1 /*
   2  * Copyright (c) 2005, 2013, 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/markSweep.hpp"
  33 #include "gc_implementation/shared/mutableSpace.hpp"
  34 #include "memory/sharedHeap.hpp"
  35 #include "oops/oop.hpp"
  36 
  37 class ParallelScavengeHeap;
  38 class PSAdaptiveSizePolicy;
  39 class PSYoungGen;
  40 class PSOldGen;
  41 class ParCompactionManager;
  42 class ParallelTaskTerminator;
  43 class PSParallelCompact;
  44 class GCTaskManager;
  45 class GCTaskQueue;
  46 class PreGCValues;
  47 class MoveAndUpdateClosure;
  48 class RefProcTaskExecutor;
  49 class ParallelOldTracer;
  50 class STWGCTimer;
  51 
  52 // The SplitInfo class holds the information needed to 'split' a source region
  53 // so that the live data can be copied to two destination *spaces*.  Normally,
  54 // all the live data in a region is copied to a single destination space (e.g.,
  55 // everything live in a region in eden is copied entirely into the old gen).
  56 // However, when the heap is nearly full, all the live data in eden may not fit
  57 // into the old gen.  Copying only some of the regions from eden to old gen
  58 // requires finding a region that does not contain a partial object (i.e., no
  59 // live object crosses the region boundary) somewhere near the last object that
  60 // does fit into the old gen.  Since it's not always possible to find such a
  61 // region, splitting is necessary for predictable behavior.
  62 //
  63 // A region is always split at the end of the partial object.  This avoids
  64 // additional tests when calculating the new location of a pointer, which is a
  65 // very hot code path.  The partial object and everything to its left will be
  66 // copied to another space (call it dest_space_1).  The live data to the right
  67 // of the partial object will be copied either within the space itself, or to a
  68 // different destination space (distinct from dest_space_1).
  69 //
  70 // Split points are identified during the summary phase, when region
  71 // destinations are computed:  data about the split, including the
  72 // partial_object_size, is recorded in a SplitInfo record and the
  73 // partial_object_size field in the summary data is set to zero.  The zeroing is
  74 // possible (and necessary) since the partial object will move to a different
  75 // destination space than anything to its right, thus the partial object should
  76 // not affect the locations of any objects to its right.
  77 //
  78 // The recorded data is used during the compaction phase, but only rarely:  when
  79 // the partial object on the split region will be copied across a destination
  80 // region boundary.  This test is made once each time a region is filled, and is
  81 // a simple address comparison, so the overhead is negligible (see
  82 // PSParallelCompact::first_src_addr()).
  83 //
  84 // Notes:
  85 //
  86 // Only regions with partial objects are split; a region without a partial
  87 // object does not need any extra bookkeeping.
  88 //
  89 // At most one region is split per space, so the amount of data required is
  90 // constant.
  91 //
  92 // A region is split only when the destination space would overflow.  Once that
  93 // happens, the destination space is abandoned and no other data (even from
  94 // other source spaces) is targeted to that destination space.  Abandoning the
  95 // destination space may leave a somewhat large unused area at the end, if a
  96 // large object caused the overflow.
  97 //
  98 // Future work:
  99 //
 100 // More bookkeeping would be required to continue to use the destination space.
 101 // The most general solution would allow data from regions in two different
 102 // source spaces to be "joined" in a single destination region.  At the very
 103 // least, additional code would be required in next_src_region() to detect the
 104 // join and skip to an out-of-order source region.  If the join region was also
 105 // the last destination region to which a split region was copied (the most
 106 // likely case), then additional work would be needed to get fill_region() to
 107 // stop iteration and switch to a new source region at the right point.  Basic
 108 // idea would be to use a fake value for the top of the source space.  It is
 109 // doable, if a bit tricky.
 110 //
 111 // A simpler (but less general) solution would fill the remainder of the
 112 // destination region with a dummy object and continue filling the next
 113 // destination region.
 114 
 115 class SplitInfo
 116 {
 117 public:
 118   // Return true if this split info is valid (i.e., if a split has been
 119   // recorded).  The very first region cannot have a partial object and thus is
 120   // never split, so 0 is the 'invalid' value.
 121   bool is_valid() const { return _src_region_idx > 0; }
 122 
 123   // Return true if this split holds data for the specified source region.
 124   inline bool is_split(size_t source_region) const;
 125 
 126   // The index of the split region, the size of the partial object on that
 127   // region and the destination of the partial object.
 128   size_t    src_region_idx() const   { return _src_region_idx; }
 129   size_t    partial_obj_size() const { return _partial_obj_size; }
 130   HeapWord* destination() const      { return _destination; }
 131 
 132   // The destination count of the partial object referenced by this split
 133   // (either 1 or 2).  This must be added to the destination count of the
 134   // remainder of the source region.
 135   unsigned int destination_count() const { return _destination_count; }
 136 
 137   // If a word within the partial object will be written to the first word of a
 138   // destination region, this is the address of the destination region;
 139   // otherwise this is NULL.
 140   HeapWord* dest_region_addr() const     { return _dest_region_addr; }
 141 
 142   // If a word within the partial object will be written to the first word of a
 143   // destination region, this is the address of that word within the partial
 144   // object; otherwise this is NULL.
 145   HeapWord* first_src_addr() const       { return _first_src_addr; }
 146 
 147   // Record the data necessary to split the region src_region_idx.
 148   void record(size_t src_region_idx, size_t partial_obj_size,
 149               HeapWord* destination);
 150 
 151   void clear();
 152 
 153   DEBUG_ONLY(void verify_clear();)
 154 
 155 private:
 156   size_t       _src_region_idx;
 157   size_t       _partial_obj_size;
 158   HeapWord*    _destination;
 159   unsigned int _destination_count;
 160   HeapWord*    _dest_region_addr;
 161   HeapWord*    _first_src_addr;
 162 };
 163 
 164 inline bool SplitInfo::is_split(size_t region_idx) const
 165 {
 166   return _src_region_idx == region_idx && is_valid();
 167 }
 168 
 169 class SpaceInfo
 170 {
 171  public:
 172   MutableSpace* space() const { return _space; }
 173 
 174   // Where the free space will start after the collection.  Valid only after the
 175   // summary phase completes.
 176   HeapWord* new_top() const { return _new_top; }
 177 
 178   // Allows new_top to be set.
 179   HeapWord** new_top_addr() { return &_new_top; }
 180 
 181   // Where the smallest allowable dense prefix ends (used only for perm gen).
 182   HeapWord* min_dense_prefix() const { return _min_dense_prefix; }
 183 
 184   // Where the dense prefix ends, or the compacted region begins.
 185   HeapWord* dense_prefix() const { return _dense_prefix; }
 186 
 187   // The start array for the (generation containing the) space, or NULL if there
 188   // is no start array.
 189   ObjectStartArray* start_array() const { return _start_array; }
 190 
 191   SplitInfo& split_info() { return _split_info; }
 192 
 193   void set_space(MutableSpace* s)           { _space = s; }
 194   void set_new_top(HeapWord* addr)          { _new_top = addr; }
 195   void set_min_dense_prefix(HeapWord* addr) { _min_dense_prefix = addr; }
 196   void set_dense_prefix(HeapWord* addr)     { _dense_prefix = addr; }
 197   void set_start_array(ObjectStartArray* s) { _start_array = s; }
 198 
 199   void publish_new_top() const              { _space->set_top(_new_top); }
 200 
 201  private:
 202   MutableSpace*     _space;
 203   HeapWord*         _new_top;
 204   HeapWord*         _min_dense_prefix;
 205   HeapWord*         _dense_prefix;
 206   ObjectStartArray* _start_array;
 207   SplitInfo         _split_info;
 208 };
 209 
 210 class ParallelCompactData
 211 {
 212 public:
 213   // Sizes are in HeapWords, unless indicated otherwise.
