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 "logging/logStream.hpp"
34 #include "runtime/timer.hpp"
35 #include "runtime/os.hpp"
36
37 static const char* Indents[5] = {"", " ", " ", " ", " "};
38
39 G1GCPhaseTimes::G1GCPhaseTimes(uint max_gc_threads) :
40 _max_gc_threads(max_gc_threads),
41 _gc_start_counter(0),
42 _gc_pause_time_ms(0.0)
43 {
44 assert(max_gc_threads > 0, "Must have some GC threads");
45
46 _gc_par_phases[GCWorkerStart] = new WorkerDataArray<double>(max_gc_threads, "GC Worker Start (ms):");
47 _gc_par_phases[ExtRootScan] = new WorkerDataArray<double>(max_gc_threads, "Ext Root Scanning (ms):");
48
49 // Root scanning phases
50 _gc_par_phases[ThreadRoots] = new WorkerDataArray<double>(max_gc_threads, "Thread Roots (ms):");
51 _gc_par_phases[StringTableRoots] = new WorkerDataArray<double>(max_gc_threads, "StringTable Roots (ms):");
52 _gc_par_phases[UniverseRoots] = new WorkerDataArray<double>(max_gc_threads, "Universe Roots (ms):");
53 _gc_par_phases[JNIRoots] = new WorkerDataArray<double>(max_gc_threads, "JNI Handles Roots (ms):");
54 _gc_par_phases[ObjectSynchronizerRoots] = new WorkerDataArray<double>(max_gc_threads, "ObjectSynchronizer Roots (ms):");
55 _gc_par_phases[FlatProfilerRoots] = new WorkerDataArray<double>(max_gc_threads, "FlatProfiler Roots (ms):");
56 _gc_par_phases[ManagementRoots] = new WorkerDataArray<double>(max_gc_threads, "Management Roots (ms):");
57 _gc_par_phases[SystemDictionaryRoots] = new WorkerDataArray<double>(max_gc_threads, "SystemDictionary Roots (ms):");
58 _gc_par_phases[CLDGRoots] = new WorkerDataArray<double>(max_gc_threads, "CLDG Roots (ms):");
59 _gc_par_phases[JVMTIRoots] = new WorkerDataArray<double>(max_gc_threads, "JVMTI Roots (ms):");
60 _gc_par_phases[CMRefRoots] = new WorkerDataArray<double>(max_gc_threads, "CM RefProcessor Roots (ms):");
61 _gc_par_phases[WaitForStrongCLD] = new WorkerDataArray<double>(max_gc_threads, "Wait For Strong CLD (ms):");
62 _gc_par_phases[WeakCLDRoots] = new WorkerDataArray<double>(max_gc_threads, "Weak CLD Roots (ms):");
63 _gc_par_phases[SATBFiltering] = new WorkerDataArray<double>(max_gc_threads, "SATB Filtering (ms):");
64
65 _gc_par_phases[UpdateRS] = new WorkerDataArray<double>(max_gc_threads, "Update RS (ms):");
66 if (G1HotCardCache::default_use_cache()) {
67 _gc_par_phases[ScanHCC] = new WorkerDataArray<double>(max_gc_threads, "Scan HCC (ms):");
68 } else {
69 _gc_par_phases[ScanHCC] = NULL;
70 }
71 _gc_par_phases[ScanRS] = new WorkerDataArray<double>(max_gc_threads, "Scan RS (ms):");
72 _gc_par_phases[CodeRoots] = new WorkerDataArray<double>(max_gc_threads, "Code Root Scanning (ms):");
73 #if INCLUDE_AOT
74 _gc_par_phases[AOTCodeRoots] = new WorkerDataArray<double>(max_gc_threads, "AOT Root Scanning (ms):");
75 #endif
76 _gc_par_phases[ObjCopy] = new WorkerDataArray<double>(max_gc_threads, "Object Copy (ms):");
77 _gc_par_phases[Termination] = new WorkerDataArray<double>(max_gc_threads, "Termination (ms):");
78 _gc_par_phases[GCWorkerTotal] = new WorkerDataArray<double>(max_gc_threads, "GC Worker Total (ms):");
79 _gc_par_phases[GCWorkerEnd] = new WorkerDataArray<double>(max_gc_threads, "GC Worker End (ms):");
80 _gc_par_phases[Other] = new WorkerDataArray<double>(max_gc_threads, "GC Worker Other (ms):");
81
82 _update_rs_processed_buffers = new WorkerDataArray<size_t>(max_gc_threads, "Processed Buffers:");
