rev 49643 : [mq]: heap8
rev 49644 : [mq]: event_rebased
rev 49649 : [mq]: heap14

   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/genCollectedHeap.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 // Thread-Local Edens support
  37 
  38 // static member initialization
  39 size_t           ThreadLocalAllocBuffer::_max_size       = 0;
  40 int              ThreadLocalAllocBuffer::_reserve_for_allocation_prefetch = 0;
  41 unsigned         ThreadLocalAllocBuffer::_target_refills = 0;
  42 GlobalTLABStats* ThreadLocalAllocBuffer::_global_stats   = NULL;
  43 
  44 void ThreadLocalAllocBuffer::clear_before_allocation() {
  45   _slow_refill_waste += (unsigned)remaining();
  46   make_parsable(true);   // also retire the TLAB
  47 }
  48 
  49 size_t ThreadLocalAllocBuffer::remaining() {
  50   if (fast_path_end() == NULL) {
  51     return 0;
  52   }
  53 
  54   return pointer_delta(hard_end(), top());
  55 }
  56 
  57 void ThreadLocalAllocBuffer::accumulate_statistics_before_gc() {
  58   global_stats()->initialize();
  59 
  60   for (JavaThreadIteratorWithHandle jtiwh; JavaThread *thread = jtiwh.next(); ) {
  61     thread->tlab().accumulate_statistics();
  62     thread->tlab().initialize_statistics();
  63   }
  64 
  65   // Publish new stats if some allocation occurred.
  66   if (global_stats()->allocation() != 0) {
  67     global_stats()->publish();
  68     global_stats()->print();
  69   }
  70 }
  71 
  72 void ThreadLocalAllocBuffer::accumulate_statistics() {
  73   Thread* thread = myThread();
  74   size_t capacity = Universe::heap()->tlab_capacity(thread);
  75   size_t used     = Universe::heap()->tlab_used(thread);
  76 
  77   _gc_waste += (unsigned)remaining();
  78   size_t total_allocated = thread->allocated_bytes();
  79   size_t allocated_since_last_gc = total_allocated - _allocated_before_last_gc;
  80   _allocated_before_last_gc = total_allocated;
  81 
  82   print_stats("gc");
  83 
  84   if (_number_of_refills > 0) {
  85     // Update allocation history if a reasonable amount of eden was allocated.
  86     bool update_allocation_history = used > 0.5 * capacity;
  87 
  88     if (update_allocation_history) {
  89       // Average the fraction of eden allocated in a tlab by this
  90       // thread for use in the next resize operation.
  91       // _gc_waste is not subtracted because it's included in
  92       // "used".
  93       // The result can be larger than 1.0 due to direct to old allocations.
  94       // These allocations should ideally not be counted but since it is not possible
  95       // to filter them out here we just cap the fraction to be at most 1.0.
  96       double alloc_frac = MIN2(1.0, (double) allocated_since_last_gc / used);
  97       _allocation_fraction.sample(alloc_frac);
  98     }
  99     global_stats()->update_allocating_threads();
 100     global_stats()->update_number_of_refills(_number_of_refills);
 101     global_stats()->update_allocation(_number_of_refills * desired_size());
 102     global_stats()->update_gc_waste(_gc_waste);
 103     global_stats()->update_slow_refill_waste(_slow_refill_waste);
 104     global_stats()->update_fast_refill_waste(_fast_refill_waste);
 105 
 106   } else {
 107     assert(_number_of_refills == 0 && _fast_refill_waste == 0 &&
 108            _slow_refill_waste == 0 && _gc_waste          == 0,
 109            "tlab stats == 0");
 110   }
 111   global_stats()->update_slow_allocations(_slow_allocations);
 112 }
 113 
 114 // Fills the current tlab with a dummy filler array to create
 115 // an illusion of a contiguous Eden and optionally retires the tlab.
 116 // Waste accounting should be done in caller as appropriate; see,
 117 // for example, clear_before_allocation().
