1 /*
   2  * Copyright (c) 2001, 2013, 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, OOM_MALLOC_ERROR, "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, OOM_MALLOC_ERROR,
  97               "Cannot create worker GC thread. Out of system resources.");
  98       return false;
  99     }
 100     if (!DisableStartThread) {
 101       os::start_thread(new_worker);
 102     }
 103   }
 104   return true;
 105 }
 106 
 107 AbstractWorkGang::~AbstractWorkGang() {
 108   if (TraceWorkGang) {
 109     tty->print_cr("Destructing work gang %s", name());
 110   }
 111   stop();   // stop all the workers
 112   for (uint worker = 0; worker < total_workers(); worker += 1) {
 113     delete gang_worker(worker);
 114   }
 115   delete gang_workers();
 116   delete monitor();
 117 }
 118 
 119 GangWorker* AbstractWorkGang::gang_worker(uint i) const {
 120   // Array index bounds checking.
 121   GangWorker* result = NULL;
 122   assert(gang_workers() != NULL, "No workers for indexing");
 123   assert(((i >= 0) && (i < total_workers())), "Worker index out of bounds");
 124   result = _gang_workers[i];
 125   assert(result != NULL, "Indexing to null worker");
 126   return result;
 127 }
 128 
 129 void WorkGang::run_task(AbstractGangTask* task) {
 130   run_task(task, total_workers());
 131 }
 132 
 133 void WorkGang::run_task(AbstractGangTask* task, uint no_of_parallel_workers) {
 134   task->set_for_termination(no_of_parallel_workers);
 135 
 136   // This thread is executed by the VM thread which does not block
 137   // on ordinary MutexLocker's.
 138   MutexLockerEx ml(monitor(), Mutex::_no_safepoint_check_flag);
 139   if (TraceWorkGang) {
 140     tty->print_cr("Running work gang %s task %s", name(), task->name());
 141   }
 142   // Tell all the workers to run a task.
 143   assert(task != NULL, "Running a null task");
 144   // Initialize.
 145   _task = task;
 146   _sequence_number += 1;
 147   _started_workers = 0;
 148   _finished_workers = 0;
 149   // Tell the workers to get to work.
 150   monitor()->notify_all();
 151   // Wait for them to be finished
 152   while (finished_workers() < no_of_parallel_workers) {
 153     if (TraceWorkGang) {
 154       tty->print_cr("Waiting in work gang %s: %d/%d finished sequence %d",
 155                     name(), finished_workers(), no_of_parallel_workers,
 156                     _sequence_number);
 157     }
 158     monitor()->wait(/* no_safepoint_check */ true);
 159   }
 160   _task = NULL;
 161   if (TraceWorkGang) {
 162     tty->print_cr("\nFinished work gang %s: %d/%d sequence %d",
 163                   name(), finished_workers(), no_of_parallel_workers,
 164                   _sequence_number);
 165     Thread* me = Thread::current();
 166     tty->print_cr("  T: 0x%x  VM_thread: %d", me, me->is_VM_thread());
 167   }
 168 }
 169 
 170 void FlexibleWorkGang::run_task(AbstractGangTask* task) {
 171   // If active_workers() is passed, _finished_workers
 172   // must only be incremented for workers that find non_null
 173   // work (as opposed to all those that just check that the
 174   // task is not null).
 175   WorkGang::run_task(task, (uint) active_workers());
 176 }
 177 
 178 void AbstractWorkGang::stop() {
 179   // Tell all workers to terminate, then wait for them to become inactive.
