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 304135038aSMarcin Slusarz int watchdog_enabled = 1; 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 /* boot commands */ 4758687acbSDon Zickus /* 4858687acbSDon Zickus * Should we panic when a soft-lockup or hard-lockup occurs: 4958687acbSDon Zickus */ 5023637d47SFrederic Weisbecker #ifdef CONFIG_HARDLOCKUP_DETECTOR 51fef2c9bcSDon Zickus static int hardlockup_panic = 52fef2c9bcSDon Zickus CONFIG_BOOTPARAM_HARDLOCKUP_PANIC_VALUE; 5358687acbSDon Zickus 5458687acbSDon Zickus static int __init hardlockup_panic_setup(char *str) 5558687acbSDon Zickus { 5658687acbSDon Zickus if (!strncmp(str, "panic", 5)) 5758687acbSDon Zickus hardlockup_panic = 1; 58fef2c9bcSDon Zickus else if (!strncmp(str, "nopanic", 7)) 59fef2c9bcSDon Zickus hardlockup_panic = 0; 605dc30558SDon Zickus else if (!strncmp(str, "0", 1)) 614135038aSMarcin Slusarz watchdog_enabled = 0; 6258687acbSDon Zickus return 1; 6358687acbSDon Zickus } 6458687acbSDon Zickus __setup("nmi_watchdog=", hardlockup_panic_setup); 6558687acbSDon Zickus #endif 6658687acbSDon Zickus 6758687acbSDon Zickus unsigned int __read_mostly softlockup_panic = 6858687acbSDon Zickus CONFIG_BOOTPARAM_SOFTLOCKUP_PANIC_VALUE; 6958687acbSDon Zickus 7058687acbSDon Zickus static int __init softlockup_panic_setup(char *str) 7158687acbSDon Zickus { 7258687acbSDon Zickus softlockup_panic = simple_strtoul(str, NULL, 0); 7358687acbSDon Zickus 7458687acbSDon Zickus return 1; 7558687acbSDon Zickus } 7658687acbSDon Zickus __setup("softlockup_panic=", softlockup_panic_setup); 7758687acbSDon Zickus 7858687acbSDon Zickus static int __init nowatchdog_setup(char *str) 7958687acbSDon Zickus { 804135038aSMarcin Slusarz watchdog_enabled = 0; 8158687acbSDon Zickus return 1; 8258687acbSDon Zickus } 8358687acbSDon Zickus __setup("nowatchdog", nowatchdog_setup); 8458687acbSDon Zickus 8558687acbSDon Zickus /* deprecated */ 8658687acbSDon Zickus static int __init nosoftlockup_setup(char *str) 8758687acbSDon Zickus { 884135038aSMarcin Slusarz watchdog_enabled = 0; 8958687acbSDon Zickus return 1; 9058687acbSDon Zickus } 9158687acbSDon Zickus __setup("nosoftlockup", nosoftlockup_setup); 9258687acbSDon Zickus /* */ 9358687acbSDon Zickus 9458687acbSDon Zickus 9558687acbSDon Zickus /* 9658687acbSDon Zickus * Returns seconds, approximately. We don't need nanosecond 9758687acbSDon Zickus * resolution, and we don't need to waste time with a big divide when 9858687acbSDon Zickus * 2^30ns == 1.074s. 9958687acbSDon Zickus */ 10058687acbSDon Zickus static unsigned long get_timestamp(int this_cpu) 10158687acbSDon Zickus { 10258687acbSDon Zickus return cpu_clock(this_cpu) >> 30LL; /* 2^30 ~= 10^9 */ 10358687acbSDon Zickus } 10458687acbSDon Zickus 10558687acbSDon Zickus static unsigned long get_sample_period(void) 10658687acbSDon Zickus { 10758687acbSDon Zickus /* 10858687acbSDon Zickus * convert softlockup_thresh from seconds to ns 10958687acbSDon Zickus * the divide by 5 is to give hrtimer 5 chances to 11058687acbSDon Zickus * increment before the hardlockup detector generates 11158687acbSDon Zickus * a warning 