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 #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 namespace simgrid {
 143 namespace simix {
 144
 145 void create_maestro(std::function<void()> code)
 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 = nullptr;
 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(code, maestro);
 164   }
 165
 166   maestro->simcall.issuer = maestro;
 167   simix_global->maestro_process = maestro;
 168 }
 169
 170 }
 171 }
 172
 173 /**
 174  * \brief Creates and runs the maestro process
 175  */
 176 void SIMIX_maestro_create(void (*code)(void*), void* data)
 177 {
 178   simgrid::simix::create_maestro(std::bind(code, data));
 179 }
 180
 181 /**
 182  * \brief Stops a process.
 183  *
 184  * Stops the process, execute all the registered on_exit functions,
 185  * register it to the list of the process to restart if needed
 186  * and stops its context.
 187  */
 188 void SIMIX_process_stop(smx_process_t arg) {
 189   /* execute the on_exit functions */
 190   SIMIX_process_on_exit_runall(arg);
 191   /* Add the process to the list of process to restart, only if the host is down */
 192   if (arg->auto_restart && arg->host->isOff()) {
 193     SIMIX_host_add_auto_restart_process(arg->host,arg->name,arg->code, arg->data,
 194                                         sg_host_get_name(arg->host),
 195                                         SIMIX_timer_get_date(arg->kill_timer),
 196                                         arg->argc,arg->argv,arg->properties,
 197                                         arg->auto_restart);
 198   }
 199   XBT_DEBUG("Process %s (%s) is dead",arg->name,sg_host_get_name(arg->host));
 200   /* stop the context */
 201   SIMIX_context_stop(arg->context);
 202 }
 203
 204 /**
 205  * \brief Same as SIMIX_process_create() but with only one argument (used by timers).
 206  * This function frees the argument.
 207  * \return the process created
 208  */
 209 smx_process_t SIMIX_process_create_from_wrapper(smx_process_arg_t args) {
 210
 211   smx_process_t process = simix_global->create_process_function(
 212                                         args->name,
 213                                         args->code,
 214                                         args->data,
 215                                         args->hostname,
 216                                         args->kill_time,
 217                                         args->argc,
 218                                         args->argv,
 219                                         args->properties,
 220                                         args->auto_restart,
 221                                         NULL);
 222   xbt_free(args);
 223   return process;
 224 }
 225
 226
 227 void* simcall_HANDLER_process_create(smx_simcall_t simcall,
 228                           const char *name,
 229                           xbt_main_func_t code,
 230                           void *data,
 231                           const char *hostname,
 232                           double kill_time,
 233                           int argc, char **argv,
 234                           xbt_dict_t properties,
 235                           int auto_restart){
 236   return (void*)SIMIX_process_create(name, code, data, hostname,
 237                        kill_time, argc, argv, properties, auto_restart,
 238                        simcall->issuer);
 239 }
 240
 241 static void kill_process(void* process)
 242 {
 243   simix_global->kill_process_function((smx_process_t) process);
 244 }
 245
 246 /**
 247  * \brief Internal function to create a process.
 248  *
 249  * This function actually creates the process.
 250  * It may be called when a SIMCALL_PROCESS_CREATE simcall occurs,
 251  * or directly for SIMIX internal purposes. The sure thing is that it's called from maestro context.
