1 /*
   2  * Copyright (c) 2013, 2015, 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/g1Log.hpp"
  29 #include "gc/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, "Worker %d is greater than max: %d", worker_i, _length);
 134     assert(_data[worker_i] == WorkerDataArray<T>::uninitialized(), "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, "Worker %d is greater than max: %d", worker_i, _length);
 146     assert(_data[worker_i] != WorkerDataArray<T>::uninitialized(), "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, "Worker %d is greater than max: %d", worker_i, _length);
 152     assert(_data[worker_i] != WorkerDataArray<T>::uninitialized(), "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            "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[CMRefRoots] = new WorkerDataArray<double>(max_gc_threads, "CM RefProcessor Roots (ms)", true, G1Log::LevelFinest, 3);
 267   _gc_par_phases[WaitForStrongCLD] = new WorkerDataArray<double>(max_gc_threads, "Wait For Strong CLD (ms)", true, G1Log::LevelFinest, 3);
 268   _gc_par_phases[WeakCLDRoots] = new WorkerDataArray<double>(max_gc_threads, "Weak CLD Roots (ms)", true, G1Log::LevelFinest, 3);
 269   _gc_par_phases[SATBFiltering] = new WorkerDataArray<double>(max_gc_threads, "SATB Filtering (ms)", true, G1Log::LevelFinest, 3);
 270 
 271   _gc_par_phases[UpdateRS] = new WorkerDataArray<double>(max_gc_threads, "Update RS (ms)", true, G1Log::LevelFiner, 2);
 272   _gc_par_phases[ScanRS] = new WorkerDataArray<double>(max_gc_threads, "Scan RS (ms)", true, G1Log::LevelFiner, 2);
 273   _gc_par_phases[CodeRoots] = new WorkerDataArray<double>(max_gc_threads, "Code Root Scanning (ms)", true, G1Log::LevelFiner, 2);
 274   _gc_par_phases[ObjCopy] = new WorkerDataArray<double>(max_gc_threads, "Object Copy (ms)", true, G1Log::LevelFiner, 2);
 275   _gc_par_phases[Termination] = new WorkerDataArray<double>(max_gc_threads, "Termination (ms)", true, G1Log::LevelFiner, 2);
 276   _gc_par_phases[GCWorkerTotal] = new WorkerDataArray<double>(max_gc_threads, "GC Worker Total (ms)", true, G1Log::LevelFiner, 2);
 277   _gc_par_phases[GCWorkerEnd] = new WorkerDataArray<double>(max_gc_threads, "GC Worker End (ms)", false, G1Log::LevelFiner, 2);
 278   _gc_par_phases[Other] = new WorkerDataArray<double>(max_gc_threads, "GC Worker Other (ms)", true, G1Log::LevelFiner, 2);
 279 
 280   _update_rs_processed_buffers = new WorkerDataArray<size_t>(max_gc_threads, "Processed Buffers", true, G1Log::LevelFiner, 3);
 281   _gc_par_phases[UpdateRS]->link_thread_work_items(_update_rs_processed_buffers);
 282 
 283   _termination_attempts = new WorkerDataArray<size_t>(max_gc_threads, "Termination Attempts", true, G1Log::LevelFinest, 3);
 284   _gc_par_phases[Termination]->link_thread_work_items(_termination_attempts);
 285 
 286   _gc_par_phases[StringDedupQueueFixup] = new WorkerDataArray<double>(max_gc_threads, "Queue Fixup (ms)", true, G1Log::LevelFiner, 2);
 287   _gc_par_phases[StringDedupTableFixup] = new WorkerDataArray<double>(max_gc_threads, "Table Fixup (ms)", true, G1Log::LevelFiner, 2);
 288 
 289   _gc_par_phases[RedirtyCards] = new WorkerDataArray<double>(max_gc_threads, "Parallel Redirty", true, G1Log::LevelFinest, 3);
 290   _redirtied_cards = new WorkerDataArray<size_t>(max_gc_threads, "Redirtied Cards", true, G1Log::LevelFinest, 3);
