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(new simgrid::kernel::activity::CommImpl());
  45   this_comm->set_type(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     other_comm->set_type(simgrid::kernel::activity::CommImpl::Type::READY);
  73   }
  74
  75   if (detached) {
  76     other_comm->detach();
  77     other_comm->clean_fun = clean_fun;
  78   } else {
  79     other_comm->clean_fun = nullptr;
  80     src_proc->activities_.emplace_back(other_comm);
  81   }
  82
  83   /* Setup the communication synchro */
  84   other_comm->src_actor_     = src_proc;
  85   other_comm->src_data_      = data;
  86   (*other_comm).set_src_buff(src_buff, src_buff_size).set_size(task_size).set_rate(rate);
  87
  88   other_comm->match_fun     = match_fun;
  89   other_comm->copy_data_fun = copy_data_fun;
  90
  91   if (MC_is_active() || MC_record_replay_is_active())
  92     other_comm->state_ = simgrid::kernel::activity::State::RUNNING;
  93   else
  94     other_comm->start();
  95
  96   return (detached ? nullptr : other_comm);
  97 }
  98
  99 XBT_PRIVATE void simcall_HANDLER_comm_recv(smx_simcall_t simcall, smx_actor_t receiver, smx_mailbox_t mbox,
 100                                            unsigned char* dst_buff, size_t* dst_buff_size,
 101                                            bool (*match_fun)(void*, void*, simgrid::kernel::activity::CommImpl*),
 102                                            void (*copy_data_fun)(simgrid::kernel::activity::CommImpl*, void*, size_t),
 103                                            void* data, double timeout, double rate)
 104 {
 105   simgrid::kernel::activity::ActivityImplPtr comm = simcall_HANDLER_comm_irecv(
 106       simcall, receiver, mbox, dst_buff, dst_buff_size, match_fun, copy_data_fun, data, rate);
 107   SIMCALL_SET_MC_VALUE(*simcall, 0);
 108   simcall_HANDLER_comm_wait(simcall, static_cast<simgrid::kernel::activity::CommImpl*>(comm.get()), timeout);
 109 }
 110
 111 XBT_PRIVATE simgrid::kernel::activity::ActivityImplPtr
 112 simcall_HANDLER_comm_irecv(smx_simcall_t /*simcall*/, smx_actor_t receiver, smx_mailbox_t mbox, unsigned char* dst_buff,
 113                            size_t* dst_buff_size, bool (*match_fun)(void*, void*, simgrid::kernel::activity::CommImpl*),
 114                            void (*copy_data_fun)(simgrid::kernel::activity::CommImpl*, void*, size_t), void* data,
 115                            double rate)
 116 {
 117   simgrid::kernel::activity::CommImplPtr this_synchro(new simgrid::kernel::activity::CommImpl());
 118   this_synchro->set_type(simgrid::kernel::activity::CommImpl::Type::RECEIVE);
 119   XBT_DEBUG("recv from mbox %p. this_synchro=%p", mbox, this_synchro.get());
 120
 121   simgrid::kernel::activity::CommImplPtr other_comm;
 122   // communication already done, get it inside the list of completed comms
 123   if (mbox->permanent_receiver_ != nullptr && not mbox->done_comm_queue_.empty()) {
 124     XBT_DEBUG("We have a comm that has probably already been received, trying to match it, to skip the communication");
 125     // find a match in the list of already received comms
 126     other_comm = mbox->find_matching_comm(simgrid::kernel::activity::CommImpl::Type::SEND, match_fun, data,
 127                                           this_synchro, /*done*/ true,
 128                                           /*remove_matching*/ true);
 129     // if not found, assume the receiver came first, register it to the mailbox in the classical way
 130     if (not other_comm) {
 131       XBT_DEBUG("We have messages in the permanent receive list, but not the one we are looking for, pushing request "
 132                 "into list");
 133       other_comm = std::move(this_synchro);
 134       mbox->push(other_comm);
 135     } else {
 136       if (other_comm->surf_action_ && other_comm->get_remaining() < 1e-12) {
 137         XBT_DEBUG("comm %p has been already sent, and is finished, destroy it", other_comm.get());
 138         other_comm->state_ = simgrid::kernel::activity::State::DONE;
 139         other_comm->set_type(simgrid::kernel::activity::CommImpl::Type::DONE).set_mailbox(nullptr);
 140       }
 141     }
 142   } else {
 143     /* Prepare a comm describing us, so that it gets passed to the user-provided filter of other side */
 144
 145     /* Look for communication activity matching our needs. We also provide a description of
 146      * ourself so that the other side also gets a chance of choosing if it wants to match with us.
