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/context/Context.hpp"
14 #include "src/kernel/resource/CpuImpl.hpp"
15 #include "src/kernel/resource/LinkImpl.hpp"
16 #include "src/kernel/resource/StandardLinkImpl.hpp"
17 #include "src/mc/mc_replay.hpp"
19 XBT_LOG_NEW_DEFAULT_SUBCATEGORY(ker_network, kernel, "Kernel network-related synchronization");
21 XBT_PRIVATE void simcall_HANDLER_comm_send(smx_simcall_t simcall, smx_actor_t src, smx_mailbox_t mbox, double task_size,
22 double rate, unsigned char* src_buff, size_t src_buff_size,
23 bool (*match_fun)(void*, void*, simgrid::kernel::activity::CommImpl*),
24 void (*copy_data_fun)(simgrid::kernel::activity::CommImpl*, void*, size_t),
25 void* data, double timeout)
27 simgrid::kernel::activity::ActivityImplPtr comm = simcall_HANDLER_comm_isend(
28 simcall, src, mbox, task_size, rate, src_buff, src_buff_size, match_fun, nullptr, copy_data_fun, data, false);
29 simcall->mc_value_ = 0;
30 comm->wait_for(simcall->issuer_, timeout);
33 XBT_PRIVATE simgrid::kernel::activity::ActivityImplPtr simcall_HANDLER_comm_isend(
34 smx_simcall_t /*simcall*/, smx_actor_t src_proc, smx_mailbox_t mbox, double task_size, double rate,
35 unsigned char* src_buff, size_t src_buff_size,
36 bool (*match_fun)(void*, void*, simgrid::kernel::activity::CommImpl*),
37 void (*clean_fun)(void*), // used to free the synchro in case of problem after a detached send
38 void (*copy_data_fun)(simgrid::kernel::activity::CommImpl*, void*, size_t), // used to copy data if not default one
39 void* data, bool detached)
41 XBT_DEBUG("send from mailbox %p", mbox);
43 /* Prepare a synchro describing us, so that it gets passed to the user-provided filter of other side */
44 simgrid::kernel::activity::CommImplPtr this_comm(
45 new simgrid::kernel::activity::CommImpl(simgrid::kernel::activity::CommImpl::Type::SEND));
47 /* Look for communication synchro matching our needs. We also provide a description of
48 * ourself so that the other side also gets a chance of choosing if it wants to match with us.
50 * If it is not found then push our communication into the rendez-vous point */
51 simgrid::kernel::activity::CommImplPtr other_comm =
52 mbox->find_matching_comm(simgrid::kernel::activity::CommImpl::Type::RECEIVE, match_fun, data, this_comm,
53 /*done*/ false, /*remove_matching*/ true);
56 other_comm = std::move(this_comm);
58 if (mbox->is_permanent()) {
59 // this mailbox is for small messages, which have to be sent right now
60 other_comm->set_state(simgrid::kernel::activity::State::READY);
61 other_comm->dst_actor_ = mbox->get_permanent_receiver().get();
62 mbox->push_done(other_comm);
63 XBT_DEBUG("pushing a message into the permanent receive list %p, comm %p", mbox, other_comm.get());
66 mbox->push(other_comm);
69 XBT_DEBUG("Receive already pushed");
71 other_comm->set_state(simgrid::kernel::activity::State::READY);
76 other_comm->clean_fun = clean_fun;
78 other_comm->clean_fun = nullptr;
79 src_proc->activities_.emplace_back(other_comm);
82 /* Setup the communication synchro */
