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/msg/msg_private.h"
  16
  17 #include "src/simix/SynchroSleep.hpp"
  18 #include "src/simix/SynchroRaw.hpp"
  19 #include "src/simix/SynchroIo.hpp"
  20
  21 #ifdef HAVE_SMPI
  22 #include "src/smpi/private.h"
  23 #endif
  24
  25 XBT_LOG_NEW_DEFAULT_SUBCATEGORY(simix_process, simix, "Logging specific to SIMIX (process)");
  26
  27 unsigned long simix_process_maxpid = 0;
  28
  29 /**
  30  * \brief Returns the current agent.
  31  *
  32  * This functions returns the currently running SIMIX process.
  33  *
  34  * \return The SIMIX process
  35  */
  36 smx_process_t SIMIX_process_self(void)
  37 {
  38   smx_context_t self_context = SIMIX_context_self();
  39
  40   return self_context ? self_context->process() : NULL;
  41 }
  42
  43 /**
  44  * \brief Returns whether a process has pending asynchronous communications.
  45  * \return true if there are asynchronous communications in this process
  46  */
  47 int SIMIX_process_has_pending_comms(smx_process_t process) {
  48
  49   return xbt_fifo_size(process->comms) > 0;
  50 }
  51
  52 /**
  53  * \brief Moves a process to the list of processes to destroy.
  54  */
  55 void SIMIX_process_cleanup(smx_process_t process)
  56 {
  57   XBT_DEBUG("Cleanup process %s (%p), waiting synchro %p",
  58       process->name, process, process->waiting_synchro);
  59
  60   SIMIX_process_on_exit_runall(process);
  61
  62   /* Unregister from the kill timer if any */
  63   if (process->kill_timer != NULL)
  64       SIMIX_timer_remove(process->kill_timer);
  65
  66   xbt_os_mutex_acquire(simix_global->mutex);
  67
  68   /* cancel non-blocking communications */
  69   smx_synchro_t synchro;
  70   while ((synchro = (smx_synchro_t) xbt_fifo_pop(process->comms))) {
  71     simgrid::simix::Comm *comm = static_cast<simgrid::simix::Comm*>(synchro);
  72
  73     /* make sure no one will finish the comm after this process is destroyed,
  74      * because src_proc or dst_proc would be an invalid pointer */
  75     comm->cancel();
  76
  77     if (comm->src_proc == process) {
  78       XBT_DEBUG("Found an unfinished send comm %p (detached = %d), state %d, src = %p, dst = %p",
  79           comm, comm->detached, (int)comm->state, comm->src_proc, comm->dst_proc);
  80       comm->src_proc = NULL;
  81
  82       /* I'm not supposed to destroy a detached comm from the sender side, */
  83       if (comm->detached)
  84         XBT_DEBUG("Don't destroy it since it's a detached comm and I'm the sender");
  85       else
  86         comm->unref();
  87
  88     }
  89     else if (comm->dst_proc == process){
  90       XBT_DEBUG("Found an unfinished recv comm %p, state %d, src = %p, dst = %p",
  91           comm, (int)comm->state, comm->src_proc, comm->dst_proc);
  92       comm->dst_proc = NULL;
  93
  94       if (comm->detached && comm->src_proc != NULL) {
  95         /* the comm will be freed right now, remove it from the sender */
  96         xbt_fifo_remove(comm->src_proc->comms, comm);
  97       }
  98
  99       comm->unref();
 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     delete 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     delete 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 = new simgrid::simix::Process();
 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(std::function<void()>(), 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   arg->context->stop();
 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   delete 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 = new simgrid::simix::Process();
