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> |
a1c6ca3c | 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> |
a98eb6f1 | 22 | #include <linux/sched/sysctl.h> |
3f07c014 | 23 | |
6a716c90 | 24 | #include <trace/events/sched.h> |
e162b39a MSB |
25 | |
26 | /* | |
ce9dbe24 | 27 | * The number of tasks checked: |
e162b39a | 28 | */ |
cd64647f | 29 | int __read_mostly sysctl_hung_task_check_count = PID_MAX_LIMIT; |
ce9dbe24 MSB |
30 | |
31 | /* | |
32 | * Limit number of tasks checked in a batch. | |
33 | * | |
34 | * This value controls the preemptibility of khungtaskd since preemption | |
35 | * is disabled during the critical section. It also controls the size of | |
36 | * the RCU grace period. So it needs to be upper-bound. | |
37 | */ | |
304ae427 | 38 | #define HUNG_TASK_LOCK_BREAK (HZ / 10) |
e162b39a MSB |
39 | |
40 | /* | |
41 | * Zero means infinite timeout - no checking done: | |
42 | */ | |
e11feaa1 | 43 | unsigned long __read_mostly sysctl_hung_task_timeout_secs = CONFIG_DEFAULT_HUNG_TASK_TIMEOUT; |
e162b39a | 44 | |
a2e51445 DV |
45 | /* |
46 | * Zero (default value) means use sysctl_hung_task_timeout_secs: | |
47 | */ | |
48 | unsigned long __read_mostly sysctl_hung_task_check_interval_secs; | |
49 | ||
270750db | 50 | int __read_mostly sysctl_hung_task_warnings = 10; |
e162b39a MSB |
51 | |
52 | static int __read_mostly did_panic; | |
780cbcf2 | 53 | static bool hung_task_show_lock; |
401c636a | 54 | static bool hung_task_call_panic; |
e162b39a MSB |
55 | |
56 | static struct task_struct *watchdog_task; | |
57 | ||
58 | /* | |
59 | * Should we panic (and reboot, if panic_timeout= is set) when a | |
60 | * hung task is detected: | |
61 | */ | |
62 | unsigned int __read_mostly sysctl_hung_task_panic = | |
63 | CONFIG_BOOTPARAM_HUNG_TASK_PANIC_VALUE; | |
64 | ||
65 | static int __init hung_task_panic_setup(char *str) | |
66 | { | |
b51dbec6 | 67 | int rc = kstrtouint(str, 0, &sysctl_hung_task_panic); |
e162b39a | 68 | |
b51dbec6 FF |
69 | if (rc) |
70 | return rc; | |
e162b39a MSB |
71 | return 1; |
72 | } | |
73 | __setup("hung_task_panic=", hung_task_panic_setup); | |
74 | ||
75 | static int | |
76 | hung_task_panic(struct notifier_block *this, unsigned long event, void *ptr) | |
77 | { | |
78 | did_panic = 1; | |
79 | ||
80 | return NOTIFY_DONE; | |
81 | } | |
82 | ||
83 | static struct notifier_block panic_block = { | |
84 | .notifier_call = hung_task_panic, | |
85 | }; | |
86 | ||
17406b82 | 87 | static void check_hung_task(struct task_struct *t, unsigned long timeout) |
e162b39a MSB |
88 | { |
89 | unsigned long switch_count = t->nvcsw + t->nivcsw; | |
90 | ||
cf2592f5 FW |
91 | /* |
92 | * Ensure the task is not frozen. | |
f9fab10b | 93 | * Also, skip vfork and any other user process that freezer should skip. |
cf2592f5 | 94 | */ |
f9fab10b MSB |
95 | if (unlikely(t->flags & (PF_FROZEN | PF_FREEZER_SKIP))) |
96 | return; | |
97 | ||
98 | /* | |
99 | * When a freshly created task is scheduled once, changes its state to | |
100 | * TASK_UNINTERRUPTIBLE without having ever been switched out once, it | |
101 | * musn't be checked. | |
102 | */ | |
103 | if (unlikely(!switch_count)) | |
e162b39a MSB |
104 | return; |
105 | ||
17406b82 | 106 | if (switch_count != t->last_switch_count) { |
e162b39a | 107 | t->last_switch_count = switch_count; |
a2e51445 | 108 | t->last_switch_time = jiffies; |
e162b39a MSB |
109 | return; |
110 | } | |
a2e51445 DV |
111 | if (time_is_after_jiffies(t->last_switch_time + timeout * HZ)) |
112 | return; | |
6a716c90 ON |
113 | |
