4 * started by Ingo Molnar, Copyright (C) 2005, 2006 Red Hat, Inc.
6 * this code detects soft lockups: incidents in where on a CPU
7 * the kernel does not reschedule for 10 seconds or more.
10 #include <linux/cpu.h>
11 #include <linux/init.h>
12 #include <linux/delay.h>
13 #include <linux/kthread.h>
14 #include <linux/notifier.h>
15 #include <linux/module.h>
17 static DEFINE_SPINLOCK(print_lock);
19 static DEFINE_PER_CPU(unsigned long, touch_timestamp);
20 static DEFINE_PER_CPU(unsigned long, print_timestamp);
21 static DEFINE_PER_CPU(struct task_struct *, watchdog_task);
23 static int did_panic = 0;
26 softlock_panic(struct notifier_block *this, unsigned long event, void *ptr)
33 static struct notifier_block panic_block = {
34 .notifier_call = softlock_panic,
38 * Returns seconds, approximately. We don't need nanosecond
39 * resolution, and we don't need to waste time with a big divide when
42 static unsigned long get_timestamp(void)
44 return sched_clock() >> 30; /* 2^30 ~= 10^9 */
47 void touch_softlockup_watchdog(void)
49 __raw_get_cpu_var(touch_timestamp) = get_timestamp();
51 EXPORT_SYMBOL(touch_softlockup_watchdog);
53 void touch_all_softlockup_watchdogs(void)
57 /* Cause each CPU to re-update its timestamp rather than complain */
58 for_each_online_cpu(cpu)
59 per_cpu(touch_timestamp, cpu) = 0;
61 EXPORT_SYMBOL(touch_all_softlockup_watchdogs);
64 * This callback runs from the timer interrupt, and checks
65 * whether the watchdog thread has hung or not:
67 void softlockup_tick(void)
69 int this_cpu = smp_processor_id();
70 unsigned long touch_timestamp = per_cpu(touch_timestamp, this_cpu);
71 unsigned long print_timestamp;
74 if (touch_timestamp == 0) {
75 touch_softlockup_watchdog();
79 print_timestamp = per_cpu(print_timestamp, this_cpu);
81 /* report at most once a second */
82 if (print_timestamp < (touch_timestamp + 1) ||
84 !per_cpu(watchdog_task, this_cpu))
87 /* do not print during early bootup: */
88 if (unlikely(system_state != SYSTEM_RUNNING)) {
89 touch_softlockup_watchdog();
93 now = get_timestamp();
95 /* Wake up the high-prio watchdog task every second: */
96 if (now > (touch_timestamp + 1))
97 wake_up_process(per_cpu(watchdog_task, this_cpu));
99 /* Warn about unreasonable 10+ seconds delays: */
100 if (now > (touch_timestamp + 10)) {
101 per_cpu(print_timestamp, this_cpu) = touch_timestamp;
103 spin_lock(&print_lock);
104 printk(KERN_ERR "BUG: soft lockup detected on CPU#%d!\n",
107 spin_unlock(&print_lock);
112 * The watchdog thread - runs every second and touches the timestamp.
114 static int watchdog(void * __bind_cpu)
116 struct sched_param param = { .sched_priority = MAX_RT_PRIO-1 };
118 sched_setscheduler(current, SCHED_FIFO, ¶m);
119 current->flags |= PF_NOFREEZE;
121 /* initialize timestamp */
122 touch_softlockup_watchdog();
125 * Run briefly once per second to reset the softlockup timestamp.
126 * If this gets delayed for more than 10 seconds then the
127 * debug-printout triggers in softlockup_tick().
129 while (!kthread_should_stop()) {
130 set_current_state(TASK_INTERRUPTIBLE);
131 touch_softlockup_watchdog();
139 * Create/destroy watchdog threads as CPUs come and go:
142 cpu_callback(struct notifier_block *nfb, unsigned long action, void *hcpu)
144 int hotcpu = (unsigned long)hcpu;
145 struct task_struct *p;
149 BUG_ON(per_cpu(watchdog_task, hotcpu));
150 p = kthread_create(watchdog, hcpu, "watchdog/%d", hotcpu);
152 printk("watchdog for %i failed\n", hotcpu);
155 per_cpu(touch_timestamp, hotcpu) = 0;
156 per_cpu(watchdog_task, hotcpu) = p;
157 kthread_bind(p, hotcpu);
160 wake_up_process(per_cpu(watchdog_task, hotcpu));
162 #ifdef CONFIG_HOTPLUG_CPU
163 case CPU_UP_CANCELED:
164 if (!per_cpu(watchdog_task, hotcpu))
166 /* Unbind so it can run. Fall thru. */
167 kthread_bind(per_cpu(watchdog_task, hotcpu),
168 any_online_cpu(cpu_online_map));
170 p = per_cpu(watchdog_task, hotcpu);
171 per_cpu(watchdog_task, hotcpu) = NULL;
174 #endif /* CONFIG_HOTPLUG_CPU */
179 static struct notifier_block __cpuinitdata cpu_nfb = {
180 .notifier_call = cpu_callback
183 __init void spawn_softlockup_task(void)
185 void *cpu = (void *)(long)smp_processor_id();
186 int err = cpu_callback(&cpu_nfb, CPU_UP_PREPARE, cpu);
188 BUG_ON(err == NOTIFY_BAD);
189 cpu_callback(&cpu_nfb, CPU_ONLINE, cpu);
190 register_cpu_notifier(&cpu_nfb);
192 atomic_notifier_chain_register(&panic_notifier_list, &panic_block);