duke@435: /*
ysr@2071:  * Copyright (c) 2001, 2010, Oracle and/or its affiliates. All rights reserved.
duke@435:  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
duke@435:  *
duke@435:  * This code is free software; you can redistribute it and/or modify it
duke@435:  * under the terms of the GNU General Public License version 2 only, as
duke@435:  * published by the Free Software Foundation.
duke@435:  *
duke@435:  * This code is distributed in the hope that it will be useful, but WITHOUT
duke@435:  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
duke@435:  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
duke@435:  * version 2 for more details (a copy is included in the LICENSE file that
duke@435:  * accompanied this code).
duke@435:  *
duke@435:  * You should have received a copy of the GNU General Public License version
duke@435:  * 2 along with this work; if not, write to the Free Software Foundation,
duke@435:  * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
duke@435:  *
trims@1907:  * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
trims@1907:  * or visit www.oracle.com if you need additional information or have any
trims@1907:  * questions.
duke@435:  *
duke@435:  */
duke@435: 
duke@435: // Classes in support of keeping track of promotions into a non-Contiguous
duke@435: // space, in this case a CompactibleFreeListSpace.
duke@435: 
duke@435: // Forward declarations
duke@435: class CompactibleFreeListSpace;
duke@435: class BlkClosure;
duke@435: class BlkClosureCareful;
duke@435: class UpwardsObjectClosure;
duke@435: class ObjectClosureCareful;
duke@435: class Klass;
duke@435: 
duke@435: class LinearAllocBlock VALUE_OBJ_CLASS_SPEC {
duke@435:  public:
duke@435:   LinearAllocBlock() : _ptr(0), _word_size(0), _refillSize(0),
duke@435:     _allocation_size_limit(0) {}
duke@435:   void set(HeapWord* ptr, size_t word_size, size_t refill_size,
duke@435:     size_t allocation_size_limit) {
duke@435:     _ptr = ptr;
duke@435:     _word_size = word_size;
duke@435:     _refillSize = refill_size;
duke@435:     _allocation_size_limit = allocation_size_limit;
duke@435:   }
duke@435:   HeapWord* _ptr;
duke@435:   size_t    _word_size;
duke@435:   size_t    _refillSize;
duke@435:   size_t    _allocation_size_limit;  // largest size that will be allocated
ysr@2071: 
ysr@2071:   void print_on(outputStream* st) const;
duke@435: };
duke@435: 
duke@435: // Concrete subclass of CompactibleSpace that implements
duke@435: // a free list space, such as used in the concurrent mark sweep
duke@435: // generation.
duke@435: 
duke@435: class CompactibleFreeListSpace: public CompactibleSpace {
duke@435:   friend class VMStructs;
duke@435:   friend class ConcurrentMarkSweepGeneration;
duke@435:   friend class ASConcurrentMarkSweepGeneration;
duke@435:   friend class CMSCollector;
duke@435:   friend class CMSPermGenGen;
duke@435:   // Local alloc buffer for promotion into this space.
duke@435:   friend class CFLS_LAB;
duke@435: 
duke@435:   // "Size" of chunks of work (executed during parallel remark phases
duke@435:   // of CMS collection); this probably belongs in CMSCollector, although
duke@435:   // it's cached here because it's used in
duke@435:   // initialize_sequential_subtasks_for_rescan() which modifies
duke@435:   // par_seq_tasks which also lives in Space. XXX
duke@435:   const size_t _rescan_task_size;
duke@435:   const size_t _marking_task_size;
duke@435: 
duke@435:   // Yet another sequential tasks done structure. This supports
duke@435:   // CMS GC, where we have threads dynamically
duke@435:   // claiming sub-tasks from a larger parallel task.
duke@435:   SequentialSubTasksDone _conc_par_seq_tasks;
duke@435: 
duke@435:   BlockOffsetArrayNonContigSpace _bt;
duke@435: 
duke@435:   CMSCollector* _collector;
duke@435:   ConcurrentMarkSweepGeneration* _gen;
duke@435: 
duke@435:   // Data structures for free blocks (used during allocation/sweeping)
duke@435: 
duke@435:   // Allocation is done linearly from two different blocks depending on
duke@435:   // whether the request is small or large, in an effort to reduce
duke@435:   // fragmentation. We assume that any locking for allocation is done
duke@435:   // by the containing generation. Thus, none of the methods in this
duke@435:   // space are re-entrant.
