src/kernel/activity/CommImpl.cpp

   1 /* Copyright (c) 2007-2020. 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, 0);
  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       simgrid::kernel::activity::CommImplPtr(new simgrid::kernel::activity::CommImpl());
  46   this_comm->set_type(simgrid::kernel::activity::CommImpl::Type::SEND);
  47
  48   /* Look for communication synchro matching our needs. We also provide a description of
  49    * ourself so that the other side also gets a chance of choosing if it wants to match with us.
  50    *
  51    * If it is not found then push our communication into the rendez-vous point */
  52   simgrid::kernel::activity::CommImplPtr other_comm =
  53       mbox->find_matching_comm(simgrid::kernel::activity::CommImpl::Type::RECEIVE, match_fun, data, this_comm,
  54                                /*done*/ false, /*remove_matching*/ true);
  55
  56   if (not other_comm) {
  57     other_comm = std::move(this_comm);
  58
  59     if (mbox->permanent_receiver_ != nullptr) {
  60       // this mailbox is for small messages, which have to be sent right now
  61       other_comm->state_     = simgrid::kernel::activity::State::READY;
  62       other_comm->dst_actor_ = mbox->permanent_receiver_.get();
  63       mbox->done_comm_queue_.push_back(other_comm);
  64       XBT_DEBUG("pushing a message into the permanent receive list %p, comm %p", mbox, other_comm.get());
  65
  66     } else {
  67       mbox->push(other_comm);
  68     }
  69   } else {
  70     XBT_DEBUG("Receive already pushed");
  71
  72     other_comm->state_ = simgrid::kernel::activity::State::READY;
  73     other_comm->set_type(simgrid::kernel::activity::CommImpl::Type::READY);
  74   }
  75
  76   if (detached) {
  77     other_comm->detach();
  78     other_comm->clean_fun = clean_fun;
  79   } else {
  80     other_comm->clean_fun = nullptr;
  81     src_proc->comms.push_back(other_comm);
  82   }
  83
  84   /* Setup the communication synchro */
  85   other_comm->src_actor_     = src_proc;
  86   other_comm->src_data_      = data;
  87   (*other_comm).set_src_buff(src_buff, src_buff_size).set_size(task_size).set_rate(rate);
  88
  89   other_comm->match_fun     = match_fun;
  90   other_comm->copy_data_fun = copy_data_fun;
  91
  92   if (MC_is_active() || MC_record_replay_is_active())
  93     other_comm->state_ = simgrid::kernel::activity::State::RUNNING;
  94   else
  95     other_comm->start();
  96
  97   return (detached ? nullptr : other_comm);
  98 }
  99
 100 XBT_PRIVATE void simcall_HANDLER_comm_recv(smx_simcall_t simcall, smx_actor_t receiver, smx_mailbox_t mbox,
 101                                            unsigned char* dst_buff, size_t* dst_buff_size,
 102                                            bool (*match_fun)(void*, void*, simgrid::kernel::activity::CommImpl*),
 103                                            void (*copy_data_fun)(simgrid::kernel::activity::CommImpl*, void*, size_t),
 104                                            void* data, double timeout, double rate)
 105 {
 106   simgrid::kernel::activity::ActivityImplPtr comm = simcall_HANDLER_comm_irecv(
 107       simcall, receiver, mbox, dst_buff, dst_buff_size, match_fun, copy_data_fun, data, rate);
 108   SIMCALL_SET_MC_VALUE(*simcall, 0);
 109   simcall_HANDLER_comm_wait(simcall, static_cast<simgrid::kernel::activity::CommImpl*>(comm.get()), timeout);
 110 }
 111
 112 XBT_PRIVATE simgrid::kernel::activity::ActivityImplPtr
