1 /*
   2  * Copyright (c) 1999, 2015, 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 //
  41 //            Heap sampling is performed via the end/actual_end fields.
  42 //            actual_end contains the real end of the tlab allocation,
  43 //            whereas end can be set to an arbitrary spot in the tlab to
  44 //            trip the return and sample the allocation.
  45 //            slow_path_end is used to track if a fast tlab refill occured
  46 //            between slowpath calls.
  47 class ThreadLocalAllocBuffer: public CHeapObj<mtThread> {
  48   friend class VMStructs;
  49   friend class JVMCIVMStructs;
  50 private:
  51   HeapWord* _start;                              // address of TLAB
  52   HeapWord* _top;                                // address after last allocation
  53   HeapWord* _pf_top;                             // allocation prefetch watermark
  54   HeapWord* _current_end;                        // allocation end (can be the sampling end point or _allocation_end)
  55   HeapWord* _allocation_end;                     // end for allocations (actual TLAB end, excluding alignment_reserve)
  56   HeapWord* _last_slow_path_end;                 // last address for slow_path_end (as opposed to _allocation_end)
  57 
  58   size_t    _desired_size;                       // desired size   (including alignment_reserve)
  59   size_t    _refill_waste_limit;                 // hold onto tlab if free() is larger than this
  60   size_t    _allocated_before_last_gc;           // total bytes allocated up until the last gc
  61   size_t    _bytes_until_sample;                 // bytes until sample.
  62 
  63   static size_t   _max_size;                          // maximum size of any TLAB
  64   static int      _reserve_for_allocation_prefetch;   // Reserve at the end of the TLAB
  65   static unsigned _target_refills;                    // expected number of refills between GCs
  66 
  67   unsigned  _number_of_refills;
  68   unsigned  _fast_refill_waste;
  69   unsigned  _slow_refill_waste;
  70   unsigned  _gc_waste;
  71   unsigned  _slow_allocations;
  72 
  73   AdaptiveWeightedAverage _allocation_fraction;  // fraction of eden allocated in tlabs
  74 
  75   void accumulate_statistics();
  76   void initialize_statistics();
  77 
  78   void set_start(HeapWord* start)                { _start = start; }
  79   void set_current_end(HeapWord* current_end)    { _current_end = current_end; }
  80   void set_allocation_end(HeapWord* ptr)         { _allocation_end = ptr; }
  81   void set_last_slow_path_end(HeapWord* ptr)     { _last_slow_path_end = ptr; }
  82   void set_top(HeapWord* top)                    { _top = top; }
  83   void set_pf_top(HeapWord* pf_top)              { _pf_top = pf_top; }
  84   void set_desired_size(size_t desired_size)     { _desired_size = desired_size; }
  85   void set_refill_waste_limit(size_t waste)      { _refill_waste_limit = waste;  }
  86   void set_bytes_until_sample(size_t bytes)      { _bytes_until_sample = bytes;  }
  87 
  88   size_t initial_refill_waste_limit()            { return desired_size() / TLABRefillWasteFraction; }
  89 
  90   static int    target_refills()                 { return _target_refills; }
  91   size_t initial_desired_size();
  92 
  93   size_t remaining();
  94 
  95   void set_sample_end();
  96   void update_end_pointers();
  97 
  98   // Make parsable and release it.
  99   void reset();
 100 
 101   // Resize based on amount of allocation, etc.
