|
0
|
1 /*
|
|
|
2 * Copyright 1999-2007 Sun Microsystems, Inc. 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 Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
|
|
|
20 * CA 95054 USA or visit www.sun.com if you need additional information or
|
|
|
21 * have any 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 MemRegion mr(top(), hard_end());
|
|
|
104 SharedHeap::fill_region_with_object(mr);
|
|
|
105
|
|
|
106 if (retire || ZeroTLAB) { // "Reset" the TLAB
|
|
|
107 set_start(NULL);
|
|
|
108 set_top(NULL);
|
|
|
109 set_pf_top(NULL);
|
|
|
110 set_end(NULL);
|
|
|
111 }
|
|
|
112 }
|
|
|
113 assert(!(retire || ZeroTLAB) ||
|
|
|
114 (start() == NULL && end() == NULL && top() == NULL),
|
|
|
115 "TLAB must be reset");
|
|
|
116 }
|
|
|
117
|
|
|
118 void ThreadLocalAllocBuffer::resize_all_tlabs() {
|
|
|
119 for(JavaThread *thread = Threads::first(); thread; thread = thread->next()) {
|
|
|
120 thread->tlab().resize();
|
|
|
121 }
|
|
|
122 }
|
|
|
123
|
|
|
124 void ThreadLocalAllocBuffer::resize() {
|
|
|
125
|
|
|
126 if (ResizeTLAB) {
|
|
|
127 // Compute the next tlab size using expected allocation amount
|
|
|
128 size_t alloc = (size_t)(_allocation_fraction.average() *
|
|
|
129 (Universe::heap()->tlab_capacity(myThread()) / HeapWordSize));
|
|
|
130 size_t new_size = alloc / _target_refills;
|
|
|
131
|
|
|
132 new_size = MIN2(MAX2(new_size, min_size()), max_size());
|
|
|
133
|
|
|
134 size_t aligned_new_size = align_object_size(new_size);
|
|
|
135
|
|
|
136 if (PrintTLAB && Verbose) {
|
|
|
137 gclog_or_tty->print("TLAB new size: thread: " INTPTR_FORMAT " [id: %2d]"
|
|
|
138 " refills %d alloc: %8.6f desired_size: " SIZE_FORMAT " -> " SIZE_FORMAT "\n",
|
|
|
139 myThread(), myThread()->osthread()->thread_id(),
|
|
|
140 _target_refills, _allocation_fraction.average(), desired_size(), aligned_new_size);
|
|
|
141 }
|
|
|
142 set_desired_size(aligned_new_size);
|
|
|
143
|
|
|
144 set_refill_waste_limit(initial_refill_waste_limit());
|
|
|
145 }
|
|
|
146 }
|
|
|
147
|
|
|
148 void ThreadLocalAllocBuffer::initialize_statistics() {
|
|
|
149 _number_of_refills = 0;
|
|
|
150 _fast_refill_waste = 0;
|
|
|
151 _slow_refill_waste = 0;
|
|
|
152 _gc_waste = 0;
|
|
|
153 _slow_allocations = 0;
|
|
|
154 }
|
|
|
155
|
|
|
156 void ThreadLocalAllocBuffer::fill(HeapWord* start,
|
|
|
157 HeapWord* top,
|
|
|
158 size_t new_size) {
|
|
|
159 _number_of_refills++;
|
|
|
160 if (PrintTLAB && Verbose) {
|
|
|
161 print_stats("fill");
|
|
|
162 }
|
|
|
163 assert(top <= start + new_size - alignment_reserve(), "size too small");
|
|
|
164 initialize(start, top, start + new_size - alignment_reserve());
|
|
|
165
|
|
|
166 // Reset amount of internal fragmentation
|
|
|
167 set_refill_waste_limit(initial_refill_waste_limit());
|
|
|
168 }
|
|
|
169
|
|
|
170 void ThreadLocalAllocBuffer::initialize(HeapWord* start,
|
|
|
171 HeapWord* top,
|
|
|
172 HeapWord* end) {
|
|
|
173 set_start(start);
|
|
|
174 set_top(top);
|
|
|
175 set_pf_top(top);
|
|
|
176 set_end(end);
|
|
|
177 invariants();
|
|
|
178 }
|
|
|
179
|
|
|
180 void ThreadLocalAllocBuffer::initialize() {
|
|
|
181 initialize(NULL, // start
|
|
|
182 NULL, // top
|
|
|
183 NULL); // end
|
|
|
184
|
|
|
185 set_desired_size(initial_desired_size());
|
|
|
186
|
|
|
187 // Following check is needed because at startup the main (primordial)
|
|
|
188 // thread is initialized before the heap is. The initialization for
|
|
|
189 // this thread is redone in startup_initialization below.
