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::activity::ActivityImplPtr comm = simgrid::kernel::activity::CommImpl::isend(
29 src, mbox, task_size, rate, src_buff, src_buff_size, match_fun, nullptr, copy_data_fun, data, false);
30 simcall->mc_value_ = 0;
31 comm->wait_for(simcall->issuer_, timeout);
34 XBT_PRIVATE simgrid::kernel::activity::ActivityImplPtr simcall_HANDLER_comm_isend(
35 smx_simcall_t /*simcall*/, smx_actor_t src_proc, smx_mailbox_t mbox, double task_size, double rate,
36 unsigned char* src_buff, size_t src_buff_size,
37 bool (*match_fun)(void*, void*, simgrid::kernel::activity::CommImpl*),
38 void (*clean_fun)(void*), // used to free the synchro in case of problem after a detached send
39 void (*copy_data_fun)(simgrid::kernel::activity::CommImpl*, void*, size_t), // used to copy data if not default one
40 void* data, bool detached)
42 return simgrid::kernel::activity::CommImpl::isend(src_proc, mbox, task_size, rate, src_buff, src_buff_size, match_fun,
43 clean_fun, copy_data_fun, data, detached);
46 XBT_PRIVATE void simcall_HANDLER_comm_recv(smx_simcall_t simcall, smx_actor_t receiver, smx_mailbox_t mbox,
47 unsigned char* dst_buff, size_t* dst_buff_size,
48 bool (*match_fun)(void*, void*, simgrid::kernel::activity::CommImpl*),
49 void (*copy_data_fun)(simgrid::kernel::activity::CommImpl*, void*, size_t),
50 void* data, double timeout, double rate)
52 simgrid::kernel::activity::ActivityImplPtr comm = simgrid::kernel::activity::CommImpl::irecv(
53 receiver, mbox, dst_buff, dst_buff_size, match_fun, copy_data_fun, data, rate);
54 simcall->mc_value_ = 0;
55 comm->wait_for(simcall->issuer_, timeout);
58 XBT_PRIVATE simgrid::kernel::activity::ActivityImplPtr
59 simcall_HANDLER_comm_irecv(smx_simcall_t /*simcall*/, smx_actor_t receiver, smx_mailbox_t mbox, unsigned char* dst_buff,
60 size_t* dst_buff_size, bool (*match_fun)(void*, void*, simgrid::kernel::activity::CommImpl*),
61 void (*copy_data_fun)(simgrid::kernel::activity::CommImpl*, void*, size_t), void* data,
64 return simgrid::kernel::activity::CommImpl::irecv(receiver, mbox, dst_buff, dst_buff_size, match_fun, copy_data_fun,
68 void simcall_HANDLER_comm_wait(smx_simcall_t simcall, simgrid::kernel::activity::CommImpl* comm, double timeout)
70 comm->wait_for(simcall->issuer_, timeout);
73 bool simcall_HANDLER_comm_test(smx_simcall_t simcall, simgrid::kernel::activity::CommImpl* comm)
75 return comm->test(simcall->issuer_);
78 ssize_t simcall_HANDLER_comm_testany(smx_simcall_t simcall, simgrid::kernel::activity::CommImpl* comms[], size_t count)
80 std::vector<simgrid::kernel::activity::ActivityImpl*> comms_vec(comms, comms + count);
81 return simgrid::kernel::activity::ActivityImpl::test_any(simcall->issuer_, comms_vec);
84 void simcall_HANDLER_comm_waitany(smx_simcall_t simcall, simgrid::kernel::activity::CommImpl* comms[], size_t count,
87 std::vector<simgrid::kernel::activity::CommImpl*> comms_vec(comms, comms + count);
88 simgrid::kernel::activity::CommImpl::wait_any_for(simcall->issuer_, comms_vec, timeout);
91 /******************************************************************************/
