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_ = [](kernel::activity::CommImpl* comm,
27 void* buff, size_t buff_size) {
28 xbt_assert((buff_size == sizeof(void*)), "Cannot copy %zu bytes: must be sizeof(void*)", buff_size);
29 *(void**)(comm->dst_buff_) = buff;
32 void CommImpl::set_copy_data_callback(const std::function<void(CommImpl*, void*, size_t)>& callback)
34 copy_data_callback_ = callback;
37 CommImpl& CommImpl::set_type(CommImplType type)
43 CommImpl& CommImpl::set_source(s4u::Host* from)
45 xbt_assert( from_ == nullptr );
51 CommImpl& CommImpl::set_destination(s4u::Host* to)
53 xbt_assert( to_ == nullptr );
59 CommImpl& CommImpl::set_size(double size)
65 CommImpl& CommImpl::set_rate(double rate)
70 CommImpl& CommImpl::set_mailbox(MailboxImpl* mbox)
73 mbox_id_ = mbox->get_id();
78 CommImpl& CommImpl::set_src_buff(unsigned char* buff, size_t size)
81 src_buff_size_ = size;
85 CommImpl& CommImpl::set_dst_buff(unsigned char* buff, size_t* size)
88 dst_buff_size_ = size;
92 CommImpl& CommImpl::detach()
95 EngineImpl::get_instance()->get_maestro()->activities_.insert(this);
101 XBT_DEBUG("Really free communication %p in state %s (detached = %d)", this, get_state_str(), detached_);
105 if (detached_ && get_state() != State::DONE) {
106 /* the communication has failed and was detached:
107 * we have to free the buffer */
109 clean_fun(src_buff_);
116 /** @brief Starts the simulation of a communication synchro. */
117 CommImpl* CommImpl::start()
119 /* If both the sender and the receiver are already there, start the communication */
120 if (get_state() == State::READY) {
121 from_ = from_ != nullptr ? from_ : src_actor_->get_host();
122 xbt_assert(from_->is_on());
123 to_ = to_ != nullptr ? to_ : dst_actor_->get_host();
124 xbt_assert(to_->is_on());
126 /* Getting the network_model from the origin host
127 * Valid while we have a single network model, otherwise we would need to change this function to first get the
128 * routes and later create the respective model actions */
129 auto net_model = from_->get_netpoint()->get_englobing_zone()->get_network_model();
131 model_action_ = net_model->communicate(from_, to_, size_, rate_, false);
132 model_action_->set_activity(this);
133 model_action_->set_category(get_tracing_category());
134 set_start_time(model_action_->get_start_time());
135 set_state(State::RUNNING);
138 XBT_DEBUG("Starting communication %p from '%s' to '%s' (model action: %p; state: %s)", this, from_->get_cname(),
139 to_->get_cname(), model_action_, get_state_str());
141 /* If a link is failed, detect it immediately */
142 if (model_action_->get_state() == resource::Action::State::FAILED) {
143 XBT_DEBUG("Communication from '%s' to '%s' failed to start because of a link failure", from_->get_cname(),
145 set_state(State::LINK_FAILURE);
148 } else if ((src_actor_ != nullptr && src_actor_->is_suspended()) ||
149 (dst_actor_ != nullptr && dst_actor_->is_suspended())) {
150 /* If any of the actor is suspended, create the synchro but stop its execution,
151 it will be restarted when the sender actor resumes */
152 if (src_actor_->is_suspended())
153 XBT_DEBUG("The communication is suspended on startup because src (%s@%s) was suspended since it initiated the "
155 src_actor_->get_cname(), src_actor_->get_host()->get_cname());
157 XBT_DEBUG("The communication is suspended on startup because dst (%s@%s) was suspended since it initiated the "
159 dst_actor_->get_cname(), dst_actor_->get_host()->get_cname());
161 model_action_->suspend();
168 std::vector<s4u::Link*> CommImpl::get_traversed_links() const
170 xbt_assert(get_state() != State::WAITING, "You cannot use %s() if your communication is not ready (%s)", __FUNCTION__,
172 std::vector<s4u::Link*> vlinks;
173 XBT_ATTRIB_UNUSED double res = 0;
174 from_->route_to(to_, vlinks, &res);
178 /** @brief Copy the communication data from the sender's buffer to the receiver's one */
179 void CommImpl::copy_data()
181 size_t buff_size = src_buff_size_;
182 /* If there is no data to copy then return */
183 if (not src_buff_ || not dst_buff_ || copied_)
186 XBT_DEBUG("Copying comm %p data from %s (%p) -> %s (%p) (%zu bytes)", this,
187 src_actor_ ? src_actor_->get_host()->get_cname() : "a finished actor", src_buff_,
188 dst_actor_ ? dst_actor_->get_host()->get_cname() : "a finished actor", dst_buff_, buff_size);
190 /* Copy at most dst_buff_size bytes of the message to receiver's buffer */
191 if (dst_buff_size_) {
192 buff_size = std::min(buff_size, *dst_buff_size_);
194 /* Update the receiver's buffer size to the copied amount */
195 *dst_buff_size_ = buff_size;
200 copy_data_fun(this, src_buff_, buff_size);
202 copy_data_callback_(this, src_buff_, buff_size);
205 /* Set the copied flag so we copy data only once */
206 /* (this function might be called from both communication ends) */
210 ActivityImplPtr CommImpl::isend(actor::CommIsendSimcall* observer)
212 auto* mbox = observer->get_mailbox();
213 XBT_DEBUG("send from mailbox %p", mbox);
215 /* Prepare a synchro describing us, so that it gets passed to the user-provided filter of other side */
216 CommImplPtr this_comm(new CommImpl());
217 this_comm->set_type(CommImplType::SEND);
219 /* Look for communication synchro matching our needs. We also provide a description of
220 * ourself so that the other side also gets a chance of choosing if it wants to match with us.
