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 void ThreadLocalAllocBuffer::accumulate_statistics_before_gc() {
49 global_stats()->initialize();
50
51 for (JavaThreadIteratorWithHandle jtiwh; JavaThread *thread = jtiwh.next(); ) {
52 thread->tlab().accumulate_statistics();
53 thread->tlab().initialize_statistics();
54 }
55
56 // Publish new stats if some allocation occurred.
57 if (global_stats()->allocation() != 0) {
58 global_stats()->publish();
59 global_stats()->print();
60 }
61 }
62
63 void ThreadLocalAllocBuffer::accumulate_statistics() {
64 Thread* thread = myThread();
65 size_t capacity = Universe::heap()->tlab_capacity(thread);
66 size_t used = Universe::heap()->tlab_used(thread);
67
68 _gc_waste += (unsigned)remaining();
69 size_t total_allocated = thread->allocated_bytes();
70 size_t allocated_since_last_gc = total_allocated - _allocated_before_last_gc;
71 _allocated_before_last_gc = total_allocated;
72
73 print_stats("gc");
74
75 if (_number_of_refills > 0) {
76 // Update allocation history if a reasonable amount of eden was allocated.
77 bool update_allocation_history = used > 0.5 * capacity;
78
79 if (update_allocation_history) {
80 // Average the fraction of eden allocated in a tlab by this
81 // thread for use in the next resize operation.
82 // _gc_waste is not subtracted because it's included in
83 // "used".
84 // The result can be larger than 1.0 due to direct to old allocations.
85 // These allocations should ideally not be counted but since it is not possible
86 // to filter them out here we just cap the fraction to be at most 1.0.
87 double alloc_frac = MIN2(1.0, (double) allocated_since_last_gc / used);
88 _allocation_fraction.sample(alloc_frac);
89 }
90 global_stats()->update_allocating_threads();
91 global_stats()->update_number_of_refills(_number_of_refills);
92 global_stats()->update_allocation(_number_of_refills * desired_size());
93 global_stats()->update_gc_waste(_gc_waste);
94 global_stats()->update_slow_refill_waste(_slow_refill_waste);
95 global_stats()->update_fast_refill_waste(_fast_refill_waste);
96
97 } else {
98 assert(_number_of_refills == 0 && _fast_refill_waste == 0 &&
99 _slow_refill_waste == 0 && _gc_waste == 0,
100 "tlab stats == 0");
101 }
102 global_stats()->update_slow_allocations(_slow_allocations);
103 }
104
105 // Fills the current tlab with a dummy filler array to create
106 // an illusion of a contiguous Eden and optionally retires the tlab.
107 // Waste accounting should be done in caller as appropriate; see,
108 // for example, clear_before_allocation().
109 void ThreadLocalAllocBuffer::make_parsable(bool retire, bool zap) {
110 if (end() != NULL) {
111 invariants();
112
113 if (retire) {
114 myThread()->incr_allocated_bytes(used_bytes());
115 }
116
117 CollectedHeap::fill_with_object(top(), hard_end(), retire && zap);
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 if (ResizeTLAB) {
133 for (JavaThreadIteratorWithHandle jtiwh; JavaThread *thread = jtiwh.next(); ) {
134 thread->tlab().resize();
135 }
136 }
137 }
138
139 void ThreadLocalAllocBuffer::resize() {
140 // Compute the next tlab size using expected allocation amount
141 assert(ResizeTLAB, "Should not call this otherwise");
142 size_t alloc = (size_t)(_allocation_fraction.average() *
143 (Universe::heap()->tlab_capacity(myThread()) / HeapWordSize));
144 size_t new_size = alloc / _target_refills;
145
146 new_size = MIN2(MAX2(new_size, min_size()), max_size());
147
