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/g1CollectedHeap.inline.hpp"
27 #include "gc/g1/g1GCPhaseTimes.hpp"
28 #include "gc/g1/g1HotCardCache.hpp"
29 #include "gc/g1/g1StringDedup.hpp"
30 #include "gc/g1/workerDataArray.inline.hpp"
31 #include "memory/resourceArea.hpp"
32 #include "logging/log.hpp"
33 #include "runtime/timer.hpp"
34 #include "runtime/os.hpp"
35
36 static const char* Indents[5] = {"", " ", " ", " ", " "};
37
38 G1GCPhaseTimes::G1GCPhaseTimes(uint max_gc_threads) :
39 _max_gc_threads(max_gc_threads),
40 _gc_start_counter(0),
41 _gc_pause_time_ms(0.0)
42 {
43 assert(max_gc_threads > 0, "Must have some GC threads");
44
45 _gc_par_phases[GCWorkerStart] = new WorkerDataArray<double>(max_gc_threads, "GC Worker Start (ms):");
46 _gc_par_phases[ExtRootScan] = new WorkerDataArray<double>(max_gc_threads, "Ext Root Scanning (ms):");
47
48 // Root scanning phases
49 _gc_par_phases[ThreadRoots] = new WorkerDataArray<double>(max_gc_threads, "Thread Roots (ms):");
50 _gc_par_phases[StringTableRoots] = new WorkerDataArray<double>(max_gc_threads, "StringTable Roots (ms):");
51 _gc_par_phases[UniverseRoots] = new WorkerDataArray<double>(max_gc_threads, "Universe Roots (ms):");
52 _gc_par_phases[JNIRoots] = new WorkerDataArray<double>(max_gc_threads, "JNI Handles Roots (ms):");
53 _gc_par_phases[ObjectSynchronizerRoots] = new WorkerDataArray<double>(max_gc_threads, "ObjectSynchronizer Roots (ms):");
54 _gc_par_phases[FlatProfilerRoots] = new WorkerDataArray<double>(max_gc_threads, "FlatProfiler Roots (ms):");
55 _gc_par_phases[ManagementRoots] = new WorkerDataArray<double>(max_gc_threads, "Management Roots (ms):");
56 _gc_par_phases[SystemDictionaryRoots] = new WorkerDataArray<double>(max_gc_threads, "SystemDictionary Roots (ms):");
57 _gc_par_phases[CLDGRoots] = new WorkerDataArray<double>(max_gc_threads, "CLDG Roots (ms):");
58 _gc_par_phases[JVMTIRoots] = new WorkerDataArray<double>(max_gc_threads, "JVMTI Roots (ms):");
59 _gc_par_phases[CMRefRoots] = new WorkerDataArray<double>(max_gc_threads, "CM RefProcessor Roots (ms):");
60 _gc_par_phases[WaitForStrongCLD] = new WorkerDataArray<double>(max_gc_threads, "Wait For Strong CLD (ms):");
61 _gc_par_phases[WeakCLDRoots] = new WorkerDataArray<double>(max_gc_threads, "Weak CLD Roots (ms):");
62 _gc_par_phases[SATBFiltering] = new WorkerDataArray<double>(max_gc_threads, "SATB Filtering (ms):");
63
64 _gc_par_phases[UpdateRS] = new WorkerDataArray<double>(max_gc_threads, "Update RS (ms):");
65 if (G1HotCardCache::default_use_cache()) {
66 _gc_par_phases[ScanHCC] = new WorkerDataArray<double>(max_gc_threads, "Scan HCC (ms):");
67 } else {
68 _gc_par_phases[ScanHCC] = NULL;
69 }
70 _gc_par_phases[ScanRS] = new WorkerDataArray<double>(max_gc_threads, "Scan RS (ms):");
71 _gc_par_phases[CodeRoots] = new WorkerDataArray<double>(max_gc_threads, "Code Root Scanning (ms):");
72 #if INCLUDE_AOT
73 _gc_par_phases[AOTCodeRoots] = new WorkerDataArray<double>(max_gc_threads, "AOT Root Scanning (ms):");
74 #endif
75 _gc_par_phases[ObjCopy] = new WorkerDataArray<double>(max_gc_threads, "Object Copy (ms):");
76 _gc_par_phases[Termination] = new WorkerDataArray<double>(max_gc_threads, "Termination (ms):");
77 _gc_par_phases[GCWorkerTotal] = new WorkerDataArray<double>(max_gc_threads, "GC Worker Total (ms):");
78 _gc_par_phases[GCWorkerEnd] = new WorkerDataArray<double>(max_gc_threads, "GC Worker End (ms):");
