1 /*
2 * Copyright (c) 1999, 2015, Oracle and/or its affiliates. All rights reserved.
3 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
4 *
5 * This code is free software; you can redistribute it and/or modify it
6 * under the terms of the GNU General Public License version 2 only, as
7 * published by the Free Software Foundation.
8 *
9 * This code is distributed in the hope that it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License
12 * version 2 for more details (a copy is included in the LICENSE file that
13 * accompanied this code).
14 *
15 * You should have received a copy of the GNU General Public License version
16 * 2 along with this work; if not, write to the Free Software Foundation,
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
18 *
19 * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
20 * or visit www.oracle.com if you need additional information or have any
21 * questions.
22 *
23 */
24
25 #include "precompiled.hpp"
26 #include "gc/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 "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 void ThreadLocalAllocBuffer::accumulate_statistics_before_gc() {
50 global_stats()->initialize();
51
52 for (JavaThread *thread = Threads::first(); thread != NULL; thread = thread->next()) {
53 thread->tlab().accumulate_statistics();
54 thread->tlab().initialize_statistics();
55 }
56
57 // Publish new stats if some allocation occurred.
58 if (global_stats()->allocation() != 0) {
59 global_stats()->publish();
60 global_stats()->print();
61 }
62 }
63
64 void ThreadLocalAllocBuffer::accumulate_statistics() {
65 Thread* thread = myThread();
66 size_t capacity = Universe::heap()->tlab_capacity(thread);
67 size_t used = Universe::heap()->tlab_used(thread);
68
69 _gc_waste += (unsigned)remaining();
70 size_t total_allocated = thread->allocated_bytes();
71 size_t allocated_since_last_gc = total_allocated - _allocated_before_last_gc;
72 _allocated_before_last_gc = total_allocated;
73
74 print_stats("gc");
75
76 if (_number_of_refills > 0) {
77 // Update allocation history if a reasonable amount of eden was allocated.
78 bool update_allocation_history = used > 0.5 * capacity;
79
80 if (update_allocation_history) {
81 // Average the fraction of eden allocated in a tlab by this
82 // thread for use in the next resize operation.
83 // _gc_waste is not subtracted because it's included in
84 // "used".
85 // The result can be larger than 1.0 due to direct to old allocations.
86 // These allocations should ideally not be counted but since it is not possible
87 // to filter them out here we just cap the fraction to be at most 1.0.
88 double alloc_frac = MIN2(1.0, (double) allocated_since_last_gc / used);
89 _allocation_fraction.sample(alloc_frac);
90 }
91 global_stats()->update_allocating_threads();
92 global_stats()->update_number_of_refills(_number_of_refills);
93 global_stats()->update_allocation(_number_of_refills * desired_size());
94 global_stats()->update_gc_waste(_gc_waste);
95 global_stats()->update_slow_refill_waste(_slow_refill_waste);
96 global_stats()->update_fast_refill_waste(_fast_refill_waste);
97
98 } else {
99 assert(_number_of_refills == 0 && _fast_refill_waste == 0 &&
100 _slow_refill_waste == 0 && _gc_waste == 0,
101 "tlab stats == 0");
102 }
103 global_stats()->update_slow_allocations(_slow_allocations);
104 }
105
106 // Fills the current tlab with a dummy filler array to create
107 // an illusion of a contiguous Eden and optionally retires the tlab.
108 // Waste accounting should be done in caller as appropriate; see,
109 // for example, clear_before_allocation().
