Mercurial > hg > graal-compiler
annotate src/share/vm/gc_implementation/parNew/parNewGeneration.cpp @ 353:9bb2c10ac07b
6723570: G1: assertion failure: p == current_top or oop(p)->is_oop(),"p is not a block start" (revisited!)
Summary: Fixed the incorrect assigment to G1OffsetTableContigSpace::_gc_time_stamp. Also added a little more paranoia to operations on a global timestamp.
Reviewed-by: tonyp
author | iveresov |
---|---|
date | Thu, 10 Jul 2008 09:29:54 -0700 |
parents | feeb96a45707 |
children | d1605aabd0a1 12eea04c8b06 |
rev | line source |
---|---|
0 | 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/_parNewGeneration.cpp.incl" | |
27 | |
28 #ifdef _MSC_VER | |
29 #pragma warning( push ) | |
30 #pragma warning( disable:4355 ) // 'this' : used in base member initializer list | |
31 #endif | |
32 ParScanThreadState::ParScanThreadState(Space* to_space_, | |
33 ParNewGeneration* gen_, | |
34 Generation* old_gen_, | |
35 int thread_num_, | |
36 ObjToScanQueueSet* work_queue_set_, | |
37 size_t desired_plab_sz_, | |
38 ParallelTaskTerminator& term_) : | |
39 _to_space(to_space_), _old_gen(old_gen_), _thread_num(thread_num_), | |
40 _work_queue(work_queue_set_->queue(thread_num_)), _to_space_full(false), | |
41 _ageTable(false), // false ==> not the global age table, no perf data. | |
42 _to_space_alloc_buffer(desired_plab_sz_), | |
43 _to_space_closure(gen_, this), _old_gen_closure(gen_, this), | |
44 _to_space_root_closure(gen_, this), _old_gen_root_closure(gen_, this), | |
45 _older_gen_closure(gen_, this), | |
46 _evacuate_followers(this, &_to_space_closure, &_old_gen_closure, | |
47 &_to_space_root_closure, gen_, &_old_gen_root_closure, | |
48 work_queue_set_, &term_), | |
49 _is_alive_closure(gen_), _scan_weak_ref_closure(gen_, this), | |
50 _keep_alive_closure(&_scan_weak_ref_closure), | |
51 _pushes(0), _pops(0), _steals(0), _steal_attempts(0), _term_attempts(0), | |
52 _strong_roots_time(0.0), _term_time(0.0) | |
53 { | |
54 _survivor_chunk_array = | |
55 (ChunkArray*) old_gen()->get_data_recorder(thread_num()); | |
56 _hash_seed = 17; // Might want to take time-based random value. | |
57 _start = os::elapsedTime(); | |
58 _old_gen_closure.set_generation(old_gen_); | |
59 _old_gen_root_closure.set_generation(old_gen_); | |
60 } | |
61 #ifdef _MSC_VER | |
62 #pragma warning( pop ) | |
63 #endif | |
64 | |
65 void ParScanThreadState::record_survivor_plab(HeapWord* plab_start, | |
66 size_t plab_word_size) { | |
67 ChunkArray* sca = survivor_chunk_array(); | |
68 if (sca != NULL) { | |
69 // A non-null SCA implies that we want the PLAB data recorded. | |
70 sca->record_sample(plab_start, plab_word_size); | |
71 } | |
72 } | |
73 | |
74 bool ParScanThreadState::should_be_partially_scanned(oop new_obj, oop old_obj) const { | |
75 return new_obj->is_objArray() && | |
76 arrayOop(new_obj)->length() > ParGCArrayScanChunk && | |
77 new_obj != old_obj; | |
78 } | |
79 | |
80 void ParScanThreadState::scan_partial_array_and_push_remainder(oop old) { | |
81 assert(old->is_objArray(), "must be obj array"); | |
82 assert(old->is_forwarded(), "must be forwarded"); | |
83 assert(Universe::heap()->is_in_reserved(old), "must be in heap."); | |
84 assert(!_old_gen->is_in(old), "must be in young generation."); | |
85 | |
86 objArrayOop obj = objArrayOop(old->forwardee()); | |
87 // Process ParGCArrayScanChunk elements now | |
88 // and push the remainder back onto queue | |
89 int start = arrayOop(old)->length(); | |
90 int end = obj->length(); | |
91 int remainder = end - start; | |
92 assert(start <= end, "just checking"); | |
93 if (remainder > 2 * ParGCArrayScanChunk) { | |
94 // Test above combines last partial chunk with a full chunk | |
95 end = start + ParGCArrayScanChunk; | |
96 arrayOop(old)->set_length(end); | |
97 // Push remainder. | |
98 bool ok = work_queue()->push(old); | |
99 assert(ok, "just popped, push must be okay"); | |
100 note_push(); | |
101 } else { | |
102 // Restore length so that it can be used if there | |
103 // is a promotion failure and forwarding pointers | |
104 // must be removed. | |
105 arrayOop(old)->set_length(end); | |
106 } | |
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
6
diff
changeset
|
107 |
0 | 108 // process our set of indices (include header in first chunk) |
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
6
diff
changeset
|
109 // should make sure end is even (aligned to HeapWord in case of compressed oops) |
0 | 110 if ((HeapWord *)obj < young_old_boundary()) { |
111 // object is in to_space | |
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
6
diff
changeset
|
112 obj->oop_iterate_range(&_to_space_closure, start, end); |
0 | 113 } else { |
114 // object is in old generation | |
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
6
diff
changeset
|
115 obj->oop_iterate_range(&_old_gen_closure, start, end); |
0 | 116 } |
117 } | |
118 | |
119 | |
120 void ParScanThreadState::trim_queues(int max_size) { | |
121 ObjToScanQueue* queue = work_queue(); | |
122 while (queue->size() > (juint)max_size) { | |
123 oop obj_to_scan; | |
124 if (queue->pop_local(obj_to_scan)) { | |
125 note_pop(); | |
126 | |
127 if ((HeapWord *)obj_to_scan < young_old_boundary()) { | |
128 if (obj_to_scan->is_objArray() && | |
129 obj_to_scan->is_forwarded() && | |
130 obj_to_scan->forwardee() != obj_to_scan) { | |
131 scan_partial_array_and_push_remainder(obj_to_scan); | |
132 } else { | |
133 // object is in to_space | |
134 obj_to_scan->oop_iterate(&_to_space_closure); | |
135 } | |
136 } else { | |
137 // object is in old generation | |
138 obj_to_scan->oop_iterate(&_old_gen_closure); | |
139 } | |
140 } | |
141 } | |
142 } | |
143 | |
144 HeapWord* ParScanThreadState::alloc_in_to_space_slow(size_t word_sz) { | |
145 | |
146 // Otherwise, if the object is small enough, try to reallocate the | |
147 // buffer. | |
148 HeapWord* obj = NULL; | |
149 if (!_to_space_full) { | |
150 ParGCAllocBuffer* const plab = to_space_alloc_buffer(); | |
151 Space* const sp = to_space(); | |
152 if (word_sz * 100 < | |
153 ParallelGCBufferWastePct * plab->word_sz()) { | |
154 // Is small enough; abandon this buffer and start a new one. | |
155 plab->retire(false, false); | |
156 size_t buf_size = plab->word_sz(); | |
157 HeapWord* buf_space = sp->par_allocate(buf_size); | |
158 if (buf_space == NULL) { | |
159 const size_t min_bytes = | |
160 ParGCAllocBuffer::min_size() << LogHeapWordSize; | |
161 size_t free_bytes = sp->free(); | |
162 while(buf_space == NULL && free_bytes >= min_bytes) { | |
163 buf_size = free_bytes >> LogHeapWordSize; | |
164 assert(buf_size == (size_t)align_object_size(buf_size), | |
165 "Invariant"); | |
166 buf_space = sp->par_allocate(buf_size); | |