|
|
|
190 if (Universe::heap() != NULL) {
|
|
|
191 size_t capacity = Universe::heap()->tlab_capacity(myThread()) / HeapWordSize;
|
|
|
192 double alloc_frac = desired_size() * target_refills() / (double) capacity;
|
|
|
193 _allocation_fraction.sample(alloc_frac);
|
|
|
194 }
|
|
|
195
|
|
|
196 set_refill_waste_limit(initial_refill_waste_limit());
|
|
|
197
|
|
|
198 initialize_statistics();
|
|
|
199 }
|
|
|
200
|
|
|
201 void ThreadLocalAllocBuffer::startup_initialization() {
|
|
|
202
|
|
|
203 // Assuming each thread's active tlab is, on average,
|
|
|
204 // 1/2 full at a GC
|
|
|
205 _target_refills = 100 / (2 * TLABWasteTargetPercent);
|
|
|
206 _target_refills = MAX2(_target_refills, (unsigned)1U);
|
|
|
207
|
|
|
208 _global_stats = new GlobalTLABStats();
|
|
|
209
|
|
|
210 // During jvm startup, the main (primordial) thread is initialized
|
|
|
211 // before the heap is initialized. So reinitialize it now.
|
|
|
212 guarantee(Thread::current()->is_Java_thread(), "tlab initialization thread not Java thread");
|
|
|
213 Thread::current()->tlab().initialize();
|
|
|
214
|
|
|
215 if (PrintTLAB && Verbose) {
|
|
|
216 gclog_or_tty->print("TLAB min: " SIZE_FORMAT " initial: " SIZE_FORMAT " max: " SIZE_FORMAT "\n",
|
|
|
217 min_size(), Thread::current()->tlab().initial_desired_size(), max_size());
|
|
|
218 }
|
|
|
219 }
|
|
|
220
|
|
|
221 size_t ThreadLocalAllocBuffer::initial_desired_size() {
|
|
|
222 size_t init_sz;
|
|
|
223
|
|
|
224 if (TLABSize > 0) {
|
|
|
225 init_sz = MIN2(TLABSize / HeapWordSize, max_size());
|
|
|
226 } else if (global_stats() == NULL) {
|
|
|
227 // Startup issue - main thread initialized before heap initialized.
|
|
|
228 init_sz = min_size();
|
|
|
229 } else {
|
|
|
230 // Initial size is a function of the average number of allocating threads.
|
|
|
231 unsigned nof_threads = global_stats()->allocating_threads_avg();
|
|
|
232
|
|
|
233 init_sz = (Universe::heap()->tlab_capacity(myThread()) / HeapWordSize) /
|
|
|
234 (nof_threads * target_refills());
|
|
|
235 init_sz = align_object_size(init_sz);
|
|
|
236 init_sz = MIN2(MAX2(init_sz, min_size()), max_size());
|
|
|
237 }
|
|
|
238 return init_sz;
|
|
|
239 }
|
|
|
240
|
|
|
241 const size_t ThreadLocalAllocBuffer::max_size() {
|
|
|
242
|
|
|
243 // TLABs can't be bigger than we can fill with a int[Integer.MAX_VALUE].
|
|
|
244 // This restriction could be removed by enabling filling with multiple arrays.
|
|
|
245 // If we compute that the reasonable way as
|
|
|
246 // header_size + ((sizeof(jint) * max_jint) / HeapWordSize)
|
|
|
247 // we'll overflow on the multiply, so we do the divide first.
|
|
|
248 // We actually lose a little by dividing first,
|
|
|
249 // but that just makes the TLAB somewhat smaller than the biggest array,
|
|
|
250 // which is fine, since we'll be able to fill that.
