1 /* 2 * Copyright (c) 2000, 2013, Oracle and/or its affiliates. All rights reserved. 3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. 4 * 5 * This code is free software; you can redistribute it and/or modify it 6 * under the terms of the GNU General Public License version 2 only, as 7 * published by the Free Software Foundation. 8 * 9 * This code is distributed in the hope that it will be useful, but WITHOUT 10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or 11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License 12 * version 2 for more details (a copy is included in the LICENSE file that 13 * accompanied this code). 14 * 15 * You should have received a copy of the GNU General Public License version 16 * 2 along with this work; if not, write to the Free Software Foundation, 17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA. 18 * 19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA 20 * or visit www.oracle.com if you need additional information or have any 21 * questions. 22 * 23 */ 24 25 #ifndef SHARE_VM_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 g1_young_gen = 32 74 }; 75 76 // a word's worth (row) of clean card values 77 static const intptr_t clean_card_row = (intptr_t)(-1); 78 79 // dirty and precleaned are equivalent wrt younger_refs_iter. 80 static bool card_is_dirty_wrt_gen_iter(jbyte cv) { 81 return cv == dirty_card || cv == precleaned_card; 82 } 83 84 // Returns "true" iff the value "cv" will cause the card containing it 85 // to be scanned in the current traversal. May be overridden by 86 // subtypes. 87 virtual bool card_will_be_scanned(jbyte cv) { 88 return CardTableModRefBS::card_is_dirty_wrt_gen_iter(cv); 89 } 90 91 // Returns "true" iff the value "cv" may have represented a dirty card at 92 // some point. 93 virtual bool card_may_have_been_dirty(jbyte cv) { 94 return card_is_dirty_wrt_gen_iter(cv); 95 } 96 97 // The declaration order of these const fields is important; see the 98 // constructor before changing. 99 const MemRegion _whole_heap; // the region covered by the card table 100 const size_t _guard_index; // index of very last element in the card 101 // table; it is set to a guard value 102 // (last_card) and should never be modified 103 const size_t _last_valid_index; // index of the last valid element 104 const size_t _page_size; // page size used when mapping _byte_map 105 const size_t _byte_map_size; // in bytes 106 jbyte* _byte_map; // the card marking array 107 108 int _cur_covered_regions; 109 // The covered regions should be in address order. 110 MemRegion* _covered; 111 // The committed regions correspond one-to-one to the covered regions. 112 // They represent the card-table memory that has been committed to service 113 // the corresponding covered region. It may be that committed region for 114 // one covered region corresponds to a larger region because of page-size 115 // roundings. Thus, a committed region for one covered region may 116 // actually extend onto the card-table space for the next covered region. 117 MemRegion* _committed; 118 119 // The last card is a guard card, and we commit the page for it so 120 // we can use the card for verification purposes. We make sure we never 121 // uncommit the MemRegion for that page. 122 MemRegion _guard_region; 123 124 protected: 125 // Initialization utilities; covered_words is the size of the covered region 126 // in, um, words. 127 inline size_t cards_required(size_t covered_words); 128 inline size_t compute_byte_map_size(); 129 130 // Finds and return the index of the region, if any, to which the given 131 // region would be contiguous. If none exists, assign a new region and 132 // returns its index. Requires that no more than the maximum number of 133 // covered regions defined in the constructor are ever in use. 134 int find_covering_region_by_base(HeapWord* base); 135 136 // Same as above, but finds the region containing the given address 137 // instead of starting at a given base address. 138 int find_covering_region_containing(HeapWord* addr); 139 140 // Resize one of the regions covered by the remembered set. 141 void resize_covered_region(MemRegion new_region); 142 143 // Returns the leftmost end of a committed region corresponding to a 144 // covered region before covered region "ind", or else "NULL" if "ind" is 145 // the first covered region. 