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