Ingo Molnar | 8446f1d | 2005-09-06 15:16:27 -0700 | [diff] [blame] | 1 | /* |
| 2 | * Detect Soft Lockups |
| 3 | * |
Ingo Molnar | 6687a97 | 2006-03-24 03:18:41 -0800 | [diff] [blame] | 4 | * started by Ingo Molnar, Copyright (C) 2005, 2006 Red Hat, Inc. |
Ingo Molnar | 8446f1d | 2005-09-06 15:16:27 -0700 | [diff] [blame] | 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 | */ |
Ingo Molnar | 8446f1d | 2005-09-06 15:16:27 -0700 | [diff] [blame] | 9 | #include <linux/mm.h> |
| 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> |
| 16 | |
| 17 | static DEFINE_SPINLOCK(print_lock); |
| 18 | |
Ingo Molnar | 6687a97 | 2006-03-24 03:18:41 -0800 | [diff] [blame] | 19 | static DEFINE_PER_CPU(unsigned long, touch_timestamp); |
| 20 | static DEFINE_PER_CPU(unsigned long, print_timestamp); |
Ingo Molnar | 8446f1d | 2005-09-06 15:16:27 -0700 | [diff] [blame] | 21 | static DEFINE_PER_CPU(struct task_struct *, watchdog_task); |
| 22 | |
| 23 | static int did_panic = 0; |
Ingo Molnar | 6687a97 | 2006-03-24 03:18:41 -0800 | [diff] [blame] | 24 | |
| 25 | static int |
| 26 | softlock_panic(struct notifier_block *this, unsigned long event, void *ptr) |
Ingo Molnar | 8446f1d | 2005-09-06 15:16:27 -0700 | [diff] [blame] | 27 | { |
| 28 | did_panic = 1; |
| 29 | |
| 30 | return NOTIFY_DONE; |
| 31 | } |
| 32 | |
| 33 | static struct notifier_block panic_block = { |
| 34 | .notifier_call = softlock_panic, |
| 35 | }; |
| 36 | |
| 37 | void touch_softlockup_watchdog(void) |
| 38 | { |
Paul Mackerras | bfe5d83 | 2006-06-25 05:47:14 -0700 | [diff] [blame] | 39 | __raw_get_cpu_var(touch_timestamp) = jiffies; |
Ingo Molnar | 8446f1d | 2005-09-06 15:16:27 -0700 | [diff] [blame] | 40 | } |
| 41 | EXPORT_SYMBOL(touch_softlockup_watchdog); |
| 42 | |
| 43 | /* |
| 44 | * This callback runs from the timer interrupt, and checks |
| 45 | * whether the watchdog thread has hung or not: |
| 46 | */ |
Ingo Molnar | 6687a97 | 2006-03-24 03:18:41 -0800 | [diff] [blame] | 47 | void softlockup_tick(void) |
Ingo Molnar | 8446f1d | 2005-09-06 15:16:27 -0700 | [diff] [blame] | 48 | { |
| 49 | int this_cpu = smp_processor_id(); |
Ingo Molnar | 6687a97 | 2006-03-24 03:18:41 -0800 | [diff] [blame] | 50 | unsigned long touch_timestamp = per_cpu(touch_timestamp, this_cpu); |
Ingo Molnar | 8446f1d | 2005-09-06 15:16:27 -0700 | [diff] [blame] | 51 | |
Ingo Molnar | 6687a97 | 2006-03-24 03:18:41 -0800 | [diff] [blame] | 52 | /* prevent double reports: */ |
| 53 | if (per_cpu(print_timestamp, this_cpu) == touch_timestamp || |
| 54 | did_panic || |
| 55 | !per_cpu(watchdog_task, this_cpu)) |
Ingo Molnar | 8446f1d | 2005-09-06 15:16:27 -0700 | [diff] [blame] | 56 | return; |
| 57 | |
Ingo Molnar | 6687a97 | 2006-03-24 03:18:41 -0800 | [diff] [blame] | 58 | /* do not print during early bootup: */ |
