X-Git-Url: http://info.iut-bm.univ-fcomte.fr/pub/gitweb/simgrid.git/blobdiff_plain/3ba5d4a966c2ac976b2b1656c52145f70189d108..d53e7db348cf100819ca9b3bc8f30406dd5077ea:/src/simix/smx_global.cpp diff --git a/src/simix/smx_global.cpp b/src/simix/smx_global.cpp index 39c1c41507..fb5a7c54b5 100644 --- a/src/simix/smx_global.cpp +++ b/src/simix/smx_global.cpp @@ -3,29 +3,32 @@ /* 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 #include #include -#include /* Signal handling */ -#include #include "src/internal_config.h" +#include /* Signal handling */ +#include +#include #include +#include #include "simgrid/s4u/Engine.hpp" #include "simgrid/s4u/Host.hpp" +#include "smx_private.hpp" #include "src/surf/surf_interface.hpp" #include "src/surf/xml/platf.hpp" -#include "smx_private.h" -#include "xbt/ex.h" /* ex_backtrace_display */ +#include "xbt/ex.h" /* ex_backtrace_display */ #include "mc/mc.h" #include "simgrid/sg_config.h" -#include "src/mc/mc_replay.h" +#include "src/mc/mc_replay.hpp" #include "src/surf/StorageImpl.hpp" -#include "src/smpi/smpi_process.hpp" +#include "src/smpi/include/smpi_process.hpp" #include "src/kernel/activity/CommImpl.hpp" #include "src/kernel/activity/ExecImpl.hpp" @@ -34,37 +37,42 @@ #include "src/kernel/activity/SynchroRaw.hpp" #if SIMGRID_HAVE_MC -#include "src/mc/mc_private.h" +#include "src/mc/mc_private.hpp" #include "src/mc/remote/Client.hpp" #include "src/mc/remote/mc_protocol.h" -#include #endif -#include "src/mc/mc_record.h" +#include "src/mc/mc_record.hpp" #if HAVE_SMPI -#include "src/smpi/private.h" +#include "src/smpi/include/private.hpp" #endif XBT_LOG_NEW_CATEGORY(simix, "All SIMIX categories"); XBT_LOG_NEW_DEFAULT_SUBCATEGORY(simix_kernel, simix, "Logging specific to SIMIX (kernel)"); std::unique_ptr simix_global; -static xbt_heap_t simix_timers = nullptr; + +namespace { +typedef std::pair TimerQelt; +boost::heap::fibonacci_heap>> simix_timers; +} /** @brief Timer datatype */ -typedef struct s_smx_timer { +class s_smx_timer_t { double date = 0.0; - simgrid::xbt::Task callback; - s_smx_timer()=default; - s_smx_timer(double date, simgrid::xbt::Task callback) : date(date), callback(std::move(callback)) {} -} s_smx_timer_t; +public: + decltype(simix_timers)::handle_type handle_; + simgrid::xbt::Task callback; + double getDate() { return date; } + s_smx_timer_t(double date, simgrid::xbt::Task callback) : date(date), callback(std::move(callback)) {} +}; void (*SMPI_switch_data_segment)(int) = nullptr; int _sg_do_verbose_exit = 1; -static void inthandler(int ignored) +static void inthandler(int) { if ( _sg_do_verbose_exit ) { XBT_INFO("CTRL-C pressed. The current status will be displayed before exit (disable that behavior with option 'verbose-exit')."); @@ -77,13 +85,13 @@ static void inthandler(int ignored) } #ifndef _WIN32 -static void segvhandler(int signum, siginfo_t *siginfo, void *context) +static void segvhandler(int signum, siginfo_t* siginfo, void* /*context*/) { if (siginfo->si_signo == SIGSEGV && siginfo->si_code == SEGV_ACCERR) { fprintf(stderr, "Access violation 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 %d KiB).\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" @@ -127,8 +135,8 @@ static void install_segvhandler() return; } if (not(old_stack.ss_flags & SS_DISABLE)) { - XBT_DEBUG("An alternate stack was already installed (sp=%p, size=%zd, flags=%x). Restore it.", - old_stack.ss_sp, old_stack.ss_size, old_stack.ss_flags); + 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); } @@ -154,7 +162,7 @@ static void install_segvhandler() /********************************* SIMIX **************************************/ double SIMIX_timer_next() { - return xbt_heap_size(simix_timers) > 0 ? xbt_heap_maxkey(simix_timers) : -1.0; + return simix_timers.empty() ? -1.0 : simix_timers.top().first; } static void kill_process(smx_actor_t process) @@ -202,11 +210,6 @@ void SIMIX_global_init(int *argc, char **argv) if (not simix_global) { simix_global = std::unique_ptr(new simgrid::simix::Global()); - - simgrid::simix::ActorImpl proc; - simix_global->process_to_run = xbt_dynar_new(sizeof(smx_actor_t), nullptr); - simix_global->process_that_ran = xbt_dynar_new(sizeof(smx_actor_t), nullptr); - simix_global->process_to_destroy = xbt_swag_new(xbt_swag_offset(proc, destroy_hookup)); simix_global->maestro_process = nullptr; simix_global->create_process_function = &SIMIX_process_create; simix_global->kill_process_function = &kill_process; @@ -235,16 +238,11 @@ void SIMIX_global_init(int *argc, char **argv) }); simgrid::surf::storageCreatedCallbacks.connect([](simgrid::surf::StorageImpl* storage) { - sg_storage_t s = simgrid::s4u::Storage::byName(storage->cname()); - xbt_assert(s != nullptr, "Storage not found for name %s", storage->cname()); + sg_storage_t s = simgrid::s4u::Storage::byName(storage->getCname()); + xbt_assert(s != nullptr, "Storage not found for name %s", storage->getCname()); }); } - if (not simix_timers) - simix_timers = xbt_heap_new(8, [](void* p) { - delete static_cast(p); - }); - if (xbt_cfg_get_boolean("clean-atexit")) atexit(SIMIX_clean); @@ -264,6 +262,15 @@ void SIMIX_clean() if (smx_cleaned) return; // to avoid double cleaning by java and C + smx_cleaned = 1; + XBT_DEBUG("SIMIX_clean called. Simulation's over."); + if (not simix_global->process_to_run.empty() && 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."); + xbt_die("Bailing out to avoid that stop-before-start madness. Please fix your code."); + } + #if HAVE_SMPI if (SIMIX_process_count()>0){ if(smpi_process()->initialized()){ @@ -275,33 +282,29 @@ void SIMIX_clean() } #endif - smx_cleaned = 1; - XBT_DEBUG("SIMIX_clean called. Simulation's over."); - if (not xbt_dynar_is_empty(simix_global->process_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."); - xbt_die("Bailing out to avoid that stop-before-start madness. Please fix your code."); - } /* Kill all processes (but maestro) */ SIMIX_process_killall(simix_global->maestro_process, 1); + SIMIX_context_runall(); + SIMIX_process_empty_trash(); /* Exit the SIMIX network module */ SIMIX_mailbox_exit(); - xbt_heap_free(simix_timers); - simix_timers = nullptr; + while (not simix_timers.empty()) { + delete simix_timers.top().second; + simix_timers.pop(); + } /* Free the remaining data structures */ - xbt_dynar_free(&simix_global->process_to_run); - xbt_dynar_free(&simix_global->process_that_ran); - xbt_swag_free(simix_global->process_to_destroy); + simix_global->process_to_run.clear(); + simix_global->process_that_ran.clear(); + simix_global->process_to_destroy.clear(); simix_global->process_list.clear(); - simix_global->process_to_destroy = nullptr; xbt_os_mutex_destroy(simix_global->mutex); simix_global->mutex = nullptr; #if SIMGRID_HAVE_MC xbt_dynar_free(&simix_global->actors_vector); + xbt_dynar_free(&simix_global->dead_actors_vector); #endif /* Let's free maestro now */ @@ -334,25 +337,12 @@ double SIMIX_get_clock() } } -static int process_syscall_color(void *p) -{ - switch ((*(smx_actor_t *)p)->simcall.call) { - case SIMCALL_NONE: - case SIMCALL_PROCESS_KILL: - return 2; - case SIMCALL_PROCESS_RESUME: - return 1; - default: - return 0; - } -} - /** Wake up all processes waiting for a Surf action to finish */ static void SIMIX_wake_processes() { surf_action_t action; - for(auto model : *all_existing_models) { + for (auto const& model : *all_existing_models) { XBT_DEBUG("Handling the processes whose action failed (if any)"); while ((action = surf_model_extract_failed_action_set(model))) { XBT_DEBUG(" Handling Action %p",action); @@ -373,18 +363,18 @@ static void SIMIX_wake_processes() static bool SIMIX_execute_timers() { bool result = false; - while (xbt_heap_size(simix_timers) > 0 && SIMIX_get_clock() >= SIMIX_timer_next()) { + while (not simix_timers.empty() && SIMIX_get_clock() >= 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 = (smx_timer_t) xbt_heap_pop(simix_timers); - try { - timer->callback(); - } - catch(...) { - xbt_die("Exception throwed ouf of timer callback"); - } - delete timer; + // FIXME: make the timers being real callbacks + // (i.e. provide dispatchers that read and expand the args) + smx_timer_t timer = simix_timers.top().second; + simix_timers.pop(); + try { + timer->callback(); + } catch (...) { + xbt_die("Exception thrown ouf of timer callback"); + } + delete timer; } return result; } @@ -421,7 +411,7 @@ static bool SIMIX_execute_tasks() */ void SIMIX_run() { - if (MC_record_path) { + if (not MC_record_path.empty()) { simgrid::mc::replay(MC_record_path); return; } @@ -429,32 +419,31 @@ void SIMIX_run() double time = 0; do { - XBT_DEBUG("New Schedule Round; size(queue)=%lu", xbt_dynar_length(simix_global->process_to_run)); + XBT_DEBUG("New Schedule Round; size(queue)=%zu", simix_global->process_to_run.size()); SIMIX_execute_tasks(); - while (not xbt_dynar_is_empty(simix_global->process_to_run)) { - XBT_DEBUG("New Sub-Schedule Round; size(queue)=%lu", xbt_dynar_length(simix_global->process_to_run)); + while (not simix_global->process_to_run.empty()) { + XBT_DEBUG("New Sub-Schedule Round; size(queue)=%zu", simix_global->process_to_run.size()); /* Run all processes that are ready to run, possibly in parallel */ SIMIX_process_runall(); - /* Move all killer processes to the end of the list, because killing a process that have an ongoing simcall is a bad idea */ - xbt_dynar_three_way_partition(simix_global->process_that_ran, process_syscall_color); - /* 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 process_to_run array, so the execution order of user contexts do not impact its order. + * Short proof: only maestro adds stuff to the process_to_run array, so the execution order of user contexts do + * not impact its order. * * Long proof: processes remain sorted through an arbitrary (implicit, complex but fixed) order in all cases. * * - if there is no kill during the simulation, processes remain sorted according by their PID. * rational: This can be proved inductively. - * Assume that process_to_run is sorted at a beginning of one round (it is at round 0: the deployment file is parsed linearly). + * Assume that process_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 a process 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 @@ -463,43 +452,46 @@ void SIMIX_run() * - If a process gets added to process_to_run because one of their blocking action constituting the meat * of a simcall terminates, we're still good. Proof: * - You are added from SIMIX_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. + * 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: + * - 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 = xbt_swag_extract(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 - * poped out of the swag does not depend on the user code's execution order. + * poped 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 = xbt_swag_extract(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 SIMIX_comm_finish() do not depend on the order in which user processes are executed. - * So, in any cases, the orders of processes within process_to_run do not depend on the order in which user processes were executed previously. + * So, in any case, the orders of calls to SIMIX_comm_finish() do not depend on the order in which user + * processes are executed. + * So, in any cases, the orders of processes within process_to_run do not depend on the order in which + * user processes were executed previously. * So, if there is no killing in the simulation, the simulation reproducibility is not jeopardized. * - If there is some process 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. + * 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 process killing is perfectly reproducible. * * So science works, bitches [http://xkcd.com/54/]. * - * We could sort the process_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). + * We could sort the process_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. */ - unsigned int iter; - smx_actor_t process; - xbt_dynar_foreach(simix_global->process_that_ran, iter, process) { + for (smx_actor_t const& process : simix_global->process_that_ran) { if (process->simcall.call != SIMCALL_NONE) { SIMIX_simcall_handle(&process->simcall, 0); } @@ -512,14 +504,14 @@ void SIMIX_run() /* 