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/concurrentG1Refine.hpp"
27 #include "gc/g1/g1CollectedHeap.inline.hpp"
28 #include "gc/g1/g1GCPhaseTimes.hpp"
29 #include "gc/g1/g1StringDedup.hpp"
30 #include "gc/g1/workerDataArray.inline.hpp"
31 #include "memory/allocation.hpp"
32 #include "logging/log.hpp"
33 #include "runtime/timer.hpp"
34 #include "runtime/os.hpp"
35
36 // Helper class for avoiding interleaved logging
37 class LineBuffer: public StackObj {
38
39 private:
40 static const int BUFFER_LEN = 1024;
41 static const int INDENT_CHARS = 3;
42 char _buffer[BUFFER_LEN];
43 int _indent_level;
44 int _cur;
45
46 void vappend(const char* format, va_list ap) ATTRIBUTE_PRINTF(2, 0) {
47 int res = vsnprintf(&_buffer[_cur], BUFFER_LEN - _cur, format, ap);
48 if (res != -1) {
49 _cur += res;
50 } else {
51 DEBUG_ONLY(warning("buffer too small in LineBuffer");)
52 _buffer[BUFFER_LEN -1] = 0;
53 _cur = BUFFER_LEN; // vsnprintf above should not add to _buffer if we are called again
54 }
55 }
56
57 public:
58 explicit LineBuffer(int indent_level): _indent_level(indent_level), _cur(0) {
59 for (; (_cur < BUFFER_LEN && _cur < (_indent_level * INDENT_CHARS)); _cur++) {
60 _buffer[_cur] = ' ';
61 }
62 }
63
64 #ifndef PRODUCT
65 ~LineBuffer() {
66 assert(_cur == _indent_level * INDENT_CHARS, "pending data in buffer - append_and_print_cr() not called?");
67 }
68 #endif
69
70 void append(const char* format, ...) ATTRIBUTE_PRINTF(2, 3) {
71 va_list ap;
72 va_start(ap, format);
73 vappend(format, ap);
74 va_end(ap);
75 }
76
77 const char* to_string() {
78 _cur = _indent_level * INDENT_CHARS;
79 return _buffer;
80 }
81 };
82
83 static const char* Indents[4] = {"", " ", " ", " "};
84
85 G1GCPhaseTimes::G1GCPhaseTimes(uint max_gc_threads) :
86 _max_gc_threads(max_gc_threads)
87 {
88 assert(max_gc_threads > 0, "Must have some GC threads");
89
90 _gc_par_phases[GCWorkerStart] = new WorkerDataArray<double>(max_gc_threads, "GC Worker Start:", false, 2);
91 _gc_par_phases[ExtRootScan] = new WorkerDataArray<double>(max_gc_threads, "Ext Root Scanning:", true, 2);
92
93 // Root scanning phases
94 _gc_par_phases[ThreadRoots] = new WorkerDataArray<double>(max_gc_threads, "Thread Roots:", true, 3);
95 _gc_par_phases[StringTableRoots] = new WorkerDataArray<double>(max_gc_threads, "StringTable Roots:", true, 3);
96 _gc_par_phases[UniverseRoots] = new WorkerDataArray<double>(max_gc_threads, "Universe Roots:", true, 3);
97 _gc_par_phases[JNIRoots] = new WorkerDataArray<double>(max_gc_threads, "JNI Handles Roots:", true, 3);
98 _gc_par_phases[ObjectSynchronizerRoots] = new WorkerDataArray<double>(max_gc_threads, "ObjectSynchronizer Roots:", true, 3);
99 _gc_par_phases[FlatProfilerRoots] = new WorkerDataArray<double>(max_gc_threads, "FlatProfiler Roots:", true, 3);
100 _gc_par_phases[ManagementRoots] = new WorkerDataArray<double>(max_gc_threads, "Management Roots:", true, 3);
101 _gc_par_phases[SystemDictionaryRoots] = new WorkerDataArray<double>(max_gc_threads, "SystemDictionary Roots:", true, 3);
102 _gc_par_phases[CLDGRoots] = new WorkerDataArray<double>(max_gc_threads, "CLDG Roots:", true, 3);
103 _gc_par_phases[JVMTIRoots] = new WorkerDataArray<double>(max_gc_threads, "JVMTI Roots:", true, 3);
104 _gc_par_phases[CMRefRoots] = new WorkerDataArray<double>(max_gc_threads, "CM RefProcessor Roots:", true, 3);
105 _gc_par_phases[WaitForStrongCLD] = new WorkerDataArray<double>(max_gc_threads, "Wait For Strong CLD:", true, 3);
106 _gc_par_phases[WeakCLDRoots] = new WorkerDataArray<double>(max_gc_threads, "Weak CLD Roots:", true, 3);
