1 /* 2 * Copyright (c) 2001, 2018, 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 #include "precompiled.hpp" 26 #include "classfile/classLoaderDataGraph.hpp" 27 #include "gc/shared/cardTableRS.hpp" 28 #include "gc/shared/genCollectedHeap.hpp" 29 #include "gc/shared/genOopClosures.hpp" 30 #include "gc/shared/generation.hpp" 31 #include "gc/shared/space.inline.hpp" 32 #include "memory/allocation.inline.hpp" 33 #include "memory/iterator.inline.hpp" 34 #include "oops/access.inline.hpp" 35 #include "oops/oop.inline.hpp" 36 #include "runtime/atomic.hpp" 37 #include "runtime/java.hpp" 38 #include "runtime/os.hpp" 39 #include "utilities/macros.hpp" 40 41 class HasAccumulatedModifiedOopsClosure : public CLDClosure { 42 bool _found; 43 public: 44 HasAccumulatedModifiedOopsClosure() : _found(false) {} 45 void do_cld(ClassLoaderData* cld) { 46 if (_found) { 47 return; 48 } 49 50 if (cld->has_accumulated_modified_oops()) { 51 _found = true; 52 } 53 } 54 bool found() { 55 return _found; 56 } 57 }; 58 59 bool CLDRemSet::mod_union_is_clear() { 60 HasAccumulatedModifiedOopsClosure closure; 61 ClassLoaderDataGraph::cld_do(&closure); 62 63 return !closure.found(); 64 } 65 66 67 class ClearCLDModUnionClosure : public CLDClosure { 68 public: 69 void do_cld(ClassLoaderData* cld) { 70 if (cld->has_accumulated_modified_oops()) { 71 cld->clear_accumulated_modified_oops(); 72 } 73 } 74 }; 75 76 void CLDRemSet::clear_mod_union() { 77 ClearCLDModUnionClosure closure; 78 ClassLoaderDataGraph::cld_do(&closure); 79 } 80 81 82 jbyte CardTableRS::find_unused_youngergenP_card_value() { 83 for (jbyte v = youngergenP1_card; 84 v < cur_youngergen_and_prev_nonclean_card; 85 v++) { 86 bool seen = false; 87 for (int g = 0; g < _regions_to_iterate; g++) { 88 if (_last_cur_val_in_gen[g] == v) { 89 seen = true; 90 break; 91 } 92 } 93 if (!seen) { 94 return v; 95 } 96 } 97 ShouldNotReachHere(); 98 return 0; 99 } 100 101 void CardTableRS::prepare_for_younger_refs_iterate(bool parallel) { 102 // Parallel or sequential, we must always set the prev to equal the 103 // last one written. 104 if (parallel) { 105 // Find a parallel value to be used next. 106 jbyte next_val = find_unused_youngergenP_card_value(); 107 set_cur_youngergen_card_val(next_val); 108 109 } else { 110 // In an sequential traversal we will always write youngergen, so that 111 // the inline barrier is correct. 112 set_cur_youngergen_card_val(youngergen_card); 113 } 114 } 115 116 void CardTableRS::younger_refs_iterate(Generation* g, 117 OopsInGenClosure* blk, 118 uint n_threads) { 119 // The indexing in this array is slightly odd. We want to access 120 // the old generation record here, which is at index 2. 121 _last_cur_val_in_gen[2] = cur_youngergen_card_val(); 122 g->younger_refs_iterate(blk, n_threads); 123 } 124 125 inline bool ClearNoncleanCardWrapper::clear_card(jbyte* entry) { 126 if (_is_par) { 127 return clear_card_parallel(entry); 128 } else { 129 return clear_card_serial(entry); 130 } 131 } 132 133 inline bool ClearNoncleanCardWrapper::clear_card_parallel(jbyte* entry) { 134 while (true) { 135 // In the parallel case, we may have to do this several times. 