92 /* SIMIX_comm_copy_data callbacks */
93 /******************************************************************************/
94 // XBT_ATTRIB_DEPRECATED_v333
95 void SIMIX_comm_set_copy_data_callback(void (*callback)(simgrid::kernel::activity::CommImpl*, void*, size_t))
97 simgrid::kernel::activity::CommImpl::set_copy_data_callback(callback);
100 // XBT_ATTRIB_DEPRECATED_v333
101 void SIMIX_comm_copy_buffer_callback(simgrid::kernel::activity::CommImpl* comm, void* buff, size_t buff_size)
103 simgrid::s4u::Comm::copy_buffer_callback(comm, buff, buff_size);
106 // XBT_ATTRIB_DEPRECATED_v333
107 void SIMIX_comm_copy_pointer_callback(simgrid::kernel::activity::CommImpl* comm, void* buff, size_t buff_size)
109 simgrid::s4u::Comm::copy_pointer_callback(comm, buff, buff_size);
115 xbt::signal<void(CommImpl const&)> CommImpl::on_start;
116 xbt::signal<void(CommImpl const&)> CommImpl::on_completion;
118 void (*CommImpl::copy_data_callback_)(CommImpl*, void*, size_t) = &s4u::Comm::copy_pointer_callback;
120 void CommImpl::set_copy_data_callback(void (*callback)(CommImpl*, void*, size_t))
122 copy_data_callback_ = callback;
125 CommImpl& CommImpl::set_size(double size)
131 CommImpl& CommImpl::set_rate(double rate)
136 CommImpl& CommImpl::set_mailbox(MailboxImpl* mbox)
142 CommImpl& CommImpl::set_src_buff(unsigned char* buff, size_t size)
145 src_buff_size_ = size;
149 CommImpl& CommImpl::set_dst_buff(unsigned char* buff, size_t* size)
152 dst_buff_size_ = size;
156 CommImpl& CommImpl::detach()
162 CommImpl::CommImpl(s4u::Host* from, s4u::Host* to, double bytes) : size_(bytes), detached_(true), from_(from), to_(to)
164 set_state(State::READY);
167 CommImpl::~CommImpl()
169 XBT_DEBUG("Really free communication %p in state %s (detached = %d)", this, get_state_str(), detached_);
173 if (detached_ && get_state() != State::DONE) {
174 /* the communication has failed and was detached:
175 * we have to free the buffer */
177 clean_fun(src_buff_);
184 /** @brief Starts the simulation of a communication synchro. */
185 CommImpl* CommImpl::start()
187 /* If both the sender and the receiver are already there, start the communication */
188 if (get_state() == State::READY) {
189 from_ = from_ != nullptr ? from_ : src_actor_->get_host();
190 to_ = to_ != nullptr ? to_ : dst_actor_->get_host();
191 /* Getting the network_model from the origin host
192 * Valid while we have a single network model, otherwise we would need to change this function to first get the
193 * routes and later create the respective surf actions */
194 auto net_model = from_->get_netpoint()->get_englobing_zone()->get_network_model();
196 surf_action_ = net_model->communicate(from_, to_, size_, rate_);
197 surf_action_->set_activity(this);
198 surf_action_->set_category(get_tracing_category());
199 set_start_time(surf_action_->get_start_time());
200 set_state(State::RUNNING);
203 XBT_DEBUG("Starting communication %p from '%s' to '%s' (surf_action: %p; state: %s)", this, from_->get_cname(),
204 to_->get_cname(), surf_action_, get_state_str());
