1/*
2 * Copyright (c) 2001, 2019, 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#ifndef SHARE_GC_CMS_COMPACTIBLEFREELISTSPACE_HPP
26#define SHARE_GC_CMS_COMPACTIBLEFREELISTSPACE_HPP
27
28#include "gc/cms/adaptiveFreeList.hpp"
29#include "gc/cms/promotionInfo.hpp"
30#include "gc/shared/blockOffsetTable.hpp"
31#include "gc/shared/cardTable.hpp"
32#include "gc/shared/space.hpp"
33#include "logging/log.hpp"
34#include "memory/binaryTreeDictionary.hpp"
35#include "memory/freeList.hpp"
36
37// Classes in support of keeping track of promotions into a non-Contiguous
38// space, in this case a CompactibleFreeListSpace.
39
40// Forward declarations
41class CMSCollector;
42class CompactibleFreeListSpace;
43class ConcurrentMarkSweepGeneration;
44class BlkClosure;
45class BlkClosureCareful;
46class FreeChunk;
47class UpwardsObjectClosure;
48class ObjectClosureCareful;
49class Klass;
50
51class AFLBinaryTreeDictionary : public BinaryTreeDictionary<FreeChunk, AdaptiveFreeList<FreeChunk> > {
52 public:
53 AFLBinaryTreeDictionary(MemRegion mr)
54 : BinaryTreeDictionary<FreeChunk, AdaptiveFreeList<FreeChunk> >(mr) {}
55
56 // Find the list with size "size" in the binary tree and update
57 // the statistics in the list according to "split" (chunk was
58 // split or coalesce) and "birth" (chunk was added or removed).
59 void dict_census_update(size_t size, bool split, bool birth);
60 // Return true if the dictionary is overpopulated (more chunks of
61 // this size than desired) for size "size".
62 bool coal_dict_over_populated(size_t size);
63 // Methods called at the beginning of a sweep to prepare the
64 // statistics for the sweep.
65 void begin_sweep_dict_census(double coalSurplusPercent,
66 float inter_sweep_current,
67 float inter_sweep_estimate,
68 float intra_sweep_estimate);
69 // Methods called after the end of a sweep to modify the
70 // statistics for the sweep.
71 void end_sweep_dict_census(double splitSurplusPercent);
72 // Accessors for statistics
73 void set_tree_surplus(double splitSurplusPercent);
74 void set_tree_hints(void);
75 // Reset statistics for all the lists in the tree.
76 void clear_tree_census(void);
77 // Print the statistics for all the lists in the tree. Also may
78 // print out summaries.
79 void print_dict_census(outputStream* st) const;
80};
81
82class LinearAllocBlock {
83 public:
84 LinearAllocBlock() : _ptr(0), _word_size(0), _refillSize(0),
85 _allocation_size_limit(0) {}
86 void set(HeapWord* ptr, size_t word_size, size_t refill_size,
87 size_t allocation_size_limit) {
88 _ptr = ptr;
89 _word_size = word_size;
90 _refillSize = refill_size;
91 _allocation_size_limit = allocation_size_limit;
92 }
93 HeapWord* _ptr;
94 size_t _word_size;
95 size_t _refillSize;
96 size_t _allocation_size_limit; // Largest size that will be allocated
97
98 void print_on(outputStream* st) const;
99};
100
101// Concrete subclass of CompactibleSpace that implements
102// a free list space, such as used in the concurrent mark sweep
103// generation.
104
105class CompactibleFreeListSpace: public CompactibleSpace {
106 friend class VMStructs;
107 friend class ConcurrentMarkSweepGeneration;
108 friend class CMSCollector;
109 // Local alloc buffer for promotion into this space.
