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 "src/surf/surf_interface.hpp"
   8 #include "smx_private.h"
   9 #include "xbt/sysdep.h"
  10 #include "xbt/log.h"
  11 #include "xbt/dict.h"
  12 #include "mc/mc.h"
  13 #include "src/mc/mc_replay.h"
  14 #include "src/mc/mc_client.h"
  15 #include "src/simix/smx_private.hpp"
  16 #include "src/msg/msg_private.h"
  17
  18 #ifdef HAVE_SMPI
  19 #include "src/smpi/private.h"
  20 #endif
  21
  22 XBT_LOG_NEW_DEFAULT_SUBCATEGORY(simix_process, simix,
  23                                 "Logging specific to SIMIX (process)");
  24
  25 unsigned long simix_process_maxpid = 0;
  26
  27 /**
  28  * \brief Returns the current agent.
  29  *
  30  * This functions returns the currently running SIMIX process.
  31  *
  32  * \return The SIMIX process
  33  */
  34 smx_process_t SIMIX_process_self(void)
  35 {
  36   smx_context_t self_context = SIMIX_context_self();
  37
  38   return self_context ? SIMIX_context_get_process(self_context) : NULL;
  39 }
  40
  41 /**
  42  * \brief Returns whether a process has pending asynchronous communications.
  43  * \return true if there are asynchronous communications in this process
  44  */
  45 int SIMIX_process_has_pending_comms(smx_process_t process) {
  46
  47   return xbt_fifo_size(process->comms) > 0;
  48 }
  49
  50 /**
  51  * \brief Moves a process to the list of processes to destroy.
  52  */
  53 void SIMIX_process_cleanup(smx_process_t process)
  54 {
  55   XBT_DEBUG("Cleanup process %s (%p), waiting synchro %p",
  56       process->name, process, process->waiting_synchro);
  57
  58   SIMIX_process_on_exit_runall(process);
  59
  60   /* Unregister from the kill timer if any */
  61   if (process->kill_timer != NULL)
  62       SIMIX_timer_remove(process->kill_timer);
  63
  64   xbt_os_mutex_acquire(simix_global->mutex);
  65
  66   /* cancel non-blocking communications */
  67   smx_synchro_t synchro;
  68   while ((synchro = (smx_synchro_t) xbt_fifo_pop(process->comms))) {
  69
  70     /* make sure no one will finish the comm after this process is destroyed,
  71      * because src_proc or dst_proc would be an invalid pointer */
  72     SIMIX_comm_cancel(synchro);
  73
  74     if (synchro->comm.src_proc == process) {
  75       XBT_DEBUG("Found an unfinished send comm %p (detached = %d), state %d, src = %p, dst = %p",
  76           synchro, synchro->comm.detached, (int)synchro->state, synchro->comm.src_proc, synchro->comm.dst_proc);
  77       synchro->comm.src_proc = NULL;
  78
  79       /* I'm not supposed to destroy a detached comm from the sender side, */
  80       if (!synchro->comm.detached)
  81         SIMIX_comm_destroy(synchro);
  82       else
  83         XBT_DEBUG("Don't destroy it since it's a detached comm");
  84
  85     }
  86     else if (synchro->comm.dst_proc == process){
  87       XBT_DEBUG("Found an unfinished recv comm %p, state %d, src = %p, dst = %p",
  88           synchro, (int)synchro->state, synchro->comm.src_proc, synchro->comm.dst_proc);
  89       synchro->comm.dst_proc = NULL;
  90
  91       if (synchro->comm.detached && synchro->comm.refcount == 1
  92           && synchro->comm.src_proc != NULL) {
  93         /* the comm will be freed right now, remove it from the sender */
  94         xbt_fifo_remove(synchro->comm.src_proc->comms, synchro);
  95       }
  96       SIMIX_comm_destroy(synchro);
  97     }
  98     else {
  99       xbt_die("Communication synchro %p is in my list but I'm not the sender "
 100           "or the receiver", synchro);
 101     }
 102   }
 103
 104   XBT_DEBUG("%p should not be run anymore",process);
 105   xbt_swag_remove(process, simix_global->process_list);
 106   xbt_swag_remove(process, sg_host_simix(process->host)->process_list);
 107   xbt_swag_insert(process, simix_global->process_to_destroy);
 108   process->context->iwannadie = 0;
 109
 110   xbt_os_mutex_release(simix_global->mutex);
 111 }
 112
 113 /**
 114  * Garbage collection
 115  *
 116  * Should be called some time to time to free the memory allocated for processes
 117  * that have finished (or killed).
