/*

 * Copyright (c) 2001, 2019, 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_GC_CMS_COMPACTIBLEFREELISTSPACE_HPP

#define SHARE_GC_CMS_COMPACTIBLEFREELISTSPACE_HPP

#include "gc/cms/adaptiveFreeList.hpp"

#include "gc/cms/promotionInfo.hpp"

#include "gc/shared/blockOffsetTable.hpp"

#include "gc/shared/cardTable.hpp"

#include "gc/shared/space.hpp"

#include "logging/log.hpp"

#include "memory/binaryTreeDictionary.hpp"

#include "memory/freeList.hpp"

// Classes in support of keeping track of promotions into a non-Contiguous

// space, in this case a CompactibleFreeListSpace.

// Forward declarations

class CMSCollector;

class CompactibleFreeListSpace;

class ConcurrentMarkSweepGeneration;

class BlkClosure;

class BlkClosureCareful;

class FreeChunk;

class UpwardsObjectClosure;

class ObjectClosureCareful;

class Klass;

class AFLBinaryTreeDictionary : public BinaryTreeDictionary<FreeChunk, AdaptiveFreeList<FreeChunk> > {

 public:

  AFLBinaryTreeDictionary(MemRegion mr)

      : BinaryTreeDictionary<FreeChunk, AdaptiveFreeList<FreeChunk> >(mr) {}

  // Find the list with size "size" in the binary tree and update

  // the statistics in the list according to "split" (chunk was

  // split or coalesce) and "birth" (chunk was added or removed).

  void       dict_census_update(size_t size, bool split, bool birth);

  // Return true if the dictionary is overpopulated (more chunks of

  // this size than desired) for size "size".

  bool       coal_dict_over_populated(size_t size);

  // Methods called at the beginning of a sweep to prepare the

  // statistics for the sweep.

  void       begin_sweep_dict_census(double coalSurplusPercent,

                                     float inter_sweep_current,

                                     float inter_sweep_estimate,

                                     float intra_sweep_estimate);

  // Methods called after the end of a sweep to modify the

  // statistics for the sweep.

  void       end_sweep_dict_census(double splitSurplusPercent);

  // Accessors for statistics

  void       set_tree_surplus(double splitSurplusPercent);

  void       set_tree_hints(void);

  // Reset statistics for all the lists in the tree.

  void       clear_tree_census(void);

  // Print the statistics for all the lists in the tree.  Also may

  // print out summaries.

  void       print_dict_census(outputStream* st) const;

};

class LinearAllocBlock {

 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 CMSCollector;

  // Local alloc buffer for promotion into this space.

  friend class CompactibleFreeListSpaceLAB;

  // Allow scan_and_* functions to call (private) overrides of the auxiliary functions on this class

  template <typename SpaceType>

  friend void CompactibleSpace::scan_and_adjust_pointers(SpaceType* space);

  template <typename SpaceType>

  friend void CompactibleSpace::scan_and_compact(SpaceType* space);

  template <typename SpaceType>

  friend void CompactibleSpace::verify_up_to_first_dead(SpaceType* space);

  template <typename SpaceType>

  friend void CompactibleSpace::scan_and_forward(SpaceType* space, CompactPoint* cp);

  // "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* _old_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;

  static size_t _min_chunk_size_in_bytes;

 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;

  AFLBinaryTreeDictionary* _dictionary;    // Pointer to dictionary for large size blocks

  // Indexed array for small size blocks

  AdaptiveFreeList<FreeChunk> _indexedFreeList[IndexSetSize];

  // Allocation strategy

  bool _fitStrategy;  // Use best fit strategy

  // 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;

  // Stable value of used().

  size_t _used_stable;

  // Used to make the young collector update the mod union table

  MemRegionClosure* _preconsumptionDirtyCardClosure;

  // 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<FreeChunk>* fl);

  // Used by par_get_chunk_of_blocks() for the chunks from the

  // indexed_free_lists.

  bool par_get_chunk_of_blocks_IFL(size_t word_sz, size_t n, AdaptiveFreeList<FreeChunk>* fl);

  // Used by par_get_chunk_of_blocks_dictionary() to get a chunk

  // evenly splittable into "n" "word_sz" chunks.  Returns that

  // evenly splittable chunk.  May split a larger chunk to get the

  // evenly splittable chunk.

