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/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 void ThreadLocalAllocBuffer::accumulate_statistics_before_gc() {
  50   global_stats()->initialize();
  51 
  52   for (JavaThreadIteratorWithHandle jtiwh; JavaThread *thread = jtiwh.next(); ) {
  53     thread->tlab().accumulate_statistics();
  54     thread->tlab().initialize_statistics();
  55   }
  56 
  57   // Publish new stats if some allocation occurred.
  58   if (global_stats()->allocation() != 0) {
  59     global_stats()->publish();
  60     global_stats()->print();
  61   }
  62 }
  63 
  64 void ThreadLocalAllocBuffer::accumulate_statistics() {
  65   Thread* thread = myThread();
  66   size_t capacity = Universe::heap()->tlab_capacity(thread);
  67   size_t used     = Universe::heap()->tlab_used(thread);
  68 
  69   _gc_waste += (unsigned)remaining();
  70   size_t total_allocated = thread->allocated_bytes();
  71   size_t allocated_since_last_gc = total_allocated - _allocated_before_last_gc;
  72   _allocated_before_last_gc = total_allocated;
  73 
  74   print_stats("gc");
  75 
  76   if (_number_of_refills > 0) {
  77     // Update allocation history if a reasonable amount of eden was allocated.
  78     bool update_allocation_history = used > 0.5 * capacity;
  79 
  80     if (update_allocation_history) {
  81       // Average the fraction of eden allocated in a tlab by this
  82       // thread for use in the next resize operation.
  83       // _gc_waste is not subtracted because it's included in
  84       // "used".
  85       // The result can be larger than 1.0 due to direct to old allocations.
  86       // These allocations should ideally not be counted but since it is not possible
  87       // to filter them out here we just cap the fraction to be at most 1.0.
  88       double alloc_frac = MIN2(1.0, (double) allocated_since_last_gc / used);
  89       _allocation_fraction.sample(alloc_frac);
  90     }
  91     global_stats()->update_allocating_threads();
  92     global_stats()->update_number_of_refills(_number_of_refills);
  93     global_stats()->update_allocation(_number_of_refills * desired_size());
  94     global_stats()->update_gc_waste(_gc_waste);
  95     global_stats()->update_slow_refill_waste(_slow_refill_waste);
  96     global_stats()->update_fast_refill_waste(_fast_refill_waste);
  97 
  98   } else {
  99     assert(_number_of_refills == 0 && _fast_refill_waste == 0 &&
 100            _slow_refill_waste == 0 && _gc_waste          == 0,
 101            "tlab stats == 0");
 102   }
 103   global_stats()->update_slow_allocations(_slow_allocations);
 104 }
 105 
 106 // Fills the current tlab with a dummy filler array to create
 107 // an illusion of a contiguous Eden and optionally retires the tlab.
 108 // Waste accounting should be done in caller as appropriate; see,
 109 // for example, clear_before_allocation().
 110 void ThreadLocalAllocBuffer::make_parsable(bool retire, bool zap) {
 111   if (end() != NULL) {
 112     invariants();
 113 
 114     if (retire) {
 115       myThread()->incr_allocated_bytes(used_bytes());
 116     }
 117 
 118     CollectedHeap::fill_with_object(top(), hard_end(), retire && zap);
 119 
 120     if (retire || ZeroTLAB) {  // "Reset" the TLAB
 121       set_start(NULL);
 122       set_top(NULL);
 123       set_pf_top(NULL);
 124       set_end(NULL);
 125     }
 126   }
 127   assert(!(retire || ZeroTLAB)  ||
 128          (start() == NULL && end() == NULL && top() == NULL),
 129          "TLAB must be reset");
 130 }
 131 
 132 void ThreadLocalAllocBuffer::resize_all_tlabs() {
 133   if (ResizeTLAB) {
 134     for (JavaThreadIteratorWithHandle jtiwh; JavaThread *thread = jtiwh.next(); ) {
 135       thread->tlab().resize();
 136     }
 137   }
 138 }
 139 
 140 void ThreadLocalAllocBuffer::resize() {
 141   // Compute the next tlab size using expected allocation amount
 142   assert(ResizeTLAB, "Should not call this otherwise");
 143   size_t alloc = (size_t)(_allocation_fraction.average() *
 144                           (Universe::heap()->tlab_capacity(myThread()) / HeapWordSize));
 145   size_t new_size = alloc / _target_refills;
 146 
 147   new_size = MIN2(MAX2(new_size, min_size()), max_size());