 118  */
 119 void SIMIX_process_empty_trash(void)
 120 {
 121   smx_process_t process = NULL;
 122
 123   while ((process = (smx_process_t) xbt_swag_extract(simix_global->process_to_destroy))) {
 124     XBT_DEBUG("Getting rid of %p",process);
 125
 126     SIMIX_context_free(process->context);
 127
 128     /* Free the exception allocated at creation time */
 129     free(process->running_ctx);
 130     xbt_dict_free(&process->properties);
 131
 132     xbt_fifo_free(process->comms);
 133
 134     xbt_dynar_free(&process->on_exit);
 135
 136     xbt_free(process->name);
 137     xbt_free(process);
 138   }
 139 }
 140
 141 /**
 142  * \brief Creates and runs the maestro process
 143  */
 144 void SIMIX_create_maestro_process()
 145 {
 146   smx_process_t maestro = NULL;
 147
 148   /* Create maestro process and intilialize it */
 149   maestro = xbt_new0(s_smx_process_t, 1);
 150   maestro->pid = simix_process_maxpid++;
 151   maestro->ppid = -1;
 152   maestro->name = (char *) "";
 153   maestro->running_ctx = (xbt_running_ctx_t*) xbt_malloc0(sizeof(xbt_running_ctx_t));
 154   XBT_RUNNING_CTX_INITIALIZE(maestro->running_ctx);
 155   maestro->context = SIMIX_context_new(NULL, 0, NULL, NULL, maestro);
 156   maestro->simcall.issuer = maestro;
 157   simix_global->maestro_process = maestro;
 158 }
 159 /**
 160  * \brief Stops a process.
 161  *
 162  * Stops the process, execute all the registered on_exit functions,
 163  * register it to the list of the process to restart if needed
 164  * and stops its context.
 165  */
 166 void SIMIX_process_stop(smx_process_t arg) {
 167   /* execute the on_exit functions */
 168   SIMIX_process_on_exit_runall(arg);
 169   /* Add the process to the list of process to restart, only if the host is down */
 170   if (arg->auto_restart && arg->host->is_off()) {
 171     SIMIX_host_add_auto_restart_process(arg->host,arg->name,arg->code, arg->data,
 172                                         sg_host_get_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_get_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 static void kill_process(void* process)
 220 {
 221   simix_global->kill_process_function((smx_process_t) process);
 222 }
 223
 224 /**
 225  * \brief Internal function to create a process.
 226  *
 227  * This function actually creates the process.
 228  * It may be called when a SIMCALL_PROCESS_CREATE simcall occurs,
 229  * or directly for SIMIX internal purposes. The sure thing is that it's called from maestro context.
 230  *
 231  * \return the process created
 232  */
 233 smx_process_t SIMIX_process_create(
 234                           const char *name,
 235                           xbt_main_func_t code,
 236                           void *data,
 237                           const char *hostname,
 238                           double kill_time,
 239                           int argc, char **argv,
 240                           xbt_dict_t properties,
 241                           int auto_restart,
 242                           smx_process_t parent_process)
 243 {
 244   smx_process_t process = NULL;
 245   sg_host_t host = sg_host_by_name(hostname);
 246
 247   XBT_DEBUG("Start process %s on host '%s'", name, hostname);
 248
 249   if (host->is_off()) {
 250     int i;
 251     XBT_WARN("Cannot launch process '%s' on failed host '%s'", name,
 252           hostname);
 253     for (i = 0; i < argc; i++)
 254       xbt_free(argv[i]);
 255     xbt_free(argv);
 256   }
 257   else {
 258     process = xbt_new0(s_smx_process_t, 1);
 259
 260     xbt_assert(((code != NULL) && (host != NULL)), "Invalid parameters");
 261     /* Process data */
 262     process->pid = simix_process_maxpid++;
 263     process->name = xbt_strdup(name);
 264     process->host = host;
 265     process->data = data;
 266     process->comms = xbt_fifo_new();
 267     process->simcall.issuer = process;
 268     /* Initiliaze data segment to default value */
 269     SIMIX_segment_index_set(process, -1);
 270
 271      if (parent_process != NULL) {
 272        process->ppid = SIMIX_process_get_PID(parent_process);
 273        /* SMPI process have their own data segment and
 274           each other inherit from their father */
 275 #ifdef HAVE_SMPI
 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 #endif
 284      } else {
 285        process->ppid = -1;
 286      }
 287
 288     /* Process data for auto-restart */
 289     process->auto_restart = auto_restart;
 290     process->code = code;
 291     process->argc = argc;
 292     process->argv = argv;
 293
 294
 295     XBT_VERB("Create context %s", process->name);
 296     process->context = SIMIX_context_new(code, argc, argv, simix_global->cleanup_process_function, process);
 297
 298     process->running_ctx = (xbt_running_ctx_t*) xbt_malloc0(sizeof(xbt_running_ctx_t));
 299     XBT_RUNNING_CTX_INITIALIZE(process->running_ctx);
 300
 301     if(MC_is_active()){
 302       MC_ignore_heap(process->running_ctx, sizeof(*process->running_ctx));
 303     }
 304
 305     /* Add properties */
 306     process->properties = properties;
 307
 308     /* Add the process to it's host process list */
 309     xbt_swag_insert(process, sg_host_simix(host)->process_list);
 310
 311     XBT_DEBUG("Start context '%s'", process->name);
 312
 313     /* Now insert it in the global process list and in the process to run list */
 314     xbt_swag_insert(process, simix_global->process_list);
 315     XBT_DEBUG("Inserting %s(%s) in the to_run list", process->name, sg_host_get_name(host));
 316     xbt_dynar_push_as(simix_global->process_to_run, smx_process_t, process);
 317
 318     if (kill_time > SIMIX_get_clock() && simix_global->kill_process_function) {
 319       XBT_DEBUG("Process %s(%s) will be kill at time %f", process->name,
 320           sg_host_get_name(process->host), kill_time);
 321       process->kill_timer = SIMIX_timer_set(kill_time, kill_process, process);
 322     }
 323
 324     /* Tracing the process creation */
 325     TRACE_msg_process_create(process->name, process->pid, process->host);
 326   }
 327   return process;
 328 }
 329
 330 /**
 331  * \brief Executes the processes from simix_global->process_to_run.
