[linux-2.6-block.git] / kernel / watchdog_hld.c

/*
 * Detect hard lockups on a system
 *
 * started by Don Zickus, Copyright (C) 2010 Red Hat, Inc.
 *
 * Note: Most of this code is borrowed heavily from the original softlockup
 * detector, so thanks to Ingo for the initial implementation.
 * Some chunks also taken from the old x86-specific nmi watchdog code, thanks
 * to those contributors as well.
 */

#define pr_fmt(fmt) "NMI watchdog: " fmt

#include <linux/nmi.h>
#include <linux/module.h>
#include <linux/sched/debug.h>

#include <asm/irq_regs.h>
#include <linux/perf_event.h>

static DEFINE_PER_CPU(bool, hard_watchdog_warn);
static DEFINE_PER_CPU(bool, watchdog_nmi_touch);
static DEFINE_PER_CPU(struct perf_event *, watchdog_ev);
static struct cpumask dead_events_mask;

static unsigned long hardlockup_allcpu_dumped;
static unsigned int watchdog_cpus;

void arch_touch_nmi_watchdog(void)
{
	/*
	 * Using __raw here because some code paths have
	 * preemption enabled.  If preemption is enabled
	 * then interrupts should be enabled too, in which
	 * case we shouldn't have to worry about the watchdog
	 * going off.
	 */
	raw_cpu_write(watchdog_nmi_touch, true);
}
EXPORT_SYMBOL(arch_touch_nmi_watchdog);

#ifdef CONFIG_HARDLOCKUP_CHECK_TIMESTAMP
static DEFINE_PER_CPU(ktime_t, last_timestamp);
static DEFINE_PER_CPU(unsigned int, nmi_rearmed);
static ktime_t watchdog_hrtimer_sample_threshold __read_mostly;

void watchdog_update_hrtimer_threshold(u64 period)
{
	/*
	 * The hrtimer runs with a period of (watchdog_threshold * 2) / 5
	 *
	 * So it runs effectively with 2.5 times the rate of the NMI
	 * watchdog. That means the hrtimer should fire 2-3 times before
	 * the NMI watchdog expires. The NMI watchdog on x86 is based on
	 * unhalted CPU cycles, so if Turbo-Mode is enabled the CPU cycles
	 * might run way faster than expected and the NMI fires in a
	 * smaller period than the one deduced from the nominal CPU
	 * frequency. Depending on the Turbo-Mode factor this might be fast
	 * enough to get the NMI period smaller than the hrtimer watchdog
	 * period and trigger false positives.
	 *
	 * The sample threshold is used to check in the NMI handler whether
	 * the minimum time between two NMI samples has elapsed. That
	 * prevents false positives.
	 *
	 * Set this to 4/5 of the actual watchdog threshold period so the
	 * hrtimer is guaranteed to fire at least once within the real
	 * watchdog threshold.
	 */
	watchdog_hrtimer_sample_threshold = period * 2;
}

static bool watchdog_check_timestamp(void)
{
	ktime_t delta, now = ktime_get_mono_fast_ns();

	delta = now - __this_cpu_read(last_timestamp);
	if (delta < watchdog_hrtimer_sample_threshold) {
		/*
		 * If ktime is jiffies based, a stalled timer would prevent
		 * jiffies from being incremented and the filter would look
		 * at a stale timestamp and never trigger.
		 */
		if (__this_cpu_inc_return(nmi_rearmed) < 10)
			return false;
	}
	__this_cpu_write(nmi_rearmed, 0);
	__this_cpu_write(last_timestamp, now);
	return true;
}
#else
static inline bool watchdog_check_timestamp(void)
{
	return true;
}
#endif

static struct perf_event_attr wd_hw_attr = {
	.type		= PERF_TYPE_HARDWARE,
	.config		= PERF_COUNT_HW_CPU_CYCLES,
	.size		= sizeof(struct perf_event_attr),
	.pinned		= 1,
	.disabled	= 1,
};

