src/simix/ActorImpl.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 <exception>
   8 #include <functional>
   9 #include <string>
  10 #include <utility>
  11
  12 #include <boost/range/algorithm.hpp>
  13
  14 #include <xbt/functional.hpp>
  15 #include <xbt/ex.hpp>
  16 #include <xbt/sysdep.h>
  17 #include <xbt/log.h>
  18 #include <xbt/dict.h>
  19
  20 #include <simgrid/s4u/host.hpp>
  21
  22 #include <mc/mc.h>
  23
  24 #include "smx_private.h"
  25 #include "src/kernel/activity/SynchroIo.hpp"
  26 #include "src/kernel/activity/SynchroRaw.hpp"
  27 #include "src/kernel/activity/SynchroSleep.hpp"
  28 #include "src/mc/mc_replay.h"
  29 #include "src/mc/remote/Client.hpp"
  30 #include "src/msg/msg_private.h"
  31 #include "src/surf/cpu_interface.hpp"
  32 #include "src/surf/surf_interface.hpp"
  33
  34 #ifdef HAVE_SMPI
  35 #include "src/smpi/private.h"
  36 #endif
  37
  38 XBT_LOG_NEW_DEFAULT_SUBCATEGORY(simix_process, simix, "Logging specific to SIMIX (process)");
  39
  40 unsigned long simix_process_maxpid = 0;
  41
  42 /** Increase the refcount for this process */
  43 smx_actor_t SIMIX_process_ref(smx_actor_t process)
  44 {
  45   if (process != nullptr)
  46     intrusive_ptr_add_ref(process);
  47   return process;
  48 }
  49
  50 /** Decrease the refcount for this process */
  51 void SIMIX_process_unref(smx_actor_t process)
  52 {
  53   if (process != nullptr)
  54     intrusive_ptr_release(process);
  55 }
  56
  57 /**
  58  * \brief Returns the current agent.
  59  *
  60  * This functions returns the currently running SIMIX process.
  61  *
  62  * \return The SIMIX process
  63  */
  64 smx_actor_t SIMIX_process_self()
  65 {
  66   smx_context_t self_context = SIMIX_context_self();
  67
  68   return (self_context != nullptr) ? self_context->process() : nullptr;
  69 }
  70
  71 /**
  72  * \brief Returns whether a process has pending asynchronous communications.
  73  * \return true if there are asynchronous communications in this process
  74  */
  75 int SIMIX_process_has_pending_comms(smx_actor_t process) {
  76
  77   return xbt_fifo_size(process->comms) > 0;
  78 }
  79
  80 /**
  81  * \brief Moves a process to the list of processes to destroy.
  82  */
  83 void SIMIX_process_cleanup(smx_actor_t process)
  84 {
  85   XBT_DEBUG("Cleanup process %s (%p), waiting synchro %p",
  86       process->name.c_str(), process, process->waiting_synchro);
  87
  88   process->finished = true;
  89   SIMIX_process_on_exit_runall(process);
  90
  91   /* Unregister from the kill timer if any */
  92   if (process->kill_timer != nullptr)
  93       SIMIX_timer_remove(process->kill_timer);
  94
  95   xbt_os_mutex_acquire(simix_global->mutex);
  96
  97   /* cancel non-blocking communications */
  98   smx_activity_t synchro = static_cast<smx_activity_t>(xbt_fifo_pop(process->comms));
  99   while (synchro != nullptr) {
 100     simgrid::kernel::activity::Comm *comm = static_cast<simgrid::kernel::activity::Comm*>(synchro);
 101
 102     /* make sure no one will finish the comm after this process is destroyed,
 103      * because src_proc or dst_proc would be an invalid pointer */
 104     comm->cancel();
 105
 106     if (comm->src_proc == process) {
 107       XBT_DEBUG("Found an unfinished send comm %p (detached = %d), state %d, src = %p, dst = %p",
 108           comm, comm->detached, (int)comm->state, comm->src_proc, comm->dst_proc);
 109       comm->src_proc = nullptr;
 110
 111       /* I'm not supposed to destroy a detached comm from the sender side, */
 112       if (comm->detached)
 113         XBT_DEBUG("Don't destroy it since it's a detached comm and I'm the sender");
 114       else
 115         comm->unref();
 116
 117     }
 118     else if (comm->dst_proc == process){
 119       XBT_DEBUG("Found an unfinished recv comm %p, state %d, src = %p, dst = %p",
 120           comm, (int)comm->state, comm->src_proc, comm->dst_proc);
 121       comm->dst_proc = nullptr;
 122
 123       if (comm->detached && comm->src_proc != nullptr) {
 124         /* the comm will be freed right now, remove it from the sender */
 125         xbt_fifo_remove(comm->src_proc->comms, comm);
 126       }
 127
 128       comm->unref();
 129     } else {
 130       xbt_die("Communication synchro %p is in my list but I'm not the sender nor the receiver", synchro);
 131     }
 132     synchro = static_cast<smx_activity_t>(xbt_fifo_pop(process->comms));
 133   }
 134
 135   XBT_DEBUG("%p should not be run anymore",process);
 136   simix_global->process_list.erase(process->pid);
 137   if (process->host)
 138     xbt_swag_remove(process, process->host->extension<simgrid::simix::Host>()->process_list);
 139   xbt_swag_insert(process, simix_global->process_to_destroy);
 140   process->context->iwannadie = 0;
 141
 142   xbt_os_mutex_release(simix_global->mutex);
 143 }
 144
 145 /**
 146  * Garbage collection
 147  *
 148  * Should be called some time to time to free the memory allocated for processes
 149  * that have finished (or killed).
