1 /* 2 * Copyright (c) 1999, 2015, 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 "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 (JavaThread *thread = Threads::first(); thread != NULL; thread = thread->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 (JavaThread *thread = Threads::first(); thread != NULL; thread = thread->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 (primordial) 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 // Need extra space at the end of TLAB, otherwise prefetching 220 // instructions will fault (due to accessing memory outside of heap). 221 // The amount of space is the max of the number of lines to 222 // prefetch for array and for instance allocations. (Extra space must be 223 // reserved to accommodate both types of allocations.) 224 // 225 // Only SPARC-specific BIS instructions are known to fault. (Those 226 // instructions are generated if AllocatePrefetchStyle==3 and 227 // AllocatePrefetchInstr==1). To be on the safe side, however, 228 // extra space is reserved for all combinations of 229 // AllocatePrefetchStyle and AllocatePrefetchInstr. 230 231 // +1 for rounding up to next cache line, +1 to be safe 232 int lines = MAX2(AllocatePrefetchLines, AllocateInstancePrefetchLines) + 2; 233 _reserve_for_allocation_prefetch = (AllocatePrefetchDistance + AllocatePrefetchStepSize * lines) / 234 (int)HeapWordSize; 235 236 // During jvm startup, the main (primordial) thread is initialized 237 // before the heap is initialized. So reinitialize it now. 238 guarantee(Thread::current()->is_Java_thread(), "tlab initialization thread not Java thread"); 239 Thread::current()->tlab().initialize(); 240 241 log_develop_trace(gc, tlab)("TLAB min: " SIZE_FORMAT " initial: " SIZE_FORMAT " max: " SIZE_FORMAT, 242 min_size(), Thread::current()->tlab().initial_desired_size(), max_size()); 243 } 244 245 size_t ThreadLocalAllocBuffer::initial_desired_size() { 246 size_t init_sz = 0; 247 248 if (TLABSize > 0) { 249 init_sz = TLABSize / HeapWordSize; 250 } else if (global_stats() != NULL) { 251 // Initial size is a function of the average number of allocating threads. 252 unsigned nof_threads = global_stats()->allocating_threads_avg(); 253 254 init_sz = (Universe::heap()->tlab_capacity(myThread()) / HeapWordSize) / 255 (nof_threads * target_refills()); 256 init_sz = align_object_size(init_sz); 257 } 258 init_sz = MIN2(MAX2(init_sz, min_size()), max_size()); 259 return init_sz; 260 } 261 262 void ThreadLocalAllocBuffer::print_stats(const char* tag) { 263 Log(gc, tlab) log; 264 if (!log.is_trace()) { 265 return; 266 } 267 268 Thread* thrd = myThread(); 269 size_t waste = _gc_waste + _slow_refill_waste + _fast_refill_waste; 270 size_t alloc = _number_of_refills * _desired_size; 271 double waste_percent = alloc == 0 ? 0.0 : 272 100.0 * waste / alloc; 273 size_t tlab_used = Universe::heap()->tlab_used(thrd); 274 log.trace("TLAB: %s thread: " INTPTR_FORMAT " [id: %2d]" 275 " desired_size: " SIZE_FORMAT "KB" 276 " slow allocs: %d refill waste: " SIZE_FORMAT "B" 277 " alloc:%8.5f %8.0fKB refills: %d waste %4.1f%% gc: %dB" 278 " slow: %dB fast: %dB", 279 tag, p2i(thrd), thrd->osthread()->thread_id(), 280 _desired_size / (K / HeapWordSize), 281 _slow_allocations, _refill_waste_limit * HeapWordSize, 282 _allocation_fraction.average(), 283 _allocation_fraction.average() * tlab_used / K, 284 _number_of_refills, waste_percent, 285 _gc_waste * HeapWordSize, 286 _slow_refill_waste * HeapWordSize, 287 _fast_refill_waste * HeapWordSize); 288 } 289 290 void ThreadLocalAllocBuffer::verify() { 291 HeapWord* p = start(); 292 HeapWord* t = top(); 293 HeapWord* prev_p = NULL; 294 while (p < t) { 295 oop(p)->verify(); 296 prev_p = p; 297 p += oop(p)->size(); 298 } 299 guarantee(p == top(), "end of last object must match end of space"); 300 } 301 302 Thread* ThreadLocalAllocBuffer::myThread() { 303 return (Thread*)(((char *)this) + 304 in_bytes(start_offset()) - 305 in_bytes(Thread::tlab_start_offset())); 306 } 307 308 309 GlobalTLABStats::GlobalTLABStats() : 310 _allocating_threads_avg(TLABAllocationWeight) { 311 312 initialize(); 313 314 _allocating_threads_avg.sample(1); // One allocating thread at startup 315 316 if (UsePerfData) { 317 318 EXCEPTION_MARK; 319 ResourceMark rm; 320 321 