1 /*
   2  * Copyright (c) 2013, 2014 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_implementation/g1/g1CollectedHeap.inline.hpp"
  27 #include "gc_implementation/g1/g1GCPhaseTimes.hpp"
  28 #include "gc_implementation/g1/g1Log.hpp"
  29 #include "gc_implementation/g1/g1StringDedup.hpp"
  30 #include "memory/allocation.hpp"
  31 #include "runtime/os.hpp"
  32 
  33 // Helper class for avoiding interleaved logging
  34 class LineBuffer: public StackObj {
  35 
  36 private:
  37   static const int BUFFER_LEN = 1024;
  38   static const int INDENT_CHARS = 3;
  39   char _buffer[BUFFER_LEN];
  40   int _indent_level;
  41   int _cur;
  42 
  43   void vappend(const char* format, va_list ap)  ATTRIBUTE_PRINTF(2, 0) {
  44     int res = vsnprintf(&_buffer[_cur], BUFFER_LEN - _cur, format, ap);
  45     if (res != -1) {
  46       _cur += res;
  47     } else {
  48       DEBUG_ONLY(warning("buffer too small in LineBuffer");)
  49       _buffer[BUFFER_LEN -1] = 0;
  50       _cur = BUFFER_LEN; // vsnprintf above should not add to _buffer if we are called again
  51     }
  52   }
  53 
  54 public:
  55   explicit LineBuffer(int indent_level): _indent_level(indent_level), _cur(0) {
  56     for (; (_cur < BUFFER_LEN && _cur < (_indent_level * INDENT_CHARS)); _cur++) {
  57       _buffer[_cur] = ' ';
  58     }
  59   }
  60 
  61 #ifndef PRODUCT
  62   ~LineBuffer() {
  63     assert(_cur == _indent_level * INDENT_CHARS, "pending data in buffer - append_and_print_cr() not called?");
  64   }
  65 #endif
  66 
  67   void append(const char* format, ...)  ATTRIBUTE_PRINTF(2, 3) {
  68     va_list ap;
  69     va_start(ap, format);
  70     vappend(format, ap);
  71     va_end(ap);
  72   }
  73 
  74   void print_cr() {
  75     gclog_or_tty->print_cr("%s", _buffer);
  76     _cur = _indent_level * INDENT_CHARS;
  77   }
  78 
  79   void append_and_print_cr(const char* format, ...)  ATTRIBUTE_PRINTF(2, 3) {
  80     va_list ap;
  81     va_start(ap, format);
  82     vappend(format, ap);
  83     va_end(ap);
  84     print_cr();
  85   }
  86 };
  87 
  88 template <class T>
  89 class WorkerDataArray  : public CHeapObj<mtGC> {
  90   friend class G1GCParPhasePrinter;
  91   T*          _data;
  92   uint        _length;
  93   const char* _title;
  94   bool        _print_sum;
  95   int         _log_level;
  96   uint        _indent_level;
  97   bool        _enabled;
  98 
  99   WorkerDataArray<size_t>* _thread_work_items;
 100 
 101   NOT_PRODUCT(T uninitialized();)
 102 
 103   // We are caching the sum and average to only have to calculate them once.
 104   // This is not done in an MT-safe way. It is intended to allow single
 105   // threaded code to call sum() and average() multiple times in any order
 106   // without having to worry about the cost.
