src/simix/smx_process.c

   1 /* Copyright (c) 2007-2014. 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 #include "mc/mc_replay.h"
  13 #include "mc/mc_client.h"
  14
  15 XBT_LOG_NEW_DEFAULT_SUBCATEGORY(simix_process, simix,
  16                                 "Logging specific to SIMIX (process)");
  17
  18 unsigned long simix_process_maxpid = 0;
  19
  20 /**
  21  * \brief Returns the current agent.
  22  *
  23  * This functions returns the currently running SIMIX process.
  24  *
  25  * \return The SIMIX process
  26  */
  27 XBT_INLINE smx_process_t SIMIX_process_self(void)
  28 {
  29   smx_context_t self_context = SIMIX_context_self();
  30
  31   return self_context ? SIMIX_context_get_process(self_context) : NULL;
  32 }
  33
  34 /**
  35  * \brief Returns whether a process has pending asynchronous communications.
  36  * \return true if there are asynchronous communications in this process
  37  */
  38 int SIMIX_process_has_pending_comms(smx_process_t process) {
  39
  40   return xbt_fifo_size(process->comms) > 0;
  41 }
  42
  43 /**
  44  * \brief Moves a process to the list of processes to destroy.
  45  */
  46 void SIMIX_process_cleanup(smx_process_t process)
  47 {
  48   XBT_DEBUG("Cleanup process %s (%p), waiting synchro %p",
  49       process->name, process, process->waiting_synchro);
  50
  51   SIMIX_process_on_exit_runall(process);
  52
  53   /* Unregister from the kill timer if any */
  54   if (process->kill_timer != NULL)
  55           SIMIX_timer_remove(process->kill_timer);
  56
  57   xbt_os_mutex_acquire(simix_global->mutex);
  58
  59   /* cancel non-blocking communications */
  60   smx_synchro_t synchro;
  61   while ((synchro = xbt_fifo_pop(process->comms))) {
  62
  63     /* make sure no one will finish the comm after this process is destroyed,
  64      * because src_proc or dst_proc would be an invalid pointer */
  65     SIMIX_comm_cancel(synchro);
  66
  67     if (synchro->comm.src_proc == process) {
  68       XBT_DEBUG("Found an unfinished send comm %p (detached = %d), state %d, src = %p, dst = %p",
  69           synchro, synchro->comm.detached, (int)synchro->state, synchro->comm.src_proc, synchro->comm.dst_proc);
  70       synchro->comm.src_proc = NULL;
  71
  72       /* I'm not supposed to destroy a detached comm from the sender side, */
  73       if (!synchro->comm.detached)
  74         SIMIX_comm_destroy(synchro);
  75       else
  76         XBT_DEBUG("Don't destroy it since it's a detached comm");
  77
  78     }
  79     else if (synchro->comm.dst_proc == process){
  80       XBT_DEBUG("Found an unfinished recv comm %p, state %d, src = %p, dst = %p",
  81           synchro, (int)synchro->state, synchro->comm.src_proc, synchro->comm.dst_proc);
  82       synchro->comm.dst_proc = NULL;
  83
  84       if (synchro->comm.detached && synchro->comm.refcount == 1
  85           && synchro->comm.src_proc != NULL) {
  86         /* the comm will be freed right now, remove it from the sender */
  87         xbt_fifo_remove(synchro->comm.src_proc->comms, synchro);
  88       }
  89       SIMIX_comm_destroy(synchro);
  90     }
  91     else {
  92       xbt_die("Communication synchro %p is in my list but I'm not the sender "
  93           "or the receiver", synchro);
  94     }
  95   }
  96
  97   XBT_DEBUG("%p should not be run anymore",process);
  98   xbt_swag_remove(process, simix_global->process_list);
  99   xbt_swag_remove(process, SIMIX_host_priv(process->smx_host)->process_list);
 100   xbt_swag_insert(process, simix_global->process_to_destroy);
 101   process->context->iwannadie = 0;
 102
 103   xbt_os_mutex_release(simix_global->mutex);
 104 }
 105
 106 /**
 107  * Garbage collection
 108  *
 109  * Should be called some time to time to free the memory allocated for processes
 110  * that have finished (or killed).