83 other_comm->src_actor_ = src_proc;
84 other_comm->src_data_ = data;
85 (*other_comm).set_src_buff(src_buff, src_buff_size).set_size(task_size).set_rate(rate);
87 other_comm->match_fun = match_fun;
88 other_comm->copy_data_fun = copy_data_fun;
90 if (MC_is_active() || MC_record_replay_is_active())
91 other_comm->set_state(simgrid::kernel::activity::State::RUNNING);
95 return (detached ? nullptr : other_comm);
98 XBT_PRIVATE void simcall_HANDLER_comm_recv(smx_simcall_t simcall, smx_actor_t receiver, smx_mailbox_t mbox,
99 unsigned char* dst_buff, size_t* dst_buff_size,
100 bool (*match_fun)(void*, void*, simgrid::kernel::activity::CommImpl*),
101 void (*copy_data_fun)(simgrid::kernel::activity::CommImpl*, void*, size_t),
102 void* data, double timeout, double rate)
104 simgrid::kernel::activity::ActivityImplPtr comm = simcall_HANDLER_comm_irecv(
105 simcall, receiver, mbox, dst_buff, dst_buff_size, match_fun, copy_data_fun, data, rate);
106 simcall->mc_value_ = 0;
107 comm->wait_for(simcall->issuer_, timeout);
110 XBT_PRIVATE simgrid::kernel::activity::ActivityImplPtr
111 simcall_HANDLER_comm_irecv(smx_simcall_t /*simcall*/, smx_actor_t receiver, smx_mailbox_t mbox, unsigned char* dst_buff,
112 size_t* dst_buff_size, bool (*match_fun)(void*, void*, simgrid::kernel::activity::CommImpl*),
113 void (*copy_data_fun)(simgrid::kernel::activity::CommImpl*, void*, size_t), void* data,
116 simgrid::kernel::activity::CommImplPtr this_synchro(
117 new simgrid::kernel::activity::CommImpl(simgrid::kernel::activity::CommImpl::Type::RECEIVE));
118 XBT_DEBUG("recv from mbox %p. this_synchro=%p", mbox, this_synchro.get());
120 simgrid::kernel::activity::CommImplPtr other_comm;
121 // communication already done, get it inside the list of completed comms
122 if (mbox->is_permanent() && mbox->has_some_done_comm()) {
123 XBT_DEBUG("We have a comm that has probably already been received, trying to match it, to skip the communication");
124 // find a match in the list of already received comms
125 other_comm = mbox->find_matching_comm(simgrid::kernel::activity::CommImpl::Type::SEND, match_fun, data,
126 this_synchro, /*done*/ true,
127 /*remove_matching*/ true);
128 // if not found, assume the receiver came first, register it to the mailbox in the classical way
129 if (not other_comm) {
130 XBT_DEBUG("We have messages in the permanent receive list, but not the one we are looking for, pushing request "
132 other_comm = std::move(this_synchro);
133 mbox->push(other_comm);
135 if (other_comm->surf_action_ && other_comm->get_remaining() < 1e-12) {
136 XBT_DEBUG("comm %p has been already sent, and is finished, destroy it", other_comm.get());
137 other_comm->set_state(simgrid::kernel::activity::State::DONE);
138 other_comm->set_mailbox(nullptr);
142 /* Prepare a comm describing us, so that it gets passed to the user-provided filter of other side */
144 /* Look for communication activity matching our needs. We also provide a description of
145 * ourself so that the other side also gets a chance of choosing if it wants to match with us.
