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