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/actor/SimcallObserver.hpp"
14 #include "src/kernel/context/Context.hpp"
15 #include "src/kernel/resource/CpuImpl.hpp"
16 #include "src/kernel/resource/LinkImpl.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 /******************************************************************************/
23 /* SIMIX_comm_copy_data callbacks */
24 /******************************************************************************/
25 // XBT_ATTRIB_DEPRECATED_v333
26 void SIMIX_comm_set_copy_data_callback(void (*callback)(simgrid::kernel::activity::CommImpl*, void*, size_t))
28 simgrid::kernel::activity::CommImpl::set_copy_data_callback(callback);
31 // XBT_ATTRIB_DEPRECATED_v333
32 void SIMIX_comm_copy_buffer_callback(simgrid::kernel::activity::CommImpl* comm, void* buff, size_t buff_size)
34 simgrid::s4u::Comm::copy_buffer_callback(comm, buff, buff_size);
37 // XBT_ATTRIB_DEPRECATED_v333
38 void SIMIX_comm_copy_pointer_callback(simgrid::kernel::activity::CommImpl* comm, void* buff, size_t buff_size)
40 simgrid::s4u::Comm::copy_pointer_callback(comm, buff, buff_size);
46 xbt::signal<void(CommImpl const&)> CommImpl::on_start;
47 xbt::signal<void(CommImpl const&)> CommImpl::on_completion;
49 void (*CommImpl::copy_data_callback_)(CommImpl*, void*, size_t) = &s4u::Comm::copy_pointer_callback;
51 void CommImpl::set_copy_data_callback(void (*callback)(CommImpl*, void*, size_t))
53 copy_data_callback_ = callback;
56 CommImpl& CommImpl::set_type(CommImplType type)
62 CommImpl& CommImpl::set_size(double size)
68 CommImpl& CommImpl::set_rate(double rate)
73 CommImpl& CommImpl::set_mailbox(MailboxImpl* mbox)
76 mbox_id_ = mbox->get_id();
81 CommImpl& CommImpl::set_src_buff(unsigned char* buff, size_t size)
84 src_buff_size_ = size;
88 CommImpl& CommImpl::set_dst_buff(unsigned char* buff, size_t* size)
91 dst_buff_size_ = size;
95 CommImpl& CommImpl::detach()
101 CommImpl::CommImpl(s4u::Host* from, s4u::Host* to, double bytes) : size_(bytes), detached_(true), from_(from), to_(to)
103 set_state(State::READY);
106 CommImpl::~CommImpl()
108 XBT_DEBUG("Really free communication %p in state %s (detached = %d)", this, get_state_str(), detached_);
112 if (detached_ && get_state() != State::DONE) {
113 /* the communication has failed and was detached:
114 * we have to free the buffer */
116 clean_fun(src_buff_);
123 /** @brief Starts the simulation of a communication synchro. */
124 CommImpl* CommImpl::start()
126 /* If both the sender and the receiver are already there, start the communication */
127 if (get_state() == State::READY) {
128 from_ = from_ != nullptr ? from_ : src_actor_->get_host();
129 to_ = to_ != nullptr ? to_ : dst_actor_->get_host();
130 /* Getting the network_model from the origin host
131 * Valid while we have a single network model, otherwise we would need to change this function to first get the
132 * routes and later create the respective surf actions */
133 auto net_model = from_->get_netpoint()->get_englobing_zone()->get_network_model();
135 surf_action_ = net_model->communicate(from_, to_, size_, rate_);
136 surf_action_->set_activity(this);
137 surf_action_->set_category(get_tracing_category());
138 set_start_time(surf_action_->get_start_time());
139 set_state(State::RUNNING);
142 XBT_DEBUG("Starting communication %p from '%s' to '%s' (surf_action: %p; state: %s)", this, from_->get_cname(),
143 to_->get_cname(), surf_action_, get_state_str());
145 /* If a link is failed, detect it immediately */
146 if (surf_action_->get_state() == resource::Action::State::FAILED) {
147 XBT_DEBUG("Communication from '%s' to '%s' failed to start because of a link failure", from_->get_cname(),
149 set_state(State::LINK_FAILURE);
152 } else if ((src_actor_ != nullptr && src_actor_->is_suspended()) ||
153 (dst_actor_ != nullptr && dst_actor_->is_suspended())) {
154 /* If any of the actor is suspended, create the synchro but stop its execution,
155 it will be restarted when the sender actor resume */
156 if (src_actor_->is_suspended())
157 XBT_DEBUG("The communication is suspended on startup because src (%s@%s) was suspended since it initiated the "
159 src_actor_->get_cname(), src_actor_->get_host()->get_cname());
161 XBT_DEBUG("The communication is suspended on startup because dst (%s@%s) was suspended since it initiated the "
163 dst_actor_->get_cname(), dst_actor_->get_host()->get_cname());
165 surf_action_->suspend();
172 std::vector<s4u::Link*> CommImpl::get_traversed_links() const
174 xbt_assert(get_state() != State::WAITING, "You cannot use %s() if your communication is not ready (%s)", __FUNCTION__,
176 std::vector<s4u::Link*> vlinks;
177 XBT_ATTRIB_UNUSED double res = 0;
178 from_->route_to(to_, vlinks, &res);
182 /** @brief Copy the communication data from the sender's buffer to the receiver's one */
