rev 52072 : 8211955: GC abstraction for LAB reserve

   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/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                                    _fast_refill_waste,
  80                                    _slow_refill_waste);
  81   } else {
  82     assert(_number_of_refills == 0 && _fast_refill_waste == 0 &&
  83            _slow_refill_waste == 0 && _gc_waste          == 0,
  84            "tlab stats == 0");
  85   }
  86 
  87   stats->update_slow_allocations(_slow_allocations);
  88 
  89   reset_statistics();
  90 }
  91 
  92 void ThreadLocalAllocBuffer::insert_filler() {
  93   assert(end() != NULL, "Must not be retired");
  94   Universe::heap()->fill_with_dummy_object(top(), hard_end(), true);
  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   _slow_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   _fast_refill_waste = 0;
 149   _slow_refill_waste = 0;
 150   _gc_waste          = 0;
 151   _slow_allocations  = 0;
 152   _allocated_size    = 0;
 153 }
 154 
 155 void ThreadLocalAllocBuffer::fill(HeapWord* start,
 156                                   HeapWord* top,
 157                                   size_t    new_size) {
 158   _number_of_refills++;
 159   _allocated_size += new_size;
 160   print_stats("fill");
 161   assert(top <= start + new_size - alignment_reserve(), "size too small");
 162 
 163   initialize(start, top, start + new_size - alignment_reserve());
 164 
 165   // Reset amount of internal fragmentation
 166   set_refill_waste_limit(initial_refill_waste_limit());
 167 }
 168 
 169 void ThreadLocalAllocBuffer::initialize(HeapWord* start,
 170                                         HeapWord* top,
 171                                         HeapWord* end) {
 172   set_start(start);
 173   set_top(top);
 174   set_pf_top(top);
 175   set_end(end);
 176   set_allocation_end(end);
 177   invariants();
 178 }
 179 
 180 void ThreadLocalAllocBuffer::initialize() {
 181   initialize(NULL,                    // start
 182              NULL,                    // top
 183              NULL);                   // end
 184 
 185   set_desired_size(initial_desired_size());
 186 
 187   size_t capacity = Universe::heap()->tlab_capacity(thread()) / HeapWordSize;
 188   double alloc_frac = desired_size() * target_refills() / (double) capacity;
 189   _allocation_fraction.sample(alloc_frac);
 190 
 191   set_refill_waste_limit(initial_refill_waste_limit());
 192 
 193   reset_statistics();
 194 }
 195 
 196 void ThreadLocalAllocBuffer::startup_initialization() {
 197   ThreadLocalAllocStats::initialize();
 198 
 199   // Assuming each thread's active tlab is, on average,
 200   // 1/2 full at a GC
 201   _target_refills = 100 / (2 * TLABWasteTargetPercent);
 202   // We need to set initial target refills to 2 to avoid a GC which causes VM
 203   // abort during VM initialization.
 204   _target_refills = MAX2(_target_refills, 2U);
 205 
 206 #ifdef COMPILER2
 207   // If the C2 compiler is present, extra space is needed at the end of
 208   // TLABs, otherwise prefetching instructions generated by the C2
 209   // compiler will fault (due to accessing memory outside of heap).
 210   // The amount of space is the max of the number of lines to
 211   // prefetch for array and for instance allocations. (Extra space must be
 212   // reserved to accommodate both types of allocations.)
 213   //
 214   // Only SPARC-specific BIS instructions are known to fault. (Those
 215   // instructions are generated if AllocatePrefetchStyle==3 and
 216   // AllocatePrefetchInstr==1). To be on the safe side, however,
 217   // extra space is reserved for all combinations of
 218   // AllocatePrefetchStyle and AllocatePrefetchInstr.
 219   //
 220   // If the C2 compiler is not present, no space is reserved.
 221 
 222   // +1 for rounding up to next cache line, +1 to be safe
 223   if (is_server_compilation_mode_vm()) {
 224     int lines =  MAX2(AllocatePrefetchLines, AllocateInstancePrefetchLines) + 2;
 225     _reserve_for_allocation_prefetch = (AllocatePrefetchDistance + AllocatePrefetchStepSize * lines) /
 226                                        (int)HeapWordSize;
 227   }
 228 #endif
 229 
 230   // During jvm startup, the main thread is initialized
 231   // before the heap is initialized.  So reinitialize it now.
