1 /*
2 * Copyright (c) 2013, 2014 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
26 #include "precompiled.hpp"
27 #include "gc_implementation/g1/g1CollectedHeap.inline.hpp"
28 #include "gc_implementation/g1/g1GCPhaseTimes.hpp"
29 #include "gc_implementation/g1/g1Log.hpp"
30 #include "gc_implementation/g1/g1StringDedup.hpp"
31
32 // Helper class for avoiding interleaved logging
33 class LineBuffer: public StackObj {
34
35 private:
36 static const int BUFFER_LEN = 1024;
37 static const int INDENT_CHARS = 3;
38 char _buffer[BUFFER_LEN];
39 int _indent_level;
40 int _cur;
41
42 void vappend(const char* format, va_list ap) ATTRIBUTE_PRINTF(2, 0) {
43 int res = vsnprintf(&_buffer[_cur], BUFFER_LEN - _cur, format, ap);
44 if (res != -1) {
45 _cur += res;
46 } else {
47 DEBUG_ONLY(warning("buffer too small in LineBuffer");)
48 _buffer[BUFFER_LEN -1] = 0;
49 _cur = BUFFER_LEN; // vsnprintf above should not add to _buffer if we are called again
50 }
51 }
52
53 public:
54 explicit LineBuffer(int indent_level): _indent_level(indent_level), _cur(0) {
55 for (; (_cur < BUFFER_LEN && _cur < (_indent_level * INDENT_CHARS)); _cur++) {
56 _buffer[_cur] = ' ';
57 }
58 }
59
60 #ifndef PRODUCT
61 ~LineBuffer() {
62 assert(_cur == _indent_level * INDENT_CHARS, "pending data in buffer - append_and_print_cr() not called?");
63 }
64 #endif
65
66 void append(const char* format, ...) ATTRIBUTE_PRINTF(2, 3) {
67 va_list ap;
68 va_start(ap, format);
69 vappend(format, ap);
70 va_end(ap);
71 }
72
73 void append_and_print_cr(const char* format, ...) ATTRIBUTE_PRINTF(2, 3) {
74 va_list ap;
75 va_start(ap, format);
76 vappend(format, ap);
77 va_end(ap);
78 gclog_or_tty->print_cr("%s", _buffer);
79 _cur = _indent_level * INDENT_CHARS;
80 }
81 };
82
83 PRAGMA_DIAG_PUSH
84 PRAGMA_FORMAT_NONLITERAL_IGNORED
85 template <class T>
86 void WorkerDataArray<T>::print(int level, const char* title) {
87 if (_length == 1) {
88 // No need for min, max, average and sum for only one worker
89 LineBuffer buf(level);
90 buf.append("[%s: ", title);
91 buf.append(_print_format, _data[0]);
92 buf.append_and_print_cr("]");
93 return;
94 }
95
96 T min = _data[0];
97 T max = _data[0];
98 T sum = 0;
99
100 LineBuffer buf(level);
101 buf.append("[%s:", title);
102 for (uint i = 0; i < _length; ++i) {
103 T val = _data[i];
104 min = MIN2(val, min);
105 max = MAX2(val, max);
106 sum += val;
107 if (G1Log::finest()) {
108 buf.append(" ");
109 buf.append(_print_format, val);
110 }
111 }
112
113 if (G1Log::finest()) {
114 buf.append_and_print_cr("%s", "");
115 }
116
117 double avg = (double)sum / (double)_length;
118 buf.append(" Min: ");
119 buf.append(_print_format, min);
120 buf.append(", Avg: ");
121 buf.append("%.1lf", avg); // Always print average as a double
122 buf.append(", Max: ");
123 buf.append(_print_format, max);
124 buf.append(", Diff: ");
125 buf.append(_print_format, max - min);
126 if (_print_sum) {
127 // for things like the start and end times the sum is not
128 // that relevant
129 buf.append(", Sum: ");
130 buf.append(_print_format, sum);
131 }
132 buf.append_and_print_cr("]");
133 }
134 PRAGMA_DIAG_POP
135
136 #ifndef PRODUCT
137
138 template <> const int WorkerDataArray<int>::_uninitialized = -1;
139 template <> const double WorkerDataArray<double>::_uninitialized = -1.0;
