1 /* 2 * Copyright (c) 2013, 2016, Oracle and/or its affiliates. All rights reserved. 3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. 4 * 5 * This code is free software; you can redistribute it and/or modify it 6 * under the terms of the GNU General Public License version 2 only, as 7 * published by the Free Software Foundation. 8 * 9 * This code is distributed in the hope that it will be useful, but WITHOUT 10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or 11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License 12 * version 2 for more details (a copy is included in the LICENSE file that 13 * accompanied this code). 14 * 15 * You should have received a copy of the GNU General Public License version 16 * 2 along with this work; if not, write to the Free Software Foundation, 17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA. 18 * 19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA 20 * or visit www.oracle.com if you need additional information or have any 21 * questions. 22 * 23 */ 24 25 #include "precompiled.hpp" 26 #include "gc/g1/concurrentG1Refine.hpp" 27 #include "gc/g1/g1CollectedHeap.inline.hpp" 28 #include "gc/g1/g1GCPhaseTimes.hpp" 29 #include "gc/g1/g1StringDedup.hpp" 30 #include "gc/g1/workerDataArray.inline.hpp" 31 #include "memory/resourceArea.hpp" 32 #include "logging/log.hpp" 33 #include "runtime/timer.hpp" 34 #include "runtime/os.hpp" 35 36 static const char* Indents[5] = {"", " ", " ", " ", " "}; 37 38 G1GCPhaseTimes::G1GCPhaseTimes(uint max_gc_threads) : 39 _max_gc_threads(max_gc_threads) 40 { 41 assert(max_gc_threads > 0, "Must have some GC threads"); 42 43 _gc_par_phases[GCWorkerStart] = new WorkerDataArray<double>(max_gc_threads, "GC Worker Start (ms):"); 44 _gc_par_phases[ExtRootScan] = new WorkerDataArray<double>(max_gc_threads, "Ext Root Scanning (ms):"); 45 46 // Root scanning phases 47 _gc_par_phases[ThreadRoots] = new WorkerDataArray<double>(max_gc_threads, "Thread Roots (ms):"); 48 _gc_par_phases[StringTableRoots] = new WorkerDataArray<double>(max_gc_threads, "StringTable Roots (ms):"); 49 _gc_par_phases[UniverseRoots] = new WorkerDataArray<double>(max_gc_threads, "Universe Roots (ms):"); 50 _gc_par_phases[JNIRoots] = new WorkerDataArray<double>(max_gc_threads, "JNI Handles Roots (ms):"); 51 _gc_par_phases[ObjectSynchronizerRoots] = new WorkerDataArray<double>(max_gc_threads, "ObjectSynchronizer Roots (ms):"); 52 _gc_par_phases[FlatProfilerRoots] = new WorkerDataArray<double>(max_gc_threads, "FlatProfiler Roots (ms):"); 53 _gc_par_phases[ManagementRoots] = new WorkerDataArray<double>(max_gc_threads, "Management Roots (ms):"); 54 _gc_par_phases[SystemDictionaryRoots] = new WorkerDataArray<double>(max_gc_threads, "SystemDictionary Roots (ms):"); 55 _gc_par_phases[CLDGRoots] = new WorkerDataArray<double>(max_gc_threads, "CLDG Roots (ms):"); 56 _gc_par_phases[JVMTIRoots] = new WorkerDataArray<double>(max_gc_threads, "JVMTI Roots (ms):"); 57 _gc_par_phases[CMRefRoots] = new WorkerDataArray<double>(max_gc_threads, "CM RefProcessor Roots (ms):"); 58 _gc_par_phases[WaitForStrongCLD] = new WorkerDataArray<double>(max_gc_threads, "Wait For Strong CLD (ms):"); 59 _gc_par_phases[WeakCLDRoots] = new WorkerDataArray<double>(max_gc_threads, "Weak CLD Roots (ms):"); 60 _gc_par_phases[SATBFiltering] = new WorkerDataArray<double>(max_gc_threads, "SATB Filtering (ms):"); 61 62 _gc_par_phases[UpdateRS] = new WorkerDataArray<double>(max_gc_threads, "Update RS (ms):"); 63 if (ConcurrentG1Refine::hot_card_cache_enabled()) { 64 _gc_par_phases[ScanHCC] = new WorkerDataArray<double>(max_gc_threads, "Scan HCC (ms):"); 65 } else { 66 _gc_par_phases[ScanHCC] = NULL; 67 } 68 _gc_par_phases[ScanRS] = new WorkerDataArray<double>(max_gc_threads, "Scan RS (ms):"); 69 _gc_par_phases[CodeRoots] = new WorkerDataArray<double>(max_gc_threads, "Code Root Scanning (ms):"); 70 _gc_par_phases[ObjCopy] = new WorkerDataArray<double>(max_gc_threads, "Object Copy (ms):"); 71 _gc_par_phases[Termination] = new WorkerDataArray<double>(max_gc_threads, "Termination (ms):"); 72 _gc_par_phases[GCWorkerTotal] = new WorkerDataArray<double>(max_gc_threads, "GC Worker Total (ms):"); 73 _gc_par_phases[GCWorkerEnd] = new WorkerDataArray<double>(max_gc_threads, "GC Worker End (ms):"); 74 _gc_par_phases[Other] = new WorkerDataArray<double>(max_gc_threads, "GC Worker Other (ms):"); 75 76 _update_rs_processed_buffers = new WorkerDataArray<size_t>(max_gc_threads, "Processed Buffers:"); 77 _gc_par_phases[UpdateRS]->link_thread_work_items(_update_rs_processed_buffers); 78 79 _termination_attempts = new WorkerDataArray<size_t>(max_gc_threads, "Termination Attempts:"); 80 _gc_par_phases[Termination]->link_thread_work_items(_termination_attempts); 81 82 if (UseStringDeduplication) { 83 _gc_par_phases[StringDedupQueueFixup] = new WorkerDataArray<double>(max_gc_threads, "Queue Fixup (ms):"); 84 _gc_par_phases[StringDedupTableFixup] = new WorkerDataArray<double>(max_gc_threads, "Table Fixup (ms):"); 85 } else { 86 _gc_par_phases[StringDedupQueueFixup] = NULL; 87 _gc_par_phases[StringDedupTableFixup] = NULL; 88 } 89 90 _gc_par_phases[RedirtyCards] = new WorkerDataArray<double>(max_gc_threads, "Parallel Redirty (ms):"); 91 _redirtied_cards = new WorkerDataArray<size_t>(max_gc_threads, "Redirtied Cards:"); 92 _gc_par_phases[RedirtyCards]->link_thread_work_items(_redirtied_cards); 93 94 _gc_par_phases[PreserveCMReferents] = new WorkerDataArray<double>(max_gc_threads, "Parallel Preserve CM Refs (ms):"); 95 } 96 97 void G1GCPhaseTimes::note_gc_start() { 98 _gc_start_counter = os::elapsed_counter(); 99 _cur_expand_heap_time_ms = 0.0; 100 _external_accounted_time_ms = 0.0; 101 102 for (int i = 0; i < GCParPhasesSentinel; i++) { 103 if (_gc_par_phases[i] != NULL) { 104 _gc_par_phases[i]->reset(); 105 } 106 } 107 } 108 109 #define ASSERT_PHASE_UNINITILIAZED(phase) \ 110 assert(_gc_par_phases[phase]->get(i) == uninitialized, "Phase " #phase " reported for thread that was not started"); 111 112 double G1GCPhaseTimes::worker_time(GCParPhases phase, uint worker) { 113 double value = _gc_par_phases[phase]->get(worker); 114 if (value != WorkerDataArray<double>::uninitialized()) { 115 return value; 116 } 117 return 0.0; 118 } 119 120 void G1GCPhaseTimes::note_gc_end() { 121 _gc_pause_time_ms = TimeHelper::counter_to_millis(os::elapsed_counter() - _gc_start_counter); 122 123 double uninitialized = WorkerDataArray<double>::uninitialized(); 124 125 for (uint i = 0; i < _max_gc_threads; i++) { 126 double worker_start = _gc_par_phases[GCWorkerStart]->get(i); 127 if (worker_start != uninitialized) { 128 assert(_gc_par_phases[GCWorkerEnd]->get(i) != uninitialized, "Worker started but not ended."); 129 double total_worker_time = _gc_par_phases[GCWorkerEnd]->get(i) - _gc_par_phases[GCWorkerStart]->get(i); 130 record_time_secs(GCWorkerTotal, i , total_worker_time); 131 132 double worker_known_time = 133 worker_time(ExtRootScan, i) 134 + worker_time(SATBFiltering, i) 135 + worker_time(UpdateRS, i) 136 + worker_time(ScanRS, i) 137 + worker_time(CodeRoots, i) 138 + worker_time(ObjCopy, i) 139 + worker_time(Termination, i); 140 141 record_time_secs(Other, i, total_worker_time - worker_known_time); 142 } else { 143 // Make sure all slots are uninitialized since this thread did not seem to have been started 144 ASSERT_PHASE_UNINITILIAZED(GCWorkerEnd); 145 ASSERT_PHASE_UNINITILIAZED(ExtRootScan); 146 ASSERT_PHASE_UNINITILIAZED(SATBFiltering); 147 ASSERT_PHASE_UNINITILIAZED(UpdateRS); 148 ASSERT_PHASE_UNINITILIAZED(ScanRS); 