79 _gc_par_phases[Other] = new WorkerDataArray<double>(max_gc_threads, "GC Worker Other (ms):");
80
81 _update_rs_processed_buffers = new WorkerDataArray<size_t>(max_gc_threads, "Processed Buffers:");
82 _gc_par_phases[UpdateRS]->link_thread_work_items(_update_rs_processed_buffers);
83
84 _termination_attempts = new WorkerDataArray<size_t>(max_gc_threads, "Termination Attempts:");
85 _gc_par_phases[Termination]->link_thread_work_items(_termination_attempts);
86
87 if (UseStringDeduplication) {
88 _gc_par_phases[StringDedupQueueFixup] = new WorkerDataArray<double>(max_gc_threads, "Queue Fixup (ms):");
89 _gc_par_phases[StringDedupTableFixup] = new WorkerDataArray<double>(max_gc_threads, "Table Fixup (ms):");
90 } else {
91 _gc_par_phases[StringDedupQueueFixup] = NULL;
92 _gc_par_phases[StringDedupTableFixup] = NULL;
93 }
94
95 _gc_par_phases[RedirtyCards] = new WorkerDataArray<double>(max_gc_threads, "Parallel Redirty (ms):");
96 _redirtied_cards = new WorkerDataArray<size_t>(max_gc_threads, "Redirtied Cards:");
97 _gc_par_phases[RedirtyCards]->link_thread_work_items(_redirtied_cards);
98
99 _gc_par_phases[YoungFreeCSet] = new WorkerDataArray<double>(max_gc_threads, "Young Free Collection Set (ms):");
100 _gc_par_phases[NonYoungFreeCSet] = new WorkerDataArray<double>(max_gc_threads, "Non-Young Free Collection Set (ms):");
101
102 _gc_par_phases[PreserveCMReferents] = new WorkerDataArray<double>(max_gc_threads, "Parallel Preserve CM Refs (ms):");
103
104 reset();
105 }
106
107 void G1GCPhaseTimes::reset() {
108 _cur_collection_par_time_ms = 0.0;
109 _cur_collection_code_root_fixup_time_ms = 0.0;
110 _cur_strong_code_root_purge_time_ms = 0.0;
111 _cur_evac_fail_recalc_used = 0.0;
112 _cur_evac_fail_restore_remsets = 0.0;
113 _cur_evac_fail_remove_self_forwards = 0.0;
114 _cur_string_dedup_fixup_time_ms = 0.0;
115 _cur_clear_ct_time_ms = 0.0;
116 _cur_expand_heap_time_ms = 0.0;
117 _cur_ref_proc_time_ms = 0.0;
118 _cur_ref_enq_time_ms = 0.0;
119 _cur_collection_start_sec = 0.0;
120 _root_region_scan_wait_time_ms = 0.0;
121 _external_accounted_time_ms = 0.0;
122 _recorded_clear_claimed_marks_time_ms = 0.0;
123 _recorded_young_cset_choice_time_ms = 0.0;
124 _recorded_non_young_cset_choice_time_ms = 0.0;
125 _recorded_redirty_logged_cards_time_ms = 0.0;
126 _recorded_preserve_cm_referents_time_ms = 0.0;
127 _recorded_merge_pss_time_ms = 0.0;
128 _recorded_total_free_cset_time_ms = 0.0;
129 _recorded_serial_free_cset_time_ms = 0.0;
130 _cur_fast_reclaim_humongous_time_ms = 0.0;
131 _cur_fast_reclaim_humongous_register_time_ms = 0.0;
132 _cur_fast_reclaim_humongous_total = 0;
133 _cur_fast_reclaim_humongous_candidates = 0;
134 _cur_fast_reclaim_humongous_reclaimed = 0;
135 _cur_verify_before_time_ms = 0.0;
136 _cur_verify_after_time_ms = 0.0;
137
138 for (int i = 0; i < GCParPhasesSentinel; i++) {
139 if (_gc_par_phases[i] != NULL) {
140 _gc_par_phases[i]->reset();
141 }
142 }
143 }
144
145 void G1GCPhaseTimes::note_gc_start() {
146 _gc_start_counter = os::elapsed_counter();
147 reset();
148 }
149
150 #define ASSERT_PHASE_UNINITIALIZED(phase) \
151 assert(_gc_par_phases[phase]->get(i) == uninitialized, "Phase " #phase " reported for thread that was not started");
152
153 double G1GCPhaseTimes::worker_time(GCParPhases phase, uint worker) {
154 double value = _gc_par_phases[phase]->get(worker);
155 if (value != WorkerDataArray<double>::uninitialized()) {