206 /* If a link is failed, detect it immediately */
207 if (surf_action_->get_state() == resource::Action::State::FAILED) {
208 XBT_DEBUG("Communication from '%s' to '%s' failed to start because of a link failure", from_->get_cname(),
210 set_state(State::LINK_FAILURE);
213 } else if ((src_actor_ != nullptr && src_actor_->is_suspended()) ||
214 (dst_actor_ != nullptr && dst_actor_->is_suspended())) {
215 /* If any of the actor is suspended, create the synchro but stop its execution,
216 it will be restarted when the sender actor resume */
217 if (src_actor_->is_suspended())
218 XBT_DEBUG("The communication is suspended on startup because src (%s@%s) was suspended since it initiated the "
220 src_actor_->get_cname(), src_actor_->get_host()->get_cname());
222 XBT_DEBUG("The communication is suspended on startup because dst (%s@%s) was suspended since it initiated the "
224 dst_actor_->get_cname(), dst_actor_->get_host()->get_cname());
226 surf_action_->suspend();
233 std::vector<s4u::Link*> CommImpl::get_traversed_links() const
235 xbt_assert(get_state() != State::WAITING, "You cannot use %s() if your communication is not ready (%s)", __FUNCTION__,
237 std::vector<s4u::Link*> vlinks;
238 XBT_ATTRIB_UNUSED double res = 0;
239 from_->route_to(to_, vlinks, &res);
243 /** @brief Copy the communication data from the sender's buffer to the receiver's one */
244 void CommImpl::copy_data()
246 size_t buff_size = src_buff_size_;
247 /* If there is no data to copy then return */
248 if (not src_buff_ || not dst_buff_ || copied_)
251 XBT_DEBUG("Copying comm %p data from %s (%p) -> %s (%p) (%zu bytes)", this,
252 src_actor_ ? src_actor_->get_host()->get_cname() : "a finished actor", src_buff_,
253 dst_actor_ ? dst_actor_->get_host()->get_cname() : "a finished actor", dst_buff_, buff_size);
255 /* Copy at most dst_buff_size bytes of the message to receiver's buffer */
256 if (dst_buff_size_) {
257 buff_size = std::min(buff_size, *dst_buff_size_);
259 /* Update the receiver's buffer size to the copied amount */
260 *dst_buff_size_ = buff_size;
265 copy_data_fun(this, src_buff_, buff_size);
267 copy_data_callback_(this, src_buff_, buff_size);
270 /* Set the copied flag so we copy data only once */
271 /* (this function might be called from both communication ends) */
276 CommImpl::isend(actor::ActorImpl* sender, MailboxImpl* mbox, double task_size, double rate, unsigned char* src_buff,
277 size_t src_buff_size, bool (*match_fun)(void*, void*, CommImpl*),
278 void (*clean_fun)(void*), // used to free the synchro in case of problem after a detached send
279 void (*copy_data_fun)(CommImpl*, void*, size_t), // used to copy data if not default one
280 void* data, bool detached)
282 XBT_DEBUG("send from mailbox %p", mbox);
284 /* Prepare a synchro describing us, so that it gets passed to the user-provided filter of other side */
285 CommImplPtr this_comm(new CommImpl(CommImpl::Type::SEND));
287 /* Look for communication synchro matching our needs. We also provide a description of
288 * ourself so that the other side also gets a chance of choosing if it wants to match with us.
