1 /*
2 * Copyright (c) 2001, 2014, Oracle and/or its affiliates. All rights reserved.
3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
4 *
5 * This code is free software; you can redistribute it and/or modify it
6 * under the terms of the GNU General Public License version 2 only, as
7 * published by the Free Software Foundation.
8 *
9 * This code is distributed in the hope that it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
12 * version 2 for more details (a copy is included in the LICENSE file that
13 * accompanied this code).
14 *
15 * You should have received a copy of the GNU General Public License version
16 * 2 along with this work; if not, write to the Free Software Foundation,
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
18 *
19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
20 * or visit www.oracle.com if you need additional information or have any
21 * questions.
22 *
23 */
24
25 #include "precompiled.hpp"
26 #include "gc_implementation/concurrentMarkSweep/cmsLockVerifier.hpp"
27 #include "gc_implementation/concurrentMarkSweep/compactibleFreeListSpace.hpp"
28 #include "gc_implementation/concurrentMarkSweep/concurrentMarkSweepGeneration.inline.hpp"
29 #include "gc_implementation/concurrentMarkSweep/concurrentMarkSweepThread.hpp"
30 #include "gc_implementation/shared/liveRange.hpp"
31 #include "gc_implementation/shared/spaceDecorator.hpp"
32 #include "gc_interface/collectedHeap.inline.hpp"
33 #include "memory/allocation.inline.hpp"
34 #include "memory/blockOffsetTable.inline.hpp"
35 #include "memory/resourceArea.hpp"
36 #include "memory/space.inline.hpp"
37 #include "memory/universe.inline.hpp"
38 #include "oops/oop.inline.hpp"
39 #include "runtime/globals.hpp"
40 #include "runtime/handles.inline.hpp"
41 #include "runtime/init.hpp"
42 #include "runtime/java.hpp"
43 #include "runtime/orderAccess.inline.hpp"
44 #include "runtime/vmThread.hpp"
45 #include "utilities/copy.hpp"
46
47 /////////////////////////////////////////////////////////////////////////
48 //// CompactibleFreeListSpace
49 /////////////////////////////////////////////////////////////////////////
50
51 // highest ranked free list lock rank
52 int CompactibleFreeListSpace::_lockRank = Mutex::leaf + 3;
53
54 // Defaults are 0 so things will break badly if incorrectly initialized.
55 size_t CompactibleFreeListSpace::IndexSetStart = 0;
56 size_t CompactibleFreeListSpace::IndexSetStride = 0;
57
58 size_t MinChunkSize = 0;
59
60 void CompactibleFreeListSpace::set_cms_values() {
61 // Set CMS global values
62 assert(MinChunkSize == 0, "already set");
63
64 // MinChunkSize should be a multiple of MinObjAlignment and be large enough
65 // for chunks to contain a FreeChunk.
66 size_t min_chunk_size_in_bytes = align_size_up(sizeof(FreeChunk), MinObjAlignmentInBytes);
67 MinChunkSize = min_chunk_size_in_bytes / BytesPerWord;
68
69 assert(IndexSetStart == 0 && IndexSetStride == 0, "already set");
70 IndexSetStart = MinChunkSize;
71 IndexSetStride = MinObjAlignment;
72 }
73
74 // Constructor
75 CompactibleFreeListSpace::CompactibleFreeListSpace(BlockOffsetSharedArray* bs,
76 MemRegion mr, bool use_adaptive_freelists,
77 FreeBlockDictionary<FreeChunk>::DictionaryChoice dictionaryChoice) :
78 _dictionaryChoice(dictionaryChoice),
79 _adaptive_freelists(use_adaptive_freelists),
80 _bt(bs, mr),
81 // free list locks are in the range of values taken by _lockRank
82 // This range currently is [_leaf+2, _leaf+3]
83 // Note: this requires that CFLspace c'tors
84 // are called serially in the order in which the locks are
85 // are acquired in the program text. This is true today.
86 _freelistLock(_lockRank--, "CompactibleFreeListSpace._lock", true),
87 _parDictionaryAllocLock(Mutex::leaf - 1, // == rank(ExpandHeap_lock) - 1
88 "CompactibleFreeListSpace._dict_par_lock", true),
89 _rescan_task_size(CardTableModRefBS::card_size_in_words * BitsPerWord *
90 CMSRescanMultiple),
91 _marking_task_size(CardTableModRefBS::card_size_in_words * BitsPerWord *
92 CMSConcMarkMultiple),
93 _collector(NULL)
94 {
95 assert(sizeof(FreeChunk) / BytesPerWord <= MinChunkSize,
96 "FreeChunk is larger than expected");
97 _bt.set_space(this);
98 initialize(mr, SpaceDecorator::Clear, SpaceDecorator::Mangle);
99 // We have all of "mr", all of which we place in the dictionary
100 // as one big chunk. We'll need to decide here which of several
101 // possible alternative dictionary implementations to use. For
102 // now the choice is easy, since we have only one working
103 // implementation, namely, the simple binary tree (splaying
104 // temporarily disabled).
105 switch (dictionaryChoice) {
106 case FreeBlockDictionary<FreeChunk>::dictionaryBinaryTree:
107 _dictionary = new AFLBinaryTreeDictionary(mr);
108 break;
109 case FreeBlockDictionary<FreeChunk>::dictionarySplayTree:
110 case FreeBlockDictionary<FreeChunk>::dictionarySkipList:
111 default:
112 warning("dictionaryChoice: selected option not understood; using"
113 " default BinaryTreeDictionary implementation instead.");
114 }
115 assert(_dictionary != NULL, "CMS dictionary initialization");
116 // The indexed free lists are initially all empty and are lazily
117 // filled in on demand. Initialize the array elements to NULL.
118 initializeIndexedFreeListArray();
119
120 // Not using adaptive free lists assumes that allocation is first
121 // from the linAB's. Also a cms perm gen which can be compacted
122 // has to have the klass's klassKlass allocated at a lower
123 // address in the heap than the klass so that the klassKlass is
124 // moved to its new location before the klass is moved.
125 // Set the _refillSize for the linear allocation blocks
126 if (!use_adaptive_freelists) {
127 FreeChunk* fc = _dictionary->get_chunk(mr.word_size(),
128 FreeBlockDictionary<FreeChunk>::atLeast);
129 // The small linAB initially has all the space and will allocate
130 // a chunk of any size.
131 HeapWord* addr = (HeapWord*) fc;
132 _smallLinearAllocBlock.set(addr, fc->size() ,
133 1024*SmallForLinearAlloc, fc->size());
134 // Note that _unallocated_block is not updated here.
135 // Allocations from the linear allocation block should
136 // update it.
137 } else {
138 _smallLinearAllocBlock.set(0, 0, 1024*SmallForLinearAlloc,
139 SmallForLinearAlloc);
140 }
141 // CMSIndexedFreeListReplenish should be at least 1
142 CMSIndexedFreeListReplenish = MAX2((uintx)1, CMSIndexedFreeListReplenish);
143 _promoInfo.setSpace(this);
144 if (UseCMSBestFit) {
145 _fitStrategy = FreeBlockBestFitFirst;
146 } else {
147 _fitStrategy = FreeBlockStrategyNone;
148 }
149 check_free_list_consistency();
150
151 // Initialize locks for parallel case.
152
153 if (CollectedHeap::use_parallel_gc_threads()) {
154 for (size_t i = IndexSetStart; i < IndexSetSize; i += IndexSetStride) {
155 _indexedFreeListParLocks[i] = new Mutex(Mutex::leaf - 1, // == ExpandHeap_lock - 1
156 "a freelist par lock",
157 true);
158 DEBUG_ONLY(
159 _indexedFreeList[i].set_protecting_lock(_indexedFreeListParLocks[i]);
160 )
161 }
162 _dictionary->set_par_lock(&_parDictionaryAllocLock);
163 }
164 }
165
166 // Like CompactibleSpace forward() but always calls cross_threshold() to
167 // update the block offset table. Removed initialize_threshold call because
168 // CFLS does not use a block offset array for contiguous spaces.
169 HeapWord* CompactibleFreeListSpace::forward(oop q, size_t size,
170 CompactPoint* cp, HeapWord* compact_top) {
171 // q is alive
172 // First check if we should switch compaction space
173 assert(this == cp->space, "'this' should be current compaction space.");
174 size_t compaction_max_size = pointer_delta(end(), compact_top);
175 assert(adjustObjectSize(size) == cp->space->adjust_object_size_v(size),
176 "virtual adjustObjectSize_v() method is not correct");
177 size_t adjusted_size = adjustObjectSize(size);
178 assert(compaction_max_size >= MinChunkSize || compaction_max_size == 0,
179 "no small fragments allowed");
180 assert(minimum_free_block_size() == MinChunkSize,
181 "for de-virtualized reference below");
182 // Can't leave a nonzero size, residual fragment smaller than MinChunkSize
183 if (adjusted_size + MinChunkSize > compaction_max_size &&
184 adjusted_size != compaction_max_size) {
185 do {
186 // switch to next compaction space
187 cp->space->set_compaction_top(compact_top);
188 cp->space = cp->space->next_compaction_space();
189 if (cp->space == NULL) {
190 cp->gen = GenCollectedHeap::heap()->young_gen();
191 assert(cp->gen != NULL, "compaction must succeed");
192 cp->space = cp->gen->first_compaction_space();
193 assert(cp->space != NULL, "generation must have a first compaction space");
194 }
195 compact_top = cp->space->bottom();
196 cp->space->set_compaction_top(compact_top);
197 // The correct adjusted_size may not be the same as that for this method
198 // (i.e., cp->space may no longer be "this" so adjust the size again.
199 // Use the virtual method which is not used above to save the virtual
200 // dispatch.
201 adjusted_size = cp->space->adjust_object_size_v(size);
202 compaction_max_size = pointer_delta(cp->space->end(), compact_top);
203 assert(cp->space->minimum_free_block_size() == 0, "just checking");
204 } while (adjusted_size > compaction_max_size);
205 }
206
207 // store the forwarding pointer into the mark word
208 if ((HeapWord*)q != compact_top) {
209 q->forward_to(oop(compact_top));
210 assert(q->is_gc_marked(), "encoding the pointer should preserve the mark");
211 } else {
212 // if the object isn't moving we can just set the mark to the default
213 // mark and handle it specially later on.
214 q->init_mark();
215 assert(q->forwardee() == NULL, "should be forwarded to NULL");
216 }
217
218 compact_top += adjusted_size;
219
220 // we need to update the offset table so that the beginnings of objects can be
221 // found during scavenge. Note that we are updating the offset table based on
222 // where the object will be once the compaction phase finishes.
223
224 // Always call cross_threshold(). A contiguous space can only call it when
225 // the compaction_top exceeds the current threshold but not for an
226 // non-contiguous space.
227 cp->threshold =
228 cp->space->cross_threshold(compact_top - adjusted_size, compact_top);
229 return compact_top;
230 }
231
232 // A modified copy of OffsetTableContigSpace::cross_threshold() with _offsets -> _bt
233 // and use of single_block instead of alloc_block. The name here is not really
234 // appropriate - maybe a more general name could be invented for both the
235 // contiguous and noncontiguous spaces.
236
237 HeapWord* CompactibleFreeListSpace::cross_threshold(HeapWord* start, HeapWord* the_end) {
238 _bt.single_block(start, the_end);
239 return end();
240 }
241
242 // Initialize them to NULL.
243 void CompactibleFreeListSpace::initializeIndexedFreeListArray() {
244 for (size_t i = 0; i < IndexSetSize; i++) {
245 // Note that on platforms where objects are double word aligned,
246 // the odd array elements are not used. It is convenient, however,
247 // to map directly from the object size to the array element.
248 _indexedFreeList[i].reset(IndexSetSize);
249 _indexedFreeList[i].set_size(i);
250 assert(_indexedFreeList[i].count() == 0, "reset check failed");
251 assert(_indexedFreeList[i].head() == NULL, "reset check failed");
252 assert(_indexedFreeList[i].tail() == NULL, "reset check failed");
253 assert(_indexedFreeList[i].hint() == IndexSetSize, "reset check failed");
254 }
255 }
256
257 void CompactibleFreeListSpace::resetIndexedFreeListArray() {
258 for (size_t i = 1; i < IndexSetSize; i++) {
259 assert(_indexedFreeList[i].size() == (size_t) i,
260 "Indexed free list sizes are incorrect");
261 _indexedFreeList[i].reset(IndexSetSize);
262 assert(_indexedFreeList[i].count() == 0, "reset check failed");
263 assert(_indexedFreeList[i].head() == NULL, "reset check failed");
264 assert(_indexedFreeList[i].tail() == NULL, "reset check failed");
265 assert(_indexedFreeList[i].hint() == IndexSetSize, "reset check failed");
266 }
267 }
268
269 void CompactibleFreeListSpace::reset(MemRegion mr) {
270 resetIndexedFreeListArray();
271 dictionary()->reset();
272 if (BlockOffsetArrayUseUnallocatedBlock) {
273 assert(end() == mr.end(), "We are compacting to the bottom of CMS gen");
274 // Everything's allocated until proven otherwise.
275 _bt.set_unallocated_block(end());
276 }
277 if (!mr.is_empty()) {
278 assert(mr.word_size() >= MinChunkSize, "Chunk size is too small");
279 _bt.single_block(mr.start(), mr.word_size());
280 FreeChunk* fc = (FreeChunk*) mr.start();
281 fc->set_size(mr.word_size());
282 if (mr.word_size() >= IndexSetSize ) {
283 returnChunkToDictionary(fc);
284 } else {
285 _bt.verify_not_unallocated((HeapWord*)fc, fc->size());
286 _indexedFreeList[mr.word_size()].return_chunk_at_head(fc);
287 }
288 coalBirth(mr.word_size());
289 }
290 _promoInfo.reset();
291 _smallLinearAllocBlock._ptr = NULL;
292 _smallLinearAllocBlock._word_size = 0;
293 }
294
295 void CompactibleFreeListSpace::reset_after_compaction() {
296 // Reset the space to the new reality - one free chunk.
297 MemRegion mr(compaction_top(), end());
298 reset(mr);
299 // Now refill the linear allocation block(s) if possible.
300 if (_adaptive_freelists) {
301 refillLinearAllocBlocksIfNeeded();
302 } else {
303 // Place as much of mr in the linAB as we can get,
304 // provided it was big enough to go into the dictionary.
305 FreeChunk* fc = dictionary()->find_largest_dict();
306 if (fc != NULL) {
307 assert(fc->size() == mr.word_size(),
308 "Why was the chunk broken up?");
309 removeChunkFromDictionary(fc);
310 HeapWord* addr = (HeapWord*) fc;
311 _smallLinearAllocBlock.set(addr, fc->size() ,
312 1024*SmallForLinearAlloc, fc->size());
313 // Note that _unallocated_block is not updated here.
314 }
315 }
316 }
317
318 // Walks the entire dictionary, returning a coterminal
319 // chunk, if it exists. Use with caution since it involves
320 // a potentially complete walk of a potentially large tree.
321 FreeChunk* CompactibleFreeListSpace::find_chunk_at_end() {
322
323 assert_lock_strong(&_freelistLock);
324
325 return dictionary()->find_chunk_ends_at(end());
326 }
327
328
329 #ifndef PRODUCT
330 void CompactibleFreeListSpace::initializeIndexedFreeListArrayReturnedBytes() {
331 for (size_t i = IndexSetStart; i < IndexSetSize; i += IndexSetStride) {
332 _indexedFreeList[i].allocation_stats()->set_returned_bytes(0);
333 }
334 }
335
336 size_t CompactibleFreeListSpace::sumIndexedFreeListArrayReturnedBytes() {
337 size_t sum = 0;
338 for (size_t i = IndexSetStart; i < IndexSetSize; i += IndexSetStride) {
339 sum += _indexedFreeList[i].allocation_stats()->returned_bytes();
340 }
341 return sum;
342 }
343
344 size_t CompactibleFreeListSpace::totalCountInIndexedFreeLists() const {
345 size_t count = 0;
346 for (size_t i = IndexSetStart; i < IndexSetSize; i++) {
347 debug_only(
348 ssize_t total_list_count = 0;
349 for (FreeChunk* fc = _indexedFreeList[i].head(); fc != NULL;
350 fc = fc->next()) {
351 total_list_count++;
352 }
353 assert(total_list_count == _indexedFreeList[i].count(),
354 "Count in list is incorrect");
355 )
356 count += _indexedFreeList[i].count();
357 }
358 return count;
359 }
360
361 size_t CompactibleFreeListSpace::totalCount() {
362 size_t num = totalCountInIndexedFreeLists();
363 num += dictionary()->total_count();
364 if (_smallLinearAllocBlock._word_size != 0) {
365 num++;
366 }
367 return num;
368 }
369 #endif
370
371 bool CompactibleFreeListSpace::is_free_block(const HeapWord* p) const {
372 FreeChunk* fc = (FreeChunk*) p;
373 return fc->is_free();
374 }
375
376 size_t CompactibleFreeListSpace::used() const {
377 return capacity() - free();
378 }
379
380 size_t CompactibleFreeListSpace::free() const {
381 // "MT-safe, but not MT-precise"(TM), if you will: i.e.
382 // if you do this while the structures are in flux you
383 // may get an approximate answer only; for instance
384 // because there is concurrent allocation either
385 // directly by mutators or for promotion during a GC.
