rev 57222 : imported patch shadow-regions-final-fixes

   1 /*
   2  * Copyright (c) 2005, 2019, Oracle and/or its affiliates. All rights reserved.
   3  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
   4  *
   5  * This code is free software; you can redistribute it and/or modify it
   6  * under the terms of the GNU General Public License version 2 only, as
   7  * published by the Free Software Foundation.
   8  *
   9  * This code is distributed in the hope that it will be useful, but WITHOUT
  10  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
  11  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
  12  * version 2 for more details (a copy is included in the LICENSE file that
  13  * accompanied this code).
  14  *
  15  * You should have received a copy of the GNU General Public License version
  16  * 2 along with this work; if not, write to the Free Software Foundation,
  17  * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
  18  *
  19  * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
  20  * or visit www.oracle.com if you need additional information or have any
  21  * questions.
  22  *
  23  */
  24 
  25 #ifndef SHARE_GC_PARALLEL_PSPARALLELCOMPACT_HPP
  26 #define SHARE_GC_PARALLEL_PSPARALLELCOMPACT_HPP
  27 
  28 #include "gc/parallel/mutableSpace.hpp"
  29 #include "gc/parallel/objectStartArray.hpp"
  30 #include "gc/parallel/parMarkBitMap.hpp"
  31 #include "gc/parallel/parallelScavengeHeap.hpp"
  32 #include "gc/shared/collectedHeap.hpp"
  33 #include "gc/shared/collectorCounters.hpp"
  34 #include "oops/oop.hpp"
  35 #include "runtime/atomic.hpp"
  36 #include "runtime/orderAccess.hpp"
  37 
  38 class ParallelScavengeHeap;
  39 class PSAdaptiveSizePolicy;
  40 class PSYoungGen;
  41 class PSOldGen;
  42 class ParCompactionManager;
  43 class ParallelTaskTerminator;
  44 class PSParallelCompact;
  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     // Reuse _source_region to store the corresponding shadow region index
 245     size_t shadow_region() const { return _source_region; }
 246 
 247     // The object (if any) starting in this region and ending in a different
 248     // region that could not be updated during the main (parallel) compaction
 249     // phase.  This is different from _partial_obj_addr, which is an object that
 250     // extends onto a source region.  However, the two uses do not overlap in
 251     // time, so the same field is used to save space.
 252     HeapWord* deferred_obj_addr() const { return _partial_obj_addr; }
 253 
 254     // The starting address of the partial object extending onto the region.
 255     HeapWord* partial_obj_addr() const { return _partial_obj_addr; }
 256 
 257     // Size of the partial object extending onto the region (words).
 258     size_t partial_obj_size() const { return _partial_obj_size; }
 259 
 260     // Size of live data that lies within this region due to objects that start
 261     // in this region (words).  This does not include the partial object
 262     // extending onto the region (if any), or the part of an object that extends
 263     // onto the next region (if any).
 264     size_t live_obj_size() const { return _dc_and_los & los_mask; }
 265 
 266     // Total live data that lies within the region (words).
 267     size_t data_size() const { return partial_obj_size() + live_obj_size(); }
 268 
 269     // The destination_count is the number of other regions to which data from
 270     // this region will be copied.  At the end of the summary phase, the valid
 271     // values of destination_count are
 272     //
 273     // 0 - data from the region will be compacted completely into itself, or the
 274     //     region is empty.  The region can be claimed and then filled.
 275     // 1 - data from the region will be compacted into 1 other region; some
 276     //     data from the region may also be compacted into the region itself.
 277     // 2 - data from the region will be copied to 2 other regions.
 278     //
 279     // During compaction as regions are emptied, the destination_count is
 280     // decremented (atomically) and when it reaches 0, it can be claimed and
 281     // then filled.
 282     //
 283     // A region is claimed for processing by atomically changing the
 284     // destination_count to the claimed value (dc_claimed).  After a region has
 285     // been filled, the destination_count should be set to the completed value
 286     // (dc_completed).
 287     inline uint destination_count() const;
 288     inline uint destination_count_raw() const;
 289 
 290     // Whether the block table for this region has been filled.
 291     inline bool blocks_filled() const;
 292 
 293     // Number of times the block table was filled.
 294     DEBUG_ONLY(inline size_t blocks_filled_count() const;)
 295 
 296     // The location of the java heap data that corresponds to this region.
 297     inline HeapWord* data_location() const;
 298 
 299     // The highest address referenced by objects in this region.
 300     inline HeapWord* highest_ref() const;
 301 
 302     // Whether this region is available to be claimed, has been claimed, or has
 303     // been completed.
 304     //
 305     // Minor subtlety:  claimed() returns true if the region is marked
 306     // completed(), which is desirable since a region must be claimed before it
 307     // can be completed.
 308     bool available() const { return _dc_and_los < dc_one; }
 309     bool claimed()   const { return _dc_and_los >= dc_claimed; }
 310     bool completed() const { return _dc_and_los >= dc_completed; }
 311 
 312     // These are not atomic.
 313     void set_destination(HeapWord* addr)       { _destination = addr; }
 314     void set_source_region(size_t region)      { _source_region = region; }
 315     void set_shadow_region(size_t region)      { _source_region = region; }
 316     void set_deferred_obj_addr(HeapWord* addr) { _partial_obj_addr = addr; }
 317     void set_partial_obj_addr(HeapWord* addr)  { _partial_obj_addr = addr; }
 318     void set_partial_obj_size(size_t words)    {
 319       _partial_obj_size = (region_sz_t) words;
 320     }
 321     inline void set_blocks_filled();
 322 
 323     inline void set_destination_count(uint count);
 324     inline void set_live_obj_size(size_t words);
 325     inline void set_data_location(HeapWord* addr);
 326     inline void set_completed();
 327     inline bool claim_unsafe();
 328 
 329     // These are atomic.
 330     inline void add_live_obj(size_t words);
 331     inline void set_highest_ref(HeapWord* addr);
 332     inline void decrement_destination_count();
 333     inline bool claim();
 334 
 335     // Possible values of _shadow_state, and transition is as follows
 336     // Normal Path:
 337     // UnusedRegion -> mark_normal() -> NormalRegion
 338     // Shadow Path:
 339     // UnusedRegion -> mark_shadow() -> ShadowRegion ->
 340     // mark_filled() -> FilledShadow -> mark_copied() -> CopiedShadow
 341     static const int UnusedRegion = 0; // The region is not collected yet
 342     static const int ShadowRegion = 1; // Stolen by an idle thread, and a shadow region is created for it
 343     static const int FilledShadow = 2; // Its shadow region has been filled and ready to be copied back
 344     static const int CopiedShadow = 3; // The data of the shadow region has been copied back
 345     static const int NormalRegion = 4; // The region will be collected by the original parallel algorithm
 346 
 347     // Mark the current region as normal or shadow to enter different processing paths
 348     inline bool mark_normal();
 349     inline bool mark_shadow();
 350     // Mark the shadow region as filled and ready to be copied back
 351     inline void mark_filled();
 352     // Mark the shadow region as copied back to avoid double copying.
