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 ? self_context->process() : 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         XBT_DEBUG("Don't destroy it since it's a detached comm and I'm the sender");
  86       else
  87         SIMIX_comm_destroy(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     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     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   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   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     simgrid::simix::Exec *exec = dynamic_cast<simgrid::simix::Exec*>(process->waiting_synchro);
 500     simgrid::simix::Comm *comm = dynamic_cast<simgrid::simix::Comm*>(process->waiting_synchro);
 501     simgrid::simix::Sleep *sleep = dynamic_cast<simgrid::simix::Sleep*>(process->waiting_synchro);
 502     simgrid::simix::Raw *raw = dynamic_cast<simgrid::simix::Raw*>(process->waiting_synchro);
 503     simgrid::simix::Io *io = dynamic_cast<simgrid::simix::Io*>(process->waiting_synchro);
 504
 505     if (exec != nullptr) {
 506       SIMIX_execution_destroy(process->waiting_synchro);
 507
 508     } else if (comm != nullptr) {
 509       xbt_fifo_remove(process->comms, process->waiting_synchro);
 510       SIMIX_comm_cancel(process->waiting_synchro);
 511       xbt_fifo_remove(process->waiting_synchro->simcalls, &process->simcall);
 512       SIMIX_comm_destroy(process->waiting_synchro);
 513
 514     } else if (sleep != nullptr) {
 515       SIMIX_process_sleep_destroy(process->waiting_synchro);
 516
 517     } else if (raw != nullptr) {
 518       SIMIX_synchro_stop_waiting(process, &process->simcall);
 519       SIMIX_synchro_destroy(process->waiting_synchro);
 520
 521     } else if (io != nullptr) {
 522       SIMIX_io_destroy(process->waiting_synchro);
 523     }
 524
 525     /*
 526     switch (process->waiting_synchro->type) {
 527     case SIMIX_SYNC_JOIN:
 528       SIMIX_process_sleep_destroy(process->waiting_synchro);
 529       break;
 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     simgrid::simix::Exec *exec = dynamic_cast<simgrid::simix::Exec*>(process->waiting_synchro);
 557     if (exec != nullptr) {
 558       SIMIX_execution_cancel(process->waiting_synchro);
 559     }
 560
 561     simgrid::simix::Comm *comm = dynamic_cast<simgrid::simix::Comm*>(process->waiting_synchro);
 562     if (comm != nullptr) {
 563       xbt_fifo_remove(process->comms, process->waiting_synchro);
 564       SIMIX_comm_cancel(process->waiting_synchro);
 565     }
 566
 567     simgrid::simix::Sleep *sleep = dynamic_cast<simgrid::simix::Sleep*>(process->waiting_synchro);
 568     if (sleep != nullptr) {
 569       SIMIX_process_sleep_destroy(process->waiting_synchro);
 570       if (!xbt_dynar_member(simix_global->process_to_run, &(process)) && process != SIMIX_process_self()) {
 571         XBT_DEBUG("Inserting %s in the to_run list", process->name);
 572         xbt_dynar_push_as(simix_global->process_to_run, smx_process_t, process);
 573       }
 574     }
 575
 576     simgrid::simix::Raw *raw = dynamic_cast<simgrid::simix::Raw*>(process->waiting_synchro);
 577     if (raw != nullptr) {
 578       SIMIX_synchro_stop_waiting(process, &process->simcall);
 579     }
 580
 581     simgrid::simix::Io *io = dynamic_cast<simgrid::simix::Io*>(process->waiting_synchro);
 582     if (io != nullptr) {
 583       SIMIX_io_destroy(process->waiting_synchro);
 584     }
 585   }
 586   process->waiting_synchro = NULL;
 587
 588 }
 589
 590 void simcall_HANDLER_process_killall(smx_simcall_t simcall, int reset_pid) {
 591   SIMIX_process_killall(simcall->issuer, reset_pid);
 592 }
 593 /**
 594  * \brief Kills all running processes.
