1 /*
   2  * Copyright (c) 1999, 2011, 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 "utilities/copy.hpp"
  32 #ifdef TARGET_OS_FAMILY_linux
  33 # include "thread_linux.inline.hpp"
  34 #endif
  35 #ifdef TARGET_OS_FAMILY_solaris
  36 # include "thread_solaris.inline.hpp"
  37 #endif
  38 #ifdef TARGET_OS_FAMILY_windows
  39 # include "thread_windows.inline.hpp"
  40 #endif
  41 
  42 // Thread-Local Edens support
  43 
  44 // static member initialization
  45 unsigned         ThreadLocalAllocBuffer::_target_refills = 0;
  46 GlobalTLABStats* ThreadLocalAllocBuffer::_global_stats   = NULL;
  47 
  48 void ThreadLocalAllocBuffer::clear_before_allocation() {
  49   _slow_refill_waste += (unsigned)remaining();
  50   make_parsable(true);   // also retire the TLAB
  51 }
  52 
  53 void ThreadLocalAllocBuffer::accumulate_statistics_before_gc() {
  54   global_stats()->initialize();
  55 
  56   for(JavaThread *thread = Threads::first(); thread; thread = thread->next()) {
  57     thread->tlab().accumulate_statistics();
  58     thread->tlab().initialize_statistics();
  59   }
  60 
  61   // Publish new stats if some allocation occurred.
  62   if (global_stats()->allocation() != 0) {
  63     global_stats()->publish();
  64     if (PrintTLAB) {
  65       global_stats()->print();
  66     }
  67   }
  68 }
  69 
  70 void ThreadLocalAllocBuffer::accumulate_statistics() {
  71   size_t capacity = Universe::heap()->tlab_capacity(myThread()) / HeapWordSize;
  72   size_t unused   = Universe::heap()->unsafe_max_tlab_alloc(myThread()) / HeapWordSize;
  73   size_t used     = capacity - unused;
  74 
  75   // Update allocation history if a reasonable amount of eden was allocated.
  76   bool update_allocation_history = used > 0.5 * capacity;
  77 
  78   _gc_waste += (unsigned)remaining();
  79 
  80   if (PrintTLAB && (_number_of_refills > 0 || Verbose)) {
  81     print_stats("gc");
  82   }
  83 
  84   if (_number_of_refills > 0) {
  85 
  86     if (update_allocation_history) {
  87       // Average the fraction of eden allocated in a tlab by this
  88       // thread for use in the next resize operation.
  89       // _gc_waste is not subtracted because it's included in
  90       // "used".
  91       size_t allocation = _number_of_refills * desired_size();
  92       double alloc_frac = allocation / (double) used;
  93       _allocation_fraction.sample(alloc_frac);
  94     }
  95     global_stats()->update_allocating_threads();
  96     global_stats()->update_number_of_refills(_number_of_refills);
  97     global_stats()->update_allocation(_number_of_refills * desired_size());
  98     global_stats()->update_gc_waste(_gc_waste);
  99     global_stats()->update_slow_refill_waste(_slow_refill_waste);
 100     global_stats()->update_fast_refill_waste(_fast_refill_waste);
 101 
 102   } else {
 103     assert(_number_of_refills == 0 && _fast_refill_waste == 0 &&
 104            _slow_refill_waste == 0 && _gc_waste          == 0,
 105            "tlab stats == 0");
 106   }
 107   global_stats()->update_slow_allocations(_slow_allocations);
 108 }
 109 
 110 // Fills the current tlab with a dummy filler array to create
 111 // an illusion of a contiguous Eden and optionally retires the tlab.
 112 // Waste accounting should be done in caller as appropriate; see,
 113 // for example, clear_before_allocation().
