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 "logging/logStream.hpp"
34 #include "runtime/timer.hpp"
35 #include "runtime/os.hpp"
36 #include "utilities/macros.hpp"
37
38 static const char* Indents[5] = {"", " ", " ", " ", " "};
39
40 G1GCPhaseTimes::G1GCPhaseTimes(uint max_gc_threads) :
41 _max_gc_threads(max_gc_threads),
42 _gc_start_counter(0),
43 _gc_pause_time_ms(0.0)
44 {
45 assert(max_gc_threads > 0, "Must have some GC threads");
46
47 _gc_par_phases[GCWorkerStart] = new WorkerDataArray<double>(max_gc_threads, "GC Worker Start (ms):");
48 _gc_par_phases[ExtRootScan] = new WorkerDataArray<double>(max_gc_threads, "Ext Root Scanning (ms):");
49
50 // Root scanning phases
51 _gc_par_phases[ThreadRoots] = new WorkerDataArray<double>(max_gc_threads, "Thread Roots (ms):");
52 _gc_par_phases[StringTableRoots] = new WorkerDataArray<double>(max_gc_threads, "StringTable Roots (ms):");
53 _gc_par_phases[UniverseRoots] = new WorkerDataArray<double>(max_gc_threads, "Universe Roots (ms):");
54 _gc_par_phases[JNIRoots] = new WorkerDataArray<double>(max_gc_threads, "JNI Handles Roots (ms):");
55 _gc_par_phases[ObjectSynchronizerRoots] = new WorkerDataArray<double>(max_gc_threads, "ObjectSynchronizer Roots (ms):");
56 _gc_par_phases[FlatProfilerRoots] = new WorkerDataArray<double>(max_gc_threads, "FlatProfiler Roots (ms):");
57 _gc_par_phases[ManagementRoots] = new WorkerDataArray<double>(max_gc_threads, "Management Roots (ms):");
58 _gc_par_phases[SystemDictionaryRoots] = new WorkerDataArray<double>(max_gc_threads, "SystemDictionary Roots (ms):");
59 _gc_par_phases[CLDGRoots] = new WorkerDataArray<double>(max_gc_threads, "CLDG Roots (ms):");
60 _gc_par_phases[JVMTIRoots] = new WorkerDataArray<double>(max_gc_threads, "JVMTI Roots (ms):");
61 _gc_par_phases[CMRefRoots] = new WorkerDataArray<double>(max_gc_threads, "CM RefProcessor Roots (ms):");
62 _gc_par_phases[WaitForStrongCLD] = new WorkerDataArray<double>(max_gc_threads, "Wait For Strong CLD (ms):");
63 _gc_par_phases[WeakCLDRoots] = new WorkerDataArray<double>(max_gc_threads, "Weak CLD Roots (ms):");
64 _gc_par_phases[SATBFiltering] = new WorkerDataArray<double>(max_gc_threads, "SATB Filtering (ms):");
65
66 _gc_par_phases[UpdateRS] = new WorkerDataArray<double>(max_gc_threads, "Update RS (ms):");
67 if (G1HotCardCache::default_use_cache()) {
68 _gc_par_phases[ScanHCC] = new WorkerDataArray<double>(max_gc_threads, "Scan HCC (ms):");
69 } else {
70 _gc_par_phases[ScanHCC] = NULL;
71 }
72 _gc_par_phases[ScanRS] = new WorkerDataArray<double>(max_gc_threads, "Scan RS (ms):");
73 _gc_par_phases[CodeRoots] = new WorkerDataArray<double>(max_gc_threads, "Code Root Scanning (ms):");
74 #if INCLUDE_AOT
75 _gc_par_phases[AOTCodeRoots] = new WorkerDataArray<double>(max_gc_threads, "AOT Root Scanning (ms):");
76 #endif
77 _gc_par_phases[ObjCopy] = new WorkerDataArray<double>(max_gc_threads, "Object Copy (ms):");
78 _gc_par_phases[Termination] = new WorkerDataArray<double>(max_gc_threads, "Termination (ms):");
