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   if (process->host)
 107     xbt_swag_remove(process, sg_host_simix(process->host)->process_list);
 108   xbt_swag_insert(process, simix_global->process_to_destroy);
 109   process->context->iwannadie = 0;
 110
 111   xbt_os_mutex_release(simix_global->mutex);
 112 }
 113
 114 /**
 115  * Garbage collection
 116  *
 117  * Should be called some time to time to free the memory allocated for processes
 118  * that have finished (or killed).
 119  */
 120 void SIMIX_process_empty_trash(void)
 121 {
 122   smx_process_t process = NULL;
 123
 124   while ((process = (smx_process_t) xbt_swag_extract(simix_global->process_to_destroy))) {
 125     XBT_DEBUG("Getting rid of %p",process);
 126
 127     SIMIX_context_free(process->context);
 128
 129     /* Free the exception allocated at creation time */
 130     free(process->running_ctx);
 131     xbt_dict_free(&process->properties);
 132
 133     xbt_fifo_free(process->comms);
 134
 135     xbt_dynar_free(&process->on_exit);
 136
 137     xbt_free(process->name);
 138     xbt_free(process);
 139   }
 140 }
 141
 142 /**
 143  * \brief Creates and runs the maestro process
 144  */
 145 void SIMIX_maestro_create(void (*code)(void*), void* data)
 146 {
 147   smx_process_t maestro = NULL;
 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->data = data;
 154   maestro->running_ctx = (xbt_running_ctx_t*) xbt_malloc0(sizeof(xbt_running_ctx_t));
 155   XBT_RUNNING_CTX_INITIALIZE(maestro->running_ctx);
 156
 157   if (!code) {
 158     maestro->context = SIMIX_context_new(NULL, 0, nullptr, NULL, maestro);
 159   } else {
 160     if (!simix_global)
 161       xbt_die("simix is not initialized, please call MSG_init first");
 162     maestro->context =
 163       simix_global->context_factory->create_maestro(
 164         std::bind(code, data), maestro);
 165   }
 166
 167   maestro->simcall.issuer = maestro;
 168   simix_global->maestro_process = maestro;
 169 }
 170
 171 /**
 172  * \brief Stops a process.
 173  *
 174  * Stops the process, execute all the registered on_exit functions,
 175  * register it to the list of the process to restart if needed
 176  * and stops its context.
 177  */
 178 void SIMIX_process_stop(smx_process_t arg) {
 179   /* execute the on_exit functions */
 180   SIMIX_process_on_exit_runall(arg);
 181   /* Add the process to the list of process to restart, only if the host is down */
 182   if (arg->auto_restart && arg->host->is_off()) {
 183     SIMIX_host_add_auto_restart_process(arg->host,arg->name,arg->code, arg->data,
 184                                         sg_host_get_name(arg->host),
 185                                         SIMIX_timer_get_date(arg->kill_timer),
 186                                         arg->argc,arg->argv,arg->properties,
 187                                         arg->auto_restart);
 188   }
 189   XBT_DEBUG("Process %s (%s) is dead",arg->name,sg_host_get_name(arg->host));
 190   /* stop the context */
 191   SIMIX_context_stop(arg->context);
 192 }
 193
 194 /**
 195  * \brief Same as SIMIX_process_create() but with only one argument (used by timers).
 196  * This function frees the argument.
 197  * \return the process created
 198  */
 199 smx_process_t SIMIX_process_create_from_wrapper(smx_process_arg_t args) {
 200
 201   smx_process_t process = simix_global->create_process_function(
 202                                         args->name,
 203                                         args->code,
 204                                         args->data,
 205                                         args->hostname,
 206                                         args->kill_time,
 207                                         args->argc,
 208                                         args->argv,
 209                                         args->properties,
 210                                         args->auto_restart,
 211                                         NULL);
 212   xbt_free(args);
 213   return process;
 214 }
 215
 216
 217 void* simcall_HANDLER_process_create(smx_simcall_t simcall,
 218                           const char *name,
 219                           xbt_main_func_t code,
 220                           void *data,
 221                           const char *hostname,
 222                           double kill_time,
 223                           int argc, char **argv,
 224                           xbt_dict_t properties,
 225                           int auto_restart){
 226   return (void*)SIMIX_process_create(name, code, data, hostname,
 227                        kill_time, argc, argv, properties, auto_restart,
 228                        simcall->issuer);
 229 }
 230
 231 static void kill_process(void* process)
 232 {
 233   simix_global->kill_process_function((smx_process_t) process);
 234 }
 235
 236 /**
 237  * \brief Internal function to create a process.
 238  *
 239  * This function actually creates the process.
 240  * It may be called when a SIMCALL_PROCESS_CREATE simcall occurs,
 241  * or directly for SIMIX internal purposes. The sure thing is that it's called from maestro context.