 114 void ThreadLocalAllocBuffer::make_parsable(bool retire) {
 115   if (end() != NULL) {
 116     invariants();
 117 
 118     if (retire) {
 119       myThread()->incr_allocated_bytes(used_bytes());
 120     }
 121 
 122     CollectedHeap::fill_with_object(top(), hard_end(), retire);
 123 
 124     if (retire || ZeroTLAB) {  // "Reset" the TLAB
 125       set_start(NULL);
 126       set_top(NULL);
 127       set_pf_top(NULL);
 128       set_end(NULL);
 129     }
 130   }
 131   assert(!(retire || ZeroTLAB)  ||
 132          (start() == NULL && end() == NULL && top() == NULL),
 133          "TLAB must be reset");
 134 }
 135 
 136 void ThreadLocalAllocBuffer::resize_all_tlabs() {
 137   for(JavaThread *thread = Threads::first(); thread; thread = thread->next()) {
 138     thread->tlab().resize();
 139   }
 140 }
 141 
 142 void ThreadLocalAllocBuffer::resize() {
 143 
 144   if (ResizeTLAB) {
 145     // Compute the next tlab size using expected allocation amount
 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 
 162     set_refill_waste_limit(initial_refill_waste_limit());
 163   }
 164 }
 165 
 166 void ThreadLocalAllocBuffer::initialize_statistics() {
 167     _number_of_refills = 0;
 168     _fast_refill_waste = 0;
 169     _slow_refill_waste = 0;
 170     _gc_waste          = 0;
 171     _slow_allocations  = 0;
 172 }
 173 
 174 void ThreadLocalAllocBuffer::fill(HeapWord* start,
 175                                   HeapWord* top,
 176                                   size_t    new_size) {
 177   _number_of_refills++;
 178   if (PrintTLAB && Verbose) {
 179     print_stats("fill");
 180   }
 181   assert(top <= start + new_size - alignment_reserve(), "size too small");
 182   initialize(start, top, start + new_size - alignment_reserve());
 183 
 184   // Reset amount of internal fragmentation
 185   set_refill_waste_limit(initial_refill_waste_limit());
 186 }
 187 
 188 void ThreadLocalAllocBuffer::initialize(HeapWord* start,
 189                                         HeapWord* top,
 190                                         HeapWord* end) {
 191   set_start(start);
 192   set_top(top);
 193   set_pf_top(top);
 194   set_end(end);
 195   invariants();
 196 }
 197 
 198 void ThreadLocalAllocBuffer::initialize() {
 199   initialize(NULL,                    // start
 200              NULL,                    // top
 201              NULL);                   // end
 202 
 203   set_desired_size(initial_desired_size());
 204 
 205   // Following check is needed because at startup the main (primordial)
 206   // thread is initialized before the heap is.  The initialization for
 207   // this thread is redone in startup_initialization below.
 208   if (Universe::heap() != NULL) {
 209     size_t capacity   = Universe::heap()->tlab_capacity(myThread()) / HeapWordSize;
 210     double alloc_frac = desired_size() * target_refills() / (double) capacity;
 211     _allocation_fraction.sample(alloc_frac);
 212   }
 213 
 214   set_refill_waste_limit(initial_refill_waste_limit());
 215 
 216   initialize_statistics();
 217 }
 218 
 219 void ThreadLocalAllocBuffer::startup_initialization() {
 220 
 221   // Assuming each thread's active tlab is, on average,
 222   // 1/2 full at a GC
 223   _target_refills = 100 / (2 * TLABWasteTargetPercent);
 224   _target_refills = MAX2(_target_refills, (unsigned)1U);
 225 
 226   _global_stats = new GlobalTLABStats();
 227 
 228   // During jvm startup, the main (primordial) thread is initialized
 229   // before the heap is initialized.  So reinitialize it now.
 230   guarantee(Thread::current()->is_Java_thread(), "tlab initialization thread not Java thread");
 231   Thread::current()->tlab().initialize();
 232 
 233   if (PrintTLAB && Verbose) {
 234     gclog_or_tty->print("TLAB min: " SIZE_FORMAT " initial: " SIZE_FORMAT " max: " SIZE_FORMAT "\n",
 235                         min_size(), Thread::current()->tlab().initial_desired_size(), max_size());
 236   }
 237 }
 238 
 239 size_t ThreadLocalAllocBuffer::initial_desired_size() {
 240   size_t init_sz;
 241 
 242   if (TLABSize > 0) {
 243     init_sz = MIN2(TLABSize / HeapWordSize, max_size());
 244   } else if (global_stats() == NULL) {
 245     // Startup issue - main thread initialized before heap initialized.
 246     init_sz = min_size();
 247   } else {
 248     // Initial size is a function of the average number of allocating threads.