110 friend class CompactibleFreeListSpaceLAB;
111 // Allow scan_and_* functions to call (private) overrides of the auxiliary functions on this class
112 template <typename SpaceType>
113 friend void CompactibleSpace::scan_and_adjust_pointers(SpaceType* space);
114 template <typename SpaceType>
115 friend void CompactibleSpace::scan_and_compact(SpaceType* space);
116 template <typename SpaceType>
117 friend void CompactibleSpace::verify_up_to_first_dead(SpaceType* space);
118 template <typename SpaceType>
119 friend void CompactibleSpace::scan_and_forward(SpaceType* space, CompactPoint* cp);
120
121 // "Size" of chunks of work (executed during parallel remark phases
122 // of CMS collection); this probably belongs in CMSCollector, although
123 // it's cached here because it's used in
124 // initialize_sequential_subtasks_for_rescan() which modifies
125 // par_seq_tasks which also lives in Space. XXX
126 const size_t _rescan_task_size;
127 const size_t _marking_task_size;
128
129 // Yet another sequential tasks done structure. This supports
130 // CMS GC, where we have threads dynamically
131 // claiming sub-tasks from a larger parallel task.
132 SequentialSubTasksDone _conc_par_seq_tasks;
133
134 BlockOffsetArrayNonContigSpace _bt;
135
136 CMSCollector* _collector;
137 ConcurrentMarkSweepGeneration* _old_gen;
138
139 // Data structures for free blocks (used during allocation/sweeping)
140
141 // Allocation is done linearly from two different blocks depending on
142 // whether the request is small or large, in an effort to reduce
143 // fragmentation. We assume that any locking for allocation is done
144 // by the containing generation. Thus, none of the methods in this
145 // space are re-entrant.
146 enum SomeConstants {
147 SmallForLinearAlloc = 16, // size < this then use _sLAB
148 SmallForDictionary = 257, // size < this then use _indexedFreeList
149 IndexSetSize = SmallForDictionary // keep this odd-sized
150 };
151 static size_t IndexSetStart;
152 static size_t IndexSetStride;
153 static size_t _min_chunk_size_in_bytes;
154
155 private:
156 enum FitStrategyOptions {
157 FreeBlockStrategyNone = 0,
158 FreeBlockBestFitFirst
159 };
160
161 PromotionInfo _promoInfo;
162
163 // Helps to impose a global total order on freelistLock ranks;
164 // assumes that CFLSpace's are allocated in global total order
165 static int _lockRank;
166
167 // A lock protecting the free lists and free blocks;
168 // mutable because of ubiquity of locking even for otherwise const methods
169 mutable Mutex _freelistLock;
170
171 // Locking verifier convenience function
172 void assert_locked() const PRODUCT_RETURN;
173 void assert_locked(const Mutex* lock) const PRODUCT_RETURN;
174
175 // Linear allocation blocks
176 LinearAllocBlock _smallLinearAllocBlock;
177
178 AFLBinaryTreeDictionary* _dictionary; // Pointer to dictionary for large size blocks
179
180 // Indexed array for small size blocks
181 AdaptiveFreeList<FreeChunk> _indexedFreeList[IndexSetSize];
182
183 // Allocation strategy
184 bool _fitStrategy; // Use best fit strategy
185
186 // This is an address close to the largest free chunk in the heap.
187 // It is currently assumed to be at the end of the heap. Free
188 // chunks with addresses greater than nearLargestChunk are coalesced
189 // in an effort to maintain a large chunk at the end of the heap.
190 HeapWord* _nearLargestChunk;
191
192 // Used to keep track of limit of sweep for the space
193 HeapWord* _sweep_limit;
194
195 // Used to make the young collector update the mod union table
196 MemRegionClosure* _preconsumptionDirtyCardClosure;
197
198 // Support for compacting cms
199 HeapWord* cross_threshold(HeapWord* start, HeapWord* end);
200 HeapWord* forward(oop q, size_t size, CompactPoint* cp, HeapWord* compact_top);
201
202 // Initialization helpers.
203 void initializeIndexedFreeListArray();
204
205 // Extra stuff to manage promotion parallelism.
206
207 // A lock protecting the dictionary during par promotion allocation.
208 mutable Mutex _parDictionaryAllocLock;
209 Mutex* parDictionaryAllocLock() const { return &_parDictionaryAllocLock; }
210
211 // Locks protecting the exact lists during par promotion allocation.
