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 <linux/sched/signal.h>
21 #include <trace/events/sched.h>
24 * The number of tasks checked:
26 int __read_mostly sysctl_hung_task_check_count = PID_MAX_LIMIT;
29 * Limit number of tasks checked in a batch.
31 * This value controls the preemptibility of khungtaskd since preemption
32 * is disabled during the critical section. It also controls the size of
33 * the RCU grace period. So it needs to be upper-bound.
35 #define HUNG_TASK_BATCHING 1024
38 * Zero means infinite timeout - no checking done:
40 unsigned long __read_mostly sysctl_hung_task_timeout_secs = CONFIG_DEFAULT_HUNG_TASK_TIMEOUT;
42 int __read_mostly sysctl_hung_task_warnings = 10;
44 static int __read_mostly did_panic;
46 static struct task_struct *watchdog_task;
49 * Should we panic (and reboot, if panic_timeout= is set) when a
50 * hung task is detected:
52 unsigned int __read_mostly sysctl_hung_task_panic =
53 CONFIG_BOOTPARAM_HUNG_TASK_PANIC_VALUE;
55 static int __init hung_task_panic_setup(char *str)
57 int rc = kstrtouint(str, 0, &sysctl_hung_task_panic);
63 __setup("hung_task_panic=", hung_task_panic_setup);
66 hung_task_panic(struct notifier_block *this, unsigned long event, void *ptr)
73 static struct notifier_block panic_block = {
74 .notifier_call = hung_task_panic,
77 static void check_hung_task(struct task_struct *t, unsigned long timeout)
79 unsigned long switch_count = t->nvcsw + t->nivcsw;
82 * Ensure the task is not frozen.
83 * Also, skip vfork and any other user process that freezer should skip.
85 if (unlikely(t->flags & (PF_FROZEN | PF_FREEZER_SKIP)))
89 * When a freshly created task is scheduled once, changes its state to
90 * TASK_UNINTERRUPTIBLE without having ever been switched out once, it
93 if (unlikely(!switch_count))
96 if (switch_count != t->last_switch_count) {
97 t->last_switch_count = switch_count;
101 trace_sched_process_hang(t);
103 if (!sysctl_hung_task_warnings && !sysctl_hung_task_panic)
107 * Ok, the task did not get scheduled for more than 2 minutes,
110 if (sysctl_hung_task_warnings) {
111 if (sysctl_hung_task_warnings > 0)
112 sysctl_hung_task_warnings--;
113 pr_err("INFO: task %s:%d blocked for more than %ld seconds.\n",
114 t->comm, t->pid, timeout);
115 pr_err(" %s %s %.*s\n",
116 print_tainted(), init_utsname()->release,
117 (int)strcspn(init_utsname()->version, " "),
118 init_utsname()->version);
119 pr_err("\"echo 0 > /proc/sys/kernel/hung_task_timeout_secs\""
120 " disables this message.\n");
122 debug_show_all_locks();
125 touch_nmi_watchdog();
127 if (sysctl_hung_task_panic) {
128 trigger_all_cpu_backtrace();
129 panic("hung_task: blocked tasks");
134 * To avoid extending the RCU grace period for an unbounded amount of time,
135 * periodically exit the critical section and enter a new one.
137 * For preemptible RCU it is sufficient to call rcu_read_unlock in order
138 * to exit the grace period. For classic RCU, a reschedule is required.
140 static bool rcu_lock_break(struct task_struct *g, struct task_struct *t)
149 can_cont = pid_alive(g) && pid_alive(t);
157 * Check whether a TASK_UNINTERRUPTIBLE does not get woken up for
158 * a really long time (120 seconds). If that happens, print out
161 static void check_hung_uninterruptible_tasks(unsigned long timeout)
163 int max_count = sysctl_hung_task_check_count;
164 int batch_count = HUNG_TASK_BATCHING;
165 struct task_struct *g, *t;
168 * If the system crashed already then all bets are off,
169 * do not report extra hung tasks:
171 if (test_taint(TAINT_DIE) || did_panic)
175 for_each_process_thread(g, t) {
178 if (!--batch_count) {
179 batch_count = HUNG_TASK_BATCHING;
180 if (!rcu_lock_break(g, t))
183 /* use "==" to skip the TASK_KILLABLE tasks waiting on NFS */
184 if (t->state == TASK_UNINTERRUPTIBLE)
185 check_hung_task(t, timeout);
191 static long hung_timeout_jiffies(unsigned long last_checked,
192 unsigned long timeout)
194 /* timeout of 0 will disable the watchdog */
195 return timeout ? last_checked - jiffies + timeout * HZ :
196 MAX_SCHEDULE_TIMEOUT;
200 * Process updating of timeout sysctl
202 int proc_dohung_task_timeout_secs(struct ctl_table *table, int write,
204 size_t *lenp, loff_t *ppos)
208 ret = proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
213 wake_up_process(watchdog_task);
219 static atomic_t reset_hung_task = ATOMIC_INIT(0);
221 void reset_hung_task_detector(void)
223 atomic_set(&reset_hung_task, 1);
225 EXPORT_SYMBOL_GPL(reset_hung_task_detector);
228 * kthread which checks for tasks stuck in D state
230 static int watchdog(void *dummy)
232 unsigned long hung_last_checked = jiffies;
234 set_user_nice(current, 0);
237 unsigned long timeout = sysctl_hung_task_timeout_secs;
238 long t = hung_timeout_jiffies(hung_last_checked, timeout);
241 if (!atomic_xchg(&reset_hung_task, 0))
242 check_hung_uninterruptible_tasks(timeout);
243 hung_last_checked = jiffies;
246 schedule_timeout_interruptible(t);
252 static int __init hung_task_init(void)
254 atomic_notifier_chain_register(&panic_notifier_list, &panic_block);
255 watchdog_task = kthread_run(watchdog, NULL, "khungtaskd");
259 subsys_initcall(hung_task_init);