290 * If it is not found then push our communication into the rendez-vous point */
291 CommImplPtr other_comm = mbox->find_matching_comm(CommImpl::Type::RECEIVE, match_fun, data, this_comm,
292 /*done*/ false, /*remove_matching*/ true);
294 if (not other_comm) {
295 other_comm = std::move(this_comm);
297 if (mbox->is_permanent()) {
298 // this mailbox is for small messages, which have to be sent right now
299 other_comm->set_state(State::READY);
300 other_comm->dst_actor_ = mbox->get_permanent_receiver().get();
301 mbox->push_done(other_comm);
302 XBT_DEBUG("pushing a message into the permanent receive list %p, comm %p", mbox, other_comm.get());
305 mbox->push(other_comm);
308 XBT_DEBUG("Receive already pushed");
310 other_comm->set_state(State::READY);
314 other_comm->detach();
315 other_comm->clean_fun = clean_fun;
317 other_comm->clean_fun = nullptr;
318 sender->activities_.emplace_back(other_comm);
321 /* Setup the communication synchro */
322 other_comm->src_actor_ = sender;
323 other_comm->src_data_ = data;
324 (*other_comm).set_src_buff(src_buff, src_buff_size).set_size(task_size).set_rate(rate);
326 other_comm->match_fun = match_fun;
327 other_comm->copy_data_fun = copy_data_fun;
329 if (MC_is_active() || MC_record_replay_is_active())
330 other_comm->set_state(simgrid::kernel::activity::State::RUNNING);
334 if (auto* observer = dynamic_cast<actor::CommIsendSimcall*>(sender->simcall_.observer_)) {
335 observer->set_result(detached ? nullptr : other_comm);
336 sender->simcall_answer();
339 return (detached ? nullptr : other_comm);
342 ActivityImplPtr CommImpl::irecv(actor::ActorImpl* receiver, MailboxImpl* mbox, unsigned char* dst_buff,
343 size_t* dst_buff_size, bool (*match_fun)(void*, void*, CommImpl*),
344 void (*copy_data_fun)(CommImpl*, void*, size_t), void* data, double rate)
346 CommImplPtr this_synchro(new CommImpl(CommImpl::Type::RECEIVE));
347 XBT_DEBUG("recv from mbox %p. this_synchro=%p", mbox, this_synchro.get());
349 CommImplPtr other_comm;
350 // communication already done, get it inside the list of completed comms
351 if (mbox->is_permanent() && mbox->has_some_done_comm()) {
352 XBT_DEBUG("We have a comm that has probably already been received, trying to match it, to skip the communication");
353 // find a match in the list of already received comms
354 other_comm = mbox->find_matching_comm(CommImpl::Type::SEND, match_fun, data, this_synchro, /*done*/ true,
355 /*remove_matching*/ true);
356 // if not found, assume the receiver came first, register it to the mailbox in the classical way
357 if (not other_comm) {
358 XBT_DEBUG("We have messages in the permanent receive list, but not the one we are looking for, pushing request "
360 other_comm = std::move(this_synchro);
361 mbox->push(other_comm);
363 if (other_comm->surf_action_ && other_comm->get_remaining() < 1e-12) {
364 XBT_DEBUG("comm %p has been already sent, and is finished, destroy it", other_comm.get());
365 other_comm->set_state(State::DONE);
366 other_comm->set_mailbox(nullptr);
370 /* Prepare a comm describing us, so that it gets passed to the user-provided filter of other side */
372 /* Look for communication activity matching our needs. We also provide a description of
373 * ourself so that the other side also gets a chance of choosing if it wants to match with us.
375 * If it is not found then push our communication into the rendez-vous point */
376 other_comm = mbox->find_matching_comm(CommImpl::Type::SEND, match_fun, data, this_synchro, /*done*/ false,
377 /*remove_matching*/ true);
379 if (other_comm == nullptr) {
380 XBT_DEBUG("Receive pushed first (%zu comm enqueued so far)", mbox->size());
381 other_comm = std::move(this_synchro);
382 mbox->push(other_comm);
