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   // Publish new stats if some allocation occurred.
  57   if (global_stats()->allocation() != 0) {
  58     global_stats()->publish();
  59     if (PrintTLAB) {
  60       global_stats()->print();
  61     }
  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   if (PrintTLAB && (_number_of_refills > 0 || Verbose)) {
  76     print_stats("gc");
  77   }
  78 
  79   if (_number_of_refills > 0) {
  80     // Update allocation history if a reasonable amount of eden was allocated.
  81     bool update_allocation_history = used > 0.5 * capacity;
  82 
  83     if (update_allocation_history) {
  84       // Average the fraction of eden allocated in a tlab by this
  85       // thread for use in the next resize operation.
  86       // _gc_waste is not subtracted because it's included in
  87       // "used".
  88       // The result can be larger than 1.0 due to direct to old allocations.
  89       // These allocations should ideally not be counted but since it is not possible
  90       // to filter them out here we just cap the fraction to be at most 1.0.
  91       double alloc_frac = MIN2(1.0, (double) allocated_since_last_gc / used);
  92       _allocation_fraction.sample(alloc_frac);
  93     }
  94     global_stats()->update_allocating_threads();
  95     global_stats()->update_number_of_refills(_number_of_refills);
  96     global_stats()->update_allocation(_number_of_refills * desired_size());
  97     global_stats()->update_gc_waste(_gc_waste);
  98     global_stats()->update_slow_refill_waste(_slow_refill_waste);
  99     global_stats()->update_fast_refill_waste(_fast_refill_waste);
 100 
 101   } else {
 102     assert(_number_of_refills == 0 && _fast_refill_waste == 0 &&
 103            _slow_refill_waste == 0 && _gc_waste          == 0,
 104            "tlab stats == 0");
 105   }
 106   global_stats()->update_slow_allocations(_slow_allocations);
 107 }
 108 
 109 // Fills the current tlab with a dummy filler array to create
 110 // an illusion of a contiguous Eden and optionally retires the tlab.
 111 // Waste accounting should be done in caller as appropriate; see,
 112 // for example, clear_before_allocation().
 113 void ThreadLocalAllocBuffer::make_parsable(bool retire) {
 114   if (end() != NULL) {
 115     invariants();
 116 
 117     if (retire) {
 118       myThread()->incr_allocated_bytes(used_bytes());
 119     }
 120 
 121     CollectedHeap::fill_with_object(top(), hard_end(), retire);
 122 
 123     if (retire || ZeroTLAB) {  // "Reset" the TLAB
 124       set_start(NULL);
 125       set_top(NULL);
 126       set_pf_top(NULL);
 127       set_end(NULL);
 128     }
 129   }
 130   assert(!(retire || ZeroTLAB)  ||
 131          (start() == NULL && end() == NULL && top() == NULL),
 132          "TLAB must be reset");
 133 }
 134 
 135 void ThreadLocalAllocBuffer::resize_all_tlabs() {
 136   if (ResizeTLAB) {
 137     for (JavaThread *thread = Threads::first(); thread != NULL; thread = thread->next()) {
 138       thread->tlab().resize();
 139     }
 140   }
 141 }
 142 
 143 void ThreadLocalAllocBuffer::resize() {
 144   // Compute the next tlab size using expected allocation amount
 145   assert(ResizeTLAB, "Should not call this otherwise");
 146   size_t alloc = (size_t)(_allocation_fraction.average() *
 147                           (Universe::heap()->tlab_capacity(myThread()) / HeapWordSize));
 148   size_t new_size = alloc / _target_refills;
 149 
 150   new_size = MIN2(MAX2(new_size, min_size()), max_size());
 151 
 152   size_t aligned_new_size = align_object_size(new_size);
 153 
 154   if (PrintTLAB && Verbose) {
 155     gclog_or_tty->print("TLAB new size: thread: " INTPTR_FORMAT " [id: %2d]"
 156                         " refills %d  alloc: %8.6f desired_size: " SIZE_FORMAT " -> " SIZE_FORMAT "\n",
 157                         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   if (PrintTLAB && Verbose) {
 177     print_stats("fill");
 178   }
 179   assert(top <= start + new_size - alignment_reserve(), "size too small");
 180   initialize(start, top, start + new_size - alignment_reserve());
 181 
 182   // Reset amount of internal fragmentation
 183   set_refill_waste_limit(initial_refill_waste_limit());
 184 }
 185 
 186 void ThreadLocalAllocBuffer::initialize(HeapWord* start,
 187                                         HeapWord* top,
 188                                         HeapWord* end) {
 189   set_start(start);
 190   set_top(top);
 191   set_pf_top(top);
 192   set_end(end);
 193   invariants();
 194 }
 195 
 196 void ThreadLocalAllocBuffer::initialize() {
 197   initialize(NULL,                    // start
 198              NULL,                    // top
 199              NULL);                   // end
 200 
 201   set_desired_size(initial_desired_size());
 202 
 203   // Following check is needed because at startup the main (primordial)
 204   // thread is initialized before the heap is.  The initialization for
 205   // this thread is redone in startup_initialization below.
