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)(smx_activity_t, void*, size_t), void* data,
  25                                            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)(smx_activity_t, 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)(smx_activity_t, void*, size_t), void* data,
 107                                            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(smx_simcall_t /*simcall*/, smx_actor_t receiver,
 116                                                       smx_mailbox_t mbox, void* dst_buff, size_t* dst_buff_size,
 117                                                       simix_match_func_t match_fun,
 118                                                       void (*copy_data_fun)(smx_activity_t, void*, size_t), void* data,
 119                                                       double rate)
 120 {
 121   simgrid::kernel::activity::CommImplPtr this_synchro = simgrid::kernel::activity::CommImplPtr(
 122       new simgrid::kernel::activity::CommImpl(simgrid::kernel::activity::CommImpl::Type::RECEIVE));
 123   XBT_DEBUG("recv from mbox %p. this_synchro=%p", mbox, this_synchro.get());
 124
 125   simgrid::kernel::activity::CommImplPtr other_comm;
 126   // communication already done, get it inside the list of completed comms
 127   if (mbox->permanent_receiver_ != nullptr && not mbox->done_comm_queue_.empty()) {
 128
 129     XBT_DEBUG("We have a comm that has probably already been received, trying to match it, to skip the communication");
 130     // find a match in the list of already received comms
 131     other_comm = mbox->find_matching_comm(simgrid::kernel::activity::CommImpl::Type::SEND, match_fun, data,
 132                                           this_synchro, /*done*/ true,
 133                                           /*remove_matching*/ true);
 134     // if not found, assume the receiver came first, register it to the mailbox in the classical way
 135     if (not other_comm) {
 136       XBT_DEBUG("We have messages in the permanent receive list, but not the one we are looking for, pushing request "
 137                 "into list");
 138       other_comm = std::move(this_synchro);
 139       mbox->push(other_comm);
 140     } else {
 141       if (other_comm->surf_action_ && other_comm->remains() < 1e-12) {
 142         XBT_DEBUG("comm %p has been already sent, and is finished, destroy it", other_comm.get());
 143         other_comm->state_ = SIMIX_DONE;
 144         other_comm->type   = simgrid::kernel::activity::CommImpl::Type::DONE;
 145         other_comm->mbox   = nullptr;
 146       }
 147     }
 148   } else {
 149     /* Prepare a comm describing us, so that it gets passed to the user-provided filter of other side */
 150
 151     /* Look for communication activity matching our needs. We also provide a description of
 152      * ourself so that the other side also gets a chance of choosing if it wants to match with us.
 153      *
 154      * If it is not found then push our communication into the rendez-vous point */
 155     other_comm = mbox->find_matching_comm(simgrid::kernel::activity::CommImpl::Type::SEND, match_fun, data,
 156                                           this_synchro, /*done*/ false,
 157                                           /*remove_matching*/ true);
 158
 159     if (other_comm == nullptr) {
 160       XBT_DEBUG("Receive pushed first (%zu comm enqueued so far)", mbox->comm_queue_.size());
 161       other_comm = std::move(this_synchro);
 162       mbox->push(other_comm);
 163     } else {
 164       XBT_DEBUG("Match my %p with the existing %p", this_synchro.get(), other_comm.get());
 165
 166       other_comm->state_ = SIMIX_READY;
 167       other_comm->type   = simgrid::kernel::activity::CommImpl::Type::READY;
 168     }
 169     receiver->comms.push_back(other_comm);
 170   }
 171
 172   /* Setup communication synchro */
