src/simix/smx_process.c

   1 /* Copyright (c) 2007-2014. The SimGrid Team.
   2  * All rights reserved.                                                     */
   3
   4 /* This program is free software; you can redistribute it and/or modify it
   5  * under the terms of the license (GNU LGPL) which comes with this package. */
   6
   7 #include "smx_private.h"
   8 #include "xbt/sysdep.h"
   9 #include "xbt/log.h"
  10 #include "xbt/dict.h"
  11 #include "mc/mc.h"
  12 #include "mc/mc_replay.h"
  13 #include "mc/mc_client.h"
  14
  15 XBT_LOG_NEW_DEFAULT_SUBCATEGORY(simix_process, simix,
  16                                 "Logging specific to SIMIX (process)");
  17
  18 unsigned long simix_process_maxpid = 0;
  19
  20 /**
  21  * \brief Returns the current agent.
  22  *
  23  * This functions returns the currently running SIMIX process.
  24  *
  25  * \return The SIMIX process
  26  */
  27 XBT_INLINE smx_process_t SIMIX_process_self(void)
  28 {
  29   smx_context_t self_context = SIMIX_context_self();
  30
  31   return self_context ? SIMIX_context_get_process(self_context) : NULL;
  32 }
  33
  34 /**
  35  * \brief Returns whether a process has pending asynchronous communications.
  36  * \return true if there are asynchronous communications in this process
  37  */
  38 int SIMIX_process_has_pending_comms(smx_process_t process) {
  39
  40   return xbt_fifo_size(process->comms) > 0;
  41 }
  42
  43 /**
  44  * \brief Moves a process to the list of processes to destroy.
  45  */
  46 void SIMIX_process_cleanup(smx_process_t process)
  47 {
  48   XBT_DEBUG("Cleanup process %s (%p), waiting synchro %p",
  49       process->name, process, process->waiting_synchro);
  50
  51   SIMIX_process_on_exit_runall(process);
  52
  53   /* cancel non-blocking communications */
  54   smx_synchro_t synchro;
  55   while ((synchro = xbt_fifo_pop(process->comms))) {
  56
  57     /* make sure no one will finish the comm after this process is destroyed,
  58      * because src_proc or dst_proc would be an invalid pointer */
  59     SIMIX_comm_cancel(synchro);
  60
  61     if (synchro->comm.src_proc == process) {
  62       XBT_DEBUG("Found an unfinished send comm %p (detached = %d), state %d, src = %p, dst = %p",
  63           synchro, synchro->comm.detached, (int)synchro->state, synchro->comm.src_proc, synchro->comm.dst_proc);
  64       synchro->comm.src_proc = NULL;
  65
  66       /* I'm not supposed to destroy a detached comm from the sender side, */
  67       if (!synchro->comm.detached)
  68         SIMIX_comm_destroy(synchro);
  69       else
  70         XBT_DEBUG("Don't destroy it since it's a detached comm");
  71
  72     }
  73     else if (synchro->comm.dst_proc == process){
  74       XBT_DEBUG("Found an unfinished recv comm %p, state %d, src = %p, dst = %p",
  75           synchro, (int)synchro->state, synchro->comm.src_proc, synchro->comm.dst_proc);
  76       synchro->comm.dst_proc = NULL;
  77
  78       if (synchro->comm.detached && synchro->comm.refcount == 1
  79           && synchro->comm.src_proc != NULL) {
  80         /* the comm will be freed right now, remove it from the sender */
  81         xbt_fifo_remove(synchro->comm.src_proc->comms, synchro);
  82       }
  83       SIMIX_comm_destroy(synchro);
  84     }
  85     else {
  86       xbt_die("Communication synchro %p is in my list but I'm not the sender "
  87           "or the receiver", synchro);
  88     }
  89   }
  90
  91   XBT_DEBUG("%p should not be run anymore",process);
  92   xbt_swag_remove(process, simix_global->process_list);
  93   xbt_swag_remove(process, SIMIX_host_priv(process->smx_host)->process_list);
  94   xbt_swag_insert(process, simix_global->process_to_destroy);
  95   process->context->iwannadie = 0;
  96 }
  97
  98 /**
  99  * Garbage collection
 100  *
 101  * Should be called some time to time to free the memory allocated for processes
 102  * that have finished (or killed).
