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