1.1 --- /dev/null Thu Jan 01 00:00:00 1970 +0000
1.2 +++ b/src/share/vm/gc_implementation/parallelScavenge/psParallelCompact.cpp Wed Apr 27 01:25:04 2016 +0800
1.3 @@ -0,0 +1,3383 @@
1.4 +/*
1.5 + * Copyright (c) 2005, 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 "classfile/symbolTable.hpp"
1.30 +#include "classfile/systemDictionary.hpp"
1.31 +#include "code/codeCache.hpp"
1.32 +#include "gc_implementation/parallelScavenge/gcTaskManager.hpp"
1.33 +#include "gc_implementation/parallelScavenge/parallelScavengeHeap.inline.hpp"
1.34 +#include "gc_implementation/parallelScavenge/pcTasks.hpp"
1.35 +#include "gc_implementation/parallelScavenge/psAdaptiveSizePolicy.hpp"
1.36 +#include "gc_implementation/parallelScavenge/psCompactionManager.inline.hpp"
1.37 +#include "gc_implementation/parallelScavenge/psMarkSweep.hpp"
1.38 +#include "gc_implementation/parallelScavenge/psMarkSweepDecorator.hpp"
1.39 +#include "gc_implementation/parallelScavenge/psOldGen.hpp"
1.40 +#include "gc_implementation/parallelScavenge/psParallelCompact.hpp"
1.41 +#include "gc_implementation/parallelScavenge/psPromotionManager.inline.hpp"
1.42 +#include "gc_implementation/parallelScavenge/psScavenge.hpp"
1.43 +#include "gc_implementation/parallelScavenge/psYoungGen.hpp"
1.44 +#include "gc_implementation/shared/gcHeapSummary.hpp"
1.45 +#include "gc_implementation/shared/gcTimer.hpp"
1.46 +#include "gc_implementation/shared/gcTrace.hpp"
1.47 +#include "gc_implementation/shared/gcTraceTime.hpp"
1.48 +#include "gc_implementation/shared/isGCActiveMark.hpp"
1.49 +#include "gc_interface/gcCause.hpp"
1.50 +#include "memory/gcLocker.inline.hpp"
1.51 +#include "memory/referencePolicy.hpp"
1.52 +#include "memory/referenceProcessor.hpp"
1.53 +#include "oops/methodData.hpp"
1.54 +#include "oops/oop.inline.hpp"
1.55 +#include "oops/oop.pcgc.inline.hpp"
1.56 +#include "runtime/fprofiler.hpp"
1.57 +#include "runtime/safepoint.hpp"
1.58 +#include "runtime/vmThread.hpp"
1.59 +#include "services/management.hpp"
1.60 +#include "services/memoryService.hpp"
1.61 +#include "services/memTracker.hpp"
1.62 +#include "utilities/events.hpp"
1.63 +#include "utilities/stack.inline.hpp"
1.64 +
1.65 +#include <math.h>
1.66 +
1.67 +PRAGMA_FORMAT_MUTE_WARNINGS_FOR_GCC
1.68 +
1.69 +// All sizes are in HeapWords.
1.70 +const size_t ParallelCompactData::Log2RegionSize = 16; // 64K words
1.71 +const size_t ParallelCompactData::RegionSize = (size_t)1 << Log2RegionSize;
1.72 +const size_t ParallelCompactData::RegionSizeBytes =
1.73 + RegionSize << LogHeapWordSize;
1.74 +const size_t ParallelCompactData::RegionSizeOffsetMask = RegionSize - 1;
1.75 +const size_t ParallelCompactData::RegionAddrOffsetMask = RegionSizeBytes - 1;
1.76 +const size_t ParallelCompactData::RegionAddrMask = ~RegionAddrOffsetMask;
1.77 +
1.78 +const size_t ParallelCompactData::Log2BlockSize = 7; // 128 words
1.79 +const size_t ParallelCompactData::BlockSize = (size_t)1 << Log2BlockSize;
1.80 +const size_t ParallelCompactData::BlockSizeBytes =
1.81 + BlockSize << LogHeapWordSize;
1.82 +const size_t ParallelCompactData::BlockSizeOffsetMask = BlockSize - 1;
1.83 +const size_t ParallelCompactData::BlockAddrOffsetMask = BlockSizeBytes - 1;
1.84 +const size_t ParallelCompactData::BlockAddrMask = ~BlockAddrOffsetMask;
1.85 +
1.86 +const size_t ParallelCompactData::BlocksPerRegion = RegionSize / BlockSize;
1.87 +const size_t ParallelCompactData::Log2BlocksPerRegion =
1.88 + Log2RegionSize - Log2BlockSize;
1.89 +
1.90 +const ParallelCompactData::RegionData::region_sz_t
1.91 +ParallelCompactData::RegionData::dc_shift = 27;
1.92 +
1.93 +const ParallelCompactData::RegionData::region_sz_t
1.94 +ParallelCompactData::RegionData::dc_mask = ~0U << dc_shift;
1.95 +
1.96 +const ParallelCompactData::RegionData::region_sz_t
1.97 +ParallelCompactData::RegionData::dc_one = 0x1U << dc_shift;
1.98 +
1.99 +const ParallelCompactData::RegionData::region_sz_t
1.100 +ParallelCompactData::RegionData::los_mask = ~dc_mask;
1.101 +
1.102 +const ParallelCompactData::RegionData::region_sz_t
1.103 +ParallelCompactData::RegionData::dc_claimed = 0x8U << dc_shift;
1.104 +
1.105 +const ParallelCompactData::RegionData::region_sz_t
1.106 +ParallelCompactData::RegionData::dc_completed = 0xcU << dc_shift;
1.107 +
1.108 +SpaceInfo PSParallelCompact::_space_info[PSParallelCompact::last_space_id];
1.109 +bool PSParallelCompact::_print_phases = false;
1.110 +
1.111 +ReferenceProcessor* PSParallelCompact::_ref_processor = NULL;
1.112 +Klass* PSParallelCompact::_updated_int_array_klass_obj = NULL;
1.113 +
1.114 +double PSParallelCompact::_dwl_mean;
1.115 +double PSParallelCompact::_dwl_std_dev;
1.116 +double PSParallelCompact::_dwl_first_term;
1.117 +double PSParallelCompact::_dwl_adjustment;
1.118 +#ifdef ASSERT
1.119 +bool PSParallelCompact::_dwl_initialized = false;
1.120 +#endif // #ifdef ASSERT
1.121 +
1.122 +void SplitInfo::record(size_t src_region_idx, size_t partial_obj_size,
1.123 + HeapWord* destination)
1.124 +{
1.125 + assert(src_region_idx != 0, "invalid src_region_idx");
1.126 + assert(partial_obj_size != 0, "invalid partial_obj_size argument");
1.127 + assert(destination != NULL, "invalid destination argument");
1.128 +
1.129 + _src_region_idx = src_region_idx;
1.130 + _partial_obj_size = partial_obj_size;
1.131 + _destination = destination;
1.132 +
1.133 + // These fields may not be updated below, so make sure they're clear.
1.134 + assert(_dest_region_addr == NULL, "should have been cleared");
1.135 + assert(_first_src_addr == NULL, "should have been cleared");
1.136 +
1.137 + // Determine the number of destination regions for the partial object.
1.138 + HeapWord* const last_word = destination + partial_obj_size - 1;
1.139 + const ParallelCompactData& sd = PSParallelCompact::summary_data();
1.140 + HeapWord* const beg_region_addr = sd.region_align_down(destination);
1.141 + HeapWord* const end_region_addr = sd.region_align_down(last_word);
1.142 +
1.143 + if (beg_region_addr == end_region_addr) {
1.144 + // One destination region.
1.145 + _destination_count = 1;
1.146 + if (end_region_addr == destination) {
1.147 + // The destination falls on a region boundary, thus the first word of the
1.148 + // partial object will be the first word copied to the destination region.
1.149 + _dest_region_addr = end_region_addr;
1.150 + _first_src_addr = sd.region_to_addr(src_region_idx);
1.151 + }
1.152 + } else {
1.153 + // Two destination regions. When copied, the partial object will cross a
1.154 + // destination region boundary, so a word somewhere within the partial
1.155 + // object will be the first word copied to the second destination region.
1.156 + _destination_count = 2;
1.157 + _dest_region_addr = end_region_addr;
1.158 + const size_t ofs = pointer_delta(end_region_addr, destination);
1.159 + assert(ofs < _partial_obj_size, "sanity");
1.160 + _first_src_addr = sd.region_to_addr(src_region_idx) + ofs;
1.161 + }
1.162 +}
1.163 +
1.164 +void SplitInfo::clear()
1.165 +{
1.166 + _src_region_idx = 0;
1.167 + _partial_obj_size = 0;
1.168 + _destination = NULL;
1.169 + _destination_count = 0;
1.170 + _dest_region_addr = NULL;
1.171 + _first_src_addr = NULL;
1.172 + assert(!is_valid(), "sanity");
1.173 +}
1.174 +
1.175 +#ifdef ASSERT
1.176 +void SplitInfo::verify_clear()
1.177 +{
1.178 + assert(_src_region_idx == 0, "not clear");
1.179 + assert(_partial_obj_size == 0, "not clear");
1.180 + assert(_destination == NULL, "not clear");
1.181 + assert(_destination_count == 0, "not clear");
1.182 + assert(_dest_region_addr == NULL, "not clear");
1.183 + assert(_first_src_addr == NULL, "not clear");
1.184 +}
1.185 +#endif // #ifdef ASSERT
1.186 +
1.187 +
1.188 +void PSParallelCompact::print_on_error(outputStream* st) {
1.189 + _mark_bitmap.print_on_error(st);
1.190 +}
1.191 +
1.192 +#ifndef PRODUCT
1.193 +const char* PSParallelCompact::space_names[] = {
1.194 + "old ", "eden", "from", "to "
1.195 +};
1.196 +
1.197 +void PSParallelCompact::print_region_ranges()
1.198 +{
1.199 + tty->print_cr("space bottom top end new_top");
1.200 + tty->print_cr("------ ---------- ---------- ---------- ----------");
1.201 +
1.202 + for (unsigned int id = 0; id < last_space_id; ++id) {
1.203 + const MutableSpace* space = _space_info[id].space();
1.204 + tty->print_cr("%u %s "
1.205 + SIZE_FORMAT_W(10) " " SIZE_FORMAT_W(10) " "
1.206 + SIZE_FORMAT_W(10) " " SIZE_FORMAT_W(10) " ",
1.207 + id, space_names[id],
1.208 + summary_data().addr_to_region_idx(space->bottom()),
1.209 + summary_data().addr_to_region_idx(space->top()),
1.210 + summary_data().addr_to_region_idx(space->end()),
1.211 + summary_data().addr_to_region_idx(_space_info[id].new_top()));
1.212 + }
1.213 +}
1.214 +
1.215 +void
1.216 +print_generic_summary_region(size_t i, const ParallelCompactData::RegionData* c)
1.217 +{
1.218 +#define REGION_IDX_FORMAT SIZE_FORMAT_W(7)
1.219 +#define REGION_DATA_FORMAT SIZE_FORMAT_W(5)
1.220 +
1.221 + ParallelCompactData& sd = PSParallelCompact::summary_data();
1.222 + size_t dci = c->destination() ? sd.addr_to_region_idx(c->destination()) : 0;
1.223 + tty->print_cr(REGION_IDX_FORMAT " " PTR_FORMAT " "
1.224 + REGION_IDX_FORMAT " " PTR_FORMAT " "
1.225 + REGION_DATA_FORMAT " " REGION_DATA_FORMAT " "
1.226 + REGION_DATA_FORMAT " " REGION_IDX_FORMAT " %d",
1.227 + i, c->data_location(), dci, c->destination(),
1.228 + c->partial_obj_size(), c->live_obj_size(),
1.229 + c->data_size(), c->source_region(), c->destination_count());
1.230 +
1.231 +#undef REGION_IDX_FORMAT
1.232 +#undef REGION_DATA_FORMAT
1.233 +}
1.234 +
1.235 +void
1.236 +print_generic_summary_data(ParallelCompactData& summary_data,
1.237 + HeapWord* const beg_addr,
1.238 + HeapWord* const end_addr)
1.239 +{
1.240 + size_t total_words = 0;
1.241 + size_t i = summary_data.addr_to_region_idx(beg_addr);
1.242 + const size_t last = summary_data.addr_to_region_idx(end_addr);
1.243 + HeapWord* pdest = 0;
1.244 +
1.245 + while (i <= last) {
1.246 + ParallelCompactData::RegionData* c = summary_data.region(i);
1.247 + if (c->data_size() != 0 || c->destination() != pdest) {
1.248 + print_generic_summary_region(i, c);
1.249 + total_words += c->data_size();
1.250 + pdest = c->destination();
1.251 + }
1.252 + ++i;
1.253 + }
1.254 +
1.255 + tty->print_cr("summary_data_bytes=" SIZE_FORMAT, total_words * HeapWordSize);
1.256 +}
1.257 +
1.258 +void
1.259 +print_generic_summary_data(ParallelCompactData& summary_data,
1.260 + SpaceInfo* space_info)
1.261 +{
1.262 + for (unsigned int id = 0; id < PSParallelCompact::last_space_id; ++id) {
1.263 + const MutableSpace* space = space_info[id].space();
1.264 + print_generic_summary_data(summary_data, space->bottom(),
1.265 + MAX2(space->top(), space_info[id].new_top()));
1.266 + }
1.267 +}
1.268 +
1.269 +void
1.270 +print_initial_summary_region(size_t i,
1.271 + const ParallelCompactData::RegionData* c,
1.272 + bool newline = true)
1.273 +{
1.274 + tty->print(SIZE_FORMAT_W(5) " " PTR_FORMAT " "
1.275 + SIZE_FORMAT_W(5) " " SIZE_FORMAT_W(5) " "
1.276 + SIZE_FORMAT_W(5) " " SIZE_FORMAT_W(5) " %d",
1.277 + i, c->destination(),
1.278 + c->partial_obj_size(), c->live_obj_size(),
1.279 + c->data_size(), c->source_region(), c->destination_count());
1.280 + if (newline) tty->cr();
1.281 +}
1.282 +
1.283 +void
1.284 +print_initial_summary_data(ParallelCompactData& summary_data,
1.285 + const MutableSpace* space) {
1.286 + if (space->top() == space->bottom()) {
1.287 + return;
1.288 + }
1.289 +
1.290 + const size_t region_size = ParallelCompactData::RegionSize;
1.291 + typedef ParallelCompactData::RegionData RegionData;
1.292 + HeapWord* const top_aligned_up = summary_data.region_align_up(space->top());
1.293 + const size_t end_region = summary_data.addr_to_region_idx(top_aligned_up);
1.294 + const RegionData* c = summary_data.region(end_region - 1);
1.295 + HeapWord* end_addr = c->destination() + c->data_size();
1.296 + const size_t live_in_space = pointer_delta(end_addr, space->bottom());
1.297 +
1.298 + // Print (and count) the full regions at the beginning of the space.
1.299 + size_t full_region_count = 0;
1.300 + size_t i = summary_data.addr_to_region_idx(space->bottom());
1.301 + while (i < end_region && summary_data.region(i)->data_size() == region_size) {
1.302 + print_initial_summary_region(i, summary_data.region(i));
1.303 + ++full_region_count;
1.304 + ++i;
1.305 + }
1.306 +
1.307 + size_t live_to_right = live_in_space - full_region_count * region_size;
1.308 +
1.309 + double max_reclaimed_ratio = 0.0;
1.310 + size_t max_reclaimed_ratio_region = 0;
1.311 + size_t max_dead_to_right = 0;
1.312 + size_t max_live_to_right = 0;
1.313 +
1.314 + // Print the 'reclaimed ratio' for regions while there is something live in
1.315 + // the region or to the right of it. The remaining regions are empty (and
1.316 + // uninteresting), and computing the ratio will result in division by 0.
1.317 + while (i < end_region && live_to_right > 0) {
1.318 + c = summary_data.region(i);
1.319 + HeapWord* const region_addr = summary_data.region_to_addr(i);
1.320 + const size_t used_to_right = pointer_delta(space->top(), region_addr);
1.321 + const size_t dead_to_right = used_to_right - live_to_right;
1.322 + const double reclaimed_ratio = double(dead_to_right) / live_to_right;
1.323 +
1.324 + if (reclaimed_ratio > max_reclaimed_ratio) {
1.325 + max_reclaimed_ratio = reclaimed_ratio;
1.326 + max_reclaimed_ratio_region = i;
1.327 + max_dead_to_right = dead_to_right;
1.328 + max_live_to_right = live_to_right;
1.329 + }
1.330 +
1.331 + print_initial_summary_region(i, c, false);
1.332 + tty->print_cr(" %12.10f " SIZE_FORMAT_W(10) " " SIZE_FORMAT_W(10),
1.333 + reclaimed_ratio, dead_to_right, live_to_right);
1.334 +
1.335 + live_to_right -= c->data_size();
1.336 + ++i;
1.337 + }
1.338 +
1.339 + // Any remaining regions are empty. Print one more if there is one.
1.340 + if (i < end_region) {
1.341 + print_initial_summary_region(i, summary_data.region(i));
1.342 + }
1.343 +
1.344 + tty->print_cr("max: " SIZE_FORMAT_W(4) " d2r=" SIZE_FORMAT_W(10) " "
1.345 + "l2r=" SIZE_FORMAT_W(10) " max_ratio=%14.12f",
1.346 + max_reclaimed_ratio_region, max_dead_to_right,
1.347 + max_live_to_right, max_reclaimed_ratio);
1.348 +}
1.349 +
1.350 +void
1.351 +print_initial_summary_data(ParallelCompactData& summary_data,
1.352 + SpaceInfo* space_info) {
1.353 + unsigned int id = PSParallelCompact::old_space_id;
1.354 + const MutableSpace* space;
1.355 + do {
1.356 + space = space_info[id].space();
1.357 + print_initial_summary_data(summary_data, space);
1.358 + } while (++id < PSParallelCompact::eden_space_id);
1.359 +
1.360 + do {
1.361 + space = space_info[id].space();
1.362 + print_generic_summary_data(summary_data, space->bottom(), space->top());
1.363 + } while (++id < PSParallelCompact::last_space_id);
1.364 +}
1.365 +#endif // #ifndef PRODUCT
1.366 +
1.367 +#ifdef ASSERT
1.368 +size_t add_obj_count;
1.369 +size_t add_obj_size;
1.370 +size_t mark_bitmap_count;
1.371 +size_t mark_bitmap_size;
1.372 +#endif // #ifdef ASSERT
1.373 +
1.374 +ParallelCompactData::ParallelCompactData()
1.375 +{
1.376 + _region_start = 0;
1.377 +
1.378 + _region_vspace = 0;
1.379 + _reserved_byte_size = 0;
1.380 + _region_data = 0;
1.381 + _region_count = 0;
1.382 +
1.383 + _block_vspace = 0;
1.384 + _block_data = 0;
1.385 + _block_count = 0;
1.386 +}
1.387 +
1.388 +bool ParallelCompactData::initialize(MemRegion covered_region)
1.389 +{
1.390 + _region_start = covered_region.start();
1.391 + const size_t region_size = covered_region.word_size();
1.392 + DEBUG_ONLY(_region_end = _region_start + region_size;)
1.393 +
1.394 + assert(region_align_down(_region_start) == _region_start,
1.395 + "region start not aligned");
1.396 + assert((region_size & RegionSizeOffsetMask) == 0,
1.397 + "region size not a multiple of RegionSize");
1.398 +
1.399 + bool result = initialize_region_data(region_size) && initialize_block_data();
1.400 + return result;
1.401 +}
1.402 +
1.403 +PSVirtualSpace*
1.404 +ParallelCompactData::create_vspace(size_t count, size_t element_size)
1.405 +{
1.406 + const size_t raw_bytes = count * element_size;
1.407 + const size_t page_sz = os::page_size_for_region(raw_bytes, raw_bytes, 10);
1.408 + const size_t granularity = os::vm_allocation_granularity();
1.409 + _reserved_byte_size = align_size_up(raw_bytes, MAX2(page_sz, granularity));
1.410 +
1.411 + const size_t rs_align = page_sz == (size_t) os::vm_page_size() ? 0 :
1.412 + MAX2(page_sz, granularity);
1.413 + ReservedSpace rs(_reserved_byte_size, rs_align, rs_align > 0);
1.414 + os::trace_page_sizes("par compact", raw_bytes, raw_bytes, page_sz, rs.base(),
1.415 + rs.size());
1.416 +
1.417 + MemTracker::record_virtual_memory_type((address)rs.base(), mtGC);
1.418 +
1.419 + PSVirtualSpace* vspace = new PSVirtualSpace(rs, page_sz);
1.420 + if (vspace != 0) {
1.421 + if (vspace->expand_by(_reserved_byte_size)) {
1.422 + return vspace;
1.423 + }
1.424 + delete vspace;
1.425 + // Release memory reserved in the space.
