1 /*
2 * Copyright (c) 2013, 2017, 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 _scan_rs_scanned_cards = new WorkerDataArray<size_t>(max_gc_threads, "Scanned Cards:");
82 _gc_par_phases[ScanRS]->link_thread_work_items(_scan_rs_scanned_cards, ScanRSScannedCards);
83 _scan_rs_claimed_cards = new WorkerDataArray<size_t>(max_gc_threads, "Claimed Cards:");
84 _gc_par_phases[ScanRS]->link_thread_work_items(_scan_rs_claimed_cards, ScanRSClaimedCards);
85 _scan_rs_skipped_cards = new WorkerDataArray<size_t>(max_gc_threads, "Skipped Cards:");
86 _gc_par_phases[ScanRS]->link_thread_work_items(_scan_rs_skipped_cards, ScanRSSkippedCards);
87
88 _update_rs_processed_buffers = new WorkerDataArray<size_t>(max_gc_threads, "Processed Buffers:");
89 _gc_par_phases[UpdateRS]->link_thread_work_items(_update_rs_processed_buffers, UpdateRSProcessedBuffers);
90 _update_rs_scanned_cards = new WorkerDataArray<size_t>(max_gc_threads, "Scanned Cards:");
91 _gc_par_phases[UpdateRS]->link_thread_work_items(_update_rs_scanned_cards, UpdateRSScannedCards);
92 _update_rs_skipped_cards = new WorkerDataArray<size_t>(max_gc_threads, "Skipped Cards:");
93 _gc_par_phases[UpdateRS]->link_thread_work_items(_update_rs_skipped_cards, ScanRSSkippedCards);
94
95 _termination_attempts = new WorkerDataArray<size_t>(max_gc_threads, "Termination Attempts:");
96 _gc_par_phases[Termination]->link_thread_work_items(_termination_attempts);
97
98 if (UseStringDeduplication) {
99 _gc_par_phases[StringDedupQueueFixup] = new WorkerDataArray<double>(max_gc_threads, "Queue Fixup (ms):");
100 _gc_par_phases[StringDedupTableFixup] = new WorkerDataArray<double>(max_gc_threads, "Table Fixup (ms):");
101 } else {
102 _gc_par_phases[StringDedupQueueFixup] = NULL;
103 _gc_par_phases[StringDedupTableFixup] = NULL;
104 }
105
106 _gc_par_phases[RedirtyCards] = new WorkerDataArray<double>(max_gc_threads, "Parallel Redirty (ms):");
107 _redirtied_cards = new WorkerDataArray<size_t>(max_gc_threads, "Redirtied Cards:");
108 _gc_par_phases[RedirtyCards]->link_thread_work_items(_redirtied_cards);
109
110 _gc_par_phases[YoungFreeCSet] = new WorkerDataArray<double>(max_gc_threads, "Young Free Collection Set (ms):");
111 _gc_par_phases[NonYoungFreeCSet] = new WorkerDataArray<double>(max_gc_threads, "Non-Young Free Collection Set (ms):");
112
113 _gc_par_phases[PreserveCMReferents] = new WorkerDataArray<double>(max_gc_threads, "Parallel Preserve CM Refs (ms):");
114
115 reset();
116 }
117
118 void G1GCPhaseTimes::reset() {
119 _cur_collection_par_time_ms = 0.0;
120 _cur_collection_code_root_fixup_time_ms = 0.0;
121 _cur_strong_code_root_purge_time_ms = 0.0;
122 _cur_evac_fail_recalc_used = 0.0;
123 _cur_evac_fail_remove_self_forwards = 0.0;
124 _cur_string_dedup_fixup_time_ms = 0.0;
125 _cur_prepare_tlab_time_ms = 0.0;
126 _cur_resize_tlab_time_ms = 0.0;
127 _cur_derived_pointer_table_update_time_ms = 0.0;
128 _cur_clear_ct_time_ms = 0.0;
129 _cur_expand_heap_time_ms = 0.0;
130 _cur_ref_proc_time_ms = 0.0;
131 _cur_ref_enq_time_ms = 0.0;
