]>
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> |
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 | */ | |
37 | #define HUNG_TASK_BATCHING 1024 | |
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 | |
a2e51445 DV |
44 | /* |
45 | * Zero (default value) means use sysctl_hung_task_timeout_secs: | |
46 | */ | |
47 | unsigned long __read_mostly sysctl_hung_task_check_interval_secs; | |
48 | ||
270750db | 49 | int __read_mostly sysctl_hung_task_warnings = 10; |
e162b39a MSB |
50 | |
51 | static int __read_mostly did_panic; | |
780cbcf2 | 52 | static bool hung_task_show_lock; |
401c636a | 53 | static bool hung_task_call_panic; |
e162b39a MSB |
54 | |
55 | static struct task_struct *watchdog_task; | |
56 | ||
57 | /* | |
58 | * Should we panic (and reboot, if panic_timeout= is set) when a | |
59 | * hung task is detected: | |
60 | */ | |
61 | unsigned int __read_mostly sysctl_hung_task_panic = | |
62 | CONFIG_BOOTPARAM_HUNG_TASK_PANIC_VALUE; | |
63 | ||
64 | static int __init hung_task_panic_setup(char *str) | |
65 | { | |
b51dbec6 | 66 | int rc = kstrtouint(str, 0, &sysctl_hung_task_panic); |
e162b39a | 67 | |
b51dbec6 FF |
68 | if (rc) |
69 | return rc; | |
e162b39a MSB |
70 | return 1; |
71 | } | |
72 | __setup("hung_task_panic=", hung_task_panic_setup); | |
73 | ||
74 | static int | |
75 | hung_task_panic(struct notifier_block *this, unsigned long event, void *ptr) | |
76 | { | |
77 | did_panic = 1; | |
78 | ||
79 | return NOTIFY_DONE; | |
80 | } | |
81 | ||
82 | static struct notifier_block panic_block = { | |
83 | .notifier_call = hung_task_panic, | |
84 | }; | |
85 | ||
17406b82 | 86 | static void check_hung_task(struct task_struct *t, unsigned long timeout) |
e162b39a MSB |
87 | { |
88 | unsigned long switch_count = t->nvcsw + t->nivcsw; | |
89 | ||
cf2592f5 FW |
90 | /* |
91 | * Ensure the task is not frozen. | |
f9fab10b | 92 | * Also, skip vfork and any other user process that freezer should skip. |
cf2592f5 | 93 | */ |
f9fab10b MSB |
94 | if (unlikely(t->flags & (PF_FROZEN | PF_FREEZER_SKIP))) |
95 | return; | |
96 | ||
97 | /* | |
98 | * When a freshly created task is scheduled once, changes its state to | |
99 | * TASK_UNINTERRUPTIBLE without having ever been switched out once, it | |
100 | * musn't be checked. | |
101 | */ | |
102 | if (unlikely(!switch_count)) | |
e162b39a MSB |
103 | return; |
104 | ||
17406b82 | 105 | if (switch_count != t->last_switch_count) { |
e162b39a | 106 | t->last_switch_count = switch_count; |
a2e51445 | 107 | t->last_switch_time = jiffies; |
e162b39a MSB |
108 | return; |
109 | } | |
a2e51445 DV |
110 | if (time_is_after_jiffies(t->last_switch_time + timeout * HZ)) |
111 | return; | |
6a716c90 ON |
112 | |
113 | trace_sched_process_hang(t); | |
114 | ||
48a6d64e | 115 | if (!sysctl_hung_task_warnings && !sysctl_hung_task_panic) |
e162b39a | 116 | return; |
270750db | 117 | |
e162b39a MSB |
118 | /* |
119 | * Ok, the task did not get scheduled for more than 2 minutes, | |
120 | * complain: | |
121 | */ | |
48a6d64e | 122 | if (sysctl_hung_task_warnings) { |
4ca5ede0 TH |
123 | if (sysctl_hung_task_warnings > 0) |
124 | sysctl_hung_task_warnings--; | |
48a6d64e JS |
125 | pr_err("INFO: task %s:%d blocked for more than %ld seconds.\n", |
126 | t->comm, t->pid, timeout); | |