 180   MutexLockerEx ml(monitor(), Mutex::_no_safepoint_check_flag);
 181   if (TraceWorkGang) {
 182     tty->print_cr("Stopping work gang %s task %s", name(), task()->name());
 183   }
 184   _task = NULL;
 185   _terminate = true;
 186   monitor()->notify_all();
 187   while (finished_workers() < active_workers()) {
 188     if (TraceWorkGang) {
 189       tty->print_cr("Waiting in work gang %s: %d/%d finished",
 190                     name(), finished_workers(), active_workers());
 191     }
 192     monitor()->wait(/* no_safepoint_check */ true);
 193   }
 194 }
 195 
 196 void AbstractWorkGang::internal_worker_poll(WorkData* data) const {
 197   assert(monitor()->owned_by_self(), "worker_poll is an internal method");
 198   assert(data != NULL, "worker data is null");
 199   data->set_terminate(terminate());
 200   data->set_task(task());
 201   data->set_sequence_number(sequence_number());
 202 }
 203 
 204 void AbstractWorkGang::internal_note_start() {
 205   assert(monitor()->owned_by_self(), "note_finish is an internal method");
 206   _started_workers += 1;
 207 }
 208 
 209 void AbstractWorkGang::internal_note_finish() {
 210   assert(monitor()->owned_by_self(), "note_finish is an internal method");
 211   _finished_workers += 1;
 212 }
 213 
 214 void AbstractWorkGang::print_worker_threads_on(outputStream* st) const {
 215   uint    num_thr = total_workers();
 216   for (uint i = 0; i < num_thr; i++) {
 217     gang_worker(i)->print_on(st);
 218     st->cr();
 219   }
 220 }
 221 
 222 void AbstractWorkGang::threads_do(ThreadClosure* tc) const {
 223   assert(tc != NULL, "Null ThreadClosure");
 224   uint num_thr = total_workers();
 225   for (uint i = 0; i < num_thr; i++) {
 226     tc->do_thread(gang_worker(i));
 227   }
 228 }
 229 
 230 // GangWorker methods.
 231 
 232 GangWorker::GangWorker(AbstractWorkGang* gang, uint id) {
 233   _gang = gang;
 234   set_id(id);
 235   set_name("Gang worker#%d (%s)", id, gang->name());
 236 }
 237 
 238 void GangWorker::run() {
 239   initialize();
 240   loop();
 241 }
 242 
 243 void GangWorker::initialize() {
 244   this->initialize_thread_local_storage();
 245   this->record_stack_base_and_size();
 246   assert(_gang != NULL, "No gang to run in");
 247   os::set_priority(this, NearMaxPriority);
 248   if (TraceWorkGang) {
 249     tty->print_cr("Running gang worker for gang %s id %d",
 250                   gang()->name(), id());
 251   }
 252   // The VM thread should not execute here because MutexLocker's are used
 253   // as (opposed to MutexLockerEx's).
 254   assert(!Thread::current()->is_VM_thread(), "VM thread should not be part"
 255          " of a work gang");
 256 }
 257 
 258 void GangWorker::loop() {
 259   int previous_sequence_number = 0;
 260   Monitor* gang_monitor = gang()->monitor();
 261   for ( ; /* !terminate() */; ) {
 262     WorkData data;
 263     int part;  // Initialized below.
 264     {
 265       // Grab the gang mutex.
 266       MutexLocker ml(gang_monitor);
 267       // Wait for something to do.
 268       // Polling outside the while { wait } avoids missed notifies
 269       // in the outer loop.
 270       gang()->internal_worker_poll(&data);
 271       if (TraceWorkGang) {
 272         tty->print("Polled outside for work in gang %s worker %d",
 273                    gang()->name(), id());
 274         tty->print("  terminate: %s",
 275                    data.terminate() ? "true" : "false");
 276         tty->print("  sequence: %d (prev: %d)",
 277                    data.sequence_number(), previous_sequence_number);
 278         if (data.task() != NULL) {
 279           tty->print("  task: %s", data.task()->name());
 280         } else {
 281           tty->print("  task: NULL");
 282         }
 283         tty->cr();
 284       }
 285       for ( ; /* break or return */; ) {
 286         // Terminate if requested.
 287         if (data.terminate()) {
 288           gang()->internal_note_finish();
 289           gang_monitor->notify_all();
 290           return;
 291         }
 292         // Check for new work.
 293         if ((data.task() != NULL) &&
 294             (data.sequence_number() != previous_sequence_number)) {
 295           if (gang()->needs_more_workers()) {
 296             gang()->internal_note_start();
 297             gang_monitor->notify_all();
 298             part = gang()->started_workers() - 1;
 299             break;
 300           }
 301         }
 302         // Nothing to do.