 291   _gc_par_phases[RedirtyCards]->link_thread_work_items(_redirtied_cards);
 292 }
 293 
 294 void G1GCPhaseTimes::note_gc_start(uint active_gc_threads, bool mark_in_progress) {
 295   assert(active_gc_threads > 0, "The number of threads must be > 0");
 296   assert(active_gc_threads <= _max_gc_threads, "The number of active threads must be <= the max number of threads");
 297   _active_gc_threads = active_gc_threads;
 298 
 299   for (int i = 0; i < GCParPhasesSentinel; i++) {
 300     _gc_par_phases[i]->reset();
 301   }
 302 
 303   _gc_par_phases[StringDedupQueueFixup]->set_enabled(G1StringDedup::is_enabled());
 304   _gc_par_phases[StringDedupTableFixup]->set_enabled(G1StringDedup::is_enabled());
 305 }
 306 
 307 void G1GCPhaseTimes::note_gc_end() {
 308   for (uint i = 0; i < _active_gc_threads; i++) {
 309     double worker_time = _gc_par_phases[GCWorkerEnd]->get(i) - _gc_par_phases[GCWorkerStart]->get(i);
 310     record_time_secs(GCWorkerTotal, i , worker_time);
 311 
 312     double worker_known_time =
 313         _gc_par_phases[ExtRootScan]->get(i) +
 314         _gc_par_phases[SATBFiltering]->get(i) +
 315         _gc_par_phases[UpdateRS]->get(i) +
 316         _gc_par_phases[ScanRS]->get(i) +
 317         _gc_par_phases[CodeRoots]->get(i) +
 318         _gc_par_phases[ObjCopy]->get(i) +
 319         _gc_par_phases[Termination]->get(i);
 320 
 321     record_time_secs(Other, i, worker_time - worker_known_time);
 322   }
 323 
 324   for (int i = 0; i < GCParPhasesSentinel; i++) {
 325     _gc_par_phases[i]->verify(_active_gc_threads);
 326   }
 327 }
 328 
 329 void G1GCPhaseTimes::print_stats(int level, const char* str, double value) {
 330   LineBuffer(level).append_and_print_cr("[%s: %.1lf ms]", str, value);
 331 }
 332 
 333 void G1GCPhaseTimes::print_stats(int level, const char* str, size_t value) {
 334   LineBuffer(level).append_and_print_cr("[%s: " SIZE_FORMAT "]", str, value);
 335 }
 336 
 337 void G1GCPhaseTimes::print_stats(int level, const char* str, double value, uint workers) {
 338   LineBuffer(level).append_and_print_cr("[%s: %.1lf ms, GC Workers: %u]", str, value, workers);
 339 }
 340 
 341 double G1GCPhaseTimes::accounted_time_ms() {
 342     // Subtract the root region scanning wait time. It's initialized to
 343     // zero at the start of the pause.
 344     double misc_time_ms = _root_region_scan_wait_time_ms;
 345 
 346     misc_time_ms += _cur_collection_par_time_ms;
 347 
 348     // Now subtract the time taken to fix up roots in generated code
 349     misc_time_ms += _cur_collection_code_root_fixup_time_ms;
 350 
 351     // Strong code root purge time
 352     misc_time_ms += _cur_strong_code_root_purge_time_ms;
 353 
 354     if (G1StringDedup::is_enabled()) {
 355       // String dedup fixup time
 356       misc_time_ms += _cur_string_dedup_fixup_time_ms;
 357     }
 358 
 359     // Subtract the time taken to clean the card table from the
 360     // current value of "other time"
 361     misc_time_ms += _cur_clear_ct_time_ms;
 362 
 363     return misc_time_ms;
 364 }
 365 
 366 // record the time a phase took in seconds
 367 void G1GCPhaseTimes::record_time_secs(GCParPhases phase, uint worker_i, double secs) {
 368   _gc_par_phases[phase]->set(worker_i, secs);
 369 }
 370 
 371 // add a number of seconds to a phase
 372 void G1GCPhaseTimes::add_time_secs(GCParPhases phase, uint worker_i, double secs) {
