1 /* Copyright (c) 2007-2022. The SimGrid Team. All rights reserved. */
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. */
6 #include <simgrid/Exception.hpp>
7 #include <simgrid/kernel/routing/NetPoint.hpp>
8 #include <simgrid/modelchecker.h>
9 #include <simgrid/s4u/Host.hpp>
11 #include "src/kernel/activity/CommImpl.hpp"
12 #include "src/kernel/activity/MailboxImpl.hpp"
13 #include "src/kernel/context/Context.hpp"
14 #include "src/kernel/resource/CpuImpl.hpp"
15 #include "src/kernel/resource/LinkImpl.hpp"
16 #include "src/kernel/resource/StandardLinkImpl.hpp"
17 #include "src/mc/mc_replay.hpp"
19 XBT_LOG_NEW_DEFAULT_SUBCATEGORY(ker_network, kernel, "Kernel network-related synchronization");
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)
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->mc_value_ = 0;
30 comm->wait_for(simcall->issuer_, timeout);
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)
41 return simgrid::kernel::activity::CommImpl::isend(src_proc, mbox, task_size, rate, src_buff, src_buff_size, match_fun,
42 clean_fun, copy_data_fun, data, detached);
45 XBT_PRIVATE void simcall_HANDLER_comm_recv(smx_simcall_t simcall, smx_actor_t receiver, smx_mailbox_t mbox,
46 unsigned char* dst_buff, size_t* dst_buff_size,
47 bool (*match_fun)(void*, void*, simgrid::kernel::activity::CommImpl*),
48 void (*copy_data_fun)(simgrid::kernel::activity::CommImpl*, void*, size_t),
49 void* data, double timeout, double rate)
51 simgrid::kernel::activity::ActivityImplPtr comm = simcall_HANDLER_comm_irecv(
52 simcall, receiver, mbox, dst_buff, dst_buff_size, match_fun, copy_data_fun, data, rate);
53 simcall->mc_value_ = 0;
54 comm->wait_for(simcall->issuer_, timeout);
57 XBT_PRIVATE simgrid::kernel::activity::ActivityImplPtr
58 simcall_HANDLER_comm_irecv(smx_simcall_t /*simcall*/, smx_actor_t receiver, smx_mailbox_t mbox, unsigned char* dst_buff,
59 size_t* dst_buff_size, bool (*match_fun)(void*, void*, simgrid::kernel::activity::CommImpl*),
60 void (*copy_data_fun)(simgrid::kernel::activity::CommImpl*, void*, size_t), void* data,
63 return simgrid::kernel::activity::CommImpl::irecv(receiver, mbox, dst_buff, dst_buff_size, match_fun, copy_data_fun,
67 void simcall_HANDLER_comm_wait(smx_simcall_t simcall, simgrid::kernel::activity::CommImpl* comm, double timeout)
69 comm->wait_for(simcall->issuer_, timeout);
72 bool simcall_HANDLER_comm_test(smx_simcall_t simcall, simgrid::kernel::activity::CommImpl* comm)
74 return comm->test(simcall->issuer_);
77 ssize_t simcall_HANDLER_comm_testany(smx_simcall_t simcall, simgrid::kernel::activity::CommImpl* comms[], size_t count)
79 std::vector<simgrid::kernel::activity::ActivityImpl*> comms_vec(comms, comms + count);
80 return simgrid::kernel::activity::ActivityImpl::test_any(simcall->issuer_, comms_vec);
83 void simcall_HANDLER_comm_waitany(smx_simcall_t simcall, simgrid::kernel::activity::CommImpl* comms[], size_t count,
86 std::vector<simgrid::kernel::activity::CommImpl*> comms_vec(comms, comms + count);
87 simgrid::kernel::activity::CommImpl::wait_any_for(simcall->issuer_, comms_vec, timeout);
90 /******************************************************************************/
91 /* SIMIX_comm_copy_data callbacks */
92 /******************************************************************************/
93 // XBT_ATTRIB_DEPRECATED_v333
94 void SIMIX_comm_set_copy_data_callback(void (*callback)(simgrid::kernel::activity::CommImpl*, void*, size_t))
96 simgrid::kernel::activity::CommImpl::set_copy_data_callback(callback);
99 // XBT_ATTRIB_DEPRECATED_v333
100 void SIMIX_comm_copy_buffer_callback(simgrid::kernel::activity::CommImpl* comm, void* buff, size_t buff_size)
102 simgrid::s4u::Comm::copy_buffer_callback(comm, buff, buff_size);
105 // XBT_ATTRIB_DEPRECATED_v333
106 void SIMIX_comm_copy_pointer_callback(simgrid::kernel::activity::CommImpl* comm, void* buff, size_t buff_size)
108 simgrid::s4u::Comm::copy_pointer_callback(comm, buff, buff_size);
114 xbt::signal<void(CommImpl const&)> CommImpl::on_start;
115 xbt::signal<void(CommImpl const&)> CommImpl::on_completion;
117 void (*CommImpl::copy_data_callback_)(CommImpl*, void*, size_t) = &s4u::Comm::copy_pointer_callback;