 150  */
 151 void SIMIX_process_empty_trash()
 152 {
 153   smx_actor_t process = nullptr;
 154
 155   while ((process = (smx_actor_t) xbt_swag_extract(simix_global->process_to_destroy))) {
 156     XBT_DEBUG("Getting rid of %p",process);
 157     intrusive_ptr_release(process);
 158   }
 159 }
 160
 161 namespace simgrid {
 162 namespace simix {
 163
 164 ActorImpl::~ActorImpl()
 165 {
 166   delete this->context;
 167   if (this->properties)
 168     xbt_dict_free(&this->properties);
 169   if (this->comms != nullptr)
 170     xbt_fifo_free(this->comms);
 171   if (this->on_exit)
 172     xbt_dynar_free(&this->on_exit);
 173 }
 174
 175 void create_maestro(std::function<void()> code)
 176 {
 177   smx_actor_t maestro = nullptr;
 178   /* Create maestro process and initialize it */
 179   maestro = new simgrid::simix::ActorImpl();
 180   maestro->pid = simix_process_maxpid++;
 181   maestro->name = "";
 182   maestro->data = nullptr;
 183
 184   if (!code) {
 185     maestro->context = SIMIX_context_new(std::function<void()>(), nullptr, maestro);
 186   } else {
 187     if (!simix_global)
 188       xbt_die("simix is not initialized, please call MSG_init first");
 189     maestro->context =
 190       simix_global->context_factory->create_maestro(code, maestro);
 191   }
 192
 193   maestro->simcall.issuer = maestro;
 194   simix_global->maestro_process = maestro;
 195 }
 196
 197 }
 198 }
 199
 200 /** @brief Creates and runs the maestro process */
 201 void SIMIX_maestro_create(void (*code)(void*), void* data)
 202 {
 203   simgrid::simix::create_maestro(std::bind(code, data));
 204 }
 205
 206 /**
 207  * \brief Internal function to create a process.
 208  *
 209  * This function actually creates the process.
 210  * It may be called when a SIMCALL_PROCESS_CREATE simcall occurs,
 211  * or directly for SIMIX internal purposes. The sure thing is that it's called from maestro context.
 212  *
 213  * \return the process created
 214  */
 215 smx_actor_t SIMIX_process_create(
 216                           const char *name,
 217                           std::function<void()> code,
 218                           void *data,
 219                           sg_host_t host,
 220                           xbt_dict_t properties,
 221                           smx_actor_t parent_process)
 222 {
 223
 224   XBT_DEBUG("Start process %s on host '%s'", name, host->cname());
 225
 226   if (host->isOff()) {
 227     XBT_WARN("Cannot launch process '%s' on failed host '%s'", name, host->cname());
 228     return nullptr;
 229   }
 230
 231   smx_actor_t process = new simgrid::simix::ActorImpl();
 232
 233   xbt_assert(code && host != nullptr, "Invalid parameters");
 234   /* Process data */
 235   process->pid            = simix_process_maxpid++;
 236   process->name           = simgrid::xbt::string(name);
 237   process->host           = host;
 238   process->data           = data;
 239   process->simcall.issuer = process;
 240
 241   if (parent_process != nullptr) {
 242     process->ppid = parent_process->pid;
 243 /* SMPI process have their own data segment and each other inherit from their father */
 244 #if HAVE_SMPI
 245     if (smpi_privatize_global_variables) {
 246       if (parent_process->pid != 0) {
 247         SIMIX_segment_index_set(process, parent_process->segment_index);
 248       } else {
 249         SIMIX_segment_index_set(process, process->pid - 1);
 250       }
 251     }
 252 #endif
 253   }
 254
 255   process->code         = code;
 256
 257   XBT_VERB("Create context %s", process->name.c_str());
 258   process->context = SIMIX_context_new(std::move(code), simix_global->cleanup_process_function, process);
 259
 260   /* Add properties */
 261   process->properties = properties;
 262
 263   /* Add the process to it's host process list */
 264   xbt_swag_insert(process, host->extension<simgrid::simix::Host>()->process_list);
 265
 266   XBT_DEBUG("Start context '%s'", process->name.c_str());
 267
 268   /* Now insert it in the global process list and in the process to run list */
 269   simix_global->process_list[process->pid] = process;
 270   XBT_DEBUG("Inserting %s(%s) in the to_run list", process->cname(), host->cname());
 271   xbt_dynar_push_as(simix_global->process_to_run, smx_actor_t, process);
 272
 273   /* Tracing the process creation */
 274   TRACE_msg_process_create(process->cname(), process->pid, process->host);
 275
 276   return process;
 277 }
 278
 279 smx_actor_t SIMIX_process_attach(const char* name, void* data, const char* hostname, xbt_dict_t properties,
 280                                  smx_actor_t parent_process)
 281 {
 282   // This is mostly a copy/paste from SIMIX_process_new(),
 283   // it'd be nice to share some code between those two functions.
