1 /*
2 * Copyright (c) 1999, 2010, 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 "utilities/copy.hpp"
32 #ifdef TARGET_OS_FAMILY_linux
33 # include "thread_linux.inline.hpp"
34 #endif
35 #ifdef TARGET_OS_FAMILY_solaris
36 # include "thread_solaris.inline.hpp"
37 #endif
38 #ifdef TARGET_OS_FAMILY_windows
39 # include "thread_windows.inline.hpp"
40 #endif
41
42 // Thread-Local Edens support
43
44 // static member initialization
45 unsigned ThreadLocalAllocBuffer::_target_refills = 0;
46 GlobalTLABStats* ThreadLocalAllocBuffer::_global_stats = NULL;
47
48 void ThreadLocalAllocBuffer::clear_before_allocation() {
49 _slow_refill_waste += (unsigned)remaining();
50 make_parsable(true); // also retire the TLAB
51 }
52
53 void ThreadLocalAllocBuffer::accumulate_statistics_before_gc() {
54 global_stats()->initialize();
55
56 for(JavaThread *thread = Threads::first(); thread; thread = thread->next()) {
57 thread->tlab().accumulate_statistics();
58 thread->tlab().initialize_statistics();
59 }
60
61 // Publish new stats if some allocation occurred.
62 if (global_stats()->allocation() != 0) {
63 global_stats()->publish();
64 if (PrintTLAB) {
65 global_stats()->print();
66 }
67 }
68 }
69
70 void ThreadLocalAllocBuffer::accumulate_statistics() {
71 size_t capacity = Universe::heap()->tlab_capacity(myThread()) / HeapWordSize;
72 size_t unused = Universe::heap()->unsafe_max_tlab_alloc(myThread()) / HeapWordSize;
73 size_t used = capacity - unused;
74
75 // Update allocation history if a reasonable amount of eden was allocated.
76 bool update_allocation_history = used > 0.5 * capacity;
77
78 _gc_waste += (unsigned)remaining();
79
80 if (PrintTLAB && (_number_of_refills > 0 || Verbose)) {
81 print_stats("gc");
82 }
83
84 if (_number_of_refills > 0) {
85
86 if (update_allocation_history) {
87 // Average the fraction of eden allocated in a tlab by this
88 // thread for use in the next resize operation.
89 // _gc_waste is not subtracted because it's included in
90 // "used".
91 size_t allocation = _number_of_refills * desired_size();
92 double alloc_frac = allocation / (double) 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 CollectedHeap::fill_with_object(top(), hard_end(), retire);
118
119 if (retire || ZeroTLAB) { // "Reset" the TLAB
120 set_start(NULL);
121 set_top(NULL);
122 set_pf_top(NULL);
123 set_end(NULL);
124 }
125 }
126 assert(!(retire || ZeroTLAB) ||
127 (start() == NULL && end() == NULL && top() == NULL),
128 "TLAB must be reset");
129 }
130
131 void ThreadLocalAllocBuffer::resize_all_tlabs() {
132 for(JavaThread *thread = Threads::first(); thread; thread = thread->next()) {
133 thread->tlab().resize();
134 }
135 }
136
137 void ThreadLocalAllocBuffer::resize() {
138
139 if (ResizeTLAB) {
140 // Compute the next tlab size using expected allocation amount
141 size_t alloc = (size_t)(_allocation_fraction.average() *
142 (Universe::heap()->tlab_capacity(myThread()) / HeapWordSize));
143 size_t new_size = alloc / _target_refills;
144
145 new_size = MIN2(MAX2(new_size, min_size()), max_size());
146
147 size_t aligned_new_size = align_object_size(new_size);
148
149 if (PrintTLAB && Verbose) {
150 gclog_or_tty->print("TLAB new size: thread: " INTPTR_FORMAT " [id: %2d]"
151 " refills %d alloc: %8.6f desired_size: " SIZE_FORMAT " -> " SIZE_FORMAT "\n",
152 myThread(), myThread()->osthread()->thread_id(),
153 _target_refills, _allocation_fraction.average(), desired_size(), aligned_new_size);