212 Mutex* _indexedFreeListParLocks[IndexSetSize];
213
214 // Attempt to obtain up to "n" blocks of the size "word_sz" (which is
215 // required to be smaller than "IndexSetSize".) If successful,
216 // adds them to "fl", which is required to be an empty free list.
217 // If the count of "fl" is negative, it's absolute value indicates a
218 // number of free chunks that had been previously "borrowed" from global
219 // list of size "word_sz", and must now be decremented.
220 void par_get_chunk_of_blocks(size_t word_sz, size_t n, AdaptiveFreeList<FreeChunk>* fl);
221
222 // Used by par_get_chunk_of_blocks() for the chunks from the
223 // indexed_free_lists.
224 bool par_get_chunk_of_blocks_IFL(size_t word_sz, size_t n, AdaptiveFreeList<FreeChunk>* fl);
225
226 // Used by par_get_chunk_of_blocks_dictionary() to get a chunk
227 // evenly splittable into "n" "word_sz" chunks. Returns that
228 // evenly splittable chunk. May split a larger chunk to get the
229 // evenly splittable chunk.
230 FreeChunk* get_n_way_chunk_to_split(size_t word_sz, size_t n);
231
232 // Used by par_get_chunk_of_blocks() for the chunks from the
233 // dictionary.
234 void par_get_chunk_of_blocks_dictionary(size_t word_sz, size_t n, AdaptiveFreeList<FreeChunk>* fl);
235
236 // Allocation helper functions
237 // Allocate using a strategy that takes from the indexed free lists
238 // first. This allocation strategy assumes a companion sweeping
239 // strategy that attempts to keep the needed number of chunks in each
240 // indexed free lists.
241 HeapWord* allocate_adaptive_freelists(size_t size);
242
243 // Gets a chunk from the linear allocation block (LinAB). If there
244 // is not enough space in the LinAB, refills it.
245 HeapWord* getChunkFromLinearAllocBlock(LinearAllocBlock* blk, size_t size);
246 HeapWord* getChunkFromSmallLinearAllocBlock(size_t size);
247 // Get a chunk from the space remaining in the linear allocation block. Do
248 // not attempt to refill if the space is not available, return NULL. Do the
249 // repairs on the linear allocation block as appropriate.
250 HeapWord* getChunkFromLinearAllocBlockRemainder(LinearAllocBlock* blk, size_t size);
251 inline HeapWord* getChunkFromSmallLinearAllocBlockRemainder(size_t size);
252
253 // Helper function for getChunkFromIndexedFreeList.
254 // Replenish the indexed free list for this "size". Do not take from an
255 // underpopulated size.
256 FreeChunk* getChunkFromIndexedFreeListHelper(size_t size, bool replenish = true);
257
258 // Get a chunk from the indexed free list. If the indexed free list
259 // does not have a free chunk, try to replenish the indexed free list
260 // then get the free chunk from the replenished indexed free list.
261 inline FreeChunk* getChunkFromIndexedFreeList(size_t size);
262
263 // The returned chunk may be larger than requested (or null).
264 FreeChunk* getChunkFromDictionary(size_t size);
265 // The returned chunk is the exact size requested (or null).
266 FreeChunk* getChunkFromDictionaryExact(size_t size);
267
268 // Find a chunk in the indexed free list that is the best
269 // fit for size "numWords".
270 FreeChunk* bestFitSmall(size_t numWords);
271 // For free list "fl" of chunks of size > numWords,
272 // remove a chunk, split off a chunk of size numWords
273 // and return it. The split off remainder is returned to
274 // the free lists. The old name for getFromListGreater
275 // was lookInListGreater.
276 FreeChunk* getFromListGreater(AdaptiveFreeList<FreeChunk>* fl, size_t numWords);
277 // Get a chunk in the indexed free list or dictionary,
278 // by considering a larger chunk and splitting it.
279 FreeChunk* getChunkFromGreater(size_t numWords);
280 // Verify that the given chunk is in the indexed free lists.
