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