src/simix/smx_process.c

   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 "mc/mc_replay.h"
  13 #include "mc/mc_client.h"
  14
  15 #ifdef HAVE_SMPI
  16 #include "smpi/private.h"
  17 #endif
  18
  19 XBT_LOG_NEW_DEFAULT_SUBCATEGORY(simix_process, simix,
  20                                 "Logging specific to SIMIX (process)");
  21
  22 unsigned long simix_process_maxpid = 0;
  23
  24 /**
  25  * \brief Returns the current agent.
  26  *
  27  * This functions returns the currently running SIMIX process.
  28  *
  29  * \return The SIMIX process
  30  */
  31 XBT_INLINE smx_process_t SIMIX_process_self(void)
  32 {
  33   smx_context_t self_context = SIMIX_context_self();
  34
  35   return self_context ? SIMIX_context_get_process(self_context) : NULL;
  36 }
  37
  38 /**
  39  * \brief Returns whether a process has pending asynchronous communications.
  40  * \return true if there are asynchronous communications in this process
  41  */
  42 int SIMIX_process_has_pending_comms(smx_process_t process) {
  43
  44   return xbt_fifo_size(process->comms) > 0;
  45 }
  46
  47 /**
  48  * \brief Moves a process to the list of processes to destroy.
  49  */
  50 void SIMIX_process_cleanup(smx_process_t process)
  51 {
  52   XBT_DEBUG("Cleanup process %s (%p), waiting synchro %p",
  53       process->name, process, process->waiting_synchro);
  54
  55   SIMIX_process_on_exit_runall(process);
  56
  57   /* Unregister from the kill timer if any */
  58   if (process->kill_timer != NULL)
  59           SIMIX_timer_remove(process->kill_timer);
  60
  61   xbt_os_mutex_acquire(simix_global->mutex);
  62
  63   /* cancel non-blocking communications */
  64   smx_synchro_t synchro;
  65   while ((synchro = xbt_fifo_pop(process->comms))) {
  66
  67     /* make sure no one will finish the comm after this process is destroyed,
  68      * because src_proc or dst_proc would be an invalid pointer */
  69     SIMIX_comm_cancel(synchro);
  70
  71     if (synchro->comm.src_proc == process) {
  72       XBT_DEBUG("Found an unfinished send comm %p (detached = %d), state %d, src = %p, dst = %p",
  73           synchro, synchro->comm.detached, (int)synchro->state, synchro->comm.src_proc, synchro->comm.dst_proc);
  74       synchro->comm.src_proc = NULL;
  75
  76       /* I'm not supposed to destroy a detached comm from the sender side, */
  77       if (!synchro->comm.detached)
  78         SIMIX_comm_destroy(synchro);
  79       else
  80         XBT_DEBUG("Don't destroy it since it's a detached comm");
  81
  82     }
  83     else if (synchro->comm.dst_proc == process){
  84       XBT_DEBUG("Found an unfinished recv comm %p, state %d, src = %p, dst = %p",
  85           synchro, (int)synchro->state, synchro->comm.src_proc, synchro->comm.dst_proc);
  86       synchro->comm.dst_proc = NULL;
  87
  88       if (synchro->comm.detached && synchro->comm.refcount == 1
  89           && synchro->comm.src_proc != NULL) {
  90         /* the comm will be freed right now, remove it from the sender */
  91         xbt_fifo_remove(synchro->comm.src_proc->comms, synchro);
  92       }
  93       SIMIX_comm_destroy(synchro);
  94     }
  95     else {
  96       xbt_die("Communication synchro %p is in my list but I'm not the sender "
  97           "or the receiver", synchro);
  98     }
  99   }
 100
 101   XBT_DEBUG("%p should not be run anymore",process);
 102   xbt_swag_remove(process, simix_global->process_list);
 103   xbt_swag_remove(process, sg_host_simix(process->host)->process_list);
 104   xbt_swag_insert(process, simix_global->process_to_destroy);
 105   process->context->iwannadie = 0;
 106
 107   xbt_os_mutex_release(simix_global->mutex);
 108 }
 109
 110 /**
 111  * Garbage collection
 112  *
 113  * Should be called some time to time to free the memory allocated for processes
 114  * that have finished (or killed).