 332  *
 333  * The processes of simix_global->process_to_run are run (in parallel if
 334  * possible).  On exit, simix_global->process_to_run is empty, and
 335  * simix_global->process_that_ran contains the list of processes that just ran.
 336  * The two lists are swapped so, be careful when using them before and after a
 337  * call to this function.
 338  */
 339 void SIMIX_process_runall(void)
 340 {
 341   SIMIX_context_runall();
 342
 343   xbt_dynar_t tmp = simix_global->process_that_ran;
 344   simix_global->process_that_ran = simix_global->process_to_run;
 345   simix_global->process_to_run = tmp;
 346   xbt_dynar_reset(simix_global->process_to_run);
 347 }
 348
 349 void simcall_HANDLER_process_kill(smx_simcall_t simcall, smx_process_t process) {
 350   SIMIX_process_kill(process, simcall->issuer);
 351 }
 352 /**
 353  * \brief Internal function to kill a SIMIX process.
 354  *
 355  * This function may be called when a SIMCALL_PROCESS_KILL simcall occurs,
 356  * or directly for SIMIX internal purposes.
 357  *
 358  * \param process poor victim
 359  * \param issuer the process which has sent the PROCESS_KILL. Important to not schedule twice the same process.
 360  */
 361 void SIMIX_process_kill(smx_process_t process, smx_process_t issuer) {
 362
 363   XBT_DEBUG("Killing process %s on %s", process->name, sg_host_get_name(process->host));
 364
 365   process->context->iwannadie = 1;
 366   process->blocked = 0;
 367   process->suspended = 0;
 368   process->doexception = 0;
 369
 370   /* destroy the blocking synchro if any */
 371   if (process->waiting_synchro) {
 372
 373     switch (process->waiting_synchro->type) {
 374
 375     case SIMIX_SYNC_EXECUTE:
 376     case SIMIX_SYNC_PARALLEL_EXECUTE:
 377       SIMIX_execution_destroy(process->waiting_synchro);
 378       break;
 379
 380     case SIMIX_SYNC_COMMUNICATE:
 381       xbt_fifo_remove(process->comms, process->waiting_synchro);
 382       SIMIX_comm_cancel(process->waiting_synchro);
 383       xbt_fifo_remove(process->waiting_synchro->simcalls, &process->simcall);
 384       SIMIX_comm_destroy(process->waiting_synchro);
 385       break;
 386
 387     case SIMIX_SYNC_SLEEP:
 388       SIMIX_process_sleep_destroy(process->waiting_synchro);
 389       break;
 390
 391     case SIMIX_SYNC_JOIN:
 392       SIMIX_process_sleep_destroy(process->waiting_synchro);
 393       break;
 394
 395     case SIMIX_SYNC_SYNCHRO:
 396       SIMIX_synchro_stop_waiting(process, &process->simcall);
 397       SIMIX_synchro_destroy(process->waiting_synchro);
 398       break;
 399
 400     case SIMIX_SYNC_IO:
 401       SIMIX_io_destroy(process->waiting_synchro);
 402       break;
 403
 404     }
 405
 406     process->waiting_synchro = NULL;
 407   }
 408   if(!xbt_dynar_member(simix_global->process_to_run, &(process)) && process != issuer) {
 409     XBT_DEBUG("Inserting %s in the to_run list", process->name);
 410     xbt_dynar_push_as(simix_global->process_to_run, smx_process_t, process);
 411   }
 412
 413 }
 414
 415 /** @brief Ask another process to raise the given exception
 416  *
 417  * @param cat category of exception
 418  * @param value value associated to the exception
 419  * @param msg string information associated to the exception
 420  */
 421 void SIMIX_process_throw(smx_process_t process, xbt_errcat_t cat, int value, const char *msg) {
 422   SMX_EXCEPTION(process, cat, value, msg);
 423
 424   if (process->suspended)
 425     SIMIX_process_resume(process,SIMIX_process_self());
 426
 427   /* cancel the blocking synchro if any */
 428   if (process->waiting_synchro) {
 429
 430     switch (process->waiting_synchro->type) {
 431
 432     case SIMIX_SYNC_EXECUTE:
 433     case SIMIX_SYNC_PARALLEL_EXECUTE:
 434       SIMIX_execution_cancel(process->waiting_synchro);
 435       break;
 436
 437     case SIMIX_SYNC_COMMUNICATE:
 438       xbt_fifo_remove(process->comms, process->waiting_synchro);