167 free_bytes = sp->free(); | |
168 } | |
169 } | |
170 if (buf_space != NULL) { | |
171 plab->set_word_size(buf_size); | |
172 plab->set_buf(buf_space); | |
173 record_survivor_plab(buf_space, buf_size); | |
174 obj = plab->allocate(word_sz); | |
175 // Note that we cannot compare buf_size < word_sz below | |
176 // because of AlignmentReserve (see ParGCAllocBuffer::allocate()). | |
177 assert(obj != NULL || plab->words_remaining() < word_sz, | |
178 "Else should have been able to allocate"); | |
179 // It's conceivable that we may be able to use the | |
180 // buffer we just grabbed for subsequent small requests | |
181 // even if not for this one. | |
182 } else { | |
183 // We're used up. | |
184 _to_space_full = true; | |
185 } | |
186 | |
187 } else { | |
188 // Too large; allocate the object individually. | |
189 obj = sp->par_allocate(word_sz); | |
190 } | |
191 } | |
192 return obj; | |
193 } | |
194 | |
195 | |
196 void ParScanThreadState::undo_alloc_in_to_space(HeapWord* obj, | |
197 size_t word_sz) { | |
198 // Is the alloc in the current alloc buffer? | |
199 if (to_space_alloc_buffer()->contains(obj)) { | |
200 assert(to_space_alloc_buffer()->contains(obj + word_sz - 1), | |
201 "Should contain whole object."); | |
202 to_space_alloc_buffer()->undo_allocation(obj, word_sz); | |
203 } else { | |
204 SharedHeap::fill_region_with_object(MemRegion(obj, word_sz)); | |
205 } | |
206 } | |
207 | |
208 class ParScanThreadStateSet: private ResourceArray { | |
209 public: | |
210 // Initializes states for the specified number of threads; | |
211 ParScanThreadStateSet(int num_threads, | |
212 Space& to_space, | |
213 ParNewGeneration& gen, | |
214 Generation& old_gen, | |
215 ObjToScanQueueSet& queue_set, | |
216 size_t desired_plab_sz, | |
217 ParallelTaskTerminator& term); | |
218 inline ParScanThreadState& thread_sate(int i); | |
219 int pushes() { return _pushes; } | |
220 int pops() { return _pops; } | |
221 int steals() { return _steals; } | |
222 void reset(); | |
223 void flush(); | |
224 private: | |
225 ParallelTaskTerminator& _term; | |
226 ParNewGeneration& _gen; | |
227 Generation& _next_gen; | |
228 // staticstics | |
229 int _pushes; | |
230 int _pops; | |
231 int _steals; | |
232 }; | |
233 | |
234 | |
235 ParScanThreadStateSet::ParScanThreadStateSet( | |
236 int num_threads, Space& to_space, ParNewGeneration& gen, | |
237 Generation& old_gen, ObjToScanQueueSet& queue_set, | |
238 size_t desired_plab_sz, ParallelTaskTerminator& term) | |
239 : ResourceArray(sizeof(ParScanThreadState), num_threads), | |
240 _gen(gen), _next_gen(old_gen), _term(term), | |
241 _pushes(0), _pops(0), _steals(0) | |
242 { | |
243 assert(num_threads > 0, "sanity check!"); | |
244 // Initialize states. | |
245 for (int i = 0; i < num_threads; ++i) { | |
246 new ((ParScanThreadState*)_data + i) | |
247 ParScanThreadState(&to_space, &gen, &old_gen, i, &queue_set, | |
248 desired_plab_sz, term); | |
249 } | |
250 } | |
251 | |
252 inline ParScanThreadState& ParScanThreadStateSet::thread_sate(int i) | |
253 { | |
254 assert(i >= 0 && i < length(), "sanity check!"); | |
255 return ((ParScanThreadState*)_data)[i]; | |
256 } | |
257 | |
258 | |
259 void ParScanThreadStateSet::reset() | |
260 { | |
261 _term.reset_for_reuse(); | |
262 } | |
263 | |
264 void ParScanThreadStateSet::flush() | |
265 { | |
266 for (int i = 0; i < length(); ++i) { | |
267 ParScanThreadState& par_scan_state = thread_sate(i); | |
268 | |
269 // Flush stats related to To-space PLAB activity and | |
270 // retire the last buffer. | |
271 par_scan_state.to_space_alloc_buffer()-> | |
272 flush_stats_and_retire(_gen.plab_stats(), | |
273 false /* !retain */); | |
274 | |
275 // Every thread has its own age table. We need to merge | |
276 // them all into one. | |
277 ageTable *local_table = par_scan_state.age_table(); | |
278 _gen.age_table()->merge(local_table); | |
279 | |
280 // Inform old gen that we're done. | |
281 _next_gen.par_promote_alloc_done(i); | |
282 _next_gen.par_oop_since_save_marks_iterate_done(i); | |
283 | |
284 // Flush stats related to work queue activity (push/pop/steal) | |
285 // This could conceivably become a bottleneck; if so, we'll put the | |
286 // stat's gathering under the flag. | |
287 if (PAR_STATS_ENABLED) { | |
288 _pushes += par_scan_state.pushes(); | |
289 _pops += par_scan_state.pops(); | |
290 _steals += par_scan_state.steals(); | |
291 if (ParallelGCVerbose) { | |
292 gclog_or_tty->print("Thread %d complete:\n" | |
293 " Pushes: %7d Pops: %7d Steals %7d (in %d attempts)\n", | |
294 i, par_scan_state.pushes(), par_scan_state.pops(), | |
295 par_scan_state.steals(), par_scan_state.steal_attempts()); | |
296 if (par_scan_state.overflow_pushes() > 0 || | |
297 par_scan_state.overflow_refills() > 0) { | |
298 gclog_or_tty->print(" Overflow pushes: %7d " | |
299 "Overflow refills: %7d for %d objs.\n", | |
300 par_scan_state.overflow_pushes(), | |
301 par_scan_state.overflow_refills(), | |
302 par_scan_state.overflow_refill_objs()); | |
303 } | |
304 | |
305 double elapsed = par_scan_state.elapsed(); | |
306 double strong_roots = par_scan_state.strong_roots_time(); | |
307 double term = par_scan_state.term_time(); | |
308 gclog_or_tty->print( | |
309 " Elapsed: %7.2f ms.\n" | |
310 " Strong roots: %7.2f ms (%6.2f%%)\n" | |
311 " Termination: %7.2f ms (%6.2f%%) (in %d entries)\n", | |
312 elapsed * 1000.0, | |
313 strong_roots * 1000.0, (strong_roots*100.0/elapsed), | |
314 term * 1000.0, (term*100.0/elapsed), | |
315 par_scan_state.term_attempts()); | |
316 } | |
317 } | |
318 } | |
319 } | |
320 | |
321 ParScanClosure::ParScanClosure(ParNewGeneration* g, | |
322 ParScanThreadState* par_scan_state) : | |
323 OopsInGenClosure(g), _par_scan_state(par_scan_state), _g(g) | |
324 { | |
325 assert(_g->level() == 0, "Optimized for youngest generation"); | |
326 _boundary = _g->reserved().end(); | |
327 } | |
328 | |
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
6
diff
changeset
|
329 void ParScanWithBarrierClosure::do_oop(oop* p) { ParScanClosure::do_oop_work(p, true, false); } |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
6
diff
changeset
|
330 void ParScanWithBarrierClosure::do_oop(narrowOop* p) { ParScanClosure::do_oop_work(p, true, false); } |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
6
diff
changeset
|
331 |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
6
diff
changeset
|
332 void ParScanWithoutBarrierClosure::do_oop(oop* p) { ParScanClosure::do_oop_work(p, false, false); } |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
6
diff
changeset
|
333 void ParScanWithoutBarrierClosure::do_oop(narrowOop* p) { ParScanClosure::do_oop_work(p, false, false); } |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
