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(uint active_gc_threads) {
  98   assert(active_gc_threads > 0, "The number of threads must be > 0");
  99   assert(active_gc_threads <= _max_gc_threads, "The number of active threads must be <= the max number of threads");
 100   _gc_start_counter = os::elapsed_counter();
 101   _active_gc_threads = active_gc_threads;
 102   _cur_expand_heap_time_ms = 0.0;
 103   _external_accounted_time_ms = 0.0;
 104 
 105   for (int i = 0; i < GCParPhasesSentinel; i++) {
 106     if (_gc_par_phases[i] != NULL) {
 107       _gc_par_phases[i]->reset();
 108     }
 109   }
 110 }
 111 
 112 void G1GCPhaseTimes::note_gc_end() {
 113   _gc_pause_time_ms = TimeHelper::counter_to_millis(os::elapsed_counter() - _gc_start_counter);
 114   for (uint i = 0; i < _active_gc_threads; i++) {
 115     double worker_time = _gc_par_phases[GCWorkerEnd]->get(i) - _gc_par_phases[GCWorkerStart]->get(i);
 116     record_time_secs(GCWorkerTotal, i , worker_time);
 117 
 118     double worker_known_time =
 119         _gc_par_phases[ExtRootScan]->get(i) +
 120         _gc_par_phases[SATBFiltering]->get(i) +
 121         _gc_par_phases[UpdateRS]->get(i) +
 122         _gc_par_phases[ScanRS]->get(i) +
 123         _gc_par_phases[CodeRoots]->get(i) +
 124         _gc_par_phases[ObjCopy]->get(i) +
 125         _gc_par_phases[Termination]->get(i);
 126 
 127     record_time_secs(Other, i, worker_time - worker_known_time);
 128   }
 129 
 130   for (int i = 0; i < GCParPhasesSentinel; i++) {
 131     if (_gc_par_phases[i] != NULL) {
 132       _gc_par_phases[i]->verify(_active_gc_threads);
 133     }
 134   }
 135 }
 136 
 137 // record the time a phase took in seconds
 138 void G1GCPhaseTimes::record_time_secs(GCParPhases phase, uint worker_i, double secs) {
 139   _gc_par_phases[phase]->set(worker_i, secs);
 140 }
 141 
 142 // add a number of seconds to a phase
 143 void G1GCPhaseTimes::add_time_secs(GCParPhases phase, uint worker_i, double secs) {
 144   _gc_par_phases[phase]->add(worker_i, secs);
 145 }
 146 
 147 void G1GCPhaseTimes::record_thread_work_item(GCParPhases phase, uint worker_i, size_t count) {
 148   _gc_par_phases[phase]->set_thread_work_item(worker_i, count);
 149 }
 150 
 151 // return the average time for a phase in milliseconds
 152 double G1GCPhaseTimes::average_time_ms(GCParPhases phase) {
 153   return _gc_par_phases[phase]->average(_active_gc_threads) * 1000.0;
 154 }
 155 
 156 size_t G1GCPhaseTimes::sum_thread_work_items(GCParPhases phase) {
 157   assert(_gc_par_phases[phase]->thread_work_items() != NULL, "No sub count");
 158   return _gc_par_phases[phase]->thread_work_items()->sum(_active_gc_threads);
 159 }
 160 
 161 template <class T>
 162 void G1GCPhaseTimes::details(T* phase, const char* indent) {
 163   LogHandle(gc, phases, task) log;
 164   if (log.is_level(LogLevel::Trace)) {
 165     outputStream* trace_out = log.trace_stream();
 166     trace_out->print("%s", indent);
 167     phase->print_details_on(trace_out, _active_gc_threads);
 168   }
 169 }
 170 
 171 void G1GCPhaseTimes::log_phase(WorkerDataArray<double>* phase, uint indent, outputStream* out, bool print_sum) {
