1 /*
2 * Copyright (c) 1999, 2020, 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 // Publish new stats if some allocation occurred.
57 if (global_stats()->allocation() != 0) {
58 global_stats()->publish();
59 if (PrintTLAB) {
60 global_stats()->print();
61 }
62 }
63 }
64
65 void ThreadLocalAllocBuffer::accumulate_statistics() {
66 Thread* thread = myThread();
67 size_t capacity = Universe::heap()->tlab_capacity(thread);
68 size_t used = Universe::heap()->tlab_used(thread);
69
70 _gc_waste += (unsigned)remaining();
71 size_t total_allocated = thread->allocated_bytes();
72 size_t allocated_since_last_gc = total_allocated - _allocated_before_last_gc;
73 _allocated_before_last_gc = total_allocated;
74
75 if (PrintTLAB && (_number_of_refills > 0 || Verbose)) {
76 print_stats("gc");
77 }
78
79 if (_number_of_refills > 0) {
80 // Update allocation history if a reasonable amount of eden was allocated.
81 bool update_allocation_history = used > 0.5 * capacity;
82
83 if (update_allocation_history) {
84 // Average the fraction of eden allocated in a tlab by this
85 // thread for use in the next resize operation.
86 // _gc_waste is not subtracted because it's included in
87 // "used".
88 // The result can be larger than 1.0 due to direct to old allocations.
89 // These allocations should ideally not be counted but since it is not possible
90 // to filter them out here we just cap the fraction to be at most 1.0.
91 // Keep alloc_frac as float and not double to avoid the double to float conversion
92 float alloc_frac = MIN2(1.0f, allocated_since_last_gc / (float) used);
93 _allocation_fraction.sample(alloc_frac);
94 }
95 global_stats()->update_allocating_threads();
96 global_stats()->update_number_of_refills(_number_of_refills);
97 global_stats()->update_allocation(_number_of_refills * desired_size());
98 global_stats()->update_gc_waste(_gc_waste);
99 global_stats()->update_slow_refill_waste(_slow_refill_waste);
100 global_stats()->update_fast_refill_waste(_fast_refill_waste);
101
102 } else {
103 assert(_number_of_refills == 0 && _fast_refill_waste == 0 &&
104 _slow_refill_waste == 0 && _gc_waste == 0,
105 "tlab stats == 0");
106 }
107 global_stats()->update_slow_allocations(_slow_allocations);
108 }
109
110 // Fills the current tlab with a dummy filler array to create
111 // an illusion of a contiguous Eden and optionally retires the tlab.
112 // Waste accounting should be done in caller as appropriate; see,
113 // for example, clear_before_allocation().
114 void ThreadLocalAllocBuffer::make_parsable(bool retire) {
115 if (end() != NULL) {
116 invariants();
117
118 if (retire) {
119 myThread()->incr_allocated_bytes(used_bytes());
120 }
121
122 CollectedHeap::fill_with_object(top(), hard_end(), retire);
123
124 if (retire || ZeroTLAB) { // "Reset" the TLAB
125 set_start(NULL);
126 set_top(NULL);
127 set_pf_top(NULL);
128 set_end(NULL);
129 }
130 }
131 assert(!(retire || ZeroTLAB) ||
132 (start() == NULL && end() == NULL && top() == NULL),
133 "TLAB must be reset");
134 }
135
136 void ThreadLocalAllocBuffer::resize_all_tlabs() {
137 if (ResizeTLAB) {
138 for (JavaThread *thread = Threads::first(); thread != NULL; thread = thread->next()) {
139 thread->tlab().resize();
140 }
141 }
142 }
143
144 void ThreadLocalAllocBuffer::resize() {
145 // Compute the next tlab size using expected allocation amount
146 assert(ResizeTLAB, "Should not call this otherwise");
147 size_t alloc = (size_t)(_allocation_fraction.average() *
148 (Universe::heap()->tlab_capacity(myThread()) / HeapWordSize));
149 size_t new_size = alloc / _target_refills;
150
151 new_size = MIN2(MAX2(new_size, min_size()), max_size());
152
153 size_t aligned_new_size = align_object_size(new_size);
154
155 if (PrintTLAB && Verbose) {
156 gclog_or_tty->print("TLAB new size: thread: " INTPTR_FORMAT " [id: %2d]"
157 " refills %d alloc: %8.6f desired_size: " SIZE_FORMAT " -> " SIZE_FORMAT "\n",
158 myThread(), myThread()->osthread()->thread_id(),
159 _target_refills, _allocation_fraction.average(), desired_size(), aligned_new_size);
160 }
161 set_desired_size(aligned_new_size);
162 set_refill_waste_limit(initial_refill_waste_limit());
163 }
164
165 void ThreadLocalAllocBuffer::initialize_statistics() {
166 _number_of_refills = 0;
167 _fast_refill_waste = 0;
168 _slow_refill_waste = 0;
169 _gc_waste = 0;
170 _slow_allocations = 0;
171 }
172
173 void ThreadLocalAllocBuffer::fill(HeapWord* start,
174 HeapWord* top,
175 size_t new_size) {
176 _number_of_refills++;
177 if (PrintTLAB && Verbose) {
178 print_stats("fill");
179 }
180 assert(top <= start + new_size - alignment_reserve(), "size too small");
181 initialize(start, top, start + new_size - alignment_reserve());
182
183 // Reset amount of internal fragmentation
184 set_refill_waste_limit(initial_refill_waste_limit());
185 }
186
187 void ThreadLocalAllocBuffer::initialize(HeapWord* start,
188 HeapWord* top,
189 HeapWord* end) {
190 set_start(start);
191 set_top(top);
192 set_pf_top(top);
193 set_end(end);
194 invariants();
195 }
196
197 void ThreadLocalAllocBuffer::initialize() {
198 initialize(NULL, // start
199 NULL, // top
200 NULL); // end
201
202 set_desired_size(initial_desired_size());
203
204 // Following check is needed because at startup the main
205 // thread is initialized before the heap is. The initialization for
206 // this thread is redone in startup_initialization below.