107 _gc_par_phases[SATBFiltering] = new WorkerDataArray<double>(max_gc_threads, "SATB Filtering:", true, 3);
108
109 _gc_par_phases[UpdateRS] = new WorkerDataArray<double>(max_gc_threads, "Update RS:", true, 2);
110 _gc_par_phases[ScanHCC] = new WorkerDataArray<double>(max_gc_threads, "Scan HCC:", true, 3);
111 _gc_par_phases[ScanHCC]->set_enabled(ConcurrentG1Refine::hot_card_cache_enabled());
112 _gc_par_phases[ScanRS] = new WorkerDataArray<double>(max_gc_threads, "Scan RS:", true, 2);
113 _gc_par_phases[CodeRoots] = new WorkerDataArray<double>(max_gc_threads, "Code Root Scanning:", true, 2);
114 _gc_par_phases[ObjCopy] = new WorkerDataArray<double>(max_gc_threads, "Object Copy:", true, 2);
115 _gc_par_phases[Termination] = new WorkerDataArray<double>(max_gc_threads, "Termination:", true, 2);
116 _gc_par_phases[GCWorkerTotal] = new WorkerDataArray<double>(max_gc_threads, "GC Worker Total:", true, 2);
117 _gc_par_phases[GCWorkerEnd] = new WorkerDataArray<double>(max_gc_threads, "GC Worker End:", false, 2);
118 _gc_par_phases[Other] = new WorkerDataArray<double>(max_gc_threads, "GC Worker Other:", true, 2);
119
120 _update_rs_processed_buffers = new WorkerDataArray<size_t>(max_gc_threads, "Processed Buffers:", true, 3);
121 _gc_par_phases[UpdateRS]->link_thread_work_items(_update_rs_processed_buffers);
122
123 _termination_attempts = new WorkerDataArray<size_t>(max_gc_threads, "Termination Attempts:", true, 3);
124 _gc_par_phases[Termination]->link_thread_work_items(_termination_attempts);
125
126 _gc_par_phases[StringDedupQueueFixup] = new WorkerDataArray<double>(max_gc_threads, "Queue Fixup:", true, 2);
127 _gc_par_phases[StringDedupTableFixup] = new WorkerDataArray<double>(max_gc_threads, "Table Fixup:", true, 2);
128
129 _gc_par_phases[RedirtyCards] = new WorkerDataArray<double>(max_gc_threads, "Parallel Redirty:", true, 3);
130 _redirtied_cards = new WorkerDataArray<size_t>(max_gc_threads, "Redirtied Cards:", true, 3);
131 _gc_par_phases[RedirtyCards]->link_thread_work_items(_redirtied_cards);
132
133 _gc_par_phases[PreserveCMReferents] = new WorkerDataArray<double>(max_gc_threads, "Parallel Preserve CM Refs:", true, 3);
134 }
135
136 void G1GCPhaseTimes::note_gc_start(uint active_gc_threads) {
137 assert(active_gc_threads > 0, "The number of threads must be > 0");
138 assert(active_gc_threads <= _max_gc_threads, "The number of active threads must be <= the max number of threads");
139 _gc_start_counter = os::elapsed_counter();
140 _active_gc_threads = active_gc_threads;
141 _cur_expand_heap_time_ms = 0.0;
142 _external_accounted_time_ms = 0.0;
143
144 for (int i = 0; i < GCParPhasesSentinel; i++) {
145 _gc_par_phases[i]->reset();
146 }
147
148 _gc_par_phases[StringDedupQueueFixup]->set_enabled(G1StringDedup::is_enabled());
149 _gc_par_phases[StringDedupTableFixup]->set_enabled(G1StringDedup::is_enabled());
150 }
151
152 void G1GCPhaseTimes::note_gc_end() {
153 _gc_pause_time_ms = TimeHelper::counter_to_millis(os::elapsed_counter() - _gc_start_counter);
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(const char* indent, const char* str, double value) {
176 log_debug(gc, phases)("%s%s: %.1lf ms", indent, str, value);
177 }
178
179 double G1GCPhaseTimes::accounted_time_ms() {
180 // First subtract any externally accounted time
181 double misc_time_ms = _external_accounted_time_ms;
182
183 // Subtract the root region scanning wait time. It's initialized to
184 // zero at the start of the pause.
