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/concurrentG1Refine.hpp"
27 #include "gc/g1/g1CollectedHeap.inline.hpp"
28 #include "gc/g1/g1GCPhaseTimes.hpp"
29 #include "gc/g1/g1Log.hpp"
30 #include "gc/g1/g1StringDedup.hpp"
31 #include "gc/g1/workerDataArray.inline.hpp"
32 #include "memory/allocation.hpp"
33 #include "runtime/os.hpp"
34
35 // Helper class for avoiding interleaved logging
36 class LineBuffer: public StackObj {
37
38 private:
39 static const int BUFFER_LEN = 1024;
40 static const int INDENT_CHARS = 3;
41 char _buffer[BUFFER_LEN];
42 int _indent_level;
43 int _cur;
44
45 void vappend(const char* format, va_list ap) ATTRIBUTE_PRINTF(2, 0) {
46 int res = vsnprintf(&_buffer[_cur], BUFFER_LEN - _cur, format, ap);
47 if (res != -1) {
48 _cur += res;
49 } else {
50 DEBUG_ONLY(warning("buffer too small in LineBuffer");)
51 _buffer[BUFFER_LEN -1] = 0;
52 _cur = BUFFER_LEN; // vsnprintf above should not add to _buffer if we are called again
53 }
54 }
55
56 public:
57 explicit LineBuffer(int indent_level): _indent_level(indent_level), _cur(0) {
58 for (; (_cur < BUFFER_LEN && _cur < (_indent_level * INDENT_CHARS)); _cur++) {
59 _buffer[_cur] = ' ';
60 }
61 }
62
63 #ifndef PRODUCT
64 ~LineBuffer() {
65 assert(_cur == _indent_level * INDENT_CHARS, "pending data in buffer - append_and_print_cr() not called?");
66 }
67 #endif
68
69 void append(const char* format, ...) ATTRIBUTE_PRINTF(2, 3) {
70 va_list ap;
71 va_start(ap, format);
72 vappend(format, ap);
73 va_end(ap);
74 }
75
76 void print_cr() {
77 gclog_or_tty->print_cr("%s", _buffer);
78 _cur = _indent_level * INDENT_CHARS;
79 }
80
81 void append_and_print_cr(const char* format, ...) ATTRIBUTE_PRINTF(2, 3) {
82 va_list ap;
83 va_start(ap, format);
84 vappend(format, ap);
85 va_end(ap);
86 print_cr();
87 }
88 };
89
90 G1GCPhaseTimes::G1GCPhaseTimes(uint max_gc_threads) :
91 _max_gc_threads(max_gc_threads)
92 {
93 assert(max_gc_threads > 0, "Must have some GC threads");
94
95 _gc_par_phases[GCWorkerStart] = new WorkerDataArray<double>(max_gc_threads, "GC Worker Start (ms)", false, G1Log::LevelFiner, 2);
96 _gc_par_phases[ExtRootScan] = new WorkerDataArray<double>(max_gc_threads, "Ext Root Scanning (ms)", true, G1Log::LevelFiner, 2);
97
98 // Root scanning phases
99 _gc_par_phases[ThreadRoots] = new WorkerDataArray<double>(max_gc_threads, "Thread Roots (ms)", true, G1Log::LevelFinest, 3);
100 _gc_par_phases[StringTableRoots] = new WorkerDataArray<double>(max_gc_threads, "StringTable Roots (ms)", true, G1Log::LevelFinest, 3);
101 _gc_par_phases[UniverseRoots] = new WorkerDataArray<double>(max_gc_threads, "Universe Roots (ms)", true, G1Log::LevelFinest, 3);
102 _gc_par_phases[JNIRoots] = new WorkerDataArray<double>(max_gc_threads, "JNI Handles Roots (ms)", true, G1Log::LevelFinest, 3);
