1 /* 2 * Copyright (c) 2001, 2015, 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_G1_CONCURRENTMARK_INLINE_HPP 26 #define SHARE_VM_GC_G1_CONCURRENTMARK_INLINE_HPP 27 28 #include "gc/g1/concurrentMark.hpp" 29 #include "gc/g1/g1CollectedHeap.inline.hpp" 30 #include "gc/shared/taskqueue.inline.hpp" 31 32 // Utility routine to set an exclusive range of cards on the given 33 // card liveness bitmap 34 inline void ConcurrentMark::set_card_bitmap_range(BitMap* card_bm, 35 BitMap::idx_t start_idx, 36 BitMap::idx_t end_idx, 37 bool is_par) { 38 39 // Set the exclusive bit range [start_idx, end_idx). 40 assert((end_idx - start_idx) > 0, "at least one card"); 41 assert(end_idx <= card_bm->size(), "sanity"); 42 43 // Silently clip the end index 44 end_idx = MIN2(end_idx, card_bm->size()); 45 46 // For small ranges use a simple loop; otherwise use set_range or 47 // use par_at_put_range (if parallel). The range is made up of the 48 // cards that are spanned by an object/mem region so 8 cards will 49 // allow up to object sizes up to 4K to be handled using the loop. 50 if ((end_idx - start_idx) <= 8) { 51 for (BitMap::idx_t i = start_idx; i < end_idx; i += 1) { 52 if (is_par) { 53 card_bm->par_set_bit(i); 54 } else { 55 card_bm->set_bit(i); 56 } 57 } 58 } else { 59 // Note BitMap::par_at_put_range() and BitMap::set_range() are exclusive. 60 if (is_par) { 61 card_bm->par_at_put_range(start_idx, end_idx, true); 62 } else { 63 card_bm->set_range(start_idx, end_idx); 64 } 65 } 66 } 67 68 // Returns the index in the liveness accounting card bitmap 69 // for the given address 70 inline BitMap::idx_t ConcurrentMark::card_bitmap_index_for(HeapWord* addr) { 71 // Below, the term "card num" means the result of shifting an address 72 // by the card shift -- address 0 corresponds to card number 0. One 73 // must subtract the card num of the bottom of the heap to obtain a 74 // card table index. 75 intptr_t card_num = intptr_t(uintptr_t(addr) >> CardTableModRefBS::card_shift); 76 return card_num - heap_bottom_card_num(); 77 } 78 79 // Counts the given memory region in the given task/worker 80 // counting data structures. 81 inline void ConcurrentMark::count_region(MemRegion mr, HeapRegion* hr, 82 size_t* marked_bytes_array, 83 BitMap* task_card_bm) { 84 G1CollectedHeap* g1h = _g1h; 85 CardTableModRefBS* ct_bs = g1h->g1_barrier_set(); 86 87 HeapWord* start = mr.start(); 88 HeapWord* end = mr.end(); 89 size_t region_size_bytes = mr.byte_size(); 90 uint index = hr->hrm_index(); 91 92 assert(!hr->is_continues_humongous(), "should not be HC region"); 93 assert(hr == g1h->heap_region_containing(start), "sanity"); 94 assert(hr == g1h->heap_region_containing(mr.last()), "sanity"); 95 assert(marked_bytes_array != NULL, "pre-condition"); 96 assert(task_card_bm != NULL, "pre-condition"); 97 98 // Add to the task local marked bytes for this region. 99 marked_bytes_array[index] += region_size_bytes; 100 101 BitMap::idx_t start_idx = card_bitmap_index_for(start); 102 BitMap::idx_t end_idx = card_bitmap_index_for(end); 103 104 // Note: if we're looking at the last region in heap - end 105 // could be actually just beyond the end of the heap; end_idx 106 // will then correspond to a (non-existent) card that is also 107 // just beyond the heap. 108 if (g1h->is_in_g1_reserved(end) && !ct_bs->is_card_aligned(end)) { 109 // end of region is not card aligned - increment to cover 110 // all the cards spanned by the region. 111 end_idx += 1; 112 } 113 // The card bitmap is task/worker specific => no need to use 114 // the 'par' BitMap routines. 115 // Set bits in the exclusive bit range [start_idx, end_idx). 