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 #include <linux/hung_task.h>
26 #include <linux/rwsem.h>
27 #include <linux/sys_info.h>
28
29 #include <trace/events/sched.h>
30
31 /*
32 * The number of tasks checked:
33 */
34 static int __read_mostly sysctl_hung_task_check_count = PID_MAX_LIMIT;
35
36 /*
37 * Total number of tasks detected as hung since boot:
38 */
39 static unsigned long __read_mostly sysctl_hung_task_detect_count;
40
41 /*
42 * Limit number of tasks checked in a batch.
43 *
44 * This value controls the preemptibility of khungtaskd since preemption
45 * is disabled during the critical section. It also controls the size of
46 * the RCU grace period. So it needs to be upper-bound.
47 */
48 #define HUNG_TASK_LOCK_BREAK (HZ / 10)
49
50 /*
51 * Zero means infinite timeout - no checking done:
52 */
53 unsigned long __read_mostly sysctl_hung_task_timeout_secs = CONFIG_DEFAULT_HUNG_TASK_TIMEOUT;
54
55 /*
56 * Zero (default value) means use sysctl_hung_task_timeout_secs:
57 */
58 static unsigned long __read_mostly sysctl_hung_task_check_interval_secs;
59
60 static int __read_mostly sysctl_hung_task_warnings = 10;
61
62 static int __read_mostly did_panic;
63 static bool hung_task_call_panic;
64
65 static struct task_struct *watchdog_task;
66
67 /*
68 * A bitmask to control what kinds of system info to be printed when
69 * a hung task is detected, it could be task, memory, lock etc. Refer
70 * include/linux/sys_info.h for detailed bit definition.
71 */
72 static unsigned long hung_task_si_mask;
73
74 #ifdef CONFIG_SMP
75 /*
76 * Should we dump all CPUs backtraces in a hung task event?
77 * Defaults to 0, can be changed via sysctl.
78 */
79 static unsigned int __read_mostly sysctl_hung_task_all_cpu_backtrace;
80 #else
81 #define sysctl_hung_task_all_cpu_backtrace 0
82 #endif /* CONFIG_SMP */
83
84 /*
85 * Should we panic (and reboot, if panic_timeout= is set) when a
86 * hung task is detected:
87 */
88 static unsigned int __read_mostly sysctl_hung_task_panic =
89 CONFIG_BOOTPARAM_HUNG_TASK_PANIC;
90
91 static int
hung_task_panic(struct notifier_block * this,unsigned long event,void * ptr)92 hung_task_panic(struct notifier_block *this, unsigned long event, void *ptr)
93 {
94 did_panic = 1;
95
96 return NOTIFY_DONE;
97 }
98
99 static struct notifier_block panic_block = {
100 .notifier_call = hung_task_panic,
101 };
102
task_is_hung(struct task_struct * t,unsigned long timeout)103 static bool task_is_hung(struct task_struct *t, unsigned long timeout)
104 {
105 unsigned long switch_count = t->nvcsw + t->nivcsw;
106 unsigned int state = READ_ONCE(t->__state);
107
108 /*
109 * skip the TASK_KILLABLE tasks -- these can be killed
110 * skip the TASK_IDLE tasks -- those are genuinely idle
111 * skip the TASK_FROZEN task -- it reasonably stops scheduling by freezer
112 */
113 if (!(state & TASK_UNINTERRUPTIBLE) ||
114 (state & (TASK_WAKEKILL | TASK_NOLOAD | TASK_FROZEN)))
115 return false;
116
117 /*
118 * When a freshly created task is scheduled once, changes its state to
119 * TASK_UNINTERRUPTIBLE without having ever been switched out once, it
120 * musn't be checked.
121 */
122 if (unlikely(!switch_count))
123 return false;
124
125 if (switch_count != t->last_switch_count) {
126 t->last_switch_count = switch_count;
127 t->last_switch_time = jiffies;
128 return false;
129 }
130 if (time_is_after_jiffies(t->last_switch_time + timeout * HZ))
131 return false;
132
133 return true;
134 }
135
136 #ifdef CONFIG_DETECT_HUNG_TASK_BLOCKER
debug_show_blocker(struct task_struct * task,unsigned long timeout)137 static void debug_show_blocker(struct task_struct *task, unsigned long timeout)
138 {
139 struct task_struct *g, *t;
140 unsigned long owner, blocker, blocker_type;
141 const char *rwsem_blocked_by, *rwsem_blocked_as;
142
143 RCU_LOCKDEP_WARN(!rcu_read_lock_held(), "No rcu lock held");
144
145 blocker = READ_ONCE(task->blocker);
146 if (!blocker)
147 return;
148
149 blocker_type = hung_task_get_blocker_type(blocker);
150
151 switch (blocker_type) {
152 case BLOCKER_TYPE_MUTEX:
153 owner = mutex_get_owner(hung_task_blocker_to_lock(blocker));
154 break;
155 case BLOCKER_TYPE_SEM:
156 owner = sem_last_holder(hung_task_blocker_to_lock(blocker));
157 break;
158 case BLOCKER_TYPE_RWSEM_READER:
159 case BLOCKER_TYPE_RWSEM_WRITER:
160 owner = (unsigned long)rwsem_owner(
161 hung_task_blocker_to_lock(blocker));
162 rwsem_blocked_as = (blocker_type == BLOCKER_TYPE_RWSEM_READER) ?
