |
483 assert(ParallelGCThreads > 1 | |
484 || to->rem_set()->contains_reference(p), | |
485 "Invariant: pushed after being added." | |
486 "(Not reliable in parallel code.)"); | |
487 oc->set_region(r); | |
488 // If "p" has already been processed concurrently, this is | |
489 // idempotent. | |
490 oc->do_oop(p); | |
491 } | |
492 } | |
493 _g1p->record_scan_new_refs_time(worker_i, | |
494 (os::elapsedTime() - scan_new_refs_start_sec) | |
495 * 1000.0); | |
496 } | |
497 | |
498 void HRInto_G1RemSet::set_par_traversal(bool b) { | |
499 _par_traversal_in_progress = b; | |
500 HeapRegionRemSet::set_par_traversal(b); | |
501 } | |
502 | |
503 void HRInto_G1RemSet::cleanupHRRS() { | |
504 HeapRegionRemSet::cleanup(); | |
505 } | |
506 | |
507 void | |
508 HRInto_G1RemSet::oops_into_collection_set_do(OopsInHeapRegionClosure* oc, | |
509 int worker_i) { | |
510 #if CARD_REPEAT_HISTO | |
511 ct_freq_update_histo_and_reset(); | |
512 #endif | |
513 if (worker_i == 0) { | |
514 _cg1r->clear_and_record_card_counts(); | |
515 } | |
516 | |
517 // Make this into a command-line flag... | |
518 if (G1RSCountHisto && (ParallelGCThreads == 0 || worker_i == 0)) { | |
519 CountRSSizeClosure count_cl; | |
520 _g1->heap_region_iterate(&count_cl); | |
521 gclog_or_tty->print_cr("Avg of %d RS counts is %f, max is %d, " | |
522 "max region is " PTR_FORMAT, | |
523 count_cl.n(), (float)count_cl.tot()/(float)count_cl.n(), | |
524 count_cl.mx(), count_cl.mxr()); | |
525 count_cl.print_histo(); | |
526 } | |
527 | |
528 if (ParallelGCThreads > 0) { | |
529 // This is a temporary change to serialize the update and scanning | |
530 // of remembered sets. There are some race conditions when this is | |
531 // done in parallel and they are causing failures. When we resolve | |
532 // said race conditions, we'll revert back to parallel remembered | |
533 // set updating and scanning. See CRs 6677707 and 6677708. | |
534 if (worker_i == 0) { | |
535 updateRS(worker_i); | |
536 scanNewRefsRS(oc, worker_i); | |
537 scanRS(oc, worker_i); | |
538 } | |
539 } else { | |
540 assert(worker_i == 0, "invariant"); | |
541 | |
542 updateRS(0); | |
543 scanRS(oc, 0); | |
544 } | |
545 } | |
546 | |
547 void HRInto_G1RemSet:: | |
548 prepare_for_oops_into_collection_set_do() { | |
549 #if G1_REM_SET_LOGGING | |
550 PrintRSClosure cl; | |
551 _g1->collection_set_iterate(&cl); | |
552 #endif | |
553 cleanupHRRS(); | |
554 ConcurrentG1Refine* cg1r = _g1->concurrent_g1_refine(); | |
555 _g1->set_refine_cte_cl_concurrency(false); | |
556 DirtyCardQueueSet& dcqs = JavaThread::dirty_card_queue_set(); | |
557 dcqs.concatenate_logs(); | |
558 | |
559 assert(!_par_traversal_in_progress, "Invariant between iterations."); | |
560 if (ParallelGCThreads > 0) { | |
561 set_par_traversal(true); | |
562 int n_workers = _g1->workers()->total_workers(); | |
563 _seq_task->set_par_threads(n_workers); | |
564 for (uint i = 0; i < ParallelGCThreads; i++) | |
565 _new_refs[i] = new (ResourceObj::C_HEAP) GrowableArray<oop*>(8192,true); | |
566 | |
567 if (cg1r->do_traversal()) { | |
568 updateRS(0); | |
569 // Have to do this again after updaters | |
570 cleanupHRRS(); | |
571 } | |
572 } | |
573 guarantee( _cards_scanned == NULL, "invariant" ); | |
574 _cards_scanned = NEW_C_HEAP_ARRAY(size_t, n_workers()); | |
575 _total_cards_scanned = 0; | |
576 } | |
577 | |
578 | |
