1 /*
   2  * Copyright (c) 2005, 2012, 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 
  50 // The SplitInfo class holds the information needed to 'split' a source region
  51 // so that the live data can be copied to two destination *spaces*.  Normally,
  52 // all the live data in a region is copied to a single destination space (e.g.,
  53 // everything live in a region in eden is copied entirely into the old gen).
  54 // However, when the heap is nearly full, all the live data in eden may not fit
  55 // into the old gen.  Copying only some of the regions from eden to old gen
  56 // requires finding a region that does not contain a partial object (i.e., no
  57 // live object crosses the region boundary) somewhere near the last object that
  58 // does fit into the old gen.  Since it's not always possible to find such a
  59 // region, splitting is necessary for predictable behavior.
  60 //
  61 // A region is always split at the end of the partial object.  This avoids
  62 // additional tests when calculating the new location of a pointer, which is a
  63 // very hot code path.  The partial object and everything to its left will be
  64 // copied to another space (call it dest_space_1).  The live data to the right
  65 // of the partial object will be copied either within the space itself, or to a
  66 // different destination space (distinct from dest_space_1).
  67 //
  68 // Split points are identified during the summary phase, when region
  69 // destinations are computed:  data about the split, including the
  70 // partial_object_size, is recorded in a SplitInfo record and the
  71 // partial_object_size field in the summary data is set to zero.  The zeroing is
  72 // possible (and necessary) since the partial object will move to a different
  73 // destination space than anything to its right, thus the partial object should
  74 // not affect the locations of any objects to its right.
  75 //
  76 // The recorded data is used during the compaction phase, but only rarely:  when
  77 // the partial object on the split region will be copied across a destination
  78 // region boundary.  This test is made once each time a region is filled, and is
  79 // a simple address comparison, so the overhead is negligible (see
  80 // PSParallelCompact::first_src_addr()).
  81 //
  82 // Notes:
  83 //
  84 // Only regions with partial objects are split; a region without a partial
  85 // object does not need any extra bookkeeping.
  86 //
  87 // At most one region is split per space, so the amount of data required is
  88 // constant.
  89 //
  90 // A region is split only when the destination space would overflow.  Once that
  91 // happens, the destination space is abandoned and no other data (even from
  92 // other source spaces) is targeted to that destination space.  Abandoning the
  93 // destination space may leave a somewhat large unused area at the end, if a
  94 // large object caused the overflow.
  95 //
  96 // Future work:
  97 //
  98 // More bookkeeping would be required to continue to use the destination space.
  99 // The most general solution would allow data from regions in two different
 100 // source spaces to be "joined" in a single destination region.  At the very
 101 // least, additional code would be required in next_src_region() to detect the
 102 // join and skip to an out-of-order source region.  If the join region was also
 103 // the last destination region to which a split region was copied (the most
 104 // likely case), then additional work would be needed to get fill_region() to
 105 // stop iteration and switch to a new source region at the right point.  Basic
 106 // idea would be to use a fake value for the top of the source space.  It is
 107 // doable, if a bit tricky.
 108 //
 109 // A simpler (but less general) solution would fill the remainder of the
 110 // destination region with a dummy object and continue filling the next
 111 // destination region.
 112 
 113 class SplitInfo
 114 {
 115 public:
 116   // Return true if this split info is valid (i.e., if a split has been
 117   // recorded).  The very first region cannot have a partial object and thus is
 118   // never split, so 0 is the 'invalid' value.
 119   bool is_valid() const { return _src_region_idx > 0; }
 120 
 121   // Return true if this split holds data for the specified source region.
 122   inline bool is_split(size_t source_region) const;
 123 
 124   // The index of the split region, the size of the partial object on that
 125   // region and the destination of the partial object.
 126   size_t    src_region_idx() const   { return _src_region_idx; }
 127   size_t    partial_obj_size() const { return _partial_obj_size; }
 128   HeapWord* destination() const      { return _destination; }
 129 
 130   // The destination count of the partial object referenced by this split
 131   // (either 1 or 2).  This must be added to the destination count of the
 132   // remainder of the source region.
 133   unsigned int destination_count() const { return _destination_count; }
 134 
 135   // If a word within the partial object will be written to the first word of a
 136   // destination region, this is the address of the destination region;
 137   // otherwise this is NULL.
 138   HeapWord* dest_region_addr() const     { return _dest_region_addr; }
 139 
 140   // If a word within the partial object will be written to the first word of a
 141   // destination region, this is the address of that word within the partial
 142   // object; otherwise this is NULL.
 143   HeapWord* first_src_addr() const       { return _first_src_addr; }
 144 
 145   // Record the data necessary to split the region src_region_idx.
 146   void record(size_t src_region_idx, size_t partial_obj_size,
 147               HeapWord* destination);
 148 
 149   void clear();
 150 
 151   DEBUG_ONLY(void verify_clear();)
 152 
 153 private:
 154   size_t       _src_region_idx;
 155   size_t       _partial_obj_size;
 156   HeapWord*    _destination;
 157   unsigned int _destination_count;
 158   HeapWord*    _dest_region_addr;
 159   HeapWord*    _first_src_addr;
 160 };
 161 
 162 inline bool SplitInfo::is_split(size_t region_idx) const
 163 {
 164   return _src_region_idx == region_idx && is_valid();
 165 }
 166 
 167 class SpaceInfo
 168 {
 169  public:
 170   MutableSpace* space() const { return _space; }
 171 
 172   // Where the free space will start after the collection.  Valid only after the
 173   // summary phase completes.
 174   HeapWord* new_top() const { return _new_top; }
 175 
 176   // Allows new_top to be set.
 177   HeapWord** new_top_addr() { return &_new_top; }
 178 
 179   // Where the smallest allowable dense prefix ends (used only for perm gen).
 180   HeapWord* min_dense_prefix() const { return _min_dense_prefix; }
 181 
 182   // Where the dense prefix ends, or the compacted region begins.
 183   HeapWord* dense_prefix() const { return _dense_prefix; }
 184 
 185   // The start array for the (generation containing the) space, or NULL if there
 186   // is no start array.
 187   ObjectStartArray* start_array() const { return _start_array; }
 188 
 189   SplitInfo& split_info() { return _split_info; }
 190 
 191   void set_space(MutableSpace* s)           { _space = s; }
 192   void set_new_top(HeapWord* addr)          { _new_top = addr; }
 193   void set_min_dense_prefix(HeapWord* addr) { _min_dense_prefix = addr; }
 194   void set_dense_prefix(HeapWord* addr)     { _dense_prefix = addr; }
 195   void set_start_array(ObjectStartArray* s) { _start_array = s; }
 196 
 197   void publish_new_top() const              { _space->set_top(_new_top); }
 198 
 199  private:
 200   MutableSpace*     _space;
 201   HeapWord*         _new_top;
 202   HeapWord*         _min_dense_prefix;
 203   HeapWord*         _dense_prefix;
 204   ObjectStartArray* _start_array;
 205   SplitInfo         _split_info;
 206 };
 207 
 208 class ParallelCompactData
 209 {
 210 public:
 211   // Sizes are in HeapWords, unless indicated otherwise.