 118 void ThreadLocalAllocBuffer::make_parsable(bool retire, bool zap) {
 119   if (fast_path_end() != NULL) {
 120     invariants();
 121 
 122     if (retire) {
 123       myThread()->incr_allocated_bytes(used_bytes());
 124     }
 125 
 126     CollectedHeap::fill_with_object(top(), hard_end(), retire && zap);
 127 
 128     if (retire || ZeroTLAB) {  // "Reset" the TLAB
 129       set_start(NULL);
 130       set_top(NULL);
 131       set_pf_top(NULL);
 132       set_fast_path_end(NULL);
 133       set_allocation_end(NULL);
 134     }
 135   }
 136   assert(!(retire || ZeroTLAB)  ||
 137          (start() == NULL && fast_path_end() == NULL && top() == NULL &&
 138           _allocation_end == NULL),
 139          "TLAB must be reset");
 140 }
 141 
 142 void ThreadLocalAllocBuffer::resize_all_tlabs() {
 143   if (ResizeTLAB) {
 144     for (JavaThreadIteratorWithHandle jtiwh; JavaThread *thread = jtiwh.next(); ) {
 145       thread->tlab().resize();
 146     }
 147   }
 148 }
 149 
 150 void ThreadLocalAllocBuffer::resize() {
 151   // Compute the next tlab size using expected allocation amount
 152   assert(ResizeTLAB, "Should not call this otherwise");
 153   size_t alloc = (size_t)(_allocation_fraction.average() *
 154                           (Universe::heap()->tlab_capacity(myThread()) / HeapWordSize));
 155   size_t new_size = alloc / _target_refills;
 156 
 157   new_size = MIN2(MAX2(new_size, min_size()), max_size());
 158 
 159   size_t aligned_new_size = align_object_size(new_size);
 160 
 161   log_trace(gc, tlab)("TLAB new size: thread: " INTPTR_FORMAT " [id: %2d]"
 162                       " refills %d  alloc: %8.6f desired_size: " SIZE_FORMAT " -> " SIZE_FORMAT,
 163                       p2i(myThread()), myThread()->osthread()->thread_id(),
 164                       _target_refills, _allocation_fraction.average(), desired_size(), aligned_new_size);
 165 
 166   set_desired_size(aligned_new_size);
 167   set_refill_waste_limit(initial_refill_waste_limit());
 168 }
 169 
 170 void ThreadLocalAllocBuffer::initialize_statistics() {
 171     _number_of_refills = 0;
 172     _fast_refill_waste = 0;
 173     _slow_refill_waste = 0;
 174     _gc_waste          = 0;
 175     _slow_allocations  = 0;
 176 }
 177 
 178 void ThreadLocalAllocBuffer::fill(HeapWord* start,
 179                                   HeapWord* top,
 180                                   size_t    new_size) {
 181   _number_of_refills++;
 182   print_stats("fill");
 183   assert(top <= start + new_size - alignment_reserve(), "size too small");
 184 
 185   initialize(start, top, start + new_size - alignment_reserve());
 186 
 187   if (ThreadHeapSampler::enabled()) {
 188     set_sample_end();
 189   }
 190 
 191   // Reset amount of internal fragmentation
 192   set_refill_waste_limit(initial_refill_waste_limit());
 193 }
 194 
 195 void ThreadLocalAllocBuffer::initialize(HeapWord* start,
 196                                         HeapWord* top,
 197                                         HeapWord* end) {
 198   set_start(start);
 199   set_top(top);
 200   set_pf_top(top);
 201   set_fast_path_end(end);
 202   set_allocation_end(end);
 203   invariants();
 204 }
 205 
 206 void ThreadLocalAllocBuffer::initialize() {
 207   initialize(NULL,                    // start
 208              NULL,                    // top
 209              NULL);                   // end
 210 
 211   set_desired_size(initial_desired_size());
 212 
 213   // Following check is needed because at startup the main
 214   // thread is initialized before the heap is.  The initialization for
 215   // this thread is redone in startup_initialization below.