132 _cur_collection_start_sec = 0.0;
133 _root_region_scan_wait_time_ms = 0.0;
134 _external_accounted_time_ms = 0.0;
135 _recorded_clear_claimed_marks_time_ms = 0.0;
136 _recorded_young_cset_choice_time_ms = 0.0;
137 _recorded_non_young_cset_choice_time_ms = 0.0;
138 _recorded_redirty_logged_cards_time_ms = 0.0;
139 _recorded_preserve_cm_referents_time_ms = 0.0;
140 _recorded_merge_pss_time_ms = 0.0;
141 _recorded_start_new_cset_time_ms = 0.0;
142 _recorded_total_free_cset_time_ms = 0.0;
143 _recorded_serial_free_cset_time_ms = 0.0;
144 _cur_fast_reclaim_humongous_time_ms = 0.0;
145 _cur_fast_reclaim_humongous_register_time_ms = 0.0;
146 _cur_fast_reclaim_humongous_total = 0;
147 _cur_fast_reclaim_humongous_candidates = 0;
148 _cur_fast_reclaim_humongous_reclaimed = 0;
149 _cur_verify_before_time_ms = 0.0;
150 _cur_verify_after_time_ms = 0.0;
151
152 for (int i = 0; i < GCParPhasesSentinel; i++) {
153 if (_gc_par_phases[i] != NULL) {
154 _gc_par_phases[i]->reset();
155 }
156 }
157 }
158
159 void G1GCPhaseTimes::note_gc_start() {
160 _gc_start_counter = os::elapsed_counter();
161 reset();
162 }
163
164 #define ASSERT_PHASE_UNINITIALIZED(phase) \
165 assert(_gc_par_phases[phase]->get(i) == uninitialized, "Phase " #phase " reported for thread that was not started");
166
167 double G1GCPhaseTimes::worker_time(GCParPhases phase, uint worker) {
168 double value = _gc_par_phases[phase]->get(worker);
169 if (value != WorkerDataArray<double>::uninitialized()) {
170 return value;
171 }
172 return 0.0;
173 }
174
175 void G1GCPhaseTimes::note_gc_end() {
176 _gc_pause_time_ms = TimeHelper::counter_to_millis(os::elapsed_counter() - _gc_start_counter);
177
178 double uninitialized = WorkerDataArray<double>::uninitialized();
179
180 for (uint i = 0; i < _max_gc_threads; i++) {
181 double worker_start = _gc_par_phases[GCWorkerStart]->get(i);
182 if (worker_start != uninitialized) {
183 assert(_gc_par_phases[GCWorkerEnd]->get(i) != uninitialized, "Worker started but not ended.");
184 double total_worker_time = _gc_par_phases[GCWorkerEnd]->get(i) - _gc_par_phases[GCWorkerStart]->get(i);
185 record_time_secs(GCWorkerTotal, i , total_worker_time);
186
187 double worker_known_time =
188 worker_time(ExtRootScan, i)
189 + worker_time(SATBFiltering, i)
190 + worker_time(UpdateRS, i)
191 + worker_time(ScanRS, i)
192 + worker_time(CodeRoots, i)
193 + worker_time(ObjCopy, i)
194 + worker_time(Termination, i);
195
196 record_time_secs(Other, i, total_worker_time - worker_known_time);
197 } else {
198 // Make sure all slots are uninitialized since this thread did not seem to have been started
199 ASSERT_PHASE_UNINITIALIZED(GCWorkerEnd);
200 ASSERT_PHASE_UNINITIALIZED(ExtRootScan);
201 ASSERT_PHASE_UNINITIALIZED(SATBFiltering);
202 ASSERT_PHASE_UNINITIALIZED(UpdateRS);
203 ASSERT_PHASE_UNINITIALIZED(ScanRS);
204 ASSERT_PHASE_UNINITIALIZED(CodeRoots);
205 ASSERT_PHASE_UNINITIALIZED(ObjCopy);
206 ASSERT_PHASE_UNINITIALIZED(Termination);
207 }
208 }
209 }
210
211 #undef ASSERT_PHASE_UNINITIALIZED
212
