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->ppid = -1;
 182   maestro->name = "";
 183   maestro->data = nullptr;
 184
 185   if (!code) {
 186     maestro->context = SIMIX_context_new(std::function<void()>(), nullptr, maestro);
 187   } else {
 188     if (!simix_global)
 189       xbt_die("simix is not initialized, please call MSG_init first");
 190     maestro->context =
 191       simix_global->context_factory->create_maestro(code, maestro);
 192   }
 193
 194   maestro->simcall.issuer = maestro;
 195   simix_global->maestro_process = maestro;
 196 }
 197
 198 }
 199 }
 200
 201 /** @brief Creates and runs the maestro process */
 202 void SIMIX_maestro_create(void (*code)(void*), void* data)
 203 {
 204   simgrid::simix::create_maestro(std::bind(code, data));
 205 }
 206
 207 /**
 208  * \brief Internal function to create a process.
 209  *
 210  * This function actually creates the process.
 211  * It may be called when a SIMCALL_PROCESS_CREATE simcall occurs,
 212  * or directly for SIMIX internal purposes. The sure thing is that it's called from maestro context.
 213  *
 214  * \return the process created
 215  */
 216 smx_actor_t SIMIX_process_create(
 217                           const char *name,
 218                           std::function<void()> code,
 219                           void *data,
 220                           sg_host_t host,
 221                           double kill_time,
 222                           xbt_dict_t properties,
 223                           int auto_restart,
 224                           smx_actor_t parent_process)
 225 {
 226   smx_actor_t process = nullptr;
 227
 228   XBT_DEBUG("Start process %s on host '%s'", name, host->cname());
 229
 230   if (host->isOff()) {
 231     XBT_WARN("Cannot launch process '%s' on failed host '%s'", name, host->cname());
 232     return nullptr;
 233   }
 234   else {
 235     process = new simgrid::simix::ActorImpl();
 236
 237     xbt_assert(code && host != nullptr, "Invalid parameters");
 238     /* Process data */
 239     process->pid = simix_process_maxpid++;
 240     process->name = simgrid::xbt::string(name);
 241     process->host = host;
 242     process->data = data;
 243     process->comms = xbt_fifo_new();
 244     process->simcall.issuer = process;
 245     /* Initiliaze data segment to default value */
 246     SIMIX_segment_index_set(process, -1);
 247
 248     if (parent_process != nullptr) {
 249       process->ppid = parent_process->pid;
 250       /* SMPI process have their own data segment and each other inherit from their father */
 251 #if HAVE_SMPI
 252        if( smpi_privatize_global_variables) {
 253          if (parent_process->pid != 0) {
 254            SIMIX_segment_index_set(process, parent_process->segment_index);
 255          } else {
 256            SIMIX_segment_index_set(process, process->pid - 1);
 257          }
 258        }
 259 #endif
 260      } else {
 261        process->ppid = -1;
 262      }
 263
 264     /* Process data for auto-restart */
 265     process->auto_restart = auto_restart;
 266     process->code = code;
 267
 268     XBT_VERB("Create context %s", process->name.c_str());
 269     process->context = SIMIX_context_new(
 270       std::move(code),
 271       simix_global->cleanup_process_function, process);
 272
 273     /* Add properties */
 274     process->properties = properties;
 275
 276     /* Add the process to it's host process list */
 277     xbt_swag_insert(process, host->extension<simgrid::simix::Host>()->process_list);
 278
 279     XBT_DEBUG("Start context '%s'", process->name.c_str());
 280
 281     /* Now insert it in the global process list and in the process to run list */
 282     simix_global->process_list[process->pid] = process;
 283     XBT_DEBUG("Inserting %s(%s) in the to_run list", process->cname(), host->cname());
 284     xbt_dynar_push_as(simix_global->process_to_run, smx_actor_t, process);
 285
 286     if (kill_time > SIMIX_get_clock() && simix_global->kill_process_function) {
 287       XBT_DEBUG("Process %s(%s) will be kill at time %f", process->cname(), process->host->cname(), kill_time);
 288       process->kill_timer = SIMIX_timer_set(kill_time, [=]() {
 289         simix_global->kill_process_function(process);
 290       });
 291     }
 292
 293     /* Tracing the process creation */
 294     TRACE_msg_process_create(process->cname(), process->pid, process->host);
 295   }
 296   return process;
 297 }
 298
 299 smx_actor_t SIMIX_process_attach(
 300   const char* name,
 301   void *data,
 302   const char* hostname,
 303   xbt_dict_t properties,
 304   smx_actor_t parent_process)
 305 {
 306   // This is mostly a copy/paste from SIMIX_process_new(),
 307   // it'd be nice to share some code between those two functions.