154 }
155 set_desired_size(aligned_new_size);
156
157 set_refill_waste_limit(initial_refill_waste_limit());
158 }
159 }
160
161 void ThreadLocalAllocBuffer::initialize_statistics() {
162 _number_of_refills = 0;
163 _fast_refill_waste = 0;
164 _slow_refill_waste = 0;
165 _gc_waste = 0;
166 _slow_allocations = 0;
167 }
168
169 void ThreadLocalAllocBuffer::fill(HeapWord* start,
170 HeapWord* top,
171 size_t new_size) {
172 _number_of_refills++;
173 if (PrintTLAB && Verbose) {
174 print_stats("fill");
175 }
176 assert(top <= start + new_size - alignment_reserve(), "size too small");
177 initialize(start, top, start + new_size - alignment_reserve());
178
179 // Reset amount of internal fragmentation
180 set_refill_waste_limit(initial_refill_waste_limit());
181 }
182
183 void ThreadLocalAllocBuffer::initialize(HeapWord* start,
184 HeapWord* top,
185 HeapWord* end) {
186 set_start(start);
187 set_top(top);
188 set_pf_top(top);
189 set_end(end);
190 invariants();
191 }
192
193 void ThreadLocalAllocBuffer::initialize() {
194 initialize(NULL, // start
195 NULL, // top
196 NULL); // end
197
198 set_desired_size(initial_desired_size());
199
200 // Following check is needed because at startup the main (primordial)
201 // thread is initialized before the heap is. The initialization for
202 // this thread is redone in startup_initialization below.
203 if (Universe::heap() != NULL) {
204 size_t capacity = Universe::heap()->tlab_capacity(myThread()) / HeapWordSize;
205 double alloc_frac = desired_size() * target_refills() / (double) capacity;
206 _allocation_fraction.sample(alloc_frac);
207 }
208
209 set_refill_waste_limit(initial_refill_waste_limit());
210
211 initialize_statistics();
212 }
213
214 void ThreadLocalAllocBuffer::startup_initialization() {
215
216 // Assuming each thread's active tlab is, on average,
217 // 1/2 full at a GC
218 _target_refills = 100 / (2 * TLABWasteTargetPercent);
219 _target_refills = MAX2(_target_refills, (unsigned)1U);
220
221 _global_stats = new GlobalTLABStats();
222
223 // During jvm startup, the main (primordial) thread is initialized
224 // before the heap is initialized. So reinitialize it now.
225 guarantee(Thread::current()->is_Java_thread(), "tlab initialization thread not Java thread");
226 Thread::current()->tlab().initialize();
227
228 if (PrintTLAB && Verbose) {
229 gclog_or_tty->print("TLAB min: " SIZE_FORMAT " initial: " SIZE_FORMAT " max: " SIZE_FORMAT "\n",
230 min_size(), Thread::current()->tlab().initial_desired_size(), max_size());
231 }
232 }
233
234 size_t ThreadLocalAllocBuffer::initial_desired_size() {
235 size_t init_sz;
236
237 if (TLABSize > 0) {
238 init_sz = MIN2(TLABSize / HeapWordSize, max_size());
239 } else if (global_stats() == NULL) {
240 // Startup issue - main thread initialized before heap initialized.
241 init_sz = min_size();
242 } else {
243 // Initial size is a function of the average number of allocating threads.
244 unsigned nof_threads = global_stats()->allocating_threads_avg();
245
246 init_sz = (Universe::heap()->tlab_capacity(myThread()) / HeapWordSize) /
247 (nof_threads * target_refills());
248 init_sz = align_object_size(init_sz);
249 init_sz = MIN2(MAX2(init_sz, min_size()), max_size());
250 }
251 return init_sz;
252 }
253
254 const size_t ThreadLocalAllocBuffer::max_size() {
255
256 // TLABs can't be bigger than we can fill with a int[Integer.MAX_VALUE].