 242  *
 243  * \return the process created
 244  */
 245 smx_process_t SIMIX_process_create(
 246                           const char *name,
 247                           xbt_main_func_t code,
 248                           void *data,
 249                           const char *hostname,
 250                           double kill_time,
 251                           int argc, char **argv,
 252                           xbt_dict_t properties,
 253                           int auto_restart,
 254                           smx_process_t parent_process)
 255 {
 256   smx_process_t process = NULL;
 257   sg_host_t host = sg_host_by_name(hostname);
 258
 259   XBT_DEBUG("Start process %s on host '%s'", name, hostname);
 260
 261   if (host->is_off()) {
 262     int i;
 263     XBT_WARN("Cannot launch process '%s' on failed host '%s'", name,
 264           hostname);
 265     for (i = 0; i < argc; i++)
 266       xbt_free(argv[i]);
 267     xbt_free(argv);
 268   }
 269   else {
 270     process = xbt_new0(s_smx_process_t, 1);
 271
 272     xbt_assert(((code != NULL) && (host != NULL)), "Invalid parameters");
 273     /* Process data */
 274     process->pid = simix_process_maxpid++;
 275     process->name = xbt_strdup(name);
 276     process->host = host;
 277     process->data = data;
 278     process->comms = xbt_fifo_new();
 279     process->simcall.issuer = process;
 280     /* Initiliaze data segment to default value */
 281     SIMIX_segment_index_set(process, -1);
 282
 283      if (parent_process != NULL) {
 284        process->ppid = SIMIX_process_get_PID(parent_process);
 285        /* SMPI process have their own data segment and
 286           each other inherit from their father */
 287 #ifdef HAVE_SMPI
 288        if(smpi_privatize_global_variables){
 289          if( parent_process->pid != 0){
 290            SIMIX_segment_index_set(process, parent_process->segment_index);
 291          } else {
 292            SIMIX_segment_index_set(process, process->pid - 1);
 293          }
 294        }
 295 #endif
 296      } else {
 297        process->ppid = -1;
 298      }
 299
 300     /* Process data for auto-restart */
 301     process->auto_restart = auto_restart;
 302     process->code = code;
 303     process->argc = argc;
 304     process->argv = argv;
 305
 306
 307     XBT_VERB("Create context %s", process->name);
 308     process->context = SIMIX_context_new(code, argc, argv, simix_global->cleanup_process_function, process);
 309
 310     process->running_ctx = (xbt_running_ctx_t*) xbt_malloc0(sizeof(xbt_running_ctx_t));
 311     XBT_RUNNING_CTX_INITIALIZE(process->running_ctx);
 312
 313     if(MC_is_active()){
 314       MC_ignore_heap(process->running_ctx, sizeof(*process->running_ctx));
 315     }
 316
 317     /* Add properties */
 318     process->properties = properties;
 319
 320     /* Add the process to it's host process list */
 321     xbt_swag_insert(process, sg_host_simix(host)->process_list);
 322
 323     XBT_DEBUG("Start context '%s'", process->name);
 324
 325     /* Now insert it in the global process list and in the process to run list */
 326     xbt_swag_insert(process, simix_global->process_list);
 327     XBT_DEBUG("Inserting %s(%s) in the to_run list", process->name, sg_host_get_name(host));
 328     xbt_dynar_push_as(simix_global->process_to_run, smx_process_t, process);
 329
 330     if (kill_time > SIMIX_get_clock() && simix_global->kill_process_function) {
 331       XBT_DEBUG("Process %s(%s) will be kill at time %f", process->name,
 332           sg_host_get_name(process->host), kill_time);
 333       process->kill_timer = SIMIX_timer_set(kill_time, kill_process, process);
 334     }
 335
 336     /* Tracing the process creation */
 337     TRACE_msg_process_create(process->name, process->pid, process->host);
 338   }
 339   return process;
 340 }
 341
 342 smx_process_t SIMIX_process_attach(
 343   const char* name,
 344   void *data,
 345   const char* hostname,
 346   xbt_dict_t properties,
 347   smx_process_t parent_process)
 348 {
 349   // This is mostly a copy/paste from SIMIX_process_new(),
 350   // it'd be nice to share some code between those two functions.
