1 /* 2 * Copyright (c) 2001, 2012, 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_GC_IMPLEMENTATION_G1_CONCURRENTMARK_INLINE_HPP 26 #define SHARE_VM_GC_IMPLEMENTATION_G1_CONCURRENTMARK_INLINE_HPP 27 28 #include "gc_implementation/g1/concurrentMark.hpp" 29 #include "gc_implementation/g1/g1CollectedHeap.inline.hpp" 30 31 // Utility routine to set an exclusive range of cards on the given 32 // card liveness bitmap 33 inline void ConcurrentMark::set_card_bitmap_range(BitMap* card_bm, 34 BitMap::idx_t start_idx, 35 BitMap::idx_t end_idx, 36 bool is_par) { 37 38 // Set the exclusive bit range [start_idx, end_idx). 39 assert((end_idx - start_idx) > 0, "at least one card"); 40 assert(end_idx <= card_bm->size(), "sanity"); 41 42 // Silently clip the end index 43 end_idx = MIN2(end_idx, card_bm->size()); 44 45 // For small ranges use a simple loop; otherwise use set_range or 46 // use par_at_put_range (if parallel). The range is made up of the 47 // cards that are spanned by an object/mem region so 8 cards will 48 // allow up to object sizes up to 4K to be handled using the loop. 49 if ((end_idx - start_idx) <= 8) { 50 for (BitMap::idx_t i = start_idx; i < end_idx; i += 1) { 51 if (is_par) { 52 card_bm->par_set_bit(i); 53 } else { 54 card_bm->set_bit(i); 55 } 56 } 57 } else { 58 // Note BitMap::par_at_put_range() and BitMap::set_range() are exclusive. 59 if (is_par) { 60 card_bm->par_at_put_range(start_idx, end_idx, true); 61 } else { 62 card_bm->set_range(start_idx, end_idx); 63 } 64 } 65 } 66 67 // Returns the index in the liveness accounting card bitmap 68 // for the given address 69 inline BitMap::idx_t ConcurrentMark::card_bitmap_index_for(HeapWord* addr) { 70 // Below, the term "card num" means the result of shifting an address 71 // by the card shift -- address 0 corresponds to card number 0. One 72 // must subtract the card num of the bottom of the heap to obtain a 73 // card table index. 74 intptr_t card_num = intptr_t(uintptr_t(addr) >> CardTableModRefBS::card_shift); 75 return card_num - heap_bottom_card_num(); 76 } 77 78 // Counts the given memory region in the given task/worker 79 // counting data structures. 80 inline void ConcurrentMark::count_region(MemRegion mr, HeapRegion* hr, 81 size_t* marked_bytes_array, 82 BitMap* task_card_bm) { 83 G1CollectedHeap* g1h = _g1h; 84 CardTableModRefBS* ct_bs = (CardTableModRefBS*) (g1h->barrier_set()); 85 86 HeapWord* start = mr.start(); 87 HeapWord* end = mr.end(); 88 size_t region_size_bytes = mr.byte_size(); 89 uint index = hr->hrs_index(); 90 91 assert(!hr->continuesHumongous(), "should not be HC region"); 92 assert(hr == g1h->heap_region_containing(start), "sanity"); 93 assert(hr == g1h->heap_region_containing(mr.last()), "sanity"); 94 assert(marked_bytes_array != NULL, "pre-condition"); 95 assert(task_card_bm != NULL, "pre-condition"); 96 97 // Add to the task local marked bytes for this region. 98 marked_bytes_array[index] += region_size_bytes; 99 100 BitMap::idx_t start_idx = card_bitmap_index_for(start); 101 BitMap::idx_t end_idx = card_bitmap_index_for(end); 102 103 // Note: if we're looking at a region that coincides with the end 104 // of the heap - end could actually be outside the heap and end_idx 105 // correspond to a card that is also outside the heap. 106 if (g1h->is_in_g1_reserved(end) && !ct_bs->is_card_aligned(end)) { 107 // end of region is not card aligned - incremement to cover 108 // all the cards spanned by the region. 109 end_idx += 1; 110 } 111 // The card bitmap is task/worker specific => no need to use 112 // the 'par' BitMap routines. 113 // Set bits in the exclusive bit range [start_idx, end_idx). 