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/Client.hpp"
  15 #include "src/simix/smx_private.hpp"
  16 #include "src/msg/msg_private.h"
  17
  18 #include "src/simix/SynchroSleep.hpp"
  19 #include "src/simix/SynchroRaw.hpp"
  20 #include "src/simix/SynchroIo.hpp"
  21
  22 #ifdef HAVE_SMPI
  23 #include "src/smpi/private.h"
  24 #endif
  25
  26 XBT_LOG_NEW_DEFAULT_SUBCATEGORY(simix_process, simix, "Logging specific to SIMIX (process)");
  27
  28 unsigned long simix_process_maxpid = 0;
  29
  30 /**
  31  * \brief Returns the current agent.
  32  *
  33  * This functions returns the currently running SIMIX process.
  34  *
  35  * \return The SIMIX process
  36  */
  37 smx_process_t SIMIX_process_self(void)
  38 {
  39   smx_context_t self_context = SIMIX_context_self();
  40
  41   return self_context ? SIMIX_context_get_process(self_context) : NULL;
  42 }
  43
  44 /**
  45  * \brief Returns whether a process has pending asynchronous communications.
  46  * \return true if there are asynchronous communications in this process
  47  */
  48 int SIMIX_process_has_pending_comms(smx_process_t process) {
  49
  50   return xbt_fifo_size(process->comms) > 0;
  51 }
  52
  53 /**
  54  * \brief Moves a process to the list of processes to destroy.
  55  */
  56 void SIMIX_process_cleanup(smx_process_t process)
  57 {
  58   XBT_DEBUG("Cleanup process %s (%p), waiting synchro %p",
  59       process->name, process, process->waiting_synchro);
  60
  61   SIMIX_process_on_exit_runall(process);
  62
  63   /* Unregister from the kill timer if any */
  64   if (process->kill_timer != NULL)
  65       SIMIX_timer_remove(process->kill_timer);
  66
  67   xbt_os_mutex_acquire(simix_global->mutex);
  68
  69   /* cancel non-blocking communications */
  70   smx_synchro_t synchro;
  71   while ((synchro = (smx_synchro_t) xbt_fifo_pop(process->comms))) {
  72     simgrid::simix::Comm *comm = static_cast<simgrid::simix::Comm*>(synchro);
  73
  74     /* make sure no one will finish the comm after this process is destroyed,
  75      * because src_proc or dst_proc would be an invalid pointer */
  76     SIMIX_comm_cancel(comm);
  77
  78     if (comm->src_proc == process) {
  79       XBT_DEBUG("Found an unfinished send comm %p (detached = %d), state %d, src = %p, dst = %p",
  80           comm, comm->detached, (int)comm->state, comm->src_proc, comm->dst_proc);
  81       comm->src_proc = NULL;
  82
  83       /* I'm not supposed to destroy a detached comm from the sender side, */
  84       if (!comm->detached)
  85         SIMIX_comm_destroy(comm);
  86       else
  87         XBT_DEBUG("Don't destroy it since it's a detached comm");
  88
  89     }
  90     else if (comm->dst_proc == process){
  91       XBT_DEBUG("Found an unfinished recv comm %p, state %d, src = %p, dst = %p",
  92           comm, (int)comm->state, comm->src_proc, comm->dst_proc);
  93       comm->dst_proc = NULL;
  94
  95       if (comm->detached && comm->refcount == 1 && comm->src_proc != NULL) {
  96         /* the comm will be freed right now, remove it from the sender */
  97         xbt_fifo_remove(comm->src_proc->comms, comm);
  98       }
  99       SIMIX_comm_destroy(comm);
 100     } else {
 101       xbt_die("Communication synchro %p is in my list but I'm not the sender nor the receiver", synchro);
 102     }
 103   }
 104
 105   XBT_DEBUG("%p should not be run anymore",process);
 106   xbt_swag_remove(process, simix_global->process_list);
 107   if (process->host)
 108     xbt_swag_remove(process, sg_host_simix(process->host)->process_list);
 109   xbt_swag_insert(process, simix_global->process_to_destroy);
 110   process->context->iwannadie = 0;
 111
 112   xbt_os_mutex_release(simix_global->mutex);
 113 }
 114
 115 /**
 116  * Garbage collection
 117  *
 118  * Should be called some time to time to free the memory allocated for processes
 119  * that have finished (or killed).
 120  */
 121 void SIMIX_process_empty_trash(void)
 122 {
 123   smx_process_t process = NULL;
 124
 125   while ((process = (smx_process_t) xbt_swag_extract(simix_global->process_to_destroy))) {
 126     XBT_DEBUG("Getting rid of %p",process);
 127
 128     SIMIX_context_free(process->context);
 129
 130     /* Free the exception allocated at creation time */
 131     free(process->running_ctx);
 132     xbt_dict_free(&process->properties);
 133
 134     xbt_fifo_free(process->comms);
 135
 136     xbt_dynar_free(&process->on_exit);
 137
 138     xbt_free(process->name);
 139     xbt_free(process);
 140   }
 141 }
 142
 143 namespace simgrid {
 144 namespace simix {
 145
 146 void create_maestro(std::function<void()> code)
 147 {
 148   smx_process_t maestro = NULL;
 149   /* Create maestro process and intilialize it */
 150   maestro = xbt_new0(s_smx_process_t, 1);
 151   maestro->pid = simix_process_maxpid++;
 152   maestro->ppid = -1;
 153   maestro->name = (char*) "";
 154   maestro->data = nullptr;
 155   maestro->running_ctx = (xbt_running_ctx_t*) xbt_malloc0(sizeof(xbt_running_ctx_t));
 156   XBT_RUNNING_CTX_INITIALIZE(maestro->running_ctx);
 157
 158   if (!code) {
 159     maestro->context = SIMIX_context_new(NULL, 0, nullptr, NULL, maestro);
 160   } else {
 161     if (!simix_global)
 162       xbt_die("simix is not initialized, please call MSG_init first");
 163     maestro->context =
 164       simix_global->context_factory->create_maestro(code, maestro);
 165   }
 166
 167   maestro->simcall.issuer = maestro;
 168   simix_global->maestro_process = maestro;
 169 }
 170
 171 }
 172 }
 173
 174 /**
 175  * \brief Creates and runs the maestro process
 176  */
 177 void SIMIX_maestro_create(void (*code)(void*), void* data)
 178 {
 179   simgrid::simix::create_maestro(std::bind(code, data));
 180 }
 181
 182 /**
 183  * \brief Stops a process.
