]>
Commit | Line | Data |
---|---|---|
e162b39a MSB |
1 | /* |
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 | /* | |
ce9dbe24 | 20 | * The number of tasks checked: |
e162b39a | 21 | */ |
ce9dbe24 MSB |
22 | unsigned long __read_mostly sysctl_hung_task_check_count = PID_MAX_LIMIT; |
23 | ||
24 | /* | |
25 | * Limit number of tasks checked in a batch. | |
26 | * | |
27 | * This value controls the preemptibility of khungtaskd since preemption | |
28 | * is disabled during the critical section. It also controls the size of | |
29 | * the RCU grace period. So it needs to be upper-bound. | |
30 | */ | |
31 | #define HUNG_TASK_BATCHING 1024 | |
e162b39a MSB |
32 | |
33 | /* | |
34 | * Zero means infinite timeout - no checking done: | |
35 | */ | |
36 | unsigned long __read_mostly sysctl_hung_task_timeout_secs = 120; | |
e162b39a MSB |
37 | |
38 | unsigned long __read_mostly sysctl_hung_task_warnings = 10; | |
39 | ||
40 | static int __read_mostly did_panic; | |
41 | ||
42 | static struct task_struct *watchdog_task; | |
43 | ||
44 | /* | |
45 | * Should we panic (and reboot, if panic_timeout= is set) when a | |
46 | * hung task is detected: | |
47 | */ | |
48 | unsigned int __read_mostly sysctl_hung_task_panic = | |
49 | CONFIG_BOOTPARAM_HUNG_TASK_PANIC_VALUE; | |
50 | ||
51 | static int __init hung_task_panic_setup(char *str) | |
52 | { | |
53 | sysctl_hung_task_panic = simple_strtoul(str, NULL, 0); | |
54 | ||
55 | return 1; | |
56 | } | |
57 | __setup("hung_task_panic=", hung_task_panic_setup); | |
58 | ||
59 | static int | |
60 | hung_task_panic(struct notifier_block *this, unsigned long event, void *ptr) | |
61 | { | |
62 | did_panic = 1; | |
63 | ||
64 | return NOTIFY_DONE; | |
65 | } | |
66 | ||
67 | static struct notifier_block panic_block = { | |
68 | .notifier_call = hung_task_panic, | |
69 | }; | |
70 | ||
17406b82 | 71 | static void check_hung_task(struct task_struct *t, unsigned long timeout) |
e162b39a MSB |
72 | { |
73 | unsigned long switch_count = t->nvcsw + t->nivcsw; | |
74 | ||
75 | if (t->flags & PF_FROZEN) | |
76 | return; | |
77 | ||
17406b82 | 78 | if (switch_count != t->last_switch_count) { |
e162b39a | 79 | t->last_switch_count = switch_count; |
e162b39a MSB |
80 | return; |
81 | } | |
e162b39a MSB |
82 | if (!sysctl_hung_task_warnings) |
83 | return; | |
84 | sysctl_hung_task_warnings--; | |
85 | ||
86 | /* | |
87 | * Ok, the task did not get scheduled for more than 2 minutes, | |
88 | * complain: | |
89 | */ | |
90 | printk(KERN_ERR "INFO: task %s:%d blocked for more than " | |
603a148f | 91 | "%ld seconds.\n", t->comm, t->pid, timeout); |
e162b39a MSB |
92 | printk(KERN_ERR "\"echo 0 > /proc/sys/kernel/hung_task_timeout_secs\"" |
93 | " disables this message.\n"); | |
94 | sched_show_task(t); | |
95 | __debug_show_held_locks(t); | |
96 | ||
e162b39a MSB |
97 | touch_nmi_watchdog(); |
98 | ||
99 | if (sysctl_hung_task_panic) | |
100 | panic("hung_task: blocked tasks"); | |
101 | } | |
102 | ||
ce9dbe24 MSB |
103 | /* |
104 | * To avoid extending the RCU grace period for an unbounded amount of time, | |
105 | * periodically exit the critical section and enter a new one. | |
106 | * | |
107 | * For preemptible RCU it is sufficient to call rcu_read_unlock in order | |
108 | * exit the grace period. For classic RCU, a reschedule is required. | |
109 | */ | |
110 | static void rcu_lock_break(struct task_struct *g, struct task_struct *t) | |