1.426 + rs.release();
1.427 + }
1.428 +
1.429 + return 0;
1.430 +}
1.431 +
1.432 +bool ParallelCompactData::initialize_region_data(size_t region_size)
1.433 +{
1.434 + const size_t count = (region_size + RegionSizeOffsetMask) >> Log2RegionSize;
1.435 + _region_vspace = create_vspace(count, sizeof(RegionData));
1.436 + if (_region_vspace != 0) {
1.437 + _region_data = (RegionData*)_region_vspace->reserved_low_addr();
1.438 + _region_count = count;
1.439 + return true;
1.440 + }
1.441 + return false;
1.442 +}
1.443 +
1.444 +bool ParallelCompactData::initialize_block_data()
1.445 +{
1.446 + assert(_region_count != 0, "region data must be initialized first");
1.447 + const size_t count = _region_count << Log2BlocksPerRegion;
1.448 + _block_vspace = create_vspace(count, sizeof(BlockData));
1.449 + if (_block_vspace != 0) {
1.450 + _block_data = (BlockData*)_block_vspace->reserved_low_addr();
1.451 + _block_count = count;
1.452 + return true;
1.453 + }
1.454 + return false;
1.455 +}
1.456 +
1.457 +void ParallelCompactData::clear()
1.458 +{
1.459 + memset(_region_data, 0, _region_vspace->committed_size());
1.460 + memset(_block_data, 0, _block_vspace->committed_size());
1.461 +}
1.462 +
1.463 +void ParallelCompactData::clear_range(size_t beg_region, size_t end_region) {
1.464 + assert(beg_region <= _region_count, "beg_region out of range");
1.465 + assert(end_region <= _region_count, "end_region out of range");
1.466 + assert(RegionSize % BlockSize == 0, "RegionSize not a multiple of BlockSize");
1.467 +
1.468 + const size_t region_cnt = end_region - beg_region;
1.469 + memset(_region_data + beg_region, 0, region_cnt * sizeof(RegionData));
1.470 +
1.471 + const size_t beg_block = beg_region * BlocksPerRegion;
1.472 + const size_t block_cnt = region_cnt * BlocksPerRegion;
1.473 + memset(_block_data + beg_block, 0, block_cnt * sizeof(BlockData));
1.474 +}
1.475 +
1.476 +HeapWord* ParallelCompactData::partial_obj_end(size_t region_idx) const
1.477 +{
1.478 + const RegionData* cur_cp = region(region_idx);
1.479 + const RegionData* const end_cp = region(region_count() - 1);
1.480 +
1.481 + HeapWord* result = region_to_addr(region_idx);
1.482 + if (cur_cp < end_cp) {
1.483 + do {
1.484 + result += cur_cp->partial_obj_size();
1.485 + } while (cur_cp->partial_obj_size() == RegionSize && ++cur_cp < end_cp);
1.486 + }
1.487 + return result;
1.488 +}
1.489 +
1.490 +void ParallelCompactData::add_obj(HeapWord* addr, size_t len)
1.491 +{
1.492 + const size_t obj_ofs = pointer_delta(addr, _region_start);
1.493 + const size_t beg_region = obj_ofs >> Log2RegionSize;
1.494 + const size_t end_region = (obj_ofs + len - 1) >> Log2RegionSize;
1.495 +
1.496 + DEBUG_ONLY(Atomic::inc_ptr(&add_obj_count);)
1.497 + DEBUG_ONLY(Atomic::add_ptr(len, &add_obj_size);)
1.498 +
1.499 + if (beg_region == end_region) {
1.500 + // All in one region.
1.501 + _region_data[beg_region].add_live_obj(len);
1.502 + return;
1.503 + }
1.504 +
1.505 + // First region.
1.506 + const size_t beg_ofs = region_offset(addr);
1.507 + _region_data[beg_region].add_live_obj(RegionSize - beg_ofs);
1.508 +
1.509 + Klass* klass = ((oop)addr)->klass();
1.510 + // Middle regions--completely spanned by this object.
1.511 + for (size_t region = beg_region + 1; region < end_region; ++region) {
1.512 + _region_data[region].set_partial_obj_size(RegionSize);
1.513 + _region_data[region].set_partial_obj_addr(addr);
1.514 + }
1.515 +
1.516 + // Last region.
1.517 + const size_t end_ofs = region_offset(addr + len - 1);
1.518 + _region_data[end_region].set_partial_obj_size(end_ofs + 1);
1.519 + _region_data[end_region].set_partial_obj_addr(addr);
1.520 +}
1.521 +
1.522 +void
1.523 +ParallelCompactData::summarize_dense_prefix(HeapWord* beg, HeapWord* end)
1.524 +{
1.525 + assert(region_offset(beg) == 0, "not RegionSize aligned");
1.526 + assert(region_offset(end) == 0, "not RegionSize aligned");
1.527 +
1.528 + size_t cur_region = addr_to_region_idx(beg);
1.529 + const size_t end_region = addr_to_region_idx(end);
1.530 + HeapWord* addr = beg;
1.531 + while (cur_region < end_region) {
1.532 + _region_data[cur_region].set_destination(addr);
1.533 + _region_data[cur_region].set_destination_count(0);
1.534 + _region_data[cur_region].set_source_region(cur_region);
1.535 + _region_data[cur_region].set_data_location(addr);
1.536 +
1.537 + // Update live_obj_size so the region appears completely full.
1.538 + size_t live_size = RegionSize - _region_data[cur_region].partial_obj_size();
1.539 + _region_data[cur_region].set_live_obj_size(live_size);
1.540 +
1.541 + ++cur_region;
1.542 + addr += RegionSize;
1.543 + }
1.544 +}
1.545 +
1.546 +// Find the point at which a space can be split and, if necessary, record the
1.547 +// split point.
1.548 +//
1.549 +// If the current src region (which overflowed the destination space) doesn't
1.550 +// have a partial object, the split point is at the beginning of the current src
1.551 +// region (an "easy" split, no extra bookkeeping required).
1.552 +//
1.553 +// If the current src region has a partial object, the split point is in the
1.554 +// region where that partial object starts (call it the split_region). If
1.555 +// split_region has a partial object, then the split point is just after that
1.556 +// partial object (a "hard" split where we have to record the split data and
1.557 +// zero the partial_obj_size field). With a "hard" split, we know that the
1.558 +// partial_obj ends within split_region because the partial object that caused
1.559 +// the overflow starts in split_region. If split_region doesn't have a partial
1.560 +// obj, then the split is at the beginning of split_region (another "easy"
1.561 +// split).
1.562 +HeapWord*
1.563 +ParallelCompactData::summarize_split_space(size_t src_region,
1.564 + SplitInfo& split_info,
1.565 + HeapWord* destination,
1.566 + HeapWord* target_end,
1.567 + HeapWord** target_next)
1.568 +{
1.569 + assert(destination <= target_end, "sanity");
1.570 + assert(destination + _region_data[src_region].data_size() > target_end,
1.571 + "region should not fit into target space");
1.572 + assert(is_region_aligned(target_end), "sanity");
1.573 +
1.574 + size_t split_region = src_region;
1.575 + HeapWord* split_destination = destination;
1.576 + size_t partial_obj_size = _region_data[src_region].partial_obj_size();
1.577 +
1.578 + if (destination + partial_obj_size > target_end) {
1.579 + // The split point is just after the partial object (if any) in the
1.580 + // src_region that contains the start of the object that overflowed the
1.581 + // destination space.
1.582 + //
1.583 + // Find the start of the "overflow" object and set split_region to the
1.584 + // region containing it.
1.585 + HeapWord* const overflow_obj = _region_data[src_region].partial_obj_addr();
1.586 + split_region = addr_to_region_idx(overflow_obj);
1.587 +
1.588 + // Clear the source_region field of all destination regions whose first word
1.589 + // came from data after the split point (a non-null source_region field
1.590 + // implies a region must be filled).
1.591 + //
1.592 + // An alternative to the simple loop below: clear during post_compact(),
1.593 + // which uses memcpy instead of individual stores, and is easy to
1.594 + // parallelize. (The downside is that it clears the entire RegionData
1.595 + // object as opposed to just one field.)
1.596 + //
1.597 + // post_compact() would have to clear the summary data up to the highest
1.598 + // address that was written during the summary phase, which would be
1.599 + //
1.600 + // max(top, max(new_top, clear_top))
1.601 + //
1.602 + // where clear_top is a new field in SpaceInfo. Would have to set clear_top
1.603 + // to target_end.
1.604 + const RegionData* const sr = region(split_region);
1.605 + const size_t beg_idx =
1.606 + addr_to_region_idx(region_align_up(sr->destination() +
1.607 + sr->partial_obj_size()));
1.608 + const size_t end_idx = addr_to_region_idx(target_end);
1.609 +
1.610 + if (TraceParallelOldGCSummaryPhase) {
1.611 + gclog_or_tty->print_cr("split: clearing source_region field in ["
1.612 + SIZE_FORMAT ", " SIZE_FORMAT ")",
1.613 + beg_idx, end_idx);
1.614 + }
1.615 + for (size_t idx = beg_idx; idx < end_idx; ++idx) {
1.616 + _region_data[idx].set_source_region(0);
1.617 + }
1.618 +
1.619 + // Set split_destination and partial_obj_size to reflect the split region.
1.620 + split_destination = sr->destination();
1.621 + partial_obj_size = sr->partial_obj_size();
1.622 + }
1.623 +
1.624 + // The split is recorded only if a partial object extends onto the region.
1.625 + if (partial_obj_size != 0) {
1.626 + _region_data[split_region].set_partial_obj_size(0);
1.627 + split_info.record(split_region, partial_obj_size, split_destination);
1.628 + }
1.629 +
1.630 + // Setup the continuation addresses.
1.631 + *target_next = split_destination + partial_obj_size;
1.632 + HeapWord* const source_next = region_to_addr(split_region) + partial_obj_size;
1.633 +
1.634 + if (TraceParallelOldGCSummaryPhase) {
1.635 + const char * split_type = partial_obj_size == 0 ? "easy" : "hard";
1.636 + gclog_or_tty->print_cr("%s split: src=" PTR_FORMAT " src_c=" SIZE_FORMAT
1.637 + " pos=" SIZE_FORMAT,
1.638 + split_type, source_next, split_region,
1.639 + partial_obj_size);
1.640 + gclog_or_tty->print_cr("%s split: dst=" PTR_FORMAT " dst_c=" SIZE_FORMAT
1.641 + " tn=" PTR_FORMAT,
1.642 + split_type, split_destination,
1.643 + addr_to_region_idx(split_destination),
1.644 + *target_next);
1.645 +
1.646 + if (partial_obj_size != 0) {
1.647 + HeapWord* const po_beg = split_info.destination();
1.648 + HeapWord* const po_end = po_beg + split_info.partial_obj_size();
1.649 + gclog_or_tty->print_cr("%s split: "
1.650 + "po_beg=" PTR_FORMAT " " SIZE_FORMAT " "
1.651 + "po_end=" PTR_FORMAT " " SIZE_FORMAT,
1.652 + split_type,
1.653 + po_beg, addr_to_region_idx(po_beg),
1.654 + po_end, addr_to_region_idx(po_end));
1.655 + }
1.656 + }
1.657 +
1.658 + return source_next;
1.659 +}
1.660 +
1.661 +bool ParallelCompactData::summarize(SplitInfo& split_info,
1.662 + HeapWord* source_beg, HeapWord* source_end,
1.663 + HeapWord** source_next,
1.664 + HeapWord* target_beg, HeapWord* target_end,
1.665 + HeapWord** target_next)
1.666 +{
1.667 + if (TraceParallelOldGCSummaryPhase) {
1.668 + HeapWord* const source_next_val = source_next == NULL ? NULL : *source_next;
1.669 + tty->print_cr("sb=" PTR_FORMAT " se=" PTR_FORMAT " sn=" PTR_FORMAT
1.670 + "tb=" PTR_FORMAT " te=" PTR_FORMAT " tn=" PTR_FORMAT,
1.671 + source_beg, source_end, source_next_val,
1.672 + target_beg, target_end, *target_next);
1.673 + }
1.674 +
1.675 + size_t cur_region = addr_to_region_idx(source_beg);
1.676 + const size_t end_region = addr_to_region_idx(region_align_up(source_end));
1.677 +
1.678 + HeapWord *dest_addr = target_beg;
1.679 + while (cur_region < end_region) {
1.680 + // The destination must be set even if the region has no data.
1.681 + _region_data[cur_region].set_destination(dest_addr);
1.682 +
1.683 + size_t words = _region_data[cur_region].data_size();
1.684 + if (words > 0) {
1.685 + // If cur_region does not fit entirely into the target space, find a point
1.686 + // at which the source space can be 'split' so that part is copied to the
1.687 + // target space and the rest is copied elsewhere.
1.688 + if (dest_addr + words > target_end) {
1.689 + assert(source_next != NULL, "source_next is NULL when splitting");
1.690 + *source_next = summarize_split_space(cur_region, split_info, dest_addr,
1.691 + target_end, target_next);
1.692 + return false;
1.693 + }
1.694 +
1.695 + // Compute the destination_count for cur_region, and if necessary, update
1.696 + // source_region for a destination region. The source_region field is
1.697 + // updated if cur_region is the first (left-most) region to be copied to a
1.698 + // destination region.
1.699 + //
1.700 + // The destination_count calculation is a bit subtle. A region that has
1.701 + // data that compacts into itself does not count itself as a destination.
1.702 + // This maintains the invariant that a zero count means the region is
1.703 + // available and can be claimed and then filled.
1.704 + uint destination_count = 0;
1.705 + if (split_info.is_split(cur_region)) {
1.706 + // The current region has been split: the partial object will be copied
1.707 + // to one destination space and the remaining data will be copied to
1.708 + // another destination space. Adjust the initial destination_count and,
1.709 + // if necessary, set the source_region field if the partial object will
1.710 + // cross a destination region boundary.
1.711 + destination_count = split_info.destination_count();
1.712 + if (destination_count == 2) {
1.713 + size_t dest_idx = addr_to_region_idx(split_info.dest_region_addr());
1.714 + _region_data[dest_idx].set_source_region(cur_region);
1.715 + }
1.716 + }
1.717 +
1.718 + HeapWord* const last_addr = dest_addr + words - 1;
1.719 + const size_t dest_region_1 = addr_to_region_idx(dest_addr);
1.720 + const size_t dest_region_2 = addr_to_region_idx(last_addr);
1.721 +
1.722 + // Initially assume that the destination regions will be the same and
1.723 + // adjust the value below if necessary. Under this assumption, if
1.724 + // cur_region == dest_region_2, then cur_region will be compacted
1.725 + // completely into itself.
1.726 + destination_count += cur_region == dest_region_2 ? 0 : 1;
1.727 + if (dest_region_1 != dest_region_2) {
1.728 + // Destination regions differ; adjust destination_count.
1.729 + destination_count += 1;
1.730 + // Data from cur_region will be copied to the start of dest_region_2.
1.731 + _region_data[dest_region_2].set_source_region(cur_region);
1.732 + } else if (region_offset(dest_addr) == 0) {
1.733 + // Data from cur_region will be copied to the start of the destination
1.734 + // region.
1.735 + _region_data[dest_region_1].set_source_region(cur_region);
1.736 + }
1.737 +
1.738 + _region_data[cur_region].set_destination_count(destination_count);
1.739 + _region_data[cur_region].set_data_location(region_to_addr(cur_region));
1.740 + dest_addr += words;
1.741 + }
1.742 +
1.743 + ++cur_region;
1.744 + }
1.745 +
1.746 + *target_next = dest_addr;
1.747 + return true;
1.748 +}
1.749 +
1.750 +HeapWord* ParallelCompactData::calc_new_pointer(HeapWord* addr) {
1.751 + assert(addr != NULL, "Should detect NULL oop earlier");
1.752 + assert(PSParallelCompact::gc_heap()->is_in(addr), "not in heap");
1.753 + assert(PSParallelCompact::mark_bitmap()->is_marked(addr), "not marked");
1.754 +
1.755 + // Region covering the object.
1.756 + RegionData* const region_ptr = addr_to_region_ptr(addr);
1.757 + HeapWord* result = region_ptr->destination();
1.758 +
1.759 + // If the entire Region is live, the new location is region->destination + the
1.760 + // offset of the object within in the Region.
1.761 +
1.762 + // Run some performance tests to determine if this special case pays off. It
1.763 + // is worth it for pointers into the dense prefix. If the optimization to
1.764 + // avoid pointer updates in regions that only point to the dense prefix is
1.765 + // ever implemented, this should be revisited.
1.766 + if (region_ptr->data_size() == RegionSize) {
1.767 + result += region_offset(addr);
1.768 + return result;
1.769 + }
1.770 +
1.771 + // Otherwise, the new location is region->destination + block offset + the
1.772 + // number of live words in the Block that are (a) to the left of addr and (b)
1.773 + // due to objects that start in the Block.
1.774 +
1.775 + // Fill in the block table if necessary. This is unsynchronized, so multiple
1.776 + // threads may fill the block table for a region (harmless, since it is
1.777 + // idempotent).
1.778 + if (!region_ptr->blocks_filled()) {
1.779 + PSParallelCompact::fill_blocks(addr_to_region_idx(addr));
1.780 + region_ptr->set_blocks_filled();
1.781 + }
1.782 +
1.783 + HeapWord* const search_start = block_align_down(addr);
1.784 + const size_t block_offset = addr_to_block_ptr(addr)->offset();
1.785 +
1.786 + const ParMarkBitMap* bitmap = PSParallelCompact::mark_bitmap();
1.787 + const size_t live = bitmap->live_words_in_range(search_start, oop(addr));
1.788 + result += block_offset + live;
1.789 + DEBUG_ONLY(PSParallelCompact::check_new_location(addr, result));
1.790 + return result;
1.791 +}
1.792 +
1.793 +#ifdef ASSERT
1.794 +void ParallelCompactData::verify_clear(const PSVirtualSpace* vspace)
1.795 +{
1.796 + const size_t* const beg = (const size_t*)vspace->committed_low_addr();
1.797 + const size_t* const end = (const size_t*)vspace->committed_high_addr();
1.798 + for (const size_t* p = beg; p < end; ++p) {
1.799 + assert(*p == 0, "not zero");
1.800 + }
1.801 +}
1.802 +
1.803 +void ParallelCompactData::verify_clear()
1.804 +{
1.805 + verify_clear(_region_vspace);
1.806 + verify_clear(_block_vspace);
1.807 +}
1.808 +#endif // #ifdef ASSERT
1.809 +
1.810 +STWGCTimer PSParallelCompact::_gc_timer;
1.811 +ParallelOldTracer PSParallelCompact::_gc_tracer;
1.812 +elapsedTimer PSParallelCompact::_accumulated_time;
1.813 +unsigned int PSParallelCompact::_total_invocations = 0;
1.814 +unsigned int PSParallelCompact::_maximum_compaction_gc_num = 0;
1.815 +jlong PSParallelCompact::_time_of_last_gc = 0;
1.816 +CollectorCounters* PSParallelCompact::_counters = NULL;
1.817 +ParMarkBitMap PSParallelCompact::_mark_bitmap;
1.818 +ParallelCompactData PSParallelCompact::_summary_data;
1.819 +
1.820 +PSParallelCompact::IsAliveClosure PSParallelCompact::_is_alive_closure;
1.821 +
1.822 +bool PSParallelCompact::IsAliveClosure::do_object_b(oop p) { return mark_bitmap()->is_marked(p); }
1.823 +
1.824 +void PSParallelCompact::KeepAliveClosure::do_oop(oop* p) { PSParallelCompact::KeepAliveClosure::do_oop_work(p); }
1.825 +void PSParallelCompact::KeepAliveClosure::do_oop(narrowOop* p) { PSParallelCompact::KeepAliveClosure::do_oop_work(p); }
1.826 +
1.827 +PSParallelCompact::AdjustPointerClosure PSParallelCompact::_adjust_pointer_closure;
1.828 +PSParallelCompact::AdjustKlassClosure PSParallelCompact::_adjust_klass_closure;
1.829 +
1.830 +void PSParallelCompact::AdjustPointerClosure::do_oop(oop* p) { adjust_pointer(p); }
1.831 +void PSParallelCompact::AdjustPointerClosure::do_oop(narrowOop* p) { adjust_pointer(p); }
1.832 +
1.833 +void PSParallelCompact::FollowStackClosure::do_void() { _compaction_manager->follow_marking_stacks(); }
1.834 +
1.835 +void PSParallelCompact::MarkAndPushClosure::do_oop(oop* p) {
1.836 + mark_and_push(_compaction_manager, p);
1.837 +}
1.838 +void PSParallelCompact::MarkAndPushClosure::do_oop(narrowOop* p) { mark_and_push(_compaction_manager, p); }
1.839 +
1.840 +void PSParallelCompact::FollowKlassClosure::do_klass(Klass* klass) {
1.841 + klass->oops_do(_mark_and_push_closure);
1.842 +}
1.843 +void PSParallelCompact::AdjustKlassClosure::do_klass(Klass* klass) {
1.844 + klass->oops_do(&PSParallelCompact::_adjust_pointer_closure);
1.845 +}
1.846 +
1.847 +void PSParallelCompact::post_initialize() {
1.848 + ParallelScavengeHeap* heap = gc_heap();
1.849 + assert(heap->kind() == CollectedHeap::ParallelScavengeHeap, "Sanity");
1.850 +
1.851 + MemRegion mr = heap->reserved_region();
1.852 + _ref_processor =
1.853 + new ReferenceProcessor(mr, // span
1.854 + ParallelRefProcEnabled && (ParallelGCThreads > 1), // mt processing
1.855 + (int) ParallelGCThreads, // mt processing degree
1.856 + true, // mt discovery
1.857 + (int) ParallelGCThreads, // mt discovery degree
1.858 + true, // atomic_discovery
1.859 + &_is_alive_closure); // non-header is alive closure
1.860 + _counters = new CollectorCounters("PSParallelCompact", 1);
1.861 +
1.862 + // Initialize static fields in ParCompactionManager.
1.863 + ParCompactionManager::initialize(mark_bitmap());
1.864 +}
1.865 +
1.866 +bool PSParallelCompact::initialize() {
1.867 + ParallelScavengeHeap* heap = gc_heap();
1.868 + assert(heap->kind() == CollectedHeap::ParallelScavengeHeap, "Sanity");
1.869 + MemRegion mr = heap->reserved_region();
1.870 +
1.871 + // Was the old gen get allocated successfully?
