src/kernel/activity/CommImpl.cpp

   1 /* Copyright (c) 2007-2021. 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, unsigned char* src_buff, size_t src_buff_size,
  23                                            bool (*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   simgrid::kernel::activity::ActivityImplPtr comm = simcall_HANDLER_comm_isend(
  28       simcall, src, mbox, task_size, rate, src_buff, src_buff_size, match_fun, nullptr, copy_data_fun, data, false);
  29   SIMCALL_SET_MC_VALUE(*simcall, 0);
  30   simcall_HANDLER_comm_wait(simcall, static_cast<simgrid::kernel::activity::CommImpl*>(comm.get()), timeout);
  31 }
  32
  33 XBT_PRIVATE simgrid::kernel::activity::ActivityImplPtr simcall_HANDLER_comm_isend(
  34     smx_simcall_t /*simcall*/, smx_actor_t src_proc, smx_mailbox_t mbox, double task_size, double rate,
  35     unsigned char* src_buff, size_t src_buff_size,
  36     bool (*match_fun)(void*, void*, simgrid::kernel::activity::CommImpl*),
  37     void (*clean_fun)(void*), // used to free the synchro in case of problem after a detached send
  38     void (*copy_data_fun)(simgrid::kernel::activity::CommImpl*, void*, size_t), // used to copy data if not default one
  39     void* data, bool detached)
  40 {
  41   XBT_DEBUG("send from mailbox %p", mbox);
  42
  43   /* Prepare a synchro describing us, so that it gets passed to the user-provided filter of other side */
  44   simgrid::kernel::activity::CommImplPtr this_comm(
  45       new simgrid::kernel::activity::CommImpl(simgrid::kernel::activity::CommImpl::Type::SEND));
  46
  47   /* Look for communication synchro matching our needs. We also provide a description of
  48    * ourself so that the other side also gets a chance of choosing if it wants to match with us.
  49    *
  50    * If it is not found then push our communication into the rendez-vous point */
  51   simgrid::kernel::activity::CommImplPtr other_comm =
  52       mbox->find_matching_comm(simgrid::kernel::activity::CommImpl::Type::RECEIVE, match_fun, data, this_comm,
  53                                /*done*/ false, /*remove_matching*/ true);
  54
  55   if (not other_comm) {
  56     other_comm = std::move(this_comm);
  57
  58     if (mbox->permanent_receiver_ != nullptr) {
  59       // this mailbox is for small messages, which have to be sent right now
  60       other_comm->state_     = simgrid::kernel::activity::State::READY;
  61       other_comm->dst_actor_ = mbox->permanent_receiver_.get();
  62       mbox->done_comm_queue_.push_back(other_comm);
  63       XBT_DEBUG("pushing a message into the permanent receive list %p, comm %p", mbox, other_comm.get());
  64
  65     } else {
  66       mbox->push(other_comm);
  67     }
  68   } else {
  69     XBT_DEBUG("Receive already pushed");
  70
  71     other_comm->state_ = simgrid::kernel::activity::State::READY;
  72   }
  73
  74   if (detached) {
  75     other_comm->detach();
  76     other_comm->clean_fun = clean_fun;
  77   } else {
  78     other_comm->clean_fun = nullptr;
  79     src_proc->activities_.emplace_back(other_comm);
  80   }
  81
  82   /* Setup the communication synchro */
  83   other_comm->src_actor_     = src_proc;
  84   other_comm->src_data_      = data;
  85   (*other_comm).set_src_buff(src_buff, src_buff_size).set_size(task_size).set_rate(rate);
  86
  87   other_comm->match_fun     = match_fun;
  88   other_comm->copy_data_fun = copy_data_fun;
  89
  90   if (MC_is_active() || MC_record_replay_is_active())
