/* 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 <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 "mc/mc.h"
-#include "simgrid/Exception.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 */
-XBT_LOG_NEW_DEFAULT_CATEGORY(ker_engine, "Logging specific to Engine (kernel)");
+#include <boost/algorithm/string/predicate.hpp>
+#ifndef _WIN32
+#include <dlfcn.h>
+#endif /* _WIN32 */
+#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<double> cfg_breakpoint{"debug/breakpoint",
"When non-negative, raise a SIGTRAP after given (simulated) time", -1.0};
+config::Flag<bool> 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
+}
+
+} // 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
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;
+
+ /* Free the remaining data structures */
+#if SIMGRID_HAVE_MC
+ xbt_dynar_free(&actors_vector_);
+ xbt_dynar_free(&dead_actors_vector_);
+#endif
+ /* clear models before freeing handle, network models can use external callback defined in the handle */
+ models_prio_.clear();
+}
+
+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<bool>("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)
+{
+ double start = xbt_os_time();
+ if (boost::algorithm::ends_with(platf, ".so") or boost::algorithm::ends_with(platf, ".dylib")) {
+#ifdef _WIN32
+ xbt_die("loading platform through shared library isn't supported on windows");
+#else
+ 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<void, std::function<int(void*)>>(handle, dlclose);
+ using load_fct_t = void (*)(const simgrid::s4u::Engine&);
+ 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());
+#endif /* _WIN32 */
+ } else {
+ parse_platform_file(platf);
+ }
+
+ double end = xbt_os_time();
+ XBT_DEBUG("PARSE TIME: %g", (end - start));
}
void EngineImpl::load_deployment(const std::string& file) const
models_prio_[model_name] = std::move(model);
}
+void EngineImpl::add_split_duplex_link(const std::string& name, std::unique_ptr<resource::SplitDuplexLinkImpl> 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
+{
+ for (auto const& model : models_) {
+ XBT_DEBUG("Handling the failed actions (if any)");
+ while (auto* action = model->extract_failed_action()) {
+ XBT_DEBUG(" Handling Action %p", action);
+ if (action->get_activity() != nullptr)
+ activity::ActivityImplPtr(action->get_activity())->post();
+ }
+ XBT_DEBUG("Handling the terminated actions (if any)");
+ while (auto* 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
+ activity::ActivityImplPtr(action->get_activity())->post();
+ }
+ }
+}
+/**
+ * @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 EngineImpl::run_all_actors()
+{
+ instance_->get_context_factory()->run_all();
+
+ actors_to_run_.swap(actors_that_ran_);
+ actors_to_run_.clear();
+}
+
+actor::ActorImpl* EngineImpl::get_actor_by_pid(aid_t pid)
+{
+ auto item = actor_list_.find(pid);
+ if (item != actor_list_.end())
+ return item->second;
+
+ // Search the trash
+ for (auto& a : actors_to_destroy_)
+ if (a.get_pid() == 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()
+{
+ if (tasks.empty())
+ return false;
+
+ std::vector<xbt::Task<void()>> tasksTemp;
+ 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 EngineImpl::remove_daemon(actor::ActorImpl* actor)
+{
+ auto it = daemons_.find(actor);
+ xbt_assert(it != daemons_.end(), "The dying daemon is not a daemon after all. Please report that bug.");
+ daemons_.erase(it);
+}
+
+void EngineImpl::add_actor_to_run_list_no_check(actor::ActorImpl* actor)
+{
+ XBT_DEBUG("Inserting [%p] %s(%s) in the to_run list", actor, actor->get_cname(), actor->get_host()->get_cname());
+ actors_to_run_.push_back(actor);
+}
+
+void EngineImpl::add_actor_to_run_list(actor::ActorImpl* actor)
+{
+ if (std::find(begin(actors_to_run_), end(actors_to_run_), actor) != end(actors_to_run_)) {
+ XBT_DEBUG("Actor %s is already in the to_run list", actor->get_cname());
+ } else {
+ XBT_DEBUG("Inserting [%p] %s(%s) in the to_run list", actor, actor->get_cname(), actor->get_host()->get_cname());
+ actors_to_run_.push_back(actor);
+ }
+}
+void EngineImpl::empty_trash()
+{
+ while (not actors_to_destroy_.empty()) {
+ 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
+}
+
+void EngineImpl::display_all_actor_status() const
+{
+ XBT_INFO("%zu actors are still running, waiting for something.", actor_list_.size());
