Merge tag 'for-linux-6.12-ofs1' of git://git.kernel.org/pub/scm/linux/kernel/git...
[linux-2.6-block.git] / kernel / hung_task.c
CommitLineData
457c8996 1// SPDX-License-Identifier: GPL-2.0-only
e162b39a
MSB
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>
9984de1a 17#include <linux/export.h>
f39650de 18#include <linux/panic_notifier.h>
e162b39a 19#include <linux/sysctl.h>
a1c6ca3c 20#include <linux/suspend.h>
41e85ce8 21#include <linux/utsname.h>
3f07c014 22#include <linux/sched/signal.h>
b17b0153 23#include <linux/sched/debug.h>
a98eb6f1 24#include <linux/sched/sysctl.h>
3f07c014 25
6a716c90 26#include <trace/events/sched.h>
e162b39a
MSB
27
28/*
ce9dbe24 29 * The number of tasks checked:
e162b39a 30 */
882c5b26 31static int __read_mostly sysctl_hung_task_check_count = PID_MAX_LIMIT;
ce9dbe24
MSB
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 */
304ae427 40#define HUNG_TASK_LOCK_BREAK (HZ / 10)
e162b39a
MSB
41
42/*
43 * Zero means infinite timeout - no checking done:
44 */
e11feaa1 45unsigned long __read_mostly sysctl_hung_task_timeout_secs = CONFIG_DEFAULT_HUNG_TASK_TIMEOUT;
5c3273ec 46EXPORT_SYMBOL_GPL(sysctl_hung_task_timeout_secs);
e162b39a 47
a2e51445
DV
48/*
49 * Zero (default value) means use sysctl_hung_task_timeout_secs:
50 */
882c5b26 51static unsigned long __read_mostly sysctl_hung_task_check_interval_secs;
a2e51445 52
882c5b26 53static int __read_mostly sysctl_hung_task_warnings = 10;
e162b39a
MSB
54
55static int __read_mostly did_panic;
780cbcf2 56static bool hung_task_show_lock;
401c636a 57static bool hung_task_call_panic;
0ec9dc9b 58static bool hung_task_show_all_bt;
e162b39a
MSB
59
60static struct task_struct *watchdog_task;
61
0ec9dc9b
GP
62#ifdef CONFIG_SMP
63/*
64 * Should we dump all CPUs backtraces in a hung task event?
65 * Defaults to 0, can be changed via sysctl.
66 */
bbe7a10e
XN
67static unsigned int __read_mostly sysctl_hung_task_all_cpu_backtrace;
68#else
69#define sysctl_hung_task_all_cpu_backtrace 0
0ec9dc9b
GP
70#endif /* CONFIG_SMP */
71
e162b39a
MSB
72/*
73 * Should we panic (and reboot, if panic_timeout= is set) when a
74 * hung task is detected:
75 */
882c5b26
TR
76static unsigned int __read_mostly sysctl_hung_task_panic =
77 IS_ENABLED(CONFIG_BOOTPARAM_HUNG_TASK_PANIC);
e162b39a 78
e162b39a
MSB
79static int
80hung_task_panic(struct notifier_block *this, unsigned long event, void *ptr)
81{
82 did_panic = 1;
83
84 return NOTIFY_DONE;
85}
86
87static struct notifier_block panic_block = {
88 .notifier_call = hung_task_panic,
89};
90
17406b82 91static void check_hung_task(struct task_struct *t, unsigned long timeout)
e162b39a
MSB
92{
93 unsigned long switch_count = t->nvcsw + t->nivcsw;
94
cf2592f5
FW
95 /*
96 * Ensure the task is not frozen.
f9fab10b 97 * Also, skip vfork and any other user process that freezer should skip.
cf2592f5 98 */
fdf756f7 99 if (unlikely(READ_ONCE(t->__state) & TASK_FROZEN))
f5d39b02 100 return;
f9fab10b
MSB
101
102 /*
103 * When a freshly created task is scheduled once, changes its state to
104 * TASK_UNINTERRUPTIBLE without having ever been switched out once, it
105 * musn't be checked.