  91     other_comm->state_ = simgrid::kernel::activity::State::RUNNING;
  92   else
  93     other_comm->start();
  94
  95   return (detached ? nullptr : other_comm);
  96 }
  97
  98 XBT_PRIVATE void simcall_HANDLER_comm_recv(smx_simcall_t simcall, smx_actor_t receiver, smx_mailbox_t mbox,
  99                                            unsigned char* dst_buff, size_t* dst_buff_size,
 100                                            bool (*match_fun)(void*, void*, simgrid::kernel::activity::CommImpl*),
 101                                            void (*copy_data_fun)(simgrid::kernel::activity::CommImpl*, void*, size_t),
 102                                            void* data, double timeout, double rate)
 103 {
 104   simgrid::kernel::activity::ActivityImplPtr comm = simcall_HANDLER_comm_irecv(
 105       simcall, receiver, mbox, dst_buff, dst_buff_size, match_fun, copy_data_fun, data, rate);
 106   SIMCALL_SET_MC_VALUE(*simcall, 0);
 107   simcall_HANDLER_comm_wait(simcall, static_cast<simgrid::kernel::activity::CommImpl*>(comm.get()), timeout);
 108 }
 109
 110 XBT_PRIVATE simgrid::kernel::activity::ActivityImplPtr
 111 simcall_HANDLER_comm_irecv(smx_simcall_t /*simcall*/, smx_actor_t receiver, smx_mailbox_t mbox, unsigned char* dst_buff,
 112                            size_t* dst_buff_size, bool (*match_fun)(void*, void*, simgrid::kernel::activity::CommImpl*),
 113                            void (*copy_data_fun)(simgrid::kernel::activity::CommImpl*, void*, size_t), void* data,
 114                            double rate)
 115 {
 116   simgrid::kernel::activity::CommImplPtr this_synchro(
 117       new simgrid::kernel::activity::CommImpl(simgrid::kernel::activity::CommImpl::Type::RECEIVE));
 118   XBT_DEBUG("recv from mbox %p. this_synchro=%p", mbox, this_synchro.get());
 119
 120   simgrid::kernel::activity::CommImplPtr other_comm;
 121   // communication already done, get it inside the list of completed comms
 122   if (mbox->permanent_receiver_ != nullptr && not mbox->done_comm_queue_.empty()) {
 123     XBT_DEBUG("We have a comm that has probably already been received, trying to match it, to skip the communication");
 124     // find a match in the list of already received comms
 125     other_comm = mbox->find_matching_comm(simgrid::kernel::activity::CommImpl::Type::SEND, match_fun, data,
 126                                           this_synchro, /*done*/ true,
 127                                           /*remove_matching*/ true);
 128     // if not found, assume the receiver came first, register it to the mailbox in the classical way
 129     if (not other_comm) {
 130       XBT_DEBUG("We have messages in the permanent receive list, but not the one we are looking for, pushing request "
 131                 "into list");
 132       other_comm = std::move(this_synchro);
 133       mbox->push(other_comm);
 134     } else {
 135       if (other_comm->surf_action_ && other_comm->get_remaining() < 1e-12) {
 136         XBT_DEBUG("comm %p has been already sent, and is finished, destroy it", other_comm.get());
 137         other_comm->state_ = simgrid::kernel::activity::State::DONE;
 138         other_comm->set_mailbox(nullptr);
 139       }
 140     }
 141   } else {
 142     /* Prepare a comm describing us, so that it gets passed to the user-provided filter of other side */
 143
 144     /* Look for communication activity matching our needs. We also provide a description of
 145      * ourself so that the other side also gets a chance of choosing if it wants to match with us.
