1 /*
   2  * Copyright (c) 1999, 2017, 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/heapMonitoring.hpp"
  33 #include "runtime/thread.inline.hpp"
  34 #include "runtime/threadSMR.hpp"
  35 #include "utilities/copy.hpp"
  36 
  37 // Thread-Local Edens support
  38 
  39 // static member initialization
  40 size_t           ThreadLocalAllocBuffer::_max_size       = 0;
  41 int              ThreadLocalAllocBuffer::_reserve_for_allocation_prefetch = 0;
  42 unsigned         ThreadLocalAllocBuffer::_target_refills = 0;
  43 GlobalTLABStats* ThreadLocalAllocBuffer::_global_stats   = NULL;
  44 
  45 void ThreadLocalAllocBuffer::clear_before_allocation() {
  46   _slow_refill_waste += (unsigned)remaining();
  47   make_parsable(true);   // also retire the TLAB
  48 }
  49 
  50 void ThreadLocalAllocBuffer::accumulate_statistics_before_gc() {
  51   global_stats()->initialize();
  52 
  53   for (JavaThreadIteratorWithHandle jtiwh; JavaThread *thread = jtiwh.next(); ) {
  54     thread->tlab().accumulate_statistics();
  55     thread->tlab().initialize_statistics();
  56   }
  57 
  58   // Publish new stats if some allocation occurred.
  59   if (global_stats()->allocation() != 0) {
  60     global_stats()->publish();
  61     global_stats()->print();
  62   }
  63 }
  64 
  65 void ThreadLocalAllocBuffer::accumulate_statistics() {
  66   Thread* thread = myThread();
  67   size_t capacity = Universe::heap()->tlab_capacity(thread);
  68   size_t used     = Universe::heap()->tlab_used(thread);
  69 
  70   _gc_waste += (unsigned)remaining();
  71   size_t total_allocated = thread->allocated_bytes();
  72   size_t allocated_since_last_gc = total_allocated - _allocated_before_last_gc;
  73   _allocated_before_last_gc = total_allocated;
  74 
  75   print_stats("gc");
  76 
  77   if (_number_of_refills > 0) {
  78     // Update allocation history if a reasonable amount of eden was allocated.
  79     bool update_allocation_history = used > 0.5 * capacity;
  80 
  81     if (update_allocation_history) {
  82       // Average the fraction of eden allocated in a tlab by this
  83       // thread for use in the next resize operation.
  84       // _gc_waste is not subtracted because it's included in
  85       // "used".
  86       // The result can be larger than 1.0 due to direct to old allocations.
  87       // These allocations should ideally not be counted but since it is not possible
  88       // to filter them out here we just cap the fraction to be at most 1.0.
  89       double alloc_frac = MIN2(1.0, (double) allocated_since_last_gc / used);
  90       _allocation_fraction.sample(alloc_frac);
  91     }
  92     global_stats()->update_allocating_threads();
  93     global_stats()->update_number_of_refills(_number_of_refills);
  94     global_stats()->update_allocation(_number_of_refills * desired_size());
  95     global_stats()->update_gc_waste(_gc_waste);
  96     global_stats()->update_slow_refill_waste(_slow_refill_waste);
  97     global_stats()->update_fast_refill_waste(_fast_refill_waste);
  98 
  99   } else {
 100     assert(_number_of_refills == 0 && _fast_refill_waste == 0 &&
 101            _slow_refill_waste == 0 && _gc_waste          == 0,
 102            "tlab stats == 0");
 103   }
 104   global_stats()->update_slow_allocations(_slow_allocations);
 105 }
 106 
 107 // Fills the current tlab with a dummy filler array to create
 108 // an illusion of a contiguous Eden and optionally retires the tlab.
 109 // Waste accounting should be done in caller as appropriate; see,
 110 // for example, clear_before_allocation().