103 _gc_par_phases[ObjectSynchronizerRoots] = new WorkerDataArray<double>(max_gc_threads, "ObjectSynchronizer Roots (ms)", true, G1Log::LevelFinest, 3);
104 _gc_par_phases[FlatProfilerRoots] = new WorkerDataArray<double>(max_gc_threads, "FlatProfiler Roots (ms)", true, G1Log::LevelFinest, 3);
105 _gc_par_phases[ManagementRoots] = new WorkerDataArray<double>(max_gc_threads, "Management Roots (ms)", true, G1Log::LevelFinest, 3);
106 _gc_par_phases[SystemDictionaryRoots] = new WorkerDataArray<double>(max_gc_threads, "SystemDictionary Roots (ms)", true, G1Log::LevelFinest, 3);
107 _gc_par_phases[CLDGRoots] = new WorkerDataArray<double>(max_gc_threads, "CLDG Roots (ms)", true, G1Log::LevelFinest, 3);
108 _gc_par_phases[JVMTIRoots] = new WorkerDataArray<double>(max_gc_threads, "JVMTI Roots (ms)", true, G1Log::LevelFinest, 3);
109 _gc_par_phases[CMRefRoots] = new WorkerDataArray<double>(max_gc_threads, "CM RefProcessor Roots (ms)", true, G1Log::LevelFinest, 3);
110 _gc_par_phases[WaitForStrongCLD] = new WorkerDataArray<double>(max_gc_threads, "Wait For Strong CLD (ms)", true, G1Log::LevelFinest, 3);
111 _gc_par_phases[WeakCLDRoots] = new WorkerDataArray<double>(max_gc_threads, "Weak CLD Roots (ms)", true, G1Log::LevelFinest, 3);
112 _gc_par_phases[SATBFiltering] = new WorkerDataArray<double>(max_gc_threads, "SATB Filtering (ms)", true, G1Log::LevelFinest, 3);
113
114 _gc_par_phases[UpdateRS] = new WorkerDataArray<double>(max_gc_threads, "Update RS (ms)", true, G1Log::LevelFiner, 2);
115 _gc_par_phases[ScanHCC] = new WorkerDataArray<double>(max_gc_threads, "Scan HCC (ms)", true, G1Log::LevelFiner, 3);
116 _gc_par_phases[ScanHCC]->set_enabled(ConcurrentG1Refine::hot_card_cache_enabled());
117 _gc_par_phases[ScanRS] = new WorkerDataArray<double>(max_gc_threads, "Scan RS (ms)", true, G1Log::LevelFiner, 2);
118 _gc_par_phases[CodeRoots] = new WorkerDataArray<double>(max_gc_threads, "Code Root Scanning (ms)", true, G1Log::LevelFiner, 2);
119 _gc_par_phases[ObjCopy] = new WorkerDataArray<double>(max_gc_threads, "Object Copy (ms)", true, G1Log::LevelFiner, 2);
120 _gc_par_phases[Termination] = new WorkerDataArray<double>(max_gc_threads, "Termination (ms)", true, G1Log::LevelFiner, 2);
121 _gc_par_phases[GCWorkerTotal] = new WorkerDataArray<double>(max_gc_threads, "GC Worker Total (ms)", true, G1Log::LevelFiner, 2);
122 _gc_par_phases[GCWorkerEnd] = new WorkerDataArray<double>(max_gc_threads, "GC Worker End (ms)", false, G1Log::LevelFiner, 2);
123 _gc_par_phases[Other] = new WorkerDataArray<double>(max_gc_threads, "GC Worker Other (ms)", true, G1Log::LevelFiner, 2);
124
125 _update_rs_processed_buffers = new WorkerDataArray<size_t>(max_gc_threads, "Processed Buffers", true, G1Log::LevelFiner, 3);
126 _gc_par_phases[UpdateRS]->link_thread_work_items(_update_rs_processed_buffers);
127