 113 simcall_HANDLER_comm_irecv(smx_simcall_t /*simcall*/, smx_actor_t receiver, smx_mailbox_t mbox, unsigned char* dst_buff,
 114                            size_t* dst_buff_size, bool (*match_fun)(void*, void*, simgrid::kernel::activity::CommImpl*),
 115                            void (*copy_data_fun)(simgrid::kernel::activity::CommImpl*, void*, size_t), void* data,
 116                            double rate)
 117 {
 118   simgrid::kernel::activity::CommImplPtr this_synchro =
 119       simgrid::kernel::activity::CommImplPtr(new simgrid::kernel::activity::CommImpl());
 120   this_synchro->set_type(simgrid::kernel::activity::CommImpl::Type::RECEIVE);
 121   XBT_DEBUG("recv from mbox %p. this_synchro=%p", mbox, this_synchro.get());
 122
 123   simgrid::kernel::activity::CommImplPtr other_comm;
 124   // communication already done, get it inside the list of completed comms
 125   if (mbox->permanent_receiver_ != nullptr && not mbox->done_comm_queue_.empty()) {
 126     XBT_DEBUG("We have a comm that has probably already been received, trying to match it, to skip the communication");
 127     // find a match in the list of already received comms
 128     other_comm = mbox->find_matching_comm(simgrid::kernel::activity::CommImpl::Type::SEND, match_fun, data,
 129                                           this_synchro, /*done*/ true,
 130                                           /*remove_matching*/ true);
 131     // if not found, assume the receiver came first, register it to the mailbox in the classical way
 132     if (not other_comm) {
 133       XBT_DEBUG("We have messages in the permanent receive list, but not the one we are looking for, pushing request "
 134                 "into list");
 135       other_comm = std::move(this_synchro);
 136       mbox->push(other_comm);
 137     } else {
 138       if (other_comm->surf_action_ && other_comm->get_remaining() < 1e-12) {
 139         XBT_DEBUG("comm %p has been already sent, and is finished, destroy it", other_comm.get());
 140         other_comm->state_ = simgrid::kernel::activity::State::DONE;
 141         other_comm->set_type(simgrid::kernel::activity::CommImpl::Type::DONE).set_mailbox(nullptr);
 142       }
 143     }
 144   } else {
 145     /* Prepare a comm describing us, so that it gets passed to the user-provided filter of other side */
 146
 147     /* Look for communication activity matching our needs. We also provide a description of
 148      * ourself so that the other side also gets a chance of choosing if it wants to match with us.
 149      *
 150      * If it is not found then push our communication into the rendez-vous point */
 151     other_comm = mbox->find_matching_comm(simgrid::kernel::activity::CommImpl::Type::SEND, match_fun, data,
 152                                           this_synchro, /*done*/ false,
 153                                           /*remove_matching*/ true);
 154
 155     if (other_comm == nullptr) {
 156       XBT_DEBUG("Receive pushed first (%zu comm enqueued so far)", mbox->comm_queue_.size());
 157       other_comm = std::move(this_synchro);
 158       mbox->push(other_comm);
 159     } else {
 160       XBT_DEBUG("Match my %p with the existing %p", this_synchro.get(), other_comm.get());
 161
 162       other_comm->state_ = simgrid::kernel::activity::State::READY;
 163       other_comm->set_type(simgrid::kernel::activity::CommImpl::Type::READY);
 164     }