147 * If it is not found then push our communication into the rendez-vous point */
148 other_comm = mbox->find_matching_comm(simgrid::kernel::activity::CommImpl::Type::SEND, match_fun, data,
149 this_synchro, /*done*/ false,
150 /*remove_matching*/ true);
152 if (other_comm == nullptr) {
153 XBT_DEBUG("Receive pushed first (%zu comm enqueued so far)", mbox->size());
154 other_comm = std::move(this_synchro);
155 mbox->push(other_comm);
157 XBT_DEBUG("Match my %p with the existing %p", this_synchro.get(), other_comm.get());
159 other_comm->set_state(simgrid::kernel::activity::State::READY);
161 receiver->activities_.emplace_back(other_comm);
164 /* Setup communication synchro */
165 other_comm->dst_actor_ = receiver;
166 other_comm->dst_data_ = data;
167 other_comm->set_dst_buff(dst_buff, dst_buff_size);
169 if (rate > -1.0 && (other_comm->get_rate() < 0.0 || rate < other_comm->get_rate()))
170 other_comm->set_rate(rate);
172 other_comm->match_fun = match_fun;
173 other_comm->copy_data_fun = copy_data_fun;
175 if (MC_is_active() || MC_record_replay_is_active()) {
176 other_comm->set_state(simgrid::kernel::activity::State::RUNNING);
183 void simcall_HANDLER_comm_wait(smx_simcall_t simcall, simgrid::kernel::activity::CommImpl* comm, double timeout)
185 comm->wait_for(simcall->issuer_, timeout);
188 bool simcall_HANDLER_comm_test(smx_simcall_t simcall, simgrid::kernel::activity::CommImpl* comm)
190 return comm->test(simcall->issuer_);
193 ssize_t simcall_HANDLER_comm_testany(smx_simcall_t simcall, simgrid::kernel::activity::CommImpl* comms[], size_t count)
195 std::vector<simgrid::kernel::activity::ActivityImpl*> comms_vec(comms, comms + count);
196 return simgrid::kernel::activity::ActivityImpl::test_any(simcall->issuer_, comms_vec);
199 void simcall_HANDLER_comm_waitany(smx_simcall_t simcall, simgrid::kernel::activity::CommImpl* comms[], size_t count,
202 std::vector<simgrid::kernel::activity::CommImpl*> comms_vec(comms, comms + count);
203 simgrid::kernel::activity::CommImpl::wait_any_for(simcall->issuer_, comms_vec, timeout);
206 /******************************************************************************/
207 /* SIMIX_comm_copy_data callbacks */
208 /******************************************************************************/
209 // XBT_ATTRIB_DEPRECATED_v333
210 void SIMIX_comm_set_copy_data_callback(void (*callback)(simgrid::kernel::activity::CommImpl*, void*, size_t))
212 simgrid::kernel::activity::CommImpl::set_copy_data_callback(callback);
215 // XBT_ATTRIB_DEPRECATED_v333
216 void SIMIX_comm_copy_buffer_callback(simgrid::kernel::activity::CommImpl* comm, void* buff, size_t buff_size)
218 simgrid::s4u::Comm::copy_buffer_callback(comm, buff, buff_size);
221 // XBT_ATTRIB_DEPRECATED_v333
222 void SIMIX_comm_copy_pointer_callback(simgrid::kernel::activity::CommImpl* comm, void* buff, size_t buff_size)
224 simgrid::s4u::Comm::copy_pointer_callback(comm, buff, buff_size);
230 xbt::signal<void(CommImpl const&)> CommImpl::on_start;
231 xbt::signal<void(CommImpl const&)> CommImpl::on_completion;
233 void (*CommImpl::copy_data_callback_)(CommImpl*, void*, size_t) = &s4u::Comm::copy_pointer_callback;
235 void CommImpl::set_copy_data_callback(void (*callback)(CommImpl*, void*, size_t))
237 copy_data_callback_ = callback;
240 CommImpl& CommImpl::set_size(double size)
246 CommImpl& CommImpl::set_rate(double rate)
251 CommImpl& CommImpl::set_mailbox(MailboxImpl* mbox)
257 CommImpl& CommImpl::set_src_buff(unsigned char* buff, size_t size)
260 src_buff_size_ = size;
264 CommImpl& CommImpl::set_dst_buff(unsigned char* buff, size_t* size)
