Mercurial > hg > truffle
annotate src/share/vm/gc_implementation/concurrentMarkSweep/compactibleFreeListSpace.cpp @ 1278:da9559b49b84
6915557: assert(_gvn.type(l)->higher_equal(type),"must constrain OSR typestate") with debug build
Reviewed-by: kvn
author | never |
---|---|
date | Thu, 25 Feb 2010 11:38:50 -0800 |
parents | 05b775309e59 |
children | a8127dc669ba |
rev | line source |
---|---|
0 | 1 /* |
579 | 2 * Copyright 2001-2009 Sun Microsystems, Inc. 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 * | |
19 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara, | |
20 * CA 95054 USA or visit www.sun.com if you need additional information or | |
21 * have any questions. | |
22 * | |
23 */ | |
24 | |
25 # include "incls/_precompiled.incl" | |
26 # include "incls/_compactibleFreeListSpace.cpp.incl" | |
27 | |
28 ///////////////////////////////////////////////////////////////////////// | |
29 //// CompactibleFreeListSpace | |
30 ///////////////////////////////////////////////////////////////////////// | |
31 | |
32 // highest ranked free list lock rank | |
33 int CompactibleFreeListSpace::_lockRank = Mutex::leaf + 3; | |
34 | |
35 // Constructor | |
36 CompactibleFreeListSpace::CompactibleFreeListSpace(BlockOffsetSharedArray* bs, | |
37 MemRegion mr, bool use_adaptive_freelists, | |
38 FreeBlockDictionary::DictionaryChoice dictionaryChoice) : | |
39 _dictionaryChoice(dictionaryChoice), | |
40 _adaptive_freelists(use_adaptive_freelists), | |
41 _bt(bs, mr), | |
42 // free list locks are in the range of values taken by _lockRank | |
43 // This range currently is [_leaf+2, _leaf+3] | |
44 // Note: this requires that CFLspace c'tors | |
45 // are called serially in the order in which the locks are | |
46 // are acquired in the program text. This is true today. | |
47 _freelistLock(_lockRank--, "CompactibleFreeListSpace._lock", true), | |
48 _parDictionaryAllocLock(Mutex::leaf - 1, // == rank(ExpandHeap_lock) - 1 | |
49 "CompactibleFreeListSpace._dict_par_lock", true), | |
50 _rescan_task_size(CardTableModRefBS::card_size_in_words * BitsPerWord * | |
51 CMSRescanMultiple), | |
52 _marking_task_size(CardTableModRefBS::card_size_in_words * BitsPerWord * | |
53 CMSConcMarkMultiple), | |
54 _collector(NULL) | |
55 { | |
56 _bt.set_space(this); | |
263
12eea04c8b06
6672698: mangle_unused_area() should not remangle the entire heap at each collection.
jmasa
parents:
187
diff
changeset
|
57 initialize(mr, SpaceDecorator::Clear, SpaceDecorator::Mangle); |
0 | 58 // We have all of "mr", all of which we place in the dictionary |
59 // as one big chunk. We'll need to decide here which of several | |
60 // possible alternative dictionary implementations to use. For | |
61 // now the choice is easy, since we have only one working | |
62 // implementation, namely, the simple binary tree (splaying | |
63 // temporarily disabled). | |
64 switch (dictionaryChoice) { | |
65 case FreeBlockDictionary::dictionarySplayTree: | |
66 case FreeBlockDictionary::dictionarySkipList: | |
67 default: | |
68 warning("dictionaryChoice: selected option not understood; using" | |
69 " default BinaryTreeDictionary implementation instead."); | |
1145
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
70 case FreeBlockDictionary::dictionaryBinaryTree: |
0 | 71 _dictionary = new BinaryTreeDictionary(mr); |
72 break; | |
73 } | |
74 assert(_dictionary != NULL, "CMS dictionary initialization"); | |
75 // The indexed free lists are initially all empty and are lazily | |
76 // filled in on demand. Initialize the array elements to NULL. | |
77 initializeIndexedFreeListArray(); | |
78 | |
79 // Not using adaptive free lists assumes that allocation is first | |
80 // from the linAB's. Also a cms perm gen which can be compacted | |
81 // has to have the klass's klassKlass allocated at a lower | |
82 // address in the heap than the klass so that the klassKlass is | |
83 // moved to its new location before the klass is moved. | |
84 // Set the _refillSize for the linear allocation blocks | |
85 if (!use_adaptive_freelists) { | |
86 FreeChunk* fc = _dictionary->getChunk(mr.word_size()); | |
87 // The small linAB initially has all the space and will allocate | |
88 // a chunk of any size. | |
89 HeapWord* addr = (HeapWord*) fc; | |
90 _smallLinearAllocBlock.set(addr, fc->size() , | |
91 1024*SmallForLinearAlloc, fc->size()); | |
92 // Note that _unallocated_block is not updated here. | |
93 // Allocations from the linear allocation block should | |
94 // update it. | |
95 } else { | |
96 _smallLinearAllocBlock.set(0, 0, 1024*SmallForLinearAlloc, | |
97 SmallForLinearAlloc); | |
98 } | |
99 // CMSIndexedFreeListReplenish should be at least 1 | |
100 CMSIndexedFreeListReplenish = MAX2((uintx)1, CMSIndexedFreeListReplenish); | |
101 _promoInfo.setSpace(this); | |
102 if (UseCMSBestFit) { | |
103 _fitStrategy = FreeBlockBestFitFirst; | |
104 } else { | |
105 _fitStrategy = FreeBlockStrategyNone; | |
106 } | |
107 checkFreeListConsistency(); | |
108 | |
109 // Initialize locks for parallel case. | |
110 if (ParallelGCThreads > 0) { | |
111 for (size_t i = IndexSetStart; i < IndexSetSize; i += IndexSetStride) { | |
112 _indexedFreeListParLocks[i] = new Mutex(Mutex::leaf - 1, // == ExpandHeap_lock - 1 | |
113 "a freelist par lock", | |
114 true); | |
115 if (_indexedFreeListParLocks[i] == NULL) | |
116 vm_exit_during_initialization("Could not allocate a par lock"); | |
117 DEBUG_ONLY( | |
118 _indexedFreeList[i].set_protecting_lock(_indexedFreeListParLocks[i]); | |
119 ) | |
120 } | |
121 _dictionary->set_par_lock(&_parDictionaryAllocLock); | |
122 } | |
123 } | |
124 | |
125 // Like CompactibleSpace forward() but always calls cross_threshold() to | |
126 // update the block offset table. Removed initialize_threshold call because | |
127 // CFLS does not use a block offset array for contiguous spaces. | |
128 HeapWord* CompactibleFreeListSpace::forward(oop q, size_t size, | |
129 CompactPoint* cp, HeapWord* compact_top) { | |
130 // q is alive | |
131 // First check if we should switch compaction space | |
132 assert(this == cp->space, "'this' should be current compaction space."); | |
133 size_t compaction_max_size = pointer_delta(end(), compact_top); | |
134 assert(adjustObjectSize(size) == cp->space->adjust_object_size_v(size), | |
135 "virtual adjustObjectSize_v() method is not correct"); | |
136 size_t adjusted_size = adjustObjectSize(size); | |
137 assert(compaction_max_size >= MinChunkSize || compaction_max_size == 0, | |
138 "no small fragments allowed"); | |
139 assert(minimum_free_block_size() == MinChunkSize, | |
140 "for de-virtualized reference below"); | |
141 // Can't leave a nonzero size, residual fragment smaller than MinChunkSize | |
142 if (adjusted_size + MinChunkSize > compaction_max_size && | |
143 adjusted_size != compaction_max_size) { | |
144 do { | |
145 // switch to next compaction space | |
146 cp->space->set_compaction_top(compact_top); | |
147 cp->space = cp->space->next_compaction_space(); | |
148 if (cp->space == NULL) { | |
149 cp->gen = GenCollectedHeap::heap()->prev_gen(cp->gen); | |
150 assert(cp->gen != NULL, "compaction must succeed"); | |
151 cp->space = cp->gen->first_compaction_space(); | |
152 assert(cp->space != NULL, "generation must have a first compaction space"); | |
153 } | |
154 compact_top = cp->space->bottom(); | |
155 cp->space->set_compaction_top(compact_top); | |
156 // The correct adjusted_size may not be the same as that for this method | |
157 // (i.e., cp->space may no longer be "this" so adjust the size again. | |
158 // Use the virtual method which is not used above to save the virtual | |
159 // dispatch. | |
160 adjusted_size = cp->space->adjust_object_size_v(size); | |
161 compaction_max_size = pointer_delta(cp->space->end(), compact_top); | |
162 assert(cp->space->minimum_free_block_size() == 0, "just checking"); | |
163 } while (adjusted_size > compaction_max_size); | |
164 } | |
165 | |
166 // store the forwarding pointer into the mark word | |
167 if ((HeapWord*)q != compact_top) { | |
168 q->forward_to(oop(compact_top)); | |
169 assert(q->is_gc_marked(), "encoding the pointer should preserve the mark"); | |
170 } else { | |
171 // if the object isn't moving we can just set the mark to the default | |
172 // mark and handle it specially later on. | |
173 q->init_mark(); | |
174 assert(q->forwardee() == NULL, "should be forwarded to NULL"); | |
175 } | |
176 | |
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
12
diff
changeset
|
177 VALIDATE_MARK_SWEEP_ONLY(MarkSweep::register_live_oop(q, adjusted_size)); |
0 | 178 compact_top += adjusted_size; |
179 | |
180 // we need to update the offset table so that the beginnings of objects can be | |
181 // found during scavenge. Note that we are updating the offset table based on | |
182 // where the object will be once the compaction phase finishes. | |
183 | |
184 // Always call cross_threshold(). A contiguous space can only call it when | |
185 // the compaction_top exceeds the current threshold but not for an | |
186 // non-contiguous space. | |
187 cp->threshold = | |
188 cp->space->cross_threshold(compact_top - adjusted_size, compact_top); | |
189 return compact_top; | |
190 } | |
191 | |
192 // A modified copy of OffsetTableContigSpace::cross_threshold() with _offsets -> _bt | |
193 // and use of single_block instead of alloc_block. The name here is not really | |
194 // appropriate - maybe a more general name could be invented for both the | |
195 // contiguous and noncontiguous spaces. | |
196 | |
197 HeapWord* CompactibleFreeListSpace::cross_threshold(HeapWord* start, HeapWord* the_end) { | |
198 _bt.single_block(start, the_end); | |
199 return end(); | |
200 } | |
201 | |
202 // Initialize them to NULL. | |
203 void CompactibleFreeListSpace::initializeIndexedFreeListArray() { | |
204 for (size_t i = 0; i < IndexSetSize; i++) { | |
205 // Note that on platforms where objects are double word aligned, | |
206 // the odd array elements are not used. It is convenient, however, | |
207 // to map directly from the object size to the array element. | |
208 _indexedFreeList[i].reset(IndexSetSize); | |
209 _indexedFreeList[i].set_size(i); | |
210 assert(_indexedFreeList[i].count() == 0, "reset check failed"); | |
211 assert(_indexedFreeList[i].head() == NULL, "reset check failed"); | |
212 assert(_indexedFreeList[i].tail() == NULL, "reset check failed"); | |
213 assert(_indexedFreeList[i].hint() == IndexSetSize, "reset check failed"); | |
214 } | |
215 } | |
216 | |
217 void CompactibleFreeListSpace::resetIndexedFreeListArray() { | |
218 for (int i = 1; i < IndexSetSize; i++) { | |
219 assert(_indexedFreeList[i].size() == (size_t) i, | |
220 "Indexed free list sizes are incorrect"); | |
221 _indexedFreeList[i].reset(IndexSetSize); | |
222 assert(_indexedFreeList[i].count() == 0, "reset check failed"); | |
223 assert(_indexedFreeList[i].head() == NULL, "reset check failed"); | |
224 assert(_indexedFreeList[i].tail() == NULL, "reset check failed"); | |
225 assert(_indexedFreeList[i].hint() == IndexSetSize, "reset check failed"); | |
226 } | |
227 } | |
228 | |
229 void CompactibleFreeListSpace::reset(MemRegion mr) { | |
230 resetIndexedFreeListArray(); | |
231 dictionary()->reset(); | |
232 if (BlockOffsetArrayUseUnallocatedBlock) { | |
233 assert(end() == mr.end(), "We are compacting to the bottom of CMS gen"); | |
234 // Everything's allocated until proven otherwise. | |
235 _bt.set_unallocated_block(end()); | |
236 } | |
237 if (!mr.is_empty()) { | |
238 assert(mr.word_size() >= MinChunkSize, "Chunk size is too small"); | |
239 _bt.single_block(mr.start(), mr.word_size()); | |
240 FreeChunk* fc = (FreeChunk*) mr.start(); | |
241 fc->setSize(mr.word_size()); | |
242 if (mr.word_size() >= IndexSetSize ) { | |
243 returnChunkToDictionary(fc); | |
244 } else { | |
245 _bt.verify_not_unallocated((HeapWord*)fc, fc->size()); | |
246 _indexedFreeList[mr.word_size()].returnChunkAtHead(fc); | |
247 } | |
248 } | |
249 _promoInfo.reset(); | |
250 _smallLinearAllocBlock._ptr = NULL; | |
251 _smallLinearAllocBlock._word_size = 0; | |
252 } | |
253 | |
254 void CompactibleFreeListSpace::reset_after_compaction() { | |
255 // Reset the space to the new reality - one free chunk. | |
256 MemRegion mr(compaction_top(), end()); | |
257 reset(mr); | |
258 // Now refill the linear allocation block(s) if possible. | |
259 if (_adaptive_freelists) { | |
260 refillLinearAllocBlocksIfNeeded(); | |
261 } else { | |
262 // Place as much of mr in the linAB as we can get, | |
263 // provided it was big enough to go into the dictionary. | |
264 FreeChunk* fc = dictionary()->findLargestDict(); | |
265 if (fc != NULL) { | |
266 assert(fc->size() == mr.word_size(), | |
267 "Why was the chunk broken up?"); | |
268 removeChunkFromDictionary(fc); | |
269 HeapWord* addr = (HeapWord*) fc; | |
270 _smallLinearAllocBlock.set(addr, fc->size() , | |
271 1024*SmallForLinearAlloc, fc->size()); | |
272 // Note that _unallocated_block is not updated here. | |
273 } | |
274 } | |
275 } | |
276 | |
277 // Walks the entire dictionary, returning a coterminal | |
278 // chunk, if it exists. Use with caution since it involves | |
279 // a potentially complete walk of a potentially large tree. | |
280 FreeChunk* CompactibleFreeListSpace::find_chunk_at_end() { | |
281 | |
282 assert_lock_strong(&_freelistLock); | |
283 | |
284 return dictionary()->find_chunk_ends_at(end()); | |
285 } | |
286 | |
287 | |
288 #ifndef PRODUCT | |
289 void CompactibleFreeListSpace::initializeIndexedFreeListArrayReturnedBytes() { | |
290 for (size_t i = IndexSetStart; i < IndexSetSize; i += IndexSetStride) { | |
291 _indexedFreeList[i].allocation_stats()->set_returnedBytes(0); | |
292 } | |
293 } | |
294 | |
295 size_t CompactibleFreeListSpace::sumIndexedFreeListArrayReturnedBytes() { | |
296 size_t sum = 0; | |
297 for (size_t i = IndexSetStart; i < IndexSetSize; i += IndexSetStride) { | |
298 sum += _indexedFreeList[i].allocation_stats()->returnedBytes(); | |
299 } | |
300 return sum; | |
301 } | |
302 | |
303 size_t CompactibleFreeListSpace::totalCountInIndexedFreeLists() const { | |
304 size_t count = 0; | |
305 for (int i = MinChunkSize; i < IndexSetSize; i++) { | |
306 debug_only( | |
307 ssize_t total_list_count = 0; | |
308 for (FreeChunk* fc = _indexedFreeList[i].head(); fc != NULL; | |
309 fc = fc->next()) { | |
310 total_list_count++; | |
311 } | |
312 assert(total_list_count == _indexedFreeList[i].count(), | |
313 "Count in list is incorrect"); | |
314 ) | |
315 count += _indexedFreeList[i].count(); | |
316 } | |
317 return count; | |
318 } | |
319 | |
320 size_t CompactibleFreeListSpace::totalCount() { | |
321 size_t num = totalCountInIndexedFreeLists(); | |
322 num += dictionary()->totalCount(); | |
323 if (_smallLinearAllocBlock._word_size != 0) { | |
324 num++; | |
325 } | |
326 return num; | |
327 } | |
328 #endif | |
329 | |
330 bool CompactibleFreeListSpace::is_free_block(const HeapWord* p) const { | |
331 FreeChunk* fc = (FreeChunk*) p; | |
332 return fc->isFree(); | |
333 } | |
334 | |
335 size_t CompactibleFreeListSpace::used() const { | |
336 return capacity() - free(); | |
337 } | |
338 | |
339 size_t CompactibleFreeListSpace::free() const { | |
340 // "MT-safe, but not MT-precise"(TM), if you will: i.e. | |
341 // if you do this while the structures are in flux you | |
342 // may get an approximate answer only; for instance | |
343 // because there is concurrent allocation either | |
344 // directly by mutators or for promotion during a GC. | |
345 // It's "MT-safe", however, in the sense that you are guaranteed | |
346 // not to crash and burn, for instance, because of walking | |
347 // pointers that could disappear as you were walking them. | |
348 // The approximation is because the various components | |
349 // that are read below are not read atomically (and | |
350 // further the computation of totalSizeInIndexedFreeLists() | |
351 // is itself a non-atomic computation. The normal use of | |
352 // this is during a resize operation at the end of GC | |
353 // and at that time you are guaranteed to get the | |
354 // correct actual value. However, for instance, this is | |
355 // also read completely asynchronously by the "perf-sampler" | |
356 // that supports jvmstat, and you are apt to see the values | |
357 // flicker in such cases. | |
358 assert(_dictionary != NULL, "No _dictionary?"); | |
359 return (_dictionary->totalChunkSize(DEBUG_ONLY(freelistLock())) + | |
360 totalSizeInIndexedFreeLists() + | |
361 _smallLinearAllocBlock._word_size) * HeapWordSize; | |
362 } | |
363 | |
364 size_t CompactibleFreeListSpace::max_alloc_in_words() const { | |
365 assert(_dictionary != NULL, "No _dictionary?"); | |
366 assert_locked(); | |
367 size_t res = _dictionary->maxChunkSize(); | |
368 res = MAX2(res, MIN2(_smallLinearAllocBlock._word_size, | |
369 (size_t) SmallForLinearAlloc - 1)); | |
370 // XXX the following could potentially be pretty slow; | |
371 // should one, pesimally for the rare cases when res | |
372 // caclulated above is less than IndexSetSize, | |
373 // just return res calculated above? My reasoning was that | |
374 // those cases will be so rare that the extra time spent doesn't | |
375 // really matter.... | |
376 // Note: do not change the loop test i >= res + IndexSetStride | |
377 // to i > res below, because i is unsigned and res may be zero. | |
378 for (size_t i = IndexSetSize - 1; i >= res + IndexSetStride; | |
379 i -= IndexSetStride) { | |
380 if (_indexedFreeList[i].head() != NULL) { | |
381 assert(_indexedFreeList[i].count() != 0, "Inconsistent FreeList"); | |
382 return i; | |
383 } | |
384 } | |
385 return res; | |
386 } | |
387 | |
1145
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
388 void CompactibleFreeListSpace::print_indexed_free_lists(outputStream* st) |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
389 const { |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
390 reportIndexedFreeListStatistics(); |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
391 gclog_or_tty->print_cr("Layout of Indexed Freelists"); |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
392 gclog_or_tty->print_cr("---------------------------"); |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
393 FreeList::print_labels_on(st, "size"); |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
394 for (size_t i = IndexSetStart; i < IndexSetSize; i += IndexSetStride) { |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
395 _indexedFreeList[i].print_on(gclog_or_tty); |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
396 for (FreeChunk* fc = _indexedFreeList[i].head(); fc != NULL; |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
397 fc = fc->next()) { |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
398 gclog_or_tty->print_cr("\t[" PTR_FORMAT "," PTR_FORMAT ") %s", |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
399 fc, (HeapWord*)fc + i, |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
400 fc->cantCoalesce() ? "\t CC" : ""); |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
401 } |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
402 } |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
403 } |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
404 |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
405 void CompactibleFreeListSpace::print_promo_info_blocks(outputStream* st) |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
406 const { |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
407 _promoInfo.print_on(st); |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
408 } |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
409 |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
410 void CompactibleFreeListSpace::print_dictionary_free_lists(outputStream* st) |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
411 const { |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
412 _dictionary->reportStatistics(); |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
413 st->print_cr("Layout of Freelists in Tree"); |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
414 st->print_cr("---------------------------"); |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
415 _dictionary->print_free_lists(st); |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
416 } |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
417 |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
418 class BlkPrintingClosure: public BlkClosure { |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
419 const CMSCollector* _collector; |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
420 const CompactibleFreeListSpace* _sp; |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
421 const CMSBitMap* _live_bit_map; |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
422 const bool _post_remark; |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
423 outputStream* _st; |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
424 public: |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
425 BlkPrintingClosure(const CMSCollector* collector, |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
426 const CompactibleFreeListSpace* sp, |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
427 const CMSBitMap* live_bit_map, |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
428 outputStream* st): |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
429 _collector(collector), |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
430 _sp(sp), |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
431 _live_bit_map(live_bit_map), |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
432 _post_remark(collector->abstract_state() > CMSCollector::FinalMarking), |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
433 _st(st) { } |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
434 size_t do_blk(HeapWord* addr); |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
435 }; |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
436 |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
437 size_t BlkPrintingClosure::do_blk(HeapWord* addr) { |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