 439       SIMIX_comm_cancel(process->waiting_synchro);
 440       break;
 441
 442     case SIMIX_SYNC_SLEEP:
 443     case SIMIX_SYNC_JOIN:
 444       SIMIX_process_sleep_destroy(process->waiting_synchro);
 445       if (!xbt_dynar_member(simix_global->process_to_run, &(process)) && process != SIMIX_process_self()) {
 446         XBT_DEBUG("Inserting %s in the to_run list", process->name);
 447         xbt_dynar_push_as(simix_global->process_to_run, smx_process_t, process);
 448       }
 449       break;
 450
 451     case SIMIX_SYNC_SYNCHRO:
 452       SIMIX_synchro_stop_waiting(process, &process->simcall);
 453       break;
 454
 455     case SIMIX_SYNC_IO:
 456       SIMIX_io_destroy(process->waiting_synchro);
 457       break;
 458
 459     }
 460   }
 461   process->waiting_synchro = NULL;
 462
 463 }
 464
 465 void simcall_HANDLER_process_killall(smx_simcall_t simcall, int reset_pid) {
 466   SIMIX_process_killall(simcall->issuer, reset_pid);
 467 }
 468 /**
 469  * \brief Kills all running processes.
 470  * \param issuer this one will not be killed
 471  */
 472 void SIMIX_process_killall(smx_process_t issuer, int reset_pid)
 473 {
 474   smx_process_t p = NULL;
 475
 476   while ((p = (smx_process_t) xbt_swag_extract(simix_global->process_list))) {
 477     if (p != issuer) {
 478       SIMIX_process_kill(p,issuer);
 479     }
 480   }
 481
 482   if (reset_pid > 0)
 483     simix_process_maxpid = reset_pid;
 484
 485   SIMIX_context_runall();
 486
 487   SIMIX_process_empty_trash();
 488 }
 489
 490 void simcall_HANDLER_process_set_host(smx_simcall_t simcall, smx_process_t process, sg_host_t dest)
 491 {
 492   process->new_host = dest;
 493 }
 494 void SIMIX_process_change_host(smx_process_t process,
 495              sg_host_t dest)
 496 {
 497   xbt_assert((process != NULL), "Invalid parameters");
 498   xbt_swag_remove(process, sg_host_simix(process->host)->process_list);
 499   process->host = dest;
 500   xbt_swag_insert(process, sg_host_simix(dest)->process_list);
 501 }
 502
 503
 504 void simcall_HANDLER_process_suspend(smx_simcall_t simcall, smx_process_t process)
 505 {
 506   smx_synchro_t sync_suspend =
 507       SIMIX_process_suspend(process, simcall->issuer);
 508
 509   if (process != simcall->issuer) {
 510     SIMIX_simcall_answer(simcall);
 511   } else {
 512     xbt_fifo_push(sync_suspend->simcalls, simcall);
 513     process->waiting_synchro = sync_suspend;
 514     SIMIX_execution_suspend(process->waiting_synchro);
 515   }
 516   /* If we are suspending ourselves, then just do not finish the simcall now */
 517 }
 518
 519 smx_synchro_t SIMIX_process_suspend(smx_process_t process, smx_process_t issuer)
 520 {
 521   xbt_assert((process != NULL), "Invalid parameters");
 522
 523   if (process->suspended) {
 524     XBT_DEBUG("Process '%s' is already suspended", process->name);
 525     return NULL;
 526   }
 527
 528   process->suspended = 1;
 529
 530   /* If we are suspending another process, and it is waiting on a sync,
 531      suspend its synchronization. */
 532   if (process != issuer) {
 533
 534     if (process->waiting_synchro) {
 535
 536       switch (process->waiting_synchro->type) {
 537
 538         case SIMIX_SYNC_EXECUTE:
 539         case SIMIX_SYNC_PARALLEL_EXECUTE:
 540           SIMIX_execution_suspend(process->waiting_synchro);
 541           break;
 542
 543         case SIMIX_SYNC_COMMUNICATE:
 544           SIMIX_comm_suspend(process->waiting_synchro);
 545           break;
 546
 547         case SIMIX_SYNC_SLEEP:
 548           SIMIX_process_sleep_suspend(process->waiting_synchro);
 549           break;
 550
 551         case SIMIX_SYNC_SYNCHRO:
 552           /* Suspension is delayed to when the process is rescheduled. */
 553           break;
 554
 555         default:
 556           xbt_die("Internal error in SIMIX_process_suspend: unexpected synchronization type %d",
