src/simix/smx_process.cpp

   1 /* Copyright (c) 2007-2015. 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 "src/mc/mc_replay.h"
  13 #include "src/mc/mc_client.h"
  14 #include "src/simix/smx_private.hpp"
  15
  16 XBT_LOG_NEW_DEFAULT_SUBCATEGORY(simix_process, simix,
  17                                 "Logging specific to SIMIX (process)");
  18
  19 unsigned long simix_process_maxpid = 0;
  20
  21 /**
  22  * \brief Returns the current agent.
  23  *
  24  * This functions returns the currently running SIMIX process.
  25  *
  26  * \return The SIMIX process
  27  */
  28 smx_process_t SIMIX_process_self(void)
  29 {
  30   smx_context_t self_context = SIMIX_context_self();
  31
  32   return self_context ? SIMIX_context_get_process(self_context) : NULL;
  33 }
  34
  35 /**
  36  * \brief Returns whether a process has pending asynchronous communications.
  37  * \return true if there are asynchronous communications in this process
  38  */
  39 int SIMIX_process_has_pending_comms(smx_process_t process) {
  40
  41   return xbt_fifo_size(process->comms) > 0;
  42 }
  43
  44 /**
  45  * \brief Moves a process to the list of processes to destroy.
  46  */
  47 void SIMIX_process_cleanup(smx_process_t process)
  48 {
  49   XBT_DEBUG("Cleanup process %s (%p), waiting synchro %p",
  50       process->name, process, process->waiting_synchro);
  51
  52   SIMIX_process_on_exit_runall(process);
  53
  54   /* Unregister from the kill timer if any */
  55   if (process->kill_timer != NULL)
  56           SIMIX_timer_remove(process->kill_timer);
  57
  58   xbt_os_mutex_acquire(simix_global->mutex);
  59
  60   /* cancel non-blocking communications */
  61   smx_synchro_t synchro;
  62   while ((synchro = (smx_synchro_t) xbt_fifo_pop(process->comms))) {
  63
  64     /* make sure no one will finish the comm after this process is destroyed,
  65      * because src_proc or dst_proc would be an invalid pointer */
  66     SIMIX_comm_cancel(synchro);
  67
  68     if (synchro->comm.src_proc == process) {
  69       XBT_DEBUG("Found an unfinished send comm %p (detached = %d), state %d, src = %p, dst = %p",
  70           synchro, synchro->comm.detached, (int)synchro->state, synchro->comm.src_proc, synchro->comm.dst_proc);
  71       synchro->comm.src_proc = NULL;
  72
  73       /* I'm not supposed to destroy a detached comm from the sender side, */
  74       if (!synchro->comm.detached)
  75         SIMIX_comm_destroy(synchro);
  76       else
  77         XBT_DEBUG("Don't destroy it since it's a detached comm");
  78
  79     }
  80     else if (synchro->comm.dst_proc == process){
  81       XBT_DEBUG("Found an unfinished recv comm %p, state %d, src = %p, dst = %p",
  82           synchro, (int)synchro->state, synchro->comm.src_proc, synchro->comm.dst_proc);
  83       synchro->comm.dst_proc = NULL;
  84
  85       if (synchro->comm.detached && synchro->comm.refcount == 1
  86           && synchro->comm.src_proc != NULL) {
  87         /* the comm will be freed right now, remove it from the sender */
  88         xbt_fifo_remove(synchro->comm.src_proc->comms, synchro);
  89       }
  90       SIMIX_comm_destroy(synchro);
  91     }
  92     else {
  93       xbt_die("Communication synchro %p is in my list but I'm not the sender "
  94           "or the receiver", synchro);
  95     }
  96   }
  97
  98   XBT_DEBUG("%p should not be run anymore",process);
  99   xbt_swag_remove(process, simix_global->process_list);
 100   xbt_swag_remove(process, sg_host_simix(process->host)->process_list);
 101   xbt_swag_insert(process, simix_global->process_to_destroy);
 102   process->context->iwannadie = 0;
 103
 104   xbt_os_mutex_release(simix_global->mutex);
 105 }
 106
 107 /**
 108  * Garbage collection
 109  *
 110  * Should be called some time to time to free the memory allocated for processes
 111  * that have finished (or killed).