 351
 352   sg_host_t host = sg_host_by_name(hostname);
 353   XBT_DEBUG("Attach process %s on host '%s'", name, hostname);
 354
 355   if (host->is_off()) {
 356     XBT_WARN("Cannot launch process '%s' on failed host '%s'",
 357       name, hostname);
 358     return nullptr;
 359   }
 360
 361   smx_process_t process = xbt_new0(s_smx_process_t, 1);
 362   /* Process data */
 363   process->pid = simix_process_maxpid++;
 364   process->name = xbt_strdup(name);
 365   process->host = host;
 366   process->data = data;
 367   process->comms = xbt_fifo_new();
 368   process->simcall.issuer = process;
 369   process->ppid = -1;
 370   /* Initiliaze data segment to default value */
 371   SIMIX_segment_index_set(process, -1);
 372   if (parent_process != NULL) {
 373     process->ppid = SIMIX_process_get_PID(parent_process);
 374    /* SMPI process have their own data segment and
 375       each other inherit from their father */
 376   #ifdef HAVE_SMPI
 377     if(smpi_privatize_global_variables){
 378       if(parent_process->pid != 0){
 379         SIMIX_segment_index_set(process, parent_process->segment_index);
 380       } else {
 381         SIMIX_segment_index_set(process, process->pid - 1);
 382       }
 383     }
 384   #endif
 385   }
 386
 387   /* Process data for auto-restart */
 388   process->auto_restart = false;
 389   process->code = nullptr;
 390   process->argc = 0;
 391   process->argv = nullptr;
 392
 393   XBT_VERB("Create context %s", process->name);
 394   if (!simix_global)
 395     xbt_die("simix is not initialized, please call MSG_init first");
 396   process->context = simix_global->context_factory->attach(
 397     simix_global->cleanup_process_function, process);
 398
 399   process->running_ctx = (xbt_running_ctx_t*) xbt_malloc0(sizeof(xbt_running_ctx_t));
 400   XBT_RUNNING_CTX_INITIALIZE(process->running_ctx);
 401
 402   if(MC_is_active()){
 403     MC_ignore_heap(process->running_ctx, sizeof(*process->running_ctx));
 404   }
 405
 406   /* Add properties */
 407   process->properties = properties;
 408
 409   /* Add the process to it's host process list */
 410   xbt_swag_insert(process, sg_host_simix(host)->process_list);
 411
 412   /* Now insert it in the global process list and in the process to run list */
 413   xbt_swag_insert(process, simix_global->process_list);
 414   XBT_DEBUG("Inserting %s(%s) in the to_run list", process->name, sg_host_get_name(host));
 415   xbt_dynar_push_as(simix_global->process_to_run, smx_process_t, process);
 416
 417   /* Tracing the process creation */
 418   TRACE_msg_process_create(process->name, process->pid, process->host);
 419
 420   auto context = dynamic_cast<simgrid::simix::AttachContext*>(process->context);
 421   if (!context)
 422     xbt_die("Not a suitable context");
 423
 424   context->attach_start();
 425   return process;
 426 }
 427
 428 void SIMIX_process_detach(void)
 429 {
 430   auto context = dynamic_cast<simgrid::simix::AttachContext*>(SIMIX_context_self());
 431   if (!context)
 432     xbt_die("Not a suitable context");
 433
 434   simix_global->cleanup_process_function(context->process());
 435
 436   // Let maestro ignore we are still alive:
 437   // xbt_swag_remove(context->process(), simix_global->process_list);
 438
 439   // TODDO, Remove from proces list:
 440   //   xbt_swag_remove(process, sg_host_simix(host)->process_list);
 441
 442   context->attach_stop();
 443   // delete context;
 444 }
 445
 446 /**
 447  * \brief Executes the processes from simix_global->process_to_run.
 448  *
 449  * The processes of simix_global->process_to_run are run (in parallel if
 450  * possible).  On exit, simix_global->process_to_run is empty, and
 451  * simix_global->process_that_ran contains the list of processes that just ran.
 452  * The two lists are swapped so, be careful when using them before and after a
 453  * call to this function.
 454  */
 455 void SIMIX_process_runall(void)
 456 {
 457   SIMIX_context_runall();
 458
 459   xbt_dynar_t tmp = simix_global->process_that_ran;
 460   simix_global->process_that_ran = simix_global->process_to_run;
 461   simix_global->process_to_run = tmp;
 462   xbt_dynar_reset(simix_global->process_to_run);
 463 }
 464
 465 void simcall_HANDLER_process_kill(smx_simcall_t simcall, smx_process_t process) {
 466   SIMIX_process_kill(process, simcall->issuer);
 467 }
 468 /**
 469  * \brief Internal function to kill a SIMIX process.
 470  *
 471  * This function may be called when a SIMCALL_PROCESS_KILL simcall occurs,
 472  * or directly for SIMIX internal purposes.
 473  *
 474  * \param process poor victim
 475  * \param issuer the process which has sent the PROCESS_KILL. Important to not schedule twice the same process.