 147      *
 148      * If it is not found then push our communication into the rendez-vous point */
 149     other_comm = mbox->find_matching_comm(simgrid::kernel::activity::CommImpl::Type::SEND, match_fun, data,
 150                                           this_synchro, /*done*/ false,
 151                                           /*remove_matching*/ true);
 152
 153     if (other_comm == nullptr) {
 154       XBT_DEBUG("Receive pushed first (%zu comm enqueued so far)", mbox->comm_queue_.size());
 155       other_comm = std::move(this_synchro);
 156       mbox->push(other_comm);
 157     } else {
 158       XBT_DEBUG("Match my %p with the existing %p", this_synchro.get(), other_comm.get());
 159
 160       other_comm->state_ = simgrid::kernel::activity::State::READY;
 161       other_comm->set_type(simgrid::kernel::activity::CommImpl::Type::READY);
 162     }
 163     receiver->activities_.emplace_back(other_comm);
 164   }
 165
 166   /* Setup communication synchro */
 167   other_comm->dst_actor_     = receiver;
 168   other_comm->dst_data_      = data;
 169   other_comm->set_dst_buff(dst_buff, dst_buff_size);
 170
 171   if (rate > -1.0 && (other_comm->get_rate() < 0.0 || rate < other_comm->get_rate()))
 172     other_comm->set_rate(rate);
 173
 174   other_comm->match_fun     = match_fun;
 175   other_comm->copy_data_fun = copy_data_fun;
 176
 177   if (MC_is_active() || MC_record_replay_is_active()) {
 178     other_comm->state_ = simgrid::kernel::activity::State::RUNNING;
 179     return other_comm;
 180   }
 181   other_comm->start();
 182   return other_comm;
 183 }
 184
 185 void simcall_HANDLER_comm_wait(smx_simcall_t simcall, simgrid::kernel::activity::CommImpl* comm, double timeout)
 186 {
 187   /* Associate this simcall to the wait synchro */
 188   XBT_DEBUG("simcall_HANDLER_comm_wait, %p", comm);
 189
 190   comm->register_simcall(simcall);
 191
 192   if (MC_is_active() || MC_record_replay_is_active()) {
 193     int idx = SIMCALL_GET_MC_VALUE(*simcall);
 194     if (idx == 0) {
 195       comm->state_ = simgrid::kernel::activity::State::DONE;
 196     } else {
 197       /* If we reached this point, the wait simcall must have a timeout */
 198       /* Otherwise it shouldn't be enabled and executed by the MC */
 199       if (timeout < 0.0)
 200         THROW_IMPOSSIBLE;
 201
 202       if (comm->src_actor_ == simcall->issuer_)
 203         comm->state_ = simgrid::kernel::activity::State::SRC_TIMEOUT;
 204       else
 205         comm->state_ = simgrid::kernel::activity::State::DST_TIMEOUT;
 206     }
 207
 208     comm->finish();
 209     return;
 210   }
 211
 212   /* If the synchro has already finish perform the error handling, */
 213   /* otherwise set up a waiting timeout on the right side          */
 214   if (comm->state_ != simgrid::kernel::activity::State::WAITING &&
 215       comm->state_ != simgrid::kernel::activity::State::RUNNING) {
 216     comm->finish();
 217   } else { /* we need a sleep action (even when there is no timeout) to be notified of host failures */
 218     simgrid::kernel::resource::Action* sleep = simcall->issuer_->get_host()->pimpl_cpu->sleep(timeout);
 219     sleep->set_activity(comm);
 220
 221     if (simcall->issuer_ == comm->src_actor_)
 222       comm->src_timeout_ = sleep;
 223     else
 224       comm->dst_timeout_ = sleep;
 225   }
 226 }
 227
 228 void simcall_HANDLER_comm_test(smx_simcall_t simcall, simgrid::kernel::activity::CommImpl* comm)
 229 {
 230   bool res;
 231
 232   if (MC_is_active() || MC_record_replay_is_active()) {
 233     res = comm->src_actor_ && comm->dst_actor_;
 234     if (res)
 235       comm->state_ = simgrid::kernel::activity::State::DONE;
 236   } else {
 237     res = comm->state_ != simgrid::kernel::activity::State::WAITING &&
 238           comm->state_ != simgrid::kernel::activity::State::RUNNING;
 239   }
 240
 241   simcall_comm_test__set__result(simcall, res);
 242   if (res) {
 243     comm->simcalls_.push_back(simcall);
 244     comm->finish();
 245   } else {
 246     simcall->issuer_->simcall_answer();
 247   }
 248 }
 249
 250 void simcall_HANDLER_comm_testany(smx_simcall_t simcall, simgrid::kernel::activity::CommImpl* comms[], size_t count)
 251 {
 252   // The default result is -1 -- this means, "nothing is ready".