 146      *
 147      * If it is not found then push our communication into the rendez-vous point */
 148     other_comm = mbox->find_matching_comm(simgrid::kernel::activity::CommImpl::Type::SEND, match_fun, data,
 149                                           this_synchro, /*done*/ false,
 150                                           /*remove_matching*/ true);
 151
 152     if (other_comm == nullptr) {
 153       XBT_DEBUG("Receive pushed first (%zu comm enqueued so far)", mbox->comm_queue_.size());
 154       other_comm = std::move(this_synchro);
 155       mbox->push(other_comm);
 156     } else {
 157       XBT_DEBUG("Match my %p with the existing %p", this_synchro.get(), other_comm.get());
 158
 159       other_comm->state_ = simgrid::kernel::activity::State::READY;
 160     }
 161     receiver->activities_.emplace_back(other_comm);
 162   }
 163
 164   /* Setup communication synchro */
 165   other_comm->dst_actor_     = receiver;
 166   other_comm->dst_data_      = data;
 167   other_comm->set_dst_buff(dst_buff, dst_buff_size);
 168
 169   if (rate > -1.0 && (other_comm->get_rate() < 0.0 || rate < other_comm->get_rate()))
 170     other_comm->set_rate(rate);
 171
 172   other_comm->match_fun     = match_fun;
 173   other_comm->copy_data_fun = copy_data_fun;
 174
 175   if (MC_is_active() || MC_record_replay_is_active()) {
 176     other_comm->state_ = simgrid::kernel::activity::State::RUNNING;
 177     return other_comm;
 178   }
 179   other_comm->start();
 180   return other_comm;
 181 }
 182
 183 void simcall_HANDLER_comm_wait(smx_simcall_t simcall, simgrid::kernel::activity::CommImpl* comm, double timeout)
 184 {
 185   /* Associate this simcall to the wait synchro */
 186   XBT_DEBUG("simcall_HANDLER_comm_wait, %p", comm);
 187
 188   comm->register_simcall(simcall);
 189
 190   if (MC_is_active() || MC_record_replay_is_active()) {
 191     int idx = SIMCALL_GET_MC_VALUE(*simcall);
 192     if (idx == 0) {
 193       comm->state_ = simgrid::kernel::activity::State::DONE;
 194     } else {
 195       /* If we reached this point, the wait simcall must have a timeout */
 196       /* Otherwise it shouldn't be enabled and executed by the MC */
 197       if (timeout < 0.0)
 198         THROW_IMPOSSIBLE;
 199
 200       if (comm->src_actor_ == simcall->issuer_)
 201         comm->state_ = simgrid::kernel::activity::State::SRC_TIMEOUT;
 202       else
 203         comm->state_ = simgrid::kernel::activity::State::DST_TIMEOUT;
 204     }
 205
 206     comm->finish();
 207     return;
 208   }
 209
 210   /* If the synchro has already finish perform the error handling, */
 211   /* otherwise set up a waiting timeout on the right side          */
 212   if (comm->state_ != simgrid::kernel::activity::State::WAITING &&
 213       comm->state_ != simgrid::kernel::activity::State::RUNNING) {
 214     comm->finish();
 215   } else { /* we need a sleep action (even when there is no timeout) to be notified of host failures */
 216     simgrid::kernel::resource::Action* sleep = simcall->issuer_->get_host()->pimpl_cpu->sleep(timeout);
 217     sleep->set_activity(comm);
 218
 219     if (simcall->issuer_ == comm->src_actor_)
 220       comm->src_timeout_ = sleep;
 221     else
 222       comm->dst_timeout_ = sleep;
 223   }
 224 }
 225
 226 void simcall_HANDLER_comm_test(smx_simcall_t simcall, simgrid::kernel::activity::CommImpl* comm)
 227 {
 228   bool res;
 229
 230   if (MC_is_active() || MC_record_replay_is_active()) {
 231     res = comm->src_actor_ && comm->dst_actor_;
 232     if (res)
 233       comm->state_ = simgrid::kernel::activity::State::DONE;
 234   } else {
 235     res = comm->state_ != simgrid::kernel::activity::State::WAITING &&
 236           comm->state_ != simgrid::kernel::activity::State::RUNNING;
 237   }
 238
 239   simcall_comm_test__set__result(simcall, res);
 240   if (res) {
 241     comm->simcalls_.push_back(simcall);
 242     comm->finish();
 243   } else {
 244     simcall->issuer_->simcall_answer();
 245   }
 246 }
 247
 248 void simcall_HANDLER_comm_testany(smx_simcall_t simcall, simgrid::kernel::activity::CommImpl* comms[], size_t count)
 249 {
 250   // The default result is -1 -- this means, "nothing is ready".
 251   // It can be changed below, but only if something matches.
