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 std::function<void(CommImpl*, void*, size_t)> CommImpl::copy_data_callback_ = [](kernel::activity::CommImpl* comm,
25 void* buff, size_t buff_size) {
26 xbt_assert((buff_size == sizeof(void*)), "Cannot copy %zu bytes: must be sizeof(void*)", buff_size);
27 *(void**)(comm->dst_buff_) = buff;
30 void CommImpl::set_copy_data_callback(const std::function<void(CommImpl*, void*, size_t)>& callback)
32 copy_data_callback_ = callback;
35 CommImpl& CommImpl::set_type(CommImplType type)
41 CommImpl& CommImpl::set_source(s4u::Host* from)
43 xbt_assert( from_ == nullptr );
49 CommImpl& CommImpl::set_destination(s4u::Host* to)
51 xbt_assert( to_ == nullptr );
57 CommImpl& CommImpl::set_size(double size)
63 CommImpl& CommImpl::set_rate(double rate)
68 CommImpl& CommImpl::set_mailbox(MailboxImpl* mbox)
71 mbox_id_ = mbox->get_id();
76 CommImpl& CommImpl::set_src_buff(unsigned char* buff, size_t size)
79 src_buff_size_ = size;
83 CommImpl& CommImpl::set_dst_buff(unsigned char* buff, size_t* size)
86 dst_buff_size_ = size;
90 CommImpl& CommImpl::detach()
93 EngineImpl::get_instance()->get_maestro()->activities_.insert(this);
99 XBT_DEBUG("Really free communication %p in state %s (detached = %d)", this, get_state_str(), detached_);
103 if (detached_ && get_state() != State::DONE) {
104 /* the communication has failed and was detached:
105 * we have to free the buffer */
107 clean_fun(src_buff_);
114 /** @brief Starts the simulation of a communication synchro. */
115 CommImpl* CommImpl::start()
117 /* If both the sender and the receiver are already there, start the communication */
118 if (get_state() == State::READY) {
119 from_ = from_ != nullptr ? from_ : src_actor_->get_host();
120 xbt_assert(from_->is_on());
121 to_ = to_ != nullptr ? to_ : dst_actor_->get_host();
122 xbt_assert(to_->is_on());
124 /* Getting the network_model from the origin host
125 * Valid while we have a single network model, otherwise we would need to change this function to first get the
126 * routes and later create the respective model actions */
127 auto net_model = from_->get_netpoint()->get_englobing_zone()->get_network_model();
129 model_action_ = net_model->communicate(from_, to_, size_, rate_, false);
130 model_action_->set_activity(this);
131 model_action_->set_category(get_tracing_category());
132 set_start_time(model_action_->get_start_time());
133 set_state(State::RUNNING);
135 XBT_DEBUG("Starting communication %p from '%s' to '%s' (model action: %p; state: %s)", this, from_->get_cname(),
136 to_->get_cname(), model_action_, get_state_str());
138 /* If a link is failed, detect it immediately */
139 if (model_action_->get_state() == resource::Action::State::FAILED) {
140 XBT_DEBUG("Communication from '%s' to '%s' failed to start because of a link failure", from_->get_cname(),
142 set_state(State::LINK_FAILURE);
145 } else if ((src_actor_ != nullptr && src_actor_->is_suspended()) ||
146 (dst_actor_ != nullptr && dst_actor_->is_suspended())) {
147 /* If any of the actor is suspended, create the synchro but stop its execution,
148 it will be restarted when the sender actor resumes */
149 if (src_actor_->is_suspended())
150 XBT_DEBUG("The communication is suspended on startup because src (%s@%s) was suspended since it initiated the "
152 src_actor_->get_cname(), src_actor_->get_host()->get_cname());
154 XBT_DEBUG("The communication is suspended on startup because dst (%s@%s) was suspended since it initiated the "
156 dst_actor_->get_cname(), dst_actor_->get_host()->get_cname());
158 model_action_->suspend();
165 std::vector<s4u::Link*> CommImpl::get_traversed_links() const
167 xbt_assert(get_state() != State::WAITING, "You cannot use %s() if your communication is not ready (%s)", __FUNCTION__,
169 std::vector<s4u::Link*> vlinks;
170 XBT_ATTRIB_UNUSED double res = 0;
171 from_->route_to(to_, vlinks, &res);
175 /** @brief Copy the communication data from the sender's buffer to the receiver's one */
176 void CommImpl::copy_data()
178 size_t buff_size = src_buff_size_;
179 /* If there is no data to copy then return */
180 if (not src_buff_ || not dst_buff_ || copied_)
183 XBT_DEBUG("Copying comm %p data from %s (%p) -> %s (%p) (%zu bytes)", this,
184 src_actor_ ? src_actor_->get_host()->get_cname() : "a finished actor", src_buff_,
185 dst_actor_ ? dst_actor_->get_host()->get_cname() : "a finished actor", dst_buff_, buff_size);
187 /* Copy at most dst_buff_size bytes of the message to receiver's buffer */
188 if (dst_buff_size_) {
189 buff_size = std::min(buff_size, *dst_buff_size_);
191 /* Update the receiver's buffer size to the copied amount */
192 *dst_buff_size_ = buff_size;
197 copy_data_fun(this, src_buff_, buff_size);
199 copy_data_callback_(this, src_buff_, buff_size);
202 /* Set the copied flag so we copy data only once */
203 /* (this function might be called from both communication ends) */
207 ActivityImplPtr CommImpl::isend(actor::CommIsendSimcall* observer)
209 auto* mbox = observer->get_mailbox();
210 XBT_DEBUG("send from mailbox %p", mbox);
212 /* Prepare a synchro describing us, so that it gets passed to the user-provided filter of other side */
213 CommImplPtr this_comm(new CommImpl());
214 this_comm->set_type(CommImplType::SEND);
216 /* Look for communication synchro matching our needs. We also provide a description of
217 * ourself so that the other side also gets a chance of choosing if it wants to match with us.