 102   void resize();
 103 
 104   void invariants() const { assert(top() >= start() && top() <= current_end(), "invalid tlab"); }
 105 
 106   void initialize(HeapWord* start, HeapWord* top, HeapWord* end);
 107 
 108   void print_stats(const char* tag);
 109 
 110   Thread* myThread();
 111 
 112   // statistics
 113 
 114   int number_of_refills() const { return _number_of_refills; }
 115   int fast_refill_waste() const { return _fast_refill_waste; }
 116   int slow_refill_waste() const { return _slow_refill_waste; }
 117   int gc_waste() const          { return _gc_waste; }
 118   int slow_allocations() const  { return _slow_allocations; }
 119 
 120   static GlobalTLABStats* _global_stats;
 121   static GlobalTLABStats* global_stats() { return _global_stats; }
 122 
 123 public:
 124   ThreadLocalAllocBuffer() : _allocation_fraction(TLABAllocationWeight), _allocated_before_last_gc(0) {
 125     // do nothing.  tlabs must be inited by initialize() calls
 126   }
 127 
 128   static size_t min_size()                       { return align_object_size(MinTLABSize / HeapWordSize) + alignment_reserve(); }
 129   static size_t max_size()                       { assert(_max_size != 0, "max_size not set up"); return _max_size; }
 130   static size_t max_size_in_bytes()              { return max_size() * BytesPerWord; }
 131   static void set_max_size(size_t max_size)      { _max_size = max_size; }
 132 
 133   HeapWord* start() const                        { return _start; }
 134   HeapWord* current_end() const                  { return _current_end; }
 135   HeapWord* top() const                          { return _top; }
 136   HeapWord* reserved_end();
 137   HeapWord* pf_top() const                       { return _pf_top; }
 138   size_t desired_size() const                    { return _desired_size; }
 139   size_t used() const                            { return pointer_delta(top(), start()); }
 140   size_t used_bytes() const                      { return pointer_delta(top(), start(), 1); }
 141   size_t free() const                            { return pointer_delta(current_end(), top()); }
 142   // Don't discard tlab if remaining space is larger than this.
 143   size_t refill_waste_limit() const              { return _refill_waste_limit; }
 144 
 145   // Allocate size HeapWords. The memory is NOT initialized to zero.
 146   inline HeapWord* allocate(size_t size);
 147 
 148   // Reserve space at the end of TLAB
 149   static size_t end_reserve() {
 150     int reserve_size = typeArrayOopDesc::header_size(T_INT);
 151     return MAX2(reserve_size, _reserve_for_allocation_prefetch);
 152   }
 153   static size_t alignment_reserve()              { return align_object_size(end_reserve()); }
 154   static size_t alignment_reserve_in_bytes()     { return alignment_reserve() * HeapWordSize; }
 155 
 156   // Return tlab size or remaining space in eden such that the
 157   // space is large enough to hold obj_size and necessary fill space.
 158   // Otherwise return 0;
 159   inline size_t compute_size(size_t obj_size);
 160 
 161   // Record slow allocation
 162   inline void record_slow_allocation(size_t obj_size);
 163 
 164   // Initialization at startup
 165   static void startup_initialization();
 166 
 167   // Make an in-use tlab parsable, optionally retiring and/or zapping it.
 168   void make_parsable(bool retire, bool zap = true);
 169 
 170   // Retire in-use tlab before allocation of a new tlab
 171   void clear_before_allocation();
 172 
 173   // Accumulate statistics across all tlabs before gc
 174   static void accumulate_statistics_before_gc();
 175 
 176   // Resize tlabs for all threads
 177   static void resize_all_tlabs();
 178 
 179   void fill(HeapWord* start, HeapWord* top, size_t new_size);
 180   void initialize();
 181 
 182   void pick_next_sample(size_t diff = 0);
 183   void set_back_allocation_end();
 184   void update_tlab_sample_point(size_t size_in_bytes);
 185   void handle_sample(Thread* thread, HeapWord* result, size_t size_in_bytes);
 186   bool should_sample() { return _bytes_until_sample == 0; }
