158687acbSDon Zickus /* 258687acbSDon Zickus * Detect hard and soft lockups on a system 358687acbSDon Zickus * 458687acbSDon Zickus * started by Don Zickus, Copyright (C) 2010 Red Hat, Inc. 558687acbSDon Zickus * 658687acbSDon Zickus * this code detects hard lockups: incidents in where on a CPU 758687acbSDon Zickus * the kernel does not respond to anything except NMI. 858687acbSDon Zickus * 958687acbSDon Zickus * Note: Most of this code is borrowed heavily from softlockup.c, 1058687acbSDon Zickus * so thanks to Ingo for the initial implementation. 1158687acbSDon Zickus * Some chunks also taken from arch/x86/kernel/apic/nmi.c, thanks 1258687acbSDon Zickus * to those contributors as well. 1358687acbSDon Zickus */ 1458687acbSDon Zickus 1558687acbSDon Zickus #include <linux/mm.h> 1658687acbSDon Zickus #include <linux/cpu.h> 1758687acbSDon Zickus #include <linux/nmi.h> 1858687acbSDon Zickus #include <linux/init.h> 1958687acbSDon Zickus #include <linux/delay.h> 2058687acbSDon Zickus #include <linux/freezer.h> 2158687acbSDon Zickus #include <linux/kthread.h> 2258687acbSDon Zickus #include <linux/lockdep.h> 2358687acbSDon Zickus #include <linux/notifier.h> 2458687acbSDon Zickus #include <linux/module.h> 2558687acbSDon Zickus #include <linux/sysctl.h> 2658687acbSDon Zickus 2758687acbSDon Zickus #include <asm/irq_regs.h> 2858687acbSDon Zickus #include <linux/perf_event.h> 2958687acbSDon Zickus 3058687acbSDon Zickus int watchdog_enabled; 3158687acbSDon Zickus int __read_mostly softlockup_thresh = 60; 3258687acbSDon Zickus 3358687acbSDon Zickus static DEFINE_PER_CPU(unsigned long, watchdog_touch_ts); 3458687acbSDon Zickus static DEFINE_PER_CPU(struct task_struct *, softlockup_watchdog); 3558687acbSDon Zickus static DEFINE_PER_CPU(struct hrtimer, watchdog_hrtimer); 3658687acbSDon Zickus static DEFINE_PER_CPU(bool, softlockup_touch_sync); 3758687acbSDon Zickus static DEFINE_PER_CPU(bool, soft_watchdog_warn); 3823637d47SFrederic Weisbecker #ifdef CONFIG_HARDLOCKUP_DETECTOR 39cafcd80dSDon Zickus static DEFINE_PER_CPU(bool, hard_watchdog_warn); 40cafcd80dSDon Zickus static DEFINE_PER_CPU(bool, watchdog_nmi_touch); 4158687acbSDon Zickus static DEFINE_PER_CPU(unsigned long, hrtimer_interrupts); 4258687acbSDon Zickus static DEFINE_PER_CPU(unsigned long, hrtimer_interrupts_saved); 4358687acbSDon Zickus static DEFINE_PER_CPU(struct perf_event *, watchdog_ev); 4458687acbSDon Zickus #endif 4558687acbSDon Zickus 4658687acbSDon Zickus static int __initdata no_watchdog; 4758687acbSDon Zickus 4858687acbSDon Zickus 4958687acbSDon Zickus /* boot commands */ 5058687acbSDon Zickus /* 5158687acbSDon Zickus * Should we panic when a soft-lockup or hard-lockup occurs: 5258687acbSDon Zickus */ 5323637d47SFrederic Weisbecker #ifdef CONFIG_HARDLOCKUP_DETECTOR 5458687acbSDon Zickus static int hardlockup_panic; 5558687acbSDon Zickus 5658687acbSDon Zickus static int __init hardlockup_panic_setup(char *str) 5758687acbSDon Zickus { 5858687acbSDon Zickus if (!strncmp(str, "panic", 5)) 5958687acbSDon Zickus hardlockup_panic = 1; 6058687acbSDon Zickus return 1; 6158687acbSDon Zickus } 6258687acbSDon Zickus __setup("nmi_watchdog=", hardlockup_panic_setup); 6358687acbSDon Zickus #endif 6458687acbSDon Zickus 6558687acbSDon Zickus unsigned