 173   other_comm->dst_actor_     = receiver;
 174   other_comm->dst_buff_      = dst_buff;
 175   other_comm->dst_buff_size_ = dst_buff_size;
 176   other_comm->dst_data_      = data;
 177
 178   if (rate > -1.0 && (other_comm->rate_ < 0.0 || rate < other_comm->rate_))
 179     other_comm->rate_ = rate;
 180
 181   other_comm->match_fun     = match_fun;
 182   other_comm->copy_data_fun = copy_data_fun;
 183
 184   if (MC_is_active() || MC_record_replay_is_active()) {
 185     other_comm->state_ = SIMIX_RUNNING;
 186     return other_comm;
 187   }
 188   other_comm->start();
 189   return other_comm;
 190 }
 191
 192 void simcall_HANDLER_comm_wait(smx_simcall_t simcall, smx_activity_t synchro, double timeout)
 193 {
 194   /* Associate this simcall to the wait synchro */
 195   XBT_DEBUG("simcall_HANDLER_comm_wait, %p", synchro.get());
 196
 197   synchro->simcalls_.push_back(simcall);
 198   simcall->issuer->waiting_synchro = synchro;
 199
 200   if (MC_is_active() || MC_record_replay_is_active()) {
 201     int idx = SIMCALL_GET_MC_VALUE(simcall);
 202     if (idx == 0) {
 203       synchro->state_ = SIMIX_DONE;
 204     } else {
 205       /* If we reached this point, the wait simcall must have a timeout */
 206       /* Otherwise it shouldn't be enabled and executed by the MC */
 207       if (timeout < 0.0)
 208         THROW_IMPOSSIBLE;
 209
 210       simgrid::kernel::activity::CommImplPtr comm =
 211           boost::static_pointer_cast<simgrid::kernel::activity::CommImpl>(synchro);
 212       if (comm->src_actor_ == simcall->issuer)
 213         comm->state_ = SIMIX_SRC_TIMEOUT;
 214       else
 215         comm->state_ = SIMIX_DST_TIMEOUT;
 216     }
 217
 218     boost::static_pointer_cast<simgrid::kernel::activity::CommImpl>(synchro)->finish();
 219     return;
 220   }
 221
 222   /* If the synchro has already finish perform the error handling, */
 223   /* otherwise set up a waiting timeout on the right side          */
 224   if (synchro->state_ != SIMIX_WAITING && synchro->state_ != SIMIX_RUNNING) {
 225     boost::static_pointer_cast<simgrid::kernel::activity::CommImpl>(synchro)->finish();
 226   } else { /* we need a sleep action (even when there is no timeout) to be notified of host failures */
 227     simgrid::kernel::resource::Action* sleep = simcall->issuer->get_host()->pimpl_cpu->sleep(timeout);
 228     sleep->set_data(synchro.get());
 229
 230     simgrid::kernel::activity::CommImplPtr comm =
 231         boost::static_pointer_cast<simgrid::kernel::activity::CommImpl>(synchro);
 232     if (simcall->issuer == comm->src_actor_)
 233       comm->src_timeout_ = sleep;
 234     else
 235       comm->dst_timeout_ = sleep;
 236   }
 237 }
 238
 239 void simcall_HANDLER_comm_test(smx_simcall_t simcall, smx_activity_t synchro)
 240 {
 241   simgrid::kernel::activity::CommImplPtr comm =
 242       boost::static_pointer_cast<simgrid::kernel::activity::CommImpl>(synchro);
 243
 244   int res;
 245
 246   if (MC_is_active() || MC_record_replay_is_active()) {
 247     res = comm->src_actor_ && comm->dst_actor_;
 248     if (res)
 249       synchro->state_ = SIMIX_DONE;
 250   } else {
 251     res = synchro->state_ != SIMIX_WAITING && synchro->state_ != SIMIX_RUNNING;
 252   }
 253
 254   simcall_comm_test__set__result(simcall, res);
 255   if (simcall_comm_test__get__result(simcall)) {
 256     synchro->simcalls_.push_back(simcall);
 257     boost::static_pointer_cast<simgrid::kernel::activity::CommImpl>(synchro)->finish();
 258   } else {
 259     SIMIX_simcall_answer(simcall);
 260   }
 261 }
 262
 263 void simcall_HANDLER_comm_testany(smx_simcall_t simcall, simgrid::kernel::activity::ActivityImplPtr comms[],
 264                                   size_t count)
 265 {
 266   // The default result is -1 -- this means, "nothing is ready".
 267   // It can be changed below, but only if something matches.