 148 
 149   size_t aligned_new_size = align_object_size(new_size);
 150 
 151   log_trace(gc, tlab)("TLAB new size: thread: " INTPTR_FORMAT " [id: %2d]"
 152                       " refills %d  alloc: %8.6f desired_size: " SIZE_FORMAT " -> " SIZE_FORMAT,
 153                       p2i(myThread()), myThread()->osthread()->thread_id(),
 154                       _target_refills, _allocation_fraction.average(), desired_size(), aligned_new_size);
 155 
 156   set_desired_size(aligned_new_size);
 157   set_refill_waste_limit(initial_refill_waste_limit());
 158 }
 159 
 160 void ThreadLocalAllocBuffer::initialize_statistics() {
 161     _number_of_refills = 0;
 162     _fast_refill_waste = 0;
 163     _slow_refill_waste = 0;
 164     _gc_waste          = 0;
 165     _slow_allocations  = 0;
 166 }
 167 
 168 void ThreadLocalAllocBuffer::fill(HeapWord* start,
 169                                   HeapWord* top,
 170                                   size_t    new_size) {
 171   _number_of_refills++;
 172   print_stats("fill");
 173   assert(top <= start + new_size - alignment_reserve(), "size too small");
 174   initialize(start, top, start + new_size - alignment_reserve());
 175 
 176   // Reset amount of internal fragmentation
 177   set_refill_waste_limit(initial_refill_waste_limit());
 178 }
 179 
 180 void ThreadLocalAllocBuffer::initialize(HeapWord* start,
 181                                         HeapWord* top,
 182                                         HeapWord* end) {
 183   set_start(start);
 184   set_top(top);
 185   set_pf_top(top);
 186   set_end(end);
 187   invariants();
 188 }
 189 
 190 void ThreadLocalAllocBuffer::initialize() {
 191   initialize(NULL,                    // start
 192              NULL,                    // top
 193              NULL);                   // end
 194 
 195   set_desired_size(initial_desired_size());
 196 
 197   // Following check is needed because at startup the main (primordial)
 198   // thread is initialized before the heap is.  The initialization for
 199   // this thread is redone in startup_initialization below.
 200   if (Universe::heap() != NULL) {
 201     size_t capacity   = Universe::heap()->tlab_capacity(myThread()) / HeapWordSize;
 202     double alloc_frac = desired_size() * target_refills() / (double) capacity;
 203     _allocation_fraction.sample(alloc_frac);
 204   }
 205 
 206   set_refill_waste_limit(initial_refill_waste_limit());
 207 
 208   initialize_statistics();
 209 }
 210 
 211 void ThreadLocalAllocBuffer::startup_initialization() {
 212 
 213   // Assuming each thread's active tlab is, on average,
 214   // 1/2 full at a GC
 215   _target_refills = 100 / (2 * TLABWasteTargetPercent);
 216   _target_refills = MAX2(_target_refills, (unsigned)1U);
 217 
 218   _global_stats = new GlobalTLABStats();
 219 
 220 #ifdef COMPILER2
 221   // If the C2 compiler is present, extra space is needed at the end of
 222   // TLABs, otherwise prefetching instructions generated by the C2
 223   // compiler will fault (due to accessing memory outside of heap).
 224   // The amount of space is the max of the number of lines to
 225   // prefetch for array and for instance allocations. (Extra space must be
 226   // reserved to accommodate both types of allocations.)
 227   //
 228   // Only SPARC-specific BIS instructions are known to fault. (Those
 229   // instructions are generated if AllocatePrefetchStyle==3 and
 230   // AllocatePrefetchInstr==1). To be on the safe side, however,
 231   // extra space is reserved for all combinations of
 232   // AllocatePrefetchStyle and AllocatePrefetchInstr.
 233   //
 234   // If the C2 compiler is not present, no space is reserved.
 235 
 236   // +1 for rounding up to next cache line, +1 to be safe
 237   if (is_server_compilation_mode_vm()) {
 238     int lines =  MAX2(AllocatePrefetchLines, AllocateInstancePrefetchLines) + 2;
 239     _reserve_for_allocation_prefetch = (AllocatePrefetchDistance + AllocatePrefetchStepSize * lines) /
 240                                        (int)HeapWordSize;
 241   }
 242 #endif
 243 
 244   // During jvm startup, the main (primordial) thread is initialized
 245   // before the heap is initialized.  So reinitialize it now.
