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>
20 #include <linux/sched/debug.h>
22 #include <trace/events/sched.h>
25 * The number of tasks checked:
27 int __read_mostly sysctl_hung_task_check_count = PID_MAX_LIMIT;
30 * Limit number of tasks checked in a batch.
32 * This value controls the preemptibility of khungtaskd since preemption
33 * is disabled during the critical section. It also controls the size of
34 * the RCU grace period. So it needs to be upper-bound.
36 #define HUNG_TASK_BATCHING 1024
39 * Zero means infinite timeout - no checking done:
41 unsigned long __read_mostly sysctl_hung_task_timeout_secs = CONFIG_DEFAULT_HUNG_TASK_TIMEOUT;
43 int __read_mostly sysctl_hung_task_warnings = 10;
45 static int __read_mostly did_panic;
46 static bool hung_task_show_lock;
48 static struct task_struct *watchdog_task;
51 * Should we panic (and reboot, if panic_timeout= is set) when a
52 * hung task is detected:
54 unsigned int __read_mostly sysctl_hung_task_panic =
55 CONFIG_BOOTPARAM_HUNG_TASK_PANIC_VALUE;
57 static int __init hung_task_panic_setup(char *str)
59 int rc = kstrtouint(str, 0, &sysctl_hung_task_panic);
65 __setup("hung_task_panic=", hung_task_panic_setup);
68 hung_task_panic(struct notifier_block *this, unsigned long event, void *ptr)
75 static struct notifier_block panic_block = {
76 .notifier_call = hung_task_panic,
79 static void check_hung_task(struct task_struct *t, unsigned long timeout)
81 unsigned long switch_count = t->nvcsw + t->nivcsw;
84 * Ensure the task is not frozen.
85 * Also, skip vfork and any other user process that freezer should skip.
87 if (unlikely(t->flags & (PF_FROZEN | PF_FREEZER_SKIP)))
91 * When a freshly created task is scheduled once, changes its state to
92 * TASK_UNINTERRUPTIBLE without having ever been switched out once, it
95 if (unlikely(!switch_count))
98 if (switch_count != t->last_switch_count) {
99 t->last_switch_count = switch_count;
103 trace_sched_process_hang(t);
105 if (!sysctl_hung_task_warnings && !sysctl_hung_task_panic)
109 * Ok, the task did not get scheduled for more than 2 minutes,
112 if (sysctl_hung_task_warnings) {
113 if (sysctl_hung_task_warnings > 0)
114 sysctl_hung_task_warnings--;
115 pr_err("INFO: task %s:%d blocked for more than %ld seconds.\n",
116 t->comm, t->pid, timeout);
117 pr_err(" %s %s %.*s\n",
118 print_tainted(), init_utsname()->release,
119 (int)strcspn(init_utsname()->version, " "),
120 init_utsname()->version);
121 pr_err("\"echo 0 > /proc/sys/kernel/hung_task_timeout_secs\""
122 " disables this message.\n");
124 hung_task_show_lock = true;
127 touch_nmi_watchdog();
129 if (sysctl_hung_task_panic) {
130 if (hung_task_show_lock)
131 debug_show_all_locks();
132 trigger_all_cpu_backtrace();
133 panic("hung_task: blocked tasks");
138 * To avoid extending the RCU grace period for an unbounded amount of time,
139 * periodically exit the critical section and enter a new one.
141 * For preemptible RCU it is sufficient to call rcu_read_unlock in order
142 * to exit the grace period. For classic RCU, a reschedule is required.
144 static bool rcu_lock_break(struct task_struct *g, struct task_struct *t)
153 can_cont = pid_alive(g) && pid_alive(t);
161 * Check whether a TASK_UNINTERRUPTIBLE does not get woken up for
162 * a really long time (120 seconds). If that happens, print out
165 static void check_hung_uninterruptible_tasks(unsigned long timeout)
167 int max_count = sysctl_hung_task_check_count;
168 int batch_count = HUNG_TASK_BATCHING;
169 struct task_struct *g, *t;
172 * If the system crashed already then all bets are off,
173 * do not report extra hung tasks:
175 if (test_taint(TAINT_DIE) || did_panic)
178 hung_task_show_lock = false;
180 for_each_process_thread(g, t) {
183 if (!--batch_count) {
184 batch_count = HUNG_TASK_BATCHING;
185 if (!rcu_lock_break(g, t))
188 /* use "==" to skip the TASK_KILLABLE tasks waiting on NFS */
189 if (t->state == TASK_UNINTERRUPTIBLE)
190 check_hung_task(t, timeout);
194 if (hung_task_show_lock)
195 debug_show_all_locks();
198 static long hung_timeout_jiffies(unsigned long last_checked,
199 unsigned long timeout)
201 /* timeout of 0 will disable the watchdog */
202 return timeout ? last_checked - jiffies + timeout * HZ :
203 MAX_SCHEDULE_TIMEOUT;
207 * Process updating of timeout sysctl
209 int proc_dohung_task_timeout_secs(struct ctl_table *table, int write,
211 size_t *lenp, loff_t *ppos)
215 ret = proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
220 wake_up_process(watchdog_task);
226 static atomic_t reset_hung_task = ATOMIC_INIT(0);
228 void reset_hung_task_detector(void)
230 atomic_set(&reset_hung_task, 1);
232 EXPORT_SYMBOL_GPL(reset_hung_task_detector);
235 * kthread which checks for tasks stuck in D state
237 static int watchdog(void *dummy)
239 unsigned long hung_last_checked = jiffies;
241 set_user_nice(current, 0);
244 unsigned long timeout = sysctl_hung_task_timeout_secs;
245 long t = hung_timeout_jiffies(hung_last_checked, timeout);
248 if (!atomic_xchg(&reset_hung_task, 0))
249 check_hung_uninterruptible_tasks(timeout);
250 hung_last_checked = jiffies;
253 schedule_timeout_interruptible(t);
259 static int __init hung_task_init(void)
261 atomic_notifier_chain_register(&panic_notifier_list, &panic_block);
262 watchdog_task = kthread_run(watchdog, NULL, "khungtaskd");
266 subsys_initcall(hung_task_init);