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 #ifdef HAVE_SMPI
  17 #include "src/smpi/private.h"
  18 #endif
  19
  20 XBT_LOG_NEW_DEFAULT_SUBCATEGORY(simix_process, simix,
  21                                 "Logging specific to SIMIX (process)");
  22
  23 unsigned long simix_process_maxpid = 0;
  24
  25 /**
  26  * \brief Returns the current agent.
  27  *
  28  * This functions returns the currently running SIMIX process.
  29  *
  30  * \return The SIMIX process
  31  */
  32 smx_process_t SIMIX_process_self(void)
  33 {
  34   smx_context_t self_context = SIMIX_context_self();
  35
  36   return self_context ? SIMIX_context_get_process(self_context) : NULL;
  37 }
  38
  39 /**
  40  * \brief Returns whether a process has pending asynchronous communications.
  41  * \return true if there are asynchronous communications in this process
  42  */
  43 int SIMIX_process_has_pending_comms(smx_process_t process) {
  44
  45   return xbt_fifo_size(process->comms) > 0;
  46 }
  47
  48 /**
  49  * \brief Moves a process to the list of processes to destroy.
  50  */
  51 void SIMIX_process_cleanup(smx_process_t process)
  52 {
  53   XBT_DEBUG("Cleanup process %s (%p), waiting synchro %p",
  54       process->name, process, process->waiting_synchro);
  55
  56   SIMIX_process_on_exit_runall(process);
  57
  58   /* Unregister from the kill timer if any */
  59   if (process->kill_timer != NULL)
  60           SIMIX_timer_remove(process->kill_timer);
  61
  62   xbt_os_mutex_acquire(simix_global->mutex);
  63
  64   /* cancel non-blocking communications */
  65   smx_synchro_t synchro;
  66   while ((synchro = (smx_synchro_t) xbt_fifo_pop(process->comms))) {
  67
  68     /* make sure no one will finish the comm after this process is destroyed,
  69      * because src_proc or dst_proc would be an invalid pointer */
  70     SIMIX_comm_cancel(synchro);
  71
  72     if (synchro->comm.src_proc == process) {
  73       XBT_DEBUG("Found an unfinished send comm %p (detached = %d), state %d, src = %p, dst = %p",
  74           synchro, synchro->comm.detached, (int)synchro->state, synchro->comm.src_proc, synchro->comm.dst_proc);
  75       synchro->comm.src_proc = NULL;
  76
  77       /* I'm not supposed to destroy a detached comm from the sender side, */
  78       if (!synchro->comm.detached)
  79         SIMIX_comm_destroy(synchro);
  80       else
  81         XBT_DEBUG("Don't destroy it since it's a detached comm");
  82
  83     }
  84     else if (synchro->comm.dst_proc == process){
  85       XBT_DEBUG("Found an unfinished recv comm %p, state %d, src = %p, dst = %p",
  86           synchro, (int)synchro->state, synchro->comm.src_proc, synchro->comm.dst_proc);
  87       synchro->comm.dst_proc = NULL;
  88
  89       if (synchro->comm.detached && synchro->comm.refcount == 1
  90           && synchro->comm.src_proc != NULL) {
  91         /* the comm will be freed right now, remove it from the sender */
  92         xbt_fifo_remove(synchro->comm.src_proc->comms, synchro);
  93       }
  94       SIMIX_comm_destroy(synchro);
  95     }
  96     else {
  97       xbt_die("Communication synchro %p is in my list but I'm not the sender "
  98           "or the receiver", synchro);
  99     }
 100   }
 101
 102   XBT_DEBUG("%p should not be run anymore",process);
 103   xbt_swag_remove(process, simix_global->process_list);
 104   xbt_swag_remove(process, sg_host_simix(process->host)->process_list);
 105   xbt_swag_insert(process, simix_global->process_to_destroy);
 106   process->context->iwannadie = 0;
 107
 108   xbt_os_mutex_release(simix_global->mutex);
 109 }
 110
 111 /**
 112  * Garbage collection
 113  *
 114  * Should be called some time to time to free the memory allocated for processes
 115  * that have finished (or killed).
 116  */
 117 void SIMIX_process_empty_trash(void)
 118 {
 119   smx_process_t process = NULL;
 120
 121   while ((process = (smx_process_t) xbt_swag_extract(simix_global->process_to_destroy))) {
 122     XBT_DEBUG("Getting rid of %p",process);
 123
 124     SIMIX_context_free(process->context);
 125
 126     /* Free the exception allocated at creation time */
 127     free(process->running_ctx);
 128     xbt_dict_free(&process->properties);
 129
 130     xbt_fifo_free(process->comms);
 131
 132     xbt_dynar_free(&process->on_exit);
 133
 134     xbt_free(process->name);
 135     xbt_free(process);
 136   }
 137 }
 138
 139 /**
 140  * \brief Creates and runs the maestro process
 141  */
 142 void SIMIX_create_maestro_process()
 143 {
 144   smx_process_t maestro = NULL;
 145
 146   /* Create maestro process and intilialize it */
 147   maestro = xbt_new0(s_smx_process_t, 1);
 148   maestro->pid = simix_process_maxpid++;
 149   maestro->ppid = -1;
 150   maestro->name = (char *) "";
 151   maestro->running_ctx = (xbt_running_ctx_t*) xbt_malloc0(sizeof(xbt_running_ctx_t));
 152   XBT_RUNNING_CTX_INITIALIZE(maestro->running_ctx);
 153   maestro->context = SIMIX_context_new(NULL, 0, NULL, NULL, maestro);
 154   maestro->simcall.issuer = maestro;
 155   simix_global->maestro_process = maestro;
 156   return;
 157 }
 158 /**
 159  * \brief Stops a process.
