duke@435: /* drchase@6680: * Copyright (c) 2007, 2014, Oracle and/or its affiliates. All rights reserved. duke@435: * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. duke@435: * duke@435: * This code is free software; you can redistribute it and/or modify it duke@435: * under the terms of the GNU General Public License version 2 only, as duke@435: * published by the Free Software Foundation. duke@435: * duke@435: * This code is distributed in the hope that it will be useful, but WITHOUT duke@435: * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or duke@435: * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License duke@435: * version 2 for more details (a copy is included in the LICENSE file that duke@435: * accompanied this code). duke@435: * duke@435: * You should have received a copy of the GNU General Public License version duke@435: * 2 along with this work; if not, write to the Free Software Foundation, duke@435: * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA. duke@435: * trims@1907: * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA trims@1907: * or visit www.oracle.com if you need additional information or have any trims@1907: * questions. duke@435: * duke@435: */ duke@435: stefank@2314: #include "precompiled.hpp" stefank@2314: #include "memory/allocation.inline.hpp" stefank@2314: #include "memory/cardTableModRefBS.hpp" stefank@2314: #include "memory/cardTableRS.hpp" stefank@2314: #include "memory/sharedHeap.hpp" stefank@2314: #include "memory/space.inline.hpp" stefank@2314: #include "memory/universe.hpp" ysr@2889: #include "oops/oop.inline.hpp" stefank@2314: #include "runtime/java.hpp" stefank@2314: #include "runtime/mutexLocker.hpp" goetz@6911: #include "runtime/orderAccess.inline.hpp" stefank@2314: #include "runtime/virtualspace.hpp" jmasa@3294: #include "runtime/vmThread.hpp" duke@435: drchase@6680: PRAGMA_FORMAT_MUTE_WARNINGS_FOR_GCC drchase@6680: ysr@2819: void CardTableModRefBS::non_clean_card_iterate_parallel_work(Space* sp, MemRegion mr, ysr@2889: OopsInGenClosure* cl, ysr@2889: CardTableRS* ct, ysr@2819: int n_threads) { ysr@2819: assert(n_threads > 0, "Error: expected n_threads > 0"); ysr@2819: assert((n_threads == 1 && ParallelGCThreads == 0) || ysr@2819: n_threads <= (int)ParallelGCThreads, ysr@2819: "# worker threads != # requested!"); jmasa@3294: assert(!Thread::current()->is_VM_thread() || (n_threads == 1), "There is only 1 VM thread"); jmasa@3294: assert(UseDynamicNumberOfGCThreads || jmasa@3294: !FLAG_IS_DEFAULT(ParallelGCThreads) || jmasa@3294: n_threads == (int)ParallelGCThreads, jmasa@3294: "# worker threads != # requested!"); ysr@2819: // Make sure the LNC array is valid for the space. ysr@2819: jbyte** lowest_non_clean; ysr@2819: uintptr_t lowest_non_clean_base_chunk_index; ysr@2819: size_t lowest_non_clean_chunk_size; ysr@2819: get_LNC_array_for_space(sp, lowest_non_clean, ysr@2819: lowest_non_clean_base_chunk_index, ysr@2819: lowest_non_clean_chunk_size); duke@435: jmasa@3357: uint n_strides = n_threads * ParGCStridesPerThread; ysr@2819: SequentialSubTasksDone* pst = sp->par_seq_tasks(); jmasa@3294: // Sets the condition for completion of the subtask (how many threads jmasa@3294: // need to finish in order to be done). ysr@2819: pst->set_n_threads(n_threads); ysr@2819: pst->set_n_tasks(n_strides); duke@435: jmasa@3357: uint stride = 0; ysr@2819: while (!pst->is_task_claimed(/* reference */ stride)) { ysr@2889: process_stride(sp, mr, stride, n_strides, cl, ct, ysr@2819: lowest_non_clean, ysr@2819: lowest_non_clean_base_chunk_index, ysr@2819: lowest_non_clean_chunk_size); ysr@2819: } ysr@2819: if (pst->all_tasks_completed()) { ysr@2819: // Clear lowest_non_clean array for next time. ysr@2819: intptr_t first_chunk_index = addr_to_chunk_index(mr.start()); ysr@2819: uintptr_t last_chunk_index = addr_to_chunk_index(mr.last()); ysr@2819: for (uintptr_t ch = first_chunk_index; ch <= last_chunk_index; ch++) { ysr@2819: intptr_t ind = ch - lowest_non_clean_base_chunk_index; ysr@2819: assert(0 <= ind && ind < (intptr_t)lowest_non_clean_chunk_size, ysr@2819: "Bounds error"); ysr@2819: lowest_non_clean[ind] = NULL; duke@435: } duke@435: } duke@435: } duke@435: duke@435: void duke@435: CardTableModRefBS:: duke@435: process_stride(Space* sp, duke@435: MemRegion used, duke@435: jint stride, int n_strides, ysr@2889: OopsInGenClosure* cl, ysr@2889: CardTableRS* ct, duke@435: jbyte** lowest_non_clean, duke@435: uintptr_t lowest_non_clean_base_chunk_index, duke@435: size_t lowest_non_clean_chunk_size) { ysr@2889: // We go from higher to lower addresses here; it wouldn't help that much ysr@2889: // because of the strided parallelism pattern used here. duke@435: duke@435: // Find the first card address of the first chunk in the stride that is duke@435: // at least "bottom" of the used region. duke@435: jbyte* start_card = byte_for(used.start()); duke@435: jbyte* end_card = byte_after(used.last()); duke@435: uintptr_t start_chunk = addr_to_chunk_index(used.start()); duke@435: uintptr_t start_chunk_stride_num = start_chunk % n_strides; duke@435: jbyte* chunk_card_start; duke@435: duke@435: if ((uintptr_t)stride >= start_chunk_stride_num) { duke@435: chunk_card_start = (jbyte*)(start_card + duke@435: (stride - start_chunk_stride_num) * ysr@2889: ParGCCardsPerStrideChunk); duke@435: } else { duke@435: // Go ahead to the next chunk group boundary, then to the requested stride. duke@435: chunk_card_start = (jbyte*)(start_card + duke@435: (n_strides - start_chunk_stride_num + stride) * ysr@2889: ParGCCardsPerStrideChunk); duke@435: } duke@435: duke@435: while (chunk_card_start < end_card) { ysr@2889: // Even though we go from lower to higher addresses below, the ysr@2889: // strided parallelism can interleave the actual processing of the ysr@2889: // dirty pages in various ways. For a specific chunk within this ysr@2889: // stride, we take care to avoid double scanning or missing a card ysr@2889: // by suitably initializing the "min_done" field in process_chunk_boundaries() ysr@2889: // below, together with the dirty region extension accomplished in ysr@2889: // DirtyCardToOopClosure::do_MemRegion(). ysr@2889: jbyte* chunk_card_end = chunk_card_start + ParGCCardsPerStrideChunk; duke@435: // Invariant: chunk_mr should be fully contained within the "used" region. duke@435: MemRegion chunk_mr = MemRegion(addr_for(chunk_card_start), duke@435: chunk_card_end >= end_card ? duke@435: used.end() : addr_for(chunk_card_end)); duke@435: assert(chunk_mr.word_size() > 0, "[chunk_card_start > used_end)"); duke@435: assert(used.contains(chunk_mr), "chunk_mr should be subset of used"); duke@435: ysr@2889: DirtyCardToOopClosure* dcto_cl = sp->new_dcto_cl(cl, precision(), ysr@2889: cl->gen_boundary()); ysr@2889: ClearNoncleanCardWrapper clear_cl(dcto_cl, ct); ysr@2889: ysr@2889: duke@435: // Process the chunk. duke@435: process_chunk_boundaries(sp, duke@435: dcto_cl, duke@435: chunk_mr, duke@435: used, duke@435: lowest_non_clean, duke@435: lowest_non_clean_base_chunk_index, duke@435: lowest_non_clean_chunk_size); duke@435: ysr@2889: // We want the LNC array updates above in process_chunk_boundaries ysr@2889: // to be visible before any of the card