 111 void ThreadLocalAllocBuffer::make_parsable(bool retire, bool zap) {
 112   if (end() != NULL) {
 113     invariants();
 114 
 115     if (retire) {
 116       myThread()->incr_allocated_bytes(used_bytes());
 117     }
 118 
 119     CollectedHeap::fill_with_object(top(), hard_end(), retire && zap);
 120 
 121     if (retire || ZeroTLAB) {  // "Reset" the TLAB
 122       set_start(NULL);
 123       set_top(NULL);
 124       set_pf_top(NULL);
 125       set_end(NULL);
 126       set_actual_end(NULL);
 127       set_slow_path_end(NULL);
 128     }
 129   }
 130   assert(!(retire || ZeroTLAB)  ||
 131          (start() == NULL && end() == NULL && top() == NULL &&
 132           _actual_end == NULL && _slow_path_end == NULL),
 133          "TLAB must be reset");
 134 }
 135 
 136 void ThreadLocalAllocBuffer::resize_all_tlabs() {
 137   if (ResizeTLAB) {
 138     for (JavaThreadIteratorWithHandle jtiwh; JavaThread *thread = jtiwh.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   log_trace(gc, tlab)("TLAB new size: thread: " INTPTR_FORMAT " [id: %2d]"
 156                       " refills %d  alloc: %8.6f desired_size: " SIZE_FORMAT " -> " SIZE_FORMAT,
 157                       p2i(myThread()), myThread()->osthread()->thread_id(),
 158                       _target_refills, _allocation_fraction.average(), desired_size(), aligned_new_size);
 159 
 160   set_desired_size(aligned_new_size);
 161   set_refill_waste_limit(initial_refill_waste_limit());
 162 }
 163 
 164 void ThreadLocalAllocBuffer::initialize_statistics() {
 165     _number_of_refills = 0;
 166     _fast_refill_waste = 0;
 167     _slow_refill_waste = 0;
 168     _gc_waste          = 0;
 169     _slow_allocations  = 0;
 170 }
 171 
 172 void ThreadLocalAllocBuffer::fill(HeapWord* start,
 173                                   HeapWord* top,
 174                                   size_t    new_size) {
 175   _number_of_refills++;
 176   print_stats("fill");
 177   assert(top <= start + new_size - alignment_reserve(), "size too small");
 178 
 179   // Remember old bytes until sample for the next tlab only if this is our first
 180   // actual refill.
 181   size_t old_bytes_until_sample = 0;
 182   if (_number_of_refills > 1) {
 183     old_bytes_until_sample = _bytes_until_sample;
 184   }
 185 
 186   initialize(start, top, start + new_size - alignment_reserve());
 187 
 188   if (old_bytes_until_sample > 0) {
 189     set_bytes_until_sample(old_bytes_until_sample);
 190     set_sample_end();
 191   }
 192 
 193   // Reset amount of internal fragmentation
 194   set_refill_waste_limit(initial_refill_waste_limit());
 195 }
 196 
 197 void ThreadLocalAllocBuffer::initialize(HeapWord* start,
 198                                         HeapWord* top,
 199                                         HeapWord* end) {
 200   set_start(start);
 201   set_top(top);
 202   set_pf_top(top);
 203   set_end(end);
 204   set_actual_end(end);
 205   set_slow_path_end(end);
 206   invariants();
 207   _bytes_until_sample = 0;
 208 }
 209 
 210 void ThreadLocalAllocBuffer::initialize() {
 211   initialize(NULL,                    // start
 212              NULL,                    // top
 213              NULL);                   // end
 214 
 215   set_desired_size(initial_desired_size());
 216 
 217   // Following check is needed because at startup the main
 218   // thread is initialized before the heap is.  The initialization for
 219   // this thread is redone in startup_initialization below.
 220   if (Universe::heap() != NULL) {
 221     size_t capacity   = Universe::heap()->tlab_capacity(myThread()) / HeapWordSize;
 222     double alloc_frac = desired_size() * target_refills() / (double) capacity;
 223     _allocation_fraction.sample(alloc_frac);
 224   }
 225 
 226   set_refill_waste_limit(initial_refill_waste_limit());
 227 
 228   initialize_statistics();
 229 }
 230 
 231 void ThreadLocalAllocBuffer::startup_initialization() {
 232 
 233   // Assuming each thread's active tlab is, on average,
 234   // 1/2 full at a GC
 235   _target_refills = 100 / (2 * TLABWasteTargetPercent);
 236   _target_refills = MAX2(_target_refills, (unsigned)1U);
 237 
 238   _global_stats = new GlobalTLABStats();
 239 
 240 #ifdef COMPILER2
 241   // If the C2 compiler is present, extra space is needed at the end of
 242   // TLABs, otherwise prefetching instructions generated by the C2
 243   // compiler will fault (due to accessing memory outside of heap).
 244   // The amount of space is the max of the number of lines to
 245   // prefetch for array and for instance allocations. (Extra space must be
 246   // reserved to accommodate both types of allocations.)
