Mandeep Singh Baines | e162b39 | 2009-01-15 11:08:40 -0800 | [diff] [blame] | 1 | /* |
| 2 | * Detect Hung Task |
| 3 | * |
| 4 | * kernel/hung_task.c - kernel thread for detecting tasks stuck in D state |
| 5 | * |
| 6 | */ |
| 7 | |
| 8 | #include <linux/mm.h> |
| 9 | #include <linux/cpu.h> |
| 10 | #include <linux/nmi.h> |
| 11 | #include <linux/init.h> |
| 12 | #include <linux/delay.h> |
| 13 | #include <linux/freezer.h> |
| 14 | #include <linux/kthread.h> |
| 15 | #include <linux/lockdep.h> |
Paul Gortmaker | 9984de1 | 2011-05-23 14:51:41 -0400 | [diff] [blame] | 16 | #include <linux/export.h> |
Mandeep Singh Baines | e162b39 | 2009-01-15 11:08:40 -0800 | [diff] [blame] | 17 | #include <linux/sysctl.h> |
Oleg Nesterov | 41e85ce | 2013-08-01 18:59:41 +0200 | [diff] [blame] | 18 | #include <linux/utsname.h> |
Oleg Nesterov | 6a716c9 | 2013-10-19 18:18:28 +0200 | [diff] [blame] | 19 | #include <trace/events/sched.h> |
Imran Khan | a9788f4 | 2017-09-18 15:48:30 +0530 | [diff] [blame^] | 20 | #include <linux/sched/sysctl.h> |
Mandeep Singh Baines | e162b39 | 2009-01-15 11:08:40 -0800 | [diff] [blame] | 21 | |
| 22 | /* |
Mandeep Singh Baines | ce9dbe2 | 2009-02-04 20:35:48 -0800 | [diff] [blame] | 23 | * The number of tasks checked: |
Mandeep Singh Baines | e162b39 | 2009-01-15 11:08:40 -0800 | [diff] [blame] | 24 | */ |
Li Zefan | cd64647 | 2013-09-23 16:43:58 +0800 | [diff] [blame] | 25 | int __read_mostly sysctl_hung_task_check_count = PID_MAX_LIMIT; |
Mandeep Singh Baines | ce9dbe2 | 2009-02-04 20:35:48 -0800 | [diff] [blame] | 26 | |
| 27 | /* |
Imran Khan | a9788f4 | 2017-09-18 15:48:30 +0530 | [diff] [blame^] | 28 | * Selective monitoring of hung tasks. |
| 29 | * |
| 30 | * if set to 1, khungtaskd skips monitoring tasks, which has |
| 31 | * task_struct->hang_detection_enabled value not set, else monitors all tasks. |
| 32 | */ |
| 33 | int sysctl_hung_task_selective_monitoring = 1; |
| 34 | |
| 35 | /* |
Mandeep Singh Baines | ce9dbe2 | 2009-02-04 20:35:48 -0800 | [diff] [blame] | 36 | * Limit number of tasks checked in a batch. |
| 37 | * |
| 38 | * This value controls the preemptibility of khungtaskd since preemption |
| 39 | * is disabled during the critical section. It also controls the size of |
| 40 | * the RCU grace period. So it needs to be upper-bound. |
| 41 | */ |
| 42 | #define HUNG_TASK_BATCHING 1024 |
Mandeep Singh Baines | e162b39 | 2009-01-15 11:08:40 -0800 | [diff] [blame] | 43 | |
| 44 | /* |
| 45 | * Zero means infinite timeout - no checking done: |
| 46 | */ |
Jeff Mahoney | e11feaa | 2011-04-27 14:27:24 -0400 | [diff] [blame] | 47 | unsigned long __read_mostly sysctl_hung_task_timeout_secs = CONFIG_DEFAULT_HUNG_TASK_TIMEOUT; |
Mandeep Singh Baines | e162b39 | 2009-01-15 11:08:40 -0800 | [diff] [blame] | 48 | |
Aaron Tomlin | 270750db | 2014-01-20 17:34:13 +0000 | [diff] [blame] | 49 | int __read_mostly sysctl_hung_task_warnings = 10; |
Mandeep Singh Baines | e162b39 | 2009-01-15 11:08:40 -0800 | [diff] [blame] | 50 | |
| 51 | static int __read_mostly did_panic; |
| 52 | |
| 53 | static struct task_struct *watchdog_task; |
| 54 | |
| 55 | /* |