281 bool verifyChunkInIndexedFreeLists(FreeChunk* fc) const;
282 // Remove the specified chunk from the indexed free lists.
283 void removeChunkFromIndexedFreeList(FreeChunk* fc);
284 // Remove the specified chunk from the dictionary.
285 void removeChunkFromDictionary(FreeChunk* fc);
286 // Split a free chunk into a smaller free chunk of size "new_size".
287 // Return the smaller free chunk and return the remainder to the
288 // free lists.
289 FreeChunk* splitChunkAndReturnRemainder(FreeChunk* chunk, size_t new_size);
290 // Add a chunk to the free lists.
291 void addChunkToFreeLists(HeapWord* chunk, size_t size);
292 // Add a chunk to the free lists, preferring to suffix it
293 // to the last free chunk at end of space if possible, and
294 // updating the block census stats as well as block offset table.
295 // Take any locks as appropriate if we are multithreaded.
296 void addChunkToFreeListsAtEndRecordingStats(HeapWord* chunk, size_t size);
297 // Add a free chunk to the indexed free lists.
298 void returnChunkToFreeList(FreeChunk* chunk);
299 // Add a free chunk to the dictionary.
300 void returnChunkToDictionary(FreeChunk* chunk);
301
302 // Functions for maintaining the linear allocation buffers (LinAB).
303 // Repairing a linear allocation block refers to operations
304 // performed on the remainder of a LinAB after an allocation
305 // has been made from it.
306 void repairLinearAllocationBlocks();
307 void repairLinearAllocBlock(LinearAllocBlock* blk);
308 void refillLinearAllocBlock(LinearAllocBlock* blk);
309 void refillLinearAllocBlockIfNeeded(LinearAllocBlock* blk);
310 void refillLinearAllocBlocksIfNeeded();
311
312 void verify_objects_initialized() const;
313
314 // Statistics reporting helper functions
315 void reportFreeListStatistics(const char* title) const;
316 void reportIndexedFreeListStatistics(outputStream* st) const;
317 size_t maxChunkSizeInIndexedFreeLists() const;
318 size_t numFreeBlocksInIndexedFreeLists() const;
319 // Accessor
320 HeapWord* unallocated_block() const {
321 if (BlockOffsetArrayUseUnallocatedBlock) {
322 HeapWord* ub = _bt.unallocated_block();
323 assert(ub >= bottom() &&
324 ub <= end(), "space invariant");
325 return ub;
326 } else {
327 return end();
328 }
329 }
330 void freed(HeapWord* start, size_t size) {
331 _bt.freed(start, size);
332 }
333
334 // Auxiliary functions for scan_and_{forward,adjust_pointers,compact} support.
335 // See comments for CompactibleSpace for more information.
336 inline HeapWord* scan_limit() const {
337 return end();
338 }
339
340 inline bool scanned_block_is_obj(const HeapWord* addr) const {
341 return CompactibleFreeListSpace::block_is_obj(addr); // Avoid virtual call
342 }
343
344 inline size_t scanned_block_size(const HeapWord* addr) const {
345 return CompactibleFreeListSpace::block_size(addr); // Avoid virtual call
346 }
347
348 inline size_t adjust_obj_size(size_t size) const {
349 return adjustObjectSize(size);
350 }
351
352 inline size_t obj_size(const HeapWord* addr) const;
353
354 protected:
355 // Reset the indexed free list to its initial empty condition.
356 void resetIndexedFreeListArray();
357 // Reset to an initial state with a single free block described
358 // by the MemRegion parameter.
359 void reset(MemRegion mr);
360 // Return the total number of words in the indexed free lists.
361 size_t totalSizeInIndexedFreeLists() const;
362
363 public:
364 // Constructor
365 CompactibleFreeListSpace(BlockOffsetSharedArray* bs, MemRegion mr);
366 // Accessors
367 bool bestFitFirst() { return _fitStrategy == FreeBlockBestFitFirst; }
368 AFLBinaryTreeDictionary* dictionary() const { return _dictionary; }
369 HeapWord* nearLargestChunk() const { return _nearLargestChunk; }
370 void set_nearLargestChunk(HeapWord* v) { _nearLargestChunk = v; }
371
372 // Set CMS global values.
