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 "TLAB must be reset");
132 }
133
134 void ThreadLocalAllocBuffer::resize_all_tlabs() {
135 if (ResizeTLAB) {
136 for (JavaThread *thread = Threads::first(); thread != NULL; thread = thread->next()) {
137 thread->tlab().resize();
138 }
139 }
140 }
141
142 void ThreadLocalAllocBuffer::resize() {
143 // Compute the next tlab size using expected allocation amount
144 assert(ResizeTLAB, "Should not call this otherwise");
145 size_t alloc = (size_t)(_allocation_fraction.average() *
146 (Universe::heap()->tlab_capacity(myThread()) / HeapWordSize));
147 size_t new_size = alloc / _target_refills;
148
149 new_size = MIN2(MAX2(new_size, min_size()), max_size());
150
151 size_t aligned_new_size = align_object_size(new_size);
152
153 log_trace(gc, tlab)("TLAB new size: thread: " INTPTR_FORMAT " [id: %2d]"
154 " refills %d alloc: %8.6f desired_size: " SIZE_FORMAT " -> " SIZE_FORMAT,
155 p2i(myThread()), myThread()->osthread()->thread_id(),
156 _target_refills, _allocation_fraction.average(), desired_size(), aligned_new_size);
157
158 set_desired_size(aligned_new_size);
159 set_refill_waste_limit(initial_refill_waste_limit());
160 }
161
162 void ThreadLocalAllocBuffer::initialize_statistics() {
163 _number_of_refills = 0;
164 _fast_refill_waste = 0;
165 _slow_refill_waste = 0;
166 _gc_waste = 0;
167 _slow_allocations = 0;
168 }
169
170 void ThreadLocalAllocBuffer::fill(HeapWord* start,
171 HeapWord* top,
172 size_t new_size) {
173 _number_of_refills++;
174 print_stats("fill");
175 assert(top <= start + new_size - alignment_reserve(), "size too small");
176 initialize(start, top, start + new_size - alignment_reserve());
177
178 // Reset amount of internal fragmentation
179 set_refill_waste_limit(initial_refill_waste_limit());
180 }
181
182 void ThreadLocalAllocBuffer::initialize(HeapWord* start,
183 HeapWord* top,
184 HeapWord* end) {
185 set_start(start);
186 set_top(top);
187 set_pf_top(top);
188 set_end(end);
189 set_actual_end(end);
190 set_slow_path_end(end);
191 invariants();
192 _bytes_until_sample = 0;
193 }
194
195 void ThreadLocalAllocBuffer::initialize() {
196 initialize(NULL, // start
197 NULL, // top
198 NULL); // end
199
200 set_desired_size(initial_desired_size());
201
202 // Following check is needed because at startup the main (primordial)
203 // thread is initialized before the heap is. The initialization for
204 // this thread is redone in startup_initialization below.
205 if (Universe::heap() != NULL) {
206 size_t capacity = Universe::heap()->tlab_capacity(myThread()) / HeapWordSize;
207 double alloc_frac = desired_size() * target_refills() / (double) capacity;
208 _allocation_fraction.sample(alloc_frac);
209 }
210
211 set_refill_waste_limit(initial_refill_waste_limit());
212
213 initialize_statistics();
214 }
215
216 void ThreadLocalAllocBuffer::startup_initialization() {
217
218 // Assuming each thread's active tlab is, on average,
219 // 1/2 full at a GC
220 _target_refills = 100 / (2 * TLABWasteTargetPercent);
221 _target_refills = MAX2(_target_refills, (unsigned)1U);
222
223 _global_stats = new GlobalTLABStats();
224
225 #ifdef COMPILER2
226 // If the C2 compiler is present, extra space is needed at the end of
227 // TLABs, otherwise prefetching instructions generated by the C2
228 // compiler will fault (due to accessing memory outside of heap).
229 // The amount of space is the max of the number of lines to
230 // prefetch for array and for instance allocations. (Extra space must be
231 // reserved to accommodate both types of allocations.)
232 //
233 // Only SPARC-specific BIS instructions are known to fault. (Those
234 // instructions are generated if AllocatePrefetchStyle==3 and
235 // AllocatePrefetchInstr==1). To be on the safe side, however,
236 // extra space is reserved for all combinations of
237 // AllocatePrefetchStyle and AllocatePrefetchInstr.
238 //
239 // If the C2 compiler is not present, no space is reserved.
240
241 // +1 for rounding up to next cache line, +1 to be safe
242 if (is_server_compilation_mode_vm()) {
243 int lines = MAX2(AllocatePrefetchLines, AllocateInstancePrefetchLines) + 2;
244 _reserve_for_allocation_prefetch = (AllocatePrefetchDistance + AllocatePrefetchStepSize * lines) /
245 (int)HeapWordSize;
246 }
247 #endif
248
249 // During jvm startup, the main (primordial) thread is initialized
250 // before the heap is initialized. So reinitialize it now.