|
|
|
251
|
|
|
252 size_t unaligned_max_size = typeArrayOopDesc::header_size(T_INT) +
|
|
|
253 sizeof(jint) *
|
|
|
254 ((juint) max_jint / (size_t) HeapWordSize);
|
|
|
255 return align_size_down(unaligned_max_size, MinObjAlignment);
|
|
|
256 }
|
|
|
257
|
|
|
258 void ThreadLocalAllocBuffer::print_stats(const char* tag) {
|
|
|
259 Thread* thrd = myThread();
|
|
|
260 size_t waste = _gc_waste + _slow_refill_waste + _fast_refill_waste;
|
|
|
261 size_t alloc = _number_of_refills * _desired_size;
|
|
|
262 double waste_percent = alloc == 0 ? 0.0 :
|
|
|
263 100.0 * waste / alloc;
|
|
|
264 size_t tlab_used = Universe::heap()->tlab_capacity(thrd) -
|
|
|
265 Universe::heap()->unsafe_max_tlab_alloc(thrd);
|
|
|
266 gclog_or_tty->print("TLAB: %s thread: " INTPTR_FORMAT " [id: %2d]"
|
|
|
267 " desired_size: " SIZE_FORMAT "KB"
|
|
|
268 " slow allocs: %d refill waste: " SIZE_FORMAT "B"
|
|
|
269 " alloc:%8.5f %8.0fKB refills: %d waste %4.1f%% gc: %dB"
|
|
|
270 " slow: %dB fast: %dB\n",
|
|
|
271 tag, thrd, thrd->osthread()->thread_id(),
|
|
|
272 _desired_size / (K / HeapWordSize),
|
|
|
273 _slow_allocations, _refill_waste_limit * HeapWordSize,
|
|
|
274 _allocation_fraction.average(),
|
|
|
275 _allocation_fraction.average() * tlab_used / K,
|
|
|
276 _number_of_refills, waste_percent,
|
|
|
277 _gc_waste * HeapWordSize,
|
|
|
278 _slow_refill_waste * HeapWordSize,
|
|
|
279 _fast_refill_waste * HeapWordSize);
|
|
|
280 }
|
|
|
281
|
|
|
282 void ThreadLocalAllocBuffer::verify() {
|
|
|
283 HeapWord* p = start();
|
|
|
284 HeapWord* t = top();
|
|
|
285 HeapWord* prev_p = NULL;
|
|
|
286 while (p < t) {
|
|
|
287 oop(p)->verify();
|
|
|
288 prev_p = p;
|
|
|
289 p += oop(p)->size();
|
|
|
290 }
|
|
|
291 guarantee(p == top(), "end of last object must match end of space");
|
|
|
292 }
|
|
|
293
|
|
|
294 Thread* ThreadLocalAllocBuffer::myThread() {
|
|
|
295 return (Thread*)(((char *)this) +
|
|
|
296 in_bytes(start_offset()) -
|
|
|
297 in_bytes(Thread::tlab_start_offset()));
|
|
|
298 }
|
|
|
299
|
|
|
300
|
|
|
301 GlobalTLABStats::GlobalTLABStats() :
|
|
|
302 _allocating_threads_avg(TLABAllocationWeight) {
|
|
|
303
|
|
|
304 initialize();
|
|
|
305
|
|
|
306 _allocating_threads_avg.sample(1); // One allocating thread at startup
|
|
|
307
|
|
|
308 if (UsePerfData) {
|
|
|
309
|
|
|
310 EXCEPTION_MARK;
|
|
|
311 ResourceMark rm;
|
|
|
312
|
|
|
313 char* cname = PerfDataManager::counter_name("tlab", "allocThreads");
|
|
|
314 _perf_allocating_threads =
|
|
|
315 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_None, CHECK);
|
|
|
316
|
|
|
317 cname = PerfDataManager::counter_name("tlab", "fills");
|
|
|
318 _perf_total_refills =
|
|
|
319 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_None, CHECK);
|
|
|
320
|
|
|
321 cname = PerfDataManager::counter_name("tlab", "maxFills");
|
|
|
322 _perf_max_refills =
|
|
|
323 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_None, CHECK);
|
|
|
324
|
|
|
325 cname = PerfDataManager::counter_name("tlab", "alloc");
|
|
|
326 _perf_allocation =
|
|
|
327 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_Bytes, CHECK);
|
|
|
328
|
|
|
329 cname = PerfDataManager::counter_name("tlab", "gcWaste");
|
|
|
330 _perf_gc_waste =
|
|
|
331 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_Bytes, CHECK);
|
|
|
332
|
|
|
333 cname = PerfDataManager::counter_name("tlab", "maxGcWaste");
|
|
|
334 _perf_max_gc_waste =
|
|
|
335 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_Bytes, CHECK);
|
|
|
336
|
|
|
337 cname = PerfDataManager::counter_name("tlab", "slowWaste");
|
|
|
338 _perf_slow_refill_waste =
