src/simix/smx_process.c

   1 /* Copyright (c) 2007, 2008, 2009, 2010. The SimGrid Team.
   2  * All rights reserved.                                                     */
   3
   4 /* This program is free software; you can redistribute it and/or modify it
   5  * under the terms of the license (GNU LGPL) which comes with this package. */
   6
   7 #include "smx_private.h"
   8 #include "xbt/sysdep.h"
   9 #include "xbt/log.h"
  10 #include "xbt/dict.h"
  11 #include "mc/mc.h"
  12
  13 XBT_LOG_NEW_DEFAULT_SUBCATEGORY(simix_process, simix,
  14                                 "Logging specific to SIMIX (process)");
  15
  16 unsigned long simix_process_maxpid = 0;
  17
  18 /**
  19  * \brief Returns the current agent.
  20  *
  21  * This functions returns the currently running SIMIX process.
  22  *
  23  * \return The SIMIX process
  24  */
  25 XBT_INLINE smx_process_t SIMIX_process_self(void)
  26 {
  27   smx_context_t self_context = SIMIX_context_self();
  28
  29   return self_context ? SIMIX_context_get_data(self_context) : NULL;
  30 }
  31
  32 /**
  33  * \brief Returns whether a process has pending asynchronous communications.
  34  * \return true if there are asynchronous communications in this process
  35  */
  36 int SIMIX_process_has_pending_comms(smx_process_t process) {
  37
  38   return xbt_fifo_size(process->comms) > 0;
  39 }
  40
  41 /**
  42  * \brief Moves a process to the list of processes to destroy.
  43  */
  44 void SIMIX_process_cleanup(smx_process_t process)
  45 {
  46   XBT_DEBUG("Cleanup process %s (%p), waiting action %p",
  47       process->name, process, process->waiting_action);
  48
  49   /* cancel non-blocking communications */
  50   smx_action_t action;
  51   while ((action = xbt_fifo_pop(process->comms))) {
  52
  53     /* make sure no one will finish the comm after this process is destroyed,
  54      * because src_proc or dst_proc would be an invalid pointer */
  55     SIMIX_comm_cancel(action);
  56
  57     if (action->comm.src_proc == process) {
  58       XBT_DEBUG("Found an unfinished send comm %p (detached = %d), state %d, src = %p, dst = %p",
  59           action, action->comm.detached, action->state, action->comm.src_proc, action->comm.dst_proc);
  60       action->comm.src_proc = NULL;
  61
  62       if (action->comm.detached) {
  63          if (action->comm.refcount == 0) {
  64            XBT_DEBUG("Increase the refcount before destroying it since it's detached");
  65            /* I'm not supposed to destroy a detached comm from the sender side,
  66             * unless there is no receiver matching the rdv */
  67            action->comm.refcount++;
  68            SIMIX_comm_destroy(action);
  69          }
  70          else {
  71            XBT_DEBUG("Don't destroy it since its refcount is %d", action->comm.refcount);
  72          }
  73       } else {
  74         SIMIX_comm_destroy(action);
  75       }
  76     }
  77     else if (action->comm.dst_proc == process){
  78       XBT_DEBUG("Found an unfinished recv comm %p, state %d, src = %p, dst = %p",
  79           action, action->state, action->comm.src_proc, action->comm.dst_proc);
  80       action->comm.dst_proc = NULL;
  81
  82       if (action->comm.detached && action->comm.refcount == 1
  83           && action->comm.src_proc != NULL) {
  84         /* the comm will be freed right now, remove it from the sender */
  85         xbt_fifo_remove(action->comm.src_proc->comms, action);
  86       }
  87       SIMIX_comm_destroy(action);
  88     }
  89     else {
  90       xbt_die("Communication action %p is in my list but I'm not the sender "
  91           "or the receiver", action);
  92     }
  93   }
  94
  95   /*xbt_swag_remove(process, simix_global->process_to_run);*/
  96   xbt_swag_remove(process, simix_global->process_list);
  97   xbt_swag_remove(process, process->smx_host->process_list);
  98   xbt_swag_insert(process, simix_global->process_to_destroy);
  99 }
 100
 101 /**
 102  * Garbage collection
 103  *
 104  * Should be called some time to time to free the memory allocated for processes
 105  * that have finished (or killed).