 107   bool   _has_new_data;
 108   T      _sum;
 109   T      _min;
 110   T      _max;
 111   double _average;
 112 
 113  public:
 114   WorkerDataArray(uint length, const char* title, bool print_sum, int log_level, uint indent_level) :
 115     _title(title), _length(0), _print_sum(print_sum), _log_level(log_level), _indent_level(indent_level),
 116     _has_new_data(true), _thread_work_items(NULL), _enabled(true) {
 117     assert(length > 0, "Must have some workers to store data for");
 118     _length = length;
 119     _data = NEW_C_HEAP_ARRAY(T, _length, mtGC);
 120   }
 121 
 122   ~WorkerDataArray() {
 123     FREE_C_HEAP_ARRAY(T, _data);
 124   }
 125 
 126   void link_thread_work_items(WorkerDataArray<size_t>* thread_work_items) {
 127     _thread_work_items = thread_work_items;
 128   }
 129 
 130   WorkerDataArray<size_t>* thread_work_items() { return _thread_work_items; }
 131 
 132   void set(uint worker_i, T value) {
 133     assert(worker_i < _length, err_msg("Worker %d is greater than max: %d", worker_i, _length));
 134     assert(_data[worker_i] == WorkerDataArray<T>::uninitialized(), err_msg("Overwriting data for worker %d in %s", worker_i, _title));
 135     _data[worker_i] = value;
 136     _has_new_data = true;
 137   }
 138 
 139   void set_thread_work_item(uint worker_i, size_t value) {
 140     assert(_thread_work_items != NULL, "No sub count");
 141     _thread_work_items->set(worker_i, value);
 142   }
 143 
 144   T get(uint worker_i) {
 145     assert(worker_i < _length, err_msg("Worker %d is greater than max: %d", worker_i, _length));
 146     assert(_data[worker_i] != WorkerDataArray<T>::uninitialized(), err_msg("No data added for worker %d", worker_i));
 147     return _data[worker_i];
 148   }
 149 
 150   void add(uint worker_i, T value) {
 151     assert(worker_i < _length, err_msg("Worker %d is greater than max: %d", worker_i, _length));
 152     assert(_data[worker_i] != WorkerDataArray<T>::uninitialized(), err_msg("No data to add to for worker %d", worker_i));
 153     _data[worker_i] += value;
 154     _has_new_data = true;
 155   }
 156 
 157   double average(uint active_threads){
 158     calculate_totals(active_threads);
 159     return _average;
 160   }
 161 
 162   T sum(uint active_threads) {
 163     calculate_totals(active_threads);
 164     return _sum;
 165   }
 166 
 167   T minimum(uint active_threads) {
 168     calculate_totals(active_threads);
 169     return _min;
 170   }
 171 
 172   T maximum(uint active_threads) {
 173     calculate_totals(active_threads);
 174     return _max;
 175   }
 176 
 177   void reset() PRODUCT_RETURN;
 178   void verify(uint active_threads) PRODUCT_RETURN;
 179 
 180   void set_enabled(bool enabled) { _enabled = enabled; }
 181 
 182   int log_level() { return _log_level;  }
 183 
 184  private:
 185 
 186   void calculate_totals(uint active_threads){
 187     if (!_has_new_data) {
 188       return;
 189     }
 190 
 191     _sum = (T)0;
 192     _min = _data[0];
 193     _max = _min;
 194     assert(active_threads <= _length, "Wrong number of active threads");
 195     for (uint i = 0; i < active_threads; ++i) {
 196       T val = _data[i];
 197       _sum += val;