table value changes as a ysr@2889: // result of the dirty card iteration below. ysr@2889: OrderAccess::storestore(); ysr@2889: ysr@2819: // We do not call the non_clean_card_iterate_serial() version because ysr@2889: // we want to clear the cards: clear_cl here does the work of finding ysr@2889: // contiguous dirty ranges of cards to process and clear. ysr@2889: clear_cl.do_MemRegion(chunk_mr); duke@435: duke@435: // Find the next chunk of the stride. ysr@2889: chunk_card_start += ParGCCardsPerStrideChunk * n_strides; duke@435: } duke@435: } duke@435: ysr@2889: ysr@2889: // If you want a talkative process_chunk_boundaries, ysr@2889: // then #define NOISY(x) x ysr@2889: #ifdef NOISY ysr@2889: #error "Encountered a global preprocessor flag, NOISY, which might clash with local definition to follow" ysr@2889: #else ysr@2889: #define NOISY(x) ysr@2889: #endif ysr@2889: duke@435: void duke@435: CardTableModRefBS:: duke@435: process_chunk_boundaries(Space* sp, duke@435: DirtyCardToOopClosure* dcto_cl, duke@435: MemRegion chunk_mr, duke@435: MemRegion used, duke@435: jbyte** lowest_non_clean, duke@435: uintptr_t lowest_non_clean_base_chunk_index, duke@435: size_t lowest_non_clean_chunk_size) duke@435: { ysr@2889: // We must worry about non-array objects that cross chunk boundaries, ysr@2889: // because such objects are both precisely and imprecisely marked: ysr@2889: // .. if the head of such an object is dirty, the entire object ysr@2889: // needs to be scanned, under the interpretation that this ysr@2889: // was an imprecise mark ysr@2889: // .. if the head of such an object is not dirty, we can assume ysr@2889: // precise marking and it's efficient to scan just the dirty ysr@2889: // cards. ysr@2889: // In either case, each scanned reference must be scanned precisely ysr@2889: // once so as to avoid cloning of a young referent. For efficiency, ysr@2889: // our closures depend on this property and do not protect against ysr@2889: // double scans. duke@435: duke@435: uintptr_t cur_chunk_index = addr_to_chunk_index(chunk_mr.start()); duke@435: cur_chunk_index = cur_chunk_index - lowest_non_clean_base_chunk_index; duke@435: ysr@2889: NOISY(tty->print_cr("===========================================================================");) ysr@2889: NOISY(tty->print_cr(" process_chunk_boundary: Called with [" PTR_FORMAT "," PTR_FORMAT ")", ysr@2889: chunk_mr.start(), chunk_mr.end());) ysr@2889: ysr@2889: // First, set "our" lowest_non_clean entry, which would be ysr@2889: // used by the thread scanning an adjoining left chunk with ysr@2889: // a non-array object straddling the mutual boundary. ysr@2889: // Find the object that spans our boundary, if one exists. ysr@2889: // first_block is the block possibly straddling our left boundary. ysr@2889: HeapWord* first_block = sp->block_start(chunk_mr.start()); ysr@2889: assert((chunk_mr.start() != used.start()) || (first_block == chunk_mr.start()), ysr@2889: "First chunk should always have a co-initial block"); ysr@2889: // Does the block straddle the chunk's left boundary, and is it ysr@2889: // a non-array object? ysr@2889: if (first_block < chunk_mr.start() // first block straddles left bdry ysr@2889: && sp->block_is_obj(first_block) // first block is an object ysr@2889: && !