X-Git-Url: http://bilbo.iut-bm.univ-fcomte.fr/pub/gitweb/simgrid.git/blobdiff_plain/eeaf1624487cd72c4e08db73e10b82e21fe083b3..9f2dfd25486ba89c4adaa54d774092c0b9756361:/src/kernel/activity/CommImpl.cpp diff --git a/src/kernel/activity/CommImpl.cpp b/src/kernel/activity/CommImpl.cpp index 842d669957..3abb315c8c 100644 --- a/src/kernel/activity/CommImpl.cpp +++ b/src/kernel/activity/CommImpl.cpp @@ -1,4 +1,4 @@ -/* Copyright (c) 2007-2022. The SimGrid Team. All rights reserved. */ +/* Copyright (c) 2007-2023. The SimGrid Team. All rights reserved. */ /* This program is free software; you can redistribute it and/or modify it * under the terms of the license (GNU LGPL) which comes with this package. */ @@ -8,111 +8,50 @@ #include #include +#include "src/kernel/EngineImpl.hpp" #include "src/kernel/activity/CommImpl.hpp" #include "src/kernel/activity/MailboxImpl.hpp" #include "src/kernel/actor/SimcallObserver.hpp" -#include "src/kernel/context/Context.hpp" #include "src/kernel/resource/CpuImpl.hpp" -#include "src/kernel/resource/LinkImpl.hpp" +#include "src/kernel/resource/NetworkModel.hpp" #include "src/kernel/resource/StandardLinkImpl.hpp" #include "src/mc/mc_replay.hpp" XBT_LOG_NEW_DEFAULT_SUBCATEGORY(ker_network, kernel, "Kernel network-related synchronization"); -XBT_PRIVATE void simcall_HANDLER_comm_send(smx_simcall_t simcall, smx_actor_t src, smx_mailbox_t mbox, double task_size, - double rate, unsigned char* src_buff, size_t src_buff_size, - bool (*match_fun)(void*, void*, simgrid::kernel::activity::CommImpl*), - void (*copy_data_fun)(simgrid::kernel::activity::CommImpl*, void*, size_t), - void* data, double timeout) -{ - simgrid::kernel::actor::CommIsendSimcall observer(src, mbox, task_size, rate, src_buff, src_buff_size, match_fun, - nullptr, copy_data_fun, data, false); - simgrid::kernel::activity::ActivityImplPtr comm = simgrid::kernel::activity::CommImpl::isend(&observer); - simcall->mc_value_ = 0; - comm->wait_for(simcall->issuer_, timeout); -} - -XBT_PRIVATE simgrid::kernel::activity::ActivityImplPtr simcall_HANDLER_comm_isend( - smx_simcall_t /*simcall*/, smx_actor_t src_proc, smx_mailbox_t mbox, double task_size, double rate, - unsigned char* src_buff, size_t src_buff_size, - bool (*match_fun)(void*, void*, simgrid::kernel::activity::CommImpl*), - void (*clean_fun)(void*), // used to free the synchro in case of problem after a detached send - void (*copy_data_fun)(simgrid::kernel::activity::CommImpl*, void*, size_t), // used to copy data if not default one - void* data, bool detached) -{ - simgrid::kernel::actor::CommIsendSimcall observer(src_proc, mbox, task_size, rate, src_buff, src_buff_size, match_fun, - clean_fun, copy_data_fun, data, detached); - return simgrid::kernel::activity::CommImpl::isend(&observer); -} +namespace simgrid::kernel::activity { -XBT_PRIVATE void simcall_HANDLER_comm_recv(smx_simcall_t simcall, smx_actor_t receiver, smx_mailbox_t mbox, - unsigned char* dst_buff, size_t* dst_buff_size, - bool (*match_fun)(void*, void*, simgrid::kernel::activity::CommImpl*), - void (*copy_data_fun)(simgrid::kernel::activity::CommImpl*, void*, size_t), - void* data, double timeout, double rate) -{ - simgrid::kernel::actor::CommIrecvSimcall observer(receiver, mbox, dst_buff, dst_buff_size, match_fun, copy_data_fun, - data, rate); - simgrid::kernel::activity::ActivityImplPtr comm = simgrid::kernel::activity::CommImpl::irecv(&observer); - simcall->mc_value_ = 0; - comm->wait_for(simcall->issuer_, timeout); -} - -XBT_PRIVATE simgrid::kernel::activity::ActivityImplPtr -simcall_HANDLER_comm_irecv(smx_simcall_t /*simcall*/, smx_actor_t receiver, smx_mailbox_t mbox, unsigned char* dst_buff, - size_t* dst_buff_size, bool (*match_fun)(void*, void*, simgrid::kernel::activity::CommImpl*), - void (*copy_data_fun)(simgrid::kernel::activity::CommImpl*, void*, size_t), void* data, - double rate) -{ - simgrid::kernel::actor::CommIrecvSimcall observer(receiver, mbox, dst_buff, dst_buff_size, match_fun, copy_data_fun, - data, rate); - return simgrid::kernel::activity::CommImpl::irecv(&observer); -} +std::function CommImpl::copy_data_callback_ = [](kernel::activity::CommImpl* comm, + void* buff, size_t buff_size) { + xbt_assert((buff_size == sizeof(void*)), "Cannot copy %zu bytes: must be sizeof(void*)", buff_size); + *(void**)(comm->dst_buff_) = buff; +}; -void simcall_HANDLER_comm_wait(smx_simcall_t simcall, simgrid::kernel::activity::CommImpl* comm, double timeout) +void CommImpl::set_copy_data_callback(const std::function& callback) { - comm->wait_for(simcall->issuer_, timeout); -} - -void simcall_HANDLER_comm_waitany(smx_simcall_t simcall, simgrid::kernel::activity::CommImpl* comms[], size_t count, - double timeout) -{ - std::vector comms_vec(comms, comms + count); - simgrid::kernel::activity::CommImpl::wait_any_for(simcall->issuer_, comms_vec, timeout); -} - -/******************************************************************************/ -/* SIMIX_comm_copy_data callbacks */ -/******************************************************************************/ -// XBT_ATTRIB_DEPRECATED_v333 -void SIMIX_comm_set_copy_data_callback(void (*callback)(simgrid::kernel::activity::CommImpl*, void*, size_t)) -{ - simgrid::kernel::activity::CommImpl::set_copy_data_callback(callback); + copy_data_callback_ = callback; } -// XBT_ATTRIB_DEPRECATED_v333 -void SIMIX_comm_copy_buffer_callback(simgrid::kernel::activity::CommImpl* comm, void* buff, size_t buff_size) +CommImpl& CommImpl::set_type(CommImplType type) { - simgrid::s4u::Comm::copy_buffer_callback(comm, buff, buff_size); + type_ = type; + return *this; } -// XBT_ATTRIB_DEPRECATED_v333 -void SIMIX_comm_copy_pointer_callback(simgrid::kernel::activity::CommImpl* comm, void* buff, size_t buff_size) +CommImpl& CommImpl::set_source(s4u::Host* from) { - simgrid::s4u::Comm::copy_pointer_callback(comm, buff, buff_size); + xbt_assert( from_ == nullptr ); + from_ = from; + add_host(from); + return *this; } -namespace simgrid { -namespace kernel { -namespace activity { -xbt::signal CommImpl::on_start; -xbt::signal CommImpl::on_completion; - -void (*CommImpl::copy_data_callback_)(CommImpl*, void*, size_t) = &s4u::Comm::copy_pointer_callback; - -void CommImpl::set_copy_data_callback(void (*callback)(CommImpl*, void*, size_t)) +CommImpl& CommImpl::set_destination(s4u::Host* to) { - copy_data_callback_ = callback; + xbt_assert( to_ == nullptr ); + to_ = to; + add_host(to_); + return *this; } CommImpl& CommImpl::set_size(double size) @@ -151,19 +90,15 @@ CommImpl& CommImpl::set_dst_buff(unsigned char* buff, size_t* size) CommImpl& CommImpl::detach() { detached_ = true; + EngineImpl::get_instance()->get_maestro()->activities_.insert(this); return *this; } -CommImpl::CommImpl(s4u::Host* from, s4u::Host* to, double bytes) : size_(bytes), detached_(true), from_(from), to_(to) -{ - set_state(State::READY); -} - CommImpl::~CommImpl() { XBT_DEBUG("Really free communication %p in state %s (detached = %d)", this, get_state_str(), detached_); - cleanup_surf(); + clean_action(); if (detached_ && get_state() != State::DONE) { /* the communication has failed and was detached: @@ -182,33 +117,35 @@ CommImpl* CommImpl::start() /* If