 252   simcall_comm_testany__set__result(simcall, -1);
 253
 254   if (MC_is_active() || MC_record_replay_is_active()) {
 255     int idx = SIMCALL_GET_MC_VALUE(*simcall);
 256     if (idx == -1) {
 257       simcall->issuer_->simcall_answer();
 258     } else {
 259       simgrid::kernel::activity::CommImpl* comm = comms[idx];
 260       simcall_comm_testany__set__result(simcall, idx);
 261       comm->simcalls_.push_back(simcall);
 262       comm->state_ = simgrid::kernel::activity::State::DONE;
 263       comm->finish();
 264     }
 265     return;
 266   }
 267
 268   for (std::size_t i = 0; i != count; ++i) {
 269     simgrid::kernel::activity::CommImpl* comm = comms[i];
 270     if (comm->state_ != simgrid::kernel::activity::State::WAITING &&
 271         comm->state_ != simgrid::kernel::activity::State::RUNNING) {
 272       simcall_comm_testany__set__result(simcall, i);
 273       comm->simcalls_.push_back(simcall);
 274       comm->finish();
 275       return;
 276     }
 277   }
 278   simcall->issuer_->simcall_answer();
 279 }
 280
 281 static void SIMIX_waitany_remove_simcall_from_actions(smx_simcall_t simcall)
 282 {
 283   simgrid::kernel::activity::CommImpl** comms = simcall_comm_waitany__get__comms(simcall);
 284   size_t count                                = simcall_comm_waitany__get__count(simcall);
 285
 286   for (size_t i = 0; i < count; i++) {
 287     // Remove the first occurrence of simcall:
 288     auto* comm = comms[i];
 289     auto j     = boost::range::find(comm->simcalls_, simcall);
 290     if (j != comm->simcalls_.end())
 291       comm->simcalls_.erase(j);
 292   }
 293 }
 294 void simcall_HANDLER_comm_waitany(smx_simcall_t simcall, simgrid::kernel::activity::CommImpl* comms[], size_t count,
 295                                   double timeout)
 296 {
 297   if (MC_is_active() || MC_record_replay_is_active()) {
 298     if (timeout > 0.0)
 299       xbt_die("Timeout not implemented for waitany in the model-checker");
 300     int idx                 = SIMCALL_GET_MC_VALUE(*simcall);
 301     auto* comm              = comms[idx];
 302     comm->simcalls_.push_back(simcall);
 303     simcall_comm_waitany__set__result(simcall, idx);
 304     comm->state_ = simgrid::kernel::activity::State::DONE;
 305     comm->finish();
 306     return;
 307   }
 308
 309   if (timeout < 0.0) {
 310     simcall->timeout_cb_ = nullptr;
 311   } else {
 312     simcall->timeout_cb_ = simgrid::simix::Timer::set(SIMIX_get_clock() + timeout, [simcall]() {
 313       simcall->timeout_cb_ = nullptr;
 314       SIMIX_waitany_remove_simcall_from_actions(simcall);
 315       simcall_comm_waitany__set__result(simcall, -1);
 316       simcall->issuer_->simcall_answer();
 317     });
 318   }
 319
 320   for (size_t i = 0; i < count; i++) {
 321     /* associate this simcall to the the synchro */
 322     auto* comm = comms[i];
 323     comm->simcalls_.push_back(simcall);
 324
 325     /* see if the synchro is already finished */
 326     if (comm->state_ != simgrid::kernel::activity::State::WAITING &&
 327         comm->state_ != simgrid::kernel::activity::State::RUNNING) {
 328       comm->finish();
 329       break;
 330     }
 331   }
 332 }
 333
 334 /******************************************************************************/
 335 /*                    SIMIX_comm_copy_data callbacks                       */
 336 /******************************************************************************/
 337 static void (*SIMIX_comm_copy_data_callback)(simgrid::kernel::activity::CommImpl*, void*,
 338                                              size_t) = &SIMIX_comm_copy_pointer_callback;
 339
 340 void SIMIX_comm_copy_buffer_callback(simgrid::kernel::activity::CommImpl* comm, void* buff, size_t buff_size)
