1 /*
2 * Copyright (c) 1999, 2018, 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/threadLocalAllocBuffer.inline.hpp"
27 #include "logging/log.hpp"
28 #include "memory/resourceArea.hpp"
29 #include "memory/universe.hpp"
30 #include "oops/oop.inline.hpp"
31 #include "runtime/thread.inline.hpp"
32 #include "runtime/threadSMR.hpp"
33 #include "utilities/copy.hpp"
34
35 // Thread-Local Edens support
36
37 // static member initialization
38 size_t ThreadLocalAllocBuffer::_max_size = 0;
39 int ThreadLocalAllocBuffer::_reserve_for_allocation_prefetch = 0;
40 unsigned ThreadLocalAllocBuffer::_target_refills = 0;
41 GlobalTLABStats* ThreadLocalAllocBuffer::_global_stats = NULL;
42
43 void ThreadLocalAllocBuffer::clear_before_allocation() {
44 _slow_refill_waste += (unsigned)remaining();
45 make_parsable(true); // also retire the TLAB
46 }
47
48 size_t ThreadLocalAllocBuffer::remaining() {
49 if (end() == NULL) {
50 return 0;
51 }
52
53 return pointer_delta(hard_end(), top());
54 }
55
56 void ThreadLocalAllocBuffer::accumulate_statistics_before_gc() {
57 global_stats()->initialize();
58
59 for (JavaThreadIteratorWithHandle jtiwh; JavaThread *thread = jtiwh.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 global_stats()->print();
68 }
69 }
70
71 void ThreadLocalAllocBuffer::accumulate_statistics() {
72 Thread* thr = thread();
73 size_t capacity = Universe::heap()->tlab_capacity(thr);
74 size_t used = Universe::heap()->tlab_used(thr);
75
76 _gc_waste += (unsigned)remaining();
77 size_t total_allocated = thr->allocated_bytes();
78 size_t allocated_since_last_gc = total_allocated - _allocated_before_last_gc;
79 _allocated_before_last_gc = total_allocated;
80
81 print_stats("gc");
82
83 if (_number_of_refills > 0) {
84 // Update allocation history if a reasonable amount of eden was allocated.
85 bool update_allocation_history = used > 0.5 * capacity;
86
87 if (update_allocation_history) {
88 // Average the fraction of eden allocated in a tlab by this
89 // thread for use in the next resize operation.
90 // _gc_waste is not subtracted because it's included in
91 // "used".
92 // The result can be larger than 1.0 due to direct to old allocations.
93 // These allocations should ideally not be counted but since it is not possible
94 // to filter them out here we just cap the fraction to be at most 1.0.
95 double alloc_frac = MIN2(1.0, (double) allocated_since_last_gc / 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(_allocated_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, bool zap) {
118 if (end() != NULL) {
119 invariants();
120
121 if (retire) {
122 thread()->incr_allocated_bytes(used_bytes());
123 }
124
125 Universe::heap()->fill_with_dummy_object(top(), hard_end(), retire && zap);
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 set_allocation_end(NULL);
133 }
134 }
135 assert(!(retire || ZeroTLAB) ||
136 (start() == NULL && end() == NULL && top() == NULL &&
137 _allocation_end == NULL),
138 "TLAB must be reset");
139 }
140
141 void ThreadLocalAllocBuffer::resize() {
142 // Compute the next tlab size using expected allocation amount
143 assert(ResizeTLAB, "Should not call this otherwise");
144 size_t alloc = (size_t)(_allocation_fraction.average() *
145 (Universe::heap()->tlab_capacity(thread()) / HeapWordSize));
146 size_t new_size = alloc / _target_refills;
147
148 new_size = MIN2(MAX2(new_size, min_size()), max_size());
149
150 size_t aligned_new_size = align_object_size(new_size);
151
152 log_trace(gc, tlab)("TLAB new size: thread: " INTPTR_FORMAT " [id: %2d]"
153 " refills %d alloc: %8.6f desired_size: " SIZE_FORMAT " -> " SIZE_FORMAT,
154 p2i(thread()), thread()->osthread()->thread_id(),
155 _target_refills, _allocation_fraction.average(), desired_size(), aligned_new_size);
156
157 set_desired_size(aligned_new_size);
158 set_refill_waste_limit(initial_refill_waste_limit());
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 _allocated_size = 0;
168 }
169
170 void ThreadLocalAllocBuffer::fill(HeapWord* start,
171 HeapWord* top,
172 size_t new_size) {
173 _number_of_refills++;
174 _allocated_size += new_size;
175 print_stats("fill");
176 assert(top <= start + new_size - alignment_reserve(), "size too small");
177
178 initialize(start, top, start + new_size - alignment_reserve());
179
180 // Reset amount of internal fragmentation
181 set_refill_waste_limit(initial_refill_waste_limit());
182 }
183
184 void ThreadLocalAllocBuffer::initialize(HeapWord* start,
185 HeapWord* top,
186 HeapWord* end) {
187 set_start(start);
188 set_top(top);
189 set_pf_top(top);
190 set_end(end);
191 set_allocation_end(end);
192 invariants();
193 }
194
195 void ThreadLocalAllocBuffer::initialize() {
196 initialize(NULL, // start
197 NULL, // top
198 NULL); // end
199
200 set_desired_size(initial_desired_size());
201
202 // Following check is needed because at startup the main
203 // thread is initialized before the heap is. The initialization for
204 // this thread is redone in startup_initialization below.