207 if (Universe::heap() != NULL) {
208 size_t capacity = Universe::heap()->tlab_capacity(myThread()) / HeapWordSize;
209 // Keep alloc_frac as float and not double to avoid the double to float conversion
210 float alloc_frac = desired_size() * target_refills() / (float) capacity;
211 _allocation_fraction.sample(alloc_frac);
212 }
213
214 set_refill_waste_limit(initial_refill_waste_limit());
215
216 initialize_statistics();
217 }
218
219 void ThreadLocalAllocBuffer::startup_initialization() {
220
221 // Assuming each thread's active tlab is, on average,
222 // 1/2 full at a GC
223 _target_refills = 100 / (2 * TLABWasteTargetPercent);
224 _target_refills = MAX2(_target_refills, (unsigned)1U);
225
226 _global_stats = new GlobalTLABStats();
227
228 // During jvm startup, the main thread is initialized
229 // before the heap is initialized. So reinitialize it now.
230 guarantee(Thread::current()->is_Java_thread(), "tlab initialization thread not Java thread");
231 Thread::current()->tlab().initialize();
232
233 if (PrintTLAB && Verbose) {
234 gclog_or_tty->print("TLAB min: " SIZE_FORMAT " initial: " SIZE_FORMAT " max: " SIZE_FORMAT "\n",
235 min_size(), Thread::current()->tlab().initial_desired_size(), max_size());
236 }
237 }
238
239 size_t ThreadLocalAllocBuffer::initial_desired_size() {
240 size_t init_sz = 0;
241
242 if (TLABSize > 0) {
243 init_sz = TLABSize / HeapWordSize;
244 } else if (global_stats() != NULL) {
245 // Initial size is a function of the average number of allocating threads.
246 unsigned nof_threads = global_stats()->allocating_threads_avg();
247
248 init_sz = (Universe::heap()->tlab_capacity(myThread()) / HeapWordSize) /
249 (nof_threads * target_refills());
250 init_sz = align_object_size(init_sz);
251 }
252 init_sz = MIN2(MAX2(init_sz, min_size()), max_size());
253 return init_sz;
254 }
255
256 void ThreadLocalAllocBuffer::print_stats(const char* tag) {
257 Thread* thrd = myThread();
258 size_t waste = _gc_waste + _slow_refill_waste + _fast_refill_waste;
259 size_t alloc = _number_of_refills * _desired_size;
260 double waste_percent = alloc == 0 ? 0.0 :
261 100.0 * waste / alloc;
262 size_t tlab_used = Universe::heap()->tlab_used(thrd);
263 gclog_or_tty->print("TLAB: %s thread: " INTPTR_FORMAT " [id: %2d]"
264 " desired_size: " SIZE_FORMAT "KB"
265 " slow allocs: %d refill waste: " SIZE_FORMAT "B"
266 " alloc:%8.5f %8.0fKB refills: %d waste %4.1f%% gc: %dB"
267 " slow: %dB fast: %dB\n",
268 tag, thrd, thrd->osthread()->thread_id(),
269 _desired_size / (K / HeapWordSize),
270 _slow_allocations, _refill_waste_limit * HeapWordSize,
271 _allocation_fraction.average(),
272 _allocation_fraction.average() * tlab_used / K,
273 _number_of_refills, waste_percent,
274 _gc_waste * HeapWordSize,
275 _slow_refill_waste * HeapWordSize,
276 _fast_refill_waste * HeapWordSize);
277 }
278
279 void ThreadLocalAllocBuffer::verify() {
280 HeapWord* p = start();
281 HeapWord* t = top();
282 HeapWord* prev_p = NULL;
283 while (p < t) {
284 oop(p)->verify();
285 prev_p = p;
286 p += oop(p)->size();
287 }
288 guarantee(p == top(), "end of last object must match end of space");
289 }
290
291 Thread* ThreadLocalAllocBuffer::myThread() {
292 return (Thread*)(((char *)this) +
293 in_bytes(start_offset()) -
294 in_bytes(Thread::tlab_start_offset()));
295 }
296
297
298 GlobalTLABStats::GlobalTLABStats() :
299 _allocating_threads_avg(TLABAllocationWeight) {
300
301 initialize();
302
303 _allocating_threads_avg.sample(1); // One allocating thread at startup
304
305 if (UsePerfData) {
306
307 EXCEPTION_MARK;
308 ResourceMark rm;
309
310 char* cname = PerfDataManager::counter_name("tlab", "allocThreads");
311 _perf_allocating_threads =
312 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_None, CHECK);
313
314 cname = PerfDataManager::counter_name("tlab", "fills");
315 _perf_total_refills =
316 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_None, CHECK);
