1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3 * Detect Hung Task
4 *
5 * kernel/hung_task.c - kernel thread for detecting tasks stuck in D state
6 *
7 */
8
9 #include <linux/mm.h>
10 #include <linux/cpu.h>
11 #include <linux/nmi.h>
12 #include <linux/init.h>
13 #include <linux/delay.h>
14 #include <linux/freezer.h>
15 #include <linux/kthread.h>
16 #include <linux/lockdep.h>
17 #include <linux/export.h>
18 #include <linux/panic_notifier.h>
19 #include <linux/sysctl.h>
20 #include <linux/suspend.h>
21 #include <linux/utsname.h>
22 #include <linux/sched/signal.h>
23 #include <linux/sched/debug.h>
24 #include <linux/sched/sysctl.h>
25
26 #include <trace/events/sched.h>
27
28 /*
29 * The number of tasks checked:
30 */
31 int __read_mostly sysctl_hung_task_check_count = PID_MAX_LIMIT;
32
33 /*
34 * Limit number of tasks checked in a batch.
35 *
36 * This value controls the preemptibility of khungtaskd since preemption
37 * is disabled during the critical section. It also controls the size of
38 * the RCU grace period. So it needs to be upper-bound.
39 */
40 #define HUNG_TASK_LOCK_BREAK (HZ / 10)
41
42 /*
43 * Zero means infinite timeout - no checking done:
44 */
45 unsigned long __read_mostly sysctl_hung_task_timeout_secs = CONFIG_DEFAULT_HUNG_TASK_TIMEOUT;
46
47 /*
48 * Zero (default value) means use sysctl_hung_task_timeout_secs:
49 */
50 unsigned long __read_mostly sysctl_hung_task_check_interval_secs;
51
52 int __read_mostly sysctl_hung_task_warnings = 10;
53
54 static int __read_mostly did_panic;
55 static bool hung_task_show_lock;
56 static bool hung_task_call_panic;
57 static bool hung_task_show_all_bt;
58
59 static struct task_struct *watchdog_task;
60
61 #ifdef CONFIG_SMP
62 /*
63 * Should we dump all CPUs backtraces in a hung task event?
64 * Defaults to 0, can be changed via sysctl.
65 */
66 static unsigned int __read_mostly sysctl_hung_task_all_cpu_backtrace;
67 #else
68 #define sysctl_hung_task_all_cpu_backtrace 0
69 #endif /* CONFIG_SMP */
70
71 /*
72 * Should we panic (and reboot, if panic_timeout= is set) when a
73 * hung task is detected:
74 */
75 unsigned int __read_mostly sysctl_hung_task_panic =
76 IS_ENABLED(CONFIG_BOOTPARAM_HUNG_TASK_PANIC);
77
78 static int
hung_task_panic(struct notifier_block * this,unsigned long event,void * ptr)79 hung_task_panic(struct notifier_block *this, unsigned long event, void *ptr)
80 {
81 did_panic = 1;
82
83 return NOTIFY_DONE;
84 }
85
86 static struct notifier_block panic_block = {
87 .notifier_call = hung_task_panic,
88 };
89
check_hung_task(struct task_struct * t,unsigned long timeout)90 static void check_hung_task(struct task_struct *t, unsigned long timeout)
91 {
92 unsigned long switch_count = t->nvcsw + t->nivcsw;
93
94 /*
95 * Ensure the task is not frozen.
96 * Also, skip vfork and any other user process that freezer should skip.
97 */
98 if (unlikely(READ_ONCE(t->__state) & TASK_FROZEN))
99 return;
100
101 /*
102 * When a freshly created task is scheduled once, changes its state to
103 * TASK_UNINTERRUPTIBLE without having ever been switched out once, it
104 * musn't be checked.
105 */
106 if (unlikely(!switch_count))
107 return;
108
109 if (switch_count != t->last_switch_count) {
110 t->last_switch_count = switch_count;
111 t->last_switch_time = jiffies;
112 return;
113 }
114 if (time_is_after_jiffies(t->last_switch_time + timeout * HZ))
115 return;
116
117 trace_sched_process_hang(t);
118
119 if (sysctl_hung_task_panic) {
120 console_verbose();
121 hung_task_show_lock = true;
122 hung_task_call_panic = true;
123 }
124
125 /*
126 * Ok, the task did not get scheduled for more than 2 minutes,
127 * complain:
128 */
129 if (sysctl_hung_task_warnings) {
130 if (sysctl_hung_task_warnings > 0)
131 sysctl_hung_task_warnings--;
132 pr_err("INFO: task %s:%d blocked for more than %ld seconds.\n",
133 t->comm, t->pid, (jiffies - t->last_switch_time) / HZ);
134 pr_err(" %s %s %.*s\n",
135 print_tainted(), init_utsname()->release,
136 (int)strcspn(init_utsname()->version, " "),
137 init_utsname()->version);
138 pr_err("\"echo 0 > /proc/sys/kernel/hung_task_timeout_secs\""
139 " disables this message.\n");
140 sched_show_task(t);
141 hung_task_show_lock = true;
142
143 if (sysctl_hung_task_all_cpu_backtrace)
144 hung_task_show_all_bt = true;
145 if (!sysctl_hung_task_warnings)
146 pr_info("Future hung task reports are suppressed, see sysctl kernel.hung_task_warnings\n");
147 }
148
149 touch_nmi_watchdog();
150 }
151
152 /*
153 * To avoid extending the RCU grace period for an unbounded amount of time,
154 * periodically exit the critical section and enter a new one.
