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 Universe::heap()->accumulate_statistics_all_gclabs();
57
58 // Publish new stats if some allocation occurred.
59 if (global_stats()->allocation() != 0) {
60 global_stats()->publish();
61 global_stats()->print();
62 }
63 }
64
65 void ThreadLocalAllocBuffer::accumulate_statistics() {
66 Thread* thread = myThread();
67 size_t capacity = Universe::heap()->tlab_capacity(thread);
68 size_t used = Universe::heap()->tlab_used(thread);
69
70 _gc_waste += (unsigned)remaining();
71 size_t total_allocated = _gclab ? thread->allocated_bytes_gclab() : thread->allocated_bytes();
72 size_t allocated_since_last_gc = total_allocated - _allocated_before_last_gc;
73 _allocated_before_last_gc = total_allocated;
74
75 print_stats("gc");
76
77 if (_number_of_refills > 0) {
78 // Update allocation history if a reasonable amount of eden was allocated.
79 bool update_allocation_history = used > 0.5 * capacity;
80
81 if (update_allocation_history) {
82 // Average the fraction of eden allocated in a tlab by this
83 // thread for use in the next resize operation.
84 // _gc_waste is not subtracted because it's included in
85 // "used".
86 // The result can be larger than 1.0 due to direct to old allocations.
87 // These allocations should ideally not be counted but since it is not possible
88 // to filter them out here we just cap the fraction to be at most 1.0.
89 double alloc_frac = MIN2(1.0, (double) allocated_since_last_gc / used);
90 _allocation_fraction.sample(alloc_frac);
91 }
92 global_stats()->update_allocating_threads();
93 global_stats()->update_number_of_refills(_number_of_refills);
94 global_stats()->update_allocation(_number_of_refills * desired_size());
95 global_stats()->update_gc_waste(_gc_waste);
96 global_stats()->update_slow_refill_waste(_slow_refill_waste);
97 global_stats()->update_fast_refill_waste(_fast_refill_waste);
98
99 } else {
100 assert(_number_of_refills == 0 && _fast_refill_waste == 0 &&
101 _slow_refill_waste == 0 && _gc_waste == 0,
102 "tlab stats == 0");
103 }
104 global_stats()->update_slow_allocations(_slow_allocations);
105 }
106
107 // Fills the current tlab with a dummy filler array to create
108 // an illusion of a contiguous Eden and optionally retires the tlab.
109 // Waste accounting should be done in caller as appropriate; see,
110 // for example, clear_before_allocation().
111 void ThreadLocalAllocBuffer::make_parsable(bool retire, bool zap) {
112 if (end() != NULL) {
113 invariants();
114
115 if (retire) {
116 if (_gclab) {
117 myThread()->incr_allocated_bytes_gclab(used_bytes());
118 } else {
119 myThread()->incr_allocated_bytes(used_bytes());
120 }
121 }
122
123 CollectedHeap::fill_with_object(top(), hard_end(), retire && zap);
124
125 if (retire || ZeroTLAB) { // "Reset" the TLAB
126 set_start(NULL);
127 set_top(NULL);
128 set_pf_top(NULL);
129 set_end(NULL);
130 }
131 }
132 assert(!(retire || ZeroTLAB) ||
133 (start() == NULL && end() == NULL && top() == NULL),
134 "TLAB must be reset");
135 }
136
137 void ThreadLocalAllocBuffer::resize_all_tlabs() {
138 if (ResizeTLAB) {
139 for (JavaThreadIteratorWithHandle jtiwh; JavaThread *thread = jtiwh.next(); ) {
140 thread->tlab().resize();
141 }
142 }
143 }
144
145 void ThreadLocalAllocBuffer::resize() {