 212   static const size_t Log2RegionSize;
 213   static const size_t RegionSize;
 214   static const size_t RegionSizeBytes;
 215 
 216   // Mask for the bits in a size_t to get an offset within a region.
 217   static const size_t RegionSizeOffsetMask;
 218   // Mask for the bits in a pointer to get an offset within a region.
 219   static const size_t RegionAddrOffsetMask;
 220   // Mask for the bits in a pointer to get the address of the start of a region.
 221   static const size_t RegionAddrMask;
 222 
 223   class RegionData
 224   {
 225   public:
 226     // Destination address of the region.
 227     HeapWord* destination() const { return _destination; }
 228 
 229     // The first region containing data destined for this region.
 230     size_t source_region() const { return _source_region; }
 231 
 232     // The object (if any) starting in this region and ending in a different
 233     // region that could not be updated during the main (parallel) compaction
 234     // phase.  This is different from _partial_obj_addr, which is an object that
 235     // extends onto a source region.  However, the two uses do not overlap in
 236     // time, so the same field is used to save space.
 237     HeapWord* deferred_obj_addr() const { return _partial_obj_addr; }
 238 
 239     // The starting address of the partial object extending onto the region.
 240     HeapWord* partial_obj_addr() const { return _partial_obj_addr; }
 241 
 242     // Size of the partial object extending onto the region (words).
 243     size_t partial_obj_size() const { return _partial_obj_size; }
 244 
 245     // Size of live data that lies within this region due to objects that start
 246     // in this region (words).  This does not include the partial object
 247     // extending onto the region (if any), or the part of an object that extends
 248     // onto the next region (if any).
 249     size_t live_obj_size() const { return _dc_and_los & los_mask; }
 250 
 251     // Total live data that lies within the region (words).
 252     size_t data_size() const { return partial_obj_size() + live_obj_size(); }
 253 
 254     // The destination_count is the number of other regions to which data from
 255     // this region will be copied.  At the end of the summary phase, the valid
 256     // values of destination_count are
 257     //
 258     // 0 - data from the region will be compacted completely into itself, or the
 259     //     region is empty.  The region can be claimed and then filled.
 260     // 1 - data from the region will be compacted into 1 other region; some
 261     //     data from the region may also be compacted into the region itself.
 262     // 2 - data from the region will be copied to 2 other regions.
 263     //
 264     // During compaction as regions are emptied, the destination_count is
 265     // decremented (atomically) and when it reaches 0, it can be claimed and
 266     // then filled.
 267     //
 268     // A region is claimed for processing by atomically changing the
 269     // destination_count to the claimed value (dc_claimed).  After a region has
 270     // been filled, the destination_count should be set to the completed value
 271     // (dc_completed).
 272     inline uint destination_count() const;
 273     inline uint destination_count_raw() const;
 274 
 275     // The location of the java heap data that corresponds to this region.
 276     inline HeapWord* data_location() const;
 277 
 278     // The highest address referenced by objects in this region.
 279     inline HeapWord* highest_ref() const;
 280 
 281     // Whether this region is available to be claimed, has been claimed, or has
 282     // been completed.
 283     //
 284     // Minor subtlety:  claimed() returns true if the region is marked
 285     // completed(), which is desirable since a region must be claimed before it
 286     // can be completed.
 287     bool available() const { return _dc_and_los < dc_one; }
 288     bool claimed() const   { return _dc_and_los >= dc_claimed; }
 289     bool completed() const { return _dc_and_los >= dc_completed; }
 290 
 291     // These are not atomic.
 292     void set_destination(HeapWord* addr)       { _destination = addr; }
 293     void set_source_region(size_t region)      { _source_region = region; }
 294     void set_deferred_obj_addr(HeapWord* addr) { _partial_obj_addr = addr; }
 295     void set_partial_obj_addr(HeapWord* addr)  { _partial_obj_addr = addr; }
 296     void set_partial_obj_size(size_t words)    {
 297       _partial_obj_size = (region_sz_t) words;
 298     }
 299 
 300     inline void set_destination_count(uint count);
 301     inline void set_live_obj_size(size_t words);
 302     inline void set_data_location(HeapWord* addr);
 303     inline void set_completed();
 304     inline bool claim_unsafe();
 305 
 306     // These are atomic.
 307     inline void add_live_obj(size_t words);
 308     inline void set_highest_ref(HeapWord* addr);
 309     inline void decrement_destination_count();
 310     inline bool claim();
 311 
 312   private:
 313     // The type used to represent object sizes within a region.
 314     typedef uint region_sz_t;
 315 
 316     // Constants for manipulating the _dc_and_los field, which holds both the
 317     // destination count and live obj size.  The live obj size lives at the
 318     // least significant end so no masking is necessary when adding.
 319     static const region_sz_t dc_shift;           // Shift amount.
 320     static const region_sz_t dc_mask;            // Mask for destination count.
 321     static const region_sz_t dc_one;             // 1, shifted appropriately.
 322     static const region_sz_t dc_claimed;         // Region has been claimed.
 323     static const region_sz_t dc_completed;       // Region has been completed.
 324     static const region_sz_t los_mask;           // Mask for live obj size.
 325 
 326     HeapWord*            _destination;
 327     size_t               _source_region;
 328     HeapWord*            _partial_obj_addr;
 329     region_sz_t          _partial_obj_size;
 330     region_sz_t volatile _dc_and_los;
 331 #ifdef ASSERT
 332     // These enable optimizations that are only partially implemented.  Use
 333     // debug builds to prevent the code fragments from breaking.
 334     HeapWord*            _data_location;
 335     HeapWord*            _highest_ref;
 336 #endif  // #ifdef ASSERT
 337 
 338 #ifdef ASSERT
 339    public:
 340     uint            _pushed;   // 0 until region is pushed onto a worker's stack
 341    private:
 342 #endif
 343   };
 344 
 345 public:
 346   ParallelCompactData();
 347   bool initialize(MemRegion covered_region);
 348 
 349   size_t region_count() const { return _region_count; }
 350   size_t reserved_byte_size() const { return _reserved_byte_size; }
 351 
 352   // Convert region indices to/from RegionData pointers.
 353   inline RegionData* region(size_t region_idx) const;
 354   inline size_t     region(const RegionData* const region_ptr) const;
 355 
 356   // Returns true if the given address is contained within the region
 357   bool region_contains(size_t region_index, HeapWord* addr);
 358 
 359   void add_obj(HeapWord* addr, size_t len);
 360   void add_obj(oop p, size_t len) { add_obj((HeapWord*)p, len); }
 361 
 362   // Fill in the regions covering [beg, end) so that no data moves; i.e., the
 363   // destination of region n is simply the start of region n.  The argument beg
 364   // must be region-aligned; end need not be.
 365   void summarize_dense_prefix(HeapWord* beg, HeapWord* end);
 366 
 367   HeapWord* summarize_split_space(size_t src_region, SplitInfo& split_info,
 368                                   HeapWord* destination, HeapWord* target_end,
 369                                   HeapWord** target_next);
 370   bool summarize(SplitInfo& split_info,
 371                  HeapWord* source_beg, HeapWord* source_end,
 372                  HeapWord** source_next,
 373                  HeapWord* target_beg, HeapWord* target_end,
 374                  HeapWord** target_next);
 375 
 376   void clear();
 377   void clear_range(size_t beg_region, size_t end_region);
 378   void clear_range(HeapWord* beg, HeapWord* end) {
 379     clear_range(addr_to_region_idx(beg), addr_to_region_idx(end));
 380   }
 381 
 382   // Return the number of words between addr and the start of the region
 383   // containing addr.