128 _termination_attempts = new WorkerDataArray<size_t>(max_gc_threads, "Termination Attempts", true, G1Log::LevelFinest, 3);
129 _gc_par_phases[Termination]->link_thread_work_items(_termination_attempts);
130
131 _gc_par_phases[StringDedupQueueFixup] = new WorkerDataArray<double>(max_gc_threads, "Queue Fixup (ms)", true, G1Log::LevelFiner, 2);
132 _gc_par_phases[StringDedupTableFixup] = new WorkerDataArray<double>(max_gc_threads, "Table Fixup (ms)", true, G1Log::LevelFiner, 2);
133
134 _gc_par_phases[RedirtyCards] = new WorkerDataArray<double>(max_gc_threads, "Parallel Redirty", true, G1Log::LevelFinest, 3);
135 _redirtied_cards = new WorkerDataArray<size_t>(max_gc_threads, "Redirtied Cards", true, G1Log::LevelFinest, 3);
136 _gc_par_phases[RedirtyCards]->link_thread_work_items(_redirtied_cards);
137 }
138
139 void G1GCPhaseTimes::note_gc_start(uint active_gc_threads) {
140 assert(active_gc_threads > 0, "The number of threads must be > 0");
141 assert(active_gc_threads <= _max_gc_threads, "The number of active threads must be <= the max number of threads");
142 _active_gc_threads = active_gc_threads;
143 _cur_expand_heap_time_ms = 0.0;
144
145 for (int i = 0; i < GCParPhasesSentinel; i++) {
146 _gc_par_phases[i]->reset();
147 }
148
149 _gc_par_phases[StringDedupQueueFixup]->set_enabled(G1StringDedup::is_enabled());
150 _gc_par_phases[StringDedupTableFixup]->set_enabled(G1StringDedup::is_enabled());
151 }
152
153 void G1GCPhaseTimes::note_gc_end() {
154 for (uint i = 0; i < _active_gc_threads; i++) {
155 double worker_time = _gc_par_phases[GCWorkerEnd]->get(i) - _gc_par_phases[GCWorkerStart]->get(i);
156 record_time_secs(GCWorkerTotal, i , worker_time);
157
158 double worker_known_time =
159 _gc_par_phases[ExtRootScan]->get(i) +
160 _gc_par_phases[SATBFiltering]->get(i) +
161 _gc_par_phases[UpdateRS]->get(i) +
162 _gc_par_phases[ScanRS]->get(i) +
163 _gc_par_phases[CodeRoots]->get(i) +
164 _gc_par_phases[ObjCopy]->get(i) +
165 _gc_par_phases[Termination]->get(i);
166
167 record_time_secs(Other, i, worker_time - worker_known_time);
168 }
169
170 for (int i = 0; i < GCParPhasesSentinel; i++) {
171 _gc_par_phases[i]->verify(_active_gc_threads);
172 }
173 }
174
175 void G1GCPhaseTimes::print_stats(int level, const char* str, double value) {
176 LineBuffer(level).append_and_print_cr("[%s: %.1lf ms]", str, value);
177 }
178
179 void G1GCPhaseTimes::print_stats(int level, const char* str, size_t value) {
180 LineBuffer(level).append_and_print_cr("[%s: " SIZE_FORMAT "]", str, value);
181 }
182
183 void G1GCPhaseTimes::print_stats(int level, const char* str, double value, uint workers) {
184 LineBuffer(level).append_and_print_cr("[%s: %.1lf ms, GC Workers: %u]", str, value, workers);
185 }
186
187 double G1GCPhaseTimes::accounted_time_ms() {
188 // Subtract the root region scanning wait time. It's initialized to
189 // zero at the start of the pause.
