1 /*
2 * Copyright (c) 1999, 2011, 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 #ifdef TARGET_OS_FAMILY_bsd
42 # include "thread_bsd.inline.hpp"
43 #endif
44
45 // Thread-Local Edens support
46
47 // static member initialization
48 unsigned ThreadLocalAllocBuffer::_target_refills = 0;
49 GlobalTLABStats* ThreadLocalAllocBuffer::_global_stats = NULL;
50
51 void ThreadLocalAllocBuffer::clear_before_allocation() {
52 _slow_refill_waste += (unsigned)remaining();
53 make_parsable(true); // also retire the TLAB
54 }
55
56 void ThreadLocalAllocBuffer::accumulate_statistics_before_gc() {
57 global_stats()->initialize();
58
59 for(JavaThread *thread = Threads::first(); thread; thread = thread->next()) {
60 thread->tlab().accumulate_statistics();
61 thread->tlab().initialize_statistics();
62 }
63
64 // Publish new stats if some allocation occurred.
65 if (global_stats()->allocation() != 0) {
66 global_stats()->publish();
67 if (PrintTLAB) {
68 global_stats()->print();
69 }
70 }
71 }
72
73 void ThreadLocalAllocBuffer::accumulate_statistics() {
74 size_t capacity = Universe::heap()->tlab_capacity(myThread()) / HeapWordSize;
75 size_t unused = Universe::heap()->unsafe_max_tlab_alloc(myThread()) / HeapWordSize;
76 size_t used = capacity - unused;
77
78 // Update allocation history if a reasonable amount of eden was allocated.
79 bool update_allocation_history = used > 0.5 * capacity;
80
81 _gc_waste += (unsigned)remaining();
82
83 if (PrintTLAB && (_number_of_refills > 0 || Verbose)) {
84 print_stats("gc");
85 }
86
87 if (_number_of_refills > 0) {
88
89 if (update_allocation_history) {
90 // Average the fraction of eden allocated in a tlab by this
91 // thread for use in the next resize operation.
92 // _gc_waste is not subtracted because it's included in
93 // "used".
94 size_t allocation = _number_of_refills * desired_size();
95 double alloc_frac = allocation / (double) used;
96 _allocation_fraction.sample(alloc_frac);
97 }
98 global_stats()->update_allocating_threads();
99 global_stats()->update_number_of_refills(_number_of_refills);
100 global_stats()->update_allocation(_number_of_refills * desired_size());
101 global_stats()->update_gc_waste(_gc_waste);
102 global_stats()->update_slow_refill_waste(_slow_refill_waste);
103 global_stats()->update_fast_refill_waste(_fast_refill_waste);
104
105 } else {
106 assert(_number_of_refills == 0 && _fast_refill_waste == 0 &&
107 _slow_refill_waste == 0 && _gc_waste == 0,
108 "tlab stats == 0");
109 }
110 global_stats()->update_slow_allocations(_slow_allocations);
111 }
112
113 // Fills the current tlab with a dummy filler array to create
114 // an illusion of a contiguous Eden and optionally retires the tlab.
115 // Waste accounting should be done in caller as appropriate; see,
116 // for example, clear_before_allocation().