 308
 309   sg_host_t host = sg_host_by_name(hostname);
 310   XBT_DEBUG("Attach process %s on host '%s'", name, hostname);
 311
 312   if (host->isOff()) {
 313     XBT_WARN("Cannot launch process '%s' on failed host '%s'",
 314       name, hostname);
 315     return nullptr;
 316   }
 317
 318   smx_actor_t process = new simgrid::simix::ActorImpl();
 319   /* Process data */
 320   process->pid = simix_process_maxpid++;
 321   process->name = std::string(name);
 322   process->host = host;
 323   process->data = data;
 324   process->comms = xbt_fifo_new();
 325   process->simcall.issuer = process;
 326   process->ppid = -1;
 327   /* Initiliaze data segment to default value */
 328   SIMIX_segment_index_set(process, -1);
 329   if (parent_process != nullptr) {
 330     process->ppid = parent_process->pid;
 331     /* SMPI process have their own data segment and each other inherit from their father */
 332 #if HAVE_SMPI
 333     if (smpi_privatize_global_variables) {
 334       if (parent_process->pid != 0) {
 335         SIMIX_segment_index_set(process, parent_process->segment_index);
 336       } else {
 337         SIMIX_segment_index_set(process, process->pid - 1);
 338       }
 339     }
 340 #endif
 341   }
 342
 343   /* Process data for auto-restart */
 344   process->auto_restart = false;
 345   process->code = nullptr;
 346
 347   XBT_VERB("Create context %s", process->name.c_str());
 348   if (!simix_global)
 349     xbt_die("simix is not initialized, please call MSG_init first");
 350   process->context = simix_global->context_factory->attach(
 351     simix_global->cleanup_process_function, process);
 352
 353   /* Add properties */
 354   process->properties = properties;
 355
 356   /* Add the process to it's host process list */
 357   xbt_swag_insert(process, host->extension<simgrid::simix::Host>()->process_list);
 358
 359   /* Now insert it in the global process list and in the process to run list */
 360   simix_global->process_list[process->pid] = process;
 361   XBT_DEBUG("Inserting %s(%s) in the to_run list", process->cname(), host->cname());
 362   xbt_dynar_push_as(simix_global->process_to_run, smx_actor_t, process);
 363
 364   /* Tracing the process creation */
 365   TRACE_msg_process_create(process->cname(), process->pid, process->host);
 366
 367   auto context = dynamic_cast<simgrid::kernel::context::AttachContext*>(process->context);
 368   if (!context)
 369     xbt_die("Not a suitable context");
 370
 371   context->attach_start();
 372   return process;
 373 }
 374
 375 void SIMIX_process_detach()
 376 {
 377   auto context = dynamic_cast<simgrid::kernel::context::AttachContext*>(SIMIX_context_self());
 378   if (!context)
 379     xbt_die("Not a suitable context");
 380
 381   simix_global->cleanup_process_function(context->process());
 382
 383   // Let maestro ignore we are still alive:
 384   // xbt_swag_remove(context->process(), simix_global->process_list);
 385
 386   // TODO, Remove from proces list:
 387   //   xbt_swag_remove(process, sg_host_simix(host)->process_list);
 388
 389   context->attach_stop();
 390   // delete context;
 391 }
 392
 393 /**
 394  * \brief Executes the processes from simix_global->process_to_run.
 395  *
 396  * The processes of simix_global->process_to_run are run (in parallel if
 397  * possible).  On exit, simix_global->process_to_run is empty, and
 398  * simix_global->process_that_ran contains the list of processes that just ran.
