]>
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> | |
82a1fcb9 | 11 | #include <linux/nmi.h> |
8446f1d3 IM |
12 | #include <linux/init.h> |
13 | #include <linux/delay.h> | |
83144186 | 14 | #include <linux/freezer.h> |
8446f1d3 | 15 | #include <linux/kthread.h> |
8d5be7f4 | 16 | #include <linux/lockdep.h> |
8446f1d3 IM |
17 | #include <linux/notifier.h> |
18 | #include <linux/module.h> | |
baf48f65 | 19 | #include <linux/sysctl.h> |
8446f1d3 | 20 | |
43581a10 IM |
21 | #include <asm/irq_regs.h> |
22 | ||
8446f1d3 IM |
23 | static DEFINE_SPINLOCK(print_lock); |
24 | ||
6687a97d IM |
25 | static DEFINE_PER_CPU(unsigned long, touch_timestamp); |
26 | static DEFINE_PER_CPU(unsigned long, print_timestamp); | |
8446f1d3 IM |
27 | static DEFINE_PER_CPU(struct task_struct *, watchdog_task); |
28 | ||
90739081 | 29 | static int __read_mostly did_panic; |
9383d967 | 30 | int __read_mostly softlockup_thresh = 60; |
6687a97d | 31 | |
9c44bc03 IM |
32 | /* |
33 | * Should we panic (and reboot, if panic_timeout= is set) when a | |
34 | * soft-lockup occurs: | |
35 | */ | |
36 | unsigned int __read_mostly softlockup_panic = | |
37 | CONFIG_BOOTPARAM_SOFTLOCKUP_PANIC_VALUE; | |
38 | ||
39 | static int __init softlockup_panic_setup(char *str) | |
40 | { | |
41 | softlockup_panic = simple_strtoul(str, NULL, 0); | |
42 | ||
43 | return 1; | |
44 | } | |
45 | __setup("softlockup_panic=", softlockup_panic_setup); | |
46 | ||
6687a97d IM |
47 | static int |
48 | softlock_panic(struct notifier_block *this, unsigned long event, void *ptr) | |
8446f1d3 IM |
49 | { |
50 | did_panic = 1; | |
51 | ||
52 | return NOTIFY_DONE; | |
53 | } | |
54 | ||
55 | static struct notifier_block panic_block = { | |
56 | .notifier_call = softlock_panic, | |
57 | }; | |
58 | ||
966812dc JF |
59 | /* |
60 | * Returns seconds, approximately. We don't need nanosecond | |
61 | * resolution, and we don't need to waste time with a big divide when | |
62 | * 2^30ns == 1.074s. | |
63 | */ | |
a3b13c23 | 64 | static unsigned long get_timestamp(int this_cpu) |
966812dc | 65 | { |
82a1fcb9 | 66 | return cpu_clock(this_cpu) >> 30LL; /* 2^30 ~= 10^9 */ |
966812dc JF |
67 | } |
68 | ||
8c2238ea | 69 | static void __touch_softlockup_watchdog(void) |
8446f1d3 | 70 | { |
a3b13c23 IM |
71 | int this_cpu = raw_smp_processor_id(); |
72 | ||
73 | __raw_get_cpu_var(touch_timestamp) = get_timestamp(this_cpu); | |
8446f1d3 | 74 | } |
8c2238ea JW |
75 | |
76 | void touch_softlockup_watchdog(void) | |
77 | { | |
78 | __raw_get_cpu_var(touch_timestamp) = 0; | |
79 | } | |
8446f1d3 IM |
80 | EXPORT_SYMBOL(touch_softlockup_watchdog); |
81 | ||
04c9167f JF |
82 | void touch_all_softlockup_watchdogs(void) |
83 | { | |
84 | int cpu; | |
85 | ||
86 | /* Cause each CPU to re-update its timestamp rather than complain */ | |
87 | for_each_online_cpu(cpu) | |
88 | per_cpu(touch_timestamp, cpu) = 0; | |
89 | } | |
90 | EXPORT_SYMBOL(touch_all_softlockup_watchdogs); | |
91 | ||
baf48f65 | 92 | int proc_dosoftlockup_thresh(struct ctl_table *table, int write, |
8d65af78 | 93 | void __user *buffer, |
baf48f65 MSB |
94 | size_t *lenp, loff_t *ppos) |
95 | { | |
96 | touch_all_softlockup_watchdogs(); | |
8d65af78 | 97 | return proc_dointvec_minmax(table, write, buffer, lenp, ppos); |
baf48f65 MSB |
98 | } |
99 | ||
8446f1d3 IM |
100 | /* |
