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 sysctl_hung_task_panic = simple_strtoul(str, NULL, 0);
59 __setup("hung_task_panic=", hung_task_panic_setup);
62 hung_task_panic(struct notifier_block *this, unsigned long event, void *ptr)
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)))
85 * When a freshly created task is scheduled once, changes its state to
86 * TASK_UNINTERRUPTIBLE without having ever been switched out once, it
89 if (unlikely(!switch_count))
92 if (switch_count != t->last_switch_count) {
93 t->last_switch_count = switch_count;
97 trace_sched_process_hang(t);
99 if (!sysctl_hung_task_warnings)
102 if (sysctl_hung_task_warnings > 0)
103 sysctl_hung_task_warnings--;
106 * Ok, the task did not get scheduled for more than 2 minutes,
109 pr_err("INFO: task %s:%d blocked for more than %ld seconds.\n",
110 t->comm, t->pid, timeout);
111 pr_err(" %s %s %.*s\n",
112 print_tainted(), init_utsname()->release,
113 (int)strcspn(init_utsname()->version, " "),
114 init_utsname()->version);
115 pr_err("\"echo 0 > /proc/sys/kernel/hung_task_timeout_secs\""
116 " disables this message.\n");
118 debug_show_held_locks(t);
120 touch_nmi_watchdog();
122 if (sysctl_hung_task_panic) {
123 trigger_all_cpu_backtrace();
124 panic("hung_task: blocked tasks");
129 * To avoid extending the RCU grace period for an unbounded amount of time,
130 * periodically exit the critical section and enter a new one.
132 * For preemptible RCU it is sufficient to call rcu_read_unlock in order
133 * to exit the grace period. For classic RCU, a reschedule is required.
135 static bool rcu_lock_break(struct task_struct *g, struct task_struct *t)
144 can_cont = pid_alive(g) && pid_alive(t);
152 * Check whether a TASK_UNINTERRUPTIBLE does not get woken up for
153 * a really long time (120 seconds). If that happens, print out
156 static void check_hung_uninterruptible_tasks(unsigned long timeout)
158 int max_count = sysctl_hung_task_check_count;
159 int batch_count = HUNG_TASK_BATCHING;
160 struct task_struct *g, *t;
163 * If the system crashed already then all bets are off,
164 * do not report extra hung tasks:
166 if (test_taint(TAINT_DIE) || did_panic)
170 do_each_thread(g, t) {
173 if (!--batch_count) {
174 batch_count = HUNG_TASK_BATCHING;
175 if (!rcu_lock_break(g, t))
178 /* use "==" to skip the TASK_KILLABLE tasks waiting on NFS */
179 if (t->state == TASK_UNINTERRUPTIBLE)
180 check_hung_task(t, timeout);
181 } while_each_thread(g, t);
186 static unsigned long timeout_jiffies(unsigned long timeout)
188 /* timeout of 0 will disable the watchdog */
189 return timeout ? timeout * HZ : MAX_SCHEDULE_TIMEOUT;
193 * Process updating of timeout sysctl
195 int proc_dohung_task_timeout_secs(struct ctl_table *table, int write,
197 size_t *lenp, loff_t *ppos)
201 ret = proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
206 wake_up_process(watchdog_task);
212 static atomic_t reset_hung_task = ATOMIC_INIT(0);
214 void reset_hung_task_detector(void)
216 atomic_set(&reset_hung_task, 1);
218 EXPORT_SYMBOL_GPL(reset_hung_task_detector);
221 * kthread which checks for tasks stuck in D state
223 static int watchdog(void *dummy)
225 set_user_nice(current, 0);
228 unsigned long timeout = sysctl_hung_task_timeout_secs;
230 while (schedule_timeout_interruptible(timeout_jiffies(timeout)))
231 timeout = sysctl_hung_task_timeout_secs;
233 if (atomic_xchg(&reset_hung_task, 0))
236 check_hung_uninterruptible_tasks(timeout);
242 static int __init hung_task_init(void)
244 atomic_notifier_chain_register(&panic_notifier_list, &panic_block);
245 watchdog_task = kthread_run(watchdog, NULL, "khungtaskd");
250 module_init(hung_task_init);