386 // It's "MT-safe", however, in the sense that you are guaranteed
387 // not to crash and burn, for instance, because of walking
388 // pointers that could disappear as you were walking them.
389 // The approximation is because the various components
390 // that are read below are not read atomically (and
391 // further the computation of totalSizeInIndexedFreeLists()
392 // is itself a non-atomic computation. The normal use of
393 // this is during a resize operation at the end of GC
394 // and at that time you are guaranteed to get the
395 // correct actual value. However, for instance, this is
396 // also read completely asynchronously by the "perf-sampler"
397 // that supports jvmstat, and you are apt to see the values
398 // flicker in such cases.
399 assert(_dictionary != NULL, "No _dictionary?");
400 return (_dictionary->total_chunk_size(DEBUG_ONLY(freelistLock())) +
401 totalSizeInIndexedFreeLists() +
402 _smallLinearAllocBlock._word_size) * HeapWordSize;
403 }
404
405 size_t CompactibleFreeListSpace::max_alloc_in_words() const {
406 assert(_dictionary != NULL, "No _dictionary?");
407 assert_locked();
408 size_t res = _dictionary->max_chunk_size();
409 res = MAX2(res, MIN2(_smallLinearAllocBlock._word_size,
410 (size_t) SmallForLinearAlloc - 1));
411 // XXX the following could potentially be pretty slow;
412 // should one, pessimistically for the rare cases when res
413 // calculated above is less than IndexSetSize,
414 // just return res calculated above? My reasoning was that
415 // those cases will be so rare that the extra time spent doesn't
416 // really matter....
417 // Note: do not change the loop test i >= res + IndexSetStride
418 // to i > res below, because i is unsigned and res may be zero.
419 for (size_t i = IndexSetSize - 1; i >= res + IndexSetStride;
420 i -= IndexSetStride) {
421 if (_indexedFreeList[i].head() != NULL) {
422 assert(_indexedFreeList[i].count() != 0, "Inconsistent FreeList");
423 return i;
424 }
425 }
426 return res;
427 }
428
429 void LinearAllocBlock::print_on(outputStream* st) const {
430 st->print_cr(" LinearAllocBlock: ptr = " PTR_FORMAT ", word_size = " SIZE_FORMAT
431 ", refillsize = " SIZE_FORMAT ", allocation_size_limit = " SIZE_FORMAT,
432 p2i(_ptr), _word_size, _refillSize, _allocation_size_limit);
433 }
434
435 void CompactibleFreeListSpace::print_on(outputStream* st) const {
436 st->print_cr("COMPACTIBLE FREELIST SPACE");
437 st->print_cr(" Space:");
438 Space::print_on(st);
439
440 st->print_cr("promoInfo:");
441 _promoInfo.print_on(st);
442
443 st->print_cr("_smallLinearAllocBlock");
444 _smallLinearAllocBlock.print_on(st);
445
446 // dump_memory_block(_smallLinearAllocBlock->_ptr, 128);
447
448 st->print_cr(" _fitStrategy = %s, _adaptive_freelists = %s",
449 _fitStrategy?"true":"false", _adaptive_freelists?"true":"false");
450 }
451
452 void CompactibleFreeListSpace::print_indexed_free_lists(outputStream* st)
453 const {
454 reportIndexedFreeListStatistics();
455 gclog_or_tty->print_cr("Layout of Indexed Freelists");
456 gclog_or_tty->print_cr("---------------------------");
457 AdaptiveFreeList<FreeChunk>::print_labels_on(st, "size");
458 for (size_t i = IndexSetStart; i < IndexSetSize; i += IndexSetStride) {
459 _indexedFreeList[i].print_on(gclog_or_tty);
460 for (FreeChunk* fc = _indexedFreeList[i].head(); fc != NULL;
461 fc = fc->next()) {
462 gclog_or_tty->print_cr("\t[" PTR_FORMAT "," PTR_FORMAT ") %s",
463 p2i(fc), p2i((HeapWord*)fc + i),
464 fc->cantCoalesce() ? "\t CC" : "");
465 }
466 }
467 }
468
469 void CompactibleFreeListSpace::print_promo_info_blocks(outputStream* st)
470 const {
471 _promoInfo.print_on(st);
472 }
473
474 void CompactibleFreeListSpace::print_dictionary_free_lists(outputStream* st)
475 const {
476 _dictionary->report_statistics();
477 st->print_cr("Layout of Freelists in Tree");
478 st->print_cr("---------------------------");
479 _dictionary->print_free_lists(st);
480 }
481
482 class BlkPrintingClosure: public BlkClosure {
483 const CMSCollector* _collector;
484 const CompactibleFreeListSpace* _sp;
485 const CMSBitMap* _live_bit_map;
486 const bool _post_remark;
487 outputStream* _st;
488 public:
489 BlkPrintingClosure(const CMSCollector* collector,
490 const CompactibleFreeListSpace* sp,
491 const CMSBitMap* live_bit_map,
492 outputStream* st):
493 _collector(collector),
494 _sp(sp),
495 _live_bit_map(live_bit_map),
496 _post_remark(collector->abstract_state() > CMSCollector::FinalMarking),
497 _st(st) { }
498 size_t do_blk(HeapWord* addr);
499 };
500
501 size_t BlkPrintingClosure::do_blk(HeapWord* addr) {
502 size_t sz = _sp->block_size_no_stall(addr, _collector);
503 assert(sz != 0, "Should always be able to compute a size");
504 if (_sp->block_is_obj(addr)) {
505 const bool dead = _post_remark && !_live_bit_map->isMarked(addr);
506 _st->print_cr(PTR_FORMAT ": %s object of size " SIZE_FORMAT "%s",
507 p2i(addr),
508 dead ? "dead" : "live",
509 sz,
510 (!dead && CMSPrintObjectsInDump) ? ":" : ".");
511 if (CMSPrintObjectsInDump && !dead) {
512 oop(addr)->print_on(_st);
513 _st->print_cr("--------------------------------------");
514 }
515 } else { // free block
516 _st->print_cr(PTR_FORMAT ": free block of size " SIZE_FORMAT "%s",
517 p2i(addr), sz, CMSPrintChunksInDump ? ":" : ".");
518 if (CMSPrintChunksInDump) {
519 ((FreeChunk*)addr)->print_on(_st);
520 _st->print_cr("--------------------------------------");
521 }
522 }
523 return sz;
524 }
525
526 void CompactibleFreeListSpace::dump_at_safepoint_with_locks(CMSCollector* c,
527 outputStream* st) {
528 st->print_cr("\n=========================");
529 st->print_cr("Block layout in CMS Heap:");
530 st->print_cr("=========================");
531 BlkPrintingClosure bpcl(c, this, c->markBitMap(), st);
532 blk_iterate(&bpcl);
533
534 st->print_cr("\n=======================================");
535 st->print_cr("Order & Layout of Promotion Info Blocks");
536 st->print_cr("=======================================");
537 print_promo_info_blocks(st);
538
539 st->print_cr("\n===========================");
540 st->print_cr("Order of Indexed Free Lists");
541 st->print_cr("=========================");
542 print_indexed_free_lists(st);
543
544 st->print_cr("\n=================================");
545 st->print_cr("Order of Free Lists in Dictionary");
546 st->print_cr("=================================");
547 print_dictionary_free_lists(st);
548 }
549
550
551 void CompactibleFreeListSpace::reportFreeListStatistics() const {
552 assert_lock_strong(&_freelistLock);
553 assert(PrintFLSStatistics != 0, "Reporting error");
554 _dictionary->report_statistics();
555 if (PrintFLSStatistics > 1) {
556 reportIndexedFreeListStatistics();
557 size_t total_size = totalSizeInIndexedFreeLists() +
558 _dictionary->total_chunk_size(DEBUG_ONLY(freelistLock()));
559 gclog_or_tty->print(" free=" SIZE_FORMAT " frag=%1.4f\n", total_size, flsFrag());
560 }
561 }
562
563 void CompactibleFreeListSpace::reportIndexedFreeListStatistics() const {
564 assert_lock_strong(&_freelistLock);
565 gclog_or_tty->print("Statistics for IndexedFreeLists:\n"
566 "--------------------------------\n");
567 size_t total_size = totalSizeInIndexedFreeLists();
568 size_t free_blocks = numFreeBlocksInIndexedFreeLists();
569 gclog_or_tty->print("Total Free Space: " SIZE_FORMAT "\n", total_size);
570 gclog_or_tty->print("Max Chunk Size: " SIZE_FORMAT "\n", maxChunkSizeInIndexedFreeLists());
571 gclog_or_tty->print("Number of Blocks: " SIZE_FORMAT "\n", free_blocks);
572 if (free_blocks != 0) {
573 gclog_or_tty->print("Av. Block Size: " SIZE_FORMAT "\n", total_size/free_blocks);
574 }
575 }
576
577 size_t CompactibleFreeListSpace::numFreeBlocksInIndexedFreeLists() const {
578 size_t res = 0;
579 for (size_t i = IndexSetStart; i < IndexSetSize; i += IndexSetStride) {
580 debug_only(
581 ssize_t recount = 0;
582 for (FreeChunk* fc = _indexedFreeList[i].head(); fc != NULL;
583 fc = fc->next()) {
584 recount += 1;
585 }
586 assert(recount == _indexedFreeList[i].count(),
587 "Incorrect count in list");
588 )
589 res += _indexedFreeList[i].count();
590 }
591 return res;
592 }
593
594 size_t CompactibleFreeListSpace::maxChunkSizeInIndexedFreeLists() const {
595 for (size_t i = IndexSetSize - 1; i != 0; i -= IndexSetStride) {
596 if (_indexedFreeList[i].head() != NULL) {
597 assert(_indexedFreeList[i].count() != 0, "Inconsistent FreeList");
598 return (size_t)i;
599 }
600 }
601 return 0;
602 }
603
604 void CompactibleFreeListSpace::set_end(HeapWord* value) {
605 HeapWord* prevEnd = end();
606 assert(prevEnd != value, "unnecessary set_end call");
607 assert(prevEnd == NULL || !BlockOffsetArrayUseUnallocatedBlock || value >= unallocated_block(),
608 "New end is below unallocated block");
609 _end = value;
610 if (prevEnd != NULL) {
611 // Resize the underlying block offset table.
612 _bt.resize(pointer_delta(value, bottom()));
613 if (value <= prevEnd) {
614 assert(!BlockOffsetArrayUseUnallocatedBlock || value >= unallocated_block(),
615 "New end is below unallocated block");
616 } else {
617 // Now, take this new chunk and add it to the free blocks.
618 // Note that the BOT has not yet been updated for this block.
619 size_t newFcSize = pointer_delta(value, prevEnd);
620 // XXX This is REALLY UGLY and should be fixed up. XXX
621 if (!_adaptive_freelists && _smallLinearAllocBlock._ptr == NULL) {
622 // Mark the boundary of the new block in BOT
623 _bt.mark_block(prevEnd, value);
624 // put it all in the linAB
625 if (ParallelGCThreads == 0) {
626 _smallLinearAllocBlock._ptr = prevEnd;
627 _smallLinearAllocBlock._word_size = newFcSize;
628 repairLinearAllocBlock(&_smallLinearAllocBlock);
629 } else { // ParallelGCThreads > 0
630 MutexLockerEx x(parDictionaryAllocLock(),
631 Mutex::_no_safepoint_check_flag);
632 _smallLinearAllocBlock._ptr = prevEnd;
633 _smallLinearAllocBlock._word_size = newFcSize;
634 repairLinearAllocBlock(&_smallLinearAllocBlock);
635 }
636 // Births of chunks put into a LinAB are not recorded. Births
637 // of chunks as they are allocated out of a LinAB are.
638 } else {
639 // Add the block to the free lists, if possible coalescing it
640 // with the last free block, and update the BOT and census data.
641 addChunkToFreeListsAtEndRecordingStats(prevEnd, newFcSize);
642 }
643 }
644 }
645 }
646
647 class FreeListSpace_DCTOC : public Filtering_DCTOC {
648 CompactibleFreeListSpace* _cfls;
649 CMSCollector* _collector;
650 protected:
651 // Override.
652 #define walk_mem_region_with_cl_DECL(ClosureType) \
653 virtual void walk_mem_region_with_cl(MemRegion mr, \
654 HeapWord* bottom, HeapWord* top, \
655 ClosureType* cl); \
656 void walk_mem_region_with_cl_par(MemRegion mr, \
657 HeapWord* bottom, HeapWord* top, \
658 ClosureType* cl); \
659 void walk_mem_region_with_cl_nopar(MemRegion mr, \
660 HeapWord* bottom, HeapWord* top, \
661 ClosureType* cl)
662 walk_mem_region_with_cl_DECL(ExtendedOopClosure);
663 walk_mem_region_with_cl_DECL(FilteringClosure);
664
665 public:
666 FreeListSpace_DCTOC(CompactibleFreeListSpace* sp,
667 CMSCollector* collector,
668 ExtendedOopClosure* cl,
669 CardTableModRefBS::PrecisionStyle precision,
670 HeapWord* boundary) :
671 Filtering_DCTOC(sp, cl, precision, boundary),
672 _cfls(sp), _collector(collector) {}
673 };
674
675 // We de-virtualize the block-related calls below, since we know that our
676 // space is a CompactibleFreeListSpace.
677
678 #define FreeListSpace_DCTOC__walk_mem_region_with_cl_DEFN(ClosureType) \
679 void FreeListSpace_DCTOC::walk_mem_region_with_cl(MemRegion mr, \
680 HeapWord* bottom, \
681 HeapWord* top, \
682 ClosureType* cl) { \
683 bool is_par = SharedHeap::heap()->n_par_threads() > 0; \
684 if (is_par) { \
685 assert(SharedHeap::heap()->n_par_threads() == \
686 SharedHeap::heap()->workers()->active_workers(), "Mismatch"); \
687 walk_mem_region_with_cl_par(mr, bottom, top, cl); \
688 } else { \
689 walk_mem_region_with_cl_nopar(mr, bottom, top, cl); \
690 } \
691 } \
692 void FreeListSpace_DCTOC::walk_mem_region_with_cl_par(MemRegion mr, \
693 HeapWord* bottom, \
694 HeapWord* top, \
695 ClosureType* cl) { \
696 /* Skip parts that are before "mr", in case "block_start" sent us \
697 back too far. */ \
698 HeapWord* mr_start = mr.start(); \
699 size_t bot_size = _cfls->CompactibleFreeListSpace::block_size(bottom); \
700 HeapWord* next = bottom + bot_size; \
701 while (next < mr_start) { \
702 bottom = next; \
703 bot_size = _cfls->CompactibleFreeListSpace::block_size(bottom); \
704 next = bottom + bot_size; \
705 } \
706 \
707 while (bottom < top) { \
708 if (_cfls->CompactibleFreeListSpace::block_is_obj(bottom) && \
709 !_cfls->CompactibleFreeListSpace::obj_allocated_since_save_marks( \
710 oop(bottom)) && \
711 !_collector->CMSCollector::is_dead_obj(oop(bottom))) { \
712 size_t word_sz = oop(bottom)->oop_iterate(cl, mr); \
713 bottom += _cfls->adjustObjectSize(word_sz); \
714 } else { \
715 bottom += _cfls->CompactibleFreeListSpace::block_size(bottom); \
716 } \
717 } \
718 } \
719 void FreeListSpace_DCTOC::walk_mem_region_with_cl_nopar(MemRegion mr, \
720 HeapWord* bottom, \
721 HeapWord* top, \
722 ClosureType* cl) { \
723 /* Skip parts that are before "mr", in case "block_start" sent us \
724 back too far. */ \
725 HeapWord* mr_start = mr.start(); \
726 size_t bot_size = _cfls->CompactibleFreeListSpace::block_size_nopar(bottom); \
727 HeapWord* next = bottom + bot_size; \
728 while (next < mr_start) { \
729 bottom = next; \
730 bot_size = _cfls->CompactibleFreeListSpace::block_size_nopar(bottom); \
731 next = bottom + bot_size; \
732 } \
733 \
734 while (bottom < top) { \
735 if (_cfls->CompactibleFreeListSpace::block_is_obj_nopar(bottom) && \
736 !_cfls->CompactibleFreeListSpace::obj_allocated_since_save_marks( \
737 oop(bottom)) && \
738 !_collector->CMSCollector::is_dead_obj(oop(bottom))) { \
739 size_t word_sz = oop(bottom)->oop_iterate(cl, mr); \
740 bottom += _cfls->adjustObjectSize(word_sz); \
741 } else { \
742 bottom += _cfls->CompactibleFreeListSpace::block_size_nopar(bottom); \
743 } \
744 } \
745 }
746
747 // (There are only two of these, rather than N, because the split is due
748 // only to the introduction of the FilteringClosure, a local part of the
749 // impl of this abstraction.)
750 FreeListSpace_DCTOC__walk_mem_region_with_cl_DEFN(ExtendedOopClosure)
751 FreeListSpace_DCTOC__walk_mem_region_with_cl_DEFN(FilteringClosure)
752
753 DirtyCardToOopClosure*
754 CompactibleFreeListSpace::new_dcto_cl(ExtendedOopClosure* cl,
755 CardTableModRefBS::PrecisionStyle precision,
756 HeapWord* boundary) {
757 return new FreeListSpace_DCTOC(this, _collector, cl, precision, boundary);
758 }
759
760
761 // Note on locking for the space iteration functions:
762 // since the collector's iteration activities are concurrent with
763 // allocation activities by mutators, absent a suitable mutual exclusion
764 // mechanism the iterators may go awry. For instance a block being iterated
765 // may suddenly be allocated or divided up and part of it allocated and
766 // so on.
