1 /*
2 * Copyright (c) 1999, 2017, 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 "gc/shared/genCollectedHeap.hpp"
27 #include "gc/shared/threadLocalAllocBuffer.inline.hpp"
28 #include "logging/log.hpp"
29 #include "memory/resourceArea.hpp"
30 #include "memory/universe.inline.hpp"
31 #include "oops/oop.inline.hpp"
32 #include "runtime/thread.inline.hpp"
33 #include "runtime/threadSMR.hpp"
34 #include "utilities/copy.hpp"
35
36 // Thread-Local Edens support
37
38 // static member initialization
39 size_t ThreadLocalAllocBuffer::_max_size = 0;
40 int ThreadLocalAllocBuffer::_reserve_for_allocation_prefetch = 0;
41 unsigned ThreadLocalAllocBuffer::_target_refills = 0;
42 GlobalTLABStats* ThreadLocalAllocBuffer::_global_stats = NULL;
43
44 void ThreadLocalAllocBuffer::clear_before_allocation() {
45 _slow_refill_waste += (unsigned)remaining();
46 make_parsable(true); // also retire the TLAB
47 }
48
49 void ThreadLocalAllocBuffer::accumulate_statistics_before_gc() {
50 global_stats()->initialize();
51
52 {
53 ThreadsListHandle tlh;
54 JavaThreadIterator jti(tlh.list());
55 for (JavaThread *thread = jti.first(); thread != NULL; thread = jti.next()) {
56 thread->tlab().accumulate_statistics();
57 thread->tlab().initialize_statistics();
58 }
59 }
60
61 // Publish new stats if some allocation occurred.
62 if (global_stats()->allocation() != 0) {
63 global_stats()->publish();
64 global_stats()->print();
65 }
66 }
67
68 void ThreadLocalAllocBuffer::accumulate_statistics() {
69 Thread* thread = myThread();
70 size_t capacity = Universe::heap()->tlab_capacity(thread);
71 size_t used = Universe::heap()->tlab_used(thread);
72
73 _gc_waste += (unsigned)remaining();
74 size_t total_allocated = thread->allocated_bytes();
75 size_t allocated_since_last_gc = total_allocated - _allocated_before_last_gc;
76 _allocated_before_last_gc = total_allocated;
77
78 print_stats("gc");
79
80 if (_number_of_refills > 0) {
81 // Update allocation history if a reasonable amount of eden was allocated.
82 bool update_allocation_history = used > 0.5 * capacity;
83
84 if (update_allocation_history) {
85 // Average the fraction of eden allocated in a tlab by this
86 // thread for use in the next resize operation.
87 // _gc_waste is not subtracted because it's included in
88 // "used".
89 // The result can be larger than 1.0 due to direct to old allocations.
90 // These allocations should ideally not be counted but since it is not possible
91 // to filter them out here we just cap the fraction to be at most 1.0.