11258687acbSDon Zickus */ 11358687acbSDon Zickus return softlockup_thresh / 5 * NSEC_PER_SEC; 11458687acbSDon Zickus } 11558687acbSDon Zickus 11658687acbSDon Zickus /* Commands for resetting the watchdog */ 11758687acbSDon Zickus static void __touch_watchdog(void) 11858687acbSDon Zickus { 11926e09c6eSDon Zickus int this_cpu = smp_processor_id(); 12058687acbSDon Zickus 121909ea964SChristoph Lameter __this_cpu_write(watchdog_touch_ts, get_timestamp(this_cpu)); 12258687acbSDon Zickus } 12358687acbSDon Zickus 124332fbdbcSDon Zickus void touch_softlockup_watchdog(void) 12558687acbSDon Zickus { 126909ea964SChristoph Lameter __this_cpu_write(watchdog_touch_ts, 0); 12758687acbSDon Zickus } 1280167c781SIngo Molnar EXPORT_SYMBOL(touch_softlockup_watchdog); 12958687acbSDon Zickus 130332fbdbcSDon Zickus void touch_all_softlockup_watchdogs(void) 13158687acbSDon Zickus { 13258687acbSDon Zickus int cpu; 13358687acbSDon Zickus 13458687acbSDon Zickus /* 13558687acbSDon Zickus * this is done lockless 13658687acbSDon Zickus * do we care if a 0 races with a timestamp? 13758687acbSDon Zickus * all it means is the softlock check starts one cycle later 13858687acbSDon Zickus */ 13958687acbSDon Zickus for_each_online_cpu(cpu) 14058687acbSDon Zickus per_cpu(watchdog_touch_ts, cpu) = 0; 14158687acbSDon Zickus } 14258687acbSDon Zickus 143cafcd80dSDon Zickus #ifdef CONFIG_HARDLOCKUP_DETECTOR 14458687acbSDon Zickus void touch_nmi_watchdog(void) 14558687acbSDon Zickus { 14668d3f1d8SDon Zickus if (watchdog_enabled) { 14768d3f1d8SDon Zickus unsigned cpu; 14868d3f1d8SDon Zickus 14968d3f1d8SDon Zickus for_each_present_cpu(cpu) { 15068d3f1d8SDon Zickus if (per_cpu(watchdog_nmi_touch, cpu) != true) 15168d3f1d8SDon Zickus per_cpu(watchdog_nmi_touch, cpu) = true; 15268d3f1d8SDon Zickus } 15368d3f1d8SDon Zickus } 154332fbdbcSDon Zickus touch_softlockup_watchdog(); 15558687acbSDon Zickus } 15658687acbSDon Zickus EXPORT_SYMBOL(touch_nmi_watchdog); 15758687acbSDon Zickus 158cafcd80dSDon Zickus #endif 159cafcd80dSDon Zickus 16058687acbSDon Zickus void touch_softlockup_watchdog_sync(void) 16158687acbSDon Zickus { 16258687acbSDon Zickus __raw_get_cpu_var(softlockup_touch_sync) = true; 16358687acbSDon Zickus __raw_get_cpu_var(watchdog_touch_ts) = 0; 16458687acbSDon Zickus } 16558687acbSDon Zickus 16623637d47SFrederic Weisbecker #ifdef CONFIG_HARDLOCKUP_DETECTOR 16758687acbSDon Zickus /* watchdog detector functions */ 16826e09c6eSDon Zickus static int is_hardlockup(void) 16958687acbSDon Zickus { 170909ea964SChristoph Lameter unsigned long hrint = __this_cpu_read(hrtimer_interrupts); 17158687acbSDon Zickus 172909ea964SChristoph Lameter if (__this_cpu_read(hrtimer_interrupts_saved) == hrint) 17358687acbSDon Zickus return 1; 17458687acbSDon Zickus 175909ea964SChristoph Lameter __this_cpu_write(hrtimer_interrupts_saved, hrint); 17658687acbSDon Zickus return 0; 17758687acbSDon Zickus } 17858687acbSDon Zickus #endif 17958687acbSDon Zickus 18026e09c6eSDon Zickus static int is_softlockup(unsigned long touch_ts) 18158687acbSDon Zickus { 18226e09c6eSDon Zickus unsigned long now = get_timestamp(smp_processor_id()); 18358687acbSDon