 384   inline size_t     region_offset(const HeapWord* addr) const;
 385 
 386   // Convert addresses to/from a region index or region pointer.
 387   inline size_t     addr_to_region_idx(const HeapWord* addr) const;
 388   inline RegionData* addr_to_region_ptr(const HeapWord* addr) const;
 389   inline HeapWord*  region_to_addr(size_t region) const;
 390   inline HeapWord*  region_to_addr(size_t region, size_t offset) const;
 391   inline HeapWord*  region_to_addr(const RegionData* region) const;
 392 
 393   inline HeapWord*  region_align_down(HeapWord* addr) const;
 394   inline HeapWord*  region_align_up(HeapWord* addr) const;
 395   inline bool       is_region_aligned(HeapWord* addr) const;
 396 
 397   // Return the address one past the end of the partial object.
 398   HeapWord* partial_obj_end(size_t region_idx) const;
 399 
 400   // Return the new location of the object p after the
 401   // the compaction.
 402   HeapWord* calc_new_pointer(HeapWord* addr);
 403 
 404   HeapWord* calc_new_pointer(oop p) {
 405     return calc_new_pointer((HeapWord*) p);
 406   }
 407 
 408 #ifdef  ASSERT
 409   void verify_clear(const PSVirtualSpace* vspace);
 410   void verify_clear();
 411 #endif  // #ifdef ASSERT
 412 
 413 private:
 414   bool initialize_region_data(size_t region_size);
 415   PSVirtualSpace* create_vspace(size_t count, size_t element_size);
 416 
 417 private:
 418   HeapWord*       _region_start;
 419 #ifdef  ASSERT
 420   HeapWord*       _region_end;
 421 #endif  // #ifdef ASSERT
 422 
 423   PSVirtualSpace* _region_vspace;
 424   size_t          _reserved_byte_size;
 425   RegionData*     _region_data;
 426   size_t          _region_count;
 427 };
 428 
 429 inline uint
 430 ParallelCompactData::RegionData::destination_count_raw() const
 431 {
 432   return _dc_and_los & dc_mask;
 433 }
 434 
 435 inline uint
 436 ParallelCompactData::RegionData::destination_count() const
 437 {
 438   return destination_count_raw() >> dc_shift;
 439 }
 440 
 441 inline void
 442 ParallelCompactData::RegionData::set_destination_count(uint count)
 443 {
 444   assert(count <= (dc_completed >> dc_shift), "count too large");
 445   const region_sz_t live_sz = (region_sz_t) live_obj_size();
 446   _dc_and_los = (count << dc_shift) | live_sz;
 447 }
 448 
 449 inline void ParallelCompactData::RegionData::set_live_obj_size(size_t words)
 450 {
 451   assert(words <= los_mask, "would overflow");
 452   _dc_and_los = destination_count_raw() | (region_sz_t)words;
 453 }
 454 
 455 inline void ParallelCompactData::RegionData::decrement_destination_count()
 456 {
 457   assert(_dc_and_los < dc_claimed, "already claimed");
 458   assert(_dc_and_los >= dc_one, "count would go negative");
 459   Atomic::add((int)dc_mask, (volatile int*)&_dc_and_los);
 460 }
 461 
 462 inline HeapWord* ParallelCompactData::RegionData::data_location() const
 463 {
 464   DEBUG_ONLY(return _data_location;)
 465   NOT_DEBUG(return NULL;)
 466 }
 467 
 468 inline HeapWord* ParallelCompactData::RegionData::highest_ref() const
 469 {
 470   DEBUG_ONLY(return _highest_ref;)
 471   NOT_DEBUG(return NULL;)
 472 }
 473 
 474 inline void ParallelCompactData::RegionData::set_data_location(HeapWord* addr)
 475 {
 476   DEBUG_ONLY(_data_location = addr;)
 477 }
 478 
 479 inline void ParallelCompactData::RegionData::set_completed()
 480 {
 481   assert(claimed(), "must be claimed first");
 482   _dc_and_los = dc_completed | (region_sz_t) live_obj_size();
 483 }
 484 
 485 // MT-unsafe claiming of a region.  Should only be used during single threaded
 486 // execution.
 487 inline bool ParallelCompactData::RegionData::claim_unsafe()
 488 {
 489   if (available()) {
 490     _dc_and_los |= dc_claimed;
 491     return true;
 492   }
 493   return false;
 494 }
 495 
 496 inline void ParallelCompactData::RegionData::add_live_obj(size_t words)
 497 {
 498   assert(words <= (size_t)los_mask - live_obj_size(), "overflow");
 499   Atomic::add((int) words, (volatile int*) &_dc_and_los);
 500 }
 501 
 502 inline void ParallelCompactData::RegionData::set_highest_ref(HeapWord* addr)
 503 {
 504 #ifdef ASSERT
 505   HeapWord* tmp = _highest_ref;
 506   while (addr > tmp) {
 507     tmp = (HeapWord*)Atomic::cmpxchg_ptr(addr, &_highest_ref, tmp);
 508   }
 509 #endif  // #ifdef ASSERT
 510 }
 511 
 512 inline bool ParallelCompactData::RegionData::claim()
 513 {
 514   const int los = (int) live_obj_size();
 515   const int old = Atomic::cmpxchg(dc_claimed | los,
 516                                   (volatile int*) &_dc_and_los, los);
 517   return old == los;
 518 }
 519 
 520 inline ParallelCompactData::RegionData*
 521 ParallelCompactData::region(size_t region_idx) const
 522 {
 523   assert(region_idx <= region_count(), "bad arg");
 524   return _region_data + region_idx;
 525 }
 526 
 527 inline size_t
 528 ParallelCompactData::region(const RegionData* const region_ptr) const
 529 {
 530   assert(region_ptr >= _region_data, "bad arg");
 531   assert(region_ptr <= _region_data + region_count(), "bad arg");
 532   return pointer_delta(region_ptr, _region_data, sizeof(RegionData));
 533 }
 534 
 535 inline size_t
 536 ParallelCompactData::region_offset(const HeapWord* addr) const
 537 {
 538   assert(addr >= _region_start, "bad addr");
 539   assert(addr <= _region_end, "bad addr");
 540   return (size_t(addr) & RegionAddrOffsetMask) >> LogHeapWordSize;
 541 }
 542 
 543 inline size_t
 544 ParallelCompactData::addr_to_region_idx(const HeapWord* addr) const
 545 {
 546   assert(addr >= _region_start, "bad addr");
 547   assert(addr <= _region_end, "bad addr");
 548   return pointer_delta(addr, _region_start) >> Log2RegionSize;
 549 }
 550 
 551 inline ParallelCompactData::RegionData*
 552 ParallelCompactData::addr_to_region_ptr(const HeapWord* addr) const
 553 {
 554   return region(addr_to_region_idx(addr));
 555 }
 556 
 557 inline HeapWord*
 558 ParallelCompactData::region_to_addr(size_t region) const
 559 {
 560   assert(region <= _region_count, "region out of range");
 561   return _region_start + (region << Log2RegionSize);
 562 }
 563 
 564 inline HeapWord*
 565 ParallelCompactData::region_to_addr(const RegionData* region) const
 566 {
 567   return region_to_addr(pointer_delta(region, _region_data,
 568                                       sizeof(RegionData)));
 569 }
 570 
 571 inline HeapWord*
 572 ParallelCompactData::region_to_addr(size_t region, size_t offset) const
 573 {
 574   assert(region <= _region_count, "region out of range");
 575   assert(offset < RegionSize, "offset too big");  // This may be too strict.