 103  */
 104 void SIMIX_process_empty_trash(void)
 105 {
 106   smx_process_t process = NULL;
 107
 108   while ((process = xbt_swag_extract(simix_global->process_to_destroy))) {
 109     XBT_DEBUG("Getting rid of %p",process);
 110
 111     SIMIX_context_free(process->context);
 112
 113     /* Free the exception allocated at creation time */
 114     free(process->running_ctx);
 115     xbt_dict_free(&process->properties);
 116
 117     xbt_fifo_free(process->comms);
 118
 119     xbt_dynar_free(&process->on_exit);
 120
 121     xbt_free(process->name);
 122     xbt_free(process);
 123   }
 124 }
 125
 126 /**
 127  * \brief Creates and runs the maestro process
 128  */
 129 void SIMIX_create_maestro_process()
 130 {
 131   smx_process_t maestro = NULL;
 132
 133   /* Create maestro process and intilialize it */
 134   maestro = xbt_new0(s_smx_process_t, 1);
 135   maestro->pid = simix_process_maxpid++;
 136   maestro->ppid = -1;
 137   maestro->name = (char *) "";
 138   maestro->running_ctx = xbt_new(xbt_running_ctx_t, 1);
 139   XBT_RUNNING_CTX_INITIALIZE(maestro->running_ctx);
 140   maestro->context = SIMIX_context_new(NULL, 0, NULL, NULL, maestro);
 141   maestro->simcall.issuer = maestro;
 142   simix_global->maestro_process = maestro;
 143   return;
 144 }
 145 /**
 146  * \brief Stops a process.
 147  *
 148  * Stops the process, execute all the registered on_exit functions,
 149  * register it to the list of the process to restart if needed
 150  * and stops its context.
 151  */
 152 void SIMIX_process_stop(smx_process_t arg) {
 153   /* execute the on_exit functions */
 154   SIMIX_process_on_exit_runall(arg);
 155   /* Add the process to the list of process to restart, only if
 156    * the host is down
 157    */
 158   if (arg->auto_restart && !SIMIX_host_get_state(arg->smx_host)) {
 159     SIMIX_host_add_auto_restart_process(arg->smx_host,arg->name,arg->code, arg->data,
 160                                         sg_host_name(arg->smx_host),
 161                                         arg->kill_time,
 162                                         arg->argc,arg->argv,arg->properties,
 163                                         arg->auto_restart);
 164   }
 165   XBT_DEBUG("Process %s (%s) is dead",arg->name,sg_host_name(arg->smx_host));
 166   /* stop the context */
 167   SIMIX_context_stop(arg->context);
 168 }
 169
 170 /**
 171  * \brief Same as SIMIX_process_create() but with only one argument (used by timers).
 172  * This function frees the argument.
 173  * \return the process created
 174  */
 175 smx_process_t SIMIX_process_create_from_wrapper(smx_process_arg_t args) {
 176
 177   smx_process_t process;
 178   simix_global->create_process_function(&process,
 179                                         args->name,
 180                                         args->code,
 181                                         args->data,
 182                                         args->hostname,
 183                                         args->kill_time,
 184                                         args->argc,
 185                                         args->argv,
 186                                         args->properties,
 187                                         args->auto_restart,
 188                                         NULL);
 189   xbt_free(args);
 190   return process;
 191 }
 192
 193
 194 void simcall_HANDLER_process_create(smx_simcall_t simcall,
 195                           smx_process_t *process,
 196                           const char *name,
 197                           xbt_main_func_t code,
 198                           void *data,
 199                           const char *hostname,
 200                           double kill_time,
 201                           int argc, char **argv,
 202                           xbt_dict_t properties,
 203                           int auto_restart){
 204   SIMIX_process_create(process, name, code, data, hostname,
 205                        kill_time, argc, argv, properties, auto_restart,
 206                        simcall->issuer);
 207 }
 208 /**
 209  * \brief Internal function to create a process.
 210  *
 211  * This function actually creates the process.
 212  * It may be called when a SIMCALL_PROCESS_CREATE simcall occurs,
 213  * or directly for SIMIX internal purposes. The sure thing is that it's called from maestro context.