83 _gc_par_phases[UpdateRS]->link_thread_work_items(_update_rs_processed_buffers);
84
85 _termination_attempts = new WorkerDataArray<size_t>(max_gc_threads, "Termination Attempts:");
86 _gc_par_phases[Termination]->link_thread_work_items(_termination_attempts);
87
88 if (UseStringDeduplication) {
89 _gc_par_phases[StringDedupQueueFixup] = new WorkerDataArray<double>(max_gc_threads, "Queue Fixup (ms):");
90 _gc_par_phases[StringDedupTableFixup] = new WorkerDataArray<double>(max_gc_threads, "Table Fixup (ms):");
91 } else {
92 _gc_par_phases[StringDedupQueueFixup] = NULL;
93 _gc_par_phases[StringDedupTableFixup] = NULL;
94 }
95
96 _gc_par_phases[RedirtyCards] = new WorkerDataArray<double>(max_gc_threads, "Parallel Redirty (ms):");
97 _redirtied_cards = new WorkerDataArray<size_t>(max_gc_threads, "Redirtied Cards:");
98 _gc_par_phases[RedirtyCards]->link_thread_work_items(_redirtied_cards);
99
100 _gc_par_phases[YoungFreeCSet] = new WorkerDataArray<double>(max_gc_threads, "Young Free Collection Set (ms):");
101 _gc_par_phases[NonYoungFreeCSet] = new WorkerDataArray<double>(max_gc_threads, "Non-Young Free Collection Set (ms):");
102
103 _gc_par_phases[PreserveCMReferents] = new WorkerDataArray<double>(max_gc_threads, "Parallel Preserve CM Refs (ms):");
104
105 reset();
106 }
107
108 void G1GCPhaseTimes::reset() {
109 _cur_collection_par_time_ms = 0.0;
110 _cur_collection_code_root_fixup_time_ms = 0.0;
111 _cur_strong_code_root_purge_time_ms = 0.0;
112 _cur_evac_fail_recalc_used = 0.0;
113 _cur_evac_fail_restore_remsets = 0.0;
114 _cur_evac_fail_remove_self_forwards = 0.0;
115 _cur_string_dedup_fixup_time_ms = 0.0;
116 _cur_prepare_tlab_time_ms = 0.0;
117 _cur_resize_tlab_time_ms = 0.0;
118 _cur_derived_pointer_table_update_time_ms = 0.0;
119 _cur_clear_ct_time_ms = 0.0;
120 _cur_expand_heap_time_ms = 0.0;
121 _cur_ref_proc_time_ms = 0.0;
122 _cur_ref_enq_time_ms = 0.0;
123 _cur_collection_start_sec = 0.0;
124 _root_region_scan_wait_time_ms = 0.0;
125 _external_accounted_time_ms = 0.0;
126 _recorded_clear_claimed_marks_time_ms = 0.0;
127 _recorded_young_cset_choice_time_ms = 0.0;
128 _recorded_non_young_cset_choice_time_ms = 0.0;
129 _recorded_redirty_logged_cards_time_ms = 0.0;
130 _recorded_preserve_cm_referents_time_ms = 0.0;
131 _recorded_merge_pss_time_ms = 0.0;
132 _recorded_start_new_cset_time_ms = 0.0;
133 _recorded_total_free_cset_time_ms = 0.0;
134 _recorded_serial_free_cset_time_ms = 0.0;
135 _cur_fast_reclaim_humongous_time_ms = 0.0;
136 _cur_fast_reclaim_humongous_register_time_ms = 0.0;
137 _cur_fast_reclaim_humongous_total = 0;
138 _cur_fast_reclaim_humongous_candidates = 0;
139 _cur_fast_reclaim_humongous_reclaimed = 0;
140 _cur_verify_before_time_ms = 0.0;
141 _cur_verify_after_time_ms = 0.0;
142
143 for (int i = 0; i < GCParPhasesSentinel; i++) {
144 if (_gc_par_phases[i] != NULL) {
145 _gc_par_phases[i]->reset();
146 }
147 }
148 }
149
150 void G1GCPhaseTimes::note_gc_start() {
151 _gc_start_counter = os::elapsed_counter();
152 reset();
153 }
154
155 #define ASSERT_PHASE_UNINITIALIZED(phase) \
156 assert(_gc_par_phases[phase]->get(i) == uninitialized, "Phase " #phase " reported for thread that was not started");
157
158 double G1GCPhaseTimes::worker_time(GCParPhases phase, uint worker) {
159 double value = _gc_par_phases[phase]->get(worker);
160 if (value != WorkerDataArray<double>::uninitialized()) {
