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   return;
 159 }
 160 /**
 161  * \brief Stops a process.
 162  *
 163  * Stops the process, execute all the registered on_exit functions,
 164  * register it to the list of the process to restart if needed
 165  * and stops its context.
 166  */
 167 void SIMIX_process_stop(smx_process_t arg) {
 168   /* execute the on_exit functions */
 169   SIMIX_process_on_exit_runall(arg);
 170   /* Add the process to the list of process to restart, only if the host is down */
 171   if (arg->auto_restart && arg->host->is_off()) {
 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 (host->is_off()) {
 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 #ifdef HAVE_SMPI
 277        if(smpi_privatize_global_variables){
 278          if( parent_process->pid != 0){
 279            SIMIX_segment_index_set(process, parent_process->segment_index);
 280          } else {
 281            SIMIX_segment_index_set(process, process->pid - 1);
 282          }
 283        }
 284 #endif
 285      } else {
 286        process->ppid = -1;
 287      }
 288
 289     /* Process data for auto-restart */
 290     process->auto_restart = auto_restart;
 291     process->code = code;
 292     process->argc = argc;
 293     process->argv = argv;
 294
 295
 296     XBT_VERB("Create context %s", process->name);
 297     process->context = SIMIX_context_new(code, argc, argv, simix_global->cleanup_process_function, process);
 298
 299     process->running_ctx = (xbt_running_ctx_t*) xbt_malloc0(sizeof(xbt_running_ctx_t));
 300     XBT_RUNNING_CTX_INITIALIZE(process->running_ctx);
 301
 302     if(MC_is_active()){
 303       MC_ignore_heap(process->running_ctx, sizeof(*process->running_ctx));
 304     }
 305
 306     /* Add properties */
 307     process->properties = properties;
 308
 309     /* Add the process to it's host process list */
 310     xbt_swag_insert(process, sg_host_simix(host)->process_list);
 311
 312     XBT_DEBUG("Start context '%s'", process->name);
 313
 314     /* Now insert it in the global process list and in the process to run list */
 315     xbt_swag_insert(process, simix_global->process_list);
 316     XBT_DEBUG("Inserting %s(%s) in the to_run list", process->name, sg_host_get_name(host));
 317     xbt_dynar_push_as(simix_global->process_to_run, smx_process_t, process);
 318
 319     if (kill_time > SIMIX_get_clock() && simix_global->kill_process_function) {
 320       XBT_DEBUG("Process %s(%s) will be kill at time %f", process->name,
 321           sg_host_get_name(process->host), kill_time);
 322       process->kill_timer = SIMIX_timer_set(kill_time, kill_process, process);
 323     }
 324
 325     /* Tracing the process creation */
 326     TRACE_msg_process_create(process->name, process->pid, process->host);
 327   }
 328   return process;
 329 }
 330
 331 /**
 332  * \brief Executes the processes from simix_global->process_to_run.
 333  *
 334  * The processes of simix_global->process_to_run are run (in parallel if
 335  * possible).  On exit, simix_global->process_to_run is empty, and
 336  * simix_global->process_that_ran contains the list of processes that just ran.
 337  * The two lists are swapped so, be careful when using them before and after a
 338  * call to this function.
 339  */
 340 void SIMIX_process_runall(void)
 341 {
 342   SIMIX_context_runall();
 343
 344   xbt_dynar_t tmp = simix_global->process_that_ran;
 345   simix_global->process_that_ran = simix_global->process_to_run;
 346   simix_global->process_to_run = tmp;
 347   xbt_dynar_reset(simix_global->process_to_run);
 348 }
 349
 350 void simcall_HANDLER_process_kill(smx_simcall_t simcall, smx_process_t process) {
 351   SIMIX_process_kill(process, simcall->issuer);
 352 }
 353 /**
 354  * \brief Internal function to kill a SIMIX process.
 355  *
 356  * This function may be called when a SIMCALL_PROCESS_KILL simcall occurs,
 357  * or directly for SIMIX internal purposes.
 358  *
 359  * \param process poor victim
 360  * \param issuer the process which has sent the PROCESS_KILL. Important to not schedule twice the same process.
