src/kernel/activity/CommImpl.cpp

   1 /* Copyright (c) 2007-2019. The SimGrid Team. All rights reserved.          */
   2
   3 /* This program is free software; you can redistribute it and/or modify it
   4  * under the terms of the license (GNU LGPL) which comes with this package. */
   5
   6 #include "src/kernel/activity/CommImpl.hpp"
   7 #include "simgrid/Exception.hpp"
   8 #include "simgrid/kernel/resource/Action.hpp"
   9 #include "simgrid/modelchecker.h"
  10 #include "simgrid/s4u/Host.hpp"
  11 #include "src/kernel/activity/MailboxImpl.hpp"
  12 #include "src/kernel/context/Context.hpp"
  13 #include "src/mc/mc_replay.hpp"
  14 #include "src/surf/cpu_interface.hpp"
  15 #include "src/surf/network_interface.hpp"
  16 #include "src/surf/surf_interface.hpp"
  17
  18 #include <boost/range/algorithm.hpp>
  19 XBT_LOG_NEW_DEFAULT_SUBCATEGORY(simix_network, simix, "SIMIX network-related synchronization");
  20
  21 XBT_PRIVATE void simcall_HANDLER_comm_send(smx_simcall_t simcall, smx_actor_t src, smx_mailbox_t mbox, double task_size,
  22                                            double rate, void* src_buff, size_t src_buff_size,
  23                                            int (*match_fun)(void*, void*, simgrid::kernel::activity::CommImpl*),
  24                                            void (*copy_data_fun)(simgrid::kernel::activity::CommImpl*, void*, size_t),
  25                                            void* data, double timeout)
  26 {
  27   smx_activity_t comm = simcall_HANDLER_comm_isend(simcall, src, mbox, task_size, rate, src_buff, src_buff_size,
  28                                                    match_fun, nullptr, copy_data_fun, data, 0);
  29   SIMCALL_SET_MC_VALUE(simcall, 0);
  30   simcall_HANDLER_comm_wait(simcall, comm, timeout);
  31 }
  32
  33 XBT_PRIVATE smx_activity_t simcall_HANDLER_comm_isend(
  34     smx_simcall_t /*simcall*/, smx_actor_t src_proc, smx_mailbox_t mbox, double task_size, double rate, void* src_buff,
  35     size_t src_buff_size, int (*match_fun)(void*, void*, simgrid::kernel::activity::CommImpl*),
  36     void (*clean_fun)(void*), // used to free the synchro in case of problem after a detached send
  37     void (*copy_data_fun)(simgrid::kernel::activity::CommImpl*, void*, size_t), // used to copy data if not default one
  38     void* data, int detached)
  39 {
  40   XBT_DEBUG("send from mailbox %p", mbox);
  41
  42   /* Prepare a synchro describing us, so that it gets passed to the user-provided filter of other side */
  43   simgrid::kernel::activity::CommImplPtr this_comm = simgrid::kernel::activity::CommImplPtr(
  44       new simgrid::kernel::activity::CommImpl(simgrid::kernel::activity::CommImpl::Type::SEND));
  45
  46   /* Look for communication synchro matching our needs. We also provide a description of
  47    * ourself so that the other side also gets a chance of choosing if it wants to match with us.
