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 template<typename Fn> 170 inline void CMMarkStack::iterate(Fn fn) { 171 assert(_saved_index == _index, 172 err_msg("saved index: %d index: %d", _saved_index, _index)); 173 for (int i = 0; i < _index; ++i) { 174 fn(_base[i]); 175 } 176 } 177 178 // It scans an object and visits its children. 179 inline void CMTask::scan_object(oop obj) { process_grey_object<true>(obj); } 180 181 inline void CMTask::push(oop obj) { 182 HeapWord* objAddr = (HeapWord*) obj; 183 assert(_g1h->is_in_g1_reserved(objAddr), "invariant"); 184 assert(!_g1h->is_on_master_free_list( 185 _g1h->heap_region_containing((HeapWord*) objAddr)), "invariant"); 186 assert(!_g1h->is_obj_ill(obj), "invariant"); 187 assert(_nextMarkBitMap->isMarked(objAddr), "invariant"); 188 189 if (_cm->verbose_high()) { 190 gclog_or_tty->print_cr("[%u] pushing " PTR_FORMAT, _worker_id, p2i((void*) obj)); 191 } 192 193 if (!_task_queue->push(obj)) { 194 // The local task queue looks full. We need to push some entries 195 // to the global stack. 196 197 if (_cm->verbose_medium()) { 198 gclog_or_tty->print_cr("[%u] task queue overflow, " 199 "moving entries to the global stack", 200 _worker_id); 201 } 202 move_entries_to_global_stack(); 203 204 // this should succeed since, even if we overflow the global 205 // stack, we should have definitely removed some entries from the 206 // local queue. So, there must be space on it. 207 bool success = _task_queue->push(obj); 208 assert(success, "invariant"); 209 } 210 211 statsOnly( size_t tmp_size = (size_t)_task_queue->size(); 212 if (tmp_size > _local_max_size) { 213 _local_max_size = tmp_size; 214 } 215 ++_local_pushes ); 216 } 217 218 inline bool CMTask::is_below_finger(oop obj, HeapWord* global_finger) const { 219 // If obj is above the global finger, then the mark bitmap scan 220 // will find it later, and no push is needed. Similarly, if we have 221 // a current region and obj is between the local finger and the 222 // end of the current region, then no push is needed. The tradeoff 223 // of checking both vs only checking the global finger is that the 224 // local check will be more accurate and so result in fewer pushes, 225 // but may also be a little slower. 226 HeapWord* objAddr = (HeapWord*)obj; 227 if (_finger != NULL) { 228 // We have a current region. 229 230 // Finger and region values are all NULL or all non-NULL. We 231 // use _finger to check since we immediately use its value. 232 assert(_curr_region != NULL, "invariant"); 233 assert(_region_limit != NULL, "invariant"); 234 assert(_region_limit <= global_finger, "invariant"); 235 236 // True if obj is less than the local finger, or is between 237 // the region limit and the global finger. 238 if (objAddr < _finger) { 239 return true; 240 } else if (objAddr < _region_limit) { 241 return false; 242 } // Else check global finger. 243 } 244 // Check global finger. 245 return objAddr < global_finger; 246 } 247 248 template<bool scan> 249 inline void CMTask::process_grey_object(oop obj) { 250 assert(scan || obj->is_typeArray(), "Skipping scan of grey non-typeArray"); 251 assert(_nextMarkBitMap->isMarked((HeapWord*) obj), "invariant"); 252 253 if (_cm->verbose_high()) { 254 gclog_or_tty->print_cr("[%u] processing grey object " PTR_FORMAT, 255 _worker_id, p2i((void*) obj)); 256 } 257 258 size_t obj_size = obj->size(); 259 _words_scanned += obj_size; 260 261 if (scan) { 262 obj->oop_iterate(_cm_oop_closure); 263 } 264 statsOnly( ++_objs_scanned ); 265 check_limits(); 266 } 267 268 269 270 inline void CMTask::make_reference_grey(oop obj, HeapRegion* hr) { 271 if (_cm->par_mark_and_count(obj, hr, _marked_bytes_array, _card_bm)) { 272 273 if (_cm->verbose_high()) { 274 gclog_or_tty->print_cr("[%u] marked object " PTR_FORMAT, 275 _worker_id, p2i(obj)); 276 } 277 278 // No OrderAccess:store_load() is needed. It is implicit in the 279 // CAS done in CMBitMap::parMark() call in the routine above. 