 353     inline bool mark_copied();
 354     // Special case: see the comment in PSParallelCompact::fill_and_update_shadow_region.
 355     // Return to the normal path here
 356     inline void shadow_to_normal();
 357 
 358 
 359     int shadow_state() { return _shadow_state; }
 360 
 361   private:
 362     // The type used to represent object sizes within a region.
 363     typedef uint region_sz_t;
 364 
 365     // Constants for manipulating the _dc_and_los field, which holds both the
 366     // destination count and live obj size.  The live obj size lives at the
 367     // least significant end so no masking is necessary when adding.
 368     static const region_sz_t dc_shift;           // Shift amount.
 369     static const region_sz_t dc_mask;            // Mask for destination count.
 370     static const region_sz_t dc_one;             // 1, shifted appropriately.
 371     static const region_sz_t dc_claimed;         // Region has been claimed.
 372     static const region_sz_t dc_completed;       // Region has been completed.
 373     static const region_sz_t los_mask;           // Mask for live obj size.
 374 
 375     HeapWord*            _destination;
 376     size_t               _source_region;
 377     HeapWord*            _partial_obj_addr;
 378     region_sz_t          _partial_obj_size;
 379     region_sz_t volatile _dc_and_los;
 380     bool        volatile _blocks_filled;
 381     int         volatile _shadow_state;
 382 
 383 #ifdef ASSERT
 384     size_t               _blocks_filled_count;   // Number of block table fills.
 385 
 386     // These enable optimizations that are only partially implemented.  Use
 387     // debug builds to prevent the code fragments from breaking.
 388     HeapWord*            _data_location;
 389     HeapWord*            _highest_ref;
 390 #endif  // #ifdef ASSERT
 391 
 392 #ifdef ASSERT
 393    public:
 394     uint                 _pushed;   // 0 until region is pushed onto a stack
 395    private:
 396 #endif
 397   };
 398 
 399   // "Blocks" allow shorter sections of the bitmap to be searched.  Each Block
 400   // holds an offset, which is the amount of live data in the Region to the left
 401   // of the first live object that starts in the Block.
 402   class BlockData
 403   {
 404   public:
 405     typedef unsigned short int blk_ofs_t;
 406 
 407     blk_ofs_t offset() const    { return _offset; }
 408     void set_offset(size_t val) { _offset = (blk_ofs_t)val; }
 409 
 410   private:
 411     blk_ofs_t _offset;
 412   };
 413 
 414 public:
 415   ParallelCompactData();
 416   bool initialize(MemRegion covered_region);
 417 
 418   size_t region_count() const { return _region_count; }
 419   size_t reserved_byte_size() const { return _reserved_byte_size; }
 420 
 421   // Convert region indices to/from RegionData pointers.
 422   inline RegionData* region(size_t region_idx) const;
 423   inline size_t     region(const RegionData* const region_ptr) const;
 424 
 425   size_t block_count() const { return _block_count; }
 426   inline BlockData* block(size_t block_idx) const;
 427   inline size_t     block(const BlockData* block_ptr) const;
 428 
 429   void add_obj(HeapWord* addr, size_t len);
 430   void add_obj(oop p, size_t len) { add_obj((HeapWord*)p, len); }
 431 
 432   // Fill in the regions covering [beg, end) so that no data moves; i.e., the
 433   // destination of region n is simply the start of region n.  The argument beg
 434   // must be region-aligned; end need not be.
 435   void summarize_dense_prefix(HeapWord* beg, HeapWord* end);
 436 
 437   HeapWord* summarize_split_space(size_t src_region, SplitInfo& split_info,
 438                                   HeapWord* destination, HeapWord* target_end,
 439                                   HeapWord** target_next);
 440   bool summarize(SplitInfo& split_info,
 441                  HeapWord* source_beg, HeapWord* source_end,
 442                  HeapWord** source_next,
 443                  HeapWord* target_beg, HeapWord* target_end,
 444                  HeapWord** target_next);
 445 
 446   void clear();
 447   void clear_range(size_t beg_region, size_t end_region);
 448   void clear_range(HeapWord* beg, HeapWord* end) {
 449     clear_range(addr_to_region_idx(beg), addr_to_region_idx(end));
 450   }
 451 
 452   // Return the number of words between addr and the start of the region
 453   // containing addr.
 454   inline size_t     region_offset(const HeapWord* addr) const;
 455 
 456   // Convert addresses to/from a region index or region pointer.
 457   inline size_t     addr_to_region_idx(const HeapWord* addr) const;
 458   inline RegionData* addr_to_region_ptr(const HeapWord* addr) const;
 459   inline HeapWord*  region_to_addr(size_t region) const;
 460   inline HeapWord*  region_to_addr(size_t region, size_t offset) const;
 461   inline HeapWord*  region_to_addr(const RegionData* region) const;
 462 
 463   inline HeapWord*  region_align_down(HeapWord* addr) const;
 464   inline HeapWord*  region_align_up(HeapWord* addr) const;
 465   inline bool       is_region_aligned(HeapWord* addr) const;
 466 
 467   // Analogous to region_offset() for blocks.
 468   size_t     block_offset(const HeapWord* addr) const;
 469   size_t     addr_to_block_idx(const HeapWord* addr) const;
 470   size_t     addr_to_block_idx(const oop obj) const {
 471     return addr_to_block_idx((HeapWord*) obj);
 472   }
 473   inline BlockData* addr_to_block_ptr(const HeapWord* addr) const;
 474   inline HeapWord*  block_to_addr(size_t block) const;
 475   inline size_t     region_to_block_idx(size_t region) const;
 476 
 477   inline HeapWord*  block_align_down(HeapWord* addr) const;
 478   inline HeapWord*  block_align_up(HeapWord* addr) const;
 479   inline bool       is_block_aligned(HeapWord* addr) const;
 480 
 481   // Return the address one past the end of the partial object.
 482   HeapWord* partial_obj_end(size_t region_idx) const;
 483 
 484   // Return the location of the object after compaction.
