-/* Copyright (c) 2007-2019. The SimGrid Team. All rights reserved. */
+/* Copyright (c) 2007-2021. 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. */
#include "mc/mc.h"
+#include "simgrid/kernel/Timer.hpp"
#include "simgrid/s4u/Engine.hpp"
#include "simgrid/s4u/Host.hpp"
#include "src/smpi/include/smpi_actor.hpp"
#include "simgrid/sg_config.hpp"
-#include "src/kernel/activity/ExecImpl.hpp"
-#include "src/kernel/activity/IoImpl.hpp"
-#include "src/kernel/activity/MailboxImpl.hpp"
-#include "src/kernel/activity/SleepImpl.hpp"
-#include "src/kernel/activity/SynchroRaw.hpp"
+#include "src/kernel/EngineImpl.hpp"
#include "src/mc/mc_record.hpp"
#include "src/mc/mc_replay.hpp"
#include "src/simix/smx_private.hpp"
-#include "src/surf/StorageImpl.hpp"
#include "src/surf/xml/platf.hpp"
+#include "simgrid/kernel/resource/Model.hpp"
+
#if SIMGRID_HAVE_MC
-#include "src/mc/remote/Client.hpp"
+#include "src/mc/remote/AppSide.hpp"
#endif
+#include <memory>
XBT_LOG_NEW_CATEGORY(simix, "All SIMIX categories");
XBT_LOG_NEW_DEFAULT_SUBCATEGORY(simix_kernel, simix, "Logging specific to SIMIX (kernel)");
namespace simgrid {
namespace simix {
-config::Flag<double> cfg_verbose_exit{"debug/verbose-exit", {"verbose-exit"}, "Display the actor status at exit", true};
+config::Flag<bool> cfg_verbose_exit{"debug/verbose-exit", "Display the actor status at exit", true};
+
+xbt_dynar_t simix_global_get_actors_addr()
+{
+#if SIMGRID_HAVE_MC
+ return kernel::EngineImpl::get_instance()->get_actors_vector();
+#else
+ xbt_die("This function is intended to be used when compiling with MC");
+#endif
+}
+xbt_dynar_t simix_global_get_dead_actors_addr()
+{
+#if SIMGRID_HAVE_MC
+ return kernel::EngineImpl::get_instance()->get_dead_actors_vector();
+#else
+ xbt_die("This function is intended to be used when compiling with MC");
+#endif
+}
+
} // namespace simix
} // namespace simgrid
if (simgrid::simix::cfg_verbose_exit) {
XBT_INFO("CTRL-C pressed. The current status will be displayed before exit (disable that behavior with option "
"'debug/verbose-exit').");
- simix_global->display_all_actor_status();
- }
- else {
+ 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);
"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) {
+ } else if (siginfo->si_signo == SIGSEGV) {
fprintf(stderr, "Segmentation fault.\n");
#if HAVE_SMPI
if (smpi_enabled() && smpi_cfg_privatization() == SmpiPrivStrategies::NONE) {
std::raise(signum);
}
-unsigned char sigsegv_stack[SIGSTKSZ]; /* alternate stack for SIGSEGV handler */
-
/**
* 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 stack;
stack_t old_stack;
- stack.ss_sp = sigsegv_stack;
- stack.ss_size = sizeof sigsegv_stack;
- stack.ss_flags = 0;
- if (sigaltstack(&stack, &old_stack) == -1) {
+ if (simgrid::kernel::context::Context::install_sigsegv_stack(&old_stack, true) == -1) {
XBT_WARN("Failed to register alternate signal stack: %s", strerror(errno));
return;
}
struct sigaction action;
struct sigaction old_action;
action.sa_sigaction = &segvhandler;
- action.sa_flags = SA_ONSTACK | SA_RESETHAND | SA_SIGINFO;
+ 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 */
#endif /* _WIN32 */
-/********************************* SIMIX **************************************/
-namespace simgrid {
-namespace simix {
-
-Timer* Timer::set(double date, xbt::Task<void()>&& callback)
-{
- Timer* timer = new Timer(date, std::move(callback));
- timer->handle_ = simix_timers.emplace(std::make_pair(date, timer));
- return timer;
-}
-
-/** @brief cancels a timer that was added earlier */
-void Timer::remove()
-{
- simix_timers.erase(handle_);
- delete this;
-}
-
-/** Execute all the tasks that are queued, e.g. `.then()` callbacks of futures. */
-bool Global::execute_tasks()
-{
- xbt_assert(tasksTemp.empty());
-
- if (tasks.empty())
- return false;
-
- do {
- // We don't want the callbacks to modify the vector we are iterating over:
- tasks.swap(tasksTemp);
-
- // Execute all the queued tasks:
- for (auto& task : tasksTemp)
- task();
-
- tasksTemp.clear();
- } while (not tasks.empty());
-
- return true;
-}
-
-void Global::empty_trash()
-{
- while (not actors_to_destroy.empty()) {
- kernel::actor::ActorImpl* actor = &actors_to_destroy.front();
- actors_to_destroy.pop_front();
- XBT_DEBUG("Getting rid of %s (refcount: %d)", actor->get_cname(), actor->get_refcount());
- intrusive_ptr_release(actor);
- }
-#if SIMGRID_HAVE_MC
- xbt_dynar_reset(dead_actors_vector);
-#endif
-}
-/**
- * @brief Executes the actors in actors_to_run.