219 * If it is not found then push our communication into the rendez-vous point */
220 CommImplPtr other_comm =
221 mbox->find_matching_comm(CommImplType::RECEIVE, observer->get_match_fun(), observer->get_payload(), this_comm,
222 /*done*/ false, /*remove_matching*/ true);
224 if (not other_comm) {
225 other_comm = std::move(this_comm);
227 if (mbox->is_permanent()) {
228 // this mailbox is for small messages, which have to be sent right now
229 other_comm->set_state(State::READY);
230 other_comm->dst_actor_ = mbox->get_permanent_receiver().get();
231 mbox->push_done(other_comm);
232 XBT_DEBUG("pushing a message into the permanent receive list %p, comm %p", mbox, other_comm.get());
235 mbox->push(other_comm);
238 XBT_DEBUG("Receive already pushed");
240 other_comm->set_state(State::READY);
242 observer->set_comm(other_comm.get());
244 if (observer->is_detached()) {
245 other_comm->detach();
246 other_comm->clean_fun = observer->get_clean_fun();
248 other_comm->clean_fun = nullptr;
249 observer->get_issuer()->activities_.insert(other_comm);
252 /* Setup the communication synchro */
253 other_comm->src_actor_ = observer->get_issuer();
254 other_comm->src_data_ = observer->get_payload();
256 .set_src_buff(observer->get_src_buff(), observer->get_src_buff_size())
257 .set_size(observer->get_payload_size())
258 .set_rate(observer->get_rate());
260 other_comm->match_fun = observer->get_match_fun();
261 other_comm->copy_data_fun = observer->get_copy_data_fun();
263 if (MC_is_active() || MC_record_replay_is_active())
264 other_comm->set_state(simgrid::kernel::activity::State::RUNNING);
268 return (observer->is_detached() ? nullptr : other_comm);
271 ActivityImplPtr CommImpl::irecv(actor::CommIrecvSimcall* observer)
273 CommImplPtr this_synchro(new CommImpl());
274 this_synchro->set_type(CommImplType::RECEIVE);
276 auto* mbox = observer->get_mailbox();
277 XBT_DEBUG("recv from mbox %p. this_synchro=%p", mbox, this_synchro.get());
279 CommImplPtr other_comm;
280 // communication already done, get it inside the list of completed comms
281 if (mbox->is_permanent() && mbox->has_some_done_comm()) {
282 XBT_DEBUG("We have a comm that has probably already been received, trying to match it, to skip the communication");
283 // find a match in the list of already received comms
284 other_comm = mbox->find_matching_comm(CommImplType::SEND, observer->get_match_fun(), observer->get_payload(),
285 this_synchro, /*done*/ true, /*remove_matching*/ true);
286 if (other_comm && other_comm->model_action_ && other_comm->get_remaining() < 1e-12) {
287 XBT_DEBUG("comm %p has been already sent, and is finished, destroy it", other_comm.get());
288 other_comm->set_state(State::DONE);
289 other_comm->set_mailbox(nullptr);
291 // if not found, assume the receiver came first, register it to the mailbox in the classical way
292 if (not other_comm) {
293 XBT_DEBUG("We have messages in the permanent receive list, but not the one we are looking for, pushing request "
295 other_comm = std::move(this_synchro);
296 mbox->push(other_comm);
298 observer->get_issuer()->activities_.insert(other_comm);
301 /* Prepare a comm describing us, so that it gets passed to the user-provided filter of other side */
303 /* Look for communication activity matching our needs. We also provide a description of
304 * ourself so that the other side also gets a chance of choosing if it wants to match with us.