 165     receiver->comms.push_back(other_comm);
 166   }
 167
 168   /* Setup communication synchro */
 169   other_comm->dst_actor_     = receiver;
 170   other_comm->dst_data_      = data;
 171   other_comm->set_dst_buff(dst_buff, dst_buff_size);
 172
 173   if (rate > -1.0 && (other_comm->get_rate() < 0.0 || rate < other_comm->get_rate()))
 174     other_comm->set_rate(rate);
 175
 176   other_comm->match_fun     = match_fun;
 177   other_comm->copy_data_fun = copy_data_fun;
 178
 179   if (MC_is_active() || MC_record_replay_is_active()) {
 180     other_comm->state_ = simgrid::kernel::activity::State::RUNNING;
 181     return other_comm;
 182   }
 183   other_comm->start();
 184   return other_comm;
 185 }
 186
 187 void simcall_HANDLER_comm_wait(smx_simcall_t simcall, simgrid::kernel::activity::CommImpl* comm, double timeout)
 188 {
 189   /* Associate this simcall to the wait synchro */
 190   XBT_DEBUG("simcall_HANDLER_comm_wait, %p", comm);
 191
 192   comm->register_simcall(simcall);
 193
 194   if (MC_is_active() || MC_record_replay_is_active()) {
 195     int idx = SIMCALL_GET_MC_VALUE(*simcall);
 196     if (idx == 0) {
 197       comm->state_ = simgrid::kernel::activity::State::DONE;
 198     } else {
 199       /* If we reached this point, the wait simcall must have a timeout */
 200       /* Otherwise it shouldn't be enabled and executed by the MC */
 201       if (timeout < 0.0)
 202         THROW_IMPOSSIBLE;
 203
 204       if (comm->src_actor_ == simcall->issuer_)
 205         comm->state_ = simgrid::kernel::activity::State::SRC_TIMEOUT;
 206       else
 207         comm->state_ = simgrid::kernel::activity::State::DST_TIMEOUT;
 208     }
 209
 210     comm->finish();
 211     return;
 212   }
 213
 214   /* If the synchro has already finish perform the error handling, */
 215   /* otherwise set up a waiting timeout on the right side          */
 216   if (comm->state_ != simgrid::kernel::activity::State::WAITING &&
 217       comm->state_ != simgrid::kernel::activity::State::RUNNING) {
 218     comm->finish();
 219   } else { /* we need a sleep action (even when there is no timeout) to be notified of host failures */
 220     simgrid::kernel::resource::Action* sleep = simcall->issuer_->get_host()->pimpl_cpu->sleep(timeout);
 221     sleep->set_activity(comm);
 222
 223     if (simcall->issuer_ == comm->src_actor_)
 224       comm->src_timeout_ = sleep;
 225     else
 226       comm->dst_timeout_ = sleep;
 227   }
 228 }
 229
 230 void simcall_HANDLER_comm_test(smx_simcall_t simcall, simgrid::kernel::activity::CommImpl* comm)
 231 {
 232   bool res;
 233
 234   if (MC_is_active() || MC_record_replay_is_active()) {
 235     res = comm->src_actor_ && comm->dst_actor_;
 236     if (res)
 237       comm->state_ = simgrid::kernel::activity::State::DONE;
 238   } else {
 239     res = comm->state_ != simgrid::kernel::activity::State::WAITING &&
 240           comm->state_ != simgrid::kernel::activity::State::RUNNING;
 241   }
 242
 243   simcall_comm_test__set__result(simcall, res);
 244   if (res) {
 245     comm->simcalls_.push_back(simcall);
 246     comm->finish();
 247   } else {
 248     simcall->issuer_->simcall_answer();
 249   }
 250 }
 251
 252 void simcall_HANDLER_comm_testany(smx_simcall_t simcall, simgrid::kernel::activity::CommImpl* comms[], size_t count)
 253 {
 254   // The default result is -1 -- this means, "nothing is ready".
 255   // It can be changed below, but only if something matches.