136 jbyte entry_val = *entry; 137 assert(entry_val != CardTableRS::clean_card_val(), 138 "We shouldn't be looking at clean cards, and this should " 139 "be the only place they get cleaned."); 140 if (CardTableRS::card_is_dirty_wrt_gen_iter(entry_val) 141 || _ct->is_prev_youngergen_card_val(entry_val)) { 142 jbyte res = 143 Atomic::cmpxchg(CardTableRS::clean_card_val(), entry, entry_val); 144 if (res == entry_val) { 145 break; 146 } else { 147 assert(res == CardTableRS::cur_youngergen_and_prev_nonclean_card, 148 "The CAS above should only fail if another thread did " 149 "a GC write barrier."); 150 } 151 } else if (entry_val == 152 CardTableRS::cur_youngergen_and_prev_nonclean_card) { 153 // Parallelism shouldn't matter in this case. Only the thread 154 // assigned to scan the card should change this value. 155 *entry = _ct->cur_youngergen_card_val(); 156 break; 157 } else { 158 assert(entry_val == _ct->cur_youngergen_card_val(), 159 "Should be the only possibility."); 160 // In this case, the card was clean before, and become 161 // cur_youngergen only because of processing of a promoted object. 162 // We don't have to look at the card. 163 return false; 164 } 165 } 166 return true; 167 } 168 169 170 inline bool ClearNoncleanCardWrapper::clear_card_serial(jbyte* entry) { 171 jbyte entry_val = *entry; 172 assert(entry_val != CardTableRS::clean_card_val(), 173 "We shouldn't be looking at clean cards, and this should " 174 "be the only place they get cleaned."); 175 assert(entry_val != CardTableRS::cur_youngergen_and_prev_nonclean_card, 176 "This should be possible in the sequential case."); 177 *entry = CardTableRS::clean_card_val(); 178 return true; 179 } 180 181 ClearNoncleanCardWrapper::ClearNoncleanCardWrapper( 182 DirtyCardToOopClosure* dirty_card_closure, CardTableRS* ct, bool is_par) : 183 _dirty_card_closure(dirty_card_closure), _ct(ct), _is_par(is_par) { 184 } 185 186 bool ClearNoncleanCardWrapper::is_word_aligned(jbyte* entry) { 187 return (((intptr_t)entry) & (BytesPerWord-1)) == 0; 188 } 189 190 // The regions are visited in *decreasing* address order. 191 // This order aids with imprecise card marking, where a dirty 192 // card may cause scanning, and summarization marking, of objects 193 // that extend onto subsequent cards. 194 void ClearNoncleanCardWrapper::do_MemRegion(MemRegion mr) { 195 assert(mr.word_size() > 0, "Error"); 196 assert(_ct->is_aligned(mr.start()), "mr.start() should be card aligned"); 197 // mr.end() may not necessarily be card aligned. 198 jbyte* cur_entry = _ct->byte_for(mr.last()); 199 const jbyte* limit = _ct->byte_for(mr.start()); 200 HeapWord* end_of_non_clean = mr.end(); 201 HeapWord* start_of_non_clean = end_of_non_clean; 202 while (cur_entry >= limit) { 203 HeapWord* cur_hw = _ct->addr_for(cur_entry); 204 if ((*cur_entry != CardTableRS::clean_card_val()) && clear_card(cur_entry)) { 205 // Continue the dirty range by opening the 206 // dirty window one card to the left. 207 start_of_non_clean = cur_hw; 208 } else { 209 // We hit a "clean" card; process any non-empty 210 // "dirty" range accumulated so far. 211 if (start_of_non_clean < end_of_non_clean) { 212 const MemRegion mrd(start_of_non_clean, end_of_non_clean); 213 _dirty_card_closure->do_MemRegion(mrd); 214 } 215 216 // fast forward through potential continuous whole-word range of clean cards beginning at a word-boundary 217 if (is_word_aligned(cur_entry)) { 218 jbyte* cur_row = cur_entry - BytesPerWord; 219 while (cur_row >= limit && *((intptr_t*)cur_row) == CardTableRS::clean_card_row_val()) { 220 cur_row -= BytesPerWord; 221 } 222 cur_entry = cur_row + BytesPerWord; 223 cur_hw = _ct->addr_for(cur_entry); 224 } 225 226 // Reset the dirty window, while continuing to look 227 // for the next dirty card that will start a 228 // new dirty window. 