 172   out->print("%s", Indents[indent]);
 173   phase->print_summary_on(out, _active_gc_threads, print_sum);
 174   details(phase, Indents[indent]);
 175 
 176   WorkerDataArray<size_t>* work_items = phase->thread_work_items();
 177   if (work_items != NULL) {
 178     out->print("%s", Indents[indent + 1]);
 179     work_items->print_summary_on(out, _active_gc_threads, true);
 180     details(work_items, Indents[indent + 1]);
 181   }
 182 }
 183 
 184 void G1GCPhaseTimes::debug_phase(WorkerDataArray<double>* phase) {
 185   LogHandle(gc, phases) log;
 186   if (log.is_level(LogLevel::Debug)) {
 187     ResourceMark rm;
 188     log_phase(phase, 2, log.debug_stream(), true);
 189   }
 190 }
 191 
 192 void G1GCPhaseTimes::trace_phase(WorkerDataArray<double>* phase, bool print_sum) {
 193   LogHandle(gc, phases) log;
 194   if (log.is_level(LogLevel::Trace)) {
 195     ResourceMark rm;
 196     log_phase(phase, 3, log.trace_stream(), print_sum);
 197   }
 198 }
 199 
 200 #define PHASE_DOUBLE_FORMAT "%s%s: %.1lfms"
 201 #define PHASE_SIZE_FORMAT "%s%s: " SIZE_FORMAT
 202 
 203 #define info_line(str, value) \
 204   log_info(gc, phases)(PHASE_DOUBLE_FORMAT, Indents[1], str, value);
 205 
 206 #define debug_line(str, value) \
 207   log_debug(gc, phases)(PHASE_DOUBLE_FORMAT, Indents[2], str, value);
 208 
 209 #define trace_line(str, value) \
 210   log_trace(gc, phases)(PHASE_DOUBLE_FORMAT, Indents[3], str, value);
 211 
 212 #define trace_line_sz(str, value) \
 213   log_trace(gc, phases)(PHASE_SIZE_FORMAT, Indents[3], str, value);
 214 
 215 #define trace_line_ms(str, value) \
 216   log_trace(gc, phases)(PHASE_SIZE_FORMAT, Indents[3], str, value);
 217 
 218 #define info_line_and_account(str, value) \
 219   info_line(str, value);                  \
 220   accounted_time_ms += value;
 221 
 222 void G1GCPhaseTimes::print() {
 223   note_gc_end();
 224 
 225   double accounted_time_ms = _external_accounted_time_ms;
 226   if (_root_region_scan_wait_time_ms > 0.0) {
 227     info_line_and_account("Root Region Scan Waiting", _root_region_scan_wait_time_ms);
 228   }
 229 
 230   info_line_and_account("Evacuate Collection Set", _cur_collection_par_time_ms);
 231   trace_phase(_gc_par_phases[GCWorkerStart], false);
 232   debug_phase(_gc_par_phases[ExtRootScan]);
 233   for (int i = ThreadRoots; i <= SATBFiltering; i++) {
 234     trace_phase(_gc_par_phases[i]);
 235   }
 236   debug_phase(_gc_par_phases[UpdateRS]);
 237   if (ConcurrentG1Refine::hot_card_cache_enabled()) {
 238     trace_phase(_gc_par_phases[ScanHCC]);
 239   }
 240   debug_phase(_gc_par_phases[ScanRS]);
 241   debug_phase(_gc_par_phases[CodeRoots]);
 242   debug_phase(_gc_par_phases[ObjCopy]);
 243   debug_phase(_gc_par_phases[Termination]);
 244   debug_phase(_gc_par_phases[Other]);
 245   debug_phase(_gc_par_phases[GCWorkerTotal]);
 246   trace_phase(_gc_par_phases[GCWorkerEnd], false);
 247 
 248   info_line_and_account("Code Roots", _cur_collection_code_root_fixup_time_ms + _cur_strong_code_root_purge_time_ms);
 249   debug_line("Code Roots Fixup", _cur_collection_code_root_fixup_time_ms);
 250   debug_line("Code Roots Purge", _cur_strong_code_root_purge_time_ms);
 251 