110 void ThreadLocalAllocBuffer::make_parsable(bool retire, bool zap) {
111 if (end() != NULL) {
112 invariants();
113
114 if (retire) {
115 myThread()->incr_allocated_bytes(used_bytes());
116 }
117
118 CollectedHeap::fill_with_object(top(), hard_end(), retire && zap);
119
120 if (retire || ZeroTLAB) { // "Reset" the TLAB
121 set_start(NULL);
122 set_top(NULL);
123 set_pf_top(NULL);
124 set_end(NULL);
125 set_actual_end(NULL);
126 set_slow_path_end(NULL);
127 }
128 }
129 assert(!(retire || ZeroTLAB) ||
130 (start() == NULL && end() == NULL && top() == NULL &&
131 actual_end() == NULL && slow_path_end() == NULL),
132 "TLAB must be reset");
133 }
134
135 void ThreadLocalAllocBuffer::resize_all_tlabs() {
136 if (ResizeTLAB) {
137 for (JavaThread *thread = Threads::first(); thread != NULL; thread = thread->next()) {
138 thread->tlab().resize();
139 }
140 }
141 }
142
143 void ThreadLocalAllocBuffer::resize() {
144 // Compute the next tlab size using expected allocation amount
145 assert(ResizeTLAB, "Should not call this otherwise");
146 size_t alloc = (size_t)(_allocation_fraction.average() *
147 (Universe::heap()->tlab_capacity(myThread()) / HeapWordSize));
148 size_t new_size = alloc / _target_refills;
149
150 new_size = MIN2(MAX2(new_size, min_size()), max_size());
151
152 size_t aligned_new_size = align_object_size(new_size);
153
154 log_trace(gc, tlab)("TLAB new size: thread: " INTPTR_FORMAT " [id: %2d]"
155 " refills %d alloc: %8.6f desired_size: " SIZE_FORMAT " -> " SIZE_FORMAT,
156 p2i(myThread()), myThread()->osthread()->thread_id(),
157 _target_refills, _allocation_fraction.average(), desired_size(), aligned_new_size);
158
159 set_desired_size(aligned_new_size);
160 set_refill_waste_limit(initial_refill_waste_limit());
161 }
162
163 void ThreadLocalAllocBuffer::initialize_statistics() {
164 _number_of_refills = 0;
165 _fast_refill_waste = 0;
166 _slow_refill_waste = 0;
167 _gc_waste = 0;
168 _slow_allocations = 0;
169 }
170
171 void ThreadLocalAllocBuffer::fill(HeapWord* start,
172 HeapWord* top,
173 size_t new_size) {
174 _number_of_refills++;
175 print_stats("fill");
176 assert(top <= start + new_size - alignment_reserve(), "size too small");
177 initialize(start, top, start + new_size - alignment_reserve());
178
179 // Reset amount of internal fragmentation
180 set_refill_waste_limit(initial_refill_waste_limit());
181 }
182
183 void ThreadLocalAllocBuffer::initialize(HeapWord* start,
184 HeapWord* top,
185 HeapWord* end) {
186 set_start(start);
187 set_top(top);
188 set_pf_top(top);
189 set_end(end);
190 set_actual_end(end);
191 set_slow_path_end(end);
192 invariants();
193 _bytes_until_sample = 0;
194 }
195
196 void ThreadLocalAllocBuffer::initialize() {
197 initialize(NULL, // start
198 NULL, // top
199 NULL); // end
200
201 set_desired_size(initial_desired_size());
202
203 // Following check is needed because at startup the main (primordial)
204 // thread is initialized before the heap is. The initialization for
205 // this thread is redone in startup_initialization below.
206 if (Universe::heap() != NULL) {
207 size_t capacity = Universe::heap()->tlab_capacity(myThread()) / HeapWordSize;
208 double alloc_frac = desired_size() * target_refills() / (double) capacity;
209 _allocation_fraction.sample(alloc_frac);
210 }
211
212 set_refill_waste_limit(initial_refill_waste_limit());
213
214 initialize_statistics();
215 }
216
217 void ThreadLocalAllocBuffer::startup_initialization() {
218
219 // Assuming each thread's active tlab is, on average,
220 // 1/2 full at a GC
221 _target_refills = 100 / (2 * TLABWasteTargetPercent);
222 _target_refills = MAX2(_target_refills, (unsigned)1U);
223
224 _global_stats = new GlobalTLABStats();
225
226 #ifdef COMPILER2
227 // If the C2 compiler is present, extra space is needed at the end of
228 // TLABs, otherwise prefetching instructions generated by the C2
229 // compiler will fault (due to accessing memory outside of heap).