 216   if (Universe::heap() != NULL) {
 217     size_t capacity   = Universe::heap()->tlab_capacity(myThread()) / HeapWordSize;
 218     double alloc_frac = desired_size() * target_refills() / (double) capacity;
 219     _allocation_fraction.sample(alloc_frac);
 220   }
 221 
 222   set_refill_waste_limit(initial_refill_waste_limit());
 223 
 224   initialize_statistics();
 225 }
 226 
 227 void ThreadLocalAllocBuffer::startup_initialization() {
 228 
 229   // Assuming each thread's active tlab is, on average,
 230   // 1/2 full at a GC
 231   _target_refills = 100 / (2 * TLABWasteTargetPercent);
 232   _target_refills = MAX2(_target_refills, (unsigned)1U);
 233 
 234   _global_stats = new GlobalTLABStats();
 235 
 236 #ifdef COMPILER2
 237   // If the C2 compiler is present, extra space is needed at the end of
 238   // TLABs, otherwise prefetching instructions generated by the C2
 239   // compiler will fault (due to accessing memory outside of heap).
 240   // The amount of space is the max of the number of lines to
 241   // prefetch for array and for instance allocations. (Extra space must be
 242   // reserved to accommodate both types of allocations.)
 243   //
 244   // Only SPARC-specific BIS instructions are known to fault. (Those
 245   // instructions are generated if AllocatePrefetchStyle==3 and
 246   // AllocatePrefetchInstr==1). To be on the safe side, however,
 247   // extra space is reserved for all combinations of
 248   // AllocatePrefetchStyle and AllocatePrefetchInstr.
 249   //
 250   // If the C2 compiler is not present, no space is reserved.
 251 
 252   // +1 for rounding up to next cache line, +1 to be safe
 253   if (is_server_compilation_mode_vm()) {
 254     int lines =  MAX2(AllocatePrefetchLines, AllocateInstancePrefetchLines) + 2;
 255     _reserve_for_allocation_prefetch = (AllocatePrefetchDistance + AllocatePrefetchStepSize * lines) /
 256                                        (int)HeapWordSize;
 257   }
 258 #endif
 259 
 260   // During jvm startup, the main thread is initialized
 261   // before the heap is initialized.  So reinitialize it now.
 262   guarantee(Thread::current()->is_Java_thread(), "tlab initialization thread not Java thread");
 263   Thread::current()->tlab().initialize();
 264 
 265   log_develop_trace(gc, tlab)("TLAB min: " SIZE_FORMAT " initial: " SIZE_FORMAT " max: " SIZE_FORMAT,
 266                                min_size(), Thread::current()->tlab().initial_desired_size(), max_size());
 267 }
 268 
 269 size_t ThreadLocalAllocBuffer::initial_desired_size() {
 270   size_t init_sz = 0;
 271 
 272   if (TLABSize > 0) {
 273     init_sz = TLABSize / HeapWordSize;
 274   } else if (global_stats() != NULL) {
 275     // Initial size is a function of the average number of allocating threads.
 276     unsigned nof_threads = global_stats()->allocating_threads_avg();
 277 
 278     init_sz  = (Universe::heap()->tlab_capacity(myThread()) / HeapWordSize) /
 279                       (nof_threads * target_refills());
 280     init_sz = align_object_size(init_sz);
 281   }
 282   init_sz = MIN2(MAX2(init_sz, min_size()), max_size());
 283   return init_sz;
 284 }
 285 
 286 void ThreadLocalAllocBuffer::print_stats(const char* tag) {
 287   Log(gc, tlab) log;
 288   if (!log.is_trace()) {
 289     return;
 290   }
 291 
 292   Thread* thrd = myThread();
 293   size_t waste = _gc_waste + _slow_refill_waste + _fast_refill_waste;
 294   size_t alloc = _number_of_refills * _desired_size;
 295   double waste_percent = percent_of(waste, alloc);
 296   size_t tlab_used  = Universe::heap()->tlab_used(thrd);
 297   log.trace("TLAB: %s thread: " INTPTR_FORMAT " [id: %2d]"
 298             " desired_size: " SIZE_FORMAT "KB"
 299             " slow allocs: %d  refill waste: " SIZE_FORMAT "B"
 300             " alloc:%8.5f %8.0fKB refills: %d waste %4.1f%% gc: %dB"
