1 /*
2 * Copyright (c) 1999, 2009, 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 // Thread-Local Edens support
26
27 # include "incls/_precompiled.incl"
28 # include "incls/_threadLocalAllocBuffer.cpp.incl"
29
30 // static member initialization
31 unsigned ThreadLocalAllocBuffer::_target_refills = 0;
32 GlobalTLABStats* ThreadLocalAllocBuffer::_global_stats = NULL;
33
34 void ThreadLocalAllocBuffer::clear_before_allocation() {
35 _slow_refill_waste += (unsigned)remaining();
36 make_parsable(true); // also retire the TLAB
37 }
38
39 void ThreadLocalAllocBuffer::accumulate_statistics_before_gc() {
40 global_stats()->initialize();
41
42 for(JavaThread *thread = Threads::first(); thread; thread = thread->next()) {
43 thread->tlab().accumulate_statistics();
44 thread->tlab().initialize_statistics();
45 }
46
47 // Publish new stats if some allocation occurred.
48 if (global_stats()->allocation() != 0) {
49 global_stats()->publish();
50 if (PrintTLAB) {
51 global_stats()->print();
52 }
53 }
54 }
55
56 void ThreadLocalAllocBuffer::accumulate_statistics() {
57 size_t capacity = Universe::heap()->tlab_capacity(myThread()) / HeapWordSize;
58 size_t unused = Universe::heap()->unsafe_max_tlab_alloc(myThread()) / HeapWordSize;
59 size_t used = capacity - unused;
60
61 // Update allocation history if a reasonable amount of eden was allocated.
62 bool update_allocation_history = used > 0.5 * capacity;
63
64 _gc_waste += (unsigned)remaining();
65
66 if (PrintTLAB && (_number_of_refills > 0 || Verbose)) {
67 print_stats("gc");
68 }
69
70 if (_number_of_refills > 0) {
71
72 if (update_allocation_history) {
73 // Average the fraction of eden allocated in a tlab by this
74 // thread for use in the next resize operation.
75 // _gc_waste is not subtracted because it's included in
76 // "used".
77 size_t allocation = _number_of_refills * desired_size();
78 double alloc_frac = allocation / (double) used;
79 _allocation_fraction.sample(alloc_frac);
80 }
81 global_stats()->update_allocating_threads();
82 global_stats()->update_number_of_refills(_number_of_refills);
83 global_stats()->update_allocation(_number_of_refills * desired_size());
84 global_stats()->update_gc_waste(_gc_waste);
85 global_stats()->update_slow_refill_waste(_slow_refill_waste);
86 global_stats()->update_fast_refill_waste(_fast_refill_waste);
87
88 } else {
89 assert(_number_of_refills == 0 && _fast_refill_waste == 0 &&
90 _slow_refill_waste == 0 && _gc_waste == 0,
91 "tlab stats == 0");
92 }
93 global_stats()->update_slow_allocations(_slow_allocations);
94 }
95
96 // Fills the current tlab with a dummy filler array to create
97 // an illusion of a contiguous Eden and optionally retires the tlab.
98 // Waste accounting should be done in caller as appropriate; see,
99 // for example, clear_before_allocation().
100 void ThreadLocalAllocBuffer::make_parsable(bool retire) {
101 if (end() != NULL) {
102 invariants();
103 CollectedHeap::fill_with_object(top(), hard_end(), retire);
104
105 if (retire || ZeroTLAB) { // "Reset" the TLAB
106 set_start(NULL);
107 set_top(NULL);
108 set_pf_top(NULL);
109 set_end(NULL);
110 }
111 }
112 assert(!(retire || ZeroTLAB) ||
113 (start() == NULL && end() == NULL && top() == NULL),
114 "TLAB must be reset");
115 }
116
117 void ThreadLocalAllocBuffer::resize_all_tlabs() {
118 for(JavaThread *thread = Threads::first(); thread; thread = thread->next()) {
119 thread->tlab().resize();
120 }
121 }
122
123 void ThreadLocalAllocBuffer::resize() {
124
125 if (ResizeTLAB) {
126 // Compute the next tlab size using expected allocation amount
127 size_t alloc = (size_t)(_allocation_fraction.average() *
128 (Universe::heap()->tlab_capacity(myThread()) / HeapWordSize));
129 size_t new_size = alloc / _target_refills;
130
131 new_size = MIN2(MAX2(new_size, min_size()), max_size());
132
133 size_t aligned_new_size = align_object_size(new_size);
134
135 if (PrintTLAB && Verbose) {
136 gclog_or_tty->print("TLAB new size: thread: " INTPTR_FORMAT " [id: %2d]"
137 " refills %d alloc: %8.6f desired_size: " SIZE_FORMAT " -> " SIZE_FORMAT "\n",
138 myThread(), myThread()->osthread()->thread_id(),
139 _target_refills, _allocation_fraction.average(), desired_size(), aligned_new_size);
140 }
141 set_desired_size(aligned_new_size);
142
143 set_refill_waste_limit(initial_refill_waste_limit());
144 }
145 }
146
147 void ThreadLocalAllocBuffer::initialize_statistics() {
148 _number_of_refills = 0;
149 _fast_refill_waste = 0;
150 _slow_refill_waste = 0;
151 _gc_waste = 0;
152 _slow_allocations = 0;
153 }
154
155 void ThreadLocalAllocBuffer::fill(HeapWord* start,
156 HeapWord* top,
157 size_t new_size) {
158 _number_of_refills++;
159 if (PrintTLAB && Verbose) {
160 print_stats("fill");
161 }
162 assert(top <= start + new_size - alignment_reserve(), "size too small");
163 initialize(start, top, start + new_size - alignment_reserve());
164
165 // Reset amount of internal fragmentation
166 set_refill_waste_limit(initial_refill_waste_limit());
167 }
168
169 void ThreadLocalAllocBuffer::initialize(HeapWord* start,
170 HeapWord* top,
171 HeapWord* end) {
172 set_start(start);
173 set_top(top);
174 set_pf_top(top);
175 set_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 // Following check is needed because at startup the main (primordial)
187 // thread is initialized before the heap is. The initialization for
188 // this thread is redone in startup_initialization below.