6
diff
changeset
|
334 |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
6
diff
changeset
|
335 void ParRootScanWithBarrierTwoGensClosure::do_oop(oop* p) { ParScanClosure::do_oop_work(p, true, true); } |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
6
diff
changeset
|
336 void ParRootScanWithBarrierTwoGensClosure::do_oop(narrowOop* p) { ParScanClosure::do_oop_work(p, true, true); } |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
6
diff
changeset
|
337 |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
6
diff
changeset
|
338 void ParRootScanWithoutBarrierClosure::do_oop(oop* p) { ParScanClosure::do_oop_work(p, false, true); } |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
6
diff
changeset
|
339 void ParRootScanWithoutBarrierClosure::do_oop(narrowOop* p) { ParScanClosure::do_oop_work(p, false, true); } |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
6
diff
changeset
|
340 |
0 | 341 ParScanWeakRefClosure::ParScanWeakRefClosure(ParNewGeneration* g, |
342 ParScanThreadState* par_scan_state) | |
343 : ScanWeakRefClosure(g), _par_scan_state(par_scan_state) | |
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
6
diff
changeset
|
344 {} |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
6
diff
changeset
|
345 |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
6
diff
changeset
|
346 void ParScanWeakRefClosure::do_oop(oop* p) { ParScanWeakRefClosure::do_oop_work(p); } |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
6
diff
changeset
|
347 void ParScanWeakRefClosure::do_oop(narrowOop* p) { ParScanWeakRefClosure::do_oop_work(p); } |
0 | 348 |
349 #ifdef WIN32 | |
350 #pragma warning(disable: 4786) /* identifier was truncated to '255' characters in the browser information */ | |
351 #endif | |
352 | |
353 ParEvacuateFollowersClosure::ParEvacuateFollowersClosure( | |
354 ParScanThreadState* par_scan_state_, | |
355 ParScanWithoutBarrierClosure* to_space_closure_, | |
356 ParScanWithBarrierClosure* old_gen_closure_, | |
357 ParRootScanWithoutBarrierClosure* to_space_root_closure_, | |
358 ParNewGeneration* par_gen_, | |
359 ParRootScanWithBarrierTwoGensClosure* old_gen_root_closure_, | |
360 ObjToScanQueueSet* task_queues_, | |
361 ParallelTaskTerminator* terminator_) : | |
362 | |
363 _par_scan_state(par_scan_state_), | |
364 _to_space_closure(to_space_closure_), | |
365 _old_gen_closure(old_gen_closure_), | |
366 _to_space_root_closure(to_space_root_closure_), | |
367 _old_gen_root_closure(old_gen_root_closure_), | |
368 _par_gen(par_gen_), | |
369 _task_queues(task_queues_), | |
370 _terminator(terminator_) | |
371 {} | |
372 | |
373 void ParEvacuateFollowersClosure::do_void() { | |
374 ObjToScanQueue* work_q = par_scan_state()->work_queue(); | |
375 | |
376 while (true) { | |
377 | |
378 // Scan to-space and old-gen objs until we run out of both. | |
379 oop obj_to_scan; | |
380 par_scan_state()->trim_queues(0); | |
381 | |
382 // We have no local work, attempt to steal from other threads. | |
383 | |
384 // attempt to steal work from promoted. | |
385 par_scan_state()->note_steal_attempt(); | |
386 if (task_queues()->steal(par_scan_state()->thread_num(), | |
387 par_scan_state()->hash_seed(), | |
388 obj_to_scan)) { | |
389 par_scan_state()->note_steal(); | |
390 bool res = work_q->push(obj_to_scan); | |
391 assert(res, "Empty queue should have room for a push."); | |
392 | |
393 par_scan_state()->note_push(); | |
394 // if successful, goto Start. | |
395 continue; | |
396 | |
397 // try global overflow list. | |
398 } else if (par_gen()->take_from_overflow_list(par_scan_state())) { | |
399 continue; | |
400 } | |
401 | |
402 // Otherwise, offer termination. | |
403 par_scan_state()->start_term_time(); | |
404 if (terminator()->offer_termination()) break; | |
405 par_scan_state()->end_term_time(); | |
406 } | |
407 // Finish the last termination pause. | |
408 par_scan_state()->end_term_time(); | |
409 } | |
410 | |
411 ParNewGenTask::ParNewGenTask(ParNewGeneration* gen, Generation* next_gen, | |
412 HeapWord* young_old_boundary, ParScanThreadStateSet* state_set) : | |
413 AbstractGangTask("ParNewGeneration collection"), | |
414 _gen(gen), _next_gen(next_gen), | |
415 _young_old_boundary(young_old_boundary), | |
416 _state_set(state_set) | |
417 {} | |
418 | |
419 void ParNewGenTask::work(int i) { | |
420 GenCollectedHeap* gch = GenCollectedHeap::heap(); | |
421 // Since this is being done in a separate thread, need new resource | |
422 // and handle marks. | |
423 ResourceMark rm; | |
424 HandleMark hm; | |
425 // We would need multiple old-gen queues otherwise. | |
426 guarantee(gch->n_gens() == 2, | |
427 "Par young collection currently only works with one older gen."); | |
428 | |
429 Generation* old_gen = gch->next_gen(_gen); | |
430 | |
431 ParScanThreadState& par_scan_state = _state_set->thread_sate(i); | |
432 par_scan_state.set_young_old_boundary(_young_old_boundary); | |
433 | |
434 par_scan_state.start_strong_roots(); | |
435 gch->gen_process_strong_roots(_gen->level(), | |
436 true, // Process younger gens, if any, | |
437 // as strong roots. | |
438 false,// not collecting perm generation. | |
439 SharedHeap::SO_AllClasses, | |
440 &par_scan_state.older_gen_closure(), | |
441 &par_scan_state.to_space_root_closure()); | |
442 par_scan_state.end_strong_roots(); | |
443 | |
444 // "evacuate followers". | |
445 par_scan_state.evacuate_followers_closure().do_void(); | |
446 } | |
447 | |
448 #ifdef _MSC_VER | |
449 #pragma warning( push ) | |
450 #pragma warning( disable:4355 ) // 'this' : used in base member initializer list | |
451 #endif | |
452 ParNewGeneration:: | |
453 ParNewGeneration(ReservedSpace rs, size_t initial_byte_size, int level) | |
454 : DefNewGeneration(rs, initial_byte_size, level, "PCopy"), | |
455 _overflow_list(NULL), | |
456 _is_alive_closure(this), | |
457 _plab_stats(YoungPLABSize, PLABWeight) | |
458 { | |
459 _task_queues = new ObjToScanQueueSet(ParallelGCThreads); | |
460 guarantee(_task_queues != NULL, "task_queues allocation failure."); | |
461 | |
462 for (uint i1 = 0; i1 < ParallelGCThreads; i1++) { | |
463 ObjToScanQueuePadded *q_padded = new ObjToScanQueuePadded(); | |
464 guarantee(q_padded != NULL, "work_queue Allocation failure."); | |
465 | |
466 _task_queues->register_queue(i1, &q_padded->work_queue); | |
467 } | |
468 | |
469 for (uint i2 = 0; i2 < ParallelGCThreads; i2++) | |
470 _task_queues->queue(i2)->initialize(); | |
471 | |
472 if (UsePerfData) { | |
473 EXCEPTION_MARK; | |
474 ResourceMark rm; | |
475 | |
476 const char* cname = | |
477 PerfDataManager::counter_name(_gen_counters->name_space(), "threads"); | |
478 PerfDataManager::create_constant(SUN_GC, cname, PerfData::U_None, | |
