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/g1CollectedHeap.inline.hpp"
  27 #include "gc/g1/g1GCPhaseTimes.hpp"
  28 #include "gc/g1/g1HotCardCache.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 (G1HotCardCache::default_use_cache()) {
  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 #if INCLUDE_AOT
  71   _gc_par_phases[AOTCodeRoots] = new WorkerDataArray<double>(max_gc_threads, "AOT Root Scanning (ms):");
  72 #endif
  73   _gc_par_phases[ObjCopy] = new WorkerDataArray<double>(max_gc_threads, "Object Copy (ms):");
  74   _gc_par_phases[Termination] = new WorkerDataArray<double>(max_gc_threads, "Termination (ms):");
  75   _gc_par_phases[GCWorkerTotal] = new WorkerDataArray<double>(max_gc_threads, "GC Worker Total (ms):");
  76   _gc_par_phases[GCWorkerEnd] = new WorkerDataArray<double>(max_gc_threads, "GC Worker End (ms):");
  77   _gc_par_phases[Other] = new WorkerDataArray<double>(max_gc_threads, "GC Worker Other (ms):");
  78 
  79   _update_rs_processed_buffers = new WorkerDataArray<size_t>(max_gc_threads, "Processed Buffers:");
  80   _gc_par_phases[UpdateRS]->link_thread_work_items(_update_rs_processed_buffers);
  81 
  82   _termination_attempts = new WorkerDataArray<size_t>(max_gc_threads, "Termination Attempts:");
  83   _gc_par_phases[Termination]->link_thread_work_items(_termination_attempts);
  84 
  85   if (UseStringDeduplication) {
  86     _gc_par_phases[StringDedupQueueFixup] = new WorkerDataArray<double>(max_gc_threads, "Queue Fixup (ms):");
  87     _gc_par_phases[StringDedupTableFixup] = new WorkerDataArray<double>(max_gc_threads, "Table Fixup (ms):");
  88   } else {
  89     _gc_par_phases[StringDedupQueueFixup] = NULL;
  90     _gc_par_phases[StringDedupTableFixup] = NULL;
  91   }
  92 
  93   _gc_par_phases[RedirtyCards] = new WorkerDataArray<double>(max_gc_threads, "Parallel Redirty (ms):");
  94   _redirtied_cards = new WorkerDataArray<size_t>(max_gc_threads, "Redirtied Cards:");
  95   _gc_par_phases[RedirtyCards]->link_thread_work_items(_redirtied_cards);
  96 
  97   _gc_par_phases[YoungFreeCSet] = new WorkerDataArray<double>(max_gc_threads, "Young Free Collection Set (ms):");
  98   _gc_par_phases[NonYoungFreeCSet] = new WorkerDataArray<double>(max_gc_threads, "Non-Young Free Collection Set (ms):");
  99 
 100   _gc_par_phases[PreserveCMReferents] = new WorkerDataArray<double>(max_gc_threads, "Parallel Preserve CM Refs (ms):");
 101 }
 102 
 103 void G1GCPhaseTimes::note_gc_start() {
 104   _gc_start_counter = os::elapsed_counter();
 105   _cur_expand_heap_time_ms = 0.0;
 106   _external_accounted_time_ms = 0.0;
 107   _recorded_clear_claimed_marks_time_ms = 0.0;
 108 
 109   for (int i = 0; i < GCParPhasesSentinel; i++) {
 110     if (_gc_par_phases[i] != NULL) {
 111       _gc_par_phases[i]->reset();
 112     }
 113   }
 114 }
 115 
 116 #define ASSERT_PHASE_UNINITIALIZED(phase) \
 117     assert(_gc_par_phases[phase]->get(i) == uninitialized, "Phase " #phase " reported for thread that was not started");
 118 
 119 double G1GCPhaseTimes::worker_time(GCParPhases phase, uint worker) {
 120   double value = _gc_par_phases[phase]->get(worker);
 121   if (value != WorkerDataArray<double>::uninitialized()) {
 122     return value;
