Mercurial > hg > truffle
annotate src/share/vm/memory/cardTableModRefBS.hpp @ 113:ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
Summary: Compressed oops in instances, arrays, and headers. Code contributors are coleenp, phh, never, swamyv
Reviewed-by: jmasa, kamg, acorn, tbell, kvn, rasbold
| author | coleenp |
|---|---|
| date | Sun, 13 Apr 2008 17:43:42 -0400 |
| parents | a61af66fc99e |
| children | d1605aabd0a1 37f87013dfd8 |
| rev | line source |
|---|---|
| 0 | 1 /* |
| 2 * Copyright 2000-2006 Sun Microsystems, Inc. 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 Sun Microsystems, Inc., 4150 Network Circle, Santa Clara, | |
| 20 * CA 95054 USA or visit www.sun.com if you need additional information or | |
| 21 * have any 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 #ifndef PRODUCT | |
| 48 // For debugging. | |
| 49 friend class GuaranteeNotModClosure; | |
| 50 #endif | |
| 51 protected: | |
| 52 | |
| 53 enum CardValues { | |
| 54 clean_card = -1, | |
| 55 dirty_card = 0, | |
| 56 precleaned_card = 1, | |
| 57 last_card = 4, | |
| 58 CT_MR_BS_last_reserved = 10 | |
| 59 }; | |
| 60 | |
| 61 // dirty and precleaned are equivalent wrt younger_refs_iter. | |
| 62 static bool card_is_dirty_wrt_gen_iter(jbyte cv) { | |
| 63 return cv == dirty_card || cv == precleaned_card; | |
| 64 } | |
| 65 | |
| 66 // Returns "true" iff the value "cv" will cause the card containing it | |
| 67 // to be scanned in the current traversal. May be overridden by | |
| 68 // subtypes. | |
| 69 virtual bool card_will_be_scanned(jbyte cv) { | |
| 70 return CardTableModRefBS::card_is_dirty_wrt_gen_iter(cv); | |
| 71 } | |
| 72 | |
| 73 // Returns "true" iff the value "cv" may have represented a dirty card at | |
| 74 // some point. | |
| 75 virtual bool card_may_have_been_dirty(jbyte cv) { | |
| 76 return card_is_dirty_wrt_gen_iter(cv); | |
| 77 } | |
| 78 | |
| 79 // The declaration order of these const fields is important; see the | |
| 80 // constructor before changing. | |
| 81 const MemRegion _whole_heap; // the region covered by the card table | |
| 82 const size_t _guard_index; // index of very last element in the card | |
| 83 // table; it is set to a guard value | |
| 84 // (last_card) and should never be modified | |
| 85 const size_t _last_valid_index; // index of the last valid element | |
| 86 const size_t _page_size; // page size used when mapping _byte_map | |
| 87 const size_t _byte_map_size; // in bytes | |
| 88 jbyte* _byte_map; // the card marking array | |
| 89 | |
| 90 int _cur_covered_regions; | |
| 91 // The covered regions should be in address order. | |
| 92 MemRegion* _covered; | |
| 93 // The committed regions correspond one-to-one to the covered regions. | |
| 94 // They represent the card-table memory that has been committed to service | |
| 95 // the corresponding covered region. It may be that committed region for | |
| 96 // one covered region corresponds to a larger region because of page-size | |
| 97 // roundings. Thus, a committed region for one covered region may | |
| 98 // actually extend onto the card-table space for the next covered region. | |
| 99 MemRegion* _committed; | |
| 100 | |
| 101 // The last card is a guard card, and we commit the page for it so | |
| 102 // we can use the card for verification purposes. We make sure we never | |
| 103 // uncommit the MemRegion for that page. | |
| 104 MemRegion _guard_region; | |
| 105 | |
| 106 protected: | |
