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 size_t ThreadLocalAllocBuffer::remaining() {
  49   if (end() == NULL) {
  50     return 0;
  51   }
  52 
  53   return pointer_delta(hard_end(), top());
  54 }
  55 
  56 void ThreadLocalAllocBuffer::accumulate_statistics_before_gc() {
  57   global_stats()->initialize();
  58 
  59   for (JavaThreadIteratorWithHandle jtiwh; JavaThread *thread = jtiwh.next(); ) {
  60     thread->tlab().accumulate_statistics();
  61     thread->tlab().initialize_statistics();
  62   }
  63 
  64   // Publish new stats if some allocation occurred.
  65   if (global_stats()->allocation() != 0) {
  66     global_stats()->publish();
  67     global_stats()->print();
  68   }
  69 }
  70 
  71 void ThreadLocalAllocBuffer::accumulate_statistics() {
  72   Thread* thread = myThread();
  73   size_t capacity = Universe::heap()->tlab_capacity(thread);
  74   size_t used     = Universe::heap()->tlab_used(thread);
  75 
  76   _gc_waste += (unsigned)remaining();
  77   size_t total_allocated = thread->allocated_bytes();
  78   size_t allocated_since_last_gc = total_allocated - _allocated_before_last_gc;
  79   _allocated_before_last_gc = total_allocated;
  80 
  81   print_stats("gc");
  82 
  83   if (_number_of_refills > 0) {
  84     // Update allocation history if a reasonable amount of eden was allocated.
  85     bool update_allocation_history = used > 0.5 * capacity;
  86 
  87     if (update_allocation_history) {
  88       // Average the fraction of eden allocated in a tlab by this
  89       // thread for use in the next resize operation.
  90       // _gc_waste is not subtracted because it's included in
  91       // "used".
  92       // The result can be larger than 1.0 due to direct to old allocations.
  93       // These allocations should ideally not be counted but since it is not possible
  94       // to filter them out here we just cap the fraction to be at most 1.0.
  95       double alloc_frac = MIN2(1.0, (double) allocated_since_last_gc / used);
  96       _allocation_fraction.sample(alloc_frac);
  97     }
  98     global_stats()->update_allocating_threads();
  99     global_stats()->update_number_of_refills(_number_of_refills);
 100     global_stats()->update_allocation(_allocated_size);
 101     global_stats()->update_gc_waste(_gc_waste);
 102     global_stats()->update_slow_refill_waste(_slow_refill_waste);
 103     global_stats()->update_fast_refill_waste(_fast_refill_waste);
 104 
 105   } else {
 106     assert(_number_of_refills == 0 && _fast_refill_waste == 0 &&
 107            _slow_refill_waste == 0 && _gc_waste          == 0,
 108            "tlab stats == 0");
 109   }
 110   global_stats()->update_slow_allocations(_slow_allocations);
 111 }
 112 
 113 // Fills the current tlab with a dummy filler array to create
 114 // an illusion of a contiguous Eden and optionally retires the tlab.
 115 // Waste accounting should be done in caller as appropriate; see,
 116 // for example, clear_before_allocation().
