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 #ifndef SHARE_VM_GC_SHARED_THREADLOCALALLOCBUFFER_HPP
  26 #define SHARE_VM_GC_SHARED_THREADLOCALALLOCBUFFER_HPP
  27 
  28 #include "gc/shared/gcUtil.hpp"
  29 #include "oops/typeArrayOop.hpp"
  30 #include "runtime/perfData.hpp"
  31 #include "runtime/vm_version.hpp"
  32 
  33 class GlobalTLABStats;
  34 
  35 // ThreadLocalAllocBuffer: a descriptor for thread-local storage used by
  36 // the threads for allocation.
  37 //            It is thread-private at any time, but maybe multiplexed over
  38 //            time across multiple threads. The park()/unpark() pair is
  39 //            used to make it available for such multiplexing.
  40 class ThreadLocalAllocBuffer: public CHeapObj<mtThread> {
  41   friend class VMStructs;
  42   friend class JVMCIVMStructs;
  43 private:
  44   HeapWord* _start;                              // address of TLAB
  45   HeapWord* _top;                                // address after last allocation
  46   HeapWord* _pf_top;                             // allocation prefetch watermark
  47   HeapWord* _end;                                // allocation end (excluding alignment_reserve)
  48   size_t    _desired_size;                       // desired size   (including alignment_reserve)
  49   size_t    _refill_waste_limit;                 // hold onto tlab if free() is larger than this
  50   size_t    _allocated_before_last_gc;           // total bytes allocated up until the last gc
  51 
  52   static size_t   _max_size;                          // maximum size of any TLAB
  53   static int      _reserve_for_allocation_prefetch;   // Reserve at the end of the TLAB
  54   static unsigned _target_refills;                    // expected number of refills between GCs
  55 
  56   unsigned  _number_of_refills;
  57   unsigned  _fast_refill_waste;
  58   unsigned  _slow_refill_waste;
  59   unsigned  _gc_waste;
  60   unsigned  _slow_allocations;
  61 
  62   AdaptiveWeightedAverage _allocation_fraction;  // fraction of eden allocated in tlabs
  63 
  64   void accumulate_statistics();
  65   void initialize_statistics();
  66 
  67   void set_start(HeapWord* start)                { _start = start; }
  68   void set_end(HeapWord* end)                    { _end = end; }
  69   void set_top(HeapWord* top)                    { _top = top; }
  70   void set_pf_top(HeapWord* pf_top)              { _pf_top = pf_top; }
  71   void set_desired_size(size_t desired_size)     { _desired_size = desired_size; }
  72   void set_refill_waste_limit(size_t waste)      { _refill_waste_limit = waste;  }
  73 
  74   size_t initial_refill_waste_limit()            { return desired_size() / TLABRefillWasteFraction; }
  75 
  76   static int    target_refills()                 { return _target_refills; }
  77   size_t initial_desired_size();
  78 
  79   size_t remaining() const                       { return end() == NULL ? 0 : pointer_delta(hard_end(), top()); }
  80 
  81   // Make parsable and release it.
  82   void reset();
  83 
  84   // Resize based on amount of allocation, etc.