Zickus 18458687acbSDon Zickus /* Warn about unreasonable delays: */ 18558687acbSDon Zickus if (time_after(now, touch_ts + softlockup_thresh)) 18658687acbSDon Zickus return now - touch_ts; 18758687acbSDon Zickus 18858687acbSDon Zickus return 0; 18958687acbSDon Zickus } 19058687acbSDon Zickus 19123637d47SFrederic Weisbecker #ifdef CONFIG_HARDLOCKUP_DETECTOR 19258687acbSDon Zickus static struct perf_event_attr wd_hw_attr = { 19358687acbSDon Zickus .type = PERF_TYPE_HARDWARE, 19458687acbSDon Zickus .config = PERF_COUNT_HW_CPU_CYCLES, 19558687acbSDon Zickus .size = sizeof(struct perf_event_attr), 19658687acbSDon Zickus .pinned = 1, 19758687acbSDon Zickus .disabled = 1, 19858687acbSDon Zickus }; 19958687acbSDon Zickus 20058687acbSDon Zickus /* Callback function for perf event subsystem */ 201277b1998SLin Ming static void watchdog_overflow_callback(struct perf_event *event, int nmi, 20258687acbSDon Zickus struct perf_sample_data *data, 20358687acbSDon Zickus struct pt_regs *regs) 20458687acbSDon Zickus { 205c6db67cdSPeter Zijlstra /* Ensure the watchdog never gets throttled */ 206c6db67cdSPeter Zijlstra event->hw.interrupts = 0; 207c6db67cdSPeter Zijlstra 208909ea964SChristoph Lameter if (__this_cpu_read(watchdog_nmi_touch) == true) { 209909ea964SChristoph Lameter __this_cpu_write(watchdog_nmi_touch, false); 21058687acbSDon Zickus return; 21158687acbSDon Zickus } 21258687acbSDon Zickus 21358687acbSDon Zickus /* check for a hardlockup 21458687acbSDon Zickus * This is done by making sure our timer interrupt 21558687acbSDon Zickus * is incrementing. The timer interrupt should have 21658687acbSDon Zickus * fired multiple times before we overflow'd. If it hasn't 21758687acbSDon Zickus * then this is a good indication the cpu is stuck 21858687acbSDon Zickus */ 21926e09c6eSDon Zickus if (is_hardlockup()) { 22026e09c6eSDon Zickus int this_cpu = smp_processor_id(); 22126e09c6eSDon Zickus 22258687acbSDon Zickus /* only print hardlockups once */ 223909ea964SChristoph Lameter if (__this_cpu_read(hard_watchdog_warn) == true) 22458687acbSDon Zickus return; 22558687acbSDon Zickus 22658687acbSDon Zickus if (hardlockup_panic) 22758687acbSDon Zickus panic("Watchdog detected hard LOCKUP on cpu %d", this_cpu); 22858687acbSDon Zickus else 22958687acbSDon Zickus WARN(1, "Watchdog detected hard LOCKUP on cpu %d", this_cpu); 23058687acbSDon Zickus 231909ea964SChristoph Lameter __this_cpu_write(hard_watchdog_warn, true); 23258687acbSDon Zickus return; 23358687acbSDon Zickus } 23458687acbSDon Zickus 235909ea964SChristoph Lameter __this_cpu_write(hard_watchdog_warn, false); 23658687acbSDon Zickus return; 23758687acbSDon Zickus } 23858687acbSDon Zickus static void watchdog_interrupt_count(void) 23958687acbSDon Zickus { 240909ea964SChristoph Lameter __this_cpu_inc(hrtimer_interrupts); 24158687acbSDon Zickus } 24258687acbSDon Zickus #else 24358687acbSDon Zickus static inline void watchdog_interrupt_count(void) { return; } 24423637d47SFrederic Weisbecker #endif /* CONFIG_HARDLOCKUP_DETECTOR */ 24558687acbSDon Zickus 24658687acbSDon Zickus /* watchdog kicker functions */ 24758687acbSDon Zickus