 301             " slow: %dB fast: %dB",
 302             tag, p2i(thrd), thrd->osthread()->thread_id(),
 303             _desired_size / (K / HeapWordSize),
 304             _slow_allocations, _refill_waste_limit * HeapWordSize,
 305             _allocation_fraction.average(),
 306             _allocation_fraction.average() * tlab_used / K,
 307             _number_of_refills, waste_percent,
 308             _gc_waste * HeapWordSize,
 309             _slow_refill_waste * HeapWordSize,
 310             _fast_refill_waste * HeapWordSize);
 311 }
 312 
 313 void ThreadLocalAllocBuffer::verify() {
 314   HeapWord* p = start();
 315   HeapWord* t = top();
 316   HeapWord* prev_p = NULL;
 317   while (p < t) {
 318     oop(p)->verify();
 319     prev_p = p;
 320     p += oop(p)->size();
 321   }
 322   guarantee(p == top(), "end of last object must match end of space");
 323 }
 324 
 325 void ThreadLocalAllocBuffer::set_sample_end() {
 326   size_t heap_words_remaining = pointer_delta(_fast_path_end, _top);
 327   size_t bytes_until_sample = myThread()->heap_sampler().bytes_until_sample();
 328   size_t words_until_sample = bytes_until_sample / HeapWordSize;;
 329 
 330   if (heap_words_remaining > words_until_sample) {
 331     HeapWord* new_end = _top + words_until_sample;
 332     set_fast_path_end(new_end);
 333     _bytes_since_last_sample_point = bytes_until_sample;
 334   } else {
 335     _bytes_since_last_sample_point = heap_words_remaining * HeapWordSize;;
 336   }
 337 }
 338 
 339 Thread* ThreadLocalAllocBuffer::myThread() {
 340   return (Thread*)(((char *)this) +
 341                    in_bytes(start_offset()) -
 342                    in_bytes(Thread::tlab_start_offset()));
 343 }
 344 
 345 void ThreadLocalAllocBuffer::set_back_allocation_end() {
 346   _fast_path_end = _allocation_end;
 347 }
 348 
 349 HeapWord* ThreadLocalAllocBuffer::allocate_sampled_object(size_t size) {
 350   Thread* thread = myThread();
 351   thread->tlab().set_back_allocation_end();
 352   HeapWord* result = thread->tlab().allocate(size);
 353 
 354   if (result) {
 355     thread->heap_sampler().check_for_sampling(result, size * HeapWordSize, _bytes_since_last_sample_point);
 356     thread->tlab().set_sample_end();
 357   }
 358 
 359   return result;
 360 }
 361 
 362 HeapWord* ThreadLocalAllocBuffer::hard_end() {
 363   return _allocation_end + alignment_reserve();
 364 }
 365 
 366 GlobalTLABStats::GlobalTLABStats() :
 367   _allocating_threads_avg(TLABAllocationWeight) {
 368 
 369   initialize();
 370 
 371   _allocating_threads_avg.sample(1); // One allocating thread at startup
 372 
 373   if (UsePerfData) {
 374 
 375     EXCEPTION_MARK;
 376     ResourceMark rm;
 377 
 378     char* cname = PerfDataManager::counter_name("tlab", "allocThreads");
 379     _perf_allocating_threads =
 380       PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_None, CHECK);
 381 
 382     cname = PerfDataManager::counter_name("tlab", "fills");
 383     _perf_total_refills =
 384       PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_None, CHECK);
 385 
 386     cname = PerfDataManager::counter_name("tlab", "maxFills");
 387     _perf_max_refills =
 388       PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_None, CHECK);
 389 
 390     cname = PerfDataManager::counter_name("tlab", "alloc");
 391     _perf_allocation =
 392       PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_Bytes, CHECK);
 393 
 394     cname = PerfDataManager::counter_name("tlab", "gcWaste");
 395     _perf_gc_waste =
 396       PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_Bytes, CHECK);
 397 
 398     cname = PerfDataManager::counter_name("tlab", "maxGcWaste");
 399     _perf_max_gc_waste =
 400       PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_Bytes, CHECK);
 401 
