1 /*
2 * Copyright (c) 2013, 2015, 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/os.hpp"
34
35 // Helper class for avoiding interleaved logging
36 class LineBuffer: public StackObj {
37
38 private:
39 static const int BUFFER_LEN = 1024;
40 static const int INDENT_CHARS = 3;
41 char _buffer[BUFFER_LEN];
42 int _indent_level;
43 int _cur;
44
45 void vappend(const char* format, va_list ap) ATTRIBUTE_PRINTF(2, 0) {
46 int res = vsnprintf(&_buffer[_cur], BUFFER_LEN - _cur, format, ap);
47 if (res != -1) {
48 _cur += res;
49 } else {
50 DEBUG_ONLY(warning("buffer too small in LineBuffer");)
51 _buffer[BUFFER_LEN -1] = 0;
52 _cur = BUFFER_LEN; // vsnprintf above should not add to _buffer if we are called again
53 }
54 }
55
56 public:
57 explicit LineBuffer(int indent_level): _indent_level(indent_level), _cur(0) {
58 for (; (_cur < BUFFER_LEN && _cur < (_indent_level * INDENT_CHARS)); _cur++) {
59 _buffer[_cur] = ' ';
60 }
61 }
62
63 #ifndef PRODUCT
64 ~LineBuffer() {
65 assert(_cur == _indent_level * INDENT_CHARS, "pending data in buffer - append_and_print_cr() not called?");
66 }
67 #endif
68
69 void append(const char* format, ...) ATTRIBUTE_PRINTF(2, 3) {
70 va_list ap;
71 va_start(ap, format);
72 vappend(format, ap);
73 va_end(ap);
74 }
75
76 const char* to_string() {
77 _cur = _indent_level * INDENT_CHARS;
78 return _buffer;
79 }
80 };
81
82 static const char* Indents[4] = {"", " ", " ", " "};
83
84 G1GCPhaseTimes::G1GCPhaseTimes(uint max_gc_threads) :
85 _max_gc_threads(max_gc_threads)
86 {
87 assert(max_gc_threads > 0, "Must have some GC threads");
88
89 _gc_par_phases[GCWorkerStart] = new WorkerDataArray<double>(max_gc_threads, "GC Worker Start:", false, 2);
90 _gc_par_phases[ExtRootScan] = new WorkerDataArray<double>(max_gc_threads, "Ext Root Scanning:", true, 2);
91
92 // Root scanning phases
93 _gc_par_phases[ThreadRoots] = new WorkerDataArray<double>(max_gc_threads, "Thread Roots:", true, 3);
94 _gc_par_phases[StringTableRoots] = new WorkerDataArray<double>(max_gc_threads, "StringTable Roots:", true, 3);
95 _gc_par_phases[UniverseRoots] = new WorkerDataArray<double>(max_gc_threads, "Universe Roots:", true, 3);
96 _gc_par_phases[JNIRoots] = new WorkerDataArray<double>(max_gc_threads, "JNI Handles Roots:", true, 3);
97 _gc_par_phases[ObjectSynchronizerRoots] = new WorkerDataArray<double>(max_gc_threads, "ObjectSynchronizer Roots:", true, 3);
98 _gc_par_phases[FlatProfilerRoots] = new WorkerDataArray<double>(max_gc_threads, "FlatProfiler Roots:", true, 3);
99 _gc_par_phases[ManagementRoots] = new WorkerDataArray<double>(max_gc_threads, "Management Roots:", true, 3);
100 _gc_par_phases[SystemDictionaryRoots] = new WorkerDataArray<double>(max_gc_threads, "SystemDictionary Roots:", true, 3);
101 _gc_par_phases[CLDGRoots] = new WorkerDataArray<double>(max_gc_threads, "CLDG Roots:", true, 3);
102 _gc_par_phases[JVMTIRoots] = new WorkerDataArray<double>(max_gc_threads, "JVMTI Roots:", true, 3);
103 _gc_par_phases[CMRefRoots] = new WorkerDataArray<double>(max_gc_threads, "CM RefProcessor Roots:", true, 3);
104 _gc_par_phases[WaitForStrongCLD] = new WorkerDataArray<double>(max_gc_threads, "Wait For Strong CLD:", true, 3);
105 _gc_par_phases[WeakCLDRoots] = new WorkerDataArray<double>(max_gc_threads, "Weak CLD Roots:", true, 3);
106 _gc_par_phases[SATBFiltering] = new WorkerDataArray<double>(max_gc_threads, "SATB Filtering:", true, 3);
107
108 _gc_par_phases[UpdateRS] = new WorkerDataArray<double>(max_gc_threads, "Update RS:", true, 2);
109 _gc_par_phases[ScanHCC] = new WorkerDataArray<double>(max_gc_threads, "Scan HCC:", true, 3);