- *
- * The actors in actors_to_run are run (in parallel if possible). On exit, actors_to_run is empty, and actors_that_ran
- * contains the list of actors that just ran. The two lists are swapped so, be careful when using them before and after
- * a call to this function.
- */
-void Global::run_all_actors()
-{
- simix_global->context_factory->run_all();
-
- actors_to_run.swap(actors_that_ran);
- actors_to_run.clear();
-}
-
-/** Wake up all actors waiting for a Surf action to finish */
-void Global::wake_all_waiting_actors()
-{
- for (auto const& model : all_existing_models) {
- kernel::resource::Action* action;
-
- XBT_DEBUG("Handling the failed actions (if any)");
- while ((action = model->extract_failed_action())) {
- XBT_DEBUG(" Handling Action %p", action);
- if (action->get_activity() != nullptr)
- kernel::activity::ActivityImplPtr(action->get_activity())->post();
- }
- XBT_DEBUG("Handling the terminated actions (if any)");
- while ((action = model->extract_done_action())) {
- XBT_DEBUG(" Handling Action %p", action);
- if (action->get_activity() == nullptr)
- XBT_DEBUG("probably vcpu's action %p, skip", action);
- else
- kernel::activity::ActivityImplPtr(action->get_activity())->post();
- }
- }
-}
-
-void Global::display_all_actor_status()
-{
- XBT_INFO("%lu actors are still running, waiting for something.", process_list.size());
- /* List the actors and their state */
- XBT_INFO("Legend of the following listing: \"Actor <pid> (<name>@<host>): <status>\"");
- for (auto const& kv : process_list) {
- kernel::actor::ActorImpl* actor = kv.second;
-
- if (actor->waiting_synchro) {
- const char* synchro_description = "unknown";
- // we don't care about the Activity type to get its name, use RawImpl
- const char* name = boost::static_pointer_cast<kernel::activity::ActivityImpl_T<kernel::activity::RawImpl>>(
- actor->waiting_synchro)
- ->get_cname();
-
- if (boost::dynamic_pointer_cast<kernel::activity::ExecImpl>(actor->waiting_synchro) != nullptr)
- synchro_description = "execution";
-
- if (boost::dynamic_pointer_cast<kernel::activity::CommImpl>(actor->waiting_synchro) != nullptr)
- synchro_description = "communication";
-
- if (boost::dynamic_pointer_cast<kernel::activity::SleepImpl>(actor->waiting_synchro) != nullptr)
- synchro_description = "sleeping";
-
- if (boost::dynamic_pointer_cast<kernel::activity::RawImpl>(actor->waiting_synchro) != nullptr)
- synchro_description = "synchronization";
-
- if (boost::dynamic_pointer_cast<kernel::activity::IoImpl>(actor->waiting_synchro) != nullptr)
- synchro_description = "I/O";
-
- XBT_INFO("Actor %ld (%s@%s): waiting for %s activity %p (%s) in state %d to finish", actor->get_pid(),
- actor->get_cname(), actor->get_host()->get_cname(), synchro_description, actor->waiting_synchro.get(),
- name, (int)actor->waiting_synchro->state_);
- } else {
- XBT_INFO("Actor %ld (%s@%s)", actor->get_pid(), actor->get_cname(), actor->get_host()->get_cname());
- }
- }
-}
-
-config::Flag<double> cfg_breakpoint{
- "debug/breakpoint", {"simix/breakpoint"}, "When non-negative, raise a SIGTRAP after given (simulated) time", -1.0};
-} // namespace simix
-} // namespace simgrid
-
-static simgrid::simix::ActorCode maestro_code;
+static simgrid::kernel::actor::ActorCode maestro_code;
void SIMIX_set_maestro(void (*code)(void*), void* data)
{
#ifdef _WIN32
- XBT_INFO("WARNING, SIMIX_set_maestro is believed to not work on windows. Please help us investigating this issue if you need that feature");
+ XBT_INFO("WARNING, SIMIX_set_maestro is believed to not work on windows. Please help us investigating this issue if "
+ "you need that feature");
#endif
maestro_code = std::bind(code, data);
}
-/**
- * @ingroup SIMIX_API
- * @brief Initialize SIMIX internal data.