 284
 285   sg_host_t host = sg_host_by_name(hostname);
 286   XBT_DEBUG("Attach process %s on host '%s'", name, hostname);
 287
 288   if (host->isOff()) {
 289     XBT_WARN("Cannot launch process '%s' on failed host '%s'",
 290       name, hostname);
 291     return nullptr;
 292   }
 293
 294   smx_actor_t process = new simgrid::simix::ActorImpl();
 295   /* Process data */
 296   process->pid = simix_process_maxpid++;
 297   process->name = std::string(name);
 298   process->host = host;
 299   process->data = data;
 300   process->simcall.issuer = process;
 301
 302   if (parent_process != nullptr) {
 303     process->ppid = parent_process->pid;
 304     /* SMPI process have their own data segment and each other inherit from their father */
 305 #if HAVE_SMPI
 306     if (smpi_privatize_global_variables) {
 307       if (parent_process->pid != 0) {
 308         SIMIX_segment_index_set(process, parent_process->segment_index);
 309       } else {
 310         SIMIX_segment_index_set(process, process->pid - 1);
 311       }
 312     }
 313 #endif
 314   }
 315
 316   /* Process data for auto-restart */
 317   process->code = nullptr;
 318
 319   XBT_VERB("Create context %s", process->name.c_str());
 320   if (!simix_global)
 321     xbt_die("simix is not initialized, please call MSG_init first");
 322   process->context = simix_global->context_factory->attach(
 323     simix_global->cleanup_process_function, process);
 324
 325   /* Add properties */
 326   process->properties = properties;
 327
 328   /* Add the process to it's host process list */
 329   xbt_swag_insert(process, host->extension<simgrid::simix::Host>()->process_list);
 330
 331   /* Now insert it in the global process list and in the process to run list */
 332   simix_global->process_list[process->pid] = process;
 333   XBT_DEBUG("Inserting %s(%s) in the to_run list", process->cname(), host->cname());
 334   xbt_dynar_push_as(simix_global->process_to_run, smx_actor_t, process);
 335
 336   /* Tracing the process creation */
 337   TRACE_msg_process_create(process->cname(), process->pid, process->host);
 338
 339   auto context = dynamic_cast<simgrid::kernel::context::AttachContext*>(process->context);
 340   if (!context)
 341     xbt_die("Not a suitable context");
 342
 343   context->attach_start();
 344   return process;
 345 }
 346
 347 void SIMIX_process_detach()
 348 {
 349   auto context = dynamic_cast<simgrid::kernel::context::AttachContext*>(SIMIX_context_self());
 350   if (!context)
 351     xbt_die("Not a suitable context");
 352
 353   simix_global->cleanup_process_function(context->process());
 354
 355   // Let maestro ignore we are still alive:
 356   // xbt_swag_remove(context->process(), simix_global->process_list);
 357
 358   // TODO, Remove from proces list:
 359   //   xbt_swag_remove(process, sg_host_simix(host)->process_list);
 360
 361   context->attach_stop();
 362   // delete context;
 363 }
 364
 365 /**
 366  * \brief Executes the processes from simix_global->process_to_run.
 367  *
 368  * The processes of simix_global->process_to_run are run (in parallel if
 369  * possible).  On exit, simix_global->process_to_run is empty, and
 370  * simix_global->process_that_ran contains the list of processes that just ran.
 371  * The two lists are swapped so, be careful when using them before and after a
 372  * call to this function.
