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 XBT_PRIVATE void simcall_HANDLER_comm_send(smx_simcall_t simcall, smx_actor_t src, smx_mailbox_t mbox, double task_size,
23 double rate, unsigned char* src_buff, size_t src_buff_size,
24 bool (*match_fun)(void*, void*, simgrid::kernel::activity::CommImpl*),
25 void (*copy_data_fun)(simgrid::kernel::activity::CommImpl*, void*, size_t),
26 void* data, double timeout)
28 simgrid::kernel::actor::CommIsendSimcall observer(src, mbox, task_size, rate, src_buff, src_buff_size, match_fun,
29 nullptr, copy_data_fun, data, false);
30 simgrid::kernel::activity::ActivityImplPtr comm = simgrid::kernel::activity::CommImpl::isend(&observer);
31 simcall->mc_value_ = 0;
32 comm->wait_for(simcall->issuer_, timeout);
35 XBT_PRIVATE simgrid::kernel::activity::ActivityImplPtr simcall_HANDLER_comm_isend(
36 smx_simcall_t /*simcall*/, smx_actor_t src_proc, smx_mailbox_t mbox, double task_size, double rate,
37 unsigned char* src_buff, size_t src_buff_size,
38 bool (*match_fun)(void*, void*, simgrid::kernel::activity::CommImpl*),
39 void (*clean_fun)(void*), // used to free the synchro in case of problem after a detached send
40 void (*copy_data_fun)(simgrid::kernel::activity::CommImpl*, void*, size_t), // used to copy data if not default one
41 void* data, bool detached)
43 simgrid::kernel::actor::CommIsendSimcall observer(src_proc, mbox, task_size, rate, src_buff, src_buff_size, match_fun,
44 clean_fun, copy_data_fun, data, detached);
45 return simgrid::kernel::activity::CommImpl::isend(&observer);
48 XBT_PRIVATE void simcall_HANDLER_comm_recv(smx_simcall_t simcall, smx_actor_t receiver, smx_mailbox_t mbox,
49 unsigned char* dst_buff, size_t* dst_buff_size,
50 bool (*match_fun)(void*, void*, simgrid::kernel::activity::CommImpl*),
51 void (*copy_data_fun)(simgrid::kernel::activity::CommImpl*, void*, size_t),
52 void* data, double timeout, double rate)
54 simgrid::kernel::actor::CommIrecvSimcall observer(receiver, mbox, dst_buff, dst_buff_size, match_fun, copy_data_fun,
56 simgrid::kernel::activity::ActivityImplPtr comm = simgrid::kernel::activity::CommImpl::irecv(&observer);
57 simcall->mc_value_ = 0;
58 comm->wait_for(simcall->issuer_, timeout);
61 XBT_PRIVATE simgrid::kernel::activity::ActivityImplPtr
62 simcall_HANDLER_comm_irecv(smx_simcall_t /*simcall*/, smx_actor_t receiver, smx_mailbox_t mbox, unsigned char* dst_buff,
63 size_t* dst_buff_size, bool (*match_fun)(void*, void*, simgrid::kernel::activity::CommImpl*),
64 void (*copy_data_fun)(simgrid::kernel::activity::CommImpl*, void*, size_t), void* data,
67 simgrid::kernel::actor::CommIrecvSimcall observer(receiver, mbox, dst_buff, dst_buff_size, match_fun, copy_data_fun,
69 return simgrid::kernel::activity::CommImpl::irecv(&observer);
72 void simcall_HANDLER_comm_wait(smx_simcall_t simcall, simgrid::kernel::activity::CommImpl* comm, double timeout)
74 comm->wait_for(simcall->issuer_, timeout);
77 void simcall_HANDLER_comm_waitany(smx_simcall_t simcall, simgrid::kernel::activity::CommImpl* comms[], size_t count,
