1.1 --- /dev/null Thu Jan 01 00:00:00 1970 +0000 1.2 +++ b/src/share/vm/gc_implementation/parNew/parCardTableModRefBS.cpp Wed Apr 27 01:25:04 2016 +0800 1.3 @@ -0,0 +1,483 @@ 1.4 +/* 1.5 + * Copyright (c) 2007, 2014, Oracle and/or its affiliates. All rights reserved. 1.6 + * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. 1.7 + * 1.8 + * This code is free software; you can redistribute it and/or modify it 1.9 + * under the terms of the GNU General Public License version 2 only, as 1.10 + * published by the Free Software Foundation. 1.11 + * 1.12 + * This code is distributed in the hope that it will be useful, but WITHOUT 1.13 + * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or 1.14 + * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License 1.15 + * version 2 for more details (a copy is included in the LICENSE file that 1.16 + * accompanied this code). 1.17 + * 1.18 + * You should have received a copy of the GNU General Public License version 1.19 + * 2 along with this work; if not, write to the Free Software Foundation, 1.20 + * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA. 1.21 + * 1.22 + * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA 1.23 + * or visit www.oracle.com if you need additional information or have any 1.24 + * questions. 1.25 + * 1.26 + */ 1.27 + 1.28 +#include "precompiled.hpp" 1.29 +#include "memory/allocation.inline.hpp" 1.30 +#include "memory/cardTableModRefBS.hpp" 1.31 +#include "memory/cardTableRS.hpp" 1.32 +#include "memory/sharedHeap.hpp" 1.33 +#include "memory/space.inline.hpp" 1.34 +#include "memory/universe.hpp" 1.35 +#include "oops/oop.inline.hpp" 1.36 +#include "runtime/java.hpp" 1.37 +#include "runtime/mutexLocker.hpp" 1.38 +#include "runtime/virtualspace.hpp" 1.39 +#include "runtime/vmThread.hpp" 1.40 + 1.41 +PRAGMA_FORMAT_MUTE_WARNINGS_FOR_GCC 1.42 + 1.43 +void CardTableModRefBS::non_clean_card_iterate_parallel_work(Space* sp, MemRegion mr, 1.44 + OopsInGenClosure* cl, 1.45 + CardTableRS* ct, 1.46 + int n_threads) { 1.47 + assert(n_threads > 0, "Error: expected n_threads > 0"); 1.48 + assert((n_threads == 1 && ParallelGCThreads == 0) || 1.49 + n_threads <= (int)ParallelGCThreads, 1.50 + "# worker threads != # requested!"); 1.51 + assert(!Thread::current()->is_VM_thread() || (n_threads == 1), "There is only 1 VM thread"); 1.52 + assert(UseDynamicNumberOfGCThreads || 1.53 + !FLAG_IS_DEFAULT(ParallelGCThreads) || 1.54 + n_threads == (int)ParallelGCThreads, 1.55 + "# worker threads != # requested!"); 1.56 + // Make sure the LNC array is valid for the space. 1.57 + jbyte** lowest_non_clean; 1.58 + uintptr_t lowest_non_clean_base_chunk_index; 1.59 + size_t lowest_non_clean_chunk_size; 1.60 + get_LNC_array_for_space(sp, lowest_non_clean, 1.61 + lowest_non_clean_base_chunk_index, 1.62 + lowest_non_clean_chunk_size); 1.63 + 1.64 + uint n_strides = n_threads * ParGCStridesPerThread; 1.65 + SequentialSubTasksDone* pst = sp->par_seq_tasks(); 1.66 + // Sets the condition for completion of the subtask (how many threads 1.67 + // need to finish in order to be done). 