 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(
 319       simgrid::simix::wrap_main(code, argc, argv),
 320       simix_global->cleanup_process_function, process);
 321
 322     process->running_ctx = (xbt_running_ctx_t*) xbt_malloc0(sizeof(xbt_running_ctx_t));
 323     XBT_RUNNING_CTX_INITIALIZE(process->running_ctx);
 324
 325     if(MC_is_active()){
 326       MC_ignore_heap(process->running_ctx, sizeof(*process->running_ctx));
 327     }
 328
 329     /* Add properties */
 330     process->properties = properties;
 331
 332     /* Add the process to it's host process list */
 333     xbt_swag_insert(process, sg_host_simix(host)->process_list);
 334
 335     XBT_DEBUG("Start context '%s'", process->name);
 336
 337     /* Now insert it in the global process list and in the process to run list */
 338     xbt_swag_insert(process, simix_global->process_list);
 339     XBT_DEBUG("Inserting %s(%s) in the to_run list", process->name, sg_host_get_name(host));
 340     xbt_dynar_push_as(simix_global->process_to_run, smx_process_t, process);
 341
 342     if (kill_time > SIMIX_get_clock() && simix_global->kill_process_function) {
 343       XBT_DEBUG("Process %s(%s) will be kill at time %f", process->name,
 344           sg_host_get_name(process->host), kill_time);
 345       process->kill_timer = SIMIX_timer_set(kill_time, kill_process, process);
 346     }
 347
 348     /* Tracing the process creation */
 349     TRACE_msg_process_create(process->name, process->pid, process->host);
 350   }
 351   return process;
 352 }
 353
 354 smx_process_t SIMIX_process_attach(
 355   const char* name,
 356   void *data,
 357   const char* hostname,
 358   xbt_dict_t properties,
 359   smx_process_t parent_process)
 360 {
 361   // This is mostly a copy/paste from SIMIX_process_new(),
 362   // it'd be nice to share some code between those two functions.
 363
 364   sg_host_t host = sg_host_by_name(hostname);
 365   XBT_DEBUG("Attach process %s on host '%s'", name, hostname);
 366
 367   if (host->isOff()) {
 368     XBT_WARN("Cannot launch process '%s' on failed host '%s'",
 369       name, hostname);
 370     return nullptr;
 371   }
 372
 373   smx_process_t process = new simgrid::simix::Process();
 374   /* Process data */
 375   process->pid = simix_process_maxpid++;
 376   process->name = xbt_strdup(name);
 377   process->host = host;
 378   process->data = data;
 379   process->comms = xbt_fifo_new();
 380   process->simcall.issuer = process;
 381   process->ppid = -1;
 382   /* Initiliaze data segment to default value */
 383   SIMIX_segment_index_set(process, -1);
 384   if (parent_process != NULL) {
 385     process->ppid = SIMIX_process_get_PID(parent_process);
 386    /* SMPI process have their own data segment and
 387       each other inherit from their father */
 388   #if HAVE_SMPI
 389     if(smpi_privatize_global_variables){
 390       if(parent_process->pid != 0){
 391         SIMIX_segment_index_set(process, parent_process->segment_index);
 392       } else {
 393         SIMIX_segment_index_set(process, process->pid - 1);
 394       }
 395     }
 396   #endif
 397   }
 398
 399   /* Process data for auto-restart */
 400   process->auto_restart = false;
 401   process->code = nullptr;
 402   process->argc = 0;
 403   process->argv = nullptr;
 404
 405   XBT_VERB("Create context %s", process->name);
 406   if (!simix_global)
 407     xbt_die("simix is not initialized, please call MSG_init first");
 408   process->context = simix_global->context_factory->attach(
 409     simix_global->cleanup_process_function, process);
 410
 411   process->running_ctx = (xbt_running_ctx_t*) xbt_malloc0(sizeof(xbt_running_ctx_t));