229 end_of_non_clean = cur_hw; 230 start_of_non_clean = cur_hw; 231 } 232 // Note that "cur_entry" leads "start_of_non_clean" in 233 // its leftward excursion after this point 234 // in the loop and, when we hit the left end of "mr", 235 // will point off of the left end of the card-table 236 // for "mr". 237 cur_entry--; 238 } 239 // If the first card of "mr" was dirty, we will have 240 // been left with a dirty window, co-initial with "mr", 241 // which we now process. 242 if (start_of_non_clean < end_of_non_clean) { 243 const MemRegion mrd(start_of_non_clean, end_of_non_clean); 244 _dirty_card_closure->do_MemRegion(mrd); 245 } 246 } 247 248 // clean (by dirty->clean before) ==> cur_younger_gen 249 // dirty ==> cur_youngergen_and_prev_nonclean_card 250 // precleaned ==> cur_youngergen_and_prev_nonclean_card 251 // prev-younger-gen ==> cur_youngergen_and_prev_nonclean_card 252 // cur-younger-gen ==> cur_younger_gen 253 // cur_youngergen_and_prev_nonclean_card ==> no change. 254 void CardTableRS::write_ref_field_gc_par(void* field, oop new_val) { 255 volatile jbyte* entry = byte_for(field); 256 do { 257 jbyte entry_val = *entry; 258 // We put this first because it's probably the most common case. 259 if (entry_val == clean_card_val()) { 260 // No threat of contention with cleaning threads. 261 *entry = cur_youngergen_card_val(); 262 return; 263 } else if (card_is_dirty_wrt_gen_iter(entry_val) 264 || is_prev_youngergen_card_val(entry_val)) { 265 // Mark it as both cur and prev youngergen; card cleaning thread will 266 // eventually remove the previous stuff. 267 jbyte new_val = cur_youngergen_and_prev_nonclean_card; 268 jbyte res = Atomic::cmpxchg(new_val, entry, entry_val); 269 // Did the CAS succeed? 270 if (res == entry_val) return; 271 // Otherwise, retry, to see the new value. 272 continue; 273 } else { 274 assert(entry_val == cur_youngergen_and_prev_nonclean_card 275 || entry_val == cur_youngergen_card_val(), 276 "should be only possibilities."); 277 return; 278 } 279 } while (true); 280 } 281 282 void CardTableRS::younger_refs_in_space_iterate(Space* sp, 283 OopsInGenClosure* cl, 284 uint n_threads) { 285 verify_used_region_at_save_marks(sp); 286 287 const MemRegion urasm = sp->used_region_at_save_marks(); 288 non_clean_card_iterate_possibly_parallel(sp, urasm, cl, this, n_threads); 289 } 290 291 #ifdef ASSERT 292 void CardTableRS::verify_used_region_at_save_marks(Space* sp) const { 293 MemRegion ur = sp->used_region(); 294 MemRegion urasm = sp->used_region_at_save_marks(); 295 296 assert(ur.contains(urasm), 297 "Did you forget to call save_marks()? " 298 "[" PTR_FORMAT ", " PTR_FORMAT ") is not contained in " 299 "[" PTR_FORMAT ", " PTR_FORMAT ")", 300 p2i(urasm.start()), p2i(urasm.end()), p2i(ur.start()), p2i(ur.end())); 301 } 302 #endif 303 304 void CardTableRS::clear_into_younger(Generation* old_gen) { 305 assert(GenCollectedHeap::heap()->is_old_gen(old_gen), 306 "Should only be called for the old generation"); 307 // The card tables for the youngest gen need never be cleared. 308 // There's a bit of subtlety in the clear() and invalidate() 309 // methods that we exploit here and in invalidate_or_clear() 310 // below to avoid missing cards at the fringes. If clear() or 311 // invalidate() are changed in the future, this code should 312 // be revisited. 