1 /*
   2  * Copyright (c) 1999, 2014, 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 "memory/genCollectedHeap.hpp"
  27 #include "memory/resourceArea.hpp"
  28 #include "memory/threadLocalAllocBuffer.inline.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 PRAGMA_FORMAT_MUTE_WARNINGS_FOR_GCC
  35 
  36 // Thread-Local Edens support
  37 
  38 // static member initialization
  39 size_t           ThreadLocalAllocBuffer::_max_size       = 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   Universe::heap()->accumulate_statistics_all_gclabs();
  57 
  58   // Publish new stats if some allocation occurred.
  59   if (global_stats()->allocation() != 0) {
  60     global_stats()->publish();
  61     if (PrintTLAB) {
  62       global_stats()->print();
  63     }
  64   }
  65 }
  66 
  67 void ThreadLocalAllocBuffer::accumulate_statistics() {
  68   Thread* thread = myThread();
  69   size_t capacity = Universe::heap()->tlab_capacity(thread);
  70   size_t used     = Universe::heap()->tlab_used(thread);
  71 
  72   _gc_waste += (unsigned)remaining();
  73   size_t total_allocated = _gclab ? thread->allocated_bytes_gclab() : thread->allocated_bytes();
  74   size_t allocated_since_last_gc = total_allocated - _allocated_before_last_gc;
  75   _allocated_before_last_gc = total_allocated;
  76 
  77   if (PrintTLAB && (_number_of_refills > 0 || Verbose)) {
  78     print_stats("gc");
  79   }
  80 
  81   if (_number_of_refills > 0) {
  82     // Update allocation history if a reasonable amount of eden was allocated.
  83     bool update_allocation_history = used > 0.5 * capacity;
  84 
  85     if (update_allocation_history) {
  86       // Average the fraction of eden allocated in a tlab by this
  87       // thread for use in the next resize operation.
  88       // _gc_waste is not subtracted because it's included in
  89       // "used".
  90       // The result can be larger than 1.0 due to direct to old allocations.
  91       // These allocations should ideally not be counted but since it is not possible
  92       // to filter them out here we just cap the fraction to be at most 1.0.
  93       double alloc_frac = MIN2(1.0, (double) allocated_since_last_gc / used);
  94       _allocation_fraction.sample(alloc_frac);
  95     }
  96     global_stats()->update_allocating_threads();
  97     global_stats()->update_number_of_refills(_number_of_refills);
  98     global_stats()->update_allocation(_number_of_refills * desired_size());
  99     global_stats()->update_gc_waste(_gc_waste);
 100     global_stats()->update_slow_refill_waste(_slow_refill_waste);
 101     global_stats()->update_fast_refill_waste(_fast_refill_waste);
 102 
 103   } else {
 104     assert(_number_of_refills == 0 && _fast_refill_waste == 0 &&
 105            _slow_refill_waste == 0 && _gc_waste          == 0,
 106            "tlab stats == 0");
 107   }
 108   global_stats()->update_slow_allocations(_slow_allocations);
 109 }
 110 
 111 // Fills the current tlab with a dummy filler array to create
 112 // an illusion of a contiguous Eden and optionally retires the tlab.
 113 // Waste accounting should be done in caller as appropriate; see,
 114 // for example, clear_before_allocation().