 252  *
 253  * \return the process created
 254  */
 255 smx_process_t SIMIX_process_create(
 256                           const char *name,
 257                           xbt_main_func_t code,
 258                           void *data,
 259                           const char *hostname,
 260                           double kill_time,
 261                           int argc, char **argv,
 262                           xbt_dict_t properties,
 263                           int auto_restart,
 264                           smx_process_t parent_process)
 265 {
 266   smx_process_t process = NULL;
 267   sg_host_t host = sg_host_by_name(hostname);
 268
 269   XBT_DEBUG("Start process %s on host '%s'", name, hostname);
 270
 271   if (host->isOff()) {
 272     int i;
 273     XBT_WARN("Cannot launch process '%s' on failed host '%s'", name,
 274           hostname);
 275     for (i = 0; i < argc; i++)
 276       xbt_free(argv[i]);
 277     xbt_free(argv);
 278   }
 279   else {
 280     process = xbt_new0(s_smx_process_t, 1);
 281
 282     xbt_assert(((code != NULL) && (host != NULL)), "Invalid parameters");
 283     /* Process data */
 284     process->pid = simix_process_maxpid++;
 285     process->name = xbt_strdup(name);
 286     process->host = host;
 287     process->data = data;
 288     process->comms = xbt_fifo_new();
 289     process->simcall.issuer = process;
 290     /* Initiliaze data segment to default value */
 291     SIMIX_segment_index_set(process, -1);
 292
 293      if (parent_process != NULL) {
 294        process->ppid = SIMIX_process_get_PID(parent_process);
 295        /* SMPI process have their own data segment and
 296           each other inherit from their father */
 297 #if HAVE_SMPI
 298        if(smpi_privatize_global_variables){
 299          if( parent_process->pid != 0){
 300            SIMIX_segment_index_set(process, parent_process->segment_index);
 301          } else {
 302            SIMIX_segment_index_set(process, process->pid - 1);
 303          }
 304        }
 305 #endif
 306      } else {
 307        process->ppid = -1;
 308      }
 309
 310     /* Process data for auto-restart */
 311     process->auto_restart = auto_restart;
 312     process->code = code;
 313     process->argc = argc;
 314     process->argv = argv;
 315
 316
 317     XBT_VERB("Create context %s", process->name);
 318     process->context = SIMIX_context_new(code, argc, argv, simix_global->cleanup_process_function, process);
 319
 320     process->running_ctx = (xbt_running_ctx_t*) xbt_malloc0(sizeof(xbt_running_ctx_t));
 321     XBT_RUNNING_CTX_INITIALIZE(process->running_ctx);
 322
 323     if(MC_is_active()){
 324       MC_ignore_heap(process->running_ctx, sizeof(*process->running_ctx));
 325     }
 326
 327     /* Add properties */
 328     process->properties = properties;
 329
 330     /* Add the process to it's host process list */
 331     xbt_swag_insert(process, sg_host_simix(host)->process_list);
 332
 333     XBT_DEBUG("Start context '%s'", process->name);
 334
 335     /* Now insert it in the global process list and in the process to run list */
 336     xbt_swag_insert(process, simix_global->process_list);
 337     XBT_DEBUG("Inserting %s(%s) in the to_run list", process->name, sg_host_get_name(host));
 338     xbt_dynar_push_as(simix_global->process_to_run, smx_process_t, process);
 339
 340     if (kill_time > SIMIX_get_clock() && simix_global->kill_process_function) {
 341       XBT_DEBUG("Process %s(%s) will be kill at time %f", process->name,
 342           sg_host_get_name(process->host), kill_time);
 343       process->kill_timer = SIMIX_timer_set(kill_time, kill_process, process);
 344     }
 345
 346     /* Tracing the process creation */
 347     TRACE_msg_process_create(process->name, process->pid, process->host);
 348   }
 349   return process;
 350 }
 351
 352 smx_process_t SIMIX_process_attach(
 353   const char* name,
 354   void *data,
 355   const char* hostname,
 356   xbt_dict_t properties,
 357   smx_process_t parent_process)
 358 {
 359   // This is mostly a copy/paste from SIMIX_process_new(),
 360   // it'd be nice to share some code between those two functions.
 361
 362   sg_host_t host = sg_host_by_name(hostname);
 363   XBT_DEBUG("Attach process %s on host '%s'", name, hostname);
 364
 365   if (host->isOff()) {
 366     XBT_WARN("Cannot launch process '%s' on failed host '%s'",
 367       name, hostname);
 368     return nullptr;