  48    *
  49    * If it is not found then push our communication into the rendez-vous point */
  50   simgrid::kernel::activity::CommImplPtr other_comm =
  51       mbox->find_matching_comm(simgrid::kernel::activity::CommImpl::Type::RECEIVE, match_fun, data, this_comm,
  52                                /*done*/ false, /*remove_matching*/ true);
  53
  54   if (not other_comm) {
  55     other_comm = std::move(this_comm);
  56
  57     if (mbox->permanent_receiver_ != nullptr) {
  58       // this mailbox is for small messages, which have to be sent right now
  59       other_comm->state_     = SIMIX_READY;
  60       other_comm->dst_actor_ = mbox->permanent_receiver_.get();
  61       mbox->done_comm_queue_.push_back(other_comm);
  62       XBT_DEBUG("pushing a message into the permanent receive list %p, comm %p", mbox, other_comm.get());
  63
  64     } else {
  65       mbox->push(other_comm);
  66     }
  67   } else {
  68     XBT_DEBUG("Receive already pushed");
  69
  70     other_comm->state_ = SIMIX_READY;
  71     other_comm->type   = simgrid::kernel::activity::CommImpl::Type::READY;
  72   }
  73   src_proc->comms.push_back(other_comm);
  74
  75   if (detached) {
  76     other_comm->detached  = true;
  77     other_comm->clean_fun = clean_fun;
  78   } else {
  79     other_comm->clean_fun = nullptr;
  80   }
  81
  82   /* Setup the communication synchro */
  83   other_comm->src_actor_     = src_proc;
  84   other_comm->task_size_     = task_size;
  85   other_comm->rate_          = rate;
  86   other_comm->src_buff_      = src_buff;
  87   other_comm->src_buff_size_ = src_buff_size;
  88   other_comm->src_data_      = data;
  89
  90   other_comm->match_fun     = match_fun;
  91   other_comm->copy_data_fun = copy_data_fun;
  92
  93   if (MC_is_active() || MC_record_replay_is_active()) {
  94     other_comm->state_ = SIMIX_RUNNING;
  95     return (detached ? nullptr : other_comm);
  96   }
  97
  98   other_comm->start();
  99
 100   return (detached ? nullptr : other_comm);
 101 }
 102
 103 XBT_PRIVATE void simcall_HANDLER_comm_recv(smx_simcall_t simcall, smx_actor_t receiver, smx_mailbox_t mbox,
 104                                            void* dst_buff, size_t* dst_buff_size,
 105                                            int (*match_fun)(void*, void*, simgrid::kernel::activity::CommImpl*),
 106                                            void (*copy_data_fun)(simgrid::kernel::activity::CommImpl*, void*, size_t),
 107                                            void* data, double timeout, double rate)
 108 {
 109   smx_activity_t comm = simcall_HANDLER_comm_irecv(simcall, receiver, mbox, dst_buff, dst_buff_size, match_fun,
 110                                                    copy_data_fun, data, rate);
 111   SIMCALL_SET_MC_VALUE(simcall, 0);
 112   simcall_HANDLER_comm_wait(simcall, comm, timeout);
 113 }
 114
 115 XBT_PRIVATE smx_activity_t simcall_HANDLER_comm_irecv(
 116     smx_simcall_t /*simcall*/, smx_actor_t receiver, smx_mailbox_t mbox, void* dst_buff, size_t* dst_buff_size,
 117     simix_match_func_t match_fun, void (*copy_data_fun)(simgrid::kernel::activity::CommImpl*, void*, size_t),
 118     void* data, double rate)
 119 {
 120   simgrid::kernel::activity::CommImplPtr this_synchro = simgrid::kernel::activity::CommImplPtr(
 121       new simgrid::kernel::activity::CommImpl(simgrid::kernel::activity::CommImpl::Type::RECEIVE));
 122   XBT_DEBUG("recv from mbox %p. this_synchro=%p", mbox, this_synchro.get());
 123
 124   simgrid::kernel::activity::CommImplPtr other_comm;
 125   // communication already done, get it inside the list of completed comms
 126   if (mbox->permanent_receiver_ != nullptr && not mbox->done_comm_queue_.empty()) {
 127
 128     XBT_DEBUG("We have a comm that has probably already been received, trying to match it, to skip the communication");
 129     // find a match in the list of already received comms
 130     other_comm = mbox->find_matching_comm(simgrid::kernel::activity::CommImpl::Type::SEND, match_fun, data,
 131                                           this_synchro, /*done*/ true,
 132                                           /*remove_matching*/ true);
 133     // if not found, assume the receiver came first, register it to the mailbox in the classical way
 134     if (not other_comm) {
 135       XBT_DEBUG("We have messages in the permanent receive list, but not the one we are looking for, pushing request "
 136                 "into list");
 137       other_comm = std::move(this_synchro);
 138       mbox->push(other_comm);
 139     } else {
 140       if (other_comm->surf_action_ && other_comm->remains() < 1e-12) {
 141         XBT_DEBUG("comm %p has been already sent, and is finished, destroy it", other_comm.get());
 142         other_comm->state_ = SIMIX_DONE;
 143         other_comm->type   = simgrid::kernel::activity::CommImpl::Type::DONE;
 144         other_comm->mbox   = nullptr;
 145       }
 146     }
 147   } else {
 148     /* Prepare a comm describing us, so that it gets passed to the user-provided filter of other side */
 149
 150     /* Look for communication activity matching our needs. We also provide a description of
 151      * ourself so that the other side also gets a chance of choosing if it wants to match with us.