114 set_card_bitmap_range(task_card_bm, start_idx, end_idx, false /* is_par */); 115 } 116 117 // Counts the given memory region in the task/worker counting 118 // data structures for the given worker id. 119 inline void ConcurrentMark::count_region(MemRegion mr, 120 HeapRegion* hr, 121 uint worker_id) { 122 size_t* marked_bytes_array = count_marked_bytes_array_for(worker_id); 123 BitMap* task_card_bm = count_card_bitmap_for(worker_id); 124 count_region(mr, hr, marked_bytes_array, task_card_bm); 125 } 126 127 // Counts the given memory region, which may be a single object, in the 128 // task/worker counting data structures for the given worker id. 129 inline void ConcurrentMark::count_region(MemRegion mr, uint worker_id) { 130 HeapWord* addr = mr.start(); 131 HeapRegion* hr = _g1h->heap_region_containing_raw(addr); 132 count_region(mr, hr, worker_id); 133 } 134 135 // Counts the given object in the given task/worker counting data structures. 136 inline void ConcurrentMark::count_object(oop obj, 137 HeapRegion* hr, 138 size_t* marked_bytes_array, 139 BitMap* task_card_bm) { 140 MemRegion mr((HeapWord*)obj, obj->size()); 141 count_region(mr, hr, marked_bytes_array, task_card_bm); 142 } 143 144 // Counts the given object in the task/worker counting data 145 // structures for the given worker id. 146 inline void ConcurrentMark::count_object(oop obj, 147 HeapRegion* hr, 148 uint worker_id) { 149 size_t* marked_bytes_array = count_marked_bytes_array_for(worker_id); 150 BitMap* task_card_bm = count_card_bitmap_for(worker_id); 151 HeapWord* addr = (HeapWord*) obj; 152 count_object(obj, hr, marked_bytes_array, task_card_bm); 153 } 154 155 // Attempts to mark the given object and, if successful, counts 156 // the object in the given task/worker counting structures. 157 inline bool ConcurrentMark::par_mark_and_count(oop obj, 158 HeapRegion* hr, 159 size_t* marked_bytes_array, 160 BitMap* task_card_bm) { 161 HeapWord* addr = (HeapWord*)obj; 162 if (_nextMarkBitMap->parMark(addr)) { 163 // Update the task specific count data for the object. 164 count_object(obj, hr, marked_bytes_array, task_card_bm); 165 return true; 166 } 167 return false; 168 } 169 170 // Attempts to mark the given object and, if successful, counts 171 // the object in the task/worker counting structures for the 172 // given worker id. 173 inline bool ConcurrentMark::par_mark_and_count(oop obj, 174 size_t word_size, 175 HeapRegion* hr, 176 uint worker_id) { 177 HeapWord* addr = (HeapWord*)obj; 178 if (_nextMarkBitMap->parMark(addr)) { 179 MemRegion mr(addr, word_size); 180 count_region(mr, hr, worker_id); 181 return true; 182 } 183 return false; 184 } 185 186 // Attempts to mark the given object and, if successful, counts 187 // the object in the task/worker counting structures for the 188 // given worker id. 189 inline bool ConcurrentMark::par_mark_and_count(oop obj, 190 HeapRegion* hr, 191 uint worker_id) { 192 HeapWord* addr = (HeapWord*)obj; 193 if (_nextMarkBitMap->parMark(addr)) { 194 // Update the task specific count data for the object. 195 count_object(obj, hr, worker_id); 196 return true; 197 } 198 return false; 199 } 200 201 // As above - but we don't know the heap region containing the 202 // object and so have to supply it. 203 inline bool ConcurrentMark::par_mark_and_count(oop obj, uint worker_id) { 204 HeapWord* addr = (HeapWord*)obj; 205 HeapRegion* hr = _g1h->heap_region_containing_raw(addr); 206 return par_mark_and_count(obj, hr, worker_id); 207 } 208 209 // Similar to the above routine but we already know the size, in words, of 210 // the object that we wish to mark/count 211 inline bool ConcurrentMark::par_mark_and_count(oop obj, 212 size_t word_size, 213 uint worker_id) { 214 HeapWord* addr = (HeapWord*)obj; 215 if (_nextMarkBitMap->parMark(addr)) { 216 // Update the task specific count data for the object. 217 MemRegion mr(addr, word_size); 218 count_region(mr, worker_id); 219 return true; 220 } 221 return false; 222 } 223 224 // Unconditionally mark the given object, and unconditinally count 225 // the object in the counting structures for worker id 0. 