| 59 | if (unlikely(system_state != SYSTEM_RUNNING)) { |
| 60 | touch_softlockup_watchdog(); |
Ingo Molnar | 8446f1d | 2005-09-06 15:16:27 -0700 | [diff] [blame] | 61 | return; |
Ingo Molnar | 6687a97 | 2006-03-24 03:18:41 -0800 | [diff] [blame] | 62 | } |
Ingo Molnar | 8446f1d | 2005-09-06 15:16:27 -0700 | [diff] [blame] | 63 | |
Ingo Molnar | 6687a97 | 2006-03-24 03:18:41 -0800 | [diff] [blame] | 64 | /* Wake up the high-prio watchdog task every second: */ |
| 65 | if (time_after(jiffies, touch_timestamp + HZ)) |
| 66 | wake_up_process(per_cpu(watchdog_task, this_cpu)); |
| 67 | |
| 68 | /* Warn about unreasonable 10+ seconds delays: */ |
| 69 | if (time_after(jiffies, touch_timestamp + 10*HZ)) { |
| 70 | per_cpu(print_timestamp, this_cpu) = touch_timestamp; |
Ingo Molnar | 8446f1d | 2005-09-06 15:16:27 -0700 | [diff] [blame] | 71 | |
| 72 | spin_lock(&print_lock); |
| 73 | printk(KERN_ERR "BUG: soft lockup detected on CPU#%d!\n", |
| 74 | this_cpu); |
Ingo Molnar | 6687a97 | 2006-03-24 03:18:41 -0800 | [diff] [blame] | 75 | dump_stack(); |
Ingo Molnar | 8446f1d | 2005-09-06 15:16:27 -0700 | [diff] [blame] | 76 | spin_unlock(&print_lock); |
| 77 | } |
| 78 | } |
| 79 | |
| 80 | /* |
| 81 | * The watchdog thread - runs every second and touches the timestamp. |
| 82 | */ |
| 83 | static int watchdog(void * __bind_cpu) |
| 84 | { |
| 85 | struct sched_param param = { .sched_priority = 99 }; |
Ingo Molnar | 8446f1d | 2005-09-06 15:16:27 -0700 | [diff] [blame] | 86 | |
| 87 | sched_setscheduler(current, SCHED_FIFO, ¶m); |
| 88 | current->flags |= PF_NOFREEZE; |
| 89 | |
Ingo Molnar | 8446f1d | 2005-09-06 15:16:27 -0700 | [diff] [blame] | 90 | /* |
Ingo Molnar | 6687a97 | 2006-03-24 03:18:41 -0800 | [diff] [blame] | 91 | * Run briefly once per second to reset the softlockup timestamp. |
| 92 | * If this gets delayed for more than 10 seconds then the |
| 93 | * debug-printout triggers in softlockup_tick(). |
Ingo Molnar | 8446f1d | 2005-09-06 15:16:27 -0700 | [diff] [blame] | 94 | */ |
| 95 | while (!kthread_should_stop()) { |
Ingo Molnar | 6687a97 | 2006-03-24 03:18:41 -0800 | [diff] [blame] | 96 | set_current_state(TASK_INTERRUPTIBLE); |
Ingo Molnar | 8446f1d | 2005-09-06 15:16:27 -0700 | [diff] [blame] | 97 | touch_softlockup_watchdog(); |
Ingo Molnar | 6687a97 | 2006-03-24 03:18:41 -0800 | [diff] [blame] | 98 | schedule(); |
Ingo Molnar | 8446f1d | 2005-09-06 15:16:27 -0700 | [diff] [blame] | 99 | } |
Ingo Molnar | 8446f1d | 2005-09-06 15:16:27 -0700 | [diff] [blame] | 100 | |
| 101 | return 0; |
| 102 | } |
| 103 | |
| 104 | /* |
| 105 | * Create/destroy watchdog threads as CPUs come and go: |
| 106 | */ |
Chandra Seetharaman | 8c78f30 | 2006-07-30 03:03:35 -0700 | [diff] [blame] | 107 | static int __cpuinit |
Ingo Molnar | 8446f1d | 2005-09-06 15:16:27 -0700 | [diff] [blame] | 108 | cpu_callback(struct notifier_block *nfb, unsigned long action, void *hcpu) |