101 | * This callback runs from the timer interrupt, and checks | |
102 | * whether the watchdog thread has hung or not: | |
103 | */ | |
6687a97d | 104 | void softlockup_tick(void) |
8446f1d3 IM |
105 | { |
106 | int this_cpu = smp_processor_id(); | |
6687a97d | 107 | unsigned long touch_timestamp = per_cpu(touch_timestamp, this_cpu); |
966812dc | 108 | unsigned long print_timestamp; |
43581a10 | 109 | struct pt_regs *regs = get_irq_regs(); |
966812dc | 110 | unsigned long now; |
8446f1d3 | 111 | |
9383d967 DS |
112 | /* Is detection switched off? */ |
113 | if (!per_cpu(watchdog_task, this_cpu) || softlockup_thresh <= 0) { | |
114 | /* Be sure we don't false trigger if switched back on */ | |
115 | if (touch_timestamp) | |
116 | per_cpu(touch_timestamp, this_cpu) = 0; | |
117 | return; | |
118 | } | |
119 | ||
04c9167f | 120 | if (touch_timestamp == 0) { |
8c2238ea | 121 | __touch_softlockup_watchdog(); |
966812dc | 122 | return; |
04c9167f | 123 | } |
966812dc JF |
124 | |
125 | print_timestamp = per_cpu(print_timestamp, this_cpu); | |
126 | ||
127 | /* report at most once a second */ | |
3e2f69fd | 128 | if (print_timestamp == touch_timestamp || did_panic) |
8446f1d3 IM |
129 | return; |
130 | ||
6687a97d IM |
131 | /* do not print during early bootup: */ |
132 | if (unlikely(system_state != SYSTEM_RUNNING)) { | |
8c2238ea | 133 | __touch_softlockup_watchdog(); |
8446f1d3 | 134 | return; |
6687a97d | 135 | } |
8446f1d3 | 136 | |
a3b13c23 | 137 | now = get_timestamp(this_cpu); |
966812dc | 138 | |
dd7a1e56 JW |
139 | /* |
140 | * Wake up the high-prio watchdog task twice per | |
141 | * threshold timespan. | |
142 | */ | |
143 | if (now > touch_timestamp + softlockup_thresh/2) | |
ed50d6cb PZ |
144 | wake_up_process(per_cpu(watchdog_task, this_cpu)); |
145 | ||
82a1fcb9 | 146 | /* Warn about unreasonable delays: */ |
c4f3b63f | 147 | if (now <= (touch_timestamp + softlockup_thresh)) |
43581a10 | 148 | return; |
8446f1d3 | 149 | |
43581a10 IM |
150 | per_cpu(print_timestamp, this_cpu) = touch_timestamp; |
151 | ||
152 | spin_lock(&print_lock); | |
c4f3b63f RT |
153 | printk(KERN_ERR "BUG: soft lockup - CPU#%d stuck for %lus! [%s:%d]\n", |
154 | this_cpu, now - touch_timestamp, | |
ba25f9dc | 155 | current->comm, task_pid_nr(current)); |
688c9175 | 156 | print_modules(); |
8d5be7f4 | 157 | print_irqtrace_events(current); |
43581a10 IM |
158 | if (regs) |
159 | show_regs(regs); | |
160 | else | |
6687a97d | 161 | dump_stack(); |
43581a10 | 162 | spin_unlock(&print_lock); |
9c44bc03 IM |
163 | |
164 | if (softlockup_panic) | |
165 | panic("softlockup: hung tasks"); | |
8446f1d3 IM |
166 | } |
167 | ||
168 | /* | |
169 | * The watchdog thread - runs every second and touches the timestamp. | |
170 | */ | |
a5f2ce3c | 171 | static int watchdog(void *__bind_cpu) |
8446f1d3 | 172 | { |
02fb6149 | 173 | struct sched_param param = { .sched_priority = MAX_RT_PRIO-1 }; |
8446f1d3 IM |
174 | |
175 | sched_setscheduler(current, SCHED_FIFO, ¶m); | |
8446f1d3 | 176 | |
966812dc | 177 | /* initialize timestamp */ |
8c2238ea | 178 | __touch_softlockup_watchdog(); |
966812dc | 179 | |
7be2a03e | 180 | set_current_state(TASK_INTERRUPTIBLE); |
8446f1d3 | 181 | /* |
6687a97d | 182 | * Run briefly once per second to reset the softlockup timestamp. |
82a1fcb9 | 183 | * If this gets delayed for more than 60 seconds then the |