 247   //
 248   // Only SPARC-specific BIS instructions are known to fault. (Those
 249   // instructions are generated if AllocatePrefetchStyle==3 and
 250   // AllocatePrefetchInstr==1). To be on the safe side, however,
 251   // extra space is reserved for all combinations of
 252   // AllocatePrefetchStyle and AllocatePrefetchInstr.
 253   //
 254   // If the C2 compiler is not present, no space is reserved.
 255 
 256   // +1 for rounding up to next cache line, +1 to be safe
 257   if (is_server_compilation_mode_vm()) {
 258     int lines =  MAX2(AllocatePrefetchLines, AllocateInstancePrefetchLines) + 2;
 259     _reserve_for_allocation_prefetch = (AllocatePrefetchDistance + AllocatePrefetchStepSize * lines) /
 260                                        (int)HeapWordSize;
 261   }
 262 #endif
 263 
 264   // During jvm startup, the main thread is initialized
 265   // before the heap is initialized.  So reinitialize it now.
 266   guarantee(Thread::current()->is_Java_thread(), "tlab initialization thread not Java thread");
 267   Thread::current()->tlab().initialize();
 268 
 269   log_develop_trace(gc, tlab)("TLAB min: " SIZE_FORMAT " initial: " SIZE_FORMAT " max: " SIZE_FORMAT,
 270                                min_size(), Thread::current()->tlab().initial_desired_size(), max_size());
 271 }
 272 
 273 size_t ThreadLocalAllocBuffer::initial_desired_size() {
 274   size_t init_sz = 0;
 275 
 276   if (TLABSize > 0) {
 277     init_sz = TLABSize / HeapWordSize;
 278   } else if (global_stats() != NULL) {
 279     // Initial size is a function of the average number of allocating threads.
 280     unsigned nof_threads = global_stats()->allocating_threads_avg();
 281 
 282     init_sz  = (Universe::heap()->tlab_capacity(myThread()) / HeapWordSize) /
 283                       (nof_threads * target_refills());
 284     init_sz = align_object_size(init_sz);
 285   }
 286   init_sz = MIN2(MAX2(init_sz, min_size()), max_size());
 287   return init_sz;
 288 }
 289 
 290 void ThreadLocalAllocBuffer::print_stats(const char* tag) {
 291   Log(gc, tlab) log;
 292   if (!log.is_trace()) {
 293     return;
 294   }
 295 
 296   Thread* thrd = myThread();
 297   size_t waste = _gc_waste + _slow_refill_waste + _fast_refill_waste;
 298   size_t alloc = _number_of_refills * _desired_size;
 299   double waste_percent = percent_of(waste, alloc);
 300   size_t tlab_used  = Universe::heap()->tlab_used(thrd);
 301   log.trace("TLAB: %s thread: " INTPTR_FORMAT " [id: %2d]"
 302             " desired_size: " SIZE_FORMAT "KB"
 303             " slow allocs: %d  refill waste: " SIZE_FORMAT "B"
 304             " alloc:%8.5f %8.0fKB refills: %d waste %4.1f%% gc: %dB"
 305             " slow: %dB fast: %dB",
 306             tag, p2i(thrd), thrd->osthread()->thread_id(),
 307             _desired_size / (K / HeapWordSize),
 308             _slow_allocations, _refill_waste_limit * HeapWordSize,
 309             _allocation_fraction.average(),
 310             _allocation_fraction.average() * tlab_used / K,
 311             _number_of_refills, waste_percent,
 312             _gc_waste * HeapWordSize,
 313             _slow_refill_waste * HeapWordSize,
 314             _fast_refill_waste * HeapWordSize);
 315 }
 316 
 317 void ThreadLocalAllocBuffer::verify() {
 318   HeapWord* p = start();
 319   HeapWord* t = top();
 320   HeapWord* prev_p = NULL;
 321   while (p < t) {
 322     oop(p)->verify();
 323     prev_p = p;
 324     p += oop(p)->size();
 325   }
 326   guarantee(p == top(), "end of last object must match end of space");
 327 }
 328 
 329 void ThreadLocalAllocBuffer::set_sample_end() {
 330   size_t heap_words_remaining = pointer_delta(_end, _top);
 331   size_t bytes_left = _bytes_until_sample;
 332   size_t words_until_sample = bytes_left / HeapWordSize;
 333 
 334   if (heap_words_remaining > words_until_sample) {
 335     HeapWord* new_end = _top + words_until_sample;
 336     set_end(new_end);
 337     set_slow_path_end(new_end);
 338     set_bytes_until_sample(0);
 339   } else {
 340     bytes_left -= heap_words_remaining * HeapWordSize;
 341     set_bytes_until_sample(bytes_left);
 342   }
 343 }
 344 
 345 void ThreadLocalAllocBuffer::pick_next_sample(size_t overflowed_words) {
 346   if (!HeapMonitoring::enabled()) {
 347     return;
 348   }
 349 
 350   if (_bytes_until_sample == 0) {
 351     HeapMonitoring::pick_next_sample(&_bytes_until_sample);
 352   }
 353 
 354   if (overflowed_words > 0) {
 355     // Try to correct sample size by removing extra space from last allocation.
