blob: 78e650fc5b411c076ce288b9926c22a1dcd239fb [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * hangcheck-timer.c
3 *
4 * Driver for a little io fencing timer.
5 *
Joel Becker696f9482005-05-01 08:59:08 -07006 * Copyright (C) 2002, 2003 Oracle. All rights reserved.
Linus Torvalds1da177e2005-04-16 15:20:36 -07007 *
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>
Joel Becker696f9482005-05-01 08:59:08 -070047#include <linux/smp_lock.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070048#include <linux/init.h>
Joel Becker696f9482005-05-01 08:59:08 -070049#include <linux/delay.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070050#include <asm/uaccess.h>
Joel Becker696f9482005-05-01 08:59:08 -070051#include <linux/sysrq.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070052
53
Joel Becker696f9482005-05-01 08:59:08 -070054#define VERSION_STR "0.9.0"
Linus Torvalds1da177e2005-04-16 15:20:36 -070055
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 */
Joel Becker696f9482005-05-01 08:59:08 -070062static int hangcheck_dump_tasks; /* Defaults to not dumping SysRQ T */
Linus Torvalds1da177e2005-04-16 15:20:36 -070063
Joel Becker696f9482005-05-01 08:59:08 -070064/* options - modular */
Linus Torvalds1da177e2005-04-16 15:20:36 -070065module_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.");
Joel Becker696f9482005-05-01 08:59:08 -070071module_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.");
Linus Torvalds1da177e2005-04-16 15:20:36 -070073
Joel Becker696f9482005-05-01 08:59:08 -070074MODULE_AUTHOR("Oracle");
Linus Torvalds1da177e2005-04-16 15:20:36 -070075MODULE_DESCRIPTION("Hangcheck-timer detects when the system has gone out to lunch past a certain margin.");
76MODULE_LICENSE("GPL");
Joel Becker696f9482005-05-01 08:59:08 -070077MODULE_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#if defined(CONFIG_X86) || defined(CONFIG_X86_64)
121# define HAVE_MONOTONIC
122# define TIMER_FREQ 1000000000ULL
123#elif defined(CONFIG_ARCH_S390)
124/* FA240000 is 1 Second in the IBM time universe (Page 4-38 Principles of Op for zSeries */
125# define TIMER_FREQ 0xFA240000ULL
126#elif defined(CONFIG_IA64)
127# define TIMER_FREQ ((unsigned long long)local_cpu_data->itc_freq)
128#elif defined(CONFIG_PPC64)
129# define TIMER_FREQ (HZ*loops_per_jiffy)
130#endif
131
132#ifdef HAVE_MONOTONIC
133extern unsigned long long monotonic_clock(void);
134#else
135static inline unsigned long long monotonic_clock(void)
136{
137# ifdef __s390__
138 /* returns the TOD. see 4-38 Principles of Op of zSeries */
139 return get_clock();
140# else
141 return get_cycles();
142# endif /* __s390__ */
143}
144#endif /* HAVE_MONOTONIC */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700145
146
147/* Last time scheduled */
148static unsigned long long hangcheck_tsc, hangcheck_tsc_margin;
149
150static void hangcheck_fire(unsigned long);
151
152static struct timer_list hangcheck_ticktock =
153 TIMER_INITIALIZER(hangcheck_fire, 0, 0);
154
Linus Torvalds1da177e2005-04-16 15:20:36 -0700155
156static void hangcheck_fire(unsigned long data)
157{
158 unsigned long long cur_tsc, tsc_diff;
159
160 cur_tsc = monotonic_clock();
161
162 if (cur_tsc > hangcheck_tsc)
163 tsc_diff = cur_tsc - hangcheck_tsc;
164 else
165 tsc_diff = (cur_tsc + (~0ULL - hangcheck_tsc)); /* or something */
166
167 if (tsc_diff > hangcheck_tsc_margin) {
Joel Becker696f9482005-05-01 08:59:08 -0700168 if (hangcheck_dump_tasks) {
169 printk(KERN_CRIT "Hangcheck: Task state:\n");
170#ifdef CONFIG_MAGIC_SYSRQ
171 handle_sysrq('t', NULL, NULL);
172#endif /* CONFIG_MAGIC_SYSRQ */
173 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700174 if (hangcheck_reboot) {
175 printk(KERN_CRIT "Hangcheck: hangcheck is restarting the machine.\n");
176 machine_restart(NULL);
177 } else {
178 printk(KERN_CRIT "Hangcheck: hangcheck value past margin!\n");
179 }
180 }
181 mod_timer(&hangcheck_ticktock, jiffies + (hangcheck_tick*HZ));
182 hangcheck_tsc = monotonic_clock();
183}
184
185
186static int __init hangcheck_init(void)
187{
188 printk("Hangcheck: starting hangcheck timer %s (tick is %d seconds, margin is %d seconds).\n",
189 VERSION_STR, hangcheck_tick, hangcheck_margin);
Joel Becker696f9482005-05-01 08:59:08 -0700190#if defined (HAVE_MONOTONIC)
191 printk("Hangcheck: Using monotonic_clock().\n");
192#elif defined(__s390__)
193 printk("Hangcheck: Using TOD.\n");
194#else
195 printk("Hangcheck: Using get_cycles().\n");
196#endif /* HAVE_MONOTONIC */
197 hangcheck_tsc_margin =
198 (unsigned long long)(hangcheck_margin + hangcheck_tick);
199 hangcheck_tsc_margin *= (unsigned long long)TIMER_FREQ;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700200
201 hangcheck_tsc = monotonic_clock();
202 mod_timer(&hangcheck_ticktock, jiffies + (hangcheck_tick*HZ));
203
204 return 0;
205}
206
207
208static void __exit hangcheck_exit(void)
209{
210 del_timer_sync(&hangcheck_ticktock);
Joel Becker696f9482005-05-01 08:59:08 -0700211 printk("Hangcheck: Stopped hangcheck timer.\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700212}
213
214module_init(hangcheck_init);
215module_exit(hangcheck_exit);