1 /*
2 * Copyright (c) 2001, 2011, 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_implementation/g1/concurrentG1Refine.hpp"
27 #include "gc_implementation/g1/concurrentMark.hpp"
28 #include "gc_implementation/g1/concurrentMarkThread.inline.hpp"
29 #include "gc_implementation/g1/g1CollectedHeap.inline.hpp"
30 #include "gc_implementation/g1/g1CollectorPolicy.hpp"
31 #include "gc_implementation/g1/g1ErgoVerbose.hpp"
32 #include "gc_implementation/g1/heapRegionRemSet.hpp"
33 #include "gc_implementation/shared/gcPolicyCounters.hpp"
34 #include "runtime/arguments.hpp"
35 #include "runtime/java.hpp"
36 #include "runtime/mutexLocker.hpp"
37 #include "utilities/debug.hpp"
38
39 #define PREDICTIONS_VERBOSE 0
40
41 // <NEW PREDICTION>
42
43 // Different defaults for different number of GC threads
44 // They were chosen by running GCOld and SPECjbb on debris with different
45 // numbers of GC threads and choosing them based on the results
46
47 // all the same
48 static double rs_length_diff_defaults[] = {
49 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0
50 };
51
52 static double cost_per_card_ms_defaults[] = {
53 0.01, 0.005, 0.005, 0.003, 0.003, 0.002, 0.002, 0.0015
54 };
55
56 // all the same
57 static double fully_young_cards_per_entry_ratio_defaults[] = {
58 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0, 1.0
59 };
60
61 static double cost_per_entry_ms_defaults[] = {
62 0.015, 0.01, 0.01, 0.008, 0.008, 0.0055, 0.0055, 0.005
63 };
64
65 static double cost_per_byte_ms_defaults[] = {
66 0.00006, 0.00003, 0.00003, 0.000015, 0.000015, 0.00001, 0.00001, 0.000009
67 };
68
69 // these should be pretty consistent
70 static double constant_other_time_ms_defaults[] = {
71 5.0, 5.0, 5.0, 5.0, 5.0, 5.0, 5.0, 5.0
72 };
73
74
75 static double young_other_cost_per_region_ms_defaults[] = {
76 0.3, 0.2, 0.2, 0.15, 0.15, 0.12, 0.12, 0.1
77 };
78
79 static double non_young_other_cost_per_region_ms_defaults[] = {
80 1.0, 0.7, 0.7, 0.5, 0.5, 0.42, 0.42, 0.30
81 };
82
83 // </NEW PREDICTION>
84
85 // Help class for avoiding interleaved logging
86 class LineBuffer: public StackObj {
87
88 private:
89 static const int BUFFER_LEN = 1024;
90 static const int INDENT_CHARS = 3;
91 char _buffer[BUFFER_LEN];
92 int _indent_level;
93 int _cur;
94
95 void vappend(const char* format, va_list ap) {
96 int res = vsnprintf(&_buffer[_cur], BUFFER_LEN - _cur, format, ap);
97 if (res != -1) {
98 _cur += res;
99 } else {
100 DEBUG_ONLY(warning("buffer too small in LineBuffer");)
101 _buffer[BUFFER_LEN -1] = 0;
102 _cur = BUFFER_LEN; // vsnprintf above should not add to _buffer if we are called again
103 }
104 }
105
106 public:
107 explicit LineBuffer(int indent_level): _indent_level(indent_level), _cur(0) {
108 for (; (_cur < BUFFER_LEN && _cur < (_indent_level * INDENT_CHARS)); _cur++) {
109 _buffer[_cur] = ' ';
110 }
111 }
112
113 #ifndef PRODUCT
114 ~LineBuffer() {
115 assert(_cur == _indent_level * INDENT_CHARS, "pending data in buffer - append_and_print_cr() not called?");
116 }
117 #endif
118
119 void append(const char* format, ...) {
120 va_list ap;
121 va_start(ap, format);
122 vappend(format, ap);
123 va_end(ap);
124 }
125
126 void append_and_print_cr(const char* format, ...) {
127 va_list ap;
128 va_start(ap, format);
129 vappend(format, ap);
130 va_end(ap);
131 gclog_or_tty->print_cr("%s", _buffer);
132 _cur = _indent_level * INDENT_CHARS;
133 }
134 };
135
136 G1CollectorPolicy::G1CollectorPolicy() :
137 _parallel_gc_threads(G1CollectedHeap::use_parallel_gc_threads()
138 ? ParallelGCThreads : 1),
139
140 _n_pauses(0),
141 _recent_rs_scan_times_ms(new TruncatedSeq(NumPrevPausesForHeuristics)),
142 _recent_pause_times_ms(new TruncatedSeq(NumPrevPausesForHeuristics)),
143 _recent_rs_sizes(new TruncatedSeq(NumPrevPausesForHeuristics)),
144 _recent_gc_times_ms(new TruncatedSeq(NumPrevPausesForHeuristics)),
145 _all_pause_times_ms(new NumberSeq()),
146 _stop_world_start(0.0),
147 _all_stop_world_times_ms(new NumberSeq()),
148 _all_yield_times_ms(new NumberSeq()),
149 _using_new_ratio_calculations(false),
150
151 _all_mod_union_times_ms(new NumberSeq()),
152
153 _summary(new Summary()),
154
155 #ifndef PRODUCT
156 _cur_clear_ct_time_ms(0.0),
157 _min_clear_cc_time_ms(-1.0),
158 _max_clear_cc_time_ms(-1.0),
159 _cur_clear_cc_time_ms(0.0),
160 _cum_clear_cc_time_ms(0.0),
161 _num_cc_clears(0L),
162 #endif
163
164 _region_num_young(0),
165 _region_num_tenured(0),
166 _prev_region_num_young(0),
167 _prev_region_num_tenured(0),
168
169 _aux_num(10),
170 _all_aux_times_ms(new NumberSeq[_aux_num]),
171 _cur_aux_start_times_ms(new double[_aux_num]),
172 _cur_aux_times_ms(new double[_aux_num]),
173 _cur_aux_times_set(new bool[_aux_num]),
174
175 _concurrent_mark_remark_times_ms(new TruncatedSeq(NumPrevPausesForHeuristics)),
176 _concurrent_mark_cleanup_times_ms(new TruncatedSeq(NumPrevPausesForHeuristics)),
177
178 // <NEW PREDICTION>
179
180 _alloc_rate_ms_seq(new TruncatedSeq(TruncatedSeqLength)),
181 _prev_collection_pause_end_ms(0.0),
182 _pending_card_diff_seq(new TruncatedSeq(TruncatedSeqLength)),
183 _rs_length_diff_seq(new TruncatedSeq(TruncatedSeqLength)),
184 _cost_per_card_ms_seq(new TruncatedSeq(TruncatedSeqLength)),
185 _fully_young_cards_per_entry_ratio_seq(new TruncatedSeq(TruncatedSeqLength)),
186 _partially_young_cards_per_entry_ratio_seq(
187 new TruncatedSeq(TruncatedSeqLength)),
188 _cost_per_entry_ms_seq(new TruncatedSeq(TruncatedSeqLength)),
189 _partially_young_cost_per_entry_ms_seq(new TruncatedSeq(TruncatedSeqLength)),
190 _cost_per_byte_ms_seq(new TruncatedSeq(TruncatedSeqLength)),
191 _cost_per_byte_ms_during_cm_seq(new TruncatedSeq(TruncatedSeqLength)),
192 _constant_other_time_ms_seq(new TruncatedSeq(TruncatedSeqLength)),
193 _young_other_cost_per_region_ms_seq(new TruncatedSeq(TruncatedSeqLength)),
194 _non_young_other_cost_per_region_ms_seq(
195 new TruncatedSeq(TruncatedSeqLength)),
196
197 _pending_cards_seq(new TruncatedSeq(TruncatedSeqLength)),
198 _scanned_cards_seq(new TruncatedSeq(TruncatedSeqLength)),
199 _rs_lengths_seq(new TruncatedSeq(TruncatedSeqLength)),
200
201 _pause_time_target_ms((double) MaxGCPauseMillis),
202
203 // </NEW PREDICTION>
204
205 _full_young_gcs(true),
206 _full_young_pause_num(0),
207 _partial_young_pause_num(0),
208
209 _during_marking(false),
210 _in_marking_window(false),
211 _in_marking_window_im(false),
212
213 _known_garbage_ratio(0.0),
214 _known_garbage_bytes(0),
215
216 _young_gc_eff_seq(new TruncatedSeq(TruncatedSeqLength)),
217
218 _recent_prev_end_times_for_all_gcs_sec(new TruncatedSeq(NumPrevPausesForHeuristics)),
219
220 _recent_CS_bytes_used_before(new TruncatedSeq(NumPrevPausesForHeuristics)),
221 _recent_CS_bytes_surviving(new TruncatedSeq(NumPrevPausesForHeuristics)),
222
223 _recent_avg_pause_time_ratio(0.0),
224 _num_markings(0),
225 _n_marks(0),
226 _n_pauses_at_mark_end(0),
227
228 _all_full_gc_times_ms(new NumberSeq()),
229
230 // G1PausesBtwnConcMark defaults to -1
231 // so the hack is to do the cast QQQ FIXME
232 _pauses_btwn_concurrent_mark((size_t)G1PausesBtwnConcMark),
233 _n_marks_since_last_pause(0),
234 _initiate_conc_mark_if_possible(false),
235 _during_initial_mark_pause(false),
236 _should_revert_to_full_young_gcs(false),
237 _last_full_young_gc(false),
238
239 _eden_bytes_before_gc(0),
240 _survivor_bytes_before_gc(0),
241 _capacity_before_gc(0),
242
243 _prev_collection_pause_used_at_end_bytes(0),
244
245 _collection_set(NULL),
246 _collection_set_size(0),
247 _collection_set_bytes_used_before(0),
248
249 // Incremental CSet attributes
250 _inc_cset_build_state(Inactive),
251 _inc_cset_head(NULL),
252 _inc_cset_tail(NULL),
253 _inc_cset_size(0),
254 _inc_cset_young_index(0),
255 _inc_cset_bytes_used_before(0),
256 _inc_cset_max_finger(NULL),
257 _inc_cset_recorded_young_bytes(0),
258 _inc_cset_recorded_rs_lengths(0),
259 _inc_cset_predicted_elapsed_time_ms(0.0),
260 _inc_cset_predicted_bytes_to_copy(0),
261
262 #ifdef _MSC_VER // the use of 'this' below gets a warning, make it go away
263 #pragma warning( disable:4355 ) // 'this' : used in base member initializer list
264 #endif // _MSC_VER
265
266 _short_lived_surv_rate_group(new SurvRateGroup(this, "Short Lived",
267 G1YoungSurvRateNumRegionsSummary)),
268 _survivor_surv_rate_group(new SurvRateGroup(this, "Survivor",
269 G1YoungSurvRateNumRegionsSummary)),
270 // add here any more surv rate groups
271 _recorded_survivor_regions(0),
272 _recorded_survivor_head(NULL),
273 _recorded_survivor_tail(NULL),
274 _survivors_age_table(true),
275
276 _gc_overhead_perc(0.0) {
277
278 // Set up the region size and associated fields. Given that the
279 // policy is created before the heap, we have to set this up here,
280 // so it's done as soon as possible.
281 HeapRegion::setup_heap_region_size(Arguments::min_heap_size());
282 HeapRegionRemSet::setup_remset_size();
283
284 G1ErgoVerbose::initialize();
285 if (PrintAdaptiveSizePolicy) {
286 // Currently, we only use a single switch for all the heuristics.
287 G1ErgoVerbose::set_enabled(true);
288 // Given that we don't currently have a verboseness level
289 // parameter, we'll hardcode this to high. This can be easily
290 // changed in the future.
291 G1ErgoVerbose::set_level(ErgoHigh);
292 } else {
293 G1ErgoVerbose::set_enabled(false);
294 }
295
296 // Verify PLAB sizes
297 const uint region_size = HeapRegion::GrainWords;
298 if (YoungPLABSize > region_size || OldPLABSize > region_size) {
299 char buffer[128];
300 jio_snprintf(buffer, sizeof(buffer), "%sPLABSize should be at most %u",
301 OldPLABSize > region_size ? "Old" : "Young", region_size);
302 vm_exit_during_initialization(buffer);
303 }
304
305 _recent_prev_end_times_for_all_gcs_sec->add(os::elapsedTime());
306 _prev_collection_pause_end_ms = os::elapsedTime() * 1000.0;
307
308 _par_last_gc_worker_start_times_ms = new double[_parallel_gc_threads];
309 _par_last_ext_root_scan_times_ms = new double[_parallel_gc_threads];
310 _par_last_mark_stack_scan_times_ms = new double[_parallel_gc_threads];
311
312 _par_last_update_rs_times_ms = new double[_parallel_gc_threads];
313 _par_last_update_rs_processed_buffers = new double[_parallel_gc_threads];
314
315 _par_last_scan_rs_times_ms = new double[_parallel_gc_threads];
316
317 _par_last_obj_copy_times_ms = new double[_parallel_gc_threads];
318
319 _par_last_termination_times_ms = new double[_parallel_gc_threads];
320 _par_last_termination_attempts = new double[_parallel_gc_threads];
321 _par_last_gc_worker_end_times_ms = new double[_parallel_gc_threads];
322 _par_last_gc_worker_times_ms = new double[_parallel_gc_threads];
323
324 // start conservatively
325 _expensive_region_limit_ms = 0.5 * (double) MaxGCPauseMillis;
326
327 // <NEW PREDICTION>
328
329 int index;
330 if (ParallelGCThreads == 0)
331 index = 0;
332 else if (ParallelGCThreads > 8)
333 index = 7;
334 else
335 index = ParallelGCThreads - 1;
336
337 _pending_card_diff_seq->add(0.0);
338 _rs_length_diff_seq->add(rs_length_diff_defaults[index]);
339 _cost_per_card_ms_seq->add(cost_per_card_ms_defaults[index]);
340 _fully_young_cards_per_entry_ratio_seq->add(
341 fully_young_cards_per_entry_ratio_defaults[index]);
342 _cost_per_entry_ms_seq->add(cost_per_entry_ms_defaults[index]);
343 _cost_per_byte_ms_seq->add(cost_per_byte_ms_defaults[index]);
344 _constant_other_time_ms_seq->add(constant_other_time_ms_defaults[index]);
345 _young_other_cost_per_region_ms_seq->add(
346 young_other_cost_per_region_ms_defaults[index]);
347 _non_young_other_cost_per_region_ms_seq->add(
348 non_young_other_cost_per_region_ms_defaults[index]);
349
350 // </NEW PREDICTION>
351
352 // Below, we might need to calculate the pause time target based on
353 // the pause interval. When we do so we are going to give G1 maximum
354 // flexibility and allow it to do pauses when it needs to. So, we'll
355 // arrange that the pause interval to be pause time target + 1 to
356 // ensure that a) the pause time target is maximized with respect to
357 // the pause interval and b) we maintain the invariant that pause
358 // time target < pause interval. If the user does not want this
359 // maximum flexibility, they will have to set the pause interval
360 // explicitly.
361
362 // First make sure that, if either parameter is set, its value is
363 // reasonable.
364 if (!FLAG_IS_DEFAULT(MaxGCPauseMillis)) {
365 if (MaxGCPauseMillis < 1) {
366 vm_exit_during_initialization("MaxGCPauseMillis should be "
367 "greater than 0");
368 }
369 }
370 if (!FLAG_IS_DEFAULT(GCPauseIntervalMillis)) {
371 if (GCPauseIntervalMillis < 1) {
372 vm_exit_during_initialization("GCPauseIntervalMillis should be "
373 "greater than 0");
374 }
375 }
376
377 // Then, if the pause time target parameter was not set, set it to
378 // the default value.
379 if (FLAG_IS_DEFAULT(MaxGCPauseMillis)) {
380 if (FLAG_IS_DEFAULT(GCPauseIntervalMillis)) {
381 // The default pause time target in G1 is 200ms
382 FLAG_SET_DEFAULT(MaxGCPauseMillis, 200);
383 } else {
384 // We do not allow the pause interval to be set without the
385 // pause time target
386 vm_exit_during_initialization("GCPauseIntervalMillis cannot be set "
387 "without setting MaxGCPauseMillis");
388 }
389 }
390
391 // Then, if the interval parameter was not set, set it according to
392 // the pause time target (this will also deal with the case when the
393 // pause time target is the default value).
394 if (FLAG_IS_DEFAULT(GCPauseIntervalMillis)) {
395 FLAG_SET_DEFAULT(GCPauseIntervalMillis, MaxGCPauseMillis + 1);
396 }
397
398 // Finally, make sure that the two parameters are consistent.