110 _gc_par_phases[ScanHCC]->set_enabled(ConcurrentG1Refine::hot_card_cache_enabled());
111 _gc_par_phases[ScanRS] = new WorkerDataArray<double>(max_gc_threads, "Scan RS:", true, 2);
112 _gc_par_phases[CodeRoots] = new WorkerDataArray<double>(max_gc_threads, "Code Root Scanning:", true, 2);
113 _gc_par_phases[ObjCopy] = new WorkerDataArray<double>(max_gc_threads, "Object Copy:", true, 2);
114 _gc_par_phases[Termination] = new WorkerDataArray<double>(max_gc_threads, "Termination:", true, 2);
115 _gc_par_phases[GCWorkerTotal] = new WorkerDataArray<double>(max_gc_threads, "GC Worker Total:", true, 2);
116 _gc_par_phases[GCWorkerEnd] = new WorkerDataArray<double>(max_gc_threads, "GC Worker End:", false, 2);
117 _gc_par_phases[Other] = new WorkerDataArray<double>(max_gc_threads, "GC Worker Other:", true, 2);
118
119 _update_rs_processed_buffers = new WorkerDataArray<size_t>(max_gc_threads, "Processed Buffers:", true, 3);
120 _gc_par_phases[UpdateRS]->link_thread_work_items(_update_rs_processed_buffers);
121
122 _termination_attempts = new WorkerDataArray<size_t>(max_gc_threads, "Termination Attempts:", true, 3);
123 _gc_par_phases[Termination]->link_thread_work_items(_termination_attempts);
124
125 _gc_par_phases[StringDedupQueueFixup] = new WorkerDataArray<double>(max_gc_threads, "Queue Fixup:", true, 2);
126 _gc_par_phases[StringDedupTableFixup] = new WorkerDataArray<double>(max_gc_threads, "Table Fixup:", true, 2);
127
128 _gc_par_phases[RedirtyCards] = new WorkerDataArray<double>(max_gc_threads, "Parallel Redirty", true, 3);
129 _redirtied_cards = new WorkerDataArray<size_t>(max_gc_threads, "Redirtied Cards:", true, 3);
130 _gc_par_phases[RedirtyCards]->link_thread_work_items(_redirtied_cards);
131 }
132
133 void G1GCPhaseTimes::note_gc_start(uint active_gc_threads) {
134 assert(active_gc_threads > 0, "The number of threads must be > 0");
135 assert(active_gc_threads <= _max_gc_threads, "The number of active threads must be <= the max number of threads");
136 _active_gc_threads = active_gc_threads;
137 _cur_expand_heap_time_ms = 0.0;
138 _external_accounted_time_ms = 0.0;
139
140 for (int i = 0; i < GCParPhasesSentinel; i++) {
141 _gc_par_phases[i]->reset();
142 }
143
144 _gc_par_phases[StringDedupQueueFixup]->set_enabled(G1StringDedup::is_enabled());
145 _gc_par_phases[StringDedupTableFixup]->set_enabled(G1StringDedup::is_enabled());
146 }
147
148 void G1GCPhaseTimes::note_gc_end() {
149 for (uint i = 0; i < _active_gc_threads; i++) {
150 double worker_time = _gc_par_phases[GCWorkerEnd]->get(i) - _gc_par_phases[GCWorkerStart]->get(i);
151 record_time_secs(GCWorkerTotal, i , worker_time);
152
153 double worker_known_time =
154 _gc_par_phases[ExtRootScan]->get(i) +
155 _gc_par_phases[SATBFiltering]->get(i) +
156 _gc_par_phases[UpdateRS]->get(i) +
157 _gc_par_phases[ScanRS]->get(i) +
158 _gc_par_phases[CodeRoots]->get(i) +
159 _gc_par_phases[ObjCopy]->get(i) +
160 _gc_par_phases[Termination]->get(i);
161
162 record_time_secs(Other, i, worker_time - worker_known_time);
163 }
164
165 for (int i = 0; i < GCParPhasesSentinel; i++) {
166 _gc_par_phases[i]->verify(_active_gc_threads);
167 }
168 }
169
170 void G1GCPhaseTimes::print_stats(const char* indent, const char* str, double value) {
171 log_debug(gc, phases)("%s%s: %.1lf ms", indent, str, value);
172 }
173
174 double G1GCPhaseTimes::accounted_time_ms() {
175 // First subtract any externally accounted time
176 double misc_time_ms = _external_accounted_time_ms;
177
178 // Subtract the root region scanning wait time. It's initialized to
179 // zero at the start of the pause.