 268   simcall_comm_testany__set__result(simcall, -1);
 269
 270   if (MC_is_active() || MC_record_replay_is_active()) {
 271     int idx = SIMCALL_GET_MC_VALUE(simcall);
 272     if (idx == -1) {
 273       SIMIX_simcall_answer(simcall);
 274     } else {
 275       simgrid::kernel::activity::ActivityImplPtr synchro = comms[idx];
 276       simcall_comm_testany__set__result(simcall, idx);
 277       synchro->simcalls_.push_back(simcall);
 278       synchro->state_ = SIMIX_DONE;
 279       boost::static_pointer_cast<simgrid::kernel::activity::CommImpl>(synchro)->finish();
 280     }
 281     return;
 282   }
 283
 284   for (std::size_t i = 0; i != count; ++i) {
 285     simgrid::kernel::activity::ActivityImplPtr synchro = comms[i];
 286     if (synchro->state_ != SIMIX_WAITING && synchro->state_ != SIMIX_RUNNING) {
 287       simcall_comm_testany__set__result(simcall, i);
 288       synchro->simcalls_.push_back(simcall);
 289       boost::static_pointer_cast<simgrid::kernel::activity::CommImpl>(synchro)->finish();
 290       return;
 291     }
 292   }
 293   SIMIX_simcall_answer(simcall);
 294 }
 295
 296 static void SIMIX_waitany_remove_simcall_from_actions(smx_simcall_t simcall)
 297 {
 298   unsigned int cursor  = 0;
 299   xbt_dynar_t synchros = simcall_comm_waitany__get__comms(simcall);
 300
 301   simgrid::kernel::activity::ActivityImpl* ptr;
 302   xbt_dynar_foreach (synchros, cursor, ptr) {
 303     smx_activity_t synchro = simgrid::kernel::activity::ActivityImplPtr(ptr);
 304
 305     // Remove the first occurence of simcall:
 306     auto i = boost::range::find(synchro->simcalls_, simcall);
 307     if (i != synchro->simcalls_.end())
 308       synchro->simcalls_.erase(i);
 309   }
 310 }
 311 void simcall_HANDLER_comm_waitany(smx_simcall_t simcall, xbt_dynar_t synchros, double timeout)
 312 {
 313   if (MC_is_active() || MC_record_replay_is_active()) {
 314     if (timeout > 0.0)
 315       xbt_die("Timeout not implemented for waitany in the model-checker");
 316     int idx                = SIMCALL_GET_MC_VALUE(simcall);
 317     smx_activity_t synchro = xbt_dynar_get_as(synchros, idx, smx_activity_t);
 318     synchro->simcalls_.push_back(simcall);
 319     simcall_comm_waitany__set__result(simcall, idx);
 320     synchro->state_ = SIMIX_DONE;
 321     boost::static_pointer_cast<simgrid::kernel::activity::CommImpl>(synchro)->finish();
 322     return;
 323   }
 324
 325   if (timeout < 0.0) {
 326     simcall->timer = NULL;
 327   } else {
 328     simcall->timer = simgrid::simix::Timer::set(SIMIX_get_clock() + timeout, [simcall]() {
 329       SIMIX_waitany_remove_simcall_from_actions(simcall);
 330       simcall_comm_waitany__set__result(simcall, -1);
 331       SIMIX_simcall_answer(simcall);
 332     });
 333   }
 334
 335   unsigned int cursor;
 336   simgrid::kernel::activity::ActivityImpl* ptr;
 337   xbt_dynar_foreach (synchros, cursor, ptr) {
 338     smx_activity_t synchro = simgrid::kernel::activity::ActivityImplPtr(ptr);
 339     /* associate this simcall to the the synchro */
 340     synchro->simcalls_.push_back(simcall);
 341
 342     /* see if the synchro is already finished */
 343     if (synchro->state_ != SIMIX_WAITING && synchro->state_ != SIMIX_RUNNING) {
 344       boost::static_pointer_cast<simgrid::kernel::activity::CommImpl>(synchro)->finish();
 345       break;
 346     }
 347   }
 348 }
 349
 350 /******************************************************************************/
 351 /*                    SIMIX_comm_copy_data callbacks                       */
 352 /******************************************************************************/
 353 static void (*SIMIX_comm_copy_data_callback)(smx_activity_t, void*, size_t) = &SIMIX_comm_copy_pointer_callback;
 354