399 if (MaxGCPauseMillis >= GCPauseIntervalMillis) {
400 char buffer[256];
401 jio_snprintf(buffer, 256,
402 "MaxGCPauseMillis (%u) should be less than "
403 "GCPauseIntervalMillis (%u)",
404 MaxGCPauseMillis, GCPauseIntervalMillis);
405 vm_exit_during_initialization(buffer);
406 }
407
408 double max_gc_time = (double) MaxGCPauseMillis / 1000.0;
409 double time_slice = (double) GCPauseIntervalMillis / 1000.0;
410 _mmu_tracker = new G1MMUTrackerQueue(time_slice, max_gc_time);
411 _sigma = (double) G1ConfidencePercent / 100.0;
412
413 // start conservatively (around 50ms is about right)
414 _concurrent_mark_remark_times_ms->add(0.05);
415 _concurrent_mark_cleanup_times_ms->add(0.20);
416 _tenuring_threshold = MaxTenuringThreshold;
417 // _max_survivor_regions will be calculated by
418 // update_young_list_target_length() during initialization.
419 _max_survivor_regions = 0;
420
421 assert(GCTimeRatio > 0,
422 "we should have set it to a default value set_g1_gc_flags() "
423 "if a user set it to 0");
424 _gc_overhead_perc = 100.0 * (1.0 / (1.0 + GCTimeRatio));
425
426 uintx reserve_perc = G1ReservePercent;
427 // Put an artificial ceiling on this so that it's not set to a silly value.
428 if (reserve_perc > 50) {
429 reserve_perc = 50;
430 warning("G1ReservePercent is set to a value that is too large, "
431 "it's been updated to %u", reserve_perc);
432 }
433 _reserve_factor = (double) reserve_perc / 100.0;
434 // This will be set when the heap is expanded
435 // for the first time during initialization.
436 _reserve_regions = 0;
437
438 initialize_all();
439 }
440
441 // Increment "i", mod "len"
442 static void inc_mod(int& i, int len) {
443 i++; if (i == len) i = 0;
444 }
445
446 void G1CollectorPolicy::initialize_flags() {
447 set_min_alignment(HeapRegion::GrainBytes);
448 set_max_alignment(GenRemSet::max_alignment_constraint(rem_set_name()));
449 if (SurvivorRatio < 1) {
450 vm_exit_during_initialization("Invalid survivor ratio specified");
451 }
452 CollectorPolicy::initialize_flags();
453 }
454
455 // The easiest way to deal with the parsing of the NewSize /
456 // MaxNewSize / etc. parameteres is to re-use the code in the
457 // TwoGenerationCollectorPolicy class. This is similar to what
458 // ParallelScavenge does with its GenerationSizer class (see
459 // ParallelScavengeHeap::initialize()). We might change this in the
460 // future, but it's a good start.
461 class G1YoungGenSizer : public TwoGenerationCollectorPolicy {
462
463 public:
464 G1YoungGenSizer() {
465 initialize_flags();
466 initialize_size_info();
467 }
468 size_t size_to_region_num(size_t byte_size) {
469 return MAX2((size_t) 1, byte_size / HeapRegion::GrainBytes);
470 }
471 size_t min_young_region_num() {
472 return size_to_region_num(_min_gen0_size);
473 }
474 size_t initial_young_region_num() {
475 return size_to_region_num(_initial_gen0_size);
476 }
477 size_t max_young_region_num() {
478 return size_to_region_num(_max_gen0_size);
479 }
480 };
481
482 void G1CollectorPolicy::update_young_list_size_using_newratio(size_t number_of_heap_regions) {
483 assert(number_of_heap_regions > 0, "Heap must be initialized");
484 size_t young_size = number_of_heap_regions / (NewRatio + 1);
485 _min_desired_young_length = young_size;
486 _max_desired_young_length = young_size;
487 }
488
489 void G1CollectorPolicy::init() {
490 // Set aside an initial future to_space.
491 _g1 = G1CollectedHeap::heap();
492
493 assert(Heap_lock->owned_by_self(), "Locking discipline.");
494
495 initialize_gc_policy_counters();
496
497 G1YoungGenSizer sizer;
498 size_t initial_region_num = sizer.initial_young_region_num();
499 _min_desired_young_length = sizer.min_young_region_num();
500 _max_desired_young_length = sizer.max_young_region_num();
501
502 if (FLAG_IS_CMDLINE(NewRatio)) {
503 if (FLAG_IS_CMDLINE(NewSize) || FLAG_IS_CMDLINE(MaxNewSize)) {
504 gclog_or_tty->print_cr("-XX:NewSize and -XX:MaxNewSize overrides -XX:NewRatio");
505 } else {
506 // Treat NewRatio as a fixed size that is only recalculated when the heap size changes
507 size_t heap_regions = sizer.size_to_region_num(_g1->n_regions());
508 update_young_list_size_using_newratio(heap_regions);
509 _using_new_ratio_calculations = true;
510 }
511 }
512
513 // GenCollectorPolicy guarantees that min <= initial <= max.
514 // Asserting here just to state that we rely on this property.
515 assert(_min_desired_young_length <= _max_desired_young_length, "Invalid min/max young gen size values");
516 assert(initial_region_num <= _max_desired_young_length, "Initial young gen size too large");
517 assert(_min_desired_young_length <= initial_region_num, "Initial young gen size too small");
518
519 set_adaptive_young_list_length(_min_desired_young_length < _max_desired_young_length);
520 if (adaptive_young_list_length()) {
521 _young_list_fixed_length = 0;
522 } else {
523 _young_list_fixed_length = initial_region_num;
524 }
525 _free_regions_at_end_of_collection = _g1->free_regions();
526 update_young_list_target_length();
527 _prev_eden_capacity = _young_list_target_length * HeapRegion::GrainBytes;
528
529 // We may immediately start allocating regions and placing them on the
530 // collection set list. Initialize the per-collection set info
531 start_incremental_cset_building();
532 }
533
534 // Create the jstat counters for the policy.
535 void G1CollectorPolicy::initialize_gc_policy_counters() {
536 _gc_policy_counters = new GCPolicyCounters("GarbageFirst", 1, 3);
537 }
538
539 bool G1CollectorPolicy::predict_will_fit(size_t young_length,
540 double base_time_ms,
541 size_t base_free_regions,
542 double target_pause_time_ms) {
543 if (young_length >= base_free_regions) {
544 // end condition 1: not enough space for the young regions
545 return false;
546 }
547
548 double accum_surv_rate = accum_yg_surv_rate_pred((int)(young_length - 1));
549 size_t bytes_to_copy =
550 (size_t) (accum_surv_rate * (double) HeapRegion::GrainBytes);
551 double copy_time_ms = predict_object_copy_time_ms(bytes_to_copy);
552 double young_other_time_ms = predict_young_other_time_ms(young_length);
553 double pause_time_ms = base_time_ms + copy_time_ms + young_other_time_ms;
554 if (pause_time_ms > target_pause_time_ms) {
555 // end condition 2: prediction is over the target pause time
556 return false;
557 }
558
559 size_t free_bytes =
560 (base_free_regions - young_length) * HeapRegion::GrainBytes;
561 if ((2.0 * sigma()) * (double) bytes_to_copy > (double) free_bytes) {
562 // end condition 3: out-of-space (conservatively!)
563 return false;
564 }
565
566 // success!
567 return true;
568 }
569
570 void G1CollectorPolicy::record_new_heap_size(size_t new_number_of_regions) {
571 // re-calculate the necessary reserve
572 double reserve_regions_d = (double) new_number_of_regions * _reserve_factor;
573 // We use ceiling so that if reserve_regions_d is > 0.0 (but
574 // smaller than 1.0) we'll get 1.
575 _reserve_regions = (size_t) ceil(reserve_regions_d);
576
577 if (_using_new_ratio_calculations) {
578 // -XX:NewRatio was specified so we need to update the
579 // young gen length when the heap size has changed.
580 update_young_list_size_using_newratio(new_number_of_regions);
581 }
582 }
583
584 size_t G1CollectorPolicy::calculate_young_list_desired_min_length(
585 size_t base_min_length) {
586 size_t desired_min_length = 0;
587 if (adaptive_young_list_length()) {
588 if (_alloc_rate_ms_seq->num() > 3) {
589 double now_sec = os::elapsedTime();
590 double when_ms = _mmu_tracker->when_max_gc_sec(now_sec) * 1000.0;
591 double alloc_rate_ms = predict_alloc_rate_ms();
592 desired_min_length = (size_t) ceil(alloc_rate_ms * when_ms);
593 } else {
594 // otherwise we don't have enough info to make the prediction
595 }
596 }
597 desired_min_length += base_min_length;
598 // make sure we don't go below any user-defined minimum bound
599 return MAX2(_min_desired_young_length, desired_min_length);
600 }
601
602 size_t G1CollectorPolicy::calculate_young_list_desired_max_length() {
603 // Here, we might want to also take into account any additional
604 // constraints (i.e., user-defined minimum bound). Currently, we
605 // effectively don't set this bound.
606 return _max_desired_young_length;
607 }
608
609 void G1CollectorPolicy::update_young_list_target_length(size_t rs_lengths) {
610 if (rs_lengths == (size_t) -1) {
611 // if it's set to the default value (-1), we should predict it;
612 // otherwise, use the given value.
613 rs_lengths = (size_t) get_new_prediction(_rs_lengths_seq);
614 }
615
616 // Calculate the absolute and desired min bounds.
617
618 // This is how many young regions we already have (currently: the survivors).
619 size_t base_min_length = recorded_survivor_regions();
620 // This is the absolute minimum young length, which ensures that we
621 // can allocate one eden region in the worst-case.
622 size_t absolute_min_length = base_min_length + 1;
623 size_t desired_min_length =
624 calculate_young_list_desired_min_length(base_min_length);
625 if (desired_min_length < absolute_min_length) {
626 desired_min_length = absolute_min_length;
627 }
628
629 // Calculate the absolute and desired max bounds.
630
631 // We will try our best not to "eat" into the reserve.
632 size_t absolute_max_length = 0;
633 if (_free_regions_at_end_of_collection > _reserve_regions) {
634 absolute_max_length = _free_regions_at_end_of_collection - _reserve_regions;
635 }
636 size_t desired_max_length = calculate_young_list_desired_max_length();
637 if (desired_max_length > absolute_max_length) {
638 desired_max_length = absolute_max_length;
639 }
640
641 size_t young_list_target_length = 0;
642 if (adaptive_young_list_length()) {
643 if (full_young_gcs()) {
644 young_list_target_length =
645 calculate_young_list_target_length(rs_lengths,
646 base_min_length,
647 desired_min_length,
648 desired_max_length);
649 _rs_lengths_prediction = rs_lengths;
650 } else {
651 // Don't calculate anything and let the code below bound it to
652 // the desired_min_length, i.e., do the next GC as soon as
653 // possible to maximize how many old regions we can add to it.
654 }
655 } else {
656 if (full_young_gcs()) {
657 young_list_target_length = _young_list_fixed_length;
658 } else {
659 // A bit arbitrary: during partially-young GCs we allocate half
660 // the young regions to try to add old regions to the CSet.
661 young_list_target_length = _young_list_fixed_length / 2;
662 // We choose to accept that we might go under the desired min
663 // length given that we intentionally ask for a smaller young gen.
664 desired_min_length = absolute_min_length;
665 }
666 }
667
668 // Make sure we don't go over the desired max length, nor under the
669 // desired min length. In case they clash, desired_min_length wins
670 // which is why that test is second.
671 if (young_list_target_length > desired_max_length) {
672 young_list_target_length = desired_max_length;
673 }
674 if (young_list_target_length < desired_min_length) {
675 young_list_target_length = desired_min_length;
676 }
677
678 assert(young_list_target_length > recorded_survivor_regions(),
679 "we should be able to allocate at least one eden region");
680 assert(young_list_target_length >= absolute_min_length, "post-condition");
681 _young_list_target_length = young_list_target_length;
682
683 update_max_gc_locker_expansion();
684 }
685
686 size_t
687 G1CollectorPolicy::calculate_young_list_target_length(size_t rs_lengths,
688 size_t base_min_length,
689 size_t desired_min_length,
690 size_t desired_max_length) {
691 assert(adaptive_young_list_length(), "pre-condition");
692 assert(full_young_gcs(), "only call this for fully-young GCs");
693
694 // In case some edge-condition makes the desired max length too small...
695 if (desired_max_length <= desired_min_length) {
696 return desired_min_length;
697 }
698
699 // We'll adjust min_young_length and max_young_length not to include
700 // the already allocated young regions (i.e., so they reflect the
701 // min and max eden regions we'll allocate). The base_min_length
702 // will be reflected in the predictions by the
703 // survivor_regions_evac_time prediction.
704 assert(desired_min_length > base_min_length, "invariant");
705 size_t min_young_length = desired_min_length - base_min_length;
706 assert(desired_max_length > base_min_length, "invariant");
707 size_t max_young_length = desired_max_length - base_min_length;
708
709 double target_pause_time_ms = _mmu_tracker->max_gc_time() * 1000.0;
710 double survivor_regions_evac_time = predict_survivor_regions_evac_time();
711 size_t pending_cards = (size_t) get_new_prediction(_pending_cards_seq);
712 size_t adj_rs_lengths = rs_lengths + predict_rs_length_diff();
713 size_t scanned_cards = predict_young_card_num(adj_rs_lengths);
714 double base_time_ms =
715 predict_base_elapsed_time_ms(pending_cards, scanned_cards) +
716 survivor_regions_evac_time;
717 size_t available_free_regions = _free_regions_at_end_of_collection;
718 size_t base_free_regions = 0;
719 if (available_free_regions > _reserve_regions) {
720 base_free_regions = available_free_regions - _reserve_regions;
721 }
722
723 // Here, we will make sure that the shortest young length that
724 // makes sense fits within the target pause time.
725
726 if (predict_will_fit(min_young_length, base_time_ms,
727 base_free_regions, target_pause_time_ms)) {
728 // The shortest young length will fit into the target pause time;
729 // we'll now check whether the absolute maximum number of young
730 // regions will fit in the target pause time. If not, we'll do
731 // a binary search between min_young_length and max_young_length.
732 if (predict_will_fit(max_young_length, base_time_ms,
733 base_free_regions, target_pause_time_ms)) {
734 // The maximum young length will fit into the target pause time.
735 // We are done so set min young length to the maximum length (as
736 // the result is assumed to be returned in min_young_length).
737 min_young_length = max_young_length;
738 } else {
739 // The maximum possible number of young regions will not fit within
740 // the target pause time so we'll search for the optimal
741 // length. The loop invariants are:
742 //
743 // min_young_length < max_young_length
744 // min_young_length is known to fit into the target pause time
745 // max_young_length is known not to fit into the target pause time
746 //
747 // Going into the loop we know the above hold as we've just
748 // checked them. Every time around the loop we check whether
749 // the middle value between min_young_length and
750 // max_young_length fits into the target pause time. If it
751 // does, it becomes the new min. If it doesn't, it becomes
752 // the new max. This way we maintain the loop invariants.
753
754 assert(min_young_length < max_young_length, "invariant");
755 size_t diff = (max_young_length - min_young_length) / 2;
756 while (diff > 0) {
757 size_t young_length = min_young_length + diff;
758 if (predict_will_fit(young_length, base_time_ms,
759 base_free_regions, target_pause_time_ms)) {
760 min_young_length = young_length;
761 } else {
762 max_young_length = young_length;
763 }
764 assert(min_young_length < max_young_length, "invariant");
765 diff = (max_young_length - min_young_length) / 2;
766 }
767 // The results is min_young_length which, according to the
768 // loop invariants, should fit within the target pause time.
769
770 // These are the post-conditions of the binary search above:
771 assert(min_young_length < max_young_length,
772 "otherwise we should have discovered that max_young_length "
773 "fits into the pause target and not done the binary search");
774 assert(predict_will_fit(min_young_length, base_time_ms,
775 base_free_regions, target_pause_time_ms),
776 "min_young_length, the result of the binary search, should "
777 "fit into the pause target");
778 assert(!predict_will_fit(min_young_length + 1, base_time_ms,
779 base_free_regions, target_pause_time_ms),
780 "min_young_length, the result of the binary search, should be "
781 "optimal, so no larger length should fit into the pause target");
782 }
783 } else {
784 // Even the minimum length doesn't fit into the pause time
785 // target, return it as the result nevertheless.
786 }
787 return base_min_length + min_young_length;
788 }
789
790 double G1CollectorPolicy::predict_survivor_regions_evac_time() {
791 double survivor_regions_evac_time = 0.0;
792 for (HeapRegion * r = _recorded_survivor_head;
793 r != NULL && r != _recorded_survivor_tail->get_next_young_region();
794 r = r->get_next_young_region()) {
795 survivor_regions_evac_time += predict_region_elapsed_time_ms(r, true);
796 }
797 return survivor_regions_evac_time;
798 }
799
800 void G1CollectorPolicy::revise_young_list_target_length_if_necessary() {
801 guarantee( adaptive_young_list_length(), "should not call this otherwise" );
802
803 size_t rs_lengths = _g1->young_list()->sampled_rs_lengths();
804 if (rs_lengths > _rs_lengths_prediction) {
805 // add 10% to avoid having to recalculate often
806 size_t rs_lengths_prediction = rs_lengths * 1100 / 1000;
807 update_young_list_target_length(rs_lengths_prediction);
808 }
809 }
810
811
812
813 HeapWord* G1CollectorPolicy::mem_allocate_work(size_t size,
814 bool is_tlab,
815 bool* gc_overhead_limit_was_exceeded) {
816 guarantee(false, "Not using this policy feature yet.");
817 return NULL;
818 }
819
820 // This method controls how a collector handles one or more
821 // of its generations being fully allocated.
