1 /*
   2  * Copyright (c) 2000, 2009, 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 // This kind of "BarrierSet" allows a "CollectedHeap" to detect and
  26 // enumerate ref fields that have been modified (since the last
  27 // enumeration.)
  28 
  29 // As it currently stands, this barrier is *imprecise*: when a ref field in
  30 // an object "o" is modified, the card table entry for the card containing
  31 // the head of "o" is dirtied, not necessarily the card containing the
  32 // modified field itself.  For object arrays, however, the barrier *is*
  33 // precise; only the card containing the modified element is dirtied.
  34 // Any MemRegionClosures used to scan dirty cards should take these
  35 // considerations into account.
  36 
  37 class Generation;
  38 class OopsInGenClosure;
  39 class DirtyCardToOopClosure;
  40 
  41 class CardTableModRefBS: public ModRefBarrierSet {
  42   // Some classes get to look at some private stuff.
  43   friend class BytecodeInterpreter;
  44   friend class VMStructs;
  45   friend class CardTableRS;
  46   friend class CheckForUnmarkedOops; // Needs access to raw card bytes.
  47   friend class SharkBuilder;
  48 #ifndef PRODUCT
  49   // For debugging.
  50   friend class GuaranteeNotModClosure;
  51 #endif
  52  protected:
  53 
  54   enum CardValues {
  55     clean_card                  = -1,
  56     // The mask contains zeros in places for all other values.
  57     clean_card_mask             = clean_card - 31,
  58 
  59     dirty_card                  =  0,
  60     precleaned_card             =  1,
  61     claimed_card                =  2,
  62     deferred_card               =  4,
  63     last_card                   =  8,
  64     CT_MR_BS_last_reserved      = 16
  65   };
  66 
  67   // dirty and precleaned are equivalent wrt younger_refs_iter.
  68   static bool card_is_dirty_wrt_gen_iter(jbyte cv) {
  69     return cv == dirty_card || cv == precleaned_card;
  70   }
  71 
  72   // Returns "true" iff the value "cv" will cause the card containing it
  73   // to be scanned in the current traversal.  May be overridden by
  74   // subtypes.
  75   virtual bool card_will_be_scanned(jbyte cv) {
  76     return CardTableModRefBS::card_is_dirty_wrt_gen_iter(cv);
  77   }
  78 
  79   // Returns "true" iff the value "cv" may have represented a dirty card at
  80   // some point.
  81   virtual bool card_may_have_been_dirty(jbyte cv) {
  82     return card_is_dirty_wrt_gen_iter(cv);
  83   }
  84 
  85   // The declaration order of these const fields is important; see the
  86   // constructor before changing.
  87   const MemRegion _whole_heap;       // the region covered by the card table
  88   const size_t    _guard_index;      // index of very last element in the card
  89                                      // table; it is set to a guard value
  90                                      // (last_card) and should never be modified
  91   const size_t    _last_valid_index; // index of the last valid element
  92   const size_t    _page_size;        // page size used when mapping _byte_map
  93   const size_t    _byte_map_size;    // in bytes
  94   jbyte*          _byte_map;         // the card marking array
  95 
  96   int _cur_covered_regions;
  97   // The covered regions should be in address order.
  98   MemRegion* _covered;
  99   // The committed regions correspond one-to-one to the covered regions.
 100   // They represent the card-table memory that has been committed to service
 101   // the corresponding covered region.  It may be that committed region for
 102   // one covered region corresponds to a larger region because of page-size
 103   // roundings.  Thus, a committed region for one covered region may
 104   // actually extend onto the card-table space for the next covered region.
 105   MemRegion* _committed;
 106 
 107   // The last card is a guard card, and we commit the page for it so
 108   // we can use the card for verification purposes. We make sure we never
 109   // uncommit the MemRegion for that page.
 110   MemRegion _guard_region;
 111 
 112  protected:
 113   // Initialization utilities; covered_words is the size of the covered region
 114   // in, um, words.