duke@435:   enum SomeConstants {
duke@435:     SmallForLinearAlloc = 16,        // size < this then use _sLAB
duke@435:     SmallForDictionary  = 257,       // size < this then use _indexedFreeList
kvn@1926:     IndexSetSize        = SmallForDictionary  // keep this odd-sized
duke@435:   };
kvn@1926:   static int IndexSetStart;
kvn@1926:   static int IndexSetStride;
duke@435: 
duke@435:  private:
duke@435:   enum FitStrategyOptions {
duke@435:     FreeBlockStrategyNone = 0,
duke@435:     FreeBlockBestFitFirst
duke@435:   };
duke@435: 
duke@435:   PromotionInfo _promoInfo;
duke@435: 
duke@435:   // helps to impose a global total order on freelistLock ranks;
duke@435:   // assumes that CFLSpace's are allocated in global total order
duke@435:   static int   _lockRank;
duke@435: 
duke@435:   // a lock protecting the free lists and free blocks;
duke@435:   // mutable because of ubiquity of locking even for otherwise const methods
duke@435:   mutable Mutex _freelistLock;
duke@435:   // locking verifier convenience function
duke@435:   void assert_locked() const PRODUCT_RETURN;
ysr@1580:   void assert_locked(const Mutex* lock) const PRODUCT_RETURN;
duke@435: 
duke@435:   // Linear allocation blocks
duke@435:   LinearAllocBlock _smallLinearAllocBlock;
duke@435: 
duke@435:   FreeBlockDictionary::DictionaryChoice _dictionaryChoice;
duke@435:   FreeBlockDictionary* _dictionary;    // ptr to dictionary for large size blocks
duke@435: 
duke@435:   FreeList _indexedFreeList[IndexSetSize];
duke@435:                                        // indexed array for small size blocks
duke@435:   // allocation stategy
duke@435:   bool       _fitStrategy;      // Use best fit strategy.
duke@435:   bool       _adaptive_freelists; // Use adaptive freelists
duke@435: 
duke@435:   // This is an address close to the largest free chunk in the heap.
duke@435:   // It is currently assumed to be at the end of the heap.  Free
duke@435:   // chunks with addresses greater than nearLargestChunk are coalesced
duke@435:   // in an effort to maintain a large chunk at the end of the heap.
duke@435:   HeapWord*  _nearLargestChunk;
duke@435: 
duke@435:   // Used to keep track of limit of sweep for the space
duke@435:   HeapWord* _sweep_limit;
duke@435: 
duke@435:   // Support for compacting cms
duke@435:   HeapWord* cross_threshold(HeapWord* start, HeapWord* end);
duke@435:   HeapWord* forward(oop q, size_t size, CompactPoint* cp, HeapWord* compact_top);
duke@435: 
duke@435:   // Initialization helpers.
duke@435:   void initializeIndexedFreeListArray();
duke@435: 
duke@435:   // Extra stuff to manage promotion parallelism.
duke@435: 
duke@435:   // a lock protecting the dictionary during par promotion allocation.
duke@435:   mutable Mutex _parDictionaryAllocLock;
duke@435:   Mutex* parDictionaryAllocLock() const { return &_parDictionaryAllocLock; }
duke@435: 
duke@435:   // Locks protecting the exact lists during par promotion allocation.
duke@435:   Mutex* _indexedFreeListParLocks[IndexSetSize];
duke@435: 
duke@435:   // Attempt to obtain up to "n" blocks of the size "word_sz" (which is
duke@435:   // required to be smaller than "IndexSetSize".)  If successful,
duke@435:   // adds them to "fl", which is required to be an empty free list.
duke@435:   // If the count of "fl" is negative, it's absolute value indicates a
duke@435:   // number of free chunks that had been previously "borrowed" from global
duke@435:   // list of size "word_sz", and must now be decremented.
duke@435:   void par_get_chunk_of_blocks(size_t word_sz, size_t n, FreeList* fl);
duke@435: 
duke@435:   // Allocation helper functions
duke@435:   // Allocate using a strategy that takes from the indexed free lists
duke@435:   // first.  This allocation strategy assumes a companion sweeping
duke@435:   // strategy that attempts to keep the needed number of chunks in each
duke@435:   // indexed free lists.
duke@435:   HeapWord* allocate_adaptive_freelists(size_t size);
duke@435:   // Allocate from the linear allocation buffers first.  This allocation
duke@435:   // strategy assumes maximal coalescing can maintain chunks large enough
duke@435:   // to be used as linear allocation buffers.
