duke@435: /*
poonam@9965:  * Copyright (c) 1999, 2020, Oracle and/or its affiliates. All rights reserved.
duke@435:  * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
duke@435:  *
duke@435:  * This code is free software; you can redistribute it and/or modify it
duke@435:  * under the terms of the GNU General Public License version 2 only, as
duke@435:  * published by the Free Software Foundation.
duke@435:  *
duke@435:  * This code is distributed in the hope that it will be useful, but WITHOUT
duke@435:  * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
duke@435:  * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License
duke@435:  * version 2 for more details (a copy is included in the LICENSE file that
duke@435:  * accompanied this code).
duke@435:  *
duke@435:  * You should have received a copy of the GNU General Public License version
duke@435:  * 2 along with this work; if not, write to the Free Software Foundation,
duke@435:  * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA.
duke@435:  *
trims@1907:  * Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA
trims@1907:  * or visit www.oracle.com if you need additional information or have any
trims@1907:  * questions.
duke@435:  *
duke@435:  */
duke@435: 
stefank@2314: #include "precompiled.hpp"
stefank@2314: #include "memory/genCollectedHeap.hpp"
stefank@2314: #include "memory/resourceArea.hpp"
stefank@2314: #include "memory/threadLocalAllocBuffer.inline.hpp"
stefank@2314: #include "memory/universe.inline.hpp"
stefank@2314: #include "oops/oop.inline.hpp"
stefank@4299: #include "runtime/thread.inline.hpp"
stefank@2314: #include "utilities/copy.hpp"
stefank@2314: 
drchase@6680: PRAGMA_FORMAT_MUTE_WARNINGS_FOR_GCC
drchase@6680: 
duke@435: // Thread-Local Edens support
duke@435: 
duke@435: // static member initialization
brutisso@6376: size_t           ThreadLocalAllocBuffer::_max_size       = 0;
duke@435: unsigned         ThreadLocalAllocBuffer::_target_refills = 0;
duke@435: GlobalTLABStats* ThreadLocalAllocBuffer::_global_stats   = NULL;
duke@435: 
duke@435: void ThreadLocalAllocBuffer::clear_before_allocation() {
duke@435:   _slow_refill_waste += (unsigned)remaining();
duke@435:   make_parsable(true);   // also retire the TLAB
duke@435: }
duke@435: 
duke@435: void ThreadLocalAllocBuffer::accumulate_statistics_before_gc() {
duke@435:   global_stats()->initialize();
duke@435: 
brutisso@6376:   for (JavaThread *thread = Threads::first(); thread != NULL; thread = thread->next()) {
duke@435:     thread->tlab().accumulate_statistics();
duke@435:     thread->tlab().initialize_statistics();
duke@435:   }
duke@435: 
duke@435:   // Publish new stats if some allocation occurred.
duke@435:   if (global_stats()->allocation() != 0) {
duke@435:     global_stats()->publish();
duke@435:     if (PrintTLAB) {
duke@435:       global_stats()->print();
duke@435:     }
duke@435:   }
duke@435: }
duke@435: 
duke@435: void ThreadLocalAllocBuffer::accumulate_statistics() {
brutisso@6376:   Thread* thread = myThread();
brutisso@6376:   size_t capacity = Universe::heap()->tlab_capacity(thread);
brutisso@6376:   size_t used     = Universe::heap()->tlab_used(thread);
duke@435: 
duke@435:   _gc_waste += (unsigned)remaining();
brutisso@6376:   size_t total_allocated = thread->allocated_bytes();
brutisso@6376:   size_t allocated_since_last_gc = total_allocated - _allocated_before_last_gc;
brutisso@6376:   _allocated_before_last_gc = total_allocated;
duke@435: 
duke@435:   if (PrintTLAB && (_number_of_refills > 0 || Verbose)) {
duke@435:     print_stats("gc");
duke@435:   }
duke@435: 
duke@435:   if (_number_of_refills > 0) {
brutisso@6376:     // Update allocation history if a reasonable amount of eden was allocated.
brutisso@6376:     bool update_allocation_history = used > 0.5 * capacity;
duke@435: 
duke@435:     if (update_allocation_history) {
duke@435:       // Average the fraction of eden allocated in a tlab by this
duke@435:       // thread for use in the next resize operation.
duke@435:       // _gc_waste is not subtracted because it's included in
duke@435:       // "used".
brutisso@6376:       // The result can be larger than 1.0 due to direct to old allocations.