 253   // It can be changed below, but only if something matches.
 254   simcall_comm_testany__set__result(simcall, -1);
 255
 256   if (MC_is_active() || MC_record_replay_is_active()) {
 257     int idx = SIMCALL_GET_MC_VALUE(*simcall);
 258     if (idx == -1) {
 259       simcall->issuer_->simcall_answer();
 260     } else {
 261       simgrid::kernel::activity::CommImpl* comm = comms[idx];
 262       simcall_comm_testany__set__result(simcall, idx);
 263       comm->simcalls_.push_back(simcall);
 264       comm->state_ = simgrid::kernel::activity::State::DONE;
 265       comm->finish();
 266     }
 267     return;
 268   }
 269
 270   for (std::size_t i = 0; i != count; ++i) {
 271     simgrid::kernel::activity::CommImpl* comm = comms[i];
 272     if (comm->state_ != simgrid::kernel::activity::State::WAITING &&
 273         comm->state_ != simgrid::kernel::activity::State::RUNNING) {
 274       simcall_comm_testany__set__result(simcall, i);
 275       comm->simcalls_.push_back(simcall);
 276       comm->finish();
 277       return;
 278     }
 279   }
 280   simcall->issuer_->simcall_answer();
 281 }
 282
 283 static void SIMIX_waitany_remove_simcall_from_actions(smx_simcall_t simcall)
 284 {
 285   simgrid::kernel::activity::CommImpl** comms = simcall_comm_waitany__get__comms(simcall);
 286   size_t count                                = simcall_comm_waitany__get__count(simcall);
 287
 288   for (size_t i = 0; i < count; i++) {
 289     // Remove the first occurrence of simcall:
 290     auto* comm = comms[i];
 291     auto j     = boost::range::find(comm->simcalls_, simcall);
 292     if (j != comm->simcalls_.end())
 293       comm->simcalls_.erase(j);
 294   }
 295 }
 296 void simcall_HANDLER_comm_waitany(smx_simcall_t simcall, simgrid::kernel::activity::CommImpl* comms[], size_t count,
 297                                   double timeout)
 298 {
 299   if (MC_is_active() || MC_record_replay_is_active()) {
 300     if (timeout > 0.0)
 301       xbt_die("Timeout not implemented for waitany in the model-checker");
 302     int idx                 = SIMCALL_GET_MC_VALUE(*simcall);
 303     auto* comm              = comms[idx];
 304     comm->simcalls_.push_back(simcall);
 305     simcall_comm_waitany__set__result(simcall, idx);
 306     comm->state_ = simgrid::kernel::activity::State::DONE;
 307     comm->finish();
 308     return;
 309   }
 310
 311   if (timeout < 0.0) {
 312     simcall->timeout_cb_ = nullptr;
 313   } else {
 314     simcall->timeout_cb_ = simgrid::simix::Timer::set(SIMIX_get_clock() + timeout, [simcall]() {
 315       simcall->timeout_cb_ = nullptr;
 316       SIMIX_waitany_remove_simcall_from_actions(simcall);
 317       simcall_comm_waitany__set__result(simcall, -1);
 318       simcall->issuer_->simcall_answer();
 319     });
 320   }
 321
 322   for (size_t i = 0; i < count; i++) {
 323     /* associate this simcall to the the synchro */
 324     auto* comm = comms[i];
 325     comm->simcalls_.push_back(simcall);
 326
 327     /* see if the synchro is already finished */
 328     if (comm->state_ != simgrid::kernel::activity::State::WAITING &&
 329         comm->state_ != simgrid::kernel::activity::State::RUNNING) {
 330       comm->finish();
 331       break;
 332     }
 333   }
 334 }
 335
 336 /******************************************************************************/
 337 /*                    SIMIX_comm_copy_data callbacks                       */
 338 /******************************************************************************/
 339 static void (*SIMIX_comm_copy_data_callback)(simgrid::kernel::activity::CommImpl*, void*,
 340                                              size_t) = &SIMIX_comm_copy_pointer_callback;
 341
 342 void SIMIX_comm_copy_buffer_callback(simgrid::kernel::activity::CommImpl* comm, void* buff, size_t buff_size)