 576   return region_to_addr(region) + offset;
 577 }
 578 
 579 inline HeapWord*
 580 ParallelCompactData::region_align_down(HeapWord* addr) const
 581 {
 582   assert(addr >= _region_start, "bad addr");
 583   assert(addr < _region_end + RegionSize, "bad addr");
 584   return (HeapWord*)(size_t(addr) & RegionAddrMask);
 585 }
 586 
 587 inline HeapWord*
 588 ParallelCompactData::region_align_up(HeapWord* addr) const
 589 {
 590   assert(addr >= _region_start, "bad addr");
 591   assert(addr <= _region_end, "bad addr");
 592   return region_align_down(addr + RegionSizeOffsetMask);
 593 }
 594 
 595 inline bool
 596 ParallelCompactData::is_region_aligned(HeapWord* addr) const
 597 {
 598   return region_offset(addr) == 0;
 599 }
 600 
 601 // Abstract closure for use with ParMarkBitMap::iterate(), which will invoke the
 602 // do_addr() method.
 603 //
 604 // The closure is initialized with the number of heap words to process
 605 // (words_remaining()), and becomes 'full' when it reaches 0.  The do_addr()
 606 // methods in subclasses should update the total as words are processed.  Since
 607 // only one subclass actually uses this mechanism to terminate iteration, the
 608 // default initial value is > 0.  The implementation is here and not in the
 609 // single subclass that uses it to avoid making is_full() virtual, and thus
 610 // adding a virtual call per live object.
 611 
 612 class ParMarkBitMapClosure: public StackObj {
 613  public:
 614   typedef ParMarkBitMap::idx_t idx_t;
 615   typedef ParMarkBitMap::IterationStatus IterationStatus;
 616 
 617  public:
 618   inline ParMarkBitMapClosure(ParMarkBitMap* mbm, ParCompactionManager* cm,
 619                               size_t words = max_uintx);
 620 
 621   inline ParCompactionManager* compaction_manager() const;
 622   inline ParMarkBitMap*        bitmap() const;
 623   inline size_t                words_remaining() const;
 624   inline bool                  is_full() const;
 625   inline HeapWord*             source() const;
 626 
 627   inline void                  set_source(HeapWord* addr);
 628 
 629   virtual IterationStatus do_addr(HeapWord* addr, size_t words) = 0;
 630 
 631  protected:
 632   inline void decrement_words_remaining(size_t words);
 633 
 634  private:
 635   ParMarkBitMap* const        _bitmap;
 636   ParCompactionManager* const _compaction_manager;
 637   DEBUG_ONLY(const size_t     _initial_words_remaining;) // Useful in debugger.
 638   size_t                      _words_remaining; // Words left to copy.
 639 
 640  protected:
 641   HeapWord*                   _source;          // Next addr that would be read.
 642 };
 643 
 644 inline
 645 ParMarkBitMapClosure::ParMarkBitMapClosure(ParMarkBitMap* bitmap,
 646                                            ParCompactionManager* cm,
 647                                            size_t words):
 648   _bitmap(bitmap), _compaction_manager(cm)
 649 #ifdef  ASSERT
 650   , _initial_words_remaining(words)
 651 #endif
 652 {
 653   _words_remaining = words;
 654   _source = NULL;
 655 }
 656 
 657 inline ParCompactionManager* ParMarkBitMapClosure::compaction_manager() const {
 658   return _compaction_manager;
 659 }
 660 
 661 inline ParMarkBitMap* ParMarkBitMapClosure::bitmap() const {
 662   return _bitmap;
 663 }
 664 
 665 inline size_t ParMarkBitMapClosure::words_remaining() const {
 666   return _words_remaining;
 667 }
 668 
 669 inline bool ParMarkBitMapClosure::is_full() const {
 670   return words_remaining() == 0;
 671 }
 672 
 673 inline HeapWord* ParMarkBitMapClosure::source() const {
 674   return _source;
 675 }
 676 
 677 inline void ParMarkBitMapClosure::set_source(HeapWord* addr) {
 678   _source = addr;
 679 }
 680 
 681 inline void ParMarkBitMapClosure::decrement_words_remaining(size_t words) {
 682   assert(_words_remaining >= words, "processed too many words");
 683   _words_remaining -= words;
 684 }
 685 
 686 // The UseParallelOldGC collector is a stop-the-world garbage collector that
 687 // does parts of the collection using parallel threads.  The collection includes
 688 // the tenured generation and the young generation.  The permanent generation is
 689 // collected at the same time as the other two generations but the permanent
 690 // generation is collect by a single GC thread.  The permanent generation is
 691 // collected serially because of the requirement that during the processing of a
 692 // klass AAA, any objects reference by AAA must already have been processed.
 693 // This requirement is enforced by a left (lower address) to right (higher
 694 // address) sliding compaction.
 695 //
 696 // There are four phases of the collection.
 697 //
 698 //      - marking phase
 699 //      - summary phase
 700 //      - compacting phase
 701 //      - clean up phase
 702 //
 703 // Roughly speaking these phases correspond, respectively, to
 704 //      - mark all the live objects
 705 //      - calculate the destination of each object at the end of the collection
 706 //      - move the objects to their destination
 707 //      - update some references and reinitialize some variables
 708 //
 709 // These three phases are invoked in PSParallelCompact::invoke_no_policy().  The
 710 // marking phase is implemented in PSParallelCompact::marking_phase() and does a
 711 // complete marking of the heap.  The summary phase is implemented in
 712 // PSParallelCompact::summary_phase().  The move and update phase is implemented
 713 // in PSParallelCompact::compact().
 714 //
 715 // A space that is being collected is divided into regions and with each region
 716 // is associated an object of type ParallelCompactData.  Each region is of a
 717 // fixed size and typically will contain more than 1 object and may have parts
 718 // of objects at the front and back of the region.
 719 //
 720 // region            -----+---------------------+----------
 721 // objects covered   [ AAA  )[ BBB )[ CCC   )[ DDD     )
 722 //
 723 // The marking phase does a complete marking of all live objects in the heap.
 724 // The marking also compiles the size of the data for all live objects covered
 725 // by the region.  This size includes the part of any live object spanning onto
 726 // the region (part of AAA if it is live) from the front, all live objects
 727 // contained in the region (BBB and/or CCC if they are live), and the part of
 728 // any live objects covered by the region that extends off the region (part of
 729 // DDD if it is live).  The marking phase uses multiple GC threads and marking
 730 // is done in a bit array of type ParMarkBitMap.  The marking of the bit map is
 731 // done atomically as is the accumulation of the size of the live objects
 732 // covered by a region.