579 class cleanUpIteratorsClosure : public HeapRegionClosure { | |
580 bool doHeapRegion(HeapRegion *r) { | |
581 HeapRegionRemSet* hrrs = r->rem_set(); | |
582 hrrs->init_for_par_iteration(); | |
583 return false; | |
584 } | |
585 }; | |
586 | |
587 void HRInto_G1RemSet::cleanup_after_oops_into_collection_set_do() { | |
588 guarantee( _cards_scanned != NULL, "invariant" ); | |
589 _total_cards_scanned = 0; | |
590 for (uint i = 0; i < n_workers(); ++i) | |
591 _total_cards_scanned += _cards_scanned[i]; | |
592 FREE_C_HEAP_ARRAY(size_t, _cards_scanned); | |
593 _cards_scanned = NULL; | |
594 // Cleanup after copy | |
595 #if G1_REM_SET_LOGGING | |
596 PrintRSClosure cl; | |
597 _g1->heap_region_iterate(&cl); | |
598 #endif | |
599 _g1->set_refine_cte_cl_concurrency(true); | |
600 cleanUpIteratorsClosure iterClosure; | |
601 _g1->collection_set_iterate(&iterClosure); | |
602 // Set all cards back to clean. | |
603 _g1->cleanUpCardTable(); | |
604 if (ParallelGCThreads > 0) { | |
605 ConcurrentG1Refine* cg1r = _g1->concurrent_g1_refine(); | |
606 if (cg1r->do_traversal()) { | |
607 cg1r->cg1rThread()->set_do_traversal(false); | |
608 } | |
609 for (uint i = 0; i < ParallelGCThreads; i++) { | |
610 delete _new_refs[i]; | |
611 } | |
612 set_par_traversal(false); | |
613 } | |
614 assert(!_par_traversal_in_progress, "Invariant between iterations."); | |
615 } | |
616 | |
617 class UpdateRSObjectClosure: public ObjectClosure { | |
618 UpdateRSOopClosure* _update_rs_oop_cl; | |
619 public: | |
620 UpdateRSObjectClosure(UpdateRSOopClosure* update_rs_oop_cl) : | |
621 _update_rs_oop_cl(update_rs_oop_cl) {} | |
622 void do_object(oop obj) { | |
623 obj->oop_iterate(_update_rs_oop_cl); | |
624 } | |
625 | |
626 }; | |
627 | |
628 class ScrubRSClosure: public HeapRegionClosure { | |
629 G1CollectedHeap* _g1h; | |
630 BitMap* _region_bm; | |
631 BitMap* _card_bm; | |
632 CardTableModRefBS* _ctbs; | |
633 public: | |
634 ScrubRSClosure(BitMap* region_bm, BitMap* card_bm) : | |
635 _g1h(G1CollectedHeap::heap()), | |
636 _region_bm(region_bm), _card_bm(card_bm), | |
637 _ctbs(NULL) | |
638 { | |
639 ModRefBarrierSet* bs = _g1h->mr_bs(); | |
640 guarantee(bs->is_a(BarrierSet::CardTableModRef), "Precondition"); | |
641 _ctbs = (CardTableModRefBS*)bs; | |
642 } | |
643 | |
644 bool doHeapRegion(HeapRegion* r) { | |
645 if (!r->continuesHumongous()) { | |
646 r->rem_set()->scrub(_ctbs, _region_bm, _card_bm); | |
647 } | |
648 return false; | |
649 } | |
650 }; | |
651 | |
652 void HRInto_G1RemSet::scrub(BitMap* region_bm, BitMap* card_bm) { | |
653 ScrubRSClosure scrub_cl(region_bm, card_bm); | |
654 _g1->heap_region_iterate(&scrub_cl); | |
655 } | |
656 | |
657 void HRInto_G1RemSet::scrub_par(BitMap* region_bm, BitMap* card_bm, | |
658 int worker_num, int claim_val) { | |
659 ScrubRSClosure scrub_cl(region_bm, card_bm); | |
660 _g1->heap_region_par_iterate_chunked(&scrub_cl, worker_num, claim_val); | |
661 } | |
662 | |
663 | |
664 class ConcRefineRegionClosure: public HeapRegionClosure { | |
665 G1CollectedHeap* _g1h; | |
666 CardTableModRefBS* _ctbs; | |
667 ConcurrentGCThread* _cgc_thrd; | |
668 ConcurrentG1Refine* _cg1r; | |
669 unsigned _cards_processed; | |
670 UpdateRSOopClosure _update_rs_oop_cl; | |
671 public: | |
672 ConcRefineRegionClosure(CardTableModRefBS* ctbs, | |
673 ConcurrentG1Refine* cg1r, | |
674 HRInto_G1RemSet* g1rs) : | |