92 double alloc_frac = MIN2(1.0, (double) allocated_since_last_gc / 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, bool zap) {
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 && zap);
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 ThreadsListHandle tlh;
139 JavaThreadIterator jti(tlh.list());
140 for (JavaThread *thread = jti.first(); thread != NULL; thread = jti.next()) {
141 thread->tlab().resize();
142 }
143 }
144 }
145
146 void ThreadLocalAllocBuffer::resize() {
147 // Compute the next tlab size using expected allocation amount
148 assert(ResizeTLAB, "Should not call this otherwise");
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 log_trace(gc, tlab)("TLAB new size: thread: " INTPTR_FORMAT " [id: %2d]"
158 " refills %d alloc: %8.6f desired_size: " SIZE_FORMAT " -> " SIZE_FORMAT,
159 p2i(myThread()), myThread()->osthread()->thread_id(),
160 _target_refills, _allocation_fraction.average(), desired_size(), aligned_new_size);
161
162 set_desired_size(aligned_new_size);
163 set_refill_waste_limit(initial_refill_waste_limit());
164 }
165
166 void ThreadLocalAllocBuffer::initialize_statistics() {
167 _number_of_refills = 0;
168 _fast_refill_waste = 0;
169 _slow_refill_waste = 0;
170 _gc_waste = 0;
171 _slow_allocations = 0;
172 }
173
174 void ThreadLocalAllocBuffer::fill(HeapWord* start,
175 HeapWord* top,
176 size_t new_size) {
177 _number_of_refills++;
178 print_stats("fill");
179 assert(top <= start + new_size - alignment_reserve(), "size too small");
180 initialize(start, top, start + new_size - alignment_reserve());
181
182 // Reset amount of internal fragmentation
183 set_refill_waste_limit(initial_refill_waste_limit());
184 }
185
186 void ThreadLocalAllocBuffer::initialize(HeapWord* start,
187 HeapWord* top,
188 HeapWord* end) {
189 set_start(start);
190 set_top(top);
191 set_pf_top(top);
192 set_end(end);
193 invariants();
194 }
195
196 void ThreadLocalAllocBuffer::initialize() {
197 initialize(NULL, // start
198 NULL, // top
199 NULL); // end
200
201 set_desired_size(initial_desired_size());
202
203 // Following check is needed because at startup the main (primordial)
204 // thread is initialized before the heap is. The initialization for
205 // this thread is redone in startup_initialization below.
206 if (Universe::heap() != NULL) {
207 size_t capacity = Universe::heap()->tlab_capacity(myThread()) / HeapWordSize;
208 double alloc_frac = desired_size() * target_refills() / (double) capacity;
209 _allocation_fraction.sample(alloc_frac);
210 }
211
212 set_refill_waste_limit(initial_refill_waste_limit());
213
214 initialize_statistics();
215 }
216
217 void ThreadLocalAllocBuffer::startup_initialization() {
218
219 // Assuming each thread's active tlab is, on average,
220 // 1/2 full at a GC
221 _target_refills = 100 / (2 * TLABWasteTargetPercent);
222 _target_refills = MAX2(_target_refills, (unsigned)1U);
223
224 _global_stats = new GlobalTLABStats();
225
226 #ifdef COMPILER2
227 // If the C2 compiler is present, extra space is needed at the end of
228 // TLABs, otherwise prefetching instructions generated by the C2
229 // compiler will fault (due to accessing memory outside of heap).
230 // The amount of space is the max of the number of lines to
231 // prefetch for array and for instance allocations. (Extra space must be
232 // reserved to accommodate both types of allocations.)
233 //
234 // Only SPARC-specific BIS instructions are known to fault. (Those
235 // instructions are generated if AllocatePrefetchStyle==3 and
236 // AllocatePrefetchInstr==1). To be on the safe side, however,
237 // extra space is reserved for all combinations of
238 // AllocatePrefetchStyle and AllocatePrefetchInstr.
239 //
240 // If the C2 compiler is not present, no space is reserved.
241
242 // +1 for rounding up to next cache line, +1 to be safe
243 if (is_server_compilation_mode_vm()) {
244 int lines = MAX2(AllocatePrefetchLines, AllocateInstancePrefetchLines) + 2;
245 _reserve_for_allocation_prefetch = (AllocatePrefetchDistance + AllocatePrefetchStepSize * lines) /
246 (int)HeapWordSize;
247 }
248 #endif
249
250 // During jvm startup, the main (primordial) thread is initialized
251 // before the heap is initialized. So reinitialize it now.
252 guarantee(Thread::current()->is_Java_thread(), "tlab initialization thread not Java thread");
253 Thread::current()->tlab().initialize();
254
255 log_develop_trace(gc, tlab)("TLAB min: " SIZE_FORMAT " initial: " SIZE_FORMAT " max: " SIZE_FORMAT,
256 min_size(), Thread::current()->tlab().initial_desired_size(), max_size());
257 }
258
259 size_t ThreadLocalAllocBuffer::initial_desired_size() {
260 size_t init_sz = 0;
261
262 if (TLABSize > 0) {
263 init_sz = TLABSize / HeapWordSize;
264 } else if (global_stats() != NULL) {
265 // Initial size is a function of the average number of allocating threads.