duke@435:   HeapWord* allocate_non_adaptive_freelists(size_t size);
duke@435: 
duke@435:   // Gets a chunk from the linear allocation block (LinAB).  If there
duke@435:   // is not enough space in the LinAB, refills it.
duke@435:   HeapWord*  getChunkFromLinearAllocBlock(LinearAllocBlock* blk, size_t size);
duke@435:   HeapWord*  getChunkFromSmallLinearAllocBlock(size_t size);
duke@435:   // Get a chunk from the space remaining in the linear allocation block.  Do
duke@435:   // not attempt to refill if the space is not available, return NULL.  Do the
duke@435:   // repairs on the linear allocation block as appropriate.
duke@435:   HeapWord*  getChunkFromLinearAllocBlockRemainder(LinearAllocBlock* blk, size_t size);
duke@435:   inline HeapWord*  getChunkFromSmallLinearAllocBlockRemainder(size_t size);
duke@435: 
duke@435:   // Helper function for getChunkFromIndexedFreeList.
duke@435:   // Replenish the indexed free list for this "size".  Do not take from an
duke@435:   // underpopulated size.
ysr@1580:   FreeChunk*  getChunkFromIndexedFreeListHelper(size_t size, bool replenish = true);
duke@435: 
duke@435:   // Get a chunk from the indexed free list.  If the indexed free list
duke@435:   // does not have a free chunk, try to replenish the indexed free list
duke@435:   // then get the free chunk from the replenished indexed free list.
duke@435:   inline FreeChunk* getChunkFromIndexedFreeList(size_t size);
duke@435: 
duke@435:   // The returned chunk may be larger than requested (or null).
duke@435:   FreeChunk* getChunkFromDictionary(size_t size);
duke@435:   // The returned chunk is the exact size requested (or null).
duke@435:   FreeChunk* getChunkFromDictionaryExact(size_t size);
duke@435: 
duke@435:   // Find a chunk in the indexed free list that is the best
duke@435:   // fit for size "numWords".
duke@435:   FreeChunk* bestFitSmall(size_t numWords);
duke@435:   // For free list "fl" of chunks of size > numWords,
duke@435:   // remove a chunk, split off a chunk of size numWords
duke@435:   // and return it.  The split off remainder is returned to
duke@435:   // the free lists.  The old name for getFromListGreater
duke@435:   // was lookInListGreater.
duke@435:   FreeChunk* getFromListGreater(FreeList* fl, size_t numWords);
duke@435:   // Get a chunk in the indexed free list or dictionary,
duke@435:   // by considering a larger chunk and splitting it.
duke@435:   FreeChunk* getChunkFromGreater(size_t numWords);
duke@435:   //  Verify that the given chunk is in the indexed free lists.
duke@435:   bool verifyChunkInIndexedFreeLists(FreeChunk* fc) const;
duke@435:   // Remove the specified chunk from the indexed free lists.
duke@435:   void       removeChunkFromIndexedFreeList(FreeChunk* fc);
duke@435:   // Remove the specified chunk from the dictionary.
duke@435:   void       removeChunkFromDictionary(FreeChunk* fc);
duke@435:   // Split a free chunk into a smaller free chunk of size "new_size".
duke@435:   // Return the smaller free chunk and return the remainder to the
duke@435:   // free lists.
duke@435:   FreeChunk* splitChunkAndReturnRemainder(FreeChunk* chunk, size_t new_size);
duke@435:   // Add a chunk to the free lists.
duke@435:   void       addChunkToFreeLists(HeapWord* chunk, size_t size);
duke@435:   // Add a chunk to the free lists, preferring to suffix it
duke@435:   // to the last free chunk at end of space if possible, and
duke@435:   // updating the block census stats as well as block offset table.
duke@435:   // Take any locks as appropriate if we are multithreaded.
duke@435:   void       addChunkToFreeListsAtEndRecordingStats(HeapWord* chunk, size_t size);
duke@435:   // Add a free chunk to the indexed free lists.
duke@435:   void       returnChunkToFreeList(FreeChunk* chunk);
duke@435:   // Add a free chunk to the dictionary.
duke@435:   void       returnChunkToDictionary(FreeChunk* chunk);
duke@435: 
duke@435:   // Functions for maintaining the linear allocation buffers (LinAB).
duke@435:   // Repairing a linear allocation block refers to operations
duke@435:   // performed on the remainder of a LinAB after an allocation
duke@435:   // has been made from it.