280 HeapWord* global_finger = _cm->finger(); 281 282 // We only need to push a newly grey object on the mark 283 // stack if it is in a section of memory the mark bitmap 284 // scan has already examined. Mark bitmap scanning 285 // maintains progress "fingers" for determining that. 286 // 287 // Notice that the global finger might be moving forward 288 // concurrently. This is not a problem. In the worst case, we 289 // mark the object while it is above the global finger and, by 290 // the time we read the global finger, it has moved forward 291 // past this object. In this case, the object will probably 292 // be visited when a task is scanning the region and will also 293 // be pushed on the stack. So, some duplicate work, but no 294 // correctness problems. 295 if (is_below_finger(obj, global_finger)) { 296 if (obj->is_typeArray()) { 297 // Immediately process arrays of primitive types, rather 298 // than pushing on the mark stack. This keeps us from 299 // adding humongous objects to the mark stack that might 300 // be reclaimed before the entry is processed - see 301 // selection of candidates for eager reclaim of humongous 302 // objects. The cost of the additional type test is 303 // mitigated by avoiding a trip through the mark stack, 304 // by only doing a bookkeeping update and avoiding the 305 // actual scan of the object - a typeArray contains no 306 // references, and the metadata is built-in. 307 process_grey_object<false>(obj); 308 } else { 309 if (_cm->verbose_high()) { 310 gclog_or_tty->print_cr("[%u] below a finger (local: " PTR_FORMAT 311 ", global: " PTR_FORMAT ") pushing " 312 PTR_FORMAT " on mark stack", 313 _worker_id, p2i(_finger), 314 p2i(global_finger), p2i(obj)); 315 } 316 push(obj); 317 } 318 } 319 } 320 } 321 322 inline void CMTask::deal_with_reference(oop obj) { 323 if (_cm->verbose_high()) { 324 gclog_or_tty->print_cr("[%u] we're dealing with reference = " PTR_FORMAT, 325 _worker_id, p2i((void*) obj)); 326 } 327 328 increment_refs_reached(); 329 330 HeapWord* objAddr = (HeapWord*) obj; 331 assert(obj->is_oop_or_null(true /* ignore mark word */), err_msg("Expected an oop or NULL at " PTR_FORMAT, p2i(obj))); 332 if (_g1h->is_in_g1_reserved(objAddr)) { 333 assert(obj != NULL, "null check is implicit"); 334 if (!_nextMarkBitMap->isMarked(objAddr)) { 335 // Only get the containing region if the object is not marked on the 336 // bitmap (otherwise, it's a waste of time since we won't do 337 // anything with it). 338 HeapRegion* hr = _g1h->heap_region_containing_raw(obj); 339 if (!hr->obj_allocated_since_next_marking(obj)) { 340 make_reference_grey(obj, hr); 341 } 342 } 343 } 344 } 345 346 inline void ConcurrentMark::markPrev(oop p) { 347 assert(!_prevMarkBitMap->isMarked((HeapWord*) p), "sanity"); 348 // Note we are overriding the read-only view of the prev map here, via 349 // the cast. 350 ((CMBitMap*)_prevMarkBitMap)->mark((HeapWord*) p); 351 } 352 353 inline void ConcurrentMark::grayRoot(oop obj, size_t word_size, 354 uint worker_id, HeapRegion* hr) { 355 assert(obj != NULL, "pre-condition"); 356 HeapWord* addr = (HeapWord*) obj; 357 if (hr == NULL) { 358 hr = _g1h->heap_region_containing_raw(addr); 359 } else { 360 assert(hr->is_in(addr), "pre-condition"); 361 } 362 assert(hr != NULL, "sanity"); 363 // Given that we're looking for a region that contains an object 364 // header it's impossible to get back a HC region. 365 assert(!hr->is_continues_humongous(), "sanity"); 366 367 // We cannot assert that word_size == obj->size() given that obj 368 // might not be in a consistent state (another thread might be in 369 // the process of copying it). So the best thing we can do is to 370 // assert that word_size is under an upper bound which is its 371 // containing region's capacity. 372 assert(word_size * HeapWordSize <= hr->capacity(), 373 err_msg("size: " SIZE_FORMAT " capacity: " SIZE_FORMAT " " HR_FORMAT, 374 word_size * HeapWordSize, hr->capacity(), 375 HR_FORMAT_PARAMS(hr))); 376 377 if (addr < hr->next_top_at_mark_start()) { 378 if (!_nextMarkBitMap->isMarked(addr)) { 379 par_mark_and_count(obj, word_size, hr, worker_id); 380 } 381 } 382 } 383 384 #endif // SHARE_VM_GC_G1_CONCURRENTMARK_INLINE_HPP