257 // This restriction could be removed by enabling filling with multiple arrays.
258 // If we compute that the reasonable way as
259 // header_size + ((sizeof(jint) * max_jint) / HeapWordSize)
260 // we'll overflow on the multiply, so we do the divide first.
261 // We actually lose a little by dividing first,
262 // but that just makes the TLAB somewhat smaller than the biggest array,
263 // which is fine, since we'll be able to fill that.
264
265 size_t unaligned_max_size = typeArrayOopDesc::header_size(T_INT) +
266 sizeof(jint) *
267 ((juint) max_jint / (size_t) HeapWordSize);
268 return align_size_down(unaligned_max_size, MinObjAlignment);
269 }
270
271 void ThreadLocalAllocBuffer::print_stats(const char* tag) {
272 Thread* thrd = myThread();
273 size_t waste = _gc_waste + _slow_refill_waste + _fast_refill_waste;
274 size_t alloc = _number_of_refills * _desired_size;
275 double waste_percent = alloc == 0 ? 0.0 :
276 100.0 * waste / alloc;
277 size_t tlab_used = Universe::heap()->tlab_capacity(thrd) -
278 Universe::heap()->unsafe_max_tlab_alloc(thrd);
279 gclog_or_tty->print("TLAB: %s thread: " INTPTR_FORMAT " [id: %2d]"
280 " desired_size: " SIZE_FORMAT "KB"
281 " slow allocs: %d refill waste: " SIZE_FORMAT "B"
282 " alloc:%8.5f %8.0fKB refills: %d waste %4.1f%% gc: %dB"
283 " slow: %dB fast: %dB\n",
284 tag, thrd, thrd->osthread()->thread_id(),
285 _desired_size / (K / HeapWordSize),
286 _slow_allocations, _refill_waste_limit * HeapWordSize,
287 _allocation_fraction.average(),
288 _allocation_fraction.average() * tlab_used / K,
289 _number_of_refills, waste_percent,
290 _gc_waste * HeapWordSize,
291 _slow_refill_waste * HeapWordSize,
292 _fast_refill_waste * HeapWordSize);
293 }
294
295 void ThreadLocalAllocBuffer::verify() {
296 HeapWord* p = start();
297 HeapWord* t = top();
298 HeapWord* prev_p = NULL;
299 while (p < t) {
300 oop(p)->verify();
301 prev_p = p;
302 p += oop(p)->size();
303 }
304 guarantee(p == top(), "end of last object must match end of space");
305 }
306
307 Thread* ThreadLocalAllocBuffer::myThread() {
308 return (Thread*)(((char *)this) +
309 in_bytes(start_offset()) -
310 in_bytes(Thread::tlab_start_offset()));
311 }
312
313
314 GlobalTLABStats::GlobalTLABStats() :
315 _allocating_threads_avg(TLABAllocationWeight) {
316
317 initialize();
318
319 _allocating_threads_avg.sample(1); // One allocating thread at startup
320
321 if (UsePerfData) {
322
323 EXCEPTION_MARK;
324 ResourceMark rm;
325
326 char* cname = PerfDataManager::counter_name("tlab", "allocThreads");
327 _perf_allocating_threads =
328 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_None, CHECK);
329
330 cname = PerfDataManager::counter_name("tlab", "fills");
331 _perf_total_refills =
332 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_None, CHECK);
333
334 cname = PerfDataManager::counter_name("tlab", "maxFills");
335 _perf_max_refills =
336 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_None, CHECK);
337
338 cname = PerfDataManager::counter_name("tlab", "alloc");
339 _perf_allocation =
340 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_Bytes, CHECK);
341
342 cname = PerfDataManager::counter_name("tlab", "gcWaste");
343 _perf_gc_waste =
344 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_Bytes, CHECK);
345
346 cname = PerfDataManager::counter_name("tlab", "maxGcWaste");