180 misc_time_ms += _root_region_scan_wait_time_ms;
181
182 misc_time_ms += _cur_collection_par_time_ms;
183
184 // Now subtract the time taken to fix up roots in generated code
185 misc_time_ms += _cur_collection_code_root_fixup_time_ms;
186
187 // Strong code root purge time
188 misc_time_ms += _cur_strong_code_root_purge_time_ms;
189
190 if (G1StringDedup::is_enabled()) {
191 // String dedup fixup time
192 misc_time_ms += _cur_string_dedup_fixup_time_ms;
193 }
194
195 // Subtract the time taken to clean the card table from the
196 // current value of "other time"
197 misc_time_ms += _cur_clear_ct_time_ms;
198
199 // Remove expand heap time from "other time"
200 misc_time_ms += _cur_expand_heap_time_ms;
201
202 return misc_time_ms;
203 }
204
205 // record the time a phase took in seconds
206 void G1GCPhaseTimes::record_time_secs(GCParPhases phase, uint worker_i, double secs) {
207 _gc_par_phases[phase]->set(worker_i, secs);
208 }
209
210 // add a number of seconds to a phase
211 void G1GCPhaseTimes::add_time_secs(GCParPhases phase, uint worker_i, double secs) {
212 _gc_par_phases[phase]->add(worker_i, secs);
213 }
214
215 void G1GCPhaseTimes::record_thread_work_item(GCParPhases phase, uint worker_i, size_t count) {
216 _gc_par_phases[phase]->set_thread_work_item(worker_i, count);
217 }
218
219 // return the average time for a phase in milliseconds
220 double G1GCPhaseTimes::average_time_ms(GCParPhases phase) {
221 return _gc_par_phases[phase]->average(_active_gc_threads) * 1000.0;
222 }
223
224 double G1GCPhaseTimes::get_time_ms(GCParPhases phase, uint worker_i) {
225 return _gc_par_phases[phase]->get(worker_i) * 1000.0;
226 }
227
228 double G1GCPhaseTimes::sum_time_ms(GCParPhases phase) {
229 return _gc_par_phases[phase]->sum(_active_gc_threads) * 1000.0;
230 }
231
232 double G1GCPhaseTimes::min_time_ms(GCParPhases phase) {
233 return _gc_par_phases[phase]->minimum(_active_gc_threads) * 1000.0;
234 }
235
236 double G1GCPhaseTimes::max_time_ms(GCParPhases phase) {
237 return _gc_par_phases[phase]->maximum(_active_gc_threads) * 1000.0;
238 }
239
240 size_t G1GCPhaseTimes::get_thread_work_item(GCParPhases phase, uint worker_i) {
241 assert(_gc_par_phases[phase]->thread_work_items() != NULL, "No sub count");
242 return _gc_par_phases[phase]->thread_work_items()->get(worker_i);
243 }
244
245 size_t G1GCPhaseTimes::sum_thread_work_items(GCParPhases phase) {
246 assert(_gc_par_phases[phase]->thread_work_items() != NULL, "No sub count");
247 return _gc_par_phases[phase]->thread_work_items()->sum(_active_gc_threads);
248 }
249
250 double G1GCPhaseTimes::average_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()->average(_active_gc_threads);
253 }
254
255 size_t G1GCPhaseTimes::min_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()->minimum(_active_gc_threads);
258 }
259
260 size_t G1GCPhaseTimes::max_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()->maximum(_active_gc_threads);
263 }
264
265 class G1GCParPhasePrinter : public StackObj {
266 G1GCPhaseTimes* _phase_times;
267 public:
268 G1GCParPhasePrinter(G1GCPhaseTimes* phase_times) : _phase_times(phase_times) {}
269
270 void print(G1GCPhaseTimes::GCParPhases phase_id) {
271 WorkerDataArray<double>* phase = _phase_times->_gc_par_phases[phase_id];
272
273 if (phase->_length == 1) {
274 print_single_length(phase_id, phase);
275 } else {
276 print_multi_length(phase_id, phase);
277 }
278 }
279
280
281 private:
282 void print_single_length(G1GCPhaseTimes::GCParPhases phase_id, WorkerDataArray<double>* phase) {