163 "reader" : "writer";
164 rwsem_blocked_by = is_rwsem_reader_owned(
165 hung_task_blocker_to_lock(blocker)) ?
166 "reader" : "writer";
167 break;
168 default:
169 WARN_ON_ONCE(1);
170 return;
171 }
172
173
174 if (unlikely(!owner)) {
175 switch (blocker_type) {
176 case BLOCKER_TYPE_MUTEX:
177 pr_err("INFO: task %s:%d is blocked on a mutex, but the owner is not found.\n",
178 task->comm, task->pid);
179 break;
180 case BLOCKER_TYPE_SEM:
181 pr_err("INFO: task %s:%d is blocked on a semaphore, but the last holder is not found.\n",
182 task->comm, task->pid);
183 break;
184 case BLOCKER_TYPE_RWSEM_READER:
185 case BLOCKER_TYPE_RWSEM_WRITER:
186 pr_err("INFO: task %s:%d is blocked on an rw-semaphore, but the owner is not found.\n",
187 task->comm, task->pid);
188 break;
189 }
190 return;
191 }
192
193 /* Ensure the owner information is correct. */
194 for_each_process_thread(g, t) {
195 if ((unsigned long)t != owner)
196 continue;
197
198 switch (blocker_type) {
199 case BLOCKER_TYPE_MUTEX:
200 pr_err("INFO: task %s:%d is blocked on a mutex likely owned by task %s:%d.\n",
201 task->comm, task->pid, t->comm, t->pid);
202 break;
203 case BLOCKER_TYPE_SEM:
204 pr_err("INFO: task %s:%d blocked on a semaphore likely last held by task %s:%d\n",
205 task->comm, task->pid, t->comm, t->pid);
206 break;
207 case BLOCKER_TYPE_RWSEM_READER:
208 case BLOCKER_TYPE_RWSEM_WRITER:
209 pr_err("INFO: task %s:%d <%s> blocked on an rw-semaphore likely owned by task %s:%d <%s>\n",
210 task->comm, task->pid, rwsem_blocked_as, t->comm,
211 t->pid, rwsem_blocked_by);
212 break;
213 }
214 /* Avoid duplicated task dump, skip if the task is also hung. */
215 if (!task_is_hung(t, timeout))
216 sched_show_task(t);
217 return;
218 }
219 }
220 #else
debug_show_blocker(struct task_struct * task,unsigned long timeout)221 static inline void debug_show_blocker(struct task_struct *task, unsigned long timeout)
222 {
223 }
224 #endif
225
check_hung_task(struct task_struct * t,unsigned long timeout,unsigned long prev_detect_count)226 static void check_hung_task(struct task_struct *t, unsigned long timeout,
227 unsigned long prev_detect_count)
228 {
229 unsigned long total_hung_task;
230
231 if (!task_is_hung(t, timeout))
232 return;
233
234 /*
235 * This counter tracks the total number of tasks detected as hung
236 * since boot.
237 */
238 sysctl_hung_task_detect_count++;
239
240 total_hung_task = sysctl_hung_task_detect_count - prev_detect_count;
241 trace_sched_process_hang(t);
242
243 if (sysctl_hung_task_panic && total_hung_task >= sysctl_hung_task_panic) {
244 console_verbose();
245 hung_task_call_panic = true;
246 }
247
248 /*
249 * Ok, the task did not get scheduled for more than 2 minutes,
250 * complain:
251 */
252 if (sysctl_hung_task_warnings || hung_task_call_panic) {
253 if (sysctl_hung_task_warnings > 0)
254 sysctl_hung_task_warnings--;
255 pr_err("INFO: task %s:%d blocked for more than %ld seconds.\n",
256 t->comm, t->pid, (jiffies - t->last_switch_time) / HZ);
257 pr_err(" %s %s %.*s\n",
258 print_tainted(), init_utsname()->release,
259 (int)strcspn(init_utsname()->version, " "),
260 init_utsname()->version);
261 if (t->flags & PF_POSTCOREDUMP)
262 pr_err(" Blocked by coredump.\n");
263 pr_err("\"echo 0 > /proc/sys/kernel/hung_task_timeout_secs\""
264 " disables this message.\n");
265 sched_show_task(t);
266 debug_show_blocker(t, timeout);
267
268 if (!sysctl_hung_task_warnings)
269 pr_info("Future hung task reports are suppressed, see sysctl kernel.hung_task_warnings\n");
270 }
271
272 touch_nmi_watchdog();
273 }
274
275 /*
276 * To avoid extending the RCU grace period for an unbounded amount of time,
277 * periodically exit the critical section and enter a new one.