 249     unsigned nof_threads = global_stats()->allocating_threads_avg();
 250 
 251     init_sz  = (Universe::heap()->tlab_capacity(myThread()) / HeapWordSize) /
 252                       (nof_threads * target_refills());
 253     init_sz = align_object_size(init_sz);
 254     init_sz = MIN2(MAX2(init_sz, min_size()), max_size());
 255   }
 256   return init_sz;
 257 }
 258 
 259 const size_t ThreadLocalAllocBuffer::max_size() {
 260 
 261   // TLABs can't be bigger than we can fill with a int[Integer.MAX_VALUE].
 262   // This restriction could be removed by enabling filling with multiple arrays.
 263   // If we compute that the reasonable way as
 264   //    header_size + ((sizeof(jint) * max_jint) / HeapWordSize)
 265   // we'll overflow on the multiply, so we do the divide first.
 266   // We actually lose a little by dividing first,
 267   // but that just makes the TLAB  somewhat smaller than the biggest array,
 268   // which is fine, since we'll be able to fill that.
 269 
 270   size_t unaligned_max_size = typeArrayOopDesc::header_size(T_INT) +
 271                               sizeof(jint) *
 272                               ((juint) max_jint / (size_t) HeapWordSize);
 273   return align_size_down(unaligned_max_size, MinObjAlignment);
 274 }
 275 
 276 void ThreadLocalAllocBuffer::print_stats(const char* tag) {
 277   Thread* thrd = myThread();
 278   size_t waste = _gc_waste + _slow_refill_waste + _fast_refill_waste;
 279   size_t alloc = _number_of_refills * _desired_size;
 280   double waste_percent = alloc == 0 ? 0.0 :
 281                       100.0 * waste / alloc;
 282   size_t tlab_used  = Universe::heap()->tlab_capacity(thrd) -
 283                       Universe::heap()->unsafe_max_tlab_alloc(thrd);
 284   gclog_or_tty->print("TLAB: %s thread: " INTPTR_FORMAT " [id: %2d]"
 285                       " desired_size: " SIZE_FORMAT "KB"
 286                       " slow allocs: %d  refill waste: " SIZE_FORMAT "B"
 287                       " alloc:%8.5f %8.0fKB refills: %d waste %4.1f%% gc: %dB"
 288                       " slow: %dB fast: %dB\n",
 289                       tag, thrd, thrd->osthread()->thread_id(),
 290                       _desired_size / (K / HeapWordSize),
 291                       _slow_allocations, _refill_waste_limit * HeapWordSize,
 292                       _allocation_fraction.average(),
 293                       _allocation_fraction.average() * tlab_used / K,
 294                       _number_of_refills, waste_percent,
 295                       _gc_waste * HeapWordSize,
 296                       _slow_refill_waste * HeapWordSize,
 297                       _fast_refill_waste * HeapWordSize);
 298 }
 299 
 300 void ThreadLocalAllocBuffer::verify() {
 301   HeapWord* p = start();
 302   HeapWord* t = top();
 303   HeapWord* prev_p = NULL;
 304   while (p < t) {
 305     oop(p)->verify();
 306     prev_p = p;
 307     p += oop(p)->size();
 308   }
 309   guarantee(p == top(), "end of last object must match end of space");
 310 }
 311 
 312 Thread* ThreadLocalAllocBuffer::myThread() {
 313   return (Thread*)(((char *)this) +
 314                    in_bytes(start_offset()) -
 315                    in_bytes(Thread::tlab_start_offset()));
 316 }
 317 
 318 
 319 GlobalTLABStats::GlobalTLABStats() :
 320   _allocating_threads_avg(TLABAllocationWeight) {
 321 
 322   initialize();
 323 
 324   _allocating_threads_avg.sample(1); // One allocating thread at startup
 325 
 326   if (UsePerfData) {
 327 
 328     EXCEPTION_MARK;
 329     ResourceMark rm;
 330 
 331     char* cname = PerfDataManager::counter_name("tlab", "allocThreads");
 332     _perf_allocating_threads =
 333       PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_None, CHECK);
 334 
 335     cname = PerfDataManager::counter_name("tlab", "fills");
 336     _perf_total_refills =
 337       PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_None, CHECK);
 338 
 339     cname = PerfDataManager::counter_name("tlab", "maxFills");
 340     _perf_max_refills =
 341       PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_None, CHECK);
 342 
 343     cname = PerfDataManager::counter_name("tlab", "alloc");
 344     _perf_allocation =
 345       PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_Bytes, CHECK);
 346 