 341 {
 342   XBT_DEBUG("Copy the data over");
 343   memcpy(comm->dst_buff_, buff, buff_size);
 344   if (comm->detached()) { // if this is a detached send, the source buffer was duplicated by SMPI sender to make the
 345                           // original buffer available to the application ASAP
 346     xbt_free(buff);
 347     comm->src_buff_ = nullptr;
 348   }
 349 }
 350
 351 void SIMIX_comm_set_copy_data_callback(void (*callback)(simgrid::kernel::activity::CommImpl*, void*, size_t))
 352 {
 353   SIMIX_comm_copy_data_callback = callback;
 354 }
 355
 356 void SIMIX_comm_copy_pointer_callback(simgrid::kernel::activity::CommImpl* comm, void* buff, size_t buff_size)
 357 {
 358   xbt_assert((buff_size == sizeof(void*)), "Cannot copy %zu bytes: must be sizeof(void*)", buff_size);
 359   *(void**)(comm->dst_buff_) = buff;
 360 }
 361
 362 namespace simgrid {
 363 namespace kernel {
 364 namespace activity {
 365
 366 CommImpl& CommImpl::set_size(double size)
 367 {
 368   size_ = size;
 369   return *this;
 370 }
 371
 372 CommImpl& CommImpl::set_rate(double rate)
 373 {
 374   rate_ = rate;
 375   return *this;
 376 }
 377 CommImpl& CommImpl::set_mailbox(MailboxImpl* mbox)
 378 {
 379   mbox_ = mbox;
 380   return *this;
 381 }
 382
 383 CommImpl& CommImpl::set_src_buff(unsigned char* buff, size_t size)
 384 {
 385   src_buff_      = buff;
 386   src_buff_size_ = size;
 387   return *this;
 388 }
 389
 390 CommImpl& CommImpl::set_dst_buff(unsigned char* buff, size_t* size)
 391 {
 392   dst_buff_      = buff;
 393   dst_buff_size_ = size;
 394   return *this;
 395 }
 396
 397 CommImpl& CommImpl::detach()
 398 {
 399   detached_ = true;
 400   return *this;
 401 }
 402
 403 CommImpl::CommImpl(s4u::Host* from, s4u::Host* to, double bytes)
 404     : size_(bytes), detached_(true), type_(Type::SEND), from_(from), to_(to)
 405 {
 406   state_ = State::READY;
 407 }
 408
 409 CommImpl::~CommImpl()
 410 {
 411   XBT_DEBUG("Really free communication %p in state %d (detached = %d)", this, static_cast<int>(state_), detached_);
 412
 413   cleanup_surf();
 414
 415   if (detached_ && state_ != State::DONE) {
 416     /* the communication has failed and was detached:
 417      * we have to free the buffer */
 418     if (clean_fun)
 419       clean_fun(src_buff_);
 420     src_buff_ = nullptr;
 421   } else if (mbox_) {
 422     mbox_->remove(this);
 423   }
 424 }
 425
 426 /**  @brief Starts the simulation of a communication synchro. */
 427 CommImpl* CommImpl::start()
 428 {
 429   /* If both the sender and the receiver are already there, start the communication */
 430   if (state_ == State::READY) {
 431     from_ = from_ != nullptr ? from_ : src_actor_->get_host();
 432     to_   = to_ != nullptr ? to_ : dst_actor_->get_host();
 433
 434     surf_action_ = surf_network_model->communicate(from_, to_, size_, rate_);
 435     surf_action_->set_activity(this);
 436     surf_action_->set_category(get_tracing_category());
 437     state_ = State::RUNNING;
 438
 439     XBT_DEBUG("Starting communication %p from '%s' to '%s' (surf_action: %p; state: %s)", this, from_->get_cname(),
 440               to_->get_cname(), surf_action_, get_state_str());
 441
 442     /* If a link is failed, detect it immediately */
 443     if (surf_action_->get_state() == resource::Action::State::FAILED) {
 444       XBT_DEBUG("Communication from '%s' to '%s' failed to start because of a link failure", from_->get_cname(),
 445                 to_->get_cname());
 446       state_ = State::LINK_FAILURE;
 447       post();
 448
 449     } else if ((src_actor_ != nullptr && src_actor_->is_suspended()) ||
 450                (dst_actor_ != nullptr && dst_actor_->is_suspended())) {