438 size_t sz = _sp->block_size_no_stall(addr, _collector); |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
439 assert(sz != 0, "Should always be able to compute a size"); |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
440 if (_sp->block_is_obj(addr)) { |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
441 const bool dead = _post_remark && !_live_bit_map->isMarked(addr); |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
442 _st->print_cr(PTR_FORMAT ": %s object of size " SIZE_FORMAT "%s", |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
443 addr, |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
444 dead ? "dead" : "live", |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
445 sz, |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
446 (!dead && CMSPrintObjectsInDump) ? ":" : "."); |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
447 if (CMSPrintObjectsInDump && !dead) { |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
448 oop(addr)->print_on(_st); |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
449 _st->print_cr("--------------------------------------"); |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
450 } |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
451 } else { // free block |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
452 _st->print_cr(PTR_FORMAT ": free block of size " SIZE_FORMAT "%s", |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
453 addr, sz, CMSPrintChunksInDump ? ":" : "."); |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
454 if (CMSPrintChunksInDump) { |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
455 ((FreeChunk*)addr)->print_on(_st); |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
456 _st->print_cr("--------------------------------------"); |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
457 } |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
458 } |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
459 return sz; |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
460 } |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
461 |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
462 void CompactibleFreeListSpace::dump_at_safepoint_with_locks(CMSCollector* c, |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
463 outputStream* st) { |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
464 st->print_cr("\n========================="); |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
465 st->print_cr("Block layout in CMS Heap:"); |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
466 st->print_cr("========================="); |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
467 BlkPrintingClosure bpcl(c, this, c->markBitMap(), st); |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
468 blk_iterate(&bpcl); |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
469 |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
470 st->print_cr("\n======================================="); |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
471 st->print_cr("Order & Layout of Promotion Info Blocks"); |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
472 st->print_cr("======================================="); |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
473 print_promo_info_blocks(st); |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
474 |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
475 st->print_cr("\n==========================="); |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
476 st->print_cr("Order of Indexed Free Lists"); |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
477 st->print_cr("========================="); |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
478 print_indexed_free_lists(st); |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
479 |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
480 st->print_cr("\n================================="); |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
481 st->print_cr("Order of Free Lists in Dictionary"); |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
482 st->print_cr("================================="); |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
483 print_dictionary_free_lists(st); |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
484 } |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
485 |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
486 |
0 | 487 void CompactibleFreeListSpace::reportFreeListStatistics() const { |
488 assert_lock_strong(&_freelistLock); | |
489 assert(PrintFLSStatistics != 0, "Reporting error"); | |
490 _dictionary->reportStatistics(); | |
491 if (PrintFLSStatistics > 1) { | |
492 reportIndexedFreeListStatistics(); | |
493 size_t totalSize = totalSizeInIndexedFreeLists() + | |
494 _dictionary->totalChunkSize(DEBUG_ONLY(freelistLock())); | |
495 gclog_or_tty->print(" free=%ld frag=%1.4f\n", totalSize, flsFrag()); | |
496 } | |
497 } | |
498 | |
499 void CompactibleFreeListSpace::reportIndexedFreeListStatistics() const { | |
500 assert_lock_strong(&_freelistLock); | |
501 gclog_or_tty->print("Statistics for IndexedFreeLists:\n" | |
502 "--------------------------------\n"); | |
503 size_t totalSize = totalSizeInIndexedFreeLists(); | |
504 size_t freeBlocks = numFreeBlocksInIndexedFreeLists(); | |
505 gclog_or_tty->print("Total Free Space: %d\n", totalSize); | |
506 gclog_or_tty->print("Max Chunk Size: %d\n", maxChunkSizeInIndexedFreeLists()); | |
507 gclog_or_tty->print("Number of Blocks: %d\n", freeBlocks); | |
508 if (freeBlocks != 0) { | |
509 gclog_or_tty->print("Av. Block Size: %d\n", totalSize/freeBlocks); | |
510 } | |
511 } | |
512 | |
513 size_t CompactibleFreeListSpace::numFreeBlocksInIndexedFreeLists() const { | |
514 size_t res = 0; | |
515 for (size_t i = IndexSetStart; i < IndexSetSize; i += IndexSetStride) { | |
516 debug_only( | |
517 ssize_t recount = 0; | |
518 for (FreeChunk* fc = _indexedFreeList[i].head(); fc != NULL; | |
519 fc = fc->next()) { | |
520 recount += 1; | |
521 } | |
522 assert(recount == _indexedFreeList[i].count(), | |
523 "Incorrect count in list"); | |
524 ) | |
525 res += _indexedFreeList[i].count(); | |
526 } | |
527 return res; | |
528 } | |
529 | |
530 size_t CompactibleFreeListSpace::maxChunkSizeInIndexedFreeLists() const { | |
531 for (size_t i = IndexSetSize - 1; i != 0; i -= IndexSetStride) { | |
532 if (_indexedFreeList[i].head() != NULL) { | |
533 assert(_indexedFreeList[i].count() != 0, "Inconsistent FreeList"); | |
534 return (size_t)i; | |
535 } | |
536 } | |
537 return 0; | |
538 } | |
539 | |
540 void CompactibleFreeListSpace::set_end(HeapWord* value) { | |
541 HeapWord* prevEnd = end(); | |
542 assert(prevEnd != value, "unnecessary set_end call"); | |
543 assert(prevEnd == NULL || value >= unallocated_block(), "New end is below unallocated block"); | |
544 _end = value; | |
545 if (prevEnd != NULL) { | |
546 // Resize the underlying block offset table. | |
547 _bt.resize(pointer_delta(value, bottom())); | |
1145
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
548 if (value <= prevEnd) { |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
549 assert(value >= unallocated_block(), "New end is below unallocated block"); |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
550 } else { |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
551 // Now, take this new chunk and add it to the free blocks. |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
552 // Note that the BOT has not yet been updated for this block. |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
553 size_t newFcSize = pointer_delta(value, prevEnd); |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
554 // XXX This is REALLY UGLY and should be fixed up. XXX |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
555 if (!_adaptive_freelists && _smallLinearAllocBlock._ptr == NULL) { |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
556 // Mark the boundary of the new block in BOT |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
557 _bt.mark_block(prevEnd, value); |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
558 // put it all in the linAB |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
559 if (ParallelGCThreads == 0) { |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
560 _smallLinearAllocBlock._ptr = prevEnd; |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
561 _smallLinearAllocBlock._word_size = newFcSize; |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
562 repairLinearAllocBlock(&_smallLinearAllocBlock); |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
563 } else { // ParallelGCThreads > 0 |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
564 MutexLockerEx x(parDictionaryAllocLock(), |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
565 Mutex::_no_safepoint_check_flag); |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
566 _smallLinearAllocBlock._ptr = prevEnd; |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
567 _smallLinearAllocBlock._word_size = newFcSize; |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
568 repairLinearAllocBlock(&_smallLinearAllocBlock); |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
569 } |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
570 // Births of chunks put into a LinAB are not recorded. Births |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
571 // of chunks as they are allocated out of a LinAB are. |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
572 } else { |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
573 // Add the block to the free lists, if possible coalescing it |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
574 // with the last free block, and update the BOT and census data. |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
575 addChunkToFreeListsAtEndRecordingStats(prevEnd, newFcSize); |
0 | 576 } |
577 } | |
578 } | |
579 } | |
580 | |
581 class FreeListSpace_DCTOC : public Filtering_DCTOC { | |
582 CompactibleFreeListSpace* _cfls; | |
583 CMSCollector* _collector; | |
584 protected: | |
585 // Override. | |
586 #define walk_mem_region_with_cl_DECL(ClosureType) \ | |
587 virtual void walk_mem_region_with_cl(MemRegion mr, \ | |
588 HeapWord* bottom, HeapWord* top, \ | |
589 ClosureType* cl); \ | |
590 void walk_mem_region_with_cl_par(MemRegion mr, \ | |
591 HeapWord* bottom, HeapWord* top, \ | |
592 ClosureType* cl); \ | |
593 void walk_mem_region_with_cl_nopar(MemRegion mr, \ | |
594 HeapWord* bottom, HeapWord* top, \ | |
595 ClosureType* cl) | |
596 walk_mem_region_with_cl_DECL(OopClosure); | |
597 walk_mem_region_with_cl_DECL(FilteringClosure); | |
598 | |
599 public: | |
600 FreeListSpace_DCTOC(CompactibleFreeListSpace* sp, | |
601 CMSCollector* collector, | |
602 OopClosure* cl, | |
603 CardTableModRefBS::PrecisionStyle precision, | |
604 HeapWord* boundary) : | |
605 Filtering_DCTOC(sp, cl, precision, boundary), | |
606 _cfls(sp), _collector(collector) {} | |
607 }; | |
608 | |
609 // We de-virtualize the block-related calls below, since we know that our | |
610 // space is a CompactibleFreeListSpace. | |
611 #define FreeListSpace_DCTOC__walk_mem_region_with_cl_DEFN(ClosureType) \ | |
612 void FreeListSpace_DCTOC::walk_mem_region_with_cl(MemRegion mr, \ | |
613 HeapWord* bottom, \ | |
614 HeapWord* top, \ | |
615 ClosureType* cl) { \ | |
616 if (SharedHeap::heap()->n_par_threads() > 0) { \ | |
617 walk_mem_region_with_cl_par(mr, bottom, top, cl); \ | |
618 } else { \ | |
619 walk_mem_region_with_cl_nopar(mr, bottom, top, cl); \ | |
620 } \ | |
621 } \ | |
622 void FreeListSpace_DCTOC::walk_mem_region_with_cl_par(MemRegion mr, \ | |
623 HeapWord* bottom, \ | |
624 HeapWord* top, \ | |
625 ClosureType* cl) { \ | |
626 /* Skip parts that are before "mr", in case "block_start" sent us \ | |
627 back too far. */ \ | |
628 HeapWord* mr_start = mr.start(); \ | |
629 size_t bot_size = _cfls->CompactibleFreeListSpace::block_size(bottom); \ | |
630 HeapWord* next = bottom + bot_size; \ | |
631 while (next < mr_start) { \ | |
632 bottom = next; \ | |
633 bot_size = _cfls->CompactibleFreeListSpace::block_size(bottom); \ | |
634 next = bottom + bot_size; \ | |
635 } \ | |
636 \ | |
637 while (bottom < top) { \ | |
638 if (_cfls->CompactibleFreeListSpace::block_is_obj(bottom) && \ | |
639 !_cfls->CompactibleFreeListSpace::obj_allocated_since_save_marks( \ | |
640 oop(bottom)) && \ | |
641 !_collector->CMSCollector::is_dead_obj(oop(bottom))) { \ | |
642 size_t word_sz = oop(bottom)->oop_iterate(cl, mr); \ | |
643 bottom += _cfls->adjustObjectSize(word_sz); \ | |
644 } else { \ | |
645 bottom += _cfls->CompactibleFreeListSpace::block_size(bottom); \ | |
646 } \ | |
647 } \ | |
648 } \ | |
649 void FreeListSpace_DCTOC::walk_mem_region_with_cl_nopar(MemRegion mr, \ | |
650 HeapWord* bottom, \ | |
651 HeapWord* top, \ | |
652 ClosureType* cl) { \ | |
653 /* Skip parts that are before "mr", in case "block_start" sent us \ | |
654 back too far. */ \ | |
655 HeapWord* mr_start = mr.start(); \ | |
656 size_t bot_size = _cfls->CompactibleFreeListSpace::block_size_nopar(bottom); \ | |
657 HeapWord* next = bottom + bot_size; \ | |
658 while (next < mr_start) { \ | |
659 bottom = next; \ | |
660 bot_size = _cfls->CompactibleFreeListSpace::block_size_nopar(bottom); \ | |
661 next = bottom + bot_size; \ | |
662 } \ | |
663 \ | |
664 while (bottom < top) { \ | |
665 if (_cfls->CompactibleFreeListSpace::block_is_obj_nopar(bottom) && \ | |
666 !_cfls->CompactibleFreeListSpace::obj_allocated_since_save_marks( \ | |
667 oop(bottom)) && \ | |
668 !_collector->CMSCollector::is_dead_obj(oop(bottom))) { \ | |
669 size_t word_sz = oop(bottom)->oop_iterate(cl, mr); \ | |
670 bottom += _cfls->adjustObjectSize(word_sz); \ | |
671 } else { \ | |
672 bottom += _cfls->CompactibleFreeListSpace::block_size_nopar(bottom); \ | |
673 } \ | |
674 } \ | |
675 } | |
676 | |
677 // (There are only two of these, rather than N, because the split is due | |
678 // only to the introduction of the FilteringClosure, a local part of the | |
679 // impl of this abstraction.) | |
680 FreeListSpace_DCTOC__walk_mem_region_with_cl_DEFN(OopClosure) | |
681 FreeListSpace_DCTOC__walk_mem_region_with_cl_DEFN(FilteringClosure) | |
682 | |
683 DirtyCardToOopClosure* | |
684 CompactibleFreeListSpace::new_dcto_cl(OopClosure* cl, | |
685 CardTableModRefBS::PrecisionStyle precision, | |
686 HeapWord* boundary) { | |
687 return new FreeListSpace_DCTOC(this, _collector, cl, precision, boundary); | |
688 } | |
689 | |
690 | |
691 // Note on locking for the space iteration functions: | |
692 // since the collector's iteration activities are concurrent with | |
693 // allocation activities by mutators, absent a suitable mutual exclusion | |
694 // mechanism the iterators may go awry. For instace a block being iterated | |
695 // may suddenly be allocated or divided up and part of it allocated and | |
696 // so on. | |
697 | |
698 // Apply the given closure to each block in the space. | |
699 void CompactibleFreeListSpace::blk_iterate_careful(BlkClosureCareful* cl) { | |
700 assert_lock_strong(freelistLock()); | |
701 HeapWord *cur, *limit; | |
702 for (cur = bottom(), limit = end(); cur < limit; | |
703 cur += cl->do_blk_careful(cur)); | |
704 } | |
705 | |
706 // Apply the given closure to each block in the space. | |
707 void CompactibleFreeListSpace::blk_iterate(BlkClosure* cl) { | |
708 assert_lock_strong(freelistLock()); | |
709 HeapWord *cur, *limit; | |
710 for (cur = bottom(), limit = end(); cur < limit; | |
711 cur += cl->do_blk(cur)); | |
712 } | |
713 | |
714 // Apply the given closure to each oop in the space. | |
715 void CompactibleFreeListSpace::oop_iterate(OopClosure* cl) { | |
716 assert_lock_strong(freelistLock()); | |
717 HeapWord *cur, *limit; | |
718 size_t curSize; | |
719 for (cur = bottom(), limit = end(); cur < limit; | |
720 cur += curSize) { | |
721 curSize = block_size(cur); | |
722 if (block_is_obj(cur)) { | |
723 oop(cur)->oop_iterate(cl); | |
724 } | |
725 } | |
726 } | |
727 | |
728 // Apply the given closure to each oop in the space \intersect memory region. | |
729 void CompactibleFreeListSpace::oop_iterate(MemRegion mr, OopClosure* cl) { | |
730 assert_lock_strong(freelistLock()); | |
731 if (is_empty()) { | |
732 return; | |
733 } | |
734 MemRegion cur = MemRegion(bottom(), end()); | |
735 mr = mr.intersection(cur); | |
736 if (mr.is_empty()) { | |
737 return; | |
738 } | |
739 if (mr.equals(cur)) { | |
740 oop_iterate(cl); | |
741 return; | |
742 } | |
743 assert(mr.end() <= end(), "just took an intersection above"); | |
744 HeapWord* obj_addr = block_start(mr.start()); | |
745 HeapWord* t = mr.end(); | |
746 | |
747 SpaceMemRegionOopsIterClosure smr_blk(cl, mr); | |
748 if (block_is_obj(obj_addr)) { | |
749 // Handle first object specially. | |
750 oop obj = oop(obj_addr); | |
751 obj_addr += adjustObjectSize(obj->oop_iterate(&smr_blk)); | |
752 } else { | |
753 FreeChunk* fc = (FreeChunk*)obj_addr; | |
754 obj_addr += fc->size(); | |
755 } | |
756 while (obj_addr < t) { | |
757 HeapWord* obj = obj_addr; | |
758 obj_addr += block_size(obj_addr); | |
759 // If "obj_addr" is not greater than top, then the | |
760 // entire object "obj" is within the region. | |
761 if (obj_addr <= t) { | |
762 if (block_is_obj(obj)) { | |
763 oop(obj)->oop_iterate(cl); | |
764 } | |
765 } else { | |
766 // "obj" extends beyond end of region | |
767 if (block_is_obj(obj)) { | |
768 oop(obj)->oop_iterate(&smr_blk); | |
769 } | |
770 break; | |
771 } | |
772 } | |
773 } | |
774 | |
775 // NOTE: In the following methods, in order to safely be able to | |
776 // apply the closure to an object, we need to be sure that the | |
777 // object has been initialized. We are guaranteed that an object | |
778 // is initialized if we are holding the Heap_lock with the | |
779 // world stopped. | |
780 void CompactibleFreeListSpace::verify_objects_initialized() const { | |
781 if (is_init_completed()) { | |
782 assert_locked_or_safepoint(Heap_lock); | |
783 if (Universe::is_fully_initialized()) { | |
784 guarantee(SafepointSynchronize::is_at_safepoint(), | |
785 "Required for objects to be initialized"); | |
786 } | |
787 } // else make a concession at vm start-up | |
788 } | |
789 | |
790 // Apply the given closure to each object in the space | |
791 void CompactibleFreeListSpace::object_iterate(ObjectClosure* blk) { | |
792 assert_lock_strong(freelistLock()); | |
793 NOT_PRODUCT(verify_objects_initialized()); | |
794 HeapWord *cur, *limit; | |
795 size_t curSize; | |
796 for (cur = bottom(), limit = end(); cur < limit; | |
797 cur += curSize) { | |
798 curSize = block_size(cur); | |
799 if (block_is_obj(cur)) { | |
800 blk->do_object(oop(cur)); | |
801 } | |
802 } | |
803 } | |
804 | |
517
e9be0e04635a
6689653: JMapPerm fails with UseConcMarkSweepIncGC and compressed oops off
jmasa
parents:
360
diff
changeset
|
805 // Apply the given closure to each live object in the space |
e9be0e04635a
6689653: JMapPerm fails with UseConcMarkSweepIncGC and compressed oops off
jmasa
parents:
360
diff
changeset
|
806 // The usage of CompactibleFreeListSpace |
e9be0e04635a
6689653: JMapPerm fails with UseConcMarkSweepIncGC and compressed oops off
jmasa
parents:
360
diff
changeset
|
807 // by the ConcurrentMarkSweepGeneration for concurrent GC's allows |
e9be0e04635a
6689653: JMapPerm fails with UseConcMarkSweepIncGC and compressed oops off
jmasa
parents:
360
diff
changeset
|
808 // objects in the space with references to objects that are no longer |
e9be0e04635a
6689653: JMapPerm fails with UseConcMarkSweepIncGC and compressed oops off
jmasa
parents:
360
diff
changeset
|
809 // valid. For example, an object may reference another object |
e9be0e04635a
6689653: JMapPerm fails with UseConcMarkSweepIncGC and compressed oops off
jmasa
parents:
360
diff
changeset
|
810 // that has already been sweep up (collected). This method uses |
e9be0e04635a
6689653: JMapPerm fails with UseConcMarkSweepIncGC and compressed oops off
jmasa
parents:
360
diff
changeset
|
811 // obj_is_alive() to determine whether it is safe to apply the closure to |
e9be0e04635a
6689653: JMapPerm fails with UseConcMarkSweepIncGC and compressed oops off
jmasa
parents:
360
diff
changeset
|
812 // an object. See obj_is_alive() for details on how liveness of an |
e9be0e04635a
6689653: JMapPerm fails with UseConcMarkSweepIncGC and compressed oops off
jmasa
parents:
360
diff
changeset
|
813 // object is decided. |
e9be0e04635a
6689653: JMapPerm fails with UseConcMarkSweepIncGC and compressed oops off
jmasa
parents:
360
diff
changeset
|
814 |
e9be0e04635a
6689653: JMapPerm fails with UseConcMarkSweepIncGC and compressed oops off
jmasa
parents:
360
diff
changeset
|
815 void CompactibleFreeListSpace::safe_object_iterate(ObjectClosure* blk) { |
e9be0e04635a
6689653: JMapPerm fails with UseConcMarkSweepIncGC and compressed oops off
jmasa
parents:
360
diff
changeset
|
816 assert_lock_strong(freelistLock()); |
e9be0e04635a
6689653: JMapPerm fails with UseConcMarkSweepIncGC and compressed oops off
jmasa
parents:
360
diff
changeset
|
817 NOT_PRODUCT(verify_objects_initialized()); |
e9be0e04635a
6689653: JMapPerm fails with UseConcMarkSweepIncGC and compressed oops off
jmasa
parents:
360
diff
changeset
|
818 HeapWord *cur, *limit; |
e9be0e04635a
6689653: JMapPerm fails with UseConcMarkSweepIncGC and compressed oops off
jmasa
parents:
360
diff
changeset
|
819 size_t curSize; |
e9be0e04635a
6689653: JMapPerm fails with UseConcMarkSweepIncGC and compressed oops off
jmasa
parents:
360
diff
changeset
|
820 for (cur = bottom(), limit = end(); cur < limit; |
e9be0e04635a
6689653: JMapPerm fails with UseConcMarkSweepIncGC and compressed oops off
jmasa
parents:
360
diff
changeset
|
821 cur += curSize) { |
e9be0e04635a
6689653: JMapPerm fails with UseConcMarkSweepIncGC and compressed oops off
jmasa
parents:
360
diff
changeset
|
822 curSize = block_size(cur); |
e9be0e04635a
6689653: JMapPerm fails with UseConcMarkSweepIncGC and compressed oops off
jmasa
parents:
360
diff
changeset
|
823 if (block_is_obj(cur) && obj_is_alive(cur)) { |
e9be0e04635a