| 56 | * Should we panic (and reboot, if panic_timeout= is set) when a |
| 57 | * hung task is detected: |
| 58 | */ |
| 59 | unsigned int __read_mostly sysctl_hung_task_panic = |
| 60 | CONFIG_BOOTPARAM_HUNG_TASK_PANIC_VALUE; |
| 61 | |
| 62 | static int __init hung_task_panic_setup(char *str) |
| 63 | { |
Fabian Frederick | b51dbec | 2014-06-04 16:11:26 -0700 | [diff] [blame] | 64 | int rc = kstrtouint(str, 0, &sysctl_hung_task_panic); |
Mandeep Singh Baines | e162b39 | 2009-01-15 11:08:40 -0800 | [diff] [blame] | 65 | |
Fabian Frederick | b51dbec | 2014-06-04 16:11:26 -0700 | [diff] [blame] | 66 | if (rc) |
| 67 | return rc; |
Mandeep Singh Baines | e162b39 | 2009-01-15 11:08:40 -0800 | [diff] [blame] | 68 | return 1; |
| 69 | } |
| 70 | __setup("hung_task_panic=", hung_task_panic_setup); |
| 71 | |
| 72 | static int |
| 73 | hung_task_panic(struct notifier_block *this, unsigned long event, void *ptr) |
| 74 | { |
| 75 | did_panic = 1; |
| 76 | |
| 77 | return NOTIFY_DONE; |
| 78 | } |
| 79 | |
| 80 | static struct notifier_block panic_block = { |
| 81 | .notifier_call = hung_task_panic, |
| 82 | }; |
| 83 | |
Mandeep Singh Baines | 17406b8 | 2009-02-06 15:37:47 -0800 | [diff] [blame] | 84 | static void check_hung_task(struct task_struct *t, unsigned long timeout) |
Mandeep Singh Baines | e162b39 | 2009-01-15 11:08:40 -0800 | [diff] [blame] | 85 | { |
| 86 | unsigned long switch_count = t->nvcsw + t->nivcsw; |
| 87 | |
Frederic Weisbecker | cf2592f | 2009-02-10 16:52:37 +0100 | [diff] [blame] | 88 | /* |
| 89 | * Ensure the task is not frozen. |
Mandeep Singh Baines | f9fab10 | 2012-01-03 14:41:13 -0800 | [diff] [blame] | 90 | * Also, skip vfork and any other user process that freezer should skip. |
Frederic Weisbecker | cf2592f | 2009-02-10 16:52:37 +0100 | [diff] [blame] | 91 | */ |
Mandeep Singh Baines | f9fab10 | 2012-01-03 14:41:13 -0800 | [diff] [blame] | 92 | if (unlikely(t->flags & (PF_FROZEN | PF_FREEZER_SKIP))) |
| 93 | return; |
| 94 | |
| 95 | /* |
| 96 | * When a freshly created task is scheduled once, changes its state to |
| 97 | * TASK_UNINTERRUPTIBLE without having ever been switched out once, it |
| 98 | * musn't be checked. |
| 99 | */ |
| 100 | if (unlikely(!switch_count)) |
Mandeep Singh Baines | e162b39 | 2009-01-15 11:08:40 -0800 | [diff] [blame] | 101 | return; |
| 102 | |
Mandeep Singh Baines | 17406b8 | 2009-02-06 15:37:47 -0800 | [diff] [blame] | 103 | if (switch_count != t->last_switch_count) { |
Mandeep Singh Baines | e162b39 | 2009-01-15 11:08:40 -0800 | [diff] [blame] | 104 | t->last_switch_count = switch_count; |
Mandeep Singh Baines | e162b39 | 2009-01-15 11:08:40 -0800 | [diff] [blame] | 105 | return; |
| 106 | } |
Oleg Nesterov | 6a716c9 | 2013-10-19 18:18:28 +0200 | [diff] [blame] | 107 | |
| 108 | trace_sched_process_hang(t); |
| 109 | |
John Siddle | 48a6d64 | 2016-10-11 13:55:56 -0700 | [diff] [blame] | 110 | if (!sysctl_hung_task_warnings && !sysctl_hung_task_panic) |
Mandeep Singh Baines | e162b39 | 2009-01-15 11:08:40 -0800 | [diff] [blame] | 111 | return; |
Aaron Tomlin | 270750db | 2014-01-20 17:34:13 +0000 | [diff] [blame] | 112 | |