675 _ctbs(ctbs), _cg1r(cg1r), _cgc_thrd(cg1r->cg1rThread()), | |
676 _update_rs_oop_cl(g1rs), _cards_processed(0), | |
677 _g1h(G1CollectedHeap::heap()) | |
678 {} | |
679 | |
680 bool doHeapRegion(HeapRegion* r) { | |
681 if (!r->in_collection_set() && | |
682 !r->continuesHumongous() && | |
683 !r->is_young()) { | |
684 _update_rs_oop_cl.set_from(r); | |
685 UpdateRSObjectClosure update_rs_obj_cl(&_update_rs_oop_cl); | |
686 | |
687 // For each run of dirty card in the region: | |
688 // 1) Clear the cards. | |
689 // 2) Process the range corresponding to the run, adding any | |
690 // necessary RS entries. | |
691 // 1 must precede 2, so that a concurrent modification redirties the | |
692 // card. If a processing attempt does not succeed, because it runs | |
693 // into an unparseable region, we will do binary search to find the | |
694 // beginning of the next parseable region. | |
695 HeapWord* startAddr = r->bottom(); | |
696 HeapWord* endAddr = r->used_region().end(); | |
697 HeapWord* lastAddr; | |
698 HeapWord* nextAddr; | |
699 | |
700 for (nextAddr = lastAddr = startAddr; | |
701 nextAddr < endAddr; | |
702 nextAddr = lastAddr) { | |
703 MemRegion dirtyRegion; | |
704 | |
705 // Get and clear dirty region from card table | |
706 MemRegion next_mr(nextAddr, endAddr); | |
707 dirtyRegion = | |
708 _ctbs->dirty_card_range_after_reset( | |
709 next_mr, | |
710 true, CardTableModRefBS::clean_card_val()); | |
711 assert(dirtyRegion.start() >= nextAddr, | |
712 "returned region inconsistent?"); | |
713 | |
714 if (!dirtyRegion.is_empty()) { | |
715 HeapWord* stop_point = | |
716 r->object_iterate_mem_careful(dirtyRegion, | |
717 &update_rs_obj_cl); | |
718 if (stop_point == NULL) { | |
719 lastAddr = dirtyRegion.end(); | |
720 _cards_processed += | |
721 (int) (dirtyRegion.word_size() / CardTableModRefBS::card_size_in_words); | |
722 } else { | |
723 // We're going to skip one or more cards that we can't parse. | |
724 HeapWord* next_parseable_card = | |
725 r->next_block_start_careful(stop_point); | |
726 // Round this up to a card boundary. | |
727 next_parseable_card = | |
728 _ctbs->addr_for(_ctbs->byte_after_const(next_parseable_card)); | |
729 // Now we invalidate the intervening cards so we'll see them | |
730 // again. | |
731 MemRegion remaining_dirty = | |
732 MemRegion(stop_point, dirtyRegion.end()); | |
733 MemRegion skipped = | |
734 MemRegion(stop_point, next_parseable_card); | |
735 _ctbs->invalidate(skipped.intersection(remaining_dirty)); | |
736 | |
737 // Now start up again where we can parse. | |
738 lastAddr = next_parseable_card; | |
739 | |
740 // Count how many we did completely. | |
741 _cards_processed += | |
742 (stop_point - dirtyRegion.start()) / | |
743 CardTableModRefBS::card_size_in_words; | |
744 } | |
745 // Allow interruption at regular intervals. | |
746 // (Might need to make them more regular, if we get big | |
747 // dirty regions.) | |
748 if (_cgc_thrd != NULL) { | |
749 if (_cgc_thrd->should_yield()) { | |
750 _cgc_thrd->yield(); | |
751 switch (_cg1r->get_pya()) { | |
752 case PYA_continue: | |
753 // This may have changed: re-read. | |
754 endAddr = r->used_region().end(); | |
755 continue; | |
756 case PYA_restart: case PYA_cancel: | |
757 return true; | |
758 } | |
759 } | |
760 } | |
761 } else { | |
762 break; | |
763 } | |
764 } | |
765 } | |
766 // A good yield opportunity. | |