| 107 // Initialization utilities; covered_words is the size of the covered region | |
| 108 // in, um, words. | |
| 109 inline size_t cards_required(size_t covered_words); | |
| 110 inline size_t compute_byte_map_size(); | |
| 111 | |
| 112 // Finds and return the index of the region, if any, to which the given | |
| 113 // region would be contiguous. If none exists, assign a new region and | |
| 114 // returns its index. Requires that no more than the maximum number of | |
| 115 // covered regions defined in the constructor are ever in use. | |
| 116 int find_covering_region_by_base(HeapWord* base); | |
| 117 | |
| 118 // Same as above, but finds the region containing the given address | |
| 119 // instead of starting at a given base address. | |
| 120 int find_covering_region_containing(HeapWord* addr); | |
| 121 | |
| 122 // Resize one of the regions covered by the remembered set. | |
| 123 void resize_covered_region(MemRegion new_region); | |
| 124 | |
| 125 // Returns the leftmost end of a committed region corresponding to a | |
| 126 // covered region before covered region "ind", or else "NULL" if "ind" is | |
| 127 // the first covered region. | |
| 128 HeapWord* largest_prev_committed_end(int ind) const; | |
| 129 | |
| 130 // Returns the part of the region mr that doesn't intersect with | |
| 131 // any committed region other than self. Used to prevent uncommitting | |
| 132 // regions that are also committed by other regions. Also protects | |
| 133 // against uncommitting the guard region. | |
| 134 MemRegion committed_unique_to_self(int self, MemRegion mr) const; | |
| 135 | |
| 136 // Mapping from address to card marking array entry | |
| 137 jbyte* byte_for(const void* p) const { | |
| 138 assert(_whole_heap.contains(p), | |
| 139 "out of bounds access to card marking array"); | |
| 140 jbyte* result = &byte_map_base[uintptr_t(p) >> card_shift]; | |
| 141 assert(result >= _byte_map && result < _byte_map + _byte_map_size, | |
| 142 "out of bounds accessor for card marking array"); | |
| 143 return result; | |
| 144 } | |
| 145 | |
| 146 // The card table byte one after the card marking array | |
| 147 // entry for argument address. Typically used for higher bounds | |
| 148 // for loops iterating through the card table. | |
| 149 jbyte* byte_after(const void* p) const { | |
| 150 return byte_for(p) + 1; | |
| 151 } | |
| 152 | |
| 153 // Mapping from card marking array entry to address of first word | |
| 154 HeapWord* addr_for(const jbyte* p) const { | |
| 155 assert(p >= _byte_map && p < _byte_map + _byte_map_size, | |
| 156 "out of bounds access to card marking array"); | |
| 157 size_t delta = pointer_delta(p, byte_map_base, sizeof(jbyte)); | |
| 158 HeapWord* result = (HeapWord*) (delta << card_shift); | |
| 159 assert(_whole_heap.contains(result), | |
| 160 "out of bounds accessor from card marking array"); | |
| 161 return result; | |
| 162 } | |
| 163 | |
| 164 // Iterate over the portion of the card-table which covers the given | |
| 165 // region mr in the given space and apply cl to any dirty sub-regions | |
| 166 // of mr. cl and dcto_cl must either be the same closure or cl must | |
| 167 // wrap dcto_cl. Both are required - neither may be NULL. Also, dcto_cl | |
| 168 // may be modified. Note that this function will operate in a parallel | |
| 169 // mode if worker threads are available. | |
| 170 void non_clean_card_iterate(Space* sp, MemRegion mr, | |
| 171 DirtyCardToOopClosure* dcto_cl, | |
| 172 MemRegionClosure* cl, | |
| 173 bool clear); | |