161 return value;
162 }
163 return 0.0;
164 }
165
166 void G1GCPhaseTimes::note_gc_end() {
167 _gc_pause_time_ms = TimeHelper::counter_to_millis(os::elapsed_counter() - _gc_start_counter);
168
169 double uninitialized = WorkerDataArray<double>::uninitialized();
170
171 for (uint i = 0; i < _max_gc_threads; i++) {
172 double worker_start = _gc_par_phases[GCWorkerStart]->get(i);
173 if (worker_start != uninitialized) {
174 assert(_gc_par_phases[GCWorkerEnd]->get(i) != uninitialized, "Worker started but not ended.");
175 double total_worker_time = _gc_par_phases[GCWorkerEnd]->get(i) - _gc_par_phases[GCWorkerStart]->get(i);
176 record_time_secs(GCWorkerTotal, i , total_worker_time);
177
178 double worker_known_time =
179 worker_time(ExtRootScan, i)
180 + worker_time(SATBFiltering, i)
181 + worker_time(UpdateRS, i)
182 + worker_time(ScanRS, i)
183 + worker_time(CodeRoots, i)
184 + worker_time(ObjCopy, i)
185 + worker_time(Termination, i);
186
187 record_time_secs(Other, i, total_worker_time - worker_known_time);
188 } else {
189 // Make sure all slots are uninitialized since this thread did not seem to have been started
190 ASSERT_PHASE_UNINITIALIZED(GCWorkerEnd);
191 ASSERT_PHASE_UNINITIALIZED(ExtRootScan);
192 ASSERT_PHASE_UNINITIALIZED(SATBFiltering);
193 ASSERT_PHASE_UNINITIALIZED(UpdateRS);
194 ASSERT_PHASE_UNINITIALIZED(ScanRS);
195 ASSERT_PHASE_UNINITIALIZED(CodeRoots);
196 ASSERT_PHASE_UNINITIALIZED(ObjCopy);
197 ASSERT_PHASE_UNINITIALIZED(Termination);
198 }
199 }
200 }
201
202 #undef ASSERT_PHASE_UNINITIALIZED
203
204 // record the time a phase took in seconds
205 void G1GCPhaseTimes::record_time_secs(GCParPhases phase, uint worker_i, double secs) {
206 _gc_par_phases[phase]->set(worker_i, secs);
207 }
208
209 // add a number of seconds to a phase
210 void G1GCPhaseTimes::add_time_secs(GCParPhases phase, uint worker_i, double secs) {
211 _gc_par_phases[phase]->add(worker_i, secs);
212 }
213
214 void G1GCPhaseTimes::record_thread_work_item(GCParPhases phase, uint worker_i, size_t count) {
215 _gc_par_phases[phase]->set_thread_work_item(worker_i, count);
216 }
217
218 // return the average time for a phase in milliseconds
219 double G1GCPhaseTimes::average_time_ms(GCParPhases phase) {
220 return _gc_par_phases[phase]->average() * 1000.0;
221 }
222
223 size_t G1GCPhaseTimes::sum_thread_work_items(GCParPhases phase) {
224 assert(_gc_par_phases[phase]->thread_work_items() != NULL, "No sub count");
225 return _gc_par_phases[phase]->thread_work_items()->sum();
226 }
227
228 template <class T>
229 void G1GCPhaseTimes::details(T* phase, const char* indent) const {
230 LogTarget(Trace, gc, phases, task) lt;
231 if (lt.is_enabled()) {
232 LogStream ls(lt);
233 ls.print("%s", indent);
234 phase->print_details_on(&ls);
235 }
236 }
237
238 void G1GCPhaseTimes::log_phase(WorkerDataArray<double>* phase, uint indent, outputStream* out, bool print_sum) const {
239 out->print("%s", Indents[indent]);
240 phase->print_summary_on(out, print_sum);
241 details(phase, Indents[indent]);
242
243 WorkerDataArray<size_t>* work_items = phase->thread_work_items();
244 if (work_items != NULL) {
245 out->print("%s", Indents[indent + 1]);
246 work_items->print_summary_on(out, true);
247 details(work_items, Indents[indent + 1]);
248 }
249 }
250
251 void G1GCPhaseTimes::debug_phase(WorkerDataArray<double>* phase) const {