 152      *
 153      * If it is not found then push our communication into the rendez-vous point */
 154     other_comm = mbox->find_matching_comm(simgrid::kernel::activity::CommImpl::Type::SEND, match_fun, data,
 155                                           this_synchro, /*done*/ false,
 156                                           /*remove_matching*/ true);
 157
 158     if (other_comm == nullptr) {
 159       XBT_DEBUG("Receive pushed first (%zu comm enqueued so far)", mbox->comm_queue_.size());
 160       other_comm = std::move(this_synchro);
 161       mbox->push(other_comm);
 162     } else {
 163       XBT_DEBUG("Match my %p with the existing %p", this_synchro.get(), other_comm.get());
 164
 165       other_comm->state_ = SIMIX_READY;
 166       other_comm->type   = simgrid::kernel::activity::CommImpl::Type::READY;
 167     }
 168     receiver->comms.push_back(other_comm);
 169   }
 170
 171   /* Setup communication synchro */
 172   other_comm->dst_actor_     = receiver;
 173   other_comm->dst_buff_      = dst_buff;
 174   other_comm->dst_buff_size_ = dst_buff_size;
 175   other_comm->dst_data_      = data;
 176
 177   if (rate > -1.0 && (other_comm->rate_ < 0.0 || rate < other_comm->rate_))
 178     other_comm->rate_ = rate;
 179
 180   other_comm->match_fun     = match_fun;
 181   other_comm->copy_data_fun = copy_data_fun;
 182
 183   if (MC_is_active() || MC_record_replay_is_active()) {
 184     other_comm->state_ = SIMIX_RUNNING;
 185     return other_comm;
 186   }
 187   other_comm->start();
 188   return other_comm;
 189 }
 190
 191 void simcall_HANDLER_comm_wait(smx_simcall_t simcall, smx_activity_t synchro, double timeout)
 192 {
 193   /* Associate this simcall to the wait synchro */
 194   XBT_DEBUG("simcall_HANDLER_comm_wait, %p", synchro.get());
 195
 196   synchro->simcalls_.push_back(simcall);
 197   simcall->issuer->waiting_synchro = synchro;
 198
 199   if (MC_is_active() || MC_record_replay_is_active()) {
 200     int idx = SIMCALL_GET_MC_VALUE(simcall);
 201     if (idx == 0) {
 202       synchro->state_ = SIMIX_DONE;
 203     } else {
 204       /* If we reached this point, the wait simcall must have a timeout */
 205       /* Otherwise it shouldn't be enabled and executed by the MC */
 206       if (timeout < 0.0)
 207         THROW_IMPOSSIBLE;
 208
 209       simgrid::kernel::activity::CommImplPtr comm =
 210           boost::static_pointer_cast<simgrid::kernel::activity::CommImpl>(synchro);
 211       if (comm->src_actor_ == simcall->issuer)
 212         comm->state_ = SIMIX_SRC_TIMEOUT;
 213       else
 214         comm->state_ = SIMIX_DST_TIMEOUT;
 215     }
 216
 217     boost::static_pointer_cast<simgrid::kernel::activity::CommImpl>(synchro)->finish();
 218     return;
 219   }
 220
 221   /* If the synchro has already finish perform the error handling, */
 222   /* otherwise set up a waiting timeout on the right side          */
 223   if (synchro->state_ != SIMIX_WAITING && synchro->state_ != SIMIX_RUNNING) {
 224     boost::static_pointer_cast<simgrid::kernel::activity::CommImpl>(synchro)->finish();
 225   } else { /* we need a sleep action (even when there is no timeout) to be notified of host failures */
 226     simgrid::kernel::resource::Action* sleep = simcall->issuer->get_host()->pimpl_cpu->sleep(timeout);
 227     sleep->set_data(synchro.get());
 228
 229     simgrid::kernel::activity::CommImplPtr comm =
 230         boost::static_pointer_cast<simgrid::kernel::activity::CommImpl>(synchro);
 231     if (simcall->issuer == comm->src_actor_)
 232       comm->src_timeout_ = sleep;
 233     else
 234       comm->dst_timeout_ = sleep;
 235   }
 236 }
 237
 238 void simcall_HANDLER_comm_test(smx_simcall_t simcall, smx_activity_t synchro)
 239 {
 240   simgrid::kernel::activity::CommImplPtr comm =
 241       boost::static_pointer_cast<simgrid::kernel::activity::CommImpl>(synchro);
 242
 243   int res;
 244
 245   if (MC_is_active() || MC_record_replay_is_active()) {
 246     res = comm->src_actor_ && comm->dst_actor_;
 247     if (res)
 248       synchro->state_ = SIMIX_DONE;
 249   } else {
 250     res = synchro->state_ != SIMIX_WAITING && synchro->state_ != SIMIX_RUNNING;
 251   }
 252
 253   simcall_comm_test__set__result(simcall, res);
 254   if (simcall_comm_test__get__result(simcall)) {
 255     synchro->simcalls_.push_back(simcall);
 256     boost::static_pointer_cast<simgrid::kernel::activity::CommImpl>(synchro)->finish();
 257   } else {
 258     SIMIX_simcall_answer(simcall);
 259   }
 260 }
 261
 262 void simcall_HANDLER_comm_testany(smx_simcall_t simcall, simgrid::kernel::activity::ActivityImplPtr comms[],
 263                                   size_t count)
 264 {
 265   // The default result is -1 -- this means, "nothing is ready".
