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 {
24 unsigned CommImpl::next_id_ = 0;
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 if (comm->dst_buff_ != nullptr) // get_async provided a buffer
30 *(void**)(comm->dst_buff_) = buff;
31 comm->payload_ = buff; // Setup what will be retrieved by s4u::Comm::get_payload()
34 void CommImpl::set_copy_data_callback(const std::function<void(CommImpl*, void*, size_t)>& callback)
36 copy_data_callback_ = callback;
39 CommImpl& CommImpl::set_type(CommImplType type)
45 CommImpl& CommImpl::set_source(s4u::Host* from)
47 xbt_assert( from_ == nullptr );
53 CommImpl& CommImpl::set_destination(s4u::Host* to)
55 xbt_assert( to_ == nullptr );
61 CommImpl& CommImpl::set_size(double size)
67 CommImpl& CommImpl::set_rate(double rate)
72 CommImpl& CommImpl::set_mailbox(MailboxImpl* mbox)
75 mbox_id_ = mbox->get_id();
80 CommImpl& CommImpl::set_src_buff(unsigned char* buff, size_t size)
83 src_buff_size_ = size;
87 CommImpl& CommImpl::set_dst_buff(unsigned char* buff, size_t* size)
90 dst_buff_size_ = size;
94 CommImpl& CommImpl::detach()
97 EngineImpl::get_instance()->get_maestro()->activities_.insert(this);
101 CommImpl::~CommImpl()
103 XBT_DEBUG("Really free communication %p in state %s (detached = %d)", this, get_state_str(), detached_);
107 if (detached_ && get_state() != State::DONE) {
108 /* the communication has failed and was detached:
109 * we have to free the buffer */
111 clean_fun(src_buff_);
118 /** @brief Starts the simulation of a communication synchro. */
119 CommImpl* CommImpl::start()
121 /* If both the sender and the receiver are already there, start the communication */
122 if (get_state() == State::READY) {
123 from_ = from_ != nullptr ? from_ : src_actor_->get_host();
124 xbt_assert(from_->is_on());
125 to_ = to_ != nullptr ? to_ : dst_actor_->get_host();
126 xbt_assert(to_->is_on());
128 /* Getting the network_model from the origin host
129 * Valid while we have a single network model, otherwise we would need to change this function to first get the
130 * routes and later create the respective model actions */
131 auto net_model = from_->get_netpoint()->get_englobing_zone()->get_network_model();
133 model_action_ = net_model->communicate(from_, to_, size_, rate_, false);
134 model_action_->set_activity(this);
135 model_action_->set_category(get_tracing_category());
136 set_start_time(model_action_->get_start_time());
137 set_state(State::RUNNING);
139 XBT_DEBUG("Starting communication %p from '%s' to '%s' (model action: %p; state: %s)", this, from_->get_cname(),
140 to_->get_cname(), model_action_, get_state_str());
142 /* If a link is failed, detect it immediately */
143 if (model_action_->get_state() == resource::Action::State::FAILED) {
144 XBT_DEBUG("Communication from '%s' to '%s' failed to start because of a link failure", from_->get_cname(),
146 set_state(State::LINK_FAILURE);
149 } else if ((src_actor_ != nullptr && src_actor_->is_suspended()) ||
150 (dst_actor_ != nullptr && dst_actor_->is_suspended())) {
151 /* If any of the actor is suspended, create the synchro but stop its execution,
152 it will be restarted when the sender actor resumes */
153 if (src_actor_->is_suspended())
154 XBT_DEBUG("The communication is suspended on startup because src (%s@%s) was suspended since it initiated the "
156 src_actor_->get_cname(), src_actor_->get_host()->get_cname());
158 XBT_DEBUG("The communication is suspended on startup because dst (%s@%s) was suspended since it initiated the "
160 dst_actor_->get_cname(), dst_actor_->get_host()->get_cname());
162 model_action_->suspend();
169 std::vector<s4u::Link*> CommImpl::get_traversed_links() const
171 xbt_assert(get_state() != State::WAITING, "You cannot use %s() if your communication is not ready (%s)", __func__,
173 std::vector<s4u::Link*> vlinks;
174 XBT_ATTRIB_UNUSED double res = 0;
175 from_->route_to(to_, vlinks, &res);
179 /** @brief Copy the communication data from the sender's buffer to the receiver's one */
180 void CommImpl::copy_data()
182 size_t buff_size = src_buff_size_;
183 /* If there is no data to copy then return */
184 if (not src_buff_ || not dst_buff_size_ || copied_)
187 XBT_DEBUG("Copying comm %p data from %s (%p) -> %s (%p) (%zu bytes)", this,
188 src_actor_ ? src_actor_->get_host()->get_cname() : "a finished actor", src_buff_,
189 dst_actor_ ? dst_actor_->get_host()->get_cname() : "a finished actor", dst_buff_, buff_size);
191 /* Copy at most dst_buff_size bytes of the message to receiver's buffer */
192 if (dst_buff_size_) {
193 buff_size = std::min(buff_size, *dst_buff_size_);
195 /* Update the receiver's buffer size to the copied amount */
196 *dst_buff_size_ = buff_size;
201 copy_data_fun(this, src_buff_, buff_size);
203 copy_data_callback_(this, src_buff_, buff_size);
206 /* Set the copied flag so we copy data only once */
207 /* (this function might be called from both communication ends) */
211 ActivityImplPtr CommImpl::isend(actor::CommIsendSimcall* observer)
213 auto* mbox = observer->get_mailbox();
214 XBT_DEBUG("send from mailbox %p", mbox);
216 /* Prepare a synchro describing us, so that it gets passed to the user-provided filter of other side */
217 CommImplPtr this_comm(new CommImpl());
218 this_comm->set_type(CommImplType::SEND);
220 /* Look for communication synchro matching our needs. We also provide a description of
221 * ourself so that the other side also gets a chance of choosing if it wants to match with us.