 485   HeapWord* calc_new_pointer(HeapWord* addr, ParCompactionManager* cm);
 486 
 487   HeapWord* calc_new_pointer(oop p, ParCompactionManager* cm) {
 488     return calc_new_pointer((HeapWord*) p, cm);
 489   }
 490 
 491 #ifdef  ASSERT
 492   void verify_clear(const PSVirtualSpace* vspace);
 493   void verify_clear();
 494 #endif  // #ifdef ASSERT
 495 
 496 private:
 497   bool initialize_block_data();
 498   bool initialize_region_data(size_t region_size);
 499   PSVirtualSpace* create_vspace(size_t count, size_t element_size);
 500 
 501 private:
 502   HeapWord*       _region_start;
 503 #ifdef  ASSERT
 504   HeapWord*       _region_end;
 505 #endif  // #ifdef ASSERT
 506 
 507   PSVirtualSpace* _region_vspace;
 508   size_t          _reserved_byte_size;
 509   RegionData*     _region_data;
 510   size_t          _region_count;
 511 
 512   PSVirtualSpace* _block_vspace;
 513   BlockData*      _block_data;
 514   size_t          _block_count;
 515 };
 516 
 517 inline uint
 518 ParallelCompactData::RegionData::destination_count_raw() const
 519 {
 520   return _dc_and_los & dc_mask;
 521 }
 522 
 523 inline uint
 524 ParallelCompactData::RegionData::destination_count() const
 525 {
 526   return destination_count_raw() >> dc_shift;
 527 }
 528 
 529 inline bool
 530 ParallelCompactData::RegionData::blocks_filled() const
 531 {
 532   bool result = _blocks_filled;
 533   OrderAccess::acquire();
 534   return result;
 535 }
 536 
 537 #ifdef ASSERT
 538 inline size_t
 539 ParallelCompactData::RegionData::blocks_filled_count() const
 540 {
 541   return _blocks_filled_count;
 542 }
 543 #endif // #ifdef ASSERT
 544 
 545 inline void
 546 ParallelCompactData::RegionData::set_blocks_filled()
 547 {
 548   OrderAccess::release();
 549   _blocks_filled = true;
 550   // Debug builds count the number of times the table was filled.
 551   DEBUG_ONLY(Atomic::inc(&_blocks_filled_count));
 552 }
 553 
 554 inline void
 555 ParallelCompactData::RegionData::set_destination_count(uint count)
 556 {
 557   assert(count <= (dc_completed >> dc_shift), "count too large");
 558   const region_sz_t live_sz = (region_sz_t) live_obj_size();
 559   _dc_and_los = (count << dc_shift) | live_sz;
 560 }
 561 
 562 inline void ParallelCompactData::RegionData::set_live_obj_size(size_t words)
 563 {
 564   assert(words <= los_mask, "would overflow");
 565   _dc_and_los = destination_count_raw() | (region_sz_t)words;
 566 }
 567 
 568 inline void ParallelCompactData::RegionData::decrement_destination_count()
 569 {
 570   assert(_dc_and_los < dc_claimed, "already claimed");
 571   assert(_dc_and_los >= dc_one, "count would go negative");
 572   Atomic::add(&_dc_and_los, dc_mask);
 573 }
 574 
 575 inline HeapWord* ParallelCompactData::RegionData::data_location() const
 576 {
 577   DEBUG_ONLY(return _data_location;)
 578   NOT_DEBUG(return NULL;)
 579 }
 580 
 581 inline HeapWord* ParallelCompactData::RegionData::highest_ref() const
 582 {
 583   DEBUG_ONLY(return _highest_ref;)
 584   NOT_DEBUG(return NULL;)
 585 }
 586 
 587 inline void ParallelCompactData::RegionData::set_data_location(HeapWord* addr)
 588 {
 589   DEBUG_ONLY(_data_location = addr;)
 590 }
 591 
 592 inline void ParallelCompactData::RegionData::set_completed()
 593 {
 594   assert(claimed(), "must be claimed first");
 595   _dc_and_los = dc_completed | (region_sz_t) live_obj_size();
 596 }
 597 
 598 // MT-unsafe claiming of a region.  Should only be used during single threaded
 599 // execution.
 600 inline bool ParallelCompactData::RegionData::claim_unsafe()
 601 {
 602   if (available()) {
 603     _dc_and_los |= dc_claimed;
 604     return true;
 605   }
 606   return false;
 607 }
 608 
 609 inline void ParallelCompactData::RegionData::add_live_obj(size_t words)
 610 {
 611   assert(words <= (size_t)los_mask - live_obj_size(), "overflow");
 612   Atomic::add(&_dc_and_los, static_cast<region_sz_t>(words));
 613 }
 614 
 615 inline void ParallelCompactData::RegionData::set_highest_ref(HeapWord* addr)
 616 {
 617 #ifdef ASSERT
 618   HeapWord* tmp = _highest_ref;
 619   while (addr > tmp) {
 620     tmp = Atomic::cmpxchg(&_highest_ref, tmp, addr);
 621   }
 622 #endif  // #ifdef ASSERT
 623 }
 624 
 625 inline bool ParallelCompactData::RegionData::claim()
 626 {
 627   const region_sz_t los = static_cast<region_sz_t>(live_obj_size());
 628   const region_sz_t old = Atomic::cmpxchg(&_dc_and_los, los, dc_claimed | los);
 629   return old == los;
 630 }
 631 
 632 inline bool ParallelCompactData::RegionData::mark_normal() {
 633   return Atomic::cmpxchg(&_shadow_state, UnusedRegion, NormalRegion, memory_order_relaxed) == UnusedRegion;
 634 }
 635 
 636 inline bool ParallelCompactData::RegionData::mark_shadow() {
 637   if (_shadow_state != UnusedRegion) return false;
 638   return Atomic::cmpxchg(&_shadow_state, UnusedRegion, ShadowRegion, memory_order_relaxed) == UnusedRegion;
 639 }
 640 
 641 inline void ParallelCompactData::RegionData::mark_filled() {
 642   int old = Atomic::cmpxchg(&_shadow_state, ShadowRegion, FilledShadow, memory_order_relaxed);
 643   assert(old == ShadowRegion, "Fail to mark the region as filled");
 644 }
 645 
 646 inline bool ParallelCompactData::RegionData::mark_copied() {
 647   return Atomic::cmpxchg(&_shadow_state, FilledShadow, CopiedShadow, memory_order_relaxed) == FilledShadow;
 648 }
 649 
 650 void ParallelCompactData::RegionData::shadow_to_normal() {
 651   int old = Atomic::cmpxchg(&_shadow_state, ShadowRegion, NormalRegion, memory_order_relaxed);