 266   // It can be changed below, but only if something matches.
 267   simcall_comm_testany__set__result(simcall, -1);
 268
 269   if (MC_is_active() || MC_record_replay_is_active()) {
 270     int idx = SIMCALL_GET_MC_VALUE(simcall);
 271     if (idx == -1) {
 272       SIMIX_simcall_answer(simcall);
 273     } else {
 274       simgrid::kernel::activity::ActivityImplPtr synchro = comms[idx];
 275       simcall_comm_testany__set__result(simcall, idx);
 276       synchro->simcalls_.push_back(simcall);
 277       synchro->state_ = SIMIX_DONE;
 278       boost::static_pointer_cast<simgrid::kernel::activity::CommImpl>(synchro)->finish();
 279     }
 280     return;
 281   }
 282
 283   for (std::size_t i = 0; i != count; ++i) {
 284     simgrid::kernel::activity::ActivityImplPtr synchro = comms[i];
 285     if (synchro->state_ != SIMIX_WAITING && synchro->state_ != SIMIX_RUNNING) {
 286       simcall_comm_testany__set__result(simcall, i);
 287       synchro->simcalls_.push_back(simcall);
 288       boost::static_pointer_cast<simgrid::kernel::activity::CommImpl>(synchro)->finish();
 289       return;
 290     }
 291   }
 292   SIMIX_simcall_answer(simcall);
 293 }
 294
 295 static void SIMIX_waitany_remove_simcall_from_actions(smx_simcall_t simcall)
 296 {
 297   unsigned int cursor  = 0;
 298   xbt_dynar_t synchros = simcall_comm_waitany__get__comms(simcall);
 299
 300   simgrid::kernel::activity::ActivityImpl* ptr;
 301   xbt_dynar_foreach (synchros, cursor, ptr) {
 302     smx_activity_t synchro = simgrid::kernel::activity::ActivityImplPtr(ptr);
 303
 304     // Remove the first occurence of simcall:
 305     auto i = boost::range::find(synchro->simcalls_, simcall);
 306     if (i != synchro->simcalls_.end())
 307       synchro->simcalls_.erase(i);
 308   }
 309 }
 310 void simcall_HANDLER_comm_waitany(smx_simcall_t simcall, xbt_dynar_t synchros, double timeout)
 311 {
 312   if (MC_is_active() || MC_record_replay_is_active()) {
 313     if (timeout > 0.0)
 314       xbt_die("Timeout not implemented for waitany in the model-checker");
 315     int idx                = SIMCALL_GET_MC_VALUE(simcall);
 316     smx_activity_t synchro = xbt_dynar_get_as(synchros, idx, smx_activity_t);
 317     synchro->simcalls_.push_back(simcall);
 318     simcall_comm_waitany__set__result(simcall, idx);
 319     synchro->state_ = SIMIX_DONE;
 320     boost::static_pointer_cast<simgrid::kernel::activity::CommImpl>(synchro)->finish();
 321     return;
 322   }
 323
 324   if (timeout < 0.0) {
 325     simcall->timer = NULL;
 326   } else {
 327     simcall->timer = simgrid::simix::Timer::set(SIMIX_get_clock() + timeout, [simcall]() {
 328       SIMIX_waitany_remove_simcall_from_actions(simcall);
 329       simcall_comm_waitany__set__result(simcall, -1);
 330       SIMIX_simcall_answer(simcall);
 331     });
 332   }
 333
 334   unsigned int cursor;
 335   simgrid::kernel::activity::ActivityImpl* ptr;
 336   xbt_dynar_foreach (synchros, cursor, ptr) {
 337     smx_activity_t synchro = simgrid::kernel::activity::ActivityImplPtr(ptr);
 338     /* associate this simcall to the the synchro */
 339     synchro->simcalls_.push_back(simcall);
 340
 341     /* see if the synchro is already finished */
 342     if (synchro->state_ != SIMIX_WAITING && synchro->state_ != SIMIX_RUNNING) {
 343       boost::static_pointer_cast<simgrid::kernel::activity::CommImpl>(synchro)->finish();
 344       break;
 345     }
 346   }
 347 }
 348
 349 /******************************************************************************/
 350 /*                    SIMIX_comm_copy_data callbacks                       */
 351 /******************************************************************************/
 352 static void (*SIMIX_comm_copy_data_callback)(simgrid::kernel::activity::CommImpl*, void*,