 412   XBT_RUNNING_CTX_INITIALIZE(process->running_ctx);
 413
 414   if(MC_is_active()){
 415     MC_ignore_heap(process->running_ctx, sizeof(*process->running_ctx));
 416   }
 417
 418   /* Add properties */
 419   process->properties = properties;
 420
 421   /* Add the process to it's host process list */
 422   xbt_swag_insert(process, sg_host_simix(host)->process_list);
 423
 424   /* Now insert it in the global process list and in the process to run list */
 425   xbt_swag_insert(process, simix_global->process_list);
 426   XBT_DEBUG("Inserting %s(%s) in the to_run list", process->name, sg_host_get_name(host));
 427   xbt_dynar_push_as(simix_global->process_to_run, smx_process_t, process);
 428
 429   /* Tracing the process creation */
 430   TRACE_msg_process_create(process->name, process->pid, process->host);
 431
 432   auto context = dynamic_cast<simgrid::simix::AttachContext*>(process->context);
 433   if (!context)
 434     xbt_die("Not a suitable context");
 435
 436   context->attach_start();
 437   return process;
 438 }
 439
 440 void SIMIX_process_detach(void)
 441 {
 442   auto context = dynamic_cast<simgrid::simix::AttachContext*>(SIMIX_context_self());
 443   if (!context)
 444     xbt_die("Not a suitable context");
 445
 446   simix_global->cleanup_process_function(context->process());
 447
 448   // Let maestro ignore we are still alive:
 449   // xbt_swag_remove(context->process(), simix_global->process_list);
 450
 451   // TODDO, Remove from proces list:
 452   //   xbt_swag_remove(process, sg_host_simix(host)->process_list);
 453
 454   context->attach_stop();
 455   // delete context;
 456 }
 457
 458 /**
 459  * \brief Executes the processes from simix_global->process_to_run.
 460  *
 461  * The processes of simix_global->process_to_run are run (in parallel if
 462  * possible).  On exit, simix_global->process_to_run is empty, and
 463  * simix_global->process_that_ran contains the list of processes that just ran.
 464  * The two lists are swapped so, be careful when using them before and after a
 465  * call to this function.
 466  */
 467 void SIMIX_process_runall(void)
 468 {
 469   SIMIX_context_runall();
 470
 471   xbt_dynar_t tmp = simix_global->process_that_ran;
 472   simix_global->process_that_ran = simix_global->process_to_run;
 473   simix_global->process_to_run = tmp;
 474   xbt_dynar_reset(simix_global->process_to_run);
 475 }
 476
 477 void simcall_HANDLER_process_kill(smx_simcall_t simcall, smx_process_t process) {
 478   SIMIX_process_kill(process, simcall->issuer);
 479 }
 480 /**
 481  * \brief Internal function to kill a SIMIX process.
 482  *
 483  * This function may be called when a SIMCALL_PROCESS_KILL simcall occurs,
 484  * or directly for SIMIX internal purposes.
 485  *
 486  * \param process poor victim
 487  * \param issuer the process which has sent the PROCESS_KILL. Important to not schedule twice the same process.
 488  */
 489 void SIMIX_process_kill(smx_process_t process, smx_process_t issuer) {
 490
 491   XBT_DEBUG("Killing process %s on %s", process->name, sg_host_get_name(process->host));
 492
 493   process->context->iwannadie = 1;
 494   process->blocked = 0;
 495   process->suspended = 0;
 496   process->doexception = 0;
 497
 498   /* destroy the blocking synchro if any */
 499   if (process->waiting_synchro) {
 500
 501     simgrid::simix::Exec *exec = dynamic_cast<simgrid::simix::Exec*>(process->waiting_synchro);
 502     simgrid::simix::Comm *comm = dynamic_cast<simgrid::simix::Comm*>(process->waiting_synchro);
 503     simgrid::simix::Sleep *sleep = dynamic_cast<simgrid::simix::Sleep*>(process->waiting_synchro);