156 return value;
157 }
158 return 0.0;
159 }
160
161 void G1GCPhaseTimes::note_gc_end() {
162 _gc_pause_time_ms = TimeHelper::counter_to_millis(os::elapsed_counter() - _gc_start_counter);
163
164 double uninitialized = WorkerDataArray<double>::uninitialized();
165
166 for (uint i = 0; i < _max_gc_threads; i++) {
167 double worker_start = _gc_par_phases[GCWorkerStart]->get(i);
168 if (worker_start != uninitialized) {
169 assert(_gc_par_phases[GCWorkerEnd]->get(i) != uninitialized, "Worker started but not ended.");
170 double total_worker_time = _gc_par_phases[GCWorkerEnd]->get(i) - _gc_par_phases[GCWorkerStart]->get(i);
171 record_time_secs(GCWorkerTotal, i , total_worker_time);
172
173 double worker_known_time =
174 worker_time(ExtRootScan, i)
175 + worker_time(SATBFiltering, i)
176 + worker_time(UpdateRS, i)
177 + worker_time(ScanRS, i)
178 + worker_time(CodeRoots, i)
179 + worker_time(ObjCopy, i)
180 + worker_time(Termination, i);
181
182 record_time_secs(Other, i, total_worker_time - worker_known_time);
183 } else {
184 // Make sure all slots are uninitialized since this thread did not seem to have been started
185 ASSERT_PHASE_UNINITIALIZED(GCWorkerEnd);
186 ASSERT_PHASE_UNINITIALIZED(ExtRootScan);
187 ASSERT_PHASE_UNINITIALIZED(SATBFiltering);
188 ASSERT_PHASE_UNINITIALIZED(UpdateRS);
189 ASSERT_PHASE_UNINITIALIZED(ScanRS);
190 ASSERT_PHASE_UNINITIALIZED(CodeRoots);
191 ASSERT_PHASE_UNINITIALIZED(ObjCopy);
192 ASSERT_PHASE_UNINITIALIZED(Termination);
193 }
194 }
195 }
196
197 #undef ASSERT_PHASE_UNINITIALIZED
198
199 // record the time a phase took in seconds
200 void G1GCPhaseTimes::record_time_secs(GCParPhases phase, uint worker_i, double secs) {
201 _gc_par_phases[phase]->set(worker_i, secs);
202 }
203
204 // add a number of seconds to a phase
205 void G1GCPhaseTimes::add_time_secs(GCParPhases phase, uint worker_i, double secs) {
206 _gc_par_phases[phase]->add(worker_i, secs);
207 }
208
209 void G1GCPhaseTimes::record_thread_work_item(GCParPhases phase, uint worker_i, size_t count) {
210 _gc_par_phases[phase]->set_thread_work_item(worker_i, count);
211 }
212
213 // return the average time for a phase in milliseconds
214 double G1GCPhaseTimes::average_time_ms(GCParPhases phase) {
215 return _gc_par_phases[phase]->average() * 1000.0;
216 }
217
218 size_t G1GCPhaseTimes::sum_thread_work_items(GCParPhases phase) {
219 assert(_gc_par_phases[phase]->thread_work_items() != NULL, "No sub count");
220 return _gc_par_phases[phase]->thread_work_items()->sum();
221 }
222
223 template <class T>
224 void G1GCPhaseTimes::details(T* phase, const char* indent) const {
225 Log(gc, phases, task) log;
226 if (log.is_level(LogLevel::Trace)) {
227 outputStream* trace_out = log.trace_stream();
228 trace_out->print("%s", indent);
229 phase->print_details_on(trace_out);
230 }
231 }
232
233 void G1GCPhaseTimes::log_phase(WorkerDataArray<double>* phase, uint indent, outputStream* out, bool print_sum) const {
234 out->print("%s", Indents[indent]);
235 phase->print_summary_on(out, print_sum);
236 details(phase, Indents[indent]);
237
238 WorkerDataArray<size_t>* work_items = phase->thread_work_items();
239 if (work_items != NULL) {
240 out->print("%s", Indents[indent + 1]);
241 work_items->print_summary_on(out, true);
242 details(work_items, Indents[indent + 1]);
243 }
244 }
245
246 void G1GCPhaseTimes::debug_phase(WorkerDataArray<double>* phase) const {
247 Log(gc, phases) log;