 476  */
 477 void SIMIX_process_kill(smx_process_t process, smx_process_t issuer) {
 478
 479   XBT_DEBUG("Killing process %s on %s", process->name, sg_host_get_name(process->host));
 480
 481   process->context->iwannadie = 1;
 482   process->blocked = 0;
 483   process->suspended = 0;
 484   process->doexception = 0;
 485
 486   /* destroy the blocking synchro if any */
 487   if (process->waiting_synchro) {
 488
 489     switch (process->waiting_synchro->type) {
 490
 491     case SIMIX_SYNC_EXECUTE:
 492     case SIMIX_SYNC_PARALLEL_EXECUTE:
 493       SIMIX_execution_destroy(process->waiting_synchro);
 494       break;
 495
 496     case SIMIX_SYNC_COMMUNICATE:
 497       xbt_fifo_remove(process->comms, process->waiting_synchro);
 498       SIMIX_comm_cancel(process->waiting_synchro);
 499       xbt_fifo_remove(process->waiting_synchro->simcalls, &process->simcall);
 500       SIMIX_comm_destroy(process->waiting_synchro);
 501       break;
 502
 503     case SIMIX_SYNC_SLEEP:
 504       SIMIX_process_sleep_destroy(process->waiting_synchro);
 505       break;
 506
 507     case SIMIX_SYNC_JOIN:
 508       SIMIX_process_sleep_destroy(process->waiting_synchro);
 509       break;
 510
 511     case SIMIX_SYNC_SYNCHRO:
 512       SIMIX_synchro_stop_waiting(process, &process->simcall);
 513       SIMIX_synchro_destroy(process->waiting_synchro);
 514       break;
 515
 516     case SIMIX_SYNC_IO:
 517       SIMIX_io_destroy(process->waiting_synchro);
 518       break;
 519
 520     }
 521
 522     process->waiting_synchro = NULL;
 523   }
 524   if(!xbt_dynar_member(simix_global->process_to_run, &(process)) && process != issuer) {
 525     XBT_DEBUG("Inserting %s in the to_run list", process->name);
 526     xbt_dynar_push_as(simix_global->process_to_run, smx_process_t, process);
 527   }
 528
 529 }
 530
 531 /** @brief Ask another process to raise the given exception
 532  *
 533  * @param cat category of exception
 534  * @param value value associated to the exception
 535  * @param msg string information associated to the exception
 536  */
 537 void SIMIX_process_throw(smx_process_t process, xbt_errcat_t cat, int value, const char *msg) {
 538   SMX_EXCEPTION(process, cat, value, msg);
 539
 540   if (process->suspended)
 541     SIMIX_process_resume(process,SIMIX_process_self());
 542
 543   /* cancel the blocking synchro if any */
 544   if (process->waiting_synchro) {
 545
 546     switch (process->waiting_synchro->type) {
 547
 548     case SIMIX_SYNC_EXECUTE:
 549     case SIMIX_SYNC_PARALLEL_EXECUTE:
 550       SIMIX_execution_cancel(process->waiting_synchro);
 551       break;
 552
 553     case SIMIX_SYNC_COMMUNICATE:
 554       xbt_fifo_remove(process->comms, process->waiting_synchro);
 555       SIMIX_comm_cancel(process->waiting_synchro);
 556       break;
 557
 558     case SIMIX_SYNC_SLEEP:
 559     case SIMIX_SYNC_JOIN:
 560       SIMIX_process_sleep_destroy(process->waiting_synchro);
 561       if (!xbt_dynar_member(simix_global->process_to_run, &(process)) && process != SIMIX_process_self()) {
 562         XBT_DEBUG("Inserting %s in the to_run list", process->name);
 563         xbt_dynar_push_as(simix_global->process_to_run, smx_process_t, process);
 564       }
 565       break;
 566
 567     case SIMIX_SYNC_SYNCHRO:
 568       SIMIX_synchro_stop_waiting(process, &process->simcall);
 569       break;
 570
 571     case SIMIX_SYNC_IO:
 572       SIMIX_io_destroy(process->waiting_synchro);
 573       break;
 574
 575     }
 576   }
 577   process->waiting_synchro = NULL;
 578
 579 }
 580
 581 void simcall_HANDLER_process_killall(smx_simcall_t simcall, int reset_pid) {
 582   SIMIX_process_killall(simcall->issuer, reset_pid);
 583 }
 584 /**
 585  * \brief Kills all running processes.
 586  * \param issuer this one will not be killed
 587  */
 588 void SIMIX_process_killall(smx_process_t issuer, int reset_pid)
 589 {
 590   smx_process_t p = NULL;
 591
 592   while ((p = (smx_process_t) xbt_swag_extract(simix_global->process_list))) {
 593     if (p != issuer) {
 594       SIMIX_process_kill(p,issuer);
 595     }
 596   }
 597
 598   if (reset_pid > 0)
 599     simix_process_maxpid = reset_pid;
 600
 601   SIMIX_context_runall();
 602
 603   SIMIX_process_empty_trash();
 604 }
 605
 606 void simcall_HANDLER_process_set_host(smx_simcall_t simcall, smx_process_t process, sg_host_t dest)
 607 {
 608   process->new_host = dest;
 609 }
 610 void SIMIX_process_change_host(smx_process_t process,
 611              sg_host_t dest)
 612 {
 613   xbt_assert((process != NULL), "Invalid parameters");
 614   xbt_swag_remove(process, sg_host_simix(process->host)->process_list);