 256   simcall_comm_testany__set__result(simcall, -1);
 257
 258   if (MC_is_active() || MC_record_replay_is_active()) {
 259     int idx = SIMCALL_GET_MC_VALUE(*simcall);
 260     if (idx == -1) {
 261       simcall->issuer_->simcall_answer();
 262     } else {
 263       simgrid::kernel::activity::CommImpl* comm = comms[idx];
 264       simcall_comm_testany__set__result(simcall, idx);
 265       comm->simcalls_.push_back(simcall);
 266       comm->state_ = simgrid::kernel::activity::State::DONE;
 267       comm->finish();
 268     }
 269     return;
 270   }
 271
 272   for (std::size_t i = 0; i != count; ++i) {
 273     simgrid::kernel::activity::CommImpl* comm = comms[i];
 274     if (comm->state_ != simgrid::kernel::activity::State::WAITING &&
 275         comm->state_ != simgrid::kernel::activity::State::RUNNING) {
 276       simcall_comm_testany__set__result(simcall, i);
 277       comm->simcalls_.push_back(simcall);
 278       comm->finish();
 279       return;
 280     }
 281   }
 282   simcall->issuer_->simcall_answer();
 283 }
 284
 285 static void SIMIX_waitany_remove_simcall_from_actions(smx_simcall_t simcall)
 286 {
 287   simgrid::kernel::activity::CommImpl** comms = simcall_comm_waitany__get__comms(simcall);
 288   size_t count                                = simcall_comm_waitany__get__count(simcall);
 289
 290   for (size_t i = 0; i < count; i++) {
 291     // Remove the first occurrence of simcall:
 292     auto* comm = comms[i];
 293     auto j     = boost::range::find(comm->simcalls_, simcall);
 294     if (j != comm->simcalls_.end())
 295       comm->simcalls_.erase(j);
 296   }
 297 }
 298 void simcall_HANDLER_comm_waitany(smx_simcall_t simcall, simgrid::kernel::activity::CommImpl* comms[], size_t count,
 299                                   double timeout)
 300 {
 301   if (MC_is_active() || MC_record_replay_is_active()) {
 302     if (timeout > 0.0)
 303       xbt_die("Timeout not implemented for waitany in the model-checker");
 304     int idx                 = SIMCALL_GET_MC_VALUE(*simcall);
 305     auto* comm              = comms[idx];
 306     comm->simcalls_.push_back(simcall);
 307     simcall_comm_waitany__set__result(simcall, idx);
 308     comm->state_ = simgrid::kernel::activity::State::DONE;
 309     comm->finish();
 310     return;
 311   }
 312
 313   if (timeout < 0.0) {
 314     simcall->timeout_cb_ = NULL;
 315   } else {
 316     simcall->timeout_cb_ = simgrid::simix::Timer::set(SIMIX_get_clock() + timeout, [simcall]() {
 317       SIMIX_waitany_remove_simcall_from_actions(simcall);
 318       simcall_comm_waitany__set__result(simcall, -1);
 319       simcall->issuer_->simcall_answer();
 320     });
 321   }
 322
 323   for (size_t i = 0; i < count; i++) {
 324     /* associate this simcall to the the synchro */
 325     auto* comm = comms[i];
 326     comm->simcalls_.push_back(simcall);
 327
 328     /* see if the synchro is already finished */
 329     if (comm->state_ != simgrid::kernel::activity::State::WAITING &&
 330         comm->state_ != simgrid::kernel::activity::State::RUNNING) {
 331       comm->finish();
 332       break;
 333     }
 334   }
 335 }
 336
 337 /******************************************************************************/
 338 /*                    SIMIX_comm_copy_data callbacks                       */
 339 /******************************************************************************/
 340 static void (*SIMIX_comm_copy_data_callback)(simgrid::kernel::activity::CommImpl*, void*,
 341                                              size_t) = &SIMIX_comm_copy_pointer_callback;
 342
 343 void SIMIX_comm_copy_buffer_callback(simgrid::kernel::activity::CommImpl* comm, void* buff, size_t buff_size)