(oop(first_block)->is_objArray() // first block is not an array (arrays are precisely dirtied) ysr@2889: || oop(first_block)->is_typeArray())) { ysr@2889: // Find our least non-clean card, so that a left neighbour ysr@2889: // does not scan an object straddling the mutual boundary ysr@2889: // too far to the right, and attempt to scan a portion of ysr@2889: // that object twice. ysr@2889: jbyte* first_dirty_card = NULL; ysr@2889: jbyte* last_card_of_first_obj = ysr@2889: byte_for(first_block + sp->block_size(first_block) - 1); ysr@2889: jbyte* first_card_of_cur_chunk = byte_for(chunk_mr.start()); ysr@2889: jbyte* last_card_of_cur_chunk = byte_for(chunk_mr.last()); ysr@2889: jbyte* last_card_to_check = ysr@2889: (jbyte*) MIN2((intptr_t) last_card_of_cur_chunk, ysr@2889: (intptr_t) last_card_of_first_obj); ysr@2889: // Note that this does not need to go beyond our last card ysr@2889: // if our first object completely straddles this chunk. ysr@2889: for (jbyte* cur = first_card_of_cur_chunk; ysr@2889: cur <= last_card_to_check; cur++) { ysr@2889: jbyte val = *cur; ysr@2889: if (card_will_be_scanned(val)) { ysr@2889: first_dirty_card = cur; break; ysr@2889: } else { ysr@2889: assert(!card_may_have_been_dirty(val), "Error"); ysr@2889: } ysr@2889: } ysr@2889: if (first_dirty_card != NULL) { ysr@2889: NOISY(tty->print_cr(" LNC: Found a dirty card at " PTR_FORMAT " in current chunk", ysr@2889: first_dirty_card);) ysr@2889: assert(0 <= cur_chunk_index && cur_chunk_index < lowest_non_clean_chunk_size, ysr@2889: "Bounds error."); ysr@2889: assert(lowest_non_clean[cur_chunk_index] == NULL, ysr@2889: "Write exactly once : value should be stable hereafter for this round"); ysr@2889: lowest_non_clean[cur_chunk_index] = first_dirty_card; ysr@2889: } NOISY(else { ysr@2889: tty->print_cr(" LNC: Found no dirty card in current chunk; leaving LNC entry NULL"); ysr@2889: // In the future, we could have this thread look for a non-NULL value to copy from its ysr@2889: // right neighbour (up to the end of the first object). ysr@2889: if (last_card_of_cur_chunk < last_card_of_first_obj) { ysr@2889: tty->print_cr(" LNC: BEWARE!!! first obj straddles past right end of chunk:\n" ysr@2889: " might be efficient to get value from right neighbour?"); ysr@2889: } ysr@2889: }) ysr@2889: } else { ysr@2889: // In this case we can help our neighbour by just asking them ysr@2889: // to stop at our first card (even though it may not be dirty). ysr@2889: NOISY(tty->print_cr(" LNC: first block is not a non-array object; setting LNC to first card of current chunk");) ysr@2889: assert(lowest_non_clean[cur_chunk_index] == NULL, "Write once : value should be stable hereafter"); ysr@2889: jbyte* first_card_of_cur_chunk = byte_for(chunk_mr.start()); ysr@2889: lowest_non_clean[cur_chunk_index] = first_card_of_cur_chunk; ysr@2889: } ysr@2889: NOISY(tty->print_cr(" process_chunk_boundary: lowest_non_clean[" INTPTR_FORMAT "] = " PTR_FORMAT ysr@2889: " which corresponds to the heap address " PTR_FORMAT, ysr@2889: cur_chunk_index, lowest_non_clean[cur_chunk_index], ysr@2889: (lowest_non_clean[cur_chunk_index] != NULL) ysr@2889: ? addr_for(lowest_non_clean[cur_chunk_index]) ysr@2889: : NULL);) ysr@2889: NOISY(tty->print_cr("---------------------------------------------------------------------------");) ysr@2889: ysr@2889: // Next, set our own max_to_do, which will strictly/exclusively bound ysr@2889: // the highest address that we will scan past the right end of our chunk. ysr@2889: HeapWord* max_to_do = NULL; duke@435: if (chunk_mr.end() < used.end()) { ysr@2889: // This is not the last chunk in the used region. ysr@2889: // What is our last block? We check the first block of ysr@2889: // the next (right) chunk rather than strictly check our last block ysr@2889: // because it's potentially more efficient to do so. ysr@2889: HeapWord* const last_block = sp->block_start(chunk_mr.end()); duke@435: assert(last_block <= chunk_mr.end(), "In case this property changes."); ysr@2889: if ((last_block == chunk_mr.end()) // our last block does not straddle boundary ysr@2889: || !sp->block_is_obj(last_block) // last_block isn't an object ysr@2889: || oop(last_block)->is_objArray() // last_block is an array (precisely marked) ysr@2889: || oop(last_block)->is_typeArray()) { duke@435: max_to_do = chunk_mr.end(); ysr@2889: NOISY(tty->print_cr(" process_chunk_boundary: Last block on this card is not a non-array object;\n" ysr@2889: " max_to_do left at " PTR_FORMAT, max_to_do);) duke@435: } else { ysr@2889: assert(last_block < chunk_mr.end(), "Tautology"); ysr@2889: // It is a non-array object that straddles the right boundary of this chunk. duke@435: // last_obj_card is the card corresponding to the start of the last object duke@435: // in the chunk. Note that the last object may not start in duke@435: // the chunk. ysr@2889: jbyte* const last_obj_card = byte_for(last_block); ysr@2889: const jbyte val = *last_obj_card; ysr@2889: if (!card_will_be_scanned(val)) { ysr@2889: assert(!card_may_have_been_dirty(val), "Error"); ysr@2889: // The card containing the head is not dirty. Any marks on duke@435: // subsequent cards still in this chunk must have been made ysr@2889: // precisely; we can cap processing at the end of our chunk. duke@435: max_to_do = chunk_mr.end(); ysr@2889: NOISY(tty->print_cr(" process_chunk_boundary: Head of last object on this card is not dirty;\n" ysr@2889: " max_to_do left at " PTR_FORMAT, ysr@2889: max_to_do);) duke@435: } else { duke@435: // The last object must be considered dirty, and extends onto the duke@435: // following chunk. Look for a dirty card in that chunk that will duke@435: // bound our processing. duke@435: jbyte* limit_card = NULL; ysr@2889: const size_t last_block_size = sp->block_size(last_block); ysr@2889: jbyte* const last_card_of_last_obj = duke@435: byte_for(last_block + last_block_size - 1); ysr@2889: jbyte* const first_card_of_next_chunk = byte_for(chunk_mr.end()); duke@435: // This search potentially goes a long distance looking ysr@2889: // for the next card that will be scanned, terminating ysr@2889: // at the end of the last_block, if no earlier dirty card ysr@2889: // is found. ysr@2889: assert(byte_for(chunk_mr.end()) - byte_for(chunk_mr.start()) == ParGCCardsPerStrideChunk, ysr@2889: "last card of next chunk may be wrong"); duke@435: for (jbyte* cur = first_card_of_next_chunk; ysr@2889: cur <= last_card_of_last_obj; cur++) { ysr@2889: const jbyte val = *cur; ysr@2889: if (card_will_be_scanned(val)) { ysr@2889: NOISY(tty->print_cr(" Found a non-clean card " PTR_FORMAT " with value 0x%x", ysr@2889: cur, (int)val);) duke@435: limit_card = cur; break; ysr@2889: } else { ysr@2889: assert(!card_may_have_been_dirty(val), "Error: card can't be skipped"); duke@435: } duke@435: } ysr@2889: if (limit_card != NULL) { ysr@2889: max_to_do = addr_for(limit_card); ysr@2889: assert(limit_card != NULL && max_to_do != NULL, "Error"); ysr@2889: NOISY(tty->print_cr(" process_chunk_boundary: Found a dirty card at " PTR_FORMAT ysr@2889: " max_to_do set at " PTR_FORMAT " which is before end of last block in chunk: " ysr@2889: PTR_FORMAT " + " PTR_FORMAT " = " PTR_FORMAT, ysr@2889: limit_card, max_to_do, last_block, last_block_size, (last_block+last_block_size));) ysr@2889: } else { ysr@2889: // The following is a pessimistic value, because it's possible ysr@2889: // that a dirty card on a subsequent chunk has been cleared by ysr@2889: // the time we get to look at it; we'll correct for that further below, ysr@2889: // using the LNC array