 214   static const size_t Log2RegionSize;
 215   static const size_t RegionSize;
 216   static const size_t RegionSizeBytes;
 217 
 218   // Mask for the bits in a size_t to get an offset within a region.
 219   static const size_t RegionSizeOffsetMask;
 220   // Mask for the bits in a pointer to get an offset within a region.
 221   static const size_t RegionAddrOffsetMask;
 222   // Mask for the bits in a pointer to get the address of the start of a region.
 223   static const size_t RegionAddrMask;
 224 
 225   class RegionData
 226   {
 227   public:
 228     // Destination address of the region.
 229     HeapWord* destination() const { return _destination; }
 230 
 231     // The first region containing data destined for this region.
 232     size_t source_region() const { return _source_region; }
 233 
 234     // The object (if any) starting in this region and ending in a different
 235     // region that could not be updated during the main (parallel) compaction
 236     // phase.  This is different from _partial_obj_addr, which is an object that
 237     // extends onto a source region.  However, the two uses do not overlap in
 238     // time, so the same field is used to save space.
 239     HeapWord* deferred_obj_addr() const { return _partial_obj_addr; }
 240 
 241     // The starting address of the partial object extending onto the region.
 242     HeapWord* partial_obj_addr() const { return _partial_obj_addr; }
 243 
 244     // Size of the partial object extending onto the region (words).
 245     size_t partial_obj_size() const { return _partial_obj_size; }
 246 
 247     // Size of live data that lies within this region due to objects that start
 248     // in this region (words).  This does not include the partial object
 249     // extending onto the region (if any), or the part of an object that extends
 250     // onto the next region (if any).
 251     size_t live_obj_size() const { return _dc_and_los & los_mask; }
 252 
 253     // Total live data that lies within the region (words).
 254     size_t data_size() const { return partial_obj_size() + live_obj_size(); }
 255 
 256     // The destination_count is the number of other regions to which data from
 257     // this region will be copied.  At the end of the summary phase, the valid
 258     // values of destination_count are
 259     //
 260     // 0 - data from the region will be compacted completely into itself, or the
 261     //     region is empty.  The region can be claimed and then filled.
 262     // 1 - data from the region will be compacted into 1 other region; some
 263     //     data from the region may also be compacted into the region itself.
 264     // 2 - data from the region will be copied to 2 other regions.
 265     //
 266     // During compaction as regions are emptied, the destination_count is
 267     // decremented (atomically) and when it reaches 0, it can be claimed and
 268     // then filled.
 269     //
 270     // A region is claimed for processing by atomically changing the
 271     // destination_count to the claimed value (dc_claimed).  After a region has
 272     // been filled, the destination_count should be set to the completed value
 273     // (dc_completed).
 274     inline uint destination_count() const;
 275     inline uint destination_count_raw() const;
 276 
 277     // The location of the java heap data that corresponds to this region.
 278     inline HeapWord* data_location() const;
 279 
 280     // The highest address referenced by objects in this region.
 281     inline HeapWord* highest_ref() const;
 282 
 283     // Whether this region is available to be claimed, has been claimed, or has
 284     // been completed.
 285     //
 286     // Minor subtlety:  claimed() returns true if the region is marked
 287     // completed(), which is desirable since a region must be claimed before it
 288     // can be completed.
 289     bool available() const { return _dc_and_los < dc_one; }
 290     bool claimed() const   { return _dc_and_los >= dc_claimed; }
 291     bool completed() const { return _dc_and_los >= dc_completed; }
 292 
 293     // These are not atomic.
 294     void set_destination(HeapWord* addr)       { _destination = addr; }
 295     void set_source_region(size_t region)      { _source_region = region; }
 296     void set_deferred_obj_addr(HeapWord* addr) { _partial_obj_addr = addr; }
 297     void set_partial_obj_addr(HeapWord* addr)  { _partial_obj_addr = addr; }
 298     void set_partial_obj_size(size_t words)    {
 299       _partial_obj_size = (region_sz_t) words;
 300     }
 301 
 302     inline void set_destination_count(uint count);
 303     inline void set_live_obj_size(size_t words);
 304     inline void set_data_location(HeapWord* addr);
 305     inline void set_completed();
 306     inline bool claim_unsafe();
 307 
 308     // These are atomic.
 309     inline void add_live_obj(size_t words);
 310     inline void set_highest_ref(HeapWord* addr);
 311     inline void decrement_destination_count();
 312     inline bool claim();
 313 
 314   private:
 315     // The type used to represent object sizes within a region.
 316     typedef uint region_sz_t;
 317 
 318     // Constants for manipulating the _dc_and_los field, which holds both the
 319     // destination count and live obj size.  The live obj size lives at the
 320     // least significant end so no masking is necessary when adding.
 321     static const region_sz_t dc_shift;           // Shift amount.
 322     static const region_sz_t dc_mask;            // Mask for destination count.
 323     static const region_sz_t dc_one;             // 1, shifted appropriately.
 324     static const region_sz_t dc_claimed;         // Region has been claimed.
 325     static const region_sz_t dc_completed;       // Region has been completed.
 326     static const region_sz_t los_mask;           // Mask for live obj size.
 327 
 328     HeapWord*            _destination;
 329     size_t               _source_region;
 330     HeapWord*            _partial_obj_addr;
 331     region_sz_t          _partial_obj_size;
 332     region_sz_t volatile _dc_and_los;
 333 #ifdef ASSERT
 334     // These enable optimizations that are only partially implemented.  Use
 335     // debug builds to prevent the code fragments from breaking.
 336     HeapWord*            _data_location;
 337     HeapWord*            _highest_ref;
 338 #endif  // #ifdef ASSERT
 339 
 340 #ifdef ASSERT
 341    public:
 342     uint            _pushed;   // 0 until region is pushed onto a worker's stack
 343    private:
 344 #endif
 345   };
 346 
 347 public:
 348   ParallelCompactData();
 349   bool initialize(MemRegion covered_region);
 350 
 351   size_t region_count() const { return _region_count; }
 352   size_t reserved_byte_size() const { return _reserved_byte_size; }
 353 
 354   // Convert region indices to/from RegionData pointers.
 355   inline RegionData* region(size_t region_idx) const;
 356   inline size_t     region(const RegionData* const region_ptr) const;
 357 
 358   // Returns true if the given address is contained within the region
 359   bool region_contains(size_t region_index, HeapWord* addr);
 360 
 361   void add_obj(HeapWord* addr, size_t len);
 362   void add_obj(oop p, size_t len) { add_obj((HeapWord*)p, len); }
 363 
 364   // Fill in the regions covering [beg, end) so that no data moves; i.e., the
 365   // destination of region n is simply the start of region n.  The argument beg
 366   // must be region-aligned; end need not be.