822 HeapWord* G1CollectorPolicy::satisfy_failed_allocation(size_t size,
823 bool is_tlab) {
824 guarantee(false, "Not using this policy feature yet.");
825 return NULL;
826 }
827
828
829 #ifndef PRODUCT
830 bool G1CollectorPolicy::verify_young_ages() {
831 HeapRegion* head = _g1->young_list()->first_region();
832 return
833 verify_young_ages(head, _short_lived_surv_rate_group);
834 // also call verify_young_ages on any additional surv rate groups
835 }
836
837 bool
838 G1CollectorPolicy::verify_young_ages(HeapRegion* head,
839 SurvRateGroup *surv_rate_group) {
840 guarantee( surv_rate_group != NULL, "pre-condition" );
841
842 const char* name = surv_rate_group->name();
843 bool ret = true;
844 int prev_age = -1;
845
846 for (HeapRegion* curr = head;
847 curr != NULL;
848 curr = curr->get_next_young_region()) {
849 SurvRateGroup* group = curr->surv_rate_group();
850 if (group == NULL && !curr->is_survivor()) {
851 gclog_or_tty->print_cr("## %s: encountered NULL surv_rate_group", name);
852 ret = false;
853 }
854
855 if (surv_rate_group == group) {
856 int age = curr->age_in_surv_rate_group();
857
858 if (age < 0) {
859 gclog_or_tty->print_cr("## %s: encountered negative age", name);
860 ret = false;
861 }
862
863 if (age <= prev_age) {
864 gclog_or_tty->print_cr("## %s: region ages are not strictly increasing "
865 "(%d, %d)", name, age, prev_age);
866 ret = false;
867 }
868 prev_age = age;
869 }
870 }
871
872 return ret;
873 }
874 #endif // PRODUCT
875
876 void G1CollectorPolicy::record_full_collection_start() {
877 _cur_collection_start_sec = os::elapsedTime();
878 // Release the future to-space so that it is available for compaction into.
879 _g1->set_full_collection();
880 }
881
882 void G1CollectorPolicy::record_full_collection_end() {
883 // Consider this like a collection pause for the purposes of allocation
884 // since last pause.
885 double end_sec = os::elapsedTime();
886 double full_gc_time_sec = end_sec - _cur_collection_start_sec;
887 double full_gc_time_ms = full_gc_time_sec * 1000.0;
888
889 _all_full_gc_times_ms->add(full_gc_time_ms);
890
891 update_recent_gc_times(end_sec, full_gc_time_ms);
892
893 _g1->clear_full_collection();
894
895 // "Nuke" the heuristics that control the fully/partially young GC
896 // transitions and make sure we start with fully young GCs after the
897 // Full GC.
898 set_full_young_gcs(true);
899 _last_full_young_gc = false;
900 _should_revert_to_full_young_gcs = false;
901 clear_initiate_conc_mark_if_possible();
902 clear_during_initial_mark_pause();
903 _known_garbage_bytes = 0;
904 _known_garbage_ratio = 0.0;
905 _in_marking_window = false;
906 _in_marking_window_im = false;
907
908 _short_lived_surv_rate_group->start_adding_regions();
909 // also call this on any additional surv rate groups
910
911 record_survivor_regions(0, NULL, NULL);
912
913 _prev_region_num_young = _region_num_young;
914 _prev_region_num_tenured = _region_num_tenured;
915
916 _free_regions_at_end_of_collection = _g1->free_regions();
917 // Reset survivors SurvRateGroup.
918 _survivor_surv_rate_group->reset();
919 update_young_list_target_length();
920 }
921
922 void G1CollectorPolicy::record_stop_world_start() {
923 _stop_world_start = os::elapsedTime();
924 }
925
926 void G1CollectorPolicy::record_collection_pause_start(double start_time_sec,
927 size_t start_used) {
928 if (PrintGCDetails) {
929 gclog_or_tty->stamp(PrintGCTimeStamps);
930 gclog_or_tty->print("[GC pause");
931 gclog_or_tty->print(" (%s)", full_young_gcs() ? "young" : "partial");
932 }
933
934 // We only need to do this here as the policy will only be applied
935 // to the GC we're about to start. so, no point is calculating this
936 // every time we calculate / recalculate the target young length.
937 update_survivors_policy();
938
939 assert(_g1->used() == _g1->recalculate_used(),
940 err_msg("sanity, used: "SIZE_FORMAT" recalculate_used: "SIZE_FORMAT,
941 _g1->used(), _g1->recalculate_used()));
942
943 double s_w_t_ms = (start_time_sec - _stop_world_start) * 1000.0;
944 _all_stop_world_times_ms->add(s_w_t_ms);
945 _stop_world_start = 0.0;
946
947 _cur_collection_start_sec = start_time_sec;
948 _cur_collection_pause_used_at_start_bytes = start_used;
949 _cur_collection_pause_used_regions_at_start = _g1->used_regions();
950 _pending_cards = _g1->pending_card_num();
951 _max_pending_cards = _g1->max_pending_card_num();
952
953 _bytes_in_collection_set_before_gc = 0;
954 _bytes_copied_during_gc = 0;
955
956 YoungList* young_list = _g1->young_list();
957 _eden_bytes_before_gc = young_list->eden_used_bytes();
958 _survivor_bytes_before_gc = young_list->survivor_used_bytes();
959 _capacity_before_gc = _g1->capacity();
960
961 #ifdef DEBUG
962 // initialise these to something well known so that we can spot
963 // if they are not set properly
964
965 for (int i = 0; i < _parallel_gc_threads; ++i) {
966 _par_last_gc_worker_start_times_ms[i] = -1234.0;
967 _par_last_ext_root_scan_times_ms[i] = -1234.0;
968 _par_last_mark_stack_scan_times_ms[i] = -1234.0;
969 _par_last_update_rs_times_ms[i] = -1234.0;
970 _par_last_update_rs_processed_buffers[i] = -1234.0;
971 _par_last_scan_rs_times_ms[i] = -1234.0;
972 _par_last_obj_copy_times_ms[i] = -1234.0;
973 _par_last_termination_times_ms[i] = -1234.0;
974 _par_last_termination_attempts[i] = -1234.0;
975 _par_last_gc_worker_end_times_ms[i] = -1234.0;
976 _par_last_gc_worker_times_ms[i] = -1234.0;
977 }
978 #endif
979
980 for (int i = 0; i < _aux_num; ++i) {
981 _cur_aux_times_ms[i] = 0.0;
982 _cur_aux_times_set[i] = false;
983 }
984
985 _satb_drain_time_set = false;
986 _last_satb_drain_processed_buffers = -1;
987
988 _last_young_gc_full = false;
989
990 // do that for any other surv rate groups
991 _short_lived_surv_rate_group->stop_adding_regions();
992 _survivors_age_table.clear();
993
994 assert( verify_young_ages(), "region age verification" );
995 }
996
997 void G1CollectorPolicy::record_mark_closure_time(double mark_closure_time_ms) {
998 _mark_closure_time_ms = mark_closure_time_ms;
999 }
1000
1001 void G1CollectorPolicy::record_concurrent_mark_init_end(double
1002 mark_init_elapsed_time_ms) {
1003 _during_marking = true;
1004 assert(!initiate_conc_mark_if_possible(), "we should have cleared it by now");
1005 clear_during_initial_mark_pause();
1006 _cur_mark_stop_world_time_ms = mark_init_elapsed_time_ms;
1007 }
1008
1009 void G1CollectorPolicy::record_concurrent_mark_remark_start() {
1010 _mark_remark_start_sec = os::elapsedTime();
1011 _during_marking = false;
1012 }
1013
1014 void G1CollectorPolicy::record_concurrent_mark_remark_end() {
1015 double end_time_sec = os::elapsedTime();
1016 double elapsed_time_ms = (end_time_sec - _mark_remark_start_sec)*1000.0;
1017 _concurrent_mark_remark_times_ms->add(elapsed_time_ms);
1018 _cur_mark_stop_world_time_ms += elapsed_time_ms;
1019 _prev_collection_pause_end_ms += elapsed_time_ms;
1020
1021 _mmu_tracker->add_pause(_mark_remark_start_sec, end_time_sec, true);
1022 }
1023
1024 void G1CollectorPolicy::record_concurrent_mark_cleanup_start() {
1025 _mark_cleanup_start_sec = os::elapsedTime();
1026 }
1027
1028 void
1029 G1CollectorPolicy::record_concurrent_mark_cleanup_end(size_t freed_bytes,
1030 size_t max_live_bytes) {
1031 record_concurrent_mark_cleanup_end_work1(freed_bytes, max_live_bytes);
1032 record_concurrent_mark_cleanup_end_work2();
1033 }
1034
1035 void
1036 G1CollectorPolicy::
1037 record_concurrent_mark_cleanup_end_work1(size_t freed_bytes,
1038 size_t max_live_bytes) {
1039 if (_n_marks < 2) {
1040 _n_marks++;
1041 }
1042 }
1043
1044 // The important thing about this is that it includes "os::elapsedTime".
1045 void G1CollectorPolicy::record_concurrent_mark_cleanup_end_work2() {
1046 double end_time_sec = os::elapsedTime();
1047 double elapsed_time_ms = (end_time_sec - _mark_cleanup_start_sec)*1000.0;
1048 _concurrent_mark_cleanup_times_ms->add(elapsed_time_ms);
1049 _cur_mark_stop_world_time_ms += elapsed_time_ms;
1050 _prev_collection_pause_end_ms += elapsed_time_ms;
1051
1052 _mmu_tracker->add_pause(_mark_cleanup_start_sec, end_time_sec, true);
1053
1054 _num_markings++;
1055 _n_pauses_at_mark_end = _n_pauses;
1056 _n_marks_since_last_pause++;
1057 }
1058
1059 void
1060 G1CollectorPolicy::record_concurrent_mark_cleanup_completed() {
1061 _should_revert_to_full_young_gcs = false;
1062 _last_full_young_gc = true;
1063 _in_marking_window = false;
1064 }
1065
1066 void G1CollectorPolicy::record_concurrent_pause() {
1067 if (_stop_world_start > 0.0) {
1068 double yield_ms = (os::elapsedTime() - _stop_world_start) * 1000.0;
1069 _all_yield_times_ms->add(yield_ms);
1070 }
1071 }
1072
1073 void G1CollectorPolicy::record_concurrent_pause_end() {
1074 }
1075
1076 template<class T>
1077 T sum_of(T* sum_arr, int start, int n, int N) {
1078 T sum = (T)0;
1079 for (int i = 0; i < n; i++) {
1080 int j = (start + i) % N;
1081 sum += sum_arr[j];
1082 }
1083 return sum;
1084 }
1085
1086 void G1CollectorPolicy::print_par_stats(int level,
1087 const char* str,
1088 double* data) {
1089 double min = data[0], max = data[0];
1090 double total = 0.0;
1091 LineBuffer buf(level);
1092 buf.append("[%s (ms):", str);
1093 for (uint i = 0; i < ParallelGCThreads; ++i) {
1094 double val = data[i];
1095 if (val < min)
1096 min = val;
1097 if (val > max)
1098 max = val;
1099 total += val;
1100 buf.append(" %3.1lf", val);
1101 }
1102 buf.append_and_print_cr("");
1103 double avg = total / (double) ParallelGCThreads;
1104 buf.append_and_print_cr(" Avg: %5.1lf, Min: %5.1lf, Max: %5.1lf, Diff: %5.1lf]",
1105 avg, min, max, max - min);
1106 }
1107
1108 void G1CollectorPolicy::print_par_sizes(int level,
1109 const char* str,
1110 double* data) {
1111 double min = data[0], max = data[0];
1112 double total = 0.0;
1113 LineBuffer buf(level);
1114 buf.append("[%s :", str);
1115 for (uint i = 0; i < ParallelGCThreads; ++i) {
1116 double val = data[i];
1117 if (val < min)
1118 min = val;
1119 if (val > max)
1120 max = val;
1121 total += val;
1122 buf.append(" %d", (int) val);
1123 }
1124 buf.append_and_print_cr("");
1125 double avg = total / (double) ParallelGCThreads;
1126 buf.append_and_print_cr(" Sum: %d, Avg: %d, Min: %d, Max: %d, Diff: %d]",
1127 (int)total, (int)avg, (int)min, (int)max, (int)max - (int)min);
1128 }
1129
1130 void G1CollectorPolicy::print_stats (int level,
1131 const char* str,
1132 double value) {
1133 LineBuffer(level).append_and_print_cr("[%s: %5.1lf ms]", str, value);
1134 }
1135
1136 void G1CollectorPolicy::print_stats (int level,
1137 const char* str,
1138 int value) {
1139 LineBuffer(level).append_and_print_cr("[%s: %d]", str, value);
1140 }
1141
1142 double G1CollectorPolicy::avg_value (double* data) {
1143 if (G1CollectedHeap::use_parallel_gc_threads()) {
1144 double ret = 0.0;
1145 for (uint i = 0; i < ParallelGCThreads; ++i)
1146 ret += data[i];
1147 return ret / (double) ParallelGCThreads;
1148 } else {
1149 return data[0];
1150 }
1151 }
1152
1153 double G1CollectorPolicy::max_value (double* data) {
1154 if (G1CollectedHeap::use_parallel_gc_threads()) {
1155 double ret = data[0];
1156 for (uint i = 1; i < ParallelGCThreads; ++i)
1157 if (data[i] > ret)
1158 ret = data[i];
1159 return ret;
1160 } else {
1161 return data[0];
1162 }
1163 }
1164
1165 double G1CollectorPolicy::sum_of_values (double* data) {
1166 if (G1CollectedHeap::use_parallel_gc_threads()) {
1167 double sum = 0.0;
1168 for (uint i = 0; i < ParallelGCThreads; i++)
1169 sum += data[i];
1170 return sum;
1171 } else {
1172 return data[0];
1173 }
1174 }
1175
1176 double G1CollectorPolicy::max_sum (double* data1,
1177 double* data2) {
1178 double ret = data1[0] + data2[0];
1179
1180 if (G1CollectedHeap::use_parallel_gc_threads()) {
1181 for (uint i = 1; i < ParallelGCThreads; ++i) {
1182 double data = data1[i] + data2[i];
1183 if (data > ret)
1184 ret = data;
1185 }
1186 }
1187 return ret;
1188 }
1189
1190 // Anything below that is considered to be zero
1191 #define MIN_TIMER_GRANULARITY 0.0000001
1192
1193 void G1CollectorPolicy::record_collection_pause_end() {
1194 double end_time_sec = os::elapsedTime();
1195 double elapsed_ms = _last_pause_time_ms;
1196 bool parallel = G1CollectedHeap::use_parallel_gc_threads();
1197 size_t rs_size =
1198 _cur_collection_pause_used_regions_at_start - collection_set_size();
1199 size_t cur_used_bytes = _g1->used();
1200 assert(cur_used_bytes == _g1->recalculate_used(), "It should!");
1201 bool last_pause_included_initial_mark = false;
1202 bool update_stats = !_g1->evacuation_failed();
1203
1204 #ifndef PRODUCT
1205 if (G1YoungSurvRateVerbose) {
1206 gclog_or_tty->print_cr("");
1207 _short_lived_surv_rate_group->print();
1208 // do that for any other surv rate groups too
1209 }
1210 #endif // PRODUCT
1211
1212 last_pause_included_initial_mark = during_initial_mark_pause();
1213 if (last_pause_included_initial_mark)
1214 record_concurrent_mark_init_end(0.0);
1215
1216 size_t marking_initiating_used_threshold =
1217 (_g1->capacity() / 100) * InitiatingHeapOccupancyPercent;
1218
1219 if (!_g1->mark_in_progress() && !_last_full_young_gc) {
1220 assert(!last_pause_included_initial_mark, "invariant");
1221 if (cur_used_bytes > marking_initiating_used_threshold) {
1222 if (cur_used_bytes > _prev_collection_pause_used_at_end_bytes) {
1223 assert(!during_initial_mark_pause(), "we should not see this here");
1224
1225 ergo_verbose3(ErgoConcCycles,
1226 "request concurrent cycle initiation",
1227 ergo_format_reason("occupancy higher than threshold")
1228 ergo_format_byte("occupancy")
1229 ergo_format_byte_perc("threshold"),
1230 cur_used_bytes,
1231 marking_initiating_used_threshold,
1232 (double) InitiatingHeapOccupancyPercent);
1233
1234 // Note: this might have already been set, if during the last
1235 // pause we decided to start a cycle but at the beginning of
1236 // this pause we decided to postpone it. That's OK.
