1 /* 2 * Copyright (c) 1999, 2019, 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/collectedHeap.hpp" 27 #include "gc/shared/threadLocalAllocBuffer.inline.hpp" 28 #include "logging/log.hpp" 29 #include "memory/resourceArea.hpp" 30 #include "memory/universe.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 size_t ThreadLocalAllocBuffer::_max_size = 0; 37 int ThreadLocalAllocBuffer::_reserve_for_allocation_prefetch = 0; 38 unsigned int ThreadLocalAllocBuffer::_target_refills = 0; 39 40 size_t ThreadLocalAllocBuffer::remaining() { 41 if (end() == NULL) { 42 return 0; 43 } 44 45 return pointer_delta(hard_end(), top()); 46 } 47 48 void ThreadLocalAllocBuffer::accumulate_and_reset_statistics(ThreadLocalAllocStats* stats) { 49 Thread* thr = thread(); 50 size_t capacity = Universe::heap()->tlab_capacity(thr); 51 size_t used = Universe::heap()->tlab_used(thr); 52 53 _gc_waste += (unsigned)remaining(); 54 size_t total_allocated = thr->allocated_bytes(); 55 size_t allocated_since_last_gc = total_allocated - _allocated_before_last_gc; 56 _allocated_before_last_gc = total_allocated; 57 58 print_stats("gc"); 59 60 if (_number_of_refills > 0) { 61 // Update allocation history if a reasonable amount of eden was allocated. 62 bool update_allocation_history = used > 0.5 * capacity; 63 64 if (update_allocation_history) { 65 // Average the fraction of eden allocated in a tlab by this 66 // thread for use in the next resize operation. 67 // _gc_waste is not subtracted because it's included in 68 // "used". 69 // The result can be larger than 1.0 due to direct to old allocations. 70 // These allocations should ideally not be counted but since it is not possible 71 // to filter them out here we just cap the fraction to be at most 1.0. 72 double alloc_frac = MIN2(1.0, (double) allocated_since_last_gc / used); 73 _allocation_fraction.sample(alloc_frac); 74 } 75 76 stats->update_fast_allocations(_number_of_refills, 77 _allocated_size, 78 _gc_waste, 79 _refill_waste); 80 } else { 81 assert(_number_of_refills == 0 && _refill_waste == 0 && _gc_waste == 0, 82 "tlab stats == 0"); 83 } 84 85 stats->update_slow_allocations(_slow_allocations); 86 87 reset_statistics(); 88 } 89 90 void ThreadLocalAllocBuffer::insert_filler() { 91 assert(end() != NULL, "Must not be retired"); 92 if (top() < hard_end()) { 93 Universe::heap()->fill_with_dummy_object(top(), hard_end(), true); 94 } 95 } 96 97 void ThreadLocalAllocBuffer::make_parsable() { 98 if (end() != NULL) { 99 invariants(); 100 if (ZeroTLAB) { 101 retire(); 102 } else { 103 insert_filler(); 104 } 105 } 106 } 107 108 void ThreadLocalAllocBuffer::retire(ThreadLocalAllocStats* stats) { 109 if (stats != NULL) { 110 accumulate_and_reset_statistics(stats); 111 } 112 113 if (end() != NULL) { 114 invariants(); 115 thread()->incr_allocated_bytes(used_bytes()); 116 insert_filler(); 117 initialize(NULL, NULL, NULL); 118 } 119 } 120 121 void ThreadLocalAllocBuffer::retire_before_allocation() { 122 _refill_waste += (unsigned int)remaining(); 123 retire(); 124 } 125 126 void ThreadLocalAllocBuffer::resize() { 127 // Compute the next tlab size using expected allocation amount 128 assert(ResizeTLAB, "Should not call this otherwise"); 129 size_t alloc = (size_t)(_allocation_fraction.average() * 130 (Universe::heap()->tlab_capacity(thread()) / HeapWordSize)); 131 size_t new_size = alloc / _target_refills; 132 133 new_size = MIN2(MAX2(new_size, min_size()), max_size()); 134 135 size_t aligned_new_size = align_object_size(new_size); 136 137 log_trace(gc, tlab)("TLAB new size: thread: " INTPTR_FORMAT " [id: %2d]" 138 " refills %d alloc: %8.6f desired_size: " SIZE_FORMAT " -> " SIZE_FORMAT, 139 p2i(thread()), thread()->osthread()->thread_id(), 140 _target_refills, _allocation_fraction.average(), desired_size(), aligned_new_size); 141 142 set_desired_size(aligned_new_size); 143 