185 misc_time_ms += _root_region_scan_wait_time_ms;
186
187 misc_time_ms += _cur_collection_par_time_ms;
188
189 // Now subtract the time taken to fix up roots in generated code
190 misc_time_ms += _cur_collection_code_root_fixup_time_ms;
191
192 // Strong code root purge time
193 misc_time_ms += _cur_strong_code_root_purge_time_ms;
194
195 if (G1StringDedup::is_enabled()) {
196 // String dedup fixup time
197 misc_time_ms += _cur_string_dedup_fixup_time_ms;
198 }
199
200 // Subtract the time taken to clean the card table from the
201 // current value of "other time"
202 misc_time_ms += _cur_clear_ct_time_ms;
203
204 // Remove expand heap time from "other time"
205 misc_time_ms += _cur_expand_heap_time_ms;
206
207 return misc_time_ms;
208 }
209
210 // record the time a phase took in seconds
211 void G1GCPhaseTimes::record_time_secs(GCParPhases phase, uint worker_i, double secs) {
212 _gc_par_phases[phase]->set(worker_i, secs);
213 }
214
215 // add a number of seconds to a phase
216 void G1GCPhaseTimes::add_time_secs(GCParPhases phase, uint worker_i, double secs) {
217 _gc_par_phases[phase]->add(worker_i, secs);
218 }
219
220 void G1GCPhaseTimes::record_thread_work_item(GCParPhases phase, uint worker_i, size_t count) {
221 _gc_par_phases[phase]->set_thread_work_item(worker_i, count);
222 }
223
224 // return the average time for a phase in milliseconds
225 double G1GCPhaseTimes::average_time_ms(GCParPhases phase) {
226 return _gc_par_phases[phase]->average(_active_gc_threads) * 1000.0;
227 }
228
229 double G1GCPhaseTimes::get_time_ms(GCParPhases phase, uint worker_i) {
230 return _gc_par_phases[phase]->get(worker_i) * 1000.0;
231 }
232
233 double G1GCPhaseTimes::sum_time_ms(GCParPhases phase) {
234 return _gc_par_phases[phase]->sum(_active_gc_threads) * 1000.0;
235 }
236
237 double G1GCPhaseTimes::min_time_ms(GCParPhases phase) {
238 return _gc_par_phases[phase]->minimum(_active_gc_threads) * 1000.0;
239 }
240
241 double G1GCPhaseTimes::max_time_ms(GCParPhases phase) {
242 return _gc_par_phases[phase]->maximum(_active_gc_threads) * 1000.0;
243 }
244
245 size_t G1GCPhaseTimes::get_thread_work_item(GCParPhases phase, uint worker_i) {
246 assert(_gc_par_phases[phase]->thread_work_items() != NULL, "No sub count");
247 return _gc_par_phases[phase]->thread_work_items()->get(worker_i);
248 }
249
250 size_t G1GCPhaseTimes::sum_thread_work_items(GCParPhases phase) {
251 assert(_gc_par_phases[phase]->thread_work_items() != NULL, "No sub count");
252 return _gc_par_phases[phase]->thread_work_items()->sum(_active_gc_threads);
253 }
254
255 double G1GCPhaseTimes::average_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()->average(_active_gc_threads);
258 }
259
260 size_t G1GCPhaseTimes::min_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()->minimum(_active_gc_threads);
263 }
264
265 size_t G1GCPhaseTimes::max_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()->maximum(_active_gc_threads);
268 }
269
270 class G1GCParPhasePrinter : public StackObj {
271 G1GCPhaseTimes* _phase_times;
272 public:
273 G1GCParPhasePrinter(G1GCPhaseTimes* phase_times) : _phase_times(phase_times) {}
274
275 void print(G1GCPhaseTimes::GCParPhases phase_id) {
276 WorkerDataArray<double>* phase = _phase_times->_gc_par_phases[phase_id];
277
278 if (phase->_length == 1) {
279 print_single_length(phase_id, phase);
280 } else {
281 print_multi_length(phase_id, phase);
282 }
283 }
284
285
286 private:
287 void print_single_length(G1GCPhaseTimes::GCParPhases phase_id, WorkerDataArray<double>* phase) {
288 // No need for min, max, average and sum for only one worker