 198       _min = MIN2(_min, val);
 199       _max = MAX2(_max, val);
 200     }
 201     _average = (double)_sum / (double)active_threads;
 202     _has_new_data = false;
 203   }
 204 };
 205 
 206 
 207 #ifndef PRODUCT
 208 
 209 template <>
 210 size_t WorkerDataArray<size_t>::uninitialized() {
 211   return (size_t)-1;
 212 }
 213 
 214 template <>
 215 double WorkerDataArray<double>::uninitialized() {
 216   return -1.0;
 217 }
 218 
 219 template <class T>
 220 void WorkerDataArray<T>::reset() {
 221   for (uint i = 0; i < _length; i++) {
 222     _data[i] = WorkerDataArray<T>::uninitialized();
 223   }
 224   if (_thread_work_items != NULL) {
 225     _thread_work_items->reset();
 226   }
 227 }
 228 
 229 template <class T>
 230 void WorkerDataArray<T>::verify(uint active_threads) {
 231   if (!_enabled) {
 232     return;
 233   }
 234 
 235   assert(active_threads <= _length, "Wrong number of active threads");
 236   for (uint i = 0; i < active_threads; i++) {
 237     assert(_data[i] != WorkerDataArray<T>::uninitialized(),
 238         err_msg("Invalid data for worker %u in '%s'", i, _title));
 239   }
 240   if (_thread_work_items != NULL) {
 241     _thread_work_items->verify(active_threads);
 242   }
 243 }
 244 
 245 #endif
 246 
 247 G1GCPhaseTimes::G1GCPhaseTimes(uint max_gc_threads) :
 248   _max_gc_threads(max_gc_threads)
 249 {
 250   assert(max_gc_threads > 0, "Must have some GC threads");
 251 
 252   _gc_par_phases[GCWorkerStart] = new WorkerDataArray<double>(max_gc_threads, "GC Worker Start (ms)", false, G1Log::LevelFiner, 2);
 253   _gc_par_phases[ExtRootScan] = new WorkerDataArray<double>(max_gc_threads, "Ext Root Scanning (ms)", true, G1Log::LevelFiner, 2);
 254 
 255   // Root scanning phases
 256   _gc_par_phases[ThreadRoots] = new WorkerDataArray<double>(max_gc_threads, "Thread Roots (ms)", true, G1Log::LevelFinest, 3);
 257   _gc_par_phases[StringTableRoots] = new WorkerDataArray<double>(max_gc_threads, "StringTable Roots (ms)", true, G1Log::LevelFinest, 3);
 258   _gc_par_phases[UniverseRoots] = new WorkerDataArray<double>(max_gc_threads, "Universe Roots (ms)", true, G1Log::LevelFinest, 3);
 259   _gc_par_phases[JNIRoots] = new WorkerDataArray<double>(max_gc_threads, "JNI Handles Roots (ms)", true, G1Log::LevelFinest, 3);
 260   _gc_par_phases[ObjectSynchronizerRoots] = new WorkerDataArray<double>(max_gc_threads, "ObjectSynchronizer Roots (ms)", true, G1Log::LevelFinest, 3);
 261   _gc_par_phases[FlatProfilerRoots] = new WorkerDataArray<double>(max_gc_threads, "FlatProfiler Roots (ms)", true, G1Log::LevelFinest, 3);
 262   _gc_par_phases[ManagementRoots] = new WorkerDataArray<double>(max_gc_threads, "Management Roots (ms)", true, G1Log::LevelFinest, 3);
 263   _gc_par_phases[SystemDictionaryRoots] = new WorkerDataArray<double>(max_gc_threads, "SystemDictionary Roots (ms)", true, G1Log::LevelFinest, 3);
 264   _gc_par_phases[CLDGRoots] = new WorkerDataArray<double>(max_gc_threads, "CLDG Roots (ms)", true, G1Log::LevelFinest, 3);
 265   _gc_par_phases[JVMTIRoots] = new WorkerDataArray<double>(max_gc_threads, "JVMTI Roots (ms)", true, G1Log::LevelFinest, 3);
 266   _gc_par_phases[CodeCacheRoots] = new WorkerDataArray<double>(max_gc_threads, "CodeCache Roots (ms)", true, G1Log::LevelFinest, 3);