 252   if (G1StringDedup::is_enabled()) {
 253     info_line_and_account("String Dedup Fixup", _cur_string_dedup_fixup_time_ms);
 254     debug_phase(_gc_par_phases[StringDedupQueueFixup]);
 255     debug_phase(_gc_par_phases[StringDedupTableFixup]);
 256   }
 257   info_line_and_account("Clear Card Table", _cur_clear_ct_time_ms);
 258   info_line_and_account("Expand Heap After Collection", _cur_expand_heap_time_ms);
 259 
 260   double free_cset_time = _recorded_young_free_cset_time_ms + _recorded_non_young_free_cset_time_ms;
 261   info_line_and_account("Free Collection Set", free_cset_time);
 262   debug_line("Young Free Collection Set", _recorded_young_free_cset_time_ms);
 263   debug_line("Non-Young Free Collection Set", _recorded_non_young_free_cset_time_ms);
 264   info_line_and_account("Merge Per-Thread State", _recorded_merge_pss_time_ms);
 265 
 266   info_line("Other", _gc_pause_time_ms - accounted_time_ms);
 267   if (_cur_verify_before_time_ms > 0.0) {
 268     debug_line("Verify Before", _cur_verify_before_time_ms);
 269   }
 270   if (G1CollectedHeap::heap()->evacuation_failed()) {
 271     double evac_fail_handling = _cur_evac_fail_recalc_used + _cur_evac_fail_remove_self_forwards +
 272       _cur_evac_fail_restore_remsets;
 273     debug_line("Evacuation Failure", evac_fail_handling);
 274     trace_line("Recalculate Used", _cur_evac_fail_recalc_used);
 275     trace_line("Remove Self Forwards",_cur_evac_fail_remove_self_forwards);
 276     trace_line("Restore RemSet", _cur_evac_fail_restore_remsets);
 277   }
 278   debug_line("Choose CSet", (_recorded_young_cset_choice_time_ms + _recorded_non_young_cset_choice_time_ms));
 279   debug_line("Preserve CM Refs", _recorded_preserve_cm_referents_time_ms);
 280   trace_phase(_gc_par_phases[PreserveCMReferents]);
 281   debug_line("Reference Processing", _cur_ref_proc_time_ms);
 282   debug_line("Reference Enqueuing", _cur_ref_enq_time_ms);
 283   debug_line("Redirty Cards", _recorded_redirty_logged_cards_time_ms);
 284   trace_phase(_gc_par_phases[RedirtyCards]);
 285   if (G1EagerReclaimHumongousObjects) {
 286     debug_line("Humongous Register", _cur_fast_reclaim_humongous_register_time_ms);
 287     trace_line_sz("Humongous Total", _cur_fast_reclaim_humongous_total);
 288     trace_line_sz("Humongous Candidate", _cur_fast_reclaim_humongous_candidates);
 289     debug_line("Humongous Reclaim", _cur_fast_reclaim_humongous_time_ms);
 290     trace_line_sz("Humongous Reclaimed", _cur_fast_reclaim_humongous_reclaimed);
 291   }
 292   if (_cur_verify_after_time_ms > 0.0) {
 293     debug_line("Verify After", _cur_verify_after_time_ms);
 294   }
 295 }
 296 
 297 G1GCParPhaseTimesTracker::G1GCParPhaseTimesTracker(G1GCPhaseTimes* phase_times, G1GCPhaseTimes::GCParPhases phase, uint worker_id) :
 298     _phase_times(phase_times), _phase(phase), _worker_id(worker_id) {
 299   if (_phase_times != NULL) {
 300     _start_time = os::elapsedTime();
 301   }
 302 }
 303 
 304 G1GCParPhaseTimesTracker::~G1GCParPhaseTimesTracker() {
 305   if (_phase_times != NULL) {
 306     _phase_times->record_time_secs(_phase, _worker_id, os::elapsedTime() - _start_time);
 307   }
 308 }
 309