 402     cname = PerfDataManager::counter_name("tlab", "slowWaste");
 403     _perf_slow_refill_waste =
 404       PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_Bytes, CHECK);
 405 
 406     cname = PerfDataManager::counter_name("tlab", "maxSlowWaste");
 407     _perf_max_slow_refill_waste =
 408       PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_Bytes, CHECK);
 409 
 410     cname = PerfDataManager::counter_name("tlab", "fastWaste");
 411     _perf_fast_refill_waste =
 412       PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_Bytes, CHECK);
 413 
 414     cname = PerfDataManager::counter_name("tlab", "maxFastWaste");
 415     _perf_max_fast_refill_waste =
 416       PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_Bytes, CHECK);
 417 
 418     cname = PerfDataManager::counter_name("tlab", "slowAlloc");
 419     _perf_slow_allocations =
 420       PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_None, CHECK);
 421 
 422     cname = PerfDataManager::counter_name("tlab", "maxSlowAlloc");
 423     _perf_max_slow_allocations =
 424       PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_None, CHECK);
 425   }
 426 }
 427 
 428 void GlobalTLABStats::initialize() {
 429   // Clear counters summarizing info from all threads
 430   _allocating_threads      = 0;
 431   _total_refills           = 0;
 432   _max_refills             = 0;
 433   _total_allocation        = 0;
 434   _total_gc_waste          = 0;
 435   _max_gc_waste            = 0;
 436   _total_slow_refill_waste = 0;
 437   _max_slow_refill_waste   = 0;
 438   _total_fast_refill_waste = 0;
 439   _max_fast_refill_waste   = 0;
 440   _total_slow_allocations  = 0;
 441   _max_slow_allocations    = 0;
 442 }
 443 
 444 void GlobalTLABStats::publish() {
 445   _allocating_threads_avg.sample(_allocating_threads);
 446   if (UsePerfData) {
 447     _perf_allocating_threads   ->set_value(_allocating_threads);
 448     _perf_total_refills        ->set_value(_total_refills);
 449     _perf_max_refills          ->set_value(_max_refills);
 450     _perf_allocation           ->set_value(_total_allocation);
 451     _perf_gc_waste             ->set_value(_total_gc_waste);
 452     _perf_max_gc_waste         ->set_value(_max_gc_waste);
 453     _perf_slow_refill_waste    ->set_value(_total_slow_refill_waste);
 454     _perf_max_slow_refill_waste->set_value(_max_slow_refill_waste);
 455     _perf_fast_refill_waste    ->set_value(_total_fast_refill_waste);
 456     _perf_max_fast_refill_waste->set_value(_max_fast_refill_waste);
 457     _perf_slow_allocations     ->set_value(_total_slow_allocations);
 458     _perf_max_slow_allocations ->set_value(_max_slow_allocations);
 459   }
 460 }
 461 
 462 void GlobalTLABStats::print() {
 463   Log(gc, tlab) log;
 464   if (!log.is_debug()) {
 465     return;
 466   }
 467 
 468   size_t waste = _total_gc_waste + _total_slow_refill_waste + _total_fast_refill_waste;
 469   double waste_percent = percent_of(waste, _total_allocation);
 470   log.debug("TLAB totals: thrds: %d  refills: %d max: %d"
 471             " slow allocs: %d max %d waste: %4.1f%%"
 472             " gc: " SIZE_FORMAT "B max: " SIZE_FORMAT "B"
 473             " slow: " SIZE_FORMAT "B max: " SIZE_FORMAT "B"
 474             " fast: " SIZE_FORMAT "B max: " SIZE_FORMAT "B",
 475             _allocating_threads,
 476             _total_refills, _max_refills,
 477             _total_slow_allocations, _max_slow_allocations,
 478             waste_percent,
 479             _total_gc_waste * HeapWordSize,
 480             _max_gc_waste * HeapWordSize,
 481             _total_slow_refill_waste * HeapWordSize,
 482             _max_slow_refill_waste * HeapWordSize,
 483             _total_fast_refill_waste * HeapWordSize,
 484             _max_fast_refill_waste * HeapWordSize);
 485 }
--- EOF ---