 344 {
 345   XBT_DEBUG("Copy the data over");
 346   memcpy(comm->dst_buff_, buff, buff_size);
 347   if (comm->detached()) { // if this is a detached send, the source buffer was duplicated by SMPI sender to make the
 348                           // original buffer available to the application ASAP
 349     xbt_free(buff);
 350     comm->src_buff_ = nullptr;
 351   }
 352 }
 353
 354 void SIMIX_comm_set_copy_data_callback(void (*callback)(simgrid::kernel::activity::CommImpl*, void*, size_t))
 355 {
 356   SIMIX_comm_copy_data_callback = callback;
 357 }
 358
 359 void SIMIX_comm_copy_pointer_callback(simgrid::kernel::activity::CommImpl* comm, void* buff, size_t buff_size)
 360 {
 361   xbt_assert((buff_size == sizeof(void*)), "Cannot copy %zu bytes: must be sizeof(void*)", buff_size);
 362   *(void**)(comm->dst_buff_) = buff;
 363 }
 364
 365 namespace simgrid {
 366 namespace kernel {
 367 namespace activity {
 368
 369 CommImpl& CommImpl::set_type(CommImpl::Type type)
 370 {
 371   type_ = type;
 372   return *this;
 373 }
 374
 375 CommImpl& CommImpl::set_size(double size)
 376 {
 377   size_ = size;
 378   return *this;
 379 }
 380
 381 CommImpl& CommImpl::set_rate(double rate)
 382 {
 383   rate_ = rate;
 384   return *this;
 385 }
 386 CommImpl& CommImpl::set_mailbox(MailboxImpl* mbox)
 387 {
 388   mbox_ = mbox;
 389   return *this;
 390 }
 391
 392 CommImpl& CommImpl::set_src_buff(unsigned char* buff, size_t size)
 393 {
 394   src_buff_      = buff;
 395   src_buff_size_ = size;
 396   return *this;
 397 }
 398
 399 CommImpl& CommImpl::set_dst_buff(unsigned char* buff, size_t* size)
 400 {
 401   dst_buff_      = buff;
 402   dst_buff_size_ = size;
 403   return *this;
 404 }
 405
 406 CommImpl& CommImpl::detach()
 407 {
 408   detached_ = true;
 409   return *this;
 410 }
 411
 412 CommImpl::~CommImpl()
 413 {
 414   XBT_DEBUG("Really free communication %p in state %d (detached = %d)", this, static_cast<int>(state_), detached_);
 415
 416   cleanupSurf();
 417
 418   if (detached_ && state_ != State::DONE) {
 419     /* the communication has failed and was detached:
 420      * we have to free the buffer */
 421     if (clean_fun)
 422       clean_fun(src_buff_);
 423     src_buff_ = nullptr;
 424   } else if (mbox_) {
 425     mbox_->remove(this);
 426   }
 427 }
 428
 429 /**  @brief Starts the simulation of a communication synchro. */
 430 CommImpl* CommImpl::start()
 431 {
 432   /* If both the sender and the receiver are already there, start the communication */
 433   if (state_ == State::READY) {
 434     s4u::Host* sender   = src_actor_->get_host();
 435     s4u::Host* receiver = dst_actor_->get_host();
 436
 437     surf_action_ = surf_network_model->communicate(sender, receiver, size_, rate_);
 438     surf_action_->set_activity(this);
 439     surf_action_->set_category(get_tracing_category());
 440     state_ = State::RUNNING;
 441
 442     XBT_DEBUG("Starting communication %p from '%s' to '%s' (surf_action: %p)", this, sender->get_cname(),
 443               receiver->get_cname(), surf_action_);
 444
 445     /* If a link is failed, detect it immediately */
 446     if (surf_action_->get_state() == resource::Action::State::FAILED) {
 447       XBT_DEBUG("Communication from '%s' to '%s' failed to start because of a link failure", sender->get_cname(),
 448                 receiver->get_cname());
 449       state_ = State::LINK_FAILURE;
 450       post();
 451
 452     } else if (src_actor_->is_suspended() || dst_actor_->is_suspended()) {
 453       /* If any of the process is suspended, create the synchro but stop its execution,
 454          it will be restarted when the sender process resume */
 455       if (src_actor_->is_suspended())
