/* * Copyright (c) 2001, 2012, Oracle and/or its affiliates. All rights reserved. * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. * * This code is free software; you can redistribute it and/or modify it * under the terms of the GNU General Public License version 2 only, as * published by the Free Software Foundation. * * This code is distributed in the hope that it will be useful, but WITHOUT * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License * version 2 for more details (a copy is included in the LICENSE file that * accompanied this code). * * You should have received a copy of the GNU General Public License version * 2 along with this work; if not, write to the Free Software Foundation, * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA. * * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA * or visit www.oracle.com if you need additional information or have any * questions. * */ #ifndef SHARE_VM_GC_IMPLEMENTATION_CONCURRENTMARKSWEEP_COMPACTIBLEFREELISTSPACE_HPP #define SHARE_VM_GC_IMPLEMENTATION_CONCURRENTMARKSWEEP_COMPACTIBLEFREELISTSPACE_HPP #include "gc_implementation/concurrentMarkSweep/adaptiveFreeList.hpp" #include "gc_implementation/concurrentMarkSweep/promotionInfo.hpp" #include "memory/binaryTreeDictionary.hpp" #include "memory/blockOffsetTable.inline.hpp" #include "memory/freeList.hpp" #include "memory/space.hpp" // Classes in support of keeping track of promotions into a non-Contiguous // space, in this case a CompactibleFreeListSpace. // Forward declarations class CompactibleFreeListSpace; class BlkClosure; class BlkClosureCareful; class FreeChunk; class UpwardsObjectClosure; class ObjectClosureCareful; class Klass; class LinearAllocBlock VALUE_OBJ_CLASS_SPEC { public: LinearAllocBlock() : _ptr(0), _word_size(0), _refillSize(0), _allocation_size_limit(0) {} void set(HeapWord* ptr, size_t word_size, size_t refill_size, size_t allocation_size_limit) { _ptr = ptr; _word_size = word_size; _refillSize = refill_size; _allocation_size_limit = allocation_size_limit; } HeapWord* _ptr; size_t _word_size; size_t _refillSize; size_t _allocation_size_limit; // largest size that will be allocated void print_on(outputStream* st) const; }; // Concrete subclass of CompactibleSpace that implements // a free list space, such as used in the concurrent mark sweep // generation. class CompactibleFreeListSpace: public CompactibleSpace { friend class VMStructs; friend class ConcurrentMarkSweepGeneration; friend class ASConcurrentMarkSweepGeneration; friend class CMSCollector; // Local alloc buffer for promotion into this space. friend class CFLS_LAB; // "Size" of chunks of work (executed during parallel remark phases // of CMS collection); this probably belongs in CMSCollector, although // it's cached here because it's used in // initialize_sequential_subtasks_for_rescan() which modifies // par_seq_tasks which also lives in Space. XXX const size_t _rescan_task_size; const size_t _marking_task_size; // Yet another sequential tasks done structure. This supports // CMS GC, where we have threads dynamically // claiming sub-tasks from a larger parallel task. SequentialSubTasksDone _conc_par_seq_tasks; BlockOffsetArrayNonContigSpace _bt; CMSCollector* _collector; ConcurrentMarkSweepGeneration* _gen; // Data structures for free blocks (used during allocation/sweeping) // Allocation is done linearly from two different blocks depending on // whether the request is small or large, in an effort to reduce // fragmentation. We assume that any locking for allocation is done // by the containing generation. Thus, none of the methods in this // space are re-entrant. enum SomeConstants { SmallForLinearAlloc = 16, // size < this then use _sLAB SmallForDictionary = 257, // size < this then use _indexedFreeList IndexSetSize = SmallForDictionary // keep this odd-sized }; static size_t IndexSetStart; static size_t IndexSetStride; private: enum FitStrategyOptions { FreeBlockStrategyNone = 0, FreeBlockBestFitFirst }; PromotionInfo _promoInfo; // helps to impose a global total order on freelistLock ranks; // assumes that CFLSpace's are allocated in global total order static int _lockRank; // a lock protecting the free lists and free blocks; // mutable because of ubiquity of locking even for otherwise const methods