 343 {
 344   XBT_DEBUG("Copy the data over");
 345   memcpy(comm->dst_buff_, buff, buff_size);
 346   if (comm->detached()) { // if this is a detached send, the source buffer was duplicated by SMPI sender to make the
 347                           // original buffer available to the application ASAP
 348     xbt_free(buff);
 349     comm->src_buff_ = nullptr;
 350   }
 351 }
 352
 353 void SIMIX_comm_set_copy_data_callback(void (*callback)(simgrid::kernel::activity::CommImpl*, void*, size_t))
 354 {
 355   SIMIX_comm_copy_data_callback = callback;
 356 }
 357
 358 void SIMIX_comm_copy_pointer_callback(simgrid::kernel::activity::CommImpl* comm, void* buff, size_t buff_size)
 359 {
 360   xbt_assert((buff_size == sizeof(void*)), "Cannot copy %zu bytes: must be sizeof(void*)", buff_size);
 361   *(void**)(comm->dst_buff_) = buff;
 362 }
 363
 364 namespace simgrid {
 365 namespace kernel {
 366 namespace activity {
 367
 368 CommImpl& CommImpl::set_type(CommImpl::Type type)
 369 {
 370   type_ = type;
 371   return *this;
 372 }
 373
 374 CommImpl& CommImpl::set_size(double size)
 375 {
 376   size_ = size;
 377   return *this;
 378 }
 379
 380 CommImpl& CommImpl::set_rate(double rate)
 381 {
 382   rate_ = rate;
 383   return *this;
 384 }
 385 CommImpl& CommImpl::set_mailbox(MailboxImpl* mbox)
 386 {
 387   mbox_ = mbox;
 388   return *this;
 389 }
 390
 391 CommImpl& CommImpl::set_src_buff(unsigned char* buff, size_t size)
 392 {
 393   src_buff_      = buff;
 394   src_buff_size_ = size;
 395   return *this;
 396 }
 397
 398 CommImpl& CommImpl::set_dst_buff(unsigned char* buff, size_t* size)
 399 {
 400   dst_buff_      = buff;
 401   dst_buff_size_ = size;
 402   return *this;
 403 }
 404
 405 CommImpl& CommImpl::detach()
 406 {
 407   detached_ = true;
 408   return *this;
 409 }
 410
 411 CommImpl::CommImpl(s4u::Host* from, s4u::Host* to, double bytes) : size_(bytes), detached_(true), from_(from), to_(to)
 412 {
 413   state_ = State::READY;
 414 }
 415
 416 CommImpl::~CommImpl()
 417 {
 418   XBT_DEBUG("Really free communication %p in state %d (detached = %d)", this, static_cast<int>(state_), detached_);
 419
 420   cleanup_surf();
 421
 422   if (detached_ && state_ != State::DONE) {
 423     /* the communication has failed and was detached:
 424      * we have to free the buffer */
 425     if (clean_fun)
 426       clean_fun(src_buff_);
 427     src_buff_ = nullptr;
 428   } else if (mbox_) {
 429     mbox_->remove(this);
 430   }
 431 }
 432
 433 /**  @brief Starts the simulation of a communication synchro. */
 434 CommImpl* CommImpl::start()
 435 {
 436   /* If both the sender and the receiver are already there, start the communication */
 437   if (state_ == State::READY) {
 438     from_ = from_ != nullptr ? from_ : src_actor_->get_host();
 439     to_   = to_ != nullptr ? to_ : dst_actor_->get_host();
 440
 441     surf_action_ = surf_network_model->communicate(from_, to_, size_, rate_);
 442     surf_action_->set_activity(this);
 443     surf_action_->set_category(get_tracing_category());
 444     state_ = State::RUNNING;
 445
 446     XBT_DEBUG("Starting communication %p from '%s' to '%s' (surf_action: %p; state: %s)", this, from_->get_cname(),
 447               to_->get_cname(), surf_action_, get_state_str());
 448
 449     /* If a link is failed, detect it immediately */
 450     if (surf_action_->get_state() == resource::Action::State::FAILED) {
 451       XBT_DEBUG("Communication from '%s' to '%s' failed to start because of a link failure", from_->get_cname(),
 452                 to_->get_cname());
 453       state_ = State::LINK_FAILURE;
 454       post();
 455
 456     } else if ((src_actor_ != nullptr && src_actor_->is_suspended()) ||
 457                (dst_actor_ != nullptr && dst_actor_->is_suspended())) {
