1 /*
   2  * Copyright (c) 1999, 2018, 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/threadLocalAllocBuffer.inline.hpp"
  27 #include "logging/log.hpp"
  28 #include "memory/resourceArea.hpp"
  29 #include "memory/universe.hpp"
  30 #include "oops/oop.inline.hpp"
  31 #include "runtime/thread.inline.hpp"
  32 #include "runtime/threadSMR.hpp"
  33 #include "utilities/copy.hpp"
  34 
  35 // Thread-Local Edens support
  36 
  37 // static member initialization
  38 size_t           ThreadLocalAllocBuffer::_max_size       = 0;
  39 int              ThreadLocalAllocBuffer::_reserve_for_allocation_prefetch = 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 (JavaThreadIteratorWithHandle jtiwh; JavaThread *thread = jtiwh.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     global_stats()->print();
  60   }
  61 }
  62 
  63 void ThreadLocalAllocBuffer::accumulate_statistics() {
  64   Thread* thread = myThread();
  65   size_t capacity = Universe::heap()->tlab_capacity(thread);
  66   size_t used     = Universe::heap()->tlab_used(thread);
  67 
  68   _gc_waste += (unsigned)remaining();
  69   size_t total_allocated = thread->allocated_bytes();
  70   size_t allocated_since_last_gc = total_allocated - _allocated_before_last_gc;
  71   _allocated_before_last_gc = total_allocated;
  72 
  73   print_stats("gc");
  74 
  75   if (_number_of_refills > 0) {
  76     // Update allocation history if a reasonable amount of eden was allocated.
  77     bool update_allocation_history = used > 0.5 * capacity;
  78 
  79     if (update_allocation_history) {
  80       // Average the fraction of eden allocated in a tlab by this
  81       // thread for use in the next resize operation.
  82       // _gc_waste is not subtracted because it's included in
  83       // "used".
  84       // The result can be larger than 1.0 due to direct to old allocations.
  85       // These allocations should ideally not be counted but since it is not possible
  86       // to filter them out here we just cap the fraction to be at most 1.0.
  87       double alloc_frac = MIN2(1.0, (double) allocated_since_last_gc / used);
  88       _allocation_fraction.sample(alloc_frac);
  89     }
  90     global_stats()->update_allocating_threads();
  91     global_stats()->update_number_of_refills(_number_of_refills);
  92     global_stats()->update_allocation(_allocated_size);
  93     global_stats()->update_gc_waste(_gc_waste);
  94     global_stats()->update_slow_refill_waste(_slow_refill_waste);
  95     global_stats()->update_fast_refill_waste(_fast_refill_waste);
  96 
  97   } else {
  98     assert(_number_of_refills == 0 && _fast_refill_waste == 0 &&
  99            _slow_refill_waste == 0 && _gc_waste          == 0,
 100            "tlab stats == 0");
 101   }
 102   global_stats()->update_slow_allocations(_slow_allocations);
 103 }
 104 
 105 // Fills the current tlab with a dummy filler array to create
 106 // an illusion of a contiguous Eden and optionally retires the tlab.
 107 // Waste accounting should be done in caller as appropriate; see,
 108 // for example, clear_before_allocation().
