1/*
2 * hangcheck-timer.c
3 *
4 * Driver for a little io fencing timer.
5 *
6 * Copyright (C) 2002, 2003 Oracle.  All rights reserved.
7 *
8 * Author: Joel Becker <joel.becker@oracle.com>
9 *
10 * This program is free software; you can redistribute it and/or
11 * modify it under the terms of the GNU General Public
12 * License version 2 as published by the Free Software Foundation.
13 *
14 * This program is distributed in the hope that it will be useful,
15 * but WITHOUT ANY WARRANTY; without even the implied warranty of
16 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
17 * General Public License for more details.
18 *
19 * You should have received a copy of the GNU General Public
20 * License along with this program; if not, write to the
21 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
22 * Boston, MA 021110-1307, USA.
23 */
24
25/*
26 * The hangcheck-timer driver uses the TSC to catch delays that
27 * jiffies does not notice.  A timer is set.  When the timer fires, it
28 * checks whether it was delayed and if that delay exceeds a given
29 * margin of error.  The hangcheck_tick module parameter takes the timer
30 * duration in seconds.  The hangcheck_margin parameter defines the
31 * margin of error, in seconds.  The defaults are 60 seconds for the
32 * timer and 180 seconds for the margin of error.  IOW, a timer is set
33 * for 60 seconds.  When the timer fires, the callback checks the
34 * actual duration that the timer waited.  If the duration exceeds the
35 * alloted time and margin (here 60 + 180, or 240 seconds), the machine
36 * is restarted.  A healthy machine will have the duration match the
37 * expected timeout very closely.
38 */
39
40#include <linux/module.h>
41#include <linux/moduleparam.h>
42#include <linux/types.h>
43#include <linux/kernel.h>
44#include <linux/fs.h>
45#include <linux/mm.h>
46#include <linux/reboot.h>
47#include <linux/init.h>
48#include <linux/delay.h>
49#include <asm/uaccess.h>
50#include <linux/sysrq.h>
51#include <linux/timer.h>
52#include <linux/hrtimer.h>
53
54#define VERSION_STR "0.9.1"
55
56#define DEFAULT_IOFENCE_MARGIN 60	/* Default fudge factor, in seconds */
57#define DEFAULT_IOFENCE_TICK 180	/* Default timer timeout, in seconds */
58
59static int hangcheck_tick = DEFAULT_IOFENCE_TICK;
60static int hangcheck_margin = DEFAULT_IOFENCE_MARGIN;
61static int hangcheck_reboot;  /* Defaults to not reboot */
62static int hangcheck_dump_tasks;  /* Defaults to not dumping SysRQ T */
63
64/* options - modular */
65module_param(hangcheck_tick, int, 0);
66MODULE_PARM_DESC(hangcheck_tick, "Timer delay.");
67module_param(hangcheck_margin, int, 0);
68MODULE_PARM_DESC(hangcheck_margin, "If the hangcheck timer has been delayed more than hangcheck_margin seconds, the driver will fire.");
69module_param(hangcheck_reboot, int, 0);
70MODULE_PARM_DESC(hangcheck_reboot, "If nonzero, the machine will reboot when the timer margin is exceeded.");
71module_param(hangcheck_dump_tasks, int, 0);
72MODULE_PARM_DESC(hangcheck_dump_tasks, "If nonzero, the machine will dump the system task state when the timer margin is exceeded.");
73
74MODULE_AUTHOR("Oracle");
75MODULE_DESCRIPTION("Hangcheck-timer detects when the system has gone out to lunch past a certain margin.");
76MODULE_LICENSE("GPL");
77MODULE_VERSION(VERSION_STR);
78
79/* options - nonmodular */
80#ifndef MODULE
81
82static int __init hangcheck_parse_tick(char *str)
83{
84	int par;
85	if (get_option(&str,&par))
86		hangcheck_tick = par;
87	return 1;
88}
89
90static int __init hangcheck_parse_margin(char *str)
91{
92	int par;
93	if (get_option(&str,&par))
94		hangcheck_margin = par;
95	return 1;
96}
97
98static int __init hangcheck_parse_reboot(char *str)
99{
100	int par;
101	if (get_option(&str,&par))
102		hangcheck_reboot = par;
103	return 1;
104}
105
106static int __init hangcheck_parse_dump_tasks(char *str)
107{
108	int par;
109	if (get_option(&str,&par))
110		hangcheck_dump_tasks = par;
111	return 1;
112}
113
114__setup("hcheck_tick", hangcheck_parse_tick);
115__setup("hcheck_margin", hangcheck_parse_margin);
116__setup("hcheck_reboot", hangcheck_parse_reboot);
117__setup("hcheck_dump_tasks", hangcheck_parse_dump_tasks);
118#endif /* not MODULE */
119
120#define TIMER_FREQ 1000000000ULL
121
122/* Last time scheduled */
123static unsigned long long hangcheck_tsc, hangcheck_tsc_margin;
124
125static void hangcheck_fire(unsigned long);
126
127static DEFINE_TIMER(hangcheck_ticktock, hangcheck_fire, 0, 0);
128
129static void hangcheck_fire(unsigned long data)
130{
131	unsigned long long cur_tsc, tsc_diff;
132
133	cur_tsc = ktime_get_ns();
134
135	if (cur_tsc > hangcheck_tsc)
136		tsc_diff = cur_tsc - hangcheck_tsc;
137	else
138		tsc_diff = (cur_tsc + (~0ULL - hangcheck_tsc)); /* or something */
139
140	if (tsc_diff > hangcheck_tsc_margin) {
141		if (hangcheck_dump_tasks) {
142			printk(KERN_CRIT "Hangcheck: Task state:\n");
143#ifdef CONFIG_MAGIC_SYSRQ
144			handle_sysrq('t');
145#endif  /* CONFIG_MAGIC_SYSRQ */
146		}
147		if (hangcheck_reboot) {
148			printk(KERN_CRIT "Hangcheck: hangcheck is restarting the machine.\n");
149			emergency_restart();
150		} else {
151			printk(KERN_CRIT "Hangcheck: hangcheck value past margin!\n");
152		}
153	}
154#if 0
155	/*
156	 * Enable to investigate delays in detail
157	 */
158	printk("Hangcheck: called %Ld ns since last time (%Ld ns overshoot)\n",
159			tsc_diff, tsc_diff - hangcheck_tick*TIMER_FREQ);
160#endif
161	mod_timer(&hangcheck_ticktock, jiffies + (hangcheck_tick*HZ));
162	hangcheck_tsc = ktime_get_ns();
163}
164
165
166static int __init hangcheck_init(void)
167{
168	printk("Hangcheck: starting hangcheck timer %s (tick is %d seconds, margin is %d seconds).\n",
169	       VERSION_STR, hangcheck_tick, hangcheck_margin);
170	hangcheck_tsc_margin =
171		(unsigned long long)hangcheck_margin + hangcheck_tick;
172	hangcheck_tsc_margin *= TIMER_FREQ;
173
174	hangcheck_tsc = ktime_get_ns();
175	mod_timer(&hangcheck_ticktock, jiffies + (hangcheck_tick*HZ));
176
177	return 0;
178}
179
180
181static void __exit hangcheck_exit(void)
182{
183	del_timer_sync(&hangcheck_ticktock);
184        printk("Hangcheck: Stopped hangcheck timer.\n");
185}
186
187module_init(hangcheck_init);
188module_exit(hangcheck_exit);
189