226 // Should *not* be called from parallel code. 227 inline bool ConcurrentMark::mark_and_count(oop obj, HeapRegion* hr) { 228 HeapWord* addr = (HeapWord*)obj; 229 _nextMarkBitMap->mark(addr); 230 // Update the task specific count data for the object. 231 count_object(obj, hr, 0 /* worker_id */); 232 return true; 233 } 234 235 // As above - but we don't have the heap region containing the 236 // object, so we have to supply it. 237 inline bool ConcurrentMark::mark_and_count(oop obj) { 238 HeapWord* addr = (HeapWord*)obj; 239 HeapRegion* hr = _g1h->heap_region_containing_raw(addr); 240 return mark_and_count(obj, hr); 241 } 242 243 inline bool CMBitMapRO::iterate(BitMapClosure* cl, MemRegion mr) { 244 HeapWord* start_addr = MAX2(startWord(), mr.start()); 245 HeapWord* end_addr = MIN2(endWord(), mr.end()); 246 247 if (end_addr > start_addr) { 248 // Right-open interval [start-offset, end-offset). 249 BitMap::idx_t start_offset = heapWordToOffset(start_addr); 250 BitMap::idx_t end_offset = heapWordToOffset(end_addr); 251 252 start_offset = _bm.get_next_one_offset(start_offset, end_offset); 253 while (start_offset < end_offset) { 254 HeapWord* obj_addr = offsetToHeapWord(start_offset); 255 oop obj = (oop) obj_addr; 256 if (!cl->do_bit(start_offset)) { 257 return false; 258 } 259 HeapWord* next_addr = MIN2(obj_addr + obj->size(), end_addr); 260 BitMap::idx_t next_offset = heapWordToOffset(next_addr); 261 start_offset = _bm.get_next_one_offset(next_offset, end_offset); 262 } 263 } 264 return true; 265 } 266 267 inline bool CMBitMapRO::iterate(BitMapClosure* cl) { 268 MemRegion mr(startWord(), sizeInWords()); 269 return iterate(cl, mr); 270 } 271 272 inline void CMTask::push(oop obj) { 273 HeapWord* objAddr = (HeapWord*) obj; 274 assert(_g1h->is_in_g1_reserved(objAddr), "invariant"); 275 assert(!_g1h->is_on_master_free_list( 276 _g1h->heap_region_containing((HeapWord*) objAddr)), "invariant"); 277 assert(!_g1h->is_obj_ill(obj), "invariant"); 278 assert(_nextMarkBitMap->isMarked(objAddr), "invariant"); 279 280 if (_cm->verbose_high()) { 281 gclog_or_tty->print_cr("[%d] pushing "PTR_FORMAT, _task_id, (void*) obj); 282 } 283 284 if (!_task_queue->push(obj)) { 285 // The local task queue looks full. We need to push some entries 286 // to the global stack. 287 288 if (_cm->verbose_medium()) { 289 gclog_or_tty->print_cr("[%d] task queue overflow, " 290 "moving entries to the global stack", 291 _task_id); 292 } 293 move_entries_to_global_stack(); 294 295 // this should succeed since, even if we overflow the global 296 // stack, we should have definitely removed some entries from the 297 // local queue. So, there must be space on it. 298 bool success = _task_queue->push(obj); 299 assert(success, "invariant"); 300 } 301 302 statsOnly( int tmp_size = _task_queue->size(); 303 if (tmp_size > _local_max_size) { 304 _local_max_size = tmp_size; 305 } 306 ++_local_pushes ); 307 } 308 309 // This determines whether the method below will check both the local 310 // and global fingers when determining whether to push on the stack a 311 // gray object (value 1) or whether it will only check the global one 312 // (value 0). The tradeoffs are that the former will be a bit more 313 // accurate and possibly push less on the stack, but it might also be 314 // a little bit slower. 315 316 #define _CHECK_BOTH_FINGERS_ 1 317 318 inline void CMTask::deal_with_reference(oop obj) { 319 if (_cm->verbose_high()) { 320 gclog_or_tty->print_cr("[%d] we're dealing with reference = "PTR_FORMAT, 321 _task_id, (void*) obj); 322 } 323 324 ++_refs_reached; 325 326 HeapWord* objAddr = (HeapWord*) obj; 327 assert(obj->is_oop_or_null(true /* ignore mark word */), "Error"); 328 if (_g1h->is_in_g1_reserved(objAddr)) { 329 assert(obj != NULL, "null check is implicit"); 330 if (!