]>
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> |
d283cf4f | 18 | #include <linux/suspend.h> |
41e85ce8 | 19 | #include <linux/utsname.h> |
3f07c014 | 20 | #include <linux/sched/signal.h> |
b17b0153 | 21 | #include <linux/sched/debug.h> |
3f07c014 | 22 | |
6a716c90 | 23 | #include <trace/events/sched.h> |
e162b39a MSB |
24 | |
25 | /* | |
ce9dbe24 | 26 | * The number of tasks checked: |
e162b39a | 27 | */ |
cd64647f | 28 | int __read_mostly sysctl_hung_task_check_count = PID_MAX_LIMIT; |
ce9dbe24 MSB |
29 | |
30 | /* | |
31 | * Limit number of tasks checked in a batch. | |
32 | * | |
33 | * This value controls the preemptibility of khungtaskd since preemption | |
34 | * is disabled during the critical section. It also controls the size of | |
35 | * the RCU grace period. So it needs to be upper-bound. | |
36 | */ | |
9434bc1d | 37 | #define HUNG_TASK_LOCK_BREAK (HZ / 10) |
e162b39a MSB |
38 | |
39 | /* | |
40 | * Zero means infinite timeout - no checking done: | |
41 | */ | |
e11feaa1 | 42 | unsigned long __read_mostly sysctl_hung_task_timeout_secs = CONFIG_DEFAULT_HUNG_TASK_TIMEOUT; |
e162b39a | 43 | |
270750db | 44 | int __read_mostly sysctl_hung_task_warnings = 10; |
e162b39a MSB |
45 | |
46 | static int __read_mostly did_panic; | |
780cbcf2 | 47 | static bool hung_task_show_lock; |
2bdd4984 | 48 | static bool hung_task_call_panic; |
e162b39a MSB |
49 | |
50 | static struct task_struct *watchdog_task; | |
51 | ||
52 | /* | |
53 | * Should we panic (and reboot, if panic_timeout= is set) when a | |
54 | * hung task is detected: | |
55 | */ | |
56 | unsigned int __read_mostly sysctl_hung_task_panic = | |
57 | CONFIG_BOOTPARAM_HUNG_TASK_PANIC_VALUE; | |
58 | ||
59 | static int __init hung_task_panic_setup(char *str) | |
60 | { | |
b51dbec6 | 61 | int rc = kstrtouint(str, 0, &sysctl_hung_task_panic); |
e162b39a | 62 | |
b51dbec6 FF |
63 | if (rc) |
64 | return rc; | |
e162b39a MSB |
65 | return 1; |
66 | } | |
67 | __setup("hung_task_panic=", hung_task_panic_setup); | |
68 | ||
69 | static int | |
70 | hung_task_panic(struct notifier_block *this, unsigned long event, void *ptr) | |
71 | { | |
72 | did_panic = 1; | |
73 | ||
74 | return NOTIFY_DONE; | |
75 | } | |
76 | ||
77 | static struct notifier_block panic_block = { | |
78 | .notifier_call = hung_task_panic, | |
79 | }; | |
80 | ||
17406b82 | 81 | static void check_hung_task(struct task_struct *t, unsigned long timeout) |
e162b39a MSB |
82 | { |
83 | unsigned long switch_count = t->nvcsw + t->nivcsw; | |
84 | ||
cf2592f5 FW |
85 | /* |
86 | * Ensure the task is not frozen. | |
f9fab10b | 87 | * Also, skip vfork and any other user process that freezer should skip. |
cf2592f5 | 88 | */ |
f9fab10b MSB |
89 | if (unlikely(t->flags & (PF_FROZEN | PF_FREEZER_SKIP))) |
90 | return; | |
91 | ||
92 | /* | |
93 | * When a freshly created task is scheduled once, changes its state to | |
94 | * TASK_UNINTERRUPTIBLE without having ever been switched out once, it | |
95 | * musn't be checked. | |
96 | */ | |
97 | if (unlikely(!switch_count)) | |
e162b39a MSB |
98 | return; |
99 | ||
17406b82 | 100 | if (switch_count != t->last_switch_count) { |
e162b39a | 101 | t->last_switch_count = switch_count; |
e162b39a MSB |
102 | return; |
103 | } | |
6a716c90 ON |
104 | |
105 | trace_sched_process_hang(t); | |
106 | ||
aad23e89 LC |
107 | if (sysctl_hung_task_panic) { |