static enum hrtimer_restart watchdog_timer_fn(struct hrtimer *hrtimer) 24858687acbSDon Zickus { 249909ea964SChristoph Lameter unsigned long touch_ts = __this_cpu_read(watchdog_touch_ts); 25058687acbSDon Zickus struct pt_regs *regs = get_irq_regs(); 25158687acbSDon Zickus int duration; 25258687acbSDon Zickus 25358687acbSDon Zickus /* kick the hardlockup detector */ 25458687acbSDon Zickus watchdog_interrupt_count(); 25558687acbSDon Zickus 25658687acbSDon Zickus /* kick the softlockup detector */ 257909ea964SChristoph Lameter wake_up_process(__this_cpu_read(softlockup_watchdog)); 25858687acbSDon Zickus 25958687acbSDon Zickus /* .. and repeat */ 26058687acbSDon Zickus hrtimer_forward_now(hrtimer, ns_to_ktime(get_sample_period())); 26158687acbSDon Zickus 26258687acbSDon Zickus if (touch_ts == 0) { 263909ea964SChristoph Lameter if (unlikely(__this_cpu_read(softlockup_touch_sync))) { 26458687acbSDon Zickus /* 26558687acbSDon Zickus * If the time stamp was touched atomically 26658687acbSDon Zickus * make sure the scheduler tick is up to date. 26758687acbSDon Zickus */ 268909ea964SChristoph Lameter __this_cpu_write(softlockup_touch_sync, false); 26958687acbSDon Zickus sched_clock_tick(); 27058687acbSDon Zickus } 27158687acbSDon Zickus __touch_watchdog(); 27258687acbSDon Zickus return HRTIMER_RESTART; 27358687acbSDon Zickus } 27458687acbSDon Zickus 27558687acbSDon Zickus /* check for a softlockup 27658687acbSDon Zickus * This is done by making sure a high priority task is 27758687acbSDon Zickus * being scheduled. The task touches the watchdog to 27858687acbSDon Zickus * indicate it is getting cpu time. If it hasn't then 27958687acbSDon Zickus * this is a good indication some task is hogging the cpu 28058687acbSDon Zickus */ 28126e09c6eSDon Zickus duration = is_softlockup(touch_ts); 28258687acbSDon Zickus if (unlikely(duration)) { 28358687acbSDon Zickus /* only warn once */ 284909ea964SChristoph Lameter if (__this_cpu_read(soft_watchdog_warn) == true) 28558687acbSDon Zickus return HRTIMER_RESTART; 28658687acbSDon Zickus 28758687acbSDon Zickus printk(KERN_ERR "BUG: soft lockup - CPU#%d stuck for %us! [%s:%d]\n", 28826e09c6eSDon Zickus smp_processor_id(), duration, 28958687acbSDon Zickus current->comm, task_pid_nr(current)); 29058687acbSDon Zickus print_modules(); 29158687acbSDon Zickus print_irqtrace_events(current); 29258687acbSDon Zickus if (regs) 29358687acbSDon Zickus show_regs(regs); 29458687acbSDon Zickus else 29558687acbSDon Zickus dump_stack(); 29658687acbSDon Zickus 29758687acbSDon Zickus if (softlockup_panic) 29858687acbSDon Zickus panic("softlockup: hung tasks"); 299909ea964SChristoph Lameter __this_cpu_write(soft_watchdog_warn, true); 30058687acbSDon Zickus } else 301909ea964SChristoph Lameter __this_cpu_write(soft_watchdog_warn, false); 30258687acbSDon Zickus 30358687acbSDon Zickus return HRTIMER_RESTART; 30458687acbSDon Zickus } 30558687acbSDon Zickus 30658687acbSDon Zickus 30758687acbSDon Zickus /* 30858687acbSDon Zickus * The watchdog thread - touches the timestamp. 