1 /*
   2  * Copyright (c) 2005, 2015, 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_VM_GC_IMPLEMENTATION_PARALLELSCAVENGE_PSPARALLELCOMPACT_HPP
  26 #define SHARE_VM_GC_IMPLEMENTATION_PARALLELSCAVENGE_PSPARALLELCOMPACT_HPP
  27 
  28 #include "gc_implementation/parallelScavenge/objectStartArray.hpp"
  29 #include "gc_implementation/parallelScavenge/parMarkBitMap.hpp"
  30 #include "gc_implementation/parallelScavenge/psCompactionManager.hpp"
  31 #include "gc_implementation/shared/collectorCounters.hpp"
  32 #include "gc_implementation/shared/mutableSpace.hpp"
  33 #include "gc_interface/collectedHeap.hpp"
  34 #include "oops/oop.hpp"
  35 
  36 class ParallelScavengeHeap;
  37 class PSAdaptiveSizePolicy;
  38 class PSYoungGen;
  39 class PSOldGen;
  40 class ParCompactionManager;
  41 class ParallelTaskTerminator;
  42 class PSParallelCompact;
  43 class GCTaskManager;
  44 class GCTaskQueue;
  45 class PreGCValues;
  46 class MoveAndUpdateClosure;
  47 class RefProcTaskExecutor;
  48 class ParallelOldTracer;
  49 class STWGCTimer;
  50 
  51 // The SplitInfo class holds the information needed to 'split' a source region
  52 // so that the live data can be copied to two destination *spaces*.  Normally,
  53 // all the live data in a region is copied to a single destination space (e.g.,
  54 // everything live in a region in eden is copied entirely into the old gen).
  55 // However, when the heap is nearly full, all the live data in eden may not fit
  56 // into the old gen.  Copying only some of the regions from eden to old gen
  57 // requires finding a region that does not contain a partial object (i.e., no
  58 // live object crosses the region boundary) somewhere near the last object that
  59 // does fit into the old gen.  Since it's not always possible to find such a
  60 // region, splitting is necessary for predictable behavior.
  61 //
  62 // A region is always split at the end of the partial object.  This avoids
  63 // additional tests when calculating the new location of a pointer, which is a
  64 // very hot code path.  The partial object and everything to its left will be
  65 // copied to another space (call it dest_space_1).  The live data to the right
  66 // of the partial object will be copied either within the space itself, or to a
  67 // different destination space (distinct from dest_space_1).
  68 //
  69 // Split points are identified during the summary phase, when region
  70 // destinations are computed:  data about the split, including the
  71 // partial_object_size, is recorded in a SplitInfo record and the
  72 // partial_object_size field in the summary data is set to zero.  The zeroing is
  73 // possible (and necessary) since the partial object will move to a different
  74 // destination space than anything to its right, thus the partial object should
  75 // not affect the locations of any objects to its right.
  76 //
  77 // The recorded data is used during the compaction phase, but only rarely:  when
  78 // the partial object on the split region will be copied across a destination
  79 // region boundary.  This test is made once each time a region is filled, and is
  80 // a simple address comparison, so the overhead is negligible (see
  81 // PSParallelCompact::first_src_addr()).
  82 //
  83 // Notes:
  84 //
  85 // Only regions with partial objects are split; a region without a partial
  86 // object does not need any extra bookkeeping.
  87 //
  88 // At most one region is split per space, so the amount of data required is
  89 // constant.
  90 //
  91 // A region is split only when the destination space would overflow.  Once that
  92 // happens, the destination space is abandoned and no other data (even from
  93 // other source spaces) is targeted to that destination space.  Abandoning the
  94 // destination space may leave a somewhat large unused area at the end, if a
  95 // large object caused the overflow.
  96 //
  97 // Future work:
  98 //
  99 // More bookkeeping would be required to continue to use the destination space.
 100 // The most general solution would allow data from regions in two different
 101 // source spaces to be "joined" in a single destination region.  At the very
 102 // least, additional code would be required in next_src_region() to detect the
 103 // join and skip to an out-of-order source region.  If the join region was also
 104 // the last destination region to which a split region was copied (the most
 105 // likely case), then additional work would be needed to get fill_region() to
 106 // stop iteration and switch to a new source region at the right point.  Basic
 107 // idea would be to use a fake value for the top of the source space.  It is
 108 // doable, if a bit tricky.
 109 //
 110 // A simpler (but less general) solution would fill the remainder of the
 111 // destination region with a dummy object and continue filling the next
 112 // destination region.
 113 
 114 class SplitInfo
 115 {
 116 public:
 117   // Return true if this split info is valid (i.e., if a split has been
 118   // recorded).  The very first region cannot have a partial object and thus is
 119   // never split, so 0 is the 'invalid' value.
 120   bool is_valid() const { return _src_region_idx > 0; }
 121 
 122   // Return true if this split holds data for the specified source region.
 123   inline bool is_split(size_t source_region) const;
 124 
 125   // The index of the split region, the size of the partial object on that
 126   // region and the destination of the partial object.
 127   size_t    src_region_idx() const   { return _src_region_idx; }
 128   size_t    partial_obj_size() const { return _partial_obj_size; }
 129   HeapWord* destination() const      { return _destination; }
 130 
 131   // The destination count of the partial object referenced by this split
 132   // (either 1 or 2).  This must be added to the destination count of the
 133   // remainder of the source region.
 134   unsigned int destination_count() const { return _destination_count; }
 135 
 136   // If a word within the partial object will be written to the first word of a
 137   // destination region, this is the address of the destination region;
 138   // otherwise this is NULL.
 139   HeapWord* dest_region_addr() const     { return _dest_region_addr; }
 140 
 141   // If a word within the partial object will be written to the first word of a
 142   // destination region, this is the address of that word within the partial
 143   // object; otherwise this is NULL.
 144   HeapWord* first_src_addr() const       { return _first_src_addr; }
 145 
 146   // Record the data necessary to split the region src_region_idx.
 147   void record(size_t src_region_idx, size_t partial_obj_size,
 148               HeapWord* destination);
 149 
 150   void clear();
 151 
 152   DEBUG_ONLY(void verify_clear();)
 153 
 154 private:
 155   size_t       _src_region_idx;
 156   size_t       _partial_obj_size;
 157   HeapWord*    _destination;
 158   unsigned int _destination_count;
 159   HeapWord*    _dest_region_addr;
 160   HeapWord*    _first_src_addr;
 161 };
 162 
 163 inline bool SplitInfo::is_split(size_t region_idx) const
 164 {
 165   return _src_region_idx == region_idx && is_valid();
 166 }
 167 
 168 class SpaceInfo
 169 {
 170  public:
 171   MutableSpace* space() const { return _space; }
 172 
 173   // Where the free space will start after the collection.  Valid only after the
 174   // summary phase completes.
 175   HeapWord* new_top() const { return _new_top; }
 176 
 177   // Allows new_top to be set.
 178   HeapWord** new_top_addr() { return &_new_top; }
 179 
 180   // Where the smallest allowable dense prefix ends (used only for perm gen).
 181   HeapWord* min_dense_prefix() const { return _min_dense_prefix; }
 182 
 183   // Where the dense prefix ends, or the compacted region begins.
 184   HeapWord* dense_prefix() const { return _dense_prefix; }
 185 
 186   // The start array for the (generation containing the) space, or NULL if there
 187   // is no start array.