767
768 // Apply the given closure to each block in the space.
769 void CompactibleFreeListSpace::blk_iterate_careful(BlkClosureCareful* cl) {
770 assert_lock_strong(freelistLock());
771 HeapWord *cur, *limit;
772 for (cur = bottom(), limit = end(); cur < limit;
773 cur += cl->do_blk_careful(cur));
774 }
775
776 // Apply the given closure to each block in the space.
777 void CompactibleFreeListSpace::blk_iterate(BlkClosure* cl) {
778 assert_lock_strong(freelistLock());
779 HeapWord *cur, *limit;
780 for (cur = bottom(), limit = end(); cur < limit;
781 cur += cl->do_blk(cur));
782 }
783
784 // Apply the given closure to each oop in the space.
785 void CompactibleFreeListSpace::oop_iterate(ExtendedOopClosure* cl) {
786 assert_lock_strong(freelistLock());
787 HeapWord *cur, *limit;
788 size_t curSize;
789 for (cur = bottom(), limit = end(); cur < limit;
790 cur += curSize) {
791 curSize = block_size(cur);
792 if (block_is_obj(cur)) {
793 oop(cur)->oop_iterate(cl);
794 }
795 }
796 }
797
798 // NOTE: In the following methods, in order to safely be able to
799 // apply the closure to an object, we need to be sure that the
800 // object has been initialized. We are guaranteed that an object
801 // is initialized if we are holding the Heap_lock with the
802 // world stopped.
803 void CompactibleFreeListSpace::verify_objects_initialized() const {
804 if (is_init_completed()) {
805 assert_locked_or_safepoint(Heap_lock);
806 if (Universe::is_fully_initialized()) {
807 guarantee(SafepointSynchronize::is_at_safepoint(),
808 "Required for objects to be initialized");
809 }
810 } // else make a concession at vm start-up
811 }
812
813 // Apply the given closure to each object in the space
814 void CompactibleFreeListSpace::object_iterate(ObjectClosure* blk) {
815 assert_lock_strong(freelistLock());
816 NOT_PRODUCT(verify_objects_initialized());
817 HeapWord *cur, *limit;
818 size_t curSize;
819 for (cur = bottom(), limit = end(); cur < limit;
820 cur += curSize) {
821 curSize = block_size(cur);
822 if (block_is_obj(cur)) {
823 blk->do_object(oop(cur));
824 }
825 }
826 }
827
828 // Apply the given closure to each live object in the space
829 // The usage of CompactibleFreeListSpace
830 // by the ConcurrentMarkSweepGeneration for concurrent GC's allows
831 // objects in the space with references to objects that are no longer
832 // valid. For example, an object may reference another object
833 // that has already been sweep up (collected). This method uses
834 // obj_is_alive() to determine whether it is safe to apply the closure to
835 // an object. See obj_is_alive() for details on how liveness of an
836 // object is decided.
837
838 void CompactibleFreeListSpace::safe_object_iterate(ObjectClosure* blk) {
839 assert_lock_strong(freelistLock());
840 NOT_PRODUCT(verify_objects_initialized());
841 HeapWord *cur, *limit;
842 size_t curSize;
843 for (cur = bottom(), limit = end(); cur < limit;
844 cur += curSize) {
845 curSize = block_size(cur);
846 if (block_is_obj(cur) && obj_is_alive(cur)) {
847 blk->do_object(oop(cur));
848 }
849 }
850 }
851
852 void CompactibleFreeListSpace::object_iterate_mem(MemRegion mr,
853 UpwardsObjectClosure* cl) {
854 assert_locked(freelistLock());
855 NOT_PRODUCT(verify_objects_initialized());
856 assert(!mr.is_empty(), "Should be non-empty");
857 // We use MemRegion(bottom(), end()) rather than used_region() below
858 // because the two are not necessarily equal for some kinds of
859 // spaces, in particular, certain kinds of free list spaces.
860 // We could use the more complicated but more precise:
861 // MemRegion(used_region().start(), round_to(used_region().end(), CardSize))
862 // but the slight imprecision seems acceptable in the assertion check.
863 assert(MemRegion(bottom(), end()).contains(mr),
864 "Should be within used space");
865 HeapWord* prev = cl->previous(); // max address from last time
866 if (prev >= mr.end()) { // nothing to do
867 return;
868 }
869 // This assert will not work when we go from cms space to perm
870 // space, and use same closure. Easy fix deferred for later. XXX YSR
871 // assert(prev == NULL || contains(prev), "Should be within space");
872
873 bool last_was_obj_array = false;
874 HeapWord *blk_start_addr, *region_start_addr;
875 if (prev > mr.start()) {
876 region_start_addr = prev;
877 blk_start_addr = prev;
878 // The previous invocation may have pushed "prev" beyond the
879 // last allocated block yet there may be still be blocks
880 // in this region due to a particular coalescing policy.
881 // Relax the assertion so that the case where the unallocated
882 // block is maintained and "prev" is beyond the unallocated
883 // block does not cause the assertion to fire.
884 assert((BlockOffsetArrayUseUnallocatedBlock &&
885 (!is_in(prev))) ||
886 (blk_start_addr == block_start(region_start_addr)), "invariant");
887 } else {
888 region_start_addr = mr.start();
889 blk_start_addr = block_start(region_start_addr);
890 }
891 HeapWord* region_end_addr = mr.end();
892 MemRegion derived_mr(region_start_addr, region_end_addr);
893 while (blk_start_addr < region_end_addr) {
894 const size_t size = block_size(blk_start_addr);
895 if (block_is_obj(blk_start_addr)) {
896 last_was_obj_array = cl->do_object_bm(oop(blk_start_addr), derived_mr);
897 } else {
898 last_was_obj_array = false;
899 }
900 blk_start_addr += size;
901 }
902 if (!last_was_obj_array) {
903 assert((bottom() <= blk_start_addr) && (blk_start_addr <= end()),
904 "Should be within (closed) used space");
905 assert(blk_start_addr > prev, "Invariant");
906 cl->set_previous(blk_start_addr); // min address for next time
907 }
908 }
909
910 // Callers of this iterator beware: The closure application should
911 // be robust in the face of uninitialized objects and should (always)
912 // return a correct size so that the next addr + size below gives us a
913 // valid block boundary. [See for instance,
914 // ScanMarkedObjectsAgainCarefullyClosure::do_object_careful()
915 // in ConcurrentMarkSweepGeneration.cpp.]
916 HeapWord*
917 CompactibleFreeListSpace::object_iterate_careful_m(MemRegion mr,
918 ObjectClosureCareful* cl) {
919 assert_lock_strong(freelistLock());
920 // Can't use used_region() below because it may not necessarily
921 // be the same as [bottom(),end()); although we could
922 // use [used_region().start(),round_to(used_region().end(),CardSize)),
923 // that appears too cumbersome, so we just do the simpler check
924 // in the assertion below.
925 assert(!mr.is_empty() && MemRegion(bottom(),end()).contains(mr),
926 "mr should be non-empty and within used space");
927 HeapWord *addr, *end;
928 size_t size;
929 for (addr = block_start_careful(mr.start()), end = mr.end();
930 addr < end; addr += size) {
931 FreeChunk* fc = (FreeChunk*)addr;
932 if (fc->is_free()) {
933 // Since we hold the free list lock, which protects direct
934 // allocation in this generation by mutators, a free object
935 // will remain free throughout this iteration code.
936 size = fc->size();
937 } else {
938 // Note that the object need not necessarily be initialized,
939 // because (for instance) the free list lock does NOT protect
940 // object initialization. The closure application below must
941 // therefore be correct in the face of uninitialized objects.
942 size = cl->do_object_careful_m(oop(addr), mr);
943 if (size == 0) {
944 // An unparsable object found. Signal early termination.
945 return addr;
946 }
947 }
948 }
949 return NULL;
950 }
951
952
953 HeapWord* CompactibleFreeListSpace::block_start_const(const void* p) const {
954 NOT_PRODUCT(verify_objects_initialized());
955 return _bt.block_start(p);
956 }
957
958 HeapWord* CompactibleFreeListSpace::block_start_careful(const void* p) const {
959 return _bt.block_start_careful(p);
960 }
961
962 size_t CompactibleFreeListSpace::block_size(const HeapWord* p) const {
963 NOT_PRODUCT(verify_objects_initialized());
964 // This must be volatile, or else there is a danger that the compiler
965 // will compile the code below into a sometimes-infinite loop, by keeping
966 // the value read the first time in a register.
967 while (true) {
968 // We must do this until we get a consistent view of the object.
969 if (FreeChunk::indicatesFreeChunk(p)) {
970 volatile FreeChunk* fc = (volatile FreeChunk*)p;
971 size_t res = fc->size();
972
973 // Bugfix for systems with weak memory model (PPC64/IA64). The
974 // block's free bit was set and we have read the size of the
975 // block. Acquire and check the free bit again. If the block is
976 // still free, the read size is correct.
977 OrderAccess::acquire();
978
979 // If the object is still a free chunk, return the size, else it
980 // has been allocated so try again.
981 if (FreeChunk::indicatesFreeChunk(p)) {
982 assert(res != 0, "Block size should not be 0");
983 return res;
984 }
985 } else {
986 // must read from what 'p' points to in each loop.
987 Klass* k = ((volatile oopDesc*)p)->klass_or_null();
988 if (k != NULL) {
989 assert(k->is_klass(), "Should really be klass oop.");
990 oop o = (oop)p;
991 assert(o->is_oop(true /* ignore mark word */), "Should be an oop.");
992
993 // Bugfix for systems with weak memory model (PPC64/IA64).
994 // The object o may be an array. Acquire to make sure that the array
995 // size (third word) is consistent.
996 OrderAccess::acquire();
997
998 size_t res = o->size_given_klass(k);
999 res = adjustObjectSize(res);
1000 assert(res != 0, "Block size should not be 0");
1001 return res;
1002 }
1003 }
1004 }
1005 }
1006
1007 // TODO: Now that is_parsable is gone, we should combine these two functions.
1008 // A variant of the above that uses the Printezis bits for
1009 // unparsable but allocated objects. This avoids any possible
1010 // stalls waiting for mutators to initialize objects, and is
1011 // thus potentially faster than the variant above. However,
1012 // this variant may return a zero size for a block that is
1013 // under mutation and for which a consistent size cannot be
1014 // inferred without stalling; see CMSCollector::block_size_if_printezis_bits().
1015 size_t CompactibleFreeListSpace::block_size_no_stall(HeapWord* p,
1016 const CMSCollector* c)
1017 const {
1018 assert(MemRegion(bottom(), end()).contains(p), "p not in space");
1019 // This must be volatile, or else there is a danger that the compiler
1020 // will compile the code below into a sometimes-infinite loop, by keeping
1021 // the value read the first time in a register.
1022 DEBUG_ONLY(uint loops = 0;)
1023 while (true) {
1024 // We must do this until we get a consistent view of the object.
1025 if (FreeChunk::indicatesFreeChunk(p)) {
1026 volatile FreeChunk* fc = (volatile FreeChunk*)p;
1027 size_t res = fc->size();
1028
1029 // Bugfix for systems with weak memory model (PPC64/IA64). The
1030 // free bit of the block was set and we have read the size of
1031 // the block. Acquire and check the free bit again. If the
1032 // block is still free, the read size is correct.
1033 OrderAccess::acquire();
1034
1035 if (FreeChunk::indicatesFreeChunk(p)) {
1036 assert(res != 0, "Block size should not be 0");
1037 assert(loops == 0, "Should be 0");
1038 return res;
1039 }
1040 } else {
1041 // must read from what 'p' points to in each loop.
1042 Klass* k = ((volatile oopDesc*)p)->klass_or_null();
1043 // We trust the size of any object that has a non-NULL
1044 // klass and (for those in the perm gen) is parsable
1045 // -- irrespective of its conc_safe-ty.
1046 if (k != NULL) {
1047 assert(k->is_klass(), "Should really be klass oop.");
1048 oop o = (oop)p;
1049 assert(o->is_oop(), "Should be an oop");
1050
1051 // Bugfix for systems with weak memory model (PPC64/IA64).
1052 // The object o may be an array. Acquire to make sure that the array
1053 // size (third word) is consistent.
1054 OrderAccess::acquire();
1055
1056 size_t res = o->size_given_klass(k);
1057 res = adjustObjectSize(res);
1058 assert(res != 0, "Block size should not be 0");
1059 return res;
1060 } else {
1061 // May return 0 if P-bits not present.
1062 return c->block_size_if_printezis_bits(p);
1063 }
1064 }
1065 assert(loops == 0, "Can loop at most once");
1066 DEBUG_ONLY(loops++;)
1067 }
1068 }
1069
1070 size_t CompactibleFreeListSpace::block_size_nopar(const HeapWord* p) const {
1071 NOT_PRODUCT(verify_objects_initialized());
1072 assert(MemRegion(bottom(), end()).contains(p), "p not in space");
1073 FreeChunk* fc = (FreeChunk*)p;
1074 if (fc->is_free()) {
1075 return fc->size();
1076 } else {
1077 // Ignore mark word because this may be a recently promoted
1078 // object whose mark word is used to chain together grey
1079 // objects (the last one would have a null value).
1080 assert(oop(p)->is_oop(true), "Should be an oop");
1081 return adjustObjectSize(oop(p)->size());
1082 }
1083 }
1084
1085 // This implementation assumes that the property of "being an object" is
1086 // stable. But being a free chunk may not be (because of parallel
1087 // promotion.)
1088 bool CompactibleFreeListSpace::block_is_obj(const HeapWord* p) const {
1089 FreeChunk* fc = (FreeChunk*)p;
1090 assert(is_in_reserved(p), "Should be in space");
1091 // When doing a mark-sweep-compact of the CMS generation, this
1092 // assertion may fail because prepare_for_compaction() uses
1093 // space that is garbage to maintain information on ranges of
1094 // live objects so that these live ranges can be moved as a whole.
1095 // Comment out this assertion until that problem can be solved
1096 // (i.e., that the block start calculation may look at objects
1097 // at address below "p" in finding the object that contains "p"
1098 // and those objects (if garbage) may have been modified to hold
1099 // live range information.
1100 // assert(CollectedHeap::use_parallel_gc_threads() || _bt.block_start(p) == p,
1101 // "Should be a block boundary");
1102 if (FreeChunk::indicatesFreeChunk(p)) return false;
1103 Klass* k = oop(p)->klass_or_null();
1104 if (k != NULL) {
1105 // Ignore mark word because it may have been used to
1106 // chain together promoted objects (the last one
1107 // would have a null value).
1108 assert(oop(p)->is_oop(true), "Should be an oop");
1109 return true;
1110 } else {
1111 return false; // Was not an object at the start of collection.
1112 }
1113 }
1114
1115 // Check if the object is alive. This fact is checked either by consulting
1116 // the main marking bitmap in the sweeping phase or, if it's a permanent
1117 // generation and we're not in the sweeping phase, by checking the
1118 // perm_gen_verify_bit_map where we store the "deadness" information if
1119 // we did not sweep the perm gen in the most recent previous GC cycle.
1120 bool CompactibleFreeListSpace::obj_is_alive(const HeapWord* p) const {
1121 assert(SafepointSynchronize::is_at_safepoint() || !is_init_completed(),
1122 "Else races are possible");
1123 assert(block_is_obj(p), "The address should point to an object");
1124
1125 // If we're sweeping, we use object liveness information from the main bit map
1126 // for both perm gen and old gen.
1127 // We don't need to lock the bitmap (live_map or dead_map below), because
1128 // EITHER we are in the middle of the sweeping phase, and the
1129 // main marking bit map (live_map below) is locked,
1130 // OR we're in other phases and perm_gen_verify_bit_map (dead_map below)
1131 // is stable, because it's mutated only in the sweeping phase.
1132 // NOTE: This method is also used by jmap where, if class unloading is
1133 // off, the results can return "false" for legitimate perm objects,
1134 // when we are not in the midst of a sweeping phase, which can result
1135 // in jmap not reporting certain perm gen objects. This will be moot
1136 // if/when the perm gen goes away in the future.
1137 if (_collector->abstract_state() == CMSCollector::Sweeping) {
1138 CMSBitMap* live_map = _collector->markBitMap();
1139 return live_map->par_isMarked((HeapWord*) p);
1140 }
1141 return true;
1142 }
1143
1144 bool CompactibleFreeListSpace::block_is_obj_nopar(const HeapWord* p) const {
1145 FreeChunk* fc = (FreeChunk*)p;
1146 assert(is_in_reserved(p), "Should be in space");
1147 assert(_bt.block_start(p) == p, "Should be a block boundary");
1148 if (!fc->is_free()) {
1149 // Ignore mark word because it may have been used to
1150 // chain together promoted objects (the last one
1151 // would have a null value).
1152 assert(oop(p)->is_oop(true), "Should be an oop");
1153 return true;
1154 }
1155 return false;
1156 }
1157
1158 // "MT-safe but not guaranteed MT-precise" (TM); you may get an
1159 // approximate answer if you don't hold the freelistlock when you call this.