 595  * \param issuer this one will not be killed
 596  */
 597 void SIMIX_process_killall(smx_process_t issuer, int reset_pid)
 598 {
 599   smx_process_t p = NULL;
 600
 601   while ((p = (smx_process_t) xbt_swag_extract(simix_global->process_list))) {
 602     if (p != issuer) {
 603       SIMIX_process_kill(p,issuer);
 604     }
 605   }
 606
 607   if (reset_pid > 0)
 608     simix_process_maxpid = reset_pid;
 609
 610   SIMIX_context_runall();
 611
 612   SIMIX_process_empty_trash();
 613 }
 614
 615 void simcall_HANDLER_process_set_host(smx_simcall_t simcall, smx_process_t process, sg_host_t dest)
 616 {
 617   process->new_host = dest;
 618 }
 619 void SIMIX_process_change_host(smx_process_t process,
 620              sg_host_t dest)
 621 {
 622   xbt_assert((process != NULL), "Invalid parameters");
 623   xbt_swag_remove(process, sg_host_simix(process->host)->process_list);
 624   process->host = dest;
 625   xbt_swag_insert(process, sg_host_simix(dest)->process_list);
 626 }
 627
 628
 629 void simcall_HANDLER_process_suspend(smx_simcall_t simcall, smx_process_t process)
 630 {
 631   smx_synchro_t sync_suspend =
 632       SIMIX_process_suspend(process, simcall->issuer);
 633
 634   if (process != simcall->issuer) {
 635     SIMIX_simcall_answer(simcall);
 636   } else {
 637     xbt_fifo_push(sync_suspend->simcalls, simcall);
 638     process->waiting_synchro = sync_suspend;
 639     SIMIX_execution_suspend(process->waiting_synchro);
 640   }
 641   /* If we are suspending ourselves, then just do not finish the simcall now */
 642 }
 643
 644 smx_synchro_t SIMIX_process_suspend(smx_process_t process, smx_process_t issuer)
 645 {
 646   if (process->suspended) {
 647     XBT_DEBUG("Process '%s' is already suspended", process->name);
 648     return NULL;
 649   }
 650
 651   process->suspended = 1;
 652
 653   /* If we are suspending another process that is waiting on a sync, suspend its synchronization. */
 654   if (process != issuer) {
 655
 656     if (process->waiting_synchro)
 657       process->waiting_synchro->suspend();
 658     /* If the other process is not waiting, its suspension is delayed to when the process is rescheduled. */
 659
 660     return NULL;
 661   } else {
 662     /* FIXME: computation size is zero. Is it okay that bound is zero ? */
 663     return SIMIX_execution_start(process, "suspend", 0.0, 1.0, 0.0, 0);
 664   }
 665 }
 666
 667 void simcall_HANDLER_process_resume(smx_simcall_t simcall, smx_process_t process){
 668   SIMIX_process_resume(process, simcall->issuer);
 669 }
 670
 671 void SIMIX_process_resume(smx_process_t process, smx_process_t issuer)
 672 {
 673   XBT_IN("process = %p, issuer = %p", process, issuer);
 674
 675   if(process->context->iwannadie) {
 676     XBT_VERB("Ignoring request to suspend a process that is currently dying.");
 677     return;
 678   }
 679
 680   if(!process->suspended) return;
 681   process->suspended = 0;
 682
 683   /* If we are resuming another process, resume the synchronization it was waiting for
 684      if any. Otherwise add it to the list of process to run in the next round. */
 685   if (process != issuer) {
 686
 687     if (process->waiting_synchro) {
 688       process->waiting_synchro->resume();
 689     }
 690   } else XBT_WARN("Strange. Process %p is trying to resume himself.", issuer);
 691
 692   XBT_OUT();
 693 }
 694
 695 int SIMIX_process_get_maxpid(void) {
 696   return simix_process_maxpid;
 697 }
 698
 699 int SIMIX_process_count(void)
 700 {
 701   return xbt_swag_size(simix_global->process_list);
 702 }
 703
 704 int SIMIX_process_get_PID(smx_process_t self){
 705   if (self == NULL)
 706     return 0;
 707   else
 708     return self->pid;
 709 }
 710
 711 int SIMIX_process_get_PPID(smx_process_t self){
 712   if (self == NULL)
 713     return 0;
 714   else
 715     return self->ppid;
 716 }
 717
 718 void* SIMIX_process_self_get_data()
 719 {
 720   smx_process_t self = SIMIX_process_self();
 721
 722   if (!self) {
 723     return NULL;
 724   }
 725   return SIMIX_process_get_data(self);
 726 }
 727
 728 void SIMIX_process_self_set_data(void *data)