289 log_debug(gc, phases)("%s%s: %.1lf", Indents[phase->_indent_level], phase->_title, _phase_times->get_time_ms(phase_id, 0));
290
291 WorkerDataArray<size_t>* work_items = phase->_thread_work_items;
292 if (work_items != NULL) {
293 log_debug(gc, phases)("%s%s: " SIZE_FORMAT, Indents[work_items->_indent_level], work_items->_title, _phase_times->sum_thread_work_items(phase_id));
294 }
295 }
296
297 void print_time_values(const char* indent, G1GCPhaseTimes::GCParPhases phase_id) {
298 if (log_is_enabled(Trace, gc)) {
299 LineBuffer buf(0);
300 uint active_length = _phase_times->_active_gc_threads;
301 for (uint i = 0; i < active_length; ++i) {
302 buf.append(" %4.1lf", _phase_times->get_time_ms(phase_id, i));
303 }
304 const char* line = buf.to_string();
305 log_trace(gc, phases)("%s%-25s%s", indent, "", line);
306 }
307 }
308
309 void print_count_values(const char* indent, G1GCPhaseTimes::GCParPhases phase_id, WorkerDataArray<size_t>* thread_work_items) {
310 if (log_is_enabled(Trace, gc)) {
311 LineBuffer buf(0);
312 uint active_length = _phase_times->_active_gc_threads;
313 for (uint i = 0; i < active_length; ++i) {
314 buf.append(" " SIZE_FORMAT, _phase_times->get_thread_work_item(phase_id, i));
315 }
316 const char* line = buf.to_string();
317 log_trace(gc, phases)("%s%-25s%s", indent, "", line);
318 }
319 }
320
321 void print_thread_work_items(G1GCPhaseTimes::GCParPhases phase_id, WorkerDataArray<size_t>* thread_work_items) {
322 const char* indent = Indents[thread_work_items->_indent_level];
323
324 assert(thread_work_items->_print_sum, "%s does not have print sum true even though it is a count", thread_work_items->_title);
325
326 log_debug(gc, phases)("%s%-25s Min: " SIZE_FORMAT ", Avg: %4.1lf, Max: " SIZE_FORMAT ", Diff: " SIZE_FORMAT ", Sum: " SIZE_FORMAT,
327 indent, thread_work_items->_title,
328 _phase_times->min_thread_work_items(phase_id), _phase_times->average_thread_work_items(phase_id), _phase_times->max_thread_work_items(phase_id),
329 _phase_times->max_thread_work_items(phase_id) - _phase_times->min_thread_work_items(phase_id), _phase_times->sum_thread_work_items(phase_id));
330
331 print_count_values(indent, phase_id, thread_work_items);
332 }
333
334 void print_multi_length(G1GCPhaseTimes::GCParPhases phase_id, WorkerDataArray<double>* phase) {
335 const char* indent = Indents[phase->_indent_level];
336
337 if (phase->_print_sum) {
338 log_debug(gc, phases)("%s%-25s Min: %4.1lf, Avg: %4.1lf, Max: %4.1lf, Diff: %4.1lf, Sum: %4.1lf",
339 indent, phase->_title,
340 _phase_times->min_time_ms(phase_id), _phase_times->average_time_ms(phase_id), _phase_times->max_time_ms(phase_id),
341 _phase_times->max_time_ms(phase_id) - _phase_times->min_time_ms(phase_id), _phase_times->sum_time_ms(phase_id));
342 } else {
343 log_debug(gc, phases)("%s%-25s Min: %4.1lf, Avg: %4.1lf, Max: %4.1lf, Diff: %4.1lf",
344 indent, phase->_title,
345 _phase_times->min_time_ms(phase_id), _phase_times->average_time_ms(phase_id), _phase_times->max_time_ms(phase_id),
346 _phase_times->max_time_ms(phase_id) - _phase_times->min_time_ms(phase_id));
347 }
348
349 print_time_values(indent, phase_id);
350
351 if (phase->_thread_work_items != NULL) {
352 print_thread_work_items(phase_id, phase->_thread_work_items);
353 }
354 }
355 };
356
357 void G1GCPhaseTimes::print() {
358 note_gc_end();
359
360 G1GCParPhasePrinter par_phase_printer(this);
361
362 if (_root_region_scan_wait_time_ms > 0.0) {
363 print_stats(Indents[1], "Root Region Scan Waiting", _root_region_scan_wait_time_ms);
364 }
365
366 print_stats(Indents[1], "Parallel Time", _cur_collection_par_time_ms);
367 for (int i = 0; i <= GCMainParPhasesLast; i++) {