127 | pr_err(" %s %s %.*s\n", | |
128 | print_tainted(), init_utsname()->release, | |
129 | (int)strcspn(init_utsname()->version, " "), | |
130 | init_utsname()->version); | |
131 | pr_err("\"echo 0 > /proc/sys/kernel/hung_task_timeout_secs\"" | |
132 | " disables this message.\n"); | |
133 | sched_show_task(t); | |
780cbcf2 | 134 | hung_task_show_lock = true; |
48a6d64e | 135 | } |
e162b39a | 136 | |
e162b39a MSB |
137 | touch_nmi_watchdog(); |
138 | ||
625056b6 | 139 | if (sysctl_hung_task_panic) { |
401c636a TH |
140 | hung_task_show_lock = true; |
141 | hung_task_call_panic = true; | |
625056b6 | 142 | } |
e162b39a MSB |
143 | } |
144 | ||
ce9dbe24 MSB |
145 | /* |
146 | * To avoid extending the RCU grace period for an unbounded amount of time, | |
147 | * periodically exit the critical section and enter a new one. | |
148 | * | |
149 | * For preemptible RCU it is sufficient to call rcu_read_unlock in order | |
6a103b0d | 150 | * to exit the grace period. For classic RCU, a reschedule is required. |
ce9dbe24 | 151 | */ |
6027ce49 | 152 | static bool rcu_lock_break(struct task_struct *g, struct task_struct *t) |
ce9dbe24 | 153 | { |
6027ce49 ON |
154 | bool can_cont; |
155 | ||
ce9dbe24 MSB |
156 | get_task_struct(g); |
157 | get_task_struct(t); | |
158 | rcu_read_unlock(); | |
159 | cond_resched(); | |
160 | rcu_read_lock(); | |
6027ce49 | 161 | can_cont = pid_alive(g) && pid_alive(t); |
ce9dbe24 MSB |
162 | put_task_struct(t); |
163 | put_task_struct(g); | |
6027ce49 ON |
164 | |
165 | return can_cont; | |
ce9dbe24 MSB |
166 | } |
167 | ||
e162b39a MSB |
168 | /* |
169 | * Check whether a TASK_UNINTERRUPTIBLE does not get woken up for | |
170 | * a really long time (120 seconds). If that happens, print out | |
171 | * a warning. | |
172 | */ | |
603a148f | 173 | static void check_hung_uninterruptible_tasks(unsigned long timeout) |
e162b39a MSB |
174 | { |
175 | int max_count = sysctl_hung_task_check_count; | |
ce9dbe24 | 176 | int batch_count = HUNG_TASK_BATCHING; |
e162b39a MSB |
177 | struct task_struct *g, *t; |
178 | ||
179 | /* | |
180 | * If the system crashed already then all bets are off, | |
181 | * do not report extra hung tasks: | |
182 | */ | |
183 | if (test_taint(TAINT_DIE) || did_panic) | |
184 | return; | |
185 | ||
780cbcf2 | 186 | hung_task_show_lock = false; |
94be52dc | 187 | rcu_read_lock(); |
972fae69 | 188 | for_each_process_thread(g, t) { |
e5af0226 | 189 | if (!max_count--) |
e162b39a | 190 | goto unlock; |
ce9dbe24 MSB |
191 | if (!--batch_count) { |
192 | batch_count = HUNG_TASK_BATCHING; | |
6027ce49 | 193 | if (!rcu_lock_break(g, t)) |
ce9dbe24 MSB |
194 | goto unlock; |
195 | } | |
e162b39a MSB |
196 | /* use "==" to skip the TASK_KILLABLE tasks waiting on NFS */ |
197 | if (t->state == TASK_UNINTERRUPTIBLE) | |
17406b82 | 198 | check_hung_task(t, timeout); |
972fae69 | 199 | } |
e162b39a | 200 | unlock: |
94be52dc | 201 | rcu_read_unlock(); |
780cbcf2 TH |
202 | if (hung_task_show_lock) |
203 | debug_show_all_locks(); | |
401c636a TH |
204 | if (hung_task_call_panic) { |
205 | trigger_all_cpu_backtrace(); | |
206 | panic("hung_task: blocked tasks"); | |
207 | } | |
e162b39a MSB |
208 | } |
209 | ||
b4aa14a6 TH |
210 | static long hung_timeout_jiffies(unsigned long last_checked, |