 112  */
 113 void SIMIX_process_empty_trash(void)
 114 {
 115   smx_process_t process = NULL;
 116
 117   while ((process = (smx_process_t) xbt_swag_extract(simix_global->process_to_destroy))) {
 118     XBT_DEBUG("Getting rid of %p",process);
 119
 120     SIMIX_context_free(process->context);
 121
 122     /* Free the exception allocated at creation time */
 123     free(process->running_ctx);
 124     xbt_dict_free(&process->properties);
 125
 126     xbt_fifo_free(process->comms);
 127
 128     xbt_dynar_free(&process->on_exit);
 129
 130     xbt_free(process->name);
 131     xbt_free(process);
 132   }
 133 }
 134
 135 /**
 136  * \brief Creates and runs the maestro process
 137  */
 138 void SIMIX_create_maestro_process()
 139 {
 140   smx_process_t maestro = NULL;
 141
 142   /* Create maestro process and intilialize it */
 143   maestro = xbt_new0(s_smx_process_t, 1);
 144   maestro->pid = simix_process_maxpid++;
 145   maestro->ppid = -1;
 146   maestro->name = (char *) "";
 147   maestro->running_ctx = (xbt_running_ctx_t*) xbt_malloc0(sizeof(xbt_running_ctx_t));
 148   XBT_RUNNING_CTX_INITIALIZE(maestro->running_ctx);
 149   maestro->context = SIMIX_context_new(NULL, 0, NULL, NULL, maestro);
 150   maestro->simcall.issuer = maestro;
 151   simix_global->maestro_process = maestro;
 152   return;
 153 }
 154 /**
 155  * \brief Stops a process.
 156  *
 157  * Stops the process, execute all the registered on_exit functions,
 158  * register it to the list of the process to restart if needed
 159  * and stops its context.
 160  */
 161 void SIMIX_process_stop(smx_process_t arg) {
 162   /* execute the on_exit functions */
 163   SIMIX_process_on_exit_runall(arg);
 164   /* Add the process to the list of process to restart, only if
 165    * the host is down
 166    */
 167   if (arg->auto_restart && !sg_host_get_state(arg->host)) {
 168     SIMIX_host_add_auto_restart_process(arg->host,arg->name,arg->code, arg->data,
 169                                         sg_host_get_name(arg->host),
 170                                         SIMIX_timer_get_date(arg->kill_timer),
 171                                         arg->argc,arg->argv,arg->properties,
 172                                         arg->auto_restart);
 173   }
 174   XBT_DEBUG("Process %s (%s) is dead",arg->name,sg_host_get_name(arg->host));
 175   /* stop the context */
 176   SIMIX_context_stop(arg->context);
 177 }
 178
 179 /**
 180  * \brief Same as SIMIX_process_create() but with only one argument (used by timers).
 181  * This function frees the argument.
 182  * \return the process created
 183  */
 184 smx_process_t SIMIX_process_create_from_wrapper(smx_process_arg_t args) {
 185
 186   smx_process_t process = simix_global->create_process_function(
 187                                         args->name,
 188                                         args->code,
 189                                         args->data,
 190                                         args->hostname,
 191                                         args->kill_time,
 192                                         args->argc,
 193                                         args->argv,
 194                                         args->properties,
 195                                         args->auto_restart,
 196                                         NULL);
 197   xbt_free(args);
 198   return process;
 199 }
 200
 201
 202 void* simcall_HANDLER_process_create(smx_simcall_t simcall,
 203                           const char *name,
 204                           xbt_main_func_t code,
 205                           void *data,
 206                           const char *hostname,
 207                           double kill_time,
 208                           int argc, char **argv,
 209                           xbt_dict_t properties,
 210                           int auto_restart){
 211   return (void*)SIMIX_process_create(name, code, data, hostname,
 212                        kill_time, argc, argv, properties, auto_restart,
 213                        simcall->issuer);
 214 }
 215
 216 static void kill_process(void* process)
 217 {
 218   simix_global->kill_process_function((smx_process_t) process);
 219 }
 220
 221 /**
 222  * \brief Internal function to create a process.
 223  *
 224  * This function actually creates the process.
 225  * It may be called when a SIMCALL_PROCESS_CREATE simcall occurs,
 226  * or directly for SIMIX internal purposes. The sure thing is that it's called from maestro context.