  85   void resize();
  86 
  87   void invariants() const { assert(top() >= start() && top() <= end(), "invalid tlab"); }
  88 
  89   void initialize(HeapWord* start, HeapWord* top, HeapWord* end);
  90 
  91   void print_stats(const char* tag);
  92 
  93   Thread* myThread();
  94 
  95   // statistics
  96 
  97   int number_of_refills() const { return _number_of_refills; }
  98   int fast_refill_waste() const { return _fast_refill_waste; }
  99   int slow_refill_waste() const { return _slow_refill_waste; }
 100   int gc_waste() const          { return _gc_waste; }
 101   int slow_allocations() const  { return _slow_allocations; }
 102 
 103   static GlobalTLABStats* _global_stats;
 104   static GlobalTLABStats* global_stats() { return _global_stats; }
 105 
 106 public:
 107   ThreadLocalAllocBuffer() : _allocation_fraction(TLABAllocationWeight), _allocated_before_last_gc(0) {
 108     // do nothing.  tlabs must be inited by initialize() calls
 109   }
 110 
 111   static size_t min_size()                       { return align_object_size(MinTLABSize / HeapWordSize) + alignment_reserve(); }
 112   static size_t max_size()                       { assert(_max_size != 0, "max_size not set up"); return _max_size; }
 113   static size_t max_size_in_bytes()              { return max_size() * BytesPerWord; }
 114   static void set_max_size(size_t max_size)      { _max_size = max_size; }
 115 
 116   HeapWord* start() const                        { return _start; }
 117   HeapWord* end() const                          { return _end; }
 118   HeapWord* hard_end() const                     { return _end + alignment_reserve(); }
 119   HeapWord* top() const                          { return _top; }
 120   HeapWord* pf_top() const                       { return _pf_top; }
 121   size_t desired_size() const                    { return _desired_size; }
 122   size_t used() const                            { return pointer_delta(top(), start()); }
 123   size_t used_bytes() const                      { return pointer_delta(top(), start(), 1); }
 124   size_t free() const                            { return pointer_delta(end(), top()); }
 125   // Don't discard tlab if remaining space is larger than this.
 126   size_t refill_waste_limit() const              { return _refill_waste_limit; }
 127 
 128   // Allocate size HeapWords. The memory is NOT initialized to zero.
 129   inline HeapWord* allocate(size_t size);
 130 
 131   // Reserve space at the end of TLAB
 132   static size_t end_reserve() {
 133     int reserve_size = typeArrayOopDesc::header_size(T_INT);
 134     return MAX2(reserve_size, _reserve_for_allocation_prefetch);
 135   }
 136   static size_t alignment_reserve()              { return align_object_size(end_reserve()); }
 137   static size_t alignment_reserve_in_bytes()     { return alignment_reserve() * HeapWordSize; }
 138 
 139   // Return tlab size or remaining space in eden such that the
 140   // space is large enough to hold obj_size and necessary fill space.
 141   // Otherwise return 0;
 142   inline size_t compute_size(size_t obj_size);
 143 
 144   // Compute the minimal needed tlab size for the given object size.
 145   static inline size_t compute_min_size(size_t obj_size);
 146 
 147   // Record slow allocation
 148   inline void record_slow_allocation(size_t obj_size);
 149 
 150   // Initialization at startup
 151   static void startup_initialization();
 152 
 153   // Make an in-use tlab parsable, optionally retiring and/or zapping it.
 154   void make_parsable(bool retire, bool zap = true);
 155 
 156   // Retire in-use tlab before allocation of a new tlab
 157   void clear_before_allocation();
 158 
 159   // Accumulate statistics across all tlabs before gc
 160   static void accumulate_statistics_before_gc();
 161 
 162   // Resize tlabs for all threads
 163   static void resize_all_tlabs();
 164 
 165   void fill(HeapWord* start, HeapWord* top, size_t new_size);
 166   void initialize();
 167 
 168   static size_t refill_waste_limit_increment()   { return TLABWasteIncrement; }
 169 
 170   template <typename T> void addresses_do(T f) {