set_refill_waste_limit(initial_refill_waste_limit()); 144 } 145 146 void ThreadLocalAllocBuffer::reset_statistics() { 147 _number_of_refills = 0; 148 _refill_waste = 0; 149 _gc_waste = 0; 150 _slow_allocations = 0; 151 _allocated_size = 0; 152 } 153 154 void ThreadLocalAllocBuffer::fill(HeapWord* start, 155 HeapWord* top, 156 size_t new_size) { 157 _number_of_refills++; 158 _allocated_size += new_size; 159 print_stats("fill"); 160 assert(top <= start + new_size - alignment_reserve(), "size too small"); 161 162 initialize(start, top, start + new_size - alignment_reserve()); 163 164 // Reset amount of internal fragmentation 165 set_refill_waste_limit(initial_refill_waste_limit()); 166 } 167 168 void ThreadLocalAllocBuffer::initialize(HeapWord* start, 169 HeapWord* top, 170 HeapWord* end) { 171 set_start(start); 172 set_top(top); 173 set_pf_top(top); 174 set_end(end); 175 set_allocation_end(end); 176 invariants(); 177 } 178 179 void ThreadLocalAllocBuffer::initialize() { 180 initialize(NULL, // start 181 NULL, // top 182 NULL); // end 183 184 set_desired_size(initial_desired_size()); 185 186 size_t capacity = Universe::heap()->tlab_capacity(thread()) / HeapWordSize; 187 double alloc_frac = desired_size() * target_refills() / (double) capacity; 188 _allocation_fraction.sample(alloc_frac); 189 190 set_refill_waste_limit(initial_refill_waste_limit()); 191 192 reset_statistics(); 193 } 194 195 void ThreadLocalAllocBuffer::startup_initialization() { 196 ThreadLocalAllocStats::initialize(); 197 198 // Assuming each thread's active tlab is, on average, 199 // 1/2 full at a GC 200 _target_refills = 100 / (2 * TLABWasteTargetPercent); 201 // We need to set initial target refills to 2 to avoid a GC which causes VM 202 // abort during VM initialization. 203 _target_refills = MAX2(_target_refills, 2U); 204 205 #ifdef COMPILER2 206 // If the C2 compiler is present, extra space is needed at the end of 207 // TLABs, otherwise prefetching instructions generated by the C2 208 // compiler will fault (due to accessing memory outside of heap). 209 // The amount of space is the max of the number of lines to 210 // prefetch for array and for instance allocations. (Extra space must be 211 // reserved to accommodate both types of allocations.) 212 // 213 // Only SPARC-specific BIS instructions are known to fault. (Those 214 // instructions are generated if AllocatePrefetchStyle==3 and 215 // AllocatePrefetchInstr==1). To be on the safe side, however, 216 // extra space is reserved for all combinations of 217 // AllocatePrefetchStyle and AllocatePrefetchInstr. 218 // 219 // If the C2 compiler is not present, no space is reserved. 220 221 // +1 for rounding up to next cache line, +1 to be safe 222 if (is_server_compilation_mode_vm()) { 223 int lines = MAX2(AllocatePrefetchLines, AllocateInstancePrefetchLines) + 2; 224 _reserve_for_allocation_prefetch = (AllocatePrefetchDistance + AllocatePrefetchStepSize * lines) / 225 (int)HeapWordSize; 226 } 227 #endif 228 229 // During jvm startup, the main thread is initialized 230 // before the heap is initialized. So reinitialize it now. 231 guarantee(Thread::current()->is_Java_thread(), "tlab initialization thread not Java thread"); 232 Thread::current()->tlab().initialize(); 233 234 log_develop_trace(gc, tlab)("TLAB min: " SIZE_FORMAT " initial: " SIZE_FORMAT " max: " SIZE_FORMAT, 235 min_size(), Thread::current()->tlab().initial_desired_size(), max_size()); 236 } 237 238 size_t ThreadLocalAllocBuffer::initial_desired_size() { 239 size_t init_sz = 0; 240 241 if (TLABSize > 0) { 242 init_sz = TLABSize / HeapWordSize; 243 } else { 244 // Initial size is a function of the average number of allocating threads. 