79 _gc_par_phases[GCWorkerTotal] = new WorkerDataArray<double>(max_gc_threads, "GC Worker Total (ms):");
80 _gc_par_phases[GCWorkerEnd] = new WorkerDataArray<double>(max_gc_threads, "GC Worker End (ms):");
81 _gc_par_phases[Other] = new WorkerDataArray<double>(max_gc_threads, "GC Worker Other (ms):");
82
83 _scan_rs_scanned_cards = new WorkerDataArray<size_t>(max_gc_threads, "Scanned Cards:");
84 _gc_par_phases[ScanRS]->link_thread_work_items(_scan_rs_scanned_cards, ScannedCards);
85 _scan_rs_claimed_cards = new WorkerDataArray<size_t>(max_gc_threads, "Claimed Cards:");
86 _gc_par_phases[ScanRS]->link_thread_work_items(_scan_rs_claimed_cards, ClaimedCards);
87 _scan_rs_skipped_cards = new WorkerDataArray<size_t>(max_gc_threads, "Skipped Cards:");
88 _gc_par_phases[ScanRS]->link_thread_work_items(_scan_rs_skipped_cards, SkippedCards);
89
90 _update_rs_processed_buffers = new WorkerDataArray<size_t>(max_gc_threads, "Processed Buffers:");
91 _gc_par_phases[UpdateRS]->link_thread_work_items(_update_rs_processed_buffers);
92
93 _termination_attempts = new WorkerDataArray<size_t>(max_gc_threads, "Termination Attempts:");
94 _gc_par_phases[Termination]->link_thread_work_items(_termination_attempts);
95
96 if (UseStringDeduplication) {
97 _gc_par_phases[StringDedupQueueFixup] = new WorkerDataArray<double>(max_gc_threads, "Queue Fixup (ms):");
98 _gc_par_phases[StringDedupTableFixup] = new WorkerDataArray<double>(max_gc_threads, "Table Fixup (ms):");
99 } else {
100 _gc_par_phases[StringDedupQueueFixup] = NULL;
101 _gc_par_phases[StringDedupTableFixup] = NULL;
102 }
103
104 _gc_par_phases[RedirtyCards] = new WorkerDataArray<double>(max_gc_threads, "Parallel Redirty (ms):");
105 _redirtied_cards = new WorkerDataArray<size_t>(max_gc_threads, "Redirtied Cards:");
106 _gc_par_phases[RedirtyCards]->link_thread_work_items(_redirtied_cards);
107
108 _gc_par_phases[YoungFreeCSet] = new WorkerDataArray<double>(max_gc_threads, "Young Free Collection Set (ms):");
109 _gc_par_phases[NonYoungFreeCSet] = new WorkerDataArray<double>(max_gc_threads, "Non-Young Free Collection Set (ms):");
110
111 _gc_par_phases[PreserveCMReferents] = new WorkerDataArray<double>(max_gc_threads, "Parallel Preserve CM Refs (ms):");
112
113 reset();
114 }
115
116 void G1GCPhaseTimes::reset() {
117 _cur_collection_par_time_ms = 0.0;
118 _cur_collection_code_root_fixup_time_ms = 0.0;
119 _cur_strong_code_root_purge_time_ms = 0.0;
120 _cur_evac_fail_recalc_used = 0.0;
121 _cur_evac_fail_remove_self_forwards = 0.0;
122 _cur_string_dedup_fixup_time_ms = 0.0;
123 _cur_prepare_tlab_time_ms = 0.0;
124 _cur_resize_tlab_time_ms = 0.0;
125 _cur_derived_pointer_table_update_time_ms = 0.0;
126 _cur_clear_ct_time_ms = 0.0;
127 _cur_expand_heap_time_ms = 0.0;
128 _cur_ref_proc_time_ms = 0.0;
129 _cur_ref_enq_time_ms = 0.0;
130 _cur_collection_start_sec = 0.0;
131 _root_region_scan_wait_time_ms = 0.0;
132 _external_accounted_time_ms = 0.0;