 456         XBT_DEBUG("The communication is suspended on startup because src (%s@%s) was suspended since it initiated the "
 457                   "communication",
 458                   src_actor_->get_cname(), src_actor_->get_host()->get_cname());
 459       else
 460         XBT_DEBUG("The communication is suspended on startup because dst (%s@%s) was suspended since it initiated the "
 461                   "communication",
 462                   dst_actor_->get_cname(), dst_actor_->get_host()->get_cname());
 463
 464       surf_action_->suspend();
 465     }
 466   }
 467
 468   return this;
 469 }
 470
 471 /** @brief Copy the communication data from the sender's buffer to the receiver's one  */
 472 void CommImpl::copy_data()
 473 {
 474   size_t buff_size = src_buff_size_;
 475   /* If there is no data to copy then return */
 476   if (not src_buff_ || not dst_buff_ || copied_)
 477     return;
 478
 479   XBT_DEBUG("Copying comm %p data from %s (%p) -> %s (%p) (%zu bytes)", this,
 480             src_actor_ ? src_actor_->get_host()->get_cname() : "a finished process", src_buff_,
 481             dst_actor_ ? dst_actor_->get_host()->get_cname() : "a finished process", dst_buff_, buff_size);
 482
 483   /* Copy at most dst_buff_size bytes of the message to receiver's buffer */
 484   if (dst_buff_size_)
 485     buff_size = std::min(buff_size, *(dst_buff_size_));
 486
 487   /* Update the receiver's buffer size to the copied amount */
 488   if (dst_buff_size_)
 489     *dst_buff_size_ = buff_size;
 490
 491   if (buff_size > 0) {
 492     if (copy_data_fun)
 493       copy_data_fun(this, src_buff_, buff_size);
 494     else
 495       SIMIX_comm_copy_data_callback(this, src_buff_, buff_size);
 496   }
 497
 498   /* Set the copied flag so we copy data only once */
 499   /* (this function might be called from both communication ends) */
 500   copied_ = true;
 501 }
 502
 503 void CommImpl::suspend()
 504 {
 505   /* FIXME: shall we suspend also the timeout synchro? */
 506   if (surf_action_)
 507     surf_action_->suspend();
 508   /* if not created yet, the action will be suspended on creation, in CommImpl::start() */
 509 }
 510
 511 void CommImpl::resume()
 512 {
 513   /*FIXME: check what happen with the timeouts */
 514   if (surf_action_)
 515     surf_action_->resume();
 516   /* in the other case, the synchro were not really suspended yet, see CommImpl::suspend() and CommImpl::start() */
 517 }
 518
 519 void CommImpl::cancel()
 520 {
 521   /* if the synchro is a waiting state means that it is still in a mbox so remove from it and delete it */
 522   if (state_ == State::WAITING) {
 523     if (not detached_) {
 524       mbox_->remove(this);
 525       state_ = State::CANCELED;
 526     }
 527   } else if (not MC_is_active() /* when running the MC there are no surf actions */
 528              && not MC_record_replay_is_active() && (state_ == State::READY || state_ == State::RUNNING)) {
 529     surf_action_->cancel();
 530   }
 531 }
 532
 533 /** @brief This is part of the cleanup process, probably an internal command */
 534 void CommImpl::cleanupSurf()
 535 {
 536   clean_action();
 537
 538   if (src_timeout_) {
 539     src_timeout_->unref();
 540     src_timeout_ = nullptr;
 541   }
 542
 543   if (dst_timeout_) {
 544     dst_timeout_->unref();
 545     dst_timeout_ = nullptr;
 546   }
 547 }
 548
 549 void CommImpl::post()
 550 {
 551   /* Update synchro state */
 552   if (src_timeout_ && src_timeout_->get_state() == resource::Action::State::FINISHED)
 553     state_ = State::SRC_TIMEOUT;
 554   else if (dst_timeout_ && dst_timeout_->get_state() == resource::Action::State::FINISHED)
 555     state_ = State::DST_TIMEOUT;