189 if (Universe::heap() != NULL) {
190 size_t capacity = Universe::heap()->tlab_capacity(myThread()) / HeapWordSize;
191 double alloc_frac = desired_size() * target_refills() / (double) capacity;
192 _allocation_fraction.sample(alloc_frac);
193 }
194
195 set_refill_waste_limit(initial_refill_waste_limit());
196
197 initialize_statistics();
198 }
199
200 void ThreadLocalAllocBuffer::startup_initialization() {
201
202 // Assuming each thread's active tlab is, on average,
203 // 1/2 full at a GC
204 _target_refills = 100 / (2 * TLABWasteTargetPercent);
205 _target_refills = MAX2(_target_refills, (unsigned)1U);
206
207 _global_stats = new GlobalTLABStats();
208
209 // During jvm startup, the main (primordial) thread is initialized
210 // before the heap is initialized. So reinitialize it now.
211 guarantee(Thread::current()->is_Java_thread(), "tlab initialization thread not Java thread");
212 Thread::current()->tlab().initialize();
213
214 if (PrintTLAB && Verbose) {
215 gclog_or_tty->print("TLAB min: " SIZE_FORMAT " initial: " SIZE_FORMAT " max: " SIZE_FORMAT "\n",
216 min_size(), Thread::current()->tlab().initial_desired_size(), max_size());
217 }
218 }
219
220 size_t ThreadLocalAllocBuffer::initial_desired_size() {
221 size_t init_sz;
222
223 if (TLABSize > 0) {
224 init_sz = MIN2(TLABSize / HeapWordSize, max_size());
225 } else if (global_stats() == NULL) {
226 // Startup issue - main thread initialized before heap initialized.
227 init_sz = min_size();
228 } else {
229 // Initial size is a function of the average number of allocating threads.
230 unsigned nof_threads = global_stats()->allocating_threads_avg();
231
232 init_sz = (Universe::heap()->tlab_capacity(myThread()) / HeapWordSize) /
233 (nof_threads * target_refills());
234 init_sz = align_object_size(init_sz);
235 init_sz = MIN2(MAX2(init_sz, min_size()), max_size());
236 }
237 return init_sz;
238 }
239
240 const size_t ThreadLocalAllocBuffer::max_size() {
241
242 // TLABs can't be bigger than we can fill with a int[Integer.MAX_VALUE].
243 // This restriction could be removed by enabling filling with multiple arrays.
244 // If we compute that the reasonable way as
245 // header_size + ((sizeof(jint) * max_jint) / HeapWordSize)
246 // we'll overflow on the multiply, so we do the divide first.
247 // We actually lose a little by dividing first,
248 // but that just makes the TLAB somewhat smaller than the biggest array,
249 // which is fine, since we'll be able to fill that.