duke@435:   void       repairLinearAllocationBlocks();
duke@435:   void       repairLinearAllocBlock(LinearAllocBlock* blk);
duke@435:   void       refillLinearAllocBlock(LinearAllocBlock* blk);
duke@435:   void       refillLinearAllocBlockIfNeeded(LinearAllocBlock* blk);
duke@435:   void       refillLinearAllocBlocksIfNeeded();
duke@435: 
duke@435:   void       verify_objects_initialized() const;
duke@435: 
duke@435:   // Statistics reporting helper functions
duke@435:   void       reportFreeListStatistics() const;
duke@435:   void       reportIndexedFreeListStatistics() const;
duke@435:   size_t     maxChunkSizeInIndexedFreeLists() const;
duke@435:   size_t     numFreeBlocksInIndexedFreeLists() const;
duke@435:   // Accessor
duke@435:   HeapWord* unallocated_block() const {
ysr@2071:     if (BlockOffsetArrayUseUnallocatedBlock) {
ysr@2071:       HeapWord* ub = _bt.unallocated_block();
ysr@2071:       assert(ub >= bottom() &&
ysr@2071:              ub <= end(), "space invariant");
ysr@2071:       return ub;
ysr@2071:     } else {
ysr@2071:       return end();
ysr@2071:     }
duke@435:   }
duke@435:   void freed(HeapWord* start, size_t size) {
duke@435:     _bt.freed(start, size);
duke@435:   }
duke@435: 
duke@435:  protected:
duke@435:   // reset the indexed free list to its initial empty condition.
duke@435:   void resetIndexedFreeListArray();
duke@435:   // reset to an initial state with a single free block described
duke@435:   // by the MemRegion parameter.
duke@435:   void reset(MemRegion mr);
duke@435:   // Return the total number of words in the indexed free lists.
duke@435:   size_t     totalSizeInIndexedFreeLists() const;
duke@435: 
duke@435:  public:
duke@435:   // Constructor...
duke@435:   CompactibleFreeListSpace(BlockOffsetSharedArray* bs, MemRegion mr,
duke@435:                            bool use_adaptive_freelists,
duke@435:                            FreeBlockDictionary::DictionaryChoice);
duke@435:   // accessors
duke@435:   bool bestFitFirst() { return _fitStrategy == FreeBlockBestFitFirst; }
duke@435:   FreeBlockDictionary* dictionary() const { return _dictionary; }
duke@435:   HeapWord* nearLargestChunk() const { return _nearLargestChunk; }
duke@435:   void set_nearLargestChunk(HeapWord* v) { _nearLargestChunk = v; }
duke@435: 
kvn@1926:   // Set CMS global values
kvn@1926:   static void set_cms_values();
kvn@1926: 
duke@435:   // Return the free chunk at the end of the space.  If no such
duke@435:   // chunk exists, return NULL.
duke@435:   FreeChunk* find_chunk_at_end();
duke@435: 
ysr@447:   bool adaptive_freelists() const { return _adaptive_freelists; }
duke@435: 
duke@435:   void set_collector(CMSCollector* collector) { _collector = collector; }
duke@435: 
duke@435:   // Support for parallelization of rescan and marking
duke@435:   const size_t rescan_task_size()  const { return _rescan_task_size;  }
duke@435:   const size_t marking_task_size() const { return _marking_task_size; }
duke@435:   SequentialSubTasksDone* conc_par_seq_tasks() {return &_conc_par_seq_tasks; }
duke@435:   void initialize_sequential_subtasks_for_rescan(int n_threads);
duke@435:   void initialize_sequential_subtasks_for_marking(int n_threads,
duke@435:          HeapWord* low = NULL);
duke@435: 
duke@435:   // Space enquiries
duke@435:   size_t used() const;
duke@435:   size_t free() const;
duke@435:   size_t max_alloc_in_words() const;
duke@435:   // XXX: should have a less conservative used_region() than that of
duke@435:   // Space; we could consider keeping track of highest allocated
duke@435:   // address and correcting that at each sweep, as the sweeper
duke@435:   // goes through the entire allocated part of the generation. We
duke@435:   // could also use that information to keep the sweeper from
duke@435:   // sweeping more than is necessary. The allocator and sweeper will
duke@435:   // of course need to synchronize on this, since the sweeper will
duke@435:   // try to bump down the address and the allocator will try to bump it up.