 369   }
 370
 371   smx_process_t process = xbt_new0(s_smx_process_t, 1);
 372   /* Process data */
 373   process->pid = simix_process_maxpid++;
 374   process->name = xbt_strdup(name);
 375   process->host = host;
 376   process->data = data;
 377   process->comms = xbt_fifo_new();
 378   process->simcall.issuer = process;
 379   process->ppid = -1;
 380   /* Initiliaze data segment to default value */
 381   SIMIX_segment_index_set(process, -1);
 382   if (parent_process != NULL) {
 383     process->ppid = SIMIX_process_get_PID(parent_process);
 384    /* SMPI process have their own data segment and
 385       each other inherit from their father */
 386   #if HAVE_SMPI
 387     if(smpi_privatize_global_variables){
 388       if(parent_process->pid != 0){
 389         SIMIX_segment_index_set(process, parent_process->segment_index);
 390       } else {
 391         SIMIX_segment_index_set(process, process->pid - 1);
 392       }
 393     }
 394   #endif
 395   }
 396
 397   /* Process data for auto-restart */
 398   process->auto_restart = false;
 399   process->code = nullptr;
 400   process->argc = 0;
 401   process->argv = nullptr;
 402
 403   XBT_VERB("Create context %s", process->name);
 404   if (!simix_global)
 405     xbt_die("simix is not initialized, please call MSG_init first");
 406   process->context = simix_global->context_factory->attach(
 407     simix_global->cleanup_process_function, process);
 408
 409   process->running_ctx = (xbt_running_ctx_t*) xbt_malloc0(sizeof(xbt_running_ctx_t));
 410   XBT_RUNNING_CTX_INITIALIZE(process->running_ctx);
 411
 412   if(MC_is_active()){
 413     MC_ignore_heap(process->running_ctx, sizeof(*process->running_ctx));
 414   }
 415
 416   /* Add properties */
 417   process->properties = properties;
 418
 419   /* Add the process to it's host process list */
 420   xbt_swag_insert(process, sg_host_simix(host)->process_list);
 421
 422   /* Now insert it in the global process list and in the process to run list */
 423   xbt_swag_insert(process, simix_global->process_list);
 424   XBT_DEBUG("Inserting %s(%s) in the to_run list", process->name, sg_host_get_name(host));
 425   xbt_dynar_push_as(simix_global->process_to_run, smx_process_t, process);
 426
 427   /* Tracing the process creation */
 428   TRACE_msg_process_create(process->name, process->pid, process->host);
 429
 430   auto context = dynamic_cast<simgrid::simix::AttachContext*>(process->context);
 431   if (!context)
 432     xbt_die("Not a suitable context");
 433
 434   context->attach_start();
 435   return process;
 436 }
 437
 438 void SIMIX_process_detach(void)
 439 {
 440   auto context = dynamic_cast<simgrid::simix::AttachContext*>(SIMIX_context_self());
 441   if (!context)
 442     xbt_die("Not a suitable context");
 443
 444   simix_global->cleanup_process_function(context->process());
 445
 446   // Let maestro ignore we are still alive:
 447   // xbt_swag_remove(context->process(), simix_global->process_list);
 448
 449   // TODDO, Remove from proces list:
 450   //   xbt_swag_remove(process, sg_host_simix(host)->process_list);
 451
 452   context->attach_stop();
 453   // delete context;
 454 }
 455
 456 /**
 457  * \brief Executes the processes from simix_global->process_to_run.
 458  *
 459  * The processes of simix_global->process_to_run are run (in parallel if
 460  * possible).  On exit, simix_global->process_to_run is empty, and
 461  * simix_global->process_that_ran contains the list of processes that just ran.
 462  * The two lists are swapped so, be careful when using them before and after a
 463  * call to this function.
 464  */
 465 void SIMIX_process_runall(void)
 466 {
 467   SIMIX_context_runall();
 468
 469   xbt_dynar_t tmp = simix_global->process_that_ran;
 470   simix_global->process_that_ran = simix_global->process_to_run;
 471   simix_global->process_to_run = tmp;
 472   xbt_dynar_reset(simix_global->process_to_run);
 473 }
 474
 475 void simcall_HANDLER_process_kill(smx_simcall_t simcall, smx_process_t process) {
 476   SIMIX_process_kill(process, simcall->issuer);
 477 }
 478 /**
 479  * \brief Internal function to kill a SIMIX process.
 480  *
 481  * This function may be called when a SIMCALL_PROCESS_KILL simcall occurs,
 482  * or directly for SIMIX internal purposes.
 483  *
 484  * \param process poor victim
 485  * \param issuer the process which has sent the PROCESS_KILL. Important to not schedule twice the same process.