 356     if (_bytes_until_sample > overflowed_words * HeapWordSize) {
 357       set_bytes_until_sample(_bytes_until_sample - overflowed_words * HeapWordSize);
 358     }
 359   }
 360 
 361   set_sample_end();
 362 
 363   log_trace(gc, tlab)("TLAB picked next sample: thread: " INTPTR_FORMAT " [id: %2d]"
 364                       " start: " INTPTR_FORMAT " top: " INTPTR_FORMAT " end: " INTPTR_FORMAT " actual_end:"
 365                       INTPTR_FORMAT " slow_path_end: " INTPTR_FORMAT,
 366                       p2i(myThread()), myThread()->osthread()->thread_id(),
 367                       p2i(start()), p2i(top()), p2i(end()),
 368                       p2i(_actual_end), p2i(_slow_path_end));
 369 }
 370 
 371 Thread* ThreadLocalAllocBuffer::myThread() {
 372   return (Thread*)(((char *)this) +
 373                    in_bytes(start_offset()) -
 374                    in_bytes(Thread::tlab_start_offset()));
 375 }
 376 
 377 void ThreadLocalAllocBuffer::set_back_actual_end() {
 378   // Did a fast TLAB refill occur?
 379   if (_slow_path_end != _end) {
 380     // Fix up the actual end to be now the end of this TLAB.
 381     _slow_path_end = _end;
 382     _actual_end = _end;
 383   } else {
 384     _end = _actual_end;
 385   }
 386 }
 387 
 388 void ThreadLocalAllocBuffer::handle_sample(Thread* thread, HeapWord* result,
 389                                            size_t size) {
 390   if (!HeapMonitoring::enabled()) {
 391     return;
 392   }
 393 
 394   size_t size_in_bytes = size * HeapWordSize;
 395   if (_bytes_until_sample > size_in_bytes) {
 396     set_bytes_until_sample(_bytes_until_sample - size_in_bytes);
 397   } else {
 398     // Technically this is not exactly right, we probably should remember how many bytes are
 399     // negative probably to then reduce our next sample size.
 400     set_bytes_until_sample(0);
 401   }
 402 
 403   // Should we sample now?
 404   if (should_sample()) {
 405     HeapMonitoring::object_alloc_do_sample(thread,
 406                                            reinterpret_cast<oopDesc*>(result),
 407                                            size_in_bytes);
 408     set_back_actual_end();
 409     pick_next_sample();
 410   }
 411 }
 412 
 413 HeapWord* ThreadLocalAllocBuffer::hard_end() {
 414   // Did a fast TLAB refill occur?
 415   if (_slow_path_end != _end) {
 416     // Fix up the actual end to be now the end of this TLAB.
