1 /*
   2  * Copyright (c) 2000, 2014, 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_MEMORY_CARDTABLEMODREFBS_HPP
  26 #define SHARE_VM_MEMORY_CARDTABLEMODREFBS_HPP
  27 
  28 #include "memory/modRefBarrierSet.hpp"
  29 #include "oops/oop.hpp"
  30 #include "oops/oop.inline2.hpp"
  31 
  32 // This kind of "BarrierSet" allows a "CollectedHeap" to detect and
  33 // enumerate ref fields that have been modified (since the last
  34 // enumeration.)
  35 
  36 // As it currently stands, this barrier is *imprecise*: when a ref field in
  37 // an object "o" is modified, the card table entry for the card containing
  38 // the head of "o" is dirtied, not necessarily the card containing the
  39 // modified field itself.  For object arrays, however, the barrier *is*
  40 // precise; only the card containing the modified element is dirtied.
  41 // Any MemRegionClosures used to scan dirty cards should take these
  42 // considerations into account.
  43 
  44 class Generation;
  45 class OopsInGenClosure;
  46 class DirtyCardToOopClosure;
  47 class ClearNoncleanCardWrapper;
  48 
  49 class CardTableModRefBS: public ModRefBarrierSet {
  50   // Some classes get to look at some private stuff.
  51   friend class BytecodeInterpreter;
  52   friend class VMStructs;
  53   friend class CardTableRS;
  54   friend class CheckForUnmarkedOops; // Needs access to raw card bytes.
  55   friend class SharkBuilder;
  56 #ifndef PRODUCT
  57   // For debugging.
  58   friend class GuaranteeNotModClosure;
  59 #endif
  60  protected:
  61 
  62   enum CardValues {
  63     clean_card                  = -1,
  64     // The mask contains zeros in places for all other values.
  65     clean_card_mask             = clean_card - 31,
  66 
  67     dirty_card                  =  0,
  68     precleaned_card             =  1,
  69     claimed_card                =  2,
  70     deferred_card               =  4,
  71     last_card                   =  8,
  72     CT_MR_BS_last_reserved      = 16
  73   };
  74 
  75   // a word's worth (row) of clean card values
  76   static const intptr_t clean_card_row = (intptr_t)(-1);
  77 
  78   // dirty and precleaned are equivalent wrt younger_refs_iter.
  79   static bool card_is_dirty_wrt_gen_iter(jbyte cv) {
  80     return cv == dirty_card || cv == precleaned_card;
  81   }
  82 
  83   // Returns "true" iff the value "cv" will cause the card containing it
  84   // to be scanned in the current traversal.  May be overridden by
  85   // subtypes.
  86   virtual bool card_will_be_scanned(jbyte cv) {
  87     return CardTableModRefBS::card_is_dirty_wrt_gen_iter(cv);
  88   }
  89 
  90   // Returns "true" iff the value "cv" may have represented a dirty card at
  91   // some point.
  92   virtual bool card_may_have_been_dirty(jbyte cv) {
  93     return card_is_dirty_wrt_gen_iter(cv);
  94   }
  95 
  96   // The declaration order of these const fields is important; see the
  97   // constructor before changing.
  98   const MemRegion _whole_heap;       // the region covered by the card table
  99   size_t          _guard_index;      // index of very last element in the card
 100                                      // table; it is set to a guard value
 101                                      // (last_card) and should never be modified
 102   size_t          _last_valid_index; // index of the last valid element
 103   const size_t    _page_size;        // page size used when mapping _byte_map
 104   size_t          _byte_map_size;    // in bytes
 105   jbyte*          _byte_map;         // the card marking array
 106 
 107   int _cur_covered_regions;
 108   // The covered regions should be in address order.
 109   MemRegion* _covered;
 110   // The committed regions correspond one-to-one to the covered regions.
 111   // They represent the card-table memory that has been committed to service
 112   // the corresponding covered region.  It may be that committed region for
 113   // one covered region corresponds to a larger region because of page-size
 114   // roundings.  Thus, a committed region for one covered region may
 115   // actually extend onto the card-table space for the next covered region.