106 */
107 if (unlikely(!switch_count))
e162b39a
MSB
108 return;
109
17406b82 110 if (switch_count != t->last_switch_count) {
e162b39a 111 t->last_switch_count = switch_count;
a2e51445 112 t->last_switch_time = jiffies;
e162b39a
MSB
113 return;
114 }
a2e51445
DV
115 if (time_is_after_jiffies(t->last_switch_time + timeout * HZ))
116 return;
6a716c90
ON
117
118 trace_sched_process_hang(t);
119
168e06f7
LC
120 if (sysctl_hung_task_panic) {
121 console_verbose();
122 hung_task_show_lock = true;
123 hung_task_call_panic = true;
124 }
270750db 125
e162b39a
MSB
126 /*
127 * Ok, the task did not get scheduled for more than 2 minutes,
128 * complain:
129 */
abd8ac05 130 if (sysctl_hung_task_warnings || hung_task_call_panic) {
4ca5ede0
TH
131 if (sysctl_hung_task_warnings > 0)
132 sysctl_hung_task_warnings--;
48a6d64e 133 pr_err("INFO: task %s:%d blocked for more than %ld seconds.\n",
b014beba 134 t->comm, t->pid, (jiffies - t->last_switch_time) / HZ);
48a6d64e
JS
135 pr_err(" %s %s %.*s\n",
136 print_tainted(), init_utsname()->release,
137 (int)strcspn(init_utsname()->version, " "),
138 init_utsname()->version);
139 pr_err("\"echo 0 > /proc/sys/kernel/hung_task_timeout_secs\""
140 " disables this message.\n");
141 sched_show_task(t);
780cbcf2 142 hung_task_show_lock = true;
0ec9dc9b
GP
143
144 if (sysctl_hung_task_all_cpu_backtrace)
145 hung_task_show_all_bt = true;
b1f712b3 146 if (!sysctl_hung_task_warnings)
147 pr_info("Future hung task reports are suppressed, see sysctl kernel.hung_task_warnings\n");
48a6d64e 148 }
e162b39a 149
e162b39a 150 touch_nmi_watchdog();
e162b39a
MSB
151}
152
ce9dbe24
MSB
153/*
154 * To avoid extending the RCU grace period for an unbounded amount of time,
155 * periodically exit the critical section and enter a new one.
156 *
157 * For preemptible RCU it is sufficient to call rcu_read_unlock in order
6a103b0d 158 * to exit the grace period. For classic RCU, a reschedule is required.
ce9dbe24 159 */
6027ce49 160static bool rcu_lock_break(struct task_struct *g, struct task_struct *t)
ce9dbe24 161{
6027ce49
ON
162 bool can_cont;
163
ce9dbe24
MSB
164 get_task_struct(g);
165 get_task_struct(t);
166 rcu_read_unlock();
167 cond_resched();
168 rcu_read_lock();
6027ce49 169 can_cont = pid_alive(g) && pid_alive(t);
ce9dbe24
MSB
170 put_task_struct(t);
171 put_task_struct(g);
6027ce49
ON
172
173 return can_cont;
ce9dbe24
MSB
174}
175
e162b39a
MSB
176/*
177 * Check whether a TASK_UNINTERRUPTIBLE does not get woken up for
178 * a really long time (120 seconds). If that happens, print out
179 * a warning.
180 */
603a148f 181static void check_hung_uninterruptible_tasks(unsigned long timeout)
e162b39a
MSB
182{
183 int max_count = sysctl_hung_task_check_count;
304ae427 184 unsigned long last_break = jiffies;
e162b39a
MSB
185 struct task_struct *g, *t;
186
187 /*
188 * If the system crashed already then all bets are off,
189 * do not report extra hung tasks:
190 */
191 if (test_taint(TAINT_DIE) || did_panic)
192 return;
193
780cbcf2 194 hung_task_show_lock = false;
94be52dc 195 rcu_read_lock();
972fae69 196 for_each_process_thread(g, t) {
5aec788a
PZ
197 unsigned int state;
198
e5af0226 199 if (!max_count--)
e162b39a 200 goto unlock;
304ae427 201 if (time_after(jiffies, last_break + HUNG_TASK_LOCK_BREAK)) {
6027ce49 202 if (!rcu_lock_break(g, t))
ce9dbe24 203 goto unlock;
304ae427 204 last_break = jiffies;
ce9dbe24 205 }
fdf756f7
PZ
206 /*
207 * skip the TASK_KILLABLE tasks -- these can be killed
208 * skip the TASK_IDLE tasks -- those are genuinely idle
209 */
5aec788a
PZ
210 state = READ_ONCE(t->__state);
211 if ((state & TASK_UNINTERRUPTIBLE) &&
fdf756f7
PZ
212 !(state & TASK_WAKEKILL) &&
213 !(state & TASK_NOLOAD))
17406b82 214 check_hung_task(t, timeout);
972fae69 215 }
e162b39a 216 unlock:
94be52dc 217 rcu_read_unlock();
07a22b61 218 if (hung_task_show_lock)
780cbcf2 219 debug_show_all_locks();
0ec9dc9b
GP
220
221 if (hung_task_show_all_bt) {
222 hung_task_show_all_bt = false;
401c636a 223 trigger_all_cpu_backtrace();
401c636a 224 }
0ec9dc9b
GP
225
226 if (hung_task_call_panic)
227 panic("hung_task: blocked tasks");
e162b39a
MSB
228}
229
b4aa14a6
TH
230static long hung_timeout_jiffies(unsigned long last_checked,
231 unsigned long timeout)