 367   void summarize_dense_prefix(HeapWord* beg, HeapWord* end);
 368 
 369   HeapWord* summarize_split_space(size_t src_region, SplitInfo& split_info,
 370                                   HeapWord* destination, HeapWord* target_end,
 371                                   HeapWord** target_next);
 372   bool summarize(SplitInfo& split_info,
 373                  HeapWord* source_beg, HeapWord* source_end,
 374                  HeapWord** source_next,
 375                  HeapWord* target_beg, HeapWord* target_end,
 376                  HeapWord** target_next);
 377 
 378   void clear();
 379   void clear_range(size_t beg_region, size_t end_region);
 380   void clear_range(HeapWord* beg, HeapWord* end) {
 381     clear_range(addr_to_region_idx(beg), addr_to_region_idx(end));
 382   }
 383 
 384   // Return the number of words between addr and the start of the region
 385   // containing addr.
 386   inline size_t     region_offset(const HeapWord* addr) const;
 387 
 388   // Convert addresses to/from a region index or region pointer.
 389   inline size_t     addr_to_region_idx(const HeapWord* addr) const;
 390   inline RegionData* addr_to_region_ptr(const HeapWord* addr) const;
 391   inline HeapWord*  region_to_addr(size_t region) const;
 392   inline HeapWord*  region_to_addr(size_t region, size_t offset) const;
 393   inline HeapWord*  region_to_addr(const RegionData* region) const;
 394 
 395   inline HeapWord*  region_align_down(HeapWord* addr) const;
 396   inline HeapWord*  region_align_up(HeapWord* addr) const;
 397   inline bool       is_region_aligned(HeapWord* addr) const;
 398 
 399   // Return the address one past the end of the partial object.
 400   HeapWord* partial_obj_end(size_t region_idx) const;
 401 
 402   // Return the new location of the object p after the
 403   // the compaction.
 404   HeapWord* calc_new_pointer(HeapWord* addr);
 405 
 406   HeapWord* calc_new_pointer(oop p) {
 407     return calc_new_pointer((HeapWord*) p);
 408   }
 409 
 410 #ifdef  ASSERT
 411   void verify_clear(const PSVirtualSpace* vspace);
 412   void verify_clear();
 413 #endif  // #ifdef ASSERT
 414 
 415 private:
 416   bool initialize_region_data(size_t region_size);
 417   PSVirtualSpace* create_vspace(size_t count, size_t element_size);
 418 
 419 private:
 420   HeapWord*       _region_start;
 421 #ifdef  ASSERT
 422   HeapWord*       _region_end;
 423 #endif  // #ifdef ASSERT
 424 
 425   PSVirtualSpace* _region_vspace;
 426   size_t          _reserved_byte_size;
 427   RegionData*     _region_data;
 428   size_t          _region_count;
 429 };
 430 
 431 inline uint
 432 ParallelCompactData::RegionData::destination_count_raw() const
 433 {
 434   return _dc_and_los & dc_mask;
 435 }
 436 
 437 inline uint
 438 ParallelCompactData::RegionData::destination_count() const
 439 {
 440   return destination_count_raw() >> dc_shift;
 441 }
 442 
 443 inline void
 444 ParallelCompactData::RegionData::set_destination_count(uint count)
 445 {
 446   assert(count <= (dc_completed >> dc_shift), "count too large");
 447   const region_sz_t live_sz = (region_sz_t) live_obj_size();
 448   _dc_and_los = (count << dc_shift) | live_sz;
 449 }
 450 
 451 inline void ParallelCompactData::RegionData::set_live_obj_size(size_t words)
 452 {
 453   assert(words <= los_mask, "would overflow");
 454   _dc_and_los = destination_count_raw() | (region_sz_t)words;
 455 }
 456 
 457 inline void ParallelCompactData::RegionData::decrement_destination_count()
 458 {
 459   assert(_dc_and_los < dc_claimed, "already claimed");
 460   assert(_dc_and_los >= dc_one, "count would go negative");
 461   Atomic::add((int)dc_mask, (volatile int*)&_dc_and_los);
 462 }
 463 
 464 inline HeapWord* ParallelCompactData::RegionData::data_location() const
 465 {
 466   DEBUG_ONLY(return _data_location;)
 467   NOT_DEBUG(return NULL;)
 468 }
 469 
 470 inline HeapWord* ParallelCompactData::RegionData::highest_ref() const
 471 {
 472   DEBUG_ONLY(return _highest_ref;)
 473   NOT_DEBUG(return NULL;)
 474 }
 475 
 476 inline void ParallelCompactData::RegionData::set_data_location(HeapWord* addr)
 477 {
 478   DEBUG_ONLY(_data_location = addr;)
 479 }
 480 
 481 inline void ParallelCompactData::RegionData::set_completed()
 482 {
 483   assert(claimed(), "must be claimed first");
 484   _dc_and_los = dc_completed | (region_sz_t) live_obj_size();
 485 }
 486 
 487 // MT-unsafe claiming of a region.  Should only be used during single threaded
 488 // execution.
 489 inline bool ParallelCompactData::RegionData::claim_unsafe()
 490 {
 491   if (available()) {
 492     _dc_and_los |= dc_claimed;
 493     return true;
 494   }
 495   return false;
 496 }
 497 
 498 inline void ParallelCompactData::RegionData::add_live_obj(size_t words)
 499 {
 500   assert(words <= (size_t)los_mask - live_obj_size(), "overflow");
 501   Atomic::add((int) words, (volatile int*) &_dc_and_los);
 502 }
 503 
 504 inline void ParallelCompactData::RegionData::set_highest_ref(HeapWord* addr)
 505 {
 506 #ifdef ASSERT
 507   HeapWord* tmp = _highest_ref;
 508   while (addr > tmp) {
 509     tmp = (HeapWord*)Atomic::cmpxchg_ptr(addr, &_highest_ref, tmp);
 510   }
 511 #endif  // #ifdef ASSERT
 512 }
 513 
 514 inline bool ParallelCompactData::RegionData::claim()
 515 {
 516   const int los = (int) live_obj_size();
 517   const int old = Atomic::cmpxchg(dc_claimed | los,
 518                                   (volatile int*) &_dc_and_los, los);
 519   return old == los;
 520 }
 521 
 522 inline ParallelCompactData::RegionData*
 523 ParallelCompactData::region(size_t region_idx) const
 524 {
 525   assert(region_idx <= region_count(), "bad arg");
 526   return _region_data + region_idx;
 527 }
 528 
 529 inline size_t
 530 ParallelCompactData::region(const RegionData* const region_ptr) const
 531 {
 532   assert(region_ptr >= _region_data, "bad arg");
 533   assert(region_ptr <= _region_data + region_count(), "bad arg");
 534   return pointer_delta(region_ptr, _region_data, sizeof(RegionData));
 535 }
 536 
 537 inline size_t
 538 ParallelCompactData::region_offset(const HeapWord* addr) const
 539 {
 540   assert(addr >= _region_start, "bad addr");
 541   assert(addr <= _region_end, "bad addr");
 542   return (size_t(addr) & RegionAddrOffsetMask) >> LogHeapWordSize;
 543 }
 544 
 545 inline size_t
 546 ParallelCompactData::addr_to_region_idx(const HeapWord* addr) const
 547 {
 548   assert(addr >= _region_start, "bad addr");
 549   assert(addr <= _region_end, "bad addr");
 550   return pointer_delta(addr, _region_start) >> Log2RegionSize;
 551 }
 552 
 553 inline ParallelCompactData::RegionData*
 554 ParallelCompactData::addr_to_region_ptr(const HeapWord* addr) const
 555 {
 556   return region(addr_to_region_idx(addr));
 557 }
 558 
 559 inline HeapWord*
 560 ParallelCompactData::region_to_addr(size_t region) const
 561 {
 562   assert(region <= _region_count, "region out of range");
 563   return _region_start + (region << Log2RegionSize);
 564 }
 565 
 566 inline HeapWord*
 567 ParallelCompactData::region_to_addr(const RegionData* region) const
 568 {
 569   return region_to_addr(pointer_delta(region, _region_data,
 570                                       sizeof(RegionData)));
 571 }
 572 
 573 inline HeapWord*
 574 ParallelCompactData::region_to_addr(size_t region, size_t offset) const
 575 {
 576   assert(region <= _region_count, "region out of range");
 577   assert(offset < RegionSize, "offset too big");  // This may be too strict.
