Merge tag 'gpio-v3.13-3' of git://git.kernel.org/pub/scm/linux/kernel/git/linusw...
[linux-2.6.git] / kernel / hung_task.c
blob9328b80eaf14c347bb188bee856154ea07b86168
1 /*
2 * Detect Hung Task
4 * kernel/hung_task.c - kernel thread for detecting tasks stuck in D state
6 */
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/export.h>
17 #include <linux/sysctl.h>
18 #include <linux/utsname.h>
19 #include <trace/events/sched.h>
22 * The number of tasks checked:
24 int __read_mostly sysctl_hung_task_check_count = PID_MAX_LIMIT;
27 * Limit number of tasks checked in a batch.
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.
33 #define HUNG_TASK_BATCHING 1024
36 * Zero means infinite timeout - no checking done:
38 unsigned long __read_mostly sysctl_hung_task_timeout_secs = CONFIG_DEFAULT_HUNG_TASK_TIMEOUT;
40 unsigned long __read_mostly sysctl_hung_task_warnings = 10;
42 static int __read_mostly did_panic;
44 static struct task_struct *watchdog_task;
47 * Should we panic (and reboot, if panic_timeout= is set) when a
48 * hung task is detected:
50 unsigned int __read_mostly sysctl_hung_task_panic =
51 CONFIG_BOOTPARAM_HUNG_TASK_PANIC_VALUE;
53 static int __init hung_task_panic_setup(char *str)
55 sysctl_hung_task_panic = simple_strtoul(str, NULL, 0);
57 return 1;
59 __setup("hung_task_panic=", hung_task_panic_setup);
61 static int
62 hung_task_panic(struct notifier_block *this, unsigned long event, void *ptr)
64 did_panic = 1;
66 return NOTIFY_DONE;
69 static struct notifier_block panic_block = {
70 .notifier_call = hung_task_panic,
73 static void check_hung_task(struct task_struct *t, unsigned long timeout)
75 unsigned long switch_count = t->nvcsw + t->nivcsw;
78 * Ensure the task is not frozen.
79 * Also, skip vfork and any other user process that freezer should skip.
81 if (unlikely(t->flags & (PF_FROZEN | PF_FREEZER_SKIP)))
82 return;
85 * When a freshly created task is scheduled once, changes its state to
86 * TASK_UNINTERRUPTIBLE without having ever been switched out once, it
87 * musn't be checked.
89 if (unlikely(!switch_count))
90 return;
92 if (switch_count != t->last_switch_count) {
93 t->last_switch_count = switch_count;
94 return;
97 trace_sched_process_hang(t);
99 if (!sysctl_hung_task_warnings)
100 return;
101 sysctl_hung_task_warnings--;
104 * Ok, the task did not get scheduled for more than 2 minutes,
105 * complain:
107 pr_err("INFO: task %s:%d blocked for more than %ld seconds.\n",
108 t->comm, t->pid, timeout);
109 pr_err(" %s %s %.*s\n",
110 print_tainted(), init_utsname()->release,
111 (int)strcspn(init_utsname()->version, " "),
112 init_utsname()->version);
113 pr_err("\"echo 0 > /proc/sys/kernel/hung_task_timeout_secs\""
114 " disables this message.\n");
115 sched_show_task(t);
116 debug_show_held_locks(t);
118 touch_nmi_watchdog();
120 if (sysctl_hung_task_panic) {
121 trigger_all_cpu_backtrace();
122 panic("hung_task: blocked tasks");
127 * To avoid extending the RCU grace period for an unbounded amount of time,
128 * periodically exit the critical section and enter a new one.
130 * For preemptible RCU it is sufficient to call rcu_read_unlock in order
131 * to exit the grace period. For classic RCU, a reschedule is required.
133 static bool rcu_lock_break(struct task_struct *g, struct task_struct *t)
135 bool can_cont;
137 get_task_struct(g);
138 get_task_struct(t);
139 rcu_read_unlock();
140 cond_resched();
141 rcu_read_lock();
142 can_cont = pid_alive(g) && pid_alive(t);
143 put_task_struct(t);
144 put_task_struct(g);
146 return can_cont;
150 * Check whether a TASK_UNINTERRUPTIBLE does not get woken up for
151 * a really long time (120 seconds). If that happens, print out
152 * a warning.
154 static void check_hung_uninterruptible_tasks(unsigned long timeout)
156 int max_count = sysctl_hung_task_check_count;
157 int batch_count = HUNG_TASK_BATCHING;
158 struct task_struct *g, *t;
161 * If the system crashed already then all bets are off,
162 * do not report extra hung tasks:
164 if (test_taint(TAINT_DIE) || did_panic)
165 return;
167 rcu_read_lock();
168 do_each_thread(g, t) {
169 if (!max_count--)
170 goto unlock;
171 if (!--batch_count) {
172 batch_count = HUNG_TASK_BATCHING;
173 if (!rcu_lock_break(g, t))
174 goto unlock;
176 /* use "==" to skip the TASK_KILLABLE tasks waiting on NFS */
177 if (t->state == TASK_UNINTERRUPTIBLE)
178 check_hung_task(t, timeout);
179 } while_each_thread(g, t);
180 unlock:
181 rcu_read_unlock();
184 static unsigned long timeout_jiffies(unsigned long timeout)
186 /* timeout of 0 will disable the watchdog */
187 return timeout ? timeout * HZ : MAX_SCHEDULE_TIMEOUT;
191 * Process updating of timeout sysctl
193 int proc_dohung_task_timeout_secs(struct ctl_table *table, int write,
194 void __user *buffer,
195 size_t *lenp, loff_t *ppos)
197 int ret;
199 ret = proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
201 if (ret || !write)
202 goto out;
204 wake_up_process(watchdog_task);
206 out:
207 return ret;
210 static atomic_t reset_hung_task = ATOMIC_INIT(0);
212 void reset_hung_task_detector(void)
214 atomic_set(&reset_hung_task, 1);
216 EXPORT_SYMBOL_GPL(reset_hung_task_detector);
219 * kthread which checks for tasks stuck in D state
221 static int watchdog(void *dummy)
223 set_user_nice(current, 0);
225 for ( ; ; ) {
226 unsigned long timeout = sysctl_hung_task_timeout_secs;
228 while (schedule_timeout_interruptible(timeout_jiffies(timeout)))
229 timeout = sysctl_hung_task_timeout_secs;
231 if (atomic_xchg(&reset_hung_task, 0))
232 continue;
234 check_hung_uninterruptible_tasks(timeout);
237 return 0;
240 static int __init hung_task_init(void)
242 atomic_notifier_chain_register(&panic_notifier_list, &panic_block);
243 watchdog_task = kthread_run(watchdog, NULL, "khungtaskd");
245 return 0;
248 module_init(hung_task_init);