 116   MemRegion* _committed;
 117 
 118   // The last card is a guard card, and we commit the page for it so
 119   // we can use the card for verification purposes. We make sure we never
 120   // uncommit the MemRegion for that page.
 121   MemRegion _guard_region;
 122 
 123  protected:
 124   // Initialization utilities; covered_words is the size of the covered region
 125   // in, um, words.
 126   inline size_t cards_required(size_t covered_words) {
 127     // Add one for a guard card, used to detect errors.
 128     const size_t words = align_size_up(covered_words, card_size_in_words);
 129     return words / card_size_in_words + 1;
 130   }
 131 
 132   inline size_t compute_byte_map_size();
 133 
 134   // Finds and return the index of the region, if any, to which the given
 135   // region would be contiguous.  If none exists, assign a new region and
 136   // returns its index.  Requires that no more than the maximum number of
 137   // covered regions defined in the constructor are ever in use.
 138   int find_covering_region_by_base(HeapWord* base);
 139 
 140   // Same as above, but finds the region containing the given address
 141   // instead of starting at a given base address.
 142   int find_covering_region_containing(HeapWord* addr);
 143 
 144   // Resize one of the regions covered by the remembered set.
 145   virtual void resize_covered_region(MemRegion new_region);
 146 
 147   // Returns the leftmost end of a committed region corresponding to a
 148   // covered region before covered region "ind", or else "NULL" if "ind" is
 149   // the first covered region.
 150   HeapWord* largest_prev_committed_end(int ind) const;
 151 
 152   // Returns the part of the region mr that doesn't intersect with
 153   // any committed region other than self.  Used to prevent uncommitting
 154   // regions that are also committed by other regions.  Also protects
 155   // against uncommitting the guard region.
 156   MemRegion committed_unique_to_self(int self, MemRegion mr) const;
 157 
 158   // Mapping from address to card marking array entry
 159   jbyte* byte_for(const void* p) const {
 160     assert(_whole_heap.contains(p),
 161            err_msg("Attempt to access p = "PTR_FORMAT" out of bounds of "
 162                    " card marking array's _whole_heap = ["PTR_FORMAT","PTR_FORMAT")",
 163                    p2i(p), p2i(_whole_heap.start()), p2i(_whole_heap.end())));
 164     jbyte* result = &byte_map_base[uintptr_t(p) >> card_shift];
 165     assert(result >= _byte_map && result < _byte_map + _byte_map_size,
 166            "out of bounds accessor for card marking array");
 167     return result;
 168   }
 169 
 170   // The card table byte one after the card marking array
 171   // entry for argument address. Typically used for higher bounds
 172   // for loops iterating through the card table.
 173   jbyte* byte_after(const void* p) const {
 174     return byte_for(p) + 1;
 175   }
 176 
 177   // Iterate over the portion of the card-table which covers the given
 178   // region mr in the given space and apply cl to any dirty sub-regions
 179   // of mr. Dirty cards are _not_ cleared by the iterator method itself,
 180   // but closures may arrange to do so on their own should they so wish.
 181   void non_clean_card_iterate_serial(MemRegion mr, MemRegionClosure* cl);
 182 
 183   // A variant of the above that will operate in a parallel mode if
 184   // worker threads are available, and clear the dirty cards as it
 185   // processes them.
 186   // XXX ??? MemRegionClosure above vs OopsInGenClosure below XXX
 187   // XXX some new_dcto_cl's take OopClosure's, plus as above there are
 188   // some MemRegionClosures. Clean this up everywhere. XXX
 189   void non_clean_card_iterate_possibly_parallel(Space* sp, MemRegion mr,
 190                                                 OopsInGenClosure* cl, CardTableRS* ct);
 191 
 192  private:
 193   // Work method used to implement non_clean_card_iterate_possibly_parallel()
 194   // above in the parallel case.
 195   void non_clean_card_iterate_parallel_work(Space* sp, MemRegion mr,
 196                                             OopsInGenClosure* cl, CardTableRS* ct,
 197                                             int n_threads);
 198 
 199  protected:
 200   // Dirty the bytes corresponding to "mr" (not all of which must be
 201   // covered.)
