X-Git-Url: http://info.iut-bm.univ-fcomte.fr/pub/gitweb/simgrid.git/blobdiff_plain/38f246c4c46920a78ea0af8233449f402c178f04..78fedc4d8f9b0daf04101bd76017c27285a77251:/src/kernel/EngineImpl.cpp diff --git a/src/kernel/EngineImpl.cpp b/src/kernel/EngineImpl.cpp index f38a4ab37b..b6269d0e46 100644 --- a/src/kernel/EngineImpl.cpp +++ b/src/kernel/EngineImpl.cpp @@ -3,22 +3,22 @@ /* 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. */ -#include "src/kernel/EngineImpl.hpp" +#include +#include +#include +#include +#include +#include + #include "mc/mc.h" -#include "simgrid/Exception.hpp" -#include "simgrid/kernel/Timer.hpp" -#include "simgrid/kernel/routing/NetPoint.hpp" -#include "simgrid/kernel/routing/NetZoneImpl.hpp" -#include "simgrid/s4u/Host.hpp" -#include "simgrid/sg_config.hpp" -#include "src/include/surf/surf.hpp" //get_clock() and surf_solve() -#include "src/kernel/resource/DiskImpl.hpp" +#include "src/kernel/EngineImpl.hpp" #include "src/kernel/resource/profile/Profile.hpp" #include "src/mc/mc_record.hpp" #include "src/mc/mc_replay.hpp" #include "src/smpi/include/smpi_actor.hpp" #include "src/surf/network_interface.hpp" -#include "src/surf/xml/platf.hpp" // FIXME: KILLME. There must be a better way than mimicking XML here +#include "src/surf/xml/platf.hpp" +#include "xbt/xbt_modinter.h" /* whether initialization was already done */ #include #ifndef _WIN32 @@ -29,7 +29,10 @@ #include "src/mc/remote/AppSide.hpp" #endif +double NOW = 0; + XBT_LOG_NEW_DEFAULT_CATEGORY(ker_engine, "Logging specific to Engine (kernel)"); + namespace simgrid { namespace kernel { EngineImpl* EngineImpl::instance_ = nullptr; /* That singleton is awful too. */ @@ -37,6 +40,56 @@ EngineImpl* EngineImpl::instance_ = nullptr; /* That singleton is awful too. */ config::Flag cfg_breakpoint{"debug/breakpoint", "When non-negative, raise a SIGTRAP after given (simulated) time", -1.0}; config::Flag cfg_verbose_exit{"debug/verbose-exit", "Display the actor status at exit", true}; + +xbt_dynar_t get_actors_addr() +{ +#if SIMGRID_HAVE_MC + return EngineImpl::get_instance()->get_actors_vector(); +#else + xbt_die("This function is intended to be used when compiling with MC"); +#endif +} + +xbt_dynar_t get_dead_actors_addr() +{ +#if SIMGRID_HAVE_MC + return EngineImpl::get_instance()->get_dead_actors_vector(); +#else + xbt_die("This function is intended to be used when compiling with MC"); +#endif +} + +constexpr std::initializer_list> context_factories = { +#if HAVE_RAW_CONTEXTS + {"raw", &context::raw_factory}, +#endif +#if HAVE_UCONTEXT_CONTEXTS + {"ucontext", &context::sysv_factory}, +#endif +#if HAVE_BOOST_CONTEXTS + {"boost", &context::boost_factory}, +#endif + {"thread", &context::thread_factory}, +}; + +static_assert(context_factories.size() > 0, "No context factories are enabled for this build"); + +// Create the list of possible contexts: +static inline std::string contexts_list() +{ + std::string res; + std::string sep = ""; + for (auto const& factory : context_factories) { + res += sep + factory.first; + sep = ", "; + } + return res; +} + +static config::Flag context_factory_name("contexts/factory", + (std::string("Possible values: ") + contexts_list()).c_str(), + context_factories.begin()->first); + } // namespace kernel } // namespace simgrid @@ -69,7 +122,7 @@ static void segvhandler(int signum, siginfo_t* siginfo, void* /*context*/) "If you think you've found a bug in SimGrid, please report it along with a\n" "Minimal Working Example (MWE) reproducing your problem and a full backtrace\n" "of the fault captured with gdb or valgrind.\n", - smx_context_stack_size / 1024); + simgrid::kernel::context::stack_size / 1024); } else if (siginfo->si_signo == SIGSEGV) { fprintf(stderr, "Segmentation fault.\n"); #if HAVE_SMPI @@ -141,9 +194,8 @@ EngineImpl::~EngineImpl() for (auto const& kv : netpoints_) delete kv.second; - for (auto const& kv : links_) - if (kv.second) - kv.second->destroy(); + while (not links_.empty()) + links_.begin()->second->destroy(); for (auto const& kv : mailboxes_) delete kv.second; @@ -169,7 +221,11 @@ void EngineImpl::initialize(int* argc, char** argv) simgrid::mc::AppSide::initialize(); #endif - surf_init(argc, argv); /* Initialize SURF structures */ + if (xbt_initialized == 0) { + xbt_init(argc, argv); + + sg_config_init(argc, argv); + } instance_->context_mod_init(); @@ -182,17 +238,84 @@ void EngineImpl::initialize(int* argc, char** argv) /* register a function to be called by SURF after the environment creation */ sg_platf_init(); - s4u::Engine::on_platform_created.connect(surf_presolve); + s4u::Engine::on_platform_created.connect([this]() { this->presolve(); }); if (config::get_value("debug/clean-atexit")) atexit(shutdown); } +void