 355 void SIMIX_comm_copy_buffer_callback(smx_activity_t synchro, void* buff, size_t buff_size)
 356 {
 357   simgrid::kernel::activity::CommImplPtr comm =
 358       boost::static_pointer_cast<simgrid::kernel::activity::CommImpl>(synchro);
 359
 360   XBT_DEBUG("Copy the data over");
 361   memcpy(comm->dst_buff_, buff, buff_size);
 362   if (comm->detached) { // if this is a detached send, the source buffer was duplicated by SMPI sender to make the
 363                         // original buffer available to the application ASAP
 364     xbt_free(buff);
 365     comm->src_buff_ = nullptr;
 366   }
 367 }
 368
 369 void SIMIX_comm_set_copy_data_callback(void (*callback)(smx_activity_t, void*, size_t))
 370 {
 371   SIMIX_comm_copy_data_callback = callback;
 372 }
 373
 374 void SIMIX_comm_copy_pointer_callback(smx_activity_t synchro, void* buff, size_t buff_size)
 375 {
 376   simgrid::kernel::activity::CommImplPtr comm =
 377       boost::static_pointer_cast<simgrid::kernel::activity::CommImpl>(synchro);
 378
 379   xbt_assert((buff_size == sizeof(void*)), "Cannot copy %zu bytes: must be sizeof(void*)", buff_size);
 380   *(void**)(comm->dst_buff_) = buff;
 381 }
 382
 383 namespace simgrid {
 384 namespace kernel {
 385 namespace activity {
 386
 387 CommImpl::CommImpl(CommImpl::Type type) : type(type)
 388 {
 389   state_   = SIMIX_WAITING;
 390   src_data_ = nullptr;
 391   dst_data_ = nullptr;
 392   XBT_DEBUG("Create comm activity %p", this);
 393 }
 394
 395 CommImpl::~CommImpl()
 396 {
 397   XBT_DEBUG("Really free communication %p", this);
 398
 399   cleanupSurf();
 400
 401   if (detached && state_ != SIMIX_DONE) {
 402     /* the communication has failed and was detached:
 403      * we have to free the buffer */
 404     if (clean_fun)
 405       clean_fun(src_buff_);
 406     src_buff_ = nullptr;
 407   }
 408
 409   if (mbox)
 410     mbox->remove(this);
 411 }
 412
 413 /**  @brief Starts the simulation of a communication synchro. */
 414 void CommImpl::start()
 415 {
 416   /* If both the sender and the receiver are already there, start the communication */
 417   if (state_ == SIMIX_READY) {
 418
 419     s4u::Host* sender   = src_actor_->get_host();
 420     s4u::Host* receiver = dst_actor_->get_host();
 421
 422     surf_action_ = surf_network_model->communicate(sender, receiver, task_size_, rate_);
 423     surf_action_->set_data(this);
 424     state_ = SIMIX_RUNNING;
 425
 426     XBT_DEBUG("Starting communication %p from '%s' to '%s' (surf_action: %p)", this, sender->get_cname(),
 427               receiver->get_cname(), surf_action_);
 428
 429     /* If a link is failed, detect it immediately */
 430     if (surf_action_->get_state() == resource::Action::State::FAILED) {
 431       XBT_DEBUG("Communication from '%s' to '%s' failed to start because of a link failure", sender->get_cname(),
 432                 receiver->get_cname());
 433       state_ = SIMIX_LINK_FAILURE;
 434       cleanupSurf();
 435
 436     } else if (src_actor_->is_suspended() || dst_actor_->is_suspended()) {
 437       /* If any of the process is suspended, create the synchro but stop its execution,
 438          it will be restarted when the sender process resume */
 439       if (src_actor_->is_suspended())
 440         XBT_DEBUG("The communication is suspended on startup because src (%s@%s) was suspended since it initiated the "
 441                   "communication",
 442                   src_actor_->get_cname(), src_actor_->get_host()->get_cname());
 443       else
 444         XBT_DEBUG("The communication is suspended on startup because dst (%s@%s) was suspended since it initiated the "
 445                   "communication",
 446                   dst_actor_->get_cname(), dst_actor_->get_host()->get_cname());
 447