 227  *
 228  * \return the process created
 229  */
 230 smx_process_t SIMIX_process_create(
 231                           const char *name,
 232                           xbt_main_func_t code,
 233                           void *data,
 234                           const char *hostname,
 235                           double kill_time,
 236                           int argc, char **argv,
 237                           xbt_dict_t properties,
 238                           int auto_restart,
 239                           smx_process_t parent_process)
 240 {
 241   smx_process_t process = NULL;
 242   sg_host_t host = sg_host_by_name(hostname);
 243
 244   XBT_DEBUG("Start process %s on host '%s'", name, hostname);
 245
 246   if (!sg_host_get_state(host)) {
 247     int i;
 248     XBT_WARN("Cannot launch process '%s' on failed host '%s'", name,
 249           hostname);
 250     for (i = 0; i < argc; i++)
 251       xbt_free(argv[i]);
 252     xbt_free(argv);
 253   }
 254   else {
 255     process = xbt_new0(s_smx_process_t, 1);
 256
 257     xbt_assert(((code != NULL) && (host != NULL)), "Invalid parameters");
 258     /* Process data */
 259     process->pid = simix_process_maxpid++;
 260     process->name = xbt_strdup(name);
 261     process->host = host;
 262     process->data = data;
 263     process->comms = xbt_fifo_new();
 264     process->simcall.issuer = process;
 265
 266      if (parent_process) {
 267        process->ppid = SIMIX_process_get_PID(parent_process);
 268      } else {
 269        process->ppid = -1;
 270      }
 271
 272     /* Process data for auto-restart */
 273     process->auto_restart = auto_restart;
 274     process->code = code;
 275     process->argc = argc;
 276     process->argv = argv;
 277
 278
 279     XBT_VERB("Create context %s", process->name);
 280     process->context = SIMIX_context_new(code, argc, argv, simix_global->cleanup_process_function, process);
 281
 282     process->running_ctx = (xbt_running_ctx_t*) xbt_malloc0(sizeof(xbt_running_ctx_t));
 283     XBT_RUNNING_CTX_INITIALIZE(process->running_ctx);
 284
 285     if(MC_is_active()){
 286       MC_ignore_heap(process->running_ctx, sizeof(*process->running_ctx));
 287     }
 288
 289     /* Add properties */
 290     process->properties = properties;
 291
 292     /* Add the process to it's host process list */
 293     xbt_swag_insert(process, sg_host_simix(host)->process_list);
 294
 295     XBT_DEBUG("Start context '%s'", process->name);
 296
 297     /* Now insert it in the global process list and in the process to run list */
 298     xbt_swag_insert(process, simix_global->process_list);
 299     XBT_DEBUG("Inserting %s(%s) in the to_run list", process->name, sg_host_get_name(host));
 300     xbt_dynar_push_as(simix_global->process_to_run, smx_process_t, process);
 301
 302     if (kill_time > SIMIX_get_clock() && simix_global->kill_process_function) {
 303       XBT_DEBUG("Process %s(%s) will be kill at time %f", process->name,
 304           sg_host_get_name(process->host), kill_time);
 305       process->kill_timer = SIMIX_timer_set(kill_time, kill_process, process);
 306     }
 307   }
 308   return process;
 309 }
 310
 311 /**
 312  * \brief Executes the processes from simix_global->process_to_run.
 313  *
 314  * The processes of simix_global->process_to_run are run (in parallel if
 315  * possible).  On exit, simix_global->process_to_run is empty, and
 316  * simix_global->process_that_ran contains the list of processes that just ran.
 317  * The two lists are swapped so, be careful when using them before and after a
 318  * call to this function.
 319  */
 320 void SIMIX_process_runall(void)
 321 {
 322   SIMIX_context_runall();
 323
 324   xbt_dynar_t tmp = simix_global->process_that_ran;
 325   simix_global->process_that_ran = simix_global->process_to_run;
 326   simix_global->process_to_run = tmp;
 327   xbt_dynar_reset(simix_global->process_to_run);
 328 }
 329
 330 void simcall_HANDLER_process_kill(smx_simcall_t simcall, smx_process_t process) {
 331   SIMIX_process_kill(process, simcall->issuer);
 332 }
 333 /**
 334  * \brief Internal function to kill a SIMIX process.
 335  *
 336  * This function may be called when a SIMCALL_PROCESS_KILL simcall occurs,
 337  * or directly for SIMIX internal purposes.
 338  *
 339  * \param process poor victim
 340  * \param issuer the process which has sent the PROCESS_KILL. Important to not schedule twice the same process.
 341  */
 342 void SIMIX_process_kill(smx_process_t process, smx_process_t issuer) {
 343
