1 /*
2 * Copyright (c) 1999, 2018, 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.hpp"
31 #include "oops/oop.inline.hpp"
32 #include "runtime/thread.inline.hpp"
33 #include "runtime/threadSMR.hpp"
34 #include "utilities/copy.hpp"
35
36 // Thread-Local Edens support
37
38 // static member initialization
39 size_t ThreadLocalAllocBuffer::_max_size = 0;
40 int ThreadLocalAllocBuffer::_reserve_for_allocation_prefetch = 0;
41 unsigned ThreadLocalAllocBuffer::_target_refills = 0;
42 GlobalTLABStats* ThreadLocalAllocBuffer::_global_stats = NULL;
43
44 void ThreadLocalAllocBuffer::clear_before_allocation() {
45 _slow_refill_waste += (unsigned)remaining();
46 make_parsable(true); // also retire the TLAB
47 }
48
49 size_t ThreadLocalAllocBuffer::remaining() {
50 if (current_end() == NULL) {
51 return 0;
52 }
53
54 // TODO: To be deprecated when FastTLABRefill is deprecated.
55 update_end_pointers();
56 return pointer_delta(reserved_end(), top());
57 }
58
59 void ThreadLocalAllocBuffer::accumulate_statistics_before_gc() {
60 global_stats()->initialize();
61
62 for (JavaThreadIteratorWithHandle jtiwh; JavaThread *thread = jtiwh.next(); ) {
63 thread->tlab().accumulate_statistics();
64 thread->tlab().initialize_statistics();
65 }
66
67 // Publish new stats if some allocation occurred.
68 if (global_stats()->allocation() != 0) {
69 global_stats()->publish();
70 global_stats()->print();
71 }
72 }
73
74 void ThreadLocalAllocBuffer::accumulate_statistics() {
75 Thread* thread = myThread();
76 size_t capacity = Universe::heap()->tlab_capacity(thread);
77 size_t used = Universe::heap()->tlab_used(thread);
78
79 _gc_waste += (unsigned)remaining();
80 size_t total_allocated = thread->allocated_bytes();
81 size_t allocated_since_last_gc = total_allocated - _allocated_before_last_gc;
82 _allocated_before_last_gc = total_allocated;
83
84 print_stats("gc");
85
86 if (_number_of_refills > 0) {
87 // Update allocation history if a reasonable amount of eden was allocated.
88 bool update_allocation_history = used > 0.5 * capacity;
89
90 if (update_allocation_history) {
91 // Average the fraction of eden allocated in a tlab by this
92 // thread for use in the next resize operation.
93 // _gc_waste is not subtracted because it's included in
94 // "used".
95 // The result can be larger than 1.0 due to direct to old allocations.
96 // These allocations should ideally not be counted but since it is not possible
97 // to filter them out here we just cap the fraction to be at most 1.0.
98 double alloc_frac = MIN2(1.0, (double) allocated_since_last_gc / used);
99 _allocation_fraction.sample(alloc_frac);
100 }
101 global_stats()->update_allocating_threads();
102 global_stats()->update_number_of_refills(_number_of_refills);
103 global_stats()->update_allocation(_number_of_refills * desired_size());
104 global_stats()->update_gc_waste(_gc_waste);
105 global_stats()->update_slow_refill_waste(_slow_refill_waste);
106 global_stats()->update_fast_refill_waste(_fast_refill_waste);
107
108 } else {
109 assert(_number_of_refills == 0 && _fast_refill_waste == 0 &&
110 _slow_refill_waste == 0 && _gc_waste == 0,
111 "tlab stats == 0");
112 }
113 global_stats()->update_slow_allocations(_slow_allocations);
114 }
115
116 // Fills the current tlab with a dummy filler array to create
117 // an illusion of a contiguous Eden and optionally retires the tlab.
118 // Waste accounting should be done in caller as appropriate; see,
119 // for example, clear_before_allocation().