 184  *
 185  * Stops the process, execute all the registered on_exit functions,
 186  * register it to the list of the process to restart if needed
 187  * and stops its context.
 188  */
 189 void SIMIX_process_stop(smx_process_t arg) {
 190   /* execute the on_exit functions */
 191   SIMIX_process_on_exit_runall(arg);
 192   /* Add the process to the list of process to restart, only if the host is down */
 193   if (arg->auto_restart && arg->host->isOff()) {
 194     SIMIX_host_add_auto_restart_process(arg->host,arg->name,arg->code, arg->data,
 195                                         sg_host_get_name(arg->host),
 196                                         SIMIX_timer_get_date(arg->kill_timer),
 197                                         arg->argc,arg->argv,arg->properties,
 198                                         arg->auto_restart);
 199   }
 200   XBT_DEBUG("Process %s (%s) is dead",arg->name,sg_host_get_name(arg->host));
 201   /* stop the context */
 202   SIMIX_context_stop(arg->context);
 203 }
 204
 205 /**
 206  * \brief Same as SIMIX_process_create() but with only one argument (used by timers).
 207  * This function frees the argument.
 208  * \return the process created
 209  */
 210 smx_process_t SIMIX_process_create_from_wrapper(smx_process_arg_t args) {
 211
 212   smx_process_t process = simix_global->create_process_function(
 213                                         args->name,
 214                                         args->code,
 215                                         args->data,
 216                                         args->hostname,
 217                                         args->kill_time,
 218                                         args->argc,
 219                                         args->argv,
 220                                         args->properties,
 221                                         args->auto_restart,
 222                                         NULL);
 223   xbt_free(args);
 224   return process;
 225 }
 226
 227
 228 void* simcall_HANDLER_process_create(smx_simcall_t simcall,
 229                           const char *name,
 230                           xbt_main_func_t code,
 231                           void *data,
 232                           const char *hostname,
 233                           double kill_time,
 234                           int argc, char **argv,
 235                           xbt_dict_t properties,
 236                           int auto_restart){
 237   return (void*)SIMIX_process_create(name, code, data, hostname,
 238                        kill_time, argc, argv, properties, auto_restart,
 239                        simcall->issuer);
 240 }
 241
 242 static void kill_process(void* process)
 243 {
 244   simix_global->kill_process_function((smx_process_t) process);
 245 }
 246
 247 /**
 248  * \brief Internal function to create a process.
 249  *
 250  * This function actually creates the process.
 251  * It may be called when a SIMCALL_PROCESS_CREATE simcall occurs,
 252  * or directly for SIMIX internal purposes. The sure thing is that it's called from maestro context.
 253  *
 254  * \return the process created
 255  */
 256 smx_process_t SIMIX_process_create(
 257                           const char *name,
 258                           xbt_main_func_t code,
 259                           void *data,
 260                           const char *hostname,
 261                           double kill_time,
 262                           int argc, char **argv,
 263                           xbt_dict_t properties,
 264                           int auto_restart,
 265                           smx_process_t parent_process)
 266 {
 267   smx_process_t process = NULL;
 268   sg_host_t host = sg_host_by_name(hostname);
 269
 270   XBT_DEBUG("Start process %s on host '%s'", name, hostname);
 271
 272   if (host->isOff()) {
 273     int i;
 274     XBT_WARN("Cannot launch process '%s' on failed host '%s'", name,
 275           hostname);
 276     for (i = 0; i < argc; i++)
 277       xbt_free(argv[i]);
 278     xbt_free(argv);
 279   }
 280   else {
 281     process = xbt_new0(s_smx_process_t, 1);
 282
 283     xbt_assert(((code != NULL) && (host != NULL)), "Invalid parameters");
 284     /* Process data */
 285     process->pid = simix_process_maxpid++;
 286     process->name = xbt_strdup(name);
 287     process->host = host;
 288     process->data = data;
 289     process->comms = xbt_fifo_new();
 290     process->simcall.issuer = process;
 291     /* Initiliaze data segment to default value */
 292     SIMIX_segment_index_set(process, -1);
 293
 294      if (parent_process != NULL) {
 295        process->ppid = SIMIX_process_get_PID(parent_process);
 296        /* SMPI process have their own data segment and
 297           each other inherit from their father */
 298 #if HAVE_SMPI
 299        if(smpi_privatize_global_variables){
 300          if( parent_process->pid != 0){
 301            SIMIX_segment_index_set(process, parent_process->segment_index);
 302          } else {
 303            SIMIX_segment_index_set(process, process->pid - 1);
 304          }
 305        }
 306 #endif
 307      } else {
 308        process->ppid = -1;
 309      }
 310
 311     /* Process data for auto-restart */
 312     process->auto_restart = auto_restart;
 313     process->code = code;
 314     process->argc = argc;
 315     process->argv = argv;