 115   inline size_t cards_required(size_t covered_words);
 116   inline size_t compute_byte_map_size();
 117 
 118   // Finds and return the index of the region, if any, to which the given
 119   // region would be contiguous.  If none exists, assign a new region and
 120   // returns its index.  Requires that no more than the maximum number of
 121   // covered regions defined in the constructor are ever in use.
 122   int find_covering_region_by_base(HeapWord* base);
 123 
 124   // Same as above, but finds the region containing the given address
 125   // instead of starting at a given base address.
 126   int find_covering_region_containing(HeapWord* addr);
 127 
 128   // Resize one of the regions covered by the remembered set.
 129   void resize_covered_region(MemRegion new_region);
 130 
 131   // Returns the leftmost end of a committed region corresponding to a
 132   // covered region before covered region "ind", or else "NULL" if "ind" is
 133   // the first covered region.
 134   HeapWord* largest_prev_committed_end(int ind) const;
 135 
 136   // Returns the part of the region mr that doesn't intersect with
 137   // any committed region other than self.  Used to prevent uncommitting
 138   // regions that are also committed by other regions.  Also protects
 139   // against uncommitting the guard region.
 140   MemRegion committed_unique_to_self(int self, MemRegion mr) const;
 141 
 142   // Mapping from address to card marking array entry
 143   jbyte* byte_for(const void* p) const {
 144     assert(_whole_heap.contains(p),
 145            "out of bounds access to card marking array");
 146     jbyte* result = &byte_map_base[uintptr_t(p) >> card_shift];
 147     assert(result >= _byte_map && result < _byte_map + _byte_map_size,
 148            "out of bounds accessor for card marking array");
 149     return result;
 150   }
 151 
 152   // The card table byte one after the card marking array
 153   // entry for argument address. Typically used for higher bounds
 154   // for loops iterating through the card table.
 155   jbyte* byte_after(const void* p) const {
 156     return byte_for(p) + 1;
 157   }
 158 
 159   // Iterate over the portion of the card-table which covers the given
 160   // region mr in the given space and apply cl to any dirty sub-regions
 161   // of mr. cl and dcto_cl must either be the same closure or cl must
 162   // wrap dcto_cl. Both are required - neither may be NULL. Also, dcto_cl
 163   // may be modified. Note that this function will operate in a parallel
 164   // mode if worker threads are available.
 165   void non_clean_card_iterate(Space* sp, MemRegion mr,
 166                               DirtyCardToOopClosure* dcto_cl,
 167                               MemRegionClosure* cl,
 168                               bool clear);
 169 
 170   // Utility function used to implement the other versions below.
 171   void non_clean_card_iterate_work(MemRegion mr, MemRegionClosure* cl,
 172                                    bool clear);
 173 
 174   void par_non_clean_card_iterate_work(Space* sp, MemRegion mr,
 175                                        DirtyCardToOopClosure* dcto_cl,
 176                                        MemRegionClosure* cl,
 177                                        bool clear,
 178                                        int n_threads);
 179 
 180   // Dirty the bytes corresponding to "mr" (not all of which must be
 181   // covered.)
 182   void dirty_MemRegion(MemRegion mr);
 183 
 184   // Clear (to clean_card) the bytes entirely contained within "mr" (not
 185   // all of which must be covered.)
 186   void clear_MemRegion(MemRegion mr);
 187 
 188   // *** Support for parallel card scanning.
 189 
 190   enum SomeConstantsForParallelism {
 191     StridesPerThread    = 2,
 192     CardsPerStrideChunk = 256
 193   };
 194 
 195   // This is an array, one element per covered region of the card table.
 196   // Each entry is itself an array, with one element per chunk in the
 197   // covered region.  Each entry of these arrays is the lowest non-clean
 198   // card of the corresponding chunk containing part of an object from the
 199   // previous chunk, or else NULL.