114 | trace_sched_process_hang(t); | |
115 | ||
168e06f7 LC |
116 | if (sysctl_hung_task_panic) { |
117 | console_verbose(); | |
118 | hung_task_show_lock = true; | |
119 | hung_task_call_panic = true; | |
120 | } | |
270750db | 121 | |
e162b39a MSB |
122 | /* |
123 | * Ok, the task did not get scheduled for more than 2 minutes, | |
124 | * complain: | |
125 | */ | |
48a6d64e | 126 | if (sysctl_hung_task_warnings) { |
4ca5ede0 TH |
127 | if (sysctl_hung_task_warnings > 0) |
128 | sysctl_hung_task_warnings--; | |
48a6d64e | 129 | pr_err("INFO: task %s:%d blocked for more than %ld seconds.\n", |
b014beba | 130 | t->comm, t->pid, (jiffies - t->last_switch_time) / HZ); |
48a6d64e JS |
131 | pr_err(" %s %s %.*s\n", |
132 | print_tainted(), init_utsname()->release, | |
133 | (int)strcspn(init_utsname()->version, " "), | |
134 | init_utsname()->version); | |
135 | pr_err("\"echo 0 > /proc/sys/kernel/hung_task_timeout_secs\"" | |
136 | " disables this message.\n"); | |
137 | sched_show_task(t); | |
780cbcf2 | 138 | hung_task_show_lock = true; |
48a6d64e | 139 | } |
e162b39a | 140 | |
e162b39a | 141 | touch_nmi_watchdog(); |
e162b39a MSB |
142 | } |
143 | ||
ce9dbe24 MSB |
144 | /* |
145 | * To avoid extending the RCU grace period for an unbounded amount of time, | |
146 | * periodically exit the critical section and enter a new one. | |
147 | * | |
148 | * For preemptible RCU it is sufficient to call rcu_read_unlock in order | |
6a103b0d | 149 | * to exit the grace period. For classic RCU, a reschedule is required. |
ce9dbe24 | 150 | */ |
6027ce49 | 151 | static bool rcu_lock_break(struct task_struct *g, struct task_struct *t) |
ce9dbe24 | 152 | { |
6027ce49 ON |
153 | bool can_cont; |
154 | ||
ce9dbe24 MSB |
155 | get_task_struct(g); |
156 | get_task_struct(t); | |
157 | rcu_read_unlock(); | |
158 | cond_resched(); | |
159 | rcu_read_lock(); | |
6027ce49 | 160 | can_cont = pid_alive(g) && pid_alive(t); |
ce9dbe24 MSB |
161 | put_task_struct(t); |
162 | put_task_struct(g); | |
6027ce49 ON |
163 | |
164 | return can_cont; | |
ce9dbe24 MSB |
165 | } |
166 | ||
e162b39a MSB |
167 | /* |
168 | * Check whether a TASK_UNINTERRUPTIBLE does not get woken up for | |
169 | * a really long time (120 seconds). If that happens, print out | |
170 | * a warning. | |
171 | */ | |
603a148f | 172 | static void check_hung_uninterruptible_tasks(unsigned long timeout) |
e162b39a MSB |
173 | { |
174 | int max_count = sysctl_hung_task_check_count; | |
304ae427 | 175 | unsigned long last_break = jiffies; |
e162b39a MSB |
176 | struct task_struct *g, *t; |
177 | ||
178 | /* | |
179 | * If the system crashed already then all bets are off, | |
180 | * do not report extra hung tasks: | |
181 | */ | |
182 | if (test_taint(TAINT_DIE) || did_panic) | |
183 | return; | |
184 | ||
780cbcf2 | 185 | hung_task_show_lock = false; |
94be52dc | 186 | rcu_read_lock(); |
972fae69 | 187 | for_each_process_thread(g, t) { |
e5af0226 | 188 | if (!max_count--) |
e162b39a | 189 | goto unlock; |
304ae427 | 190 | if (time_after(jiffies, last_break + HUNG_TASK_LOCK_BREAK)) { |
6027ce49 | 191 | if (!rcu_lock_break(g, t)) |
ce9dbe24 | 192 | goto unlock; |
304ae427 | 193 | last_break = jiffies; |
ce9dbe24 | 194 | } |
e162b39a MSB |
195 | /* use "==" to skip the TASK_KILLABLE tasks waiting on NFS */ |
196 | if (t->state == TASK_UNINTERRUPTIBLE) | |
17406b82 | 197 | check_hung_task(t, timeout); |
972fae69 | 198 | } |
e162b39a | 199 | unlock: |
94be52dc | 200 | rcu_read_unlock(); |
780cbcf2 TH |
201 | if (hung_task_show_lock) |
202 | debug_show_all_locks(); | |