183 void CommImpl::copy_data()
185 size_t buff_size = src_buff_size_;
186 /* If there is no data to copy then return */
187 if (not src_buff_ || not dst_buff_ || copied_)
190 XBT_DEBUG("Copying comm %p data from %s (%p) -> %s (%p) (%zu bytes)", this,
191 src_actor_ ? src_actor_->get_host()->get_cname() : "a finished actor", src_buff_,
192 dst_actor_ ? dst_actor_->get_host()->get_cname() : "a finished actor", dst_buff_, buff_size);
194 /* Copy at most dst_buff_size bytes of the message to receiver's buffer */
195 if (dst_buff_size_) {
196 buff_size = std::min(buff_size, *dst_buff_size_);
198 /* Update the receiver's buffer size to the copied amount */
199 *dst_buff_size_ = buff_size;
204 copy_data_fun(this, src_buff_, buff_size);
206 copy_data_callback_(this, src_buff_, buff_size);
209 /* Set the copied flag so we copy data only once */
210 /* (this function might be called from both communication ends) */
214 ActivityImplPtr CommImpl::isend(actor::CommIsendSimcall* observer)
216 auto* mbox = observer->get_mailbox();
217 XBT_DEBUG("send from mailbox %p", mbox);
219 /* Prepare a synchro describing us, so that it gets passed to the user-provided filter of other side */
220 CommImplPtr this_comm(new CommImpl());
221 this_comm->set_type(CommImplType::SEND);
223 /* Look for communication synchro matching our needs. We also provide a description of
224 * ourself so that the other side also gets a chance of choosing if it wants to match with us.
226 * If it is not found then push our communication into the rendez-vous point */
227 CommImplPtr other_comm =
228 mbox->find_matching_comm(CommImplType::RECEIVE, observer->get_match_fun(), observer->get_payload(), this_comm,
229 /*done*/ false, /*remove_matching*/ true);
231 if (not other_comm) {
232 other_comm = std::move(this_comm);
234 if (mbox->is_permanent()) {
235 // this mailbox is for small messages, which have to be sent right now
236 other_comm->set_state(State::READY);
237 other_comm->dst_actor_ = mbox->get_permanent_receiver().get();
238 mbox->push_done(other_comm);
239 XBT_DEBUG("pushing a message into the permanent receive list %p, comm %p", mbox, other_comm.get());
242 mbox->push(other_comm);
245 XBT_DEBUG("Receive already pushed");
247 other_comm->set_state(State::READY);
249 observer->set_comm(other_comm.get());
251 if (observer->is_detached()) {
252 other_comm->detach();
253 other_comm->clean_fun = observer->get_clean_fun();
255 other_comm->clean_fun = nullptr;
256 observer->get_issuer()->activities_.emplace_back(other_comm);
259 /* Setup the communication synchro */
260 other_comm->src_actor_ = observer->get_issuer();
261 other_comm->src_data_ = observer->get_payload();
263 .set_src_buff(observer->get_src_buff(), observer->get_src_buff_size())
264 .set_size(observer->get_payload_size())
265 .set_rate(observer->get_rate());
267 other_comm->match_fun = observer->get_match_fun();
268 other_comm->copy_data_fun = observer->get_copy_data_fun();
270 if (MC_is_active() || MC_record_replay_is_active())
271 other_comm->set_state(simgrid::kernel::activity::State::RUNNING);
275 return (observer->is_detached() ? nullptr : other_comm);
278 ActivityImplPtr CommImpl::irecv(actor::CommIrecvSimcall* observer)
280 CommImplPtr this_synchro(new CommImpl());
281 this_synchro->set_type(CommImplType::RECEIVE);
283 auto* mbox = observer->get_mailbox();
284 XBT_DEBUG("recv from mbox %p. this_synchro=%p", mbox, this_synchro.get());
286 CommImplPtr other_comm;
287 // communication already done, get it inside the list of completed comms
288 if (mbox->is_permanent() && mbox->has_some_done_comm()) {
289 XBT_DEBUG("We have a comm that has probably already been received, trying to match it, to skip the communication");
290 // find a match in the list of already received comms
291 other_comm = mbox->find_matching_comm(CommImplType::SEND, observer->get_match_fun(), observer->get_payload(),
292 this_synchro, /*done*/ true, /*remove_matching*/ true);
293 // if not found, assume the receiver came first, register it to the mailbox in the classical way
294 if (not other_comm) {
295 XBT_DEBUG("We have messages in the permanent receive list, but not the one we are looking for, pushing request "
297 other_comm = std::move(this_synchro);
298 mbox->push(other_comm);
300 if (other_comm->surf_action_ && other_comm->get_remaining() < 1e-12) {
301 XBT_DEBUG("comm %p has been already sent, and is finished, destroy it", other_comm.get());
302 other_comm->set_state(State::DONE);
303 other_comm->set_mailbox(nullptr);
307 /* Prepare a comm describing us, so that it gets passed to the user-provided filter of other side */
309 /* Look for communication activity matching our needs. We also provide a description of
310 * ourself so that the other side also gets a chance of choosing if it wants to match with us.