 200   typedef jbyte*  CardPtr;
 201   typedef CardPtr* CardArr;
 202   CardArr* _lowest_non_clean;
 203   size_t*  _lowest_non_clean_chunk_size;
 204   uintptr_t* _lowest_non_clean_base_chunk_index;
 205   int* _last_LNC_resizing_collection;
 206 
 207   // Initializes "lowest_non_clean" to point to the array for the region
 208   // covering "sp", and "lowest_non_clean_base_chunk_index" to the chunk
 209   // index of the corresponding to the first element of that array.
 210   // Ensures that these arrays are of sufficient size, allocating if necessary.
 211   // May be called by several threads concurrently.
 212   void get_LNC_array_for_space(Space* sp,
 213                                jbyte**& lowest_non_clean,
 214                                uintptr_t& lowest_non_clean_base_chunk_index,
 215                                size_t& lowest_non_clean_chunk_size);
 216 
 217   // Returns the number of chunks necessary to cover "mr".
 218   size_t chunks_to_cover(MemRegion mr) {
 219     return (size_t)(addr_to_chunk_index(mr.last()) -
 220                     addr_to_chunk_index(mr.start()) + 1);
 221   }
 222 
 223   // Returns the index of the chunk in a stride which
 224   // covers the given address.
 225   uintptr_t addr_to_chunk_index(const void* addr) {
 226     uintptr_t card = (uintptr_t) byte_for(addr);
 227     return card / CardsPerStrideChunk;
 228   }
 229 
 230   // Apply cl, which must either itself apply dcto_cl or be dcto_cl,
 231   // to the cards in the stride (of n_strides) within the given space.
 232   void process_stride(Space* sp,
 233                       MemRegion used,
 234                       jint stride, int n_strides,
 235                       DirtyCardToOopClosure* dcto_cl,
 236                       MemRegionClosure* cl,
 237                       bool clear,
 238                       jbyte** lowest_non_clean,
 239                       uintptr_t lowest_non_clean_base_chunk_index,
 240                       size_t lowest_non_clean_chunk_size);
 241 
 242   // Makes sure that chunk boundaries are handled appropriately, by
 243   // adjusting the min_done of dcto_cl, and by using a special card-table
 244   // value to indicate how min_done should be set.
 245   void process_chunk_boundaries(Space* sp,
 246                                 DirtyCardToOopClosure* dcto_cl,
 247                                 MemRegion chunk_mr,
 248                                 MemRegion used,
 249                                 jbyte** lowest_non_clean,
 250                                 uintptr_t lowest_non_clean_base_chunk_index,
 251                                 size_t    lowest_non_clean_chunk_size);
 252 
 253 public:
 254   // Constants
 255   enum SomePublicConstants {
 256     card_shift                  = 9,
 257     card_size                   = 1 << card_shift,
 258     card_size_in_words          = card_size / sizeof(HeapWord)
 259   };
 260 
 261   static int clean_card_val()      { return clean_card; }
 262   static int clean_card_mask_val() { return clean_card_mask; }
 263   static int dirty_card_val()      { return dirty_card; }
 264   static int claimed_card_val()    { return claimed_card; }
 265   static int precleaned_card_val() { return precleaned_card; }
 266   static int deferred_card_val()   { return deferred_card; }
 267 
 268   // For RTTI simulation.
 269   bool is_a(BarrierSet::Name bsn) {
 270     return bsn == BarrierSet::CardTableModRef || ModRefBarrierSet::is_a(bsn);
 271   }
 272 
 273   CardTableModRefBS(MemRegion whole_heap, int max_covered_regions);
 274 
 275   // *** Barrier set functions.
 276 
 277   bool has_write_ref_pre_barrier() { return false; }
 278 
 279   inline bool write_ref_needs_barrier(void* field, oop new_val) {
 280     // Note that this assumes the perm gen is the highest generation
 281     // in the address space
 282     return new_val != NULL && !new_val->is_perm();
 283   }
 284 
 285   // Record a reference update. Note that these versions are precise!
 286   // The scanning code has to handle the fact that the write barrier may be
 287   // either precise or imprecise. We make non-virtual inline variants of
 288   // these functions here for performance.