 106  */
 107 void SIMIX_process_empty_trash(void)
 108 {
 109   smx_process_t process = NULL;
 110
 111   while ((process = xbt_swag_extract(simix_global->process_to_destroy))) {
 112     SIMIX_context_free(process->context);
 113
 114     /* Free the exception allocated at creation time */
 115     free(process->running_ctx);
 116     xbt_dict_free(&process->properties);
 117
 118     xbt_fifo_free(process->comms);
 119
 120     free(process->name);
 121     free(process);
 122   }
 123 }
 124
 125 /**
 126  * \brief Creates and runs the maestro process
 127  */
 128 void SIMIX_create_maestro_process()
 129 {
 130   smx_process_t maestro = NULL;
 131
 132   /* Create maestro process and intilialize it */
 133   maestro = xbt_new0(s_smx_process_t, 1);
 134   maestro->pid = simix_process_maxpid++;
 135   maestro->name = (char *) "";
 136   maestro->running_ctx = xbt_new(xbt_running_ctx_t, 1);
 137   XBT_RUNNING_CTX_INITIALIZE(maestro->running_ctx);
 138   maestro->context = SIMIX_context_new(NULL, 0, NULL, NULL, maestro);
 139   maestro->simcall.issuer = maestro;
 140
 141   simix_global->maestro_process = maestro;
 142   return;
 143 }
 144
 145 /**
 146  * \brief Same as SIMIX_process_create() but with only one argument (used by timers).
 147  * \return the process created
 148  */
 149 smx_process_t SIMIX_process_create_from_wrapper(smx_process_arg_t args) {
 150
 151   smx_process_t process;
 152   simix_global->create_process_function(
 153       &process,
 154       args->name,
 155       args->code,
 156       args->data,
 157       args->hostname,
 158       args->argc,
 159       args->argv,
 160       args->properties);
 161
 162   return process;
 163 }
 164
 165 /**
 166  * \brief Internal function to create a process.
 167  *
 168  * This function actually creates the process.
 169  * It may be called when a SIMCALL_PROCESS_CREATE simcall occurs,
 170  * or directly for SIMIX internal purposes.
 171  *
 172  * \return the process created
 173  */
 174 void SIMIX_process_create(smx_process_t *process,
 175                           const char *name,
 176                           xbt_main_func_t code,
 177                           void *data,
 178                           const char *hostname,
 179                           int argc, char **argv,
 180                           xbt_dict_t properties) {
 181
 182   *process = NULL;
 183   smx_host_t host = SIMIX_host_get_by_name(hostname);
 184
 185   XBT_DEBUG("Start process %s on host %s", name, hostname);
 186
 187   if (!SIMIX_host_get_state(host)) {
 188     XBT_WARN("Cannot launch process '%s' on failed host '%s'", name,
 189           hostname);
 190   }
 191   else {
 192     *process = xbt_new0(s_smx_process_t, 1);
 193
 194     xbt_assert(((code != NULL) && (host != NULL)), "Invalid parameters");
 195
 196     /* Process data */
 197     (*process)->pid = simix_process_maxpid++;
 198     (*process)->name = xbt_strdup(name);
 199     (*process)->smx_host = host;
 200     (*process)->data = data;
 201     (*process)->comms = xbt_fifo_new();
 202     (*process)->simcall.issuer = *process;
 203
 204     XBT_VERB("Create context %s", (*process)->name);
 205     (*process)->context = SIMIX_context_new(code, argc, argv,
 206         simix_global->cleanup_process_function, *process);
 207
 208     (*process)->running_ctx = xbt_new(xbt_running_ctx_t, 1);
 209     XBT_RUNNING_CTX_INITIALIZE((*process)->running_ctx);
 210
 211     /* Add properties */
 212     (*process)->properties = properties;
 213
 214     /* Add the process to it's host process list */
 215     xbt_swag_insert(*process, host->process_list);
 216
 217     XBT_DEBUG("Start context '%s'", (*process)->name);
 218
 219     /* Now insert it in the global process list and in the process to run list */
 220     xbt_swag_insert(*process, simix_global->process_list);
 221     XBT_DEBUG("Inserting %s(%s) in the to_run list", (*process)->name, host->name);
 222     xbt_dynar_push_as(simix_global->process_to_run, smx_process_t, *process);
 223   }
 224 }
 225
 226 /**
 227  * \brief Executes the processes from simix_global->process_to_run.