767 if (_cgc_thrd != NULL) { | |
768 if (_cgc_thrd->should_yield()) { | |
769 _cgc_thrd->yield(); | |
770 switch (_cg1r->get_pya()) { | |
771 case PYA_restart: case PYA_cancel: | |
772 return true; | |
773 default: | |
774 break; | |
775 } | |
776 | |
777 } | |
778 } | |
779 return false; | |
780 } | |
781 | |
782 unsigned cards_processed() { return _cards_processed; } | |
783 }; | |
784 | |
785 | |
786 void HRInto_G1RemSet::concurrentRefinementPass(ConcurrentG1Refine* cg1r) { | |
787 ConcRefineRegionClosure cr_cl(ct_bs(), cg1r, this); | |
788 _g1->heap_region_iterate(&cr_cl); | |
789 _conc_refine_traversals++; | |
790 _conc_refine_cards += cr_cl.cards_processed(); | |
791 } | |
792 | |
793 static IntHistogram out_of_histo(50, 50); | |
794 | |
795 | |
796 | |
797 void HRInto_G1RemSet::concurrentRefineOneCard(jbyte* card_ptr, int worker_i) { | |
798 // If the card is no longer dirty, nothing to do. | |
799 if (*card_ptr != CardTableModRefBS::dirty_card_val()) return; | |
800 | |
801 // Construct the region representing the card. | |
802 HeapWord* start = _ct_bs->addr_for(card_ptr); | |
803 // And find the region containing it. | |
804 HeapRegion* r = _g1->heap_region_containing(start); | |
805 if (r == NULL) { | |
806 guarantee(_g1->is_in_permanent(start), "Or else where?"); | |
807 return; // Not in the G1 heap (might be in perm, for example.) | |
808 } | |
809 // Why do we have to check here whether a card is on a young region, | |
810 // given that we dirty young regions and, as a result, the | |
811 // post-barrier is supposed to filter them out and never to enqueue | |
812 // them? When we allocate a new region as the "allocation region" we | |
813 // actually dirty its cards after we release the lock, since card | |
814 // dirtying while holding the lock was a performance bottleneck. So, | |
815 // as a result, it is possible for other threads to actually | |
816 // allocate objects in the region (after the acquire the lock) | |
817 // before all the cards on the region are dirtied. This is unlikely, | |
818 // and it doesn't happen often, but it can happen. So, the extra | |
819 // check below filters out those cards. | |
820 if (r->is_young()) { | |
821 return; | |
822 } | |
823 // While we are processing RSet buffers during the collection, we | |
824 // actually don't want to scan any cards on the collection set, | |
825 // since we don't want to update remebered sets with entries that | |
826 // point into the collection set, given that live objects from the | |
827 // collection set are about to move and such entries will be stale | |
828 // very soon. This change also deals with a reliability issue which | |
829 // involves scanning a card in the collection set and coming across | |
830 // an array that was being chunked and looking malformed. Note, | |
831 // however, that if evacuation fails, we have to scan any objects | |
832 // that were not moved and create any missing entries. | |
833 if (r->in_collection_set()) { | |
834 return; | |
835 } | |
836 | |
837 // Should we defer it? | |
838 if (_cg1r->use_cache()) { | |
839 card_ptr = _cg1r->cache_insert(card_ptr); | |
840 // If it was not an eviction, nothing to do. | |
841 if (card_ptr == NULL) return; | |
842 | |
843 // OK, we have to reset the card start, region, etc. | |
844 start = _ct_bs->addr_for(card_ptr); | |
845 r = _g1->heap_region_containing(start); | |