373 static void set_cms_values();
374
375 // Return the free chunk at the end of the space. If no such
376 // chunk exists, return NULL.
377 FreeChunk* find_chunk_at_end();
378
379 void set_collector(CMSCollector* collector) { _collector = collector; }
380
381 // Support for parallelization of rescan and marking.
382 const size_t rescan_task_size() const { return _rescan_task_size; }
383 const size_t marking_task_size() const { return _marking_task_size; }
384 // Return ergonomic max size for CMSRescanMultiple and CMSConcMarkMultiple.
385 const size_t max_flag_size_for_task_size() const;
386 SequentialSubTasksDone* conc_par_seq_tasks() {return &_conc_par_seq_tasks; }
387 void initialize_sequential_subtasks_for_rescan(int n_threads);
388 void initialize_sequential_subtasks_for_marking(int n_threads,
389 HeapWord* low = NULL);
390
391 virtual MemRegionClosure* preconsumptionDirtyCardClosure() const {
392 return _preconsumptionDirtyCardClosure;
393 }
394
395 void setPreconsumptionDirtyCardClosure(MemRegionClosure* cl) {
396 _preconsumptionDirtyCardClosure = cl;
397 }
398
399 // Space enquiries
400 size_t used() const;
401 size_t free() const;
402 size_t max_alloc_in_words() const;
403 // XXX: should have a less conservative used_region() than that of
404 // Space; we could consider keeping track of highest allocated
405 // address and correcting that at each sweep, as the sweeper
406 // goes through the entire allocated part of the generation. We
407 // could also use that information to keep the sweeper from
408 // sweeping more than is necessary. The allocator and sweeper will
409 // of course need to synchronize on this, since the sweeper will
410 // try to bump down the address and the allocator will try to bump it up.
411 // For now, however, we'll just use the default used_region()
412 // which overestimates the region by returning the entire
413 // committed region (this is safe, but inefficient).
414
415 // Returns a subregion of the space containing all the objects in
416 // the space.
417 MemRegion used_region() const {
418 return MemRegion(bottom(),
419 BlockOffsetArrayUseUnallocatedBlock ?
420 unallocated_block() : end());
421 }
422
423 virtual bool is_free_block(const HeapWord* p) const;
424
425 // Resizing support
426 void set_end(HeapWord* value); // override
427
428 // Never mangle CompactibleFreeListSpace
429 void mangle_unused_area() {}
430 void mangle_unused_area_complete() {}
431
432 // Mutual exclusion support
433 Mutex* freelistLock() const { return &_freelistLock; }
434
435 // Iteration support
436 void oop_iterate(OopIterateClosure* cl);
437
438 void object_iterate(ObjectClosure* blk);
439 // Apply the closure to each object in the space whose references
440 // point to objects in the heap. The usage of CompactibleFreeListSpace
441 // by the ConcurrentMarkSweepGeneration for concurrent GC's allows
442 // objects in the space with references to objects that are no longer
443 // valid. For example, an object may reference another object
444 // that has already been sweep up (collected). This method uses
445 // obj_is_alive() to determine whether it is safe to iterate of
446 // an object.
447 void safe_object_iterate(ObjectClosure* blk);
448
449 // Iterate over all objects that intersect with mr, calling "cl->do_object"
450 // on each. There is an exception to this: if this closure has already
451 // been invoked on an object, it may skip such objects in some cases. This is
452 // Most likely to happen in an "upwards" (ascending address) iteration of
453 // MemRegions.
454 void object_iterate_mem(MemRegion mr, UpwardsObjectClosure* cl);
455
456 // Requires that "mr" be entirely within the space.
457 // Apply "cl->do_object" to all objects that intersect with "mr".
458 // If the iteration encounters an unparseable portion of the region,
459 // terminate the iteration and return the address of the start of the
460 // subregion that isn't done. Return of "NULL" indicates that the
461 // iteration completed.
