1 /* Copyright (c) 2007-2023. 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/EngineImpl.hpp"
12 #include "src/kernel/activity/CommImpl.hpp"
13 #include "src/kernel/activity/MailboxImpl.hpp"
14 #include "src/kernel/actor/SimcallObserver.hpp"
15 #include "src/kernel/resource/CpuImpl.hpp"
16 #include "src/kernel/resource/NetworkModel.hpp"
17 #include "src/kernel/resource/StandardLinkImpl.hpp"
18 #include "src/mc/mc_replay.hpp"
20 XBT_LOG_NEW_DEFAULT_SUBCATEGORY(ker_network, kernel, "Kernel network-related synchronization");
22 namespace simgrid::kernel::activity {
23 xbt::signal<void(CommImpl const&)> CommImpl::on_start;
24 xbt::signal<void(CommImpl const&)> CommImpl::on_completion;
26 std::function<void(CommImpl*, void*, size_t)> CommImpl::copy_data_callback_ = &s4u::Comm::copy_pointer_callback;
28 void CommImpl::set_copy_data_callback(const std::function<void(CommImpl*, void*, size_t)>& callback)
30 copy_data_callback_ = callback;
33 CommImpl& CommImpl::set_type(CommImplType type)
39 CommImpl& CommImpl::set_source(s4u::Host* from)
41 xbt_assert( from_ == nullptr );
47 CommImpl& CommImpl::set_destination(s4u::Host* to)
49 xbt_assert( to_ == nullptr );
55 CommImpl& CommImpl::set_size(double size)
61 CommImpl& CommImpl::set_rate(double rate)
66 CommImpl& CommImpl::set_mailbox(MailboxImpl* mbox)
69 mbox_id_ = mbox->get_id();
74 CommImpl& CommImpl::set_src_buff(unsigned char* buff, size_t size)
77 src_buff_size_ = size;
81 CommImpl& CommImpl::set_dst_buff(unsigned char* buff, size_t* size)
84 dst_buff_size_ = size;
88 CommImpl& CommImpl::detach()
91 EngineImpl::get_instance()->get_maestro()->activities_.insert(this);
97 XBT_DEBUG("Really free communication %p in state %s (detached = %d)", this, get_state_str(), detached_);
101 if (detached_ && get_state() != State::DONE) {
102 /* the communication has failed and was detached:
103 * we have to free the buffer */
105 clean_fun(src_buff_);
112 /** @brief Starts the simulation of a communication synchro. */
113 CommImpl* CommImpl::start()
115 /* If both the sender and the receiver are already there, start the communication */
116 if (get_state() == State::READY) {
117 from_ = from_ != nullptr ? from_ : src_actor_->get_host();
118 xbt_assert(from_->is_on());
119 to_ = to_ != nullptr ? to_ : dst_actor_->get_host();
120 xbt_assert(to_->is_on());
122 /* Getting the network_model from the origin host
123 * Valid while we have a single network model, otherwise we would need to change this function to first get the
124 * routes and later create the respective surf actions */
125 auto net_model = from_->get_netpoint()->get_englobing_zone()->get_network_model();
127 surf_action_ = net_model->communicate(from_, to_, size_, rate_, false);
128 surf_action_->set_activity(this);
129 surf_action_->set_category(get_tracing_category());
130 set_start_time(surf_action_->get_start_time());
131 set_state(State::RUNNING);
134 XBT_DEBUG("Starting communication %p from '%s' to '%s' (surf_action: %p; state: %s)", this, from_->get_cname(),
135 to_->get_cname(), surf_action_, get_state_str());
137 /* If a link is failed, detect it immediately */
138 if (surf_action_->get_state() == resource::Action::State::FAILED) {
139 XBT_DEBUG("Communication from '%s' to '%s' failed to start because of a link failure", from_->get_cname(),
141 set_state(State::LINK_FAILURE);
144 } else if ((src_actor_ != nullptr && src_actor_->is_suspended()) ||
145 (dst_actor_ != nullptr && dst_actor_->is_suspended())) {
146 /* If any of the actor is suspended, create the synchro but stop its execution,
147 it will be restarted when the sender actor resumes */
148 if (src_actor_->is_suspended())
149 XBT_DEBUG("The communication is suspended on startup because src (%s@%s) was suspended since it initiated the "
151 src_actor_->get_cname(), src_actor_->get_host()->get_cname());
153 XBT_DEBUG("The communication is suspended on startup because dst (%s@%s) was suspended since it initiated the "
155 dst_actor_->get_cname(), dst_actor_->get_host()->get_cname());
157 surf_action_->suspend();
164 std::vector<s4u::Link*> CommImpl::get_traversed_links() const
166 xbt_assert(get_state() != State::WAITING, "You cannot use %s() if your communication is not ready (%s)", __FUNCTION__,
168 std::vector<s4u::Link*> vlinks;
169 XBT_ATTRIB_UNUSED double res = 0;
170 from_->route_to(to_, vlinks, &res);
174 /** @brief Copy the communication data from the sender's buffer to the receiver's one */
175 void CommImpl::copy_data()
177 size_t buff_size = src_buff_size_;
178 /* If there is no data to copy then return */
179 if (not src_buff_ || not dst_buff_ || copied_)