 448       surf_action_->suspend();
 449     }
 450   }
 451 }
 452
 453 /** @brief Copy the communication data from the sender's buffer to the receiver's one  */
 454 void CommImpl::copy_data()
 455 {
 456   size_t buff_size = src_buff_size_;
 457   /* If there is no data to copy then return */
 458   if (not src_buff_ || not dst_buff_ || copied)
 459     return;
 460
 461   XBT_DEBUG("Copying comm %p data from %s (%p) -> %s (%p) (%zu bytes)", this,
 462             src_actor_ ? src_actor_->get_host()->get_cname() : "a finished process", src_buff_,
 463             dst_actor_ ? dst_actor_->get_host()->get_cname() : "a finished process", dst_buff_, buff_size);
 464
 465   /* Copy at most dst_buff_size bytes of the message to receiver's buffer */
 466   if (dst_buff_size_)
 467     buff_size = std::min(buff_size, *(dst_buff_size_));
 468
 469   /* Update the receiver's buffer size to the copied amount */
 470   if (dst_buff_size_)
 471     *dst_buff_size_ = buff_size;
 472
 473   if (buff_size > 0) {
 474     if (copy_data_fun)
 475       copy_data_fun(this, src_buff_, buff_size);
 476     else
 477       SIMIX_comm_copy_data_callback(this, src_buff_, buff_size);
 478   }
 479
 480   /* Set the copied flag so we copy data only once */
 481   /* (this function might be called from both communication ends) */
 482   copied = true;
 483 }
 484
 485 void CommImpl::suspend()
 486 {
 487   /* FIXME: shall we suspend also the timeout synchro? */
 488   if (surf_action_)
 489     surf_action_->suspend();
 490   /* if not created yet, the action will be suspended on creation, in CommImpl::start() */
 491 }
 492
 493 void CommImpl::resume()
 494 {
 495   /*FIXME: check what happen with the timeouts */
 496   if (surf_action_)
 497     surf_action_->resume();
 498   /* in the other case, the synchro were not really suspended yet, see CommImpl::suspend() and CommImpl::start() */
 499 }
 500
 501 void CommImpl::cancel()
 502 {
 503   /* if the synchro is a waiting state means that it is still in a mbox so remove from it and delete it */
 504   if (state_ == SIMIX_WAITING) {
 505     mbox->remove(this);
 506     state_ = SIMIX_CANCELED;
 507   } else if (not MC_is_active() /* when running the MC there are no surf actions */
 508              && not MC_record_replay_is_active() && (state_ == SIMIX_READY || state_ == SIMIX_RUNNING)) {
 509     surf_action_->cancel();
 510   }
 511 }
 512
 513 /**  @brief get the amount remaining from the communication */
 514 double CommImpl::remains()
 515 {
 516   return surf_action_->get_remains();
 517 }
 518
 519 /** @brief This is part of the cleanup process, probably an internal command */
 520 void CommImpl::cleanupSurf()
 521 {
 522   if (surf_action_) {
 523     surf_action_->unref();
 524     surf_action_ = nullptr;
 525   }
 526
 527   if (src_timeout_) {
 528     src_timeout_->unref();
 529     src_timeout_ = nullptr;
 530   }
 531
 532   if (dst_timeout_) {
 533     dst_timeout_->unref();
 534     dst_timeout_ = nullptr;
 535   }
 536 }
 537
 538 void CommImpl::post()
 539 {
 540   /* Update synchro state */
 541   if (src_timeout_ && src_timeout_->get_state() == resource::Action::State::FINISHED)
 542     state_ = SIMIX_SRC_TIMEOUT;
 543   else if (dst_timeout_ && dst_timeout_->get_state() == resource::Action::State::FINISHED)
 544     state_ = SIMIX_DST_TIMEOUT;
 545   else if (src_timeout_ && src_timeout_->get_state() == resource::Action::State::FAILED)
 546     state_ = SIMIX_SRC_HOST_FAILURE;
 547   else if (dst_timeout_ && dst_timeout_->get_state() == resource::Action::State::FAILED)
 548     state_ = SIMIX_DST_HOST_FAILURE;
 549   else if (surf_action_ && surf_action_->get_state() == resource::Action::State::FAILED) {
 550     state_ = SIMIX_LINK_FAILURE;
 551   } else
 552     state_ = SIMIX_DONE;