int __read_mostly softlockup_panic = 6658687acbSDon Zickus CONFIG_BOOTPARAM_SOFTLOCKUP_PANIC_VALUE; 6758687acbSDon Zickus 6858687acbSDon Zickus static int __init softlockup_panic_setup(char *str) 6958687acbSDon Zickus { 7058687acbSDon Zickus softlockup_panic = simple_strtoul(str, NULL, 0); 7158687acbSDon Zickus 7258687acbSDon Zickus return 1; 7358687acbSDon Zickus } 7458687acbSDon Zickus __setup("softlockup_panic=", softlockup_panic_setup); 7558687acbSDon Zickus 7658687acbSDon Zickus static int __init nowatchdog_setup(char *str) 7758687acbSDon Zickus { 7858687acbSDon Zickus no_watchdog = 1; 7958687acbSDon Zickus return 1; 8058687acbSDon Zickus } 8158687acbSDon Zickus __setup("nowatchdog", nowatchdog_setup); 8258687acbSDon Zickus 8358687acbSDon Zickus /* deprecated */ 8458687acbSDon Zickus static int __init nosoftlockup_setup(char *str) 8558687acbSDon Zickus { 8658687acbSDon Zickus no_watchdog = 1; 8758687acbSDon Zickus return 1; 8858687acbSDon Zickus } 8958687acbSDon Zickus __setup("nosoftlockup", nosoftlockup_setup); 9058687acbSDon Zickus /* */ 9158687acbSDon Zickus 9258687acbSDon Zickus 9358687acbSDon Zickus /* 9458687acbSDon Zickus * Returns seconds, approximately. We don't need nanosecond 9558687acbSDon Zickus * resolution, and we don't need to waste time with a big divide when 9658687acbSDon Zickus * 2^30ns == 1.074s. 9758687acbSDon Zickus */ 9858687acbSDon Zickus static unsigned long get_timestamp(int this_cpu) 9958687acbSDon Zickus { 10058687acbSDon Zickus return cpu_clock(this_cpu) >> 30LL; /* 2^30 ~= 10^9 */ 10158687acbSDon Zickus } 10258687acbSDon Zickus 10358687acbSDon Zickus static unsigned long get_sample_period(void) 10458687acbSDon Zickus { 10558687acbSDon Zickus /* 10658687acbSDon Zickus * convert softlockup_thresh from seconds to ns 10758687acbSDon Zickus * the divide by 5 is to give hrtimer 5 chances to 10858687acbSDon Zickus * increment before the hardlockup detector generates 10958687acbSDon Zickus * a warning 11058687acbSDon Zickus */ 11158687acbSDon Zickus return softlockup_thresh / 5 * NSEC_PER_SEC; 11258687acbSDon Zickus } 11358687acbSDon Zickus 11458687acbSDon Zickus /* Commands for resetting the watchdog */ 11558687acbSDon Zickus static void __touch_watchdog(void) 11658687acbSDon Zickus { 11726e09c6eSDon Zickus int this_cpu = smp_processor_id(); 11858687acbSDon Zickus 11958687acbSDon Zickus __get_cpu_var(watchdog_touch_ts) = get_timestamp(this_cpu); 12058687acbSDon Zickus } 12158687acbSDon Zickus 122332fbdbcSDon Zickus void touch_softlockup_watchdog(void) 12358687acbSDon Zickus { 12468d3f1d8SDon Zickus __raw_get_cpu_var(watchdog_touch_ts) = 0; 12558687acbSDon Zickus } 1260167c781SIngo Molnar EXPORT_SYMBOL(touch_softlockup_watchdog); 12758687acbSDon Zickus 128332fbdbcSDon Zickus void touch_all_softlockup_watchdogs(void) 12958687acbSDon Zickus { 13058687acbSDon Zickus int cpu; 13158687acbSDon Zickus 13258687acbSDon Zickus /* 13358687acbSDon Zickus * this is done lockless 13458687acbSDon Zickus * do we care if a 0 races with a timestamp? 