116 set_card_bitmap_range(task_card_bm, start_idx, end_idx, false /* is_par */); 117 } 118 119 // Counts the given memory region in the task/worker counting 120 // data structures for the given worker id. 121 inline void ConcurrentMark::count_region(MemRegion mr, 122 HeapRegion* hr, 123 uint worker_id) { 124 size_t* marked_bytes_array = count_marked_bytes_array_for(worker_id); 125 BitMap* task_card_bm = count_card_bitmap_for(worker_id); 126 count_region(mr, hr, marked_bytes_array, task_card_bm); 127 } 128 129 // Counts the given object in the given task/worker counting data structures. 130 inline void ConcurrentMark::count_object(oop obj, 131 HeapRegion* hr, 132 size_t* marked_bytes_array, 133 BitMap* task_card_bm) { 134 MemRegion mr((HeapWord*)obj, obj->size()); 135 count_region(mr, hr, marked_bytes_array, task_card_bm); 136 } 137 138 // Attempts to mark the given object and, if successful, counts 139 // the object in the given task/worker counting structures. 140 inline bool ConcurrentMark::par_mark_and_count(oop obj, 141 HeapRegion* hr, 142 size_t* marked_bytes_array, 143 BitMap* task_card_bm) { 144 HeapWord* addr = (HeapWord*)obj; 145 if (_nextMarkBitMap->parMark(addr)) { 146 // Update the task specific count data for the object. 147 count_object(obj, hr, marked_bytes_array, task_card_bm); 148 return true; 149 } 150 return false; 151 } 152 153 // Attempts to mark the given object and, if successful, counts 154 // the object in the task/worker counting structures for the 155 // given worker id. 156 inline bool ConcurrentMark::par_mark_and_count(oop obj, 157 size_t word_size, 158 HeapRegion* hr, 159 uint worker_id) { 160 HeapWord* addr = (HeapWord*)obj; 161 if (_nextMarkBitMap->parMark(addr)) { 162 MemRegion mr(addr, word_size); 163 count_region(mr, hr, worker_id); 164 return true; 165 } 166 return false; 167 } 168 169 inline bool CMBitMapRO::iterate(BitMapClosure* cl, MemRegion mr) { 170 HeapWord* start_addr = MAX2(startWord(), mr.start()); 171 HeapWord* end_addr = MIN2(endWord(), mr.end()); 172 173 if (end_addr > start_addr) { 174 // Right-open interval [start-offset, end-offset). 175 BitMap::idx_t start_offset = heapWordToOffset(start_addr); 176 BitMap::idx_t end_offset = heapWordToOffset(end_addr); 177 178 start_offset = _bm.get_next_one_offset(start_offset, end_offset); 179 while (start_offset < end_offset) { 180 if (!cl->do_bit(start_offset)) { 181 return false; 182 } 183 HeapWord* next_addr = MIN2(nextObject(offsetToHeapWord(start_offset)), end_addr); 184 BitMap::idx_t next_offset = heapWordToOffset(next_addr); 185 start_offset = _bm.get_next_one_offset(next_offset, end_offset); 186 } 187 } 188 return true; 189 } 190 191 inline bool CMBitMapRO::iterate(BitMapClosure* cl) { 192 MemRegion mr(startWord(), sizeInWords()); 193 return iterate(cl, mr); 194 } 195 196 #define check_mark(addr) \ 197 assert(_bmStartWord <= (addr) && (addr) < (_bmStartWord + _bmWordSize), \ 198 "outside underlying space?"); \ 199 assert(G1CollectedHeap::heap()->is_in_exact(addr), \ 200 err_msg("Trying to access not available bitmap "PTR_FORMAT \ 201 " corresponding to "PTR_FORMAT" (%u)", \ 202 p2i(this), p2i(addr), G1CollectedHeap::heap()->addr_to_region(addr))); 203 204 inline void CMBitMap::mark(HeapWord* addr) { 205 check_mark(addr); 206 _bm.set_bit(heapWordToOffset(addr)); 207 } 208 209 inline void CMBitMap::clear(HeapWord* addr) { 210 check_mark(addr); 211 _bm.clear_bit(heapWordToOffset(addr)); 212 } 213 214 inline bool CMBitMap::parMark(HeapWord* addr) { 215 check_mark(addr); 216 return _bm.par_set_bit(heapWordToOffset(addr)); 217 } 218 219 inline bool CMBitMap::parClear(HeapWord* addr) { 220 check_mark(addr); 221 return _bm.par_clear_bit(heapWordToOffset(addr)); 222 } 223 224 #undef check_mark 225 226 inline void CMTask::push(oop obj) { 227 HeapWord* objAddr = (HeapWord*) obj; 228 assert(_g1h->is_in_g1_reserved(objAddr), "invariant"); 229 assert(!