230 // The amount of space is the max of the number of lines to
231 // prefetch for array and for instance allocations. (Extra space must be
232 // reserved to accommodate both types of allocations.)
233 //
234 // Only SPARC-specific BIS instructions are known to fault. (Those
235 // instructions are generated if AllocatePrefetchStyle==3 and
236 // AllocatePrefetchInstr==1). To be on the safe side, however,
237 // extra space is reserved for all combinations of
238 // AllocatePrefetchStyle and AllocatePrefetchInstr.
239 //
240 // If the C2 compiler is not present, no space is reserved.
241
242 // +1 for rounding up to next cache line, +1 to be safe
243 if (is_server_compilation_mode_vm()) {
244 int lines = MAX2(AllocatePrefetchLines, AllocateInstancePrefetchLines) + 2;
245 _reserve_for_allocation_prefetch = (AllocatePrefetchDistance + AllocatePrefetchStepSize * lines) /
246 (int)HeapWordSize;
247 }
248 #endif
249
250 // During jvm startup, the main (primordial) thread is initialized
251 // before the heap is initialized. So reinitialize it now.
252 guarantee(Thread::current()->is_Java_thread(), "tlab initialization thread not Java thread");
253 Thread::current()->tlab().initialize();
254
255 log_develop_trace(gc, tlab)("TLAB min: " SIZE_FORMAT " initial: " SIZE_FORMAT " max: " SIZE_FORMAT,
256 min_size(), Thread::current()->tlab().initial_desired_size(), max_size());
257 }
258
259 size_t ThreadLocalAllocBuffer::initial_desired_size() {
260 size_t init_sz = 0;
261
262 if (TLABSize > 0) {
263 init_sz = TLABSize / HeapWordSize;
264 } else if (global_stats() != NULL) {
265 // Initial size is a function of the average number of allocating threads.
266 unsigned nof_threads = global_stats()->allocating_threads_avg();
267
268 init_sz = (Universe::heap()->tlab_capacity(myThread()) / HeapWordSize) /
269 (nof_threads * target_refills());
270 init_sz = align_object_size(init_sz);
271 }
272 init_sz = MIN2(MAX2(init_sz, min_size()), max_size());
273 return init_sz;
274 }
275
276 void ThreadLocalAllocBuffer::print_stats(const char* tag) {
277 Log(gc, tlab) log;
278 if (!log.is_trace()) {
279 return;
280 }
281
282 Thread* thrd = myThread();
283 size_t waste = _gc_waste + _slow_refill_waste + _fast_refill_waste;
284 size_t alloc = _number_of_refills * _desired_size;
285 double waste_percent = alloc == 0 ? 0.0 :
286 100.0 * waste / alloc;
287 size_t tlab_used = Universe::heap()->tlab_used(thrd);
288 log.trace("TLAB: %s thread: " INTPTR_FORMAT " [id: %2d]"
289 " desired_size: " SIZE_FORMAT "KB"
290 " slow allocs: %d refill waste: " SIZE_FORMAT "B"
291 " alloc:%8.5f %8.0fKB refills: %d waste %4.1f%% gc: %dB"
292 " slow: %dB fast: %dB",
293 tag, p2i(thrd), thrd->osthread()->thread_id(),
294 _desired_size / (K / HeapWordSize),
295 _slow_allocations, _refill_waste_limit * HeapWordSize,
296 _allocation_fraction.average(),
297 _allocation_fraction.average() * tlab_used / K,
298 _number_of_refills, waste_percent,
299 _gc_waste * HeapWordSize,
300 _slow_refill_waste * HeapWordSize,
301 _fast_refill_waste * HeapWordSize);
302 }
303
304 void ThreadLocalAllocBuffer::verify() {
305 HeapWord* p = start();
306 HeapWord* t = top();
307 HeapWord* prev_p = NULL;
308 while (p < t) {
309 oop(p)->verify();
310 prev_p = p;
311 p += oop(p)->size();
312 }
313 guarantee(p == top(), "end of last object must match end of space");
314 }
315
316 void ThreadLocalAllocBuffer::pick_next_sample() {
317 if (!HeapMonitoring::enabled()) {
318 return;
319 }
320
321 if (bytes_until_sample() == 0) {
322 HeapMonitoring::pick_next_sample(bytes_until_sample_addr());
323 }
324
325 // Finally, fix up the sampling bytes left and _end.