 206   if (Universe::heap() != NULL) {
 207     size_t capacity   = Universe::heap()->tlab_capacity(myThread()) / HeapWordSize;
 208     double alloc_frac = desired_size() * target_refills() / (double) capacity;
 209     _allocation_fraction.sample(alloc_frac);
 210   }
 211 
 212   set_refill_waste_limit(initial_refill_waste_limit());
 213 
 214   initialize_statistics();
 215 }
 216 
 217 void ThreadLocalAllocBuffer::startup_initialization() {
 218 
 219   // Assuming each thread's active tlab is, on average,
 220   // 1/2 full at a GC
 221   _target_refills = 100 / (2 * TLABWasteTargetPercent);
 222   _target_refills = MAX2(_target_refills, (unsigned)1U);
 223 
 224   _global_stats = new GlobalTLABStats();
 225 
 226   // During jvm startup, the main (primordial) thread is initialized
 227   // before the heap is initialized.  So reinitialize it now.
 228   guarantee(Thread::current()->is_Java_thread(), "tlab initialization thread not Java thread");
 229   Thread::current()->tlab().initialize();
 230 
 231   if (PrintTLAB && Verbose) {
 232     gclog_or_tty->print("TLAB min: " SIZE_FORMAT " initial: " SIZE_FORMAT " max: " SIZE_FORMAT "\n",
 233                         min_size(), Thread::current()->tlab().initial_desired_size(), max_size());
 234   }
 235 }
 236 
 237 size_t ThreadLocalAllocBuffer::initial_desired_size() {
 238   size_t init_sz;
 239 
 240   if (TLABSize > 0) {
 241     init_sz = MIN2((size_t)(TLABSize / HeapWordSize), max_size());
 242   } else if (global_stats() == NULL) {
 243     // Startup issue - main thread initialized before heap initialized.
 244     init_sz = min_size();
 245   } else {
 246     // Initial size is a function of the average number of allocating threads.
 247     unsigned nof_threads = global_stats()->allocating_threads_avg();
 248 
 249     init_sz  = (Universe::heap()->tlab_capacity(myThread()) / HeapWordSize) /
 250                       (nof_threads * target_refills());
 251     init_sz = align_object_size(init_sz);
 252     init_sz = MIN2(MAX2(init_sz, min_size()), max_size());
 253   }
 254   return init_sz;
 255 }
 256 
 257 void ThreadLocalAllocBuffer::print_stats(const char* tag) {
 258   Thread* thrd = myThread();
 259   size_t waste = _gc_waste + _slow_refill_waste + _fast_refill_waste;
 260   size_t alloc = _number_of_refills * _desired_size;
 261   double waste_percent = alloc == 0 ? 0.0 :
 262                       100.0 * waste / alloc;
 263   size_t tlab_used  = Universe::heap()->tlab_used(thrd);
 264   gclog_or_tty->print("TLAB: %s thread: " INTPTR_FORMAT " [id: %2d]"
 265                       " desired_size: " SIZE_FORMAT "KB"
 266                       " slow allocs: %d  refill waste: " SIZE_FORMAT "B"
 267                       " alloc:%8.5f %8.0fKB refills: %d waste %4.1f%% gc: %dB"
 268                       " slow: %dB fast: %dB\n",
 269                       tag, thrd, thrd->osthread()->thread_id(),
 270                       _desired_size / (K / HeapWordSize),
 271                       _slow_allocations, _refill_waste_limit * HeapWordSize,
 272                       _allocation_fraction.average(),
 273                       _allocation_fraction.average() * tlab_used / K,
 274                       _number_of_refills, waste_percent,
 275                       _gc_waste * HeapWordSize,
 276                       _slow_refill_waste * HeapWordSize,
 277                       _fast_refill_waste * HeapWordSize);
 278 }
 279 
 280 void ThreadLocalAllocBuffer::verify() {
 281   HeapWord* p = start();
 282   HeapWord* t = top();
 283   HeapWord* prev_p = NULL;
 284   while (p < t) {
 285     oop(p)->verify();
 286     prev_p = p;
 287     p += oop(p)->size();
 288   }
 289   guarantee(p == top(), "end of last object must match end of space");
 290 }
 291 
 292 Thread* ThreadLocalAllocBuffer::myThread() {
 293   return (Thread*)(((char *)this) +
 294                    in_bytes(start_offset()) -
 295                    in_bytes(Thread::tlab_start_offset()));
 296 }
 297 
 298 
 299 GlobalTLABStats::GlobalTLABStats() :
 300   _allocating_threads_avg(TLABAllocationWeight) {
 301 
 302   initialize();
 303 
 304   _allocating_threads_avg.sample(1); // One allocating thread at startup
 305 
 306   if (UsePerfData) {
 307 
 308     EXCEPTION_MARK;
 309     ResourceMark rm;
 310 
 311     char* cname = PerfDataManager::counter_name("tlab", "allocThreads");
 312     _perf_allocating_threads =
 313       PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_None, CHECK);
 314 
 315     cname = PerfDataManager::counter_name("tlab", "fills");
 316     _perf_total_refills =
 317       PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_None, CHECK);