 115  */
 116 void SIMIX_process_empty_trash(void)
 117 {
 118   smx_process_t process = NULL;
 119
 120   while ((process = xbt_swag_extract(simix_global->process_to_destroy))) {
 121     XBT_DEBUG("Getting rid of %p",process);
 122
 123     SIMIX_context_free(process->context);
 124
 125     /* Free the exception allocated at creation time */
 126     free(process->running_ctx);
 127     xbt_dict_free(&process->properties);
 128
 129     xbt_fifo_free(process->comms);
 130
 131     xbt_dynar_free(&process->on_exit);
 132
 133     xbt_free(process->name);
 134     xbt_free(process);
 135   }
 136 }
 137
 138 /**
 139  * \brief Creates and runs the maestro process
 140  */
 141 void SIMIX_create_maestro_process()
 142 {
 143   smx_process_t maestro = NULL;
 144
 145   /* Create maestro process and intilialize it */
 146   maestro = xbt_new0(s_smx_process_t, 1);
 147   maestro->pid = simix_process_maxpid++;
 148   maestro->ppid = -1;
 149   maestro->name = (char *) "";
 150   maestro->running_ctx = xbt_new(xbt_running_ctx_t, 1);
 151   XBT_RUNNING_CTX_INITIALIZE(maestro->running_ctx);
 152   maestro->context = SIMIX_context_new(NULL, 0, NULL, NULL, maestro);
 153   maestro->simcall.issuer = maestro;
 154   simix_global->maestro_process = maestro;
 155   return;
 156 }
 157 /**
 158  * \brief Stops a process.
 159  *
 160  * Stops the process, execute all the registered on_exit functions,
 161  * register it to the list of the process to restart if needed
 162  * and stops its context.
 163  */
 164 void SIMIX_process_stop(smx_process_t arg) {
 165   /* execute the on_exit functions */
 166   SIMIX_process_on_exit_runall(arg);
 167   /* Add the process to the list of process to restart, only if
 168    * the host is down
 169    */
 170   if (arg->auto_restart && !SIMIX_host_get_state(arg->host)) {
 171     SIMIX_host_add_auto_restart_process(arg->host,arg->name,arg->code, arg->data,
 172                                         sg_host_name(arg->host),
 173                                         SIMIX_timer_get_date(arg->kill_timer),
 174                                         arg->argc,arg->argv,arg->properties,
 175                                         arg->auto_restart);
 176   }
 177   XBT_DEBUG("Process %s (%s) is dead",arg->name,sg_host_name(arg->host));
 178   /* stop the context */
 179   SIMIX_context_stop(arg->context);
 180 }
 181
 182 /**
 183  * \brief Same as SIMIX_process_create() but with only one argument (used by timers).
 184  * This function frees the argument.
 185  * \return the process created
 186  */
 187 smx_process_t SIMIX_process_create_from_wrapper(smx_process_arg_t args) {
 188
 189   smx_process_t process = simix_global->create_process_function(
 190                                         args->name,
 191                                         args->code,
 192                                         args->data,
 193                                         args->hostname,
 194                                         args->kill_time,
 195                                         args->argc,
 196                                         args->argv,
 197                                         args->properties,
 198                                         args->auto_restart,
 199                                         NULL);
 200   xbt_free(args);
 201   return process;
 202 }
 203
 204
 205 void* simcall_HANDLER_process_create(smx_simcall_t simcall,
 206                           const char *name,
 207                           xbt_main_func_t code,
 208                           void *data,
 209                           const char *hostname,
 210                           double kill_time,
 211                           int argc, char **argv,
 212                           xbt_dict_t properties,
 213                           int auto_restart){
 214   return (void*)SIMIX_process_create(name, code, data, hostname,
 215                        kill_time, argc, argv, properties, auto_restart,
 216                        simcall->issuer);
 217 }
 218 /**
 219  * \brief Internal function to create a process.
 220  *
 221  * This function actually creates the process.
 222  * It may be called when a SIMCALL_PROCESS_CREATE simcall occurs,
 223  * or directly for SIMIX internal purposes. The sure thing is that it's called from maestro context.
 224  *
 225  * \return the process created
 226  */
 227 smx_process_t SIMIX_process_create(
 228                           const char *name,
 229                           xbt_main_func_t code,
 230                           void *data,
 231                           const char *hostname,