 451       /* If any of the process is suspended, create the synchro but stop its execution,
 452          it will be restarted when the sender process resume */
 453       if (src_actor_->is_suspended())
 454         XBT_DEBUG("The communication is suspended on startup because src (%s@%s) was suspended since it initiated the "
 455                   "communication",
 456                   src_actor_->get_cname(), src_actor_->get_host()->get_cname());
 457       else
 458         XBT_DEBUG("The communication is suspended on startup because dst (%s@%s) was suspended since it initiated the "
 459                   "communication",
 460                   dst_actor_->get_cname(), dst_actor_->get_host()->get_cname());
 461
 462       surf_action_->suspend();
 463     }
 464   }
 465
 466   return this;
 467 }
 468
 469 /** @brief Copy the communication data from the sender's buffer to the receiver's one  */
 470 void CommImpl::copy_data()
 471 {
 472   size_t buff_size = src_buff_size_;
 473   /* If there is no data to copy then return */
 474   if (not src_buff_ || not dst_buff_ || copied_)
 475     return;
 476
 477   XBT_DEBUG("Copying comm %p data from %s (%p) -> %s (%p) (%zu bytes)", this,
 478             src_actor_ ? src_actor_->get_host()->get_cname() : "a finished process", src_buff_,
 479             dst_actor_ ? dst_actor_->get_host()->get_cname() : "a finished process", dst_buff_, buff_size);
 480
 481   /* Copy at most dst_buff_size bytes of the message to receiver's buffer */
 482   if (dst_buff_size_) {
 483     buff_size = std::min(buff_size, *(dst_buff_size_));
 484
 485     /* Update the receiver's buffer size to the copied amount */
 486     *dst_buff_size_ = buff_size;
 487   }
 488
 489   if (buff_size > 0) {
 490     if (copy_data_fun)
 491       copy_data_fun(this, src_buff_, buff_size);
 492     else
 493       SIMIX_comm_copy_data_callback(this, src_buff_, buff_size);
 494   }
 495
 496   /* Set the copied flag so we copy data only once */
 497   /* (this function might be called from both communication ends) */
 498   copied_ = true;
 499 }
 500
 501 void CommImpl::suspend()
 502 {
 503   /* FIXME: shall we suspend also the timeout synchro? */
 504   if (surf_action_)
 505     surf_action_->suspend();
 506   /* if not created yet, the action will be suspended on creation, in CommImpl::start() */
 507 }
 508
 509 void CommImpl::resume()
 510 {
 511   /*FIXME: check what happen with the timeouts */
 512   if (surf_action_)
 513     surf_action_->resume();
 514   /* in the other case, the synchro were not really suspended yet, see CommImpl::suspend() and CommImpl::start() */
 515 }
 516
 517 void CommImpl::cancel()
 518 {
 519   /* if the synchro is a waiting state means that it is still in a mbox so remove from it and delete it */
 520   if (state_ == State::WAITING) {
 521     if (not detached_) {
 522       mbox_->remove(this);
 523       state_ = State::CANCELED;
 524     }
 525   } else if (not MC_is_active() /* when running the MC there are no surf actions */
 526              && not MC_record_replay_is_active() && (state_ == State::READY || state_ == State::RUNNING)) {
 527     surf_action_->cancel();
 528   }
 529 }
 530
 531 /** @brief This is part of the cleanup process, probably an internal command */
 532 void CommImpl::cleanup_surf()
 533 {
 534   clean_action();
 535
 536   if (src_timeout_) {
 537     src_timeout_->unref();
 538     src_timeout_ = nullptr;
 539   }
 540
 541   if (dst_timeout_) {
 542     dst_timeout_->unref();
 543     dst_timeout_ = nullptr;
 544   }
 545 }
 546
 547 void CommImpl::post()
 548 {
 549   /* Update synchro state */
 550   if (src_timeout_ && src_timeout_->get_state() == resource::Action::State::FINISHED)
 551     state_ = State::SRC_TIMEOUT;