 111  */
 112 void SIMIX_process_empty_trash(void)
 113 {
 114   smx_process_t process = NULL;
 115
 116   while ((process = xbt_swag_extract(simix_global->process_to_destroy))) {
 117     XBT_DEBUG("Getting rid of %p",process);
 118
 119     SIMIX_context_free(process->context);
 120
 121     /* Free the exception allocated at creation time */
 122     free(process->running_ctx);
 123     xbt_dict_free(&process->properties);
 124
 125     xbt_fifo_free(process->comms);
 126
 127     xbt_dynar_free(&process->on_exit);
 128
 129     xbt_free(process->name);
 130     xbt_free(process);
 131   }
 132 }
 133
 134 /**
 135  * \brief Creates and runs the maestro process
 136  */
 137 void SIMIX_create_maestro_process()
 138 {
 139   smx_process_t maestro = NULL;
 140
 141   /* Create maestro process and intilialize it */
 142   maestro = xbt_new0(s_smx_process_t, 1);
 143   maestro->pid = simix_process_maxpid++;
 144   maestro->ppid = -1;
 145   maestro->name = (char *) "";
 146   maestro->running_ctx = xbt_new(xbt_running_ctx_t, 1);
 147   XBT_RUNNING_CTX_INITIALIZE(maestro->running_ctx);
 148   maestro->context = SIMIX_context_new(NULL, 0, NULL, NULL, maestro);
 149   maestro->simcall.issuer = maestro;
 150   simix_global->maestro_process = maestro;
 151   return;
 152 }
 153 /**
 154  * \brief Stops a process.
 155  *
 156  * Stops the process, execute all the registered on_exit functions,
 157  * register it to the list of the process to restart if needed
 158  * and stops its context.
 159  */
 160 void SIMIX_process_stop(smx_process_t arg) {
 161   /* execute the on_exit functions */
 162   SIMIX_process_on_exit_runall(arg);
 163   /* Add the process to the list of process to restart, only if
 164    * the host is down
 165    */
 166   if (arg->auto_restart && !SIMIX_host_get_state(arg->smx_host)) {
 167     SIMIX_host_add_auto_restart_process(arg->smx_host,arg->name,arg->code, arg->data,
 168                                         sg_host_name(arg->smx_host),
 169                                         SIMIX_timer_get_date(arg->kill_timer),
 170                                         arg->argc,arg->argv,arg->properties,
 171                                         arg->auto_restart);
 172   }
 173   XBT_DEBUG("Process %s (%s) is dead",arg->name,sg_host_name(arg->smx_host));
 174   /* stop the context */
 175   SIMIX_context_stop(arg->context);
 176 }
 177
 178 /**
 179  * \brief Same as SIMIX_process_create() but with only one argument (used by timers).
 180  * This function frees the argument.
 181  * \return the process created
 182  */
 183 smx_process_t SIMIX_process_create_from_wrapper(smx_process_arg_t args) {
 184
 185   smx_process_t process;
 186   simix_global->create_process_function(&process,
 187                                         args->name,
 188                                         args->code,
 189                                         args->data,
 190                                         args->hostname,
 191                                         args->kill_time,
 192                                         args->argc,
 193                                         args->argv,
 194                                         args->properties,
 195                                         args->auto_restart,
 196                                         NULL);
 197   xbt_free(args);
 198   return process;
 199 }
 200
 201
 202 void simcall_HANDLER_process_create(smx_simcall_t simcall,
 203                           smx_process_t *process,
 204                           const char *name,
 205                           xbt_main_func_t code,
 206                           void *data,
 207                           const char *hostname,
 208                           double kill_time,
 209                           int argc, char **argv,
 210                           xbt_dict_t properties,
 211                           int auto_restart){
 212   SIMIX_process_create(process, name, code, data, hostname,
 213                        kill_time, argc, argv, properties, auto_restart,
 214                        simcall->issuer);
 215 }
 216 /**
 217  * \brief Internal function to create a process.
 218  *
 219  * This function actually creates the process.
 220  * It may be called when a SIMCALL_PROCESS_CREATE simcall occurs,
 221  * or directly for SIMIX internal purposes. The sure thing is that it's called from maestro context.