205 if (Universe::heap() != NULL) {
206 size_t capacity = Universe::heap()->tlab_capacity(thread()) / HeapWordSize;
207 double alloc_frac = desired_size() * target_refills() / (double) capacity;
208 _allocation_fraction.sample(alloc_frac);
209 }
210
211 set_refill_waste_limit(initial_refill_waste_limit());
212
213 initialize_statistics();
214 }
215
216 void ThreadLocalAllocBuffer::startup_initialization() {
217
218 // Assuming each thread's active tlab is, on average,
219 // 1/2 full at a GC
220 _target_refills = 100 / (2 * TLABWasteTargetPercent);
221 // We need to set initial target refills to 2 to avoid a GC which causes VM
222 // abort during VM initialization.
223 _target_refills = MAX2(_target_refills, 2U);
224
225 _global_stats = new GlobalTLABStats();
226
227 #ifdef COMPILER2
228 // If the C2 compiler is present, extra space is needed at the end of
229 // TLABs, otherwise prefetching instructions generated by the C2
230 // compiler will fault (due to accessing memory outside of heap).
231 // The amount of space is the max of the number of lines to
232 // prefetch for array and for instance allocations. (Extra space must be
233 // reserved to accommodate both types of allocations.)
234 //
235 // Only SPARC-specific BIS instructions are known to fault. (Those
236 // instructions are generated if AllocatePrefetchStyle==3 and
237 // AllocatePrefetchInstr==1). To be on the safe side, however,
238 // extra space is reserved for all combinations of
239 // AllocatePrefetchStyle and AllocatePrefetchInstr.
240 //
241 // If the C2 compiler is not present, no space is reserved.
242
243 // +1 for rounding up to next cache line, +1 to be safe
244 if (is_server_compilation_mode_vm()) {
245 int lines = MAX2(AllocatePrefetchLines, AllocateInstancePrefetchLines) + 2;
246 _reserve_for_allocation_prefetch = (AllocatePrefetchDistance + AllocatePrefetchStepSize * lines) /
247 (int)HeapWordSize;
248 }
249 #endif
250
251 // During jvm startup, the main thread is initialized
252 // before the heap is initialized. So reinitialize it now.
253 guarantee(Thread::current()->is_Java_thread(), "tlab initialization thread not Java thread");
254 Thread::current()->tlab().initialize();
255
256 log_develop_trace(gc, tlab)("TLAB min: " SIZE_FORMAT " initial: " SIZE_FORMAT " max: " SIZE_FORMAT,
257 min_size(), Thread::current()->tlab().initial_desired_size(), max_size());
258 }
259
260 size_t ThreadLocalAllocBuffer::initial_desired_size() {
261 size_t init_sz = 0;
262
263 if (TLABSize > 0) {
264 init_sz = TLABSize / HeapWordSize;
265 } else if (global_stats() != NULL) {
266 // Initial size is a function of the average number of allocating threads.
267 unsigned nof_threads = global_stats()->allocating_threads_avg();
268
269 init_sz = (Universe::heap()->tlab_capacity(thread()) / HeapWordSize) /
270 (nof_threads * target_refills());
271 init_sz = align_object_size(init_sz);
272 }
273 init_sz = MIN2(MAX2(init_sz, min_size()), max_size());
274 return init_sz;
275 }
276
277 void ThreadLocalAllocBuffer::print_stats(const char* tag) {
278 Log(gc, tlab) log;
279 if (!log.is_trace()) {
280 return;
281 }
282
283 Thread* thrd = thread();
284 size_t waste = _gc_waste + _slow_refill_waste + _fast_refill_waste;
285 double waste_percent = percent_of(waste, _allocated_size);
286 size_t tlab_used = Universe::heap()->tlab_used(thrd);
287 log.trace("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",
292 tag, p2i(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 oopDesc::verify(oop(p));
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 void ThreadLocalAllocBuffer::set_sample_end() {
316 size_t heap_words_remaining = pointer_delta(_end, _top);
317 size_t bytes_until_sample = thread()->heap_sampler().bytes_until_sample();
318 size_t words_until_sample = bytes_until_sample / HeapWordSize;
319
320 if (heap_words_remaining > words_until_sample) {
321 HeapWord* new_end = _top + words_until_sample;
322 set_end(new_end);
323 _bytes_since_last_sample_point = bytes_until_sample;
324 } else {
325 _bytes_since_last_sample_point = heap_words_remaining * HeapWordSize;
326 }
327 }
328
329 Thread* ThreadLocalAllocBuffer::thread() {
330 return (Thread*)(((char*)this) + in_bytes(start_offset()) - in_bytes(Thread::tlab_start_offset()));
331 }
332
333 void ThreadLocalAllocBuffer::set_back_allocation_end() {
334 _end = _allocation_end;
335 }
336
337 HeapWord* ThreadLocalAllocBuffer::hard_end() {
338 return _allocation_end + alignment_reserve();
339 }
340
341 GlobalTLABStats::GlobalTLABStats() :