 399  * The two lists are swapped so, be careful when using them before and after a
 400  * call to this function.
 401  */
 402 void SIMIX_process_runall()
 403 {
 404   SIMIX_context_runall();
 405
 406   xbt_dynar_t tmp = simix_global->process_that_ran;
 407   simix_global->process_that_ran = simix_global->process_to_run;
 408   simix_global->process_to_run = tmp;
 409   xbt_dynar_reset(simix_global->process_to_run);
 410 }
 411
 412 void simcall_HANDLER_process_kill(smx_simcall_t simcall, smx_actor_t process) {
 413   SIMIX_process_kill(process, simcall->issuer);
 414 }
 415 /**
 416  * \brief Internal function to kill a SIMIX process.
 417  *
 418  * This function may be called when a SIMCALL_PROCESS_KILL simcall occurs,
 419  * or directly for SIMIX internal purposes.
 420  *
 421  * \param process poor victim
 422  * \param issuer the process which has sent the PROCESS_KILL. Important to not schedule twice the same process.
 423  */
 424 void SIMIX_process_kill(smx_actor_t process, smx_actor_t issuer) {
 425
 426   XBT_DEBUG("Killing process %s@%s", process->cname(), process->host->cname());
 427
 428   process->context->iwannadie = 1;
 429   process->blocked = 0;
 430   process->suspended = 0;
 431   process->exception = nullptr;
 432
 433   /* destroy the blocking synchro if any */
 434   if (process->waiting_synchro) {
 435
 436     simgrid::kernel::activity::Exec *exec = dynamic_cast<simgrid::kernel::activity::Exec*>(process->waiting_synchro);
 437     simgrid::kernel::activity::Comm *comm = dynamic_cast<simgrid::kernel::activity::Comm*>(process->waiting_synchro);
 438     simgrid::kernel::activity::Sleep *sleep = dynamic_cast<simgrid::kernel::activity::Sleep*>(process->waiting_synchro);
 439     simgrid::kernel::activity::Raw *raw = dynamic_cast<simgrid::kernel::activity::Raw*>(process->waiting_synchro);
 440     simgrid::kernel::activity::Io *io = dynamic_cast<simgrid::kernel::activity::Io*>(process->waiting_synchro);
 441
 442     if (exec != nullptr) {
 443       exec->unref();
 444
 445     } else if (comm != nullptr) {
 446       xbt_fifo_remove(process->comms, process->waiting_synchro);
 447       comm->cancel();
 448
 449       // Remove first occurrence of &process->simcall:
 450       auto i = boost::range::find(
 451         process->waiting_synchro->simcalls,
 452         &process->simcall);
 453       if (i != process->waiting_synchro->simcalls.end())
 454         process->waiting_synchro->simcalls.remove(&process->simcall);
 455
 456       comm->unref();
 457
 458     } else if (sleep != nullptr) {
 459       SIMIX_process_sleep_destroy(process->waiting_synchro);
 460
 461     } else if (raw != nullptr) {
 462       SIMIX_synchro_stop_waiting(process, &process->simcall);
 463       delete process->waiting_synchro;
 464
 465     } else if (io != nullptr) {
 466       SIMIX_io_destroy(process->waiting_synchro);
 467     }
 468
 469     /*
 470     switch (process->waiting_synchro->type) {
 471     case SIMIX_SYNC_JOIN:
 472       SIMIX_process_sleep_destroy(process->waiting_synchro);
 473       break;
 474     } */
 475
 476     process->waiting_synchro = nullptr;
 477   }
 478   if(!xbt_dynar_member(simix_global->process_to_run, &(process)) && process != issuer) {
 479     XBT_DEBUG("Inserting %s in the to_run list", process->name.c_str());
 480     xbt_dynar_push_as(simix_global->process_to_run, smx_actor_t, process);
 481   }
 482 }
 483
 484 /** @brief Ask another process to raise the given exception