462 HeapWord* object_iterate_careful_m(MemRegion mr,
463 ObjectClosureCareful* cl);
464
465 // Override: provides a DCTO_CL specific to this kind of space.
466 DirtyCardToOopClosure* new_dcto_cl(OopIterateClosure* cl,
467 CardTable::PrecisionStyle precision,
468 HeapWord* boundary,
469 bool parallel);
470
471 void blk_iterate(BlkClosure* cl);
472 void blk_iterate_careful(BlkClosureCareful* cl);
473 HeapWord* block_start_const(const void* p) const;
474 HeapWord* block_start_careful(const void* p) const;
475 size_t block_size(const HeapWord* p) const;
476 size_t block_size_no_stall(HeapWord* p, const CMSCollector* c) const;
477 bool block_is_obj(const HeapWord* p) const;
478 bool obj_is_alive(const HeapWord* p) const;
479 size_t block_size_nopar(const HeapWord* p) const;
480 bool block_is_obj_nopar(const HeapWord* p) const;
481
482 // Iteration support for promotion
483 void save_marks();
484 bool no_allocs_since_save_marks();
485
486 // Iteration support for sweeping
487 void save_sweep_limit() {
488 _sweep_limit = BlockOffsetArrayUseUnallocatedBlock ?
489 unallocated_block() : end();
490 log_develop_trace(gc, sweep)(">>>>> Saving sweep limit " PTR_FORMAT
491 " for space [" PTR_FORMAT "," PTR_FORMAT ") <<<<<<",
492 p2i(_sweep_limit), p2i(bottom()), p2i(end()));
493 }
494 NOT_PRODUCT(
495 void clear_sweep_limit() { _sweep_limit = NULL; }
496 )
497 HeapWord* sweep_limit() { return _sweep_limit; }
498
499 // Apply "blk->do_oop" to the addresses of all reference fields in objects
500 // promoted into this generation since the most recent save_marks() call.
501 // Fields in objects allocated by applications of the closure
502 // *are* included in the iteration. Thus, when the iteration completes
503 // there should be no further such objects remaining.
504 template <typename OopClosureType>
505 void oop_since_save_marks_iterate(OopClosureType* blk);
506
507 // Allocation support
508 HeapWord* allocate(size_t size);
509 HeapWord* par_allocate(size_t size);
510
511 oop promote(oop obj, size_t obj_size);
512 void gc_prologue();
513 void gc_epilogue();
514
515 // This call is used by a containing CMS generation / collector
516 // to inform the CFLS space that a sweep has been completed
517 // and that the space can do any related house-keeping functions.
518 void sweep_completed();
519
520 // For an object in this space, the mark-word's two
521 // LSB's having the value [11] indicates that it has been
522 // promoted since the most recent call to save_marks() on
523 // this generation and has not subsequently been iterated
524 // over (using oop_since_save_marks_iterate() above).
525 // This property holds only for single-threaded collections,
526 // and is typically used for Cheney scans; for MT scavenges,
527 // the property holds for all objects promoted during that
528 // scavenge for the duration of the scavenge and is used
529 // by card-scanning to avoid scanning objects (being) promoted
530 // during that scavenge.
531 bool obj_allocated_since_save_marks(const oop obj) const {
532 assert(is_in_reserved(obj), "Wrong space?");
533 return ((PromotedObject*)obj)->hasPromotedMark();
534 }
535
536 // A worst-case estimate of the space required (in HeapWords) to expand the
537 // heap when promoting an obj of size obj_size.
538 size_t expansionSpaceRequired(size_t obj_size) const;
539
540 FreeChunk* allocateScratch(size_t size);
541
542 // Returns true if either the small or large linear allocation buffer is empty.
543 bool linearAllocationWouldFail() const;
544
545 // Adjust the chunk for the minimum size. This version is called in
546 // most cases in CompactibleFreeListSpace methods.
547 inline static size_t adjustObjectSize(size_t size) {
548 return align_object_size(MAX2(size, (size_t)MinChunkSize));
549 }
550 // This is a virtual version of adjustObjectSize() that is called
551 // only occasionally when the compaction space changes and the type
552 // of the new compaction space is is only known to be CompactibleSpace.