 232                           double kill_time,
 233                           int argc, char **argv,
 234                           xbt_dict_t properties,
 235                           int auto_restart,
 236                           smx_process_t parent_process)
 237 {
 238   smx_process_t process = NULL;
 239   sg_host_t host = sg_host_by_name(hostname);
 240
 241   XBT_DEBUG("Start process %s on host '%s'", name, hostname);
 242
 243   if (!SIMIX_host_get_state(host)) {
 244     int i;
 245     XBT_WARN("Cannot launch process '%s' on failed host '%s'", name,
 246           hostname);
 247     for (i = 0; i < argc; i++)
 248       xbt_free(argv[i]);
 249     xbt_free(argv);
 250   }
 251   else {
 252     process = xbt_new0(s_smx_process_t, 1);
 253
 254     xbt_assert(((code != NULL) && (host != NULL)), "Invalid parameters");
 255     /* Process data */
 256     process->pid = simix_process_maxpid++;
 257     process->name = xbt_strdup(name);
 258     process->host = host;
 259     process->data = data;
 260     process->comms = xbt_fifo_new();
 261     process->simcall.issuer = process;
 262     /* Initiliaze data segment to default value */
 263     SIMIX_segment_index_set(process, -1);
 264
 265      if (parent_process != NULL) {
 266        process->ppid = SIMIX_process_get_PID(parent_process);
 267        /* SMPI process have their own data segment and
 268           each other inherit from their father */
 269        if(smpi_privatize_global_variables){
 270          if( parent_process->pid != 0){
 271            SIMIX_segment_index_set(process, parent_process->segment_index);
 272          } else {
 273            SIMIX_segment_index_set(process, process->pid - 1);
 274          }
 275        }
 276      } else {
 277        process->ppid = -1;
 278      }
 279
 280     /* Process data for auto-restart */
 281     process->auto_restart = auto_restart;
 282     process->code = code;
 283     process->argc = argc;
 284     process->argv = argv;
 285
 286
 287     XBT_VERB("Create context %s", process->name);
 288     process->context = SIMIX_context_new(code, argc, argv, simix_global->cleanup_process_function, process);
 289
 290     process->running_ctx = xbt_new(xbt_running_ctx_t, 1);
 291     XBT_RUNNING_CTX_INITIALIZE(process->running_ctx);
 292
 293     if(MC_is_active()){
 294       MC_ignore_heap(process->running_ctx, sizeof(*process->running_ctx));
 295     }
 296
 297     /* Add properties */
 298     process->properties = properties;
 299
 300     /* Add the process to it's host process list */
 301     xbt_swag_insert(process, sg_host_simix(host)->process_list);
 302
 303     XBT_DEBUG("Start context '%s'", process->name);
 304
 305     /* Now insert it in the global process list and in the process to run list */
 306     xbt_swag_insert(process, simix_global->process_list);
 307     XBT_DEBUG("Inserting %s(%s) in the to_run list", process->name, sg_host_name(host));
 308     xbt_dynar_push_as(simix_global->process_to_run, smx_process_t, process);
 309
 310     if (kill_time > SIMIX_get_clock() && simix_global->kill_process_function) {
 311       XBT_DEBUG("Process %s(%s) will be kill at time %f", process->name,
 312           sg_host_name(process->host), kill_time);
 313       process->kill_timer = SIMIX_timer_set(kill_time, simix_global->kill_process_function, process);
 314     }
 315   }
 316   return process;
 317 }
 318
 319 /**
 320  * \brief Executes the processes from simix_global->process_to_run.
 321  *
 322  * The processes of simix_global->process_to_run are run (in parallel if
 323  * possible).  On exit, simix_global->process_to_run is empty, and
 324  * simix_global->process_that_ran contains the list of processes that just ran.
 325  * The two lists are swapped so, be careful when using them before and after a
 326  * call to this function.
 327  */
 328 void SIMIX_process_runall(void)
 329 {
 330   SIMIX_context_runall();
 331
 332   xbt_dynar_t tmp = simix_global->process_that_ran;
 333   simix_global->process_that_ran = simix_global->process_to_run;
 334   simix_global->process_to_run = tmp;
 335   xbt_dynar_reset(simix_global->process_to_run);
 336 }
 337
 338 void simcall_HANDLER_process_kill(smx_simcall_t simcall, smx_process_t process) {
 339   SIMIX_process_kill(process, simcall->issuer);
 340 }
 341 /**
 342  * \brief Internal function to kill a SIMIX process.
 343  *
 344  * This function may be called when a SIMCALL_PROCESS_KILL simcall occurs,
 345  * or directly for SIMIX internal purposes.
 346  *
 347  * \param process poor victim
 348  * \param issuer the process which has sent the PROCESS_KILL. Important to not schedule twice the same process.