140 template <> const size_t WorkerDataArray<size_t>::_uninitialized = (size_t)-1;
141
142 template <class T>
143 void WorkerDataArray<T>::reset() {
144 for (uint i = 0; i < _length; i++) {
145 _data[i] = (T)_uninitialized;
146 }
147 }
148
149 template <class T>
150 void WorkerDataArray<T>::verify() {
151 for (uint i = 0; i < _length; i++) {
152 assert(_data[i] != _uninitialized,
153 err_msg("Invalid data for worker " UINT32_FORMAT ", data: %lf, uninitialized: %lf",
154 i, (double)_data[i], (double)_uninitialized));
155 }
156 }
157
158 #endif
159
160 G1GCPhaseTimes::G1GCPhaseTimes(uint max_gc_threads) :
161 _max_gc_threads(max_gc_threads),
162 _last_gc_worker_start_times_ms(_max_gc_threads, "%.1lf", false),
163 _last_ext_root_scan_times_ms(_max_gc_threads, "%.1lf"),
164 _last_satb_filtering_times_ms(_max_gc_threads, "%.1lf"),
165 _last_update_rs_times_ms(_max_gc_threads, "%.1lf"),
166 _last_update_rs_processed_buffers(_max_gc_threads, "%d"),
167 _last_scan_rs_times_ms(_max_gc_threads, "%.1lf"),
168 _last_strong_code_root_scan_times_ms(_max_gc_threads, "%.1lf"),
169 _last_strong_code_root_mark_times_ms(_max_gc_threads, "%.1lf"),
170 _last_obj_copy_times_ms(_max_gc_threads, "%.1lf"),
171 _last_termination_times_ms(_max_gc_threads, "%.1lf"),
172 _last_termination_attempts(_max_gc_threads, SIZE_FORMAT),
173 _last_gc_worker_end_times_ms(_max_gc_threads, "%.1lf", false),
174 _last_gc_worker_times_ms(_max_gc_threads, "%.1lf"),
175 _last_gc_worker_other_times_ms(_max_gc_threads, "%.1lf"),
176 _cur_string_dedup_queue_fixup_worker_times_ms(_max_gc_threads, "%.1lf"),
177 _cur_string_dedup_table_fixup_worker_times_ms(_max_gc_threads, "%.1lf")
178 {
179 assert(max_gc_threads > 0, "Must have some GC threads");
180 }
181
182 void G1GCPhaseTimes::note_gc_start(uint active_gc_threads) {
183 assert(active_gc_threads > 0, "The number of threads must be > 0");
184 assert(active_gc_threads <= _max_gc_threads, "The number of active threads must be <= the max nubmer of threads");
185 _active_gc_threads = active_gc_threads;
186
187 _last_gc_worker_start_times_ms.reset();
188 _last_ext_root_scan_times_ms.reset();
189 _last_satb_filtering_times_ms.reset();
190 _last_update_rs_times_ms.reset();
191 _last_update_rs_processed_buffers.reset();
192 _last_scan_rs_times_ms.reset();
193 _last_strong_code_root_scan_times_ms.reset();
194 _last_strong_code_root_mark_times_ms.reset();
195 _last_obj_copy_times_ms.reset();
196 _last_termination_times_ms.reset();
197 _last_termination_attempts.reset();
198 _last_gc_worker_end_times_ms.reset();
199 _last_gc_worker_times_ms.reset();
200 _last_gc_worker_other_times_ms.reset();
201 }
202
203 void G1GCPhaseTimes::note_gc_end() {
204 _last_gc_worker_start_times_ms.verify();
205 _last_ext_root_scan_times_ms.verify();
206 _last_satb_filtering_times_ms.verify();
207 _last_update_rs_times_ms.verify();
208 _last_update_rs_processed_buffers.verify();
209 _last_scan_rs_times_ms.verify();
210 _last_strong_code_root_scan_times_ms.verify();
211 _last_strong_code_root_mark_times_ms.verify();
212 _last_obj_copy_times_ms.verify();
213 _last_termination_times_ms.verify();
214 _last_termination_attempts.verify();
215 _last_gc_worker_end_times_ms.verify();
216
217 for (uint i = 0; i < _active_gc_threads; i++) {