 316
 317
 318     XBT_VERB("Create context %s", process->name);
 319     process->context = SIMIX_context_new(code, argc, argv, simix_global->cleanup_process_function, process);
 320
 321     process->running_ctx = (xbt_running_ctx_t*) xbt_malloc0(sizeof(xbt_running_ctx_t));
 322     XBT_RUNNING_CTX_INITIALIZE(process->running_ctx);
 323
 324     if(MC_is_active()){
 325       MC_ignore_heap(process->running_ctx, sizeof(*process->running_ctx));
 326     }
 327
 328     /* Add properties */
 329     process->properties = properties;
 330
 331     /* Add the process to it's host process list */
 332     xbt_swag_insert(process, sg_host_simix(host)->process_list);
 333
 334     XBT_DEBUG("Start context '%s'", process->name);
 335
 336     /* Now insert it in the global process list and in the process to run list */
 337     xbt_swag_insert(process, simix_global->process_list);
 338     XBT_DEBUG("Inserting %s(%s) in the to_run list", process->name, sg_host_get_name(host));
 339     xbt_dynar_push_as(simix_global->process_to_run, smx_process_t, process);
 340
 341     if (kill_time > SIMIX_get_clock() && simix_global->kill_process_function) {
 342       XBT_DEBUG("Process %s(%s) will be kill at time %f", process->name,
 343           sg_host_get_name(process->host), kill_time);
 344       process->kill_timer = SIMIX_timer_set(kill_time, kill_process, process);
 345     }
 346
 347     /* Tracing the process creation */
 348     TRACE_msg_process_create(process->name, process->pid, process->host);
 349   }
 350   return process;
 351 }
 352
 353 smx_process_t SIMIX_process_attach(
 354   const char* name,
 355   void *data,
 356   const char* hostname,
 357   xbt_dict_t properties,
 358   smx_process_t parent_process)
 359 {
 360   // This is mostly a copy/paste from SIMIX_process_new(),
 361   // it'd be nice to share some code between those two functions.
 362
 363   sg_host_t host = sg_host_by_name(hostname);
 364   XBT_DEBUG("Attach process %s on host '%s'", name, hostname);
 365
 366   if (host->isOff()) {
 367     XBT_WARN("Cannot launch process '%s' on failed host '%s'",
 368       name, hostname);
 369     return nullptr;
 370   }
 371
 372   smx_process_t process = xbt_new0(s_smx_process_t, 1);
 373   /* Process data */
 374   process->pid = simix_process_maxpid++;
 375   process->name = xbt_strdup(name);
 376   process->host = host;
 377   process->data = data;
 378   process->comms = xbt_fifo_new();
 379   process->simcall.issuer = process;
 380   process->ppid = -1;
 381   /* Initiliaze data segment to default value */
 382   SIMIX_segment_index_set(process, -1);
 383   if (parent_process != NULL) {
 384     process->ppid = SIMIX_process_get_PID(parent_process);
 385    /* SMPI process have their own data segment and
 386       each other inherit from their father */
 387   #if HAVE_SMPI
 388     if(smpi_privatize_global_variables){
 389       if(parent_process->pid != 0){
 390         SIMIX_segment_index_set(process, parent_process->segment_index);
 391       } else {
 392         SIMIX_segment_index_set(process, process->pid - 1);
 393       }
 394     }
 395   #endif
 396   }
 397
 398   /* Process data for auto-restart */
 399   process->auto_restart = false;
 400   process->code = nullptr;
 401   process->argc = 0;
 402   process->argv = nullptr;
 403
 404   XBT_VERB("Create context %s", process->name);
 405   if (!simix_global)
 406     xbt_die("simix is not initialized, please call MSG_init first");
 407   process->context = simix_global->context_factory->attach(
 408     simix_global->cleanup_process_function, process);
 409
 410   process->running_ctx = (xbt_running_ctx_t*) xbt_malloc0(sizeof(xbt_running_ctx_t));
 411   XBT_RUNNING_CTX_INITIALIZE(process->running_ctx);
 412
 413   if(MC_is_active()){
 414     MC_ignore_heap(process->running_ctx, sizeof(*process->running_ctx));
 415   }
 416
 417   /* Add properties */
 418   process->properties = properties;
 419
 420   /* Add the process to it's host process list */
 421   xbt_swag_insert(process, sg_host_simix(host)->process_list);
 422
 423   /* Now insert it in the global process list and in the process to run list */
 424   xbt_swag_insert(process, simix_global->process_list);
 425   XBT_DEBUG("Inserting %s(%s) in the to_run list", process->name, sg_host_get_name(host));
 426   xbt_dynar_push_as(simix_global->process_to_run, smx_process_t, process);
 427
 428   /* Tracing the process creation */
 429   TRACE_msg_process_create(process->name, process->pid, process->host);
 430
 431   auto context = dynamic_cast<simgrid::simix::AttachContext*>(process->context);
 432   if (!context)
 433     xbt_die("Not a suitable context");
 434
 435   context->attach_start();
 436   return process;
 437 }
 438
 439 void SIMIX_process_detach(void)
 440 {
 441   auto context = dynamic_cast<simgrid::simix::AttachContext*>(SIMIX_context_self());
 442   if (!context)
 443     xbt_die("Not a suitable context");
 444
 445   simix_global->cleanup_process_function(context->process());
 446
 447   // Let maestro ignore we are still alive:
 448   // xbt_swag_remove(context->process(), simix_global->process_list);
 449
 450   // TODDO, Remove from proces list:
 451   //   xbt_swag_remove(process, sg_host_simix(host)->process_list);
 452
 453   context->attach_stop();
 454   // delete context;
 455 }
 456
 457 /**
 458  * \brief Executes the processes from simix_global->process_to_run.