213 // record the time a phase took in seconds
214 void G1GCPhaseTimes::record_time_secs(GCParPhases phase, uint worker_i, double secs) {
215 _gc_par_phases[phase]->set(worker_i, secs);
216 }
217
218 // add a number of seconds to a phase
219 void G1GCPhaseTimes::add_time_secs(GCParPhases phase, uint worker_i, double secs) {
220 _gc_par_phases[phase]->add(worker_i, secs);
221 }
222
223 void G1GCPhaseTimes::record_thread_work_item(GCParPhases phase, uint worker_i, size_t count, uint index) {
224 _gc_par_phases[phase]->set_thread_work_item(worker_i, count, index);
225 }
226
227 // return the average time for a phase in milliseconds
228 double G1GCPhaseTimes::average_time_ms(GCParPhases phase) {
229 return _gc_par_phases[phase]->average() * 1000.0;
230 }
231
232 size_t G1GCPhaseTimes::sum_thread_work_items(GCParPhases phase, uint index) {
233 assert(_gc_par_phases[phase]->thread_work_items(index) != NULL, "No sub count");
234 return _gc_par_phases[phase]->thread_work_items(index)->sum();
235 }
236
237 template <class T>
238 void G1GCPhaseTimes::details(T* phase, const char* indent) const {
239 Log(gc, phases, task) log;
240 if (log.is_level(LogLevel::Trace)) {
241 outputStream* trace_out = log.trace_stream();
242 trace_out->print("%s", indent);
243 phase->print_details_on(trace_out);
244 }
245 }
246
247 void G1GCPhaseTimes::log_phase(WorkerDataArray<double>* phase, uint indent, outputStream* out, bool print_sum) const {
248 out->print("%s", Indents[indent]);
249 phase->print_summary_on(out, print_sum);
250 details(phase, Indents[indent]);
251
252 for (uint i = 0; i < phase->MaxThreadWorkItems; i++) {
253 WorkerDataArray<size_t>* work_items = phase->thread_work_items(i);
254 if (work_items != NULL) {
255 out->print("%s", Indents[indent + 1]);
256 work_items->print_summary_on(out, true);
257 details(work_items, Indents[indent + 1]);
258 }
259 }
260 }
261
262 void G1GCPhaseTimes::debug_phase(WorkerDataArray<double>* phase) const {
263 Log(gc, phases) log;
264 if (log.is_level(LogLevel::Debug)) {
265 ResourceMark rm;
266 log_phase(phase, 2, log.debug_stream(), true);
267 }
268 }
269
270 void G1GCPhaseTimes::trace_phase(WorkerDataArray<double>* phase, bool print_sum) const {
271 Log(gc, phases) log;
272 if (log.is_level(LogLevel::Trace)) {
273 ResourceMark rm;
274 log_phase(phase, 3, log.trace_stream(), print_sum);
275 }
276 }
277
278 #define TIME_FORMAT "%.1lfms"
279
280 void G1GCPhaseTimes::info_time(const char* name, double value) const {
281 log_info(gc, phases)("%s%s: " TIME_FORMAT, Indents[1], name, value);
282 }
283
284 void G1GCPhaseTimes::debug_time(const char* name, double value) const {
285 log_debug(gc, phases)("%s%s: " TIME_FORMAT, Indents[2], name, value);
286 }
287
288 void G1GCPhaseTimes::trace_time(const char* name, double value) const {
289 log_trace(gc, phases)("%s%s: " TIME_FORMAT, Indents[3], name, value);
290 }
291
292 void G1GCPhaseTimes::trace_count(const char* name, size_t value) const {
293 log_trace(gc, phases)("%s%s: " SIZE_FORMAT, Indents[3], name, value);