312 * If it is not found then push our communication into the rendez-vous point */
313 other_comm = mbox->find_matching_comm(CommImplType::SEND, observer->get_match_fun(), observer->get_payload(),
314 this_synchro, /*done*/ false, /*remove_matching*/ true);
316 if (other_comm == nullptr) {
317 XBT_DEBUG("Receive pushed first (%zu comm enqueued so far)", mbox->size());
318 other_comm = std::move(this_synchro);
319 mbox->push(other_comm);
321 XBT_DEBUG("Match my %p with the existing %p", this_synchro.get(), other_comm.get());
323 other_comm->set_state(simgrid::kernel::activity::State::READY);
325 observer->get_issuer()->activities_.emplace_back(other_comm);
327 observer->set_comm(other_comm.get());
329 /* Setup communication synchro */
330 other_comm->dst_actor_ = observer->get_issuer();
331 other_comm->dst_data_ = observer->get_payload();
332 other_comm->set_dst_buff(observer->get_dst_buff(), observer->get_dst_buff_size());
334 if (observer->get_rate() > -1.0 && (other_comm->get_rate() < 0.0 || observer->get_rate() < other_comm->get_rate()))
335 other_comm->set_rate(observer->get_rate());
337 other_comm->match_fun = observer->get_match_fun();
338 other_comm->copy_data_fun = observer->get_copy_data_fun();
340 if (MC_is_active() || MC_record_replay_is_active()) {
341 other_comm->set_state(State::RUNNING);
349 bool CommImpl::test(actor::ActorImpl* issuer)
351 if ((MC_is_active() || MC_record_replay_is_active()) && src_actor_ && dst_actor_)
352 set_state(State::DONE);
353 return ActivityImpl::test(issuer);
356 void CommImpl::wait_for(actor::ActorImpl* issuer, double timeout)
358 XBT_DEBUG("CommImpl::wait_for(%g), %p, state %s", timeout, this, get_state_str());
360 /* Associate this simcall to the wait synchro */
361 register_simcall(&issuer->simcall_);
362 if (MC_is_active() || MC_record_replay_is_active()) {
363 // FIXME: what about timeouts?
364 set_state(State::DONE);
368 ActivityImpl::wait_for(issuer, timeout);
371 void CommImpl::wait_any_for(actor::ActorImpl* issuer, const std::vector<CommImpl*>& comms, double timeout)
373 std::vector<ActivityImpl*> activities(comms.begin(), comms.end());
374 ActivityImpl::wait_any_for(issuer, activities, timeout);
377 void CommImpl::suspend()
379 /* FIXME: shall we suspend also the timeout synchro? */
381 surf_action_->suspend();
382 /* if not created yet, the action will be suspended on creation, in CommImpl::start() */
385 void CommImpl::resume()
387 /*FIXME: check what happen with the timeouts */
389 surf_action_->resume();
390 /* in the other case, the synchro were not really suspended yet, see CommImpl::suspend() and CommImpl::start() */
393 void CommImpl::cancel()
395 /* if the synchro is a waiting state means that it is still in a mbox so remove from it and delete it */
396 if (get_state() == State::WAITING) {
399 set_state(State::CANCELED);
401 } else if (not MC_is_active() /* when running the MC there are no surf actions */
402 && not MC_record_replay_is_active() && (get_state() == State::READY || get_state() == State::RUNNING)) {
403 surf_action_->cancel();
407 /** @brief This is part of the cleanup process, probably an internal command */
408 void CommImpl::cleanup_surf()
413 src_timeout_->unref();
414 src_timeout_ = nullptr;
418 dst_timeout_->unref();
419 dst_timeout_ = nullptr;
423 void CommImpl::post()
425 on_completion(*this);
427 /* Update synchro state */
428 if (src_timeout_ && src_timeout_->get_state() == resource::Action::State::FINISHED)
429 set_state(State::SRC_TIMEOUT);
430 else if (dst_timeout_ && dst_timeout_->get_state() == resource::Action::State::FINISHED)
431 set_state(State::DST_TIMEOUT);
432 else if (src_timeout_ && src_timeout_->get_state() == resource::Action::State::FAILED)
433 set_state(State::SRC_HOST_FAILURE);
434 else if (dst_timeout_ && dst_timeout_->get_state() == resource::Action::State::FAILED)
435 set_state(State::DST_HOST_FAILURE);