190 double misc_time_ms = _root_region_scan_wait_time_ms;
191
192 misc_time_ms += _cur_collection_par_time_ms;
193
194 // Now subtract the time taken to fix up roots in generated code
195 misc_time_ms += _cur_collection_code_root_fixup_time_ms;
196
197 // Strong code root purge time
198 misc_time_ms += _cur_strong_code_root_purge_time_ms;
199
200 if (G1StringDedup::is_enabled()) {
201 // String dedup fixup time
202 misc_time_ms += _cur_string_dedup_fixup_time_ms;
203 }
204
205 // Subtract the time taken to clean the card table from the
206 // current value of "other time"
207 misc_time_ms += _cur_clear_ct_time_ms;
208
209 // Remove expand heap time from "other time"
210 misc_time_ms += _cur_expand_heap_time_ms;
211
212 return misc_time_ms;
213 }
214
215 // record the time a phase took in seconds
216 void G1GCPhaseTimes::record_time_secs(GCParPhases phase, uint worker_i, double secs) {
217 _gc_par_phases[phase]->set(worker_i, secs);
218 }
219
220 // add a number of seconds to a phase
221 void G1GCPhaseTimes::add_time_secs(GCParPhases phase, uint worker_i, double secs) {
222 _gc_par_phases[phase]->add(worker_i, secs);
223 }
224
225 void G1GCPhaseTimes::record_thread_work_item(GCParPhases phase, uint worker_i, size_t count) {
226 _gc_par_phases[phase]->set_thread_work_item(worker_i, count);
227 }
228
229 // return the average time for a phase in milliseconds
230 double G1GCPhaseTimes::average_time_ms(GCParPhases phase) {
231 return _gc_par_phases[phase]->average(_active_gc_threads) * 1000.0;
232 }
233
234 double G1GCPhaseTimes::get_time_ms(GCParPhases phase, uint worker_i) {
235 return _gc_par_phases[phase]->get(worker_i) * 1000.0;
236 }
237
238 double G1GCPhaseTimes::sum_time_ms(GCParPhases phase) {
239 return _gc_par_phases[phase]->sum(_active_gc_threads) * 1000.0;
240 }
241
242 double G1GCPhaseTimes::min_time_ms(GCParPhases phase) {
243 return _gc_par_phases[phase]->minimum(_active_gc_threads) * 1000.0;
244 }
245
246 double G1GCPhaseTimes::max_time_ms(GCParPhases phase) {
247 return _gc_par_phases[phase]->maximum(_active_gc_threads) * 1000.0;
248 }
249
250 size_t G1GCPhaseTimes::get_thread_work_item(GCParPhases phase, uint worker_i) {
251 assert(_gc_par_phases[phase]->thread_work_items() != NULL, "No sub count");
252 return _gc_par_phases[phase]->thread_work_items()->get(worker_i);
253 }
254
255 size_t G1GCPhaseTimes::sum_thread_work_items(GCParPhases phase) {
256 assert(_gc_par_phases[phase]->thread_work_items() != NULL, "No sub count");
257 return _gc_par_phases[phase]->thread_work_items()->sum(_active_gc_threads);
258 }
259
260 double G1GCPhaseTimes::average_thread_work_items(GCParPhases phase) {
261 assert(_gc_par_phases[phase]->thread_work_items() != NULL, "No sub count");
262 return _gc_par_phases[phase]->thread_work_items()->average(_active_gc_threads);
263 }
264
265 size_t G1GCPhaseTimes::min_thread_work_items(GCParPhases phase) {
266 assert(_gc_par_phases[phase]->thread_work_items() != NULL, "No sub count");
267 return _gc_par_phases[phase]->thread_work_items()->minimum(_active_gc_threads);
268 }
269
270 size_t G1GCPhaseTimes::max_thread_work_items(GCParPhases phase) {
271 assert(_gc_par_phases[phase]->thread_work_items() != NULL, "No sub count");
272 return _gc_par_phases[phase]->thread_work_items()->maximum(_active_gc_threads);
273 }
274
275 class G1GCParPhasePrinter : public StackObj {
276 G1GCPhaseTimes* _phase_times;
277 public:
278 G1GCParPhasePrinter(G1GCPhaseTimes* phase_times) : _phase_times(phase_times) {}
279
