1 /*
   2  * Copyright (c) 2001, 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 #include "precompiled.hpp"
  26 #include "memory/allocation.hpp"
  27 #include "memory/allocation.inline.hpp"
  28 #include "runtime/os.hpp"
  29 #include "utilities/workgroup.hpp"
  30 
  31 // Definitions of WorkGang methods.
  32 
  33 AbstractWorkGang::AbstractWorkGang(const char* name,
  34                                    bool  are_GC_task_threads,
  35                                    bool  are_ConcurrentGC_threads) :
  36   _name(name),
  37   _are_GC_task_threads(are_GC_task_threads),
  38   _are_ConcurrentGC_threads(are_ConcurrentGC_threads) {
  39 
  40   assert(!(are_GC_task_threads && are_ConcurrentGC_threads),
  41          "They cannot both be STW GC and Concurrent threads" );
  42 
  43   // Other initialization.
  44   _monitor = new Monitor(/* priority */       Mutex::leaf,
  45                          /* name */           "WorkGroup monitor",
  46                          /* allow_vm_block */ are_GC_task_threads);
  47   assert(monitor() != NULL, "Failed to allocate monitor");
  48   _terminate = false;
  49   _task = NULL;
  50   _sequence_number = 0;
  51   _started_workers = 0;
  52   _finished_workers = 0;
  53 }
  54 
  55 WorkGang::WorkGang(const char* name,
  56                    uint        workers,
  57                    bool        are_GC_task_threads,
  58                    bool        are_ConcurrentGC_threads) :
  59   AbstractWorkGang(name, are_GC_task_threads, are_ConcurrentGC_threads) {
  60   _total_workers = workers;
  61 }
  62 
  63 GangWorker* WorkGang::allocate_worker(uint which) {
  64   GangWorker* new_worker = new GangWorker(this, which);
  65   return new_worker;
  66 }
  67 
  68 // The current implementation will exit if the allocation
  69 // of any worker fails.  Still, return a boolean so that
  70 // a future implementation can possibly do a partial
  71 // initialization of the workers and report such to the
  72 // caller.
  73 bool WorkGang::initialize_workers() {
  74 
  75   if (TraceWorkGang) {
  76     tty->print_cr("Constructing work gang %s with %d threads",
  77                   name(),
  78                   total_workers());
  79   }
  80   _gang_workers = NEW_C_HEAP_ARRAY(GangWorker*, total_workers(), mtInternal);
  81   if (gang_workers() == NULL) {
  82     vm_exit_out_of_memory(0, "Cannot create GangWorker array.");
  83     return false;
  84   }
  85   os::ThreadType worker_type;
  86   if (are_ConcurrentGC_threads()) {
  87     worker_type = os::cgc_thread;
  88   } else {
  89     worker_type = os::pgc_thread;
  90   }
  91   for (uint worker = 0; worker < total_workers(); worker += 1) {
  92     GangWorker* new_worker = allocate_worker(worker);
  93     assert(new_worker != NULL, "Failed to allocate GangWorker");
  94     _gang_workers[worker] = new_worker;
  95     if (new_worker == NULL || !os::create_thread(new_worker, worker_type)) {
  96       vm_exit_out_of_memory(0, "Cannot create worker GC thread. Out of system resources.");
  97       return false;
  98     }
  99     if (!DisableStartThread) {
 100       os::start_thread(new_worker);
 101     }
 102   }
 103   return true;
 104 }
 105 
 106 AbstractWorkGang::~AbstractWorkGang() {
 107   if (TraceWorkGang) {
 108     tty->print_cr("Destructing work gang %s", name());
 109   }
 110   stop();   // stop all the workers
 111   for (uint worker = 0; worker < total_workers(); worker += 1) {
 112     delete gang_worker(worker);
 113   }
 114   delete gang_workers();
 115   delete monitor();
 116 }
 117 
 118 GangWorker* AbstractWorkGang::gang_worker(uint i) const {
 119   // Array index bounds checking.