 361  */
 362 void SIMIX_process_kill(smx_process_t process, smx_process_t issuer) {
 363
 364   XBT_DEBUG("Killing process %s on %s", process->name, sg_host_get_name(process->host));
 365
 366   process->context->iwannadie = 1;
 367   process->blocked = 0;
 368   process->suspended = 0;
 369   process->doexception = 0;
 370
 371   /* destroy the blocking synchro if any */
 372   if (process->waiting_synchro) {
 373
 374     switch (process->waiting_synchro->type) {
 375
 376     case SIMIX_SYNC_EXECUTE:
 377     case SIMIX_SYNC_PARALLEL_EXECUTE:
 378       SIMIX_process_execution_destroy(process->waiting_synchro);
 379       break;
 380
 381     case SIMIX_SYNC_COMMUNICATE:
 382       xbt_fifo_remove(process->comms, process->waiting_synchro);
 383       SIMIX_comm_cancel(process->waiting_synchro);
 384       xbt_fifo_remove(process->waiting_synchro->simcalls, &process->simcall);
 385       SIMIX_comm_destroy(process->waiting_synchro);
 386       break;
 387
 388     case SIMIX_SYNC_SLEEP:
 389       SIMIX_process_sleep_destroy(process->waiting_synchro);
 390       break;
 391
 392     case SIMIX_SYNC_JOIN:
 393       SIMIX_process_sleep_destroy(process->waiting_synchro);
 394       break;
 395
 396     case SIMIX_SYNC_SYNCHRO:
 397       SIMIX_synchro_stop_waiting(process, &process->simcall);
 398       SIMIX_synchro_destroy(process->waiting_synchro);
 399       break;
 400
 401     case SIMIX_SYNC_IO:
 402       SIMIX_io_destroy(process->waiting_synchro);
 403       break;
 404
 405     }
 406
 407     process->waiting_synchro = NULL;
 408   }
 409   if(!xbt_dynar_member(simix_global->process_to_run, &(process)) && process != issuer) {
 410     XBT_DEBUG("Inserting %s in the to_run list", process->name);
 411     xbt_dynar_push_as(simix_global->process_to_run, smx_process_t, process);
 412   }
 413
 414 }
 415
 416 /** @brief Ask another process to raise the given exception
 417  *
 418  * @param cat category of exception
 419  * @param value value associated to the exception
 420  * @param msg string information associated to the exception
 421  */
 422 void SIMIX_process_throw(smx_process_t process, xbt_errcat_t cat, int value, const char *msg) {
 423   SMX_EXCEPTION(process, cat, value, msg);
 424
 425   if (process->suspended)
 426     SIMIX_process_resume(process,SIMIX_process_self());
 427
 428   /* cancel the blocking synchro if any */
 429   if (process->waiting_synchro) {
 430
 431     switch (process->waiting_synchro->type) {
 432
 433     case SIMIX_SYNC_EXECUTE:
 434     case SIMIX_SYNC_PARALLEL_EXECUTE:
 435       SIMIX_process_execution_cancel(process->waiting_synchro);
 436       break;
 437
 438     case SIMIX_SYNC_COMMUNICATE:
 439       xbt_fifo_remove(process->comms, process->waiting_synchro);
 440       SIMIX_comm_cancel(process->waiting_synchro);
 441       break;
 442
 443     case SIMIX_SYNC_SLEEP:
 444     case SIMIX_SYNC_JOIN:
 445       SIMIX_process_sleep_destroy(process->waiting_synchro);
 446       if (!xbt_dynar_member(simix_global->process_to_run, &(process)) && process != SIMIX_process_self()) {
 447         XBT_DEBUG("Inserting %s in the to_run list", process->name);
 448         xbt_dynar_push_as(simix_global->process_to_run, smx_process_t, process);
 449       }
 450       break;
 451
 452     case SIMIX_SYNC_SYNCHRO:
 453       SIMIX_synchro_stop_waiting(process, &process->simcall);
 454       break;
 455
 456     case SIMIX_SYNC_IO:
 457       SIMIX_io_destroy(process->waiting_synchro);
 458       break;
 459
 460     }
 461   }
 462   process->waiting_synchro = NULL;
 463
 464 }
 465
 466 void simcall_HANDLER_process_killall(smx_simcall_t simcall, int reset_pid) {
 467   SIMIX_process_killall(simcall->issuer, reset_pid);
 468 }
 469 /**
 470  * \brief Kills all running processes.