218 double worker_time = _last_gc_worker_end_times_ms.get(i) - _last_gc_worker_start_times_ms.get(i);
219 _last_gc_worker_times_ms.set(i, worker_time);
220
221 double worker_known_time = _last_ext_root_scan_times_ms.get(i) +
222 _last_satb_filtering_times_ms.get(i) +
223 _last_update_rs_times_ms.get(i) +
224 _last_scan_rs_times_ms.get(i) +
225 _last_strong_code_root_scan_times_ms.get(i) +
226 _last_strong_code_root_mark_times_ms.get(i) +
227 _last_obj_copy_times_ms.get(i) +
228 _last_termination_times_ms.get(i);
229
230 double worker_other_time = worker_time - worker_known_time;
231 _last_gc_worker_other_times_ms.set(i, worker_other_time);
232 }
233
234 _last_gc_worker_times_ms.verify();
235 _last_gc_worker_other_times_ms.verify();
236 }
237
238 void G1GCPhaseTimes::note_string_dedup_fixup_start() {
239 _cur_string_dedup_queue_fixup_worker_times_ms.reset();
240 _cur_string_dedup_table_fixup_worker_times_ms.reset();
241 }
242
243 void G1GCPhaseTimes::note_string_dedup_fixup_end() {
244 _cur_string_dedup_queue_fixup_worker_times_ms.verify();
245 _cur_string_dedup_table_fixup_worker_times_ms.verify();
246 }
247
248 void G1GCPhaseTimes::print_stats(int level, const char* str, double value) {
249 LineBuffer(level).append_and_print_cr("[%s: %.1lf ms]", str, value);
250 }
251
252 void G1GCPhaseTimes::print_stats(int level, const char* str, double value, uint workers) {
253 LineBuffer(level).append_and_print_cr("[%s: %.1lf ms, GC Workers: " UINT32_FORMAT "]", str, value, workers);
254 }
255
256 double G1GCPhaseTimes::accounted_time_ms() {
257 // Subtract the root region scanning wait time. It's initialized to
258 // zero at the start of the pause.
259 double misc_time_ms = _root_region_scan_wait_time_ms;
260
261 misc_time_ms += _cur_collection_par_time_ms;
262
263 // Now subtract the time taken to fix up roots in generated code
264 misc_time_ms += _cur_collection_code_root_fixup_time_ms;
265
266 // Strong code root migration time
267 misc_time_ms += _cur_strong_code_root_migration_time_ms;
268
269 // Strong code root purge time
270 misc_time_ms += _cur_strong_code_root_purge_time_ms;
271
272 if (G1StringDedup::is_enabled()) {
273 // String dedup fixup time
274 misc_time_ms += _cur_string_dedup_fixup_time_ms;
275 }
276
277 // Subtract the time taken to clean the card table from the
278 // current value of "other time"
279 misc_time_ms += _cur_clear_ct_time_ms;
280
281 return misc_time_ms;
282 }
283
284 void G1GCPhaseTimes::print(double pause_time_sec) {
285 if (_root_region_scan_wait_time_ms > 0.0) {
286 print_stats(1, "Root Region Scan Waiting", _root_region_scan_wait_time_ms);
287 }
288 if (G1CollectedHeap::use_parallel_gc_threads()) {
289 print_stats(1, "Parallel Time", _cur_collection_par_time_ms, _active_gc_threads);
290 _last_gc_worker_start_times_ms.print(2, "GC Worker Start (ms)");
291 _last_ext_root_scan_times_ms.print(2, "Ext Root Scanning (ms)");
292 if (_last_satb_filtering_times_ms.sum() > 0.0) {
293 _last_satb_filtering_times_ms.print(2, "SATB Filtering (ms)");
294 }
295 if (_last_strong_code_root_mark_times_ms.sum() > 0.0) {
296 _last_strong_code_root_mark_times_ms.print(2, "Code Root Marking (ms)");
297 }
298 _last_update_rs_times_ms.print(2, "Update RS (ms)");
299 _last_update_rs_processed_buffers.print(3, "Processed Buffers");
300 _last_scan_rs_times_ms.print(2, "Scan RS (ms)");