 344   XBT_DEBUG("Killing process %s on %s", process->name, sg_host_get_name(process->host));
 345
 346   process->context->iwannadie = 1;
 347   process->blocked = 0;
 348   process->suspended = 0;
 349   process->doexception = 0;
 350
 351   /* destroy the blocking synchro if any */
 352   if (process->waiting_synchro) {
 353
 354     switch (process->waiting_synchro->type) {
 355
 356     case SIMIX_SYNC_EXECUTE:
 357     case SIMIX_SYNC_PARALLEL_EXECUTE:
 358       SIMIX_process_execution_destroy(process->waiting_synchro);
 359       break;
 360
 361     case SIMIX_SYNC_COMMUNICATE:
 362       xbt_fifo_remove(process->comms, process->waiting_synchro);
 363       SIMIX_comm_cancel(process->waiting_synchro);
 364       xbt_fifo_remove(process->waiting_synchro->simcalls, &process->simcall);
 365       SIMIX_comm_destroy(process->waiting_synchro);
 366       break;
 367
 368     case SIMIX_SYNC_SLEEP:
 369       SIMIX_process_sleep_destroy(process->waiting_synchro);
 370       break;
 371
 372     case SIMIX_SYNC_JOIN:
 373       SIMIX_process_sleep_destroy(process->waiting_synchro);
 374       break;
 375
 376     case SIMIX_SYNC_SYNCHRO:
 377       SIMIX_synchro_stop_waiting(process, &process->simcall);
 378       SIMIX_synchro_destroy(process->waiting_synchro);
 379       break;
 380
 381     case SIMIX_SYNC_IO:
 382       SIMIX_io_destroy(process->waiting_synchro);
 383       break;
 384
 385     }
 386
 387     process->waiting_synchro = NULL;
 388   }
 389   if(!xbt_dynar_member(simix_global->process_to_run, &(process)) && process != issuer) {
 390     XBT_DEBUG("Inserting %s in the to_run list", process->name);
 391     xbt_dynar_push_as(simix_global->process_to_run, smx_process_t, process);
 392   }
 393
 394 }
 395
 396 /** @brief Ask another process to raise the given exception
 397  *
 398  * @param cat category of exception
 399  * @param value value associated to the exception
 400  * @param msg string information associated to the exception
 401  */
 402 void SIMIX_process_throw(smx_process_t process, xbt_errcat_t cat, int value, const char *msg) {
 403   SMX_EXCEPTION(process, cat, value, msg);
 404
 405   if (process->suspended)
 406     SIMIX_process_resume(process,SIMIX_process_self());
 407
 408   /* cancel the blocking synchro if any */
 409   if (process->waiting_synchro) {
 410
 411     switch (process->waiting_synchro->type) {
 412
 413     case SIMIX_SYNC_EXECUTE:
 414     case SIMIX_SYNC_PARALLEL_EXECUTE:
 415       SIMIX_process_execution_cancel(process->waiting_synchro);
 416       break;
 417
 418     case SIMIX_SYNC_COMMUNICATE:
 419       xbt_fifo_remove(process->comms, process->waiting_synchro);
 420       SIMIX_comm_cancel(process->waiting_synchro);
 421       break;
 422
 423     case SIMIX_SYNC_SLEEP:
 424     case SIMIX_SYNC_JOIN:
 425       SIMIX_process_sleep_destroy(process->waiting_synchro);
 426       if (!xbt_dynar_member(simix_global->process_to_run, &(process)) && process != SIMIX_process_self()) {
 427         XBT_DEBUG("Inserting %s in the to_run list", process->name);
 428         xbt_dynar_push_as(simix_global->process_to_run, smx_process_t, process);
 429       }
 430       break;
 431
 432     case SIMIX_SYNC_SYNCHRO:
 433       SIMIX_synchro_stop_waiting(process, &process->simcall);
 434       break;
 435
 436     case SIMIX_SYNC_IO:
 437       SIMIX_io_destroy(process->waiting_synchro);
 438       break;
 439
 440     }
 441   }
 442   process->waiting_synchro = NULL;
 443
 444 }
 445
 446 void simcall_HANDLER_process_killall(smx_simcall_t simcall, int reset_pid) {
 447   SIMIX_process_killall(simcall->issuer, reset_pid);
 448 }
 449 /**
 450  * \brief Kills all running processes.
 451  * \param issuer this one will not be killed
 452  */
 453 void SIMIX_process_killall(smx_process_t issuer, int reset_pid)
 454 {
 455   smx_process_t p = NULL;
 456
 457   while ((p = (smx_process_t) xbt_swag_extract(simix_global->process_list))) {
 458     if (p != issuer) {
 459       SIMIX_process_kill(p,issuer);
 460     }
 461   }
 462
 463   if (reset_pid > 0)
 464     simix_process_maxpid = reset_pid;
 465
 466   SIMIX_context_runall();
 467
 468   SIMIX_process_empty_trash();
 469 }
 470
 471 void simcall_HANDLER_process_set_host(smx_simcall_t simcall, smx_process_t process, sg_host_t dest)
 472 {
 473   process->new_host = dest;
 474 }
 475 void SIMIX_process_change_host(smx_process_t process,
 476              sg_host_t dest)
 477 {
 478   xbt_assert((process != NULL), "Invalid parameters");
 479   xbt_swag_remove(process, sg_host_simix(process->host)->process_list);
 480   process->host = dest;
