Mercurial > hg > graal-compiler
annotate src/share/vm/gc_implementation/parNew/parNewGeneration.cpp @ 14308:870aedf4ba4f
8032379: Remove the is_scavenging flag to process_strong_roots
Summary: Refactor the strong root processing to avoid using a boolean in addition to the ScanOption enum.
Reviewed-by: stefank, tschatzl, ehelin, jmasa
author | mgerdin |
---|---|
date | Tue, 14 Jan 2014 16:40:33 +0100 |
parents | 86e6d691f2e1 |
children | 63a4eb8bcd23 |
rev | line source |
---|---|
0 | 1 /* |
10405 | 2 * Copyright (c) 2001, 2013, Oracle and/or its affiliates. All rights reserved. |
0 | 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 * | |
1552
c18cbe5936b8
6941466: Oracle rebranding changes for Hotspot repositories
trims
parents:
1387
diff
changeset
|
19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA |
c18cbe5936b8
6941466: Oracle rebranding changes for Hotspot repositories
trims
parents:
1387
diff
changeset
|
20 * or visit www.oracle.com if you need additional information or have any |
c18cbe5936b8
6941466: Oracle rebranding changes for Hotspot repositories
trims
parents:
1387
diff
changeset
|
21 * questions. |
0 | 22 * |
23 */ | |
24 | |
1972 | 25 #include "precompiled.hpp" |
26 #include "gc_implementation/concurrentMarkSweep/concurrentMarkSweepGeneration.hpp" | |
27 #include "gc_implementation/parNew/parNewGeneration.hpp" | |
28 #include "gc_implementation/parNew/parOopClosures.inline.hpp" | |
29 #include "gc_implementation/shared/adaptiveSizePolicy.hpp" | |
30 #include "gc_implementation/shared/ageTable.hpp" | |
6595 | 31 #include "gc_implementation/shared/parGCAllocBuffer.hpp" |
10405 | 32 #include "gc_implementation/shared/gcHeapSummary.hpp" |
33 #include "gc_implementation/shared/gcTimer.hpp" | |
34 #include "gc_implementation/shared/gcTrace.hpp" | |
35 #include "gc_implementation/shared/gcTraceTime.hpp" | |
36 #include "gc_implementation/shared/copyFailedInfo.hpp" | |
1972 | 37 #include "gc_implementation/shared/spaceDecorator.hpp" |
38 #include "memory/defNewGeneration.inline.hpp" | |
39 #include "memory/genCollectedHeap.hpp" | |
40 #include "memory/genOopClosures.inline.hpp" | |
41 #include "memory/generation.hpp" | |
42 #include "memory/generation.inline.hpp" | |
43 #include "memory/referencePolicy.hpp" | |
44 #include "memory/resourceArea.hpp" | |
45 #include "memory/sharedHeap.hpp" | |
46 #include "memory/space.hpp" | |
47 #include "oops/objArrayOop.hpp" | |
48 #include "oops/oop.inline.hpp" | |
49 #include "oops/oop.pcgc.inline.hpp" | |
50 #include "runtime/handles.hpp" | |
51 #include "runtime/handles.inline.hpp" | |
52 #include "runtime/java.hpp" | |
53 #include "runtime/thread.hpp" | |
54 #include "utilities/copy.hpp" | |
55 #include "utilities/globalDefinitions.hpp" | |
56 #include "utilities/workgroup.hpp" | |
0 | 57 |
58 #ifdef _MSC_VER | |
59 #pragma warning( push ) | |
60 #pragma warning( disable:4355 ) // 'this' : used in base member initializer list | |
61 #endif | |
62 ParScanThreadState::ParScanThreadState(Space* to_space_, | |
63 ParNewGeneration* gen_, | |
64 Generation* old_gen_, | |
65 int thread_num_, | |
66 ObjToScanQueueSet* work_queue_set_, | |
6197 | 67 Stack<oop, mtGC>* overflow_stacks_, |
0 | 68 size_t desired_plab_sz_, |
69 ParallelTaskTerminator& term_) : | |
679
cea947c8a988
6819891: ParNew: Fix work queue overflow code to deal correctly with +UseCompressedOops
ysr
parents:
579
diff
changeset
|
70 _to_space(to_space_), _old_gen(old_gen_), _young_gen(gen_), _thread_num(thread_num_), |
0 | 71 _work_queue(work_queue_set_->queue(thread_num_)), _to_space_full(false), |
1836
894b1d7c7e01
6423256: GC stacks should use a better data structure
jcoomes
parents:
1833
diff
changeset
|
72 _overflow_stack(overflow_stacks_ ? overflow_stacks_ + thread_num_ : NULL), |
0 | 73 _ageTable(false), // false ==> not the global age table, no perf data. |
74 _to_space_alloc_buffer(desired_plab_sz_), | |
75 _to_space_closure(gen_, this), _old_gen_closure(gen_, this), | |
76 _to_space_root_closure(gen_, this), _old_gen_root_closure(gen_, this), | |
77 _older_gen_closure(gen_, this), | |
78 _evacuate_followers(this, &_to_space_closure, &_old_gen_closure, | |
79 &_to_space_root_closure, gen_, &_old_gen_root_closure, | |
80 work_queue_set_, &term_), | |
81 _is_alive_closure(gen_), _scan_weak_ref_closure(gen_, this), | |
82 _keep_alive_closure(&_scan_weak_ref_closure), | |
83 _strong_roots_time(0.0), _term_time(0.0) | |
84 { | |
1710 | 85 #if TASKQUEUE_STATS |
86 _term_attempts = 0; | |
87 _overflow_refills = 0; | |
88 _overflow_refill_objs = 0; | |
89 #endif // TASKQUEUE_STATS | |
90 | |
0 | 91 _survivor_chunk_array = |
92 (ChunkArray*) old_gen()->get_data_recorder(thread_num()); | |
93 _hash_seed = 17; // Might want to take time-based random value. | |
94 _start = os::elapsedTime(); | |
95 _old_gen_closure.set_generation(old_gen_); | |
96 _old_gen_root_closure.set_generation(old_gen_); | |
97 } | |
98 #ifdef _MSC_VER | |
99 #pragma warning( pop ) | |
100 #endif | |
101 | |
102 void ParScanThreadState::record_survivor_plab(HeapWord* plab_start, | |
103 size_t plab_word_size) { | |
104 ChunkArray* sca = survivor_chunk_array(); | |
105 if (sca != NULL) { | |
106 // A non-null SCA implies that we want the PLAB data recorded. | |
107 sca->record_sample(plab_start, plab_word_size); | |
108 } | |
109 } | |
110 | |
111 bool ParScanThreadState::should_be_partially_scanned(oop new_obj, oop old_obj) const { | |
112 return new_obj->is_objArray() && | |
113 arrayOop(new_obj)->length() > ParGCArrayScanChunk && | |
114 new_obj != old_obj; | |
115 } | |
116 | |
117 void ParScanThreadState::scan_partial_array_and_push_remainder(oop old) { | |
118 assert(old->is_objArray(), "must be obj array"); | |
119 assert(old->is_forwarded(), "must be forwarded"); | |
120 assert(Universe::heap()->is_in_reserved(old), "must be in heap."); | |
679
cea947c8a988
6819891: ParNew: Fix work queue overflow code to deal correctly with +UseCompressedOops
ysr
parents:
579
diff
changeset
|
121 assert(!old_gen()->is_in(old), "must be in young generation."); |
0 | 122 |
123 objArrayOop obj = objArrayOop(old->forwardee()); | |
124 // Process ParGCArrayScanChunk elements now | |
125 // and push the remainder back onto queue | |
126 int start = arrayOop(old)->length(); | |
127 int end = obj->length(); | |
128 int remainder = end - start; | |
129 assert(start <= end, "just checking"); | |
130 if (remainder > 2 * ParGCArrayScanChunk) { | |
131 // Test above combines last partial chunk with a full chunk | |
132 end = start + ParGCArrayScanChunk; | |
133 arrayOop(old)->set_length(end); | |
134 // Push remainder. | |
135 bool ok = work_queue()->push(old); | |
136 assert(ok, "just popped, push must be okay"); | |
137 } else { | |
138 // Restore length so that it can be used if there | |
139 // is a promotion failure and forwarding pointers | |
140 // must be removed. | |
141 arrayOop(old)->set_length(end); | |
142 } | |
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
6
diff
changeset
|
143 |
0 | 144 // 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
|
145 // should make sure end is even (aligned to HeapWord in case of compressed oops) |
0 | 146 if ((HeapWord *)obj < young_old_boundary()) { |
147 // 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
|
148 obj->oop_iterate_range(&_to_space_closure, start, end); |
0 | 149 } else { |
150 // 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
|
151 obj->oop_iterate_range(&_old_gen_closure, start, end); |
0 | 152 } |
153 } | |
154 | |
155 | |
156 void ParScanThreadState::trim_queues(int max_size) { | |
157 ObjToScanQueue* queue = work_queue(); | |
679
cea947c8a988
6819891: ParNew: Fix work queue overflow code to deal correctly with +UseCompressedOops
ysr
parents:
579
diff
changeset
|
158 do { |
cea947c8a988
6819891: ParNew: Fix work queue overflow code to deal correctly with +UseCompressedOops
ysr
parents:
579
diff
changeset
|
159 while (queue->size() > (juint)max_size) { |
cea947c8a988
6819891: ParNew: Fix work queue overflow code to deal correctly with +UseCompressedOops
ysr
parents:
579
diff
changeset
|
160 oop obj_to_scan; |
cea947c8a988
6819891: ParNew: Fix work queue overflow code to deal correctly with +UseCompressedOops
ysr
parents:
579
diff
changeset
|
161 if (queue->pop_local(obj_to_scan)) { |
cea947c8a988
6819891: ParNew: Fix work queue overflow code to deal correctly with +UseCompressedOops
ysr
parents:
579
diff
changeset
|
162 if ((HeapWord *)obj_to_scan < young_old_boundary()) { |
cea947c8a988
6819891: ParNew: Fix work queue overflow code to deal correctly with +UseCompressedOops
ysr
parents:
579
diff
changeset
|
163 if (obj_to_scan->is_objArray() && |
cea947c8a988
6819891: ParNew: Fix work queue overflow code to deal correctly with +UseCompressedOops
ysr
parents:
579
diff
changeset
|
164 obj_to_scan->is_forwarded() && |
cea947c8a988
6819891: ParNew: Fix work queue overflow code to deal correctly with +UseCompressedOops
ysr
parents:
579
diff
changeset
|
165 obj_to_scan->forwardee() != obj_to_scan) { |
cea947c8a988
6819891: ParNew: Fix work queue overflow code to deal correctly with +UseCompressedOops
ysr
parents:
579
diff
changeset
|
166 scan_partial_array_and_push_remainder(obj_to_scan); |
cea947c8a988
6819891: ParNew: Fix work queue overflow code to deal correctly with +UseCompressedOops
ysr
parents:
579
diff
changeset
|
167 } else { |
cea947c8a988
6819891: ParNew: Fix work queue overflow code to deal correctly with +UseCompressedOops
ysr
parents:
579
diff
changeset
|
168 // object is in to_space |
cea947c8a988
6819891: ParNew: Fix work queue overflow code to deal correctly with +UseCompressedOops
ysr
parents:
579
diff
changeset
|
169 obj_to_scan->oop_iterate(&_to_space_closure); |
cea947c8a988
6819891: ParNew: Fix work queue overflow code to deal correctly with +UseCompressedOops
ysr
parents:
579
diff
changeset
|
170 } |
0 | 171 } else { |
679
cea947c8a988
6819891: ParNew: Fix work queue overflow code to deal correctly with +UseCompressedOops
ysr
parents:
579
diff
changeset
|
172 // object is in old generation |
cea947c8a988
6819891: ParNew: Fix work queue overflow code to deal correctly with +UseCompressedOops
ysr
parents:
579
diff
changeset
|
173 obj_to_scan->oop_iterate(&_old_gen_closure); |
0 | 174 } |
175 } | |
176 } | |
679
cea947c8a988
6819891: ParNew: Fix work queue overflow code to deal correctly with +UseCompressedOops
ysr
parents:
579
diff
changeset
|
177 // For the case of compressed oops, we have a private, non-shared |
cea947c8a988
6819891: ParNew: Fix work queue overflow code to deal correctly with +UseCompressedOops
ysr
parents:
579
diff
changeset
|
178 // overflow stack, so we eagerly drain it so as to more evenly |
cea947c8a988
6819891: ParNew: Fix work queue overflow code to deal correctly with +UseCompressedOops
ysr
parents:
579
diff
changeset
|
179 // distribute load early. Note: this may be good to do in |
cea947c8a988
6819891: ParNew: Fix work queue overflow code to deal correctly with +UseCompressedOops
ysr
parents:
579
diff
changeset
|
180 // general rather than delay for the final stealing phase. |
cea947c8a988
6819891: ParNew: Fix work queue overflow code to deal correctly with +UseCompressedOops
ysr
parents:
579
diff
changeset
|
181 // If applicable, we'll transfer a set of objects over to our |
cea947c8a988
6819891: ParNew: Fix work queue overflow code to deal correctly with +UseCompressedOops
ysr
parents:
579
diff
changeset
|
182 // work queue, allowing them to be stolen and draining our |
cea947c8a988
6819891: ParNew: Fix work queue overflow code to deal correctly with +UseCompressedOops
ysr
parents:
579
diff
changeset
|
183 // private overflow stack. |
cea947c8a988
6819891: ParNew: Fix work queue overflow code to deal correctly with +UseCompressedOops
ysr
parents:
579
diff
changeset
|
184 } while (ParGCTrimOverflow && young_gen()->take_from_overflow_list(this)); |
cea947c8a988
6819891: ParNew: Fix work queue overflow code to deal correctly with +UseCompressedOops
ysr
parents:
579
diff
changeset
|
185 } |
cea947c8a988
6819891: ParNew: Fix work queue overflow code to deal correctly with +UseCompressedOops
ysr
parents:
579
diff
changeset
|
186 |
cea947c8a988
6819891: ParNew: Fix work queue overflow code to deal correctly with +UseCompressedOops
ysr
parents:
579
diff
changeset
|
187 bool ParScanThreadState::take_from_overflow_stack() { |
695 | 188 assert(ParGCUseLocalOverflow, "Else should not call"); |
679
cea947c8a988
6819891: ParNew: Fix work queue overflow code to deal correctly with +UseCompressedOops
ysr
parents:
579
diff
changeset
|
189 assert(young_gen()->overflow_list() == NULL, "Error"); |