266 unsigned nof_threads = global_stats()->allocating_threads_avg();
267
268 init_sz = (Universe::heap()->tlab_capacity(myThread()) / HeapWordSize) /
269 (nof_threads * target_refills());
270 init_sz = align_object_size(init_sz);
271 }
272 init_sz = MIN2(MAX2(init_sz, min_size()), max_size());
273 return init_sz;
274 }
275
276 void ThreadLocalAllocBuffer::print_stats(const char* tag) {
277 Log(gc, tlab) log;
278 if (!log.is_trace()) {
279 return;
280 }
281
282 Thread* thrd = myThread();
283 size_t waste = _gc_waste + _slow_refill_waste + _fast_refill_waste;
284 size_t alloc = _number_of_refills * _desired_size;
285 double waste_percent = alloc == 0 ? 0.0 :
286 100.0 * waste / alloc;
287 size_t tlab_used = Universe::heap()->tlab_used(thrd);
288 log.trace("TLAB: %s thread: " INTPTR_FORMAT " [id: %2d]"
289 " desired_size: " SIZE_FORMAT "KB"
290 " slow allocs: %d refill waste: " SIZE_FORMAT "B"
291 " alloc:%8.5f %8.0fKB refills: %d waste %4.1f%% gc: %dB"
292 " slow: %dB fast: %dB",
293 tag, p2i(thrd), thrd->osthread()->thread_id(),
294 _desired_size / (K / HeapWordSize),
295 _slow_allocations, _refill_waste_limit * HeapWordSize,
296 _allocation_fraction.average(),
297 _allocation_fraction.average() * tlab_used / K,
298 _number_of_refills, waste_percent,
299 _gc_waste * HeapWordSize,
300 _slow_refill_waste * HeapWordSize,
301 _fast_refill_waste * HeapWordSize);
302 }
303
304 void ThreadLocalAllocBuffer::verify() {
305 HeapWord* p = start();
306 HeapWord* t = top();
307 HeapWord* prev_p = NULL;
308 while (p < t) {
309 oop(p)->verify();
310 prev_p = p;
311 p += oop(p)->size();
312 }
313 guarantee(p == top(), "end of last object must match end of space");
314 }
315
316 Thread* ThreadLocalAllocBuffer::myThread() {
317 return (Thread*)(((char *)this) +
318 in_bytes(start_offset()) -
319 in_bytes(Thread::tlab_start_offset()));
320 }
321
322
323 GlobalTLABStats::GlobalTLABStats() :
324 _allocating_threads_avg(TLABAllocationWeight) {
325
326 initialize();
327
328 _allocating_threads_avg.sample(1); // One allocating thread at startup
329
330 if (UsePerfData) {
331
332 EXCEPTION_MARK;
333 ResourceMark rm;
334
335 char* cname = PerfDataManager::counter_name("tlab", "allocThreads");
336 _perf_allocating_threads =
337 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_None, CHECK);
338
339 cname = PerfDataManager::counter_name("tlab", "fills");
340 _perf_total_refills =
341 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_None, CHECK);
342
343 cname = PerfDataManager::counter_name("tlab", "maxFills");
344 _perf_max_refills =
345 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_None, CHECK);
346
347 cname = PerfDataManager::counter_name("tlab", "alloc");
348 _perf_allocation =
349 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_Bytes, CHECK);
350
351 cname = PerfDataManager::counter_name("tlab", "gcWaste");
352 _perf_gc_waste =
353 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_Bytes, CHECK);
354
355 cname = PerfDataManager::counter_name("tlab", "maxGcWaste");
356 _perf_max_gc_waste =
357 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_Bytes, CHECK);
358
359 cname = PerfDataManager::counter_name("tlab", "slowWaste");
360 _perf_slow_refill_waste =