 222  *
 223  * \return the process created
 224  */
 225 void SIMIX_process_create(smx_process_t *process,
 226                           const char *name,
 227                           xbt_main_func_t code,
 228                           void *data,
 229                           const char *hostname,
 230                           double kill_time,
 231                           int argc, char **argv,
 232                           xbt_dict_t properties,
 233                           int auto_restart,
 234                           smx_process_t parent_process)
 235 {
 236   *process = NULL;
 237   smx_host_t host = SIMIX_host_get_by_name(hostname);
 238
 239   XBT_DEBUG("Start process %s on host '%s'", name, hostname);
 240
 241   if (!SIMIX_host_get_state(host)) {
 242     int i;
 243     XBT_WARN("Cannot launch process '%s' on failed host '%s'", name,
 244           hostname);
 245     for (i = 0; i < argc; i++)
 246       xbt_free(argv[i]);
 247     xbt_free(argv);
 248   }
 249   else {
 250     *process = xbt_new0(s_smx_process_t, 1);
 251
 252     xbt_assert(((code != NULL) && (host != NULL)), "Invalid parameters");
 253     /* Process data */
 254     (*process)->pid = simix_process_maxpid++;
 255     (*process)->name = xbt_strdup(name);
 256     (*process)->smx_host = host;
 257     (*process)->data = data;
 258     (*process)->comms = xbt_fifo_new();
 259     (*process)->simcall.issuer = *process;
 260
 261      if (parent_process) {
 262        (*process)->ppid = SIMIX_process_get_PID(parent_process);
 263      } else {
 264        (*process)->ppid = -1;
 265      }
 266
 267     /* Process data for auto-restart */
 268     (*process)->auto_restart = auto_restart;
 269     (*process)->code = code;
 270     (*process)->argc = argc;
 271     (*process)->argv = argv;
 272
 273
 274     XBT_VERB("Create context %s", (*process)->name);
 275     (*process)->context = SIMIX_context_new(code, argc, argv,
 276       simix_global->cleanup_process_function, *process);
 277
 278     (*process)->running_ctx = xbt_new(xbt_running_ctx_t, 1);
 279     XBT_RUNNING_CTX_INITIALIZE((*process)->running_ctx);
 280
 281     if(MC_is_active()){
 282       MC_ignore_heap((*process)->running_ctx, sizeof(*(*process)->running_ctx));
 283     }
 284
 285     /* Add properties */
 286     (*process)->properties = properties;
 287
 288     /* Add the process to it's host process list */
 289     xbt_swag_insert(*process, SIMIX_host_priv(host)->process_list);
 290
 291     XBT_DEBUG("Start context '%s'", (*process)->name);
 292
 293     /* Now insert it in the global process list and in the process to run list */
 294     xbt_swag_insert(*process, simix_global->process_list);
 295     XBT_DEBUG("Inserting %s(%s) in the to_run list", (*process)->name, sg_host_name(host));
 296     xbt_dynar_push_as(simix_global->process_to_run, smx_process_t, *process);
 297
 298     if (kill_time > SIMIX_get_clock() && simix_global->kill_process_function) {
 299       XBT_DEBUG("Process %s(%s) will be kill at time %f", (*process)->name,
 300           sg_host_name((*process)->smx_host), kill_time);
 301       (*process)->kill_timer = SIMIX_timer_set(kill_time, simix_global->kill_process_function, *process);
 302     }
 303   }
 304 }
 305
 306 /**
 307  * \brief Executes the processes from simix_global->process_to_run.
 308  *
 309  * The processes of simix_global->process_to_run are run (in parallel if
 310  * possible).  On exit, simix_global->process_to_run is empty, and
 311  * simix_global->process_that_ran contains the list of processes that just ran.
 312  * The two lists are swapped so, be careful when using them before and after a
 313  * call to this function.
 314  */
 315 void SIMIX_process_runall(void)
 316 {
 317   SIMIX_context_runall();
 318
 319   xbt_dynar_t tmp = simix_global->process_that_ran;
 320   simix_global->process_that_ran = simix_global->process_to_run;
 321   simix_global->process_to_run = tmp;
 322   xbt_dynar_reset(simix_global->process_to_run);
 323 }
 324
 325 void simcall_HANDLER_process_kill(smx_simcall_t simcall, smx_process_t process) {
 326   SIMIX_process_kill(process, simcall->issuer);
 327 }
 328 /**
 329  * \brief Internal function to kill a SIMIX process.
 330  *
 331  * This function may be called when a SIMCALL_PROCESS_KILL simcall occurs,
 332  * or directly for SIMIX internal purposes.
 333  *
 334  * \param process poor victim
 335  * \param issuer the process which has sent the PROCESS_KILL. Important to not schedule twice the same process.
 336  */