 109 void ThreadLocalAllocBuffer::make_parsable(bool retire, bool zap) {
 110   if (end() != NULL) {
 111     invariants();
 112 
 113     if (retire) {
 114       myThread()->incr_allocated_bytes(used_bytes());
 115     }
 116 
 117     CollectedHeap::fill_with_object(top(), hard_end(), retire && zap);
 118 
 119     if (retire || ZeroTLAB) {  // "Reset" the TLAB
 120       set_start(NULL);
 121       set_top(NULL);
 122       set_pf_top(NULL);
 123       set_end(NULL);
 124     }
 125   }
 126   assert(!(retire || ZeroTLAB)  ||
 127          (start() == NULL && end() == NULL && top() == NULL),
 128          "TLAB must be reset");
 129 }
 130 
 131 void ThreadLocalAllocBuffer::resize_all_tlabs() {
 132   if (ResizeTLAB) {
 133     for (JavaThreadIteratorWithHandle jtiwh; JavaThread *thread = jtiwh.next(); ) {
 134       thread->tlab().resize();
 135     }
 136   }
 137 }
 138 
 139 void ThreadLocalAllocBuffer::resize() {
 140   // Compute the next tlab size using expected allocation amount
 141   assert(ResizeTLAB, "Should not call this otherwise");
 142   size_t alloc = (size_t)(_allocation_fraction.average() *
 143                           (Universe::heap()->tlab_capacity(myThread()) / HeapWordSize));
 144   size_t new_size = alloc / _target_refills;
 145 
 146   new_size = MIN2(MAX2(new_size, min_size()), max_size());
 147 
 148   size_t aligned_new_size = align_object_size(new_size);
 149 
 150   log_trace(gc, tlab)("TLAB new size: thread: " INTPTR_FORMAT " [id: %2d]"
 151                       " refills %d  alloc: %8.6f desired_size: " SIZE_FORMAT " -> " SIZE_FORMAT,
 152                       p2i(myThread()), myThread()->osthread()->thread_id(),
 153                       _target_refills, _allocation_fraction.average(), desired_size(), aligned_new_size);
 154 
 155   set_desired_size(aligned_new_size);
 156   set_refill_waste_limit(initial_refill_waste_limit());
 157 }
 158 
 159 void ThreadLocalAllocBuffer::initialize_statistics() {
 160     _number_of_refills = 0;
 161     _fast_refill_waste = 0;
 162     _slow_refill_waste = 0;
 163     _gc_waste          = 0;
 164     _slow_allocations  = 0;
 165     _allocated_size    = 0;
 166 }
 167 
 168 void ThreadLocalAllocBuffer::fill(HeapWord* start,
 169                                   HeapWord* top,
 170                                   size_t    new_size) {
 171   _number_of_refills++;
 172   _allocated_size += new_size;
 173   print_stats("fill");
 174   assert(top <= start + new_size - alignment_reserve(), "size too small");
 175   initialize(start, top, start + new_size - alignment_reserve());
 176 
 177   // Reset amount of internal fragmentation
 178   set_refill_waste_limit(initial_refill_waste_limit());
 179 }
 180 
 181 void ThreadLocalAllocBuffer::initialize(HeapWord* start,
 182                                         HeapWord* top,
 183                                         HeapWord* end) {
 184   set_start(start);
 185   set_top(top);
 186   set_pf_top(top);
 187   set_end(end);
 188   invariants();
 189 }
 190 
 191 void ThreadLocalAllocBuffer::initialize() {
 192   initialize(NULL,                    // start
 193              NULL,                    // top
 194              NULL);                   // end
 195 
 196   set_desired_size(initial_desired_size());
 197 
 198   // Following check is needed because at startup the main
 199   // thread is initialized before the heap is.  The initialization for
 200   // this thread is redone in startup_initialization below.
 201   if (Universe::heap() != NULL) {
 202     size_t capacity   = Universe::heap()->tlab_capacity(myThread()) / HeapWordSize;
 203     double alloc_frac = desired_size() * target_refills() / (double) capacity;
 204     _allocation_fraction.sample(alloc_frac);
 205   }
 206 
 207   set_refill_waste_limit(initial_refill_waste_limit());
 208 
 209   initialize_statistics();
 210 }
 211 
 212 void ThreadLocalAllocBuffer::startup_initialization() {
 213 
 214   // Assuming each thread's active tlab is, on average,
 215   // 1/2 full at a GC
 216   _target_refills = 100 / (2 * TLABWasteTargetPercent);
 217   _target_refills = MAX2(_target_refills, (unsigned)1U);
 218 
 219   _global_stats = new GlobalTLABStats();
 220 
 221 #ifdef COMPILER2
 222   // If the C2 compiler is present, extra space is needed at the end of
 223   // TLABs, otherwise prefetching instructions generated by the C2
 224   // compiler will fault (due to accessing memory outside of heap).
 225   // The amount of space is the max of the number of lines to
 226   // prefetch for array and for instance allocations. (Extra space must be
 227   // reserved to accommodate both types of allocations.)
 228   //
 229   // Only SPARC-specific BIS instructions are known to fault. (Those
 230   // instructions are generated if AllocatePrefetchStyle==3 and
 231   // AllocatePrefetchInstr==1). To be on the safe side, however,
 232   // extra space is reserved for all combinations of
 233   // AllocatePrefetchStyle and AllocatePrefetchInstr.
 234   //
 235   // If the C2 compiler is not present, no space is reserved.
 236 
 237   // +1 for rounding up to next cache line, +1 to be safe
 238   if (is_server_compilation_mode_vm()) {
 239     int lines =  MAX2(AllocatePrefetchLines, AllocateInstancePrefetchLines) + 2;
 240     _reserve_for_allocation_prefetch = (AllocatePrefetchDistance + AllocatePrefetchStepSize * lines) /
 241                                        (int)HeapWordSize;
 242   }
 243 #endif
 244 
 245   // During jvm startup, the main thread is initialized
 246   // before the heap is initialized.  So reinitialize it now.