306 * If it is not found then push our communication into the rendez-vous point */
307 other_comm = mbox->find_matching_comm(CommImplType::SEND, observer->get_match_fun(), observer->get_payload(),
308 this_synchro, /*done*/ false, /*remove_matching*/ true);
310 if (other_comm == nullptr) {
311 XBT_DEBUG("Receive pushed first (%zu comm enqueued so far)", mbox->size());
312 other_comm = std::move(this_synchro);
313 mbox->push(other_comm);
315 XBT_DEBUG("Match my %p with the existing %p", this_synchro.get(), other_comm.get());
317 other_comm->set_state(simgrid::kernel::activity::State::READY);
319 observer->get_issuer()->activities_.insert(other_comm);
321 observer->set_comm(other_comm.get());
323 /* Setup communication synchro */
324 other_comm->dst_actor_ = observer->get_issuer();
325 other_comm->dst_data_ = observer->get_payload();
326 other_comm->set_dst_buff(observer->get_dst_buff(), observer->get_dst_buff_size());
328 if (observer->get_rate() > -1.0 && (other_comm->get_rate() < 0.0 || observer->get_rate() < other_comm->get_rate()))
329 other_comm->set_rate(observer->get_rate());
331 other_comm->match_fun = observer->get_match_fun();
332 other_comm->copy_data_fun = observer->get_copy_data_fun();
334 if (MC_is_active() || MC_record_replay_is_active()) {
335 other_comm->set_state(State::RUNNING);
343 bool CommImpl::test(actor::ActorImpl* issuer)
345 if ((MC_is_active() || MC_record_replay_is_active()) && src_actor_ && dst_actor_)
346 set_state(State::DONE);
347 return ActivityImpl::test(issuer);
350 void CommImpl::wait_for(actor::ActorImpl* issuer, double timeout)
352 XBT_DEBUG("CommImpl::wait_for(%g), %p, state %s", timeout, this, get_state_str());
354 /* Associate this simcall to the wait synchro */
355 register_simcall(&issuer->simcall_);
356 if (MC_is_active() || MC_record_replay_is_active()) {
357 // FIXME: what about timeouts?
358 set_state(State::DONE);
362 ActivityImpl::wait_for(issuer, timeout);
365 void CommImpl::wait_any_for(actor::ActorImpl* issuer, const std::vector<CommImpl*>& comms, double timeout)
367 std::vector<ActivityImpl*> activities(comms.begin(), comms.end());
368 ActivityImpl::wait_any_for(issuer, activities, timeout);
371 void CommImpl::suspend()
373 /* FIXME: shall we suspend also the timeout synchro? */
375 model_action_->suspend();
376 /* if not created yet, the action will be suspended on creation, in CommImpl::start() */
379 void CommImpl::resume()
381 /*FIXME: check what happen with the timeouts */
383 model_action_->resume();
384 /* in the other case, the synchro were not really suspended yet, see CommImpl::suspend() and CommImpl::start() */
387 void CommImpl::cancel()
389 /* if the synchro is a waiting state means that it is still in a mbox so remove from it and delete it */
390 if (get_state() == State::WAITING) {
393 set_state(State::CANCELED);
395 } else if (not MC_is_active() /* when running the MC there are no model actions */
396 && not MC_record_replay_is_active() && (get_state() == State::READY || get_state() == State::RUNNING)) {
397 model_action_->cancel();
401 void CommImpl::set_exception(actor::ActorImpl* issuer)
403 switch (get_state()) {
405 issuer->exception_ = std::make_exception_ptr(NetworkFailureException(XBT_THROW_POINT, "Remote peer failed"));
407 case State::SRC_TIMEOUT:
409 std::make_exception_ptr(TimeoutException(XBT_THROW_POINT, "Communication timeouted because of the sender"));
412 case State::DST_TIMEOUT:
414 std::make_exception_ptr(TimeoutException(XBT_THROW_POINT, "Communication timeouted because of the receiver"));
417 case State::SRC_HOST_FAILURE:
418 if (issuer == src_actor_)
419 issuer->set_wannadie();
421 set_state(State::FAILED);
422 issuer->exception_ = std::make_exception_ptr(NetworkFailureException(XBT_THROW_POINT, "Remote peer failed"));
426 case State::DST_HOST_FAILURE:
427 if (issuer == dst_actor_)
428 issuer->set_wannadie();
430 set_state(State::FAILED);