6689653: JMapPerm fails with UseConcMarkSweepIncGC and compressed oops off
jmasa
parents:
360
diff
changeset
|
824 blk->do_object(oop(cur)); |
e9be0e04635a
6689653: JMapPerm fails with UseConcMarkSweepIncGC and compressed oops off
jmasa
parents:
360
diff
changeset
|
825 } |
e9be0e04635a
6689653: JMapPerm fails with UseConcMarkSweepIncGC and compressed oops off
jmasa
parents:
360
diff
changeset
|
826 } |
e9be0e04635a
6689653: JMapPerm fails with UseConcMarkSweepIncGC and compressed oops off
jmasa
parents:
360
diff
changeset
|
827 } |
e9be0e04635a
6689653: JMapPerm fails with UseConcMarkSweepIncGC and compressed oops off
jmasa
parents:
360
diff
changeset
|
828 |
0 | 829 void CompactibleFreeListSpace::object_iterate_mem(MemRegion mr, |
830 UpwardsObjectClosure* cl) { | |
1145
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
831 assert_locked(freelistLock()); |
0 | 832 NOT_PRODUCT(verify_objects_initialized()); |
833 Space::object_iterate_mem(mr, cl); | |
834 } | |
835 | |
836 // Callers of this iterator beware: The closure application should | |
837 // be robust in the face of uninitialized objects and should (always) | |
838 // return a correct size so that the next addr + size below gives us a | |
839 // valid block boundary. [See for instance, | |
840 // ScanMarkedObjectsAgainCarefullyClosure::do_object_careful() | |
841 // in ConcurrentMarkSweepGeneration.cpp.] | |
842 HeapWord* | |
843 CompactibleFreeListSpace::object_iterate_careful(ObjectClosureCareful* cl) { | |
844 assert_lock_strong(freelistLock()); | |
845 HeapWord *addr, *last; | |
846 size_t size; | |
847 for (addr = bottom(), last = end(); | |
848 addr < last; addr += size) { | |
849 FreeChunk* fc = (FreeChunk*)addr; | |
850 if (fc->isFree()) { | |
851 // Since we hold the free list lock, which protects direct | |
852 // allocation in this generation by mutators, a free object | |
853 // will remain free throughout this iteration code. | |
854 size = fc->size(); | |
855 } else { | |
856 // Note that the object need not necessarily be initialized, | |
857 // because (for instance) the free list lock does NOT protect | |
858 // object initialization. The closure application below must | |
859 // therefore be correct in the face of uninitialized objects. | |
860 size = cl->do_object_careful(oop(addr)); | |
861 if (size == 0) { | |
862 // An unparsable object found. Signal early termination. | |
863 return addr; | |
864 } | |
865 } | |
866 } | |
867 return NULL; | |
868 } | |
869 | |
870 // Callers of this iterator beware: The closure application should | |
871 // be robust in the face of uninitialized objects and should (always) | |
872 // return a correct size so that the next addr + size below gives us a | |
873 // valid block boundary. [See for instance, | |
874 // ScanMarkedObjectsAgainCarefullyClosure::do_object_careful() | |
875 // in ConcurrentMarkSweepGeneration.cpp.] | |
876 HeapWord* | |
877 CompactibleFreeListSpace::object_iterate_careful_m(MemRegion mr, | |
878 ObjectClosureCareful* cl) { | |
879 assert_lock_strong(freelistLock()); | |
880 // Can't use used_region() below because it may not necessarily | |
881 // be the same as [bottom(),end()); although we could | |
882 // use [used_region().start(),round_to(used_region().end(),CardSize)), | |
883 // that appears too cumbersome, so we just do the simpler check | |
884 // in the assertion below. | |
885 assert(!mr.is_empty() && MemRegion(bottom(),end()).contains(mr), | |
886 "mr should be non-empty and within used space"); | |
887 HeapWord *addr, *end; | |
888 size_t size; | |
889 for (addr = block_start_careful(mr.start()), end = mr.end(); | |
890 addr < end; addr += size) { | |
891 FreeChunk* fc = (FreeChunk*)addr; | |
892 if (fc->isFree()) { | |
893 // Since we hold the free list lock, which protects direct | |
894 // allocation in this generation by mutators, a free object | |
895 // will remain free throughout this iteration code. | |
896 size = fc->size(); | |
897 } else { | |
898 // Note that the object need not necessarily be initialized, | |
899 // because (for instance) the free list lock does NOT protect | |
900 // object initialization. The closure application below must | |
901 // therefore be correct in the face of uninitialized objects. | |
902 size = cl->do_object_careful_m(oop(addr), mr); | |
903 if (size == 0) { | |
904 // An unparsable object found. Signal early termination. | |
905 return addr; | |
906 } | |
907 } | |
908 } | |
909 return NULL; | |
910 } | |
911 | |
912 | |
342
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
113
diff
changeset
|
913 HeapWord* CompactibleFreeListSpace::block_start_const(const void* p) const { |
0 | 914 NOT_PRODUCT(verify_objects_initialized()); |
915 return _bt.block_start(p); | |
916 } | |
917 | |
918 HeapWord* CompactibleFreeListSpace::block_start_careful(const void* p) const { | |
919 return _bt.block_start_careful(p); | |
920 } | |
921 | |
922 size_t CompactibleFreeListSpace::block_size(const HeapWord* p) const { | |
923 NOT_PRODUCT(verify_objects_initialized()); | |
924 assert(MemRegion(bottom(), end()).contains(p), "p not in space"); | |
925 // This must be volatile, or else there is a danger that the compiler | |
926 // will compile the code below into a sometimes-infinite loop, by keeping | |
927 // the value read the first time in a register. | |
928 while (true) { | |
929 // We must do this until we get a consistent view of the object. | |
187 | 930 if (FreeChunk::indicatesFreeChunk(p)) { |
931 volatile FreeChunk* fc = (volatile FreeChunk*)p; | |
932 size_t res = fc->size(); | |
933 // If the object is still a free chunk, return the size, else it | |
934 // has been allocated so try again. | |
935 if (FreeChunk::indicatesFreeChunk(p)) { | |
0 | 936 assert(res != 0, "Block size should not be 0"); |
937 return res; | |
938 } | |
187 | 939 } else { |
940 // must read from what 'p' points to in each loop. | |
941 klassOop k = ((volatile oopDesc*)p)->klass_or_null(); | |
942 if (k != NULL) { | |
943 assert(k->is_oop(true /* ignore mark word */), "Should really be klass oop."); | |
944 oop o = (oop)p; | |
945 assert(o->is_parsable(), "Should be parsable"); | |
946 assert(o->is_oop(true /* ignore mark word */), "Should be an oop."); | |
947 size_t res = o->size_given_klass(k->klass_part()); | |
948 res = adjustObjectSize(res); | |
949 assert(res != 0, "Block size should not be 0"); | |
950 return res; | |
951 } | |
0 | 952 } |
953 } | |
954 } | |
955 | |
956 // A variant of the above that uses the Printezis bits for | |
957 // unparsable but allocated objects. This avoids any possible | |
958 // stalls waiting for mutators to initialize objects, and is | |
959 // thus potentially faster than the variant above. However, | |
960 // this variant may return a zero size for a block that is | |
961 // under mutation and for which a consistent size cannot be | |
962 // inferred without stalling; see CMSCollector::block_size_if_printezis_bits(). | |
963 size_t CompactibleFreeListSpace::block_size_no_stall(HeapWord* p, | |
964 const CMSCollector* c) | |
965 const { | |
966 assert(MemRegion(bottom(), end()).contains(p), "p not in space"); | |
967 // This must be volatile, or else there is a danger that the compiler | |
968 // will compile the code below into a sometimes-infinite loop, by keeping | |
969 // the value read the first time in a register. | |
970 DEBUG_ONLY(uint loops = 0;) | |
971 while (true) { | |
972 // We must do this until we get a consistent view of the object. | |
187 | 973 if (FreeChunk::indicatesFreeChunk(p)) { |
974 volatile FreeChunk* fc = (volatile FreeChunk*)p; | |
975 size_t res = fc->size(); | |
976 if (FreeChunk::indicatesFreeChunk(p)) { | |
0 | 977 assert(res != 0, "Block size should not be 0"); |
978 assert(loops == 0, "Should be 0"); | |
979 return res; | |
980 } | |
981 } else { | |
187 | 982 // must read from what 'p' points to in each loop. |
983 klassOop k = ((volatile oopDesc*)p)->klass_or_null(); | |
518
0af8b0718fc9
6692899: CMS: many vm.parallel_class_loading tests fail with assert "missing Printezis mark"
jmasa
parents:
517
diff
changeset
|
984 if (k != NULL && |
0af8b0718fc9
6692899: CMS: many vm.parallel_class_loading tests fail with assert "missing Printezis mark"
jmasa
parents:
517
diff
changeset
|
985 ((oopDesc*)p)->is_parsable() && |
0af8b0718fc9
6692899: CMS: many vm.parallel_class_loading tests fail with assert "missing Printezis mark"
jmasa
parents:
517
diff
changeset
|
986 ((oopDesc*)p)->is_conc_safe()) { |
187 | 987 assert(k->is_oop(), "Should really be klass oop."); |
988 oop o = (oop)p; | |
989 assert(o->is_oop(), "Should be an oop"); | |
990 size_t res = o->size_given_klass(k->klass_part()); | |
991 res = adjustObjectSize(res); | |
992 assert(res != 0, "Block size should not be 0"); | |
993 return res; | |
994 } else { | |
995 return c->block_size_if_printezis_bits(p); | |
996 } | |
0 | 997 } |
998 assert(loops == 0, "Can loop at most once"); | |
999 DEBUG_ONLY(loops++;) | |
1000 } | |
1001 } | |
1002 | |
1003 size_t CompactibleFreeListSpace::block_size_nopar(const HeapWord* p) const { | |
1004 NOT_PRODUCT(verify_objects_initialized()); | |
1005 assert(MemRegion(bottom(), end()).contains(p), "p not in space"); | |
1006 FreeChunk* fc = (FreeChunk*)p; | |
1007 if (fc->isFree()) { | |
1008 return fc->size(); | |
1009 } else { | |
1010 // Ignore mark word because this may be a recently promoted | |
1011 // object whose mark word is used to chain together grey | |
1012 // objects (the last one would have a null value). | |
1013 assert(oop(p)->is_oop(true), "Should be an oop"); | |
1014 return adjustObjectSize(oop(p)->size()); | |
1015 } | |
1016 } | |
1017 | |
1018 // This implementation assumes that the property of "being an object" is | |
1019 // stable. But being a free chunk may not be (because of parallel | |
1020 // promotion.) | |
1021 bool CompactibleFreeListSpace::block_is_obj(const HeapWord* p) const { | |
1022 FreeChunk* fc = (FreeChunk*)p; | |
1023 assert(is_in_reserved(p), "Should be in space"); | |
1024 // When doing a mark-sweep-compact of the CMS generation, this | |
1025 // assertion may fail because prepare_for_compaction() uses | |
1026 // space that is garbage to maintain information on ranges of | |
1027 // live objects so that these live ranges can be moved as a whole. | |
1028 // Comment out this assertion until that problem can be solved | |
1029 // (i.e., that the block start calculation may look at objects | |
1030 // at address below "p" in finding the object that contains "p" | |
1031 // and those objects (if garbage) may have been modified to hold | |
1032 // live range information. | |
1033 // assert(ParallelGCThreads > 0 || _bt.block_start(p) == p, "Should be a block boundary"); | |
187 | 1034 if (FreeChunk::indicatesFreeChunk(p)) return false; |
1035 klassOop k = oop(p)->klass_or_null(); | |
0 | 1036 if (k != NULL) { |
1037 // Ignore mark word because it may have been used to | |
1038 // chain together promoted objects (the last one | |
1039 // would have a null value). | |
1040 assert(oop(p)->is_oop(true), "Should be an oop"); | |
1041 return true; | |
1042 } else { | |
1043 return false; // Was not an object at the start of collection. | |
1044 } | |
1045 } | |
1046 | |
1047 // Check if the object is alive. This fact is checked either by consulting | |
1048 // the main marking bitmap in the sweeping phase or, if it's a permanent | |
1049 // generation and we're not in the sweeping phase, by checking the | |
1050 // perm_gen_verify_bit_map where we store the "deadness" information if | |
1051 // we did not sweep the perm gen in the most recent previous GC cycle. | |
1052 bool CompactibleFreeListSpace::obj_is_alive(const HeapWord* p) const { | |
1053 assert (block_is_obj(p), "The address should point to an object"); | |
1054 | |
1055 // If we're sweeping, we use object liveness information from the main bit map | |
1056 // for both perm gen and old gen. | |
1057 // We don't need to lock the bitmap (live_map or dead_map below), because | |
1058 // EITHER we are in the middle of the sweeping phase, and the | |
1059 // main marking bit map (live_map below) is locked, | |
1060 // OR we're in other phases and perm_gen_verify_bit_map (dead_map below) | |
1061 // is stable, because it's mutated only in the sweeping phase. | |
1062 if (_collector->abstract_state() == CMSCollector::Sweeping) { | |
1063 CMSBitMap* live_map = _collector->markBitMap(); | |
1064 return live_map->isMarked((HeapWord*) p); | |
1065 } else { | |
1066 // If we're not currently sweeping and we haven't swept the perm gen in | |
1067 // the previous concurrent cycle then we may have dead but unswept objects | |
1068 // in the perm gen. In this case, we use the "deadness" information | |
1069 // that we had saved in perm_gen_verify_bit_map at the last sweep. | |
1070 if (!CMSClassUnloadingEnabled && _collector->_permGen->reserved().contains(p)) { | |
1071 if (_collector->verifying()) { | |
1072 CMSBitMap* dead_map = _collector->perm_gen_verify_bit_map(); | |
1073 // Object is marked in the dead_map bitmap at the previous sweep | |
1074 // when we know that it's dead; if the bitmap is not allocated then | |
1075 // the object is alive. | |
1076 return (dead_map->sizeInBits() == 0) // bit_map has been allocated | |
1077 || !dead_map->par_isMarked((HeapWord*) p); | |
1078 } else { | |
1079 return false; // We can't say for sure if it's live, so we say that it's dead. | |
1080 } | |
1081 } | |
1082 } | |
1083 return true; | |
1084 } | |
1085 | |
1086 bool CompactibleFreeListSpace::block_is_obj_nopar(const HeapWord* p) const { | |
1087 FreeChunk* fc = (FreeChunk*)p; | |
1088 assert(is_in_reserved(p), "Should be in space"); | |
1089 assert(_bt.block_start(p) == p, "Should be a block boundary"); | |
1090 if (!fc->isFree()) { | |
1091 // Ignore mark word because it may have been used to | |
1092 // chain together promoted objects (the last one | |
1093 // would have a null value). | |
1094 assert(oop(p)->is_oop(true), "Should be an oop"); | |
1095 return true; | |
1096 } | |
1097 return false; | |
1098 } | |
1099 | |
1100 // "MT-safe but not guaranteed MT-precise" (TM); you may get an | |
1101 // approximate answer if you don't hold the freelistlock when you call this. | |
1102 size_t CompactibleFreeListSpace::totalSizeInIndexedFreeLists() const { | |
1103 size_t size = 0; | |
1104 for (size_t i = IndexSetStart; i < IndexSetSize; i += IndexSetStride) { | |
1105 debug_only( | |
1106 // We may be calling here without the lock in which case we | |
1107 // won't do this modest sanity check. | |
1108 if (freelistLock()->owned_by_self()) { | |
1109 size_t total_list_size = 0; | |
1110 for (FreeChunk* fc = _indexedFreeList[i].head(); fc != NULL; | |
1111 fc = fc->next()) { | |
1112 total_list_size += i; | |
1113 } | |
1114 assert(total_list_size == i * _indexedFreeList[i].count(), | |
1115 "Count in list is incorrect"); | |
1116 } | |
1117 ) | |
1118 size += i * _indexedFreeList[i].count(); | |
1119 } | |
1120 return size; | |
1121 } | |
1122 | |
1123 HeapWord* CompactibleFreeListSpace::par_allocate(size_t size) { | |
1124 MutexLockerEx x(freelistLock(), Mutex::_no_safepoint_check_flag); | |
1125 return allocate(size); | |
1126 } | |
1127 | |
1128 HeapWord* | |
1129 CompactibleFreeListSpace::getChunkFromSmallLinearAllocBlockRemainder(size_t size) { | |
1130 return getChunkFromLinearAllocBlockRemainder(&_smallLinearAllocBlock, size); | |
1131 } | |
1132 | |
1133 HeapWord* CompactibleFreeListSpace::allocate(size_t size) { | |
1134 assert_lock_strong(freelistLock()); | |
1135 HeapWord* res = NULL; | |
1136 assert(size == adjustObjectSize(size), | |
1137 "use adjustObjectSize() before calling into allocate()"); | |
1138 | |
1139 if (_adaptive_freelists) { | |
1140 res = allocate_adaptive_freelists(size); | |
1141 } else { // non-adaptive free lists | |
1142 res = allocate_non_adaptive_freelists(size); | |
1143 } | |
1144 | |
1145 if (res != NULL) { | |
1146 // check that res does lie in this space! | |
1147 assert(is_in_reserved(res), "Not in this space!"); | |
1148 assert(is_aligned((void*)res), "alignment check"); | |
1149 | |
1150 FreeChunk* fc = (FreeChunk*)res; | |
1151 fc->markNotFree(); | |
1152 assert(!fc->isFree(), "shouldn't be marked free"); | |
187 | 1153 assert(oop(fc)->klass_or_null() == NULL, "should look uninitialized"); |
0 | 1154 // Verify that the block offset table shows this to |
1155 // be a single block, but not one which is unallocated. | |
1156 _bt.verify_single_block(res, size); | |
1157 _bt.verify_not_unallocated(res, size); | |
1158 // mangle a just allocated object with a distinct pattern. | |
1159 debug_only(fc->mangleAllocated(size)); | |
1160 } | |
1161 | |
1162 return res; | |
1163 } | |
1164 | |
1165 HeapWord* CompactibleFreeListSpace::allocate_non_adaptive_freelists(size_t size) { | |
1166 HeapWord* res = NULL; | |
1167 // try and use linear allocation for smaller blocks | |
1168 if (size < _smallLinearAllocBlock._allocation_size_limit) { | |
1169 // if successful, the following also adjusts block offset table | |
1170 res = getChunkFromSmallLinearAllocBlock(size); | |
1171 } | |
1172 // Else triage to indexed lists for smaller sizes | |
1173 if (res == NULL) { | |
1174 if (size < SmallForDictionary) { | |
1175 res = (HeapWord*) getChunkFromIndexedFreeList(size); | |
1176 } else { | |
1177 // else get it from the big dictionary; if even this doesn't | |
1178 // work we are out of luck. | |
1179 res = (HeapWord*)getChunkFromDictionaryExact(size); | |
1180 } | |
1181 } | |
1182 | |
1183 return res; | |
1184 } | |
1185 | |
1186 HeapWord* CompactibleFreeListSpace::allocate_adaptive_freelists(size_t size) { | |
1187 assert_lock_strong(freelistLock()); | |
1188 HeapWord* res = NULL; | |
1189 assert(size == adjustObjectSize(size), | |
1190 "use adjustObjectSize() before calling into allocate()"); | |
1191 | |
1192 // Strategy | |
1193 // if small | |
1194 // exact size from small object indexed list if small | |
1195 // small or large linear allocation block (linAB) as appropriate | |
1196 // take from lists of greater sized chunks | |
1197 // else | |
1198 // dictionary | |
1199 // small or large linear allocation block if it has the space | |
1200 // Try allocating exact size from indexTable first | |
1201 if (size < IndexSetSize) { | |
1202 res = (HeapWord*) getChunkFromIndexedFreeList(size); | |
1203 if(res != NULL) { | |
1204 assert(res != (HeapWord*)_indexedFreeList[size].head(), | |
1205 "Not removed from free list"); | |
1206 // no block offset table adjustment is necessary on blocks in | |
1207 // the indexed lists. | |
1208 | |
1209 // Try allocating from the small LinAB | |
1210 } else if (size < _smallLinearAllocBlock._allocation_size_limit && | |
1211 (res = getChunkFromSmallLinearAllocBlock(size)) != NULL) { | |
1212 // if successful, the above also adjusts block offset table | |
1213 // Note that this call will refill the LinAB to | |
1214 // satisfy the request. This is different that | |
1215 // evm. | |
1216 // Don't record chunk off a LinAB? smallSplitBirth(size); | |
1217 | |
1218 } else { | |
1219 // Raid the exact free lists larger than size, even if they are not | |
1220 // overpopulated. | |
1221 res = (HeapWord*) getChunkFromGreater(size); | |
1222 } | |
1223 } else { | |
1224 // Big objects get allocated directly from the dictionary. | |
1225 res = (HeapWord*) getChunkFromDictionaryExact(size); | |
1226 if (res == NULL) { | |
1227 // Try hard not to fail since an allocation failure will likely | |
1228 // trigger a synchronous GC. Try to get the space from the | |
1229 // allocation blocks. | |
1230 res = getChunkFromSmallLinearAllocBlockRemainder(size); | |
1231 } | |
1232 } | |
1233 | |
1234 return res; | |
1235 } | |
1236 | |
1237 // A worst-case estimate of the space required (in HeapWords) to expand the heap | |
1238 // when promoting obj. | |
1239 size_t CompactibleFreeListSpace::expansionSpaceRequired(size_t obj_size) const { | |
1240 // Depending on the object size, expansion may require refilling either a | |
1241 // bigLAB or a smallLAB plus refilling a PromotionInfo object. MinChunkSize | |
1242 // is added because the dictionary may over-allocate to avoid fragmentation. | |
1243 size_t space = obj_size; | |
1244 if (!_adaptive_freelists) { | |
1245 space = MAX2(space, _smallLinearAllocBlock._refillSize); | |
1246 } | |
1247 space += _promoInfo.refillSize() + 2 * MinChunkSize; | |
1248 return space; | |
1249 } | |
1250 | |
1251 FreeChunk* CompactibleFreeListSpace::getChunkFromGreater(size_t numWords) { | |
1252 FreeChunk* ret; | |
1253 | |
1254 assert(numWords >= MinChunkSize, "Size is less than minimum"); | |
1255 assert(linearAllocationWouldFail() || bestFitFirst(), | |
1256 "Should not be here"); | |
1257 | |
1258 size_t i; | |
1259 size_t currSize = numWords + MinChunkSize; | |
1260 assert(currSize % MinObjAlignment == 0, "currSize should be aligned"); | |
1261 for (i = currSize; i < IndexSetSize; i += IndexSetStride) { | |
1262 FreeList* fl = &_indexedFreeList[i]; | |
1263 if (fl->head()) { | |
1264 ret = getFromListGreater(fl, numWords); | |
1265 assert(ret == NULL || ret->isFree(), "Should be returning a free chunk"); | |
1266 return ret; | |
1267 } | |
1268 } | |
1269 | |
1270 currSize = MAX2((size_t)SmallForDictionary, | |
1271 (size_t)(numWords + MinChunkSize)); | |
1272 | |
1273 /* Try to get a chunk that satisfies request, while avoiding | |
1274 fragmentation that can't be handled. */ | |
1275 { | |
1276 ret = dictionary()->getChunk(currSize); | |
1277 if (ret != NULL) { | |
1278 assert(ret->size() - numWords >= MinChunkSize, | |
1279 "Chunk is too small"); | |
1280 _bt.allocated((HeapWord*)ret, ret->size()); | |
1281 /* Carve returned chunk. */ | |
1282 (void) splitChunkAndReturnRemainder(ret, numWords); | |
1283 /* Label this as no longer a free chunk. */ | |
1284 assert(ret->isFree(), "This chunk should be free"); | |
1285 ret->linkPrev(NULL); | |
1286 } | |
1287 assert(ret == NULL || ret->isFree(), "Should be returning a free chunk"); | |
1288 return ret; | |
1289 } | |
1290 ShouldNotReachHere(); | |
1291 } | |
1292 | |
1293 bool CompactibleFreeListSpace::verifyChunkInIndexedFreeLists(FreeChunk* fc) | |
1294 const { | |