 202   void dirty_MemRegion(MemRegion mr);
 203 
 204   // Clear (to clean_card) the bytes entirely contained within "mr" (not
 205   // all of which must be covered.)
 206   void clear_MemRegion(MemRegion mr);
 207 
 208   // *** Support for parallel card scanning.
 209 
 210   // This is an array, one element per covered region of the card table.
 211   // Each entry is itself an array, with one element per chunk in the
 212   // covered region.  Each entry of these arrays is the lowest non-clean
 213   // card of the corresponding chunk containing part of an object from the
 214   // previous chunk, or else NULL.
 215   typedef jbyte*  CardPtr;
 216   typedef CardPtr* CardArr;
 217   CardArr* _lowest_non_clean;
 218   size_t*  _lowest_non_clean_chunk_size;
 219   uintptr_t* _lowest_non_clean_base_chunk_index;
 220   int* _last_LNC_resizing_collection;
 221 
 222   // Initializes "lowest_non_clean" to point to the array for the region
 223   // covering "sp", and "lowest_non_clean_base_chunk_index" to the chunk
 224   // index of the corresponding to the first element of that array.
 225   // Ensures that these arrays are of sufficient size, allocating if necessary.
 226   // May be called by several threads concurrently.
 227   void get_LNC_array_for_space(Space* sp,
 228                                jbyte**& lowest_non_clean,
 229                                uintptr_t& lowest_non_clean_base_chunk_index,
 230                                size_t& lowest_non_clean_chunk_size);
 231 
 232   // Returns the number of chunks necessary to cover "mr".
 233   size_t chunks_to_cover(MemRegion mr) {
 234     return (size_t)(addr_to_chunk_index(mr.last()) -
 235                     addr_to_chunk_index(mr.start()) + 1);
 236   }
 237 
 238   // Returns the index of the chunk in a stride which
 239   // covers the given address.
 240   uintptr_t addr_to_chunk_index(const void* addr) {
 241     uintptr_t card = (uintptr_t) byte_for(addr);
 242     return card / ParGCCardsPerStrideChunk;
 243   }
 244 
 245   // Apply cl, which must either itself apply dcto_cl or be dcto_cl,
 246   // to the cards in the stride (of n_strides) within the given space.
 247   void process_stride(Space* sp,
 248                       MemRegion used,
 249                       jint stride, int n_strides,
 250                       OopsInGenClosure* cl,
 251                       CardTableRS* ct,
 252                       jbyte** lowest_non_clean,
 253                       uintptr_t lowest_non_clean_base_chunk_index,
 254                       size_t lowest_non_clean_chunk_size);
 255 
 256   // Makes sure that chunk boundaries are handled appropriately, by
 257   // adjusting the min_done of dcto_cl, and by using a special card-table
 258   // value to indicate how min_done should be set.
 259   void process_chunk_boundaries(Space* sp,
 260                                 DirtyCardToOopClosure* dcto_cl,
 261                                 MemRegion chunk_mr,
 262                                 MemRegion used,
 263                                 jbyte** lowest_non_clean,
 264                                 uintptr_t lowest_non_clean_base_chunk_index,
 265                                 size_t    lowest_non_clean_chunk_size);
 266 
 267 public:
 268   // Constants
 269   enum SomePublicConstants {
 270     card_shift                  = 9,
 271     card_size                   = 1 << card_shift,
 272     card_size_in_words          = card_size / sizeof(HeapWord)
 273   };
 274 
 275   static int clean_card_val()      { return clean_card; }
 276   static int clean_card_mask_val() { return clean_card_mask; }
 277   static int dirty_card_val()      { return dirty_card; }
 278   static int claimed_card_val()    { return claimed_card; }
 279   static int precleaned_card_val() { return precleaned_card; }
 280   static int deferred_card_val()   { return deferred_card; }
 281 
 282   // For RTTI simulation.
 283   bool is_a(BarrierSet::Name bsn) {
 284     return bsn == BarrierSet::CardTableModRef || ModRefBarrierSet::is_a(bsn);
 285   }
 286 
 287   CardTableModRefBS(MemRegion whole_heap);
 288   ~CardTableModRefBS();
 289 
 290   virtual void initialize();
 291 
 292   // *** Barrier set functions.