479 ParallelGCThreads, CHECK); | |
480 } | |
481 } | |
482 #ifdef _MSC_VER | |
483 #pragma warning( pop ) | |
484 #endif | |
485 | |
486 // ParNewGeneration:: | |
487 ParKeepAliveClosure::ParKeepAliveClosure(ParScanWeakRefClosure* cl) : | |
488 DefNewGeneration::KeepAliveClosure(cl), _par_cl(cl) {} | |
489 | |
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
6
diff
changeset
|
490 template <class T> |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
6
diff
changeset
|
491 void /*ParNewGeneration::*/ParKeepAliveClosure::do_oop_work(T* p) { |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
6
diff
changeset
|
492 #ifdef ASSERT |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
6
diff
changeset
|
493 { |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
6
diff
changeset
|
494 assert(!oopDesc::is_null(*p), "expected non-null ref"); |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
6
diff
changeset
|
495 oop obj = oopDesc::load_decode_heap_oop_not_null(p); |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
6
diff
changeset
|
496 // We never expect to see a null reference being processed |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
6
diff
changeset
|
497 // as a weak reference. |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
6
diff
changeset
|
498 assert(obj->is_oop(), "expected an oop while scanning weak refs"); |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
6
diff
changeset
|
499 } |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
6
diff
changeset
|
500 #endif // ASSERT |
0 | 501 |
502 _par_cl->do_oop_nv(p); | |
503 | |
504 if (Universe::heap()->is_in_reserved(p)) { | |
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
6
diff
changeset
|
505 oop obj = oopDesc::load_decode_heap_oop_not_null(p); |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
6
diff
changeset
|
506 _rs->write_ref_field_gc_par(p, obj); |
0 | 507 } |
508 } | |
509 | |
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
6
diff
changeset
|
510 void /*ParNewGeneration::*/ParKeepAliveClosure::do_oop(oop* p) { ParKeepAliveClosure::do_oop_work(p); } |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
6
diff
changeset
|
511 void /*ParNewGeneration::*/ParKeepAliveClosure::do_oop(narrowOop* p) { ParKeepAliveClosure::do_oop_work(p); } |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
6
diff
changeset
|
512 |
0 | 513 // ParNewGeneration:: |
514 KeepAliveClosure::KeepAliveClosure(ScanWeakRefClosure* cl) : | |
515 DefNewGeneration::KeepAliveClosure(cl) {} | |
516 | |
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
6
diff
changeset
|
517 template <class T> |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
6
diff
changeset
|
518 void /*ParNewGeneration::*/KeepAliveClosure::do_oop_work(T* p) { |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
6
diff
changeset
|
519 #ifdef ASSERT |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
6
diff
changeset
|
520 { |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
6
diff
changeset
|
521 assert(!oopDesc::is_null(*p), "expected non-null ref"); |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
6
diff
changeset
|
522 oop obj = oopDesc::load_decode_heap_oop_not_null(p); |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
6
diff
changeset
|
523 // We never expect to see a null reference being processed |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
6
diff
changeset
|
524 // as a weak reference. |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
6
diff
changeset
|
525 assert(obj->is_oop(), "expected an oop while scanning weak refs"); |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
6
diff
changeset
|
526 } |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
6
diff
changeset
|
527 #endif // ASSERT |
0 | 528 |
529 _cl->do_oop_nv(p); | |
530 | |
531 if (Universe::heap()->is_in_reserved(p)) { | |
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
6
diff
changeset
|
532 oop obj = oopDesc::load_decode_heap_oop_not_null(p); |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
6
diff
changeset
|
533 _rs->write_ref_field_gc_par(p, obj); |
0 | 534 } |
535 } | |
536 | |
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
6
diff
changeset
|
537 void /*ParNewGeneration::*/KeepAliveClosure::do_oop(oop* p) { KeepAliveClosure::do_oop_work(p); } |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
6
diff
changeset
|
538 void /*ParNewGeneration::*/KeepAliveClosure::do_oop(narrowOop* p) { KeepAliveClosure::do_oop_work(p); } |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
6
diff
changeset
|
539 |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
6
diff
changeset
|
540 template <class T> void ScanClosureWithParBarrier::do_oop_work(T* p) { |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
6
diff
changeset
|
541 T heap_oop = oopDesc::load_heap_oop(p); |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
6
diff
changeset
|
542 if (!oopDesc::is_null(heap_oop)) { |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
6
diff
changeset
|
543 oop obj = oopDesc::decode_heap_oop_not_null(heap_oop); |
0 | 544 if ((HeapWord*)obj < _boundary) { |
545 assert(!_g->to()->is_in_reserved(obj), "Scanning field twice?"); | |
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
6
diff
changeset
|
546 oop new_obj = obj->is_forwarded() |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
6
diff
changeset
|
547 ? obj->forwardee() |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
6
diff
changeset
|
548 : _g->DefNewGeneration::copy_to_survivor_space(obj); |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
6
diff
changeset
|
549 oopDesc::encode_store_heap_oop_not_null(p, new_obj); |
0 | 550 } |
551 if (_gc_barrier) { | |
552 // If p points to a younger generation, mark the card. | |
553 if ((HeapWord*)obj < _gen_boundary) { | |
554 _rs->write_ref_field_gc_par(p, obj); | |
555 } | |
556 } | |
557 } | |
558 } | |
559 | |
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
6
diff
changeset
|
560 void ScanClosureWithParBarrier::do_oop(oop* p) { ScanClosureWithParBarrier::do_oop_work(p); } |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
6
diff
changeset
|
561 void ScanClosureWithParBarrier::do_oop(narrowOop* p) { ScanClosureWithParBarrier::do_oop_work(p); } |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
6
diff
changeset
|
562 |
0 | 563 class ParNewRefProcTaskProxy: public AbstractGangTask { |
564 typedef AbstractRefProcTaskExecutor::ProcessTask ProcessTask; | |
565 public: | |
566 ParNewRefProcTaskProxy(ProcessTask& task, ParNewGeneration& gen, | |
567 Generation& next_gen, | |
568 HeapWord* young_old_boundary, | |
569 ParScanThreadStateSet& state_set); | |
570 | |
