hangcheck-timer.c

  1// SPDX-License-Identifier: GPL-2.0-only
  2/*
  3 * hangcheck-timer.c
  4 *
  5 * Driver for a little io fencing timer.
  6 *
  7 * Copyright (C) 2002, 2003 Oracle.  All rights reserved.
  8 *
  9 * Author: Joel Becker <joel.becker@oracle.com>
 10 */
 11
 12/*
 13 * The hangcheck-timer driver uses the TSC to catch delays that
 14 * jiffies does not notice.  A timer is set.  When the timer fires, it
 15 * checks whether it was delayed and if that delay exceeds a given
 16 * margin of error.  The hangcheck_tick module parameter takes the timer
 17 * duration in seconds.  The hangcheck_margin parameter defines the
 18 * margin of error, in seconds.  The defaults are 60 seconds for the
 19 * timer and 180 seconds for the margin of error.  IOW, a timer is set
 20 * for 60 seconds.  When the timer fires, the callback checks the
 21 * actual duration that the timer waited.  If the duration exceeds the
 22 * allotted time and margin (here 60 + 180, or 240 seconds), the machine
 23 * is restarted.  A healthy machine will have the duration match the
 24 * expected timeout very closely.
 25 */
 26
 27#include <linux/module.h>
 28#include <linux/moduleparam.h>
 29#include <linux/types.h>
 30#include <linux/kernel.h>
 31#include <linux/fs.h>
 32#include <linux/mm.h>
 33#include <linux/reboot.h>
 34#include <linux/init.h>
 35#include <linux/delay.h>
 36#include <linux/uaccess.h>
 37#include <linux/sysrq.h>
 38#include <linux/timer.h>
 39#include <linux/hrtimer.h>
 40
 41#define VERSION_STR "0.9.1"
 42
 43#define DEFAULT_IOFENCE_MARGIN 60	/* Default fudge factor, in seconds */
 44#define DEFAULT_IOFENCE_TICK 180	/* Default timer timeout, in seconds */
 45
 46static int hangcheck_tick = DEFAULT_IOFENCE_TICK;
 47static int hangcheck_margin = DEFAULT_IOFENCE_MARGIN;
 48static int hangcheck_reboot;  /* Defaults to not reboot */
 49static int hangcheck_dump_tasks;  /* Defaults to not dumping SysRQ T */
 50
 51/* options - modular */
 52module_param(hangcheck_tick, int, 0);
 53MODULE_PARM_DESC(hangcheck_tick, "Timer delay.");
 54module_param(hangcheck_margin, int, 0);
 55MODULE_PARM_DESC(hangcheck_margin, "If the hangcheck timer has been delayed more than hangcheck_margin seconds, the driver will fire.");
 56module_param(hangcheck_reboot, int, 0);
 57MODULE_PARM_DESC(hangcheck_reboot, "If nonzero, the machine will reboot when the timer margin is exceeded.");
 58module_param(hangcheck_dump_tasks, int, 0);
 59MODULE_PARM_DESC(hangcheck_dump_tasks, "If nonzero, the machine will dump the system task state when the timer margin is exceeded.");
 60
 61MODULE_AUTHOR("Oracle");
 62MODULE_DESCRIPTION("Hangcheck-timer detects when the system has gone out to lunch past a certain margin.");
 63MODULE_LICENSE("GPL");
 64MODULE_VERSION(VERSION_STR);
 65
 66/* options - nonmodular */
 67#ifndef MODULE
 68
 69static int __init hangcheck_parse_tick(char *str)
 70{
 71	int par;
 72
 73	if (get_option(&str, &par))
 74		hangcheck_tick = par;
 75	return 1;
 76}
 77
 78static int __init hangcheck_parse_margin(char *str)
 79{
 80	int par;
 81
 82	if (get_option(&str, &par))
 83		hangcheck_margin = par;
 84	return 1;
 85}
 86
 87static int __init hangcheck_parse_reboot(char *str)
 88{
 89	int par;
 90
 91	if (get_option(&str, &par))
 92		hangcheck_reboot = par;
 93	return 1;
 94}
 95
 96static int __init hangcheck_parse_dump_tasks(char *str)
 97{
 98	int par;
 99
100	if (get_option(&str, &par))
101		hangcheck_dump_tasks = par;
102	return 1;
103}
104
105__setup("hcheck_tick", hangcheck_parse_tick);
106__setup("hcheck_margin", hangcheck_parse_margin);
107__setup("hcheck_reboot", hangcheck_parse_reboot);
108__setup("hcheck_dump_tasks", hangcheck_parse_dump_tasks);
109#endif /* not MODULE */
110
111#define TIMER_FREQ 1000000000ULL
112
113/* Last time scheduled */
114static unsigned long long hangcheck_tsc, hangcheck_tsc_margin;
115
116static void hangcheck_fire(struct timer_list *);
117
118static DEFINE_TIMER(hangcheck_ticktock, hangcheck_fire);
119
120static void hangcheck_fire(struct timer_list *unused)
121{
122	unsigned long long cur_tsc, tsc_diff;
123
124	cur_tsc = ktime_get_ns();
125
126	if (cur_tsc > hangcheck_tsc)
127		tsc_diff = cur_tsc - hangcheck_tsc;
128	else
129		tsc_diff = (cur_tsc + (~0ULL - hangcheck_tsc)); /* or something */
130
131	if (tsc_diff > hangcheck_tsc_margin) {
132		if (hangcheck_dump_tasks) {
133			pr_crit("Hangcheck: Task state:\n");
134#ifdef CONFIG_MAGIC_SYSRQ
135			handle_sysrq('t');
136#endif  /* CONFIG_MAGIC_SYSRQ */
137		}
138		if (hangcheck_reboot) {
139			pr_crit("Hangcheck: hangcheck is restarting the machine.\n");
140			emergency_restart();
141		} else {
142			pr_crit("Hangcheck: hangcheck value past margin!\n");
143		}
144	}
145#if 0
146	/*
147	 * Enable to investigate delays in detail
148	 */
149	pr_debug("Hangcheck: called %lld ns since last time (%lld ns overshoot)\n",
150			tsc_diff, tsc_diff - hangcheck_tick*TIMER_FREQ);
151#endif
152	mod_timer(&hangcheck_ticktock, jiffies + (hangcheck_tick*HZ));
153	hangcheck_tsc = ktime_get_ns();
154}
155
156
157static int __init hangcheck_init(void)
158{
159	pr_debug("Hangcheck: starting hangcheck timer %s (tick is %d seconds, margin is %d seconds).\n",
160	       VERSION_STR, hangcheck_tick, hangcheck_margin);
161	hangcheck_tsc_margin =
162		(unsigned long long)hangcheck_margin + hangcheck_tick;
163	hangcheck_tsc_margin *= TIMER_FREQ;
164
165	hangcheck_tsc = ktime_get_ns();
166	mod_timer(&hangcheck_ticktock, jiffies + (hangcheck_tick*HZ));
167
168	return 0;
169}
170
171
172static void __exit hangcheck_exit(void)
173{
174	timer_delete_sync(&hangcheck_ticktock);
175	pr_debug("Hangcheck: Stopped hangcheck timer.\n");
176}
177
178module_init(hangcheck_init);
179module_exit(hangcheck_exit);