223 * If it is not found then push our communication into the rendez-vous point */
224 CommImplPtr other_comm =
225 mbox->find_matching_comm(CommImplType::RECEIVE, observer->get_match_fun(), observer->get_payload(), this_comm,
226 /*done*/ false, /*remove_matching*/ true);
228 if (not other_comm) {
229 other_comm = std::move(this_comm);
231 if (mbox->is_permanent()) {
232 // this mailbox is for small messages, which have to be sent right now
233 other_comm->set_state(State::READY);
234 other_comm->dst_actor_ = mbox->get_permanent_receiver().get();
235 mbox->push_done(other_comm);
236 XBT_DEBUG("pushing a message into the permanent receive list %p, comm %p", mbox, other_comm.get());
239 mbox->push(other_comm);
242 XBT_DEBUG("Receive already pushed");
244 other_comm->set_state(State::READY);
246 observer->set_comm(other_comm.get());
248 if (observer->is_detached()) {
249 other_comm->detach();
250 other_comm->clean_fun = observer->get_clean_fun();
252 other_comm->clean_fun = nullptr;
253 observer->get_issuer()->activities_.insert(other_comm);
256 /* Setup the communication synchro */
257 other_comm->src_actor_ = observer->get_issuer();
258 other_comm->src_data_ = observer->get_payload();
260 .set_src_buff(observer->get_src_buff(), observer->get_src_buff_size())
261 .set_size(observer->get_payload_size())
262 .set_rate(observer->get_rate());
264 other_comm->match_fun = observer->get_match_fun();
265 other_comm->copy_data_fun = observer->get_copy_data_fun();
267 if (MC_is_active() || MC_record_replay_is_active())
268 other_comm->set_state(simgrid::kernel::activity::State::RUNNING);
272 return (observer->is_detached() ? nullptr : other_comm);
275 ActivityImplPtr CommImpl::irecv(actor::CommIrecvSimcall* observer)
277 CommImplPtr this_synchro(new CommImpl());
278 this_synchro->set_type(CommImplType::RECEIVE);
280 auto* mbox = observer->get_mailbox();
281 XBT_DEBUG("recv from mbox %p. this_synchro=%p", mbox, this_synchro.get());
283 CommImplPtr other_comm;
284 // communication already done, get it inside the list of completed comms
285 if (mbox->is_permanent() && mbox->has_some_done_comm()) {
286 XBT_DEBUG("We have a comm that has probably already been received, trying to match it, to skip the communication");
287 // find a match in the list of already received comms
288 other_comm = mbox->find_matching_comm(CommImplType::SEND, observer->get_match_fun(), observer->get_payload(),
289 this_synchro, /*done*/ true, /*remove_matching*/ true);
290 if (other_comm && other_comm->model_action_ && other_comm->get_remaining() < 1e-12) {
291 XBT_DEBUG("comm %p has been already sent, and is finished, destroy it", other_comm.get());
292 other_comm->set_state(State::DONE);
293 other_comm->set_mailbox(nullptr);
295 // if not found, assume the receiver came first, register it to the mailbox in the classical way
296 if (not other_comm) {
297 XBT_DEBUG("We have messages in the permanent receive list, but not the one we are looking for, pushing request "
299 other_comm = std::move(this_synchro);
300 mbox->push(other_comm);
302 observer->get_issuer()->activities_.insert(other_comm);
305 /* Prepare a comm describing us, so that it gets passed to the user-provided filter of other side */
307 /* Look for communication activity matching our needs. We also provide a description of
308 * ourself so that the other side also gets a chance of choosing if it wants to match with us.