 485  *
 486  * @param process The process that should raise that exception
 487  * @param cat category of exception
 488  * @param value value associated to the exception
 489  * @param msg string information associated to the exception
 490  */
 491 void SIMIX_process_throw(smx_actor_t process, xbt_errcat_t cat, int value, const char *msg) {
 492   SMX_EXCEPTION(process, cat, value, msg);
 493
 494   if (process->suspended)
 495     SIMIX_process_resume(process);
 496
 497   /* cancel the blocking synchro if any */
 498   if (process->waiting_synchro) {
 499
 500     simgrid::kernel::activity::Exec *exec = dynamic_cast<simgrid::kernel::activity::Exec*>(process->waiting_synchro);
 501     if (exec != nullptr) {
 502       SIMIX_execution_cancel(process->waiting_synchro);
 503     }
 504
 505     simgrid::kernel::activity::Comm *comm = dynamic_cast<simgrid::kernel::activity::Comm*>(process->waiting_synchro);
 506     if (comm != nullptr) {
 507       xbt_fifo_remove(process->comms, comm);
 508       comm->cancel();
 509     }
 510
 511     simgrid::kernel::activity::Sleep *sleep = dynamic_cast<simgrid::kernel::activity::Sleep*>(process->waiting_synchro);
 512     if (sleep != nullptr) {
 513       SIMIX_process_sleep_destroy(process->waiting_synchro);
 514       if (!xbt_dynar_member(simix_global->process_to_run, &(process)) && process != SIMIX_process_self()) {
 515         XBT_DEBUG("Inserting %s in the to_run list", process->name.c_str());
 516         xbt_dynar_push_as(simix_global->process_to_run, smx_actor_t, process);
 517       }
 518     }
 519
 520     simgrid::kernel::activity::Raw *raw = dynamic_cast<simgrid::kernel::activity::Raw*>(process->waiting_synchro);
 521     if (raw != nullptr) {
 522       SIMIX_synchro_stop_waiting(process, &process->simcall);
 523     }
 524
 525     simgrid::kernel::activity::Io *io = dynamic_cast<simgrid::kernel::activity::Io*>(process->waiting_synchro);
 526     if (io != nullptr) {
 527       SIMIX_io_destroy(process->waiting_synchro);
 528     }
 529   }
 530   process->waiting_synchro = nullptr;
 531
 532 }
 533
 534 void simcall_HANDLER_process_killall(smx_simcall_t simcall, int reset_pid) {
 535   SIMIX_process_killall(simcall->issuer, reset_pid);
 536 }
 537 /**
 538  * \brief Kills all running processes.
 539  * \param issuer this one will not be killed
 540  */
 541 void SIMIX_process_killall(smx_actor_t issuer, int reset_pid)
 542 {
 543   for (auto kv : simix_global->process_list)
 544     if (kv.second != issuer)
 545       SIMIX_process_kill(kv.second, issuer);
 546
 547   if (reset_pid > 0)
 548     simix_process_maxpid = reset_pid;
 549
 550   SIMIX_context_runall();
 551
 552   SIMIX_process_empty_trash();
 553 }
 554
 555 void simcall_HANDLER_process_set_host(smx_simcall_t simcall, smx_actor_t process, sg_host_t dest)
 556 {
 557   process->new_host = dest;
 558 }
 559 void SIMIX_process_change_host(smx_actor_t process, sg_host_t dest)
 560 {
 561   xbt_assert((process != nullptr), "Invalid parameters");
 562   xbt_swag_remove(process, process->host->extension<simgrid::simix::Host>()->process_list);
 563   process->host = dest;
 564   xbt_swag_insert(process, dest->extension<simgrid::simix::Host>()->process_list);
 565 }
 566
 567
 568 void simcall_HANDLER_process_suspend(smx_simcall_t simcall, smx_actor_t process)
 569 {
 570   smx_activity_t sync_suspend = SIMIX_process_suspend(process, simcall->issuer);
 571
 572   if (process != simcall->issuer) {