 247   guarantee(Thread::current()->is_Java_thread(), "tlab initialization thread not Java thread");
 248   Thread::current()->tlab().initialize();
 249 
 250   log_develop_trace(gc, tlab)("TLAB min: " SIZE_FORMAT " initial: " SIZE_FORMAT " max: " SIZE_FORMAT,
 251                                min_size(), Thread::current()->tlab().initial_desired_size(), max_size());
 252 }
 253 
 254 size_t ThreadLocalAllocBuffer::initial_desired_size() {
 255   size_t init_sz = 0;
 256 
 257   if (TLABSize > 0) {
 258     init_sz = TLABSize / HeapWordSize;
 259   } else if (global_stats() != NULL) {
 260     // Initial size is a function of the average number of allocating threads.
 261     unsigned nof_threads = global_stats()->allocating_threads_avg();
 262 
 263     init_sz  = (Universe::heap()->tlab_capacity(myThread()) / HeapWordSize) /
 264                       (nof_threads * target_refills());
 265     init_sz = align_object_size(init_sz);
 266   }
 267   init_sz = MIN2(MAX2(init_sz, min_size()), max_size());
 268   return init_sz;
 269 }
 270 
 271 void ThreadLocalAllocBuffer::print_stats(const char* tag) {
 272   Log(gc, tlab) log;
 273   if (!log.is_trace()) {
 274     return;
 275   }
 276 
 277   Thread* thrd = myThread();
 278   size_t waste = _gc_waste + _slow_refill_waste + _fast_refill_waste;
 279   double waste_percent = percent_of(waste, _allocated_size);
 280   size_t tlab_used  = Universe::heap()->tlab_used(thrd);
 281   log.trace("TLAB: %s thread: " INTPTR_FORMAT " [id: %2d]"
 282             " desired_size: " SIZE_FORMAT "KB"
 283             " slow allocs: %d  refill waste: " SIZE_FORMAT "B"
 284             " alloc:%8.5f %8.0fKB refills: %d waste %4.1f%% gc: %dB"
 285             " slow: %dB fast: %dB",
 286             tag, p2i(thrd), thrd->osthread()->thread_id(),
 287             _desired_size / (K / HeapWordSize),
 288             _slow_allocations, _refill_waste_limit * HeapWordSize,
 289             _allocation_fraction.average(),
 290             _allocation_fraction.average() * tlab_used / K,
 291             _number_of_refills, waste_percent,
 292             _gc_waste * HeapWordSize,
 293             _slow_refill_waste * HeapWordSize,
 294             _fast_refill_waste * HeapWordSize);
 295 }
 296 
 297 void ThreadLocalAllocBuffer::verify() {
 298   HeapWord* p = start();
 299   HeapWord* t = top();
 300   HeapWord* prev_p = NULL;
 301   while (p < t) {
 302     oop(p)->verify();
 303     prev_p = p;
 304     p += oop(p)->size();
 305   }
 306   guarantee(p == top(), "end of last object must match end of space");
 307 }
 308 
 309 Thread* ThreadLocalAllocBuffer::myThread() {
 310   return (Thread*)(((char *)this) +
 311                    in_bytes(start_offset()) -
 312                    in_bytes(Thread::tlab_start_offset()));
 313 }
 314 
 315 
 316 GlobalTLABStats::GlobalTLABStats() :
 317   _allocating_threads_avg(TLABAllocationWeight) {
 318 
 319   initialize();
 320 
 321   _allocating_threads_avg.sample(1); // One allocating thread at startup
 322 
 323   if (UsePerfData) {
 324 
 325     EXCEPTION_MARK;
 326     ResourceMark rm;
 327 
 328     char* cname = PerfDataManager::counter_name("tlab", "allocThreads");
 329     _perf_allocating_threads =
 330       PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_None, CHECK);
 331 
 332     cname = PerfDataManager::counter_name("tlab", "fills");
 333     _perf_total_refills =
 334       PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_None, CHECK);
 335 
 336     cname = PerfDataManager::counter_name("tlab", "maxFills");
 337     _perf_max_refills =
 338       PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_None, CHECK);
 339 
 340     cname = PerfDataManager::counter_name("tlab", "alloc");
 341     _perf_allocation =
 342       PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_Bytes, CHECK);
 343 
 344     cname = PerfDataManager::counter_name("tlab", "gcWaste");
 345     _perf_gc_waste =
 346       PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_Bytes, CHECK);