 349  */
 350 void SIMIX_process_kill(smx_process_t process, smx_process_t issuer) {
 351
 352   XBT_DEBUG("Killing process %s on %s", process->name, sg_host_name(process->host));
 353
 354   process->context->iwannadie = 1;
 355   process->blocked = 0;
 356   process->suspended = 0;
 357   process->doexception = 0;
 358
 359   /* destroy the blocking synchro if any */
 360   if (process->waiting_synchro) {
 361
 362     switch (process->waiting_synchro->type) {
 363
 364     case SIMIX_SYNC_EXECUTE:
 365     case SIMIX_SYNC_PARALLEL_EXECUTE:
 366       SIMIX_process_execution_destroy(process->waiting_synchro);
 367       break;
 368
 369     case SIMIX_SYNC_COMMUNICATE:
 370       xbt_fifo_remove(process->comms, process->waiting_synchro);
 371       SIMIX_comm_cancel(process->waiting_synchro);
 372       xbt_fifo_remove(process->waiting_synchro->simcalls, &process->simcall);
 373       SIMIX_comm_destroy(process->waiting_synchro);
 374       break;
 375
 376     case SIMIX_SYNC_SLEEP:
 377       SIMIX_process_sleep_destroy(process->waiting_synchro);
 378       break;
 379
 380     case SIMIX_SYNC_JOIN:
 381       SIMIX_process_sleep_destroy(process->waiting_synchro);
 382       break;
 383
 384     case SIMIX_SYNC_SYNCHRO:
 385       SIMIX_synchro_stop_waiting(process, &process->simcall);
 386       SIMIX_synchro_destroy(process->waiting_synchro);
 387       break;
 388
 389     case SIMIX_SYNC_IO:
 390       SIMIX_io_destroy(process->waiting_synchro);
 391       break;
 392
 393     }
 394
 395     process->waiting_synchro = NULL;
 396   }
 397   if(!xbt_dynar_member(simix_global->process_to_run, &(process)) && process != issuer) {
 398     XBT_DEBUG("Inserting %s in the to_run list", process->name);
 399     xbt_dynar_push_as(simix_global->process_to_run, smx_process_t, process);
 400   }
 401
 402 }
 403
 404 /** @brief Ask another process to raise the given exception
 405  *
 406  * @param cat category of exception
 407  * @param value value associated to the exception
 408  * @param msg string information associated to the exception
 409  */
 410 void SIMIX_process_throw(smx_process_t process, xbt_errcat_t cat, int value, const char *msg) {
 411   SMX_EXCEPTION(process, cat, value, msg);
 412
 413   if (process->suspended)
 414     SIMIX_process_resume(process,SIMIX_process_self());
 415
 416   /* cancel the blocking synchro if any */
 417   if (process->waiting_synchro) {
 418
 419     switch (process->waiting_synchro->type) {
 420
 421     case SIMIX_SYNC_EXECUTE:
 422     case SIMIX_SYNC_PARALLEL_EXECUTE:
 423       SIMIX_process_execution_cancel(process->waiting_synchro);
 424       break;
 425
 426     case SIMIX_SYNC_COMMUNICATE:
 427       xbt_fifo_remove(process->comms, process->waiting_synchro);
 428       SIMIX_comm_cancel(process->waiting_synchro);
 429       break;
 430
 431     case SIMIX_SYNC_SLEEP:
 432     case SIMIX_SYNC_JOIN:
 433       SIMIX_process_sleep_destroy(process->waiting_synchro);
 434       if (!xbt_dynar_member(simix_global->process_to_run, &(process)) && process != SIMIX_process_self()) {
 435         XBT_DEBUG("Inserting %s in the to_run list", process->name);
 436         xbt_dynar_push_as(simix_global->process_to_run, smx_process_t, process);
 437       }
 438       break;
 439
 440     case SIMIX_SYNC_SYNCHRO:
 441       SIMIX_synchro_stop_waiting(process, &process->simcall);
 442       break;
 443
 444     case SIMIX_SYNC_IO:
 445       SIMIX_io_destroy(process->waiting_synchro);
 446       break;
 447
 448     }
 449   }
 450   process->waiting_synchro = NULL;
 451
 452 }
 453
 454 void simcall_HANDLER_process_killall(smx_simcall_t simcall, int reset_pid) {
 455   SIMIX_process_killall(simcall->issuer, reset_pid);
 456 }
 457 /**
 458  * \brief Kills all running processes.
 459  * \param issuer this one will not be killed
 460  */
 461 void SIMIX_process_killall(smx_process_t issuer, int reset_pid)
 462 {
 463   smx_process_t p = NULL;
 464
 465   while ((p = xbt_swag_extract(simix_global->process_list))) {
 466     if (p != issuer) {
 467       SIMIX_process_kill(p,issuer);
 468     }
 469   }
 470
 471   if (reset_pid > 0)
 472     simix_process_maxpid = reset_pid;
 473
 474   SIMIX_context_runall();
 475
 476   SIMIX_process_empty_trash();
 477 }
 478
 479 void simcall_HANDLER_process_set_host(smx_simcall_t simcall, smx_process_t process, sg_host_t dest)
 480 {
 481   process->new_host = dest;
 482 }
 483 void SIMIX_process_change_host(smx_process_t process,
 484              sg_host_t dest)
 485 {
 486   xbt_assert((process != NULL), "Invalid parameters");
 487   xbt_swag_remove(process, sg_host_simix(process->host)->process_list);