which records the least non-clean card ysr@2889: // before cards were cleared in a particular chunk. ysr@2889: limit_card = last_card_of_last_obj; ysr@2889: max_to_do = last_block + last_block_size; ysr@2889: assert(limit_card != NULL && max_to_do != NULL, "Error"); ysr@2889: NOISY(tty->print_cr(" process_chunk_boundary: Found no dirty card before end of last block in chunk\n" ysr@2889: " Setting limit_card to " PTR_FORMAT ysr@2889: " and max_to_do " PTR_FORMAT " + " PTR_FORMAT " = " PTR_FORMAT, ysr@2889: limit_card, last_block, last_block_size, max_to_do);) ysr@2889: } ysr@2889: assert(0 < cur_chunk_index+1 && cur_chunk_index+1 < lowest_non_clean_chunk_size, duke@435: "Bounds error."); ysr@2889: // It is possible that a dirty card for the last object may have been ysr@2889: // cleared before we had a chance to examine it. In that case, the value ysr@2889: // will have been logged in the LNC for that chunk. ysr@2889: // We need to examine as many chunks to the right as this object ysr@2959: // covers. However, we need to bound this checking to the largest ysr@2959: // entry in the LNC array: this is because the heap may expand ysr@2959: // after the LNC array has been created but before we reach this point, ysr@2959: // and the last block in our chunk may have been expanded to include ysr@2959: // the expansion delta (and possibly subsequently allocated from, so ysr@2959: // it wouldn't be sufficient to check whether that last block was ysr@2959: // or was not an object at this point). ysr@2959: uintptr_t last_chunk_index_to_check = addr_to_chunk_index(last_block + last_block_size - 1) ysr@2959: - lowest_non_clean_base_chunk_index; ysr@2959: const uintptr_t last_chunk_index = addr_to_chunk_index(used.last()) ysr@2959: - lowest_non_clean_base_chunk_index; ysr@2959: if (last_chunk_index_to_check > last_chunk_index) { ysr@2959: assert(last_block + last_block_size > used.end(), ysr@2959: err_msg("Inconsistency detected: last_block [" PTR_FORMAT "," PTR_FORMAT "]" ysr@2959: " does not exceed used.end() = " PTR_FORMAT "," ysr@2959: " yet last_chunk_index_to_check " INTPTR_FORMAT ysr@2959: " exceeds last_chunk_index " INTPTR_FORMAT, mikael@4668: last_block, last_block + last_block_size, mikael@4668: used.end(), ysr@2959: last_chunk_index_to_check, last_chunk_index)); ysr@2959: assert(sp->used_region().end() > used.end(), ysr@2959: err_msg("Expansion did not happen: " ysr@2959: "[" PTR_FORMAT "," PTR_FORMAT ") -> [" PTR_FORMAT "," PTR_FORMAT ")", ysr@2959: sp->used_region().start(), sp->used_region().end(), used.start(), used.end())); ysr@2959: NOISY(tty->print_cr(" process_chunk_boundary: heap expanded; explicitly bounding last_chunk");) ysr@2959: last_chunk_index_to_check = last_chunk_index; ysr@2959: } ysr@2889: for (uintptr_t lnc_index = cur_chunk_index + 1; ysr@2889: lnc_index <= last_chunk_index_to_check; ysr@2889: lnc_index++) { ysr@2889: jbyte* lnc_card = lowest_non_clean[lnc_index]; ysr@2889: if (lnc_card != NULL) { ysr@2889: // we can stop at the first non-NULL entry we find ysr@2889: if (lnc_card <= limit_card) { ysr@2889: NOISY(tty->print_cr(" process_chunk_boundary: LNC card " PTR_FORMAT " is lower than limit_card " PTR_FORMAT, ysr@2889: " max_to_do will be lowered to " PTR_FORMAT " from " PTR_FORMAT, ysr@2889: lnc_card, limit_card, addr_for(lnc_card), max_to_do);) ysr@2889: limit_card = lnc_card; ysr@2889: max_to_do = addr_for(limit_card); ysr@2889: assert(limit_card != NULL && max_to_do != NULL, "Error"); ysr@2889: } ysr@2889: // In any case, we break now ysr@2889: break; ysr@2889: } // else continue to look for a non-NULL