1.872 + if (!heap->old_gen()->is_allocated()) {
1.873 + return false;
1.874 + }
1.875 +
1.876 + initialize_space_info();
1.877 + initialize_dead_wood_limiter();
1.878 +
1.879 + if (!_mark_bitmap.initialize(mr)) {
1.880 + vm_shutdown_during_initialization(
1.881 + err_msg("Unable to allocate " SIZE_FORMAT "KB bitmaps for parallel "
1.882 + "garbage collection for the requested " SIZE_FORMAT "KB heap.",
1.883 + _mark_bitmap.reserved_byte_size()/K, mr.byte_size()/K));
1.884 + return false;
1.885 + }
1.886 +
1.887 + if (!_summary_data.initialize(mr)) {
1.888 + vm_shutdown_during_initialization(
1.889 + err_msg("Unable to allocate " SIZE_FORMAT "KB card tables for parallel "
1.890 + "garbage collection for the requested " SIZE_FORMAT "KB heap.",
1.891 + _summary_data.reserved_byte_size()/K, mr.byte_size()/K));
1.892 + return false;
1.893 + }
1.894 +
1.895 + return true;
1.896 +}
1.897 +
1.898 +void PSParallelCompact::initialize_space_info()
1.899 +{
1.900 + memset(&_space_info, 0, sizeof(_space_info));
1.901 +
1.902 + ParallelScavengeHeap* heap = gc_heap();
1.903 + PSYoungGen* young_gen = heap->young_gen();
1.904 +
1.905 + _space_info[old_space_id].set_space(heap->old_gen()->object_space());
1.906 + _space_info[eden_space_id].set_space(young_gen->eden_space());
1.907 + _space_info[from_space_id].set_space(young_gen->from_space());
1.908 + _space_info[to_space_id].set_space(young_gen->to_space());
1.909 +
1.910 + _space_info[old_space_id].set_start_array(heap->old_gen()->start_array());
1.911 +}
1.912 +
1.913 +void PSParallelCompact::initialize_dead_wood_limiter()
1.914 +{
1.915 + const size_t max = 100;
1.916 + _dwl_mean = double(MIN2(ParallelOldDeadWoodLimiterMean, max)) / 100.0;
1.917 + _dwl_std_dev = double(MIN2(ParallelOldDeadWoodLimiterStdDev, max)) / 100.0;
1.918 + _dwl_first_term = 1.0 / (sqrt(2.0 * M_PI) * _dwl_std_dev);
1.919 + DEBUG_ONLY(_dwl_initialized = true;)
1.920 + _dwl_adjustment = normal_distribution(1.0);
1.921 +}
1.922 +
1.923 +// Simple class for storing info about the heap at the start of GC, to be used
1.924 +// after GC for comparison/printing.
1.925 +class PreGCValues {
1.926 +public:
1.927 + PreGCValues() { }
1.928 + PreGCValues(ParallelScavengeHeap* heap) { fill(heap); }
1.929 +
1.930 + void fill(ParallelScavengeHeap* heap) {
1.931 + _heap_used = heap->used();
1.932 + _young_gen_used = heap->young_gen()->used_in_bytes();
1.933 + _old_gen_used = heap->old_gen()->used_in_bytes();
1.934 + _metadata_used = MetaspaceAux::used_bytes();
1.935 + };
1.936 +
1.937 + size_t heap_used() const { return _heap_used; }
1.938 + size_t young_gen_used() const { return _young_gen_used; }
1.939 + size_t old_gen_used() const { return _old_gen_used; }
1.940 + size_t metadata_used() const { return _metadata_used; }
1.941 +
1.942 +private:
1.943 + size_t _heap_used;
1.944 + size_t _young_gen_used;
1.945 + size_t _old_gen_used;
1.946 + size_t _metadata_used;
1.947 +};
1.948 +
1.949 +void
1.950 +PSParallelCompact::clear_data_covering_space(SpaceId id)
1.951 +{
1.952 + // At this point, top is the value before GC, new_top() is the value that will
1.953 + // be set at the end of GC. The marking bitmap is cleared to top; nothing
1.954 + // should be marked above top. The summary data is cleared to the larger of
1.955 + // top & new_top.
1.956 + MutableSpace* const space = _space_info[id].space();
1.957 + HeapWord* const bot = space->bottom();
1.958 + HeapWord* const top = space->top();
1.959 + HeapWord* const max_top = MAX2(top, _space_info[id].new_top());
1.960 +
1.961 + const idx_t beg_bit = _mark_bitmap.addr_to_bit(bot);
1.962 + const idx_t end_bit = BitMap::word_align_up(_mark_bitmap.addr_to_bit(top));
1.963 + _mark_bitmap.clear_range(beg_bit, end_bit);
1.964 +
1.965 + const size_t beg_region = _summary_data.addr_to_region_idx(bot);
1.966 + const size_t end_region =
1.967 + _summary_data.addr_to_region_idx(_summary_data.region_align_up(max_top));
1.968 + _summary_data.clear_range(beg_region, end_region);
1.969 +
1.970 + // Clear the data used to 'split' regions.
1.971 + SplitInfo& split_info = _space_info[id].split_info();
1.972 + if (split_info.is_valid()) {
1.973 + split_info.clear();
1.974 + }
1.975 + DEBUG_ONLY(split_info.verify_clear();)
1.976 +}
1.977 +
1.978 +void PSParallelCompact::pre_compact(PreGCValues* pre_gc_values)
1.979 +{
1.980 + // Update the from & to space pointers in space_info, since they are swapped
1.981 + // at each young gen gc. Do the update unconditionally (even though a
1.982 + // promotion failure does not swap spaces) because an unknown number of minor
1.983 + // collections will have swapped the spaces an unknown number of times.
1.984 + GCTraceTime tm("pre compact", print_phases(), true, &_gc_timer);
1.985 + ParallelScavengeHeap* heap = gc_heap();
1.986 + _space_info[from_space_id].set_space(heap->young_gen()->from_space());
1.987 + _space_info[to_space_id].set_space(heap->young_gen()->to_space());
1.988 +
1.989 + pre_gc_values->fill(heap);
1.990 +
1.991 + DEBUG_ONLY(add_obj_count = add_obj_size = 0;)
1.992 + DEBUG_ONLY(mark_bitmap_count = mark_bitmap_size = 0;)
1.993 +
1.994 + // Increment the invocation count
1.995 + heap->increment_total_collections(true);
1.996 +
1.997 + // We need to track unique mark sweep invocations as well.
1.998 + _total_invocations++;
1.999 +
1.1000 + heap->print_heap_before_gc();
1.1001 + heap->trace_heap_before_gc(&_gc_tracer);
1.1002 +
1.1003 + // Fill in TLABs
1.1004 + heap->accumulate_statistics_all_tlabs();
1.1005 + heap->ensure_parsability(true); // retire TLABs
1.1006 +
1.1007 + if (VerifyBeforeGC && heap->total_collections() >= VerifyGCStartAt) {
1.1008 + HandleMark hm; // Discard invalid handles created during verification
1.1009 + Universe::verify(" VerifyBeforeGC:");
1.1010 + }
1.1011 +
1.1012 + // Verify object start arrays
1.1013 + if (VerifyObjectStartArray &&
1.1014 + VerifyBeforeGC) {
1.1015 + heap->old_gen()->verify_object_start_array();
1.1016 + }
1.1017 +
1.1018 + DEBUG_ONLY(mark_bitmap()->verify_clear();)
1.1019 + DEBUG_ONLY(summary_data().verify_clear();)
1.1020 +
1.1021 + // Have worker threads release resources the next time they run a task.
1.1022 + gc_task_manager()->release_all_resources();
1.1023 +}
1.1024 +
1.1025 +void PSParallelCompact::post_compact()
1.1026 +{
1.1027 + GCTraceTime tm("post compact", print_phases(), true, &_gc_timer);
1.1028 +
1.1029 + for (unsigned int id = old_space_id; id < last_space_id; ++id) {
1.1030 + // Clear the marking bitmap, summary data and split info.
1.1031 + clear_data_covering_space(SpaceId(id));
1.1032 + // Update top(). Must be done after clearing the bitmap and summary data.
1.1033 + _space_info[id].publish_new_top();
1.1034 + }
1.1035 +
1.1036 + MutableSpace* const eden_space = _space_info[eden_space_id].space();
1.1037 + MutableSpace* const from_space = _space_info[from_space_id].space();
1.1038 + MutableSpace* const to_space = _space_info[to_space_id].space();
1.1039 +
1.1040 + ParallelScavengeHeap* heap = gc_heap();
1.1041 + bool eden_empty = eden_space->is_empty();
1.1042 + if (!eden_empty) {
1.1043 + eden_empty = absorb_live_data_from_eden(heap->size_policy(),
1.1044 + heap->young_gen(), heap->old_gen());
1.1045 + }
1.1046 +
1.1047 + // Update heap occupancy information which is used as input to the soft ref
1.1048 + // clearing policy at the next gc.
1.1049 + Universe::update_heap_info_at_gc();
1.1050 +
1.1051 + bool young_gen_empty = eden_empty && from_space->is_empty() &&
1.1052 + to_space->is_empty();
1.1053 +
1.1054 + BarrierSet* bs = heap->barrier_set();
1.1055 + if (bs->is_a(BarrierSet::ModRef)) {
1.1056 + ModRefBarrierSet* modBS = (ModRefBarrierSet*)bs;
1.1057 + MemRegion old_mr = heap->old_gen()->reserved();
1.1058 +
1.1059 + if (young_gen_empty) {
1.1060 + modBS->clear(MemRegion(old_mr.start(), old_mr.end()));
1.1061 + } else {
1.1062 + modBS->invalidate(MemRegion(old_mr.start(), old_mr.end()));
1.1063 + }
1.1064 + }
1.1065 +
1.1066 + // Delete metaspaces for unloaded class loaders and clean up loader_data graph
1.1067 + ClassLoaderDataGraph::purge();
1.1068 + MetaspaceAux::verify_metrics();
1.1069 +
1.1070 + Threads::gc_epilogue();
1.1071 + CodeCache::gc_epilogue();
1.1072 + JvmtiExport::gc_epilogue();
1.1073 +
1.1074 + COMPILER2_PRESENT(DerivedPointerTable::update_pointers());
1.1075 +
1.1076 + ref_processor()->enqueue_discovered_references(NULL);
1.1077 +
1.1078 + if (ZapUnusedHeapArea) {
1.1079 + heap->gen_mangle_unused_area();
1.1080 + }
1.1081 +
1.1082 + // Update time of last GC
1.1083 + reset_millis_since_last_gc();
1.1084 +}
1.1085 +
1.1086 +HeapWord*
1.1087 +PSParallelCompact::compute_dense_prefix_via_density(const SpaceId id,
1.1088 + bool maximum_compaction)
1.1089 +{
1.1090 + const size_t region_size = ParallelCompactData::RegionSize;
1.1091 + const ParallelCompactData& sd = summary_data();
1.1092 +
1.1093 + const MutableSpace* const space = _space_info[id].space();
1.1094 + HeapWord* const top_aligned_up = sd.region_align_up(space->top());
1.1095 + const RegionData* const beg_cp = sd.addr_to_region_ptr(space->bottom());
1.1096 + const RegionData* const end_cp = sd.addr_to_region_ptr(top_aligned_up);
1.1097 +
1.1098 + // Skip full regions at the beginning of the space--they are necessarily part
1.1099 + // of the dense prefix.
1.1100 + size_t full_count = 0;
1.1101 + const RegionData* cp;
1.1102 + for (cp = beg_cp; cp < end_cp && cp->data_size() == region_size; ++cp) {
1.1103 + ++full_count;
1.1104 + }
1.1105 +
1.1106 + assert(total_invocations() >= _maximum_compaction_gc_num, "sanity");
1.1107 + const size_t gcs_since_max = total_invocations() - _maximum_compaction_gc_num;
1.1108 + const bool interval_ended = gcs_since_max > HeapMaximumCompactionInterval;
1.1109 + if (maximum_compaction || cp == end_cp || interval_ended) {
1.1110 + _maximum_compaction_gc_num = total_invocations();
1.1111 + return sd.region_to_addr(cp);
1.1112 + }
1.1113 +
1.1114 + HeapWord* const new_top = _space_info[id].new_top();
1.1115 + const size_t space_live = pointer_delta(new_top, space->bottom());
1.1116 + const size_t space_used = space->used_in_words();
1.1117 + const size_t space_capacity = space->capacity_in_words();
1.1118 +
1.1119 + const double cur_density = double(space_live) / space_capacity;
1.1120 + const double deadwood_density =
1.1121 + (1.0 - cur_density) * (1.0 - cur_density) * cur_density * cur_density;
1.1122 + const size_t deadwood_goal = size_t(space_capacity * deadwood_density);
1.1123 +
1.1124 + if (TraceParallelOldGCDensePrefix) {
1.1125 + tty->print_cr("cur_dens=%5.3f dw_dens=%5.3f dw_goal=" SIZE_FORMAT,
1.1126 + cur_density, deadwood_density, deadwood_goal);
1.1127 + tty->print_cr("space_live=" SIZE_FORMAT " " "space_used=" SIZE_FORMAT " "
1.1128 + "space_cap=" SIZE_FORMAT,
1.1129 + space_live, space_used,
1.1130 + space_capacity);
1.1131 + }
1.1132 +
1.1133 + // XXX - Use binary search?
1.1134 + HeapWord* dense_prefix = sd.region_to_addr(cp);
1.1135 + const RegionData* full_cp = cp;
1.1136 + const RegionData* const top_cp = sd.addr_to_region_ptr(space->top() - 1);
1.1137 + while (cp < end_cp) {
1.1138 + HeapWord* region_destination = cp->destination();
1.1139 + const size_t cur_deadwood = pointer_delta(dense_prefix, region_destination);
1.1140 + if (TraceParallelOldGCDensePrefix && Verbose) {
1.1141 + tty->print_cr("c#=" SIZE_FORMAT_W(4) " dst=" PTR_FORMAT " "
1.1142 + "dp=" SIZE_FORMAT_W(8) " " "cdw=" SIZE_FORMAT_W(8),
1.1143 + sd.region(cp), region_destination,
1.1144 + dense_prefix, cur_deadwood);
1.1145 + }
1.1146 +
1.1147 + if (cur_deadwood >= deadwood_goal) {
1.1148 + // Found the region that has the correct amount of deadwood to the left.
1.1149 + // This typically occurs after crossing a fairly sparse set of regions, so
1.1150 + // iterate backwards over those sparse regions, looking for the region
1.1151 + // that has the lowest density of live objects 'to the right.'
1.1152 + size_t space_to_left = sd.region(cp) * region_size;
1.1153 + size_t live_to_left = space_to_left - cur_deadwood;
1.1154 + size_t space_to_right = space_capacity - space_to_left;
1.1155 + size_t live_to_right = space_live - live_to_left;
1.1156 + double density_to_right = double(live_to_right) / space_to_right;
1.1157 + while (cp > full_cp) {
1.1158 + --cp;
1.1159 + const size_t prev_region_live_to_right = live_to_right -
1.1160 + cp->data_size();
1.1161 + const size_t prev_region_space_to_right = space_to_right + region_size;
1.1162 + double prev_region_density_to_right =
1.1163 + double(prev_region_live_to_right) / prev_region_space_to_right;
1.1164 + if (density_to_right <= prev_region_density_to_right) {
1.1165 + return dense_prefix;
1.1166 + }
1.1167 + if (TraceParallelOldGCDensePrefix && Verbose) {
1.1168 + tty->print_cr("backing up from c=" SIZE_FORMAT_W(4) " d2r=%10.8f "
1.1169 + "pc_d2r=%10.8f", sd.region(cp), density_to_right,
1.1170 + prev_region_density_to_right);
1.1171 + }
1.1172 + dense_prefix -= region_size;
1.1173 + live_to_right = prev_region_live_to_right;
1.1174 + space_to_right = prev_region_space_to_right;
1.1175 + density_to_right = prev_region_density_to_right;
1.1176 + }
1.1177 + return dense_prefix;
1.1178 + }
1.1179 +
1.1180 + dense_prefix += region_size;
1.1181 + ++cp;
1.1182 + }
1.1183 +
1.1184 + return dense_prefix;
1.1185 +}
1.1186 +
1.1187 +#ifndef PRODUCT
1.1188 +void PSParallelCompact::print_dense_prefix_stats(const char* const algorithm,
1.1189 + const SpaceId id,
1.1190 + const bool maximum_compaction,
1.1191 + HeapWord* const addr)
1.1192 +{
1.1193 + const size_t region_idx = summary_data().addr_to_region_idx(addr);
1.1194 + RegionData* const cp = summary_data().region(region_idx);
1.1195 + const MutableSpace* const space = _space_info[id].space();
1.1196 + HeapWord* const new_top = _space_info[id].new_top();
1.1197 +
1.1198 + const size_t space_live = pointer_delta(new_top, space->bottom());
1.1199 + const size_t dead_to_left = pointer_delta(addr, cp->destination());
1.1200 + const size_t space_cap = space->capacity_in_words();
1.1201 + const double dead_to_left_pct = double(dead_to_left) / space_cap;
1.1202 + const size_t live_to_right = new_top - cp->destination();
1.1203 + const size_t dead_to_right = space->top() - addr - live_to_right;
1.1204 +
1.1205 + tty->print_cr("%s=" PTR_FORMAT " dpc=" SIZE_FORMAT_W(5) " "
1.1206 + "spl=" SIZE_FORMAT " "
1.1207 + "d2l=" SIZE_FORMAT " d2l%%=%6.4f "
1.1208 + "d2r=" SIZE_FORMAT " l2r=" SIZE_FORMAT
1.1209 + " ratio=%10.8f",
1.1210 + algorithm, addr, region_idx,
1.1211 + space_live,
1.1212 + dead_to_left, dead_to_left_pct,
1.1213 + dead_to_right, live_to_right,
1.1214 + double(dead_to_right) / live_to_right);
1.1215 +}
1.1216 +#endif // #ifndef PRODUCT
1.1217 +
1.1218 +// Return a fraction indicating how much of the generation can be treated as
1.1219 +// "dead wood" (i.e., not reclaimed). The function uses a normal distribution
1.1220 +// based on the density of live objects in the generation to determine a limit,
1.1221 +// which is then adjusted so the return value is min_percent when the density is
1.1222 +// 1.
1.1223 +//
1.1224 +// The following table shows some return values for a different values of the
1.1225 +// standard deviation (ParallelOldDeadWoodLimiterStdDev); the mean is 0.5 and
1.1226 +// min_percent is 1.
1.1227 +//
1.1228 +// fraction allowed as dead wood
1.1229 +// -----------------------------------------------------------------
1.1230 +// density std_dev=70 std_dev=75 std_dev=80 std_dev=85 std_dev=90 std_dev=95
1.1231 +// ------- ---------- ---------- ---------- ---------- ---------- ----------
1.1232 +// 0.00000 0.01000000 0.01000000 0.01000000 0.01000000 0.01000000 0.01000000
1.1233 +// 0.05000 0.03193096 0.02836880 0.02550828 0.02319280 0.02130337 0.01974941
1.1234 +// 0.10000 0.05247504 0.04547452 0.03988045 0.03537016 0.03170171 0.02869272
1.1235 +// 0.15000 0.07135702 0.06111390 0.05296419 0.04641639 0.04110601 0.03676066
1.1236 +// 0.20000 0.08831616 0.07509618 0.06461766 0.05622444 0.04943437 0.04388975
1.1237 +// 0.25000 0.10311208 0.08724696 0.07471205 0.06469760 0.05661313 0.05002313
1.1238 +// 0.30000 0.11553050 0.09741183 0.08313394 0.07175114 0.06257797 0.05511132
1.1239 +// 0.35000 0.12538832 0.10545958 0.08978741 0.07731366 0.06727491 0.05911289
1.1240 +// 0.40000 0.13253818 0.11128511 0.09459590 0.08132834 0.07066107 0.06199500
1.1241 +// 0.45000 0.13687208 0.11481163 0.09750361 0.08375387 0.07270534 0.06373386
1.1242 +// 0.50000 0.13832410 0.11599237 0.09847664 0.08456518 0.07338887 0.06431510
1.1243 +// 0.55000 0.13687208 0.11481163 0.09750361 0.08375387 0.07270534 0.06373386
1.1244 +// 0.60000 0.13253818 0.11128511 0.09459590 0.08132834 0.07066107 0.06199500
1.1245 +// 0.65000 0.12538832 0.10545958 0.08978741 0.07731366 0.06727491 0.05911289
1.1246 +// 0.70000 0.11553050 0.09741183 0.08313394 0.07175114 0.06257797 0.05511132
1.1247 +// 0.75000 0.10311208 0.08724696 0.07471205 0.06469760 0.05661313 0.05002313
1.1248 +// 0.80000 0.08831616 0.07509618 0.06461766 0.05622444 0.04943437 0.04388975
1.1249 +// 0.85000 0.07135702 0.06111390 0.05296419 0.04641639 0.04110601 0.03676066
1.1250 +// 0.90000 0.05247504 0.04547452 0.03988045 0.03537016 0.03170171 0.02869272
1.1251 +// 0.95000 0.03193096 0.02836880 0.02550828 0.02319280 0.02130337 0.01974941
1.1252 +// 1.00000 0.01000000 0.01000000 0.01000000 0.01000000 0.01000000 0.01000000
1.1253 +
1.1254 +double PSParallelCompact::dead_wood_limiter(double density, size_t min_percent)
1.1255 +{
1.1256 + assert(_dwl_initialized, "uninitialized");
1.1257 +
1.1258 + // The raw limit is the value of the normal distribution at x = density.
1.1259 + const double raw_limit = normal_distribution(density);
1.1260 +
1.1261 + // Adjust the raw limit so it becomes the minimum when the density is 1.
1.1262 + //
1.1263 + // First subtract the adjustment value (which is simply the precomputed value
1.1264 + // normal_distribution(1.0)); this yields a value of 0 when the density is 1.
1.1265 + // Then add the minimum value, so the minimum is returned when the density is
1.1266 + // 1. Finally, prevent negative values, which occur when the mean is not 0.5.
1.1267 + const double min = double(min_percent) / 100.0;
1.1268 + const double limit = raw_limit - _dwl_adjustment + min;
1.1269 + return MAX2(limit, 0.0);
1.1270 +}
1.1271 +
1.1272 +ParallelCompactData::RegionData*
1.1273 +PSParallelCompact::first_dead_space_region(const RegionData* beg,
1.1274 + const RegionData* end)
1.1275 +{
1.1276 + const size_t region_size = ParallelCompactData::RegionSize;
1.1277 + ParallelCompactData& sd = summary_data();
1.1278 + size_t left = sd.region(beg);
1.1279 + size_t right = end > beg ? sd.region(end) - 1 : left;
1.1280 +
1.1281 + // Binary search.
1.1282 + while (left < right) {
1.1283 + // Equivalent to (left + right) / 2, but does not overflow.