 504     simgrid::simix::Raw *raw = dynamic_cast<simgrid::simix::Raw*>(process->waiting_synchro);
 505     simgrid::simix::Io *io = dynamic_cast<simgrid::simix::Io*>(process->waiting_synchro);
 506
 507     if (exec != nullptr) {
 508       exec->unref();
 509
 510     } else if (comm != nullptr) {
 511       xbt_fifo_remove(process->comms, process->waiting_synchro);
 512       comm->cancel();
 513       xbt_fifo_remove(process->waiting_synchro->simcalls, &process->simcall);
 514       comm->unref();
 515
 516     } else if (sleep != nullptr) {
 517       SIMIX_process_sleep_destroy(process->waiting_synchro);
 518
 519     } else if (raw != nullptr) {
 520       SIMIX_synchro_stop_waiting(process, &process->simcall);
 521       delete process->waiting_synchro;
 522
 523     } else if (io != nullptr) {
 524       SIMIX_io_destroy(process->waiting_synchro);
 525     }
 526
 527     /*
 528     switch (process->waiting_synchro->type) {
 529     case SIMIX_SYNC_JOIN:
 530       SIMIX_process_sleep_destroy(process->waiting_synchro);
 531       break;
 532     } */
 533
 534     process->waiting_synchro = NULL;
 535   }
 536   if(!xbt_dynar_member(simix_global->process_to_run, &(process)) && process != issuer) {
 537     XBT_DEBUG("Inserting %s in the to_run list", process->name);
 538     xbt_dynar_push_as(simix_global->process_to_run, smx_process_t, process);
 539   }
 540
 541 }
 542
 543 /** @brief Ask another process to raise the given exception
 544  *
 545  * @param cat category of exception
 546  * @param value value associated to the exception
 547  * @param msg string information associated to the exception
 548  */
 549 void SIMIX_process_throw(smx_process_t process, xbt_errcat_t cat, int value, const char *msg) {
 550   SMX_EXCEPTION(process, cat, value, msg);
 551
 552   if (process->suspended)
 553     SIMIX_process_resume(process,SIMIX_process_self());
 554
 555   /* cancel the blocking synchro if any */
 556   if (process->waiting_synchro) {
 557
 558     simgrid::simix::Exec *exec = dynamic_cast<simgrid::simix::Exec*>(process->waiting_synchro);
 559     if (exec != nullptr) {
 560       SIMIX_execution_cancel(process->waiting_synchro);
 561     }
 562
 563     simgrid::simix::Comm *comm = dynamic_cast<simgrid::simix::Comm*>(process->waiting_synchro);
 564     if (comm != nullptr) {
 565       xbt_fifo_remove(process->comms, comm);
 566       comm->cancel();
 567     }
 568
 569     simgrid::simix::Sleep *sleep = dynamic_cast<simgrid::simix::Sleep*>(process->waiting_synchro);
 570     if (sleep != nullptr) {
 571       SIMIX_process_sleep_destroy(process->waiting_synchro);
 572       if (!xbt_dynar_member(simix_global->process_to_run, &(process)) && process != SIMIX_process_self()) {
 573         XBT_DEBUG("Inserting %s in the to_run list", process->name);
 574         xbt_dynar_push_as(simix_global->process_to_run, smx_process_t, process);
 575       }
 576     }
 577
 578     simgrid::simix::Raw *raw = dynamic_cast<simgrid::simix::Raw*>(process->waiting_synchro);
 579     if (raw != nullptr) {
 580       SIMIX_synchro_stop_waiting(process, &process->simcall);
 581     }
 582
 583     simgrid::simix::Io *io = dynamic_cast<simgrid::simix::Io*>(process->waiting_synchro);
 584     if (io != nullptr) {
 585       SIMIX_io_destroy(process->waiting_synchro);
 586     }
 587   }
 588   process->waiting_synchro = NULL;
 589
 590 }
 591
 592 void simcall_HANDLER_process_killall(smx_simcall_t simcall, int reset_pid) {
 593   SIMIX_process_killall(simcall->issuer, reset_pid);
 594 }
 595 /**
 596  * \brief Kills all running processes.