mutable Mutex _freelistLock; // locking verifier convenience function void assert_locked() const PRODUCT_RETURN; void assert_locked(const Mutex* lock) const PRODUCT_RETURN; // Linear allocation blocks LinearAllocBlock _smallLinearAllocBlock; FreeBlockDictionary::DictionaryChoice _dictionaryChoice; FreeBlockDictionary* _dictionary; // ptr to dictionary for large size blocks AdaptiveFreeList _indexedFreeList[IndexSetSize]; // indexed array for small size blocks // allocation stategy bool _fitStrategy; // Use best fit strategy. bool _adaptive_freelists; // Use adaptive freelists // This is an address close to the largest free chunk in the heap. // It is currently assumed to be at the end of the heap. Free // chunks with addresses greater than nearLargestChunk are coalesced // in an effort to maintain a large chunk at the end of the heap. HeapWord* _nearLargestChunk; // Used to keep track of limit of sweep for the space HeapWord* _sweep_limit; // Support for compacting cms HeapWord* cross_threshold(HeapWord* start, HeapWord* end); HeapWord* forward(oop q, size_t size, CompactPoint* cp, HeapWord* compact_top); // Initialization helpers. void initializeIndexedFreeListArray(); // Extra stuff to manage promotion parallelism. // a lock protecting the dictionary during par promotion allocation. mutable Mutex _parDictionaryAllocLock; Mutex* parDictionaryAllocLock() const { return &_parDictionaryAllocLock; } // Locks protecting the exact lists during par promotion allocation. Mutex* _indexedFreeListParLocks[IndexSetSize]; // Attempt to obtain up to "n" blocks of the size "word_sz" (which is // required to be smaller than "IndexSetSize".) If successful, // adds them to "fl", which is required to be an empty free list. // If the count of "fl" is negative, it's absolute value indicates a // number of free chunks that had been previously "borrowed" from global // list of size "word_sz", and must now be decremented. void par_get_chunk_of_blocks(size_t word_sz, size_t n, AdaptiveFreeList* fl); // Allocation helper functions // Allocate using a strategy that takes from the indexed free lists // first. This allocation strategy assumes a companion sweeping // strategy that attempts to keep the needed number of chunks in each // indexed free lists. HeapWord* allocate_adaptive_freelists(size_t size); // Allocate from the linear allocation buffers first. This allocation // strategy assumes maximal coalescing can maintain chunks large enough // to be used as linear allocation buffers. HeapWord* allocate_non_adaptive_freelists(size_t size); // Gets a chunk from the linear allocation block (LinAB). If there // is not enough space in the LinAB, refills it. HeapWord* getChunkFromLinearAllocBlock(LinearAllocBlock* blk, size_t size); HeapWord* getChunkFromSmallLinearAllocBlock(size_t size); // Get a chunk from the space remaining in the linear allocation block. Do // not attempt to refill if the space is not available, return NULL. Do the // repairs on the linear allocation block as appropriate. HeapWord* getChunkFromLinearAllocBlockRemainder(LinearAllocBlock* blk, size_t size); inline HeapWord* getChunkFromSmallLinearAllocBlockRemainder(size_t size); // Helper function for getChunkFromIndexedFreeList. // Replenish the indexed free list for this "size". Do not take from an // underpopulated size. FreeChunk* getChunkFromIndexedFreeListHelper(size_t size, bool replenish = true); // Get a chunk from the indexed free list. If the indexed free list // does not have a free chunk, try to replenish the indexed free list // then get the free chunk from the replenished indexed free list. inline FreeChunk* getChunkFromIndexedFreeList(size_t size); // The returned chunk may be larger than requested (or null). FreeChunk* getChunkFromDictionary(size_t size); // The returned chunk is the exact size requested (or null). FreeChunk* getChunkFromDictionaryExact(size_t size); // Find a chunk in the indexed free list that is the best // fit for size "numWords". FreeChunk* bestFitSmall(size_t numWords); // For free list "fl" of chunks of size > numWords, // remove a chunk, split off a chunk of size numWords // and return it. The split off remainder is returned to // the free lists. The old name for getFromListGreater // was