119 void CommImpl::set_copy_data_callback(void (*callback)(CommImpl*, void*, size_t))
121 copy_data_callback_ = callback;
124 CommImpl& CommImpl::set_size(double size)
130 CommImpl& CommImpl::set_rate(double rate)
135 CommImpl& CommImpl::set_mailbox(MailboxImpl* mbox)
141 CommImpl& CommImpl::set_src_buff(unsigned char* buff, size_t size)
144 src_buff_size_ = size;
148 CommImpl& CommImpl::set_dst_buff(unsigned char* buff, size_t* size)
151 dst_buff_size_ = size;
155 CommImpl& CommImpl::detach()
161 CommImpl::CommImpl(s4u::Host* from, s4u::Host* to, double bytes) : size_(bytes), detached_(true), from_(from), to_(to)
163 set_state(State::READY);
166 CommImpl::~CommImpl()
168 XBT_DEBUG("Really free communication %p in state %s (detached = %d)", this, get_state_str(), detached_);
172 if (detached_ && get_state() != State::DONE) {
173 /* the communication has failed and was detached:
174 * we have to free the buffer */
176 clean_fun(src_buff_);
183 /** @brief Starts the simulation of a communication synchro. */
184 CommImpl* CommImpl::start()
186 /* If both the sender and the receiver are already there, start the communication */
187 if (get_state() == State::READY) {
188 from_ = from_ != nullptr ? from_ : src_actor_->get_host();
189 to_ = to_ != nullptr ? to_ : dst_actor_->get_host();
190 /* Getting the network_model from the origin host
191 * Valid while we have a single network model, otherwise we would need to change this function to first get the
192 * routes and later create the respective surf actions */
193 auto net_model = from_->get_netpoint()->get_englobing_zone()->get_network_model();
195 surf_action_ = net_model->communicate(from_, to_, size_, rate_);
196 surf_action_->set_activity(this);
197 surf_action_->set_category(get_tracing_category());
198 set_start_time(surf_action_->get_start_time());
199 set_state(State::RUNNING);
202 XBT_DEBUG("Starting communication %p from '%s' to '%s' (surf_action: %p; state: %s)", this, from_->get_cname(),
203 to_->get_cname(), surf_action_, get_state_str());
205 /* If a link is failed, detect it immediately */
206 if (surf_action_->get_state() == resource::Action::State::FAILED) {
207 XBT_DEBUG("Communication from '%s' to '%s' failed to start because of a link failure", from_->get_cname(),
209 set_state(State::LINK_FAILURE);
212 } else if ((src_actor_ != nullptr && src_actor_->is_suspended()) ||
213 (dst_actor_ != nullptr && dst_actor_->is_suspended())) {
214 /* If any of the actor is suspended, create the synchro but stop its execution,
215 it will be restarted when the sender actor resume */
216 if (src_actor_->is_suspended())
217 XBT_DEBUG("The communication is suspended on startup because src (%s@%s) was suspended since it initiated the "
219 src_actor_->get_cname(), src_actor_->get_host()->get_cname());
221 XBT_DEBUG("The communication is suspended on startup because dst (%s@%s) was suspended since it initiated the "
223 dst_actor_->get_cname(), dst_actor_->get_host()->get_cname());
225 surf_action_->suspend();
232 std::vector<s4u::Link*> CommImpl::get_traversed_links() const
234 xbt_assert(get_state() != State::WAITING, "You cannot use %s() if your communication is not ready (%s)", __FUNCTION__,
236 std::vector<s4u::Link*> vlinks;
237 XBT_ATTRIB_UNUSED double res = 0;
238 from_->route_to(to_, vlinks, &res);
242 /** @brief Copy the communication data from the sender's buffer to the receiver's one */
243 void CommImpl::copy_data()
245 size_t buff_size = src_buff_size_;
246 /* If there is no data to copy then return */
247 if (not src_buff_ || not dst_buff_ || copied_)
250 XBT_DEBUG("Copying comm %p data from %s (%p) -> %s (%p) (%zu bytes)", this,
251 src_actor_ ? src_actor_->get_host()->get_cname() : "a finished actor", src_buff_,
252 dst_actor_ ? dst_actor_->get_host()->get_cname() : "a finished actor", dst_buff_, buff_size);
254 /* Copy at most dst_buff_size bytes of the message to receiver's buffer */
255 if (dst_buff_size_) {
256 buff_size = std::min(buff_size, *dst_buff_size_);
258 /* Update the receiver's buffer size to the copied amount */
259 *dst_buff_size_ = buff_size;