30958687acbSDon Zickus */ 31026e09c6eSDon Zickus static int watchdog(void *unused) 31158687acbSDon Zickus { 312fe7de49fSKOSAKI Motohiro static struct sched_param param = { .sched_priority = MAX_RT_PRIO-1 }; 31326e09c6eSDon Zickus struct hrtimer *hrtimer = &__raw_get_cpu_var(watchdog_hrtimer); 31458687acbSDon Zickus 31558687acbSDon Zickus sched_setscheduler(current, SCHED_FIFO, ¶m); 31658687acbSDon Zickus 31758687acbSDon Zickus /* initialize timestamp */ 31858687acbSDon Zickus __touch_watchdog(); 31958687acbSDon Zickus 32058687acbSDon Zickus /* kick off the timer for the hardlockup detector */ 32158687acbSDon Zickus /* done here because hrtimer_start can only pin to smp_processor_id() */ 32258687acbSDon Zickus hrtimer_start(hrtimer, ns_to_ktime(get_sample_period()), 32358687acbSDon Zickus HRTIMER_MODE_REL_PINNED); 32458687acbSDon Zickus 32558687acbSDon Zickus set_current_state(TASK_INTERRUPTIBLE); 32658687acbSDon Zickus /* 32758687acbSDon Zickus * Run briefly once per second to reset the softlockup timestamp. 32858687acbSDon Zickus * If this gets delayed for more than 60 seconds then the 32926e09c6eSDon Zickus * debug-printout triggers in watchdog_timer_fn(). 33058687acbSDon Zickus */ 33158687acbSDon Zickus while (!kthread_should_stop()) { 33258687acbSDon Zickus __touch_watchdog(); 33358687acbSDon Zickus schedule(); 33458687acbSDon Zickus 33558687acbSDon Zickus if (kthread_should_stop()) 33658687acbSDon Zickus break; 33758687acbSDon Zickus 33858687acbSDon Zickus set_current_state(TASK_INTERRUPTIBLE); 33958687acbSDon Zickus } 34058687acbSDon Zickus __set_current_state(TASK_RUNNING); 34158687acbSDon Zickus 34258687acbSDon Zickus return 0; 34358687acbSDon Zickus } 34458687acbSDon Zickus 34558687acbSDon Zickus 34623637d47SFrederic Weisbecker #ifdef CONFIG_HARDLOCKUP_DETECTOR 34758687acbSDon Zickus static int watchdog_nmi_enable(int cpu) 34858687acbSDon Zickus { 34958687acbSDon Zickus struct perf_event_attr *wd_attr; 35058687acbSDon Zickus struct perf_event *event = per_cpu(watchdog_ev, cpu); 35158687acbSDon Zickus 35258687acbSDon Zickus /* is it already setup and enabled? */ 35358687acbSDon Zickus if (event && event->state > PERF_EVENT_STATE_OFF) 35458687acbSDon Zickus goto out; 35558687acbSDon Zickus 35658687acbSDon Zickus /* it is setup but not enabled */ 35758687acbSDon Zickus if (event != NULL) 35858687acbSDon Zickus goto out_enable; 35958687acbSDon Zickus 36058687acbSDon Zickus /* Try to register using hardware perf events */ 36158687acbSDon Zickus wd_attr = &wd_hw_attr; 36258687acbSDon Zickus wd_attr->sample_period = hw_nmi_get_sample_period(); 36338a81da2SMatt Helsley event = perf_event_create_kernel_counter(wd_attr, cpu, NULL, watchdog_overflow_callback); 36458687acbSDon Zickus if (!IS_ERR(event)) { 36558687acbSDon Zickus printk(KERN_INFO "NMI watchdog enabled, takes one hw-pmu counter.\n"); 36658687acbSDon Zickus goto out_save; 36758687acbSDon Zickus } 36858687acbSDon Zickus 3695651f7f4SDon Zickus 3705651f7f4SDon Zickus /* vary the KERN level based on the returned errno */ 3715651f7f4SDon Zickus if (PTR_ERR(event) == -EOPNOTSUPP) 3725651f7f4SDon Zickus printk(KERN_INFO "NMI watchdog disabled (cpu%i): not supported (no LAPIC?)