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