1160 size_t CompactibleFreeListSpace::totalSizeInIndexedFreeLists() const {
1161 size_t size = 0;
1162 for (size_t i = IndexSetStart; i < IndexSetSize; i += IndexSetStride) {
1163 debug_only(
1164 // We may be calling here without the lock in which case we
1165 // won't do this modest sanity check.
1166 if (freelistLock()->owned_by_self()) {
1167 size_t total_list_size = 0;
1168 for (FreeChunk* fc = _indexedFreeList[i].head(); fc != NULL;
1169 fc = fc->next()) {
1170 total_list_size += i;
1171 }
1172 assert(total_list_size == i * _indexedFreeList[i].count(),
1173 "Count in list is incorrect");
1174 }
1175 )
1176 size += i * _indexedFreeList[i].count();
1177 }
1178 return size;
1179 }
1180
1181 HeapWord* CompactibleFreeListSpace::par_allocate(size_t size) {
1182 MutexLockerEx x(freelistLock(), Mutex::_no_safepoint_check_flag);
1183 return allocate(size);
1184 }
1185
1186 HeapWord*
1187 CompactibleFreeListSpace::getChunkFromSmallLinearAllocBlockRemainder(size_t size) {
1188 return getChunkFromLinearAllocBlockRemainder(&_smallLinearAllocBlock, size);
1189 }
1190
1191 HeapWord* CompactibleFreeListSpace::allocate(size_t size) {
1192 assert_lock_strong(freelistLock());
1193 HeapWord* res = NULL;
1194 assert(size == adjustObjectSize(size),
1195 "use adjustObjectSize() before calling into allocate()");
1196
1197 if (_adaptive_freelists) {
1198 res = allocate_adaptive_freelists(size);
1199 } else { // non-adaptive free lists
1200 res = allocate_non_adaptive_freelists(size);
1201 }
1202
1203 if (res != NULL) {
1204 // check that res does lie in this space!
1205 assert(is_in_reserved(res), "Not in this space!");
1206 assert(is_aligned((void*)res), "alignment check");
1207
1208 FreeChunk* fc = (FreeChunk*)res;
1209 fc->markNotFree();
1210 assert(!fc->is_free(), "shouldn't be marked free");
1211 assert(oop(fc)->klass_or_null() == NULL, "should look uninitialized");
1212 // Verify that the block offset table shows this to
1213 // be a single block, but not one which is unallocated.
1214 _bt.verify_single_block(res, size);
1215 _bt.verify_not_unallocated(res, size);
1216 // mangle a just allocated object with a distinct pattern.
1217 debug_only(fc->mangleAllocated(size));
1218 }
1219
1220 return res;
1221 }
1222
1223 HeapWord* CompactibleFreeListSpace::allocate_non_adaptive_freelists(size_t size) {
1224 HeapWord* res = NULL;
1225 // try and use linear allocation for smaller blocks
1226 if (size < _smallLinearAllocBlock._allocation_size_limit) {
1227 // if successful, the following also adjusts block offset table
1228 res = getChunkFromSmallLinearAllocBlock(size);
1229 }
1230 // Else triage to indexed lists for smaller sizes
1231 if (res == NULL) {
1232 if (size < SmallForDictionary) {
1233 res = (HeapWord*) getChunkFromIndexedFreeList(size);
1234 } else {
1235 // else get it from the big dictionary; if even this doesn't
1236 // work we are out of luck.
1237 res = (HeapWord*)getChunkFromDictionaryExact(size);
1238 }
1239 }
1240
1241 return res;
1242 }
1243
1244 HeapWord* CompactibleFreeListSpace::allocate_adaptive_freelists(size_t size) {
1245 assert_lock_strong(freelistLock());
1246 HeapWord* res = NULL;
1247 assert(size == adjustObjectSize(size),
1248 "use adjustObjectSize() before calling into allocate()");
1249
1250 // Strategy
1251 // if small
1252 // exact size from small object indexed list if small
1253 // small or large linear allocation block (linAB) as appropriate
1254 // take from lists of greater sized chunks
1255 // else
1256 // dictionary
1257 // small or large linear allocation block if it has the space
1258 // Try allocating exact size from indexTable first
1259 if (size < IndexSetSize) {
1260 res = (HeapWord*) getChunkFromIndexedFreeList(size);
1261 if(res != NULL) {
1262 assert(res != (HeapWord*)_indexedFreeList[size].head(),
1263 "Not removed from free list");
1264 // no block offset table adjustment is necessary on blocks in
1265 // the indexed lists.
1266
1267 // Try allocating from the small LinAB
1268 } else if (size < _smallLinearAllocBlock._allocation_size_limit &&
1269 (res = getChunkFromSmallLinearAllocBlock(size)) != NULL) {
1270 // if successful, the above also adjusts block offset table
1271 // Note that this call will refill the LinAB to
1272 // satisfy the request. This is different that
1273 // evm.
1274 // Don't record chunk off a LinAB? smallSplitBirth(size);
1275 } else {
1276 // Raid the exact free lists larger than size, even if they are not
1277 // overpopulated.
1278 res = (HeapWord*) getChunkFromGreater(size);
1279 }
1280 } else {
1281 // Big objects get allocated directly from the dictionary.
1282 res = (HeapWord*) getChunkFromDictionaryExact(size);
1283 if (res == NULL) {
1284 // Try hard not to fail since an allocation failure will likely
1285 // trigger a synchronous GC. Try to get the space from the
1286 // allocation blocks.
1287 res = getChunkFromSmallLinearAllocBlockRemainder(size);
1288 }
1289 }
1290
1291 return res;
1292 }
1293
1294 // A worst-case estimate of the space required (in HeapWords) to expand the heap
1295 // when promoting obj.
1296 size_t CompactibleFreeListSpace::expansionSpaceRequired(size_t obj_size) const {
1297 // Depending on the object size, expansion may require refilling either a
1298 // bigLAB or a smallLAB plus refilling a PromotionInfo object. MinChunkSize
1299 // is added because the dictionary may over-allocate to avoid fragmentation.
1300 size_t space = obj_size;
1301 if (!_adaptive_freelists) {
1302 space = MAX2(space, _smallLinearAllocBlock._refillSize);
1303 }
1304 space += _promoInfo.refillSize() + 2 * MinChunkSize;
1305 return space;
1306 }
1307
1308 FreeChunk* CompactibleFreeListSpace::getChunkFromGreater(size_t numWords) {
1309 FreeChunk* ret;
1310
1311 assert(numWords >= MinChunkSize, "Size is less than minimum");
1312 assert(linearAllocationWouldFail() || bestFitFirst(),
1313 "Should not be here");
1314
1315 size_t i;
1316 size_t currSize = numWords + MinChunkSize;
1317 assert(currSize % MinObjAlignment == 0, "currSize should be aligned");
1318 for (i = currSize; i < IndexSetSize; i += IndexSetStride) {
1319 AdaptiveFreeList<FreeChunk>* fl = &_indexedFreeList[i];
1320 if (fl->head()) {
1321 ret = getFromListGreater(fl, numWords);
1322 assert(ret == NULL || ret->is_free(), "Should be returning a free chunk");
1323 return ret;
1324 }
1325 }
1326
1327 currSize = MAX2((size_t)SmallForDictionary,
1328 (size_t)(numWords + MinChunkSize));
1329
1330 /* Try to get a chunk that satisfies request, while avoiding
1331 fragmentation that can't be handled. */
1332 {
1333 ret = dictionary()->get_chunk(currSize);
1334 if (ret != NULL) {
1335 assert(ret->size() - numWords >= MinChunkSize,
1336 "Chunk is too small");
1337 _bt.allocated((HeapWord*)ret, ret->size());
1338 /* Carve returned chunk. */
1339 (void) splitChunkAndReturnRemainder(ret, numWords);
1340 /* Label this as no longer a free chunk. */
1341 assert(ret->is_free(), "This chunk should be free");
1342 ret->link_prev(NULL);
1343 }
1344 assert(ret == NULL || ret->is_free(), "Should be returning a free chunk");
1345 return ret;
1346 }
1347 ShouldNotReachHere();
1348 }
1349
1350 bool CompactibleFreeListSpace::verifyChunkInIndexedFreeLists(FreeChunk* fc) const {
1351 assert(fc->size() < IndexSetSize, "Size of chunk is too large");
1352 return _indexedFreeList[fc->size()].verify_chunk_in_free_list(fc);
1353 }
1354
1355 bool CompactibleFreeListSpace::verify_chunk_is_linear_alloc_block(FreeChunk* fc) const {
1356 assert((_smallLinearAllocBlock._ptr != (HeapWord*)fc) ||
1357 (_smallLinearAllocBlock._word_size == fc->size()),
1358 "Linear allocation block shows incorrect size");
1359 return ((_smallLinearAllocBlock._ptr == (HeapWord*)fc) &&
1360 (_smallLinearAllocBlock._word_size == fc->size()));
1361 }
1362
1363 // Check if the purported free chunk is present either as a linear
1364 // allocation block, the size-indexed table of (smaller) free blocks,
1365 // or the larger free blocks kept in the binary tree dictionary.
1366 bool CompactibleFreeListSpace::verify_chunk_in_free_list(FreeChunk* fc) const {
1367 if (verify_chunk_is_linear_alloc_block(fc)) {
1368 return true;
1369 } else if (fc->size() < IndexSetSize) {
1370 return verifyChunkInIndexedFreeLists(fc);
1371 } else {
1372 return dictionary()->verify_chunk_in_free_list(fc);
1373 }
1374 }
1375
1376 #ifndef PRODUCT
1377 void CompactibleFreeListSpace::assert_locked() const {
1378 CMSLockVerifier::assert_locked(freelistLock(), parDictionaryAllocLock());
1379 }
1380
1381 void CompactibleFreeListSpace::assert_locked(const Mutex* lock) const {
1382 CMSLockVerifier::assert_locked(lock);
1383 }
1384 #endif
1385
1386 FreeChunk* CompactibleFreeListSpace::allocateScratch(size_t size) {
1387 // In the parallel case, the main thread holds the free list lock
1388 // on behalf the parallel threads.
1389 FreeChunk* fc;
1390 {
1391 // If GC is parallel, this might be called by several threads.
1392 // This should be rare enough that the locking overhead won't affect
1393 // the sequential code.
1394 MutexLockerEx x(parDictionaryAllocLock(),
1395 Mutex::_no_safepoint_check_flag);
1396 fc = getChunkFromDictionary(size);
1397 }
1398 if (fc != NULL) {
1399 fc->dontCoalesce();
1400 assert(fc->is_free(), "Should be free, but not coalescable");
1401 // Verify that the block offset table shows this to
1402 // be a single block, but not one which is unallocated.
1403 _bt.verify_single_block((HeapWord*)fc, fc->size());
1404 _bt.verify_not_unallocated((HeapWord*)fc, fc->size());
1405 }
1406 return fc;
1407 }
1408
1409 oop CompactibleFreeListSpace::promote(oop obj, size_t obj_size) {
1410 assert(obj_size == (size_t)obj->size(), "bad obj_size passed in");
1411 assert_locked();
1412
1413 // if we are tracking promotions, then first ensure space for
1414 // promotion (including spooling space for saving header if necessary).
1415 // then allocate and copy, then track promoted info if needed.
1416 // When tracking (see PromotionInfo::track()), the mark word may
1417 // be displaced and in this case restoration of the mark word
1418 // occurs in the (oop_since_save_marks_)iterate phase.
1419 if (_promoInfo.tracking() && !_promoInfo.ensure_spooling_space()) {
1420 return NULL;
1421 }
1422 // Call the allocate(size_t, bool) form directly to avoid the
1423 // additional call through the allocate(size_t) form. Having
1424 // the compile inline the call is problematic because allocate(size_t)
1425 // is a virtual method.
1426 HeapWord* res = allocate(adjustObjectSize(obj_size));
1427 if (res != NULL) {
1428 Copy::aligned_disjoint_words((HeapWord*)obj, res, obj_size);
1429 // if we should be tracking promotions, do so.
1430 if (_promoInfo.tracking()) {
1431 _promoInfo.track((PromotedObject*)res);
1432 }
1433 }
1434 return oop(res);
1435 }
1436
1437 HeapWord*
1438 CompactibleFreeListSpace::getChunkFromSmallLinearAllocBlock(size_t size) {
1439 assert_locked();
1440 assert(size >= MinChunkSize, "minimum chunk size");
1441 assert(size < _smallLinearAllocBlock._allocation_size_limit,
1442 "maximum from smallLinearAllocBlock");
1443 return getChunkFromLinearAllocBlock(&_smallLinearAllocBlock, size);
1444 }
1445
1446 HeapWord*
1447 CompactibleFreeListSpace::getChunkFromLinearAllocBlock(LinearAllocBlock *blk,
1448 size_t size) {
1449 assert_locked();
1450 assert(size >= MinChunkSize, "too small");
1451 HeapWord* res = NULL;
1452 // Try to do linear allocation from blk, making sure that
1453 if (blk->_word_size == 0) {
1454 // We have probably been unable to fill this either in the prologue or
1455 // when it was exhausted at the last linear allocation. Bail out until
1456 // next time.
1457 assert(blk->_ptr == NULL, "consistency check");
1458 return NULL;
1459 }
1460 assert(blk->_word_size != 0 && blk->_ptr != NULL, "consistency check");
1461 res = getChunkFromLinearAllocBlockRemainder(blk, size);
1462 if (res != NULL) return res;
1463
1464 // about to exhaust this linear allocation block
1465 if (blk->_word_size == size) { // exactly satisfied
1466 res = blk->_ptr;
1467 _bt.allocated(res, blk->_word_size);
1468 } else if (size + MinChunkSize <= blk->_refillSize) {
1469 size_t sz = blk->_word_size;
1470 // Update _unallocated_block if the size is such that chunk would be
1471 // returned to the indexed free list. All other chunks in the indexed
1472 // free lists are allocated from the dictionary so that _unallocated_block
1473 // has already been adjusted for them. Do it here so that the cost
1474 // for all chunks added back to the indexed free lists.
1475 if (sz < SmallForDictionary) {
1476 _bt.allocated(blk->_ptr, sz);
1477 }
1478 // Return the chunk that isn't big enough, and then refill below.
1479 addChunkToFreeLists(blk->_ptr, sz);
1480 split_birth(sz);
1481 // Don't keep statistics on adding back chunk from a LinAB.
1482 } else {
1483 // A refilled block would not satisfy the request.
1484 return NULL;
1485 }
1486
1487 blk->_ptr = NULL; blk->_word_size = 0;
1488 refillLinearAllocBlock(blk);
1489 assert(blk->_ptr == NULL || blk->_word_size >= size + MinChunkSize,
1490 "block was replenished");
1491 if (res != NULL) {
1492 split_birth(size);
1493 repairLinearAllocBlock(blk);
1494 } else if (blk->_ptr != NULL) {
1495 res = blk->_ptr;
1496 size_t blk_size = blk->_word_size;
1497 blk->_word_size -= size;
1498 blk->_ptr += size;
1499 split_birth(size);
1500 repairLinearAllocBlock(blk);
1501 // Update BOT last so that other (parallel) GC threads see a consistent
1502 // view of the BOT and free blocks.
1503 // Above must occur before BOT is updated below.
1504 OrderAccess::storestore();
1505 _bt.split_block(res, blk_size, size); // adjust block offset table
1506 }
1507 return res;
1508 }
1509
1510 HeapWord* CompactibleFreeListSpace::getChunkFromLinearAllocBlockRemainder(
1511 LinearAllocBlock* blk,
1512 size_t size) {
1513 assert_locked();
1514 assert(size >= MinChunkSize, "too small");
1515
1516 HeapWord* res = NULL;
1517 // This is the common case. Keep it simple.
1518 if (blk->_word_size >= size + MinChunkSize) {
1519 assert(blk->_ptr != NULL, "consistency check");
1520 res = blk->_ptr;
1521 // Note that the BOT is up-to-date for the linAB before allocation. It
1522 // indicates the start of the linAB. The split_block() updates the
1523 // BOT for the linAB after the allocation (indicates the start of the
1524 // next chunk to be allocated).
1525 size_t blk_size = blk->_word_size;
1526 blk->_word_size -= size;
1527 blk->_ptr += size;
1528 split_birth(size);
1529 repairLinearAllocBlock(blk);
1530 // Update BOT last so that other (parallel) GC threads see a consistent
1531 // view of the BOT and free blocks.
1532 // Above must occur before BOT is updated below.
1533 OrderAccess::storestore();
1534 _bt.split_block(res, blk_size, size); // adjust block offset table
1535 _bt.allocated(res, size);
1536 }
1537 return res;
1538 }
1539
1540 FreeChunk*
1541 CompactibleFreeListSpace::getChunkFromIndexedFreeList(size_t size) {
1542 assert_locked();
1543 assert(size < SmallForDictionary, "just checking");
1544 FreeChunk* res;
1545 res = _indexedFreeList[size].get_chunk_at_head();
1546 if (res == NULL) {
1547 res = getChunkFromIndexedFreeListHelper(size);
1548 }
1549 _bt.verify_not_unallocated((HeapWord*) res, size);
1550 assert(res == NULL || res->size() == size, "Incorrect block size");
1551 return res;
1552 }
1553
1554 FreeChunk*
1555 CompactibleFreeListSpace::getChunkFromIndexedFreeListHelper(size_t size,
1556 bool replenish) {
1557 assert_locked();
1558 FreeChunk* fc = NULL;
1559 if (size < SmallForDictionary) {
1560 assert(_indexedFreeList[size].head() == NULL ||
1561 _indexedFreeList[size].surplus() <= 0,
1562 "List for this size should be empty or under populated");
1563 // Try best fit in exact lists before replenishing the list
1564 if (!bestFitFirst() || (fc = bestFitSmall(size)) == NULL) {
1565 // Replenish list.