 117 void ThreadLocalAllocBuffer::make_parsable(bool retire, bool zap) {
 118   if (end() != NULL) {
 119     invariants();
 120 
 121     if (retire) {
 122       myThread()->incr_allocated_bytes(used_bytes());
 123     }
 124 
 125     CollectedHeap::fill_with_object(top(), hard_end(), retire && zap);
 126 
 127     if (retire || ZeroTLAB) {  // "Reset" the TLAB
 128       set_start(NULL);
 129       set_top(NULL);
 130       set_pf_top(NULL);
 131       set_end(NULL);
 132       set_allocation_end(NULL);
 133     }
 134   }
 135   assert(!(retire || ZeroTLAB)  ||
 136          (start() == NULL && end() == NULL && top() == NULL &&
 137           _allocation_end == NULL),
 138          "TLAB must be reset");
 139 }
 140 
 141 void ThreadLocalAllocBuffer::resize_all_tlabs() {
 142   if (ResizeTLAB) {
 143     for (JavaThreadIteratorWithHandle jtiwh; JavaThread *thread = jtiwh.next(); ) {
 144       thread->tlab().resize();
 145     }
 146   }
 147 }
 148 
 149 void ThreadLocalAllocBuffer::resize() {
 150   // Compute the next tlab size using expected allocation amount
 151   assert(ResizeTLAB, "Should not call this otherwise");
 152   size_t alloc = (size_t)(_allocation_fraction.average() *
 153                           (Universe::heap()->tlab_capacity(myThread()) / HeapWordSize));
 154   size_t new_size = alloc / _target_refills;
 155 
 156   new_size = MIN2(MAX2(new_size, min_size()), max_size());
 157 
 158   size_t aligned_new_size = align_object_size(new_size);
 159 
 160   log_trace(gc, tlab)("TLAB new size: thread: " INTPTR_FORMAT " [id: %2d]"
 161                       " refills %d  alloc: %8.6f desired_size: " SIZE_FORMAT " -> " SIZE_FORMAT,
 162                       p2i(myThread()), myThread()->osthread()->thread_id(),
 163                       _target_refills, _allocation_fraction.average(), desired_size(), aligned_new_size);
 164 
 165   set_desired_size(aligned_new_size);
 166   set_refill_waste_limit(initial_refill_waste_limit());
 167 }
 168 
 169 void ThreadLocalAllocBuffer::initialize_statistics() {
 170   _number_of_refills = 0;
 171   _fast_refill_waste = 0;
 172   _slow_refill_waste = 0;
 173   _gc_waste          = 0;
 174   _slow_allocations  = 0;
 175   _allocated_size    = 0;
 176 }
 177 
 178 void ThreadLocalAllocBuffer::fill(HeapWord* start,
 179                                   HeapWord* top,
 180                                   size_t    new_size) {
 181   _number_of_refills++;
 182   _allocated_size += new_size;
 183   print_stats("fill");
 184   assert(top <= start + new_size - alignment_reserve(), "size too small");
 185 
 186   initialize(start, top, start + new_size - alignment_reserve());
 187 
 188   if (ThreadHeapSampler::enabled()) {
 189     set_sample_end();
 190   }
 191 
 192   // Reset amount of internal fragmentation
 193   set_refill_waste_limit(initial_refill_waste_limit());
 194 }
 195 
 196 void ThreadLocalAllocBuffer::initialize(HeapWord* start,
 197                                         HeapWord* top,
 198                                         HeapWord* end) {
 199   set_start(start);
 200   set_top(top);
 201   set_pf_top(top);
 202   set_end(end);
 203   set_allocation_end(end);
 204   invariants();
 205 }
 206 
 207 void ThreadLocalAllocBuffer::initialize() {
 208   initialize(NULL,                    // start
 209              NULL,                    // top
 210              NULL);                   // end
 211 
 212   set_desired_size(initial_desired_size());
 213 
 214   // Following check is needed because at startup the main
 215   // thread is initialized before the heap is.  The initialization for
 216   // this thread is redone in startup_initialization below.
 217   if (Universe::heap() != NULL) {
 218     size_t capacity   = Universe::heap()->tlab_capacity(myThread()) / HeapWordSize;
 219     double alloc_frac = desired_size() * target_refills() / (double) capacity;
 220     _allocation_fraction.sample(alloc_frac);
 221   }
 222 
 223   set_refill_waste_limit(initial_refill_waste_limit());
 224 
 225   initialize_statistics();
 226 }
 227 
 228 void ThreadLocalAllocBuffer::startup_initialization() {
 229 
 230   // Assuming each thread's active tlab is, on average,
 231   // 1/2 full at a GC
 232   _target_refills = 100 / (2 * TLABWasteTargetPercent);
 233   _target_refills = MAX2(_target_refills, (unsigned)1U);
 234 
 235   _global_stats = new GlobalTLABStats();
 236 
 237 #ifdef COMPILER2
 238   // If the C2 compiler is present, extra space is needed at the end of
 239   // TLABs, otherwise prefetching instructions generated by the C2
 240   // compiler will fault (due to accessing memory outside of heap).