- */
-void SIMIX_global_init(int *argc, char **argv)
+void SIMIX_global_init(int* argc, char** argv)
{
+ if (simix_global != nullptr)
+ return;
+
+ simix_global = std::make_unique<simgrid::simix::Global>();
+
#if SIMGRID_HAVE_MC
- // The communication initialization is done ASAP.
- // We need to communicate initialization of the different layers to the model-checker.
- simgrid::mc::Client::initialize();
+ // 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 (simix_global == nullptr) {
- surf_init(argc, argv); /* Initialize SURF structures */
+ surf_init(argc, argv); /* Initialize SURF structures */
- simix_global.reset(new simgrid::simix::Global());
- simix_global->maestro_ = nullptr;
- SIMIX_context_mod_init();
+ simix_global->maestro_ = nullptr;
+ SIMIX_context_mod_init();
- // Either create a new context with maestro or create
- // a context object with the current context maestro):
- simgrid::kernel::actor::create_maestro(maestro_code);
+ // Either create a new context with maestro or create
+ // a context object with the current context maestro):
+ simgrid::kernel::actor::create_maestro(maestro_code);
- /* Prepare to display some more info when dying on Ctrl-C pressing */
- std::signal(SIGINT, inthandler);
+ /* Prepare to display some more info when dying on Ctrl-C pressing */
+ std::signal(SIGINT, inthandler);
#ifndef _WIN32
- install_segvhandler();
+ install_segvhandler();
#endif
- /* register a function to be called by SURF after the environment creation */
- sg_platf_init();
- simgrid::s4u::Engine::on_platform_created.connect(surf_presolve);
-
- simgrid::s4u::Storage::on_creation.connect([](simgrid::s4u::Storage const& storage) {
- sg_storage_t s = simgrid::s4u::Storage::by_name(storage.get_name());
- xbt_assert(s != nullptr, "Storage not found for name %s", storage.get_cname());
- });
- }
+ /* register a function to be called by SURF after the environment creation */
+ sg_platf_init();
+ simgrid::s4u::Engine::on_platform_created.connect(surf_presolve);
if (simgrid::config::get_value<bool>("debug/clean-atexit"))
atexit(SIMIX_clean);
}
-int smx_cleaned = 0;
/**
* @ingroup SIMIX_API
* @brief Clean the SIMIX simulation
*/
void SIMIX_clean()
{
+ static bool smx_cleaned = false;
if (smx_cleaned)
return; // to avoid double cleaning by java and C
- smx_cleaned = 1;
+ smx_cleaned = true;
XBT_DEBUG("SIMIX_clean called. Simulation's over.");
- if (not simix_global->actors_to_run.empty() && SIMIX_get_clock() <= 0.0) {
+ auto* engine = simgrid::kernel::EngineImpl::get_instance();
+ if (engine->has_actors_to_run() && SIMIX_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.");
}
#if HAVE_SMPI
- if (simix_global->process_list.size() > 0) {
- if(smpi_process()->initialized()){
+ if (not engine->get_actor_list().empty()) {
+ if (smpi_process()->initialized()) {
xbt_die("Process exited without calling MPI_Finalize - Killing simulation");
- }else{
+ } else {
XBT_WARN("Process called exit when leaving - Skipping cleanups");
return;
}
/* Kill all processes (but maestro) */
simix_global->maestro_->kill_all();
- simix_global->run_all_actors();
- simix_global->empty_trash();
-
- /* Exit the SIMIX network module */
- SIMIX_mailbox_exit();
-
- while (not simgrid::simix::simix_timers.empty()) {
- delete simgrid::simix::simix_timers.top().second;
- simgrid::simix::simix_timers.pop();
- }
- /* Free the remaining data structures */
- simix_global->actors_to_run.clear();
- simix_global->actors_that_ran.clear();
- simix_global->actors_to_destroy.clear();
- simix_global->process_list.clear();
-
-#if SIMGRID_HAVE_MC
- xbt_dynar_free(&simix_global->actors_vector);
- xbt_dynar_free(&simix_global->dead_actors_vector);
-#endif
+ engine->run_all_actors();
+ engine->empty_trash();
/* Let's free maestro now */
delete simix_global->maestro_;
simix_global->maestro_ = nullptr;
/* Finish context module and SURF */
- SIMIX_context_mod_exit();
+ simix_global->destroy_context_factory();
surf_exit();
*/
double SIMIX_get_clock()
{
- if(MC_is_active() || MC_record_replay_is_active()){
+ if (MC_is_active() || MC_record_replay_is_active()) {
return MC_process_clock_get(SIMIX_process_self());
- }else{
+ } else {
return surf_get_clock();
}
}
-/** Handle any pending timer. Returns if something was actually run. */
-static bool SIMIX_execute_timers()
+void SIMIX_run() // XBT_ATTRIB_DEPRECATED_v332
{
- bool result = false;
- while (not simgrid::simix::simix_timers.empty() && SIMIX_get_clock() >= simgrid::simix::simix_timers.top().first) {
- result = true;
- // FIXME: make the timers being real callbacks (i.e. provide dispatchers that read and expand the args)
- smx_timer_t timer = simgrid::simix::simix_timers.top().second;
- simgrid::simix::simix_timers.pop();
- timer->callback();
- delete timer;
- }
- return result;
-}
-
-/**
- * @ingroup SIMIX_API
- * @brief Run the main simulation loop.