 267   _gc_par_phases[CMRefRoots] = new WorkerDataArray<double>(max_gc_threads, "CM RefProcessor Roots (ms)", true, G1Log::LevelFinest, 3);
 268   _gc_par_phases[WaitForStrongCLD] = new WorkerDataArray<double>(max_gc_threads, "Wait For Strong CLD (ms)", true, G1Log::LevelFinest, 3);
 269   _gc_par_phases[WeakCLDRoots] = new WorkerDataArray<double>(max_gc_threads, "Weak CLD Roots (ms)", true, G1Log::LevelFinest, 3);
 270   _gc_par_phases[SATBFiltering] = new WorkerDataArray<double>(max_gc_threads, "SATB Filtering (ms)", true, G1Log::LevelFinest, 3);
 271 
 272   _gc_par_phases[UpdateRS] = new WorkerDataArray<double>(max_gc_threads, "Update RS (ms)", true, G1Log::LevelFiner, 2);
 273   _gc_par_phases[ScanRS] = new WorkerDataArray<double>(max_gc_threads, "Scan RS (ms)", true, G1Log::LevelFiner, 2);
 274   _gc_par_phases[CodeRoots] = new WorkerDataArray<double>(max_gc_threads, "Code Root Scanning (ms)", true, G1Log::LevelFiner, 2);
 275   _gc_par_phases[ObjCopy] = new WorkerDataArray<double>(max_gc_threads, "Object Copy (ms)", true, G1Log::LevelFiner, 2);
 276   _gc_par_phases[Termination] = new WorkerDataArray<double>(max_gc_threads, "Termination (ms)", true, G1Log::LevelFiner, 2);
 277   _gc_par_phases[GCWorkerTotal] = new WorkerDataArray<double>(max_gc_threads, "GC Worker Total (ms)", true, G1Log::LevelFiner, 2);
 278   _gc_par_phases[GCWorkerEnd] = new WorkerDataArray<double>(max_gc_threads, "GC Worker End (ms)", false, G1Log::LevelFiner, 2);
 279   _gc_par_phases[Other] = new WorkerDataArray<double>(max_gc_threads, "GC Worker Other (ms)", true, G1Log::LevelFiner, 2);
 280 
 281   _update_rs_processed_buffers = new WorkerDataArray<size_t>(max_gc_threads, "Processed Buffers", true, G1Log::LevelFiner, 3);
 282   _gc_par_phases[UpdateRS]->link_thread_work_items(_update_rs_processed_buffers);
 283 
 284   _termination_attempts = new WorkerDataArray<size_t>(max_gc_threads, "Termination Attempts", true, G1Log::LevelFinest, 3);
 285   _gc_par_phases[Termination]->link_thread_work_items(_termination_attempts);
 286 
 287   _gc_par_phases[StringDedupQueueFixup] = new WorkerDataArray<double>(max_gc_threads, "Queue Fixup (ms)", true, G1Log::LevelFiner, 2);
 288   _gc_par_phases[StringDedupTableFixup] = new WorkerDataArray<double>(max_gc_threads, "Table Fixup (ms)", true, G1Log::LevelFiner, 2);
 289 
 290   _gc_par_phases[RedirtyCards] = new WorkerDataArray<double>(max_gc_threads, "Parallel Redirty", true, G1Log::LevelFinest, 3);
 291   _redirtied_cards = new WorkerDataArray<size_t>(max_gc_threads, "Redirtied Cards", true, G1Log::LevelFinest, 3);
 292   _gc_par_phases[RedirtyCards]->link_thread_work_items(_redirtied_cards);
 293 }
 294 
 295 void G1GCPhaseTimes::note_gc_start(uint active_gc_threads, bool mark_in_progress) {
 296   assert(active_gc_threads > 0, "The number of threads must be > 0");
 297   assert(active_gc_threads <= _max_gc_threads, "The number of active threads must be <= the max number of threads");
 298   _active_gc_threads = active_gc_threads;
 299 
 300   for (int i = 0; i < GCParPhasesSentinel; i++) {
 301     _gc_par_phases[i]->reset();
 302   }
 303 
 304   _gc_par_phases[StringDedupQueueFixup]->set_enabled(G1StringDedup::is_enabled());
