1 /*
2 * Copyright (c) 1999, 2017, Oracle and/or its affiliates. All rights reserved.
3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
4 *
5 * This code is free software; you can redistribute it and/or modify it
6 * under the terms of the GNU General Public License version 2 only, as
7 * published by the Free Software Foundation.
8 *
9 * This code is distributed in the hope that it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
12 * version 2 for more details (a copy is included in the LICENSE file that
13 * accompanied this code).
14 *
15 * You should have received a copy of the GNU General Public License version
16 * 2 along with this work; if not, write to the Free Software Foundation,
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
18 *
19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
20 * or visit www.oracle.com if you need additional information or have any
21 * questions.
22 *
23 */
24
25 #include "precompiled.hpp"
26 #include "gc/shared/genCollectedHeap.hpp"
27 #include "gc/shared/threadLocalAllocBuffer.inline.hpp"
28 #include "logging/log.hpp"
29 #include "memory/resourceArea.hpp"
30 #include "memory/universe.inline.hpp"
31 #include "oops/oop.inline.hpp"
32 #include "runtime/heapMonitoring.hpp"
33 #include "runtime/thread.inline.hpp"
34 #include "runtime/threadSMR.hpp"
35 #include "utilities/copy.hpp"
36
37 // Thread-Local Edens support
38
39 // static member initialization
40 size_t ThreadLocalAllocBuffer::_max_size = 0;
41 int ThreadLocalAllocBuffer::_reserve_for_allocation_prefetch = 0;
42 unsigned ThreadLocalAllocBuffer::_target_refills = 0;
43 GlobalTLABStats* ThreadLocalAllocBuffer::_global_stats = NULL;
44
45 void ThreadLocalAllocBuffer::clear_before_allocation() {
46 _slow_refill_waste += (unsigned)remaining();
47 make_parsable(true); // also retire the TLAB
48 }
49
50 size_t ThreadLocalAllocBuffer::remaining() {
51 if (current_end() == NULL) {
52 return 0;
53 }
54
55 // TODO: To be deprecated when FastTLABRefill is deprecated.
56 update_end_pointers();
57 return pointer_delta(reserved_end(), top());
58 }
59
60 void ThreadLocalAllocBuffer::accumulate_statistics_before_gc() {
61 global_stats()->initialize();
62
63 for (JavaThreadIteratorWithHandle jtiwh; JavaThread *thread = jtiwh.next(); ) {
64 thread->tlab().accumulate_statistics();
65 thread->tlab().initialize_statistics();
66 }
67
68 // Publish new stats if some allocation occurred.
69 if (global_stats()->allocation() != 0) {
70 global_stats()->publish();
71 global_stats()->print();
72 }
73 }
74
75 void ThreadLocalAllocBuffer::accumulate_statistics() {
76 Thread* thread = myThread();
77 size_t capacity = Universe::heap()->tlab_capacity(thread);
78 size_t used = Universe::heap()->tlab_used(thread);
79
80 _gc_waste += (unsigned)remaining();
81 size_t total_allocated = thread->allocated_bytes();
82 size_t allocated_since_last_gc = total_allocated - _allocated_before_last_gc;
83 _allocated_before_last_gc = total_allocated;
84
85 print_stats("gc");
86
87 if (_number_of_refills > 0) {
88 // Update allocation history if a reasonable amount of eden was allocated.
89 bool update_allocation_history = used > 0.5 * capacity;
90
91 if (update_allocation_history) {
92 // Average the fraction of eden allocated in a tlab by this
93 // thread for use in the next resize operation.
94 // _gc_waste is not subtracted because it's included in
95 // "used".
96 // The result can be larger than 1.0 due to direct to old allocations.
97 // These allocations should ideally not be counted but since it is not possible
98 // to filter them out here we just cap the fraction to be at most 1.0.
99 double alloc_frac = MIN2(1.0, (double) allocated_since_last_gc / used);
100 _allocation_fraction.sample(alloc_frac);
101 }
102 global_stats()->update_allocating_threads();
103 global_stats()->update_number_of_refills(_number_of_refills);
104 global_stats()->update_allocation(_number_of_refills * desired_size());
105 global_stats()->update_gc_waste(_gc_waste);
106 global_stats()->update_slow_refill_waste(_slow_refill_waste);
107 global_stats()->update_fast_refill_waste(_fast_refill_waste);
108
109 } else {
110 assert(_number_of_refills == 0 && _fast_refill_waste == 0 &&
111 _slow_refill_waste == 0 && _gc_waste == 0,
112 "tlab stats == 0");
113 }
114 global_stats()->update_slow_allocations(_slow_allocations);
115 }
116
117 // Fills the current tlab with a dummy filler array to create
118 // an illusion of a contiguous Eden and optionally retires the tlab.