133 _recorded_clear_claimed_marks_time_ms = 0.0;
134 _recorded_young_cset_choice_time_ms = 0.0;
135 _recorded_non_young_cset_choice_time_ms = 0.0;
136 _recorded_redirty_logged_cards_time_ms = 0.0;
137 _recorded_preserve_cm_referents_time_ms = 0.0;
138 _recorded_merge_pss_time_ms = 0.0;
139 _recorded_start_new_cset_time_ms = 0.0;
140 _recorded_total_free_cset_time_ms = 0.0;
141 _recorded_serial_free_cset_time_ms = 0.0;
142 _cur_fast_reclaim_humongous_time_ms = 0.0;
143 _cur_fast_reclaim_humongous_register_time_ms = 0.0;
144 _cur_fast_reclaim_humongous_total = 0;
145 _cur_fast_reclaim_humongous_candidates = 0;
146 _cur_fast_reclaim_humongous_reclaimed = 0;
147 _cur_verify_before_time_ms = 0.0;
148 _cur_verify_after_time_ms = 0.0;
149
150 for (int i = 0; i < GCParPhasesSentinel; i++) {
151 if (_gc_par_phases[i] != NULL) {
152 _gc_par_phases[i]->reset();
153 }
154 }
155 }
156
157 void G1GCPhaseTimes::note_gc_start() {
158 _gc_start_counter = os::elapsed_counter();
159 reset();
160 }
161
162 #define ASSERT_PHASE_UNINITIALIZED(phase) \
163 assert(_gc_par_phases[phase]->get(i) == uninitialized, "Phase " #phase " reported for thread that was not started");
164
165 double G1GCPhaseTimes::worker_time(GCParPhases phase, uint worker) {
166 double value = _gc_par_phases[phase]->get(worker);
167 if (value != WorkerDataArray<double>::uninitialized()) {
168 return value;
169 }
170 return 0.0;
171 }
172
173 void G1GCPhaseTimes::note_gc_end() {
174 _gc_pause_time_ms = TimeHelper::counter_to_millis(os::elapsed_counter() - _gc_start_counter);
175
176 double uninitialized = WorkerDataArray<double>::uninitialized();
177
178 for (uint i = 0; i < _max_gc_threads; i++) {
179 double worker_start = _gc_par_phases[GCWorkerStart]->get(i);
180 if (worker_start != uninitialized) {
181 assert(_gc_par_phases[GCWorkerEnd]->get(i) != uninitialized, "Worker started but not ended.");
182 double total_worker_time = _gc_par_phases[GCWorkerEnd]->get(i) - _gc_par_phases[GCWorkerStart]->get(i);
183 record_time_secs(GCWorkerTotal, i , total_worker_time);
184
185 double worker_known_time =
186 worker_time(ExtRootScan, i)
187 + worker_time(SATBFiltering, i)
188 + worker_time(UpdateRS, i)
189 + worker_time(ScanRS, i)
190 + worker_time(CodeRoots, i)
191 + worker_time(ObjCopy, i)
192 + worker_time(Termination, i);
193
194 record_time_secs(Other, i, total_worker_time - worker_known_time);
195 } else {
196 // Make sure all slots are uninitialized since this thread did not seem to have been started
197 ASSERT_PHASE_UNINITIALIZED(GCWorkerEnd);
198 ASSERT_PHASE_UNINITIALIZED(ExtRootScan);
199 ASSERT_PHASE_UNINITIALIZED(SATBFiltering);
200 ASSERT_PHASE_UNINITIALIZED(UpdateRS);
201 ASSERT_PHASE_UNINITIALIZED(ScanRS);
202 ASSERT_PHASE_UNINITIALIZED(CodeRoots);
203 ASSERT_PHASE_UNINITIALIZED(ObjCopy);
204 ASSERT_PHASE_UNINITIALIZED(Termination);
205 }
206 }
207 }
208
209 #undef ASSERT_PHASE_UNINITIALIZED
210
211 // record the time a phase took in seconds
212 void G1GCPhaseTimes::record_time_secs(GCParPhases phase, uint worker_i, double secs) {