117 void ThreadLocalAllocBuffer::make_parsable(bool retire) {
118 if (end() != NULL) {
119 invariants();
120
121 if (retire) {
122 myThread()->incr_allocated_bytes(used_bytes());
123 }
124
125 CollectedHeap::fill_with_object(top(), hard_end(), retire);
126
127 if (retire || ZeroTLAB) { // "Reset" the TLAB
128 set_start(NULL);
129 set_top(NULL);
130 set_pf_top(NULL);
131 set_end(NULL);
132 }
133 }
134 assert(!(retire || ZeroTLAB) ||
135 (start() == NULL && end() == NULL && top() == NULL),
136 "TLAB must be reset");
137 }
138
139 void ThreadLocalAllocBuffer::resize_all_tlabs() {
140 for(JavaThread *thread = Threads::first(); thread; thread = thread->next()) {
141 thread->tlab().resize();
142 }
143 }
144
145 void ThreadLocalAllocBuffer::resize() {
146
147 if (ResizeTLAB) {
148 // Compute the next tlab size using expected allocation amount
149 size_t alloc = (size_t)(_allocation_fraction.average() *
150 (Universe::heap()->tlab_capacity(myThread()) / HeapWordSize));
151 size_t new_size = alloc / _target_refills;
152
153 new_size = MIN2(MAX2(new_size, min_size()), max_size());
154
155 size_t aligned_new_size = align_object_size(new_size);
156
157 if (PrintTLAB && Verbose) {
158 gclog_or_tty->print("TLAB new size: thread: " INTPTR_FORMAT " [id: %2d]"
159 " refills %d alloc: %8.6f desired_size: " SIZE_FORMAT " -> " SIZE_FORMAT "\n",
160 myThread(), myThread()->osthread()->thread_id(),
161 _target_refills, _allocation_fraction.average(), desired_size(), aligned_new_size);
162 }
163 set_desired_size(aligned_new_size);
164
165 set_refill_waste_limit(initial_refill_waste_limit());
166 }
167 }
168
169 void ThreadLocalAllocBuffer::initialize_statistics() {
170 _number_of_refills = 0;
171 _fast_refill_waste = 0;
172 _slow_refill_waste = 0;
173 _gc_waste = 0;
174 _slow_allocations = 0;
175 }
176
177 void ThreadLocalAllocBuffer::fill(HeapWord* start,
178 HeapWord* top,
179 size_t new_size) {
180 _number_of_refills++;
181 if (PrintTLAB && Verbose) {
182 print_stats("fill");
183 }
184 assert(top <= start + new_size - alignment_reserve(), "size too small");
185 initialize(start, top, start + new_size - alignment_reserve());
186
187 // Reset amount of internal fragmentation
188 set_refill_waste_limit(initial_refill_waste_limit());
189 }
190
191 void ThreadLocalAllocBuffer::initialize(HeapWord* start,
192 HeapWord* top,
193 HeapWord* end) {
194 set_start(start);
195 set_top(top);
196 set_pf_top(top);
197 set_end(end);
198 invariants();
199 }
200
201 void ThreadLocalAllocBuffer::initialize() {
202 initialize(NULL, // start
203 NULL, // top
204 NULL); // end
205
206 set_desired_size(initial_desired_size());
207
208 // Following check is needed because at startup the main (primordial)
209 // thread is initialized before the heap is. The initialization for
210 // this thread is redone in startup_initialization below.
211 if (Universe::heap() != NULL) {
212 size_t capacity = Universe::heap()->tlab_capacity(myThread()) / HeapWordSize;
213 double alloc_frac = desired_size() * target_refills() / (double) capacity;
214 _allocation_fraction.sample(alloc_frac);
215 }
216
217 set_refill_waste_limit(initial_refill_waste_limit());
218
219 initialize_statistics();
220 }
221
222 void ThreadLocalAllocBuffer::startup_initialization() {
223
224 // Assuming each thread's active tlab is, on average,
225 // 1/2 full at a GC
226 _target_refills = 100 / (2 * TLABWasteTargetPercent);
227 _target_refills = MAX2(_target_refills, (unsigned)1U);
228
229 _global_stats = new GlobalTLABStats();
230
231 // During jvm startup, the main (primordial) thread is initialized
232 // before the heap is initialized. So reinitialize it now.
233 guarantee(Thread::current()->is_Java_thread(), "tlab initialization thread not Java thread");
234 Thread::current()->tlab().initialize();
235
236 if (PrintTLAB && Verbose) {
237 gclog_or_tty->print("TLAB min: " SIZE_FORMAT " initial: " SIZE_FORMAT " max: " SIZE_FORMAT "\n",
238 min_size(), Thread::current()->tlab().initial_desired_size(), max_size());
239 }
240 }
241
242 size_t ThreadLocalAllocBuffer::initial_desired_size() {
243 size_t init_sz;
244
245 if (TLABSize > 0) {
246 init_sz = MIN2(TLABSize / HeapWordSize, max_size());
247 } else if (global_stats() == NULL) {
248 // Startup issue - main thread initialized before heap initialized.
249 init_sz = min_size();
250 } else {
251 // Initial size is a function of the average number of allocating threads.