1.68 + pst->set_n_threads(n_threads); 1.69 + pst->set_n_tasks(n_strides); 1.70 + 1.71 + uint stride = 0; 1.72 + while (!pst->is_task_claimed(/* reference */ stride)) { 1.73 + process_stride(sp, mr, stride, n_strides, cl, ct, 1.74 + lowest_non_clean, 1.75 + lowest_non_clean_base_chunk_index, 1.76 + lowest_non_clean_chunk_size); 1.77 + } 1.78 + if (pst->all_tasks_completed()) { 1.79 + // Clear lowest_non_clean array for next time. 1.80 + intptr_t first_chunk_index = addr_to_chunk_index(mr.start()); 1.81 + uintptr_t last_chunk_index = addr_to_chunk_index(mr.last()); 1.82 + for (uintptr_t ch = first_chunk_index; ch <= last_chunk_index; ch++) { 1.83 + intptr_t ind = ch - lowest_non_clean_base_chunk_index; 1.84 + assert(0 <= ind && ind < (intptr_t)lowest_non_clean_chunk_size, 1.85 + "Bounds error"); 1.86 + lowest_non_clean[ind] = NULL; 1.87 + } 1.88 + } 1.89 +} 1.90 + 1.91 +void 1.92 +CardTableModRefBS:: 1.93 +process_stride(Space* sp, 1.94 + MemRegion used, 1.95 + jint stride, int n_strides, 1.96 + OopsInGenClosure* cl, 1.97 + CardTableRS* ct, 1.98 + jbyte** lowest_non_clean, 1.99 + uintptr_t lowest_non_clean_base_chunk_index, 1.100 + size_t lowest_non_clean_chunk_size) { 1.101 + // We go from higher to lower addresses here; it wouldn't help that much 1.102 + // because of the strided parallelism pattern used here. 1.103 + 1.104 + // Find the first card address of the first chunk in the stride that is 1.105 + // at least "bottom" of the used region. 1.106 + jbyte* start_card = byte_for(used.start()); 1.107 + jbyte* end_card = byte_after(used.last()); 1.108 + uintptr_t start_chunk = addr_to_chunk_index(used.start()); 1.109 + uintptr_t start_chunk_stride_num = start_chunk % n_strides; 1.110 + jbyte* chunk_card_start; 1.111 + 1.112 + if ((uintptr_t)stride >= start_chunk_stride_num) { 1.113 + chunk_card_start = (jbyte*)(start_card + 1.114 + (stride - start_chunk_stride_num) * 1.115 + ParGCCardsPerStrideChunk); 1.116 + } else { 1.117 + // Go ahead to the next chunk group boundary, then to the requested stride. 1.118 + chunk_card_start = (jbyte*)(start_card + 1.119 + (n_strides - start_chunk_stride_num + stride) * 1.120 + ParGCCardsPerStrideChunk); 1.121 + } 1.122 + 1.123 + while (chunk_card_start < end_card) { 1.124 + // Even though we go from lower to higher addresses below, the 1.125 + // strided parallelism can interleave the actual processing of the 1.126 + // dirty pages in various ways. For a specific chunk within this 1.127 + // stride, we take care to avoid double scanning or missing a card 1.128 + // by suitably initializing the "min_done" field in process_chunk_boundaries() 1.129 + // below, together with the dirty region extension accomplished in 1.130 + // DirtyCardToOopClosure::do_MemRegion(). 1.131 + jbyte* chunk_card_end = chunk_card_start + ParGCCardsPerStrideChunk; 1.132 + // Invariant: chunk_mr should be fully contained within the "used" region. 1.133 + MemRegion chunk_mr = MemRegion(addr_for(chunk_card_start), 1.134 + chunk_card_end >= end_card ? 1.135 + used.end() : addr_for(chunk_card_end)); 1.136 + assert(chunk_mr.word_size() > 0, "[chunk_card_start > used_end)"); 1.137 + assert(used.contains(chunk_mr), "chunk_mr should be subset of used"); 1.138 + 1.139 + DirtyCardToOopClosure* dcto_cl = sp->new_dcto_cl(cl, precision(), 1.140 + cl->gen_boundary()); 1.141 + ClearNoncleanCardWrapper clear_cl(dcto_cl, ct); 1.142 + 1.143 + 1.144 + // Process the chunk. 1.145 + process_chunk_boundaries(sp, 1.146 + dcto_cl, 1.147 + chunk_mr, 1.148 + used, 1.149 + lowest_non_clean, 1.150 + lowest_non_clean_base_chunk_index, 1.151 + lowest_non_clean_chunk_size); 1.152 + 1.153 + // We want the LNC array updates above in process_chunk_boundaries 1.154 + // to be visible before any of the card table value changes as a 1.155 + // result of the dirty card iteration below. 