80 std::vector<simgrid::kernel::activity::CommImpl*> comms_vec(comms, comms + count);
81 simgrid::kernel::activity::CommImpl::wait_any_for(simcall->issuer_, comms_vec, timeout);
84 /******************************************************************************/
85 /* SIMIX_comm_copy_data callbacks */
86 /******************************************************************************/
87 // XBT_ATTRIB_DEPRECATED_v333
88 void SIMIX_comm_set_copy_data_callback(void (*callback)(simgrid::kernel::activity::CommImpl*, void*, size_t))
90 simgrid::kernel::activity::CommImpl::set_copy_data_callback(callback);
93 // XBT_ATTRIB_DEPRECATED_v333
94 void SIMIX_comm_copy_buffer_callback(simgrid::kernel::activity::CommImpl* comm, void* buff, size_t buff_size)
96 simgrid::s4u::Comm::copy_buffer_callback(comm, buff, buff_size);
99 // XBT_ATTRIB_DEPRECATED_v333
100 void SIMIX_comm_copy_pointer_callback(simgrid::kernel::activity::CommImpl* comm, void* buff, size_t buff_size)
102 simgrid::s4u::Comm::copy_pointer_callback(comm, buff, buff_size);
108 xbt::signal<void(CommImpl const&)> CommImpl::on_start;
109 xbt::signal<void(CommImpl const&)> CommImpl::on_completion;
111 void (*CommImpl::copy_data_callback_)(CommImpl*, void*, size_t) = &s4u::Comm::copy_pointer_callback;
113 void CommImpl::set_copy_data_callback(void (*callback)(CommImpl*, void*, size_t))
115 copy_data_callback_ = callback;
118 CommImpl& CommImpl::set_size(double size)
124 CommImpl& CommImpl::set_rate(double rate)
129 CommImpl& CommImpl::set_mailbox(MailboxImpl* mbox)
132 mbox_id_ = mbox->get_id();
137 CommImpl& CommImpl::set_src_buff(unsigned char* buff, size_t size)
140 src_buff_size_ = size;
144 CommImpl& CommImpl::set_dst_buff(unsigned char* buff, size_t* size)
147 dst_buff_size_ = size;
151 CommImpl& CommImpl::detach()
157 CommImpl::CommImpl(s4u::Host* from, s4u::Host* to, double bytes) : size_(bytes), detached_(true), from_(from), to_(to)
159 set_state(State::READY);
162 CommImpl::~CommImpl()
164 XBT_DEBUG("Really free communication %p in state %s (detached = %d)", this, get_state_str(), detached_);
168 if (detached_ && get_state() != State::DONE) {
169 /* the communication has failed and was detached:
170 * we have to free the buffer */
172 clean_fun(src_buff_);
179 /** @brief Starts the simulation of a communication synchro. */
180 CommImpl* CommImpl::start()
182 /* If both the sender and the receiver are already there, start the communication */
183 if (get_state() == State::READY) {
184 from_ = from_ != nullptr ? from_ : src_actor_->get_host();
185 to_ = to_ != nullptr ? to_ : dst_actor_->get_host();
186 /* Getting the network_model from the origin host
187 * Valid while we have a single network model, otherwise we would need to change this function to first get the
188 * routes and later create the respective surf actions */
189 auto net_model = from_->get_netpoint()->get_englobing_zone()->get_network_model();
191 surf_action_ = net_model->communicate(from_, to_, size_, rate_);
192 surf_action_->set_activity(this);
193 surf_action_->set_category(get_tracing_category());