cea947c8a988
6819891: ParNew: Fix work queue overflow code to deal correctly with +UseCompressedOops
ysr
parents:
579
diff
changeset
|
190 ObjToScanQueue* queue = work_queue(); |
6197 | 191 Stack<oop, mtGC>* const of_stack = overflow_stack(); |
1836
894b1d7c7e01
6423256: GC stacks should use a better data structure
jcoomes
parents:
1833
diff
changeset
|
192 const size_t num_overflow_elems = of_stack->size(); |
894b1d7c7e01
6423256: GC stacks should use a better data structure
jcoomes
parents:
1833
diff
changeset
|
193 const size_t space_available = queue->max_elems() - queue->size(); |
894b1d7c7e01
6423256: GC stacks should use a better data structure
jcoomes
parents:
1833
diff
changeset
|
194 const size_t num_take_elems = MIN3(space_available / 4, |
894b1d7c7e01
6423256: GC stacks should use a better data structure
jcoomes
parents:
1833
diff
changeset
|
195 ParGCDesiredObjsFromOverflowList, |
894b1d7c7e01
6423256: GC stacks should use a better data structure
jcoomes
parents:
1833
diff
changeset
|
196 num_overflow_elems); |
679
cea947c8a988
6819891: ParNew: Fix work queue overflow code to deal correctly with +UseCompressedOops
ysr
parents:
579
diff
changeset
|
197 // Transfer the most recent num_take_elems from the overflow |
cea947c8a988
6819891: ParNew: Fix work queue overflow code to deal correctly with +UseCompressedOops
ysr
parents:
579
diff
changeset
|
198 // stack to our work queue. |
cea947c8a988
6819891: ParNew: Fix work queue overflow code to deal correctly with +UseCompressedOops
ysr
parents:
579
diff
changeset
|
199 for (size_t i = 0; i != num_take_elems; i++) { |
cea947c8a988
6819891: ParNew: Fix work queue overflow code to deal correctly with +UseCompressedOops
ysr
parents:
579
diff
changeset
|
200 oop cur = of_stack->pop(); |
cea947c8a988
6819891: ParNew: Fix work queue overflow code to deal correctly with +UseCompressedOops
ysr
parents:
579
diff
changeset
|
201 oop obj_to_push = cur->forwardee(); |
cea947c8a988
6819891: ParNew: Fix work queue overflow code to deal correctly with +UseCompressedOops
ysr
parents:
579
diff
changeset
|
202 assert(Universe::heap()->is_in_reserved(cur), "Should be in heap"); |
cea947c8a988
6819891: ParNew: Fix work queue overflow code to deal correctly with +UseCompressedOops
ysr
parents:
579
diff
changeset
|
203 assert(!old_gen()->is_in_reserved(cur), "Should be in young gen"); |
cea947c8a988
6819891: ParNew: Fix work queue overflow code to deal correctly with +UseCompressedOops
ysr
parents:
579
diff
changeset
|
204 assert(Universe::heap()->is_in_reserved(obj_to_push), "Should be in heap"); |
cea947c8a988
6819891: ParNew: Fix work queue overflow code to deal correctly with +UseCompressedOops
ysr
parents:
579
diff
changeset
|
205 if (should_be_partially_scanned(obj_to_push, cur)) { |
cea947c8a988
6819891: ParNew: Fix work queue overflow code to deal correctly with +UseCompressedOops
ysr
parents:
579
diff
changeset
|
206 assert(arrayOop(cur)->length() == 0, "entire array remaining to be scanned"); |
cea947c8a988
6819891: ParNew: Fix work queue overflow code to deal correctly with +UseCompressedOops
ysr
parents:
579
diff
changeset
|
207 obj_to_push = cur; |
cea947c8a988
6819891: ParNew: Fix work queue overflow code to deal correctly with +UseCompressedOops
ysr
parents:
579
diff
changeset
|
208 } |
cea947c8a988
6819891: ParNew: Fix work queue overflow code to deal correctly with +UseCompressedOops
ysr
parents:
579
diff
changeset
|
209 bool ok = queue->push(obj_to_push); |
cea947c8a988
6819891: ParNew: Fix work queue overflow code to deal correctly with +UseCompressedOops
ysr
parents:
579
diff
changeset
|
210 assert(ok, "Should have succeeded"); |
0 | 211 } |
679
cea947c8a988
6819891: ParNew: Fix work queue overflow code to deal correctly with +UseCompressedOops
ysr
parents:
579
diff
changeset
|
212 assert(young_gen()->overflow_list() == NULL, "Error"); |
cea947c8a988
6819891: ParNew: Fix work queue overflow code to deal correctly with +UseCompressedOops
ysr
parents:
579
diff
changeset
|
213 return num_take_elems > 0; // was something transferred? |
cea947c8a988
6819891: ParNew: Fix work queue overflow code to deal correctly with +UseCompressedOops
ysr
parents:
579
diff
changeset
|
214 } |
cea947c8a988
6819891: ParNew: Fix work queue overflow code to deal correctly with +UseCompressedOops
ysr
parents:
579
diff
changeset
|
215 |
cea947c8a988
6819891: ParNew: Fix work queue overflow code to deal correctly with +UseCompressedOops
ysr
parents:
579
diff
changeset
|
216 void ParScanThreadState::push_on_overflow_stack(oop p) { |
695 | 217 assert(ParGCUseLocalOverflow, "Else should not call"); |
679
cea947c8a988
6819891: ParNew: Fix work queue overflow code to deal correctly with +UseCompressedOops
ysr
parents:
579
diff
changeset
|
218 overflow_stack()->push(p); |
cea947c8a988
6819891: ParNew: Fix work queue overflow code to deal correctly with +UseCompressedOops
ysr
parents:
579
diff
changeset
|
219 assert(young_gen()->overflow_list() == NULL, "Error"); |
0 | 220 } |
221 | |
222 HeapWord* ParScanThreadState::alloc_in_to_space_slow(size_t word_sz) { | |
223 | |
224 // Otherwise, if the object is small enough, try to reallocate the | |
225 // buffer. | |
226 HeapWord* obj = NULL; | |
227 if (!_to_space_full) { | |
228 ParGCAllocBuffer* const plab = to_space_alloc_buffer(); | |
229 Space* const sp = to_space(); | |
230 if (word_sz * 100 < | |
231 ParallelGCBufferWastePct * plab->word_sz()) { | |
232 // Is small enough; abandon this buffer and start a new one. | |
233 plab->retire(false, false); | |
234 size_t buf_size = plab->word_sz(); | |
235 HeapWord* buf_space = sp->par_allocate(buf_size); | |
236 if (buf_space == NULL) { | |
237 const size_t min_bytes = | |
238 ParGCAllocBuffer::min_size() << LogHeapWordSize; | |
239 size_t free_bytes = sp->free(); | |
240 while(buf_space == NULL && free_bytes >= min_bytes) { | |
241 buf_size = free_bytes >> LogHeapWordSize; | |
242 assert(buf_size == (size_t)align_object_size(buf_size), | |
243 "Invariant"); | |
244 buf_space = sp->par_allocate(buf_size); | |
245 free_bytes = sp->free(); | |
246 } | |
247 } | |
248 if (buf_space != NULL) { | |
249 plab->set_word_size(buf_size); | |
250 plab->set_buf(buf_space); | |
251 record_survivor_plab(buf_space, buf_size); | |
252 obj = plab->allocate(word_sz); | |
253 // Note that we cannot compare buf_size < word_sz below | |
254 // because of AlignmentReserve (see ParGCAllocBuffer::allocate()). | |
255 assert(obj != NULL || plab->words_remaining() < word_sz, | |
256 "Else should have been able to allocate"); | |
257 // It's conceivable that we may be able to use the | |
258 // buffer we just grabbed for subsequent small requests | |
259 // even if not for this one. | |
260 } else { | |
261 // We're used up. | |
262 _to_space_full = true; | |
263 } | |
264 | |
265 } else { | |
266 // Too large; allocate the object individually. | |
267 obj = sp->par_allocate(word_sz); | |
268 } | |
269 } | |
270 return obj; | |
271 } | |
272 | |
273 | |
274 void ParScanThreadState::undo_alloc_in_to_space(HeapWord* obj, | |
275 size_t word_sz) { | |
276 // Is the alloc in the current alloc buffer? | |
277 if (to_space_alloc_buffer()->contains(obj)) { | |
278 assert(to_space_alloc_buffer()->contains(obj + word_sz - 1), | |
279 "Should contain whole object."); | |
280 to_space_alloc_buffer()->undo_allocation(obj, word_sz); | |
281 } else { | |
481
7d7a7c599c17
6578152: fill_region_with_object has usability and safety issues
jcoomes
parents:
457
diff
changeset
|
282 CollectedHeap::fill_with_object(obj, word_sz); |
0 | 283 } |
284 } | |
285 | |
10405 | 286 void ParScanThreadState::print_promotion_failure_size() { |
287 if (_promotion_failed_info.has_failed() && PrintPromotionFailure) { | |
288 gclog_or_tty->print(" (%d: promotion failure size = " SIZE_FORMAT ") ", | |
289 _thread_num, _promotion_failed_info.first_size()); | |
1145
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
989
diff
changeset
|
290 } |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
989
diff
changeset
|
291 } |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
989
diff
changeset
|
292 |
0 | 293 class ParScanThreadStateSet: private ResourceArray { |
294 public: | |
295 // Initializes states for the specified number of threads; | |
296 ParScanThreadStateSet(int num_threads, | |
297 Space& to_space, | |
298 ParNewGeneration& gen, | |
299 Generation& old_gen, | |
300 ObjToScanQueueSet& queue_set, | |
6197 | 301 Stack<oop, mtGC>* overflow_stacks_, |
0 | 302 size_t desired_plab_sz, |
303 ParallelTaskTerminator& term); | |
1710 | 304 |
305 ~ParScanThreadStateSet() { TASKQUEUE_STATS_ONLY(reset_stats()); } | |
306 | |
1145
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
989
diff
changeset
|
307 inline ParScanThreadState& thread_state(int i); |
1710 | 308 |
10405 | 309 void trace_promotion_failed(YoungGCTracer& gc_tracer); |
4095
bca17e38de00
6593758: RFE: Enhance GC ergonomics to dynamically choose ParallelGCThreads
jmasa
parents:
2369
diff
changeset
|
310 void reset(int active_workers, bool promotion_failed); |
0 | 311 void flush(); |
1710 | 312 |
313 #if TASKQUEUE_STATS | |
314 static void | |
315 print_termination_stats_hdr(outputStream* const st = gclog_or_tty); | |
316 void print_termination_stats(outputStream* const st = gclog_or_tty); | |
317 static void | |
318 print_taskqueue_stats_hdr(outputStream* const st = gclog_or_tty); | |
319 void print_taskqueue_stats(outputStream* const st = gclog_or_tty); | |
320 void reset_stats(); | |
321 #endif // TASKQUEUE_STATS | |
322 | |
0 | 323 private: |
324 ParallelTaskTerminator& _term; | |
325 ParNewGeneration& _gen; | |
326 Generation& _next_gen; | |
4095
bca17e38de00
6593758: RFE: Enhance GC ergonomics to dynamically choose ParallelGCThreads
jmasa
parents:
2369
diff
changeset
|
327 public: |
bca17e38de00
6593758: RFE: Enhance GC ergonomics to dynamically choose ParallelGCThreads
jmasa
parents:
2369
diff
changeset
|
328 bool is_valid(int id) const { return id < length(); } |
bca17e38de00
6593758: RFE: Enhance GC ergonomics to dynamically choose ParallelGCThreads
jmasa
parents:
2369
diff
changeset
|
329 ParallelTaskTerminator* terminator() { return &_term; } |
0 | 330 }; |
331 | |
332 | |
333 ParScanThreadStateSet::ParScanThreadStateSet( | |
334 int num_threads, Space& to_space, ParNewGeneration& gen, | |
335 Generation& old_gen, ObjToScanQueueSet& queue_set, | |
6197 | 336 Stack<oop, mtGC>* overflow_stacks, |
0 | 337 size_t desired_plab_sz, ParallelTaskTerminator& term) |
338 : ResourceArray(sizeof(ParScanThreadState), num_threads), | |
1710 | 339 _gen(gen), _next_gen(old_gen), _term(term) |
0 | 340 { |
341 assert(num_threads > 0, "sanity check!"); | |
1836
894b1d7c7e01
6423256: GC stacks should use a better data structure
jcoomes
parents:
1833
diff
changeset
|
342 assert(ParGCUseLocalOverflow == (overflow_stacks != NULL), |
894b1d7c7e01
6423256: GC stacks should use a better data structure
jcoomes
parents:
1833
diff
changeset
|
343 "overflow_stack allocation mismatch"); |
0 | 344 // Initialize states. |
345 for (int i = 0; i < num_threads; ++i) { | |
346 new ((ParScanThreadState*)_data + i) | |
347 ParScanThreadState(&to_space, &gen, &old_gen, i, &queue_set, | |
1836
894b1d7c7e01
6423256: GC stacks should use a better data structure
jcoomes
parents:
1833
diff
changeset
|
348 overflow_stacks, desired_plab_sz, term); |
0 | 349 } |
350 } | |
351 | |
1145
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
989
diff
changeset
|
352 inline ParScanThreadState& ParScanThreadStateSet::thread_state(int i) |
0 | 353 { |
354 assert(i >= 0 && i < length(), "sanity check!"); | |
355 return ((ParScanThreadState*)_data)[i]; | |
356 } | |
357 | |
10405 | 358 void ParScanThreadStateSet::trace_promotion_failed(YoungGCTracer& gc_tracer) { |
359 for (int i = 0; i < length(); ++i) { | |
360 if (thread_state(i).promotion_failed()) { | |
361 gc_tracer.report_promotion_failed(thread_state(i).promotion_failed_info()); | |
362 thread_state(i).promotion_failed_info().reset(); | |
363 } | |
364 } | |
365 } | |
0 | 366 |
4095
bca17e38de00
6593758: RFE: Enhance GC ergonomics to dynamically choose ParallelGCThreads
jmasa
parents:
2369
diff
changeset
|
367 void ParScanThreadStateSet::reset(int active_threads, bool promotion_failed) |
0 | 368 { |
4095
bca17e38de00
6593758: RFE: Enhance GC ergonomics to dynamically choose ParallelGCThreads
jmasa
parents:
2369
diff
changeset
|
369 _term.reset_for_reuse(active_threads); |
1145
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
989
diff
changeset
|
370 if (promotion_failed) { |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
989
diff