 733 //
 734 // The summary phase calculates the total live data to the left of each region
 735 // XXX.  Based on that total and the bottom of the space, it can calculate the
 736 // starting location of the live data in XXX.  The summary phase calculates for
 737 // each region XXX quantites such as
 738 //
 739 //      - the amount of live data at the beginning of a region from an object
 740 //        entering the region.
 741 //      - the location of the first live data on the region
 742 //      - a count of the number of regions receiving live data from XXX.
 743 //
 744 // See ParallelCompactData for precise details.  The summary phase also
 745 // calculates the dense prefix for the compaction.  The dense prefix is a
 746 // portion at the beginning of the space that is not moved.  The objects in the
 747 // dense prefix do need to have their object references updated.  See method
 748 // summarize_dense_prefix().
 749 //
 750 // The summary phase is done using 1 GC thread.
 751 //
 752 // The compaction phase moves objects to their new location and updates all
 753 // references in the object.
 754 //
 755 // A current exception is that objects that cross a region boundary are moved
 756 // but do not have their references updated.  References are not updated because
 757 // it cannot easily be determined if the klass pointer KKK for the object AAA
 758 // has been updated.  KKK likely resides in a region to the left of the region
 759 // containing AAA.  These AAA's have there references updated at the end in a
 760 // clean up phase.  See the method PSParallelCompact::update_deferred_objects().
 761 // An alternate strategy is being investigated for this deferral of updating.
 762 //
 763 // Compaction is done on a region basis.  A region that is ready to be filled is
 764 // put on a ready list and GC threads take region off the list and fill them.  A
 765 // region is ready to be filled if it empty of live objects.  Such a region may
 766 // have been initially empty (only contained dead objects) or may have had all
 767 // its live objects copied out already.  A region that compacts into itself is
 768 // also ready for filling.  The ready list is initially filled with empty
 769 // regions and regions compacting into themselves.  There is always at least 1
 770 // region that can be put on the ready list.  The regions are atomically added
 771 // and removed from the ready list.
 772 
 773 class PSParallelCompact : AllStatic {
 774  public:
 775   // Convenient access to type names.
 776   typedef ParMarkBitMap::idx_t idx_t;
 777   typedef ParallelCompactData::RegionData RegionData;
 778 
 779   typedef enum {
 780     old_space_id, eden_space_id,
 781     from_space_id, to_space_id, last_space_id
 782   } SpaceId;
 783 
 784  public:
 785   // Inline closure decls
 786   //
 787   class IsAliveClosure: public BoolObjectClosure {
 788    public:
 789     virtual void do_object(oop p);
 790     virtual bool do_object_b(oop p);
 791   };
 792 
 793   class KeepAliveClosure: public OopClosure {
 794    private:
 795     ParCompactionManager* _compaction_manager;
 796    protected:
 797     template <class T> inline void do_oop_work(T* p);
 798    public:
 799     KeepAliveClosure(ParCompactionManager* cm) : _compaction_manager(cm) { }
 800     virtual void do_oop(oop* p);
 801     virtual void do_oop(narrowOop* p);
 802   };
 803 
 804   class FollowStackClosure: public VoidClosure {
 805    private:
 806     ParCompactionManager* _compaction_manager;
 807    public:
 808     FollowStackClosure(ParCompactionManager* cm) : _compaction_manager(cm) { }
 809     virtual void do_void();
 810   };
 811 
 812   class AdjustPointerClosure: public OopClosure {
 813    public:
 814     virtual void do_oop(oop* p);
 815     virtual void do_oop(narrowOop* p);
 816     // do not walk from thread stacks to the code cache on this phase
 817     virtual void do_code_blob(CodeBlob* cb) const { }
 818   };
 819 
 820   class AdjustKlassClosure : public KlassClosure {
 821    public:
 822     void do_klass(Klass* klass);
 823   };
 824 
 825   friend class KeepAliveClosure;
 826   friend class FollowStackClosure;
 827   friend class AdjustPointerClosure;
 828   friend class AdjustKlassClosure;
 829   friend class FollowKlassClosure;
 830   friend class InstanceClassLoaderKlass;
 831   friend class RefProcTaskProxy;
 832 
 833  private:
 834   static elapsedTimer         _accumulated_time;
 835   static unsigned int         _total_invocations;
 836   static unsigned int         _maximum_compaction_gc_num;
 837   static jlong                _time_of_last_gc;   // ms
 838   static CollectorCounters*   _counters;
 839   static ParMarkBitMap        _mark_bitmap;
 840   static ParallelCompactData  _summary_data;
 841   static IsAliveClosure       _is_alive_closure;
 842   static SpaceInfo            _space_info[last_space_id];
 843   static bool                 _print_phases;
 844   static AdjustPointerClosure _adjust_pointer_closure;
 845   static AdjustKlassClosure   _adjust_klass_closure;
 846 
 847   // Reference processing (used in ...follow_contents)
 848   static ReferenceProcessor*  _ref_processor;
 849 
 850   // Updated location of intArrayKlassObj.
 851   static Klass* _updated_int_array_klass_obj;
 852 
 853   // Values computed at initialization and used by dead_wood_limiter().
 854   static double _dwl_mean;
 855   static double _dwl_std_dev;
 856   static double _dwl_first_term;
 857   static double _dwl_adjustment;
 858 #ifdef  ASSERT
 859   static bool   _dwl_initialized;
 860 #endif  // #ifdef ASSERT
 861 
 862  private:
 863 
 864   static void initialize_space_info();
 865 
 866   // Return true if details about individual phases should be printed.
 867   static inline bool print_phases();
 868 
 869   // Clear the marking bitmap and summary data that cover the specified space.
 870   static void clear_data_covering_space(SpaceId id);
 871 
 872   static void pre_compact(PreGCValues* pre_gc_values);
 873   static void post_compact();
 874 
 875   // Mark live objects
 876   static void marking_phase(ParCompactionManager* cm,
 877                             bool maximum_heap_compaction);
 878 
 879   template <class T>
 880   static inline void follow_root(ParCompactionManager* cm, T* p);
 881 
 882   // Compute the dense prefix for the designated space.  This is an experimental
 883   // implementation currently not used in production.
 884   static HeapWord* compute_dense_prefix_via_density(const SpaceId id,
 885                                                     bool maximum_compaction);
 886 
 887   // Methods used to compute the dense prefix.
 888 
 889   // Compute the value of the normal distribution at x = density.  The mean and
 890   // standard deviation are values saved by initialize_dead_wood_limiter().
 891   static inline double normal_distribution(double density);
 892 
 893   // Initialize the static vars used by dead_wood_limiter().
 894   static void initialize_dead_wood_limiter();
 895 
 896   // Return the percentage of space that can be treated as "dead wood" (i.e.,
 897   // not reclaimed).
 898   static double dead_wood_limiter(double density, size_t min_percent);
 899 
 900   // Find the first (left-most) region in the range [beg, end) that has at least
 901   // dead_words of dead space to the left.  The argument beg must be the first
 902   // region in the space that is not completely live.
 903   static RegionData* dead_wood_limit_region(const RegionData* beg,
 904                                             const RegionData* end,
 905                                             size_t dead_words);
 906 
 907   // Return a pointer to the first region in the range [beg, end) that is not
 908   // completely full.