384 XBT_DEBUG("Match my %p with the existing %p", this_synchro.get(), other_comm.get());
386 other_comm->set_state(simgrid::kernel::activity::State::READY);
388 receiver->activities_.emplace_back(other_comm);
391 /* Setup communication synchro */
392 other_comm->dst_actor_ = receiver;
393 other_comm->dst_data_ = data;
394 other_comm->set_dst_buff(dst_buff, dst_buff_size);
396 if (rate > -1.0 && (other_comm->get_rate() < 0.0 || rate < other_comm->get_rate()))
397 other_comm->set_rate(rate);
399 other_comm->match_fun = match_fun;
400 other_comm->copy_data_fun = copy_data_fun;
402 if (MC_is_active() || MC_record_replay_is_active()) {
403 other_comm->set_state(State::RUNNING);
408 if (auto* observer = dynamic_cast<actor::CommIrecvSimcall*>(receiver->simcall_.observer_)) {
409 observer->set_result(other_comm);
410 receiver->simcall_answer();
415 bool CommImpl::test(actor::ActorImpl* issuer)
417 if ((MC_is_active() || MC_record_replay_is_active()) && src_actor_ && dst_actor_)
418 set_state(State::DONE);
419 return ActivityImpl::test(issuer);
422 void CommImpl::wait_for(actor::ActorImpl* issuer, double timeout)
424 XBT_DEBUG("CommImpl::wait_for(%g), %p, state %s", timeout, this, get_state_str());
426 /* Associate this simcall to the wait synchro */
427 register_simcall(&issuer->simcall_);
428 if (MC_is_active() || MC_record_replay_is_active()) {
429 int idx = issuer->simcall_.mc_value_;
431 set_state(State::DONE);
433 /* If we reached this point, the wait simcall must have a timeout */
434 /* Otherwise it shouldn't be enabled and executed by the MC */
437 set_state(issuer == src_actor_ ? State::SRC_TIMEOUT : State::DST_TIMEOUT);
442 ActivityImpl::wait_for(issuer, timeout);
445 void CommImpl::wait_any_for(actor::ActorImpl* issuer, const std::vector<CommImpl*>& comms, double timeout)
447 std::vector<ActivityImpl*> activities(comms.begin(), comms.end());
448 ActivityImpl::wait_any_for(issuer, activities, timeout);
451 void CommImpl::suspend()
453 /* FIXME: shall we suspend also the timeout synchro? */
455 surf_action_->suspend();
456 /* if not created yet, the action will be suspended on creation, in CommImpl::start() */
459 void CommImpl::resume()
461 /*FIXME: check what happen with the timeouts */
463 surf_action_->resume();
464 /* in the other case, the synchro were not really suspended yet, see CommImpl::suspend() and CommImpl::start() */
467 void CommImpl::cancel()
469 /* if the synchro is a waiting state means that it is still in a mbox so remove from it and delete it */
470 if (get_state() == State::WAITING) {
473 set_state(State::CANCELED);
475 } else if (not MC_is_active() /* when running the MC there are no surf actions */
476 && not MC_record_replay_is_active() && (get_state() == State::READY || get_state() == State::RUNNING)) {
477 surf_action_->cancel();
481 /** @brief This is part of the cleanup process, probably an internal command */
482 void CommImpl::cleanup_surf()
487 src_timeout_->unref();
488 src_timeout_ = nullptr;
492 dst_timeout_->unref();
493 dst_timeout_ = nullptr;
497 void CommImpl::post()
499 on_completion(*this);
501 /* Update synchro state */
502 if (src_timeout_ && src_timeout_->get_state() == resource::Action::State::FINISHED)
503 set_state(State::SRC_TIMEOUT);
504 else if (dst_timeout_ && dst_timeout_->get_state() == resource::Action::State::FINISHED)
505 set_state(State::DST_TIMEOUT);
506 else if (src_timeout_ && src_timeout_->get_state() == resource::Action::State::FAILED)
507 set_state(State::SRC_HOST_FAILURE);
508 else if (dst_timeout_ && dst_timeout_->get_state() == resource::Action::State::FAILED)
509 set_state(State::DST_HOST_FAILURE);
510 else if (surf_action_ && surf_action_->get_state() == resource::Action::State::FAILED) {
511 set_state(State::LINK_FAILURE);