1237 set_initiate_conc_mark_if_possible();
1238 } else {
1239 ergo_verbose2(ErgoConcCycles,
1240 "do not request concurrent cycle initiation",
1241 ergo_format_reason("occupancy lower than previous occupancy")
1242 ergo_format_byte("occupancy")
1243 ergo_format_byte("previous occupancy"),
1244 cur_used_bytes,
1245 _prev_collection_pause_used_at_end_bytes);
1246 }
1247 }
1248 }
1249
1250 _prev_collection_pause_used_at_end_bytes = cur_used_bytes;
1251
1252 _mmu_tracker->add_pause(end_time_sec - elapsed_ms/1000.0,
1253 end_time_sec, false);
1254
1255 guarantee(_cur_collection_pause_used_regions_at_start >=
1256 collection_set_size(),
1257 "Negative RS size?");
1258
1259 // This assert is exempted when we're doing parallel collection pauses,
1260 // because the fragmentation caused by the parallel GC allocation buffers
1261 // can lead to more memory being used during collection than was used
1262 // before. Best leave this out until the fragmentation problem is fixed.
1263 // Pauses in which evacuation failed can also lead to negative
1264 // collections, since no space is reclaimed from a region containing an
1265 // object whose evacuation failed.
1266 // Further, we're now always doing parallel collection. But I'm still
1267 // leaving this here as a placeholder for a more precise assertion later.
1268 // (DLD, 10/05.)
1269 assert((true || parallel) // Always using GC LABs now.
1270 || _g1->evacuation_failed()
1271 || _cur_collection_pause_used_at_start_bytes >= cur_used_bytes,
1272 "Negative collection");
1273
1274 size_t freed_bytes =
1275 _cur_collection_pause_used_at_start_bytes - cur_used_bytes;
1276 size_t surviving_bytes = _collection_set_bytes_used_before - freed_bytes;
1277
1278 double survival_fraction =
1279 (double)surviving_bytes/
1280 (double)_collection_set_bytes_used_before;
1281
1282 _n_pauses++;
1283
1284 double ext_root_scan_time = avg_value(_par_last_ext_root_scan_times_ms);
1285 double mark_stack_scan_time = avg_value(_par_last_mark_stack_scan_times_ms);
1286 double update_rs_time = avg_value(_par_last_update_rs_times_ms);
1287 double update_rs_processed_buffers =
1288 sum_of_values(_par_last_update_rs_processed_buffers);
1289 double scan_rs_time = avg_value(_par_last_scan_rs_times_ms);
1290 double obj_copy_time = avg_value(_par_last_obj_copy_times_ms);
1291 double termination_time = avg_value(_par_last_termination_times_ms);
1292
1293 double parallel_known_time = update_rs_time +
1294 ext_root_scan_time +
1295 mark_stack_scan_time +
1296 scan_rs_time +
1297 obj_copy_time +
1298 termination_time;
1299
1300 double parallel_other_time = _cur_collection_par_time_ms - parallel_known_time;
1301
1302 PauseSummary* summary = _summary;
1303
1304 if (update_stats) {
1305 _recent_rs_scan_times_ms->add(scan_rs_time);
1306 _recent_pause_times_ms->add(elapsed_ms);
1307 _recent_rs_sizes->add(rs_size);
1308
1309 MainBodySummary* body_summary = summary->main_body_summary();
1310 guarantee(body_summary != NULL, "should not be null!");
1311
1312 if (_satb_drain_time_set)
1313 body_summary->record_satb_drain_time_ms(_cur_satb_drain_time_ms);
1314 else
1315 body_summary->record_satb_drain_time_ms(0.0);
1316
1317 body_summary->record_ext_root_scan_time_ms(ext_root_scan_time);
1318 body_summary->record_mark_stack_scan_time_ms(mark_stack_scan_time);
1319 body_summary->record_update_rs_time_ms(update_rs_time);
1320 body_summary->record_scan_rs_time_ms(scan_rs_time);
1321 body_summary->record_obj_copy_time_ms(obj_copy_time);
1322 if (parallel) {
1323 body_summary->record_parallel_time_ms(_cur_collection_par_time_ms);
1324 body_summary->record_clear_ct_time_ms(_cur_clear_ct_time_ms);
1325 body_summary->record_termination_time_ms(termination_time);
1326 body_summary->record_parallel_other_time_ms(parallel_other_time);
1327 }
1328 body_summary->record_mark_closure_time_ms(_mark_closure_time_ms);
1329
1330 // We exempt parallel collection from this check because Alloc Buffer
1331 // fragmentation can produce negative collections. Same with evac
1332 // failure.
1333 // Further, we're now always doing parallel collection. But I'm still
1334 // leaving this here as a placeholder for a more precise assertion later.
1335 // (DLD, 10/05.
1336 assert((true || parallel)
1337 || _g1->evacuation_failed()
1338 || surviving_bytes <= _collection_set_bytes_used_before,
1339 "Or else negative collection!");
1340 _recent_CS_bytes_used_before->add(_collection_set_bytes_used_before);
1341 _recent_CS_bytes_surviving->add(surviving_bytes);
1342
1343 // this is where we update the allocation rate of the application
1344 double app_time_ms =
1345 (_cur_collection_start_sec * 1000.0 - _prev_collection_pause_end_ms);
1346 if (app_time_ms < MIN_TIMER_GRANULARITY) {
1347 // This usually happens due to the timer not having the required
1348 // granularity. Some Linuxes are the usual culprits.
1349 // We'll just set it to something (arbitrarily) small.
1350 app_time_ms = 1.0;
1351 }
1352 size_t regions_allocated =
1353 (_region_num_young - _prev_region_num_young) +
1354 (_region_num_tenured - _prev_region_num_tenured);
1355 double alloc_rate_ms = (double) regions_allocated / app_time_ms;
1356 _alloc_rate_ms_seq->add(alloc_rate_ms);
1357 _prev_region_num_young = _region_num_young;
1358 _prev_region_num_tenured = _region_num_tenured;
1359
1360 double interval_ms =
1361 (end_time_sec - _recent_prev_end_times_for_all_gcs_sec->oldest()) * 1000.0;
1362 update_recent_gc_times(end_time_sec, elapsed_ms);
1363 _recent_avg_pause_time_ratio = _recent_gc_times_ms->sum()/interval_ms;
1364 if (recent_avg_pause_time_ratio() < 0.0 ||
1365 (recent_avg_pause_time_ratio() - 1.0 > 0.0)) {
1366 #ifndef PRODUCT
1367 // Dump info to allow post-facto debugging
1368 gclog_or_tty->print_cr("recent_avg_pause_time_ratio() out of bounds");
1369 gclog_or_tty->print_cr("-------------------------------------------");
1370 gclog_or_tty->print_cr("Recent GC Times (ms):");
1371 _recent_gc_times_ms->dump();
1372 gclog_or_tty->print_cr("(End Time=%3.3f) Recent GC End Times (s):", end_time_sec);
1373 _recent_prev_end_times_for_all_gcs_sec->dump();
1374 gclog_or_tty->print_cr("GC = %3.3f, Interval = %3.3f, Ratio = %3.3f",
1375 _recent_gc_times_ms->sum(), interval_ms, recent_avg_pause_time_ratio());
1376 // In debug mode, terminate the JVM if the user wants to debug at this point.
1377 assert(!G1FailOnFPError, "Debugging data for CR 6898948 has been dumped above");
1378 #endif // !PRODUCT
1379 // Clip ratio between 0.0 and 1.0, and continue. This will be fixed in
1380 // CR 6902692 by redoing the manner in which the ratio is incrementally computed.
1381 if (_recent_avg_pause_time_ratio < 0.0) {
1382 _recent_avg_pause_time_ratio = 0.0;
1383 } else {
1384 assert(_recent_avg_pause_time_ratio - 1.0 > 0.0, "Ctl-point invariant");
1385 _recent_avg_pause_time_ratio = 1.0;
1386 }
1387 }
1388 }
1389
1390 if (G1PolicyVerbose > 1) {
1391 gclog_or_tty->print_cr(" Recording collection pause(%d)", _n_pauses);
1392 }
1393
1394 if (G1PolicyVerbose > 1) {
1395 gclog_or_tty->print_cr(" ET: %10.6f ms (avg: %10.6f ms)\n"
1396 " ET-RS: %10.6f ms (avg: %10.6f ms)\n"
1397 " |RS|: " SIZE_FORMAT,
1398 elapsed_ms, recent_avg_time_for_pauses_ms(),
1399 scan_rs_time, recent_avg_time_for_rs_scan_ms(),
1400 rs_size);
1401
1402 gclog_or_tty->print_cr(" Used at start: " SIZE_FORMAT"K"
1403 " At end " SIZE_FORMAT "K\n"
1404 " garbage : " SIZE_FORMAT "K"
1405 " of " SIZE_FORMAT "K\n"
1406 " survival : %6.2f%% (%6.2f%% avg)",
1407 _cur_collection_pause_used_at_start_bytes/K,
1408 _g1->used()/K, freed_bytes/K,
1409 _collection_set_bytes_used_before/K,
1410 survival_fraction*100.0,
1411 recent_avg_survival_fraction()*100.0);
1412 gclog_or_tty->print_cr(" Recent %% gc pause time: %6.2f",
1413 recent_avg_pause_time_ratio() * 100.0);
1414 }
1415
1416 double other_time_ms = elapsed_ms;
1417
1418 if (_satb_drain_time_set) {
1419 other_time_ms -= _cur_satb_drain_time_ms;
1420 }
1421
1422 if (parallel) {
1423 other_time_ms -= _cur_collection_par_time_ms + _cur_clear_ct_time_ms;
1424 } else {
1425 other_time_ms -=
1426 update_rs_time +
1427 ext_root_scan_time + mark_stack_scan_time +
1428 scan_rs_time + obj_copy_time;
1429 }
1430
1431 if (PrintGCDetails) {
1432 gclog_or_tty->print_cr("%s, %1.8lf secs]",
1433 (last_pause_included_initial_mark) ? " (initial-mark)" : "",
1434 elapsed_ms / 1000.0);
1435
1436 if (_satb_drain_time_set) {
1437 print_stats(1, "SATB Drain Time", _cur_satb_drain_time_ms);
1438 }
1439 if (_last_satb_drain_processed_buffers >= 0) {
1440 print_stats(2, "Processed Buffers", _last_satb_drain_processed_buffers);
1441 }
1442 if (parallel) {
1443 print_stats(1, "Parallel Time", _cur_collection_par_time_ms);
1444 print_par_stats(2, "GC Worker Start Time", _par_last_gc_worker_start_times_ms);
1445 print_par_stats(2, "Update RS", _par_last_update_rs_times_ms);
1446 print_par_sizes(3, "Processed Buffers", _par_last_update_rs_processed_buffers);
1447 print_par_stats(2, "Ext Root Scanning", _par_last_ext_root_scan_times_ms);
1448 print_par_stats(2, "Mark Stack Scanning", _par_last_mark_stack_scan_times_ms);
1449 print_par_stats(2, "Scan RS", _par_last_scan_rs_times_ms);
1450 print_par_stats(2, "Object Copy", _par_last_obj_copy_times_ms);
1451 print_par_stats(2, "Termination", _par_last_termination_times_ms);
1452 print_par_sizes(3, "Termination Attempts", _par_last_termination_attempts);
1453 print_par_stats(2, "GC Worker End Time", _par_last_gc_worker_end_times_ms);
1454
1455 for (int i = 0; i < _parallel_gc_threads; i++) {
1456 _par_last_gc_worker_times_ms[i] = _par_last_gc_worker_end_times_ms[i] - _par_last_gc_worker_start_times_ms[i];
1457 }
1458 print_par_stats(2, "GC Worker Times", _par_last_gc_worker_times_ms);
1459
1460 print_stats(2, "Parallel Other", parallel_other_time);
1461 print_stats(1, "Clear CT", _cur_clear_ct_time_ms);
1462 } else {
1463 print_stats(1, "Update RS", update_rs_time);
1464 print_stats(2, "Processed Buffers",
1465 (int)update_rs_processed_buffers);
1466 print_stats(1, "Ext Root Scanning", ext_root_scan_time);
1467 print_stats(1, "Mark Stack Scanning", mark_stack_scan_time);
1468 print_stats(1, "Scan RS", scan_rs_time);
1469 print_stats(1, "Object Copying", obj_copy_time);
1470 }
1471 #ifndef PRODUCT
1472 print_stats(1, "Cur Clear CC", _cur_clear_cc_time_ms);
1473 print_stats(1, "Cum Clear CC", _cum_clear_cc_time_ms);
1474 print_stats(1, "Min Clear CC", _min_clear_cc_time_ms);
1475 print_stats(1, "Max Clear CC", _max_clear_cc_time_ms);
1476 if (_num_cc_clears > 0) {
1477 print_stats(1, "Avg Clear CC", _cum_clear_cc_time_ms / ((double)_num_cc_clears));
1478 }
1479 #endif
1480 print_stats(1, "Other", other_time_ms);
1481 print_stats(2, "Choose CSet", _recorded_young_cset_choice_time_ms);
1482
1483 for (int i = 0; i < _aux_num; ++i) {
1484 if (_cur_aux_times_set[i]) {
1485 char buffer[96];
1486 sprintf(buffer, "Aux%d", i);
1487 print_stats(1, buffer, _cur_aux_times_ms[i]);
1488 }
1489 }
1490 }
1491
1492 _all_pause_times_ms->add(elapsed_ms);
1493 if (update_stats) {
1494 summary->record_total_time_ms(elapsed_ms);
1495 summary->record_other_time_ms(other_time_ms);
1496 }
1497 for (int i = 0; i < _aux_num; ++i)
1498 if (_cur_aux_times_set[i])
1499 _all_aux_times_ms[i].add(_cur_aux_times_ms[i]);
1500
1501 // Reset marks-between-pauses counter.
1502 _n_marks_since_last_pause = 0;
1503
1504 // Update the efficiency-since-mark vars.
1505 double proc_ms = elapsed_ms * (double) _parallel_gc_threads;
1506 if (elapsed_ms < MIN_TIMER_GRANULARITY) {
1507 // This usually happens due to the timer not having the required
1508 // granularity. Some Linuxes are the usual culprits.
1509 // We'll just set it to something (arbitrarily) small.