148 size_t aligned_new_size = align_object_size(new_size);
149
150 log_trace(gc, tlab)("TLAB new size: thread: " INTPTR_FORMAT " [id: %2d]"
151 " refills %d alloc: %8.6f desired_size: " SIZE_FORMAT " -> " SIZE_FORMAT,
152 p2i(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 set_refill_waste_limit(initial_refill_waste_limit());
157 }
158
159 void ThreadLocalAllocBuffer::initialize_statistics() {
160 _number_of_refills = 0;
161 _fast_refill_waste = 0;
162 _slow_refill_waste = 0;
163 _gc_waste = 0;
164 _slow_allocations = 0;
165 }
166
167 void ThreadLocalAllocBuffer::fill(HeapWord* start,
168 HeapWord* top,
169 size_t new_size) {
170 _number_of_refills++;
171 print_stats("fill");
172 assert(top <= start + new_size - alignment_reserve(), "size too small");
173 initialize(start, top, start + new_size - alignment_reserve());
174
175 // Reset amount of internal fragmentation
176 set_refill_waste_limit(initial_refill_waste_limit());
177 }
178
179 void ThreadLocalAllocBuffer::initialize(HeapWord* start,
180 HeapWord* top,
181 HeapWord* end) {
182 set_start(start);
183 set_top(top);
184 set_pf_top(top);
185 set_end(end);
186 invariants();
187 }
188
189 void ThreadLocalAllocBuffer::initialize() {
190 initialize(NULL, // start
191 NULL, // top
192 NULL); // end
193
194 set_desired_size(initial_desired_size());
195
196 // Following check is needed because at startup the main
197 // thread is initialized before the heap is. The initialization for
198 // this thread is redone in startup_initialization below.
199 if (Universe::heap() != NULL) {
200 size_t capacity = Universe::heap()->tlab_capacity(myThread()) / HeapWordSize;
201 double alloc_frac = desired_size() * target_refills() / (double) capacity;
202 _allocation_fraction.sample(alloc_frac);
203 }
204
205 set_refill_waste_limit(initial_refill_waste_limit());
206
207 initialize_statistics();
208 }
209
210 void ThreadLocalAllocBuffer::startup_initialization() {
211
212 // Assuming each thread's active tlab is, on average,
213 // 1/2 full at a GC
214 _target_refills = 100 / (2 * TLABWasteTargetPercent);
215 _target_refills = MAX2(_target_refills, (unsigned)1U);
216
217 _global_stats = new GlobalTLABStats();
218
219 #ifdef COMPILER2
220 // If the C2 compiler is present, extra space is needed at the end of
221 // TLABs, otherwise prefetching instructions generated by the C2
222 // compiler will fault (due to accessing memory outside of heap).
223 // The amount of space is the max of the number of lines to
224 // prefetch for array and for instance allocations. (Extra space must be
225 // reserved to accommodate both types of allocations.)
226 //
227 // Only SPARC-specific BIS instructions are known to fault. (Those
228 // instructions are generated if AllocatePrefetchStyle==3 and
229 // AllocatePrefetchInstr==1). To be on the safe side, however,
230 // extra space is reserved for all combinations of
231 // AllocatePrefetchStyle and AllocatePrefetchInstr.
232 //
233 // If the C2 compiler is not present, no space is reserved.
234
235 // +1 for rounding up to next cache line, +1 to be safe
236 if (is_server_compilation_mode_vm()) {
237 int lines = MAX2(AllocatePrefetchLines, AllocateInstancePrefetchLines) + 2;
238 _reserve_for_allocation_prefetch = (AllocatePrefetchDistance + AllocatePrefetchStepSize * lines) /
239 (int)HeapWordSize;
240 }
241 #endif
242
243 // During jvm startup, the main thread is initialized
244 // before the heap is initialized. So reinitialize it now.