211 | unsigned long timeout) | |
e162b39a MSB |
212 | { |
213 | /* timeout of 0 will disable the watchdog */ | |
b4aa14a6 TH |
214 | return timeout ? last_checked - jiffies + timeout * HZ : |
215 | MAX_SCHEDULE_TIMEOUT; | |
e162b39a MSB |
216 | } |
217 | ||
218 | /* | |
219 | * Process updating of timeout sysctl | |
220 | */ | |
221 | int proc_dohung_task_timeout_secs(struct ctl_table *table, int write, | |
8d65af78 | 222 | void __user *buffer, |
e162b39a MSB |
223 | size_t *lenp, loff_t *ppos) |
224 | { | |
225 | int ret; | |
226 | ||
8d65af78 | 227 | ret = proc_doulongvec_minmax(table, write, buffer, lenp, ppos); |
e162b39a MSB |
228 | |
229 | if (ret || !write) | |
230 | goto out; | |
231 | ||
e162b39a MSB |
232 | wake_up_process(watchdog_task); |
233 | ||
234 | out: | |
235 | return ret; | |
236 | } | |
237 | ||
8b414521 MT |
238 | static atomic_t reset_hung_task = ATOMIC_INIT(0); |
239 | ||
240 | void reset_hung_task_detector(void) | |
241 | { | |
242 | atomic_set(&reset_hung_task, 1); | |
243 | } | |
244 | EXPORT_SYMBOL_GPL(reset_hung_task_detector); | |
245 | ||
a1c6ca3c VK |
246 | static bool hung_detector_suspended; |
247 | ||
248 | static int hungtask_pm_notify(struct notifier_block *self, | |
249 | unsigned long action, void *hcpu) | |
250 | { | |
251 | switch (action) { | |
252 | case PM_SUSPEND_PREPARE: | |
253 | case PM_HIBERNATION_PREPARE: | |
254 | case PM_RESTORE_PREPARE: | |
255 | hung_detector_suspended = true; | |
256 | break; | |
257 | case PM_POST_SUSPEND: | |
258 | case PM_POST_HIBERNATION: | |
259 | case PM_POST_RESTORE: | |
260 | hung_detector_suspended = false; | |
261 | break; | |
262 | default: | |
263 | break; | |
264 | } | |
265 | return NOTIFY_OK; | |
266 | } | |
267 | ||
e162b39a MSB |
268 | /* |
269 | * kthread which checks for tasks stuck in D state | |
270 | */ | |
271 | static int watchdog(void *dummy) | |
272 | { | |
b4aa14a6 TH |
273 | unsigned long hung_last_checked = jiffies; |
274 | ||
e162b39a | 275 | set_user_nice(current, 0); |
e162b39a MSB |
276 | |
277 | for ( ; ; ) { | |
17406b82 | 278 | unsigned long timeout = sysctl_hung_task_timeout_secs; |
a2e51445 DV |
279 | unsigned long interval = sysctl_hung_task_check_interval_secs; |
280 | long t; | |
603a148f | 281 | |
a2e51445 DV |
282 | if (interval == 0) |
283 | interval = timeout; | |
284 | interval = min_t(unsigned long, interval, timeout); | |
285 | t = hung_timeout_jiffies(hung_last_checked, interval); | |
b4aa14a6 | 286 | if (t <= 0) { |
a1c6ca3c VK |
287 | if (!atomic_xchg(&reset_hung_task, 0) && |
288 | !hung_detector_suspended) | |
b4aa14a6 TH |
289 | check_hung_uninterruptible_tasks(timeout); |
290 | hung_last_checked = jiffies; | |
8b414521 | 291 | continue; |
b4aa14a6 TH |
292 | } |
293 | schedule_timeout_interruptible(t); | |
e162b39a MSB |
294 | } |
295 | ||
296 | return 0; | |
297 | } | |
298 | ||
299 | static int __init hung_task_init(void) | |
300 | { | |
301 | atomic_notifier_chain_register(&panic_notifier_list, &panic_block); | |
a1c6ca3c VK |
302 | |
303 | /* Disable hung task detector on suspend */ | |
304 | pm_notifier(hungtask_pm_notify, 0); | |
305 | ||
e162b39a MSB |
306 | watchdog_task = kthread_run(watchdog, NULL, "khungtaskd"); |
307 | ||
308 | return 0; | |
309 | } | |
c96d6660 | 310 | subsys_initcall(hung_task_init); |