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 paramter 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
53#define VERSION_STR "0.9.0"
54
55#define DEFAULT_IOFENCE_MARGIN 60	/* Default fudge factor, in seconds */
56#define DEFAULT_IOFENCE_TICK 180	/* Default timer timeout, in seconds */
57
58static int hangcheck_tick = DEFAULT_IOFENCE_TICK;
59static int hangcheck_margin = DEFAULT_IOFENCE_MARGIN;
60static int hangcheck_reboot;  /* Defaults to not reboot */
61static int hangcheck_dump_tasks;  /* Defaults to not dumping SysRQ T */
62
63/* options - modular */
64module_param(hangcheck_tick, int, 0);
65MODULE_PARM_DESC(hangcheck_tick, "Timer delay.");
66module_param(hangcheck_margin, int, 0);
67MODULE_PARM_DESC(hangcheck_margin, "If the hangcheck timer has been delayed more than hangcheck_margin seconds, the driver will fire.");
68module_param(hangcheck_reboot, int, 0);
69MODULE_PARM_DESC(hangcheck_reboot, "If nonzero, the machine will reboot when the timer margin is exceeded.");
70module_param(hangcheck_dump_tasks, int, 0);
71MODULE_PARM_DESC(hangcheck_dump_tasks, "If nonzero, the machine will dump the system task state when the timer margin is exceeded.");
72
73MODULE_AUTHOR("Oracle");
74MODULE_DESCRIPTION("Hangcheck-timer detects when the system has gone out to lunch past a certain margin.");
75MODULE_LICENSE("GPL");
76MODULE_VERSION(VERSION_STR);
77
78/* options - nonmodular */
79#ifndef MODULE
80
81static int __init hangcheck_parse_tick(char *str)
82{
83	int par;
84	if (get_option(&str,&par))
85		hangcheck_tick = par;
86	return 1;
87}
88
89static int __init hangcheck_parse_margin(char *str)
90{
91	int par;
92	if (get_option(&str,&par))
93		hangcheck_margin = par;
94	return 1;
95}
96
97static int __init hangcheck_parse_reboot(char *str)
98{
99	int par;
100	if (get_option(&str,&par))
101		hangcheck_reboot = par;
102	return 1;
103}
104
105static int __init hangcheck_parse_dump_tasks(char *str)
106{
107	int par;
108	if (get_option(&str,&par))
109		hangcheck_dump_tasks = par;
110	return 1;
111}
112
113__setup("hcheck_tick", hangcheck_parse_tick);
114__setup("hcheck_margin", hangcheck_parse_margin);
115__setup("hcheck_reboot", hangcheck_parse_reboot);
116__setup("hcheck_dump_tasks", hangcheck_parse_dump_tasks);
117#endif /* not MODULE */
118
119#if defined(CONFIG_S390)
120# define HAVE_MONOTONIC
121# define TIMER_FREQ 1000000000ULL
122#elif defined(CONFIG_IA64)
123# define TIMER_FREQ ((unsigned long long)local_cpu_data->itc_freq)
124#else
125# define TIMER_FREQ (HZ*loops_per_jiffy)
126#endif
127
128#ifdef HAVE_MONOTONIC
129extern unsigned long long monotonic_clock(void);
130#else
131static inline unsigned long long monotonic_clock(void)
132{
133	return get_cycles();
134}
135#endif  /* HAVE_MONOTONIC */
136
137
138/* Last time scheduled */
139static unsigned long long hangcheck_tsc, hangcheck_tsc_margin;
140
141static void hangcheck_fire(unsigned long);
142
143static DEFINE_TIMER(hangcheck_ticktock, hangcheck_fire, 0, 0);
144
145
146static void hangcheck_fire(unsigned long data)
147{
148	unsigned long long cur_tsc, tsc_diff;
149
150	cur_tsc = monotonic_clock();
151
152	if (cur_tsc > hangcheck_tsc)
153		tsc_diff = cur_tsc - hangcheck_tsc;
154	else
155		tsc_diff = (cur_tsc + (~0ULL - hangcheck_tsc)); /* or something */
156
157	if (tsc_diff > hangcheck_tsc_margin) {
158		if (hangcheck_dump_tasks) {
159			printk(KERN_CRIT "Hangcheck: Task state:\n");
160#ifdef CONFIG_MAGIC_SYSRQ
161			handle_sysrq('t', NULL);
162#endif  /* CONFIG_MAGIC_SYSRQ */
163		}
164		if (hangcheck_reboot) {
165			printk(KERN_CRIT "Hangcheck: hangcheck is restarting the machine.\n");
166			emergency_restart();
167		} else {
168			printk(KERN_CRIT "Hangcheck: hangcheck value past margin!\n");
169		}
170	}
171	mod_timer(&hangcheck_ticktock, jiffies + (hangcheck_tick*HZ));
172	hangcheck_tsc = monotonic_clock();
173}
174
175
176static int __init hangcheck_init(void)
177{
178	printk("Hangcheck: starting hangcheck timer %s (tick is %d seconds, margin is %d seconds).\n",
179	       VERSION_STR, hangcheck_tick, hangcheck_margin);
180#if defined(HAVE_MONOTONIC)
181	printk("Hangcheck: Using monotonic_clock().\n");
182#else
183	printk("Hangcheck: Using get_cycles().\n");
184#endif  /* HAVE_MONOTONIC */
185	hangcheck_tsc_margin =
186		(unsigned long long)(hangcheck_margin + hangcheck_tick);
187	hangcheck_tsc_margin *= (unsigned long long)TIMER_FREQ;
188
189	hangcheck_tsc = monotonic_clock();
190	mod_timer(&hangcheck_ticktock, jiffies + (hangcheck_tick*HZ));
191
192	return 0;
193}
194
195
196static void __exit hangcheck_exit(void)
197{
198	del_timer_sync(&hangcheck_ticktock);
199        printk("Hangcheck: Stopped hangcheck timer.\n");
200}
201
202module_init(hangcheck_init);
203module_exit(hangcheck_exit);
204