1.1284 + const size_t middle = left + (right - left) / 2;
1.1285 + RegionData* const middle_ptr = sd.region(middle);
1.1286 + HeapWord* const dest = middle_ptr->destination();
1.1287 + HeapWord* const addr = sd.region_to_addr(middle);
1.1288 + assert(dest != NULL, "sanity");
1.1289 + assert(dest <= addr, "must move left");
1.1290 +
1.1291 + if (middle > left && dest < addr) {
1.1292 + right = middle - 1;
1.1293 + } else if (middle < right && middle_ptr->data_size() == region_size) {
1.1294 + left = middle + 1;
1.1295 + } else {
1.1296 + return middle_ptr;
1.1297 + }
1.1298 + }
1.1299 + return sd.region(left);
1.1300 +}
1.1301 +
1.1302 +ParallelCompactData::RegionData*
1.1303 +PSParallelCompact::dead_wood_limit_region(const RegionData* beg,
1.1304 + const RegionData* end,
1.1305 + size_t dead_words)
1.1306 +{
1.1307 + ParallelCompactData& sd = summary_data();
1.1308 + size_t left = sd.region(beg);
1.1309 + size_t right = end > beg ? sd.region(end) - 1 : left;
1.1310 +
1.1311 + // Binary search.
1.1312 + while (left < right) {
1.1313 + // Equivalent to (left + right) / 2, but does not overflow.
1.1314 + const size_t middle = left + (right - left) / 2;
1.1315 + RegionData* const middle_ptr = sd.region(middle);
1.1316 + HeapWord* const dest = middle_ptr->destination();
1.1317 + HeapWord* const addr = sd.region_to_addr(middle);
1.1318 + assert(dest != NULL, "sanity");
1.1319 + assert(dest <= addr, "must move left");
1.1320 +
1.1321 + const size_t dead_to_left = pointer_delta(addr, dest);
1.1322 + if (middle > left && dead_to_left > dead_words) {
1.1323 + right = middle - 1;
1.1324 + } else if (middle < right && dead_to_left < dead_words) {
1.1325 + left = middle + 1;
1.1326 + } else {
1.1327 + return middle_ptr;
1.1328 + }
1.1329 + }
1.1330 + return sd.region(left);
1.1331 +}
1.1332 +
1.1333 +// The result is valid during the summary phase, after the initial summarization
1.1334 +// of each space into itself, and before final summarization.
1.1335 +inline double
1.1336 +PSParallelCompact::reclaimed_ratio(const RegionData* const cp,
1.1337 + HeapWord* const bottom,
1.1338 + HeapWord* const top,
1.1339 + HeapWord* const new_top)
1.1340 +{
1.1341 + ParallelCompactData& sd = summary_data();
1.1342 +
1.1343 + assert(cp != NULL, "sanity");
1.1344 + assert(bottom != NULL, "sanity");
1.1345 + assert(top != NULL, "sanity");
1.1346 + assert(new_top != NULL, "sanity");
1.1347 + assert(top >= new_top, "summary data problem?");
1.1348 + assert(new_top > bottom, "space is empty; should not be here");
1.1349 + assert(new_top >= cp->destination(), "sanity");
1.1350 + assert(top >= sd.region_to_addr(cp), "sanity");
1.1351 +
1.1352 + HeapWord* const destination = cp->destination();
1.1353 + const size_t dense_prefix_live = pointer_delta(destination, bottom);
1.1354 + const size_t compacted_region_live = pointer_delta(new_top, destination);
1.1355 + const size_t compacted_region_used = pointer_delta(top,
1.1356 + sd.region_to_addr(cp));
1.1357 + const size_t reclaimable = compacted_region_used - compacted_region_live;
1.1358 +
1.1359 + const double divisor = dense_prefix_live + 1.25 * compacted_region_live;
1.1360 + return double(reclaimable) / divisor;
1.1361 +}
1.1362 +
1.1363 +// Return the address of the end of the dense prefix, a.k.a. the start of the
1.1364 +// compacted region. The address is always on a region boundary.
1.1365 +//
1.1366 +// Completely full regions at the left are skipped, since no compaction can
1.1367 +// occur in those regions. Then the maximum amount of dead wood to allow is
1.1368 +// computed, based on the density (amount live / capacity) of the generation;
1.1369 +// the region with approximately that amount of dead space to the left is
1.1370 +// identified as the limit region. Regions between the last completely full
1.1371 +// region and the limit region are scanned and the one that has the best
1.1372 +// (maximum) reclaimed_ratio() is selected.
1.1373 +HeapWord*
1.1374 +PSParallelCompact::compute_dense_prefix(const SpaceId id,
1.1375 + bool maximum_compaction)
1.1376 +{
1.1377 + if (ParallelOldGCSplitALot) {
1.1378 + if (_space_info[id].dense_prefix() != _space_info[id].space()->bottom()) {
1.1379 + // The value was chosen to provoke splitting a young gen space; use it.
1.1380 + return _space_info[id].dense_prefix();
1.1381 + }
1.1382 + }
1.1383 +
1.1384 + const size_t region_size = ParallelCompactData::RegionSize;
1.1385 + const ParallelCompactData& sd = summary_data();
1.1386 +
1.1387 + const MutableSpace* const space = _space_info[id].space();
1.1388 + HeapWord* const top = space->top();
1.1389 + HeapWord* const top_aligned_up = sd.region_align_up(top);
1.1390 + HeapWord* const new_top = _space_info[id].new_top();
1.1391 + HeapWord* const new_top_aligned_up = sd.region_align_up(new_top);
1.1392 + HeapWord* const bottom = space->bottom();
1.1393 + const RegionData* const beg_cp = sd.addr_to_region_ptr(bottom);
1.1394 + const RegionData* const top_cp = sd.addr_to_region_ptr(top_aligned_up);
1.1395 + const RegionData* const new_top_cp =
1.1396 + sd.addr_to_region_ptr(new_top_aligned_up);
1.1397 +
1.1398 + // Skip full regions at the beginning of the space--they are necessarily part
1.1399 + // of the dense prefix.
1.1400 + const RegionData* const full_cp = first_dead_space_region(beg_cp, new_top_cp);
1.1401 + assert(full_cp->destination() == sd.region_to_addr(full_cp) ||
1.1402 + space->is_empty(), "no dead space allowed to the left");
1.1403 + assert(full_cp->data_size() < region_size || full_cp == new_top_cp - 1,
1.1404 + "region must have dead space");
1.1405 +
1.1406 + // The gc number is saved whenever a maximum compaction is done, and used to
1.1407 + // determine when the maximum compaction interval has expired. This avoids
1.1408 + // successive max compactions for different reasons.
1.1409 + assert(total_invocations() >= _maximum_compaction_gc_num, "sanity");
1.1410 + const size_t gcs_since_max = total_invocations() - _maximum_compaction_gc_num;
1.1411 + const bool interval_ended = gcs_since_max > HeapMaximumCompactionInterval ||
1.1412 + total_invocations() == HeapFirstMaximumCompactionCount;
1.1413 + if (maximum_compaction || full_cp == top_cp || interval_ended) {
1.1414 + _maximum_compaction_gc_num = total_invocations();
1.1415 + return sd.region_to_addr(full_cp);
1.1416 + }
1.1417 +
1.1418 + const size_t space_live = pointer_delta(new_top, bottom);
1.1419 + const size_t space_used = space->used_in_words();
1.1420 + const size_t space_capacity = space->capacity_in_words();
1.1421 +
1.1422 + const double density = double(space_live) / double(space_capacity);
1.1423 + const size_t min_percent_free = MarkSweepDeadRatio;
1.1424 + const double limiter = dead_wood_limiter(density, min_percent_free);
1.1425 + const size_t dead_wood_max = space_used - space_live;
1.1426 + const size_t dead_wood_limit = MIN2(size_t(space_capacity * limiter),
1.1427 + dead_wood_max);
1.1428 +
1.1429 + if (TraceParallelOldGCDensePrefix) {
1.1430 + tty->print_cr("space_live=" SIZE_FORMAT " " "space_used=" SIZE_FORMAT " "
1.1431 + "space_cap=" SIZE_FORMAT,
1.1432 + space_live, space_used,
1.1433 + space_capacity);
1.1434 + tty->print_cr("dead_wood_limiter(%6.4f, %d)=%6.4f "
1.1435 + "dead_wood_max=" SIZE_FORMAT " dead_wood_limit=" SIZE_FORMAT,
1.1436 + density, min_percent_free, limiter,
1.1437 + dead_wood_max, dead_wood_limit);
1.1438 + }
1.1439 +
1.1440 + // Locate the region with the desired amount of dead space to the left.
1.1441 + const RegionData* const limit_cp =
1.1442 + dead_wood_limit_region(full_cp, top_cp, dead_wood_limit);
1.1443 +
1.1444 + // Scan from the first region with dead space to the limit region and find the
1.1445 + // one with the best (largest) reclaimed ratio.
1.1446 + double best_ratio = 0.0;
1.1447 + const RegionData* best_cp = full_cp;
1.1448 + for (const RegionData* cp = full_cp; cp < limit_cp; ++cp) {
1.1449 + double tmp_ratio = reclaimed_ratio(cp, bottom, top, new_top);
1.1450 + if (tmp_ratio > best_ratio) {
1.1451 + best_cp = cp;
1.1452 + best_ratio = tmp_ratio;
1.1453 + }
1.1454 + }
1.1455 +
1.1456 +#if 0
1.1457 + // Something to consider: if the region with the best ratio is 'close to' the
1.1458 + // first region w/free space, choose the first region with free space
1.1459 + // ("first-free"). The first-free region is usually near the start of the
1.1460 + // heap, which means we are copying most of the heap already, so copy a bit
1.1461 + // more to get complete compaction.
1.1462 + if (pointer_delta(best_cp, full_cp, sizeof(RegionData)) < 4) {
1.1463 + _maximum_compaction_gc_num = total_invocations();
1.1464 + best_cp = full_cp;
1.1465 + }
1.1466 +#endif // #if 0
1.1467 +
1.1468 + return sd.region_to_addr(best_cp);
1.1469 +}
1.1470 +
1.1471 +#ifndef PRODUCT
1.1472 +void
1.1473 +PSParallelCompact::fill_with_live_objects(SpaceId id, HeapWord* const start,
1.1474 + size_t words)
1.1475 +{
1.1476 + if (TraceParallelOldGCSummaryPhase) {
1.1477 + tty->print_cr("fill_with_live_objects [" PTR_FORMAT " " PTR_FORMAT ") "
1.1478 + SIZE_FORMAT, start, start + words, words);
1.1479 + }
1.1480 +
1.1481 + ObjectStartArray* const start_array = _space_info[id].start_array();
1.1482 + CollectedHeap::fill_with_objects(start, words);
1.1483 + for (HeapWord* p = start; p < start + words; p += oop(p)->size()) {
1.1484 + _mark_bitmap.mark_obj(p, words);
1.1485 + _summary_data.add_obj(p, words);
1.1486 + start_array->allocate_block(p);
1.1487 + }
1.1488 +}
1.1489 +
1.1490 +void
1.1491 +PSParallelCompact::summarize_new_objects(SpaceId id, HeapWord* start)
1.1492 +{
1.1493 + ParallelCompactData& sd = summary_data();
1.1494 + MutableSpace* space = _space_info[id].space();
1.1495 +
1.1496 + // Find the source and destination start addresses.
1.1497 + HeapWord* const src_addr = sd.region_align_down(start);
1.1498 + HeapWord* dst_addr;
1.1499 + if (src_addr < start) {
1.1500 + dst_addr = sd.addr_to_region_ptr(src_addr)->destination();
1.1501 + } else if (src_addr > space->bottom()) {
1.1502 + // The start (the original top() value) is aligned to a region boundary so
1.1503 + // the associated region does not have a destination. Compute the
1.1504 + // destination from the previous region.
1.1505 + RegionData* const cp = sd.addr_to_region_ptr(src_addr) - 1;
1.1506 + dst_addr = cp->destination() + cp->data_size();
1.1507 + } else {
1.1508 + // Filling the entire space.
1.1509 + dst_addr = space->bottom();
1.1510 + }
1.1511 + assert(dst_addr != NULL, "sanity");
1.1512 +
1.1513 + // Update the summary data.
1.1514 + bool result = _summary_data.summarize(_space_info[id].split_info(),
1.1515 + src_addr, space->top(), NULL,
1.1516 + dst_addr, space->end(),
1.1517 + _space_info[id].new_top_addr());
1.1518 + assert(result, "should not fail: bad filler object size");
1.1519 +}
1.1520 +
1.1521 +void
1.1522 +PSParallelCompact::provoke_split_fill_survivor(SpaceId id)
1.1523 +{
1.1524 + if (total_invocations() % (ParallelOldGCSplitInterval * 3) != 0) {
1.1525 + return;
1.1526 + }
1.1527 +
1.1528 + MutableSpace* const space = _space_info[id].space();
1.1529 + if (space->is_empty()) {
1.1530 + HeapWord* b = space->bottom();
1.1531 + HeapWord* t = b + space->capacity_in_words() / 2;
1.1532 + space->set_top(t);
1.1533 + if (ZapUnusedHeapArea) {
1.1534 + space->set_top_for_allocations();
1.1535 + }
1.1536 +
1.1537 + size_t min_size = CollectedHeap::min_fill_size();
1.1538 + size_t obj_len = min_size;
1.1539 + while (b + obj_len <= t) {
1.1540 + CollectedHeap::fill_with_object(b, obj_len);
1.1541 + mark_bitmap()->mark_obj(b, obj_len);
1.1542 + summary_data().add_obj(b, obj_len);
1.1543 + b += obj_len;
1.1544 + obj_len = (obj_len & (min_size*3)) + min_size; // 8 16 24 32 8 16 24 32 ...
1.1545 + }
1.1546 + if (b < t) {
1.1547 + // The loop didn't completely fill to t (top); adjust top downward.
1.1548 + space->set_top(b);
1.1549 + if (ZapUnusedHeapArea) {
1.1550 + space->set_top_for_allocations();
1.1551 + }
1.1552 + }
1.1553 +
1.1554 + HeapWord** nta = _space_info[id].new_top_addr();
1.1555 + bool result = summary_data().summarize(_space_info[id].split_info(),
1.1556 + space->bottom(), space->top(), NULL,
1.1557 + space->bottom(), space->end(), nta);
1.1558 + assert(result, "space must fit into itself");
1.1559 + }
1.1560 +}
1.1561 +
1.1562 +void
1.1563 +PSParallelCompact::provoke_split(bool & max_compaction)
1.1564 +{
1.1565 + if (total_invocations() % ParallelOldGCSplitInterval != 0) {
1.1566 + return;
1.1567 + }
1.1568 +
1.1569 + const size_t region_size = ParallelCompactData::RegionSize;
1.1570 + ParallelCompactData& sd = summary_data();
1.1571 +
1.1572 + MutableSpace* const eden_space = _space_info[eden_space_id].space();
1.1573 + MutableSpace* const from_space = _space_info[from_space_id].space();
1.1574 + const size_t eden_live = pointer_delta(eden_space->top(),
1.1575 + _space_info[eden_space_id].new_top());
1.1576 + const size_t from_live = pointer_delta(from_space->top(),
1.1577 + _space_info[from_space_id].new_top());
1.1578 +
1.1579 + const size_t min_fill_size = CollectedHeap::min_fill_size();
1.1580 + const size_t eden_free = pointer_delta(eden_space->end(), eden_space->top());
1.1581 + const size_t eden_fillable = eden_free >= min_fill_size ? eden_free : 0;
1.1582 + const size_t from_free = pointer_delta(from_space->end(), from_space->top());
1.1583 + const size_t from_fillable = from_free >= min_fill_size ? from_free : 0;
1.1584 +
1.1585 + // Choose the space to split; need at least 2 regions live (or fillable).
1.1586 + SpaceId id;
1.1587 + MutableSpace* space;
1.1588 + size_t live_words;
1.1589 + size_t fill_words;
1.1590 + if (eden_live + eden_fillable >= region_size * 2) {
1.1591 + id = eden_space_id;
1.1592 + space = eden_space;
1.1593 + live_words = eden_live;
1.1594 + fill_words = eden_fillable;
1.1595 + } else if (from_live + from_fillable >= region_size * 2) {
1.1596 + id = from_space_id;
1.1597 + space = from_space;
1.1598 + live_words = from_live;
1.1599 + fill_words = from_fillable;
1.1600 + } else {
1.1601 + return; // Give up.
1.1602 + }
1.1603 + assert(fill_words == 0 || fill_words >= min_fill_size, "sanity");
1.1604 +
1.1605 + if (live_words < region_size * 2) {
1.1606 + // Fill from top() to end() w/live objects of mixed sizes.
1.1607 + HeapWord* const fill_start = space->top();
1.1608 + live_words += fill_words;
1.1609 +
1.1610 + space->set_top(fill_start + fill_words);
1.1611 + if (ZapUnusedHeapArea) {
1.1612 + space->set_top_for_allocations();
1.1613 + }
1.1614 +
1.1615 + HeapWord* cur_addr = fill_start;
1.1616 + while (fill_words > 0) {
1.1617 + const size_t r = (size_t)os::random() % (region_size / 2) + min_fill_size;
1.1618 + size_t cur_size = MIN2(align_object_size_(r), fill_words);
1.1619 + if (fill_words - cur_size < min_fill_size) {
1.1620 + cur_size = fill_words; // Avoid leaving a fragment too small to fill.
1.1621 + }
1.1622 +
1.1623 + CollectedHeap::fill_with_object(cur_addr, cur_size);
1.1624 + mark_bitmap()->mark_obj(cur_addr, cur_size);
1.1625 + sd.add_obj(cur_addr, cur_size);
1.1626 +
1.1627 + cur_addr += cur_size;
1.1628 + fill_words -= cur_size;
1.1629 + }
1.1630 +
1.1631 + summarize_new_objects(id, fill_start);
1.1632 + }
1.1633 +
1.1634 + max_compaction = false;
1.1635 +
1.1636 + // Manipulate the old gen so that it has room for about half of the live data
1.1637 + // in the target young gen space (live_words / 2).
1.1638 + id = old_space_id;
1.1639 + space = _space_info[id].space();
1.1640 + const size_t free_at_end = space->free_in_words();
1.1641 + const size_t free_target = align_object_size(live_words / 2);
1.1642 + const size_t dead = pointer_delta(space->top(), _space_info[id].new_top());
1.1643 +
1.1644 + if (free_at_end >= free_target + min_fill_size) {
1.1645 + // Fill space above top() and set the dense prefix so everything survives.
1.1646 + HeapWord* const fill_start = space->top();
1.1647 + const size_t fill_size = free_at_end - free_target;
1.1648 + space->set_top(space->top() + fill_size);
1.1649 + if (ZapUnusedHeapArea) {
1.1650 + space->set_top_for_allocations();
1.1651 + }
1.1652 + fill_with_live_objects(id, fill_start, fill_size);
1.1653 + summarize_new_objects(id, fill_start);
1.1654 + _space_info[id].set_dense_prefix(sd.region_align_down(space->top()));
1.1655 + } else if (dead + free_at_end > free_target) {
1.1656 + // Find a dense prefix that makes the right amount of space available.
1.1657 + HeapWord* cur = sd.region_align_down(space->top());
1.1658 + HeapWord* cur_destination = sd.addr_to_region_ptr(cur)->destination();
1.1659 + size_t dead_to_right = pointer_delta(space->end(), cur_destination);
1.1660 + while (dead_to_right < free_target) {
1.1661 + cur -= region_size;
1.1662 + cur_destination = sd.addr_to_region_ptr(cur)->destination();
1.1663 + dead_to_right = pointer_delta(space->end(), cur_destination);
1.1664 + }
1.1665 + _space_info[id].set_dense_prefix(cur);
1.1666 + }
1.1667 +}
1.1668 +#endif // #ifndef PRODUCT
1.1669 +
1.1670 +void PSParallelCompact::summarize_spaces_quick()
1.1671 +{
1.1672 + for (unsigned int i = 0; i < last_space_id; ++i) {
1.1673 + const MutableSpace* space = _space_info[i].space();
1.1674 + HeapWord** nta = _space_info[i].new_top_addr();
1.1675 + bool result = _summary_data.summarize(_space_info[i].split_info(),
1.1676 + space->bottom(), space->top(), NULL,
1.1677 + space->bottom(), space->end(), nta);
1.1678 + assert(result, "space must fit into itself");
1.1679 + _space_info[i].set_dense_prefix(space->bottom());
1.1680 + }
1.1681 +
1.1682 +#ifndef PRODUCT
1.1683 + if (ParallelOldGCSplitALot) {
1.1684 + provoke_split_fill_survivor(to_space_id);
1.1685 + }
1.1686 +#endif // #ifndef PRODUCT
1.1687 +}
1.1688 +
1.1689 +void PSParallelCompact::fill_dense_prefix_end(SpaceId id)
1.1690 +{
1.1691 + HeapWord* const dense_prefix_end = dense_prefix(id);
1.1692 + const RegionData* region = _summary_data.addr_to_region_ptr(dense_prefix_end);
1.1693 + const idx_t dense_prefix_bit = _mark_bitmap.addr_to_bit(dense_prefix_end);
1.1694 + if (dead_space_crosses_boundary(region, dense_prefix_bit)) {
1.1695 + // Only enough dead space is filled so that any remaining dead space to the
1.1696 + // left is larger than the minimum filler object. (The remainder is filled
1.1697 + // during the copy/update phase.)
1.1698 + //
1.1699 + // The size of the dead space to the right of the boundary is not a
1.1700 + // concern, since compaction will be able to use whatever space is
1.1701 + // available.
1.1702 + //
1.1703 + // Here '||' is the boundary, 'x' represents a don't care bit and a box
1.1704 + // surrounds the space to be filled with an object.
1.1705 + //
1.1706 + // In the 32-bit VM, each bit represents two 32-bit words:
1.1707 + // +---+
1.1708 + // a) beg_bits: ... x x x | 0 | || 0 x x ...
1.1709 + // end_bits: ... x x x | 0 | || 0 x x ...
1.1710 + // +---+
1.1711 + //
1.1712 + // In the 64-bit VM, each bit represents one 64-bit word:
1.1713 + // +------------+
1.1714 + // b) beg_bits: ... x x x | 0 || 0 | x x ...
1.1715 + // end_bits: ... x x 1 | 0 || 0 | x x ...