 228  *
 229  * The processes of simix_global->process_to_run are run (in parallel if
 230  * possible).  On exit, simix_global->process_to_run is empty, and
 231  * simix_global->process_that_ran contains the list of processes that just ran.
 232  * The two lists are swapped so, be careful when using them before and after a
 233  * call to this function.
 234  */
 235 void SIMIX_process_runall(void)
 236 {
 237   SIMIX_context_runall();
 238
 239   xbt_dynar_t tmp = simix_global->process_that_ran;
 240   simix_global->process_that_ran = simix_global->process_to_run;
 241   simix_global->process_to_run = tmp;
 242   xbt_dynar_reset(simix_global->process_to_run);
 243 }
 244
 245 /**
 246  * \brief Internal function to kill a SIMIX process.
 247  *
 248  * This function may be called when a SIMCALL_PROCESS_KILL simcall occurs,
 249  * or directly for SIMIX internal purposes.
 250  *
 251  * \param process poor victim
 252  */
 253 void SIMIX_process_kill(smx_process_t process) {
 254
 255   XBT_DEBUG("Killing process %s on %s", process->name, process->smx_host->name);
 256
 257   process->context->iwannadie = 1;
 258   process->blocked = 0;
 259   process->suspended = 0;
 260   /* FIXME: set doexception to 0 also? */
 261
 262   /* destroy the blocking action if any */
 263   if (process->waiting_action) {
 264
 265     switch (process->waiting_action->type) {
 266
 267       case SIMIX_ACTION_EXECUTE:
 268       case SIMIX_ACTION_PARALLEL_EXECUTE:
 269         SIMIX_host_execution_destroy(process->waiting_action);
 270         break;
 271
 272       case SIMIX_ACTION_COMMUNICATE:
 273         xbt_fifo_remove(process->comms, process->waiting_action);
 274         SIMIX_comm_destroy(process->waiting_action);
 275         break;
 276
 277       case SIMIX_ACTION_SLEEP:
 278         SIMIX_process_sleep_destroy(process->waiting_action);
 279         break;
 280
 281       case SIMIX_ACTION_SYNCHRO:
 282         SIMIX_synchro_stop_waiting(process, &process->simcall);
 283         SIMIX_synchro_destroy(process->waiting_action);
 284         break;
 285
 286       case SIMIX_ACTION_IO:
 287         SIMIX_io_destroy(process->waiting_action);
 288         break;
 289     }
 290   }
 291
 292   xbt_dynar_push_as(simix_global->process_to_run, smx_process_t, process);
 293 }
 294
 295 /**
 296  * \brief Kills all running processes.
