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