1566 //
1567 // Things tried that failed.
1568 // Tried allocating out of the two LinAB's first before
1569 // replenishing lists.
1570 // Tried small linAB of size 256 (size in indexed list)
1571 // and replenishing indexed lists from the small linAB.
1572 //
1573 FreeChunk* newFc = NULL;
1574 const size_t replenish_size = CMSIndexedFreeListReplenish * size;
1575 if (replenish_size < SmallForDictionary) {
1576 // Do not replenish from an underpopulated size.
1577 if (_indexedFreeList[replenish_size].surplus() > 0 &&
1578 _indexedFreeList[replenish_size].head() != NULL) {
1579 newFc = _indexedFreeList[replenish_size].get_chunk_at_head();
1580 } else if (bestFitFirst()) {
1581 newFc = bestFitSmall(replenish_size);
1582 }
1583 }
1584 if (newFc == NULL && replenish_size > size) {
1585 assert(CMSIndexedFreeListReplenish > 1, "ctl pt invariant");
1586 newFc = getChunkFromIndexedFreeListHelper(replenish_size, false);
1587 }
1588 // Note: The stats update re split-death of block obtained above
1589 // will be recorded below precisely when we know we are going to
1590 // be actually splitting it into more than one pieces below.
1591 if (newFc != NULL) {
1592 if (replenish || CMSReplenishIntermediate) {
1593 // Replenish this list and return one block to caller.
1594 size_t i;
1595 FreeChunk *curFc, *nextFc;
1596 size_t num_blk = newFc->size() / size;
1597 assert(num_blk >= 1, "Smaller than requested?");
1598 assert(newFc->size() % size == 0, "Should be integral multiple of request");
1599 if (num_blk > 1) {
1600 // we are sure we will be splitting the block just obtained
1601 // into multiple pieces; record the split-death of the original
1602 splitDeath(replenish_size);
1603 }
1604 // carve up and link blocks 0, ..., num_blk - 2
1605 // The last chunk is not added to the lists but is returned as the
1606 // free chunk.
1607 for (curFc = newFc, nextFc = (FreeChunk*)((HeapWord*)curFc + size),
1608 i = 0;
1609 i < (num_blk - 1);
1610 curFc = nextFc, nextFc = (FreeChunk*)((HeapWord*)nextFc + size),
1611 i++) {
1612 curFc->set_size(size);
1613 // Don't record this as a return in order to try and
1614 // determine the "returns" from a GC.
1615 _bt.verify_not_unallocated((HeapWord*) fc, size);
1616 _indexedFreeList[size].return_chunk_at_tail(curFc, false);
1617 _bt.mark_block((HeapWord*)curFc, size);
1618 split_birth(size);
1619 // Don't record the initial population of the indexed list
1620 // as a split birth.
1621 }
1622
1623 // check that the arithmetic was OK above
1624 assert((HeapWord*)nextFc == (HeapWord*)newFc + num_blk*size,
1625 "inconsistency in carving newFc");
1626 curFc->set_size(size);
1627 _bt.mark_block((HeapWord*)curFc, size);
1628 split_birth(size);
1629 fc = curFc;
1630 } else {
1631 // Return entire block to caller
1632 fc = newFc;
1633 }
1634 }
1635 }
1636 } else {
1637 // Get a free chunk from the free chunk dictionary to be returned to
1638 // replenish the indexed free list.
1639 fc = getChunkFromDictionaryExact(size);
1640 }
1641 // assert(fc == NULL || fc->is_free(), "Should be returning a free chunk");
1642 return fc;
1643 }
1644
1645 FreeChunk*
1646 CompactibleFreeListSpace::getChunkFromDictionary(size_t size) {
1647 assert_locked();
1648 FreeChunk* fc = _dictionary->get_chunk(size,
1649 FreeBlockDictionary<FreeChunk>::atLeast);
1650 if (fc == NULL) {
1651 return NULL;
1652 }
1653 _bt.allocated((HeapWord*)fc, fc->size());
1654 if (fc->size() >= size + MinChunkSize) {
1655 fc = splitChunkAndReturnRemainder(fc, size);
1656 }
1657 assert(fc->size() >= size, "chunk too small");
1658 assert(fc->size() < size + MinChunkSize, "chunk too big");
1659 _bt.verify_single_block((HeapWord*)fc, fc->size());
1660 return fc;
1661 }
1662
1663 FreeChunk*
1664 CompactibleFreeListSpace::getChunkFromDictionaryExact(size_t size) {
1665 assert_locked();
1666 FreeChunk* fc = _dictionary->get_chunk(size,
1667 FreeBlockDictionary<FreeChunk>::atLeast);
1668 if (fc == NULL) {
1669 return fc;
1670 }
1671 _bt.allocated((HeapWord*)fc, fc->size());
1672 if (fc->size() == size) {
1673 _bt.verify_single_block((HeapWord*)fc, size);
1674 return fc;
1675 }
1676 assert(fc->size() > size, "get_chunk() guarantee");
1677 if (fc->size() < size + MinChunkSize) {
1678 // Return the chunk to the dictionary and go get a bigger one.
1679 returnChunkToDictionary(fc);
1680 fc = _dictionary->get_chunk(size + MinChunkSize,
1681 FreeBlockDictionary<FreeChunk>::atLeast);
1682 if (fc == NULL) {
1683 return NULL;
1684 }
1685 _bt.allocated((HeapWord*)fc, fc->size());
1686 }
1687 assert(fc->size() >= size + MinChunkSize, "tautology");
1688 fc = splitChunkAndReturnRemainder(fc, size);
1689 assert(fc->size() == size, "chunk is wrong size");
1690 _bt.verify_single_block((HeapWord*)fc, size);
1691 return fc;
1692 }
1693
1694 void
1695 CompactibleFreeListSpace::returnChunkToDictionary(FreeChunk* chunk) {
1696 assert_locked();
1697
1698 size_t size = chunk->size();
1699 _bt.verify_single_block((HeapWord*)chunk, size);
1700 // adjust _unallocated_block downward, as necessary
1701 _bt.freed((HeapWord*)chunk, size);
1702 _dictionary->return_chunk(chunk);
1703 #ifndef PRODUCT
1704 if (CMSCollector::abstract_state() != CMSCollector::Sweeping) {
1705 TreeChunk<FreeChunk, AdaptiveFreeList<FreeChunk> >* tc = TreeChunk<FreeChunk, AdaptiveFreeList<FreeChunk> >::as_TreeChunk(chunk);
1706 TreeList<FreeChunk, AdaptiveFreeList<FreeChunk> >* tl = tc->list();
1707 tl->verify_stats();
1708 }
1709 #endif // PRODUCT
1710 }
1711
1712 void
1713 CompactibleFreeListSpace::returnChunkToFreeList(FreeChunk* fc) {
1714 assert_locked();
1715 size_t size = fc->size();
1716 _bt.verify_single_block((HeapWord*) fc, size);
1717 _bt.verify_not_unallocated((HeapWord*) fc, size);
1718 if (_adaptive_freelists) {
1719 _indexedFreeList[size].return_chunk_at_tail(fc);
1720 } else {
1721 _indexedFreeList[size].return_chunk_at_head(fc);
1722 }
1723 #ifndef PRODUCT
1724 if (CMSCollector::abstract_state() != CMSCollector::Sweeping) {
1725 _indexedFreeList[size].verify_stats();
1726 }
1727 #endif // PRODUCT
1728 }
1729
1730 // Add chunk to end of last block -- if it's the largest
1731 // block -- and update BOT and census data. We would
1732 // of course have preferred to coalesce it with the
1733 // last block, but it's currently less expensive to find the
1734 // largest block than it is to find the last.
1735 void
1736 CompactibleFreeListSpace::addChunkToFreeListsAtEndRecordingStats(
1737 HeapWord* chunk, size_t size) {
1738 // check that the chunk does lie in this space!
1739 assert(chunk != NULL && is_in_reserved(chunk), "Not in this space!");
1740 // One of the parallel gc task threads may be here
1741 // whilst others are allocating.
1742 Mutex* lock = NULL;
1743 if (ParallelGCThreads != 0) {
1744 lock = &_parDictionaryAllocLock;
1745 }
1746 FreeChunk* ec;
1747 {
1748 MutexLockerEx x(lock, Mutex::_no_safepoint_check_flag);
1749 ec = dictionary()->find_largest_dict(); // get largest block
1750 if (ec != NULL && ec->end() == (uintptr_t*) chunk) {
1751 // It's a coterminal block - we can coalesce.
1752 size_t old_size = ec->size();
1753 coalDeath(old_size);
1754 removeChunkFromDictionary(ec);
1755 size += old_size;
1756 } else {
1757 ec = (FreeChunk*)chunk;
1758 }
1759 }
1760 ec->set_size(size);
1761 debug_only(ec->mangleFreed(size));
1762 if (size < SmallForDictionary && ParallelGCThreads != 0) {
1763 lock = _indexedFreeListParLocks[size];
1764 }
1765 MutexLockerEx x(lock, Mutex::_no_safepoint_check_flag);
1766 addChunkAndRepairOffsetTable((HeapWord*)ec, size, true);
1767 // record the birth under the lock since the recording involves
1768 // manipulation of the list on which the chunk lives and
1769 // if the chunk is allocated and is the last on the list,
1770 // the list can go away.
1771 coalBirth(size);
1772 }
1773
1774 void
1775 CompactibleFreeListSpace::addChunkToFreeLists(HeapWord* chunk,
1776 size_t size) {
1777 // check that the chunk does lie in this space!
1778 assert(chunk != NULL && is_in_reserved(chunk), "Not in this space!");
1779 assert_locked();
1780 _bt.verify_single_block(chunk, size);
1781
1782 FreeChunk* fc = (FreeChunk*) chunk;
1783 fc->set_size(size);
1784 debug_only(fc->mangleFreed(size));
1785 if (size < SmallForDictionary) {
1786 returnChunkToFreeList(fc);
1787 } else {
1788 returnChunkToDictionary(fc);
1789 }
1790 }
1791
1792 void
1793 CompactibleFreeListSpace::addChunkAndRepairOffsetTable(HeapWord* chunk,
1794 size_t size, bool coalesced) {
1795 assert_locked();
1796 assert(chunk != NULL, "null chunk");
1797 if (coalesced) {
1798 // repair BOT
1799 _bt.single_block(chunk, size);
1800 }
1801 addChunkToFreeLists(chunk, size);
1802 }
1803
1804 // We _must_ find the purported chunk on our free lists;
1805 // we assert if we don't.
1806 void
1807 CompactibleFreeListSpace::removeFreeChunkFromFreeLists(FreeChunk* fc) {
1808 size_t size = fc->size();
1809 assert_locked();
1810 debug_only(verifyFreeLists());
1811 if (size < SmallForDictionary) {
1812 removeChunkFromIndexedFreeList(fc);
1813 } else {
1814 removeChunkFromDictionary(fc);
1815 }
1816 _bt.verify_single_block((HeapWord*)fc, size);
1817 debug_only(verifyFreeLists());
1818 }
1819
1820 void
1821 CompactibleFreeListSpace::removeChunkFromDictionary(FreeChunk* fc) {
1822 size_t size = fc->size();
1823 assert_locked();
1824 assert(fc != NULL, "null chunk");
1825 _bt.verify_single_block((HeapWord*)fc, size);
1826 _dictionary->remove_chunk(fc);
1827 // adjust _unallocated_block upward, as necessary
1828 _bt.allocated((HeapWord*)fc, size);
1829 }
1830
1831 void
1832 CompactibleFreeListSpace::removeChunkFromIndexedFreeList(FreeChunk* fc) {
1833 assert_locked();
1834 size_t size = fc->size();
1835 _bt.verify_single_block((HeapWord*)fc, size);
1836 NOT_PRODUCT(
1837 if (FLSVerifyIndexTable) {
1838 verifyIndexedFreeList(size);
1839 }
1840 )
1841 _indexedFreeList[size].remove_chunk(fc);
1842 NOT_PRODUCT(
1843 if (FLSVerifyIndexTable) {
1844 verifyIndexedFreeList(size);
1845 }
1846 )
1847 }
1848
1849 FreeChunk* CompactibleFreeListSpace::bestFitSmall(size_t numWords) {
1850 /* A hint is the next larger size that has a surplus.
1851 Start search at a size large enough to guarantee that
1852 the excess is >= MIN_CHUNK. */
1853 size_t start = align_object_size(numWords + MinChunkSize);
1854 if (start < IndexSetSize) {
1855 AdaptiveFreeList<FreeChunk>* it = _indexedFreeList;
1856 size_t hint = _indexedFreeList[start].hint();
1857 while (hint < IndexSetSize) {
1858 assert(hint % MinObjAlignment == 0, "hint should be aligned");
1859 AdaptiveFreeList<FreeChunk> *fl = &_indexedFreeList[hint];
1860 if (fl->surplus() > 0 && fl->head() != NULL) {
1861 // Found a list with surplus, reset original hint
1862 // and split out a free chunk which is returned.
1863 _indexedFreeList[start].set_hint(hint);
1864 FreeChunk* res = getFromListGreater(fl, numWords);
1865 assert(res == NULL || res->is_free(),
1866 "Should be returning a free chunk");
1867 return res;
1868 }
1869 hint = fl->hint(); /* keep looking */
1870 }
1871 /* None found. */
1872 it[start].set_hint(IndexSetSize);
1873 }
1874 return NULL;
1875 }
1876
1877 /* Requires fl->size >= numWords + MinChunkSize */
1878 FreeChunk* CompactibleFreeListSpace::getFromListGreater(AdaptiveFreeList<FreeChunk>* fl,
1879 size_t numWords) {
1880 FreeChunk *curr = fl->head();
1881 size_t oldNumWords = curr->size();
1882 assert(numWords >= MinChunkSize, "Word size is too small");
1883 assert(curr != NULL, "List is empty");
1884 assert(oldNumWords >= numWords + MinChunkSize,
1885 "Size of chunks in the list is too small");
1886
1887 fl->remove_chunk(curr);
1888 // recorded indirectly by splitChunkAndReturnRemainder -
1889 // smallSplit(oldNumWords, numWords);
1890 FreeChunk* new_chunk = splitChunkAndReturnRemainder(curr, numWords);
1891 // Does anything have to be done for the remainder in terms of
1892 // fixing the card table?
1893 assert(new_chunk == NULL || new_chunk->is_free(),
1894 "Should be returning a free chunk");
1895 return new_chunk;
1896 }
1897
1898 FreeChunk*
1899 CompactibleFreeListSpace::splitChunkAndReturnRemainder(FreeChunk* chunk,
1900 size_t new_size) {
1901 assert_locked();
1902 size_t size = chunk->size();
1903 assert(size > new_size, "Split from a smaller block?");
1904 assert(is_aligned(chunk), "alignment problem");
1905 assert(size == adjustObjectSize(size), "alignment problem");
1906 size_t rem_sz = size - new_size;
1907 assert(rem_sz == adjustObjectSize(rem_sz), "alignment problem");
1908 assert(rem_sz >= MinChunkSize, "Free chunk smaller than minimum");
1909 FreeChunk* ffc = (FreeChunk*)((HeapWord*)chunk + new_size);
1910 assert(is_aligned(ffc), "alignment problem");
1911 ffc->set_size(rem_sz);
1912 ffc->link_next(NULL);
1913 ffc->link_prev(NULL); // Mark as a free block for other (parallel) GC threads.
1914 // Above must occur before BOT is updated below.
1915 // adjust block offset table
1916 OrderAccess::storestore();
1917 assert(chunk->is_free() && ffc->is_free(), "Error");
1918 _bt.split_block((HeapWord*)chunk, chunk->size(), new_size);
1919 if (rem_sz < SmallForDictionary) {
1920 bool is_par = (SharedHeap::heap()->n_par_threads() > 0);
1921 if (is_par) _indexedFreeListParLocks[rem_sz]->lock();
1922 assert(!is_par ||
1923 (SharedHeap::heap()->n_par_threads() ==
1924 SharedHeap::heap()->workers()->active_workers()), "Mismatch");
1925 returnChunkToFreeList(ffc);
1926 split(size, rem_sz);
1927 if (is_par) _indexedFreeListParLocks[rem_sz]->unlock();
1928 } else {
1929 returnChunkToDictionary(ffc);
1930 split(size, rem_sz);
1931 }
1932 chunk->set_size(new_size);
1933 return chunk;
1934 }
1935
1936 void
1937 CompactibleFreeListSpace::sweep_completed() {
1938 // Now that space is probably plentiful, refill linear
1939 // allocation blocks as needed.