 297  * \param issuer this one will not be killed
 298  */
 299 void SIMIX_process_killall(smx_process_t issuer)
 300 {
 301   smx_process_t p = NULL;
 302
 303   while ((p = xbt_swag_extract(simix_global->process_list))) {
 304     if (p != issuer) {
 305       SIMIX_process_kill(p);
 306     }
 307   }
 308
 309   SIMIX_context_runall(simix_global->process_to_run);
 310
 311   SIMIX_process_empty_trash();
 312 }
 313
 314 void SIMIX_process_change_host(smx_process_t process,
 315                                smx_host_t dest)
 316 {
 317   xbt_assert((process != NULL), "Invalid parameters");
 318   xbt_swag_remove(process, process->smx_host->process_list);
 319   process->smx_host = dest;
 320   xbt_swag_insert(process, dest->process_list);
 321 }
 322
 323 void SIMIX_pre_process_change_host(smx_process_t process, smx_host_t dest)
 324 {
 325   process->new_host = dest;
 326 }
 327
 328 void SIMIX_pre_process_suspend(smx_simcall_t simcall)
 329 {
 330   smx_process_t process = simcall->process_suspend.process;
 331   SIMIX_process_suspend(process, simcall->issuer);
 332
 333   if (process != simcall->issuer) {
 334     SIMIX_simcall_answer(simcall);
 335   }
 336   /* If we are suspending ourselves, then just do not finish the simcall now */
 337 }
 338
 339 void SIMIX_process_suspend(smx_process_t process, smx_process_t issuer)
 340 {
 341   xbt_assert((process != NULL), "Invalid parameters");
 342
 343   if (process->suspended) {
 344     XBT_DEBUG("Process '%s' is already suspended", process->name);
 345     return;
 346   }
 347
 348   process->suspended = 1;
 349
 350   /* If we are suspending another process, and it is waiting on an action,
 351      suspend its action. */
 352   if (process != issuer) {
 353
 354     if (process->waiting_action) {
 355
 356       switch (process->waiting_action->type) {
 357
 358         case SIMIX_ACTION_EXECUTE:
 359         case SIMIX_ACTION_PARALLEL_EXECUTE:
 360           SIMIX_host_execution_suspend(process->waiting_action);
 361           break;
 362
 363         case SIMIX_ACTION_COMMUNICATE:
 364           SIMIX_comm_suspend(process->waiting_action);
 365           break;
 366
 367         case SIMIX_ACTION_SLEEP:
 368           SIMIX_process_sleep_suspend(process->waiting_action);
 369           break;
 370
 371         default:
 372           xbt_die("Internal error in SIMIX_process_suspend: unexpected action type %d",
 373               process->waiting_action->type);
 374       }
 375     }
 376   }
 377 }
 378
 379 void SIMIX_process_resume(smx_process_t process, smx_process_t issuer)
 380 {
 381   xbt_assert((process != NULL), "Invalid parameters");
 382
 383   if (!process->suspended) {
 384     XBT_DEBUG("Process '%s' is not suspended", process->name);
 385     return;
 386   }
 387
 388   process->suspended = 0;
 389
 390   /* If we are resuming another process, resume the action it was waiting for
 391      if any. Otherwise add it to the list of process to run in the next round. */
 392   if (process != issuer) {
 393
 394     if (process->waiting_action) {
 395
 396       switch (process->waiting_action->type) {
 397
 398         case SIMIX_ACTION_EXECUTE:
 399         case SIMIX_ACTION_PARALLEL_EXECUTE:
 400           SIMIX_host_execution_resume(process->waiting_action);
 401           break;
 402