_nextMarkBitMap->isMarked(objAddr)) { 331 // Only get the containing region if the object is not marked on the 332 // bitmap (otherwise, it's a waste of time since we won't do 333 // anything with it). 334 HeapRegion* hr = _g1h->heap_region_containing_raw(obj); 335 if (!hr->obj_allocated_since_next_marking(obj)) { 336 if (_cm->verbose_high()) { 337 gclog_or_tty->print_cr("[%d] "PTR_FORMAT" is not considered marked", 338 _task_id, (void*) obj); 339 } 340 341 // we need to mark it first 342 if (_cm->par_mark_and_count(obj, hr, _marked_bytes_array, _card_bm)) { 343 // No OrderAccess:store_load() is needed. It is implicit in the 344 // CAS done in CMBitMap::parMark() call in the routine above. 345 HeapWord* global_finger = _cm->finger(); 346 347 #if _CHECK_BOTH_FINGERS_ 348 // we will check both the local and global fingers 349 350 if (_finger != NULL && objAddr < _finger) { 351 if (_cm->verbose_high()) { 352 gclog_or_tty->print_cr("[%d] below the local finger ("PTR_FORMAT"), " 353 "pushing it", _task_id, _finger); 354 } 355 push(obj); 356 } else if (_curr_region != NULL && objAddr < _region_limit) { 357 // do nothing 358 } else if (objAddr < global_finger) { 359 // Notice that the global finger might be moving forward 360 // concurrently. This is not a problem. In the worst case, we 361 // mark the object while it is above the global finger and, by 362 // the time we read the global finger, it has moved forward 363 // passed this object. In this case, the object will probably 364 // be visited when a task is scanning the region and will also 365 // be pushed on the stack. So, some duplicate work, but no 366 // correctness problems. 367 368 if (_cm->verbose_high()) { 369 gclog_or_tty->print_cr("[%d] below the global finger " 370 "("PTR_FORMAT"), pushing it", 371 _task_id, global_finger); 372 } 373 push(obj); 374 } else { 375 // do nothing 376 } 377 #else // _CHECK_BOTH_FINGERS_ 378 // we will only check the global finger 379 380 if (objAddr < global_finger) { 381 // see long comment above 382 383 if (_cm->verbose_high()) { 384 gclog_or_tty->print_cr("[%d] below the global finger " 385 "("PTR_FORMAT"), pushing it", 386 _task_id, global_finger); 387 } 388 push(obj); 389 } 390 #endif // _CHECK_BOTH_FINGERS_ 391 } 392 } 393 } 394 } 395 } 396 397 inline void ConcurrentMark::markPrev(oop p) { 398 assert(!_prevMarkBitMap->isMarked((HeapWord*) p), "sanity"); 399 // Note we are overriding the read-only view of the prev map here, via 400 // the cast. 401 ((CMBitMap*)_prevMarkBitMap)->mark((HeapWord*) p); 402 } 403 404 inline void ConcurrentMark::grayRoot(oop obj, size_t word_size, 405 uint worker_id, HeapRegion* hr) { 406 assert(obj != NULL, "pre-condition"); 407 HeapWord* addr = (HeapWord*) obj; 408 if (hr == NULL) { 409 hr = _g1h->heap_region_containing_raw(addr); 410 } else { 411 assert(hr->is_in(addr), "pre-condition"); 412 } 413 assert(hr != NULL, "sanity"); 414 // Given that we're looking for a region that contains an object 415 // header it's impossible to get back a HC region. 416 assert(!hr->continuesHumongous(), "sanity"); 417 418 // We cannot assert that word_size == obj->size() given that obj 419 // might not be in a consistent state (another thread might be in 420 // the process of copying it). So the best thing we can do is to 421 // assert that word_size is under an upper bound which is its 422 // containing region's capacity. 423 assert(word_size * HeapWordSize <= hr->capacity(), 424 err_msg("size: "SIZE_FORMAT" capacity: "SIZE_FORMAT" "HR_FORMAT, 425 word_size * HeapWordSize, hr->capacity(), 426 HR_FORMAT_PARAMS(hr))); 427 428 if (addr < hr->next_top_at_mark_start()) { 429 if (!_nextMarkBitMap->isMarked(addr)) { 430 par_mark_and_count(obj, word_size, hr, worker_id); 431 } 432 } 433 } 434 435 #endif // SHARE_VM_GC_IMPLEMENTATION_G1_CONCURRENTMARK_INLINE_HPP