 553
 554   XBT_DEBUG("SIMIX_post_comm: comm %p, state %d, src_proc %p, dst_proc %p, detached: %d", this, (int)state_,
 555             src_actor_.get(), dst_actor_.get(), detached);
 556
 557   /* destroy the surf actions associated with the Simix communication */
 558   cleanupSurf();
 559
 560   /* if there are simcalls associated with the synchro, then answer them */
 561   if (not simcalls_.empty()) {
 562     finish();
 563   }
 564 }
 565
 566 void CommImpl::finish()
 567 {
 568   smx_activity_t synchro = this;
 569   while (not simcalls_.empty()) {
 570     smx_simcall_t simcall = simcalls_.front();
 571     simcalls_.pop_front();
 572
 573     /* If a waitany simcall is waiting for this synchro to finish, then remove it from the other synchros in the waitany
 574      * list. Afterwards, get the position of the actual synchro in the waitany dynar and return it as the result of the
 575      * simcall */
 576
 577     if (simcall->call == SIMCALL_NONE) // FIXME: maybe a better way to handle this case
 578       continue;                        // if process handling comm is killed
 579     if (simcall->call == SIMCALL_COMM_WAITANY) {
 580       SIMIX_waitany_remove_simcall_from_actions(simcall);
 581       if (simcall->timer) {
 582         simcall->timer->remove();
 583         simcall->timer = nullptr;
 584       }
 585       if (not MC_is_active() && not MC_record_replay_is_active())
 586         simcall_comm_waitany__set__result(simcall,
 587                                           xbt_dynar_search(simcall_comm_waitany__get__comms(simcall), &synchro));
 588     }
 589
 590     /* If the synchro is still in a rendez-vous point then remove from it */
 591     if (mbox)
 592       mbox->remove(this);
 593
 594     XBT_DEBUG("CommImpl::finish(): synchro state = %d", static_cast<int>(state_));
 595
 596     /* Check out for errors */
 597
 598     if (not simcall->issuer->get_host()->is_on()) {
 599       simcall->issuer->context_->iwannadie = true;
 600       simcall->issuer->exception_ =
 601           std::make_exception_ptr(simgrid::HostFailureException(XBT_THROW_POINT, "Host failed"));
 602     } else {
 603       switch (state_) {
 604
 605         case SIMIX_DONE:
 606           XBT_DEBUG("Communication %p complete!", this);
 607           copy_data();
 608           break;
 609
 610         case SIMIX_SRC_TIMEOUT:
 611           simcall->issuer->exception_ = std::make_exception_ptr(
 612               simgrid::TimeoutError(XBT_THROW_POINT, "Communication timeouted because of the sender"));
 613           break;
 614
 615         case SIMIX_DST_TIMEOUT:
 616           simcall->issuer->exception_ = std::make_exception_ptr(
 617               simgrid::TimeoutError(XBT_THROW_POINT, "Communication timeouted because of the receiver"));
 618           break;
 619
 620         case SIMIX_SRC_HOST_FAILURE:
 621           if (simcall->issuer == src_actor_)
 622             simcall->issuer->context_->iwannadie = true;
 623           else
 624             simcall->issuer->exception_ =
 625                 std::make_exception_ptr(simgrid::NetworkFailureException(XBT_THROW_POINT, "Remote peer failed"));
 626           break;
 627
 628         case SIMIX_DST_HOST_FAILURE:
 629           if (simcall->issuer == dst_actor_)
 630             simcall->issuer->context_->iwannadie = true;
 631           else
 632             simcall->issuer->exception_ =
 633                 std::make_exception_ptr(simgrid::NetworkFailureException(XBT_THROW_POINT, "Remote peer failed"));
 634           break;
 635
 636         case SIMIX_LINK_FAILURE:
 637           XBT_DEBUG("Link failure in synchro %p between '%s' and '%s': posting an exception to the issuer: %s (%p) "
 638                     "detached:%d",