 373  */
 374 void SIMIX_process_runall()
 375 {
 376   SIMIX_context_runall();
 377
 378   xbt_dynar_t tmp = simix_global->process_that_ran;
 379   simix_global->process_that_ran = simix_global->process_to_run;
 380   simix_global->process_to_run = tmp;
 381   xbt_dynar_reset(simix_global->process_to_run);
 382 }
 383
 384 void simcall_HANDLER_process_kill(smx_simcall_t simcall, smx_actor_t process) {
 385   SIMIX_process_kill(process, simcall->issuer);
 386 }
 387 /**
 388  * \brief Internal function to kill a SIMIX process.
 389  *
 390  * This function may be called when a SIMCALL_PROCESS_KILL simcall occurs,
 391  * or directly for SIMIX internal purposes.
 392  *
 393  * \param process poor victim
 394  * \param issuer the process which has sent the PROCESS_KILL. Important to not schedule twice the same process.
 395  */
 396 void SIMIX_process_kill(smx_actor_t process, smx_actor_t issuer) {
 397
 398   XBT_DEBUG("Killing process %s@%s", process->cname(), process->host->cname());
 399
 400   process->context->iwannadie = 1;
 401   process->blocked = 0;
 402   process->suspended = 0;
 403   process->exception = nullptr;
 404
 405   /* destroy the blocking synchro if any */
 406   if (process->waiting_synchro) {
 407
 408     simgrid::kernel::activity::Exec *exec = dynamic_cast<simgrid::kernel::activity::Exec*>(process->waiting_synchro);
 409     simgrid::kernel::activity::Comm *comm = dynamic_cast<simgrid::kernel::activity::Comm*>(process->waiting_synchro);
 410     simgrid::kernel::activity::Sleep *sleep = dynamic_cast<simgrid::kernel::activity::Sleep*>(process->waiting_synchro);
 411     simgrid::kernel::activity::Raw *raw = dynamic_cast<simgrid::kernel::activity::Raw*>(process->waiting_synchro);
 412     simgrid::kernel::activity::Io *io = dynamic_cast<simgrid::kernel::activity::Io*>(process->waiting_synchro);
 413
 414     if (exec != nullptr) {
 415       exec->unref();
 416
 417     } else if (comm != nullptr) {
 418       xbt_fifo_remove(process->comms, process->waiting_synchro);
 419       comm->cancel();
 420
 421       // Remove first occurrence of &process->simcall:
 422       auto i = boost::range::find(
 423         process->waiting_synchro->simcalls,
 424         &process->simcall);
 425       if (i != process->waiting_synchro->simcalls.end())
 426         process->waiting_synchro->simcalls.remove(&process->simcall);
 427
 428       comm->unref();
 429
 430     } else if (sleep != nullptr) {
 431       SIMIX_process_sleep_destroy(process->waiting_synchro);
 432
 433     } else if (raw != nullptr) {
 434       SIMIX_synchro_stop_waiting(process, &process->simcall);
 435       delete process->waiting_synchro;
 436
 437     } else if (io != nullptr) {
 438       SIMIX_io_destroy(process->waiting_synchro);
 439     }
 440
 441     /*
 442     switch (process->waiting_synchro->type) {
 443     case SIMIX_SYNC_JOIN:
 444       SIMIX_process_sleep_destroy(process->waiting_synchro);
 445       break;
 446     } */
 447
 448     process->waiting_synchro = nullptr;
 449   }
 450   if(!xbt_dynar_member(simix_global->process_to_run, &(process)) && process != issuer) {
 451     XBT_DEBUG("Inserting %s in the to_run list", process->name.c_str());
 452     xbt_dynar_push_as(simix_global->process_to_run, smx_actor_t, process);
 453   }
 454 }
 455
 456 /** @brief Ask another process to raise the given exception
 457  *
 458  * @param process The process that should raise that exception
 459  * @param cat category of exception
 460  * @param value value associated to the exception
 461  * @param msg string information associated to the exception
 462  */
 463 void SIMIX_process_throw(smx_actor_t process, xbt_errcat_t cat, int value, const char *msg) {
 464   SMX_EXCEPTION(process, cat, value, msg);
 465
 466   if (process->suspended)
 467     SIMIX_process_resume(process);
 468
 469   /* cancel the blocking synchro if any */
 470   if (process->waiting_synchro) {
 471
 472     simgrid::kernel::activity::Exec *exec = dynamic_cast<simgrid::kernel::activity::Exec*>(process->waiting_synchro);
 473     if (exec != nullptr) {
 474       SIMIX_execution_cancel(process->waiting_synchro);
 475     }
 476
 477     simgrid::kernel::activity::Comm *comm = dynamic_cast<simgrid::kernel::activity::Comm*>(process->waiting_synchro);
 478     if (comm != nullptr) {
 479       xbt_fifo_remove(process->comms, comm);
 480       comm->cancel();
 481     }
 482
 483     simgrid::kernel::activity::Sleep *sleep = dynamic_cast<simgrid::kernel::activity::Sleep*>(process->waiting_synchro);
 484     if (sleep != nullptr) {
 485       SIMIX_process_sleep_destroy(process->waiting_synchro);
 486       if (!xbt_dynar_member(simix_global->process_to_run, &(process)) && process != SIMIX_process_self()) {
 487         XBT_DEBUG("Inserting %s in the to_run list", process->name.c_str());
 488         xbt_dynar_push_as(simix_global->process_to_run, smx_actor_t, process);
 489       }
 490     }
 491
 492     simgrid::kernel::activity::Raw *raw = dynamic_cast<simgrid::kernel::activity::Raw*>(process->waiting_synchro);
 493     if (raw != nullptr) {
 494       SIMIX_synchro_stop_waiting(process, &process->simcall);
 495     }
 496
 497     simgrid::kernel::activity::Io *io = dynamic_cast<simgrid::kernel::activity::Io*>(process->waiting_synchro);
 498     if (io != nullptr) {
 499       SIMIX_io_destroy(process->waiting_synchro);
 500     }
 501   }
 502   process->waiting_synchro = nullptr;
 503
 504 }
 505
 506 void simcall_HANDLER_process_killall(smx_simcall_t simcall, int reset_pid) {
 507   SIMIX_process_killall(simcall->issuer, reset_pid);
 508 }
 509 /**
 510  * \brief Kills all running processes.