 615   process->host = dest;
 616   xbt_swag_insert(process, sg_host_simix(dest)->process_list);
 617 }
 618
 619
 620 void simcall_HANDLER_process_suspend(smx_simcall_t simcall, smx_process_t process)
 621 {
 622   smx_synchro_t sync_suspend =
 623       SIMIX_process_suspend(process, simcall->issuer);
 624
 625   if (process != simcall->issuer) {
 626     SIMIX_simcall_answer(simcall);
 627   } else {
 628     xbt_fifo_push(sync_suspend->simcalls, simcall);
 629     process->waiting_synchro = sync_suspend;
 630     SIMIX_execution_suspend(process->waiting_synchro);
 631   }
 632   /* If we are suspending ourselves, then just do not finish the simcall now */
 633 }
 634
 635 smx_synchro_t SIMIX_process_suspend(smx_process_t process, smx_process_t issuer)
 636 {
 637   xbt_assert((process != NULL), "Invalid parameters");
 638
 639   if (process->suspended) {
 640     XBT_DEBUG("Process '%s' is already suspended", process->name);
 641     return NULL;
 642   }
 643
 644   process->suspended = 1;
 645
 646   /* If we are suspending another process, and it is waiting on a sync,
 647      suspend its synchronization. */
 648   if (process != issuer) {
 649
 650     if (process->waiting_synchro) {
 651
 652       switch (process->waiting_synchro->type) {
 653
 654         case SIMIX_SYNC_EXECUTE:
 655         case SIMIX_SYNC_PARALLEL_EXECUTE:
 656           SIMIX_execution_suspend(process->waiting_synchro);
 657           break;
 658
 659         case SIMIX_SYNC_COMMUNICATE:
 660           SIMIX_comm_suspend(process->waiting_synchro);
 661           break;
 662
 663         case SIMIX_SYNC_SLEEP:
 664           SIMIX_process_sleep_suspend(process->waiting_synchro);
 665           break;
 666
 667         case SIMIX_SYNC_SYNCHRO:
 668           /* Suspension is delayed to when the process is rescheduled. */
 669           break;
 670
 671         default:
 672           xbt_die("Internal error in SIMIX_process_suspend: unexpected synchronization type %d",
 673               (int)process->waiting_synchro->type);
 674       }
 675       return NULL;
 676     } else {
 677       /* Suspension is delayed to when the process is rescheduled. */
 678       return NULL;
 679     }
 680   } else {
 681     /* FIXME: computation size is zero. Is it okay that bound is zero ? */
 682     return SIMIX_execution_start(process, "suspend", 0.0, 1.0, 0.0, 0);
 683   }
 684 }
 685
 686 void simcall_HANDLER_process_resume(smx_simcall_t simcall, smx_process_t process){
 687   SIMIX_process_resume(process, simcall->issuer);
 688 }
 689
 690 void SIMIX_process_resume(smx_process_t process, smx_process_t issuer)
 691 {
 692   XBT_IN("process = %p, issuer = %p", process, issuer);
 693
 694   if(process->context->iwannadie) {
 695     XBT_VERB("Ignoring request to suspend a process that is currently dying.");
 696     return;
 697   }
 698
 699   if(!process->suspended) return;
 700   process->suspended = 0;
 701
 702   /* If we are resuming another process, resume the synchronization it was waiting for
 703      if any. Otherwise add it to the list of process to run in the next round. */
 704   if (process != issuer) {
 705
 706     if (process->waiting_synchro) {
 707
 708       switch (process->waiting_synchro->type) {
 709
 710         case SIMIX_SYNC_EXECUTE:
 711         case SIMIX_SYNC_PARALLEL_EXECUTE:
 712           SIMIX_execution_resume(process->waiting_synchro);
 713           break;
 714
 715         case SIMIX_SYNC_COMMUNICATE:
 716           SIMIX_comm_resume(process->waiting_synchro);
 717           break;
 718
 719         case SIMIX_SYNC_SLEEP:
 720           SIMIX_process_sleep_resume(process->waiting_synchro);
 721           break;
 722
 723         case SIMIX_SYNC_SYNCHRO:
 724           /* I cannot resume it now. This is delayed to when the process is rescheduled at
 725            * the end of the synchro. */
 726           break;
 727
 728         default:
 729           xbt_die("Internal error in SIMIX_process_resume: unexpected synchronization type %d",
 730               (int)process->waiting_synchro->type);
 731       }
 732     }
 733   } else XBT_WARN("Strange. Process %p is trying to resume himself.", issuer);
 734
 735   XBT_OUT();
 736 }
 737
 738 int SIMIX_process_get_maxpid(void) {
 739   return simix_process_maxpid;
 740 }
 741
 742 int SIMIX_process_count(void)
 743 {
 744   return xbt_swag_size(simix_global->process_list);
 745 }
 746
 747 int SIMIX_process_get_PID(smx_process_t self){
 748   if (self == NULL)
 749     return 0;
 750   else
 751     return self->pid;
 752 }
 753
 754 int SIMIX_process_get_PPID(smx_process_t self){