252 unsigned nof_threads = global_stats()->allocating_threads_avg();
253
254 init_sz = (Universe::heap()->tlab_capacity(myThread()) / HeapWordSize) /
255 (nof_threads * target_refills());
256 init_sz = align_object_size(init_sz);
257 init_sz = MIN2(MAX2(init_sz, min_size()), max_size());
258 }
259 return init_sz;
260 }
261
262 const size_t ThreadLocalAllocBuffer::max_size() {
263
264 // TLABs can't be bigger than we can fill with a int[Integer.MAX_VALUE].
265 // This restriction could be removed by enabling filling with multiple arrays.
266 // If we compute that the reasonable way as
267 // header_size + ((sizeof(jint) * max_jint) / HeapWordSize)
268 // we'll overflow on the multiply, so we do the divide first.
269 // We actually lose a little by dividing first,
270 // but that just makes the TLAB somewhat smaller than the biggest array,
271 // which is fine, since we'll be able to fill that.
272
273 size_t unaligned_max_size = typeArrayOopDesc::header_size(T_INT) +
274 sizeof(jint) *
275 ((juint) max_jint / (size_t) HeapWordSize);
276 return align_size_down(unaligned_max_size, MinObjAlignment);
277 }
278
279 void ThreadLocalAllocBuffer::print_stats(const char* tag) {
280 Thread* thrd = myThread();
281 size_t waste = _gc_waste + _slow_refill_waste + _fast_refill_waste;
282 size_t alloc = _number_of_refills * _desired_size;
283 double waste_percent = alloc == 0 ? 0.0 :
284 100.0 * waste / alloc;
285 size_t tlab_used = Universe::heap()->tlab_capacity(thrd) -
286 Universe::heap()->unsafe_max_tlab_alloc(thrd);
287 gclog_or_tty->print("TLAB: %s thread: " INTPTR_FORMAT " [id: %2d]"
288 " desired_size: " SIZE_FORMAT "KB"
289 " slow allocs: %d refill waste: " SIZE_FORMAT "B"
290 " alloc:%8.5f %8.0fKB refills: %d waste %4.1f%% gc: %dB"
291 " slow: %dB fast: %dB\n",
292 tag, thrd, thrd->osthread()->thread_id(),
293 _desired_size / (K / HeapWordSize),
294 _slow_allocations, _refill_waste_limit * HeapWordSize,
295 _allocation_fraction.average(),
296 _allocation_fraction.average() * tlab_used / K,
297 _number_of_refills, waste_percent,
298 _gc_waste * HeapWordSize,
299 _slow_refill_waste * HeapWordSize,
300 _fast_refill_waste * HeapWordSize);
301 }
302
303 void ThreadLocalAllocBuffer::verify() {
304 HeapWord* p = start();
305 HeapWord* t = top();
306 HeapWord* prev_p = NULL;
307 while (p < t) {
308 oop(p)->verify();
309 prev_p = p;
310 p += oop(p)->size();
311 }
312 guarantee(p == top(), "end of last object must match end of space");
313 }
314
315 Thread* ThreadLocalAllocBuffer::myThread() {
316 return (Thread*)(((char *)this) +
317 in_bytes(start_offset()) -
318 in_bytes(Thread::tlab_start_offset()));
319 }
320
321
322 GlobalTLABStats::GlobalTLABStats() :
323 _allocating_threads_avg(TLABAllocationWeight) {
324
325 initialize();
326
327 _allocating_threads_avg.sample(1); // One allocating thread at startup
328
329 if (UsePerfData) {
330
331 EXCEPTION_MARK;
332 ResourceMark rm;
333
334 char* cname = PerfDataManager::counter_name("tlab", "allocThreads");
335 _perf_allocating_threads =
336 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_None, CHECK);
337
338 cname = PerfDataManager::counter_name("tlab", "fills");
339 _perf_total_refills =
340 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_None, CHECK);
341
342 cname = PerfDataManager::counter_name("tlab", "maxFills");
343 _perf_max_refills =
344 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_None, CHECK);
345
346 cname = PerfDataManager::counter_name("tlab", "alloc");
347 _perf_allocation =
348 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_Bytes, CHECK);