 353                                              size_t) = &SIMIX_comm_copy_pointer_callback;
 354
 355 void SIMIX_comm_copy_buffer_callback(simgrid::kernel::activity::CommImpl* comm, void* buff, size_t buff_size)
 356 {
 357   XBT_DEBUG("Copy the data over");
 358   memcpy(comm->dst_buff_, buff, buff_size);
 359   if (comm->detached) { // if this is a detached send, the source buffer was duplicated by SMPI sender to make the
 360                         // original buffer available to the application ASAP
 361     xbt_free(buff);
 362     comm->src_buff_ = nullptr;
 363   }
 364 }
 365
 366 void SIMIX_comm_set_copy_data_callback(void (*callback)(simgrid::kernel::activity::CommImpl*, void*, size_t))
 367 {
 368   SIMIX_comm_copy_data_callback = callback;
 369 }
 370
 371 void SIMIX_comm_copy_pointer_callback(simgrid::kernel::activity::CommImpl* comm, void* buff, size_t buff_size)
 372 {
 373   xbt_assert((buff_size == sizeof(void*)), "Cannot copy %zu bytes: must be sizeof(void*)", buff_size);
 374   *(void**)(comm->dst_buff_) = buff;
 375 }
 376
 377 namespace simgrid {
 378 namespace kernel {
 379 namespace activity {
 380
 381 CommImpl::CommImpl(CommImpl::Type type) : type(type)
 382 {
 383   state_   = SIMIX_WAITING;
 384   src_data_ = nullptr;
 385   dst_data_ = nullptr;
 386   XBT_DEBUG("Create comm activity %p", this);
 387 }
 388
 389 CommImpl::~CommImpl()
 390 {
 391   XBT_DEBUG("Really free communication %p", this);
 392
 393   cleanupSurf();
 394
 395   if (detached && state_ != SIMIX_DONE) {
 396     /* the communication has failed and was detached:
 397      * we have to free the buffer */
 398     if (clean_fun)
 399       clean_fun(src_buff_);
 400     src_buff_ = nullptr;
 401   }
 402
 403   if (mbox)
 404     mbox->remove(this);
 405 }
 406
 407 /**  @brief Starts the simulation of a communication synchro. */
 408 void CommImpl::start()
 409 {
 410   /* If both the sender and the receiver are already there, start the communication */
 411   if (state_ == SIMIX_READY) {
 412
 413     s4u::Host* sender   = src_actor_->get_host();
 414     s4u::Host* receiver = dst_actor_->get_host();
 415
 416     surf_action_ = surf_network_model->communicate(sender, receiver, task_size_, rate_);
 417     surf_action_->set_data(this);
 418     state_ = SIMIX_RUNNING;
 419
 420     XBT_DEBUG("Starting communication %p from '%s' to '%s' (surf_action: %p)", this, sender->get_cname(),
 421               receiver->get_cname(), surf_action_);
 422
 423     /* If a link is failed, detect it immediately */
 424     if (surf_action_->get_state() == resource::Action::State::FAILED) {
 425       XBT_DEBUG("Communication from '%s' to '%s' failed to start because of a link failure", sender->get_cname(),
 426                 receiver->get_cname());
 427       state_ = SIMIX_LINK_FAILURE;
 428       cleanupSurf();
 429
 430     } else if (src_actor_->is_suspended() || dst_actor_->is_suspended()) {
 431       /* If any of the process is suspended, create the synchro but stop its execution,
 432          it will be restarted when the sender process resume */
 433       if (src_actor_->is_suspended())
 434         XBT_DEBUG("The communication is suspended on startup because src (%s@%s) was suspended since it initiated the "
 435                   "communication",
 436                   src_actor_->get_cname(), src_actor_->get_host()->get_cname());
 437       else
 438         XBT_DEBUG("The communication is suspended on startup because dst (%s@%s) was suspended since it initiated the "
 439                   "communication",
 440                   dst_actor_->get_cname(), dst_actor_->get_host()->get_cname());
 441
 442       surf_action_->suspend();
 443     }
 444   }
 445 }