 123   }
 124   return 0.0;
 125 }
 126 
 127 void G1GCPhaseTimes::note_gc_end() {
 128   _gc_pause_time_ms = TimeHelper::counter_to_millis(os::elapsed_counter() - _gc_start_counter);
 129 
 130   double uninitialized = WorkerDataArray<double>::uninitialized();
 131 
 132   for (uint i = 0; i < _max_gc_threads; i++) {
 133     double worker_start = _gc_par_phases[GCWorkerStart]->get(i);
 134     if (worker_start != uninitialized) {
 135       assert(_gc_par_phases[GCWorkerEnd]->get(i) != uninitialized, "Worker started but not ended.");
 136       double total_worker_time = _gc_par_phases[GCWorkerEnd]->get(i) - _gc_par_phases[GCWorkerStart]->get(i);
 137       record_time_secs(GCWorkerTotal, i , total_worker_time);
 138 
 139       double worker_known_time =
 140           worker_time(ExtRootScan, i)
 141           + worker_time(SATBFiltering, i)
 142           + worker_time(UpdateRS, i)
 143           + worker_time(ScanRS, i)
 144           + worker_time(CodeRoots, i)
 145           + worker_time(ObjCopy, i)
 146           + worker_time(Termination, i);
 147 
 148       record_time_secs(Other, i, total_worker_time - worker_known_time);
 149     } else {
 150       // Make sure all slots are uninitialized since this thread did not seem to have been started
 151       ASSERT_PHASE_UNINITIALIZED(GCWorkerEnd);
 152       ASSERT_PHASE_UNINITIALIZED(ExtRootScan);
 153       ASSERT_PHASE_UNINITIALIZED(SATBFiltering);
 154       ASSERT_PHASE_UNINITIALIZED(UpdateRS);
 155       ASSERT_PHASE_UNINITIALIZED(ScanRS);
 156       ASSERT_PHASE_UNINITIALIZED(CodeRoots);
 157       ASSERT_PHASE_UNINITIALIZED(ObjCopy);
 158       ASSERT_PHASE_UNINITIALIZED(Termination);
 159     }
 160   }
 161 }
 162 
 163 #undef ASSERT_PHASE_UNINITIALIZED
 164 
 165 // record the time a phase took in seconds
 166 void G1GCPhaseTimes::record_time_secs(GCParPhases phase, uint worker_i, double secs) {
 167   _gc_par_phases[phase]->set(worker_i, secs);
 168 }
 169 
 170 // add a number of seconds to a phase
 171 void G1GCPhaseTimes::add_time_secs(GCParPhases phase, uint worker_i, double secs) {
 172   _gc_par_phases[phase]->add(worker_i, secs);
 173 }
 174 
 175 void G1GCPhaseTimes::record_thread_work_item(GCParPhases phase, uint worker_i, size_t count) {
 176   _gc_par_phases[phase]->set_thread_work_item(worker_i, count);
 177 }
 178 
 179 // return the average time for a phase in milliseconds
 180 double G1GCPhaseTimes::average_time_ms(GCParPhases phase) {
 181   return _gc_par_phases[phase]->average() * 1000.0;
 182 }
 183 
 184 size_t G1GCPhaseTimes::sum_thread_work_items(GCParPhases phase) {
 185   assert(_gc_par_phases[phase]->thread_work_items() != NULL, "No sub count");
 186   return _gc_par_phases[phase]->thread_work_items()->sum();
 187 }
 188 
 189 template <class T>
 190 void G1GCPhaseTimes::details(T* phase, const char* indent) {
 191   Log(gc, phases, task) log;
 192   if (log.is_level(LogLevel::Trace)) {
 193     outputStream* trace_out = log.trace_stream();
 194     trace_out->print("%s", indent);
 195     phase->print_details_on(trace_out);
 196   }
 197 }
 198 
 199 void G1GCPhaseTimes::log_phase(WorkerDataArray<double>* phase, uint indent, outputStream* out, bool print_sum) {