  FreeChunk* get_n_way_chunk_to_split(size_t word_sz, size_t n);

  // Used by par_get_chunk_of_blocks() for the chunks from the

  // dictionary.

  void par_get_chunk_of_blocks_dictionary(size_t word_sz, size_t n, AdaptiveFreeList<FreeChunk>* 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);

  // 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<FreeChunk>* 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);

  // 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);

  // Add a free chunk to the dictionary.

  void       returnChunkToDictionary(FreeChunk* chunk);

  // 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 char* title) const;

  void       reportIndexedFreeListStatistics(outputStream* st) 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);

  }

  // Auxiliary functions for scan_and_{forward,adjust_pointers,compact} support.

  // See comments for CompactibleSpace for more information.

  inline HeapWord* scan_limit() const {

    return end();

  }

  inline bool scanned_block_is_obj(const HeapWord* addr) const {

    return CompactibleFreeListSpace::block_is_obj(addr); // Avoid virtual call

  }

  inline size_t scanned_block_size(const HeapWord* addr) const {

    return CompactibleFreeListSpace::block_size(addr); // Avoid virtual call

  }

  inline size_t adjust_obj_size(size_t size) const {

    return adjustObjectSize(size);

  }

  inline size_t obj_size(const HeapWord* addr) const;

 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);

  // Accessors

  bool bestFitFirst() { return _fitStrategy == FreeBlockBestFitFirst; }

  AFLBinaryTreeDictionary* 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();

  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; }

  // Return ergonomic max size for CMSRescanMultiple and CMSConcMarkMultiple.

  const size_t max_flag_size_for_task_size() const;

  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);

  virtual MemRegionClosure* preconsumptionDirtyCardClosure() const {

    return _preconsumptionDirtyCardClosure;

  }

  void setPreconsumptionDirtyCardClosure(MemRegionClosure* cl) {

    _preconsumptionDirtyCardClosure = cl;

  }

  // 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 monotonically increasing stable used space bytes for CMS.

  // This is required for jstat and other memory monitoring tools

  // that might otherwise see inconsistent used space values during a garbage

  // collection, promotion or allocation into compactibleFreeListSpace.

  // The value returned by this function might be smaller than the

  // actual value.

  size_t used_stable() const;

  // Recalculate and cache the current stable used() value. Only to be called

  // in places where we can be sure that the result is stable.

  void recalculate_used_stable();

  // Returns a subregion of the space containing all the objects in

  // the space.

  MemRegion used_region() const {

    return MemRegion(bottom(),

                     BlockOffsetArrayUseUnallocatedBlock ?

                     unallocated_block() : end());

  }

  virtual bool is_free_block(const HeapWord* p) const;

  // Resizing support

  void set_end(HeapWord* value);  // override

  // Never mangle CompactibleFreeListSpace

  void mangle_unused_area() {}

  void mangle_unused_area_complete() {}

  // Mutual exclusion support

  Mutex* freelistLock() const { return &_freelistLock; }

  // Iteration support

  void oop_iterate(OopIterateClosure* 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);

  // Iterate over all objects that intersect with mr, calling "cl->do_object"

  // on each.  There is an exception to this: if this closure has already

  // been invoked on an object, it may skip such objects in some cases.  This is

  // Most likely to happen in an "upwards" (ascending address) iteration of

  // MemRegions.

  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

  // iteration completed.

  HeapWord* object_iterate_careful_m(MemRegion mr,

                                     ObjectClosureCareful* cl);

  // Override: provides a DCTO_CL specific to this kind of space.

  DirtyCardToOopClosure* new_dcto_cl(OopIterateClosure* cl,

                                     CardTable::PrecisionStyle precision,

                                     HeapWord* boundary,

                                     bool parallel);

  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();

  // Iteration support for sweeping

  void save_sweep_limit() {

    _sweep_limit = BlockOffsetArrayUseUnallocatedBlock ?

                   unallocated_block() : end();

    log_develop_trace(gc, sweep)(">>>>> Saving sweep limit " PTR_FORMAT

                                 "  for space [" PTR_FORMAT "," PTR_FORMAT ") <<<<<<",

                                 p2i(_sweep_limit), p2i(bottom()), p2i(end()));

  }