213 _gc_par_phases[phase]->set(worker_i, secs);
214 }
215
216 // add a number of seconds to a phase
217 void G1GCPhaseTimes::add_time_secs(GCParPhases phase, uint worker_i, double secs) {
218 _gc_par_phases[phase]->add(worker_i, secs);
219 }
220
221 void G1GCPhaseTimes::record_thread_work_item(GCParPhases phase, uint worker_i, size_t count, uint index) {
222 _gc_par_phases[phase]->set_thread_work_item(worker_i, count, index);
223 }
224
225 // return the average time for a phase in milliseconds
226 double G1GCPhaseTimes::average_time_ms(GCParPhases phase) {
227 return _gc_par_phases[phase]->average() * 1000.0;
228 }
229
230 size_t G1GCPhaseTimes::sum_thread_work_items(GCParPhases phase, uint index) {
231 assert(_gc_par_phases[phase]->thread_work_items(index) != NULL, "No sub count");
232 return _gc_par_phases[phase]->thread_work_items(index)->sum();
233 }
234
235 template <class T>
236 void G1GCPhaseTimes::details(T* phase, const char* indent) const {
237 LogTarget(Trace, gc, phases, task) lt;
238 if (lt.is_enabled()) {
239 LogStream ls(lt);
240 ls.print("%s", indent);
241 phase->print_details_on(&ls);
242 }
243 }
244
245 void G1GCPhaseTimes::log_phase(WorkerDataArray<double>* phase, uint indent, outputStream* out, bool print_sum) const {
246 out->print("%s", Indents[indent]);
247 phase->print_summary_on(out, print_sum);
248 details(phase, Indents[indent]);
249
250 for (uint i = 0; i < phase->MaxThreadWorkItems; i++) {
251 WorkerDataArray<size_t>* work_items = phase->thread_work_items(i);
252 if (work_items != NULL) {
253 out->print("%s", Indents[indent + 1]);
254 work_items->print_summary_on(out, true);
255 details(work_items, Indents[indent + 1]);
256 }
257 }
258 }
259
260 void G1GCPhaseTimes::debug_phase(WorkerDataArray<double>* phase) const {
261 Log(gc, phases) log;
262 if (log.is_level(LogLevel::Debug)) {
263 ResourceMark rm;
264 LogStream ls(log.debug());
265 log_phase(phase, 2, &ls, true);
266 }
267 }
268
269 void G1GCPhaseTimes::trace_phase(WorkerDataArray<double>* phase, bool print_sum) const {
270 LogTarget(Trace, gc, phases) lt;
271 if (lt.is_enabled()) {
272 LogStream ls(lt);
273 log_phase(phase, 3, &ls, print_sum);
274 }
275 }
276
277 #define TIME_FORMAT "%.1lfms"
278
279 void G1GCPhaseTimes::info_time(const char* name, double value) const {
280 log_info(gc, phases)("%s%s: " TIME_FORMAT, Indents[1], name, value);
281 }
282
283 void G1GCPhaseTimes::debug_time(const char* name, double value) const {
284 log_debug(gc, phases)("%s%s: " TIME_FORMAT, Indents[2], name, value);
285 }
286
287 void G1GCPhaseTimes::trace_time(const char* name, double value) const {
288 log_trace(gc, phases)("%s%s: " TIME_FORMAT, Indents[3], name, value);
289 }
290
291 void G1GCPhaseTimes::trace_count(const char* name, size_t value) const {
292 log_trace(gc, phases)("%s%s: " SIZE_FORMAT, Indents[3], name, value);
293 }
294
295 double G1GCPhaseTimes::print_pre_evacuate_collection_set() const {