194 set_start_time(surf_action_->get_start_time());
195 set_state(State::RUNNING);
198 XBT_DEBUG("Starting communication %p from '%s' to '%s' (surf_action: %p; state: %s)", this, from_->get_cname(),
199 to_->get_cname(), surf_action_, get_state_str());
201 /* If a link is failed, detect it immediately */
202 if (surf_action_->get_state() == resource::Action::State::FAILED) {
203 XBT_DEBUG("Communication from '%s' to '%s' failed to start because of a link failure", from_->get_cname(),
205 set_state(State::LINK_FAILURE);
208 } else if ((src_actor_ != nullptr && src_actor_->is_suspended()) ||
209 (dst_actor_ != nullptr && dst_actor_->is_suspended())) {
210 /* If any of the actor is suspended, create the synchro but stop its execution,
211 it will be restarted when the sender actor resume */
212 if (src_actor_->is_suspended())
213 XBT_DEBUG("The communication is suspended on startup because src (%s@%s) was suspended since it initiated the "
215 src_actor_->get_cname(), src_actor_->get_host()->get_cname());
217 XBT_DEBUG("The communication is suspended on startup because dst (%s@%s) was suspended since it initiated the "
219 dst_actor_->get_cname(), dst_actor_->get_host()->get_cname());
221 surf_action_->suspend();
228 std::vector<s4u::Link*> CommImpl::get_traversed_links() const
230 xbt_assert(get_state() != State::WAITING, "You cannot use %s() if your communication is not ready (%s)", __FUNCTION__,
232 std::vector<s4u::Link*> vlinks;
233 XBT_ATTRIB_UNUSED double res = 0;
234 from_->route_to(to_, vlinks, &res);
238 /** @brief Copy the communication data from the sender's buffer to the receiver's one */
239 void CommImpl::copy_data()
241 size_t buff_size = src_buff_size_;
242 /* If there is no data to copy then return */
243 if (not src_buff_ || not dst_buff_ || copied_)
246 XBT_DEBUG("Copying comm %p data from %s (%p) -> %s (%p) (%zu bytes)", this,
247 src_actor_ ? src_actor_->get_host()->get_cname() : "a finished actor", src_buff_,
248 dst_actor_ ? dst_actor_->get_host()->get_cname() : "a finished actor", dst_buff_, buff_size);
250 /* Copy at most dst_buff_size bytes of the message to receiver's buffer */
251 if (dst_buff_size_) {
252 buff_size = std::min(buff_size, *dst_buff_size_);
254 /* Update the receiver's buffer size to the copied amount */
255 *dst_buff_size_ = buff_size;
260 copy_data_fun(this, src_buff_, buff_size);
262 copy_data_callback_(this, src_buff_, buff_size);
265 /* Set the copied flag so we copy data only once */
266 /* (this function might be called from both communication ends) */
270 ActivityImplPtr CommImpl::isend(actor::CommIsendSimcall* observer)
272 auto* mbox = observer->get_mailbox();
273 XBT_DEBUG("send from mailbox %p", mbox);
275 /* Prepare a synchro describing us, so that it gets passed to the user-provided filter of other side */
276 CommImplPtr this_comm(new CommImpl(CommImpl::Type::SEND));
278 /* Look for communication synchro matching our needs. We also provide a description of
279 * ourself so that the other side also gets a chance of choosing if it wants to match with us.