361 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_Bytes, CHECK);
362
363 cname = PerfDataManager::counter_name("tlab", "maxSlowWaste");
364 _perf_max_slow_refill_waste =
365 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_Bytes, CHECK);
366
367 cname = PerfDataManager::counter_name("tlab", "fastWaste");
368 _perf_fast_refill_waste =
369 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_Bytes, CHECK);
370
371 cname = PerfDataManager::counter_name("tlab", "maxFastWaste");
372 _perf_max_fast_refill_waste =
373 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_Bytes, CHECK);
374
375 cname = PerfDataManager::counter_name("tlab", "slowAlloc");
376 _perf_slow_allocations =
377 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_None, CHECK);
378
379 cname = PerfDataManager::counter_name("tlab", "maxSlowAlloc");
380 _perf_max_slow_allocations =
381 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_None, CHECK);
382 }
383 }
384
385 void GlobalTLABStats::initialize() {
386 // Clear counters summarizing info from all threads
387 _allocating_threads = 0;
388 _total_refills = 0;
389 _max_refills = 0;
390 _total_allocation = 0;
391 _total_gc_waste = 0;
392 _max_gc_waste = 0;
393 _total_slow_refill_waste = 0;
394 _max_slow_refill_waste = 0;
395 _total_fast_refill_waste = 0;
396 _max_fast_refill_waste = 0;
397 _total_slow_allocations = 0;
398 _max_slow_allocations = 0;
399 }
400
401 void GlobalTLABStats::publish() {
402 _allocating_threads_avg.sample(_allocating_threads);
403 if (UsePerfData) {
404 _perf_allocating_threads ->set_value(_allocating_threads);
405 _perf_total_refills ->set_value(_total_refills);
406 _perf_max_refills ->set_value(_max_refills);
407 _perf_allocation ->set_value(_total_allocation);
408 _perf_gc_waste ->set_value(_total_gc_waste);
409 _perf_max_gc_waste ->set_value(_max_gc_waste);
410 _perf_slow_refill_waste ->set_value(_total_slow_refill_waste);
411 _perf_max_slow_refill_waste->set_value(_max_slow_refill_waste);
412 _perf_fast_refill_waste ->set_value(_total_fast_refill_waste);
413 _perf_max_fast_refill_waste->set_value(_max_fast_refill_waste);
414 _perf_slow_allocations ->set_value(_total_slow_allocations);
415 _perf_max_slow_allocations ->set_value(_max_slow_allocations);
416 }
417 }
418
419 void GlobalTLABStats::print() {
420 Log(gc, tlab) log;
421 if (!log.is_debug()) {
422 return;
423 }
424
425 size_t waste = _total_gc_waste + _total_slow_refill_waste + _total_fast_refill_waste;
426 double waste_percent = _total_allocation == 0 ? 0.0 :
427 100.0 * waste / _total_allocation;
428 log.debug("TLAB totals: thrds: %d refills: %d max: %d"
429 " slow allocs: %d max %d waste: %4.1f%%"
430 " gc: " SIZE_FORMAT "B max: " SIZE_FORMAT "B"
431 " slow: " SIZE_FORMAT "B max: " SIZE_FORMAT "B"
432 " fast: " SIZE_FORMAT "B max: " SIZE_FORMAT "B",
433 _allocating_threads,
434 _total_refills, _max_refills,
435 _total_slow_allocations, _max_slow_allocations,
436 waste_percent,
437 _total_gc_waste * HeapWordSize,
438 _max_gc_waste * HeapWordSize,
439 _total_slow_refill_waste * HeapWordSize,
440 _max_slow_refill_waste * HeapWordSize,
441 _total_fast_refill_waste * HeapWordSize,
442 _max_fast_refill_waste * HeapWordSize);
443 }