 115 void ThreadLocalAllocBuffer::make_parsable(bool retire) {
 116   if (end() != NULL) {
 117     invariants();
 118 
 119     if (retire) {
 120       if (_gclab) {
 121         myThread()->incr_allocated_bytes_gclab(used_bytes());
 122       } else {
 123         myThread()->incr_allocated_bytes(used_bytes());
 124       }
 125     }
 126 
 127     HeapWord* obj = Universe::heap()->tlab_post_allocation_setup(top());
 128     CollectedHeap::fill_with_object(obj, hard_end(), retire);
 129 
 130     if (retire || ZeroTLAB) {  // "Reset" the TLAB
 131       set_start(NULL);
 132       set_top(NULL);
 133       set_pf_top(NULL);
 134       set_end(NULL);
 135     }
 136   }
 137   assert(!(retire || ZeroTLAB)  ||
 138          (start() == NULL && end() == NULL && top() == NULL),
 139          "TLAB must be reset");
 140 }
 141 
 142 void ThreadLocalAllocBuffer::resize_all_tlabs() {
 143   if (ResizeTLAB) {
 144     for (JavaThread *thread = Threads::first(); thread != NULL; thread = thread->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   if (PrintTLAB && Verbose) {
 162     gclog_or_tty->print("TLAB new size: thread: " INTPTR_FORMAT " [id: %2d]"
 163                         " refills %d  alloc: %8.6f desired_size: " SIZE_FORMAT " -> " SIZE_FORMAT "\n",
 164                         myThread(), myThread()->osthread()->thread_id(),
 165                         _target_refills, _allocation_fraction.average(), desired_size(), aligned_new_size);
 166   }
 167   set_desired_size(aligned_new_size);
 168   set_refill_waste_limit(initial_refill_waste_limit());
 169 }
 170 
 171 void ThreadLocalAllocBuffer::initialize_statistics() {
 172     _number_of_refills = 0;
 173     _fast_refill_waste = 0;
 174     _slow_refill_waste = 0;
 175     _gc_waste          = 0;
 176     _slow_allocations  = 0;
 177 }
 178 
 179 void ThreadLocalAllocBuffer::fill(HeapWord* start,
 180                                   HeapWord* top,
 181                                   size_t    new_size) {
 182   _number_of_refills++;
 183   if (PrintTLAB && Verbose) {
 184     print_stats("fill");
 185   }
 186   assert(top <= start + new_size - alignment_reserve(), "size too small");
 187   initialize(start, top, start + new_size - alignment_reserve());
 188 
 189   // Reset amount of internal fragmentation
 190   set_refill_waste_limit(initial_refill_waste_limit());
 191 }
 192 
 193 void ThreadLocalAllocBuffer::initialize(HeapWord* start,
 194                                         HeapWord* top,
 195                                         HeapWord* end) {
 196   set_start(start);
 197   set_top(top);
 198   set_pf_top(top);
 199   set_end(end);
 200   invariants();
 201 }
 202 
 203 void ThreadLocalAllocBuffer::initialize(bool gclab) {
 204   _initialized = true;
 205   _gclab = gclab;
 206   initialize(NULL,                    // start
 207              NULL,                    // top
 208              NULL);                   // end
 209 
 210   set_desired_size(initial_desired_size());
 211 
 212   // Following check is needed because at startup the main (primordial)
 213   // thread is initialized before the heap is.  The initialization for
 214   // this thread is redone in startup_initialization below.
 215   if (Universe::heap() != NULL) {
 216     size_t capacity   = Universe::heap()->tlab_capacity(myThread()) / HeapWordSize;
 217     double alloc_frac = desired_size() * target_refills() / (double) capacity;
 218     _allocation_fraction.sample(alloc_frac);
 219   }
 220 
 221   set_refill_waste_limit(initial_refill_waste_limit());
 222 
 223   initialize_statistics();
 224 }
 225 
 226 void ThreadLocalAllocBuffer::startup_initialization() {
 227 
 228   // Assuming each thread's active tlab is, on average,
 229   // 1/2 full at a GC
 230   _target_refills = 100 / (2 * TLABWasteTargetPercent);
 231   _target_refills = MAX2(_target_refills, (unsigned)1U);
 232 
 233   _global_stats = new GlobalTLABStats();
 234 
 235   // During jvm startup, the main (primordial) thread is initialized
 236   // before the heap is initialized.  So reinitialize it now.
 237   guarantee(Thread::current()->is_Java_thread(), "tlab initialization thread not Java thread");
 238   Thread::current()->tlab().initialize(false);
 239   if (UseShenandoahGC) {
 240     Thread::current()->gclab().initialize(true);
 241   }
 242 
 243   if (PrintTLAB && Verbose) {
 244     gclog_or_tty->print("TLAB min: " SIZE_FORMAT " initial: " SIZE_FORMAT " max: " SIZE_FORMAT "\n",
 245                         min_size(), Thread::current()->tlab().initial_desired_size(), max_size());
 246   }
 247 }
 248 
 249 size_t ThreadLocalAllocBuffer::initial_desired_size() {
 250   size_t init_sz = 0;
 251 
 252   if (TLABSize > 0) {
 253     init_sz = TLABSize / HeapWordSize;
 254   } else if (global_stats() != NULL) {
 255     // Initial size is a function of the average number of allocating threads.
