kthread: rename probe_kthread_data() to kthread_probe_data()
[linux-2.6-block.git] / kernel / hung_task.c
CommitLineData
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>
9984de1a 16#include <linux/export.h>
e162b39a 17#include <linux/sysctl.h>
41e85ce8 18#include <linux/utsname.h>
6a716c90 19#include <trace/events/sched.h>
e162b39a
MSB
20
21/*
ce9dbe24 22 * The number of tasks checked:
e162b39a 23 */
cd64647f 24int __read_mostly sysctl_hung_task_check_count = PID_MAX_LIMIT;
ce9dbe24
MSB
25
26/*
27 * Limit number of tasks checked in a batch.
28 *
29 * This value controls the preemptibility of khungtaskd since preemption
30 * is disabled during the critical section. It also controls the size of
31 * the RCU grace period. So it needs to be upper-bound.
32 */
33#define HUNG_TASK_BATCHING 1024
e162b39a
MSB
34
35/*
36 * Zero means infinite timeout - no checking done:
37 */
e11feaa1 38unsigned long __read_mostly sysctl_hung_task_timeout_secs = CONFIG_DEFAULT_HUNG_TASK_TIMEOUT;
e162b39a 39
270750db 40int __read_mostly sysctl_hung_task_warnings = 10;
e162b39a
MSB
41
42static int __read_mostly did_panic;
43
44static struct task_struct *watchdog_task;
45
46/*
47 * Should we panic (and reboot, if panic_timeout= is set) when a
48 * hung task is detected:
49 */
50unsigned int __read_mostly sysctl_hung_task_panic =
51 CONFIG_BOOTPARAM_HUNG_TASK_PANIC_VALUE;
52
53static int __init hung_task_panic_setup(char *str)
54{
b51dbec6 55 int rc = kstrtouint(str, 0, &sysctl_hung_task_panic);
e162b39a 56
b51dbec6
FF
57 if (rc)
58 return rc;
e162b39a
MSB
59 return 1;
60}
61__setup("hung_task_panic=", hung_task_panic_setup);
62
63static int
64hung_task_panic(struct notifier_block *this, unsigned long event, void *ptr)
65{
66 did_panic = 1;
67
68 return NOTIFY_DONE;
69}
70
71static struct notifier_block panic_block = {
72 .notifier_call = hung_task_panic,
73};
74
17406b82 75static void check_hung_task(struct task_struct *t, unsigned long timeout)
e162b39a
MSB
76{
77 unsigned long switch_count = t->nvcsw + t->nivcsw;
78
cf2592f5
FW
79 /*
80 * Ensure the task is not frozen.
f9fab10b 81 * Also, skip vfork and any other user process that freezer should skip.
cf2592f5 82 */
f9fab10b
MSB
83 if (unlikely(t->flags & (PF_FROZEN | PF_FREEZER_SKIP)))
84 return;
85
86 /*
87 * When a freshly created task is scheduled once, changes its state to
88 * TASK_UNINTERRUPTIBLE without having ever been switched out once, it
89 * musn't be checked.
90 */
91 if (unlikely(!switch_count))
e162b39a
MSB
92 return;
93
17406b82 94 if (switch_count != t->last_switch_count) {
e162b39a 95 t->last_switch_count = switch_count;
e162b39a
MSB
96 return;
97 }
6a716c90
ON
98
99 trace_sched_process_hang(t);
100
e162b39a
MSB
101 if (!sysctl_hung_task_warnings)
102 return;
270750db
AT
103
104 if (sysctl_hung_task_warnings > 0)
105 sysctl_hung_task_warnings--;
e162b39a
MSB
106
107 /*
108 * Ok, the task did not get scheduled for more than 2 minutes,
109 * complain:
110 */
41e85ce8
ON
111 pr_err("INFO: task %s:%d blocked for more than %ld seconds.\n",
112 t->comm, t->pid, timeout);
113 pr_err(" %s %s %.*s\n",
114 print_tainted(), init_utsname()->release,
115 (int)strcspn(init_utsname()->version, " "),
116 init_utsname()->version);
117 pr_err("\"echo 0 > /proc/sys/kernel/hung_task_timeout_secs\""
118 " disables this message.\n");
e162b39a 119 sched_show_task(t);
b2d4c2ed 120 debug_show_all_locks();
e162b39a 121
e162b39a
MSB
122 touch_nmi_watchdog();
123
625056b6
SL
124 if (sysctl_hung_task_panic) {
125 trigger_all_cpu_backtrace();
e162b39a 126 panic("hung_task: blocked tasks");
625056b6 127 }
e162b39a
MSB
128}
129
ce9dbe24
MSB
130/*
131 * To avoid extending the RCU grace period for an unbounded amount of time,
132 * periodically exit the critical section and enter a new one.