 120   GangWorker* result = NULL;
 121   assert(gang_workers() != NULL, "No workers for indexing");
 122   assert(((i >= 0) && (i < total_workers())), "Worker index out of bounds");
 123   result = _gang_workers[i];
 124   assert(result != NULL, "Indexing to null worker");
 125   return result;
 126 }
 127 
 128 void WorkGang::run_task(AbstractGangTask* task) {
 129   run_task(task, total_workers());
 130 }
 131 
 132 void WorkGang::run_task(AbstractGangTask* task, uint no_of_parallel_workers) {
 133   task->set_for_termination(no_of_parallel_workers);
 134 
 135   // This thread is executed by the VM thread which does not block
 136   // on ordinary MutexLocker's.
 137   MutexLockerEx ml(monitor(), Mutex::_no_safepoint_check_flag);
 138   if (TraceWorkGang) {
 139     tty->print_cr("Running work gang %s task %s", name(), task->name());
 140   }
 141   // Tell all the workers to run a task.
 142   assert(task != NULL, "Running a null task");
 143   // Initialize.
 144   _task = task;
 145   _sequence_number += 1;
 146   _started_workers = 0;
 147   _finished_workers = 0;
 148   // Tell the workers to get to work.
 149   monitor()->notify_all();
 150   // Wait for them to be finished
 151   while (finished_workers() < no_of_parallel_workers) {
 152     if (TraceWorkGang) {
 153       tty->print_cr("Waiting in work gang %s: %d/%d finished sequence %d",
 154                     name(), finished_workers(), no_of_parallel_workers,
 155                     _sequence_number);
 156     }
 157     monitor()->wait(/* no_safepoint_check */ true);
 158   }
 159   _task = NULL;
 160   if (TraceWorkGang) {
 161     tty->print_cr("\nFinished work gang %s: %d/%d sequence %d",
 162                   name(), finished_workers(), no_of_parallel_workers,
 163                   _sequence_number);
 164     Thread* me = Thread::current();
 165     tty->print_cr("  T: 0x%x  VM_thread: %d", me, me->is_VM_thread());
 166   }
 167 }
 168 
 169 void FlexibleWorkGang::run_task(AbstractGangTask* task) {
 170   // If active_workers() is passed, _finished_workers
 171   // must only be incremented for workers that find non_null
 172   // work (as opposed to all those that just check that the
 173   // task is not null).
 174   WorkGang::run_task(task, (uint) active_workers());
 175 }
 176 
 177 void AbstractWorkGang::stop() {
 178   // Tell all workers to terminate, then wait for them to become inactive.
 179   MutexLockerEx ml(monitor(), Mutex::_no_safepoint_check_flag);
 180   if (TraceWorkGang) {
 181     tty->print_cr("Stopping work gang %s task %s", name(), task()->name());
 182   }
 183   _task = NULL;
 184   _terminate = true;
 185   monitor()->notify_all();
 186   while (finished_workers() < active_workers()) {
 187     if (TraceWorkGang) {
 188       tty->print_cr("Waiting in work gang %s: %d/%d finished",
 189                     name(), finished_workers(), active_workers());
 190     }
 191     monitor()->wait(/* no_safepoint_check */ true);
 192   }
 193 }
 194 
 195 void AbstractWorkGang::internal_worker_poll(WorkData* data) const {
 196   assert(monitor()->owned_by_self(), "worker_poll is an internal method");
 197   assert(data != NULL, "worker data is null");
 198   data->set_terminate(terminate());
 199   data->set_task(task());
 200   data->set_sequence_number(sequence_number());
 201 }
 202 
 203 void AbstractWorkGang::internal_note_start() {
 204   assert(monitor()->owned_by_self(), "note_finish is an internal method");
 205   _started_workers += 1;
 206 }
 207 
 208 void AbstractWorkGang::internal_note_finish() {
 209   assert(monitor()->owned_by_self(), "note_finish is an internal method");
 210   _finished_workers += 1;
 211 }
 212 
 213 void AbstractWorkGang::print_worker_threads_on(outputStream* st) const {
 214   uint    num_thr = total_workers();
 215   for (uint i = 0; i < num_thr; i++) {
 216     gang_worker(i)->print_on(st);
 217     st->cr();
 218   }
 219 }
 220 
 221 void AbstractWorkGang::threads_do(ThreadClosure* tc) const {
 222   assert(tc != NULL, "Null ThreadClosure");
 223   uint num_thr = total_workers();
 224   for (uint i = 0; i < num_thr; i++) {
 225     tc->do_thread(gang_worker(i));
 226   }
 227 }
 228 
 229 // GangWorker methods.