 578   return region_to_addr(region) + offset;
 579 }
 580 
 581 inline HeapWord*
 582 ParallelCompactData::region_align_down(HeapWord* addr) const
 583 {
 584   assert(addr >= _region_start, "bad addr");
 585   assert(addr < _region_end + RegionSize, "bad addr");
 586   return (HeapWord*)(size_t(addr) & RegionAddrMask);
 587 }
 588 
 589 inline HeapWord*
 590 ParallelCompactData::region_align_up(HeapWord* addr) const
 591 {
 592   assert(addr >= _region_start, "bad addr");
 593   assert(addr <= _region_end, "bad addr");
 594   return region_align_down(addr + RegionSizeOffsetMask);
 595 }
 596 
 597 inline bool
 598 ParallelCompactData::is_region_aligned(HeapWord* addr) const
 599 {
 600   return region_offset(addr) == 0;
 601 }
 602 
 603 // Abstract closure for use with ParMarkBitMap::iterate(), which will invoke the
 604 // do_addr() method.
 605 //
 606 // The closure is initialized with the number of heap words to process
 607 // (words_remaining()), and becomes 'full' when it reaches 0.  The do_addr()
 608 // methods in subclasses should update the total as words are processed.  Since
 609 // only one subclass actually uses this mechanism to terminate iteration, the
 610 // default initial value is > 0.  The implementation is here and not in the
 611 // single subclass that uses it to avoid making is_full() virtual, and thus
 612 // adding a virtual call per live object.
 613 
 614 class ParMarkBitMapClosure: public StackObj {
 615  public:
 616   typedef ParMarkBitMap::idx_t idx_t;
 617   typedef ParMarkBitMap::IterationStatus IterationStatus;
 618 
 619  public:
 620   inline ParMarkBitMapClosure(ParMarkBitMap* mbm, ParCompactionManager* cm,
 621                               size_t words = max_uintx);
 622 
 623   inline ParCompactionManager* compaction_manager() const;
 624   inline ParMarkBitMap*        bitmap() const;
 625   inline size_t                words_remaining() const;
 626   inline bool                  is_full() const;
 627   inline HeapWord*             source() const;
 628 
 629   inline void                  set_source(HeapWord* addr);
 630 
 631   virtual IterationStatus do_addr(HeapWord* addr, size_t words) = 0;
 632 
 633  protected:
 634   inline void decrement_words_remaining(size_t words);
 635 
 636  private:
 637   ParMarkBitMap* const        _bitmap;
 638   ParCompactionManager* const _compaction_manager;
 639   DEBUG_ONLY(const size_t     _initial_words_remaining;) // Useful in debugger.
 640   size_t                      _words_remaining; // Words left to copy.
 641 
 642  protected:
 643   HeapWord*                   _source;          // Next addr that would be read.
 644 };
 645 
 646 inline
 647 ParMarkBitMapClosure::ParMarkBitMapClosure(ParMarkBitMap* bitmap,
 648                                            ParCompactionManager* cm,
 649                                            size_t words):
 650   _bitmap(bitmap), _compaction_manager(cm)
 651 #ifdef  ASSERT
 652   , _initial_words_remaining(words)
 653 #endif
 654 {
 655   _words_remaining = words;
 656   _source = NULL;
 657 }
 658 
 659 inline ParCompactionManager* ParMarkBitMapClosure::compaction_manager() const {
 660   return _compaction_manager;
 661 }
 662 
 663 inline ParMarkBitMap* ParMarkBitMapClosure::bitmap() const {
 664   return _bitmap;
 665 }
 666 
 667 inline size_t ParMarkBitMapClosure::words_remaining() const {
 668   return _words_remaining;
 669 }
 670 
 671 inline bool ParMarkBitMapClosure::is_full() const {
 672   return words_remaining() == 0;
 673 }
 674 
 675 inline HeapWord* ParMarkBitMapClosure::source() const {
 676   return _source;
 677 }
 678 
 679 inline void ParMarkBitMapClosure::set_source(HeapWord* addr) {
 680   _source = addr;
 681 }
 682 
 683 inline void ParMarkBitMapClosure::decrement_words_remaining(size_t words) {
 684   assert(_words_remaining >= words, "processed too many words");
 685   _words_remaining -= words;
 686 }
 687 
 688 // The UseParallelOldGC collector is a stop-the-world garbage collector that
 689 // does parts of the collection using parallel threads.  The collection includes
 690 // the tenured generation and the young generation.  The permanent generation is
 691 // collected at the same time as the other two generations but the permanent
 692 // generation is collect by a single GC thread.  The permanent generation is
 693 // collected serially because of the requirement that during the processing of a
 694 // klass AAA, any objects reference by AAA must already have been processed.
 695 // This requirement is enforced by a left (lower address) to right (higher
 696 // address) sliding compaction.
 697 //
 698 // There are four phases of the collection.
 699 //
 700 //      - marking phase
 701 //      - summary phase
 702 //      - compacting phase
 703 //      - clean up phase
 704 //
 705 // Roughly speaking these phases correspond, respectively, to
 706 //      - mark all the live objects
 707 //      - calculate the destination of each object at the end of the collection
 708 //      - move the objects to their destination
 709 //      - update some references and reinitialize some variables
 710 //
 711 // These three phases are invoked in PSParallelCompact::invoke_no_policy().  The
 712 // marking phase is implemented in PSParallelCompact::marking_phase() and does a
 713 // complete marking of the heap.  The summary phase is implemented in
 714 // PSParallelCompact::summary_phase().  The move and update phase is implemented
 715 // in PSParallelCompact::compact().
 716 //
 717 // A space that is being collected is divided into regions and with each region
 718 // is associated an object of type ParallelCompactData.  Each region is of a
 719 // fixed size and typically will contain more than 1 object and may have parts
 720 // of objects at the front and back of the region.
 721 //
 722 // region            -----+---------------------+----------
 723 // objects covered   [ AAA  )[ BBB )[ CCC   )[ DDD     )
 724 //
 725 // The marking phase does a complete marking of all live objects in the heap.
 726 // The marking also compiles the size of the data for all live objects covered
 727 // by the region.  This size includes the part of any live object spanning onto
 728 // the region (part of AAA if it is live) from the front, all live objects
 729 // contained in the region (BBB and/or CCC if they are live), and the part of
 730 // any live objects covered by the region that extends off the region (part of
 731 // DDD if it is live).  The marking phase uses multiple GC threads and marking
 732 // is done in a bit array of type ParMarkBitMap.  The marking of the bit map is
 733 // done atomically as is the accumulation of the size of the live objects
 734 // covered by a region.