 481   xbt_swag_insert(process, sg_host_simix(dest)->process_list);
 482 }
 483
 484
 485 void simcall_HANDLER_process_suspend(smx_simcall_t simcall, smx_process_t process)
 486 {
 487   smx_synchro_t sync_suspend =
 488       SIMIX_process_suspend(process, simcall->issuer);
 489
 490   if (process != simcall->issuer) {
 491     SIMIX_simcall_answer(simcall);
 492   } else {
 493     xbt_fifo_push(sync_suspend->simcalls, simcall);
 494     process->waiting_synchro = sync_suspend;
 495     SIMIX_host_execution_suspend(process->waiting_synchro);
 496   }
 497   /* If we are suspending ourselves, then just do not finish the simcall now */
 498 }
 499
 500 smx_synchro_t SIMIX_process_suspend(smx_process_t process, smx_process_t issuer)
 501 {
 502   xbt_assert((process != NULL), "Invalid parameters");
 503
 504   if (process->suspended) {
 505     XBT_DEBUG("Process '%s' is already suspended", process->name);
 506     return NULL;
 507   }
 508
 509   process->suspended = 1;
 510
 511   /* If we are suspending another process, and it is waiting on a sync,
 512      suspend its synchronization. */
 513   if (process != issuer) {
 514
 515     if (process->waiting_synchro) {
 516
 517       switch (process->waiting_synchro->type) {
 518
 519         case SIMIX_SYNC_EXECUTE:
 520         case SIMIX_SYNC_PARALLEL_EXECUTE:
 521           SIMIX_host_execution_suspend(process->waiting_synchro);
 522           break;
 523
 524         case SIMIX_SYNC_COMMUNICATE:
 525           SIMIX_comm_suspend(process->waiting_synchro);
 526           break;
 527
 528         case SIMIX_SYNC_SLEEP:
 529           SIMIX_process_sleep_suspend(process->waiting_synchro);
 530           break;
 531
 532         case SIMIX_SYNC_SYNCHRO:
 533           /* Suspension is delayed to when the process is rescheduled. */
 534           break;
 535
 536         default:
 537           xbt_die("Internal error in SIMIX_process_suspend: unexpected synchronization type %d",
 538               (int)process->waiting_synchro->type);
 539       }
 540       return NULL;
 541     } else {
 542       /* Suspension is delayed to when the process is rescheduled. */
 543       return NULL;
 544     }
 545   } else {
 546     /* FIXME: computation size is zero. Is it okay that bound is zero ? */
 547     return SIMIX_process_execute(process, "suspend", 0.0, 1.0, 0.0, 0);
 548   }
 549 }
 550
 551 void simcall_HANDLER_process_resume(smx_simcall_t simcall, smx_process_t process){
 552   SIMIX_process_resume(process, simcall->issuer);
 553 }
 554
 555 void SIMIX_process_resume(smx_process_t process, smx_process_t issuer)
 556 {
 557   XBT_IN("process = %p, issuer = %p", process, issuer);
 558
 559   if(process->context->iwannadie) {
 560     XBT_VERB("Ignoring request to suspend a process that is currently dying.");
 561     return;
 562   }
 563
 564   if(!process->suspended) return;
 565   process->suspended = 0;
 566
 567   /* If we are resuming another process, resume the synchronization it was waiting for
 568      if any. Otherwise add it to the list of process to run in the next round. */
 569   if (process != issuer) {
 570
 571     if (process->waiting_synchro) {
 572
 573       switch (process->waiting_synchro->type) {
 574
 575         case SIMIX_SYNC_EXECUTE:
 576         case SIMIX_SYNC_PARALLEL_EXECUTE:
 577           SIMIX_host_execution_resume(process->waiting_synchro);
 578           break;
 579
 580         case SIMIX_SYNC_COMMUNICATE:
 581           SIMIX_comm_resume(process->waiting_synchro);
 582           break;
 583
 584         case SIMIX_SYNC_SLEEP:
 585           SIMIX_process_sleep_resume(process->waiting_synchro);
 586           break;
 587
 588         case SIMIX_SYNC_SYNCHRO:
 589           /* I cannot resume it now. This is delayed to when the process is rescheduled at
 590            * the end of the synchro. */
 591           break;
 592
 593         default:
 594           xbt_die("Internal error in SIMIX_process_resume: unexpected synchronization type %d",
 595               (int)process->waiting_synchro->type);
 596       }
 597     }
 598   } else XBT_WARN("Strange. Process %p is trying to resume himself.", issuer);
 599
 600   XBT_OUT();
 601 }
 602
 603 int SIMIX_process_get_maxpid(void) {
 604   return simix_process_maxpid;
 605 }
 606
 607 int SIMIX_process_count(void)
 608 {
 609   return xbt_swag_size(simix_global->process_list);
 610 }
 611
 612 int SIMIX_process_get_PID(smx_process_t self){
 613   if (self == NULL)
 614     return 0;
 615   else
 616     return self->pid;
 617 }
 618