571 private: | |
572 virtual void work(int i); | |
573 | |
574 private: | |
575 ParNewGeneration& _gen; | |
576 ProcessTask& _task; | |
577 Generation& _next_gen; | |
578 HeapWord* _young_old_boundary; | |
579 ParScanThreadStateSet& _state_set; | |
580 }; | |
581 | |
582 ParNewRefProcTaskProxy::ParNewRefProcTaskProxy( | |
583 ProcessTask& task, ParNewGeneration& gen, | |
584 Generation& next_gen, | |
585 HeapWord* young_old_boundary, | |
586 ParScanThreadStateSet& state_set) | |
587 : AbstractGangTask("ParNewGeneration parallel reference processing"), | |
588 _gen(gen), | |
589 _task(task), | |
590 _next_gen(next_gen), | |
591 _young_old_boundary(young_old_boundary), | |
592 _state_set(state_set) | |
593 { | |
594 } | |
595 | |
596 void ParNewRefProcTaskProxy::work(int i) | |
597 { | |
598 ResourceMark rm; | |
599 HandleMark hm; | |
600 ParScanThreadState& par_scan_state = _state_set.thread_sate(i); | |
601 par_scan_state.set_young_old_boundary(_young_old_boundary); | |
602 _task.work(i, par_scan_state.is_alive_closure(), | |
603 par_scan_state.keep_alive_closure(), | |
604 par_scan_state.evacuate_followers_closure()); | |
605 } | |
606 | |
607 class ParNewRefEnqueueTaskProxy: public AbstractGangTask { | |
608 typedef AbstractRefProcTaskExecutor::EnqueueTask EnqueueTask; | |
609 EnqueueTask& _task; | |
610 | |
611 public: | |
612 ParNewRefEnqueueTaskProxy(EnqueueTask& task) | |
613 : AbstractGangTask("ParNewGeneration parallel reference enqueue"), | |
614 _task(task) | |
615 { } | |
616 | |
617 virtual void work(int i) | |
618 { | |
619 _task.work(i); | |
620 } | |
621 }; | |
622 | |
623 | |
624 void ParNewRefProcTaskExecutor::execute(ProcessTask& task) | |
625 { | |
626 GenCollectedHeap* gch = GenCollectedHeap::heap(); | |
627 assert(gch->kind() == CollectedHeap::GenCollectedHeap, | |
628 "not a generational heap"); | |
629 WorkGang* workers = gch->workers(); | |
630 assert(workers != NULL, "Need parallel worker threads."); | |
631 ParNewRefProcTaskProxy rp_task(task, _generation, *_generation.next_gen(), | |
632 _generation.reserved().end(), _state_set); | |
633 workers->run_task(&rp_task); | |
634 _state_set.reset(); | |
635 } | |
636 | |
637 void ParNewRefProcTaskExecutor::execute(EnqueueTask& task) | |
638 { | |
639 GenCollectedHeap* gch = GenCollectedHeap::heap(); | |
640 WorkGang* workers = gch->workers(); | |
641 assert(workers != NULL, "Need parallel worker threads."); | |
642 ParNewRefEnqueueTaskProxy enq_task(task); | |
643 workers->run_task(&enq_task); | |
644 } | |
645 | |
646 void ParNewRefProcTaskExecutor::set_single_threaded_mode() | |
647 { | |
648 _state_set.flush(); | |
649 GenCollectedHeap* gch = GenCollectedHeap::heap(); | |
650 gch->set_par_threads(0); // 0 ==> non-parallel. | |
651 gch->save_marks(); | |
652 } | |
653 | |
654 ScanClosureWithParBarrier:: | |
655 ScanClosureWithParBarrier(ParNewGeneration* g, bool gc_barrier) : | |
656 ScanClosure(g, gc_barrier) {} | |
657 | |
658 EvacuateFollowersClosureGeneral:: | |
659 EvacuateFollowersClosureGeneral(GenCollectedHeap* gch, int level, | |
660 OopsInGenClosure* cur, | |
661 OopsInGenClosure* older) : | |
662 _gch(gch), _level(level), | |
663 _scan_cur_or_nonheap(cur), _scan_older(older) | |
664 {} | |
665 | |
666 void EvacuateFollowersClosureGeneral::do_void() { | |
667 do { | |
668 // Beware: this call will lead to closure applications via virtual | |
669 // calls. | |
670 _gch->oop_since_save_marks_iterate(_level, | |
671 _scan_cur_or_nonheap, | |
672 _scan_older); | |
673 } while (!_gch->no_allocs_since_save_marks(_level)); | |
674 } | |
675 | |
676 | |
677 bool ParNewGeneration::_avoid_promotion_undo = false; | |
678 | |
679 void ParNewGeneration::adjust_desired_tenuring_threshold() { | |
680 // Set the desired survivor size to half the real survivor space | |
681 _tenuring_threshold = | |
682 age_table()->compute_tenuring_threshold(to()->capacity()/HeapWordSize); | |
683 } | |
684 | |
685 // A Generation that does parallel young-gen collection. | |
686 | |
687 void ParNewGeneration::collect(bool full, | |
688 bool clear_all_soft_refs, | |
689 size_t size, | |
690 bool is_tlab) { | |
691 assert(full || size > 0, "otherwise we don't want to collect"); | |
692 GenCollectedHeap* gch = GenCollectedHeap::heap(); | |
693 assert(gch->kind() == CollectedHeap::GenCollectedHeap, | |
694 "not a CMS generational heap"); | |
695 AdaptiveSizePolicy* size_policy = gch->gen_policy()->size_policy(); | |
696 WorkGang* workers = gch->workers(); | |
697 _next_gen = gch->next_gen(this); | |
698 assert(_next_gen != NULL, | |
699 "This must be the youngest gen, and not the only gen"); | |
700 assert(gch->n_gens() == 2, | |
701 "Par collection currently only works with single older gen."); | |
702 // Do we have to avoid promotion_undo? | |
703 if (gch->collector_policy()->is_concurrent_mark_sweep_policy()) { | |
704 set_avoid_promotion_undo(true); | |
705 } | |
706 | |
707 // If the next generation is too full to accomodate worst-case promotion | |
708 // from this generation, pass on collection; let the next generation | |
709 // do it. | |
710 if (!collection_attempt_is_safe()) { | |
711 gch->set_incremental_collection_will_fail(); | |
712 return; | |
713 } | |
714 assert(to()->is_empty(), "Else not collection_attempt_is_safe"); | |
715 | |
716 init_assuming_no_promotion_failure(); | |
717 | |
718 if (UseAdaptiveSizePolicy) { | |
719 set_survivor_overflow(false); | |
720 size_policy->minor_collection_begin(); | |
721 } | |
722 | |
723 TraceTime t1("GC", PrintGC && !PrintGCDetails, true, gclog_or_tty); | |
724 // Capture heap used before collection (for printing). | |
725 size_t gch_prev_used = gch->used(); | |
726 | |
727 SpecializationStats::clear(); | |
728 | |
729 age_table()->clear(); | |
730 to()->clear(); | |
731 | |
732 gch->save_marks(); | |
733 assert(workers != NULL, "Need parallel worker threads."); | |
734 ParallelTaskTerminator _term(workers->total_workers(), task_queues()); | |
735 ParScanThreadStateSet thread_state_set(workers->total_workers(), | |
736 *to(), *this, *_next_gen, *task_queues(), | |
737 desired_plab_sz(), _term); | |
738 | |
739 ParNewGenTask tsk(this, _next_gen, reserved().end(), &thread_state_set); | |
740 int n_workers = workers->total_workers(); | |
741 gch->set_par_threads(n_workers); | |
742 gch->change_strong_roots_parity(); | |
743 gch->rem_set()->prepare_for_younger_refs_iterate(true); | |
744 // It turns out that even when we're using 1 thread, doing the work in a | |
745 // separate thread causes wide variance in run times. We can't help this | |
746 // in the multi-threaded case, but we special-case n=1 here to get | |
747 // repeatable measurements of the 1-thread overhead of the parallel code. | |