278 *
279 * For preemptible RCU it is sufficient to call rcu_read_unlock in order
280 * to exit the grace period. For classic RCU, a reschedule is required.
281 */
rcu_lock_break(struct task_struct * g,struct task_struct * t)282 static bool rcu_lock_break(struct task_struct *g, struct task_struct *t)
283 {
284 bool can_cont;
285
286 get_task_struct(g);
287 get_task_struct(t);
288 rcu_read_unlock();
289 cond_resched();
290 rcu_read_lock();
291 can_cont = pid_alive(g) && pid_alive(t);
292 put_task_struct(t);
293 put_task_struct(g);
294
295 return can_cont;
296 }
297
298 /*
299 * Check whether a TASK_UNINTERRUPTIBLE does not get woken up for
300 * a really long time (120 seconds). If that happens, print out
301 * a warning.
302 */
check_hung_uninterruptible_tasks(unsigned long timeout)303 static void check_hung_uninterruptible_tasks(unsigned long timeout)
304 {
305 int max_count = sysctl_hung_task_check_count;
306 unsigned long last_break = jiffies;
307 struct task_struct *g, *t;
308 unsigned long prev_detect_count = sysctl_hung_task_detect_count;
309 int need_warning = sysctl_hung_task_warnings;
310 unsigned long si_mask = hung_task_si_mask;
311
312 /*
313 * If the system crashed already then all bets are off,
314 * do not report extra hung tasks:
315 */
316 if (test_taint(TAINT_DIE) || did_panic)
317 return;
318
319
320 rcu_read_lock();
321 for_each_process_thread(g, t) {
322
323 if (!max_count--)
324 goto unlock;
325 if (time_after(jiffies, last_break + HUNG_TASK_LOCK_BREAK)) {
326 if (!rcu_lock_break(g, t))
327 goto unlock;
328 last_break = jiffies;
329 }
330
331 check_hung_task(t, timeout, prev_detect_count);
332 }
333 unlock:
334 rcu_read_unlock();
335
336 if (!(sysctl_hung_task_detect_count - prev_detect_count))
337 return;
338
339 if (need_warning || hung_task_call_panic) {
340 si_mask |= SYS_INFO_LOCKS;
341
342 if (sysctl_hung_task_all_cpu_backtrace)
343 si_mask |= SYS_INFO_ALL_BT;
344 }
345
346 sys_info(si_mask);
347
348 if (hung_task_call_panic)
349 panic("hung_task: blocked tasks");
350 }
351
hung_timeout_jiffies(unsigned long last_checked,unsigned long timeout)352 static long hung_timeout_jiffies(unsigned long last_checked,
353 unsigned long timeout)
354 {
355 /* timeout of 0 will disable the watchdog */
356 return timeout ? last_checked - jiffies + timeout * HZ :
357 MAX_SCHEDULE_TIMEOUT;
358 }
359
360 #ifdef CONFIG_SYSCTL
361 /*
362 * Process updating of timeout sysctl
363 */
proc_dohung_task_timeout_secs(const struct ctl_table * table,int write,void * buffer,size_t * lenp,loff_t * ppos)364 static int proc_dohung_task_timeout_secs(const struct ctl_table *table, int write,
365 void *buffer,
366 size_t *lenp, loff_t *ppos)
367 {
368 int ret;
369
370 ret = proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
371
372 if (ret || !write)
373 goto out;
374
375 wake_up_process(watchdog_task);
376
377 out:
378 return ret;
379 }
380
381 /*
382 * This is needed for proc_doulongvec_minmax of sysctl_hung_task_timeout_secs
383 * and hung_task_check_interval_secs
384 */
385 static const unsigned long hung_task_timeout_max = (LONG_MAX / HZ);
386 static const struct ctl_table hung_task_sysctls[] = {
387 #ifdef CONFIG_SMP
388 {
389 .procname = "hung_task_all_cpu_backtrace",
390 .data = &sysctl_hung_task_all_cpu_backtrace,
391 .maxlen = sizeof(int),
392 .mode = 0644,
393 .proc_handler = proc_dointvec_minmax,
394 .extra1 = SYSCTL_ZERO,
395 .extra2 = SYSCTL_ONE,
396 },
397 #endif /* CONFIG_SMP */
398 {
399 .procname = "hung_task_panic",
400 .data = &sysctl_hung_task_panic,
401 .maxlen = sizeof(int),
402 .mode = 0644,
403 .proc_handler = proc_dointvec_minmax,
404 .extra1 = SYSCTL_ZERO,