 735 //
 736 // The summary phase calculates the total live data to the left of each region
 737 // XXX.  Based on that total and the bottom of the space, it can calculate the
 738 // starting location of the live data in XXX.  The summary phase calculates for
 739 // each region XXX quantites such as
 740 //
 741 //      - the amount of live data at the beginning of a region from an object
 742 //        entering the region.
 743 //      - the location of the first live data on the region
 744 //      - a count of the number of regions receiving live data from XXX.
 745 //
 746 // See ParallelCompactData for precise details.  The summary phase also
 747 // calculates the dense prefix for the compaction.  The dense prefix is a
 748 // portion at the beginning of the space that is not moved.  The objects in the
 749 // dense prefix do need to have their object references updated.  See method
 750 // summarize_dense_prefix().
 751 //
 752 // The summary phase is done using 1 GC thread.
 753 //
 754 // The compaction phase moves objects to their new location and updates all
 755 // references in the object.
 756 //
 757 // A current exception is that objects that cross a region boundary are moved
 758 // but do not have their references updated.  References are not updated because
 759 // it cannot easily be determined if the klass pointer KKK for the object AAA
 760 // has been updated.  KKK likely resides in a region to the left of the region
 761 // containing AAA.  These AAA's have there references updated at the end in a
 762 // clean up phase.  See the method PSParallelCompact::update_deferred_objects().
 763 // An alternate strategy is being investigated for this deferral of updating.
 764 //
 765 // Compaction is done on a region basis.  A region that is ready to be filled is
 766 // put on a ready list and GC threads take region off the list and fill them.  A
 767 // region is ready to be filled if it empty of live objects.  Such a region may
 768 // have been initially empty (only contained dead objects) or may have had all
 769 // its live objects copied out already.  A region that compacts into itself is
 770 // also ready for filling.  The ready list is initially filled with empty
 771 // regions and regions compacting into themselves.  There is always at least 1
 772 // region that can be put on the ready list.  The regions are atomically added
 773 // and removed from the ready list.
 774 
 775 class PSParallelCompact : AllStatic {
 776  public:
 777   // Convenient access to type names.
 778   typedef ParMarkBitMap::idx_t idx_t;
 779   typedef ParallelCompactData::RegionData RegionData;
 780 
 781   typedef enum {
 782     old_space_id, eden_space_id,
 783     from_space_id, to_space_id, last_space_id
 784   } SpaceId;
 785 
 786  public:
 787   // Inline closure decls
 788   //
 789   class IsAliveClosure: public BoolObjectClosure {
 790    public:
 791     virtual bool do_object_b(oop p);
 792   };
 793 
 794   class KeepAliveClosure: public OopClosure {
 795    private:
 796     ParCompactionManager* _compaction_manager;
 797    protected:
 798     template <class T> inline void do_oop_work(T* p);
 799    public:
 800     KeepAliveClosure(ParCompactionManager* cm) : _compaction_manager(cm) { }
 801     virtual void do_oop(oop* p);
 802     virtual void do_oop(narrowOop* p);
 803   };
 804 
 805   class FollowStackClosure: public VoidClosure {
 806    private:
 807     ParCompactionManager* _compaction_manager;
 808    public:
 809     FollowStackClosure(ParCompactionManager* cm) : _compaction_manager(cm) { }
 810     virtual void do_void();
 811   };
 812 
 813   class AdjustPointerClosure: public OopClosure {
 814    public:
 815     virtual void do_oop(oop* p);
 816     virtual void do_oop(narrowOop* p);
 817     // do not walk from thread stacks to the code cache on this phase
 818     virtual void do_code_blob(CodeBlob* cb) const { }
 819   };
 820 
 821   class AdjustKlassClosure : public KlassClosure {
 822    public:
 823     void do_klass(Klass* klass);
 824   };
 825 
 826   friend class KeepAliveClosure;
 827   friend class FollowStackClosure;
 828   friend class AdjustPointerClosure;
 829   friend class AdjustKlassClosure;
 830   friend class FollowKlassClosure;
 831   friend class InstanceClassLoaderKlass;
 832   friend class RefProcTaskProxy;
 833 
 834  private:
 835   static STWGCTimer           _gc_timer;
 836   static ParallelOldTracer    _gc_tracer;
 837   static elapsedTimer         _accumulated_time;
 838   static unsigned int         _total_invocations;
 839   static unsigned int         _maximum_compaction_gc_num;
 840   static jlong                _time_of_last_gc;   // ms
 841   static CollectorCounters*   _counters;
 842   static ParMarkBitMap        _mark_bitmap;
 843   static ParallelCompactData  _summary_data;
 844   static IsAliveClosure       _is_alive_closure;
 845   static SpaceInfo            _space_info[last_space_id];
 846   static bool                 _print_phases;
 847   static AdjustPointerClosure _adjust_pointer_closure;
 848   static AdjustKlassClosure   _adjust_klass_closure;
 849 
 850   // Reference processing (used in ...follow_contents)
 851   static ReferenceProcessor*  _ref_processor;
 852 
 853   // Updated location of intArrayKlassObj.
 854   static Klass* _updated_int_array_klass_obj;
 855 
 856   // Values computed at initialization and used by dead_wood_limiter().
 857   static double _dwl_mean;
 858   static double _dwl_std_dev;
 859   static double _dwl_first_term;
 860   static double _dwl_adjustment;
 861 #ifdef  ASSERT
 862   static bool   _dwl_initialized;
 863 #endif  // #ifdef ASSERT
 864 
 865  private:
 866 
 867   static void initialize_space_info();
 868 
 869   // Return true if details about individual phases should be printed.
 870   static inline bool print_phases();
 871 
 872   // Clear the marking bitmap and summary data that cover the specified space.
 873   static void clear_data_covering_space(SpaceId id);
 874 
 875   static void pre_compact(PreGCValues* pre_gc_values);
 876   static void post_compact();
 877 
 878   // Mark live objects
 879   static void marking_phase(ParCompactionManager* cm,
 880                             bool maximum_heap_compaction,
 881                             ParallelOldTracer *gc_tracer);
 882 
 883   template <class T>
 884   static inline void follow_root(ParCompactionManager* cm, T* p);
 885 
 886   // Compute the dense prefix for the designated space.  This is an experimental
 887   // implementation currently not used in production.
 888   static HeapWord* compute_dense_prefix_via_density(const SpaceId id,
 889                                                     bool maximum_compaction);
 890 
 891   // Methods used to compute the dense prefix.
 892 
 893   // Compute the value of the normal distribution at x = density.  The mean and
 894   // standard deviation are values saved by initialize_dead_wood_limiter().
 895   static inline double normal_distribution(double density);
 896 
 897   // Initialize the static vars used by dead_wood_limiter().
 898   static void initialize_dead_wood_limiter();
 899 
 900   // Return the percentage of space that can be treated as "dead wood" (i.e.,
 901   // not reclaimed).
 902   static double dead_wood_limiter(double density, size_t min_percent);
 903 
 904   // Find the first (left-most) region in the range [beg, end) that has at least
 905   // dead_words of dead space to the left.  The argument beg must be the first
 906   // region in the space that is not completely live.