 471  * \param issuer this one will not be killed
 472  */
 473 void SIMIX_process_killall(smx_process_t issuer, int reset_pid)
 474 {
 475   smx_process_t p = NULL;
 476
 477   while ((p = (smx_process_t) xbt_swag_extract(simix_global->process_list))) {
 478     if (p != issuer) {
 479       SIMIX_process_kill(p,issuer);
 480     }
 481   }
 482
 483   if (reset_pid > 0)
 484     simix_process_maxpid = reset_pid;
 485
 486   SIMIX_context_runall();
 487
 488   SIMIX_process_empty_trash();
 489 }
 490
 491 void simcall_HANDLER_process_set_host(smx_simcall_t simcall, smx_process_t process, sg_host_t dest)
 492 {
 493   process->new_host = dest;
 494 }
 495 void SIMIX_process_change_host(smx_process_t process,
 496              sg_host_t dest)
 497 {
 498   xbt_assert((process != NULL), "Invalid parameters");
 499   xbt_swag_remove(process, sg_host_simix(process->host)->process_list);
 500   process->host = dest;
 501   xbt_swag_insert(process, sg_host_simix(dest)->process_list);
 502 }
 503
 504
 505 void simcall_HANDLER_process_suspend(smx_simcall_t simcall, smx_process_t process)
 506 {
 507   smx_synchro_t sync_suspend =
 508       SIMIX_process_suspend(process, simcall->issuer);
 509
 510   if (process != simcall->issuer) {
 511     SIMIX_simcall_answer(simcall);
 512   } else {
 513     xbt_fifo_push(sync_suspend->simcalls, simcall);
 514     process->waiting_synchro = sync_suspend;
 515     SIMIX_host_execution_suspend(process->waiting_synchro);
 516   }
 517   /* If we are suspending ourselves, then just do not finish the simcall now */
 518 }
 519
 520 smx_synchro_t SIMIX_process_suspend(smx_process_t process, smx_process_t issuer)
 521 {
 522   xbt_assert((process != NULL), "Invalid parameters");
 523
 524   if (process->suspended) {
 525     XBT_DEBUG("Process '%s' is already suspended", process->name);
 526     return NULL;
 527   }
 528
 529   process->suspended = 1;
 530
 531   /* If we are suspending another process, and it is waiting on a sync,
 532      suspend its synchronization. */
 533   if (process != issuer) {
 534
 535     if (process->waiting_synchro) {
 536
 537       switch (process->waiting_synchro->type) {
 538
 539         case SIMIX_SYNC_EXECUTE:
 540         case SIMIX_SYNC_PARALLEL_EXECUTE:
 541           SIMIX_host_execution_suspend(process->waiting_synchro);
 542           break;
 543
 544         case SIMIX_SYNC_COMMUNICATE:
 545           SIMIX_comm_suspend(process->waiting_synchro);
 546           break;
 547
 548         case SIMIX_SYNC_SLEEP:
 549           SIMIX_process_sleep_suspend(process->waiting_synchro);
 550           break;
 551
 552         case SIMIX_SYNC_SYNCHRO:
 553           /* Suspension is delayed to when the process is rescheduled. */
 554           break;
 555
 556         default:
 557           xbt_die("Internal error in SIMIX_process_suspend: unexpected synchronization type %d",
 558               (int)process->waiting_synchro->type);
 559       }
 560       return NULL;
 561     } else {
 562       /* Suspension is delayed to when the process is rescheduled. */
 563       return NULL;
 564     }
 565   } else {
 566     /* FIXME: computation size is zero. Is it okay that bound is zero ? */
 567     return SIMIX_process_execute(process, "suspend", 0.0, 1.0, 0.0, 0);
 568   }
 569 }
 570
 571 void simcall_HANDLER_process_resume(smx_simcall_t simcall, smx_process_t process){
 572   SIMIX_process_resume(process, simcall->issuer);
 573 }
 574
 575 void SIMIX_process_resume(smx_process_t process, smx_process_t issuer)
 576 {
 577   XBT_IN("process = %p, issuer = %p", process, issuer);
 578
 579   if(process->context->iwannadie) {
 580     XBT_VERB("Ignoring request to suspend a process that is currently dying.");