brutisso@6376:       // These allocations should ideally not be counted but since it is not possible
brutisso@6376:       // to filter them out here we just cap the fraction to be at most 1.0.
poonam@9965:       // Keep alloc_frac as float and not double to avoid the double to float conversion
poonam@9965:       float alloc_frac = MIN2(1.0f, allocated_since_last_gc / (float) used);
duke@435:       _allocation_fraction.sample(alloc_frac);
duke@435:     }
duke@435:     global_stats()->update_allocating_threads();
duke@435:     global_stats()->update_number_of_refills(_number_of_refills);
duke@435:     global_stats()->update_allocation(_number_of_refills * desired_size());
duke@435:     global_stats()->update_gc_waste(_gc_waste);
duke@435:     global_stats()->update_slow_refill_waste(_slow_refill_waste);
duke@435:     global_stats()->update_fast_refill_waste(_fast_refill_waste);
duke@435: 
duke@435:   } else {
duke@435:     assert(_number_of_refills == 0 && _fast_refill_waste == 0 &&
duke@435:            _slow_refill_waste == 0 && _gc_waste          == 0,
duke@435:            "tlab stats == 0");
duke@435:   }
duke@435:   global_stats()->update_slow_allocations(_slow_allocations);
duke@435: }
duke@435: 
duke@435: // Fills the current tlab with a dummy filler array to create
duke@435: // an illusion of a contiguous Eden and optionally retires the tlab.
duke@435: // Waste accounting should be done in caller as appropriate; see,
duke@435: // for example, clear_before_allocation().
duke@435: void ThreadLocalAllocBuffer::make_parsable(bool retire) {
duke@435:   if (end() != NULL) {
duke@435:     invariants();
phh@2423: 
phh@2423:     if (retire) {
phh@2423:       myThread()->incr_allocated_bytes(used_bytes());
phh@2423:     }
phh@2423: 
johnc@1600:     CollectedHeap::fill_with_object(top(), hard_end(), retire);
duke@435: 
duke@435:     if (retire || ZeroTLAB) {  // "Reset" the TLAB
duke@435:       set_start(NULL);
duke@435:       set_top(NULL);
duke@435:       set_pf_top(NULL);
duke@435:       set_end(NULL);
duke@435:     }
duke@435:   }
duke@435:   assert(!(retire || ZeroTLAB)  ||
duke@435:          (start() == NULL && end() == NULL && top() == NULL),
duke@435:          "TLAB must be reset");
duke@435: }
duke@435: 
duke@435: void ThreadLocalAllocBuffer::resize_all_tlabs() {
brutisso@6376:   if (ResizeTLAB) {
brutisso@6376:     for (JavaThread *thread = Threads::first(); thread != NULL; thread = thread->next()) {
brutisso@6376:       thread->tlab().resize();
brutisso@6376:     }
duke@435:   }
duke@435: }
duke@435: 
duke@435: void ThreadLocalAllocBuffer::resize() {
brutisso@6376:   // Compute the next tlab size using expected allocation amount
brutisso@6376:   assert(ResizeTLAB, "Should not call this otherwise");
brutisso@6376:   size_t alloc = (size_t)(_allocation_fraction.average() *
brutisso@6376:                           (Universe::heap()->tlab_capacity(myThread()) / HeapWordSize));
brutisso@6376:   size_t new_size = alloc / _target_refills;
duke@435: 
brutisso@6376:   new_size = MIN2(MAX2(new_size, min_size()), max_size());
duke@435: 
brutisso@6376:   size_t aligned_new_size = align_object_size(new_size);
duke@435: 
brutisso@6376:   if (PrintTLAB && Verbose) {
brutisso@6376:     gclog_or_tty->print("TLAB new size: thread: " INTPTR_FORMAT " [id: %2d]"
brutisso@6376:                         " refills %d  alloc: %8.6f desired_size: " SIZE_FORMAT " -> " SIZE_FORMAT "\n",
brutisso@6376:                         myThread(), myThread()->osthread()->thread_id(),
brutisso@6376:                         _target_refills, _allocation_fraction.average(), desired_size(), aligned_new_size);
duke@435:   }
brutisso@6376:   set_desired_size(aligned_new_size);