 289 protected:
 290   void write_ref_field_work(oop obj, size_t offset, oop newVal);
 291   virtual void write_ref_field_work(void* field, oop newVal);
 292 public:
 293 
 294   bool has_write_ref_array_opt() { return true; }
 295   bool has_write_region_opt() { return true; }
 296 
 297   inline void inline_write_region(MemRegion mr) {
 298     dirty_MemRegion(mr);
 299   }
 300 protected:
 301   void write_region_work(MemRegion mr) {
 302     inline_write_region(mr);
 303   }
 304 public:
 305 
 306   inline void inline_write_ref_array(MemRegion mr) {
 307     dirty_MemRegion(mr);
 308   }
 309 protected:
 310   void write_ref_array_work(MemRegion mr) {
 311     inline_write_ref_array(mr);
 312   }
 313 public:
 314 
 315   bool is_aligned(HeapWord* addr) {
 316     return is_card_aligned(addr);
 317   }
 318 
 319   // *** Card-table-barrier-specific things.
 320 
 321   template <class T> inline void inline_write_ref_field_pre(T* field, oop newVal) {}
 322 
 323   template <class T> inline void inline_write_ref_field(T* field, oop newVal) {
 324     jbyte* byte = byte_for((void*)field);
 325     *byte = dirty_card;
 326   }
 327 
 328   // These are used by G1, when it uses the card table as a temporary data
 329   // structure for card claiming.
 330   bool is_card_dirty(size_t card_index) {
 331     return _byte_map[card_index] == dirty_card_val();
 332   }
 333 
 334   void mark_card_dirty(size_t card_index) {
 335     _byte_map[card_index] = dirty_card_val();
 336   }
 337 
 338   bool is_card_claimed(size_t card_index) {
 339     jbyte val = _byte_map[card_index];
 340     return (val & (clean_card_mask_val() | claimed_card_val())) == claimed_card_val();
 341   }
 342 
 343   void set_card_claimed(size_t card_index) {
 344       jbyte val = _byte_map[card_index];
 345       if (val == clean_card_val()) {
 346         val = (jbyte)claimed_card_val();
 347       } else {
 348         val |= (jbyte)claimed_card_val();
 349       }
 350       _byte_map[card_index] = val;
 351   }
 352 
 353   bool claim_card(size_t card_index);
 354 
 355   bool is_card_clean(size_t card_index) {
 356     return _byte_map[card_index] == clean_card_val();
 357   }
 358 
 359   bool is_card_deferred(size_t card_index) {
 360     jbyte val = _byte_map[card_index];
 361     return (val & (clean_card_mask_val() | deferred_card_val())) == deferred_card_val();
 362   }
 363 
 364   bool mark_card_deferred(size_t card_index);
 365 
 366   // Card marking array base (adjusted for heap low boundary)
 367   // This would be the 0th element of _byte_map, if the heap started at 0x0.
 368   // But since the heap starts at some higher address, this points to somewhere
 369   // before the beginning of the actual _byte_map.
 370   jbyte* byte_map_base;
 371 
 372   // Return true if "p" is at the start of a card.
 373   bool is_card_aligned(HeapWord* p) {
 374     jbyte* pcard = byte_for(p);
 375     return (addr_for(pcard) == p);
 376   }
 377 
 378   // The kinds of precision a CardTableModRefBS may offer.
 379   enum PrecisionStyle {
 380     Precise,
 381     ObjHeadPreciseArray
 382   };
 383 
 384   // Tells what style of precision this card table offers.
 385   PrecisionStyle precision() {
 386     return ObjHeadPreciseArray; // Only one supported for now.
 387   }
 388 
 389   // ModRefBS functions.
 390   virtual void invalidate(MemRegion mr, bool whole_heap = false);
 391   void clear(MemRegion mr);
 392   void dirty(MemRegion mr);
 393   void mod_oop_in_space_iterate(Space* sp, OopClosure* cl,
 394                                 bool clear = false,
 395                                 bool before_save_marks = false);
 396 
 397   // *** Card-table-RemSet-specific things.
