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 "memory/genCollectedHeap.hpp"
27 #include "memory/resourceArea.hpp"
28 #include "memory/threadLocalAllocBuffer.inline.hpp"
29 #include "memory/universe.inline.hpp"
30 #include "oops/oop.inline.hpp"
31 #include "runtime/thread.inline.hpp"
32 #include "utilities/copy.hpp"
33
34 PRAGMA_FORMAT_MUTE_WARNINGS_FOR_GCC
35
36 // Thread-Local Edens support
37
38 // static member initialization
39 size_t ThreadLocalAllocBuffer::_max_size = 0;
40 unsigned ThreadLocalAllocBuffer::_target_refills = 0;
41 GlobalTLABStats* ThreadLocalAllocBuffer::_global_stats = NULL;
42
43 void ThreadLocalAllocBuffer::clear_before_allocation() {
44 _slow_refill_waste += (unsigned)remaining();
45 make_parsable(true); // also retire the TLAB
46 }
47
48 void ThreadLocalAllocBuffer::accumulate_statistics_before_gc() {
49 global_stats()->initialize();
50
51 for (JavaThread *thread = Threads::first(); thread != NULL; thread = thread->next()) {
52 thread->tlab().accumulate_statistics();
53 thread->tlab().initialize_statistics();
54 }
55
56 Universe::heap()->accumulate_statistics_all_gclabs();
57
58 // Publish new stats if some allocation occurred.
59 if (global_stats()->allocation() != 0) {
60 global_stats()->publish();
61 if (PrintTLAB) {
62 global_stats()->print();
63 }
64 }
65 }
66
67 void ThreadLocalAllocBuffer::accumulate_statistics() {
68 Thread* thread = myThread();
69 size_t capacity = Universe::heap()->tlab_capacity(thread);
70 size_t used = Universe::heap()->tlab_used(thread);
71
72 _gc_waste += (unsigned)remaining();
73 size_t total_allocated = _gclab ? thread->allocated_bytes_gclab() : thread->allocated_bytes();
74 size_t allocated_since_last_gc = total_allocated - _allocated_before_last_gc;
75 _allocated_before_last_gc = total_allocated;
76
77 if (PrintTLAB && (_number_of_refills > 0 || Verbose)) {
78 print_stats("gc");
79 }
80
81 if (_number_of_refills > 0) {
82 // Update allocation history if a reasonable amount of eden was allocated.
83 bool update_allocation_history = used > 0.5 * capacity;
84
85 if (update_allocation_history) {
86 // Average the fraction of eden allocated in a tlab by this
87 // thread for use in the next resize operation.
88 // _gc_waste is not subtracted because it's included in
89 // "used".
90 // The result can be larger than 1.0 due to direct to old allocations.
91 // These allocations should ideally not be counted but since it is not possible
92 // to filter them out here we just cap the fraction to be at most 1.0.
93 double alloc_frac = MIN2(1.0, (double) allocated_since_last_gc / used);
94 _allocation_fraction.sample(alloc_frac);
95 }
96 global_stats()->update_allocating_threads();
97 global_stats()->update_number_of_refills(_number_of_refills);
98 global_stats()->update_allocation(_number_of_refills * desired_size());
99 global_stats()->update_gc_waste(_gc_waste);
100 global_stats()->update_slow_refill_waste(_slow_refill_waste);
101 global_stats()->update_fast_refill_waste(_fast_refill_waste);
102
103 } else {
104 assert(_number_of_refills == 0 && _fast_refill_waste == 0 &&
105 _slow_refill_waste == 0 && _gc_waste == 0,
106 "tlab stats == 0");
107 }
108 global_stats()->update_slow_allocations(_slow_allocations);
109 }
110
111 // Fills the current tlab with a dummy filler array to create
112 // an illusion of a contiguous Eden and optionally retires the tlab.
113 // Waste accounting should be done in caller as appropriate; see,
114 // for example, clear_before_allocation().