\n", cpu); 3735651f7f4SDon Zickus else if (PTR_ERR(event) == -ENOENT) 3745651f7f4SDon Zickus printk(KERN_WARNING "NMI watchdog disabled (cpu%i): hardware events not enabled\n", cpu); 3755651f7f4SDon Zickus else 3765651f7f4SDon Zickus printk(KERN_ERR "NMI watchdog disabled (cpu%i): unable to create perf event: %ld\n", cpu, PTR_ERR(event)); 377eac24335SAkinobu Mita return PTR_ERR(event); 37858687acbSDon Zickus 37958687acbSDon Zickus /* success path */ 38058687acbSDon Zickus out_save: 38158687acbSDon Zickus per_cpu(watchdog_ev, cpu) = event; 38258687acbSDon Zickus out_enable: 38358687acbSDon Zickus perf_event_enable(per_cpu(watchdog_ev, cpu)); 38458687acbSDon Zickus out: 38558687acbSDon Zickus return 0; 38658687acbSDon Zickus } 38758687acbSDon Zickus 38858687acbSDon Zickus static void watchdog_nmi_disable(int cpu) 38958687acbSDon Zickus { 39058687acbSDon Zickus struct perf_event *event = per_cpu(watchdog_ev, cpu); 39158687acbSDon Zickus 39258687acbSDon Zickus if (event) { 39358687acbSDon Zickus perf_event_disable(event); 39458687acbSDon Zickus per_cpu(watchdog_ev, cpu) = NULL; 39558687acbSDon Zickus 39658687acbSDon Zickus /* should be in cleanup, but blocks oprofile */ 39758687acbSDon Zickus perf_event_release_kernel(event); 39858687acbSDon Zickus } 39958687acbSDon Zickus return; 40058687acbSDon Zickus } 40158687acbSDon Zickus #else 40258687acbSDon Zickus static int watchdog_nmi_enable(int cpu) { return 0; } 40358687acbSDon Zickus static void watchdog_nmi_disable(int cpu) { return; } 40423637d47SFrederic Weisbecker #endif /* CONFIG_HARDLOCKUP_DETECTOR */ 40558687acbSDon Zickus 40658687acbSDon Zickus /* prepare/enable/disable routines */ 40758687acbSDon Zickus static int watchdog_prepare_cpu(int cpu) 40858687acbSDon Zickus { 40958687acbSDon Zickus struct hrtimer *hrtimer = &per_cpu(watchdog_hrtimer, cpu); 41058687acbSDon Zickus 41158687acbSDon Zickus WARN_ON(per_cpu(softlockup_watchdog, cpu)); 41258687acbSDon Zickus hrtimer_init(hrtimer, CLOCK_MONOTONIC, HRTIMER_MODE_REL); 41358687acbSDon Zickus hrtimer->function = watchdog_timer_fn; 41458687acbSDon Zickus 41558687acbSDon Zickus return 0; 41658687acbSDon Zickus } 41758687acbSDon Zickus 41858687acbSDon Zickus static int watchdog_enable(int cpu) 41958687acbSDon Zickus { 42058687acbSDon Zickus struct task_struct *p = per_cpu(softlockup_watchdog, cpu); 421f99a9933SDon Zickus int err = 0; 42258687acbSDon Zickus 42358687acbSDon Zickus /* enable the perf event */ 424eac24335SAkinobu Mita err = watchdog_nmi_enable(cpu); 425f99a9933SDon Zickus 426f99a9933SDon Zickus /* Regardless of err above, fall through and start softlockup */ 42758687acbSDon Zickus 42858687acbSDon Zickus /* create the watchdog thread */ 42958687acbSDon Zickus if (!p) { 43058687acbSDon Zickus p = kthread_create(watchdog, (void *)(unsigned long)cpu, "watchdog/%d", cpu); 43158687acbSDon Zickus if (IS_ERR(p)) { 43258687acbSDon Zickus printk(KERN_ERR "softlockup watchdog for %i failed\n", cpu); 433f99a9933SDon Zickus if (!err) 434f99a9933SDon Zickus /* if hardlockup hasn't already set this */ 435f99a9933SDon Zickus err = PTR_ERR(p); 436f99a9933SDon Zickus goto out; 43758687acbSDon