326 size_t heap_words_remaining = _end - _top;
327 size_t bytes_left = bytes_until_sample();
328 size_t words_until_sample = bytes_left / HeapWordSize;
329
330 if (heap_words_remaining > words_until_sample) {
331 HeapWord* new_end = _top + words_until_sample;
332 set_end(new_end);
333 set_slow_path_end(new_end);
334 set_bytes_until_sample(0);
335 } else {
336 bytes_left -= heap_words_remaining * HeapWordSize;
337 set_bytes_until_sample(bytes_left);
338 }
339
340 log_trace(gc, tlab)("TLAB picked next sample: thread: " INTPTR_FORMAT " [id: %2d]"
341 " start: %p top: %p end: %p actual_end: %p slow_path_end: %p",
342 p2i(myThread()), myThread()->osthread()->thread_id(),
343 start(), top(), end(),
344 actual_end(), slow_path_end());
345 }
346
347 Thread* ThreadLocalAllocBuffer::myThread() {
348 return (Thread*)(((char *)this) +
349 in_bytes(start_offset()) -
350 in_bytes(Thread::tlab_start_offset()));
351 }
352
353 void ThreadLocalAllocBuffer::set_back_actual_end() {
354 // Did a fast TLAB refill occur?
355 if (_slow_path_end != _end) {
356 // Fix up the actual end to be now the end of this TLAB.
357 _slow_path_end = _end;
358 _actual_end = _end;
359 } else {
360 _end = _actual_end;
361 }
362 }
363
364 void ThreadLocalAllocBuffer::handle_sample(Thread* thread, HeapWord* result,
365 size_t size) {
366 if (!HeapMonitoring::enabled()) {
367 return;
368 }
369
370 set_bytes_until_sample(bytes_until_sample() - size);
371
372 // Should we sample now?
373 set_back_actual_end();
374 if (should_sample()) {
375 HeapMonitoring::object_alloc_do_sample(thread,
376 reinterpret_cast<oopDesc*>(result),
377 size);
378 }
379 pick_next_sample();
380 }
381
382 HeapWord* ThreadLocalAllocBuffer::hard_end() {
383 // Did a fast TLAB refill occur?
384 if (_slow_path_end != _end) {
385 // Fix up the actual end to be now the end of this TLAB.
386 _slow_path_end = _end;
387 _actual_end = _end;
388 }
389
390 return _actual_end + alignment_reserve();
391 }
392 GlobalTLABStats::GlobalTLABStats() :
393 _allocating_threads_avg(TLABAllocationWeight) {
394
395 initialize();
396
397 _allocating_threads_avg.sample(1); // One allocating thread at startup
398
399 if (UsePerfData) {
400
401 EXCEPTION_MARK;
402 ResourceMark rm;
403
404 char* cname = PerfDataManager::counter_name("tlab", "allocThreads");
405 _perf_allocating_threads =
406 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_None, CHECK);
407
408 cname = PerfDataManager::counter_name("tlab", "fills");
409 _perf_total_refills =
410 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_None, CHECK);
411
412 cname = PerfDataManager::counter_name("tlab", "maxFills");
413 _perf_max_refills =
414 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_None, CHECK);
415
416 cname = PerfDataManager::counter_name("tlab", "alloc");
417 _perf_allocation =
418 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_Bytes, CHECK);
419
420 cname = PerfDataManager::counter_name("tlab", "gcWaste");
421 _perf_gc_waste =
422 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_Bytes, CHECK);
423
424 cname = PerfDataManager::counter_name("tlab", "maxGcWaste");
425 _perf_max_gc_waste =
426 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_Bytes, CHECK);
427
428 cname = PerfDataManager::counter_name("tlab", "slowWaste");