553 size_t adjust_object_size_v(size_t size) const {
554 return adjustObjectSize(size);
555 }
556 // Minimum size of a free block.
557 virtual size_t minimum_free_block_size() const { return MinChunkSize; }
558 void removeFreeChunkFromFreeLists(FreeChunk* chunk);
559 void addChunkAndRepairOffsetTable(HeapWord* chunk, size_t size,
560 bool coalesced);
561
562 // Support for compaction.
563 void prepare_for_compaction(CompactPoint* cp);
564 void adjust_pointers();
565 void compact();
566 // Reset the space to reflect the fact that a compaction of the
567 // space has been done.
568 virtual void reset_after_compaction();
569
570 // Debugging support.
571 void print() const;
572 void print_on(outputStream* st) const;
573 void prepare_for_verify();
574 void verify() const;
575 void verifyFreeLists() const PRODUCT_RETURN;
576 void verifyIndexedFreeLists() const;
577 void verifyIndexedFreeList(size_t size) const;
578 // Verify that the given chunk is in the free lists:
579 // i.e. either the binary tree dictionary, the indexed free lists
580 // or the linear allocation block.
581 bool verify_chunk_in_free_list(FreeChunk* fc) const;
582 // Verify that the given chunk is the linear allocation block.
583 bool verify_chunk_is_linear_alloc_block(FreeChunk* fc) const;
584 // Do some basic checks on the the free lists.
585 void check_free_list_consistency() const PRODUCT_RETURN;
586
587 // Printing support
588 void dump_at_safepoint_with_locks(CMSCollector* c, outputStream* st);
589 void print_indexed_free_lists(outputStream* st) const;
590 void print_dictionary_free_lists(outputStream* st) const;
591 void print_promo_info_blocks(outputStream* st) const;
592
593 NOT_PRODUCT (
594 void initializeIndexedFreeListArrayReturnedBytes();
595 size_t sumIndexedFreeListArrayReturnedBytes();
596 // Return the total number of chunks in the indexed free lists.
597 size_t totalCountInIndexedFreeLists() const;
598 // Return the total number of chunks in the space.
599 size_t totalCount();
600 )
601
602 // The census consists of counts of the quantities such as
603 // the current count of the free chunks, number of chunks
604 // created as a result of the split of a larger chunk or
605 // coalescing of smaller chucks, etc. The counts in the
606 // census is used to make decisions on splitting and
607 // coalescing of chunks during the sweep of garbage.
608
609 // Print the statistics for the free lists.
610 void printFLCensus(size_t sweep_count) const;
611
612 // Statistics functions
613 // Initialize census for lists before the sweep.
614 void beginSweepFLCensus(float inter_sweep_current,
615 float inter_sweep_estimate,
616 float intra_sweep_estimate);
617 // Set the surplus for each of the free lists.
618 void setFLSurplus();
619 // Set the hint for each of the free lists.
620 void setFLHints();
621 // Clear the census for each of the free lists.
622 void clearFLCensus();
623 // Perform functions for the census after the end of the sweep.
624 void endSweepFLCensus(size_t sweep_count);
625 // Return true if the count of free chunks is greater
626 // than the desired number of free chunks.
627 bool coalOverPopulated(size_t size);
628
629// Record (for each size):
630//
631// split-births = #chunks added due to splits in (prev-sweep-end,
632// this-sweep-start)
633// split-deaths = #chunks removed for splits in (prev-sweep-end,
634// this-sweep-start)
635// num-curr = #chunks at start of this sweep
636// num-prev = #chunks at end of previous sweep
637//
638// The above are quantities that are measured. Now define:
639//
640// num-desired := num-prev + split-births - split-deaths - num-curr
641//
642// Roughly, num-prev + split-births is the supply,
643// split-deaths is demand due to other sizes
644// and num-curr is what we have left.
645//
646// Thus, num-desired is roughly speaking the "legitimate demand"
647// for blocks of this size and what we are striving to reach at the
648// end of the current sweep.