146 HeapWord* largest_prev_committed_end(int ind) const; 147 148 // Returns the part of the region mr that doesn't intersect with 149 // any committed region other than self. Used to prevent uncommitting 150 // regions that are also committed by other regions. Also protects 151 // against uncommitting the guard region. 152 MemRegion committed_unique_to_self(int self, MemRegion mr) const; 153 154 // Mapping from address to card marking array entry 155 jbyte* byte_for(const void* p) const { 156 assert(_whole_heap.contains(p), 157 err_msg("Attempt to access p = "PTR_FORMAT" out of bounds of " 158 " card marking array's _whole_heap = ["PTR_FORMAT","PTR_FORMAT")", 159 p, _whole_heap.start(), _whole_heap.end())); 160 jbyte* result = &byte_map_base[uintptr_t(p) >> card_shift]; 161 assert(result >= _byte_map && result < _byte_map + _byte_map_size, 162 "out of bounds accessor for card marking array"); 163 return result; 164 } 165 166 // The card table byte one after the card marking array 167 // entry for argument address. Typically used for higher bounds 168 // for loops iterating through the card table. 169 jbyte* byte_after(const void* p) const { 170 return byte_for(p) + 1; 171 } 172 173 // Iterate over the portion of the card-table which covers the given 174 // region mr in the given space and apply cl to any dirty sub-regions 175 // of mr. Dirty cards are _not_ cleared by the iterator method itself, 176 // but closures may arrange to do so on their own should they so wish. 177 void non_clean_card_iterate_serial(MemRegion mr, MemRegionClosure* cl); 178 179 // A variant of the above that will operate in a parallel mode if 180 // worker threads are available, and clear the dirty cards as it 181 // processes them. 182 // XXX ??? MemRegionClosure above vs OopsInGenClosure below XXX 183 // XXX some new_dcto_cl's take OopClosure's, plus as above there are 184 // some MemRegionClosures. Clean this up everywhere. XXX 185 void non_clean_card_iterate_possibly_parallel(Space* sp, MemRegion mr, 186 OopsInGenClosure* cl, CardTableRS* ct); 187 188 private: 189 // Work method used to implement non_clean_card_iterate_possibly_parallel() 190 // above in the parallel case. 191 void non_clean_card_iterate_parallel_work(Space* sp, MemRegion mr, 192 OopsInGenClosure* cl, CardTableRS* ct, 193 int n_threads); 194 195 protected: 196 // Dirty the bytes corresponding to "mr" (not all of which must be 197 // covered.) 198 void dirty_MemRegion(MemRegion mr); 199 200 // Clear (to clean_card) the bytes entirely contained within "mr" (not 201 // all of which must be covered.) 202 void clear_MemRegion(MemRegion mr); 203 204 // *** Support for parallel card scanning. 205 206 // This is an array, one element per covered region of the card table. 207 // Each entry is itself an array, with one element per chunk in the 208 // covered region. Each entry of these arrays is the lowest non-clean 209 // card of the corresponding chunk containing part of an object from the 210 // previous chunk, or else NULL. 211 typedef jbyte* CardPtr; 212 typedef CardPtr* CardArr; 213 CardArr* _lowest_non_clean; 214 size_t* _lowest_non_clean_chunk_size; 215 uintptr_t* _lowest_non_clean_base_chunk_index; 216 int* _last_LNC_resizing_collection; 217 218 // Initializes "lowest_non_clean" to point to the array for the region 219 // covering "sp", and "lowest_non_clean_base_chunk_index" to the chunk 220 // index of the corresponding to the first element of that array. 221 // Ensures that these arrays are of sufficient size, allocating if necessary. 222 // May be called by several threads concurrently. 223 void get_LNC_array_for_space(Space* sp, 224 jbyte**& lowest_non_clean, 225 uintptr_t& lowest_non_clean_base_chunk_index, 226 size_t& lowest_non_clean_chunk_size); 227 228 // Returns the number of chunks necessary to cover "mr". 229 size_t chunks_to_cover(MemRegion mr) { 230 return (size_t)(addr_to_chunk_index(mr.last()) - 231 addr_to_chunk_index(mr.start()) + 1); 232 } 233 234 // Returns the index of the chunk in a stride which 235 // covers the given address. 236 uintptr_t addr_to_chunk_index(const void* addr) { 237 uintptr_t card = (uintptr_t) byte_for(addr); 238 return card / ParGCCardsPerStrideChunk; 239 } 240 241 // Apply cl, which must either itself apply dcto_cl or be dcto_cl, 242 // to the cards in the stride (of n_strides) within the given space. 