 907   static RegionData* dead_wood_limit_region(const RegionData* beg,
 908                                             const RegionData* end,
 909                                             size_t dead_words);
 910 
 911   // Return a pointer to the first region in the range [beg, end) that is not
 912   // completely full.
 913   static RegionData* first_dead_space_region(const RegionData* beg,
 914                                              const RegionData* end);
 915 
 916   // Return a value indicating the benefit or 'yield' if the compacted region
 917   // were to start (or equivalently if the dense prefix were to end) at the
 918   // candidate region.  Higher values are better.
 919   //
 920   // The value is based on the amount of space reclaimed vs. the costs of (a)
 921   // updating references in the dense prefix plus (b) copying objects and
 922   // updating references in the compacted region.
 923   static inline double reclaimed_ratio(const RegionData* const candidate,
 924                                        HeapWord* const bottom,
 925                                        HeapWord* const top,
 926                                        HeapWord* const new_top);
 927 
 928   // Compute the dense prefix for the designated space.
 929   static HeapWord* compute_dense_prefix(const SpaceId id,
 930                                         bool maximum_compaction);
 931 
 932   // Return true if dead space crosses onto the specified Region; bit must be
 933   // the bit index corresponding to the first word of the Region.
 934   static inline bool dead_space_crosses_boundary(const RegionData* region,
 935                                                  idx_t bit);
 936 
 937   // Summary phase utility routine to fill dead space (if any) at the dense
 938   // prefix boundary.  Should only be called if the the dense prefix is
 939   // non-empty.
 940   static void fill_dense_prefix_end(SpaceId id);
 941 
 942   // Clear the summary data source_region field for the specified addresses.
 943   static void clear_source_region(HeapWord* beg_addr, HeapWord* end_addr);
 944 
 945 #ifndef PRODUCT
 946   // Routines to provoke splitting a young gen space (ParallelOldGCSplitALot).
 947 
 948   // Fill the region [start, start + words) with live object(s).  Only usable
 949   // for the old and permanent generations.
 950   static void fill_with_live_objects(SpaceId id, HeapWord* const start,
 951                                      size_t words);
 952   // Include the new objects in the summary data.
 953   static void summarize_new_objects(SpaceId id, HeapWord* start);
 954 
 955   // Add live objects to a survivor space since it's rare that both survivors
 956   // are non-empty.
 957   static void provoke_split_fill_survivor(SpaceId id);
 958 
 959   // Add live objects and/or choose the dense prefix to provoke splitting.
 960   static void provoke_split(bool & maximum_compaction);
 961 #endif
 962 
 963   static void summarize_spaces_quick();
 964   static void summarize_space(SpaceId id, bool maximum_compaction);
 965   static void summary_phase(ParCompactionManager* cm, bool maximum_compaction);
 966 
 967   // Adjust addresses in roots.  Does not adjust addresses in heap.
 968   static void adjust_roots();
 969 
 970   // Move objects to new locations.
 971   static void compact_perm(ParCompactionManager* cm);
 972   static void compact();
 973 
 974   // Add available regions to the stack and draining tasks to the task queue.
 975   static void enqueue_region_draining_tasks(GCTaskQueue* q,
 976                                             uint parallel_gc_threads);
 977 
 978   // Add dense prefix update tasks to the task queue.
 979   static void enqueue_dense_prefix_tasks(GCTaskQueue* q,
 980                                          uint parallel_gc_threads);
 981 
 982   // Add region stealing tasks to the task queue.
 983   static void enqueue_region_stealing_tasks(
 984                                        GCTaskQueue* q,
 985                                        ParallelTaskTerminator* terminator_ptr,
 986                                        uint parallel_gc_threads);
 987 
 988   // If objects are left in eden after a collection, try to move the boundary
 989   // and absorb them into the old gen.  Returns true if eden was emptied.
 990   static bool absorb_live_data_from_eden(PSAdaptiveSizePolicy* size_policy,
 991                                          PSYoungGen* young_gen,
 992                                          PSOldGen* old_gen);
 993 
 994   // Reset time since last full gc
 995   static void reset_millis_since_last_gc();
 996 
 997  public:
 998   class MarkAndPushClosure: public OopClosure {
 999    private:
1000     ParCompactionManager* _compaction_manager;
1001    public:
1002     MarkAndPushClosure(ParCompactionManager* cm) : _compaction_manager(cm) { }
1003     virtual void do_oop(oop* p);
1004     virtual void do_oop(narrowOop* p);
1005   };
1006 
1007   // The one and only place to start following the classes.
1008   // Should only be applied to the ClassLoaderData klasses list.
1009   class FollowKlassClosure : public KlassClosure {
1010    private:
1011     MarkAndPushClosure* _mark_and_push_closure;
1012    public:
1013     FollowKlassClosure(MarkAndPushClosure* mark_and_push_closure) :
1014         _mark_and_push_closure(mark_and_push_closure) { }
1015     void do_klass(Klass* klass);
1016   };
1017 
1018   PSParallelCompact();
1019 
1020   // Convenient accessor for Universe::heap().
1021   static ParallelScavengeHeap* gc_heap() {
1022     return (ParallelScavengeHeap*)Universe::heap();
1023   }
1024 
1025   static void invoke(bool maximum_heap_compaction);
1026   static bool invoke_no_policy(bool maximum_heap_compaction);
1027 
1028   static void post_initialize();
1029   // Perform initialization for PSParallelCompact that requires
1030   // allocations.  This should be called during the VM initialization
1031   // at a pointer where it would be appropriate to return a JNI_ENOMEM
1032   // in the event of a failure.
1033   static bool initialize();
1034 
1035   // Closure accessors
1036   static OopClosure* adjust_pointer_closure()      { return (OopClosure*)&_adjust_pointer_closure; }
1037   static KlassClosure* adjust_klass_closure()      { return (KlassClosure*)&_adjust_klass_closure; }
1038   static BoolObjectClosure* is_alive_closure()     { return (BoolObjectClosure*)&_is_alive_closure; }
1039 
1040   // Public accessors
1041   static elapsedTimer* accumulated_time() { return &_accumulated_time; }
1042   static unsigned int total_invocations() { return _total_invocations; }
1043   static CollectorCounters* counters()    { return _counters; }
1044 
1045   // Used to add tasks
1046   static GCTaskManager* const gc_task_manager();
1047   static Klass* updated_int_array_klass_obj() {
1048     return _updated_int_array_klass_obj;
1049   }
1050 
1051   // Marking support
1052   static inline bool mark_obj(oop obj);
1053   static inline bool is_marked(oop obj);
1054   // Check mark and maybe push on marking stack
1055   template <class T> static inline void mark_and_push(ParCompactionManager* cm,
1056                                                       T* p);
1057   template <class T> static inline void adjust_pointer(T* p);
1058 
1059   static void follow_klass(ParCompactionManager* cm, Klass* klass);
1060   static void adjust_klass(ParCompactionManager* cm, Klass* klass);
1061 
1062   static void follow_class_loader(ParCompactionManager* cm,
1063                                   ClassLoaderData* klass);
1064   static void adjust_class_loader(ParCompactionManager* cm,
1065                                   ClassLoaderData* klass);
1066 
1067   // Compaction support.
1068   // Return true if p is in the range [beg_addr, end_addr).