310 * If it is not found then push our communication into the rendez-vous point */
311 other_comm = mbox->find_matching_comm(CommImplType::SEND, observer->get_match_fun(), observer->get_payload(),
312 this_synchro, /*done*/ false, /*remove_matching*/ true);
314 if (other_comm == nullptr) {
315 XBT_DEBUG("Receive pushed first (%zu comm enqueued so far)", mbox->size());
316 other_comm = std::move(this_synchro);
317 mbox->push(other_comm);
319 XBT_DEBUG("Match my %p with the existing %p", this_synchro.get(), other_comm.get());
321 other_comm->set_state(simgrid::kernel::activity::State::READY);
323 observer->get_issuer()->activities_.insert(other_comm);
325 observer->set_comm(other_comm.get());
327 /* Setup communication synchro */
328 other_comm->dst_actor_ = observer->get_issuer();
329 other_comm->dst_data_ = observer->get_payload();
330 other_comm->set_dst_buff(observer->get_dst_buff(), observer->get_dst_buff_size());
332 if (observer->get_rate() > -1.0 && (other_comm->get_rate() < 0.0 || observer->get_rate() < other_comm->get_rate()))
333 other_comm->set_rate(observer->get_rate());
335 other_comm->match_fun = observer->get_match_fun();
336 other_comm->copy_data_fun = observer->get_copy_data_fun();
338 if (MC_is_active() || MC_record_replay_is_active()) {
339 other_comm->set_state(State::RUNNING);
347 bool CommImpl::test(actor::ActorImpl* issuer)
349 if ((MC_is_active() || MC_record_replay_is_active()) && src_actor_ && dst_actor_)
350 set_state(State::DONE);
351 return ActivityImpl::test(issuer);
354 void CommImpl::wait_for(actor::ActorImpl* issuer, double timeout)
356 XBT_DEBUG("CommImpl::wait_for(%g), %p, state %s", timeout, this, get_state_str());
358 /* Associate this simcall to the wait synchro */
359 register_simcall(&issuer->simcall_);
360 if (MC_is_active() || MC_record_replay_is_active()) {
361 // FIXME: what about timeouts?
362 set_state(State::DONE);
366 ActivityImpl::wait_for(issuer, timeout);
369 void CommImpl::wait_any_for(actor::ActorImpl* issuer, const std::vector<CommImpl*>& comms, double timeout)
371 std::vector<ActivityImpl*> activities(comms.begin(), comms.end());
372 ActivityImpl::wait_any_for(issuer, activities, timeout);
375 void CommImpl::suspend()
377 /* FIXME: shall we suspend also the timeout synchro? */
379 model_action_->suspend();
380 /* if not created yet, the action will be suspended on creation, in CommImpl::start() */
383 void CommImpl::resume()
385 /*FIXME: check what happen with the timeouts */
387 model_action_->resume();
388 /* in the other case, the synchro were not really suspended yet, see CommImpl::suspend() and CommImpl::start() */
391 void CommImpl::cancel()
393 /* if the synchro is a waiting state means that it is still in a mbox so remove from it and delete it */
394 if (get_state() == State::WAITING) {
397 set_state(State::CANCELED);
399 } else if (not MC_is_active() /* when running the MC there are no model actions */
400 && not MC_record_replay_is_active() && (get_state() == State::READY || get_state() == State::RUNNING)) {
401 model_action_->cancel();
405 void CommImpl::set_exception(actor::ActorImpl* issuer)
407 switch (get_state()) {
409 issuer->exception_ = std::make_exception_ptr(NetworkFailureException(XBT_THROW_POINT, "Remote peer failed"));
411 case State::SRC_TIMEOUT:
413 std::make_exception_ptr(TimeoutException(XBT_THROW_POINT, "Communication timeouted because of the sender"));
416 case State::DST_TIMEOUT:
418 std::make_exception_ptr(TimeoutException(XBT_THROW_POINT, "Communication timeouted because of the receiver"));
421 case State::SRC_HOST_FAILURE:
422 if (issuer == src_actor_)
423 issuer->set_wannadie();
425 set_state(State::FAILED);
426 issuer->exception_ = std::make_exception_ptr(NetworkFailureException(XBT_THROW_POINT, "Remote peer failed"));
430 case State::DST_HOST_FAILURE:
431 if (issuer == dst_actor_)
432 issuer->set_wannadie();
434 set_state(State::FAILED);