222 * If it is not found then push our communication into the rendez-vous point */
223 CommImplPtr other_comm =
224 mbox->find_matching_comm(CommImplType::RECEIVE, observer->get_match_fun(), observer->get_payload(), this_comm,
225 /*done*/ false, /*remove_matching*/ true);
227 if (not other_comm) {
228 other_comm = std::move(this_comm);
230 if (mbox->is_permanent()) {
231 // this mailbox is for small messages, which have to be sent right now
232 other_comm->set_state(State::READY);
233 other_comm->dst_actor_ = mbox->get_permanent_receiver().get();
234 mbox->push_done(other_comm);
235 XBT_DEBUG("pushing a message into the permanent receive list %p, comm %p", mbox, other_comm.get());
238 mbox->push(other_comm);
241 XBT_DEBUG("Receive already pushed");
243 other_comm->set_state(State::READY);
245 observer->set_comm(other_comm.get());
247 if (observer->is_detached()) {
248 other_comm->detach();
249 other_comm->clean_fun = observer->get_clean_fun();
251 other_comm->clean_fun = nullptr;
252 observer->get_issuer()->activities_.insert(other_comm);
255 /* Setup the communication synchro */
256 other_comm->src_actor_ = observer->get_issuer();
257 other_comm->src_data_ = observer->get_payload();
259 .set_src_buff(observer->get_src_buff(), observer->get_src_buff_size())
260 .set_size(observer->get_payload_size())
261 .set_rate(observer->get_rate());
263 other_comm->match_fun = observer->get_match_fun();
264 other_comm->copy_data_fun = observer->get_copy_data_fun();
266 if (MC_is_active() || MC_record_replay_is_active())
267 other_comm->set_state(simgrid::kernel::activity::State::RUNNING);
271 return (observer->is_detached() ? nullptr : other_comm);
274 ActivityImplPtr CommImpl::irecv(actor::CommIrecvSimcall* observer)
276 CommImplPtr this_synchro(new CommImpl());
277 this_synchro->set_type(CommImplType::RECEIVE);
279 auto* mbox = observer->get_mailbox();
280 XBT_DEBUG("recv from mbox %p. this_synchro=%p", mbox, this_synchro.get());
282 CommImplPtr other_comm;
283 // communication already done, get it inside the list of completed comms
284 if (mbox->is_permanent() && mbox->has_some_done_comm()) {
285 XBT_DEBUG("We have a comm that has probably already been received, trying to match it, to skip the communication");
286 // find a match in the list of already received comms
287 other_comm = mbox->find_matching_comm(CommImplType::SEND, observer->get_match_fun(), observer->get_payload(),
288 this_synchro, /*done*/ true, /*remove_matching*/ true);
289 if (other_comm && other_comm->model_action_ && other_comm->get_remaining() < 1e-12) {
290 XBT_DEBUG("comm %p has been already sent, and is finished, destroy it", other_comm.get());
291 other_comm->set_state(State::DONE);
292 other_comm->set_mailbox(nullptr);
294 // if not found, assume the receiver came first, register it to the mailbox in the classical way
295 if (not other_comm) {
296 XBT_DEBUG("We have messages in the permanent receive list, but not the one we are looking for, pushing request "
298 other_comm = std::move(this_synchro);
299 mbox->push(other_comm);
301 observer->get_issuer()->activities_.insert(other_comm);
304 /* Prepare a comm describing us, so that it gets passed to the user-provided filter of other side */
306 /* Look for communication activity matching our needs. We also provide a description of
307 * ourself so that the other side also gets a chance of choosing if it wants to match with us.