changeset
|
371 for (int i = 0; i < length(); ++i) { |
10405 | 372 thread_state(i).print_promotion_failure_size(); |
1145
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
989
diff
changeset
|
373 } |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
989
diff
changeset
|
374 } |
0 | 375 } |
376 | |
1710 | 377 #if TASKQUEUE_STATS |
378 void | |
379 ParScanThreadState::reset_stats() | |
380 { | |
381 taskqueue_stats().reset(); | |
382 _term_attempts = 0; | |
383 _overflow_refills = 0; | |
384 _overflow_refill_objs = 0; | |
385 } | |
386 | |
387 void ParScanThreadStateSet::reset_stats() | |
388 { | |
389 for (int i = 0; i < length(); ++i) { | |
390 thread_state(i).reset_stats(); | |
391 } | |
392 } | |
393 | |
394 void | |
395 ParScanThreadStateSet::print_termination_stats_hdr(outputStream* const st) | |
396 { | |
397 st->print_raw_cr("GC Termination Stats"); | |
398 st->print_raw_cr(" elapsed --strong roots-- " | |
399 "-------termination-------"); | |
400 st->print_raw_cr("thr ms ms % " | |
401 " ms % attempts"); | |
402 st->print_raw_cr("--- --------- --------- ------ " | |
403 "--------- ------ --------"); | |
404 } | |
405 | |
406 void ParScanThreadStateSet::print_termination_stats(outputStream* const st) | |
407 { | |
408 print_termination_stats_hdr(st); | |
409 | |
410 for (int i = 0; i < length(); ++i) { | |
411 const ParScanThreadState & pss = thread_state(i); | |
412 const double elapsed_ms = pss.elapsed_time() * 1000.0; | |
413 const double s_roots_ms = pss.strong_roots_time() * 1000.0; | |
414 const double term_ms = pss.term_time() * 1000.0; | |
415 st->print_cr("%3d %9.2f %9.2f %6.2f " | |
416 "%9.2f %6.2f " SIZE_FORMAT_W(8), | |
417 i, elapsed_ms, s_roots_ms, s_roots_ms * 100 / elapsed_ms, | |
418 term_ms, term_ms * 100 / elapsed_ms, pss.term_attempts()); | |
419 } | |
420 } | |
421 | |
422 // Print stats related to work queue activity. | |
423 void ParScanThreadStateSet::print_taskqueue_stats_hdr(outputStream* const st) | |
424 { | |
425 st->print_raw_cr("GC Task Stats"); | |
426 st->print_raw("thr "); TaskQueueStats::print_header(1, st); st->cr(); | |
427 st->print_raw("--- "); TaskQueueStats::print_header(2, st); st->cr(); | |
428 } | |
429 | |
430 void ParScanThreadStateSet::print_taskqueue_stats(outputStream* const st) | |
431 { | |
432 print_taskqueue_stats_hdr(st); | |
433 | |
434 TaskQueueStats totals; | |
435 for (int i = 0; i < length(); ++i) { | |
436 const ParScanThreadState & pss = thread_state(i); | |
437 const TaskQueueStats & stats = pss.taskqueue_stats(); | |
438 st->print("%3d ", i); stats.print(st); st->cr(); | |
439 totals += stats; | |
440 | |
441 if (pss.overflow_refills() > 0) { | |
442 st->print_cr(" " SIZE_FORMAT_W(10) " overflow refills " | |
443 SIZE_FORMAT_W(10) " overflow objects", | |
444 pss.overflow_refills(), pss.overflow_refill_objs()); | |
445 } | |
446 } | |
447 st->print("tot "); totals.print(st); st->cr(); | |
448 | |
449 DEBUG_ONLY(totals.verify()); | |
450 } | |
451 #endif // TASKQUEUE_STATS | |
452 | |
0 | 453 void ParScanThreadStateSet::flush() |
454 { | |
1145
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
989
diff
changeset
|
455 // Work in this loop should be kept as lightweight as |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
989
diff
changeset
|
456 // possible since this might otherwise become a bottleneck |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
989
diff
changeset
|
457 // to scaling. Should we add heavy-weight work into this |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
989
diff
changeset
|
458 // loop, consider parallelizing the loop into the worker threads. |
0 | 459 for (int i = 0; i < length(); ++i) { |
1145
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
989
diff
changeset
|
460 ParScanThreadState& par_scan_state = thread_state(i); |
0 | 461 |
462 // Flush stats related to To-space PLAB activity and | |
463 // retire the last buffer. | |
464 par_scan_state.to_space_alloc_buffer()-> | |
465 flush_stats_and_retire(_gen.plab_stats(), | |
6595 | 466 true /* end_of_gc */, |
467 false /* retain */); | |
0 | 468 |
469 // Every thread has its own age table. We need to merge | |
470 // them all into one. | |
471 ageTable *local_table = par_scan_state.age_table(); | |
472 _gen.age_table()->merge(local_table); | |
473 | |
474 // Inform old gen that we're done. | |
475 _next_gen.par_promote_alloc_done(i); | |
476 _next_gen.par_oop_since_save_marks_iterate_done(i); | |
1710 | 477 } |
0 | 478 |
1145
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
989
diff
changeset
|
479 if (UseConcMarkSweepGC && ParallelGCThreads > 0) { |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
989
diff
changeset
|
480 // We need to call this even when ResizeOldPLAB is disabled |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
989
diff
changeset
|
481 // so as to avoid breaking some asserts. While we may be able |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
989
diff
changeset
|
482 // to avoid this by reorganizing the code a bit, I am loathe |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
989
diff
changeset
|
483 // to do that unless we find cases where ergo leads to bad |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
989
diff
changeset
|
484 // performance. |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
989
diff
changeset
|
485 CFLS_LAB::compute_desired_plab_size(); |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
989
diff
changeset
|
486 } |
0 | 487 } |
488 | |
489 ParScanClosure::ParScanClosure(ParNewGeneration* g, | |
490 ParScanThreadState* par_scan_state) : | |
6725
da91efe96a93
6964458: Reimplement class meta-data storage to use native memory
coleenp
parents:
6595
diff
changeset
|
491 OopsInKlassOrGenClosure(g), _par_scan_state(par_scan_state), _g(g) |
0 | 492 { |
493 assert(_g->level() == 0, "Optimized for youngest generation"); | |
494 _boundary = _g->reserved().end(); | |
495 } | |
496 | |
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
6
diff
changeset
|
497 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
|
498 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
|
499 |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
6
diff
changeset
|
500 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
|
501 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
|
502 |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
6
diff
changeset
|
503 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
|
504 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
|
505 |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
6
diff
changeset
|
506 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
|
507 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
|
508 |
0 | 509 ParScanWeakRefClosure::ParScanWeakRefClosure(ParNewGeneration* g, |
510 ParScanThreadState* par_scan_state) | |
511 : 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
|
512 {} |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
6
diff
changeset
|
513 |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
6
diff
changeset
|
514 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
|
515 void ParScanWeakRefClosure::do_oop(narrowOop* p) { ParScanWeakRefClosure::do_oop_work(p); } |
0 | 516 |
517 #ifdef WIN32 | |
518 #pragma warning(disable: 4786) /* identifier was truncated to '255' characters in the browser information */ | |
519 #endif | |
520 | |
521 ParEvacuateFollowersClosure::ParEvacuateFollowersClosure( | |
522 ParScanThreadState* par_scan_state_, | |
523 ParScanWithoutBarrierClosure* to_space_closure_, | |
524 ParScanWithBarrierClosure* old_gen_closure_, | |
525 ParRootScanWithoutBarrierClosure* to_space_root_closure_, | |
526 ParNewGeneration* par_gen_, | |
527 ParRootScanWithBarrierTwoGensClosure* old_gen_root_closure_, | |
528 ObjToScanQueueSet* task_queues_, | |
529 ParallelTaskTerminator* terminator_) : | |
530 | |
531 _par_scan_state(par_scan_state_), | |
532 _to_space_closure(to_space_closure_), | |
533 _old_gen_closure(old_gen_closure_), | |
534 _to_space_root_closure(to_space_root_closure_), | |
535 _old_gen_root_closure(old_gen_root_closure_), | |
536 _par_gen(par_gen_), | |
537 _task_queues(task_queues_), | |
538 _terminator(terminator_) | |
539 {} | |
540 | |
541 void ParEvacuateFollowersClosure::do_void() { | |
542 ObjToScanQueue* work_q = par_scan_state()->work_queue(); | |
543 | |
544 while (true) { | |
545 | |
546 // Scan to-space and old-gen objs until we run out of both. | |
547 oop obj_to_scan; | |
548 par_scan_state()->trim_queues(0); | |
549 | |
550 // We have no local work, attempt to steal from other threads. | |
551 | |
552 // attempt to steal work from promoted. | |
553 if (task_queues()->steal(par_scan_state()->thread_num(), | |
554 par_scan_state()->hash_seed(), | |
555 obj_to_scan)) { | |
556 bool res = work_q->push(obj_to_scan); | |
557 assert(res, "Empty queue should have room for a push."); | |
558 | |
559 // if successful, goto Start. | |
560 continue; | |
561 | |
562 // try global overflow list. | |
563 } else if (par_gen()->take_from_overflow_list(par_scan_state())) { | |
564 continue; | |
565 } | |
566 | |
567 // Otherwise, offer termination. | |
568 par_scan_state()->start_term_time(); | |
569 if (terminator()->offer_termination()) break; | |
570 par_scan_state()->end_term_time(); | |
571 } | |
534 | 572 assert(par_gen()->_overflow_list == NULL && par_gen()->_num_par_pushes == 0, |
573 "Broken overflow list?"); | |
0 | 574 // Finish the last termination pause. |
575 par_scan_state()->end_term_time(); | |
576 } | |
577 | |
578 ParNewGenTask::ParNewGenTask(ParNewGeneration* gen, Generation* next_gen, | |
579 HeapWord* young_old_boundary, ParScanThreadStateSet* state_set) : | |
580 AbstractGangTask("ParNewGeneration collection"), | |
581 _gen(gen), _next_gen(next_gen), | |
582 _young_old_boundary(young_old_boundary), | |
583 _state_set(state_set) | |
584 {} | |
585 | |
4095
bca17e38de00
6593758: RFE: Enhance GC ergonomics to dynamically choose ParallelGCThreads
jmasa
parents:
2369
diff
changeset
|
586 // Reset the terminator for the given number of |
bca17e38de00
6593758: RFE: Enhance GC ergonomics to dynamically choose ParallelGCThreads
jmasa
parents:
2369
diff
changeset
|
587 // active threads. |
bca17e38de00
6593758: RFE: Enhance GC ergonomics to dynamically choose ParallelGCThreads
jmasa
parents:
2369
diff
changeset
|
588 void ParNewGenTask::set_for_termination(int active_workers) { |
bca17e38de00
6593758: RFE: Enhance GC ergonomics to dynamically choose ParallelGCThreads
jmasa
parents:
2369
diff
changeset
|
589 _state_set->reset(active_workers, _gen->promotion_failed()); |
bca17e38de00
6593758: RFE: Enhance GC ergonomics to dynamically choose ParallelGCThreads
jmasa
parents:
2369
diff
changeset
|
590 // Should the heap be passed in? There's only 1 for now so |
bca17e38de00
6593758: RFE: Enhance GC ergonomics to dynamically choose ParallelGCThreads
jmasa
parents:
2369
diff
changeset
|
591 // grab it instead. |
bca17e38de00
6593758: RFE: Enhance GC ergonomics to dynamically choose ParallelGCThreads
jmasa
parents:
2369
diff
changeset
|
592 GenCollectedHeap* gch = GenCollectedHeap::heap(); |
bca17e38de00
6593758: RFE: Enhance GC ergonomics to dynamically choose ParallelGCThreads
jmasa
parents:
2369
diff
changeset
|
593 gch->set_n_termination(active_workers); |
bca17e38de00
6593758: RFE: Enhance GC ergonomics to dynamically choose ParallelGCThreads
jmasa
parents:
2369
diff
changeset
|
594 } |
bca17e38de00
6593758: RFE: Enhance GC ergonomics to dynamically choose ParallelGCThreads
jmasa
parents:
2369
diff
changeset
|
595 |
4728
441e946dc1af
7121618: Change type of number of GC workers to unsigned int.
jmasa
parents:
4095
diff
changeset
|
596 void ParNewGenTask::work(uint worker_id) { |
0 | 597 GenCollectedHeap* gch = GenCollectedHeap::heap(); |
598 // Since this is being done in a separate thread, need new resource | |
599 // and handle marks. | |
600 ResourceMark rm; | |
601 HandleMark hm; | |
602 // We would need multiple old-gen queues otherwise. | |
679
cea947c8a988
6819891: ParNew: Fix work queue overflow code to deal correctly with +UseCompressedOops
ysr
parents:
579
diff
changeset
|
603 assert(gch->n_gens() == 2, "Par young collection currently only works with one older gen."); |
0 | 604 |
605 Generation* old_gen = gch->next_gen(_gen); | |
606 | |
4728
441e946dc1af
7121618: Change type of number of GC workers to unsigned int.
jmasa
parents:
4095
diff
changeset
|
607 ParScanThreadState& par_scan_state = _state_set->thread_state(worker_id); |
441e946dc1af
7121618: Change type of number of GC workers to unsigned int.