431 issuer->exception_ = std::make_exception_ptr(NetworkFailureException(XBT_THROW_POINT, "Remote peer failed"));
435 case State::LINK_FAILURE:
436 XBT_DEBUG("Link failure in synchro %p between '%s' and '%s': posting an exception to the issuer: %s (%p) "
438 this, src_actor_ ? src_actor_->get_host()->get_cname() : nullptr,
439 dst_actor_ ? dst_actor_->get_host()->get_cname() : nullptr, issuer->get_cname(), issuer, detached_);
440 if (src_actor_ == issuer) {
441 XBT_DEBUG("I'm source");
442 } else if (dst_actor_ == issuer) {
443 XBT_DEBUG("I'm dest");
445 XBT_DEBUG("I'm neither source nor dest");
447 set_state(State::FAILED);
448 issuer->throw_exception(std::make_exception_ptr(NetworkFailureException(XBT_THROW_POINT, "Link failure")));
451 case State::CANCELED:
452 if (issuer == dst_actor_)
454 std::make_exception_ptr(CancelException(XBT_THROW_POINT, "Communication canceled by the sender"));
457 std::make_exception_ptr(CancelException(XBT_THROW_POINT, "Communication canceled by the receiver"));
461 xbt_assert(get_state() == State::DONE, "Internal error in CommImpl::finish(): unexpected synchro state %s",
466 void CommImpl::finish()
468 XBT_DEBUG("CommImpl::finish() comm %p, state %s, src_proc %p, dst_proc %p, detached: %d", this, get_state_str(),
469 src_actor_.get(), dst_actor_.get(), detached_);
472 const auto& piface = static_cast<const s4u::Comm&>(*get_iface());
473 set_iface(nullptr); // reset iface to protect against multiple trigger of the on_completion signals
474 s4u::Comm::on_completion(piface);
475 piface.on_this_completion(piface);
478 /* Update synchro state */
479 if (src_timeout_ && src_timeout_->get_state() == resource::Action::State::FINISHED)
480 set_state(State::SRC_TIMEOUT);
481 else if (dst_timeout_ && dst_timeout_->get_state() == resource::Action::State::FINISHED)
482 set_state(State::DST_TIMEOUT);
483 else if ((from_ && not from_->is_on()) ||
484 (src_timeout_ && src_timeout_->get_state() == resource::Action::State::FAILED))
485 set_state(State::SRC_HOST_FAILURE);
486 else if ((to_ && not to_->is_on()) || (dst_timeout_ && dst_timeout_->get_state() == resource::Action::State::FAILED))
487 set_state(State::DST_HOST_FAILURE);
488 else if (model_action_ && model_action_->get_state() == resource::Action::State::FAILED) {
489 set_state(State::LINK_FAILURE);
490 } else if (get_state() == State::RUNNING) {
491 xbt_assert(from_ && from_->is_on());
492 xbt_assert(to_ && to_->is_on());
493 set_state(State::DONE);
495 src_timeout_ = nullptr;
496 dst_timeout_ = nullptr;
498 /* destroy the model actions associated with the communication activity */
501 /* If the synchro is still in a rendez-vous point then remove from it */
505 if (get_state() == State::DONE)
509 EngineImpl::get_instance()->get_maestro()->activities_.erase(this);
511 while (not simcalls_.empty()) {
512 actor::Simcall* simcall = simcalls_.front();
513 simcalls_.pop_front();
515 /* If a waitany simcall is waiting for this synchro to finish, then remove it from the other synchros in the waitany
516 * list. Afterwards, get the position of the actual synchro in the waitany list and return it as the result of the
519 if (simcall->call_ == actor::Simcall::Type::NONE) // FIXME: maybe a better way to handle this case
520 continue; // if actor handling comm is killed
522 handle_activity_waitany(simcall);
524 /* Check out for errors */
526 if (not simcall->issuer_->get_host()->is_on()) {
527 simcall->issuer_->set_wannadie();
529 // Do not answer to dying actors
530 if (not simcall->issuer_->wannadie()) {
531 set_exception(simcall->issuer_);
532 simcall->issuer_->simcall_answer();
536 simcall->issuer_->waiting_synchro_ = nullptr;
537 simcall->issuer_->activities_.erase(this);
539 if (simcall->issuer_ != dst_actor_ && dst_actor_ != nullptr)
540 dst_actor_->activities_.erase(this);
541 if (simcall->issuer_ != src_actor_ && src_actor_ != nullptr)
542 src_actor_->activities_.erase(this);
547 } // namespace simgrid::kernel::activity