155 *
156 * For preemptible RCU it is sufficient to call rcu_read_unlock in order
157 * to exit the grace period. For classic RCU, a reschedule is required.
158 */
rcu_lock_break(struct task_struct * g,struct task_struct * t)159 static bool rcu_lock_break(struct task_struct *g, struct task_struct *t)
160 {
161 bool can_cont;
162
163 get_task_struct(g);
164 get_task_struct(t);
165 rcu_read_unlock();
166 cond_resched();
167 rcu_read_lock();
168 can_cont = pid_alive(g) && pid_alive(t);
169 put_task_struct(t);
170 put_task_struct(g);
171
172 return can_cont;
173 }
174
175 /*
176 * Check whether a TASK_UNINTERRUPTIBLE does not get woken up for
177 * a really long time (120 seconds). If that happens, print out
178 * a warning.
179 */
check_hung_uninterruptible_tasks(unsigned long timeout)180 static void check_hung_uninterruptible_tasks(unsigned long timeout)
181 {
182 int max_count = sysctl_hung_task_check_count;
183 unsigned long last_break = jiffies;
184 struct task_struct *g, *t;
185
186 /*
187 * If the system crashed already then all bets are off,
188 * do not report extra hung tasks:
189 */
190 if (test_taint(TAINT_DIE) || did_panic)
191 return;
192
193 hung_task_show_lock = false;
194 rcu_read_lock();
195 for_each_process_thread(g, t) {
196 unsigned int state;
197
198 if (!max_count--)
199 goto unlock;
200 if (time_after(jiffies, last_break + HUNG_TASK_LOCK_BREAK)) {
201 if (!rcu_lock_break(g, t))
202 goto unlock;
203 last_break = jiffies;
204 }
205 /*
206 * skip the TASK_KILLABLE tasks -- these can be killed
207 * skip the TASK_IDLE tasks -- those are genuinely idle
208 */
209 state = READ_ONCE(t->__state);
210 if ((state & TASK_UNINTERRUPTIBLE) &&
211 !(state & TASK_WAKEKILL) &&
212 !(state & TASK_NOLOAD))
213 check_hung_task(t, timeout);
214 }
215 unlock:
216 rcu_read_unlock();
217 if (hung_task_show_lock)
218 debug_show_all_locks();
219
220 if (hung_task_show_all_bt) {
221 hung_task_show_all_bt = false;
222 trigger_all_cpu_backtrace();
223 }
224
225 if (hung_task_call_panic)
226 panic("hung_task: blocked tasks");
227 }
228
hung_timeout_jiffies(unsigned long last_checked,unsigned long timeout)229 static long hung_timeout_jiffies(unsigned long last_checked,
230 unsigned long timeout)
231 {
232 /* timeout of 0 will disable the watchdog */
233 return timeout ? last_checked - jiffies + timeout * HZ :
234 MAX_SCHEDULE_TIMEOUT;
235 }
236
237 #ifdef CONFIG_SYSCTL
238 /*
239 * Process updating of timeout sysctl
240 */
proc_dohung_task_timeout_secs(struct ctl_table * table,int write,void * buffer,size_t * lenp,loff_t * ppos)241 static int proc_dohung_task_timeout_secs(struct ctl_table *table, int write,
242 void *buffer,
243 size_t *lenp, loff_t *ppos)
244 {
245 int ret;
246
247 ret = proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
248
249 if (ret || !write)
250 goto out;
251
252 wake_up_process(watchdog_task);
253
254 out:
255 return ret;
256 }
257
258 /*
259 * This is needed for proc_doulongvec_minmax of sysctl_hung_task_timeout_secs
260 * and hung_task_check_interval_secs
261 */
262 static const unsigned long hung_task_timeout_max = (LONG_MAX / HZ);
263 static struct ctl_table hung_task_sysctls[] = {
264 #ifdef CONFIG_SMP
265 {
266 .procname = "hung_task_all_cpu_backtrace",
267 .data = &sysctl_hung_task_all_cpu_backtrace,
268 .maxlen = sizeof(int),
269 .mode = 0644,
270 .proc_handler = proc_dointvec_minmax,
271 .extra1 = SYSCTL_ZERO,
272 .extra2 = SYSCTL_ONE,
273 },
274 #endif /* CONFIG_SMP */