 597  * \param issuer this one will not be killed
 598  */
 599 void SIMIX_process_killall(smx_process_t issuer, int reset_pid)
 600 {
 601   smx_process_t p = NULL;
 602
 603   while ((p = (smx_process_t) xbt_swag_extract(simix_global->process_list))) {
 604     if (p != issuer) {
 605       SIMIX_process_kill(p,issuer);
 606     }
 607   }
 608
 609   if (reset_pid > 0)
 610     simix_process_maxpid = reset_pid;
 611
 612   SIMIX_context_runall();
 613
 614   SIMIX_process_empty_trash();
 615 }
 616
 617 void simcall_HANDLER_process_set_host(smx_simcall_t simcall, smx_process_t process, sg_host_t dest)
 618 {
 619   process->new_host = dest;
 620 }
 621 void SIMIX_process_change_host(smx_process_t process,
 622              sg_host_t dest)
 623 {
 624   xbt_assert((process != NULL), "Invalid parameters");
 625   xbt_swag_remove(process, sg_host_simix(process->host)->process_list);
 626   process->host = dest;
 627   xbt_swag_insert(process, sg_host_simix(dest)->process_list);
 628 }
 629
 630
 631 void simcall_HANDLER_process_suspend(smx_simcall_t simcall, smx_process_t process)
 632 {
 633   smx_synchro_t sync_suspend = 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     process->waiting_synchro->suspend();
 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   if (process->suspended) {
 648     XBT_DEBUG("Process '%s' is already suspended", process->name);
 649     return NULL;
 650   }
 651
 652   process->suspended = 1;
 653
 654   /* If we are suspending another process that is waiting on a sync, suspend its synchronization. */
 655   if (process != issuer) {
 656
 657     if (process->waiting_synchro)
 658       process->waiting_synchro->suspend();
 659     /* If the other process is not waiting, its suspension is delayed to when the process is rescheduled. */
 660
 661     return NULL;
 662   } else {
 663     /* FIXME: computation size is zero. Is it okay that bound is zero ? */
 664     return SIMIX_execution_start(process, "suspend", 0.0, 1.0, 0.0, 0);
 665   }
 666 }
 667
 668 void simcall_HANDLER_process_resume(smx_simcall_t simcall, smx_process_t process){
 669   SIMIX_process_resume(process, simcall->issuer);
 670 }
 671
 672 void SIMIX_process_resume(smx_process_t process, smx_process_t issuer)
 673 {
 674   XBT_IN("process = %p, issuer = %p", process, issuer);
 675
 676   if(process->context->iwannadie) {
 677     XBT_VERB("Ignoring request to suspend a process that is currently dying.");
 678     return;
 679   }
 680
 681   if(!process->suspended) return;
 682   process->suspended = 0;
 683
 684   /* If we are resuming another process, resume the synchronization it was waiting for
 685      if any. Otherwise add it to the list of process to run in the next round. */
 686   if (process != issuer) {
 687
 688     if (process->waiting_synchro) {
 689       process->waiting_synchro->resume();
 690     }
 691   } else XBT_WARN("Strange. Process %p is trying to resume himself.", issuer);
 692
 693   XBT_OUT();
 694 }
 695
 696 int SIMIX_process_get_maxpid(void) {
 697   return simix_process_maxpid;
 698 }
 699
 700 int SIMIX_process_count(void)
 701 {
 702   return xbt_swag_size(simix_global->process_list);
 703 }
 704
 705 int SIMIX_process_get_PID(smx_process_t self){
 706   if (self == NULL)
 707     return 0;
 708   else
 709     return self->pid;
 710 }
 711
 712 int SIMIX_process_get_PPID(smx_process_t self){
 713   if (self == NULL)
 714     return 0;
 715   else
 716     return self->ppid;
 717 }
 718
 719 void* SIMIX_process_self_get_data()
 720 {
 721   smx_process_t self = SIMIX_process_self();
 722
 723   if (!self) {
 724     return NULL;
 725   }
 726   return SIMIX_process_get_data(self);
 727 }
 728
 729 void SIMIX_process_self_set_data(void *data)
 730 {
 731   smx_process_t self = SIMIX_process_self();
 732
 733   SIMIX_process_set_data(self, data);
 734 }
 735
 736 void* SIMIX_process_get_data(smx_process_t process)
 737 {
 738   return process->data;
 739 }
 740
 741 void SIMIX_process_set_data(smx_process_t process, void *data)
 742 {
 743   process->data = data;
 744 }
 745
 746 sg_host_t SIMIX_process_get_host(smx_process_t process)
 747 {
 748   return process->host;
 749 }
 750