281 * If it is not found then push our communication into the rendez-vous point */
282 CommImplPtr other_comm =
283 mbox->find_matching_comm(CommImpl::Type::RECEIVE, observer->get_match_fun(), observer->get_payload(), this_comm,
284 /*done*/ false, /*remove_matching*/ true);
286 if (not other_comm) {
287 other_comm = std::move(this_comm);
289 if (mbox->is_permanent()) {
290 // this mailbox is for small messages, which have to be sent right now
291 other_comm->set_state(State::READY);
292 other_comm->dst_actor_ = mbox->get_permanent_receiver().get();
293 mbox->push_done(other_comm);
294 XBT_DEBUG("pushing a message into the permanent receive list %p, comm %p", mbox, other_comm.get());
297 mbox->push(other_comm);
300 XBT_DEBUG("Receive already pushed");
302 other_comm->set_state(State::READY);
304 observer->set_comm(other_comm.get());
306 if (observer->is_detached()) {
307 other_comm->detach();
308 other_comm->clean_fun = observer->get_clean_fun();
310 other_comm->clean_fun = nullptr;
311 observer->get_issuer()->activities_.emplace_back(other_comm);
314 /* Setup the communication synchro */
315 other_comm->src_actor_ = observer->get_issuer();
316 other_comm->src_data_ = observer->get_payload();
318 .set_src_buff(observer->get_src_buff(), observer->get_src_buff_size())
319 .set_size(observer->get_payload_size())
320 .set_rate(observer->get_rate());
322 other_comm->match_fun = observer->get_match_fun();
323 other_comm->copy_data_fun = observer->get_copy_data_fun();
325 if (MC_is_active() || MC_record_replay_is_active())
326 other_comm->set_state(simgrid::kernel::activity::State::RUNNING);
330 return (observer->is_detached() ? nullptr : other_comm);
333 ActivityImplPtr CommImpl::irecv(actor::CommIrecvSimcall* observer)
335 CommImplPtr this_synchro(new CommImpl(CommImpl::Type::RECEIVE));
336 auto* mbox = observer->get_mailbox();
337 XBT_DEBUG("recv from mbox %p. this_synchro=%p", mbox, this_synchro.get());
339 CommImplPtr other_comm;
340 // communication already done, get it inside the list of completed comms
341 if (mbox->is_permanent() && mbox->has_some_done_comm()) {
342 XBT_DEBUG("We have a comm that has probably already been received, trying to match it, to skip the communication");
343 // find a match in the list of already received comms
344 other_comm = mbox->find_matching_comm(CommImpl::Type::SEND, observer->get_match_fun(), observer->get_payload(),
345 this_synchro, /*done*/ true, /*remove_matching*/ true);
346 // if not found, assume the receiver came first, register it to the mailbox in the classical way
347 if (not other_comm) {
348 XBT_DEBUG("We have messages in the permanent receive list, but not the one we are looking for, pushing request "
350 other_comm = std::move(this_synchro);
351 mbox->push(other_comm);
353 if (other_comm->surf_action_ && other_comm->get_remaining() < 1e-12) {
354 XBT_DEBUG("comm %p has been already sent, and is finished, destroy it", other_comm.get());
355 other_comm->set_state(State::DONE);
356 other_comm->set_mailbox(nullptr);
360 /* Prepare a comm describing us, so that it gets passed to the user-provided filter of other side */
362 /* Look for communication activity matching our needs. We also provide a description of
363 * ourself so that the other side also gets a chance of choosing if it wants to match with us.
365 * If it is not found then push our communication into the rendez-vous point */
366 other_comm = mbox->find_matching_comm(CommImpl::Type::SEND, observer->get_match_fun(), observer->get_payload(),
367 this_synchro, /*done*/ false, /*remove_matching*/ true);
369 if (other_comm == nullptr) {
370 XBT_DEBUG("Receive pushed first (%zu comm enqueued so far)", mbox->size());
371 other_comm = std::move(this_synchro);
372 mbox->push(other_comm);
374 XBT_DEBUG("Match my %p with the existing %p", this_synchro.get(), other_comm.get());