 373   _gc_par_phases[phase]->add(worker_i, secs);
 374 }
 375 
 376 void G1GCPhaseTimes::record_thread_work_item(GCParPhases phase, uint worker_i, size_t count) {
 377   _gc_par_phases[phase]->set_thread_work_item(worker_i, count);
 378 }
 379 
 380 // return the average time for a phase in milliseconds
 381 double G1GCPhaseTimes::average_time_ms(GCParPhases phase) {
 382   return _gc_par_phases[phase]->average(_active_gc_threads) * 1000.0;
 383 }
 384 
 385 double G1GCPhaseTimes::get_time_ms(GCParPhases phase, uint worker_i) {
 386   return _gc_par_phases[phase]->get(worker_i) * 1000.0;
 387 }
 388 
 389 double G1GCPhaseTimes::sum_time_ms(GCParPhases phase) {
 390   return _gc_par_phases[phase]->sum(_active_gc_threads) * 1000.0;
 391 }
 392 
 393 double G1GCPhaseTimes::min_time_ms(GCParPhases phase) {
 394   return _gc_par_phases[phase]->minimum(_active_gc_threads) * 1000.0;
 395 }
 396 
 397 double G1GCPhaseTimes::max_time_ms(GCParPhases phase) {
 398   return _gc_par_phases[phase]->maximum(_active_gc_threads) * 1000.0;
 399 }
 400 
 401 size_t G1GCPhaseTimes::get_thread_work_item(GCParPhases phase, uint worker_i) {
 402   assert(_gc_par_phases[phase]->thread_work_items() != NULL, "No sub count");
 403   return _gc_par_phases[phase]->thread_work_items()->get(worker_i);
 404 }
 405 
 406 size_t G1GCPhaseTimes::sum_thread_work_items(GCParPhases phase) {
 407   assert(_gc_par_phases[phase]->thread_work_items() != NULL, "No sub count");
 408   return _gc_par_phases[phase]->thread_work_items()->sum(_active_gc_threads);
 409 }
 410 
 411 double G1GCPhaseTimes::average_thread_work_items(GCParPhases phase) {
 412   assert(_gc_par_phases[phase]->thread_work_items() != NULL, "No sub count");
 413   return _gc_par_phases[phase]->thread_work_items()->average(_active_gc_threads);
 414 }
 415 
 416 size_t G1GCPhaseTimes::min_thread_work_items(GCParPhases phase) {
 417   assert(_gc_par_phases[phase]->thread_work_items() != NULL, "No sub count");
 418   return _gc_par_phases[phase]->thread_work_items()->minimum(_active_gc_threads);
 419 }
 420 
 421 size_t G1GCPhaseTimes::max_thread_work_items(GCParPhases phase) {
 422   assert(_gc_par_phases[phase]->thread_work_items() != NULL, "No sub count");
 423   return _gc_par_phases[phase]->thread_work_items()->maximum(_active_gc_threads);
 424 }
 425 
 426 class G1GCParPhasePrinter : public StackObj {
 427   G1GCPhaseTimes* _phase_times;
 428  public:
 429   G1GCParPhasePrinter(G1GCPhaseTimes* phase_times) : _phase_times(phase_times) {}
 430 
 431   void print(G1GCPhaseTimes::GCParPhases phase_id) {
 432     WorkerDataArray<double>* phase = _phase_times->_gc_par_phases[phase_id];
 433 
 434     if (phase->_log_level > G1Log::level() || !phase->_enabled) {
 435       return;
 436     }
 437 
 438     if (phase->_length == 1) {
 439       print_single_length(phase_id, phase);
 440     } else {
 441       print_multi_length(phase_id, phase);
 442     }
 443   }
 444 
 445  private:
 446 
 447   void print_single_length(G1GCPhaseTimes::GCParPhases phase_id, WorkerDataArray<double>* phase) {
 448     // No need for min, max, average and sum for only one worker
 449     LineBuffer buf(phase->_indent_level);
 450     buf.append_and_print_cr("[%s:  %.1lf]", phase->_title, _phase_times->get_time_ms(phase_id, 0));
 451 
 452     if (phase->_thread_work_items != NULL) {