 619 int SIMIX_process_get_PPID(smx_process_t self){
 620   if (self == NULL)
 621     return 0;
 622   else
 623     return self->ppid;
 624 }
 625
 626 void* SIMIX_process_self_get_data(smx_process_t self)
 627 {
 628   xbt_assert(self == SIMIX_process_self(), "This is not the current process");
 629
 630   if (!self) {
 631     return NULL;
 632   }
 633   return SIMIX_process_get_data(self);
 634 }
 635
 636 void SIMIX_process_self_set_data(smx_process_t self, void *data)
 637 {
 638   xbt_assert(self == SIMIX_process_self(), "This is not the current process");
 639
 640   SIMIX_process_set_data(self, data);
 641 }
 642
 643 void* SIMIX_process_get_data(smx_process_t process)
 644 {
 645   return process->data;
 646 }
 647
 648 void SIMIX_process_set_data(smx_process_t process, void *data)
 649 {
 650   process->data = data;
 651 }
 652
 653 sg_host_t SIMIX_process_get_host(smx_process_t process)
 654 {
 655   return process->host;
 656 }
 657
 658 /* needs to be public and without simcall because it is called
 659    by exceptions and logging events */
 660 const char* SIMIX_process_self_get_name(void) {
 661
 662   smx_process_t process = SIMIX_process_self();
 663   if (process == NULL || process == simix_global->maestro_process)
 664     return "";
 665
 666   return SIMIX_process_get_name(process);
 667 }
 668
 669 const char* SIMIX_process_get_name(smx_process_t process)
 670 {
 671   return process->name;
 672 }
 673
 674 smx_process_t SIMIX_process_get_by_name(const char* name)
 675 {
 676   smx_process_t proc;
 677
 678   xbt_swag_foreach(proc, simix_global->process_list)
 679   {
 680     if(!strcmp(name, proc->name))
 681       return proc;
 682   }
 683   return NULL;
 684 }
 685
 686 int SIMIX_process_is_suspended(smx_process_t process)
 687 {
 688   return process->suspended;
 689 }
 690
 691 xbt_dict_t SIMIX_process_get_properties(smx_process_t process)
 692 {
 693   return process->properties;
 694 }
 695
 696 void simcall_HANDLER_process_join(smx_simcall_t simcall, smx_process_t process, double timeout)
 697 {
 698   smx_synchro_t sync = SIMIX_process_join(simcall->issuer, process, timeout);
 699   xbt_fifo_push(sync->simcalls, simcall);
 700   simcall->issuer->waiting_synchro = sync;
 701 }
 702
 703 static int SIMIX_process_join_finish(smx_process_exit_status_t status, smx_synchro_t sync){
 704   if (sync->sleep.surf_sleep) {
 705     surf_action_cancel(sync->sleep.surf_sleep);
 706
 707     smx_simcall_t simcall;
 708     while ((simcall = (smx_simcall_t) xbt_fifo_shift(sync->simcalls))) {
 709       simcall_process_sleep__set__result(simcall, SIMIX_DONE);
 710       simcall->issuer->waiting_synchro = NULL;
 711       if (simcall->issuer->suspended) {
 712         XBT_DEBUG("Wait! This process is suspended and can't wake up now.");
 713         simcall->issuer->suspended = 0;
 714         simcall_HANDLER_process_suspend(simcall, simcall->issuer);
 715       } else {
 716         SIMIX_simcall_answer(simcall);
 717       }
 718     }
 719     surf_action_unref(sync->sleep.surf_sleep);
 720     sync->sleep.surf_sleep = NULL;
 721   }
 722   xbt_mallocator_release(simix_global->synchro_mallocator, sync);
 723   return 0;
 724 }
 725
 726 smx_synchro_t SIMIX_process_join(smx_process_t issuer, smx_process_t process, double timeout)
 727 {
 728   smx_synchro_t res = SIMIX_process_sleep(issuer, timeout);
 729   res->type = SIMIX_SYNC_JOIN;
 730   SIMIX_process_on_exit(process, (int_f_pvoid_pvoid_t)SIMIX_process_join_finish, res);
 731   return res;
 732 }
 733
 734 void simcall_HANDLER_process_sleep(smx_simcall_t simcall, double duration)
 735 {
 736   if (MC_is_active() || MC_record_replay_is_active()) {
 737     MC_process_clock_add(simcall->issuer, duration);
 738     simcall_process_sleep__set__result(simcall, SIMIX_DONE);
 739     SIMIX_simcall_answer(simcall);
 740     return;
 741   }
 742   smx_synchro_t sync = SIMIX_process_sleep(simcall->issuer, duration);
 743   xbt_fifo_push(sync->simcalls, simcall);
 744   simcall->issuer->waiting_synchro = sync;
 745 }
 746
 747 smx_synchro_t SIMIX_process_sleep(smx_process_t process, double duration)
 748 {
 749   sg_host_t host = process->host;
 750
 751   /* check if the host is active */
 752   if (surf_host_get_state(surf_host_resource_priv(host)) != SURF_RESOURCE_ON) {
 753     THROWF(host_error, 0, "Host %s failed, you cannot call this function",
 754            sg_host_get_name(host));
 755   }
 756
 757   smx_synchro_t synchro = (smx_synchro_t) xbt_mallocator_get(simix_global->synchro_mallocator);
 758   synchro->type = SIMIX_SYNC_SLEEP;