283 // No need for min, max, average and sum for only one worker
284 log_debug(gc, phases)("%s%s: %.1lf", Indents[phase->_indent_level], phase->_title, _phase_times->get_time_ms(phase_id, 0));
285
286 WorkerDataArray<size_t>* work_items = phase->_thread_work_items;
287 if (work_items != NULL) {
288 log_debug(gc, phases)("%s%s: " SIZE_FORMAT, Indents[work_items->_indent_level], work_items->_title, _phase_times->sum_thread_work_items(phase_id));
289 }
290 }
291
292 void print_time_values(const char* indent, G1GCPhaseTimes::GCParPhases phase_id) {
293 if (log_is_enabled(Trace, gc)) {
294 LineBuffer buf(0);
295 uint active_length = _phase_times->_active_gc_threads;
296 for (uint i = 0; i < active_length; ++i) {
297 buf.append(" %4.1lf", _phase_times->get_time_ms(phase_id, i));
298 }
299 const char* line = buf.to_string();
300 log_trace(gc, phases)("%s%-25s%s", indent, "", line);
301 }
302 }
303
304 void print_count_values(const char* indent, G1GCPhaseTimes::GCParPhases phase_id, WorkerDataArray<size_t>* thread_work_items) {
305 if (log_is_enabled(Trace, gc)) {
306 LineBuffer buf(0);
307 uint active_length = _phase_times->_active_gc_threads;
308 for (uint i = 0; i < active_length; ++i) {
309 buf.append(" " SIZE_FORMAT, _phase_times->get_thread_work_item(phase_id, i));
310 }
311 const char* line = buf.to_string();
312 log_trace(gc, phases)("%s%-25s%s", indent, "", line);
313 }
314 }
315
316 void print_thread_work_items(G1GCPhaseTimes::GCParPhases phase_id, WorkerDataArray<size_t>* thread_work_items) {
317 const char* indent = Indents[thread_work_items->_indent_level];
318
319 assert(thread_work_items->_print_sum, "%s does not have print sum true even though it is a count", thread_work_items->_title);
320
321 log_debug(gc, phases)("%s%-25s Min: " SIZE_FORMAT ", Avg: %4.1lf, Max: " SIZE_FORMAT ", Diff: " SIZE_FORMAT ", Sum: " SIZE_FORMAT,
322 indent, thread_work_items->_title,
323 _phase_times->min_thread_work_items(phase_id), _phase_times->average_thread_work_items(phase_id), _phase_times->max_thread_work_items(phase_id),
324 _phase_times->max_thread_work_items(phase_id) - _phase_times->min_thread_work_items(phase_id), _phase_times->sum_thread_work_items(phase_id));
325
326 print_count_values(indent, phase_id, thread_work_items);
327 }
328
329 void print_multi_length(G1GCPhaseTimes::GCParPhases phase_id, WorkerDataArray<double>* phase) {
330 const char* indent = Indents[phase->_indent_level];
331
332 if (phase->_print_sum) {
333 log_debug(gc, phases)("%s%-25s Min: %4.1lf, Avg: %4.1lf, Max: %4.1lf, Diff: %4.1lf, Sum: %4.1lf",
334 indent, phase->_title,
335 _phase_times->min_time_ms(phase_id), _phase_times->average_time_ms(phase_id), _phase_times->max_time_ms(phase_id),
336 _phase_times->max_time_ms(phase_id) - _phase_times->min_time_ms(phase_id), _phase_times->sum_time_ms(phase_id));
337 } else {
338 log_debug(gc, phases)("%s%-25s Min: %4.1lf, Avg: %4.1lf, Max: %4.1lf, Diff: %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));
342 }
343
344 print_time_values(indent, phase_id);
345
346 if (phase->_thread_work_items != NULL) {
347 print_thread_work_items(phase_id, phase->_thread_work_items);
348 }
349 }
350 };
351
352 void G1GCPhaseTimes::print(double pause_time_sec) {
353 note_gc_end();
354
355 G1GCParPhasePrinter par_phase_printer(this);
356
357 if (_root_region_scan_wait_time_ms > 0.0) {
358 print_stats(Indents[1], "Root Region Scan Waiting", _root_region_scan_wait_time_ms);
359 }
360