both the sender and the receiver are already there, start the communication */ if (get_state() == State::READY) { from_ = from_ != nullptr ? from_ : src_actor_->get_host(); + xbt_assert(from_->is_on()); to_ = to_ != nullptr ? to_ : dst_actor_->get_host(); + xbt_assert(to_->is_on()); + /* Getting the network_model from the origin host * Valid while we have a single network model, otherwise we would need to change this function to first get the - * routes and later create the respective surf actions */ + * routes and later create the respective model actions */ auto net_model = from_->get_netpoint()->get_englobing_zone()->get_network_model(); - surf_action_ = net_model->communicate(from_, to_, size_, rate_); - surf_action_->set_activity(this); - surf_action_->set_category(get_tracing_category()); - set_start_time(surf_action_->get_start_time()); + model_action_ = net_model->communicate(from_, to_, size_, rate_, false); + model_action_->set_activity(this); + model_action_->set_category(get_tracing_category()); + set_start_time(model_action_->get_start_time()); set_state(State::RUNNING); - on_start(*this); - XBT_DEBUG("Starting communication %p from '%s' to '%s' (surf_action: %p; state: %s)", this, from_->get_cname(), - to_->get_cname(), surf_action_, get_state_str()); + XBT_DEBUG("Starting communication %p from '%s' to '%s' (model action: %p; state: %s)", this, from_->get_cname(), + to_->get_cname(), model_action_, get_state_str()); /* If a link is failed, detect it immediately */ - if (surf_action_->get_state() == resource::Action::State::FAILED) { + if (model_action_->get_state() == resource::Action::State::FAILED) { XBT_DEBUG("Communication from '%s' to '%s' failed to start because of a link failure", from_->get_cname(), to_->get_cname()); set_state(State::LINK_FAILURE); - post(); + finish(); } else if ((src_actor_ != nullptr && src_actor_->is_suspended()) || (dst_actor_ != nullptr && dst_actor_->is_suspended())) { /* If any of the actor is suspended, create the synchro but stop its execution, - it will be restarted when the sender actor resume */ + it will be restarted when the sender actor resumes */ if (src_actor_->is_suspended()) XBT_DEBUG("The communication is suspended on startup because src (%s@%s) was suspended since it initiated the " "communication", @@ -218,7 +155,7 @@ CommImpl* CommImpl::start() "communication", dst_actor_->get_cname(), dst_actor_->get_host()->get_cname()); - surf_action_->suspend(); + model_action_->suspend(); } } @@ -273,14 +210,15 @@ ActivityImplPtr CommImpl::isend(actor::CommIsendSimcall* observer) XBT_DEBUG("send from mailbox %p", mbox); /* Prepare a synchro describing us, so that it gets passed to the user-provided filter of other side */ - CommImplPtr this_comm(new CommImpl(CommImpl::Type::SEND)); + CommImplPtr this_comm(new CommImpl()); + this_comm->set_type(CommImplType::SEND); /* Look for communication synchro matching our needs. We also provide a description of * ourself so that the other side also gets a chance of choosing if it wants to match with us. * * If it is not found then push our communication into the rendez-vous point */ CommImplPtr other_comm = - mbox->find_matching_comm(CommImpl::Type::RECEIVE, observer->get_match_fun(), observer->get_payload(), this_comm, + mbox->find_matching_comm(CommImplType::RECEIVE, observer->get_match_fun(), observer->get_payload(), this_comm, /*done*/ false, /*remove_matching*/ true); if (not other_comm) { @@ -308,7 +246,7 @@ ActivityImplPtr CommImpl::isend(actor::CommIsendSimcall* observer) other_comm->clean_fun = observer->get_clean_fun(); } else { other_comm->clean_fun = nullptr; - observer->get_issuer()->activities_.emplace_back(other_comm); + observer->get_issuer()->activities_.insert(other_comm); } /* Setup the communication synchro */ @@ -332,7 +270,9 @@ ActivityImplPtr CommImpl::isend(actor::CommIsendSimcall* observer) ActivityImplPtr CommImpl::irecv(actor::CommIrecvSimcall* observer) { - CommImplPtr this_synchro(new CommImpl(CommImpl::Type::RECEIVE)); + CommImplPtr this_synchro(new CommImpl()); + this_synchro->set_type(CommImplType::RECEIVE); + auto* mbox = observer->get_mailbox(); XBT_DEBUG("recv from mbox %p. this_synchro=%p", mbox, this_synchro.get()); @@ -341,20 +281,21 @@ ActivityImplPtr CommImpl::irecv(actor::CommIrecvSimcall* observer) if (mbox->is_permanent() && mbox->has_some_done_comm()) { XBT_DEBUG("We have a comm that has probably already been received, trying to match it, to skip the communication"); // find a match in the list of already received comms - other_comm = mbox->find_matching_comm(CommImpl::Type::SEND, observer->get_match_fun(), observer->get_payload(), + other_comm = mbox->find_matching_comm(CommImplType::SEND, observer->get_match_fun(), observer->get_payload(), this_synchro, /*done*/ true, /*remove_matching*/ true); - // if not found, assume the receiver came first, register it to the mailbox in the classical way - if (not other_comm) { - XBT_DEBUG("We have messages in the permanent receive list, but not the one we are looking for, pushing request " - "into list"); - other_comm = std::move(this_synchro); - mbox->push(other_comm); + if (other_comm && other_comm->model_action_ && other_comm->get_remaining() < 1e-12) { + XBT_DEBUG("comm %p has been already sent, and is finished, destroy it", other_comm.get()); + other_comm->set_state(State::DONE); + other_comm->set_mailbox(nullptr); } else { - if (other_comm->surf_action_ && other_comm->get_remaining() < 1e-12) { - XBT_DEBUG("comm %p has been already sent, and is finished, destroy it", other_comm.get()); - other_comm->set_state(State::DONE); - other_comm->set_mailbox(nullptr); + // if not found, assume the receiver came first, register it to the mailbox in the classical way + if (not other_comm) { + XBT_DEBUG("We have messages in the permanent receive list, but not the one we are looking for, pushing request " + "into list"); + other_comm = std::move(this_synchro); + mbox->push(other_comm); } + observer->get_issuer()->activities_.insert(other_comm); } } else { /* Prepare a comm describing us, so that it gets passed to the user-provided filter of other side */ @@ -363,7 +304,7 @@ ActivityImplPtr CommImpl::irecv(actor::CommIrecvSimcall* observer) * ourself so that the other side also gets a chance of choosing if it wants to match with us. * * If it is not found then push our communication into the rendez-vous point */ - other_comm = mbox->find_matching_comm(CommImpl::Type::SEND, observer->get_match_fun(), observer->get_payload(), + other_comm = mbox->find_matching_comm(CommImplType::SEND, observer->get_match_fun(), observer->get_payload(), this_synchro, /*done*/ false, /*remove_matching*/ true); if (other_comm == nullptr) { @@ -375,7 +316,7 @@ ActivityImplPtr CommImpl::irecv(actor::CommIrecvSimcall* observer) other_comm->set_state(simgrid::kernel::activity::State::READY); } - observer->get_issuer()->activities_.emplace_back(other_comm); + observer->get_issuer()->activities_.insert(other_comm); } observer->set_comm(other_comm.get()); @@ -413,16 +354,8 @@ void CommImpl::wait_for(actor::ActorImpl* issuer, double timeout) /* Associate this simcall to the wait synchro */ register_simcall(&issuer->simcall_); if (MC_is_active() || MC_record_replay_is_active()) { - int idx = issuer->simcall_.mc_value_; - if (idx == 0) { - set_state(State::DONE); - } else { - /* If we reached this point, the wait simcall must have a timeout */ - /* Otherwise it shouldn't be enabled and executed by the MC */ - xbt_assert(timeout >= 0.0, - "The checker asked me to raise a timeout on a communication that is not expecting any timeout"); - set_state(issuer == src_actor_ ? State::SRC_TIMEOUT : State::DST_TIMEOUT); - } + // FIXME: what about timeouts? + set_state(State::DONE); finish(); return; } @@ -438,16 +371,16 @@ void CommImpl::wait_any_for(actor::ActorImpl* issuer, const std::vectorsuspend(); + if (model_action_) + model_action_->suspend(); /* if not created yet, the action will be suspended on creation, in CommImpl::start() */ } void CommImpl::resume() { /*FIXME: check what happen with the timeouts */ - if (surf_action_) - surf_action_->resume(); + if (model_action_) + model_action_->resume(); /* in the other case, the synchro were not really suspended yet, see CommImpl::suspend() and CommImpl::start() */ } @@ -459,55 +392,12 @@ void CommImpl::cancel() mbox_->remove(this); set_state(State::CANCELED); } - } else if (not MC_is_active() /* when running the MC there are no surf actions */ + } else if (not MC_is_active() /* when running the MC there are no model actions */ && not MC_record_replay_is_active() && (get_state() == State::READY || get_state() == State::RUNNING)) { - surf_action_->cancel(); - } -} - -/** @brief This is part of the cleanup process, probably an internal command */ -void CommImpl::cleanup_surf() -{ - clean_action(); - - if (src_timeout_) { - src_timeout_->unref(); - src_timeout_ = nullptr; - } - - if (dst_timeout_) { - dst_timeout_->unref(); - dst_timeout_ = nullptr; + model_action_->cancel(); } } -void CommImpl::post() -{ - on_completion(*this); - - /* Update synchro state */ - if (src_timeout_ && src_timeout_->get_state() == resource::Action::State::FINISHED) - set_state(State::SRC_TIMEOUT); - else if (dst_timeout_ && dst_timeout_->get_state() == resource::Action::State::FINISHED) - set_state(State::DST_TIMEOUT); - else if (src_timeout_ && src_timeout_->get_state() == resource::Action::State::FAILED) - set_state(State::SRC_HOST_FAILURE); - else if (dst_timeout_ && dst_timeout_->get_state() == resource::Action::State::FAILED) - set_state(State::DST_HOST_FAILURE); - else if (surf_action_ && surf_action_->get_state() == resource::Action::State::FAILED) { - set_state(State::LINK_FAILURE); - } else - set_state(State::DONE); - - XBT_DEBUG("CommImpl::post(): comm %p, state %s, src_proc %p, dst_proc %p, detached: %d", this, get_state_str(), - src_actor_.get(), dst_actor_.get(), detached_); - - /* destroy the surf actions associated with the Simix communication */ - cleanup_surf(); - - /* Answer all simcalls associated with the synchro */ - finish(); -} void CommImpl::set_exception(actor::ActorImpl* issuer) { switch (get_state()) { @@ -526,7 +416,7 @@ void CommImpl::set_exception(actor::ActorImpl* issuer) case State::SRC_HOST_FAILURE: if (issuer == src_actor_) - issuer->context_->set_wannadie(); + issuer->set_wannadie(); else { set_state(State::FAILED); issuer->exception_ = std::make_exception_ptr(NetworkFailureException(XBT_THROW_POINT, "Remote peer failed")); @@ -535,7 +425,7 @@ void CommImpl::set_exception(actor::ActorImpl* issuer) case State::DST_HOST_FAILURE: if (issuer == dst_actor_) - issuer->context_->set_wannadie(); + issuer->set_wannadie(); else { set_state(State::FAILED); issuer->exception_ = std::make_exception_ptr(NetworkFailureException(XBT_THROW_POINT, "Remote peer failed")); @@ -575,7 +465,39 @@ void CommImpl::set_exception(actor::ActorImpl* issuer) void CommImpl::finish() { - XBT_DEBUG("CommImpl::finish() in state %s", get_state_str()); + XBT_DEBUG("CommImpl::finish() comm %p, state %s, src_proc %p, dst_proc %p, detached: %d", this, get_state_str(), + src_actor_.get(), dst_actor_.get(), detached_); + + if (get_iface()) { + const auto& piface = static_cast(*get_iface()); + s4u::Comm::on_completion(piface); + if (not MC_is_active()) // MC doesn't like when we access user interface from the kernel. + piface.on_this_completion(piface); + } + + /* Update synchro state */ + if (src_timeout_ && src_timeout_->get_state() == resource::Action::State::FINISHED) + set_state(State::SRC_TIMEOUT); + else if (dst_timeout_ && dst_timeout_->get_state() == resource::Action::State::FINISHED) + set_state(State::DST_TIMEOUT); + else if ((from_ && not from_->is_on()) || + (src_timeout_ && src_timeout_->get_state() == resource::Action::State::FAILED)) + set_state(State::SRC_HOST_FAILURE); + else if ((to_ && not to_->is_on()) || (dst_timeout_ && dst_timeout_->get_state() == resource::Action::State::FAILED)) + set_state(State::DST_HOST_FAILURE); + else if (model_action_ && model_action_->get_state() == resource::Action::State::FAILED) { + set_state(State::LINK_FAILURE); + } else if (get_state() == State::RUNNING) { + xbt_assert(from_ && from_->is_on()); + xbt_assert(to_ && to_->is_on()); + set_state(State::DONE); + } + src_timeout_ = nullptr; + dst_timeout_ = nullptr; + + /* destroy the model actions associated with the communication activity */ + clean_action(); + /* If the synchro is still in a rendez-vous point then remove from it */ if (mbox_) mbox_->remove(this); @@ -583,42 +505,43 @@ void CommImpl::finish() if (get_state() == State::DONE) copy_data(); + if (detached_) + EngineImpl::get_instance()->get_maestro()->activities_.erase(this); + while (not simcalls_.empty()) { - smx_simcall_t simcall = simcalls_.front(); + actor::Simcall* simcall = simcalls_.front(); simcalls_.pop_front(); /* If a waitany simcall is waiting for this synchro to finish, then remove it from the other synchros in the waitany * list. Afterwards, get the position of the actual synchro in the waitany list and return it as the result of the * simcall */ - if (simcall->call_ == simix::Simcall::NONE) // FIXME: maybe a better way to handle this case - continue; // if actor handling comm is killed + if (simcall->call_ == actor::Simcall::Type::NONE) // FIXME: maybe a better way to handle this case + continue; // if actor handling comm is killed handle_activity_waitany(simcall); /* Check out for errors */ if (not simcall->issuer_->get_host()->is_on()) { - simcall->issuer_->context_->set_wannadie(); + simcall->issuer_->set_wannadie(); } else { // Do not answer to dying actors - if (not simcall->issuer_->context_->wannadie()) { + if (not simcall->issuer_->wannadie()) { set_exception(simcall->issuer_); simcall->issuer_->simcall_answer(); } } simcall->issuer_->waiting_synchro_ = nullptr; - simcall->issuer_->activities_.remove(this); + simcall->issuer_->activities_.erase(this); if (detached_) { if (simcall->issuer_ != dst_actor_ && dst_actor_ != nullptr) - dst_actor_->activities_.remove(this); + dst_actor_->activities_.erase(this); if (simcall->issuer_ != src_actor_ && src_actor_ != nullptr) - src_actor_->activities_.remove(this); + src_actor_->activities_.erase(this); } } } -} // namespace activity -} // namespace kernel -} // namespace simgrid +} // namespace simgrid::kernel::activity