 652   assert(old == ShadowRegion, "Fail to mark the region as finish");
 653 }
 654 
 655 inline ParallelCompactData::RegionData*
 656 ParallelCompactData::region(size_t region_idx) const
 657 {
 658   assert(region_idx <= region_count(), "bad arg");
 659   return _region_data + region_idx;
 660 }
 661 
 662 inline size_t
 663 ParallelCompactData::region(const RegionData* const region_ptr) const
 664 {
 665   assert(region_ptr >= _region_data, "bad arg");
 666   assert(region_ptr <= _region_data + region_count(), "bad arg");
 667   return pointer_delta(region_ptr, _region_data, sizeof(RegionData));
 668 }
 669 
 670 inline ParallelCompactData::BlockData*
 671 ParallelCompactData::block(size_t n) const {
 672   assert(n < block_count(), "bad arg");
 673   return _block_data + n;
 674 }
 675 
 676 inline size_t
 677 ParallelCompactData::region_offset(const HeapWord* addr) const
 678 {
 679   assert(addr >= _region_start, "bad addr");
 680   assert(addr <= _region_end, "bad addr");
 681   return (size_t(addr) & RegionAddrOffsetMask) >> LogHeapWordSize;
 682 }
 683 
 684 inline size_t
 685 ParallelCompactData::addr_to_region_idx(const HeapWord* addr) const
 686 {
 687   assert(addr >= _region_start, "bad addr " PTR_FORMAT " _region_start " PTR_FORMAT, p2i(addr), p2i(_region_start));
 688   assert(addr <= _region_end, "bad addr " PTR_FORMAT " _region_end " PTR_FORMAT, p2i(addr), p2i(_region_end));
 689   return pointer_delta(addr, _region_start) >> Log2RegionSize;
 690 }
 691 
 692 inline ParallelCompactData::RegionData*
 693 ParallelCompactData::addr_to_region_ptr(const HeapWord* addr) const
 694 {
 695   return region(addr_to_region_idx(addr));
 696 }
 697 
 698 inline HeapWord*
 699 ParallelCompactData::region_to_addr(size_t region) const
 700 {
 701   assert(region <= _region_count, "region out of range");
 702   return _region_start + (region << Log2RegionSize);
 703 }
 704 
 705 inline HeapWord*
 706 ParallelCompactData::region_to_addr(const RegionData* region) const
 707 {
 708   return region_to_addr(pointer_delta(region, _region_data,
 709                                       sizeof(RegionData)));
 710 }
 711 
 712 inline HeapWord*
 713 ParallelCompactData::region_to_addr(size_t region, size_t offset) const
 714 {
 715   assert(region <= _region_count, "region out of range");
 716   assert(offset < RegionSize, "offset too big");  // This may be too strict.
 717   return region_to_addr(region) + offset;
 718 }
 719 
 720 inline HeapWord*
 721 ParallelCompactData::region_align_down(HeapWord* addr) const
 722 {
 723   assert(addr >= _region_start, "bad addr");
 724   assert(addr < _region_end + RegionSize, "bad addr");
 725   return (HeapWord*)(size_t(addr) & RegionAddrMask);
 726 }
 727 
 728 inline HeapWord*
 729 ParallelCompactData::region_align_up(HeapWord* addr) const
 730 {
 731   assert(addr >= _region_start, "bad addr");
 732   assert(addr <= _region_end, "bad addr");
 733   return region_align_down(addr + RegionSizeOffsetMask);
 734 }
 735 
 736 inline bool
 737 ParallelCompactData::is_region_aligned(HeapWord* addr) const
 738 {
 739   return region_offset(addr) == 0;
 740 }
 741 
 742 inline size_t
 743 ParallelCompactData::block_offset(const HeapWord* addr) const
 744 {
 745   assert(addr >= _region_start, "bad addr");
 746   assert(addr <= _region_end, "bad addr");
 747   return (size_t(addr) & BlockAddrOffsetMask) >> LogHeapWordSize;
 748 }
 749 
 750 inline size_t
 751 ParallelCompactData::addr_to_block_idx(const HeapWord* addr) const
 752 {
 753   assert(addr >= _region_start, "bad addr");
 754   assert(addr <= _region_end, "bad addr");
 755   return pointer_delta(addr, _region_start) >> Log2BlockSize;
 756 }
 757 
 758 inline ParallelCompactData::BlockData*
 759 ParallelCompactData::addr_to_block_ptr(const HeapWord* addr) const
 760 {
 761   return block(addr_to_block_idx(addr));
 762 }
 763 
 764 inline HeapWord*
 765 ParallelCompactData::block_to_addr(size_t block) const
 766 {
 767   assert(block < _block_count, "block out of range");
 768   return _region_start + (block << Log2BlockSize);
 769 }
 770 
 771 inline size_t
 772 ParallelCompactData::region_to_block_idx(size_t region) const
 773 {
 774   return region << Log2BlocksPerRegion;
 775 }
 776 
 777 inline HeapWord*
 778 ParallelCompactData::block_align_down(HeapWord* addr) const
 779 {
 780   assert(addr >= _region_start, "bad addr");
 781   assert(addr < _region_end + RegionSize, "bad addr");
 782   return (HeapWord*)(size_t(addr) & BlockAddrMask);
 783 }
 784 
 785 inline HeapWord*
 786 ParallelCompactData::block_align_up(HeapWord* addr) const
 787 {
 788   assert(addr >= _region_start, "bad addr");
 789   assert(addr <= _region_end, "bad addr");
 790   return block_align_down(addr + BlockSizeOffsetMask);
 791 }
 792 
 793 inline bool
 794 ParallelCompactData::is_block_aligned(HeapWord* addr) const
 795 {
 796   return block_offset(addr) == 0;
 797 }
 798 
 799 // Abstract closure for use with ParMarkBitMap::iterate(), which will invoke the
 800 // do_addr() method.
 801 //
 802 // The closure is initialized with the number of heap words to process
 803 // (words_remaining()), and becomes 'full' when it reaches 0.  The do_addr()
 804 // methods in subclasses should update the total as words are processed.  Since
 805 // only one subclass actually uses this mechanism to terminate iteration, the
 806 // default initial value is > 0.  The implementation is here and not in the
 807 // single subclass that uses it to avoid making is_full() virtual, and thus
 808 // adding a virtual call per live object.