 909   static RegionData* first_dead_space_region(const RegionData* beg,
 910                                              const RegionData* end);
 911 
 912   // Return a value indicating the benefit or 'yield' if the compacted region
 913   // were to start (or equivalently if the dense prefix were to end) at the
 914   // candidate region.  Higher values are better.
 915   //
 916   // The value is based on the amount of space reclaimed vs. the costs of (a)
 917   // updating references in the dense prefix plus (b) copying objects and
 918   // updating references in the compacted region.
 919   static inline double reclaimed_ratio(const RegionData* const candidate,
 920                                        HeapWord* const bottom,
 921                                        HeapWord* const top,
 922                                        HeapWord* const new_top);
 923 
 924   // Compute the dense prefix for the designated space.
 925   static HeapWord* compute_dense_prefix(const SpaceId id,
 926                                         bool maximum_compaction);
 927 
 928   // Return true if dead space crosses onto the specified Region; bit must be
 929   // the bit index corresponding to the first word of the Region.
 930   static inline bool dead_space_crosses_boundary(const RegionData* region,
 931                                                  idx_t bit);
 932 
 933   // Summary phase utility routine to fill dead space (if any) at the dense
 934   // prefix boundary.  Should only be called if the the dense prefix is
 935   // non-empty.
 936   static void fill_dense_prefix_end(SpaceId id);
 937 
 938   // Clear the summary data source_region field for the specified addresses.
 939   static void clear_source_region(HeapWord* beg_addr, HeapWord* end_addr);
 940 
 941 #ifndef PRODUCT
 942   // Routines to provoke splitting a young gen space (ParallelOldGCSplitALot).
 943 
 944   // Fill the region [start, start + words) with live object(s).  Only usable
 945   // for the old and permanent generations.
 946   static void fill_with_live_objects(SpaceId id, HeapWord* const start,
 947                                      size_t words);
 948   // Include the new objects in the summary data.
 949   static void summarize_new_objects(SpaceId id, HeapWord* start);
 950 
 951   // Add live objects to a survivor space since it's rare that both survivors
 952   // are non-empty.
 953   static void provoke_split_fill_survivor(SpaceId id);
 954 
 955   // Add live objects and/or choose the dense prefix to provoke splitting.
 956   static void provoke_split(bool & maximum_compaction);
 957 #endif
 958 
 959   static void summarize_spaces_quick();
 960   static void summarize_space(SpaceId id, bool maximum_compaction);
 961   static void summary_phase(ParCompactionManager* cm, bool maximum_compaction);
 962 
 963   // Adjust addresses in roots.  Does not adjust addresses in heap.
 964   static void adjust_roots();
 965 
 966   // Move objects to new locations.
 967   static void compact_perm(ParCompactionManager* cm);
 968   static void compact();
 969 
 970   // Add available regions to the stack and draining tasks to the task queue.
 971   static void enqueue_region_draining_tasks(GCTaskQueue* q,
 972                                             uint parallel_gc_threads);
 973 
 974   // Add dense prefix update tasks to the task queue.
 975   static void enqueue_dense_prefix_tasks(GCTaskQueue* q,
 976                                          uint parallel_gc_threads);
 977 
 978   // Add region stealing tasks to the task queue.
 979   static void enqueue_region_stealing_tasks(
 980                                        GCTaskQueue* q,
 981                                        ParallelTaskTerminator* terminator_ptr,
 982                                        uint parallel_gc_threads);
 983 
 984   // If objects are left in eden after a collection, try to move the boundary
 985   // and absorb them into the old gen.  Returns true if eden was emptied.
 986   static bool absorb_live_data_from_eden(PSAdaptiveSizePolicy* size_policy,
 987                                          PSYoungGen* young_gen,
 988                                          PSOldGen* old_gen);
 989 
 990   // Reset time since last full gc
 991   static void reset_millis_since_last_gc();
 992 
 993  public:
 994   class MarkAndPushClosure: public OopClosure {
 995    private:
 996     ParCompactionManager* _compaction_manager;
 997    public:
 998     MarkAndPushClosure(ParCompactionManager* cm) : _compaction_manager(cm) { }
 999     virtual void do_oop(oop* p);
1000     virtual void do_oop(narrowOop* p);
1001   };
1002 
1003   // The one and only place to start following the classes.
1004   // Should only be applied to the ClassLoaderData klasses list.
1005   class FollowKlassClosure : public KlassClosure {
1006    private:
1007     MarkAndPushClosure* _mark_and_push_closure;
1008    public:
1009     FollowKlassClosure(MarkAndPushClosure* mark_and_push_closure) :
1010         _mark_and_push_closure(mark_and_push_closure) { }
1011     void do_klass(Klass* klass);
1012   };
1013 
1014   PSParallelCompact();
1015 
1016   // Convenient accessor for Universe::heap().
1017   static ParallelScavengeHeap* gc_heap() {
1018     return (ParallelScavengeHeap*)Universe::heap();
1019   }
1020 
1021   static void invoke(bool maximum_heap_compaction);
1022   static bool invoke_no_policy(bool maximum_heap_compaction);
1023 
1024   static void post_initialize();
1025   // Perform initialization for PSParallelCompact that requires
1026   // allocations.  This should be called during the VM initialization
1027   // at a pointer where it would be appropriate to return a JNI_ENOMEM
1028   // in the event of a failure.
1029   static bool initialize();
1030 
1031   // Closure accessors
1032   static OopClosure* adjust_pointer_closure()      { return (OopClosure*)&_adjust_pointer_closure; }
1033   static KlassClosure* adjust_klass_closure()      { return (KlassClosure*)&_adjust_klass_closure; }
1034   static BoolObjectClosure* is_alive_closure()     { return (BoolObjectClosure*)&_is_alive_closure; }
1035 
1036   // Public accessors
1037   static elapsedTimer* accumulated_time() { return &_accumulated_time; }
1038   static unsigned int total_invocations() { return _total_invocations; }
1039   static CollectorCounters* counters()    { return _counters; }
1040 
1041   // Used to add tasks
1042   static GCTaskManager* const gc_task_manager();
1043   static Klass* updated_int_array_klass_obj() {
1044     return _updated_int_array_klass_obj;
1045   }
1046 
1047   // Marking support
1048   static inline bool mark_obj(oop obj);
1049   static inline bool is_marked(oop obj);
1050   // Check mark and maybe push on marking stack
1051   template <class T> static inline void mark_and_push(ParCompactionManager* cm,
1052                                                       T* p);
1053   template <class T> static inline void adjust_pointer(T* p);
1054 
1055   static void follow_klass(ParCompactionManager* cm, Klass* klass);
1056   static void adjust_klass(ParCompactionManager* cm, Klass* klass);
1057 
1058   static void follow_class_loader(ParCompactionManager* cm,
1059                                   ClassLoaderData* klass);
1060   static void adjust_class_loader(ParCompactionManager* cm,
1061                                   ClassLoaderData* klass);
1062 
1063   // Compaction support.
1064   // Return true if p is in the range [beg_addr, end_addr).
1065   static inline bool is_in(HeapWord* p, HeapWord* beg_addr, HeapWord* end_addr);
1066   static inline bool is_in(oop* p, HeapWord* beg_addr, HeapWord* end_addr);
1067 
1068   // Convenience wrappers for per-space data kept in _space_info.