 556   else if (src_timeout_ && src_timeout_->get_state() == resource::Action::State::FAILED)
 557     state_ = State::SRC_HOST_FAILURE;
 558   else if (dst_timeout_ && dst_timeout_->get_state() == resource::Action::State::FAILED)
 559     state_ = State::DST_HOST_FAILURE;
 560   else if (surf_action_ && surf_action_->get_state() == resource::Action::State::FAILED) {
 561     state_ = State::LINK_FAILURE;
 562   } else
 563     state_ = State::DONE;
 564
 565   XBT_DEBUG("SIMIX_post_comm: comm %p, state %d, src_proc %p, dst_proc %p, detached: %d", this, (int)state_,
 566             src_actor_.get(), dst_actor_.get(), detached_);
 567
 568   /* destroy the surf actions associated with the Simix communication */
 569   cleanupSurf();
 570
 571   /* Answer all simcalls associated with the synchro */
 572   finish();
 573 }
 574
 575 void CommImpl::finish()
 576 {
 577   while (not simcalls_.empty()) {
 578     smx_simcall_t simcall = simcalls_.front();
 579     simcalls_.pop_front();
 580
 581     /* If a waitany simcall is waiting for this synchro to finish, then remove it from the other synchros in the waitany
 582      * list. Afterwards, get the position of the actual synchro in the waitany list and return it as the result of the
 583      * simcall */
 584
 585     if (simcall->call_ == SIMCALL_NONE) // FIXME: maybe a better way to handle this case
 586       continue;                         // if actor handling comm is killed
 587     if (simcall->call_ == SIMCALL_COMM_WAITANY) {
 588       SIMIX_waitany_remove_simcall_from_actions(simcall);
 589       if (simcall->timeout_cb_) {
 590         simcall->timeout_cb_->remove();
 591         simcall->timeout_cb_ = nullptr;
 592       }
 593       if (not MC_is_active() && not MC_record_replay_is_active()) {
 594         CommImpl** comms   = simcall_comm_waitany__get__comms(simcall);
 595         size_t count       = simcall_comm_waitany__get__count(simcall);
 596         CommImpl** element = std::find(comms, comms + count, this);
 597         int rank           = (element != comms + count) ? element - comms : -1;
 598         simcall_comm_waitany__set__result(simcall, rank);
 599       }
 600     }
 601
 602     /* If the synchro is still in a rendez-vous point then remove from it */
 603     if (mbox_)
 604       mbox_->remove(this);
 605
 606     XBT_DEBUG("CommImpl::finish(): synchro state = %d", static_cast<int>(state_));
 607
 608     /* Check out for errors */
 609
 610     if (not simcall->issuer_->get_host()->is_on()) {
 611       simcall->issuer_->context_->set_wannadie();
 612     } else {
 613       switch (state_) {
 614         case State::DONE:
 615           XBT_DEBUG("Communication %p complete!", this);
 616           copy_data();
 617           break;
 618
 619         case State::SRC_TIMEOUT:
 620           simcall->issuer_->exception_ = std::make_exception_ptr(
 621               simgrid::TimeoutException(XBT_THROW_POINT, "Communication timeouted because of the sender"));
 622           break;
 623
 624         case State::DST_TIMEOUT:
 625           simcall->issuer_->exception_ = std::make_exception_ptr(
 626               simgrid::TimeoutException(XBT_THROW_POINT, "Communication timeouted because of the receiver"));
 627           break;
 628
 629         case State::SRC_HOST_FAILURE:
 630           if (simcall->issuer_ == src_actor_)
 631             simcall->issuer_->context_->set_wannadie();
 632           else
 633             simcall->issuer_->exception_ =
 634                 std::make_exception_ptr(simgrid::NetworkFailureException(XBT_THROW_POINT, "Remote peer failed"));
 635           break;
 636
 637         case State::DST_HOST_FAILURE:
 638           if (simcall->issuer_ == dst_actor_)