 809 
 810 class ParMarkBitMapClosure: public StackObj {
 811  public:
 812   typedef ParMarkBitMap::idx_t idx_t;
 813   typedef ParMarkBitMap::IterationStatus IterationStatus;
 814 
 815  public:
 816   inline ParMarkBitMapClosure(ParMarkBitMap* mbm, ParCompactionManager* cm,
 817                               size_t words = max_uintx);
 818 
 819   inline ParCompactionManager* compaction_manager() const;
 820   inline ParMarkBitMap*        bitmap() const;
 821   inline size_t                words_remaining() const;
 822   inline bool                  is_full() const;
 823   inline HeapWord*             source() const;
 824 
 825   inline void                  set_source(HeapWord* addr);
 826 
 827   virtual IterationStatus do_addr(HeapWord* addr, size_t words) = 0;
 828 
 829  protected:
 830   inline void decrement_words_remaining(size_t words);
 831 
 832  private:
 833   ParMarkBitMap* const        _bitmap;
 834   ParCompactionManager* const _compaction_manager;
 835   DEBUG_ONLY(const size_t     _initial_words_remaining;) // Useful in debugger.
 836   size_t                      _words_remaining; // Words left to copy.
 837 
 838  protected:
 839   HeapWord*                   _source;          // Next addr that would be read.
 840 };
 841 
 842 inline
 843 ParMarkBitMapClosure::ParMarkBitMapClosure(ParMarkBitMap* bitmap,
 844                                            ParCompactionManager* cm,
 845                                            size_t words):
 846   _bitmap(bitmap), _compaction_manager(cm)
 847 #ifdef  ASSERT
 848   , _initial_words_remaining(words)
 849 #endif
 850 {
 851   _words_remaining = words;
 852   _source = NULL;
 853 }
 854 
 855 inline ParCompactionManager* ParMarkBitMapClosure::compaction_manager() const {
 856   return _compaction_manager;
 857 }
 858 
 859 inline ParMarkBitMap* ParMarkBitMapClosure::bitmap() const {
 860   return _bitmap;
 861 }
 862 
 863 inline size_t ParMarkBitMapClosure::words_remaining() const {
 864   return _words_remaining;
 865 }
 866 
 867 inline bool ParMarkBitMapClosure::is_full() const {
 868   return words_remaining() == 0;
 869 }
 870 
 871 inline HeapWord* ParMarkBitMapClosure::source() const {
 872   return _source;
 873 }
 874 
 875 inline void ParMarkBitMapClosure::set_source(HeapWord* addr) {
 876   _source = addr;
 877 }
 878 
 879 inline void ParMarkBitMapClosure::decrement_words_remaining(size_t words) {
 880   assert(_words_remaining >= words, "processed too many words");
 881   _words_remaining -= words;
 882 }
 883 
 884 // The UseParallelOldGC collector is a stop-the-world garbage collector that
 885 // does parts of the collection using parallel threads.  The collection includes
 886 // the tenured generation and the young generation.  The permanent generation is
 887 // collected at the same time as the other two generations but the permanent
 888 // generation is collect by a single GC thread.  The permanent generation is
 889 // collected serially because of the requirement that during the processing of a
 890 // klass AAA, any objects reference by AAA must already have been processed.
 891 // This requirement is enforced by a left (lower address) to right (higher
 892 // address) sliding compaction.
 893 //
 894 // There are four phases of the collection.
 895 //
 896 //      - marking phase
 897 //      - summary phase
 898 //      - compacting phase
 899 //      - clean up phase
 900 //
 901 // Roughly speaking these phases correspond, respectively, to
 902 //      - mark all the live objects
 903 //      - calculate the destination of each object at the end of the collection
 904 //      - move the objects to their destination
 905 //      - update some references and reinitialize some variables
 906 //
 907 // These three phases are invoked in PSParallelCompact::invoke_no_policy().  The
 908 // marking phase is implemented in PSParallelCompact::marking_phase() and does a
 909 // complete marking of the heap.  The summary phase is implemented in
 910 // PSParallelCompact::summary_phase().  The move and update phase is implemented
 911 // in PSParallelCompact::compact().
 912 //
 913 // A space that is being collected is divided into regions and with each region
 914 // is associated an object of type ParallelCompactData.  Each region is of a
 915 // fixed size and typically will contain more than 1 object and may have parts
 916 // of objects at the front and back of the region.
 917 //
 918 // region            -----+---------------------+----------
 919 // objects covered   [ AAA  )[ BBB )[ CCC   )[ DDD     )
 920 //
 921 // The marking phase does a complete marking of all live objects in the heap.
 922 // The marking also compiles the size of the data for all live objects covered
 923 // by the region.  This size includes the part of any live object spanning onto
 924 // the region (part of AAA if it is live) from the front, all live objects
 925 // contained in the region (BBB and/or CCC if they are live), and the part of
 926 // any live objects covered by the region that extends off the region (part of
 927 // DDD if it is live).  The marking phase uses multiple GC threads and marking
 928 // is done in a bit array of type ParMarkBitMap.  The marking of the bit map is
 929 // done atomically as is the accumulation of the size of the live objects
 930 // covered by a region.
 931 //
 932 // The summary phase calculates the total live data to the left of each region
 933 // XXX.  Based on that total and the bottom of the space, it can calculate the
 934 // starting location of the live data in XXX.  The summary phase calculates for
 935 // each region XXX quantities such as
 936 //
 937 //      - the amount of live data at the beginning of a region from an object
 938 //        entering the region.
 939 //      - the location of the first live data on the region
 940 //      - a count of the number of regions receiving live data from XXX.
 941 //
 942 // See ParallelCompactData for precise details.  The summary phase also
 943 // calculates the dense prefix for the compaction.  The dense prefix is a
 944 // portion at the beginning of the space that is not moved.  The objects in the
 945 // dense prefix do need to have their object references updated.  See method
 946 // summarize_dense_prefix().
 947 //
 948 // The summary phase is done using 1 GC thread.
 949 //
 950 // The compaction phase moves objects to their new location and updates all
 951 // references in the object.
 952 //
 953 // A current exception is that objects that cross a region boundary are moved
 954 // but do not have their references updated.  References are not updated because
 955 // it cannot easily be determined if the klass pointer KKK for the object AAA
 956 // has been updated.  KKK likely resides in a region to the left of the region
 957 // containing AAA.  These AAA's have there references updated at the end in a
 958 // clean up phase.  See the method PSParallelCompact::update_deferred_objects().
 959 // An alternate strategy is being investigated for this deferral of updating.