108 | console_verbose(); | |
109 | hung_task_show_lock = true; | |
110 | hung_task_call_panic = true; | |
111 | } | |
270750db | 112 | |
e162b39a MSB |
113 | /* |
114 | * Ok, the task did not get scheduled for more than 2 minutes, | |
115 | * complain: | |
116 | */ | |
48a6d64e | 117 | if (sysctl_hung_task_warnings) { |
4ca5ede0 TH |
118 | if (sysctl_hung_task_warnings > 0) |
119 | sysctl_hung_task_warnings--; | |
48a6d64e JS |
120 | pr_err("INFO: task %s:%d blocked for more than %ld seconds.\n", |
121 | t->comm, t->pid, timeout); | |
122 | pr_err(" %s %s %.*s\n", | |
123 | print_tainted(), init_utsname()->release, | |
124 | (int)strcspn(init_utsname()->version, " "), | |
125 | init_utsname()->version); | |
126 | pr_err("\"echo 0 > /proc/sys/kernel/hung_task_timeout_secs\"" | |
127 | " disables this message.\n"); | |
128 | sched_show_task(t); | |
780cbcf2 | 129 | hung_task_show_lock = true; |
48a6d64e | 130 | } |
e162b39a | 131 | |
e162b39a | 132 | touch_nmi_watchdog(); |
e162b39a MSB |
133 | } |
134 | ||
ce9dbe24 MSB |
135 | /* |
136 | * To avoid extending the RCU grace period for an unbounded amount of time, | |
137 | * periodically exit the critical section and enter a new one. | |
138 | * | |
139 | * For preemptible RCU it is sufficient to call rcu_read_unlock in order | |
6a103b0d | 140 | * to exit the grace period. For classic RCU, a reschedule is required. |
ce9dbe24 | 141 | */ |
6027ce49 | 142 | static bool rcu_lock_break(struct task_struct *g, struct task_struct *t) |
ce9dbe24 | 143 | { |
6027ce49 ON |
144 | bool can_cont; |
145 | ||
ce9dbe24 MSB |
146 | get_task_struct(g); |
147 | get_task_struct(t); | |
148 | rcu_read_unlock(); | |
149 | cond_resched(); | |
150 | rcu_read_lock(); | |
6027ce49 | 151 | can_cont = pid_alive(g) && pid_alive(t); |
ce9dbe24 MSB |
152 | put_task_struct(t); |
153 | put_task_struct(g); | |
6027ce49 ON |
154 | |
155 | return can_cont; | |
ce9dbe24 MSB |
156 | } |
157 | ||
e162b39a MSB |
158 | /* |
159 | * Check whether a TASK_UNINTERRUPTIBLE does not get woken up for | |
160 | * a really long time (120 seconds). If that happens, print out | |
161 | * a warning. | |
162 | */ | |
603a148f | 163 | static void check_hung_uninterruptible_tasks(unsigned long timeout) |
e162b39a MSB |
164 | { |
165 | int max_count = sysctl_hung_task_check_count; | |
9434bc1d | 166 | unsigned long last_break = jiffies; |
e162b39a MSB |
167 | struct task_struct *g, *t; |
168 | ||
169 | /* | |
170 | * If the system crashed already then all bets are off, | |
171 | * do not report extra hung tasks: | |
172 | */ | |
173 | if (test_taint(TAINT_DIE) || did_panic) | |
174 | return; | |
175 | ||
780cbcf2 | 176 | hung_task_show_lock = false; |
94be52dc | 177 | rcu_read_lock(); |
972fae69 | 178 | for_each_process_thread(g, t) { |
e5af0226 | 179 | if (!max_count--) |
e162b39a | 180 | goto unlock; |
9434bc1d | 181 | if (time_after(jiffies, last_break + HUNG_TASK_LOCK_BREAK)) { |
6027ce49 | 182 | if (!rcu_lock_break(g, t)) |
ce9dbe24 | 183 | goto unlock; |
9434bc1d | 184 | last_break = jiffies; |
ce9dbe24 | 185 | } |
e162b39a MSB |
186 | /* use "==" to skip the TASK_KILLABLE tasks waiting on NFS */ |
187 | if (t->state == TASK_UNINTERRUPTIBLE) | |
17406b82 | 188 | check_hung_task(t, timeout); |
972fae69 | 189 | } |
e162b39a | 190 | unlock: |
94be52dc | 191 | rcu_read_unlock(); |
780cbcf2 TH |
192 | if (hung_task_show_lock) |