 188   ObjectStartArray* start_array() const { return _start_array; }
 189 
 190   SplitInfo& split_info() { return _split_info; }
 191 
 192   void set_space(MutableSpace* s)           { _space = s; }
 193   void set_new_top(HeapWord* addr)          { _new_top = addr; }
 194   void set_min_dense_prefix(HeapWord* addr) { _min_dense_prefix = addr; }
 195   void set_dense_prefix(HeapWord* addr)     { _dense_prefix = addr; }
 196   void set_start_array(ObjectStartArray* s) { _start_array = s; }
 197 
 198   void publish_new_top() const              { _space->set_top(_new_top); }
 199 
 200  private:
 201   MutableSpace*     _space;
 202   HeapWord*         _new_top;
 203   HeapWord*         _min_dense_prefix;
 204   HeapWord*         _dense_prefix;
 205   ObjectStartArray* _start_array;
 206   SplitInfo         _split_info;
 207 };
 208 
 209 class ParallelCompactData
 210 {
 211 public:
 212   // Sizes are in HeapWords, unless indicated otherwise.
 213   static const size_t Log2RegionSize;
 214   static const size_t RegionSize;
 215   static const size_t RegionSizeBytes;
 216 
 217   // Mask for the bits in a size_t to get an offset within a region.
 218   static const size_t RegionSizeOffsetMask;
 219   // Mask for the bits in a pointer to get an offset within a region.
 220   static const size_t RegionAddrOffsetMask;
 221   // Mask for the bits in a pointer to get the address of the start of a region.
 222   static const size_t RegionAddrMask;
 223 
 224   static const size_t Log2BlockSize;
 225   static const size_t BlockSize;
 226   static const size_t BlockSizeBytes;
 227 
 228   static const size_t BlockSizeOffsetMask;
 229   static const size_t BlockAddrOffsetMask;
 230   static const size_t BlockAddrMask;
 231 
 232   static const size_t BlocksPerRegion;
 233   static const size_t Log2BlocksPerRegion;
 234 
 235   class RegionData
 236   {
 237   public:
 238     // Destination address of the region.
 239     HeapWord* destination() const { return _destination; }
 240 
 241     // The first region containing data destined for this region.
 242     size_t source_region() const { return _source_region; }
 243 
 244     // The object (if any) starting in this region and ending in a different
 245     // region that could not be updated during the main (parallel) compaction
 246     // phase.  This is different from _partial_obj_addr, which is an object that
 247     // extends onto a source region.  However, the two uses do not overlap in
 248     // time, so the same field is used to save space.
 249     HeapWord* deferred_obj_addr() const { return _partial_obj_addr; }
 250 
 251     // The starting address of the partial object extending onto the region.
 252     HeapWord* partial_obj_addr() const { return _partial_obj_addr; }
 253 
 254     // Size of the partial object extending onto the region (words).
 255     size_t partial_obj_size() const { return _partial_obj_size; }
 256 
 257     // Size of live data that lies within this region due to objects that start
 258     // in this region (words).  This does not include the partial object
 259     // extending onto the region (if any), or the part of an object that extends
 260     // onto the next region (if any).
 261     size_t live_obj_size() const { return _dc_and_los & los_mask; }
 262 
 263     // Total live data that lies within the region (words).
 264     size_t data_size() const { return partial_obj_size() + live_obj_size(); }
 265 
 266     // The destination_count is the number of other regions to which data from
 267     // this region will be copied.  At the end of the summary phase, the valid
 268     // values of destination_count are
 269     //
 270     // 0 - data from the region will be compacted completely into itself, or the
 271     //     region is empty.  The region can be claimed and then filled.
 272     // 1 - data from the region will be compacted into 1 other region; some
 273     //     data from the region may also be compacted into the region itself.
 274     // 2 - data from the region will be copied to 2 other regions.
 275     //
 276     // During compaction as regions are emptied, the destination_count is
 277     // decremented (atomically) and when it reaches 0, it can be claimed and
 278     // then filled.
 279     //
 280     // A region is claimed for processing by atomically changing the
 281     // destination_count to the claimed value (dc_claimed).  After a region has
 282     // been filled, the destination_count should be set to the completed value
 283     // (dc_completed).
 284     inline uint destination_count() const;
 285     inline uint destination_count_raw() const;
 286 
 287     // Whether the block table for this region has been filled.
 288     inline bool blocks_filled() const;
 289 
 290     // Number of times the block table was filled.
 291     DEBUG_ONLY(inline size_t blocks_filled_count() const;)
 292 
 293     // The location of the java heap data that corresponds to this region.
 294     inline HeapWord* data_location() const;
 295 
 296     // The highest address referenced by objects in this region.
 297     inline HeapWord* highest_ref() const;
 298 
 299     // Whether this region is available to be claimed, has been claimed, or has
 300     // been completed.
 301     //
 302     // Minor subtlety:  claimed() returns true if the region is marked
 303     // completed(), which is desirable since a region must be claimed before it
 304     // can be completed.
 305     bool available() const { return _dc_and_los < dc_one; }
 306     bool claimed() const   { return _dc_and_los >= dc_claimed; }
 307     bool completed() const { return _dc_and_los >= dc_completed; }
 308 
 309     // These are not atomic.
 310     void set_destination(HeapWord* addr)       { _destination = addr; }
 311     void set_source_region(size_t region)      { _source_region = region; }
 312     void set_deferred_obj_addr(HeapWord* addr) { _partial_obj_addr = addr; }
 313     void set_partial_obj_addr(HeapWord* addr)  { _partial_obj_addr = addr; }
 314     void set_partial_obj_size(size_t words)    {
 315       _partial_obj_size = (region_sz_t) words;
 316     }
 317     inline void set_blocks_filled();
 318 
 319     inline void set_destination_count(uint count);
 320     inline void set_live_obj_size(size_t words);
 321     inline void set_data_location(HeapWord* addr);
 322     inline void set_completed();
 323     inline bool claim_unsafe();
 324 
 325     // These are atomic.
 326     inline void add_live_obj(size_t words);
 327     inline void set_highest_ref(HeapWord* addr);
 328     inline void decrement_destination_count();
 329     inline bool claim();
 330 
 331   private:
 332     // The type used to represent object sizes within a region.
 333     typedef uint region_sz_t;
 334 
 335     // Constants for manipulating the _dc_and_los field, which holds both the
 336     // destination count and live obj size.  The live obj size lives at the
 337     // least significant end so no masking is necessary when adding.
 338     static const region_sz_t dc_shift;           // Shift amount.
 339     static const region_sz_t dc_mask;            // Mask for destination count.
 340     static const region_sz_t dc_one;             // 1, shifted appropriately.
 341     static const region_sz_t dc_claimed;         // Region has been claimed.
 342     static const region_sz_t dc_completed;       // Region has been completed.
 343     static const region_sz_t los_mask;           // Mask for live obj size.
 344 
 345     HeapWord*            _destination;
 346     size_t               _source_region;
 347     HeapWord*            _partial_obj_addr;
 348     region_sz_t          _partial_obj_size;
 349     region_sz_t volatile _dc_and_los;
 350     bool                 _blocks_filled;
 351 
 352 #ifdef ASSERT
 353     size_t               _blocks_filled_count;   // Number of block table fills.
 354 
 355     // These enable optimizations that are only partially implemented.  Use
 356     // debug builds to prevent the code fragments from breaking.
 357     HeapWord*            _data_location;
 358     HeapWord*            _highest_ref;
 359 #endif  // #ifdef ASSERT
 360 
 361 #ifdef ASSERT
 362    public:
 363     uint                 _pushed;   // 0 until region is pushed onto a stack
 364    private:
 365 #endif
 366   };
 367 
 368   // "Blocks" allow shorter sections of the bitmap to be searched.  Each Block
 369   // holds an offset, which is the amount of live data in the Region to the left
 370   // of the first live object that starts in the Block.
 371   class BlockData
 372   {
 373   public:
 374     typedef unsigned short int blk_ofs_t;
 375 
 376     blk_ofs_t offset() const    { return _offset; }
 377     void set_offset(size_t val) { _offset = (blk_ofs_t)val; }
 378 
 379   private:
 380     blk_ofs_t _offset;
 381   };
 382 
 383 public:
 384   ParallelCompactData();
 385   bool initialize(MemRegion covered_region);
 386 
 387   size_t region_count() const { return _region_count; }
 388   size_t reserved_byte_size() const { return _reserved_byte_size; }
 389 
 390   // Convert region indices to/from RegionData pointers.