 459  *
 460  * The processes of simix_global->process_to_run are run (in parallel if
 461  * possible).  On exit, simix_global->process_to_run is empty, and
 462  * simix_global->process_that_ran contains the list of processes that just ran.
 463  * The two lists are swapped so, be careful when using them before and after a
 464  * call to this function.
 465  */
 466 void SIMIX_process_runall(void)
 467 {
 468   SIMIX_context_runall();
 469
 470   xbt_dynar_t tmp = simix_global->process_that_ran;
 471   simix_global->process_that_ran = simix_global->process_to_run;
 472   simix_global->process_to_run = tmp;
 473   xbt_dynar_reset(simix_global->process_to_run);
 474 }
 475
 476 void simcall_HANDLER_process_kill(smx_simcall_t simcall, smx_process_t process) {
 477   SIMIX_process_kill(process, simcall->issuer);
 478 }
 479 /**
 480  * \brief Internal function to kill a SIMIX process.
 481  *
 482  * This function may be called when a SIMCALL_PROCESS_KILL simcall occurs,
 483  * or directly for SIMIX internal purposes.
 484  *
 485  * \param process poor victim
 486  * \param issuer the process which has sent the PROCESS_KILL. Important to not schedule twice the same process.
 487  */
 488 void SIMIX_process_kill(smx_process_t process, smx_process_t issuer) {
 489
 490   XBT_DEBUG("Killing process %s on %s", process->name, sg_host_get_name(process->host));
 491
 492   process->context->iwannadie = 1;
 493   process->blocked = 0;
 494   process->suspended = 0;
 495   process->doexception = 0;
 496
 497   /* destroy the blocking synchro if any */
 498   if (process->waiting_synchro) {
 499
 500     simgrid::simix::Exec *exec = dynamic_cast<simgrid::simix::Exec*>(process->waiting_synchro);
 501     simgrid::simix::Comm *comm = dynamic_cast<simgrid::simix::Comm*>(process->waiting_synchro);
 502     simgrid::simix::Sleep *sleep = dynamic_cast<simgrid::simix::Sleep*>(process->waiting_synchro);
 503     simgrid::simix::Raw *raw = dynamic_cast<simgrid::simix::Raw*>(process->waiting_synchro);
 504     simgrid::simix::Io *io = dynamic_cast<simgrid::simix::Io*>(process->waiting_synchro);
 505
 506     if (exec != nullptr) {
 507       SIMIX_execution_destroy(process->waiting_synchro);
 508
 509     } else if (comm != nullptr) {
 510       xbt_fifo_remove(process->comms, process->waiting_synchro);
 511       SIMIX_comm_cancel(process->waiting_synchro);
 512       xbt_fifo_remove(process->waiting_synchro->simcalls, &process->simcall);
 513       SIMIX_comm_destroy(process->waiting_synchro);
 514
 515     } else if (sleep != nullptr) {
 516       SIMIX_process_sleep_destroy(process->waiting_synchro);
 517
 518     } else if (raw != nullptr) {
 519       SIMIX_synchro_stop_waiting(process, &process->simcall);
 520       SIMIX_synchro_destroy(process->waiting_synchro);
 521
 522     } else if (io != nullptr) {
 523       SIMIX_io_destroy(process->waiting_synchro);
 524     }
 525
 526     /*
 527     switch (process->waiting_synchro->type) {
 528     case SIMIX_SYNC_JOIN:
 529       SIMIX_process_sleep_destroy(process->waiting_synchro);
 530       break;
 531     } */
 532
 533     process->waiting_synchro = NULL;
 534   }
 535   if(!xbt_dynar_member(simix_global->process_to_run, &(process)) && process != issuer) {
 536     XBT_DEBUG("Inserting %s in the to_run list", process->name);
 537     xbt_dynar_push_as(simix_global->process_to_run, smx_process_t, process);
 538   }
 539
 540 }
 541
 542 /** @brief Ask another process to raise the given exception
 543  *
 544  * @param cat category of exception
 545  * @param value value associated to the exception
 546  * @param msg string information associated to the exception
 547  */
 548 void SIMIX_process_throw(smx_process_t process, xbt_errcat_t cat, int value, const char *msg) {
 549   SMX_EXCEPTION(process, cat, value, msg);
 550
 551   if (process->suspended)
 552     SIMIX_process_resume(process,SIMIX_process_self());
 553
 554   /* cancel the blocking synchro if any */
 555   if (process->waiting_synchro) {
 556
 557     simgrid::simix::Exec *exec = dynamic_cast<simgrid::simix::Exec*>(process->waiting_synchro);
 558     if (exec != nullptr) {
 559       SIMIX_execution_cancel(process->waiting_synchro);
 560     }
 561
 562     simgrid::simix::Comm *comm = dynamic_cast<simgrid::simix::Comm*>(process->waiting_synchro);
 563     if (comm != nullptr) {
 564       xbt_fifo_remove(process->comms, process->waiting_synchro);
 565       SIMIX_comm_cancel(process->waiting_synchro);
 566     }
 567
 568     simgrid::simix::Sleep *sleep = dynamic_cast<simgrid::simix::Sleep*>(process->waiting_synchro);
 569     if (sleep != nullptr) {
 570       SIMIX_process_sleep_destroy(process->waiting_synchro);
 571       if (!xbt_dynar_member(simix_global->process_to_run, &(process)) && process != SIMIX_process_self()) {
 572         XBT_DEBUG("Inserting %s in the to_run list", process->name);
 573         xbt_dynar_push_as(simix_global->process_to_run, smx_process_t, process);
 574       }
 575     }
 576
 577     simgrid::simix::Raw *raw = dynamic_cast<simgrid::simix::Raw*>(process->waiting_synchro);
 578     if (raw != nullptr) {
 579       SIMIX_synchro_stop_waiting(process, &process->simcall);
 580     }
 581
 582     simgrid::simix::Io *io = dynamic_cast<simgrid::simix::Io*>(process->waiting_synchro);
 583     if (io != nullptr) {
 584       SIMIX_io_destroy(process->waiting_synchro);
 585     }
 586   }
 587   process->waiting_synchro = NULL;
 588
 589 }
 590
 591 void simcall_HANDLER_process_killall(smx_simcall_t simcall, int reset_pid) {
 592   SIMIX_process_killall(simcall->issuer, reset_pid);
 593 }
 594 /**
 595  * \brief Kills all running processes.