jmasa
parents:
4095
diff
changeset
|
608 assert(_state_set->is_valid(worker_id), "Should not have been called"); |
4095
bca17e38de00
6593758: RFE: Enhance GC ergonomics to dynamically choose ParallelGCThreads
jmasa
parents:
2369
diff
changeset
|
609 |
0 | 610 par_scan_state.set_young_old_boundary(_young_old_boundary); |
611 | |
6725
da91efe96a93
6964458: Reimplement class meta-data storage to use native memory
coleenp
parents:
6595
diff
changeset
|
612 KlassScanClosure klass_scan_closure(&par_scan_state.to_space_root_closure(), |
da91efe96a93
6964458: Reimplement class meta-data storage to use native memory
coleenp
parents:
6595
diff
changeset
|
613 gch->rem_set()->klass_rem_set()); |
da91efe96a93
6964458: Reimplement class meta-data storage to use native memory
coleenp
parents:
6595
diff
changeset
|
614 |
14308
870aedf4ba4f
8032379: Remove the is_scavenging flag to process_strong_roots
mgerdin
parents:
13400
diff
changeset
|
615 int so = SharedHeap::SO_AllClasses | SharedHeap::SO_Strings | SharedHeap::SO_ScavengeCodeCache; |
6725
da91efe96a93
6964458: Reimplement class meta-data storage to use native memory
coleenp
parents:
6595
diff
changeset
|
616 |
0 | 617 par_scan_state.start_strong_roots(); |
618 gch->gen_process_strong_roots(_gen->level(), | |
989
148e5441d916
6863023: need non-perm oops in code cache for JSR 292
jrose
parents:
695
diff
changeset
|
619 true, // Process younger gens, if any, |
148e5441d916
6863023: need non-perm oops in code cache for JSR 292
jrose
parents:
695
diff
changeset
|
620 // as strong roots. |
148e5441d916
6863023: need non-perm oops in code cache for JSR 292
jrose
parents:
695
diff
changeset
|
621 false, // no scope; this is parallel code |
6725
da91efe96a93
6964458: Reimplement class meta-data storage to use native memory
coleenp
parents:
6595
diff
changeset
|
622 SharedHeap::ScanningOption(so), |
989
148e5441d916
6863023: need non-perm oops in code cache for JSR 292
jrose
parents:
695
diff
changeset
|
623 &par_scan_state.to_space_root_closure(), |
148e5441d916
6863023: need non-perm oops in code cache for JSR 292
jrose
parents:
695
diff
changeset
|
624 true, // walk *all* scavengable nmethods |
6725
da91efe96a93
6964458: Reimplement class meta-data storage to use native memory
coleenp
parents:
6595
diff
changeset
|
625 &par_scan_state.older_gen_closure(), |
da91efe96a93
6964458: Reimplement class meta-data storage to use native memory
coleenp
parents:
6595
diff
changeset
|
626 &klass_scan_closure); |
0 | 627 par_scan_state.end_strong_roots(); |
628 | |
629 // "evacuate followers". | |
630 par_scan_state.evacuate_followers_closure().do_void(); | |
631 } | |
632 | |
633 #ifdef _MSC_VER | |
634 #pragma warning( push ) | |
635 #pragma warning( disable:4355 ) // 'this' : used in base member initializer list | |
636 #endif | |
637 ParNewGeneration:: | |
638 ParNewGeneration(ReservedSpace rs, size_t initial_byte_size, int level) | |
639 : DefNewGeneration(rs, initial_byte_size, level, "PCopy"), | |
640 _overflow_list(NULL), | |
641 _is_alive_closure(this), | |
642 _plab_stats(YoungPLABSize, PLABWeight) | |
643 { | |
534 | 644 NOT_PRODUCT(_overflow_counter = ParGCWorkQueueOverflowInterval;) |
645 NOT_PRODUCT(_num_par_pushes = 0;) | |
0 | 646 _task_queues = new ObjToScanQueueSet(ParallelGCThreads); |
647 guarantee(_task_queues != NULL, "task_queues allocation failure."); | |
648 | |
649 for (uint i1 = 0; i1 < ParallelGCThreads; i1++) { | |
1665 | 650 ObjToScanQueue *q = new ObjToScanQueue(); |
651 guarantee(q != NULL, "work_queue Allocation failure."); | |
652 _task_queues->register_queue(i1, q); | |
0 | 653 } |
654 | |
655 for (uint i2 = 0; i2 < ParallelGCThreads; i2++) | |
656 _task_queues->queue(i2)->initialize(); | |
657 | |
1836
894b1d7c7e01
6423256: GC stacks should use a better data structure
jcoomes
parents:
1833
diff
changeset
|
658 _overflow_stacks = NULL; |
894b1d7c7e01
6423256: GC stacks should use a better data structure
jcoomes
parents:
1833
diff
changeset
|
659 if (ParGCUseLocalOverflow) { |
6197 | 660 |
661 // typedef to workaround NEW_C_HEAP_ARRAY macro, which can not deal | |
662 // with ',' | |
663 typedef Stack<oop, mtGC> GCOopStack; | |
664 | |
665 _overflow_stacks = NEW_C_HEAP_ARRAY(GCOopStack, ParallelGCThreads, mtGC); | |
1836
894b1d7c7e01
6423256: GC stacks should use a better data structure
jcoomes
parents:
1833
diff
changeset
|
666 for (size_t i = 0; i < ParallelGCThreads; ++i) { |
6197 | 667 new (_overflow_stacks + i) Stack<oop, mtGC>(); |
695 | 668 } |
669 } | |
670 | |
0 | 671 if (UsePerfData) { |
672 EXCEPTION_MARK; | |
673 ResourceMark rm; | |
674 | |
675 const char* cname = | |
676 PerfDataManager::counter_name(_gen_counters->name_space(), "threads"); | |
677 PerfDataManager::create_constant(SUN_GC, cname, PerfData::U_None, | |
678 ParallelGCThreads, CHECK); | |
679 } | |
680 } | |
681 #ifdef _MSC_VER | |
682 #pragma warning( pop ) | |
683 #endif | |
684 | |
685 // ParNewGeneration:: | |
686 ParKeepAliveClosure::ParKeepAliveClosure(ParScanWeakRefClosure* cl) : | |
687 DefNewGeneration::KeepAliveClosure(cl), _par_cl(cl) {} | |
688 | |
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
6
diff
changeset
|
689 template <class T> |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
6
diff
changeset
|
690 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
|
691 #ifdef ASSERT |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
6
diff
changeset
|
692 { |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
6
diff
changeset
|
693 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
|
694 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
|
695 // 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
|
696 // as a weak reference. |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
6
diff
changeset
|
697 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
|
698 } |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
6
diff
changeset
|
699 #endif // ASSERT |
0 | 700 |
701 _par_cl->do_oop_nv(p); | |
702 | |
703 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
|
704 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
|
705 _rs->write_ref_field_gc_par(p, obj); |
0 | 706 } |
707 } | |
708 | |
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
6
diff
changeset
|
709 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
|
710 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
|
711 |
0 | 712 // ParNewGeneration:: |
713 KeepAliveClosure::KeepAliveClosure(ScanWeakRefClosure* cl) : | |
714 DefNewGeneration::KeepAliveClosure(cl) {} | |
715 | |
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
6
diff
changeset
|
716 template <class T> |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
6
diff
changeset
|
717 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
|
718 #ifdef ASSERT |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
6
diff
changeset
|
719 { |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
6
diff
changeset
|
720 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
|
721 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
|
722 // 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
|
723 // as a weak reference. |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
6
diff
changeset
|
724 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
|
725 } |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
6
diff
changeset
|
726 #endif // ASSERT |
0 | 727 |
728 _cl->do_oop_nv(p); | |
729 | |
730 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
|
731 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
|
732 _rs->write_ref_field_gc_par(p, obj); |
0 | 733 } |
734 } | |
735 | |
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
6
diff
changeset
|
736 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
|
737 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
|
738 |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
6
diff
changeset
|
739 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
|
740 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
|
741 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
|
742 oop obj = oopDesc::decode_heap_oop_not_null(heap_oop); |
0 | 743 if ((HeapWord*)obj < _boundary) { |
744 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
|
745 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
|
746 ? obj->forwardee() |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
6
diff
changeset
|
747 : _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
|
748 oopDesc::encode_store_heap_oop_not_null(p, new_obj); |
0 | 749 } |
750 if (_gc_barrier) { | |
751 // If p points to a younger generation, mark the card. | |
752 if ((HeapWord*)obj < _gen_boundary) { | |
753 _rs->write_ref_field_gc_par(p, obj); | |
754 } | |
755 } | |
756 } | |
757 } | |
758 | |
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
6
diff
changeset
|
759 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
|
760 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
|
761 |
0 | 762 class ParNewRefProcTaskProxy: public AbstractGangTask { |
763 typedef AbstractRefProcTaskExecutor::ProcessTask ProcessTask; | |
764 public: | |
765 ParNewRefProcTaskProxy(ProcessTask& task, ParNewGeneration& gen, | |
766 Generation& next_gen, | |
767 HeapWord* young_old_boundary, | |
768 ParScanThreadStateSet& state_set); | |
769 | |
770 private: | |
4728
441e946dc1af
7121618: Change type of number of GC workers to unsigned int.
jmasa
parents:
4095
diff
changeset
|
771 virtual void work(uint worker_id); |
4095
bca17e38de00
6593758: RFE: Enhance GC ergonomics to dynamically choose ParallelGCThreads
jmasa
parents:
2369
diff
changeset
|
772 virtual void set_for_termination(int active_workers) { |
bca17e38de00
6593758: RFE: Enhance GC ergonomics to dynamically choose ParallelGCThreads
jmasa
parents:
2369
diff
changeset
|
773 _state_set.terminator()->reset_for_reuse(active_workers); |
bca17e38de00
6593758: RFE: Enhance GC ergonomics to dynamically choose ParallelGCThreads
jmasa
parents:
2369
diff
changeset
|
774 } |
0 | 775 private: |
776 ParNewGeneration& _gen; | |
777 ProcessTask& _task; | |
778 Generation& _next_gen; | |
779 HeapWord* _young_old_boundary; | |
780 ParScanThreadStateSet& _state_set; | |
781 }; | |
782 | |
783 ParNewRefProcTaskProxy::ParNewRefProcTaskProxy( | |
784 ProcessTask& task, ParNewGeneration& gen, | |
785 Generation& next_gen, | |
786 HeapWord* young_old_boundary, | |
787 ParScanThreadStateSet& state_set) | |
788 : AbstractGangTask("ParNewGeneration parallel reference processing"), | |
789 _gen(gen), | |
790 _task(task), | |
791 _next_gen(next_gen), | |
792 _young_old_boundary(young_old_boundary), | |
793 _state_set(state_set) | |
794 { | |
795 } | |
796 | |
4728
441e946dc1af
7121618: Change type of number of GC workers to unsigned int.
jmasa
parents:
4095
diff
changeset
|
797 void ParNewRefProcTaskProxy::work(uint worker_id) |
0 | 798 { |
799 ResourceMark rm; | |
800 HandleMark hm; | |
4728
441e946dc1af
7121618: Change type of number of GC workers to unsigned int.
jmasa
parents:
4095
diff
changeset
|
801 ParScanThreadState& par_scan_state = _state_set.thread_state(worker_id); |
0 | 802 par_scan_state.set_young_old_boundary(_young_old_boundary); |
4728
441e946dc1af
7121618: Change type of number of GC workers to unsigned int.
jmasa
parents:
4095
diff
changeset
|
803 _task.work(worker_id, par_scan_state.is_alive_closure(), |
0 | 804 par_scan_state.keep_alive_closure(), |
805 par_scan_state.evacuate_followers_closure()); | |
806 } | |
807 | |
808 class ParNewRefEnqueueTaskProxy: public AbstractGangTask { | |
809 typedef AbstractRefProcTaskExecutor::EnqueueTask EnqueueTask; | |
810 EnqueueTask& _task; | |
811 | |
812 public: | |
813 ParNewRefEnqueueTaskProxy(EnqueueTask& task) | |
814 : AbstractGangTask("ParNewGeneration parallel reference enqueue"), | |
815 _task(task) | |
816 { } | |
817 | |
4728
441e946dc1af
7121618: Change type of number of GC workers to unsigned int.
jmasa
parents:
4095
diff
changeset
|
818 virtual void work(uint worker_id) |
0 | 819 { |
4728
441e946dc1af
7121618: Change type of number of GC workers to unsigned int.
jmasa
parents:
4095
diff
changeset
|
820 _task.work(worker_id); |
0 | 821 } |
822 }; | |
823 | |
824 | |
825 void ParNewRefProcTaskExecutor::execute(ProcessTask& task) | |
826 { | |
827 GenCollectedHeap* gch = GenCollectedHeap::heap(); | |
828 assert(gch->kind() == CollectedHeap::GenCollectedHeap, | |
829 "not a generational heap"); | |
4095
bca17e38de00
6593758: RFE: Enhance GC ergonomics to dynamically choose ParallelGCThreads
jmasa
parents:
2369
diff
changeset
|
830 FlexibleWorkGang* workers = gch->workers(); |
0 | 831 assert(workers != NULL, "Need parallel worker threads."); |
4095
bca17e38de00
6593758: RFE: Enhance GC ergonomics to dynamically choose ParallelGCThreads
jmasa
parents:
2369
diff
changeset
|
832 _state_set.reset(workers->active_workers(), _generation.promotion_failed()); |
0 | 833 ParNewRefProcTaskProxy rp_task(task, _generation, *_generation.next_gen(), |
834 _generation.reserved().end(), _state_set); | |
835 workers->run_task(&rp_task); | |
4095
bca17e38de00
6593758: RFE: Enhance GC ergonomics to dynamically choose ParallelGCThreads
jmasa
parents:
2369
diff
changeset
|
836 _state_set.reset(0 /* bad value in debug if not reset */, |
bca17e38de00
6593758: RFE: Enhance GC ergonomics to dynamically choose ParallelGCThreads
jmasa
parents:
2369
diff
changeset
|
837 _generation.promotion_failed()); |
0 | 838 } |
839 | |
840 void ParNewRefProcTaskExecutor::execute(EnqueueTask& task) | |
841 { | |
842 GenCollectedHeap* gch = GenCollectedHeap::heap(); | |
4095
bca17e38de00
6593758: RFE: Enhance GC ergonomics to dynamically choose ParallelGCThreads
jmasa
parents:
2369
diff
changeset
|
843 FlexibleWorkGang* workers = gch->workers(); |
0 | 844 assert(workers != NULL, "Need parallel worker threads."); |
845 ParNewRefEnqueueTaskProxy enq_task(task); | |
846 workers->run_task(&enq_task); | |
847 } | |
848 | |
849 void ParNewRefProcTaskExecutor::set_single_threaded_mode() | |
850 { | |
851 _state_set.flush(); | |
852 GenCollectedHeap* gch = GenCollectedHeap::heap(); | |
853 gch->set_par_threads(0); // 0 ==> non-parallel. | |
854 gch->save_marks(); | |
855 } | |
856 | |
857 ScanClosureWithParBarrier:: | |
858 ScanClosureWithParBarrier(ParNewGeneration* g, bool gc_barrier) : | |
859 ScanClosure(g, gc_barrier) {} | |
860 | |
861 EvacuateFollowersClosureGeneral:: | |
862 EvacuateFollowersClosureGeneral(GenCollectedHeap* gch, int level, | |
863 OopsInGenClosure* cur, | |
864 OopsInGenClosure* older) : | |
865 _gch(gch), _level(level), | |
866 _scan_cur_or_nonheap(cur), _scan_older(older) | |
867 {} | |
868 | |
869 void EvacuateFollowersClosureGeneral::do_void() { | |
870 do { | |
871 // Beware: this call will lead to closure applications via virtual | |
872 // calls. | |
873 _gch->oop_since_save_marks_iterate(_level, | |
874 _scan_cur_or_nonheap, | |
875 _scan_older); | |
876 } while (!_gch->no_allocs_since_save_marks(_level)); | |
877 } | |
878 | |
879 | |
10405 | 880 // A Generation that does parallel young-gen collection. |
881 | |
0 | 882 bool ParNewGeneration::_avoid_promotion_undo = false; |
883 | |
10405 | 884 void ParNewGeneration::handle_promotion_failed(GenCollectedHeap* gch, ParScanThreadStateSet& thread_state_set, ParNewTracer& gc_tracer) { |
885 assert(_promo_failure_scan_stack.is_empty(), "post condition"); | |
886 _promo_failure_scan_stack.clear(true); // Clear cached segments. | |
887 | |
888 remove_forwarding_pointers(); | |
889 if (PrintGCDetails) { | |
890 gclog_or_tty->print(" (promotion failed)"); | |
891 } | |
892 // All the spaces are in play for mark-sweep. | |
893 swap_spaces(); // Make life simpler for CMS || rescan; see 6483690. | |
894 from()->set_next_compaction_space(to()); | |
895 gch->set_incremental_collection_failed(); | |
896 // Inform the next generation that a promotion failure occurred. | |
897 _next_gen->promotion_failure_occurred(); | |
898 | |
899 // Trace promotion failure in the parallel GC threads | |
900 thread_state_set.trace_promotion_failed(gc_tracer); | |
901 // Single threaded code may have reported promotion failure to the global state | |
902 if (_promotion_failed_info.has_failed()) { | |
903 gc_tracer.report_promotion_failed(_promotion_failed_info); | |
904 } | |
905 // Reset the PromotionFailureALot counters. | |
906 NOT_PRODUCT(Universe::heap()->reset_promotion_should_fail();) | |
907 } | |
0 | 908 |
909 void ParNewGeneration::collect(bool full, | |
910 bool clear_all_soft_refs, | |
911 size_t size, | |
912 bool is_tlab) { | |
913 assert(full || size > 0, "otherwise we don't want to collect"); | |
10405 | 914 |
0 | 915 GenCollectedHeap* gch = GenCollectedHeap::heap(); |
10405 | 916 |
13400
86e6d691f2e1
8028128: Add a type safe alternative for working with counter based data
mgronlun
parents:
12316
diff
changeset
|
917 _gc_timer->register_gc_start(); |
10405 | 918 |
0 | 919 assert(gch->kind() == CollectedHeap::GenCollectedHeap, |
920 "not a CMS generational heap"); | |
921 AdaptiveSizePolicy* size_policy = gch->gen_policy()->size_policy(); | |
4095
bca17e38de00
6593758: RFE: Enhance GC ergonomics to dynamically choose ParallelGCThreads
jmasa
parents:
2369
diff
changeset
|
922 FlexibleWorkGang* workers = gch->workers(); |
bca17e38de00
6593758: RFE: Enhance GC ergonomics to dynamically choose ParallelGCThreads
jmasa
parents:
2369
diff
changeset
|
923 assert(workers != NULL, "Need workgang for parallel work"); |
bca17e38de00
6593758: RFE: Enhance GC ergonomics to dynamically choose ParallelGCThreads
jmasa
parents:
2369
diff
changeset
|
924 int active_workers = |
bca17e38de00
6593758: RFE: Enhance GC ergonomics to dynamically choose ParallelGCThreads
jmasa
parents:
2369
diff
changeset
|
925 AdaptiveSizePolicy::calc_active_workers(workers->total_workers(), |
bca17e38de00
6593758: RFE: Enhance GC ergonomics to dynamically choose ParallelGCThreads
jmasa
parents:
2369
diff
changeset
|
926 workers->active_workers(), |
bca17e38de00
6593758: RFE: Enhance GC ergonomics to dynamically choose ParallelGCThreads
jmasa
parents:
2369
diff
changeset
|
927 Threads::number_of_non_daemon_threads()); |
bca17e38de00
6593758: RFE: Enhance GC ergonomics to dynamically choose ParallelGCThreads
jmasa
parents:
2369
diff
changeset
|
928 workers->set_active_workers(active_workers); |
0 | 929 assert(gch->n_gens() == 2, |
930 "Par collection currently only works with single older gen."); | |
12030
330dfb0476f4
8022800: Use specific generations rather than generation iteration
brutisso
parents:
10405
diff
changeset
|
931 _next_gen = gch->next_gen(this); |
0 | 932 // Do we have to avoid promotion_undo? |
933 if (gch->collector_policy()->is_concurrent_mark_sweep_policy()) { | |
934 set_avoid_promotion_undo(true); | |
935 } | |
936 | |
10405 | 937 // If the next generation is too full to accommodate worst-case promotion |
0 | 938 // from this generation, pass on collection; let the next generation |
939 // do it. | |
940 if (!collection_attempt_is_safe()) { | |
1888
a7214d79fcf1
6896603: CMS/GCH: collection_attempt_is_safe() ergo should use more recent data
ysr
parents:
1836
diff
changeset
|
941 gch->set_incremental_collection_failed(); // slight lie, in that we did not even attempt one |
0 | 942 return; |
943 } | |
944 assert(to()->is_empty(), "Else not collection_attempt_is_safe"); | |
945 | |
10405 | 946 ParNewTracer gc_tracer; |
947 gc_tracer.report_gc_start(gch->gc_cause(), _gc_timer->gc_start()); | |
948 gch->trace_heap_before_gc(&gc_tracer); | |
949 | |
0 | 950 init_assuming_no_promotion_failure(); |
951 | |
952 if (UseAdaptiveSizePolicy) { | |
953 set_survivor_overflow(false); | |
954 size_policy->minor_collection_begin(); | |
955 } | |
956 | |
10405 | 957 GCTraceTime t1(GCCauseString("GC", gch->gc_cause()), PrintGC && !PrintGCDetails, true, NULL); |
0 | 958 // Capture heap used before collection (for printing). |
959 size_t gch_prev_used = gch->used(); | |
960 | |
961 SpecializationStats::clear(); | |
962 | |
963 age_table()->clear(); | |
263
12eea04c8b06
6672698: mangle_unused_area() should not remangle the entire heap at each collection.