 391   inline RegionData* region(size_t region_idx) const;
 392   inline size_t     region(const RegionData* const region_ptr) const;
 393 
 394   size_t block_count() const { return _block_count; }
 395   inline BlockData* block(size_t block_idx) const;
 396   inline size_t     block(const BlockData* block_ptr) const;
 397 
 398   void add_obj(HeapWord* addr, size_t len);
 399   void add_obj(oop p, size_t len) { add_obj((HeapWord*)p, len); }
 400 
 401   // Fill in the regions covering [beg, end) so that no data moves; i.e., the
 402   // destination of region n is simply the start of region n.  The argument beg
 403   // must be region-aligned; end need not be.
 404   void summarize_dense_prefix(HeapWord* beg, HeapWord* end);
 405 
 406   HeapWord* summarize_split_space(size_t src_region, SplitInfo& split_info,
 407                                   HeapWord* destination, HeapWord* target_end,
 408                                   HeapWord** target_next);
 409   bool summarize(SplitInfo& split_info,
 410                  HeapWord* source_beg, HeapWord* source_end,
 411                  HeapWord** source_next,
 412                  HeapWord* target_beg, HeapWord* target_end,
 413                  HeapWord** target_next);
 414 
 415   void clear();
 416   void clear_range(size_t beg_region, size_t end_region);
 417   void clear_range(HeapWord* beg, HeapWord* end) {
 418     clear_range(addr_to_region_idx(beg), addr_to_region_idx(end));
 419   }
 420 
 421   // Return the number of words between addr and the start of the region
 422   // containing addr.
 423   inline size_t     region_offset(const HeapWord* addr) const;
 424 
 425   // Convert addresses to/from a region index or region pointer.
 426   inline size_t     addr_to_region_idx(const HeapWord* addr) const;
 427   inline RegionData* addr_to_region_ptr(const HeapWord* addr) const;
 428   inline HeapWord*  region_to_addr(size_t region) const;
 429   inline HeapWord*  region_to_addr(size_t region, size_t offset) const;
 430   inline HeapWord*  region_to_addr(const RegionData* region) const;
 431 
 432   inline HeapWord*  region_align_down(HeapWord* addr) const;
 433   inline HeapWord*  region_align_up(HeapWord* addr) const;
 434   inline bool       is_region_aligned(HeapWord* addr) const;
 435 
 436   // Analogous to region_offset() for blocks.
 437   size_t     block_offset(const HeapWord* addr) const;
 438   size_t     addr_to_block_idx(const HeapWord* addr) const;
 439   size_t     addr_to_block_idx(const oop obj) const {
 440     return addr_to_block_idx((HeapWord*) obj);
 441   }
 442   inline BlockData* addr_to_block_ptr(const HeapWord* addr) const;
 443   inline HeapWord*  block_to_addr(size_t block) const;
 444   inline size_t     region_to_block_idx(size_t region) const;
 445 
 446   inline HeapWord*  block_align_down(HeapWord* addr) const;
 447   inline HeapWord*  block_align_up(HeapWord* addr) const;
 448   inline bool       is_block_aligned(HeapWord* addr) const;
 449 
 450   // Return the address one past the end of the partial object.
 451   HeapWord* partial_obj_end(size_t region_idx) const;
 452 
 453   // Return the location of the object after compaction.
 454   HeapWord* calc_new_pointer(HeapWord* addr);
 455 
 456   HeapWord* calc_new_pointer(oop p) {
 457     return calc_new_pointer((HeapWord*) p);
 458   }
 459 
 460 #ifdef  ASSERT
 461   void verify_clear(const PSVirtualSpace* vspace);
 462   void verify_clear();
 463 #endif  // #ifdef ASSERT
 464 
 465 private:
 466   bool initialize_block_data();
 467   bool initialize_region_data(size_t region_size);
 468   PSVirtualSpace* create_vspace(size_t count, size_t element_size);
 469 
 470 private:
 471   HeapWord*       _region_start;
 472 #ifdef  ASSERT
 473   HeapWord*       _region_end;
 474 #endif  // #ifdef ASSERT
 475 
 476   PSVirtualSpace* _region_vspace;
 477   size_t          _reserved_byte_size;
 478   RegionData*     _region_data;
 479   size_t          _region_count;
 480 
 481   PSVirtualSpace* _block_vspace;
 482   BlockData*      _block_data;
 483   size_t          _block_count;
 484 };
 485 
 486 inline uint
 487 ParallelCompactData::RegionData::destination_count_raw() const
 488 {
 489   return _dc_and_los & dc_mask;
 490 }
 491 
 492 inline uint
 493 ParallelCompactData::RegionData::destination_count() const
 494 {
 495   return destination_count_raw() >> dc_shift;
 496 }
 497 
 498 inline bool
 499 ParallelCompactData::RegionData::blocks_filled() const
 500 {
 501   return _blocks_filled;
 502 }
 503 
 504 #ifdef ASSERT
 505 inline size_t
 506 ParallelCompactData::RegionData::blocks_filled_count() const
 507 {
 508   return _blocks_filled_count;
 509 }
 510 #endif // #ifdef ASSERT
 511 
 512 inline void
 513 ParallelCompactData::RegionData::set_blocks_filled()
 514 {
 515   _blocks_filled = true;
 516   // Debug builds count the number of times the table was filled.
 517   DEBUG_ONLY(Atomic::inc_ptr(&_blocks_filled_count));
 518 }
 519 
 520 inline void
 521 ParallelCompactData::RegionData::set_destination_count(uint count)
 522 {
 523   assert(count <= (dc_completed >> dc_shift), "count too large");
 524   const region_sz_t live_sz = (region_sz_t) live_obj_size();
 525   _dc_and_los = (count << dc_shift) | live_sz;
 526 }
 527 
 528 inline void ParallelCompactData::RegionData::set_live_obj_size(size_t words)
 529 {
 530   assert(words <= los_mask, "would overflow");
 531   _dc_and_los = destination_count_raw() | (region_sz_t)words;
 532 }
 533 
 534 inline void ParallelCompactData::RegionData::decrement_destination_count()
 535 {
 536   assert(_dc_and_los < dc_claimed, "already claimed");
 537   assert(_dc_and_los >= dc_one, "count would go negative");
 538   Atomic::add((int)dc_mask, (volatile int*)&_dc_and_los);
 539 }
 540 
 541 inline HeapWord* ParallelCompactData::RegionData::data_location() const
 542 {
 543   DEBUG_ONLY(return _data_location;)
 544   NOT_DEBUG(return NULL;)
 545 }
 546 
 547 inline HeapWord* ParallelCompactData::RegionData::highest_ref() const
 548 {
 549   DEBUG_ONLY(return _highest_ref;)
 550   NOT_DEBUG(return NULL;)
 551 }
 552 
 553 inline void ParallelCompactData::RegionData::set_data_location(HeapWord* addr)
 554 {
 555   DEBUG_ONLY(_data_location = addr;)
 556 }
 557 
 558 inline void ParallelCompactData::RegionData::set_completed()
 559 {
 560   assert(claimed(), "must be claimed first");
 561   _dc_and_los = dc_completed | (region_sz_t) live_obj_size();
 562 }
 563 
 564 // MT-unsafe claiming of a region.  Should only be used during single threaded
 565 // execution.