1295 assert(fc->size() < IndexSetSize, "Size of chunk is too large"); | |
1296 return _indexedFreeList[fc->size()].verifyChunkInFreeLists(fc); | |
1297 } | |
1298 | |
1299 bool CompactibleFreeListSpace::verifyChunkInFreeLists(FreeChunk* fc) const { | |
1300 if (fc->size() >= IndexSetSize) { | |
1301 return dictionary()->verifyChunkInFreeLists(fc); | |
1302 } else { | |
1303 return verifyChunkInIndexedFreeLists(fc); | |
1304 } | |
1305 } | |
1306 | |
1307 #ifndef PRODUCT | |
1308 void CompactibleFreeListSpace::assert_locked() const { | |
1309 CMSLockVerifier::assert_locked(freelistLock(), parDictionaryAllocLock()); | |
1310 } | |
1145
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
1311 |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
1312 void CompactibleFreeListSpace::assert_locked(const Mutex* lock) const { |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
1313 CMSLockVerifier::assert_locked(lock); |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
1314 } |
0 | 1315 #endif |
1316 | |
1317 FreeChunk* CompactibleFreeListSpace::allocateScratch(size_t size) { | |
1318 // In the parallel case, the main thread holds the free list lock | |
1319 // on behalf the parallel threads. | |
1320 FreeChunk* fc; | |
1321 { | |
1322 // If GC is parallel, this might be called by several threads. | |
1323 // This should be rare enough that the locking overhead won't affect | |
1324 // the sequential code. | |
1325 MutexLockerEx x(parDictionaryAllocLock(), | |
1326 Mutex::_no_safepoint_check_flag); | |
1327 fc = getChunkFromDictionary(size); | |
1328 } | |
1329 if (fc != NULL) { | |
1330 fc->dontCoalesce(); | |
1331 assert(fc->isFree(), "Should be free, but not coalescable"); | |
1332 // Verify that the block offset table shows this to | |
1333 // be a single block, but not one which is unallocated. | |
1334 _bt.verify_single_block((HeapWord*)fc, fc->size()); | |
1335 _bt.verify_not_unallocated((HeapWord*)fc, fc->size()); | |
1336 } | |
1337 return fc; | |
1338 } | |
1339 | |
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
12
diff
changeset
|
1340 oop CompactibleFreeListSpace::promote(oop obj, size_t obj_size) { |
0 | 1341 assert(obj_size == (size_t)obj->size(), "bad obj_size passed in"); |
1342 assert_locked(); | |
1343 | |
1344 // if we are tracking promotions, then first ensure space for | |
1345 // promotion (including spooling space for saving header if necessary). | |
1346 // then allocate and copy, then track promoted info if needed. | |
1347 // When tracking (see PromotionInfo::track()), the mark word may | |
1348 // be displaced and in this case restoration of the mark word | |
1349 // occurs in the (oop_since_save_marks_)iterate phase. | |
1350 if (_promoInfo.tracking() && !_promoInfo.ensure_spooling_space()) { | |
1351 return NULL; | |
1352 } | |
1353 // Call the allocate(size_t, bool) form directly to avoid the | |
1354 // additional call through the allocate(size_t) form. Having | |
1355 // the compile inline the call is problematic because allocate(size_t) | |
1356 // is a virtual method. | |
1357 HeapWord* res = allocate(adjustObjectSize(obj_size)); | |
1358 if (res != NULL) { | |
1359 Copy::aligned_disjoint_words((HeapWord*)obj, res, obj_size); | |
1360 // if we should be tracking promotions, do so. | |
1361 if (_promoInfo.tracking()) { | |
1362 _promoInfo.track((PromotedObject*)res); | |
1363 } | |
1364 } | |
1365 return oop(res); | |
1366 } | |
1367 | |
1368 HeapWord* | |
1369 CompactibleFreeListSpace::getChunkFromSmallLinearAllocBlock(size_t size) { | |
1370 assert_locked(); | |
1371 assert(size >= MinChunkSize, "minimum chunk size"); | |
1372 assert(size < _smallLinearAllocBlock._allocation_size_limit, | |
1373 "maximum from smallLinearAllocBlock"); | |
1374 return getChunkFromLinearAllocBlock(&_smallLinearAllocBlock, size); | |
1375 } | |
1376 | |
1377 HeapWord* | |
1378 CompactibleFreeListSpace::getChunkFromLinearAllocBlock(LinearAllocBlock *blk, | |
1379 size_t size) { | |
1380 assert_locked(); | |
1381 assert(size >= MinChunkSize, "too small"); | |
1382 HeapWord* res = NULL; | |
1383 // Try to do linear allocation from blk, making sure that | |
1384 if (blk->_word_size == 0) { | |
1385 // We have probably been unable to fill this either in the prologue or | |
1386 // when it was exhausted at the last linear allocation. Bail out until | |
1387 // next time. | |
1388 assert(blk->_ptr == NULL, "consistency check"); | |
1389 return NULL; | |
1390 } | |
1391 assert(blk->_word_size != 0 && blk->_ptr != NULL, "consistency check"); | |
1392 res = getChunkFromLinearAllocBlockRemainder(blk, size); | |
1393 if (res != NULL) return res; | |
1394 | |
1395 // about to exhaust this linear allocation block | |
1396 if (blk->_word_size == size) { // exactly satisfied | |
1397 res = blk->_ptr; | |
1398 _bt.allocated(res, blk->_word_size); | |
1399 } else if (size + MinChunkSize <= blk->_refillSize) { | |
1145
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
1400 size_t sz = blk->_word_size; |
0 | 1401 // Update _unallocated_block if the size is such that chunk would be |
1402 // returned to the indexed free list. All other chunks in the indexed | |
1403 // free lists are allocated from the dictionary so that _unallocated_block | |
1404 // has already been adjusted for them. Do it here so that the cost | |
1405 // for all chunks added back to the indexed free lists. | |
1145
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
1406 if (sz < SmallForDictionary) { |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
1407 _bt.allocated(blk->_ptr, sz); |
0 | 1408 } |
1409 // Return the chunk that isn't big enough, and then refill below. | |
1145
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
1410 addChunkToFreeLists(blk->_ptr, sz); |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
1411 splitBirth(sz); |
0 | 1412 // Don't keep statistics on adding back chunk from a LinAB. |
1413 } else { | |
1414 // A refilled block would not satisfy the request. | |
1415 return NULL; | |
1416 } | |
1417 | |
1418 blk->_ptr = NULL; blk->_word_size = 0; | |
1419 refillLinearAllocBlock(blk); | |
1420 assert(blk->_ptr == NULL || blk->_word_size >= size + MinChunkSize, | |
1421 "block was replenished"); | |
1422 if (res != NULL) { | |
1423 splitBirth(size); | |
1424 repairLinearAllocBlock(blk); | |
1425 } else if (blk->_ptr != NULL) { | |
1426 res = blk->_ptr; | |
1427 size_t blk_size = blk->_word_size; | |
1428 blk->_word_size -= size; | |
1429 blk->_ptr += size; | |
1430 splitBirth(size); | |
1431 repairLinearAllocBlock(blk); | |
1432 // Update BOT last so that other (parallel) GC threads see a consistent | |
1433 // view of the BOT and free blocks. | |
1434 // Above must occur before BOT is updated below. | |
1435 _bt.split_block(res, blk_size, size); // adjust block offset table | |
1436 } | |
1437 return res; | |
1438 } | |
1439 | |
1440 HeapWord* CompactibleFreeListSpace::getChunkFromLinearAllocBlockRemainder( | |
1441 LinearAllocBlock* blk, | |
1442 size_t size) { | |
1443 assert_locked(); | |
1444 assert(size >= MinChunkSize, "too small"); | |
1445 | |
1446 HeapWord* res = NULL; | |
1447 // This is the common case. Keep it simple. | |
1448 if (blk->_word_size >= size + MinChunkSize) { | |
1449 assert(blk->_ptr != NULL, "consistency check"); | |
1450 res = blk->_ptr; | |
1451 // Note that the BOT is up-to-date for the linAB before allocation. It | |
1452 // indicates the start of the linAB. The split_block() updates the | |
1453 // BOT for the linAB after the allocation (indicates the start of the | |
1454 // next chunk to be allocated). | |
1455 size_t blk_size = blk->_word_size; | |
1456 blk->_word_size -= size; | |
1457 blk->_ptr += size; | |
1458 splitBirth(size); | |
1459 repairLinearAllocBlock(blk); | |
1460 // Update BOT last so that other (parallel) GC threads see a consistent | |
1461 // view of the BOT and free blocks. | |
1462 // Above must occur before BOT is updated below. | |
1463 _bt.split_block(res, blk_size, size); // adjust block offset table | |
1464 _bt.allocated(res, size); | |
1465 } | |
1466 return res; | |
1467 } | |
1468 | |
1469 FreeChunk* | |
1470 CompactibleFreeListSpace::getChunkFromIndexedFreeList(size_t size) { | |
1471 assert_locked(); | |
1472 assert(size < SmallForDictionary, "just checking"); | |
1473 FreeChunk* res; | |
1474 res = _indexedFreeList[size].getChunkAtHead(); | |
1475 if (res == NULL) { | |
1476 res = getChunkFromIndexedFreeListHelper(size); | |
1477 } | |
1478 _bt.verify_not_unallocated((HeapWord*) res, size); | |
1145
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
1479 assert(res == NULL || res->size() == size, "Incorrect block size"); |
0 | 1480 return res; |
1481 } | |
1482 | |
1483 FreeChunk* | |
1145
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
1484 CompactibleFreeListSpace::getChunkFromIndexedFreeListHelper(size_t size, |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
1485 bool replenish) { |
0 | 1486 assert_locked(); |
1487 FreeChunk* fc = NULL; | |
1488 if (size < SmallForDictionary) { | |
1489 assert(_indexedFreeList[size].head() == NULL || | |
1490 _indexedFreeList[size].surplus() <= 0, | |
1491 "List for this size should be empty or under populated"); | |
1492 // Try best fit in exact lists before replenishing the list | |
1493 if (!bestFitFirst() || (fc = bestFitSmall(size)) == NULL) { | |
1494 // Replenish list. | |
1495 // | |
1496 // Things tried that failed. | |
1497 // Tried allocating out of the two LinAB's first before | |
1498 // replenishing lists. | |
1499 // Tried small linAB of size 256 (size in indexed list) | |
1500 // and replenishing indexed lists from the small linAB. | |
1501 // | |
1502 FreeChunk* newFc = NULL; | |
1145
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
1503 const size_t replenish_size = CMSIndexedFreeListReplenish * size; |
0 | 1504 if (replenish_size < SmallForDictionary) { |
1505 // Do not replenish from an underpopulated size. | |
1506 if (_indexedFreeList[replenish_size].surplus() > 0 && | |
1507 _indexedFreeList[replenish_size].head() != NULL) { | |
1145
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
1508 newFc = _indexedFreeList[replenish_size].getChunkAtHead(); |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
1509 } else if (bestFitFirst()) { |
0 | 1510 newFc = bestFitSmall(replenish_size); |
1511 } | |
1512 } | |
1145
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
1513 if (newFc == NULL && replenish_size > size) { |
0 | 1514 assert(CMSIndexedFreeListReplenish > 1, "ctl pt invariant"); |
1145
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
1515 newFc = getChunkFromIndexedFreeListHelper(replenish_size, false); |
0 | 1516 } |
1145
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
1517 // Note: The stats update re split-death of block obtained above |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
1518 // will be recorded below precisely when we know we are going to |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
1519 // be actually splitting it into more than one pieces below. |
0 | 1520 if (newFc != NULL) { |
1145
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
1521 if (replenish || CMSReplenishIntermediate) { |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
1522 // Replenish this list and return one block to caller. |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
1523 size_t i; |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
1524 FreeChunk *curFc, *nextFc; |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
1525 size_t num_blk = newFc->size() / size; |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
1526 assert(num_blk >= 1, "Smaller than requested?"); |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
1527 assert(newFc->size() % size == 0, "Should be integral multiple of request"); |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
1528 if (num_blk > 1) { |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
1529 // we are sure we will be splitting the block just obtained |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
1530 // into multiple pieces; record the split-death of the original |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
1531 splitDeath(replenish_size); |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
1532 } |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
1533 // carve up and link blocks 0, ..., num_blk - 2 |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
1534 // The last chunk is not added to the lists but is returned as the |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
1535 // free chunk. |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
1536 for (curFc = newFc, nextFc = (FreeChunk*)((HeapWord*)curFc + size), |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
1537 i = 0; |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
1538 i < (num_blk - 1); |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
1539 curFc = nextFc, nextFc = (FreeChunk*)((HeapWord*)nextFc + size), |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
1540 i++) { |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
1541 curFc->setSize(size); |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
1542 // Don't record this as a return in order to try and |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
1543 // determine the "returns" from a GC. |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
1544 _bt.verify_not_unallocated((HeapWord*) fc, size); |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
1545 _indexedFreeList[size].returnChunkAtTail(curFc, false); |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
1546 _bt.mark_block((HeapWord*)curFc, size); |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
1547 splitBirth(size); |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
1548 // Don't record the initial population of the indexed list |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
1549 // as a split birth. |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
1550 } |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
1551 |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
1552 // check that the arithmetic was OK above |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
1553 assert((HeapWord*)nextFc == (HeapWord*)newFc + num_blk*size, |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
1554 "inconsistency in carving newFc"); |
0 | 1555 curFc->setSize(size); |
1556 _bt.mark_block((HeapWord*)curFc, size); | |
1557 splitBirth(size); | |
1145
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
1558 fc = curFc; |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
1559 } else { |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
1560 // Return entire block to caller |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
1561 fc = newFc; |
0 | 1562 } |
1563 } | |
1564 } | |
1565 } else { | |
1566 // Get a free chunk from the free chunk dictionary to be returned to | |
1567 // replenish the indexed free list. | |
1568 fc = getChunkFromDictionaryExact(size); | |
1569 } | |
1145
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
1570 // assert(fc == NULL || fc->isFree(), "Should be returning a free chunk"); |
0 | 1571 return fc; |
1572 } | |
1573 | |
1574 FreeChunk* | |
1575 CompactibleFreeListSpace::getChunkFromDictionary(size_t size) { | |
1576 assert_locked(); | |
1577 FreeChunk* fc = _dictionary->getChunk(size); | |
1578 if (fc == NULL) { | |
1579 return NULL; | |
1580 } | |
1581 _bt.allocated((HeapWord*)fc, fc->size()); | |
1582 if (fc->size() >= size + MinChunkSize) { | |
1583 fc = splitChunkAndReturnRemainder(fc, size); | |
1584 } | |
1585 assert(fc->size() >= size, "chunk too small"); | |
1586 assert(fc->size() < size + MinChunkSize, "chunk too big"); | |
1587 _bt.verify_single_block((HeapWord*)fc, fc->size()); | |
1588 return fc; | |
1589 } | |
1590 | |
1591 FreeChunk* | |
1592 CompactibleFreeListSpace::getChunkFromDictionaryExact(size_t size) { | |
1593 assert_locked(); | |
1594 FreeChunk* fc = _dictionary->getChunk(size); | |
1595 if (fc == NULL) { | |
1596 return fc; | |
1597 } | |
1598 _bt.allocated((HeapWord*)fc, fc->size()); | |
1599 if (fc->size() == size) { | |
1600 _bt.verify_single_block((HeapWord*)fc, size); | |
1601 return fc; | |
1602 } | |
1603 assert(fc->size() > size, "getChunk() guarantee"); | |
1604 if (fc->size() < size + MinChunkSize) { | |
1605 // Return the chunk to the dictionary and go get a bigger one. | |
1606 returnChunkToDictionary(fc); | |
1607 fc = _dictionary->getChunk(size + MinChunkSize); | |
1608 if (fc == NULL) { | |
1609 return NULL; | |
1610 } | |
1611 _bt.allocated((HeapWord*)fc, fc->size()); | |
1612 } | |
1613 assert(fc->size() >= size + MinChunkSize, "tautology"); | |
1614 fc = splitChunkAndReturnRemainder(fc, size); | |
1615 assert(fc->size() == size, "chunk is wrong size"); | |
1616 _bt.verify_single_block((HeapWord*)fc, size); | |
1617 return fc; | |
1618 } | |
1619 | |
1620 void | |
1621 CompactibleFreeListSpace::returnChunkToDictionary(FreeChunk* chunk) { | |
1622 assert_locked(); | |
1623 | |
1624 size_t size = chunk->size(); | |
1625 _bt.verify_single_block((HeapWord*)chunk, size); | |
1626 // adjust _unallocated_block downward, as necessary | |
1627 _bt.freed((HeapWord*)chunk, size); | |
1628 _dictionary->returnChunk(chunk); | |
1145
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
1629 #ifndef PRODUCT |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
1630 if (CMSCollector::abstract_state() != CMSCollector::Sweeping) { |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
1631 TreeChunk::as_TreeChunk(chunk)->list()->verify_stats(); |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
1632 } |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
1633 #endif // PRODUCT |
0 | 1634 } |
1635 | |
1636 void | |
1637 CompactibleFreeListSpace::returnChunkToFreeList(FreeChunk* fc) { | |
1638 assert_locked(); | |
1639 size_t size = fc->size(); | |
1640 _bt.verify_single_block((HeapWord*) fc, size); | |
1641 _bt.verify_not_unallocated((HeapWord*) fc, size); | |
1642 if (_adaptive_freelists) { | |
1643 _indexedFreeList[size].returnChunkAtTail(fc); | |
1644 } else { | |
1645 _indexedFreeList[size].returnChunkAtHead(fc); | |
1646 } | |
1145
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
1647 #ifndef PRODUCT |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
1648 if (CMSCollector::abstract_state() != CMSCollector::Sweeping) { |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
1649 _indexedFreeList[size].verify_stats(); |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
1650 } |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
1651 #endif // PRODUCT |
0 | 1652 } |
1653 | |
1654 // Add chunk to end of last block -- if it's the largest | |
1655 // block -- and update BOT and census data. We would | |
1656 // of course have preferred to coalesce it with the | |
1657 // last block, but it's currently less expensive to find the | |
1658 // largest block than it is to find the last. | |
1659 void | |
1660 CompactibleFreeListSpace::addChunkToFreeListsAtEndRecordingStats( | |
1661 HeapWord* chunk, size_t size) { | |
1662 // check that the chunk does lie in this space! | |
1663 assert(chunk != NULL && is_in_reserved(chunk), "Not in this space!"); | |
1664 // One of the parallel gc task threads may be here | |
1665 // whilst others are allocating. | |
1666 Mutex* lock = NULL; | |
1667 if (ParallelGCThreads != 0) { | |
1668 lock = &_parDictionaryAllocLock; | |
1669 } | |
1670 FreeChunk* ec; | |
1671 { | |
1672 MutexLockerEx x(lock, Mutex::_no_safepoint_check_flag); | |
1673 ec = dictionary()->findLargestDict(); // get largest block | |
1674 if (ec != NULL && ec->end() == chunk) { | |
1675 // It's a coterminal block - we can coalesce. | |
1676 size_t old_size = ec->size(); | |
1677 coalDeath(old_size); | |
1678 removeChunkFromDictionary(ec); | |
1679 size += old_size; | |
1680 } else { | |
1681 ec = (FreeChunk*)chunk; | |
1682 } | |
1683 } | |
1684 ec->setSize(size); | |
1685 debug_only(ec->mangleFreed(size)); | |
1686 if (size < SmallForDictionary) { | |
1687 lock = _indexedFreeListParLocks[size]; | |
1688 } | |
1689 MutexLockerEx x(lock, Mutex::_no_safepoint_check_flag); | |
1690 addChunkAndRepairOffsetTable((HeapWord*)ec, size, true); | |
1691 // record the birth under the lock since the recording involves | |
1692 // manipulation of the list on which the chunk lives and | |
1693 // if the chunk is allocated and is the last on the list, | |
1694 // the list can go away. | |
1695 coalBirth(size); | |
1696 } | |
1697 | |
1698 void | |
1699 CompactibleFreeListSpace::addChunkToFreeLists(HeapWord* chunk, | |
1700 size_t size) { | |
1701 // check that the chunk does lie in this space! | |
1702 assert(chunk != NULL && is_in_reserved(chunk), "Not in this space!"); | |
1703 assert_locked(); | |
1704 _bt.verify_single_block(chunk, size); | |
1705 | |
1706 FreeChunk* fc = (FreeChunk*) chunk; | |
1707 fc->setSize(size); | |
1708 debug_only(fc->mangleFreed(size)); | |
1709 if (size < SmallForDictionary) { | |
1710 returnChunkToFreeList(fc); | |
1711 } else { | |
1712 returnChunkToDictionary(fc); | |
1713 } | |
1714 } | |
1715 | |
1716 void | |
1717 CompactibleFreeListSpace::addChunkAndRepairOffsetTable(HeapWord* chunk, | |
1718 size_t size, bool coalesced) { | |
1719 assert_locked(); | |
1720 assert(chunk != NULL, "null chunk"); | |
1721 if (coalesced) { | |
1722 // repair BOT | |
1723 _bt.single_block(chunk, size); | |
1724 } | |
1725 addChunkToFreeLists(chunk, size); | |
1726 } | |
1727 | |
1728 // We _must_ find the purported chunk on our free lists; | |
1729 // we assert if we don't. | |
1730 void | |
1731 CompactibleFreeListSpace::removeFreeChunkFromFreeLists(FreeChunk* fc) { | |
1732 size_t size = fc->size(); | |
1733 assert_locked(); | |
1734 debug_only(verifyFreeLists()); | |
1735 if (size < SmallForDictionary) { | |
1736 removeChunkFromIndexedFreeList(fc); | |
1737 } else { | |
1738 removeChunkFromDictionary(fc); | |
1739 } | |
1740 _bt.verify_single_block((HeapWord*)fc, size); | |
1741 debug_only(verifyFreeLists()); | |
1742 } | |
1743 | |
1744 void | |
1745 CompactibleFreeListSpace::removeChunkFromDictionary(FreeChunk* fc) { | |
1746 size_t size = fc->size(); | |