252 Log(gc, phases) log;
253 if (log.is_level(LogLevel::Debug)) {
254 ResourceMark rm;
255 LogStream ls(log.debug());
256 log_phase(phase, 2, &ls, true);
257 }
258 }
259
260 void G1GCPhaseTimes::trace_phase(WorkerDataArray<double>* phase, bool print_sum) const {
261 LogTarget(Trace, gc, phases) lt;
262 if (lt.is_enabled()) {
263 LogStream ls(lt);
264 log_phase(phase, 3, &ls, print_sum);
265 }
266 }
267
268 #define TIME_FORMAT "%.1lfms"
269
270 void G1GCPhaseTimes::info_time(const char* name, double value) const {
271 log_info(gc, phases)("%s%s: " TIME_FORMAT, Indents[1], name, value);
272 }
273
274 void G1GCPhaseTimes::debug_time(const char* name, double value) const {
275 log_debug(gc, phases)("%s%s: " TIME_FORMAT, Indents[2], name, value);
276 }
277
278 void G1GCPhaseTimes::trace_time(const char* name, double value) const {
279 log_trace(gc, phases)("%s%s: " TIME_FORMAT, Indents[3], name, value);
280 }
281
282 void G1GCPhaseTimes::trace_count(const char* name, size_t value) const {
283 log_trace(gc, phases)("%s%s: " SIZE_FORMAT, Indents[3], name, value);
284 }
285
286 double G1GCPhaseTimes::print_pre_evacuate_collection_set() const {
287 const double sum_ms = _root_region_scan_wait_time_ms +
288 _recorded_young_cset_choice_time_ms +
289 _recorded_non_young_cset_choice_time_ms +
290 _cur_fast_reclaim_humongous_register_time_ms +
291 _recorded_clear_claimed_marks_time_ms;
292
293 info_time("Pre Evacuate Collection Set", sum_ms);
294
295 if (_root_region_scan_wait_time_ms > 0.0) {
296 debug_time("Root Region Scan Waiting", _root_region_scan_wait_time_ms);
297 }
298 debug_time("Prepare TLABs", _cur_prepare_tlab_time_ms);
299 debug_time("Choose Collection Set", (_recorded_young_cset_choice_time_ms + _recorded_non_young_cset_choice_time_ms));
300 if (G1EagerReclaimHumongousObjects) {
301 debug_time("Humongous Register", _cur_fast_reclaim_humongous_register_time_ms);
302 trace_count("Humongous Total", _cur_fast_reclaim_humongous_total);
303 trace_count("Humongous Candidate", _cur_fast_reclaim_humongous_candidates);
304 }
305
306 if (_recorded_clear_claimed_marks_time_ms > 0.0) {
307 debug_time("Clear Claimed Marks", _recorded_clear_claimed_marks_time_ms);
308 }
309 return sum_ms;
310 }
311
312 double G1GCPhaseTimes::print_evacuate_collection_set() const {
313 const double sum_ms = _cur_collection_par_time_ms;
314
315 info_time("Evacuate Collection Set", sum_ms);
316
317 trace_phase(_gc_par_phases[GCWorkerStart], false);
318 debug_phase(_gc_par_phases[ExtRootScan]);
319 for (int i = ThreadRoots; i <= SATBFiltering; i++) {
320 trace_phase(_gc_par_phases[i]);
321 }
322 debug_phase(_gc_par_phases[UpdateRS]);
323 if (G1HotCardCache::default_use_cache()) {
324 trace_phase(_gc_par_phases[ScanHCC]);
325 }
326 debug_phase(_gc_par_phases[ScanRS]);
327 debug_phase(_gc_par_phases[CodeRoots]);
328 #if INCLUDE_AOT
329 debug_phase(_gc_par_phases[AOTCodeRoots]);
330 #endif
331 debug_phase(_gc_par_phases[ObjCopy]);
332 debug_phase(_gc_par_phases[Termination]);
333 debug_phase(_gc_par_phases[Other]);
334 debug_phase(_gc_par_phases[GCWorkerTotal]);
335 trace_phase(_gc_par_phases[GCWorkerEnd], false);
336
337 return sum_ms;
338 }
339
340 double G1GCPhaseTimes::print_post_evacuate_collection_set() const {
341 const double evac_fail_handling = _cur_evac_fail_recalc_used +
342 _cur_evac_fail_remove_self_forwards +