 160  *
 161  * Stops the process, execute all the registered on_exit functions,
 162  * register it to the list of the process to restart if needed
 163  * and stops its context.
 164  */
 165 void SIMIX_process_stop(smx_process_t arg) {
 166   /* execute the on_exit functions */
 167   SIMIX_process_on_exit_runall(arg);
 168   /* Add the process to the list of process to restart, only if
 169    * the host is down
 170    */
 171   if (arg->auto_restart && !sg_host_get_state(arg->host)) {
 172     SIMIX_host_add_auto_restart_process(arg->host,arg->name,arg->code, arg->data,
 173                                         sg_host_get_name(arg->host),
 174                                         SIMIX_timer_get_date(arg->kill_timer),
 175                                         arg->argc,arg->argv,arg->properties,
 176                                         arg->auto_restart);
 177   }
 178   XBT_DEBUG("Process %s (%s) is dead",arg->name,sg_host_get_name(arg->host));
 179   /* stop the context */
 180   SIMIX_context_stop(arg->context);
 181 }
 182
 183 /**
 184  * \brief Same as SIMIX_process_create() but with only one argument (used by timers).
 185  * This function frees the argument.
 186  * \return the process created
 187  */
 188 smx_process_t SIMIX_process_create_from_wrapper(smx_process_arg_t args) {
 189
 190   smx_process_t process = simix_global->create_process_function(
 191                                         args->name,
 192                                         args->code,
 193                                         args->data,
 194                                         args->hostname,
 195                                         args->kill_time,
 196                                         args->argc,
 197                                         args->argv,
 198                                         args->properties,
 199                                         args->auto_restart,
 200                                         NULL);
 201   xbt_free(args);
 202   return process;
 203 }
 204
 205
 206 void* simcall_HANDLER_process_create(smx_simcall_t simcall,
 207                           const char *name,
 208                           xbt_main_func_t code,
 209                           void *data,
 210                           const char *hostname,
 211                           double kill_time,
 212                           int argc, char **argv,
 213                           xbt_dict_t properties,
 214                           int auto_restart){
 215   return (void*)SIMIX_process_create(name, code, data, hostname,
 216                        kill_time, argc, argv, properties, auto_restart,
 217                        simcall->issuer);
 218 }
 219
 220 static void kill_process(void* process)
 221 {
 222   simix_global->kill_process_function((smx_process_t) process);
 223 }
 224
 225 /**
 226  * \brief Internal function to create a process.
 227  *
 228  * This function actually creates the process.
 229  * It may be called when a SIMCALL_PROCESS_CREATE simcall occurs,
 230  * or directly for SIMIX internal purposes. The sure thing is that it's called from maestro context.
 231  *
 232  * \return the process created
 233  */
 234 smx_process_t SIMIX_process_create(
 235                           const char *name,
 236                           xbt_main_func_t code,
 237                           void *data,
 238                           const char *hostname,
 239                           double kill_time,
 240                           int argc, char **argv,
 241                           xbt_dict_t properties,
 242                           int auto_restart,
 243                           smx_process_t parent_process)
 244 {
 245   smx_process_t process = NULL;
 246   sg_host_t host = sg_host_by_name(hostname);
 247
 248   XBT_DEBUG("Start process %s on host '%s'", name, hostname);
 249
 250   if (!sg_host_get_state(host)) {
 251     int i;
 252     XBT_WARN("Cannot launch process '%s' on failed host '%s'", name,
 253           hostname);
 254     for (i = 0; i < argc; i++)
 255       xbt_free(argv[i]);
 256     xbt_free(argv);
 257   }
 258   else {
 259     process = xbt_new0(s_smx_process_t, 1);
 260
 261     xbt_assert(((code != NULL) && (host != NULL)), "Invalid parameters");
 262     /* Process data */
 263     process->pid = simix_process_maxpid++;
 264     process->name = xbt_strdup(name);
 265     process->host = host;
 266     process->data = data;