748 if (n_workers > 1) { | |
749 workers->run_task(&tsk); | |
750 } else { | |
751 tsk.work(0); | |
752 } | |
753 thread_state_set.reset(); | |
754 | |
755 if (PAR_STATS_ENABLED && ParallelGCVerbose) { | |
756 gclog_or_tty->print("Thread totals:\n" | |
757 " Pushes: %7d Pops: %7d Steals %7d (sum = %7d).\n", | |
758 thread_state_set.pushes(), thread_state_set.pops(), | |
759 thread_state_set.steals(), | |
760 thread_state_set.pops()+thread_state_set.steals()); | |
761 } | |
762 assert(thread_state_set.pushes() == thread_state_set.pops() + thread_state_set.steals(), | |
763 "Or else the queues are leaky."); | |
764 | |
765 // For now, process discovered weak refs sequentially. | |
766 #ifdef COMPILER2 | |
767 ReferencePolicy *soft_ref_policy = new LRUMaxHeapPolicy(); | |
768 #else | |
769 ReferencePolicy *soft_ref_policy = new LRUCurrentHeapPolicy(); | |
770 #endif // COMPILER2 | |
771 | |
772 // Process (weak) reference objects found during scavenge. | |
773 IsAliveClosure is_alive(this); | |
774 ScanWeakRefClosure scan_weak_ref(this); | |
775 KeepAliveClosure keep_alive(&scan_weak_ref); | |
776 ScanClosure scan_without_gc_barrier(this, false); | |
777 ScanClosureWithParBarrier scan_with_gc_barrier(this, true); | |
778 set_promo_failure_scan_stack_closure(&scan_without_gc_barrier); | |
779 EvacuateFollowersClosureGeneral evacuate_followers(gch, _level, | |
780 &scan_without_gc_barrier, &scan_with_gc_barrier); | |
781 if (ref_processor()->processing_is_mt()) { | |
782 ParNewRefProcTaskExecutor task_executor(*this, thread_state_set); | |
783 ref_processor()->process_discovered_references( | |
784 soft_ref_policy, &is_alive, &keep_alive, &evacuate_followers, | |
785 &task_executor); | |
786 } else { | |
787 thread_state_set.flush(); | |
788 gch->set_par_threads(0); // 0 ==> non-parallel. | |
789 gch->save_marks(); | |
790 ref_processor()->process_discovered_references( | |
791 soft_ref_policy, &is_alive, &keep_alive, &evacuate_followers, | |
792 NULL); | |
793 } | |
794 if (!promotion_failed()) { | |
795 // Swap the survivor spaces. | |
796 eden()->clear(); | |
797 from()->clear(); | |
798 swap_spaces(); | |
799 | |
800 assert(to()->is_empty(), "to space should be empty now"); | |
801 } else { | |
802 assert(HandlePromotionFailure, | |
803 "Should only be here if promotion failure handling is on"); | |
804 if (_promo_failure_scan_stack != NULL) { | |
805 // Can be non-null because of reference processing. | |
806 // Free stack with its elements. | |
807 delete _promo_failure_scan_stack; | |
808 _promo_failure_scan_stack = NULL; | |
809 } | |
810 remove_forwarding_pointers(); | |
811 if (PrintGCDetails) { | |
812 gclog_or_tty->print(" (promotion failed)"); | |
813 } | |
814 // All the spaces are in play for mark-sweep. | |
815 swap_spaces(); // Make life simpler for CMS || rescan; see 6483690. | |
816 from()->set_next_compaction_space(to()); | |
817 gch->set_incremental_collection_will_fail(); | |
6
73e96e5c30df
6624765: Guarantee failure "Unexpected dirty card found"
jmasa
parents:
0
diff
changeset
|
818 |
73e96e5c30df
6624765: Guarantee failure "Unexpected dirty card found"
jmasa
parents:
0
diff
changeset
|
819 // Reset the PromotionFailureALot counters. |
73e96e5c30df
6624765: Guarantee failure "Unexpected dirty card found"
jmasa
parents:
0
diff
changeset
|
820 NOT_PRODUCT(Universe::heap()->reset_promotion_should_fail();) |
0 | 821 } |
822 // set new iteration safe limit for the survivor spaces | |
823 from()->set_concurrent_iteration_safe_limit(from()->top()); | |
824 to()->set_concurrent_iteration_safe_limit(to()->top()); | |
825 | |
826 adjust_desired_tenuring_threshold(); | |
827 if (ResizePLAB) { | |
828 plab_stats()->adjust_desired_plab_sz(); | |
829 } | |
830 | |
831 if (PrintGC && !PrintGCDetails) { | |
832 gch->print_heap_change(gch_prev_used); | |
833 } | |
834 | |
835 if (UseAdaptiveSizePolicy) { | |
836 size_policy->minor_collection_end(gch->gc_cause()); | |
837 size_policy->avg_survived()->sample(from()->used()); | |
838 } | |
839 | |
840 update_time_of_last_gc(os::javaTimeMillis()); | |
841 | |
842 SpecializationStats::print(); | |
843 | |
844 ref_processor()->set_enqueuing_is_done(true); | |
845 if (ref_processor()->processing_is_mt()) { | |
846 ParNewRefProcTaskExecutor task_executor(*this, thread_state_set); | |
847 ref_processor()->enqueue_discovered_references(&task_executor); | |
848 } else { | |
849 ref_processor()->enqueue_discovered_references(NULL); | |
850 } | |
851 ref_processor()->verify_no_references_recorded(); | |
852 } | |
853 | |
854 static int sum; | |
855 void ParNewGeneration::waste_some_time() { | |
856 for (int i = 0; i < 100; i++) { | |
857 sum += i; | |
858 } | |
859 } | |
860 | |
861 static const oop ClaimedForwardPtr = oop(0x4); | |
862 | |
863 // Because of concurrency, there are times where an object for which | |
864 // "is_forwarded()" is true contains an "interim" forwarding pointer | |
865 // value. Such a value will soon be overwritten with a real value. | |
866 // This method requires "obj" to have a forwarding pointer, and waits, if | |
867 // necessary for a real one to be inserted, and returns it. | |
868 | |
869 oop ParNewGeneration::real_forwardee(oop obj) { | |
870 oop forward_ptr = obj->forwardee(); | |
871 if (forward_ptr != ClaimedForwardPtr) { | |
872 return forward_ptr; | |
873 } else { | |
874 return real_forwardee_slow(obj); | |
875 } | |
876 } | |
877 | |
878 oop ParNewGeneration::real_forwardee_slow(oop obj) { | |
879 // Spin-read if it is claimed but not yet written by another thread. | |
880 oop forward_ptr = obj->forwardee(); | |
881 while (forward_ptr == ClaimedForwardPtr) { | |
882 waste_some_time(); | |
883 assert(obj->is_forwarded(), "precondition"); | |
884 forward_ptr = obj->forwardee(); | |
885 } | |
886 return forward_ptr; | |
887 } | |
888 | |
889 #ifdef ASSERT | |
890 bool ParNewGeneration::is_legal_forward_ptr(oop p) { | |
891 return | |
892 (_avoid_promotion_undo && p == ClaimedForwardPtr) | |
893 || Universe::heap()->is_in_reserved(p); | |
894 } | |
895 #endif | |
896 | |
897 void ParNewGeneration::preserve_mark_if_necessary(oop obj, markOop m) { | |
898 if ((m != markOopDesc::prototype()) && | |
899 (!UseBiasedLocking || (m != markOopDesc::biased_locking_prototype()))) { | |
900 MutexLocker ml(ParGCRareEvent_lock); | |
901 DefNewGeneration::preserve_mark_if_necessary(obj, m); | |
902 } | |
903 } | |
904 | |
905 // Multiple GC threads may try to promote an object. If the object | |
906 // is successfully promoted, a forwarding pointer will be installed in | |
907 // the object in the young generation. This method claims the right | |
908 // to install the forwarding pointer before it copies the object, | |