376 other_comm->set_state(simgrid::kernel::activity::State::READY);
378 observer->get_issuer()->activities_.emplace_back(other_comm);
380 observer->set_comm(other_comm.get());
382 /* Setup communication synchro */
383 other_comm->dst_actor_ = observer->get_issuer();
384 other_comm->dst_data_ = observer->get_payload();
385 other_comm->set_dst_buff(observer->get_dst_buff(), observer->get_dst_buff_size());
387 if (observer->get_rate() > -1.0 && (other_comm->get_rate() < 0.0 || observer->get_rate() < other_comm->get_rate()))
388 other_comm->set_rate(observer->get_rate());
390 other_comm->match_fun = observer->get_match_fun();
391 other_comm->copy_data_fun = observer->get_copy_data_fun();
393 if (MC_is_active() || MC_record_replay_is_active()) {
394 other_comm->set_state(State::RUNNING);
402 bool CommImpl::test(actor::ActorImpl* issuer)
404 if ((MC_is_active() || MC_record_replay_is_active()) && src_actor_ && dst_actor_)
405 set_state(State::DONE);
406 return ActivityImpl::test(issuer);
409 void CommImpl::wait_for(actor::ActorImpl* issuer, double timeout)
411 XBT_DEBUG("CommImpl::wait_for(%g), %p, state %s", timeout, this, get_state_str());
413 /* Associate this simcall to the wait synchro */
414 register_simcall(&issuer->simcall_);
415 if (MC_is_active() || MC_record_replay_is_active()) {
416 int idx = issuer->simcall_.mc_value_;
418 set_state(State::DONE);
420 /* If we reached this point, the wait simcall must have a timeout */
421 /* Otherwise it shouldn't be enabled and executed by the MC */
422 xbt_assert(timeout >= 0.0,
423 "The checker asked me to raise a timeout on a communication that is not expecting any timeout");
424 set_state(issuer == src_actor_ ? State::SRC_TIMEOUT : State::DST_TIMEOUT);
429 ActivityImpl::wait_for(issuer, timeout);
432 void CommImpl::wait_any_for(actor::ActorImpl* issuer, const std::vector<CommImpl*>& comms, double timeout)
434 std::vector<ActivityImpl*> activities(comms.begin(), comms.end());
435 ActivityImpl::wait_any_for(issuer, activities, timeout);
438 void CommImpl::suspend()
440 /* FIXME: shall we suspend also the timeout synchro? */
442 surf_action_->suspend();
443 /* if not created yet, the action will be suspended on creation, in CommImpl::start() */
446 void CommImpl::resume()
448 /*FIXME: check what happen with the timeouts */
450 surf_action_->resume();
451 /* in the other case, the synchro were not really suspended yet, see CommImpl::suspend() and CommImpl::start() */
454 void CommImpl::cancel()
456 /* if the synchro is a waiting state means that it is still in a mbox so remove from it and delete it */
457 if (get_state() == State::WAITING) {
460 set_state(State::CANCELED);
462 } else if (not MC_is_active() /* when running the MC there are no surf actions */
463 && not MC_record_replay_is_active() && (get_state() == State::READY || get_state() == State::RUNNING)) {
464 surf_action_->cancel();
468 /** @brief This is part of the cleanup process, probably an internal command */
469 void CommImpl::cleanup_surf()
474 src_timeout_->unref();
475 src_timeout_ = nullptr;
479 dst_timeout_->unref();
480 dst_timeout_ = nullptr;
484 void CommImpl::post()
486 on_completion(*this);
488 /* Update synchro state */
489 if (src_timeout_ && src_timeout_->get_state() == resource::Action::State::FINISHED)
490 set_state(State::SRC_TIMEOUT);
491 else if (dst_timeout_ && dst_timeout_->get_state() == resource::Action::State::FINISHED)
492 set_state(State::DST_TIMEOUT);
493 else if (src_timeout_ && src_timeout_->get_state() == resource::Action::State::FAILED)
494 set_state(State::SRC_HOST_FAILURE);
495 else if (dst_timeout_ && dst_timeout_->get_state() == resource::Action::State::FAILED)
496 set_state(State::DST_HOST_FAILURE);
497 else if (surf_action_ && surf_action_->get_state() == resource::Action::State::FAILED) {