115 void ThreadLocalAllocBuffer::make_parsable(bool retire) {
116 if (end() != NULL) {
117 invariants();
118
119 if (retire) {
120 if (_gclab) {
121 myThread()->incr_allocated_bytes_gclab(used_bytes());
122 } else {
123 myThread()->incr_allocated_bytes(used_bytes());
124 }
125 }
126
127 CollectedHeap::fill_with_object(top(), hard_end(), retire);
128
129 if (retire || ZeroTLAB) { // "Reset" the TLAB
130 set_start(NULL);
131 set_top(NULL);
132 set_pf_top(NULL);
133 set_end(NULL);
134 }
135 }
136 assert(!(retire || ZeroTLAB) ||
137 (start() == NULL && end() == NULL && top() == NULL),
138 "TLAB must be reset");
139 }
140
141 void ThreadLocalAllocBuffer::resize_all_tlabs() {
142 if (ResizeTLAB) {
143 for (JavaThread *thread = Threads::first(); thread != NULL; thread = thread->next()) {
144 thread->tlab().resize();
145 }
146 }
147 }
148
149 void ThreadLocalAllocBuffer::resize() {
150 // Compute the next tlab size using expected allocation amount
151 assert(ResizeTLAB, "Should not call this otherwise");
152 size_t alloc = (size_t)(_allocation_fraction.average() *
153 (Universe::heap()->tlab_capacity(myThread()) / HeapWordSize));
154 size_t new_size = alloc / _target_refills;
155
156 new_size = MIN2(MAX2(new_size, min_size()), max_size());
157
158 size_t aligned_new_size = align_object_size(new_size);
159
160 if (PrintTLAB && Verbose) {
161 gclog_or_tty->print("TLAB new size: thread: " INTPTR_FORMAT " [id: %2d]"
162 " refills %d alloc: %8.6f desired_size: " SIZE_FORMAT " -> " SIZE_FORMAT "\n",
163 myThread(), myThread()->osthread()->thread_id(),
164 _target_refills, _allocation_fraction.average(), desired_size(), aligned_new_size);
165 }
166 set_desired_size(aligned_new_size);
167 set_refill_waste_limit(initial_refill_waste_limit());
168 }
169
170 void ThreadLocalAllocBuffer::initialize_statistics() {
171 _number_of_refills = 0;
172 _fast_refill_waste = 0;
173 _slow_refill_waste = 0;
174 _gc_waste = 0;
175 _slow_allocations = 0;
176 }
177
178 void ThreadLocalAllocBuffer::fill(HeapWord* start,
179 HeapWord* top,
180 size_t new_size) {
181 _number_of_refills++;
182 if (PrintTLAB && Verbose) {
183 print_stats("fill");
184 }
185 assert(top <= start + new_size - alignment_reserve(), "size too small");
186 initialize(start, top, start + new_size - alignment_reserve());
187
188 // Reset amount of internal fragmentation
189 set_refill_waste_limit(initial_refill_waste_limit());
190 }
191
192 void ThreadLocalAllocBuffer::initialize(HeapWord* start,
193 HeapWord* top,
194 HeapWord* end) {
195 set_start(start);
196 set_top(top);
197 set_pf_top(top);
198 set_end(end);
199 invariants();
200 }
201
202 void ThreadLocalAllocBuffer::initialize(bool gclab) {
203 _initialized = true;
204 _gclab = gclab;
205 initialize(NULL, // start
206 NULL, // top
207 NULL); // end
208
209 set_desired_size(initial_desired_size());
210
211 // Following check is needed because at startup the main
212 // thread is initialized before the heap is. The initialization for
213 // this thread is redone in startup_initialization below.
214 if (Universe::heap() != NULL) {
215 size_t capacity = Universe::heap()->tlab_capacity(myThread()) / HeapWordSize;
216 double alloc_frac = desired_size() * target_refills() / (double) capacity;
217 _allocation_fraction.sample(alloc_frac);
218 }
219
220 set_refill_waste_limit(initial_refill_waste_limit());
221
222 initialize_statistics();
223 }
224
225 void ThreadLocalAllocBuffer::startup_initialization() {
226
227 // Assuming each thread's active tlab is, on average,
228 // 1/2 full at a GC
229 _target_refills = 100 / (2 * TLABWasteTargetPercent);
230 _target_refills = MAX2(_target_refills, (unsigned)1U);
231
232 _global_stats = new GlobalTLABStats();
233
234 // During jvm startup, the main thread is initialized
235 // before the heap is initialized. So reinitialize it now.