_g1h->is_on_master_free_list( 230 _g1h->heap_region_containing((HeapWord*) objAddr)), "invariant"); 231 assert(!_g1h->is_obj_ill(obj), "invariant"); 232 assert(_nextMarkBitMap->isMarked(objAddr), "invariant"); 233 234 if (_cm->verbose_high()) { 235 gclog_or_tty->print_cr("[%u] pushing " PTR_FORMAT, _worker_id, p2i((void*) obj)); 236 } 237 238 if (!_task_queue->push(obj)) { 239 // The local task queue looks full. We need to push some entries 240 // to the global stack. 241 242 if (_cm->verbose_medium()) { 243 gclog_or_tty->print_cr("[%u] task queue overflow, " 244 "moving entries to the global stack", 245 _worker_id); 246 } 247 move_entries_to_global_stack(); 248 249 // this should succeed since, even if we overflow the global 250 // stack, we should have definitely removed some entries from the 251 // local queue. So, there must be space on it. 252 bool success = _task_queue->push(obj); 253 assert(success, "invariant"); 254 } 255 256 statsOnly( size_t tmp_size = (size_t)_task_queue->size(); 257 if (tmp_size > _local_max_size) { 258 _local_max_size = tmp_size; 259 } 260 ++_local_pushes ); 261 } 262 263 inline bool CMTask::is_below_finger(oop obj, HeapWord* global_finger) const { 264 // If obj is above the global finger, then the mark bitmap scan 265 // will find it later, and no push is needed. Similarly, if we have 266 // a current region and obj is between the local finger and the 267 // end of the current region, then no push is needed. The tradeoff 268 // of checking both vs only checking the global finger is that the 269 // local check will be more accurate and so result in fewer pushes, 270 // but may also be a little slower. 271 HeapWord* objAddr = (HeapWord*)obj; 272 if (_finger != NULL) { 273 // We have a current region. 274 275 // Finger and region values are all NULL or all non-NULL. We 276 // use _finger to check since we immediately use its value. 277 assert(_curr_region != NULL, "invariant"); 278 assert(_region_limit != NULL, "invariant"); 279 assert(_region_limit <= global_finger, "invariant"); 280 281 // True if obj is less than the local finger, or is between 282 // the region limit and the global finger. 283 if (objAddr < _finger) { 284 return true; 285 } else if (objAddr < _region_limit) { 286 return false; 287 } // Else check global finger. 288 } 289 // Check global finger. 290 return objAddr < global_finger; 291 } 292 293 inline void CMTask::make_reference_grey(oop obj, HeapRegion* hr) { 294 if (_cm->par_mark_and_count(obj, hr, _marked_bytes_array, _card_bm)) { 295 296 if (_cm->verbose_high()) { 297 gclog_or_tty->print_cr("[%u] marked object " PTR_FORMAT, 298 _worker_id, p2i(obj)); 299 } 300 301 // No OrderAccess:store_load() is needed. It is implicit in the 302 // CAS done in CMBitMap::parMark() call in the routine above. 303 HeapWord* global_finger = _cm->finger(); 304 305 // We only need to push a newly grey object on the mark 306 // stack if it is in a section of memory the mark bitmap 307 // scan has already examined. Mark bitmap scanning 308 // maintains progress "fingers" for determining that. 309 // 310 // Notice that the global finger might be moving forward 311 // concurrently. This is not a problem. In the worst case, we 312 // mark the object while it is above the global finger and, by 313 // the time we read the global finger, it has moved forward 314 // past this object. In this case, the object will probably 315 // be visited when a task is scanning the region and will also 316 // be pushed on the stack. So, some duplicate work, but no 317 // correctness problems. 