 256     unsigned nof_threads = global_stats()->allocating_threads_avg();
 257 
 258     init_sz  = (Universe::heap()->tlab_capacity(myThread()) / HeapWordSize) /
 259                       (nof_threads * target_refills());
 260     init_sz = align_object_size(init_sz);
 261   }
 262   init_sz = MIN2(MAX2(init_sz, min_size()), max_size());
 263   return init_sz;
 264 }
 265 
 266 void ThreadLocalAllocBuffer::print_stats(const char* tag) {
 267   Thread* thrd = myThread();
 268   size_t waste = _gc_waste + _slow_refill_waste + _fast_refill_waste;
 269   size_t alloc = _number_of_refills * _desired_size;
 270   double waste_percent = alloc == 0 ? 0.0 :
 271                       100.0 * waste / alloc;
 272   size_t tlab_used  = Universe::heap()->tlab_used(thrd);
 273   gclog_or_tty->print("TLAB: %s %s thread: " INTPTR_FORMAT " [id: %2d]"
 274                       " desired_size: " SIZE_FORMAT "KB"
 275                       " slow allocs: %d  refill waste: " SIZE_FORMAT "B"
 276                       " alloc:%8.5f %8.0fKB refills: %d waste %4.1f%% gc: %dB"
 277                       " slow: %dB fast: %dB\n",
 278                       tag, _gclab ? "gclab" : "tlab ", p2i(thrd), thrd->osthread()->thread_id(),
 279                       _desired_size / (K / HeapWordSize),
 280                       _slow_allocations, _refill_waste_limit * HeapWordSize,
 281                       _allocation_fraction.average(),
 282                       _allocation_fraction.average() * tlab_used / K,
 283                       _number_of_refills, waste_percent,
 284                       _gc_waste * HeapWordSize,
 285                       _slow_refill_waste * HeapWordSize,
 286                       _fast_refill_waste * HeapWordSize);
 287 }
 288 
 289 void ThreadLocalAllocBuffer::verify() {
 290   HeapWord* p = start();
 291   HeapWord* t = top();
 292   HeapWord* prev_p = NULL;
 293   while (p < t) {
 294     oop(p)->verify();
 295     prev_p = p;
 296     p += oop(p)->size();
 297   }
 298   guarantee(p == top(), "end of last object must match end of space");
 299 }
 300 
 301 Thread* ThreadLocalAllocBuffer::myThread() {
 302   ByteSize gclab_offset = Thread::gclab_start_offset();
 303   ByteSize tlab_offset = Thread::tlab_start_offset();
 304   ByteSize offs = _gclab ? gclab_offset : tlab_offset;
 305   Thread* thread = (Thread*)(((char *)this) +
 306                    in_bytes(start_offset()) - in_bytes(offs));
 307 #ifdef ASSERT
 308   assert(this == (_gclab ? &thread->gclab() : &thread->tlab()), "must be");
 309 #endif
 310   return thread;
 311 }
 312 
 313 size_t ThreadLocalAllocBuffer::end_reserve() {
 314   int reserve_size = typeArrayOopDesc::header_size(T_INT) + Universe::heap()->oop_extra_words();
 315   return MAX2(reserve_size, VM_Version::reserve_for_allocation_prefetch());
 316 }
 317 
 318 void ThreadLocalAllocBuffer::rollback(size_t size) {
 319   HeapWord* old_top = top();
 320   if (old_top != NULL) { // Pathological case: we accept that we can't rollback.