 230 
 231 GangWorker::GangWorker(AbstractWorkGang* gang, uint id) {
 232   _gang = gang;
 233   set_id(id);
 234   set_name("Gang worker#%d (%s)", id, gang->name());
 235 }
 236 
 237 void GangWorker::run() {
 238   initialize();
 239   loop();
 240 }
 241 
 242 void GangWorker::initialize() {
 243   this->initialize_thread_local_storage();
 244   this->record_stack_base_and_size();
 245   assert(_gang != NULL, "No gang to run in");
 246   os::set_priority(this, NearMaxPriority);
 247   if (TraceWorkGang) {
 248     tty->print_cr("Running gang worker for gang %s id %d",
 249                   gang()->name(), id());
 250   }
 251   // The VM thread should not execute here because MutexLocker's are used
 252   // as (opposed to MutexLockerEx's).
 253   assert(!Thread::current()->is_VM_thread(), "VM thread should not be part"
 254          " of a work gang");
 255 }
 256 
 257 void GangWorker::loop() {
 258   int previous_sequence_number = 0;
 259   Monitor* gang_monitor = gang()->monitor();
 260   for ( ; /* !terminate() */; ) {
 261     WorkData data;
 262     int part;  // Initialized below.
 263     {
 264       // Grab the gang mutex.
 265       MutexLocker ml(gang_monitor);
 266       // Wait for something to do.
 267       // Polling outside the while { wait } avoids missed notifies
 268       // in the outer loop.
 269       gang()->internal_worker_poll(&data);
 270       if (TraceWorkGang) {
 271         tty->print("Polled outside for work in gang %s worker %d",
 272                    gang()->name(), id());
 273         tty->print("  terminate: %s",
 274                    data.terminate() ? "true" : "false");
 275         tty->print("  sequence: %d (prev: %d)",
 276                    data.sequence_number(), previous_sequence_number);
 277         if (data.task() != NULL) {
 278           tty->print("  task: %s", data.task()->name());
 279         } else {
 280           tty->print("  task: NULL");
 281         }
 282         tty->cr();
 283       }
 284       for ( ; /* break or return */; ) {
 285         // Terminate if requested.
 286         if (data.terminate()) {
 287           gang()->internal_note_finish();
 288           gang_monitor->notify_all();
 289           return;
 290         }
 291         // Check for new work.
 292         if ((data.task() != NULL) &&
 293             (data.sequence_number() != previous_sequence_number)) {
 294           if (gang()->needs_more_workers()) {
 295             gang()->internal_note_start();
 296             gang_monitor->notify_all();
 297             part = gang()->started_workers() - 1;
 298             break;
 299           }
 300         }
 301         // Nothing to do.
 302         gang_monitor->wait(/* no_safepoint_check */ true);
 303         gang()->internal_worker_poll(&data);
 304         if (TraceWorkGang) {
 305           tty->print("Polled inside for work in gang %s worker %d",
 306                      gang()->name(), id());
 307           tty->print("  terminate: %s",
 308                      data.terminate() ? "true" : "false");
 309           tty->print("  sequence: %d (prev: %d)",
 310                      data.sequence_number(), previous_sequence_number);
 311           if (data.task() != NULL) {
 312             tty->print("  task: %s", data.task()->name());
 313           } else {
 314             tty->print("  task: NULL");
 315           }
 316           tty->cr();
 317         }
 318       }
 319       // Drop gang mutex.