119 // Waste accounting should be done in caller as appropriate; see,
120 // for example, clear_before_allocation().
121 void ThreadLocalAllocBuffer::make_parsable(bool retire, bool zap) {
122 if (current_end() != NULL) {
123 invariants();
124
125 if (retire) {
126 myThread()->incr_allocated_bytes(used_bytes());
127 }
128
129 // TODO: To be deprecated when FastTLABRefill is deprecated.
130 update_end_pointers();
131 CollectedHeap::fill_with_object(top(), reserved_end(), retire && zap);
132
133 if (retire || ZeroTLAB) { // "Reset" the TLAB
134 set_start(NULL);
135 set_top(NULL);
136 set_pf_top(NULL);
137 set_current_end(NULL);
138 set_allocation_end(NULL);
139 set_last_slow_path_end(NULL);
140 }
141 }
142 assert(!(retire || ZeroTLAB) ||
143 (start() == NULL && current_end() == NULL && top() == NULL &&
144 _allocation_end == NULL && _last_slow_path_end == NULL),
145 "TLAB must be reset");
146 }
147
148 void ThreadLocalAllocBuffer::resize_all_tlabs() {
149 if (ResizeTLAB) {
150 for (JavaThreadIteratorWithHandle jtiwh; JavaThread *thread = jtiwh.next(); ) {
151 thread->tlab().resize();
152 }
153 }
154 }
155
156 void ThreadLocalAllocBuffer::resize() {
157 // Compute the next tlab size using expected allocation amount
158 assert(ResizeTLAB, "Should not call this otherwise");
159 size_t alloc = (size_t)(_allocation_fraction.average() *
160 (Universe::heap()->tlab_capacity(myThread()) / HeapWordSize));
161 size_t new_size = alloc / _target_refills;
162
163 new_size = MIN2(MAX2(new_size, min_size()), max_size());
164
165 size_t aligned_new_size = align_object_size(new_size);
166
167 log_trace(gc, tlab)("TLAB new size: thread: " INTPTR_FORMAT " [id: %2d]"
168 " refills %d alloc: %8.6f desired_size: " SIZE_FORMAT " -> " SIZE_FORMAT,
169 p2i(myThread()), myThread()->osthread()->thread_id(),
170 _target_refills, _allocation_fraction.average(), desired_size(), aligned_new_size);
171
172 set_desired_size(aligned_new_size);
173 set_refill_waste_limit(initial_refill_waste_limit());
174 }
175
176 void ThreadLocalAllocBuffer::initialize_statistics() {
177 _number_of_refills = 0;
178 _fast_refill_waste = 0;
179 _slow_refill_waste = 0;
180 _gc_waste = 0;
181 _slow_allocations = 0;
182 }
183
184 void ThreadLocalAllocBuffer::fill(HeapWord* start,
185 HeapWord* top,
186 size_t new_size) {
187 _number_of_refills++;
188 print_stats("fill");
189 assert(top <= start + new_size - alignment_reserve(), "size too small");
190
191 // Remember old bytes until sample for the next tlab only if this is our first
192 // actual refill.
193 size_t old_bytes_until_sample = 0;
194 if (_number_of_refills > 1) {
195 old_bytes_until_sample = _bytes_until_sample;
196 }
197
198 initialize(start, top, start + new_size - alignment_reserve());
199
200 if (old_bytes_until_sample > 0) {
201 set_bytes_until_sample(old_bytes_until_sample);
202 set_sample_end();
203 }
204
205 // Reset amount of internal fragmentation
206 set_refill_waste_limit(initial_refill_waste_limit());
207 }
208
209 void ThreadLocalAllocBuffer::initialize(HeapWord* start,
210 HeapWord* top,
211 HeapWord* end) {
212 set_start(start);
213 set_top(top);
214 set_pf_top(top);
215 set_current_end(end);
216 set_allocation_end(end);
217 set_last_slow_path_end(end);
218 invariants();
219 _bytes_until_sample = 0;
220 }
221
222 void ThreadLocalAllocBuffer::initialize() {
223 initialize(NULL, // start
224 NULL, // top
225 NULL); // end
226
227 set_desired_size(initial_desired_size());
228
229 // Following check is needed because at startup the main
230 // thread is initialized before the heap is. The initialization for
231 // this thread is redone in startup_initialization below.