309 * If it is not found then push our communication into the rendez-vous point */
310 other_comm = mbox->find_matching_comm(CommImplType::SEND, observer->get_match_fun(), observer->get_payload(),
311 this_synchro, /*done*/ false, /*remove_matching*/ true);
313 if (other_comm == nullptr) {
314 XBT_DEBUG("Receive pushed first (%zu comm enqueued so far)", mbox->size());
315 other_comm = std::move(this_synchro);
316 mbox->push(other_comm);
318 XBT_DEBUG("Match my %p with the existing %p", this_synchro.get(), other_comm.get());
320 other_comm->set_state(simgrid::kernel::activity::State::READY);
322 observer->get_issuer()->activities_.insert(other_comm);
324 observer->set_comm(other_comm.get());
326 /* Setup communication synchro */
327 other_comm->dst_actor_ = observer->get_issuer();
328 other_comm->dst_data_ = observer->get_payload();
329 other_comm->set_dst_buff(observer->get_dst_buff(), observer->get_dst_buff_size());
331 if (observer->get_rate() > -1.0 && (other_comm->get_rate() < 0.0 || observer->get_rate() < other_comm->get_rate()))
332 other_comm->set_rate(observer->get_rate());
334 other_comm->match_fun = observer->get_match_fun();
335 other_comm->copy_data_fun = observer->get_copy_data_fun();
337 if (MC_is_active() || MC_record_replay_is_active()) {
338 other_comm->set_state(State::RUNNING);
346 bool CommImpl::test(actor::ActorImpl* issuer)
348 if ((MC_is_active() || MC_record_replay_is_active()) && src_actor_ && dst_actor_)
349 set_state(State::DONE);
350 return ActivityImpl::test(issuer);
353 void CommImpl::wait_for(actor::ActorImpl* issuer, double timeout)
355 XBT_DEBUG("CommImpl::wait_for(%g), %p, state %s", timeout, this, get_state_str());
357 /* Associate this simcall to the wait synchro */
358 register_simcall(&issuer->simcall_);
359 if (MC_is_active() || MC_record_replay_is_active()) {
360 // FIXME: what about timeouts?
361 set_state(State::DONE);
365 ActivityImpl::wait_for(issuer, timeout);
368 void CommImpl::wait_any_for(actor::ActorImpl* issuer, const std::vector<CommImpl*>& comms, double timeout)
370 std::vector<ActivityImpl*> activities(comms.begin(), comms.end());
371 ActivityImpl::wait_any_for(issuer, activities, timeout);
374 void CommImpl::suspend()
376 /* FIXME: shall we suspend also the timeout synchro? */
378 model_action_->suspend();
379 /* if not created yet, the action will be suspended on creation, in CommImpl::start() */
382 void CommImpl::resume()
384 /*FIXME: check what happen with the timeouts */
386 model_action_->resume();
387 /* in the other case, the synchro were not really suspended yet, see CommImpl::suspend() and CommImpl::start() */
390 void CommImpl::cancel()
392 /* if the synchro is a waiting state means that it is still in a mbox so remove from it and delete it */
393 if (get_state() == State::WAITING) {
396 set_state(State::CANCELED);
398 } else if (not MC_is_active() /* when running the MC there are no model actions */
399 && not MC_record_replay_is_active() && (get_state() == State::READY || get_state() == State::RUNNING)) {
400 model_action_->cancel();
404 void CommImpl::set_exception(actor::ActorImpl* issuer)
406 switch (get_state()) {
408 issuer->exception_ = std::make_exception_ptr(NetworkFailureException(XBT_THROW_POINT, "Remote peer failed"));
410 case State::SRC_TIMEOUT:
412 std::make_exception_ptr(TimeoutException(XBT_THROW_POINT, "Communication timeouted because of the sender"));
415 case State::DST_TIMEOUT:
417 std::make_exception_ptr(TimeoutException(XBT_THROW_POINT, "Communication timeouted because of the receiver"));
420 case State::SRC_HOST_FAILURE:
421 if (issuer == src_actor_)
422 issuer->set_wannadie();
424 set_state(State::FAILED);
425 issuer->exception_ = std::make_exception_ptr(NetworkFailureException(XBT_THROW_POINT, "Remote peer failed"));
429 case State::DST_HOST_FAILURE:
430 if (issuer == dst_actor_)
431 issuer->set_wannadie();
433 set_state(State::FAILED);
434 issuer->exception_ = std::make_exception_ptr(NetworkFailureException(XBT_THROW_POINT, "Remote peer failed"));
438 case State::LINK_FAILURE:
439 XBT_DEBUG("Link failure in synchro %p between '%s' and '%s': posting an exception to the issuer: %s (%p) "
441 this, src_actor_ ? src_actor_->get_host()->get_cname() : nullptr,
442 dst_actor_ ? dst_actor_->get_host()->get_cname() : nullptr, issuer->get_cname(), issuer, detached_);
443 if (src_actor_ == issuer) {
444 XBT_DEBUG("I'm source");
445 } else if (dst_actor_ == issuer) {
446 XBT_DEBUG("I'm dest");
448 XBT_DEBUG("I'm neither source nor dest");
450 set_state(State::FAILED);
451 issuer->throw_exception(std::make_exception_ptr(NetworkFailureException(XBT_THROW_POINT, "Link failure")));
454 case State::CANCELED:
455 if (issuer == dst_actor_)
457 std::make_exception_ptr(CancelException(XBT_THROW_POINT, "Communication canceled by the sender"));
460 std::make_exception_ptr(CancelException(XBT_THROW_POINT, "Communication canceled by the receiver"));
464 xbt_assert(get_state() == State::DONE, "Internal error in CommImpl::finish(): unexpected synchro state %s",
469 void CommImpl::finish()
471 XBT_DEBUG("CommImpl::finish() comm %p, state %s, src_proc %p, dst_proc %p, detached: %d", this, get_state_str(),
472 src_actor_.get(), dst_actor_.get(), detached_);
474 on_completion(*this);
476 /* Update synchro state */
477 if (src_timeout_ && src_timeout_->get_state() == resource::Action::State::FINISHED)
478 set_state(State::SRC_TIMEOUT);
479 else if (dst_timeout_ && dst_timeout_->get_state() == resource::Action::State::FINISHED)
480 set_state(State::DST_TIMEOUT);
481 else if ((from_ && not from_->is_on()) ||
482 (src_timeout_ && src_timeout_->get_state() == resource::Action::State::FAILED))
483 set_state(State::SRC_HOST_FAILURE);
484 else if ((to_ && not to_->is_on()) || (dst_timeout_ && dst_timeout_->get_state() == resource::Action::State::FAILED))
485 set_state(State::DST_HOST_FAILURE);
486 else if (model_action_ && model_action_->get_state() == resource::Action::State::FAILED) {
487 set_state(State::LINK_FAILURE);
488 } else if (get_state() == State::RUNNING) {
489 xbt_assert(from_ && from_->is_on());
490 xbt_assert(to_ && to_->is_on());
491 set_state(State::DONE);
493 src_timeout_ = nullptr;
494 dst_timeout_ = nullptr;
496 /* destroy the model actions associated with the communication activity */
499 /* If the synchro is still in a rendez-vous point then remove from it */
503 if (get_state() == State::DONE)
507 EngineImpl::get_instance()->get_maestro()->activities_.erase(this);
509 while (not simcalls_.empty()) {
510 actor::Simcall* simcall = simcalls_.front();
511 simcalls_.pop_front();
513 /* If a waitany simcall is waiting for this synchro to finish, then remove it from the other synchros in the waitany
514 * list. Afterwards, get the position of the actual synchro in the waitany list and return it as the result of the
517 if (simcall->call_ == actor::Simcall::Type::NONE) // FIXME: maybe a better way to handle this case
518 continue; // if actor handling comm is killed
520 handle_activity_waitany(simcall);
522 /* Check out for errors */
524 if (not simcall->issuer_->get_host()->is_on()) {
525 simcall->issuer_->set_wannadie();
527 // Do not answer to dying actors
528 if (not simcall->issuer_->wannadie()) {
529 set_exception(simcall->issuer_);
530 simcall->issuer_->simcall_answer();
534 simcall->issuer_->waiting_synchro_ = nullptr;
535 simcall->issuer_->activities_.erase(this);
537 if (simcall->issuer_ != dst_actor_ && dst_actor_ != nullptr)
538 dst_actor_->activities_.erase(this);
539 if (simcall->issuer_ != src_actor_ && src_actor_ != nullptr)
540 src_actor_->activities_.erase(this);
545 } // namespace simgrid::kernel::activity