EngineImpl::context_mod_init() const +{ + xbt_assert(not instance_->has_context_factory()); + +#if HAVE_SMPI && defined(__NetBSD__) + smpi_init_options_internal(false); + std::string priv = config::get_value("smpi/privatization"); + if (context_factory_name == "thread" && (priv == "dlopen" || priv == "yes" || priv == "default" || priv == "1")) { + XBT_WARN("dlopen+thread broken on Apple and BSD. Switching to raw contexts."); + context_factory_name = "raw"; + } +#endif + +#if HAVE_SMPI && defined(__FreeBSD__) + smpi_init_options_internal(false); + if (context_factory_name == "thread" && config::get_value("smpi/privatization") != "no") { + XBT_WARN("mmap broken on FreeBSD, but dlopen+thread broken too. Switching to dlopen+raw contexts."); + context_factory_name = "raw"; + } +#endif + + /* select the context factory to use to create the contexts */ + if (context::factory_initializer != nullptr) { // Give Java a chance to hijack the factory mechanism + instance_->set_context_factory(context::factory_initializer()); + return; + } + /* use the factory specified by --cfg=contexts/factory:value */ + for (auto const& factory : context_factories) + if (context_factory_name == factory.first) { + instance_->set_context_factory(factory.second()); + break; + } + + if (not instance_->has_context_factory()) { + XBT_ERROR("Invalid context factory specified. Valid factories on this machine:"); +#if HAVE_RAW_CONTEXTS + XBT_ERROR(" raw: high performance context factory implemented specifically for SimGrid"); +#else + XBT_ERROR(" (raw contexts were disabled at compilation time on this machine -- check configure logs for details)"); +#endif +#if HAVE_UCONTEXT_CONTEXTS + XBT_ERROR(" ucontext: classical system V contexts (implemented with makecontext, swapcontext and friends)"); +#else + XBT_ERROR(" (ucontext was disabled at compilation time on this machine -- check configure logs for details)"); +#endif +#if HAVE_BOOST_CONTEXTS + XBT_ERROR(" boost: this uses the boost libraries context implementation"); +#else + XBT_ERROR(" (boost was disabled at compilation time on this machine -- check configure logs for details. Did you " + "install the libboost-context-dev package?)"); +#endif + XBT_ERROR(" thread: slow portability layer using pthreads as provided by gcc"); + xbt_die("Please use a valid factory."); + } +} + void EngineImpl::shutdown() { if (EngineImpl::instance_ == nullptr) return; XBT_DEBUG("EngineImpl::shutdown() called. Simulation's over."); +#if HAVE_SMPI + if (not instance_->actor_list_.empty()) { + if (smpi_process()->initialized()) { + xbt_die("Process exited without calling MPI_Finalize - Killing simulation"); + } else { + XBT_WARN("Process called exit when leaving - Skipping cleanups"); + return; + } + } +#endif + if (instance_->has_actors_to_run() && simgrid_get_clock() <= 0.0) { XBT_CRITICAL(" "); XBT_CRITICAL("The time is still 0, and you still have processes ready to run."); @@ -205,17 +328,6 @@ void EngineImpl::shutdown() instance_->run_all_actors(); instance_->empty_trash(); -#if HAVE_SMPI - if (not instance_->actor_list_.empty()) { - if (smpi_process()->initialized()) { - xbt_die("Process exited without calling MPI_Finalize - Killing simulation"); - } else { - XBT_WARN("Process called exit when leaving - Skipping cleanups"); - return; - } - } -#endif - /* Let's free maestro now */ instance_->destroy_maestro(); @@ -437,6 +549,129 @@ void EngineImpl::display_all_actor_status() const } } +void EngineImpl::presolve() const +{ + XBT_DEBUG("Consume all trace events occurring before the starting time."); + double next_event_date; + while ((next_event_date = profile::future_evt_set.next_date()) != -1.0) { + if (next_event_date > NOW) + break; + + double value = -1.0; + resource::Resource* resource = nullptr; + while (auto* event = profile::future_evt_set.pop_leq(next_event_date, &value, &resource)) { + if (value >= 0) + resource->apply_event(event, value); + } + } + + XBT_DEBUG("Set every models in the right state by updating them to 0."); + for (auto const& model : models_) + model->update_actions_state(NOW, 0.0); +} + +double EngineImpl::solve(double max_date) const +{ + double time_delta = -1.0; /* duration */ + double value = -1.0; + resource::Resource* resource = nullptr; + + if (max_date != -1.0) { + xbt_assert(max_date >= NOW, "You asked to simulate up to %f, but that's in the past already", max_date); + + time_delta = max_date - NOW; + } + + XBT_DEBUG("Looking for next event in all models"); + for (auto model : models_) { + if (not model->next_occurring_event_is_idempotent()) { + continue; + } + double next_event = model->next_occurring_event(NOW); + if ((time_delta < 0.0 || next_event < time_delta) && next_event >= 0.0) { + time_delta = next_event; + } + } + + XBT_DEBUG("Min for resources (remember that NS3 don't update that value): %f", time_delta); + + XBT_DEBUG("Looking for next trace event"); + + while (true) { // Handle next occurring events until none remains + double next_event_date = profile::future_evt_set.next_date(); + XBT_DEBUG("Next TRACE event: %f", next_event_date); + + for (auto model : models_) { + /* Skip all idempotent models, they were already treated above + * NS3 is the one to handled here */ + if (model->next_occurring_event_is_idempotent()) + continue; + + if (next_event_date != -1.0) { + time_delta = std::min(next_event_date - NOW, time_delta); + } else { + time_delta = std::max(next_event_date - NOW, time_delta); // Get the positive component + } + + XBT_DEBUG("Run the NS3 network at most %fs", time_delta); + // run until min or next flow + double model_next_action_end = model->next_occurring_event(time_delta); + + XBT_DEBUG("Min for network : %f", model_next_action_end); + if (model_next_action_end >= 0.0) + time_delta = model_next_action_end; + } + + if (next_event_date < 0.0 || (next_event_date > NOW + time_delta)) { + // next event may have already occurred or will after the next resource change, then bail out + XBT_DEBUG("no next usable TRACE event. Stop searching for it"); + break; + } + + XBT_DEBUG("Updating models (min = %g, NOW = %g, next_event_date = %g)", time_delta, NOW, next_event_date); + + while (auto* event = profile::future_evt_set.pop_leq(next_event_date, &value, &resource)) { + if (resource->is_used() || (watched_hosts().find(resource->get_cname()) != watched_hosts().end())) { + time_delta = next_event_date - NOW; + XBT_DEBUG("This event invalidates the next_occurring_event() computation of models. Next event set to %f", + time_delta); + } + // FIXME: I'm too lame to update NOW live, so I change it and restore it so that the real update with surf_min + // will work + double round_start = NOW; + NOW = next_event_date; + /* update state of the corresponding resource to the new value. Does not touch lmm. + It will be modified if needed when updating actions */ + XBT_DEBUG("Calling update_resource_state for resource %s", resource->get_cname()); + resource->apply_event(event, value); + NOW = round_start; + } + } + + /* FIXME: Moved this test to here to avoid stopping simulation if there are actions running on cpus and all cpus are + * with availability = 0. This may cause an infinite loop if one cpu has a trace with periodicity = 0 and the other a + * trace with periodicity > 0. + * The options are: all traces with same periodicity(0 or >0) or we need to change the way how the events are managed + */ + if (time_delta < 0) { + XBT_DEBUG("No next event at all. Bail out now."); + return -1.0; + } + + XBT_DEBUG("Duration set to %f", time_delta); + + // Bump the time: jump into the future + NOW = NOW + time_delta; + + // Inform the models of the date change + for (auto const& model : models_) + model->update_actions_state(NOW, time_delta); + + s4u::Engine::on_time_advance(time_delta); + + return time_delta; +} + void EngineImpl::run() { if (MC_record_replay_is_active()) { @@ -450,7 +685,7 @@ void EngineImpl::run() do { XBT_DEBUG("New Schedule Round; size(queue)=%zu", actors_to_run_.size()); - if (cfg_breakpoint >= 0.0 && surf_get_clock() >= cfg_breakpoint) { + if (cfg_breakpoint >= 0.0 && simgrid_get_clock() >= cfg_breakpoint) { XBT_DEBUG("Breakpoint reached (%g)", cfg_breakpoint.get()); cfg_breakpoint = -1.0; #ifdef SIGTRAP @@ -493,7 +728,7 @@ void EngineImpl::run() * - You are added from ActorImpl::simcall_answer() only. When this function is called depends on the * resource kind (network, cpu, disk, whatever), but the same arguments hold. Let's take communications * as an example. - * - For communications, this function is called from SIMIX_comm_finish(). + * - For communications, this function is called from CommImpl::finish(). * This function itself don't mess with the order since simcalls are handled in FIFO order. * The function is called: * - before the comm starts (invalid parameters, or resource already dead or whatever). @@ -551,8 +786,8 @@ void EngineImpl::run() time = timer::Timer::next(); if (time > -1.0 || not actor_list_.empty()) { - XBT_DEBUG("Calling surf_solve"); - time = surf_solve(time); + XBT_DEBUG("Calling solve"); + time = solve(time); XBT_DEBUG("Moving time ahead : %g", time); } @@ -595,5 +830,15 @@ void EngineImpl::run() simgrid::s4u::Engine::on_simulation_end(); } + +double EngineImpl::get_clock() +{ + return NOW; +} } // namespace kernel } // namespace simgrid + +void SIMIX_run() // XBT_ATTRIB_DEPRECATED_v332 +{ + simgrid::kernel::EngineImpl::get_instance()->run(); +}