435 issuer->exception_ = std::make_exception_ptr(NetworkFailureException(XBT_THROW_POINT, "Remote peer failed"));
439 case State::LINK_FAILURE:
440 XBT_DEBUG("Link failure in synchro %p between '%s' and '%s': posting an exception to the issuer: %s (%p) "
442 this, src_actor_ ? src_actor_->get_host()->get_cname() : nullptr,
443 dst_actor_ ? dst_actor_->get_host()->get_cname() : nullptr, issuer->get_cname(), issuer, detached_);
444 if (src_actor_ == issuer) {
445 XBT_DEBUG("I'm source");
446 } else if (dst_actor_ == issuer) {
447 XBT_DEBUG("I'm dest");
449 XBT_DEBUG("I'm neither source nor dest");
451 set_state(State::FAILED);
452 issuer->throw_exception(std::make_exception_ptr(NetworkFailureException(XBT_THROW_POINT, "Link failure")));
455 case State::CANCELED:
456 if (issuer == dst_actor_)
458 std::make_exception_ptr(CancelException(XBT_THROW_POINT, "Communication canceled by the sender"));
461 std::make_exception_ptr(CancelException(XBT_THROW_POINT, "Communication canceled by the receiver"));
465 xbt_assert(get_state() == State::DONE, "Internal error in CommImpl::finish(): unexpected synchro state %s",
470 void CommImpl::finish()
472 XBT_DEBUG("CommImpl::finish() comm %p, state %s, src_proc %p, dst_proc %p, detached: %d", this, get_state_str(),
473 src_actor_.get(), dst_actor_.get(), detached_);
476 const auto& piface = static_cast<const s4u::Comm&>(*get_iface());
477 set_iface(nullptr); // reset iface to protect against multiple trigger of the on_completion signals
478 piface.fire_on_completion_for_real();
479 piface.fire_on_this_completion_for_real();
482 /* Update synchro state */
483 if (src_timeout_ && src_timeout_->get_state() == resource::Action::State::FINISHED)
484 set_state(State::SRC_TIMEOUT);
485 else if (dst_timeout_ && dst_timeout_->get_state() == resource::Action::State::FINISHED)
486 set_state(State::DST_TIMEOUT);
487 else if ((from_ && not from_->is_on()) ||
488 (src_timeout_ && src_timeout_->get_state() == resource::Action::State::FAILED))
489 set_state(State::SRC_HOST_FAILURE);
490 else if ((to_ && not to_->is_on()) || (dst_timeout_ && dst_timeout_->get_state() == resource::Action::State::FAILED))
491 set_state(State::DST_HOST_FAILURE);
492 else if (model_action_ && model_action_->get_state() == resource::Action::State::FAILED) {
493 set_state(State::LINK_FAILURE);
494 } else if (get_state() == State::RUNNING) {
495 xbt_assert(from_ && from_->is_on());
496 xbt_assert(to_ && to_->is_on());
497 set_state(State::DONE);
499 src_timeout_ = nullptr;
500 dst_timeout_ = nullptr;
502 /* destroy the model actions associated with the communication activity */
505 /* If the synchro is still in a rendez-vous point then remove from it */
509 if (get_state() == State::DONE)
513 EngineImpl::get_instance()->get_maestro()->activities_.erase(this);
515 while (not simcalls_.empty()) {
516 actor::Simcall* simcall = simcalls_.front();
517 simcalls_.pop_front();
519 /* If a waitany simcall is waiting for this synchro to finish, then remove it from the other synchros in the waitany
520 * list. Afterwards, get the position of the actual synchro in the waitany list and return it as the result of the
523 if (simcall->call_ == actor::Simcall::Type::NONE) // FIXME: maybe a better way to handle this case
524 continue; // if actor handling comm is killed
526 handle_activity_waitany(simcall);
528 /* Check out for errors */
530 if (not simcall->issuer_->get_host()->is_on()) {
531 simcall->issuer_->set_wannadie();
533 // Do not answer to dying actors
534 if (not simcall->issuer_->wannadie()) {
535 set_exception(simcall->issuer_);
536 simcall->issuer_->simcall_answer();
540 simcall->issuer_->waiting_synchro_ = nullptr;
541 simcall->issuer_->activities_.erase(this);
543 if (simcall->issuer_ != dst_actor_ && dst_actor_ != nullptr)
544 dst_actor_->activities_.erase(this);
545 if (simcall->issuer_ != src_actor_ && src_actor_ != nullptr)
546 src_actor_->activities_.erase(this);
551 } // namespace simgrid::kernel::activity