 557               (int)process->waiting_synchro->type);
 558       }
 559       return NULL;
 560     } else {
 561       /* Suspension is delayed to when the process is rescheduled. */
 562       return NULL;
 563     }
 564   } else {
 565     /* FIXME: computation size is zero. Is it okay that bound is zero ? */
 566     return SIMIX_execution_start(process, "suspend", 0.0, 1.0, 0.0, 0);
 567   }
 568 }
 569
 570 void simcall_HANDLER_process_resume(smx_simcall_t simcall, smx_process_t process){
 571   SIMIX_process_resume(process, simcall->issuer);
 572 }
 573
 574 void SIMIX_process_resume(smx_process_t process, smx_process_t issuer)
 575 {
 576   XBT_IN("process = %p, issuer = %p", process, issuer);
 577
 578   if(process->context->iwannadie) {
 579     XBT_VERB("Ignoring request to suspend a process that is currently dying.");
 580     return;
 581   }
 582
 583   if(!process->suspended) return;
 584   process->suspended = 0;
 585
 586   /* If we are resuming another process, resume the synchronization it was waiting for
 587      if any. Otherwise add it to the list of process to run in the next round. */
 588   if (process != issuer) {
 589
 590     if (process->waiting_synchro) {
 591
 592       switch (process->waiting_synchro->type) {
 593
 594         case SIMIX_SYNC_EXECUTE:
 595         case SIMIX_SYNC_PARALLEL_EXECUTE:
 596           SIMIX_execution_resume(process->waiting_synchro);
 597           break;
 598
 599         case SIMIX_SYNC_COMMUNICATE:
 600           SIMIX_comm_resume(process->waiting_synchro);
 601           break;
 602
 603         case SIMIX_SYNC_SLEEP:
 604           SIMIX_process_sleep_resume(process->waiting_synchro);
 605           break;
 606
 607         case SIMIX_SYNC_SYNCHRO:
 608           /* I cannot resume it now. This is delayed to when the process is rescheduled at
 609            * the end of the synchro. */
 610           break;
 611
 612         default:
 613           xbt_die("Internal error in SIMIX_process_resume: unexpected synchronization type %d",
 614               (int)process->waiting_synchro->type);
 615       }
 616     }
 617   } else XBT_WARN("Strange. Process %p is trying to resume himself.", issuer);
 618
 619   XBT_OUT();
 620 }
 621
 622 int SIMIX_process_get_maxpid(void) {
 623   return simix_process_maxpid;
 624 }
 625
 626 int SIMIX_process_count(void)
 627 {
 628   return xbt_swag_size(simix_global->process_list);
 629 }
 630
 631 int SIMIX_process_get_PID(smx_process_t self){
 632   if (self == NULL)
 633     return 0;
 634   else
 635     return self->pid;
 636 }
 637
 638 int SIMIX_process_get_PPID(smx_process_t self){
 639   if (self == NULL)
 640     return 0;
 641   else
 642     return self->ppid;
 643 }
 644
 645 void* SIMIX_process_self_get_data(smx_process_t self)
 646 {
 647   xbt_assert(self == SIMIX_process_self(), "This is not the current process");
 648
 649   if (!self) {
 650     return NULL;
 651   }
 652   return SIMIX_process_get_data(self);
 653 }
 654
 655 void SIMIX_process_self_set_data(smx_process_t self, void *data)
 656 {
 657   xbt_assert(self == SIMIX_process_self(), "This is not the current process");
 658
 659   SIMIX_process_set_data(self, data);
 660 }
 661
 662 void* SIMIX_process_get_data(smx_process_t process)
 663 {
 664   return process->data;
 665 }
 666
 667 void SIMIX_process_set_data(smx_process_t process, void *data)
 668 {
 669   process->data = data;
 670 }
 671
 672 sg_host_t SIMIX_process_get_host(smx_process_t process)
 673 {
 674   return process->host;
 675 }
 676
 677 xbt_main_func_t SIMIX_process_get_code(void){
 678   return SIMIX_process_self()->code;
 679 }
 680
 681 /* needs to be public and without simcall because it is called
 682    by exceptions and logging events */
 683 const char* SIMIX_process_self_get_name(void) {
 684
 685   smx_process_t process = SIMIX_process_self();
 686   if (process == NULL || process == simix_global->maestro_process)