20040107.ysr 313 clear(old_gen->prev_used_region()); 314 } 315 316 void CardTableRS::invalidate_or_clear(Generation* old_gen) { 317 assert(GenCollectedHeap::heap()->is_old_gen(old_gen), 318 "Should only be called for the old generation"); 319 // Invalidate the cards for the currently occupied part of 320 // the old generation and clear the cards for the 321 // unoccupied part of the generation (if any, making use 322 // of that generation's prev_used_region to determine that 323 // region). No need to do anything for the youngest 324 // generation. Also see note#20040107.ysr above. 325 MemRegion used_mr = old_gen->used_region(); 326 MemRegion to_be_cleared_mr = old_gen->prev_used_region().minus(used_mr); 327 if (!to_be_cleared_mr.is_empty()) { 328 clear(to_be_cleared_mr); 329 } 330 invalidate(used_mr); 331 } 332 333 334 class VerifyCleanCardClosure: public BasicOopIterateClosure { 335 private: 336 HeapWord* _boundary; 337 HeapWord* _begin; 338 HeapWord* _end; 339 protected: 340 template <class T> void do_oop_work(T* p) { 341 HeapWord* jp = (HeapWord*)p; 342 assert(jp >= _begin && jp < _end, 343 "Error: jp " PTR_FORMAT " should be within " 344 "[_begin, _end) = [" PTR_FORMAT "," PTR_FORMAT ")", 345 p2i(jp), p2i(_begin), p2i(_end)); 346 oop obj = RawAccess<>::oop_load(p); 347 guarantee(obj == NULL || (HeapWord*)obj >= _boundary, 348 "pointer " PTR_FORMAT " at " PTR_FORMAT " on " 349 "clean card crosses boundary" PTR_FORMAT, 350 p2i(obj), p2i(jp), p2i(_boundary)); 351 } 352 353 public: 354 VerifyCleanCardClosure(HeapWord* b, HeapWord* begin, HeapWord* end) : 355 _boundary(b), _begin(begin), _end(end) { 356 assert(b <= begin, 357 "Error: boundary " PTR_FORMAT " should be at or below begin " PTR_FORMAT, 358 p2i(b), p2i(begin)); 359 assert(begin <= end, 360 "Error: begin " PTR_FORMAT " should be strictly below end " PTR_FORMAT, 361 p2i(begin), p2i(end)); 362 } 363 364 virtual void do_oop(oop* p) { VerifyCleanCardClosure::do_oop_work(p); } 365 virtual void do_oop(narrowOop* p) { VerifyCleanCardClosure::do_oop_work(p); } 366 }; 367 368 class VerifyCTSpaceClosure: public SpaceClosure { 369 private: 370 CardTableRS* _ct; 371 HeapWord* _boundary; 372 public: 373 VerifyCTSpaceClosure(CardTableRS* ct, HeapWord* boundary) : 374 _ct(ct), _boundary(boundary) {} 375 virtual void do_space(Space* s) { _ct->verify_space(s, _boundary); } 376 }; 377 378 class VerifyCTGenClosure: public GenCollectedHeap::GenClosure { 379 CardTableRS* _ct; 380 public: 381 VerifyCTGenClosure(CardTableRS* ct) : _ct(ct) {} 382 void do_generation(Generation* gen) { 383 // Skip the youngest generation. 384 if (GenCollectedHeap::heap()->is_young_gen(gen)) { 385 return; 386 } 387 // Normally, we're interested in pointers to younger generations. 388 VerifyCTSpaceClosure blk(_ct, gen->reserved().start()); 389 gen->space_iterate(&blk, true); 390 } 391 }; 392 393 void CardTableRS::verify_space(Space* s, HeapWord* gen_boundary) { 394 // We don't need to do young-gen spaces. 395 if (s->end() <= gen_boundary) return; 396 MemRegion used = s->used_region(); 397 398 jbyte* cur_entry = byte_for(used.start()); 399 jbyte* limit = byte_after(used.last()); 400 while (cur_entry < limit) { 401 if (*cur_entry == clean_card_val()) { 402 jbyte* first_dirty = cur_entry+1; 403 while (first_dirty < limit && 404 *first_dirty == clean_card_val()) { 405 first_dirty++; 406 } 407 // If the first object is a regular object, and it has a 408 // young-to-old field, that would mark the previous card. 409 HeapWord* boundary = addr_for(cur_entry); 410 HeapWord* end = (first_dirty >= limit) ? used.end() : addr_for(first_dirty); 411 HeapWord* boundary_block = s->block_start(boundary); 412 HeapWord* begin = boundary; // Until proven otherwise. 