 446
 447 /** @brief Copy the communication data from the sender's buffer to the receiver's one  */
 448 void CommImpl::copy_data()
 449 {
 450   size_t buff_size = src_buff_size_;
 451   /* If there is no data to copy then return */
 452   if (not src_buff_ || not dst_buff_ || copied)
 453     return;
 454
 455   XBT_DEBUG("Copying comm %p data from %s (%p) -> %s (%p) (%zu bytes)", this,
 456             src_actor_ ? src_actor_->get_host()->get_cname() : "a finished process", src_buff_,
 457             dst_actor_ ? dst_actor_->get_host()->get_cname() : "a finished process", dst_buff_, buff_size);
 458
 459   /* Copy at most dst_buff_size bytes of the message to receiver's buffer */
 460   if (dst_buff_size_)
 461     buff_size = std::min(buff_size, *(dst_buff_size_));
 462
 463   /* Update the receiver's buffer size to the copied amount */
 464   if (dst_buff_size_)
 465     *dst_buff_size_ = buff_size;
 466
 467   if (buff_size > 0) {
 468     if (copy_data_fun)
 469       copy_data_fun(this, src_buff_, buff_size);
 470     else
 471       SIMIX_comm_copy_data_callback(this, src_buff_, buff_size);
 472   }
 473
 474   /* Set the copied flag so we copy data only once */
 475   /* (this function might be called from both communication ends) */
 476   copied = true;
 477 }
 478
 479 void CommImpl::suspend()
 480 {
 481   /* FIXME: shall we suspend also the timeout synchro? */
 482   if (surf_action_)
 483     surf_action_->suspend();
 484   /* if not created yet, the action will be suspended on creation, in CommImpl::start() */
 485 }
 486
 487 void CommImpl::resume()
 488 {
 489   /*FIXME: check what happen with the timeouts */
 490   if (surf_action_)
 491     surf_action_->resume();
 492   /* in the other case, the synchro were not really suspended yet, see CommImpl::suspend() and CommImpl::start() */
 493 }
 494
 495 void CommImpl::cancel()
 496 {
 497   /* if the synchro is a waiting state means that it is still in a mbox so remove from it and delete it */
 498   if (state_ == SIMIX_WAITING) {
 499     mbox->remove(this);
 500     state_ = SIMIX_CANCELED;
 501   } else if (not MC_is_active() /* when running the MC there are no surf actions */
 502              && not MC_record_replay_is_active() && (state_ == SIMIX_READY || state_ == SIMIX_RUNNING)) {
 503     surf_action_->cancel();
 504   }
 505 }
 506
 507 /**  @brief get the amount remaining from the communication */
 508 double CommImpl::remains()
 509 {
 510   return surf_action_->get_remains();
 511 }
 512
 513 /** @brief This is part of the cleanup process, probably an internal command */
 514 void CommImpl::cleanupSurf()
 515 {
 516   if (surf_action_) {
 517     surf_action_->unref();
 518     surf_action_ = nullptr;
 519   }
 520
 521   if (src_timeout_) {
 522     src_timeout_->unref();
 523     src_timeout_ = nullptr;
 524   }
 525
 526   if (dst_timeout_) {
 527     dst_timeout_->unref();
 528     dst_timeout_ = nullptr;
 529   }
 530 }
 531
 532 void CommImpl::post()
 533 {
 534   /* Update synchro state */
 535   if (src_timeout_ && src_timeout_->get_state() == resource::Action::State::FINISHED)
 536     state_ = SIMIX_SRC_TIMEOUT;
 537   else if (dst_timeout_ && dst_timeout_->get_state() == resource::Action::State::FINISHED)
 538     state_ = SIMIX_DST_TIMEOUT;
 539   else if (src_timeout_ && src_timeout_->get_state() == resource::Action::State::FAILED)
 540     state_ = SIMIX_SRC_HOST_FAILURE;
 541   else if (dst_timeout_ && dst_timeout_->get_state() == resource::Action::State::FAILED)
 542     state_ = SIMIX_DST_HOST_FAILURE;
 543   else if (surf_action_ && surf_action_->get_state() == resource::Action::State::FAILED) {
 544     state_ = SIMIX_LINK_FAILURE;
 545   } else
 546     state_ = SIMIX_DONE;
 547
