/***************************************************************************** * Product: QF/C port to POSIX/P-threads, GNU * Last Updated for Version: 5.2.1 * Date of the Last Update: Jan 06, 2014 * * Q u a n t u m L e a P s * --------------------------- * innovating embedded systems * * Copyright (C) 2002-2014 Quantum Leaps, LLC. All rights reserved. * * This program is open source software: you can redistribute it and/or * modify it under the terms of the GNU General Public License as published * by the Free Software Foundation, either version 3 of the License, or * (at your option) any later version. * * Alternatively, this program may be distributed and modified under the * terms of Quantum Leaps commercial licenses, which expressly supersede * the GNU General Public License and are specifically designed for * licensees interested in retaining the proprietary status of their code. * * This program is distributed in the hope that it will be useful, * but WITHOUT ANY WARRANTY; without even the implied warranty of * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the * GNU General Public License for more details. * * You should have received a copy of the GNU General Public License * along with this program. If not, see . * * Contact information: * Quantum Leaps Web sites: http://www.quantum-leaps.com * http://www.state-machine.com * e-mail: info@quantum-leaps.com *****************************************************************************/ #include "qf_pkg.h" #include "qassert.h" #include /* for PTHREAD_STACK_MIN */ #include /* for mlockall() */ #include Q_DEFINE_THIS_MODULE("qf_port") /* Global objects ----------------------------------------------------------*/ pthread_mutex_t QF_pThreadMutex_ = PTHREAD_MUTEX_INITIALIZER; /* Local objects -----------------------------------------------------------*/ static long int l_tickUsec = 10000UL; /* clock tick in usec (for tv_usec) */ static int_t l_running; /*..........................................................................*/ void QF_init(void) { /* lock memory so we're never swapped out to disk */ /*mlockall(MCL_CURRENT | MCL_FUTURE); uncomment when supported */ } /*..........................................................................*/ int_t QF_run(void) { struct sched_param sparam; struct timeval timeout = { 0 }; /* timeout for select() */ QF_onStartup(); /* invoke startup callback */ /* try to maximize the priority of the ticker thread, see NOTE01 */ sparam.sched_priority = sched_get_priority_max(SCHED_FIFO); if (pthread_setschedparam(pthread_self(), SCHED_FIFO, &sparam) == 0) { /* success, this application has sufficient privileges */ } else { /* setting priority failed, probably due to insufficient privieges */ } l_running = (int_t)1; while (l_running) { QF_onClockTick(); /* clock tick callback (must call QF_TICK_X()) */ timeout.tv_usec = l_tickUsec; /* set the desired tick interval */ select(0, 0, 0, 0, &timeout); /* sleep for the desired tick, NOTE05 */ } QF_onCleanup(); /* invoke cleanup callback */ pthread_mutex_destroy(&QF_pThreadMutex_); return (int_t)0; /* return success */ } /*..........................................................................*/ void QF_setTickRate(uint32_t ticksPerSec) { l_tickUsec = 1000000UL / ticksPerSec; } /*..........................................................................*/ void QF_stop(void) { l_running = (int_t)0; /* stop the loop in QF_run() */ } /*..........................................................................*/ static void *thread_routine(void *arg) { /* the expected POSIX signature */ QActive *act = (QActive *)arg; do { /* loop until m_thread is cleared in QActive_stop() */ QEvt const *e = QActive_get_(act); /* wait for the event */ QMSM_DISPATCH(&act->super, e); /* dispatch to the SM */ QF_gc(e); /* check if the event is garbage, and collect it if so */ } while (act->thread != (uint8_t)0); QF_remove_(act); /* remove this object from any subscriptions */ pthread_cond_destroy(&act->osObject); /* cleanup the condition variable */ return (void *)0; /* return success */ } /*..........................................................................