13558687acbSDon Zickus * all it means is the softlock check starts one cycle later 13658687acbSDon Zickus */ 13758687acbSDon Zickus for_each_online_cpu(cpu) 13858687acbSDon Zickus per_cpu(watchdog_touch_ts, cpu) = 0; 13958687acbSDon Zickus } 14058687acbSDon Zickus 141cafcd80dSDon Zickus #ifdef CONFIG_HARDLOCKUP_DETECTOR 14258687acbSDon Zickus void touch_nmi_watchdog(void) 14358687acbSDon Zickus { 14468d3f1d8SDon Zickus if (watchdog_enabled) { 14568d3f1d8SDon Zickus unsigned cpu; 14668d3f1d8SDon Zickus 14768d3f1d8SDon Zickus for_each_present_cpu(cpu) { 14868d3f1d8SDon Zickus if (per_cpu(watchdog_nmi_touch, cpu) != true) 14968d3f1d8SDon Zickus per_cpu(watchdog_nmi_touch, cpu) = true; 15068d3f1d8SDon Zickus } 15168d3f1d8SDon Zickus } 152332fbdbcSDon Zickus touch_softlockup_watchdog(); 15358687acbSDon Zickus } 15458687acbSDon Zickus EXPORT_SYMBOL(touch_nmi_watchdog); 15558687acbSDon Zickus 156cafcd80dSDon Zickus #endif 157cafcd80dSDon Zickus 15858687acbSDon Zickus void touch_softlockup_watchdog_sync(void) 15958687acbSDon Zickus { 16058687acbSDon Zickus __raw_get_cpu_var(softlockup_touch_sync) = true; 16158687acbSDon Zickus __raw_get_cpu_var(watchdog_touch_ts) = 0; 16258687acbSDon Zickus } 16358687acbSDon Zickus 16423637d47SFrederic Weisbecker #ifdef CONFIG_HARDLOCKUP_DETECTOR 16558687acbSDon Zickus /* watchdog detector functions */ 16626e09c6eSDon Zickus static int is_hardlockup(void) 16758687acbSDon Zickus { 16826e09c6eSDon Zickus unsigned long hrint = __get_cpu_var(hrtimer_interrupts); 16958687acbSDon Zickus 17026e09c6eSDon Zickus if (__get_cpu_var(hrtimer_interrupts_saved) == hrint) 17158687acbSDon Zickus return 1; 17258687acbSDon Zickus 17326e09c6eSDon Zickus __get_cpu_var(hrtimer_interrupts_saved) = hrint; 17458687acbSDon Zickus return 0; 17558687acbSDon Zickus } 17658687acbSDon Zickus #endif 17758687acbSDon Zickus 17826e09c6eSDon Zickus static int is_softlockup(unsigned long touch_ts) 17958687acbSDon Zickus { 18026e09c6eSDon Zickus unsigned long now = get_timestamp(smp_processor_id()); 18158687acbSDon Zickus 18258687acbSDon Zickus /* Warn about unreasonable delays: */ 18358687acbSDon Zickus if (time_after(now, touch_ts + softlockup_thresh)) 18458687acbSDon Zickus return now - touch_ts; 18558687acbSDon Zickus 18658687acbSDon Zickus return 0; 18758687acbSDon Zickus } 18858687acbSDon Zickus 18923637d47SFrederic Weisbecker #ifdef CONFIG_HARDLOCKUP_DETECTOR 19058687acbSDon Zickus static struct perf_event_attr wd_hw_attr = { 19158687acbSDon Zickus .type = PERF_TYPE_HARDWARE, 19258687acbSDon Zickus .config = PERF_COUNT_HW_CPU_CYCLES, 19358687acbSDon Zickus .size = sizeof(struct perf_event_attr), 19458687acbSDon Zickus .pinned = 1, 19558687acbSDon Zickus .disabled = 1, 19658687acbSDon Zickus }; 19758687acbSDon Zickus 19858687acbSDon Zickus /* Callback function for perf event subsystem */ 19958687acbSDon Zickus void watchdog_overflow_callback(struct perf_event *event, int nmi, 20058687acbSDon Zickus struct perf_sample_data *data, 20158687acbSDon Zickus struct pt_regs *regs) 20258687acbSDon Zickus { 203c6db67cdSPeter Zijlstra /* Ensure the watchdog never gets throttled */ 204c6db67cdSPeter Zijlstra event->hw.interrupts = 0; 205c6db67cdSPeter Zijlstra 206d7c54733SDon Zickus if (__get_cpu_var(watchdog_nmi_touch) == true) { 207d7c54733SDon Zickus __get_cpu_var(watchdog_nmi_touch) = false; 20858687acbSDon Zickus return; 20958687acbSDon Zickus } 21058687acbSDon Zickus 21158687acbSDon Zickus /* check for a hardlockup 21258687acbSDon