| 174 | |
| 175 // Utility function used to implement the other versions below. | |
| 176 void non_clean_card_iterate_work(MemRegion mr, MemRegionClosure* cl, | |
| 177 bool clear); | |
| 178 | |
| 179 void par_non_clean_card_iterate_work(Space* sp, MemRegion mr, | |
| 180 DirtyCardToOopClosure* dcto_cl, | |
| 181 MemRegionClosure* cl, | |
| 182 bool clear, | |
| 183 int n_threads); | |
| 184 | |
| 185 // Dirty the bytes corresponding to "mr" (not all of which must be | |
| 186 // covered.) | |
| 187 void dirty_MemRegion(MemRegion mr); | |
| 188 | |
| 189 // Clear (to clean_card) the bytes entirely contained within "mr" (not | |
| 190 // all of which must be covered.) | |
| 191 void clear_MemRegion(MemRegion mr); | |
| 192 | |
| 193 // *** Support for parallel card scanning. | |
| 194 | |
| 195 enum SomeConstantsForParallelism { | |
| 196 StridesPerThread = 2, | |
| 197 CardsPerStrideChunk = 256 | |
| 198 }; | |
| 199 | |
| 200 // This is an array, one element per covered region of the card table. | |
| 201 // Each entry is itself an array, with one element per chunk in the | |
| 202 // covered region. Each entry of these arrays is the lowest non-clean | |
| 203 // card of the corresponding chunk containing part of an object from the | |
| 204 // previous chunk, or else NULL. | |
| 205 typedef jbyte* CardPtr; | |
| 206 typedef CardPtr* CardArr; | |
| 207 CardArr* _lowest_non_clean; | |
| 208 size_t* _lowest_non_clean_chunk_size; | |
| 209 uintptr_t* _lowest_non_clean_base_chunk_index; | |
| 210 int* _last_LNC_resizing_collection; | |
| 211 | |
| 212 // Initializes "lowest_non_clean" to point to the array for the region | |
| 213 // covering "sp", and "lowest_non_clean_base_chunk_index" to the chunk | |
| 214 // index of the corresponding to the first element of that array. | |
| 215 // Ensures that these arrays are of sufficient size, allocating if necessary. | |
| 216 // May be called by several threads concurrently. | |
| 217 void get_LNC_array_for_space(Space* sp, | |
| 218 jbyte**& lowest_non_clean, | |
| 219 uintptr_t& lowest_non_clean_base_chunk_index, | |
| 220 size_t& lowest_non_clean_chunk_size); | |
| 221 | |
| 222 // Returns the number of chunks necessary to cover "mr". | |
| 223 size_t chunks_to_cover(MemRegion mr) { | |
| 224 return (size_t)(addr_to_chunk_index(mr.last()) - | |
| 225 addr_to_chunk_index(mr.start()) + 1); | |
| 226 } | |
| 227 | |
| 228 // Returns the index of the chunk in a stride which | |
| 229 // covers the given address. | |
| 230 uintptr_t addr_to_chunk_index(const void* addr) { | |
| 231 uintptr_t card = (uintptr_t) byte_for(addr); | |
| 232 return card / CardsPerStrideChunk; | |
| 233 } | |
| 234 | |
| 235 // Apply cl, which must either itself apply dcto_cl or be dcto_cl, | |
| 236 // to the cards in the stride (of n_strides) within the given space. | |
| 237 void process_stride(Space* sp, | |
| 238 MemRegion used, | |
| 239 jint stride, int n_strides, | |
| 240 DirtyCardToOopClosure* dcto_cl, | |
| 241 MemRegionClosure* cl, | |
| 242 bool clear, | |
| 243 jbyte** lowest_non_clean, | |
| 244 uintptr_t lowest_non_clean_base_chunk_index, | |
| 245 size_t lowest_non_clean_chunk_size); | |
| 246 | |
| 247 // Makes sure that chunk boundaries are handled appropriately, by | |
| 248 // adjusting the min_done of dcto_cl, and by using a special card-table | |
| 249 // value to indicate how min_done should be set. | |
| 250 void process_chunk_boundaries(Space* sp, | |
| 251 DirtyCardToOopClosure* dcto_cl, | |