275 {
276 .procname = "hung_task_panic",
277 .data = &sysctl_hung_task_panic,
278 .maxlen = sizeof(int),
279 .mode = 0644,
280 .proc_handler = proc_dointvec_minmax,
281 .extra1 = SYSCTL_ZERO,
282 .extra2 = SYSCTL_ONE,
283 },
284 {
285 .procname = "hung_task_check_count",
286 .data = &sysctl_hung_task_check_count,
287 .maxlen = sizeof(int),
288 .mode = 0644,
289 .proc_handler = proc_dointvec_minmax,
290 .extra1 = SYSCTL_ZERO,
291 },
292 {
293 .procname = "hung_task_timeout_secs",
294 .data = &sysctl_hung_task_timeout_secs,
295 .maxlen = sizeof(unsigned long),
296 .mode = 0644,
297 .proc_handler = proc_dohung_task_timeout_secs,
298 .extra2 = (void *)&hung_task_timeout_max,
299 },
300 {
301 .procname = "hung_task_check_interval_secs",
302 .data = &sysctl_hung_task_check_interval_secs,
303 .maxlen = sizeof(unsigned long),
304 .mode = 0644,
305 .proc_handler = proc_dohung_task_timeout_secs,
306 .extra2 = (void *)&hung_task_timeout_max,
307 },
308 {
309 .procname = "hung_task_warnings",
310 .data = &sysctl_hung_task_warnings,
311 .maxlen = sizeof(int),
312 .mode = 0644,
313 .proc_handler = proc_dointvec_minmax,
314 .extra1 = SYSCTL_NEG_ONE,
315 },
316 {}
317 };
318
hung_task_sysctl_init(void)319 static void __init hung_task_sysctl_init(void)
320 {
321 register_sysctl_init("kernel", hung_task_sysctls);
322 }
323 #else
324 #define hung_task_sysctl_init() do { } while (0)
325 #endif /* CONFIG_SYSCTL */
326
327
328 static atomic_t reset_hung_task = ATOMIC_INIT(0);
329
reset_hung_task_detector(void)330 void reset_hung_task_detector(void)
331 {
332 atomic_set(&reset_hung_task, 1);
333 }
334 EXPORT_SYMBOL_GPL(reset_hung_task_detector);
335
336 static bool hung_detector_suspended;
337
hungtask_pm_notify(struct notifier_block * self,unsigned long action,void * hcpu)338 static int hungtask_pm_notify(struct notifier_block *self,
339 unsigned long action, void *hcpu)
340 {
341 switch (action) {
342 case PM_SUSPEND_PREPARE:
343 case PM_HIBERNATION_PREPARE:
344 case PM_RESTORE_PREPARE:
345 hung_detector_suspended = true;
346 break;
347 case PM_POST_SUSPEND:
348 case PM_POST_HIBERNATION:
349 case PM_POST_RESTORE:
350 hung_detector_suspended = false;
351 break;
352 default:
353 break;
354 }
355 return NOTIFY_OK;
356 }
357
358 /*
359 * kthread which checks for tasks stuck in D state
360 */
watchdog(void * dummy)361 static int watchdog(void *dummy)
362 {
363 unsigned long hung_last_checked = jiffies;
364
365 set_user_nice(current, 0);
366
367 for ( ; ; ) {
368 unsigned long timeout = sysctl_hung_task_timeout_secs;
369 unsigned long interval = sysctl_hung_task_check_interval_secs;
370 long t;
371
372 if (interval == 0)
373 interval = timeout;
374 interval = min_t(unsigned long, interval, timeout);
375 t = hung_timeout_jiffies(hung_last_checked, interval);
376 if (t <= 0) {
377 if (!atomic_xchg(&reset_hung_task, 0) &&
378 !hung_detector_suspended)
379 check_hung_uninterruptible_tasks(timeout);
380 hung_last_checked = jiffies;
381 continue;
382 }
383 schedule_timeout_interruptible(t);
384 }
385
386 return 0;
387 }
388
hung_task_init(void)389 static int __init hung_task_init(void)
390 {
391 atomic_notifier_chain_register(&panic_notifier_list, &panic_block);
392
393 /* Disable hung task detector on suspend */
394 pm_notifier(hungtask_pm_notify, 0);
395
396 watchdog_task = kthread_run(watchdog, NULL, "khungtaskd");
397 hung_task_sysctl_init();
398
399 return 0;
400 }
401 subsys_initcall(hung_task_init);
402