 566 inline bool ParallelCompactData::RegionData::claim_unsafe()
 567 {
 568   if (available()) {
 569     _dc_and_los |= dc_claimed;
 570     return true;
 571   }
 572   return false;
 573 }
 574 
 575 inline void ParallelCompactData::RegionData::add_live_obj(size_t words)
 576 {
 577   assert(words <= (size_t)los_mask - live_obj_size(), "overflow");
 578   Atomic::add((int) words, (volatile int*) &_dc_and_los);
 579 }
 580 
 581 inline void ParallelCompactData::RegionData::set_highest_ref(HeapWord* addr)
 582 {
 583 #ifdef ASSERT
 584   HeapWord* tmp = _highest_ref;
 585   while (addr > tmp) {
 586     tmp = (HeapWord*)Atomic::cmpxchg_ptr(addr, &_highest_ref, tmp);
 587   }
 588 #endif  // #ifdef ASSERT
 589 }
 590 
 591 inline bool ParallelCompactData::RegionData::claim()
 592 {
 593   const int los = (int) live_obj_size();
 594   const int old = Atomic::cmpxchg(dc_claimed | los,
 595                                   (volatile int*) &_dc_and_los, los);
 596   return old == los;
 597 }
 598 
 599 inline ParallelCompactData::RegionData*
 600 ParallelCompactData::region(size_t region_idx) const
 601 {
 602   assert(region_idx <= region_count(), "bad arg");
 603   return _region_data + region_idx;
 604 }
 605 
 606 inline size_t
 607 ParallelCompactData::region(const RegionData* const region_ptr) const
 608 {
 609   assert(region_ptr >= _region_data, "bad arg");
 610   assert(region_ptr <= _region_data + region_count(), "bad arg");
 611   return pointer_delta(region_ptr, _region_data, sizeof(RegionData));
 612 }
 613 
 614 inline ParallelCompactData::BlockData*
 615 ParallelCompactData::block(size_t n) const {
 616   assert(n < block_count(), "bad arg");
 617   return _block_data + n;
 618 }
 619 
 620 inline size_t
 621 ParallelCompactData::region_offset(const HeapWord* addr) const
 622 {
 623   assert(addr >= _region_start, "bad addr");
 624   assert(addr <= _region_end, "bad addr");
 625   return (size_t(addr) & RegionAddrOffsetMask) >> LogHeapWordSize;
 626 }
 627 
 628 inline size_t
 629 ParallelCompactData::addr_to_region_idx(const HeapWord* addr) const
 630 {
 631   assert(addr >= _region_start, "bad addr");
 632   assert(addr <= _region_end, "bad addr");
 633   return pointer_delta(addr, _region_start) >> Log2RegionSize;
 634 }
 635 
 636 inline ParallelCompactData::RegionData*
 637 ParallelCompactData::addr_to_region_ptr(const HeapWord* addr) const
 638 {
 639   return region(addr_to_region_idx(addr));
 640 }
 641 
 642 inline HeapWord*
 643 ParallelCompactData::region_to_addr(size_t region) const
 644 {
 645   assert(region <= _region_count, "region out of range");
 646   return _region_start + (region << Log2RegionSize);
 647 }
 648 
 649 inline HeapWord*
 650 ParallelCompactData::region_to_addr(const RegionData* region) const
 651 {
 652   return region_to_addr(pointer_delta(region, _region_data,
 653                                       sizeof(RegionData)));
 654 }
 655 
 656 inline HeapWord*
 657 ParallelCompactData::region_to_addr(size_t region, size_t offset) const
 658 {
 659   assert(region <= _region_count, "region out of range");
 660   assert(offset < RegionSize, "offset too big");  // This may be too strict.
 661   return region_to_addr(region) + offset;
 662 }
 663 
 664 inline HeapWord*
 665 ParallelCompactData::region_align_down(HeapWord* addr) const
 666 {
 667   assert(addr >= _region_start, "bad addr");
 668   assert(addr < _region_end + RegionSize, "bad addr");
 669   return (HeapWord*)(size_t(addr) & RegionAddrMask);
 670 }
 671 
 672 inline HeapWord*
 673 ParallelCompactData::region_align_up(HeapWord* addr) const
 674 {
 675   assert(addr >= _region_start, "bad addr");
 676   assert(addr <= _region_end, "bad addr");
 677   return region_align_down(addr + RegionSizeOffsetMask);
 678 }
 679 
 680 inline bool
 681 ParallelCompactData::is_region_aligned(HeapWord* addr) const
 682 {
 683   return region_offset(addr) == 0;
 684 }
 685 
 686 inline size_t
 687 ParallelCompactData::block_offset(const HeapWord* addr) const
 688 {
 689   assert(addr >= _region_start, "bad addr");
 690   assert(addr <= _region_end, "bad addr");
 691   return (size_t(addr) & BlockAddrOffsetMask) >> LogHeapWordSize;
 692 }
 693 
 694 inline size_t
 695 ParallelCompactData::addr_to_block_idx(const HeapWord* addr) const
 696 {
 697   assert(addr >= _region_start, "bad addr");
 698   assert(addr <= _region_end, "bad addr");
 699   return pointer_delta(addr, _region_start) >> Log2BlockSize;
 700 }
 701 
 702 inline ParallelCompactData::BlockData*
 703 ParallelCompactData::addr_to_block_ptr(const HeapWord* addr) const
 704 {
 705   return block(addr_to_block_idx(addr));
 706 }
 707 
 708 inline HeapWord*
 709 ParallelCompactData::block_to_addr(size_t block) const
 710 {
 711   assert(block < _block_count, "block out of range");
 712   return _region_start + (block << Log2BlockSize);
 713 }
 714 
 715 inline size_t
 716 ParallelCompactData::region_to_block_idx(size_t region) const
 717 {
 718   return region << Log2BlocksPerRegion;
 719 }
 720 
 721 inline HeapWord*
 722 ParallelCompactData::block_align_down(HeapWord* addr) const
 723 {
 724   assert(addr >= _region_start, "bad addr");
 725   assert(addr < _region_end + RegionSize, "bad addr");
 726   return (HeapWord*)(size_t(addr) & BlockAddrMask);
 727 }
 728 
 729 inline HeapWord*
 730 ParallelCompactData::block_align_up(HeapWord* addr) const
 731 {
 732   assert(addr >= _region_start, "bad addr");
 733   assert(addr <= _region_end, "bad addr");
 734   return block_align_down(addr + BlockSizeOffsetMask);
 735 }
 736 
 737 inline bool
 738 ParallelCompactData::is_block_aligned(HeapWord* addr) const
 739 {
 740   return block_offset(addr) == 0;
 741 }
 742 
 743 // Abstract closure for use with ParMarkBitMap::iterate(), which will invoke the
 744 // do_addr() method.
 745 //
 746 // The closure is initialized with the number of heap words to process
 747 // (words_remaining()), and becomes 'full' when it reaches 0.  The do_addr()
 748 // methods in subclasses should update the total as words are processed.  Since
 749 // only one subclass actually uses this mechanism to terminate iteration, the
 750 // default initial value is > 0.  The implementation is here and not in the
 751 // single subclass that uses it to avoid making is_full() virtual, and thus
 752 // adding a virtual call per live object.
 753 
 754 class ParMarkBitMapClosure: public StackObj {
 755  public:
 756   typedef ParMarkBitMap::idx_t idx_t;
 757   typedef ParMarkBitMap::IterationStatus IterationStatus;
 758 
 759  public:
 760   inline ParMarkBitMapClosure(ParMarkBitMap* mbm, ParCompactionManager* cm,
 761                               size_t words = max_uintx);
 762 
 763   inline ParCompactionManager* compaction_manager() const;
 764   inline ParMarkBitMap*        bitmap() const;
 765   inline size_t                words_remaining() const;
 766   inline bool                  is_full() const;
 767   inline HeapWord*             source() const;
 768 
 769   inline void                  set_source(HeapWord* addr);
 770 
 771   virtual IterationStatus do_addr(HeapWord* addr, size_t words) = 0;
 772 
 773  protected:
 774   inline void decrement_words_remaining(size_t words);
 775 
 776  private:
 777   ParMarkBitMap* const        _bitmap;
 778   ParCompactionManager* const _compaction_manager;
 779   DEBUG_ONLY(const size_t     _initial_words_remaining;) // Useful in debugger.
 780   size_t                      _words_remaining; // Words left to copy.