 755   if (self == NULL)
 756     return 0;
 757   else
 758     return self->ppid;
 759 }
 760
 761 void* SIMIX_process_self_get_data(smx_process_t self)
 762 {
 763   xbt_assert(self == SIMIX_process_self(), "This is not the current process");
 764
 765   if (!self) {
 766     return NULL;
 767   }
 768   return SIMIX_process_get_data(self);
 769 }
 770
 771 void SIMIX_process_self_set_data(smx_process_t self, void *data)
 772 {
 773   xbt_assert(self == SIMIX_process_self(), "This is not the current process");
 774
 775   SIMIX_process_set_data(self, data);
 776 }
 777
 778 void* SIMIX_process_get_data(smx_process_t process)
 779 {
 780   return process->data;
 781 }
 782
 783 void SIMIX_process_set_data(smx_process_t process, void *data)
 784 {
 785   process->data = data;
 786 }
 787
 788 sg_host_t SIMIX_process_get_host(smx_process_t process)
 789 {
 790   return process->host;
 791 }
 792
 793 xbt_main_func_t SIMIX_process_get_code(void){
 794   return SIMIX_process_self()->code;
 795 }
 796
 797 /* needs to be public and without simcall because it is called
 798    by exceptions and logging events */
 799 const char* SIMIX_process_self_get_name(void) {
 800
 801   smx_process_t process = SIMIX_process_self();
 802   if (process == NULL || process == simix_global->maestro_process)
 803     return "maestro";
 804
 805   return SIMIX_process_get_name(process);
 806 }
 807
 808 const char* SIMIX_process_get_name(smx_process_t process)
 809 {
 810   return process->name;
 811 }
 812
 813 smx_process_t SIMIX_process_get_by_name(const char* name)
 814 {
 815   smx_process_t proc;
 816
 817   xbt_swag_foreach(proc, simix_global->process_list)
 818   {
 819     if(!strcmp(name, proc->name))
 820       return proc;
 821   }
 822   return NULL;
 823 }
 824
 825 int SIMIX_process_is_suspended(smx_process_t process)
 826 {
 827   return process->suspended;
 828 }
 829
 830 xbt_dict_t SIMIX_process_get_properties(smx_process_t process)
 831 {
 832   return process->properties;
 833 }
 834
 835 void simcall_HANDLER_process_join(smx_simcall_t simcall, smx_process_t process, double timeout)
 836 {
 837   smx_synchro_t sync = SIMIX_process_join(simcall->issuer, process, timeout);
 838   xbt_fifo_push(sync->simcalls, simcall);
 839   simcall->issuer->waiting_synchro = sync;
 840 }
 841
 842 static int SIMIX_process_join_finish(smx_process_exit_status_t status, smx_synchro_t sync){
 843   if (sync->sleep.surf_sleep) {
 844     sync->sleep.surf_sleep->cancel();
 845
 846     smx_simcall_t simcall;
 847     while ((simcall = (smx_simcall_t) xbt_fifo_shift(sync->simcalls))) {
 848       simcall_process_sleep__set__result(simcall, SIMIX_DONE);
 849       simcall->issuer->waiting_synchro = NULL;
 850       if (simcall->issuer->suspended) {
 851         XBT_DEBUG("Wait! This process is suspended and can't wake up now.");
 852         simcall->issuer->suspended = 0;
 853         simcall_HANDLER_process_suspend(simcall, simcall->issuer);
 854       } else {
 855         SIMIX_simcall_answer(simcall);
 856       }
 857     }
 858     sync->sleep.surf_sleep->unref();
 859     sync->sleep.surf_sleep = NULL;
 860   }
 861   xbt_mallocator_release(simix_global->synchro_mallocator, sync);
 862   return 0;
 863 }
 864
 865 smx_synchro_t SIMIX_process_join(smx_process_t issuer, smx_process_t process, double timeout)
 866 {
 867   smx_synchro_t res = SIMIX_process_sleep(issuer, timeout);
 868   res->type = SIMIX_SYNC_JOIN;
 869   SIMIX_process_on_exit(process, (int_f_pvoid_pvoid_t)SIMIX_process_join_finish, res);
 870   return res;
 871 }
 872
 873 void simcall_HANDLER_process_sleep(smx_simcall_t simcall, double duration)
 874 {
 875   if (MC_is_active() || MC_record_replay_is_active()) {
 876     MC_process_clock_add(simcall->issuer, duration);
 877     simcall_process_sleep__set__result(simcall, SIMIX_DONE);
 878     SIMIX_simcall_answer(simcall);
 879     return;
 880   }
 881   smx_synchro_t sync = SIMIX_process_sleep(simcall->issuer, duration);
 882   xbt_fifo_push(sync->simcalls, simcall);
 883   simcall->issuer->waiting_synchro = sync;
 884 }
 885
 886 smx_synchro_t SIMIX_process_sleep(smx_process_t process, double duration)
 887 {
 888   sg_host_t host = process->host;
 889
 890   /* check if the host is active */
 891   if (host->is_off()) {
 892     THROWF(host_error, 0, "Host %s failed, you cannot call this function",
 893            sg_host_get_name(host));
 894   }
 895
 896   smx_synchro_t synchro = (smx_synchro_t) xbt_mallocator_get(simix_global->synchro_mallocator);
 897   synchro->type = SIMIX_SYNC_SLEEP;
 898   synchro->name = NULL;
 899   synchro->category = NULL;
