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