 305   _gc_par_phases[StringDedupTableFixup]->set_enabled(G1StringDedup::is_enabled());
 306 }
 307 
 308 void G1GCPhaseTimes::note_gc_end() {
 309   for (uint i = 0; i < _active_gc_threads; i++) {
 310     double worker_time = _gc_par_phases[GCWorkerEnd]->get(i) - _gc_par_phases[GCWorkerStart]->get(i);
 311     record_time_secs(GCWorkerTotal, i , worker_time);
 312 
 313     double worker_known_time =
 314         _gc_par_phases[ExtRootScan]->get(i) +
 315         _gc_par_phases[SATBFiltering]->get(i) +
 316         _gc_par_phases[UpdateRS]->get(i) +
 317         _gc_par_phases[ScanRS]->get(i) +
 318         _gc_par_phases[CodeRoots]->get(i) +
 319         _gc_par_phases[ObjCopy]->get(i) +
 320         _gc_par_phases[Termination]->get(i);
 321 
 322     record_time_secs(Other, i, worker_time - worker_known_time);
 323   }
 324 
 325   for (int i = 0; i < GCParPhasesSentinel; i++) {
 326     _gc_par_phases[i]->verify(_active_gc_threads);
 327   }
 328 }
 329 
 330 void G1GCPhaseTimes::print_stats(int level, const char* str, double value) {
 331   LineBuffer(level).append_and_print_cr("[%s: %.1lf ms]", str, value);
 332 }
 333 
 334 void G1GCPhaseTimes::print_stats(int level, const char* str, size_t value) {
 335   LineBuffer(level).append_and_print_cr("[%s: "SIZE_FORMAT"]", str, value);
 336 }
 337 
 338 void G1GCPhaseTimes::print_stats(int level, const char* str, double value, uint workers) {
 339   LineBuffer(level).append_and_print_cr("[%s: %.1lf ms, GC Workers: %u]", str, value, workers);
 340 }
 341 
 342 double G1GCPhaseTimes::accounted_time_ms() {
 343     // Subtract the root region scanning wait time. It's initialized to
 344     // zero at the start of the pause.
 345     double misc_time_ms = _root_region_scan_wait_time_ms;
 346 
 347     misc_time_ms += _cur_collection_par_time_ms;
 348 
 349     // Now subtract the time taken to fix up roots in generated code
 350     misc_time_ms += _cur_collection_code_root_fixup_time_ms;
 351 
 352     // Strong code root purge time
 353     misc_time_ms += _cur_strong_code_root_purge_time_ms;
 354 
 355     if (G1StringDedup::is_enabled()) {
 356       // String dedup fixup time
 357       misc_time_ms += _cur_string_dedup_fixup_time_ms;
 358     }
 359 
 360     // Subtract the time taken to clean the card table from the
 361     // current value of "other time"
 362     misc_time_ms += _cur_clear_ct_time_ms;
 363 
 364     return misc_time_ms;
 365 }
 366 
 367 // record the time a phase took in seconds
 368 void G1GCPhaseTimes::record_time_secs(GCParPhases phase, uint worker_i, double secs) {
 369   _gc_par_phases[phase]->set(worker_i, secs);
 370 }
 371 
 372 // add a number of seconds to a phase
 373 void G1GCPhaseTimes::add_time_secs(GCParPhases phase, uint worker_i, double secs) {
 374   _gc_par_phases[phase]->add(worker_i, secs);
 375 }
 376 
 377 void G1GCPhaseTimes::record_thread_work_item(GCParPhases phase, uint worker_i, size_t count) {
 378   _gc_par_phases[phase]->set_thread_work_item(worker_i, count);
 379 }
 380 
 381 // return the average time for a phase in milliseconds
 382 double G1GCPhaseTimes::average_time_ms(GCParPhases phase) {
 383   return _gc_par_phases[phase]->average(_active_gc_threads) * 1000.0;
