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 }
127 }
128 assert(!(retire || ZeroTLAB) ||
129 (start() == NULL && end() == NULL && top() == NULL),
130 "TLAB must be reset");
131 }
132
133 void ThreadLocalAllocBuffer::resize_all_tlabs() {
134 if (ResizeTLAB) {
135 for (JavaThread *thread = Threads::first(); thread != NULL; thread = thread->next()) {
136 thread->tlab().resize();
137 }
138 }
139 }
140
141 void ThreadLocalAllocBuffer::resize() {
142 // Compute the next tlab size using expected allocation amount
143 assert(ResizeTLAB, "Should not call this otherwise");
144 size_t alloc = (size_t)(_allocation_fraction.average() *
145 (Universe::heap()->tlab_capacity(myThread()) / HeapWordSize));
146 size_t new_size = alloc / _target_refills;
147
148 new_size = MIN2(MAX2(new_size, min_size()), max_size());
149
150 size_t aligned_new_size = align_object_size(new_size);
151
152 log_trace(gc, tlab)("TLAB new size: thread: " INTPTR_FORMAT " [id: %2d]"
153 " refills %d alloc: %8.6f desired_size: " SIZE_FORMAT " -> " SIZE_FORMAT,
154 p2i(myThread()), myThread()->osthread()->thread_id(),
155 _target_refills, _allocation_fraction.average(), desired_size(), aligned_new_size);
156
157 set_desired_size(aligned_new_size);
158 set_refill_waste_limit(initial_refill_waste_limit());
159 }
160
161 void ThreadLocalAllocBuffer::initialize_statistics() {
162 _number_of_refills = 0;
163 _fast_refill_waste = 0;
164 _slow_refill_waste = 0;
165 _gc_waste = 0;
166 _slow_allocations = 0;
167 }
168
169 void ThreadLocalAllocBuffer::fill(HeapWord* start,
170 HeapWord* top,
171 size_t new_size) {
172 _number_of_refills++;
173 print_stats("fill");
174 assert(top <= start + new_size - alignment_reserve(), "size too small");
175 initialize(start, top, start + new_size - alignment_reserve());
176
177 // Reset amount of internal fragmentation
178 set_refill_waste_limit(initial_refill_waste_limit());
179 }
180
181 void ThreadLocalAllocBuffer::initialize(HeapWord* start,
182 HeapWord* top,
183 HeapWord* end) {
184 set_start(start);
185 set_top(top);
186 set_pf_top(top);
187 set_end(end);
188 set_actual_end(end);
189 invariants();
190 _bytes_until_sample = 0;
191 }
192
193 void ThreadLocalAllocBuffer::initialize() {
194 initialize(NULL, // start
195 NULL, // top
196 NULL); // end
197
198 set_desired_size(initial_desired_size());
199
200 // Following check is needed because at startup the main (primordial)
201 // thread is initialized before the heap is. The initialization for
202 // this thread is redone in startup_initialization below.
203 if (Universe::heap() != NULL) {
204 size_t capacity = Universe::heap()->tlab_capacity(myThread()) / HeapWordSize;
205 double alloc_frac = desired_size() * target_refills() / (double) capacity;
206 _allocation_fraction.sample(alloc_frac);
207 }
208
209 set_refill_waste_limit(initial_refill_waste_limit());
210
211 initialize_statistics();
212 }
213
214 void ThreadLocalAllocBuffer::startup_initialization() {
215
216 // Assuming each thread's active tlab is, on average,
217 // 1/2 full at a GC
218 _target_refills = 100 / (2 * TLABWasteTargetPercent);
219 _target_refills = MAX2(_target_refills, (unsigned)1U);
220
221 _global_stats = new GlobalTLABStats();
222
223 #ifdef COMPILER2
224 // If the C2 compiler is present, extra space is needed at the end of
225 // TLABs, otherwise prefetching instructions generated by the C2
226 // compiler will fault (due to accessing memory outside of heap).
227 // The amount of space is the max of the number of lines to
228 // prefetch for array and for instance allocations. (Extra space must be
229 // reserved to accommodate both types of allocations.)
230 //
231 // Only SPARC-specific BIS instructions are known to fault. (Those
232 // instructions are generated if AllocatePrefetchStyle==3 and
233 // AllocatePrefetchInstr==1). To be on the safe side, however,
234 // extra space is reserved for all combinations of
235 // AllocatePrefetchStyle and AllocatePrefetchInstr.
236 //
237 // If the C2 compiler is not present, no space is reserved.
238
239 // +1 for rounding up to next cache line, +1 to be safe
240 if (is_server_compilation_mode_vm()) {
241 int lines = MAX2(AllocatePrefetchLines, AllocateInstancePrefetchLines) + 2;
242 _reserve_for_allocation_prefetch = (AllocatePrefetchDistance + AllocatePrefetchStepSize * lines) /
243 (int)HeapWordSize;
244 }
245 #endif
246
247 // During jvm startup, the main (primordial) thread is initialized
248 // before the heap is initialized. So reinitialize it now.