 187 
 188   static size_t refill_waste_limit_increment()   { return TLABWasteIncrement; }
 189 
 190   // Code generation support
 191   static ByteSize start_offset()                 { return byte_offset_of(ThreadLocalAllocBuffer, _start); }
 192   static ByteSize current_end_offset()           { return byte_offset_of(ThreadLocalAllocBuffer, _current_end  ); }
 193   static ByteSize top_offset()                   { return byte_offset_of(ThreadLocalAllocBuffer, _top  ); }
 194   static ByteSize pf_top_offset()                { return byte_offset_of(ThreadLocalAllocBuffer, _pf_top  ); }
 195   static ByteSize size_offset()                  { return byte_offset_of(ThreadLocalAllocBuffer, _desired_size ); }
 196   static ByteSize refill_waste_limit_offset()    { return byte_offset_of(ThreadLocalAllocBuffer, _refill_waste_limit ); }
 197 
 198   static ByteSize number_of_refills_offset()     { return byte_offset_of(ThreadLocalAllocBuffer, _number_of_refills ); }
 199   static ByteSize fast_refill_waste_offset()     { return byte_offset_of(ThreadLocalAllocBuffer, _fast_refill_waste ); }
 200   static ByteSize slow_allocations_offset()      { return byte_offset_of(ThreadLocalAllocBuffer, _slow_allocations ); }
 201 
 202   void verify();
 203 };
 204 
 205 class GlobalTLABStats: public CHeapObj<mtThread> {
 206 private:
 207 
 208   // Accumulate perfdata in private variables because
 209   // PerfData should be write-only for security reasons
 210   // (see perfData.hpp)
 211   unsigned _allocating_threads;
 212   unsigned _total_refills;
 213   unsigned _max_refills;
 214   size_t   _total_allocation;
 215   size_t   _total_gc_waste;
 216   size_t   _max_gc_waste;
 217   size_t   _total_slow_refill_waste;
 218   size_t   _max_slow_refill_waste;
 219   size_t   _total_fast_refill_waste;
 220   size_t   _max_fast_refill_waste;
 221   unsigned _total_slow_allocations;
 222   unsigned _max_slow_allocations;
 223 
 224   PerfVariable* _perf_allocating_threads;
 225   PerfVariable* _perf_total_refills;
 226   PerfVariable* _perf_max_refills;
 227   PerfVariable* _perf_allocation;
 228   PerfVariable* _perf_gc_waste;
 229   PerfVariable* _perf_max_gc_waste;
 230   PerfVariable* _perf_slow_refill_waste;
 231   PerfVariable* _perf_max_slow_refill_waste;
 232   PerfVariable* _perf_fast_refill_waste;
 233   PerfVariable* _perf_max_fast_refill_waste;
 234   PerfVariable* _perf_slow_allocations;
 235   PerfVariable* _perf_max_slow_allocations;
 236 
 237   AdaptiveWeightedAverage _allocating_threads_avg;
 238 
 239 public:
 240   GlobalTLABStats();
 241 
 242   // Initialize all counters
 243   void initialize();
 244 
 245   // Write all perf counters to the perf_counters
 246   void publish();
 247 
 248   void print();
 249 
 250   // Accessors
 251   unsigned allocating_threads_avg() {
 252     return MAX2((unsigned)(_allocating_threads_avg.average() + 0.5), 1U);
 253   }
 254 
 255   size_t allocation() {
 256     return _total_allocation;
 257   }
 258 
 259   // Update methods
 260 
 261   void update_allocating_threads() {
 262     _allocating_threads++;
 263   }
 264   void update_number_of_refills(unsigned value) {
 265     _total_refills += value;
 266     _max_refills    = MAX2(_max_refills, value);
 267   }
 268   void update_allocation(size_t value) {
 269     _total_allocation += value;
 270   }
 271   void update_gc_waste(size_t value) {
 272     _total_gc_waste += value;
 273     _max_gc_waste    = MAX2(_max_gc_waste, value);
 274   }
 275   void update_fast_refill_waste(size_t value) {
 276     _total_fast_refill_waste += value;
 277     _max_fast_refill_waste    = MAX2(_max_fast_refill_waste, value);
 278   }
 279   void update_slow_refill_waste(size_t value) {
 280     _total_slow_refill_waste += value;
 281     _max_slow_refill_waste    = MAX2(_max_slow_refill_waste, value);
 282   }
 283   void update_slow_allocations(unsigned value) {
 284     _total_slow_allocations += value;
 285     _max_slow_allocations    = MAX2(_max_slow_allocations, value);
 286   }
 287 };
 288 
 289 #endif // SHARE_VM_GC_SHARED_THREADLOCALALLOCBUFFER_HPP