243 void process_stride(Space* sp, 244 MemRegion used, 245 jint stride, int n_strides, 246 OopsInGenClosure* cl, 247 CardTableRS* ct, 248 jbyte** lowest_non_clean, 249 uintptr_t lowest_non_clean_base_chunk_index, 250 size_t lowest_non_clean_chunk_size); 251 252 // Makes sure that chunk boundaries are handled appropriately, by 253 // adjusting the min_done of dcto_cl, and by using a special card-table 254 // value to indicate how min_done should be set. 255 void process_chunk_boundaries(Space* sp, 256 DirtyCardToOopClosure* dcto_cl, 257 MemRegion chunk_mr, 258 MemRegion used, 259 jbyte** lowest_non_clean, 260 uintptr_t lowest_non_clean_base_chunk_index, 261 size_t lowest_non_clean_chunk_size); 262 263 public: 264 // Constants 265 enum SomePublicConstants { 266 card_shift = 9, 267 card_size = 1 << card_shift, 268 card_size_in_words = card_size / sizeof(HeapWord) 269 }; 270 271 static int clean_card_val() { return clean_card; } 272 static int clean_card_mask_val() { return clean_card_mask; } 273 static int dirty_card_val() { return dirty_card; } 274 static int claimed_card_val() { return claimed_card; } 275 static int precleaned_card_val() { return precleaned_card; } 276 static int deferred_card_val() { return deferred_card; } 277 static int g1_young_card_val() { return g1_young_gen; } 278 279 // For RTTI simulation. 280 bool is_a(BarrierSet::Name bsn) { 281 return bsn == BarrierSet::CardTableModRef || ModRefBarrierSet::is_a(bsn); 282 } 283 284 CardTableModRefBS(MemRegion whole_heap, int max_covered_regions); 285 ~CardTableModRefBS(); 286 287 // *** Barrier set functions. 288 289 bool has_write_ref_pre_barrier() { return false; } 290 291 // Record a reference update. Note that these versions are precise! 292 // The scanning code has to handle the fact that the write barrier may be 293 // either precise or imprecise. We make non-virtual inline variants of 294 // these functions here for performance. 295 protected: 296 void write_ref_field_work(oop obj, size_t offset, oop newVal); 297 virtual void write_ref_field_work(void* field, oop newVal); 298 public: 299 300 bool has_write_ref_array_opt() { return true; } 301 bool has_write_region_opt() { return true; } 302 303 inline void inline_write_region(MemRegion mr) { 304 dirty_MemRegion(mr); 305 } 306 protected: 307 void write_region_work(MemRegion mr) { 308 inline_write_region(mr); 309 } 310 public: 311 312 inline void inline_write_ref_array(MemRegion mr) { 313 dirty_MemRegion(mr); 314 } 315 protected: 316 void write_ref_array_work(MemRegion mr) { 317 inline_write_ref_array(mr); 318 } 319 public: 320 321 bool is_aligned(HeapWord* addr) { 322 return is_card_aligned(addr); 323 } 324 325 // *** Card-table-barrier-specific things. 326 327 template <class T> inline void inline_write_ref_field_pre(T* field, oop newVal) {} 328 329 template <class T> inline void inline_write_ref_field(T* field, oop newVal) { 330 jbyte* byte = byte_for((void*)field); 331 *byte = dirty_card; 332 } 333 334 // These are used by G1, when it uses the card table as a temporary data 335 // structure for card claiming. 336 bool is_card_dirty(size_t card_index) { 337 return _byte_map[card_index] == dirty_card_val(); 338 } 339 340 void mark_card_dirty(size_t card_index) { 341 _byte_map[card_index] = dirty_card_val(); 342 } 343 344 bool is_card_claimed(size_t card_index) { 345 jbyte val = _byte_map[card_index]; 346 return (val & (clean_card_mask_val() | claimed_card_val())) == claimed_card_val(); 347 } 348 349 void set_card_claimed(size_t card_index) { 350 jbyte val = _byte_map[card_index]; 351 if (val == clean_card_val()) { 352 val = (jbyte)claimed_card_val(); 353 } else { 354 val |= (jbyte)claimed_card_val(); 355 } 356 _byte_map[card_index] = val; 357 } 358 359 bool claim_card(size_t card_index); 360 361 bool is_card_clean(size_t card_index) { 362 return _byte_map[card_index] == clean_card_val(); 363 } 364 365 bool is_card_deferred(size_t card_index) { 366 jbyte val = _byte_map[card_index]; 367 return (val & (clean_card_mask_val() | deferred_card_val())) == deferred_card_val(); 368 } 369 370 bool mark_card_deferred(size_t card_index); 371 372 // Card marking array base (adjusted for heap low boundary) 373 // This would be the 0th element of _byte_map, if the heap started at 0x0. 374 // But since the heap starts at some higher address, this points to somewhere 375 // before the beginning of the actual _byte_map. 376 jbyte* byte_map_base; 377 378 // Return true if "p" is at the start of a card. 379 bool is_card_aligned(HeapWord* p) { 380 jbyte* pcard = byte_for(p); 381 return (addr_for(pcard) == p); 382 } 383 384 HeapWord* align_to_card_boundary(HeapWord* p) { 385 jbyte* pcard = byte_for(p + card_size_in_words - 1); 386 return addr_for(pcard); 387 } 388 389 // The kinds of precision a CardTableModRefBS may offer. 