brutisso@6376:   set_refill_waste_limit(initial_refill_waste_limit());
duke@435: }
duke@435: 
duke@435: void ThreadLocalAllocBuffer::initialize_statistics() {
duke@435:     _number_of_refills = 0;
duke@435:     _fast_refill_waste = 0;
duke@435:     _slow_refill_waste = 0;
duke@435:     _gc_waste          = 0;
duke@435:     _slow_allocations  = 0;
duke@435: }
duke@435: 
duke@435: void ThreadLocalAllocBuffer::fill(HeapWord* start,
duke@435:                                   HeapWord* top,
duke@435:                                   size_t    new_size) {
duke@435:   _number_of_refills++;
duke@435:   if (PrintTLAB && Verbose) {
duke@435:     print_stats("fill");
duke@435:   }
duke@435:   assert(top <= start + new_size - alignment_reserve(), "size too small");
duke@435:   initialize(start, top, start + new_size - alignment_reserve());
duke@435: 
duke@435:   // Reset amount of internal fragmentation
duke@435:   set_refill_waste_limit(initial_refill_waste_limit());
duke@435: }
duke@435: 
duke@435: void ThreadLocalAllocBuffer::initialize(HeapWord* start,
duke@435:                                         HeapWord* top,
duke@435:                                         HeapWord* end) {
duke@435:   set_start(start);
duke@435:   set_top(top);
duke@435:   set_pf_top(top);
duke@435:   set_end(end);
duke@435:   invariants();
duke@435: }
duke@435: 
duke@435: void ThreadLocalAllocBuffer::initialize() {
duke@435:   initialize(NULL,                    // start
duke@435:              NULL,                    // top
duke@435:              NULL);                   // end
duke@435: 
duke@435:   set_desired_size(initial_desired_size());
duke@435: 
dbuck@9289:   // Following check is needed because at startup the main
duke@435:   // thread is initialized before the heap is.  The initialization for
duke@435:   // this thread is redone in startup_initialization below.
duke@435:   if (Universe::heap() != NULL) {
duke@435:     size_t capacity   = Universe::heap()->tlab_capacity(myThread()) / HeapWordSize;
poonam@9965:     // Keep alloc_frac as float and not double to avoid the double to float conversion
poonam@9965:     float alloc_frac = desired_size() * target_refills() / (float) capacity;
duke@435:     _allocation_fraction.sample(alloc_frac);
duke@435:   }
duke@435: 
duke@435:   set_refill_waste_limit(initial_refill_waste_limit());
duke@435: 
duke@435:   initialize_statistics();
duke@435: }
duke@435: 
duke@435: void ThreadLocalAllocBuffer::startup_initialization() {
duke@435: 
duke@435:   // Assuming each thread's active tlab is, on average,
duke@435:   // 1/2 full at a GC
duke@435:   _target_refills = 100 / (2 * TLABWasteTargetPercent);
duke@435:   _target_refills = MAX2(_target_refills, (unsigned)1U);
duke@435: 
duke@435:   _global_stats = new GlobalTLABStats();
duke@435: 
dbuck@9289:   // During jvm startup, the main thread is initialized
duke@435:   // before the heap is initialized.  So reinitialize it now.
duke@435:   guarantee(Thread::current()->is_Java_thread(), "tlab initialization thread not Java thread");
duke@435:   Thread::current()->tlab().initialize();
duke@435: 
duke@435:   if (PrintTLAB && Verbose) {
duke@435:     gclog_or_tty->print("TLAB min: " SIZE_FORMAT " initial: " SIZE_FORMAT " max: " SIZE_FORMAT "\n",
duke@435:                         min_size(), Thread::current()->tlab().initial_desired_size(), max_size());
duke@435:   }
duke@435: }
duke@435: 
duke@435: size_t ThreadLocalAllocBuffer::initial_desired_size() {
mgerdin@7470:   size_t init_sz = 0;
duke@435: 
duke@435:   if (TLABSize > 0) {
mgerdin@7470:     init_sz = TLABSize / HeapWordSize;
mgerdin@7470:   } else if (global_stats() != NULL) {
duke@435:     // Initial size is a function of the average number of allocating threads.
