Commit | Line | Data |
---|---|---|
8446f1d3 IM |
1 | /* |
2 | * Detect Soft Lockups | |
3 | * | |
6687a97d | 4 | * started by Ingo Molnar, Copyright (C) 2005, 2006 Red Hat, Inc. |
8446f1d3 IM |
5 | * |
6 | * this code detects soft lockups: incidents in where on a CPU | |
7 | * the kernel does not reschedule for 10 seconds or more. | |
8 | */ | |
8446f1d3 IM |
9 | #include <linux/mm.h> |
10 | #include <linux/cpu.h> | |
11 | #include <linux/init.h> | |
12 | #include <linux/delay.h> | |
83144186 | 13 | #include <linux/freezer.h> |
8446f1d3 IM |
14 | #include <linux/kthread.h> |
15 | #include <linux/notifier.h> | |
16 | #include <linux/module.h> | |
17 | ||
43581a10 IM |
18 | #include <asm/irq_regs.h> |
19 | ||
8446f1d3 IM |
20 | static DEFINE_SPINLOCK(print_lock); |
21 | ||
6687a97d IM |
22 | static DEFINE_PER_CPU(unsigned long, touch_timestamp); |
23 | static DEFINE_PER_CPU(unsigned long, print_timestamp); | |
8446f1d3 IM |
24 | static DEFINE_PER_CPU(struct task_struct *, watchdog_task); |
25 | ||
a5f2ce3c | 26 | static int did_panic; |
6687a97d IM |
27 | |
28 | static int | |
29 | softlock_panic(struct notifier_block *this, unsigned long event, void *ptr) | |
8446f1d3 IM |
30 | { |
31 | did_panic = 1; | |
32 | ||
33 | return NOTIFY_DONE; | |
34 | } | |
35 | ||
36 | static struct notifier_block panic_block = { | |
37 | .notifier_call = softlock_panic, | |
38 | }; | |
39 | ||
966812dc JF |
40 | /* |
41 | * Returns seconds, approximately. We don't need nanosecond | |
42 | * resolution, and we don't need to waste time with a big divide when | |
43 | * 2^30ns == 1.074s. | |
44 | */ | |
a3b13c23 | 45 | static unsigned long get_timestamp(int this_cpu) |
966812dc | 46 | { |
a3b13c23 | 47 | return cpu_clock(this_cpu) >> 30; /* 2^30 ~= 10^9 */ |
966812dc JF |
48 | } |
49 | ||
8446f1d3 IM |
50 | void touch_softlockup_watchdog(void) |
51 | { | |
a3b13c23 IM |
52 | int this_cpu = raw_smp_processor_id(); |
53 | ||
54 | __raw_get_cpu_var(touch_timestamp) = get_timestamp(this_cpu); | |
8446f1d3 IM |
55 | } |
56 | EXPORT_SYMBOL(touch_softlockup_watchdog); | |
57 | ||
04c9167f JF |
58 | void touch_all_softlockup_watchdogs(void) |
59 | { | |
60 | int cpu; | |
61 | ||
62 | /* Cause each CPU to re-update its timestamp rather than complain */ | |
63 | for_each_online_cpu(cpu) | |
64 | per_cpu(touch_timestamp, cpu) = 0; | |
65 | } | |
66 | EXPORT_SYMBOL(touch_all_softlockup_watchdogs); | |
67 | ||
8446f1d3 IM |
68 | /* |
69 | * This callback runs from the timer interrupt, and checks | |
70 | * whether the watchdog thread has hung or not: | |
71 | */ | |
6687a97d | 72 | void softlockup_tick(void) |
8446f1d3 IM |
73 | { |
74 | int this_cpu = smp_processor_id(); | |
6687a97d | 75 | unsigned long touch_timestamp = per_cpu(touch_timestamp, this_cpu); |
966812dc | 76 | unsigned long print_timestamp; |
43581a10 | 77 | struct pt_regs *regs = get_irq_regs(); |
966812dc | 78 | unsigned long now; |
8446f1d3 | 79 | |
04c9167f JF |
80 | if (touch_timestamp == 0) { |
81 | touch_softlockup_watchdog(); | |
966812dc | 82 | return; |
04c9167f | 83 | } |
966812dc JF |
84 | |
85 | print_timestamp = per_cpu(print_timestamp, this_cpu); | |
86 | ||
87 | /* report at most once a second */ | |
a115d5ca IM |
88 | if ((print_timestamp >= touch_timestamp && |
89 | print_timestamp < (touch_timestamp + 1)) || | |
90 | did_panic || !per_cpu(watchdog_task, this_cpu)) { | |
8446f1d3 | 91 | return; |
a115d5ca | 92 | } |
8446f1d3 | 93 | |
6687a97d IM |
94 | /* do not print during early bootup: */ |
95 | if (unlikely(system_state != SYSTEM_RUNNING)) { | |
96 | touch_softlockup_watchdog(); | |
8446f1d3 | 97 | return; |
6687a97d | 98 | } |
8446f1d3 | 99 | |
a3b13c23 | 100 | now = get_timestamp(this_cpu); |
966812dc | 101 | |
6687a97d | 102 | /* Wake up the high-prio watchdog task every second: */ |
966812dc | 103 | if (now > (touch_timestamp + 1)) |
6687a97d IM |
104 | wake_up_process(per_cpu(watchdog_task, this_cpu)); |