 337 void SIMIX_process_kill(smx_process_t process, smx_process_t issuer) {
 338
 339   XBT_DEBUG("Killing process %s on %s", process->name, sg_host_name(process->smx_host));
 340
 341   process->context->iwannadie = 1;
 342   process->blocked = 0;
 343   process->suspended = 0;
 344   process->doexception = 0;
 345
 346   /* destroy the blocking synchro if any */
 347   if (process->waiting_synchro) {
 348
 349     switch (process->waiting_synchro->type) {
 350
 351     case SIMIX_SYNC_EXECUTE:
 352     case SIMIX_SYNC_PARALLEL_EXECUTE:
 353       SIMIX_host_execution_destroy(process->waiting_synchro);
 354       break;
 355
 356     case SIMIX_SYNC_COMMUNICATE:
 357       xbt_fifo_remove(process->comms, process->waiting_synchro);
 358       SIMIX_comm_cancel(process->waiting_synchro);
 359       SIMIX_comm_destroy(process->waiting_synchro);
 360       break;
 361
 362     case SIMIX_SYNC_SLEEP:
 363       SIMIX_process_sleep_destroy(process->waiting_synchro);
 364       break;
 365
 366     case SIMIX_SYNC_JOIN:
 367       SIMIX_process_sleep_destroy(process->waiting_synchro);
 368       break;
 369
 370     case SIMIX_SYNC_SYNCHRO:
 371       SIMIX_synchro_stop_waiting(process, &process->simcall);
 372       SIMIX_synchro_destroy(process->waiting_synchro);
 373       break;
 374
 375     case SIMIX_SYNC_IO:
 376       SIMIX_io_destroy(process->waiting_synchro);
 377       break;
 378
 379     }
 380   }
 381   if(!xbt_dynar_member(simix_global->process_to_run, &(process)) && process != issuer) {
 382     XBT_DEBUG("Inserting %s in the to_run list", process->name);
 383     xbt_dynar_push_as(simix_global->process_to_run, smx_process_t, process);
 384   }
 385
 386 }
 387
 388 /** @brief Ask another process to raise the given exception
 389  *
 390  * @param cat category of exception
 391  * @param value value associated to the exception
 392  * @param msg string information associated to the exception
 393  */
 394 void SIMIX_process_throw(smx_process_t process, xbt_errcat_t cat, int value, const char *msg) {
 395   SMX_EXCEPTION(process, cat, value, msg);
 396
 397   if (process->suspended)
 398     SIMIX_process_resume(process,SIMIX_process_self());
 399
 400   /* cancel the blocking synchro if any */
 401   if (process->waiting_synchro) {
 402
 403     switch (process->waiting_synchro->type) {
 404
 405     case SIMIX_SYNC_EXECUTE:
 406     case SIMIX_SYNC_PARALLEL_EXECUTE:
 407       SIMIX_host_execution_cancel(process->waiting_synchro);
 408       break;
 409
 410     case SIMIX_SYNC_COMMUNICATE:
 411       xbt_fifo_remove(process->comms, process->waiting_synchro);
 412       SIMIX_comm_cancel(process->waiting_synchro);
 413       break;
 414
 415     case SIMIX_SYNC_SLEEP:
 416     case SIMIX_SYNC_JOIN:
 417       SIMIX_process_sleep_destroy(process->waiting_synchro);
 418       if (!xbt_dynar_member(simix_global->process_to_run, &(process)) && process != SIMIX_process_self()) {
 419         XBT_DEBUG("Inserting %s in the to_run list", process->name);
 420         xbt_dynar_push_as(simix_global->process_to_run, smx_process_t, process);
 421       }
 422       break;
 423
 424     case SIMIX_SYNC_SYNCHRO:
 425       SIMIX_synchro_stop_waiting(process, &process->simcall);
 426       break;
 427
 428     case SIMIX_SYNC_IO:
 429       SIMIX_io_destroy(process->waiting_synchro);
 430       break;
 431
 432     }
 433   }
 434   process->waiting_synchro = NULL;
 435
 436 }
 437
 438 void simcall_HANDLER_process_killall(smx_simcall_t simcall, int reset_pid) {
 439   SIMIX_process_killall(simcall->issuer, reset_pid);
 440 }
 441 /**
 442  * \brief Kills all running processes.
 443  * \param issuer this one will not be killed
 444  */
 445 void SIMIX_process_killall(smx_process_t issuer, int reset_pid)
 446 {
 447   smx_process_t p = NULL;
 448
 449   while ((p = xbt_swag_extract(simix_global->process_list))) {
 450     if (p != issuer) {
 451       SIMIX_process_kill(p,issuer);
 452     }
 453   }
 454
 455   if (reset_pid > 0)
 456     simix_process_maxpid = reset_pid;
 457
 458   SIMIX_context_runall();
 459
 460   SIMIX_process_empty_trash();
 461 }
 462
 463 void simcall_HANDLER_process_change_host(smx_simcall_t simcall, smx_process_t process,
 464                                    smx_host_t dest)
 465 {
 466   process->new_host = dest;
 467 }
 468 void SIMIX_process_change_host(smx_process_t process,
 469              smx_host_t dest)
 470 {
 471   xbt_assert((process != NULL), "Invalid parameters");
 472   xbt_swag_remove(process, SIMIX_host_priv(process->smx_host)->process_list);
 473   process->smx_host = dest;