249 guarantee(Thread::current()->is_Java_thread(), "tlab initialization thread not Java thread");
250 Thread::current()->tlab().initialize();
251
252 log_develop_trace(gc, tlab)("TLAB min: " SIZE_FORMAT " initial: " SIZE_FORMAT " max: " SIZE_FORMAT,
253 min_size(), Thread::current()->tlab().initial_desired_size(), max_size());
254 }
255
256 size_t ThreadLocalAllocBuffer::initial_desired_size() {
257 size_t init_sz = 0;
258
259 if (TLABSize > 0) {
260 init_sz = TLABSize / HeapWordSize;
261 } else if (global_stats() != NULL) {
262 // Initial size is a function of the average number of allocating threads.
263 unsigned nof_threads = global_stats()->allocating_threads_avg();
264
265 init_sz = (Universe::heap()->tlab_capacity(myThread()) / HeapWordSize) /
266 (nof_threads * target_refills());
267 init_sz = align_object_size(init_sz);
268 }
269 init_sz = MIN2(MAX2(init_sz, min_size()), max_size());
270 return init_sz;
271 }
272
273 void ThreadLocalAllocBuffer::print_stats(const char* tag) {
274 Log(gc, tlab) log;
275 if (!log.is_trace()) {
276 return;
277 }
278
279 Thread* thrd = myThread();
280 size_t waste = _gc_waste + _slow_refill_waste + _fast_refill_waste;
281 size_t alloc = _number_of_refills * _desired_size;
282 double waste_percent = alloc == 0 ? 0.0 :
283 100.0 * waste / alloc;
284 size_t tlab_used = Universe::heap()->tlab_used(thrd);
285 log.trace("TLAB: %s thread: " INTPTR_FORMAT " [id: %2d]"
286 " desired_size: " SIZE_FORMAT "KB"
287 " slow allocs: %d refill waste: " SIZE_FORMAT "B"
288 " alloc:%8.5f %8.0fKB refills: %d waste %4.1f%% gc: %dB"
289 " slow: %dB fast: %dB",
290 tag, p2i(thrd), thrd->osthread()->thread_id(),
291 _desired_size / (K / HeapWordSize),
292 _slow_allocations, _refill_waste_limit * HeapWordSize,
293 _allocation_fraction.average(),
294 _allocation_fraction.average() * tlab_used / K,
295 _number_of_refills, waste_percent,
296 _gc_waste * HeapWordSize,
297 _slow_refill_waste * HeapWordSize,
298 _fast_refill_waste * HeapWordSize);
299 }
300
301 void ThreadLocalAllocBuffer::verify() {
302 HeapWord* p = start();
303 HeapWord* t = top();
304 HeapWord* prev_p = NULL;
305 while (p < t) {
306 oop(p)->verify();
307 prev_p = p;
308 p += oop(p)->size();
309 }
310 guarantee(p == top(), "end of last object must match end of space");
311 }
312
313 void ThreadLocalAllocBuffer::pick_next_sample() {
314 if (!HeapMonitoring::enabled()) {
315 return;
316 }
317
318 if (bytes_until_sample() == 0) {
319 HeapMonitoring::pick_next_sample(bytes_until_sample_addr());
320 }
321
322 // Finally, fix up the sampling bytes left and _end.
323 size_t heap_words_remaining = _end - _top;
324 size_t bytes_left = bytes_until_sample();
325 size_t words_until_sample = bytes_left / HeapWordSize;
326
327 if (heap_words_remaining > words_until_sample) {
328 set_end(_top + words_until_sample);
329 set_bytes_until_sample(0);
330 } else {
331 // TODO(jcbeyler): this is not exact actually since there can be left over
332 // space, this will need to get fixed...
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::handle_sample(Thread* thread, HeapWord* result,
345 size_t size) {
346 if (!HeapMonitoring::enabled()) {
347 return;
348 }
349
350 set_bytes_until_sample(bytes_until_sample() - size);
351
352 // Should we sample now?
353 if (should_sample()) {
354 set_back_actual_end();
355 HeapMonitoring::object_alloc_do_sample(thread,
356 reinterpret_cast<oopDesc*>(result),
357 size);
358 pick_next_sample();
359 } else {
360 // Update the size and end fields.