236 guarantee(Thread::current()->is_Java_thread(), "tlab initialization thread not Java thread");
237 Thread::current()->tlab().initialize(false);
238 if (UseShenandoahGC) {
239 Thread::current()->gclab().initialize(true);
240 }
241
242 if (PrintTLAB && Verbose) {
243 gclog_or_tty->print("TLAB min: " SIZE_FORMAT " initial: " SIZE_FORMAT " max: " SIZE_FORMAT "\n",
244 min_size(), Thread::current()->tlab().initial_desired_size(), max_size());
245 }
246 }
247
248 size_t ThreadLocalAllocBuffer::initial_desired_size() {
249 size_t init_sz = 0;
250
251 if (TLABSize > 0) {
252 init_sz = TLABSize / HeapWordSize;
253 } else if (global_stats() != NULL) {
254 // Initial size is a function of the average number of allocating threads.
255 unsigned nof_threads = global_stats()->allocating_threads_avg();
256
257 init_sz = (Universe::heap()->tlab_capacity(myThread()) / HeapWordSize) /
258 (nof_threads * target_refills());
259 init_sz = align_object_size(init_sz);
260 }
261 init_sz = MIN2(MAX2(init_sz, min_size()), max_size());
262 return init_sz;
263 }
264
265 void ThreadLocalAllocBuffer::print_stats(const char* tag) {
266 Thread* thrd = myThread();
267 size_t waste = _gc_waste + _slow_refill_waste + _fast_refill_waste;
268 size_t alloc = _number_of_refills * _desired_size;
269 double waste_percent = alloc == 0 ? 0.0 :
270 100.0 * waste / alloc;
271 size_t tlab_used = Universe::heap()->tlab_used(thrd);
272 gclog_or_tty->print("TLAB: %s %s thread: " INTPTR_FORMAT " [id: %2d]"
273 " desired_size: " SIZE_FORMAT "KB"
274 " slow allocs: %d refill waste: " SIZE_FORMAT "B"
275 " alloc:%8.5f %8.0fKB refills: %d waste %4.1f%% gc: %dB"
276 " slow: %dB fast: %dB\n",
277 tag, _gclab ? "gclab" : "tlab ", p2i(thrd), thrd->osthread()->thread_id(),
278 _desired_size / (K / HeapWordSize),
279 _slow_allocations, _refill_waste_limit * HeapWordSize,
280 _allocation_fraction.average(),
281 _allocation_fraction.average() * tlab_used / K,
282 _number_of_refills, waste_percent,
283 _gc_waste * HeapWordSize,
284 _slow_refill_waste * HeapWordSize,
285 _fast_refill_waste * HeapWordSize);
286 }
287
288 void ThreadLocalAllocBuffer::verify() {
289 HeapWord* p = start();
290 HeapWord* t = top();
291 HeapWord* prev_p = NULL;
292 while (p < t) {
293 oop(p)->verify();
294 prev_p = p;
295 p += oop(p)->size();
296 }
297 guarantee(p == top(), "end of last object must match end of space");
298 }
299
300 Thread* ThreadLocalAllocBuffer::myThread() {
301 ByteSize gclab_offset = Thread::gclab_start_offset();
302 ByteSize tlab_offset = Thread::tlab_start_offset();
303 ByteSize offs = _gclab ? gclab_offset : tlab_offset;
304 Thread* thread = (Thread*)(((char *)this) +
305 in_bytes(start_offset()) - in_bytes(offs));
306 #ifdef ASSERT
307 assert(this == (_gclab ? &thread->gclab() : &thread->tlab()), "must be");
308 #endif
309 return thread;
310 }
311
312 void ThreadLocalAllocBuffer::rollback(size_t size) {
313 HeapWord* old_top = top();
314 if (old_top != NULL) { // Pathological case: we accept that we can't rollback.