 573     SIMIX_simcall_answer(simcall);
 574   } else {
 575     sync_suspend->simcalls.push_back(simcall);
 576     process->waiting_synchro = sync_suspend;
 577     process->waiting_synchro->suspend();
 578   }
 579   /* If we are suspending ourselves, then just do not finish the simcall now */
 580 }
 581
 582 smx_activity_t SIMIX_process_suspend(smx_actor_t process, smx_actor_t issuer)
 583 {
 584   if (process->suspended) {
 585     XBT_DEBUG("Process '%s' is already suspended", process->name.c_str());
 586     return nullptr;
 587   }
 588
 589   process->suspended = 1;
 590
 591   /* If we are suspending another process that is waiting on a sync, suspend its synchronization. */
 592   if (process != issuer) {
 593
 594     if (process->waiting_synchro)
 595       process->waiting_synchro->suspend();
 596     /* If the other process is not waiting, its suspension is delayed to when the process is rescheduled. */
 597
 598     return nullptr;
 599   } else {
 600     return SIMIX_execution_start(process, "suspend", 0.0, 1.0, 0.0);
 601   }
 602 }
 603
 604 void SIMIX_process_resume(smx_actor_t process)
 605 {
 606   XBT_IN("process = %p", process);
 607
 608   if(process->context->iwannadie) {
 609     XBT_VERB("Ignoring request to suspend a process that is currently dying.");
 610     return;
 611   }
 612
 613   if(!process->suspended) return;
 614   process->suspended = 0;
 615
 616   /* resume the synchronization that was blocking the resumed process. */
 617   if (process->waiting_synchro)
 618     process->waiting_synchro->resume();
 619
 620   XBT_OUT();
 621 }
 622
 623 int SIMIX_process_get_maxpid() {
 624   return simix_process_maxpid;
 625 }
 626
 627 int SIMIX_process_count()
 628 {
 629   return simix_global->process_list.size();
 630 }
 631
 632 int SIMIX_process_get_PID(smx_actor_t self)
 633 {
 634   if (self == nullptr)
 635     return 0;
 636   else
 637     return self->pid;
 638 }
 639
 640 void* SIMIX_process_self_get_data()
 641 {
 642   smx_actor_t self = SIMIX_process_self();
 643
 644   if (!self) {
 645     return nullptr;
 646   }
 647   return SIMIX_process_get_data(self);
 648 }
 649
 650 void SIMIX_process_self_set_data(void *data)
 651 {
 652   smx_actor_t self = SIMIX_process_self();
 653
 654   SIMIX_process_set_data(self, data);
 655 }
 656
 657 void* SIMIX_process_get_data(smx_actor_t process)
 658 {
 659   return process->data;
 660 }
 661
 662 void SIMIX_process_set_data(smx_actor_t process, void *data)
 663 {
 664   process->data = data;
 665 }
 666
 667 /* needs to be public and without simcall because it is called
 668    by exceptions and logging events */
 669 const char* SIMIX_process_self_get_name() {
 670
 671   smx_actor_t process = SIMIX_process_self();
 672   if (process == nullptr || process == simix_global->maestro_process)
 673     return "maestro";
 674
 675   return process->name.c_str();
 676 }
 677
 678 smx_actor_t SIMIX_process_get_by_name(const char* name)
 679 {
 680   for (auto kv : simix_global->process_list)
 681     if (kv.second->name == name)
 682       return kv.second;
 683   return nullptr;
 684 }
 685
 686 int SIMIX_process_is_suspended(smx_actor_t process)
 687 {
 688   return process->suspended;
 689 }
 690
 691 xbt_dict_t SIMIX_process_get_properties(smx_actor_t process)
 692 {
 693   return process->properties;
 694 }
 695
 696 void simcall_HANDLER_process_join(smx_simcall_t simcall, smx_actor_t process, double timeout)
 697 {
 698   if (process->finished) {