182 XBT_DEBUG("Copying comm %p data from %s (%p) -> %s (%p) (%zu bytes)", this,
183 src_actor_ ? src_actor_->get_host()->get_cname() : "a finished actor", src_buff_,
184 dst_actor_ ? dst_actor_->get_host()->get_cname() : "a finished actor", dst_buff_, buff_size);
186 /* Copy at most dst_buff_size bytes of the message to receiver's buffer */
187 if (dst_buff_size_) {
188 buff_size = std::min(buff_size, *dst_buff_size_);
190 /* Update the receiver's buffer size to the copied amount */
191 *dst_buff_size_ = buff_size;
196 copy_data_fun(this, src_buff_, buff_size);
198 copy_data_callback_(this, src_buff_, buff_size);
201 /* Set the copied flag so we copy data only once */
202 /* (this function might be called from both communication ends) */
206 ActivityImplPtr CommImpl::isend(actor::CommIsendSimcall* observer)
208 auto* mbox = observer->get_mailbox();
209 XBT_DEBUG("send from mailbox %p", mbox);
211 /* Prepare a synchro describing us, so that it gets passed to the user-provided filter of other side */
212 CommImplPtr this_comm(new CommImpl());
213 this_comm->set_type(CommImplType::SEND);
215 /* Look for communication synchro matching our needs. We also provide a description of
216 * ourself so that the other side also gets a chance of choosing if it wants to match with us.
218 * If it is not found then push our communication into the rendez-vous point */
219 CommImplPtr other_comm =
220 mbox->find_matching_comm(CommImplType::RECEIVE, observer->get_match_fun(), observer->get_payload(), this_comm,
221 /*done*/ false, /*remove_matching*/ true);
223 if (not other_comm) {
224 other_comm = std::move(this_comm);
226 if (mbox->is_permanent()) {
227 // this mailbox is for small messages, which have to be sent right now
228 other_comm->set_state(State::READY);
229 other_comm->dst_actor_ = mbox->get_permanent_receiver().get();
230 mbox->push_done(other_comm);
231 XBT_DEBUG("pushing a message into the permanent receive list %p, comm %p", mbox, other_comm.get());
234 mbox->push(other_comm);
237 XBT_DEBUG("Receive already pushed");
239 other_comm->set_state(State::READY);
241 observer->set_comm(other_comm.get());
243 if (observer->is_detached()) {
244 other_comm->detach();
245 other_comm->clean_fun = observer->get_clean_fun();
247 other_comm->clean_fun = nullptr;
248 observer->get_issuer()->activities_.insert(other_comm);
251 /* Setup the communication synchro */
252 other_comm->src_actor_ = observer->get_issuer();
253 other_comm->src_data_ = observer->get_payload();
255 .set_src_buff(observer->get_src_buff(), observer->get_src_buff_size())
256 .set_size(observer->get_payload_size())
257 .set_rate(observer->get_rate());
259 other_comm->match_fun = observer->get_match_fun();
260 other_comm->copy_data_fun = observer->get_copy_data_fun();
262 if (MC_is_active() || MC_record_replay_is_active())
263 other_comm->set_state(simgrid::kernel::activity::State::RUNNING);
267 return (observer->is_detached() ? nullptr : other_comm);
270 ActivityImplPtr CommImpl::irecv(actor::CommIrecvSimcall* observer)
272 CommImplPtr this_synchro(new CommImpl());
273 this_synchro->set_type(CommImplType::RECEIVE);
275 auto* mbox = observer->get_mailbox();
276 XBT_DEBUG("recv from mbox %p. this_synchro=%p", mbox, this_synchro.get());
278 CommImplPtr other_comm;
279 // communication already done, get it inside the list of completed comms
280 if (mbox->is_permanent() && mbox->has_some_done_comm()) {
281 XBT_DEBUG("We have a comm that has probably already been received, trying to match it, to skip the communication");
282 // find a match in the list of already received comms
283 other_comm = mbox->find_matching_comm(CommImplType::SEND, observer->get_match_fun(), observer->get_payload(),
284 this_synchro, /*done*/ true, /*remove_matching*/ true);
285 if (other_comm && other_comm->surf_action_ && other_comm->get_remaining() < 1e-12) {
286 XBT_DEBUG("comm %p has been already sent, and is finished, destroy it", other_comm.get());
287 other_comm->set_state(State::DONE);
288 other_comm->set_mailbox(nullptr);
290 // if not found, assume the receiver came first, register it to the mailbox in the classical way
291 if (not other_comm) {
292 XBT_DEBUG("We have messages in the permanent receive list, but not the one we are looking for, pushing request "
294 other_comm = std::move(this_synchro);
295 mbox->push(other_comm);
297 observer->get_issuer()->activities_.insert(other_comm);
300 /* Prepare a comm describing us, so that it gets passed to the user-provided filter of other side */
302 /* Look for communication activity matching our needs. We also provide a description of
303 * ourself so that the other side also gets a chance of choosing if it wants to match with us.