 759   synchro->name = NULL;
 760   synchro->category = NULL;
 761
 762   synchro->sleep.host = host;
 763   synchro->sleep.surf_sleep = surf_host_sleep(host, duration);
 764
 765   surf_action_set_data(synchro->sleep.surf_sleep, synchro);
 766   XBT_DEBUG("Create sleep synchronization %p", synchro);
 767
 768   return synchro;
 769 }
 770
 771 void SIMIX_post_process_sleep(smx_synchro_t synchro)
 772 {
 773   smx_simcall_t simcall;
 774   e_smx_state_t state;
 775   xbt_assert(synchro->type == SIMIX_SYNC_SLEEP || synchro->type == SIMIX_SYNC_JOIN);
 776
 777   while ((simcall = (smx_simcall_t) xbt_fifo_shift(synchro->simcalls))) {
 778
 779     switch(surf_action_get_state(synchro->sleep.surf_sleep)){
 780       case SURF_ACTION_FAILED:
 781         simcall->issuer->context->iwannadie = 1;
 782         //SMX_EXCEPTION(simcall->issuer, host_error, 0, "Host failed");
 783         state = SIMIX_SRC_HOST_FAILURE;
 784         break;
 785
 786       case SURF_ACTION_DONE:
 787         state = SIMIX_DONE;
 788         break;
 789
 790       default:
 791         THROW_IMPOSSIBLE;
 792         break;
 793     }
 794     if (surf_host_get_state(surf_host_resource_priv(simcall->issuer->host)) != SURF_RESOURCE_ON) {
 795       simcall->issuer->context->iwannadie = 1;
 796     }
 797     simcall_process_sleep__set__result(simcall, state);
 798     simcall->issuer->waiting_synchro = NULL;
 799     if (simcall->issuer->suspended) {
 800       XBT_DEBUG("Wait! This process is suspended and can't wake up now.");
 801       simcall->issuer->suspended = 0;
 802       simcall_HANDLER_process_suspend(simcall, simcall->issuer);
 803     } else {
 804       SIMIX_simcall_answer(simcall);
 805     }
 806   }
 807
 808   SIMIX_process_sleep_destroy(synchro);
 809 }
 810
 811 void SIMIX_process_sleep_destroy(smx_synchro_t synchro)
 812 {
 813   XBT_DEBUG("Destroy synchro %p", synchro);
 814   xbt_assert(synchro->type == SIMIX_SYNC_SLEEP || synchro->type == SIMIX_SYNC_JOIN);
 815
 816   if (synchro->sleep.surf_sleep) {
 817     surf_action_unref(synchro->sleep.surf_sleep);
 818     synchro->sleep.surf_sleep = NULL;
 819   }
 820   if (synchro->type == SIMIX_SYNC_SLEEP)
 821     xbt_mallocator_release(simix_global->synchro_mallocator, synchro);
 822 }
 823
 824 void SIMIX_process_sleep_suspend(smx_synchro_t synchro)
 825 {
 826   xbt_assert(synchro->type == SIMIX_SYNC_SLEEP);
 827   surf_action_suspend(synchro->sleep.surf_sleep);
 828 }
 829
 830 void SIMIX_process_sleep_resume(smx_synchro_t synchro)
 831 {
 832   XBT_DEBUG("Synchro state is %d on process_sleep_resume.", synchro->state);
 833   xbt_assert(synchro->type == SIMIX_SYNC_SLEEP);
 834   surf_action_resume(synchro->sleep.surf_sleep);
 835 }
 836
 837 /**
 838  * \brief Calling this function makes the process to yield.
 839  *
 840  * Only the current process can call this function, giving back the control to
 841  * maestro.
 842  *
 843  * \param self the current process
 844  */
 845 void SIMIX_process_yield(smx_process_t self)
 846 {
 847   XBT_DEBUG("Yield process '%s'", self->name);
 848
 849   /* Go into sleep and return control to maestro */
 850   SIMIX_context_suspend(self->context);
 851
 852   /* Ok, maestro returned control to us */
 853   XBT_DEBUG("Control returned to me: '%s'", self->name);
 854
 855   if (self->new_host) {
 856     SIMIX_process_change_host(self, self->new_host);
 857     self->new_host = NULL;
 858   }
 859
 860   if (self->context->iwannadie){
 861     XBT_DEBUG("I wanna die!");
 862     SIMIX_process_stop(self);
 863   }
 864
 865   if (self->suspended) {
 866     XBT_DEBUG("Hey! I'm suspended.");
 867     xbt_assert(!self->doexception, "Gasp! This exception may be lost by subsequent calls.");
 868     self->suspended = 0;
 869     SIMIX_process_suspend(self, self);
 870   }
 871
 872   if (self->doexception) {
 873     XBT_DEBUG("Wait, maestro left me an exception");
 874     self->doexception = 0;
 875     SMX_THROW();
 876   }
 877
 878 }
 879
 880 /* callback: context fetching */
 881 xbt_running_ctx_t *SIMIX_process_get_running_context(void)
 882 {
 883   return SIMIX_process_self()->running_ctx;
 884 }
 885
 886 /* callback: termination */
 887 void SIMIX_process_exception_terminate(xbt_ex_t * e)
 888 {
 889   xbt_ex_display(e);
 890   xbt_abort();
 891 }
 892
 893 smx_context_t SIMIX_process_get_context(smx_process_t p) {
 894   return p->context;
 895 }
 896
 897 void SIMIX_process_set_context(smx_process_t p,smx_context_t c) {
 898   p->context = c;
 899 }
 900
 901 /**
 902  * \brief Returns the list of processes to run.