Zickus } 43858687acbSDon Zickus kthread_bind(p, cpu); 43958687acbSDon Zickus per_cpu(watchdog_touch_ts, cpu) = 0; 44058687acbSDon Zickus per_cpu(softlockup_watchdog, cpu) = p; 44158687acbSDon Zickus wake_up_process(p); 44258687acbSDon Zickus } 44358687acbSDon Zickus 444f99a9933SDon Zickus out: 445f99a9933SDon Zickus return err; 44658687acbSDon Zickus } 44758687acbSDon Zickus 44858687acbSDon Zickus static void watchdog_disable(int cpu) 44958687acbSDon Zickus { 45058687acbSDon Zickus struct task_struct *p = per_cpu(softlockup_watchdog, cpu); 45158687acbSDon Zickus struct hrtimer *hrtimer = &per_cpu(watchdog_hrtimer, cpu); 45258687acbSDon Zickus 45358687acbSDon Zickus /* 45458687acbSDon Zickus * cancel the timer first to stop incrementing the stats 45558687acbSDon Zickus * and waking up the kthread 45658687acbSDon Zickus */ 45758687acbSDon Zickus hrtimer_cancel(hrtimer); 45858687acbSDon Zickus 45958687acbSDon Zickus /* disable the perf event */ 46058687acbSDon Zickus watchdog_nmi_disable(cpu); 46158687acbSDon Zickus 46258687acbSDon Zickus /* stop the watchdog thread */ 46358687acbSDon Zickus if (p) { 46458687acbSDon Zickus per_cpu(softlockup_watchdog, cpu) = NULL; 46558687acbSDon Zickus kthread_stop(p); 46658687acbSDon Zickus } 46758687acbSDon Zickus } 46858687acbSDon Zickus 46958687acbSDon Zickus static void watchdog_enable_all_cpus(void) 47058687acbSDon Zickus { 47158687acbSDon Zickus int cpu; 47239735766SMarcin Slusarz 47339735766SMarcin Slusarz watchdog_enabled = 0; 47458687acbSDon Zickus 47558687acbSDon Zickus for_each_online_cpu(cpu) 47639735766SMarcin Slusarz if (!watchdog_enable(cpu)) 47739735766SMarcin Slusarz /* if any cpu succeeds, watchdog is considered 47839735766SMarcin Slusarz enabled for the system */ 47939735766SMarcin Slusarz watchdog_enabled = 1; 48058687acbSDon Zickus 48139735766SMarcin Slusarz if (!watchdog_enabled) 48258687acbSDon Zickus printk(KERN_ERR "watchdog: failed to be enabled on some cpus\n"); 48358687acbSDon Zickus 48458687acbSDon Zickus } 48558687acbSDon Zickus 48658687acbSDon Zickus static void watchdog_disable_all_cpus(void) 48758687acbSDon Zickus { 48858687acbSDon Zickus int cpu; 48958687acbSDon Zickus 49058687acbSDon Zickus for_each_online_cpu(cpu) 49158687acbSDon Zickus watchdog_disable(cpu); 49258687acbSDon Zickus 49358687acbSDon Zickus /* if all watchdogs are disabled, then they are disabled for the system */ 49458687acbSDon Zickus watchdog_enabled = 0; 49558687acbSDon Zickus } 49658687acbSDon Zickus 49758687acbSDon Zickus 49858687acbSDon Zickus /* sysctl functions */ 49958687acbSDon Zickus #ifdef CONFIG_SYSCTL 50058687acbSDon Zickus /* 50158687acbSDon Zickus * proc handler for /proc/sys/kernel/nmi_watchdog 50258687acbSDon Zickus */ 50358687acbSDon Zickus 50458687acbSDon Zickus int proc_dowatchdog_enabled(struct ctl_table *table, int write, 50558687acbSDon Zickus void __user *buffer, size_t *length, loff_t *ppos) 50658687acbSDon Zickus { 50758687acbSDon Zickus proc_dointvec(table, write, buffer, length, ppos); 50858687acbSDon Zickus 5099ffdc6c3SMarcin Slusarz if (write) { 51058687acbSDon Zickus if (watchdog_enabled) 