 729 {
 730   smx_process_t self = SIMIX_process_self();
 731
 732   SIMIX_process_set_data(self, data);
 733 }
 734
 735 void* SIMIX_process_get_data(smx_process_t process)
 736 {
 737   return process->data;
 738 }
 739
 740 void SIMIX_process_set_data(smx_process_t process, void *data)
 741 {
 742   process->data = data;
 743 }
 744
 745 sg_host_t SIMIX_process_get_host(smx_process_t process)
 746 {
 747   return process->host;
 748 }
 749
 750 xbt_main_func_t SIMIX_process_get_code(void){
 751   return SIMIX_process_self()->code;
 752 }
 753
 754 /* needs to be public and without simcall because it is called
 755    by exceptions and logging events */
 756 const char* SIMIX_process_self_get_name(void) {
 757
 758   smx_process_t process = SIMIX_process_self();
 759   if (process == NULL || process == simix_global->maestro_process)
 760     return "maestro";
 761
 762   return SIMIX_process_get_name(process);
 763 }
 764
 765 const char* SIMIX_process_get_name(smx_process_t process)
 766 {
 767   return process->name;
 768 }
 769
 770 smx_process_t SIMIX_process_get_by_name(const char* name)
 771 {
 772   smx_process_t proc;
 773
 774   xbt_swag_foreach(proc, simix_global->process_list)
 775   {
 776     if(!strcmp(name, proc->name))
 777       return proc;
 778   }
 779   return NULL;
 780 }
 781
 782 int SIMIX_process_is_suspended(smx_process_t process)
 783 {
 784   return process->suspended;
 785 }
 786
 787 xbt_dict_t SIMIX_process_get_properties(smx_process_t process)
 788 {
 789   return process->properties;
 790 }
 791
 792 void simcall_HANDLER_process_join(smx_simcall_t simcall, smx_process_t process, double timeout)
 793 {
 794   smx_synchro_t sync = SIMIX_process_join(simcall->issuer, process, timeout);
 795   xbt_fifo_push(sync->simcalls, simcall);
 796   simcall->issuer->waiting_synchro = sync;
 797 }
 798
 799 static int SIMIX_process_join_finish(smx_process_exit_status_t status, smx_synchro_t synchro){
 800   simgrid::simix::Sleep *sleep = static_cast<simgrid::simix::Sleep*>(synchro);
 801
 802   if (sleep->surf_sleep) {
 803     sleep->surf_sleep->cancel();
 804
 805     smx_simcall_t simcall;
 806     while ((simcall = (smx_simcall_t) xbt_fifo_shift(sleep->simcalls))) {
 807       simcall_process_sleep__set__result(simcall, SIMIX_DONE);
 808       simcall->issuer->waiting_synchro = NULL;
 809       if (simcall->issuer->suspended) {
 810         XBT_DEBUG("Wait! This process is suspended and can't wake up now.");
 811         simcall->issuer->suspended = 0;
 812         simcall_HANDLER_process_suspend(simcall, simcall->issuer);
 813       } else {
 814         SIMIX_simcall_answer(simcall);
 815       }
 816     }
 817     sleep->surf_sleep->unref();
 818     sleep->surf_sleep = NULL;
 819   }
 820   delete sleep;
 821   return 0;
 822 }
 823
 824 smx_synchro_t SIMIX_process_join(smx_process_t issuer, smx_process_t process, double timeout)
 825 {
 826   smx_synchro_t res = SIMIX_process_sleep(issuer, timeout);
 827   SIMIX_process_on_exit(process, (int_f_pvoid_pvoid_t)SIMIX_process_join_finish, res);
 828   return res;
 829 }
 830
 831 void simcall_HANDLER_process_sleep(smx_simcall_t simcall, double duration)
 832 {
 833   if (MC_is_active() || MC_record_replay_is_active()) {
 834     MC_process_clock_add(simcall->issuer, duration);
 835     simcall_process_sleep__set__result(simcall, SIMIX_DONE);
 836     SIMIX_simcall_answer(simcall);
 837     return;
 838   }
 839   smx_synchro_t sync = SIMIX_process_sleep(simcall->issuer, duration);
 840   xbt_fifo_push(sync->simcalls, simcall);
 841   simcall->issuer->waiting_synchro = sync;
 842 }
 843
 844 smx_synchro_t SIMIX_process_sleep(smx_process_t process, double duration)
 845 {
 846   sg_host_t host = process->host;
 847
 848   /* check if the host is active */
 849   if (host->isOff())
 850     THROWF(host_error, 0, "Host %s failed, you cannot call this function", sg_host_get_name(host));
 851
 852   simgrid::simix::Sleep *synchro = new simgrid::simix::Sleep();
 853   synchro->name = NULL;
 854
 855   synchro->host = host;
 856   synchro->surf_sleep = surf_host_sleep(host, duration);