Mandeep Singh Baines | e162b39 | 2009-01-15 11:08:40 -0800 | [diff] [blame] | 113 | /* |
| 114 | * Ok, the task did not get scheduled for more than 2 minutes, |
| 115 | * complain: |
| 116 | */ |
John Siddle | 48a6d64 | 2016-10-11 13:55:56 -0700 | [diff] [blame] | 117 | if (sysctl_hung_task_warnings) { |
| 118 | sysctl_hung_task_warnings--; |
| 119 | pr_err("INFO: task %s:%d blocked for more than %ld seconds.\n", |
| 120 | t->comm, t->pid, timeout); |
| 121 | pr_err(" %s %s %.*s\n", |
| 122 | print_tainted(), init_utsname()->release, |
| 123 | (int)strcspn(init_utsname()->version, " "), |
| 124 | init_utsname()->version); |
| 125 | pr_err("\"echo 0 > /proc/sys/kernel/hung_task_timeout_secs\"" |
| 126 | " disables this message.\n"); |
| 127 | sched_show_task(t); |
| 128 | debug_show_all_locks(); |
| 129 | } |
Mandeep Singh Baines | e162b39 | 2009-01-15 11:08:40 -0800 | [diff] [blame] | 130 | |
Mandeep Singh Baines | e162b39 | 2009-01-15 11:08:40 -0800 | [diff] [blame] | 131 | touch_nmi_watchdog(); |
| 132 | |
Sasha Levin | 625056b | 2012-03-15 17:47:20 -0400 | [diff] [blame] | 133 | if (sysctl_hung_task_panic) { |
| 134 | trigger_all_cpu_backtrace(); |
Mandeep Singh Baines | e162b39 | 2009-01-15 11:08:40 -0800 | [diff] [blame] | 135 | panic("hung_task: blocked tasks"); |
Sasha Levin | 625056b | 2012-03-15 17:47:20 -0400 | [diff] [blame] | 136 | } |
Mandeep Singh Baines | e162b39 | 2009-01-15 11:08:40 -0800 | [diff] [blame] | 137 | } |
| 138 | |
| 139 | /* |
Mandeep Singh Baines | ce9dbe2 | 2009-02-04 20:35:48 -0800 | [diff] [blame] | 140 | * To avoid extending the RCU grace period for an unbounded amount of time, |
| 141 | * periodically exit the critical section and enter a new one. |
| 142 | * |
| 143 | * For preemptible RCU it is sufficient to call rcu_read_unlock in order |
John Kacur | 6a103b0 | 2010-08-05 17:10:54 +0200 | [diff] [blame] | 144 | * to exit the grace period. For classic RCU, a reschedule is required. |
Mandeep Singh Baines | ce9dbe2 | 2009-02-04 20:35:48 -0800 | [diff] [blame] | 145 | */ |
Oleg Nesterov | 6027ce4 | 2012-03-05 14:59:14 -0800 | [diff] [blame] | 146 | static bool rcu_lock_break(struct task_struct *g, struct task_struct *t) |
Mandeep Singh Baines | ce9dbe2 | 2009-02-04 20:35:48 -0800 | [diff] [blame] | 147 | { |
Oleg Nesterov | 6027ce4 | 2012-03-05 14:59:14 -0800 | [diff] [blame] | 148 | bool can_cont; |
| 149 | |
Mandeep Singh Baines | ce9dbe2 | 2009-02-04 20:35:48 -0800 | [diff] [blame] | 150 | get_task_struct(g); |
| 151 | get_task_struct(t); |
| 152 | rcu_read_unlock(); |
| 153 | cond_resched(); |
| 154 | rcu_read_lock(); |
Oleg Nesterov | 6027ce4 | 2012-03-05 14:59:14 -0800 | [diff] [blame] | 155 | can_cont = pid_alive(g) && pid_alive(t); |
Mandeep Singh Baines | ce9dbe2 | 2009-02-04 20:35:48 -0800 | [diff] [blame] | 156 | put_task_struct(t); |
| 157 | put_task_struct(g); |
Oleg Nesterov | 6027ce4 | 2012-03-05 14:59:14 -0800 | [diff] [blame] | 158 | |
| 159 | return can_cont; |
Mandeep Singh Baines | ce9dbe2 | 2009-02-04 20:35:48 -0800 | [diff] [blame] | 160 | } |
| 161 | |
| 162 | /* |