1.156 + OrderAccess::storestore(); 1.157 + 1.158 + // We do not call the non_clean_card_iterate_serial() version because 1.159 + // we want to clear the cards: clear_cl here does the work of finding 1.160 + // contiguous dirty ranges of cards to process and clear. 1.161 + clear_cl.do_MemRegion(chunk_mr); 1.162 + 1.163 + // Find the next chunk of the stride. 1.164 + chunk_card_start += ParGCCardsPerStrideChunk * n_strides; 1.165 + } 1.166 +} 1.167 + 1.168 + 1.169 +// If you want a talkative process_chunk_boundaries, 1.170 +// then #define NOISY(x) x 1.171 +#ifdef NOISY 1.172 +#error "Encountered a global preprocessor flag, NOISY, which might clash with local definition to follow" 1.173 +#else 1.174 +#define NOISY(x) 1.175 +#endif 1.176 + 1.177 +void 1.178 +CardTableModRefBS:: 1.179 +process_chunk_boundaries(Space* sp, 1.180 + DirtyCardToOopClosure* dcto_cl, 1.181 + MemRegion chunk_mr, 1.182 + MemRegion used, 1.183 + jbyte** lowest_non_clean, 1.184 + uintptr_t lowest_non_clean_base_chunk_index, 1.185 + size_t lowest_non_clean_chunk_size) 1.186 +{ 1.187 + // We must worry about non-array objects that cross chunk boundaries, 1.188 + // because such objects are both precisely and imprecisely marked: 1.189 + // .. if the head of such an object is dirty, the entire object 1.190 + // needs to be scanned, under the interpretation that this 1.191 + // was an imprecise mark 1.192 + // .. if the head of such an object is not dirty, we can assume 1.193 + // precise marking and it's efficient to scan just the dirty 1.194 + // cards. 1.195 + // In either case, each scanned reference must be scanned precisely 1.196 + // once so as to avoid cloning of a young referent. For efficiency, 1.197 + // our closures depend on this property and do not protect against 1.198 + // double scans. 1.199 + 1.200 + uintptr_t cur_chunk_index = addr_to_chunk_index(chunk_mr.start()); 1.201 + cur_chunk_index = cur_chunk_index - lowest_non_clean_base_chunk_index; 1.202 + 1.203 + NOISY(tty->print_cr("===========================================================================");) 1.204 + NOISY(tty->print_cr(" process_chunk_boundary: Called with [" PTR_FORMAT "," PTR_FORMAT ")", 1.205 + chunk_mr.start(), chunk_mr.end());) 1.206 + 1.207 + // First, set "our" lowest_non_clean entry, which would be 1.208 + // used by the thread scanning an adjoining left chunk with 1.209 + // a non-array object straddling the mutual boundary. 1.210 + // Find the object that spans our boundary, if one exists. 1.211 + // first_block is the block possibly straddling our left boundary. 1.212 + HeapWord* first_block = sp->block_start(chunk_mr.start()); 1.213 + assert((chunk_mr.start() != used.start()) || (first_block == chunk_mr.start()), 1.214 + "First chunk should always have a co-initial block"); 1.215 + // Does the block straddle the chunk's left boundary, and is it 1.216 + // a non-array object? 1.217 + if (first_block < chunk_mr.start() // first block straddles left bdry 1.218 + && sp->block_is_obj(first_block) // first block is an object 1.219 + && !(oop(first_block)->is_objArray() // first block is not an array (arrays are precisely dirtied) 1.220 + || oop(first_block)->is_typeArray())) { 1.221 + // Find our least non-clean card, so that a left neighbour 1.222 + // does not scan an object straddling the mutual boundary 1.223 + // too far to the right, and attempt to scan a portion of 1.224 + // that object twice. 