jmasa
parents:
167
diff
changeset
|
964 to()->clear(SpaceDecorator::Mangle); |
0 | 965 |
966 gch->save_marks(); | |
967 assert(workers != NULL, "Need parallel worker threads."); | |
4095
bca17e38de00
6593758: RFE: Enhance GC ergonomics to dynamically choose ParallelGCThreads
jmasa
parents:
2369
diff
changeset
|
968 int n_workers = active_workers; |
bca17e38de00
6593758: RFE: Enhance GC ergonomics to dynamically choose ParallelGCThreads
jmasa
parents:
2369
diff
changeset
|
969 |
bca17e38de00
6593758: RFE: Enhance GC ergonomics to dynamically choose ParallelGCThreads
jmasa
parents:
2369
diff
changeset
|
970 // Set the correct parallelism (number of queues) in the reference processor |
bca17e38de00
6593758: RFE: Enhance GC ergonomics to dynamically choose ParallelGCThreads
jmasa
parents:
2369
diff
changeset
|
971 ref_processor()->set_active_mt_degree(n_workers); |
bca17e38de00
6593758: RFE: Enhance GC ergonomics to dynamically choose ParallelGCThreads
jmasa
parents:
2369
diff
changeset
|
972 |
bca17e38de00
6593758: RFE: Enhance GC ergonomics to dynamically choose ParallelGCThreads
jmasa
parents:
2369
diff
changeset
|
973 // Always set the terminator for the active number of workers |
bca17e38de00
6593758: RFE: Enhance GC ergonomics to dynamically choose ParallelGCThreads
jmasa
parents:
2369
diff
changeset
|
974 // because only those workers go through the termination protocol. |
bca17e38de00
6593758: RFE: Enhance GC ergonomics to dynamically choose ParallelGCThreads
jmasa
parents:
2369
diff
changeset
|
975 ParallelTaskTerminator _term(n_workers, task_queues()); |
bca17e38de00
6593758: RFE: Enhance GC ergonomics to dynamically choose ParallelGCThreads
jmasa
parents:
2369
diff
changeset
|
976 ParScanThreadStateSet thread_state_set(workers->active_workers(), |
0 | 977 *to(), *this, *_next_gen, *task_queues(), |
695 | 978 _overflow_stacks, desired_plab_sz(), _term); |
0 | 979 |
980 ParNewGenTask tsk(this, _next_gen, reserved().end(), &thread_state_set); | |
981 gch->set_par_threads(n_workers); | |
982 gch->rem_set()->prepare_for_younger_refs_iterate(true); | |
983 // It turns out that even when we're using 1 thread, doing the work in a | |
984 // separate thread causes wide variance in run times. We can't help this | |
985 // in the multi-threaded case, but we special-case n=1 here to get | |
986 // repeatable measurements of the 1-thread overhead of the parallel code. | |
987 if (n_workers > 1) { | |
989
148e5441d916
6863023: need non-perm oops in code cache for JSR 292
jrose
parents:
695
diff
changeset
|
988 GenCollectedHeap::StrongRootsScope srs(gch); |
0 | 989 workers->run_task(&tsk); |
990 } else { | |
989
148e5441d916
6863023: need non-perm oops in code cache for JSR 292
jrose
parents:
695
diff
changeset
|
991 GenCollectedHeap::StrongRootsScope srs(gch); |
0 | 992 tsk.work(0); |
993 } | |
4095
bca17e38de00
6593758: RFE: Enhance GC ergonomics to dynamically choose ParallelGCThreads
jmasa
parents:
2369
diff
changeset
|
994 thread_state_set.reset(0 /* Bad value in debug if not reset */, |
bca17e38de00
6593758: RFE: Enhance GC ergonomics to dynamically choose ParallelGCThreads
jmasa
parents:
2369
diff
changeset
|
995 promotion_failed()); |
0 | 996 |
997 // Process (weak) reference objects found during scavenge. | |
453
c96030fff130
6684579: SoftReference processing can be made more efficient
ysr
parents:
269
diff
changeset
|
998 ReferenceProcessor* rp = ref_processor(); |
0 | 999 IsAliveClosure is_alive(this); |
1000 ScanWeakRefClosure scan_weak_ref(this); | |
1001 KeepAliveClosure keep_alive(&scan_weak_ref); | |
1002 ScanClosure scan_without_gc_barrier(this, false); | |
1003 ScanClosureWithParBarrier scan_with_gc_barrier(this, true); | |
1004 set_promo_failure_scan_stack_closure(&scan_without_gc_barrier); | |
1005 EvacuateFollowersClosureGeneral evacuate_followers(gch, _level, | |
1006 &scan_without_gc_barrier, &scan_with_gc_barrier); | |
457
27a80744a83b
6778647: snap(), snap_policy() should be renamed setup(), setup_policy()
ysr
parents:
454
diff
changeset
|
1007 rp->setup_policy(clear_all_soft_refs); |
4095
bca17e38de00
6593758: RFE: Enhance GC ergonomics to dynamically choose ParallelGCThreads
jmasa
parents:
2369
diff
changeset
|
1008 // Can the mt_degree be set later (at run_task() time would be best)? |
bca17e38de00
6593758: RFE: Enhance GC ergonomics to dynamically choose ParallelGCThreads
jmasa
parents:
2369
diff
changeset
|
1009 rp->set_active_mt_degree(active_workers); |
10405 | 1010 ReferenceProcessorStats stats; |
453
c96030fff130
6684579: SoftReference processing can be made more efficient
ysr
parents:
269
diff
changeset
|
1011 if (rp->processing_is_mt()) { |
0 | 1012 ParNewRefProcTaskExecutor task_executor(*this, thread_state_set); |
10405 | 1013 stats = rp->process_discovered_references(&is_alive, &keep_alive, |
1014 &evacuate_followers, &task_executor, | |
1015 _gc_timer); | |
0 | 1016 } else { |
1017 thread_state_set.flush(); | |
1018 gch->set_par_threads(0); // 0 ==> non-parallel. | |
1019 gch->save_marks(); | |
10405 | 1020 stats = rp->process_discovered_references(&is_alive, &keep_alive, |
1021 &evacuate_followers, NULL, | |
1022 _gc_timer); | |
0 | 1023 } |
10405 | 1024 gc_tracer.report_gc_reference_stats(stats); |
0 | 1025 if (!promotion_failed()) { |
1026 // Swap the survivor spaces. | |
263
12eea04c8b06
6672698: mangle_unused_area() should not remangle the entire heap at each collection.
jmasa
parents:
167
diff
changeset
|
1027 eden()->clear(SpaceDecorator::Mangle); |
12eea04c8b06
6672698: mangle_unused_area() should not remangle the entire heap at each collection.
jmasa
parents:
167
diff
changeset
|
1028 from()->clear(SpaceDecorator::Mangle); |
12eea04c8b06
6672698: mangle_unused_area() should not remangle the entire heap at each collection.
jmasa
parents:
167
diff
changeset
|
1029 if (ZapUnusedHeapArea) { |
12eea04c8b06
6672698: mangle_unused_area() should not remangle the entire heap at each collection.
jmasa
parents:
167
diff
changeset
|
1030 // This is now done here because of the piece-meal mangling which |
12eea04c8b06
6672698: mangle_unused_area() should not remangle the entire heap at each collection.
jmasa
parents:
167
diff
changeset
|
1031 // can check for valid mangling at intermediate points in the |
12eea04c8b06
6672698: mangle_unused_area() should not remangle the entire heap at each collection.
jmasa
parents:
167
diff
changeset
|
1032 // collection(s). When a minor collection fails to collect |
12eea04c8b06
6672698: mangle_unused_area() should not remangle the entire heap at each collection.
jmasa
parents:
167
diff
changeset
|
1033 // sufficient space resizing of the young generation can occur |
12eea04c8b06
6672698: mangle_unused_area() should not remangle the entire heap at each collection.
jmasa
parents:
167
diff
changeset
|
1034 // an redistribute the spaces in the young generation. Mangle |
12eea04c8b06
6672698: mangle_unused_area() should not remangle the entire heap at each collection.
jmasa
parents:
167
diff
changeset
|
1035 // here so that unzapped regions don't get distributed to |
12eea04c8b06
6672698: mangle_unused_area() should not remangle the entire heap at each collection.
jmasa
parents:
167
diff
changeset
|
1036 // other spaces. |
12eea04c8b06
6672698: mangle_unused_area() should not remangle the entire heap at each collection.
jmasa
parents:
167
diff
changeset
|
1037 to()->mangle_unused_area(); |
12eea04c8b06
6672698: mangle_unused_area() should not remangle the entire heap at each collection.
jmasa
parents:
167
diff
changeset
|
1038 } |
0 | 1039 swap_spaces(); |
1040 | |
1387
0bfd3fb24150
6858496: Clear all SoftReferences before an out-of-memory due to GC overhead limit.
jmasa
parents:
1145
diff
changeset
|
1041 // A successful scavenge should restart the GC time limit count which is |
0bfd3fb24150
6858496: Clear all SoftReferences before an out-of-memory due to GC overhead limit.
jmasa
parents:
1145
diff
changeset
|
1042 // for full GC's. |
0bfd3fb24150
6858496: Clear all SoftReferences before an out-of-memory due to GC overhead limit.
jmasa
parents:
1145
diff
changeset
|
1043 size_policy->reset_gc_overhead_limit_count(); |
0bfd3fb24150
6858496: Clear all SoftReferences before an out-of-memory due to GC overhead limit.