120 void ThreadLocalAllocBuffer::make_parsable(bool retire, bool zap) {
121 if (current_end() != NULL) {
122 invariants();
123
124 if (retire) {
125 myThread()->incr_allocated_bytes(used_bytes());
126 }
127
128 // TODO: To be deprecated when FastTLABRefill is deprecated.
129 update_end_pointers();
130 CollectedHeap::fill_with_object(top(), reserved_end(), retire && zap);
131
132 if (retire || ZeroTLAB) { // "Reset" the TLAB
133 set_start(NULL);
134 set_top(NULL);
135 set_pf_top(NULL);
136 set_current_end(NULL);
137 set_allocation_end(NULL);
138 set_last_slow_path_end(NULL);
139 }
140 }
141 assert(!(retire || ZeroTLAB) ||
142 (start() == NULL && current_end() == NULL && top() == NULL &&
143 _allocation_end == NULL && _last_slow_path_end == NULL),
144 "TLAB must be reset");
145 }
146
147 void ThreadLocalAllocBuffer::resize_all_tlabs() {
148 if (ResizeTLAB) {
149 for (JavaThreadIteratorWithHandle jtiwh; JavaThread *thread = jtiwh.next(); ) {
150 thread->tlab().resize();
151 }
152 }
153 }
154
155 void ThreadLocalAllocBuffer::resize() {
156 // Compute the next tlab size using expected allocation amount
157 assert(ResizeTLAB, "Should not call this otherwise");
158 size_t alloc = (size_t)(_allocation_fraction.average() *
159 (Universe::heap()->tlab_capacity(myThread()) / HeapWordSize));
160 size_t new_size = alloc / _target_refills;
161
162 new_size = MIN2(MAX2(new_size, min_size()), max_size());
163
164 size_t aligned_new_size = align_object_size(new_size);
165
166 log_trace(gc, tlab)("TLAB new size: thread: " INTPTR_FORMAT " [id: %2d]"
167 " refills %d alloc: %8.6f desired_size: " SIZE_FORMAT " -> " SIZE_FORMAT,
168 p2i(myThread()), myThread()->osthread()->thread_id(),
169 _target_refills, _allocation_fraction.average(), desired_size(), aligned_new_size);
170
171 set_desired_size(aligned_new_size);
172 set_refill_waste_limit(initial_refill_waste_limit());
173 }
174
175 void ThreadLocalAllocBuffer::initialize_statistics() {
176 _number_of_refills = 0;
177 _fast_refill_waste = 0;
178 _slow_refill_waste = 0;
179 _gc_waste = 0;
180 _slow_allocations = 0;
181 }
182
183 void ThreadLocalAllocBuffer::fill(HeapWord* start,
184 HeapWord* top,
185 size_t new_size) {
186 _number_of_refills++;
187 print_stats("fill");
188 assert(top <= start + new_size - alignment_reserve(), "size too small");
189
190 initialize(start, top, start + new_size - alignment_reserve());
191
192 if (ThreadHeapSampler::enabled()) {
193 set_sample_end();
194 }
195
196 // Reset amount of internal fragmentation
197 set_refill_waste_limit(initial_refill_waste_limit());
198 }
199
200 void ThreadLocalAllocBuffer::initialize(HeapWord* start,
201 HeapWord* top,
202 HeapWord* end) {
203 set_start(start);
204 set_top(top);
205 set_pf_top(top);
206 set_current_end(end);
207 set_allocation_end(end);
208 set_last_slow_path_end(end);
209 invariants();
210 }
211
212 void ThreadLocalAllocBuffer::initialize() {
213 initialize(NULL, // start
214 NULL, // top
215 NULL); // end
216
217 set_desired_size(initial_desired_size());
218
219 // Following check is needed because at startup the main
220 // thread is initialized before the heap is. The initialization for
221 // this thread is redone in startup_initialization below.