 214  *
 215  * \return the process created
 216  */
 217 void SIMIX_process_create(smx_process_t *process,
 218                           const char *name,
 219                           xbt_main_func_t code,
 220                           void *data,
 221                           const char *hostname,
 222                           double kill_time,
 223                           int argc, char **argv,
 224                           xbt_dict_t properties,
 225                           int auto_restart,
 226                           smx_process_t parent_process)
 227 {
 228   *process = NULL;
 229   smx_host_t host = SIMIX_host_get_by_name(hostname);
 230
 231   XBT_DEBUG("Start process %s on host '%s'", name, hostname);
 232
 233   if (!SIMIX_host_get_state(host)) {
 234     int i;
 235     XBT_WARN("Cannot launch process '%s' on failed host '%s'", name,
 236           hostname);
 237     for (i = 0; i < argc; i++)
 238       xbt_free(argv[i]);
 239     xbt_free(argv);
 240   }
 241   else {
 242     *process = xbt_new0(s_smx_process_t, 1);
 243
 244     xbt_assert(((code != NULL) && (host != NULL)), "Invalid parameters");
 245     /* Process data */
 246     (*process)->pid = simix_process_maxpid++;
 247     (*process)->name = xbt_strdup(name);
 248     (*process)->smx_host = host;
 249     (*process)->data = data;
 250     (*process)->comms = xbt_fifo_new();
 251     (*process)->simcall.issuer = *process;
 252
 253      if (parent_process) {
 254        (*process)->ppid = SIMIX_process_get_PID(parent_process);
 255      } else {
 256        (*process)->ppid = -1;
 257      }
 258
 259     /* Process data for auto-restart */
 260     (*process)->auto_restart = auto_restart;
 261     (*process)->code = code;
 262     (*process)->argc = argc;
 263     (*process)->argv = argv;
 264     (*process)->kill_time = kill_time;
 265
 266
 267     XBT_VERB("Create context %s", (*process)->name);
 268     (*process)->context = SIMIX_context_new(code, argc, argv,
 269       simix_global->cleanup_process_function, *process);
 270
 271     (*process)->running_ctx = xbt_new(xbt_running_ctx_t, 1);
 272     XBT_RUNNING_CTX_INITIALIZE((*process)->running_ctx);
 273
 274     if(MC_is_active()){
 275       MC_ignore_heap((*process)->running_ctx, sizeof(*(*process)->running_ctx));
 276     }
 277
 278     /* Add properties */
 279     (*process)->properties = properties;
 280
 281     /* Add the process to it's host process list */
 282     xbt_swag_insert(*process, SIMIX_host_priv(host)->process_list);
 283
 284     XBT_DEBUG("Start context '%s'", (*process)->name);
 285
 286     /* Now insert it in the global process list and in the process to run list */
 287     xbt_swag_insert(*process, simix_global->process_list);
 288     XBT_DEBUG("Inserting %s(%s) in the to_run list", (*process)->name, sg_host_name(host));
 289     xbt_dynar_push_as(simix_global->process_to_run, smx_process_t, *process);
 290
 291     if (kill_time > SIMIX_get_clock() && simix_global->kill_process_function) {
 292       XBT_DEBUG("Process %s(%s) will be kill at time %f", (*process)->name,
 293           sg_host_name((*process)->smx_host), kill_time);
 294       SIMIX_timer_set(kill_time, simix_global->kill_process_function, *process);
 295     }
 296   }
 297 }
 298
 299 /**
 300  * \brief Executes the processes from simix_global->process_to_run.
 301  *
 302  * The processes of simix_global->process_to_run are run (in parallel if
 303  * possible).  On exit, simix_global->process_to_run is empty, and
 304  * simix_global->process_that_ran contains the list of processes that just ran.
 305  * The two lists are swapped so, be careful when using them before and after a
 306  * call to this function.
 307  */
 308 void SIMIX_process_runall(void)
 309 {
 310   SIMIX_context_runall();
 311
 312   xbt_dynar_t tmp = simix_global->process_that_ran;
 313   simix_global->process_that_ran = simix_global->process_to_run;
 314   simix_global->process_to_run = tmp;
 315   xbt_dynar_reset(simix_global->process_to_run);
 316 }
 317
 318 void simcall_HANDLER_process_kill(smx_simcall_t simcall, smx_process_t process) {
 319   SIMIX_process_kill(process, simcall->issuer);
 320 }
 321 /**
 322  * \brief Internal function to kill a SIMIX process.
 323  *
 324  * This function may be called when a SIMCALL_PROCESS_KILL simcall occurs,
 325  * or directly for SIMIX internal purposes.
 326  *
 327  * \param process poor victim
 328  * \param issuer the process which has sent the PROCESS_KILL. Important to not schedule twice the same process.
 329  */