 548   XBT_DEBUG("SIMIX_post_comm: comm %p, state %d, src_proc %p, dst_proc %p, detached: %d", this, (int)state_,
 549             src_actor_.get(), dst_actor_.get(), detached);
 550
 551   /* destroy the surf actions associated with the Simix communication */
 552   cleanupSurf();
 553
 554   /* if there are simcalls associated with the synchro, then answer them */
 555   if (not simcalls_.empty()) {
 556     finish();
 557   }
 558 }
 559
 560 void CommImpl::finish()
 561 {
 562   smx_activity_t synchro = this;
 563   while (not simcalls_.empty()) {
 564     smx_simcall_t simcall = simcalls_.front();
 565     simcalls_.pop_front();
 566
 567     /* If a waitany simcall is waiting for this synchro to finish, then remove it from the other synchros in the waitany
 568      * list. Afterwards, get the position of the actual synchro in the waitany dynar and return it as the result of the
 569      * simcall */
 570
 571     if (simcall->call == SIMCALL_NONE) // FIXME: maybe a better way to handle this case
 572       continue;                        // if process handling comm is killed
 573     if (simcall->call == SIMCALL_COMM_WAITANY) {
 574       SIMIX_waitany_remove_simcall_from_actions(simcall);
 575       if (simcall->timer) {
 576         simcall->timer->remove();
 577         simcall->timer = nullptr;
 578       }
 579       if (not MC_is_active() && not MC_record_replay_is_active())
 580         simcall_comm_waitany__set__result(simcall,
 581                                           xbt_dynar_search(simcall_comm_waitany__get__comms(simcall), &synchro));
 582     }
 583
 584     /* If the synchro is still in a rendez-vous point then remove from it */
 585     if (mbox)
 586       mbox->remove(this);
 587
 588     XBT_DEBUG("CommImpl::finish(): synchro state = %d", static_cast<int>(state_));
 589
 590     /* Check out for errors */
 591
 592     if (not simcall->issuer->get_host()->is_on()) {
 593       simcall->issuer->context_->iwannadie = true;
 594       simcall->issuer->exception_ =
 595           std::make_exception_ptr(simgrid::HostFailureException(XBT_THROW_POINT, "Host failed"));
 596     } else {
 597       switch (state_) {
 598
 599         case SIMIX_DONE:
 600           XBT_DEBUG("Communication %p complete!", this);
 601           copy_data();
 602           break;
 603
 604         case SIMIX_SRC_TIMEOUT:
 605           simcall->issuer->exception_ = std::make_exception_ptr(
 606               simgrid::TimeoutError(XBT_THROW_POINT, "Communication timeouted because of the sender"));
 607           break;
 608
 609         case SIMIX_DST_TIMEOUT:
 610           simcall->issuer->exception_ = std::make_exception_ptr(
 611               simgrid::TimeoutError(XBT_THROW_POINT, "Communication timeouted because of the receiver"));
 612           break;
 613
 614         case SIMIX_SRC_HOST_FAILURE:
 615           if (simcall->issuer == src_actor_)
 616             simcall->issuer->context_->iwannadie = true;
 617           else
 618             simcall->issuer->exception_ =
 619                 std::make_exception_ptr(simgrid::NetworkFailureException(XBT_THROW_POINT, "Remote peer failed"));
 620           break;
 621
 622         case SIMIX_DST_HOST_FAILURE:
 623           if (simcall->issuer == dst_actor_)
 624             simcall->issuer->context_->iwannadie = true;
 625           else
 626             simcall->issuer->exception_ =
 627                 std::make_exception_ptr(simgrid::NetworkFailureException(XBT_THROW_POINT, "Remote peer failed"));
 628           break;
 629
 630         case SIMIX_LINK_FAILURE:
 631           XBT_DEBUG("Link failure in synchro %p between '%s' and '%s': posting an exception to the issuer: %s (%p) "
 632                     "detached:%d",