 293 
 294   bool has_write_ref_pre_barrier() { return false; }
 295 
 296   // Record a reference update. Note that these versions are precise!
 297   // The scanning code has to handle the fact that the write barrier may be
 298   // either precise or imprecise. We make non-virtual inline variants of
 299   // these functions here for performance.
 300 protected:
 301   void write_ref_field_work(oop obj, size_t offset, oop newVal);
 302   virtual void write_ref_field_work(void* field, oop newVal, bool release = false);
 303 public:
 304 
 305   bool has_write_ref_array_opt() { return true; }
 306   bool has_write_region_opt() { return true; }
 307 
 308   inline void inline_write_region(MemRegion mr) {
 309     dirty_MemRegion(mr);
 310   }
 311 protected:
 312   void write_region_work(MemRegion mr) {
 313     inline_write_region(mr);
 314   }
 315 public:
 316 
 317   inline void inline_write_ref_array(MemRegion mr) {
 318     dirty_MemRegion(mr);
 319   }
 320 protected:
 321   void write_ref_array_work(MemRegion mr) {
 322     inline_write_ref_array(mr);
 323   }
 324 public:
 325 
 326   bool is_aligned(HeapWord* addr) {
 327     return is_card_aligned(addr);
 328   }
 329 
 330   // *** Card-table-barrier-specific things.
 331 
 332   template <class T> inline void inline_write_ref_field_pre(T* field, oop newVal) {}
 333 
 334   template <class T> inline void inline_write_ref_field(T* field, oop newVal, bool release) {
 335     jbyte* byte = byte_for((void*)field);
 336     if (release) {
 337       // Perform a releasing store if requested.
 338       OrderAccess::release_store((volatile jbyte*) byte, dirty_card);
 339     } else {
 340       *byte = dirty_card;
 341     }
 342   }
 343 
 344   // These are used by G1, when it uses the card table as a temporary data
 345   // structure for card claiming.
 346   bool is_card_dirty(size_t card_index) {
 347     return _byte_map[card_index] == dirty_card_val();
 348   }
 349 
 350   void mark_card_dirty(size_t card_index) {
 351     _byte_map[card_index] = dirty_card_val();
 352   }
 353 
 354   bool is_card_clean(size_t card_index) {
 355     return _byte_map[card_index] == clean_card_val();
 356   }
 357 
 358   // Card marking array base (adjusted for heap low boundary)
 359   // This would be the 0th element of _byte_map, if the heap started at 0x0.
 360   // But since the heap starts at some higher address, this points to somewhere
 361   // before the beginning of the actual _byte_map.
 362   jbyte* byte_map_base;
 363 
 364   // Return true if "p" is at the start of a card.
 365   bool is_card_aligned(HeapWord* p) {
 366     jbyte* pcard = byte_for(p);
 367     return (addr_for(pcard) == p);
 368   }
 369 
 370   HeapWord* align_to_card_boundary(HeapWord* p) {
 371     jbyte* pcard = byte_for(p + card_size_in_words - 1);
 372     return addr_for(pcard);
 373   }
 374 
 375   // The kinds of precision a CardTableModRefBS may offer.
 376   enum PrecisionStyle {
 377     Precise,
 378     ObjHeadPreciseArray
 379   };
 380 
 381   // Tells what style of precision this card table offers.
 382   PrecisionStyle precision() {
 383     return ObjHeadPreciseArray; // Only one supported for now.
 384   }
 385 
 386   // ModRefBS functions.
 387   virtual void invalidate(MemRegion mr, bool whole_heap = false);
 388   void clear(MemRegion mr);
 389   void dirty(MemRegion mr);
 390 
 391   // *** Card-table-RemSet-specific things.
 392 
 393   // Invoke "cl.do_MemRegion" on a set of MemRegions that collectively
 394   // includes all the modified cards (expressing each card as a
 395   // MemRegion).  Thus, several modified cards may be lumped into one
 396   // region.  The regions are non-overlapping, and are visited in
 397   // *decreasing* address order.  (This order aids with imprecise card
 398   // marking, where a dirty card may cause scanning, and summarization
 399   // marking, of objects that extend onto subsequent cards.)