1069   static inline bool is_in(HeapWord* p, HeapWord* beg_addr, HeapWord* end_addr);
1070   static inline bool is_in(oop* p, HeapWord* beg_addr, HeapWord* end_addr);
1071 
1072   // Convenience wrappers for per-space data kept in _space_info.
1073   static inline MutableSpace*     space(SpaceId space_id);
1074   static inline HeapWord*         new_top(SpaceId space_id);
1075   static inline HeapWord*         dense_prefix(SpaceId space_id);
1076   static inline ObjectStartArray* start_array(SpaceId space_id);
1077 
1078   // Move and update the live objects in the specified space.
1079   static void move_and_update(ParCompactionManager* cm, SpaceId space_id);
1080 
1081   // Process the end of the given region range in the dense prefix.
1082   // This includes saving any object not updated.
1083   static void dense_prefix_regions_epilogue(ParCompactionManager* cm,
1084                                             size_t region_start_index,
1085                                             size_t region_end_index,
1086                                             idx_t exiting_object_offset,
1087                                             idx_t region_offset_start,
1088                                             idx_t region_offset_end);
1089 
1090   // Update a region in the dense prefix.  For each live object
1091   // in the region, update it's interior references.  For each
1092   // dead object, fill it with deadwood. Dead space at the end
1093   // of a region range will be filled to the start of the next
1094   // live object regardless of the region_index_end.  None of the
1095   // objects in the dense prefix move and dead space is dead
1096   // (holds only dead objects that don't need any processing), so
1097   // dead space can be filled in any order.
1098   static void update_and_deadwood_in_dense_prefix(ParCompactionManager* cm,
1099                                                   SpaceId space_id,
1100                                                   size_t region_index_start,
1101                                                   size_t region_index_end);
1102 
1103   // Return the address of the count + 1st live word in the range [beg, end).
1104   static HeapWord* skip_live_words(HeapWord* beg, HeapWord* end, size_t count);
1105 
1106   // Return the address of the word to be copied to dest_addr, which must be
1107   // aligned to a region boundary.
1108   static HeapWord* first_src_addr(HeapWord* const dest_addr,
1109                                   SpaceId src_space_id,
1110                                   size_t src_region_idx);
1111 
1112   // Determine the next source region, set closure.source() to the start of the
1113   // new region return the region index.  Parameter end_addr is the address one
1114   // beyond the end of source range just processed.  If necessary, switch to a
1115   // new source space and set src_space_id (in-out parameter) and src_space_top
1116   // (out parameter) accordingly.
1117   static size_t next_src_region(MoveAndUpdateClosure& closure,
1118                                 SpaceId& src_space_id,
1119                                 HeapWord*& src_space_top,
1120                                 HeapWord* end_addr);
1121 
1122   // Decrement the destination count for each non-empty source region in the
1123   // range [beg_region, region(region_align_up(end_addr))).  If the destination
1124   // count for a region goes to 0 and it needs to be filled, enqueue it.
1125   static void decrement_destination_counts(ParCompactionManager* cm,
1126                                            SpaceId src_space_id,
1127                                            size_t beg_region,
1128                                            HeapWord* end_addr);
1129 
1130   // Fill a region, copying objects from one or more source regions.
1131   static void fill_region(ParCompactionManager* cm, size_t region_idx);
1132   static void fill_and_update_region(ParCompactionManager* cm, size_t region) {
1133     fill_region(cm, region);
1134   }
1135 
1136   // Update the deferred objects in the space.
1137   static void update_deferred_objects(ParCompactionManager* cm, SpaceId id);
1138 
1139   static ParMarkBitMap* mark_bitmap() { return &_mark_bitmap; }
1140   static ParallelCompactData& summary_data() { return _summary_data; }
1141 
1142   // Reference Processing
1143   static ReferenceProcessor* const ref_processor() { return _ref_processor; }
1144 
1145   static STWGCTimer* gc_timer() { return &_gc_timer; }
1146 
1147   // Return the SpaceId for the given address.
1148   static SpaceId space_id(HeapWord* addr);
1149 
1150   // Time since last full gc (in milliseconds).
1151   static jlong millis_since_last_gc();
1152 
1153   static void print_on_error(outputStream* st);
1154 
1155 #ifndef PRODUCT
1156   // Debugging support.
1157   static const char* space_names[last_space_id];
1158   static void print_region_ranges();
1159   static void print_dense_prefix_stats(const char* const algorithm,
1160                                        const SpaceId id,
1161                                        const bool maximum_compaction,
1162                                        HeapWord* const addr);
1163   static void summary_phase_msg(SpaceId dst_space_id,
1164                                 HeapWord* dst_beg, HeapWord* dst_end,
1165                                 SpaceId src_space_id,
1166                                 HeapWord* src_beg, HeapWord* src_end);
1167 #endif  // #ifndef PRODUCT
1168 
1169 #ifdef  ASSERT
1170   // Sanity check the new location of a word in the heap.
1171   static inline void check_new_location(HeapWord* old_addr, HeapWord* new_addr);
1172   // Verify that all the regions have been emptied.
1173   static void verify_complete(SpaceId space_id);
1174 #endif  // #ifdef ASSERT
1175 };
1176 
1177 inline bool PSParallelCompact::mark_obj(oop obj) {
1178   const int obj_size = obj->size();
1179   if (mark_bitmap()->mark_obj(obj, obj_size)) {
1180     _summary_data.add_obj(obj, obj_size);
1181     return true;
1182   } else {
1183     return false;
1184   }
1185 }
1186 
1187 inline bool PSParallelCompact::is_marked(oop obj) {
1188   return mark_bitmap()->is_marked(obj);
1189 }
1190 
1191 template <class T>
1192 inline void PSParallelCompact::follow_root(ParCompactionManager* cm, T* p) {
1193   assert(!Universe::heap()->is_in_reserved(p),
1194          "roots shouldn't be things within the heap");
1195 
1196   T heap_oop = oopDesc::load_heap_oop(p);
1197   if (!oopDesc::is_null(heap_oop)) {
1198     oop obj = oopDesc::decode_heap_oop_not_null(heap_oop);
1199     if (mark_bitmap()->is_unmarked(obj)) {
1200       if (mark_obj(obj)) {
1201         obj->follow_contents(cm);
1202       }
1203     }
1204   }
1205   cm->follow_marking_stacks();
1206 }
1207 
1208 template <class T>
1209 inline void PSParallelCompact::mark_and_push(ParCompactionManager* cm, T* p) {
1210   T heap_oop = oopDesc::load_heap_oop(p);
1211   if (!oopDesc::is_null(heap_oop)) {
1212     oop obj = oopDesc::decode_heap_oop_not_null(heap_oop);
1213     if (mark_bitmap()->is_unmarked(obj) && mark_obj(obj)) {
1214       cm->push(obj);
1215     }
1216   }
1217 }
1218 
1219 template <class T>
1220 inline void PSParallelCompact::adjust_pointer(T* p) {
1221   T heap_oop = oopDesc::load_heap_oop(p);
1222   if (!oopDesc::is_null(heap_oop)) {
1223     oop obj     = oopDesc::decode_heap_oop_not_null(heap_oop);
1224     oop new_obj = (oop)summary_data().calc_new_pointer(obj);
1225     assert(new_obj != NULL,                    // is forwarding ptr?
1226            "should be forwarded");
1227     // Just always do the update unconditionally?