Zickus * This is done by making sure our timer interrupt 21358687acbSDon Zickus * is incrementing. The timer interrupt should have 21458687acbSDon Zickus * fired multiple times before we overflow'd. If it hasn't 21558687acbSDon Zickus * then this is a good indication the cpu is stuck 21658687acbSDon Zickus */ 21726e09c6eSDon Zickus if (is_hardlockup()) { 21826e09c6eSDon Zickus int this_cpu = smp_processor_id(); 21926e09c6eSDon Zickus 22058687acbSDon Zickus /* only print hardlockups once */ 22158687acbSDon Zickus if (__get_cpu_var(hard_watchdog_warn) == true) 22258687acbSDon Zickus return; 22358687acbSDon Zickus 22458687acbSDon Zickus if (hardlockup_panic) 22558687acbSDon Zickus panic("Watchdog detected hard LOCKUP on cpu %d", this_cpu); 22658687acbSDon Zickus else 22758687acbSDon Zickus WARN(1, "Watchdog detected hard LOCKUP on cpu %d", this_cpu); 22858687acbSDon Zickus 22958687acbSDon Zickus __get_cpu_var(hard_watchdog_warn) = true; 23058687acbSDon Zickus return; 23158687acbSDon Zickus } 23258687acbSDon Zickus 23358687acbSDon Zickus __get_cpu_var(hard_watchdog_warn) = false; 23458687acbSDon Zickus return; 23558687acbSDon Zickus } 23658687acbSDon Zickus static void watchdog_interrupt_count(void) 23758687acbSDon Zickus { 23858687acbSDon Zickus __get_cpu_var(hrtimer_interrupts)++; 23958687acbSDon Zickus } 24058687acbSDon Zickus #else 24158687acbSDon Zickus static inline void watchdog_interrupt_count(void) { return; } 24223637d47SFrederic Weisbecker #endif /* CONFIG_HARDLOCKUP_DETECTOR */ 24358687acbSDon Zickus 24458687acbSDon Zickus /* watchdog kicker functions */ 24558687acbSDon Zickus static enum hrtimer_restart watchdog_timer_fn(struct hrtimer *hrtimer) 24658687acbSDon Zickus { 24758687acbSDon Zickus unsigned long touch_ts = __get_cpu_var(watchdog_touch_ts); 24858687acbSDon Zickus struct pt_regs *regs = get_irq_regs(); 24958687acbSDon Zickus int duration; 25058687acbSDon Zickus 25158687acbSDon Zickus /* kick the hardlockup detector */ 25258687acbSDon Zickus watchdog_interrupt_count(); 25358687acbSDon Zickus 25458687acbSDon Zickus /* kick the softlockup detector */ 25558687acbSDon Zickus wake_up_process(__get_cpu_var(softlockup_watchdog)); 25658687acbSDon Zickus 25758687acbSDon Zickus /* .. and repeat */ 25858687acbSDon Zickus hrtimer_forward_now(hrtimer, ns_to_ktime(get_sample_period())); 25958687acbSDon Zickus 26058687acbSDon Zickus if (touch_ts == 0) { 26126e09c6eSDon Zickus if (unlikely(__get_cpu_var(softlockup_touch_sync))) { 26258687acbSDon Zickus /* 26358687acbSDon Zickus * If the time stamp was touched atomically 26458687acbSDon Zickus * make sure the scheduler tick is up to date. 26558687acbSDon Zickus */ 26626e09c6eSDon Zickus __get_cpu_var(softlockup_touch_sync) = false; 26758687acbSDon Zickus sched_clock_tick(); 26858687acbSDon Zickus } 26958687acbSDon Zickus __touch_watchdog(); 27058687acbSDon Zickus return HRTIMER_RESTART; 27158687acbSDon Zickus } 27258687acbSDon Zickus 27358687acbSDon Zickus /* check for a softlockup 27458687acbSDon Zickus * This is done by making sure a high priority task is 27558687acbSDon Zickus * being scheduled. The task touches the watchdog to 27658687acbSDon Zickus * indicate it is getting cpu time. If it hasn't then 27758687acbSDon Zickus * this is a