 552   else if (dst_timeout_ && dst_timeout_->get_state() == resource::Action::State::FINISHED)
 553     state_ = State::DST_TIMEOUT;
 554   else if (src_timeout_ && src_timeout_->get_state() == resource::Action::State::FAILED)
 555     state_ = State::SRC_HOST_FAILURE;
 556   else if (dst_timeout_ && dst_timeout_->get_state() == resource::Action::State::FAILED)
 557     state_ = State::DST_HOST_FAILURE;
 558   else if (surf_action_ && surf_action_->get_state() == resource::Action::State::FAILED) {
 559     state_ = State::LINK_FAILURE;
 560   } else
 561     state_ = State::DONE;
 562
 563   XBT_DEBUG("CommImpl::post(): comm %p, state %s, src_proc %p, dst_proc %p, detached: %d", this, get_state_str(),
 564             src_actor_.get(), dst_actor_.get(), detached_);
 565
 566   /* destroy the surf actions associated with the Simix communication */
 567   cleanup_surf();
 568
 569   /* Answer all simcalls associated with the synchro */
 570   finish();
 571 }
 572
 573 void CommImpl::finish()
 574 {
 575   while (not simcalls_.empty()) {
 576     smx_simcall_t simcall = simcalls_.front();
 577     simcalls_.pop_front();
 578
 579     /* If a waitany simcall is waiting for this synchro to finish, then remove it from the other synchros in the waitany
 580      * list. Afterwards, get the position of the actual synchro in the waitany list and return it as the result of the
 581      * simcall */
 582
 583     if (simcall->call_ == simix::Simcall::NONE) // FIXME: maybe a better way to handle this case
 584       continue;                                 // if actor handling comm is killed
 585     if (simcall->call_ == simix::Simcall::COMM_WAITANY) {
 586       SIMIX_waitany_remove_simcall_from_actions(simcall);
 587       if (simcall->timeout_cb_) {
 588         simcall->timeout_cb_->remove();
 589         simcall->timeout_cb_ = nullptr;
 590       }
 591       if (not MC_is_active() && not MC_record_replay_is_active()) {
 592         CommImpl** comms   = simcall_comm_waitany__get__comms(simcall);
 593         size_t count       = simcall_comm_waitany__get__count(simcall);
 594         CommImpl** element = std::find(comms, comms + count, this);
 595         int rank           = (element != comms + count) ? element - comms : -1;
 596         simcall_comm_waitany__set__result(simcall, rank);
 597       }
 598     }
 599
 600     /* If the synchro is still in a rendez-vous point then remove from it */
 601     if (mbox_)
 602       mbox_->remove(this);
 603
 604     XBT_DEBUG("CommImpl::finish(): synchro state = %d", static_cast<int>(state_));
 605
 606     /* Check out for errors */
 607
 608     if (not simcall->issuer_->get_host()->is_on()) {
 609       simcall->issuer_->context_->set_wannadie();
 610     } else {
 611       switch (state_) {
 612         case State::DONE:
 613           XBT_DEBUG("Communication %p complete!", this);
 614           copy_data();
 615           break;
 616
 617         case State::SRC_TIMEOUT:
 618           simcall->issuer_->exception_ = std::make_exception_ptr(
 619               simgrid::TimeoutException(XBT_THROW_POINT, "Communication timeouted because of the sender"));
 620           break;
 621
 622         case State::DST_TIMEOUT:
 623           simcall->issuer_->exception_ = std::make_exception_ptr(
 624               simgrid::TimeoutException(XBT_THROW_POINT, "Communication timeouted because of the receiver"));
 625           break;
 626
 627         case State::SRC_HOST_FAILURE:
 628           if (simcall->issuer_ == src_actor_)
 629             simcall->issuer_->context_->set_wannadie();
 630           else
 631             simcall->issuer_->exception_ =
 632                 std::make_exception_ptr(simgrid::NetworkFailureException(XBT_THROW_POINT, "Remote peer failed"));