 751 xbt_main_func_t SIMIX_process_get_code(void){
 752   return SIMIX_process_self()->code;
 753 }
 754
 755 /* needs to be public and without simcall because it is called
 756    by exceptions and logging events */
 757 const char* SIMIX_process_self_get_name(void) {
 758
 759   smx_process_t process = SIMIX_process_self();
 760   if (process == NULL || process == simix_global->maestro_process)
 761     return "maestro";
 762
 763   return SIMIX_process_get_name(process);
 764 }
 765
 766 const char* SIMIX_process_get_name(smx_process_t process)
 767 {
 768   return process->name;
 769 }
 770
 771 smx_process_t SIMIX_process_get_by_name(const char* name)
 772 {
 773   smx_process_t proc;
 774
 775   xbt_swag_foreach(proc, simix_global->process_list)
 776   {
 777     if(!strcmp(name, proc->name))
 778       return proc;
 779   }
 780   return NULL;
 781 }
 782
 783 int SIMIX_process_is_suspended(smx_process_t process)
 784 {
 785   return process->suspended;
 786 }
 787
 788 xbt_dict_t SIMIX_process_get_properties(smx_process_t process)
 789 {
 790   return process->properties;
 791 }
 792
 793 void simcall_HANDLER_process_join(smx_simcall_t simcall, smx_process_t process, double timeout)
 794 {
 795   smx_synchro_t sync = SIMIX_process_join(simcall->issuer, process, timeout);
 796   xbt_fifo_push(sync->simcalls, simcall);
 797   simcall->issuer->waiting_synchro = sync;
 798 }
 799
 800 static int SIMIX_process_join_finish(smx_process_exit_status_t status, smx_synchro_t synchro){
 801   simgrid::simix::Sleep *sleep = static_cast<simgrid::simix::Sleep*>(synchro);
 802
 803   if (sleep->surf_sleep) {
 804     sleep->surf_sleep->cancel();
 805
 806     smx_simcall_t simcall;
 807     while ((simcall = (smx_simcall_t) xbt_fifo_shift(sleep->simcalls))) {
 808       simcall_process_sleep__set__result(simcall, SIMIX_DONE);
 809       simcall->issuer->waiting_synchro = NULL;
 810       if (simcall->issuer->suspended) {
 811         XBT_DEBUG("Wait! This process is suspended and can't wake up now.");
 812         simcall->issuer->suspended = 0;
 813         simcall_HANDLER_process_suspend(simcall, simcall->issuer);
 814       } else {
 815         SIMIX_simcall_answer(simcall);
 816       }
 817     }
 818     sleep->surf_sleep->unref();
 819     sleep->surf_sleep = NULL;
 820   }
 821   delete sleep;
 822   return 0;
 823 }
 824
 825 smx_synchro_t SIMIX_process_join(smx_process_t issuer, smx_process_t process, double timeout)
 826 {
 827   smx_synchro_t res = SIMIX_process_sleep(issuer, timeout);
 828   SIMIX_process_on_exit(process, (int_f_pvoid_pvoid_t)SIMIX_process_join_finish, res);
 829   return res;
 830 }
 831
 832 void simcall_HANDLER_process_sleep(smx_simcall_t simcall, double duration)
 833 {
 834   if (MC_is_active() || MC_record_replay_is_active()) {
 835     MC_process_clock_add(simcall->issuer, duration);
 836     simcall_process_sleep__set__result(simcall, SIMIX_DONE);
 837     SIMIX_simcall_answer(simcall);
 838     return;
 839   }
 840   smx_synchro_t sync = SIMIX_process_sleep(simcall->issuer, duration);
 841   xbt_fifo_push(sync->simcalls, simcall);
 842   simcall->issuer->waiting_synchro = sync;
 843 }
 844
 845 smx_synchro_t SIMIX_process_sleep(smx_process_t process, double duration)
 846 {
 847   sg_host_t host = process->host;
 848
 849   /* check if the host is active */
 850   if (host->isOff())
 851     THROWF(host_error, 0, "Host %s failed, you cannot call this function", sg_host_get_name(host));
 852
 853   simgrid::simix::Sleep *synchro = new simgrid::simix::Sleep();
 854   synchro->name = NULL;
 855
 856   synchro->host = host;
 857   synchro->surf_sleep = surf_host_sleep(host, duration);
 858   synchro->surf_sleep->setData(synchro);
 859   XBT_DEBUG("Create sleep synchronization %p", synchro);
 860
 861   return synchro;
 862 }
 863
 864 void SIMIX_process_sleep_destroy(smx_synchro_t synchro)
 865 {
 866   XBT_DEBUG("Destroy synchro %p", synchro);
 867   simgrid::simix::Sleep *sleep = static_cast<simgrid::simix::Sleep*>(synchro);
 868
 869   if (sleep->surf_sleep) {
 870     sleep->surf_sleep->unref();
 871     sleep->surf_sleep = NULL;
 872   }
 873 }
 874
 875 /**
 876  * \brief Calling this function makes the process to yield.