- */
-void SIMIX_run()
-{
- if (MC_record_replay_is_active()) {
- simgrid::mc::replay(MC_record_path);
- return;
- }
-
- double time = 0;
-
- do {
- XBT_DEBUG("New Schedule Round; size(queue)=%zu", simix_global->actors_to_run.size());
-
- if (simgrid::simix::cfg_breakpoint >= 0.0 && surf_get_clock() >= simgrid::simix::cfg_breakpoint) {
- XBT_DEBUG("Breakpoint reached (%g)", simgrid::simix::cfg_breakpoint.get());
- simgrid::simix::cfg_breakpoint = -1.0;
-#ifdef SIGTRAP
- std::raise(SIGTRAP);
-#else
- std::raise(SIGABRT);
-#endif
- }
-
- simix_global->execute_tasks();
-
- while (not simix_global->actors_to_run.empty()) {
- XBT_DEBUG("New Sub-Schedule Round; size(queue)=%zu", simix_global->actors_to_run.size());
-
- /* Run all processes that are ready to run, possibly in parallel */
- simix_global->run_all_actors();
-
- /* answer sequentially and in a fixed arbitrary order all the simcalls that were issued during that sub-round */
-
- /* WARNING, the order *must* be fixed or you'll jeopardize the simulation reproducibility (see RR-7653) */
-
- /* Here, the order is ok because:
- *
- * Short proof: only maestro adds stuff to the actors_to_run array, so the execution order of user contexts do
- * not impact its order.
- *
- * Long proof: actors remain sorted through an arbitrary (implicit, complex but fixed) order in all cases.
- *
- * - if there is no kill during the simulation, actors remain sorted according by their PID.
- * Rationale: This can be proved inductively.
- * Assume that actors_to_run is sorted at a beginning of one round (it is at round 0: the deployment file
- * is parsed linearly).
- * Let's show that it is still so at the end of this round.
- * - if an actor is added when being created, that's from maestro. It can be either at startup
- * time (and then in PID order), or in response to a process_create simcall. Since simcalls are handled
- * in arbitrary order (inductive hypothesis), we are fine.
- * - If an actor is added because it's getting killed, its subsequent actions shouldn't matter
- * - If an actor gets added to actors_to_run because one of their blocking action constituting the meat
- * of a simcall terminates, we're still good. Proof:
- * - 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().
- * 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).
- * The order then trivial holds since maestro didn't interrupt its handling of the simcall yet
- * - because the communication failed or were canceled after startup. In this case, it's called from
- * the function we are in, by the chunk:
- * set = model->states.failed_action_set;
- * while ((synchro = extract(set)))
- * SIMIX_simcall_post((smx_synchro_t) synchro->data);
- * This order is also fixed because it depends of the order in which the surf actions were
- * added to the system, and only maestro can add stuff this way, through simcalls.
- * We thus use the inductive hypothesis once again to conclude that the order in which synchros are
- * popped out of the set does not depend on the user code's execution order.
- * - because the communication terminated. In this case, synchros are served in the order given by
- * set = model->states.done_action_set;
- * while ((synchro = extract(set)))
- * SIMIX_simcall_post((smx_synchro_t) synchro->data);
- * and the argument is very similar to the previous one.