 900
 901   synchro->sleep.host = host;
 902   synchro->sleep.surf_sleep = surf_host_sleep(host, duration);
 903
 904   synchro->sleep.surf_sleep->setData(synchro);
 905   XBT_DEBUG("Create sleep synchronization %p", synchro);
 906
 907   return synchro;
 908 }
 909
 910 void SIMIX_post_process_sleep(smx_synchro_t synchro)
 911 {
 912   smx_simcall_t simcall;
 913   e_smx_state_t state;
 914   xbt_assert(synchro->type == SIMIX_SYNC_SLEEP || synchro->type == SIMIX_SYNC_JOIN);
 915
 916   while ((simcall = (smx_simcall_t) xbt_fifo_shift(synchro->simcalls))) {
 917
 918     switch (synchro->sleep.surf_sleep->getState()){
 919       case SURF_ACTION_FAILED:
 920         simcall->issuer->context->iwannadie = 1;
 921         //SMX_EXCEPTION(simcall->issuer, host_error, 0, "Host failed");
 922         state = SIMIX_SRC_HOST_FAILURE;
 923         break;
 924
 925       case SURF_ACTION_DONE:
 926         state = SIMIX_DONE;
 927         break;
 928
 929       default:
 930         THROW_IMPOSSIBLE;
 931         break;
 932     }
 933     if (simcall->issuer->host->is_off()) {
 934       simcall->issuer->context->iwannadie = 1;
 935     }
 936     simcall_process_sleep__set__result(simcall, state);
 937     simcall->issuer->waiting_synchro = NULL;
 938     if (simcall->issuer->suspended) {
 939       XBT_DEBUG("Wait! This process is suspended and can't wake up now.");
 940       simcall->issuer->suspended = 0;
 941       simcall_HANDLER_process_suspend(simcall, simcall->issuer);
 942     } else {
 943       SIMIX_simcall_answer(simcall);
 944     }
 945   }
 946
 947   SIMIX_process_sleep_destroy(synchro);
 948 }
 949
 950 void SIMIX_process_sleep_destroy(smx_synchro_t synchro)
 951 {
 952   XBT_DEBUG("Destroy synchro %p", synchro);
 953   xbt_assert(synchro->type == SIMIX_SYNC_SLEEP || synchro->type == SIMIX_SYNC_JOIN);
 954
 955   if (synchro->sleep.surf_sleep) {
 956     synchro->sleep.surf_sleep->unref();
 957     synchro->sleep.surf_sleep = NULL;
 958   }
 959   if (synchro->type == SIMIX_SYNC_SLEEP)
 960     xbt_mallocator_release(simix_global->synchro_mallocator, synchro);
 961 }
 962
 963 void SIMIX_process_sleep_suspend(smx_synchro_t synchro)
 964 {
 965   xbt_assert(synchro->type == SIMIX_SYNC_SLEEP);
 966   synchro->sleep.surf_sleep->suspend();
 967 }
 968
 969 void SIMIX_process_sleep_resume(smx_synchro_t synchro)
 970 {
 971   XBT_DEBUG("Synchro state is %d on process_sleep_resume.", synchro->state);
 972   xbt_assert(synchro->type == SIMIX_SYNC_SLEEP);
 973   synchro->sleep.surf_sleep->resume();
 974 }
 975
 976 /**
 977  * \brief Calling this function makes the process to yield.
 978  *
 979  * Only the current process can call this function, giving back the control to
 980  * maestro.
 981  *
 982  * \param self the current process
 983  */
 984 void SIMIX_process_yield(smx_process_t self)
 985 {
 986   XBT_DEBUG("Yield process '%s'", self->name);
 987
 988   /* Go into sleep and return control to maestro */
 989   SIMIX_context_suspend(self->context);
 990
 991   /* Ok, maestro returned control to us */
 992   XBT_DEBUG("Control returned to me: '%s'", self->name);
 993
 994   if (self->new_host) {
 995     SIMIX_process_change_host(self, self->new_host);
 996     self->new_host = NULL;
 997   }
 998
 999   if (self->context->iwannadie){
1000     XBT_DEBUG("I wanna die!");
1001     SIMIX_process_stop(self);
1002   }
1003
1004   if (self->suspended) {
1005     XBT_DEBUG("Hey! I'm suspended.");
1006     xbt_assert(!self->doexception, "Gasp! This exception may be lost by subsequent calls.");
1007     self->suspended = 0;
1008     SIMIX_process_suspend(self, self);
1009   }
1010
1011   if (self->doexception) {
1012     XBT_DEBUG("Wait, maestro left me an exception");
1013     self->doexception = 0;
1014     SMX_THROW();
1015   }
1016
1017   if(SMPI_switch_data_segment && self->segment_index != -1){
1018     SMPI_switch_data_segment(self->segment_index);
1019   }
1020 }
1021
1022 /* callback: context fetching */
1023 xbt_running_ctx_t *SIMIX_process_get_running_context(void)
1024 {
1025   smx_process_t process = SIMIX_process_self();
1026   if (process)
1027     return process->running_ctx;
1028   else
1029     return nullptr;
1030 }
1031
1032 /* callback: termination */
1033 void SIMIX_process_exception_terminate(xbt_ex_t * e)
1034 {
1035   xbt_ex_display(e);
1036   xbt_abort();
1037 }
1038
1039 smx_context_t SIMIX_process_get_context(smx_process_t p) {
1040   return p->context;
1041 }
1042
1043 void SIMIX_process_set_context(smx_process_t p,smx_context_t c) {
1044   p->context = c;
1045 }
1046
1047 /**
1048  * \brief Returns the list of processes to run.