 488   process->host = dest;
 489   xbt_swag_insert(process, sg_host_simix(dest)->process_list);
 490 }
 491
 492
 493 void simcall_HANDLER_process_suspend(smx_simcall_t simcall, smx_process_t process)
 494 {
 495   smx_synchro_t sync_suspend =
 496       SIMIX_process_suspend(process, simcall->issuer);
 497
 498   if (process != simcall->issuer) {
 499     SIMIX_simcall_answer(simcall);
 500   } else {
 501     xbt_fifo_push(sync_suspend->simcalls, simcall);
 502     process->waiting_synchro = sync_suspend;
 503     SIMIX_host_execution_suspend(process->waiting_synchro);
 504   }
 505   /* If we are suspending ourselves, then just do not finish the simcall now */
 506 }
 507
 508 smx_synchro_t SIMIX_process_suspend(smx_process_t process, smx_process_t issuer)
 509 {
 510   xbt_assert((process != NULL), "Invalid parameters");
 511
 512   if (process->suspended) {
 513     XBT_DEBUG("Process '%s' is already suspended", process->name);
 514     return NULL;
 515   }
 516
 517   process->suspended = 1;
 518
 519   /* If we are suspending another process, and it is waiting on a sync,
 520      suspend its synchronization. */
 521   if (process != issuer) {
 522
 523     if (process->waiting_synchro) {
 524
 525       switch (process->waiting_synchro->type) {
 526
 527         case SIMIX_SYNC_EXECUTE:
 528         case SIMIX_SYNC_PARALLEL_EXECUTE:
 529           SIMIX_host_execution_suspend(process->waiting_synchro);
 530           break;
 531
 532         case SIMIX_SYNC_COMMUNICATE:
 533           SIMIX_comm_suspend(process->waiting_synchro);
 534           break;
 535
 536         case SIMIX_SYNC_SLEEP:
 537           SIMIX_process_sleep_suspend(process->waiting_synchro);
 538           break;
 539
 540         case SIMIX_SYNC_SYNCHRO:
 541           /* Suspension is delayed to when the process is rescheduled. */
 542           break;
 543
 544         default:
 545           xbt_die("Internal error in SIMIX_process_suspend: unexpected synchronization type %d",
 546               (int)process->waiting_synchro->type);
 547       }
 548       return NULL;
 549     } else {
 550       /* Suspension is delayed to when the process is rescheduled. */
 551       return NULL;
 552     }
 553   } else {
 554     /* FIXME: computation size is zero. Is it okay that bound is zero ? */
 555     return SIMIX_process_execute(process, "suspend", 0.0, 1.0, 0.0, 0);
 556   }
 557 }
 558
 559 void simcall_HANDLER_process_resume(smx_simcall_t simcall, smx_process_t process){
 560   SIMIX_process_resume(process, simcall->issuer);
 561 }
 562
 563 void SIMIX_process_resume(smx_process_t process, smx_process_t issuer)
 564 {
 565   XBT_IN("process = %p, issuer = %p", process, issuer);
 566
 567   if(process->context->iwannadie) {
 568     XBT_VERB("Ignoring request to suspend a process that is currently dying.");
 569     return;
 570   }
 571
 572   if(!process->suspended) return;
 573   process->suspended = 0;
 574
 575   /* If we are resuming another process, resume the synchronization it was waiting for
 576      if any. Otherwise add it to the list of process to run in the next round. */
 577   if (process != issuer) {
 578
 579     if (process->waiting_synchro) {
 580
 581       switch (process->waiting_synchro->type) {
 582
 583         case SIMIX_SYNC_EXECUTE:
 584         case SIMIX_SYNC_PARALLEL_EXECUTE:
 585           SIMIX_host_execution_resume(process->waiting_synchro);
 586           break;
 587
 588         case SIMIX_SYNC_COMMUNICATE:
 589           SIMIX_comm_resume(process->waiting_synchro);
 590           break;
 591
 592         case SIMIX_SYNC_SLEEP:
 593           SIMIX_process_sleep_resume(process->waiting_synchro);
 594           break;
 595
 596         case SIMIX_SYNC_SYNCHRO:
 597           /* I cannot resume it now. This is delayed to when the process is rescheduled at
 598            * the end of the synchro. */
 599           break;
 600
 601         default:
 602           xbt_die("Internal error in SIMIX_process_resume: unexpected synchronization type %d",
 603               (int)process->waiting_synchro->type);
 604       }
 605     }
 606   } else XBT_WARN("Strange. Process %p is trying to resume himself.", issuer);
 607
 608   XBT_OUT();
 609 }
 610
 611 int SIMIX_process_get_maxpid(void) {
 612   return simix_process_maxpid;
 613 }
 614
 615 int SIMIX_process_count(void)
 616 {
 617   return xbt_swag_size(simix_global->process_list);
 618 }
 619
 620 int SIMIX_process_get_PID(smx_process_t self){
 621   if (self == NULL)
 622     return 0;
 623   else
 624     return self->pid;
 625 }
 626
 627 int SIMIX_process_get_PPID(smx_process_t self){