duke@435:   // For now, however, we'll just use the default used_region()
duke@435:   // which overestimates the region by returning the entire
duke@435:   // committed region (this is safe, but inefficient).
duke@435: 
duke@435:   // Returns a subregion of the space containing all the objects in
duke@435:   // the space.
duke@435:   MemRegion used_region() const {
duke@435:     return MemRegion(bottom(),
duke@435:                      BlockOffsetArrayUseUnallocatedBlock ?
duke@435:                      unallocated_block() : end());
duke@435:   }
duke@435: 
duke@435:   // This is needed because the default implementation uses block_start()
duke@435:   // which can;t be used at certain times (for example phase 3 of mark-sweep).
duke@435:   // A better fix is to change the assertions in phase 3 of mark-sweep to
duke@435:   // use is_in_reserved(), but that is deferred since the is_in() assertions
duke@435:   // are buried through several layers of callers and are used elsewhere
duke@435:   // as well.
duke@435:   bool is_in(const void* p) const {
duke@435:     return used_region().contains(p);
duke@435:   }
duke@435: 
duke@435:   virtual bool is_free_block(const HeapWord* p) const;
duke@435: 
duke@435:   // Resizing support
duke@435:   void set_end(HeapWord* value);  // override
duke@435: 
duke@435:   // mutual exclusion support
duke@435:   Mutex* freelistLock() const { return &_freelistLock; }
duke@435: 
duke@435:   // Iteration support
duke@435:   void oop_iterate(MemRegion mr, OopClosure* cl);
duke@435:   void oop_iterate(OopClosure* cl);
duke@435: 
duke@435:   void object_iterate(ObjectClosure* blk);
jmasa@952:   // Apply the closure to each object in the space whose references
jmasa@952:   // point to objects in the heap.  The usage of CompactibleFreeListSpace
jmasa@952:   // by the ConcurrentMarkSweepGeneration for concurrent GC's allows
jmasa@952:   // objects in the space with references to objects that are no longer
jmasa@952:   // valid.  For example, an object may reference another object
jmasa@952:   // that has already been sweep up (collected).  This method uses
jmasa@952:   // obj_is_alive() to determine whether it is safe to iterate of
jmasa@952:   // an object.
jmasa@952:   void safe_object_iterate(ObjectClosure* blk);
duke@435:   void object_iterate_mem(MemRegion mr, UpwardsObjectClosure* cl);
duke@435: 
duke@435:   // Requires that "mr" be entirely within the space.
duke@435:   // Apply "cl->do_object" to all objects that intersect with "mr".
duke@435:   // If the iteration encounters an unparseable portion of the region,
duke@435:   // terminate the iteration and return the address of the start of the
duke@435:   // subregion that isn't done.  Return of "NULL" indicates that the
duke@435:   // interation completed.
duke@435:   virtual HeapWord*
duke@435:        object_iterate_careful_m(MemRegion mr,
duke@435:                                 ObjectClosureCareful* cl);
duke@435:   virtual HeapWord*
duke@435:        object_iterate_careful(ObjectClosureCareful* cl);
duke@435: 
duke@435:   // Override: provides a DCTO_CL specific to this kind of space.
duke@435:   DirtyCardToOopClosure* new_dcto_cl(OopClosure* cl,
duke@435:                                      CardTableModRefBS::PrecisionStyle precision,
duke@435:                                      HeapWord* boundary);
duke@435: 
duke@435:   void blk_iterate(BlkClosure* cl);
duke@435:   void blk_iterate_careful(BlkClosureCareful* cl);
ysr@777:   HeapWord* block_start_const(const void* p) const;
duke@435:   HeapWord* block_start_careful(const void* p) const;
duke@435:   size_t block_size(const HeapWord* p) const;
duke@435:   size_t block_size_no_stall(HeapWord* p, const CMSCollector* c) const;
duke@435:   bool block_is_obj(const HeapWord* p) const;
duke@435:   bool obj_is_alive(const HeapWord* p) const;
duke@435:   size_t block_size_nopar(const HeapWord* p) const;
duke@435:   bool block_is_obj_nopar(const HeapWord* p) const;
duke@435: 
duke@435:   // iteration support for promotion
duke@435:   void save_marks();
duke@435:   bool no_allocs_since_save_marks();
duke@435:   void object_iterate_since_last_GC(ObjectClosure* cl);
duke@435: 
duke@435:   // iteration support for sweeping
duke@435:   void save_sweep_limit() {
duke@435:     _sweep_limit = BlockOffsetArrayUseUnallocatedBlock ?
