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 "memory/resourceArea.hpp" 29 #include "memory/universe.inline.hpp" 30 #include "oops/oop.inline.hpp" 31 #include "runtime/thread.inline.hpp" 32 #include "utilities/copy.hpp" 33 34 // Thread-Local Edens support 35 36 // static member initialization 37 size_t ThreadLocalAllocBuffer::_max_size = 0; 38 unsigned ThreadLocalAllocBuffer::_target_refills = 0; 39 GlobalTLABStats* ThreadLocalAllocBuffer::_global_stats = NULL; 40 41 void ThreadLocalAllocBuffer::clear_before_allocation() { 42 _slow_refill_waste += (unsigned)remaining(); 43 // In debug mode we expect the storage above top to be uninitialized 44 // or filled with a padding object. 45 assert(!ZapUnusedHeapArea || VM_Version::reserve_for_allocation_prefetch() > 0 || top() == NULL || *(intptr_t*)top() != 0, "overzeroing detected"); 46 make_parsable(true); // also retire the TLAB 47 } 48 49 void ThreadLocalAllocBuffer::accumulate_statistics_before_gc() { 50 global_stats()->initialize(); 51 52 for (JavaThread *thread = Threads::first(); thread != NULL; thread = thread->next()) { 53 thread->tlab().accumulate_statistics(); 54 thread->tlab().initialize_statistics(); 55 } 56 57 // Publish new stats if some allocation occurred. 58 if (global_stats()->allocation() != 0) { 59 global_stats()->publish(); 60 if (PrintTLAB) { 61 global_stats()->print(); 62 } 63 } 64 } 65 66 void ThreadLocalAllocBuffer::accumulate_statistics() { 67 Thread* thread = myThread(); 68 size_t capacity = Universe::heap()->tlab_capacity(thread); 69 size_t used = Universe::heap()->tlab_used(thread); 70 71 _gc_waste += (unsigned)remaining(); 72 size_t total_allocated = thread->allocated_bytes(); 73 size_t allocated_since_last_gc = total_allocated - _allocated_before_last_gc; 74 _allocated_before_last_gc = total_allocated; 75 76 if (PrintTLAB && (_number_of_refills > 0 || Verbose)) { 77 print_stats("gc"); 78 } 79 80 if (_number_of_refills > 0) { 81 // Update allocation history if a reasonable amount of eden was allocated. 82 bool update_allocation_history = used > 0.5 * capacity; 83 84 if (update_allocation_history) { 85 // Average the fraction of eden allocated in a tlab by this 86 // thread for use in the next resize operation. 87 // _gc_waste is not subtracted because it's included in 88 // "used". 89 // The result can be larger than 1.0 due to direct to old allocations. 90 // These allocations should ideally not be counted but since it is not possible 91 // to filter them out here we just cap the fraction to be at most 1.0. 92 double alloc_frac = MIN2(1.0, (double) allocated_since_last_gc / used); 93 _allocation_fraction.sample(alloc_frac); 94 } 95 global_stats()->update_allocating_threads(); 96 global_stats()->update_number_of_refills(_number_of_refills); 97 global_stats()->update_allocation(_number_of_refills * desired_size()); 98 global_stats()->update_gc_waste(_gc_waste); 99 global_stats()->update_slow_refill_waste(_slow_refill_waste); 100 global_stats()->update_fast_refill_waste(_fast_refill_waste); 101 102 } else { 103 assert(_number_of_refills == 0 && _fast_refill_waste == 0 && 104 _slow_refill_waste == 0 && _gc_waste == 0, 105 "tlab stats == 0"); 106 } 107 global_stats()->update_slow_allocations(_slow_allocations); 108 } 109 110 // Fills the current tlab with a dummy filler array to create 111 // an illusion of a contiguous Eden and optionally retires the tlab. 112 // Waste accounting should be done in caller as appropriate; see, 113 // for example, clear_before_allocation(). 114 void ThreadLocalAllocBuffer::make_parsable(bool retire) { 115 if (end() != NULL) { 116 invariants(); 117 118 if (retire) { 119 myThread()->incr_allocated_bytes(used_bytes()); 120 } 121 122 CollectedHeap::fill_with_object(top(), hard_end(), retire); 123 124 if (retire || ZeroTLAB) { // "Reset" the TLAB 125 set_start(NULL); 126 set_top(NULL); 127 set_pf_top(NULL); 128 set_end(NULL); 129 } 130 } 131 assert(!