 877  *
 878  * Only the current process can call this function, giving back the control to
 879  * maestro.
 880  *
 881  * \param self the current process
 882  */
 883 void SIMIX_process_yield(smx_process_t self)
 884 {
 885   XBT_DEBUG("Yield process '%s'", self->name);
 886
 887   /* Go into sleep and return control to maestro */
 888   self->context->suspend();
 889
 890   /* Ok, maestro returned control to us */
 891   XBT_DEBUG("Control returned to me: '%s'", self->name);
 892
 893   if (self->new_host) {
 894     SIMIX_process_change_host(self, self->new_host);
 895     self->new_host = NULL;
 896   }
 897
 898   if (self->context->iwannadie){
 899     XBT_DEBUG("I wanna die!");
 900     SIMIX_process_stop(self);
 901   }
 902
 903   if (self->suspended) {
 904     XBT_DEBUG("Hey! I'm suspended.");
 905     xbt_assert(!self->doexception, "Gasp! This exception may be lost by subsequent calls.");
 906     self->suspended = 0;
 907     SIMIX_process_suspend(self, self);
 908   }
 909
 910   if (self->doexception) {
 911     XBT_DEBUG("Wait, maestro left me an exception");
 912     self->doexception = 0;
 913     RETHROW;
 914   }
 915
 916   if(SMPI_switch_data_segment && self->segment_index != -1){
 917     SMPI_switch_data_segment(self->segment_index);
 918   }
 919 }
 920
 921 /* callback: context fetching */
 922 xbt_running_ctx_t *SIMIX_process_get_running_context(void)
 923 {
 924   smx_process_t process = SIMIX_process_self();
 925   if (process)
 926     return process->running_ctx;
 927   else
 928     return nullptr;
 929 }
 930
 931 /* callback: termination */
 932 void SIMIX_process_exception_terminate(xbt_ex_t * e)
 933 {
 934   xbt_ex_display(e);
 935   xbt_abort();
 936 }
 937
 938 smx_context_t SIMIX_process_get_context(smx_process_t p) {
 939   return p->context;
 940 }
 941
 942 void SIMIX_process_set_context(smx_process_t p,smx_context_t c) {
 943   p->context = c;
 944 }
 945
 946 /**
 947  * \brief Returns the list of processes to run.
 948  */
 949 xbt_dynar_t SIMIX_process_get_runnable(void)
 950 {
 951   return simix_global->process_to_run;
 952 }
 953
 954 /**
 955  * \brief Returns the process from PID.
 956  */
 957 smx_process_t SIMIX_process_from_PID(int PID)
 958 {
 959   smx_process_t proc;
 960   xbt_swag_foreach(proc, simix_global->process_list) {
 961    if (proc->pid == (unsigned long) PID)
 962     return proc;
 963   }
 964   return NULL;
 965 }
 966
 967 /** @brief returns a dynar containg all currently existing processes */
 968 xbt_dynar_t SIMIX_processes_as_dynar(void) {
 969   smx_process_t proc;
 970   xbt_dynar_t res = xbt_dynar_new(sizeof(smx_process_t),NULL);
 971   xbt_swag_foreach(proc, simix_global->process_list) {
 972     xbt_dynar_push(res,&proc);
 973   }
 974   return res;
 975 }
 976
 977
 978 void SIMIX_process_on_exit_runall(smx_process_t process) {
 979   s_smx_process_exit_fun_t exit_fun;
 980   smx_process_exit_status_t exit_status = (process->context->iwannadie) ?
