X-Git-Url: http://info.iut-bm.univ-fcomte.fr/pub/gitweb/simgrid.git/blobdiff_plain/db2d173d5f0a8dfa069697279d937ac7203234a5..64e7e8c55cb98ef667ebe13284e1a215fe1a8862:/src/kernel/EngineImpl.cpp diff --git a/src/kernel/EngineImpl.cpp b/src/kernel/EngineImpl.cpp index 5832916af5..bcb35621eb 100644 --- a/src/kernel/EngineImpl.cpp +++ b/src/kernel/EngineImpl.cpp @@ -3,43 +3,153 @@ /* 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/simix/smx_private.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 "surf/surf.hpp" //surf_presolve() and surf_solve() +#include "xbt/xbt_modinter.h" /* whether initialization was already done */ #include #ifndef _WIN32 #include #endif /* _WIN32 */ -XBT_LOG_NEW_DEFAULT_CATEGORY(ker_engine, "Logging specific to Engine (kernel)"); +#if SIMGRID_HAVE_MC +#include "src/mc/remote/AppSide.hpp" +#endif +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. */ config::Flag cfg_breakpoint{"debug/breakpoint", "When non-negative, raise a SIGTRAP after given (simulated) time", -1.0}; -EngineImpl::~EngineImpl() +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 +} - while (not timer::kernel_timers().empty()) { - delete timer::kernel_timers().top().second; - timer::kernel_timers().pop(); +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 +} + +} // namespace kernel +} // namespace simgrid + +XBT_ATTRIB_NORETURN static void inthandler(int) +{ + if (simgrid::kernel::cfg_verbose_exit) { + XBT_INFO("CTRL-C pressed. The current status will be displayed before exit (disable that behavior with option " + "'debug/verbose-exit')."); + simgrid::kernel::EngineImpl::get_instance()->display_all_actor_status(); + } else { + XBT_INFO("CTRL-C pressed, exiting. Hiding the current process status since 'debug/verbose-exit' is set to false."); } + exit(1); +} + +#ifndef _WIN32 +static void segvhandler(int signum, siginfo_t* siginfo, void* /*context*/) +{ + if ((siginfo->si_signo == SIGSEGV && siginfo->si_code == SEGV_ACCERR) || siginfo->si_signo == SIGBUS) { + fprintf(stderr, + "Access violation or Bus error detected.\n" + "This probably comes from a programming error in your code, or from a stack\n" + "overflow. If you are certain of your code, try increasing the stack size\n" + " --cfg=contexts/stack-size:XXX (current size is %u KiB).\n" + "\n" + "If it does not help, this may have one of the following causes:\n" + "a bug in SimGrid, a bug in the OS or a bug in a third-party libraries.\n" + "Failing hardware can sometimes generate such errors too.\n" + "\n" + "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); + } else if (siginfo->si_signo == SIGSEGV) { + fprintf(stderr, "Segmentation fault.\n"); +#if HAVE_SMPI + if (smpi_enabled() && smpi_cfg_privatization() == SmpiPrivStrategies::NONE) { +#if HAVE_PRIVATIZATION + fprintf(stderr, "Try to enable SMPI variable privatization with --cfg=smpi/privatization:yes.\n"); +#else + fprintf(stderr, "Sadly, your system does not support --cfg=smpi/privatization:yes (yet).\n"); +#endif /* HAVE_PRIVATIZATION */ + } +#endif /* HAVE_SMPI */ + } + std::raise(signum); +} + +/** + * Install signal handler for SIGSEGV. Check that nobody has already installed + * its own handler. For example, the Java VM does this. + */ +static void install_segvhandler() +{ + stack_t old_stack; + + if (simgrid::kernel::context::Context::install_sigsegv_stack(&old_stack, true) == -1) { + XBT_WARN("Failed to register alternate signal stack: %s", strerror(errno)); + return; + } + if (not(old_stack.ss_flags & SS_DISABLE)) { + XBT_DEBUG("An alternate stack was already installed (sp=%p, size=%zu, flags=%x). Restore it.", old_stack.ss_sp, + old_stack.ss_size, (unsigned)old_stack.ss_flags); + sigaltstack(&old_stack, nullptr); + } + + struct sigaction action; + struct sigaction old_action; + action.sa_sigaction = &segvhandler; + action.sa_flags = SA_ONSTACK | SA_RESETHAND | SA_SIGINFO; + sigemptyset(&action.sa_mask); + + /* Linux tend to raise only SIGSEGV where other systems also raise SIGBUS on severe error */ + for (int sig : {SIGSEGV, SIGBUS}) { + if (sigaction(sig, &action, &old_action) == -1) { + XBT_WARN("Failed to register signal handler for signal %d: %s", sig, strerror(errno)); + continue; + } + if ((old_action.sa_flags & SA_SIGINFO) || old_action.sa_handler != SIG_DFL) { + XBT_DEBUG("A signal handler was already installed for signal %d (%p). Restore it.", sig, + (old_action.sa_flags & SA_SIGINFO) ? (void*)old_action.sa_sigaction : (void*)old_action.sa_handler); + sigaction(sig, &old_action, nullptr); + } + } +} + +#endif /* _WIN32 */ +namespace simgrid { +namespace kernel { + +EngineImpl::~EngineImpl() +{ /* Since hosts_ is a std::map, the hosts are destroyed in the lexicographic order, which ensures that the output is * reproducible. */ @@ -51,9 +161,8 @@ 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; @@ -65,8 +174,87 @@ EngineImpl::~EngineImpl() #endif /* clear models before freeing handle, network models can use external callback defined in the handle */ models_prio_.clear(); - if (platf_handle_) - dlclose(platf_handle_); +} + +void EngineImpl::initialize(int* argc, char** argv) +{ + xbt_assert(EngineImpl::instance_ == nullptr, + "It is currently forbidden to create more than one instance of kernel::EngineImpl"); + EngineImpl::instance_ = this; +#if SIMGRID_HAVE_MC + // The communication initialization is done ASAP, as we need to get some init parameters from the MC for different + // layers. But simix_global needs to be created, as we send the address of some of its fields to the MC that wants to + // read them directly. + simgrid::mc::AppSide::initialize(); +#endif + + if (xbt_initialized == 0) { + xbt_init(argc, argv); + + sg_config_init(argc, argv); + } + + instance_->context_mod_init(); + + /* Prepare to display some more info when dying on Ctrl-C pressing */ + std::signal(SIGINT, inthandler); + +#ifndef _WIN32 + install_segvhandler(); +#endif + + /* register a function to be called by SURF after the environment creation */ + sg_platf_init(); + s4u::Engine::on_platform_created.connect(surf_presolve); + + if (config::get_value("debug/clean-atexit")) + atexit(shutdown); +} + +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."); + XBT_CRITICAL("It seems that you forgot to run the simulation that you setup."); + xbt_die("Bailing out to avoid that stop-before-start madness. Please fix your code."); + } + + /* Kill all actors (but maestro) */ + instance_->maestro_->kill_all(); + instance_->run_all_actors(); + instance_->empty_trash(); + + /* Let's free maestro now */ + instance_->destroy_maestro(); + + /* Finish context module and SURF */ + instance_->destroy_context_factory(); + + while (not timer::kernel_timers().empty()) { + delete timer::kernel_timers().top().second; + timer::kernel_timers().pop(); + } + + tmgr_finalize(); + sg_platf_exit(); + + delete instance_; + instance_ = nullptr; } void EngineImpl::load_platform(const std::string& platf) @@ -76,11 +264,11 @@ void EngineImpl::load_platform(const std::string& platf) #ifdef _WIN32 xbt_die("loading platform through shared library isn't supported on windows"); #else - platf_handle_ = dlopen(platf.c_str(), RTLD_LAZY); - xbt_assert(platf_handle_, "Impossible to open platform file: %s", platf.c_str()); + void* handle = dlopen(platf.c_str(), RTLD_LAZY); + xbt_assert(handle, "Impossible to open platform file: %s", platf.c_str()); + platf_handle_ = std::unique_ptr>(handle, dlclose); using load_fct_t = void (*)(const simgrid::s4u::Engine&); - dlerror(); - auto callable = (load_fct_t)dlsym(platf_handle_, "load_platform"); + auto callable = (load_fct_t)dlsym(platf_handle_.get(), "load_platform"); const char* dlsym_error = dlerror(); xbt_assert(not dlsym_error, "Error: %s", dlsym_error); callable(*simgrid::s4u::Engine::get_instance()); @@ -126,6 +314,11 @@ void EngineImpl::add_model(std::shared_ptr model, const std::ve models_prio_[model_name] = std::move(model); } +void EngineImpl::add_split_duplex_link(const std::string& name, std::unique_ptr link) +{ + split_duplex_links_[name] = std::move(link); +} + /** Wake up all actors waiting for a Surf action to finish */ void EngineImpl::wake_all_waiting_actors() const { @@ -155,7 +348,7 @@ void EngineImpl::wake_all_waiting_actors() const */ void EngineImpl::run_all_actors() { - simix_global->get_context_factory()->run_all(); + instance_->get_context_factory()->run_all(); actors_to_run_.swap(actors_that_ran_); actors_to_run_.clear(); @@ -173,14 +366,14 @@ actor::ActorImpl* EngineImpl::get_actor_by_pid(aid_t pid) return &a; return nullptr; // Not found, even in the trash } + /** Execute all the tasks that are queued, e.g. `.then()` callbacks of futures. */ bool EngineImpl::execute_tasks() { - xbt_assert(tasksTemp.empty()); - if (tasks.empty()) return false; + std::vector> tasksTemp; do { // We don't want the callbacks to modify the vector we are iterating over: tasks.swap(tasksTemp); @@ -280,7 +473,7 @@ 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 @@ -375,7 +568,7 @@ void EngineImpl::run() if (actor_list_.size() == daemons_.size()) for (auto const& dmon : daemons_) { XBT_DEBUG("Kill %s", dmon->get_cname()); - simix_global->get_maestro()->kill(dmon); + maestro_->kill(dmon); } } @@ -415,7 +608,7 @@ void EngineImpl::run() simgrid::s4u::Engine::on_deadlock(); for (auto const& kv : actor_list_) { XBT_DEBUG("Kill %s", kv.second->get_cname()); - simix_global->get_maestro()->kill(kv.second); + maestro_->kill(kv.second); } } } while (time > -1.0 || has_actors_to_run()); @@ -427,3 +620,8 @@ void EngineImpl::run() } } // namespace kernel } // namespace simgrid + +void SIMIX_run() // XBT_ATTRIB_DEPRECATED_v332 +{ + simgrid::kernel::EngineImpl::get_instance()->run(); +}