1.1716 + // +------------+
1.1717 + // +-------+
1.1718 + // c) beg_bits: ... x x | 0 0 | || 0 x x ...
1.1719 + // end_bits: ... x 1 | 0 0 | || 0 x x ...
1.1720 + // +-------+
1.1721 + // +-----------+
1.1722 + // d) beg_bits: ... x | 0 0 0 | || 0 x x ...
1.1723 + // end_bits: ... 1 | 0 0 0 | || 0 x x ...
1.1724 + // +-----------+
1.1725 + // +-------+
1.1726 + // e) beg_bits: ... 0 0 | 0 0 | || 0 x x ...
1.1727 + // end_bits: ... 0 0 | 0 0 | || 0 x x ...
1.1728 + // +-------+
1.1729 +
1.1730 + // Initially assume case a, c or e will apply.
1.1731 + size_t obj_len = CollectedHeap::min_fill_size();
1.1732 + HeapWord* obj_beg = dense_prefix_end - obj_len;
1.1733 +
1.1734 +#ifdef _LP64
1.1735 + if (MinObjAlignment > 1) { // object alignment > heap word size
1.1736 + // Cases a, c or e.
1.1737 + } else if (_mark_bitmap.is_obj_end(dense_prefix_bit - 2)) {
1.1738 + // Case b above.
1.1739 + obj_beg = dense_prefix_end - 1;
1.1740 + } else if (!_mark_bitmap.is_obj_end(dense_prefix_bit - 3) &&
1.1741 + _mark_bitmap.is_obj_end(dense_prefix_bit - 4)) {
1.1742 + // Case d above.
1.1743 + obj_beg = dense_prefix_end - 3;
1.1744 + obj_len = 3;
1.1745 + }
1.1746 +#endif // #ifdef _LP64
1.1747 +
1.1748 + CollectedHeap::fill_with_object(obj_beg, obj_len);
1.1749 + _mark_bitmap.mark_obj(obj_beg, obj_len);
1.1750 + _summary_data.add_obj(obj_beg, obj_len);
1.1751 + assert(start_array(id) != NULL, "sanity");
1.1752 + start_array(id)->allocate_block(obj_beg);
1.1753 + }
1.1754 +}
1.1755 +
1.1756 +void
1.1757 +PSParallelCompact::clear_source_region(HeapWord* beg_addr, HeapWord* end_addr)
1.1758 +{
1.1759 + RegionData* const beg_ptr = _summary_data.addr_to_region_ptr(beg_addr);
1.1760 + HeapWord* const end_aligned_up = _summary_data.region_align_up(end_addr);
1.1761 + RegionData* const end_ptr = _summary_data.addr_to_region_ptr(end_aligned_up);
1.1762 + for (RegionData* cur = beg_ptr; cur < end_ptr; ++cur) {
1.1763 + cur->set_source_region(0);
1.1764 + }
1.1765 +}
1.1766 +
1.1767 +void
1.1768 +PSParallelCompact::summarize_space(SpaceId id, bool maximum_compaction)
1.1769 +{
1.1770 + assert(id < last_space_id, "id out of range");
1.1771 + assert(_space_info[id].dense_prefix() == _space_info[id].space()->bottom() ||
1.1772 + ParallelOldGCSplitALot && id == old_space_id,
1.1773 + "should have been reset in summarize_spaces_quick()");
1.1774 +
1.1775 + const MutableSpace* space = _space_info[id].space();
1.1776 + if (_space_info[id].new_top() != space->bottom()) {
1.1777 + HeapWord* dense_prefix_end = compute_dense_prefix(id, maximum_compaction);
1.1778 + _space_info[id].set_dense_prefix(dense_prefix_end);
1.1779 +
1.1780 +#ifndef PRODUCT
1.1781 + if (TraceParallelOldGCDensePrefix) {
1.1782 + print_dense_prefix_stats("ratio", id, maximum_compaction,
1.1783 + dense_prefix_end);
1.1784 + HeapWord* addr = compute_dense_prefix_via_density(id, maximum_compaction);
1.1785 + print_dense_prefix_stats("density", id, maximum_compaction, addr);
1.1786 + }
1.1787 +#endif // #ifndef PRODUCT
1.1788 +
1.1789 + // Recompute the summary data, taking into account the dense prefix. If
1.1790 + // every last byte will be reclaimed, then the existing summary data which
1.1791 + // compacts everything can be left in place.
1.1792 + if (!maximum_compaction && dense_prefix_end != space->bottom()) {
1.1793 + // If dead space crosses the dense prefix boundary, it is (at least
1.1794 + // partially) filled with a dummy object, marked live and added to the
1.1795 + // summary data. This simplifies the copy/update phase and must be done
1.1796 + // before the final locations of objects are determined, to prevent
1.1797 + // leaving a fragment of dead space that is too small to fill.
1.1798 + fill_dense_prefix_end(id);
1.1799 +
1.1800 + // Compute the destination of each Region, and thus each object.
1.1801 + _summary_data.summarize_dense_prefix(space->bottom(), dense_prefix_end);
1.1802 + _summary_data.summarize(_space_info[id].split_info(),
1.1803 + dense_prefix_end, space->top(), NULL,
1.1804 + dense_prefix_end, space->end(),
1.1805 + _space_info[id].new_top_addr());
1.1806 + }
1.1807 + }
1.1808 +
1.1809 + if (TraceParallelOldGCSummaryPhase) {
1.1810 + const size_t region_size = ParallelCompactData::RegionSize;
1.1811 + HeapWord* const dense_prefix_end = _space_info[id].dense_prefix();
1.1812 + const size_t dp_region = _summary_data.addr_to_region_idx(dense_prefix_end);
1.1813 + const size_t dp_words = pointer_delta(dense_prefix_end, space->bottom());
1.1814 + HeapWord* const new_top = _space_info[id].new_top();
1.1815 + const HeapWord* nt_aligned_up = _summary_data.region_align_up(new_top);
1.1816 + const size_t cr_words = pointer_delta(nt_aligned_up, dense_prefix_end);
1.1817 + tty->print_cr("id=%d cap=" SIZE_FORMAT " dp=" PTR_FORMAT " "
1.1818 + "dp_region=" SIZE_FORMAT " " "dp_count=" SIZE_FORMAT " "
1.1819 + "cr_count=" SIZE_FORMAT " " "nt=" PTR_FORMAT,
1.1820 + id, space->capacity_in_words(), dense_prefix_end,
1.1821 + dp_region, dp_words / region_size,
1.1822 + cr_words / region_size, new_top);
1.1823 + }
1.1824 +}
1.1825 +
1.1826 +#ifndef PRODUCT
1.1827 +void PSParallelCompact::summary_phase_msg(SpaceId dst_space_id,
1.1828 + HeapWord* dst_beg, HeapWord* dst_end,
1.1829 + SpaceId src_space_id,
1.1830 + HeapWord* src_beg, HeapWord* src_end)
1.1831 +{
1.1832 + if (TraceParallelOldGCSummaryPhase) {
1.1833 + tty->print_cr("summarizing %d [%s] into %d [%s]: "
1.1834 + "src=" PTR_FORMAT "-" PTR_FORMAT " "
1.1835 + SIZE_FORMAT "-" SIZE_FORMAT " "
1.1836 + "dst=" PTR_FORMAT "-" PTR_FORMAT " "
1.1837 + SIZE_FORMAT "-" SIZE_FORMAT,
1.1838 + src_space_id, space_names[src_space_id],
1.1839 + dst_space_id, space_names[dst_space_id],
1.1840 + src_beg, src_end,
1.1841 + _summary_data.addr_to_region_idx(src_beg),
1.1842 + _summary_data.addr_to_region_idx(src_end),
1.1843 + dst_beg, dst_end,
1.1844 + _summary_data.addr_to_region_idx(dst_beg),
1.1845 + _summary_data.addr_to_region_idx(dst_end));
1.1846 + }
1.1847 +}
1.1848 +#endif // #ifndef PRODUCT
1.1849 +
1.1850 +void PSParallelCompact::summary_phase(ParCompactionManager* cm,
1.1851 + bool maximum_compaction)
1.1852 +{
1.1853 + GCTraceTime tm("summary phase", print_phases(), true, &_gc_timer);
1.1854 + // trace("2");
1.1855 +
1.1856 +#ifdef ASSERT
1.1857 + if (TraceParallelOldGCMarkingPhase) {
1.1858 + tty->print_cr("add_obj_count=" SIZE_FORMAT " "
1.1859 + "add_obj_bytes=" SIZE_FORMAT,
1.1860 + add_obj_count, add_obj_size * HeapWordSize);
1.1861 + tty->print_cr("mark_bitmap_count=" SIZE_FORMAT " "
1.1862 + "mark_bitmap_bytes=" SIZE_FORMAT,
1.1863 + mark_bitmap_count, mark_bitmap_size * HeapWordSize);
1.1864 + }
1.1865 +#endif // #ifdef ASSERT
1.1866 +
1.1867 + // Quick summarization of each space into itself, to see how much is live.
1.1868 + summarize_spaces_quick();
1.1869 +
1.1870 + if (TraceParallelOldGCSummaryPhase) {
1.1871 + tty->print_cr("summary_phase: after summarizing each space to self");
1.1872 + Universe::print();
1.1873 + NOT_PRODUCT(print_region_ranges());
1.1874 + if (Verbose) {
1.1875 + NOT_PRODUCT(print_initial_summary_data(_summary_data, _space_info));
1.1876 + }
1.1877 + }
1.1878 +
1.1879 + // The amount of live data that will end up in old space (assuming it fits).
1.1880 + size_t old_space_total_live = 0;
1.1881 + for (unsigned int id = old_space_id; id < last_space_id; ++id) {
1.1882 + old_space_total_live += pointer_delta(_space_info[id].new_top(),
1.1883 + _space_info[id].space()->bottom());
1.1884 + }
1.1885 +
1.1886 + MutableSpace* const old_space = _space_info[old_space_id].space();
1.1887 + const size_t old_capacity = old_space->capacity_in_words();
1.1888 + if (old_space_total_live > old_capacity) {
1.1889 + // XXX - should also try to expand
1.1890 + maximum_compaction = true;
1.1891 + }
1.1892 +#ifndef PRODUCT
1.1893 + if (ParallelOldGCSplitALot && old_space_total_live < old_capacity) {
1.1894 + provoke_split(maximum_compaction);
1.1895 + }
1.1896 +#endif // #ifndef PRODUCT
1.1897 +
1.1898 + // Old generations.
1.1899 + summarize_space(old_space_id, maximum_compaction);
1.1900 +
1.1901 + // Summarize the remaining spaces in the young gen. The initial target space
1.1902 + // is the old gen. If a space does not fit entirely into the target, then the
1.1903 + // remainder is compacted into the space itself and that space becomes the new
1.1904 + // target.
1.1905 + SpaceId dst_space_id = old_space_id;
1.1906 + HeapWord* dst_space_end = old_space->end();
1.1907 + HeapWord** new_top_addr = _space_info[dst_space_id].new_top_addr();
1.1908 + for (unsigned int id = eden_space_id; id < last_space_id; ++id) {
1.1909 + const MutableSpace* space = _space_info[id].space();
1.1910 + const size_t live = pointer_delta(_space_info[id].new_top(),
1.1911 + space->bottom());
1.1912 + const size_t available = pointer_delta(dst_space_end, *new_top_addr);
1.1913 +
1.1914 + NOT_PRODUCT(summary_phase_msg(dst_space_id, *new_top_addr, dst_space_end,
1.1915 + SpaceId(id), space->bottom(), space->top());)
1.1916 + if (live > 0 && live <= available) {
1.1917 + // All the live data will fit.
1.1918 + bool done = _summary_data.summarize(_space_info[id].split_info(),
1.1919 + space->bottom(), space->top(),
1.1920 + NULL,
1.1921 + *new_top_addr, dst_space_end,
1.1922 + new_top_addr);
1.1923 + assert(done, "space must fit into old gen");
1.1924 +
1.1925 + // Reset the new_top value for the space.
1.1926 + _space_info[id].set_new_top(space->bottom());
1.1927 + } else if (live > 0) {
1.1928 + // Attempt to fit part of the source space into the target space.
1.1929 + HeapWord* next_src_addr = NULL;
1.1930 + bool done = _summary_data.summarize(_space_info[id].split_info(),
1.1931 + space->bottom(), space->top(),
1.1932 + &next_src_addr,
1.1933 + *new_top_addr, dst_space_end,
1.1934 + new_top_addr);
1.1935 + assert(!done, "space should not fit into old gen");
1.1936 + assert(next_src_addr != NULL, "sanity");
1.1937 +
1.1938 + // The source space becomes the new target, so the remainder is compacted
1.1939 + // within the space itself.
1.1940 + dst_space_id = SpaceId(id);
1.1941 + dst_space_end = space->end();
1.1942 + new_top_addr = _space_info[id].new_top_addr();
1.1943 + NOT_PRODUCT(summary_phase_msg(dst_space_id,
1.1944 + space->bottom(), dst_space_end,
1.1945 + SpaceId(id), next_src_addr, space->top());)
1.1946 + done = _summary_data.summarize(_space_info[id].split_info(),
1.1947 + next_src_addr, space->top(),
1.1948 + NULL,
1.1949 + space->bottom(), dst_space_end,
1.1950 + new_top_addr);
1.1951 + assert(done, "space must fit when compacted into itself");
1.1952 + assert(*new_top_addr <= space->top(), "usage should not grow");
1.1953 + }
1.1954 + }
1.1955 +
1.1956 + if (TraceParallelOldGCSummaryPhase) {
1.1957 + tty->print_cr("summary_phase: after final summarization");
1.1958 + Universe::print();
1.1959 + NOT_PRODUCT(print_region_ranges());
1.1960 + if (Verbose) {
1.1961 + NOT_PRODUCT(print_generic_summary_data(_summary_data, _space_info));
1.1962 + }
1.1963 + }
1.1964 +}
1.1965 +
1.1966 +// This method should contain all heap-specific policy for invoking a full
1.1967 +// collection. invoke_no_policy() will only attempt to compact the heap; it
1.1968 +// will do nothing further. If we need to bail out for policy reasons, scavenge
1.1969 +// before full gc, or any other specialized behavior, it needs to be added here.
1.1970 +//
1.1971 +// Note that this method should only be called from the vm_thread while at a
1.1972 +// safepoint.
1.1973 +//
1.1974 +// Note that the all_soft_refs_clear flag in the collector policy
1.1975 +// may be true because this method can be called without intervening
1.1976 +// activity. For example when the heap space is tight and full measure
1.1977 +// are being taken to free space.
1.1978 +void PSParallelCompact::invoke(bool maximum_heap_compaction) {
1.1979 + assert(SafepointSynchronize::is_at_safepoint(), "should be at safepoint");
1.1980 + assert(Thread::current() == (Thread*)VMThread::vm_thread(),
1.1981 + "should be in vm thread");
1.1982 +
1.1983 + ParallelScavengeHeap* heap = gc_heap();
1.1984 + GCCause::Cause gc_cause = heap->gc_cause();
1.1985 + assert(!heap->is_gc_active(), "not reentrant");
1.1986 +
1.1987 + PSAdaptiveSizePolicy* policy = heap->size_policy();
1.1988 + IsGCActiveMark mark;
1.1989 +
1.1990 + if (ScavengeBeforeFullGC) {
1.1991 + PSScavenge::invoke_no_policy();
1.1992 + }
1.1993 +
1.1994 + const bool clear_all_soft_refs =
1.1995 + heap->collector_policy()->should_clear_all_soft_refs();
1.1996 +
1.1997 + PSParallelCompact::invoke_no_policy(clear_all_soft_refs ||
1.1998 + maximum_heap_compaction);
1.1999 +}
1.2000 +
1.2001 +// This method contains no policy. You should probably
1.2002 +// be calling invoke() instead.
1.2003 +bool PSParallelCompact::invoke_no_policy(bool maximum_heap_compaction) {
1.2004 + assert(SafepointSynchronize::is_at_safepoint(), "must be at a safepoint");
1.2005 + assert(ref_processor() != NULL, "Sanity");
1.2006 +
1.2007 + if (GC_locker::check_active_before_gc()) {
1.2008 + return false;
1.2009 + }
1.2010 +
1.2011 + ParallelScavengeHeap* heap = gc_heap();
1.2012 +
1.2013 + _gc_timer.register_gc_start();
1.2014 + _gc_tracer.report_gc_start(heap->gc_cause(), _gc_timer.gc_start());
1.2015 +
1.2016 + TimeStamp marking_start;
1.2017 + TimeStamp compaction_start;
1.2018 + TimeStamp collection_exit;
1.2019 +
1.2020 + GCCause::Cause gc_cause = heap->gc_cause();
1.2021 + PSYoungGen* young_gen = heap->young_gen();
1.2022 + PSOldGen* old_gen = heap->old_gen();
1.2023 + PSAdaptiveSizePolicy* size_policy = heap->size_policy();
1.2024 +
1.2025 + // The scope of casr should end after code that can change
1.2026 + // CollectorPolicy::_should_clear_all_soft_refs.
1.2027 + ClearedAllSoftRefs casr(maximum_heap_compaction,
1.2028 + heap->collector_policy());
1.2029 +
1.2030 + if (ZapUnusedHeapArea) {
1.2031 + // Save information needed to minimize mangling
1.2032 + heap->record_gen_tops_before_GC();
1.2033 + }
1.2034 +
1.2035 + heap->pre_full_gc_dump(&_gc_timer);
1.2036 +
1.2037 + _print_phases = PrintGCDetails && PrintParallelOldGCPhaseTimes;
1.2038 +
1.2039 + // Make sure data structures are sane, make the heap parsable, and do other
1.2040 + // miscellaneous bookkeeping.
1.2041 + PreGCValues pre_gc_values;
1.2042 + pre_compact(&pre_gc_values);
1.2043 +
1.2044 + // Get the compaction manager reserved for the VM thread.
1.2045 + ParCompactionManager* const vmthread_cm =
1.2046 + ParCompactionManager::manager_array(gc_task_manager()->workers());
1.2047 +
1.2048 + // Place after pre_compact() where the number of invocations is incremented.
1.2049 + AdaptiveSizePolicyOutput(size_policy, heap->total_collections());
1.2050 +
1.2051 + {
1.2052 + ResourceMark rm;
1.2053 + HandleMark hm;
1.2054 +
1.2055 + // Set the number of GC threads to be used in this collection
1.2056 + gc_task_manager()->set_active_gang();
1.2057 + gc_task_manager()->task_idle_workers();
1.2058 + heap->set_par_threads(gc_task_manager()->active_workers());
1.2059 +
1.2060 + gclog_or_tty->date_stamp(PrintGC && PrintGCDateStamps);
1.2061 + TraceCPUTime tcpu(PrintGCDetails, true, gclog_or_tty);
1.2062 + GCTraceTime t1(GCCauseString("Full GC", gc_cause), PrintGC, !PrintGCDetails, NULL);
1.2063 + TraceCollectorStats tcs(counters());
1.2064 + TraceMemoryManagerStats tms(true /* Full GC */,gc_cause);
1.2065 +
1.2066 + if (TraceGen1Time) accumulated_time()->start();
1.2067 +
1.2068 + // Let the size policy know we're starting
1.2069 + size_policy->major_collection_begin();
1.2070 +
1.2071 + CodeCache::gc_prologue();
1.2072 + Threads::gc_prologue();
1.2073 +
1.2074 + COMPILER2_PRESENT(DerivedPointerTable::clear());
1.2075 +
1.2076 + ref_processor()->enable_discovery(true /*verify_disabled*/, true /*verify_no_refs*/);
1.2077 + ref_processor()->setup_policy(maximum_heap_compaction);
1.2078 +
1.2079 + bool marked_for_unloading = false;
1.2080 +
1.2081 + marking_start.update();
1.2082 + marking_phase(vmthread_cm, maximum_heap_compaction, &_gc_tracer);
1.2083 +
1.2084 + bool max_on_system_gc = UseMaximumCompactionOnSystemGC
1.2085 + && gc_cause == GCCause::_java_lang_system_gc;
1.2086 + summary_phase(vmthread_cm, maximum_heap_compaction || max_on_system_gc);
1.2087 +
1.2088 + COMPILER2_PRESENT(assert(DerivedPointerTable::is_active(), "Sanity"));
1.2089 + COMPILER2_PRESENT(DerivedPointerTable::set_active(false));
1.2090 +
1.2091 + // adjust_roots() updates Universe::_intArrayKlassObj which is
1.2092 + // needed by the compaction for filling holes in the dense prefix.
1.2093 + adjust_roots();
1.2094 +
1.2095 + compaction_start.update();
1.2096 + compact();
1.2097 +
1.2098 + // Reset the mark bitmap, summary data, and do other bookkeeping. Must be
1.2099 + // done before resizing.
1.2100 + post_compact();
1.2101 +
1.2102 + // Let the size policy know we're done
1.2103 + size_policy->major_collection_end(old_gen->used_in_bytes(), gc_cause);
1.2104 +
1.2105 + if (UseAdaptiveSizePolicy) {
1.2106 + if (PrintAdaptiveSizePolicy) {
1.2107 + gclog_or_tty->print("AdaptiveSizeStart: ");
1.2108 + gclog_or_tty->stamp();
1.2109 + gclog_or_tty->print_cr(" collection: %d ",
1.2110 + heap->total_collections());
1.2111 + if (Verbose) {
1.2112 + gclog_or_tty->print("old_gen_capacity: %d young_gen_capacity: %d",
1.2113 + old_gen->capacity_in_bytes(), young_gen->capacity_in_bytes());
1.2114 + }
1.2115 + }
1.2116 +
1.2117 + // Don't check if the size_policy is ready here. Let
1.2118 + // the size_policy check that internally.