good indication some task is hogging the cpu 27858687acbSDon Zickus */ 27926e09c6eSDon Zickus duration = is_softlockup(touch_ts); 28058687acbSDon Zickus if (unlikely(duration)) { 28158687acbSDon Zickus /* only warn once */ 28258687acbSDon Zickus if (__get_cpu_var(soft_watchdog_warn) == true) 28358687acbSDon Zickus return HRTIMER_RESTART; 28458687acbSDon Zickus 28558687acbSDon Zickus printk(KERN_ERR "BUG: soft lockup - CPU#%d stuck for %us! [%s:%d]\n", 28626e09c6eSDon Zickus smp_processor_id(), duration, 28758687acbSDon Zickus current->comm, task_pid_nr(current)); 28858687acbSDon Zickus print_modules(); 28958687acbSDon Zickus print_irqtrace_events(current); 29058687acbSDon Zickus if (regs) 29158687acbSDon Zickus show_regs(regs); 29258687acbSDon Zickus else 29358687acbSDon Zickus dump_stack(); 29458687acbSDon Zickus 29558687acbSDon Zickus if (softlockup_panic) 29658687acbSDon Zickus panic("softlockup: hung tasks"); 29758687acbSDon Zickus __get_cpu_var(soft_watchdog_warn) = true; 29858687acbSDon Zickus } else 29958687acbSDon Zickus __get_cpu_var(soft_watchdog_warn) = false; 30058687acbSDon Zickus 30158687acbSDon Zickus return HRTIMER_RESTART; 30258687acbSDon Zickus } 30358687acbSDon Zickus 30458687acbSDon Zickus 30558687acbSDon Zickus /* 30658687acbSDon Zickus * The watchdog thread - touches the timestamp. 30758687acbSDon Zickus */ 30826e09c6eSDon Zickus static int watchdog(void *unused) 30958687acbSDon Zickus { 31058687acbSDon Zickus struct sched_param param = { .sched_priority = MAX_RT_PRIO-1 }; 31126e09c6eSDon Zickus struct hrtimer *hrtimer = &__raw_get_cpu_var(watchdog_hrtimer); 31258687acbSDon Zickus 31358687acbSDon Zickus sched_setscheduler(current, SCHED_FIFO, ¶m); 31458687acbSDon Zickus 31558687acbSDon Zickus /* initialize timestamp */ 31658687acbSDon Zickus __touch_watchdog(); 31758687acbSDon Zickus 31858687acbSDon Zickus /* kick off the timer for the hardlockup detector */ 31958687acbSDon Zickus /* done here because hrtimer_start can only pin to smp_processor_id() */ 32058687acbSDon Zickus hrtimer_start(hrtimer, ns_to_ktime(get_sample_period()), 32158687acbSDon Zickus HRTIMER_MODE_REL_PINNED); 32258687acbSDon Zickus 32358687acbSDon Zickus set_current_state(TASK_INTERRUPTIBLE); 32458687acbSDon Zickus /* 32558687acbSDon Zickus * Run briefly once per second to reset the softlockup timestamp. 32658687acbSDon Zickus * If this gets delayed for more than 60 seconds then the 32726e09c6eSDon Zickus * debug-printout triggers in watchdog_timer_fn(). 32858687acbSDon Zickus */ 32958687acbSDon Zickus while (!kthread_should_stop()) { 33058687acbSDon Zickus __touch_watchdog(); 33158687acbSDon Zickus schedule(); 33258687acbSDon Zickus 33358687acbSDon Zickus if (kthread_should_stop()) 33458687acbSDon Zickus break; 33558687acbSDon Zickus 33658687acbSDon Zickus set_current_state(TASK_INTERRUPTIBLE); 33758687acbSDon Zickus } 33858687acbSDon Zickus __set_current_state(TASK_RUNNING); 33958687acbSDon Zickus 34058687acbSDon Zickus return 0; 34158687acbSDon Zickus } 34258687acbSDon Zickus 34358687acbSDon Zickus 34423637d47SFrederic Weisbecker #ifdef CONFIG_HARDLOCKUP_DETECTOR 34558687acbSDon Zickus static int watchdog_nmi_enable(int cpu) 34658687acbSDon Zickus { 34758687acbSDon Zickus struct perf_event_attr *wd_attr; 34858687acbSDon Zickus struct perf_event *event = per_cpu(watchdog_ev, cpu); 34958687acbSDon Zickus 35058687acbSDon Zickus /* is it already setup and enabled? */ 35158687acbSDon Zickus if (event && event->state > PERF_EVENT_STATE_OFF) 35258687acbSDon Zickus goto out; 35358687acbSDon Zickus 35458687acbSDon Zickus /* it is setup but not enabled */ 35558687acbSDon Zickus if (event != NULL) 35658687acbSDon Zickus goto out_enable; 35758687acbSDon Zickus 35858687acbSDon Zickus /* Try to register using hardware perf events */ 35958687acbSDon Zickus wd_attr = &wd_hw_attr; 36058687acbSDon Zickus wd_attr->sample_period = hw_nmi_get_sample_period(); 36138a81da2SMatt Helsley event = perf_event_create_kernel_counter(wd_attr, cpu, NULL, watchdog_overflow_callback); 36258687acbSDon Zickus if (!IS_ERR(event)) { 36358687acbSDon Zickus printk(KERN_INFO "NMI watchdog enabled, takes one hw-pmu counter.\n"); 36458687acbSDon Zickus goto out_save; 36558687acbSDon Zickus } 36658687acbSDon Zickus 36758687acbSDon Zickus printk(KERN_ERR "NMI watchdog failed to create perf event on cpu%i: %p\n", cpu, event); 368eac24335SAkinobu Mita return PTR_ERR(event); 36958687acbSDon Zickus 37058687acbSDon Zickus /* success path */ 37158687acbSDon Zickus out_save: 37258687acbSDon Zickus per_cpu(watchdog_ev, cpu) = event; 37358687acbSDon Zickus out_enable: 37458687acbSDon Zickus perf_event_enable(per_cpu(watchdog_ev, cpu)); 37558687acbSDon Zickus out: 37658687acbSDon Zickus return 0; 37758687acbSDon Zickus } 37858687acbSDon Zickus 37958687acbSDon Zickus static void watchdog_nmi_disable(int cpu) 38058687acbSDon Zickus { 38158687acbSDon Zickus struct perf_event *event = per_cpu(watchdog_ev, cpu); 38258687acbSDon Zickus 38358687acbSDon Zickus if (event) { 38458687acbSDon Zickus perf_event_disable(event); 38558687acbSDon Zickus per_cpu(watchdog_ev, cpu) = NULL; 38658687acbSDon Zickus 38758687acbSDon Zickus /* should be in cleanup, but blocks oprofile */ 38858687acbSDon Zickus perf_event_release_kernel(event); 38958687acbSDon Zickus } 39058687acbSDon Zickus return; 39158687acbSDon Zickus } 39258687acbSDon Zickus #else 39358687acbSDon Zickus static int watchdog_nmi_enable(int cpu) { return 0; } 39458687acbSDon Zickus static void watchdog_nmi_disable(int cpu) { return; } 39523637d47SFrederic Weisbecker #endif /* CONFIG_HARDLOCKUP_DETECTOR */ 39658687acbSDon Zickus 39758687acbSDon Zickus /* prepare/enable/disable routines */ 39858687acbSDon Zickus static int watchdog_prepare_cpu(int cpu) 39958687acbSDon Zickus { 40058687acbSDon Zickus struct hrtimer *hrtimer = &per_cpu(watchdog_hrtimer, cpu); 40158687acbSDon Zickus 40258687acbSDon Zickus WARN_ON(per_cpu(softlockup_watchdog, cpu)); 40358687acbSDon Zickus hrtimer_init(hrtimer, CLOCK_MONOTONIC, HRTIMER_MODE_REL); 40458687acbSDon Zickus hrtimer->function = watchdog_timer_fn; 40558687acbSDon Zickus 40658687acbSDon Zickus return 0; 40758687acbSDon Zickus } 40858687acbSDon Zickus 40958687acbSDon Zickus static int watchdog_enable(int cpu) 41058687acbSDon Zickus { 41158687acbSDon Zickus struct task_struct *p = per_cpu(softlockup_watchdog, cpu); 412eac24335SAkinobu Mita int err; 41358687acbSDon Zickus 41458687acbSDon Zickus /* enable the perf event */ 415eac24335SAkinobu Mita err = watchdog_nmi_enable(cpu); 416eac24335SAkinobu Mita if (err) 417eac24335SAkinobu Mita return err; 41858687acbSDon Zickus 41958687acbSDon Zickus /* create the watchdog thread */ 42058687acbSDon Zickus if (!p) { 42158687acbSDon Zickus p = kthread_create(watchdog, (void *)(unsigned long)cpu, "watchdog/%d", cpu); 42258687acbSDon Zickus if (IS_ERR(p)) { 42358687acbSDon Zickus printk(KERN_ERR "softlockup watchdog for %i failed\n", cpu); 424eac24335SAkinobu Mita return PTR_ERR(p); 42558687acbSDon Zickus } 42658687acbSDon Zickus kthread_bind(p, cpu); 42758687acbSDon Zickus per_cpu(watchdog_touch_ts, cpu) = 0; 42858687acbSDon Zickus per_cpu(softlockup_watchdog, cpu) = p; 42958687acbSDon Zickus wake_up_process(p); 43058687acbSDon Zickus } 43158687acbSDon Zickus 43268d3f1d8SDon Zickus /* if any cpu succeeds, watchdog is considered enabled for the system */ 43368d3f1d8SDon Zickus watchdog_enabled = 1; 43468d3f1d8SDon Zickus 43558687acbSDon Zickus return 0; 43658687acbSDon Zickus } 43758687acbSDon Zickus 43858687acbSDon Zickus static void watchdog_disable(int cpu) 43958687acbSDon Zickus { 44058687acbSDon Zickus struct task_struct *p = per_cpu(softlockup_watchdog, cpu); 44158687acbSDon Zickus struct hrtimer *hrtimer = &per_cpu(watchdog_hrtimer, cpu); 44258687acbSDon Zickus 44358687acbSDon Zickus /* 44458687acbSDon Zickus * cancel the timer first to stop incrementing the stats 44558687acbSDon Zickus * and waking up the kthread 44658687acbSDon Zickus */ 44758687acbSDon Zickus hrtimer_cancel(hrtimer); 44858687acbSDon Zickus 44958687acbSDon Zickus /* disable the perf event */ 45058687acbSDon Zickus watchdog_nmi_disable(cpu); 45158687acbSDon Zickus 45258687acbSDon Zickus /* stop the watchdog thread */ 45358687acbSDon Zickus if (p) { 45458687acbSDon Zickus per_cpu(softlockup_watchdog, cpu) = NULL; 45558687acbSDon Zickus kthread_stop(p); 45658687acbSDon Zickus } 45758687acbSDon Zickus } 45858687acbSDon Zickus 45958687acbSDon Zickus static void watchdog_enable_all_cpus(void) 46058687acbSDon Zickus { 46158687acbSDon Zickus int cpu; 462eb703f98SKulikov Vasiliy int result = 0; 46358687acbSDon Zickus 46458687acbSDon Zickus for_each_online_cpu(cpu) 46558687acbSDon Zickus result += watchdog_enable(cpu); 46658687acbSDon Zickus 46758687acbSDon Zickus if (result) 46858687acbSDon Zickus printk(KERN_ERR "watchdog: failed to be enabled on some cpus\n"); 46958687acbSDon Zickus 47058687acbSDon Zickus } 47158687acbSDon Zickus 47258687acbSDon Zickus static void watchdog_disable_all_cpus(void) 47358687acbSDon Zickus { 47458687acbSDon Zickus int cpu; 47558687acbSDon Zickus 476d9ca07a0SStephane Eranian if (no_watchdog) 477d9ca07a0SStephane Eranian return; 478d9ca07a0SStephane Eranian 47958687acbSDon Zickus for_each_online_cpu(cpu) 48058687acbSDon Zickus watchdog_disable(cpu); 48158687acbSDon Zickus 48258687acbSDon Zickus /* if all watchdogs are disabled, then they are disabled for the system */ 48358687acbSDon Zickus watchdog_enabled = 0; 48458687acbSDon Zickus } 48558687acbSDon Zickus 48658687acbSDon Zickus 48758687acbSDon Zickus /* sysctl functions */ 48858687acbSDon Zickus #ifdef CONFIG_SYSCTL 48958687acbSDon Zickus /* 49058687acbSDon Zickus * proc handler for /proc/sys/kernel/nmi_watchdog 49158687acbSDon Zickus */ 49258687acbSDon Zickus 