/* Callback function for perf event subsystem */
static void watchdog_overflow_callback(struct perf_event *event,
				       struct perf_sample_data *data,
				       struct pt_regs *regs)
{
	/* Ensure the watchdog never gets throttled */
	event->hw.interrupts = 0;

	if (__this_cpu_read(watchdog_nmi_touch) == true) {
		__this_cpu_write(watchdog_nmi_touch, false);
		return;
	}

	if (!watchdog_check_timestamp())
		return;

	/* check for a hardlockup
	 * This is done by making sure our timer interrupt
	 * is incrementing.  The timer interrupt should have
	 * fired multiple times before we overflow'd.  If it hasn't
	 * then this is a good indication the cpu is stuck
	 */
	if (is_hardlockup()) {
		int this_cpu = smp_processor_id();

		/* only print hardlockups once */
		if (__this_cpu_read(hard_watchdog_warn) == true)
			return;

		pr_emerg("Watchdog detected hard LOCKUP on cpu %d", this_cpu);
		print_modules();
		print_irqtrace_events(current);
		if (regs)
			show_regs(regs);
		else
			dump_stack();

		/*
		 * Perform all-CPU dump only once to avoid multiple hardlockups
		 * generating interleaving traces
		 */
		if (sysctl_hardlockup_all_cpu_backtrace &&
				!test_and_set_bit(0, &hardlockup_allcpu_dumped))
			trigger_allbutself_cpu_backtrace();

		if (hardlockup_panic)
			nmi_panic(regs, "Hard LOCKUP");

		__this_cpu_write(hard_watchdog_warn, true);
		return;
	}

	__this_cpu_write(hard_watchdog_warn, false);
	return;
}

static int hardlockup_detector_event_create(void)
{
	unsigned int cpu = smp_processor_id();
	struct perf_event_attr *wd_attr;
	struct perf_event *evt;

	wd_attr = &wd_hw_attr;
	wd_attr->sample_period = hw_nmi_get_sample_period(watchdog_thresh);

	/* Try to register using hardware perf events */
	evt = perf_event_create_kernel_counter(wd_attr, cpu, NULL,
					       watchdog_overflow_callback, NULL);
	if (IS_ERR(evt)) {
		pr_info("Perf event create on CPU %d failed with %ld\n", cpu,
			PTR_ERR(evt));
		return PTR_ERR(evt);
	}
	this_cpu_write(watchdog_ev, evt);
	return 0;
}

/**
 * hardlockup_detector_perf_enable - Enable the local event
 */
void hardlockup_detector_perf_enable(void)
{
	if (hardlockup_detector_event_create())
		return;

	if (!watchdog_cpus++)
		pr_info("Enabled. Permanently consumes one hw-PMU counter.\n");

	perf_event_enable(this_cpu_read(watchdog_ev));
}

/**
 * hardlockup_detector_perf_disable - Disable the local event
 */
void hardlockup_detector_perf_disable(void)
{
	struct perf_event *event = this_cpu_read(watchdog_ev);

	if (event) {
		perf_event_disable(event);
		cpumask_set_cpu(smp_processor_id(), &dead_events_mask);
		watchdog_cpus--;
	}
}

/**
 * hardlockup_detector_perf_cleanup - Cleanup disabled events and destroy them
 *
 * Called from lockup_detector_cleanup(). Serialized by the caller.
 */
void hardlockup_detector_perf_cleanup(void)
{
	int cpu;

	for_each_cpu(cpu, &dead_events_mask) {
		struct perf_event *event = per_cpu(watchdog_ev, cpu);

		/*
		 * Required because for_each_cpu() reports  unconditionally
		 * CPU0 as set on UP kernels. Sigh.
		 */
		if (event)
			perf_event_release_kernel(event);
		per_cpu(watchdog_ev, cpu) = NULL;
	}
	cpumask_clear(&dead_events_mask);
}