lookInListGreater. FreeChunk* getFromListGreater(AdaptiveFreeList* fl, size_t numWords); // Get a chunk in the indexed free list or dictionary, // by considering a larger chunk and splitting it. FreeChunk* getChunkFromGreater(size_t numWords); // Verify that the given chunk is in the indexed free lists. bool verifyChunkInIndexedFreeLists(FreeChunk* fc) const; // Remove the specified chunk from the indexed free lists. void removeChunkFromIndexedFreeList(FreeChunk* fc); // Remove the specified chunk from the dictionary. void removeChunkFromDictionary(FreeChunk* fc); // Split a free chunk into a smaller free chunk of size "new_size". // Return the smaller free chunk and return the remainder to the // free lists. FreeChunk* splitChunkAndReturnRemainder(FreeChunk* chunk, size_t new_size); // Add a chunk to the free lists. void addChunkToFreeLists(HeapWord* chunk, size_t size, bool deallocate_pages); // Add a chunk to the free lists, preferring to suffix it // to the last free chunk at end of space if possible, and // updating the block census stats as well as block offset table. // Take any locks as appropriate if we are multithreaded. void addChunkToFreeListsAtEndRecordingStats(HeapWord* chunk, size_t size); // Add a free chunk to the indexed free lists. void returnChunkToFreeList(FreeChunk* chunk, bool deallocate_pages); // Add a free chunk to the dictionary. void returnChunkToDictionary(FreeChunk* chunk, bool deallocate_pages); // Functions for maintaining the linear allocation buffers (LinAB). // Repairing a linear allocation block refers to operations // performed on the remainder of a LinAB after an allocation // has been made from it. void repairLinearAllocationBlocks(); void repairLinearAllocBlock(LinearAllocBlock* blk); void refillLinearAllocBlock(LinearAllocBlock* blk); void refillLinearAllocBlockIfNeeded(LinearAllocBlock* blk); void refillLinearAllocBlocksIfNeeded(); void verify_objects_initialized() const; // Statistics reporting helper functions void reportFreeListStatistics() const; void reportIndexedFreeListStatistics() const; size_t maxChunkSizeInIndexedFreeLists() const; size_t numFreeBlocksInIndexedFreeLists() const; // Accessor HeapWord* unallocated_block() const { if (BlockOffsetArrayUseUnallocatedBlock) { HeapWord* ub = _bt.unallocated_block(); assert(ub >= bottom() && ub <= end(), "space invariant"); return ub; } else { return end(); } } void freed(HeapWord* start, size_t size) { _bt.freed(start, size); } protected: // reset the indexed free list to its initial empty condition. void resetIndexedFreeListArray(); // reset to an initial state with a single free block described // by the MemRegion parameter. void reset(MemRegion mr); // Return the total number of words in the indexed free lists. size_t totalSizeInIndexedFreeLists() const; public: // Constructor... CompactibleFreeListSpace(BlockOffsetSharedArray* bs, MemRegion mr, bool use_adaptive_freelists, FreeBlockDictionary::DictionaryChoice); // accessors bool bestFitFirst() { return _fitStrategy == FreeBlockBestFitFirst; } FreeBlockDictionary* dictionary() const { return _dictionary; } HeapWord* nearLargestChunk() const { return _nearLargestChunk; } void set_nearLargestChunk(HeapWord* v) { _nearLargestChunk = v; } // Set CMS global values static void set_cms_values(); // Return the free chunk at the end of the space. If no such // chunk exists, return NULL. FreeChunk* find_chunk_at_end(); bool adaptive_freelists() const { return _adaptive_freelists; } void set_collector(CMSCollector* collector) { _collector = collector; } // Support for parallelization of rescan and marking const size_t rescan_task_size() const { return _rescan_task_size; } const size_t marking_task_size() const { return _marking_task_size; } SequentialSubTasksDone* conc_par_seq_tasks() {return &_conc_par_seq_tasks; } void initialize_sequential_subtasks_for_rescan(int n_threads); void initialize_sequential_subtasks_for_marking(int n_threads, HeapWord* low = NULL); // Space enquiries size_t used() const; size_t free() const; size_t max_alloc_in_words() const; // XXX: should have a less conservative used_region() than that of // Space; we could consider keeping track of highest allocated // address and correcting that at each sweep, as the sweeper // goes through the entire allocated part of the generation. We // could also use that information to keep the sweeper from // sweeping more than is necessary. The allocator and sweeper will // of course need to synchronize on this, since the sweeper will // try to bump down the address and the allocator will try to bump it up. // For now, however, we'll just use the default used_region() // which overestimates the region by returning the entire // committed region (this is safe, but inefficient). // Returns a subregion of the space containing all the objects in // the space. MemRegion used_region() const { return MemRegion(bottom(), BlockOffsetArrayUseUnallocatedBlock ? unallocated_block() : end()); } bool is_in(const void* p) const { return used_region().contains(p); } virtual bool is_free_block(const HeapWord* p) const; // Resizing support void set_end(HeapWord* value); // override // mutual exclusion support Mutex* freelistLock() const { return &_freelistLock; } // Iteration support void oop_iterate(MemRegion mr, ExtendedOopClosure* cl); void oop_iterate(ExtendedOopClosure* cl); void object_iterate(ObjectClosure* blk); // Apply the closure to each object in the space whose references // point to objects in the heap. The usage of CompactibleFreeListSpace // by the ConcurrentMarkSweepGeneration for concurrent GC's allows // objects in the space with references to objects that are no longer // valid. For example, an object may reference another object // that has already been sweep up (collected). This method uses // obj_is_alive() to determine whether it is safe to iterate of // an object. void safe_object_iterate(ObjectClosure* blk); void object_iterate_mem(MemRegion mr, UpwardsObjectClosure* cl); // Requires that "mr" be entirely within the space. // Apply "cl->do_object" to all objects that intersect with "mr". // If the iteration encounters an unparseable portion of the region, // terminate the iteration and return the address of the start of the // subregion that isn't done. Return of "NULL" indicates that the // interation completed. virtual HeapWord* object_iterate_careful_m(MemRegion mr, ObjectClosureCareful* cl); virtual HeapWord* object_iterate_careful(ObjectClosureCareful* cl); // Override: provides a DCTO_CL specific to this kind of space. DirtyCardToOopClosure* new_dcto_cl(ExtendedOopClosure* cl, CardTableModRefBS::PrecisionStyle precision, HeapWord* boundary); void blk_iterate(BlkClosure* cl); void blk_iterate_careful(BlkClosureCareful* cl); HeapWord* block_start_const(const void* p) const; HeapWord* block_start_careful(const void* p) const; size_t block_size(const HeapWord* p) const; size_t block_size_no_stall(HeapWord* p, const CMSCollector* c) const; bool block_is_obj(const HeapWord* p) const; bool obj_is_alive(const HeapWord* p) const; size_t block_size_nopar(const HeapWord* p) const; bool block_is_obj_nopar(const HeapWord* p) const; // iteration support for promotion void save_marks(); bool no_allocs_since_save_marks(); void object_iterate_since_last_GC(ObjectClosure* cl); // iteration support for sweeping void save_sweep_limit() { _sweep_limit = BlockOffsetArrayUseUnallocatedBlock ? unallocated_block() : end(); if (CMSTraceSweeper) { gclog_or_tty->print_cr(">>>>> Saving sweep limit " PTR_FORMAT " for space [" PTR_FORMAT "," PTR_FORMAT ") <<<<<<", _sweep_limit, bottom(), end()); } } NOT_PRODUCT( void clear_sweep_limit() { _sweep_limit = NULL; } ) HeapWord* sweep_limit() { return _sweep_limit; } // Apply "blk->do_oop" to the addresses of all reference fields in objects // promoted into this generation since the most recent save_marks() call. // Fields in objects allocated by applications of the closure // *are* included in the iteration. Thus, when the iteration completes // there should be no further such objects remaining. #define CFLS_OOP_SINCE_SAVE_MARKS_DECL(OopClosureType, nv_suffix) \ void oop_since_save_marks_iterate##nv_suffix(OopClosureType* blk); ALL_SINCE_SAVE_MARKS_CLOSURES(CFLS_OOP_SINCE_SAVE_MARKS_DECL) #undef CFLS_OOP_SINCE_SAVE_MARKS_DECL // Allocation support HeapWord* allocate(size_t size); HeapWord* par_allocate(size_t size); oop promote(oop obj, size_t obj_size); void gc_prologue(); void gc_epilogue(); // This call is used by a containing CMS generation / collector // to inform the CFLS space that a sweep has been completed // and