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 
  26 #include "precompiled.hpp"
  27 #include "gc_implementation/g1/g1CollectedHeap.inline.hpp"
  28 #include "gc_implementation/g1/g1GCPhaseTimes.hpp"
  29 #include "gc_implementation/g1/g1Log.hpp"
  30 #include "gc_implementation/g1/g1StringDedup.hpp"
  31 #include "runtime/atomic.inline.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 append_and_print_cr(const char* format, ...)  ATTRIBUTE_PRINTF(2, 3) {
  75     va_list ap;
  76     va_start(ap, format);
  77     vappend(format, ap);
  78     va_end(ap);
  79     gclog_or_tty->print_cr("%s", _buffer);
  80     _cur = _indent_level * INDENT_CHARS;
  81   }
  82 };
  83 
  84 PRAGMA_DIAG_PUSH
  85 PRAGMA_FORMAT_NONLITERAL_IGNORED
  86 template <class T>
  87 void WorkerDataArray<T>::print(int level, const char* title) {
  88   if (_length == 1) {
  89     // No need for min, max, average and sum for only one worker
  90     LineBuffer buf(level);
  91     buf.append("[%s:  ", title);
  92     buf.append(_print_format, _data[0]);
  93     buf.append_and_print_cr("]");
  94     return;
  95   }
  96 
  97   T min = _data[0];
  98   T max = _data[0];
  99   T sum = 0;
 100 
 101   LineBuffer buf(level);
 102   buf.append("[%s:", title);
 103   for (uint i = 0; i < _length; ++i) {
 104     T val = _data[i];
 105     min = MIN2(val, min);
 106     max = MAX2(val, max);
 107     sum += val;
 108     if (G1Log::finest()) {
 109       buf.append("  ");
 110       buf.append(_print_format, val);
 111     }
 112   }
 113 
 114   if (G1Log::finest()) {
 115     buf.append_and_print_cr("%s", "");
 116   }
 117 
 118   double avg = (double)sum / (double)_length;
 119   buf.append(" Min: ");
 120   buf.append(_print_format, min);
 121   buf.append(", Avg: ");
 122   buf.append("%.1lf", avg); // Always print average as a double
 123   buf.append(", Max: ");
 124   buf.append(_print_format, max);
 125   buf.append(", Diff: ");
 126   buf.append(_print_format, max - min);
 127   if (_print_sum) {
 128     // for things like the start and end times the sum is not
 129     // that relevant
 130     buf.append(", Sum: ");
 131     buf.append(_print_format, sum);
 132   }
 133   buf.append_and_print_cr("]");
 134 }
 135 PRAGMA_DIAG_POP
 136 
 137 #ifndef PRODUCT
 138 
 139 template <> const int WorkerDataArray<int>::_uninitialized = -1;
 140 template <> const double WorkerDataArray<double>::_uninitialized = -1.0;
 141 template <> const size_t WorkerDataArray<size_t>::_uninitialized = (size_t)-1;
 142 
 143 template <class T>
 144 void WorkerDataArray<T>::reset() {
 145   for (uint i = 0; i < _length; i++) {
 146     _data[i] = (T)_uninitialized;
 147   }
 148 }
 149 
 150 template <class T>
 151 void WorkerDataArray<T>::verify() {
 152   for (uint i = 0; i < _length; i++) {
 153     assert(_data[i] != _uninitialized,
 154         err_msg("Invalid data for worker %u, data: %lf, uninitialized: %lf",
 155             i, (double)_data[i], (double)_uninitialized));
 156   }
 157 }
 158 
 159 #endif
 160 
 161 G1GCPhaseTimes::G1GCPhaseTimes(uint max_gc_threads, uint num_ext_root_scan_phases) :
 162   _max_gc_threads(max_gc_threads),
 163   _num_ext_root_scan_phases(num_ext_root_scan_phases),
 164   _last_gc_worker_start_times_ms(_max_gc_threads, "%.1lf", false),
 165   _last_ext_root_scan_times_ms(_max_gc_threads, "%.1lf"),
 166   _last_ext_root_scan_phase_times_ms(NULL),
 167   _last_satb_filtering_times_ms(_max_gc_threads, "%.1lf"),
 168   _last_update_rs_times_ms(_max_gc_threads, "%.1lf"),
 169   _last_update_rs_processed_buffers(_max_gc_threads, "%d"),