 628   if (self == NULL)
 629     return 0;
 630   else
 631     return self->ppid;
 632 }
 633
 634 void* SIMIX_process_self_get_data(smx_process_t self)
 635 {
 636   xbt_assert(self == SIMIX_process_self(), "This is not the current process");
 637
 638   if (!self) {
 639     return NULL;
 640   }
 641   return SIMIX_process_get_data(self);
 642 }
 643
 644 void SIMIX_process_self_set_data(smx_process_t self, void *data)
 645 {
 646   xbt_assert(self == SIMIX_process_self(), "This is not the current process");
 647
 648   SIMIX_process_set_data(self, data);
 649 }
 650
 651 void* SIMIX_process_get_data(smx_process_t process)
 652 {
 653   return process->data;
 654 }
 655
 656 void SIMIX_process_set_data(smx_process_t process, void *data)
 657 {
 658   process->data = data;
 659 }
 660
 661 sg_host_t SIMIX_process_get_host(smx_process_t process)
 662 {
 663   return process->host;
 664 }
 665
 666 xbt_main_func_t SIMIX_process_get_code(void){
 667   return SIMIX_process_self()->code;
 668 }
 669
 670 /* needs to be public and without simcall because it is called
 671    by exceptions and logging events */
 672 const char* SIMIX_process_self_get_name(void) {
 673
 674   smx_process_t process = SIMIX_process_self();
 675   if (process == NULL || process == simix_global->maestro_process)
 676     return "";
 677
 678   return SIMIX_process_get_name(process);
 679 }
 680
 681 const char* SIMIX_process_get_name(smx_process_t process)
 682 {
 683   return process->name;
 684 }
 685
 686 smx_process_t SIMIX_process_get_by_name(const char* name)
 687 {
 688   smx_process_t proc;
 689
 690   xbt_swag_foreach(proc, simix_global->process_list)
 691   {
 692     if(!strcmp(name, proc->name))
 693       return proc;
 694   }
 695   return NULL;
 696 }
 697
 698 int SIMIX_process_is_suspended(smx_process_t process)
 699 {
 700   return process->suspended;
 701 }
 702
 703 xbt_dict_t SIMIX_process_get_properties(smx_process_t process)
 704 {
 705   return process->properties;
 706 }
 707
 708 void simcall_HANDLER_process_join(smx_simcall_t simcall, smx_process_t process, double timeout)
 709 {
 710   smx_synchro_t sync = SIMIX_process_join(simcall->issuer, process, timeout);
 711   xbt_fifo_push(sync->simcalls, simcall);
 712   simcall->issuer->waiting_synchro = sync;
 713 }
 714
 715 static int SIMIX_process_join_finish(smx_process_exit_status_t status, smx_synchro_t sync){
 716   if (sync->sleep.surf_sleep) {
 717     surf_action_cancel(sync->sleep.surf_sleep);
 718
 719     smx_simcall_t simcall;
 720     while ((simcall = xbt_fifo_shift(sync->simcalls))) {
 721       simcall_process_sleep__set__result(simcall, SIMIX_DONE);
 722       simcall->issuer->waiting_synchro = NULL;
 723       if (simcall->issuer->suspended) {
 724         XBT_DEBUG("Wait! This process is suspended and can't wake up now.");
 725         simcall->issuer->suspended = 0;
 726         simcall_HANDLER_process_suspend(simcall, simcall->issuer);
 727       } else {
 728         SIMIX_simcall_answer(simcall);
 729       }
 730     }
 731     surf_action_unref(sync->sleep.surf_sleep);
 732     sync->sleep.surf_sleep = NULL;
 733   }
 734   xbt_mallocator_release(simix_global->synchro_mallocator, sync);
 735   return 0;
 736 }
 737
 738 smx_synchro_t SIMIX_process_join(smx_process_t issuer, smx_process_t process, double timeout)
 739 {
 740   smx_synchro_t res = SIMIX_process_sleep(issuer, timeout);
 741   res->type = SIMIX_SYNC_JOIN;
 742   SIMIX_process_on_exit(process, (int_f_pvoid_pvoid_t)SIMIX_process_join_finish, res);
 743   return res;
 744 }
 745
 746 void simcall_HANDLER_process_sleep(smx_simcall_t simcall, double duration)
 747 {
 748   if (MC_is_active() || MC_record_replay_is_active()) {
 749     MC_process_clock_add(simcall->issuer, duration);
 750     simcall_process_sleep__set__result(simcall, SIMIX_DONE);
 751     SIMIX_simcall_answer(simcall);
 752     return;
 753   }
 754   smx_synchro_t sync = SIMIX_process_sleep(simcall->issuer, duration);
 755   xbt_fifo_push(sync->simcalls, simcall);
 756   simcall->issuer->waiting_synchro = sync;
 757 }
 758
 759 smx_synchro_t SIMIX_process_sleep(smx_process_t process, double duration)
 760 {
 761   smx_synchro_t synchro;
 762   sg_host_t host = process->host;
 763
 764   /* check if the host is active */
 765   if (surf_host_get_state(surf_host_resource_priv(host)) != SURF_RESOURCE_ON) {
 766     THROWF(host_error, 0, "Host %s failed, you cannot call this function",
 767            sg_host_name(host));
 768   }
 769
 770   synchro = xbt_mallocator_get(simix_global->synchro_mallocator);
 771   synchro->type = SIMIX_SYNC_SLEEP;
 772   synchro->name = NULL;