 303         gang_monitor->wait(/* no_safepoint_check */ true);
 304         gang()->internal_worker_poll(&data);
 305         if (TraceWorkGang) {
 306           tty->print("Polled inside for work in gang %s worker %d",
 307                      gang()->name(), id());
 308           tty->print("  terminate: %s",
 309                      data.terminate() ? "true" : "false");
 310           tty->print("  sequence: %d (prev: %d)",
 311                      data.sequence_number(), previous_sequence_number);
 312           if (data.task() != NULL) {
 313             tty->print("  task: %s", data.task()->name());
 314           } else {
 315             tty->print("  task: NULL");
 316           }
 317           tty->cr();
 318         }
 319       }
 320       // Drop gang mutex.
 321     }
 322     if (TraceWorkGang) {
 323       tty->print("Work for work gang %s id %d task %s part %d",
 324                  gang()->name(), id(), data.task()->name(), part);
 325     }
 326     assert(data.task() != NULL, "Got null task");
 327     data.task()->work(part);
 328     {
 329       if (TraceWorkGang) {
 330         tty->print("Finish for work gang %s id %d task %s part %d",
 331                    gang()->name(), id(), data.task()->name(), part);
 332       }
 333       // Grab the gang mutex.
 334       MutexLocker ml(gang_monitor);
 335       gang()->internal_note_finish();
 336       // Tell the gang you are done.
 337       gang_monitor->notify_all();
 338       // Drop the gang mutex.
 339     }
 340     previous_sequence_number = data.sequence_number();
 341   }
 342 }
 343 
 344 bool GangWorker::is_GC_task_thread() const {
 345   return gang()->are_GC_task_threads();
 346 }
 347 
 348 bool GangWorker::is_ConcurrentGC_thread() const {
 349   return gang()->are_ConcurrentGC_threads();
 350 }
 351 
 352 void GangWorker::print_on(outputStream* st) const {
 353   st->print("\"%s\" ", name());
 354   Thread::print_on(st);
 355   st->cr();
 356 }
 357 
 358 // Printing methods
 359 
 360 const char* AbstractWorkGang::name() const {
 361   return _name;
 362 }
 363 
 364 #ifndef PRODUCT
 365 
 366 const char* AbstractGangTask::name() const {
 367   return _name;
 368 }
 369 
 370 #endif /* PRODUCT */
 371 
 372 // FlexibleWorkGang
 373 
 374 
 375 // *** WorkGangBarrierSync
 376 
 377 WorkGangBarrierSync::WorkGangBarrierSync()
 378   : _monitor(Mutex::safepoint, "work gang barrier sync", true),
 379     _n_workers(0), _n_completed(0), _should_reset(false), _aborted(false) {
 380 }
 381 
 382 WorkGangBarrierSync::WorkGangBarrierSync(uint n_workers, const char* name)
 383   : _monitor(Mutex::safepoint, name, true),
 384     _n_workers(n_workers), _n_completed(0), _should_reset(false), _aborted(false) {
 385 }
 386 
 387 void WorkGangBarrierSync::set_n_workers(uint n_workers) {
 388   _n_workers    = n_workers;
 389   _n_completed  = 0;
 390   _should_reset = false;
 391   _aborted      = false;
 392 }
 393 
 394 bool WorkGangBarrierSync::enter() {
 395   MutexLockerEx x(monitor(), Mutex::_no_safepoint_check_flag);
 396   if (should_reset()) {
 397     // The should_reset() was set and we are the first worker to enter
 398     // the sync barrier. We will zero the n_completed() count which
 399     // effectively resets the barrier.
 400     zero_completed();
 401     set_should_reset(false);
 402   }
 403   inc_completed();
 404   if (n_completed() == n_workers()) {
 405     // At this point we would like to reset the barrier to be ready in
 406     // case it is used again. However, we cannot set n_completed() to
 407     // 0, even after the notify_all(), given that some other workers
 408     // might still be waiting for n_completed() to become ==
 409     // n_workers(). So, if we set n_completed() to 0, those workers
 410     // will get stuck (as they will wake up, see that n_completed() !=
 411     // n_workers() and go back to sleep). Instead, we raise the
 412     // should_reset() flag and the barrier will be reset the first
 413     // time a worker enters it again.
 414     set_should_reset(true);
 415     monitor()->notify_all();
 416   } else {
 417     while (n_completed() != n_workers() && !aborted()) {
 418       monitor()->wait(/* no_safepoint_check */ true);
 419     }
 420   }
 421   return !aborted();
 422 }
 423 
 424 void WorkGangBarrierSync::abort() {
 425   MutexLockerEx x(monitor(), Mutex::_no_safepoint_check_flag);
 426   set_aborted();
 427   monitor()->notify_all();
 428 }
 429 
 430 // SubTasksDone functions.