280 void print(G1GCPhaseTimes::GCParPhases phase_id) {
281 WorkerDataArray<double>* phase = _phase_times->_gc_par_phases[phase_id];
282
283 if (phase->_log_level > G1Log::level() || !phase->_enabled) {
284 return;
285 }
286
287 if (phase->_length == 1) {
288 print_single_length(phase_id, phase);
289 } else {
290 print_multi_length(phase_id, phase);
291 }
292 }
293
294 private:
295
296 void print_single_length(G1GCPhaseTimes::GCParPhases phase_id, WorkerDataArray<double>* phase) {
297 // No need for min, max, average and sum for only one worker
298 LineBuffer buf(phase->_indent_level);
299 buf.append_and_print_cr("[%s: %.1lf]", phase->_title, _phase_times->get_time_ms(phase_id, 0));
300
301 if (phase->_thread_work_items != NULL) {
302 LineBuffer buf2(phase->_thread_work_items->_indent_level);
303 buf2.append_and_print_cr("[%s: " SIZE_FORMAT "]", phase->_thread_work_items->_title, _phase_times->sum_thread_work_items(phase_id));
304 }
305 }
306
307 void print_time_values(LineBuffer& buf, G1GCPhaseTimes::GCParPhases phase_id, WorkerDataArray<double>* phase) {
308 uint active_length = _phase_times->_active_gc_threads;
309 for (uint i = 0; i < active_length; ++i) {
310 buf.append(" %.1lf", _phase_times->get_time_ms(phase_id, i));
311 }
312 buf.print_cr();
313 }
314
315 void print_count_values(LineBuffer& buf, G1GCPhaseTimes::GCParPhases phase_id, WorkerDataArray<size_t>* thread_work_items) {
316 uint active_length = _phase_times->_active_gc_threads;
317 for (uint i = 0; i < active_length; ++i) {
318 buf.append(" " SIZE_FORMAT, _phase_times->get_thread_work_item(phase_id, i));
319 }
320 buf.print_cr();
321 }
322
323 void print_thread_work_items(G1GCPhaseTimes::GCParPhases phase_id, WorkerDataArray<size_t>* thread_work_items) {
324 LineBuffer buf(thread_work_items->_indent_level);
325 buf.append("[%s:", thread_work_items->_title);
326
327 if (G1Log::finest()) {
328 print_count_values(buf, phase_id, thread_work_items);
329 }
330
331 assert(thread_work_items->_print_sum, "%s does not have print sum true even though it is a count", thread_work_items->_title);
332
333 buf.append_and_print_cr(" Min: " SIZE_FORMAT ", Avg: %.1lf, Max: " SIZE_FORMAT ", Diff: " SIZE_FORMAT ", Sum: " SIZE_FORMAT "]",
334 _phase_times->min_thread_work_items(phase_id), _phase_times->average_thread_work_items(phase_id), _phase_times->max_thread_work_items(phase_id),
335 _phase_times->max_thread_work_items(phase_id) - _phase_times->min_thread_work_items(phase_id), _phase_times->sum_thread_work_items(phase_id));
336 }
337
338 void print_multi_length(G1GCPhaseTimes::GCParPhases phase_id, WorkerDataArray<double>* phase) {
339 LineBuffer buf(phase->_indent_level);
340 buf.append("[%s:", phase->_title);
341
342 if (G1Log::finest()) {
343 print_time_values(buf, phase_id, phase);
344 }
345
346 buf.append(" Min: %.1lf, Avg: %.1lf, Max: %.1lf, Diff: %.1lf",
347 _phase_times->min_time_ms(phase_id), _phase_times->average_time_ms(phase_id), _phase_times->max_time_ms(phase_id),
348 _phase_times->max_time_ms(phase_id) - _phase_times->min_time_ms(phase_id));
349
350 if (phase->_print_sum) {
351 // for things like the start and end times the sum is not
352 // that relevant
353 buf.append(", Sum: %.1lf", _phase_times->sum_time_ms(phase_id));
354 }
355
356 buf.append_and_print_cr("]");
357
358 if (phase->_thread_work_items != NULL) {
359 print_thread_work_items(phase_id, phase->_thread_work_items);
360 }
361 }
362 };
363
364 void G1GCPhaseTimes::print(double pause_time_sec) {
365 note_gc_end();