 330 void SIMIX_process_kill(smx_process_t process, smx_process_t issuer) {
 331
 332   XBT_DEBUG("Killing process %s on %s", process->name, sg_host_name(process->smx_host));
 333
 334   process->context->iwannadie = 1;
 335   process->blocked = 0;
 336   process->suspended = 0;
 337   process->doexception = 0;
 338
 339   /* destroy the blocking synchro if any */
 340   if (process->waiting_synchro) {
 341
 342     switch (process->waiting_synchro->type) {
 343
 344     case SIMIX_SYNC_EXECUTE:
 345     case SIMIX_SYNC_PARALLEL_EXECUTE:
 346       SIMIX_host_execution_destroy(process->waiting_synchro);
 347       break;
 348
 349     case SIMIX_SYNC_COMMUNICATE:
 350       xbt_fifo_remove(process->comms, process->waiting_synchro);
 351       SIMIX_comm_cancel(process->waiting_synchro);
 352       SIMIX_comm_destroy(process->waiting_synchro);
 353       break;
 354
 355     case SIMIX_SYNC_SLEEP:
 356       SIMIX_process_sleep_destroy(process->waiting_synchro);
 357       break;
 358
 359     case SIMIX_SYNC_JOIN:
 360       SIMIX_process_sleep_destroy(process->waiting_synchro);
 361       break;
 362
 363     case SIMIX_SYNC_SYNCHRO:
 364       SIMIX_synchro_stop_waiting(process, &process->simcall);
 365       SIMIX_synchro_destroy(process->waiting_synchro);
 366       break;
 367
 368     case SIMIX_SYNC_IO:
 369       SIMIX_io_destroy(process->waiting_synchro);
 370       break;
 371
 372     }
 373   }
 374   if(!xbt_dynar_member(simix_global->process_to_run, &(process)) && process != issuer) {
 375     XBT_DEBUG("Inserting %s in the to_run list", process->name);
 376     xbt_dynar_push_as(simix_global->process_to_run, smx_process_t, process);
 377   }
 378
 379 }
 380
 381 /** @brief Ask another process to raise the given exception
 382  *
 383  * @param cat category of exception
 384  * @param value value associated to the exception
 385  * @param msg string information associated to the exception
 386  */
 387 void SIMIX_process_throw(smx_process_t process, xbt_errcat_t cat, int value, const char *msg) {
 388   SMX_EXCEPTION(process, cat, value, msg);
 389
 390   if (process->suspended)
 391     SIMIX_process_resume(process,SIMIX_process_self());
 392
 393   /* cancel the blocking synchro if any */
 394   if (process->waiting_synchro) {
 395
 396     switch (process->waiting_synchro->type) {
 397
 398     case SIMIX_SYNC_EXECUTE:
 399     case SIMIX_SYNC_PARALLEL_EXECUTE:
 400       SIMIX_host_execution_cancel(process->waiting_synchro);
 401       break;
 402
 403     case SIMIX_SYNC_COMMUNICATE:
 404       xbt_fifo_remove(process->comms, process->waiting_synchro);
 405       SIMIX_comm_cancel(process->waiting_synchro);
 406       break;
 407
 408     case SIMIX_SYNC_SLEEP:
 409     case SIMIX_SYNC_JOIN:
 410       SIMIX_process_sleep_destroy(process->waiting_synchro);
 411       if (!xbt_dynar_member(simix_global->process_to_run, &(process)) && process != SIMIX_process_self()) {
 412         XBT_DEBUG("Inserting %s in the to_run list", process->name);
 413         xbt_dynar_push_as(simix_global->process_to_run, smx_process_t, process);
 414       }
 415       break;
 416
 417     case SIMIX_SYNC_SYNCHRO:
 418       SIMIX_synchro_stop_waiting(process, &process->simcall);
 419       break;
 420
 421     case SIMIX_SYNC_IO:
 422       SIMIX_io_destroy(process->waiting_synchro);
 423       break;
 424
 425     }
 426   }
 427   process->waiting_synchro = NULL;
 428
 429 }
 430
 431 void simcall_HANDLER_process_killall(smx_simcall_t simcall, int reset_pid) {
 432   SIMIX_process_killall(simcall->issuer, reset_pid);
 433 }
 434 /**
 435  * \brief Kills all running processes.
 436  * \param issuer this one will not be killed
 437  */
 438 void SIMIX_process_killall(smx_process_t issuer, int reset_pid)
 439 {
 440   smx_process_t p = NULL;
 441
 442   while ((p = xbt_swag_extract(simix_global->process_list))) {
 443     if (p != issuer) {
 444       SIMIX_process_kill(p,issuer);
 445     }
 446   }
 447
 448   if (reset_pid > 0)
 449     simix_process_maxpid = reset_pid;
 450
 451   SIMIX_context_runall();
 452
 453   SIMIX_process_empty_trash();
 454 }
 455
 456 void simcall_HANDLER_process_change_host(smx_simcall_t simcall, smx_process_t process,
 457                                    smx_host_t dest)
 458 {
 459   process->new_host = dest;
 460 }
 461 void SIMIX_process_change_host(smx_process_t process,
 462              smx_host_t dest)
 463 {
 464   xbt_assert((process != NULL), "Invalid parameters");
 465   xbt_swag_remove(process, SIMIX_host_priv(process->smx_host)->process_list);
 466   process->smx_host = dest;