 241   // The amount of space is the max of the number of lines to
 242   // prefetch for array and for instance allocations. (Extra space must be
 243   // reserved to accommodate both types of allocations.)
 244   //
 245   // Only SPARC-specific BIS instructions are known to fault. (Those
 246   // instructions are generated if AllocatePrefetchStyle==3 and
 247   // AllocatePrefetchInstr==1). To be on the safe side, however,
 248   // extra space is reserved for all combinations of
 249   // AllocatePrefetchStyle and AllocatePrefetchInstr.
 250   //
 251   // If the C2 compiler is not present, no space is reserved.
 252 
 253   // +1 for rounding up to next cache line, +1 to be safe
 254   if (is_server_compilation_mode_vm()) {
 255     int lines =  MAX2(AllocatePrefetchLines, AllocateInstancePrefetchLines) + 2;
 256     _reserve_for_allocation_prefetch = (AllocatePrefetchDistance + AllocatePrefetchStepSize * lines) /
 257                                        (int)HeapWordSize;
 258   }
 259 #endif
 260 
 261   // During jvm startup, the main thread is initialized
 262   // before the heap is initialized.  So reinitialize it now.
 263   guarantee(Thread::current()->is_Java_thread(), "tlab initialization thread not Java thread");
 264   Thread::current()->tlab().initialize();
 265 
 266   log_develop_trace(gc, tlab)("TLAB min: " SIZE_FORMAT " initial: " SIZE_FORMAT " max: " SIZE_FORMAT,
 267                                min_size(), Thread::current()->tlab().initial_desired_size(), max_size());
 268 }
 269 
 270 size_t ThreadLocalAllocBuffer::initial_desired_size() {
 271   size_t init_sz = 0;
 272 
 273   if (TLABSize > 0) {
 274     init_sz = TLABSize / HeapWordSize;
 275   } else if (global_stats() != NULL) {
 276     // Initial size is a function of the average number of allocating threads.
 277     unsigned nof_threads = global_stats()->allocating_threads_avg();
 278 
 279     init_sz  = (Universe::heap()->tlab_capacity(myThread()) / HeapWordSize) /
 280                       (nof_threads * target_refills());
 281     init_sz = align_object_size(init_sz);
 282   }
 283   init_sz = MIN2(MAX2(init_sz, min_size()), max_size());
 284   return init_sz;
 285 }
 286 
 287 void ThreadLocalAllocBuffer::print_stats(const char* tag) {
 288   Log(gc, tlab) log;
 289   if (!log.is_trace()) {
 290     return;
 291   }
 292 
 293   Thread* thrd = myThread();
 294   size_t waste = _gc_waste + _slow_refill_waste + _fast_refill_waste;
 295   double waste_percent = percent_of(waste, _allocated_size);
 296   size_t tlab_used  = Universe::heap()->tlab_used(thrd);
 297   log.trace("TLAB: %s thread: " INTPTR_FORMAT " [id: %2d]"
 298             " desired_size: " SIZE_FORMAT "KB"
 299             " slow allocs: %d  refill waste: " SIZE_FORMAT "B"
 300             " alloc:%8.5f %8.0fKB refills: %d waste %4.1f%% gc: %dB"
 301             " slow: %dB fast: %dB",
 302             tag, p2i(thrd), thrd->osthread()->thread_id(),
 303             _desired_size / (K / HeapWordSize),
 304             _slow_allocations, _refill_waste_limit * HeapWordSize,
 305             _allocation_fraction.average(),
 306             _allocation_fraction.average() * tlab_used / K,
 307             _number_of_refills, waste_percent,
 308             _gc_waste * HeapWordSize,
 309             _slow_refill_waste * HeapWordSize,
 310             _fast_refill_waste * HeapWordSize);
 311 }
 312 
 313 void ThreadLocalAllocBuffer::verify() {