846 if (r == NULL) { | |
847 guarantee(_g1->is_in_permanent(start), "Or else where?"); | |
848 return; // Not in the G1 heap (might be in perm, for example.) | |
849 } | |
850 guarantee(!r->is_young(), "It was evicted in the current minor cycle."); | |
851 } | |
852 | |
853 HeapWord* end = _ct_bs->addr_for(card_ptr + 1); | |
854 MemRegion dirtyRegion(start, end); | |
855 | |
856 #if CARD_REPEAT_HISTO | |
857 init_ct_freq_table(_g1->g1_reserved_obj_bytes()); | |
858 ct_freq_note_card(_ct_bs->index_for(start)); | |
859 #endif | |
860 | |
861 UpdateRSOopClosure update_rs_oop_cl(this, worker_i); | |
862 update_rs_oop_cl.set_from(r); | |
863 FilterOutOfRegionClosure filter_then_update_rs_oop_cl(r, &update_rs_oop_cl); | |
864 | |
865 // Undirty the card. | |
866 *card_ptr = CardTableModRefBS::clean_card_val(); | |
867 // We must complete this write before we do any of the reads below. | |
868 OrderAccess::storeload(); | |
869 // And process it, being careful of unallocated portions of TLAB's. | |
870 HeapWord* stop_point = | |
871 r->oops_on_card_seq_iterate_careful(dirtyRegion, | |
872 &filter_then_update_rs_oop_cl); | |
873 // If stop_point is non-null, then we encountered an unallocated region | |
874 // (perhaps the unfilled portion of a TLAB.) For now, we'll dirty the | |
875 // card and re-enqueue: if we put off the card until a GC pause, then the | |
876 // unallocated portion will be filled in. Alternatively, we might try | |
877 // the full complexity of the technique used in "regular" precleaning. | |
878 if (stop_point != NULL) { | |
879 // The card might have gotten re-dirtied and re-enqueued while we | |
880 // worked. (In fact, it's pretty likely.) | |
881 if (*card_ptr != CardTableModRefBS::dirty_card_val()) { | |
882 *card_ptr = CardTableModRefBS::dirty_card_val(); | |
883 MutexLockerEx x(Shared_DirtyCardQ_lock, | |
884 Mutex::_no_safepoint_check_flag); | |
885 DirtyCardQueue* sdcq = | |
886 JavaThread::dirty_card_queue_set().shared_dirty_card_queue(); | |
887 sdcq->enqueue(card_ptr); | |
888 } | |
889 } else { | |
890 out_of_histo.add_entry(filter_then_update_rs_oop_cl.out_of_region()); | |
891 _conc_refine_cards++; | |
892 } | |
893 } | |
894 | |
895 class HRRSStatsIter: public HeapRegionClosure { | |
896 size_t _occupied; | |
897 size_t _total_mem_sz; | |
898 size_t _max_mem_sz; | |
899 HeapRegion* _max_mem_sz_region; | |
900 public: | |
901 HRRSStatsIter() : | |
902 _occupied(0), | |
903 _total_mem_sz(0), | |
904 _max_mem_sz(0), | |
905 _max_mem_sz_region(NULL) | |
906 {} | |
907 | |
908 bool doHeapRegion(HeapRegion* r) { | |
909 if (r->continuesHumongous()) return false; | |
910 size_t mem_sz = r->rem_set()->mem_size(); | |
911 if (mem_sz > _max_mem_sz) { | |
912 _max_mem_sz = mem_sz; | |
913 _max_mem_sz_region = r; | |
914 } | |
915 _total_mem_sz += mem_sz; | |
916 size_t occ = r->rem_set()->occupied(); | |
917 _occupied += occ; | |
918 return false; | |
919 } | |
920 size_t total_mem_sz() { return _total_mem_sz; } | |
921 size_t max_mem_sz() { return _max_mem_sz; } | |
922 size_t occupied() { return _occupied; } | |
923 HeapRegion* max_mem_sz_region() { return _max_mem_sz_region; } | |
924 }; | |
925 | |
926 void HRInto_G1RemSet::print_summary_info() { | |
927 G1CollectedHeap* g1 = G1CollectedHeap::heap(); | |
928 ConcurrentG1RefineThread* cg1r_thrd = | |
929 g1->concurrent_g1_refine()->cg1rThread(); | |
930 | |