 398 
 399   // Invoke "cl.do_MemRegion" on a set of MemRegions that collectively
 400   // includes all the modified cards (expressing each card as a
 401   // MemRegion).  Thus, several modified cards may be lumped into one
 402   // region.  The regions are non-overlapping, and are visited in
 403   // *decreasing* address order.  (This order aids with imprecise card
 404   // marking, where a dirty card may cause scanning, and summarization
 405   // marking, of objects that extend onto subsequent cards.)
 406   // If "clear" is true, the card is (conceptually) marked unmodified before
 407   // applying the closure.
 408   void mod_card_iterate(MemRegionClosure* cl, bool clear = false) {
 409     non_clean_card_iterate_work(_whole_heap, cl, clear);
 410   }
 411 
 412   // Like the "mod_cards_iterate" above, except only invokes the closure
 413   // for cards within the MemRegion "mr" (which is required to be
 414   // card-aligned and sized.)
 415   void mod_card_iterate(MemRegion mr, MemRegionClosure* cl,
 416                         bool clear = false) {
 417     non_clean_card_iterate_work(mr, cl, clear);
 418   }
 419 
 420   static uintx ct_max_alignment_constraint();
 421 
 422   // Apply closure "cl" to the dirty cards containing some part of
 423   // MemRegion "mr".
 424   void dirty_card_iterate(MemRegion mr, MemRegionClosure* cl);
 425 
 426   // Return the MemRegion corresponding to the first maximal run
 427   // of dirty cards lying completely within MemRegion mr.
 428   // If reset is "true", then sets those card table entries to the given
 429   // value.
 430   MemRegion dirty_card_range_after_reset(MemRegion mr, bool reset,
 431                                          int reset_val);
 432 
 433   // Set all the dirty cards in the given region to precleaned state.
 434   void preclean_dirty_cards(MemRegion mr);
 435 
 436   // Provide read-only access to the card table array.
 437   const jbyte* byte_for_const(const void* p) const {
 438     return byte_for(p);
 439   }
 440   const jbyte* byte_after_const(const void* p) const {
 441     return byte_after(p);
 442   }
 443 
 444   // Mapping from card marking array entry to address of first word
 445   HeapWord* addr_for(const jbyte* p) const {
 446     assert(p >= _byte_map && p < _byte_map + _byte_map_size,
 447            "out of bounds access to card marking array");
 448     size_t delta = pointer_delta(p, byte_map_base, sizeof(jbyte));
 449     HeapWord* result = (HeapWord*) (delta << card_shift);
 450     assert(_whole_heap.contains(result),
 451            "out of bounds accessor from card marking array");
 452     return result;
 453   }
 454 
 455   // Mapping from address to card marking array index.
 456   size_t index_for(void* p) {
 457     assert(_whole_heap.contains(p),
 458            "out of bounds access to card marking array");
 459     return byte_for(p) - _byte_map;
 460   }
 461 
 462   const jbyte* byte_for_index(const size_t card_index) const {
 463     return _byte_map + card_index;
 464   }
 465 
 466   void verify();
 467   void verify_guard();
 468 
 469   void verify_clean_region(MemRegion mr) PRODUCT_RETURN;
 470   void verify_dirty_region(MemRegion mr) PRODUCT_RETURN;
 471 
 472   static size_t par_chunk_heapword_alignment() {
 473     return CardsPerStrideChunk * card_size_in_words;
 474   }
 475 
 476 };
 477 
 478 class CardTableRS;
 479 
 480 // A specialization for the CardTableRS gen rem set.
 481 class CardTableModRefBSForCTRS: public CardTableModRefBS {
 482   CardTableRS* _rs;
 483 protected:
 484   bool card_will_be_scanned(jbyte cv);
 485   bool card_may_have_been_dirty(jbyte cv);
 486 public:
 487   CardTableModRefBSForCTRS(MemRegion whole_heap,
 488                            int max_covered_regions) :
 489     CardTableModRefBS(whole_heap, max_covered_regions) {}
 490 
 491   void set_CTRS(CardTableRS* rs) { _rs = rs; }
 492 };