 596  * \param issuer this one will not be killed
 597  */
 598 void SIMIX_process_killall(smx_process_t issuer, int reset_pid)
 599 {
 600   smx_process_t p = NULL;
 601
 602   while ((p = (smx_process_t) xbt_swag_extract(simix_global->process_list))) {
 603     if (p != issuer) {
 604       SIMIX_process_kill(p,issuer);
 605     }
 606   }
 607
 608   if (reset_pid > 0)
 609     simix_process_maxpid = reset_pid;
 610
 611   SIMIX_context_runall();
 612
 613   SIMIX_process_empty_trash();
 614 }
 615
 616 void simcall_HANDLER_process_set_host(smx_simcall_t simcall, smx_process_t process, sg_host_t dest)
 617 {
 618   process->new_host = dest;
 619 }
 620 void SIMIX_process_change_host(smx_process_t process,
 621              sg_host_t dest)
 622 {
 623   xbt_assert((process != NULL), "Invalid parameters");
 624   xbt_swag_remove(process, sg_host_simix(process->host)->process_list);
 625   process->host = dest;
 626   xbt_swag_insert(process, sg_host_simix(dest)->process_list);
 627 }
 628
 629
 630 void simcall_HANDLER_process_suspend(smx_simcall_t simcall, smx_process_t process)
 631 {
 632   smx_synchro_t sync_suspend =
 633       SIMIX_process_suspend(process, simcall->issuer);
 634
 635   if (process != simcall->issuer) {
 636     SIMIX_simcall_answer(simcall);
 637   } else {
 638     xbt_fifo_push(sync_suspend->simcalls, simcall);
 639     process->waiting_synchro = sync_suspend;
 640     SIMIX_execution_suspend(process->waiting_synchro);
 641   }
 642   /* If we are suspending ourselves, then just do not finish the simcall now */
 643 }
 644
 645 smx_synchro_t SIMIX_process_suspend(smx_process_t process, smx_process_t issuer)
 646 {
 647   xbt_assert((process != NULL), "Invalid parameters");
 648
 649   if (process->suspended) {
 650     XBT_DEBUG("Process '%s' is already suspended", process->name);
 651     return NULL;
 652   }
 653
 654   process->suspended = 1;
 655
 656   /* If we are suspending another process, and it is waiting on a sync, suspend its synchronization. */
 657   if (process != issuer) {
 658
 659     if (process->waiting_synchro) {
 660
 661       simgrid::simix::Exec *exec = dynamic_cast<simgrid::simix::Exec*>(process->waiting_synchro);
 662       if (exec != nullptr) {
 663         SIMIX_execution_suspend(process->waiting_synchro);
 664       }
 665
 666       simgrid::simix::Comm *comm = dynamic_cast<simgrid::simix::Comm*>(process->waiting_synchro);
 667       if (comm != nullptr) {
 668         SIMIX_comm_suspend(process->waiting_synchro);
 669       }
 670
 671       simgrid::simix::Sleep *sleep = dynamic_cast<simgrid::simix::Sleep*>(process->waiting_synchro);
 672       if (sleep != nullptr) {
 673         SIMIX_process_sleep_suspend(process->waiting_synchro);
 674       }
 675
 676       /* The suspension of raw synchros is delayed to when the process is rescheduled. */
 677       return NULL;
 678     } else {
 679       /* If the other process is not waiting, its suspension is delayed to when the process is rescheduled. */
 680       return NULL;
 681     }
 682   } else {
 683     /* FIXME: computation size is zero. Is it okay that bound is zero ? */
 684     return SIMIX_execution_start(process, "suspend", 0.0, 1.0, 0.0, 0);
 685   }
 686 }
 687
 688 void simcall_HANDLER_process_resume(smx_simcall_t simcall, smx_process_t process){
 689   SIMIX_process_resume(process, simcall->issuer);
 690 }
 691
 692 void SIMIX_process_resume(smx_process_t process, smx_process_t issuer)
 693 {
 694   XBT_IN("process = %p, issuer = %p", process, issuer);
 695
 696   if(process->context->iwannadie) {
 697     XBT_VERB("Ignoring request to suspend a process that is currently dying.");
 698     return;
 699   }
 700
 701   if(!process->suspended) return;
 702   process->suspended = 0;
 703
 704   /* If we are resuming another process, resume the synchronization it was waiting for
 705      if any. Otherwise add it to the list of process to run in the next round. */
 706   if (process != issuer) {
 707
 708     if (process->waiting_synchro) {
 709     simgrid::simix::Exec *exec = dynamic_cast<simgrid::simix::Exec*>(process->waiting_synchro);
 710     if (exec != nullptr) {
 711       SIMIX_execution_resume(process->waiting_synchro);
 712     }
 713
 714     simgrid::simix::Comm *comm = dynamic_cast<simgrid::simix::Comm*>(process->waiting_synchro);
 715     if (comm != nullptr) {
 716       SIMIX_comm_resume(process->waiting_synchro);
 717     }
 718
 719     simgrid::simix::Sleep *sleep = dynamic_cast<simgrid::simix::Sleep*>(process->waiting_synchro);
 720     if (sleep != nullptr) {
 721       SIMIX_process_sleep_resume(process->waiting_synchro);
 722     }
 723
 724     /* I cannot resume raw synchros now. This is delayed to when the process is rescheduled at