jmasa
parents:
1145
diff
changeset
|
1044 |
0 | 1045 assert(to()->is_empty(), "to space should be empty now"); |
7609
a30e7b564541
8005972: ParNew should not update the tenuring threshold when promotion failed has occurred
brutisso
parents:
7451
diff
changeset
|
1046 |
a30e7b564541
8005972: ParNew should not update the tenuring threshold when promotion failed has occurred
brutisso
parents:
7451
diff
changeset
|
1047 adjust_desired_tenuring_threshold(); |
0 | 1048 } else { |
10405 | 1049 handle_promotion_failed(gch, thread_state_set, gc_tracer); |
0 | 1050 } |
1051 // set new iteration safe limit for the survivor spaces | |
1052 from()->set_concurrent_iteration_safe_limit(from()->top()); | |
1053 to()->set_concurrent_iteration_safe_limit(to()->top()); | |
1054 | |
1055 if (ResizePLAB) { | |
6819 | 1056 plab_stats()->adjust_desired_plab_sz(n_workers); |
0 | 1057 } |
1058 | |
1059 if (PrintGC && !PrintGCDetails) { | |
1060 gch->print_heap_change(gch_prev_used); | |
1061 } | |
1062 | |
1712
2d6b74c9a797
6976378: ParNew: stats are printed unconditionally in debug builds
jcoomes
parents:
1710
diff
changeset
|
1063 if (PrintGCDetails && ParallelGCVerbose) { |
2d6b74c9a797
6976378: ParNew: stats are printed unconditionally in debug builds
jcoomes
parents:
1710
diff
changeset
|
1064 TASKQUEUE_STATS_ONLY(thread_state_set.print_termination_stats()); |
2d6b74c9a797
6976378: ParNew: stats are printed unconditionally in debug builds
jcoomes
parents:
1710
diff
changeset
|
1065 TASKQUEUE_STATS_ONLY(thread_state_set.print_taskqueue_stats()); |
2d6b74c9a797
6976378: ParNew: stats are printed unconditionally in debug builds
jcoomes
parents:
1710
diff
changeset
|
1066 } |
1710 | 1067 |
0 | 1068 if (UseAdaptiveSizePolicy) { |
1069 size_policy->minor_collection_end(gch->gc_cause()); | |
1070 size_policy->avg_survived()->sample(from()->used()); | |
1071 } | |
1072 | |
4911 | 1073 // We need to use a monotonically non-deccreasing time in ms |
1074 // or we will see time-warp warnings and os::javaTimeMillis() | |
1075 // does not guarantee monotonicity. | |
1076 jlong now = os::javaTimeNanos() / NANOSECS_PER_MILLISEC; | |
1077 update_time_of_last_gc(now); | |
0 | 1078 |
1079 SpecializationStats::print(); | |
1080 | |
453
c96030fff130
6684579: SoftReference processing can be made more efficient
ysr
parents:
269
diff
changeset
|
1081 rp->set_enqueuing_is_done(true); |
c96030fff130
6684579: SoftReference processing can be made more efficient
ysr
parents:
269
diff
changeset
|
1082 if (rp->processing_is_mt()) { |
0 | 1083 ParNewRefProcTaskExecutor task_executor(*this, thread_state_set); |
453
c96030fff130
6684579: SoftReference processing can be made more efficient
ysr
parents:
269
diff
changeset
|
1084 rp->enqueue_discovered_references(&task_executor); |
0 | 1085 } else { |
453
c96030fff130
6684579: SoftReference processing can be made more efficient
ysr
parents:
269
diff
changeset
|
1086 rp->enqueue_discovered_references(NULL); |
0 | 1087 } |
453
c96030fff130
6684579: SoftReference processing can be made more efficient
ysr
parents:
269
diff
changeset
|
1088 rp->verify_no_references_recorded(); |
10405 | 1089 |
1090 gch->trace_heap_after_gc(&gc_tracer); | |
1091 gc_tracer.report_tenuring_threshold(tenuring_threshold()); | |
1092 | |
13400
86e6d691f2e1
8028128: Add a type safe alternative for working with counter based data
mgronlun
parents:
12316
diff
changeset
|
1093 _gc_timer->register_gc_end(); |
10405 | 1094 |
1095 gc_tracer.report_gc_end(_gc_timer->gc_end(), _gc_timer->time_partitions()); | |
0 | 1096 } |
1097 | |
1098 static int sum; | |
1099 void ParNewGeneration::waste_some_time() { | |
1100 for (int i = 0; i < 100; i++) { | |
1101 sum += i; | |
1102 } | |
1103 } | |
1104 | |
12316
190899198332
7195622: CheckUnhandledOops has limited usefulness now
hseigel
parents:
12030
diff
changeset
|
1105 static const oop ClaimedForwardPtr = cast_to_oop<intptr_t>(0x4); |
0 | 1106 |
1107 // Because of concurrency, there are times where an object for which | |
1108 // "is_forwarded()" is true contains an "interim" forwarding pointer | |
1109 // value. Such a value will soon be overwritten with a real value. | |
1110 // This method requires "obj" to have a forwarding pointer, and waits, if | |
1111 // necessary for a real one to be inserted, and returns it. | |
1112 | |
1113 oop ParNewGeneration::real_forwardee(oop obj) { | |
1114 oop forward_ptr = obj->forwardee(); | |
1115 if (forward_ptr != ClaimedForwardPtr) { | |
1116 return forward_ptr; | |
1117 } else { | |
1118 return real_forwardee_slow(obj); | |
1119 } | |
1120 } | |
1121 | |
1122 oop ParNewGeneration::real_forwardee_slow(oop obj) { | |
1123 // Spin-read if it is claimed but not yet written by another thread. | |
1124 oop forward_ptr = obj->forwardee(); | |
1125 while (forward_ptr == ClaimedForwardPtr) { | |
1126 waste_some_time(); | |
1127 assert(obj->is_forwarded(), "precondition"); | |
1128 forward_ptr = obj->forwardee(); | |
1129 } | |
1130 return forward_ptr; | |
1131 } | |
1132 | |
1133 #ifdef ASSERT | |
1134 bool ParNewGeneration::is_legal_forward_ptr(oop p) { | |
1135 return | |
1136 (_avoid_promotion_undo && p == ClaimedForwardPtr) | |
1137 || Universe::heap()->is_in_reserved(p); | |
1138 } | |
1139 #endif | |
1140 | |
1141 void ParNewGeneration::preserve_mark_if_necessary(oop obj, markOop m) { | |
2038
74ee0db180fa
6807801: CMS: could save/restore fewer header words during scavenge
ysr
parents:
1972
diff
changeset
|
1142 if (m->must_be_preserved_for_promotion_failure(obj)) { |
74ee0db180fa
6807801: CMS: could save/restore fewer header words during scavenge
ysr
parents:
1972
diff
changeset
|
1143 // We should really have separate per-worker stacks, rather |
74ee0db180fa
6807801: CMS: could save/restore fewer header words during scavenge
ysr
parents:
1972
diff
changeset
|
1144 // than use locking of a common pair of stacks. |
0 | 1145 MutexLocker ml(ParGCRareEvent_lock); |
2038
74ee0db180fa
6807801: CMS: could save/restore fewer header words during scavenge
ysr
parents:
1972
diff
changeset
|
1146 preserve_mark(obj, m); |
0 | 1147 } |
1148 } | |
1149 | |
1150 // Multiple GC threads may try to promote an object. If the object | |
1151 // is successfully promoted, a forwarding pointer will be installed in | |
1152 // the object in the young generation. This method claims the right | |
1153 // to install the forwarding pointer before it copies the object, | |
1154 // thus avoiding the need to undo the copy as in | |
1155 // copy_to_survivor_space_avoiding_with_undo. | |
1156 | |
1157 oop ParNewGeneration::copy_to_survivor_space_avoiding_promotion_undo( | |
1158 ParScanThreadState* par_scan_state, oop old, size_t sz, markOop m) { | |
1159 // In the sequential version, this assert also says that the object is | |
1160 // not forwarded. That might not be the case here. It is the case that | |
1161 // the caller observed it to be not forwarded at some time in the past. | |
1162 assert(is_in_reserved(old), "shouldn't be scavenging this oop"); | |
1163 | |
1164 // The sequential code read "old->age()" below. That doesn't work here, | |
1165 // since the age is in the mark word, and that might be overwritten with | |
1166 // a forwarding pointer by a parallel thread. So we must save the mark | |
1167 // word in a local and then analyze it. | |
1168 oopDesc dummyOld; | |
1169 dummyOld.set_mark(m); | |
1170 assert(!dummyOld.is_forwarded(), | |
1171 "should not be called with forwarding pointer mark word."); | |
1172 | |
1173 oop new_obj = NULL; | |
1174 oop forward_ptr; | |
1175 | |
1176 // Try allocating obj in to-space (unless too old) | |
1177 if (dummyOld.age() < tenuring_threshold()) { | |
1178 new_obj = (oop)par_scan_state->alloc_in_to_space(sz); | |
1179 if (new_obj == NULL) { | |
1180 set_survivor_overflow(true); | |
1181 } | |
1182 } | |
1183 | |
1184 if (new_obj == NULL) { | |
1185 // Either to-space is full or we decided to promote | |
1186 // try allocating obj tenured | |
1187 | |
1188 // Attempt to install a null forwarding pointer (atomically), | |
1189 // to claim the right to install the real forwarding pointer. | |
1190 forward_ptr = old->forward_to_atomic(ClaimedForwardPtr); | |
1191 if (forward_ptr != NULL) { | |
1192 // someone else beat us to it. | |
1193 return real_forwardee(old); | |
1194 } | |
1195 | |
1196 new_obj = _next_gen->par_promote(par_scan_state->thread_num(), | |
1197 old, m, sz); | |
1198 | |
1199 if (new_obj == NULL) { | |
1200 // promotion failed, forward to self | |
1201 _promotion_failed = true; | |
1202 new_obj = old; | |
1203 | |
1204 preserve_mark_if_necessary(old, m); | |
10405 | 1205 par_scan_state->register_promotion_failure(sz); |
0 | 1206 } |
1207 | |
1208 old->forward_to(new_obj); | |
1209 forward_ptr = NULL; | |
1210 } else { | |
1211 // Is in to-space; do copying ourselves. | |
1212 Copy::aligned_disjoint_words((HeapWord*)old, (HeapWord*)new_obj, sz); | |
1213 forward_ptr = old->forward_to_atomic(new_obj); | |
1214 // Restore the mark word copied above. | |
1215 new_obj->set_mark(m); | |
1216 // Increment age if obj still in new generation | |
1217 new_obj->incr_age(); | |
1218 par_scan_state->age_table()->add(new_obj, sz); | |
1219 } | |
1220 assert(new_obj != NULL, "just checking"); | |
1221 | |
6725
da91efe96a93
6964458: Reimplement class meta-data storage to use native memory
coleenp
parents:
6595
diff
changeset
|
1222 #ifndef PRODUCT |
da91efe96a93
6964458: Reimplement class meta-data storage to use native memory
coleenp
parents:
6595
diff
changeset
|
1223 // This code must come after the CAS test, or it will print incorrect |
da91efe96a93
6964458: Reimplement class meta-data storage to use native memory
coleenp
parents:
6595
diff
changeset
|
1224 // information. |
da91efe96a93
6964458: Reimplement class meta-data storage to use native memory
coleenp
parents:
6595
diff
changeset
|
1225 if (TraceScavenge) { |
da91efe96a93
6964458: Reimplement class meta-data storage to use native memory
coleenp
parents:
6595
diff
changeset
|
1226 gclog_or_tty->print_cr("{%s %s " PTR_FORMAT " -> " PTR_FORMAT " (%d)}", |
da91efe96a93
6964458: Reimplement class meta-data storage to use native memory
coleenp
parents:
6595
diff
changeset
|
1227 is_in_reserved(new_obj) ? "copying" : "tenuring", |
12316
190899198332
7195622: CheckUnhandledOops has limited usefulness now
hseigel
parents:
12030
diff
changeset
|
1228 new_obj->klass()->internal_name(), (void *)old, (void *)new_obj, new_obj->size()); |
6725
da91efe96a93
6964458: Reimplement class meta-data storage to use native memory
coleenp
parents:
6595
diff
changeset
|
1229 } |
da91efe96a93
6964458: Reimplement class meta-data storage to use native memory
coleenp
parents:
6595
diff
changeset
|
1230 #endif |
da91efe96a93
6964458: Reimplement class meta-data storage to use native memory
coleenp
parents:
6595
diff
changeset
|
1231 |
0 | 1232 if (forward_ptr == NULL) { |
1233 oop obj_to_push = new_obj; | |
1234 if (par_scan_state->should_be_partially_scanned(obj_to_push, old)) { | |
1235 // Length field used as index of next element to be scanned. | |
1236 // Real length can be obtained from real_forwardee() | |
1237 arrayOop(old)->set_length(0); | |
1238 obj_to_push = old; | |
1239 assert(obj_to_push->is_forwarded() && obj_to_push->forwardee() != obj_to_push, | |
1240 "push forwarded object"); | |
1241 } | |
1242 // Push it on one of the queues of to-be-scanned objects. | |
534 | 1243 bool simulate_overflow = false; |
1244 NOT_PRODUCT( | |
1245 if (ParGCWorkQueueOverflowALot && should_simulate_overflow()) { | |
1246 // simulate a stack overflow | |
1247 simulate_overflow = true; | |
1248 } | |
1249 ) | |
1250 if (simulate_overflow || !par_scan_state->work_queue()->push(obj_to_push)) { | |
0 | 1251 // Add stats for overflow pushes. |
1252 if (Verbose && PrintGCDetails) { | |
1253 gclog_or_tty->print("queue overflow!\n"); | |
1254 } | |
534 | 1255 push_on_overflow_list(old, par_scan_state); |
1710 | 1256 TASKQUEUE_STATS_ONLY(par_scan_state->taskqueue_stats().record_overflow(0)); |
0 | 1257 } |
1258 | |
1259 return new_obj; | |
1260 } | |
1261 | |
1262 // Oops. Someone beat us to it. Undo the allocation. Where did we | |
1263 // allocate it? | |
1264 if (is_in_reserved(new_obj)) { | |
1265 // Must be in to_space. | |
1266 assert(to()->is_in_reserved(new_obj), "Checking"); | |
1267 if (forward_ptr == ClaimedForwardPtr) { | |
1268 // Wait to get the real forwarding pointer value. | |
1269 forward_ptr = real_forwardee(old); | |
1270 } | |
1271 par_scan_state->undo_alloc_in_to_space((HeapWord*)new_obj, sz); | |
1272 } | |
1273 | |
1274 return forward_ptr; | |
1275 } | |
1276 | |
1277 | |
1278 // Multiple GC threads may try to promote the same object. If two | |
1279 // or more GC threads copy the object, only one wins the race to install | |
1280 // the forwarding pointer. The other threads have to undo their copy. | |
1281 | |
1282 oop ParNewGeneration::copy_to_survivor_space_with_undo( | |
1283 ParScanThreadState* par_scan_state, oop old, size_t sz, markOop m) { | |
1284 | |
1285 // In the sequential version, this assert also says that the object is | |
1286 // not forwarded. That might not be the case here. It is the case that | |
1287 // the caller observed it to be not forwarded at some time in the past. | |
1288 assert(is_in_reserved(old), "shouldn't be scavenging this oop"); | |
1289 | |
1290 // The sequential code read "old->age()" below. That doesn't work here, | |
1291 // since the age is in the mark word, and that might be overwritten with | |
1292 // a forwarding pointer by a parallel thread. So we must save the mark | |
1293 // word here, install it in a local oopDesc, and then analyze it. | |
1294 oopDesc dummyOld; | |
1295 dummyOld.set_mark(m); | |
