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