 903  */
 904 xbt_dynar_t SIMIX_process_get_runnable(void)
 905 {
 906   return simix_global->process_to_run;
 907 }
 908
 909 /**
 910  * \brief Returns the process from PID.
 911  */
 912 smx_process_t SIMIX_process_from_PID(int PID)
 913 {
 914   smx_process_t proc;
 915   xbt_swag_foreach(proc, simix_global->process_list) {
 916    if (proc->pid == (unsigned long) PID)
 917     return proc;
 918   }
 919   return NULL;
 920 }
 921
 922 /** @brief returns a dynar containg all currently existing processes */
 923 xbt_dynar_t SIMIX_processes_as_dynar(void) {
 924   smx_process_t proc;
 925   xbt_dynar_t res = xbt_dynar_new(sizeof(smx_process_t),NULL);
 926   xbt_swag_foreach(proc, simix_global->process_list) {
 927     xbt_dynar_push(res,&proc);
 928   }
 929   return res;
 930 }
 931
 932
 933 void SIMIX_process_on_exit_runall(smx_process_t process) {
 934   s_smx_process_exit_fun_t exit_fun;
 935   smx_process_exit_status_t exit_status = (process->context->iwannadie) ?
 936                                          SMX_EXIT_FAILURE : SMX_EXIT_SUCCESS;
 937   while (!xbt_dynar_is_empty(process->on_exit)) {
 938     exit_fun = xbt_dynar_pop_as(process->on_exit,s_smx_process_exit_fun_t);
 939     (exit_fun.fun)((void*)exit_status, exit_fun.arg);
 940   }
 941 }
 942
 943 void SIMIX_process_on_exit(smx_process_t process, int_f_pvoid_pvoid_t fun, void *data) {
 944   xbt_assert(process, "current process not found: are you in maestro context ?");
 945
 946   if (!process->on_exit) {
 947     process->on_exit = xbt_dynar_new(sizeof(s_smx_process_exit_fun_t), NULL);
 948   }
 949
 950   s_smx_process_exit_fun_t exit_fun = {fun, data};
 951
 952   xbt_dynar_push_as(process->on_exit,s_smx_process_exit_fun_t,exit_fun);
 953 }
 954
 955 /**
 956  * \brief Sets the auto-restart status of the process.
 957  * If set to 1, the process will be automatically restarted when its host
 958  * comes back.
 959  */
 960 void SIMIX_process_auto_restart_set(smx_process_t process, int auto_restart) {
 961   process->auto_restart = auto_restart;
 962 }
 963
 964 smx_process_t simcall_HANDLER_process_restart(smx_simcall_t simcall, smx_process_t process) {
 965   return SIMIX_process_restart(process, simcall->issuer);
 966 }
 967 /** @brief Restart a process, starting it again from the beginning. */
 968 smx_process_t SIMIX_process_restart(smx_process_t process, smx_process_t issuer) {
 969   XBT_DEBUG("Restarting process %s on %s", process->name, sg_host_get_name(process->host));
 970   //retrieve the arguments of the old process
 971   //FIXME: Factorize this with SIMIX_host_add_auto_restart_process ?
 972   s_smx_process_arg_t arg;
 973   arg.code = process->code;
 974   arg.hostname = sg_host_get_name(process->host);
 975   arg.kill_time = SIMIX_timer_get_date(process->kill_timer);
 976   arg.argc = process->argc;
 977   arg.data = process->data;
 978   int i;
 979   arg.argv = xbt_new(char*,process->argc + 1);
 980   for (i = 0; i < arg.argc; i++) {
 981     arg.argv[i] = xbt_strdup(process->argv[i]);
 982   }
 983   arg.argv[process->argc] = NULL;
 984   arg.properties = NULL;
 985   arg.auto_restart = process->auto_restart;
 986   //kill the old process
 987   SIMIX_process_kill(process,issuer);
 988   //start the new process
 989   smx_process_t new_process;
 990   if (simix_global->create_process_function) {
 991     new_process = simix_global->create_process_function(
 992                                           arg.argv[0],
 993                                           arg.code,
 994                                           arg.data,
 995                                           arg.hostname,
 996                                           arg.kill_time,
 997                                           arg.argc,
 998                                           arg.argv,
 999                                           arg.properties,
1000                                           arg.auto_restart,
1001                                           NULL);
1002   } else {
1003     new_process = simcall_process_create(
1004                            arg.argv[0],
1005                            arg.code,
1006                            arg.data,
1007                            arg.hostname,
1008                            arg.kill_time,
1009                            arg.argc,
1010                            arg.argv,
1011                            arg.properties,
1012                            arg.auto_restart);
1013
1014   }
1015   return new_process;
1016 }