 200   out->print("%s", Indents[indent]);
 201   phase->print_summary_on(out, print_sum);
 202   details(phase, Indents[indent]);
 203 
 204   WorkerDataArray<size_t>* work_items = phase->thread_work_items();
 205   if (work_items != NULL) {
 206     out->print("%s", Indents[indent + 1]);
 207     work_items->print_summary_on(out, true);
 208     details(work_items, Indents[indent + 1]);
 209   }
 210 }
 211 
 212 void G1GCPhaseTimes::debug_phase(WorkerDataArray<double>* phase) {
 213   Log(gc, phases) log;
 214   if (log.is_level(LogLevel::Debug)) {
 215     ResourceMark rm;
 216     log_phase(phase, 2, log.debug_stream(), true);
 217   }
 218 }
 219 
 220 void G1GCPhaseTimes::trace_phase(WorkerDataArray<double>* phase, bool print_sum) {
 221   Log(gc, phases) log;
 222   if (log.is_level(LogLevel::Trace)) {
 223     ResourceMark rm;
 224     log_phase(phase, 3, log.trace_stream(), print_sum);
 225   }
 226 }
 227 
 228 #define PHASE_DOUBLE_FORMAT "%s%s: %.1lfms"
 229 #define PHASE_SIZE_FORMAT "%s%s: " SIZE_FORMAT
 230 
 231 #define info_line(str, value) \
 232   log_info(gc, phases)(PHASE_DOUBLE_FORMAT, Indents[1], str, value);
 233 
 234 #define debug_line(str, value) \
 235   log_debug(gc, phases)(PHASE_DOUBLE_FORMAT, Indents[2], str, value);
 236 
 237 #define trace_line(str, value) \
 238   log_trace(gc, phases)(PHASE_DOUBLE_FORMAT, Indents[3], str, value);
 239 
 240 #define trace_line_sz(str, value) \
 241   log_trace(gc, phases)(PHASE_SIZE_FORMAT, Indents[3], str, value);
 242 
 243 #define trace_line_ms(str, value) \
 244   log_trace(gc, phases)(PHASE_SIZE_FORMAT, Indents[3], str, value);
 245 
 246 #define info_line_and_account(str, value) \
 247   info_line(str, value);                  \
 248   accounted_time_ms += value;
 249 
 250 void G1GCPhaseTimes::print() {
 251   note_gc_end();
 252 
 253   double accounted_time_ms = _external_accounted_time_ms;
 254   if (_root_region_scan_wait_time_ms > 0.0) {
 255     info_line_and_account("Root Region Scan Waiting", _root_region_scan_wait_time_ms);
 256   }
 257 
 258   info_line_and_account("Evacuate Collection Set", _cur_collection_par_time_ms);
 259   trace_phase(_gc_par_phases[GCWorkerStart], false);
 260   debug_phase(_gc_par_phases[ExtRootScan]);
 261   for (int i = ThreadRoots; i <= SATBFiltering; i++) {
 262     trace_phase(_gc_par_phases[i]);
 263   }
 264   debug_phase(_gc_par_phases[UpdateRS]);
 265   if (G1HotCardCache::default_use_cache()) {
 266     trace_phase(_gc_par_phases[ScanHCC]);
 267   }
 268   debug_phase(_gc_par_phases[ScanRS]);
 269   debug_phase(_gc_par_phases[CodeRoots]);
 270 #if INCLUDE_AOT
 271   debug_phase(_gc_par_phases[AOTCodeRoots]);
 272 #endif
 273   debug_phase(_gc_par_phases[ObjCopy]);
 274   debug_phase(_gc_par_phases[Termination]);
 275   debug_phase(_gc_par_phases[Other]);
 276   debug_phase(_gc_par_phases[GCWorkerTotal]);
 277   trace_phase(_gc_par_phases[GCWorkerEnd], false);
 278 
 279   info_line_and_account("Code Roots", _cur_collection_code_root_fixup_time_ms + _cur_strong_code_root_purge_time_ms);
 280   debug_line("Code Roots Fixup", _cur_collection_code_root_fixup_time_ms);
 281   debug_line("Code Roots Purge", _cur_strong_code_root_purge_time_ms);
 282 
 283   if (G1StringDedup::is_enabled()) {