1.2119 + if (UseAdaptiveGenerationSizePolicyAtMajorCollection &&
1.2120 + ((gc_cause != GCCause::_java_lang_system_gc) ||
1.2121 + UseAdaptiveSizePolicyWithSystemGC)) {
1.2122 + // Calculate optimal free space amounts
1.2123 + assert(young_gen->max_size() >
1.2124 + young_gen->from_space()->capacity_in_bytes() +
1.2125 + young_gen->to_space()->capacity_in_bytes(),
1.2126 + "Sizes of space in young gen are out-of-bounds");
1.2127 +
1.2128 + size_t young_live = young_gen->used_in_bytes();
1.2129 + size_t eden_live = young_gen->eden_space()->used_in_bytes();
1.2130 + size_t old_live = old_gen->used_in_bytes();
1.2131 + size_t cur_eden = young_gen->eden_space()->capacity_in_bytes();
1.2132 + size_t max_old_gen_size = old_gen->max_gen_size();
1.2133 + size_t max_eden_size = young_gen->max_size() -
1.2134 + young_gen->from_space()->capacity_in_bytes() -
1.2135 + young_gen->to_space()->capacity_in_bytes();
1.2136 +
1.2137 + // Used for diagnostics
1.2138 + size_policy->clear_generation_free_space_flags();
1.2139 +
1.2140 + size_policy->compute_generations_free_space(young_live,
1.2141 + eden_live,
1.2142 + old_live,
1.2143 + cur_eden,
1.2144 + max_old_gen_size,
1.2145 + max_eden_size,
1.2146 + true /* full gc*/);
1.2147 +
1.2148 + size_policy->check_gc_overhead_limit(young_live,
1.2149 + eden_live,
1.2150 + max_old_gen_size,
1.2151 + max_eden_size,
1.2152 + true /* full gc*/,
1.2153 + gc_cause,
1.2154 + heap->collector_policy());
1.2155 +
1.2156 + size_policy->decay_supplemental_growth(true /* full gc*/);
1.2157 +
1.2158 + heap->resize_old_gen(
1.2159 + size_policy->calculated_old_free_size_in_bytes());
1.2160 +
1.2161 + // Don't resize the young generation at an major collection. A
1.2162 + // desired young generation size may have been calculated but
1.2163 + // resizing the young generation complicates the code because the
1.2164 + // resizing of the old generation may have moved the boundary
1.2165 + // between the young generation and the old generation. Let the
1.2166 + // young generation resizing happen at the minor collections.
1.2167 + }
1.2168 + if (PrintAdaptiveSizePolicy) {
1.2169 + gclog_or_tty->print_cr("AdaptiveSizeStop: collection: %d ",
1.2170 + heap->total_collections());
1.2171 + }
1.2172 + }
1.2173 +
1.2174 + if (UsePerfData) {
1.2175 + PSGCAdaptivePolicyCounters* const counters = heap->gc_policy_counters();
1.2176 + counters->update_counters();
1.2177 + counters->update_old_capacity(old_gen->capacity_in_bytes());
1.2178 + counters->update_young_capacity(young_gen->capacity_in_bytes());
1.2179 + }
1.2180 +
1.2181 + heap->resize_all_tlabs();
1.2182 +
1.2183 + // Resize the metaspace capactiy after a collection
1.2184 + MetaspaceGC::compute_new_size();
1.2185 +
1.2186 + if (TraceGen1Time) accumulated_time()->stop();
1.2187 +
1.2188 + if (PrintGC) {
1.2189 + if (PrintGCDetails) {
1.2190 + // No GC timestamp here. This is after GC so it would be confusing.
1.2191 + young_gen->print_used_change(pre_gc_values.young_gen_used());
1.2192 + old_gen->print_used_change(pre_gc_values.old_gen_used());
1.2193 + heap->print_heap_change(pre_gc_values.heap_used());
1.2194 + MetaspaceAux::print_metaspace_change(pre_gc_values.metadata_used());
1.2195 + } else {
1.2196 + heap->print_heap_change(pre_gc_values.heap_used());
1.2197 + }
1.2198 + }
1.2199 +
1.2200 + // Track memory usage and detect low memory
1.2201 + MemoryService::track_memory_usage();
1.2202 + heap->update_counters();
1.2203 + gc_task_manager()->release_idle_workers();
1.2204 + }
1.2205 +
1.2206 +#ifdef ASSERT
1.2207 + for (size_t i = 0; i < ParallelGCThreads + 1; ++i) {
1.2208 + ParCompactionManager* const cm =
1.2209 + ParCompactionManager::manager_array(int(i));
1.2210 + assert(cm->marking_stack()->is_empty(), "should be empty");
1.2211 + assert(ParCompactionManager::region_list(int(i))->is_empty(), "should be empty");
1.2212 + }
1.2213 +#endif // ASSERT
1.2214 +
1.2215 + if (VerifyAfterGC && heap->total_collections() >= VerifyGCStartAt) {
1.2216 + HandleMark hm; // Discard invalid handles created during verification
1.2217 + Universe::verify(" VerifyAfterGC:");
1.2218 + }
1.2219 +
1.2220 + // Re-verify object start arrays
1.2221 + if (VerifyObjectStartArray &&
1.2222 + VerifyAfterGC) {
1.2223 + old_gen->verify_object_start_array();
1.2224 + }
1.2225 +
1.2226 + if (ZapUnusedHeapArea) {
1.2227 + old_gen->object_space()->check_mangled_unused_area_complete();
1.2228 + }
1.2229 +
1.2230 + NOT_PRODUCT(ref_processor()->verify_no_references_recorded());
1.2231 +
1.2232 + collection_exit.update();
1.2233 +
1.2234 + heap->print_heap_after_gc();
1.2235 + heap->trace_heap_after_gc(&_gc_tracer);
1.2236 +
1.2237 + if (PrintGCTaskTimeStamps) {
1.2238 + gclog_or_tty->print_cr("VM-Thread " INT64_FORMAT " " INT64_FORMAT " "
1.2239 + INT64_FORMAT,
1.2240 + marking_start.ticks(), compaction_start.ticks(),
1.2241 + collection_exit.ticks());
1.2242 + gc_task_manager()->print_task_time_stamps();
1.2243 + }
1.2244 +
1.2245 + heap->post_full_gc_dump(&_gc_timer);
1.2246 +
1.2247 +#ifdef TRACESPINNING
1.2248 + ParallelTaskTerminator::print_termination_counts();
1.2249 +#endif
1.2250 +
1.2251 + _gc_timer.register_gc_end();
1.2252 +
1.2253 + _gc_tracer.report_dense_prefix(dense_prefix(old_space_id));
1.2254 + _gc_tracer.report_gc_end(_gc_timer.gc_end(), _gc_timer.time_partitions());
1.2255 +
1.2256 + return true;
1.2257 +}
1.2258 +
1.2259 +bool PSParallelCompact::absorb_live_data_from_eden(PSAdaptiveSizePolicy* size_policy,
1.2260 + PSYoungGen* young_gen,
1.2261 + PSOldGen* old_gen) {
1.2262 + MutableSpace* const eden_space = young_gen->eden_space();
1.2263 + assert(!eden_space->is_empty(), "eden must be non-empty");
1.2264 + assert(young_gen->virtual_space()->alignment() ==
1.2265 + old_gen->virtual_space()->alignment(), "alignments do not match");
1.2266 +
1.2267 + if (!(UseAdaptiveSizePolicy && UseAdaptiveGCBoundary)) {
1.2268 + return false;
1.2269 + }
1.2270 +
1.2271 + // Both generations must be completely committed.
1.2272 + if (young_gen->virtual_space()->uncommitted_size() != 0) {
1.2273 + return false;
1.2274 + }
1.2275 + if (old_gen->virtual_space()->uncommitted_size() != 0) {
1.2276 + return false;
1.2277 + }
1.2278 +
1.2279 + // Figure out how much to take from eden. Include the average amount promoted
1.2280 + // in the total; otherwise the next young gen GC will simply bail out to a
1.2281 + // full GC.
1.2282 + const size_t alignment = old_gen->virtual_space()->alignment();
1.2283 + const size_t eden_used = eden_space->used_in_bytes();
1.2284 + const size_t promoted = (size_t)size_policy->avg_promoted()->padded_average();
1.2285 + const size_t absorb_size = align_size_up(eden_used + promoted, alignment);
1.2286 + const size_t eden_capacity = eden_space->capacity_in_bytes();
1.2287 +
1.2288 + if (absorb_size >= eden_capacity) {
1.2289 + return false; // Must leave some space in eden.
1.2290 + }
1.2291 +
1.2292 + const size_t new_young_size = young_gen->capacity_in_bytes() - absorb_size;
1.2293 + if (new_young_size < young_gen->min_gen_size()) {
1.2294 + return false; // Respect young gen minimum size.
1.2295 + }
1.2296 +
1.2297 + if (TraceAdaptiveGCBoundary && Verbose) {
1.2298 + gclog_or_tty->print(" absorbing " SIZE_FORMAT "K: "
1.2299 + "eden " SIZE_FORMAT "K->" SIZE_FORMAT "K "
1.2300 + "from " SIZE_FORMAT "K, to " SIZE_FORMAT "K "
1.2301 + "young_gen " SIZE_FORMAT "K->" SIZE_FORMAT "K ",
1.2302 + absorb_size / K,
1.2303 + eden_capacity / K, (eden_capacity - absorb_size) / K,
1.2304 + young_gen->from_space()->used_in_bytes() / K,
1.2305 + young_gen->to_space()->used_in_bytes() / K,
1.2306 + young_gen->capacity_in_bytes() / K, new_young_size / K);
1.2307 + }
1.2308 +
1.2309 + // Fill the unused part of the old gen.
1.2310 + MutableSpace* const old_space = old_gen->object_space();
1.2311 + HeapWord* const unused_start = old_space->top();
1.2312 + size_t const unused_words = pointer_delta(old_space->end(), unused_start);
1.2313 +
1.2314 + if (unused_words > 0) {
1.2315 + if (unused_words < CollectedHeap::min_fill_size()) {
1.2316 + return false; // If the old gen cannot be filled, must give up.
1.2317 + }
1.2318 + CollectedHeap::fill_with_objects(unused_start, unused_words);
1.2319 + }
1.2320 +
1.2321 + // Take the live data from eden and set both top and end in the old gen to
1.2322 + // eden top. (Need to set end because reset_after_change() mangles the region
1.2323 + // from end to virtual_space->high() in debug builds).
1.2324 + HeapWord* const new_top = eden_space->top();
1.2325 + old_gen->virtual_space()->expand_into(young_gen->virtual_space(),
1.2326 + absorb_size);
1.2327 + young_gen->reset_after_change();
1.2328 + old_space->set_top(new_top);
1.2329 + old_space->set_end(new_top);
1.2330 + old_gen->reset_after_change();
1.2331 +
1.2332 + // Update the object start array for the filler object and the data from eden.
1.2333 + ObjectStartArray* const start_array = old_gen->start_array();
1.2334 + for (HeapWord* p = unused_start; p < new_top; p += oop(p)->size()) {
1.2335 + start_array->allocate_block(p);
1.2336 + }
1.2337 +
1.2338 + // Could update the promoted average here, but it is not typically updated at
1.2339 + // full GCs and the value to use is unclear. Something like
1.2340 + //
1.2341 + // cur_promoted_avg + absorb_size / number_of_scavenges_since_last_full_gc.
1.2342 +
1.2343 + size_policy->set_bytes_absorbed_from_eden(absorb_size);
1.2344 + return true;
1.2345 +}
1.2346 +
1.2347 +GCTaskManager* const PSParallelCompact::gc_task_manager() {
1.2348 + assert(ParallelScavengeHeap::gc_task_manager() != NULL,
1.2349 + "shouldn't return NULL");
1.2350 + return ParallelScavengeHeap::gc_task_manager();
1.2351 +}
1.2352 +
1.2353 +void PSParallelCompact::marking_phase(ParCompactionManager* cm,
1.2354 + bool maximum_heap_compaction,
1.2355 + ParallelOldTracer *gc_tracer) {
1.2356 + // Recursively traverse all live objects and mark them
1.2357 + GCTraceTime tm("marking phase", print_phases(), true, &_gc_timer);
1.2358 +
1.2359 + ParallelScavengeHeap* heap = gc_heap();
1.2360 + uint parallel_gc_threads = heap->gc_task_manager()->workers();
1.2361 + uint active_gc_threads = heap->gc_task_manager()->active_workers();
1.2362 + TaskQueueSetSuper* qset = ParCompactionManager::region_array();
1.2363 + ParallelTaskTerminator terminator(active_gc_threads, qset);
1.2364 +
1.2365 + PSParallelCompact::MarkAndPushClosure mark_and_push_closure(cm);
1.2366 + PSParallelCompact::FollowStackClosure follow_stack_closure(cm);
1.2367 +
1.2368 + // Need new claim bits before marking starts.
1.2369 + ClassLoaderDataGraph::clear_claimed_marks();
1.2370 +
1.2371 + {
1.2372 + GCTraceTime tm_m("par mark", print_phases(), true, &_gc_timer);
1.2373 +
1.2374 + ParallelScavengeHeap::ParStrongRootsScope psrs;
1.2375 +
1.2376 + GCTaskQueue* q = GCTaskQueue::create();
1.2377 +
1.2378 + q->enqueue(new MarkFromRootsTask(MarkFromRootsTask::universe));
1.2379 + q->enqueue(new MarkFromRootsTask(MarkFromRootsTask::jni_handles));
1.2380 + // We scan the thread roots in parallel
1.2381 + Threads::create_thread_roots_marking_tasks(q);
1.2382 + q->enqueue(new MarkFromRootsTask(MarkFromRootsTask::object_synchronizer));
1.2383 + q->enqueue(new MarkFromRootsTask(MarkFromRootsTask::flat_profiler));
1.2384 + q->enqueue(new MarkFromRootsTask(MarkFromRootsTask::management));
1.2385 + q->enqueue(new MarkFromRootsTask(MarkFromRootsTask::system_dictionary));
1.2386 + q->enqueue(new MarkFromRootsTask(MarkFromRootsTask::class_loader_data));
1.2387 + q->enqueue(new MarkFromRootsTask(MarkFromRootsTask::jvmti));
1.2388 + q->enqueue(new MarkFromRootsTask(MarkFromRootsTask::code_cache));
1.2389 +
1.2390 + if (active_gc_threads > 1) {
1.2391 + for (uint j = 0; j < active_gc_threads; j++) {
1.2392 + q->enqueue(new StealMarkingTask(&terminator));
1.2393 + }
1.2394 + }
1.2395 +
1.2396 + gc_task_manager()->execute_and_wait(q);
1.2397 + }
1.2398 +
1.2399 + // Process reference objects found during marking
1.2400 + {
1.2401 + GCTraceTime tm_r("reference processing", print_phases(), true, &_gc_timer);
1.2402 +
1.2403 + ReferenceProcessorStats stats;
1.2404 + if (ref_processor()->processing_is_mt()) {
1.2405 + RefProcTaskExecutor task_executor;
1.2406 + stats = ref_processor()->process_discovered_references(
1.2407 + is_alive_closure(), &mark_and_push_closure, &follow_stack_closure,
1.2408 + &task_executor, &_gc_timer);
1.2409 + } else {
1.2410 + stats = ref_processor()->process_discovered_references(
1.2411 + is_alive_closure(), &mark_and_push_closure, &follow_stack_closure, NULL,
1.2412 + &_gc_timer);
1.2413 + }
1.2414 +
1.2415 + gc_tracer->report_gc_reference_stats(stats);
1.2416 + }
1.2417 +
1.2418 + GCTraceTime tm_c("class unloading", print_phases(), true, &_gc_timer);
1.2419 +
1.2420 + // This is the point where the entire marking should have completed.
1.2421 + assert(cm->marking_stacks_empty(), "Marking should have completed");
1.2422 +
1.2423 + // Follow system dictionary roots and unload classes.
1.2424 + bool purged_class = SystemDictionary::do_unloading(is_alive_closure());
1.2425 +
1.2426 + // Unload nmethods.
1.2427 + CodeCache::do_unloading(is_alive_closure(), purged_class);
1.2428 +
1.2429 + // Prune dead klasses from subklass/sibling/implementor lists.
1.2430 + Klass::clean_weak_klass_links(is_alive_closure());
1.2431 +
1.2432 + // Delete entries for dead interned strings.
1.2433 + StringTable::unlink(is_alive_closure());
1.2434 +
1.2435 + // Clean up unreferenced symbols in symbol table.
1.2436 + SymbolTable::unlink();
1.2437 + _gc_tracer.report_object_count_after_gc(is_alive_closure());
1.2438 +}
1.2439 +
1.2440 +void PSParallelCompact::follow_class_loader(ParCompactionManager* cm,
1.2441 + ClassLoaderData* cld) {
1.2442 + PSParallelCompact::MarkAndPushClosure mark_and_push_closure(cm);
1.2443 + PSParallelCompact::FollowKlassClosure follow_klass_closure(&mark_and_push_closure);
1.2444 +
1.2445 + cld->oops_do(&mark_and_push_closure, &follow_klass_closure, true);
1.2446 +}
1.2447 +
1.2448 +// This should be moved to the shared markSweep code!
1.2449 +class PSAlwaysTrueClosure: public BoolObjectClosure {
1.2450 +public:
1.2451 + bool do_object_b(oop p) { return true; }
1.2452 +};
1.2453 +static PSAlwaysTrueClosure always_true;
1.2454 +
1.2455 +void PSParallelCompact::adjust_roots() {
1.2456 + // Adjust the pointers to reflect the new locations
1.2457 + GCTraceTime tm("adjust roots", print_phases(), true, &_gc_timer);
1.2458 +
1.2459 + // Need new claim bits when tracing through and adjusting pointers.
1.2460 + ClassLoaderDataGraph::clear_claimed_marks();
1.2461 +
1.2462 + // General strong roots.
1.2463 + Universe::oops_do(adjust_pointer_closure());
1.2464 + JNIHandles::oops_do(adjust_pointer_closure()); // Global (strong) JNI handles
1.2465 + CLDToOopClosure adjust_from_cld(adjust_pointer_closure());
1.2466 + Threads::oops_do(adjust_pointer_closure(), &adjust_from_cld, NULL);
1.2467 + ObjectSynchronizer::oops_do(adjust_pointer_closure());
1.2468 + FlatProfiler::oops_do(adjust_pointer_closure());
1.2469 + Management::oops_do(adjust_pointer_closure());
1.2470 + JvmtiExport::oops_do(adjust_pointer_closure());
1.2471 + // SO_AllClasses
1.2472 + SystemDictionary::oops_do(adjust_pointer_closure());
1.2473 + ClassLoaderDataGraph::oops_do(adjust_pointer_closure(), adjust_klass_closure(), true);
1.2474 +
1.2475 + // Now adjust pointers in remaining weak roots. (All of which should
1.2476 + // have been cleared if they pointed to non-surviving objects.)
1.2477 + // Global (weak) JNI handles
1.2478 + JNIHandles::weak_oops_do(&always_true, adjust_pointer_closure());
1.2479 +
1.2480 + CodeCache::oops_do(adjust_pointer_closure());
1.2481 + StringTable::oops_do(adjust_pointer_closure());
1.2482 + ref_processor()->weak_oops_do(adjust_pointer_closure());
1.2483 + // Roots were visited so references into the young gen in roots
1.2484 + // may have been scanned. Process them also.
1.2485 + // Should the reference processor have a span that excludes
1.2486 + // young gen objects?
1.2487 + PSScavenge::reference_processor()->weak_oops_do(adjust_pointer_closure());
1.2488 +}
1.2489 +
1.2490 +void PSParallelCompact::enqueue_region_draining_tasks(GCTaskQueue* q,
1.2491 + uint parallel_gc_threads)
1.2492 +{
1.2493 + GCTraceTime tm("drain task setup", print_phases(), true, &_gc_timer);
1.2494 +
1.2495 + // Find the threads that are active
1.2496 + unsigned int which = 0;
1.2497 +
1.2498 + const uint task_count = MAX2(parallel_gc_threads, 1U);
1.2499 + for (uint j = 0; j < task_count; j++) {
1.2500 + q->enqueue(new DrainStacksCompactionTask(j));
1.2501 + ParCompactionManager::verify_region_list_empty(j);
1.2502 + // Set the region stacks variables to "no" region stack values
1.2503 + // so that they will be recognized and needing a region stack
1.2504 + // in the stealing tasks if they do not get one by executing
1.2505 + // a draining stack.
1.2506 + ParCompactionManager* cm = ParCompactionManager::manager_array(j);
1.2507 + cm->set_region_stack(NULL);
1.2508 + cm->set_region_stack_index((uint)max_uintx);
1.2509 + }
1.2510 + ParCompactionManager::reset_recycled_stack_index();
1.2511 +
1.2512 + // Find all regions that are available (can be filled immediately) and
1.2513 + // distribute them to the thread stacks. The iteration is done in reverse
1.2514 + // order (high to low) so the regions will be removed in ascending order.
1.2515 +
1.2516 + const ParallelCompactData& sd = PSParallelCompact::summary_data();
1.2517 +
1.2518 + size_t fillable_regions = 0; // A count for diagnostic purposes.
1.2519 + // A region index which corresponds to the tasks created above.
1.2520 + // "which" must be 0 <= which < task_count
1.2521 +
1.2522 + which = 0;
1.2523 + // id + 1 is used to test termination so unsigned can
1.2524 + // be used with an old_space_id == 0.
1.2525 + for (unsigned int id = to_space_id; id + 1 > old_space_id; --id) {
1.2526 + SpaceInfo* const space_info = _space_info + id;
1.2527 + MutableSpace* const space = space_info->space();
1.2528 + HeapWord* const new_top = space_info->new_top();
1.2529 +
1.2530 + const size_t beg_region = sd.addr_to_region_idx(space_info->dense_prefix());
1.2531 + const size_t end_region =
1.2532 + sd.addr_to_region_idx(sd.region_align_up(new_top));
1.2533 +
1.2534 + for (size_t cur = end_region - 1; cur + 1 > beg_region; --cur) {
1.2535 + if (sd.region(cur)->claim_unsafe()) {
1.2536 + ParCompactionManager::region_list_push(which, cur);
1.2537 +
1.2538 + if (TraceParallelOldGCCompactionPhase && Verbose) {
1.2539 + const size_t count_mod_8 = fillable_regions & 7;
1.2540 + if (count_mod_8 == 0) gclog_or_tty->print("fillable: ");
1.2541 + gclog_or_tty->print(" " SIZE_FORMAT_W(7), cur);
1.2542 + if (count_mod_8 == 7) gclog_or_tty->cr();
1.2543 + }
1.2544 +
1.2545 + NOT_PRODUCT(++fillable_regions;)
1.2546 +
1.2547 + // Assign regions to tasks in round-robin fashion.