267 dst_buff_size_ = size;
271 CommImpl& CommImpl::detach()
277 CommImpl::CommImpl(s4u::Host* from, s4u::Host* to, double bytes) : size_(bytes), detached_(true), from_(from), to_(to)
279 set_state(State::READY);
282 CommImpl::~CommImpl()
284 XBT_DEBUG("Really free communication %p in state %s (detached = %d)", this, get_state_str(), detached_);
288 if (detached_ && get_state() != State::DONE) {
289 /* the communication has failed and was detached:
290 * we have to free the buffer */
292 clean_fun(src_buff_);
299 /** @brief Starts the simulation of a communication synchro. */
300 CommImpl* CommImpl::start()
302 /* If both the sender and the receiver are already there, start the communication */
303 if (get_state() == State::READY) {
304 from_ = from_ != nullptr ? from_ : src_actor_->get_host();
305 to_ = to_ != nullptr ? to_ : dst_actor_->get_host();
306 /* Getting the network_model from the origin host
307 * Valid while we have a single network model, otherwise we would need to change this function to first get the
308 * routes and later create the respective surf actions */
309 auto net_model = from_->get_netpoint()->get_englobing_zone()->get_network_model();
311 surf_action_ = net_model->communicate(from_, to_, size_, rate_);
312 surf_action_->set_activity(this);
313 surf_action_->set_category(get_tracing_category());
314 set_start_time(surf_action_->get_start_time());
315 set_state(State::RUNNING);
318 XBT_DEBUG("Starting communication %p from '%s' to '%s' (surf_action: %p; state: %s)", this, from_->get_cname(),
319 to_->get_cname(), surf_action_, get_state_str());
321 /* If a link is failed, detect it immediately */
322 if (surf_action_->get_state() == resource::Action::State::FAILED) {
323 XBT_DEBUG("Communication from '%s' to '%s' failed to start because of a link failure", from_->get_cname(),
325 set_state(State::LINK_FAILURE);
328 } else if ((src_actor_ != nullptr && src_actor_->is_suspended()) ||
329 (dst_actor_ != nullptr && dst_actor_->is_suspended())) {
330 /* If any of the actor is suspended, create the synchro but stop its execution,
331 it will be restarted when the sender actor resume */
332 if (src_actor_->is_suspended())
333 XBT_DEBUG("The communication is suspended on startup because src (%s@%s) was suspended since it initiated the "
335 src_actor_->get_cname(), src_actor_->get_host()->get_cname());
337 XBT_DEBUG("The communication is suspended on startup because dst (%s@%s) was suspended since it initiated the "
339 dst_actor_->get_cname(), dst_actor_->get_host()->get_cname());
341 surf_action_->suspend();
348 std::vector<s4u::Link*> CommImpl::get_traversed_links() const
350 xbt_assert(get_state() != State::WAITING, "You cannot use %s() if your communication is not ready (%s)", __FUNCTION__,
352 std::vector<s4u::Link*> vlinks;
353 XBT_ATTRIB_UNUSED double res = 0;
354 from_->route_to(to_, vlinks, &res);
358 /** @brief Copy the communication data from the sender's buffer to the receiver's one */
359 void CommImpl::copy_data()
361 size_t buff_size = src_buff_size_;
362 /* If there is no data to copy then return */
363 if (not src_buff_ || not dst_buff_ || copied_)
366 XBT_DEBUG("Copying comm %p data from %s (%p) -> %s (%p) (%zu bytes)", this,
367 src_actor_ ? src_actor_->get_host()->get_cname() : "a finished actor", src_buff_,
368 dst_actor_ ? dst_actor_->get_host()->get_cname() : "a finished actor", dst_buff_, buff_size);
370 /* Copy at most dst_buff_size bytes of the message to receiver's buffer */
371 if (dst_buff_size_) {
372 buff_size = std::min(buff_size, *dst_buff_size_);
374 /* Update the receiver's buffer size to the copied amount */