361 set_back_actual_end();
362 pick_next_sample();
363 }
364 }
365
366 GlobalTLABStats::GlobalTLABStats() :
367 _allocating_threads_avg(TLABAllocationWeight) {
368
369 initialize();
370
371 _allocating_threads_avg.sample(1); // One allocating thread at startup
372
373 if (UsePerfData) {
374
375 EXCEPTION_MARK;
376 ResourceMark rm;
377
378 char* cname = PerfDataManager::counter_name("tlab", "allocThreads");
379 _perf_allocating_threads =
380 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_None, CHECK);
381
382 cname = PerfDataManager::counter_name("tlab", "fills");
383 _perf_total_refills =
384 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_None, CHECK);
385
386 cname = PerfDataManager::counter_name("tlab", "maxFills");
387 _perf_max_refills =
388 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_None, CHECK);
389
390 cname = PerfDataManager::counter_name("tlab", "alloc");
391 _perf_allocation =
392 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_Bytes, CHECK);
393
394 cname = PerfDataManager::counter_name("tlab", "gcWaste");
395 _perf_gc_waste =
396 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_Bytes, CHECK);
397
398 cname = PerfDataManager::counter_name("tlab", "maxGcWaste");
399 _perf_max_gc_waste =
400 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_Bytes, CHECK);
401
402 cname = PerfDataManager::counter_name("tlab", "slowWaste");
403 _perf_slow_refill_waste =
404 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_Bytes, CHECK);
405
406 cname = PerfDataManager::counter_name("tlab", "maxSlowWaste");
407 _perf_max_slow_refill_waste =
408 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_Bytes, CHECK);
409
410 cname = PerfDataManager::counter_name("tlab", "fastWaste");
411 _perf_fast_refill_waste =
412 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_Bytes, CHECK);
413
414 cname = PerfDataManager::counter_name("tlab", "maxFastWaste");
415 _perf_max_fast_refill_waste =
416 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_Bytes, CHECK);
417
418 cname = PerfDataManager::counter_name("tlab", "slowAlloc");
419 _perf_slow_allocations =
420 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_None, CHECK);
421
422 cname = PerfDataManager::counter_name("tlab", "maxSlowAlloc");
423 _perf_max_slow_allocations =
424 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_None, CHECK);
425 }
426 }
427
428 void GlobalTLABStats::initialize() {
429 // Clear counters summarizing info from all threads
430 _allocating_threads = 0;
431 _total_refills = 0;
432 _max_refills = 0;
433 _total_allocation = 0;
434 _total_gc_waste = 0;
435 _max_gc_waste = 0;
436 _total_slow_refill_waste = 0;
437 _max_slow_refill_waste = 0;
438 _total_fast_refill_waste = 0;
439 _max_fast_refill_waste = 0;
440 _total_slow_allocations = 0;
441 _max_slow_allocations = 0;
442 }
443
444 void GlobalTLABStats::publish() {
445 _allocating_threads_avg.sample(_allocating_threads);
446 if (UsePerfData) {
447 _perf_allocating_threads ->set_value(_allocating_threads);
448 _perf_total_refills ->set_value(_total_refills);
449 _perf_max_refills ->set_value(_max_refills);
450 _perf_allocation ->set_value(_total_allocation);
451 _perf_gc_waste ->set_value(_total_gc_waste);
452 _perf_max_gc_waste ->set_value(_max_gc_waste);
453 _perf_slow_refill_waste ->set_value(_total_slow_refill_waste);
454 _perf_max_slow_refill_waste->set_value(_max_slow_refill_waste);
455 _perf_fast_refill_waste ->set_value(_total_fast_refill_waste);
456 _perf_max_fast_refill_waste->set_value(_max_fast_refill_waste);
457 _perf_slow_allocations ->set_value(_total_slow_allocations);
458 _perf_max_slow_allocations ->set_value(_max_slow_allocations);
459 }
460 }
461
462 void GlobalTLABStats::print() {
463 Log(gc, tlab) log;
464 if (!log.is_debug()) {
465 return;
466 }
467
468 size_t waste = _total_gc_waste + _total_slow_refill_waste + _total_fast_refill_waste;
469 double waste_percent = _total_allocation == 0 ? 0.0 :
470 100.0 * waste / _total_allocation;
471 log.debug("TLAB totals: thrds: %d refills: %d max: %d"
472 " slow allocs: %d max %d waste: %4.1f%%"
473 " gc: " SIZE_FORMAT "B max: " SIZE_FORMAT "B"
474 " slow: " SIZE_FORMAT "B max: " SIZE_FORMAT "B"
475 " fast: " SIZE_FORMAT "B max: " SIZE_FORMAT "B",
476 _allocating_threads,
477 _total_refills, _max_refills,
478 _total_slow_allocations, _max_slow_allocations,
479 waste_percent,
480 _total_gc_waste * HeapWordSize,
481 _max_gc_waste * HeapWordSize,
482 _total_slow_refill_waste * HeapWordSize,
483 _max_slow_refill_waste * HeapWordSize,
484 _total_fast_refill_waste * HeapWordSize,
485 _max_fast_refill_waste * HeapWordSize);
486 }