 458       /* If any of the process is suspended, create the synchro but stop its execution,
 459          it will be restarted when the sender process resume */
 460       if (src_actor_->is_suspended())
 461         XBT_DEBUG("The communication is suspended on startup because src (%s@%s) was suspended since it initiated the "
 462                   "communication",
 463                   src_actor_->get_cname(), src_actor_->get_host()->get_cname());
 464       else
 465         XBT_DEBUG("The communication is suspended on startup because dst (%s@%s) was suspended since it initiated the "
 466                   "communication",
 467                   dst_actor_->get_cname(), dst_actor_->get_host()->get_cname());
 468
 469       surf_action_->suspend();
 470     }
 471   }
 472
 473   return this;
 474 }
 475
 476 /** @brief Copy the communication data from the sender's buffer to the receiver's one  */
 477 void CommImpl::copy_data()
 478 {
 479   size_t buff_size = src_buff_size_;
 480   /* If there is no data to copy then return */
 481   if (not src_buff_ || not dst_buff_ || copied_)
 482     return;
 483
 484   XBT_DEBUG("Copying comm %p data from %s (%p) -> %s (%p) (%zu bytes)", this,
 485             src_actor_ ? src_actor_->get_host()->get_cname() : "a finished process", src_buff_,
 486             dst_actor_ ? dst_actor_->get_host()->get_cname() : "a finished process", dst_buff_, buff_size);
 487
 488   /* Copy at most dst_buff_size bytes of the message to receiver's buffer */
 489   if (dst_buff_size_) {
 490     buff_size = std::min(buff_size, *(dst_buff_size_));
 491
 492     /* Update the receiver's buffer size to the copied amount */
 493     *dst_buff_size_ = buff_size;
 494   }
 495
 496   if (buff_size > 0) {
 497     if (copy_data_fun)
 498       copy_data_fun(this, src_buff_, buff_size);
 499     else
 500       SIMIX_comm_copy_data_callback(this, src_buff_, buff_size);
 501   }
 502
 503   /* Set the copied flag so we copy data only once */
 504   /* (this function might be called from both communication ends) */
 505   copied_ = true;
 506 }
 507
 508 void CommImpl::suspend()
 509 {
 510   /* FIXME: shall we suspend also the timeout synchro? */
 511   if (surf_action_)
 512     surf_action_->suspend();
 513   /* if not created yet, the action will be suspended on creation, in CommImpl::start() */
 514 }
 515
 516 void CommImpl::resume()
 517 {
 518   /*FIXME: check what happen with the timeouts */
 519   if (surf_action_)
 520     surf_action_->resume();
 521   /* in the other case, the synchro were not really suspended yet, see CommImpl::suspend() and CommImpl::start() */
 522 }
 523
 524 void CommImpl::cancel()
 525 {
 526   /* if the synchro is a waiting state means that it is still in a mbox so remove from it and delete it */
 527   if (state_ == State::WAITING) {
 528     if (not detached_) {
 529       mbox_->remove(this);
 530       state_ = State::CANCELED;
 531     }
 532   } else if (not MC_is_active() /* when running the MC there are no surf actions */
 533              && not MC_record_replay_is_active() && (state_ == State::READY || state_ == State::RUNNING)) {
 534     surf_action_->cancel();
 535   }
 536 }
 537
 538 /** @brief This is part of the cleanup process, probably an internal command */
 539 void CommImpl::cleanup_surf()
 540 {
 541   clean_action();
 542
 543   if (src_timeout_) {
 544     src_timeout_->unref();
 545     src_timeout_ = nullptr;
 546   }
 547
 548   if (dst_timeout_) {
 549     dst_timeout_->unref();
 550     dst_timeout_ = nullptr;
 551   }
 552 }
 553
 554 void CommImpl::post()
 555 {
 556   /* Update synchro state */
 557   if (src_timeout_ && src_timeout_->get_state() == resource::Action::State::FINISHED)
 558     state_ = State::SRC_TIMEOUT;
 559   else if (dst_timeout_ && dst_timeout_->get_state() == resource::Action::State::FINISHED)
