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