1.225 + jbyte* first_dirty_card = NULL; 1.226 + jbyte* last_card_of_first_obj = 1.227 + byte_for(first_block + sp->block_size(first_block) - 1); 1.228 + jbyte* first_card_of_cur_chunk = byte_for(chunk_mr.start()); 1.229 + jbyte* last_card_of_cur_chunk = byte_for(chunk_mr.last()); 1.230 + jbyte* last_card_to_check = 1.231 + (jbyte*) MIN2((intptr_t) last_card_of_cur_chunk, 1.232 + (intptr_t) last_card_of_first_obj); 1.233 + // Note that this does not need to go beyond our last card 1.234 + // if our first object completely straddles this chunk. 1.235 + for (jbyte* cur = first_card_of_cur_chunk; 1.236 + cur <= last_card_to_check; cur++) { 1.237 + jbyte val = *cur; 1.238 + if (card_will_be_scanned(val)) { 1.239 + first_dirty_card = cur; break; 1.240 + } else { 1.241 + assert(!card_may_have_been_dirty(val), "Error"); 1.242 + } 1.243 + } 1.244 + if (first_dirty_card != NULL) { 1.245 + NOISY(tty->print_cr(" LNC: Found a dirty card at " PTR_FORMAT " in current chunk", 1.246 + first_dirty_card);) 1.247 + assert(0 <= cur_chunk_index && cur_chunk_index < lowest_non_clean_chunk_size, 1.248 + "Bounds error."); 1.249 + assert(lowest_non_clean[cur_chunk_index] == NULL, 1.250 + "Write exactly once : value should be stable hereafter for this round"); 1.251 + lowest_non_clean[cur_chunk_index] = first_dirty_card; 1.252 + } NOISY(else { 1.253 + tty->print_cr(" LNC: Found no dirty card in current chunk; leaving LNC entry NULL"); 1.254 + // In the future, we could have this thread look for a non-NULL value to copy from its 1.255 + // right neighbour (up to the end of the first object). 1.256 + if (last_card_of_cur_chunk < last_card_of_first_obj) { 1.257 + tty->print_cr(" LNC: BEWARE!!! first obj straddles past right end of chunk:\n" 1.258 + " might be efficient to get value from right neighbour?"); 1.259 + } 1.260 + }) 1.261 + } else { 1.262 + // In this case we can help our neighbour by just asking them 1.263 + // to stop at our first card (even though it may not be dirty). 1.264 + NOISY(tty->print_cr(" LNC: first block is not a non-array object; setting LNC to first card of current chunk");) 1.265 + assert(lowest_non_clean[cur_chunk_index] == NULL, "Write once : value should be stable hereafter"); 1.266 + jbyte* first_card_of_cur_chunk = byte_for(chunk_mr.start()); 1.267 + lowest_non_clean[cur_chunk_index] = first_card_of_cur_chunk; 1.268 + } 1.269 + NOISY(tty->print_cr(" process_chunk_boundary: lowest_non_clean[" INTPTR_FORMAT "] = " PTR_FORMAT 1.270 + " which corresponds to the heap address " PTR_FORMAT, 1.271 + cur_chunk_index, lowest_non_clean[cur_chunk_index], 1.272 + (lowest_non_clean[cur_chunk_index] != NULL) 1.273 + ? addr_for(lowest_non_clean[cur_chunk_index]) 1.274 + : NULL);) 1.275 + NOISY(tty->print_cr("---------------------------------------------------------------------------");) 1.276 + 1.277 + // Next, set our own max_to_do, which will strictly/exclusively bound 1.278 + // the highest address that we will scan past the right end of our chunk. 1.279 + HeapWord* max_to_do = NULL; 1.280 + if (chunk_mr.end() < used.end()) { 1.281 + // This is not the last chunk in the used region. 