 960 //
 961 // Compaction is done on a region basis.  A region that is ready to be filled is
 962 // put on a ready list and GC threads take region off the list and fill them.  A
 963 // region is ready to be filled if it empty of live objects.  Such a region may
 964 // have been initially empty (only contained dead objects) or may have had all
 965 // its live objects copied out already.  A region that compacts into itself is
 966 // also ready for filling.  The ready list is initially filled with empty
 967 // regions and regions compacting into themselves.  There is always at least 1
 968 // region that can be put on the ready list.  The regions are atomically added
 969 // and removed from the ready list.
 970 
 971 class TaskQueue;
 972 
 973 class PSParallelCompact : AllStatic {
 974  public:
 975   // Convenient access to type names.
 976   typedef ParMarkBitMap::idx_t idx_t;
 977   typedef ParallelCompactData::RegionData RegionData;
 978   typedef ParallelCompactData::BlockData BlockData;
 979 
 980   typedef enum {
 981     old_space_id, eden_space_id,
 982     from_space_id, to_space_id, last_space_id
 983   } SpaceId;
 984 
 985   struct UpdateDensePrefixTask : public CHeapObj<mtGC> {
 986     SpaceId _space_id;
 987     size_t _region_index_start;
 988     size_t _region_index_end;
 989 
 990     UpdateDensePrefixTask() :
 991         _space_id(SpaceId(0)),
 992         _region_index_start(0),
 993         _region_index_end(0) {}
 994 
 995     UpdateDensePrefixTask(SpaceId space_id,
 996                           size_t region_index_start,
 997                           size_t region_index_end) :
 998         _space_id(space_id),
 999         _region_index_start(region_index_start),
1000         _region_index_end(region_index_end) {}
1001   };
1002 
1003  public:
1004   // Inline closure decls
1005   //
1006   class IsAliveClosure: public BoolObjectClosure {
1007    public:
1008     virtual bool do_object_b(oop p);
1009   };
1010 
1011   friend class RefProcTaskProxy;
1012   friend class PSParallelCompactTest;
1013 
1014  private:
1015   static STWGCTimer           _gc_timer;
1016   static ParallelOldTracer    _gc_tracer;
1017   static elapsedTimer         _accumulated_time;
1018   static unsigned int         _total_invocations;
1019   static unsigned int         _maximum_compaction_gc_num;
1020   static jlong                _time_of_last_gc;   // ms
1021   static CollectorCounters*   _counters;
1022   static ParMarkBitMap        _mark_bitmap;
1023   static ParallelCompactData  _summary_data;
1024   static IsAliveClosure       _is_alive_closure;
1025   static SpaceInfo            _space_info[last_space_id];
1026 
1027   // Reference processing (used in ...follow_contents)
1028   static SpanSubjectToDiscoveryClosure  _span_based_discoverer;
1029   static ReferenceProcessor*  _ref_processor;
1030 
1031   // Values computed at initialization and used by dead_wood_limiter().
1032   static double _dwl_mean;
1033   static double _dwl_std_dev;
1034   static double _dwl_first_term;
1035   static double _dwl_adjustment;
1036 #ifdef  ASSERT
1037   static bool   _dwl_initialized;
1038 #endif  // #ifdef ASSERT
1039 
1040  public:
1041   static ParallelOldTracer* gc_tracer() { return &_gc_tracer; }
1042 
1043  private:
1044 
1045   static void initialize_space_info();
1046 
1047   // Clear the marking bitmap and summary data that cover the specified space.
1048   static void clear_data_covering_space(SpaceId id);
1049 
1050   static void pre_compact();
1051   static void post_compact();
1052 
1053   // Mark live objects
1054   static void marking_phase(ParCompactionManager* cm,
1055                             bool maximum_heap_compaction,
1056                             ParallelOldTracer *gc_tracer);
1057 
1058   // Compute the dense prefix for the designated space.  This is an experimental
1059   // implementation currently not used in production.
1060   static HeapWord* compute_dense_prefix_via_density(const SpaceId id,
1061                                                     bool maximum_compaction);
1062 
1063   // Methods used to compute the dense prefix.
1064 
1065   // Compute the value of the normal distribution at x = density.  The mean and
1066   // standard deviation are values saved by initialize_dead_wood_limiter().
1067   static inline double normal_distribution(double density);
1068 
1069   // Initialize the static vars used by dead_wood_limiter().
1070   static void initialize_dead_wood_limiter();
1071 
1072   // Return the percentage of space that can be treated as "dead wood" (i.e.,
1073   // not reclaimed).
1074   static double dead_wood_limiter(double density, size_t min_percent);
1075 
1076   // Find the first (left-most) region in the range [beg, end) that has at least
1077   // dead_words of dead space to the left.  The argument beg must be the first
1078   // region in the space that is not completely live.
1079   static RegionData* dead_wood_limit_region(const RegionData* beg,
1080                                             const RegionData* end,
1081                                             size_t dead_words);
1082 
1083   // Return a pointer to the first region in the range [beg, end) that is not
1084   // completely full.
1085   static RegionData* first_dead_space_region(const RegionData* beg,
1086                                              const RegionData* end);
1087 
1088   // Return a value indicating the benefit or 'yield' if the compacted region
1089   // were to start (or equivalently if the dense prefix were to end) at the
1090   // candidate region.  Higher values are better.
1091   //
1092   // The value is based on the amount of space reclaimed vs. the costs of (a)
1093   // updating references in the dense prefix plus (b) copying objects and
1094   // updating references in the compacted region.
1095   static inline double reclaimed_ratio(const RegionData* const candidate,
1096                                        HeapWord* const bottom,
1097                                        HeapWord* const top,
1098                                        HeapWord* const new_top);
1099 
1100   // Compute the dense prefix for the designated space.
1101   static HeapWord* compute_dense_prefix(const SpaceId id,
1102                                         bool maximum_compaction);
1103 
1104   // Return true if dead space crosses onto the specified Region; bit must be
1105   // the bit index corresponding to the first word of the Region.
1106   static inline bool dead_space_crosses_boundary(const RegionData* region,
1107                                                  idx_t bit);
1108 
1109   // Summary phase utility routine to fill dead space (if any) at the dense
1110   // prefix boundary.  Should only be called if the the dense prefix is
1111   // non-empty.
1112   static void fill_dense_prefix_end(SpaceId id);
1113 
1114   static void summarize_spaces_quick();
1115   static void summarize_space(SpaceId id, bool maximum_compaction);
1116   static void summary_phase(ParCompactionManager* cm, bool maximum_compaction);
1117 
1118   // Adjust addresses in roots.  Does not adjust addresses in heap.
1119   static void adjust_roots(ParCompactionManager* cm);
1120 
1121   DEBUG_ONLY(static void write_block_fill_histogram();)
1122 
1123   // Move objects to new locations.