| 252 MemRegion chunk_mr, | |
| 253 MemRegion used, | |
| 254 jbyte** lowest_non_clean, | |
| 255 uintptr_t lowest_non_clean_base_chunk_index, | |
| 256 size_t lowest_non_clean_chunk_size); | |
| 257 | |
| 258 public: | |
| 259 // Constants | |
| 260 enum SomePublicConstants { | |
| 261 card_shift = 9, | |
| 262 card_size = 1 << card_shift, | |
| 263 card_size_in_words = card_size / sizeof(HeapWord) | |
| 264 }; | |
| 265 | |
| 266 // For RTTI simulation. | |
| 267 BarrierSet::Name kind() { return BarrierSet::CardTableModRef; } | |
| 268 bool is_a(BarrierSet::Name bsn) { | |
| 269 return bsn == BarrierSet::CardTableModRef || bsn == BarrierSet::ModRef; | |
| 270 } | |
| 271 | |
| 272 CardTableModRefBS(MemRegion whole_heap, int max_covered_regions); | |
| 273 | |
| 274 // *** Barrier set functions. | |
| 275 | |
|
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
276 inline bool write_ref_needs_barrier(void* field, oop new_val) { |
| 0 | 277 // Note that this assumes the perm gen is the highest generation |
| 278 // in the address space | |
| 279 return new_val != NULL && !new_val->is_perm(); | |
| 280 } | |
| 281 | |
| 282 // Record a reference update. Note that these versions are precise! | |
| 283 // The scanning code has to handle the fact that the write barrier may be | |
| 284 // either precise or imprecise. We make non-virtual inline variants of | |
| 285 // these functions here for performance. | |
| 286 protected: | |
| 287 void write_ref_field_work(oop obj, size_t offset, oop newVal); | |
|
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
288 void write_ref_field_work(void* field, oop newVal); |
| 0 | 289 public: |
| 290 | |
| 291 bool has_write_ref_array_opt() { return true; } | |
| 292 bool has_write_region_opt() { return true; } | |
| 293 | |
| 294 inline void inline_write_region(MemRegion mr) { | |
| 295 dirty_MemRegion(mr); | |
| 296 } | |
| 297 protected: | |
| 298 void write_region_work(MemRegion mr) { | |
| 299 inline_write_region(mr); | |
| 300 } | |
| 301 public: | |
| 302 | |
| 303 inline void inline_write_ref_array(MemRegion mr) { | |
| 304 dirty_MemRegion(mr); | |
| 305 } | |
| 306 protected: | |
| 307 void write_ref_array_work(MemRegion mr) { | |
| 308 inline_write_ref_array(mr); | |
| 309 } | |
| 310 public: | |
| 311 | |
| 312 bool is_aligned(HeapWord* addr) { | |
| 313 return is_card_aligned(addr); | |
| 314 } | |
| 315 | |
| 316 // *** Card-table-barrier-specific things. | |
| 317 | |
|
113
ba764ed4b6f2
6420645: Create a vm that uses compressed oops for up to 32gb heapsizes
coleenp
parents:
0
diff
changeset
|
318 inline void inline_write_ref_field(void* field, oop newVal) { |
| 0 | 319 jbyte* byte = byte_for(field); |
| 320 *byte = dirty_card; | |
| 321 } | |
| 322 | |
| 323 // Card marking array base (adjusted for heap low boundary) | |
| 324 // This would be the 0th element of _byte_map, if the heap started at 0x0. | |
| 325 // But since the heap starts at some higher address, this points to somewhere | |
| 326 // before the beginning of the actual _byte_map. | |
| 327 jbyte* byte_map_base; | |
| 328 | |
| 329 // Return true if "p" is at the start of a card. | |
| 330 bool is_card_aligned(HeapWord* p) { | |
| 331 jbyte* pcard = byte_for(p); | |
| 332 return (addr_for(pcard) == p); | |
| 333 } | |
| 334 | |
| 335 // The kinds of precision a CardTableModRefBS may offer. | |
| 336 enum PrecisionStyle { | |
| 337 Precise, | |
| 338 ObjHeadPreciseArray | |
| 339 }; | |
| 340 | |