 474   xbt_swag_insert(process, SIMIX_host_priv(dest)->process_list);
 475 }
 476
 477
 478 void simcall_HANDLER_process_suspend(smx_simcall_t simcall, smx_process_t process)
 479 {
 480   smx_synchro_t sync_suspend =
 481       SIMIX_process_suspend(process, simcall->issuer);
 482
 483   if (process != simcall->issuer) {
 484     SIMIX_simcall_answer(simcall);
 485   } else {
 486     xbt_fifo_push(sync_suspend->simcalls, simcall);
 487     process->waiting_synchro = sync_suspend;
 488     SIMIX_host_execution_suspend(process->waiting_synchro);
 489   }
 490   /* If we are suspending ourselves, then just do not finish the simcall now */
 491 }
 492
 493 smx_synchro_t SIMIX_process_suspend(smx_process_t process, smx_process_t issuer)
 494 {
 495   xbt_assert((process != NULL), "Invalid parameters");
 496
 497   if (process->suspended) {
 498     XBT_DEBUG("Process '%s' is already suspended", process->name);
 499     return NULL;
 500   }
 501
 502   process->suspended = 1;
 503
 504   /* If we are suspending another process, and it is waiting on a sync,
 505      suspend its synchronization. */
 506   if (process != issuer) {
 507
 508     if (process->waiting_synchro) {
 509
 510       switch (process->waiting_synchro->type) {
 511
 512         case SIMIX_SYNC_EXECUTE:
 513         case SIMIX_SYNC_PARALLEL_EXECUTE:
 514           SIMIX_host_execution_suspend(process->waiting_synchro);
 515           break;
 516
 517         case SIMIX_SYNC_COMMUNICATE:
 518           SIMIX_comm_suspend(process->waiting_synchro);
 519           break;
 520
 521         case SIMIX_SYNC_SLEEP:
 522           SIMIX_process_sleep_suspend(process->waiting_synchro);
 523           break;
 524
 525         case SIMIX_SYNC_SYNCHRO:
 526           /* Suspension is delayed to when the process is rescheduled. */
 527           break;
 528
 529         default:
 530           xbt_die("Internal error in SIMIX_process_suspend: unexpected synchronization type %d",
 531               (int)process->waiting_synchro->type);
 532       }
 533       return NULL;
 534     } else {
 535       /* Suspension is delayed to when the process is rescheduled. */
 536       return NULL;
 537     }
 538   } else {
 539     /* FIXME: computation size is zero. Is it okay that bound is zero ? */
 540     return SIMIX_host_execute("suspend", process->smx_host, 0.0, 1.0, 0.0, 0);
 541   }
 542 }
 543
 544 void simcall_HANDLER_process_resume(smx_simcall_t simcall, smx_process_t process){
 545   SIMIX_process_resume(process, simcall->issuer);
 546 }
 547
 548 void SIMIX_process_resume(smx_process_t process, smx_process_t issuer)
 549 {
 550   XBT_IN("process = %p, issuer = %p", process, issuer);
 551
 552   if(process->context->iwannadie) {
 553     XBT_VERB("Ignoring request to suspend a process that is currently dying.");
 554     return;
 555   }
 556
 557   if(!process->suspended) return;
 558   process->suspended = 0;
 559
 560   /* If we are resuming another process, resume the synchronization it was waiting for
 561      if any. Otherwise add it to the list of process to run in the next round. */
 562   if (process != issuer) {
 563
 564     if (process->waiting_synchro) {
 565
 566       switch (process->waiting_synchro->type) {
 567
 568         case SIMIX_SYNC_EXECUTE:
 569         case SIMIX_SYNC_PARALLEL_EXECUTE:
 570           SIMIX_host_execution_resume(process->waiting_synchro);
 571           break;
 572
 573         case SIMIX_SYNC_COMMUNICATE:
 574           SIMIX_comm_resume(process->waiting_synchro);
 575           break;
 576
 577         case SIMIX_SYNC_SLEEP:
 578           SIMIX_process_sleep_resume(process->waiting_synchro);
 579           break;
 580
 581         case SIMIX_SYNC_SYNCHRO:
 582           /* I cannot resume it now. This is delayed to when the process is rescheduled at
 583            * the end of the synchro. */
 584           break;
 585
 586         default:
 587           xbt_die("Internal error in SIMIX_process_resume: unexpected synchronization type %d",
 588               (int)process->waiting_synchro->type);
 589       }
 590     }
 591   } else XBT_WARN("Strange. Process %p is trying to resume himself.", issuer);
 592
 593   XBT_OUT();
 594 }
 595
 596 int SIMIX_process_get_maxpid(void) {
 597   return simix_process_maxpid;
 598 }
 599
 600 int SIMIX_process_count(void)
 601 {
 602   return xbt_swag_size(simix_global->process_list);
 603 }
 604
 605 int SIMIX_process_get_PID(smx_process_t self){
 606   if (self == NULL)
 607     return 0;
 608   else
 609     return self->pid;
 610 }
 611
 612 int SIMIX_process_get_PPID(smx_process_t self){