 314   HeapWord* p = start();
 315   HeapWord* t = top();
 316   HeapWord* prev_p = NULL;
 317   while (p < t) {
 318     oop(p)->verify();
 319     prev_p = p;
 320     p += oop(p)->size();
 321   }
 322   guarantee(p == top(), "end of last object must match end of space");
 323 }
 324 
 325 void ThreadLocalAllocBuffer::set_sample_end() {
 326   size_t heap_words_remaining = pointer_delta(_end, _top);
 327   size_t bytes_until_sample = myThread()->heap_sampler().bytes_until_sample();
 328   size_t words_until_sample = bytes_until_sample / HeapWordSize;;
 329 
 330   if (heap_words_remaining > words_until_sample) {
 331     HeapWord* new_end = _top + words_until_sample;
 332     set_end(new_end);
 333     _bytes_since_last_sample_point = bytes_until_sample;
 334   } else {
 335     _bytes_since_last_sample_point = heap_words_remaining * HeapWordSize;;
 336   }
 337 }
 338 
 339 Thread* ThreadLocalAllocBuffer::myThread() {
 340   return (Thread*)(((char *)this) +
 341                    in_bytes(start_offset()) -
 342                    in_bytes(Thread::tlab_start_offset()));
 343 }
 344 
 345 void ThreadLocalAllocBuffer::set_back_allocation_end() {
 346   _end = _allocation_end;
 347 }
 348 
 349 HeapWord* ThreadLocalAllocBuffer::allocate_sampled_object(size_t size) {
 350   set_back_allocation_end();
 351   HeapWord* result = allocate(size);
 352 
 353   if (result) {
 354     myThread()->heap_sampler().check_for_sampling(result, size * HeapWordSize, _bytes_since_last_sample_point);
 355     set_sample_end();
 356   }
 357 
 358   return result;
 359 }
 360 
 361 HeapWord* ThreadLocalAllocBuffer::hard_end() {
 362   return _allocation_end + alignment_reserve();
 363 }
 364 
 365 GlobalTLABStats::GlobalTLABStats() :
 366   _allocating_threads_avg(TLABAllocationWeight) {
 367 
 368   initialize();
 369 
 370   _allocating_threads_avg.sample(1); // One allocating thread at startup
 371 
 372   if (UsePerfData) {
 373 
 374     EXCEPTION_MARK;
 375     ResourceMark rm;
 376 
 377     char* cname = PerfDataManager::counter_name("tlab", "allocThreads");
 378     _perf_allocating_threads =
 379       PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_None, CHECK);
 380 
 381     cname = PerfDataManager::counter_name("tlab", "fills");
 382     _perf_total_refills =
 383       PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_None, CHECK);
 384 
 385     cname = PerfDataManager::counter_name("tlab", "maxFills");
 386     _perf_max_refills =
 387       PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_None, CHECK);
 388 
 389     cname = PerfDataManager::counter_name("tlab", "alloc");
 390     _perf_allocation =
 391       PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_Bytes, CHECK);
 392 
 393     cname = PerfDataManager::counter_name("tlab", "gcWaste");
 394     _perf_gc_waste =
 395       PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_Bytes, CHECK);
 396 
 397     cname = PerfDataManager::counter_name("tlab", "maxGcWaste");
 398     _perf_max_gc_waste =
 399       PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_Bytes, CHECK);
 400 
 401     cname = PerfDataManager::counter_name("tlab", "slowWaste");
 402     _perf_slow_refill_waste =
 403       PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_Bytes, CHECK);
 404 
 405     cname = PerfDataManager::counter_name("tlab", "maxSlowWaste");
 406     _perf_max_slow_refill_waste =
 407       PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_Bytes, CHECK);