 699     // The joined process is already finished, just wake up the issuer process right away
 700     simcall_process_sleep__set__result(simcall, SIMIX_DONE);
 701     SIMIX_simcall_answer(simcall);
 702     return;
 703   }
 704   smx_activity_t sync = SIMIX_process_join(simcall->issuer, process, timeout);
 705   sync->simcalls.push_back(simcall);
 706   simcall->issuer->waiting_synchro = sync;
 707 }
 708
 709 static int SIMIX_process_join_finish(smx_process_exit_status_t status, smx_activity_t synchro){
 710   simgrid::kernel::activity::Sleep *sleep = static_cast<simgrid::kernel::activity::Sleep*>(synchro);
 711
 712   if (sleep->surf_sleep) {
 713     sleep->surf_sleep->cancel();
 714
 715     while (!sleep->simcalls.empty()) {
 716       smx_simcall_t simcall = sleep->simcalls.front();
 717       sleep->simcalls.pop_front();
 718       simcall_process_sleep__set__result(simcall, SIMIX_DONE);
 719       simcall->issuer->waiting_synchro = nullptr;
 720       if (simcall->issuer->suspended) {
 721         XBT_DEBUG("Wait! This process is suspended and can't wake up now.");
 722         simcall->issuer->suspended = 0;
 723         simcall_HANDLER_process_suspend(simcall, simcall->issuer);
 724       } else {
 725         SIMIX_simcall_answer(simcall);
 726       }
 727     }
 728     sleep->surf_sleep->unref();
 729     sleep->surf_sleep = nullptr;
 730   }
 731   sleep->unref();
 732   return 0;
 733 }
 734
 735 smx_activity_t SIMIX_process_join(smx_actor_t issuer, smx_actor_t process, double timeout)
 736 {
 737   smx_activity_t res = SIMIX_process_sleep(issuer, timeout);
 738   static_cast<simgrid::kernel::activity::ActivityImpl*>(res)->ref();
 739   SIMIX_process_on_exit(process, (int_f_pvoid_pvoid_t)SIMIX_process_join_finish, res);
 740   return res;
 741 }
 742
 743 void simcall_HANDLER_process_sleep(smx_simcall_t simcall, double duration)
 744 {
 745   if (MC_is_active() || MC_record_replay_is_active()) {
 746     MC_process_clock_add(simcall->issuer, duration);
 747     simcall_process_sleep__set__result(simcall, SIMIX_DONE);
 748     SIMIX_simcall_answer(simcall);
 749     return;
 750   }
 751   smx_activity_t sync = SIMIX_process_sleep(simcall->issuer, duration);
 752   sync->simcalls.push_back(simcall);
 753   simcall->issuer->waiting_synchro = sync;
 754 }
 755
 756 smx_activity_t SIMIX_process_sleep(smx_actor_t process, double duration)
 757 {
 758   sg_host_t host = process->host;
 759
 760   if (host->isOff())
 761     THROWF(host_error, 0, "Host %s failed, you cannot sleep there.", host->cname());
 762
 763   simgrid::kernel::activity::Sleep *synchro = new simgrid::kernel::activity::Sleep();
 764   synchro->host = host;
 765   synchro->surf_sleep                       = host->pimpl_cpu->sleep(duration);
 766   synchro->surf_sleep->setData(synchro);
 767   XBT_DEBUG("Create sleep synchronization %p", synchro);
 768
 769   return synchro;
 770 }
 771
 772 void SIMIX_process_sleep_destroy(smx_activity_t synchro)
 773 {
 774   XBT_DEBUG("Destroy synchro %p", synchro);
 775   simgrid::kernel::activity::Sleep *sleep = static_cast<simgrid::kernel::activity::Sleep*>(synchro);
 776
 777   if (sleep->surf_sleep) {
 778     sleep->surf_sleep->unref();
 779     sleep->surf_sleep = nullptr;
 780     sleep->unref();
 781   }
 782 }
 783
 784 /**
 785  * \brief Calling this function makes the process to yield.
 786  *
 787  * Only the current process can call this function, giving back the control to
 788  * maestro.
