Commit | Line | Data |
---|---|---|
e162b39a MSB |
1 | /* |
2 | * Detect Hung Task | |
3 | * | |
4 | * kernel/hung_task.c - kernel thread for detecting tasks stuck in D state | |
5 | * | |
6 | */ | |
7 | ||
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> | |
9984de1a | 16 | #include <linux/export.h> |
e162b39a | 17 | #include <linux/sysctl.h> |
41e85ce8 | 18 | #include <linux/utsname.h> |
6a716c90 | 19 | #include <trace/events/sched.h> |
e162b39a MSB |
20 | |
21 | /* | |
ce9dbe24 | 22 | * The number of tasks checked: |
e162b39a | 23 | */ |
cd64647f | 24 | int __read_mostly sysctl_hung_task_check_count = PID_MAX_LIMIT; |
ce9dbe24 MSB |
25 | |
26 | /* | |
27 | * Limit number of tasks checked in a batch. | |
28 | * | |
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. | |
32 | */ | |
33 | #define HUNG_TASK_BATCHING 1024 | |
e162b39a MSB |
34 | |
35 | /* | |
36 | * Zero means infinite timeout - no checking done: | |
37 | */ | |
e11feaa1 | 38 | unsigned long __read_mostly sysctl_hung_task_timeout_secs = CONFIG_DEFAULT_HUNG_TASK_TIMEOUT; |
e162b39a MSB |
39 | |
40 | unsigned long __read_mostly sysctl_hung_task_warnings = 10; | |
41 | ||
42 | static int __read_mostly did_panic; | |
43 | ||
44 | static struct task_struct *watchdog_task; | |
45 | ||
46 | /* | |
47 | * Should we panic (and reboot, if panic_timeout= is set) when a | |
48 | * hung task is detected: | |
49 | */ | |
50 | unsigned int __read_mostly sysctl_hung_task_panic = | |
51 | CONFIG_BOOTPARAM_HUNG_TASK_PANIC_VALUE; | |
52 | ||
53 | static int __init hung_task_panic_setup(char *str) | |
54 | { | |
55 | sysctl_hung_task_panic = simple_strtoul(str, NULL, 0); | |
56 | ||
57 | return 1; | |
58 | } | |
59 | __setup("hung_task_panic=", hung_task_panic_setup); | |
60 | ||
61 | static int | |
62 | hung_task_panic(struct notifier_block *this, unsigned long event, void *ptr) | |
63 | { | |
64 | did_panic = 1; | |
65 | ||
66 | return NOTIFY_DONE; | |
67 | } | |
68 | ||
69 | static struct notifier_block panic_block = { | |
70 | .notifier_call = hung_task_panic, | |
71 | }; | |
72 | ||
17406b82 | 73 | static void check_hung_task(struct task_struct *t, unsigned long timeout) |
e162b39a MSB |
74 | { |
75 | unsigned long switch_count = t->nvcsw + t->nivcsw; | |
76 | ||
cf2592f5 FW |
77 | /* |
78 | * Ensure the task is not frozen. | |
f9fab10b | 79 | * Also, skip vfork and any other user process that freezer should skip. |
cf2592f5 | 80 | */ |
f9fab10b MSB |
81 | if (unlikely(t->flags & (PF_FROZEN | PF_FREEZER_SKIP))) |
82 | return; | |
83 | ||
84 | /* | |
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. | |
88 | */ | |
89 | if (unlikely(!switch_count)) | |
e162b39a MSB |
90 | return; |
91 | ||
17406b82 | 92 | if (switch_count != t->last_switch_count) { |
e162b39a | 93 | t->last_switch_count = switch_count; |
e162b39a MSB |
94 | return; |
95 | } | |
6a716c90 ON |
96 | |
97 | trace_sched_process_hang(t); | |
98 | ||
e162b39a MSB |
99 | if (!sysctl_hung_task_warnings) |
100 | return; | |
101 | sysctl_hung_task_warnings--; | |
102 | ||
103 | /* | |
104 | * Ok, the task did not get scheduled for more than 2 minutes, | |
105 | * complain: | |
106 | */ | |
41e85ce8 ON |
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"); | |
e162b39a | 115 | sched_show_task(t); |
f1b499f0 | 116 | debug_show_held_locks(t); |
e162b39a | 117 | |
e162b39a MSB |
118 | touch_nmi_watchdog(); |
119 | ||
625056b6 SL |
120 | if (sysctl_hung_task_panic) { |
121 | trigger_all_cpu_backtrace(); | |
e162b39a | 122 | panic("hung_task: blocked tasks"); |
625056b6 | 123 | } |
e162b39a MSB |
124 | } |
125 | ||
ce9dbe24 MSB |
126 | /* |
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. | |
129 | * | |
130 | * For preemptible RCU it is sufficient to call rcu_read_unlock in order | |
6a103b0d | 131 | * to exit the grace period. For classic RCU, a reschedule is required. |
ce9dbe24 | 132 | */ |