(retire || ZeroTLAB) || 132 (start() == NULL && end() == NULL && top() == NULL), 133 "TLAB must be reset"); 134 } 135 136 void ThreadLocalAllocBuffer::resize_all_tlabs() { 137 if (ResizeTLAB) { 138 for (JavaThread *thread = Threads::first(); thread != NULL; thread = thread->next()) { 139 thread->tlab().resize(); 140 } 141 } 142 } 143 144 void ThreadLocalAllocBuffer::resize() { 145 // Compute the next tlab size using expected allocation amount 146 assert(ResizeTLAB, "Should not call this otherwise"); 147 size_t alloc = (size_t)(_allocation_fraction.average() * 148 (Universe::heap()->tlab_capacity(myThread()) / HeapWordSize)); 149 size_t new_size = alloc / _target_refills; 150 151 new_size = MIN2(MAX2(new_size, min_size()), max_size()); 152 153 size_t aligned_new_size = align_object_size(new_size); 154 155 if (PrintTLAB && Verbose) { 156 gclog_or_tty->print("TLAB new size: thread: " INTPTR_FORMAT " [id: %2d]" 157 " refills %d alloc: %8.6f desired_size: " SIZE_FORMAT " -> " SIZE_FORMAT "\n", 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 if (PrintTLAB && Verbose) { 178 print_stats("fill"); 179 } 180 assert(top <= start + new_size - alignment_reserve(), "size too small"); 181 initialize(start, top, start + new_size - alignment_reserve()); 182 183 // Reset amount of internal fragmentation 184 set_refill_waste_limit(initial_refill_waste_limit()); 185 } 186 187 void ThreadLocalAllocBuffer::initialize(HeapWord* start, 188 HeapWord* top, 189 HeapWord* end) { 190 set_start(start); 191 set_top(top); 192 set_pf_top(top); 193 set_end(end); 194 invariants(); 195 } 196 197 void ThreadLocalAllocBuffer::initialize() { 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 (primordial) 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 // During jvm startup, the main (primordial) thread is initialized 228 // before the heap is initialized. So reinitialize it now. 229 guarantee(Thread::current()->is_Java_thread(), "tlab initialization thread not Java thread"); 230 Thread::current()->tlab().initialize(); 231 232 if (PrintTLAB && Verbose) { 233 gclog_or_tty->print("TLAB min: " SIZE_FORMAT " initial: " SIZE_FORMAT " max: " SIZE_FORMAT "\n", 234 min_size(), Thread::current()->tlab().initial_desired_size(), max_size()); 235 } 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 if (global_stats() != NULL) { 244 // Initial size is a function of the average number of allocating threads. 245 unsigned nof_threads = global_stats()->allocating_threads_avg(); 246 247 init_sz = (Universe::heap()->tlab_capacity(myThread()) / 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 Thread* thrd = myThread(); 257 size_t waste = _gc_waste + _slow_refill_waste + _fast_refill_waste; 258 size_t alloc = _number_of_refills * _desired_size; 259 double waste_percent = alloc == 0 ? 0.0 : 260 100.0 * waste / alloc; 261 size_t tlab_used = Universe::heap()->tlab_used(thrd); 262 gclog_or_tty->print("TLAB: %s thread: " INTPTR_FORMAT " [id: %2d]" 263 " desired_size: " SIZE_FORMAT "KB" 264 " slow allocs: %d refill waste: " SIZE_FORMAT "B" 265 " alloc:%8.5f %8.0fKB refills: %d waste %4.1f%% gc: %dB" 266 " slow: %dB fast: %dB\n", 267 tag, p2i(thrd), thrd->osthread()->thread_id(), 268 _desired_size / (K / HeapWordSize), 269 _slow_allocations, _refill_waste_limit * HeapWordSize, 270 _allocation_fraction.average(), 271 _allocation_fraction.average() * tlab_used / K, 272 _number_of_refills, waste_percent, 273 _gc_waste * HeapWordSize, 274 _slow_refill_waste * HeapWordSize, 275 _fast_refill_waste * HeapWordSize); 276 } 277 278 void ThreadLocalAllocBuffer::verify() { 279 HeapWord* p = start(); 280 HeapWord* t = top(); 281 HeapWord* prev_p = NULL; 282 while (p < t) { 283 oop(p)->verify(); 284 prev_p = p; 285 p += oop(p)->size(); 286 } 287 guarantee(p == top(), "end of last object must match end of space"); 288 } 289 290 Thread* ThreadLocalAllocBuffer::myThread() { 291 return (Thread*)(((char *)this) + 292 in_bytes(start_offset()) - 293 in_bytes(Thread::tlab_start_offset())); 294 } 295 296 297 GlobalTLABStats::GlobalTLABStats() : 298 _allocating_threads_avg(TLABAllocationWeight) { 299 300 initialize(); 301 302 _allocating_threads_avg.sample(1); // One allocating thread at startup 303 304 if (UsePerfData) { 305 306 EXCEPTION_MARK; 307 ResourceMark rm; 308 309 char* cname = PerfDataManager::counter_name("tlab", "allocThreads"); 310 _perf_allocating_threads = 311 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_None, CHECK); 312 313 cname = PerfDataManager::counter_name("tlab", "fills"); 314 _perf_total_refills = 315 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_None, CHECK); 316 317 cname = PerfDataManager::counter_name("tlab", "maxFills"); 318 _perf_max_refills = 319 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_None, CHECK); 320 321 cname = PerfDataManager::counter_name("tlab", "alloc"); 322 _perf_allocation = 323 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_Bytes, CHECK); 324 325 cname = PerfDataManager::counter_name("tlab", "gcWaste"); 326 _perf_gc_waste = 327 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_Bytes, CHECK); 328 329 cname = PerfDataManager::counter_name("tlab", "maxGcWaste"); 330 _perf_max_gc_waste = 331 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_Bytes, CHECK); 332 333 cname = PerfDataManager::counter_name("tlab", "slowWaste"); 334 _perf_slow_refill_waste = 335 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_Bytes, CHECK); 336 337 cname = PerfDataManager::counter_name("tlab", "maxSlowWaste"); 338 _perf_max_slow_refill_waste = 339 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_Bytes, CHECK); 340 341 cname = PerfDataManager::counter_name("tlab", "fastWaste"); 342 _perf_fast_refill_waste = 343 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_Bytes, CHECK); 344 345 cname = PerfDataManager::counter_name("tlab", "maxFastWaste"); 346 _perf_max_fast_refill_waste = 347 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_Bytes, CHECK); 348 349 cname = PerfDataManager::counter_name("tlab", "slowAlloc"); 350 _perf_slow_allocations = 351 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_None, CHECK); 352 353 cname = PerfDataManager::counter_name("tlab", "maxSlowAlloc"); 354 _perf_max_slow_allocations = 355 PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_None, CHECK); 356 } 357 } 358 359 void GlobalTLABStats::initialize() { 360 // Clear counters summarizing info from all threads 361 _allocating_threads = 0; 362 _total_refills = 0; 363 _max_refills = 0; 364 _total_allocation = 0; 365 _total_gc_waste = 0; 366 _max_gc_waste = 0; 367 _total_slow_refill_waste = 0; 368 _max_slow_refill_waste = 0; 369 _total_fast_refill_waste = 0; 370 _max_fast_refill_waste = 0; 371 _total_slow_allocations = 0; 372 _max_slow_allocations = 0; 373 } 374 375 void GlobalTLABStats::publish() { 376 _allocating_threads_avg.sample(_allocating_threads); 377 if (UsePerfData) { 378 _perf_allocating_threads ->set_value(_allocating_threads); 379 _perf_total_refills ->set_value(_total_refills); 380 _perf_max_refills ->set_value(_max_refills); 381 _perf_allocation ->set_value(_total_allocation); 382 _perf_gc_waste ->set_value(_total_gc_waste); 383 _perf_max_gc_waste ->set_value(_max_gc_waste); 384 _perf_slow_refill_waste ->set_value(_total_slow_refill_waste); 385 _perf_max_slow_refill_waste->set_value(_max_slow_refill_waste); 386 _perf_fast_refill_waste ->set_value(_total_fast_refill_waste); 387 _perf_max_fast_refill_waste->set_value(_max_fast_refill_waste); 388 _perf_slow_allocations ->set_value(_total_slow_allocations); 389 _perf_max_slow_allocations ->set_value(_max_slow_allocations); 390 } 391 } 392 393 void GlobalTLABStats::print() { 394 size_t waste = _total_gc_waste + _total_slow_refill_waste + _total_fast_refill_waste; 395 double waste_percent = _total_allocation == 0 ? 0.0 : 396 100.0 * waste / _total_allocation; 397 gclog_or_tty->print("TLAB totals: thrds: %d refills: %d max: %d" 398 " slow allocs: %d max %d waste: %4.1f%%" 399 " gc: " SIZE_FORMAT "B max: " SIZE_FORMAT "B" 400 " slow: " SIZE_FORMAT "B max: " SIZE_FORMAT "B" 401 " fast: " SIZE_FORMAT "B max: " SIZE_FORMAT "B\n", 402 _allocating_threads, 403 _total_refills, _max_refills, 404 _total_slow_allocations, _max_slow_allocations, 405 waste_percent, 406 _total_gc_waste * HeapWordSize, 407 _max_gc_waste * HeapWordSize, 408 _total_slow_refill_waste * HeapWordSize, 409 _max_slow_refill_waste * HeapWordSize, 410 _total_fast_refill_waste * HeapWordSize, 411 _max_fast_refill_waste * HeapWordSize); 412 }