 560     state_ = State::DST_TIMEOUT;
 561   else if (src_timeout_ && src_timeout_->get_state() == resource::Action::State::FAILED)
 562     state_ = State::SRC_HOST_FAILURE;
 563   else if (dst_timeout_ && dst_timeout_->get_state() == resource::Action::State::FAILED)
 564     state_ = State::DST_HOST_FAILURE;
 565   else if (surf_action_ && surf_action_->get_state() == resource::Action::State::FAILED) {
 566     state_ = State::LINK_FAILURE;
 567   } else
 568     state_ = State::DONE;
 569
 570   XBT_DEBUG("CommImpl::post(): comm %p, state %s, src_proc %p, dst_proc %p, detached: %d", this, get_state_str(),
 571             src_actor_.get(), dst_actor_.get(), detached_);
 572
 573   /* destroy the surf actions associated with the Simix communication */
 574   cleanup_surf();
 575
 576   /* Answer all simcalls associated with the synchro */
 577   finish();
 578 }
 579
 580 void CommImpl::finish()
 581 {
 582   while (not simcalls_.empty()) {
 583     smx_simcall_t simcall = simcalls_.front();
 584     simcalls_.pop_front();
 585
 586     /* If a waitany simcall is waiting for this synchro to finish, then remove it from the other synchros in the waitany
 587      * list. Afterwards, get the position of the actual synchro in the waitany list and return it as the result of the
 588      * simcall */
 589
 590     if (simcall->call_ == simix::Simcall::NONE) // FIXME: maybe a better way to handle this case
 591       continue;                                 // if actor handling comm is killed
 592     if (simcall->call_ == simix::Simcall::COMM_WAITANY) {
 593       SIMIX_waitany_remove_simcall_from_actions(simcall);
 594       if (simcall->timeout_cb_) {
 595         simcall->timeout_cb_->remove();
 596         simcall->timeout_cb_ = nullptr;
 597       }
 598       if (not MC_is_active() && not MC_record_replay_is_active()) {
 599         CommImpl** comms   = simcall_comm_waitany__get__comms(simcall);
 600         size_t count       = simcall_comm_waitany__get__count(simcall);
 601         CommImpl** element = std::find(comms, comms + count, this);
 602         int rank           = (element != comms + count) ? element - comms : -1;
 603         simcall_comm_waitany__set__result(simcall, rank);
 604       }
 605     }
 606
 607     /* If the synchro is still in a rendez-vous point then remove from it */
 608     if (mbox_)
 609       mbox_->remove(this);
 610
 611     XBT_DEBUG("CommImpl::finish(): synchro state = %d", static_cast<int>(state_));
 612
 613     /* Check out for errors */
 614
 615     if (not simcall->issuer_->get_host()->is_on()) {
 616       simcall->issuer_->context_->set_wannadie();
 617     } else {
 618       switch (state_) {
 619         case State::DONE:
 620           XBT_DEBUG("Communication %p complete!", this);
 621           copy_data();
 622           break;
 623
 624         case State::SRC_TIMEOUT:
 625           simcall->issuer_->exception_ = std::make_exception_ptr(
 626               simgrid::TimeoutException(XBT_THROW_POINT, "Communication timeouted because of the sender"));
 627           break;
 628
 629         case State::DST_TIMEOUT:
 630           simcall->issuer_->exception_ = std::make_exception_ptr(
 631               simgrid::TimeoutException(XBT_THROW_POINT, "Communication timeouted because of the receiver"));
 632           break;
 633
 634         case State::SRC_HOST_FAILURE:
 635           if (simcall->issuer_ == src_actor_)
 636             simcall->issuer_->context_->set_wannadie();
 637           else
 638             simcall->issuer_->exception_ =
 639                 std::make_exception_ptr(simgrid::NetworkFailureException(XBT_THROW_POINT, "Remote peer failed"));
 640           break;
 641
 642         case State::DST_HOST_FAILURE:
 643           if (simcall->issuer_ == dst_actor_)
 644             simcall->issuer_->context_->set_wannadie();