 613   if (self == NULL)
 614     return 0;
 615   else
 616     return self->ppid;
 617 }
 618
 619 void* SIMIX_process_self_get_data(smx_process_t self)
 620 {
 621   xbt_assert(self == SIMIX_process_self(), "This is not the current process");
 622
 623   if (!self) {
 624     return NULL;
 625   }
 626   return SIMIX_process_get_data(self);
 627 }
 628
 629 void SIMIX_process_self_set_data(smx_process_t self, void *data)
 630 {
 631   xbt_assert(self == SIMIX_process_self(), "This is not the current process");
 632
 633   SIMIX_process_set_data(self, data);
 634 }
 635
 636 void* SIMIX_process_get_data(smx_process_t process)
 637 {
 638   return process->data;
 639 }
 640
 641 void SIMIX_process_set_data(smx_process_t process, void *data)
 642 {
 643   process->data = data;
 644 }
 645
 646 smx_host_t SIMIX_process_get_host(smx_process_t process)
 647 {
 648   return process->smx_host;
 649 }
 650
 651 /* needs to be public and without simcall because it is called
 652    by exceptions and logging events */
 653 const char* SIMIX_process_self_get_name(void) {
 654
 655   smx_process_t process = SIMIX_process_self();
 656   if (process == NULL || process == simix_global->maestro_process)
 657     return "";
 658
 659   return SIMIX_process_get_name(process);
 660 }
 661
 662 const char* SIMIX_process_get_name(smx_process_t process)
 663 {
 664   return process->name;
 665 }
 666
 667 smx_process_t SIMIX_process_get_by_name(const char* name)
 668 {
 669   smx_process_t proc;
 670
 671   xbt_swag_foreach(proc, simix_global->process_list)
 672   {
 673     if(!strcmp(name, proc->name))
 674       return proc;
 675   }
 676   return NULL;
 677 }
 678
 679 int SIMIX_process_is_suspended(smx_process_t process)
 680 {
 681   return process->suspended;
 682 }
 683
 684 xbt_dict_t SIMIX_process_get_properties(smx_process_t process)
 685 {
 686   return process->properties;
 687 }
 688
 689 void simcall_HANDLER_process_join(smx_simcall_t simcall, smx_process_t process, double timeout)
 690 {
 691   smx_synchro_t sync = SIMIX_process_join(simcall->issuer, process, timeout);
 692   xbt_fifo_push(sync->simcalls, simcall);
 693   simcall->issuer->waiting_synchro = sync;
 694 }
 695
 696 static int SIMIX_process_join_finish(smx_process_exit_status_t status, smx_synchro_t sync){
 697   if (sync->sleep.surf_sleep) {
 698     surf_action_cancel(sync->sleep.surf_sleep);
 699
 700     smx_simcall_t simcall;
 701     while ((simcall = xbt_fifo_shift(sync->simcalls))) {
 702       simcall_process_sleep__set__result(simcall, SIMIX_DONE);
 703       simcall->issuer->waiting_synchro = NULL;
 704       if (simcall->issuer->suspended) {
 705         XBT_DEBUG("Wait! This process is suspended and can't wake up now.");
 706         simcall->issuer->suspended = 0;
 707         simcall_HANDLER_process_suspend(simcall, simcall->issuer);
 708       } else {
 709         SIMIX_simcall_answer(simcall);
 710       }
 711     }
 712     surf_action_unref(sync->sleep.surf_sleep);
 713     sync->sleep.surf_sleep = NULL;
 714   }
 715   xbt_mallocator_release(simix_global->synchro_mallocator, sync);
 716   return 0;
 717 }
 718
 719 smx_synchro_t SIMIX_process_join(smx_process_t issuer, smx_process_t process, double timeout)
 720 {
 721   smx_synchro_t res = SIMIX_process_sleep(issuer, timeout);
 722   res->type = SIMIX_SYNC_JOIN;
 723   SIMIX_process_on_exit(process, (int_f_pvoid_pvoid_t)SIMIX_process_join_finish, res);
 724   return res;
 725 }
 726
 727 void simcall_HANDLER_process_sleep(smx_simcall_t simcall, double duration)
 728 {
 729   if (MC_is_active() || MC_record_replay_is_active()) {
 730     MC_process_clock_add(simcall->issuer, duration);
 731     simcall_process_sleep__set__result(simcall, SIMIX_DONE);
 732     SIMIX_simcall_answer(simcall);
 733     return;
 734   }
 735   smx_synchro_t sync = SIMIX_process_sleep(simcall->issuer, duration);
 736   xbt_fifo_push(sync->simcalls, simcall);
 737   simcall->issuer->waiting_synchro = sync;
 738 }
 739
 740 smx_synchro_t SIMIX_process_sleep(smx_process_t process, double duration)
 741 {
 742   smx_synchro_t synchro;
 743   smx_host_t host = process->smx_host;
 744
 745   /* check if the host is active */
 746   if (surf_resource_get_state(surf_workstation_resource_priv(host)) != SURF_RESOURCE_ON) {
 747     THROWF(host_error, 0, "Host %s failed, you cannot call this function",
 748            sg_host_name(host));
 749   }
 750
 751   synchro = xbt_mallocator_get(simix_global->synchro_mallocator);
 752   synchro->type = SIMIX_SYNC_SLEEP;
 753   synchro->name = NULL;