1940 refillLinearAllocBlocksIfNeeded();
1941 }
1942
1943 void
1944 CompactibleFreeListSpace::gc_prologue() {
1945 assert_locked();
1946 if (PrintFLSStatistics != 0) {
1947 gclog_or_tty->print("Before GC:\n");
1948 reportFreeListStatistics();
1949 }
1950 refillLinearAllocBlocksIfNeeded();
1951 }
1952
1953 void
1954 CompactibleFreeListSpace::gc_epilogue() {
1955 assert_locked();
1956 if (PrintGCDetails && Verbose && !_adaptive_freelists) {
1957 if (_smallLinearAllocBlock._word_size == 0)
1958 warning("CompactibleFreeListSpace(epilogue):: Linear allocation failure");
1959 }
1960 assert(_promoInfo.noPromotions(), "_promoInfo inconsistency");
1961 _promoInfo.stopTrackingPromotions();
1962 repairLinearAllocationBlocks();
1963 // Print Space's stats
1964 if (PrintFLSStatistics != 0) {
1965 gclog_or_tty->print("After GC:\n");
1966 reportFreeListStatistics();
1967 }
1968 }
1969
1970 // Iteration support, mostly delegated from a CMS generation
1971
1972 void CompactibleFreeListSpace::save_marks() {
1973 assert(Thread::current()->is_VM_thread(),
1974 "Global variable should only be set when single-threaded");
1975 // Mark the "end" of the used space at the time of this call;
1976 // note, however, that promoted objects from this point
1977 // on are tracked in the _promoInfo below.
1978 set_saved_mark_word(unallocated_block());
1979 #ifdef ASSERT
1980 // Check the sanity of save_marks() etc.
1981 MemRegion ur = used_region();
1982 MemRegion urasm = used_region_at_save_marks();
1983 assert(ur.contains(urasm),
1984 err_msg(" Error at save_marks(): [" PTR_FORMAT "," PTR_FORMAT ")"
1985 " should contain [" PTR_FORMAT "," PTR_FORMAT ")",
1986 p2i(ur.start()), p2i(ur.end()), p2i(urasm.start()), p2i(urasm.end())));
1987 #endif
1988 // inform allocator that promotions should be tracked.
1989 assert(_promoInfo.noPromotions(), "_promoInfo inconsistency");
1990 _promoInfo.startTrackingPromotions();
1991 }
1992
1993 bool CompactibleFreeListSpace::no_allocs_since_save_marks() {
1994 assert(_promoInfo.tracking(), "No preceding save_marks?");
1995 assert(SharedHeap::heap()->n_par_threads() == 0,
1996 "Shouldn't be called if using parallel gc.");
1997 return _promoInfo.noPromotions();
1998 }
1999
2000 #define CFLS_OOP_SINCE_SAVE_MARKS_DEFN(OopClosureType, nv_suffix) \
2001 \
2002 void CompactibleFreeListSpace:: \
2003 oop_since_save_marks_iterate##nv_suffix(OopClosureType* blk) { \
2004 assert(SharedHeap::heap()->n_par_threads() == 0, \
2005 "Shouldn't be called (yet) during parallel part of gc."); \
2006 _promoInfo.promoted_oops_iterate##nv_suffix(blk); \
2007 /* \
2008 * This also restores any displaced headers and removes the elements from \
2009 * the iteration set as they are processed, so that we have a clean slate \
2010 * at the end of the iteration. Note, thus, that if new objects are \
2011 * promoted as a result of the iteration they are iterated over as well. \
2012 */ \
2013 assert(_promoInfo.noPromotions(), "_promoInfo inconsistency"); \
2014 }
2015
2016 ALL_SINCE_SAVE_MARKS_CLOSURES(CFLS_OOP_SINCE_SAVE_MARKS_DEFN)
2017
2018 bool CompactibleFreeListSpace::linearAllocationWouldFail() const {
2019 return _smallLinearAllocBlock._word_size == 0;
2020 }
2021
2022 void CompactibleFreeListSpace::repairLinearAllocationBlocks() {
2023 // Fix up linear allocation blocks to look like free blocks
2024 repairLinearAllocBlock(&_smallLinearAllocBlock);
2025 }
2026
2027 void CompactibleFreeListSpace::repairLinearAllocBlock(LinearAllocBlock* blk) {
2028 assert_locked();
2029 if (blk->_ptr != NULL) {
2030 assert(blk->_word_size != 0 && blk->_word_size >= MinChunkSize,
2031 "Minimum block size requirement");
2032 FreeChunk* fc = (FreeChunk*)(blk->_ptr);
2033 fc->set_size(blk->_word_size);
2034 fc->link_prev(NULL); // mark as free
2035 fc->dontCoalesce();
2036 assert(fc->is_free(), "just marked it free");
2037 assert(fc->cantCoalesce(), "just marked it uncoalescable");
2038 }
2039 }
2040
2041 void CompactibleFreeListSpace::refillLinearAllocBlocksIfNeeded() {
2042 assert_locked();
2043 if (_smallLinearAllocBlock._ptr == NULL) {
2044 assert(_smallLinearAllocBlock._word_size == 0,
2045 "Size of linAB should be zero if the ptr is NULL");
2046 // Reset the linAB refill and allocation size limit.
2047 _smallLinearAllocBlock.set(0, 0, 1024*SmallForLinearAlloc, SmallForLinearAlloc);
2048 }
2049 refillLinearAllocBlockIfNeeded(&_smallLinearAllocBlock);
2050 }
2051
2052 void
2053 CompactibleFreeListSpace::refillLinearAllocBlockIfNeeded(LinearAllocBlock* blk) {
2054 assert_locked();
2055 assert((blk->_ptr == NULL && blk->_word_size == 0) ||
2056 (blk->_ptr != NULL && blk->_word_size >= MinChunkSize),
2057 "blk invariant");
2058 if (blk->_ptr == NULL) {
2059 refillLinearAllocBlock(blk);
2060 }
2061 if (PrintMiscellaneous && Verbose) {
2062 if (blk->_word_size == 0) {
2063 warning("CompactibleFreeListSpace(prologue):: Linear allocation failure");
2064 }
2065 }
2066 }
2067
2068 void
2069 CompactibleFreeListSpace::refillLinearAllocBlock(LinearAllocBlock* blk) {
2070 assert_locked();
2071 assert(blk->_word_size == 0 && blk->_ptr == NULL,
2072 "linear allocation block should be empty");
2073 FreeChunk* fc;
2074 if (blk->_refillSize < SmallForDictionary &&
2075 (fc = getChunkFromIndexedFreeList(blk->_refillSize)) != NULL) {
2076 // A linAB's strategy might be to use small sizes to reduce
2077 // fragmentation but still get the benefits of allocation from a
2078 // linAB.
2079 } else {
2080 fc = getChunkFromDictionary(blk->_refillSize);
2081 }
2082 if (fc != NULL) {
2083 blk->_ptr = (HeapWord*)fc;
2084 blk->_word_size = fc->size();
2085 fc->dontCoalesce(); // to prevent sweeper from sweeping us up
2086 }
2087 }
2088
2089 // Support for concurrent collection policy decisions.
2090 bool CompactibleFreeListSpace::should_concurrent_collect() const {
2091 // In the future we might want to add in fragmentation stats --
2092 // including erosion of the "mountain" into this decision as well.
2093 return !adaptive_freelists() && linearAllocationWouldFail();
2094 }
2095
2096 // Support for compaction
2097
2098 void CompactibleFreeListSpace::prepare_for_compaction(CompactPoint* cp) {
2099 SCAN_AND_FORWARD(cp,end,block_is_obj,block_size);
2100 // Prepare_for_compaction() uses the space between live objects
2101 // so that later phase can skip dead space quickly. So verification
2102 // of the free lists doesn't work after.
2103 }
2104
2105 #define obj_size(q) adjustObjectSize(oop(q)->size())
2106 #define adjust_obj_size(s) adjustObjectSize(s)
2107
2108 void CompactibleFreeListSpace::adjust_pointers() {
2109 // In other versions of adjust_pointers(), a bail out
2110 // based on the amount of live data in the generation
2111 // (i.e., if 0, bail out) may be used.
2112 // Cannot test used() == 0 here because the free lists have already
2113 // been mangled by the compaction.
2114
2115 SCAN_AND_ADJUST_POINTERS(adjust_obj_size);
2116 // See note about verification in prepare_for_compaction().
2117 }
2118
2119 void CompactibleFreeListSpace::compact() {
2120 SCAN_AND_COMPACT(obj_size);
2121 }
2122
2123 // Fragmentation metric = 1 - [sum of (fbs**2) / (sum of fbs)**2]
2124 // where fbs is free block sizes
2125 double CompactibleFreeListSpace::flsFrag() const {
2126 size_t itabFree = totalSizeInIndexedFreeLists();
2127 double frag = 0.0;
2128 size_t i;
2129
2130 for (i = IndexSetStart; i < IndexSetSize; i += IndexSetStride) {
2131 double sz = i;
2132 frag += _indexedFreeList[i].count() * (sz * sz);
2133 }
2134
2135 double totFree = itabFree +
2136 _dictionary->total_chunk_size(DEBUG_ONLY(freelistLock()));
2137 if (totFree > 0) {
2138 frag = ((frag + _dictionary->sum_of_squared_block_sizes()) /
2139 (totFree * totFree));
2140 frag = (double)1.0 - frag;
2141 } else {
2142 assert(frag == 0.0, "Follows from totFree == 0");
2143 }
2144 return frag;
2145 }
2146
2147 void CompactibleFreeListSpace::beginSweepFLCensus(
2148 float inter_sweep_current,
2149 float inter_sweep_estimate,
2150 float intra_sweep_estimate) {
2151 assert_locked();
2152 size_t i;
2153 for (i = IndexSetStart; i < IndexSetSize; i += IndexSetStride) {
2154 AdaptiveFreeList<FreeChunk>* fl = &_indexedFreeList[i];
2155 if (PrintFLSStatistics > 1) {
2156 gclog_or_tty->print("size[" SIZE_FORMAT "] : ", i);
2157 }
2158 fl->compute_desired(inter_sweep_current, inter_sweep_estimate, intra_sweep_estimate);
2159 fl->set_coal_desired((ssize_t)((double)fl->desired() * CMSSmallCoalSurplusPercent));
2160 fl->set_before_sweep(fl->count());
2161 fl->set_bfr_surp(fl->surplus());
2162 }
2163 _dictionary->begin_sweep_dict_census(CMSLargeCoalSurplusPercent,
2164 inter_sweep_current,
2165 inter_sweep_estimate,
2166 intra_sweep_estimate);
2167 }
2168
2169 void CompactibleFreeListSpace::setFLSurplus() {
2170 assert_locked();
2171 size_t i;
2172 for (i = IndexSetStart; i < IndexSetSize; i += IndexSetStride) {
2173 AdaptiveFreeList<FreeChunk> *fl = &_indexedFreeList[i];
2174 fl->set_surplus(fl->count() -
2175 (ssize_t)((double)fl->desired() * CMSSmallSplitSurplusPercent));
2176 }
2177 }
2178
2179 void CompactibleFreeListSpace::setFLHints() {
2180 assert_locked();
2181 size_t i;
2182 size_t h = IndexSetSize;
2183 for (i = IndexSetSize - 1; i != 0; i -= IndexSetStride) {
2184 AdaptiveFreeList<FreeChunk> *fl = &_indexedFreeList[i];
2185 fl->set_hint(h);
2186 if (fl->surplus() > 0) {
2187 h = i;
2188 }
2189 }
2190 }
2191
2192 void CompactibleFreeListSpace::clearFLCensus() {
2193 assert_locked();
2194 size_t i;
2195 for (i = IndexSetStart; i < IndexSetSize; i += IndexSetStride) {
2196 AdaptiveFreeList<FreeChunk> *fl = &_indexedFreeList[i];
2197 fl->set_prev_sweep(fl->count());
2198 fl->set_coal_births(0);
2199 fl->set_coal_deaths(0);
2200 fl->set_split_births(0);
2201 fl->set_split_deaths(0);
2202 }
2203 }
2204
2205 void CompactibleFreeListSpace::endSweepFLCensus(size_t sweep_count) {
2206 if (PrintFLSStatistics > 0) {
2207 HeapWord* largestAddr = (HeapWord*) dictionary()->find_largest_dict();
2208 gclog_or_tty->print_cr("CMS: Large block " PTR_FORMAT,
2209 p2i(largestAddr));
2210 }
2211 setFLSurplus();
2212 setFLHints();
2213 if (PrintGC && PrintFLSCensus > 0) {
2214 printFLCensus(sweep_count);
2215 }
2216 clearFLCensus();
2217 assert_locked();
2218 _dictionary->end_sweep_dict_census(CMSLargeSplitSurplusPercent);
2219 }
2220
2221 bool CompactibleFreeListSpace::coalOverPopulated(size_t size) {
2222 if (size < SmallForDictionary) {
2223 AdaptiveFreeList<FreeChunk> *fl = &_indexedFreeList[size];
2224 return (fl->coal_desired() < 0) ||
2225 ((int)fl->count() > fl->coal_desired());
2226 } else {
2227 return dictionary()->coal_dict_over_populated(size);
2228 }
2229 }
2230
2231 void CompactibleFreeListSpace::smallCoalBirth(size_t size) {
2232 assert(size < SmallForDictionary, "Size too large for indexed list");
2233 AdaptiveFreeList<FreeChunk> *fl = &_indexedFreeList[size];
2234 fl->increment_coal_births();
2235 fl->increment_surplus();
2236 }
2237
2238 void CompactibleFreeListSpace::smallCoalDeath(size_t size) {
2239 assert(size < SmallForDictionary, "Size too large for indexed list");
2240 AdaptiveFreeList<FreeChunk> *fl = &_indexedFreeList[size];
2241 fl->increment_coal_deaths();
2242 fl->decrement_surplus();
2243 }
2244
2245 void CompactibleFreeListSpace::coalBirth(size_t size) {
2246 if (size < SmallForDictionary) {
2247 smallCoalBirth(size);
2248 } else {
2249 dictionary()->dict_census_update(size,
2250 false /* split */,
2251 true /* birth */);
2252 }
2253 }
2254
2255 void CompactibleFreeListSpace::coalDeath(size_t size) {
2256 if(size < SmallForDictionary) {
2257 smallCoalDeath(size);
2258 } else {
2259 dictionary()->dict_census_update(size,
2260 false /* split */,
2261 false /* birth */);
2262 }
2263 }
2264
2265 void CompactibleFreeListSpace::smallSplitBirth(size_t size) {
2266 assert(size < SmallForDictionary, "Size too large for indexed list");
2267 AdaptiveFreeList<FreeChunk> *fl = &_indexedFreeList[size];
2268 fl->increment_split_births();
2269 fl->increment_surplus();
2270 }
2271
2272 void CompactibleFreeListSpace::smallSplitDeath(size_t size) {
2273 assert(size < SmallForDictionary, "Size too large for indexed list");
2274 AdaptiveFreeList<FreeChunk> *fl = &_indexedFreeList[size];
2275 fl->increment_split_deaths();
2276 fl->decrement_surplus();
2277 }
2278
2279 void CompactibleFreeListSpace::split_birth(size_t size) {
2280 if (size < SmallForDictionary) {
2281 smallSplitBirth(size);
2282 } else {
2283 dictionary()->dict_census_update(size,
2284 true /* split */,
2285 true /* birth */);
2286 }
2287 }
2288
2289 void CompactibleFreeListSpace::splitDeath(size_t size) {
2290 if (size < SmallForDictionary) {
2291 smallSplitDeath(size);
2292 } else {
2293 dictionary()->dict_census_update(size,
2294 true /* split */,
2295 false /* birth */);
2296 }
2297 }
2298
2299 void CompactibleFreeListSpace::split(size_t from, size_t to1) {
2300 size_t to2 = from - to1;
2301 splitDeath(from);
2302 split_birth(to1);
2303 split_birth(to2);
2304 }
2305
2306 void CompactibleFreeListSpace::print() const {
2307 print_on(tty);
2308 }
2309
2310 void CompactibleFreeListSpace::prepare_for_verify() {
2311 assert_locked();
2312 repairLinearAllocationBlocks();
2313 // Verify that the SpoolBlocks look like free blocks of
2314 // appropriate sizes... To be done ...