 486  */
 487 void SIMIX_process_kill(smx_process_t process, smx_process_t issuer) {
 488
 489   XBT_DEBUG("Killing process %s on %s", process->name, sg_host_get_name(process->host));
 490
 491   process->context->iwannadie = 1;
 492   process->blocked = 0;
 493   process->suspended = 0;
 494   process->doexception = 0;
 495
 496   /* destroy the blocking synchro if any */
 497   if (process->waiting_synchro) {
 498
 499     switch (process->waiting_synchro->type) {
 500
 501     case SIMIX_SYNC_EXECUTE:
 502     case SIMIX_SYNC_PARALLEL_EXECUTE:
 503       SIMIX_execution_destroy(process->waiting_synchro);
 504       break;
 505
 506     case SIMIX_SYNC_COMMUNICATE:
 507       xbt_fifo_remove(process->comms, process->waiting_synchro);
 508       SIMIX_comm_cancel(process->waiting_synchro);
 509       xbt_fifo_remove(process->waiting_synchro->simcalls, &process->simcall);
 510       SIMIX_comm_destroy(process->waiting_synchro);
 511       break;
 512
 513     case SIMIX_SYNC_SLEEP:
 514       SIMIX_process_sleep_destroy(process->waiting_synchro);
 515       break;
 516
 517     case SIMIX_SYNC_JOIN:
 518       SIMIX_process_sleep_destroy(process->waiting_synchro);
 519       break;
 520
 521     case SIMIX_SYNC_SYNCHRO:
 522       SIMIX_synchro_stop_waiting(process, &process->simcall);
 523       SIMIX_synchro_destroy(process->waiting_synchro);
 524       break;
 525
 526     case SIMIX_SYNC_IO:
 527       SIMIX_io_destroy(process->waiting_synchro);
 528       break;
 529
 530     }
 531
 532     process->waiting_synchro = NULL;
 533   }
 534   if(!xbt_dynar_member(simix_global->process_to_run, &(process)) && process != issuer) {
 535     XBT_DEBUG("Inserting %s in the to_run list", process->name);
 536     xbt_dynar_push_as(simix_global->process_to_run, smx_process_t, process);
 537   }
 538
 539 }
 540
 541 /** @brief Ask another process to raise the given exception
 542  *
 543  * @param cat category of exception
 544  * @param value value associated to the exception
 545  * @param msg string information associated to the exception
 546  */
 547 void SIMIX_process_throw(smx_process_t process, xbt_errcat_t cat, int value, const char *msg) {
 548   SMX_EXCEPTION(process, cat, value, msg);
 549
 550   if (process->suspended)
 551     SIMIX_process_resume(process,SIMIX_process_self());
 552
 553   /* cancel the blocking synchro if any */
 554   if (process->waiting_synchro) {
 555
 556     switch (process->waiting_synchro->type) {
 557
 558     case SIMIX_SYNC_EXECUTE:
 559     case SIMIX_SYNC_PARALLEL_EXECUTE:
 560       SIMIX_execution_cancel(process->waiting_synchro);
 561       break;
 562
 563     case SIMIX_SYNC_COMMUNICATE:
 564       xbt_fifo_remove(process->comms, process->waiting_synchro);
 565       SIMIX_comm_cancel(process->waiting_synchro);
 566       break;
 567
 568     case SIMIX_SYNC_SLEEP:
 569     case SIMIX_SYNC_JOIN:
 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       break;
 576
 577     case SIMIX_SYNC_SYNCHRO:
 578       SIMIX_synchro_stop_waiting(process, &process->simcall);
 579       break;
 580
 581     case SIMIX_SYNC_IO:
 582       SIMIX_io_destroy(process->waiting_synchro);
 583       break;
 584
 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,
 657      suspend its synchronization. */
 658   if (process != issuer) {
 659
 660     if (process->waiting_synchro) {
 661
 662       switch (process->waiting_synchro->type) {
 663
 664         case SIMIX_SYNC_EXECUTE:
 665         case SIMIX_SYNC_PARALLEL_EXECUTE:
 666           SIMIX_execution_suspend(process->waiting_synchro);
 667           break;
 668
 669         case SIMIX_SYNC_COMMUNICATE:
 670           SIMIX_comm_suspend(process->waiting_synchro);
 671           break;
 672
 673         case SIMIX_SYNC_SLEEP:
 674           SIMIX_process_sleep_suspend(process->waiting_synchro);
 675           break;
 676
 677         case SIMIX_SYNC_SYNCHRO:
 678           /* Suspension is delayed to when the process is rescheduled. */
 679           break;
 680
 681         default:
 682           xbt_die("Internal error in SIMIX_process_suspend: unexpected synchronization type %d",
 683               (int)process->waiting_synchro->type);
 684       }
 685       return NULL;
 686     } else {
 687       /* Suspension is delayed to when the process is rescheduled. */
 688       return NULL;
 689     }
 690   } else {
 691     /* FIXME: computation size is zero. Is it okay that bound is zero ? */
 692     return SIMIX_execution_start(process, "suspend", 0.0, 1.0, 0.0, 0);
 693   }
 694 }
 695
 696 void simcall_HANDLER_process_resume(smx_simcall_t simcall, smx_process_t process){
 697   SIMIX_process_resume(process, simcall->issuer);
 698 }
 699
 700 void SIMIX_process_resume(smx_process_t process, smx_process_t issuer)
 701 {
 702   XBT_IN("process = %p, issuer = %p", process, issuer);
 703
 704   if(process->context->iwannadie) {
 705     XBT_VERB("Ignoring request to suspend a process that is currently dying.");
 706     return;
 707   }
 708
 709   if(!process->suspended) return;
 710   process->suspended = 0;
 711
 712   /* If we are resuming another process, resume the synchronization it was waiting for
 713      if any. Otherwise add it to the list of process to run in the next round. */
 714   if (process != issuer) {
 715
 716     if (process->waiting_synchro) {
 717
 718       switch (process->waiting_synchro->type) {
 719
 720         case SIMIX_SYNC_EXECUTE:
 721         case SIMIX_SYNC_PARALLEL_EXECUTE:
 722           SIMIX_execution_resume(process->waiting_synchro);
 723           break;
 724
 725         case SIMIX_SYNC_COMMUNICATE:
 726           SIMIX_comm_resume(process->waiting_synchro);
 727           break;
 728
 729         case SIMIX_SYNC_SLEEP:
 730           SIMIX_process_sleep_resume(process->waiting_synchro);
 731           break;
 732
 733         case SIMIX_SYNC_SYNCHRO:
 734           /* I cannot resume it now. This is delayed to when the process is rescheduled at
 735            * the end of the synchro. */
 736           break;
 737
 738         default:
 739           xbt_die("Internal error in SIMIX_process_resume: unexpected synchronization type %d",
 740               (int)process->waiting_synchro->type);
 741       }
 742     }
 743   } else XBT_WARN("Strange. Process %p is trying to resume himself.", issuer);
 744
 745   XBT_OUT();
 746 }
 747
 748 int SIMIX_process_get_maxpid(void) {
 749   return simix_process_maxpid;
 750 }
 751
 752 int SIMIX_process_count(void)
 753 {
 754   return xbt_swag_size(simix_global->process_list);
 755 }
 756
 757 int SIMIX_process_get_PID(smx_process_t self){
 758   if (self == NULL)
 759     return 0;
 760   else
 761     return self->pid;
 762 }
 763
 764 int SIMIX_process_get_PPID(smx_process_t self){
 765   if (self == NULL)
 766     return 0;
 767   else
 768     return self->ppid;
 769 }
 770
 771 void* SIMIX_process_self_get_data()
 772 {
 773   smx_process_t self = SIMIX_process_self();
 774
 775   if (!self) {
 776     return NULL;
 777   }
 778   return SIMIX_process_get_data(self);
 779 }
 780
 781 void SIMIX_process_self_set_data(void *data)
 782 {
 783   smx_process_t self = SIMIX_process_self();
 784
 785   SIMIX_process_set_data(self, data);
 786 }
 787
 788 void* SIMIX_process_get_data(smx_process_t process)
 789 {
 790   return process->data;
 791 }
 792
 793 void SIMIX_process_set_data(smx_process_t process, void *data)
 794 {
 795   process->data = data;
 796 }
 797
 798 sg_host_t SIMIX_process_get_host(smx_process_t process)
 799 {
 800   return process->host;
 801 }
 802
 803 xbt_main_func_t SIMIX_process_get_code(void){
 804   return SIMIX_process_self()->code;
 805 }
 806
 807 /* needs to be public and without simcall because it is called
 808    by exceptions and logging events */
 809 const char* SIMIX_process_self_get_name(void) {
 810
 811   smx_process_t process = SIMIX_process_self();
 812   if (process == NULL || process == simix_global->maestro_process)
 813     return "maestro";
 814
 815   return SIMIX_process_get_name(process);
 816 }
 817