232 if (Universe::heap() != NULL) {
233 size_t capacity = Universe::heap()->tlab_capacity(myThread()) / HeapWordSize;
234 double alloc_frac = desired_size() * target_refills() / (double) capacity;
235 _allocation_fraction.sample(alloc_frac);
236 }
237
238 set_refill_waste_limit(initial_refill_waste_limit());
239
240 initialize_statistics();
241 }
242
243 void ThreadLocalAllocBuffer::startup_initialization() {
244
245 // Assuming each thread's active tlab is, on average,
246 // 1/2 full at a GC
247 _target_refills = 100 / (2 * TLABWasteTargetPercent);
248 _target_refills = MAX2(_target_refills, (unsigned)1U);
249
250 _global_stats = new GlobalTLABStats();
251
252 #ifdef COMPILER2
253 // If the C2 compiler is present, extra space is needed at the end of
254 // TLABs, otherwise prefetching instructions generated by the C2
255 // compiler will fault (due to accessing memory outside of heap).
256 // The amount of space is the max of the number of lines to
257 // prefetch for array and for instance allocations. (Extra space must be
258 // reserved to accommodate both types of allocations.)
259 //
260 // Only SPARC-specific BIS instructions are known to fault. (Those
261 // instructions are generated if AllocatePrefetchStyle==3 and
262 // AllocatePrefetchInstr==1). To be on the safe side, however,
263 // extra space is reserved for all combinations of
264 // AllocatePrefetchStyle and AllocatePrefetchInstr.
265 //
266 // If the C2 compiler is not present, no space is reserved.
267
268 // +1 for rounding up to next cache line, +1 to be safe
269 if (is_server_compilation_mode_vm()) {
270 int lines = MAX2(AllocatePrefetchLines, AllocateInstancePrefetchLines) + 2;
271 _reserve_for_allocation_prefetch = (AllocatePrefetchDistance + AllocatePrefetchStepSize * lines) /
272 (int)HeapWordSize;
273 }
274 #endif
275
276 // During jvm startup, the main thread is initialized
277 // before the heap is initialized. So reinitialize it now.
278 guarantee(Thread::current()->is_Java_thread(), "tlab initialization thread not Java thread");
279 Thread::current()->tlab().initialize();
280
281 log_develop_trace(gc, tlab)("TLAB min: " SIZE_FORMAT " initial: " SIZE_FORMAT " max: " SIZE_FORMAT,
282 min_size(), Thread::current()->tlab().initial_desired_size(), max_size());
283 }
284
285 size_t ThreadLocalAllocBuffer::initial_desired_size() {
286 size_t init_sz = 0;
287
288 if (TLABSize > 0) {
289 init_sz = TLABSize / HeapWordSize;
290 } else if (global_stats() != NULL) {
291 // Initial size is a function of the average number of allocating threads.
292 unsigned nof_threads = global_stats()->allocating_threads_avg();
293
294 init_sz = (Universe::heap()->tlab_capacity(myThread()) / HeapWordSize) /
295 (nof_threads * target_refills());
296 init_sz = align_object_size(init_sz);
297 }
298 init_sz = MIN2(MAX2(init_sz, min_size()), max_size());
299 return init_sz;
300 }
301
302 void ThreadLocalAllocBuffer::print_stats(const char* tag) {
303 Log(gc, tlab) log;
304 if (!log.is_trace()) {
305 return;
306 }
307
308 Thread* thrd = myThread();
309 size_t waste = _gc_waste + _slow_refill_waste + _fast_refill_waste;
310 size_t alloc = _number_of_refills * _desired_size;
311 double waste_percent = percent_of(waste, alloc);
312 size_t tlab_used = Universe::heap()->tlab_used(thrd);
313 log.trace("TLAB: %s thread: " INTPTR_FORMAT " [id: %2d]"
314 " desired_size: " SIZE_FORMAT "KB"
315 " slow allocs: %d refill waste: " SIZE_FORMAT "B"
316 " alloc:%8.5f %8.0fKB refills: %d waste %4.1f%% gc: %dB"
317 " slow: %dB fast: %dB",
318 tag, p2i(thrd), thrd->osthread()->thread_id(),
319 _desired_size / (K / HeapWordSize),
320 _slow_allocations, _refill_waste_limit * HeapWordSize,
321 _allocation_fraction.average(),
322 _allocation_fraction.average() * tlab_used / K,
323 _number_of_refills, waste_percent,
324 _gc_waste * HeapWordSize,
325 _slow_refill_waste * HeapWordSize,
326 _fast_refill_waste * HeapWordSize);
327 }
328
329 void ThreadLocalAllocBuffer::verify() {
330 HeapWord* p = start();
331 HeapWord* t = top();
332 HeapWord* prev_p = NULL;
333 while (p < t) {
334 oop(p)->verify();
335 prev_p = p;
336 p += oop(p)->size();
337 }
338 guarantee(p == top(), "end of last object must match end of space");
339 }
340
341 void ThreadLocalAllocBuffer::set_sample_end() {
342 size_t heap_words_remaining = pointer_delta(_current_end, _top);
343 size_t bytes_left = _bytes_until_sample;
344 size_t words_until_sample = bytes_left / HeapWordSize;
345
346 if (heap_words_remaining > words_until_sample) {
347 HeapWord* new_end = _top + words_until_sample;
348 set_current_end(new_end);
349 set_last_slow_path_end(new_end);
350 set_bytes_until_sample(0);
351 } else {
352 bytes_left -= heap_words_remaining * HeapWordSize;
353 set_bytes_until_sample(bytes_left);
354 }
355 }
356
357 void ThreadLocalAllocBuffer::pick_next_sample(size_t overflowed_words) {
358 if (!HeapMonitoring::enabled()) {
359 return;
360 }
361
362 if (_bytes_until_sample == 0) {
363 HeapMonitoring::pick_next_sample(&_bytes_until_sample);
364 }
365
366 if (overflowed_words > 0) {
367 // Try to correct sample size by removing extra space from last allocation.