1049  */
1050 xbt_dynar_t SIMIX_process_get_runnable(void)
1051 {
1052   return simix_global->process_to_run;
1053 }
1054
1055 /**
1056  * \brief Returns the process from PID.
1057  */
1058 smx_process_t SIMIX_process_from_PID(int PID)
1059 {
1060   smx_process_t proc;
1061   xbt_swag_foreach(proc, simix_global->process_list) {
1062    if (proc->pid == (unsigned long) PID)
1063     return proc;
1064   }
1065   return NULL;
1066 }
1067
1068 /** @brief returns a dynar containg all currently existing processes */
1069 xbt_dynar_t SIMIX_processes_as_dynar(void) {
1070   smx_process_t proc;
1071   xbt_dynar_t res = xbt_dynar_new(sizeof(smx_process_t),NULL);
1072   xbt_swag_foreach(proc, simix_global->process_list) {
1073     xbt_dynar_push(res,&proc);
1074   }
1075   return res;
1076 }
1077
1078
1079 void SIMIX_process_on_exit_runall(smx_process_t process) {
1080   s_smx_process_exit_fun_t exit_fun;
1081   smx_process_exit_status_t exit_status = (process->context->iwannadie) ?
1082                                          SMX_EXIT_FAILURE : SMX_EXIT_SUCCESS;
1083   while (!xbt_dynar_is_empty(process->on_exit)) {
1084     exit_fun = xbt_dynar_pop_as(process->on_exit,s_smx_process_exit_fun_t);
1085     (exit_fun.fun)((void*)exit_status, exit_fun.arg);
1086   }
1087 }
1088
1089 void SIMIX_process_on_exit(smx_process_t process, int_f_pvoid_pvoid_t fun, void *data) {
1090   xbt_assert(process, "current process not found: are you in maestro context ?");
1091
1092   if (!process->on_exit) {
1093     process->on_exit = xbt_dynar_new(sizeof(s_smx_process_exit_fun_t), NULL);
1094   }
1095
1096   s_smx_process_exit_fun_t exit_fun = {fun, data};
1097
1098   xbt_dynar_push_as(process->on_exit,s_smx_process_exit_fun_t,exit_fun);
1099 }
1100
1101 /**
1102  * \brief Sets the auto-restart status of the process.
1103  * If set to 1, the process will be automatically restarted when its host
1104  * comes back.
1105  */
1106 void SIMIX_process_auto_restart_set(smx_process_t process, int auto_restart) {
1107   process->auto_restart = auto_restart;
1108 }
1109
1110 smx_process_t simcall_HANDLER_process_restart(smx_simcall_t simcall, smx_process_t process) {
1111   return SIMIX_process_restart(process, simcall->issuer);
1112 }
1113 /** @brief Restart a process, starting it again from the beginning. */
1114 smx_process_t SIMIX_process_restart(smx_process_t process, smx_process_t issuer) {
1115   XBT_DEBUG("Restarting process %s on %s", process->name, sg_host_get_name(process->host));
1116   //retrieve the arguments of the old process
1117   //FIXME: Factorize this with SIMIX_host_add_auto_restart_process ?
1118   s_smx_process_arg_t arg;
1119   arg.code = process->code;
1120   arg.hostname = sg_host_get_name(process->host);
1121   arg.kill_time = SIMIX_timer_get_date(process->kill_timer);
1122   arg.argc = process->argc;
1123   arg.data = process->data;
1124   int i;
1125   arg.argv = xbt_new(char*,process->argc + 1);
1126   for (i = 0; i < arg.argc; i++) {
1127     arg.argv[i] = xbt_strdup(process->argv[i]);
1128   }
1129   arg.argv[process->argc] = NULL;
1130   arg.properties = NULL;
1131   arg.auto_restart = process->auto_restart;
1132   //kill the old process
1133   SIMIX_process_kill(process,issuer);
1134   //start the new process
1135   smx_process_t new_process;
1136   if (simix_global->create_process_function) {
1137     new_process = simix_global->create_process_function(
1138                                           arg.argv[0],
1139                                           arg.code,
1140                                           arg.data,
1141                                           arg.hostname,
1142                                           arg.kill_time,
1143                                           arg.argc,
1144                                           arg.argv,
1145                                           arg.properties,
1146                                           arg.auto_restart,
1147                                           NULL);
1148   } else {
1149     new_process = simcall_process_create(
1150                            arg.argv[0],
1151                            arg.code,
1152                            arg.data,
1153                            arg.hostname,
1154                            arg.kill_time,
1155                            arg.argc,
1156                            arg.argv,
1157                            arg.properties,
1158                            arg.auto_restart);
1159
1160   }
1161   return new_process;
1162 }
1163
1164 void SIMIX_segment_index_set(smx_process_t proc, int index){
1165   proc->segment_index = index;
1166 }