 857   synchro->surf_sleep->setData(synchro);
 858   XBT_DEBUG("Create sleep synchronization %p", synchro);
 859
 860   return synchro;
 861 }
 862
 863 void SIMIX_post_process_sleep(smx_synchro_t synchro)
 864 {
 865   smx_simcall_t simcall;
 866   e_smx_state_t state;
 867   simgrid::simix::Sleep *sleep = static_cast<simgrid::simix::Sleep*>(synchro);
 868
 869   while ((simcall = (smx_simcall_t) xbt_fifo_shift(synchro->simcalls))) {
 870
 871     switch (sleep->surf_sleep->getState()){
 872       case simgrid::surf::Action::State::failed:
 873         simcall->issuer->context->iwannadie = 1;
 874         //SMX_EXCEPTION(simcall->issuer, host_error, 0, "Host failed");
 875         state = SIMIX_SRC_HOST_FAILURE;
 876         break;
 877
 878       case simgrid::surf::Action::State::done:
 879         state = SIMIX_DONE;
 880         break;
 881
 882       default:
 883         THROW_IMPOSSIBLE;
 884         break;
 885     }
 886     if (simcall->issuer->host->isOff()) {
 887       simcall->issuer->context->iwannadie = 1;
 888     }
 889     simcall_process_sleep__set__result(simcall, state);
 890     simcall->issuer->waiting_synchro = NULL;
 891     if (simcall->issuer->suspended) {
 892       XBT_DEBUG("Wait! This process is suspended and can't wake up now.");
 893       simcall->issuer->suspended = 0;
 894       simcall_HANDLER_process_suspend(simcall, simcall->issuer);
 895     } else {
 896       SIMIX_simcall_answer(simcall);
 897     }
 898   }
 899
 900   SIMIX_process_sleep_destroy(synchro);
 901 }
 902
 903 void SIMIX_process_sleep_destroy(smx_synchro_t synchro)
 904 {
 905   XBT_DEBUG("Destroy synchro %p", synchro);
 906   simgrid::simix::Sleep *sleep = static_cast<simgrid::simix::Sleep*>(synchro);
 907
 908   if (sleep->surf_sleep) {
 909     sleep->surf_sleep->unref();
 910     sleep->surf_sleep = NULL;
 911   }
 912 }
 913
 914 /**
 915  * \brief Calling this function makes the process to yield.
 916  *
 917  * Only the current process can call this function, giving back the control to
 918  * maestro.
 919  *
 920  * \param self the current process
 921  */
 922 void SIMIX_process_yield(smx_process_t self)
 923 {
 924   XBT_DEBUG("Yield process '%s'", self->name);
 925
 926   /* Go into sleep and return control to maestro */
 927   self->context->suspend();
 928
 929   /* Ok, maestro returned control to us */
 930   XBT_DEBUG("Control returned to me: '%s'", self->name);
 931
 932   if (self->new_host) {
 933     SIMIX_process_change_host(self, self->new_host);
 934     self->new_host = NULL;
 935   }
 936
 937   if (self->context->iwannadie){
 938     XBT_DEBUG("I wanna die!");
 939     SIMIX_process_stop(self);
 940   }
 941
 942   if (self->suspended) {
 943     XBT_DEBUG("Hey! I'm suspended.");
 944     xbt_assert(!self->doexception, "Gasp! This exception may be lost by subsequent calls.");
 945     self->suspended = 0;
 946     SIMIX_process_suspend(self, self);
 947   }
 948
 949   if (self->doexception) {
 950     XBT_DEBUG("Wait, maestro left me an exception");
 951     self->doexception = 0;
 952     RETHROW;
 953   }
 954
 955   if(SMPI_switch_data_segment && self->segment_index != -1){
 956     SMPI_switch_data_segment(self->segment_index);
 957   }
 958 }
 959
 960 /* callback: context fetching */
 961 xbt_running_ctx_t *SIMIX_process_get_running_context(void)
 962 {
 963   smx_process_t process = SIMIX_process_self();
 964   if (process)
 965     return process->running_ctx;
 966   else
 967     return nullptr;
 968 }
 969
 970 /* callback: termination */
 971 void SIMIX_process_exception_terminate(xbt_ex_t * e)
 972 {
 973   xbt_ex_display(e);
 974   xbt_abort();
 975 }
 976
 977 smx_context_t SIMIX_process_get_context(smx_process_t p) {
 978   return p->context;
 979 }
 980
 981 void SIMIX_process_set_context(smx_process_t p,smx_context_t c) {
 982   p->context = c;
 983 }
 984
 985 /**
 986  * \brief Returns the list of processes to run.
 987  */
 988 xbt_dynar_t SIMIX_process_get_runnable(void)
 989 {
 990   return simix_global->process_to_run;
 991 }
 992
 993 /**
 994  * \brief Returns the process from PID.