245 unsigned int nof_threads = ThreadLocalAllocStats::allocating_threads_avg(); 246 247 init_sz = (Universe::heap()->tlab_capacity(thread()) / HeapWordSize) / 248 (nof_threads * target_refills()); 249 init_sz = align_object_size(init_sz); 250 } 251 init_sz = MIN2(MAX2(init_sz, min_size()), max_size()); 252 return init_sz; 253 } 254 255 void ThreadLocalAllocBuffer::print_stats(const char* tag) { 256 Log(gc, tlab) log; 257 if (!log.is_trace()) { 258 return; 259 } 260 261 Thread* thrd = thread(); 262 size_t waste = _gc_waste + _refill_waste; 263 double waste_percent = percent_of(waste, _allocated_size); 264 size_t tlab_used = Universe::heap()->tlab_used(thrd); 265 log.trace("TLAB: %s thread: " INTPTR_FORMAT " [id: %2d]" 266 " desired_size: " SIZE_FORMAT "KB" 267 " slow allocs: %d refill waste: " SIZE_FORMAT "B" 268 " alloc:%8.5f %8.0fKB refills: %d waste %4.1f%% gc: %dB" 269 " refill: %dB", 270 tag, p2i(thrd), thrd->osthread()->thread_id(), 271 _desired_size / (K / HeapWordSize), 272 _slow_allocations, _refill_waste_limit * HeapWordSize, 273 _allocation_fraction.average(), 274 _allocation_fraction.average() * tlab_used / K, 275 _number_of_refills, waste_percent, 276 _gc_waste * HeapWordSize, 277 _refill_waste * HeapWordSize); 278 } 279 280 void ThreadLocalAllocBuffer::set_sample_end() { 281 size_t heap_words_remaining = pointer_delta(_end, _top); 282 size_t bytes_until_sample = thread()->heap_sampler().bytes_until_sample(); 283 size_t words_until_sample = bytes_until_sample / HeapWordSize; 284 285 if (heap_words_remaining > words_until_sample) { 286 HeapWord* new_end = _top + words_until_sample; 287 set_end(new_end); 288 _bytes_since_last_sample_point = bytes_until_sample; 289 } else { 290 _bytes_since_last_sample_point = heap_words_remaining * HeapWordSize; 291 } 292 } 293 294 Thread* ThreadLocalAllocBuffer::thread() { 295 return (Thread*)(((char*)this) + in_bytes(start_offset()) - in_bytes(Thread::tlab_start_offset())); 296 } 297 298 void ThreadLocalAllocBuffer::set_back_allocation_end() { 299 _end = _allocation_end; 300 } 301 302 HeapWord* ThreadLocalAllocBuffer::hard_end() { 303 return _allocation_end + alignment_reserve(); 304 } 305 306 PerfVariable* ThreadLocalAllocStats::_perf_allocating_threads; 307 PerfVariable* ThreadLocalAllocStats::_perf_total_refills; 308 PerfVariable* ThreadLocalAllocStats::_perf_max_refills; 309 PerfVariable* ThreadLocalAllocStats::_perf_total_allocations; 310 PerfVariable* ThreadLocalAllocStats::_perf_total_gc_waste; 311 PerfVariable* ThreadLocalAllocStats::_perf_max_gc_waste; 312 PerfVariable* ThreadLocalAllocStats::_perf_total_refill_waste; 313 PerfVariable* ThreadLocalAllocStats::_perf_max_refill_waste; 314 PerfVariable* ThreadLocalAllocStats::_perf_total_slow_allocations; 315 PerfVariable* ThreadLocalAllocStats::_perf_max_slow_allocations; 316 AdaptiveWeightedAverage ThreadLocalAllocStats::_allocating_threads_avg(0); 317 318 static PerfVariable* create_perf_variable(const char* name, PerfData::Units unit, TRAPS) { 319 ResourceMark rm; 320 return PerfDataManager::create_variable(SUN_GC, PerfDataManager::counter_name("tlab", name), unit, THREAD); 321 } 322 323 void ThreadLocalAllocStats::initialize() { 324 _allocating_threads_avg = AdaptiveWeightedAverage(TLABAllocationWeight); 325 _allocating_threads_avg.sample(1); // One allocating thread at startup 326 327 if (UsePerfData) { 328 EXCEPTION_MARK; 329 _perf_allocating_threads = create_perf_variable("allocThreads", PerfData::U_None, CHECK); 330 _perf_total_refills = create_perf_variable("fills", PerfData::U_None, CHECK); 331 _perf_max_refills = create_perf_variable("maxFills", PerfData::U_None, CHECK); 332 _perf_total_allocations = create_perf_variable("alloc", PerfData::U_Bytes, CHECK); 333 _perf_total_gc_waste = create_perf_variable("gcWaste", PerfData::U_Bytes, CHECK); 334 _perf_max_gc_waste = create_perf_variable("maxGcWaste", PerfData::U_Bytes, CHECK); 335 _perf_total_refill_waste = create_perf_variable("refillWaste", PerfData::U_Bytes, CHECK); 336 _perf_max_refill_waste = create_perf_variable("maxRefillWaste", PerfData::U_Bytes, CHECK); 