6687a97d | 184 | * debug-printout triggers in softlockup_tick(). |
8446f1d3 IM |
185 | */ |
186 | while (!kthread_should_stop()) { | |
8c2238ea | 187 | __touch_softlockup_watchdog(); |
ed50d6cb PZ |
188 | schedule(); |
189 | ||
190 | if (kthread_should_stop()) | |
191 | break; | |
82a1fcb9 | 192 | |
7be2a03e | 193 | set_current_state(TASK_INTERRUPTIBLE); |
8446f1d3 | 194 | } |
7be2a03e | 195 | __set_current_state(TASK_RUNNING); |
8446f1d3 IM |
196 | |
197 | return 0; | |
198 | } | |
199 | ||
200 | /* | |
201 | * Create/destroy watchdog threads as CPUs come and go: | |
202 | */ | |
8c78f307 | 203 | static int __cpuinit |
8446f1d3 IM |
204 | cpu_callback(struct notifier_block *nfb, unsigned long action, void *hcpu) |
205 | { | |
206 | int hotcpu = (unsigned long)hcpu; | |
207 | struct task_struct *p; | |
208 | ||
209 | switch (action) { | |
210 | case CPU_UP_PREPARE: | |
8bb78442 | 211 | case CPU_UP_PREPARE_FROZEN: |
8446f1d3 IM |
212 | BUG_ON(per_cpu(watchdog_task, hotcpu)); |
213 | p = kthread_create(watchdog, hcpu, "watchdog/%d", hotcpu); | |
214 | if (IS_ERR(p)) { | |
a5f2ce3c | 215 | printk(KERN_ERR "watchdog for %i failed\n", hotcpu); |
8446f1d3 IM |
216 | return NOTIFY_BAD; |
217 | } | |
a5f2ce3c IM |
218 | per_cpu(touch_timestamp, hotcpu) = 0; |
219 | per_cpu(watchdog_task, hotcpu) = p; | |
8446f1d3 | 220 | kthread_bind(p, hotcpu); |
a5f2ce3c | 221 | break; |
8446f1d3 | 222 | case CPU_ONLINE: |
8bb78442 | 223 | case CPU_ONLINE_FROZEN: |
8446f1d3 IM |
224 | wake_up_process(per_cpu(watchdog_task, hotcpu)); |
225 | break; | |
226 | #ifdef CONFIG_HOTPLUG_CPU | |
ed50d6cb PZ |
227 | case CPU_UP_CANCELED: |
228 | case CPU_UP_CANCELED_FROZEN: | |
229 | if (!per_cpu(watchdog_task, hotcpu)) | |
230 | break; | |
231 | /* Unbind so it can run. Fall thru. */ | |
232 | kthread_bind(per_cpu(watchdog_task, hotcpu), | |
f1fc057c | 233 | cpumask_any(cpu_online_mask)); |
8446f1d3 | 234 | case CPU_DEAD: |
8bb78442 | 235 | case CPU_DEAD_FROZEN: |
8446f1d3 IM |
236 | p = per_cpu(watchdog_task, hotcpu); |
237 | per_cpu(watchdog_task, hotcpu) = NULL; | |
238 | kthread_stop(p); | |
239 | break; | |
240 | #endif /* CONFIG_HOTPLUG_CPU */ | |
a5f2ce3c | 241 | } |
8446f1d3 IM |
242 | return NOTIFY_OK; |
243 | } | |
244 | ||
8c78f307 | 245 | static struct notifier_block __cpuinitdata cpu_nfb = { |
8446f1d3 IM |
246 | .notifier_call = cpu_callback |
247 | }; | |
248 | ||
7babe8db EGM |
249 | static int __initdata nosoftlockup; |
250 | ||
251 | static int __init nosoftlockup_setup(char *str) | |
252 | { | |
253 | nosoftlockup = 1; | |
254 | return 1; | |
255 | } | |
256 | __setup("nosoftlockup", nosoftlockup_setup); | |
257 | ||
258 | static int __init spawn_softlockup_task(void) | |
8446f1d3 IM |
259 | { |
260 | void *cpu = (void *)(long)smp_processor_id(); | |
7babe8db | 261 | int err; |
8446f1d3 | 262 | |
7babe8db EGM |
263 | if (nosoftlockup) |
264 | return 0; | |
265 | ||
266 | err = cpu_callback(&cpu_nfb, CPU_UP_PREPARE, cpu); | |
267 | if (err == NOTIFY_BAD) { | |
268 | BUG(); | |
269 | return 1; | |
270 | } | |
8446f1d3 IM |
271 | cpu_callback(&cpu_nfb, CPU_ONLINE, cpu); |
272 | register_cpu_notifier(&cpu_nfb); | |
273 | ||
e041c683 | 274 | atomic_notifier_chain_register(&panic_notifier_list, &panic_block); |
7babe8db EGM |
275 | |
276 | return 0; | |
8446f1d3 | 277 | } |
7babe8db | 278 | early_initcall(spawn_softlockup_task); |