 981                                          SMX_EXIT_FAILURE : SMX_EXIT_SUCCESS;
 982   while (!xbt_dynar_is_empty(process->on_exit)) {
 983     exit_fun = xbt_dynar_pop_as(process->on_exit,s_smx_process_exit_fun_t);
 984     (exit_fun.fun)((void*)exit_status, exit_fun.arg);
 985   }
 986 }
 987
 988 void SIMIX_process_on_exit(smx_process_t process, int_f_pvoid_pvoid_t fun, void *data) {
 989   xbt_assert(process, "current process not found: are you in maestro context ?");
 990
 991   if (!process->on_exit) {
 992     process->on_exit = xbt_dynar_new(sizeof(s_smx_process_exit_fun_t), NULL);
 993   }
 994
 995   s_smx_process_exit_fun_t exit_fun = {fun, data};
 996
 997   xbt_dynar_push_as(process->on_exit,s_smx_process_exit_fun_t,exit_fun);
 998 }
 999
1000 /**
1001  * \brief Sets the auto-restart status of the process.
1002  * If set to 1, the process will be automatically restarted when its host
1003  * comes back.
1004  */
1005 void SIMIX_process_auto_restart_set(smx_process_t process, int auto_restart) {
1006   process->auto_restart = auto_restart;
1007 }
1008
1009 smx_process_t simcall_HANDLER_process_restart(smx_simcall_t simcall, smx_process_t process) {
1010   return SIMIX_process_restart(process, simcall->issuer);
1011 }
1012 /** @brief Restart a process, starting it again from the beginning. */
1013 smx_process_t SIMIX_process_restart(smx_process_t process, smx_process_t issuer) {
1014   XBT_DEBUG("Restarting process %s on %s", process->name, sg_host_get_name(process->host));
1015   //retrieve the arguments of the old process
1016   //FIXME: Factorize this with SIMIX_host_add_auto_restart_process ?
1017   s_smx_process_arg_t arg;
1018   arg.code = process->code;
1019   arg.hostname = sg_host_get_name(process->host);
1020   arg.kill_time = SIMIX_timer_get_date(process->kill_timer);
1021   arg.argc = process->argc;
1022   arg.data = process->data;
1023   int i;
1024   arg.argv = xbt_new(char*,process->argc + 1);
1025   for (i = 0; i < arg.argc; i++) {
1026     arg.argv[i] = xbt_strdup(process->argv[i]);
1027   }
1028   arg.argv[process->argc] = NULL;
1029   arg.properties = NULL;
1030   arg.auto_restart = process->auto_restart;
1031   //kill the old process
1032   SIMIX_process_kill(process,issuer);
1033   //start the new process
1034   smx_process_t new_process;
1035   if (simix_global->create_process_function) {
1036     new_process = simix_global->create_process_function(
1037                                           arg.argv[0],
1038                                           arg.code,
1039                                           arg.data,
1040                                           arg.hostname,
1041                                           arg.kill_time,
1042                                           arg.argc,
1043                                           arg.argv,
1044                                           arg.properties,
1045                                           arg.auto_restart,
1046                                           NULL);
1047   } else {
1048     new_process = simcall_process_create(
1049                            arg.argv[0],
1050                            arg.code,
1051                            arg.data,
1052                            arg.hostname,
1053                            arg.kill_time,
1054                            arg.argc,
1055                            arg.argv,
1056                            arg.properties,
1057                            arg.auto_restart);
1058
1059   }
1060   return new_process;
1061 }
1062
1063 void SIMIX_segment_index_set(smx_process_t proc, int index){
1064   proc->segment_index = index;
1065 }