 347 
 348     cname = PerfDataManager::counter_name("tlab", "maxGcWaste");
 349     _perf_max_gc_waste =
 350       PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_Bytes, CHECK);
 351 
 352     cname = PerfDataManager::counter_name("tlab", "slowWaste");
 353     _perf_slow_refill_waste =
 354       PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_Bytes, CHECK);
 355 
 356     cname = PerfDataManager::counter_name("tlab", "maxSlowWaste");
 357     _perf_max_slow_refill_waste =
 358       PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_Bytes, CHECK);
 359 
 360     cname = PerfDataManager::counter_name("tlab", "fastWaste");
 361     _perf_fast_refill_waste =
 362       PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_Bytes, CHECK);
 363 
 364     cname = PerfDataManager::counter_name("tlab", "maxFastWaste");
 365     _perf_max_fast_refill_waste =
 366       PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_Bytes, CHECK);
 367 
 368     cname = PerfDataManager::counter_name("tlab", "slowAlloc");
 369     _perf_slow_allocations =
 370       PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_None, CHECK);
 371 
 372     cname = PerfDataManager::counter_name("tlab", "maxSlowAlloc");
 373     _perf_max_slow_allocations =
 374       PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_None, CHECK);
 375   }
 376 }
 377 
 378 void GlobalTLABStats::initialize() {
 379   // Clear counters summarizing info from all threads
 380   _allocating_threads      = 0;
 381   _total_refills           = 0;
 382   _max_refills             = 0;
 383   _total_allocation        = 0;
 384   _total_gc_waste          = 0;
 385   _max_gc_waste            = 0;
 386   _total_slow_refill_waste = 0;
 387   _max_slow_refill_waste   = 0;
 388   _total_fast_refill_waste = 0;
 389   _max_fast_refill_waste   = 0;
 390   _total_slow_allocations  = 0;
 391   _max_slow_allocations    = 0;
 392 }
 393 
 394 void GlobalTLABStats::publish() {
 395   _allocating_threads_avg.sample(_allocating_threads);
 396   if (UsePerfData) {
 397     _perf_allocating_threads   ->set_value(_allocating_threads);
 398     _perf_total_refills        ->set_value(_total_refills);
 399     _perf_max_refills          ->set_value(_max_refills);
 400     _perf_allocation           ->set_value(_total_allocation);
 401     _perf_gc_waste             ->set_value(_total_gc_waste);
 402     _perf_max_gc_waste         ->set_value(_max_gc_waste);
 403     _perf_slow_refill_waste    ->set_value(_total_slow_refill_waste);
 404     _perf_max_slow_refill_waste->set_value(_max_slow_refill_waste);
 405     _perf_fast_refill_waste    ->set_value(_total_fast_refill_waste);
 406     _perf_max_fast_refill_waste->set_value(_max_fast_refill_waste);
 407     _perf_slow_allocations     ->set_value(_total_slow_allocations);
 408     _perf_max_slow_allocations ->set_value(_max_slow_allocations);
 409   }
 410 }
 411 
 412 void GlobalTLABStats::print() {
 413   Log(gc, tlab) log;
 414   if (!log.is_debug()) {
 415     return;
 416   }
 417 
 418   size_t waste = _total_gc_waste + _total_slow_refill_waste + _total_fast_refill_waste;
 419   double waste_percent = percent_of(waste, _total_allocation);
 420   log.debug("TLAB totals: thrds: %d  refills: %d max: %d"
 421             " slow allocs: %d max %d waste: %4.1f%%"
 422             " gc: " SIZE_FORMAT "B max: " SIZE_FORMAT "B"
 423             " slow: " SIZE_FORMAT "B max: " SIZE_FORMAT "B"
 424             " fast: " SIZE_FORMAT "B max: " SIZE_FORMAT "B",
 425             _allocating_threads,
 426             _total_refills, _max_refills,
 427             _total_slow_allocations, _max_slow_allocations,
 428             waste_percent,
 429             _total_gc_waste * HeapWordSize,
 430             _max_gc_waste * HeapWordSize,
 431             _total_slow_refill_waste * HeapWordSize,
 432             _max_slow_refill_waste * HeapWordSize,
 433             _total_fast_refill_waste * HeapWordSize,
 434             _max_fast_refill_waste * HeapWordSize);
 435 }