347 _perf_max_gc_waste =
348 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_Bytes, CHECK);
349
350 cname = PerfDataManager::counter_name("tlab", "slowWaste");
351 _perf_slow_refill_waste =
352 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_Bytes, CHECK);
353
354 cname = PerfDataManager::counter_name("tlab", "maxSlowWaste");
355 _perf_max_slow_refill_waste =
356 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_Bytes, CHECK);
357
358 cname = PerfDataManager::counter_name("tlab", "fastWaste");
359 _perf_fast_refill_waste =
360 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_Bytes, CHECK);
361
362 cname = PerfDataManager::counter_name("tlab", "maxFastWaste");
363 _perf_max_fast_refill_waste =
364 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_Bytes, CHECK);
365
366 cname = PerfDataManager::counter_name("tlab", "slowAlloc");
367 _perf_slow_allocations =
368 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_None, CHECK);
369
370 cname = PerfDataManager::counter_name("tlab", "maxSlowAlloc");
371 _perf_max_slow_allocations =
372 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_None, CHECK);
373 }
374 }
375
376 void GlobalTLABStats::initialize() {
377 // Clear counters summarizing info from all threads
378 _allocating_threads = 0;
379 _total_refills = 0;
380 _max_refills = 0;
381 _total_allocation = 0;
382 _total_gc_waste = 0;
383 _max_gc_waste = 0;
384 _total_slow_refill_waste = 0;
385 _max_slow_refill_waste = 0;
386 _total_fast_refill_waste = 0;
387 _max_fast_refill_waste = 0;
388 _total_slow_allocations = 0;
389 _max_slow_allocations = 0;
390 }
391
392 void GlobalTLABStats::publish() {
393 _allocating_threads_avg.sample(_allocating_threads);
394 if (UsePerfData) {
395 _perf_allocating_threads ->set_value(_allocating_threads);
396 _perf_total_refills ->set_value(_total_refills);
397 _perf_max_refills ->set_value(_max_refills);
398 _perf_allocation ->set_value(_total_allocation);
399 _perf_gc_waste ->set_value(_total_gc_waste);
400 _perf_max_gc_waste ->set_value(_max_gc_waste);
401 _perf_slow_refill_waste ->set_value(_total_slow_refill_waste);
402 _perf_max_slow_refill_waste->set_value(_max_slow_refill_waste);
403 _perf_fast_refill_waste ->set_value(_total_fast_refill_waste);
404 _perf_max_fast_refill_waste->set_value(_max_fast_refill_waste);
405 _perf_slow_allocations ->set_value(_total_slow_allocations);
406 _perf_max_slow_allocations ->set_value(_max_slow_allocations);
407 }
408 }
409
410 void GlobalTLABStats::print() {
411 size_t waste = _total_gc_waste + _total_slow_refill_waste + _total_fast_refill_waste;
412 double waste_percent = _total_allocation == 0 ? 0.0 :
413 100.0 * waste / _total_allocation;
414 gclog_or_tty->print("TLAB totals: thrds: %d refills: %d max: %d"
415 " slow allocs: %d max %d waste: %4.1f%%"
416 " gc: " SIZE_FORMAT "B max: " SIZE_FORMAT "B"
417 " slow: " SIZE_FORMAT "B max: " SIZE_FORMAT "B"
418 " fast: " SIZE_FORMAT "B max: " SIZE_FORMAT "B\n",
419 _allocating_threads,
420 _total_refills, _max_refills,
421 _total_slow_allocations, _max_slow_allocations,
422 waste_percent,
423 _total_gc_waste * HeapWordSize,
424 _max_gc_waste * HeapWordSize,
425 _total_slow_refill_waste * HeapWordSize,
426 _max_slow_refill_waste * HeapWordSize,
427 _total_fast_refill_waste * HeapWordSize,
428 _max_fast_refill_waste * HeapWordSize);
429 }