251 guarantee(Thread::current()->is_Java_thread(), "tlab initialization thread not Java thread");
252 Thread::current()->tlab().initialize();
253
254 log_develop_trace(gc, tlab)("TLAB min: " SIZE_FORMAT " initial: " SIZE_FORMAT " max: " SIZE_FORMAT,
255 min_size(), Thread::current()->tlab().initial_desired_size(), max_size());
256 }
257
258 size_t ThreadLocalAllocBuffer::initial_desired_size() {
259 size_t init_sz = 0;
260
261 if (TLABSize > 0) {
262 init_sz = TLABSize / HeapWordSize;
263 } else if (global_stats() != NULL) {
264 // Initial size is a function of the average number of allocating threads.
265 unsigned nof_threads = global_stats()->allocating_threads_avg();
266
267 init_sz = (Universe::heap()->tlab_capacity(myThread()) / HeapWordSize) /
268 (nof_threads * target_refills());
269 init_sz = align_object_size(init_sz);
270 }
271 init_sz = MIN2(MAX2(init_sz, min_size()), max_size());
272 return init_sz;
273 }
274
275 void ThreadLocalAllocBuffer::print_stats(const char* tag) {
276 Log(gc, tlab) log;
277 if (!log.is_trace()) {
278 return;
279 }
280
281 Thread* thrd = myThread();
282 size_t waste = _gc_waste + _slow_refill_waste + _fast_refill_waste;
283 size_t alloc = _number_of_refills * _desired_size;
284 double waste_percent = alloc == 0 ? 0.0 :
285 100.0 * waste / alloc;
286 size_t tlab_used = Universe::heap()->tlab_used(thrd);
287 log.trace("TLAB: %s thread: " INTPTR_FORMAT " [id: %2d]"
288 " desired_size: " SIZE_FORMAT "KB"
289 " slow allocs: %d refill waste: " SIZE_FORMAT "B"
290 " alloc:%8.5f %8.0fKB refills: %d waste %4.1f%% gc: %dB"
291 " slow: %dB fast: %dB",
292 tag, p2i(thrd), thrd->osthread()->thread_id(),
293 _desired_size / (K / HeapWordSize),
294 _slow_allocations, _refill_waste_limit * HeapWordSize,
295 _allocation_fraction.average(),
296 _allocation_fraction.average() * tlab_used / K,
297 _number_of_refills, waste_percent,
298 _gc_waste * HeapWordSize,
299 _slow_refill_waste * HeapWordSize,
300 _fast_refill_waste * HeapWordSize);
301 }
302
303 void ThreadLocalAllocBuffer::verify() {
304 HeapWord* p = start();
305 HeapWord* t = top();
306 HeapWord* prev_p = NULL;
307 while (p < t) {
308 oop(p)->verify();
309 prev_p = p;
310 p += oop(p)->size();
311 }
312 guarantee(p == top(), "end of last object must match end of space");
313 }
314
315 void ThreadLocalAllocBuffer::pick_next_sample() {
316 if (!HeapMonitoring::enabled()) {
317 return;
318 }
319
320 if (bytes_until_sample() == 0) {
321 HeapMonitoring::pick_next_sample(bytes_until_sample_addr());
322 }
323
324 // Finally, fix up the sampling bytes left and _end.
325 size_t heap_words_remaining = _end - _top;
326 size_t bytes_left = bytes_until_sample();
327 size_t words_until_sample = bytes_left / HeapWordSize;
328
329 if (heap_words_remaining > words_until_sample) {
330 set_end(_top + words_until_sample);
331 set_bytes_until_sample(0);
332 } else {
333 bytes_left -= heap_words_remaining * HeapWordSize;
334 set_bytes_until_sample(bytes_left);
335 }
336 }
337
338 Thread* ThreadLocalAllocBuffer::myThread() {
339 return (Thread*)(((char *)this) +
340 in_bytes(start_offset()) -
341 in_bytes(Thread::tlab_start_offset()));
342 }
343
344 void ThreadLocalAllocBuffer::set_back_actual_end() {
345 // Did a fast TLAB refill occur?
346 if (_slow_path_end != _end) {
347 // Fix up the actual end to be now the end of this TLAB.
348 _slow_path_end = _end;
349 _actual_end = _end;
350 } else {
351 _end = _actual_end;
352 }
353 }
354
355 void ThreadLocalAllocBuffer::handle_sample(Thread* thread, HeapWord* result,
356 size_t size) {
357 if (!HeapMonitoring::enabled()) {
358 return;
359 }
360
361 set_bytes_until_sample(bytes_until_sample() - size);
362
363 // Should we sample now?
364 set_back_actual_end();
365 if (should_sample()) {
366 HeapMonitoring::object_alloc_do_sample(thread,
367 reinterpret_cast<oopDesc*>(result),
368 size);
369 }
370 pick_next_sample();
371 }
372
373 HeapWord* ThreadLocalAllocBuffer::hard_end() {
374 // Did a fast TLAB refill occur?
375 if (_slow_path_end != _end) {
376 // Fix up the actual end to be now the end of this TLAB.