513 set_state(State::DONE);
515 XBT_DEBUG("CommImpl::post(): comm %p, state %s, src_proc %p, dst_proc %p, detached: %d", this, get_state_str(),
516 src_actor_.get(), dst_actor_.get(), detached_);
518 /* destroy the surf actions associated with the Simix communication */
521 /* Answer all simcalls associated with the synchro */
524 void CommImpl::set_exception(actor::ActorImpl* issuer)
526 switch (get_state()) {
528 issuer->exception_ = std::make_exception_ptr(NetworkFailureException(XBT_THROW_POINT, "Remote peer failed"));
530 case State::SRC_TIMEOUT:
532 std::make_exception_ptr(TimeoutException(XBT_THROW_POINT, "Communication timeouted because of the sender"));
535 case State::DST_TIMEOUT:
537 std::make_exception_ptr(TimeoutException(XBT_THROW_POINT, "Communication timeouted because of the receiver"));
540 case State::SRC_HOST_FAILURE:
541 if (issuer == src_actor_)
542 issuer->context_->set_wannadie();
544 set_state(State::FAILED);
545 issuer->exception_ = std::make_exception_ptr(NetworkFailureException(XBT_THROW_POINT, "Remote peer failed"));
549 case State::DST_HOST_FAILURE:
550 if (issuer == dst_actor_)
551 issuer->context_->set_wannadie();
553 set_state(State::FAILED);
554 issuer->exception_ = std::make_exception_ptr(NetworkFailureException(XBT_THROW_POINT, "Remote peer failed"));
558 case State::LINK_FAILURE:
559 XBT_DEBUG("Link failure in synchro %p between '%s' and '%s': posting an exception to the issuer: %s (%p) "
561 this, src_actor_ ? src_actor_->get_host()->get_cname() : nullptr,
562 dst_actor_ ? dst_actor_->get_host()->get_cname() : nullptr, issuer->get_cname(), issuer, detached_);
563 if (src_actor_ == issuer) {
564 XBT_DEBUG("I'm source");
565 } else if (dst_actor_ == issuer) {
566 XBT_DEBUG("I'm dest");
568 XBT_DEBUG("I'm neither source nor dest");
570 set_state(State::FAILED);
571 issuer->throw_exception(std::make_exception_ptr(NetworkFailureException(XBT_THROW_POINT, "Link failure")));
574 case State::CANCELED:
575 if (issuer == dst_actor_)
577 std::make_exception_ptr(CancelException(XBT_THROW_POINT, "Communication canceled by the sender"));
580 std::make_exception_ptr(CancelException(XBT_THROW_POINT, "Communication canceled by the receiver"));
584 xbt_assert(get_state() == State::DONE, "Internal error in CommImpl::finish(): unexpected synchro state %s",
589 void CommImpl::finish()
591 XBT_DEBUG("CommImpl::finish() in state %s", get_state_str());
592 /* If the synchro is still in a rendez-vous point then remove from it */
596 if (get_state() == State::DONE)
599 while (not simcalls_.empty()) {
600 smx_simcall_t simcall = simcalls_.front();
601 simcalls_.pop_front();
603 /* If a waitany simcall is waiting for this synchro to finish, then remove it from the other synchros in the waitany
604 * list. Afterwards, get the position of the actual synchro in the waitany list and return it as the result of the
607 if (simcall->call_ == simix::Simcall::NONE) // FIXME: maybe a better way to handle this case
608 continue; // if actor handling comm is killed
610 handle_activity_waitany(simcall);
612 /* Check out for errors */
614 if (not simcall->issuer_->get_host()->is_on()) {
615 simcall->issuer_->context_->set_wannadie();
617 // Do not answer to dying actors
618 if (not simcall->issuer_->context_->wannadie()) {
619 set_exception(simcall->issuer_);
620 simcall->issuer_->simcall_answer();
624 simcall->issuer_->waiting_synchro_ = nullptr;
625 simcall->issuer_->activities_.remove(this);
627 if (simcall->issuer_ != dst_actor_ && dst_actor_ != nullptr)
628 dst_actor_->activities_.remove(this);
629 if (simcall->issuer_ != src_actor_ && src_actor_ != nullptr)
630 src_actor_->activities_.remove(this);
635 } // namespace activity
636 } // namespace kernel
637 } // namespace simgrid