1069   static inline MutableSpace*     space(SpaceId space_id);
1070   static inline HeapWord*         new_top(SpaceId space_id);
1071   static inline HeapWord*         dense_prefix(SpaceId space_id);
1072   static inline ObjectStartArray* start_array(SpaceId space_id);
1073 
1074   // Move and update the live objects in the specified space.
1075   static void move_and_update(ParCompactionManager* cm, SpaceId space_id);
1076 
1077   // Process the end of the given region range in the dense prefix.
1078   // This includes saving any object not updated.
1079   static void dense_prefix_regions_epilogue(ParCompactionManager* cm,
1080                                             size_t region_start_index,
1081                                             size_t region_end_index,
1082                                             idx_t exiting_object_offset,
1083                                             idx_t region_offset_start,
1084                                             idx_t region_offset_end);
1085 
1086   // Update a region in the dense prefix.  For each live object
1087   // in the region, update it's interior references.  For each
1088   // dead object, fill it with deadwood. Dead space at the end
1089   // of a region range will be filled to the start of the next
1090   // live object regardless of the region_index_end.  None of the
1091   // objects in the dense prefix move and dead space is dead
1092   // (holds only dead objects that don't need any processing), so
1093   // dead space can be filled in any order.
1094   static void update_and_deadwood_in_dense_prefix(ParCompactionManager* cm,
1095                                                   SpaceId space_id,
1096                                                   size_t region_index_start,
1097                                                   size_t region_index_end);
1098 
1099   // Return the address of the count + 1st live word in the range [beg, end).
1100   static HeapWord* skip_live_words(HeapWord* beg, HeapWord* end, size_t count);
1101 
1102   // Return the address of the word to be copied to dest_addr, which must be
1103   // aligned to a region boundary.
1104   static HeapWord* first_src_addr(HeapWord* const dest_addr,
1105                                   SpaceId src_space_id,
1106                                   size_t src_region_idx);
1107 
1108   // Determine the next source region, set closure.source() to the start of the
1109   // new region return the region index.  Parameter end_addr is the address one
1110   // beyond the end of source range just processed.  If necessary, switch to a
1111   // new source space and set src_space_id (in-out parameter) and src_space_top
1112   // (out parameter) accordingly.
1113   static size_t next_src_region(MoveAndUpdateClosure& closure,
1114                                 SpaceId& src_space_id,
1115                                 HeapWord*& src_space_top,
1116                                 HeapWord* end_addr);
1117 
1118   // Decrement the destination count for each non-empty source region in the
1119   // range [beg_region, region(region_align_up(end_addr))).  If the destination
1120   // count for a region goes to 0 and it needs to be filled, enqueue it.
1121   static void decrement_destination_counts(ParCompactionManager* cm,
1122                                            SpaceId src_space_id,
1123                                            size_t beg_region,
1124                                            HeapWord* end_addr);
1125 
1126   // Fill a region, copying objects from one or more source regions.
1127   static void fill_region(ParCompactionManager* cm, size_t region_idx);
1128   static void fill_and_update_region(ParCompactionManager* cm, size_t region) {
1129     fill_region(cm, region);
1130   }
1131 
1132   // Update the deferred objects in the space.
1133   static void update_deferred_objects(ParCompactionManager* cm, SpaceId id);
1134 
1135   static ParMarkBitMap* mark_bitmap() { return &_mark_bitmap; }
1136   static ParallelCompactData& summary_data() { return _summary_data; }
1137 
1138   // Reference Processing
1139   static ReferenceProcessor* const ref_processor() { return _ref_processor; }
1140 
1141   // Return the SpaceId for the given address.
1142   static SpaceId space_id(HeapWord* addr);
1143 
1144   // Time since last full gc (in milliseconds).
1145   static jlong millis_since_last_gc();
1146 
1147   static void print_on_error(outputStream* st);
1148 
1149 #ifndef PRODUCT
1150   // Debugging support.
1151   static const char* space_names[last_space_id];
1152   static void print_region_ranges();
1153   static void print_dense_prefix_stats(const char* const algorithm,
1154                                        const SpaceId id,
1155                                        const bool maximum_compaction,
1156                                        HeapWord* const addr);
1157   static void summary_phase_msg(SpaceId dst_space_id,
1158                                 HeapWord* dst_beg, HeapWord* dst_end,
1159                                 SpaceId src_space_id,
1160                                 HeapWord* src_beg, HeapWord* src_end);
1161 #endif  // #ifndef PRODUCT
1162 
1163 #ifdef  ASSERT
1164   // Sanity check the new location of a word in the heap.
1165   static inline void check_new_location(HeapWord* old_addr, HeapWord* new_addr);
1166   // Verify that all the regions have been emptied.
1167   static void verify_complete(SpaceId space_id);
1168 #endif  // #ifdef ASSERT
1169 };
1170 
1171 inline bool PSParallelCompact::mark_obj(oop obj) {
1172   const int obj_size = obj->size();
1173   if (mark_bitmap()->mark_obj(obj, obj_size)) {
1174     _summary_data.add_obj(obj, obj_size);
1175     return true;
1176   } else {
1177     return false;
1178   }
1179 }
1180 
1181 inline bool PSParallelCompact::is_marked(oop obj) {
1182   return mark_bitmap()->is_marked(obj);
1183 }
1184 
1185 template <class T>
1186 inline void PSParallelCompact::follow_root(ParCompactionManager* cm, T* p) {
1187   assert(!Universe::heap()->is_in_reserved(p),
1188          "roots shouldn't be things within the heap");
1189 
1190   T heap_oop = oopDesc::load_heap_oop(p);
1191   if (!oopDesc::is_null(heap_oop)) {
1192     oop obj = oopDesc::decode_heap_oop_not_null(heap_oop);
1193     if (mark_bitmap()->is_unmarked(obj)) {
1194       if (mark_obj(obj)) {
1195         obj->follow_contents(cm);
1196       }
1197     }
1198   }
1199   cm->follow_marking_stacks();
1200 }
1201 
1202 template <class T>
1203 inline void PSParallelCompact::mark_and_push(ParCompactionManager* cm, T* p) {
1204   T heap_oop = oopDesc::load_heap_oop(p);
1205   if (!oopDesc::is_null(heap_oop)) {
1206     oop obj = oopDesc::decode_heap_oop_not_null(heap_oop);
1207     if (mark_bitmap()->is_unmarked(obj) && mark_obj(obj)) {
1208       cm->push(obj);
1209     }
1210   }
1211 }
1212 
1213 template <class T>
1214 inline void PSParallelCompact::adjust_pointer(T* p) {
1215   T heap_oop = oopDesc::load_heap_oop(p);
1216   if (!oopDesc::is_null(heap_oop)) {
1217     oop obj     = oopDesc::decode_heap_oop_not_null(heap_oop);
1218     oop new_obj = (oop)summary_data().calc_new_pointer(obj);
1219     assert(new_obj != NULL,                    // is forwarding ptr?
1220            "should be forwarded");
1221     // Just always do the update unconditionally?