909 // thus avoiding the need to undo the copy as in | |
910 // copy_to_survivor_space_avoiding_with_undo. | |
911 | |
912 oop ParNewGeneration::copy_to_survivor_space_avoiding_promotion_undo( | |
913 ParScanThreadState* par_scan_state, oop old, size_t sz, markOop m) { | |
914 // In the sequential version, this assert also says that the object is | |
915 // not forwarded. That might not be the case here. It is the case that | |
916 // the caller observed it to be not forwarded at some time in the past. | |
917 assert(is_in_reserved(old), "shouldn't be scavenging this oop"); | |
918 | |
919 // The sequential code read "old->age()" below. That doesn't work here, | |
920 // since the age is in the mark word, and that might be overwritten with | |
921 // a forwarding pointer by a parallel thread. So we must save the mark | |
922 // word in a local and then analyze it. | |
923 oopDesc dummyOld; | |
924 dummyOld.set_mark(m); | |
925 assert(!dummyOld.is_forwarded(), | |
926 "should not be called with forwarding pointer mark word."); | |
927 | |
928 oop new_obj = NULL; | |
929 oop forward_ptr; | |
930 | |
931 // Try allocating obj in to-space (unless too old) | |
932 if (dummyOld.age() < tenuring_threshold()) { | |
933 new_obj = (oop)par_scan_state->alloc_in_to_space(sz); | |
934 if (new_obj == NULL) { | |
935 set_survivor_overflow(true); | |
936 } | |
937 } | |
938 | |
939 if (new_obj == NULL) { | |
940 // Either to-space is full or we decided to promote | |
941 // try allocating obj tenured | |
942 | |
943 // Attempt to install a null forwarding pointer (atomically), | |
944 // to claim the right to install the real forwarding pointer. | |
945 forward_ptr = old->forward_to_atomic(ClaimedForwardPtr); | |
946 if (forward_ptr != NULL) { | |
947 // someone else beat us to it. | |
948 return real_forwardee(old); | |
949 } | |
950 | |
951 new_obj = _next_gen->par_promote(par_scan_state->thread_num(), | |
952 old, m, sz); | |
953 | |
954 if (new_obj == NULL) { | |
955 if (!HandlePromotionFailure) { | |
956 // A failed promotion likely means the MaxLiveObjectEvacuationRatio flag | |
957 // is incorrectly set. In any case, its seriously wrong to be here! | |
958 vm_exit_out_of_memory(sz*wordSize, "promotion"); | |
959 } | |
960 // promotion failed, forward to self | |
961 _promotion_failed = true; | |
962 new_obj = old; | |
963 | |
964 preserve_mark_if_necessary(old, m); | |
965 } | |
966 | |
967 old->forward_to(new_obj); | |
968 forward_ptr = NULL; | |
969 } else { | |
970 // Is in to-space; do copying ourselves. | |
971 Copy::aligned_disjoint_words((HeapWord*)old, (HeapWord*)new_obj, sz); | |
972 forward_ptr = old->forward_to_atomic(new_obj); | |
973 // Restore the mark word copied above. | |
974 new_obj->set_mark(m); | |
975 // Increment age if obj still in new generation | |
976 new_obj->incr_age(); | |
977 par_scan_state->age_table()->add(new_obj, sz); | |
978 } | |
979 assert(new_obj != NULL, "just checking"); | |
980 | |
981 if (forward_ptr == NULL) { | |
982 oop obj_to_push = new_obj; | |
983 if (par_scan_state->should_be_partially_scanned(obj_to_push, old)) { | |
984 // Length field used as index of next element to be scanned. | |
985 // Real length can be obtained from real_forwardee() | |
986 arrayOop(old)->set_length(0); | |
987 obj_to_push = old; | |
988 assert(obj_to_push->is_forwarded() && obj_to_push->forwardee() != obj_to_push, | |
989 "push forwarded object"); | |
990 } | |
991 // Push it on one of the queues of to-be-scanned objects. | |
992 if (!par_scan_state->work_queue()->push(obj_to_push)) { | |
993 // Add stats for overflow pushes. | |
994 if (Verbose && PrintGCDetails) { | |
995 gclog_or_tty->print("queue overflow!\n"); | |
996 } | |
997 push_on_overflow_list(old); | |
998 par_scan_state->note_overflow_push(); | |
999 } | |
1000 par_scan_state->note_push(); | |
1001 | |
1002 return new_obj; | |
1003 } | |
1004 | |
1005 // Oops. Someone beat us to it. Undo the allocation. Where did we | |
1006 // allocate it? | |
1007 if (is_in_reserved(new_obj)) { | |
1008 // Must be in to_space. | |
1009 assert(to()->is_in_reserved(new_obj), "Checking"); | |
1010 if (forward_ptr == ClaimedForwardPtr) { | |
1011 // Wait to get the real forwarding pointer value. | |
1012 forward_ptr = real_forwardee(old); | |
1013 } | |
1014 par_scan_state->undo_alloc_in_to_space((HeapWord*)new_obj, sz); | |
1015 } | |
1016 | |
1017 return forward_ptr; | |
1018 } | |
1019 | |
1020 | |
1021 // Multiple GC threads may try to promote the same object. If two | |
1022 // or more GC threads copy the object, only one wins the race to install | |
1023 // the forwarding pointer. The other threads have to undo their copy. | |
1024 | |
1025 oop ParNewGeneration::copy_to_survivor_space_with_undo( | |
1026 ParScanThreadState* par_scan_state, oop old, size_t sz, markOop m) { | |
1027 | |
1028 // In the sequential version, this assert also says that the object is | |
1029 // not forwarded. That might not be the case here. It is the case that | |
1030 // the caller observed it to be not forwarded at some time in the past. | |
1031 assert(is_in_reserved(old), "shouldn't be scavenging this oop"); | |
1032 | |
1033 // The sequential code read "old->age()" below. That doesn't work here, | |
1034 // since the age is in the mark word, and that might be overwritten with | |
1035 // a forwarding pointer by a parallel thread. So we must save the mark | |
1036 // word here, install it in a local oopDesc, and then analyze it. | |
1037 oopDesc dummyOld; | |
1038 dummyOld.set_mark(m); | |
1039 assert(!dummyOld.is_forwarded(), | |
1040 "should not be called with forwarding pointer mark word."); | |
1041 | |
1042 bool failed_to_promote = false; | |
1043 oop new_obj = NULL; | |
1044 oop forward_ptr; | |
1045 | |
1046 // Try allocating obj in to-space (unless too old) | |
1047 if (dummyOld.age() < tenuring_threshold()) { | |
1048 new_obj = (oop)par_scan_state->alloc_in_to_space(sz); | |
1049 if (new_obj == NULL) { | |
1050 set_survivor_overflow(true); | |
1051 } | |
1052 } | |
1053 | |
1054 if (new_obj == NULL) { | |
1055 // Either to-space is full or we decided to promote | |
1056 // try allocating obj tenured | |
1057 new_obj = _next_gen->par_promote(par_scan_state->thread_num(), | |
1058 old, m, sz); | |
1059 | |
1060 if (new_obj == NULL) { | |
1061 if (!HandlePromotionFailure) { | |
1062 // A failed promotion likely means the MaxLiveObjectEvacuationRatio | |
1063 // flag is incorrectly set. In any case, its seriously wrong to be | |
1064 // here! | |
1065 vm_exit_out_of_memory(sz*wordSize, "promotion"); | |
1066 } | |
1067 // promotion failed, forward to self | |
1068 forward_ptr = old->forward_to_atomic(old); | |
1069 new_obj = old; | |
1070 | |