250
251 size_t unaligned_max_size = typeArrayOopDesc::header_size(T_INT) +
252 sizeof(jint) *
253 ((juint) max_jint / (size_t) HeapWordSize);
254 return align_size_down(unaligned_max_size, MinObjAlignment);
255 }
256
257 void ThreadLocalAllocBuffer::print_stats(const char* tag) {
258 Thread* thrd = myThread();
259 size_t waste = _gc_waste + _slow_refill_waste + _fast_refill_waste;
260 size_t alloc = _number_of_refills * _desired_size;
261 double waste_percent = alloc == 0 ? 0.0 :
262 100.0 * waste / alloc;
263 size_t tlab_used = Universe::heap()->tlab_capacity(thrd) -
264 Universe::heap()->unsafe_max_tlab_alloc(thrd);
265 gclog_or_tty->print("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\n",
270 tag, 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 oop(p)->verify();
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 Thread* ThreadLocalAllocBuffer::myThread() {
294 return (Thread*)(((char *)this) +
295 in_bytes(start_offset()) -
296 in_bytes(Thread::tlab_start_offset()));
297 }
298
299
300 GlobalTLABStats::GlobalTLABStats() :
301 _allocating_threads_avg(TLABAllocationWeight) {
302
303 initialize();
304
305 _allocating_threads_avg.sample(1); // One allocating thread at startup
306
307 if (UsePerfData) {
308
309 EXCEPTION_MARK;
310 ResourceMark rm;
311
312 char* cname = PerfDataManager::counter_name("tlab", "allocThreads");
313 _perf_allocating_threads =
314 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_None, CHECK);
315
316 cname = PerfDataManager::counter_name("tlab", "fills");
317 _perf_total_refills =
318 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_None, CHECK);
319
320 cname = PerfDataManager::counter_name("tlab", "maxFills");
321 _perf_max_refills =
322 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_None, CHECK);
323
324 cname = PerfDataManager::counter_name("tlab", "alloc");
325 _perf_allocation =
326 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_Bytes, CHECK);
327
328 cname = PerfDataManager::counter_name("tlab", "gcWaste");
329 _perf_gc_waste =
330 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_Bytes, CHECK);
331
332 cname = PerfDataManager::counter_name("tlab", "maxGcWaste");
333 _perf_max_gc_waste =
334 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_Bytes, CHECK);
335
336 cname = PerfDataManager::counter_name("tlab", "slowWaste");
337 _perf_slow_refill_waste =
338 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_Bytes, CHECK);
339
340 cname = PerfDataManager::counter_name("tlab", "maxSlowWaste");
341 _perf_max_slow_refill_waste =
342 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_Bytes, CHECK);
343
344 cname = PerfDataManager::counter_name("tlab", "fastWaste");
345 _perf_fast_refill_waste =
346 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_Bytes, CHECK);
347
348 cname = PerfDataManager::counter_name("tlab", "maxFastWaste");
349 _perf_max_fast_refill_waste =
350 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_Bytes, CHECK);
351
352 cname = PerfDataManager::counter_name("tlab", "slowAlloc");
353 _perf_slow_allocations =
354 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_None, CHECK);
355
356 cname = PerfDataManager::counter_name("tlab", "maxSlowAlloc");
357 _perf_max_slow_allocations =
358 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_None, CHECK);
359 }
360 }
361
362 void GlobalTLABStats::initialize() {
363 // Clear counters summarizing info from all threads
364 _allocating_threads = 0;
365 _total_refills = 0;
366 _max_refills = 0;
367 _total_allocation = 0;
368 _total_gc_waste = 0;
369 _max_gc_waste = 0;
370 _total_slow_refill_waste = 0;
371 _max_slow_refill_waste = 0;
372 _total_fast_refill_waste = 0;
373 _max_fast_refill_waste = 0;
374 _total_slow_allocations = 0;
375 _max_slow_allocations = 0;
376 }
377
378 void GlobalTLABStats::publish() {
379 _allocating_threads_avg.sample(_allocating_threads);
380 if (UsePerfData) {
381 _perf_allocating_threads ->set_value(_allocating_threads);
382 _perf_total_refills ->set_value(_total_refills);
383 _perf_max_refills ->set_value(_max_refills);
384 _perf_allocation ->set_value(_total_allocation);
385 _perf_gc_waste ->set_value(_total_gc_waste);
386 _perf_max_gc_waste ->set_value(_max_gc_waste);
387 _perf_slow_refill_waste ->set_value(_total_slow_refill_waste);
388 _perf_max_slow_refill_waste->set_value(_max_slow_refill_waste);
389 _perf_fast_refill_waste ->set_value(_total_fast_refill_waste);
390 _perf_max_fast_refill_waste->set_value(_max_fast_refill_waste);
391 _perf_slow_allocations ->set_value(_total_slow_allocations);
392 _perf_max_slow_allocations ->set_value(_max_slow_allocations);
393 }
394 }
395
396 void GlobalTLABStats::print() {
397 size_t waste = _total_gc_waste + _total_slow_refill_waste + _total_fast_refill_waste;
398 double waste_percent = _total_allocation == 0 ? 0.0 :
399 100.0 * waste / _total_allocation;
400 gclog_or_tty->print("TLAB totals: thrds: %d refills: %d max: %d"
401 " slow allocs: %d max %d waste: %4.1f%%"
402 " gc: " SIZE_FORMAT "B max: " SIZE_FORMAT "B"
403 " slow: " SIZE_FORMAT "B max: " SIZE_FORMAT "B"
404 " fast: " SIZE_FORMAT "B max: " SIZE_FORMAT "B\n",
405 _allocating_threads,
406 _total_refills, _max_refills,
407 _total_slow_allocations, _max_slow_allocations,
408 waste_percent,
409 _total_gc_waste * HeapWordSize,
410 _max_gc_waste * HeapWordSize,
411 _total_slow_refill_waste * HeapWordSize,
412 _max_slow_refill_waste * HeapWordSize,
413 _total_fast_refill_waste * HeapWordSize,
414 _max_fast_refill_waste * HeapWordSize);
415 }