 773   synchro->category = NULL;
 774
 775   synchro->sleep.host = host;
 776   synchro->sleep.surf_sleep = surf_host_sleep(host, duration);
 777
 778   surf_action_set_data(synchro->sleep.surf_sleep, synchro);
 779   XBT_DEBUG("Create sleep synchronization %p", synchro);
 780
 781   return synchro;
 782 }
 783
 784 void SIMIX_post_process_sleep(smx_synchro_t synchro)
 785 {
 786   smx_simcall_t simcall;
 787   e_smx_state_t state;
 788   xbt_assert(synchro->type == SIMIX_SYNC_SLEEP || synchro->type == SIMIX_SYNC_JOIN);
 789
 790   while ((simcall = xbt_fifo_shift(synchro->simcalls))) {
 791
 792     switch(surf_action_get_state(synchro->sleep.surf_sleep)){
 793       case SURF_ACTION_FAILED:
 794         simcall->issuer->context->iwannadie = 1;
 795         //SMX_EXCEPTION(simcall->issuer, host_error, 0, "Host failed");
 796         state = SIMIX_SRC_HOST_FAILURE;
 797         break;
 798
 799       case SURF_ACTION_DONE:
 800         state = SIMIX_DONE;
 801         break;
 802
 803       default:
 804         THROW_IMPOSSIBLE;
 805         break;
 806     }
 807     if (surf_host_get_state(surf_host_resource_priv(simcall->issuer->host)) != SURF_RESOURCE_ON) {
 808       simcall->issuer->context->iwannadie = 1;
 809     }
 810     simcall_process_sleep__set__result(simcall, state);
 811     simcall->issuer->waiting_synchro = NULL;
 812     if (simcall->issuer->suspended) {
 813       XBT_DEBUG("Wait! This process is suspended and can't wake up now.");
 814       simcall->issuer->suspended = 0;
 815       simcall_HANDLER_process_suspend(simcall, simcall->issuer);
 816     } else {
 817       SIMIX_simcall_answer(simcall);
 818     }
 819   }
 820
 821   SIMIX_process_sleep_destroy(synchro);
 822 }
 823
 824 void SIMIX_process_sleep_destroy(smx_synchro_t synchro)
 825 {
 826   XBT_DEBUG("Destroy synchro %p", synchro);
 827   xbt_assert(synchro->type == SIMIX_SYNC_SLEEP || synchro->type == SIMIX_SYNC_JOIN);
 828
 829   if (synchro->sleep.surf_sleep) {
 830     surf_action_unref(synchro->sleep.surf_sleep);
 831     synchro->sleep.surf_sleep = NULL;
 832   }
 833   if (synchro->type == SIMIX_SYNC_SLEEP)
 834     xbt_mallocator_release(simix_global->synchro_mallocator, synchro);
 835 }
 836
 837 void SIMIX_process_sleep_suspend(smx_synchro_t synchro)
 838 {
 839   xbt_assert(synchro->type == SIMIX_SYNC_SLEEP);
 840   surf_action_suspend(synchro->sleep.surf_sleep);
 841 }
 842
 843 void SIMIX_process_sleep_resume(smx_synchro_t synchro)
 844 {
 845   XBT_DEBUG("Synchro state is %d on process_sleep_resume.", synchro->state);
 846   xbt_assert(synchro->type == SIMIX_SYNC_SLEEP);
 847   surf_action_resume(synchro->sleep.surf_sleep);
 848 }
 849
 850 /**
 851  * \brief Calling this function makes the process to yield.
 852  *
 853  * Only the current process can call this function, giving back the control to
 854  * maestro.
 855  *
 856  * \param self the current process
 857  */
 858 void SIMIX_process_yield(smx_process_t self)
 859 {
 860   XBT_DEBUG("Yield process '%s'", self->name);
 861
 862   /* Go into sleep and return control to maestro */
 863   SIMIX_context_suspend(self->context);
 864
 865   /* Ok, maestro returned control to us */
 866   XBT_DEBUG("Control returned to me: '%s'", self->name);
 867
 868   if (self->new_host) {
 869     SIMIX_process_change_host(self, self->new_host);
 870     self->new_host = NULL;
 871   }
 872
 873   if (self->context->iwannadie){
 874     XBT_DEBUG("I wanna die!");
 875     SIMIX_process_stop(self);
 876   }
 877
 878   if (self->suspended) {
 879     XBT_DEBUG("Hey! I'm suspended.");
 880     xbt_assert(!self->doexception, "Gasp! This exception may be lost by subsequent calls.");
 881     self->suspended = 0;
 882     SIMIX_process_suspend(self, self);
 883   }
 884
 885   if (self->doexception) {
 886     XBT_DEBUG("Wait, maestro left me an exception");
 887     self->doexception = 0;
 888     SMX_THROW();
 889   }
 890
 891   if(SMPI_switch_data_segment && self->segment_index != -1){
 892     SMPI_switch_data_segment(self->segment_index);
 893   }
 894 }
 895
 896 /* callback: context fetching */
 897 xbt_running_ctx_t *SIMIX_process_get_running_context(void)
 898 {
 899   return SIMIX_process_self()->running_ctx;
 900 }
 901
 902 /* callback: termination */
 903 void SIMIX_process_exception_terminate(xbt_ex_t * e)
 904 {
 905   xbt_ex_display(e);
 906   xbt_abort();
 907 }
 908
 909 smx_context_t SIMIX_process_get_context(smx_process_t p) {
 910   return p->context;
 911 }
 912
 913 void SIMIX_process_set_context(smx_process_t p,smx_context_t c) {
 914   p->context = c;
 915 }
 916
 917 /**
 918  * \brief Returns the list of processes to run.