 318 
 319     cname = PerfDataManager::counter_name("tlab", "maxFills");
 320     _perf_max_refills =
 321       PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_None, CHECK);
 322 
 323     cname = PerfDataManager::counter_name("tlab", "alloc");
 324     _perf_allocation =
 325       PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_Bytes, CHECK);
 326 
 327     cname = PerfDataManager::counter_name("tlab", "gcWaste");
 328     _perf_gc_waste =
 329       PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_Bytes, CHECK);
 330 
 331     cname = PerfDataManager::counter_name("tlab", "maxGcWaste");
 332     _perf_max_gc_waste =
 333       PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_Bytes, CHECK);
 334 
 335     cname = PerfDataManager::counter_name("tlab", "slowWaste");
 336     _perf_slow_refill_waste =
 337       PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_Bytes, CHECK);
 338 
 339     cname = PerfDataManager::counter_name("tlab", "maxSlowWaste");
 340     _perf_max_slow_refill_waste =
 341       PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_Bytes, CHECK);
 342 
 343     cname = PerfDataManager::counter_name("tlab", "fastWaste");
 344     _perf_fast_refill_waste =
 345       PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_Bytes, CHECK);
 346 
 347     cname = PerfDataManager::counter_name("tlab", "maxFastWaste");
 348     _perf_max_fast_refill_waste =
 349       PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_Bytes, CHECK);
 350 
 351     cname = PerfDataManager::counter_name("tlab", "slowAlloc");
 352     _perf_slow_allocations =
 353       PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_None, CHECK);
 354 
 355     cname = PerfDataManager::counter_name("tlab", "maxSlowAlloc");
 356     _perf_max_slow_allocations =
 357       PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_None, CHECK);
 358   }
 359 }
 360 
 361 void GlobalTLABStats::initialize() {
 362   // Clear counters summarizing info from all threads
 363   _allocating_threads      = 0;
 364   _total_refills           = 0;
 365   _max_refills             = 0;
 366   _total_allocation        = 0;
 367   _total_gc_waste          = 0;
 368   _max_gc_waste            = 0;
 369   _total_slow_refill_waste = 0;
 370   _max_slow_refill_waste   = 0;
 371   _total_fast_refill_waste = 0;
 372   _max_fast_refill_waste   = 0;
 373   _total_slow_allocations  = 0;
 374   _max_slow_allocations    = 0;
 375 }
 376 
 377 void GlobalTLABStats::publish() {
 378   _allocating_threads_avg.sample(_allocating_threads);
 379   if (UsePerfData) {
 380     _perf_allocating_threads   ->set_value(_allocating_threads);
 381     _perf_total_refills        ->set_value(_total_refills);
 382     _perf_max_refills          ->set_value(_max_refills);
 383     _perf_allocation           ->set_value(_total_allocation);
 384     _perf_gc_waste             ->set_value(_total_gc_waste);
 385     _perf_max_gc_waste         ->set_value(_max_gc_waste);
 386     _perf_slow_refill_waste    ->set_value(_total_slow_refill_waste);
 387     _perf_max_slow_refill_waste->set_value(_max_slow_refill_waste);
 388     _perf_fast_refill_waste    ->set_value(_total_fast_refill_waste);
 389     _perf_max_fast_refill_waste->set_value(_max_fast_refill_waste);
 390     _perf_slow_allocations     ->set_value(_total_slow_allocations);
 391     _perf_max_slow_allocations ->set_value(_max_slow_allocations);
 392   }
 393 }
 394 
 395 void GlobalTLABStats::print() {
 396   size_t waste = _total_gc_waste + _total_slow_refill_waste + _total_fast_refill_waste;
 397   double waste_percent = _total_allocation == 0 ? 0.0 :
 398                          100.0 * waste / _total_allocation;
 399   gclog_or_tty->print("TLAB totals: thrds: %d  refills: %d max: %d"
 400                       " slow allocs: %d max %d waste: %4.1f%%"
 401                       " gc: " SIZE_FORMAT "B max: " SIZE_FORMAT "B"
 402                       " slow: " SIZE_FORMAT "B max: " SIZE_FORMAT "B"
 403                       " fast: " SIZE_FORMAT "B max: " SIZE_FORMAT "B\n",
 404                       _allocating_threads,
 405                       _total_refills, _max_refills,
 406                       _total_slow_allocations, _max_slow_allocations,
 407                       waste_percent,
 408                       _total_gc_waste * HeapWordSize,
 409                       _max_gc_waste * HeapWordSize,
 410                       _total_slow_refill_waste * HeapWordSize,
 411                       _max_slow_refill_waste * HeapWordSize,
 412                       _total_fast_refill_waste * HeapWordSize,
 413                       _max_fast_refill_waste * HeapWordSize);
 414 }