 246   guarantee(Thread::current()->is_Java_thread(), "tlab initialization thread not Java thread");
 247   Thread::current()->tlab().initialize();
 248 
 249   log_develop_trace(gc, tlab)("TLAB min: " SIZE_FORMAT " initial: " SIZE_FORMAT " max: " SIZE_FORMAT,
 250                                min_size(), Thread::current()->tlab().initial_desired_size(), max_size());
 251 }
 252 
 253 size_t ThreadLocalAllocBuffer::initial_desired_size() {
 254   size_t init_sz = 0;
 255 
 256   if (TLABSize > 0) {
 257     init_sz = TLABSize / HeapWordSize;
 258   } else if (global_stats() != NULL) {
 259     // Initial size is a function of the average number of allocating threads.
 260     unsigned nof_threads = global_stats()->allocating_threads_avg();
 261 
 262     init_sz  = (Universe::heap()->tlab_capacity(myThread()) / HeapWordSize) /
 263                       (nof_threads * target_refills());
 264     init_sz = align_object_size(init_sz);
 265   }
 266   init_sz = MIN2(MAX2(init_sz, min_size()), max_size());
 267   return init_sz;
 268 }
 269 
 270 void ThreadLocalAllocBuffer::print_stats(const char* tag) {
 271   Log(gc, tlab) log;
 272   if (!log.is_trace()) {
 273     return;
 274   }
 275 
 276   Thread* thrd = myThread();
 277   size_t waste = _gc_waste + _slow_refill_waste + _fast_refill_waste;
 278   size_t alloc = _number_of_refills * _desired_size;
 279   double waste_percent = alloc == 0 ? 0.0 :
 280                       100.0 * waste / alloc;
 281   size_t tlab_used  = Universe::heap()->tlab_used(thrd);
 282   log.trace("TLAB: %s thread: " INTPTR_FORMAT " [id: %2d]"
 283             " desired_size: " SIZE_FORMAT "KB"
 284             " slow allocs: %d  refill waste: " SIZE_FORMAT "B"
 285             " alloc:%8.5f %8.0fKB refills: %d waste %4.1f%% gc: %dB"
 286             " slow: %dB fast: %dB",
 287             tag, p2i(thrd), thrd->osthread()->thread_id(),
 288             _desired_size / (K / HeapWordSize),
 289             _slow_allocations, _refill_waste_limit * HeapWordSize,
 290             _allocation_fraction.average(),
 291             _allocation_fraction.average() * tlab_used / K,
 292             _number_of_refills, waste_percent,
 293             _gc_waste * HeapWordSize,
 294             _slow_refill_waste * HeapWordSize,
 295             _fast_refill_waste * HeapWordSize);
 296 }
 297 
 298 void ThreadLocalAllocBuffer::verify() {
 299   HeapWord* p = start();
 300   HeapWord* t = top();
 301   HeapWord* prev_p = NULL;
 302   while (p < t) {
 303     oop(p)->verify();
 304     prev_p = p;
 305     p += oop(p)->size();
 306   }
 307   guarantee(p == top(), "end of last object must match end of space");
 308 }
 309 
 310 Thread* ThreadLocalAllocBuffer::myThread() {
 311   return (Thread*)(((char *)this) +
 312                    in_bytes(start_offset()) -
 313                    in_bytes(Thread::tlab_start_offset()));
 314 }
 315 
 316 
 317 GlobalTLABStats::GlobalTLABStats() :
 318   _allocating_threads_avg(TLABAllocationWeight) {
 319 
 320   initialize();
 321 
 322   _allocating_threads_avg.sample(1); // One allocating thread at startup
 323 
 324   if (UsePerfData) {
 325 
 326     EXCEPTION_MARK;
 327     ResourceMark rm;
 328 
 329     char* cname = PerfDataManager::counter_name("tlab", "allocThreads");
 330     _perf_allocating_threads =
 331       PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_None, CHECK);
 332 
 333     cname = PerfDataManager::counter_name("tlab", "fills");
 334     _perf_total_refills =
 335       PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_None, CHECK);
 336 
 337     cname = PerfDataManager::counter_name("tlab", "maxFills");
 338     _perf_max_refills =
 339       PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_None, CHECK);
 340 
 341     cname = PerfDataManager::counter_name("tlab", "alloc");
 342     _perf_allocation =
 343       PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_Bytes, CHECK);
 344 
 345     cname = PerfDataManager::counter_name("tlab", "gcWaste");
 346     _perf_gc_waste =
 347       PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_Bytes, CHECK);