that the space can do any related house-keeping functions. void sweep_completed(); // For an object in this space, the mark-word's two // LSB's having the value [11] indicates that it has been // promoted since the most recent call to save_marks() on // this generation and has not subsequently been iterated // over (using oop_since_save_marks_iterate() above). // This property holds only for single-threaded collections, // and is typically used for Cheney scans; for MT scavenges, // the property holds for all objects promoted during that // scavenge for the duration of the scavenge and is used // by card-scanning to avoid scanning objects (being) promoted // during that scavenge. bool obj_allocated_since_save_marks(const oop obj) const { assert(is_in_reserved(obj), "Wrong space?"); return ((PromotedObject*)obj)->hasPromotedMark(); } // A worst-case estimate of the space required (in HeapWords) to expand the // heap when promoting an obj of size obj_size. size_t expansionSpaceRequired(size_t obj_size) const; FreeChunk* allocateScratch(size_t size); // returns true if either the small or large linear allocation buffer is empty. bool linearAllocationWouldFail() const; // Adjust the chunk for the minimum size. This version is called in // most cases in CompactibleFreeListSpace methods. inline static size_t adjustObjectSize(size_t size) { return (size_t) align_object_size(MAX2(size, (size_t)MinChunkSize)); } // This is a virtual version of adjustObjectSize() that is called // only occasionally when the compaction space changes and the type // of the new compaction space is is only known to be CompactibleSpace. size_t adjust_object_size_v(size_t size) const { return adjustObjectSize(size); } // Minimum size of a free block. virtual size_t minimum_free_block_size() const { return MinChunkSize; } void removeFreeChunkFromFreeLists(FreeChunk* chunk); void addChunkAndRepairOffsetTable(HeapWord* chunk, size_t size, bool coalesced, bool deallocate_pages); // Support for decisions regarding concurrent collection policy bool should_concurrent_collect() const; // Support for compaction void prepare_for_compaction(CompactPoint* cp); void adjust_pointers(); void compact(); // reset the space to reflect the fact that a compaction of the // space has been done. virtual void reset_after_compaction(); // Debugging support void print() const; void print_on(outputStream* st) const; void prepare_for_verify(); void verify() const; void verifyFreeLists() const PRODUCT_RETURN; void verifyIndexedFreeLists() const; void verifyIndexedFreeList(size_t size) const; // Verify that the given chunk is in the free lists: // i.e. either the binary tree dictionary, the indexed free lists // or the linear allocation block. bool verify_chunk_in_free_list(FreeChunk* fc) const; // Verify that the given chunk is the linear allocation block bool verify_chunk_is_linear_alloc_block(FreeChunk* fc) const; // Do some basic checks on the the free lists. void check_free_list_consistency() const PRODUCT_RETURN; // Printing support void dump_at_safepoint_with_locks(CMSCollector* c, outputStream* st); void print_indexed_free_lists(outputStream* st) const; void print_dictionary_free_lists(outputStream* st) const; void print_promo_info_blocks(outputStream* st) const; NOT_PRODUCT ( void initializeIndexedFreeListArrayReturnedBytes(); size_t sumIndexedFreeListArrayReturnedBytes(); // Return the total number of chunks in the indexed free lists. size_t totalCountInIndexedFreeLists() const; // Return the total numberof chunks in the space. size_t totalCount(); ) // The census consists of counts of the quantities such as // the current count of the free chunks, number of chunks // created as a result of the split of a larger chunk or // coalescing of smaller chucks, etc. The counts in the // census is used to make decisions on splitting and // coalescing of chunks during the sweep of garbage. // Print the statistics for the free lists. void printFLCensus(size_t sweep_count) const; // Statistics functions // Initialize census for lists before the sweep. void beginSweepFLCensus(float inter_sweep_current, float inter_sweep_estimate, float intra_sweep_estimate); // Set the surplus for each of the free lists. void setFLSurplus(); // Set the hint for each of the free lists. void