| 109 | { |
| 110 | int hotcpu = (unsigned long)hcpu; |
| 111 | struct task_struct *p; |
| 112 | |
| 113 | switch (action) { |
| 114 | case CPU_UP_PREPARE: |
| 115 | BUG_ON(per_cpu(watchdog_task, hotcpu)); |
| 116 | p = kthread_create(watchdog, hcpu, "watchdog/%d", hotcpu); |
| 117 | if (IS_ERR(p)) { |
| 118 | printk("watchdog for %i failed\n", hotcpu); |
| 119 | return NOTIFY_BAD; |
| 120 | } |
Andrew Morton | 185ae6d | 2006-03-25 03:06:32 -0800 | [diff] [blame] | 121 | per_cpu(touch_timestamp, hotcpu) = jiffies; |
Ingo Molnar | 8446f1d | 2005-09-06 15:16:27 -0700 | [diff] [blame] | 122 | per_cpu(watchdog_task, hotcpu) = p; |
| 123 | kthread_bind(p, hotcpu); |
| 124 | break; |
| 125 | case CPU_ONLINE: |
Ingo Molnar | 8446f1d | 2005-09-06 15:16:27 -0700 | [diff] [blame] | 126 | wake_up_process(per_cpu(watchdog_task, hotcpu)); |
| 127 | break; |
| 128 | #ifdef CONFIG_HOTPLUG_CPU |
| 129 | case CPU_UP_CANCELED: |
Heiko Carstens | fc75cdf | 2006-06-25 05:49:10 -0700 | [diff] [blame] | 130 | if (!per_cpu(watchdog_task, hotcpu)) |
| 131 | break; |
Ingo Molnar | 8446f1d | 2005-09-06 15:16:27 -0700 | [diff] [blame] | 132 | /* Unbind so it can run. Fall thru. */ |
Heiko Carstens | a4c4af7 | 2005-11-07 00:58:38 -0800 | [diff] [blame] | 133 | kthread_bind(per_cpu(watchdog_task, hotcpu), |
| 134 | any_online_cpu(cpu_online_map)); |
Ingo Molnar | 8446f1d | 2005-09-06 15:16:27 -0700 | [diff] [blame] | 135 | case CPU_DEAD: |
| 136 | p = per_cpu(watchdog_task, hotcpu); |
| 137 | per_cpu(watchdog_task, hotcpu) = NULL; |
| 138 | kthread_stop(p); |
| 139 | break; |
| 140 | #endif /* CONFIG_HOTPLUG_CPU */ |
| 141 | } |
| 142 | return NOTIFY_OK; |
| 143 | } |
| 144 | |
Chandra Seetharaman | 8c78f30 | 2006-07-30 03:03:35 -0700 | [diff] [blame] | 145 | static struct notifier_block __cpuinitdata cpu_nfb = { |
Ingo Molnar | 8446f1d | 2005-09-06 15:16:27 -0700 | [diff] [blame] | 146 | .notifier_call = cpu_callback |
| 147 | }; |
| 148 | |
| 149 | __init void spawn_softlockup_task(void) |
| 150 | { |
| 151 | void *cpu = (void *)(long)smp_processor_id(); |
Akinobu Mita | 07dccf3 | 2006-09-29 02:00:22 -0700 | [diff] [blame] | 152 | int err = cpu_callback(&cpu_nfb, CPU_UP_PREPARE, cpu); |
Ingo Molnar | 8446f1d | 2005-09-06 15:16:27 -0700 | [diff] [blame] | 153 | |
Akinobu Mita | 07dccf3 | 2006-09-29 02:00:22 -0700 | [diff] [blame] | 154 | BUG_ON(err == NOTIFY_BAD); |
Ingo Molnar | 8446f1d | 2005-09-06 15:16:27 -0700 | [diff] [blame] | 155 | cpu_callback(&cpu_nfb, CPU_ONLINE, cpu); |
| 156 | register_cpu_notifier(&cpu_nfb); |
| 157 | |
Alan Stern | e041c68 | 2006-03-27 01:16:30 -0800 | [diff] [blame] | 158 | atomic_notifier_chain_register(&panic_notifier_list, &panic_block); |
Ingo Molnar | 8446f1d | 2005-09-06 15:16:27 -0700 | [diff] [blame] | 159 | } |