entry if any duke@435: } ysr@2889: assert(limit_card != NULL && max_to_do != NULL, "Error"); duke@435: } ysr@2889: assert(max_to_do != NULL, "OOPS 1 !"); duke@435: } ysr@2889: assert(max_to_do != NULL, "OOPS 2!"); duke@435: } else { duke@435: max_to_do = used.end(); ysr@2889: NOISY(tty->print_cr(" process_chunk_boundary: Last chunk of this space;\n" ysr@2889: " max_to_do left at " PTR_FORMAT, ysr@2889: max_to_do);) duke@435: } ysr@2889: assert(max_to_do != NULL, "OOPS 3!"); duke@435: // Now we can set the closure we're using so it doesn't to beyond duke@435: // max_to_do. duke@435: dcto_cl->set_min_done(max_to_do); duke@435: #ifndef PRODUCT duke@435: dcto_cl->set_last_bottom(max_to_do); duke@435: #endif ysr@2889: NOISY(tty->print_cr("===========================================================================\n");) ysr@2889: } duke@435: ysr@2889: #undef NOISY duke@435: duke@435: void duke@435: CardTableModRefBS:: duke@435: get_LNC_array_for_space(Space* sp, duke@435: jbyte**& lowest_non_clean, duke@435: uintptr_t& lowest_non_clean_base_chunk_index, duke@435: size_t& lowest_non_clean_chunk_size) { duke@435: duke@435: int i = find_covering_region_containing(sp->bottom()); duke@435: MemRegion covered = _covered[i]; duke@435: size_t n_chunks = chunks_to_cover(covered); duke@435: duke@435: // Only the first thread to obtain the lock will resize the duke@435: // LNC array for the covered region. Any later expansion can't affect duke@435: // the used_at_save_marks region. duke@435: // (I observed a bug in which the first thread to execute this would ysr@2889: // resize, and then it would cause "expand_and_allocate" that would ysr@2889: // increase the number of chunks in the covered region. Then a second duke@435: // thread would come and execute this, see that the size didn't match, duke@435: // and free and allocate again. So the first thread would be using a duke@435: // freed "_lowest_non_clean" array.) duke@435: duke@435: // Do a dirty read here. If we pass the conditional then take the rare duke@435: // event lock and do the read again in case some other thread had already duke@435: // succeeded and done the resize. duke@435: int cur_collection = Universe::heap()->total_collections(); duke@435: if (_last_LNC_resizing_collection[i] != cur_collection) { duke@435: MutexLocker x(ParGCRareEvent_lock); duke@435: if (_last_LNC_resizing_collection[i] != cur_collection) { duke@435: if (_lowest_non_clean[i] == NULL || duke@435: n_chunks != _lowest_non_clean_chunk_size[i]) { duke@435: duke@435: // Should we delete the old? duke@435: if (_lowest_non_clean[i] != NULL) { duke@435: assert(n_chunks != _lowest_non_clean_chunk_size[i], duke@435: "logical consequence"); zgu@3900: FREE_C_HEAP_ARRAY(CardPtr, _lowest_non_clean[i], mtGC); duke@435: _lowest_non_clean[i] = NULL; duke@435: } duke@435: // Now allocate a new one if necessary. duke@435: if (_lowest_non_clean[i] == NULL) { zgu@3900: _lowest_non_clean[i] = NEW_C_HEAP_ARRAY(CardPtr, n_chunks, mtGC); duke@435: _lowest_non_clean_chunk_size[i] = n_chunks; duke@435: _lowest_non_clean_base_chunk_index[i] = addr_to_chunk_index(covered.start()); duke@435: for (int j = 0; j < (int)n_chunks; j++) duke@435: _lowest_non_clean[i][j] = NULL; duke@435: } duke@435: } duke@435: _last_LNC_resizing_collection[i] = cur_collection; duke@435: } duke@435: } duke@435: // In any case, now do the initialization. duke@435: lowest_non_clean = _lowest_non_clean[i]; duke@435: lowest_non_clean_base_chunk_index = _lowest_non_clean_base_chunk_index[i]; duke@435: lowest_non_clean_chunk_size = _lowest_non_clean_chunk_size[i]; duke@435: }