1296 assert(!dummyOld.is_forwarded(), | |
1297 "should not be called with forwarding pointer mark word."); | |
1298 | |
1299 bool failed_to_promote = false; | |
1300 oop new_obj = NULL; | |
1301 oop forward_ptr; | |
1302 | |
1303 // Try allocating obj in to-space (unless too old) | |
1304 if (dummyOld.age() < tenuring_threshold()) { | |
1305 new_obj = (oop)par_scan_state->alloc_in_to_space(sz); | |
1306 if (new_obj == NULL) { | |
1307 set_survivor_overflow(true); | |
1308 } | |
1309 } | |
1310 | |
1311 if (new_obj == NULL) { | |
1312 // Either to-space is full or we decided to promote | |
1313 // try allocating obj tenured | |
1314 new_obj = _next_gen->par_promote(par_scan_state->thread_num(), | |
1315 old, m, sz); | |
1316 | |
1317 if (new_obj == NULL) { | |
1318 // promotion failed, forward to self | |
1319 forward_ptr = old->forward_to_atomic(old); | |
1320 new_obj = old; | |
1321 | |
1322 if (forward_ptr != NULL) { | |
1323 return forward_ptr; // someone else succeeded | |
1324 } | |
1325 | |
1326 _promotion_failed = true; | |
1327 failed_to_promote = true; | |
1328 | |
1329 preserve_mark_if_necessary(old, m); | |
10405 | 1330 par_scan_state->register_promotion_failure(sz); |
0 | 1331 } |
1332 } else { | |
1333 // Is in to-space; do copying ourselves. | |
1334 Copy::aligned_disjoint_words((HeapWord*)old, (HeapWord*)new_obj, sz); | |
1335 // Restore the mark word copied above. | |
1336 new_obj->set_mark(m); | |
1337 // Increment age if new_obj still in new generation | |
1338 new_obj->incr_age(); | |
1339 par_scan_state->age_table()->add(new_obj, sz); | |
1340 } | |
1341 assert(new_obj != NULL, "just checking"); | |
1342 | |
6725
da91efe96a93
6964458: Reimplement class meta-data storage to use native memory
coleenp
parents:
6595
diff
changeset
|
1343 #ifndef PRODUCT |
da91efe96a93
6964458: Reimplement class meta-data storage to use native memory
coleenp
parents:
6595
diff
changeset
|
1344 // This code must come after the CAS test, or it will print incorrect |
da91efe96a93
6964458: Reimplement class meta-data storage to use native memory
coleenp
parents:
6595
diff
changeset
|
1345 // information. |
da91efe96a93
6964458: Reimplement class meta-data storage to use native memory
coleenp
parents:
6595
diff
changeset
|
1346 if (TraceScavenge) { |
da91efe96a93
6964458: Reimplement class meta-data storage to use native memory
coleenp
parents:
6595
diff
changeset
|
1347 gclog_or_tty->print_cr("{%s %s " PTR_FORMAT " -> " PTR_FORMAT " (%d)}", |
da91efe96a93
6964458: Reimplement class meta-data storage to use native memory
coleenp
parents:
6595
diff
changeset
|
1348 is_in_reserved(new_obj) ? "copying" : "tenuring", |
12316
190899198332
7195622: CheckUnhandledOops has limited usefulness now
hseigel
parents:
12030
diff
changeset
|
1349 new_obj->klass()->internal_name(), (void *)old, (void *)new_obj, new_obj->size()); |
6725
da91efe96a93
6964458: Reimplement class meta-data storage to use native memory
coleenp
parents:
6595
diff
changeset
|
1350 } |
da91efe96a93
6964458: Reimplement class meta-data storage to use native memory
coleenp
parents:
6595
diff
changeset
|
1351 #endif |
da91efe96a93
6964458: Reimplement class meta-data storage to use native memory
coleenp
parents:
6595
diff
changeset
|
1352 |
0 | 1353 // Now attempt to install the forwarding pointer (atomically). |
1354 // We have to copy the mark word before overwriting with forwarding | |
1355 // ptr, so we can restore it below in the copy. | |
1356 if (!failed_to_promote) { | |
1357 forward_ptr = old->forward_to_atomic(new_obj); | |
1358 } | |
1359 | |
1360 if (forward_ptr == NULL) { | |
1361 oop obj_to_push = new_obj; | |
1362 if (par_scan_state->should_be_partially_scanned(obj_to_push, old)) { | |
1363 // Length field used as index of next element to be scanned. | |
1364 // Real length can be obtained from real_forwardee() | |
1365 arrayOop(old)->set_length(0); | |
1366 obj_to_push = old; | |
1367 assert(obj_to_push->is_forwarded() && obj_to_push->forwardee() != obj_to_push, | |
1368 "push forwarded object"); | |
1369 } | |
1370 // Push it on one of the queues of to-be-scanned objects. | |
534 | 1371 bool simulate_overflow = false; |
1372 NOT_PRODUCT( | |
1373 if (ParGCWorkQueueOverflowALot && should_simulate_overflow()) { | |
1374 // simulate a stack overflow | |
1375 simulate_overflow = true; | |
1376 } | |
1377 ) | |
1378 if (simulate_overflow || !par_scan_state->work_queue()->push(obj_to_push)) { | |
0 | 1379 // Add stats for overflow pushes. |
534 | 1380 push_on_overflow_list(old, par_scan_state); |
1710 | 1381 TASKQUEUE_STATS_ONLY(par_scan_state->taskqueue_stats().record_overflow(0)); |
0 | 1382 } |
1383 | |
1384 return new_obj; | |
1385 } | |
1386 | |
1387 // Oops. Someone beat us to it. Undo the allocation. Where did we | |
1388 // allocate it? | |
1389 if (is_in_reserved(new_obj)) { | |
1390 // Must be in to_space. | |
1391 assert(to()->is_in_reserved(new_obj), "Checking"); | |
1392 par_scan_state->undo_alloc_in_to_space((HeapWord*)new_obj, sz); | |
1393 } else { | |
1394 assert(!_avoid_promotion_undo, "Should not be here if avoiding."); | |
1395 _next_gen->par_promote_alloc_undo(par_scan_state->thread_num(), | |
1396 (HeapWord*)new_obj, sz); | |
1397 } | |
1398 | |
1399 return forward_ptr; | |
1400 } | |
1401 | |
534 | 1402 #ifndef PRODUCT |
1403 // It's OK to call this multi-threaded; the worst thing | |
1404 // that can happen is that we'll get a bunch of closely | |
1405 // spaced simulated oveflows, but that's OK, in fact | |
1406 // probably good as it would exercise the overflow code | |
1407 // under contention. | |
1408 bool ParNewGeneration::should_simulate_overflow() { | |
1409 if (_overflow_counter-- <= 0) { // just being defensive | |
1410 _overflow_counter = ParGCWorkQueueOverflowInterval; | |
1411 return true; | |
1412 } else { | |
1413 return false; | |
1414 } | |
1415 } | |
1416 #endif | |
1417 | |
679
cea947c8a988
6819891: ParNew: Fix work queue overflow code to deal correctly with +UseCompressedOops
ysr
parents:
579
diff
changeset
|
1418 // In case we are using compressed oops, we need to be careful. |
cea947c8a988
6819891: ParNew: Fix work queue overflow code to deal correctly with +UseCompressedOops
ysr
parents:
579
diff
changeset
|
1419 // If the object being pushed is an object array, then its length |
cea947c8a988
6819891: ParNew: Fix work queue overflow code to deal correctly with +UseCompressedOops
ysr
parents:
579
diff
changeset
|
1420 // field keeps track of the "grey boundary" at which the next |
cea947c8a988
6819891: ParNew: Fix work queue overflow code to deal correctly with +UseCompressedOops
ysr
parents:
579
diff
changeset
|
1421 // incremental scan will be done (see ParGCArrayScanChunk). |
cea947c8a988
6819891: ParNew: Fix work queue overflow code to deal correctly with +UseCompressedOops
ysr
parents:
579
diff
changeset
|
1422 // When using compressed oops, this length field is kept in the |
cea947c8a988
6819891: ParNew: Fix work queue overflow code to deal correctly with +UseCompressedOops
ysr
parents:
579
diff
changeset
|
1423 // lower 32 bits of the erstwhile klass word and cannot be used |
cea947c8a988
6819891: ParNew: Fix work queue overflow code to deal correctly with +UseCompressedOops
ysr
parents:
579
diff
changeset
|
1424 // for the overflow chaining pointer (OCP below). As such the OCP |
cea947c8a988
6819891: ParNew: Fix work queue overflow code to deal correctly with +UseCompressedOops
ysr
parents:
579
diff
changeset
|
1425 // would itself need to be compressed into the top 32-bits in this |
cea947c8a988
6819891: ParNew: Fix work queue overflow code to deal correctly with +UseCompressedOops
ysr
parents:
579
diff
changeset
|
1426 // case. Unfortunately, see below, in the event that we have a |
cea947c8a988
6819891: ParNew: Fix work queue overflow code to deal correctly with +UseCompressedOops
ysr
parents:
579
diff
changeset
|
1427 // promotion failure, the node to be pushed on the list can be |
cea947c8a988
6819891: ParNew: Fix work queue overflow code to deal correctly with +UseCompressedOops
ysr
parents:
579
diff
changeset
|
1428 // outside of the Java heap, so the heap-based pointer compression |
cea947c8a988
6819891: ParNew: Fix work queue overflow code to deal correctly with +UseCompressedOops
ysr
parents:
579
diff
changeset
|
1429 // would not work (we would have potential aliasing between C-heap |
cea947c8a988
6819891: ParNew: Fix work queue overflow code to deal correctly with +UseCompressedOops
ysr
parents:
579
diff
changeset
|
1430 // and Java-heap pointers). For this reason, when using compressed |
cea947c8a988
6819891: ParNew: Fix work queue overflow code to deal correctly with +UseCompressedOops
ysr
parents:
579
diff
changeset
|
1431 // oops, we simply use a worker-thread-local, non-shared overflow |
cea947c8a988
6819891: ParNew: Fix work queue overflow code to deal correctly with +UseCompressedOops
ysr
parents:
579
diff
changeset
|
1432 // list in the form of a growable array, with a slightly different |
cea947c8a988
6819891: ParNew: Fix work queue overflow code to deal correctly with +UseCompressedOops
ysr
parents:
579
diff
changeset
|
1433 // overflow stack draining strategy. If/when we start using fat |
cea947c8a988
6819891: ParNew: Fix work queue overflow code to deal correctly with +UseCompressedOops
ysr
parents:
579
diff
changeset
|
1434 // stacks here, we can go back to using (fat) pointer chains |
cea947c8a988
6819891: ParNew: Fix work queue overflow code to deal correctly with +UseCompressedOops
ysr
parents:
579
diff
changeset
|
1435 // (although some performance comparisons would be useful since |
cea947c8a988
6819891: ParNew: Fix work queue overflow code to deal correctly with +UseCompressedOops
ysr
parents:
579
diff
changeset
|
1436 // single global lists have their own performance disadvantages |
cea947c8a988
6819891: ParNew: Fix work queue overflow code to deal correctly with +UseCompressedOops
ysr
parents:
579
diff
changeset
|
1437 // as we were made painfully aware not long ago, see 6786503). |
12316
190899198332
7195622: CheckUnhandledOops has limited usefulness now
hseigel
parents:
12030
diff
changeset
|
1438 #define BUSY (cast_to_oop<intptr_t>(0x1aff1aff)) |
534 | 1439 void ParNewGeneration::push_on_overflow_list(oop from_space_obj, ParScanThreadState* par_scan_state) { |
679
cea947c8a988
6819891: ParNew: Fix work queue overflow code to deal correctly with +UseCompressedOops
ysr
parents:
579
diff
changeset
|
1440 assert(is_in_reserved(from_space_obj), "Should be from this generation"); |
695 | 1441 if (ParGCUseLocalOverflow) { |
679
cea947c8a988
6819891: ParNew: Fix work queue overflow code to deal correctly with +UseCompressedOops
ysr
parents:
579
diff
changeset
|
1442 // In the case of compressed oops, we use a private, not-shared |
cea947c8a988
6819891: ParNew: Fix work queue overflow code to deal correctly with +UseCompressedOops
ysr
parents:
579
diff
changeset
|
1443 // overflow stack. |
cea947c8a988
6819891: ParNew: Fix work queue overflow code to deal correctly with +UseCompressedOops
ysr
parents:
579
diff
changeset
|
1444 par_scan_state->push_on_overflow_stack(from_space_obj); |
cea947c8a988
6819891: ParNew: Fix work queue overflow code to deal correctly with +UseCompressedOops
ysr
parents:
579
diff
changeset
|
1445 } else { |
695 | 1446 assert(!UseCompressedOops, "Error"); |
679
cea947c8a988
6819891: ParNew: Fix work queue overflow code to deal correctly with +UseCompressedOops
ysr
parents:
579
diff
changeset
|
1447 // if the object has been forwarded to itself, then we cannot |
cea947c8a988
6819891: ParNew: Fix work queue overflow code to deal correctly with +UseCompressedOops
ysr
parents:
579
diff
changeset
|
1448 // use the klass pointer for the linked list. Instead we have |
cea947c8a988
6819891: ParNew: Fix work queue overflow code to deal correctly with +UseCompressedOops
ysr
parents:
579
diff
changeset
|
1449 // to allocate an oopDesc in the C-Heap and use that for the linked list. |
cea947c8a988
6819891: ParNew: Fix work queue overflow code to deal correctly with +UseCompressedOops
ysr
parents:
579
diff
changeset
|
1450 // XXX This is horribly inefficient when a promotion failure occurs |
cea947c8a988
6819891: ParNew: Fix work queue overflow code to deal correctly with +UseCompressedOops
ysr
parents:
579
diff
changeset
|
1451 // and should be fixed. XXX FIX ME !!! |
534 | 1452 #ifndef PRODUCT |
679
cea947c8a988
6819891: ParNew: Fix work queue overflow code to deal correctly with +UseCompressedOops
ysr
parents:
579
diff
changeset
|
1453 Atomic::inc_ptr(&_num_par_pushes); |
cea947c8a988
6819891: ParNew: Fix work queue overflow code to deal correctly with +UseCompressedOops
ysr
parents:
579
diff
changeset
|
1454 assert(_num_par_pushes > 0, "Tautology"); |
534 | 1455 #endif |
679
cea947c8a988
6819891: ParNew: Fix work queue overflow code to deal correctly with +UseCompressedOops
ysr
parents:
579
diff
changeset
|
1456 if (from_space_obj->forwardee() == from_space_obj) { |
6197 | 1457 oopDesc* listhead = NEW_C_HEAP_ARRAY(oopDesc, 1, mtGC); |
679
cea947c8a988
6819891: ParNew: Fix work queue overflow code to deal correctly with +UseCompressedOops
ysr
parents:
579
diff
changeset
|
1458 listhead->forward_to(from_space_obj); |
cea947c8a988
6819891: ParNew: Fix work queue overflow code to deal correctly with +UseCompressedOops
ysr
parents:
579
diff
changeset
|
1459 from_space_obj = listhead; |
cea947c8a988
6819891: ParNew: Fix work queue overflow code to deal correctly with +UseCompressedOops
ysr
parents:
579
diff
changeset
|
1460 } |
cea947c8a988
6819891: ParNew: Fix work queue overflow code to deal correctly with +UseCompressedOops
ysr
parents:
579
diff
changeset
|
1461 oop observed_overflow_list = _overflow_list; |
cea947c8a988