 320     }
 321     if (TraceWorkGang) {
 322       tty->print("Work for work gang %s id %d task %s part %d",
 323                  gang()->name(), id(), data.task()->name(), part);
 324     }
 325     assert(data.task() != NULL, "Got null task");
 326     data.task()->work(part);
 327     {
 328       if (TraceWorkGang) {
 329         tty->print("Finish for work gang %s id %d task %s part %d",
 330                    gang()->name(), id(), data.task()->name(), part);
 331       }
 332       // Grab the gang mutex.
 333       MutexLocker ml(gang_monitor);
 334       gang()->internal_note_finish();
 335       // Tell the gang you are done.
 336       gang_monitor->notify_all();
 337       // Drop the gang mutex.
 338     }
 339     previous_sequence_number = data.sequence_number();
 340   }
 341 }
 342 
 343 bool GangWorker::is_GC_task_thread() const {
 344   return gang()->are_GC_task_threads();
 345 }
 346 
 347 bool GangWorker::is_ConcurrentGC_thread() const {
 348   return gang()->are_ConcurrentGC_threads();
 349 }
 350 
 351 void GangWorker::print_on(outputStream* st) const {
 352   st->print("\"%s\" ", name());
 353   Thread::print_on(st);
 354   st->cr();
 355 }
 356 
 357 // Printing methods
 358 
 359 const char* AbstractWorkGang::name() const {
 360   return _name;
 361 }
 362 
 363 #ifndef PRODUCT
 364 
 365 const char* AbstractGangTask::name() const {
 366   return _name;
 367 }
 368 
 369 #endif /* PRODUCT */
 370 
 371 // FlexibleWorkGang
 372 
 373 
 374 // *** WorkGangBarrierSync
 375 
 376 WorkGangBarrierSync::WorkGangBarrierSync()
 377   : _monitor(Mutex::safepoint, "work gang barrier sync", true),
 378     _n_workers(0), _n_completed(0), _should_reset(false) {
 379 }
 380 
 381 WorkGangBarrierSync::WorkGangBarrierSync(uint n_workers, const char* name)
 382   : _monitor(Mutex::safepoint, name, true),
 383     _n_workers(n_workers), _n_completed(0), _should_reset(false) {
 384 }
 385 
 386 void WorkGangBarrierSync::set_n_workers(uint n_workers) {
 387   _n_workers   = n_workers;
 388   _n_completed = 0;
 389   _should_reset = false;
 390 }
 391 
 392 void WorkGangBarrierSync::enter() {
 393   MutexLockerEx x(monitor(), Mutex::_no_safepoint_check_flag);
 394   if (should_reset()) {
 395     // The should_reset() was set and we are the first worker to enter
 396     // the sync barrier. We will zero the n_completed() count which
 397     // effectively resets the barrier.
 398     zero_completed();
 399     set_should_reset(false);
 400   }
 401   inc_completed();
 402   if (n_completed() == n_workers()) {
 403     // At this point we would like to reset the barrier to be ready in
 404     // case it is used again. However, we cannot set n_completed() to
 405     // 0, even after the notify_all(), given that some other workers
 406     // might still be waiting for n_completed() to become ==
 407     // n_workers(). So, if we set n_completed() to 0, those workers
 408     // will get stuck (as they will wake up, see that n_completed() !=
 409     // n_workers() and go back to sleep). Instead, we raise the
 410     // should_reset() flag and the barrier will be reset the first
 411     // time a worker enters it again.
 412     set_should_reset(true);
 413     monitor()->notify_all();
 414   } else {
 415     while (n_completed() != n_workers()) {
 416       monitor()->wait(/* no_safepoint_check */ true);
 417     }
 418   }
 419 }
 420 
 421 // SubTasksDone functions.