296 const double sum_ms = _root_region_scan_wait_time_ms +
297 _recorded_young_cset_choice_time_ms +
298 _recorded_non_young_cset_choice_time_ms +
299 _cur_fast_reclaim_humongous_register_time_ms +
300 _recorded_clear_claimed_marks_time_ms;
301
302 info_time("Pre Evacuate Collection Set", sum_ms);
303
304 if (_root_region_scan_wait_time_ms > 0.0) {
305 debug_time("Root Region Scan Waiting", _root_region_scan_wait_time_ms);
306 }
307 debug_time("Prepare TLABs", _cur_prepare_tlab_time_ms);
308 debug_time("Choose Collection Set", (_recorded_young_cset_choice_time_ms + _recorded_non_young_cset_choice_time_ms));
309 if (G1EagerReclaimHumongousObjects) {
310 debug_time("Humongous Register", _cur_fast_reclaim_humongous_register_time_ms);
311 trace_count("Humongous Total", _cur_fast_reclaim_humongous_total);
312 trace_count("Humongous Candidate", _cur_fast_reclaim_humongous_candidates);
313 }
314
315 if (_recorded_clear_claimed_marks_time_ms > 0.0) {
316 debug_time("Clear Claimed Marks", _recorded_clear_claimed_marks_time_ms);
317 }
318 return sum_ms;
319 }
320
321 double G1GCPhaseTimes::print_evacuate_collection_set() const {
322 const double sum_ms = _cur_collection_par_time_ms;
323
324 info_time("Evacuate Collection Set", sum_ms);
325
326 trace_phase(_gc_par_phases[GCWorkerStart], false);
327 debug_phase(_gc_par_phases[ExtRootScan]);
328 for (int i = ThreadRoots; i <= SATBFiltering; i++) {
329 trace_phase(_gc_par_phases[i]);
330 }
331 debug_phase(_gc_par_phases[UpdateRS]);
332 if (G1HotCardCache::default_use_cache()) {
333 trace_phase(_gc_par_phases[ScanHCC]);
334 }
335 debug_phase(_gc_par_phases[ScanRS]);
336 debug_phase(_gc_par_phases[CodeRoots]);
337 #if INCLUDE_AOT
338 debug_phase(_gc_par_phases[AOTCodeRoots]);
339 #endif
340 debug_phase(_gc_par_phases[ObjCopy]);
341 debug_phase(_gc_par_phases[Termination]);
342 debug_phase(_gc_par_phases[Other]);
343 debug_phase(_gc_par_phases[GCWorkerTotal]);
344 trace_phase(_gc_par_phases[GCWorkerEnd], false);
345
346 return sum_ms;
347 }
348
349 double G1GCPhaseTimes::print_post_evacuate_collection_set() const {
350 const double evac_fail_handling = _cur_evac_fail_recalc_used +
351 _cur_evac_fail_remove_self_forwards;
352 const double sum_ms = evac_fail_handling +
353 _cur_collection_code_root_fixup_time_ms +
354 _recorded_preserve_cm_referents_time_ms +
355 _cur_ref_proc_time_ms +
356 _cur_ref_enq_time_ms +
357 _cur_clear_ct_time_ms +
358 _recorded_merge_pss_time_ms +
359 _cur_strong_code_root_purge_time_ms +
360 _recorded_redirty_logged_cards_time_ms +
361 _recorded_total_free_cset_time_ms +
362 _cur_fast_reclaim_humongous_time_ms +
363 _cur_expand_heap_time_ms +
364 _cur_string_dedup_fixup_time_ms;
365
366 info_time("Post Evacuate Collection Set", sum_ms);
367
368 debug_time("Code Roots Fixup", _cur_collection_code_root_fixup_time_ms);
369
370 debug_time("Preserve CM Refs", _recorded_preserve_cm_referents_time_ms);
371 trace_phase(_gc_par_phases[PreserveCMReferents]);
372