- * So, in any case, the orders of calls to CommImpl::finish() do not depend on the order in which user
- * actors are executed.
- * So, in any cases, the orders of actors within actors_to_run do not depend on the order in which
- * user actors were executed previously.
- * So, if there is no killing in the simulation, the simulation reproducibility is not jeopardized.
- * - If there is some actor killings, the order is changed by this decision that comes from user-land
- * But this decision may not have been motivated by a situation that were different because the simulation is
- * not reproducible.
- * So, even the order change induced by the actor killing is perfectly reproducible.
- *
- * So science works, bitches [http://xkcd.com/54/].
- *
- * We could sort the actors_that_ran array completely so that we can describe the order in which simcalls are
- * handled (like "according to the PID of issuer"), but it's not mandatory (order is fixed already even if
- * unfriendly).
- * That would thus be a pure waste of time.
- */
-
- for (auto const& actor : simix_global->actors_that_ran) {
- if (actor->simcall.call_ != SIMCALL_NONE) {
- actor->simcall_handle(0);
- }
- }
-
- simix_global->execute_tasks();
- do {
- simix_global->wake_all_waiting_actors();
- } while (simix_global->execute_tasks());
-
- /* If only daemon processes remain, cancel their actions, mark them to die and reschedule them */
- if (simix_global->process_list.size() == simix_global->daemons.size())
- for (auto const& dmon : simix_global->daemons) {
- XBT_DEBUG("Kill %s", dmon->get_cname());
- simix_global->maestro_->kill(dmon);
- }
- }
-
- time = simgrid::simix::Timer::next();
- if (time > -1.0 || not simix_global->process_list.empty()) {
- XBT_DEBUG("Calling surf_solve");
- time = surf_solve(time);
- XBT_DEBUG("Moving time ahead : %g", time);
- }
-
- /* Notify all the hosts that have failed */
- /* FIXME: iterate through the list of failed host and mark each of them */
- /* as failed. On each host, signal all the running processes with host_fail */
-
- // Execute timers and tasks until there isn't anything to be done:
- bool again = false;
- do {
- again = SIMIX_execute_timers();
- if (simix_global->execute_tasks())
- again = true;
- simix_global->wake_all_waiting_actors();
- } while (again);
-
- /* Clean actors to destroy */
- simix_global->empty_trash();
-
- XBT_DEBUG("### time %f, #processes %zu, #to_run %zu", time, simix_global->process_list.size(),
- simix_global->actors_to_run.size());
- } while (time > -1.0 || not simix_global->actors_to_run.empty());
-
- if (not simix_global->process_list.empty()) {
- if (simix_global->process_list.size() <= simix_global->daemons.size()) {
- XBT_CRITICAL("Oops! Daemon actors cannot do any blocking activity (communications, synchronization, etc) "
- "once the simulation is over. Please fix your on_exit() functions.");
- } else {
- XBT_CRITICAL("Oops! Deadlock or code not perfectly clean.");
- }
- simix_global->display_all_actor_status();
- simgrid::s4u::Engine::on_deadlock();
- xbt_abort();
- }
- simgrid::s4u::Engine::on_simulation_end();
+ simgrid::kernel::EngineImpl::get_instance()->run();
}
double SIMIX_timer_next() // XBT_ATTRIB_DEPRECATED_v329
{
- return simgrid::simix::Timer::next();
+ return simgrid::kernel::timer::Timer::next();
}
smx_timer_t SIMIX_timer_set(double date, void (*callback)(void*), void* arg) // XBT_ATTRIB_DEPRECATED_v329
{
- return simgrid::simix::Timer::set(date, std::bind(callback, arg));
+ return simgrid::kernel::timer::Timer::set(date, std::bind(callback, arg));
}
/** @brief cancels a timer that was added earlier */
/** @brief Returns the date at which the timer will trigger (or 0 if nullptr timer) */
double SIMIX_timer_get_date(smx_timer_t timer) // XBT_ATTRIB_DEPRECATED_v329
{
- return timer ? timer->get_date() : 0;
+ return timer ? timer->get_date() : 0.0;
}
void SIMIX_display_process_status() // XBT_ATTRIB_DEPRECATED_v329
{
- simix_global->display_all_actor_status();
+ simgrid::kernel::EngineImpl::get_instance()->display_all_actor_status();
}
int SIMIX_is_maestro()
{
if (simix_global == nullptr) // SimDag
return true;
- simgrid::kernel::actor::ActorImpl* self = SIMIX_process_self();
+ const simgrid::kernel::actor::ActorImpl* self = SIMIX_process_self();
return self == nullptr || self == simix_global->maestro_;
}