 511  * \param issuer this one will not be killed
 512  */
 513 void SIMIX_process_killall(smx_actor_t issuer, int reset_pid)
 514 {
 515   for (auto kv : simix_global->process_list)
 516     if (kv.second != issuer)
 517       SIMIX_process_kill(kv.second, issuer);
 518
 519   if (reset_pid > 0)
 520     simix_process_maxpid = reset_pid;
 521
 522   SIMIX_context_runall();
 523
 524   SIMIX_process_empty_trash();
 525 }
 526
 527 void simcall_HANDLER_process_set_host(smx_simcall_t simcall, smx_actor_t process, sg_host_t dest)
 528 {
 529   process->new_host = dest;
 530 }
 531 void SIMIX_process_change_host(smx_actor_t process, sg_host_t dest)
 532 {
 533   xbt_assert((process != nullptr), "Invalid parameters");
 534   xbt_swag_remove(process, process->host->extension<simgrid::simix::Host>()->process_list);
 535   process->host = dest;
 536   xbt_swag_insert(process, dest->extension<simgrid::simix::Host>()->process_list);
 537 }
 538
 539
 540 void simcall_HANDLER_process_suspend(smx_simcall_t simcall, smx_actor_t process)
 541 {
 542   smx_activity_t sync_suspend = SIMIX_process_suspend(process, simcall->issuer);
 543
 544   if (process != simcall->issuer) {
 545     SIMIX_simcall_answer(simcall);
 546   } else {
 547     sync_suspend->simcalls.push_back(simcall);
 548     process->waiting_synchro = sync_suspend;
 549     process->waiting_synchro->suspend();
 550   }
 551   /* If we are suspending ourselves, then just do not finish the simcall now */
 552 }
 553
 554 smx_activity_t SIMIX_process_suspend(smx_actor_t process, smx_actor_t issuer)
 555 {
 556   if (process->suspended) {
 557     XBT_DEBUG("Process '%s' is already suspended", process->name.c_str());
 558     return nullptr;
 559   }
 560
 561   process->suspended = 1;
 562
 563   /* If we are suspending another process that is waiting on a sync, suspend its synchronization. */
 564   if (process != issuer) {
 565
 566     if (process->waiting_synchro)
 567       process->waiting_synchro->suspend();
 568     /* If the other process is not waiting, its suspension is delayed to when the process is rescheduled. */
 569
 570     return nullptr;
 571   } else {
 572     return SIMIX_execution_start(process, "suspend", 0.0, 1.0, 0.0);
 573   }
 574 }
 575
 576 void SIMIX_process_resume(smx_actor_t process)
 577 {
 578   XBT_IN("process = %p", process);
 579
 580   if(process->context->iwannadie) {
 581     XBT_VERB("Ignoring request to suspend a process that is currently dying.");
 582     return;
 583   }
 584
 585   if(!process->suspended) return;
 586   process->suspended = 0;
 587
 588   /* resume the synchronization that was blocking the resumed process. */
 589   if (process->waiting_synchro)
 590     process->waiting_synchro->resume();
 591
 592   XBT_OUT();
 593 }
 594
 595 int SIMIX_process_get_maxpid() {
 596   return simix_process_maxpid;
 597 }
 598
 599 int SIMIX_process_count()
 600 {
 601   return simix_global->process_list.size();
 602 }
 603
 604 int SIMIX_process_get_PID(smx_actor_t self)
 605 {
 606   if (self == nullptr)
 607     return 0;
 608   else
 609     return self->pid;
 610 }
 611
 612 void* SIMIX_process_self_get_data()
 613 {
 614   smx_actor_t self = SIMIX_process_self();
 615
 616   if (!self) {
 617     return nullptr;
 618   }
 619   return self->data;
 620 }
 621
 622 void SIMIX_process_self_set_data(void *data)
 623 {
 624   smx_actor_t self = SIMIX_process_self();
 625
 626   SIMIX_process_set_data(self, data);
 627 }
 628
 629 void SIMIX_process_set_data(smx_actor_t process, void *data)
 630 {
 631   process->data = data;
 632 }
 633
 634 /* needs to be public and without simcall because it is called
 635    by exceptions and logging events */
 636 const char* SIMIX_process_self_get_name() {
 637
 638   smx_actor_t process = SIMIX_process_self();
 639   if (process == nullptr || process == simix_global->maestro_process)
 640     return "maestro";
 641
 642   return process->name.c_str();
 643 }
 644