377 _slow_path_end = _end;
378 _actual_end = _end;
379 }
380
381 return _actual_end + alignment_reserve();
382 }
383 GlobalTLABStats::GlobalTLABStats() :
384 _allocating_threads_avg(TLABAllocationWeight) {
385
386 initialize();
387
388 _allocating_threads_avg.sample(1); // One allocating thread at startup
389
390 if (UsePerfData) {
391
392 EXCEPTION_MARK;
393 ResourceMark rm;
394
395 char* cname = PerfDataManager::counter_name("tlab", "allocThreads");
396 _perf_allocating_threads =
397 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_None, CHECK);
398
399 cname = PerfDataManager::counter_name("tlab", "fills");
400 _perf_total_refills =
401 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_None, CHECK);
402
403 cname = PerfDataManager::counter_name("tlab", "maxFills");
404 _perf_max_refills =
405 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_None, CHECK);
406
407 cname = PerfDataManager::counter_name("tlab", "alloc");
408 _perf_allocation =
409 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_Bytes, CHECK);
410
411 cname = PerfDataManager::counter_name("tlab", "gcWaste");
412 _perf_gc_waste =
413 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_Bytes, CHECK);
414
415 cname = PerfDataManager::counter_name("tlab", "maxGcWaste");
416 _perf_max_gc_waste =
417 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_Bytes, CHECK);
418
419 cname = PerfDataManager::counter_name("tlab", "slowWaste");
420 _perf_slow_refill_waste =
421 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_Bytes, CHECK);
422
423 cname = PerfDataManager::counter_name("tlab", "maxSlowWaste");
424 _perf_max_slow_refill_waste =
425 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_Bytes, CHECK);
426
427 cname = PerfDataManager::counter_name("tlab", "fastWaste");
428 _perf_fast_refill_waste =
429 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_Bytes, CHECK);
430
431 cname = PerfDataManager::counter_name("tlab", "maxFastWaste");
432 _perf_max_fast_refill_waste =
433 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_Bytes, CHECK);
434
435 cname = PerfDataManager::counter_name("tlab", "slowAlloc");
436 _perf_slow_allocations =
437 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_None, CHECK);
438
439 cname = PerfDataManager::counter_name("tlab", "maxSlowAlloc");
440 _perf_max_slow_allocations =
441 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_None, CHECK);
442 }
443 }
444
445 void GlobalTLABStats::initialize() {
446 // Clear counters summarizing info from all threads
447 _allocating_threads = 0;
448 _total_refills = 0;
449 _max_refills = 0;
450 _total_allocation = 0;
451 _total_gc_waste = 0;
452 _max_gc_waste = 0;
453 _total_slow_refill_waste = 0;
454 _max_slow_refill_waste = 0;
455 _total_fast_refill_waste = 0;
456 _max_fast_refill_waste = 0;
457 _total_slow_allocations = 0;
458 _max_slow_allocations = 0;
459 }
460
461 void GlobalTLABStats::publish() {
462 _allocating_threads_avg.sample(_allocating_threads);
463 if (UsePerfData) {
464 _perf_allocating_threads ->set_value(_allocating_threads);
465 _perf_total_refills ->set_value(_total_refills);
466 _perf_max_refills ->set_value(_max_refills);
467 _perf_allocation ->set_value(_total_allocation);
468 _perf_gc_waste ->set_value(_total_gc_waste);
469 _perf_max_gc_waste ->set_value(_max_gc_waste);
470 _perf_slow_refill_waste ->set_value(_total_slow_refill_waste);
471 _perf_max_slow_refill_waste->set_value(_max_slow_refill_waste);
472 _perf_fast_refill_waste ->set_value(_total_fast_refill_waste);
473 _perf_max_fast_refill_waste->set_value(_max_fast_refill_waste);
474 _perf_slow_allocations ->set_value(_total_slow_allocations);
475 _perf_max_slow_allocations ->set_value(_max_slow_allocations);
476 }
477 }
478
479 void GlobalTLABStats::print() {
480 Log(gc, tlab) log;
481 if (!log.is_debug()) {
482 return;
483 }
484
485 size_t waste = _total_gc_waste + _total_slow_refill_waste + _total_fast_refill_waste;
486 double waste_percent = _total_allocation == 0 ? 0.0 :
487 100.0 * waste / _total_allocation;
488 log.debug("TLAB totals: thrds: %d refills: %d max: %d"
489 " slow allocs: %d max %d waste: %4.1f%%"
490 " gc: " SIZE_FORMAT "B max: " SIZE_FORMAT "B"
491 " slow: " SIZE_FORMAT "B max: " SIZE_FORMAT "B"
492 " fast: " SIZE_FORMAT "B max: " SIZE_FORMAT "B",
493 _allocating_threads,
494 _total_refills, _max_refills,
495 _total_slow_allocations, _max_slow_allocations,
496 waste_percent,
497 _total_gc_waste * HeapWordSize,
498 _max_gc_waste * HeapWordSize,
499 _total_slow_refill_waste * HeapWordSize,
500 _max_slow_refill_waste * HeapWordSize,
501 _total_fast_refill_waste * HeapWordSize,
502 _max_fast_refill_waste * HeapWordSize);
503 }