1747 assert_locked(); | |
1748 assert(fc != NULL, "null chunk"); | |
1749 _bt.verify_single_block((HeapWord*)fc, size); | |
1750 _dictionary->removeChunk(fc); | |
1751 // adjust _unallocated_block upward, as necessary | |
1752 _bt.allocated((HeapWord*)fc, size); | |
1753 } | |
1754 | |
1755 void | |
1756 CompactibleFreeListSpace::removeChunkFromIndexedFreeList(FreeChunk* fc) { | |
1757 assert_locked(); | |
1758 size_t size = fc->size(); | |
1759 _bt.verify_single_block((HeapWord*)fc, size); | |
1760 NOT_PRODUCT( | |
1761 if (FLSVerifyIndexTable) { | |
1762 verifyIndexedFreeList(size); | |
1763 } | |
1764 ) | |
1765 _indexedFreeList[size].removeChunk(fc); | |
1766 debug_only(fc->clearNext()); | |
1767 debug_only(fc->clearPrev()); | |
1768 NOT_PRODUCT( | |
1769 if (FLSVerifyIndexTable) { | |
1770 verifyIndexedFreeList(size); | |
1771 } | |
1772 ) | |
1773 } | |
1774 | |
1775 FreeChunk* CompactibleFreeListSpace::bestFitSmall(size_t numWords) { | |
1776 /* A hint is the next larger size that has a surplus. | |
1777 Start search at a size large enough to guarantee that | |
1778 the excess is >= MIN_CHUNK. */ | |
1779 size_t start = align_object_size(numWords + MinChunkSize); | |
1780 if (start < IndexSetSize) { | |
1781 FreeList* it = _indexedFreeList; | |
1782 size_t hint = _indexedFreeList[start].hint(); | |
1783 while (hint < IndexSetSize) { | |
1784 assert(hint % MinObjAlignment == 0, "hint should be aligned"); | |
1785 FreeList *fl = &_indexedFreeList[hint]; | |
1786 if (fl->surplus() > 0 && fl->head() != NULL) { | |
1787 // Found a list with surplus, reset original hint | |
1788 // and split out a free chunk which is returned. | |
1789 _indexedFreeList[start].set_hint(hint); | |
1790 FreeChunk* res = getFromListGreater(fl, numWords); | |
1791 assert(res == NULL || res->isFree(), | |
1792 "Should be returning a free chunk"); | |
1793 return res; | |
1794 } | |
1795 hint = fl->hint(); /* keep looking */ | |
1796 } | |
1797 /* None found. */ | |
1798 it[start].set_hint(IndexSetSize); | |
1799 } | |
1800 return NULL; | |
1801 } | |
1802 | |
1803 /* Requires fl->size >= numWords + MinChunkSize */ | |
1804 FreeChunk* CompactibleFreeListSpace::getFromListGreater(FreeList* fl, | |
1805 size_t numWords) { | |
1806 FreeChunk *curr = fl->head(); | |
1807 size_t oldNumWords = curr->size(); | |
1808 assert(numWords >= MinChunkSize, "Word size is too small"); | |
1809 assert(curr != NULL, "List is empty"); | |
1810 assert(oldNumWords >= numWords + MinChunkSize, | |
1811 "Size of chunks in the list is too small"); | |
1812 | |
1813 fl->removeChunk(curr); | |
1814 // recorded indirectly by splitChunkAndReturnRemainder - | |
1815 // smallSplit(oldNumWords, numWords); | |
1816 FreeChunk* new_chunk = splitChunkAndReturnRemainder(curr, numWords); | |
1817 // Does anything have to be done for the remainder in terms of | |
1818 // fixing the card table? | |
1819 assert(new_chunk == NULL || new_chunk->isFree(), | |
1820 "Should be returning a free chunk"); | |
1821 return new_chunk; | |
1822 } | |
1823 | |
1824 FreeChunk* | |
1825 CompactibleFreeListSpace::splitChunkAndReturnRemainder(FreeChunk* chunk, | |
1826 size_t new_size) { | |
1827 assert_locked(); | |
1828 size_t size = chunk->size(); | |
1829 assert(size > new_size, "Split from a smaller block?"); | |
1830 assert(is_aligned(chunk), "alignment problem"); | |
1831 assert(size == adjustObjectSize(size), "alignment problem"); | |
1832 size_t rem_size = size - new_size; | |
1833 assert(rem_size == adjustObjectSize(rem_size), "alignment problem"); | |
1834 assert(rem_size >= MinChunkSize, "Free chunk smaller than minimum"); | |
1835 FreeChunk* ffc = (FreeChunk*)((HeapWord*)chunk + new_size); | |
1836 assert(is_aligned(ffc), "alignment problem"); | |
1837 ffc->setSize(rem_size); | |
1838 ffc->linkNext(NULL); | |
1839 ffc->linkPrev(NULL); // Mark as a free block for other (parallel) GC threads. | |
1840 // Above must occur before BOT is updated below. | |
1841 // adjust block offset table | |
1842 _bt.split_block((HeapWord*)chunk, chunk->size(), new_size); | |
1843 if (rem_size < SmallForDictionary) { | |
1844 bool is_par = (SharedHeap::heap()->n_par_threads() > 0); | |
1845 if (is_par) _indexedFreeListParLocks[rem_size]->lock(); | |
1846 returnChunkToFreeList(ffc); | |
1847 split(size, rem_size); | |
1848 if (is_par) _indexedFreeListParLocks[rem_size]->unlock(); | |
1849 } else { | |
1850 returnChunkToDictionary(ffc); | |
1851 split(size ,rem_size); | |
1852 } | |
1853 chunk->setSize(new_size); | |
1854 return chunk; | |
1855 } | |
1856 | |
1857 void | |
1858 CompactibleFreeListSpace::sweep_completed() { | |
1859 // Now that space is probably plentiful, refill linear | |
1860 // allocation blocks as needed. | |
1861 refillLinearAllocBlocksIfNeeded(); | |
1862 } | |
1863 | |
1864 void | |
1865 CompactibleFreeListSpace::gc_prologue() { | |
1866 assert_locked(); | |
1867 if (PrintFLSStatistics != 0) { | |
1868 gclog_or_tty->print("Before GC:\n"); | |
1869 reportFreeListStatistics(); | |
1870 } | |
1871 refillLinearAllocBlocksIfNeeded(); | |
1872 } | |
1873 | |
1874 void | |
1875 CompactibleFreeListSpace::gc_epilogue() { | |
1876 assert_locked(); | |
1877 if (PrintGCDetails && Verbose && !_adaptive_freelists) { | |
1878 if (_smallLinearAllocBlock._word_size == 0) | |
1879 warning("CompactibleFreeListSpace(epilogue):: Linear allocation failure"); | |
1880 } | |
1881 assert(_promoInfo.noPromotions(), "_promoInfo inconsistency"); | |
1882 _promoInfo.stopTrackingPromotions(); | |
1883 repairLinearAllocationBlocks(); | |
1884 // Print Space's stats | |
1885 if (PrintFLSStatistics != 0) { | |
1886 gclog_or_tty->print("After GC:\n"); | |
1887 reportFreeListStatistics(); | |
1888 } | |
1889 } | |
1890 | |
1891 // Iteration support, mostly delegated from a CMS generation | |
1892 | |
1893 void CompactibleFreeListSpace::save_marks() { | |
1894 // mark the "end" of the used space at the time of this call; | |
1895 // note, however, that promoted objects from this point | |
1896 // on are tracked in the _promoInfo below. | |
1897 set_saved_mark_word(BlockOffsetArrayUseUnallocatedBlock ? | |
1898 unallocated_block() : end()); | |
1899 // inform allocator that promotions should be tracked. | |
1900 assert(_promoInfo.noPromotions(), "_promoInfo inconsistency"); | |
1901 _promoInfo.startTrackingPromotions(); | |
1902 } | |
1903 | |
1904 bool CompactibleFreeListSpace::no_allocs_since_save_marks() { | |
1905 assert(_promoInfo.tracking(), "No preceding save_marks?"); | |
1906 guarantee(SharedHeap::heap()->n_par_threads() == 0, | |
1907 "Shouldn't be called (yet) during parallel part of gc."); | |
1908 return _promoInfo.noPromotions(); | |
1909 } | |
1910 | |
1911 #define CFLS_OOP_SINCE_SAVE_MARKS_DEFN(OopClosureType, nv_suffix) \ | |
1912 \ | |
1913 void CompactibleFreeListSpace:: \ | |
1914 oop_since_save_marks_iterate##nv_suffix(OopClosureType* blk) { \ | |
1915 assert(SharedHeap::heap()->n_par_threads() == 0, \ | |
1916 "Shouldn't be called (yet) during parallel part of gc."); \ | |
1917 _promoInfo.promoted_oops_iterate##nv_suffix(blk); \ | |
1918 /* \ | |
1919 * This also restores any displaced headers and removes the elements from \ | |
1920 * the iteration set as they are processed, so that we have a clean slate \ | |
1921 * at the end of the iteration. Note, thus, that if new objects are \ | |
1922 * promoted as a result of the iteration they are iterated over as well. \ | |
1923 */ \ | |
1924 assert(_promoInfo.noPromotions(), "_promoInfo inconsistency"); \ | |
1925 } | |
1926 | |
1927 ALL_SINCE_SAVE_MARKS_CLOSURES(CFLS_OOP_SINCE_SAVE_MARKS_DEFN) | |
1928 | |
1929 ////////////////////////////////////////////////////////////////////////////// | |
1930 // We go over the list of promoted objects, removing each from the list, | |
1931 // and applying the closure (this may, in turn, add more elements to | |
1932 // the tail of the promoted list, and these newly added objects will | |
1933 // also be processed) until the list is empty. | |
1934 // To aid verification and debugging, in the non-product builds | |
1935 // we actually forward _promoHead each time we process a promoted oop. | |
1936 // Note that this is not necessary in general (i.e. when we don't need to | |
1937 // call PromotionInfo::verify()) because oop_iterate can only add to the | |
1938 // end of _promoTail, and never needs to look at _promoHead. | |
1939 | |
1940 #define PROMOTED_OOPS_ITERATE_DEFN(OopClosureType, nv_suffix) \ | |
1941 \ | |
1942 void PromotionInfo::promoted_oops_iterate##nv_suffix(OopClosureType* cl) { \ | |
1943 NOT_PRODUCT(verify()); \ | |
1944 PromotedObject *curObj, *nextObj; \ | |
1945 for (curObj = _promoHead; curObj != NULL; curObj = nextObj) { \ | |
1946 if ((nextObj = curObj->next()) == NULL) { \ | |
1947 /* protect ourselves against additions due to closure application \ | |
1948 below by resetting the list. */ \ | |
1949 assert(_promoTail == curObj, "Should have been the tail"); \ | |
1950 _promoHead = _promoTail = NULL; \ | |
1951 } \ | |
1952 if (curObj->hasDisplacedMark()) { \ | |
1953 /* restore displaced header */ \ | |
1954 oop(curObj)->set_mark(nextDisplacedHeader()); \ | |
1955 } else { \ | |
1956 /* restore prototypical header */ \ | |
1957 oop(curObj)->init_mark(); \ | |
1958 } \ | |
1959 /* The "promoted_mark" should now not be set */ \ | |
1960 assert(!curObj->hasPromotedMark(), \ | |
1961 "Should have been cleared by restoring displaced mark-word"); \ | |
1962 NOT_PRODUCT(_promoHead = nextObj); \ | |
1963 if (cl != NULL) oop(curObj)->oop_iterate(cl); \ | |
1964 if (nextObj == NULL) { /* start at head of list reset above */ \ | |
1965 nextObj = _promoHead; \ | |
1966 } \ | |
1967 } \ | |
1968 assert(noPromotions(), "post-condition violation"); \ | |
1969 assert(_promoHead == NULL && _promoTail == NULL, "emptied promoted list");\ | |
1970 assert(_spoolHead == _spoolTail, "emptied spooling buffers"); \ | |
1971 assert(_firstIndex == _nextIndex, "empty buffer"); \ | |
1972 } | |
1973 | |
1974 // This should have been ALL_SINCE_...() just like the others, | |
1975 // but, because the body of the method above is somehwat longer, | |
1976 // the MSVC compiler cannot cope; as a workaround, we split the | |
1977 // macro into its 3 constituent parts below (see original macro | |
1978 // definition in specializedOopClosures.hpp). | |
1979 SPECIALIZED_SINCE_SAVE_MARKS_CLOSURES_YOUNG(PROMOTED_OOPS_ITERATE_DEFN) | |
1980 PROMOTED_OOPS_ITERATE_DEFN(OopsInGenClosure,_v) | |
1981 | |
1982 | |
1983 void CompactibleFreeListSpace::object_iterate_since_last_GC(ObjectClosure* cl) { | |
1984 // ugghh... how would one do this efficiently for a non-contiguous space? | |
1985 guarantee(false, "NYI"); | |
1986 } | |
1987 | |
12
6432c3bb6240
6668743: CMS: Consolidate block statistics reporting code
ysr
parents:
0
diff
changeset
|
1988 bool CompactibleFreeListSpace::linearAllocationWouldFail() const { |
0 | 1989 return _smallLinearAllocBlock._word_size == 0; |
1990 } | |
1991 | |
1992 void CompactibleFreeListSpace::repairLinearAllocationBlocks() { | |
1993 // Fix up linear allocation blocks to look like free blocks | |
1994 repairLinearAllocBlock(&_smallLinearAllocBlock); | |
1995 } | |
1996 | |
1997 void CompactibleFreeListSpace::repairLinearAllocBlock(LinearAllocBlock* blk) { | |
1998 assert_locked(); | |
1999 if (blk->_ptr != NULL) { | |
2000 assert(blk->_word_size != 0 && blk->_word_size >= MinChunkSize, | |
2001 "Minimum block size requirement"); | |
2002 FreeChunk* fc = (FreeChunk*)(blk->_ptr); | |
2003 fc->setSize(blk->_word_size); | |
2004 fc->linkPrev(NULL); // mark as free | |
2005 fc->dontCoalesce(); | |
2006 assert(fc->isFree(), "just marked it free"); | |
2007 assert(fc->cantCoalesce(), "just marked it uncoalescable"); | |
2008 } | |
2009 } | |
2010 | |
2011 void CompactibleFreeListSpace::refillLinearAllocBlocksIfNeeded() { | |
2012 assert_locked(); | |
2013 if (_smallLinearAllocBlock._ptr == NULL) { | |
2014 assert(_smallLinearAllocBlock._word_size == 0, | |
2015 "Size of linAB should be zero if the ptr is NULL"); | |
2016 // Reset the linAB refill and allocation size limit. | |
2017 _smallLinearAllocBlock.set(0, 0, 1024*SmallForLinearAlloc, SmallForLinearAlloc); | |
2018 } | |
2019 refillLinearAllocBlockIfNeeded(&_smallLinearAllocBlock); | |
2020 } | |
2021 | |
2022 void | |
2023 CompactibleFreeListSpace::refillLinearAllocBlockIfNeeded(LinearAllocBlock* blk) { | |
2024 assert_locked(); | |
2025 assert((blk->_ptr == NULL && blk->_word_size == 0) || | |
2026 (blk->_ptr != NULL && blk->_word_size >= MinChunkSize), | |
2027 "blk invariant"); | |
2028 if (blk->_ptr == NULL) { | |
2029 refillLinearAllocBlock(blk); | |
2030 } | |
2031 if (PrintMiscellaneous && Verbose) { | |
2032 if (blk->_word_size == 0) { | |
2033 warning("CompactibleFreeListSpace(prologue):: Linear allocation failure"); | |
2034 } | |
2035 } | |
2036 } | |
2037 | |
2038 void | |
2039 CompactibleFreeListSpace::refillLinearAllocBlock(LinearAllocBlock* blk) { | |
2040 assert_locked(); | |
2041 assert(blk->_word_size == 0 && blk->_ptr == NULL, | |
2042 "linear allocation block should be empty"); | |
2043 FreeChunk* fc; | |
2044 if (blk->_refillSize < SmallForDictionary && | |
2045 (fc = getChunkFromIndexedFreeList(blk->_refillSize)) != NULL) { | |
2046 // A linAB's strategy might be to use small sizes to reduce | |
2047 // fragmentation but still get the benefits of allocation from a | |
2048 // linAB. | |
2049 } else { | |
2050 fc = getChunkFromDictionary(blk->_refillSize); | |
2051 } | |
2052 if (fc != NULL) { | |
2053 blk->_ptr = (HeapWord*)fc; | |
2054 blk->_word_size = fc->size(); | |
2055 fc->dontCoalesce(); // to prevent sweeper from sweeping us up | |
2056 } | |
2057 } | |
2058 | |
12
6432c3bb6240
6668743: CMS: Consolidate block statistics reporting code
ysr
parents:
0
diff
changeset
|
2059 // Support for concurrent collection policy decisions. |
6432c3bb6240
6668743: CMS: Consolidate block statistics reporting code
ysr
parents:
0
diff
changeset
|
2060 bool CompactibleFreeListSpace::should_concurrent_collect() const { |
6432c3bb6240
6668743: CMS: Consolidate block statistics reporting code
ysr
parents:
0
diff
changeset
|
2061 // In the future we might want to add in frgamentation stats -- |
6432c3bb6240
6668743: CMS: Consolidate block statistics reporting code
ysr
parents:
0
diff
changeset
|
2062 // including erosion of the "mountain" into this decision as well. |
6432c3bb6240
6668743: CMS: Consolidate block statistics reporting code
ysr
parents:
0
diff
changeset
|
2063 return !adaptive_freelists() && linearAllocationWouldFail(); |
6432c3bb6240
6668743: CMS: Consolidate block statistics reporting code
ysr
parents:
0
diff
changeset
|
2064 } |
6432c3bb6240
6668743: CMS: Consolidate block statistics reporting code
ysr
parents:
0
diff
changeset
|
2065 |
0 | 2066 // Support for compaction |
2067 | |
2068 void CompactibleFreeListSpace::prepare_for_compaction(CompactPoint* cp) { | |
2069 SCAN_AND_FORWARD(cp,end,block_is_obj,block_size); | |
2070 // prepare_for_compaction() uses the space between live objects | |
2071 // so that later phase can skip dead space quickly. So verification | |
2072 // of the free lists doesn't work after. | |
2073 } | |
2074 | |
2075 #define obj_size(q) adjustObjectSize(oop(q)->size()) | |
2076 #define adjust_obj_size(s) adjustObjectSize(s) | |
2077 | |
2078 void CompactibleFreeListSpace::adjust_pointers() { | |
2079 // In other versions of adjust_pointers(), a bail out | |
2080 // based on the amount of live data in the generation | |
2081 // (i.e., if 0, bail out) may be used. | |
2082 // Cannot test used() == 0 here because the free lists have already | |
2083 // been mangled by the compaction. | |
2084 | |
2085 SCAN_AND_ADJUST_POINTERS(adjust_obj_size); | |
2086 // See note about verification in prepare_for_compaction(). | |
2087 } | |
2088 | |
2089 void CompactibleFreeListSpace::compact() { | |
2090 SCAN_AND_COMPACT(obj_size); | |
2091 } | |
2092 | |
2093 // fragmentation_metric = 1 - [sum of (fbs**2) / (sum of fbs)**2] | |
2094 // where fbs is free block sizes | |
2095 double CompactibleFreeListSpace::flsFrag() const { | |
2096 size_t itabFree = totalSizeInIndexedFreeLists(); | |
2097 double frag = 0.0; | |
2098 size_t i; | |
2099 | |
2100 for (i = IndexSetStart; i < IndexSetSize; i += IndexSetStride) { | |
2101 double sz = i; | |
2102 frag += _indexedFreeList[i].count() * (sz * sz); | |
2103 } | |
2104 | |
2105 double totFree = itabFree + | |
2106 _dictionary->totalChunkSize(DEBUG_ONLY(freelistLock())); | |
2107 if (totFree > 0) { | |
2108 frag = ((frag + _dictionary->sum_of_squared_block_sizes()) / | |
2109 (totFree * totFree)); | |
2110 frag = (double)1.0 - frag; | |
2111 } else { | |
2112 assert(frag == 0.0, "Follows from totFree == 0"); | |
2113 } | |
2114 return frag; | |
2115 } | |
2116 | |
2117 void CompactibleFreeListSpace::beginSweepFLCensus( | |
2118 float inter_sweep_current, | |
1145
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2119 float inter_sweep_estimate, |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2120 float intra_sweep_estimate) { |
0 | 2121 assert_locked(); |
2122 size_t i; | |
2123 for (i = IndexSetStart; i < IndexSetSize; i += IndexSetStride) { | |
2124 FreeList* fl = &_indexedFreeList[i]; | |
1145
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2125 if (PrintFLSStatistics > 1) { |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2126 gclog_or_tty->print("size[%d] : ", i); |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2127 } |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2128 fl->compute_desired(inter_sweep_current, inter_sweep_estimate, intra_sweep_estimate); |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2129 fl->set_coalDesired((ssize_t)((double)fl->desired() * CMSSmallCoalSurplusPercent)); |
0 | 2130 fl->set_beforeSweep(fl->count()); |
2131 fl->set_bfrSurp(fl->surplus()); | |
2132 } | |
1145
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2133 _dictionary->beginSweepDictCensus(CMSLargeCoalSurplusPercent, |
0 | 2134 inter_sweep_current, |
1145
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2135 inter_sweep_estimate, |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2136 intra_sweep_estimate); |
0 | 2137 } |
2138 | |
2139 void CompactibleFreeListSpace::setFLSurplus() { | |
2140 assert_locked(); | |
2141 size_t i; | |
2142 for (i = IndexSetStart; i < IndexSetSize; i += IndexSetStride) { | |
2143 FreeList *fl = &_indexedFreeList[i]; | |
2144 fl->set_surplus(fl->count() - | |
1145
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2145 (ssize_t)((double)fl->desired() * CMSSmallSplitSurplusPercent)); |
0 | 2146 } |
2147 } | |
2148 | |
2149 void CompactibleFreeListSpace::setFLHints() { | |
2150 assert_locked(); | |
2151 size_t i; | |
2152 size_t h = IndexSetSize; | |
2153 for (i = IndexSetSize - 1; i != 0; i -= IndexSetStride) { | |
2154 FreeList *fl = &_indexedFreeList[i]; | |
2155 fl->set_hint(h); | |
2156 if (fl->surplus() > 0) { | |
2157 h = i; | |
2158 } | |
2159 } | |
2160 } | |
2161 | |
2162 void CompactibleFreeListSpace::clearFLCensus() { | |
2163 assert_locked(); | |
2164 int i; | |
2165 for (i = IndexSetStart; i < IndexSetSize; i += IndexSetStride) { | |
2166 FreeList *fl = &_indexedFreeList[i]; | |
2167 fl->set_prevSweep(fl->count()); | |
2168 fl->set_coalBirths(0); | |
2169 fl->set_coalDeaths(0); | |
2170 fl->set_splitBirths(0); | |
2171 fl->set_splitDeaths(0); | |
2172 } | |
2173 } | |
2174 | |
12
6432c3bb6240
6668743: CMS: Consolidate block statistics reporting code
ysr
parents:
0
diff
changeset
|
2175 void CompactibleFreeListSpace::endSweepFLCensus(size_t sweep_count) { |
1145
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2176 if (PrintFLSStatistics > 0) { |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2177 HeapWord* largestAddr = (HeapWord*) dictionary()->findLargestDict(); |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2178 gclog_or_tty->print_cr("CMS: Large block " PTR_FORMAT, |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2179 largestAddr); |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2180 } |
0 | 2181 setFLSurplus(); |
2182 setFLHints(); | |
2183 if (PrintGC && PrintFLSCensus > 0) { | |
12
6432c3bb6240
6668743: CMS: Consolidate block statistics reporting code
ysr
parents:
0
diff
changeset
|
2184 printFLCensus(sweep_count); |
0 | 2185 } |
2186 clearFLCensus(); | |
2187 assert_locked(); | |
1145
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2188 _dictionary->endSweepDictCensus(CMSLargeSplitSurplusPercent); |
0 | 2189 } |
2190 | |
2191 bool CompactibleFreeListSpace::coalOverPopulated(size_t size) { | |
2192 if (size < SmallForDictionary) { | |
2193 FreeList *fl = &_indexedFreeList[size]; | |
2194 return (fl->coalDesired() < 0) || | |
2195 ((int)fl->count() > fl->coalDesired()); | |
2196 } else { | |
2197 return dictionary()->coalDictOverPopulated(size); | |
2198 } | |
2199 } | |
2200 | |
2201 void CompactibleFreeListSpace::smallCoalBirth(size_t size) { | |
2202 assert(size < SmallForDictionary, "Size too large for indexed list"); | |
2203 FreeList *fl = &_indexedFreeList[size]; | |
2204 fl->increment_coalBirths(); | |
2205 fl->increment_surplus(); | |
2206 } | |
2207 | |
2208 void CompactibleFreeListSpace::smallCoalDeath(size_t size) { | |
2209 assert(size < SmallForDictionary, "Size too large for indexed list"); | |
2210 FreeList *fl = &_indexedFreeList[size]; | |
2211 fl->increment_coalDeaths(); | |
2212 fl->decrement_surplus(); | |
2213 } | |
2214 | |
2215 void CompactibleFreeListSpace::coalBirth(size_t size) { | |
2216 if (size < SmallForDictionary) { | |
2217 smallCoalBirth(size); | |
2218 } else { | |
2219 dictionary()->dictCensusUpdate(size, | |
2220 false /* split */, | |
2221 true /* birth */); | |
2222 } | |
2223 } | |
2224 | |
2225 void CompactibleFreeListSpace::coalDeath(size_t size) { | |
2226 if(size < SmallForDictionary) { | |
2227 smallCoalDeath(size); | |
2228 } else { | |
2229 dictionary()->dictCensusUpdate(size, | |
2230 false /* split */, | |
2231 false /* birth */); | |
2232 } | |
2233 } | |
2234 | |
2235 void CompactibleFreeListSpace::smallSplitBirth(size_t size) { | |
2236 assert(size < SmallForDictionary, "Size too large for indexed list"); | |