264 copy_data_fun(this, src_buff_, buff_size);
266 copy_data_callback_(this, src_buff_, buff_size);
269 /* Set the copied flag so we copy data only once */
270 /* (this function might be called from both communication ends) */
275 CommImpl::isend(actor::ActorImpl* src_proc, MailboxImpl* mbox, double task_size, double rate, unsigned char* src_buff,
276 size_t src_buff_size, bool (*match_fun)(void*, void*, CommImpl*),
277 void (*clean_fun)(void*), // used to free the synchro in case of problem after a detached send
278 void (*copy_data_fun)(CommImpl*, void*, size_t), // used to copy data if not default one
279 void* data, bool detached)
281 XBT_DEBUG("send from mailbox %p", mbox);
283 /* Prepare a synchro describing us, so that it gets passed to the user-provided filter of other side */
284 CommImplPtr this_comm(new CommImpl(CommImpl::Type::SEND));
286 /* Look for communication synchro matching our needs. We also provide a description of
287 * ourself so that the other side also gets a chance of choosing if it wants to match with us.
289 * If it is not found then push our communication into the rendez-vous point */
290 CommImplPtr other_comm = mbox->find_matching_comm(CommImpl::Type::RECEIVE, match_fun, data, this_comm,
291 /*done*/ false, /*remove_matching*/ true);
293 if (not other_comm) {
294 other_comm = std::move(this_comm);
296 if (mbox->is_permanent()) {
297 // this mailbox is for small messages, which have to be sent right now
298 other_comm->set_state(State::READY);
299 other_comm->dst_actor_ = mbox->get_permanent_receiver().get();
300 mbox->push_done(other_comm);
301 XBT_DEBUG("pushing a message into the permanent receive list %p, comm %p", mbox, other_comm.get());
304 mbox->push(other_comm);
307 XBT_DEBUG("Receive already pushed");
309 other_comm->set_state(State::READY);
313 other_comm->detach();
314 other_comm->clean_fun = clean_fun;
316 other_comm->clean_fun = nullptr;
317 src_proc->activities_.emplace_back(other_comm);
320 /* Setup the communication synchro */
321 other_comm->src_actor_ = src_proc;
322 other_comm->src_data_ = data;
323 (*other_comm).set_src_buff(src_buff, src_buff_size).set_size(task_size).set_rate(rate);
325 other_comm->match_fun = match_fun;
326 other_comm->copy_data_fun = copy_data_fun;
328 if (MC_is_active() || MC_record_replay_is_active())
329 other_comm->set_state(simgrid::kernel::activity::State::RUNNING);
333 return (detached ? nullptr : other_comm);
336 ActivityImplPtr CommImpl::irecv(actor::ActorImpl* receiver, MailboxImpl* mbox, unsigned char* dst_buff,
337 size_t* dst_buff_size, bool (*match_fun)(void*, void*, CommImpl*),
338 void (*copy_data_fun)(CommImpl*, void*, size_t), void* data, double rate)
340 CommImplPtr this_synchro(new CommImpl(CommImpl::Type::RECEIVE));
341 XBT_DEBUG("recv from mbox %p. this_synchro=%p", mbox, this_synchro.get());
343 CommImplPtr other_comm;
344 // communication already done, get it inside the list of completed comms
345 if (mbox->is_permanent() && mbox->has_some_done_comm()) {
346 XBT_DEBUG("We have a comm that has probably already been received, trying to match it, to skip the communication");
347 // find a match in the list of already received comms
348 other_comm = mbox->find_matching_comm(CommImpl::Type::SEND, match_fun, data, this_synchro, /*done*/ true,
349 /*remove_matching*/ true);
350 // if not found, assume the receiver came first, register it to the mailbox in the classical way
351 if (not other_comm) {
352 XBT_DEBUG("We have messages in the permanent receive list, but not the one we are looking for, pushing request "
354 other_comm = std::move(this_synchro);
355 mbox->push(other_comm);
357 if (other_comm->surf_action_ && other_comm->get_remaining() < 1e-12) {
358 XBT_DEBUG("comm %p has been already sent, and is finished, destroy it", other_comm.get());
359 other_comm->set_state(State::DONE);
360 other_comm->set_mailbox(nullptr);
364 /* Prepare a comm describing us, so that it gets passed to the user-provided filter of other side */
366 /* Look for communication activity matching our needs. We also provide a description of
367 * ourself so that the other side also gets a chance of choosing if it wants to match with us.