 789  *
 790  * \param self the current process
 791  */
 792 void SIMIX_process_yield(smx_actor_t self)
 793 {
 794   XBT_DEBUG("Yield actor '%s'", self->cname());
 795
 796   /* Go into sleep and return control to maestro */
 797   self->context->suspend();
 798
 799   /* Ok, maestro returned control to us */
 800   XBT_DEBUG("Control returned to me: '%s'", self->name.c_str());
 801
 802   if (self->new_host) {
 803     SIMIX_process_change_host(self, self->new_host);
 804     self->new_host = nullptr;
 805   }
 806
 807   if (self->context->iwannadie){
 808     XBT_DEBUG("I wanna die!");
 809     self->finished = true;
 810     /* execute the on_exit functions */
 811     SIMIX_process_on_exit_runall(self);
 812     /* Add the process to the list of process to restart, only if the host is down */
 813     if (self->auto_restart && self->host->isOff()) {
 814       SIMIX_host_add_auto_restart_process(self->host, self->cname(),
 815                                           self->code, self->data,
 816                                           SIMIX_timer_get_date(self->kill_timer),
 817                                           self->properties,
 818                                           self->auto_restart);
 819     }
 820     XBT_DEBUG("Process %s@%s is dead", self->cname(), self->host->cname());
 821     self->context->stop();
 822   }
 823
 824   if (self->suspended) {
 825     XBT_DEBUG("Hey! I'm suspended.");
 826     xbt_assert(self->exception != nullptr, "Gasp! This exception may be lost by subsequent calls.");
 827     self->suspended = 0;
 828     SIMIX_process_suspend(self, self);
 829   }
 830
 831   if (self->exception != nullptr) {
 832     XBT_DEBUG("Wait, maestro left me an exception");
 833     std::exception_ptr exception = std::move(self->exception);
 834     self->exception = nullptr;
 835     std::rethrow_exception(std::move(exception));
 836   }
 837
 838   if(SMPI_switch_data_segment && self->segment_index != -1){
 839     SMPI_switch_data_segment(self->segment_index);
 840   }
 841 }
 842
 843 /* callback: termination */
 844 void SIMIX_process_exception_terminate(xbt_ex_t * e)
 845 {
 846   xbt_ex_display(e);
 847   xbt_abort();
 848 }
 849
 850 smx_context_t SIMIX_process_get_context(smx_actor_t p) {
 851   return p->context;
 852 }
 853
 854 void SIMIX_process_set_context(smx_actor_t p,smx_context_t c) {
 855   p->context = c;
 856 }
 857
 858 /**
 859  * \brief Returns the list of processes to run.
 860  */
 861 xbt_dynar_t SIMIX_process_get_runnable()
 862 {
 863   return simix_global->process_to_run;
 864 }
 865
 866 /**
 867  * \brief Returns the process from PID.
 868  */
 869 smx_actor_t SIMIX_process_from_PID(int PID)
 870 {
 871   if (simix_global->process_list.find(PID) == simix_global->process_list.end())
 872     return nullptr;
 873   return simix_global->process_list.at(PID);
 874 }
 875
 876 /** @brief returns a dynar containing all currently existing processes */
 877 xbt_dynar_t SIMIX_processes_as_dynar() {
 878   xbt_dynar_t res = xbt_dynar_new(sizeof(smx_actor_t),nullptr);
 879   for (auto kv : simix_global->process_list) {
 880     smx_actor_t proc = kv.second;
 881     xbt_dynar_push(res,&proc);
 882   }
 883   return res;
 884 }
 885
 886 void SIMIX_process_on_exit_runall(smx_actor_t process) {
 887   s_smx_process_exit_fun_t exit_fun;
 888   smx_process_exit_status_t exit_status = (process->context->iwannadie) ? SMX_EXIT_FAILURE : SMX_EXIT_SUCCESS;
 889   while (!xbt_dynar_is_empty(process->on_exit)) {
 890     exit_fun = xbt_dynar_pop_as(process->on_exit,s_smx_process_exit_fun_t);
 891     (exit_fun.fun)((void*)exit_status, exit_fun.arg);
 892   }
 893 }
 894
 895 void SIMIX_process_on_exit(smx_actor_t process, int_f_pvoid_pvoid_t fun, void *data) {
 896   xbt_assert(process, "current process not found: are you in maestro context ?");
 897
 898   if (!process->on_exit) {
 899     process->on_exit = xbt_dynar_new(sizeof(s_smx_process_exit_fun_t), nullptr);
 900   }
 901
 902   s_smx_process_exit_fun_t exit_fun = {fun, data};
 903
 904   xbt_dynar_push_as(process->on_exit,s_smx_process_exit_fun_t,exit_fun);
 905 }
 906
 907 /**
 908  * \brief Sets the auto-restart status of the process.