146 // Compute the next tlab size using expected allocation amount
147 assert(ResizeTLAB, "Should not call this otherwise");
148 size_t alloc = (size_t)(_allocation_fraction.average() *
149 (Universe::heap()->tlab_capacity(myThread()) / HeapWordSize));
150 size_t new_size = alloc / _target_refills;
151
152 new_size = MIN2(MAX2(new_size, min_size()), max_size());
153
154 size_t aligned_new_size = align_object_size(new_size);
155
156 log_trace(gc, tlab)("TLAB new size: thread: " INTPTR_FORMAT " [id: %2d]"
157 " refills %d alloc: %8.6f desired_size: " SIZE_FORMAT " -> " SIZE_FORMAT,
158 p2i(myThread()), myThread()->osthread()->thread_id(),
159 _target_refills, _allocation_fraction.average(), desired_size(), aligned_new_size);
160
161 set_desired_size(aligned_new_size);
162 set_refill_waste_limit(initial_refill_waste_limit());
163 }
164
165 void ThreadLocalAllocBuffer::initialize_statistics() {
166 _number_of_refills = 0;
167 _fast_refill_waste = 0;
168 _slow_refill_waste = 0;
169 _gc_waste = 0;
170 _slow_allocations = 0;
171 }
172
173 void ThreadLocalAllocBuffer::fill(HeapWord* start,
174 HeapWord* top,
175 size_t new_size) {
176 _number_of_refills++;
177 print_stats("fill");
178 assert(top <= start + new_size - alignment_reserve(), "size too small");
179 initialize(start, top, start + new_size - alignment_reserve());
180
181 // Reset amount of internal fragmentation
182 set_refill_waste_limit(initial_refill_waste_limit());
183 }
184
185 void ThreadLocalAllocBuffer::initialize(HeapWord* start,
186 HeapWord* top,
187 HeapWord* end) {
188 set_start(start);
189 set_top(top);
190 set_pf_top(top);
191 set_end(end);
192 invariants();
193 }
194
195 void ThreadLocalAllocBuffer::initialize(bool gclab) {
196 _initialized = true;
197 _gclab = gclab;
198 initialize(NULL, // start
199 NULL, // top
200 NULL); // end
201
202 set_desired_size(initial_desired_size());
203
204 // Following check is needed because at startup the main
205 // thread is initialized before the heap is. The initialization for
206 // this thread is redone in startup_initialization below.
207 if (Universe::heap() != NULL) {
208 size_t capacity = Universe::heap()->tlab_capacity(myThread()) / HeapWordSize;
209 double alloc_frac = desired_size() * target_refills() / (double) capacity;
210 _allocation_fraction.sample(alloc_frac);
211 }
212
213 set_refill_waste_limit(initial_refill_waste_limit());
214
215 initialize_statistics();
216 }
217
218 void ThreadLocalAllocBuffer::startup_initialization() {
219
220 // Assuming each thread's active tlab is, on average,
221 // 1/2 full at a GC
222 _target_refills = 100 / (2 * TLABWasteTargetPercent);
223 _target_refills = MAX2(_target_refills, (unsigned)1U);
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(false);
255 Thread::current()->gclab().initialize(true);
256
257 log_develop_trace(gc, tlab)("TLAB min: " SIZE_FORMAT " initial: " SIZE_FORMAT " max: " SIZE_FORMAT,
258 min_size(), Thread::current()->tlab().initial_desired_size(), max_size());
259 }
260
261 size_t ThreadLocalAllocBuffer::initial_desired_size() {
262 size_t init_sz = 0;
263
264 if (TLABSize > 0) {
265 init_sz = TLABSize / HeapWordSize;
266 } else if (global_stats() != NULL) {
267 // Initial size is a function of the average number of allocating threads.