1071 if (forward_ptr != NULL) { | |
1072 return forward_ptr; // someone else succeeded | |
1073 } | |
1074 | |
1075 _promotion_failed = true; | |
1076 failed_to_promote = true; | |
1077 | |
1078 preserve_mark_if_necessary(old, m); | |
1079 } | |
1080 } else { | |
1081 // Is in to-space; do copying ourselves. | |
1082 Copy::aligned_disjoint_words((HeapWord*)old, (HeapWord*)new_obj, sz); | |
1083 // Restore the mark word copied above. | |
1084 new_obj->set_mark(m); | |
1085 // Increment age if new_obj still in new generation | |
1086 new_obj->incr_age(); | |
1087 par_scan_state->age_table()->add(new_obj, sz); | |
1088 } | |
1089 assert(new_obj != NULL, "just checking"); | |
1090 | |
1091 // Now attempt to install the forwarding pointer (atomically). | |
1092 // We have to copy the mark word before overwriting with forwarding | |
1093 // ptr, so we can restore it below in the copy. | |
1094 if (!failed_to_promote) { | |
1095 forward_ptr = old->forward_to_atomic(new_obj); | |
1096 } | |
1097 | |
1098 if (forward_ptr == NULL) { | |
1099 oop obj_to_push = new_obj; | |
1100 if (par_scan_state->should_be_partially_scanned(obj_to_push, old)) { | |
1101 // Length field used as index of next element to be scanned. | |
1102 // Real length can be obtained from real_forwardee() | |
1103 arrayOop(old)->set_length(0); | |
1104 obj_to_push = old; | |
1105 assert(obj_to_push->is_forwarded() && obj_to_push->forwardee() != obj_to_push, | |
1106 "push forwarded object"); | |
1107 } | |
1108 // Push it on one of the queues of to-be-scanned objects. | |
1109 if (!par_scan_state->work_queue()->push(obj_to_push)) { | |
1110 // Add stats for overflow pushes. | |
1111 push_on_overflow_list(old); | |
1112 par_scan_state->note_overflow_push(); | |
1113 } | |
1114 par_scan_state->note_push(); | |
1115 | |
1116 return new_obj; | |
1117 } | |
1118 | |
1119 // Oops. Someone beat us to it. Undo the allocation. Where did we | |
1120 // allocate it? | |
1121 if (is_in_reserved(new_obj)) { | |
1122 // Must be in to_space. | |
1123 assert(to()->is_in_reserved(new_obj), "Checking"); | |
1124 par_scan_state->undo_alloc_in_to_space((HeapWord*)new_obj, sz); | |
1125 } else { | |
1126 assert(!_avoid_promotion_undo, "Should not be here if avoiding."); | |
1127 _next_gen->par_promote_alloc_undo(par_scan_state->thread_num(), | |
1128 (HeapWord*)new_obj, sz); | |
1129 } | |
1130 | |
1131 return forward_ptr; | |
1132 } | |
1133 | |
1134 void ParNewGeneration::push_on_overflow_list(oop from_space_obj) { | |
1135 oop cur_overflow_list = _overflow_list; | |
1136 // if the object has been forwarded to itself, then we cannot | |
1137 // use the klass pointer for the linked list. Instead we have | |
1138 // to allocate an oopDesc in the C-Heap and use that for the linked list. | |
1139 if (from_space_obj->forwardee() == from_space_obj) { | |
1140 oopDesc* listhead = NEW_C_HEAP_ARRAY(oopDesc, 1); | |
1141 listhead->forward_to(from_space_obj); | |
1142 from_space_obj = listhead; | |
1143 } | |
1144 while (true) { | |
1145 from_space_obj->set_klass_to_list_ptr(cur_overflow_list); | |
1146 oop observed_overflow_list = | |
1147 (oop)Atomic::cmpxchg_ptr(from_space_obj, &_overflow_list, cur_overflow_list); | |
1148 if (observed_overflow_list == cur_overflow_list) break; | |
1149 // Otherwise... | |
1150 cur_overflow_list = observed_overflow_list; | |
1151 } | |
1152 } | |
1153 | |
1154 bool | |
1155 ParNewGeneration::take_from_overflow_list(ParScanThreadState* par_scan_state) { | |
1156 ObjToScanQueue* work_q = par_scan_state->work_queue(); | |
1157 // How many to take? | |
1158 int objsFromOverflow = MIN2(work_q->max_elems()/4, | |
1159 (juint)ParGCDesiredObjsFromOverflowList); | |
1160 | |
1161 if (_overflow_list == NULL) return false; | |
1162 | |
1163 // Otherwise, there was something there; try claiming the list. | |
1164 oop prefix = (oop)Atomic::xchg_ptr(NULL, &_overflow_list); | |
1165 | |
1166 if (prefix == NULL) { | |
1167 return false; | |
1168 } | |
1169 // Trim off a prefix of at most objsFromOverflow items | |
1170 int i = 1; | |
1171 oop cur = prefix; | |
167
feeb96a45707
6696264: assert("narrow oop can never be zero") for GCBasher & ParNewGC
coleenp
parents:
113
diff
changeset
|
1172 while (i < objsFromOverflow && cur->klass_or_null() != NULL) { |
0 | 1173 i++; cur = oop(cur->klass()); |
1174 } | |
1175 | |
1176 // Reattach remaining (suffix) to overflow list | |
167
feeb96a45707
6696264: assert("narrow oop can never be zero") for GCBasher & ParNewGC
coleenp
parents:
113
diff
changeset
|
1177 if (cur->klass_or_null() != NULL) { |
0 | 1178 oop suffix = oop(cur->klass()); |
1179 cur->set_klass_to_list_ptr(NULL); | |
1180 | |
1181 // Find last item of suffix list | |
1182 oop last = suffix; | |
167
feeb96a45707
6696264: assert("narrow oop can never be zero") for GCBasher & ParNewGC
coleenp
parents:
113
diff
changeset
|
1183 while (last->klass_or_null() != NULL) { |
0 | 1184 last = oop(last->klass()); |
1185 } | |
1186 // Atomically prepend suffix to current overflow list | |
1187 oop cur_overflow_list = _overflow_list; | |
1188 while (true) { | |
1189 last->set_klass_to_list_ptr(cur_overflow_list); | |
1190 oop observed_overflow_list = | |
1191 (oop)Atomic::cmpxchg_ptr(suffix, &_overflow_list, cur_overflow_list); | |
1192 if (observed_overflow_list == cur_overflow_list) break; | |
1193 // Otherwise... | |
1194 cur_overflow_list = observed_overflow_list; | |
1195 } | |
1196 } | |
1197 | |
1198 // Push objects on prefix list onto this thread's work queue | |
1199 assert(cur != NULL, "program logic"); | |
1200 cur = prefix; | |
1201 int n = 0; | |
1202 while (cur != NULL) { | |
1203 oop obj_to_push = cur->forwardee(); | |
1204 oop next = oop(cur->klass()); | |
1205 cur->set_klass(obj_to_push->klass()); | |
1206 if (par_scan_state->should_be_partially_scanned(obj_to_push, cur)) { | |
1207 obj_to_push = cur; | |
1208 assert(arrayOop(cur)->length() == 0, "entire array remaining to be scanned"); | |
1209 } | |
1210 work_q->push(obj_to_push); | |
1211 cur = next; | |
1212 n++; | |
1213 } | |
1214 par_scan_state->note_overflow_refill(n); | |
1215 return true; | |
1216 } | |
1217 | |
1218 void ParNewGeneration::ref_processor_init() | |
1219 { | |
1220 if (_ref_processor == NULL) { | |
1221 // Allocate and initialize a reference processor | |
1222 _ref_processor = ReferenceProcessor::create_ref_processor( | |
1223 _reserved, // span | |
1224 refs_discovery_is_atomic(), // atomic_discovery | |
1225 refs_discovery_is_mt(), // mt_discovery | |
1226 NULL, // is_alive_non_header | |
1227 ParallelGCThreads, | |
1228 ParallelRefProcEnabled); | |
1229 } | |
1230 } | |
1231 | |
1232 const char* ParNewGeneration::name() const { | |
1233 return "par new generation"; | |
1234 } |