149 ASSERT_PHASE_UNINITILIAZED(CodeRoots); 150 ASSERT_PHASE_UNINITILIAZED(ObjCopy); 151 ASSERT_PHASE_UNINITILIAZED(Termination); 152 } 153 } 154 } 155 156 #undef ASSERT_PHASE_UNINITILIAZED 157 158 // record the time a phase took in seconds 159 void G1GCPhaseTimes::record_time_secs(GCParPhases phase, uint worker_i, double secs) { 160 _gc_par_phases[phase]->set(worker_i, secs); 161 } 162 163 // add a number of seconds to a phase 164 void G1GCPhaseTimes::add_time_secs(GCParPhases phase, uint worker_i, double secs) { 165 _gc_par_phases[phase]->add(worker_i, secs); 166 } 167 168 void G1GCPhaseTimes::record_thread_work_item(GCParPhases phase, uint worker_i, size_t count) { 169 _gc_par_phases[phase]->set_thread_work_item(worker_i, count); 170 } 171 172 // return the average time for a phase in milliseconds 173 double G1GCPhaseTimes::average_time_ms(GCParPhases phase) { 174 return _gc_par_phases[phase]->average() * 1000.0; 175 } 176 177 size_t G1GCPhaseTimes::sum_thread_work_items(GCParPhases phase) { 178 assert(_gc_par_phases[phase]->thread_work_items() != NULL, "No sub count"); 179 return _gc_par_phases[phase]->thread_work_items()->sum(); 180 } 181 182 template <class T> 183 void G1GCPhaseTimes::details(T* phase, const char* indent) { 184 LogHandle(gc, phases, task) log; 185 if (log.is_level(LogLevel::Trace)) { 186 outputStream* trace_out = log.trace_stream(); 187 trace_out->print("%s", indent); 188 phase->print_details_on(trace_out); 189 } 190 } 191 192 void G1GCPhaseTimes::log_phase(WorkerDataArray<double>* phase, uint indent, outputStream* out, bool print_sum) { 193 out->print("%s", Indents[indent]); 194 phase->print_summary_on(out, print_sum); 195 details(phase, Indents[indent]); 196 197 WorkerDataArray<size_t>* work_items = phase->thread_work_items(); 198 if (work_items != NULL) { 199 out->print("%s", Indents[indent + 1]); 200 work_items->print_summary_on(out, true); 201 details(work_items, Indents[indent + 1]); 202 } 203 } 204 205 void G1GCPhaseTimes::debug_phase(WorkerDataArray<double>* phase) { 206 LogHandle(gc, phases) log; 207 if (log.is_level(LogLevel::Debug)) { 208 ResourceMark rm; 209 log_phase(phase, 2, log.debug_stream(), true); 210 } 211 } 212 213 void G1GCPhaseTimes::trace_phase(WorkerDataArray<double>* phase, bool print_sum) { 214 LogHandle(gc, phases) log; 215 if (log.is_level(LogLevel::Trace)) { 216 ResourceMark rm; 217 log_phase(phase, 3, log.trace_stream(), print_sum); 218 } 219 } 220 221 #define PHASE_DOUBLE_FORMAT "%s%s: %.1lfms" 222 #define PHASE_SIZE_FORMAT "%s%s: " SIZE_FORMAT 223 224 #define info_line(str, value) \ 225 log_info(gc, phases)(PHASE_DOUBLE_FORMAT, Indents[1], str, value); 226 227 #define debug_line(str, value) \ 228 log_debug(gc, phases)(PHASE_DOUBLE_FORMAT, Indents[2], str, value); 229 230 #define trace_line(str, value) \ 231 log_trace(gc, phases)(PHASE_DOUBLE_FORMAT, Indents[3], str, value); 232 233 #define trace_line_sz(str, value) \ 234 log_trace(gc, phases)(PHASE_SIZE_FORMAT, Indents[3], str, value); 235 236 #define trace_line_ms(str, value) \ 237 log_trace(gc, phases)(PHASE_SIZE_FORMAT, Indents[3], str, value); 238 239 #define info_line_and_account(str, value) \ 240 info_line(str, value); \ 241 accounted_time_ms += value; 242 243 void G1GCPhaseTimes::print() { 244 note_gc_end(); 245 246 double accounted_time_ms = _external_accounted_time_ms; 247 if (_root_region_scan_wait_time_ms > 0.0) { 248 info_line_and_account("Root Region Scan Waiting", _root_region_scan_wait_time_ms); 249 } 250 251 info_line_and_account("Evacuate Collection Set", _cur_collection_par_time_ms); 252 trace_phase(_gc_par_phases[GCWorkerStart], false); 253 debug_phase(_gc_par_phases[ExtRootScan]); 254 for (int i = ThreadRoots; i <= SATBFiltering; i++) { 255 