 725      * the end of the synchro. */
 726
 727     }
 728   } else XBT_WARN("Strange. Process %p is trying to resume himself.", issuer);
 729
 730   XBT_OUT();
 731 }
 732
 733 int SIMIX_process_get_maxpid(void) {
 734   return simix_process_maxpid;
 735 }
 736
 737 int SIMIX_process_count(void)
 738 {
 739   return xbt_swag_size(simix_global->process_list);
 740 }
 741
 742 int SIMIX_process_get_PID(smx_process_t self){
 743   if (self == NULL)
 744     return 0;
 745   else
 746     return self->pid;
 747 }
 748
 749 int SIMIX_process_get_PPID(smx_process_t self){
 750   if (self == NULL)
 751     return 0;
 752   else
 753     return self->ppid;
 754 }
 755
 756 void* SIMIX_process_self_get_data()
 757 {
 758   smx_process_t self = SIMIX_process_self();
 759
 760   if (!self) {
 761     return NULL;
 762   }
 763   return SIMIX_process_get_data(self);
 764 }
 765
 766 void SIMIX_process_self_set_data(void *data)
 767 {
 768   smx_process_t self = SIMIX_process_self();
 769
 770   SIMIX_process_set_data(self, data);
 771 }
 772
 773 void* SIMIX_process_get_data(smx_process_t process)
 774 {
 775   return process->data;
 776 }
 777
 778 void SIMIX_process_set_data(smx_process_t process, void *data)
 779 {
 780   process->data = data;
 781 }
 782
 783 sg_host_t SIMIX_process_get_host(smx_process_t process)
 784 {
 785   return process->host;
 786 }
 787
 788 xbt_main_func_t SIMIX_process_get_code(void){
 789   return SIMIX_process_self()->code;
 790 }
 791
 792 /* needs to be public and without simcall because it is called
 793    by exceptions and logging events */
 794 const char* SIMIX_process_self_get_name(void) {
 795
 796   smx_process_t process = SIMIX_process_self();
 797   if (process == NULL || process == simix_global->maestro_process)
 798     return "maestro";
 799
 800   return SIMIX_process_get_name(process);
 801 }
 802
 803 const char* SIMIX_process_get_name(smx_process_t process)
 804 {
 805   return process->name;
 806 }
 807
 808 smx_process_t SIMIX_process_get_by_name(const char* name)
 809 {
 810   smx_process_t proc;
 811
 812   xbt_swag_foreach(proc, simix_global->process_list)
 813   {
 814     if(!strcmp(name, proc->name))
 815       return proc;
 816   }
 817   return NULL;
 818 }
 819
 820 int SIMIX_process_is_suspended(smx_process_t process)
 821 {
 822   return process->suspended;
 823 }
 824
 825 xbt_dict_t SIMIX_process_get_properties(smx_process_t process)
 826 {
 827   return process->properties;
 828 }
 829
 830 void simcall_HANDLER_process_join(smx_simcall_t simcall, smx_process_t process, double timeout)
 831 {
 832   smx_synchro_t sync = SIMIX_process_join(simcall->issuer, process, timeout);
 833   xbt_fifo_push(sync->simcalls, simcall);
 834   simcall->issuer->waiting_synchro = sync;
 835 }
 836
 837 static int SIMIX_process_join_finish(smx_process_exit_status_t status, smx_synchro_t synchro){
 838   simgrid::simix::Sleep *sleep = static_cast<simgrid::simix::Sleep*>(synchro);
 839
 840   if (sleep->surf_sleep) {
 841     sleep->surf_sleep->cancel();
 842
 843     smx_simcall_t simcall;
 844     while ((simcall = (smx_simcall_t) xbt_fifo_shift(sleep->simcalls))) {
 845       simcall_process_sleep__set__result(simcall, SIMIX_DONE);
 846       simcall->issuer->waiting_synchro = NULL;
 847       if (simcall->issuer->suspended) {
 848         XBT_DEBUG("Wait! This process is suspended and can't wake up now.");
 849         simcall->issuer->suspended = 0;
 850         simcall_HANDLER_process_suspend(simcall, simcall->issuer);
 851       } else {
 852         SIMIX_simcall_answer(simcall);
 853       }
 854     }
 855     sleep->surf_sleep->unref();
 856     sleep->surf_sleep = NULL;
 857   }
 858   delete sleep;
 859   return 0;
 860 }
 861
 862 smx_synchro_t SIMIX_process_join(smx_process_t issuer, smx_process_t process, double timeout)
 863 {
 864   smx_synchro_t res = SIMIX_process_sleep(issuer, timeout);
 865   SIMIX_process_on_exit(process, (int_f_pvoid_pvoid_t)SIMIX_process_join_finish, res);
 866   return res;
 867 }
 868
 869 void simcall_HANDLER_process_sleep(smx_simcall_t simcall, double duration)
 870 {
 871   if (MC_is_active() || MC_record_replay_is_active()) {
 872     MC_process_clock_add(simcall->issuer, duration);
 873     simcall_process_sleep__set__result(simcall, SIMIX_DONE);
 874     SIMIX_simcall_answer(simcall);
 875     return;
 876   }
 877   smx_synchro_t sync = SIMIX_process_sleep(simcall->issuer, duration);
 878   xbt_fifo_push(sync->simcalls, simcall);
 879   simcall->issuer->waiting_synchro = sync;
 880 }
 881
 882 smx_synchro_t SIMIX_process_sleep(smx_process_t process, double duration)
 883 {
 884   sg_host_t host = process->host;
 885
 886   /* check if the host is active */
 887   if (host->isOff())
