blob: ba5a77cad3bb4d79fdb8d0e0ce0fba120f280726 [file] [log] [blame]
Mandeep Singh Bainese162b392009-01-15 11:08:40 -08001/*
2 * Detect Hung Task
3 *
4 * kernel/hung_task.c - kernel thread for detecting tasks stuck in D state
5 *
6 */
7
8#include <linux/mm.h>
9#include <linux/cpu.h>
10#include <linux/nmi.h>
11#include <linux/init.h>
12#include <linux/delay.h>
13#include <linux/freezer.h>
14#include <linux/kthread.h>
15#include <linux/lockdep.h>
16#include <linux/module.h>
17#include <linux/sysctl.h>
18
19/*
20 * Have a reasonable limit on the number of tasks checked:
21 */
22unsigned long __read_mostly sysctl_hung_task_check_count = 1024;
23
24/*
25 * Zero means infinite timeout - no checking done:
26 */
27unsigned long __read_mostly sysctl_hung_task_timeout_secs = 120;
28static unsigned long __read_mostly hung_task_poll_jiffies;
29
30unsigned long __read_mostly sysctl_hung_task_warnings = 10;
31
32static int __read_mostly did_panic;
33
34static struct task_struct *watchdog_task;
35
36/*
37 * Should we panic (and reboot, if panic_timeout= is set) when a
38 * hung task is detected:
39 */
40unsigned int __read_mostly sysctl_hung_task_panic =
41 CONFIG_BOOTPARAM_HUNG_TASK_PANIC_VALUE;
42
43static int __init hung_task_panic_setup(char *str)
44{
45 sysctl_hung_task_panic = simple_strtoul(str, NULL, 0);
46
47 return 1;
48}
49__setup("hung_task_panic=", hung_task_panic_setup);
50
51static int
52hung_task_panic(struct notifier_block *this, unsigned long event, void *ptr)
53{
54 did_panic = 1;
55
56 return NOTIFY_DONE;
57}
58
59static struct notifier_block panic_block = {
60 .notifier_call = hung_task_panic,
61};
62
63/*
64 * Returns seconds, approximately. We don't need nanosecond
65 * resolution, and we don't need to waste time with a big divide when
66 * 2^30ns == 1.074s.
67 */
68static unsigned long get_timestamp(void)
69{
70 int this_cpu = raw_smp_processor_id();
71
72 return cpu_clock(this_cpu) >> 30LL; /* 2^30 ~= 10^9 */
73}
74
75static void check_hung_task(struct task_struct *t, unsigned long now)
76{
77 unsigned long switch_count = t->nvcsw + t->nivcsw;
78
79 if (t->flags & PF_FROZEN)
80 return;
81
82 if (switch_count != t->last_switch_count || !t->last_switch_timestamp) {
83 t->last_switch_count = switch_count;
84 t->last_switch_timestamp = now;
85 return;
86 }
87 if ((long)(now - t->last_switch_timestamp) <
88 sysctl_hung_task_timeout_secs)
89 return;
90 if (!sysctl_hung_task_warnings)
91 return;
92 sysctl_hung_task_warnings--;
93
94 /*
95 * Ok, the task did not get scheduled for more than 2 minutes,
96 * complain:
97 */
98 printk(KERN_ERR "INFO: task %s:%d blocked for more than "
99 "%ld seconds.\n", t->comm, t->pid,
100 sysctl_hung_task_timeout_secs);
101 printk(KERN_ERR "\"echo 0 > /proc/sys/kernel/hung_task_timeout_secs\""
102 " disables this message.\n");
103 sched_show_task(t);
104 __debug_show_held_locks(t);
105
106 t->last_switch_timestamp = now;
107 touch_nmi_watchdog();
108
109 if (sysctl_hung_task_panic)
110 panic("hung_task: blocked tasks");
111}
112
113/*
114 * Check whether a TASK_UNINTERRUPTIBLE does not get woken up for
115 * a really long time (120 seconds). If that happens, print out
116 * a warning.
117 */
118static void check_hung_uninterruptible_tasks(void)
119{
120 int max_count = sysctl_hung_task_check_count;
121 unsigned long now = get_timestamp();
122 struct task_struct *g, *t;
123
124 /*
125 * If the system crashed already then all bets are off,
126 * do not report extra hung tasks:
127 */
128 if (test_taint(TAINT_DIE) || did_panic)
129 return;
130
131 read_lock(&tasklist_lock);
132 do_each_thread(g, t) {
133 if (!--max_count)
134 goto unlock;
135 /* use "==" to skip the TASK_KILLABLE tasks waiting on NFS */
136 if (t->state == TASK_UNINTERRUPTIBLE)
137 check_hung_task(t, now);
138 } while_each_thread(g, t);
139 unlock:
140 read_unlock(&tasklist_lock);
141}
142
143static void update_poll_jiffies(void)
144{
145 /* timeout of 0 will disable the watchdog */
146 if (sysctl_hung_task_timeout_secs == 0)
147 hung_task_poll_jiffies = MAX_SCHEDULE_TIMEOUT;
148 else
149 hung_task_poll_jiffies = sysctl_hung_task_timeout_secs * HZ / 2;
150}
151
152/*
153 * Process updating of timeout sysctl
154 */
155int proc_dohung_task_timeout_secs(struct ctl_table *table, int write,
156 struct file *filp, void __user *buffer,
157 size_t *lenp, loff_t *ppos)
158{
159 int ret;
160
161 ret = proc_doulongvec_minmax(table, write, filp, buffer, lenp, ppos);
162
163 if (ret || !write)
164 goto out;
165
166 update_poll_jiffies();
167
168 wake_up_process(watchdog_task);
169
170 out:
171 return ret;
172}
173
174/*
175 * kthread which checks for tasks stuck in D state
176 */
177static int watchdog(void *dummy)
178{
179 set_user_nice(current, 0);
180 update_poll_jiffies();
181
182 for ( ; ; ) {
183 while (schedule_timeout_interruptible(hung_task_poll_jiffies));
184 check_hung_uninterruptible_tasks();
185 }
186
187 return 0;
188}
189
190static int __init hung_task_init(void)
191{
192 atomic_notifier_chain_register(&panic_notifier_list, &panic_block);
193 watchdog_task = kthread_run(watchdog, NULL, "khungtaskd");
194
195 return 0;
196}
197
198module_init(hung_task_init);