 321     set_top(old_top - size);
 322   }
 323 }
 324 
 325 
 326 GlobalTLABStats::GlobalTLABStats() :
 327   _allocating_threads_avg(TLABAllocationWeight) {
 328 
 329   initialize();
 330 
 331   _allocating_threads_avg.sample(1); // One allocating thread at startup
 332 
 333   if (UsePerfData) {
 334 
 335     EXCEPTION_MARK;
 336     ResourceMark rm;
 337 
 338     char* cname = PerfDataManager::counter_name("tlab", "allocThreads");
 339     _perf_allocating_threads =
 340       PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_None, CHECK);
 341 
 342     cname = PerfDataManager::counter_name("tlab", "fills");
 343     _perf_total_refills =
 344       PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_None, CHECK);
 345 
 346     cname = PerfDataManager::counter_name("tlab", "maxFills");
 347     _perf_max_refills =
 348       PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_None, CHECK);
 349 
 350     cname = PerfDataManager::counter_name("tlab", "alloc");
 351     _perf_allocation =
 352       PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_Bytes, CHECK);
 353 
 354     cname = PerfDataManager::counter_name("tlab", "gcWaste");
 355     _perf_gc_waste =
 356       PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_Bytes, CHECK);
 357 
 358     cname = PerfDataManager::counter_name("tlab", "maxGcWaste");
 359     _perf_max_gc_waste =
 360       PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_Bytes, CHECK);
 361 
 362     cname = PerfDataManager::counter_name("tlab", "slowWaste");
 363     _perf_slow_refill_waste =
 364       PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_Bytes, CHECK);
 365 
 366     cname = PerfDataManager::counter_name("tlab", "maxSlowWaste");
 367     _perf_max_slow_refill_waste =
 368       PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_Bytes, CHECK);
 369 
 370     cname = PerfDataManager::counter_name("tlab", "fastWaste");
 371     _perf_fast_refill_waste =
 372       PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_Bytes, CHECK);
 373 
 374     cname = PerfDataManager::counter_name("tlab", "maxFastWaste");
 375     _perf_max_fast_refill_waste =
 376       PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_Bytes, CHECK);
 377 
 378     cname = PerfDataManager::counter_name("tlab", "slowAlloc");
 379     _perf_slow_allocations =
 380       PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_None, CHECK);
 381 
 382     cname = PerfDataManager::counter_name("tlab", "maxSlowAlloc");
 383     _perf_max_slow_allocations =
 384       PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_None, CHECK);
 385   }
 386 }
 387 
 388 void GlobalTLABStats::initialize() {
 389   // Clear counters summarizing info from all threads
 390   _allocating_threads      = 0;
 391   _total_refills           = 0;
 392   _max_refills             = 0;
 393   _total_allocation        = 0;
 394   _total_gc_waste          = 0;
 395   _max_gc_waste            = 0;
 396   _total_slow_refill_waste = 0;
 397   _max_slow_refill_waste   = 0;
 398   _total_fast_refill_waste = 0;
 399   _max_fast_refill_waste   = 0;
 400   _total_slow_allocations  = 0;
 401   _max_slow_allocations    = 0;
 402 }
 403 
 404 void GlobalTLABStats::publish() {
 405   _allocating_threads_avg.sample(_allocating_threads);
 406   if (UsePerfData) {
 407     _perf_allocating_threads   ->set_value(_allocating_threads);
 408     _perf_total_refills        ->set_value(_total_refills);
 409     _perf_max_refills          ->set_value(_max_refills);
 410     _perf_allocation           ->set_value(_total_allocation);
 411     _perf_gc_waste             ->set_value(_total_gc_waste);
 412     _perf_max_gc_waste         ->set_value(_max_gc_waste);
 413     _perf_slow_refill_waste    ->set_value(_total_slow_refill_waste);
 414     _perf_max_slow_refill_waste->set_value(_max_slow_refill_waste);
 415     _perf_fast_refill_waste    ->set_value(_total_fast_refill_waste);
 416     _perf_max_fast_refill_waste->set_value(_max_fast_refill_waste);
 417     _perf_slow_allocations     ->set_value(_total_slow_allocations);
 418     _perf_max_slow_allocations ->set_value(_max_slow_allocations);
 419   }
 420 }
 421 
 422 void GlobalTLABStats::print() {
 423   size_t waste = _total_gc_waste + _total_slow_refill_waste + _total_fast_refill_waste;
 424   double waste_percent = _total_allocation == 0 ? 0.0 :
 425                          100.0 * waste / _total_allocation;
 426   gclog_or_tty->print("TLAB totals: thrds: %d  refills: %d max: %d"
 427                       " slow allocs: %d max %d waste: %4.1f%%"
 428                       " gc: " SIZE_FORMAT "B max: " SIZE_FORMAT "B"
 429                       " slow: " SIZE_FORMAT "B max: " SIZE_FORMAT "B"
 430                       " fast: " SIZE_FORMAT "B max: " SIZE_FORMAT "B\n",
 431                       _allocating_threads,
 432                       _total_refills, _max_refills,
 433                       _total_slow_allocations, _max_slow_allocations,
 434                       waste_percent,
 435                       _total_gc_waste * HeapWordSize,
 436                       _max_gc_waste * HeapWordSize,
 437                       _total_slow_refill_waste * HeapWordSize,
 438                       _max_slow_refill_waste * HeapWordSize,
 439                       _total_fast_refill_waste * HeapWordSize,
 440                       _max_fast_refill_waste * HeapWordSize);
 441 }