Mandeep Singh Baines | e162b39 | 2009-01-15 11:08:40 -0800 | [diff] [blame] | 163 | * Check whether a TASK_UNINTERRUPTIBLE does not get woken up for |
| 164 | * a really long time (120 seconds). If that happens, print out |
| 165 | * a warning. |
| 166 | */ |
Mandeep Singh Baines | 603a148 | 2009-01-17 10:31:48 -0800 | [diff] [blame] | 167 | static void check_hung_uninterruptible_tasks(unsigned long timeout) |
Mandeep Singh Baines | e162b39 | 2009-01-15 11:08:40 -0800 | [diff] [blame] | 168 | { |
| 169 | int max_count = sysctl_hung_task_check_count; |
Mandeep Singh Baines | ce9dbe2 | 2009-02-04 20:35:48 -0800 | [diff] [blame] | 170 | int batch_count = HUNG_TASK_BATCHING; |
Mandeep Singh Baines | e162b39 | 2009-01-15 11:08:40 -0800 | [diff] [blame] | 171 | struct task_struct *g, *t; |
| 172 | |
| 173 | /* |
| 174 | * If the system crashed already then all bets are off, |
| 175 | * do not report extra hung tasks: |
| 176 | */ |
| 177 | if (test_taint(TAINT_DIE) || did_panic) |
| 178 | return; |
| 179 | |
Mandeep Singh Baines | 94be52d | 2009-02-05 09:56:08 -0800 | [diff] [blame] | 180 | rcu_read_lock(); |
Aaron Tomlin | 972fae6 | 2015-04-15 16:16:47 -0700 | [diff] [blame] | 181 | for_each_process_thread(g, t) { |
Anton Blanchard | e5af022 | 2009-11-27 13:28:20 +1100 | [diff] [blame] | 182 | if (!max_count--) |
Mandeep Singh Baines | e162b39 | 2009-01-15 11:08:40 -0800 | [diff] [blame] | 183 | goto unlock; |
Mandeep Singh Baines | ce9dbe2 | 2009-02-04 20:35:48 -0800 | [diff] [blame] | 184 | if (!--batch_count) { |
| 185 | batch_count = HUNG_TASK_BATCHING; |
Oleg Nesterov | 6027ce4 | 2012-03-05 14:59:14 -0800 | [diff] [blame] | 186 | if (!rcu_lock_break(g, t)) |
Mandeep Singh Baines | ce9dbe2 | 2009-02-04 20:35:48 -0800 | [diff] [blame] | 187 | goto unlock; |
| 188 | } |
Mandeep Singh Baines | e162b39 | 2009-01-15 11:08:40 -0800 | [diff] [blame] | 189 | /* use "==" to skip the TASK_KILLABLE tasks waiting on NFS */ |
| 190 | if (t->state == TASK_UNINTERRUPTIBLE) |
Imran Khan | a9788f4 | 2017-09-18 15:48:30 +0530 | [diff] [blame^] | 191 | /* Check for selective monitoring */ |
| 192 | if (!sysctl_hung_task_selective_monitoring || |
| 193 | t->hang_detection_enabled) |
| 194 | check_hung_task(t, timeout); |
Aaron Tomlin | 972fae6 | 2015-04-15 16:16:47 -0700 | [diff] [blame] | 195 | } |
Mandeep Singh Baines | e162b39 | 2009-01-15 11:08:40 -0800 | [diff] [blame] | 196 | unlock: |
Mandeep Singh Baines | 94be52d | 2009-02-05 09:56:08 -0800 | [diff] [blame] | 197 | rcu_read_unlock(); |
Mandeep Singh Baines | e162b39 | 2009-01-15 11:08:40 -0800 | [diff] [blame] | 198 | } |
| 199 | |
Tetsuo Handa | b4aa14a | 2016-03-22 14:24:39 -0700 | [diff] [blame] | 200 | static long hung_timeout_jiffies(unsigned long last_checked, |
| 201 | unsigned long timeout) |
Mandeep Singh Baines | e162b39 | 2009-01-15 11:08:40 -0800 | [diff] [blame] | 202 | { |
| 203 | /* timeout of 0 will disable the watchdog */ |
Tetsuo Handa | b4aa14a | 2016-03-22 14:24:39 -0700 | [diff] [blame] | 204 | return timeout ? last_checked - jiffies + timeout * HZ : |
| 205 | MAX_SCHEDULE_TIMEOUT; |
Mandeep Singh Baines | e162b39 | 2009-01-15 11:08:40 -0800 | [diff] [blame] | 206 | } |