 408 
 409     cname = PerfDataManager::counter_name("tlab", "fastWaste");
 410     _perf_fast_refill_waste =
 411       PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_Bytes, CHECK);
 412 
 413     cname = PerfDataManager::counter_name("tlab", "maxFastWaste");
 414     _perf_max_fast_refill_waste =
 415       PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_Bytes, CHECK);
 416 
 417     cname = PerfDataManager::counter_name("tlab", "slowAlloc");
 418     _perf_slow_allocations =
 419       PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_None, CHECK);
 420 
 421     cname = PerfDataManager::counter_name("tlab", "maxSlowAlloc");
 422     _perf_max_slow_allocations =
 423       PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_None, CHECK);
 424   }
 425 }
 426 
 427 void GlobalTLABStats::initialize() {
 428   // Clear counters summarizing info from all threads
 429   _allocating_threads      = 0;
 430   _total_refills           = 0;
 431   _max_refills             = 0;
 432   _total_allocation        = 0;
 433   _total_gc_waste          = 0;
 434   _max_gc_waste            = 0;
 435   _total_slow_refill_waste = 0;
 436   _max_slow_refill_waste   = 0;
 437   _total_fast_refill_waste = 0;
 438   _max_fast_refill_waste   = 0;
 439   _total_slow_allocations  = 0;
 440   _max_slow_allocations    = 0;
 441 }
 442 
 443 void GlobalTLABStats::publish() {
 444   _allocating_threads_avg.sample(_allocating_threads);
 445   if (UsePerfData) {
 446     _perf_allocating_threads   ->set_value(_allocating_threads);
 447     _perf_total_refills        ->set_value(_total_refills);
 448     _perf_max_refills          ->set_value(_max_refills);
 449     _perf_allocation           ->set_value(_total_allocation);
 450     _perf_gc_waste             ->set_value(_total_gc_waste);
 451     _perf_max_gc_waste         ->set_value(_max_gc_waste);
 452     _perf_slow_refill_waste    ->set_value(_total_slow_refill_waste);
 453     _perf_max_slow_refill_waste->set_value(_max_slow_refill_waste);
 454     _perf_fast_refill_waste    ->set_value(_total_fast_refill_waste);
 455     _perf_max_fast_refill_waste->set_value(_max_fast_refill_waste);
 456     _perf_slow_allocations     ->set_value(_total_slow_allocations);
 457     _perf_max_slow_allocations ->set_value(_max_slow_allocations);
 458   }
 459 }
 460 
 461 void GlobalTLABStats::print() {
 462   Log(gc, tlab) log;
 463   if (!log.is_debug()) {
 464     return;
 465   }
 466 
 467   size_t waste = _total_gc_waste + _total_slow_refill_waste + _total_fast_refill_waste;
 468   double waste_percent = percent_of(waste, _total_allocation);
 469   log.debug("TLAB totals: thrds: %d  refills: %d max: %d"
 470             " slow allocs: %d max %d waste: %4.1f%%"
 471             " gc: " SIZE_FORMAT "B max: " SIZE_FORMAT "B"
 472             " slow: " SIZE_FORMAT "B max: " SIZE_FORMAT "B"
 473             " fast: " SIZE_FORMAT "B max: " SIZE_FORMAT "B",
 474             _allocating_threads,
 475             _total_refills, _max_refills,
 476             _total_slow_allocations, _max_slow_allocations,
 477             waste_percent,
 478             _total_gc_waste * HeapWordSize,
 479             _max_gc_waste * HeapWordSize,
 480             _total_slow_refill_waste * HeapWordSize,
 481             _max_slow_refill_waste * HeapWordSize,
 482             _total_fast_refill_waste * HeapWordSize,
 483             _max_fast_refill_waste * HeapWordSize);
 484 }