1510 proc_ms = 1.0;
1511 }
1512 double cur_efficiency = (double) freed_bytes / proc_ms;
1513
1514 bool new_in_marking_window = _in_marking_window;
1515 bool new_in_marking_window_im = false;
1516 if (during_initial_mark_pause()) {
1517 new_in_marking_window = true;
1518 new_in_marking_window_im = true;
1519 }
1520
1521 if (_last_full_young_gc) {
1522 ergo_verbose2(ErgoPartiallyYoungGCs,
1523 "start partially-young GCs",
1524 ergo_format_byte_perc("known garbage"),
1525 _known_garbage_bytes, _known_garbage_ratio * 100.0);
1526 set_full_young_gcs(false);
1527 _last_full_young_gc = false;
1528 }
1529
1530 if ( !_last_young_gc_full ) {
1531 if (_should_revert_to_full_young_gcs) {
1532 ergo_verbose2(ErgoPartiallyYoungGCs,
1533 "end partially-young GCs",
1534 ergo_format_reason("partially-young GCs end requested")
1535 ergo_format_byte_perc("known garbage"),
1536 _known_garbage_bytes, _known_garbage_ratio * 100.0);
1537 set_full_young_gcs(true);
1538 } else if (_known_garbage_ratio < 0.05) {
1539 ergo_verbose3(ErgoPartiallyYoungGCs,
1540 "end partially-young GCs",
1541 ergo_format_reason("known garbage percent lower than threshold")
1542 ergo_format_byte_perc("known garbage")
1543 ergo_format_perc("threshold"),
1544 _known_garbage_bytes, _known_garbage_ratio * 100.0,
1545 0.05 * 100.0);
1546 set_full_young_gcs(true);
1547 } else if (adaptive_young_list_length() &&
1548 (get_gc_eff_factor() * cur_efficiency < predict_young_gc_eff())) {
1549 ergo_verbose5(ErgoPartiallyYoungGCs,
1550 "end partially-young GCs",
1551 ergo_format_reason("current GC efficiency lower than "
1552 "predicted fully-young GC efficiency")
1553 ergo_format_double("GC efficiency factor")
1554 ergo_format_double("current GC efficiency")
1555 ergo_format_double("predicted fully-young GC efficiency")
1556 ergo_format_byte_perc("known garbage"),
1557 get_gc_eff_factor(), cur_efficiency,
1558 predict_young_gc_eff(),
1559 _known_garbage_bytes, _known_garbage_ratio * 100.0);
1560 set_full_young_gcs(true);
1561 }
1562 }
1563 _should_revert_to_full_young_gcs = false;
1564
1565 if (_last_young_gc_full && !_during_marking) {
1566 _young_gc_eff_seq->add(cur_efficiency);
1567 }
1568
1569 _short_lived_surv_rate_group->start_adding_regions();
1570 // do that for any other surv rate groupsx
1571
1572 // <NEW PREDICTION>
1573
1574 if (update_stats) {
1575 double pause_time_ms = elapsed_ms;
1576
1577 size_t diff = 0;
1578 if (_max_pending_cards >= _pending_cards)
1579 diff = _max_pending_cards - _pending_cards;
1580 _pending_card_diff_seq->add((double) diff);
1581
1582 double cost_per_card_ms = 0.0;
1583 if (_pending_cards > 0) {
1584 cost_per_card_ms = update_rs_time / (double) _pending_cards;
1585 _cost_per_card_ms_seq->add(cost_per_card_ms);
1586 }
1587
1588 size_t cards_scanned = _g1->cards_scanned();
1589
1590 double cost_per_entry_ms = 0.0;
1591 if (cards_scanned > 10) {
1592 cost_per_entry_ms = scan_rs_time / (double) cards_scanned;
1593 if (_last_young_gc_full)
1594 _cost_per_entry_ms_seq->add(cost_per_entry_ms);
1595 else
1596 _partially_young_cost_per_entry_ms_seq->add(cost_per_entry_ms);
1597 }
1598
1599 if (_max_rs_lengths > 0) {
1600 double cards_per_entry_ratio =
1601 (double) cards_scanned / (double) _max_rs_lengths;
1602 if (_last_young_gc_full)
1603 _fully_young_cards_per_entry_ratio_seq->add(cards_per_entry_ratio);
1604 else
1605 _partially_young_cards_per_entry_ratio_seq->add(cards_per_entry_ratio);
1606 }
1607
1608 size_t rs_length_diff = _max_rs_lengths - _recorded_rs_lengths;
1609 if (rs_length_diff >= 0)
1610 _rs_length_diff_seq->add((double) rs_length_diff);
1611
1612 size_t copied_bytes = surviving_bytes;
1613 double cost_per_byte_ms = 0.0;
1614 if (copied_bytes > 0) {
1615 cost_per_byte_ms = obj_copy_time / (double) copied_bytes;
1616 if (_in_marking_window)
1617 _cost_per_byte_ms_during_cm_seq->add(cost_per_byte_ms);
1618 else
1619 _cost_per_byte_ms_seq->add(cost_per_byte_ms);
1620 }
1621
1622 double all_other_time_ms = pause_time_ms -
1623 (update_rs_time + scan_rs_time + obj_copy_time +
1624 _mark_closure_time_ms + termination_time);
1625
1626 double young_other_time_ms = 0.0;
1627 if (_recorded_young_regions > 0) {
1628 young_other_time_ms =
1629 _recorded_young_cset_choice_time_ms +
1630 _recorded_young_free_cset_time_ms;
1631 _young_other_cost_per_region_ms_seq->add(young_other_time_ms /
1632 (double) _recorded_young_regions);
1633 }
1634 double non_young_other_time_ms = 0.0;
1635 if (_recorded_non_young_regions > 0) {
1636 non_young_other_time_ms =
1637 _recorded_non_young_cset_choice_time_ms +
1638 _recorded_non_young_free_cset_time_ms;
1639
1640 _non_young_other_cost_per_region_ms_seq->add(non_young_other_time_ms /
1641 (double) _recorded_non_young_regions);
1642 }
1643
1644 double constant_other_time_ms = all_other_time_ms -
1645 (young_other_time_ms + non_young_other_time_ms);
1646 _constant_other_time_ms_seq->add(constant_other_time_ms);
1647
1648 double survival_ratio = 0.0;
1649 if (_bytes_in_collection_set_before_gc > 0) {
1650 survival_ratio = (double) _bytes_copied_during_gc /
1651 (double) _bytes_in_collection_set_before_gc;
1652 }
1653
1654 _pending_cards_seq->add((double) _pending_cards);
1655 _scanned_cards_seq->add((double) cards_scanned);
1656 _rs_lengths_seq->add((double) _max_rs_lengths);
1657
1658 double expensive_region_limit_ms =
1659 (double) MaxGCPauseMillis - predict_constant_other_time_ms();
1660 if (expensive_region_limit_ms < 0.0) {
1661 // this means that the other time was predicted to be longer than
1662 // than the max pause time
1663 expensive_region_limit_ms = (double) MaxGCPauseMillis;
1664 }
1665 _expensive_region_limit_ms = expensive_region_limit_ms;
1666
1667 if (PREDICTIONS_VERBOSE) {
1668 gclog_or_tty->print_cr("");
1669 gclog_or_tty->print_cr("PREDICTIONS %1.4lf %d "
1670 "REGIONS %d %d %d "
1671 "PENDING_CARDS %d %d "
1672 "CARDS_SCANNED %d %d "
1673 "RS_LENGTHS %d %d "
1674 "RS_UPDATE %1.6lf %1.6lf RS_SCAN %1.6lf %1.6lf "
1675 "SURVIVAL_RATIO %1.6lf %1.6lf "
1676 "OBJECT_COPY %1.6lf %1.6lf OTHER_CONSTANT %1.6lf %1.6lf "
1677 "OTHER_YOUNG %1.6lf %1.6lf "
1678 "OTHER_NON_YOUNG %1.6lf %1.6lf "
1679 "VTIME_DIFF %1.6lf TERMINATION %1.6lf "
1680 "ELAPSED %1.6lf %1.6lf ",
1681 _cur_collection_start_sec,
1682 (!_last_young_gc_full) ? 2 :
1683 (last_pause_included_initial_mark) ? 1 : 0,
1684 _recorded_region_num,
1685 _recorded_young_regions,
1686 _recorded_non_young_regions,
1687 _predicted_pending_cards, _pending_cards,
1688 _predicted_cards_scanned, cards_scanned,
1689 _predicted_rs_lengths, _max_rs_lengths,
1690 _predicted_rs_update_time_ms, update_rs_time,
1691 _predicted_rs_scan_time_ms, scan_rs_time,
1692 _predicted_survival_ratio, survival_ratio,
1693 _predicted_object_copy_time_ms, obj_copy_time,
1694 _predicted_constant_other_time_ms, constant_other_time_ms,
1695 _predicted_young_other_time_ms, young_other_time_ms,
1696 _predicted_non_young_other_time_ms,
1697 non_young_other_time_ms,
1698 _vtime_diff_ms, termination_time,
1699 _predicted_pause_time_ms, elapsed_ms);
1700 }
1701
1702 if (G1PolicyVerbose > 0) {
1703 gclog_or_tty->print_cr("Pause Time, predicted: %1.4lfms (predicted %s), actual: %1.4lfms",
1704 _predicted_pause_time_ms,
1705 (_within_target) ? "within" : "outside",
1706 elapsed_ms);
1707 }
1708
1709 }
1710
1711 _in_marking_window = new_in_marking_window;
1712 _in_marking_window_im = new_in_marking_window_im;
1713 _free_regions_at_end_of_collection = _g1->free_regions();
1714 update_young_list_target_length();
1715
1716 // Note that _mmu_tracker->max_gc_time() returns the time in seconds.
1717 double update_rs_time_goal_ms = _mmu_tracker->max_gc_time() * MILLIUNITS * G1RSetUpdatingPauseTimePercent / 100.0;
1718 adjust_concurrent_refinement(update_rs_time, update_rs_processed_buffers, update_rs_time_goal_ms);
1719 // </NEW PREDICTION>
1720 }
1721
1722 #define EXT_SIZE_FORMAT "%d%s"
1723 #define EXT_SIZE_PARAMS(bytes) \
1724 byte_size_in_proper_unit((bytes)), \
1725 proper_unit_for_byte_size((bytes))
1726
1727 void G1CollectorPolicy::print_heap_transition() {
1728 if (PrintGCDetails) {
1729 YoungList* young_list = _g1->young_list();
1730 size_t eden_bytes = young_list->eden_used_bytes();
1731 size_t survivor_bytes = young_list->survivor_used_bytes();
1732 size_t used_before_gc = _cur_collection_pause_used_at_start_bytes;
1733 size_t used = _g1->used();
1734 size_t capacity = _g1->capacity();
1735 size_t eden_capacity =
1736 (_young_list_target_length * HeapRegion::GrainBytes) - survivor_bytes;
1737
1738 gclog_or_tty->print_cr(
1739 " [Eden: "EXT_SIZE_FORMAT"("EXT_SIZE_FORMAT")->"EXT_SIZE_FORMAT"("EXT_SIZE_FORMAT") "
1740 "Survivors: "EXT_SIZE_FORMAT"->"EXT_SIZE_FORMAT" "
1741 "Heap: "EXT_SIZE_FORMAT"("EXT_SIZE_FORMAT")->"
1742 EXT_SIZE_FORMAT"("EXT_SIZE_FORMAT")]",
1743 EXT_SIZE_PARAMS(_eden_bytes_before_gc),
1744 EXT_SIZE_PARAMS(_prev_eden_capacity),
1745 EXT_SIZE_PARAMS(eden_bytes),
1746 EXT_SIZE_PARAMS(eden_capacity),
1747 EXT_SIZE_PARAMS(_survivor_bytes_before_gc),
1748 EXT_SIZE_PARAMS(survivor_bytes),
1749 EXT_SIZE_PARAMS(used_before_gc),
1750 EXT_SIZE_PARAMS(_capacity_before_gc),
1751 EXT_SIZE_PARAMS(used),
1752 EXT_SIZE_PARAMS(capacity));
1753
1754 _prev_eden_capacity = eden_capacity;
1755 } else if (PrintGC) {
1756 _g1->print_size_transition(gclog_or_tty,
1757 _cur_collection_pause_used_at_start_bytes,
1758 _g1->used(), _g1->capacity());
1759 }
1760 }
1761
1762 // <NEW PREDICTION>
1763
1764 void G1CollectorPolicy::adjust_concurrent_refinement(double update_rs_time,
1765 double update_rs_processed_buffers,
1766 double goal_ms) {
1767 DirtyCardQueueSet& dcqs = JavaThread::dirty_card_queue_set();
1768 ConcurrentG1Refine *cg1r = G1CollectedHeap::heap()->concurrent_g1_refine();
1769
1770 if (G1UseAdaptiveConcRefinement) {
1771 const int k_gy = 3, k_gr = 6;
1772 const double inc_k = 1.1, dec_k = 0.9;
1773
1774 int g = cg1r->green_zone();
1775 if (update_rs_time > goal_ms) {
1776 g = (int)(g * dec_k); // Can become 0, that's OK. That would mean a mutator-only processing.
1777 } else {
1778 if (update_rs_time < goal_ms && update_rs_processed_buffers > g) {
1779 g = (int)MAX2(g * inc_k, g + 1.0);
1780 }
1781 }
1782 // Change the refinement threads params
1783 cg1r->set_green_zone(g);
1784 cg1r->set_yellow_zone(g * k_gy);
1785 cg1r->set_red_zone(g * k_gr);
1786 cg1r->reinitialize_threads();
1787
1788 int processing_threshold_delta = MAX2((int)(cg1r->green_zone() * sigma()), 1);
1789 int processing_threshold = MIN2(cg1r->green_zone() + processing_threshold_delta,
1790 cg1r->yellow_zone());
1791 // Change the barrier params
1792 dcqs.set_process_completed_threshold(processing_threshold);
1793 dcqs.set_max_completed_queue(cg1r->red_zone());
1794 }
1795
1796 int curr_queue_size = dcqs.completed_buffers_num();
1797 if (curr_queue_size >= cg1r->yellow_zone()) {
1798 dcqs.set_completed_queue_padding(curr_queue_size);
1799 } else {
1800 dcqs.set_completed_queue_padding(0);
1801 }
1802 dcqs.notify_if_necessary();
1803 }
1804
1805 double
1806 G1CollectorPolicy::
1807 predict_young_collection_elapsed_time_ms(size_t adjustment) {
1808 guarantee( adjustment == 0 || adjustment == 1, "invariant" );
1809
1810 G1CollectedHeap* g1h = G1CollectedHeap::heap();
1811 size_t young_num = g1h->young_list()->length();
1812 if (young_num == 0)
1813 return 0.0;
1814
1815 young_num += adjustment;
1816 size_t pending_cards = predict_pending_cards();
1817 size_t rs_lengths = g1h->young_list()->sampled_rs_lengths() +
1818 predict_rs_length_diff();
1819 size_t card_num;
1820 if (full_young_gcs())
1821 card_num = predict_young_card_num(rs_lengths);
1822 else
1823 card_num = predict_non_young_card_num(rs_lengths);
1824 size_t young_byte_size = young_num * HeapRegion::GrainBytes;
1825 double accum_yg_surv_rate =
1826 _short_lived_surv_rate_group->accum_surv_rate(adjustment);
1827
1828 size_t bytes_to_copy =
1829 (size_t) (accum_yg_surv_rate * (double) HeapRegion::GrainBytes);
1830
1831 return
1832 predict_rs_update_time_ms(pending_cards) +
1833 predict_rs_scan_time_ms(card_num) +
1834 predict_object_copy_time_ms(bytes_to_copy) +
1835 predict_young_other_time_ms(young_num) +
1836 predict_constant_other_time_ms();
1837 }
1838
1839 double
1840 G1CollectorPolicy::predict_base_elapsed_time_ms(size_t pending_cards) {
1841 size_t rs_length = predict_rs_length_diff();
1842 size_t card_num;
1843 if (full_young_gcs())
1844 card_num = predict_young_card_num(rs_length);
1845 else
1846 card_num = predict_non_young_card_num(rs_length);
1847 return predict_base_elapsed_time_ms(pending_cards, card_num);
1848 }
1849
1850 double
1851 G1CollectorPolicy::predict_base_elapsed_time_ms(size_t pending_cards,
1852 size_t scanned_cards) {
1853 return
1854 predict_rs_update_time_ms(pending_cards) +
1855 predict_rs_scan_time_ms(scanned_cards) +
1856 predict_constant_other_time_ms();
1857 }
1858
1859 double
1860 G1CollectorPolicy::predict_region_elapsed_time_ms(HeapRegion* hr,
1861 bool young) {
1862 size_t rs_length = hr->rem_set()->occupied();
1863 size_t card_num;
1864 if (full_young_gcs())
1865 card_num = predict_young_card_num(rs_length);
1866 else
1867 card_num = predict_non_young_card_num(rs_length);
1868 size_t bytes_to_copy = predict_bytes_to_copy(hr);
1869
1870 double region_elapsed_time_ms =
1871 predict_rs_scan_time_ms(card_num) +
1872 predict_object_copy_time_ms(bytes_to_copy);
1873
1874 if (young)
1875 region_elapsed_time_ms += predict_young_other_time_ms(1);
1876 else
1877 region_elapsed_time_ms += predict_non_young_other_time_ms(1);
1878
1879 return region_elapsed_time_ms;
1880 }
1881
1882 size_t
1883 G1CollectorPolicy::predict_bytes_to_copy(HeapRegion* hr) {
1884 size_t bytes_to_copy;
1885 if (hr->is_marked())
1886 bytes_to_copy = hr->max_live_bytes();
1887 else {
1888 guarantee( hr->is_young() && hr->age_in_surv_rate_group() != -1,
1889 "invariant" );
1890 int age = hr->age_in_surv_rate_group();
1891 double yg_surv_rate = predict_yg_surv_rate(age, hr->surv_rate_group());
1892 bytes_to_copy = (size_t) ((double) hr->used() * yg_surv_rate);
1893 }
1894
1895 return bytes_to_copy;
1896 }
1897
1898 void
1899 G1CollectorPolicy::start_recording_regions() {
1900 _recorded_rs_lengths = 0;
1901 _recorded_young_regions = 0;
1902 _recorded_non_young_regions = 0;
1903
1904 #if PREDICTIONS_VERBOSE
1905 _recorded_marked_bytes = 0;
1906 _recorded_young_bytes = 0;
1907 _predicted_bytes_to_copy = 0;
1908 _predicted_rs_lengths = 0;
1909 _predicted_cards_scanned = 0;
1910 #endif // PREDICTIONS_VERBOSE
1911 }
1912
1913 void
1914 G1CollectorPolicy::record_cset_region_info(HeapRegion* hr, bool young) {
1915 #if PREDICTIONS_VERBOSE
1916 if (!young) {
1917 _recorded_marked_bytes += hr->max_live_bytes();
1918 }
1919 _predicted_bytes_to_copy += predict_bytes_to_copy(hr);
1920 #endif // PREDICTIONS_VERBOSE
1921
1922 size_t rs_length = hr->rem_set()->occupied();
1923 _recorded_rs_lengths += rs_length;
1924 }
1925
1926 void
1927 G1CollectorPolicy::record_non_young_cset_region(HeapRegion* hr) {
1928 assert(!hr->is_young(), "should not call this");
1929 ++_recorded_non_young_regions;
1930 record_cset_region_info(hr, false);
1931 }
1932
1933 void
1934 G1CollectorPolicy::set_recorded_young_regions(size_t n_regions) {
1935 _recorded_young_regions = n_regions;
1936 }
1937
1938 void G1CollectorPolicy::set_recorded_young_bytes(size_t bytes) {
1939 #if PREDICTIONS_VERBOSE
1940 _recorded_young_bytes = bytes;
1941 #endif // PREDICTIONS_VERBOSE
1942 }
1943
1944 void G1CollectorPolicy::set_recorded_rs_lengths(size_t rs_lengths) {
1945 _recorded_rs_lengths = rs_lengths;
1946 }
1947
1948 void G1CollectorPolicy::set_predicted_bytes_to_copy(size_t bytes) {
1949 _predicted_bytes_to_copy = bytes;
1950 }
1951
1952 void
1953 G1CollectorPolicy::end_recording_regions() {
1954 // The _predicted_pause_time_ms field is referenced in code
1955 // not under PREDICTIONS_VERBOSE. Let's initialize it.