343 _cur_evac_fail_restore_remsets;
344 const double sum_ms = evac_fail_handling +
345 _cur_collection_code_root_fixup_time_ms +
346 _recorded_preserve_cm_referents_time_ms +
347 _cur_ref_proc_time_ms +
348 _cur_ref_enq_time_ms +
349 _cur_clear_ct_time_ms +
350 _recorded_merge_pss_time_ms +
351 _cur_strong_code_root_purge_time_ms +
352 _recorded_redirty_logged_cards_time_ms +
353 _recorded_total_free_cset_time_ms +
354 _cur_fast_reclaim_humongous_time_ms +
355 _cur_expand_heap_time_ms +
356 _cur_string_dedup_fixup_time_ms;
357
358 info_time("Post Evacuate Collection Set", sum_ms);
359
360 debug_time("Code Roots Fixup", _cur_collection_code_root_fixup_time_ms);
361
362 debug_time("Preserve CM Refs", _recorded_preserve_cm_referents_time_ms);
363 trace_phase(_gc_par_phases[PreserveCMReferents]);
364
365 debug_time("Reference Processing", _cur_ref_proc_time_ms);
366
367 if (G1StringDedup::is_enabled()) {
368 debug_time("String Dedup Fixup", _cur_string_dedup_fixup_time_ms);
369 debug_phase(_gc_par_phases[StringDedupQueueFixup]);
370 debug_phase(_gc_par_phases[StringDedupTableFixup]);
371 }
372
373 debug_time("Clear Card Table", _cur_clear_ct_time_ms);
374
375 if (G1CollectedHeap::heap()->evacuation_failed()) {
376 debug_time("Evacuation Failure", evac_fail_handling);
377 trace_time("Recalculate Used", _cur_evac_fail_recalc_used);
378 trace_time("Remove Self Forwards",_cur_evac_fail_remove_self_forwards);
379 trace_time("Restore RemSet", _cur_evac_fail_restore_remsets);
380 }
381
382 debug_time("Reference Enqueuing", _cur_ref_enq_time_ms);
383
384 debug_time("Merge Per-Thread State", _recorded_merge_pss_time_ms);
385 debug_time("Code Roots Purge", _cur_strong_code_root_purge_time_ms);
386
387 debug_time("Redirty Cards", _recorded_redirty_logged_cards_time_ms);
388 trace_phase(_gc_par_phases[RedirtyCards]);
389 #if defined(COMPILER2) || INCLUDE_JVMCI
390 debug_time("DerivedPointerTable Update", _cur_derived_pointer_table_update_time_ms);
391 #endif
392
393 debug_time("Free Collection Set", _recorded_total_free_cset_time_ms);
394 trace_time("Free Collection Set Serial", _recorded_serial_free_cset_time_ms);
395 trace_phase(_gc_par_phases[YoungFreeCSet]);
396 trace_phase(_gc_par_phases[NonYoungFreeCSet]);
397
398 if (G1EagerReclaimHumongousObjects) {
399 debug_time("Humongous Reclaim", _cur_fast_reclaim_humongous_time_ms);
400 trace_count("Humongous Reclaimed", _cur_fast_reclaim_humongous_reclaimed);
401 }
402 debug_time("Start New Collection Set", _recorded_start_new_cset_time_ms);
403 if (UseTLAB && ResizeTLAB) {
404 debug_time("Resize TLABs", _cur_resize_tlab_time_ms);
405 }
406 debug_time("Expand Heap After Collection", _cur_expand_heap_time_ms);
407
408
409 return sum_ms;
410 }
411
412 void G1GCPhaseTimes::print_other(double accounted_ms) const {
413 info_time("Other", _gc_pause_time_ms - accounted_ms);
414 }
415
416 void G1GCPhaseTimes::print() {
417 note_gc_end();
418
419 if (_cur_verify_before_time_ms > 0.0) {
420 debug_time("Verify Before", _cur_verify_before_time_ms);
421 }
422
423 double accounted_ms = 0.0;
424 accounted_ms += print_pre_evacuate_collection_set();
425 accounted_ms += print_evacuate_collection_set();
426 accounted_ms += print_post_evacuate_collection_set();
427 print_other(accounted_ms);
428
429 if (_cur_verify_after_time_ms > 0.0) {
430 debug_time("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