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