 422 
 423 SubTasksDone::SubTasksDone(uint n) :
 424   _n_tasks(n), _n_threads(1), _tasks(NULL) {
 425   _tasks = NEW_C_HEAP_ARRAY(uint, n, mtInternal);
 426   guarantee(_tasks != NULL, "alloc failure");
 427   clear();
 428 }
 429 
 430 bool SubTasksDone::valid() {
 431   return _tasks != NULL;
 432 }
 433 
 434 void SubTasksDone::set_n_threads(uint t) {
 435   assert(_claimed == 0 || _threads_completed == _n_threads,
 436          "should not be called while tasks are being processed!");
 437   _n_threads = (t == 0 ? 1 : t);
 438 }
 439 
 440 void SubTasksDone::clear() {
 441   for (uint i = 0; i < _n_tasks; i++) {
 442     _tasks[i] = 0;
 443   }
 444   _threads_completed = 0;
 445 #ifdef ASSERT
 446   _claimed = 0;
 447 #endif
 448 }
 449 
 450 bool SubTasksDone::is_task_claimed(uint t) {
 451   assert(0 <= t && t < _n_tasks, "bad task id.");
 452   uint old = _tasks[t];
 453   if (old == 0) {
 454     old = Atomic::cmpxchg(1, &_tasks[t], 0);
 455   }
 456   assert(_tasks[t] == 1, "What else?");
 457   bool res = old != 0;
 458 #ifdef ASSERT
 459   if (!res) {
 460     assert(_claimed < _n_tasks, "Too many tasks claimed; missing clear?");
 461     Atomic::inc((volatile jint*) &_claimed);
 462   }
 463 #endif
 464   return res;
 465 }
 466 
 467 void SubTasksDone::all_tasks_completed() {
 468   jint observed = _threads_completed;
 469   jint old;
 470   do {
 471     old = observed;
 472     observed = Atomic::cmpxchg(old+1, &_threads_completed, old);
 473   } while (observed != old);
 474   // If this was the last thread checking in, clear the tasks.
 475   if (observed+1 == (jint)_n_threads) clear();
 476 }
 477 
 478 
 479 SubTasksDone::~SubTasksDone() {
 480   if (_tasks != NULL) FREE_C_HEAP_ARRAY(jint, _tasks, mtInternal);
 481 }
 482 
 483 // *** SequentialSubTasksDone
 484 
 485 void SequentialSubTasksDone::clear() {
 486   _n_tasks   = _n_claimed   = 0;
 487   _n_threads = _n_completed = 0;
 488 }
 489 
 490 bool SequentialSubTasksDone::valid() {
 491   return _n_threads > 0;
 492 }
 493 
 494 bool SequentialSubTasksDone::is_task_claimed(uint& t) {
 495   uint* n_claimed_ptr = &_n_claimed;
 496   t = *n_claimed_ptr;
 497   while (t < _n_tasks) {
 498     jint res = Atomic::cmpxchg(t+1, n_claimed_ptr, t);
 499     if (res == (jint)t) {
 500       return false;
 501     }
 502     t = *n_claimed_ptr;
 503   }
 504   return true;
 505 }
 506 
 507 bool SequentialSubTasksDone::all_tasks_completed() {
 508   uint* n_completed_ptr = &_n_completed;
 509   uint  complete        = *n_completed_ptr;
 510   while (true) {
 511     uint res = Atomic::cmpxchg(complete+1, n_completed_ptr, complete);
 512     if (res == complete) {
 513       break;
 514     }
 515     complete = res;
 516   }
 517   if (complete+1 == _n_threads) {
 518     clear();
 519     return true;
 520   }
 521   return false;
 522 }
 523 
 524 bool FreeIdSet::_stat_init = false;
 525 FreeIdSet* FreeIdSet::_sets[NSets];
 526 bool FreeIdSet::_safepoint;
 527 
 528 FreeIdSet::FreeIdSet(int sz, Monitor* mon) :
 529   _sz(sz), _mon(mon), _hd(0), _waiters(0), _index(-1), _claimed(0)
 530 {
 531   _ids = new int[sz];
 532   for (int i = 0; i < sz; i++) _ids[i] = i+1;
 533   _ids[sz-1] = end_of_list; // end of list.