6819891: ParNew: Fix work queue overflow code to deal correctly with +UseCompressedOops
ysr
parents:
579
diff
changeset
|
1462 oop cur_overflow_list; |
cea947c8a988
6819891: ParNew: Fix work queue overflow code to deal correctly with +UseCompressedOops
ysr
parents:
579
diff
changeset
|
1463 do { |
cea947c8a988
6819891: ParNew: Fix work queue overflow code to deal correctly with +UseCompressedOops
ysr
parents:
579
diff
changeset
|
1464 cur_overflow_list = observed_overflow_list; |
cea947c8a988
6819891: ParNew: Fix work queue overflow code to deal correctly with +UseCompressedOops
ysr
parents:
579
diff
changeset
|
1465 if (cur_overflow_list != BUSY) { |
cea947c8a988
6819891: ParNew: Fix work queue overflow code to deal correctly with +UseCompressedOops
ysr
parents:
579
diff
changeset
|
1466 from_space_obj->set_klass_to_list_ptr(cur_overflow_list); |
cea947c8a988
6819891: ParNew: Fix work queue overflow code to deal correctly with +UseCompressedOops
ysr
parents:
579
diff
changeset
|
1467 } else { |
cea947c8a988
6819891: ParNew: Fix work queue overflow code to deal correctly with +UseCompressedOops
ysr
parents:
579
diff
changeset
|
1468 from_space_obj->set_klass_to_list_ptr(NULL); |
cea947c8a988
6819891: ParNew: Fix work queue overflow code to deal correctly with +UseCompressedOops
ysr
parents:
579
diff
changeset
|
1469 } |
cea947c8a988
6819891: ParNew: Fix work queue overflow code to deal correctly with +UseCompressedOops
ysr
parents:
579
diff
changeset
|
1470 observed_overflow_list = |
cea947c8a988
6819891: ParNew: Fix work queue overflow code to deal correctly with +UseCompressedOops
ysr
parents:
579
diff
changeset
|
1471 (oop)Atomic::cmpxchg_ptr(from_space_obj, &_overflow_list, cur_overflow_list); |
cea947c8a988
6819891: ParNew: Fix work queue overflow code to deal correctly with +UseCompressedOops
ysr
parents:
579
diff
changeset
|
1472 } while (cur_overflow_list != observed_overflow_list); |
0 | 1473 } |
1474 } | |
1475 | |
679
cea947c8a988
6819891: ParNew: Fix work queue overflow code to deal correctly with +UseCompressedOops
ysr
parents:
579
diff
changeset
|
1476 bool ParNewGeneration::take_from_overflow_list(ParScanThreadState* par_scan_state) { |
cea947c8a988
6819891: ParNew: Fix work queue overflow code to deal correctly with +UseCompressedOops
ysr
parents:
579
diff
changeset
|
1477 bool res; |
cea947c8a988
6819891: ParNew: Fix work queue overflow code to deal correctly with +UseCompressedOops
ysr
parents:
579
diff
changeset
|
1478 |
695 | 1479 if (ParGCUseLocalOverflow) { |
679
cea947c8a988
6819891: ParNew: Fix work queue overflow code to deal correctly with +UseCompressedOops
ysr
parents:
579
diff
changeset
|
1480 res = par_scan_state->take_from_overflow_stack(); |
cea947c8a988
6819891: ParNew: Fix work queue overflow code to deal correctly with +UseCompressedOops
ysr
parents:
579
diff
changeset
|
1481 } else { |
695 | 1482 assert(!UseCompressedOops, "Error"); |
679
cea947c8a988
6819891: ParNew: Fix work queue overflow code to deal correctly with +UseCompressedOops
ysr
parents:
579
diff
changeset
|
1483 res = take_from_overflow_list_work(par_scan_state); |
cea947c8a988
6819891: ParNew: Fix work queue overflow code to deal correctly with +UseCompressedOops
ysr
parents:
579
diff
changeset
|
1484 } |
cea947c8a988
6819891: ParNew: Fix work queue overflow code to deal correctly with +UseCompressedOops
ysr
parents:
579
diff
changeset
|
1485 return res; |
cea947c8a988
6819891: ParNew: Fix work queue overflow code to deal correctly with +UseCompressedOops
ysr
parents:
579
diff
changeset
|
1486 } |
cea947c8a988
6819891: ParNew: Fix work queue overflow code to deal correctly with +UseCompressedOops
ysr
parents:
579
diff
changeset
|
1487 |
cea947c8a988
6819891: ParNew: Fix work queue overflow code to deal correctly with +UseCompressedOops
ysr
parents:
579
diff
changeset
|
1488 |
534 | 1489 // *NOTE*: The overflow list manipulation code here and |
1490 // in CMSCollector:: are very similar in shape, | |
1491 // except that in the CMS case we thread the objects | |
1492 // directly into the list via their mark word, and do | |
1493 // not need to deal with special cases below related | |
1494 // to chunking of object arrays and promotion failure | |
1495 // handling. | |
1496 // CR 6797058 has been filed to attempt consolidation of | |
1497 // the common code. | |
1498 // Because of the common code, if you make any changes in | |
1499 // the code below, please check the CMS version to see if | |
1500 // similar changes might be needed. | |
1501 // See CMSCollector::par_take_from_overflow_list() for | |
1502 // more extensive documentation comments. | |
679
cea947c8a988
6819891: ParNew: Fix work queue overflow code to deal correctly with +UseCompressedOops
ysr
parents:
579
diff
changeset
|
1503 bool ParNewGeneration::take_from_overflow_list_work(ParScanThreadState* par_scan_state) { |
0 | 1504 ObjToScanQueue* work_q = par_scan_state->work_queue(); |
1505 // How many to take? | |
679
cea947c8a988
6819891: ParNew: Fix work queue overflow code to deal correctly with +UseCompressedOops
ysr
parents:
579
diff
changeset
|
1506 size_t objsFromOverflow = MIN2((size_t)(work_q->max_elems() - work_q->size())/4, |
534 | 1507 (size_t)ParGCDesiredObjsFromOverflowList); |
0 | 1508 |
1836
894b1d7c7e01
6423256: GC stacks should use a better data structure
jcoomes
parents:
1833
diff
changeset
|
1509 assert(!UseCompressedOops, "Error"); |
679
cea947c8a988
6819891: ParNew: Fix work queue overflow code to deal correctly with +UseCompressedOops
ysr
parents:
579
diff
changeset
|
1510 assert(par_scan_state->overflow_stack() == NULL, "Error"); |
0 | 1511 if (_overflow_list == NULL) return false; |
1512 | |
1513 // Otherwise, there was something there; try claiming the list. | |
12316
190899198332
7195622: CheckUnhandledOops has limited usefulness now
hseigel
parents:
12030
diff
changeset
|
1514 oop prefix = cast_to_oop(Atomic::xchg_ptr(BUSY, &_overflow_list)); |
534 | 1515 // Trim off a prefix of at most objsFromOverflow items |
1516 Thread* tid = Thread::current(); | |
1517 size_t spin_count = (size_t)ParallelGCThreads; | |
1518 size_t sleep_time_millis = MAX2((size_t)1, objsFromOverflow/100); | |
1519 for (size_t spin = 0; prefix == BUSY && spin < spin_count; spin++) { | |
1520 // someone grabbed it before we did ... | |
1521 // ... we spin for a short while... | |
1522 os::sleep(tid, sleep_time_millis, false); | |
1523 if (_overflow_list == NULL) { | |
1524 // nothing left to take | |
1525 return false; | |
1526 } else if (_overflow_list != BUSY) { | |
1527 // try and grab the prefix | |
12316
190899198332
7195622: CheckUnhandledOops has limited usefulness now
hseigel
parents:
12030
diff
changeset
|
1528 prefix = cast_to_oop(Atomic::xchg_ptr(BUSY, &_overflow_list)); |
534 | 1529 } |
0 | 1530 } |
534 | 1531 if (prefix == NULL || prefix == BUSY) { |
1532 // Nothing to take or waited long enough | |
1533 if (prefix == NULL) { | |
1534 // Write back the NULL in case we overwrote it with BUSY above | |
1535 // and it is still the same value. | |
1536 (void) Atomic::cmpxchg_ptr(NULL, &_overflow_list, BUSY); | |
1537 } | |
1538 return false; | |
1539 } | |
1540 assert(prefix != NULL && prefix != BUSY, "Error"); | |
1541 size_t i = 1; | |
0 | 1542 oop cur = prefix; |
167
feeb96a45707
6696264: assert("narrow oop can never be zero") for GCBasher & ParNewGC
coleenp
parents:
113
diff
changeset
|
1543 while (i < objsFromOverflow && cur->klass_or_null() != NULL) { |
6725
da91efe96a93
6964458: Reimplement class meta-data storage to use native memory
coleenp
parents:
6595
diff
changeset
|
1544 i++; cur = cur->list_ptr_from_klass(); |
0 | 1545 } |
1546 | |
1547 // Reattach remaining (suffix) to overflow list | |
534 | 1548 if (cur->klass_or_null() == NULL) { |
1549 // Write back the NULL in lieu of the BUSY we wrote | |
1550 // above and it is still the same value. | |
1551 if (_overflow_list == BUSY) { | |
1552 (void) Atomic::cmpxchg_ptr(NULL, &_overflow_list, BUSY); | |
0 | 1553 } |
534 | 1554 } else { |
6725
da91efe96a93
6964458: Reimplement class meta-data storage to use native memory
coleenp
parents:
6595
diff
changeset
|
1555 assert(cur->klass_or_null() != (Klass*)(address)BUSY, "Error"); |
da91efe96a93
6964458: Reimplement class meta-data storage to use native memory
coleenp
parents:
6595
diff
changeset
|
1556 oop suffix = cur->list_ptr_from_klass(); // suffix will be put back on global list |
534 | 1557 cur->set_klass_to_list_ptr(NULL); // break off suffix |
1558 // It's possible that the list is still in the empty(busy) state | |
1559 // we left it in a short while ago; in that case we may be | |
1560 // able to place back the suffix. | |
1561 oop observed_overflow_list = _overflow_list; | |
1562 oop cur_overflow_list = observed_overflow_list; | |
1563 bool attached = false; | |
1564 while (observed_overflow_list == BUSY || observed_overflow_list == NULL) { | |
1565 observed_overflow_list = | |
1566 (oop) Atomic::cmpxchg_ptr(suffix, &_overflow_list, cur_overflow_list); | |
1567 if (cur_overflow_list == observed_overflow_list) { | |
1568 attached = true; | |
1569 break; | |
1570 } else cur_overflow_list = observed_overflow_list; | |
1571 } | |
1572 if (!attached) { | |
1573 // Too bad, someone else got in in between; we'll need to do a splice. | |
1574 // Find the last item of suffix list | |
1575 oop last = suffix; | |
1576 while (last->klass_or_null() != NULL) { | |
6725
da91efe96a93
6964458: Reimplement class meta-data storage to use native memory
coleenp
parents:
6595
diff
changeset
|
1577 last = last->list_ptr_from_klass(); |
534 | 1578 } |
1579 // Atomically prepend suffix to current overflow list | |
1580 observed_overflow_list = _overflow_list; | |
1581 do { | |
1582 cur_overflow_list = observed_overflow_list; | |
1583 if (cur_overflow_list != BUSY) { | |
1584 // Do the splice ... | |
1585 last->set_klass_to_list_ptr(cur_overflow_list); | |
1586 } else { // cur_overflow_list == BUSY | |
1587 last->set_klass_to_list_ptr(NULL); | |
1588 } | |
1589 observed_overflow_list = | |
1590 (oop)Atomic::cmpxchg_ptr(suffix, &_overflow_list, cur_overflow_list); | |
1591 } while (cur_overflow_list != observed_overflow_list); | |
0 | 1592 } |
1593 } | |
1594 | |
1595 // Push objects on prefix list onto this thread's work queue | |
534 | 1596 assert(prefix != NULL && prefix != BUSY, "program logic"); |
0 | 1597 cur = prefix; |
534 | 1598 ssize_t n = 0; |
0 | 1599 while (cur != NULL) { |
1600 oop obj_to_push = cur->forwardee(); | |
6725
da91efe96a93
6964458: Reimplement class meta-data storage to use native memory
coleenp
parents:
6595
diff
changeset
|
1601 oop next = cur->list_ptr_from_klass(); |
0 | 1602 cur->set_klass(obj_to_push->klass()); |
534 | 1603 // This may be an array object that is self-forwarded. In that case, the list pointer |
1604 // space, cur, is not in the Java heap, but rather in the C-heap and should be freed. | |
1605 if (!is_in_reserved(cur)) { | |
1606 // This can become a scaling bottleneck when there is work queue overflow coincident | |
1607 // with promotion failure. | |
1608 oopDesc* f = cur; | |
6197 | 1609 FREE_C_HEAP_ARRAY(oopDesc, f, mtGC); |
534 | 1610 } else if (par_scan_state->should_be_partially_scanned(obj_to_push, cur)) { |
0 | 1611 assert(arrayOop(cur)->length() == 0, "entire array remaining to be scanned"); |
534 | 1612 obj_to_push = cur; |
0 | 1613 } |
534 | 1614 bool ok = work_q->push(obj_to_push); |
1615 assert(ok, "Should have succeeded"); | |
0 | 1616 cur = next; |
1617 n++; | |
1618 } | |
1710 | 1619 TASKQUEUE_STATS_ONLY(par_scan_state->note_overflow_refill(n)); |
534 | 1620 #ifndef PRODUCT |
1621 assert(_num_par_pushes >= n, "Too many pops?"); | |
1622 Atomic::add_ptr(-(intptr_t)n, &_num_par_pushes); | |
1623 #endif | |
0 | 1624 return true; |
1625 } | |
534 | 1626 #undef BUSY |
0 | 1627 |
10405 | 1628 void ParNewGeneration::ref_processor_init() { |
0 | 1629 if (_ref_processor == NULL) { |
1630 // Allocate and initialize a reference processor | |
2369
92da084fefc9
6668573: CMS: reference processing crash if ParallelCMSThreads > ParallelGCThreads
ysr
parents:
2038
diff
changeset
|
1631 _ref_processor = |
92da084fefc9
6668573: CMS: reference processing crash if ParallelCMSThreads > ParallelGCThreads
ysr
parents:
2038
diff
changeset
|
1632 new ReferenceProcessor(_reserved, // span |
92da084fefc9
6668573: CMS: reference processing crash if ParallelCMSThreads > ParallelGCThreads
ysr
parents:
2038
diff
changeset
|
1633 ParallelRefProcEnabled && (ParallelGCThreads > 1), // mt processing |
92da084fefc9
6668573: CMS: reference processing crash if ParallelCMSThreads > ParallelGCThreads
ysr
parents:
2038
diff
changeset
|
1634 (int) ParallelGCThreads, // mt processing degree |
92da084fefc9
6668573: CMS: reference processing crash if ParallelCMSThreads > ParallelGCThreads
ysr
parents:
2038
diff
changeset
|
1635 refs_discovery_is_mt(), // mt discovery |
92da084fefc9
6668573: CMS: reference processing crash if ParallelCMSThreads > ParallelGCThreads
ysr
parents:
2038
diff
changeset
|
1636 (int) ParallelGCThreads, // mt discovery degree |
92da084fefc9
6668573: CMS: reference processing crash if ParallelCMSThreads > ParallelGCThreads
ysr
parents:
2038
diff
changeset
|
1637 refs_discovery_is_atomic(), // atomic_discovery |
92da084fefc9
6668573: CMS: reference processing crash if ParallelCMSThreads > ParallelGCThreads
ysr
parents:
2038
diff
changeset
|
1638 NULL, // is_alive_non_header |
92da084fefc9
6668573: CMS: reference processing crash if ParallelCMSThreads > ParallelGCThreads
ysr
parents:
2038
diff
changeset
|
1639 false); // write barrier for next field updates |
0 | 1640 } |
1641 } | |
1642 | |
1643 const char* ParNewGeneration::name() const { | |
1644 return "par new generation"; | |
1645 } |