413 HeapWord* start_block = boundary_block; // Until proven otherwise. 414 if (boundary_block < boundary) { 415 if (s->block_is_obj(boundary_block) && s->obj_is_alive(boundary_block)) { 416 oop boundary_obj = oop(boundary_block); 417 if (!boundary_obj->is_objArray() && 418 !boundary_obj->is_typeArray()) { 419 guarantee(cur_entry > byte_for(used.start()), 420 "else boundary would be boundary_block"); 421 if (*byte_for(boundary_block) != clean_card_val()) { 422 begin = boundary_block + s->block_size(boundary_block); 423 start_block = begin; 424 } 425 } 426 } 427 } 428 // Now traverse objects until end. 429 if (begin < end) { 430 MemRegion mr(begin, end); 431 VerifyCleanCardClosure verify_blk(gen_boundary, begin, end); 432 for (HeapWord* cur = start_block; cur < end; cur += s->block_size(cur)) { 433 if (s->block_is_obj(cur) && s->obj_is_alive(cur)) { 434 oop(cur)->oop_iterate(&verify_blk, mr); 435 } 436 } 437 } 438 cur_entry = first_dirty; 439 } else { 440 // We'd normally expect that cur_youngergen_and_prev_nonclean_card 441 // is a transient value, that cannot be in the card table 442 // except during GC, and thus assert that: 443 // guarantee(*cur_entry != cur_youngergen_and_prev_nonclean_card, 444 // "Illegal CT value"); 445 // That however, need not hold, as will become clear in the 446 // following... 447 448 // We'd normally expect that if we are in the parallel case, 449 // we can't have left a prev value (which would be different 450 // from the current value) in the card table, and so we'd like to 451 // assert that: 452 // guarantee(cur_youngergen_card_val() == youngergen_card 453 // || !is_prev_youngergen_card_val(*cur_entry), 454 // "Illegal CT value"); 455 // That, however, may not hold occasionally, because of 456 // CMS or MSC in the old gen. To wit, consider the 457 // following two simple illustrative scenarios: 458 // (a) CMS: Consider the case where a large object L 459 // spanning several cards is allocated in the old 460 // gen, and has a young gen reference stored in it, dirtying 461 // some interior cards. A young collection scans the card, 462 // finds a young ref and installs a youngergenP_n value. 463 // L then goes dead. Now a CMS collection starts, 464 // finds L dead and sweeps it up. Assume that L is 465 // abutting _unallocated_blk, so _unallocated_blk is 466 // adjusted down to (below) L. Assume further that 467 // no young collection intervenes during this CMS cycle. 468 // The next young gen cycle will not get to look at this 469 // youngergenP_n card since it lies in the unoccupied 470 // part of the space. 471 // Some young collections later the blocks on this 472 // card can be re-allocated either due to direct allocation 473 // or due to absorbing promotions. At this time, the 474 // before-gc verification will fail the above assert. 475 // (b) MSC: In this case, an object L with a young reference 476 // is on a card that (therefore) holds a youngergen_n value. 477 // Suppose also that L lies towards the end of the used 478 // the used space before GC. An MSC collection 479 // occurs that compacts to such an extent that this 480 // card is no longer in the occupied part of the space. 