 467   xbt_swag_insert(process, SIMIX_host_priv(dest)->process_list);
 468 }
 469
 470
 471 void simcall_HANDLER_process_suspend(smx_simcall_t simcall, smx_process_t process)
 472 {
 473   smx_synchro_t sync_suspend =
 474       SIMIX_process_suspend(process, simcall->issuer);
 475
 476   if (process != simcall->issuer) {
 477     SIMIX_simcall_answer(simcall);
 478   } else {
 479     xbt_fifo_push(sync_suspend->simcalls, simcall);
 480     process->waiting_synchro = sync_suspend;
 481     SIMIX_host_execution_suspend(process->waiting_synchro);
 482   }
 483   /* If we are suspending ourselves, then just do not finish the simcall now */
 484 }
 485
 486 smx_synchro_t SIMIX_process_suspend(smx_process_t process, smx_process_t issuer)
 487 {
 488   xbt_assert((process != NULL), "Invalid parameters");
 489
 490   if (process->suspended) {
 491     XBT_DEBUG("Process '%s' is already suspended", process->name);
 492     return NULL;
 493   }
 494
 495   process->suspended = 1;
 496
 497   /* If we are suspending another process, and it is waiting on a sync,
 498      suspend its synchronization. */
 499   if (process != issuer) {
 500
 501     if (process->waiting_synchro) {
 502
 503       switch (process->waiting_synchro->type) {
 504
 505         case SIMIX_SYNC_EXECUTE:
 506         case SIMIX_SYNC_PARALLEL_EXECUTE:
 507           SIMIX_host_execution_suspend(process->waiting_synchro);
 508           break;
 509
 510         case SIMIX_SYNC_COMMUNICATE:
 511           SIMIX_comm_suspend(process->waiting_synchro);
 512           break;
 513
 514         case SIMIX_SYNC_SLEEP:
 515           SIMIX_process_sleep_suspend(process->waiting_synchro);
 516           break;
 517
 518         case SIMIX_SYNC_SYNCHRO:
 519           /* Suspension is delayed to when the process is rescheduled. */
 520           break;
 521
 522         default:
 523           xbt_die("Internal error in SIMIX_process_suspend: unexpected synchronization type %d",
 524               (int)process->waiting_synchro->type);
 525       }
 526       return NULL;
 527     } else {
 528       /* Suspension is delayed to when the process is rescheduled. */
 529       return NULL;
 530     }
 531   } else {
 532     /* FIXME: computation size is zero. Is it okay that bound is zero ? */
 533     return SIMIX_host_execute("suspend", process->smx_host, 0.0, 1.0, 0.0, 0);
 534   }
 535 }
 536
 537 void simcall_HANDLER_process_resume(smx_simcall_t simcall, smx_process_t process){
 538   SIMIX_process_resume(process, simcall->issuer);
 539 }
 540
 541 void SIMIX_process_resume(smx_process_t process, smx_process_t issuer)
 542 {
 543   XBT_IN("process = %p, issuer = %p", process, issuer);
 544
 545   if(process->context->iwannadie) {
 546     XBT_VERB("Ignoring request to suspend a process that is currently dying.");
 547     return;
 548   }
 549
 550   if(!process->suspended) return;
 551   process->suspended = 0;
 552
 553   /* If we are resuming another process, resume the synchronization it was waiting for
 554      if any. Otherwise add it to the list of process to run in the next round. */
 555   if (process != issuer) {
 556
 557     if (process->waiting_synchro) {
 558
 559       switch (process->waiting_synchro->type) {
 560
 561         case SIMIX_SYNC_EXECUTE:
 562         case SIMIX_SYNC_PARALLEL_EXECUTE:
 563           SIMIX_host_execution_resume(process->waiting_synchro);
 564           break;
 565
 566         case SIMIX_SYNC_COMMUNICATE:
 567           SIMIX_comm_resume(process->waiting_synchro);
 568           break;
 569
 570         case SIMIX_SYNC_SLEEP:
 571           SIMIX_process_sleep_resume(process->waiting_synchro);
 572           break;
 573
 574         case SIMIX_SYNC_SYNCHRO:
 575           /* I cannot resume it now. This is delayed to when the process is rescheduled at
 576            * the end of the synchro. */
 577           break;
 578
 579         default:
 580           xbt_die("Internal error in SIMIX_process_resume: unexpected synchronization type %d",
 581               (int)process->waiting_synchro->type);
 582       }
 583     }
 584   } else XBT_WARN("Strange. Process %p is trying to resume himself.", issuer);
 585
 586   XBT_OUT();
 587 }
 588
 589 int SIMIX_process_get_maxpid(void) {
 590   return simix_process_maxpid;
 591 }
 592
 593 int SIMIX_process_count(void)
 594 {
 595   return xbt_swag_size(simix_global->process_list);
 596 }
 597
 598 int SIMIX_process_get_PID(smx_process_t self){
 599   if (self == NULL)
 600     return 0;
 601   else
 602     return self->pid;
 603 }
 604