 645 smx_actor_t SIMIX_process_get_by_name(const char* name)
 646 {
 647   for (auto kv : simix_global->process_list)
 648     if (kv.second->name == name)
 649       return kv.second;
 650   return nullptr;
 651 }
 652
 653 int SIMIX_process_is_suspended(smx_actor_t process)
 654 {
 655   return process->suspended;
 656 }
 657
 658 xbt_dict_t SIMIX_process_get_properties(smx_actor_t process)
 659 {
 660   return process->properties;
 661 }
 662
 663 void simcall_HANDLER_process_join(smx_simcall_t simcall, smx_actor_t process, double timeout)
 664 {
 665   if (process->finished) {
 666     // The joined process is already finished, just wake up the issuer process right away
 667     simcall_process_sleep__set__result(simcall, SIMIX_DONE);
 668     SIMIX_simcall_answer(simcall);
 669     return;
 670   }
 671   smx_activity_t sync = SIMIX_process_join(simcall->issuer, process, timeout);
 672   sync->simcalls.push_back(simcall);
 673   simcall->issuer->waiting_synchro = sync;
 674 }
 675
 676 static int SIMIX_process_join_finish(smx_process_exit_status_t status, smx_activity_t synchro){
 677   simgrid::kernel::activity::Sleep *sleep = static_cast<simgrid::kernel::activity::Sleep*>(synchro);
 678
 679   if (sleep->surf_sleep) {
 680     sleep->surf_sleep->cancel();
 681
 682     while (!sleep->simcalls.empty()) {
 683       smx_simcall_t simcall = sleep->simcalls.front();
 684       sleep->simcalls.pop_front();
 685       simcall_process_sleep__set__result(simcall, SIMIX_DONE);
 686       simcall->issuer->waiting_synchro = nullptr;
 687       if (simcall->issuer->suspended) {
 688         XBT_DEBUG("Wait! This process is suspended and can't wake up now.");
 689         simcall->issuer->suspended = 0;
 690         simcall_HANDLER_process_suspend(simcall, simcall->issuer);
 691       } else {
 692         SIMIX_simcall_answer(simcall);
 693       }
 694     }
 695     sleep->surf_sleep->unref();
 696     sleep->surf_sleep = nullptr;
 697   }
 698   sleep->unref();
 699   return 0;
 700 }
 701
 702 smx_activity_t SIMIX_process_join(smx_actor_t issuer, smx_actor_t process, double timeout)
 703 {
 704   smx_activity_t res = SIMIX_process_sleep(issuer, timeout);
 705   static_cast<simgrid::kernel::activity::ActivityImpl*>(res)->ref();
 706   SIMIX_process_on_exit(process, (int_f_pvoid_pvoid_t)SIMIX_process_join_finish, res);
 707   return res;
 708 }
 709
 710 void simcall_HANDLER_process_sleep(smx_simcall_t simcall, double duration)
 711 {
 712   if (MC_is_active() || MC_record_replay_is_active()) {
 713     MC_process_clock_add(simcall->issuer, duration);
 714     simcall_process_sleep__set__result(simcall, SIMIX_DONE);
 715     SIMIX_simcall_answer(simcall);
 716     return;
 717   }
 718   smx_activity_t sync = SIMIX_process_sleep(simcall->issuer, duration);
 719   sync->simcalls.push_back(simcall);
 720   simcall->issuer->waiting_synchro = sync;
 721 }
 722
 723 smx_activity_t SIMIX_process_sleep(smx_actor_t process, double duration)
 724 {
 725   sg_host_t host = process->host;
 726
 727   if (host->isOff())
 728     THROWF(host_error, 0, "Host %s failed, you cannot sleep there.", host->cname());
 729
 730   simgrid::kernel::activity::Sleep *synchro = new simgrid::kernel::activity::Sleep();
 731   synchro->host = host;
 732   synchro->surf_sleep                       = host->pimpl_cpu->sleep(duration);
 733   synchro->surf_sleep->setData(synchro);
 734   XBT_DEBUG("Create sleep synchronization %p", synchro);
 735
 736   return synchro;
 737 }
 738
 739 void SIMIX_process_sleep_destroy(smx_activity_t synchro)
 740 {
 741   XBT_DEBUG("Destroy synchro %p", synchro);
 742   simgrid::kernel::activity::Sleep *sleep = static_cast<simgrid::kernel::activity::Sleep*>(synchro);
 743
 744   if (sleep->surf_sleep) {
 745     sleep->surf_sleep->unref();
 746     sleep->surf_sleep = nullptr;
 747     sleep->unref();
 748   }
 749 }
 750
 751 /**
 752  * \brief Calling this function makes the process to yield.