 687     return "maestro";
 688
 689   return SIMIX_process_get_name(process);
 690 }
 691
 692 const char* SIMIX_process_get_name(smx_process_t process)
 693 {
 694   return process->name;
 695 }
 696
 697 smx_process_t SIMIX_process_get_by_name(const char* name)
 698 {
 699   smx_process_t proc;
 700
 701   xbt_swag_foreach(proc, simix_global->process_list)
 702   {
 703     if(!strcmp(name, proc->name))
 704       return proc;
 705   }
 706   return NULL;
 707 }
 708
 709 int SIMIX_process_is_suspended(smx_process_t process)
 710 {
 711   return process->suspended;
 712 }
 713
 714 xbt_dict_t SIMIX_process_get_properties(smx_process_t process)
 715 {
 716   return process->properties;
 717 }
 718
 719 void simcall_HANDLER_process_join(smx_simcall_t simcall, smx_process_t process, double timeout)
 720 {
 721   smx_synchro_t sync = SIMIX_process_join(simcall->issuer, process, timeout);
 722   xbt_fifo_push(sync->simcalls, simcall);
 723   simcall->issuer->waiting_synchro = sync;
 724 }
 725
 726 static int SIMIX_process_join_finish(smx_process_exit_status_t status, smx_synchro_t sync){
 727   if (sync->sleep.surf_sleep) {
 728     sync->sleep.surf_sleep->cancel();
 729
 730     smx_simcall_t simcall;
 731     while ((simcall = (smx_simcall_t) xbt_fifo_shift(sync->simcalls))) {
 732       simcall_process_sleep__set__result(simcall, SIMIX_DONE);
 733       simcall->issuer->waiting_synchro = NULL;
 734       if (simcall->issuer->suspended) {
 735         XBT_DEBUG("Wait! This process is suspended and can't wake up now.");
 736         simcall->issuer->suspended = 0;
 737         simcall_HANDLER_process_suspend(simcall, simcall->issuer);
 738       } else {
 739         SIMIX_simcall_answer(simcall);
 740       }
 741     }
 742     sync->sleep.surf_sleep->unref();
 743     sync->sleep.surf_sleep = NULL;
 744   }
 745   xbt_mallocator_release(simix_global->synchro_mallocator, sync);
 746   return 0;
 747 }
 748
 749 smx_synchro_t SIMIX_process_join(smx_process_t issuer, smx_process_t process, double timeout)
 750 {
 751   smx_synchro_t res = SIMIX_process_sleep(issuer, timeout);
 752   res->type = SIMIX_SYNC_JOIN;
 753   SIMIX_process_on_exit(process, (int_f_pvoid_pvoid_t)SIMIX_process_join_finish, res);
 754   return res;
 755 }
 756
 757 void simcall_HANDLER_process_sleep(smx_simcall_t simcall, double duration)
 758 {
 759   if (MC_is_active() || MC_record_replay_is_active()) {
 760     MC_process_clock_add(simcall->issuer, duration);
 761     simcall_process_sleep__set__result(simcall, SIMIX_DONE);
 762     SIMIX_simcall_answer(simcall);
 763     return;
 764   }
 765   smx_synchro_t sync = SIMIX_process_sleep(simcall->issuer, duration);
 766   xbt_fifo_push(sync->simcalls, simcall);
 767   simcall->issuer->waiting_synchro = sync;
 768 }
 769
 770 smx_synchro_t SIMIX_process_sleep(smx_process_t process, double duration)
 771 {
 772   sg_host_t host = process->host;
 773
 774   /* check if the host is active */
 775   if (host->is_off()) {
 776     THROWF(host_error, 0, "Host %s failed, you cannot call this function",
 777            sg_host_get_name(host));
 778   }
 779
 780   smx_synchro_t synchro = (smx_synchro_t) xbt_mallocator_get(simix_global->synchro_mallocator);
 781   synchro->type = SIMIX_SYNC_SLEEP;
 782   synchro->name = NULL;
 783   synchro->category = NULL;
 784
 785   synchro->sleep.host = host;
 786   synchro->sleep.surf_sleep = surf_host_sleep(host, duration);
 787
 788   synchro->sleep.surf_sleep->setData(synchro);
 789   XBT_DEBUG("Create sleep synchronization %p", synchro);
 790
 791   return synchro;
 792 }
 793
 794 void SIMIX_post_process_sleep(smx_synchro_t synchro)
 795 {
 796   smx_simcall_t simcall;
 797   e_smx_state_t state;
 798   xbt_assert(synchro->type == SIMIX_SYNC_SLEEP || synchro->type == SIMIX_SYNC_JOIN);
 799
 800   while ((simcall = (smx_simcall_t) xbt_fifo_shift(synchro->simcalls))) {
 801
 802     switch (synchro->sleep.surf_sleep->getState()){