 581     return;
 582   }
 583
 584   if(!process->suspended) return;
 585   process->suspended = 0;
 586
 587   /* If we are resuming another process, resume the synchronization it was waiting for
 588      if any. Otherwise add it to the list of process to run in the next round. */
 589   if (process != issuer) {
 590
 591     if (process->waiting_synchro) {
 592
 593       switch (process->waiting_synchro->type) {
 594
 595         case SIMIX_SYNC_EXECUTE:
 596         case SIMIX_SYNC_PARALLEL_EXECUTE:
 597           SIMIX_host_execution_resume(process->waiting_synchro);
 598           break;
 599
 600         case SIMIX_SYNC_COMMUNICATE:
 601           SIMIX_comm_resume(process->waiting_synchro);
 602           break;
 603
 604         case SIMIX_SYNC_SLEEP:
 605           SIMIX_process_sleep_resume(process->waiting_synchro);
 606           break;
 607
 608         case SIMIX_SYNC_SYNCHRO:
 609           /* I cannot resume it now. This is delayed to when the process is rescheduled at
 610            * the end of the synchro. */
 611           break;
 612
 613         default:
 614           xbt_die("Internal error in SIMIX_process_resume: unexpected synchronization type %d",
 615               (int)process->waiting_synchro->type);
 616       }
 617     }
 618   } else XBT_WARN("Strange. Process %p is trying to resume himself.", issuer);
 619
 620   XBT_OUT();
 621 }
 622
 623 int SIMIX_process_get_maxpid(void) {
 624   return simix_process_maxpid;
 625 }
 626
 627 int SIMIX_process_count(void)
 628 {
 629   return xbt_swag_size(simix_global->process_list);
 630 }
 631
 632 int SIMIX_process_get_PID(smx_process_t self){
 633   if (self == NULL)
 634     return 0;
 635   else
 636     return self->pid;
 637 }
 638
 639 int SIMIX_process_get_PPID(smx_process_t self){
 640   if (self == NULL)
 641     return 0;
 642   else
 643     return self->ppid;
 644 }
 645
 646 void* SIMIX_process_self_get_data(smx_process_t self)
 647 {
 648   xbt_assert(self == SIMIX_process_self(), "This is not the current process");
 649
 650   if (!self) {
 651     return NULL;
 652   }
 653   return SIMIX_process_get_data(self);
 654 }
 655
 656 void SIMIX_process_self_set_data(smx_process_t self, void *data)
 657 {
 658   xbt_assert(self == SIMIX_process_self(), "This is not the current process");
 659
 660   SIMIX_process_set_data(self, data);
 661 }
 662
 663 void* SIMIX_process_get_data(smx_process_t process)
 664 {
 665   return process->data;
 666 }
 667
 668 void SIMIX_process_set_data(smx_process_t process, void *data)
 669 {
 670   process->data = data;
 671 }
 672
 673 sg_host_t SIMIX_process_get_host(smx_process_t process)
 674 {
 675   return process->host;
 676 }
 677
 678 xbt_main_func_t SIMIX_process_get_code(void){
 679   return SIMIX_process_self()->code;
 680 }
 681
 682 /* needs to be public and without simcall because it is called
 683    by exceptions and logging events */
 684 const char* SIMIX_process_self_get_name(void) {
 685
 686   smx_process_t process = SIMIX_process_self();
 687   if (process == NULL || process == simix_global->maestro_process)
 688     return "";
 689
 690   return SIMIX_process_get_name(process);
 691 }
 692
 693 const char* SIMIX_process_get_name(smx_process_t process)
 694 {
 695   return process->name;
 696 }
 697
 698 smx_process_t SIMIX_process_get_by_name(const char* name)
 699 {
 700   smx_process_t proc;
 701
 702   xbt_swag_foreach(proc, simix_global->process_list)
 703   {
 704     if(!strcmp(name, proc->name))
 705       return proc;
 706   }
 707   return NULL;
 708 }
 709
 710 int SIMIX_process_is_suspended(smx_process_t process)
 711 {
 712   return process->suspended;
 713 }
 714