436 else if (surf_action_ && surf_action_->get_state() == resource::Action::State::FAILED) {
437 set_state(State::LINK_FAILURE);
439 set_state(State::DONE);
441 XBT_DEBUG("CommImpl::post(): comm %p, state %s, src_proc %p, dst_proc %p, detached: %d", this, get_state_str(),
442 src_actor_.get(), dst_actor_.get(), detached_);
444 /* destroy the surf actions associated with the Simix communication */
447 /* Answer all simcalls associated with the synchro */
450 void CommImpl::set_exception(actor::ActorImpl* issuer)
452 switch (get_state()) {
454 issuer->exception_ = std::make_exception_ptr(NetworkFailureException(XBT_THROW_POINT, "Remote peer failed"));
456 case State::SRC_TIMEOUT:
458 std::make_exception_ptr(TimeoutException(XBT_THROW_POINT, "Communication timeouted because of the sender"));
461 case State::DST_TIMEOUT:
463 std::make_exception_ptr(TimeoutException(XBT_THROW_POINT, "Communication timeouted because of the receiver"));
466 case State::SRC_HOST_FAILURE:
467 if (issuer == src_actor_)
468 issuer->context_->set_wannadie();
470 set_state(State::FAILED);
471 issuer->exception_ = std::make_exception_ptr(NetworkFailureException(XBT_THROW_POINT, "Remote peer failed"));
475 case State::DST_HOST_FAILURE:
476 if (issuer == dst_actor_)
477 issuer->context_->set_wannadie();
479 set_state(State::FAILED);
480 issuer->exception_ = std::make_exception_ptr(NetworkFailureException(XBT_THROW_POINT, "Remote peer failed"));
484 case State::LINK_FAILURE:
485 XBT_DEBUG("Link failure in synchro %p between '%s' and '%s': posting an exception to the issuer: %s (%p) "
487 this, src_actor_ ? src_actor_->get_host()->get_cname() : nullptr,
488 dst_actor_ ? dst_actor_->get_host()->get_cname() : nullptr, issuer->get_cname(), issuer, detached_);
489 if (src_actor_ == issuer) {
490 XBT_DEBUG("I'm source");
491 } else if (dst_actor_ == issuer) {
492 XBT_DEBUG("I'm dest");
494 XBT_DEBUG("I'm neither source nor dest");
496 set_state(State::FAILED);
497 issuer->throw_exception(std::make_exception_ptr(NetworkFailureException(XBT_THROW_POINT, "Link failure")));
500 case State::CANCELED:
501 if (issuer == dst_actor_)
503 std::make_exception_ptr(CancelException(XBT_THROW_POINT, "Communication canceled by the sender"));
506 std::make_exception_ptr(CancelException(XBT_THROW_POINT, "Communication canceled by the receiver"));
510 xbt_assert(get_state() == State::DONE, "Internal error in CommImpl::finish(): unexpected synchro state %s",
515 void CommImpl::finish()
517 XBT_DEBUG("CommImpl::finish() in state %s", get_state_str());
518 /* If the synchro is still in a rendez-vous point then remove from it */
522 if (get_state() == State::DONE)
525 while (not simcalls_.empty()) {
526 smx_simcall_t simcall = simcalls_.front();
527 simcalls_.pop_front();
529 /* If a waitany simcall is waiting for this synchro to finish, then remove it from the other synchros in the waitany
530 * list. Afterwards, get the position of the actual synchro in the waitany list and return it as the result of the
533 if (simcall->call_ == simix::Simcall::NONE) // FIXME: maybe a better way to handle this case
534 continue; // if actor handling comm is killed
536 handle_activity_waitany(simcall);
538 /* Check out for errors */
540 if (not simcall->issuer_->get_host()->is_on()) {
541 simcall->issuer_->context_->set_wannadie();
543 // Do not answer to dying actors
544 if (not simcall->issuer_->context_->wannadie()) {
545 set_exception(simcall->issuer_);
546 simcall->issuer_->simcall_answer();
550 simcall->issuer_->waiting_synchro_ = nullptr;
551 simcall->issuer_->activities_.remove(this);
553 if (simcall->issuer_ != dst_actor_ && dst_actor_ != nullptr)
554 dst_actor_->activities_.remove(this);
555 if (simcall->issuer_ != src_actor_ && src_actor_ != nullptr)
556 src_actor_->activities_.remove(this);
561 } // namespace activity
562 } // namespace kernel
563 } // namespace simgrid