481 // Since current code in MSC does not always clear cards 482 // in the unused part of old gen, this stale youngergen_n 483 // value is left behind and can later be covered by 484 // an object when promotion or direct allocation 485 // re-allocates that part of the heap. 486 // 487 // Fortunately, the presence of such stale card values is 488 // "only" a minor annoyance in that subsequent young collections 489 // might needlessly scan such cards, but would still never corrupt 490 // the heap as a result. However, it's likely not to be a significant 491 // performance inhibitor in practice. For instance, 492 // some recent measurements with unoccupied cards eagerly cleared 493 // out to maintain this invariant, showed next to no 494 // change in young collection times; of course one can construct 495 // degenerate examples where the cost can be significant.) 496 // Note, in particular, that if the "stale" card is modified 497 // after re-allocation, it would be dirty, not "stale". Thus, 498 // we can never have a younger ref in such a card and it is 499 // safe not to scan that card in any collection. [As we see 500 // below, we do some unnecessary scanning 501 // in some cases in the current parallel scanning algorithm.] 502 // 503 // The main point below is that the parallel card scanning code 504 // deals correctly with these stale card values. There are two main 505 // cases to consider where we have a stale "young gen" value and a 506 // "derivative" case to consider, where we have a stale 507 // "cur_younger_gen_and_prev_non_clean" value, as will become 508 // apparent in the case analysis below. 509 // o Case 1. If the stale value corresponds to a younger_gen_n 510 // value other than the cur_younger_gen value then the code 511 // treats this as being tantamount to a prev_younger_gen 512 // card. This means that the card may be unnecessarily scanned. 513 // There are two sub-cases to consider: 514 // o Case 1a. Let us say that the card is in the occupied part 515 // of the generation at the time the collection begins. In 516 // that case the card will be either cleared when it is scanned 517 // for young pointers, or will be set to cur_younger_gen as a 518 // result of promotion. (We have elided the normal case where 519 // the scanning thread and the promoting thread interleave 520 // possibly resulting in a transient 521 // cur_younger_gen_and_prev_non_clean value before settling 522 // to cur_younger_gen. [End Case 1a.] 523 // o Case 1b. Consider now the case when the card is in the unoccupied 524 // part of the space which becomes occupied because of promotions 525 // into it during the current young GC. In this case the card 526 // will never be scanned for young references. The current 527 // code will set the card value to either 528 // cur_younger_gen_and_prev_non_clean or leave 529 // it with its stale value -- because the promotions didn't 530 // result in any younger refs on that card. Of these two 531 // cases, the latter will be covered in Case 1a during 532 // a subsequent scan. To deal with the former case, we need 533 // to further consider how we deal with a stale value of 534 // cur_younger_gen_and_prev_non_clean in our case analysis 535 // below. This we do in Case 3 below. [End Case 1b] 536 // [End Case 1] 537 // o Case 2. If the stale value corresponds to cur_younger_gen being 538 // a value not necessarily written by a current promotion, the 539 // card will not be scanned by the younger refs scanning code. 540 // (This is OK since as we argued above such cards cannot contain 541 // any younger refs.) The result is that this value will be 542 // treated as a prev_younger_gen value in a subsequent collection, 543 // which is addressed in Case 1 above. [End Case 2] 544 // o Case 3. We here consider the "derivative" case from Case 1b. above 545 // because of which we may find a stale 546 // cur_younger_gen_and_prev_non_clean card value in the table. 