390 enum PrecisionStyle { 391 Precise, 392 ObjHeadPreciseArray 393 }; 394 395 // Tells what style of precision this card table offers. 396 PrecisionStyle precision() { 397 return ObjHeadPreciseArray; // Only one supported for now. 398 } 399 400 // ModRefBS functions. 401 virtual void invalidate(MemRegion mr, bool whole_heap = false); 402 void clear(MemRegion mr); 403 void dirty(MemRegion mr); 404 405 // *** Card-table-RemSet-specific things. 406 407 // Invoke "cl.do_MemRegion" on a set of MemRegions that collectively 408 // includes all the modified cards (expressing each card as a 409 // MemRegion). Thus, several modified cards may be lumped into one 410 // region. The regions are non-overlapping, and are visited in 411 // *decreasing* address order. (This order aids with imprecise card 412 // marking, where a dirty card may cause scanning, and summarization 413 // marking, of objects that extend onto subsequent cards.) 414 void mod_card_iterate(MemRegionClosure* cl) { 415 non_clean_card_iterate_serial(_whole_heap, cl); 416 } 417 418 // Like the "mod_cards_iterate" above, except only invokes the closure 419 // for cards within the MemRegion "mr" (which is required to be 420 // card-aligned and sized.) 421 void mod_card_iterate(MemRegion mr, MemRegionClosure* cl) { 422 non_clean_card_iterate_serial(mr, cl); 423 } 424 425 static uintx ct_max_alignment_constraint(); 426 427 // Apply closure "cl" to the dirty cards containing some part of 428 // MemRegion "mr". 429 void dirty_card_iterate(MemRegion mr, MemRegionClosure* cl); 430 431 // Return the MemRegion corresponding to the first maximal run 432 // of dirty cards lying completely within MemRegion mr. 433 // If reset is "true", then sets those card table entries to the given 434 // value. 435 MemRegion dirty_card_range_after_reset(MemRegion mr, bool reset, 436 int reset_val); 437 438 // Provide read-only access to the card table array. 439 const jbyte* byte_for_const(const void* p) const { 440 return byte_for(p); 441 } 442 const jbyte* byte_after_const(const void* p) const { 443 return byte_after(p); 444 } 445 446 // Mapping from card marking array entry to address of first word 447 HeapWord* addr_for(const jbyte* p) const { 448 assert(p >= _byte_map && p < _byte_map + _byte_map_size, 449 "out of bounds access to card marking array"); 450 size_t delta = pointer_delta(p, byte_map_base, sizeof(jbyte)); 451 HeapWord* result = (HeapWord*) (delta << card_shift); 452 assert(_whole_heap.contains(result), 453 err_msg("Returning result = "PTR_FORMAT" out of bounds of " 454 " card marking array's _whole_heap = ["PTR_FORMAT","PTR_FORMAT")", 455 result, _whole_heap.start(), _whole_heap.end())); 456 return result; 457 } 458 459 // Mapping from address to card marking array index. 460 size_t index_for(void* p) { 461 assert(_whole_heap.contains(p), 462 err_msg("Attempt to access p = "PTR_FORMAT" out of bounds of " 463 " card marking array's _whole_heap = ["PTR_FORMAT","PTR_FORMAT")", 464 p, _whole_heap.start(), _whole_heap.end())); 465 return byte_for(p) - _byte_map; 466 } 467 468 const jbyte* byte_for_index(const size_t card_index) const { 469 return _byte_map + card_index; 470 } 471 472 // Print a description of the memory for the barrier set 473 virtual void print_on(outputStream* st) const; 474 475 void verify(); 476 void verify_guard(); 477 478 // val_equals -> it will check that all cards covered by mr equal val 479 // !val_equals -> it will check that all cards covered by mr do not equal val 480 void verify_region(MemRegion mr, jbyte val, bool val_equals) PRODUCT_RETURN; 481 void verify_not_dirty_region(MemRegion mr) PRODUCT_RETURN; 482 void verify_dirty_region(MemRegion mr) PRODUCT_RETURN; 483 void verify_g1_young_region(MemRegion mr) PRODUCT_RETURN; 484 485 static size_t par_chunk_heapword_alignment() { 486 return ParGCCardsPerStrideChunk * card_size_in_words; 487 } 488 489 }; 490 491 class CardTableRS; 492 493 // A specialization for the CardTableRS gen rem set. 494 class CardTableModRefBSForCTRS: public CardTableModRefBS { 495 CardTableRS* _rs; 496 protected: 497 bool card_will_be_scanned(jbyte cv); 498 bool card_may_have_been_dirty(jbyte cv); 499 public: 500 CardTableModRefBSForCTRS(MemRegion whole_heap, 501 int max_covered_regions) : 502 CardTableModRefBS(whole_heap, max_covered_regions) {} 503 504 void set_CTRS(CardTableRS* rs) { _rs = rs; } 505 }; 506 507 508 #endif // SHARE_VM_MEMORY_CARDTABLEMODREFBS_HPP