duke@435:     unsigned nof_threads = global_stats()->allocating_threads_avg();
duke@435: 
duke@435:     init_sz  = (Universe::heap()->tlab_capacity(myThread()) / HeapWordSize) /
duke@435:                       (nof_threads * target_refills());
duke@435:     init_sz = align_object_size(init_sz);
duke@435:   }
mgerdin@7470:   init_sz = MIN2(MAX2(init_sz, min_size()), max_size());
duke@435:   return init_sz;
duke@435: }
duke@435: 
duke@435: void ThreadLocalAllocBuffer::print_stats(const char* tag) {
duke@435:   Thread* thrd = myThread();
duke@435:   size_t waste = _gc_waste + _slow_refill_waste + _fast_refill_waste;
duke@435:   size_t alloc = _number_of_refills * _desired_size;
duke@435:   double waste_percent = alloc == 0 ? 0.0 :
duke@435:                       100.0 * waste / alloc;
brutisso@6376:   size_t tlab_used  = Universe::heap()->tlab_used(thrd);
duke@435:   gclog_or_tty->print("TLAB: %s thread: " INTPTR_FORMAT " [id: %2d]"
duke@435:                       " desired_size: " SIZE_FORMAT "KB"
duke@435:                       " slow allocs: %d  refill waste: " SIZE_FORMAT "B"
duke@435:                       " alloc:%8.5f %8.0fKB refills: %d waste %4.1f%% gc: %dB"
duke@435:                       " slow: %dB fast: %dB\n",
duke@435:                       tag, thrd, thrd->osthread()->thread_id(),
duke@435:                       _desired_size / (K / HeapWordSize),
duke@435:                       _slow_allocations, _refill_waste_limit * HeapWordSize,
duke@435:                       _allocation_fraction.average(),
duke@435:                       _allocation_fraction.average() * tlab_used / K,
duke@435:                       _number_of_refills, waste_percent,
duke@435:                       _gc_waste * HeapWordSize,
duke@435:                       _slow_refill_waste * HeapWordSize,
duke@435:                       _fast_refill_waste * HeapWordSize);
duke@435: }
duke@435: 
duke@435: void ThreadLocalAllocBuffer::verify() {
duke@435:   HeapWord* p = start();
duke@435:   HeapWord* t = top();
duke@435:   HeapWord* prev_p = NULL;
duke@435:   while (p < t) {
duke@435:     oop(p)->verify();
duke@435:     prev_p = p;
duke@435:     p += oop(p)->size();
duke@435:   }
duke@435:   guarantee(p == top(), "end of last object must match end of space");
duke@435: }
duke@435: 
duke@435: Thread* ThreadLocalAllocBuffer::myThread() {
duke@435:   return (Thread*)(((char *)this) +
duke@435:                    in_bytes(start_offset()) -
duke@435:                    in_bytes(Thread::tlab_start_offset()));
duke@435: }
duke@435: 
duke@435: 
duke@435: GlobalTLABStats::GlobalTLABStats() :
duke@435:   _allocating_threads_avg(TLABAllocationWeight) {
duke@435: 
duke@435:   initialize();
duke@435: 
duke@435:   _allocating_threads_avg.sample(1); // One allocating thread at startup
duke@435: 
duke@435:   if (UsePerfData) {
duke@435: 
duke@435:     EXCEPTION_MARK;
duke@435:     ResourceMark rm;
duke@435: 
duke@435:     char* cname = PerfDataManager::counter_name("tlab", "allocThreads");
duke@435:     _perf_allocating_threads =
duke@435:       PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_None, CHECK);
duke@435: 
duke@435:     cname = PerfDataManager::counter_name("tlab", "fills");
duke@435:     _perf_total_refills =
duke@435:       PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_None, CHECK);
duke@435: 
duke@435:     cname = PerfDataManager::counter_name("tlab", "maxFills");
duke@435:     _perf_max_refills =
duke@435:       PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_None, CHECK);
duke@435: 
duke@435:     cname = PerfDataManager::counter_name("tlab", "alloc");
duke@435:     _perf_allocation =
duke@435:       PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_Bytes, CHECK);
duke@435: 
duke@435:     cname = PerfDataManager::counter_name("tlab", "gcWaste");
duke@435:     _perf_gc_waste =
duke@435:       PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_Bytes, CHECK);
duke@435: 
duke@435:     cname = PerfDataManager::counter_name("tlab", "maxGcWaste");
duke@435:     _perf_max_gc_waste =
duke@435:       PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_Bytes, CHECK);
duke@435: 
duke@435:     cname = PerfDataManager::counter_name("tlab", "slowWaste");
duke@435:     _perf_slow_refill_waste =