 605 int SIMIX_process_get_PPID(smx_process_t self){
 606   if (self == NULL)
 607     return 0;
 608   else
 609     return self->ppid;
 610 }
 611
 612 void* SIMIX_process_self_get_data(smx_process_t self)
 613 {
 614   xbt_assert(self == SIMIX_process_self(), "This is not the current process");
 615
 616   if (!self) {
 617     return NULL;
 618   }
 619   return SIMIX_process_get_data(self);
 620 }
 621
 622 void SIMIX_process_self_set_data(smx_process_t self, void *data)
 623 {
 624   xbt_assert(self == SIMIX_process_self(), "This is not the current process");
 625
 626   SIMIX_process_set_data(self, data);
 627 }
 628
 629 void* SIMIX_process_get_data(smx_process_t process)
 630 {
 631   return process->data;
 632 }
 633
 634 void SIMIX_process_set_data(smx_process_t process, void *data)
 635 {
 636   process->data = data;
 637 }
 638
 639 smx_host_t SIMIX_process_get_host(smx_process_t process)
 640 {
 641   return process->smx_host;
 642 }
 643
 644 /* needs to be public and without simcall because it is called
 645    by exceptions and logging events */
 646 const char* SIMIX_process_self_get_name(void) {
 647
 648   smx_process_t process = SIMIX_process_self();
 649   if (process == NULL || process == simix_global->maestro_process)
 650     return "";
 651
 652   return SIMIX_process_get_name(process);
 653 }
 654
 655 const char* SIMIX_process_get_name(smx_process_t process)
 656 {
 657   return process->name;
 658 }
 659
 660 smx_process_t SIMIX_process_get_by_name(const char* name)
 661 {
 662   smx_process_t proc;
 663
 664   xbt_swag_foreach(proc, simix_global->process_list)
 665   {
 666     if(!strcmp(name, proc->name))
 667       return proc;
 668   }
 669   return NULL;
 670 }
 671
 672 int SIMIX_process_is_suspended(smx_process_t process)
 673 {
 674   return process->suspended;
 675 }
 676
 677 xbt_dict_t SIMIX_process_get_properties(smx_process_t process)
 678 {
 679   return process->properties;
 680 }
 681
 682 void simcall_HANDLER_process_join(smx_simcall_t simcall, smx_process_t process, double timeout)
 683 {
 684   smx_synchro_t sync = SIMIX_process_join(simcall->issuer, process, timeout);
 685   xbt_fifo_push(sync->simcalls, simcall);
 686   simcall->issuer->waiting_synchro = sync;
 687 }
 688
 689 static int SIMIX_process_join_finish(smx_process_exit_status_t status, smx_synchro_t sync){
 690   if (sync->sleep.surf_sleep) {
 691     surf_action_cancel(sync->sleep.surf_sleep);
 692
 693     smx_simcall_t simcall;
 694     while ((simcall = xbt_fifo_shift(sync->simcalls))) {
 695       simcall_process_sleep__set__result(simcall, SIMIX_DONE);
 696       simcall->issuer->waiting_synchro = NULL;
 697       if (simcall->issuer->suspended) {
 698         XBT_DEBUG("Wait! This process is suspended and can't wake up now.");
 699         simcall->issuer->suspended = 0;
 700         simcall_HANDLER_process_suspend(simcall, simcall->issuer);
 701       } else {
 702         SIMIX_simcall_answer(simcall);
 703       }
 704     }
 705     surf_action_unref(sync->sleep.surf_sleep);
 706     sync->sleep.surf_sleep = NULL;
 707   }
 708   xbt_mallocator_release(simix_global->synchro_mallocator, sync);
 709   return 0;
 710 }
 711
 712 smx_synchro_t SIMIX_process_join(smx_process_t issuer, smx_process_t process, double timeout)
 713 {
 714   smx_synchro_t res = SIMIX_process_sleep(issuer, timeout);
 715   res->type = SIMIX_SYNC_JOIN;
 716   SIMIX_process_on_exit(process, (int_f_pvoid_pvoid_t)SIMIX_process_join_finish, res);
 717   return res;
 718 }
 719
 720 void simcall_HANDLER_process_sleep(smx_simcall_t simcall, double duration)
 721 {
 722   if (MC_is_active() || MC_record_replay_is_active()) {
 723     MC_process_clock_add(simcall->issuer, duration);
 724     simcall_process_sleep__set__result(simcall, SIMIX_DONE);
 725     SIMIX_simcall_answer(simcall);
 726     return;
 727   }
 728   smx_synchro_t sync = SIMIX_process_sleep(simcall->issuer, duration);
 729   xbt_fifo_push(sync->simcalls, simcall);
 730   simcall->issuer->waiting_synchro = sync;
 731 }
 732
 733 smx_synchro_t SIMIX_process_sleep(smx_process_t process, double duration)
 734 {
 735   smx_synchro_t synchro;
 736   smx_host_t host = process->smx_host;
 737
 738   /* check if the host is active */
 739   if (surf_resource_get_state(surf_workstation_resource_priv(host)) != SURF_RESOURCE_ON) {
 740     THROWF(host_error, 0, "Host %s failed, you cannot call this function",
 741            sg_host_name(host));
 742   }
 743
 744   synchro = xbt_mallocator_get(simix_global->synchro_mallocator);
 745   synchro->type = SIMIX_SYNC_SLEEP;