 995  */
 996 smx_process_t SIMIX_process_from_PID(int PID)
 997 {
 998   smx_process_t proc;
 999   xbt_swag_foreach(proc, simix_global->process_list) {
1000    if (proc->pid == (unsigned long) PID)
1001     return proc;
1002   }
1003   return NULL;
1004 }
1005
1006 /** @brief returns a dynar containg all currently existing processes */
1007 xbt_dynar_t SIMIX_processes_as_dynar(void) {
1008   smx_process_t proc;
1009   xbt_dynar_t res = xbt_dynar_new(sizeof(smx_process_t),NULL);
1010   xbt_swag_foreach(proc, simix_global->process_list) {
1011     xbt_dynar_push(res,&proc);
1012   }
1013   return res;
1014 }
1015
1016
1017 void SIMIX_process_on_exit_runall(smx_process_t process) {
1018   s_smx_process_exit_fun_t exit_fun;
1019   smx_process_exit_status_t exit_status = (process->context->iwannadie) ?
1020                                          SMX_EXIT_FAILURE : SMX_EXIT_SUCCESS;
1021   while (!xbt_dynar_is_empty(process->on_exit)) {
1022     exit_fun = xbt_dynar_pop_as(process->on_exit,s_smx_process_exit_fun_t);
1023     (exit_fun.fun)((void*)exit_status, exit_fun.arg);
1024   }
1025 }
1026
1027 void SIMIX_process_on_exit(smx_process_t process, int_f_pvoid_pvoid_t fun, void *data) {
1028   xbt_assert(process, "current process not found: are you in maestro context ?");
1029
1030   if (!process->on_exit) {
1031     process->on_exit = xbt_dynar_new(sizeof(s_smx_process_exit_fun_t), NULL);
1032   }
1033
1034   s_smx_process_exit_fun_t exit_fun = {fun, data};
1035
1036   xbt_dynar_push_as(process->on_exit,s_smx_process_exit_fun_t,exit_fun);
1037 }
1038
1039 /**
1040  * \brief Sets the auto-restart status of the process.
1041  * If set to 1, the process will be automatically restarted when its host
1042  * comes back.
1043  */
1044 void SIMIX_process_auto_restart_set(smx_process_t process, int auto_restart) {
1045   process->auto_restart = auto_restart;
1046 }
1047
1048 smx_process_t simcall_HANDLER_process_restart(smx_simcall_t simcall, smx_process_t process) {
1049   return SIMIX_process_restart(process, simcall->issuer);
1050 }
1051 /** @brief Restart a process, starting it again from the beginning. */
1052 smx_process_t SIMIX_process_restart(smx_process_t process, smx_process_t issuer) {
1053   XBT_DEBUG("Restarting process %s on %s", process->name, sg_host_get_name(process->host));
1054   //retrieve the arguments of the old process
1055   //FIXME: Factorize this with SIMIX_host_add_auto_restart_process ?
1056   s_smx_process_arg_t arg;
1057   arg.code = process->code;
1058   arg.hostname = sg_host_get_name(process->host);
1059   arg.kill_time = SIMIX_timer_get_date(process->kill_timer);
1060   arg.argc = process->argc;
1061   arg.data = process->data;
1062   int i;
1063   arg.argv = xbt_new(char*,process->argc + 1);
1064   for (i = 0; i < arg.argc; i++) {
1065     arg.argv[i] = xbt_strdup(process->argv[i]);
1066   }
1067   arg.argv[process->argc] = NULL;
1068   arg.properties = NULL;
1069   arg.auto_restart = process->auto_restart;
1070   //kill the old process
1071   SIMIX_process_kill(process,issuer);
1072   //start the new process
1073   smx_process_t new_process;
1074   if (simix_global->create_process_function) {
1075     new_process = simix_global->create_process_function(
1076                                           arg.argv[0],
1077                                           arg.code,
1078                                           arg.data,
1079                                           arg.hostname,
1080                                           arg.kill_time,
1081                                           arg.argc,
1082                                           arg.argv,
1083                                           arg.properties,
1084                                           arg.auto_restart,
1085                                           NULL);
1086   } else {
1087     new_process = simcall_process_create(
1088                            arg.argv[0],
1089                            arg.code,
1090                            arg.data,
1091                            arg.hostname,
1092                            arg.kill_time,
1093                            arg.argc,
1094                            arg.argv,
1095                            arg.properties,
1096                            arg.auto_restart);
1097
1098   }
1099   return new_process;
1100 }
1101
1102 void SIMIX_segment_index_set(smx_process_t proc, int index){
1103   proc->segment_index = index;
1104 }