1*1da177e4SLinus Torvalds /* 2*1da177e4SLinus Torvalds * hangcheck-timer.c 3*1da177e4SLinus Torvalds * 4*1da177e4SLinus Torvalds * Driver for a little io fencing timer. 5*1da177e4SLinus Torvalds * 6*1da177e4SLinus Torvalds * Copyright (C) 2002 Oracle Corporation. All rights reserved. 7*1da177e4SLinus Torvalds * 8*1da177e4SLinus Torvalds * Author: Joel Becker <joel.becker@oracle.com> 9*1da177e4SLinus Torvalds * 10*1da177e4SLinus Torvalds * This program is free software; you can redistribute it and/or 11*1da177e4SLinus Torvalds * modify it under the terms of the GNU General Public 12*1da177e4SLinus Torvalds * License version 2 as published by the Free Software Foundation. 13*1da177e4SLinus Torvalds * 14*1da177e4SLinus Torvalds * This program is distributed in the hope that it will be useful, 15*1da177e4SLinus Torvalds * but WITHOUT ANY WARRANTY; without even the implied warranty of 16*1da177e4SLinus Torvalds * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU 17*1da177e4SLinus Torvalds * General Public License for more details. 18*1da177e4SLinus Torvalds * 19*1da177e4SLinus Torvalds * You should have received a copy of the GNU General Public 20*1da177e4SLinus Torvalds * License along with this program; if not, write to the 21*1da177e4SLinus Torvalds * Free Software Foundation, Inc., 59 Temple Place - Suite 330, 22*1da177e4SLinus Torvalds * Boston, MA 021110-1307, USA. 23*1da177e4SLinus Torvalds */ 24*1da177e4SLinus Torvalds 25*1da177e4SLinus Torvalds /* 26*1da177e4SLinus Torvalds * The hangcheck-timer driver uses the TSC to catch delays that 27*1da177e4SLinus Torvalds * jiffies does not notice. A timer is set. When the timer fires, it 28*1da177e4SLinus Torvalds * checks whether it was delayed and if that delay exceeds a given 29*1da177e4SLinus Torvalds * margin of error. The hangcheck_tick module paramter takes the timer 30*1da177e4SLinus Torvalds * duration in seconds. The hangcheck_margin parameter defines the 31*1da177e4SLinus Torvalds * margin of error, in seconds. The defaults are 60 seconds for the 32*1da177e4SLinus Torvalds * timer and 180 seconds for the margin of error. IOW, a timer is set 33*1da177e4SLinus Torvalds * for 60 seconds. When the timer fires, the callback checks the 34*1da177e4SLinus Torvalds * actual duration that the timer waited. If the duration exceeds the 35*1da177e4SLinus Torvalds * alloted time and margin (here 60 + 180, or 240 seconds), the machine 36*1da177e4SLinus Torvalds * is restarted. A healthy machine will have the duration match the 37*1da177e4SLinus Torvalds * expected timeout very closely. 38*1da177e4SLinus Torvalds */ 39*1da177e4SLinus Torvalds 40*1da177e4SLinus Torvalds #include <linux/module.h> 41*1da177e4SLinus Torvalds #include <linux/moduleparam.h> 42*1da177e4SLinus Torvalds #include <linux/types.h> 43*1da177e4SLinus Torvalds #include <linux/kernel.h> 44*1da177e4SLinus Torvalds #include <linux/fs.h> 45*1da177e4SLinus Torvalds #include <linux/mm.h> 46*1da177e4SLinus Torvalds #include <linux/reboot.h> 47*1da177e4SLinus Torvalds #include <linux/init.h> 48*1da177e4SLinus Torvalds #include <asm/uaccess.h> 49*1da177e4SLinus Torvalds 50*1da177e4SLinus Torvalds 51*1da177e4SLinus Torvalds #define VERSION_STR "0.5.0" 52*1da177e4SLinus Torvalds 53*1da177e4SLinus Torvalds #define DEFAULT_IOFENCE_MARGIN 60 /* Default fudge factor, in seconds */ 54*1da177e4SLinus Torvalds #define DEFAULT_IOFENCE_TICK 180 /* Default timer timeout, in seconds */ 55*1da177e4SLinus Torvalds 56*1da177e4SLinus Torvalds static int hangcheck_tick = DEFAULT_IOFENCE_TICK; 57*1da177e4SLinus Torvalds static int hangcheck_margin = DEFAULT_IOFENCE_MARGIN; 58*1da177e4SLinus Torvalds static int hangcheck_reboot; /* Defaults to not reboot */ 59*1da177e4SLinus Torvalds 60*1da177e4SLinus Torvalds /* Driver options */ 61*1da177e4SLinus Torvalds