 754   synchro->category = NULL;
 755
 756   synchro->sleep.host = host;
 757   synchro->sleep.surf_sleep =
 758       surf_workstation_sleep(host, duration);
 759
 760   surf_action_set_data(synchro->sleep.surf_sleep, synchro);
 761   XBT_DEBUG("Create sleep synchronization %p", synchro);
 762
 763   return synchro;
 764 }
 765
 766 void SIMIX_post_process_sleep(smx_synchro_t synchro)
 767 {
 768   smx_simcall_t simcall;
 769   e_smx_state_t state;
 770   xbt_assert(synchro->type == SIMIX_SYNC_SLEEP || synchro->type == SIMIX_SYNC_JOIN);
 771
 772   while ((simcall = xbt_fifo_shift(synchro->simcalls))) {
 773
 774     switch(surf_action_get_state(synchro->sleep.surf_sleep)){
 775       case SURF_ACTION_FAILED:
 776         simcall->issuer->context->iwannadie = 1;
 777         //SMX_EXCEPTION(simcall->issuer, host_error, 0, "Host failed");
 778         state = SIMIX_SRC_HOST_FAILURE;
 779         break;
 780
 781       case SURF_ACTION_DONE:
 782         state = SIMIX_DONE;
 783         break;
 784
 785       default:
 786         THROW_IMPOSSIBLE;
 787         break;
 788     }
 789     if (surf_resource_get_state(surf_workstation_resource_priv(simcall->issuer->smx_host)) != SURF_RESOURCE_ON) {
 790       simcall->issuer->context->iwannadie = 1;
 791     }
 792     simcall_process_sleep__set__result(simcall, state);
 793     simcall->issuer->waiting_synchro = NULL;
 794     if (simcall->issuer->suspended) {
 795       XBT_DEBUG("Wait! This process is suspended and can't wake up now.");
 796       simcall->issuer->suspended = 0;
 797       simcall_HANDLER_process_suspend(simcall, simcall->issuer);
 798     } else {
 799       SIMIX_simcall_answer(simcall);
 800     }
 801   }
 802
 803   SIMIX_process_sleep_destroy(synchro);
 804 }
 805
 806 void SIMIX_process_sleep_destroy(smx_synchro_t synchro)
 807 {
 808   XBT_DEBUG("Destroy synchro %p", synchro);
 809   xbt_assert(synchro->type == SIMIX_SYNC_SLEEP || synchro->type == SIMIX_SYNC_JOIN);
 810
 811   if (synchro->sleep.surf_sleep) {
 812     surf_action_unref(synchro->sleep.surf_sleep);
 813     synchro->sleep.surf_sleep = NULL;
 814   }
 815   if (synchro->type == SIMIX_SYNC_SLEEP)
 816     xbt_mallocator_release(simix_global->synchro_mallocator, synchro);
 817 }
 818
 819 void SIMIX_process_sleep_suspend(smx_synchro_t synchro)
 820 {
 821   xbt_assert(synchro->type == SIMIX_SYNC_SLEEP);
 822   surf_action_suspend(synchro->sleep.surf_sleep);
 823 }
 824
 825 void SIMIX_process_sleep_resume(smx_synchro_t synchro)
 826 {
 827   XBT_DEBUG("Synchro state is %d on process_sleep_resume.", synchro->state);
 828   xbt_assert(synchro->type == SIMIX_SYNC_SLEEP);
 829   surf_action_resume(synchro->sleep.surf_sleep);
 830 }
 831
 832 /**
 833  * \brief Calling this function makes the process to yield.
 834  *
 835  * Only the current process can call this function, giving back the control to
 836  * maestro.
 837  *
 838  * \param self the current process
 839  */
 840 void SIMIX_process_yield(smx_process_t self)
 841 {
 842   XBT_DEBUG("Yield process '%s'", self->name);
 843
 844   /* Go into sleep and return control to maestro */
 845   SIMIX_context_suspend(self->context);
 846
 847   /* Ok, maestro returned control to us */
 848   XBT_DEBUG("Control returned to me: '%s'", self->name);
 849
 850   if (self->new_host) {
 851     SIMIX_process_change_host(self, self->new_host);
 852     self->new_host = NULL;
 853   }
 854
 855   if (self->context->iwannadie){
 856     XBT_DEBUG("I wanna die!");
 857     SIMIX_process_stop(self);
 858   }
 859
 860   if (self->suspended) {
 861     XBT_DEBUG("Hey! I'm suspended.");
 862     xbt_assert(!self->doexception, "Gasp! This exception may be lost by subsequent calls.");
 863     self->suspended = 0;
 864     SIMIX_process_suspend(self, self);
 865   }
 866
 867   if (self->doexception) {
 868     XBT_DEBUG("Wait, maestro left me an exception");
 869     self->doexception = 0;
 870     SMX_THROW();
 871   }
 872
 873 }
 874
 875 /* callback: context fetching */
 876 xbt_running_ctx_t *SIMIX_process_get_running_context(void)
 877 {
 878   return SIMIX_process_self()->running_ctx;
 879 }
 880
 881 /* callback: termination */
 882 void SIMIX_process_exception_terminate(xbt_ex_t * e)
 883 {
 884   xbt_ex_display(e);
 885   xbt_abort();
 886 }
 887
 888 smx_context_t SIMIX_process_get_context(smx_process_t p) {
 889   return p->context;
 890 }
 891
 892 void SIMIX_process_set_context(smx_process_t p,smx_context_t c) {
 893   p->context = c;
 894 }
 895
 896 /**
 897  * \brief Returns the list of processes to run.