342 _allocating_threads_avg(TLABAllocationWeight) {
343
344 initialize();
345
346 _allocating_threads_avg.sample(1); // One allocating thread at startup
347
348 if (UsePerfData) {
349
350 EXCEPTION_MARK;
351 ResourceMark rm;
352
353 char* cname = PerfDataManager::counter_name("tlab", "allocThreads");
354 _perf_allocating_threads =
355 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_None, CHECK);
356
357 cname = PerfDataManager::counter_name("tlab", "fills");
358 _perf_total_refills =
359 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_None, CHECK);
360
361 cname = PerfDataManager::counter_name("tlab", "maxFills");
362 _perf_max_refills =
363 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_None, CHECK);
364
365 cname = PerfDataManager::counter_name("tlab", "alloc");
366 _perf_allocation =
367 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_Bytes, CHECK);
368
369 cname = PerfDataManager::counter_name("tlab", "gcWaste");
370 _perf_gc_waste =
371 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_Bytes, CHECK);
372
373 cname = PerfDataManager::counter_name("tlab", "maxGcWaste");
374 _perf_max_gc_waste =
375 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_Bytes, CHECK);
376
377 cname = PerfDataManager::counter_name("tlab", "slowWaste");
378 _perf_slow_refill_waste =
379 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_Bytes, CHECK);
380
381 cname = PerfDataManager::counter_name("tlab", "maxSlowWaste");
382 _perf_max_slow_refill_waste =
383 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_Bytes, CHECK);
384
385 cname = PerfDataManager::counter_name("tlab", "fastWaste");
386 _perf_fast_refill_waste =
387 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_Bytes, CHECK);
388
389 cname = PerfDataManager::counter_name("tlab", "maxFastWaste");
390 _perf_max_fast_refill_waste =
391 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_Bytes, CHECK);
392
393 cname = PerfDataManager::counter_name("tlab", "slowAlloc");
394 _perf_slow_allocations =
395 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_None, CHECK);
396
397 cname = PerfDataManager::counter_name("tlab", "maxSlowAlloc");
398 _perf_max_slow_allocations =
399 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_None, CHECK);
400 }
401 }
402
403 void GlobalTLABStats::initialize() {
404 // Clear counters summarizing info from all threads
405 _allocating_threads = 0;
406 _total_refills = 0;
407 _max_refills = 0;
408 _total_allocation = 0;
409 _total_gc_waste = 0;
410 _max_gc_waste = 0;
411 _total_slow_refill_waste = 0;
412 _max_slow_refill_waste = 0;
413 _total_fast_refill_waste = 0;
414 _max_fast_refill_waste = 0;
415 _total_slow_allocations = 0;
416 _max_slow_allocations = 0;
417 }
418
419 void GlobalTLABStats::publish() {
420 _allocating_threads_avg.sample(_allocating_threads);
421 if (UsePerfData) {
422 _perf_allocating_threads ->set_value(_allocating_threads);
423 _perf_total_refills ->set_value(_total_refills);
424 _perf_max_refills ->set_value(_max_refills);
425 _perf_allocation ->set_value(_total_allocation);
426 _perf_gc_waste ->set_value(_total_gc_waste);
427 _perf_max_gc_waste ->set_value(_max_gc_waste);
428 _perf_slow_refill_waste ->set_value(_total_slow_refill_waste);
429 _perf_max_slow_refill_waste->set_value(_max_slow_refill_waste);
430 _perf_fast_refill_waste ->set_value(_total_fast_refill_waste);
431 _perf_max_fast_refill_waste->set_value(_max_fast_refill_waste);
432 _perf_slow_allocations ->set_value(_total_slow_allocations);
433 _perf_max_slow_allocations ->set_value(_max_slow_allocations);
434 }
435 }
436
437 void GlobalTLABStats::print() {
438 Log(gc, tlab) log;
439 if (!log.is_debug()) {
440 return;
441 }
442
443 size_t waste = _total_gc_waste + _total_slow_refill_waste + _total_fast_refill_waste;
444 double waste_percent = percent_of(waste, _total_allocation);
445 log.debug("TLAB totals: thrds: %d refills: %d max: %d"
446 " slow allocs: %d max %d waste: %4.1f%%"
447 " gc: " SIZE_FORMAT "B max: " SIZE_FORMAT "B"
448 " slow: " SIZE_FORMAT "B max: " SIZE_FORMAT "B"
449 " fast: " SIZE_FORMAT "B max: " SIZE_FORMAT "B",
450 _allocating_threads,
451 _total_refills, _max_refills,
452 _total_slow_allocations, _max_slow_allocations,
453 waste_percent,
454 _total_gc_waste * HeapWordSize,
455 _max_gc_waste * HeapWordSize,
456 _total_slow_refill_waste * HeapWordSize,
457 _max_slow_refill_waste * HeapWordSize,
458 _total_fast_refill_waste * HeapWordSize,
459 _max_fast_refill_waste * HeapWordSize);
460 }