1956 _predicted_pause_time_ms = -1.0;
1957
1958 #if PREDICTIONS_VERBOSE
1959 _predicted_pending_cards = predict_pending_cards();
1960 _predicted_rs_lengths = _recorded_rs_lengths + predict_rs_length_diff();
1961 if (full_young_gcs())
1962 _predicted_cards_scanned += predict_young_card_num(_predicted_rs_lengths);
1963 else
1964 _predicted_cards_scanned +=
1965 predict_non_young_card_num(_predicted_rs_lengths);
1966 _recorded_region_num = _recorded_young_regions + _recorded_non_young_regions;
1967
1968 _predicted_rs_update_time_ms =
1969 predict_rs_update_time_ms(_g1->pending_card_num());
1970 _predicted_rs_scan_time_ms =
1971 predict_rs_scan_time_ms(_predicted_cards_scanned);
1972 _predicted_object_copy_time_ms =
1973 predict_object_copy_time_ms(_predicted_bytes_to_copy);
1974 _predicted_constant_other_time_ms =
1975 predict_constant_other_time_ms();
1976 _predicted_young_other_time_ms =
1977 predict_young_other_time_ms(_recorded_young_regions);
1978 _predicted_non_young_other_time_ms =
1979 predict_non_young_other_time_ms(_recorded_non_young_regions);
1980
1981 _predicted_pause_time_ms =
1982 _predicted_rs_update_time_ms +
1983 _predicted_rs_scan_time_ms +
1984 _predicted_object_copy_time_ms +
1985 _predicted_constant_other_time_ms +
1986 _predicted_young_other_time_ms +
1987 _predicted_non_young_other_time_ms;
1988 #endif // PREDICTIONS_VERBOSE
1989 }
1990
1991 void G1CollectorPolicy::check_if_region_is_too_expensive(double
1992 predicted_time_ms) {
1993 // I don't think we need to do this when in young GC mode since
1994 // marking will be initiated next time we hit the soft limit anyway...
1995 if (predicted_time_ms > _expensive_region_limit_ms) {
1996 ergo_verbose2(ErgoPartiallyYoungGCs,
1997 "request partially-young GCs end",
1998 ergo_format_reason("predicted region time higher than threshold")
1999 ergo_format_ms("predicted region time")
2000 ergo_format_ms("threshold"),
2001 predicted_time_ms, _expensive_region_limit_ms);
2002 // no point in doing another partial one
2003 _should_revert_to_full_young_gcs = true;
2004 }
2005 }
2006
2007 // </NEW PREDICTION>
2008
2009
2010 void G1CollectorPolicy::update_recent_gc_times(double end_time_sec,
2011 double elapsed_ms) {
2012 _recent_gc_times_ms->add(elapsed_ms);
2013 _recent_prev_end_times_for_all_gcs_sec->add(end_time_sec);
2014 _prev_collection_pause_end_ms = end_time_sec * 1000.0;
2015 }
2016
2017 double G1CollectorPolicy::recent_avg_time_for_pauses_ms() {
2018 if (_recent_pause_times_ms->num() == 0) {
2019 return (double) MaxGCPauseMillis;
2020 }
2021 return _recent_pause_times_ms->avg();
2022 }
2023
2024 double G1CollectorPolicy::recent_avg_time_for_rs_scan_ms() {
2025 if (_recent_rs_scan_times_ms->num() == 0) {
2026 return (double)MaxGCPauseMillis/3.0;
2027 }
2028 return _recent_rs_scan_times_ms->avg();
2029 }
2030
2031 int G1CollectorPolicy::number_of_recent_gcs() {
2032 assert(_recent_rs_scan_times_ms->num() ==
2033 _recent_pause_times_ms->num(), "Sequence out of sync");
2034 assert(_recent_pause_times_ms->num() ==
2035 _recent_CS_bytes_used_before->num(), "Sequence out of sync");
2036 assert(_recent_CS_bytes_used_before->num() ==
2037 _recent_CS_bytes_surviving->num(), "Sequence out of sync");
2038
2039 return _recent_pause_times_ms->num();
2040 }
2041
2042 double G1CollectorPolicy::recent_avg_survival_fraction() {
2043 return recent_avg_survival_fraction_work(_recent_CS_bytes_surviving,
2044 _recent_CS_bytes_used_before);
2045 }
2046
2047 double G1CollectorPolicy::last_survival_fraction() {
2048 return last_survival_fraction_work(_recent_CS_bytes_surviving,
2049 _recent_CS_bytes_used_before);
2050 }
2051
2052 double
2053 G1CollectorPolicy::recent_avg_survival_fraction_work(TruncatedSeq* surviving,
2054 TruncatedSeq* before) {
2055 assert(surviving->num() == before->num(), "Sequence out of sync");
2056 if (before->sum() > 0.0) {
2057 double recent_survival_rate = surviving->sum() / before->sum();
2058 // We exempt parallel collection from this check because Alloc Buffer
2059 // fragmentation can produce negative collections.
2060 // Further, we're now always doing parallel collection. But I'm still
2061 // leaving this here as a placeholder for a more precise assertion later.
2062 // (DLD, 10/05.)
2063 assert((true || G1CollectedHeap::use_parallel_gc_threads()) ||
2064 _g1->evacuation_failed() ||
2065 recent_survival_rate <= 1.0, "Or bad frac");
2066 return recent_survival_rate;
2067 } else {
2068 return 1.0; // Be conservative.
2069 }
2070 }
2071
2072 double
2073 G1CollectorPolicy::last_survival_fraction_work(TruncatedSeq* surviving,
2074 TruncatedSeq* before) {
2075 assert(surviving->num() == before->num(), "Sequence out of sync");
2076 if (surviving->num() > 0 && before->last() > 0.0) {
2077 double last_survival_rate = surviving->last() / before->last();
2078 // We exempt parallel collection from this check because Alloc Buffer
2079 // fragmentation can produce negative collections.
2080 // Further, we're now always doing parallel collection. But I'm still
2081 // leaving this here as a placeholder for a more precise assertion later.
2082 // (DLD, 10/05.)
2083 assert((true || G1CollectedHeap::use_parallel_gc_threads()) ||
2084 last_survival_rate <= 1.0, "Or bad frac");
2085 return last_survival_rate;
2086 } else {
2087 return 1.0;
2088 }
2089 }
2090
2091 static const int survival_min_obs = 5;
2092 static double survival_min_obs_limits[] = { 0.9, 0.7, 0.5, 0.3, 0.1 };
2093 static const double min_survival_rate = 0.1;
2094
2095 double
2096 G1CollectorPolicy::conservative_avg_survival_fraction_work(double avg,
2097 double latest) {
2098 double res = avg;
2099 if (number_of_recent_gcs() < survival_min_obs) {
2100 res = MAX2(res, survival_min_obs_limits[number_of_recent_gcs()]);
2101 }
2102 res = MAX2(res, latest);
2103 res = MAX2(res, min_survival_rate);
2104 // In the parallel case, LAB fragmentation can produce "negative
2105 // collections"; so can evac failure. Cap at 1.0
2106 res = MIN2(res, 1.0);
2107 return res;
2108 }
2109
2110 size_t G1CollectorPolicy::expansion_amount() {
2111 double recent_gc_overhead = recent_avg_pause_time_ratio() * 100.0;
2112 double threshold = _gc_overhead_perc;
2113 if (recent_gc_overhead > threshold) {
2114 // We will double the existing space, or take
2115 // G1ExpandByPercentOfAvailable % of the available expansion
2116 // space, whichever is smaller, bounded below by a minimum
2117 // expansion (unless that's all that's left.)
2118 const size_t min_expand_bytes = 1*M;
2119 size_t reserved_bytes = _g1->max_capacity();
2120 size_t committed_bytes = _g1->capacity();
2121 size_t uncommitted_bytes = reserved_bytes - committed_bytes;
2122 size_t expand_bytes;
2123 size_t expand_bytes_via_pct =
2124 uncommitted_bytes * G1ExpandByPercentOfAvailable / 100;
2125 expand_bytes = MIN2(expand_bytes_via_pct, committed_bytes);
2126 expand_bytes = MAX2(expand_bytes, min_expand_bytes);
2127 expand_bytes = MIN2(expand_bytes, uncommitted_bytes);
2128
2129 ergo_verbose5(ErgoHeapSizing,
2130 "attempt heap expansion",
2131 ergo_format_reason("recent GC overhead higher than "
2132 "threshold after GC")
2133 ergo_format_perc("recent GC overhead")
2134 ergo_format_perc("threshold")
2135 ergo_format_byte("uncommitted")
2136 ergo_format_byte_perc("calculated expansion amount"),
2137 recent_gc_overhead, threshold,
2138 uncommitted_bytes,
2139 expand_bytes_via_pct, (double) G1ExpandByPercentOfAvailable);
2140
2141 return expand_bytes;
2142 } else {
2143 return 0;
2144 }
2145 }
2146
2147 void G1CollectorPolicy::note_start_of_mark_thread() {
2148 _mark_thread_startup_sec = os::elapsedTime();
2149 }
2150
2151 class CountCSClosure: public HeapRegionClosure {
2152 G1CollectorPolicy* _g1_policy;
2153 public:
2154 CountCSClosure(G1CollectorPolicy* g1_policy) :
2155 _g1_policy(g1_policy) {}
2156 bool doHeapRegion(HeapRegion* r) {
2157 _g1_policy->_bytes_in_collection_set_before_gc += r->used();
2158 return false;
2159 }
2160 };
2161
2162 void G1CollectorPolicy::count_CS_bytes_used() {
2163 CountCSClosure cs_closure(this);
2164 _g1->collection_set_iterate(&cs_closure);
2165 }
2166
2167 void G1CollectorPolicy::print_summary (int level,
2168 const char* str,
2169 NumberSeq* seq) const {
2170 double sum = seq->sum();
2171 LineBuffer(level + 1).append_and_print_cr("%-24s = %8.2lf s (avg = %8.2lf ms)",
2172 str, sum / 1000.0, seq->avg());
2173 }
2174
2175 void G1CollectorPolicy::print_summary_sd (int level,
2176 const char* str,
2177 NumberSeq* seq) const {
2178 print_summary(level, str, seq);
2179 LineBuffer(level + 6).append_and_print_cr("(num = %5d, std dev = %8.2lf ms, max = %8.2lf ms)",
2180 seq->num(), seq->sd(), seq->maximum());
2181 }
2182
2183 void G1CollectorPolicy::check_other_times(int level,
2184 NumberSeq* other_times_ms,
2185 NumberSeq* calc_other_times_ms) const {
2186 bool should_print = false;
2187 LineBuffer buf(level + 2);
2188
2189 double max_sum = MAX2(fabs(other_times_ms->sum()),
2190 fabs(calc_other_times_ms->sum()));
2191 double min_sum = MIN2(fabs(other_times_ms->sum()),
2192 fabs(calc_other_times_ms->sum()));
2193 double sum_ratio = max_sum / min_sum;
2194 if (sum_ratio > 1.1) {
2195 should_print = true;
2196 buf.append_and_print_cr("## CALCULATED OTHER SUM DOESN'T MATCH RECORDED ###");
2197 }
2198
2199 double max_avg = MAX2(fabs(other_times_ms->avg()),
2200 fabs(calc_other_times_ms->avg()));
2201 double min_avg = MIN2(fabs(other_times_ms->avg()),
2202 fabs(calc_other_times_ms->avg()));
2203 double avg_ratio = max_avg / min_avg;
2204 if (avg_ratio > 1.1) {
2205 should_print = true;
2206 buf.append_and_print_cr("## CALCULATED OTHER AVG DOESN'T MATCH RECORDED ###");
2207 }
2208
2209 if (other_times_ms->sum() < -0.01) {
2210 buf.append_and_print_cr("## RECORDED OTHER SUM IS NEGATIVE ###");
2211 }
2212
2213 if (other_times_ms->avg() < -0.01) {
2214 buf.append_and_print_cr("## RECORDED OTHER AVG IS NEGATIVE ###");
2215 }
2216
2217 if (calc_other_times_ms->sum() < -0.01) {
2218 should_print = true;
2219 buf.append_and_print_cr("## CALCULATED OTHER SUM IS NEGATIVE ###");
2220 }
2221
2222 if (calc_other_times_ms->avg() < -0.01) {
2223 should_print = true;
2224 buf.append_and_print_cr("## CALCULATED OTHER AVG IS NEGATIVE ###");
2225 }
2226
2227 if (should_print)
2228 print_summary(level, "Other(Calc)", calc_other_times_ms);
2229 }
2230
2231 void G1CollectorPolicy::print_summary(PauseSummary* summary) const {
2232 bool parallel = G1CollectedHeap::use_parallel_gc_threads();
2233 MainBodySummary* body_summary = summary->main_body_summary();
2234 if (summary->get_total_seq()->num() > 0) {
2235 print_summary_sd(0, "Evacuation Pauses", summary->get_total_seq());
2236 if (body_summary != NULL) {
2237 print_summary(1, "SATB Drain", body_summary->get_satb_drain_seq());
2238 if (parallel) {
2239 print_summary(1, "Parallel Time", body_summary->get_parallel_seq());
2240 print_summary(2, "Update RS", body_summary->get_update_rs_seq());
2241 print_summary(2, "Ext Root Scanning",
2242 body_summary->get_ext_root_scan_seq());
2243 print_summary(2, "Mark Stack Scanning",
2244 body_summary->get_mark_stack_scan_seq());
2245 print_summary(2, "Scan RS", body_summary->get_scan_rs_seq());
2246 print_summary(2, "Object Copy", body_summary->get_obj_copy_seq());
2247 print_summary(2, "Termination", body_summary->get_termination_seq());
2248 print_summary(2, "Other", body_summary->get_parallel_other_seq());
2249 {
2250 NumberSeq* other_parts[] = {
2251 body_summary->get_update_rs_seq(),
2252 body_summary->get_ext_root_scan_seq(),
2253 body_summary->get_mark_stack_scan_seq(),
2254 body_summary->get_scan_rs_seq(),
2255 body_summary->get_obj_copy_seq(),
2256 body_summary->get_termination_seq()
2257 };
2258 NumberSeq calc_other_times_ms(body_summary->get_parallel_seq(),
2259 6, other_parts);
2260 check_other_times(2, body_summary->get_parallel_other_seq(),
2261 &calc_other_times_ms);
2262 }
2263 print_summary(1, "Mark Closure", body_summary->get_mark_closure_seq());
2264 print_summary(1, "Clear CT", body_summary->get_clear_ct_seq());
2265 } else {
2266 print_summary(1, "Update RS", body_summary->get_update_rs_seq());
2267 print_summary(1, "Ext Root Scanning",
2268 body_summary->get_ext_root_scan_seq());
2269 print_summary(1, "Mark Stack Scanning",
2270 body_summary->get_mark_stack_scan_seq());
2271 print_summary(1, "Scan RS", body_summary->get_scan_rs_seq());
2272 print_summary(1, "Object Copy", body_summary->get_obj_copy_seq());
2273 }
2274 }
2275 print_summary(1, "Other", summary->get_other_seq());
2276 {
2277 if (body_summary != NULL) {
2278 NumberSeq calc_other_times_ms;
2279 if (parallel) {
2280 // parallel
2281 NumberSeq* other_parts[] = {
2282 body_summary->get_satb_drain_seq(),
2283 body_summary->get_parallel_seq(),
2284 body_summary->get_clear_ct_seq()
2285 };
2286 calc_other_times_ms = NumberSeq(summary->get_total_seq(),
2287 3, other_parts);
2288 } else {
2289 // serial
2290 NumberSeq* other_parts[] = {
2291 body_summary->get_satb_drain_seq(),
2292 body_summary->get_update_rs_seq(),
2293 body_summary->get_ext_root_scan_seq(),
2294 body_summary->get_mark_stack_scan_seq(),
2295 body_summary->get_scan_rs_seq(),
2296 body_summary->get_obj_copy_seq()
2297 };
2298 calc_other_times_ms = NumberSeq(summary->get_total_seq(),
2299 6, other_parts);
2300 }
2301 check_other_times(1, summary->get_other_seq(), &calc_other_times_ms);
2302 }
2303 }
2304 } else {
2305 LineBuffer(1).append_and_print_cr("none");
2306 }
2307 LineBuffer(0).append_and_print_cr("");
2308 }
2309
2310 void G1CollectorPolicy::print_tracing_info() const {
2311 if (TraceGen0Time) {
2312 gclog_or_tty->print_cr("ALL PAUSES");
2313 print_summary_sd(0, "Total", _all_pause_times_ms);
2314 gclog_or_tty->print_cr("");
2315 gclog_or_tty->print_cr("");
2316 gclog_or_tty->print_cr(" Full Young GC Pauses: %8d", _full_young_pause_num);
2317 gclog_or_tty->print_cr(" Partial Young GC Pauses: %8d", _partial_young_pause_num);
2318 gclog_or_tty->print_cr("");
2319
2320 gclog_or_tty->print_cr("EVACUATION PAUSES");
2321 print_summary(_summary);
2322
2323 gclog_or_tty->print_cr("MISC");
2324 print_summary_sd(0, "Stop World", _all_stop_world_times_ms);
2325 print_summary_sd(0, "Yields", _all_yield_times_ms);
2326 for (int i = 0; i < _aux_num; ++i) {
2327 if (_all_aux_times_ms[i].num() > 0) {
2328 char buffer[96];
2329 sprintf(buffer, "Aux%d", i);
2330 print_summary_sd(0, buffer, &_all_aux_times_ms[i]);
2331 }
2332 }
2333
2334 size_t all_region_num = _region_num_young + _region_num_tenured;
2335 gclog_or_tty->print_cr(" New Regions %8d, Young %8d (%6.2lf%%), "
2336 "Tenured %8d (%6.2lf%%)",
2337 all_region_num,
2338 _region_num_young,
2339 (double) _region_num_young / (double) all_region_num * 100.0,
2340 _region_num_tenured,
2341 (double) _region_num_tenured / (double) all_region_num * 100.0);
2342 }
2343 if (TraceGen1Time) {
2344 if (_all_full_gc_times_ms->num() > 0) {
2345 gclog_or_tty->print("\n%4d full_gcs: total time = %8.2f s",
2346 _all_full_gc_times_ms->num(),
2347 _all_full_gc_times_ms->sum() / 1000.0);
2348 gclog_or_tty->print_cr(" (avg = %8.2fms).", _all_full_gc_times_ms->avg());
2349 gclog_or_tty->print_cr(" [std. dev = %8.2f ms, max = %8.2f ms]",
2350 _all_full_gc_times_ms->sd(),
2351 _all_full_gc_times_ms->maximum());
2352 }
2353 }
2354 }
2355
2356 void G1CollectorPolicy::print_yg_surv_rate_info() const {
2357 #ifndef PRODUCT
2358 _short_lived_surv_rate_group->print_surv_rate_summary();
2359 // add this call for any other surv rate groups
2360 #endif // PRODUCT
2361 }
2362
2363 void G1CollectorPolicy::update_region_num(bool young) {
2364 if (young) {
2365 ++_region_num_young;
2366 } else {
2367 ++_region_num_tenured;
2368 }
2369 }
2370
2371 #ifndef PRODUCT
2372 // for debugging, bit of a hack...