duke@435:       PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_Bytes, CHECK);
duke@435: 
duke@435:     cname = PerfDataManager::counter_name("tlab", "maxSlowWaste");
duke@435:     _perf_max_slow_refill_waste =
duke@435:       PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_Bytes, CHECK);
duke@435: 
duke@435:     cname = PerfDataManager::counter_name("tlab", "fastWaste");
duke@435:     _perf_fast_refill_waste =
duke@435:       PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_Bytes, CHECK);
duke@435: 
duke@435:     cname = PerfDataManager::counter_name("tlab", "maxFastWaste");
duke@435:     _perf_max_fast_refill_waste =
duke@435:       PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_Bytes, CHECK);
duke@435: 
duke@435:     cname = PerfDataManager::counter_name("tlab", "slowAlloc");
duke@435:     _perf_slow_allocations =
duke@435:       PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_None, CHECK);
duke@435: 
duke@435:     cname = PerfDataManager::counter_name("tlab", "maxSlowAlloc");
duke@435:     _perf_max_slow_allocations =
duke@435:       PerfDataManager::create_variable(SUN_GC, cname, PerfData::U_None, CHECK);
duke@435:   }
duke@435: }
duke@435: 
duke@435: void GlobalTLABStats::initialize() {
duke@435:   // Clear counters summarizing info from all threads
duke@435:   _allocating_threads      = 0;
duke@435:   _total_refills           = 0;
duke@435:   _max_refills             = 0;
duke@435:   _total_allocation        = 0;
duke@435:   _total_gc_waste          = 0;
duke@435:   _max_gc_waste            = 0;
duke@435:   _total_slow_refill_waste = 0;
duke@435:   _max_slow_refill_waste   = 0;
duke@435:   _total_fast_refill_waste = 0;
duke@435:   _max_fast_refill_waste   = 0;
duke@435:   _total_slow_allocations  = 0;
duke@435:   _max_slow_allocations    = 0;
duke@435: }
duke@435: 
duke@435: void GlobalTLABStats::publish() {
duke@435:   _allocating_threads_avg.sample(_allocating_threads);
duke@435:   if (UsePerfData) {
duke@435:     _perf_allocating_threads   ->set_value(_allocating_threads);
duke@435:     _perf_total_refills        ->set_value(_total_refills);
duke@435:     _perf_max_refills          ->set_value(_max_refills);
duke@435:     _perf_allocation           ->set_value(_total_allocation);
duke@435:     _perf_gc_waste             ->set_value(_total_gc_waste);
duke@435:     _perf_max_gc_waste         ->set_value(_max_gc_waste);
duke@435:     _perf_slow_refill_waste    ->set_value(_total_slow_refill_waste);
duke@435:     _perf_max_slow_refill_waste->set_value(_max_slow_refill_waste);
duke@435:     _perf_fast_refill_waste    ->set_value(_total_fast_refill_waste);
duke@435:     _perf_max_fast_refill_waste->set_value(_max_fast_refill_waste);
duke@435:     _perf_slow_allocations     ->set_value(_total_slow_allocations);
duke@435:     _perf_max_slow_allocations ->set_value(_max_slow_allocations);
duke@435:   }
duke@435: }
duke@435: 
duke@435: void GlobalTLABStats::print() {
duke@435:   size_t waste = _total_gc_waste + _total_slow_refill_waste + _total_fast_refill_waste;
duke@435:   double waste_percent = _total_allocation == 0 ? 0.0 :
duke@435:                          100.0 * waste / _total_allocation;
duke@435:   gclog_or_tty->print("TLAB totals: thrds: %d  refills: %d max: %d"
duke@435:                       " slow allocs: %d max %d waste: %4.1f%%"
duke@435:                       " gc: " SIZE_FORMAT "B max: " SIZE_FORMAT "B"
duke@435:                       " slow: " SIZE_FORMAT "B max: " SIZE_FORMAT "B"
duke@435:                       " fast: " SIZE_FORMAT "B max: " SIZE_FORMAT "B\n",
duke@435:                       _allocating_threads,
duke@435:                       _total_refills, _max_refills,
duke@435:                       _total_slow_allocations, _max_slow_allocations,
duke@435:                       waste_percent,
duke@435:                       _total_gc_waste * HeapWordSize,
duke@435:                       _max_gc_waste * HeapWordSize,
duke@435:                       _total_slow_refill_waste * HeapWordSize,
duke@435:                       _max_slow_refill_waste * HeapWordSize,
duke@435:                       _total_fast_refill_waste * HeapWordSize,
duke@435:                       _max_fast_refill_waste * HeapWordSize);
duke@435: }