 803       case SURF_ACTION_FAILED:
 804         simcall->issuer->context->iwannadie = 1;
 805         //SMX_EXCEPTION(simcall->issuer, host_error, 0, "Host failed");
 806         state = SIMIX_SRC_HOST_FAILURE;
 807         break;
 808
 809       case SURF_ACTION_DONE:
 810         state = SIMIX_DONE;
 811         break;
 812
 813       default:
 814         THROW_IMPOSSIBLE;
 815         break;
 816     }
 817     if (simcall->issuer->host->is_off()) {
 818       simcall->issuer->context->iwannadie = 1;
 819     }
 820     simcall_process_sleep__set__result(simcall, state);
 821     simcall->issuer->waiting_synchro = NULL;
 822     if (simcall->issuer->suspended) {
 823       XBT_DEBUG("Wait! This process is suspended and can't wake up now.");
 824       simcall->issuer->suspended = 0;
 825       simcall_HANDLER_process_suspend(simcall, simcall->issuer);
 826     } else {
 827       SIMIX_simcall_answer(simcall);
 828     }
 829   }
 830
 831   SIMIX_process_sleep_destroy(synchro);
 832 }
 833
 834 void SIMIX_process_sleep_destroy(smx_synchro_t synchro)
 835 {
 836   XBT_DEBUG("Destroy synchro %p", synchro);
 837   xbt_assert(synchro->type == SIMIX_SYNC_SLEEP || synchro->type == SIMIX_SYNC_JOIN);
 838
 839   if (synchro->sleep.surf_sleep) {
 840     synchro->sleep.surf_sleep->unref();
 841     synchro->sleep.surf_sleep = NULL;
 842   }
 843   if (synchro->type == SIMIX_SYNC_SLEEP)
 844     xbt_mallocator_release(simix_global->synchro_mallocator, synchro);
 845 }
 846
 847 void SIMIX_process_sleep_suspend(smx_synchro_t synchro)
 848 {
 849   xbt_assert(synchro->type == SIMIX_SYNC_SLEEP);
 850   synchro->sleep.surf_sleep->suspend();
 851 }
 852
 853 void SIMIX_process_sleep_resume(smx_synchro_t synchro)
 854 {
 855   XBT_DEBUG("Synchro state is %d on process_sleep_resume.", synchro->state);
 856   xbt_assert(synchro->type == SIMIX_SYNC_SLEEP);
 857   synchro->sleep.surf_sleep->resume();
 858 }
 859
 860 /**
 861  * \brief Calling this function makes the process to yield.
 862  *
 863  * Only the current process can call this function, giving back the control to
 864  * maestro.
 865  *
 866  * \param self the current process
 867  */
 868 void SIMIX_process_yield(smx_process_t self)
 869 {
 870   XBT_DEBUG("Yield process '%s'", self->name);
 871
 872   /* Go into sleep and return control to maestro */
 873   SIMIX_context_suspend(self->context);
 874
 875   /* Ok, maestro returned control to us */
 876   XBT_DEBUG("Control returned to me: '%s'", self->name);
 877
 878   if (self->new_host) {
 879     SIMIX_process_change_host(self, self->new_host);
 880     self->new_host = NULL;
 881   }
 882
 883   if (self->context->iwannadie){
 884     XBT_DEBUG("I wanna die!");
 885     SIMIX_process_stop(self);
 886   }
 887
 888   if (self->suspended) {
 889     XBT_DEBUG("Hey! I'm suspended.");
 890     xbt_assert(!self->doexception, "Gasp! This exception may be lost by subsequent calls.");
 891     self->suspended = 0;
 892     SIMIX_process_suspend(self, self);
 893   }
 894
 895   if (self->doexception) {
 896     XBT_DEBUG("Wait, maestro left me an exception");
 897     self->doexception = 0;
 898     SMX_THROW();
 899   }
 900
 901   if(SMPI_switch_data_segment && self->segment_index != -1){
 902     SMPI_switch_data_segment(self->segment_index);
 903   }
 904 }
 905
 906 /* callback: context fetching */
 907 xbt_running_ctx_t *SIMIX_process_get_running_context(void)
 908 {
 909   return SIMIX_process_self()->running_ctx;
 910 }
 911
 912 /* callback: termination */
 913 void SIMIX_process_exception_terminate(xbt_ex_t * e)
 914 {
 915   xbt_ex_display(e);
 916   xbt_abort();
 917 }
 918
 919 smx_context_t SIMIX_process_get_context(smx_process_t p) {
 920   return p->context;
 921 }
 922
 923 void SIMIX_process_set_context(smx_process_t p,smx_context_t c) {
 924   p->context = c;
 925 }
 926
 927 /**
 928  * \brief Returns the list of processes to run.