2373 static char*
2374 region_num_to_mbs(int length) {
2375 static char buffer[64];
2376 double bytes = (double) (length * HeapRegion::GrainBytes);
2377 double mbs = bytes / (double) (1024 * 1024);
2378 sprintf(buffer, "%7.2lfMB", mbs);
2379 return buffer;
2380 }
2381 #endif // PRODUCT
2382
2383 size_t G1CollectorPolicy::max_regions(int purpose) {
2384 switch (purpose) {
2385 case GCAllocForSurvived:
2386 return _max_survivor_regions;
2387 case GCAllocForTenured:
2388 return REGIONS_UNLIMITED;
2389 default:
2390 ShouldNotReachHere();
2391 return REGIONS_UNLIMITED;
2392 };
2393 }
2394
2395 void G1CollectorPolicy::update_max_gc_locker_expansion() {
2396 size_t expansion_region_num = 0;
2397 if (GCLockerEdenExpansionPercent > 0) {
2398 double perc = (double) GCLockerEdenExpansionPercent / 100.0;
2399 double expansion_region_num_d = perc * (double) _young_list_target_length;
2400 // We use ceiling so that if expansion_region_num_d is > 0.0 (but
2401 // less than 1.0) we'll get 1.
2402 expansion_region_num = (size_t) ceil(expansion_region_num_d);
2403 } else {
2404 assert(expansion_region_num == 0, "sanity");
2405 }
2406 _young_list_max_length = _young_list_target_length + expansion_region_num;
2407 assert(_young_list_target_length <= _young_list_max_length, "post-condition");
2408 }
2409
2410 // Calculates survivor space parameters.
2411 void G1CollectorPolicy::update_survivors_policy() {
2412 double max_survivor_regions_d =
2413 (double) _young_list_target_length / (double) SurvivorRatio;
2414 // We use ceiling so that if max_survivor_regions_d is > 0.0 (but
2415 // smaller than 1.0) we'll get 1.
2416 _max_survivor_regions = (size_t) ceil(max_survivor_regions_d);
2417
2418 _tenuring_threshold = _survivors_age_table.compute_tenuring_threshold(
2419 HeapRegion::GrainWords * _max_survivor_regions);
2420 }
2421
2422 #ifndef PRODUCT
2423 class HRSortIndexIsOKClosure: public HeapRegionClosure {
2424 CollectionSetChooser* _chooser;
2425 public:
2426 HRSortIndexIsOKClosure(CollectionSetChooser* chooser) :
2427 _chooser(chooser) {}
2428
2429 bool doHeapRegion(HeapRegion* r) {
2430 if (!r->continuesHumongous()) {
2431 assert(_chooser->regionProperlyOrdered(r), "Ought to be.");
2432 }
2433 return false;
2434 }
2435 };
2436
2437 bool G1CollectorPolicy_BestRegionsFirst::assertMarkedBytesDataOK() {
2438 HRSortIndexIsOKClosure cl(_collectionSetChooser);
2439 _g1->heap_region_iterate(&cl);
2440 return true;
2441 }
2442 #endif
2443
2444 bool G1CollectorPolicy::force_initial_mark_if_outside_cycle(
2445 GCCause::Cause gc_cause) {
2446 bool during_cycle = _g1->concurrent_mark()->cmThread()->during_cycle();
2447 if (!during_cycle) {
2448 ergo_verbose1(ErgoConcCycles,
2449 "request concurrent cycle initiation",
2450 ergo_format_reason("requested by GC cause")
2451 ergo_format_str("GC cause"),
2452 GCCause::to_string(gc_cause));
2453 set_initiate_conc_mark_if_possible();
2454 return true;
2455 } else {
2456 ergo_verbose1(ErgoConcCycles,
2457 "do not request concurrent cycle initiation",
2458 ergo_format_reason("concurrent cycle already in progress")
2459 ergo_format_str("GC cause"),
2460 GCCause::to_string(gc_cause));
2461 return false;
2462 }
2463 }
2464
2465 void
2466 G1CollectorPolicy::decide_on_conc_mark_initiation() {
2467 // We are about to decide on whether this pause will be an
2468 // initial-mark pause.
2469
2470 // First, during_initial_mark_pause() should not be already set. We
2471 // will set it here if we have to. However, it should be cleared by
2472 // the end of the pause (it's only set for the duration of an
2473 // initial-mark pause).
2474 assert(!during_initial_mark_pause(), "pre-condition");
2475
2476 if (initiate_conc_mark_if_possible()) {
2477 // We had noticed on a previous pause that the heap occupancy has
2478 // gone over the initiating threshold and we should start a
2479 // concurrent marking cycle. So we might initiate one.
2480
2481 bool during_cycle = _g1->concurrent_mark()->cmThread()->during_cycle();
2482 if (!during_cycle) {
2483 // The concurrent marking thread is not "during a cycle", i.e.,
2484 // it has completed the last one. So we can go ahead and
2485 // initiate a new cycle.
2486
2487 set_during_initial_mark_pause();
2488
2489 // And we can now clear initiate_conc_mark_if_possible() as
2490 // we've already acted on it.
2491 clear_initiate_conc_mark_if_possible();
2492
2493 ergo_verbose0(ErgoConcCycles,
2494 "initiate concurrent cycle",
2495 ergo_format_reason("concurrent cycle initiation requested"));
2496 } else {
2497 // The concurrent marking thread is still finishing up the
2498 // previous cycle. If we start one right now the two cycles
2499 // overlap. In particular, the concurrent marking thread might
2500 // be in the process of clearing the next marking bitmap (which
2501 // we will use for the next cycle if we start one). Starting a
2502 // cycle now will be bad given that parts of the marking
2503 // information might get cleared by the marking thread. And we
2504 // cannot wait for the marking thread to finish the cycle as it
2505 // periodically yields while clearing the next marking bitmap
2506 // and, if it's in a yield point, it's waiting for us to
2507 // finish. So, at this point we will not start a cycle and we'll
2508 // let the concurrent marking thread complete the last one.
2509 ergo_verbose0(ErgoConcCycles,
2510 "do not initiate concurrent cycle",
2511 ergo_format_reason("concurrent cycle already in progress"));
2512 }
2513 }
2514 }
2515
2516 void
2517 G1CollectorPolicy_BestRegionsFirst::
2518 record_collection_pause_start(double start_time_sec, size_t start_used) {
2519 G1CollectorPolicy::record_collection_pause_start(start_time_sec, start_used);
2520 }
2521
2522 class KnownGarbageClosure: public HeapRegionClosure {
2523 CollectionSetChooser* _hrSorted;
2524
2525 public:
2526 KnownGarbageClosure(CollectionSetChooser* hrSorted) :
2527 _hrSorted(hrSorted)
2528 {}
2529
2530 bool doHeapRegion(HeapRegion* r) {
2531 // We only include humongous regions in collection
2532 // sets when concurrent mark shows that their contained object is
2533 // unreachable.
2534
2535 // Do we have any marking information for this region?
2536 if (r->is_marked()) {
2537 // We don't include humongous regions in collection
2538 // sets because we collect them immediately at the end of a marking
2539 // cycle. We also don't include young regions because we *must*
2540 // include them in the next collection pause.
2541 if (!r->isHumongous() && !r->is_young()) {
2542 _hrSorted->addMarkedHeapRegion(r);
2543 }
2544 }
2545 return false;
2546 }
2547 };
2548
2549 class ParKnownGarbageHRClosure: public HeapRegionClosure {
2550 CollectionSetChooser* _hrSorted;
2551 jint _marked_regions_added;
2552 jint _chunk_size;
2553 jint _cur_chunk_idx;
2554 jint _cur_chunk_end; // Cur chunk [_cur_chunk_idx, _cur_chunk_end)
2555 int _worker;
2556 int _invokes;
2557
2558 void get_new_chunk() {
2559 _cur_chunk_idx = _hrSorted->getParMarkedHeapRegionChunk(_chunk_size);
2560 _cur_chunk_end = _cur_chunk_idx + _chunk_size;
2561 }
2562 void add_region(HeapRegion* r) {
2563 if (_cur_chunk_idx == _cur_chunk_end) {
2564 get_new_chunk();
2565 }
2566 assert(_cur_chunk_idx < _cur_chunk_end, "postcondition");
2567 _hrSorted->setMarkedHeapRegion(_cur_chunk_idx, r);
2568 _marked_regions_added++;
2569 _cur_chunk_idx++;
2570 }
2571
2572 public:
2573 ParKnownGarbageHRClosure(CollectionSetChooser* hrSorted,
2574 jint chunk_size,
2575 int worker) :
2576 _hrSorted(hrSorted), _chunk_size(chunk_size), _worker(worker),
2577 _marked_regions_added(0), _cur_chunk_idx(0), _cur_chunk_end(0),
2578 _invokes(0)
2579 {}
2580
2581 bool doHeapRegion(HeapRegion* r) {
2582 // We only include humongous regions in collection
2583 // sets when concurrent mark shows that their contained object is
2584 // unreachable.
2585 _invokes++;
2586
2587 // Do we have any marking information for this region?
2588 if (r->is_marked()) {
2589 // We don't include humongous regions in collection
2590 // sets because we collect them immediately at the end of a marking
2591 // cycle.
2592 // We also do not include young regions in collection sets
2593 if (!r->isHumongous() && !r->is_young()) {
2594 add_region(r);
2595 }
2596 }
2597 return false;
2598 }
2599 jint marked_regions_added() { return _marked_regions_added; }
2600 int invokes() { return _invokes; }
2601 };
2602
2603 class ParKnownGarbageTask: public AbstractGangTask {
2604 CollectionSetChooser* _hrSorted;
2605 jint _chunk_size;
2606 G1CollectedHeap* _g1;
2607 public:
2608 ParKnownGarbageTask(CollectionSetChooser* hrSorted, jint chunk_size) :
2609 AbstractGangTask("ParKnownGarbageTask"),
2610 _hrSorted(hrSorted), _chunk_size(chunk_size),
2611 _g1(G1CollectedHeap::heap())
2612 {}
2613
2614 void work(int i) {
2615 ParKnownGarbageHRClosure parKnownGarbageCl(_hrSorted, _chunk_size, i);
2616 // Back to zero for the claim value.
2617 _g1->heap_region_par_iterate_chunked(&parKnownGarbageCl, i,
2618 HeapRegion::InitialClaimValue);
2619 jint regions_added = parKnownGarbageCl.marked_regions_added();
2620 _hrSorted->incNumMarkedHeapRegions(regions_added);
2621 if (G1PrintParCleanupStats) {
2622 gclog_or_tty->print_cr(" Thread %d called %d times, added %d regions to list.",
2623 i, parKnownGarbageCl.invokes(), regions_added);
2624 }
2625 }
2626 };
2627
2628 void
2629 G1CollectorPolicy_BestRegionsFirst::
2630 record_concurrent_mark_cleanup_end(size_t freed_bytes,
2631 size_t max_live_bytes) {
2632 double start;
2633 if (G1PrintParCleanupStats) start = os::elapsedTime();
2634 record_concurrent_mark_cleanup_end_work1(freed_bytes, max_live_bytes);
2635
2636 _collectionSetChooser->clearMarkedHeapRegions();
2637 double clear_marked_end;
2638 if (G1PrintParCleanupStats) {
2639 clear_marked_end = os::elapsedTime();
2640 gclog_or_tty->print_cr(" clear marked regions + work1: %8.3f ms.",
2641 (clear_marked_end - start)*1000.0);
2642 }
2643 if (G1CollectedHeap::use_parallel_gc_threads()) {
2644 const size_t OverpartitionFactor = 4;
2645 const size_t MinWorkUnit = 8;
2646 const size_t WorkUnit =
2647 MAX2(_g1->n_regions() / (ParallelGCThreads * OverpartitionFactor),
2648 MinWorkUnit);
2649 _collectionSetChooser->prepareForAddMarkedHeapRegionsPar(_g1->n_regions(),
2650 WorkUnit);
2651 ParKnownGarbageTask parKnownGarbageTask(_collectionSetChooser,
2652 (int) WorkUnit);
2653 _g1->workers()->run_task(&parKnownGarbageTask);
2654
2655 assert(_g1->check_heap_region_claim_values(HeapRegion::InitialClaimValue),
2656 "sanity check");
2657 } else {
2658 KnownGarbageClosure knownGarbagecl(_collectionSetChooser);
2659 _g1->heap_region_iterate(&knownGarbagecl);
2660 }
2661 double known_garbage_end;
2662 if (G1PrintParCleanupStats) {
2663 known_garbage_end = os::elapsedTime();
2664 gclog_or_tty->print_cr(" compute known garbage: %8.3f ms.",
2665 (known_garbage_end - clear_marked_end)*1000.0);
2666 }
2667 _collectionSetChooser->sortMarkedHeapRegions();
2668 double sort_end;
2669 if (G1PrintParCleanupStats) {
2670 sort_end = os::elapsedTime();
2671 gclog_or_tty->print_cr(" sorting: %8.3f ms.",
2672 (sort_end - known_garbage_end)*1000.0);
2673 }
2674
2675 record_concurrent_mark_cleanup_end_work2();
2676 double work2_end;
2677 if (G1PrintParCleanupStats) {
2678 work2_end = os::elapsedTime();
2679 gclog_or_tty->print_cr(" work2: %8.3f ms.",
2680 (work2_end - sort_end)*1000.0);
2681 }
2682 }
2683
2684 // Add the heap region at the head of the non-incremental collection set
2685 void G1CollectorPolicy::
2686 add_to_collection_set(HeapRegion* hr) {
2687 assert(_inc_cset_build_state == Active, "Precondition");
2688 assert(!hr->is_young(), "non-incremental add of young region");
2689
2690 if (_g1->mark_in_progress())
2691 _g1->concurrent_mark()->registerCSetRegion(hr);
2692
2693 assert(!hr->in_collection_set(), "should not already be in the CSet");
2694 hr->set_in_collection_set(true);
2695 hr->set_next_in_collection_set(_collection_set);
2696 _collection_set = hr;
2697 _collection_set_size++;
2698 _collection_set_bytes_used_before += hr->used();
2699 _g1->register_region_with_in_cset_fast_test(hr);
2700 }
2701
2702 // Initialize the per-collection-set information
2703 void G1CollectorPolicy::start_incremental_cset_building() {
2704 assert(_inc_cset_build_state == Inactive, "Precondition");
2705
2706 _inc_cset_head = NULL;
2707 _inc_cset_tail = NULL;
2708 _inc_cset_size = 0;
2709 _inc_cset_bytes_used_before = 0;
2710
2711 _inc_cset_young_index = 0;
2712
2713 _inc_cset_max_finger = 0;
2714 _inc_cset_recorded_young_bytes = 0;
2715 _inc_cset_recorded_rs_lengths = 0;
2716 _inc_cset_predicted_elapsed_time_ms = 0;
2717 _inc_cset_predicted_bytes_to_copy = 0;
2718 _inc_cset_build_state = Active;
2719 }
2720
2721 void G1CollectorPolicy::add_to_incremental_cset_info(HeapRegion* hr, size_t rs_length) {
2722 // This routine is used when:
2723 // * adding survivor regions to the incremental cset at the end of an
2724 // evacuation pause,
2725 // * adding the current allocation region to the incremental cset
2726 // when it is retired, and
2727 // * updating existing policy information for a region in the
2728 // incremental cset via young list RSet sampling.
