1 /*
   2  * Copyright (c) 2001, 2010, 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 "oops/oop.inline.hpp"
  27 #include "runtime/os.hpp"
  28 #include "utilities/debug.hpp"
  29 #include "utilities/stack.inline.hpp"
  30 #include "utilities/taskqueue.hpp"
  31 #ifdef TARGET_OS_FAMILY_linux
  32 # include "thread_linux.inline.hpp"
  33 #endif
  34 #ifdef TARGET_OS_FAMILY_solaris
  35 # include "thread_solaris.inline.hpp"
  36 #endif
  37 #ifdef TARGET_OS_FAMILY_windows
  38 # include "thread_windows.inline.hpp"
  39 #endif
  40 #ifdef TARGET_OS_FAMILY_bsd
  41 # include "thread_bsd.inline.hpp"
  42 #endif
  43 
  44 #ifdef TRACESPINNING
  45 uint ParallelTaskTerminator::_total_yields = 0;
  46 uint ParallelTaskTerminator::_total_spins = 0;
  47 uint ParallelTaskTerminator::_total_peeks = 0;
  48 #endif
  49 
  50 #if TASKQUEUE_STATS
  51 const char * const TaskQueueStats::_names[last_stat_id] = {
  52   "qpush", "qpop", "qpop-s", "qattempt", "qsteal", "opush", "omax"
  53 };
  54 
  55 TaskQueueStats & TaskQueueStats::operator +=(const TaskQueueStats & addend)
  56 {
  57   for (unsigned int i = 0; i < last_stat_id; ++i) {
  58     _stats[i] += addend._stats[i];
  59   }
  60   return *this;
  61 }
  62 
  63 void TaskQueueStats::print_header(unsigned int line, outputStream* const stream,
  64                                   unsigned int width)
  65 {
  66   // Use a width w: 1 <= w <= max_width
  67   const unsigned int max_width = 40;
  68   const unsigned int w = MAX2(MIN2(width, max_width), 1U);
  69 
  70   if (line == 0) { // spaces equal in width to the header
  71     const unsigned int hdr_width = w * last_stat_id + last_stat_id - 1;
  72     stream->print("%*s", hdr_width, " ");
  73   } else if (line == 1) { // labels
  74     stream->print("%*s", w, _names[0]);
  75     for (unsigned int i = 1; i < last_stat_id; ++i) {
  76       stream->print(" %*s", w, _names[i]);
  77     }
  78   } else if (line == 2) { // dashed lines
  79     char dashes[max_width + 1];
  80     memset(dashes, '-', w);
  81     dashes[w] = '\0';
  82     stream->print("%s", dashes);
  83     for (unsigned int i = 1; i < last_stat_id; ++i) {
  84       stream->print(" %s", dashes);
  85     }
  86   }
  87 }
  88 
  89 void TaskQueueStats::print(outputStream* stream, unsigned int width) const
  90 {
  91   #define FMT SIZE_FORMAT_W(*)
  92   stream->print(FMT, width, _stats[0]);
  93   for (unsigned int i = 1; i < last_stat_id; ++i) {
  94     stream->print(" " FMT, width, _stats[i]);
  95   }
  96   #undef FMT
  97 }
  98 
  99 #ifdef ASSERT
 100 // Invariants which should hold after a TaskQueue has been emptied and is
 101 // quiescent; they do not hold at arbitrary times.
 102 void TaskQueueStats::verify() const
 103 {
 104   assert(get(push) == get(pop) + get(steal),
 105          err_msg("push=" SIZE_FORMAT " pop=" SIZE_FORMAT " steal=" SIZE_FORMAT,
 106                  get(push), get(pop), get(steal)));
 107   assert(get(pop_slow) <= get(pop),
 108          err_msg("pop_slow=" SIZE_FORMAT " pop=" SIZE_FORMAT,
 109                  get(pop_slow), get(pop)));
 110   assert(get(steal) <= get(steal_attempt),
 111          err_msg("steal=" SIZE_FORMAT " steal_attempt=" SIZE_FORMAT,
 112                  get(steal), get(steal_attempt)));
 113   assert(get(overflow) == 0 || get(push) != 0,
 114          err_msg("overflow=" SIZE_FORMAT " push=" SIZE_FORMAT,
 115                  get(overflow), get(push)));
 116   assert(get(overflow_max_len) == 0 || get(overflow) != 0,
 117          err_msg("overflow_max_len=" SIZE_FORMAT " overflow=" SIZE_FORMAT,
 118                  get(overflow_max_len), get(overflow)));
 119 }
 120 #endif // ASSERT
 121 #endif // TASKQUEUE_STATS
 122 
 123 int TaskQueueSetSuper::randomParkAndMiller(int *seed0) {
 124   const int a =      16807;
 125   const int m = 2147483647;
 126   const int q =     127773;  /* m div a */
 127   const int r =       2836;  /* m mod a */
 128   assert(sizeof(int) == 4, "I think this relies on that");
 129   int seed = *seed0;
 130   int hi   = seed / q;
 131   int lo   = seed % q;
 132   int test = a * lo - r * hi;
 133   if (test > 0)
 134     seed = test;
 135   else
 136     seed = test + m;
 137   *seed0 = seed;
 138   return seed;
 139 }
 140 
 141 ParallelTaskTerminator::
 142 ParallelTaskTerminator(int n_threads, TaskQueueSetSuper* queue_set) :
 143   _n_threads(n_threads),
 144   _queue_set(queue_set),
 145   _offered_termination(0) {}
 146 
 147 bool ParallelTaskTerminator::peek_in_queue_set() {
 148   return _queue_set->peek();
 149 }
 150 
 151 void ParallelTaskTerminator::yield() {
 152   assert(_offered_termination <= _n_threads, "Invariant");
 153   os::yield();
 154 }
 155 
 156 void ParallelTaskTerminator::sleep(uint millis) {
 157   assert(_offered_termination <= _n_threads, "Invariant");
 158   os::sleep(Thread::current(), millis, false);
 159 }
 160 
 161 bool
 162 ParallelTaskTerminator::offer_termination(TerminatorTerminator* terminator) {
 163   assert(_n_threads > 0, "Initialization is incorrect");
 164   assert(_offered_termination < _n_threads, "Invariant");
 165   Atomic::inc(&_offered_termination);
 166 
 167   uint yield_count = 0;
 168   // Number of hard spin loops done since last yield
 169   uint hard_spin_count = 0;
 170   // Number of iterations in the hard spin loop.