369 * If it is not found then push our communication into the rendez-vous point */
370 other_comm = mbox->find_matching_comm(CommImpl::Type::SEND, match_fun, data, this_synchro, /*done*/ false,
371 /*remove_matching*/ true);
373 if (other_comm == nullptr) {
374 XBT_DEBUG("Receive pushed first (%zu comm enqueued so far)", mbox->size());
375 other_comm = std::move(this_synchro);
376 mbox->push(other_comm);
378 XBT_DEBUG("Match my %p with the existing %p", this_synchro.get(), other_comm.get());
380 other_comm->set_state(simgrid::kernel::activity::State::READY);
382 receiver->activities_.emplace_back(other_comm);
385 /* Setup communication synchro */
386 other_comm->dst_actor_ = receiver;
387 other_comm->dst_data_ = data;
388 other_comm->set_dst_buff(dst_buff, dst_buff_size);
390 if (rate > -1.0 && (other_comm->get_rate() < 0.0 || rate < other_comm->get_rate()))
391 other_comm->set_rate(rate);
393 other_comm->match_fun = match_fun;
394 other_comm->copy_data_fun = copy_data_fun;
396 if (MC_is_active() || MC_record_replay_is_active()) {
397 other_comm->set_state(State::RUNNING);
403 bool CommImpl::test(actor::ActorImpl* issuer)
405 if ((MC_is_active() || MC_record_replay_is_active()) && src_actor_ && dst_actor_)
406 set_state(State::DONE);
407 return ActivityImpl::test(issuer);
410 void CommImpl::wait_for(actor::ActorImpl* issuer, double timeout)
412 XBT_DEBUG("CommImpl::wait_for(%g), %p, state %s", timeout, this, get_state_str());
414 /* Associate this simcall to the wait synchro */
415 register_simcall(&issuer->simcall_);
416 if (MC_is_active() || MC_record_replay_is_active()) {
417 int idx = issuer->simcall_.mc_value_;
419 set_state(State::DONE);
421 /* If we reached this point, the wait simcall must have a timeout */
422 /* Otherwise it shouldn't be enabled and executed by the MC */
425 set_state(issuer == src_actor_ ? State::SRC_TIMEOUT : State::DST_TIMEOUT);
430 ActivityImpl::wait_for(issuer, timeout);
433 void CommImpl::wait_any_for(actor::ActorImpl* issuer, const std::vector<CommImpl*>& comms, double timeout)
435 std::vector<ActivityImpl*> activities(comms.begin(), comms.end());
436 ActivityImpl::wait_any_for(issuer, activities, timeout);
439 void CommImpl::suspend()
441 /* FIXME: shall we suspend also the timeout synchro? */
443 surf_action_->suspend();
444 /* if not created yet, the action will be suspended on creation, in CommImpl::start() */
447 void CommImpl::resume()
449 /*FIXME: check what happen with the timeouts */
451 surf_action_->resume();
452 /* in the other case, the synchro were not really suspended yet, see CommImpl::suspend() and CommImpl::start() */
455 void CommImpl::cancel()
457 /* if the synchro is a waiting state means that it is still in a mbox so remove from it and delete it */
458 if (get_state() == State::WAITING) {
461 set_state(State::CANCELED);
463 } else if (not MC_is_active() /* when running the MC there are no surf actions */
464 && not MC_record_replay_is_active() && (get_state() == State::READY || get_state() == State::RUNNING)) {
465 surf_action_->cancel();
469 /** @brief This is part of the cleanup process, probably an internal command */
470 void CommImpl::cleanup_surf()
475 src_timeout_->unref();
476 src_timeout_ = nullptr;
480 dst_timeout_->unref();
481 dst_timeout_ = nullptr;
485 void CommImpl::post()
487 on_completion(*this);
489 /* Update synchro state */
490 if (src_timeout_ && src_timeout_->get_state() == resource::Action::State::FINISHED)
491 set_state(State::SRC_TIMEOUT);
492 else if (dst_timeout_ && dst_timeout_->get_state() == resource::Action::State::FINISHED)
493 set_state(State::DST_TIMEOUT);
494 else if (src_timeout_ && src_timeout_->get_state() == resource::Action::State::FAILED)
495 set_state(State::SRC_HOST_FAILURE);
496 else if (dst_timeout_ && dst_timeout_->get_state() == resource::Action::State::FAILED)
497 set_state(State::DST_HOST_FAILURE);