361 print_stats(Indents[1], "Parallel Time", _cur_collection_par_time_ms);
362 for (int i = 0; i <= GCMainParPhasesLast; i++) {
363 par_phase_printer.print((GCParPhases) i);
364 }
365
366 print_stats(Indents[1], "Code Root Fixup", _cur_collection_code_root_fixup_time_ms);
367 print_stats(Indents[1], "Code Root Purge", _cur_strong_code_root_purge_time_ms);
368 if (G1StringDedup::is_enabled()) {
369 print_stats(Indents[1], "String Dedup Fixup", _cur_string_dedup_fixup_time_ms);
370 for (int i = StringDedupPhasesFirst; i <= StringDedupPhasesLast; i++) {
371 par_phase_printer.print((GCParPhases) i);
372 }
373 }
374 print_stats(Indents[1], "Clear CT", _cur_clear_ct_time_ms);
375 print_stats(Indents[1], "Expand Heap After Collection", _cur_expand_heap_time_ms);
376 double misc_time_ms = pause_time_sec * MILLIUNITS - accounted_time_ms();
377 print_stats(Indents[1], "Other", misc_time_ms);
378 if (_cur_verify_before_time_ms > 0.0) {
379 print_stats(Indents[2], "Verify Before", _cur_verify_before_time_ms);
380 }
381 if (G1CollectedHeap::heap()->evacuation_failed()) {
382 double evac_fail_handling = _cur_evac_fail_recalc_used + _cur_evac_fail_remove_self_forwards +
383 _cur_evac_fail_restore_remsets;
384 print_stats(Indents[2], "Evacuation Failure", evac_fail_handling);
385 log_trace(gc, phases)("%sRecalculate Used: %.1lf ms", Indents[3], _cur_evac_fail_recalc_used);
386 log_trace(gc, phases)("%sRemove Self Forwards: %.1lf ms", Indents[3], _cur_evac_fail_remove_self_forwards);
387 log_trace(gc, phases)("%sRestore RemSet: %.1lf ms", Indents[3], _cur_evac_fail_restore_remsets);
388 }
389 print_stats(Indents[2], "Choose CSet",
390 (_recorded_young_cset_choice_time_ms +
391 _recorded_non_young_cset_choice_time_ms));
392 print_stats(Indents[2], "Ref Proc", _cur_ref_proc_time_ms);
393 print_stats(Indents[2], "Ref Enq", _cur_ref_enq_time_ms);
394 print_stats(Indents[2], "Redirty Cards", _recorded_redirty_logged_cards_time_ms);
395 par_phase_printer.print(RedirtyCards);
396 if (G1EagerReclaimHumongousObjects) {
397 print_stats(Indents[2], "Humongous Register", _cur_fast_reclaim_humongous_register_time_ms);
398
399 log_trace(gc, phases)("%sHumongous Total: " SIZE_FORMAT, Indents[3], _cur_fast_reclaim_humongous_total);
400 log_trace(gc, phases)("%sHumongous Candidate: " SIZE_FORMAT, Indents[3], _cur_fast_reclaim_humongous_candidates);
401 print_stats(Indents[2], "Humongous Reclaim", _cur_fast_reclaim_humongous_time_ms);
402 log_trace(gc, phases)("%sHumongous Reclaimed: " SIZE_FORMAT, Indents[3], _cur_fast_reclaim_humongous_reclaimed);
403 }
404 print_stats(Indents[2], "Free CSet",
405 (_recorded_young_free_cset_time_ms +
406 _recorded_non_young_free_cset_time_ms));
407 log_trace(gc, phases)("%sYoung Free CSet: %.1lf ms", Indents[3], _recorded_young_free_cset_time_ms);
408 log_trace(gc, phases)("%sNon-Young Free CSet: %.1lf ms", Indents[3], _recorded_non_young_free_cset_time_ms);
409 if (_cur_verify_after_time_ms > 0.0) {
410 print_stats(Indents[2], "Verify After", _cur_verify_after_time_ms);
411 }
412 }
413
414 G1GCParPhaseTimesTracker::G1GCParPhaseTimesTracker(G1GCPhaseTimes* phase_times, G1GCPhaseTimes::GCParPhases phase, uint worker_id) :
415 _phase_times(phase_times), _phase(phase), _worker_id(worker_id) {
416 if (_phase_times != NULL) {
417 _start_time = os::elapsedTime();
418 }
419 }
420
421 G1GCParPhaseTimesTracker::~G1GCParPhaseTimesTracker() {
422 if (_phase_times != NULL) {
423 _phase_times->record_time_secs(_phase, _worker_id, os::elapsedTime() - _start_time);
424 }
425 }
426