 403         case SIMIX_ACTION_COMMUNICATE:
 404           SIMIX_comm_resume(process->waiting_action);
 405           break;
 406
 407         case SIMIX_ACTION_SLEEP:
 408           SIMIX_process_sleep_resume(process->waiting_action);
 409           break;
 410
 411         default:
 412           xbt_die("Internal error in SIMIX_process_resume: unexpected action type %d",
 413               process->waiting_action->type);
 414       }
 415     }
 416     else {
 417       xbt_dynar_push_as(simix_global->process_to_run, smx_process_t, process);
 418     }
 419   }
 420 }
 421
 422 int SIMIX_process_get_maxpid(void) {
 423   return simix_process_maxpid;
 424 }
 425
 426 int SIMIX_process_count(void)
 427 {
 428   return xbt_swag_size(simix_global->process_list);
 429 }
 430
 431 void* SIMIX_process_self_get_data(smx_process_t self)
 432 {
 433   xbt_assert(self == SIMIX_process_self(), "This is not the current process");
 434
 435   if (!self) {
 436     return NULL;
 437   }
 438   return SIMIX_process_get_data(self);
 439 }
 440
 441 void SIMIX_process_self_set_data(smx_process_t self, void *data)
 442 {
 443   xbt_assert(self == SIMIX_process_self(), "This is not the current process");
 444
 445   SIMIX_process_set_data(self, data);
 446 }
 447
 448 void* SIMIX_process_get_data(smx_process_t process)
 449 {
 450   return process->data;
 451 }
 452
 453 void SIMIX_process_set_data(smx_process_t process, void *data)
 454 {
 455   process->data = data;
 456 }
 457
 458 smx_host_t SIMIX_process_get_host(smx_process_t process)
 459 {
 460   return process->smx_host;
 461 }
 462
 463 /* needs to be public and without simcall because it is called
 464    by exceptions and logging events */
 465 const char* SIMIX_process_self_get_name(void) {
 466
 467   smx_process_t process = SIMIX_process_self();
 468   if (process == NULL || process == simix_global->maestro_process)
 469     return "";
 470
 471   return SIMIX_process_get_name(process);
 472 }
 473
 474 const char* SIMIX_process_get_name(smx_process_t process)
 475 {
 476   return process->name;
 477 }
 478
 479 smx_process_t SIMIX_process_get_by_name(const char* name)
 480 {
 481   smx_process_t proc;
 482
 483   xbt_swag_foreach(proc, simix_global->process_list)
 484   {
 485     if(!strcmp(name, proc->name))
 486       return proc;
 487   }
 488   return NULL;
 489 }
 490
 491 int SIMIX_process_is_suspended(smx_process_t process)
 492 {
 493   return process->suspended;
 494 }
 495
 496 xbt_dict_t SIMIX_process_get_properties(smx_process_t process)
 497 {
 498   return process->properties;
 499 }
 500
 501 void SIMIX_pre_process_sleep(smx_simcall_t simcall)
 502 {
 503   if (MC_IS_ENABLED) {
 504     MC_process_clock_add(simcall->issuer, simcall->process_sleep.duration);
 505     simcall->process_sleep.result = SIMIX_DONE;
 506     SIMIX_simcall_answer(simcall);
 507     return;
 508   }
 509   smx_action_t action = SIMIX_process_sleep(simcall->issuer, simcall->process_sleep.duration);
 510   xbt_fifo_push(action->simcalls, simcall);
 511   simcall->issuer->waiting_action = action;
 512 }
 513
 514 smx_action_t SIMIX_process_sleep(smx_process_t process, double duration)
 515 {
 516   smx_action_t action;
 517   smx_host_t host = process->smx_host;
 518
 519   /* check if the host is active */
 520   if (surf_workstation_model->extension.
