1 /*
   2  * Copyright (c) 1999, 2012, 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 #ifndef SHARE_VM_MEMORY_THREADLOCALALLOCBUFFER_HPP
  26 #define SHARE_VM_MEMORY_THREADLOCALALLOCBUFFER_HPP
  27 
  28 #include "gc_implementation/shared/gcUtil.hpp"
  29 #include "oops/typeArrayOop.hpp"
  30 #include "runtime/perfData.hpp"
  31 
  32 class GlobalTLABStats;
  33 
  34 // ThreadLocalAllocBuffer: a descriptor for thread-local storage used by
  35 // the threads for allocation.
  36 //            It is thread-private at any time, but maybe multiplexed over
  37 //            time across multiple threads. The park()/unpark() pair is
  38 //            used to make it available for such multiplexing.
  39 class ThreadLocalAllocBuffer: public CHeapObj<mtThread> {
  40   friend class VMStructs;
  41 private:
  42   HeapWord* _start;                              // address of TLAB
  43   HeapWord* _top;                                // address after last allocation
  44   HeapWord* _pf_top;                             // allocation prefetch watermark
  45   HeapWord* _end;                                // allocation end (excluding alignment_reserve)
  46   size_t    _desired_size;                       // desired size   (including alignment_reserve)
  47   size_t    _refill_waste_limit;                 // hold onto tlab if free() is larger than this
  48   size_t    _allocated_before_last_gc;           // total bytes allocated up until the last gc
  49 
  50   static size_t   _max_size;                     // maximum size of any TLAB
  51   static unsigned _target_refills;               // expected number of refills between GCs
  52 
  53   unsigned  _number_of_refills;
  54   unsigned  _fast_refill_waste;
  55   unsigned  _slow_refill_waste;
  56   unsigned  _gc_waste;
  57   unsigned  _slow_allocations;
  58 
  59   AdaptiveWeightedAverage _allocation_fraction;  // fraction of eden allocated in tlabs
  60 
  61   bool _gclab;
  62   bool _initialized;
  63 
  64   void set_start(HeapWord* start)                { _start = start; }
  65   void set_end(HeapWord* end)                    { _end = end; }
  66   void set_top(HeapWord* top)                    { _top = top; }
  67   void set_pf_top(HeapWord* pf_top)              { _pf_top = pf_top; }
  68   void set_desired_size(size_t desired_size)     { _desired_size = desired_size; }
  69   void set_refill_waste_limit(size_t waste)      { _refill_waste_limit = waste;  }
  70 
  71   size_t initial_refill_waste_limit()            { return desired_size() / TLABRefillWasteFraction; }
  72 
  73   static int    target_refills()                 { return _target_refills; }
  74   size_t initial_desired_size();
  75 
  76   size_t remaining() const                       { return end() == NULL ? 0 : pointer_delta(hard_end(), top()); }
  77 
  78   // Make parsable and release it.