429 _perf_slow_refill_waste =
430 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_Bytes, CHECK);
431
432 cname = PerfDataManager::counter_name("tlab", "maxSlowWaste");
433 _perf_max_slow_refill_waste =
434 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_Bytes, CHECK);
435
436 cname = PerfDataManager::counter_name("tlab", "fastWaste");
437 _perf_fast_refill_waste =
438 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_Bytes, CHECK);
439
440 cname = PerfDataManager::counter_name("tlab", "maxFastWaste");
441 _perf_max_fast_refill_waste =
442 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_Bytes, CHECK);
443
444 cname = PerfDataManager::counter_name("tlab", "slowAlloc");
445 _perf_slow_allocations =
446 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_None, CHECK);
447
448 cname = PerfDataManager::counter_name("tlab", "maxSlowAlloc");
449 _perf_max_slow_allocations =
450 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_None, CHECK);
451 }
452 }
453
454 void GlobalTLABStats::initialize() {
455 // Clear counters summarizing info from all threads
456 _allocating_threads = 0;
457 _total_refills = 0;
458 _max_refills = 0;
459 _total_allocation = 0;
460 _total_gc_waste = 0;
461 _max_gc_waste = 0;
462 _total_slow_refill_waste = 0;
463 _max_slow_refill_waste = 0;
464 _total_fast_refill_waste = 0;
465 _max_fast_refill_waste = 0;
466 _total_slow_allocations = 0;
467 _max_slow_allocations = 0;
468 }
469
470 void GlobalTLABStats::publish() {
471 _allocating_threads_avg.sample(_allocating_threads);
472 if (UsePerfData) {
473 _perf_allocating_threads ->set_value(_allocating_threads);
474 _perf_total_refills ->set_value(_total_refills);
475 _perf_max_refills ->set_value(_max_refills);
476 _perf_allocation ->set_value(_total_allocation);
477 _perf_gc_waste ->set_value(_total_gc_waste);
478 _perf_max_gc_waste ->set_value(_max_gc_waste);
479 _perf_slow_refill_waste ->set_value(_total_slow_refill_waste);
480 _perf_max_slow_refill_waste->set_value(_max_slow_refill_waste);
481 _perf_fast_refill_waste ->set_value(_total_fast_refill_waste);
482 _perf_max_fast_refill_waste->set_value(_max_fast_refill_waste);
483 _perf_slow_allocations ->set_value(_total_slow_allocations);
484 _perf_max_slow_allocations ->set_value(_max_slow_allocations);
485 }
486 }
487
488 void GlobalTLABStats::print() {
489 Log(gc, tlab) log;
490 if (!log.is_debug()) {
491 return;
492 }
493
494 size_t waste = _total_gc_waste + _total_slow_refill_waste + _total_fast_refill_waste;
495 double waste_percent = _total_allocation == 0 ? 0.0 :
496 100.0 * waste / _total_allocation;
497 log.debug("TLAB totals: thrds: %d refills: %d max: %d"
498 " slow allocs: %d max %d waste: %4.1f%%"
499 " gc: " SIZE_FORMAT "B max: " SIZE_FORMAT "B"
500 " slow: " SIZE_FORMAT "B max: " SIZE_FORMAT "B"
501 " fast: " SIZE_FORMAT "B max: " SIZE_FORMAT "B",
502 _allocating_threads,
503 _total_refills, _max_refills,
504 _total_slow_allocations, _max_slow_allocations,
505 waste_percent,
506 _total_gc_waste * HeapWordSize,
507 _max_gc_waste * HeapWordSize,
508 _total_slow_refill_waste * HeapWordSize,
509 _max_slow_refill_waste * HeapWordSize,
510 _total_fast_refill_waste * HeapWordSize,
511 _max_fast_refill_waste * HeapWordSize);
512 }