 431 
 432 SubTasksDone::SubTasksDone(uint n) :
 433   _n_tasks(n), _n_threads(1), _tasks(NULL) {
 434   _tasks = NEW_C_HEAP_ARRAY(uint, n, mtInternal);
 435   guarantee(_tasks != NULL, "alloc failure");
 436   clear();
 437 }
 438 
 439 bool SubTasksDone::valid() {
 440   return _tasks != NULL;
 441 }
 442 
 443 void SubTasksDone::set_n_threads(uint t) {
 444   assert(_claimed == 0 || _threads_completed == _n_threads,
 445          "should not be called while tasks are being processed!");
 446   _n_threads = (t == 0 ? 1 : t);
 447 }
 448 
 449 void SubTasksDone::clear() {
 450   for (uint i = 0; i < _n_tasks; i++) {
 451     _tasks[i] = 0;
 452   }
 453   _threads_completed = 0;
 454 #ifdef ASSERT
 455   _claimed = 0;
 456 #endif
 457 }
 458 
 459 bool SubTasksDone::is_task_claimed(uint t) {
 460   assert(0 <= t && t < _n_tasks, "bad task id.");
 461   uint old = _tasks[t];
 462   if (old == 0) {
 463     old = Atomic::cmpxchg(1, &_tasks[t], 0);
 464   }
 465   assert(_tasks[t] == 1, "What else?");
 466   bool res = old != 0;
 467 #ifdef ASSERT
 468   if (!res) {
 469     assert(_claimed < _n_tasks, "Too many tasks claimed; missing clear?");
 470     Atomic::inc((volatile jint*) &_claimed);
 471   }
 472 #endif
 473   return res;
 474 }
 475 
 476 void SubTasksDone::all_tasks_completed() {
 477   jint observed = _threads_completed;
 478   jint old;
 479   do {
 480     old = observed;
 481     observed = Atomic::cmpxchg(old+1, &_threads_completed, old);
 482   } while (observed != old);
 483   // If this was the last thread checking in, clear the tasks.
 484   if (observed+1 == (jint)_n_threads) clear();
 485 }
 486 
 487 
 488 SubTasksDone::~SubTasksDone() {
 489   if (_tasks != NULL) FREE_C_HEAP_ARRAY(jint, _tasks, mtInternal);
 490 }
 491 
 492 // *** SequentialSubTasksDone
 493 
 494 void SequentialSubTasksDone::clear() {
 495   _n_tasks   = _n_claimed   = 0;
 496   _n_threads = _n_completed = 0;
 497 }
 498 
 499 bool SequentialSubTasksDone::valid() {
 500   return _n_threads > 0;
 501 }
 502 
 503 bool SequentialSubTasksDone::is_task_claimed(uint& t) {
 504   uint* n_claimed_ptr = &_n_claimed;
 505   t = *n_claimed_ptr;
 506   while (t < _n_tasks) {
 507     jint res = Atomic::cmpxchg(t+1, n_claimed_ptr, t);
 508     if (res == (jint)t) {
 509       return false;
 510     }
 511     t = *n_claimed_ptr;
 512   }
 513   return true;
 514 }
 515 
 516 bool SequentialSubTasksDone::all_tasks_completed() {
 517   uint* n_completed_ptr = &_n_completed;
 518   uint  complete        = *n_completed_ptr;
 519   while (true) {
 520     uint res = Atomic::cmpxchg(complete+1, n_completed_ptr, complete);
 521     if (res == complete) {
 522       break;
 523     }
 524     complete = res;
 525   }
 526   if (complete+1 == _n_threads) {
 527     clear();
 528     return true;
 529   }
 530   return false;
 531 }
 532 
 533 bool FreeIdSet::_stat_init = false;
 534 FreeIdSet* FreeIdSet::_sets[NSets];
 535 bool FreeIdSet::_safepoint;
 536 
 537 FreeIdSet::FreeIdSet(int sz, Monitor* mon) :
 538   _sz(sz), _mon(mon), _hd(0), _waiters(0), _index(-1), _claimed(0)
 539 {
 540   _ids = NEW_C_HEAP_ARRAY(int, sz, mtInternal);
 541   for (int i = 0; i < sz; i++) _ids[i] = i+1;
 542   _ids[sz-1] = end_of_list; // end of list.