  79   void reset();
  80 
  81   void invariants() const { assert(top() >= start() && top() <= end(), "invalid tlab"); }
  82 
  83   void initialize(HeapWord* start, HeapWord* top, HeapWord* end);
  84 
  85   void print_stats(const char* tag);
  86 
  87   Thread* myThread();
  88 
  89   // statistics
  90 
  91   int number_of_refills() const { return _number_of_refills; }
  92   int fast_refill_waste() const { return _fast_refill_waste; }
  93   int slow_refill_waste() const { return _slow_refill_waste; }
  94   int gc_waste() const          { return _gc_waste; }
  95   int slow_allocations() const  { return _slow_allocations; }
  96 
  97   static GlobalTLABStats* _global_stats;
  98   static GlobalTLABStats* global_stats() { return _global_stats; }
  99 
 100 public:
 101   ThreadLocalAllocBuffer() : _allocation_fraction(TLABAllocationWeight), _allocated_before_last_gc(0), _initialized(false) {
 102     // do nothing.  tlabs must be inited by initialize() calls
 103   }
 104 
 105   bool is_initialized() const { return _initialized; };
 106 
 107   static const size_t min_size()                 { return align_object_size(MinTLABSize / HeapWordSize) + alignment_reserve(); }
 108   static const size_t max_size()                 { assert(_max_size != 0, "max_size not set up"); return _max_size; }
 109   static void set_max_size(size_t max_size)      { _max_size = max_size; }
 110 
 111   HeapWord* start() const                        { return _start; }
 112   HeapWord* end() const                          { return _end; }
 113   HeapWord* hard_end() const                     { return _end + alignment_reserve(); }
 114   HeapWord* top() const                          { return _top; }
 115   HeapWord* pf_top() const                       { return _pf_top; }
 116   size_t desired_size() const                    { return _desired_size; }
 117   size_t used() const                            { return pointer_delta(top(), start()); }
 118   size_t used_bytes() const                      { return pointer_delta(top(), start(), 1); }
 119   size_t free() const                            { return pointer_delta(end(), top()); }
 120   // Don't discard tlab if remaining space is larger than this.
 121   size_t refill_waste_limit() const              { return _refill_waste_limit; }
 122 
 123   // Allocate size HeapWords. The memory is NOT initialized to zero.
 124   inline HeapWord* allocate(size_t size);
 125 
 126   // Reserve space at the end of TLAB
 127   static size_t end_reserve() {
 128     int reserve_size = typeArrayOopDesc::header_size(T_INT);
 129     return MAX2(reserve_size, VM_Version::reserve_for_allocation_prefetch());
 130   }
 131 
 132   // Resize based on amount of allocation, etc.
 133   void resize();
 134 
 135   void accumulate_statistics();
 136   void initialize_statistics();
 137 
 138   // Rolls back a single allocation of the given size.
 139   void rollback(size_t size);
 140 
 141   static size_t alignment_reserve()              { return align_object_size(end_reserve()); }
 142   static size_t alignment_reserve_in_bytes()     { return alignment_reserve() * HeapWordSize; }
 143 
 144   // Return tlab size or remaining space in eden such that the
 145   // space is large enough to hold obj_size and necessary fill space.
 146   // Otherwise return 0;
 147   inline size_t compute_size(size_t obj_size);
 148 
 149   // Record slow allocation
 150   inline void record_slow_allocation(size_t obj_size);
 151 
 152   // Initialization at startup
 153   static void startup_initialization();
 154 
 155   // Make an in-use tlab parsable, optionally also retiring it.
 156   void make_parsable(bool retire);
 157 
 158   // Retire in-use tlab before allocation of a new tlab
 159   void clear_before_allocation();
 160 
 161   // Accumulate statistics across all tlabs before gc
 162   static void accumulate_statistics_before_gc();
 163 
 164   // Resize tlabs for all threads
 165   static void resize_all_tlabs();
 166 
 167   void fill(HeapWord* start, HeapWord* top, size_t new_size);
 168   void initialize(bool gclab);