 284     info_line_and_account("String Dedup Fixup", _cur_string_dedup_fixup_time_ms);
 285     debug_phase(_gc_par_phases[StringDedupQueueFixup]);
 286     debug_phase(_gc_par_phases[StringDedupTableFixup]);
 287   }
 288   info_line_and_account("Clear Card Table", _cur_clear_ct_time_ms);
 289   info_line_and_account("Expand Heap After Collection", _cur_expand_heap_time_ms);
 290 
 291   info_line_and_account("Free Collection Set", _recorded_total_free_cset_time_ms);
 292   debug_line("Free Collection Set Serial", _recorded_serial_free_cset_time_ms);
 293   debug_phase(_gc_par_phases[YoungFreeCSet]);
 294   debug_phase(_gc_par_phases[NonYoungFreeCSet]);
 295 
 296   info_line_and_account("Merge Per-Thread State", _recorded_merge_pss_time_ms);
 297 
 298   info_line("Other", _gc_pause_time_ms - accounted_time_ms);
 299   if (_cur_verify_before_time_ms > 0.0) {
 300     debug_line("Verify Before", _cur_verify_before_time_ms);
 301   }
 302   if (G1CollectedHeap::heap()->evacuation_failed()) {
 303     double evac_fail_handling = _cur_evac_fail_recalc_used + _cur_evac_fail_remove_self_forwards +
 304       _cur_evac_fail_restore_remsets;
 305     debug_line("Evacuation Failure", evac_fail_handling);
 306     trace_line("Recalculate Used", _cur_evac_fail_recalc_used);
 307     trace_line("Remove Self Forwards",_cur_evac_fail_remove_self_forwards);
 308     trace_line("Restore RemSet", _cur_evac_fail_restore_remsets);
 309   }
 310   debug_line("Choose CSet", (_recorded_young_cset_choice_time_ms + _recorded_non_young_cset_choice_time_ms));
 311   debug_line("Preserve CM Refs", _recorded_preserve_cm_referents_time_ms);
 312   trace_phase(_gc_par_phases[PreserveCMReferents]);
 313   debug_line("Reference Processing", _cur_ref_proc_time_ms);
 314   debug_line("Reference Enqueuing", _cur_ref_enq_time_ms);
 315   debug_line("Redirty Cards", _recorded_redirty_logged_cards_time_ms);
 316   if (_recorded_clear_claimed_marks_time_ms > 0.0) {
 317     debug_line("Clear Claimed Marks", _recorded_clear_claimed_marks_time_ms);
 318   }
 319 
 320   trace_phase(_gc_par_phases[RedirtyCards]);
 321   if (G1EagerReclaimHumongousObjects) {
 322     debug_line("Humongous Register", _cur_fast_reclaim_humongous_register_time_ms);
 323     trace_line_sz("Humongous Total", _cur_fast_reclaim_humongous_total);
 324     trace_line_sz("Humongous Candidate", _cur_fast_reclaim_humongous_candidates);
 325     debug_line("Humongous Reclaim", _cur_fast_reclaim_humongous_time_ms);
 326     trace_line_sz("Humongous Reclaimed", _cur_fast_reclaim_humongous_reclaimed);
 327   }
 328   if (_cur_verify_after_time_ms > 0.0) {
 329     debug_line("Verify After", _cur_verify_after_time_ms);
 330   }
 331 }
 332 
 333 G1GCParPhaseTimesTracker::G1GCParPhaseTimesTracker(G1GCPhaseTimes* phase_times, G1GCPhaseTimes::GCParPhases phase, uint worker_id) :
 334     _phase_times(phase_times), _phase(phase), _worker_id(worker_id) {
 335   if (_phase_times != NULL) {
 336     _start_time = os::elapsedTime();
 337   }
 338 }
 339 
 340 G1GCParPhaseTimesTracker::~G1GCParPhaseTimesTracker() {
 341   if (_phase_times != NULL) {
 342     _phase_times->record_time_secs(_phase, _worker_id, os::elapsedTime() - _start_time);
 343   }
 344 }
 345