268 unsigned nof_threads = global_stats()->allocating_threads_avg();
269
270 init_sz = (Universe::heap()->tlab_capacity(myThread()) / HeapWordSize) /
271 (nof_threads * target_refills());
272 init_sz = align_object_size(init_sz);
273 }
274 init_sz = MIN2(MAX2(init_sz, min_size()), max_size());
275 return init_sz;
276 }
277
278 void ThreadLocalAllocBuffer::print_stats(const char* tag) {
279 Log(gc, tlab) log;
280 if (!log.is_trace()) {
281 return;
282 }
283
284 Thread* thrd = myThread();
285 size_t waste = _gc_waste + _slow_refill_waste + _fast_refill_waste;
286 size_t alloc = _number_of_refills * _desired_size;
287 double waste_percent = percent_of(waste, alloc);
288 size_t tlab_used = Universe::heap()->tlab_used(thrd);
289 log.trace("TLAB: %s thread: " INTPTR_FORMAT " [id: %2d]"
290 " desired_size: " SIZE_FORMAT "KB"
291 " slow allocs: %d refill waste: " SIZE_FORMAT "B"
292 " alloc:%8.5f %8.0fKB refills: %d waste %4.1f%% gc: %dB"
293 " slow: %dB fast: %dB",
294 tag, p2i(thrd), thrd->osthread()->thread_id(),
295 _desired_size / (K / HeapWordSize),
296 _slow_allocations, _refill_waste_limit * HeapWordSize,
297 _allocation_fraction.average(),
298 _allocation_fraction.average() * tlab_used / K,
299 _number_of_refills, waste_percent,
300 _gc_waste * HeapWordSize,
301 _slow_refill_waste * HeapWordSize,
302 _fast_refill_waste * HeapWordSize);
303 }
304
305 void ThreadLocalAllocBuffer::verify() {
306 HeapWord* p = start();
307 HeapWord* t = top();
308 HeapWord* prev_p = NULL;
309 while (p < t) {
310 oop(p)->verify();
311 prev_p = p;
312 p += oop(p)->size();
313 }
314 guarantee(p == top(), "end of last object must match end of space");
315 }
316
317 Thread* ThreadLocalAllocBuffer::myThread() {
318 ByteSize gclab_offset = Thread::gclab_start_offset();
319 ByteSize tlab_offset = Thread::tlab_start_offset();
320 ByteSize offs = _gclab ? gclab_offset : tlab_offset;
321 Thread* thread = (Thread*)(((char *)this) +
322 in_bytes(start_offset()) - in_bytes(offs));
323 #ifdef ASSERT
324 assert(this == (_gclab ? &thread->gclab() : &thread->tlab()), "must be");
325 #endif
326 return thread;
327 }
328
329 size_t ThreadLocalAllocBuffer::end_reserve() {
330 int reserve_size = typeArrayOopDesc::header_size(T_INT);
331 return MAX2(reserve_size, _reserve_for_allocation_prefetch);
332 }
333
334 void ThreadLocalAllocBuffer::rollback(size_t size) {
335 HeapWord* old_top = top();
336 if (old_top != NULL) { // Pathological case: we accept that we can't rollback.