366
367 G1GCParPhasePrinter par_phase_printer(this);
368
369 if (_root_region_scan_wait_time_ms > 0.0) {
370 print_stats(1, "Root Region Scan Waiting", _root_region_scan_wait_time_ms);
371 }
372
373 print_stats(1, "Parallel Time", _cur_collection_par_time_ms, _active_gc_threads);
374 for (int i = 0; i <= GCMainParPhasesLast; i++) {
375 par_phase_printer.print((GCParPhases) i);
376 }
377
378 print_stats(1, "Code Root Fixup", _cur_collection_code_root_fixup_time_ms);
379 print_stats(1, "Code Root Purge", _cur_strong_code_root_purge_time_ms);
380 if (G1StringDedup::is_enabled()) {
381 print_stats(1, "String Dedup Fixup", _cur_string_dedup_fixup_time_ms, _active_gc_threads);
382 for (int i = StringDedupPhasesFirst; i <= StringDedupPhasesLast; i++) {
383 par_phase_printer.print((GCParPhases) i);
384 }
385 }
386 print_stats(1, "Clear CT", _cur_clear_ct_time_ms);
387 print_stats(1, "Expand Heap After Collection", _cur_expand_heap_time_ms);
388
389 double misc_time_ms = pause_time_sec * MILLIUNITS - accounted_time_ms();
390 print_stats(1, "Other", misc_time_ms);
391 if (_cur_verify_before_time_ms > 0.0) {
392 print_stats(2, "Verify Before", _cur_verify_before_time_ms);
393 }
394 if (G1CollectedHeap::heap()->evacuation_failed()) {
395 double evac_fail_handling = _cur_evac_fail_recalc_used + _cur_evac_fail_remove_self_forwards +
396 _cur_evac_fail_restore_remsets;
397 print_stats(2, "Evacuation Failure", evac_fail_handling);
398 if (G1Log::finest()) {
399 print_stats(3, "Recalculate Used", _cur_evac_fail_recalc_used);
400 print_stats(3, "Remove Self Forwards", _cur_evac_fail_remove_self_forwards);
401 print_stats(3, "Restore RemSet", _cur_evac_fail_restore_remsets);
402 }
403 }
404 print_stats(2, "Choose CSet",
405 (_recorded_young_cset_choice_time_ms +
406 _recorded_non_young_cset_choice_time_ms));
407 print_stats(2, "Ref Proc", _cur_ref_proc_time_ms);
408 print_stats(2, "Ref Enq", _cur_ref_enq_time_ms);
409 print_stats(2, "Redirty Cards", _recorded_redirty_logged_cards_time_ms);
410 par_phase_printer.print(RedirtyCards);
411 if (G1EagerReclaimHumongousObjects) {
412 print_stats(2, "Humongous Register", _cur_fast_reclaim_humongous_register_time_ms);
413 if (G1Log::finest()) {
414 print_stats(3, "Humongous Total", _cur_fast_reclaim_humongous_total);
415 print_stats(3, "Humongous Candidate", _cur_fast_reclaim_humongous_candidates);
416 }
417 print_stats(2, "Humongous Reclaim", _cur_fast_reclaim_humongous_time_ms);
418 if (G1Log::finest()) {
419 print_stats(3, "Humongous Reclaimed", _cur_fast_reclaim_humongous_reclaimed);
420 }
421 }
422 print_stats(2, "Free CSet",
423 (_recorded_young_free_cset_time_ms +
424 _recorded_non_young_free_cset_time_ms));
425 if (G1Log::finest()) {
426 print_stats(3, "Young Free CSet", _recorded_young_free_cset_time_ms);
427 print_stats(3, "Non-Young Free CSet", _recorded_non_young_free_cset_time_ms);
428 }
429 if (_cur_verify_after_time_ms > 0.0) {
430 print_stats(2, "Verify After", _cur_verify_after_time_ms);
431 }
432 }
433
434 G1GCParPhaseTimesTracker::G1GCParPhaseTimesTracker(G1GCPhaseTimes* phase_times, G1GCPhaseTimes::GCParPhases phase, uint worker_id) :
435 _phase_times(phase_times), _phase(phase), _worker_id(worker_id) {
436 if (_phase_times != NULL) {
437 _start_time = os::elapsedTime();
438 }
439 }
440
441 G1GCParPhaseTimesTracker::~G1GCParPhaseTimesTracker() {
442 if (_phase_times != NULL) {
443 _phase_times->record_time_secs(_phase, _worker_id, os::elapsedTime() - _start_time);
444 }
445 }
446