305 * If it is not found then push our communication into the rendez-vous point */
306 other_comm = mbox->find_matching_comm(CommImplType::SEND, observer->get_match_fun(), observer->get_payload(),
307 this_synchro, /*done*/ false, /*remove_matching*/ true);
309 if (other_comm == nullptr) {
310 XBT_DEBUG("Receive pushed first (%zu comm enqueued so far)", mbox->size());
311 other_comm = std::move(this_synchro);
312 mbox->push(other_comm);
314 XBT_DEBUG("Match my %p with the existing %p", this_synchro.get(), other_comm.get());
316 other_comm->set_state(simgrid::kernel::activity::State::READY);
318 observer->get_issuer()->activities_.insert(other_comm);
320 observer->set_comm(other_comm.get());
322 /* Setup communication synchro */
323 other_comm->dst_actor_ = observer->get_issuer();
324 other_comm->dst_data_ = observer->get_payload();
325 other_comm->set_dst_buff(observer->get_dst_buff(), observer->get_dst_buff_size());
327 if (observer->get_rate() > -1.0 && (other_comm->get_rate() < 0.0 || observer->get_rate() < other_comm->get_rate()))
328 other_comm->set_rate(observer->get_rate());
330 other_comm->match_fun = observer->get_match_fun();
331 other_comm->copy_data_fun = observer->get_copy_data_fun();
333 if (MC_is_active() || MC_record_replay_is_active()) {
334 other_comm->set_state(State::RUNNING);
342 bool CommImpl::test(actor::ActorImpl* issuer)
344 if ((MC_is_active() || MC_record_replay_is_active()) && src_actor_ && dst_actor_)
345 set_state(State::DONE);
346 return ActivityImpl::test(issuer);
349 void CommImpl::wait_for(actor::ActorImpl* issuer, double timeout)
351 XBT_DEBUG("CommImpl::wait_for(%g), %p, state %s", timeout, this, get_state_str());
353 /* Associate this simcall to the wait synchro */
354 register_simcall(&issuer->simcall_);
355 if (MC_is_active() || MC_record_replay_is_active()) {
356 // FIXME: what about timeouts?
357 set_state(State::DONE);
361 ActivityImpl::wait_for(issuer, timeout);
364 void CommImpl::wait_any_for(actor::ActorImpl* issuer, const std::vector<CommImpl*>& comms, double timeout)
366 std::vector<ActivityImpl*> activities(comms.begin(), comms.end());
367 ActivityImpl::wait_any_for(issuer, activities, timeout);
370 void CommImpl::suspend()
372 /* FIXME: shall we suspend also the timeout synchro? */
374 surf_action_->suspend();
375 /* if not created yet, the action will be suspended on creation, in CommImpl::start() */
378 void CommImpl::resume()
380 /*FIXME: check what happen with the timeouts */
382 surf_action_->resume();
383 /* in the other case, the synchro were not really suspended yet, see CommImpl::suspend() and CommImpl::start() */
386 void CommImpl::cancel()
388 /* if the synchro is a waiting state means that it is still in a mbox so remove from it and delete it */
389 if (get_state() == State::WAITING) {
392 set_state(State::CANCELED);
394 } else if (not MC_is_active() /* when running the MC there are no surf actions */
395 && not MC_record_replay_is_active() && (get_state() == State::READY || get_state() == State::RUNNING)) {
396 surf_action_->cancel();
400 /** @brief This is part of the cleanup process, probably an internal command */
401 void CommImpl::cleanup_surf()
406 src_timeout_->unref();
407 src_timeout_ = nullptr;
411 dst_timeout_->unref();
412 dst_timeout_ = nullptr;
416 void CommImpl::post()
418 on_completion(*this);
420 /* Update synchro state */
421 if (src_timeout_ && src_timeout_->get_state() == resource::Action::State::FINISHED)
422 set_state(State::SRC_TIMEOUT);
423 else if (dst_timeout_ && dst_timeout_->get_state() == resource::Action::State::FINISHED)
424 set_state(State::DST_TIMEOUT);
425 else if ((from_ && not from_->is_on()) || (src_timeout_ && src_timeout_->get_state() == resource::Action::State::FAILED))
426 set_state(State::SRC_HOST_FAILURE);
427 else if ((to_ && not to_->is_on()) || (dst_timeout_ && dst_timeout_->get_state() == resource::Action::State::FAILED))
428 set_state(State::DST_HOST_FAILURE);