2729 // Therefore this routine may be called at a safepoint by the
2730 // VM thread, or in-between safepoints by mutator threads (when
2731 // retiring the current allocation region) or a concurrent
2732 // refine thread (RSet sampling).
2733
2734 double region_elapsed_time_ms = predict_region_elapsed_time_ms(hr, true);
2735 size_t used_bytes = hr->used();
2736
2737 _inc_cset_recorded_rs_lengths += rs_length;
2738 _inc_cset_predicted_elapsed_time_ms += region_elapsed_time_ms;
2739
2740 _inc_cset_bytes_used_before += used_bytes;
2741
2742 // Cache the values we have added to the aggregated informtion
2743 // in the heap region in case we have to remove this region from
2744 // the incremental collection set, or it is updated by the
2745 // rset sampling code
2746 hr->set_recorded_rs_length(rs_length);
2747 hr->set_predicted_elapsed_time_ms(region_elapsed_time_ms);
2748
2749 #if PREDICTIONS_VERBOSE
2750 size_t bytes_to_copy = predict_bytes_to_copy(hr);
2751 _inc_cset_predicted_bytes_to_copy += bytes_to_copy;
2752
2753 // Record the number of bytes used in this region
2754 _inc_cset_recorded_young_bytes += used_bytes;
2755
2756 // Cache the values we have added to the aggregated informtion
2757 // in the heap region in case we have to remove this region from
2758 // the incremental collection set, or it is updated by the
2759 // rset sampling code
2760 hr->set_predicted_bytes_to_copy(bytes_to_copy);
2761 #endif // PREDICTIONS_VERBOSE
2762 }
2763
2764 void G1CollectorPolicy::remove_from_incremental_cset_info(HeapRegion* hr) {
2765 // This routine is currently only called as part of the updating of
2766 // existing policy information for regions in the incremental cset that
2767 // is performed by the concurrent refine thread(s) as part of young list
2768 // RSet sampling. Therefore we should not be at a safepoint.
2769
2770 assert(!SafepointSynchronize::is_at_safepoint(), "should not be at safepoint");
2771 assert(hr->is_young(), "it should be");
2772
2773 size_t used_bytes = hr->used();
2774 size_t old_rs_length = hr->recorded_rs_length();
2775 double old_elapsed_time_ms = hr->predicted_elapsed_time_ms();
2776
2777 // Subtract the old recorded/predicted policy information for
2778 // the given heap region from the collection set info.
2779 _inc_cset_recorded_rs_lengths -= old_rs_length;
2780 _inc_cset_predicted_elapsed_time_ms -= old_elapsed_time_ms;
2781
2782 _inc_cset_bytes_used_before -= used_bytes;
2783
2784 // Clear the values cached in the heap region
2785 hr->set_recorded_rs_length(0);
2786 hr->set_predicted_elapsed_time_ms(0);
2787
2788 #if PREDICTIONS_VERBOSE
2789 size_t old_predicted_bytes_to_copy = hr->predicted_bytes_to_copy();
2790 _inc_cset_predicted_bytes_to_copy -= old_predicted_bytes_to_copy;
2791
2792 // Subtract the number of bytes used in this region
2793 _inc_cset_recorded_young_bytes -= used_bytes;
2794
2795 // Clear the values cached in the heap region
2796 hr->set_predicted_bytes_to_copy(0);
2797 #endif // PREDICTIONS_VERBOSE
2798 }
2799
2800 void G1CollectorPolicy::update_incremental_cset_info(HeapRegion* hr, size_t new_rs_length) {
2801 // Update the collection set information that is dependent on the new RS length
2802 assert(hr->is_young(), "Precondition");
2803
2804 remove_from_incremental_cset_info(hr);
2805 add_to_incremental_cset_info(hr, new_rs_length);
2806 }
2807
2808 void G1CollectorPolicy::add_region_to_incremental_cset_common(HeapRegion* hr) {
2809 assert( hr->is_young(), "invariant");
2810 assert( hr->young_index_in_cset() == -1, "invariant" );
2811 assert(_inc_cset_build_state == Active, "Precondition");
2812
2813 // We need to clear and set the cached recorded/cached collection set
2814 // information in the heap region here (before the region gets added
2815 // to the collection set). An individual heap region's cached values
2816 // are calculated, aggregated with the policy collection set info,
2817 // and cached in the heap region here (initially) and (subsequently)
2818 // by the Young List sampling code.
2819
2820 size_t rs_length = hr->rem_set()->occupied();
2821 add_to_incremental_cset_info(hr, rs_length);
2822
2823 HeapWord* hr_end = hr->end();
2824 _inc_cset_max_finger = MAX2(_inc_cset_max_finger, hr_end);
2825
2826 assert(!hr->in_collection_set(), "invariant");
2827 hr->set_in_collection_set(true);
2828 assert( hr->next_in_collection_set() == NULL, "invariant");
2829
2830 _inc_cset_size++;
2831 _g1->register_region_with_in_cset_fast_test(hr);
2832
2833 hr->set_young_index_in_cset((int) _inc_cset_young_index);
2834 ++_inc_cset_young_index;
2835 }
2836
2837 // Add the region at the RHS of the incremental cset
2838 void G1CollectorPolicy::add_region_to_incremental_cset_rhs(HeapRegion* hr) {
2839 // We should only ever be appending survivors at the end of a pause
2840 assert( hr->is_survivor(), "Logic");
2841
2842 // Do the 'common' stuff
2843 add_region_to_incremental_cset_common(hr);
2844
2845 // Now add the region at the right hand side
2846 if (_inc_cset_tail == NULL) {
2847 assert(_inc_cset_head == NULL, "invariant");
2848 _inc_cset_head = hr;
2849 } else {
2850 _inc_cset_tail->set_next_in_collection_set(hr);
2851 }
2852 _inc_cset_tail = hr;
2853 }
2854
2855 // Add the region to the LHS of the incremental cset
2856 void G1CollectorPolicy::add_region_to_incremental_cset_lhs(HeapRegion* hr) {
2857 // Survivors should be added to the RHS at the end of a pause
2858 assert(!hr->is_survivor(), "Logic");
2859
2860 // Do the 'common' stuff
2861 add_region_to_incremental_cset_common(hr);
2862
2863 // Add the region at the left hand side
2864 hr->set_next_in_collection_set(_inc_cset_head);
2865 if (_inc_cset_head == NULL) {
2866 assert(_inc_cset_tail == NULL, "Invariant");
2867 _inc_cset_tail = hr;
2868 }
2869 _inc_cset_head = hr;
2870 }
2871
2872 #ifndef PRODUCT
2873 void G1CollectorPolicy::print_collection_set(HeapRegion* list_head, outputStream* st) {
2874 assert(list_head == inc_cset_head() || list_head == collection_set(), "must be");
2875
2876 st->print_cr("\nCollection_set:");
2877 HeapRegion* csr = list_head;
2878 while (csr != NULL) {
2879 HeapRegion* next = csr->next_in_collection_set();
2880 assert(csr->in_collection_set(), "bad CS");
2881 st->print_cr(" [%08x-%08x], t: %08x, P: %08x, N: %08x, C: %08x, "
2882 "age: %4d, y: %d, surv: %d",
2883 csr->bottom(), csr->end(),
2884 csr->top(),
2885 csr->prev_top_at_mark_start(),
2886 csr->next_top_at_mark_start(),
2887 csr->top_at_conc_mark_count(),
2888 csr->age_in_surv_rate_group_cond(),
2889 csr->is_young(),
2890 csr->is_survivor());
2891 csr = next;
2892 }
2893 }
2894 #endif // !PRODUCT
2895
2896 void
2897 G1CollectorPolicy_BestRegionsFirst::choose_collection_set(
2898 double target_pause_time_ms) {
2899 // Set this here - in case we're not doing young collections.
2900 double non_young_start_time_sec = os::elapsedTime();
2901
2902 YoungList* young_list = _g1->young_list();
2903
2904 start_recording_regions();
2905
2906 guarantee(target_pause_time_ms > 0.0,
2907 err_msg("target_pause_time_ms = %1.6lf should be positive",
2908 target_pause_time_ms));
2909 guarantee(_collection_set == NULL, "Precondition");
2910
2911 double base_time_ms = predict_base_elapsed_time_ms(_pending_cards);
2912 double predicted_pause_time_ms = base_time_ms;
2913
2914 double time_remaining_ms = target_pause_time_ms - base_time_ms;
2915
2916 ergo_verbose3(ErgoCSetConstruction | ErgoHigh,
2917 "start choosing CSet",
2918 ergo_format_ms("predicted base time")
2919 ergo_format_ms("remaining time")
2920 ergo_format_ms("target pause time"),
2921 base_time_ms, time_remaining_ms, target_pause_time_ms);
2922
2923 // the 10% and 50% values are arbitrary...
2924 double threshold = 0.10 * target_pause_time_ms;
2925 if (time_remaining_ms < threshold) {
2926 double prev_time_remaining_ms = time_remaining_ms;
2927 time_remaining_ms = 0.50 * target_pause_time_ms;
2928 _within_target = false;
2929 ergo_verbose3(ErgoCSetConstruction,
2930 "adjust remaining time",
2931 ergo_format_reason("remaining time lower than threshold")
2932 ergo_format_ms("remaining time")
2933 ergo_format_ms("threshold")
2934 ergo_format_ms("adjusted remaining time"),
2935 prev_time_remaining_ms, threshold, time_remaining_ms);
2936 } else {
2937 _within_target = true;
2938 }
2939
2940 size_t expansion_bytes = _g1->expansion_regions() * HeapRegion::GrainBytes;
2941
2942 HeapRegion* hr;
2943 double young_start_time_sec = os::elapsedTime();
2944
2945 _collection_set_bytes_used_before = 0;
2946 _collection_set_size = 0;
2947 _young_cset_length = 0;
2948 _last_young_gc_full = full_young_gcs() ? true : false;
2949
2950 if (_last_young_gc_full) {
2951 ++_full_young_pause_num;
2952 } else {
2953 ++_partial_young_pause_num;
2954 }
2955
2956 // The young list is laid with the survivor regions from the previous
2957 // pause are appended to the RHS of the young list, i.e.
2958 // [Newly Young Regions ++ Survivors from last pause].
2959
2960 size_t survivor_region_num = young_list->survivor_length();
2961 size_t eden_region_num = young_list->length() - survivor_region_num;
2962 size_t old_region_num = 0;
2963 hr = young_list->first_survivor_region();
2964 while (hr != NULL) {
2965 assert(hr->is_survivor(), "badly formed young list");
2966 hr->set_young();
2967 hr = hr->get_next_young_region();
2968 }
2969
2970 // Clear the fields that point to the survivor list - they are all young now.
2971 young_list->clear_survivors();
2972
2973 if (_g1->mark_in_progress())
2974 _g1->concurrent_mark()->register_collection_set_finger(_inc_cset_max_finger);
2975
2976 _young_cset_length = _inc_cset_young_index;
2977 _collection_set = _inc_cset_head;
2978 _collection_set_size = _inc_cset_size;
2979 _collection_set_bytes_used_before = _inc_cset_bytes_used_before;
2980 time_remaining_ms -= _inc_cset_predicted_elapsed_time_ms;
2981 predicted_pause_time_ms += _inc_cset_predicted_elapsed_time_ms;
2982
2983 ergo_verbose3(ErgoCSetConstruction | ErgoHigh,
2984 "add young regions to CSet",
2985 ergo_format_region("eden")
2986 ergo_format_region("survivors")
2987 ergo_format_ms("predicted young region time"),
2988 eden_region_num, survivor_region_num,
2989 _inc_cset_predicted_elapsed_time_ms);
2990
2991 // The number of recorded young regions is the incremental
2992 // collection set's current size
2993 set_recorded_young_regions(_inc_cset_size);
2994 set_recorded_rs_lengths(_inc_cset_recorded_rs_lengths);
2995 set_recorded_young_bytes(_inc_cset_recorded_young_bytes);
2996 #if PREDICTIONS_VERBOSE
2997 set_predicted_bytes_to_copy(_inc_cset_predicted_bytes_to_copy);
2998 #endif // PREDICTIONS_VERBOSE
2999
3000 assert(_inc_cset_size == young_list->length(), "Invariant");
3001
3002 double young_end_time_sec = os::elapsedTime();
3003 _recorded_young_cset_choice_time_ms =
3004 (young_end_time_sec - young_start_time_sec) * 1000.0;
3005
3006 // We are doing young collections so reset this.
3007 non_young_start_time_sec = young_end_time_sec;
3008
3009 if (!full_young_gcs()) {
3010 bool should_continue = true;
3011 NumberSeq seq;
3012 double avg_prediction = 100000000000000000.0; // something very large
3013
3014 size_t prev_collection_set_size = _collection_set_size;
3015 double prev_predicted_pause_time_ms = predicted_pause_time_ms;
3016 do {
3017 hr = _collectionSetChooser->getNextMarkedRegion(time_remaining_ms,
3018 avg_prediction);
3019 if (hr != NULL) {
3020 double predicted_time_ms = predict_region_elapsed_time_ms(hr, false);
3021 time_remaining_ms -= predicted_time_ms;
3022 predicted_pause_time_ms += predicted_time_ms;
3023 add_to_collection_set(hr);
3024 record_non_young_cset_region(hr);
3025 seq.add(predicted_time_ms);
3026 avg_prediction = seq.avg() + seq.sd();
3027 }
3028
3029 should_continue = true;
3030 if (hr == NULL) {
3031 // No need for an ergo verbose message here,
3032 // getNextMarkRegion() does this when it returns NULL.
3033 should_continue = false;
3034 } else {
3035 if (adaptive_young_list_length()) {
3036 if (time_remaining_ms < 0.0) {
3037 ergo_verbose1(ErgoCSetConstruction,
3038 "stop adding old regions to CSet",
3039 ergo_format_reason("remaining time is lower than 0")
3040 ergo_format_ms("remaining time"),
3041 time_remaining_ms);
3042 should_continue = false;
3043 }
3044 } else {
3045 if (_collection_set_size >= _young_list_fixed_length) {
3046 ergo_verbose2(ErgoCSetConstruction,
3047 "stop adding old regions to CSet",
3048 ergo_format_reason("CSet length reached target")
3049 ergo_format_region("CSet")
3050 ergo_format_region("young target"),
3051 _collection_set_size, _young_list_fixed_length);
3052 should_continue = false;
3053 }
3054 }
3055 }
3056 } while (should_continue);
3057
3058 if (!adaptive_young_list_length() &&
3059 _collection_set_size < _young_list_fixed_length) {
3060 ergo_verbose2(ErgoCSetConstruction,
3061 "request partially-young GCs end",
3062 ergo_format_reason("CSet length lower than target")
3063 ergo_format_region("CSet")
3064 ergo_format_region("young target"),
3065 _collection_set_size, _young_list_fixed_length);
3066 _should_revert_to_full_young_gcs = true;
3067 }
3068
3069 old_region_num = _collection_set_size - prev_collection_set_size;
3070
3071 ergo_verbose2(ErgoCSetConstruction | ErgoHigh,
3072 "add old regions to CSet",
3073 ergo_format_region("old")
3074 ergo_format_ms("predicted old region time"),
3075 old_region_num,
3076 predicted_pause_time_ms - prev_predicted_pause_time_ms);
3077 }
3078
3079 stop_incremental_cset_building();
3080
3081 count_CS_bytes_used();
3082
3083 end_recording_regions();
3084
3085 ergo_verbose5(ErgoCSetConstruction,
3086 "finish choosing CSet",
3087 ergo_format_region("eden")
3088 ergo_format_region("survivors")
3089 ergo_format_region("old")
3090 ergo_format_ms("predicted pause time")
3091 ergo_format_ms("target pause time"),
3092 eden_region_num, survivor_region_num, old_region_num,
3093 predicted_pause_time_ms, target_pause_time_ms);
3094
3095 double non_young_end_time_sec = os::elapsedTime();
3096 _recorded_non_young_cset_choice_time_ms =
3097 (non_young_end_time_sec - non_young_start_time_sec) * 1000.0;
3098 }
3099
3100 void G1CollectorPolicy_BestRegionsFirst::record_full_collection_end() {
3101 G1CollectorPolicy::record_full_collection_end();
3102 _collectionSetChooser->updateAfterFullCollection();
3103 }
3104
3105 void G1CollectorPolicy_BestRegionsFirst::
3106 record_collection_pause_end() {
3107 G1CollectorPolicy::record_collection_pause_end();
3108 assert(assertMarkedBytesDataOK(), "Marked regions not OK at pause end.");
3109 }