 781 
 782  protected:
 783   HeapWord*                   _source;          // Next addr that would be read.
 784 };
 785 
 786 inline
 787 ParMarkBitMapClosure::ParMarkBitMapClosure(ParMarkBitMap* bitmap,
 788                                            ParCompactionManager* cm,
 789                                            size_t words):
 790   _bitmap(bitmap), _compaction_manager(cm)
 791 #ifdef  ASSERT
 792   , _initial_words_remaining(words)
 793 #endif
 794 {
 795   _words_remaining = words;
 796   _source = NULL;
 797 }
 798 
 799 inline ParCompactionManager* ParMarkBitMapClosure::compaction_manager() const {
 800   return _compaction_manager;
 801 }
 802 
 803 inline ParMarkBitMap* ParMarkBitMapClosure::bitmap() const {
 804   return _bitmap;
 805 }
 806 
 807 inline size_t ParMarkBitMapClosure::words_remaining() const {
 808   return _words_remaining;
 809 }
 810 
 811 inline bool ParMarkBitMapClosure::is_full() const {
 812   return words_remaining() == 0;
 813 }
 814 
 815 inline HeapWord* ParMarkBitMapClosure::source() const {
 816   return _source;
 817 }
 818 
 819 inline void ParMarkBitMapClosure::set_source(HeapWord* addr) {
 820   _source = addr;
 821 }
 822 
 823 inline void ParMarkBitMapClosure::decrement_words_remaining(size_t words) {
 824   assert(_words_remaining >= words, "processed too many words");
 825   _words_remaining -= words;
 826 }
 827 
 828 // The UseParallelOldGC collector is a stop-the-world garbage collector that
 829 // does parts of the collection using parallel threads.  The collection includes
 830 // the tenured generation and the young generation.  The permanent generation is
 831 // collected at the same time as the other two generations but the permanent
 832 // generation is collect by a single GC thread.  The permanent generation is
 833 // collected serially because of the requirement that during the processing of a
 834 // klass AAA, any objects reference by AAA must already have been processed.
 835 // This requirement is enforced by a left (lower address) to right (higher
 836 // address) sliding compaction.
 837 //
 838 // There are four phases of the collection.
 839 //
 840 //      - marking phase
 841 //      - summary phase
 842 //      - compacting phase
 843 //      - clean up phase
 844 //
 845 // Roughly speaking these phases correspond, respectively, to
 846 //      - mark all the live objects
 847 //      - calculate the destination of each object at the end of the collection
 848 //      - move the objects to their destination
 849 //      - update some references and reinitialize some variables
 850 //
 851 // These three phases are invoked in PSParallelCompact::invoke_no_policy().  The
 852 // marking phase is implemented in PSParallelCompact::marking_phase() and does a
 853 // complete marking of the heap.  The summary phase is implemented in
 854 // PSParallelCompact::summary_phase().  The move and update phase is implemented
 855 // in PSParallelCompact::compact().
 856 //
 857 // A space that is being collected is divided into regions and with each region
 858 // is associated an object of type ParallelCompactData.  Each region is of a
 859 // fixed size and typically will contain more than 1 object and may have parts
 860 // of objects at the front and back of the region.
 861 //
 862 // region            -----+---------------------+----------
 863 // objects covered   [ AAA  )[ BBB )[ CCC   )[ DDD     )
 864 //
 865 // The marking phase does a complete marking of all live objects in the heap.
 866 // The marking also compiles the size of the data for all live objects covered
 867 // by the region.  This size includes the part of any live object spanning onto
 868 // the region (part of AAA if it is live) from the front, all live objects
 869 // contained in the region (BBB and/or CCC if they are live), and the part of
 870 // any live objects covered by the region that extends off the region (part of
 871 // DDD if it is live).  The marking phase uses multiple GC threads and marking
 872 // is done in a bit array of type ParMarkBitMap.  The marking of the bit map is
 873 // done atomically as is the accumulation of the size of the live objects
 874 // covered by a region.
 875 //
 876 // The summary phase calculates the total live data to the left of each region
 877 // XXX.  Based on that total and the bottom of the space, it can calculate the
 878 // starting location of the live data in XXX.  The summary phase calculates for
 879 // each region XXX quantities such as
 880 //
 881 //      - the amount of live data at the beginning of a region from an object
 882 //        entering the region.
 883 //      - the location of the first live data on the region
 884 //      - a count of the number of regions receiving live data from XXX.
 885 //
 886 // See ParallelCompactData for precise details.  The summary phase also
 887 // calculates the dense prefix for the compaction.  The dense prefix is a
 888 // portion at the beginning of the space that is not moved.  The objects in the
 889 // dense prefix do need to have their object references updated.  See method
 890 // summarize_dense_prefix().
 891 //
 892 // The summary phase is done using 1 GC thread.
 893 //
 894 // The compaction phase moves objects to their new location and updates all
 895 // references in the object.
 896 //
 897 // A current exception is that objects that cross a region boundary are moved
 898 // but do not have their references updated.  References are not updated because
 899 // it cannot easily be determined if the klass pointer KKK for the object AAA
 900 // has been updated.  KKK likely resides in a region to the left of the region
 901 // containing AAA.  These AAA's have there references updated at the end in a
 902 // clean up phase.  See the method PSParallelCompact::update_deferred_objects().
 903 // An alternate strategy is being investigated for this deferral of updating.
 904 //
 905 // Compaction is done on a region basis.  A region that is ready to be filled is
 906 // put on a ready list and GC threads take region off the list and fill them.  A
 907 // region is ready to be filled if it empty of live objects.  Such a region may
 908 // have been initially empty (only contained dead objects) or may have had all
 909 // its live objects copied out already.  A region that compacts into itself is
 910 // also ready for filling.  The ready list is initially filled with empty
 911 // regions and regions compacting into themselves.  There is always at least 1
 912 // region that can be put on the ready list.  The regions are atomically added
 913 // and removed from the ready list.
 914 
 915 class PSParallelCompact : AllStatic {
 916  public:
 917   // Convenient access to type names.
 918   typedef ParMarkBitMap::idx_t idx_t;
 919   typedef ParallelCompactData::RegionData RegionData;
 920   typedef ParallelCompactData::BlockData BlockData;
 921 
 922   typedef enum {
 923     old_space_id, eden_space_id,
 924     from_space_id, to_space_id, last_space_id
 925   } SpaceId;
 926 
 927  public:
 928   // Inline closure decls
 929   //
 930   class IsAliveClosure: public BoolObjectClosure {
 931    public:
 932     virtual bool do_object_b(oop p);
 933   };
 934 
 935   class KeepAliveClosure: public OopClosure {
 936    private:
 937     ParCompactionManager* _compaction_manager;
 938    protected:
 939     template <class T> inline void do_oop_work(T* p);
 940    public:
 941     KeepAliveClosure(ParCompactionManager* cm) : _compaction_manager(cm) { }
 942     virtual void do_oop(oop* p);
 943     virtual void do_oop(narrowOop* p);
 944   };
 945 
 946   class FollowStackClosure: public VoidClosure {
 947    private:
 948     ParCompactionManager* _compaction_manager;
 949    public:
 950     FollowStackClosure(ParCompactionManager* cm) : _compaction_manager(cm) { }
 951     virtual void do_void();
 952   };
 953 
 954   class AdjustPointerClosure: public ExtendedOopClosure {
 955    public:
 956     template <typename T> void do_oop_nv(T* p);
 957     virtual void do_oop(oop* p);
 958     virtual void do_oop(narrowOop* p);
 959 
 960     // This closure provides its own oop verification code.