e162b39a
MSB
232{
233 /* timeout of 0 will disable the watchdog */
b4aa14a6
TH
234 return timeout ? last_checked - jiffies + timeout * HZ :
235 MAX_SCHEDULE_TIMEOUT;
e162b39a
MSB
236}
237
bbe7a10e 238#ifdef CONFIG_SYSCTL
e162b39a
MSB
239/*
240 * Process updating of timeout sysctl
241 */
78eb4ea2 242static int proc_dohung_task_timeout_secs(const struct ctl_table *table, int write,
591c32bd 243 void *buffer,
bbe7a10e 244 size_t *lenp, loff_t *ppos)
e162b39a
MSB
245{
246 int ret;
247
8d65af78 248 ret = proc_doulongvec_minmax(table, write, buffer, lenp, ppos);
e162b39a
MSB
249
250 if (ret || !write)
251 goto out;
252
e162b39a
MSB
253 wake_up_process(watchdog_task);
254
255 out:
256 return ret;
257}
258
bbe7a10e
XN
259/*
260 * This is needed for proc_doulongvec_minmax of sysctl_hung_task_timeout_secs
261 * and hung_task_check_interval_secs
262 */
263static const unsigned long hung_task_timeout_max = (LONG_MAX / HZ);
264static struct ctl_table hung_task_sysctls[] = {
265#ifdef CONFIG_SMP
266 {
267 .procname = "hung_task_all_cpu_backtrace",
268 .data = &sysctl_hung_task_all_cpu_backtrace,
269 .maxlen = sizeof(int),
270 .mode = 0644,
271 .proc_handler = proc_dointvec_minmax,
272 .extra1 = SYSCTL_ZERO,
273 .extra2 = SYSCTL_ONE,
274 },
275#endif /* CONFIG_SMP */
276 {
277 .procname = "hung_task_panic",
278 .data = &sysctl_hung_task_panic,
279 .maxlen = sizeof(int),
280 .mode = 0644,
281 .proc_handler = proc_dointvec_minmax,
282 .extra1 = SYSCTL_ZERO,
283 .extra2 = SYSCTL_ONE,
284 },
285 {
286 .procname = "hung_task_check_count",
287 .data = &sysctl_hung_task_check_count,
288 .maxlen = sizeof(int),
289 .mode = 0644,
290 .proc_handler = proc_dointvec_minmax,
291 .extra1 = SYSCTL_ZERO,
292 },
293 {
294 .procname = "hung_task_timeout_secs",
295 .data = &sysctl_hung_task_timeout_secs,
296 .maxlen = sizeof(unsigned long),
297 .mode = 0644,
298 .proc_handler = proc_dohung_task_timeout_secs,
299 .extra2 = (void *)&hung_task_timeout_max,
300 },
301 {
302 .procname = "hung_task_check_interval_secs",
303 .data = &sysctl_hung_task_check_interval_secs,
304 .maxlen = sizeof(unsigned long),
305 .mode = 0644,
306 .proc_handler = proc_dohung_task_timeout_secs,
307 .extra2 = (void *)&hung_task_timeout_max,
308 },
309 {
310 .procname = "hung_task_warnings",
311 .data = &sysctl_hung_task_warnings,
312 .maxlen = sizeof(int),
313 .mode = 0644,
314 .proc_handler = proc_dointvec_minmax,
315 .extra1 = SYSCTL_NEG_ONE,
316 },
bbe7a10e
XN
317};
318
319static 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
8b414521
MT
328static atomic_t reset_hung_task = ATOMIC_INIT(0);
329
330void reset_hung_task_detector(void)
331{
332 atomic_set(&reset_hung_task, 1);
333}
334EXPORT_SYMBOL_GPL(reset_hung_task_detector);
335
a1c6ca3c
VK
336static bool hung_detector_suspended;
337
338static 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
e162b39a
MSB
358/*
359 * kthread which checks for tasks stuck in D state
360 */
361static int watchdog(void *dummy)
362{
b4aa14a6
TH
363 unsigned long hung_last_checked = jiffies;
364
e162b39a 365 set_user_nice(current, 0);
e162b39a
MSB
366
367 for ( ; ; ) {
17406b82 368 unsigned long timeout = sysctl_hung_task_timeout_secs;
a2e51445
DV
369 unsigned long interval = sysctl_hung_task_check_interval_secs;
370 long t;
603a148f 371
a2e51445
DV
372 if (interval == 0)
373 interval = timeout;
374 interval = min_t(unsigned long, interval, timeout);
375 t = hung_timeout_jiffies(hung_last_checked, interval);
b4aa14a6 376 if (t <= 0) {
a1c6ca3c
VK
377 if (!atomic_xchg(&reset_hung_task, 0) &&
378 !hung_detector_suspended)
b4aa14a6
TH
379 check_hung_uninterruptible_tasks(timeout);
380 hung_last_checked = jiffies;
8b414521 381 continue;
b4aa14a6
TH
382 }
383 schedule_timeout_interruptible(t);
e162b39a
MSB
384 }
385
386 return 0;
387}
388
389static int __init hung_task_init(void)
390{
391 atomic_notifier_chain_register(&panic_notifier_list, &panic_block);
a1c6ca3c
VK
392
393 /* Disable hung task detector on suspend */
394 pm_notifier(hungtask_pm_notify, 0);
395
e162b39a 396 watchdog_task = kthread_run(watchdog, NULL, "khungtaskd");
bbe7a10e 397 hung_task_sysctl_init();
e162b39a
MSB
398
399 return 0;
400}
c96d6660 401subsys_initcall(hung_task_init);