375 *dst_buff_size_ = buff_size;
380 copy_data_fun(this, src_buff_, buff_size);
382 copy_data_callback_(this, src_buff_, buff_size);
385 /* Set the copied flag so we copy data only once */
386 /* (this function might be called from both communication ends) */
390 bool CommImpl::test(actor::ActorImpl* issuer)
392 if ((MC_is_active() || MC_record_replay_is_active()) && src_actor_ && dst_actor_)
393 set_state(State::DONE);
394 return ActivityImpl::test(issuer);
397 void CommImpl::wait_for(actor::ActorImpl* issuer, double timeout)
399 XBT_DEBUG("CommImpl::wait_for(%g), %p, state %s", timeout, this, get_state_str());
401 /* Associate this simcall to the wait synchro */
402 register_simcall(&issuer->simcall_);
403 if (MC_is_active() || MC_record_replay_is_active()) {
404 int idx = issuer->simcall_.mc_value_;
406 set_state(State::DONE);
408 /* If we reached this point, the wait simcall must have a timeout */
409 /* Otherwise it shouldn't be enabled and executed by the MC */
412 set_state(issuer == src_actor_ ? State::SRC_TIMEOUT : State::DST_TIMEOUT);
417 ActivityImpl::wait_for(issuer, timeout);
420 void CommImpl::wait_any_for(actor::ActorImpl* issuer, const std::vector<CommImpl*>& comms, double timeout)
422 if (MC_is_active() || MC_record_replay_is_active()) {
423 xbt_assert(timeout <= 0.0, "Timeout not implemented for waitany in the model-checker");
424 int idx = issuer->simcall_.mc_value_;
425 auto* comm = comms[idx];
426 comm->simcalls_.push_back(&issuer->simcall_);
427 simcall_comm_waitany__set__result(&issuer->simcall_, idx);
428 comm->set_state(State::DONE);
432 std::vector<ActivityImpl*> activities(comms.begin(), comms.end());
433 ActivityImpl::wait_any_for(issuer, activities, timeout);
436 void CommImpl::suspend()
438 /* FIXME: shall we suspend also the timeout synchro? */
440 surf_action_->suspend();
441 /* if not created yet, the action will be suspended on creation, in CommImpl::start() */
444 void CommImpl::resume()
446 /*FIXME: check what happen with the timeouts */
448 surf_action_->resume();
449 /* in the other case, the synchro were not really suspended yet, see CommImpl::suspend() and CommImpl::start() */
452 void CommImpl::cancel()
454 /* if the synchro is a waiting state means that it is still in a mbox so remove from it and delete it */
455 if (get_state() == State::WAITING) {
458 set_state(State::CANCELED);
460 } else if (not MC_is_active() /* when running the MC there are no surf actions */
461 && not MC_record_replay_is_active() && (get_state() == State::READY || get_state() == State::RUNNING)) {
462 surf_action_->cancel();
466 /** @brief This is part of the cleanup process, probably an internal command */
467 void CommImpl::cleanup_surf()
472 src_timeout_->unref();
473 src_timeout_ = nullptr;
477 dst_timeout_->unref();
478 dst_timeout_ = nullptr;
482 void CommImpl::post()
484 on_completion(*this);
486 /* Update synchro state */
487 if (src_timeout_ && src_timeout_->get_state() == resource::Action::State::FINISHED)
488 set_state(State::SRC_TIMEOUT);
489 else if (dst_timeout_ && dst_timeout_->get_state() == resource::Action::State::FINISHED)
490 set_state(State::DST_TIMEOUT);
491 else if (src_timeout_ && src_timeout_->get_state() == resource::Action::State::FAILED)
492 set_state(State::SRC_HOST_FAILURE);
493 else if (dst_timeout_ && dst_timeout_->get_state() == resource::Action::State::FAILED)
494 set_state(State::DST_HOST_FAILURE);
495 else if (surf_action_ && surf_action_->get_state() == resource::Action::State::FAILED) {
496 set_state(State::LINK_FAILURE);
498 set_state(State::DONE);