 267     process->comms = xbt_fifo_new();
 268     process->simcall.issuer = process;
 269     /* Initiliaze data segment to default value */
 270     SIMIX_segment_index_set(process, -1);
 271
 272      if (parent_process != NULL) {
 273        process->ppid = SIMIX_process_get_PID(parent_process);
 274        /* SMPI process have their own data segment and
 275           each other inherit from their father */
 276        if(smpi_privatize_global_variables){
 277          if( parent_process->pid != 0){
 278            SIMIX_segment_index_set(process, parent_process->segment_index);
 279          } else {
 280            SIMIX_segment_index_set(process, process->pid - 1);
 281          }
 282        }
 283      } else {
 284        process->ppid = -1;
 285      }
 286
 287     /* Process data for auto-restart */
 288     process->auto_restart = auto_restart;
 289     process->code = code;
 290     process->argc = argc;
 291     process->argv = argv;
 292
 293
 294     XBT_VERB("Create context %s", process->name);
 295     process->context = SIMIX_context_new(code, argc, argv, simix_global->cleanup_process_function, process);
 296
 297     process->running_ctx = (xbt_running_ctx_t*) xbt_malloc0(sizeof(xbt_running_ctx_t));
 298     XBT_RUNNING_CTX_INITIALIZE(process->running_ctx);
 299
 300     if(MC_is_active()){
 301       MC_ignore_heap(process->running_ctx, sizeof(*process->running_ctx));
 302     }
 303
 304     /* Add properties */
 305     process->properties = properties;
 306
 307     /* Add the process to it's host process list */
 308     xbt_swag_insert(process, sg_host_simix(host)->process_list);
 309
 310     XBT_DEBUG("Start context '%s'", process->name);
 311
 312     /* Now insert it in the global process list and in the process to run list */
 313     xbt_swag_insert(process, simix_global->process_list);
 314     XBT_DEBUG("Inserting %s(%s) in the to_run list", process->name, sg_host_get_name(host));
 315     xbt_dynar_push_as(simix_global->process_to_run, smx_process_t, process);
 316
 317     if (kill_time > SIMIX_get_clock() && simix_global->kill_process_function) {
 318       XBT_DEBUG("Process %s(%s) will be kill at time %f", process->name,
 319           sg_host_get_name(process->host), kill_time);
 320       process->kill_timer = SIMIX_timer_set(kill_time, kill_process, process);
 321     }
 322   }
 323   return process;
 324 }
 325
 326 /**
 327  * \brief Executes the processes from simix_global->process_to_run.
 328  *
 329  * The processes of simix_global->process_to_run are run (in parallel if
 330  * possible).  On exit, simix_global->process_to_run is empty, and
 331  * simix_global->process_that_ran contains the list of processes that just ran.
 332  * The two lists are swapped so, be careful when using them before and after a
 333  * call to this function.
 334  */
 335 void SIMIX_process_runall(void)
 336 {
 337   SIMIX_context_runall();
 338
 339   xbt_dynar_t tmp = simix_global->process_that_ran;
 340   simix_global->process_that_ran = simix_global->process_to_run;
 341   simix_global->process_to_run = tmp;
 342   xbt_dynar_reset(simix_global->process_to_run);
 343 }
 344
 345 void simcall_HANDLER_process_kill(smx_simcall_t simcall, smx_process_t process) {
 346   SIMIX_process_kill(process, simcall->issuer);
 347 }
 348 /**
 349  * \brief Internal function to kill a SIMIX process.
 350  *
 351  * This function may be called when a SIMCALL_PROCESS_KILL simcall occurs,
 352  * or directly for SIMIX internal purposes.
 353  *
 354  * \param process poor victim
 355  * \param issuer the process which has sent the PROCESS_KILL. Important to not schedule twice the same process.
 356  */
 357 void SIMIX_process_kill(smx_process_t process, smx_process_t issuer) {
 358
 359   XBT_DEBUG("Killing process %s on %s", process->name, sg_host_get_name(process->host));
 360
 361   process->context->iwannadie = 1;
 362   process->blocked = 0;
 363   process->suspended = 0;
 364   process->doexception = 0;
 365
 366   /* destroy the blocking synchro if any */
 367   if (process->waiting_synchro) {
 368
 369     switch (process->waiting_synchro->type) {
 370
 371     case SIMIX_SYNC_EXECUTE:
 372     case SIMIX_SYNC_PARALLEL_EXECUTE:
 373       SIMIX_process_execution_destroy(process->waiting_synchro);
 374       break;
 375
 376     case SIMIX_SYNC_COMMUNICATE:
 377       xbt_fifo_remove(process->comms, process->waiting_synchro);
 378       SIMIX_comm_cancel(process->waiting_synchro);