1.2548 + if (++which == task_count) {
1.2549 + assert(which <= parallel_gc_threads,
1.2550 + "Inconsistent number of workers");
1.2551 + which = 0;
1.2552 + }
1.2553 + }
1.2554 + }
1.2555 + }
1.2556 +
1.2557 + if (TraceParallelOldGCCompactionPhase) {
1.2558 + if (Verbose && (fillable_regions & 7) != 0) gclog_or_tty->cr();
1.2559 + gclog_or_tty->print_cr("%u initially fillable regions", fillable_regions);
1.2560 + }
1.2561 +}
1.2562 +
1.2563 +#define PAR_OLD_DENSE_PREFIX_OVER_PARTITIONING 4
1.2564 +
1.2565 +void PSParallelCompact::enqueue_dense_prefix_tasks(GCTaskQueue* q,
1.2566 + uint parallel_gc_threads) {
1.2567 + GCTraceTime tm("dense prefix task setup", print_phases(), true, &_gc_timer);
1.2568 +
1.2569 + ParallelCompactData& sd = PSParallelCompact::summary_data();
1.2570 +
1.2571 + // Iterate over all the spaces adding tasks for updating
1.2572 + // regions in the dense prefix. Assume that 1 gc thread
1.2573 + // will work on opening the gaps and the remaining gc threads
1.2574 + // will work on the dense prefix.
1.2575 + unsigned int space_id;
1.2576 + for (space_id = old_space_id; space_id < last_space_id; ++ space_id) {
1.2577 + HeapWord* const dense_prefix_end = _space_info[space_id].dense_prefix();
1.2578 + const MutableSpace* const space = _space_info[space_id].space();
1.2579 +
1.2580 + if (dense_prefix_end == space->bottom()) {
1.2581 + // There is no dense prefix for this space.
1.2582 + continue;
1.2583 + }
1.2584 +
1.2585 + // The dense prefix is before this region.
1.2586 + size_t region_index_end_dense_prefix =
1.2587 + sd.addr_to_region_idx(dense_prefix_end);
1.2588 + RegionData* const dense_prefix_cp =
1.2589 + sd.region(region_index_end_dense_prefix);
1.2590 + assert(dense_prefix_end == space->end() ||
1.2591 + dense_prefix_cp->available() ||
1.2592 + dense_prefix_cp->claimed(),
1.2593 + "The region after the dense prefix should always be ready to fill");
1.2594 +
1.2595 + size_t region_index_start = sd.addr_to_region_idx(space->bottom());
1.2596 +
1.2597 + // Is there dense prefix work?
1.2598 + size_t total_dense_prefix_regions =
1.2599 + region_index_end_dense_prefix - region_index_start;
1.2600 + // How many regions of the dense prefix should be given to
1.2601 + // each thread?
1.2602 + if (total_dense_prefix_regions > 0) {
1.2603 + uint tasks_for_dense_prefix = 1;
1.2604 + if (total_dense_prefix_regions <=
1.2605 + (parallel_gc_threads * PAR_OLD_DENSE_PREFIX_OVER_PARTITIONING)) {
1.2606 + // Don't over partition. This assumes that
1.2607 + // PAR_OLD_DENSE_PREFIX_OVER_PARTITIONING is a small integer value
1.2608 + // so there are not many regions to process.
1.2609 + tasks_for_dense_prefix = parallel_gc_threads;
1.2610 + } else {
1.2611 + // Over partition
1.2612 + tasks_for_dense_prefix = parallel_gc_threads *
1.2613 + PAR_OLD_DENSE_PREFIX_OVER_PARTITIONING;
1.2614 + }
1.2615 + size_t regions_per_thread = total_dense_prefix_regions /
1.2616 + tasks_for_dense_prefix;
1.2617 + // Give each thread at least 1 region.
1.2618 + if (regions_per_thread == 0) {
1.2619 + regions_per_thread = 1;
1.2620 + }
1.2621 +
1.2622 + for (uint k = 0; k < tasks_for_dense_prefix; k++) {
1.2623 + if (region_index_start >= region_index_end_dense_prefix) {
1.2624 + break;
1.2625 + }
1.2626 + // region_index_end is not processed
1.2627 + size_t region_index_end = MIN2(region_index_start + regions_per_thread,
1.2628 + region_index_end_dense_prefix);
1.2629 + q->enqueue(new UpdateDensePrefixTask(SpaceId(space_id),
1.2630 + region_index_start,
1.2631 + region_index_end));
1.2632 + region_index_start = region_index_end;
1.2633 + }
1.2634 + }
1.2635 + // This gets any part of the dense prefix that did not
1.2636 + // fit evenly.
1.2637 + if (region_index_start < region_index_end_dense_prefix) {
1.2638 + q->enqueue(new UpdateDensePrefixTask(SpaceId(space_id),
1.2639 + region_index_start,
1.2640 + region_index_end_dense_prefix));
1.2641 + }
1.2642 + }
1.2643 +}
1.2644 +
1.2645 +void PSParallelCompact::enqueue_region_stealing_tasks(
1.2646 + GCTaskQueue* q,
1.2647 + ParallelTaskTerminator* terminator_ptr,
1.2648 + uint parallel_gc_threads) {
1.2649 + GCTraceTime tm("steal task setup", print_phases(), true, &_gc_timer);
1.2650 +
1.2651 + // Once a thread has drained it's stack, it should try to steal regions from
1.2652 + // other threads.
1.2653 + if (parallel_gc_threads > 1) {
1.2654 + for (uint j = 0; j < parallel_gc_threads; j++) {
1.2655 + q->enqueue(new StealRegionCompactionTask(terminator_ptr));
1.2656 + }
1.2657 + }
1.2658 +}
1.2659 +
1.2660 +#ifdef ASSERT
1.2661 +// Write a histogram of the number of times the block table was filled for a
1.2662 +// region.
1.2663 +void PSParallelCompact::write_block_fill_histogram(outputStream* const out)
1.2664 +{
1.2665 + if (!TraceParallelOldGCCompactionPhase) return;
1.2666 +
1.2667 + typedef ParallelCompactData::RegionData rd_t;
1.2668 + ParallelCompactData& sd = summary_data();
1.2669 +
1.2670 + for (unsigned int id = old_space_id; id < last_space_id; ++id) {
1.2671 + MutableSpace* const spc = _space_info[id].space();
1.2672 + if (spc->bottom() != spc->top()) {
1.2673 + const rd_t* const beg = sd.addr_to_region_ptr(spc->bottom());
1.2674 + HeapWord* const top_aligned_up = sd.region_align_up(spc->top());
1.2675 + const rd_t* const end = sd.addr_to_region_ptr(top_aligned_up);
1.2676 +
1.2677 + size_t histo[5] = { 0, 0, 0, 0, 0 };
1.2678 + const size_t histo_len = sizeof(histo) / sizeof(size_t);
1.2679 + const size_t region_cnt = pointer_delta(end, beg, sizeof(rd_t));
1.2680 +
1.2681 + for (const rd_t* cur = beg; cur < end; ++cur) {
1.2682 + ++histo[MIN2(cur->blocks_filled_count(), histo_len - 1)];
1.2683 + }
1.2684 + out->print("%u %-4s" SIZE_FORMAT_W(5), id, space_names[id], region_cnt);
1.2685 + for (size_t i = 0; i < histo_len; ++i) {
1.2686 + out->print(" " SIZE_FORMAT_W(5) " %5.1f%%",
1.2687 + histo[i], 100.0 * histo[i] / region_cnt);
1.2688 + }
1.2689 + out->cr();
1.2690 + }
1.2691 + }
1.2692 +}
1.2693 +#endif // #ifdef ASSERT
1.2694 +
1.2695 +void PSParallelCompact::compact() {
1.2696 + // trace("5");
1.2697 + GCTraceTime tm("compaction phase", print_phases(), true, &_gc_timer);
1.2698 +
1.2699 + ParallelScavengeHeap* heap = (ParallelScavengeHeap*)Universe::heap();
1.2700 + assert(heap->kind() == CollectedHeap::ParallelScavengeHeap, "Sanity");
1.2701 + PSOldGen* old_gen = heap->old_gen();
1.2702 + old_gen->start_array()->reset();
1.2703 + uint parallel_gc_threads = heap->gc_task_manager()->workers();
1.2704 + uint active_gc_threads = heap->gc_task_manager()->active_workers();
1.2705 + TaskQueueSetSuper* qset = ParCompactionManager::region_array();
1.2706 + ParallelTaskTerminator terminator(active_gc_threads, qset);
1.2707 +
1.2708 + GCTaskQueue* q = GCTaskQueue::create();
1.2709 + enqueue_region_draining_tasks(q, active_gc_threads);
1.2710 + enqueue_dense_prefix_tasks(q, active_gc_threads);
1.2711 + enqueue_region_stealing_tasks(q, &terminator, active_gc_threads);
1.2712 +
1.2713 + {
1.2714 + GCTraceTime tm_pc("par compact", print_phases(), true, &_gc_timer);
1.2715 +
1.2716 + gc_task_manager()->execute_and_wait(q);
1.2717 +
1.2718 +#ifdef ASSERT
1.2719 + // Verify that all regions have been processed before the deferred updates.
1.2720 + for (unsigned int id = old_space_id; id < last_space_id; ++id) {
1.2721 + verify_complete(SpaceId(id));
1.2722 + }
1.2723 +#endif
1.2724 + }
1.2725 +
1.2726 + {
1.2727 + // Update the deferred objects, if any. Any compaction manager can be used.
1.2728 + GCTraceTime tm_du("deferred updates", print_phases(), true, &_gc_timer);
1.2729 + ParCompactionManager* cm = ParCompactionManager::manager_array(0);
1.2730 + for (unsigned int id = old_space_id; id < last_space_id; ++id) {
1.2731 + update_deferred_objects(cm, SpaceId(id));
1.2732 + }
1.2733 + }
1.2734 +
1.2735 + DEBUG_ONLY(write_block_fill_histogram(gclog_or_tty));
1.2736 +}
1.2737 +
1.2738 +#ifdef ASSERT
1.2739 +void PSParallelCompact::verify_complete(SpaceId space_id) {
1.2740 + // All Regions between space bottom() to new_top() should be marked as filled
1.2741 + // and all Regions between new_top() and top() should be available (i.e.,
1.2742 + // should have been emptied).
1.2743 + ParallelCompactData& sd = summary_data();
1.2744 + SpaceInfo si = _space_info[space_id];
1.2745 + HeapWord* new_top_addr = sd.region_align_up(si.new_top());
1.2746 + HeapWord* old_top_addr = sd.region_align_up(si.space()->top());
1.2747 + const size_t beg_region = sd.addr_to_region_idx(si.space()->bottom());
1.2748 + const size_t new_top_region = sd.addr_to_region_idx(new_top_addr);
1.2749 + const size_t old_top_region = sd.addr_to_region_idx(old_top_addr);
1.2750 +
1.2751 + bool issued_a_warning = false;
1.2752 +
1.2753 + size_t cur_region;
1.2754 + for (cur_region = beg_region; cur_region < new_top_region; ++cur_region) {
1.2755 + const RegionData* const c = sd.region(cur_region);
1.2756 + if (!c->completed()) {
1.2757 + warning("region " SIZE_FORMAT " not filled: "
1.2758 + "destination_count=" SIZE_FORMAT,
1.2759 + cur_region, c->destination_count());
1.2760 + issued_a_warning = true;
1.2761 + }
1.2762 + }
1.2763 +
1.2764 + for (cur_region = new_top_region; cur_region < old_top_region; ++cur_region) {
1.2765 + const RegionData* const c = sd.region(cur_region);
1.2766 + if (!c->available()) {
1.2767 + warning("region " SIZE_FORMAT " not empty: "
1.2768 + "destination_count=" SIZE_FORMAT,
1.2769 + cur_region, c->destination_count());
1.2770 + issued_a_warning = true;
1.2771 + }
1.2772 + }
1.2773 +
1.2774 + if (issued_a_warning) {
1.2775 + print_region_ranges();
1.2776 + }
1.2777 +}
1.2778 +#endif // #ifdef ASSERT
1.2779 +
1.2780 +// Update interior oops in the ranges of regions [beg_region, end_region).
1.2781 +void
1.2782 +PSParallelCompact::update_and_deadwood_in_dense_prefix(ParCompactionManager* cm,
1.2783 + SpaceId space_id,
1.2784 + size_t beg_region,
1.2785 + size_t end_region) {
1.2786 + ParallelCompactData& sd = summary_data();
1.2787 + ParMarkBitMap* const mbm = mark_bitmap();
1.2788 +
1.2789 + HeapWord* beg_addr = sd.region_to_addr(beg_region);
1.2790 + HeapWord* const end_addr = sd.region_to_addr(end_region);
1.2791 + assert(beg_region <= end_region, "bad region range");
1.2792 + assert(end_addr <= dense_prefix(space_id), "not in the dense prefix");
1.2793 +
1.2794 +#ifdef ASSERT
1.2795 + // Claim the regions to avoid triggering an assert when they are marked as
1.2796 + // filled.
1.2797 + for (size_t claim_region = beg_region; claim_region < end_region; ++claim_region) {
1.2798 + assert(sd.region(claim_region)->claim_unsafe(), "claim() failed");
1.2799 + }
1.2800 +#endif // #ifdef ASSERT
1.2801 +
1.2802 + if (beg_addr != space(space_id)->bottom()) {
1.2803 + // Find the first live object or block of dead space that *starts* in this
1.2804 + // range of regions. If a partial object crosses onto the region, skip it;
1.2805 + // it will be marked for 'deferred update' when the object head is
1.2806 + // processed. If dead space crosses onto the region, it is also skipped; it
1.2807 + // will be filled when the prior region is processed. If neither of those
1.2808 + // apply, the first word in the region is the start of a live object or dead
1.2809 + // space.
1.2810 + assert(beg_addr > space(space_id)->bottom(), "sanity");
1.2811 + const RegionData* const cp = sd.region(beg_region);
1.2812 + if (cp->partial_obj_size() != 0) {
1.2813 + beg_addr = sd.partial_obj_end(beg_region);
1.2814 + } else if (dead_space_crosses_boundary(cp, mbm->addr_to_bit(beg_addr))) {
1.2815 + beg_addr = mbm->find_obj_beg(beg_addr, end_addr);
1.2816 + }
1.2817 + }
1.2818 +
1.2819 + if (beg_addr < end_addr) {
1.2820 + // A live object or block of dead space starts in this range of Regions.
1.2821 + HeapWord* const dense_prefix_end = dense_prefix(space_id);
1.2822 +
1.2823 + // Create closures and iterate.
1.2824 + UpdateOnlyClosure update_closure(mbm, cm, space_id);
1.2825 + FillClosure fill_closure(cm, space_id);
1.2826 + ParMarkBitMap::IterationStatus status;
1.2827 + status = mbm->iterate(&update_closure, &fill_closure, beg_addr, end_addr,
1.2828 + dense_prefix_end);
1.2829 + if (status == ParMarkBitMap::incomplete) {
1.2830 + update_closure.do_addr(update_closure.source());
1.2831 + }
1.2832 + }
1.2833 +
1.2834 + // Mark the regions as filled.
1.2835 + RegionData* const beg_cp = sd.region(beg_region);
1.2836 + RegionData* const end_cp = sd.region(end_region);
1.2837 + for (RegionData* cp = beg_cp; cp < end_cp; ++cp) {
1.2838 + cp->set_completed();
1.2839 + }
1.2840 +}
1.2841 +
1.2842 +// Return the SpaceId for the space containing addr. If addr is not in the
1.2843 +// heap, last_space_id is returned. In debug mode it expects the address to be
1.2844 +// in the heap and asserts such.
1.2845 +PSParallelCompact::SpaceId PSParallelCompact::space_id(HeapWord* addr) {
1.2846 + assert(Universe::heap()->is_in_reserved(addr), "addr not in the heap");
1.2847 +
1.2848 + for (unsigned int id = old_space_id; id < last_space_id; ++id) {
1.2849 + if (_space_info[id].space()->contains(addr)) {
1.2850 + return SpaceId(id);
1.2851 + }
1.2852 + }
1.2853 +
1.2854 + assert(false, "no space contains the addr");
1.2855 + return last_space_id;
1.2856 +}
1.2857 +
1.2858 +void PSParallelCompact::update_deferred_objects(ParCompactionManager* cm,
1.2859 + SpaceId id) {
1.2860 + assert(id < last_space_id, "bad space id");
1.2861 +
1.2862 + ParallelCompactData& sd = summary_data();
1.2863 + const SpaceInfo* const space_info = _space_info + id;
1.2864 + ObjectStartArray* const start_array = space_info->start_array();
1.2865 +
1.2866 + const MutableSpace* const space = space_info->space();
1.2867 + assert(space_info->dense_prefix() >= space->bottom(), "dense_prefix not set");
1.2868 + HeapWord* const beg_addr = space_info->dense_prefix();
1.2869 + HeapWord* const end_addr = sd.region_align_up(space_info->new_top());
1.2870 +
1.2871 + const RegionData* const beg_region = sd.addr_to_region_ptr(beg_addr);
1.2872 + const RegionData* const end_region = sd.addr_to_region_ptr(end_addr);
1.2873 + const RegionData* cur_region;
1.2874 + for (cur_region = beg_region; cur_region < end_region; ++cur_region) {
1.2875 + HeapWord* const addr = cur_region->deferred_obj_addr();
1.2876 + if (addr != NULL) {
1.2877 + if (start_array != NULL) {
1.2878 + start_array->allocate_block(addr);
1.2879 + }
1.2880 + oop(addr)->update_contents(cm);
1.2881 + assert(oop(addr)->is_oop_or_null(), "should be an oop now");
1.2882 + }
1.2883 + }
1.2884 +}
1.2885 +
1.2886 +// Skip over count live words starting from beg, and return the address of the
1.2887 +// next live word. Unless marked, the word corresponding to beg is assumed to
1.2888 +// be dead. Callers must either ensure beg does not correspond to the middle of
1.2889 +// an object, or account for those live words in some other way. Callers must
1.2890 +// also ensure that there are enough live words in the range [beg, end) to skip.
1.2891 +HeapWord*
1.2892 +PSParallelCompact::skip_live_words(HeapWord* beg, HeapWord* end, size_t count)
1.2893 +{
1.2894 + assert(count > 0, "sanity");
1.2895 +
1.2896 + ParMarkBitMap* m = mark_bitmap();
1.2897 + idx_t bits_to_skip = m->words_to_bits(count);
1.2898 + idx_t cur_beg = m->addr_to_bit(beg);
1.2899 + const idx_t search_end = BitMap::word_align_up(m->addr_to_bit(end));
1.2900 +
1.2901 + do {
1.2902 + cur_beg = m->find_obj_beg(cur_beg, search_end);
1.2903 + idx_t cur_end = m->find_obj_end(cur_beg, search_end);
1.2904 + const size_t obj_bits = cur_end - cur_beg + 1;
1.2905 + if (obj_bits > bits_to_skip) {
1.2906 + return m->bit_to_addr(cur_beg + bits_to_skip);
1.2907 + }
1.2908 + bits_to_skip -= obj_bits;
1.2909 + cur_beg = cur_end + 1;
1.2910 + } while (bits_to_skip > 0);
1.2911 +
1.2912 + // Skipping the desired number of words landed just past the end of an object.
1.2913 + // Find the start of the next object.
1.2914 + cur_beg = m->find_obj_beg(cur_beg, search_end);
1.2915 + assert(cur_beg < m->addr_to_bit(end), "not enough live words to skip");
1.2916 + return m->bit_to_addr(cur_beg);
1.2917 +}
1.2918 +
1.2919 +HeapWord* PSParallelCompact::first_src_addr(HeapWord* const dest_addr,
1.2920 + SpaceId src_space_id,
1.2921 + size_t src_region_idx)
1.2922 +{
1.2923 + assert(summary_data().is_region_aligned(dest_addr), "not aligned");
1.2924 +
1.2925 + const SplitInfo& split_info = _space_info[src_space_id].split_info();
1.2926 + if (split_info.dest_region_addr() == dest_addr) {
1.2927 + // The partial object ending at the split point contains the first word to
1.2928 + // be copied to dest_addr.
1.2929 + return split_info.first_src_addr();
1.2930 + }
1.2931 +
1.2932 + const ParallelCompactData& sd = summary_data();
1.2933 + ParMarkBitMap* const bitmap = mark_bitmap();
1.2934 + const size_t RegionSize = ParallelCompactData::RegionSize;
1.2935 +
1.2936 + assert(sd.is_region_aligned(dest_addr), "not aligned");
1.2937 + const RegionData* const src_region_ptr = sd.region(src_region_idx);
1.2938 + const size_t partial_obj_size = src_region_ptr->partial_obj_size();
1.2939 + HeapWord* const src_region_destination = src_region_ptr->destination();
1.2940 +
1.2941 + assert(dest_addr >= src_region_destination, "wrong src region");
1.2942 + assert(src_region_ptr->data_size() > 0, "src region cannot be empty");
1.2943 +
1.2944 + HeapWord* const src_region_beg = sd.region_to_addr(src_region_idx);
1.2945 + HeapWord* const src_region_end = src_region_beg + RegionSize;
1.2946 +
1.2947 + HeapWord* addr = src_region_beg;
1.2948 + if (dest_addr == src_region_destination) {
1.2949 + // Return the first live word in the source region.
1.2950 + if (partial_obj_size == 0) {
1.2951 + addr = bitmap->find_obj_beg(addr, src_region_end);
1.2952 + assert(addr < src_region_end, "no objects start in src region");
1.2953 + }
1.2954 + return addr;
1.2955 + }
1.2956 +
1.2957 + // Must skip some live data.
1.2958 + size_t words_to_skip = dest_addr - src_region_destination;
1.2959 + assert(src_region_ptr->data_size() > words_to_skip, "wrong src region");
1.2960 +
1.2961 + if (partial_obj_size >= words_to_skip) {
1.2962 + // All the live words to skip are part of the partial object.