 170   _last_scan_rs_times_ms(_max_gc_threads, "%.1lf"),
 171   _last_strong_code_root_scan_times_ms(_max_gc_threads, "%.1lf"),
 172   _last_obj_copy_times_ms(_max_gc_threads, "%.1lf"),
 173   _last_termination_times_ms(_max_gc_threads, "%.1lf"),
 174   _last_termination_attempts(_max_gc_threads, SIZE_FORMAT),
 175   _last_gc_worker_end_times_ms(_max_gc_threads, "%.1lf", false),
 176   _last_gc_worker_times_ms(_max_gc_threads, "%.1lf"),
 177   _last_gc_worker_other_times_ms(_max_gc_threads, "%.1lf"),
 178   _last_redirty_logged_cards_time_ms(_max_gc_threads, "%.1lf"),
 179   _last_redirty_logged_cards_processed_cards(_max_gc_threads, SIZE_FORMAT),
 180   _cur_string_dedup_queue_fixup_worker_times_ms(_max_gc_threads, "%.1lf"),
 181   _cur_string_dedup_table_fixup_worker_times_ms(_max_gc_threads, "%.1lf")
 182 {
 183   assert(max_gc_threads > 0, "Must have some GC threads");
 184   if (track_ext_root_scan_phases()) {
 185     _last_ext_root_scan_phase_times_ms = NEW_C_HEAP_ARRAY(WorkerDataArray<double>*, num_ext_root_scan_phases, mtGC);
 186     for (uint i = 0; i < num_ext_root_scan_phases; i++) {
 187       _last_ext_root_scan_phase_times_ms[i] = new WorkerDataArray<double>(_max_gc_threads, "%.1lf");
 188     }
 189   }
 190 }
 191 
 192 G1GCPhaseTimes::~G1GCPhaseTimes() {
 193   if (track_ext_root_scan_phases()) {
 194     for (uint i = 0; i < _num_ext_root_scan_phases; i++) {
 195       delete _last_ext_root_scan_phase_times_ms[i];
 196     }
 197     FREE_C_HEAP_ARRAY(WorkerDataArray<double>*, _last_ext_root_scan_phase_times_ms);
 198   }
 199 }
 200 
 201 void G1GCPhaseTimes::note_gc_start(uint active_gc_threads) {
 202   assert(active_gc_threads > 0, "The number of threads must be > 0");
 203   assert(active_gc_threads <= _max_gc_threads, "The number of active threads must be <= the max nubmer of threads");
 204   _active_gc_threads = active_gc_threads;
 205 
 206   _last_gc_worker_start_times_ms.reset();
 207   _last_ext_root_scan_times_ms.reset();
 208   _last_satb_filtering_times_ms.reset();
 209   _last_update_rs_times_ms.reset();
 210   _last_update_rs_processed_buffers.reset();
 211   _last_scan_rs_times_ms.reset();
 212   _last_strong_code_root_scan_times_ms.reset();
 213   _last_obj_copy_times_ms.reset();
 214   _last_termination_times_ms.reset();
 215   _last_termination_attempts.reset();
 216   _last_gc_worker_end_times_ms.reset();
 217   _last_gc_worker_times_ms.reset();
 218   _last_gc_worker_other_times_ms.reset();
 219 
 220   _last_redirty_logged_cards_time_ms.reset();
 221   _last_redirty_logged_cards_processed_cards.reset();
 222 
 223   for (uint i = 0; i < _num_ext_root_scan_phases; i++) {
 224     _last_ext_root_scan_phase_times_ms[i]->reset();
 225   }
 226 }
 227 
 228 void G1GCPhaseTimes::note_gc_end() {
 229   _last_gc_worker_start_times_ms.verify();
 230   _last_ext_root_scan_times_ms.verify();
 231   _last_satb_filtering_times_ms.verify();
 232   _last_update_rs_times_ms.verify();
 233   _last_update_rs_processed_buffers.verify();
 234   _last_scan_rs_times_ms.verify();
 235   _last_strong_code_root_scan_times_ms.verify();
 236   _last_obj_copy_times_ms.verify();
 237   _last_termination_times_ms.verify();
 238   _last_termination_attempts.verify();
 239   _last_gc_worker_end_times_ms.verify();
 240 
 241   for (uint i = 0; i < _active_gc_threads; i++) {
 242     double worker_time = _last_gc_worker_end_times_ms.get(i) - _last_gc_worker_start_times_ms.get(i);
 243     _last_gc_worker_times_ms.set(i, worker_time);
 244 
 245     double worker_known_time = _last_ext_root_scan_times_ms.get(i) +