 379       xbt_fifo_remove(process->waiting_synchro->simcalls, &process->simcall);
 380       SIMIX_comm_destroy(process->waiting_synchro);
 381       break;
 382
 383     case SIMIX_SYNC_SLEEP:
 384       SIMIX_process_sleep_destroy(process->waiting_synchro);
 385       break;
 386
 387     case SIMIX_SYNC_JOIN:
 388       SIMIX_process_sleep_destroy(process->waiting_synchro);
 389       break;
 390
 391     case SIMIX_SYNC_SYNCHRO:
 392       SIMIX_synchro_stop_waiting(process, &process->simcall);
 393       SIMIX_synchro_destroy(process->waiting_synchro);
 394       break;
 395
 396     case SIMIX_SYNC_IO:
 397       SIMIX_io_destroy(process->waiting_synchro);
 398       break;
 399
 400     }
 401
 402     process->waiting_synchro = NULL;
 403   }
 404   if(!xbt_dynar_member(simix_global->process_to_run, &(process)) && process != issuer) {
 405     XBT_DEBUG("Inserting %s in the to_run list", process->name);
 406     xbt_dynar_push_as(simix_global->process_to_run, smx_process_t, process);
 407   }
 408
 409 }
 410
 411 /** @brief Ask another process to raise the given exception
 412  *
 413  * @param cat category of exception
 414  * @param value value associated to the exception
 415  * @param msg string information associated to the exception
 416  */
 417 void SIMIX_process_throw(smx_process_t process, xbt_errcat_t cat, int value, const char *msg) {
 418   SMX_EXCEPTION(process, cat, value, msg);
 419
 420   if (process->suspended)
 421     SIMIX_process_resume(process,SIMIX_process_self());
 422
 423   /* cancel the blocking synchro if any */
 424   if (process->waiting_synchro) {
 425
 426     switch (process->waiting_synchro->type) {
 427
 428     case SIMIX_SYNC_EXECUTE:
 429     case SIMIX_SYNC_PARALLEL_EXECUTE:
 430       SIMIX_process_execution_cancel(process->waiting_synchro);
 431       break;
 432
 433     case SIMIX_SYNC_COMMUNICATE:
 434       xbt_fifo_remove(process->comms, process->waiting_synchro);
 435       SIMIX_comm_cancel(process->waiting_synchro);
 436       break;
 437
 438     case SIMIX_SYNC_SLEEP:
 439     case SIMIX_SYNC_JOIN:
 440       SIMIX_process_sleep_destroy(process->waiting_synchro);
 441       if (!xbt_dynar_member(simix_global->process_to_run, &(process)) && process != SIMIX_process_self()) {
 442         XBT_DEBUG("Inserting %s in the to_run list", process->name);
 443         xbt_dynar_push_as(simix_global->process_to_run, smx_process_t, process);
 444       }
 445       break;
 446
 447     case SIMIX_SYNC_SYNCHRO:
 448       SIMIX_synchro_stop_waiting(process, &process->simcall);
 449       break;
 450
 451     case SIMIX_SYNC_IO:
 452       SIMIX_io_destroy(process->waiting_synchro);
 453       break;
 454
 455     }
 456   }
 457   process->waiting_synchro = NULL;
 458
 459 }
 460
 461 void simcall_HANDLER_process_killall(smx_simcall_t simcall, int reset_pid) {
 462   SIMIX_process_killall(simcall->issuer, reset_pid);
 463 }
 464 /**
 465  * \brief Kills all running processes.
 466  * \param issuer this one will not be killed
 467  */
 468 void SIMIX_process_killall(smx_process_t issuer, int reset_pid)
 469 {
 470   smx_process_t p = NULL;
 471
 472   while ((p = (smx_process_t) xbt_swag_extract(simix_global->process_list))) {
 473     if (p != issuer) {
 474       SIMIX_process_kill(p,issuer);
 475     }
 476   }
 477
 478   if (reset_pid > 0)
 479     simix_process_maxpid = reset_pid;
 480
 481   SIMIX_context_runall();
 482
 483   SIMIX_process_empty_trash();
 484 }
 485
 486 void simcall_HANDLER_process_set_host(smx_simcall_t simcall, smx_process_t process, sg_host_t dest)
 487 {
 488   process->new_host = dest;
 489 }
 490 void SIMIX_process_change_host(smx_process_t process,
 491              sg_host_t dest)
 492 {
 493   xbt_assert((process != NULL), "Invalid parameters");
 494   xbt_swag_remove(process, sg_host_simix(process->host)->process_list);
 495   process->host = dest;
 496   xbt_swag_insert(process, sg_host_simix(dest)->process_list);
 497 }
 498
 499
 500 void simcall_HANDLER_process_suspend(smx_simcall_t simcall, smx_process_t process)
 501 {
 502   smx_synchro_t sync_suspend =
 503       SIMIX_process_suspend(process, simcall->issuer);
 504
 505   if (process != simcall->issuer) {
 506     SIMIX_simcall_answer(simcall);
 507   } else {
 508     xbt_fifo_push(sync_suspend->simcalls, simcall);
 509     process->waiting_synchro = sync_suspend;
 510     SIMIX_host_execution_suspend(process->waiting_synchro);