51158687acbSDon Zickus watchdog_enable_all_cpus(); 51258687acbSDon Zickus else 51358687acbSDon Zickus watchdog_disable_all_cpus(); 5149ffdc6c3SMarcin Slusarz } 51558687acbSDon Zickus return 0; 51658687acbSDon Zickus } 51758687acbSDon Zickus 51858687acbSDon Zickus int proc_dowatchdog_thresh(struct ctl_table *table, int write, 51958687acbSDon Zickus void __user *buffer, 52058687acbSDon Zickus size_t *lenp, loff_t *ppos) 52158687acbSDon Zickus { 52258687acbSDon Zickus return proc_dointvec_minmax(table, write, buffer, lenp, ppos); 52358687acbSDon Zickus } 52458687acbSDon Zickus #endif /* CONFIG_SYSCTL */ 52558687acbSDon Zickus 52658687acbSDon Zickus 52758687acbSDon Zickus /* 52858687acbSDon Zickus * Create/destroy watchdog threads as CPUs come and go: 52958687acbSDon Zickus */ 53058687acbSDon Zickus static int __cpuinit 53158687acbSDon Zickus cpu_callback(struct notifier_block *nfb, unsigned long action, void *hcpu) 53258687acbSDon Zickus { 53358687acbSDon Zickus int hotcpu = (unsigned long)hcpu; 534eac24335SAkinobu Mita int err = 0; 53558687acbSDon Zickus 53658687acbSDon Zickus switch (action) { 53758687acbSDon Zickus case CPU_UP_PREPARE: 53858687acbSDon Zickus case CPU_UP_PREPARE_FROZEN: 539eac24335SAkinobu Mita err = watchdog_prepare_cpu(hotcpu); 54058687acbSDon Zickus break; 54158687acbSDon Zickus case CPU_ONLINE: 54258687acbSDon Zickus case CPU_ONLINE_FROZEN: 5434135038aSMarcin Slusarz if (watchdog_enabled) 544eac24335SAkinobu Mita err = watchdog_enable(hotcpu); 54558687acbSDon Zickus break; 54658687acbSDon Zickus #ifdef CONFIG_HOTPLUG_CPU 54758687acbSDon Zickus case CPU_UP_CANCELED: 54858687acbSDon Zickus case CPU_UP_CANCELED_FROZEN: 54958687acbSDon Zickus watchdog_disable(hotcpu); 55058687acbSDon Zickus break; 55158687acbSDon Zickus case CPU_DEAD: 55258687acbSDon Zickus case CPU_DEAD_FROZEN: 55358687acbSDon Zickus watchdog_disable(hotcpu); 55458687acbSDon Zickus break; 55558687acbSDon Zickus #endif /* CONFIG_HOTPLUG_CPU */ 55658687acbSDon Zickus } 557f99a9933SDon Zickus 558f99a9933SDon Zickus /* 559f99a9933SDon Zickus * hardlockup and softlockup are not important enough 560f99a9933SDon Zickus * to block cpu bring up. Just always succeed and 561f99a9933SDon Zickus * rely on printk output to flag problems. 562f99a9933SDon Zickus */ 563f99a9933SDon Zickus return NOTIFY_OK; 56458687acbSDon Zickus } 56558687acbSDon Zickus 56658687acbSDon Zickus static struct notifier_block __cpuinitdata cpu_nfb = { 56758687acbSDon Zickus .notifier_call = cpu_callback 56858687acbSDon Zickus }; 56958687acbSDon Zickus 570004417a6SPeter Zijlstra void __init lockup_detector_init(void) 57158687acbSDon Zickus { 57258687acbSDon Zickus void *cpu = (void *)(long)smp_processor_id(); 57358687acbSDon Zickus int err; 57458687acbSDon Zickus 57558687acbSDon Zickus err = cpu_callback(&cpu_nfb, CPU_UP_PREPARE, cpu); 576eac24335SAkinobu Mita WARN_ON(notifier_to_errno(err)); 57758687acbSDon Zickus 57858687acbSDon Zickus cpu_callback(&cpu_nfb, CPU_ONLINE, cpu); 57958687acbSDon Zickus register_cpu_notifier(&cpu_nfb); 58058687acbSDon Zickus 581004417a6SPeter Zijlstra return; 58258687acbSDon Zickus } 583