547 // Once again, as in Case 1, we consider two subcases, depending 548 // on whether the card lies in the occupied or unoccupied part 549 // of the space at the start of the young collection. 550 // o Case 3a. Let us say the card is in the occupied part of 551 // the old gen at the start of the young collection. In that 552 // case, the card will be scanned by the younger refs scanning 553 // code which will set it to cur_younger_gen. In a subsequent 554 // scan, the card will be considered again and get its final 555 // correct value. [End Case 3a] 556 // o Case 3b. Now consider the case where the card is in the 557 // unoccupied part of the old gen, and is occupied as a result 558 // of promotions during thus young gc. In that case, 559 // the card will not be scanned for younger refs. The presence 560 // of newly promoted objects on the card will then result in 561 // its keeping the value cur_younger_gen_and_prev_non_clean 562 // value, which we have dealt with in Case 3 here. [End Case 3b] 563 // [End Case 3] 564 // 565 // (Please refer to the code in the helper class 566 // ClearNonCleanCardWrapper and in CardTable for details.) 567 // 568 // The informal arguments above can be tightened into a formal 569 // correctness proof and it behooves us to write up such a proof, 570 // or to use model checking to prove that there are no lingering 571 // concerns. 572 // 573 // Clearly because of Case 3b one cannot bound the time for 574 // which a card will retain what we have called a "stale" value. 575 // However, one can obtain a Loose upper bound on the redundant 576 // work as a result of such stale values. Note first that any 577 // time a stale card lies in the occupied part of the space at 578 // the start of the collection, it is scanned by younger refs 579 // code and we can define a rank function on card values that 580 // declines when this is so. Note also that when a card does not 581 // lie in the occupied part of the space at the beginning of a 582 // young collection, its rank can either decline or stay unchanged. 583 // In this case, no extra work is done in terms of redundant 584 // younger refs scanning of that card. 585 // Then, the case analysis above reveals that, in the worst case, 586 // any such stale card will be scanned unnecessarily at most twice. 587 // 588 // It is nonetheless advisable to try and get rid of some of this 589 // redundant work in a subsequent (low priority) re-design of 590 // the card-scanning code, if only to simplify the underlying 591 // state machine analysis/proof. ysr 1/28/2002. XXX 592 cur_entry++; 593 } 594 } 595 } 596 597 void CardTableRS::verify() { 598 // At present, we only know how to verify the card table RS for 599 // generational heaps. 600 VerifyCTGenClosure blk(this); 601 GenCollectedHeap::heap()->generation_iterate(&blk, false); 602 CardTable::verify(); 603 } 604 605 CardTableRS::CardTableRS(MemRegion whole_heap, bool scanned_concurrently) : 606 CardTable(whole_heap, scanned_concurrently), 607 _cur_youngergen_card_val(youngergenP1_card), 608 // LNC functionality 609 _lowest_non_clean(NULL), 610 _lowest_non_clean_chunk_size(NULL), 611 _lowest_non_clean_base_chunk_index(NULL), 612 _last_LNC_resizing_collection(NULL) 613 { 614 // max_gens is really