duke@435:                    unallocated_block() : end();
duke@435:   }
duke@435:   NOT_PRODUCT(
duke@435:     void clear_sweep_limit() { _sweep_limit = NULL; }
duke@435:   )
duke@435:   HeapWord* sweep_limit() { return _sweep_limit; }
duke@435: 
duke@435:   // Apply "blk->do_oop" to the addresses of all reference fields in objects
duke@435:   // promoted into this generation since the most recent save_marks() call.
duke@435:   // Fields in objects allocated by applications of the closure
duke@435:   // *are* included in the iteration. Thus, when the iteration completes
duke@435:   // there should be no further such objects remaining.
duke@435:   #define CFLS_OOP_SINCE_SAVE_MARKS_DECL(OopClosureType, nv_suffix)  \
duke@435:     void oop_since_save_marks_iterate##nv_suffix(OopClosureType* blk);
duke@435:   ALL_SINCE_SAVE_MARKS_CLOSURES(CFLS_OOP_SINCE_SAVE_MARKS_DECL)
duke@435:   #undef CFLS_OOP_SINCE_SAVE_MARKS_DECL
duke@435: 
duke@435:   // Allocation support
duke@435:   HeapWord* allocate(size_t size);
duke@435:   HeapWord* par_allocate(size_t size);
duke@435: 
coleenp@548:   oop       promote(oop obj, size_t obj_size);
duke@435:   void      gc_prologue();
duke@435:   void      gc_epilogue();
duke@435: 
duke@435:   // This call is used by a containing CMS generation / collector
duke@435:   // to inform the CFLS space that a sweep has been completed
duke@435:   // and that the space can do any related house-keeping functions.
duke@435:   void      sweep_completed();
duke@435: 
duke@435:   // For an object in this space, the mark-word's two
duke@435:   // LSB's having the value [11] indicates that it has been
duke@435:   // promoted since the most recent call to save_marks() on
duke@435:   // this generation and has not subsequently been iterated
duke@435:   // over (using oop_since_save_marks_iterate() above).
ysr@1876:   // This property holds only for single-threaded collections,
ysr@1876:   // and is typically used for Cheney scans; for MT scavenges,
ysr@1876:   // the property holds for all objects promoted during that
ysr@1876:   // scavenge for the duration of the scavenge and is used
ysr@1876:   // by card-scanning to avoid scanning objects (being) promoted
ysr@1876:   // during that scavenge.
duke@435:   bool obj_allocated_since_save_marks(const oop obj) const {
duke@435:     assert(is_in_reserved(obj), "Wrong space?");
duke@435:     return ((PromotedObject*)obj)->hasPromotedMark();
duke@435:   }
duke@435: 
duke@435:   // A worst-case estimate of the space required (in HeapWords) to expand the
duke@435:   // heap when promoting an obj of size obj_size.
duke@435:   size_t expansionSpaceRequired(size_t obj_size) const;
duke@435: 
duke@435:   FreeChunk* allocateScratch(size_t size);
duke@435: 
duke@435:   // returns true if either the small or large linear allocation buffer is empty.
ysr@447:   bool       linearAllocationWouldFail() const;
duke@435: 
duke@435:   // Adjust the chunk for the minimum size.  This version is called in
duke@435:   // most cases in CompactibleFreeListSpace methods.
duke@435:   inline static size_t adjustObjectSize(size_t size) {
duke@435:     return (size_t) align_object_size(MAX2(size, (size_t)MinChunkSize));
duke@435:   }
duke@435:   // This is a virtual version of adjustObjectSize() that is called
duke@435:   // only occasionally when the compaction space changes and the type
duke@435:   // of the new compaction space is is only known to be CompactibleSpace.
duke@435:   size_t adjust_object_size_v(size_t size) const {
duke@435:     return adjustObjectSize(size);
duke@435:   }
duke@435:   // Minimum size of a free block.
duke@435:   virtual size_t minimum_free_block_size() const { return MinChunkSize; }
duke@435:   void      removeFreeChunkFromFreeLists(FreeChunk* chunk);
duke@435:   void      addChunkAndRepairOffsetTable(HeapWord* chunk, size_t size,
duke@435:               bool coalesced);
duke@435: 
ysr@447:   // Support for decisions regarding concurrent collection policy
ysr@447:   bool should_concurrent_collect() const;
ysr@447: 
duke@435:   // Support for compaction
duke@435:   void prepare_for_compaction(CompactPoint* cp);
duke@435:   void adjust_pointers();
duke@435:   void compact();
duke@435:   // reset the space to reflect the fact that a compaction of the
duke@435:   // space has been done.