368 par_phase_printer.print((GCParPhases) i);
369 }
370
371 print_stats(Indents[1], "Code Root Fixup", _cur_collection_code_root_fixup_time_ms);
372 print_stats(Indents[1], "Code Root Purge", _cur_strong_code_root_purge_time_ms);
373 if (G1StringDedup::is_enabled()) {
374 print_stats(Indents[1], "String Dedup Fixup", _cur_string_dedup_fixup_time_ms);
375 for (int i = StringDedupPhasesFirst; i <= StringDedupPhasesLast; i++) {
376 par_phase_printer.print((GCParPhases) i);
377 }
378 }
379 print_stats(Indents[1], "Clear CT", _cur_clear_ct_time_ms);
380 print_stats(Indents[1], "Expand Heap After Collection", _cur_expand_heap_time_ms);
381 double misc_time_ms = _gc_pause_time_ms - accounted_time_ms();
382 print_stats(Indents[1], "Other", misc_time_ms);
383 if (_cur_verify_before_time_ms > 0.0) {
384 print_stats(Indents[2], "Verify Before", _cur_verify_before_time_ms);
385 }
386 if (G1CollectedHeap::heap()->evacuation_failed()) {
387 double evac_fail_handling = _cur_evac_fail_recalc_used + _cur_evac_fail_remove_self_forwards +
388 _cur_evac_fail_restore_remsets;
389 print_stats(Indents[2], "Evacuation Failure", evac_fail_handling);
390 log_trace(gc, phases)("%sRecalculate Used: %.1lf ms", Indents[3], _cur_evac_fail_recalc_used);
391 log_trace(gc, phases)("%sRemove Self Forwards: %.1lf ms", Indents[3], _cur_evac_fail_remove_self_forwards);
392 log_trace(gc, phases)("%sRestore RemSet: %.1lf ms", Indents[3], _cur_evac_fail_restore_remsets);
393 }
394 print_stats(Indents[2], "Choose CSet",
395 (_recorded_young_cset_choice_time_ms +
396 _recorded_non_young_cset_choice_time_ms));
397 print_stats(Indents[2], "Preserve CM Refs", _recorded_preserve_cm_referents_time_ms);
398 print_stats(Indents[2], "Ref Proc", _cur_ref_proc_time_ms);
399 print_stats(Indents[2], "Ref Enq", _cur_ref_enq_time_ms);
400 print_stats(Indents[2], "Redirty Cards", _recorded_redirty_logged_cards_time_ms);
401 par_phase_printer.print(RedirtyCards);
402 par_phase_printer.print(PreserveCMReferents);
403 if (G1EagerReclaimHumongousObjects) {
404 print_stats(Indents[2], "Humongous Register", _cur_fast_reclaim_humongous_register_time_ms);
405
406 log_trace(gc, phases)("%sHumongous Total: " SIZE_FORMAT, Indents[3], _cur_fast_reclaim_humongous_total);
407 log_trace(gc, phases)("%sHumongous Candidate: " SIZE_FORMAT, Indents[3], _cur_fast_reclaim_humongous_candidates);
408 print_stats(Indents[2], "Humongous Reclaim", _cur_fast_reclaim_humongous_time_ms);
409 log_trace(gc, phases)("%sHumongous Reclaimed: " SIZE_FORMAT, Indents[3], _cur_fast_reclaim_humongous_reclaimed);
410 }
411 print_stats(Indents[2], "Free CSet",
412 (_recorded_young_free_cset_time_ms +
413 _recorded_non_young_free_cset_time_ms));
414 log_trace(gc, phases)("%sYoung Free CSet: %.1lf ms", Indents[3], _recorded_young_free_cset_time_ms);
415 log_trace(gc, phases)("%sNon-Young Free CSet: %.1lf ms", Indents[3], _recorded_non_young_free_cset_time_ms);
416 print_stats(Indents[2], "Merge Per-Thread State", _recorded_merge_pss_time_ms);
417 if (_cur_verify_after_time_ms > 0.0) {
418 print_stats(Indents[2], "Verify After", _cur_verify_after_time_ms);
419 }
420 }
421
422 G1GCParPhaseTimesTracker::G1GCParPhaseTimesTracker(G1GCPhaseTimes* phase_times, G1GCPhaseTimes::GCParPhases phase, uint worker_id) :
423 _phase_times(phase_times), _phase(phase), _worker_id(worker_id) {
424 if (_phase_times != NULL) {
425 _start_time = os::elapsedTime();
426 }
427 }
428
429 G1GCParPhaseTimesTracker::~G1GCParPhaseTimesTracker() {
430 if (_phase_times != NULL) {
431 _phase_times->record_time_secs(_phase, _worker_id, os::elapsedTime() - _start_time);
432 }
433 }
434