193 | debug_show_all_locks(); | |
2bdd4984 TH |
194 | if (hung_task_call_panic) { |
195 | trigger_all_cpu_backtrace(); | |
196 | panic("hung_task: blocked tasks"); | |
197 | } | |
e162b39a MSB |
198 | } |
199 | ||
b4aa14a6 TH |
200 | static long hung_timeout_jiffies(unsigned long last_checked, |
201 | unsigned long timeout) | |
e162b39a MSB |
202 | { |
203 | /* timeout of 0 will disable the watchdog */ | |
b4aa14a6 TH |
204 | return timeout ? last_checked - jiffies + timeout * HZ : |
205 | MAX_SCHEDULE_TIMEOUT; | |
e162b39a MSB |
206 | } |
207 | ||
208 | /* | |
209 | * Process updating of timeout sysctl | |
210 | */ | |
211 | int proc_dohung_task_timeout_secs(struct ctl_table *table, int write, | |
8d65af78 | 212 | void __user *buffer, |
e162b39a MSB |
213 | size_t *lenp, loff_t *ppos) |
214 | { | |
215 | int ret; | |
216 | ||
8d65af78 | 217 | ret = proc_doulongvec_minmax(table, write, buffer, lenp, ppos); |
e162b39a MSB |
218 | |
219 | if (ret || !write) | |
220 | goto out; | |
221 | ||
e162b39a MSB |
222 | wake_up_process(watchdog_task); |
223 | ||
224 | out: | |
225 | return ret; | |
226 | } | |
227 | ||
8b414521 MT |
228 | static atomic_t reset_hung_task = ATOMIC_INIT(0); |
229 | ||
230 | void reset_hung_task_detector(void) | |
231 | { | |
232 | atomic_set(&reset_hung_task, 1); | |
233 | } | |
234 | EXPORT_SYMBOL_GPL(reset_hung_task_detector); | |
235 | ||
d283cf4f VK |
236 | static bool hung_detector_suspended; |
237 | ||
238 | static int hungtask_pm_notify(struct notifier_block *self, | |
239 | unsigned long action, void *hcpu) | |
240 | { | |
241 | switch (action) { | |
242 | case PM_SUSPEND_PREPARE: | |
243 | case PM_HIBERNATION_PREPARE: | |
244 | case PM_RESTORE_PREPARE: | |
245 | hung_detector_suspended = true; | |
246 | break; | |
247 | case PM_POST_SUSPEND: | |
248 | case PM_POST_HIBERNATION: | |
249 | case PM_POST_RESTORE: | |
250 | hung_detector_suspended = false; | |
251 | break; | |
252 | default: | |
253 | break; | |
254 | } | |
255 | return NOTIFY_OK; | |
256 | } | |
257 | ||
e162b39a MSB |
258 | /* |
259 | * kthread which checks for tasks stuck in D state | |
260 | */ | |
261 | static int watchdog(void *dummy) | |
262 | { | |
b4aa14a6 TH |
263 | unsigned long hung_last_checked = jiffies; |
264 | ||
e162b39a | 265 | set_user_nice(current, 0); |
e162b39a MSB |
266 | |
267 | for ( ; ; ) { | |
17406b82 | 268 | unsigned long timeout = sysctl_hung_task_timeout_secs; |
b4aa14a6 | 269 | long t = hung_timeout_jiffies(hung_last_checked, timeout); |
603a148f | 270 | |
b4aa14a6 | 271 | if (t <= 0) { |
d283cf4f VK |
272 | if (!atomic_xchg(&reset_hung_task, 0) && |
273 | !hung_detector_suspended) | |
b4aa14a6 TH |
274 | check_hung_uninterruptible_tasks(timeout); |
275 | hung_last_checked = jiffies; | |
8b414521 | 276 | continue; |
b4aa14a6 TH |
277 | } |
278 | schedule_timeout_interruptible(t); | |
e162b39a MSB |
279 | } |
280 | ||
281 | return 0; | |
282 | } | |
283 | ||
284 | static int __init hung_task_init(void) | |
285 | { | |
286 | atomic_notifier_chain_register(&panic_notifier_list, &panic_block); | |
d283cf4f VK |
287 | |
288 | /* Disable hung task detector on suspend */ | |
289 | pm_notifier(hungtask_pm_notify, 0); | |
290 | ||
e162b39a MSB |
291 | watchdog_task = kthread_run(watchdog, NULL, "khungtaskd"); |
292 | ||
293 | return 0; | |
294 | } | |
c96d6660 | 295 | subsys_initcall(hung_task_init); |