trace_phase(_gc_par_phases[i]); 256 } 257 debug_phase(_gc_par_phases[UpdateRS]); 258 if (ConcurrentG1Refine::hot_card_cache_enabled()) { 259 trace_phase(_gc_par_phases[ScanHCC]); 260 } 261 debug_phase(_gc_par_phases[ScanRS]); 262 debug_phase(_gc_par_phases[CodeRoots]); 263 debug_phase(_gc_par_phases[ObjCopy]); 264 debug_phase(_gc_par_phases[Termination]); 265 debug_phase(_gc_par_phases[Other]); 266 debug_phase(_gc_par_phases[GCWorkerTotal]); 267 trace_phase(_gc_par_phases[GCWorkerEnd], false); 268 269 info_line_and_account("Code Roots", _cur_collection_code_root_fixup_time_ms + _cur_strong_code_root_purge_time_ms); 270 debug_line("Code Roots Fixup", _cur_collection_code_root_fixup_time_ms); 271 debug_line("Code Roots Purge", _cur_strong_code_root_purge_time_ms); 272 273 if (G1StringDedup::is_enabled()) { 274 info_line_and_account("String Dedup Fixup", _cur_string_dedup_fixup_time_ms); 275 debug_phase(_gc_par_phases[StringDedupQueueFixup]); 276 debug_phase(_gc_par_phases[StringDedupTableFixup]); 277 } 278 info_line_and_account("Clear Card Table", _cur_clear_ct_time_ms); 279 info_line_and_account("Expand Heap After Collection", _cur_expand_heap_time_ms); 280 281 double free_cset_time = _recorded_young_free_cset_time_ms + _recorded_non_young_free_cset_time_ms; 282 info_line_and_account("Free Collection Set", free_cset_time); 283 debug_line("Young Free Collection Set", _recorded_young_free_cset_time_ms); 284 debug_line("Non-Young Free Collection Set", _recorded_non_young_free_cset_time_ms); 285 info_line_and_account("Merge Per-Thread State", _recorded_merge_pss_time_ms); 286 287 info_line("Other", _gc_pause_time_ms - accounted_time_ms); 288 if (_cur_verify_before_time_ms > 0.0) { 289 debug_line("Verify Before", _cur_verify_before_time_ms); 290 } 291 if (G1CollectedHeap::heap()->evacuation_failed()) { 292 double evac_fail_handling = _cur_evac_fail_recalc_used + _cur_evac_fail_remove_self_forwards + 293 _cur_evac_fail_restore_remsets; 294 debug_line("Evacuation Failure", evac_fail_handling); 295 trace_line("Recalculate Used", _cur_evac_fail_recalc_used); 296 trace_line("Remove Self Forwards",_cur_evac_fail_remove_self_forwards); 297 trace_line("Restore RemSet", _cur_evac_fail_restore_remsets); 298 } 299 debug_line("Choose CSet", (_recorded_young_cset_choice_time_ms + _recorded_non_young_cset_choice_time_ms)); 300 debug_line("Preserve CM Refs", _recorded_preserve_cm_referents_time_ms); 301 debug_line("Reference Processing", _cur_ref_proc_time_ms); 302 debug_line("Reference Enqueuing", _cur_ref_enq_time_ms); 303 debug_line("Redirty Cards", _recorded_redirty_logged_cards_time_ms); 304 trace_phase(_gc_par_phases[RedirtyCards]); 305 trace_phase(_gc_par_phases[PreserveCMReferents]); 306 if (G1EagerReclaimHumongousObjects) { 307 debug_line("Humongous Register", _cur_fast_reclaim_humongous_register_time_ms); 308 trace_line_sz("Humongous Total", _cur_fast_reclaim_humongous_total); 309 trace_line_sz("Humongous Candidate", _cur_fast_reclaim_humongous_candidates); 310 debug_line("Humongous Reclaim", _cur_fast_reclaim_humongous_time_ms); 311 trace_line_sz("Humongous Reclaimed", _cur_fast_reclaim_humongous_reclaimed); 312 } 313 if (_cur_verify_after_time_ms > 0.0) { 314 debug_line("Verify After", _cur_verify_after_time_ms); 315 } 316 } 317 318 G1GCParPhaseTimesTracker::G1GCParPhaseTimesTracker(G1GCPhaseTimes* phase_times, G1GCPhaseTimes::GCParPhases phase, uint worker_id) : 319 _phase_times(phase_times), _phase(phase), _worker_id(worker_id) { 320 if (_phase_times != NULL) { 321 _start_time = os::elapsedTime(); 322 } 323 } 324 325 G1GCParPhaseTimesTracker::~G1GCParPhaseTimesTracker() { 326 if (_phase_times != NULL) { 327 _phase_times->record_time_secs(_phase, _worker_id, os::elapsedTime() - _start_time); 328 } 329 } 330