49358687acbSDon Zickus int proc_dowatchdog_enabled(struct ctl_table *table, int write, 49458687acbSDon Zickus void __user *buffer, size_t *length, loff_t *ppos) 49558687acbSDon Zickus { 49658687acbSDon Zickus proc_dointvec(table, write, buffer, length, ppos); 49758687acbSDon Zickus 49858687acbSDon Zickus if (watchdog_enabled) 49958687acbSDon Zickus watchdog_enable_all_cpus(); 50058687acbSDon Zickus else 50158687acbSDon Zickus watchdog_disable_all_cpus(); 50258687acbSDon Zickus return 0; 50358687acbSDon Zickus } 50458687acbSDon Zickus 50558687acbSDon Zickus int proc_dowatchdog_thresh(struct ctl_table *table, int write, 50658687acbSDon Zickus void __user *buffer, 50758687acbSDon Zickus size_t *lenp, loff_t *ppos) 50858687acbSDon Zickus { 50958687acbSDon Zickus return proc_dointvec_minmax(table, write, buffer, lenp, ppos); 51058687acbSDon Zickus } 51158687acbSDon Zickus #endif /* CONFIG_SYSCTL */ 51258687acbSDon Zickus 51358687acbSDon Zickus 51458687acbSDon Zickus /* 51558687acbSDon Zickus * Create/destroy watchdog threads as CPUs come and go: 51658687acbSDon Zickus */ 51758687acbSDon Zickus static int __cpuinit 51858687acbSDon Zickus cpu_callback(struct notifier_block *nfb, unsigned long action, void *hcpu) 51958687acbSDon Zickus { 52058687acbSDon Zickus int hotcpu = (unsigned long)hcpu; 521eac24335SAkinobu Mita int err = 0; 52258687acbSDon Zickus 52358687acbSDon Zickus switch (action) { 52458687acbSDon Zickus case CPU_UP_PREPARE: 52558687acbSDon Zickus case CPU_UP_PREPARE_FROZEN: 526eac24335SAkinobu Mita err = watchdog_prepare_cpu(hotcpu); 52758687acbSDon Zickus break; 52858687acbSDon Zickus case CPU_ONLINE: 52958687acbSDon Zickus case CPU_ONLINE_FROZEN: 530eac24335SAkinobu Mita err = watchdog_enable(hotcpu); 53158687acbSDon Zickus break; 53258687acbSDon Zickus #ifdef CONFIG_HOTPLUG_CPU 53358687acbSDon Zickus case CPU_UP_CANCELED: 53458687acbSDon Zickus case CPU_UP_CANCELED_FROZEN: 53558687acbSDon Zickus watchdog_disable(hotcpu); 53658687acbSDon Zickus break; 53758687acbSDon Zickus case CPU_DEAD: 53858687acbSDon Zickus case CPU_DEAD_FROZEN: 53958687acbSDon Zickus watchdog_disable(hotcpu); 54058687acbSDon Zickus break; 54158687acbSDon Zickus #endif /* CONFIG_HOTPLUG_CPU */ 54258687acbSDon Zickus } 543eac24335SAkinobu Mita return notifier_from_errno(err); 54458687acbSDon Zickus } 54558687acbSDon Zickus 54658687acbSDon Zickus static struct notifier_block __cpuinitdata cpu_nfb = { 54758687acbSDon Zickus .notifier_call = cpu_callback 54858687acbSDon Zickus }; 54958687acbSDon Zickus 55058687acbSDon Zickus static int __init spawn_watchdog_task(void) 55158687acbSDon Zickus { 55258687acbSDon Zickus void *cpu = (void *)(long)smp_processor_id(); 55358687acbSDon Zickus int err; 55458687acbSDon Zickus 55558687acbSDon Zickus if (no_watchdog) 55658687acbSDon Zickus return 0; 55758687acbSDon Zickus 55858687acbSDon Zickus err = cpu_callback(&cpu_nfb, CPU_UP_PREPARE, cpu); 559eac24335SAkinobu Mita WARN_ON(notifier_to_errno(err)); 56058687acbSDon Zickus 56158687acbSDon Zickus cpu_callback(&cpu_nfb, CPU_ONLINE, cpu); 56258687acbSDon Zickus register_cpu_notifier(&cpu_nfb); 56358687acbSDon Zickus 56458687acbSDon Zickus return 0; 56558687acbSDon Zickus } 56658687acbSDon Zickus early_initcall(spawn_watchdog_task); 567