649//
650// For a given list, let num-len be its current population.
651// Define, for a free list of a given size:
652//
653// coal-overpopulated := num-len >= num-desired * coal-surplus
654// (coal-surplus is set to 1.05, i.e. we allow a little slop when
655// coalescing -- we do not coalesce unless we think that the current
656// supply has exceeded the estimated demand by more than 5%).
657//
658// For the set of sizes in the binary tree, which is neither dense nor
659// closed, it may be the case that for a particular size we have never
660// had, or do not now have, or did not have at the previous sweep,
661// chunks of that size. We need to extend the definition of
662// coal-overpopulated to such sizes as well:
663//
664// For a chunk in/not in the binary tree, extend coal-overpopulated
665// defined above to include all sizes as follows:
666//
667// . a size that is non-existent is coal-overpopulated
668// . a size that has a num-desired <= 0 as defined above is
669// coal-overpopulated.
670//
671// Also define, for a chunk heap-offset C and mountain heap-offset M:
672//
673// close-to-mountain := C >= 0.99 * M
674//
675// Now, the coalescing strategy is:
676//
677// Coalesce left-hand chunk with right-hand chunk if and
678// only if:
679//
680// EITHER
681// . left-hand chunk is of a size that is coal-overpopulated
682// OR
683// . right-hand chunk is close-to-mountain
684 void smallCoalBirth(size_t size);
685 void smallCoalDeath(size_t size);
686 void coalBirth(size_t size);
687 void coalDeath(size_t size);
688 void smallSplitBirth(size_t size);
689 void smallSplitDeath(size_t size);
690 void split_birth(size_t size);
691 void splitDeath(size_t size);
692 void split(size_t from, size_t to1);
693
694 double flsFrag() const;
695};
696
697// A parallel-GC-thread-local allocation buffer for allocation into a
698// CompactibleFreeListSpace.
699class CompactibleFreeListSpaceLAB : public CHeapObj<mtGC> {
700 // The space that this buffer allocates into.
701 CompactibleFreeListSpace* _cfls;
702
703 // Our local free lists.
704 AdaptiveFreeList<FreeChunk> _indexedFreeList[CompactibleFreeListSpace::IndexSetSize];
705
706 // Initialized from a command-line arg.
707
708 // Allocation statistics in support of dynamic adjustment of
709 // #blocks to claim per get_from_global_pool() call below.
710 static AdaptiveWeightedAverage
711 _blocks_to_claim [CompactibleFreeListSpace::IndexSetSize];
712 static size_t _global_num_blocks [CompactibleFreeListSpace::IndexSetSize];
713 static uint _global_num_workers[CompactibleFreeListSpace::IndexSetSize];
714 size_t _num_blocks [CompactibleFreeListSpace::IndexSetSize];
715
716 // Internal work method
717 void get_from_global_pool(size_t word_sz, AdaptiveFreeList<FreeChunk>* fl);
718
719public:
720 static const int _default_dynamic_old_plab_size = 16;
721 static const int _default_static_old_plab_size = 50;
722
723 CompactibleFreeListSpaceLAB(CompactibleFreeListSpace* cfls);
724
725 // Allocate and return a block of the given size, or else return NULL.
726 HeapWord* alloc(size_t word_sz);
727
728 // Return any unused portions of the buffer to the global pool.
729 void retire(int tid);
730
731 // Dynamic OldPLABSize sizing
732 static void compute_desired_plab_size();
733 // When the settings are modified from default static initialization
734 static void modify_initialization(size_t n, unsigned wt);
735};
736
737size_t PromotionInfo::refillSize() const {
738 const size_t CMSSpoolBlockSize = 256;
739 const size_t sz = heap_word_size(sizeof(SpoolBlock) + sizeof(markOop)
740 * CMSSpoolBlockSize);
741 return CompactibleFreeListSpace::adjustObjectSize(sz);
742}
743
744#endif // SHARE_GC_CMS_COMPACTIBLEFREELISTSPACE_HPP
745