 715 xbt_dict_t SIMIX_process_get_properties(smx_process_t process)
 716 {
 717   return process->properties;
 718 }
 719
 720 void simcall_HANDLER_process_join(smx_simcall_t simcall, smx_process_t process, double timeout)
 721 {
 722   smx_synchro_t sync = SIMIX_process_join(simcall->issuer, process, timeout);
 723   xbt_fifo_push(sync->simcalls, simcall);
 724   simcall->issuer->waiting_synchro = sync;
 725 }
 726
 727 static int SIMIX_process_join_finish(smx_process_exit_status_t status, smx_synchro_t sync){
 728   if (sync->sleep.surf_sleep) {
 729     sync->sleep.surf_sleep->cancel();
 730
 731     smx_simcall_t simcall;
 732     while ((simcall = (smx_simcall_t) xbt_fifo_shift(sync->simcalls))) {
 733       simcall_process_sleep__set__result(simcall, SIMIX_DONE);
 734       simcall->issuer->waiting_synchro = NULL;
 735       if (simcall->issuer->suspended) {
 736         XBT_DEBUG("Wait! This process is suspended and can't wake up now.");
 737         simcall->issuer->suspended = 0;
 738         simcall_HANDLER_process_suspend(simcall, simcall->issuer);
 739       } else {
 740         SIMIX_simcall_answer(simcall);
 741       }
 742     }
 743     sync->sleep.surf_sleep->unref();
 744     sync->sleep.surf_sleep = NULL;
 745   }
 746   xbt_mallocator_release(simix_global->synchro_mallocator, sync);
 747   return 0;
 748 }
 749
 750 smx_synchro_t SIMIX_process_join(smx_process_t issuer, smx_process_t process, double timeout)
 751 {
 752   smx_synchro_t res = SIMIX_process_sleep(issuer, timeout);
 753   res->type = SIMIX_SYNC_JOIN;
 754   SIMIX_process_on_exit(process, (int_f_pvoid_pvoid_t)SIMIX_process_join_finish, res);
 755   return res;
 756 }
 757
 758 void simcall_HANDLER_process_sleep(smx_simcall_t simcall, double duration)
 759 {
 760   if (MC_is_active() || MC_record_replay_is_active()) {
 761     MC_process_clock_add(simcall->issuer, duration);
 762     simcall_process_sleep__set__result(simcall, SIMIX_DONE);
 763     SIMIX_simcall_answer(simcall);
 764     return;
 765   }
 766   smx_synchro_t sync = SIMIX_process_sleep(simcall->issuer, duration);
 767   xbt_fifo_push(sync->simcalls, simcall);
 768   simcall->issuer->waiting_synchro = sync;
 769 }
 770
 771 smx_synchro_t SIMIX_process_sleep(smx_process_t process, double duration)
 772 {
 773   sg_host_t host = process->host;
 774
 775   /* check if the host is active */
 776   if (host->is_off()) {
 777     THROWF(host_error, 0, "Host %s failed, you cannot call this function",
 778            sg_host_get_name(host));
 779   }
 780
 781   smx_synchro_t synchro = (smx_synchro_t) xbt_mallocator_get(simix_global->synchro_mallocator);
 782   synchro->type = SIMIX_SYNC_SLEEP;
 783   synchro->name = NULL;
 784   synchro->category = NULL;
 785
 786   synchro->sleep.host = host;
 787   synchro->sleep.surf_sleep = surf_host_sleep(host, duration);
 788
 789   synchro->sleep.surf_sleep->setData(synchro);
 790   XBT_DEBUG("Create sleep synchronization %p", synchro);
 791
 792   return synchro;
 793 }
 794
 795 void SIMIX_post_process_sleep(smx_synchro_t synchro)
 796 {
 797   smx_simcall_t simcall;
 798   e_smx_state_t state;
 799   xbt_assert(synchro->type == SIMIX_SYNC_SLEEP || synchro->type == SIMIX_SYNC_JOIN);
 800
 801   while ((simcall = (smx_simcall_t) xbt_fifo_shift(synchro->simcalls))) {
 802
 803     switch (synchro->sleep.surf_sleep->getState()){
 804       case SURF_ACTION_FAILED:
 805         simcall->issuer->context->iwannadie = 1;
 806         //SMX_EXCEPTION(simcall->issuer, host_error, 0, "Host failed");
 807         state = SIMIX_SRC_HOST_FAILURE;
 808         break;
 809
 810       case SURF_ACTION_DONE:
 811         state = SIMIX_DONE;
 812         break;
 813
 814       default:
 815         THROW_IMPOSSIBLE;