1.282 + // What is our last block? We check the first block of 1.283 + // the next (right) chunk rather than strictly check our last block 1.284 + // because it's potentially more efficient to do so. 1.285 + HeapWord* const last_block = sp->block_start(chunk_mr.end()); 1.286 + assert(last_block <= chunk_mr.end(), "In case this property changes."); 1.287 + if ((last_block == chunk_mr.end()) // our last block does not straddle boundary 1.288 + || !sp->block_is_obj(last_block) // last_block isn't an object 1.289 + || oop(last_block)->is_objArray() // last_block is an array (precisely marked) 1.290 + || oop(last_block)->is_typeArray()) { 1.291 + max_to_do = chunk_mr.end(); 1.292 + NOISY(tty->print_cr(" process_chunk_boundary: Last block on this card is not a non-array object;\n" 1.293 + " max_to_do left at " PTR_FORMAT, max_to_do);) 1.294 + } else { 1.295 + assert(last_block < chunk_mr.end(), "Tautology"); 1.296 + // It is a non-array object that straddles the right boundary of this chunk. 1.297 + // last_obj_card is the card corresponding to the start of the last object 1.298 + // in the chunk. Note that the last object may not start in 1.299 + // the chunk. 1.300 + jbyte* const last_obj_card = byte_for(last_block); 1.301 + const jbyte val = *last_obj_card; 1.302 + if (!card_will_be_scanned(val)) { 1.303 + assert(!card_may_have_been_dirty(val), "Error"); 1.304 + // The card containing the head is not dirty. Any marks on 1.305 + // subsequent cards still in this chunk must have been made 1.306 + // precisely; we can cap processing at the end of our chunk. 1.307 + max_to_do = chunk_mr.end(); 1.308 + NOISY(tty->print_cr(" process_chunk_boundary: Head of last object on this card is not dirty;\n" 1.309 + " max_to_do left at " PTR_FORMAT, 1.310 + max_to_do);) 1.311 + } else { 1.312 + // The last object must be considered dirty, and extends onto the 1.313 + // following chunk. Look for a dirty card in that chunk that will 1.314 + // bound our processing. 1.315 + jbyte* limit_card = NULL; 1.316 + const size_t last_block_size = sp->block_size(last_block); 1.317 + jbyte* const last_card_of_last_obj = 1.318 + byte_for(last_block + last_block_size - 1); 1.319 + jbyte* const first_card_of_next_chunk = byte_for(chunk_mr.end()); 1.320 + // This search potentially goes a long distance looking 1.321 + // for the next card that will be scanned, terminating 1.322 + // at the end of the last_block, if no earlier dirty card 1.323 + // is found. 1.324 + assert(byte_for(chunk_mr.end()) - byte_for(chunk_mr.start()) == ParGCCardsPerStrideChunk, 1.325 + "last card of next chunk may be wrong"); 1.326 + for (jbyte* cur = first_card_of_next_chunk; 1.327 + cur <= last_card_of_last_obj; cur++) { 1.328 + const jbyte val = *cur; 1.329 + if (card_will_be_scanned(val)) { 1.330 + NOISY(tty->print_cr(" Found a non-clean card " PTR_FORMAT " with value 0x%x", 1.331 + cur, (int)val);) 1.332 + limit_card = cur; break; 1.333 + } else { 1.334 + assert(!card_may_have_been_dirty(val), "Error: card can't be skipped"); 1.335 + } 1.336 + } 1.337 + if (limit_card != NULL) { 1.338 + max_to_do = addr_for(limit_card); 1.339 + assert(limit_card != NULL && max_to_do != NULL, "Error"); 1.340 + NOISY(tty->print_cr(" process_chunk_boundary: Found a dirty card at " PTR_FORMAT 1.341 + " max_to_do set at " PTR_FORMAT " which is before end of last block in chunk: " 1.342 + PTR_FORMAT " + " PTR_FORMAT " = " PTR_FORMAT, 1.343 + limit_card, max_to_do, last_block, last_block_size, (last_block+last_block_size));) 1.344 + } else { 1.345 + // The following is a pessimistic value, because it's possible 1.346 + // that a dirty card on a subsequent chunk has been cleared by 1.347 + // the time we get to look at it; we'll correct for that further below, 1.348 + // using the LNC array which records the least non-clean card 1.349 + // before cards were cleared in a particular chunk. 