337 _perf_total_slow_allocations = create_perf_variable("slowAlloc", PerfData::U_None, CHECK); 338 _perf_max_slow_allocations = create_perf_variable("maxSlowAlloc", PerfData::U_None, CHECK); 339 } 340 } 341 342 ThreadLocalAllocStats::ThreadLocalAllocStats() : 343 _allocating_threads(0), 344 _total_refills(0), 345 _max_refills(0), 346 _total_allocations(0), 347 _total_gc_waste(0), 348 _max_gc_waste(0), 349 _total_refill_waste(0), 350 _max_refill_waste(0), 351 _total_slow_allocations(0), 352 _max_slow_allocations(0) {} 353 354 unsigned int ThreadLocalAllocStats::allocating_threads_avg() { 355 return MAX2((unsigned int)(_allocating_threads_avg.average() + 0.5), 1U); 356 } 357 358 void ThreadLocalAllocStats::update_fast_allocations(unsigned int refills, 359 size_t allocations, 360 size_t gc_waste, 361 size_t refill_waste) { 362 _allocating_threads += 1; 363 _total_refills += refills; 364 _max_refills = MAX2(_max_refills, refills); 365 _total_allocations += allocations; 366 _total_gc_waste += gc_waste; 367 _max_gc_waste = MAX2(_max_gc_waste, gc_waste); 368 _total_refill_waste += refill_waste; 369 _max_refill_waste = MAX2(_max_refill_waste, refill_waste); 370 } 371 372 void ThreadLocalAllocStats::update_slow_allocations(unsigned int allocations) { 373 _total_slow_allocations += allocations; 374 _max_slow_allocations = MAX2(_max_slow_allocations, allocations); 375 } 376 377 void ThreadLocalAllocStats::update(const ThreadLocalAllocStats& other) { 378 _allocating_threads += other._allocating_threads; 379 _total_refills += other._total_refills; 380 _max_refills = MAX2(_max_refills, other._max_refills); 381 _total_allocations += other._total_allocations; 382 _total_gc_waste += other._total_gc_waste; 383 _max_gc_waste = MAX2(_max_gc_waste, other._max_gc_waste); 384 _total_refill_waste += other._total_refill_waste; 385 _max_refill_waste = MAX2(_max_refill_waste, other._max_refill_waste); 386 _total_slow_allocations += other._total_slow_allocations; 387 _max_slow_allocations = MAX2(_max_slow_allocations, other._max_slow_allocations); 388 } 389 390 void ThreadLocalAllocStats::reset() { 391 _allocating_threads = 0; 392 _total_refills = 0; 393 _max_refills = 0; 394 _total_allocations = 0; 395 _total_gc_waste = 0; 396 _max_gc_waste = 0; 397 _total_refill_waste = 0; 398 _max_refill_waste = 0; 399 _total_slow_allocations = 0; 400 _max_slow_allocations = 0; 401 } 402 403 void ThreadLocalAllocStats::publish() { 404 if (_total_allocations == 0) { 405 return; 406 } 407 408 _allocating_threads_avg.sample(_allocating_threads); 409 410 const size_t waste = _total_gc_waste + _total_refill_waste; 411 const double waste_percent = percent_of(waste, _total_allocations); 412 log_debug(gc, tlab)("TLAB totals: thrds: %d refills: %d max: %d" 413 " slow allocs: %d max %d waste: %4.1f%%" 414 " gc: " SIZE_FORMAT "B max: " SIZE_FORMAT "B" 415 " refill: " SIZE_FORMAT "B max: " SIZE_FORMAT "B", 416 _allocating_threads, _total_refills, _max_refills, 417 _total_slow_allocations, _max_slow_allocations, waste_percent, 418 _total_gc_waste * HeapWordSize, _max_gc_waste * HeapWordSize, 419 _total_refill_waste * HeapWordSize, _max_refill_waste * HeapWordSize); 420 421 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_total_allocations ->set_value(_total_allocations); 426 _perf_total_gc_waste ->set_value(_total_gc_waste); 427 _perf_max_gc_waste ->set_value(_max_gc_waste); 428 _perf_total_refill_waste ->set_value(_total_refill_waste); 429 _perf_max_refill_waste ->set_value(_max_refill_waste); 430 _perf_total_slow_allocations ->set_value(_total_slow_allocations); 431 _perf_max_slow_allocations ->set_value(_max_slow_allocations); 432 } 433 } 434 435 size_t ThreadLocalAllocBuffer::end_reserve() { 436 size_t reserve_size = Universe::heap()->tlab_alloc_reserve(); 437 return MAX2(reserve_size, (size_t)_reserve_for_allocation_prefetch); 438 }