 232   guarantee(Thread::current()->is_Java_thread(), "tlab initialization thread not Java thread");
 233   Thread::current()->tlab().initialize();
 234 
 235   log_develop_trace(gc, tlab)("TLAB min: " SIZE_FORMAT " initial: " SIZE_FORMAT " max: " SIZE_FORMAT,
 236                                min_size(), Thread::current()->tlab().initial_desired_size(), max_size());
 237 }
 238 
 239 size_t ThreadLocalAllocBuffer::initial_desired_size() {
 240   size_t init_sz = 0;
 241 
 242   if (TLABSize > 0) {
 243     init_sz = TLABSize / HeapWordSize;
 244   } else {
 245     // Initial size is a function of the average number of allocating threads.
 246     unsigned int nof_threads = ThreadLocalAllocStats::allocating_threads_avg();
 247 
 248     init_sz  = (Universe::heap()->tlab_capacity(thread()) / HeapWordSize) /
 249                       (nof_threads * target_refills());
 250     init_sz = align_object_size(init_sz);
 251   }
 252   init_sz = MIN2(MAX2(init_sz, min_size()), max_size());
 253   return init_sz;
 254 }
 255 
 256 void ThreadLocalAllocBuffer::print_stats(const char* tag) {
 257   Log(gc, tlab) log;
 258   if (!log.is_trace()) {
 259     return;
 260   }
 261 
 262   Thread* thrd = thread();
 263   size_t waste = _gc_waste + _slow_refill_waste + _fast_refill_waste;
 264   double waste_percent = percent_of(waste, _allocated_size);
 265   size_t tlab_used  = Universe::heap()->tlab_used(thrd);
 266   log.trace("TLAB: %s thread: " INTPTR_FORMAT " [id: %2d]"
 267             " desired_size: " SIZE_FORMAT "KB"
 268             " slow allocs: %d  refill waste: " SIZE_FORMAT "B"
 269             " alloc:%8.5f %8.0fKB refills: %d waste %4.1f%% gc: %dB"
 270             " slow: %dB fast: %dB",
 271             tag, p2i(thrd), thrd->osthread()->thread_id(),
 272             _desired_size / (K / HeapWordSize),
 273             _slow_allocations, _refill_waste_limit * HeapWordSize,
 274             _allocation_fraction.average(),
 275             _allocation_fraction.average() * tlab_used / K,
 276             _number_of_refills, waste_percent,
 277             _gc_waste * HeapWordSize,
 278             _slow_refill_waste * HeapWordSize,
 279             _fast_refill_waste * HeapWordSize);
 280 }
 281 
 282 void ThreadLocalAllocBuffer::verify() {
 283   HeapWord* p = start();
 284   HeapWord* t = top();
 285   HeapWord* prev_p = NULL;
 286   while (p < t) {
 287     oopDesc::verify(oop(p));
 288     prev_p = p;
 289     p += oop(p)->size();
 290   }
 291   guarantee(p == top(), "end of last object must match end of space");
 292 }
 293 
 294 void ThreadLocalAllocBuffer::set_sample_end() {
 295   size_t heap_words_remaining = pointer_delta(_end, _top);
 296   size_t bytes_until_sample = thread()->heap_sampler().bytes_until_sample();
 297   size_t words_until_sample = bytes_until_sample / HeapWordSize;
 298 
 299   if (heap_words_remaining > words_until_sample) {
 300     HeapWord* new_end = _top + words_until_sample;
 301     set_end(new_end);
 302     _bytes_since_last_sample_point = bytes_until_sample;
 303   } else {
 304     _bytes_since_last_sample_point = heap_words_remaining * HeapWordSize;
 305   }
 306 }
 307 
 308 Thread* ThreadLocalAllocBuffer::thread() {
 309   return (Thread*)(((char*)this) + in_bytes(start_offset()) - in_bytes(Thread::tlab_start_offset()));
 310 }
 311 
 312 void ThreadLocalAllocBuffer::set_back_allocation_end() {
 313   _end = _allocation_end;
 314 }
 315 
 316 HeapWord* ThreadLocalAllocBuffer::hard_end() {
 317   return _allocation_end + alignment_reserve();
 318 }
 319 
 320 PerfVariable* ThreadLocalAllocStats::_perf_allocating_threads;
 321 PerfVariable* ThreadLocalAllocStats::_perf_total_refills;
 322 PerfVariable* ThreadLocalAllocStats::_perf_max_refills;
 323 PerfVariable* ThreadLocalAllocStats::_perf_total_allocations;
 324 PerfVariable* ThreadLocalAllocStats::_perf_total_gc_waste;
 325 PerfVariable* ThreadLocalAllocStats::_perf_max_gc_waste;