 633           break;
 634
 635         case State::DST_HOST_FAILURE:
 636           if (simcall->issuer_ == dst_actor_)
 637             simcall->issuer_->context_->set_wannadie();
 638           else
 639             simcall->issuer_->exception_ =
 640                 std::make_exception_ptr(simgrid::NetworkFailureException(XBT_THROW_POINT, "Remote peer failed"));
 641           break;
 642
 643         case State::LINK_FAILURE:
 644           XBT_DEBUG("Link failure in synchro %p between '%s' and '%s': posting an exception to the issuer: %s (%p) "
 645                     "detached:%d",
 646                     this, src_actor_ ? src_actor_->get_host()->get_cname() : nullptr,
 647                     dst_actor_ ? dst_actor_->get_host()->get_cname() : nullptr, simcall->issuer_->get_cname(),
 648                     simcall->issuer_, detached_);
 649           if (src_actor_ == simcall->issuer_) {
 650             XBT_DEBUG("I'm source");
 651           } else if (dst_actor_ == simcall->issuer_) {
 652             XBT_DEBUG("I'm dest");
 653           } else {
 654             XBT_DEBUG("I'm neither source nor dest");
 655           }
 656           simcall->issuer_->throw_exception(
 657               std::make_exception_ptr(simgrid::NetworkFailureException(XBT_THROW_POINT, "Link failure")));
 658           break;
 659
 660         case State::CANCELED:
 661           if (simcall->issuer_ == dst_actor_)
 662             simcall->issuer_->exception_ = std::make_exception_ptr(
 663                 simgrid::CancelException(XBT_THROW_POINT, "Communication canceled by the sender"));
 664           else
 665             simcall->issuer_->exception_ = std::make_exception_ptr(
 666                 simgrid::CancelException(XBT_THROW_POINT, "Communication canceled by the receiver"));
 667           break;
 668
 669         default:
 670           xbt_die("Unexpected synchro state in CommImpl::finish: %d", static_cast<int>(state_));
 671       }
 672       simcall->issuer_->simcall_answer();
 673     }
 674     /* if there is an exception during a waitany or a testany, indicate the position of the failed communication */
 675     if (simcall->issuer_->exception_ &&
 676         (simcall->call_ == simix::Simcall::COMM_WAITANY || simcall->call_ == simix::Simcall::COMM_TESTANY)) {
 677       // First retrieve the rank of our failing synchro
 678       CommImpl** comms;
 679       size_t count;
 680       if (simcall->call_ == simix::Simcall::COMM_WAITANY) {
 681         comms = simcall_comm_waitany__get__comms(simcall);
 682         count = simcall_comm_waitany__get__count(simcall);
 683       } else {
 684         /* simcall->call_ == simix::Simcall::COMM_TESTANY */
 685         comms = simcall_comm_testany__get__comms(simcall);
 686         count = simcall_comm_testany__get__count(simcall);
 687       }
 688       CommImpl** element = std::find(comms, comms + count, this);
 689       int rank           = (element != comms + count) ? element - comms : -1;
 690
 691       // In order to modify the exception we have to rethrow it:
 692       try {
 693         std::rethrow_exception(simcall->issuer_->exception_);
 694       } catch (simgrid::Exception& e) {
 695         e.value = rank;
 696       }
 697     }
 698
 699     simcall->issuer_->waiting_synchro_ = nullptr;
 700     simcall->issuer_->activities_.remove(this);
 701     if (detached_) {
 702       if (simcall->issuer_ != dst_actor_ && dst_actor_ != nullptr)
 703         dst_actor_->activities_.remove(this);
 704       if (simcall->issuer_ != src_actor_ && src_actor_ != nullptr)
 705         src_actor_->activities_.remove(this);
 706     }
 707   }
 708 }
 709
 710 } // namespace activity
 711 } // namespace kernel
 712 } // namespace simgrid