 534   if (_stat_init) {
 535     for (int j = 0; j < NSets; j++) _sets[j] = NULL;
 536     _stat_init = true;
 537   }
 538   // Add to sets.  (This should happen while the system is still single-threaded.)
 539   for (int j = 0; j < NSets; j++) {
 540     if (_sets[j] == NULL) {
 541       _sets[j] = this;
 542       _index = j;
 543       break;
 544     }
 545   }
 546   guarantee(_index != -1, "Too many FreeIdSets in use!");
 547 }
 548 
 549 FreeIdSet::~FreeIdSet() {
 550   _sets[_index] = NULL;
 551 }
 552 
 553 void FreeIdSet::set_safepoint(bool b) {
 554   _safepoint = b;
 555   if (b) {
 556     for (int j = 0; j < NSets; j++) {
 557       if (_sets[j] != NULL && _sets[j]->_waiters > 0) {
 558         Monitor* mon = _sets[j]->_mon;
 559         mon->lock_without_safepoint_check();
 560         mon->notify_all();
 561         mon->unlock();
 562       }
 563     }
 564   }
 565 }
 566 
 567 #define FID_STATS 0
 568 
 569 int FreeIdSet::claim_par_id() {
 570 #if FID_STATS
 571   thread_t tslf = thr_self();
 572   tty->print("claim_par_id[%d]: sz = %d, claimed = %d\n", tslf, _sz, _claimed);
 573 #endif
 574   MutexLockerEx x(_mon, Mutex::_no_safepoint_check_flag);
 575   while (!_safepoint && _hd == end_of_list) {
 576     _waiters++;
 577 #if FID_STATS
 578     if (_waiters > 5) {
 579       tty->print("claim_par_id waiting[%d]: %d waiters, %d claimed.\n",
 580                  tslf, _waiters, _claimed);
 581     }
 582 #endif
 583     _mon->wait(Mutex::_no_safepoint_check_flag);
 584     _waiters--;
 585   }
 586   if (_hd == end_of_list) {
 587 #if FID_STATS
 588     tty->print("claim_par_id[%d]: returning EOL.\n", tslf);
 589 #endif
 590     return -1;
 591   } else {
 592     int res = _hd;
 593     _hd = _ids[res];
 594     _ids[res] = claimed;  // For debugging.
 595     _claimed++;
 596 #if FID_STATS
 597     tty->print("claim_par_id[%d]: returning %d, claimed = %d.\n",
 598                tslf, res, _claimed);
 599 #endif
 600     return res;
 601   }
 602 }
 603 
 604 bool FreeIdSet::claim_perm_id(int i) {
 605   assert(0 <= i && i < _sz, "Out of range.");
 606   MutexLockerEx x(_mon, Mutex::_no_safepoint_check_flag);
 607   int prev = end_of_list;
 608   int cur = _hd;
 609   while (cur != end_of_list) {
 610     if (cur == i) {
 611       if (prev == end_of_list) {
 612         _hd = _ids[cur];
 613       } else {
 614         _ids[prev] = _ids[cur];
 615       }
 616       _ids[cur] = claimed;
 617       _claimed++;
 618       return true;
 619     } else {
 620       prev = cur;
 621       cur = _ids[cur];
 622     }
 623   }
 624   return false;
 625 
 626 }
 627 
 628 void FreeIdSet::release_par_id(int id) {
 629   MutexLockerEx x(_mon, Mutex::_no_safepoint_check_flag);
 630   assert(_ids[id] == claimed, "Precondition.");
 631   _ids[id] = _hd;
 632   _hd = id;
 633   _claimed--;
 634 #if FID_STATS
 635   tty->print("[%d] release_par_id(%d), waiters =%d,  claimed = %d.\n",
 636              thr_self(), id, _waiters, _claimed);
 637 #endif
 638   if (_waiters > 0)
 639     // Notify all would be safer, but this is OK, right?
 640     _mon->notify_all();
 641 }