 888     THROWF(host_error, 0, "Host %s failed, you cannot call this function", sg_host_get_name(host));
 889
 890   simgrid::simix::Sleep *synchro = new simgrid::simix::Sleep();
 891   synchro->name = NULL;
 892
 893   synchro->host = host;
 894   synchro->surf_sleep = surf_host_sleep(host, duration);
 895   synchro->surf_sleep->setData(synchro);
 896   XBT_DEBUG("Create sleep synchronization %p", synchro);
 897
 898   return synchro;
 899 }
 900
 901 void SIMIX_post_process_sleep(smx_synchro_t synchro)
 902 {
 903   smx_simcall_t simcall;
 904   e_smx_state_t state;
 905   simgrid::simix::Sleep *sleep = static_cast<simgrid::simix::Sleep*>(synchro);
 906
 907   while ((simcall = (smx_simcall_t) xbt_fifo_shift(synchro->simcalls))) {
 908
 909     switch (sleep->surf_sleep->getState()){
 910       case simgrid::surf::Action::State::failed:
 911         simcall->issuer->context->iwannadie = 1;
 912         //SMX_EXCEPTION(simcall->issuer, host_error, 0, "Host failed");
 913         state = SIMIX_SRC_HOST_FAILURE;
 914         break;
 915
 916       case simgrid::surf::Action::State::done:
 917         state = SIMIX_DONE;
 918         break;
 919
 920       default:
 921         THROW_IMPOSSIBLE;
 922         break;
 923     }
 924     if (simcall->issuer->host->isOff()) {
 925       simcall->issuer->context->iwannadie = 1;
 926     }
 927     simcall_process_sleep__set__result(simcall, state);
 928     simcall->issuer->waiting_synchro = NULL;
 929     if (simcall->issuer->suspended) {
 930       XBT_DEBUG("Wait! This process is suspended and can't wake up now.");
 931       simcall->issuer->suspended = 0;
 932       simcall_HANDLER_process_suspend(simcall, simcall->issuer);
 933     } else {
 934       SIMIX_simcall_answer(simcall);
 935     }
 936   }
 937
 938   SIMIX_process_sleep_destroy(synchro);
 939 }
 940
 941 void SIMIX_process_sleep_destroy(smx_synchro_t synchro)
 942 {
 943   XBT_DEBUG("Destroy synchro %p", synchro);
 944   simgrid::simix::Sleep *sleep = static_cast<simgrid::simix::Sleep*>(synchro);
 945
 946   if (sleep->surf_sleep) {
 947     sleep->surf_sleep->unref();
 948     sleep->surf_sleep = NULL;
 949   }
 950 }
 951
 952 void SIMIX_process_sleep_suspend(smx_synchro_t synchro)
 953 {
 954   simgrid::simix::Sleep *sleep = static_cast<simgrid::simix::Sleep*>(synchro);
 955   sleep->surf_sleep->suspend();
 956 }
 957
 958 void SIMIX_process_sleep_resume(smx_synchro_t synchro)
 959 {
 960   XBT_DEBUG("Synchro state is %d on process_sleep_resume.", synchro->state);
 961   simgrid::simix::Sleep *sleep = static_cast<simgrid::simix::Sleep*>(synchro);
 962
 963   sleep->surf_sleep->resume();
 964 }
 965
 966 /**
 967  * \brief Calling this function makes the process to yield.
 968  *
 969  * Only the current process can call this function, giving back the control to
 970  * maestro.
 971  *
 972  * \param self the current process
 973  */
 974 void SIMIX_process_yield(smx_process_t self)
 975 {
 976   XBT_DEBUG("Yield process '%s'", self->name);
 977
 978   /* Go into sleep and return control to maestro */
 979   SIMIX_context_suspend(self->context);
 980
 981   /* Ok, maestro returned control to us */
 982   XBT_DEBUG("Control returned to me: '%s'", self->name);
 983
 984   if (self->new_host) {
 985     SIMIX_process_change_host(self, self->new_host);
 986     self->new_host = NULL;
 987   }
 988
 989   if (self->context->iwannadie){
 990     XBT_DEBUG("I wanna die!");
 991     SIMIX_process_stop(self);
 992   }
 993
 994   if (self->suspended) {
 995     XBT_DEBUG("Hey! I'm suspended.");
 996     xbt_assert(!self->doexception, "Gasp! This exception may be lost by subsequent calls.");
 997     self->suspended = 0;
 998     SIMIX_process_suspend(self, self);
 999   }
1000
1001   if (self->doexception) {
1002     XBT_DEBUG("Wait, maestro left me an exception");
1003     self->doexception = 0;
1004     RETHROW;
1005   }
1006
1007   if(SMPI_switch_data_segment && self->segment_index != -1){
1008     SMPI_switch_data_segment(self->segment_index);
1009   }
1010 }
1011
1012 /* callback: context fetching */
1013 xbt_running_ctx_t *SIMIX_process_get_running_context(void)
1014 {
1015   smx_process_t process = SIMIX_process_self();
1016   if (process)
1017     return process->running_ctx;
1018   else
1019     return nullptr;
1020 }
1021
1022 /* callback: termination */
1023 void SIMIX_process_exception_terminate(xbt_ex_t * e)
1024 {
1025   xbt_ex_display(e);
1026   xbt_abort();
1027 }
1028
1029 smx_context_t SIMIX_process_get_context(smx_process_t p) {
1030   return p->context;
1031 }
1032
1033 void SIMIX_process_set_context(smx_process_t p,smx_context_t c) {
1034   p->context = c;
1035 }
1036
1037 /**
1038  * \brief Returns the list of processes to run.