module_param(hangcheck_tick, int, 0); 62*1da177e4SLinus Torvalds MODULE_PARM_DESC(hangcheck_tick, "Timer delay."); 63*1da177e4SLinus Torvalds module_param(hangcheck_margin, int, 0); 64*1da177e4SLinus Torvalds MODULE_PARM_DESC(hangcheck_margin, "If the hangcheck timer has been delayed more than hangcheck_margin seconds, the driver will fire."); 65*1da177e4SLinus Torvalds module_param(hangcheck_reboot, int, 0); 66*1da177e4SLinus Torvalds MODULE_PARM_DESC(hangcheck_reboot, "If nonzero, the machine will reboot when the timer margin is exceeded."); 67*1da177e4SLinus Torvalds 68*1da177e4SLinus Torvalds MODULE_AUTHOR("Joel Becker"); 69*1da177e4SLinus Torvalds MODULE_DESCRIPTION("Hangcheck-timer detects when the system has gone out to lunch past a certain margin."); 70*1da177e4SLinus Torvalds MODULE_LICENSE("GPL"); 71*1da177e4SLinus Torvalds 72*1da177e4SLinus Torvalds 73*1da177e4SLinus Torvalds /* Last time scheduled */ 74*1da177e4SLinus Torvalds static unsigned long long hangcheck_tsc, hangcheck_tsc_margin; 75*1da177e4SLinus Torvalds 76*1da177e4SLinus Torvalds static void hangcheck_fire(unsigned long); 77*1da177e4SLinus Torvalds 78*1da177e4SLinus Torvalds static struct timer_list hangcheck_ticktock = 79*1da177e4SLinus Torvalds TIMER_INITIALIZER(hangcheck_fire, 0, 0); 80*1da177e4SLinus Torvalds 81*1da177e4SLinus Torvalds extern unsigned long long monotonic_clock(void); 82*1da177e4SLinus Torvalds 83*1da177e4SLinus Torvalds static void hangcheck_fire(unsigned long data) 84*1da177e4SLinus Torvalds { 85*1da177e4SLinus Torvalds unsigned long long cur_tsc, tsc_diff; 86*1da177e4SLinus Torvalds 87*1da177e4SLinus Torvalds cur_tsc = monotonic_clock(); 88*1da177e4SLinus Torvalds 89*1da177e4SLinus Torvalds if (cur_tsc > hangcheck_tsc) 90*1da177e4SLinus Torvalds tsc_diff = cur_tsc - hangcheck_tsc; 91*1da177e4SLinus Torvalds else 92*1da177e4SLinus Torvalds tsc_diff = (cur_tsc + (~0ULL - hangcheck_tsc)); /* or something */ 93*1da177e4SLinus Torvalds 94*1da177e4SLinus Torvalds if (tsc_diff > hangcheck_tsc_margin) { 95*1da177e4SLinus Torvalds if (hangcheck_reboot) { 96*1da177e4SLinus Torvalds printk(KERN_CRIT "Hangcheck: hangcheck is restarting the machine.\n"); 97*1da177e4SLinus Torvalds machine_restart(NULL); 98*1da177e4SLinus Torvalds } else { 99*1da177e4SLinus Torvalds printk(KERN_CRIT "Hangcheck: hangcheck value past margin!\n"); 100*1da177e4SLinus Torvalds } 101*1da177e4SLinus Torvalds } 102*1da177e4SLinus Torvalds mod_timer(&hangcheck_ticktock, jiffies + (hangcheck_tick*HZ)); 103*1da177e4SLinus Torvalds hangcheck_tsc = monotonic_clock(); 104*1da177e4SLinus Torvalds } 105*1da177e4SLinus Torvalds 106*1da177e4SLinus Torvalds 107*1da177e4SLinus Torvalds static int __init hangcheck_init(void) 108*1da177e4SLinus Torvalds { 109*1da177e4SLinus Torvalds printk("Hangcheck: starting hangcheck timer %s (tick is %d seconds, margin is %d seconds).\n", 110*1da177e4SLinus Torvalds VERSION_STR, hangcheck_tick, hangcheck_margin); 111*1da177e4SLinus Torvalds 112*1da177e4SLinus Torvalds hangcheck_tsc_margin = hangcheck_margin + hangcheck_tick; 113*1da177e4SLinus Torvalds hangcheck_tsc_margin *= 1000000000; 114*1da177e4SLinus Torvalds 115*1da177e4SLinus Torvalds 116*1da177e4SLinus Torvalds hangcheck_tsc = monotonic_clock(); 117*1da177e4SLinus Torvalds mod_timer(&hangcheck_ticktock, jiffies + (hangcheck_tick*HZ)); 118*1da177e4SLinus Torvalds 119*1da177e4SLinus Torvalds return 0; 120*1da177e4SLinus Torvalds } 121*1da177e4SLinus Torvalds 122*1da177e4SLinus Torvalds 123*1da177e4SLinus Torvalds static void __exit hangcheck_exit(void) 124*1da177e4SLinus Torvalds { 125*1da177e4SLinus Torvalds del_timer_sync(&hangcheck_ticktock); 126*1da177e4SLinus Torvalds } 127*1da177e4SLinus Torvalds 128*1da177e4SLinus Torvalds module_init(hangcheck_init); 129*1da177e4SLinus Torvalds module_exit(hangcheck_exit); 130