 511   }
 512   /* If we are suspending ourselves, then just do not finish the simcall now */
 513 }
 514
 515 smx_synchro_t SIMIX_process_suspend(smx_process_t process, smx_process_t issuer)
 516 {
 517   xbt_assert((process != NULL), "Invalid parameters");
 518
 519   if (process->suspended) {
 520     XBT_DEBUG("Process '%s' is already suspended", process->name);
 521     return NULL;
 522   }
 523
 524   process->suspended = 1;
 525
 526   /* If we are suspending another process, and it is waiting on a sync,
 527      suspend its synchronization. */
 528   if (process != issuer) {
 529
 530     if (process->waiting_synchro) {
 531
 532       switch (process->waiting_synchro->type) {
 533
 534         case SIMIX_SYNC_EXECUTE:
 535         case SIMIX_SYNC_PARALLEL_EXECUTE:
 536           SIMIX_host_execution_suspend(process->waiting_synchro);
 537           break;
 538
 539         case SIMIX_SYNC_COMMUNICATE:
 540           SIMIX_comm_suspend(process->waiting_synchro);
 541           break;
 542
 543         case SIMIX_SYNC_SLEEP:
 544           SIMIX_process_sleep_suspend(process->waiting_synchro);
 545           break;
 546
 547         case SIMIX_SYNC_SYNCHRO:
 548           /* Suspension is delayed to when the process is rescheduled. */
 549           break;
 550
 551         default:
 552           xbt_die("Internal error in SIMIX_process_suspend: unexpected synchronization type %d",
 553               (int)process->waiting_synchro->type);
 554       }
 555       return NULL;
 556     } else {
 557       /* Suspension is delayed to when the process is rescheduled. */
 558       return NULL;
 559     }
 560   } else {
 561     /* FIXME: computation size is zero. Is it okay that bound is zero ? */
 562     return SIMIX_process_execute(process, "suspend", 0.0, 1.0, 0.0, 0);
 563   }
 564 }
 565
 566 void simcall_HANDLER_process_resume(smx_simcall_t simcall, smx_process_t process){
 567   SIMIX_process_resume(process, simcall->issuer);
 568 }
 569
 570 void SIMIX_process_resume(smx_process_t process, smx_process_t issuer)
 571 {
 572   XBT_IN("process = %p, issuer = %p", process, issuer);
 573
 574   if(process->context->iwannadie) {
 575     XBT_VERB("Ignoring request to suspend a process that is currently dying.");
 576     return;
 577   }
 578
 579   if(!process->suspended) return;
 580   process->suspended = 0;
 581
 582   /* If we are resuming another process, resume the synchronization it was waiting for
 583      if any. Otherwise add it to the list of process to run in the next round. */
 584   if (process != issuer) {
 585
 586     if (process->waiting_synchro) {
 587
 588       switch (process->waiting_synchro->type) {
 589
 590         case SIMIX_SYNC_EXECUTE:
 591         case SIMIX_SYNC_PARALLEL_EXECUTE:
 592           SIMIX_host_execution_resume(process->waiting_synchro);
 593           break;
 594
 595         case SIMIX_SYNC_COMMUNICATE:
 596           SIMIX_comm_resume(process->waiting_synchro);
 597           break;
 598
 599         case SIMIX_SYNC_SLEEP:
 600           SIMIX_process_sleep_resume(process->waiting_synchro);
 601           break;
 602
 603         case SIMIX_SYNC_SYNCHRO:
 604           /* I cannot resume it now. This is delayed to when the process is rescheduled at
 605            * the end of the synchro. */
 606           break;
 607
 608         default:
 609           xbt_die("Internal error in SIMIX_process_resume: unexpected synchronization type %d",
 610               (int)process->waiting_synchro->type);
 611       }
 612     }
 613   } else XBT_WARN("Strange. Process %p is trying to resume himself.", issuer);
 614
 615   XBT_OUT();
 616 }
 617
 618 int SIMIX_process_get_maxpid(void) {
 619   return simix_process_maxpid;
 620 }
 621
 622 int SIMIX_process_count(void)
 623 {
 624   return xbt_swag_size(simix_global->process_list);
 625 }
 626
 627 int SIMIX_process_get_PID(smx_process_t self){
 628   if (self == NULL)
 629     return 0;
 630   else
 631     return self->pid;
 632 }
 633
 634 int SIMIX_process_get_PPID(smx_process_t self){
 635   if (self == NULL)
 636     return 0;
 637   else
 638     return self->ppid;
 639 }
 640
 641 void* SIMIX_process_self_get_data(smx_process_t self)
 642 {
 643   xbt_assert(self == SIMIX_process_self(), "This is not the current process");
 644
 645   if (!self) {
 646     return NULL;
 647   }
 648   return SIMIX_process_get_data(self);
 649 }
 650
 651 void SIMIX_process_self_set_data(smx_process_t self, void *data)
