4 * kernel/hung_task.c - kernel thread for detecting tasks stuck in D state
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 int __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 int rc = kstrtouint(str, 0, &sysctl_hung_task_panic);
61 __setup("hung_task_panic=", hung_task_panic_setup);
64 hung_task_panic(struct notifier_block *this, unsigned long event, void *ptr)
71 static struct notifier_block panic_block = {
72 .notifier_call = hung_task_panic,
75 static void check_hung_task(struct task_struct *t, unsigned long timeout)
77 unsigned long switch_count = t->nvcsw + t->nivcsw;
80 * Ensure the task is not frozen.
81 * Also, skip vfork and any other user process that freezer should skip.
83 if (unlikely(t->flags & (PF_FROZEN | PF_FREEZER_SKIP)))
87 * When a freshly created task is scheduled once, changes its state to
88 * TASK_UNINTERRUPTIBLE without having ever been switched out once, it
91 if (unlikely(!switch_count))
94 if (switch_count != t->last_switch_count) {
95 t->last_switch_count = switch_count;
99 trace_sched_process_hang(t);
101 if (!sysctl_hung_task_warnings)
104 if (sysctl_hung_task_warnings > 0)
105 sysctl_hung_task_warnings--;
108 * Ok, the task did not get scheduled for more than 2 minutes,
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");
120 debug_show_held_locks(t);
122 touch_nmi_watchdog();
124 if (sysctl_hung_task_panic) {
125 trigger_all_cpu_backtrace();
126 panic("hung_task: blocked tasks");
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.
134 * For preemptible RCU it is sufficient to call rcu_read_unlock in order
135 * to exit the grace period. For classic RCU, a reschedule is required.
137 static bool rcu_lock_break(struct task_struct *g, struct task_struct *t)
146 can_cont = pid_alive(g) && pid_alive(t);
154 * Check whether a TASK_UNINTERRUPTIBLE does not get woken up for
155 * a really long time (120 seconds). If that happens, print out
158 static void check_hung_uninterruptible_tasks(unsigned long timeout)
160 int max_count = sysctl_hung_task_check_count;
161 int batch_count = HUNG_TASK_BATCHING;
162 struct task_struct *g, *t;
165 * If the system crashed already then all bets are off,
166 * do not report extra hung tasks:
168 if (test_taint(TAINT_DIE) || did_panic)
172 for_each_process_thread(g, t) {
175 if (!--batch_count) {
176 batch_count = HUNG_TASK_BATCHING;
177 if (!rcu_lock_break(g, t))
180 /* use "==" to skip the TASK_KILLABLE tasks waiting on NFS */
181 if (t->state == TASK_UNINTERRUPTIBLE)
182 check_hung_task(t, timeout);
188 static unsigned long timeout_jiffies(unsigned long timeout)
190 /* timeout of 0 will disable the watchdog */
191 return timeout ? timeout * HZ : MAX_SCHEDULE_TIMEOUT;
195 * Process updating of timeout sysctl
197 int proc_dohung_task_timeout_secs(struct ctl_table *table, int write,
199 size_t *lenp, loff_t *ppos)
203 ret = proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
208 wake_up_process(watchdog_task);
214 static atomic_t reset_hung_task = ATOMIC_INIT(0);
216 void reset_hung_task_detector(void)
218 atomic_set(&reset_hung_task, 1);
220 EXPORT_SYMBOL_GPL(reset_hung_task_detector);
223 * kthread which checks for tasks stuck in D state
225 static int watchdog(void *dummy)
227 set_user_nice(current, 0);
230 unsigned long timeout = sysctl_hung_task_timeout_secs;
232 while (schedule_timeout_interruptible(timeout_jiffies(timeout)))
233 timeout = sysctl_hung_task_timeout_secs;
235 if (atomic_xchg(&reset_hung_task, 0))
238 check_hung_uninterruptible_tasks(timeout);
244 static int __init hung_task_init(void)
246 atomic_notifier_chain_register(&panic_notifier_list, &panic_block);
247 watchdog_task = kthread_run(watchdog, NULL, "khungtaskd");
251 subsys_initcall(hung_task_init);