 453       LineBuffer buf2(phase->_thread_work_items->_indent_level);
 454       buf2.append_and_print_cr("[%s:  " SIZE_FORMAT "]", phase->_thread_work_items->_title, _phase_times->sum_thread_work_items(phase_id));
 455     }
 456   }
 457 
 458   void print_time_values(LineBuffer& buf, G1GCPhaseTimes::GCParPhases phase_id, WorkerDataArray<double>* phase) {
 459     uint active_length = _phase_times->_active_gc_threads;
 460     for (uint i = 0; i < active_length; ++i) {
 461       buf.append("  %.1lf", _phase_times->get_time_ms(phase_id, i));
 462     }
 463     buf.print_cr();
 464   }
 465 
 466   void print_count_values(LineBuffer& buf, G1GCPhaseTimes::GCParPhases phase_id, WorkerDataArray<size_t>* thread_work_items) {
 467     uint active_length = _phase_times->_active_gc_threads;
 468     for (uint i = 0; i < active_length; ++i) {
 469       buf.append("  " SIZE_FORMAT, _phase_times->get_thread_work_item(phase_id, i));
 470     }
 471     buf.print_cr();
 472   }
 473 
 474   void print_thread_work_items(G1GCPhaseTimes::GCParPhases phase_id, WorkerDataArray<size_t>* thread_work_items) {
 475     LineBuffer buf(thread_work_items->_indent_level);
 476     buf.append("[%s:", thread_work_items->_title);
 477 
 478     if (G1Log::finest()) {
 479       print_count_values(buf, phase_id, thread_work_items);
 480     }
 481 
 482     assert(thread_work_items->_print_sum, "%s does not have print sum true even though it is a count", thread_work_items->_title);
 483 
 484     buf.append_and_print_cr(" Min: " SIZE_FORMAT ", Avg: %.1lf, Max: " SIZE_FORMAT ", Diff: " SIZE_FORMAT ", Sum: " SIZE_FORMAT "]",
 485         _phase_times->min_thread_work_items(phase_id), _phase_times->average_thread_work_items(phase_id), _phase_times->max_thread_work_items(phase_id),
 486         _phase_times->max_thread_work_items(phase_id) - _phase_times->min_thread_work_items(phase_id), _phase_times->sum_thread_work_items(phase_id));
 487   }
 488 
 489   void print_multi_length(G1GCPhaseTimes::GCParPhases phase_id, WorkerDataArray<double>* phase) {
 490     LineBuffer buf(phase->_indent_level);
 491     buf.append("[%s:", phase->_title);
 492 
 493     if (G1Log::finest()) {
 494       print_time_values(buf, phase_id, phase);
 495     }
 496 
 497     buf.append(" Min: %.1lf, Avg: %.1lf, Max: %.1lf, Diff: %.1lf",
 498         _phase_times->min_time_ms(phase_id), _phase_times->average_time_ms(phase_id), _phase_times->max_time_ms(phase_id),
 499         _phase_times->max_time_ms(phase_id) - _phase_times->min_time_ms(phase_id));
 500 
 501     if (phase->_print_sum) {
 502       // for things like the start and end times the sum is not
 503       // that relevant
 504       buf.append(", Sum: %.1lf", _phase_times->sum_time_ms(phase_id));
 505     }
 506 
 507     buf.append_and_print_cr("]");
 508 
 509     if (phase->_thread_work_items != NULL) {
 510       print_thread_work_items(phase_id, phase->_thread_work_items);
 511     }
 512   }
 513 };
 514 
 515 void G1GCPhaseTimes::print(double pause_time_sec) {
 516   G1GCParPhasePrinter par_phase_printer(this);
 517 
 518   if (_root_region_scan_wait_time_ms > 0.0) {
 519     print_stats(1, "Root Region Scan Waiting", _root_region_scan_wait_time_ms);
 520   }
 521 
 522   print_stats(1, "Parallel Time", _cur_collection_par_time_ms, _active_gc_threads);
 523   for (int i = 0; i <= GCMainParPhasesLast; i++) {