294 }
295
296 double G1GCPhaseTimes::print_pre_evacuate_collection_set() const {
297 const double sum_ms = _root_region_scan_wait_time_ms +
298 _recorded_young_cset_choice_time_ms +
299 _recorded_non_young_cset_choice_time_ms +
300 _cur_fast_reclaim_humongous_register_time_ms +
301 _recorded_clear_claimed_marks_time_ms;
302
303 info_time("Pre Evacuate Collection Set", sum_ms);
304
305 if (_root_region_scan_wait_time_ms > 0.0) {
306 debug_time("Root Region Scan Waiting", _root_region_scan_wait_time_ms);
307 }
308 debug_time("Prepare TLABs", _cur_prepare_tlab_time_ms);
309 debug_time("Choose Collection Set", (_recorded_young_cset_choice_time_ms + _recorded_non_young_cset_choice_time_ms));
310 if (G1EagerReclaimHumongousObjects) {
311 debug_time("Humongous Register", _cur_fast_reclaim_humongous_register_time_ms);
312 trace_count("Humongous Total", _cur_fast_reclaim_humongous_total);
313 trace_count("Humongous Candidate", _cur_fast_reclaim_humongous_candidates);
314 }
315
316 if (_recorded_clear_claimed_marks_time_ms > 0.0) {
317 debug_time("Clear Claimed Marks", _recorded_clear_claimed_marks_time_ms);
318 }
319 return sum_ms;
320 }
321
322 double G1GCPhaseTimes::print_evacuate_collection_set() const {
323 const double sum_ms = _cur_collection_par_time_ms;
324
325 info_time("Evacuate Collection Set", sum_ms);
326
327 trace_phase(_gc_par_phases[GCWorkerStart], false);
328 debug_phase(_gc_par_phases[ExtRootScan]);
329 for (int i = ThreadRoots; i <= SATBFiltering; i++) {
330 trace_phase(_gc_par_phases[i]);
331 }
332 debug_phase(_gc_par_phases[UpdateRS]);
333 if (G1HotCardCache::default_use_cache()) {
334 trace_phase(_gc_par_phases[ScanHCC]);
335 }
336 debug_phase(_gc_par_phases[ScanRS]);
337 debug_phase(_gc_par_phases[CodeRoots]);
338 #if INCLUDE_AOT
339 debug_phase(_gc_par_phases[AOTCodeRoots]);
340 #endif
341 debug_phase(_gc_par_phases[ObjCopy]);
342 debug_phase(_gc_par_phases[Termination]);
343 debug_phase(_gc_par_phases[Other]);
344 debug_phase(_gc_par_phases[GCWorkerTotal]);
345 trace_phase(_gc_par_phases[GCWorkerEnd], false);
346
347 return sum_ms;
348 }
349
350 double G1GCPhaseTimes::print_post_evacuate_collection_set() const {
351 const double evac_fail_handling = _cur_evac_fail_recalc_used +
352 _cur_evac_fail_remove_self_forwards;
353 const double sum_ms = evac_fail_handling +
354 _cur_collection_code_root_fixup_time_ms +
355 _recorded_preserve_cm_referents_time_ms +
356 _cur_ref_proc_time_ms +
357 _cur_ref_enq_time_ms +
358 _cur_clear_ct_time_ms +
359 _recorded_merge_pss_time_ms +
360 _cur_strong_code_root_purge_time_ms +
361 _recorded_redirty_logged_cards_time_ms +
362 _recorded_total_free_cset_time_ms +
363 _cur_fast_reclaim_humongous_time_ms +
364 _cur_expand_heap_time_ms +
365 _cur_string_dedup_fixup_time_ms;
366
367 info_time("Post Evacuate Collection Set", sum_ms);
368
369 debug_time("Code Roots Fixup", _cur_collection_code_root_fixup_time_ms);
370
371 debug_time("Preserve CM Refs", _recorded_preserve_cm_referents_time_ms);
372 trace_phase(_gc_par_phases[PreserveCMReferents]);