2237 FreeList *fl = &_indexedFreeList[size]; | |
2238 fl->increment_splitBirths(); | |
2239 fl->increment_surplus(); | |
2240 } | |
2241 | |
2242 void CompactibleFreeListSpace::smallSplitDeath(size_t size) { | |
2243 assert(size < SmallForDictionary, "Size too large for indexed list"); | |
2244 FreeList *fl = &_indexedFreeList[size]; | |
2245 fl->increment_splitDeaths(); | |
2246 fl->decrement_surplus(); | |
2247 } | |
2248 | |
2249 void CompactibleFreeListSpace::splitBirth(size_t size) { | |
2250 if (size < SmallForDictionary) { | |
2251 smallSplitBirth(size); | |
2252 } else { | |
2253 dictionary()->dictCensusUpdate(size, | |
2254 true /* split */, | |
2255 true /* birth */); | |
2256 } | |
2257 } | |
2258 | |
2259 void CompactibleFreeListSpace::splitDeath(size_t size) { | |
2260 if (size < SmallForDictionary) { | |
2261 smallSplitDeath(size); | |
2262 } else { | |
2263 dictionary()->dictCensusUpdate(size, | |
2264 true /* split */, | |
2265 false /* birth */); | |
2266 } | |
2267 } | |
2268 | |
2269 void CompactibleFreeListSpace::split(size_t from, size_t to1) { | |
2270 size_t to2 = from - to1; | |
2271 splitDeath(from); | |
2272 splitBirth(to1); | |
2273 splitBirth(to2); | |
2274 } | |
2275 | |
2276 void CompactibleFreeListSpace::print() const { | |
2277 tty->print(" CompactibleFreeListSpace"); | |
2278 Space::print(); | |
2279 } | |
2280 | |
2281 void CompactibleFreeListSpace::prepare_for_verify() { | |
2282 assert_locked(); | |
2283 repairLinearAllocationBlocks(); | |
2284 // Verify that the SpoolBlocks look like free blocks of | |
2285 // appropriate sizes... To be done ... | |
2286 } | |
2287 | |
2288 class VerifyAllBlksClosure: public BlkClosure { | |
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
12
diff
changeset
|
2289 private: |
0 | 2290 const CompactibleFreeListSpace* _sp; |
2291 const MemRegion _span; | |
2292 | |
2293 public: | |
2294 VerifyAllBlksClosure(const CompactibleFreeListSpace* sp, | |
2295 MemRegion span) : _sp(sp), _span(span) { } | |
2296 | |
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
12
diff
changeset
|
2297 virtual size_t do_blk(HeapWord* addr) { |
0 | 2298 size_t res; |
2299 if (_sp->block_is_obj(addr)) { | |
2300 oop p = oop(addr); | |
2301 guarantee(p->is_oop(), "Should be an oop"); | |
2302 res = _sp->adjustObjectSize(p->size()); | |
2303 if (_sp->obj_is_alive(addr)) { | |
2304 p->verify(); | |
2305 } | |
2306 } else { | |
2307 FreeChunk* fc = (FreeChunk*)addr; | |
2308 res = fc->size(); | |
2309 if (FLSVerifyLists && !fc->cantCoalesce()) { | |
2310 guarantee(_sp->verifyChunkInFreeLists(fc), | |
2311 "Chunk should be on a free list"); | |
2312 } | |
2313 } | |
2314 guarantee(res != 0, "Livelock: no rank reduction!"); | |
2315 return res; | |
2316 } | |
2317 }; | |
2318 | |
2319 class VerifyAllOopsClosure: public OopClosure { | |
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
12
diff
changeset
|
2320 private: |
0 | 2321 const CMSCollector* _collector; |
2322 const CompactibleFreeListSpace* _sp; | |
2323 const MemRegion _span; | |
2324 const bool _past_remark; | |
2325 const CMSBitMap* _bit_map; | |
2326 | |
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
12
diff
changeset
|
2327 protected: |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
12
diff
changeset
|
2328 void do_oop(void* p, oop obj) { |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
12
diff
changeset
|
2329 if (_span.contains(obj)) { // the interior oop points into CMS heap |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
12
diff
changeset
|
2330 if (!_span.contains(p)) { // reference from outside CMS heap |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
12
diff
changeset
|
2331 // Should be a valid object; the first disjunct below allows |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
12
diff
changeset
|
2332 // us to sidestep an assertion in block_is_obj() that insists |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
12
diff
changeset
|
2333 // that p be in _sp. Note that several generations (and spaces) |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
12
diff
changeset
|
2334 // are spanned by _span (CMS heap) above. |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
12
diff
changeset
|
2335 guarantee(!_sp->is_in_reserved(obj) || |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
12
diff
changeset
|
2336 _sp->block_is_obj((HeapWord*)obj), |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
12
diff
changeset
|
2337 "Should be an object"); |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
12
diff
changeset
|
2338 guarantee(obj->is_oop(), "Should be an oop"); |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
12
diff
changeset
|
2339 obj->verify(); |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
12
diff
changeset
|
2340 if (_past_remark) { |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
12
diff
changeset
|
2341 // Remark has been completed, the object should be marked |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
12
diff
changeset
|
2342 _bit_map->isMarked((HeapWord*)obj); |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
12
diff
changeset
|
2343 } |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
12
diff
changeset
|
2344 } else { // reference within CMS heap |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
12
diff
changeset
|
2345 if (_past_remark) { |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
12
diff
changeset
|
2346 // Remark has been completed -- so the referent should have |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
12
diff
changeset
|
2347 // been marked, if referring object is. |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
12
diff
changeset
|
2348 if (_bit_map->isMarked(_collector->block_start(p))) { |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
12
diff
changeset
|
2349 guarantee(_bit_map->isMarked((HeapWord*)obj), "Marking error?"); |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
12
diff
changeset
|
2350 } |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
12
diff
changeset
|
2351 } |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
12
diff
changeset
|
2352 } |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
12
diff
changeset
|
2353 } else if (_sp->is_in_reserved(p)) { |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
12
diff
changeset
|
2354 // the reference is from FLS, and points out of FLS |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
12
diff
changeset
|
2355 guarantee(obj->is_oop(), "Should be an oop"); |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
12
diff
changeset
|
2356 obj->verify(); |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
12
diff
changeset
|
2357 } |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
12
diff
changeset
|
2358 } |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
12
diff
changeset
|
2359 |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
12
diff
changeset
|
2360 template <class T> void do_oop_work(T* p) { |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
12
diff
changeset
|
2361 T heap_oop = oopDesc::load_heap_oop(p); |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
12
diff
changeset
|
2362 if (!oopDesc::is_null(heap_oop)) { |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
12
diff
changeset
|
2363 oop obj = oopDesc::decode_heap_oop_not_null(heap_oop); |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
12
diff
changeset
|
2364 do_oop(p, obj); |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
12
diff
changeset
|
2365 } |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
12
diff
changeset
|
2366 } |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
12
diff
changeset
|
2367 |
0 | 2368 public: |
2369 VerifyAllOopsClosure(const CMSCollector* collector, | |
2370 const CompactibleFreeListSpace* sp, MemRegion span, | |
2371 bool past_remark, CMSBitMap* bit_map) : | |
2372 OopClosure(), _collector(collector), _sp(sp), _span(span), | |
2373 _past_remark(past_remark), _bit_map(bit_map) { } | |
2374 | |
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
12
diff
changeset
|
2375 virtual void do_oop(oop* p) { VerifyAllOopsClosure::do_oop_work(p); } |
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
12
diff
changeset
|
2376 virtual void do_oop(narrowOop* p) { VerifyAllOopsClosure::do_oop_work(p); } |
0 | 2377 }; |
2378 | |
2379 void CompactibleFreeListSpace::verify(bool ignored) const { | |
2380 assert_lock_strong(&_freelistLock); | |
2381 verify_objects_initialized(); | |
2382 MemRegion span = _collector->_span; | |
2383 bool past_remark = (_collector->abstract_state() == | |
2384 CMSCollector::Sweeping); | |
2385 | |
2386 ResourceMark rm; | |
2387 HandleMark hm; | |
2388 | |
2389 // Check integrity of CFL data structures | |
2390 _promoInfo.verify(); | |
2391 _dictionary->verify(); | |
2392 if (FLSVerifyIndexTable) { | |
2393 verifyIndexedFreeLists(); | |
2394 } | |
2395 // Check integrity of all objects and free blocks in space | |
2396 { | |
2397 VerifyAllBlksClosure cl(this, span); | |
2398 ((CompactibleFreeListSpace*)this)->blk_iterate(&cl); // cast off const | |
2399 } | |
2400 // Check that all references in the heap to FLS | |
2401 // are to valid objects in FLS or that references in | |
2402 // FLS are to valid objects elsewhere in the heap | |
2403 if (FLSVerifyAllHeapReferences) | |
2404 { | |
2405 VerifyAllOopsClosure cl(_collector, this, span, past_remark, | |
2406 _collector->markBitMap()); | |
2407 CollectedHeap* ch = Universe::heap(); | |
2408 ch->oop_iterate(&cl); // all oops in generations | |
2409 ch->permanent_oop_iterate(&cl); // all oops in perm gen | |
2410 } | |
2411 | |
2412 if (VerifyObjectStartArray) { | |
2413 // Verify the block offset table | |
2414 _bt.verify(); | |
2415 } | |
2416 } | |
2417 | |
2418 #ifndef PRODUCT | |
2419 void CompactibleFreeListSpace::verifyFreeLists() const { | |
2420 if (FLSVerifyLists) { | |
2421 _dictionary->verify(); | |
2422 verifyIndexedFreeLists(); | |
2423 } else { | |
2424 if (FLSVerifyDictionary) { | |
2425 _dictionary->verify(); | |
2426 } | |
2427 if (FLSVerifyIndexTable) { | |
2428 verifyIndexedFreeLists(); | |
2429 } | |
2430 } | |
2431 } | |
2432 #endif | |
2433 | |
2434 void CompactibleFreeListSpace::verifyIndexedFreeLists() const { | |
2435 size_t i = 0; | |
2436 for (; i < MinChunkSize; i++) { | |
2437 guarantee(_indexedFreeList[i].head() == NULL, "should be NULL"); | |
2438 } | |
2439 for (; i < IndexSetSize; i++) { | |
2440 verifyIndexedFreeList(i); | |
2441 } | |
2442 } | |
2443 | |
2444 void CompactibleFreeListSpace::verifyIndexedFreeList(size_t size) const { | |
1145
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2445 FreeChunk* fc = _indexedFreeList[size].head(); |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2446 FreeChunk* tail = _indexedFreeList[size].tail(); |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2447 size_t num = _indexedFreeList[size].count(); |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2448 size_t n = 0; |
342
37f87013dfd8
6711316: Open source the Garbage-First garbage collector
ysr
parents:
113
diff
changeset
|
2449 guarantee((size % 2 == 0) || fc == NULL, "Odd slots should be empty"); |
1145
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2450 for (; fc != NULL; fc = fc->next(), n++) { |
0 | 2451 guarantee(fc->size() == size, "Size inconsistency"); |
2452 guarantee(fc->isFree(), "!free?"); | |
2453 guarantee(fc->next() == NULL || fc->next()->prev() == fc, "Broken list"); | |
1145
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2454 guarantee((fc->next() == NULL) == (fc == tail), "Incorrect tail"); |
0 | 2455 } |
1145
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2456 guarantee(n == num, "Incorrect count"); |
0 | 2457 } |
2458 | |
2459 #ifndef PRODUCT | |
2460 void CompactibleFreeListSpace::checkFreeListConsistency() const { | |
2461 assert(_dictionary->minSize() <= IndexSetSize, | |
2462 "Some sizes can't be allocated without recourse to" | |
2463 " linear allocation buffers"); | |
2464 assert(MIN_TREE_CHUNK_SIZE*HeapWordSize == sizeof(TreeChunk), | |
2465 "else MIN_TREE_CHUNK_SIZE is wrong"); | |
2466 assert((IndexSetStride == 2 && IndexSetStart == 2) || | |
2467 (IndexSetStride == 1 && IndexSetStart == 1), "just checking"); | |
2468 assert((IndexSetStride != 2) || (MinChunkSize % 2 == 0), | |
2469 "Some for-loops may be incorrectly initialized"); | |
2470 assert((IndexSetStride != 2) || (IndexSetSize % 2 == 1), | |
2471 "For-loops that iterate over IndexSet with stride 2 may be wrong"); | |
2472 } | |
2473 #endif | |
2474 | |
12
6432c3bb6240
6668743: CMS: Consolidate block statistics reporting code
ysr
parents:
0
diff
changeset
|
2475 void CompactibleFreeListSpace::printFLCensus(size_t sweep_count) const { |
0 | 2476 assert_lock_strong(&_freelistLock); |
12
6432c3bb6240
6668743: CMS: Consolidate block statistics reporting code
ysr
parents:
0
diff
changeset
|
2477 FreeList total; |
6432c3bb6240
6668743: CMS: Consolidate block statistics reporting code
ysr
parents:
0
diff
changeset
|
2478 gclog_or_tty->print("end sweep# " SIZE_FORMAT "\n", sweep_count); |
6432c3bb6240
6668743: CMS: Consolidate block statistics reporting code
ysr
parents:
0
diff
changeset
|
2479 FreeList::print_labels_on(gclog_or_tty, "size"); |
0 | 2480 size_t totalFree = 0; |
2481 for (size_t i = IndexSetStart; i < IndexSetSize; i += IndexSetStride) { | |
2482 const FreeList *fl = &_indexedFreeList[i]; | |
12
6432c3bb6240
6668743: CMS: Consolidate block statistics reporting code
ysr
parents:
0
diff
changeset
|
2483 totalFree += fl->count() * fl->size(); |
6432c3bb6240
6668743: CMS: Consolidate block statistics reporting code
ysr
parents:
0
diff
changeset
|
2484 if (i % (40*IndexSetStride) == 0) { |
6432c3bb6240
6668743: CMS: Consolidate block statistics reporting code
ysr
parents:
0
diff
changeset
|
2485 FreeList::print_labels_on(gclog_or_tty, "size"); |
6432c3bb6240
6668743: CMS: Consolidate block statistics reporting code
ysr
parents:
0
diff
changeset
|
2486 } |
6432c3bb6240
6668743: CMS: Consolidate block statistics reporting code
ysr
parents:
0
diff
changeset
|
2487 fl->print_on(gclog_or_tty); |
6432c3bb6240
6668743: CMS: Consolidate block statistics reporting code
ysr
parents:
0
diff
changeset
|
2488 total.set_bfrSurp( total.bfrSurp() + fl->bfrSurp() ); |
6432c3bb6240
6668743: CMS: Consolidate block statistics reporting code
ysr
parents:
0
diff
changeset
|
2489 total.set_surplus( total.surplus() + fl->surplus() ); |
6432c3bb6240
6668743: CMS: Consolidate block statistics reporting code
ysr
parents:
0
diff
changeset
|
2490 total.set_desired( total.desired() + fl->desired() ); |
6432c3bb6240
6668743: CMS: Consolidate block statistics reporting code
ysr
parents:
0
diff
changeset
|
2491 total.set_prevSweep( total.prevSweep() + fl->prevSweep() ); |
6432c3bb6240
6668743: CMS: Consolidate block statistics reporting code
ysr
parents:
0
diff
changeset
|
2492 total.set_beforeSweep(total.beforeSweep() + fl->beforeSweep()); |
6432c3bb6240
6668743: CMS: Consolidate block statistics reporting code
ysr
parents:
0
diff
changeset
|
2493 total.set_count( total.count() + fl->count() ); |
6432c3bb6240
6668743: CMS: Consolidate block statistics reporting code
ysr
parents:
0
diff
changeset
|
2494 total.set_coalBirths( total.coalBirths() + fl->coalBirths() ); |
6432c3bb6240
6668743: CMS: Consolidate block statistics reporting code
ysr
parents:
0
diff
changeset
|
2495 total.set_coalDeaths( total.coalDeaths() + fl->coalDeaths() ); |
6432c3bb6240
6668743: CMS: Consolidate block statistics reporting code
ysr
parents:
0
diff
changeset
|
2496 total.set_splitBirths(total.splitBirths() + fl->splitBirths()); |
6432c3bb6240
6668743: CMS: Consolidate block statistics reporting code
ysr
parents:
0
diff
changeset
|
2497 total.set_splitDeaths(total.splitDeaths() + fl->splitDeaths()); |
0 | 2498 } |
12
6432c3bb6240
6668743: CMS: Consolidate block statistics reporting code
ysr
parents:
0
diff
changeset
|
2499 total.print_on(gclog_or_tty, "TOTAL"); |
6432c3bb6240
6668743: CMS: Consolidate block statistics reporting code
ysr
parents:
0
diff
changeset
|
2500 gclog_or_tty->print_cr("Total free in indexed lists " |
6432c3bb6240
6668743: CMS: Consolidate block statistics reporting code
ysr
parents:
0
diff
changeset
|
2501 SIZE_FORMAT " words", totalFree); |
0 | 2502 gclog_or_tty->print("growth: %8.5f deficit: %8.5f\n", |
12
6432c3bb6240
6668743: CMS: Consolidate block statistics reporting code
ysr
parents:
0
diff
changeset
|
2503 (double)(total.splitBirths()+total.coalBirths()-total.splitDeaths()-total.coalDeaths())/ |
6432c3bb6240
6668743: CMS: Consolidate block statistics reporting code
ysr
parents:
0
diff
changeset
|
2504 (total.prevSweep() != 0 ? (double)total.prevSweep() : 1.0), |
6432c3bb6240
6668743: CMS: Consolidate block statistics reporting code
ysr
parents:
0
diff
changeset
|
2505 (double)(total.desired() - total.count())/(total.desired() != 0 ? (double)total.desired() : 1.0)); |
0 | 2506 _dictionary->printDictCensus(); |
2507 } | |
2508 | |
2509 // Return the next displaced header, incrementing the pointer and | |
2510 // recycling spool area as necessary. | |
2511 markOop PromotionInfo::nextDisplacedHeader() { | |
2512 assert(_spoolHead != NULL, "promotionInfo inconsistency"); | |
2513 assert(_spoolHead != _spoolTail || _firstIndex < _nextIndex, | |
2514 "Empty spool space: no displaced header can be fetched"); | |
2515 assert(_spoolHead->bufferSize > _firstIndex, "Off by one error at head?"); | |
2516 markOop hdr = _spoolHead->displacedHdr[_firstIndex]; | |
2517 // Spool forward | |
2518 if (++_firstIndex == _spoolHead->bufferSize) { // last location in this block | |
2519 // forward to next block, recycling this block into spare spool buffer | |
2520 SpoolBlock* tmp = _spoolHead->nextSpoolBlock; | |
2521 assert(_spoolHead != _spoolTail, "Spooling storage mix-up"); | |
2522 _spoolHead->nextSpoolBlock = _spareSpool; | |
2523 _spareSpool = _spoolHead; | |
2524 _spoolHead = tmp; | |
2525 _firstIndex = 1; | |
2526 NOT_PRODUCT( | |
2527 if (_spoolHead == NULL) { // all buffers fully consumed | |
2528 assert(_spoolTail == NULL && _nextIndex == 1, | |
2529 "spool buffers processing inconsistency"); | |
2530 } | |
2531 ) | |
2532 } | |
2533 return hdr; | |
2534 } | |
2535 | |
2536 void PromotionInfo::track(PromotedObject* trackOop) { | |
2537 track(trackOop, oop(trackOop)->klass()); | |
2538 } | |
2539 | |
2540 void PromotionInfo::track(PromotedObject* trackOop, klassOop klassOfOop) { | |
2541 // make a copy of header as it may need to be spooled | |
2542 markOop mark = oop(trackOop)->mark(); | |
2543 trackOop->clearNext(); | |
2544 if (mark->must_be_preserved_for_cms_scavenge(klassOfOop)) { | |
2545 // save non-prototypical header, and mark oop | |
2546 saveDisplacedHeader(mark); | |
2547 trackOop->setDisplacedMark(); | |
2548 } else { | |
2549 // we'd like to assert something like the following: | |
2550 // assert(mark == markOopDesc::prototype(), "consistency check"); | |
2551 // ... but the above won't work because the age bits have not (yet) been | |
2552 // cleared. The remainder of the check would be identical to the | |
2553 // condition checked in must_be_preserved() above, so we don't really | |
2554 // have anything useful to check here! | |
2555 } | |
2556 if (_promoTail != NULL) { | |
2557 assert(_promoHead != NULL, "List consistency"); | |
2558 _promoTail->setNext(trackOop); | |
2559 _promoTail = trackOop; | |
2560 } else { | |
2561 assert(_promoHead == NULL, "List consistency"); | |
2562 _promoHead = _promoTail = trackOop; | |
2563 } | |
2564 // Mask as newly promoted, so we can skip over such objects | |
2565 // when scanning dirty cards | |
2566 assert(!trackOop->hasPromotedMark(), "Should not have been marked"); | |
2567 trackOop->setPromotedMark(); | |
2568 } | |
2569 | |
2570 // Save the given displaced header, incrementing the pointer and | |
2571 // obtaining more spool area as necessary. | |
2572 void PromotionInfo::saveDisplacedHeader(markOop hdr) { | |
2573 assert(_spoolHead != NULL && _spoolTail != NULL, | |
2574 "promotionInfo inconsistency"); | |
2575 assert(_spoolTail->bufferSize > _nextIndex, "Off by one error at tail?"); | |
2576 _spoolTail->displacedHdr[_nextIndex] = hdr; | |
2577 // Spool forward | |
2578 if (++_nextIndex == _spoolTail->bufferSize) { // last location in this block | |
2579 // get a new spooling block | |
2580 assert(_spoolTail->nextSpoolBlock == NULL, "tail should terminate spool list"); | |
2581 _splice_point = _spoolTail; // save for splicing | |
2582 _spoolTail->nextSpoolBlock = getSpoolBlock(); // might fail | |
2583 _spoolTail = _spoolTail->nextSpoolBlock; // might become NULL ... | |
2584 // ... but will attempt filling before next promotion attempt | |
2585 _nextIndex = 1; | |
2586 } | |
2587 } | |
2588 | |
2589 // Ensure that spooling space exists. Return false if spooling space | |
2590 // could not be obtained. | |
2591 bool PromotionInfo::ensure_spooling_space_work() { | |
2592 assert(!has_spooling_space(), "Only call when there is no spooling space"); | |
2593 // Try and obtain more spooling space | |
2594 SpoolBlock* newSpool = getSpoolBlock(); | |
2595 assert(newSpool == NULL || | |
2596 (newSpool->bufferSize != 0 && newSpool->nextSpoolBlock == NULL), | |
2597 "getSpoolBlock() sanity check"); | |
2598 if (newSpool == NULL) { | |
2599 return false; | |
2600 } | |
2601 _nextIndex = 1; | |
2602 if (_spoolTail == NULL) { | |
2603 _spoolTail = newSpool; | |
2604 if (_spoolHead == NULL) { | |
2605 _spoolHead = newSpool; | |
2606 _firstIndex = 1; | |
2607 } else { | |
2608 assert(_splice_point != NULL && _splice_point->nextSpoolBlock == NULL, | |
2609 "Splice point invariant"); | |
2610 // Extra check that _splice_point is connected to list | |
2611 #ifdef ASSERT | |
2612 { | |
2613 SpoolBlock* blk = _spoolHead; | |
2614 for (; blk->nextSpoolBlock != NULL; | |
2615 blk = blk->nextSpoolBlock); | |
2616 assert(blk != NULL && blk == _splice_point, | |
2617 "Splice point incorrect"); | |
2618 } | |
2619 #endif // ASSERT | |
2620 _splice_point->nextSpoolBlock = newSpool; | |
2621 } | |
2622 } else { | |
2623 assert(_spoolHead != NULL, "spool list consistency"); | |