 652 {
 653   xbt_assert(self == SIMIX_process_self(), "This is not the current process");
 654
 655   SIMIX_process_set_data(self, data);
 656 }
 657
 658 void* SIMIX_process_get_data(smx_process_t process)
 659 {
 660   return process->data;
 661 }
 662
 663 void SIMIX_process_set_data(smx_process_t process, void *data)
 664 {
 665   process->data = data;
 666 }
 667
 668 sg_host_t SIMIX_process_get_host(smx_process_t process)
 669 {
 670   return process->host;
 671 }
 672
 673 xbt_main_func_t SIMIX_process_get_code(void){
 674   return SIMIX_process_self()->code;
 675 }
 676
 677 /* needs to be public and without simcall because it is called
 678    by exceptions and logging events */
 679 const char* SIMIX_process_self_get_name(void) {
 680
 681   smx_process_t process = SIMIX_process_self();
 682   if (process == NULL || process == simix_global->maestro_process)
 683     return "";
 684
 685   return SIMIX_process_get_name(process);
 686 }
 687
 688 const char* SIMIX_process_get_name(smx_process_t process)
 689 {
 690   return process->name;
 691 }
 692
 693 smx_process_t SIMIX_process_get_by_name(const char* name)
 694 {
 695   smx_process_t proc;
 696
 697   xbt_swag_foreach(proc, simix_global->process_list)
 698   {
 699     if(!strcmp(name, proc->name))
 700       return proc;
 701   }
 702   return NULL;
 703 }
 704
 705 int SIMIX_process_is_suspended(smx_process_t process)
 706 {
 707   return process->suspended;
 708 }
 709
 710 xbt_dict_t SIMIX_process_get_properties(smx_process_t process)
 711 {
 712   return process->properties;
 713 }
 714
 715 void simcall_HANDLER_process_join(smx_simcall_t simcall, smx_process_t process, double timeout)
 716 {
 717   smx_synchro_t sync = SIMIX_process_join(simcall->issuer, process, timeout);
 718   xbt_fifo_push(sync->simcalls, simcall);
 719   simcall->issuer->waiting_synchro = sync;
 720 }
 721
 722 static int SIMIX_process_join_finish(smx_process_exit_status_t status, smx_synchro_t sync){
 723   if (sync->sleep.surf_sleep) {
 724     surf_action_cancel(sync->sleep.surf_sleep);
 725
 726     smx_simcall_t simcall;
 727     while ((simcall = (smx_simcall_t) xbt_fifo_shift(sync->simcalls))) {
 728       simcall_process_sleep__set__result(simcall, SIMIX_DONE);
 729       simcall->issuer->waiting_synchro = NULL;
 730       if (simcall->issuer->suspended) {
 731         XBT_DEBUG("Wait! This process is suspended and can't wake up now.");
 732         simcall->issuer->suspended = 0;
 733         simcall_HANDLER_process_suspend(simcall, simcall->issuer);
 734       } else {
 735         SIMIX_simcall_answer(simcall);
 736       }
 737     }
 738     surf_action_unref(sync->sleep.surf_sleep);
 739     sync->sleep.surf_sleep = NULL;
 740   }
 741   xbt_mallocator_release(simix_global->synchro_mallocator, sync);
 742   return 0;
 743 }
 744
 745 smx_synchro_t SIMIX_process_join(smx_process_t issuer, smx_process_t process, double timeout)
 746 {
 747   smx_synchro_t res = SIMIX_process_sleep(issuer, timeout);
 748   res->type = SIMIX_SYNC_JOIN;
 749   SIMIX_process_on_exit(process, (int_f_pvoid_pvoid_t)SIMIX_process_join_finish, res);
 750   return res;
 751 }
 752
 753 void simcall_HANDLER_process_sleep(smx_simcall_t simcall, double duration)
 754 {
 755   if (MC_is_active() || MC_record_replay_is_active()) {
 756     MC_process_clock_add(simcall->issuer, duration);
 757     simcall_process_sleep__set__result(simcall, SIMIX_DONE);
 758     SIMIX_simcall_answer(simcall);
 759     return;
 760   }
 761   smx_synchro_t sync = SIMIX_process_sleep(simcall->issuer, duration);
 762   xbt_fifo_push(sync->simcalls, simcall);
 763   simcall->issuer->waiting_synchro = sync;
 764 }
 765
 766 smx_synchro_t SIMIX_process_sleep(smx_process_t process, double duration)
 767 {
 768   sg_host_t host = process->host;
 769
 770   /* check if the host is active */
 771   if (surf_host_get_state(surf_host_resource_priv(host)) != SURF_RESOURCE_ON) {
 772     THROWF(host_error, 0, "Host %s failed, you cannot call this function",
 773            sg_host_get_name(host));
 774   }
 775
 776   smx_synchro_t synchro = (smx_synchro_t) xbt_mallocator_get(simix_global->synchro_mallocator);
 777   synchro->type = SIMIX_SYNC_SLEEP;
 778   synchro->name = NULL;
 779   synchro->category = NULL;
 780
 781   synchro->sleep.host = host;
 782   synchro->sleep.surf_sleep = surf_host_sleep(host, duration);
 783
 784   surf_action_set_data(synchro->sleep.surf_sleep, synchro);