GenCollectedHeap::heap()->gen_policy()->number_of_generations() 615 // (which is always 2, young & old), but GenCollectedHeap has not been initialized yet. 616 uint max_gens = 2; 617 _last_cur_val_in_gen = NEW_C_HEAP_ARRAY3(jbyte, max_gens + 1, 618 mtGC, CURRENT_PC, AllocFailStrategy::RETURN_NULL); 619 if (_last_cur_val_in_gen == NULL) { 620 vm_exit_during_initialization("Could not create last_cur_val_in_gen array."); 621 } 622 for (uint i = 0; i < max_gens + 1; i++) { 623 _last_cur_val_in_gen[i] = clean_card_val(); 624 } 625 } 626 627 CardTableRS::~CardTableRS() { 628 if (_last_cur_val_in_gen) { 629 FREE_C_HEAP_ARRAY(jbyte, _last_cur_val_in_gen); 630 _last_cur_val_in_gen = NULL; 631 } 632 if (_lowest_non_clean) { 633 FREE_C_HEAP_ARRAY(CardArr, _lowest_non_clean); 634 _lowest_non_clean = NULL; 635 } 636 if (_lowest_non_clean_chunk_size) { 637 FREE_C_HEAP_ARRAY(size_t, _lowest_non_clean_chunk_size); 638 _lowest_non_clean_chunk_size = NULL; 639 } 640 if (_lowest_non_clean_base_chunk_index) { 641 FREE_C_HEAP_ARRAY(uintptr_t, _lowest_non_clean_base_chunk_index); 642 _lowest_non_clean_base_chunk_index = NULL; 643 } 644 if (_last_LNC_resizing_collection) { 645 FREE_C_HEAP_ARRAY(int, _last_LNC_resizing_collection); 646 _last_LNC_resizing_collection = NULL; 647 } 648 } 649 650 void CardTableRS::initialize() { 651 CardTable::initialize(); 652 _lowest_non_clean = 653 NEW_C_HEAP_ARRAY(CardArr, _max_covered_regions, mtGC); 654 _lowest_non_clean_chunk_size = 655 NEW_C_HEAP_ARRAY(size_t, _max_covered_regions, mtGC); 656 _lowest_non_clean_base_chunk_index = 657 NEW_C_HEAP_ARRAY(uintptr_t, _max_covered_regions, mtGC); 658 _last_LNC_resizing_collection = 659 NEW_C_HEAP_ARRAY(int, _max_covered_regions, mtGC); 660 if (_lowest_non_clean == NULL 661 || _lowest_non_clean_chunk_size == NULL 662 || _lowest_non_clean_base_chunk_index == NULL 663 || _last_LNC_resizing_collection == NULL) 664 vm_exit_during_initialization("couldn't allocate an LNC array."); 665 for (int i = 0; i < _max_covered_regions; i++) { 666 _lowest_non_clean[i] = NULL; 667 _lowest_non_clean_chunk_size[i] = 0; 668 _last_LNC_resizing_collection[i] = -1; 669 } 670 } 671 672 bool CardTableRS::card_will_be_scanned(jbyte cv) { 673 return card_is_dirty_wrt_gen_iter(cv) || is_prev_nonclean_card_val(cv); 674 } 675 676 bool CardTableRS::card_may_have_been_dirty(jbyte cv) { 677 return 678 cv != clean_card && 679 (card_is_dirty_wrt_gen_iter(cv) || 680 CardTableRS::youngergen_may_have_been_dirty(cv)); 681 } 682 683 void CardTableRS::non_clean_card_iterate_possibly_parallel( 684 Space* sp, 685 MemRegion mr, 686 OopsInGenClosure* cl, 687 CardTableRS* ct, 688 uint n_threads) 689 { 690 if (!mr.is_empty()) { 691 if (n_threads > 0) { 692 non_clean_card_iterate_parallel_work(sp, mr, cl, ct, n_threads); 693 } else { 694 // clear_cl finds contiguous dirty ranges of cards to process and clear. 695 696 // This is the single-threaded version used by DefNew. 697 const bool parallel = false; 698 699 DirtyCardToOopClosure* dcto_cl = sp->new_dcto_cl(cl, precision(), cl->gen_boundary(), parallel); 700 ClearNoncleanCardWrapper clear_cl(dcto_cl, ct, parallel); 701 702 clear_cl.do_MemRegion(mr); 703 } 704 } 705 } 706 707 void CardTableRS::non_clean_card_iterate_parallel_work(Space* sp, MemRegion mr, 708 OopsInGenClosure* cl, CardTableRS* ct, 709 uint n_threads) { 710 fatal("Parallel gc not supported here."); 711 } 712 713 bool CardTableRS::is_in_young(oop obj) const { 714 return GenCollectedHeap::heap()->is_in_young(obj); 715 }