 326 PerfVariable* ThreadLocalAllocStats::_perf_total_slow_refill_waste;
 327 PerfVariable* ThreadLocalAllocStats::_perf_max_slow_refill_waste;
 328 PerfVariable* ThreadLocalAllocStats::_perf_total_fast_refill_waste;
 329 PerfVariable* ThreadLocalAllocStats::_perf_max_fast_refill_waste;
 330 PerfVariable* ThreadLocalAllocStats::_perf_total_slow_allocations;
 331 PerfVariable* ThreadLocalAllocStats::_perf_max_slow_allocations;
 332 AdaptiveWeightedAverage ThreadLocalAllocStats::_allocating_threads_avg(0);
 333 
 334 static PerfVariable* create_perf_variable(const char* name, PerfData::Units unit, TRAPS) {
 335   ResourceMark rm;
 336   return PerfDataManager::create_variable(SUN_GC, PerfDataManager::counter_name("tlab", name), unit, THREAD);
 337 }
 338 
 339 void ThreadLocalAllocStats::initialize() {
 340   _allocating_threads_avg = AdaptiveWeightedAverage(TLABAllocationWeight);
 341   _allocating_threads_avg.sample(1); // One allocating thread at startup
 342 
 343   if (UsePerfData) {
 344     EXCEPTION_MARK;
 345     _perf_allocating_threads      = create_perf_variable("allocThreads", PerfData::U_None,  CHECK);
 346     _perf_total_refills           = create_perf_variable("fills",        PerfData::U_None,  CHECK);
 347     _perf_max_refills             = create_perf_variable("maxFills",     PerfData::U_None,  CHECK);
 348     _perf_total_allocations       = create_perf_variable("alloc",        PerfData::U_Bytes, CHECK);
 349     _perf_total_gc_waste          = create_perf_variable("gcWaste",      PerfData::U_Bytes, CHECK);
 350     _perf_max_gc_waste            = create_perf_variable("maxGcWaste",   PerfData::U_Bytes, CHECK);
 351     _perf_total_slow_refill_waste = create_perf_variable("slowWaste",    PerfData::U_Bytes, CHECK);
 352     _perf_max_slow_refill_waste   = create_perf_variable("maxSlowWaste", PerfData::U_Bytes, CHECK);
 353     _perf_total_fast_refill_waste = create_perf_variable("fastWaste",    PerfData::U_Bytes, CHECK);
 354     _perf_max_fast_refill_waste   = create_perf_variable("maxFastWaste", PerfData::U_Bytes, CHECK);
 355     _perf_total_slow_allocations  = create_perf_variable("slowAlloc",    PerfData::U_None,  CHECK);
 356     _perf_max_slow_allocations    = create_perf_variable("maxSlowAlloc", PerfData::U_None,  CHECK);
 357   }
 358 }
 359 
 360 ThreadLocalAllocStats::ThreadLocalAllocStats() :
 361     _allocating_threads(0),
 362     _total_refills(0),
 363     _max_refills(0),
 364     _total_allocations(0),
 365     _total_gc_waste(0),
 366     _max_gc_waste(0),
 367     _total_fast_refill_waste(0),
 368     _max_fast_refill_waste(0),
 369     _total_slow_refill_waste(0),
 370     _max_slow_refill_waste(0),
 371     _total_slow_allocations(0),
 372     _max_slow_allocations(0) {}
 373 
 374 unsigned int ThreadLocalAllocStats::allocating_threads_avg() {
 375   return MAX2((unsigned int)(_allocating_threads_avg.average() + 0.5), 1U);
 376 }
 377 
 378 void ThreadLocalAllocStats::update_fast_allocations(unsigned int refills,
 379                                        size_t allocations,
 380                                        size_t gc_waste,
 381                                        size_t fast_refill_waste,
 382                                        size_t slow_refill_waste) {
 383   _allocating_threads      += 1;
 384   _total_refills           += refills;
 385   _max_refills              = MAX2(_max_refills, refills);
 386   _total_allocations       += allocations;
 387   _total_gc_waste          += gc_waste;
 388   _max_gc_waste             = MAX2(_max_gc_waste, gc_waste);
 389   _total_fast_refill_waste += fast_refill_waste;
 390   _max_fast_refill_waste    = MAX2(_max_fast_refill_waste, fast_refill_waste);
 391   _total_slow_refill_waste += slow_refill_waste;
 392   _max_slow_refill_waste    = MAX2(_max_slow_refill_waste, slow_refill_waste);
 393 }
 394 
 395 void ThreadLocalAllocStats::update_slow_allocations(unsigned int allocations) {