| 341 // Tells what style of precision this card table offers. | |
| 342 PrecisionStyle precision() { | |
| 343 return ObjHeadPreciseArray; // Only one supported for now. | |
| 344 } | |
| 345 | |
| 346 // ModRefBS functions. | |
| 347 void invalidate(MemRegion mr); | |
| 348 void clear(MemRegion mr); | |
| 349 void mod_oop_in_space_iterate(Space* sp, OopClosure* cl, | |
| 350 bool clear = false, | |
| 351 bool before_save_marks = false); | |
| 352 | |
| 353 // *** Card-table-RemSet-specific things. | |
| 354 | |
| 355 // Invoke "cl.do_MemRegion" on a set of MemRegions that collectively | |
| 356 // includes all the modified cards (expressing each card as a | |
| 357 // MemRegion). Thus, several modified cards may be lumped into one | |
| 358 // region. The regions are non-overlapping, and are visited in | |
| 359 // *decreasing* address order. (This order aids with imprecise card | |
| 360 // marking, where a dirty card may cause scanning, and summarization | |
| 361 // marking, of objects that extend onto subsequent cards.) | |
| 362 // If "clear" is true, the card is (conceptually) marked unmodified before | |
| 363 // applying the closure. | |
| 364 void mod_card_iterate(MemRegionClosure* cl, bool clear = false) { | |
| 365 non_clean_card_iterate_work(_whole_heap, cl, clear); | |
| 366 } | |
| 367 | |
| 368 // Like the "mod_cards_iterate" above, except only invokes the closure | |
| 369 // for cards within the MemRegion "mr" (which is required to be | |
| 370 // card-aligned and sized.) | |
| 371 void mod_card_iterate(MemRegion mr, MemRegionClosure* cl, | |
| 372 bool clear = false) { | |
| 373 non_clean_card_iterate_work(mr, cl, clear); | |
| 374 } | |
| 375 | |
| 376 static uintx ct_max_alignment_constraint(); | |
| 377 | |
| 378 // Apply closure cl to the dirty cards lying completely | |
| 379 // within MemRegion mr, setting the cards to precleaned. | |
| 380 void dirty_card_iterate(MemRegion mr, MemRegionClosure* cl); | |
| 381 | |
| 382 // Return the MemRegion corresponding to the first maximal run | |
| 383 // of dirty cards lying completely within MemRegion mr, after | |
| 384 // marking those cards precleaned. | |
| 385 MemRegion dirty_card_range_after_preclean(MemRegion mr); | |
| 386 | |
| 387 // Set all the dirty cards in the given region to precleaned state. | |
| 388 void preclean_dirty_cards(MemRegion mr); | |
| 389 | |
| 390 // Mapping from address to card marking array index. | |
| 391 int index_for(void* p) { | |
| 392 assert(_whole_heap.contains(p), | |
| 393 "out of bounds access to card marking array"); | |
| 394 return byte_for(p) - _byte_map; | |
| 395 } | |
| 396 | |
| 397 void verify(); | |
| 398 void verify_guard(); | |
| 399 | |
| 400 void verify_clean_region(MemRegion mr) PRODUCT_RETURN; | |
| 401 | |
| 402 static size_t par_chunk_heapword_alignment() { | |
| 403 return CardsPerStrideChunk * card_size_in_words; | |
| 404 } | |
| 405 }; | |
| 406 | |
| 407 class CardTableRS; | |
| 408 | |
| 409 // A specialization for the CardTableRS gen rem set. | |
| 410 class CardTableModRefBSForCTRS: public CardTableModRefBS { | |
| 411 CardTableRS* _rs; | |
| 412 protected: | |
| 413 bool card_will_be_scanned(jbyte cv); | |
| 414 bool card_may_have_been_dirty(jbyte cv); | |
| 415 public: | |
| 416 CardTableModRefBSForCTRS(MemRegion whole_heap, | |
| 417 int max_covered_regions) : | |
| 418 CardTableModRefBS(whole_heap, max_covered_regions) {} | |
| 419 | |
| 420 void set_CTRS(CardTableRS* rs) { _rs = rs; } | |
| 421 }; |