1222     if (new_obj != NULL) {
1223       assert(Universe::heap()->is_in_reserved(new_obj),
1224              "should be in object space");
1225       oopDesc::encode_store_heap_oop_not_null(p, new_obj);
1226     }
1227   }
1228 }
1229 
1230 template <class T>
1231 inline void PSParallelCompact::KeepAliveClosure::do_oop_work(T* p) {
1232   mark_and_push(_compaction_manager, p);
1233 }
1234 
1235 inline bool PSParallelCompact::print_phases() {
1236   return _print_phases;
1237 }
1238 
1239 inline double PSParallelCompact::normal_distribution(double density) {
1240   assert(_dwl_initialized, "uninitialized");
1241   const double squared_term = (density - _dwl_mean) / _dwl_std_dev;
1242   return _dwl_first_term * exp(-0.5 * squared_term * squared_term);
1243 }
1244 
1245 inline bool
1246 PSParallelCompact::dead_space_crosses_boundary(const RegionData* region,
1247                                                idx_t bit)
1248 {
1249   assert(bit > 0, "cannot call this for the first bit/region");
1250   assert(_summary_data.region_to_addr(region) == _mark_bitmap.bit_to_addr(bit),
1251          "sanity check");
1252 
1253   // Dead space crosses the boundary if (1) a partial object does not extend
1254   // onto the region, (2) an object does not start at the beginning of the
1255   // region, and (3) an object does not end at the end of the prior region.
1256   return region->partial_obj_size() == 0 &&
1257     !_mark_bitmap.is_obj_beg(bit) &&
1258     !_mark_bitmap.is_obj_end(bit - 1);
1259 }
1260 
1261 inline bool
1262 PSParallelCompact::is_in(HeapWord* p, HeapWord* beg_addr, HeapWord* end_addr) {
1263   return p >= beg_addr && p < end_addr;
1264 }
1265 
1266 inline bool
1267 PSParallelCompact::is_in(oop* p, HeapWord* beg_addr, HeapWord* end_addr) {
1268   return is_in((HeapWord*)p, beg_addr, end_addr);
1269 }
1270 
1271 inline MutableSpace* PSParallelCompact::space(SpaceId id) {
1272   assert(id < last_space_id, "id out of range");
1273   return _space_info[id].space();
1274 }
1275 
1276 inline HeapWord* PSParallelCompact::new_top(SpaceId id) {
1277   assert(id < last_space_id, "id out of range");
1278   return _space_info[id].new_top();
1279 }
1280 
1281 inline HeapWord* PSParallelCompact::dense_prefix(SpaceId id) {
1282   assert(id < last_space_id, "id out of range");
1283   return _space_info[id].dense_prefix();
1284 }
1285 
1286 inline ObjectStartArray* PSParallelCompact::start_array(SpaceId id) {
1287   assert(id < last_space_id, "id out of range");
1288   return _space_info[id].start_array();
1289 }
1290 
1291 #ifdef ASSERT
1292 inline void
1293 PSParallelCompact::check_new_location(HeapWord* old_addr, HeapWord* new_addr)
1294 {
1295   assert(old_addr >= new_addr || space_id(old_addr) != space_id(new_addr),
1296          "must move left or to a different space");
1297   assert(is_object_aligned((intptr_t)old_addr) && is_object_aligned((intptr_t)new_addr),
1298          "checking alignment");
1299 }
1300 #endif // ASSERT
1301 
1302 class MoveAndUpdateClosure: public ParMarkBitMapClosure {
1303  public:
1304   inline MoveAndUpdateClosure(ParMarkBitMap* bitmap, ParCompactionManager* cm,
1305                               ObjectStartArray* start_array,
1306                               HeapWord* destination, size_t words);
1307 
1308   // Accessors.
1309   HeapWord* destination() const         { return _destination; }
1310 
1311   // If the object will fit (size <= words_remaining()), copy it to the current
1312   // destination, update the interior oops and the start array and return either
1313   // full (if the closure is full) or incomplete.  If the object will not fit,
1314   // return would_overflow.
1315   virtual IterationStatus do_addr(HeapWord* addr, size_t size);
1316 
1317   // Copy enough words to fill this closure, starting at source().  Interior
1318   // oops and the start array are not updated.  Return full.
1319   IterationStatus copy_until_full();
1320 
1321   // Copy enough words to fill this closure or to the end of an object,
1322   // whichever is smaller, starting at source().  Interior oops and the start
1323   // array are not updated.
1324   void copy_partial_obj();
1325 
1326  protected:
1327   // Update variables to indicate that word_count words were processed.
1328   inline void update_state(size_t word_count);
1329 
1330  protected:
1331   ObjectStartArray* const _start_array;
1332   HeapWord*               _destination;         // Next addr to be written.
1333 };
1334 
1335 inline
1336 MoveAndUpdateClosure::MoveAndUpdateClosure(ParMarkBitMap* bitmap,
1337                                            ParCompactionManager* cm,
1338                                            ObjectStartArray* start_array,
1339                                            HeapWord* destination,
1340                                            size_t words) :
1341   ParMarkBitMapClosure(bitmap, cm, words), _start_array(start_array)
1342 {
1343   _destination = destination;
1344 }
1345 
1346 inline void MoveAndUpdateClosure::update_state(size_t words)
1347 {
1348   decrement_words_remaining(words);
1349   _source += words;
1350   _destination += words;
1351 }
1352 
1353 class UpdateOnlyClosure: public ParMarkBitMapClosure {
1354  private:
1355   const PSParallelCompact::SpaceId _space_id;
1356   ObjectStartArray* const          _start_array;
1357 
1358  public:
1359   UpdateOnlyClosure(ParMarkBitMap* mbm,
1360                     ParCompactionManager* cm,
1361                     PSParallelCompact::SpaceId space_id);
1362 
1363   // Update the object.
1364   virtual IterationStatus do_addr(HeapWord* addr, size_t words);
1365 
1366   inline void do_addr(HeapWord* addr);
1367 };
1368 
1369 inline void UpdateOnlyClosure::do_addr(HeapWord* addr)
1370 {
1371   _start_array->allocate_block(addr);
1372   oop(addr)->update_contents(compaction_manager());
1373 }
1374 
1375 class FillClosure: public ParMarkBitMapClosure
1376 {
1377 public:
1378   FillClosure(ParCompactionManager* cm, PSParallelCompact::SpaceId space_id) :
1379     ParMarkBitMapClosure(PSParallelCompact::mark_bitmap(), cm),
1380     _start_array(PSParallelCompact::start_array(space_id))
1381   {
1382     assert(space_id == PSParallelCompact::old_space_id,
1383            "cannot use FillClosure in the young gen");
1384   }
1385 
1386   virtual IterationStatus do_addr(HeapWord* addr, size_t size) {
1387     CollectedHeap::fill_with_objects(addr, size);
1388     HeapWord* const end = addr + size;
1389     do {
1390       _start_array->allocate_block(addr);
1391       addr += oop(addr)->size();
1392     } while (addr < end);
1393     return ParMarkBitMap::incomplete;
1394   }
1395 
1396 private:
1397   ObjectStartArray* const _start_array;
1398 };
1399 
1400 #endif // SHARE_VM_GC_IMPLEMENTATION_PARALLELSCAVENGE_PSPARALLELCOMPACT_HPP