 348 
 349     cname = PerfDataManager::counter_name("tlab", "maxGcWaste");
 350     _perf_max_gc_waste =
 351       PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_Bytes, CHECK);
 352 
 353     cname = PerfDataManager::counter_name("tlab", "slowWaste");
 354     _perf_slow_refill_waste =
 355       PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_Bytes, CHECK);
 356 
 357     cname = PerfDataManager::counter_name("tlab", "maxSlowWaste");
 358     _perf_max_slow_refill_waste =
 359       PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_Bytes, CHECK);
 360 
 361     cname = PerfDataManager::counter_name("tlab", "fastWaste");
 362     _perf_fast_refill_waste =
 363       PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_Bytes, CHECK);
 364 
 365     cname = PerfDataManager::counter_name("tlab", "maxFastWaste");
 366     _perf_max_fast_refill_waste =
 367       PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_Bytes, CHECK);
 368 
 369     cname = PerfDataManager::counter_name("tlab", "slowAlloc");
 370     _perf_slow_allocations =
 371       PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_None, CHECK);
 372 
 373     cname = PerfDataManager::counter_name("tlab", "maxSlowAlloc");
 374     _perf_max_slow_allocations =
 375       PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_None, CHECK);
 376   }
 377 }
 378 
 379 void GlobalTLABStats::initialize() {
 380   // Clear counters summarizing info from all threads
 381   _allocating_threads      = 0;
 382   _total_refills           = 0;
 383   _max_refills             = 0;
 384   _total_allocation        = 0;
 385   _total_gc_waste          = 0;
 386   _max_gc_waste            = 0;
 387   _total_slow_refill_waste = 0;
 388   _max_slow_refill_waste   = 0;
 389   _total_fast_refill_waste = 0;
 390   _max_fast_refill_waste   = 0;
 391   _total_slow_allocations  = 0;
 392   _max_slow_allocations    = 0;
 393 }
 394 
 395 void GlobalTLABStats::publish() {
 396   _allocating_threads_avg.sample(_allocating_threads);
 397   if (UsePerfData) {
 398     _perf_allocating_threads   ->set_value(_allocating_threads);
 399     _perf_total_refills        ->set_value(_total_refills);
 400     _perf_max_refills          ->set_value(_max_refills);
 401     _perf_allocation           ->set_value(_total_allocation);
 402     _perf_gc_waste             ->set_value(_total_gc_waste);
 403     _perf_max_gc_waste         ->set_value(_max_gc_waste);
 404     _perf_slow_refill_waste    ->set_value(_total_slow_refill_waste);
 405     _perf_max_slow_refill_waste->set_value(_max_slow_refill_waste);
 406     _perf_fast_refill_waste    ->set_value(_total_fast_refill_waste);
 407     _perf_max_fast_refill_waste->set_value(_max_fast_refill_waste);
 408     _perf_slow_allocations     ->set_value(_total_slow_allocations);
 409     _perf_max_slow_allocations ->set_value(_max_slow_allocations);
 410   }
 411 }
 412 
 413 void GlobalTLABStats::print() {
 414   Log(gc, tlab) log;
 415   if (!log.is_debug()) {
 416     return;
 417   }
 418 
 419   size_t waste = _total_gc_waste + _total_slow_refill_waste + _total_fast_refill_waste;
 420   double waste_percent = _total_allocation == 0 ? 0.0 :
 421                          100.0 * waste / _total_allocation;
 422   log.debug("TLAB totals: thrds: %d  refills: %d max: %d"
 423             " slow allocs: %d max %d waste: %4.1f%%"
 424             " gc: " SIZE_FORMAT "B max: " SIZE_FORMAT "B"
 425             " slow: " SIZE_FORMAT "B max: " SIZE_FORMAT "B"
 426             " fast: " SIZE_FORMAT "B max: " SIZE_FORMAT "B",
 427             _allocating_threads,
 428             _total_refills, _max_refills,
 429             _total_slow_allocations, _max_slow_allocations,
 430             waste_percent,
 431             _total_gc_waste * HeapWordSize,
 432             _max_gc_waste * HeapWordSize,
 433             _total_slow_refill_waste * HeapWordSize,
 434             _max_slow_refill_waste * HeapWordSize,
 435             _total_fast_refill_waste * HeapWordSize,
 436             _max_fast_refill_waste * HeapWordSize);
 437 }