373
374 debug_time("Reference Processing", _cur_ref_proc_time_ms);
375
376 if (G1StringDedup::is_enabled()) {
377 debug_time("String Dedup Fixup", _cur_string_dedup_fixup_time_ms);
378 debug_phase(_gc_par_phases[StringDedupQueueFixup]);
379 debug_phase(_gc_par_phases[StringDedupTableFixup]);
380 }
381
382 debug_time("Clear Card Table", _cur_clear_ct_time_ms);
383
384 if (G1CollectedHeap::heap()->evacuation_failed()) {
385 debug_time("Evacuation Failure", evac_fail_handling);
386 trace_time("Recalculate Used", _cur_evac_fail_recalc_used);
387 trace_time("Remove Self Forwards",_cur_evac_fail_remove_self_forwards);
388 }
389
390 debug_time("Reference Enqueuing", _cur_ref_enq_time_ms);
391
392 debug_time("Merge Per-Thread State", _recorded_merge_pss_time_ms);
393 debug_time("Code Roots Purge", _cur_strong_code_root_purge_time_ms);
394
395 debug_time("Redirty Cards", _recorded_redirty_logged_cards_time_ms);
396 trace_phase(_gc_par_phases[RedirtyCards]);
397 #if defined(COMPILER2) || INCLUDE_JVMCI
398 debug_time("DerivedPointerTable Update", _cur_derived_pointer_table_update_time_ms);
399 #endif
400
401 debug_time("Free Collection Set", _recorded_total_free_cset_time_ms);
402 trace_time("Free Collection Set Serial", _recorded_serial_free_cset_time_ms);
403 trace_phase(_gc_par_phases[YoungFreeCSet]);
404 trace_phase(_gc_par_phases[NonYoungFreeCSet]);
405
406 if (G1EagerReclaimHumongousObjects) {
407 debug_time("Humongous Reclaim", _cur_fast_reclaim_humongous_time_ms);
408 trace_count("Humongous Reclaimed", _cur_fast_reclaim_humongous_reclaimed);
409 }
410 debug_time("Start New Collection Set", _recorded_start_new_cset_time_ms);
411 if (UseTLAB && ResizeTLAB) {
412 debug_time("Resize TLABs", _cur_resize_tlab_time_ms);
413 }
414 debug_time("Expand Heap After Collection", _cur_expand_heap_time_ms);
415
416
417 return sum_ms;
418 }
419
420 void G1GCPhaseTimes::print_other(double accounted_ms) const {
421 info_time("Other", _gc_pause_time_ms - accounted_ms);
422 }
423
424 void G1GCPhaseTimes::print() {
425 note_gc_end();
426
427 if (_cur_verify_before_time_ms > 0.0) {
428 debug_time("Verify Before", _cur_verify_before_time_ms);
429 }
430
431 double accounted_ms = 0.0;
432 accounted_ms += print_pre_evacuate_collection_set();
433 accounted_ms += print_evacuate_collection_set();
434 accounted_ms += print_post_evacuate_collection_set();
435 print_other(accounted_ms);
436
437 if (_cur_verify_after_time_ms > 0.0) {
438 debug_time("Verify After", _cur_verify_after_time_ms);
439 }
440 }
441
442 G1GCParPhaseTimesTracker::G1GCParPhaseTimesTracker(G1GCPhaseTimes* phase_times, G1GCPhaseTimes::GCParPhases phase, uint worker_id) :
443 _phase_times(phase_times), _phase(phase), _worker_id(worker_id) {
444 if (_phase_times != NULL) {
445 _start_time = os::elapsedTime();
446 }
447 }
448
449 G1GCParPhaseTimesTracker::~G1GCParPhaseTimesTracker() {
450 if (_phase_times != NULL) {
451 _phase_times->record_time_secs(_phase, _worker_id, os::elapsedTime() - _start_time);
452 }
453 }
454