 171   uint hard_spin_limit = WorkStealingHardSpins;
 172 
 173   // If WorkStealingSpinToYieldRatio is 0, no hard spinning is done.
 174   // If it is greater than 0, then start with a small number
 175   // of spins and increase number with each turn at spinning until
 176   // the count of hard spins exceeds WorkStealingSpinToYieldRatio.
 177   // Then do a yield() call and start spinning afresh.
 178   if (WorkStealingSpinToYieldRatio > 0) {
 179     hard_spin_limit = WorkStealingHardSpins >> WorkStealingSpinToYieldRatio;
 180     hard_spin_limit = MAX2(hard_spin_limit, 1U);
 181   }
 182   // Remember the initial spin limit.
 183   uint hard_spin_start = hard_spin_limit;
 184 
 185   // Loop waiting for all threads to offer termination or
 186   // more work.
 187   while (true) {
 188     assert(_offered_termination <= _n_threads, "Invariant");
 189     // Are all threads offering termination?
 190     if (_offered_termination == _n_threads) {
 191       return true;
 192     } else {
 193       // Look for more work.
 194       // Periodically sleep() instead of yield() to give threads
 195       // waiting on the cores the chance to grab this code
 196       if (yield_count <= WorkStealingYieldsBeforeSleep) {
 197         // Do a yield or hardspin.  For purposes of deciding whether
 198         // to sleep, count this as a yield.
 199         yield_count++;
 200 
 201         // Periodically call yield() instead spinning
 202         // After WorkStealingSpinToYieldRatio spins, do a yield() call
 203         // and reset the counts and starting limit.
 204         if (hard_spin_count > WorkStealingSpinToYieldRatio) {
 205           yield();
 206           hard_spin_count = 0;
 207           hard_spin_limit = hard_spin_start;
 208 #ifdef TRACESPINNING
 209           _total_yields++;
 210 #endif
 211         } else {
 212           // Hard spin this time
 213           // Increase the hard spinning period but only up to a limit.
 214           hard_spin_limit = MIN2(2*hard_spin_limit,
 215                                  (uint) WorkStealingHardSpins);
 216           for (uint j = 0; j < hard_spin_limit; j++) {
 217             SpinPause();
 218           }
 219           hard_spin_count++;
 220 #ifdef TRACESPINNING
 221           _total_spins++;
 222 #endif
 223         }
 224       } else {
 225         if (PrintGCDetails && Verbose) {
 226          gclog_or_tty->print_cr("ParallelTaskTerminator::offer_termination() "
 227            "thread %d sleeps after %d yields",
 228            Thread::current(), yield_count);
 229         }
 230         yield_count = 0;
 231         // A sleep will cause this processor to seek work on another processor's
 232         // runqueue, if it has nothing else to run (as opposed to the yield
 233         // which may only move the thread to the end of the this processor's
 234         // runqueue).
 235         sleep(WorkStealingSleepMillis);
 236       }
 237 
 238 #ifdef TRACESPINNING
 239       _total_peeks++;
 240 #endif
 241       if (peek_in_queue_set() ||
 242           (terminator != NULL && terminator->should_exit_termination())) {
 243         Atomic::dec(&_offered_termination);
 244         assert(_offered_termination < _n_threads, "Invariant");
 245         return false;
 246       }
 247     }
 248   }
 249 }
 250 
 251 #ifdef TRACESPINNING
 252 void ParallelTaskTerminator::print_termination_counts() {
 253   gclog_or_tty->print_cr("ParallelTaskTerminator Total yields: %lld  "
 254     "Total spins: %lld  Total peeks: %lld",
 255     total_yields(),
 256     total_spins(),
 257     total_peeks());
 258 }
 259 #endif
 260 
 261 void ParallelTaskTerminator::reset_for_reuse() {
 262   if (_offered_termination != 0) {
 263     assert(_offered_termination == _n_threads,
 264            "Terminator may still be in use");
 265     _offered_termination = 0;
 266   }
 267 }
 268 
 269 #ifdef ASSERT
 270 bool ObjArrayTask::is_valid() const {
 271   return _obj != NULL && _obj->is_objArray() && _index > 0 &&
 272     _index < objArrayOop(_obj)->length();
 273 }
 274 #endif // ASSERT
 275 
 276 void ParallelTaskTerminator::reset_for_reuse(int n_threads) {
 277   reset_for_reuse();
 278   _n_threads = n_threads;
 279 }