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