133 *
134 * For preemptible RCU it is sufficient to call rcu_read_unlock in order
6a103b0d 135 * to exit the grace period. For classic RCU, a reschedule is required.
ce9dbe24 136 */
6027ce49 137static bool rcu_lock_break(struct task_struct *g, struct task_struct *t)
ce9dbe24 138{
6027ce49
ON
139 bool can_cont;
140
ce9dbe24
MSB
141 get_task_struct(g);
142 get_task_struct(t);
143 rcu_read_unlock();
144 cond_resched();
145 rcu_read_lock();
6027ce49 146 can_cont = pid_alive(g) && pid_alive(t);
ce9dbe24
MSB
147 put_task_struct(t);
148 put_task_struct(g);
6027ce49
ON
149
150 return can_cont;
ce9dbe24
MSB
151}
152
e162b39a
MSB
153/*
154 * Check whether a TASK_UNINTERRUPTIBLE does not get woken up for
155 * a really long time (120 seconds). If that happens, print out
156 * a warning.
157 */
603a148f 158static void check_hung_uninterruptible_tasks(unsigned long timeout)
e162b39a
MSB
159{
160 int max_count = sysctl_hung_task_check_count;
ce9dbe24 161 int batch_count = HUNG_TASK_BATCHING;
e162b39a
MSB
162 struct task_struct *g, *t;
163
164 /*
165 * If the system crashed already then all bets are off,
166 * do not report extra hung tasks:
167 */
168 if (test_taint(TAINT_DIE) || did_panic)
169 return;
170
94be52dc 171 rcu_read_lock();
972fae69 172 for_each_process_thread(g, t) {
e5af0226 173 if (!max_count--)
e162b39a 174 goto unlock;
ce9dbe24
MSB
175 if (!--batch_count) {
176 batch_count = HUNG_TASK_BATCHING;
6027ce49 177 if (!rcu_lock_break(g, t))
ce9dbe24
MSB
178 goto unlock;
179 }
e162b39a
MSB
180 /* use "==" to skip the TASK_KILLABLE tasks waiting on NFS */
181 if (t->state == TASK_UNINTERRUPTIBLE)
17406b82 182 check_hung_task(t, timeout);
972fae69 183 }
e162b39a 184 unlock:
94be52dc 185 rcu_read_unlock();
e162b39a
MSB
186}
187
b4aa14a6
TH
188static long hung_timeout_jiffies(unsigned long last_checked,
189 unsigned long timeout)
e162b39a
MSB
190{
191 /* timeout of 0 will disable the watchdog */
b4aa14a6
TH
192 return timeout ? last_checked - jiffies + timeout * HZ :
193 MAX_SCHEDULE_TIMEOUT;
e162b39a
MSB
194}
195
196/*
197 * Process updating of timeout sysctl
198 */
199int proc_dohung_task_timeout_secs(struct ctl_table *table, int write,
8d65af78 200 void __user *buffer,
e162b39a
MSB
201 size_t *lenp, loff_t *ppos)
202{
203 int ret;
204
8d65af78 205 ret = proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
e162b39a
MSB
206
207 if (ret || !write)
208 goto out;
209
e162b39a
MSB
210 wake_up_process(watchdog_task);
211
212 out:
213 return ret;
214}
215
8b414521
MT
216static atomic_t reset_hung_task = ATOMIC_INIT(0);
217
218void reset_hung_task_detector(void)
219{
220 atomic_set(&reset_hung_task, 1);
221}
222EXPORT_SYMBOL_GPL(reset_hung_task_detector);
223
e162b39a
MSB
224/*
225 * kthread which checks for tasks stuck in D state
226 */
227static int watchdog(void *dummy)
228{
b4aa14a6
TH
229 unsigned long hung_last_checked = jiffies;
230
e162b39a 231 set_user_nice(current, 0);
e162b39a
MSB
232
233 for ( ; ; ) {
17406b82 234 unsigned long timeout = sysctl_hung_task_timeout_secs;
b4aa14a6 235 long t = hung_timeout_jiffies(hung_last_checked, timeout);
603a148f 236
b4aa14a6
TH
237 if (t <= 0) {
238 if (!atomic_xchg(&reset_hung_task, 0))
239 check_hung_uninterruptible_tasks(timeout);
240 hung_last_checked = jiffies;
8b414521 241 continue;
b4aa14a6
TH
242 }
243 schedule_timeout_interruptible(t);
e162b39a
MSB
244 }
245
246 return 0;
247}
248
249static int __init hung_task_init(void)
250{
251 atomic_notifier_chain_register(&panic_notifier_list, &panic_block);
252 watchdog_task = kthread_run(watchdog, NULL, "khungtaskd");
253
254 return 0;
255}
c96d6660 256subsys_initcall(hung_task_init);