 543   if (_stat_init) {
 544     for (int j = 0; j < NSets; j++) _sets[j] = NULL;
 545     _stat_init = true;
 546   }
 547   // Add to sets.  (This should happen while the system is still single-threaded.)
 548   for (int j = 0; j < NSets; j++) {
 549     if (_sets[j] == NULL) {
 550       _sets[j] = this;
 551       _index = j;
 552       break;
 553     }
 554   }
 555   guarantee(_index != -1, "Too many FreeIdSets in use!");
 556 }
 557 
 558 FreeIdSet::~FreeIdSet() {
 559   _sets[_index] = NULL;
 560   FREE_C_HEAP_ARRAY(int, _ids, mtInternal);
 561 }
 562 
 563 void FreeIdSet::set_safepoint(bool b) {
 564   _safepoint = b;
 565   if (b) {
 566     for (int j = 0; j < NSets; j++) {
 567       if (_sets[j] != NULL && _sets[j]->_waiters > 0) {
 568         Monitor* mon = _sets[j]->_mon;
 569         mon->lock_without_safepoint_check();
 570         mon->notify_all();
 571         mon->unlock();
 572       }
 573     }
 574   }
 575 }
 576 
 577 #define FID_STATS 0
 578 
 579 int FreeIdSet::claim_par_id() {
 580 #if FID_STATS
 581   thread_t tslf = thr_self();
 582   tty->print("claim_par_id[%d]: sz = %d, claimed = %d\n", tslf, _sz, _claimed);
 583 #endif
 584   MutexLockerEx x(_mon, Mutex::_no_safepoint_check_flag);
 585   while (!_safepoint && _hd == end_of_list) {
 586     _waiters++;
 587 #if FID_STATS
 588     if (_waiters > 5) {
 589       tty->print("claim_par_id waiting[%d]: %d waiters, %d claimed.\n",
 590                  tslf, _waiters, _claimed);
 591     }
 592 #endif
 593     _mon->wait(Mutex::_no_safepoint_check_flag);
 594     _waiters--;
 595   }
 596   if (_hd == end_of_list) {
 597 #if FID_STATS
 598     tty->print("claim_par_id[%d]: returning EOL.\n", tslf);
 599 #endif
 600     return -1;
 601   } else {
 602     int res = _hd;
 603     _hd = _ids[res];
 604     _ids[res] = claimed;  // For debugging.
 605     _claimed++;
 606 #if FID_STATS
 607     tty->print("claim_par_id[%d]: returning %d, claimed = %d.\n",
 608                tslf, res, _claimed);
 609 #endif
 610     return res;
 611   }
 612 }
 613 
 614 bool FreeIdSet::claim_perm_id(int i) {
 615   assert(0 <= i && i < _sz, "Out of range.");
 616   MutexLockerEx x(_mon, Mutex::_no_safepoint_check_flag);
 617   int prev = end_of_list;
 618   int cur = _hd;
 619   while (cur != end_of_list) {
 620     if (cur == i) {
 621       if (prev == end_of_list) {
 622         _hd = _ids[cur];
 623       } else {
 624         _ids[prev] = _ids[cur];
 625       }
 626       _ids[cur] = claimed;
 627       _claimed++;
 628       return true;
 629     } else {
 630       prev = cur;
 631       cur = _ids[cur];
 632     }
 633   }
 634   return false;
 635 
 636 }
 637 
 638 void FreeIdSet::release_par_id(int id) {
 639   MutexLockerEx x(_mon, Mutex::_no_safepoint_check_flag);
 640   assert(_ids[id] == claimed, "Precondition.");
 641   _ids[id] = _hd;
 642   _hd = id;
 643   _claimed--;
 644 #if FID_STATS
 645   tty->print("[%d] release_par_id(%d), waiters =%d,  claimed = %d.\n",
 646              thr_self(), id, _waiters, _claimed);
 647 #endif
 648   if (_waiters > 0)
 649     // Notify all would be safer, but this is OK, right?
 650     _mon->notify_all();
 651 }