|
|
|
339 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_Bytes, CHECK);
|
|
|
340
|
|
|
341 cname = PerfDataManager::counter_name("tlab", "maxSlowWaste");
|
|
|
342 _perf_max_slow_refill_waste =
|
|
|
343 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_Bytes, CHECK);
|
|
|
344
|
|
|
345 cname = PerfDataManager::counter_name("tlab", "fastWaste");
|
|
|
346 _perf_fast_refill_waste =
|
|
|
347 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_Bytes, CHECK);
|
|
|
348
|
|
|
349 cname = PerfDataManager::counter_name("tlab", "maxFastWaste");
|
|
|
350 _perf_max_fast_refill_waste =
|
|
|
351 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_Bytes, CHECK);
|
|
|
352
|
|
|
353 cname = PerfDataManager::counter_name("tlab", "slowAlloc");
|
|
|
354 _perf_slow_allocations =
|
|
|
355 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_None, CHECK);
|
|
|
356
|
|
|
357 cname = PerfDataManager::counter_name("tlab", "maxSlowAlloc");
|
|
|
358 _perf_max_slow_allocations =
|
|
|
359 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_None, CHECK);
|
|
|
360 }
|
|
|
361 }
|
|
|
362
|
|
|
363 void GlobalTLABStats::initialize() {
|
|
|
364 // Clear counters summarizing info from all threads
|
|
|
365 _allocating_threads = 0;
|
|
|
366 _total_refills = 0;
|
|
|
367 _max_refills = 0;
|
|
|
368 _total_allocation = 0;
|
|
|
369 _total_gc_waste = 0;
|
|
|
370 _max_gc_waste = 0;
|
|
|
371 _total_slow_refill_waste = 0;
|
|
|
372 _max_slow_refill_waste = 0;
|
|
|
373 _total_fast_refill_waste = 0;
|
|
|
374 _max_fast_refill_waste = 0;
|
|
|
375 _total_slow_allocations = 0;
|
|
|
376 _max_slow_allocations = 0;
|
|
|
377 }
|
|
|
378
|
|
|
379 void GlobalTLABStats::publish() {
|
|
|
380 _allocating_threads_avg.sample(_allocating_threads);
|
|
|
381 if (UsePerfData) {
|
|
|
382 _perf_allocating_threads ->set_value(_allocating_threads);
|
|
|
383 _perf_total_refills ->set_value(_total_refills);
|
|
|
384 _perf_max_refills ->set_value(_max_refills);
|
|
|
385 _perf_allocation ->set_value(_total_allocation);
|
|
|
386 _perf_gc_waste ->set_value(_total_gc_waste);
|
|
|
387 _perf_max_gc_waste ->set_value(_max_gc_waste);
|
|
|
388 _perf_slow_refill_waste ->set_value(_total_slow_refill_waste);
|
|
|
389 _perf_max_slow_refill_waste->set_value(_max_slow_refill_waste);
|
|
|
390 _perf_fast_refill_waste ->set_value(_total_fast_refill_waste);
|
|
|
391 _perf_max_fast_refill_waste->set_value(_max_fast_refill_waste);
|
|
|
392 _perf_slow_allocations ->set_value(_total_slow_allocations);
|
|
|
393 _perf_max_slow_allocations ->set_value(_max_slow_allocations);
|
|
|
394 }
|
|
|
395 }
|
|
|
396
|
|
|
397 void GlobalTLABStats::print() {
|
|
|
398 size_t waste = _total_gc_waste + _total_slow_refill_waste + _total_fast_refill_waste;
|
|
|
399 double waste_percent = _total_allocation == 0 ? 0.0 :
|
|
|
400 100.0 * waste / _total_allocation;
|
|
|
401 gclog_or_tty->print("TLAB totals: thrds: %d refills: %d max: %d"
|
|
|
402 " slow allocs: %d max %d waste: %4.1f%%"
|
|
|
403 " gc: " SIZE_FORMAT "B max: " SIZE_FORMAT "B"
|
|
|
404 " slow: " SIZE_FORMAT "B max: " SIZE_FORMAT "B"
|
|
|
405 " fast: " SIZE_FORMAT "B max: " SIZE_FORMAT "B\n",
|
|
|
406 _allocating_threads,
|
|
|
407 _total_refills, _max_refills,
|
|
|
408 _total_slow_allocations, _max_slow_allocations,
|
|
|
409 waste_percent,
|
|
|
410 _total_gc_waste * HeapWordSize,
|
|
|
411 _max_gc_waste * HeapWordSize,
|
|
|
412 _total_slow_refill_waste * HeapWordSize,
|
|
|
413 _max_slow_refill_waste * HeapWordSize,
|
|
|
414 _total_fast_refill_waste * HeapWordSize,
|
|
|
415 _max_fast_refill_waste * HeapWordSize);
|
|
|
416 }
|