498 set_state(State::LINK_FAILURE);
500 set_state(State::DONE);
502 XBT_DEBUG("CommImpl::post(): comm %p, state %s, src_proc %p, dst_proc %p, detached: %d", this, get_state_str(),
503 src_actor_.get(), dst_actor_.get(), detached_);
505 /* destroy the surf actions associated with the Simix communication */
508 /* Answer all simcalls associated with the synchro */
511 void CommImpl::set_exception(actor::ActorImpl* issuer)
513 switch (get_state()) {
515 issuer->exception_ = std::make_exception_ptr(NetworkFailureException(XBT_THROW_POINT, "Remote peer failed"));
517 case State::SRC_TIMEOUT:
519 std::make_exception_ptr(TimeoutException(XBT_THROW_POINT, "Communication timeouted because of the sender"));
522 case State::DST_TIMEOUT:
524 std::make_exception_ptr(TimeoutException(XBT_THROW_POINT, "Communication timeouted because of the receiver"));
527 case State::SRC_HOST_FAILURE:
528 if (issuer == src_actor_)
529 issuer->context_->set_wannadie();
531 set_state(State::FAILED);
532 issuer->exception_ = std::make_exception_ptr(NetworkFailureException(XBT_THROW_POINT, "Remote peer failed"));
536 case State::DST_HOST_FAILURE:
537 if (issuer == dst_actor_)
538 issuer->context_->set_wannadie();
540 set_state(State::FAILED);
541 issuer->exception_ = std::make_exception_ptr(NetworkFailureException(XBT_THROW_POINT, "Remote peer failed"));
545 case State::LINK_FAILURE:
546 XBT_DEBUG("Link failure in synchro %p between '%s' and '%s': posting an exception to the issuer: %s (%p) "
548 this, src_actor_ ? src_actor_->get_host()->get_cname() : nullptr,
549 dst_actor_ ? dst_actor_->get_host()->get_cname() : nullptr, issuer->get_cname(), issuer, detached_);
550 if (src_actor_ == issuer) {
551 XBT_DEBUG("I'm source");
552 } else if (dst_actor_ == issuer) {
553 XBT_DEBUG("I'm dest");
555 XBT_DEBUG("I'm neither source nor dest");
557 set_state(State::FAILED);
558 issuer->throw_exception(std::make_exception_ptr(NetworkFailureException(XBT_THROW_POINT, "Link failure")));
561 case State::CANCELED:
562 if (issuer == dst_actor_)
564 std::make_exception_ptr(CancelException(XBT_THROW_POINT, "Communication canceled by the sender"));
567 std::make_exception_ptr(CancelException(XBT_THROW_POINT, "Communication canceled by the receiver"));
571 xbt_assert(get_state() == State::DONE, "Internal error in CommImpl::finish(): unexpected synchro state %s",
576 void CommImpl::finish()
578 XBT_DEBUG("CommImpl::finish() in state %s", get_state_str());
579 /* If the synchro is still in a rendez-vous point then remove from it */
583 if (get_state() == State::DONE)
586 while (not simcalls_.empty()) {
587 smx_simcall_t simcall = simcalls_.front();
588 simcalls_.pop_front();
590 /* If a waitany simcall is waiting for this synchro to finish, then remove it from the other synchros in the waitany
591 * list. Afterwards, get the position of the actual synchro in the waitany list and return it as the result of the
594 if (simcall->call_ == simix::Simcall::NONE) // FIXME: maybe a better way to handle this case
595 continue; // if actor handling comm is killed
597 handle_activity_waitany(simcall);
599 /* Check out for errors */
601 if (not simcall->issuer_->get_host()->is_on()) {
602 simcall->issuer_->context_->set_wannadie();
604 // Do not answer to dying actors
605 if (not simcall->issuer_->context_->wannadie()) {
606 set_exception(simcall->issuer_);
607 simcall->issuer_->simcall_answer();
611 simcall->issuer_->waiting_synchro_ = nullptr;
612 simcall->issuer_->activities_.remove(this);
614 if (simcall->issuer_ != dst_actor_ && dst_actor_ != nullptr)
615 dst_actor_->activities_.remove(this);
616 if (simcall->issuer_ != src_actor_ && src_actor_ != nullptr)
617 src_actor_->activities_.remove(this);
622 } // namespace activity
623 } // namespace kernel
624 } // namespace simgrid