 633                     this, src_actor_ ? src_actor_->get_host()->get_cname() : nullptr,
 634                     dst_actor_ ? dst_actor_->get_host()->get_cname() : nullptr, simcall->issuer->get_cname(),
 635                     simcall->issuer, detached);
 636           if (src_actor_ == simcall->issuer) {
 637             XBT_DEBUG("I'm source");
 638           } else if (dst_actor_ == simcall->issuer) {
 639             XBT_DEBUG("I'm dest");
 640           } else {
 641             XBT_DEBUG("I'm neither source nor dest");
 642           }
 643           simcall->issuer->throw_exception(
 644               std::make_exception_ptr(simgrid::NetworkFailureException(XBT_THROW_POINT, "Link failure")));
 645           break;
 646
 647         case SIMIX_CANCELED:
 648           if (simcall->issuer == dst_actor_)
 649             simcall->issuer->exception_ = std::make_exception_ptr(
 650                 simgrid::CancelException(XBT_THROW_POINT, "Communication canceled by the sender"));
 651           else
 652             simcall->issuer->exception_ = std::make_exception_ptr(
 653                 simgrid::CancelException(XBT_THROW_POINT, "Communication canceled by the receiver"));
 654           break;
 655
 656         default:
 657           xbt_die("Unexpected synchro state in CommImpl::finish: %d", static_cast<int>(state_));
 658       }
 659     }
 660
 661     /* if there is an exception during a waitany or a testany, indicate the position of the failed communication */
 662     if (simcall->issuer->exception_ &&
 663         (simcall->call == SIMCALL_COMM_WAITANY || simcall->call == SIMCALL_COMM_TESTANY)) {
 664       // First retrieve the rank of our failing synchro
 665       int rank = -1;
 666       if (simcall->call == SIMCALL_COMM_WAITANY) {
 667         rank = xbt_dynar_search(simcall_comm_waitany__get__comms(simcall), &synchro);
 668       } else if (simcall->call == SIMCALL_COMM_TESTANY) {
 669         rank         = -1;
 670         auto* comms  = simcall_comm_testany__get__comms(simcall);
 671         auto count   = simcall_comm_testany__get__count(simcall);
 672         auto element = std::find(comms, comms + count, this);
 673         if (element == comms + count)
 674           rank = -1;
 675         else
 676           rank = element - comms;
 677       }
 678
 679       // In order to modify the exception we have to rethrow it:
 680       try {
 681         std::rethrow_exception(simcall->issuer->exception_);
 682       } catch (simgrid::TimeoutError& e) {
 683         e.value                     = rank;
 684         simcall->issuer->exception_ = std::make_exception_ptr(e);
 685       } catch (simgrid::NetworkFailureException& e) {
 686         e.value                     = rank;
 687         simcall->issuer->exception_ = std::make_exception_ptr(e);
 688       } catch (simgrid::CancelException& e) {
 689         e.value                     = rank;
 690         simcall->issuer->exception_ = std::make_exception_ptr(e);
 691       }
 692     }
 693
 694     simcall->issuer->waiting_synchro = nullptr;
 695     simcall->issuer->comms.remove(this);
 696     if (detached) {
 697       if (simcall->issuer == src_actor_) {
 698         if (dst_actor_)
 699           dst_actor_->comms.remove(this);
 700       } else if (simcall->issuer == dst_actor_) {
 701         if (src_actor_)
 702           src_actor_->comms.remove(this);
 703       } else {
 704         dst_actor_->comms.remove(this);
 705         src_actor_->comms.remove(this);
 706       }
 707     }
 708
 709     if (simcall->issuer->get_host()->is_on())
 710       SIMIX_simcall_answer(simcall);
 711     else
 712       simcall->issuer->context_->iwannadie = true;
 713   }
 714 }
 715
 716 } // namespace activity
 717 } // namespace kernel
 718 } // namespace simgrid