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