429 else if (surf_action_ && surf_action_->get_state() == resource::Action::State::FAILED) {
430 set_state(State::LINK_FAILURE);
431 } else if (get_state() == State::RUNNING) {
432 xbt_assert(from_ && from_->is_on());
433 xbt_assert(to_ && to_->is_on());
434 set_state(State::DONE);
437 XBT_DEBUG("CommImpl::post(): comm %p, state %s, src_proc %p, dst_proc %p, detached: %d", this, get_state_str(),
438 src_actor_.get(), dst_actor_.get(), detached_);
440 /* destroy the surf actions associated with the Simix communication */
443 /* Answer all simcalls associated with the synchro */
446 void CommImpl::set_exception(actor::ActorImpl* issuer)
448 switch (get_state()) {
450 issuer->exception_ = std::make_exception_ptr(NetworkFailureException(XBT_THROW_POINT, "Remote peer failed"));
452 case State::SRC_TIMEOUT:
454 std::make_exception_ptr(TimeoutException(XBT_THROW_POINT, "Communication timeouted because of the sender"));
457 case State::DST_TIMEOUT:
459 std::make_exception_ptr(TimeoutException(XBT_THROW_POINT, "Communication timeouted because of the receiver"));
462 case State::SRC_HOST_FAILURE:
463 if (issuer == src_actor_)
464 issuer->set_wannadie();
466 set_state(State::FAILED);
467 issuer->exception_ = std::make_exception_ptr(NetworkFailureException(XBT_THROW_POINT, "Remote peer failed"));
471 case State::DST_HOST_FAILURE:
472 if (issuer == dst_actor_)
473 issuer->set_wannadie();
475 set_state(State::FAILED);
476 issuer->exception_ = std::make_exception_ptr(NetworkFailureException(XBT_THROW_POINT, "Remote peer failed"));
480 case State::LINK_FAILURE:
481 XBT_DEBUG("Link failure in synchro %p between '%s' and '%s': posting an exception to the issuer: %s (%p) "
483 this, src_actor_ ? src_actor_->get_host()->get_cname() : nullptr,
484 dst_actor_ ? dst_actor_->get_host()->get_cname() : nullptr, issuer->get_cname(), issuer, detached_);
485 if (src_actor_ == issuer) {
486 XBT_DEBUG("I'm source");
487 } else if (dst_actor_ == issuer) {
488 XBT_DEBUG("I'm dest");
490 XBT_DEBUG("I'm neither source nor dest");
492 set_state(State::FAILED);
493 issuer->throw_exception(std::make_exception_ptr(NetworkFailureException(XBT_THROW_POINT, "Link failure")));
496 case State::CANCELED:
497 if (issuer == dst_actor_)
499 std::make_exception_ptr(CancelException(XBT_THROW_POINT, "Communication canceled by the sender"));
502 std::make_exception_ptr(CancelException(XBT_THROW_POINT, "Communication canceled by the receiver"));
506 xbt_assert(get_state() == State::DONE, "Internal error in CommImpl::finish(): unexpected synchro state %s",
511 void CommImpl::finish()
513 XBT_DEBUG("CommImpl::finish() in state %s", get_state_str());
514 /* If the synchro is still in a rendez-vous point then remove from it */
518 if (get_state() == State::DONE)
522 EngineImpl::get_instance()->get_maestro()->activities_.erase(this);
524 while (not simcalls_.empty()) {
525 actor::Simcall* simcall = simcalls_.front();
526 simcalls_.pop_front();
528 /* If a waitany simcall is waiting for this synchro to finish, then remove it from the other synchros in the waitany
529 * list. Afterwards, get the position of the actual synchro in the waitany list and return it as the result of the
532 if (simcall->call_ == actor::Simcall::Type::NONE) // FIXME: maybe a better way to handle this case
533 continue; // if actor handling comm is killed
535 handle_activity_waitany(simcall);
537 /* Check out for errors */
539 if (not simcall->issuer_->get_host()->is_on()) {
540 simcall->issuer_->set_wannadie();
542 // Do not answer to dying actors
543 if (not simcall->issuer_->wannadie()) {
544 set_exception(simcall->issuer_);
545 simcall->issuer_->simcall_answer();
549 simcall->issuer_->waiting_synchro_ = nullptr;
550 simcall->issuer_->activities_.erase(this);
552 if (simcall->issuer_ != dst_actor_ && dst_actor_ != nullptr)
553 dst_actor_->activities_.erase(this);
554 if (simcall->issuer_ != src_actor_ && src_actor_ != nullptr)
555 src_actor_->activities_.erase(this);
560 } // namespace simgrid::kernel::activity