 753  *
 754  * Only the current process can call this function, giving back the control to
 755  * maestro.
 756  *
 757  * \param self the current process
 758  */
 759 void SIMIX_process_yield(smx_actor_t self)
 760 {
 761   XBT_DEBUG("Yield actor '%s'", self->cname());
 762
 763   /* Go into sleep and return control to maestro */
 764   self->context->suspend();
 765
 766   /* Ok, maestro returned control to us */
 767   XBT_DEBUG("Control returned to me: '%s'", self->name.c_str());
 768
 769   if (self->new_host) {
 770     SIMIX_process_change_host(self, self->new_host);
 771     self->new_host = nullptr;
 772   }
 773
 774   if (self->context->iwannadie){
 775     XBT_DEBUG("I wanna die!");
 776     self->finished = true;
 777     /* execute the on_exit functions */
 778     SIMIX_process_on_exit_runall(self);
 779     /* Add the process to the list of process to restart, only if the host is down */
 780     if (self->auto_restart && self->host->isOff()) {
 781       SIMIX_host_add_auto_restart_process(self->host, self->cname(),
 782                                           self->code, self->data,
 783                                           SIMIX_timer_get_date(self->kill_timer),
 784                                           self->properties,
 785                                           self->auto_restart);
 786     }
 787     XBT_DEBUG("Process %s@%s is dead", self->cname(), self->host->cname());
 788     self->context->stop();
 789   }
 790
 791   if (self->suspended) {
 792     XBT_DEBUG("Hey! I'm suspended.");
 793     xbt_assert(self->exception != nullptr, "Gasp! This exception may be lost by subsequent calls.");
 794     self->suspended = 0;
 795     SIMIX_process_suspend(self, self);
 796   }
 797
 798   if (self->exception != nullptr) {
 799     XBT_DEBUG("Wait, maestro left me an exception");
 800     std::exception_ptr exception = std::move(self->exception);
 801     self->exception = nullptr;
 802     std::rethrow_exception(std::move(exception));
 803   }
 804
 805   if(SMPI_switch_data_segment && self->segment_index != -1){
 806     SMPI_switch_data_segment(self->segment_index);
 807   }
 808 }
 809
 810 /* callback: termination */
 811 void SIMIX_process_exception_terminate(xbt_ex_t * e)
 812 {
 813   xbt_ex_display(e);
 814   xbt_abort();
 815 }
 816
 817 smx_context_t SIMIX_process_get_context(smx_actor_t p) {
 818   return p->context;
 819 }
 820
 821 void SIMIX_process_set_context(smx_actor_t p,smx_context_t c) {
 822   p->context = c;
 823 }
 824
 825 /**
 826  * \brief Returns the list of processes to run.
 827  */
 828 xbt_dynar_t SIMIX_process_get_runnable()
 829 {
 830   return simix_global->process_to_run;
 831 }
 832
 833 /**
 834  * \brief Returns the process from PID.
 835  */
 836 smx_actor_t SIMIX_process_from_PID(int PID)
 837 {
 838   if (simix_global->process_list.find(PID) == simix_global->process_list.end())
 839     return nullptr;
 840   return simix_global->process_list.at(PID);
 841 }
 842
 843 /** @brief returns a dynar containing all currently existing processes */
 844 xbt_dynar_t SIMIX_processes_as_dynar() {
 845   xbt_dynar_t res = xbt_dynar_new(sizeof(smx_actor_t),nullptr);
 846   for (auto kv : simix_global->process_list) {
 847     smx_actor_t proc = kv.second;
 848     xbt_dynar_push(res,&proc);
 849   }
 850   return res;
 851 }
 852
 853 void SIMIX_process_on_exit_runall(smx_actor_t process) {
 854   s_smx_process_exit_fun_t exit_fun;
 855   smx_process_exit_status_t exit_status = (process->context->iwannadie) ? SMX_EXIT_FAILURE : SMX_EXIT_SUCCESS;
 856   while (!xbt_dynar_is_empty(process->on_exit)) {
 857     exit_fun = xbt_dynar_pop_as(process->on_exit,s_smx_process_exit_fun_t);
 858     (exit_fun.fun)((void*)exit_status, exit_fun.arg);
 859   }
 860 }
 861
 862 void SIMIX_process_on_exit(smx_actor_t process, int_f_pvoid_pvoid_t fun, void *data) {
 863   xbt_assert(process, "current process not found: are you in maestro context ?");
 864
 865   if (!process->on_exit) {
 866     process->on_exit = xbt_dynar_new(sizeof(s_smx_process_exit_fun_t), nullptr);
 867   }
 868
 869   s_smx_process_exit_fun_t exit_fun = {fun, data};
 870
 871   xbt_dynar_push_as(process->on_exit,s_smx_process_exit_fun_t,exit_fun);
 872 }
 873
 874 /**
 875  * \brief Sets the auto-restart status of the process.