duke@435:   virtual void reset_after_compaction();
duke@435: 
duke@435:   // Debugging support
duke@435:   void print()                            const;
ysr@2071:   void print_on(outputStream* st)         const;
duke@435:   void prepare_for_verify();
duke@435:   void verify(bool allow_dirty)           const;
duke@435:   void verifyFreeLists()                  const PRODUCT_RETURN;
duke@435:   void verifyIndexedFreeLists()           const;
duke@435:   void verifyIndexedFreeList(size_t size) const;
duke@435:   // verify that the given chunk is in the free lists.
duke@435:   bool verifyChunkInFreeLists(FreeChunk* fc) const;
duke@435:   // Do some basic checks on the the free lists.
duke@435:   void checkFreeListConsistency()         const PRODUCT_RETURN;
duke@435: 
ysr@1580:   // Printing support
ysr@1580:   void dump_at_safepoint_with_locks(CMSCollector* c, outputStream* st);
ysr@1580:   void print_indexed_free_lists(outputStream* st) const;
ysr@1580:   void print_dictionary_free_lists(outputStream* st) const;
ysr@1580:   void print_promo_info_blocks(outputStream* st) const;
ysr@1580: 
duke@435:   NOT_PRODUCT (
duke@435:     void initializeIndexedFreeListArrayReturnedBytes();
duke@435:     size_t sumIndexedFreeListArrayReturnedBytes();
duke@435:     // Return the total number of chunks in the indexed free lists.
duke@435:     size_t totalCountInIndexedFreeLists() const;
duke@435:     // Return the total numberof chunks in the space.
duke@435:     size_t totalCount();
duke@435:   )
duke@435: 
duke@435:   // The census consists of counts of the quantities such as
duke@435:   // the current count of the free chunks, number of chunks
duke@435:   // created as a result of the split of a larger chunk or
duke@435:   // coalescing of smaller chucks, etc.  The counts in the
duke@435:   // census is used to make decisions on splitting and
duke@435:   // coalescing of chunks during the sweep of garbage.
duke@435: 
duke@435:   // Print the statistics for the free lists.
ysr@447:   void printFLCensus(size_t sweep_count) const;
duke@435: 
duke@435:   // Statistics functions
duke@435:   // Initialize census for lists before the sweep.
ysr@1580:   void beginSweepFLCensus(float inter_sweep_current,
ysr@1580:                           float inter_sweep_estimate,
ysr@1580:                           float intra_sweep_estimate);
duke@435:   // Set the surplus for each of the free lists.
duke@435:   void setFLSurplus();
duke@435:   // Set the hint for each of the free lists.
duke@435:   void setFLHints();
duke@435:   // Clear the census for each of the free lists.
duke@435:   void clearFLCensus();
duke@435:   // Perform functions for the census after the end of the sweep.
ysr@447:   void endSweepFLCensus(size_t sweep_count);
duke@435:   // Return true if the count of free chunks is greater
duke@435:   // than the desired number of free chunks.
duke@435:   bool coalOverPopulated(size_t size);
duke@435: 
duke@435: // Record (for each size):
duke@435: //
duke@435: //   split-births = #chunks added due to splits in (prev-sweep-end,
duke@435: //      this-sweep-start)
duke@435: //   split-deaths = #chunks removed for splits in (prev-sweep-end,
duke@435: //      this-sweep-start)
duke@435: //   num-curr     = #chunks at start of this sweep
duke@435: //   num-prev     = #chunks at end of previous sweep
duke@435: //
duke@435: // The above are quantities that are measured. Now define:
duke@435: //
duke@435: //   num-desired := num-prev + split-births - split-deaths - num-curr
duke@435: //
duke@435: // Roughly, num-prev + split-births is the supply,
duke@435: // split-deaths is demand due to other sizes
duke@435: // and num-curr is what we have left.
duke@435: //
duke@435: // Thus, num-desired is roughly speaking the "legitimate demand"
duke@435: // for blocks of this size and what we are striving to reach at the
duke@435: // end of the current sweep.
duke@435: //
duke@435: // For a given list, let num-len be its current population.