245 guarantee(Thread::current()->is_Java_thread(), "tlab initialization thread not Java thread");
246 Thread::current()->tlab().initialize();
247
248 log_develop_trace(gc, tlab)("TLAB min: " SIZE_FORMAT " initial: " SIZE_FORMAT " max: " SIZE_FORMAT,
249 min_size(), Thread::current()->tlab().initial_desired_size(), max_size());
250 }
251
252 size_t ThreadLocalAllocBuffer::initial_desired_size() {
253 size_t init_sz = 0;
254
255 if (TLABSize > 0) {
256 init_sz = TLABSize / HeapWordSize;
257 } else if (global_stats() != NULL) {
258 // Initial size is a function of the average number of allocating threads.
259 unsigned nof_threads = global_stats()->allocating_threads_avg();
260
261 init_sz = (Universe::heap()->tlab_capacity(myThread()) / HeapWordSize) /
262 (nof_threads * target_refills());
263 init_sz = align_object_size(init_sz);
264 }
265 init_sz = MIN2(MAX2(init_sz, min_size()), max_size());
266 return init_sz;
267 }
268
269 void ThreadLocalAllocBuffer::print_stats(const char* tag) {
270 Log(gc, tlab) log;
271 if (!log.is_trace()) {
272 return;
273 }
274
275 Thread* thrd = myThread();
276 size_t waste = _gc_waste + _slow_refill_waste + _fast_refill_waste;
277 size_t alloc = _number_of_refills * _desired_size;
278 double waste_percent = percent_of(waste, alloc);
279 size_t tlab_used = Universe::heap()->tlab_used(thrd);
280 log.trace("TLAB: %s thread: " INTPTR_FORMAT " [id: %2d]"
281 " desired_size: " SIZE_FORMAT "KB"
282 " slow allocs: %d refill waste: " SIZE_FORMAT "B"
283 " alloc:%8.5f %8.0fKB refills: %d waste %4.1f%% gc: %dB"
284 " slow: %dB fast: %dB",
285 tag, p2i(thrd), thrd->osthread()->thread_id(),
286 _desired_size / (K / HeapWordSize),
287 _slow_allocations, _refill_waste_limit * HeapWordSize,
288 _allocation_fraction.average(),
289 _allocation_fraction.average() * tlab_used / K,
290 _number_of_refills, waste_percent,
291 _gc_waste * HeapWordSize,
292 _slow_refill_waste * HeapWordSize,
293 _fast_refill_waste * HeapWordSize);
294 }
295
296 void ThreadLocalAllocBuffer::verify() {
297 HeapWord* p = start();
298 HeapWord* t = top();
299 HeapWord* prev_p = NULL;
300 while (p < t) {
301 oop(p)->verify();
302 prev_p = p;
303 p += oop(p)->size();
304 }
305 guarantee(p == top(), "end of last object must match end of space");
306 }
307
308 Thread* ThreadLocalAllocBuffer::myThread() {
309 return (Thread*)(((char *)this) +
310 in_bytes(start_offset()) -
311 in_bytes(Thread::tlab_start_offset()));
312 }
313
314
315 GlobalTLABStats::GlobalTLABStats() :
316 _allocating_threads_avg(TLABAllocationWeight) {
317
318 initialize();
319
320 _allocating_threads_avg.sample(1); // One allocating thread at startup
321
322 if (UsePerfData) {
323
324 EXCEPTION_MARK;
325 ResourceMark rm;
326
327 char* cname = PerfDataManager::counter_name("tlab", "allocThreads");
328 _perf_allocating_threads =
329 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_None, CHECK);
330
331 cname = PerfDataManager::counter_name("tlab", "fills");
332 _perf_total_refills =
333 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_None, CHECK);
334
335 cname = PerfDataManager::counter_name("tlab", "maxFills");
336 _perf_max_refills =
337 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_None, CHECK);
338
339 cname = PerfDataManager::counter_name("tlab", "alloc");
340 _perf_allocation =
341 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_Bytes, CHECK);
342
343 cname = PerfDataManager::counter_name("tlab", "gcWaste");
344 _perf_gc_waste =