222 if (Universe::heap() != NULL) {
223 size_t capacity = Universe::heap()->tlab_capacity(myThread()) / HeapWordSize;
224 double alloc_frac = desired_size() * target_refills() / (double) capacity;
225 _allocation_fraction.sample(alloc_frac);
226 }
227
228 set_refill_waste_limit(initial_refill_waste_limit());
229
230 initialize_statistics();
231 }
232
233 void ThreadLocalAllocBuffer::startup_initialization() {
234
235 // Assuming each thread's active tlab is, on average,
236 // 1/2 full at a GC
237 _target_refills = 100 / (2 * TLABWasteTargetPercent);
238 _target_refills = MAX2(_target_refills, (unsigned)1U);
239
240 _global_stats = new GlobalTLABStats();
241
242 #ifdef COMPILER2
243 // If the C2 compiler is present, extra space is needed at the end of
244 // TLABs, otherwise prefetching instructions generated by the C2
245 // compiler will fault (due to accessing memory outside of heap).
246 // The amount of space is the max of the number of lines to
247 // prefetch for array and for instance allocations. (Extra space must be
248 // reserved to accommodate both types of allocations.)
249 //
250 // Only SPARC-specific BIS instructions are known to fault. (Those
251 // instructions are generated if AllocatePrefetchStyle==3 and
252 // AllocatePrefetchInstr==1). To be on the safe side, however,
253 // extra space is reserved for all combinations of
254 // AllocatePrefetchStyle and AllocatePrefetchInstr.
255 //
256 // If the C2 compiler is not present, no space is reserved.
257
258 // +1 for rounding up to next cache line, +1 to be safe
259 if (is_server_compilation_mode_vm()) {
260 int lines = MAX2(AllocatePrefetchLines, AllocateInstancePrefetchLines) + 2;
261 _reserve_for_allocation_prefetch = (AllocatePrefetchDistance + AllocatePrefetchStepSize * lines) /
262 (int)HeapWordSize;
263 }
264 #endif
265
266 // During jvm startup, the main thread is initialized
267 // before the heap is initialized. So reinitialize it now.
268 guarantee(Thread::current()->is_Java_thread(), "tlab initialization thread not Java thread");
269 Thread::current()->tlab().initialize();
270
271 log_develop_trace(gc, tlab)("TLAB min: " SIZE_FORMAT " initial: " SIZE_FORMAT " max: " SIZE_FORMAT,
272 min_size(), Thread::current()->tlab().initial_desired_size(), max_size());
273 }
274
275 size_t ThreadLocalAllocBuffer::initial_desired_size() {
276 size_t init_sz = 0;
277
278 if (TLABSize > 0) {
279 init_sz = TLABSize / HeapWordSize;
280 } else if (global_stats() != NULL) {
281 // Initial size is a function of the average number of allocating threads.
282 unsigned nof_threads = global_stats()->allocating_threads_avg();
283
284 init_sz = (Universe::heap()->tlab_capacity(myThread()) / HeapWordSize) /
285 (nof_threads * target_refills());
286 init_sz = align_object_size(init_sz);
287 }
288 init_sz = MIN2(MAX2(init_sz, min_size()), max_size());
289 return init_sz;
290 }
291
292 void ThreadLocalAllocBuffer::print_stats(const char* tag) {
293 Log(gc, tlab) log;
294 if (!log.is_trace()) {
295 return;
296 }
297
298 Thread* thrd = myThread();
299 size_t waste = _gc_waste + _slow_refill_waste + _fast_refill_waste;
300 size_t alloc = _number_of_refills * _desired_size;
301 double waste_percent = percent_of(waste, alloc);
302 size_t tlab_used = Universe::heap()->tlab_used(thrd);
303 log.trace("TLAB: %s thread: " INTPTR_FORMAT " [id: %2d]"
304 " desired_size: " SIZE_FORMAT "KB"
305 " slow allocs: %d refill waste: " SIZE_FORMAT "B"
306 " alloc:%8.5f %8.0fKB refills: %d waste %4.1f%% gc: %dB"
307 " slow: %dB fast: %dB",
308 tag, p2i(thrd), thrd->osthread()->thread_id(),