1228     if (new_obj != NULL) {
1229       assert(Universe::heap()->is_in_reserved(new_obj),
1230              "should be in object space");
1231       oopDesc::encode_store_heap_oop_not_null(p, new_obj);
1232     }
1233   }
1234 }
1235 
1236 template <class T>
1237 inline void PSParallelCompact::KeepAliveClosure::do_oop_work(T* p) {
1238   mark_and_push(_compaction_manager, p);
1239 }
1240 
1241 inline bool PSParallelCompact::print_phases() {
1242   return _print_phases;
1243 }
1244 
1245 inline double PSParallelCompact::normal_distribution(double density) {
1246   assert(_dwl_initialized, "uninitialized");
1247   const double squared_term = (density - _dwl_mean) / _dwl_std_dev;
1248   return _dwl_first_term * exp(-0.5 * squared_term * squared_term);
1249 }
1250 
1251 inline bool
1252 PSParallelCompact::dead_space_crosses_boundary(const RegionData* region,
1253                                                idx_t bit)
1254 {
1255   assert(bit > 0, "cannot call this for the first bit/region");
1256   assert(_summary_data.region_to_addr(region) == _mark_bitmap.bit_to_addr(bit),
1257          "sanity check");
1258 
1259   // Dead space crosses the boundary if (1) a partial object does not extend
1260   // onto the region, (2) an object does not start at the beginning of the
1261   // region, and (3) an object does not end at the end of the prior region.
1262   return region->partial_obj_size() == 0 &&
1263     !_mark_bitmap.is_obj_beg(bit) &&
1264     !_mark_bitmap.is_obj_end(bit - 1);
1265 }
1266 
1267 inline bool
1268 PSParallelCompact::is_in(HeapWord* p, HeapWord* beg_addr, HeapWord* end_addr) {
1269   return p >= beg_addr && p < end_addr;
1270 }
1271 
1272 inline bool
1273 PSParallelCompact::is_in(oop* p, HeapWord* beg_addr, HeapWord* end_addr) {
1274   return is_in((HeapWord*)p, beg_addr, end_addr);
1275 }
1276 
1277 inline MutableSpace* PSParallelCompact::space(SpaceId id) {
1278   assert(id < last_space_id, "id out of range");
1279   return _space_info[id].space();
1280 }
1281 
1282 inline HeapWord* PSParallelCompact::new_top(SpaceId id) {
1283   assert(id < last_space_id, "id out of range");
1284   return _space_info[id].new_top();
1285 }
1286 
1287 inline HeapWord* PSParallelCompact::dense_prefix(SpaceId id) {
1288   assert(id < last_space_id, "id out of range");
1289   return _space_info[id].dense_prefix();
1290 }
1291 
1292 inline ObjectStartArray* PSParallelCompact::start_array(SpaceId id) {
1293   assert(id < last_space_id, "id out of range");
1294   return _space_info[id].start_array();
1295 }
1296 
1297 #ifdef ASSERT
1298 inline void
1299 PSParallelCompact::check_new_location(HeapWord* old_addr, HeapWord* new_addr)
1300 {
1301   assert(old_addr >= new_addr || space_id(old_addr) != space_id(new_addr),
1302          "must move left or to a different space");
1303   assert(is_object_aligned((intptr_t)old_addr) && is_object_aligned((intptr_t)new_addr),
1304          "checking alignment");
1305 }
1306 #endif // ASSERT
1307 
1308 class MoveAndUpdateClosure: public ParMarkBitMapClosure {
1309  public:
1310   inline MoveAndUpdateClosure(ParMarkBitMap* bitmap, ParCompactionManager* cm,
1311                               ObjectStartArray* start_array,
1312                               HeapWord* destination, size_t words);
1313 
1314   // Accessors.
1315   HeapWord* destination() const         { return _destination; }
1316 
1317   // If the object will fit (size <= words_remaining()), copy it to the current
1318   // destination, update the interior oops and the start array and return either
1319   // full (if the closure is full) or incomplete.  If the object will not fit,
1320   // return would_overflow.
1321   virtual IterationStatus do_addr(HeapWord* addr, size_t size);
1322 
1323   // Copy enough words to fill this closure, starting at source().  Interior
1324   // oops and the start array are not updated.  Return full.
1325   IterationStatus copy_until_full();
1326 
1327   // Copy enough words to fill this closure or to the end of an object,
1328   // whichever is smaller, starting at source().  Interior oops and the start
1329   // array are not updated.
1330   void copy_partial_obj();
1331 
1332  protected:
1333   // Update variables to indicate that word_count words were processed.
1334   inline void update_state(size_t word_count);
1335 
1336  protected:
1337   ObjectStartArray* const _start_array;
1338   HeapWord*               _destination;         // Next addr to be written.
1339 };
1340 
1341 inline
1342 MoveAndUpdateClosure::MoveAndUpdateClosure(ParMarkBitMap* bitmap,
1343                                            ParCompactionManager* cm,
1344                                            ObjectStartArray* start_array,
1345                                            HeapWord* destination,
1346                                            size_t words) :
1347   ParMarkBitMapClosure(bitmap, cm, words), _start_array(start_array)
1348 {
1349   _destination = destination;
1350 }
1351 
1352 inline void MoveAndUpdateClosure::update_state(size_t words)
1353 {
1354   decrement_words_remaining(words);
1355   _source += words;
1356   _destination += words;
1357 }
1358 
1359 class UpdateOnlyClosure: public ParMarkBitMapClosure {
1360  private:
1361   const PSParallelCompact::SpaceId _space_id;
1362   ObjectStartArray* const          _start_array;
1363 
1364  public:
1365   UpdateOnlyClosure(ParMarkBitMap* mbm,
1366                     ParCompactionManager* cm,
1367                     PSParallelCompact::SpaceId space_id);
1368 
1369   // Update the object.
1370   virtual IterationStatus do_addr(HeapWord* addr, size_t words);
1371 
1372   inline void do_addr(HeapWord* addr);
1373 };
1374 
1375 inline void UpdateOnlyClosure::do_addr(HeapWord* addr)
1376 {
1377   _start_array->allocate_block(addr);
1378   oop(addr)->update_contents(compaction_manager());
1379 }
1380 
1381 class FillClosure: public ParMarkBitMapClosure
1382 {
1383 public:
1384   FillClosure(ParCompactionManager* cm, PSParallelCompact::SpaceId space_id) :
1385     ParMarkBitMapClosure(PSParallelCompact::mark_bitmap(), cm),
1386     _start_array(PSParallelCompact::start_array(space_id))
1387   {
1388     assert(space_id == PSParallelCompact::old_space_id,
1389            "cannot use FillClosure in the young gen");
1390   }
1391 
1392   virtual IterationStatus do_addr(HeapWord* addr, size_t size) {
1393     CollectedHeap::fill_with_objects(addr, size);
1394     HeapWord* const end = addr + size;
1395     do {
1396       _start_array->allocate_block(addr);
1397       addr += oop(addr)->size();
1398     } while (addr < end);
1399     return ParMarkBitMap::incomplete;
1400   }
1401 
1402 private:
1403   ObjectStartArray* const _start_array;
1404 };
1405 
1406 #endif // SHARE_VM_GC_IMPLEMENTATION_PARALLELSCAVENGE_PSPARALLELCOMPACT_HPP