 639             simcall->issuer_->context_->set_wannadie();
 640           else
 641             simcall->issuer_->exception_ =
 642                 std::make_exception_ptr(simgrid::NetworkFailureException(XBT_THROW_POINT, "Remote peer failed"));
 643           break;
 644
 645         case State::LINK_FAILURE:
 646           XBT_DEBUG("Link failure in synchro %p between '%s' and '%s': posting an exception to the issuer: %s (%p) "
 647                     "detached:%d",
 648                     this, src_actor_ ? src_actor_->get_host()->get_cname() : nullptr,
 649                     dst_actor_ ? dst_actor_->get_host()->get_cname() : nullptr, simcall->issuer_->get_cname(),
 650                     simcall->issuer_, detached_);
 651           if (src_actor_ == simcall->issuer_) {
 652             XBT_DEBUG("I'm source");
 653           } else if (dst_actor_ == simcall->issuer_) {
 654             XBT_DEBUG("I'm dest");
 655           } else {
 656             XBT_DEBUG("I'm neither source nor dest");
 657           }
 658           simcall->issuer_->throw_exception(
 659               std::make_exception_ptr(simgrid::NetworkFailureException(XBT_THROW_POINT, "Link failure")));
 660           break;
 661
 662         case State::CANCELED:
 663           if (simcall->issuer_ == dst_actor_)
 664             simcall->issuer_->exception_ = std::make_exception_ptr(
 665                 simgrid::CancelException(XBT_THROW_POINT, "Communication canceled by the sender"));
 666           else
 667             simcall->issuer_->exception_ = std::make_exception_ptr(
 668                 simgrid::CancelException(XBT_THROW_POINT, "Communication canceled by the receiver"));
 669           break;
 670
 671         default:
 672           xbt_die("Unexpected synchro state in CommImpl::finish: %d", static_cast<int>(state_));
 673       }
 674       simcall->issuer_->simcall_answer();
 675     }
 676     /* if there is an exception during a waitany or a testany, indicate the position of the failed communication */
 677     if (simcall->issuer_->exception_ &&
 678         (simcall->call_ == SIMCALL_COMM_WAITANY || simcall->call_ == SIMCALL_COMM_TESTANY)) {
 679       // First retrieve the rank of our failing synchro
 680       CommImpl** comms;
 681       size_t count;
 682       if (simcall->call_ == SIMCALL_COMM_WAITANY) {
 683         comms = simcall_comm_waitany__get__comms(simcall);
 684         count = simcall_comm_waitany__get__count(simcall);
 685       } else {
 686         /* simcall->call_ == SIMCALL_COMM_TESTANY */
 687         comms = simcall_comm_testany__get__comms(simcall);
 688         count = simcall_comm_testany__get__count(simcall);
 689       }
 690       CommImpl** element = std::find(comms, comms + count, this);
 691       int rank           = (element != comms + count) ? element - comms : -1;
 692
 693       // In order to modify the exception we have to rethrow it:
 694       try {
 695         std::rethrow_exception(simcall->issuer_->exception_);
 696       } catch (simgrid::Exception& e) {
 697         e.value = rank;
 698       }
 699     }
 700
 701     simcall->issuer_->waiting_synchro = nullptr;
 702     simcall->issuer_->comms.remove(this);
 703     if (detached_) {
 704       if (simcall->issuer_ == src_actor_) {
 705         if (dst_actor_)
 706           dst_actor_->comms.remove(this);
 707       } else if (simcall->issuer_ == dst_actor_) {
 708         if (src_actor_)
 709           src_actor_->comms.remove(this);
 710       } else {
 711         dst_actor_->comms.remove(this);
 712         src_actor_->comms.remove(this);
 713       }
 714     }
 715   }
 716 }
 717
 718 } // namespace activity
 719 } // namespace kernel
 720 } // namespace simgrid