1124   static void compact_perm(ParCompactionManager* cm);
1125   static void compact();
1126 
1127   // Add available regions to the stack and draining tasks to the task queue.
1128   static void prepare_region_draining_tasks(uint parallel_gc_threads);
1129 
1130   // Add dense prefix update tasks to the task queue.
1131   static void enqueue_dense_prefix_tasks(TaskQueue& task_queue,
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 #ifndef PRODUCT
1144   // Print generic summary data
1145   static void print_generic_summary_data(ParallelCompactData& summary_data,
1146                                          HeapWord* const beg_addr,
1147                                          HeapWord* const end_addr);
1148 #endif  // #ifndef PRODUCT
1149 
1150  public:
1151 
1152   PSParallelCompact();
1153 
1154   static void invoke(bool maximum_heap_compaction);
1155   static bool invoke_no_policy(bool maximum_heap_compaction);
1156 
1157   static void post_initialize();
1158   // Perform initialization for PSParallelCompact that requires
1159   // allocations.  This should be called during the VM initialization
1160   // at a pointer where it would be appropriate to return a JNI_ENOMEM
1161   // in the event of a failure.
1162   static bool initialize();
1163 
1164   // Closure accessors
1165   static BoolObjectClosure* is_alive_closure()     { return (BoolObjectClosure*)&_is_alive_closure; }
1166 
1167   // Public accessors
1168   static elapsedTimer* accumulated_time() { return &_accumulated_time; }
1169   static unsigned int total_invocations() { return _total_invocations; }
1170   static CollectorCounters* counters()    { return _counters; }
1171 
1172   // Marking support
1173   static inline bool mark_obj(oop obj);
1174   static inline bool is_marked(oop obj);
1175 
1176   template <class T> static inline void adjust_pointer(T* p, ParCompactionManager* cm);
1177 
1178   // Compaction support.
1179   // Return true if p is in the range [beg_addr, end_addr).
1180   static inline bool is_in(HeapWord* p, HeapWord* beg_addr, HeapWord* end_addr);
1181   static inline bool is_in(oop* p, HeapWord* beg_addr, HeapWord* end_addr);
1182 
1183   // Convenience wrappers for per-space data kept in _space_info.
1184   static inline MutableSpace*     space(SpaceId space_id);
1185   static inline HeapWord*         new_top(SpaceId space_id);
1186   static inline HeapWord*         dense_prefix(SpaceId space_id);
1187   static inline ObjectStartArray* start_array(SpaceId space_id);
1188 
1189   // Process the end of the given region range in the dense prefix.
1190   // This includes saving any object not updated.
1191   static void dense_prefix_regions_epilogue(ParCompactionManager* cm,
1192                                             size_t region_start_index,
1193                                             size_t region_end_index,
1194                                             idx_t exiting_object_offset,
1195                                             idx_t region_offset_start,
1196                                             idx_t region_offset_end);
1197 
1198   // Update a region in the dense prefix.  For each live object
1199   // in the region, update it's interior references.  For each
1200   // dead object, fill it with deadwood. Dead space at the end
1201   // of a region range will be filled to the start of the next
1202   // live object regardless of the region_index_end.  None of the
1203   // objects in the dense prefix move and dead space is dead
1204   // (holds only dead objects that don't need any processing), so
1205   // dead space can be filled in any order.
1206   static void update_and_deadwood_in_dense_prefix(ParCompactionManager* cm,
1207                                                   SpaceId space_id,
1208                                                   size_t region_index_start,
1209                                                   size_t region_index_end);
1210 
1211   // Return the address of the count + 1st live word in the range [beg, end).
1212   static HeapWord* skip_live_words(HeapWord* beg, HeapWord* end, size_t count);
1213 
1214   // Return the address of the word to be copied to dest_addr, which must be
1215   // aligned to a region boundary.
1216   static HeapWord* first_src_addr(HeapWord* const dest_addr,
1217                                   SpaceId src_space_id,
1218                                   size_t src_region_idx);
1219 
1220   // Determine the next source region, set closure.source() to the start of the
1221   // new region return the region index.  Parameter end_addr is the address one
1222   // beyond the end of source range just processed.  If necessary, switch to a
1223   // new source space and set src_space_id (in-out parameter) and src_space_top
1224   // (out parameter) accordingly.
1225   static size_t next_src_region(MoveAndUpdateClosure& closure,
1226                                 SpaceId& src_space_id,
1227                                 HeapWord*& src_space_top,
1228                                 HeapWord* end_addr);
1229 
1230   // Decrement the destination count for each non-empty source region in the
1231   // range [beg_region, region(region_align_up(end_addr))).  If the destination
1232   // count for a region goes to 0 and it needs to be filled, enqueue it.
1233   static void decrement_destination_counts(ParCompactionManager* cm,
1234                                            SpaceId src_space_id,
1235                                            size_t beg_region,
1236                                            HeapWord* end_addr);
1237 
1238   static void fill_region(ParCompactionManager* cm, MoveAndUpdateClosure& closure, size_t region);
1239   static void fill_and_update_region(ParCompactionManager* cm, size_t region);
1240 
1241   static bool steal_unavailable_region(ParCompactionManager* cm, size_t& region_idx);
1242   static void fill_and_update_shadow_region(ParCompactionManager* cm, size_t region);
1243   // Copy the content of a shadow region back to its corresponding heap region
1244   static void copy_back(HeapWord* shadow_addr, HeapWord* region_addr);
1245   // Collect empty regions as shadow regions and initialize the
1246   // _next_shadow_region filed for each compact manager
1247   static void initialize_shadow_regions(uint parallel_gc_threads);
1248 
1249   // Fill in the block table for the specified region.
1250   static void fill_blocks(size_t region_idx);
1251 
1252   // Update the deferred objects in the space.
1253   static void update_deferred_objects(ParCompactionManager* cm, SpaceId id);
1254 
1255   static ParMarkBitMap* mark_bitmap() { return &_mark_bitmap; }
1256   static ParallelCompactData& summary_data() { return _summary_data; }
1257 
1258   // Reference Processing
1259   static ReferenceProcessor* const ref_processor() { return _ref_processor; }
1260 
1261   static STWGCTimer* gc_timer() { return &_gc_timer; }
1262 
1263   // Return the SpaceId for the given address.
1264   static SpaceId space_id(HeapWord* addr);
1265 
1266   // Time since last full gc (in milliseconds).