 818 const char* SIMIX_process_get_name(smx_process_t process)
 819 {
 820   return process->name;
 821 }
 822
 823 smx_process_t SIMIX_process_get_by_name(const char* name)
 824 {
 825   smx_process_t proc;
 826
 827   xbt_swag_foreach(proc, simix_global->process_list)
 828   {
 829     if(!strcmp(name, proc->name))
 830       return proc;
 831   }
 832   return NULL;
 833 }
 834
 835 int SIMIX_process_is_suspended(smx_process_t process)
 836 {
 837   return process->suspended;
 838 }
 839
 840 xbt_dict_t SIMIX_process_get_properties(smx_process_t process)
 841 {
 842   return process->properties;
 843 }
 844
 845 void simcall_HANDLER_process_join(smx_simcall_t simcall, smx_process_t process, double timeout)
 846 {
 847   smx_synchro_t sync = SIMIX_process_join(simcall->issuer, process, timeout);
 848   xbt_fifo_push(sync->simcalls, simcall);
 849   simcall->issuer->waiting_synchro = sync;
 850 }
 851
 852 static int SIMIX_process_join_finish(smx_process_exit_status_t status, smx_synchro_t sync){
 853   if (sync->sleep.surf_sleep) {
 854     sync->sleep.surf_sleep->cancel();
 855
 856     smx_simcall_t simcall;
 857     while ((simcall = (smx_simcall_t) xbt_fifo_shift(sync->simcalls))) {
 858       simcall_process_sleep__set__result(simcall, SIMIX_DONE);
 859       simcall->issuer->waiting_synchro = NULL;
 860       if (simcall->issuer->suspended) {
 861         XBT_DEBUG("Wait! This process is suspended and can't wake up now.");
 862         simcall->issuer->suspended = 0;
 863         simcall_HANDLER_process_suspend(simcall, simcall->issuer);
 864       } else {
 865         SIMIX_simcall_answer(simcall);
 866       }
 867     }
 868     sync->sleep.surf_sleep->unref();
 869     sync->sleep.surf_sleep = NULL;
 870   }
 871   xbt_mallocator_release(simix_global->synchro_mallocator, sync);
 872   return 0;
 873 }
 874
 875 smx_synchro_t SIMIX_process_join(smx_process_t issuer, smx_process_t process, double timeout)
 876 {
 877   smx_synchro_t res = SIMIX_process_sleep(issuer, timeout);
 878   res->type = SIMIX_SYNC_JOIN;
 879   SIMIX_process_on_exit(process, (int_f_pvoid_pvoid_t)SIMIX_process_join_finish, res);
 880   return res;
 881 }
 882
 883 void simcall_HANDLER_process_sleep(smx_simcall_t simcall, double duration)
 884 {
 885   if (MC_is_active() || MC_record_replay_is_active()) {
 886     MC_process_clock_add(simcall->issuer, duration);
 887     simcall_process_sleep__set__result(simcall, SIMIX_DONE);
 888     SIMIX_simcall_answer(simcall);
 889     return;
 890   }
 891   smx_synchro_t sync = SIMIX_process_sleep(simcall->issuer, duration);
 892   xbt_fifo_push(sync->simcalls, simcall);
 893   simcall->issuer->waiting_synchro = sync;
 894 }
 895
 896 smx_synchro_t SIMIX_process_sleep(smx_process_t process, double duration)
 897 {
 898   sg_host_t host = process->host;
 899
 900   /* check if the host is active */
 901   if (host->isOff()) {
 902     THROWF(host_error, 0, "Host %s failed, you cannot call this function",
 903            sg_host_get_name(host));
 904   }
 905
 906   smx_synchro_t synchro = (smx_synchro_t) xbt_mallocator_get(simix_global->synchro_mallocator);
 907   synchro->type = SIMIX_SYNC_SLEEP;
 908   synchro->name = NULL;
 909   synchro->category = NULL;
 910
 911   synchro->sleep.host = host;
 912   synchro->sleep.surf_sleep = surf_host_sleep(host, duration);
 913
 914   synchro->sleep.surf_sleep->setData(synchro);
 915   XBT_DEBUG("Create sleep synchronization %p", synchro);
 916
 917   return synchro;
 918 }
 919
 920 void SIMIX_post_process_sleep(smx_synchro_t synchro)
 921 {
 922   smx_simcall_t simcall;
 923   e_smx_state_t state;
 924   xbt_assert(synchro->type == SIMIX_SYNC_SLEEP || synchro->type == SIMIX_SYNC_JOIN);
 925
 926   while ((simcall = (smx_simcall_t) xbt_fifo_shift(synchro->simcalls))) {
 927
 928     switch (synchro->sleep.surf_sleep->getState()){
 929       case simgrid::surf::Action::State::failed:
 930         simcall->issuer->context->iwannadie = 1;
 931         //SMX_EXCEPTION(simcall->issuer, host_error, 0, "Host failed");