309 _desired_size / (K / HeapWordSize),
310 _slow_allocations, _refill_waste_limit * HeapWordSize,
311 _allocation_fraction.average(),
312 _allocation_fraction.average() * tlab_used / K,
313 _number_of_refills, waste_percent,
314 _gc_waste * HeapWordSize,
315 _slow_refill_waste * HeapWordSize,
316 _fast_refill_waste * HeapWordSize);
317 }
318
319 void ThreadLocalAllocBuffer::verify() {
320 HeapWord* p = start();
321 HeapWord* t = top();
322 HeapWord* prev_p = NULL;
323 while (p < t) {
324 oop(p)->verify();
325 prev_p = p;
326 p += oop(p)->size();
327 }
328 guarantee(p == top(), "end of last object must match end of space");
329 }
330
331 void ThreadLocalAllocBuffer::set_sample_end() {
332 size_t heap_words_remaining = pointer_delta(_current_end, _top);
333 size_t bytes_until_sample = myThread()->heap_sampler().bytes_until_sample();
334 size_t words_until_sample = bytes_until_sample / HeapWordSize;;
335
336 if (heap_words_remaining > words_until_sample) {
337 HeapWord* new_end = _top + words_until_sample;
338 set_current_end(new_end);
339 set_last_slow_path_end(new_end);
340 _bytes_since_last_sample_point = bytes_until_sample;
341 } else {
342 _bytes_since_last_sample_point = heap_words_remaining * HeapWordSize;;
343 }
344 }
345
346 Thread* ThreadLocalAllocBuffer::myThread() {
347 return (Thread*)(((char *)this) +
348 in_bytes(start_offset()) -
349 in_bytes(Thread::tlab_start_offset()));
350 }
351
352 void ThreadLocalAllocBuffer::set_back_allocation_end() {
353 update_end_pointers();
354 _current_end = _allocation_end;
355 }
356
357 void ThreadLocalAllocBuffer::update_end_pointers() {
358 // Did a fast TLAB refill occur? (This will be deprecated when fast TLAB
359 // refill disappears).
360 if (_last_slow_path_end != _current_end) {
361 // Fix up the last slow path end to be now the end of this TLAB.
362 _last_slow_path_end = _current_end;
363 _allocation_end = _current_end;
364 }
365 }
366
367 HeapWord* ThreadLocalAllocBuffer::allocate_sampled_object(size_t size) {
368 Thread* thread = myThread();
369 thread->tlab().set_back_allocation_end();
370 HeapWord* result = thread->tlab().allocate(size);
371
372 if (result) {
373 thread->heap_sampler().check_for_sampling(result, size * HeapWordSize, _bytes_since_last_sample_point);
374 thread->tlab().set_sample_end();
375 }
376
377 return result;
378 }
379
380 HeapWord* ThreadLocalAllocBuffer::reserved_end() {
381 assert (_last_slow_path_end == _current_end,
382 "Have to call update_end_pointers before reserved_end.");
383 return _allocation_end + alignment_reserve();
384 }
385
386 GlobalTLABStats::GlobalTLABStats() :
387 _allocating_threads_avg(TLABAllocationWeight) {
388
389 initialize();
390
391 _allocating_threads_avg.sample(1); // One allocating thread at startup
392
393 if (UsePerfData) {
394
395 EXCEPTION_MARK;
396 ResourceMark rm;
397
398 char* cname = PerfDataManager::counter_name("tlab", "allocThreads");
399 _perf_allocating_threads =
400 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_None, CHECK);
401
402 cname = PerfDataManager::counter_name("tlab", "fills");
403 _perf_total_refills =
404 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_None, CHECK);
405
406 cname = PerfDataManager::counter_name("tlab", "maxFills");
407 _perf_max_refills =
408 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_None, CHECK);
409
410 cname = PerfDataManager::counter_name("tlab", "alloc");
411 _perf_allocation =
412 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_Bytes, CHECK);
413
414 cname = PerfDataManager::counter_name("tlab", "gcWaste");