315 set_top(old_top - size);
316 }
317 }
318
319
320 GlobalTLABStats::GlobalTLABStats() :
321 _allocating_threads_avg(TLABAllocationWeight) {
322
323 initialize();
324
325 _allocating_threads_avg.sample(1); // One allocating thread at startup
326
327 if (UsePerfData) {
328
329 EXCEPTION_MARK;
330 ResourceMark rm;
331
332 char* cname = PerfDataManager::counter_name("tlab", "allocThreads");
333 _perf_allocating_threads =
334 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_None, CHECK);
335
336 cname = PerfDataManager::counter_name("tlab", "fills");
337 _perf_total_refills =
338 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_None, CHECK);
339
340 cname = PerfDataManager::counter_name("tlab", "maxFills");
341 _perf_max_refills =
342 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_None, CHECK);
343
344 cname = PerfDataManager::counter_name("tlab", "alloc");
345 _perf_allocation =
346 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_Bytes, CHECK);
347
348 cname = PerfDataManager::counter_name("tlab", "gcWaste");
349 _perf_gc_waste =
350 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_Bytes, CHECK);
351
352 cname = PerfDataManager::counter_name("tlab", "maxGcWaste");
353 _perf_max_gc_waste =
354 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_Bytes, CHECK);
355
356 cname = PerfDataManager::counter_name("tlab", "slowWaste");
357 _perf_slow_refill_waste =
358 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_Bytes, CHECK);
359
360 cname = PerfDataManager::counter_name("tlab", "maxSlowWaste");
361 _perf_max_slow_refill_waste =
362 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_Bytes, CHECK);
363
364 cname = PerfDataManager::counter_name("tlab", "fastWaste");
365 _perf_fast_refill_waste =
366 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_Bytes, CHECK);
367
368 cname = PerfDataManager::counter_name("tlab", "maxFastWaste");
369 _perf_max_fast_refill_waste =
370 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_Bytes, CHECK);
371
372 cname = PerfDataManager::counter_name("tlab", "slowAlloc");
373 _perf_slow_allocations =
374 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_None, CHECK);
375
376 cname = PerfDataManager::counter_name("tlab", "maxSlowAlloc");
377 _perf_max_slow_allocations =
378 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_None, CHECK);
379 }
380 }
381
382 void GlobalTLABStats::initialize() {
383 // Clear counters summarizing info from all threads
384 _allocating_threads = 0;
385 _total_refills = 0;
386 _max_refills = 0;
387 _total_allocation = 0;
388 _total_gc_waste = 0;
389 _max_gc_waste = 0;
390 _total_slow_refill_waste = 0;
391 _max_slow_refill_waste = 0;
392 _total_fast_refill_waste = 0;
393 _max_fast_refill_waste = 0;
394 _total_slow_allocations = 0;
395 _max_slow_allocations = 0;
396 }
397
398 void GlobalTLABStats::publish() {
399 _allocating_threads_avg.sample(_allocating_threads);
400 if (UsePerfData) {
401 _perf_allocating_threads ->set_value(_allocating_threads);
402 _perf_total_refills ->set_value(_total_refills);
403 _perf_max_refills ->set_value(_max_refills);
404 _perf_allocation ->set_value(_total_allocation);
405 _perf_gc_waste ->set_value(_total_gc_waste);
406 _perf_max_gc_waste ->set_value(_max_gc_waste);
407 _perf_slow_refill_waste ->set_value(_total_slow_refill_waste);
408 _perf_max_slow_refill_waste->set_value(_max_slow_refill_waste);
409 _perf_fast_refill_waste ->set_value(_total_fast_refill_waste);
410 _perf_max_fast_refill_waste->set_value(_max_fast_refill_waste);
411 _perf_slow_allocations ->set_value(_total_slow_allocations);
412 _perf_max_slow_allocations ->set_value(_max_slow_allocations);
413 }
414 }
415
416 void GlobalTLABStats::print() {
417 size_t waste = _total_gc_waste + _total_slow_refill_waste + _total_fast_refill_waste;
418 double waste_percent = _total_allocation == 0 ? 0.0 :
419 100.0 * waste / _total_allocation;
420 gclog_or_tty->print("TLAB totals: thrds: %d refills: %d max: %d"
421 " slow allocs: %d max %d waste: %4.1f%%"
422 " gc: " SIZE_FORMAT "B max: " SIZE_FORMAT "B"
423 " slow: " SIZE_FORMAT "B max: " SIZE_FORMAT "B"
424 " fast: " SIZE_FORMAT "B max: " SIZE_FORMAT "B\n",
425 _allocating_threads,
426 _total_refills, _max_refills,
427 _total_slow_allocations, _max_slow_allocations,
428 waste_percent,
429 _total_gc_waste * HeapWordSize,
430 _max_gc_waste * HeapWordSize,
431 _total_slow_refill_waste * HeapWordSize,
432 _max_slow_refill_waste * HeapWordSize,
433 _total_fast_refill_waste * HeapWordSize,
434 _max_fast_refill_waste * HeapWordSize);
435 }