1267   static jlong millis_since_last_gc();
1268 
1269   static void print_on_error(outputStream* st);
1270 
1271 #ifndef PRODUCT
1272   // Debugging support.
1273   static const char* space_names[last_space_id];
1274   static void print_region_ranges();
1275   static void print_dense_prefix_stats(const char* const algorithm,
1276                                        const SpaceId id,
1277                                        const bool maximum_compaction,
1278                                        HeapWord* const addr);
1279   static void summary_phase_msg(SpaceId dst_space_id,
1280                                 HeapWord* dst_beg, HeapWord* dst_end,
1281                                 SpaceId src_space_id,
1282                                 HeapWord* src_beg, HeapWord* src_end);
1283 #endif  // #ifndef PRODUCT
1284 
1285 #ifdef  ASSERT
1286   // Sanity check the new location of a word in the heap.
1287   static inline void check_new_location(HeapWord* old_addr, HeapWord* new_addr);
1288   // Verify that all the regions have been emptied.
1289   static void verify_complete(SpaceId space_id);
1290 #endif  // #ifdef ASSERT
1291 };
1292 
1293 class MoveAndUpdateClosure: public ParMarkBitMapClosure {
1294   static inline size_t calculate_words_remaining(size_t region);
1295  public:
1296   inline MoveAndUpdateClosure(ParMarkBitMap* bitmap, ParCompactionManager* cm,
1297                               size_t region);

1298 
1299   // Accessors.
1300   HeapWord* destination() const         { return _destination; }
1301   HeapWord* copy_destination() const    { return _destination + _offset; }
1302 
1303   // If the object will fit (size <= words_remaining()), copy it to the current
1304   // destination, update the interior oops and the start array and return either
1305   // full (if the closure is full) or incomplete.  If the object will not fit,
1306   // return would_overflow.
1307   IterationStatus do_addr(HeapWord* addr, size_t size);
1308 
1309   // Copy enough words to fill this closure, starting at source().  Interior
1310   // oops and the start array are not updated.  Return full.
1311   IterationStatus copy_until_full();
1312 
1313   // Copy enough words to fill this closure or to the end of an object,
1314   // whichever is smaller, starting at source().  Interior oops and the start
1315   // array are not updated.
1316   void copy_partial_obj();
1317 
1318   virtual void complete_region(ParCompactionManager* cm, HeapWord* dest_addr,
1319                                PSParallelCompact::RegionData* region_ptr);
1320 
1321 protected:
1322   // Update variables to indicate that word_count words were processed.
1323   inline void update_state(size_t word_count);
1324 
1325  protected:

1326   HeapWord*               _destination;         // Next addr to be written.
1327   ObjectStartArray* const _start_array;
1328   size_t                  _offset;
1329 };
1330 
1331 inline size_t MoveAndUpdateClosure::calculate_words_remaining(size_t region) {
1332   HeapWord* dest_addr = PSParallelCompact::summary_data().region_to_addr(region);
1333   PSParallelCompact::SpaceId dest_space_id = PSParallelCompact::space_id(dest_addr);
1334   HeapWord* new_top = PSParallelCompact::new_top(dest_space_id);
1335   assert(dest_addr < new_top, "sanity");
1336 
1337   return MIN2(pointer_delta(new_top, dest_addr), ParallelCompactData::RegionSize);
1338 }
1339 
1340 inline
1341 MoveAndUpdateClosure::MoveAndUpdateClosure(ParMarkBitMap* bitmap,
1342                                            ParCompactionManager* cm,
1343                                            size_t region_idx) :
1344   ParMarkBitMapClosure(bitmap, cm, calculate_words_remaining(region_idx)),
1345   _destination(PSParallelCompact::summary_data().region_to_addr(region_idx)),
1346   _start_array(PSParallelCompact::start_array(PSParallelCompact::space_id(_destination))),
1347   _offset(0) { }
1348 

1349 
1350 inline void MoveAndUpdateClosure::update_state(size_t words)
1351 {
1352   decrement_words_remaining(words);
1353   _source += words;
1354   _destination += words;
1355 }
1356 
1357 class MoveAndUpdateShadowClosure: public MoveAndUpdateClosure {
1358   inline size_t calculate_shadow_offset(size_t region_idx, size_t shadow_idx);
1359 public:
1360   inline MoveAndUpdateShadowClosure(ParMarkBitMap* bitmap, ParCompactionManager* cm,
1361                        size_t region, size_t shadow);
1362 
1363   virtual void complete_region(ParCompactionManager* cm, HeapWord* dest_addr,
1364                                PSParallelCompact::RegionData* region_ptr);
1365 
1366 private:
1367   size_t _shadow;
1368 };
1369 
1370 inline size_t MoveAndUpdateShadowClosure::calculate_shadow_offset(size_t region_idx, size_t shadow_idx) {
1371   ParallelCompactData& sd = PSParallelCompact::summary_data();
1372   HeapWord* dest_addr = sd.region_to_addr(region_idx);
1373   HeapWord* shadow_addr = sd.region_to_addr(shadow_idx);
1374   return pointer_delta(shadow_addr, dest_addr);
1375 }
1376 
1377 inline
1378 MoveAndUpdateShadowClosure::MoveAndUpdateShadowClosure(ParMarkBitMap *bitmap,
1379                                                        ParCompactionManager *cm,
1380                                                        size_t region,
1381                                                        size_t shadow) :
1382   MoveAndUpdateClosure(bitmap, cm, region),
1383   _shadow(shadow) {
1384   _offset = calculate_shadow_offset(region, shadow);
1385 }
1386 
1387 class UpdateOnlyClosure: public ParMarkBitMapClosure {
1388  private:
1389   const PSParallelCompact::SpaceId _space_id;
1390   ObjectStartArray* const          _start_array;
1391 
1392  public:
1393   UpdateOnlyClosure(ParMarkBitMap* mbm,
1394                     ParCompactionManager* cm,
1395                     PSParallelCompact::SpaceId space_id);
1396 
1397   // Update the object.
1398   virtual IterationStatus do_addr(HeapWord* addr, size_t words);
1399 
1400   inline void do_addr(HeapWord* addr);
1401 };
1402 
1403 class FillClosure: public ParMarkBitMapClosure {
1404  public:
1405   FillClosure(ParCompactionManager* cm, PSParallelCompact::SpaceId space_id);
1406 
1407   virtual IterationStatus do_addr(HeapWord* addr, size_t size);
1408 
1409  private:
1410   ObjectStartArray* const _start_array;
1411 };
1412 
1413 #endif // SHARE_GC_PARALLEL_PSPARALLELCOMPACT_HPP
--- EOF ---