2315 }
2316
2317 class VerifyAllBlksClosure: public BlkClosure {
2318 private:
2319 const CompactibleFreeListSpace* _sp;
2320 const MemRegion _span;
2321 HeapWord* _last_addr;
2322 size_t _last_size;
2323 bool _last_was_obj;
2324 bool _last_was_live;
2325
2326 public:
2327 VerifyAllBlksClosure(const CompactibleFreeListSpace* sp,
2328 MemRegion span) : _sp(sp), _span(span),
2329 _last_addr(NULL), _last_size(0),
2330 _last_was_obj(false), _last_was_live(false) { }
2331
2332 virtual size_t do_blk(HeapWord* addr) {
2333 size_t res;
2334 bool was_obj = false;
2335 bool was_live = false;
2336 if (_sp->block_is_obj(addr)) {
2337 was_obj = true;
2338 oop p = oop(addr);
2339 guarantee(p->is_oop(), "Should be an oop");
2340 res = _sp->adjustObjectSize(p->size());
2341 if (_sp->obj_is_alive(addr)) {
2342 was_live = true;
2343 p->verify();
2344 }
2345 } else {
2346 FreeChunk* fc = (FreeChunk*)addr;
2347 res = fc->size();
2348 if (FLSVerifyLists && !fc->cantCoalesce()) {
2349 guarantee(_sp->verify_chunk_in_free_list(fc),
2350 "Chunk should be on a free list");
2351 }
2352 }
2353 if (res == 0) {
2354 gclog_or_tty->print_cr("Livelock: no rank reduction!");
2355 gclog_or_tty->print_cr(
2356 " Current: addr = " PTR_FORMAT ", size = " SIZE_FORMAT ", obj = %s, live = %s \n"
2357 " Previous: addr = " PTR_FORMAT ", size = " SIZE_FORMAT ", obj = %s, live = %s \n",
2358 p2i(addr), res, was_obj ?"true":"false", was_live ?"true":"false",
2359 p2i(_last_addr), _last_size, _last_was_obj?"true":"false", _last_was_live?"true":"false");
2360 _sp->print_on(gclog_or_tty);
2361 guarantee(false, "Seppuku!");
2362 }
2363 _last_addr = addr;
2364 _last_size = res;
2365 _last_was_obj = was_obj;
2366 _last_was_live = was_live;
2367 return res;
2368 }
2369 };
2370
2371 class VerifyAllOopsClosure: public OopClosure {
2372 private:
2373 const CMSCollector* _collector;
2374 const CompactibleFreeListSpace* _sp;
2375 const MemRegion _span;
2376 const bool _past_remark;
2377 const CMSBitMap* _bit_map;
2378
2379 protected:
2380 void do_oop(void* p, oop obj) {
2381 if (_span.contains(obj)) { // the interior oop points into CMS heap
2382 if (!_span.contains(p)) { // reference from outside CMS heap
2383 // Should be a valid object; the first disjunct below allows
2384 // us to sidestep an assertion in block_is_obj() that insists
2385 // that p be in _sp. Note that several generations (and spaces)
2386 // are spanned by _span (CMS heap) above.
2387 guarantee(!_sp->is_in_reserved(obj) ||
2388 _sp->block_is_obj((HeapWord*)obj),
2389 "Should be an object");
2390 guarantee(obj->is_oop(), "Should be an oop");
2391 obj->verify();
2392 if (_past_remark) {
2393 // Remark has been completed, the object should be marked
2394 _bit_map->isMarked((HeapWord*)obj);
2395 }
2396 } else { // reference within CMS heap
2397 if (_past_remark) {
2398 // Remark has been completed -- so the referent should have
2399 // been marked, if referring object is.
2400 if (_bit_map->isMarked(_collector->block_start(p))) {
2401 guarantee(_bit_map->isMarked((HeapWord*)obj), "Marking error?");
2402 }
2403 }
2404 }
2405 } else if (_sp->is_in_reserved(p)) {
2406 // the reference is from FLS, and points out of FLS
2407 guarantee(obj->is_oop(), "Should be an oop");
2408 obj->verify();
2409 }
2410 }
2411
2412 template <class T> void do_oop_work(T* p) {
2413 T heap_oop = oopDesc::load_heap_oop(p);
2414 if (!oopDesc::is_null(heap_oop)) {
2415 oop obj = oopDesc::decode_heap_oop_not_null(heap_oop);
2416 do_oop(p, obj);
2417 }
2418 }
2419
2420 public:
2421 VerifyAllOopsClosure(const CMSCollector* collector,
2422 const CompactibleFreeListSpace* sp, MemRegion span,
2423 bool past_remark, CMSBitMap* bit_map) :
2424 _collector(collector), _sp(sp), _span(span),
2425 _past_remark(past_remark), _bit_map(bit_map) { }
2426
2427 virtual void do_oop(oop* p) { VerifyAllOopsClosure::do_oop_work(p); }
2428 virtual void do_oop(narrowOop* p) { VerifyAllOopsClosure::do_oop_work(p); }
2429 };
2430
2431 void CompactibleFreeListSpace::verify() const {
2432 assert_lock_strong(&_freelistLock);
2433 verify_objects_initialized();
2434 MemRegion span = _collector->_span;
2435 bool past_remark = (_collector->abstract_state() ==
2436 CMSCollector::Sweeping);
2437
2438 ResourceMark rm;
2439 HandleMark hm;
2440
2441 // Check integrity of CFL data structures
2442 _promoInfo.verify();
2443 _dictionary->verify();
2444 if (FLSVerifyIndexTable) {
2445 verifyIndexedFreeLists();
2446 }
2447 // Check integrity of all objects and free blocks in space
2448 {
2449 VerifyAllBlksClosure cl(this, span);
2450 ((CompactibleFreeListSpace*)this)->blk_iterate(&cl); // cast off const
2451 }
2452 // Check that all references in the heap to FLS
2453 // are to valid objects in FLS or that references in
2454 // FLS are to valid objects elsewhere in the heap
2455 if (FLSVerifyAllHeapReferences)
2456 {
2457 VerifyAllOopsClosure cl(_collector, this, span, past_remark,
2458 _collector->markBitMap());
2459 CollectedHeap* ch = Universe::heap();
2460
2461 // Iterate over all oops in the heap. Uses the _no_header version
2462 // since we are not interested in following the klass pointers.
2463 ch->oop_iterate_no_header(&cl);
2464 }
2465
2466 if (VerifyObjectStartArray) {
2467 // Verify the block offset table
2468 _bt.verify();
2469 }
2470 }
2471
2472 #ifndef PRODUCT
2473 void CompactibleFreeListSpace::verifyFreeLists() const {
2474 if (FLSVerifyLists) {
2475 _dictionary->verify();
2476 verifyIndexedFreeLists();
2477 } else {
2478 if (FLSVerifyDictionary) {
2479 _dictionary->verify();
2480 }
2481 if (FLSVerifyIndexTable) {
2482 verifyIndexedFreeLists();
2483 }
2484 }
2485 }
2486 #endif
2487
2488 void CompactibleFreeListSpace::verifyIndexedFreeLists() const {
2489 size_t i = 0;
2490 for (; i < IndexSetStart; i++) {
2491 guarantee(_indexedFreeList[i].head() == NULL, "should be NULL");
2492 }
2493 for (; i < IndexSetSize; i++) {
2494 verifyIndexedFreeList(i);
2495 }
2496 }
2497
2498 void CompactibleFreeListSpace::verifyIndexedFreeList(size_t size) const {
2499 FreeChunk* fc = _indexedFreeList[size].head();
2500 FreeChunk* tail = _indexedFreeList[size].tail();
2501 size_t num = _indexedFreeList[size].count();
2502 size_t n = 0;
2503 guarantee(((size >= IndexSetStart) && (size % IndexSetStride == 0)) || fc == NULL,
2504 "Slot should have been empty");
2505 for (; fc != NULL; fc = fc->next(), n++) {
2506 guarantee(fc->size() == size, "Size inconsistency");
2507 guarantee(fc->is_free(), "!free?");
2508 guarantee(fc->next() == NULL || fc->next()->prev() == fc, "Broken list");
2509 guarantee((fc->next() == NULL) == (fc == tail), "Incorrect tail");
2510 }
2511 guarantee(n == num, "Incorrect count");
2512 }
2513
2514 #ifndef PRODUCT
2515 void CompactibleFreeListSpace::check_free_list_consistency() const {
2516 assert((TreeChunk<FreeChunk, AdaptiveFreeList<FreeChunk> >::min_size() <= IndexSetSize),
2517 "Some sizes can't be allocated without recourse to"
2518 " linear allocation buffers");
2519 assert((TreeChunk<FreeChunk, AdaptiveFreeList<FreeChunk> >::min_size()*HeapWordSize == sizeof(TreeChunk<FreeChunk, AdaptiveFreeList<FreeChunk> >)),
2520 "else MIN_TREE_CHUNK_SIZE is wrong");
2521 assert(IndexSetStart != 0, "IndexSetStart not initialized");
2522 assert(IndexSetStride != 0, "IndexSetStride not initialized");
2523 }
2524 #endif
2525
2526 void CompactibleFreeListSpace::printFLCensus(size_t sweep_count) const {
2527 assert_lock_strong(&_freelistLock);
2528 AdaptiveFreeList<FreeChunk> total;
2529 gclog_or_tty->print("end sweep# " SIZE_FORMAT "\n", sweep_count);
2530 AdaptiveFreeList<FreeChunk>::print_labels_on(gclog_or_tty, "size");
2531 size_t total_free = 0;
2532 for (size_t i = IndexSetStart; i < IndexSetSize; i += IndexSetStride) {
2533 const AdaptiveFreeList<FreeChunk> *fl = &_indexedFreeList[i];
2534 total_free += fl->count() * fl->size();
2535 if (i % (40*IndexSetStride) == 0) {
2536 AdaptiveFreeList<FreeChunk>::print_labels_on(gclog_or_tty, "size");
2537 }
2538 fl->print_on(gclog_or_tty);
2539 total.set_bfr_surp( total.bfr_surp() + fl->bfr_surp() );
2540 total.set_surplus( total.surplus() + fl->surplus() );
2541 total.set_desired( total.desired() + fl->desired() );
2542 total.set_prev_sweep( total.prev_sweep() + fl->prev_sweep() );
2543 total.set_before_sweep(total.before_sweep() + fl->before_sweep());
2544 total.set_count( total.count() + fl->count() );
2545 total.set_coal_births( total.coal_births() + fl->coal_births() );
2546 total.set_coal_deaths( total.coal_deaths() + fl->coal_deaths() );
2547 total.set_split_births(total.split_births() + fl->split_births());
2548 total.set_split_deaths(total.split_deaths() + fl->split_deaths());
2549 }
2550 total.print_on(gclog_or_tty, "TOTAL");
2551 gclog_or_tty->print_cr("Total free in indexed lists "
2552 SIZE_FORMAT " words", total_free);
2553 gclog_or_tty->print("growth: %8.5f deficit: %8.5f\n",
2554 (double)(total.split_births()+total.coal_births()-total.split_deaths()-total.coal_deaths())/
2555 (total.prev_sweep() != 0 ? (double)total.prev_sweep() : 1.0),
2556 (double)(total.desired() - total.count())/(total.desired() != 0 ? (double)total.desired() : 1.0));
2557 _dictionary->print_dict_census();
2558 }
2559
2560 ///////////////////////////////////////////////////////////////////////////
2561 // CFLS_LAB
2562 ///////////////////////////////////////////////////////////////////////////
2563
2564 #define VECTOR_257(x) \
2565 /* 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 */ \
2566 { 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, \
2567 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, \
2568 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, \
2569 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, \
2570 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, \
2571 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, \
2572 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, \
2573 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, \
2574 x }
2575
2576 // Initialize with default setting of CMSParPromoteBlocksToClaim, _not_
2577 // OldPLABSize, whose static default is different; if overridden at the
2578 // command-line, this will get reinitialized via a call to
2579 // modify_initialization() below.
2580 AdaptiveWeightedAverage CFLS_LAB::_blocks_to_claim[] =
2581 VECTOR_257(AdaptiveWeightedAverage(OldPLABWeight, (float)CMSParPromoteBlocksToClaim));
2582 size_t CFLS_LAB::_global_num_blocks[] = VECTOR_257(0);
2583 uint CFLS_LAB::_global_num_workers[] = VECTOR_257(0);
2584
2585 CFLS_LAB::CFLS_LAB(CompactibleFreeListSpace* cfls) :
2586 _cfls(cfls)
2587 {
2588 assert(CompactibleFreeListSpace::IndexSetSize == 257, "Modify VECTOR_257() macro above");
2589 for (size_t i = CompactibleFreeListSpace::IndexSetStart;
2590 i < CompactibleFreeListSpace::IndexSetSize;
2591 i += CompactibleFreeListSpace::IndexSetStride) {
2592 _indexedFreeList[i].set_size(i);
2593 _num_blocks[i] = 0;
2594 }
2595 }
2596
2597 static bool _CFLS_LAB_modified = false;
2598
2599 void CFLS_LAB::modify_initialization(size_t n, unsigned wt) {
2600 assert(!_CFLS_LAB_modified, "Call only once");
2601 _CFLS_LAB_modified = true;
2602 for (size_t i = CompactibleFreeListSpace::IndexSetStart;
2603 i < CompactibleFreeListSpace::IndexSetSize;
2604 i += CompactibleFreeListSpace::IndexSetStride) {
2605 _blocks_to_claim[i].modify(n, wt, true /* force */);
2606 }
2607 }
2608
2609 HeapWord* CFLS_LAB::alloc(size_t word_sz) {
2610 FreeChunk* res;
2611 assert(word_sz == _cfls->adjustObjectSize(word_sz), "Error");
2612 if (word_sz >= CompactibleFreeListSpace::IndexSetSize) {
2613 // This locking manages sync with other large object allocations.
2614 MutexLockerEx x(_cfls->parDictionaryAllocLock(),
2615 Mutex::_no_safepoint_check_flag);
2616 res = _cfls->getChunkFromDictionaryExact(word_sz);
2617 if (res == NULL) return NULL;
2618 } else {
2619 AdaptiveFreeList<FreeChunk>* fl = &_indexedFreeList[word_sz];
2620 if (fl->count() == 0) {
2621 // Attempt to refill this local free list.
2622 get_from_global_pool(word_sz, fl);
2623 // If it didn't work, give up.
2624 if (fl->count() == 0) return NULL;
2625 }
2626 res = fl->get_chunk_at_head();
2627 assert(res != NULL, "Why was count non-zero?");
2628 }
2629 res->markNotFree();
2630 assert(!res->is_free(), "shouldn't be marked free");
2631 assert(oop(res)->klass_or_null() == NULL, "should look uninitialized");
2632 // mangle a just allocated object with a distinct pattern.
2633 debug_only(res->mangleAllocated(word_sz));
2634 return (HeapWord*)res;
2635 }
2636
2637 // Get a chunk of blocks of the right size and update related
2638 // book-keeping stats
2639 void CFLS_LAB::get_from_global_pool(size_t word_sz, AdaptiveFreeList<FreeChunk>* fl) {
2640 // Get the #blocks we want to claim
2641 size_t n_blks = (size_t)_blocks_to_claim[word_sz].average();
2642 assert(n_blks > 0, "Error");
2643 assert(ResizePLAB || n_blks == OldPLABSize, "Error");
2644 // In some cases, when the application has a phase change,
2645 // there may be a sudden and sharp shift in the object survival
2646 // profile, and updating the counts at the end of a scavenge
2647 // may not be quick enough, giving rise to large scavenge pauses
2648 // during these phase changes. It is beneficial to detect such
2649 // changes on-the-fly during a scavenge and avoid such a phase-change
2650 // pothole. The following code is a heuristic attempt to do that.
2651 // It is protected by a product flag until we have gained
2652 // enough experience with this heuristic and fine-tuned its behavior.
2653 // WARNING: This might increase fragmentation if we overreact to
2654 // small spikes, so some kind of historical smoothing based on
2655 // previous experience with the greater reactivity might be useful.
2656 // Lacking sufficient experience, CMSOldPLABResizeQuicker is disabled by
2657 // default.
2658 if (ResizeOldPLAB && CMSOldPLABResizeQuicker) {
2659 size_t multiple = _num_blocks[word_sz]/(CMSOldPLABToleranceFactor*CMSOldPLABNumRefills*n_blks);
2660 n_blks += CMSOldPLABReactivityFactor*multiple*n_blks;
2661 n_blks = MIN2(n_blks, CMSOldPLABMax);
2662 }
2663 assert(n_blks > 0, "Error");
2664 _cfls->par_get_chunk_of_blocks(word_sz, n_blks, fl);
2665 // Update stats table entry for this block size
2666 _num_blocks[word_sz] += fl->count();
2667 }
2668
2669 void CFLS_LAB::compute_desired_plab_size() {
2670 for (size_t i = CompactibleFreeListSpace::IndexSetStart;
2671 i < CompactibleFreeListSpace::IndexSetSize;
2672 i += CompactibleFreeListSpace::IndexSetStride) {
2673 assert((_global_num_workers[i] == 0) == (_global_num_blocks[i] == 0),
2674 "Counter inconsistency");
2675 if (_global_num_workers[i] > 0) {
2676 // Need to smooth wrt historical average
2677 if (ResizeOldPLAB) {
2678 _blocks_to_claim[i].sample(
2679 MAX2((size_t)CMSOldPLABMin,
2680 MIN2((size_t)CMSOldPLABMax,
2681 _global_num_blocks[i]/(_global_num_workers[i]*CMSOldPLABNumRefills))));
2682 }
2683 // Reset counters for next round
2684 _global_num_workers[i] = 0;
2685 _global_num_blocks[i] = 0;
2686 if (PrintOldPLAB) {
2687 gclog_or_tty->print_cr("[" SIZE_FORMAT "]: " SIZE_FORMAT,
2688 i, (size_t)_blocks_to_claim[i].average());
2689 }
2690 }
2691 }
2692 }
2693
2694 // If this is changed in the future to allow parallel
2695 // access, one would need to take the FL locks and,
2696 // depending on how it is used, stagger access from
2697 // parallel threads to reduce contention.
2698 void CFLS_LAB::retire(int tid) {
2699 // We run this single threaded with the world stopped;
2700 // so no need for locks and such.