318 if (is_below_finger(obj, global_finger)) { 319 if (obj->is_typeArray()) { 320 // Immediately process arrays of primitive types, rather 321 // than pushing on the mark stack. This keeps us from 322 // adding humongous objects to the mark stack that might 323 // be reclaimed before the entry is processed - see 324 // selection of candidates for eager reclaim of humongous 325 // objects. The cost of the additional type test is 326 // mitigated by avoiding a trip through the mark stack, 327 // by only doing a bookkeeping update and avoiding the 328 // actual scan of the object - a typeArray contains no 329 // references, and the metadata is built-in. 330 process_grey_object<false>(obj); 331 } else { 332 if (_cm->verbose_high()) { 333 gclog_or_tty->print_cr("[%u] below a finger (local: " PTR_FORMAT 334 ", global: " PTR_FORMAT ") pushing " 335 PTR_FORMAT " on mark stack", 336 _worker_id, p2i(_finger), 337 p2i(global_finger), p2i(obj)); 338 } 339 push(obj); 340 } 341 } 342 } 343 } 344 345 inline void CMTask::deal_with_reference(oop obj) { 346 if (_cm->verbose_high()) { 347 gclog_or_tty->print_cr("[%u] we're dealing with reference = "PTR_FORMAT, 348 _worker_id, p2i((void*) obj)); 349 } 350 351 increment_refs_reached(); 352 353 HeapWord* objAddr = (HeapWord*) obj; 354 assert(obj->is_oop_or_null(true /* ignore mark word */), err_msg("Expected an oop or NULL at " PTR_FORMAT, p2i(obj))); 355 if (_g1h->is_in_g1_reserved(objAddr)) { 356 assert(obj != NULL, "null check is implicit"); 357 if (!_nextMarkBitMap->isMarked(objAddr)) { 358 // Only get the containing region if the object is not marked on the 359 // bitmap (otherwise, it's a waste of time since we won't do 360 // anything with it). 361 HeapRegion* hr = _g1h->heap_region_containing_raw(obj); 362 if (!hr->obj_allocated_since_next_marking(obj)) { 363 make_reference_grey(obj, hr); 364 } 365 } 366 } 367 } 368 369 inline void ConcurrentMark::markPrev(oop p) { 370 assert(!_prevMarkBitMap->isMarked((HeapWord*) p), "sanity"); 371 // Note we are overriding the read-only view of the prev map here, via 372 // the cast. 373 ((CMBitMap*)_prevMarkBitMap)->mark((HeapWord*) p); 374 } 375 376 inline void ConcurrentMark::grayRoot(oop obj, size_t word_size, 377 uint worker_id, HeapRegion* hr) { 378 assert(obj != NULL, "pre-condition"); 379 HeapWord* addr = (HeapWord*) obj; 380 if (hr == NULL) { 381 hr = _g1h->heap_region_containing_raw(addr); 382 } else { 383 assert(hr->is_in(addr), "pre-condition"); 384 } 385 assert(hr != NULL, "sanity"); 386 // Given that we're looking for a region that contains an object 387 // header it's impossible to get back a HC region. 388 assert(!hr->is_continues_humongous(), "sanity"); 389 390 // We cannot assert that word_size == obj->size() given that obj 391 // might not be in a consistent state (another thread might be in 392 // the process of copying it). So the best thing we can do is to 393 // assert that word_size is under an upper bound which is its 394 // containing region's capacity. 395 assert(word_size * HeapWordSize <= hr->capacity(), 396 err_msg("size: "SIZE_FORMAT" capacity: "SIZE_FORMAT" "HR_FORMAT, 397 word_size * HeapWordSize, hr->capacity(), 398 HR_FORMAT_PARAMS(hr))); 399 400 if (addr < hr->next_top_at_mark_start()) { 401 if (!_nextMarkBitMap->isMarked(addr)) { 402 par_mark_and_count(obj, word_size, hr, worker_id); 403 } 404 } 405 } 406 407 #endif // SHARE_VM_GC_G1_CONCURRENTMARK_INLINE_HPP