setFLHints(); // Clear the census for each of the free lists. void clearFLCensus(); // Perform functions for the census after the end of the sweep. void endSweepFLCensus(size_t sweep_count); // Return true if the count of free chunks is greater // than the desired number of free chunks. bool coalOverPopulated(size_t size); // Record (for each size): // // split-births = #chunks added due to splits in (prev-sweep-end, // this-sweep-start) // split-deaths = #chunks removed for splits in (prev-sweep-end, // this-sweep-start) // num-curr = #chunks at start of this sweep // num-prev = #chunks at end of previous sweep // // The above are quantities that are measured. Now define: // // num-desired := num-prev + split-births - split-deaths - num-curr // // Roughly, num-prev + split-births is the supply, // split-deaths is demand due to other sizes // and num-curr is what we have left. // // Thus, num-desired is roughly speaking the "legitimate demand" // for blocks of this size and what we are striving to reach at the // end of the current sweep. // // For a given list, let num-len be its current population. // Define, for a free list of a given size: // // coal-overpopulated := num-len >= num-desired * coal-surplus // (coal-surplus is set to 1.05, i.e. we allow a little slop when // coalescing -- we do not coalesce unless we think that the current // supply has exceeded the estimated demand by more than 5%). // // For the set of sizes in the binary tree, which is neither dense nor // closed, it may be the case that for a particular size we have never // had, or do not now have, or did not have at the previous sweep, // chunks of that size. We need to extend the definition of // coal-overpopulated to such sizes as well: // // For a chunk in/not in the binary tree, extend coal-overpopulated // defined above to include all sizes as follows: // // . a size that is non-existent is coal-overpopulated // . a size that has a num-desired <= 0 as defined above is // coal-overpopulated. // // Also define, for a chunk heap-offset C and mountain heap-offset M: // // close-to-mountain := C >= 0.99 * M // // Now, the coalescing strategy is: // // Coalesce left-hand chunk with right-hand chunk if and // only if: // // EITHER // . left-hand chunk is of a size that is coal-overpopulated // OR // . right-hand chunk is close-to-mountain void smallCoalBirth(size_t size); void smallCoalDeath(size_t size); void coalBirth(size_t size); void coalDeath(size_t size); void smallSplitBirth(size_t size); void smallSplitDeath(size_t size); void split_birth(size_t size); void splitDeath(size_t size); void split(size_t from, size_t to1); double flsFrag() const; }; // A parallel-GC-thread-local allocation buffer for allocation into a // CompactibleFreeListSpace. class CFLS_LAB : public CHeapObj { // The space that this buffer allocates into. CompactibleFreeListSpace* _cfls; // Our local free lists. AdaptiveFreeList _indexedFreeList[CompactibleFreeListSpace::IndexSetSize]; // Initialized from a command-line arg. // Allocation statistics in support of dynamic adjustment of // #blocks to claim per get_from_global_pool() call below. static AdaptiveWeightedAverage _blocks_to_claim [CompactibleFreeListSpace::IndexSetSize]; static size_t _global_num_blocks [CompactibleFreeListSpace::IndexSetSize]; static uint _global_num_workers[CompactibleFreeListSpace::IndexSetSize]; size_t _num_blocks [CompactibleFreeListSpace::IndexSetSize]; // Internal work method void get_from_global_pool(size_t word_sz, AdaptiveFreeList* fl); public: CFLS_LAB(CompactibleFreeListSpace* cfls); // Allocate and return a block of the given size, or else return NULL. HeapWord* alloc(size_t word_sz); // Return any unused portions of the buffer to the global pool. void retire(int tid); // Dynamic OldPLABSize sizing static void compute_desired_plab_size(); // When the settings are modified from default static initialization static void modify_initialization(size_t n, unsigned wt); }; size_t PromotionInfo::refillSize() const { const size_t CMSSpoolBlockSize = 256; const size_t sz = heap_word_size(sizeof(SpoolBlock) + sizeof(markOop) * CMSSpoolBlockSize); return CompactibleFreeListSpace::adjustObjectSize(sz); } #endif // SHARE_VM_GC_IMPLEMENTATION_CONCURRENTMARKSWEEP_COMPACTIBLEFREELISTSPACE_HPP