 400   void mod_card_iterate(MemRegionClosure* cl) {
 401     non_clean_card_iterate_serial(_whole_heap, cl);
 402   }
 403 
 404   // Like the "mod_cards_iterate" above, except only invokes the closure
 405   // for cards within the MemRegion "mr" (which is required to be
 406   // card-aligned and sized.)
 407   void mod_card_iterate(MemRegion mr, MemRegionClosure* cl) {
 408     non_clean_card_iterate_serial(mr, cl);
 409   }
 410 
 411   static uintx ct_max_alignment_constraint();
 412 
 413   // Apply closure "cl" to the dirty cards containing some part of
 414   // MemRegion "mr".
 415   void dirty_card_iterate(MemRegion mr, MemRegionClosure* cl);
 416 
 417   // Return the MemRegion corresponding to the first maximal run
 418   // of dirty cards lying completely within MemRegion mr.
 419   // If reset is "true", then sets those card table entries to the given
 420   // value.
 421   MemRegion dirty_card_range_after_reset(MemRegion mr, bool reset,
 422                                          int reset_val);
 423 
 424   // Provide read-only access to the card table array.
 425   const jbyte* byte_for_const(const void* p) const {
 426     return byte_for(p);
 427   }
 428   const jbyte* byte_after_const(const void* p) const {
 429     return byte_after(p);
 430   }
 431 
 432   // Mapping from card marking array entry to address of first word
 433   HeapWord* addr_for(const jbyte* p) const {
 434     assert(p >= _byte_map && p < _byte_map + _byte_map_size,
 435            "out of bounds access to card marking array");
 436     size_t delta = pointer_delta(p, byte_map_base, sizeof(jbyte));
 437     HeapWord* result = (HeapWord*) (delta << card_shift);
 438     assert(_whole_heap.contains(result),
 439            err_msg("Returning result = "PTR_FORMAT" out of bounds of "
 440                    " card marking array's _whole_heap = ["PTR_FORMAT","PTR_FORMAT")",
 441                    p2i(result), p2i(_whole_heap.start()), p2i(_whole_heap.end())));
 442     return result;
 443   }
 444 
 445   // Mapping from address to card marking array index.
 446   size_t index_for(void* p) {
 447     assert(_whole_heap.contains(p),
 448            err_msg("Attempt to access p = "PTR_FORMAT" out of bounds of "
 449                    " card marking array's _whole_heap = ["PTR_FORMAT","PTR_FORMAT")",
 450                    p2i(p), p2i(_whole_heap.start()), p2i(_whole_heap.end())));
 451     return byte_for(p) - _byte_map;
 452   }
 453 
 454   const jbyte* byte_for_index(const size_t card_index) const {
 455     return _byte_map + card_index;
 456   }
 457 
 458   // Print a description of the memory for the barrier set
 459   virtual void print_on(outputStream* st) const;
 460 
 461   void verify();
 462   void verify_guard();
 463 
 464   // val_equals -> it will check that all cards covered by mr equal val
 465   // !val_equals -> it will check that all cards covered by mr do not equal val
 466   void verify_region(MemRegion mr, jbyte val, bool val_equals) PRODUCT_RETURN;
 467   void verify_not_dirty_region(MemRegion mr) PRODUCT_RETURN;
 468   void verify_dirty_region(MemRegion mr) PRODUCT_RETURN;
 469 };
 470 
 471 class CardTableRS;
 472 
 473 // A specialization for the CardTableRS gen rem set.
 474 class CardTableModRefBSForCTRS: public CardTableModRefBS {
 475   CardTableRS* _rs;
 476 protected:
 477   bool card_will_be_scanned(jbyte cv);
 478   bool card_may_have_been_dirty(jbyte cv);
 479 public:
 480   CardTableModRefBSForCTRS(MemRegion whole_heap) :
 481     CardTableModRefBS(whole_heap) {}
 482 
 483   void set_CTRS(CardTableRS* rs) { _rs = rs; }
 484 };
 485 
 486 
 487 #endif // SHARE_VM_MEMORY_CARDTABLEMODREFBS_HPP