 169 
 170   static size_t refill_waste_limit_increment()   { return TLABWasteIncrement; }
 171 
 172   // Code generation support
 173   static ByteSize start_offset()                 { return byte_offset_of(ThreadLocalAllocBuffer, _start); }
 174   static ByteSize end_offset()                   { return byte_offset_of(ThreadLocalAllocBuffer, _end  ); }
 175   static ByteSize top_offset()                   { return byte_offset_of(ThreadLocalAllocBuffer, _top  ); }
 176   static ByteSize pf_top_offset()                { return byte_offset_of(ThreadLocalAllocBuffer, _pf_top  ); }
 177   static ByteSize size_offset()                  { return byte_offset_of(ThreadLocalAllocBuffer, _desired_size ); }
 178   static ByteSize refill_waste_limit_offset()    { return byte_offset_of(ThreadLocalAllocBuffer, _refill_waste_limit ); }
 179 
 180   static ByteSize number_of_refills_offset()     { return byte_offset_of(ThreadLocalAllocBuffer, _number_of_refills ); }
 181   static ByteSize fast_refill_waste_offset()     { return byte_offset_of(ThreadLocalAllocBuffer, _fast_refill_waste ); }
 182   static ByteSize slow_allocations_offset()      { return byte_offset_of(ThreadLocalAllocBuffer, _slow_allocations ); }
 183 
 184   void verify();
 185 };
 186 
 187 class GlobalTLABStats: public CHeapObj<mtThread> {
 188 private:
 189 
 190   // Accumulate perfdata in private variables because
 191   // PerfData should be write-only for security reasons
 192   // (see perfData.hpp)
 193   unsigned _allocating_threads;
 194   unsigned _total_refills;
 195   unsigned _max_refills;
 196   size_t   _total_allocation;
 197   size_t   _total_gc_waste;
 198   size_t   _max_gc_waste;
 199   size_t   _total_slow_refill_waste;
 200   size_t   _max_slow_refill_waste;
 201   size_t   _total_fast_refill_waste;
 202   size_t   _max_fast_refill_waste;
 203   unsigned _total_slow_allocations;
 204   unsigned _max_slow_allocations;
 205 
 206   PerfVariable* _perf_allocating_threads;
 207   PerfVariable* _perf_total_refills;
 208   PerfVariable* _perf_max_refills;
 209   PerfVariable* _perf_allocation;
 210   PerfVariable* _perf_gc_waste;
 211   PerfVariable* _perf_max_gc_waste;
 212   PerfVariable* _perf_slow_refill_waste;
 213   PerfVariable* _perf_max_slow_refill_waste;
 214   PerfVariable* _perf_fast_refill_waste;
 215   PerfVariable* _perf_max_fast_refill_waste;
 216   PerfVariable* _perf_slow_allocations;
 217   PerfVariable* _perf_max_slow_allocations;
 218 
 219   AdaptiveWeightedAverage _allocating_threads_avg;
 220 
 221 public:
 222   GlobalTLABStats();
 223 
 224   // Initialize all counters
 225   void initialize();
 226 
 227   // Write all perf counters to the perf_counters
 228   void publish();
 229 
 230   void print();
 231 
 232   // Accessors
 233   unsigned allocating_threads_avg() {
 234     return MAX2((unsigned)(_allocating_threads_avg.average() + 0.5), 1U);
 235   }
 236 
 237   size_t allocation() {
 238     return _total_allocation;
 239   }
 240 
 241   // Update methods
 242 
 243   void update_allocating_threads() {
 244     _allocating_threads++;
 245   }
 246   void update_number_of_refills(unsigned value) {
 247     _total_refills += value;
 248     _max_refills    = MAX2(_max_refills, value);
 249   }
 250   void update_allocation(size_t value) {
 251     _total_allocation += value;
 252   }
 253   void update_gc_waste(size_t value) {
 254     _total_gc_waste += value;
 255     _max_gc_waste    = MAX2(_max_gc_waste, value);
 256   }
 257   void update_fast_refill_waste(size_t value) {
 258     _total_fast_refill_waste += value;
 259     _max_fast_refill_waste    = MAX2(_max_fast_refill_waste, value);
 260   }
 261   void update_slow_refill_waste(size_t value) {
 262     _total_slow_refill_waste += value;
 263     _max_slow_refill_waste    = MAX2(_max_slow_refill_waste, value);
 264   }
 265   void update_slow_allocations(unsigned value) {
 266     _total_slow_allocations += value;
 267     _max_slow_allocations    = MAX2(_max_slow_allocations, value);
 268   }
 269 };
 270 
 271 #endif // SHARE_VM_MEMORY_THREADLOCALALLOCBUFFER_HPP