 347     cname = PerfDataManager::counter_name("tlab", "gcWaste");
 348     _perf_gc_waste =
 349       PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_Bytes, CHECK);
 350 
 351     cname = PerfDataManager::counter_name("tlab", "maxGcWaste");
 352     _perf_max_gc_waste =
 353       PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_Bytes, CHECK);
 354 
 355     cname = PerfDataManager::counter_name("tlab", "slowWaste");
 356     _perf_slow_refill_waste =
 357       PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_Bytes, CHECK);
 358 
 359     cname = PerfDataManager::counter_name("tlab", "maxSlowWaste");
 360     _perf_max_slow_refill_waste =
 361       PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_Bytes, CHECK);
 362 
 363     cname = PerfDataManager::counter_name("tlab", "fastWaste");
 364     _perf_fast_refill_waste =
 365       PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_Bytes, CHECK);
 366 
 367     cname = PerfDataManager::counter_name("tlab", "maxFastWaste");
 368     _perf_max_fast_refill_waste =
 369       PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_Bytes, CHECK);
 370 
 371     cname = PerfDataManager::counter_name("tlab", "slowAlloc");
 372     _perf_slow_allocations =
 373       PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_None, CHECK);
 374 
 375     cname = PerfDataManager::counter_name("tlab", "maxSlowAlloc");
 376     _perf_max_slow_allocations =
 377       PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_None, CHECK);
 378   }
 379 }
 380 
 381 void GlobalTLABStats::initialize() {
 382   // Clear counters summarizing info from all threads
 383   _allocating_threads      = 0;
 384   _total_refills           = 0;
 385   _max_refills             = 0;
 386   _total_allocation        = 0;
 387   _total_gc_waste          = 0;
 388   _max_gc_waste            = 0;
 389   _total_slow_refill_waste = 0;
 390   _max_slow_refill_waste   = 0;
 391   _total_fast_refill_waste = 0;
 392   _max_fast_refill_waste   = 0;
 393   _total_slow_allocations  = 0;
 394   _max_slow_allocations    = 0;
 395 }
 396 
 397 void GlobalTLABStats::publish() {
 398   _allocating_threads_avg.sample(_allocating_threads);
 399   if (UsePerfData) {
 400     _perf_allocating_threads   ->set_value(_allocating_threads);
 401     _perf_total_refills        ->set_value(_total_refills);
 402     _perf_max_refills          ->set_value(_max_refills);
 403     _perf_allocation           ->set_value(_total_allocation);
 404     _perf_gc_waste             ->set_value(_total_gc_waste);
 405     _perf_max_gc_waste         ->set_value(_max_gc_waste);
 406     _perf_slow_refill_waste    ->set_value(_total_slow_refill_waste);
 407     _perf_max_slow_refill_waste->set_value(_max_slow_refill_waste);
 408     _perf_fast_refill_waste    ->set_value(_total_fast_refill_waste);
 409     _perf_max_fast_refill_waste->set_value(_max_fast_refill_waste);
 410     _perf_slow_allocations     ->set_value(_total_slow_allocations);
 411     _perf_max_slow_allocations ->set_value(_max_slow_allocations);
 412   }
 413 }
 414 
 415 void GlobalTLABStats::print() {
 416   size_t waste = _total_gc_waste + _total_slow_refill_waste + _total_fast_refill_waste;
 417   double waste_percent = _total_allocation == 0 ? 0.0 :
 418                          100.0 * waste / _total_allocation;
 419   gclog_or_tty->print("TLAB totals: thrds: %d  refills: %d max: %d"
 420                       " slow allocs: %d max %d waste: %4.1f%%"
 421                       " gc: " SIZE_FORMAT "B max: " SIZE_FORMAT "B"
 422                       " slow: " SIZE_FORMAT "B max: " SIZE_FORMAT "B"
 423                       " fast: " SIZE_FORMAT "B max: " SIZE_FORMAT "B\n",
 424                       _allocating_threads,
 425                       _total_refills, _max_refills,
 426                       _total_slow_allocations, _max_slow_allocations,
 427                       waste_percent,
 428                       _total_gc_waste * HeapWordSize,
 429                       _max_gc_waste * HeapWordSize,
 430                       _total_slow_refill_waste * HeapWordSize,
 431                       _max_slow_refill_waste * HeapWordSize,
 432                       _total_fast_refill_waste * HeapWordSize,
 433                       _max_fast_refill_waste * HeapWordSize);
 434 }