 384 }
 385 
 386 double G1GCPhaseTimes::get_time_ms(GCParPhases phase, uint worker_i) {
 387   return _gc_par_phases[phase]->get(worker_i) * 1000.0;
 388 }
 389 
 390 double G1GCPhaseTimes::sum_time_ms(GCParPhases phase) {
 391   return _gc_par_phases[phase]->sum(_active_gc_threads) * 1000.0;
 392 }
 393 
 394 double G1GCPhaseTimes::min_time_ms(GCParPhases phase) {
 395   return _gc_par_phases[phase]->minimum(_active_gc_threads) * 1000.0;
 396 }
 397 
 398 double G1GCPhaseTimes::max_time_ms(GCParPhases phase) {
 399   return _gc_par_phases[phase]->maximum(_active_gc_threads) * 1000.0;
 400 }
 401 
 402 size_t G1GCPhaseTimes::get_thread_work_item(GCParPhases phase, uint worker_i) {
 403   assert(_gc_par_phases[phase]->thread_work_items() != NULL, "No sub count");
 404   return _gc_par_phases[phase]->thread_work_items()->get(worker_i);
 405 }
 406 
 407 size_t G1GCPhaseTimes::sum_thread_work_items(GCParPhases phase) {
 408   assert(_gc_par_phases[phase]->thread_work_items() != NULL, "No sub count");
 409   return _gc_par_phases[phase]->thread_work_items()->sum(_active_gc_threads);
 410 }
 411 
 412 double G1GCPhaseTimes::average_thread_work_items(GCParPhases phase) {
 413   assert(_gc_par_phases[phase]->thread_work_items() != NULL, "No sub count");
 414   return _gc_par_phases[phase]->thread_work_items()->average(_active_gc_threads);
 415 }
 416 
 417 size_t G1GCPhaseTimes::min_thread_work_items(GCParPhases phase) {
 418   assert(_gc_par_phases[phase]->thread_work_items() != NULL, "No sub count");
 419   return _gc_par_phases[phase]->thread_work_items()->minimum(_active_gc_threads);
 420 }
 421 
 422 size_t G1GCPhaseTimes::max_thread_work_items(GCParPhases phase) {
 423   assert(_gc_par_phases[phase]->thread_work_items() != NULL, "No sub count");
 424   return _gc_par_phases[phase]->thread_work_items()->maximum(_active_gc_threads);
 425 }
 426 
 427 class G1GCParPhasePrinter : public StackObj {
 428   G1GCPhaseTimes* _phase_times;
 429  public:
 430   G1GCParPhasePrinter(G1GCPhaseTimes* phase_times) : _phase_times(phase_times) {}
 431 
 432   void print(G1GCPhaseTimes::GCParPhases phase_id) {
 433     WorkerDataArray<double>* phase = _phase_times->_gc_par_phases[phase_id];
 434 
 435     if (phase->_log_level > G1Log::level() || !phase->_enabled) {
 436       return;
 437     }
 438 
 439     if (phase->_length == 1) {
 440       print_single_length(phase_id, phase);
 441     } else {
 442       print_multi_length(phase_id, phase);
 443     }
 444   }
 445 
 446  private:
 447 
 448   void print_single_length(G1GCPhaseTimes::GCParPhases phase_id, WorkerDataArray<double>* phase) {
 449     // No need for min, max, average and sum for only one worker
 450     LineBuffer buf(phase->_indent_level);
 451     buf.append_and_print_cr("[%s:  %.1lf]", phase->_title, _phase_times->get_time_ms(phase_id, 0));
 452 
 453     if (phase->_thread_work_items != NULL) {
 454       LineBuffer buf2(phase->_thread_work_items->_indent_level);
 455       buf2.append_and_print_cr("[%s:  "SIZE_FORMAT"]", phase->_thread_work_items->_title, _phase_times->sum_thread_work_items(phase_id));
 456     }
 457   }
 458 