301 _last_strong_code_root_scan_times_ms.print(2, "Code Root Scanning (ms)");
302 _last_obj_copy_times_ms.print(2, "Object Copy (ms)");
303 _last_termination_times_ms.print(2, "Termination (ms)");
304 if (G1Log::finest()) {
305 _last_termination_attempts.print(3, "Termination Attempts");
306 }
307 _last_gc_worker_other_times_ms.print(2, "GC Worker Other (ms)");
308 _last_gc_worker_times_ms.print(2, "GC Worker Total (ms)");
309 _last_gc_worker_end_times_ms.print(2, "GC Worker End (ms)");
310 } else {
311 _last_ext_root_scan_times_ms.print(1, "Ext Root Scanning (ms)");
312 if (_last_satb_filtering_times_ms.sum() > 0.0) {
313 _last_satb_filtering_times_ms.print(1, "SATB Filtering (ms)");
314 }
315 if (_last_strong_code_root_mark_times_ms.sum() > 0.0) {
316 _last_strong_code_root_mark_times_ms.print(1, "Code Root Marking (ms)");
317 }
318 _last_update_rs_times_ms.print(1, "Update RS (ms)");
319 _last_update_rs_processed_buffers.print(2, "Processed Buffers");
320 _last_scan_rs_times_ms.print(1, "Scan RS (ms)");
321 _last_strong_code_root_scan_times_ms.print(1, "Code Root Scanning (ms)");
322 _last_obj_copy_times_ms.print(1, "Object Copy (ms)");
323 }
324 print_stats(1, "Code Root Fixup", _cur_collection_code_root_fixup_time_ms);
325 print_stats(1, "Code Root Migration", _cur_strong_code_root_migration_time_ms);
326 print_stats(1, "Code Root Purge", _cur_strong_code_root_purge_time_ms);
327 if (G1StringDedup::is_enabled()) {
328 print_stats(1, "String Dedup Fixup", _cur_string_dedup_fixup_time_ms, _active_gc_threads);
329 _cur_string_dedup_queue_fixup_worker_times_ms.print(2, "Queue Fixup (ms)");
330 _cur_string_dedup_table_fixup_worker_times_ms.print(2, "Table Fixup (ms)");
331 }
332 print_stats(1, "Clear CT", _cur_clear_ct_time_ms);
333 double misc_time_ms = pause_time_sec * MILLIUNITS - accounted_time_ms();
334 print_stats(1, "Other", misc_time_ms);
335 if (_cur_verify_before_time_ms > 0.0) {
336 print_stats(2, "Verify Before", _cur_verify_before_time_ms);
337 }
338 if (G1CollectedHeap::heap()->evacuation_failed()) {
339 double evac_fail_handling = _cur_evac_fail_recalc_used + _cur_evac_fail_remove_self_forwards +
340 _cur_evac_fail_restore_remsets;
341 print_stats(2, "Evacuation Failure", evac_fail_handling);
342 if (G1Log::finest()) {
343 print_stats(3, "Recalculate Used", _cur_evac_fail_recalc_used);
344 print_stats(3, "Remove Self Forwards", _cur_evac_fail_remove_self_forwards);
345 print_stats(3, "Restore RemSet", _cur_evac_fail_restore_remsets);
346 }
347 }
348 print_stats(2, "Choose CSet",
349 (_recorded_young_cset_choice_time_ms +
350 _recorded_non_young_cset_choice_time_ms));
351 print_stats(2, "Ref Proc", _cur_ref_proc_time_ms);
352 print_stats(2, "Ref Enq", _cur_ref_enq_time_ms);
353 if (G1DeferredRSUpdate) {
354 print_stats(2, "Redirty Cards", _recorded_redirty_logged_cards_time_ms);
355 }
356 print_stats(2, "Free CSet",
357 (_recorded_young_free_cset_time_ms +
358 _recorded_non_young_free_cset_time_ms));
359 if (G1Log::finest()) {
360 print_stats(3, "Young Free CSet", _recorded_young_free_cset_time_ms);
361 print_stats(3, "Non-Young Free CSet", _recorded_non_young_free_cset_time_ms);
362 }
363 if (_cur_verify_after_time_ms > 0.0) {
364 print_stats(2, "Verify After", _cur_verify_after_time_ms);
365 }
366 }