 746   synchro->name = NULL;
 747   synchro->category = NULL;
 748
 749   synchro->sleep.host = host;
 750   synchro->sleep.surf_sleep =
 751       surf_workstation_sleep(host, duration);
 752
 753   surf_action_set_data(synchro->sleep.surf_sleep, synchro);
 754   XBT_DEBUG("Create sleep synchronization %p", synchro);
 755
 756   return synchro;
 757 }
 758
 759 void SIMIX_post_process_sleep(smx_synchro_t synchro)
 760 {
 761   smx_simcall_t simcall;
 762   e_smx_state_t state;
 763   xbt_assert(synchro->type == SIMIX_SYNC_SLEEP || synchro->type == SIMIX_SYNC_JOIN);
 764
 765   while ((simcall = xbt_fifo_shift(synchro->simcalls))) {
 766
 767     switch(surf_action_get_state(synchro->sleep.surf_sleep)){
 768       case SURF_ACTION_FAILED:
 769         simcall->issuer->context->iwannadie = 1;
 770         //SMX_EXCEPTION(simcall->issuer, host_error, 0, "Host failed");
 771         state = SIMIX_SRC_HOST_FAILURE;
 772         break;
 773
 774       case SURF_ACTION_DONE:
 775         state = SIMIX_DONE;
 776         break;
 777
 778       default:
 779         THROW_IMPOSSIBLE;
 780         break;
 781     }
 782     if (surf_resource_get_state(surf_workstation_resource_priv(simcall->issuer->smx_host)) != SURF_RESOURCE_ON) {
 783       simcall->issuer->context->iwannadie = 1;
 784     }
 785     simcall_process_sleep__set__result(simcall, state);
 786     simcall->issuer->waiting_synchro = NULL;
 787     if (simcall->issuer->suspended) {
 788       XBT_DEBUG("Wait! This process is suspended and can't wake up now.");
 789       simcall->issuer->suspended = 0;
 790       simcall_HANDLER_process_suspend(simcall, simcall->issuer);
 791     } else {
 792       SIMIX_simcall_answer(simcall);
 793     }
 794   }
 795
 796   SIMIX_process_sleep_destroy(synchro);
 797 }
 798
 799 void SIMIX_process_sleep_destroy(smx_synchro_t synchro)
 800 {
 801   XBT_DEBUG("Destroy synchro %p", synchro);
 802   xbt_assert(synchro->type == SIMIX_SYNC_SLEEP || synchro->type == SIMIX_SYNC_JOIN);
 803
 804   if (synchro->sleep.surf_sleep) {
 805     surf_action_unref(synchro->sleep.surf_sleep);
 806     synchro->sleep.surf_sleep = NULL;
 807   }
 808   if (synchro->type == SIMIX_SYNC_SLEEP)
 809     xbt_mallocator_release(simix_global->synchro_mallocator, synchro);
 810 }
 811
 812 void SIMIX_process_sleep_suspend(smx_synchro_t synchro)
 813 {
 814   xbt_assert(synchro->type == SIMIX_SYNC_SLEEP);
 815   surf_action_suspend(synchro->sleep.surf_sleep);
 816 }
 817
 818 void SIMIX_process_sleep_resume(smx_synchro_t synchro)
 819 {
 820   XBT_DEBUG("Synchro state is %d on process_sleep_resume.", synchro->state);
 821   xbt_assert(synchro->type == SIMIX_SYNC_SLEEP);
 822   surf_action_resume(synchro->sleep.surf_sleep);
 823 }
 824
 825 /**
 826  * \brief Calling this function makes the process to yield.
 827  *
 828  * Only the current process can call this function, giving back the control to
 829  * maestro.
 830  *
 831  * \param self the current process
 832  */
 833 void SIMIX_process_yield(smx_process_t self)
 834 {
 835   XBT_DEBUG("Yield process '%s'", self->name);
 836
 837   /* Go into sleep and return control to maestro */
 838   SIMIX_context_suspend(self->context);
 839
 840   /* Ok, maestro returned control to us */
 841   XBT_DEBUG("Control returned to me: '%s'", self->name);
 842
 843   if (self->new_host) {
 844     SIMIX_process_change_host(self, self->new_host);
 845     self->new_host = NULL;
 846   }
 847
 848   if (self->context->iwannadie){
 849     XBT_DEBUG("I wanna die!");
 850     SIMIX_process_stop(self);
 851   }
 852
 853   if (self->suspended) {
 854     XBT_DEBUG("Hey! I'm suspended.");
 855     xbt_assert(!self->doexception, "Gasp! This exception may be lost by subsequent calls.");
 856     self->suspended = 0;
 857     SIMIX_process_suspend(self, self);
 858   }
 859
 860   if (self->doexception) {
 861     XBT_DEBUG("Wait, maestro left me an exception");
 862     self->doexception = 0;
 863     SMX_THROW();
 864   }
 865
 866 }
 867
 868 /* callback: context fetching */
 869 xbt_running_ctx_t *SIMIX_process_get_running_context(void)
 870 {
 871   return SIMIX_process_self()->running_ctx;
 872 }
 873
 874 /* callback: termination */
 875 void SIMIX_process_exception_terminate(xbt_ex_t * e)
 876 {
 877   xbt_ex_display(e);
 878   xbt_abort();
 879 }
 880
 881 smx_context_t SIMIX_process_get_context(smx_process_t p) {
 882   return p->context;
 883 }
 884
 885 void SIMIX_process_set_context(smx_process_t p,smx_context_t c) {
 886   p->context = c;
 887 }
 888
 889 /**
 890  * \brief Returns the list of processes to run.