105 | ||
106 | /* Warn about unreasonable 10+ seconds delays: */ | |
43581a10 IM |
107 | if (now <= (touch_timestamp + 10)) |
108 | return; | |
8446f1d3 | 109 | |
43581a10 IM |
110 | per_cpu(print_timestamp, this_cpu) = touch_timestamp; |
111 | ||
112 | spin_lock(&print_lock); | |
113 | printk(KERN_ERR "BUG: soft lockup detected on CPU#%d!\n", this_cpu); | |
114 | if (regs) | |
115 | show_regs(regs); | |
116 | else | |
6687a97d | 117 | dump_stack(); |
43581a10 | 118 | spin_unlock(&print_lock); |
8446f1d3 IM |
119 | } |
120 | ||
121 | /* | |
122 | * The watchdog thread - runs every second and touches the timestamp. | |
123 | */ | |
a5f2ce3c | 124 | static int watchdog(void *__bind_cpu) |
8446f1d3 | 125 | { |
02fb6149 | 126 | struct sched_param param = { .sched_priority = MAX_RT_PRIO-1 }; |
8446f1d3 IM |
127 | |
128 | sched_setscheduler(current, SCHED_FIFO, ¶m); | |
8446f1d3 | 129 | |
966812dc JF |
130 | /* initialize timestamp */ |
131 | touch_softlockup_watchdog(); | |
132 | ||
8446f1d3 | 133 | /* |
6687a97d IM |
134 | * Run briefly once per second to reset the softlockup timestamp. |
135 | * If this gets delayed for more than 10 seconds then the | |
136 | * debug-printout triggers in softlockup_tick(). | |
8446f1d3 IM |
137 | */ |
138 | while (!kthread_should_stop()) { | |
6687a97d | 139 | set_current_state(TASK_INTERRUPTIBLE); |
8446f1d3 | 140 | touch_softlockup_watchdog(); |
6687a97d | 141 | schedule(); |
8446f1d3 | 142 | } |
8446f1d3 IM |
143 | |
144 | return 0; | |
145 | } | |
146 | ||
147 | /* | |
148 | * Create/destroy watchdog threads as CPUs come and go: | |
149 | */ | |
8c78f307 | 150 | static int __cpuinit |
8446f1d3 IM |
151 | cpu_callback(struct notifier_block *nfb, unsigned long action, void *hcpu) |
152 | { | |
153 | int hotcpu = (unsigned long)hcpu; | |
154 | struct task_struct *p; | |
155 | ||
156 | switch (action) { | |
157 | case CPU_UP_PREPARE: | |
8bb78442 | 158 | case CPU_UP_PREPARE_FROZEN: |
8446f1d3 IM |
159 | BUG_ON(per_cpu(watchdog_task, hotcpu)); |
160 | p = kthread_create(watchdog, hcpu, "watchdog/%d", hotcpu); | |
161 | if (IS_ERR(p)) { | |
a5f2ce3c | 162 | printk(KERN_ERR "watchdog for %i failed\n", hotcpu); |
8446f1d3 IM |
163 | return NOTIFY_BAD; |
164 | } | |
a5f2ce3c IM |
165 | per_cpu(touch_timestamp, hotcpu) = 0; |
166 | per_cpu(watchdog_task, hotcpu) = p; | |
8446f1d3 | 167 | kthread_bind(p, hotcpu); |
a5f2ce3c | 168 | break; |
8446f1d3 | 169 | case CPU_ONLINE: |
8bb78442 | 170 | case CPU_ONLINE_FROZEN: |
8446f1d3 IM |
171 | wake_up_process(per_cpu(watchdog_task, hotcpu)); |
172 | break; | |
173 | #ifdef CONFIG_HOTPLUG_CPU | |
174 | case CPU_UP_CANCELED: | |
8bb78442 | 175 | case CPU_UP_CANCELED_FROZEN: |
fc75cdfa HC |
176 | if (!per_cpu(watchdog_task, hotcpu)) |
177 | break; | |
8446f1d3 | 178 | /* Unbind so it can run. Fall thru. */ |
a4c4af7c HC |
179 | kthread_bind(per_cpu(watchdog_task, hotcpu), |
180 | any_online_cpu(cpu_online_map)); | |
8446f1d3 | 181 | case CPU_DEAD: |
8bb78442 | 182 | case CPU_DEAD_FROZEN: |
8446f1d3 IM |
183 | p = per_cpu(watchdog_task, hotcpu); |
184 | per_cpu(watchdog_task, hotcpu) = NULL; | |
185 | kthread_stop(p); | |
186 | break; | |
187 | #endif /* CONFIG_HOTPLUG_CPU */ | |
a5f2ce3c | 188 | } |
8446f1d3 IM |
189 | return NOTIFY_OK; |
190 | } | |
191 | ||
8c78f307 | 192 | static struct notifier_block __cpuinitdata cpu_nfb = { |
8446f1d3 IM |
193 | .notifier_call = cpu_callback |
194 | }; | |
195 | ||
196 | __init void spawn_softlockup_task(void) | |
197 | { | |
198 | void *cpu = (void *)(long)smp_processor_id(); | |
07dccf33 | 199 | int err = cpu_callback(&cpu_nfb, CPU_UP_PREPARE, cpu); |
8446f1d3 | 200 | |
07dccf33 | 201 | BUG_ON(err == NOTIFY_BAD); |
8446f1d3 IM |
202 | cpu_callback(&cpu_nfb, CPU_ONLINE, cpu); |
203 | register_cpu_notifier(&cpu_nfb); | |
204 | ||
e041c683 | 205 | atomic_notifier_chain_register(&panic_notifier_list, &panic_block); |
8446f1d3 | 206 | } |