 396   _total_slow_allocations += allocations;
 397   _max_slow_allocations    = MAX2(_max_slow_allocations, allocations);
 398 }
 399 
 400 void ThreadLocalAllocStats::update(const ThreadLocalAllocStats& other) {
 401   _allocating_threads      += other._allocating_threads;
 402   _total_refills           += other._total_refills;
 403   _max_refills              = MAX2(_max_refills, other._max_refills);
 404   _total_allocations       += other._total_allocations;
 405   _total_gc_waste          += other._total_gc_waste;
 406   _max_gc_waste             = MAX2(_max_gc_waste, other._max_gc_waste);
 407   _total_fast_refill_waste += other._total_fast_refill_waste;
 408   _max_fast_refill_waste    = MAX2(_max_fast_refill_waste, other._max_fast_refill_waste);
 409   _total_slow_refill_waste += other._total_slow_refill_waste;
 410   _max_slow_refill_waste    = MAX2(_max_slow_refill_waste, other._max_slow_refill_waste);
 411   _total_slow_allocations  += other._total_slow_allocations;
 412   _max_slow_allocations     = MAX2(_max_slow_allocations, other._max_slow_allocations);
 413 }
 414 
 415 void ThreadLocalAllocStats::reset() {
 416   _allocating_threads      = 0;
 417   _total_refills           = 0;
 418   _max_refills             = 0;
 419   _total_allocations       = 0;
 420   _total_gc_waste          = 0;
 421   _max_gc_waste            = 0;
 422   _total_fast_refill_waste = 0;
 423   _max_fast_refill_waste   = 0;
 424   _total_slow_refill_waste = 0;
 425   _max_slow_refill_waste   = 0;
 426   _total_slow_allocations  = 0;
 427   _max_slow_allocations    = 0;
 428 }
 429 
 430 void ThreadLocalAllocStats::publish() {
 431   if (_total_allocations == 0) {
 432     return;
 433   }
 434 
 435   _allocating_threads_avg.sample(_allocating_threads);
 436 
 437   const size_t waste = _total_gc_waste + _total_slow_refill_waste + _total_fast_refill_waste;
 438   const double waste_percent = percent_of(waste, _total_allocations);
 439   log_debug(gc, tlab)("TLAB totals: thrds: %d  refills: %d max: %d"
 440                       " slow allocs: %d max %d waste: %4.1f%%"
 441                       " gc: " SIZE_FORMAT "B max: " SIZE_FORMAT "B"
 442                       " slow: " SIZE_FORMAT "B max: " SIZE_FORMAT "B"
 443                       " fast: " SIZE_FORMAT "B max: " SIZE_FORMAT "B",
 444                       _allocating_threads, _total_refills, _max_refills,
 445                       _total_slow_allocations, _max_slow_allocations, waste_percent,
 446                       _total_gc_waste * HeapWordSize, _max_gc_waste * HeapWordSize,
 447                       _total_slow_refill_waste * HeapWordSize, _max_slow_refill_waste * HeapWordSize,
 448                       _total_fast_refill_waste * HeapWordSize, _max_fast_refill_waste * HeapWordSize);
 449 
 450   if (UsePerfData) {
 451     _perf_allocating_threads      ->set_value(_allocating_threads);
 452     _perf_total_refills           ->set_value(_total_refills);
 453     _perf_max_refills             ->set_value(_max_refills);
 454     _perf_total_allocations       ->set_value(_total_allocations);
 455     _perf_total_gc_waste          ->set_value(_total_gc_waste);
 456     _perf_max_gc_waste            ->set_value(_max_gc_waste);
 457     _perf_total_slow_refill_waste ->set_value(_total_slow_refill_waste);
 458     _perf_max_slow_refill_waste   ->set_value(_max_slow_refill_waste);
 459     _perf_total_fast_refill_waste ->set_value(_total_fast_refill_waste);
 460     _perf_max_fast_refill_waste   ->set_value(_max_fast_refill_waste);
 461     _perf_total_slow_allocations  ->set_value(_total_slow_allocations);
 462     _perf_max_slow_allocations    ->set_value(_max_slow_allocations);
 463   }
 464 }
 465 
 466 size_t ThreadLocalAllocBuffer::end_reserve() {
 467   size_t reserve_size = Universe::heap()->tlab_alloc_reserve();
 468   return MAX2(reserve_size, (size_t)_reserve_for_allocation_prefetch);
 469 }
--- EOF ---