498 else if (surf_action_ && surf_action_->get_state() == resource::Action::State::FAILED) {
499 set_state(State::LINK_FAILURE);
501 set_state(State::DONE);
503 XBT_DEBUG("CommImpl::post(): comm %p, state %s, src_proc %p, dst_proc %p, detached: %d", this, get_state_str(),
504 src_actor_.get(), dst_actor_.get(), detached_);
506 /* destroy the surf actions associated with the Simix communication */
509 /* Answer all simcalls associated with the synchro */
512 void CommImpl::set_exception(actor::ActorImpl* issuer)
514 switch (get_state()) {
516 issuer->exception_ = std::make_exception_ptr(NetworkFailureException(XBT_THROW_POINT, "Remote peer failed"));
518 case State::SRC_TIMEOUT:
520 std::make_exception_ptr(TimeoutException(XBT_THROW_POINT, "Communication timeouted because of the sender"));
523 case State::DST_TIMEOUT:
525 std::make_exception_ptr(TimeoutException(XBT_THROW_POINT, "Communication timeouted because of the receiver"));
528 case State::SRC_HOST_FAILURE:
529 if (issuer == src_actor_)
530 issuer->context_->set_wannadie();
532 set_state(State::FAILED);
533 issuer->exception_ = std::make_exception_ptr(NetworkFailureException(XBT_THROW_POINT, "Remote peer failed"));
537 case State::DST_HOST_FAILURE:
538 if (issuer == dst_actor_)
539 issuer->context_->set_wannadie();
541 set_state(State::FAILED);
542 issuer->exception_ = std::make_exception_ptr(NetworkFailureException(XBT_THROW_POINT, "Remote peer failed"));
546 case State::LINK_FAILURE:
547 XBT_DEBUG("Link failure in synchro %p between '%s' and '%s': posting an exception to the issuer: %s (%p) "
549 this, src_actor_ ? src_actor_->get_host()->get_cname() : nullptr,
550 dst_actor_ ? dst_actor_->get_host()->get_cname() : nullptr, issuer->get_cname(), issuer, detached_);
551 if (src_actor_ == issuer) {
552 XBT_DEBUG("I'm source");
553 } else if (dst_actor_ == issuer) {
554 XBT_DEBUG("I'm dest");
556 XBT_DEBUG("I'm neither source nor dest");
558 set_state(State::FAILED);
559 issuer->throw_exception(std::make_exception_ptr(NetworkFailureException(XBT_THROW_POINT, "Link failure")));
562 case State::CANCELED:
563 if (issuer == dst_actor_)
565 std::make_exception_ptr(CancelException(XBT_THROW_POINT, "Communication canceled by the sender"));
568 std::make_exception_ptr(CancelException(XBT_THROW_POINT, "Communication canceled by the receiver"));
572 xbt_assert(get_state() == State::DONE, "Internal error in CommImpl::finish(): unexpected synchro state %s",
577 void CommImpl::finish()
579 XBT_DEBUG("CommImpl::finish() in state %s", get_state_str());
580 /* If the synchro is still in a rendez-vous point then remove from it */
584 if (get_state() == State::DONE)
587 while (not simcalls_.empty()) {
588 smx_simcall_t simcall = simcalls_.front();
589 simcalls_.pop_front();
591 /* If a waitany simcall is waiting for this synchro to finish, then remove it from the other synchros in the waitany
592 * list. Afterwards, get the position of the actual synchro in the waitany list and return it as the result of the
595 if (simcall->call_ == simix::Simcall::NONE) // FIXME: maybe a better way to handle this case
596 continue; // if actor handling comm is killed
598 handle_activity_waitany(simcall);
600 /* Check out for errors */
602 if (not simcall->issuer_->get_host()->is_on()) {
603 simcall->issuer_->context_->set_wannadie();
605 // Do not answer to dying actors
606 if (not simcall->issuer_->context_->wannadie()) {
607 set_exception(simcall->issuer_);
608 simcall->issuer_->simcall_answer();
612 simcall->issuer_->waiting_synchro_ = nullptr;
613 simcall->issuer_->activities_.remove(this);
615 if (simcall->issuer_ != dst_actor_ && dst_actor_ != nullptr)
616 dst_actor_->activities_.remove(this);
617 if (simcall->issuer_ != src_actor_ && src_actor_ != nullptr)
618 src_actor_->activities_.remove(this);
623 } // namespace activity
624 } // namespace kernel
625 } // namespace simgrid