 929  */
 930 xbt_dynar_t SIMIX_process_get_runnable(void)
 931 {
 932   return simix_global->process_to_run;
 933 }
 934
 935 /**
 936  * \brief Returns the process from PID.
 937  */
 938 smx_process_t SIMIX_process_from_PID(int PID)
 939 {
 940   smx_process_t proc;
 941   xbt_swag_foreach(proc, simix_global->process_list) {
 942    if (proc->pid == (unsigned long) PID)
 943     return proc;
 944   }
 945   return NULL;
 946 }
 947
 948 /** @brief returns a dynar containg all currently existing processes */
 949 xbt_dynar_t SIMIX_processes_as_dynar(void) {
 950   smx_process_t proc;
 951   xbt_dynar_t res = xbt_dynar_new(sizeof(smx_process_t),NULL);
 952   xbt_swag_foreach(proc, simix_global->process_list) {
 953     xbt_dynar_push(res,&proc);
 954   }
 955   return res;
 956 }
 957
 958
 959 void SIMIX_process_on_exit_runall(smx_process_t process) {
 960   s_smx_process_exit_fun_t exit_fun;
 961   smx_process_exit_status_t exit_status = (process->context->iwannadie) ?
 962                                          SMX_EXIT_FAILURE : SMX_EXIT_SUCCESS;
 963   while (!xbt_dynar_is_empty(process->on_exit)) {
 964     exit_fun = xbt_dynar_pop_as(process->on_exit,s_smx_process_exit_fun_t);
 965     (exit_fun.fun)((void*)exit_status, exit_fun.arg);
 966   }
 967 }
 968
 969 void SIMIX_process_on_exit(smx_process_t process, int_f_pvoid_pvoid_t fun, void *data) {
 970   xbt_assert(process, "current process not found: are you in maestro context ?");
 971
 972   if (!process->on_exit) {
 973     process->on_exit = xbt_dynar_new(sizeof(s_smx_process_exit_fun_t), NULL);
 974   }
 975
 976   s_smx_process_exit_fun_t exit_fun = {fun, data};
 977
 978   xbt_dynar_push_as(process->on_exit,s_smx_process_exit_fun_t,exit_fun);
 979 }
 980
 981 /**
 982  * \brief Sets the auto-restart status of the process.
 983  * If set to 1, the process will be automatically restarted when its host
 984  * comes back.
 985  */
 986 void SIMIX_process_auto_restart_set(smx_process_t process, int auto_restart) {
 987   process->auto_restart = auto_restart;
 988 }
 989
 990 smx_process_t simcall_HANDLER_process_restart(smx_simcall_t simcall, smx_process_t process) {
 991   return SIMIX_process_restart(process, simcall->issuer);
 992 }
 993 /** @brief Restart a process, starting it again from the beginning. */
 994 smx_process_t SIMIX_process_restart(smx_process_t process, smx_process_t issuer) {
 995   XBT_DEBUG("Restarting process %s on %s", process->name, sg_host_get_name(process->host));
 996   //retrieve the arguments of the old process
 997   //FIXME: Factorize this with SIMIX_host_add_auto_restart_process ?
 998   s_smx_process_arg_t arg;
 999   arg.code = process->code;
1000   arg.hostname = sg_host_get_name(process->host);
1001   arg.kill_time = SIMIX_timer_get_date(process->kill_timer);
1002   arg.argc = process->argc;
1003   arg.data = process->data;
1004   int i;
1005   arg.argv = xbt_new(char*,process->argc + 1);
1006   for (i = 0; i < arg.argc; i++) {
1007     arg.argv[i] = xbt_strdup(process->argv[i]);
1008   }
1009   arg.argv[process->argc] = NULL;
1010   arg.properties = NULL;
1011   arg.auto_restart = process->auto_restart;
1012   //kill the old process
1013   SIMIX_process_kill(process,issuer);
1014   //start the new process
1015   smx_process_t new_process;
1016   if (simix_global->create_process_function) {
1017     new_process = simix_global->create_process_function(
1018                                           arg.argv[0],
1019                                           arg.code,
1020                                           arg.data,
1021                                           arg.hostname,
1022                                           arg.kill_time,
1023                                           arg.argc,
1024                                           arg.argv,
1025                                           arg.properties,
1026                                           arg.auto_restart,
1027                                           NULL);
1028   } else {
1029     new_process = simcall_process_create(
1030                            arg.argv[0],
1031                            arg.code,
1032                            arg.data,
1033                            arg.hostname,
1034                            arg.kill_time,
1035                            arg.argc,
1036                            arg.argv,
1037                            arg.properties,
1038                            arg.auto_restart);
1039
1040   }
1041   return new_process;
1042 }
1043
1044 void SIMIX_segment_index_set(smx_process_t proc, int index){
1045   proc->segment_index = index;
1046 }