 932         state = SIMIX_SRC_HOST_FAILURE;
 933         break;
 934
 935       case simgrid::surf::Action::State::done:
 936         state = SIMIX_DONE;
 937         break;
 938
 939       default:
 940         THROW_IMPOSSIBLE;
 941         break;
 942     }
 943     if (simcall->issuer->host->isOff()) {
 944       simcall->issuer->context->iwannadie = 1;
 945     }
 946     simcall_process_sleep__set__result(simcall, state);
 947     simcall->issuer->waiting_synchro = NULL;
 948     if (simcall->issuer->suspended) {
 949       XBT_DEBUG("Wait! This process is suspended and can't wake up now.");
 950       simcall->issuer->suspended = 0;
 951       simcall_HANDLER_process_suspend(simcall, simcall->issuer);
 952     } else {
 953       SIMIX_simcall_answer(simcall);
 954     }
 955   }
 956
 957   SIMIX_process_sleep_destroy(synchro);
 958 }
 959
 960 void SIMIX_process_sleep_destroy(smx_synchro_t synchro)
 961 {
 962   XBT_DEBUG("Destroy synchro %p", synchro);
 963   xbt_assert(synchro->type == SIMIX_SYNC_SLEEP || synchro->type == SIMIX_SYNC_JOIN);
 964
 965   if (synchro->sleep.surf_sleep) {
 966     synchro->sleep.surf_sleep->unref();
 967     synchro->sleep.surf_sleep = NULL;
 968   }
 969   if (synchro->type == SIMIX_SYNC_SLEEP)
 970     xbt_mallocator_release(simix_global->synchro_mallocator, synchro);
 971 }
 972
 973 void SIMIX_process_sleep_suspend(smx_synchro_t synchro)
 974 {
 975   xbt_assert(synchro->type == SIMIX_SYNC_SLEEP);
 976   synchro->sleep.surf_sleep->suspend();
 977 }
 978
 979 void SIMIX_process_sleep_resume(smx_synchro_t synchro)
 980 {
 981   XBT_DEBUG("Synchro state is %d on process_sleep_resume.", synchro->state);
 982   xbt_assert(synchro->type == SIMIX_SYNC_SLEEP);
 983   synchro->sleep.surf_sleep->resume();
 984 }
 985
 986 /**
 987  * \brief Calling this function makes the process to yield.
 988  *
 989  * Only the current process can call this function, giving back the control to
 990  * maestro.
 991  *
 992  * \param self the current process
 993  */
 994 void SIMIX_process_yield(smx_process_t self)
 995 {
 996   XBT_DEBUG("Yield process '%s'", self->name);
 997
 998   /* Go into sleep and return control to maestro */
 999   SIMIX_context_suspend(self->context);
1000
1001   /* Ok, maestro returned control to us */
1002   XBT_DEBUG("Control returned to me: '%s'", self->name);
1003
1004   if (self->new_host) {
1005     SIMIX_process_change_host(self, self->new_host);
1006     self->new_host = NULL;
1007   }
1008
1009   if (self->context->iwannadie){
1010     XBT_DEBUG("I wanna die!");
1011     SIMIX_process_stop(self);
1012   }
1013
1014   if (self->suspended) {
1015     XBT_DEBUG("Hey! I'm suspended.");
1016     xbt_assert(!self->doexception, "Gasp! This exception may be lost by subsequent calls.");
1017     self->suspended = 0;
1018     SIMIX_process_suspend(self, self);
1019   }
1020
1021   if (self->doexception) {
1022     XBT_DEBUG("Wait, maestro left me an exception");
1023     self->doexception = 0;
1024     RETHROW;
1025   }
1026
1027   if(SMPI_switch_data_segment && self->segment_index != -1){
1028     SMPI_switch_data_segment(self->segment_index);
1029   }
1030 }
1031
1032 /* callback: context fetching */
1033 xbt_running_ctx_t *SIMIX_process_get_running_context(void)
1034 {
1035   smx_process_t process = SIMIX_process_self();
1036   if (process)
1037     return process->running_ctx;
1038   else
1039     return nullptr;
1040 }
1041
1042 /* callback: termination */
1043 void SIMIX_process_exception_terminate(xbt_ex_t * e)
1044 {
1045   xbt_ex_display(e);
1046   xbt_abort();
1047 }
1048
1049 smx_context_t SIMIX_process_get_context(smx_process_t p) {
1050   return p->context;
1051 }
1052
1053 void SIMIX_process_set_context(smx_process_t p,smx_context_t c) {
1054   p->context = c;
1055 }
1056
1057 /**
1058  * \brief Returns the list of processes to run.
1059  */
1060 xbt_dynar_t SIMIX_process_get_runnable(void)
1061 {
1062   return simix_global->process_to_run;
1063 }
1064
1065 /**
1066  * \brief Returns the process from PID.