6027ce49 | 133 | static bool rcu_lock_break(struct task_struct *g, struct task_struct *t) |
ce9dbe24 | 134 | { |
6027ce49 ON |
135 | bool can_cont; |
136 | ||
ce9dbe24 MSB |
137 | get_task_struct(g); |
138 | get_task_struct(t); | |
139 | rcu_read_unlock(); | |
140 | cond_resched(); | |
141 | rcu_read_lock(); | |
6027ce49 | 142 | can_cont = pid_alive(g) && pid_alive(t); |
ce9dbe24 MSB |
143 | put_task_struct(t); |
144 | put_task_struct(g); | |
6027ce49 ON |
145 | |
146 | return can_cont; | |
ce9dbe24 MSB |
147 | } |
148 | ||
e162b39a MSB |
149 | /* |
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. | |
153 | */ | |
603a148f | 154 | static void check_hung_uninterruptible_tasks(unsigned long timeout) |
e162b39a MSB |
155 | { |
156 | int max_count = sysctl_hung_task_check_count; | |
ce9dbe24 | 157 | int batch_count = HUNG_TASK_BATCHING; |
e162b39a MSB |
158 | struct task_struct *g, *t; |
159 | ||
160 | /* | |
161 | * If the system crashed already then all bets are off, | |
162 | * do not report extra hung tasks: | |
163 | */ | |
164 | if (test_taint(TAINT_DIE) || did_panic) | |
165 | return; | |
166 | ||
94be52dc | 167 | rcu_read_lock(); |
e162b39a | 168 | do_each_thread(g, t) { |
e5af0226 | 169 | if (!max_count--) |
e162b39a | 170 | goto unlock; |
ce9dbe24 MSB |
171 | if (!--batch_count) { |
172 | batch_count = HUNG_TASK_BATCHING; | |
6027ce49 | 173 | if (!rcu_lock_break(g, t)) |
ce9dbe24 MSB |
174 | goto unlock; |
175 | } | |
e162b39a MSB |
176 | /* use "==" to skip the TASK_KILLABLE tasks waiting on NFS */ |
177 | if (t->state == TASK_UNINTERRUPTIBLE) | |
17406b82 | 178 | check_hung_task(t, timeout); |
e162b39a MSB |
179 | } while_each_thread(g, t); |
180 | unlock: | |
94be52dc | 181 | rcu_read_unlock(); |
e162b39a MSB |
182 | } |
183 | ||
17406b82 | 184 | static unsigned long timeout_jiffies(unsigned long timeout) |
e162b39a MSB |
185 | { |
186 | /* timeout of 0 will disable the watchdog */ | |
17406b82 | 187 | return timeout ? timeout * HZ : MAX_SCHEDULE_TIMEOUT; |
e162b39a MSB |
188 | } |
189 | ||
190 | /* | |
191 | * Process updating of timeout sysctl | |
192 | */ | |
193 | int proc_dohung_task_timeout_secs(struct ctl_table *table, int write, | |
8d65af78 | 194 | void __user *buffer, |
e162b39a MSB |
195 | size_t *lenp, loff_t *ppos) |
196 | { | |
197 | int ret; | |
198 | ||
8d65af78 | 199 | ret = proc_doulongvec_minmax(table, write, buffer, lenp, ppos); |
e162b39a MSB |
200 | |
201 | if (ret || !write) | |
202 | goto out; | |
203 | ||
e162b39a MSB |
204 | wake_up_process(watchdog_task); |
205 | ||
206 | out: | |
207 | return ret; | |
208 | } | |
209 | ||
8b414521 MT |
210 | static atomic_t reset_hung_task = ATOMIC_INIT(0); |
211 | ||
212 | void reset_hung_task_detector(void) | |
213 | { | |
214 | atomic_set(&reset_hung_task, 1); | |
215 | } | |
216 | EXPORT_SYMBOL_GPL(reset_hung_task_detector); | |
217 | ||
e162b39a MSB |
218 | /* |
219 | * kthread which checks for tasks stuck in D state | |
220 | */ | |
221 | static int watchdog(void *dummy) | |
222 | { | |
223 | set_user_nice(current, 0); | |
e162b39a MSB |
224 | |
225 | for ( ; ; ) { | |
17406b82 | 226 | unsigned long timeout = sysctl_hung_task_timeout_secs; |
603a148f | 227 | |
17406b82 MSB |
228 | while (schedule_timeout_interruptible(timeout_jiffies(timeout))) |
229 | timeout = sysctl_hung_task_timeout_secs; | |
603a148f | 230 | |
8b414521 MT |
231 | if (atomic_xchg(&reset_hung_task, 0)) |
232 | continue; | |
233 | ||
17406b82 | 234 | check_hung_uninterruptible_tasks(timeout); |
e162b39a MSB |
235 | } |
236 | ||
237 | return 0; | |
238 | } | |
239 | ||
240 | static int __init hung_task_init(void) | |
241 | { | |
242 | atomic_notifier_chain_register(&panic_notifier_list, &panic_block); | |
243 | watchdog_task = kthread_run(watchdog, NULL, "khungtaskd"); | |
244 | ||
245 | return 0; | |
246 | } | |
247 | ||
248 | module_init(hung_task_init); |