 524     par_phase_printer.print((GCParPhases) i);
 525   }
 526 
 527   print_stats(1, "Code Root Fixup", _cur_collection_code_root_fixup_time_ms);
 528   print_stats(1, "Code Root Purge", _cur_strong_code_root_purge_time_ms);
 529   if (G1StringDedup::is_enabled()) {
 530     print_stats(1, "String Dedup Fixup", _cur_string_dedup_fixup_time_ms, _active_gc_threads);
 531     for (int i = StringDedupPhasesFirst; i <= StringDedupPhasesLast; i++) {
 532       par_phase_printer.print((GCParPhases) i);
 533     }
 534   }
 535   print_stats(1, "Clear CT", _cur_clear_ct_time_ms);
 536   double misc_time_ms = pause_time_sec * MILLIUNITS - accounted_time_ms();
 537   print_stats(1, "Other", misc_time_ms);
 538   if (_cur_verify_before_time_ms > 0.0) {
 539     print_stats(2, "Verify Before", _cur_verify_before_time_ms);
 540   }
 541   if (G1CollectedHeap::heap()->evacuation_failed()) {
 542     double evac_fail_handling = _cur_evac_fail_recalc_used + _cur_evac_fail_remove_self_forwards +
 543       _cur_evac_fail_restore_remsets;
 544     print_stats(2, "Evacuation Failure", evac_fail_handling);
 545     if (G1Log::finest()) {
 546       print_stats(3, "Recalculate Used", _cur_evac_fail_recalc_used);
 547       print_stats(3, "Remove Self Forwards", _cur_evac_fail_remove_self_forwards);
 548       print_stats(3, "Restore RemSet", _cur_evac_fail_restore_remsets);
 549     }
 550   }
 551   print_stats(2, "Choose CSet",
 552     (_recorded_young_cset_choice_time_ms +
 553     _recorded_non_young_cset_choice_time_ms));
 554   print_stats(2, "Ref Proc", _cur_ref_proc_time_ms);
 555   print_stats(2, "Ref Enq", _cur_ref_enq_time_ms);
 556   print_stats(2, "Redirty Cards", _recorded_redirty_logged_cards_time_ms);
 557   par_phase_printer.print(RedirtyCards);
 558   if (G1EagerReclaimHumongousObjects) {
 559     print_stats(2, "Humongous Register", _cur_fast_reclaim_humongous_register_time_ms);
 560     if (G1Log::finest()) {
 561       print_stats(3, "Humongous Total", _cur_fast_reclaim_humongous_total);
 562       print_stats(3, "Humongous Candidate", _cur_fast_reclaim_humongous_candidates);
 563     }
 564     print_stats(2, "Humongous Reclaim", _cur_fast_reclaim_humongous_time_ms);
 565     if (G1Log::finest()) {
 566       print_stats(3, "Humongous Reclaimed", _cur_fast_reclaim_humongous_reclaimed);
 567     }
 568   }
 569   print_stats(2, "Free CSet",
 570     (_recorded_young_free_cset_time_ms +
 571     _recorded_non_young_free_cset_time_ms));
 572   if (G1Log::finest()) {
 573     print_stats(3, "Young Free CSet", _recorded_young_free_cset_time_ms);
 574     print_stats(3, "Non-Young Free CSet", _recorded_non_young_free_cset_time_ms);
 575   }
 576   if (_cur_verify_after_time_ms > 0.0) {
 577     print_stats(2, "Verify After", _cur_verify_after_time_ms);
 578   }
 579 }
 580 
 581 G1GCParPhaseTimesTracker::G1GCParPhaseTimesTracker(G1GCPhaseTimes* phase_times, G1GCPhaseTimes::GCParPhases phase, uint worker_id) :
 582     _phase_times(phase_times), _phase(phase), _worker_id(worker_id) {
 583   if (_phase_times != NULL) {
 584     _start_time = os::elapsedTime();
 585   }
 586 }
 587 
 588 G1GCParPhaseTimesTracker::~G1GCParPhaseTimesTracker() {
 589   if (_phase_times != NULL) {
 590     _phase_times->record_time_secs(_phase, _worker_id, os::elapsedTime() - _start_time);
 591   }
 592 }
 593