500 XBT_DEBUG("CommImpl::post(): comm %p, state %s, src_proc %p, dst_proc %p, detached: %d", this, get_state_str(),
501 src_actor_.get(), dst_actor_.get(), detached_);
503 /* destroy the surf actions associated with the Simix communication */
506 /* Answer all simcalls associated with the synchro */
509 void CommImpl::set_exception(actor::ActorImpl* issuer)
511 switch (get_state()) {
513 issuer->exception_ = std::make_exception_ptr(NetworkFailureException(XBT_THROW_POINT, "Remote peer failed"));
515 case State::SRC_TIMEOUT:
517 std::make_exception_ptr(TimeoutException(XBT_THROW_POINT, "Communication timeouted because of the sender"));
520 case State::DST_TIMEOUT:
522 std::make_exception_ptr(TimeoutException(XBT_THROW_POINT, "Communication timeouted because of the receiver"));
525 case State::SRC_HOST_FAILURE:
526 if (issuer == src_actor_)
527 issuer->context_->set_wannadie();
529 set_state(State::FAILED);
530 issuer->exception_ = std::make_exception_ptr(NetworkFailureException(XBT_THROW_POINT, "Remote peer failed"));
534 case State::DST_HOST_FAILURE:
535 if (issuer == dst_actor_)
536 issuer->context_->set_wannadie();
538 set_state(State::FAILED);
539 issuer->exception_ = std::make_exception_ptr(NetworkFailureException(XBT_THROW_POINT, "Remote peer failed"));
543 case State::LINK_FAILURE:
544 XBT_DEBUG("Link failure in synchro %p between '%s' and '%s': posting an exception to the issuer: %s (%p) "
546 this, src_actor_ ? src_actor_->get_host()->get_cname() : nullptr,
547 dst_actor_ ? dst_actor_->get_host()->get_cname() : nullptr, issuer->get_cname(), issuer, detached_);
548 if (src_actor_ == issuer) {
549 XBT_DEBUG("I'm source");
550 } else if (dst_actor_ == issuer) {
551 XBT_DEBUG("I'm dest");
553 XBT_DEBUG("I'm neither source nor dest");
555 set_state(State::FAILED);
556 issuer->throw_exception(std::make_exception_ptr(NetworkFailureException(XBT_THROW_POINT, "Link failure")));
559 case State::CANCELED:
560 if (issuer == dst_actor_)
562 std::make_exception_ptr(CancelException(XBT_THROW_POINT, "Communication canceled by the sender"));
565 std::make_exception_ptr(CancelException(XBT_THROW_POINT, "Communication canceled by the receiver"));
569 xbt_assert(get_state() == State::DONE, "Internal error in CommImpl::finish(): unexpected synchro state %s",
574 void CommImpl::finish()
576 XBT_DEBUG("CommImpl::finish() in state %s", get_state_str());
577 /* If the synchro is still in a rendez-vous point then remove from it */
581 if (get_state() == State::DONE)
584 while (not simcalls_.empty()) {
585 smx_simcall_t simcall = simcalls_.front();
586 simcalls_.pop_front();
588 /* If a waitany simcall is waiting for this synchro to finish, then remove it from the other synchros in the waitany
589 * list. Afterwards, get the position of the actual synchro in the waitany list and return it as the result of the
592 if (simcall->call_ == simix::Simcall::NONE) // FIXME: maybe a better way to handle this case
593 continue; // if actor handling comm is killed
595 handle_activity_waitany(simcall);
597 /* Check out for errors */
599 if (not simcall->issuer_->get_host()->is_on()) {
600 simcall->issuer_->context_->set_wannadie();
602 set_exception(simcall->issuer_);
603 simcall->issuer_->simcall_answer();
606 simcall->issuer_->waiting_synchro_ = nullptr;
607 simcall->issuer_->activities_.remove(this);
609 if (simcall->issuer_ != dst_actor_ && dst_actor_ != nullptr)
610 dst_actor_->activities_.remove(this);
611 if (simcall->issuer_ != src_actor_ && src_actor_ != nullptr)
612 src_actor_->activities_.remove(this);
617 } // namespace activity
618 } // namespace kernel
619 } // namespace simgrid