1.350 + limit_card = last_card_of_last_obj; 1.351 + max_to_do = last_block + last_block_size; 1.352 + assert(limit_card != NULL && max_to_do != NULL, "Error"); 1.353 + NOISY(tty->print_cr(" process_chunk_boundary: Found no dirty card before end of last block in chunk\n" 1.354 + " Setting limit_card to " PTR_FORMAT 1.355 + " and max_to_do " PTR_FORMAT " + " PTR_FORMAT " = " PTR_FORMAT, 1.356 + limit_card, last_block, last_block_size, max_to_do);) 1.357 + } 1.358 + assert(0 < cur_chunk_index+1 && cur_chunk_index+1 < lowest_non_clean_chunk_size, 1.359 + "Bounds error."); 1.360 + // It is possible that a dirty card for the last object may have been 1.361 + // cleared before we had a chance to examine it. In that case, the value 1.362 + // will have been logged in the LNC for that chunk. 1.363 + // We need to examine as many chunks to the right as this object 1.364 + // covers. However, we need to bound this checking to the largest 1.365 + // entry in the LNC array: this is because the heap may expand 1.366 + // after the LNC array has been created but before we reach this point, 1.367 + // and the last block in our chunk may have been expanded to include 1.368 + // the expansion delta (and possibly subsequently allocated from, so 1.369 + // it wouldn't be sufficient to check whether that last block was 1.370 + // or was not an object at this point). 1.371 + uintptr_t last_chunk_index_to_check = addr_to_chunk_index(last_block + last_block_size - 1) 1.372 + - lowest_non_clean_base_chunk_index; 1.373 + const uintptr_t last_chunk_index = addr_to_chunk_index(used.last()) 1.374 + - lowest_non_clean_base_chunk_index; 1.375 + if (last_chunk_index_to_check > last_chunk_index) { 1.376 + assert(last_block + last_block_size > used.end(), 1.377 + err_msg("Inconsistency detected: last_block [" PTR_FORMAT "," PTR_FORMAT "]" 1.378 + " does not exceed used.end() = " PTR_FORMAT "," 1.379 + " yet last_chunk_index_to_check " INTPTR_FORMAT 1.380 + " exceeds last_chunk_index " INTPTR_FORMAT, 1.381 + last_block, last_block + last_block_size, 1.382 + used.end(), 1.383 + last_chunk_index_to_check, last_chunk_index)); 1.384 + assert(sp->used_region().end() > used.end(), 1.385 + err_msg("Expansion did not happen: " 1.386 + "[" PTR_FORMAT "," PTR_FORMAT ") -> [" PTR_FORMAT "," PTR_FORMAT ")", 1.387 + sp->used_region().start(), sp->used_region().end(), used.start(), used.end())); 1.388 + NOISY(tty->print_cr(" process_chunk_boundary: heap expanded; explicitly bounding last_chunk");) 1.389 + last_chunk_index_to_check = last_chunk_index; 1.390 + } 1.391 + for (uintptr_t lnc_index = cur_chunk_index + 1; 1.392 + lnc_index <= last_chunk_index_to_check; 1.393 + lnc_index++) { 1.394 + jbyte* lnc_card = lowest_non_clean[lnc_index]; 1.395 + if (lnc_card != NULL) { 1.396 + // we can stop at the first non-NULL entry we find 1.397 + if (lnc_card <= limit_card) { 1.398 + NOISY(tty->print_cr(" process_chunk_boundary: LNC card " PTR_FORMAT " is lower than limit_card " PTR_FORMAT, 1.399 + " max_to_do will be lowered to " PTR_FORMAT " from " PTR_FORMAT, 1.400 + lnc_card, limit_card, addr_for(lnc_card), max_to_do);) 1.401 + limit_card = lnc_card; 1.402 + max_to_do = addr_for(limit_card); 1.403 + assert(limit_card != NULL && max_to_do != NULL, "Error"); 1.404 + } 1.405 + // In any case, we break now 1.406 + break; 1.407 + } // else continue to look for