111 | { | |
112 | get_task_struct(g); | |
113 | get_task_struct(t); | |
114 | rcu_read_unlock(); | |
115 | cond_resched(); | |
116 | rcu_read_lock(); | |
117 | put_task_struct(t); | |
118 | put_task_struct(g); | |
119 | } | |
120 | ||
e162b39a MSB |
121 | /* |
122 | * Check whether a TASK_UNINTERRUPTIBLE does not get woken up for | |
123 | * a really long time (120 seconds). If that happens, print out | |
124 | * a warning. | |
125 | */ | |
603a148f | 126 | static void check_hung_uninterruptible_tasks(unsigned long timeout) |
e162b39a MSB |
127 | { |
128 | int max_count = sysctl_hung_task_check_count; | |
ce9dbe24 | 129 | int batch_count = HUNG_TASK_BATCHING; |
e162b39a MSB |
130 | struct task_struct *g, *t; |
131 | ||
132 | /* | |
133 | * If the system crashed already then all bets are off, | |
134 | * do not report extra hung tasks: | |
135 | */ | |
136 | if (test_taint(TAINT_DIE) || did_panic) | |
137 | return; | |
138 | ||
94be52dc | 139 | rcu_read_lock(); |
e162b39a MSB |
140 | do_each_thread(g, t) { |
141 | if (!--max_count) | |
142 | goto unlock; | |
ce9dbe24 MSB |
143 | if (!--batch_count) { |
144 | batch_count = HUNG_TASK_BATCHING; | |
145 | rcu_lock_break(g, t); | |
146 | /* Exit if t or g was unhashed during refresh. */ | |
147 | if (t->state == TASK_DEAD || g->state == TASK_DEAD) | |
148 | goto unlock; | |
149 | } | |
e162b39a MSB |
150 | /* use "==" to skip the TASK_KILLABLE tasks waiting on NFS */ |
151 | if (t->state == TASK_UNINTERRUPTIBLE) | |
17406b82 | 152 | check_hung_task(t, timeout); |
e162b39a MSB |
153 | } while_each_thread(g, t); |
154 | unlock: | |
94be52dc | 155 | rcu_read_unlock(); |
e162b39a MSB |
156 | } |
157 | ||
17406b82 | 158 | static unsigned long timeout_jiffies(unsigned long timeout) |
e162b39a MSB |
159 | { |
160 | /* timeout of 0 will disable the watchdog */ | |
17406b82 | 161 | return timeout ? timeout * HZ : MAX_SCHEDULE_TIMEOUT; |
e162b39a MSB |
162 | } |
163 | ||
164 | /* | |
165 | * Process updating of timeout sysctl | |
166 | */ | |
167 | int proc_dohung_task_timeout_secs(struct ctl_table *table, int write, | |
168 | struct file *filp, void __user *buffer, | |
169 | size_t *lenp, loff_t *ppos) | |
170 | { | |
171 | int ret; | |
172 | ||
173 | ret = proc_doulongvec_minmax(table, write, filp, buffer, lenp, ppos); | |
174 | ||
175 | if (ret || !write) | |
176 | goto out; | |
177 | ||
e162b39a MSB |
178 | wake_up_process(watchdog_task); |
179 | ||
180 | out: | |
181 | return ret; | |
182 | } | |
183 | ||
184 | /* | |
185 | * kthread which checks for tasks stuck in D state | |
186 | */ | |
187 | static int watchdog(void *dummy) | |
188 | { | |
189 | set_user_nice(current, 0); | |
e162b39a MSB |
190 | |
191 | for ( ; ; ) { | |
17406b82 | 192 | unsigned long timeout = sysctl_hung_task_timeout_secs; |
603a148f | 193 | |
17406b82 MSB |
194 | while (schedule_timeout_interruptible(timeout_jiffies(timeout))) |
195 | timeout = sysctl_hung_task_timeout_secs; | |
603a148f | 196 | |
17406b82 | 197 | check_hung_uninterruptible_tasks(timeout); |
e162b39a MSB |
198 | } |
199 | ||
200 | return 0; | |
201 | } | |
202 | ||
203 | static int __init hung_task_init(void) | |
204 | { | |
205 | atomic_notifier_chain_register(&panic_notifier_list, &panic_block); | |
206 | watchdog_task = kthread_run(watchdog, NULL, "khungtaskd"); | |
207 | ||
208 | return 0; | |
209 | } | |
210 | ||
211 | module_init(hung_task_init); |