1.2963 + addr += words_to_skip;
1.2964 + if (partial_obj_size == words_to_skip) {
1.2965 + // Find the first live word past the partial object.
1.2966 + addr = bitmap->find_obj_beg(addr, src_region_end);
1.2967 + assert(addr < src_region_end, "wrong src region");
1.2968 + }
1.2969 + return addr;
1.2970 + }
1.2971 +
1.2972 + // Skip over the partial object (if any).
1.2973 + if (partial_obj_size != 0) {
1.2974 + words_to_skip -= partial_obj_size;
1.2975 + addr += partial_obj_size;
1.2976 + }
1.2977 +
1.2978 + // Skip over live words due to objects that start in the region.
1.2979 + addr = skip_live_words(addr, src_region_end, words_to_skip);
1.2980 + assert(addr < src_region_end, "wrong src region");
1.2981 + return addr;
1.2982 +}
1.2983 +
1.2984 +void PSParallelCompact::decrement_destination_counts(ParCompactionManager* cm,
1.2985 + SpaceId src_space_id,
1.2986 + size_t beg_region,
1.2987 + HeapWord* end_addr)
1.2988 +{
1.2989 + ParallelCompactData& sd = summary_data();
1.2990 +
1.2991 +#ifdef ASSERT
1.2992 + MutableSpace* const src_space = _space_info[src_space_id].space();
1.2993 + HeapWord* const beg_addr = sd.region_to_addr(beg_region);
1.2994 + assert(src_space->contains(beg_addr) || beg_addr == src_space->end(),
1.2995 + "src_space_id does not match beg_addr");
1.2996 + assert(src_space->contains(end_addr) || end_addr == src_space->end(),
1.2997 + "src_space_id does not match end_addr");
1.2998 +#endif // #ifdef ASSERT
1.2999 +
1.3000 + RegionData* const beg = sd.region(beg_region);
1.3001 + RegionData* const end = sd.addr_to_region_ptr(sd.region_align_up(end_addr));
1.3002 +
1.3003 + // Regions up to new_top() are enqueued if they become available.
1.3004 + HeapWord* const new_top = _space_info[src_space_id].new_top();
1.3005 + RegionData* const enqueue_end =
1.3006 + sd.addr_to_region_ptr(sd.region_align_up(new_top));
1.3007 +
1.3008 + for (RegionData* cur = beg; cur < end; ++cur) {
1.3009 + assert(cur->data_size() > 0, "region must have live data");
1.3010 + cur->decrement_destination_count();
1.3011 + if (cur < enqueue_end && cur->available() && cur->claim()) {
1.3012 + cm->push_region(sd.region(cur));
1.3013 + }
1.3014 + }
1.3015 +}
1.3016 +
1.3017 +size_t PSParallelCompact::next_src_region(MoveAndUpdateClosure& closure,
1.3018 + SpaceId& src_space_id,
1.3019 + HeapWord*& src_space_top,
1.3020 + HeapWord* end_addr)
1.3021 +{
1.3022 + typedef ParallelCompactData::RegionData RegionData;
1.3023 +
1.3024 + ParallelCompactData& sd = PSParallelCompact::summary_data();
1.3025 + const size_t region_size = ParallelCompactData::RegionSize;
1.3026 +
1.3027 + size_t src_region_idx = 0;
1.3028 +
1.3029 + // Skip empty regions (if any) up to the top of the space.
1.3030 + HeapWord* const src_aligned_up = sd.region_align_up(end_addr);
1.3031 + RegionData* src_region_ptr = sd.addr_to_region_ptr(src_aligned_up);
1.3032 + HeapWord* const top_aligned_up = sd.region_align_up(src_space_top);
1.3033 + const RegionData* const top_region_ptr =
1.3034 + sd.addr_to_region_ptr(top_aligned_up);
1.3035 + while (src_region_ptr < top_region_ptr && src_region_ptr->data_size() == 0) {
1.3036 + ++src_region_ptr;
1.3037 + }
1.3038 +
1.3039 + if (src_region_ptr < top_region_ptr) {
1.3040 + // The next source region is in the current space. Update src_region_idx
1.3041 + // and the source address to match src_region_ptr.
1.3042 + src_region_idx = sd.region(src_region_ptr);
1.3043 + HeapWord* const src_region_addr = sd.region_to_addr(src_region_idx);
1.3044 + if (src_region_addr > closure.source()) {
1.3045 + closure.set_source(src_region_addr);
1.3046 + }
1.3047 + return src_region_idx;
1.3048 + }
1.3049 +
1.3050 + // Switch to a new source space and find the first non-empty region.
1.3051 + unsigned int space_id = src_space_id + 1;
1.3052 + assert(space_id < last_space_id, "not enough spaces");
1.3053 +
1.3054 + HeapWord* const destination = closure.destination();
1.3055 +
1.3056 + do {
1.3057 + MutableSpace* space = _space_info[space_id].space();
1.3058 + HeapWord* const bottom = space->bottom();
1.3059 + const RegionData* const bottom_cp = sd.addr_to_region_ptr(bottom);
1.3060 +
1.3061 + // Iterate over the spaces that do not compact into themselves.
1.3062 + if (bottom_cp->destination() != bottom) {
1.3063 + HeapWord* const top_aligned_up = sd.region_align_up(space->top());
1.3064 + const RegionData* const top_cp = sd.addr_to_region_ptr(top_aligned_up);
1.3065 +
1.3066 + for (const RegionData* src_cp = bottom_cp; src_cp < top_cp; ++src_cp) {
1.3067 + if (src_cp->live_obj_size() > 0) {
1.3068 + // Found it.
1.3069 + assert(src_cp->destination() == destination,
1.3070 + "first live obj in the space must match the destination");
1.3071 + assert(src_cp->partial_obj_size() == 0,
1.3072 + "a space cannot begin with a partial obj");
1.3073 +
1.3074 + src_space_id = SpaceId(space_id);
1.3075 + src_space_top = space->top();
1.3076 + const size_t src_region_idx = sd.region(src_cp);
1.3077 + closure.set_source(sd.region_to_addr(src_region_idx));
1.3078 + return src_region_idx;
1.3079 + } else {
1.3080 + assert(src_cp->data_size() == 0, "sanity");
1.3081 + }
1.3082 + }
1.3083 + }
1.3084 + } while (++space_id < last_space_id);
1.3085 +
1.3086 + assert(false, "no source region was found");
1.3087 + return 0;
1.3088 +}
1.3089 +
1.3090 +void PSParallelCompact::fill_region(ParCompactionManager* cm, size_t region_idx)
1.3091 +{
1.3092 + typedef ParMarkBitMap::IterationStatus IterationStatus;
1.3093 + const size_t RegionSize = ParallelCompactData::RegionSize;
1.3094 + ParMarkBitMap* const bitmap = mark_bitmap();
1.3095 + ParallelCompactData& sd = summary_data();
1.3096 + RegionData* const region_ptr = sd.region(region_idx);
1.3097 +
1.3098 + // Get the items needed to construct the closure.
1.3099 + HeapWord* dest_addr = sd.region_to_addr(region_idx);
1.3100 + SpaceId dest_space_id = space_id(dest_addr);
1.3101 + ObjectStartArray* start_array = _space_info[dest_space_id].start_array();
1.3102 + HeapWord* new_top = _space_info[dest_space_id].new_top();
1.3103 + assert(dest_addr < new_top, "sanity");
1.3104 + const size_t words = MIN2(pointer_delta(new_top, dest_addr), RegionSize);
1.3105 +
1.3106 + // Get the source region and related info.
1.3107 + size_t src_region_idx = region_ptr->source_region();
1.3108 + SpaceId src_space_id = space_id(sd.region_to_addr(src_region_idx));
1.3109 + HeapWord* src_space_top = _space_info[src_space_id].space()->top();
1.3110 +
1.3111 + MoveAndUpdateClosure closure(bitmap, cm, start_array, dest_addr, words);
1.3112 + closure.set_source(first_src_addr(dest_addr, src_space_id, src_region_idx));
1.3113 +
1.3114 + // Adjust src_region_idx to prepare for decrementing destination counts (the
1.3115 + // destination count is not decremented when a region is copied to itself).
1.3116 + if (src_region_idx == region_idx) {
1.3117 + src_region_idx += 1;
1.3118 + }
1.3119 +
1.3120 + if (bitmap->is_unmarked(closure.source())) {
1.3121 + // The first source word is in the middle of an object; copy the remainder
1.3122 + // of the object or as much as will fit. The fact that pointer updates were
1.3123 + // deferred will be noted when the object header is processed.
1.3124 + HeapWord* const old_src_addr = closure.source();
1.3125 + closure.copy_partial_obj();
1.3126 + if (closure.is_full()) {
1.3127 + decrement_destination_counts(cm, src_space_id, src_region_idx,
1.3128 + closure.source());
1.3129 + region_ptr->set_deferred_obj_addr(NULL);
1.3130 + region_ptr->set_completed();
1.3131 + return;
1.3132 + }
1.3133 +
1.3134 + HeapWord* const end_addr = sd.region_align_down(closure.source());
1.3135 + if (sd.region_align_down(old_src_addr) != end_addr) {
1.3136 + // The partial object was copied from more than one source region.
1.3137 + decrement_destination_counts(cm, src_space_id, src_region_idx, end_addr);
1.3138 +
1.3139 + // Move to the next source region, possibly switching spaces as well. All
1.3140 + // args except end_addr may be modified.
1.3141 + src_region_idx = next_src_region(closure, src_space_id, src_space_top,
1.3142 + end_addr);
1.3143 + }
1.3144 + }
1.3145 +
1.3146 + do {
1.3147 + HeapWord* const cur_addr = closure.source();
1.3148 + HeapWord* const end_addr = MIN2(sd.region_align_up(cur_addr + 1),
1.3149 + src_space_top);
1.3150 + IterationStatus status = bitmap->iterate(&closure, cur_addr, end_addr);
1.3151 +
1.3152 + if (status == ParMarkBitMap::incomplete) {
1.3153 + // The last obj that starts in the source region does not end in the
1.3154 + // region.
1.3155 + assert(closure.source() < end_addr, "sanity");
1.3156 + HeapWord* const obj_beg = closure.source();
1.3157 + HeapWord* const range_end = MIN2(obj_beg + closure.words_remaining(),
1.3158 + src_space_top);
1.3159 + HeapWord* const obj_end = bitmap->find_obj_end(obj_beg, range_end);
1.3160 + if (obj_end < range_end) {
1.3161 + // The end was found; the entire object will fit.
1.3162 + status = closure.do_addr(obj_beg, bitmap->obj_size(obj_beg, obj_end));
1.3163 + assert(status != ParMarkBitMap::would_overflow, "sanity");
1.3164 + } else {
1.3165 + // The end was not found; the object will not fit.
1.3166 + assert(range_end < src_space_top, "obj cannot cross space boundary");
1.3167 + status = ParMarkBitMap::would_overflow;
1.3168 + }
1.3169 + }
1.3170 +
1.3171 + if (status == ParMarkBitMap::would_overflow) {
1.3172 + // The last object did not fit. Note that interior oop updates were
1.3173 + // deferred, then copy enough of the object to fill the region.
1.3174 + region_ptr->set_deferred_obj_addr(closure.destination());
1.3175 + status = closure.copy_until_full(); // copies from closure.source()
1.3176 +
1.3177 + decrement_destination_counts(cm, src_space_id, src_region_idx,
1.3178 + closure.source());
1.3179 + region_ptr->set_completed();
1.3180 + return;
1.3181 + }
1.3182 +
1.3183 + if (status == ParMarkBitMap::full) {
1.3184 + decrement_destination_counts(cm, src_space_id, src_region_idx,
1.3185 + closure.source());
1.3186 + region_ptr->set_deferred_obj_addr(NULL);
1.3187 + region_ptr->set_completed();
1.3188 + return;
1.3189 + }
1.3190 +
1.3191 + decrement_destination_counts(cm, src_space_id, src_region_idx, end_addr);
1.3192 +
1.3193 + // Move to the next source region, possibly switching spaces as well. All
1.3194 + // args except end_addr may be modified.
1.3195 + src_region_idx = next_src_region(closure, src_space_id, src_space_top,
1.3196 + end_addr);
1.3197 + } while (true);
1.3198 +}
1.3199 +
1.3200 +void PSParallelCompact::fill_blocks(size_t region_idx)
1.3201 +{
1.3202 + // Fill in the block table elements for the specified region. Each block
1.3203 + // table element holds the number of live words in the region that are to the
1.3204 + // left of the first object that starts in the block. Thus only blocks in
1.3205 + // which an object starts need to be filled.
1.3206 + //
1.3207 + // The algorithm scans the section of the bitmap that corresponds to the
1.3208 + // region, keeping a running total of the live words. When an object start is
1.3209 + // found, if it's the first to start in the block that contains it, the
1.3210 + // current total is written to the block table element.
1.3211 + const size_t Log2BlockSize = ParallelCompactData::Log2BlockSize;
1.3212 + const size_t Log2RegionSize = ParallelCompactData::Log2RegionSize;
1.3213 + const size_t RegionSize = ParallelCompactData::RegionSize;
1.3214 +
1.3215 + ParallelCompactData& sd = summary_data();
1.3216 + const size_t partial_obj_size = sd.region(region_idx)->partial_obj_size();
1.3217 + if (partial_obj_size >= RegionSize) {
1.3218 + return; // No objects start in this region.
1.3219 + }
1.3220 +
1.3221 + // Ensure the first loop iteration decides that the block has changed.
1.3222 + size_t cur_block = sd.block_count();
1.3223 +
1.3224 + const ParMarkBitMap* const bitmap = mark_bitmap();
1.3225 +
1.3226 + const size_t Log2BitsPerBlock = Log2BlockSize - LogMinObjAlignment;
1.3227 + assert((size_t)1 << Log2BitsPerBlock ==
1.3228 + bitmap->words_to_bits(ParallelCompactData::BlockSize), "sanity");
1.3229 +
1.3230 + size_t beg_bit = bitmap->words_to_bits(region_idx << Log2RegionSize);
1.3231 + const size_t range_end = beg_bit + bitmap->words_to_bits(RegionSize);
1.3232 + size_t live_bits = bitmap->words_to_bits(partial_obj_size);
1.3233 + beg_bit = bitmap->find_obj_beg(beg_bit + live_bits, range_end);
1.3234 + while (beg_bit < range_end) {
1.3235 + const size_t new_block = beg_bit >> Log2BitsPerBlock;
1.3236 + if (new_block != cur_block) {
1.3237 + cur_block = new_block;
1.3238 + sd.block(cur_block)->set_offset(bitmap->bits_to_words(live_bits));
1.3239 + }
1.3240 +
1.3241 + const size_t end_bit = bitmap->find_obj_end(beg_bit, range_end);
1.3242 + if (end_bit < range_end - 1) {
1.3243 + live_bits += end_bit - beg_bit + 1;
1.3244 + beg_bit = bitmap->find_obj_beg(end_bit + 1, range_end);
1.3245 + } else {
1.3246 + return;
1.3247 + }
1.3248 + }
1.3249 +}
1.3250 +
1.3251 +void
1.3252 +PSParallelCompact::move_and_update(ParCompactionManager* cm, SpaceId space_id) {
1.3253 + const MutableSpace* sp = space(space_id);
1.3254 + if (sp->is_empty()) {
1.3255 + return;
1.3256 + }
1.3257 +
1.3258 + ParallelCompactData& sd = PSParallelCompact::summary_data();
1.3259 + ParMarkBitMap* const bitmap = mark_bitmap();
1.3260 + HeapWord* const dp_addr = dense_prefix(space_id);
1.3261 + HeapWord* beg_addr = sp->bottom();
1.3262 + HeapWord* end_addr = sp->top();
1.3263 +
1.3264 + assert(beg_addr <= dp_addr && dp_addr <= end_addr, "bad dense prefix");
1.3265 +
1.3266 + const size_t beg_region = sd.addr_to_region_idx(beg_addr);
1.3267 + const size_t dp_region = sd.addr_to_region_idx(dp_addr);
1.3268 + if (beg_region < dp_region) {
1.3269 + update_and_deadwood_in_dense_prefix(cm, space_id, beg_region, dp_region);
1.3270 + }
1.3271 +
1.3272 + // The destination of the first live object that starts in the region is one
1.3273 + // past the end of the partial object entering the region (if any).
1.3274 + HeapWord* const dest_addr = sd.partial_obj_end(dp_region);
1.3275 + HeapWord* const new_top = _space_info[space_id].new_top();
1.3276 + assert(new_top >= dest_addr, "bad new_top value");
1.3277 + const size_t words = pointer_delta(new_top, dest_addr);
1.3278 +
1.3279 + if (words > 0) {
1.3280 + ObjectStartArray* start_array = _space_info[space_id].start_array();
1.3281 + MoveAndUpdateClosure closure(bitmap, cm, start_array, dest_addr, words);
1.3282 +
1.3283 + ParMarkBitMap::IterationStatus status;
1.3284 + status = bitmap->iterate(&closure, dest_addr, end_addr);
1.3285 + assert(status == ParMarkBitMap::full, "iteration not complete");
1.3286 + assert(bitmap->find_obj_beg(closure.source(), end_addr) == end_addr,
1.3287 + "live objects skipped because closure is full");
1.3288 + }
1.3289 +}
1.3290 +
1.3291 +jlong PSParallelCompact::millis_since_last_gc() {
1.3292 + // We need a monotonically non-deccreasing time in ms but
1.3293 + // os::javaTimeMillis() does not guarantee monotonicity.
1.3294 + jlong now = os::javaTimeNanos() / NANOSECS_PER_MILLISEC;
1.3295 + jlong ret_val = now - _time_of_last_gc;
1.3296 + // XXX See note in genCollectedHeap::millis_since_last_gc().
1.3297 + if (ret_val < 0) {
1.3298 + NOT_PRODUCT(warning("time warp: "INT64_FORMAT, ret_val);)
1.3299 + return 0;
1.3300 + }
1.3301 + return ret_val;
1.3302 +}
1.3303 +
1.3304 +void PSParallelCompact::reset_millis_since_last_gc() {
1.3305 + // We need a monotonically non-deccreasing time in ms but
1.3306 + // os::javaTimeMillis() does not guarantee monotonicity.
1.3307 + _time_of_last_gc = os::javaTimeNanos() / NANOSECS_PER_MILLISEC;
1.3308 +}
1.3309 +
1.3310 +ParMarkBitMap::IterationStatus MoveAndUpdateClosure::copy_until_full()
1.3311 +{
1.3312 + if (source() != destination()) {
1.3313 + DEBUG_ONLY(PSParallelCompact::check_new_location(source(), destination());)
1.3314 + Copy::aligned_conjoint_words(source(), destination(), words_remaining());
1.3315 + }
1.3316 + update_state(words_remaining());
1.3317 + assert(is_full(), "sanity");
1.3318 + return ParMarkBitMap::full;
1.3319 +}
1.3320 +
1.3321 +void MoveAndUpdateClosure::copy_partial_obj()
1.3322 +{
1.3323 + size_t words = words_remaining();
1.3324 +
1.3325 + HeapWord* const range_end = MIN2(source() + words, bitmap()->region_end());
1.3326 + HeapWord* const end_addr = bitmap()->find_obj_end(source(), range_end);
1.3327 + if (end_addr < range_end) {
1.3328 + words = bitmap()->obj_size(source(), end_addr);
1.3329 + }
1.3330 +
1.3331 + // This test is necessary; if omitted, the pointer updates to a partial object
1.3332 + // that crosses the dense prefix boundary could be overwritten.
1.3333 + if (source() != destination()) {
1.3334 + DEBUG_ONLY(PSParallelCompact::check_new_location(source(), destination());)
1.3335 + Copy::aligned_conjoint_words(source(), destination(), words);
1.3336 + }
1.3337 + update_state(words);
1.3338 +}
1.3339 +
1.3340 +ParMarkBitMapClosure::IterationStatus
1.3341 +MoveAndUpdateClosure::do_addr(HeapWord* addr, size_t words) {
1.3342 + assert(destination() != NULL, "sanity");
1.3343 + assert(bitmap()->obj_size(addr) == words, "bad size");
1.3344 +
1.3345 + _source = addr;
1.3346 + assert(PSParallelCompact::summary_data().calc_new_pointer(source()) ==
1.3347 + destination(), "wrong destination");
1.3348 +
1.3349 + if (words > words_remaining()) {
1.3350 + return ParMarkBitMap::would_overflow;
1.3351 + }
1.3352 +
1.3353 + // The start_array must be updated even if the object is not moving.
1.3354 + if (_start_array != NULL) {
1.3355 + _start_array->allocate_block(destination());
1.3356 + }
1.3357 +
1.3358 + if (destination() != source()) {
1.3359 + DEBUG_ONLY(PSParallelCompact::check_new_location(source(), destination());)
1.3360 + Copy::aligned_conjoint_words(source(), destination(), words);
1.3361 + }
1.3362 +
1.3363 + oop moved_oop = (oop) destination();
1.3364 + moved_oop->update_contents(compaction_manager());
1.3365 + assert(moved_oop->is_oop_or_null(), "Object should be whole at this point");
1.3366 +
1.3367 + update_state(words);
1.3368 + assert(destination() == (HeapWord*)moved_oop + moved_oop->size(), "sanity");
1.3369 + return is_full() ? ParMarkBitMap::full : ParMarkBitMap::incomplete;
1.3370 +}
1.3371 +
1.3372 +UpdateOnlyClosure::UpdateOnlyClosure(ParMarkBitMap* mbm,
1.3373 + ParCompactionManager* cm,
1.3374 + PSParallelCompact::SpaceId space_id) :
1.3375 + ParMarkBitMapClosure(mbm, cm),
1.3376 + _space_id(space_id),
1.3377 + _start_array(PSParallelCompact::start_array(space_id))
1.3378 +{
1.3379 +}
1.3380 +
1.3381 +// Updates the references in the object to their new values.
1.3382 +ParMarkBitMapClosure::IterationStatus
1.3383 +UpdateOnlyClosure::do_addr(HeapWord* addr, size_t words) {
1.3384 + do_addr(addr);
1.3385 + return ParMarkBitMap::incomplete;
1.3386 +}