*/ void QActive_start_(QActive *me, uint_t prio, QEvt const *qSto[], uint_t qLen, void *stkSto, uint_t stkSize, QEvt const *ie) { pthread_t thread; pthread_attr_t attr; struct sched_param param; Q_REQUIRE(stkSto == (void *)0); /* p-threads allocate stack internally */ QEQueue_init(&me->eQueue, qSto, qLen); pthread_cond_init(&me->osObject, 0); me->prio = (uint8_t)prio; QF_add_(me); /* make QF aware of this active object */ QMSM_INIT(&me->super, ie); /* execute the initial transition */ pthread_attr_init(&attr); /* SCHED_FIFO corresponds to real-time preemptive priority-based scheduler * NOTE: This scheduling policy requires the superuser privileges */ pthread_attr_setschedpolicy(&attr, SCHED_FIFO); /* see NOTE04 */ param.sched_priority = prio + (sched_get_priority_max(SCHED_FIFO) - QF_MAX_ACTIVE - 3); pthread_attr_setschedparam(&attr, ¶m); pthread_attr_setdetachstate(&attr, PTHREAD_CREATE_DETACHED); if (stkSize == 0U) { /* stack size not provided? */ stkSize = (uint_t)PTHREAD_STACK_MIN; /* set the allowed minimum */ } pthread_attr_setstacksize(&attr, (size_t)stkSize); if (pthread_create(&thread, &attr, &thread_routine, me) != 0) { /* Creating the p-thread with the SCHED_FIFO policy failed. Most * probably this application has no superuser privileges, so we just * fall back to the default SCHED_OTHER policy and priority 0. */ pthread_attr_setschedpolicy(&attr, SCHED_OTHER); param.sched_priority = 0; pthread_attr_setschedparam(&attr, ¶m); Q_ALLEGE(pthread_create(&thread, &attr, &thread_routine, me)== 0); } pthread_attr_destroy(&attr); me->thread = (uint8_t)1; } /*..........................................................................*/ void QActive_stop(QActive *me) { me->thread = (uint8_t)0; /* stop the QActive thread loop */ } /***************************************************************************** * NOTE01: * In Linux, the scheduler policy closest to real-time is the SCHED_FIFO * policy, available only with superuser privileges. QF_run() attempts to set * this policy as well as to maximize its priority, so that the ticking * occurrs in the most timely manner (as close to an interrupt as possible). * However, setting the SCHED_FIFO policy might fail, most probably due to * insufficient privileges. * * NOTE02: * On some Linux systems nanosleep() might actually not deliver the finest * time granularity. For example, on some Linux implementations, nanosleep() * could not block for shorter intervals than 20ms, while the underlying * clock tick period was only 10ms. Sometimes, the select() system call can * provide a finer granularity. * * NOTE03: * Any blocking system call, such as nanosleep() or select() system call can * be interrupted by a signal, such as ^C from the keyboard. In this case this * QF port breaks out of the event-loop and returns to main() that exits and * terminates all spawned p-threads. * * NOTE04: * According to the man pages (for pthread_attr_setschedpolicy) the only value * supported in the Linux p-threads implementation is PTHREAD_SCOPE_SYSTEM, * meaning that the threads contend for CPU time with all processes running on * the machine. In particular, thread priorities are interpreted relative to * the priorities of all other processes on the machine. * * This is good, because it seems that if we set the priorities high enough, * no other process (or thread running within) can gain control over the CPU. * * However, QF limits the number of priority levels to QF_MAX_ACTIVE. * Assuming that a QF application will be real-time, this port reserves the * three highest Linux priorities for the ISR-like threads (e.g., the ticker, * I/O), and the rest highest-priorities for the active objects. * * NOTE05: * The select() system call seems to deliver the finest time granularity of * 1 clock tick. The timeout value passed to select() is rounded up to the * nearest tick (10 ms on desktop Linux). The timeout cannot be too short, * because the system might choose to busy-wait for very short timeouts. * An alternative, POSIX nanosleep() system call seems to deliver only 20ms * granularity. * * Here the select() call is used not just as a fairly portable way to sleep * with subsecond precision. The select() call is also used to detect any * characters typed on the console. * * Also according to man pages, on Linux, the function select() modifies * timeout to reflect the amount of time not slept; most other implementations * do not do this. This causes problems both when Linux code which reads * timeout is ported to other operating systems, and when code is ported to * Linux that reuses a struct timeval for multiple selects in a loop without * reinitializing it. Here the microsecond part of the structure is re- * initialized before each select() call. */