a non-NULL entry if any 1.408 + } 1.409 + assert(limit_card != NULL && max_to_do != NULL, "Error"); 1.410 + } 1.411 + assert(max_to_do != NULL, "OOPS 1 !"); 1.412 + } 1.413 + assert(max_to_do != NULL, "OOPS 2!"); 1.414 + } else { 1.415 + max_to_do = used.end(); 1.416 + NOISY(tty->print_cr(" process_chunk_boundary: Last chunk of this space;\n" 1.417 + " max_to_do left at " PTR_FORMAT, 1.418 + max_to_do);) 1.419 + } 1.420 + assert(max_to_do != NULL, "OOPS 3!"); 1.421 + // Now we can set the closure we're using so it doesn't to beyond 1.422 + // max_to_do. 1.423 + dcto_cl->set_min_done(max_to_do); 1.424 +#ifndef PRODUCT 1.425 + dcto_cl->set_last_bottom(max_to_do); 1.426 +#endif 1.427 + NOISY(tty->print_cr("===========================================================================\n");) 1.428 +} 1.429 + 1.430 +#undef NOISY 1.431 + 1.432 +void 1.433 +CardTableModRefBS:: 1.434 +get_LNC_array_for_space(Space* sp, 1.435 + jbyte**& lowest_non_clean, 1.436 + uintptr_t& lowest_non_clean_base_chunk_index, 1.437 + size_t& lowest_non_clean_chunk_size) { 1.438 + 1.439 + int i = find_covering_region_containing(sp->bottom()); 1.440 + MemRegion covered = _covered[i]; 1.441 + size_t n_chunks = chunks_to_cover(covered); 1.442 + 1.443 + // Only the first thread to obtain the lock will resize the 1.444 + // LNC array for the covered region. Any later expansion can't affect 1.445 + // the used_at_save_marks region. 1.446 + // (I observed a bug in which the first thread to execute this would 1.447 + // resize, and then it would cause "expand_and_allocate" that would 1.448 + // increase the number of chunks in the covered region. Then a second 1.449 + // thread would come and execute this, see that the size didn't match, 1.450 + // and free and allocate again. So the first thread would be using a 1.451 + // freed "_lowest_non_clean" array.) 1.452 + 1.453 + // Do a dirty read here. If we pass the conditional then take the rare 1.454 + // event lock and do the read again in case some other thread had already 1.455 + // succeeded and done the resize. 1.456 + int cur_collection = Universe::heap()->total_collections(); 1.457 + if (_last_LNC_resizing_collection[i] != cur_collection) { 1.458 + MutexLocker x(ParGCRareEvent_lock); 1.459 + if (_last_LNC_resizing_collection[i] != cur_collection) { 1.460 + if (_lowest_non_clean[i] == NULL || 1.461 + n_chunks != _lowest_non_clean_chunk_size[i]) { 1.462 + 1.463 + // Should we delete the old? 1.464 + if (_lowest_non_clean[i] != NULL) { 1.465 + assert(n_chunks != _lowest_non_clean_chunk_size[i], 1.466 + "logical consequence"); 1.467 + FREE_C_HEAP_ARRAY(CardPtr, _lowest_non_clean[i], mtGC); 1.468 + _lowest_non_clean[i] = NULL; 1.469 + } 1.470 + // Now allocate a new one if necessary. 1.471 + if (_lowest_non_clean[i] == NULL) { 1.472 + _lowest_non_clean[i] = NEW_C_HEAP_ARRAY(CardPtr, n_chunks, mtGC); 1.473 + _lowest_non_clean_chunk_size[i] = n_chunks; 1.474 + _lowest_non_clean_base_chunk_index[i] = addr_to_chunk_index(covered.start()); 1.475 + for (int j = 0; j < (int)n_chunks; j++) 1.476 + _lowest_non_clean[i][j] = NULL; 1.477 + } 1.478 + } 1.479 + _last_LNC_resizing_collection[i] = cur_collection; 1.480 + } 1.481 + } 1.482 + // In any case, now do the initialization. 1.483 + lowest_non_clean = _lowest_non_clean[i]; 1.484 + lowest_non_clean_base_chunk_index = _lowest_non_clean_base_chunk_index[i]; 1.485 + lowest_non_clean_chunk_size = _lowest_non_clean_chunk_size[i]; 1.486 +}