 961     debug_only(virtual bool should_verify_oops() { return false; })
 962   };
 963 
 964   class AdjustKlassClosure : public KlassClosure {
 965    public:
 966     void do_klass(Klass* klass);
 967   };
 968 
 969   friend class KeepAliveClosure;
 970   friend class FollowStackClosure;
 971   friend class AdjustPointerClosure;
 972   friend class AdjustKlassClosure;
 973   friend class FollowKlassClosure;
 974   friend class InstanceClassLoaderKlass;
 975   friend class RefProcTaskProxy;
 976 
 977  private:
 978   static STWGCTimer           _gc_timer;
 979   static ParallelOldTracer    _gc_tracer;
 980   static elapsedTimer         _accumulated_time;
 981   static unsigned int         _total_invocations;
 982   static unsigned int         _maximum_compaction_gc_num;
 983   static jlong                _time_of_last_gc;   // ms
 984   static CollectorCounters*   _counters;
 985   static ParMarkBitMap        _mark_bitmap;
 986   static ParallelCompactData  _summary_data;
 987   static IsAliveClosure       _is_alive_closure;
 988   static SpaceInfo            _space_info[last_space_id];
 989   static bool                 _print_phases;
 990   static AdjustPointerClosure _adjust_pointer_closure;
 991   static AdjustKlassClosure   _adjust_klass_closure;
 992 
 993   // Reference processing (used in ...follow_contents)
 994   static ReferenceProcessor*  _ref_processor;
 995 
 996   // Updated location of intArrayKlassObj.
 997   static Klass* _updated_int_array_klass_obj;
 998 
 999   // Values computed at initialization and used by dead_wood_limiter().
1000   static double _dwl_mean;
1001   static double _dwl_std_dev;
1002   static double _dwl_first_term;
1003   static double _dwl_adjustment;
1004 #ifdef  ASSERT
1005   static bool   _dwl_initialized;
1006 #endif  // #ifdef ASSERT
1007 
1008 
1009  public:
1010   static ParallelOldTracer* gc_tracer() { return &_gc_tracer; }
1011 
1012  private:
1013 
1014   static void initialize_space_info();
1015 
1016   // Return true if details about individual phases should be printed.
1017   static inline bool print_phases();
1018 
1019   // Clear the marking bitmap and summary data that cover the specified space.
1020   static void clear_data_covering_space(SpaceId id);
1021 
1022   static void pre_compact(PreGCValues* pre_gc_values);
1023   static void post_compact();
1024 
1025   // Mark live objects
1026   static void marking_phase(ParCompactionManager* cm,
1027                             bool maximum_heap_compaction,
1028                             ParallelOldTracer *gc_tracer);
1029 
1030   // Compute the dense prefix for the designated space.  This is an experimental
1031   // implementation currently not used in production.
1032   static HeapWord* compute_dense_prefix_via_density(const SpaceId id,
1033                                                     bool maximum_compaction);
1034 
1035   // Methods used to compute the dense prefix.
1036 
1037   // Compute the value of the normal distribution at x = density.  The mean and
1038   // standard deviation are values saved by initialize_dead_wood_limiter().
1039   static inline double normal_distribution(double density);
1040 
1041   // Initialize the static vars used by dead_wood_limiter().
1042   static void initialize_dead_wood_limiter();
1043 
1044   // Return the percentage of space that can be treated as "dead wood" (i.e.,
1045   // not reclaimed).
1046   static double dead_wood_limiter(double density, size_t min_percent);
1047 
1048   // Find the first (left-most) region in the range [beg, end) that has at least
1049   // dead_words of dead space to the left.  The argument beg must be the first
1050   // region in the space that is not completely live.
1051   static RegionData* dead_wood_limit_region(const RegionData* beg,
1052                                             const RegionData* end,
1053                                             size_t dead_words);
1054 
1055   // Return a pointer to the first region in the range [beg, end) that is not
1056   // completely full.
1057   static RegionData* first_dead_space_region(const RegionData* beg,
1058                                              const RegionData* end);
1059 
1060   // Return a value indicating the benefit or 'yield' if the compacted region
1061   // were to start (or equivalently if the dense prefix were to end) at the
1062   // candidate region.  Higher values are better.
1063   //
1064   // The value is based on the amount of space reclaimed vs. the costs of (a)
1065   // updating references in the dense prefix plus (b) copying objects and
1066   // updating references in the compacted region.
1067   static inline double reclaimed_ratio(const RegionData* const candidate,
1068                                        HeapWord* const bottom,
1069                                        HeapWord* const top,
1070                                        HeapWord* const new_top);
1071 
1072   // Compute the dense prefix for the designated space.
1073   static HeapWord* compute_dense_prefix(const SpaceId id,
1074                                         bool maximum_compaction);
1075 
1076   // Return true if dead space crosses onto the specified Region; bit must be
1077   // the bit index corresponding to the first word of the Region.
1078   static inline bool dead_space_crosses_boundary(const RegionData* region,
1079                                                  idx_t bit);
1080 
1081   // Summary phase utility routine to fill dead space (if any) at the dense
1082   // prefix boundary.  Should only be called if the the dense prefix is
1083   // non-empty.
1084   static void fill_dense_prefix_end(SpaceId id);
1085 
1086   // Clear the summary data source_region field for the specified addresses.
1087   static void clear_source_region(HeapWord* beg_addr, HeapWord* end_addr);
1088 
1089 #ifndef PRODUCT
1090   // Routines to provoke splitting a young gen space (ParallelOldGCSplitALot).
1091 
1092   // Fill the region [start, start + words) with live object(s).  Only usable
1093   // for the old and permanent generations.
1094   static void fill_with_live_objects(SpaceId id, HeapWord* const start,
1095                                      size_t words);
1096   // Include the new objects in the summary data.
1097   static void summarize_new_objects(SpaceId id, HeapWord* start);
1098 
1099   // Add live objects to a survivor space since it's rare that both survivors
1100   // are non-empty.
1101   static void provoke_split_fill_survivor(SpaceId id);
1102 
1103   // Add live objects and/or choose the dense prefix to provoke splitting.
1104   static void provoke_split(bool & maximum_compaction);
1105 #endif
1106 
1107   static void summarize_spaces_quick();
1108   static void summarize_space(SpaceId id, bool maximum_compaction);
1109   static void summary_phase(ParCompactionManager* cm, bool maximum_compaction);
1110 
1111   // Adjust addresses in roots.  Does not adjust addresses in heap.
1112   static void adjust_roots();
1113 
1114   DEBUG_ONLY(static void write_block_fill_histogram(outputStream* const out);)
1115 
1116   // Move objects to new locations.
1117   static void compact_perm(ParCompactionManager* cm);
1118   static void compact();
1119 
1120   // Add available regions to the stack and draining tasks to the task queue.
1121   static void enqueue_region_draining_tasks(GCTaskQueue* q,
1122                                             uint parallel_gc_threads);
1123 
1124   // Add dense prefix update tasks to the task queue.
1125   static void enqueue_dense_prefix_tasks(GCTaskQueue* q,
1126                                          uint parallel_gc_threads);
1127 
1128   // Add region stealing tasks to the task queue.
1129   static void enqueue_region_stealing_tasks(
1130                                        GCTaskQueue* q,
1131                                        ParallelTaskTerminator* terminator_ptr,
1132                                        uint parallel_gc_threads);
1133 
1134   // If objects are left in eden after a collection, try to move the boundary
1135   // and absorb them into the old gen.  Returns true if eden was emptied.
1136   static bool absorb_live_data_from_eden(PSAdaptiveSizePolicy* size_policy,
1137                                          PSYoungGen* young_gen,
1138                                          PSOldGen* old_gen);
1139 
1140   // Reset time since last full gc
1141   static void reset_millis_since_last_gc();
1142 
1143  public:
1144   class MarkAndPushClosure: public ExtendedOopClosure {
1145    private:
1146     ParCompactionManager* _compaction_manager;
1147    public:
1148     MarkAndPushClosure(ParCompactionManager* cm) : _compaction_manager(cm) { }
1149 
1150     template <typename T> void do_oop_nv(T* p);
1151     virtual void do_oop(oop* p);
1152     virtual void do_oop(narrowOop* p);
1153 
1154     // This closure provides its own oop verification code.