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 (const auto& dmon : simix_global->daemons) { - XBT_DEBUG("Kill %s", dmon->cname()); + for (auto const& dmon : simix_global->daemons) { + XBT_DEBUG("Kill %s", dmon->getCname()); SIMIX_process_kill(dmon, simix_global->maestro_process); } } time = SIMIX_timer_next(); - if (time > -1.0 || simix_global->process_list.empty() == false) { + 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); @@ -539,8 +531,8 @@ void SIMIX_run() } while (again); /* Autorestart all process */ - for (auto host: host_that_restart) { - XBT_INFO("Restart processes on host %s", host->cname()); + for (auto const& host : host_that_restart) { + XBT_INFO("Restart processes on host %s", host->getCname()); SIMIX_host_autorestart(host); } host_that_restart.clear(); @@ -548,15 +540,15 @@ void SIMIX_run() /* Clean processes to destroy */ SIMIX_process_empty_trash(); - XBT_DEBUG("### time %f, #processes %zu, #to_run %lu", time, simix_global->process_list.size(), - xbt_dynar_length(simix_global->process_to_run)); + XBT_DEBUG("### time %f, #processes %zu, #to_run %zu", time, simix_global->process_list.size(), + simix_global->process_to_run.size()); - if (xbt_dynar_is_empty(simix_global->process_to_run) && not simix_global->process_list.empty()) + if (simix_global->process_to_run.empty() && not simix_global->process_list.empty()) simgrid::simix::onDeadlock(); - } while (time > -1.0 || not xbt_dynar_is_empty(simix_global->process_to_run)); + } while (time > -1.0 || not simix_global->process_to_run.empty()); - if (simix_global->process_list.size() != 0) { + if (not simix_global->process_list.empty()) { TRACE_end(); @@ -578,26 +570,27 @@ void SIMIX_run() */ smx_timer_t SIMIX_timer_set(double date, void (*callback)(void*), void *arg) { - smx_timer_t timer = new s_smx_timer_t(date, [=](){ callback(arg); }); - xbt_heap_push(simix_timers, timer, date); + smx_timer_t timer = new s_smx_timer_t(date, [callback, arg]() { callback(arg); }); + timer->handle_ = simix_timers.emplace(std::make_pair(date, timer)); return timer; } smx_timer_t SIMIX_timer_set(double date, simgrid::xbt::Task callback) { smx_timer_t timer = new s_smx_timer_t(date, std::move(callback)); - xbt_heap_push(simix_timers, timer, date); + timer->handle_ = simix_timers.emplace(std::make_pair(date, timer)); return timer; } /** @brief cancels a timer that was added earlier */ void SIMIX_timer_remove(smx_timer_t timer) { - xbt_heap_rm_elm(simix_timers, timer, timer->date); + simix_timers.erase(timer->handle_); + delete timer; } /** @brief Returns the date at which the timer will trigger (or 0 if nullptr timer) */ double SIMIX_timer_get_date(smx_timer_t timer) { - return timer?timer->date:0; + return timer ? timer->getDate() : 0; } /** @@ -646,7 +639,7 @@ void SIMIX_display_process_status() XBT_INFO("%d processes are still running, waiting for something.", nbprocess); /* List the process and their state */ XBT_INFO("Legend of the following listing: \"Process (@): \""); - for (auto kv : simix_global->process_list) { + for (auto const& kv : simix_global->process_list) { smx_actor_t process = kv.second; if (process->waiting_synchro) { @@ -668,24 +661,12 @@ void SIMIX_display_process_status() if (boost::dynamic_pointer_cast(process->waiting_synchro) != nullptr) synchro_description = "I/O"; - - /* - switch (process->waiting_synchro->type) { - case SIMIX_SYNC_PARALLEL_EXECUTE: - synchro_description = "parallel execution"; - break; - - case SIMIX_SYNC_JOIN: - synchro_description = "joining"; - break; -*/ - XBT_INFO("Process %lu (%s@%s): waiting for %s synchro %p (%s) in state %d to finish", process->pid, - process->cname(), process->host->cname(), synchro_description, process->waiting_synchro.get(), + process->getCname(), process->host->getCname(), synchro_description, process->waiting_synchro.get(), process->waiting_synchro->name.c_str(), (int)process->waiting_synchro->state); } else { - XBT_INFO("Process %lu (%s@%s)", process->pid, process->cname(), process->host->cname()); + XBT_INFO("Process %lu (%s@%s)", process->pid, process->getCname(), process->host->getCname()); } } }