248 if (log.is_level(LogLevel::Debug)) {
249 ResourceMark rm;
250 log_phase(phase, 2, log.debug_stream(), true);
251 }
252 }
253
254 void G1GCPhaseTimes::trace_phase(WorkerDataArray<double>* phase, bool print_sum) const {
255 Log(gc, phases) log;
256 if (log.is_level(LogLevel::Trace)) {
257 ResourceMark rm;
258 log_phase(phase, 3, log.trace_stream(), print_sum);
259 }
260 }
261
262 #define TIME_FORMAT "%.1lfms"
263
264 void G1GCPhaseTimes::info_time(const char* name, double value) const {
265 log_info(gc, phases)("%s%s: " TIME_FORMAT, Indents[1], name, value);
266 }
267
268 void G1GCPhaseTimes::debug_time(const char* name, double value) const {
269 log_debug(gc, phases)("%s%s: " TIME_FORMAT, Indents[2], name, value);
270 }
271
272 void G1GCPhaseTimes::trace_time(const char* name, double value) const {
273 log_trace(gc, phases)("%s%s: " TIME_FORMAT, Indents[3], name, value);
274 }
275
276 void G1GCPhaseTimes::trace_count(const char* name, size_t value) const {
277 log_trace(gc, phases)("%s%s: " SIZE_FORMAT, Indents[3], name, value);
278 }
279
280 double G1GCPhaseTimes::print_pre_evacuate_collection_set() const {
281 const double sum_ms = _root_region_scan_wait_time_ms +
282 _recorded_young_cset_choice_time_ms +
283 _recorded_non_young_cset_choice_time_ms +
284 _cur_fast_reclaim_humongous_register_time_ms;
285
286 info_time("Pre Evacuate Collection Set", sum_ms);
287
288 if (_root_region_scan_wait_time_ms > 0.0) {
289 debug_time("Root Region Scan Waiting", _root_region_scan_wait_time_ms);
290 }
291 debug_time("Choose Collection Set", (_recorded_young_cset_choice_time_ms + _recorded_non_young_cset_choice_time_ms));
292 if (G1EagerReclaimHumongousObjects) {
293 debug_time("Humongous Register", _cur_fast_reclaim_humongous_register_time_ms);
294 trace_count("Humongous Total", _cur_fast_reclaim_humongous_total);
295 trace_count("Humongous Candidate", _cur_fast_reclaim_humongous_candidates);
296 }
297
298 return sum_ms;
299 }
300
301 double G1GCPhaseTimes::print_evacuate_collection_set() const {
302 const double sum_ms = _cur_collection_par_time_ms;
303
304 info_time("Evacuate Collection Set", sum_ms);
305
306 trace_phase(_gc_par_phases[GCWorkerStart], false);
307 debug_phase(_gc_par_phases[ExtRootScan]);
308 for (int i = ThreadRoots; i <= SATBFiltering; i++) {
309 trace_phase(_gc_par_phases[i]);
310 }
311 debug_phase(_gc_par_phases[UpdateRS]);
312 if (G1HotCardCache::default_use_cache()) {
313 trace_phase(_gc_par_phases[ScanHCC]);
314 }
315 debug_phase(_gc_par_phases[ScanRS]);
316 debug_phase(_gc_par_phases[CodeRoots]);
317 #if INCLUDE_AOT
318 debug_phase(_gc_par_phases[AOTCodeRoots]);
319 #endif
320 debug_phase(_gc_par_phases[ObjCopy]);
321 debug_phase(_gc_par_phases[Termination]);
322 debug_phase(_gc_par_phases[Other]);
323 debug_phase(_gc_par_phases[GCWorkerTotal]);
324 trace_phase(_gc_par_phases[GCWorkerEnd], false);
325
326 return sum_ms;
327 }
328
329 double G1GCPhaseTimes::print_post_evacuate_collection_set() const {
330 const double evac_fail_handling = _cur_evac_fail_recalc_used +
331 _cur_evac_fail_remove_self_forwards +
332 _cur_evac_fail_restore_remsets;
333 const double sum_ms = evac_fail_handling +
334 _cur_collection_code_root_fixup_time_ms +
335 _recorded_preserve_cm_referents_time_ms +
336 _cur_ref_proc_time_ms +
337 _cur_ref_enq_time_ms +
338 _cur_clear_ct_time_ms +
339 _recorded_merge_pss_time_ms +
340 _cur_strong_code_root_purge_time_ms +