+ /* List the actors and their state */
+ XBT_INFO("Legend of the following listing: \"Actor <pid> (<name>@<host>): <status>\"");
+ for (auto const& kv : actor_list_) {
+ actor::ActorImpl* actor = kv.second;
+
+ if (actor->waiting_synchro_) {
+ const char* synchro_description = "unknown";
+
+ 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 %#zx (%s) in state %d to finish", actor->get_pid(),
+ actor->get_cname(), actor->get_host()->get_cname(), synchro_description,
+ (xbt_log_no_loc ? (size_t)0xDEADBEEF : (size_t)actor->waiting_synchro_.get()),
+ actor->waiting_synchro_->get_cname(), (int)actor->waiting_synchro_->state_);
+ } else {
+ XBT_INFO("Actor %ld (%s@%s) simcall %s", actor->get_pid(), actor->get_cname(), actor->get_host()->get_cname(),
+ SIMIX_simcall_name(actor->simcall_));
+ }
+ }
+}
+
void EngineImpl::run()
{
if (MC_record_replay_is_active()) {
mc::replay(MC_record_path());
+ empty_trash();
return;
}
double time = 0;
do {
- XBT_DEBUG("New Schedule Round; size(queue)=%zu", simix_global->actors_to_run.size());
+ 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
#endif
}
- simix_global->execute_tasks();
+ 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());
+ while (not actors_to_run_.empty()) {
+ XBT_DEBUG("New Sub-Schedule Round; size(queue)=%zu", actors_to_run_.size());
- /* Run all processes that are ready to run, possibly in parallel */
- simix_global->run_all_actors();
+ /* Run all actors that are ready to run, possibly in parallel */
+ run_all_actors();
/* answer sequentially and in a fixed arbitrary order all the simcalls that were issued during that sub-round */
* That would thus be a pure waste of time.
*/
- for (auto const& actor : simix_global->actors_that_ran) {
+ for (auto const& actor : actors_that_ran_) {
if (actor->simcall_.call_ != simix::Simcall::NONE) {
actor->simcall_handle(0);
}
}
- simix_global->execute_tasks();
+ execute_tasks();
do {
- simix_global->wake_all_waiting_actors();
- } while (simix_global->execute_tasks());
+ wake_all_waiting_actors();
+ } while (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) {
+ /* If only daemon actors remain, cancel their actions, mark them to die and reschedule them */
+ if (actor_list_.size() == daemons_.size())
+ for (auto const& dmon : daemons_) {
XBT_DEBUG("Kill %s", dmon->get_cname());
- simix_global->maestro_->kill(dmon);
+ maestro_->kill(dmon);
}
}
time = timer::Timer::next();
- if (time > -1.0 || not simix_global->process_list.empty()) {
+ if (time > -1.0 || not actor_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 */
+ /* as failed. On each host, signal all the running actors with host_fail */
// Execute timers and tasks until there isn't anything to be done:
bool again = false;
do {
again = timer::Timer::execute_all();
- if (simix_global->execute_tasks())
+ if (execute_tasks())
again = true;
- simix_global->wake_all_waiting_actors();
+ wake_all_waiting_actors();
} while (again);
/* Clean actors to destroy */
- simix_global->empty_trash();
+ empty_trash();
- XBT_DEBUG("### time %f, #processes %zu, #to_run %zu", time, simix_global->process_list.size(),
- simix_global->actors_to_run.size());
+ XBT_DEBUG("### time %f, #actors %zu, #to_run %zu", time, actor_list_.size(), actors_to_run_.size());
- if (time < 0. && simix_global->actors_to_run.empty() && not simix_global->process_list.empty()) {
- if (simix_global->process_list.size() <= simix_global->daemons.size()) {
+ if (time < 0. && actors_to_run_.empty() && not actor_list_.empty()) {
+ if (actor_list_.size() <= 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();
+ display_all_actor_status();
simgrid::s4u::Engine::on_deadlock();
- for (auto const& kv : simix_global->process_list) {
+ for (auto const& kv : actor_list_) {
XBT_DEBUG("Kill %s", kv.second->get_cname());
- simix_global->maestro_->kill(kv.second);
+ maestro_->kill(kv.second);
}
}
- } while (time > -1.0 || not simix_global->actors_to_run.empty());
+ } while (time > -1.0 || has_actors_to_run());
- if (not simix_global->process_list.empty())
+ if (not actor_list_.empty())
THROW_IMPOSSIBLE;
simgrid::s4u::Engine::on_simulation_end();
}
} // namespace kernel
} // namespace simgrid
+
+void SIMIX_run() // XBT_ATTRIB_DEPRECATED_v332
+{
+ simgrid::kernel::EngineImpl::get_instance()->run();
+}