 876  * If set to 1, the process will be automatically restarted when its host
 877  * comes back.
 878  */
 879 void SIMIX_process_auto_restart_set(smx_actor_t process, int auto_restart) {
 880   process->auto_restart = auto_restart;
 881 }
 882
 883 smx_actor_t simcall_HANDLER_process_restart(smx_simcall_t simcall, smx_actor_t process) {
 884   return SIMIX_process_restart(process, simcall->issuer);
 885 }
 886 /** @brief Restart a process, starting it again from the beginning. */
 887 smx_actor_t SIMIX_process_restart(smx_actor_t process, smx_actor_t issuer) {
 888   XBT_DEBUG("Restarting process %s on %s", process->cname(), process->host->cname());
 889
 890   //retrieve the arguments of the old process
 891   //FIXME: Factorize this with SIMIX_host_add_auto_restart_process ?
 892   simgrid::simix::ProcessArg arg;
 893   arg.name = process->name;
 894   arg.code = process->code;
 895   arg.host = process->host;
 896   arg.kill_time = SIMIX_timer_get_date(process->kill_timer);
 897   arg.data = process->data;
 898   arg.properties = nullptr;
 899   arg.auto_restart = process->auto_restart;
 900
 901   //kill the old process
 902   SIMIX_process_kill(process, issuer);
 903
 904   //start the new process
 905   smx_actor_t actor = simix_global->create_process_function(arg.name.c_str(), std::move(arg.code), arg.data, arg.host,
 906                                                             arg.properties, nullptr);
 907   if (arg.kill_time >= 0)
 908     simcall_process_set_kill_time(actor, arg.kill_time);
 909   if (arg.auto_restart)
 910     simcall_process_auto_restart_set(actor, arg.auto_restart);
 911
 912   return actor;
 913 }
 914
 915 void SIMIX_segment_index_set(smx_actor_t proc, int index){
 916   proc->segment_index = index;
 917 }
 918
 919 /**
 920  * \ingroup simix_process_management
 921  * \brief Creates and runs a new SIMIX process.
 922  *
 923  * The structure and the corresponding thread are created and put in the list of ready processes.
 924  *
 925  * \param name a name for the process. It is for user-level information and can be nullptr.
 926  * \param code the main function of the process
 927  * \param data a pointer to any data one may want to attach to the new object. It is for user-level information and can be nullptr.
 928  * It can be retrieved with the function \ref simcall_process_get_data.
 929  * \param host where the new agent is executed.
 930  * \param kill_time time when the process is killed
 931  * \param argc first argument passed to \a code
 932  * \param argv second argument passed to \a code
 933  * \param properties the properties of the process
 934  * \param auto_restart either it is autorestarting or not.
 935  */
 936 smx_actor_t simcall_process_create(const char* name, xbt_main_func_t code, void* data, sg_host_t host, int argc,
 937                                    char** argv, xbt_dict_t properties)
 938 {
 939   if (name == nullptr)
 940     name = "";
 941   auto wrapped_code = simgrid::xbt::wrapMain(code, argc, argv);
 942   for (int i = 0; i != argc; ++i)
 943     xbt_free(argv[i]);
 944   xbt_free(argv);
 945   smx_actor_t res = simcall_process_create(name, std::move(wrapped_code), data, host, properties);
 946   return res;
 947 }
 948
 949 smx_actor_t simcall_process_create(const char* name, std::function<void()> code, void* data, sg_host_t host,
 950                                    xbt_dict_t properties)
 951 {
 952   if (name == nullptr)
 953     name = "";
 954   smx_actor_t self = SIMIX_process_self();
 955   return simgrid::simix::kernelImmediate([name, code, data, host, properties, self] {
 956     return SIMIX_process_create(name, std::move(code), data, host, properties, self);
 957   });
 958 }