 246                                _last_satb_filtering_times_ms.get(i) +
 247                                _last_update_rs_times_ms.get(i) +
 248                                _last_scan_rs_times_ms.get(i) +
 249                                _last_strong_code_root_scan_times_ms.get(i) +
 250                                _last_obj_copy_times_ms.get(i) +
 251                                _last_termination_times_ms.get(i);
 252 
 253     double worker_other_time = worker_time - worker_known_time;
 254     _last_gc_worker_other_times_ms.set(i, worker_other_time);
 255   }
 256 
 257   _last_gc_worker_times_ms.verify();
 258   _last_gc_worker_other_times_ms.verify();
 259 
 260   _last_redirty_logged_cards_time_ms.verify();
 261   _last_redirty_logged_cards_processed_cards.verify();
 262 }
 263 
 264 void G1GCPhaseTimes::note_string_dedup_fixup_start() {
 265   _cur_string_dedup_queue_fixup_worker_times_ms.reset();
 266   _cur_string_dedup_table_fixup_worker_times_ms.reset();
 267 }
 268 
 269 void G1GCPhaseTimes::note_string_dedup_fixup_end() {
 270   _cur_string_dedup_queue_fixup_worker_times_ms.verify();
 271   _cur_string_dedup_table_fixup_worker_times_ms.verify();
 272 }
 273 
 274 void G1GCPhaseTimes::print_stats(int level, const char* str, double value) {
 275   LineBuffer(level).append_and_print_cr("[%s: %.1lf ms]", str, value);
 276 }
 277 
 278 void G1GCPhaseTimes::print_stats(int level, const char* str, size_t value) {
 279   LineBuffer(level).append_and_print_cr("[%s: "SIZE_FORMAT"]", str, value);
 280 }
 281 
 282 void G1GCPhaseTimes::print_stats(int level, const char* str, double value, uint workers) {
 283   LineBuffer(level).append_and_print_cr("[%s: %.1lf ms, GC Workers: %u]", str, value, workers);
 284 }
 285 
 286 double G1GCPhaseTimes::accounted_time_ms() {
 287     // Subtract the root region scanning wait time. It's initialized to
 288     // zero at the start of the pause.
 289     double misc_time_ms = _root_region_scan_wait_time_ms;
 290 
 291     misc_time_ms += _cur_collection_par_time_ms;
 292 
 293     // Now subtract the time taken to fix up roots in generated code
 294     misc_time_ms += _cur_collection_code_root_fixup_time_ms;
 295 
 296     // Strong code root purge time
 297     misc_time_ms += _cur_strong_code_root_purge_time_ms;
 298 
 299     if (G1StringDedup::is_enabled()) {
 300       // String dedup fixup time
 301       misc_time_ms += _cur_string_dedup_fixup_time_ms;
 302     }
 303 
 304     // Subtract the time taken to clean the card table from the
 305     // current value of "other time"
 306     misc_time_ms += _cur_clear_ct_time_ms;
 307 
 308     return misc_time_ms;
 309 }
 310 
 311 void G1GCPhaseTimes::print(double pause_time_sec) {
 312   if (_root_region_scan_wait_time_ms > 0.0) {
 313     print_stats(1, "Root Region Scan Waiting", _root_region_scan_wait_time_ms);
 314   }
 315   print_stats(1, "Parallel Time", _cur_collection_par_time_ms, _active_gc_threads);
 316   _last_gc_worker_start_times_ms.print(2, "GC Worker Start (ms)");
 317   _last_ext_root_scan_times_ms.print(2, "Ext Root Scanning (ms)");
 318   if (track_ext_root_scan_phases()) {
 319     for (uint i = 0; i < _num_ext_root_scan_phases; i++) {
 320       WorkerDataArray<double>* data = _last_ext_root_scan_phase_times_ms[i];
 321       data->print(3, G1CollectedHeap::ext_roots_task_string(i));
 322     }
 323   }
 324   if (_last_satb_filtering_times_ms.sum() > 0.0) {
 325     _last_satb_filtering_times_ms.print(2, "SATB Filtering (ms)");
 326   }
 327   _last_update_rs_times_ms.print(2, "Update RS (ms)");