 521       workstation.get_state(host->host) != SURF_RESOURCE_ON) {
 522     THROWF(host_error, 0, "Host %s failed, you cannot call this function",
 523            host->name);
 524   }
 525
 526   action = xbt_mallocator_get(simix_global->action_mallocator);
 527   action->type = SIMIX_ACTION_SLEEP;
 528   action->name = NULL;
 529 #ifdef HAVE_TRACING
 530   action->category = NULL;
 531 #endif
 532
 533   action->sleep.host = host;
 534   action->sleep.surf_sleep =
 535       surf_workstation_model->extension.workstation.sleep(host->host, duration);
 536
 537   surf_workstation_model->action_data_set(action->sleep.surf_sleep, action);
 538   XBT_DEBUG("Create sleep action %p", action);
 539
 540   return action;
 541 }
 542
 543 void SIMIX_post_process_sleep(smx_action_t action)
 544 {
 545   smx_simcall_t simcall;
 546   e_smx_state_t state;
 547
 548   while ((simcall = xbt_fifo_shift(action->simcalls))) {
 549
 550     switch(surf_workstation_model->action_state_get(action->sleep.surf_sleep)){
 551       case SURF_ACTION_FAILED:
 552         state = SIMIX_SRC_HOST_FAILURE;
 553         break;
 554
 555       case SURF_ACTION_DONE:
 556         state = SIMIX_DONE;
 557         break;
 558
 559       default:
 560         THROW_IMPOSSIBLE;
 561         break;
 562     }
 563     simcall->process_sleep.result = state;
 564     simcall->issuer->waiting_action = NULL;
 565     SIMIX_simcall_answer(simcall);
 566   }
 567   SIMIX_process_sleep_destroy(action);
 568 }
 569
 570 void SIMIX_process_sleep_destroy(smx_action_t action)
 571 {
 572   XBT_DEBUG("Destroy action %p", action);
 573   if (action->sleep.surf_sleep)
 574     action->sleep.surf_sleep->model_type->action_unref(action->sleep.surf_sleep);
 575   xbt_mallocator_release(simix_global->action_mallocator, action);
 576 }
 577
 578 void SIMIX_process_sleep_suspend(smx_action_t action)
 579 {
 580   surf_workstation_model->suspend(action->sleep.surf_sleep);
 581 }
 582
 583 void SIMIX_process_sleep_resume(smx_action_t action)
 584 {
 585   surf_workstation_model->resume(action->sleep.surf_sleep);
 586 }
 587
 588 /**
 589  * \brief Calling this function makes the process to yield.
 590  *
 591  * Only the current process can call this function, giving back the control to
 592  * maestro.
 593  *
 594  * \param self the current process
 595  */
 596 void SIMIX_process_yield(smx_process_t self)
 597 {
 598   XBT_DEBUG("Yield process '%s'", self->name);
 599
 600   /* Go into sleep and return control to maestro */
 601   SIMIX_context_suspend(self->context);
 602
 603   /* Ok, maestro returned control to us */
 604   XBT_DEBUG("Control returned to me: '%s'", self->name);
 605
 606   if (self->context->iwannadie){
 607     XBT_DEBUG("I wanna die!");
 608     SIMIX_context_stop(self->context);
 609   }
 610
 611   if (self->doexception) {
 612     XBT_DEBUG("Wait, maestro left me an exception");
 613     self->doexception = 0;
 614     SMX_THROW();
 615   }
 616
 617   if (self->new_host) {
 618     SIMIX_process_change_host(self, self->new_host);
 619     self->new_host = NULL;
 620   }
 621 }
 622
 623 /* callback: context fetching */
 624 xbt_running_ctx_t *SIMIX_process_get_running_context(void)
 625 {
 626   return SIMIX_process_self()->running_ctx;
 627 }
 628
 629 /* callback: termination */
 630 void SIMIX_process_exception_terminate(xbt_ex_t * e)
 631 {
 632   xbt_ex_display(e);
 633   abort();
 634 }
 635
 636 smx_context_t SIMIX_process_get_context(smx_process_t p) {
 637   return p->context;
 638 }
 639
 640 void SIMIX_process_set_context(smx_process_t p,smx_context_t c) {
 641   p->context = c;
 642 }
 643
 644 /**
 645  * \brief Returns the list of processes to run.
 646  */
 647 xbt_dynar_t SIMIX_process_get_runnable(void)
 648 {
 649   return simix_global->process_to_run;
 650 }
 651
 652 /**
 653  * \brief Returns the process from PID.
 654  */
 655 smx_process_t SIMIX_process_from_PID(int PID)
 656 {
 657         smx_process_t proc;
 658         xbt_swag_foreach(proc, simix_global->process_list)
 659         {
 660          if(proc->pid == PID)
 661          return proc;
 662         }
 663         return NULL;
 664 }