 171     f(&_start);
 172     f(&_top);
 173     f(&_pf_top);
 174     f(&_end);
 175   }
 176 
 177   // Code generation support
 178   static ByteSize start_offset()                 { return byte_offset_of(ThreadLocalAllocBuffer, _start); }
 179   static ByteSize end_offset()                   { return byte_offset_of(ThreadLocalAllocBuffer, _end  ); }
 180   static ByteSize top_offset()                   { return byte_offset_of(ThreadLocalAllocBuffer, _top  ); }
 181   static ByteSize pf_top_offset()                { return byte_offset_of(ThreadLocalAllocBuffer, _pf_top  ); }
 182   static ByteSize size_offset()                  { return byte_offset_of(ThreadLocalAllocBuffer, _desired_size ); }
 183   static ByteSize refill_waste_limit_offset()    { return byte_offset_of(ThreadLocalAllocBuffer, _refill_waste_limit ); }
 184 
 185   static ByteSize number_of_refills_offset()     { return byte_offset_of(ThreadLocalAllocBuffer, _number_of_refills ); }
 186   static ByteSize fast_refill_waste_offset()     { return byte_offset_of(ThreadLocalAllocBuffer, _fast_refill_waste ); }
 187   static ByteSize slow_allocations_offset()      { return byte_offset_of(ThreadLocalAllocBuffer, _slow_allocations ); }
 188 
 189   void verify();
 190 };
 191 
 192 class GlobalTLABStats: public CHeapObj<mtThread> {
 193 private:
 194 
 195   // Accumulate perfdata in private variables because
 196   // PerfData should be write-only for security reasons
 197   // (see perfData.hpp)
 198   unsigned _allocating_threads;
 199   unsigned _total_refills;
 200   unsigned _max_refills;
 201   size_t   _total_allocation;
 202   size_t   _total_gc_waste;
 203   size_t   _max_gc_waste;
 204   size_t   _total_slow_refill_waste;
 205   size_t   _max_slow_refill_waste;
 206   size_t   _total_fast_refill_waste;
 207   size_t   _max_fast_refill_waste;
 208   unsigned _total_slow_allocations;
 209   unsigned _max_slow_allocations;
 210 
 211   PerfVariable* _perf_allocating_threads;
 212   PerfVariable* _perf_total_refills;
 213   PerfVariable* _perf_max_refills;
 214   PerfVariable* _perf_allocation;
 215   PerfVariable* _perf_gc_waste;
 216   PerfVariable* _perf_max_gc_waste;
 217   PerfVariable* _perf_slow_refill_waste;
 218   PerfVariable* _perf_max_slow_refill_waste;
 219   PerfVariable* _perf_fast_refill_waste;
 220   PerfVariable* _perf_max_fast_refill_waste;
 221   PerfVariable* _perf_slow_allocations;
 222   PerfVariable* _perf_max_slow_allocations;
 223 
 224   AdaptiveWeightedAverage _allocating_threads_avg;
 225 
 226 public:
 227   GlobalTLABStats();
 228 
 229   // Initialize all counters
 230   void initialize();
 231 
 232   // Write all perf counters to the perf_counters
 233   void publish();
 234 
 235   void print();
 236 
 237   // Accessors
 238   unsigned allocating_threads_avg() {
 239     return MAX2((unsigned)(_allocating_threads_avg.average() + 0.5), 1U);
 240   }
 241 
 242   size_t allocation() {
 243     return _total_allocation;
 244   }
 245 
 246   // Update methods
 247 
 248   void update_allocating_threads() {
 249     _allocating_threads++;
 250   }
 251   void update_number_of_refills(unsigned value) {
 252     _total_refills += value;
 253     _max_refills    = MAX2(_max_refills, value);
 254   }
 255   void update_allocation(size_t value) {
 256     _total_allocation += value;
 257   }
 258   void update_gc_waste(size_t value) {
 259     _total_gc_waste += value;
 260     _max_gc_waste    = MAX2(_max_gc_waste, value);
 261   }
 262   void update_fast_refill_waste(size_t value) {
 263     _total_fast_refill_waste += value;
 264     _max_fast_refill_waste    = MAX2(_max_fast_refill_waste, value);
 265   }
 266   void update_slow_refill_waste(size_t value) {
 267     _total_slow_refill_waste += value;
 268     _max_slow_refill_waste    = MAX2(_max_slow_refill_waste, value);
 269   }
 270   void update_slow_allocations(unsigned value) {
 271     _total_slow_allocations += value;
 272     _max_slow_allocations    = MAX2(_max_slow_allocations, value);
 273   }
 274 };
 275 
 276 #endif // SHARE_VM_GC_SHARED_THREADLOCALALLOCBUFFER_HPP