2624 _spoolTail->nextSpoolBlock = newSpool; | |
2625 _spoolTail = newSpool; | |
2626 } | |
2627 return true; | |
2628 } | |
2629 | |
2630 // Get a free spool buffer from the free pool, getting a new block | |
2631 // from the heap if necessary. | |
2632 SpoolBlock* PromotionInfo::getSpoolBlock() { | |
2633 SpoolBlock* res; | |
2634 if ((res = _spareSpool) != NULL) { | |
2635 _spareSpool = _spareSpool->nextSpoolBlock; | |
2636 res->nextSpoolBlock = NULL; | |
2637 } else { // spare spool exhausted, get some from heap | |
2638 res = (SpoolBlock*)(space()->allocateScratch(refillSize())); | |
2639 if (res != NULL) { | |
2640 res->init(); | |
2641 } | |
2642 } | |
2643 assert(res == NULL || res->nextSpoolBlock == NULL, "postcondition"); | |
2644 return res; | |
2645 } | |
2646 | |
2647 void PromotionInfo::startTrackingPromotions() { | |
2648 assert(_spoolHead == _spoolTail && _firstIndex == _nextIndex, | |
2649 "spooling inconsistency?"); | |
2650 _firstIndex = _nextIndex = 1; | |
2651 _tracking = true; | |
2652 } | |
2653 | |
1145
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2654 #define CMSPrintPromoBlockInfo 1 |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2655 |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2656 void PromotionInfo::stopTrackingPromotions(uint worker_id) { |
0 | 2657 assert(_spoolHead == _spoolTail && _firstIndex == _nextIndex, |
2658 "spooling inconsistency?"); | |
2659 _firstIndex = _nextIndex = 1; | |
2660 _tracking = false; | |
1145
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2661 if (CMSPrintPromoBlockInfo > 1) { |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2662 print_statistics(worker_id); |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2663 } |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2664 } |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2665 |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2666 void PromotionInfo::print_statistics(uint worker_id) const { |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2667 assert(_spoolHead == _spoolTail && _firstIndex == _nextIndex, |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2668 "Else will undercount"); |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2669 assert(CMSPrintPromoBlockInfo > 0, "Else unnecessary call"); |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2670 // Count the number of blocks and slots in the free pool |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2671 size_t slots = 0; |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2672 size_t blocks = 0; |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2673 for (SpoolBlock* cur_spool = _spareSpool; |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2674 cur_spool != NULL; |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2675 cur_spool = cur_spool->nextSpoolBlock) { |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2676 // the first entry is just a self-pointer; indices 1 through |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2677 // bufferSize - 1 are occupied (thus, bufferSize - 1 slots). |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2678 guarantee((void*)cur_spool->displacedHdr == (void*)&cur_spool->displacedHdr, |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2679 "first entry of displacedHdr should be self-referential"); |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2680 slots += cur_spool->bufferSize - 1; |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2681 blocks++; |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2682 } |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2683 if (_spoolHead != NULL) { |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2684 slots += _spoolHead->bufferSize - 1; |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2685 blocks++; |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2686 } |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2687 gclog_or_tty->print_cr(" [worker %d] promo_blocks = %d, promo_slots = %d ", |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2688 worker_id, blocks, slots); |
0 | 2689 } |
2690 | |
2691 // When _spoolTail is not NULL, then the slot <_spoolTail, _nextIndex> | |
2692 // points to the next slot available for filling. | |
2693 // The set of slots holding displaced headers are then all those in the | |
2694 // right-open interval denoted by: | |
2695 // | |
2696 // [ <_spoolHead, _firstIndex>, <_spoolTail, _nextIndex> ) | |
2697 // | |
2698 // When _spoolTail is NULL, then the set of slots with displaced headers | |
2699 // is all those starting at the slot <_spoolHead, _firstIndex> and | |
2700 // going up to the last slot of last block in the linked list. | |
2701 // In this lartter case, _splice_point points to the tail block of | |
2702 // this linked list of blocks holding displaced headers. | |
2703 void PromotionInfo::verify() const { | |
2704 // Verify the following: | |
2705 // 1. the number of displaced headers matches the number of promoted | |
2706 // objects that have displaced headers | |
2707 // 2. each promoted object lies in this space | |
2708 debug_only( | |
2709 PromotedObject* junk = NULL; | |
2710 assert(junk->next_addr() == (void*)(oop(junk)->mark_addr()), | |
2711 "Offset of PromotedObject::_next is expected to align with " | |
2712 " the OopDesc::_mark within OopDesc"); | |
2713 ) | |
2714 // FIXME: guarantee???? | |
2715 guarantee(_spoolHead == NULL || _spoolTail != NULL || | |
2716 _splice_point != NULL, "list consistency"); | |
2717 guarantee(_promoHead == NULL || _promoTail != NULL, "list consistency"); | |
2718 // count the number of objects with displaced headers | |
2719 size_t numObjsWithDisplacedHdrs = 0; | |
2720 for (PromotedObject* curObj = _promoHead; curObj != NULL; curObj = curObj->next()) { | |
2721 guarantee(space()->is_in_reserved((HeapWord*)curObj), "Containment"); | |
2722 // the last promoted object may fail the mark() != NULL test of is_oop(). | |
2723 guarantee(curObj->next() == NULL || oop(curObj)->is_oop(), "must be an oop"); | |
2724 if (curObj->hasDisplacedMark()) { | |
2725 numObjsWithDisplacedHdrs++; | |
2726 } | |
2727 } | |
2728 // Count the number of displaced headers | |
2729 size_t numDisplacedHdrs = 0; | |
2730 for (SpoolBlock* curSpool = _spoolHead; | |
2731 curSpool != _spoolTail && curSpool != NULL; | |
2732 curSpool = curSpool->nextSpoolBlock) { | |
2733 // the first entry is just a self-pointer; indices 1 through | |
2734 // bufferSize - 1 are occupied (thus, bufferSize - 1 slots). | |
2735 guarantee((void*)curSpool->displacedHdr == (void*)&curSpool->displacedHdr, | |
2736 "first entry of displacedHdr should be self-referential"); | |
2737 numDisplacedHdrs += curSpool->bufferSize - 1; | |
2738 } | |
2739 guarantee((_spoolHead == _spoolTail) == (numDisplacedHdrs == 0), | |
2740 "internal consistency"); | |
2741 guarantee(_spoolTail != NULL || _nextIndex == 1, | |
2742 "Inconsistency between _spoolTail and _nextIndex"); | |
2743 // We overcounted (_firstIndex-1) worth of slots in block | |
2744 // _spoolHead and we undercounted (_nextIndex-1) worth of | |
2745 // slots in block _spoolTail. We make an appropriate | |
2746 // adjustment by subtracting the first and adding the | |
2747 // second: - (_firstIndex - 1) + (_nextIndex - 1) | |
2748 numDisplacedHdrs += (_nextIndex - _firstIndex); | |
2749 guarantee(numDisplacedHdrs == numObjsWithDisplacedHdrs, "Displaced hdr count"); | |
2750 } | |
2751 | |
1145
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2752 void PromotionInfo::print_on(outputStream* st) const { |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2753 SpoolBlock* curSpool = NULL; |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2754 size_t i = 0; |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2755 st->print_cr("start & end indices: [" SIZE_FORMAT ", " SIZE_FORMAT ")", |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2756 _firstIndex, _nextIndex); |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2757 for (curSpool = _spoolHead; curSpool != _spoolTail && curSpool != NULL; |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2758 curSpool = curSpool->nextSpoolBlock) { |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2759 curSpool->print_on(st); |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2760 st->print_cr(" active "); |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2761 i++; |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2762 } |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2763 for (curSpool = _spoolTail; curSpool != NULL; |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2764 curSpool = curSpool->nextSpoolBlock) { |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2765 curSpool->print_on(st); |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2766 st->print_cr(" inactive "); |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2767 i++; |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2768 } |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2769 for (curSpool = _spareSpool; curSpool != NULL; |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2770 curSpool = curSpool->nextSpoolBlock) { |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2771 curSpool->print_on(st); |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2772 st->print_cr(" free "); |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2773 i++; |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2774 } |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2775 st->print_cr(SIZE_FORMAT " header spooling blocks", i); |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2776 } |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2777 |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2778 void SpoolBlock::print_on(outputStream* st) const { |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2779 st->print("[" PTR_FORMAT "," PTR_FORMAT "), " SIZE_FORMAT " HeapWords -> " PTR_FORMAT, |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2780 this, (HeapWord*)displacedHdr + bufferSize, |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2781 bufferSize, nextSpoolBlock); |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2782 } |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2783 |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2784 /////////////////////////////////////////////////////////////////////////// |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2785 // CFLS_LAB |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2786 /////////////////////////////////////////////////////////////////////////// |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2787 |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2788 #define VECTOR_257(x) \ |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2789 /* 1 2 3 4 5 6 7 8 9 1x 11 12 13 14 15 16 17 18 19 2x 21 22 23 24 25 26 27 28 29 3x 31 32 */ \ |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2790 { x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, \ |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2791 x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, \ |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2792 x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, \ |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2793 x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, \ |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2794 x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, \ |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2795 x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, \ |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2796 x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, \ |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2797 x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, x, \ |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2798 x } |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2799 |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2800 // Initialize with default setting of CMSParPromoteBlocksToClaim, _not_ |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2801 // OldPLABSize, whose static default is different; if overridden at the |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2802 // command-line, this will get reinitialized via a call to |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2803 // modify_initialization() below. |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2804 AdaptiveWeightedAverage CFLS_LAB::_blocks_to_claim[] = |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2805 VECTOR_257(AdaptiveWeightedAverage(OldPLABWeight, (float)CMSParPromoteBlocksToClaim)); |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2806 size_t CFLS_LAB::_global_num_blocks[] = VECTOR_257(0); |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2807 int CFLS_LAB::_global_num_workers[] = VECTOR_257(0); |
0 | 2808 |
2809 CFLS_LAB::CFLS_LAB(CompactibleFreeListSpace* cfls) : | |
2810 _cfls(cfls) | |
2811 { | |
1145
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2812 assert(CompactibleFreeListSpace::IndexSetSize == 257, "Modify VECTOR_257() macro above"); |
0 | 2813 for (size_t i = CompactibleFreeListSpace::IndexSetStart; |
2814 i < CompactibleFreeListSpace::IndexSetSize; | |
2815 i += CompactibleFreeListSpace::IndexSetStride) { | |
2816 _indexedFreeList[i].set_size(i); | |
1145
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2817 _num_blocks[i] = 0; |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2818 } |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2819 } |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2820 |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2821 static bool _CFLS_LAB_modified = false; |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2822 |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2823 void CFLS_LAB::modify_initialization(size_t n, unsigned wt) { |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2824 assert(!_CFLS_LAB_modified, "Call only once"); |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2825 _CFLS_LAB_modified = true; |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2826 for (size_t i = CompactibleFreeListSpace::IndexSetStart; |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2827 i < CompactibleFreeListSpace::IndexSetSize; |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2828 i += CompactibleFreeListSpace::IndexSetStride) { |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2829 _blocks_to_claim[i].modify(n, wt, true /* force */); |
0 | 2830 } |
2831 } | |
2832 | |
2833 HeapWord* CFLS_LAB::alloc(size_t word_sz) { | |
2834 FreeChunk* res; | |
2835 word_sz = _cfls->adjustObjectSize(word_sz); | |
2836 if (word_sz >= CompactibleFreeListSpace::IndexSetSize) { | |
2837 // This locking manages sync with other large object allocations. | |
2838 MutexLockerEx x(_cfls->parDictionaryAllocLock(), | |
2839 Mutex::_no_safepoint_check_flag); | |
2840 res = _cfls->getChunkFromDictionaryExact(word_sz); | |
2841 if (res == NULL) return NULL; | |
2842 } else { | |
2843 FreeList* fl = &_indexedFreeList[word_sz]; | |
2844 if (fl->count() == 0) { | |
2845 // Attempt to refill this local free list. | |
1145
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2846 get_from_global_pool(word_sz, fl); |
0 | 2847 // If it didn't work, give up. |
2848 if (fl->count() == 0) return NULL; | |
2849 } | |
2850 res = fl->getChunkAtHead(); | |
2851 assert(res != NULL, "Why was count non-zero?"); | |
2852 } | |
2853 res->markNotFree(); | |
2854 assert(!res->isFree(), "shouldn't be marked free"); | |
187 | 2855 assert(oop(res)->klass_or_null() == NULL, "should look uninitialized"); |
0 | 2856 // mangle a just allocated object with a distinct pattern. |
2857 debug_only(res->mangleAllocated(word_sz)); | |
2858 return (HeapWord*)res; | |
2859 } | |
2860 | |
1145
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2861 // Get a chunk of blocks of the right size and update related |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2862 // book-keeping stats |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2863 void CFLS_LAB::get_from_global_pool(size_t word_sz, FreeList* fl) { |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2864 // Get the #blocks we want to claim |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2865 size_t n_blks = (size_t)_blocks_to_claim[word_sz].average(); |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2866 assert(n_blks > 0, "Error"); |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2867 assert(ResizePLAB || n_blks == OldPLABSize, "Error"); |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2868 // In some cases, when the application has a phase change, |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2869 // there may be a sudden and sharp shift in the object survival |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2870 // profile, and updating the counts at the end of a scavenge |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2871 // may not be quick enough, giving rise to large scavenge pauses |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2872 // during these phase changes. It is beneficial to detect such |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2873 // changes on-the-fly during a scavenge and avoid such a phase-change |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2874 // pothole. The following code is a heuristic attempt to do that. |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2875 // It is protected by a product flag until we have gained |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2876 // enough experience with this heuristic and fine-tuned its behaviour. |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2877 // WARNING: This might increase fragmentation if we overreact to |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2878 // small spikes, so some kind of historical smoothing based on |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2879 // previous experience with the greater reactivity might be useful. |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2880 // Lacking sufficient experience, CMSOldPLABResizeQuicker is disabled by |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2881 // default. |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2882 if (ResizeOldPLAB && CMSOldPLABResizeQuicker) { |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2883 size_t multiple = _num_blocks[word_sz]/(CMSOldPLABToleranceFactor*CMSOldPLABNumRefills*n_blks); |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2884 n_blks += CMSOldPLABReactivityFactor*multiple*n_blks; |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2885 n_blks = MIN2(n_blks, CMSOldPLABMax); |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2886 } |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2887 assert(n_blks > 0, "Error"); |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2888 _cfls->par_get_chunk_of_blocks(word_sz, n_blks, fl); |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2889 // Update stats table entry for this block size |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2890 _num_blocks[word_sz] += fl->count(); |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2891 } |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2892 |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2893 void CFLS_LAB::compute_desired_plab_size() { |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2894 for (size_t i = CompactibleFreeListSpace::IndexSetStart; |
0 | 2895 i < CompactibleFreeListSpace::IndexSetSize; |
2896 i += CompactibleFreeListSpace::IndexSetStride) { | |
1145
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2897 assert((_global_num_workers[i] == 0) == (_global_num_blocks[i] == 0), |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2898 "Counter inconsistency"); |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2899 if (_global_num_workers[i] > 0) { |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2900 // Need to smooth wrt historical average |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2901 if (ResizeOldPLAB) { |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2902 _blocks_to_claim[i].sample( |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2903 MAX2((size_t)CMSOldPLABMin, |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2904 MIN2((size_t)CMSOldPLABMax, |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2905 _global_num_blocks[i]/(_global_num_workers[i]*CMSOldPLABNumRefills)))); |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2906 } |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2907 // Reset counters for next round |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2908 _global_num_workers[i] = 0; |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2909 _global_num_blocks[i] = 0; |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2910 if (PrintOldPLAB) { |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2911 gclog_or_tty->print_cr("[%d]: %d", i, (size_t)_blocks_to_claim[i].average()); |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2912 } |
0 | 2913 } |
2914 } | |
2915 } | |
2916 | |
1145