1067  */
1068 smx_process_t SIMIX_process_from_PID(int PID)
1069 {
1070   smx_process_t proc;
1071   xbt_swag_foreach(proc, simix_global->process_list) {
1072    if (proc->pid == (unsigned long) PID)
1073     return proc;
1074   }
1075   return NULL;
1076 }
1077
1078 /** @brief returns a dynar containg all currently existing processes */
1079 xbt_dynar_t SIMIX_processes_as_dynar(void) {
1080   smx_process_t proc;
1081   xbt_dynar_t res = xbt_dynar_new(sizeof(smx_process_t),NULL);
1082   xbt_swag_foreach(proc, simix_global->process_list) {
1083     xbt_dynar_push(res,&proc);
1084   }
1085   return res;
1086 }
1087
1088
1089 void SIMIX_process_on_exit_runall(smx_process_t process) {
1090   s_smx_process_exit_fun_t exit_fun;
1091   smx_process_exit_status_t exit_status = (process->context->iwannadie) ?
1092                                          SMX_EXIT_FAILURE : SMX_EXIT_SUCCESS;
1093   while (!xbt_dynar_is_empty(process->on_exit)) {
1094     exit_fun = xbt_dynar_pop_as(process->on_exit,s_smx_process_exit_fun_t);
1095     (exit_fun.fun)((void*)exit_status, exit_fun.arg);
1096   }
1097 }
1098
1099 void SIMIX_process_on_exit(smx_process_t process, int_f_pvoid_pvoid_t fun, void *data) {
1100   xbt_assert(process, "current process not found: are you in maestro context ?");
1101
1102   if (!process->on_exit) {
1103     process->on_exit = xbt_dynar_new(sizeof(s_smx_process_exit_fun_t), NULL);
1104   }
1105
1106   s_smx_process_exit_fun_t exit_fun = {fun, data};
1107
1108   xbt_dynar_push_as(process->on_exit,s_smx_process_exit_fun_t,exit_fun);
1109 }
1110
1111 /**
1112  * \brief Sets the auto-restart status of the process.
1113  * If set to 1, the process will be automatically restarted when its host
1114  * comes back.
1115  */
1116 void SIMIX_process_auto_restart_set(smx_process_t process, int auto_restart) {
1117   process->auto_restart = auto_restart;
1118 }
1119
1120 smx_process_t simcall_HANDLER_process_restart(smx_simcall_t simcall, smx_process_t process) {
1121   return SIMIX_process_restart(process, simcall->issuer);
1122 }
1123 /** @brief Restart a process, starting it again from the beginning. */
1124 smx_process_t SIMIX_process_restart(smx_process_t process, smx_process_t issuer) {
1125   XBT_DEBUG("Restarting process %s on %s", process->name, sg_host_get_name(process->host));
1126   //retrieve the arguments of the old process
1127   //FIXME: Factorize this with SIMIX_host_add_auto_restart_process ?
1128   s_smx_process_arg_t arg;
1129   arg.code = process->code;
1130   arg.hostname = sg_host_get_name(process->host);
1131   arg.kill_time = SIMIX_timer_get_date(process->kill_timer);
1132   arg.argc = process->argc;
1133   arg.data = process->data;
1134   int i;
1135   arg.argv = xbt_new(char*,process->argc + 1);
1136   for (i = 0; i < arg.argc; i++) {
1137     arg.argv[i] = xbt_strdup(process->argv[i]);
1138   }
1139   arg.argv[process->argc] = NULL;
1140   arg.properties = NULL;
1141   arg.auto_restart = process->auto_restart;
1142   //kill the old process
1143   SIMIX_process_kill(process,issuer);
1144   //start the new process
1145   smx_process_t new_process;
1146   if (simix_global->create_process_function) {
1147     new_process = simix_global->create_process_function(
1148                                           arg.argv[0],
1149                                           arg.code,
1150                                           arg.data,
1151                                           arg.hostname,
1152                                           arg.kill_time,
1153                                           arg.argc,
1154                                           arg.argv,
1155                                           arg.properties,
1156                                           arg.auto_restart,
1157                                           NULL);
1158   } else {
1159     new_process = simcall_process_create(
1160                            arg.argv[0],
1161                            arg.code,
1162                            arg.data,
1163                            arg.hostname,
1164                            arg.kill_time,
1165                            arg.argc,
1166                            arg.argv,
1167                            arg.properties,
1168                            arg.auto_restart);
1169
1170   }
1171   return new_process;
1172 }
1173
1174 void SIMIX_segment_index_set(smx_process_t proc, int index){
1175   proc->segment_index = index;
1176 }