 785   XBT_DEBUG("Create sleep synchronization %p", synchro);
 786
 787   return synchro;
 788 }
 789
 790 void SIMIX_post_process_sleep(smx_synchro_t synchro)
 791 {
 792   smx_simcall_t simcall;
 793   e_smx_state_t state;
 794   xbt_assert(synchro->type == SIMIX_SYNC_SLEEP || synchro->type == SIMIX_SYNC_JOIN);
 795
 796   while ((simcall = (smx_simcall_t) xbt_fifo_shift(synchro->simcalls))) {
 797
 798     switch(surf_action_get_state(synchro->sleep.surf_sleep)){
 799       case SURF_ACTION_FAILED:
 800         simcall->issuer->context->iwannadie = 1;
 801         //SMX_EXCEPTION(simcall->issuer, host_error, 0, "Host failed");
 802         state = SIMIX_SRC_HOST_FAILURE;
 803         break;
 804
 805       case SURF_ACTION_DONE:
 806         state = SIMIX_DONE;
 807         break;
 808
 809       default:
 810         THROW_IMPOSSIBLE;
 811         break;
 812     }
 813     if (surf_host_get_state(surf_host_resource_priv(simcall->issuer->host)) != SURF_RESOURCE_ON) {
 814       simcall->issuer->context->iwannadie = 1;
 815     }
 816     simcall_process_sleep__set__result(simcall, state);
 817     simcall->issuer->waiting_synchro = NULL;
 818     if (simcall->issuer->suspended) {
 819       XBT_DEBUG("Wait! This process is suspended and can't wake up now.");
 820       simcall->issuer->suspended = 0;
 821       simcall_HANDLER_process_suspend(simcall, simcall->issuer);
 822     } else {
 823       SIMIX_simcall_answer(simcall);
 824     }
 825   }
 826
 827   SIMIX_process_sleep_destroy(synchro);
 828 }
 829
 830 void SIMIX_process_sleep_destroy(smx_synchro_t synchro)
 831 {
 832   XBT_DEBUG("Destroy synchro %p", synchro);
 833   xbt_assert(synchro->type == SIMIX_SYNC_SLEEP || synchro->type == SIMIX_SYNC_JOIN);
 834
 835   if (synchro->sleep.surf_sleep) {
 836     surf_action_unref(synchro->sleep.surf_sleep);
 837     synchro->sleep.surf_sleep = NULL;
 838   }
 839   if (synchro->type == SIMIX_SYNC_SLEEP)
 840     xbt_mallocator_release(simix_global->synchro_mallocator, synchro);
 841 }
 842
 843 void SIMIX_process_sleep_suspend(smx_synchro_t synchro)
 844 {
 845   xbt_assert(synchro->type == SIMIX_SYNC_SLEEP);
 846   surf_action_suspend(synchro->sleep.surf_sleep);
 847 }
 848
 849 void SIMIX_process_sleep_resume(smx_synchro_t synchro)
 850 {
 851   XBT_DEBUG("Synchro state is %d on process_sleep_resume.", synchro->state);
 852   xbt_assert(synchro->type == SIMIX_SYNC_SLEEP);
 853   surf_action_resume(synchro->sleep.surf_sleep);
 854 }
 855
 856 /**
 857  * \brief Calling this function makes the process to yield.
 858  *
 859  * Only the current process can call this function, giving back the control to
 860  * maestro.
 861  *
 862  * \param self the current process
 863  */
 864 void SIMIX_process_yield(smx_process_t self)
 865 {
 866   XBT_DEBUG("Yield process '%s'", self->name);
 867
 868   /* Go into sleep and return control to maestro */
 869   SIMIX_context_suspend(self->context);
 870
 871   /* Ok, maestro returned control to us */
 872   XBT_DEBUG("Control returned to me: '%s'", self->name);
 873
 874   if (self->new_host) {
 875     SIMIX_process_change_host(self, self->new_host);
 876     self->new_host = NULL;
 877   }
 878
 879   if (self->context->iwannadie){
 880     XBT_DEBUG("I wanna die!");
 881     SIMIX_process_stop(self);
 882   }
 883
 884   if (self->suspended) {
 885     XBT_DEBUG("Hey! I'm suspended.");
 886     xbt_assert(!self->doexception, "Gasp! This exception may be lost by subsequent calls.");
 887     self->suspended = 0;
 888     SIMIX_process_suspend(self, self);
 889   }
 890
 891   if (self->doexception) {
 892     XBT_DEBUG("Wait, maestro left me an exception");
 893     self->doexception = 0;
 894     SMX_THROW();
 895   }
 896
 897   if(SMPI_switch_data_segment && self->segment_index != -1){
 898     SMPI_switch_data_segment(self->segment_index);
 899   }
 900 }
 901
 902 /* callback: context fetching */
 903 xbt_running_ctx_t *SIMIX_process_get_running_context(void)
 904 {
 905   return SIMIX_process_self()->running_ctx;
 906 }
 907
 908 /* callback: termination */
 909 void SIMIX_process_exception_terminate(xbt_ex_t * e)
 910 {
 911   xbt_ex_display(e);
 912   xbt_abort();
 913 }
 914
 915 smx_context_t SIMIX_process_get_context(smx_process_t p) {
 916   return p->context;
 917 }
 918
 919 void SIMIX_process_set_context(smx_process_t p,smx_context_t c) {
 920   p->context = c;
 921 }
 922
 923 /**
 924  * \brief Returns the list of processes to run.