373 debug_time("Reference Processing", _cur_ref_proc_time_ms);
374
375 if (G1StringDedup::is_enabled()) {
376 debug_time("String Dedup Fixup", _cur_string_dedup_fixup_time_ms);
377 debug_phase(_gc_par_phases[StringDedupQueueFixup]);
378 debug_phase(_gc_par_phases[StringDedupTableFixup]);
379 }
380
381 debug_time("Clear Card Table", _cur_clear_ct_time_ms);
382
383 if (G1CollectedHeap::heap()->evacuation_failed()) {
384 debug_time("Evacuation Failure", evac_fail_handling);
385 trace_time("Recalculate Used", _cur_evac_fail_recalc_used);
386 trace_time("Remove Self Forwards",_cur_evac_fail_remove_self_forwards);
387 }
388
389 debug_time("Reference Enqueuing", _cur_ref_enq_time_ms);
390
391 debug_time("Merge Per-Thread State", _recorded_merge_pss_time_ms);
392 debug_time("Code Roots Purge", _cur_strong_code_root_purge_time_ms);
393
394 debug_time("Redirty Cards", _recorded_redirty_logged_cards_time_ms);
395 trace_phase(_gc_par_phases[RedirtyCards]);
396 #if defined(COMPILER2) || INCLUDE_JVMCI
397 debug_time("DerivedPointerTable Update", _cur_derived_pointer_table_update_time_ms);
398 #endif
399
400 debug_time("Free Collection Set", _recorded_total_free_cset_time_ms);
401 trace_time("Free Collection Set Serial", _recorded_serial_free_cset_time_ms);
402 trace_phase(_gc_par_phases[YoungFreeCSet]);
403 trace_phase(_gc_par_phases[NonYoungFreeCSet]);
404
405 if (G1EagerReclaimHumongousObjects) {
406 debug_time("Humongous Reclaim", _cur_fast_reclaim_humongous_time_ms);
407 trace_count("Humongous Reclaimed", _cur_fast_reclaim_humongous_reclaimed);
408 }
409 debug_time("Start New Collection Set", _recorded_start_new_cset_time_ms);
410 if (UseTLAB && ResizeTLAB) {
411 debug_time("Resize TLABs", _cur_resize_tlab_time_ms);
412 }
413 debug_time("Expand Heap After Collection", _cur_expand_heap_time_ms);
414
415
416 return sum_ms;
417 }
418
419 void G1GCPhaseTimes::print_other(double accounted_ms) const {
420 info_time("Other", _gc_pause_time_ms - accounted_ms);
421 }
422
423 void G1GCPhaseTimes::print() {
424 note_gc_end();
425
426 if (_cur_verify_before_time_ms > 0.0) {
427 debug_time("Verify Before", _cur_verify_before_time_ms);
428 }
429
430 double accounted_ms = 0.0;
431 accounted_ms += print_pre_evacuate_collection_set();
432 accounted_ms += print_evacuate_collection_set();
433 accounted_ms += print_post_evacuate_collection_set();
434 print_other(accounted_ms);
435
436 if (_cur_verify_after_time_ms > 0.0) {
437 debug_time("Verify After", _cur_verify_after_time_ms);
438 }
439 }
440
441 G1GCParPhaseTimesTracker::G1GCParPhaseTimesTracker(G1GCPhaseTimes* phase_times, G1GCPhaseTimes::GCParPhases phase, uint worker_id) :
442 _phase_times(phase_times), _phase(phase), _worker_id(worker_id) {
443 if (_phase_times != NULL) {
444 _start_time = os::elapsedTime();
445 }
446 }
447
448 G1GCParPhaseTimesTracker::~G1GCParPhaseTimesTracker() {
449 if (_phase_times != NULL) {
450 _phase_times->record_time_secs(_phase, _worker_id, os::elapsedTime() - _start_time);
451 }
452 }
453