 909  * If set to 1, the process will be automatically restarted when its host
 910  * comes back.
 911  */
 912 void SIMIX_process_auto_restart_set(smx_actor_t process, int auto_restart) {
 913   process->auto_restart = auto_restart;
 914 }
 915
 916 smx_actor_t simcall_HANDLER_process_restart(smx_simcall_t simcall, smx_actor_t process) {
 917   return SIMIX_process_restart(process, simcall->issuer);
 918 }
 919 /** @brief Restart a process, starting it again from the beginning. */
 920 smx_actor_t SIMIX_process_restart(smx_actor_t process, smx_actor_t issuer) {
 921   XBT_DEBUG("Restarting process %s on %s", process->cname(), process->host->cname());
 922
 923   //retrieve the arguments of the old process
 924   //FIXME: Factorize this with SIMIX_host_add_auto_restart_process ?
 925   simgrid::simix::ProcessArg arg;
 926   arg.name = process->name;
 927   arg.code = process->code;
 928   arg.host = process->host;
 929   arg.kill_time = SIMIX_timer_get_date(process->kill_timer);
 930   arg.data = process->data;
 931   arg.properties = nullptr;
 932   arg.auto_restart = process->auto_restart;
 933
 934   //kill the old process
 935   SIMIX_process_kill(process, issuer);
 936
 937   //start the new process
 938   return simix_global->create_process_function(arg.name.c_str(), std::move(arg.code), arg.data, arg.host,
 939       arg.kill_time, arg.properties, arg.auto_restart, nullptr);
 940 }
 941
 942 void SIMIX_segment_index_set(smx_actor_t proc, int index){
 943   proc->segment_index = index;
 944 }
 945
 946 /**
 947  * \ingroup simix_process_management
 948  * \brief Creates and runs a new SIMIX process.
 949  *
 950  * The structure and the corresponding thread are created and put in the list of ready processes.
 951  *
 952  * \param name a name for the process. It is for user-level information and can be nullptr.
 953  * \param code the main function of the process
 954  * \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.
 955  * It can be retrieved with the function \ref simcall_process_get_data.
 956  * \param host where the new agent is executed.
 957  * \param kill_time time when the process is killed
 958  * \param argc first argument passed to \a code
 959  * \param argv second argument passed to \a code
 960  * \param properties the properties of the process
 961  * \param auto_restart either it is autorestarting or not.
 962  */
 963 smx_actor_t simcall_process_create(const char *name,
 964                               xbt_main_func_t code,
 965                               void *data,
 966                               sg_host_t host,
 967                               double kill_time,
 968                               int argc, char **argv,
 969                               xbt_dict_t properties,
 970                               int auto_restart)
 971 {
 972   if (name == nullptr)
 973     name = "";
 974   auto wrapped_code = simgrid::xbt::wrapMain(code, argc, argv);
 975   for (int i = 0; i != argc; ++i)
 976     xbt_free(argv[i]);
 977   xbt_free(argv);
 978   smx_actor_t res = simcall_process_create(name,
 979     std::move(wrapped_code),
 980     data, host, kill_time, properties, auto_restart);
 981   return res;
 982 }
 983
 984 smx_actor_t simcall_process_create(
 985   const char *name, std::function<void()> code, void *data,
 986   sg_host_t host, double kill_time,
 987   xbt_dict_t properties, int auto_restart)
 988 {
 989   if (name == nullptr)
 990     name = "";
 991   smx_actor_t self = SIMIX_process_self();
 992   return simgrid::simix::kernelImmediate([name, code, data, host, kill_time, properties, auto_restart, self] {
 993     return SIMIX_process_create(name, std::move(code), data, host, kill_time, properties, auto_restart, self);
 994   });
 995 }