 328     _last_update_rs_processed_buffers.print(3, "Processed Buffers");
 329   _last_scan_rs_times_ms.print(2, "Scan RS (ms)");
 330   _last_strong_code_root_scan_times_ms.print(2, "Code Root Scanning (ms)");
 331   _last_obj_copy_times_ms.print(2, "Object Copy (ms)");
 332   _last_termination_times_ms.print(2, "Termination (ms)");
 333   if (G1Log::finest()) {
 334     _last_termination_attempts.print(3, "Termination Attempts");
 335   }
 336   _last_gc_worker_other_times_ms.print(2, "GC Worker Other (ms)");
 337   _last_gc_worker_times_ms.print(2, "GC Worker Total (ms)");
 338   _last_gc_worker_end_times_ms.print(2, "GC Worker End (ms)");
 339 
 340   print_stats(1, "Code Root Fixup", _cur_collection_code_root_fixup_time_ms);
 341   print_stats(1, "Code Root Purge", _cur_strong_code_root_purge_time_ms);
 342   if (G1StringDedup::is_enabled()) {
 343     print_stats(1, "String Dedup Fixup", _cur_string_dedup_fixup_time_ms, _active_gc_threads);
 344     _cur_string_dedup_queue_fixup_worker_times_ms.print(2, "Queue Fixup (ms)");
 345     _cur_string_dedup_table_fixup_worker_times_ms.print(2, "Table Fixup (ms)");
 346   }
 347   print_stats(1, "Clear CT", _cur_clear_ct_time_ms);
 348   double misc_time_ms = pause_time_sec * MILLIUNITS - accounted_time_ms();
 349   print_stats(1, "Other", misc_time_ms);
 350   if (_cur_verify_before_time_ms > 0.0) {
 351     print_stats(2, "Verify Before", _cur_verify_before_time_ms);
 352   }
 353   if (G1CollectedHeap::heap()->evacuation_failed()) {
 354     double evac_fail_handling = _cur_evac_fail_recalc_used + _cur_evac_fail_remove_self_forwards +
 355       _cur_evac_fail_restore_remsets;
 356     print_stats(2, "Evacuation Failure", evac_fail_handling);
 357     if (G1Log::finest()) {
 358       print_stats(3, "Recalculate Used", _cur_evac_fail_recalc_used);
 359       print_stats(3, "Remove Self Forwards", _cur_evac_fail_remove_self_forwards);
 360       print_stats(3, "Restore RemSet", _cur_evac_fail_restore_remsets);
 361     }
 362   }
 363   print_stats(2, "Choose CSet",
 364     (_recorded_young_cset_choice_time_ms +
 365     _recorded_non_young_cset_choice_time_ms));
 366   print_stats(2, "Ref Proc", _cur_ref_proc_time_ms);
 367   print_stats(2, "Ref Enq", _cur_ref_enq_time_ms);
 368   print_stats(2, "Redirty Cards", _recorded_redirty_logged_cards_time_ms);
 369   if (G1Log::finest()) {
 370     _last_redirty_logged_cards_time_ms.print(3, "Parallel Redirty");
 371     _last_redirty_logged_cards_processed_cards.print(3, "Redirtied Cards");
 372   }
 373   if (G1EagerReclaimHumongousObjects) {
 374     print_stats(2, "Humongous Register", _cur_fast_reclaim_humongous_register_time_ms);
 375     if (G1Log::finest()) {
 376       print_stats(3, "Humongous Total", _cur_fast_reclaim_humongous_total);
 377       print_stats(3, "Humongous Candidate", _cur_fast_reclaim_humongous_candidates);
 378     }
 379     print_stats(2, "Humongous Reclaim", _cur_fast_reclaim_humongous_time_ms);
 380     if (G1Log::finest()) {
 381       print_stats(3, "Humongous Reclaimed", _cur_fast_reclaim_humongous_reclaimed);
 382     }
 383   }
 384   print_stats(2, "Free CSet",
 385     (_recorded_young_free_cset_time_ms +
 386     _recorded_non_young_free_cset_time_ms));
 387   if (G1Log::finest()) {
 388     print_stats(3, "Young Free CSet", _recorded_young_free_cset_time_ms);
 389     print_stats(3, "Non-Young Free CSet", _recorded_non_young_free_cset_time_ms);
 390   }
 391   if (_cur_verify_after_time_ms > 0.0) {
 392     print_stats(2, "Verify After", _cur_verify_after_time_ms);
 393   }
 394 }