415 _perf_gc_waste =
416 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_Bytes, CHECK);
417
418 cname = PerfDataManager::counter_name("tlab", "maxGcWaste");
419 _perf_max_gc_waste =
420 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_Bytes, CHECK);
421
422 cname = PerfDataManager::counter_name("tlab", "slowWaste");
423 _perf_slow_refill_waste =
424 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_Bytes, CHECK);
425
426 cname = PerfDataManager::counter_name("tlab", "maxSlowWaste");
427 _perf_max_slow_refill_waste =
428 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_Bytes, CHECK);
429
430 cname = PerfDataManager::counter_name("tlab", "fastWaste");
431 _perf_fast_refill_waste =
432 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_Bytes, CHECK);
433
434 cname = PerfDataManager::counter_name("tlab", "maxFastWaste");
435 _perf_max_fast_refill_waste =
436 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_Bytes, CHECK);
437
438 cname = PerfDataManager::counter_name("tlab", "slowAlloc");
439 _perf_slow_allocations =
440 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_None, CHECK);
441
442 cname = PerfDataManager::counter_name("tlab", "maxSlowAlloc");
443 _perf_max_slow_allocations =
444 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_None, CHECK);
445 }
446 }
447
448 void GlobalTLABStats::initialize() {
449 // Clear counters summarizing info from all threads
450 _allocating_threads = 0;
451 _total_refills = 0;
452 _max_refills = 0;
453 _total_allocation = 0;
454 _total_gc_waste = 0;
455 _max_gc_waste = 0;
456 _total_slow_refill_waste = 0;
457 _max_slow_refill_waste = 0;
458 _total_fast_refill_waste = 0;
459 _max_fast_refill_waste = 0;
460 _total_slow_allocations = 0;
461 _max_slow_allocations = 0;
462 }
463
464 void GlobalTLABStats::publish() {
465 _allocating_threads_avg.sample(_allocating_threads);
466 if (UsePerfData) {
467 _perf_allocating_threads ->set_value(_allocating_threads);
468 _perf_total_refills ->set_value(_total_refills);
469 _perf_max_refills ->set_value(_max_refills);
470 _perf_allocation ->set_value(_total_allocation);
471 _perf_gc_waste ->set_value(_total_gc_waste);
472 _perf_max_gc_waste ->set_value(_max_gc_waste);
473 _perf_slow_refill_waste ->set_value(_total_slow_refill_waste);
474 _perf_max_slow_refill_waste->set_value(_max_slow_refill_waste);
475 _perf_fast_refill_waste ->set_value(_total_fast_refill_waste);
476 _perf_max_fast_refill_waste->set_value(_max_fast_refill_waste);
477 _perf_slow_allocations ->set_value(_total_slow_allocations);
478 _perf_max_slow_allocations ->set_value(_max_slow_allocations);
479 }
480 }
481
482 void GlobalTLABStats::print() {
483 Log(gc, tlab) log;
484 if (!log.is_debug()) {
485 return;
486 }
487
488 size_t waste = _total_gc_waste + _total_slow_refill_waste + _total_fast_refill_waste;
489 double waste_percent = percent_of(waste, _total_allocation);
490 log.debug("TLAB totals: thrds: %d refills: %d max: %d"
491 " slow allocs: %d max %d waste: %4.1f%%"
492 " gc: " SIZE_FORMAT "B max: " SIZE_FORMAT "B"
493 " slow: " SIZE_FORMAT "B max: " SIZE_FORMAT "B"
494 " fast: " SIZE_FORMAT "B max: " SIZE_FORMAT "B",
495 _allocating_threads,
496 _total_refills, _max_refills,
497 _total_slow_allocations, _max_slow_allocations,
498 waste_percent,
499 _total_gc_waste * HeapWordSize,
500 _max_gc_waste * HeapWordSize,
501 _total_slow_refill_waste * HeapWordSize,
502 _max_slow_refill_waste * HeapWordSize,
503 _total_fast_refill_waste * HeapWordSize,
504 _max_fast_refill_waste * HeapWordSize);
505 }