1155     debug_only(virtual bool should_verify_oops() { return false; })
1156   };
1157 
1158   // The one and only place to start following the classes.
1159   // Should only be applied to the ClassLoaderData klasses list.
1160   class FollowKlassClosure : public KlassClosure {
1161    private:
1162     MarkAndPushClosure* _mark_and_push_closure;
1163    public:
1164     FollowKlassClosure(MarkAndPushClosure* mark_and_push_closure) :
1165         _mark_and_push_closure(mark_and_push_closure) { }
1166     void do_klass(Klass* klass);
1167   };
1168 
1169   PSParallelCompact();
1170 
1171   // Convenient accessor for Universe::heap().
1172   static ParallelScavengeHeap* gc_heap() {
1173     return (ParallelScavengeHeap*)Universe::heap();
1174   }
1175 
1176   static void invoke(bool maximum_heap_compaction);
1177   static bool invoke_no_policy(bool maximum_heap_compaction);
1178 
1179   static void post_initialize();
1180   // Perform initialization for PSParallelCompact that requires
1181   // allocations.  This should be called during the VM initialization
1182   // at a pointer where it would be appropriate to return a JNI_ENOMEM
1183   // in the event of a failure.
1184   static bool initialize();
1185 
1186   // Closure accessors
1187   static PSParallelCompact::AdjustPointerClosure* adjust_pointer_closure() {
1188     return &_adjust_pointer_closure;
1189   }
1190   static KlassClosure* adjust_klass_closure()      { return (KlassClosure*)&_adjust_klass_closure; }
1191   static BoolObjectClosure* is_alive_closure()     { return (BoolObjectClosure*)&_is_alive_closure; }
1192 
1193   // Public accessors
1194   static elapsedTimer* accumulated_time() { return &_accumulated_time; }
1195   static unsigned int total_invocations() { return _total_invocations; }
1196   static CollectorCounters* counters()    { return _counters; }
1197 
1198   // Used to add tasks
1199   static GCTaskManager* const gc_task_manager();
1200   static Klass* updated_int_array_klass_obj() {
1201     return _updated_int_array_klass_obj;
1202   }
1203 
1204   // Marking support
1205   static inline bool mark_obj(oop obj);
1206   static inline bool is_marked(oop obj);
1207   // Check mark and maybe push on marking stack
1208   template <class T> static inline void mark_and_push(ParCompactionManager* cm,
1209                                                       T* p);
1210   template <class T> static inline void adjust_pointer(T* p);
1211 
1212   static inline void follow_klass(ParCompactionManager* cm, Klass* klass);
1213 
1214   static void follow_class_loader(ParCompactionManager* cm,
1215                                   ClassLoaderData* klass);
1216 
1217   // Compaction support.
1218   // Return true if p is in the range [beg_addr, end_addr).
1219   static inline bool is_in(HeapWord* p, HeapWord* beg_addr, HeapWord* end_addr);
1220   static inline bool is_in(oop* p, HeapWord* beg_addr, HeapWord* end_addr);
1221 
1222   // Convenience wrappers for per-space data kept in _space_info.
1223   static inline MutableSpace*     space(SpaceId space_id);
1224   static inline HeapWord*         new_top(SpaceId space_id);
1225   static inline HeapWord*         dense_prefix(SpaceId space_id);
1226   static inline ObjectStartArray* start_array(SpaceId space_id);
1227 
1228   // Move and update the live objects in the specified space.
1229   static void move_and_update(ParCompactionManager* cm, SpaceId space_id);
1230 
1231   // Process the end of the given region range in the dense prefix.
1232   // This includes saving any object not updated.
1233   static void dense_prefix_regions_epilogue(ParCompactionManager* cm,
1234                                             size_t region_start_index,
1235                                             size_t region_end_index,
1236                                             idx_t exiting_object_offset,
1237                                             idx_t region_offset_start,
1238                                             idx_t region_offset_end);
1239 
1240   // Update a region in the dense prefix.  For each live object
1241   // in the region, update it's interior references.  For each
1242   // dead object, fill it with deadwood. Dead space at the end
1243   // of a region range will be filled to the start of the next
1244   // live object regardless of the region_index_end.  None of the
1245   // objects in the dense prefix move and dead space is dead
1246   // (holds only dead objects that don't need any processing), so
1247   // dead space can be filled in any order.
1248   static void update_and_deadwood_in_dense_prefix(ParCompactionManager* cm,
1249                                                   SpaceId space_id,
1250                                                   size_t region_index_start,
1251                                                   size_t region_index_end);
1252 
1253   // Return the address of the count + 1st live word in the range [beg, end).
1254   static HeapWord* skip_live_words(HeapWord* beg, HeapWord* end, size_t count);
1255 
1256   // Return the address of the word to be copied to dest_addr, which must be
1257   // aligned to a region boundary.
1258   static HeapWord* first_src_addr(HeapWord* const dest_addr,
1259                                   SpaceId src_space_id,
1260                                   size_t src_region_idx);
1261 
1262   // Determine the next source region, set closure.source() to the start of the
1263   // new region return the region index.  Parameter end_addr is the address one
1264   // beyond the end of source range just processed.  If necessary, switch to a
1265   // new source space and set src_space_id (in-out parameter) and src_space_top
1266   // (out parameter) accordingly.
1267   static size_t next_src_region(MoveAndUpdateClosure& closure,
1268                                 SpaceId& src_space_id,
1269                                 HeapWord*& src_space_top,
1270                                 HeapWord* end_addr);
1271 
1272   // Decrement the destination count for each non-empty source region in the
1273   // range [beg_region, region(region_align_up(end_addr))).  If the destination
1274   // count for a region goes to 0 and it needs to be filled, enqueue it.
1275   static void decrement_destination_counts(ParCompactionManager* cm,
1276                                            SpaceId src_space_id,
1277                                            size_t beg_region,
1278                                            HeapWord* end_addr);
1279 
1280   // Fill a region, copying objects from one or more source regions.
1281   static void fill_region(ParCompactionManager* cm, size_t region_idx);
1282   static void fill_and_update_region(ParCompactionManager* cm, size_t region) {
1283     fill_region(cm, region);
1284   }
1285 
1286   // Fill in the block table for the specified region.
1287   static void fill_blocks(size_t region_idx);
1288 
1289   // Update the deferred objects in the space.
1290   static void update_deferred_objects(ParCompactionManager* cm, SpaceId id);
1291 
1292   static ParMarkBitMap* mark_bitmap() { return &_mark_bitmap; }
1293   static ParallelCompactData& summary_data() { return _summary_data; }
1294 
1295   // Reference Processing
1296   static ReferenceProcessor* const ref_processor() { return _ref_processor; }
1297 
1298   static STWGCTimer* gc_timer() { return &_gc_timer; }
1299 
1300   // Return the SpaceId for the given address.
1301   static SpaceId space_id(HeapWord* addr);
1302 
1303   // Time since last full gc (in milliseconds).
1304   static jlong millis_since_last_gc();
1305 
1306   static void print_on_error(outputStream* st);
1307 
1308 #ifndef PRODUCT
1309   // Debugging support.
1310   static const char* space_names[last_space_id];
1311   static void print_region_ranges();
1312   static void print_dense_prefix_stats(const char* const algorithm,
1313                                        const SpaceId id,
1314                                        const bool maximum_compaction,
1315                                        HeapWord* const addr);
1316   static void summary_phase_msg(SpaceId dst_space_id,
1317                                 HeapWord* dst_beg, HeapWord* dst_end,
1318                                 SpaceId src_space_id,
1319                                 HeapWord* src_beg, HeapWord* src_end);
1320 #endif  // #ifndef PRODUCT
1321 
1322 #ifdef  ASSERT
1323   // Sanity check the new location of a word in the heap.