duke@435: // Define, for a free list of a given size:
duke@435: //
duke@435: //   coal-overpopulated := num-len >= num-desired * coal-surplus
duke@435: // (coal-surplus is set to 1.05, i.e. we allow a little slop when
duke@435: // coalescing -- we do not coalesce unless we think that the current
duke@435: // supply has exceeded the estimated demand by more than 5%).
duke@435: //
duke@435: // For the set of sizes in the binary tree, which is neither dense nor
duke@435: // closed, it may be the case that for a particular size we have never
duke@435: // had, or do not now have, or did not have at the previous sweep,
duke@435: // chunks of that size. We need to extend the definition of
duke@435: // coal-overpopulated to such sizes as well:
duke@435: //
duke@435: //   For a chunk in/not in the binary tree, extend coal-overpopulated
duke@435: //   defined above to include all sizes as follows:
duke@435: //
duke@435: //   . a size that is non-existent is coal-overpopulated
duke@435: //   . a size that has a num-desired <= 0 as defined above is
duke@435: //     coal-overpopulated.
duke@435: //
duke@435: // Also define, for a chunk heap-offset C and mountain heap-offset M:
duke@435: //
duke@435: //   close-to-mountain := C >= 0.99 * M
duke@435: //
duke@435: // Now, the coalescing strategy is:
duke@435: //
duke@435: //    Coalesce left-hand chunk with right-hand chunk if and
duke@435: //    only if:
duke@435: //
duke@435: //      EITHER
duke@435: //        . left-hand chunk is of a size that is coal-overpopulated
duke@435: //      OR
duke@435: //        . right-hand chunk is close-to-mountain
duke@435:   void smallCoalBirth(size_t size);
duke@435:   void smallCoalDeath(size_t size);
duke@435:   void coalBirth(size_t size);
duke@435:   void coalDeath(size_t size);
duke@435:   void smallSplitBirth(size_t size);
duke@435:   void smallSplitDeath(size_t size);
duke@435:   void splitBirth(size_t size);
duke@435:   void splitDeath(size_t size);
duke@435:   void split(size_t from, size_t to1);
duke@435: 
duke@435:   double flsFrag() const;
duke@435: };
duke@435: 
duke@435: // A parallel-GC-thread-local allocation buffer for allocation into a
duke@435: // CompactibleFreeListSpace.
duke@435: class CFLS_LAB : public CHeapObj {
duke@435:   // The space that this buffer allocates into.
duke@435:   CompactibleFreeListSpace* _cfls;
duke@435: 
duke@435:   // Our local free lists.
duke@435:   FreeList _indexedFreeList[CompactibleFreeListSpace::IndexSetSize];
duke@435: 
duke@435:   // Initialized from a command-line arg.
duke@435: 
ysr@1580:   // Allocation statistics in support of dynamic adjustment of
ysr@1580:   // #blocks to claim per get_from_global_pool() call below.
ysr@1580:   static AdaptiveWeightedAverage
ysr@1580:                  _blocks_to_claim  [CompactibleFreeListSpace::IndexSetSize];
ysr@1580:   static size_t _global_num_blocks [CompactibleFreeListSpace::IndexSetSize];
ysr@1580:   static int    _global_num_workers[CompactibleFreeListSpace::IndexSetSize];
ysr@1580:   size_t        _num_blocks        [CompactibleFreeListSpace::IndexSetSize];
ysr@1580: 
ysr@1580:   // Internal work method
ysr@1580:   void get_from_global_pool(size_t word_sz, FreeList* fl);
duke@435: 
duke@435: public:
duke@435:   CFLS_LAB(CompactibleFreeListSpace* cfls);
duke@435: 
duke@435:   // Allocate and return a block of the given size, or else return NULL.
duke@435:   HeapWord* alloc(size_t word_sz);
duke@435: 
duke@435:   // Return any unused portions of the buffer to the global pool.
ysr@1580:   void retire(int tid);
ysr@1580: 
ysr@1580:   // Dynamic OldPLABSize sizing
ysr@1580:   static void compute_desired_plab_size();
ysr@1580:   // When the settings are modified from default static initialization
ysr@1580:   static void modify_initialization(size_t n, unsigned wt);
duke@435: };
duke@435: 
duke@435: size_t PromotionInfo::refillSize() const {
duke@435:   const size_t CMSSpoolBlockSize = 256;
duke@435:   const size_t sz = heap_word_size(sizeof(SpoolBlock) + sizeof(markOop)
duke@435:                                    * CMSSpoolBlockSize);
duke@435:   return CompactibleFreeListSpace::adjustObjectSize(sz);
duke@435: }