 919  */
 920 xbt_dynar_t SIMIX_process_get_runnable(void)
 921 {
 922   return simix_global->process_to_run;
 923 }
 924
 925 /**
 926  * \brief Returns the process from PID.
 927  */
 928 smx_process_t SIMIX_process_from_PID(int PID)
 929 {
 930   smx_process_t proc;
 931   xbt_swag_foreach(proc, simix_global->process_list)
 932   {
 933    if(proc->pid == PID)
 934    return proc;
 935   }
 936   return NULL;
 937 }
 938
 939 /** @brief returns a dynar containg all currently existing processes */
 940 xbt_dynar_t SIMIX_processes_as_dynar(void) {
 941   smx_process_t proc;
 942   xbt_dynar_t res = xbt_dynar_new(sizeof(smx_process_t),NULL);
 943   xbt_swag_foreach(proc, simix_global->process_list) {
 944     xbt_dynar_push(res,&proc);
 945   }
 946   return res;
 947 }
 948
 949
 950 void SIMIX_process_on_exit_runall(smx_process_t process) {
 951   s_smx_process_exit_fun_t exit_fun;
 952   smx_process_exit_status_t exit_status = (process->context->iwannadie) ?
 953                                          SMX_EXIT_FAILURE : SMX_EXIT_SUCCESS;
 954   while (!xbt_dynar_is_empty(process->on_exit)) {
 955     exit_fun = xbt_dynar_pop_as(process->on_exit,s_smx_process_exit_fun_t);
 956     (exit_fun.fun)((void*)exit_status, exit_fun.arg);
 957   }
 958 }
 959
 960 void SIMIX_process_on_exit(smx_process_t process, int_f_pvoid_pvoid_t fun, void *data) {
 961   xbt_assert(process, "current process not found: are you in maestro context ?");
 962
 963   if (!process->on_exit) {
 964     process->on_exit = xbt_dynar_new(sizeof(s_smx_process_exit_fun_t), NULL);
 965   }
 966
 967   s_smx_process_exit_fun_t exit_fun = {fun, data};
 968
 969   xbt_dynar_push_as(process->on_exit,s_smx_process_exit_fun_t,exit_fun);
 970 }
 971
 972 /**
 973  * \brief Sets the auto-restart status of the process.
 974  * If set to 1, the process will be automatically restarted when its host
 975  * comes back.
 976  */
 977 void SIMIX_process_auto_restart_set(smx_process_t process, int auto_restart) {
 978   process->auto_restart = auto_restart;
 979 }
 980
 981 smx_process_t simcall_HANDLER_process_restart(smx_simcall_t simcall, smx_process_t process) {
 982   return SIMIX_process_restart(process, simcall->issuer);
 983 }
 984 /** @brief Restart a process, starting it again from the beginning. */
 985 smx_process_t SIMIX_process_restart(smx_process_t process, smx_process_t issuer) {
 986   XBT_DEBUG("Restarting process %s on %s", process->name, sg_host_name(process->host));
 987   //retrieve the arguments of the old process
 988   //FIXME: Factorize this with SIMIX_host_add_auto_restart_process ?
 989   s_smx_process_arg_t arg;
 990   arg.code = process->code;
 991   arg.hostname = sg_host_name(process->host);
 992   arg.kill_time = SIMIX_timer_get_date(process->kill_timer);
 993   arg.argc = process->argc;
 994   arg.data = process->data;
 995   int i;
 996   arg.argv = xbt_new(char*,process->argc + 1);
 997   for (i = 0; i < arg.argc; i++) {
 998     arg.argv[i] = xbt_strdup(process->argv[i]);
 999   }
1000   arg.argv[process->argc] = NULL;
1001   arg.properties = NULL;
1002   arg.auto_restart = process->auto_restart;
1003   //kill the old process
1004   SIMIX_process_kill(process,issuer);
1005   //start the new process
1006   smx_process_t new_process;
1007   if (simix_global->create_process_function) {
1008     new_process = simix_global->create_process_function(
1009                                           arg.argv[0],
1010                                           arg.code,
1011                                           arg.data,
1012                                           arg.hostname,
1013                                           arg.kill_time,
1014                                           arg.argc,
1015                                           arg.argv,
1016                                           arg.properties,
1017                                           arg.auto_restart,
1018                                           NULL);
1019   } else {
1020     new_process = simcall_process_create(
1021                            arg.argv[0],
1022                            arg.code,
1023                            arg.data,
1024                            arg.hostname,
1025                            arg.kill_time,
1026                            arg.argc,
1027                            arg.argv,
1028                            arg.properties,
1029                            arg.auto_restart);
1030
1031   }
1032   return new_process;
1033 }
1034
1035 void SIMIX_segment_index_set(smx_process_t proc, int index){
1036   proc->segment_index = index;
1037 }