931 #if CARD_REPEAT_HISTO | |
932 gclog_or_tty->print_cr("\nG1 card_repeat count histogram: "); | |
933 gclog_or_tty->print_cr(" # of repeats --> # of cards with that number."); | |
934 card_repeat_count.print_on(gclog_or_tty); | |
935 #endif | |
936 | |
937 if (FILTEROUTOFREGIONCLOSURE_DOHISTOGRAMCOUNT) { | |
938 gclog_or_tty->print_cr("\nG1 rem-set out-of-region histogram: "); | |
939 gclog_or_tty->print_cr(" # of CS ptrs --> # of cards with that number."); | |
940 out_of_histo.print_on(gclog_or_tty); | |
941 } | |
942 gclog_or_tty->print_cr("\n Concurrent RS processed %d cards in " | |
943 "%5.2fs.", | |
944 _conc_refine_cards, cg1r_thrd->vtime_accum()); | |
945 | |
946 DirtyCardQueueSet& dcqs = JavaThread::dirty_card_queue_set(); | |
947 jint tot_processed_buffers = | |
948 dcqs.processed_buffers_mut() + dcqs.processed_buffers_rs_thread(); | |
949 gclog_or_tty->print_cr(" Of %d completed buffers:", tot_processed_buffers); | |
950 gclog_or_tty->print_cr(" %8d (%5.1f%%) by conc RS thread.", | |
951 dcqs.processed_buffers_rs_thread(), | |
952 100.0*(float)dcqs.processed_buffers_rs_thread()/ | |
953 (float)tot_processed_buffers); | |
954 gclog_or_tty->print_cr(" %8d (%5.1f%%) by mutator threads.", | |
955 dcqs.processed_buffers_mut(), | |
956 100.0*(float)dcqs.processed_buffers_mut()/ | |
957 (float)tot_processed_buffers); | |
958 gclog_or_tty->print_cr(" Did %d concurrent refinement traversals.", | |
959 _conc_refine_traversals); | |
960 if (!G1RSBarrierUseQueue) { | |
961 gclog_or_tty->print_cr(" Scanned %8.2f cards/traversal.", | |
962 _conc_refine_traversals > 0 ? | |
963 (float)_conc_refine_cards/(float)_conc_refine_traversals : | |
964 0); | |
965 } | |
966 gclog_or_tty->print_cr(""); | |
967 if (G1UseHRIntoRS) { | |
968 HRRSStatsIter blk; | |
969 g1->heap_region_iterate(&blk); | |
970 gclog_or_tty->print_cr(" Total heap region rem set sizes = " SIZE_FORMAT "K." | |
971 " Max = " SIZE_FORMAT "K.", | |
972 blk.total_mem_sz()/K, blk.max_mem_sz()/K); | |
973 gclog_or_tty->print_cr(" Static structures = " SIZE_FORMAT "K," | |
974 " free_lists = " SIZE_FORMAT "K.", | |
975 HeapRegionRemSet::static_mem_size()/K, | |
976 HeapRegionRemSet::fl_mem_size()/K); | |
977 gclog_or_tty->print_cr(" %d occupied cards represented.", | |
978 blk.occupied()); | |
979 gclog_or_tty->print_cr(" Max sz region = [" PTR_FORMAT ", " PTR_FORMAT " )" | |
980 " %s, cap = " SIZE_FORMAT "K, occ = " SIZE_FORMAT "K.", | |
981 blk.max_mem_sz_region()->bottom(), blk.max_mem_sz_region()->end(), | |
982 (blk.max_mem_sz_region()->popular() ? "POP" : ""), | |
983 (blk.max_mem_sz_region()->rem_set()->mem_size() + K - 1)/K, | |
984 (blk.max_mem_sz_region()->rem_set()->occupied() + K - 1)/K); | |
985 gclog_or_tty->print_cr(" Did %d coarsenings.", | |
986 HeapRegionRemSet::n_coarsenings()); | |
987 | |
988 } | |
989 } | |
990 void HRInto_G1RemSet::prepare_for_verify() { | |
991 if (G1HRRSFlushLogBuffersOnVerify && VerifyBeforeGC && !_g1->full_collection()) { | |
992 cleanupHRRS(); | |
993 _g1->set_refine_cte_cl_concurrency(false); | |
994 if (SafepointSynchronize::is_at_safepoint()) { | |
995 DirtyCardQueueSet& dcqs = JavaThread::dirty_card_queue_set(); | |
996 dcqs.concatenate_logs(); | |
997 } | |
998 bool cg1r_use_cache = _cg1r->use_cache(); | |
999 _cg1r->set_use_cache(false); | |
1000 updateRS(0); | |
1001 _cg1r->set_use_cache(cg1r_use_cache); | |
1002 } | |
1003 } |