 891  */
 892 xbt_dynar_t SIMIX_process_get_runnable(void)
 893 {
 894   return simix_global->process_to_run;
 895 }
 896
 897 /**
 898  * \brief Returns the process from PID.
 899  */
 900 smx_process_t SIMIX_process_from_PID(int PID)
 901 {
 902   smx_process_t proc;
 903   xbt_swag_foreach(proc, simix_global->process_list)
 904   {
 905    if(proc->pid == PID)
 906    return proc;
 907   }
 908   return NULL;
 909 }
 910
 911 /** @brief returns a dynar containg all currently existing processes */
 912 xbt_dynar_t SIMIX_processes_as_dynar(void) {
 913   smx_process_t proc;
 914   xbt_dynar_t res = xbt_dynar_new(sizeof(smx_process_t),NULL);
 915   xbt_swag_foreach(proc, simix_global->process_list) {
 916     xbt_dynar_push(res,&proc);
 917   }
 918   return res;
 919 }
 920
 921
 922 void SIMIX_process_on_exit_runall(smx_process_t process) {
 923   s_smx_process_exit_fun_t exit_fun;
 924   smx_process_exit_status_t exit_status = (process->context->iwannadie) ?
 925                                          SMX_EXIT_FAILURE : SMX_EXIT_SUCCESS;
 926   while (!xbt_dynar_is_empty(process->on_exit)) {
 927     exit_fun = xbt_dynar_pop_as(process->on_exit,s_smx_process_exit_fun_t);
 928     (exit_fun.fun)((void*)exit_status, exit_fun.arg);
 929   }
 930 }
 931
 932 void SIMIX_process_on_exit(smx_process_t process, int_f_pvoid_pvoid_t fun, void *data) {
 933   xbt_assert(process, "current process not found: are you in maestro context ?");
 934
 935   if (!process->on_exit) {
 936     process->on_exit = xbt_dynar_new(sizeof(s_smx_process_exit_fun_t), NULL);
 937   }
 938
 939   s_smx_process_exit_fun_t exit_fun = {fun, data};
 940
 941   xbt_dynar_push_as(process->on_exit,s_smx_process_exit_fun_t,exit_fun);
 942 }
 943
 944 /**
 945  * \brief Sets the auto-restart status of the process.
 946  * If set to 1, the process will be automatically restarted when its host
 947  * comes back.
 948  */
 949 void SIMIX_process_auto_restart_set(smx_process_t process, int auto_restart) {
 950   process->auto_restart = auto_restart;
 951 }
 952
 953 smx_process_t simcall_HANDLER_process_restart(smx_simcall_t simcall, smx_process_t process) {
 954   return SIMIX_process_restart(process, simcall->issuer);
 955 }
 956 /**
 957  * \brief Restart a process.
 958  * Restart a process, starting it again from the beginning.
 959  */
 960 smx_process_t SIMIX_process_restart(smx_process_t process, smx_process_t issuer) {
 961   XBT_DEBUG("Restarting process %s on %s", process->name, sg_host_name(process->smx_host));
 962   //retrieve the arguments of the old process
 963   //FIXME: Factorise this with SIMIX_host_add_auto_restart_process ?
 964   s_smx_process_arg_t arg;
 965   arg.code = process->code;
 966   arg.hostname = sg_host_name(process->smx_host);
 967   arg.kill_time = process->kill_time;
 968   arg.argc = process->argc;
 969   arg.data = process->data;
 970   int i;
 971   arg.argv = xbt_new(char*,process->argc + 1);
 972   for (i = 0; i < arg.argc; i++) {
 973     arg.argv[i] = xbt_strdup(process->argv[i]);
 974   }
 975   arg.argv[process->argc] = NULL;
 976   arg.properties = NULL;
 977   arg.auto_restart = process->auto_restart;
 978   //kill the old process
 979   SIMIX_process_kill(process,issuer);
 980   //start the new process
 981   smx_process_t new_process;
 982   if (simix_global->create_process_function) {
 983     simix_global->create_process_function(&new_process,
 984                                           arg.argv[0],
 985                                           arg.code,
 986                                           arg.data,
 987                                           arg.hostname,
 988                                           arg.kill_time,
 989                                           arg.argc,
 990                                           arg.argv,
 991                                           arg.properties,
 992                                           arg.auto_restart,
 993                                           NULL);
 994   } else {
 995     simcall_process_create(&new_process,
 996                            arg.argv[0],
 997                            arg.code,
 998                            arg.data,
 999                            arg.hostname,
1000                            arg.kill_time,
1001                            arg.argc,
1002                            arg.argv,
1003                            arg.properties,
1004                            arg.auto_restart);
1005
1006   }
1007   return new_process;
1008 }