 816         break;
 817     }
 818     if (simcall->issuer->host->is_off()) {
 819       simcall->issuer->context->iwannadie = 1;
 820     }
 821     simcall_process_sleep__set__result(simcall, state);
 822     simcall->issuer->waiting_synchro = NULL;
 823     if (simcall->issuer->suspended) {
 824       XBT_DEBUG("Wait! This process is suspended and can't wake up now.");
 825       simcall->issuer->suspended = 0;
 826       simcall_HANDLER_process_suspend(simcall, simcall->issuer);
 827     } else {
 828       SIMIX_simcall_answer(simcall);
 829     }
 830   }
 831
 832   SIMIX_process_sleep_destroy(synchro);
 833 }
 834
 835 void SIMIX_process_sleep_destroy(smx_synchro_t synchro)
 836 {
 837   XBT_DEBUG("Destroy synchro %p", synchro);
 838   xbt_assert(synchro->type == SIMIX_SYNC_SLEEP || synchro->type == SIMIX_SYNC_JOIN);
 839
 840   if (synchro->sleep.surf_sleep) {
 841     synchro->sleep.surf_sleep->unref();
 842     synchro->sleep.surf_sleep = NULL;
 843   }
 844   if (synchro->type == SIMIX_SYNC_SLEEP)
 845     xbt_mallocator_release(simix_global->synchro_mallocator, synchro);
 846 }
 847
 848 void SIMIX_process_sleep_suspend(smx_synchro_t synchro)
 849 {
 850   xbt_assert(synchro->type == SIMIX_SYNC_SLEEP);
 851   synchro->sleep.surf_sleep->suspend();
 852 }
 853
 854 void SIMIX_process_sleep_resume(smx_synchro_t synchro)
 855 {
 856   XBT_DEBUG("Synchro state is %d on process_sleep_resume.", synchro->state);
 857   xbt_assert(synchro->type == SIMIX_SYNC_SLEEP);
 858   synchro->sleep.surf_sleep->resume();
 859 }
 860
 861 /**
 862  * \brief Calling this function makes the process to yield.
 863  *
 864  * Only the current process can call this function, giving back the control to
 865  * maestro.
 866  *
 867  * \param self the current process
 868  */
 869 void SIMIX_process_yield(smx_process_t self)
 870 {
 871   XBT_DEBUG("Yield process '%s'", self->name);
 872
 873   /* Go into sleep and return control to maestro */
 874   SIMIX_context_suspend(self->context);
 875
 876   /* Ok, maestro returned control to us */
 877   XBT_DEBUG("Control returned to me: '%s'", self->name);
 878
 879   if (self->new_host) {
 880     SIMIX_process_change_host(self, self->new_host);
 881     self->new_host = NULL;
 882   }
 883
 884   if (self->context->iwannadie){
 885     XBT_DEBUG("I wanna die!");
 886     SIMIX_process_stop(self);
 887   }
 888
 889   if (self->suspended) {
 890     XBT_DEBUG("Hey! I'm suspended.");
 891     xbt_assert(!self->doexception, "Gasp! This exception may be lost by subsequent calls.");
 892     self->suspended = 0;
 893     SIMIX_process_suspend(self, self);
 894   }
 895
 896   if (self->doexception) {
 897     XBT_DEBUG("Wait, maestro left me an exception");
 898     self->doexception = 0;
 899     SMX_THROW();
 900   }
 901
 902   if(SMPI_switch_data_segment && self->segment_index != -1){
 903     SMPI_switch_data_segment(self->segment_index);
 904   }
 905 }
 906
 907 /* callback: context fetching */
 908 xbt_running_ctx_t *SIMIX_process_get_running_context(void)
 909 {
 910   return SIMIX_process_self()->running_ctx;
 911 }
 912
 913 /* callback: termination */
 914 void SIMIX_process_exception_terminate(xbt_ex_t * e)
 915 {
 916   xbt_ex_display(e);
 917   xbt_abort();
 918 }
 919
 920 smx_context_t SIMIX_process_get_context(smx_process_t p) {
 921   return p->context;
 922 }
 923
 924 void SIMIX_process_set_context(smx_process_t p,smx_context_t c) {
 925   p->context = c;
 926 }
 927
 928 /**
 929  * \brief Returns the list of processes to run.
 930  */
 931 xbt_dynar_t SIMIX_process_get_runnable(void)
 932 {
 933   return simix_global->process_to_run;
 934 }
 935
 936 /**
 937  * \brief Returns the process from PID.