405 .extra2 = SYSCTL_INT_MAX,
406 },
407 {
408 .procname = "hung_task_check_count",
409 .data = &sysctl_hung_task_check_count,
410 .maxlen = sizeof(int),
411 .mode = 0644,
412 .proc_handler = proc_dointvec_minmax,
413 .extra1 = SYSCTL_ZERO,
414 },
415 {
416 .procname = "hung_task_timeout_secs",
417 .data = &sysctl_hung_task_timeout_secs,
418 .maxlen = sizeof(unsigned long),
419 .mode = 0644,
420 .proc_handler = proc_dohung_task_timeout_secs,
421 .extra2 = (void *)&hung_task_timeout_max,
422 },
423 {
424 .procname = "hung_task_check_interval_secs",
425 .data = &sysctl_hung_task_check_interval_secs,
426 .maxlen = sizeof(unsigned long),
427 .mode = 0644,
428 .proc_handler = proc_dohung_task_timeout_secs,
429 .extra2 = (void *)&hung_task_timeout_max,
430 },
431 {
432 .procname = "hung_task_warnings",
433 .data = &sysctl_hung_task_warnings,
434 .maxlen = sizeof(int),
435 .mode = 0644,
436 .proc_handler = proc_dointvec_minmax,
437 .extra1 = SYSCTL_NEG_ONE,
438 },
439 {
440 .procname = "hung_task_detect_count",
441 .data = &sysctl_hung_task_detect_count,
442 .maxlen = sizeof(unsigned long),
443 .mode = 0444,
444 .proc_handler = proc_doulongvec_minmax,
445 },
446 {
447 .procname = "hung_task_sys_info",
448 .data = &hung_task_si_mask,
449 .maxlen = sizeof(hung_task_si_mask),
450 .mode = 0644,
451 .proc_handler = sysctl_sys_info_handler,
452 },
453 };
454
hung_task_sysctl_init(void)455 static void __init hung_task_sysctl_init(void)
456 {
457 register_sysctl_init("kernel", hung_task_sysctls);
458 }
459 #else
460 #define hung_task_sysctl_init() do { } while (0)
461 #endif /* CONFIG_SYSCTL */
462
463
464 static atomic_t reset_hung_task = ATOMIC_INIT(0);
465
reset_hung_task_detector(void)466 void reset_hung_task_detector(void)
467 {
468 atomic_set(&reset_hung_task, 1);
469 }
470 EXPORT_SYMBOL_GPL(reset_hung_task_detector);
471
472 static bool hung_detector_suspended;
473
hungtask_pm_notify(struct notifier_block * self,unsigned long action,void * hcpu)474 static int hungtask_pm_notify(struct notifier_block *self,
475 unsigned long action, void *hcpu)
476 {
477 switch (action) {
478 case PM_SUSPEND_PREPARE:
479 case PM_HIBERNATION_PREPARE:
480 case PM_RESTORE_PREPARE:
481 hung_detector_suspended = true;
482 break;
483 case PM_POST_SUSPEND:
484 case PM_POST_HIBERNATION:
485 case PM_POST_RESTORE:
486 hung_detector_suspended = false;
487 break;
488 default:
489 break;
490 }
491 return NOTIFY_OK;
492 }
493
494 /*
495 * kthread which checks for tasks stuck in D state
496 */
watchdog(void * dummy)497 static int watchdog(void *dummy)
498 {
499 unsigned long hung_last_checked = jiffies;
500
501 set_user_nice(current, 0);
502
503 for ( ; ; ) {
504 unsigned long timeout = sysctl_hung_task_timeout_secs;
505 unsigned long interval = sysctl_hung_task_check_interval_secs;
506 long t;
507
508 if (interval == 0)
509 interval = timeout;
510 interval = min_t(unsigned long, interval, timeout);
511 t = hung_timeout_jiffies(hung_last_checked, interval);
512 if (t <= 0) {
513 if (!atomic_xchg(&reset_hung_task, 0) &&
514 !hung_detector_suspended)
515 check_hung_uninterruptible_tasks(timeout);
516 hung_last_checked = jiffies;
517 continue;
518 }
519 schedule_timeout_interruptible(t);
520 }
521
522 return 0;
523 }
524
hung_task_init(void)525 static int __init hung_task_init(void)
526 {
527 atomic_notifier_chain_register(&panic_notifier_list, &panic_block);
528
529 /* Disable hung task detector on suspend */
530 pm_notifier(hungtask_pm_notify, 0);
531
532 watchdog_task = kthread_run(watchdog, NULL, "khungtaskd");
533 hung_task_sysctl_init();
534
535 return 0;
536 }
537 subsys_initcall(hung_task_init);
538