341 _recorded_redirty_logged_cards_time_ms +
342 _recorded_clear_claimed_marks_time_ms +
343 _recorded_total_free_cset_time_ms +
344 _cur_fast_reclaim_humongous_time_ms +
345 _cur_expand_heap_time_ms +
346 _cur_string_dedup_fixup_time_ms;
347
348 info_time("Post Evacuate Collection Set", sum_ms);
349
350 debug_time("Code Roots Fixup", _cur_collection_code_root_fixup_time_ms);
351
352 debug_time("Preserve CM Refs", _recorded_preserve_cm_referents_time_ms);
353 trace_phase(_gc_par_phases[PreserveCMReferents]);
354
355 debug_time("Reference Processing", _cur_ref_proc_time_ms);
356
357 if (G1StringDedup::is_enabled()) {
358 debug_time("String Dedup Fixup", _cur_string_dedup_fixup_time_ms);
359 debug_phase(_gc_par_phases[StringDedupQueueFixup]);
360 debug_phase(_gc_par_phases[StringDedupTableFixup]);
361 }
362
363 debug_time("Clear Card Table", _cur_clear_ct_time_ms);
364
365 if (G1CollectedHeap::heap()->evacuation_failed()) {
366 debug_time("Evacuation Failure", evac_fail_handling);
367 trace_time("Recalculate Used", _cur_evac_fail_recalc_used);
368 trace_time("Remove Self Forwards",_cur_evac_fail_remove_self_forwards);
369 trace_time("Restore RemSet", _cur_evac_fail_restore_remsets);
370 }
371
372 debug_time("Reference Enqueuing", _cur_ref_enq_time_ms);
373
374 debug_time("Merge Per-Thread State", _recorded_merge_pss_time_ms);
375 debug_time("Code Roots Purge", _cur_strong_code_root_purge_time_ms);
376
377 debug_time("Redirty Cards", _recorded_redirty_logged_cards_time_ms);
378 if (_recorded_clear_claimed_marks_time_ms > 0.0) {
379 debug_time("Clear Claimed Marks", _recorded_clear_claimed_marks_time_ms);
380 }
381
382 trace_phase(_gc_par_phases[RedirtyCards]);
383
384 debug_time("Free Collection Set", _recorded_total_free_cset_time_ms);
385 trace_time("Free Collection Set Serial", _recorded_serial_free_cset_time_ms);
386 trace_phase(_gc_par_phases[YoungFreeCSet]);
387 trace_phase(_gc_par_phases[NonYoungFreeCSet]);
388
389 if (G1EagerReclaimHumongousObjects) {
390 debug_time("Humongous Reclaim", _cur_fast_reclaim_humongous_time_ms);
391 trace_count("Humongous Reclaimed", _cur_fast_reclaim_humongous_reclaimed);
392 }
393 debug_time("Expand Heap After Collection", _cur_expand_heap_time_ms);
394
395
396 return sum_ms;
397 }
398
399 void G1GCPhaseTimes::print_other(double accounted_ms) const {
400 info_time("Other", _gc_pause_time_ms - accounted_ms);
401 }
402
403 void G1GCPhaseTimes::print() {
404 note_gc_end();
405
406 if (_cur_verify_before_time_ms > 0.0) {
407 debug_time("Verify Before", _cur_verify_before_time_ms);
408 }
409
410 double accounted_ms = 0.0;
411 accounted_ms += print_pre_evacuate_collection_set();
412 accounted_ms += print_evacuate_collection_set();
413 accounted_ms += print_post_evacuate_collection_set();
414 print_other(accounted_ms);
415
416 if (_cur_verify_after_time_ms > 0.0) {
417 debug_time("Verify After", _cur_verify_after_time_ms);
418 }
419 }
420
421 G1GCParPhaseTimesTracker::G1GCParPhaseTimesTracker(G1GCPhaseTimes* phase_times, G1GCPhaseTimes::GCParPhases phase, uint worker_id) :
422 _phase_times(phase_times), _phase(phase), _worker_id(worker_id) {
423 if (_phase_times != NULL) {
424 _start_time = os::elapsedTime();
425 }
426 }
427
428 G1GCParPhaseTimesTracker::~G1GCParPhaseTimesTracker() {
429 if (_phase_times != NULL) {
430 _phase_times->record_time_secs(_phase, _worker_id, os::elapsedTime() - _start_time);
431 }
432 }
433