 938  */
 939 smx_process_t SIMIX_process_from_PID(int PID)
 940 {
 941   smx_process_t proc;
 942   xbt_swag_foreach(proc, simix_global->process_list) {
 943    if (proc->pid == (unsigned long) PID)
 944     return proc;
 945   }
 946   return NULL;
 947 }
 948
 949 /** @brief returns a dynar containg all currently existing processes */
 950 xbt_dynar_t SIMIX_processes_as_dynar(void) {
 951   smx_process_t proc;
 952   xbt_dynar_t res = xbt_dynar_new(sizeof(smx_process_t),NULL);
 953   xbt_swag_foreach(proc, simix_global->process_list) {
 954     xbt_dynar_push(res,&proc);
 955   }
 956   return res;
 957 }
 958
 959
 960 void SIMIX_process_on_exit_runall(smx_process_t process) {
 961   s_smx_process_exit_fun_t exit_fun;
 962   smx_process_exit_status_t exit_status = (process->context->iwannadie) ?
 963                                          SMX_EXIT_FAILURE : SMX_EXIT_SUCCESS;
 964   while (!xbt_dynar_is_empty(process->on_exit)) {
 965     exit_fun = xbt_dynar_pop_as(process->on_exit,s_smx_process_exit_fun_t);
 966     (exit_fun.fun)((void*)exit_status, exit_fun.arg);
 967   }
 968 }
 969
 970 void SIMIX_process_on_exit(smx_process_t process, int_f_pvoid_pvoid_t fun, void *data) {
 971   xbt_assert(process, "current process not found: are you in maestro context ?");
 972
 973   if (!process->on_exit) {
 974     process->on_exit = xbt_dynar_new(sizeof(s_smx_process_exit_fun_t), NULL);
 975   }
 976
 977   s_smx_process_exit_fun_t exit_fun = {fun, data};
 978
 979   xbt_dynar_push_as(process->on_exit,s_smx_process_exit_fun_t,exit_fun);
 980 }
 981
 982 /**
 983  * \brief Sets the auto-restart status of the process.
 984  * If set to 1, the process will be automatically restarted when its host
 985  * comes back.
 986  */
 987 void SIMIX_process_auto_restart_set(smx_process_t process, int auto_restart) {
 988   process->auto_restart = auto_restart;
 989 }
 990
 991 smx_process_t simcall_HANDLER_process_restart(smx_simcall_t simcall, smx_process_t process) {
 992   return SIMIX_process_restart(process, simcall->issuer);
 993 }
 994 /** @brief Restart a process, starting it again from the beginning. */
 995 smx_process_t SIMIX_process_restart(smx_process_t process, smx_process_t issuer) {
 996   XBT_DEBUG("Restarting process %s on %s", process->name, sg_host_get_name(process->host));
 997   //retrieve the arguments of the old process
 998   //FIXME: Factorize this with SIMIX_host_add_auto_restart_process ?
 999   s_smx_process_arg_t arg;
1000   arg.code = process->code;
1001   arg.hostname = sg_host_get_name(process->host);
1002   arg.kill_time = SIMIX_timer_get_date(process->kill_timer);
1003   arg.argc = process->argc;
1004   arg.data = process->data;
1005   int i;
1006   arg.argv = xbt_new(char*,process->argc + 1);
1007   for (i = 0; i < arg.argc; i++) {
1008     arg.argv[i] = xbt_strdup(process->argv[i]);
1009   }
1010   arg.argv[process->argc] = NULL;
1011   arg.properties = NULL;
1012   arg.auto_restart = process->auto_restart;
1013   //kill the old process
1014   SIMIX_process_kill(process,issuer);
1015   //start the new process
1016   smx_process_t new_process;
1017   if (simix_global->create_process_function) {
1018     new_process = simix_global->create_process_function(
1019                                           arg.argv[0],
1020                                           arg.code,
1021                                           arg.data,
1022                                           arg.hostname,
1023                                           arg.kill_time,
1024                                           arg.argc,
1025                                           arg.argv,
1026                                           arg.properties,
1027                                           arg.auto_restart,
1028                                           NULL);
1029   } else {
1030     new_process = simcall_process_create(
1031                            arg.argv[0],
1032                            arg.code,
1033                            arg.data,
1034                            arg.hostname,
1035                            arg.kill_time,
1036                            arg.argc,
1037                            arg.argv,
1038                            arg.properties,
1039                            arg.auto_restart);
1040
1041   }
1042   return new_process;
1043 }
1044
1045 void SIMIX_segment_index_set(smx_process_t proc, int index){
1046   proc->segment_index = index;
1047 }