368 if (_bytes_until_sample > overflowed_words * HeapWordSize) {
369 set_bytes_until_sample(_bytes_until_sample - overflowed_words * HeapWordSize);
370 }
371 }
372
373 set_sample_end();
374
375 log_trace(gc, tlab)("TLAB picked next sample: thread: " INTPTR_FORMAT " [id: %2d]"
376 " start: " INTPTR_FORMAT " top: " INTPTR_FORMAT " end: "
377 INTPTR_FORMAT " allocation_end:"
378 INTPTR_FORMAT " last_slow_path_end: " INTPTR_FORMAT,
379 p2i(myThread()), myThread()->osthread()->thread_id(),
380 p2i(start()), p2i(top()), p2i(current_end()),
381 p2i(_allocation_end), p2i(_last_slow_path_end));
382 }
383
384 Thread* ThreadLocalAllocBuffer::myThread() {
385 return (Thread*)(((char *)this) +
386 in_bytes(start_offset()) -
387 in_bytes(Thread::tlab_start_offset()));
388 }
389
390 void ThreadLocalAllocBuffer::set_back_allocation_end() {
391 // Did a fast TLAB refill occur?
392 if (_last_slow_path_end != _current_end) {
393 // Fix up the actual end to be now the end of this TLAB.
394 _last_slow_path_end = _current_end;
395 _allocation_end = _current_end;
396 } else {
397 _current_end = _allocation_end;
398 }
399 }
400
401 void ThreadLocalAllocBuffer::handle_sample(Thread* thread, HeapWord* result,
402 size_t size_in_bytes) {
403 if (!HeapMonitoring::enabled()) {
404 return;
405 }
406
407 if (_bytes_until_sample < size_in_bytes) {
408 HeapMonitoring::object_alloc_do_sample(thread,
409 reinterpret_cast<oopDesc*>(result),
410 size_in_bytes);
411 }
412
413 update_tlab_sample_point(size_in_bytes);
414 }
415
416 void ThreadLocalAllocBuffer::update_tlab_sample_point(size_t size_in_bytes) {
417 if (_bytes_until_sample > size_in_bytes) {
418 _bytes_until_sample -= size_in_bytes;
419 return;
420 }
421
422 // We sampled here, so reset it all and start a new sample point.
423 set_bytes_until_sample(0);
424 set_back_allocation_end();
425 pick_next_sample();
426 }
427
428 void ThreadLocalAllocBuffer::update_end_pointers() {
429 // Did a fast TLAB refill occur? (This will be deprecated when fast TLAB
430 // refill disappears).
431 if (_last_slow_path_end != _current_end) {
432 // Fix up the last slow path end to be now the end of this TLAB.