 459   void print_time_values(LineBuffer& buf, G1GCPhaseTimes::GCParPhases phase_id, WorkerDataArray<double>* phase) {
 460     uint active_length = _phase_times->_active_gc_threads;
 461     for (uint i = 0; i < active_length; ++i) {
 462       buf.append("  %.1lf", _phase_times->get_time_ms(phase_id, i));
 463     }
 464     buf.print_cr();
 465   }
 466 
 467   void print_count_values(LineBuffer& buf, G1GCPhaseTimes::GCParPhases phase_id, WorkerDataArray<size_t>* thread_work_items) {
 468     uint active_length = _phase_times->_active_gc_threads;
 469     for (uint i = 0; i < active_length; ++i) {
 470       buf.append("  " SIZE_FORMAT, _phase_times->get_thread_work_item(phase_id, i));
 471     }
 472     buf.print_cr();
 473   }
 474 
 475   void print_thread_work_items(G1GCPhaseTimes::GCParPhases phase_id, WorkerDataArray<size_t>* thread_work_items) {
 476     LineBuffer buf(thread_work_items->_indent_level);
 477     buf.append("[%s:", thread_work_items->_title);
 478 
 479     if (G1Log::finest()) {
 480       print_count_values(buf, phase_id, thread_work_items);
 481     }
 482 
 483     assert(thread_work_items->_print_sum, err_msg("%s does not have print sum true even though it is a count", thread_work_items->_title));
 484 
 485     buf.append_and_print_cr(" Min: " SIZE_FORMAT ", Avg: %.1lf, Max: " SIZE_FORMAT ", Diff: " SIZE_FORMAT ", Sum: " SIZE_FORMAT "]",
 486         _phase_times->min_thread_work_items(phase_id), _phase_times->average_thread_work_items(phase_id), _phase_times->max_thread_work_items(phase_id),
 487         _phase_times->max_thread_work_items(phase_id) - _phase_times->min_thread_work_items(phase_id), _phase_times->sum_thread_work_items(phase_id));
 488   }
 489 
 490   void print_multi_length(G1GCPhaseTimes::GCParPhases phase_id, WorkerDataArray<double>* phase) {
 491     LineBuffer buf(phase->_indent_level);
 492     buf.append("[%s:", phase->_title);
 493 
 494     if (G1Log::finest()) {
 495       print_time_values(buf, phase_id, phase);
 496     }
 497 
 498     buf.append(" Min: %.1lf, Avg: %.1lf, Max: %.1lf, Diff: %.1lf",
 499         _phase_times->min_time_ms(phase_id), _phase_times->average_time_ms(phase_id), _phase_times->max_time_ms(phase_id),
 500         _phase_times->max_time_ms(phase_id) - _phase_times->min_time_ms(phase_id));
 501 
 502     if (phase->_print_sum) {
 503       // for things like the start and end times the sum is not
 504       // that relevant
 505       buf.append(", Sum: %.1lf", _phase_times->sum_time_ms(phase_id));
 506     }
 507 
 508     buf.append_and_print_cr("]");
 509 
 510     if (phase->_thread_work_items != NULL) {
 511       print_thread_work_items(phase_id, phase->_thread_work_items);
 512     }
 513   }
 514 };
 515 
 516 void G1GCPhaseTimes::print(double pause_time_sec) {
 517   G1GCParPhasePrinter par_phase_printer(this);
 518 
 519   if (_root_region_scan_wait_time_ms > 0.0) {
 520     print_stats(1, "Root Region Scan Waiting", _root_region_scan_wait_time_ms);
 521   }
 522 
 523   print_stats(1, "Parallel Time", _cur_collection_par_time_ms, _active_gc_threads);
 524   for (int i = 0; i <= GCMainParPhasesLast; i++) {
 525     par_phase_printer.print((GCParPhases) i);
 526   }
 527 
 528   print_stats(1, "Code Root Fixup", _cur_collection_code_root_fixup_time_ms);