 898  */
 899 xbt_dynar_t SIMIX_process_get_runnable(void)
 900 {
 901   return simix_global->process_to_run;
 902 }
 903
 904 /**
 905  * \brief Returns the process from PID.
 906  */
 907 smx_process_t SIMIX_process_from_PID(int PID)
 908 {
 909   smx_process_t proc;
 910   xbt_swag_foreach(proc, simix_global->process_list)
 911   {
 912    if(proc->pid == PID)
 913    return proc;
 914   }
 915   return NULL;
 916 }
 917
 918 /** @brief returns a dynar containg all currently existing processes */
 919 xbt_dynar_t SIMIX_processes_as_dynar(void) {
 920   smx_process_t proc;
 921   xbt_dynar_t res = xbt_dynar_new(sizeof(smx_process_t),NULL);
 922   xbt_swag_foreach(proc, simix_global->process_list) {
 923     xbt_dynar_push(res,&proc);
 924   }
 925   return res;
 926 }
 927
 928
 929 void SIMIX_process_on_exit_runall(smx_process_t process) {
 930   s_smx_process_exit_fun_t exit_fun;
 931   smx_process_exit_status_t exit_status = (process->context->iwannadie) ?
 932                                          SMX_EXIT_FAILURE : SMX_EXIT_SUCCESS;
 933   while (!xbt_dynar_is_empty(process->on_exit)) {
 934     exit_fun = xbt_dynar_pop_as(process->on_exit,s_smx_process_exit_fun_t);
 935     (exit_fun.fun)((void*)exit_status, exit_fun.arg);
 936   }
 937 }
 938
 939 void SIMIX_process_on_exit(smx_process_t process, int_f_pvoid_pvoid_t fun, void *data) {
 940   xbt_assert(process, "current process not found: are you in maestro context ?");
 941
 942   if (!process->on_exit) {
 943     process->on_exit = xbt_dynar_new(sizeof(s_smx_process_exit_fun_t), NULL);
 944   }
 945
 946   s_smx_process_exit_fun_t exit_fun = {fun, data};
 947
 948   xbt_dynar_push_as(process->on_exit,s_smx_process_exit_fun_t,exit_fun);
 949 }
 950
 951 /**
 952  * \brief Sets the auto-restart status of the process.
 953  * If set to 1, the process will be automatically restarted when its host
 954  * comes back.
 955  */
 956 void SIMIX_process_auto_restart_set(smx_process_t process, int auto_restart) {
 957   process->auto_restart = auto_restart;
 958 }
 959
 960 smx_process_t simcall_HANDLER_process_restart(smx_simcall_t simcall, smx_process_t process) {
 961   return SIMIX_process_restart(process, simcall->issuer);
 962 }
 963 /** @brief Restart a process, starting it again from the beginning. */
 964 smx_process_t SIMIX_process_restart(smx_process_t process, smx_process_t issuer) {
 965   XBT_DEBUG("Restarting process %s on %s", process->name, sg_host_name(process->smx_host));
 966   //retrieve the arguments of the old process
 967   //FIXME: Factorize this with SIMIX_host_add_auto_restart_process ?
 968   s_smx_process_arg_t arg;
 969   arg.code = process->code;
 970   arg.hostname = sg_host_name(process->smx_host);
 971   arg.kill_time = SIMIX_timer_get_date(process->kill_timer);
 972   arg.argc = process->argc;
 973   arg.data = process->data;
 974   int i;
 975   arg.argv = xbt_new(char*,process->argc + 1);
 976   for (i = 0; i < arg.argc; i++) {
 977     arg.argv[i] = xbt_strdup(process->argv[i]);
 978   }
 979   arg.argv[process->argc] = NULL;
 980   arg.properties = NULL;
 981   arg.auto_restart = process->auto_restart;
 982   //kill the old process
 983   SIMIX_process_kill(process,issuer);
 984   //start the new process
 985   smx_process_t new_process;
 986   if (simix_global->create_process_function) {
 987     simix_global->create_process_function(&new_process,
 988                                           arg.argv[0],
 989                                           arg.code,
 990                                           arg.data,
 991                                           arg.hostname,
 992                                           arg.kill_time,
 993                                           arg.argc,
 994                                           arg.argv,
 995                                           arg.properties,
 996                                           arg.auto_restart,
 997                                           NULL);
 998   } else {
 999     simcall_process_create(&new_process,
1000                            arg.argv[0],
1001                            arg.code,
1002                            arg.data,
1003                            arg.hostname,
1004                            arg.kill_time,
1005                            arg.argc,
1006                            arg.argv,
1007                            arg.properties,
1008                            arg.auto_restart);
1009
1010   }
1011   return new_process;
1012 }