345 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_Bytes, CHECK);
346
347 cname = PerfDataManager::counter_name("tlab", "maxGcWaste");
348 _perf_max_gc_waste =
349 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_Bytes, CHECK);
350
351 cname = PerfDataManager::counter_name("tlab", "slowWaste");
352 _perf_slow_refill_waste =
353 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_Bytes, CHECK);
354
355 cname = PerfDataManager::counter_name("tlab", "maxSlowWaste");
356 _perf_max_slow_refill_waste =
357 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_Bytes, CHECK);
358
359 cname = PerfDataManager::counter_name("tlab", "fastWaste");
360 _perf_fast_refill_waste =
361 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_Bytes, CHECK);
362
363 cname = PerfDataManager::counter_name("tlab", "maxFastWaste");
364 _perf_max_fast_refill_waste =
365 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_Bytes, CHECK);
366
367 cname = PerfDataManager::counter_name("tlab", "slowAlloc");
368 _perf_slow_allocations =
369 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_None, CHECK);
370
371 cname = PerfDataManager::counter_name("tlab", "maxSlowAlloc");
372 _perf_max_slow_allocations =
373 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_None, CHECK);
374 }
375 }
376
377 void GlobalTLABStats::initialize() {
378 // Clear counters summarizing info from all threads
379 _allocating_threads = 0;
380 _total_refills = 0;
381 _max_refills = 0;
382 _total_allocation = 0;
383 _total_gc_waste = 0;
384 _max_gc_waste = 0;
385 _total_slow_refill_waste = 0;
386 _max_slow_refill_waste = 0;
387 _total_fast_refill_waste = 0;
388 _max_fast_refill_waste = 0;
389 _total_slow_allocations = 0;
390 _max_slow_allocations = 0;
391 }
392
393 void GlobalTLABStats::publish() {
394 _allocating_threads_avg.sample(_allocating_threads);
395 if (UsePerfData) {
396 _perf_allocating_threads ->set_value(_allocating_threads);
397 _perf_total_refills ->set_value(_total_refills);
398 _perf_max_refills ->set_value(_max_refills);
399 _perf_allocation ->set_value(_total_allocation);
400 _perf_gc_waste ->set_value(_total_gc_waste);
401 _perf_max_gc_waste ->set_value(_max_gc_waste);
402 _perf_slow_refill_waste ->set_value(_total_slow_refill_waste);
403 _perf_max_slow_refill_waste->set_value(_max_slow_refill_waste);
404 _perf_fast_refill_waste ->set_value(_total_fast_refill_waste);
405 _perf_max_fast_refill_waste->set_value(_max_fast_refill_waste);
406 _perf_slow_allocations ->set_value(_total_slow_allocations);
407 _perf_max_slow_allocations ->set_value(_max_slow_allocations);
408 }
409 }
410
411 void GlobalTLABStats::print() {
412 Log(gc, tlab) log;
413 if (!log.is_debug()) {
414 return;
415 }
416
417 size_t waste = _total_gc_waste + _total_slow_refill_waste + _total_fast_refill_waste;
418 double waste_percent = percent_of(waste, _total_allocation);
419 log.debug("TLAB totals: thrds: %d refills: %d max: %d"
420 " slow allocs: %d max %d waste: %4.1f%%"
421 " gc: " SIZE_FORMAT "B max: " SIZE_FORMAT "B"
422 " slow: " SIZE_FORMAT "B max: " SIZE_FORMAT "B"
423 " fast: " SIZE_FORMAT "B max: " SIZE_FORMAT "B",
424 _allocating_threads,
425 _total_refills, _max_refills,
426 _total_slow_allocations, _max_slow_allocations,
427 waste_percent,
428 _total_gc_waste * HeapWordSize,
429 _max_gc_waste * HeapWordSize,
430 _total_slow_refill_waste * HeapWordSize,
431 _max_slow_refill_waste * HeapWordSize,
432 _total_fast_refill_waste * HeapWordSize,
433 _max_fast_refill_waste * HeapWordSize);
434 }
--- EOF ---