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