| 207 | |
| 208 | /* |
| 209 | * Process updating of timeout sysctl |
| 210 | */ |
| 211 | int proc_dohung_task_timeout_secs(struct ctl_table *table, int write, |
Alexey Dobriyan | 8d65af7 | 2009-09-23 15:57:19 -0700 | [diff] [blame] | 212 | void __user *buffer, |
Mandeep Singh Baines | e162b39 | 2009-01-15 11:08:40 -0800 | [diff] [blame] | 213 | size_t *lenp, loff_t *ppos) |
| 214 | { |
| 215 | int ret; |
| 216 | |
Alexey Dobriyan | 8d65af7 | 2009-09-23 15:57:19 -0700 | [diff] [blame] | 217 | ret = proc_doulongvec_minmax(table, write, buffer, lenp, ppos); |
Mandeep Singh Baines | e162b39 | 2009-01-15 11:08:40 -0800 | [diff] [blame] | 218 | |
| 219 | if (ret || !write) |
| 220 | goto out; |
| 221 | |
Mandeep Singh Baines | e162b39 | 2009-01-15 11:08:40 -0800 | [diff] [blame] | 222 | wake_up_process(watchdog_task); |
| 223 | |
| 224 | out: |
| 225 | return ret; |
| 226 | } |
| 227 | |
Marcelo Tosatti | 8b41452 | 2013-10-11 21:39:26 -0300 | [diff] [blame] | 228 | static atomic_t reset_hung_task = ATOMIC_INIT(0); |
| 229 | |
| 230 | void reset_hung_task_detector(void) |
| 231 | { |
| 232 | atomic_set(&reset_hung_task, 1); |
| 233 | } |
| 234 | EXPORT_SYMBOL_GPL(reset_hung_task_detector); |
| 235 | |
Mandeep Singh Baines | e162b39 | 2009-01-15 11:08:40 -0800 | [diff] [blame] | 236 | /* |
| 237 | * kthread which checks for tasks stuck in D state |
| 238 | */ |
| 239 | static int watchdog(void *dummy) |
| 240 | { |
Tetsuo Handa | b4aa14a | 2016-03-22 14:24:39 -0700 | [diff] [blame] | 241 | unsigned long hung_last_checked = jiffies; |
| 242 | |
Mandeep Singh Baines | e162b39 | 2009-01-15 11:08:40 -0800 | [diff] [blame] | 243 | set_user_nice(current, 0); |
Mandeep Singh Baines | e162b39 | 2009-01-15 11:08:40 -0800 | [diff] [blame] | 244 | |
| 245 | for ( ; ; ) { |
Mandeep Singh Baines | 17406b8 | 2009-02-06 15:37:47 -0800 | [diff] [blame] | 246 | unsigned long timeout = sysctl_hung_task_timeout_secs; |
Tetsuo Handa | b4aa14a | 2016-03-22 14:24:39 -0700 | [diff] [blame] | 247 | long t = hung_timeout_jiffies(hung_last_checked, timeout); |
Mandeep Singh Baines | 603a148 | 2009-01-17 10:31:48 -0800 | [diff] [blame] | 248 | |
Tetsuo Handa | b4aa14a | 2016-03-22 14:24:39 -0700 | [diff] [blame] | 249 | if (t <= 0) { |
| 250 | if (!atomic_xchg(&reset_hung_task, 0)) |
| 251 | check_hung_uninterruptible_tasks(timeout); |
| 252 | hung_last_checked = jiffies; |
Marcelo Tosatti | 8b41452 | 2013-10-11 21:39:26 -0300 | [diff] [blame] | 253 | continue; |
Tetsuo Handa | b4aa14a | 2016-03-22 14:24:39 -0700 | [diff] [blame] | 254 | } |
| 255 | schedule_timeout_interruptible(t); |
Mandeep Singh Baines | e162b39 | 2009-01-15 11:08:40 -0800 | [diff] [blame] | 256 | } |
| 257 | |
| 258 | return 0; |
| 259 | } |
| 260 | |
| 261 | static int __init hung_task_init(void) |
| 262 | { |
| 263 | atomic_notifier_chain_register(&panic_notifier_list, &panic_block); |
| 264 | watchdog_task = kthread_run(watchdog, NULL, "khungtaskd"); |
| 265 | |
| 266 | return 0; |
| 267 | } |
Paul Gortmaker | c96d666 | 2014-04-03 14:48:35 -0700 | [diff] [blame] | 268 | subsys_initcall(hung_task_init); |