1324   static inline void check_new_location(HeapWord* old_addr, HeapWord* new_addr);
1325   // Verify that all the regions have been emptied.
1326   static void verify_complete(SpaceId space_id);
1327 #endif  // #ifdef ASSERT
1328 };
1329 
1330 inline bool PSParallelCompact::mark_obj(oop obj) {
1331   const int obj_size = obj->size();
1332   if (mark_bitmap()->mark_obj(obj, obj_size)) {
1333     _summary_data.add_obj(obj, obj_size);
1334     return true;
1335   } else {
1336     return false;
1337   }
1338 }
1339 
1340 inline bool PSParallelCompact::is_marked(oop obj) {
1341   return mark_bitmap()->is_marked(obj);
1342 }
1343 
1344 template <class T>
1345 inline void PSParallelCompact::KeepAliveClosure::do_oop_work(T* p) {
1346   mark_and_push(_compaction_manager, p);
1347 }
1348 
1349 inline bool PSParallelCompact::print_phases() {
1350   return _print_phases;
1351 }
1352 
1353 inline double PSParallelCompact::normal_distribution(double density) {
1354   assert(_dwl_initialized, "uninitialized");
1355   const double squared_term = (density - _dwl_mean) / _dwl_std_dev;
1356   return _dwl_first_term * exp(-0.5 * squared_term * squared_term);
1357 }
1358 
1359 inline bool
1360 PSParallelCompact::dead_space_crosses_boundary(const RegionData* region,
1361                                                idx_t bit)
1362 {
1363   assert(bit > 0, "cannot call this for the first bit/region");
1364   assert(_summary_data.region_to_addr(region) == _mark_bitmap.bit_to_addr(bit),
1365          "sanity check");
1366 
1367   // Dead space crosses the boundary if (1) a partial object does not extend
1368   // onto the region, (2) an object does not start at the beginning of the
1369   // region, and (3) an object does not end at the end of the prior region.
1370   return region->partial_obj_size() == 0 &&
1371     !_mark_bitmap.is_obj_beg(bit) &&
1372     !_mark_bitmap.is_obj_end(bit - 1);
1373 }
1374 
1375 inline bool
1376 PSParallelCompact::is_in(HeapWord* p, HeapWord* beg_addr, HeapWord* end_addr) {
1377   return p >= beg_addr && p < end_addr;
1378 }
1379 
1380 inline bool
1381 PSParallelCompact::is_in(oop* p, HeapWord* beg_addr, HeapWord* end_addr) {
1382   return is_in((HeapWord*)p, beg_addr, end_addr);
1383 }
1384 
1385 inline MutableSpace* PSParallelCompact::space(SpaceId id) {
1386   assert(id < last_space_id, "id out of range");
1387   return _space_info[id].space();
1388 }
1389 
1390 inline HeapWord* PSParallelCompact::new_top(SpaceId id) {
1391   assert(id < last_space_id, "id out of range");
1392   return _space_info[id].new_top();
1393 }
1394 
1395 inline HeapWord* PSParallelCompact::dense_prefix(SpaceId id) {
1396   assert(id < last_space_id, "id out of range");
1397   return _space_info[id].dense_prefix();
1398 }
1399 
1400 inline ObjectStartArray* PSParallelCompact::start_array(SpaceId id) {
1401   assert(id < last_space_id, "id out of range");
1402   return _space_info[id].start_array();
1403 }
1404 
1405 #ifdef ASSERT
1406 inline void
1407 PSParallelCompact::check_new_location(HeapWord* old_addr, HeapWord* new_addr)
1408 {
1409   assert(old_addr >= new_addr || space_id(old_addr) != space_id(new_addr),
1410          "must move left or to a different space");
1411   assert(is_object_aligned((intptr_t)old_addr) && is_object_aligned((intptr_t)new_addr),
1412          "checking alignment");
1413 }
1414 #endif // ASSERT
1415 
1416 class MoveAndUpdateClosure: public ParMarkBitMapClosure {
1417  public:
1418   inline MoveAndUpdateClosure(ParMarkBitMap* bitmap, ParCompactionManager* cm,
1419                               ObjectStartArray* start_array,
1420                               HeapWord* destination, size_t words);
1421 
1422   // Accessors.
1423   HeapWord* destination() const         { return _destination; }
1424 
1425   // If the object will fit (size <= words_remaining()), copy it to the current
1426   // destination, update the interior oops and the start array and return either
1427   // full (if the closure is full) or incomplete.  If the object will not fit,
1428   // return would_overflow.
1429   virtual IterationStatus do_addr(HeapWord* addr, size_t size);
1430 
1431   // Copy enough words to fill this closure, starting at source().  Interior
1432   // oops and the start array are not updated.  Return full.
1433   IterationStatus copy_until_full();
1434 
1435   // Copy enough words to fill this closure or to the end of an object,
1436   // whichever is smaller, starting at source().  Interior oops and the start
1437   // array are not updated.
1438   void copy_partial_obj();
1439 
1440  protected:
1441   // Update variables to indicate that word_count words were processed.
1442   inline void update_state(size_t word_count);
1443 
1444  protected:
1445   ObjectStartArray* const _start_array;
1446   HeapWord*               _destination;         // Next addr to be written.
1447 };
1448 
1449 inline
1450 MoveAndUpdateClosure::MoveAndUpdateClosure(ParMarkBitMap* bitmap,
1451                                            ParCompactionManager* cm,
1452                                            ObjectStartArray* start_array,
1453                                            HeapWord* destination,
1454                                            size_t words) :
1455   ParMarkBitMapClosure(bitmap, cm, words), _start_array(start_array)
1456 {
1457   _destination = destination;
1458 }
1459 
1460 inline void MoveAndUpdateClosure::update_state(size_t words)
1461 {
1462   decrement_words_remaining(words);
1463   _source += words;
1464   _destination += words;
1465 }
1466 
1467 class UpdateOnlyClosure: public ParMarkBitMapClosure {
1468  private:
1469   const PSParallelCompact::SpaceId _space_id;
1470   ObjectStartArray* const          _start_array;
1471 
1472  public:
1473   UpdateOnlyClosure(ParMarkBitMap* mbm,
1474                     ParCompactionManager* cm,
1475                     PSParallelCompact::SpaceId space_id);
1476 
1477   // Update the object.
1478   virtual IterationStatus do_addr(HeapWord* addr, size_t words);
1479 
1480   inline void do_addr(HeapWord* addr);
1481 };
1482 
1483 class FillClosure: public ParMarkBitMapClosure
1484 {
1485 public:
1486   FillClosure(ParCompactionManager* cm, PSParallelCompact::SpaceId space_id) :
1487     ParMarkBitMapClosure(PSParallelCompact::mark_bitmap(), cm),
1488     _start_array(PSParallelCompact::start_array(space_id))
1489   {
1490     assert(space_id == PSParallelCompact::old_space_id,
1491            "cannot use FillClosure in the young gen");
1492   }
1493 
1494   virtual IterationStatus do_addr(HeapWord* addr, size_t size) {
1495     CollectedHeap::fill_with_objects(addr, size);
1496     HeapWord* const end = addr + size;
1497     do {
1498       _start_array->allocate_block(addr);
1499       addr += oop(addr)->size();
1500     } while (addr < end);
1501     return ParMarkBitMap::incomplete;
1502   }
1503 
1504 private:
1505   ObjectStartArray* const _start_array;
1506 };
1507 
1508 #endif // SHARE_VM_GC_IMPLEMENTATION_PARALLELSCAVENGE_PSPARALLELCOMPACT_HPP