433 _last_slow_path_end = _current_end;
434 _allocation_end = _current_end;
435 }
436 }
437
438 HeapWord* ThreadLocalAllocBuffer::reserved_end() {
439 assert (_last_slow_path_end == _current_end,
440 "Have to call update_end_pointers before reserved_end.");
441 return _allocation_end + alignment_reserve();
442 }
443
444 GlobalTLABStats::GlobalTLABStats() :
445 _allocating_threads_avg(TLABAllocationWeight) {
446
447 initialize();
448
449 _allocating_threads_avg.sample(1); // One allocating thread at startup
450
451 if (UsePerfData) {
452
453 EXCEPTION_MARK;
454 ResourceMark rm;
455
456 char* cname = PerfDataManager::counter_name("tlab", "allocThreads");
457 _perf_allocating_threads =
458 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_None, CHECK);
459
460 cname = PerfDataManager::counter_name("tlab", "fills");
461 _perf_total_refills =
462 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_None, CHECK);
463
464 cname = PerfDataManager::counter_name("tlab", "maxFills");
465 _perf_max_refills =
466 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_None, CHECK);
467
468 cname = PerfDataManager::counter_name("tlab", "alloc");
469 _perf_allocation =
470 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_Bytes, CHECK);
471
472 cname = PerfDataManager::counter_name("tlab", "gcWaste");
473 _perf_gc_waste =
474 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_Bytes, CHECK);
475
476 cname = PerfDataManager::counter_name("tlab", "maxGcWaste");
477 _perf_max_gc_waste =
478 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_Bytes, CHECK);
479
480 cname = PerfDataManager::counter_name("tlab", "slowWaste");
481 _perf_slow_refill_waste =
482 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_Bytes, CHECK);
483
484 cname = PerfDataManager::counter_name("tlab", "maxSlowWaste");
485 _perf_max_slow_refill_waste =
486 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_Bytes, CHECK);
487
488 cname = PerfDataManager::counter_name("tlab", "fastWaste");
489 _perf_fast_refill_waste =
490 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_Bytes, CHECK);
491
492 cname = PerfDataManager::counter_name("tlab", "maxFastWaste");
493 _perf_max_fast_refill_waste =
494 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_Bytes, CHECK);
495
496 cname = PerfDataManager::counter_name("tlab", "slowAlloc");
497 _perf_slow_allocations =
498 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_None, CHECK);
499
500 cname = PerfDataManager::counter_name("tlab", "maxSlowAlloc");
501 _perf_max_slow_allocations =
502 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_None, CHECK);
503 }
504 }
505
506 void GlobalTLABStats::initialize() {
507 // Clear counters summarizing info from all threads
508 _allocating_threads = 0;
509 _total_refills = 0;
510 _max_refills = 0;
511 _total_allocation = 0;
512 _total_gc_waste = 0;
513 _max_gc_waste = 0;
514 _total_slow_refill_waste = 0;
515 _max_slow_refill_waste = 0;
516 _total_fast_refill_waste = 0;
517 _max_fast_refill_waste = 0;
518 _total_slow_allocations = 0;
519 _max_slow_allocations = 0;
520 }
521
522 void GlobalTLABStats::publish() {
523 _allocating_threads_avg.sample(_allocating_threads);
524 if (UsePerfData) {
525 _perf_allocating_threads ->set_value(_allocating_threads);
526 _perf_total_refills ->set_value(_total_refills);
527 _perf_max_refills ->set_value(_max_refills);
528 _perf_allocation ->set_value(_total_allocation);
529 _perf_gc_waste ->set_value(_total_gc_waste);
530 _perf_max_gc_waste ->set_value(_max_gc_waste);
531 _perf_slow_refill_waste ->set_value(_total_slow_refill_waste);
532 _perf_max_slow_refill_waste->set_value(_max_slow_refill_waste);
533 _perf_fast_refill_waste ->set_value(_total_fast_refill_waste);
534 _perf_max_fast_refill_waste->set_value(_max_fast_refill_waste);
535 _perf_slow_allocations ->set_value(_total_slow_allocations);
536 _perf_max_slow_allocations ->set_value(_max_slow_allocations);
537 }
538 }
539
540 void GlobalTLABStats::print() {
541 Log(gc, tlab) log;
542 if (!log.is_debug()) {
543 return;
544 }
545
546 size_t waste = _total_gc_waste + _total_slow_refill_waste + _total_fast_refill_waste;
547 double waste_percent = percent_of(waste, _total_allocation);
548 log.debug("TLAB totals: thrds: %d refills: %d max: %d"
549 " slow allocs: %d max %d waste: %4.1f%%"
550 " gc: " SIZE_FORMAT "B max: " SIZE_FORMAT "B"
551 " slow: " SIZE_FORMAT "B max: " SIZE_FORMAT "B"
552 " fast: " SIZE_FORMAT "B max: " SIZE_FORMAT "B",
553 _allocating_threads,
554 _total_refills, _max_refills,
555 _total_slow_allocations, _max_slow_allocations,
556 waste_percent,
557 _total_gc_waste * HeapWordSize,
558 _max_gc_waste * HeapWordSize,
559 _total_slow_refill_waste * HeapWordSize,
560 _max_slow_refill_waste * HeapWordSize,
561 _total_fast_refill_waste * HeapWordSize,
562 _max_fast_refill_waste * HeapWordSize);
563 }