 645           else
 646             simcall->issuer_->exception_ =
 647                 std::make_exception_ptr(simgrid::NetworkFailureException(XBT_THROW_POINT, "Remote peer failed"));
 648           break;
 649
 650         case State::LINK_FAILURE:
 651           XBT_DEBUG("Link failure in synchro %p between '%s' and '%s': posting an exception to the issuer: %s (%p) "
 652                     "detached:%d",
 653                     this, src_actor_ ? src_actor_->get_host()->get_cname() : nullptr,
 654                     dst_actor_ ? dst_actor_->get_host()->get_cname() : nullptr, simcall->issuer_->get_cname(),
 655                     simcall->issuer_, detached_);
 656           if (src_actor_ == simcall->issuer_) {
 657             XBT_DEBUG("I'm source");
 658           } else if (dst_actor_ == simcall->issuer_) {
 659             XBT_DEBUG("I'm dest");
 660           } else {
 661             XBT_DEBUG("I'm neither source nor dest");
 662           }
 663           simcall->issuer_->throw_exception(
 664               std::make_exception_ptr(simgrid::NetworkFailureException(XBT_THROW_POINT, "Link failure")));
 665           break;
 666
 667         case State::CANCELED:
 668           if (simcall->issuer_ == dst_actor_)
 669             simcall->issuer_->exception_ = std::make_exception_ptr(
 670                 simgrid::CancelException(XBT_THROW_POINT, "Communication canceled by the sender"));
 671           else
 672             simcall->issuer_->exception_ = std::make_exception_ptr(
 673                 simgrid::CancelException(XBT_THROW_POINT, "Communication canceled by the receiver"));
 674           break;
 675
 676         default:
 677           xbt_die("Unexpected synchro state in CommImpl::finish: %d", static_cast<int>(state_));
 678       }
 679       simcall->issuer_->simcall_answer();
 680     }
 681     /* if there is an exception during a waitany or a testany, indicate the position of the failed communication */
 682     if (simcall->issuer_->exception_ &&
 683         (simcall->call_ == simix::Simcall::COMM_WAITANY || simcall->call_ == simix::Simcall::COMM_TESTANY)) {
 684       // First retrieve the rank of our failing synchro
 685       CommImpl** comms;
 686       size_t count;
 687       if (simcall->call_ == simix::Simcall::COMM_WAITANY) {
 688         comms = simcall_comm_waitany__get__comms(simcall);
 689         count = simcall_comm_waitany__get__count(simcall);
 690       } else {
 691         /* simcall->call_ == simix::Simcall::COMM_TESTANY */
 692         comms = simcall_comm_testany__get__comms(simcall);
 693         count = simcall_comm_testany__get__count(simcall);
 694       }
 695       CommImpl** element = std::find(comms, comms + count, this);
 696       int rank           = (element != comms + count) ? element - comms : -1;
 697
 698       // In order to modify the exception we have to rethrow it:
 699       try {
 700         std::rethrow_exception(simcall->issuer_->exception_);
 701       } catch (simgrid::Exception& e) {
 702         e.value = rank;
 703       }
 704     }
 705
 706     simcall->issuer_->waiting_synchro_ = nullptr;
 707     simcall->issuer_->activities_.remove(this);
 708     if (detached_) {
 709       if (simcall->issuer_ == src_actor_) {
 710         if (dst_actor_)
 711           dst_actor_->activities_.remove(this);
 712       } else if (simcall->issuer_ == dst_actor_) {
 713         if (src_actor_)
 714           src_actor_->activities_.remove(this);
 715       } else {
 716         if (dst_actor_ != nullptr)
 717           dst_actor_->activities_.remove(this);
 718         if (src_actor_ != nullptr)
 719           src_actor_->activities_.remove(this);
 720       }
 721     }
 722   }
 723 }
 724
 725 } // namespace activity
 726 } // namespace kernel
 727 } // namespace simgrid