401c636a TH |
203 | if (hung_task_call_panic) { |
204 | trigger_all_cpu_backtrace(); | |
205 | panic("hung_task: blocked tasks"); | |
206 | } | |
e162b39a MSB |
207 | } |
208 | ||
b4aa14a6 TH |
209 | static long hung_timeout_jiffies(unsigned long last_checked, |
210 | unsigned long timeout) | |
e162b39a MSB |
211 | { |
212 | /* timeout of 0 will disable the watchdog */ | |
b4aa14a6 TH |
213 | return timeout ? last_checked - jiffies + timeout * HZ : |
214 | MAX_SCHEDULE_TIMEOUT; | |
e162b39a MSB |
215 | } |
216 | ||
217 | /* | |
218 | * Process updating of timeout sysctl | |
219 | */ | |
220 | int proc_dohung_task_timeout_secs(struct ctl_table *table, int write, | |
8d65af78 | 221 | void __user *buffer, |
e162b39a MSB |
222 | size_t *lenp, loff_t *ppos) |
223 | { | |
224 | int ret; | |
225 | ||
8d65af78 | 226 | ret = proc_doulongvec_minmax(table, write, buffer, lenp, ppos); |
e162b39a MSB |
227 | |
228 | if (ret || !write) | |
229 | goto out; | |
230 | ||
e162b39a MSB |
231 | wake_up_process(watchdog_task); |
232 | ||
233 | out: | |
234 | return ret; | |
235 | } | |
236 | ||
8b414521 MT |
237 | static atomic_t reset_hung_task = ATOMIC_INIT(0); |
238 | ||
239 | void reset_hung_task_detector(void) | |
240 | { | |
241 | atomic_set(&reset_hung_task, 1); | |
242 | } | |
243 | EXPORT_SYMBOL_GPL(reset_hung_task_detector); | |
244 | ||
a1c6ca3c VK |
245 | static bool hung_detector_suspended; |
246 | ||
247 | static int hungtask_pm_notify(struct notifier_block *self, | |
248 | unsigned long action, void *hcpu) | |
249 | { | |
250 | switch (action) { | |
251 | case PM_SUSPEND_PREPARE: | |
252 | case PM_HIBERNATION_PREPARE: | |
253 | case PM_RESTORE_PREPARE: | |
254 | hung_detector_suspended = true; | |
255 | break; | |
256 | case PM_POST_SUSPEND: | |
257 | case PM_POST_HIBERNATION: | |
258 | case PM_POST_RESTORE: | |
259 | hung_detector_suspended = false; | |
260 | break; | |
261 | default: | |
262 | break; | |
263 | } | |
264 | return NOTIFY_OK; | |
265 | } | |
266 | ||
e162b39a MSB |
267 | /* |
268 | * kthread which checks for tasks stuck in D state | |
269 | */ | |
270 | static int watchdog(void *dummy) | |
271 | { | |
b4aa14a6 TH |
272 | unsigned long hung_last_checked = jiffies; |
273 | ||
e162b39a | 274 | set_user_nice(current, 0); |
e162b39a MSB |
275 | |
276 | for ( ; ; ) { | |
17406b82 | 277 | unsigned long timeout = sysctl_hung_task_timeout_secs; |
a2e51445 DV |
278 | unsigned long interval = sysctl_hung_task_check_interval_secs; |
279 | long t; | |
603a148f | 280 | |
a2e51445 DV |
281 | if (interval == 0) |
282 | interval = timeout; | |
283 | interval = min_t(unsigned long, interval, timeout); | |
284 | t = hung_timeout_jiffies(hung_last_checked, interval); | |
b4aa14a6 | 285 | if (t <= 0) { |
a1c6ca3c VK |
286 | if (!atomic_xchg(&reset_hung_task, 0) && |
287 | !hung_detector_suspended) | |
b4aa14a6 TH |
288 | check_hung_uninterruptible_tasks(timeout); |
289 | hung_last_checked = jiffies; | |
8b414521 | 290 | continue; |
b4aa14a6 TH |
291 | } |
292 | schedule_timeout_interruptible(t); | |
e162b39a MSB |
293 | } |
294 | ||
295 | return 0; | |
296 | } | |
297 | ||
298 | static int __init hung_task_init(void) | |
299 | { | |
300 | atomic_notifier_chain_register(&panic_notifier_list, &panic_block); | |
a1c6ca3c VK |
301 | |
302 | /* Disable hung task detector on suspend */ | |
303 | pm_notifier(hungtask_pm_notify, 0); | |
304 | ||
e162b39a MSB |
305 | watchdog_task = kthread_run(watchdog, NULL, "khungtaskd"); |
306 | ||
307 | return 0; | |
308 | } | |
c96d6660 | 309 | subsys_initcall(hung_task_init); |