 529   print_stats(1, "Code Root Purge", _cur_strong_code_root_purge_time_ms);
 530   if (G1StringDedup::is_enabled()) {
 531     print_stats(1, "String Dedup Fixup", _cur_string_dedup_fixup_time_ms, _active_gc_threads);
 532     for (int i = StringDedupPhasesFirst; i <= StringDedupPhasesLast; i++) {
 533       par_phase_printer.print((GCParPhases) i);
 534     }
 535   }
 536   print_stats(1, "Clear CT", _cur_clear_ct_time_ms);
 537   double misc_time_ms = pause_time_sec * MILLIUNITS - accounted_time_ms();
 538   print_stats(1, "Other", misc_time_ms);
 539   if (_cur_verify_before_time_ms > 0.0) {
 540     print_stats(2, "Verify Before", _cur_verify_before_time_ms);
 541   }
 542   if (G1CollectedHeap::heap()->evacuation_failed()) {
 543     double evac_fail_handling = _cur_evac_fail_recalc_used + _cur_evac_fail_remove_self_forwards +
 544       _cur_evac_fail_restore_remsets;
 545     print_stats(2, "Evacuation Failure", evac_fail_handling);
 546     if (G1Log::finest()) {
 547       print_stats(3, "Recalculate Used", _cur_evac_fail_recalc_used);
 548       print_stats(3, "Remove Self Forwards", _cur_evac_fail_remove_self_forwards);
 549       print_stats(3, "Restore RemSet", _cur_evac_fail_restore_remsets);
 550     }
 551   }
 552   print_stats(2, "Choose CSet",
 553     (_recorded_young_cset_choice_time_ms +
 554     _recorded_non_young_cset_choice_time_ms));
 555   print_stats(2, "Ref Proc", _cur_ref_proc_time_ms);
 556   print_stats(2, "Ref Enq", _cur_ref_enq_time_ms);
 557   print_stats(2, "Redirty Cards", _recorded_redirty_logged_cards_time_ms);
 558   par_phase_printer.print(RedirtyCards);
 559   if (G1EagerReclaimHumongousObjects) {
 560     print_stats(2, "Humongous Register", _cur_fast_reclaim_humongous_register_time_ms);
 561     if (G1Log::finest()) {
 562       print_stats(3, "Humongous Total", _cur_fast_reclaim_humongous_total);
 563       print_stats(3, "Humongous Candidate", _cur_fast_reclaim_humongous_candidates);
 564     }
 565     print_stats(2, "Humongous Reclaim", _cur_fast_reclaim_humongous_time_ms);
 566     if (G1Log::finest()) {
 567       print_stats(3, "Humongous Reclaimed", _cur_fast_reclaim_humongous_reclaimed);
 568     }
 569   }
 570   print_stats(2, "Free CSet",
 571     (_recorded_young_free_cset_time_ms +
 572     _recorded_non_young_free_cset_time_ms));
 573   if (G1Log::finest()) {
 574     print_stats(3, "Young Free CSet", _recorded_young_free_cset_time_ms);
 575     print_stats(3, "Non-Young Free CSet", _recorded_non_young_free_cset_time_ms);
 576   }
 577   if (_cur_verify_after_time_ms > 0.0) {
 578     print_stats(2, "Verify After", _cur_verify_after_time_ms);
 579   }
 580 }
 581 
 582 G1GCParPhaseTimesTracker::G1GCParPhaseTimesTracker(G1GCPhaseTimes* phase_times, G1GCPhaseTimes::GCParPhases phase, uint worker_id) :
 583     _phase_times(phase_times), _phase(phase), _worker_id(worker_id) {
 584   if (_phase_times != NULL) {
 585     _start_time = os::elapsedTime();
 586   }
 587 }
 588 
 589 G1GCParPhaseTimesTracker::~G1GCParPhaseTimesTracker() {
 590   if (_phase_times != NULL) {
 591     _phase_times->record_time_secs(_phase, _worker_id, os::elapsedTime() - _start_time);
 592   }
 593 }
 594