 925  */
 926 xbt_dynar_t SIMIX_process_get_runnable(void)
 927 {
 928   return simix_global->process_to_run;
 929 }
 930
 931 /**
 932  * \brief Returns the process from PID.
 933  */
 934 smx_process_t SIMIX_process_from_PID(int PID)
 935 {
 936   smx_process_t proc;
 937   xbt_swag_foreach(proc, simix_global->process_list) {
 938    if (proc->pid == (unsigned long) PID)
 939     return proc;
 940   }
 941   return NULL;
 942 }
 943
 944 /** @brief returns a dynar containg all currently existing processes */
 945 xbt_dynar_t SIMIX_processes_as_dynar(void) {
 946   smx_process_t proc;
 947   xbt_dynar_t res = xbt_dynar_new(sizeof(smx_process_t),NULL);
 948   xbt_swag_foreach(proc, simix_global->process_list) {
 949     xbt_dynar_push(res,&proc);
 950   }
 951   return res;
 952 }
 953
 954
 955 void SIMIX_process_on_exit_runall(smx_process_t process) {
 956   s_smx_process_exit_fun_t exit_fun;
 957   smx_process_exit_status_t exit_status = (process->context->iwannadie) ?
 958                                          SMX_EXIT_FAILURE : SMX_EXIT_SUCCESS;
 959   while (!xbt_dynar_is_empty(process->on_exit)) {
 960     exit_fun = xbt_dynar_pop_as(process->on_exit,s_smx_process_exit_fun_t);
 961     (exit_fun.fun)((void*)exit_status, exit_fun.arg);
 962   }
 963 }
 964
 965 void SIMIX_process_on_exit(smx_process_t process, int_f_pvoid_pvoid_t fun, void *data) {
 966   xbt_assert(process, "current process not found: are you in maestro context ?");
 967
 968   if (!process->on_exit) {
 969     process->on_exit = xbt_dynar_new(sizeof(s_smx_process_exit_fun_t), NULL);
 970   }
 971
 972   s_smx_process_exit_fun_t exit_fun = {fun, data};
 973
 974   xbt_dynar_push_as(process->on_exit,s_smx_process_exit_fun_t,exit_fun);
 975 }
 976
 977 /**
 978  * \brief Sets the auto-restart status of the process.
 979  * If set to 1, the process will be automatically restarted when its host
 980  * comes back.
 981  */
 982 void SIMIX_process_auto_restart_set(smx_process_t process, int auto_restart) {
 983   process->auto_restart = auto_restart;
 984 }
 985
 986 smx_process_t simcall_HANDLER_process_restart(smx_simcall_t simcall, smx_process_t process) {
 987   return SIMIX_process_restart(process, simcall->issuer);
 988 }
 989 /** @brief Restart a process, starting it again from the beginning. */
 990 smx_process_t SIMIX_process_restart(smx_process_t process, smx_process_t issuer) {
 991   XBT_DEBUG("Restarting process %s on %s", process->name, sg_host_get_name(process->host));
 992   //retrieve the arguments of the old process
 993   //FIXME: Factorize this with SIMIX_host_add_auto_restart_process ?
 994   s_smx_process_arg_t arg;
 995   arg.code = process->code;
 996   arg.hostname = sg_host_get_name(process->host);
 997   arg.kill_time = SIMIX_timer_get_date(process->kill_timer);
 998   arg.argc = process->argc;
 999   arg.data = process->data;
1000   int i;
1001   arg.argv = xbt_new(char*,process->argc + 1);
1002   for (i = 0; i < arg.argc; i++) {
1003     arg.argv[i] = xbt_strdup(process->argv[i]);
1004   }
1005   arg.argv[process->argc] = NULL;
1006   arg.properties = NULL;
1007   arg.auto_restart = process->auto_restart;
1008   //kill the old process
1009   SIMIX_process_kill(process,issuer);
1010   //start the new process
1011   smx_process_t new_process;
1012   if (simix_global->create_process_function) {
1013     new_process = simix_global->create_process_function(
1014                                           arg.argv[0],
1015                                           arg.code,
1016                                           arg.data,
1017                                           arg.hostname,
1018                                           arg.kill_time,
1019                                           arg.argc,
1020                                           arg.argv,
1021                                           arg.properties,
1022                                           arg.auto_restart,
1023                                           NULL);
1024   } else {
1025     new_process = simcall_process_create(
1026                            arg.argv[0],
1027                            arg.code,
1028                            arg.data,
1029                            arg.hostname,
1030                            arg.kill_time,
1031                            arg.argc,
1032                            arg.argv,
1033                            arg.properties,
1034                            arg.auto_restart);
1035
1036   }
1037   return new_process;
1038 }
1039
1040 void SIMIX_segment_index_set(smx_process_t proc, int index){
1041   proc->segment_index = index;
1042 }