char* cname = PerfDataManager::counter_name("tlab", "allocThreads"); 322 _perf_allocating_threads = 323 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_None, CHECK); 324 325 cname = PerfDataManager::counter_name("tlab", "fills"); 326 _perf_total_refills = 327 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_None, CHECK); 328 329 cname = PerfDataManager::counter_name("tlab", "maxFills"); 330 _perf_max_refills = 331 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_None, CHECK); 332 333 cname = PerfDataManager::counter_name("tlab", "alloc"); 334 _perf_allocation = 335 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_Bytes, CHECK); 336 337 cname = PerfDataManager::counter_name("tlab", "gcWaste"); 338 _perf_gc_waste = 339 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_Bytes, CHECK); 340 341 cname = PerfDataManager::counter_name("tlab", "maxGcWaste"); 342 _perf_max_gc_waste = 343 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_Bytes, CHECK); 344 345 cname = PerfDataManager::counter_name("tlab", "slowWaste"); 346 _perf_slow_refill_waste = 347 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_Bytes, CHECK); 348 349 cname = PerfDataManager::counter_name("tlab", "maxSlowWaste"); 350 _perf_max_slow_refill_waste = 351 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_Bytes, CHECK); 352 353 cname = PerfDataManager::counter_name("tlab", "fastWaste"); 354 _perf_fast_refill_waste = 355 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_Bytes, CHECK); 356 357 cname = PerfDataManager::counter_name("tlab", "maxFastWaste"); 358 _perf_max_fast_refill_waste = 359 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_Bytes, CHECK); 360 361 cname = PerfDataManager::counter_name("tlab", "slowAlloc"); 362 _perf_slow_allocations = 363 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_None, CHECK); 364 365 cname = PerfDataManager::counter_name("tlab", "maxSlowAlloc"); 366 _perf_max_slow_allocations = 367 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_None, CHECK); 368 } 369 } 370 371 void GlobalTLABStats::initialize() { 372 // Clear counters summarizing info from all threads 373 _allocating_threads = 0; 374 _total_refills = 0; 375 _max_refills = 0; 376 _total_allocation = 0; 377 _total_gc_waste = 0; 378 _max_gc_waste = 0; 379 _total_slow_refill_waste = 0; 380 _max_slow_refill_waste = 0; 381 _total_fast_refill_waste = 0; 382 _max_fast_refill_waste = 0; 383 _total_slow_allocations = 0; 384 _max_slow_allocations = 0; 385 } 386 387 void GlobalTLABStats::publish() { 388 _allocating_threads_avg.sample(_allocating_threads); 389 if (UsePerfData) { 390 _perf_allocating_threads ->set_value(_allocating_threads); 391 _perf_total_refills ->set_value(_total_refills); 392 _perf_max_refills ->set_value(_max_refills); 393 _perf_allocation ->set_value(_total_allocation); 394 _perf_gc_waste ->set_value(_total_gc_waste); 395 _perf_max_gc_waste ->set_value(_max_gc_waste); 396 _perf_slow_refill_waste ->set_value(_total_slow_refill_waste); 397 _perf_max_slow_refill_waste->set_value(_max_slow_refill_waste); 398 _perf_fast_refill_waste ->set_value(_total_fast_refill_waste); 399 _perf_max_fast_refill_waste->set_value(_max_fast_refill_waste); 400 _perf_slow_allocations ->set_value(_total_slow_allocations); 401 _perf_max_slow_allocations ->set_value(_max_slow_allocations); 402 } 403 } 404 405 void GlobalTLABStats::print() { 406 Log(gc, tlab) log; 407 if (!log.is_debug()) { 408 return; 409 } 410 411 size_t waste = _total_gc_waste + _total_slow_refill_waste + _total_fast_refill_waste; 412 double waste_percent = _total_allocation == 0 ? 0.0 : 413 100.0 * waste / _total_allocation; 414 log.debug("TLAB totals: thrds: %d refills: %d max: %d" 415 " slow allocs: %d max %d waste: %4.1f%%" 416 " gc: " SIZE_FORMAT "B max: " SIZE_FORMAT "B" 417 " slow: " SIZE_FORMAT "B max: " SIZE_FORMAT "B" 418 " fast: " SIZE_FORMAT "B max: " SIZE_FORMAT "B", 419 _allocating_threads, 420 _total_refills, _max_refills, 421 _total_slow_allocations, _max_slow_allocations, 422 waste_percent, 423 _total_gc_waste * HeapWordSize, 424 _max_gc_waste * HeapWordSize, 425 _total_slow_refill_waste * HeapWordSize, 426 _max_slow_refill_waste * HeapWordSize, 427 _total_fast_refill_waste * HeapWordSize, 428 _max_fast_refill_waste * HeapWordSize); 429 }