 417     _slow_path_end = _end;
 418     _actual_end = _end;
 419   }
 420 
 421   return _actual_end + alignment_reserve();
 422 }
 423 
 424 GlobalTLABStats::GlobalTLABStats() :
 425   _allocating_threads_avg(TLABAllocationWeight) {
 426 
 427   initialize();
 428 
 429   _allocating_threads_avg.sample(1); // One allocating thread at startup
 430 
 431   if (UsePerfData) {
 432 
 433     EXCEPTION_MARK;
 434     ResourceMark rm;
 435 
 436     char* cname = PerfDataManager::counter_name("tlab", "allocThreads");
 437     _perf_allocating_threads =
 438       PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_None, CHECK);
 439 
 440     cname = PerfDataManager::counter_name("tlab", "fills");
 441     _perf_total_refills =
 442       PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_None, CHECK);
 443 
 444     cname = PerfDataManager::counter_name("tlab", "maxFills");
 445     _perf_max_refills =
 446       PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_None, CHECK);
 447 
 448     cname = PerfDataManager::counter_name("tlab", "alloc");
 449     _perf_allocation =
 450       PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_Bytes, CHECK);
 451 
 452     cname = PerfDataManager::counter_name("tlab", "gcWaste");
 453     _perf_gc_waste =
 454       PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_Bytes, CHECK);
 455 
 456     cname = PerfDataManager::counter_name("tlab", "maxGcWaste");
 457     _perf_max_gc_waste =
 458       PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_Bytes, CHECK);
 459 
 460     cname = PerfDataManager::counter_name("tlab", "slowWaste");
 461     _perf_slow_refill_waste =
 462       PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_Bytes, CHECK);
 463 
 464     cname = PerfDataManager::counter_name("tlab", "maxSlowWaste");
 465     _perf_max_slow_refill_waste =
 466       PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_Bytes, CHECK);
 467 
 468     cname = PerfDataManager::counter_name("tlab", "fastWaste");
 469     _perf_fast_refill_waste =
 470       PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_Bytes, CHECK);
 471 
 472     cname = PerfDataManager::counter_name("tlab", "maxFastWaste");
 473     _perf_max_fast_refill_waste =
 474       PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_Bytes, CHECK);
 475 
 476     cname = PerfDataManager::counter_name("tlab", "slowAlloc");
 477     _perf_slow_allocations =
 478       PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_None, CHECK);
 479 
 480     cname = PerfDataManager::counter_name("tlab", "maxSlowAlloc");
 481     _perf_max_slow_allocations =
 482       PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_None, CHECK);
 483   }
 484 }
 485 
 486 void GlobalTLABStats::initialize() {
 487   // Clear counters summarizing info from all threads
 488   _allocating_threads      = 0;
 489   _total_refills           = 0;
 490   _max_refills             = 0;
 491   _total_allocation        = 0;
 492   _total_gc_waste          = 0;
 493   _max_gc_waste            = 0;
 494   _total_slow_refill_waste = 0;
 495   _max_slow_refill_waste   = 0;
 496   _total_fast_refill_waste = 0;
 497   _max_fast_refill_waste   = 0;
 498   _total_slow_allocations  = 0;
 499   _max_slow_allocations    = 0;
 500 }
 501 
 502 void GlobalTLABStats::publish() {
 503   _allocating_threads_avg.sample(_allocating_threads);
 504   if (UsePerfData) {
 505     _perf_allocating_threads   ->set_value(_allocating_threads);
 506     _perf_total_refills        ->set_value(_total_refills);
 507     _perf_max_refills          ->set_value(_max_refills);
 508     _perf_allocation           ->set_value(_total_allocation);
 509     _perf_gc_waste             ->set_value(_total_gc_waste);
 510     _perf_max_gc_waste         ->set_value(_max_gc_waste);
 511     _perf_slow_refill_waste    ->set_value(_total_slow_refill_waste);
 512     _perf_max_slow_refill_waste->set_value(_max_slow_refill_waste);
 513     _perf_fast_refill_waste    ->set_value(_total_fast_refill_waste);
 514     _perf_max_fast_refill_waste->set_value(_max_fast_refill_waste);
 515     _perf_slow_allocations     ->set_value(_total_slow_allocations);
 516     _perf_max_slow_allocations ->set_value(_max_slow_allocations);
 517   }
 518 }
 519 
 520 void GlobalTLABStats::print() {
 521   Log(gc, tlab) log;
 522   if (!log.is_debug()) {
 523     return;
 524   }
 525 
 526   size_t waste = _total_gc_waste + _total_slow_refill_waste + _total_fast_refill_waste;
 527   double waste_percent = percent_of(waste, _total_allocation);
 528   log.debug("TLAB totals: thrds: %d  refills: %d max: %d"
 529             " slow allocs: %d max %d waste: %4.1f%%"
 530             " gc: " SIZE_FORMAT "B max: " SIZE_FORMAT "B"
 531             " slow: " SIZE_FORMAT "B max: " SIZE_FORMAT "B"
 532             " fast: " SIZE_FORMAT "B max: " SIZE_FORMAT "B",
 533             _allocating_threads,
 534             _total_refills, _max_refills,
 535             _total_slow_allocations, _max_slow_allocations,
 536             waste_percent,
 537             _total_gc_waste * HeapWordSize,
 538             _max_gc_waste * HeapWordSize,
 539             _total_slow_refill_waste * HeapWordSize,
 540             _max_slow_refill_waste * HeapWordSize,
 541             _total_fast_refill_waste * HeapWordSize,
 542             _max_fast_refill_waste * HeapWordSize);
 543 }