317
318 cname = PerfDataManager::counter_name("tlab", "maxFills");
319 _perf_max_refills =
320 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_None, CHECK);
321
322 cname = PerfDataManager::counter_name("tlab", "alloc");
323 _perf_allocation =
324 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_Bytes, CHECK);
325
326 cname = PerfDataManager::counter_name("tlab", "gcWaste");
327 _perf_gc_waste =
328 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_Bytes, CHECK);
329
330 cname = PerfDataManager::counter_name("tlab", "maxGcWaste");
331 _perf_max_gc_waste =
332 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_Bytes, CHECK);
333
334 cname = PerfDataManager::counter_name("tlab", "slowWaste");
335 _perf_slow_refill_waste =
336 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_Bytes, CHECK);
337
338 cname = PerfDataManager::counter_name("tlab", "maxSlowWaste");
339 _perf_max_slow_refill_waste =
340 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_Bytes, CHECK);
341
342 cname = PerfDataManager::counter_name("tlab", "fastWaste");
343 _perf_fast_refill_waste =
344 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_Bytes, CHECK);
345
346 cname = PerfDataManager::counter_name("tlab", "maxFastWaste");
347 _perf_max_fast_refill_waste =
348 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_Bytes, CHECK);
349
350 cname = PerfDataManager::counter_name("tlab", "slowAlloc");
351 _perf_slow_allocations =
352 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_None, CHECK);
353
354 cname = PerfDataManager::counter_name("tlab", "maxSlowAlloc");
355 _perf_max_slow_allocations =
356 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_None, CHECK);
357 }
358 }
359
360 void GlobalTLABStats::initialize() {
361 // Clear counters summarizing info from all threads
362 _allocating_threads = 0;
363 _total_refills = 0;
364 _max_refills = 0;
365 _total_allocation = 0;
366 _total_gc_waste = 0;
367 _max_gc_waste = 0;
368 _total_slow_refill_waste = 0;
369 _max_slow_refill_waste = 0;
370 _total_fast_refill_waste = 0;
371 _max_fast_refill_waste = 0;
372 _total_slow_allocations = 0;
373 _max_slow_allocations = 0;
374 }
375
376 void GlobalTLABStats::publish() {
377 _allocating_threads_avg.sample(_allocating_threads);
378 if (UsePerfData) {
379 _perf_allocating_threads ->set_value(_allocating_threads);
380 _perf_total_refills ->set_value(_total_refills);
381 _perf_max_refills ->set_value(_max_refills);
382 _perf_allocation ->set_value(_total_allocation);
383 _perf_gc_waste ->set_value(_total_gc_waste);
384 _perf_max_gc_waste ->set_value(_max_gc_waste);
385 _perf_slow_refill_waste ->set_value(_total_slow_refill_waste);
386 _perf_max_slow_refill_waste->set_value(_max_slow_refill_waste);
387 _perf_fast_refill_waste ->set_value(_total_fast_refill_waste);
388 _perf_max_fast_refill_waste->set_value(_max_fast_refill_waste);
389 _perf_slow_allocations ->set_value(_total_slow_allocations);
390 _perf_max_slow_allocations ->set_value(_max_slow_allocations);
391 }
392 }
393
394 void GlobalTLABStats::print() {
395 size_t waste = _total_gc_waste + _total_slow_refill_waste + _total_fast_refill_waste;
396 double waste_percent = _total_allocation == 0 ? 0.0 :
397 100.0 * waste / _total_allocation;
398 gclog_or_tty->print("TLAB totals: thrds: %d refills: %d max: %d"
399 " slow allocs: %d max %d waste: %4.1f%%"
400 " gc: " SIZE_FORMAT "B max: " SIZE_FORMAT "B"
401 " slow: " SIZE_FORMAT "B max: " SIZE_FORMAT "B"
402 " fast: " SIZE_FORMAT "B max: " SIZE_FORMAT "B\n",
403 _allocating_threads,
404 _total_refills, _max_refills,
405 _total_slow_allocations, _max_slow_allocations,
406 waste_percent,
407 _total_gc_waste * HeapWordSize,
408 _max_gc_waste * HeapWordSize,
409 _total_slow_refill_waste * HeapWordSize,
410 _max_slow_refill_waste * HeapWordSize,
411 _total_fast_refill_waste * HeapWordSize,
412 _max_fast_refill_waste * HeapWordSize);
413 }