349
350 cname = PerfDataManager::counter_name("tlab", "gcWaste");
351 _perf_gc_waste =
352 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_Bytes, CHECK);
353
354 cname = PerfDataManager::counter_name("tlab", "maxGcWaste");
355 _perf_max_gc_waste =
356 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_Bytes, CHECK);
357
358 cname = PerfDataManager::counter_name("tlab", "slowWaste");
359 _perf_slow_refill_waste =
360 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_Bytes, CHECK);
361
362 cname = PerfDataManager::counter_name("tlab", "maxSlowWaste");
363 _perf_max_slow_refill_waste =
364 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_Bytes, CHECK);
365
366 cname = PerfDataManager::counter_name("tlab", "fastWaste");
367 _perf_fast_refill_waste =
368 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_Bytes, CHECK);
369
370 cname = PerfDataManager::counter_name("tlab", "maxFastWaste");
371 _perf_max_fast_refill_waste =
372 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_Bytes, CHECK);
373
374 cname = PerfDataManager::counter_name("tlab", "slowAlloc");
375 _perf_slow_allocations =
376 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_None, CHECK);
377
378 cname = PerfDataManager::counter_name("tlab", "maxSlowAlloc");
379 _perf_max_slow_allocations =
380 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_None, CHECK);
381 }
382 }
383
384 void GlobalTLABStats::initialize() {
385 // Clear counters summarizing info from all threads
386 _allocating_threads = 0;
387 _total_refills = 0;
388 _max_refills = 0;
389 _total_allocation = 0;
390 _total_gc_waste = 0;
391 _max_gc_waste = 0;
392 _total_slow_refill_waste = 0;
393 _max_slow_refill_waste = 0;
394 _total_fast_refill_waste = 0;
395 _max_fast_refill_waste = 0;
396 _total_slow_allocations = 0;
397 _max_slow_allocations = 0;
398 }
399
400 void GlobalTLABStats::publish() {
401 _allocating_threads_avg.sample(_allocating_threads);
402 if (UsePerfData) {
403 _perf_allocating_threads ->set_value(_allocating_threads);
404 _perf_total_refills ->set_value(_total_refills);
405 _perf_max_refills ->set_value(_max_refills);
406 _perf_allocation ->set_value(_total_allocation);
407 _perf_gc_waste ->set_value(_total_gc_waste);
408 _perf_max_gc_waste ->set_value(_max_gc_waste);
409 _perf_slow_refill_waste ->set_value(_total_slow_refill_waste);
410 _perf_max_slow_refill_waste->set_value(_max_slow_refill_waste);
411 _perf_fast_refill_waste ->set_value(_total_fast_refill_waste);
412 _perf_max_fast_refill_waste->set_value(_max_fast_refill_waste);
413 _perf_slow_allocations ->set_value(_total_slow_allocations);
414 _perf_max_slow_allocations ->set_value(_max_slow_allocations);
415 }
416 }
417
418 void GlobalTLABStats::print() {
419 size_t waste = _total_gc_waste + _total_slow_refill_waste + _total_fast_refill_waste;
420 double waste_percent = _total_allocation == 0 ? 0.0 :
421 100.0 * waste / _total_allocation;
422 gclog_or_tty->print("TLAB totals: thrds: %d refills: %d max: %d"
423 " slow allocs: %d max %d waste: %4.1f%%"
424 " gc: " SIZE_FORMAT "B max: " SIZE_FORMAT "B"
425 " slow: " SIZE_FORMAT "B max: " SIZE_FORMAT "B"
426 " fast: " SIZE_FORMAT "B max: " SIZE_FORMAT "B\n",
427 _allocating_threads,
428 _total_refills, _max_refills,
429 _total_slow_allocations, _max_slow_allocations,
430 waste_percent,
431 _total_gc_waste * HeapWordSize,
432 _max_gc_waste * HeapWordSize,
433 _total_slow_refill_waste * HeapWordSize,
434 _max_slow_refill_waste * HeapWordSize,
435 _total_fast_refill_waste * HeapWordSize,
436 _max_fast_refill_waste * HeapWordSize);
437 }