1039  */
1040 xbt_dynar_t SIMIX_process_get_runnable(void)
1041 {
1042   return simix_global->process_to_run;
1043 }
1044
1045 /**
1046  * \brief Returns the process from PID.
1047  */
1048 smx_process_t SIMIX_process_from_PID(int PID)
1049 {
1050   smx_process_t proc;
1051   xbt_swag_foreach(proc, simix_global->process_list) {
1052    if (proc->pid == (unsigned long) PID)
1053     return proc;
1054   }
1055   return NULL;
1056 }
1057
1058 /** @brief returns a dynar containg all currently existing processes */
1059 xbt_dynar_t SIMIX_processes_as_dynar(void) {
1060   smx_process_t proc;
1061   xbt_dynar_t res = xbt_dynar_new(sizeof(smx_process_t),NULL);
1062   xbt_swag_foreach(proc, simix_global->process_list) {
1063     xbt_dynar_push(res,&proc);
1064   }
1065   return res;
1066 }
1067
1068
1069 void SIMIX_process_on_exit_runall(smx_process_t process) {
1070   s_smx_process_exit_fun_t exit_fun;
1071   smx_process_exit_status_t exit_status = (process->context->iwannadie) ?
1072                                          SMX_EXIT_FAILURE : SMX_EXIT_SUCCESS;
1073   while (!xbt_dynar_is_empty(process->on_exit)) {
1074     exit_fun = xbt_dynar_pop_as(process->on_exit,s_smx_process_exit_fun_t);
1075     (exit_fun.fun)((void*)exit_status, exit_fun.arg);
1076   }
1077 }
1078
1079 void SIMIX_process_on_exit(smx_process_t process, int_f_pvoid_pvoid_t fun, void *data) {
1080   xbt_assert(process, "current process not found: are you in maestro context ?");
1081
1082   if (!process->on_exit) {
1083     process->on_exit = xbt_dynar_new(sizeof(s_smx_process_exit_fun_t), NULL);
1084   }
1085
1086   s_smx_process_exit_fun_t exit_fun = {fun, data};
1087
1088   xbt_dynar_push_as(process->on_exit,s_smx_process_exit_fun_t,exit_fun);
1089 }
1090
1091 /**
1092  * \brief Sets the auto-restart status of the process.
1093  * If set to 1, the process will be automatically restarted when its host
1094  * comes back.
1095  */
1096 void SIMIX_process_auto_restart_set(smx_process_t process, int auto_restart) {
1097   process->auto_restart = auto_restart;
1098 }
1099
1100 smx_process_t simcall_HANDLER_process_restart(smx_simcall_t simcall, smx_process_t process) {
1101   return SIMIX_process_restart(process, simcall->issuer);
1102 }
1103 /** @brief Restart a process, starting it again from the beginning. */
1104 smx_process_t SIMIX_process_restart(smx_process_t process, smx_process_t issuer) {
1105   XBT_DEBUG("Restarting process %s on %s", process->name, sg_host_get_name(process->host));
1106   //retrieve the arguments of the old process
1107   //FIXME: Factorize this with SIMIX_host_add_auto_restart_process ?
1108   s_smx_process_arg_t arg;
1109   arg.code = process->code;
1110   arg.hostname = sg_host_get_name(process->host);
1111   arg.kill_time = SIMIX_timer_get_date(process->kill_timer);
1112   arg.argc = process->argc;
1113   arg.data = process->data;
1114   int i;
1115   arg.argv = xbt_new(char*,process->argc + 1);
1116   for (i = 0; i < arg.argc; i++) {
1117     arg.argv[i] = xbt_strdup(process->argv[i]);
1118   }
1119   arg.argv[process->argc] = NULL;
1120   arg.properties = NULL;
1121   arg.auto_restart = process->auto_restart;
1122   //kill the old process
1123   SIMIX_process_kill(process,issuer);
1124   //start the new process
1125   smx_process_t new_process;
1126   if (simix_global->create_process_function) {
1127     new_process = simix_global->create_process_function(
1128                                           arg.argv[0],
1129                                           arg.code,
1130                                           arg.data,
1131                                           arg.hostname,
1132                                           arg.kill_time,
1133                                           arg.argc,
1134                                           arg.argv,
1135                                           arg.properties,
1136                                           arg.auto_restart,
1137                                           NULL);
1138   } else {
1139     new_process = simcall_process_create(
1140                            arg.argv[0],
1141                            arg.code,
1142                            arg.data,
1143                            arg.hostname,
1144                            arg.kill_time,
1145                            arg.argc,
1146                            arg.argv,
1147                            arg.properties,
1148                            arg.auto_restart);
1149
1150   }
1151   return new_process;
1152 }
1153
1154 void SIMIX_segment_index_set(smx_process_t proc, int index){
1155   proc->segment_index = index;
1156 }