2701 NOT_PRODUCT(Thread* t = Thread::current();)
2702 assert(Thread::current()->is_VM_thread(), "Error");
2703 for (size_t i = CompactibleFreeListSpace::IndexSetStart;
2704 i < CompactibleFreeListSpace::IndexSetSize;
2705 i += CompactibleFreeListSpace::IndexSetStride) {
2706 assert(_num_blocks[i] >= (size_t)_indexedFreeList[i].count(),
2707 "Can't retire more than what we obtained");
2708 if (_num_blocks[i] > 0) {
2709 size_t num_retire = _indexedFreeList[i].count();
2710 assert(_num_blocks[i] > num_retire, "Should have used at least one");
2711 {
2712 // MutexLockerEx x(_cfls->_indexedFreeListParLocks[i],
2713 // Mutex::_no_safepoint_check_flag);
2714
2715 // Update globals stats for num_blocks used
2716 _global_num_blocks[i] += (_num_blocks[i] - num_retire);
2717 _global_num_workers[i]++;
2718 assert(_global_num_workers[i] <= ParallelGCThreads, "Too big");
2719 if (num_retire > 0) {
2720 _cfls->_indexedFreeList[i].prepend(&_indexedFreeList[i]);
2721 // Reset this list.
2722 _indexedFreeList[i] = AdaptiveFreeList<FreeChunk>();
2723 _indexedFreeList[i].set_size(i);
2724 }
2725 }
2726 if (PrintOldPLAB) {
2727 gclog_or_tty->print_cr("%d[" SIZE_FORMAT "]: " SIZE_FORMAT "/" SIZE_FORMAT "/" SIZE_FORMAT,
2728 tid, i, num_retire, _num_blocks[i], (size_t)_blocks_to_claim[i].average());
2729 }
2730 // Reset stats for next round
2731 _num_blocks[i] = 0;
2732 }
2733 }
2734 }
2735
2736 // Used by par_get_chunk_of_blocks() for the chunks from the
2737 // indexed_free_lists. Looks for a chunk with size that is a multiple
2738 // of "word_sz" and if found, splits it into "word_sz" chunks and add
2739 // to the free list "fl". "n" is the maximum number of chunks to
2740 // be added to "fl".
2741 bool CompactibleFreeListSpace:: par_get_chunk_of_blocks_IFL(size_t word_sz, size_t n, AdaptiveFreeList<FreeChunk>* fl) {
2742
2743 // We'll try all multiples of word_sz in the indexed set, starting with
2744 // word_sz itself and, if CMSSplitIndexedFreeListBlocks, try larger multiples,
2745 // then try getting a big chunk and splitting it.
2746 {
2747 bool found;
2748 int k;
2749 size_t cur_sz;
2750 for (k = 1, cur_sz = k * word_sz, found = false;
2751 (cur_sz < CompactibleFreeListSpace::IndexSetSize) &&
2752 (CMSSplitIndexedFreeListBlocks || k <= 1);
2753 k++, cur_sz = k * word_sz) {
2754 AdaptiveFreeList<FreeChunk> fl_for_cur_sz; // Empty.
2755 fl_for_cur_sz.set_size(cur_sz);
2756 {
2757 MutexLockerEx x(_indexedFreeListParLocks[cur_sz],
2758 Mutex::_no_safepoint_check_flag);
2759 AdaptiveFreeList<FreeChunk>* gfl = &_indexedFreeList[cur_sz];
2760 if (gfl->count() != 0) {
2761 // nn is the number of chunks of size cur_sz that
2762 // we'd need to split k-ways each, in order to create
2763 // "n" chunks of size word_sz each.
2764 const size_t nn = MAX2(n/k, (size_t)1);
2765 gfl->getFirstNChunksFromList(nn, &fl_for_cur_sz);
2766 found = true;
2767 if (k > 1) {
2768 // Update split death stats for the cur_sz-size blocks list:
2769 // we increment the split death count by the number of blocks
2770 // we just took from the cur_sz-size blocks list and which
2771 // we will be splitting below.
2772 ssize_t deaths = gfl->split_deaths() +
2773 fl_for_cur_sz.count();
2774 gfl->set_split_deaths(deaths);
2775 }
2776 }
2777 }
2778 // Now transfer fl_for_cur_sz to fl. Common case, we hope, is k = 1.
2779 if (found) {
2780 if (k == 1) {
2781 fl->prepend(&fl_for_cur_sz);
2782 } else {
2783 // Divide each block on fl_for_cur_sz up k ways.
2784 FreeChunk* fc;
2785 while ((fc = fl_for_cur_sz.get_chunk_at_head()) != NULL) {
2786 // Must do this in reverse order, so that anybody attempting to
2787 // access the main chunk sees it as a single free block until we
2788 // change it.
2789 size_t fc_size = fc->size();
2790 assert(fc->is_free(), "Error");
2791 for (int i = k-1; i >= 0; i--) {
2792 FreeChunk* ffc = (FreeChunk*)((HeapWord*)fc + i * word_sz);
2793 assert((i != 0) ||
2794 ((fc == ffc) && ffc->is_free() &&
2795 (ffc->size() == k*word_sz) && (fc_size == word_sz)),
2796 "Counting error");
2797 ffc->set_size(word_sz);
2798 ffc->link_prev(NULL); // Mark as a free block for other (parallel) GC threads.
2799 ffc->link_next(NULL);
2800 // Above must occur before BOT is updated below.
2801 OrderAccess::storestore();
2802 // splitting from the right, fc_size == i * word_sz
2803 _bt.mark_block((HeapWord*)ffc, word_sz, true /* reducing */);
2804 fc_size -= word_sz;
2805 assert(fc_size == i*word_sz, "Error");
2806 _bt.verify_not_unallocated((HeapWord*)ffc, word_sz);
2807 _bt.verify_single_block((HeapWord*)fc, fc_size);
2808 _bt.verify_single_block((HeapWord*)ffc, word_sz);
2809 // Push this on "fl".
2810 fl->return_chunk_at_head(ffc);
2811 }
2812 // TRAP
2813 assert(fl->tail()->next() == NULL, "List invariant.");
2814 }
2815 }
2816 // Update birth stats for this block size.
2817 size_t num = fl->count();
2818 MutexLockerEx x(_indexedFreeListParLocks[word_sz],
2819 Mutex::_no_safepoint_check_flag);
2820 ssize_t births = _indexedFreeList[word_sz].split_births() + num;
2821 _indexedFreeList[word_sz].set_split_births(births);
2822 return true;
2823 }
2824 }
2825 return found;
2826 }
2827 }
2828
2829 FreeChunk* CompactibleFreeListSpace::get_n_way_chunk_to_split(size_t word_sz, size_t n) {
2830
2831 FreeChunk* fc = NULL;
2832 FreeChunk* rem_fc = NULL;
2833 size_t rem;
2834 {
2835 MutexLockerEx x(parDictionaryAllocLock(),
2836 Mutex::_no_safepoint_check_flag);
2837 while (n > 0) {
2838 fc = dictionary()->get_chunk(MAX2(n * word_sz, _dictionary->min_size()),
2839 FreeBlockDictionary<FreeChunk>::atLeast);
2840 if (fc != NULL) {
2841 break;
2842 } else {
2843 n--;
2844 }
2845 }
2846 if (fc == NULL) return NULL;
2847 // Otherwise, split up that block.
2848 assert((ssize_t)n >= 1, "Control point invariant");
2849 assert(fc->is_free(), "Error: should be a free block");
2850 _bt.verify_single_block((HeapWord*)fc, fc->size());
2851 const size_t nn = fc->size() / word_sz;
2852 n = MIN2(nn, n);
2853 assert((ssize_t)n >= 1, "Control point invariant");
2854 rem = fc->size() - n * word_sz;
2855 // If there is a remainder, and it's too small, allocate one fewer.
2856 if (rem > 0 && rem < MinChunkSize) {
2857 n--; rem += word_sz;
2858 }
2859 // Note that at this point we may have n == 0.
2860 assert((ssize_t)n >= 0, "Control point invariant");
2861
2862 // If n is 0, the chunk fc that was found is not large
2863 // enough to leave a viable remainder. We are unable to
2864 // allocate even one block. Return fc to the
2865 // dictionary and return, leaving "fl" empty.
2866 if (n == 0) {
2867 returnChunkToDictionary(fc);
2868 return NULL;
2869 }
2870
2871 _bt.allocated((HeapWord*)fc, fc->size(), true /* reducing */); // update _unallocated_blk
2872 dictionary()->dict_census_update(fc->size(),
2873 true /*split*/,
2874 false /*birth*/);
2875
2876 // First return the remainder, if any.
2877 // Note that we hold the lock until we decide if we're going to give
2878 // back the remainder to the dictionary, since a concurrent allocation
2879 // may otherwise see the heap as empty. (We're willing to take that
2880 // hit if the block is a small block.)
2881 if (rem > 0) {
2882 size_t prefix_size = n * word_sz;
2883 rem_fc = (FreeChunk*)((HeapWord*)fc + prefix_size);
2884 rem_fc->set_size(rem);
2885 rem_fc->link_prev(NULL); // Mark as a free block for other (parallel) GC threads.
2886 rem_fc->link_next(NULL);
2887 // Above must occur before BOT is updated below.
2888 assert((ssize_t)n > 0 && prefix_size > 0 && rem_fc > fc, "Error");
2889 OrderAccess::storestore();
2890 _bt.split_block((HeapWord*)fc, fc->size(), prefix_size);
2891 assert(fc->is_free(), "Error");
2892 fc->set_size(prefix_size);
2893 if (rem >= IndexSetSize) {
2894 returnChunkToDictionary(rem_fc);
2895 dictionary()->dict_census_update(rem, true /*split*/, true /*birth*/);
2896 rem_fc = NULL;
2897 }
2898 // Otherwise, return it to the small list below.
2899 }
2900 }
2901 if (rem_fc != NULL) {
2902 MutexLockerEx x(_indexedFreeListParLocks[rem],
2903 Mutex::_no_safepoint_check_flag);
2904 _bt.verify_not_unallocated((HeapWord*)rem_fc, rem_fc->size());
2905 _indexedFreeList[rem].return_chunk_at_head(rem_fc);
2906 smallSplitBirth(rem);
2907 }
2908 assert(n * word_sz == fc->size(),
2909 err_msg("Chunk size " SIZE_FORMAT " is not exactly splittable by "
2910 SIZE_FORMAT " sized chunks of size " SIZE_FORMAT,
2911 fc->size(), n, word_sz));
2912 return fc;
2913 }
2914
2915 void CompactibleFreeListSpace:: par_get_chunk_of_blocks_dictionary(size_t word_sz, size_t targetted_number_of_chunks, AdaptiveFreeList<FreeChunk>* fl) {
2916
2917 FreeChunk* fc = get_n_way_chunk_to_split(word_sz, targetted_number_of_chunks);
2918
2919 if (fc == NULL) {
2920 return;
2921 }
2922
2923 size_t n = fc->size() / word_sz;
2924
2925 assert((ssize_t)n > 0, "Consistency");
2926 // Now do the splitting up.
2927 // Must do this in reverse order, so that anybody attempting to
2928 // access the main chunk sees it as a single free block until we
2929 // change it.
2930 size_t fc_size = n * word_sz;
2931 // All but first chunk in this loop
2932 for (ssize_t i = n-1; i > 0; i--) {
2933 FreeChunk* ffc = (FreeChunk*)((HeapWord*)fc + i * word_sz);
2934 ffc->set_size(word_sz);
2935 ffc->link_prev(NULL); // Mark as a free block for other (parallel) GC threads.
2936 ffc->link_next(NULL);
2937 // Above must occur before BOT is updated below.
2938 OrderAccess::storestore();
2939 // splitting from the right, fc_size == (n - i + 1) * wordsize
2940 _bt.mark_block((HeapWord*)ffc, word_sz, true /* reducing */);
2941 fc_size -= word_sz;
2942 _bt.verify_not_unallocated((HeapWord*)ffc, ffc->size());
2943 _bt.verify_single_block((HeapWord*)ffc, ffc->size());
2944 _bt.verify_single_block((HeapWord*)fc, fc_size);
2945 // Push this on "fl".
2946 fl->return_chunk_at_head(ffc);
2947 }
2948 // First chunk
2949 assert(fc->is_free() && fc->size() == n*word_sz, "Error: should still be a free block");
2950 // The blocks above should show their new sizes before the first block below
2951 fc->set_size(word_sz);
2952 fc->link_prev(NULL); // idempotent wrt free-ness, see assert above
2953 fc->link_next(NULL);
2954 _bt.verify_not_unallocated((HeapWord*)fc, fc->size());
2955 _bt.verify_single_block((HeapWord*)fc, fc->size());
2956 fl->return_chunk_at_head(fc);
2957
2958 assert((ssize_t)n > 0 && (ssize_t)n == fl->count(), "Incorrect number of blocks");
2959 {
2960 // Update the stats for this block size.
2961 MutexLockerEx x(_indexedFreeListParLocks[word_sz],
2962 Mutex::_no_safepoint_check_flag);
2963 const ssize_t births = _indexedFreeList[word_sz].split_births() + n;
2964 _indexedFreeList[word_sz].set_split_births(births);
2965 // ssize_t new_surplus = _indexedFreeList[word_sz].surplus() + n;
2966 // _indexedFreeList[word_sz].set_surplus(new_surplus);
2967 }
2968
2969 // TRAP
2970 assert(fl->tail()->next() == NULL, "List invariant.");
2971 }
2972
2973 void CompactibleFreeListSpace:: par_get_chunk_of_blocks(size_t word_sz, size_t n, AdaptiveFreeList<FreeChunk>* fl) {
2974 assert(fl->count() == 0, "Precondition.");
2975 assert(word_sz < CompactibleFreeListSpace::IndexSetSize,
2976 "Precondition");
2977
2978 if (par_get_chunk_of_blocks_IFL(word_sz, n, fl)) {
2979 // Got it
2980 return;
2981 }
2982
2983 // Otherwise, we'll split a block from the dictionary.
2984 par_get_chunk_of_blocks_dictionary(word_sz, n, fl);
2985 }
2986
2987 // Set up the space's par_seq_tasks structure for work claiming
2988 // for parallel rescan. See CMSParRemarkTask where this is currently used.
2989 // XXX Need to suitably abstract and generalize this and the next
2990 // method into one.
2991 void
2992 CompactibleFreeListSpace::
2993 initialize_sequential_subtasks_for_rescan(int n_threads) {
2994 // The "size" of each task is fixed according to rescan_task_size.
2995 assert(n_threads > 0, "Unexpected n_threads argument");
2996 const size_t task_size = rescan_task_size();
2997 size_t n_tasks = (used_region().word_size() + task_size - 1)/task_size;
2998 assert((n_tasks == 0) == used_region().is_empty(), "n_tasks incorrect");
2999 assert(n_tasks == 0 ||
3000 ((used_region().start() + (n_tasks - 1)*task_size < used_region().end()) &&
3001 (used_region().start() + n_tasks*task_size >= used_region().end())),
3002 "n_tasks calculation incorrect");
3003 SequentialSubTasksDone* pst = conc_par_seq_tasks();
3004 assert(!pst->valid(), "Clobbering existing data?");
3005 // Sets the condition for completion of the subtask (how many threads
3006 // need to finish in order to be done).
3007 pst->set_n_threads(n_threads);
3008 pst->set_n_tasks((int)n_tasks);
3009 }
3010
3011 // Set up the space's par_seq_tasks structure for work claiming
3012 // for parallel concurrent marking. See CMSConcMarkTask where this is currently used.
3013 void
3014 CompactibleFreeListSpace::
3015 initialize_sequential_subtasks_for_marking(int n_threads,
3016 HeapWord* low) {
3017 // The "size" of each task is fixed according to rescan_task_size.
3018 assert(n_threads > 0, "Unexpected n_threads argument");
3019 const size_t task_size = marking_task_size();
3020 assert(task_size > CardTableModRefBS::card_size_in_words &&
3021 (task_size % CardTableModRefBS::card_size_in_words == 0),
3022 "Otherwise arithmetic below would be incorrect");
3023 MemRegion span = _gen->reserved();
3024 if (low != NULL) {
3025 if (span.contains(low)) {
3026 // Align low down to a card boundary so that
3027 // we can use block_offset_careful() on span boundaries.
3028 HeapWord* aligned_low = (HeapWord*)align_size_down((uintptr_t)low,
3029 CardTableModRefBS::card_size);
3030 // Clip span prefix at aligned_low
3031 span = span.intersection(MemRegion(aligned_low, span.end()));
3032 } else if (low > span.end()) {
3033 span = MemRegion(low, low); // Null region
3034 } // else use entire span
3035 }
3036 assert(span.is_empty() ||
3037 ((uintptr_t)span.start() % CardTableModRefBS::card_size == 0),
3038 "span should start at a card boundary");
3039 size_t n_tasks = (span.word_size() + task_size - 1)/task_size;
3040 assert((n_tasks == 0) == span.is_empty(), "Inconsistency");
3041 assert(n_tasks == 0 ||
3042 ((span.start() + (n_tasks - 1)*task_size < span.end()) &&
3043 (span.start() + n_tasks*task_size >= span.end())),
3044 "n_tasks calculation incorrect");
3045 SequentialSubTasksDone* pst = conc_par_seq_tasks();
3046 assert(!pst->valid(), "Clobbering existing data?");
3047 // Sets the condition for completion of the subtask (how many threads
3048 // need to finish in order to be done).
3049 pst->set_n_threads(n_threads);
3050 pst->set_n_tasks((int)n_tasks);
3051 }