337 set_top(old_top - size);
338 }
339 }
340
341
342 GlobalTLABStats::GlobalTLABStats() :
343 _allocating_threads_avg(TLABAllocationWeight) {
344
345 initialize();
346
347 _allocating_threads_avg.sample(1); // One allocating thread at startup
348
349 if (UsePerfData) {
350
351 EXCEPTION_MARK;
352 ResourceMark rm;
353
354 char* cname = PerfDataManager::counter_name("tlab", "allocThreads");
355 _perf_allocating_threads =
356 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_None, CHECK);
357
358 cname = PerfDataManager::counter_name("tlab", "fills");
359 _perf_total_refills =
360 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_None, CHECK);
361
362 cname = PerfDataManager::counter_name("tlab", "maxFills");
363 _perf_max_refills =
364 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_None, CHECK);
365
366 cname = PerfDataManager::counter_name("tlab", "alloc");
367 _perf_allocation =
368 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_Bytes, CHECK);
369
370 cname = PerfDataManager::counter_name("tlab", "gcWaste");
371 _perf_gc_waste =
372 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_Bytes, CHECK);
373
374 cname = PerfDataManager::counter_name("tlab", "maxGcWaste");
375 _perf_max_gc_waste =
376 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_Bytes, CHECK);
377
378 cname = PerfDataManager::counter_name("tlab", "slowWaste");
379 _perf_slow_refill_waste =
380 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_Bytes, CHECK);
381
382 cname = PerfDataManager::counter_name("tlab", "maxSlowWaste");
383 _perf_max_slow_refill_waste =
384 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_Bytes, CHECK);
385
386 cname = PerfDataManager::counter_name("tlab", "fastWaste");
387 _perf_fast_refill_waste =
388 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_Bytes, CHECK);
389
390 cname = PerfDataManager::counter_name("tlab", "maxFastWaste");
391 _perf_max_fast_refill_waste =
392 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_Bytes, CHECK);
393
394 cname = PerfDataManager::counter_name("tlab", "slowAlloc");
395 _perf_slow_allocations =
396 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_None, CHECK);
397
398 cname = PerfDataManager::counter_name("tlab", "maxSlowAlloc");
399 _perf_max_slow_allocations =
400 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_None, CHECK);
401 }
402 }
403
404 void GlobalTLABStats::initialize() {
405 // Clear counters summarizing info from all threads
406 _allocating_threads = 0;
407 _total_refills = 0;
408 _max_refills = 0;
409 _total_allocation = 0;
410 _total_gc_waste = 0;
411 _max_gc_waste = 0;
412 _total_slow_refill_waste = 0;
413 _max_slow_refill_waste = 0;
414 _total_fast_refill_waste = 0;
415 _max_fast_refill_waste = 0;
416 _total_slow_allocations = 0;
417 _max_slow_allocations = 0;
418 }
419
420 void GlobalTLABStats::publish() {
421 _allocating_threads_avg.sample(_allocating_threads);
422 if (UsePerfData) {
423 _perf_allocating_threads ->set_value(_allocating_threads);
424 _perf_total_refills ->set_value(_total_refills);
425 _perf_max_refills ->set_value(_max_refills);
426 _perf_allocation ->set_value(_total_allocation);
427 _perf_gc_waste ->set_value(_total_gc_waste);
428 _perf_max_gc_waste ->set_value(_max_gc_waste);
429 _perf_slow_refill_waste ->set_value(_total_slow_refill_waste);
430 _perf_max_slow_refill_waste->set_value(_max_slow_refill_waste);
431 _perf_fast_refill_waste ->set_value(_total_fast_refill_waste);
432 _perf_max_fast_refill_waste->set_value(_max_fast_refill_waste);
433 _perf_slow_allocations ->set_value(_total_slow_allocations);
434 _perf_max_slow_allocations ->set_value(_max_slow_allocations);
435 }
436 }
437
438 void GlobalTLABStats::print() {
439 Log(gc, tlab) log;
440 if (!log.is_debug()) {
441 return;
442 }
443
444 size_t waste = _total_gc_waste + _total_slow_refill_waste + _total_fast_refill_waste;
445 double waste_percent = percent_of(waste, _total_allocation);
446 log.debug("TLAB totals: thrds: %d refills: %d max: %d"
447 " slow allocs: %d max %d waste: %4.1f%%"
448 " gc: " SIZE_FORMAT "B max: " SIZE_FORMAT "B"
449 " slow: " SIZE_FORMAT "B max: " SIZE_FORMAT "B"
450 " fast: " SIZE_FORMAT "B max: " SIZE_FORMAT "B",
451 _allocating_threads,
452 _total_refills, _max_refills,
453 _total_slow_allocations, _max_slow_allocations,
454 waste_percent,
455 _total_gc_waste * HeapWordSize,
456 _max_gc_waste * HeapWordSize,
457 _total_slow_refill_waste * HeapWordSize,
458 _max_slow_refill_waste * HeapWordSize,
459 _total_fast_refill_waste * HeapWordSize,
460 _max_fast_refill_waste * HeapWordSize);
461 }