 639                     this, src_actor_ ? src_actor_->get_host()->get_cname() : nullptr,
 640                     dst_actor_ ? dst_actor_->get_host()->get_cname() : nullptr, simcall->issuer->get_cname(),
 641                     simcall->issuer, detached);
 642           if (src_actor_ == simcall->issuer) {
 643             XBT_DEBUG("I'm source");
 644           } else if (dst_actor_ == simcall->issuer) {
 645             XBT_DEBUG("I'm dest");
 646           } else {
 647             XBT_DEBUG("I'm neither source nor dest");
 648           }
 649           simcall->issuer->throw_exception(
 650               std::make_exception_ptr(simgrid::NetworkFailureException(XBT_THROW_POINT, "Link failure")));
 651           break;
 652
 653         case SIMIX_CANCELED:
 654           if (simcall->issuer == dst_actor_)
 655             simcall->issuer->exception_ = std::make_exception_ptr(
 656                 simgrid::CancelException(XBT_THROW_POINT, "Communication canceled by the sender"));
 657           else
 658             simcall->issuer->exception_ = std::make_exception_ptr(
 659                 simgrid::CancelException(XBT_THROW_POINT, "Communication canceled by the receiver"));
 660           break;
 661
 662         default:
 663           xbt_die("Unexpected synchro state in CommImpl::finish: %d", static_cast<int>(state_));
 664       }
 665     }
 666
 667     /* if there is an exception during a waitany or a testany, indicate the position of the failed communication */
 668     if (simcall->issuer->exception_ &&
 669         (simcall->call == SIMCALL_COMM_WAITANY || simcall->call == SIMCALL_COMM_TESTANY)) {
 670       // First retrieve the rank of our failing synchro
 671       int rank = -1;
 672       if (simcall->call == SIMCALL_COMM_WAITANY) {
 673         rank = xbt_dynar_search(simcall_comm_waitany__get__comms(simcall), &synchro);
 674       } else if (simcall->call == SIMCALL_COMM_TESTANY) {
 675         rank         = -1;
 676         auto* comms  = simcall_comm_testany__get__comms(simcall);
 677         auto count   = simcall_comm_testany__get__count(simcall);
 678         auto element = std::find(comms, comms + count, this);
 679         if (element == comms + count)
 680           rank = -1;
 681         else
 682           rank = element - comms;
 683       }
 684
 685       // In order to modify the exception we have to rethrow it:
 686       try {
 687         std::rethrow_exception(simcall->issuer->exception_);
 688       } catch (simgrid::TimeoutError& e) {
 689         e.value                     = rank;
 690         simcall->issuer->exception_ = std::make_exception_ptr(e);
 691       } catch (simgrid::NetworkFailureException& e) {
 692         e.value                     = rank;
 693         simcall->issuer->exception_ = std::make_exception_ptr(e);
 694       } catch (simgrid::CancelException& e) {
 695         e.value                     = rank;
 696         simcall->issuer->exception_ = std::make_exception_ptr(e);
 697       }
 698     }
 699
 700     simcall->issuer->waiting_synchro = nullptr;
 701     simcall->issuer->comms.remove(this);
 702     if (detached) {
 703       if (simcall->issuer == src_actor_) {
 704         if (dst_actor_)
 705           dst_actor_->comms.remove(this);
 706       } else if (simcall->issuer == dst_actor_) {
 707         if (src_actor_)
 708           src_actor_->comms.remove(this);
 709       } else {
 710         dst_actor_->comms.remove(this);
 711         src_actor_->comms.remove(this);
 712       }
 713     }
 714
 715     if (simcall->issuer->get_host()->is_on())
 716       SIMIX_simcall_answer(simcall);
 717     else
 718       simcall->issuer->context_->iwannadie = true;
 719   }
 720 }
 721
 722 } // namespace activity
 723 } // namespace kernel
 724 } // namespace simgrid