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2917 void CFLS_LAB::retire(int tid) { |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2918 // We run this single threaded with the world stopped; |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2919 // so no need for locks and such. |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2920 #define CFLS_LAB_PARALLEL_ACCESS 0 |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2921 NOT_PRODUCT(Thread* t = Thread::current();) |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2922 assert(Thread::current()->is_VM_thread(), "Error"); |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2923 assert(CompactibleFreeListSpace::IndexSetStart == CompactibleFreeListSpace::IndexSetStride, |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2924 "Will access to uninitialized slot below"); |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2925 #if CFLS_LAB_PARALLEL_ACCESS |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2926 for (size_t i = CompactibleFreeListSpace::IndexSetSize - 1; |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2927 i > 0; |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2928 i -= CompactibleFreeListSpace::IndexSetStride) { |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2929 #else // CFLS_LAB_PARALLEL_ACCESS |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2930 for (size_t i = CompactibleFreeListSpace::IndexSetStart; |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2931 i < CompactibleFreeListSpace::IndexSetSize; |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2932 i += CompactibleFreeListSpace::IndexSetStride) { |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2933 #endif // !CFLS_LAB_PARALLEL_ACCESS |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2934 assert(_num_blocks[i] >= (size_t)_indexedFreeList[i].count(), |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2935 "Can't retire more than what we obtained"); |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2936 if (_num_blocks[i] > 0) { |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2937 size_t num_retire = _indexedFreeList[i].count(); |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2938 assert(_num_blocks[i] > num_retire, "Should have used at least one"); |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2939 { |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2940 #if CFLS_LAB_PARALLEL_ACCESS |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2941 MutexLockerEx x(_cfls->_indexedFreeListParLocks[i], |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2942 Mutex::_no_safepoint_check_flag); |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2943 #endif // CFLS_LAB_PARALLEL_ACCESS |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2944 // Update globals stats for num_blocks used |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2945 _global_num_blocks[i] += (_num_blocks[i] - num_retire); |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2946 _global_num_workers[i]++; |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2947 assert(_global_num_workers[i] <= (ssize_t)ParallelGCThreads, "Too big"); |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2948 if (num_retire > 0) { |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2949 _cfls->_indexedFreeList[i].prepend(&_indexedFreeList[i]); |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2950 // Reset this list. |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2951 _indexedFreeList[i] = FreeList(); |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2952 _indexedFreeList[i].set_size(i); |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2953 } |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2954 } |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2955 if (PrintOldPLAB) { |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2956 gclog_or_tty->print_cr("%d[%d]: %d/%d/%d", |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2957 tid, i, num_retire, _num_blocks[i], (size_t)_blocks_to_claim[i].average()); |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2958 } |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2959 // Reset stats for next round |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2960 _num_blocks[i] = 0; |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2961 } |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2962 } |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2963 } |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2964 |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2965 void CompactibleFreeListSpace:: par_get_chunk_of_blocks(size_t word_sz, size_t n, FreeList* fl) { |
0 | 2966 assert(fl->count() == 0, "Precondition."); |
2967 assert(word_sz < CompactibleFreeListSpace::IndexSetSize, | |
2968 "Precondition"); | |
2969 | |
1145
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2970 // We'll try all multiples of word_sz in the indexed set, starting with |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2971 // word_sz itself and, if CMSSplitIndexedFreeListBlocks, try larger multiples, |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2972 // then try getting a big chunk and splitting it. |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2973 { |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2974 bool found; |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2975 int k; |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2976 size_t cur_sz; |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2977 for (k = 1, cur_sz = k * word_sz, found = false; |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2978 (cur_sz < CompactibleFreeListSpace::IndexSetSize) && |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2979 (CMSSplitIndexedFreeListBlocks || k <= 1); |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2980 k++, cur_sz = k * word_sz) { |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2981 FreeList* gfl = &_indexedFreeList[cur_sz]; |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2982 FreeList fl_for_cur_sz; // Empty. |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2983 fl_for_cur_sz.set_size(cur_sz); |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2984 { |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2985 MutexLockerEx x(_indexedFreeListParLocks[cur_sz], |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2986 Mutex::_no_safepoint_check_flag); |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2987 if (gfl->count() != 0) { |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2988 // nn is the number of chunks of size cur_sz that |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2989 // we'd need to split k-ways each, in order to create |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2990 // "n" chunks of size word_sz each. |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2991 const size_t nn = MAX2(n/k, (size_t)1); |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2992 gfl->getFirstNChunksFromList(nn, &fl_for_cur_sz); |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2993 found = true; |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2994 if (k > 1) { |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2995 // Update split death stats for the cur_sz-size blocks list: |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2996 // we increment the split death count by the number of blocks |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2997 // we just took from the cur_sz-size blocks list and which |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2998 // we will be splitting below. |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
2999 ssize_t deaths = _indexedFreeList[cur_sz].splitDeaths() + |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
3000 fl_for_cur_sz.count(); |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
3001 _indexedFreeList[cur_sz].set_splitDeaths(deaths); |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
3002 } |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
3003 } |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
3004 } |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
3005 // Now transfer fl_for_cur_sz to fl. Common case, we hope, is k = 1. |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
3006 if (found) { |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
3007 if (k == 1) { |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
3008 fl->prepend(&fl_for_cur_sz); |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
3009 } else { |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
3010 // Divide each block on fl_for_cur_sz up k ways. |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
3011 FreeChunk* fc; |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
3012 while ((fc = fl_for_cur_sz.getChunkAtHead()) != NULL) { |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
3013 // Must do this in reverse order, so that anybody attempting to |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
3014 // access the main chunk sees it as a single free block until we |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
3015 // change it. |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
3016 size_t fc_size = fc->size(); |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
3017 for (int i = k-1; i >= 0; i--) { |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
3018 FreeChunk* ffc = (FreeChunk*)((HeapWord*)fc + i * word_sz); |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
3019 ffc->setSize(word_sz); |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
3020 ffc->linkNext(NULL); |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
3021 ffc->linkPrev(NULL); // Mark as a free block for other (parallel) GC threads. |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
3022 // Above must occur before BOT is updated below. |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
3023 // splitting from the right, fc_size == (k - i + 1) * wordsize |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
3024 _bt.mark_block((HeapWord*)ffc, word_sz); |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
3025 fc_size -= word_sz; |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
3026 _bt.verify_not_unallocated((HeapWord*)ffc, ffc->size()); |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
3027 _bt.verify_single_block((HeapWord*)fc, fc_size); |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
3028 _bt.verify_single_block((HeapWord*)ffc, ffc->size()); |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
3029 // Push this on "fl". |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
3030 fl->returnChunkAtHead(ffc); |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
3031 } |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
3032 // TRAP |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
3033 assert(fl->tail()->next() == NULL, "List invariant."); |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
3034 } |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
3035 } |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
3036 // Update birth stats for this block size. |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
3037 size_t num = fl->count(); |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
3038 MutexLockerEx x(_indexedFreeListParLocks[word_sz], |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
3039 Mutex::_no_safepoint_check_flag); |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
3040 ssize_t births = _indexedFreeList[word_sz].splitBirths() + num; |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
3041 _indexedFreeList[word_sz].set_splitBirths(births); |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
3042 return; |
0 | 3043 } |
3044 } | |
3045 } | |
3046 // Otherwise, we'll split a block from the dictionary. | |
3047 FreeChunk* fc = NULL; | |
3048 FreeChunk* rem_fc = NULL; | |
3049 size_t rem; | |
3050 { | |
3051 MutexLockerEx x(parDictionaryAllocLock(), | |
3052 Mutex::_no_safepoint_check_flag); | |
3053 while (n > 0) { | |
3054 fc = dictionary()->getChunk(MAX2(n * word_sz, | |
3055 _dictionary->minSize()), | |
3056 FreeBlockDictionary::atLeast); | |
3057 if (fc != NULL) { | |
3058 _bt.allocated((HeapWord*)fc, fc->size()); // update _unallocated_blk | |
3059 dictionary()->dictCensusUpdate(fc->size(), | |
3060 true /*split*/, | |
3061 false /*birth*/); | |
3062 break; | |
3063 } else { | |
3064 n--; | |
3065 } | |
3066 } | |
3067 if (fc == NULL) return; | |
1145
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
3068 assert((ssize_t)n >= 1, "Control point invariant"); |
0 | 3069 // Otherwise, split up that block. |
1145
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
3070 const size_t nn = fc->size() / word_sz; |
0 | 3071 n = MIN2(nn, n); |
1145
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
3072 assert((ssize_t)n >= 1, "Control point invariant"); |
0 | 3073 rem = fc->size() - n * word_sz; |
3074 // If there is a remainder, and it's too small, allocate one fewer. | |
3075 if (rem > 0 && rem < MinChunkSize) { | |
3076 n--; rem += word_sz; | |
3077 } | |
1148
05b775309e59
6912018: CMS: guarantee(head() != 0,"The head of the list cannot be NULL")
jmasa
parents:
1145
diff
changeset
|
3078 // Note that at this point we may have n == 0. |
05b775309e59
6912018: CMS: guarantee(head() != 0,"The head of the list cannot be NULL")
jmasa
parents:
1145
diff
changeset
|
3079 assert((ssize_t)n >= 0, "Control point invariant"); |
05b775309e59
6912018: CMS: guarantee(head() != 0,"The head of the list cannot be NULL")
jmasa
parents:
1145
diff
changeset
|
3080 |
05b775309e59
6912018: CMS: guarantee(head() != 0,"The head of the list cannot be NULL")
jmasa
parents:
1145
diff
changeset
|
3081 // If n is 0, the chunk fc that was found is not large |
05b775309e59
6912018: CMS: guarantee(head() != 0,"The head of the list cannot be NULL")
jmasa
parents:
1145
diff
changeset
|
3082 // enough to leave a viable remainder. We are unable to |
05b775309e59
6912018: CMS: guarantee(head() != 0,"The head of the list cannot be NULL")
jmasa
parents:
1145
diff
changeset
|
3083 // allocate even one block. Return fc to the |
05b775309e59
6912018: CMS: guarantee(head() != 0,"The head of the list cannot be NULL")
jmasa
parents:
1145
diff
changeset
|
3084 // dictionary and return, leaving "fl" empty. |
05b775309e59
6912018: CMS: guarantee(head() != 0,"The head of the list cannot be NULL")
jmasa
parents:
1145
diff
changeset
|
3085 if (n == 0) { |
05b775309e59
6912018: CMS: guarantee(head() != 0,"The head of the list cannot be NULL")
jmasa
parents:
1145
diff
changeset
|
3086 returnChunkToDictionary(fc); |
05b775309e59
6912018: CMS: guarantee(head() != 0,"The head of the list cannot be NULL")
jmasa
parents:
1145
diff
changeset
|
3087 return; |
05b775309e59
6912018: CMS: guarantee(head() != 0,"The head of the list cannot be NULL")
jmasa
parents:
1145
diff
changeset
|
3088 } |
05b775309e59
6912018: CMS: guarantee(head() != 0,"The head of the list cannot be NULL")
jmasa
parents:
1145
diff
changeset
|
3089 |
0 | 3090 // First return the remainder, if any. |
3091 // Note that we hold the lock until we decide if we're going to give | |
1145
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
3092 // back the remainder to the dictionary, since a concurrent allocation |
0 | 3093 // may otherwise see the heap as empty. (We're willing to take that |
3094 // hit if the block is a small block.) | |
3095 if (rem > 0) { | |
3096 size_t prefix_size = n * word_sz; | |
3097 rem_fc = (FreeChunk*)((HeapWord*)fc + prefix_size); | |
3098 rem_fc->setSize(rem); | |
3099 rem_fc->linkNext(NULL); | |
3100 rem_fc->linkPrev(NULL); // Mark as a free block for other (parallel) GC threads. | |
3101 // Above must occur before BOT is updated below. | |
1145
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
3102 assert((ssize_t)n > 0 && prefix_size > 0 && rem_fc > fc, "Error"); |
0 | 3103 _bt.split_block((HeapWord*)fc, fc->size(), prefix_size); |
3104 if (rem >= IndexSetSize) { | |
3105 returnChunkToDictionary(rem_fc); | |
1145
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
3106 dictionary()->dictCensusUpdate(rem, true /*split*/, true /*birth*/); |
0 | 3107 rem_fc = NULL; |
3108 } | |
3109 // Otherwise, return it to the small list below. | |
3110 } | |
3111 } | |
3112 if (rem_fc != NULL) { | |
3113 MutexLockerEx x(_indexedFreeListParLocks[rem], | |
3114 Mutex::_no_safepoint_check_flag); | |
3115 _bt.verify_not_unallocated((HeapWord*)rem_fc, rem_fc->size()); | |
3116 _indexedFreeList[rem].returnChunkAtHead(rem_fc); | |
3117 smallSplitBirth(rem); | |
3118 } | |
1145
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
3119 assert((ssize_t)n > 0 && fc != NULL, "Consistency"); |
0 | 3120 // Now do the splitting up. |
3121 // Must do this in reverse order, so that anybody attempting to | |
3122 // access the main chunk sees it as a single free block until we | |
3123 // change it. | |
3124 size_t fc_size = n * word_sz; | |
3125 // All but first chunk in this loop | |
3126 for (ssize_t i = n-1; i > 0; i--) { | |
3127 FreeChunk* ffc = (FreeChunk*)((HeapWord*)fc + i * word_sz); | |
3128 ffc->setSize(word_sz); | |
3129 ffc->linkNext(NULL); | |
3130 ffc->linkPrev(NULL); // Mark as a free block for other (parallel) GC threads. | |
3131 // Above must occur before BOT is updated below. | |
3132 // splitting from the right, fc_size == (n - i + 1) * wordsize | |
3133 _bt.mark_block((HeapWord*)ffc, word_sz); | |
3134 fc_size -= word_sz; | |
3135 _bt.verify_not_unallocated((HeapWord*)ffc, ffc->size()); | |
3136 _bt.verify_single_block((HeapWord*)ffc, ffc->size()); | |
3137 _bt.verify_single_block((HeapWord*)fc, fc_size); | |
3138 // Push this on "fl". | |
3139 fl->returnChunkAtHead(ffc); | |
3140 } | |
3141 // First chunk | |
3142 fc->setSize(word_sz); | |
3143 fc->linkNext(NULL); | |
3144 fc->linkPrev(NULL); | |
3145 _bt.verify_not_unallocated((HeapWord*)fc, fc->size()); | |
3146 _bt.verify_single_block((HeapWord*)fc, fc->size()); | |
3147 fl->returnChunkAtHead(fc); | |
3148 | |
1145
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
3149 assert((ssize_t)n > 0 && (ssize_t)n == fl->count(), "Incorrect number of blocks"); |
0 | 3150 { |
1145
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
3151 // Update the stats for this block size. |
0 | 3152 MutexLockerEx x(_indexedFreeListParLocks[word_sz], |
3153 Mutex::_no_safepoint_check_flag); | |
1145
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
3154 const ssize_t births = _indexedFreeList[word_sz].splitBirths() + n; |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
3155 _indexedFreeList[word_sz].set_splitBirths(births); |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
3156 // ssize_t new_surplus = _indexedFreeList[word_sz].surplus() + n; |
e018e6884bd8
6631166: CMS: better heuristics when combatting fragmentation
ysr
parents:
579
diff
changeset
|
3157 // _indexedFreeList[word_sz].set_surplus(new_surplus); |
0 | 3158 } |
3159 | |
3160 // TRAP | |
3161 assert(fl->tail()->next() == NULL, "List invariant."); | |
3162 } | |
3163 | |
3164 // Set up the space's par_seq_tasks structure for work claiming | |
3165 // for parallel rescan. See CMSParRemarkTask where this is currently used. | |
3166 // XXX Need to suitably abstract and generalize this and the next | |
3167 // method into one. | |
3168 void | |
3169 CompactibleFreeListSpace:: | |
3170 initialize_sequential_subtasks_for_rescan(int n_threads) { | |
3171 // The "size" of each task is fixed according to rescan_task_size. | |
3172 assert(n_threads > 0, "Unexpected n_threads argument"); | |
3173 const size_t task_size = rescan_task_size(); | |
3174 size_t n_tasks = (used_region().word_size() + task_size - 1)/task_size; | |
340
ebeb6490b814
6722116: CMS: Incorrect overflow handling when using parallel concurrent marking
ysr
parents:
269
diff
changeset
|
3175 assert((n_tasks == 0) == used_region().is_empty(), "n_tasks incorrect"); |
ebeb6490b814
6722116: CMS: Incorrect overflow handling when using parallel concurrent marking
ysr
parents:
269
diff
changeset
|
3176 assert(n_tasks == 0 || |
ebeb6490b814
6722116: CMS: Incorrect overflow handling when using parallel concurrent marking
ysr
parents:
269
diff
changeset
|
3177 ((used_region().start() + (n_tasks - 1)*task_size < used_region().end()) && |
ebeb6490b814
6722116: CMS: Incorrect overflow handling when using parallel concurrent marking
ysr
parents:
269
diff
changeset
|
3178 (used_region().start() + n_tasks*task_size >= used_region().end())), |
ebeb6490b814
6722116: CMS: Incorrect overflow handling when using parallel concurrent marking
ysr
parents:
269
diff
changeset
|
3179 "n_tasks calculation incorrect"); |
0 | 3180 SequentialSubTasksDone* pst = conc_par_seq_tasks(); |
3181 assert(!pst->valid(), "Clobbering existing data?"); | |
3182 pst->set_par_threads(n_threads); | |
3183 pst->set_n_tasks((int)n_tasks); | |
3184 } | |
3185 | |
3186 // Set up the space's par_seq_tasks structure for work claiming | |
3187 // for parallel concurrent marking. See CMSConcMarkTask where this is currently used. | |
3188 void | |
3189 CompactibleFreeListSpace:: | |
3190 initialize_sequential_subtasks_for_marking(int n_threads, | |
3191 HeapWord* low) { | |
3192 // The "size" of each task is fixed according to rescan_task_size. | |
3193 assert(n_threads > 0, "Unexpected n_threads argument"); | |
3194 const size_t task_size = marking_task_size(); | |
3195 assert(task_size > CardTableModRefBS::card_size_in_words && | |
3196 (task_size % CardTableModRefBS::card_size_in_words == 0), | |
3197 "Otherwise arithmetic below would be incorrect"); | |
3198 MemRegion span = _gen->reserved(); | |
3199 if (low != NULL) { | |
3200 if (span.contains(low)) { | |
3201 // Align low down to a card boundary so that | |
3202 // we can use block_offset_careful() on span boundaries. | |
3203 HeapWord* aligned_low = (HeapWord*)align_size_down((uintptr_t)low, | |
3204 CardTableModRefBS::card_size); | |
3205 // Clip span prefix at aligned_low | |
3206 span = span.intersection(MemRegion(aligned_low, span.end())); | |
3207 } else if (low > span.end()) { | |
3208 span = MemRegion(low, low); // Null region | |
3209 } // else use entire span | |
3210 } | |
3211 assert(span.is_empty() || | |
3212 ((uintptr_t)span.start() % CardTableModRefBS::card_size == 0), | |
3213 "span should start at a card boundary"); | |
3214 size_t n_tasks = (span.word_size() + task_size - 1)/task_size; | |
3215 assert((n_tasks == 0) == span.is_empty(), "Inconsistency"); | |
3216 assert(n_tasks == 0 || | |
3217 ((span.start() + (n_tasks - 1)*task_size < span.end()) && | |
3218 (span.start() + n_tasks*task_size >= span.end())), | |
340
ebeb6490b814
6722116: CMS: Incorrect overflow handling when using parallel concurrent marking
ysr
parents:
269
diff
changeset
|
3219 "n_tasks calculation incorrect"); |
0 | 3220 SequentialSubTasksDone* pst = conc_par_seq_tasks(); |
3221 assert(!pst->valid(), "Clobbering existing data?"); | |
3222 pst->set_par_threads(n_threads); | |
3223 pst->set_n_tasks((int)n_tasks); | |
3224 } |