/**
 * hardlockup_detector_perf_stop - Globally stop watchdog events
 *
 * Special interface for x86 to handle the perf HT bug.
 */
void __init hardlockup_detector_perf_stop(void)
{
	int cpu;

	lockdep_assert_cpus_held();

	for_each_online_cpu(cpu) {
		struct perf_event *event = per_cpu(watchdog_ev, cpu);

		if (event)
			perf_event_disable(event);
	}
}

/**
 * hardlockup_detector_perf_restart - Globally restart watchdog events
 *
 * Special interface for x86 to handle the perf HT bug.
 */
void __init hardlockup_detector_perf_restart(void)
{
	int cpu;

	lockdep_assert_cpus_held();

	if (!(watchdog_enabled & NMI_WATCHDOG_ENABLED))
		return;

	for_each_online_cpu(cpu) {
		struct perf_event *event = per_cpu(watchdog_ev, cpu);

		if (event)
			perf_event_enable(event);
	}
}

/**
 * hardlockup_detector_perf_init - Probe whether NMI event is available at all
 */
int __init hardlockup_detector_perf_init(void)
{
	int ret = hardlockup_detector_event_create();

	if (ret) {
		pr_info("Perf NMI watchdog permanently disabled\n");
	} else {
		perf_event_release_kernel(this_cpu_read(watchdog_ev));
		this_cpu_write(watchdog_ev, NULL);
	}
	return ret;
}
Commit	Line	Data
73ce0511 BM	1	/*
	2	* Detect hard lockups on a system
	3	*
	4	* started by Don Zickus, Copyright (C) 2010 Red Hat, Inc.
	5	*
	6	* Note: Most of this code is borrowed heavily from the original softlockup
	7	* detector, so thanks to Ingo for the initial implementation.
	8	* Some chunks also taken from the old x86-specific nmi watchdog code, thanks
	9	* to those contributors as well.
	10	*/
	11
	12	#define pr_fmt(fmt) "NMI watchdog: " fmt
	13
	14	#include <linux/nmi.h>
	15	#include <linux/module.h>
b17b0153 IM	16	#include <linux/sched/debug.h>
b17b0153 IM	17
73ce0511 BM	18	#include <asm/irq_regs.h>
	19	#include <linux/perf_event.h>
	20
	21	static DEFINE_PER_CPU(bool, hard_watchdog_warn);
	22	static DEFINE_PER_CPU(bool, watchdog_nmi_touch);
	23	static DEFINE_PER_CPU(struct perf_event *, watchdog_ev);
941154bd	24	static struct cpumask dead_events_mask;
73ce0511	25
73ce0511	26	static unsigned long hardlockup_allcpu_dumped;
146c9d0e	27	static unsigned int watchdog_cpus;
73ce0511	28
f2e0cff8	29	void arch_touch_nmi_watchdog(void)
73ce0511 BM	30	{
	31	/*
	32	* Using __raw here because some code paths have
	33	* preemption enabled. If preemption is enabled
	34	* then interrupts should be enabled too, in which
	35	* case we shouldn't have to worry about the watchdog
	36	* going off.
	37	*/
	38	raw_cpu_write(watchdog_nmi_touch, true);
73ce0511	39	}
f2e0cff8	40	EXPORT_SYMBOL(arch_touch_nmi_watchdog);
73ce0511	41
7edaeb68 TG	42	#ifdef CONFIG_HARDLOCKUP_CHECK_TIMESTAMP
	43	static DEFINE_PER_CPU(ktime_t, last_timestamp);
	44	static DEFINE_PER_CPU(unsigned int, nmi_rearmed);
	45	static ktime_t watchdog_hrtimer_sample_threshold __read_mostly;
	46
	47	void watchdog_update_hrtimer_threshold(u64 period)
	48	{
	49	/*
	50	* The hrtimer runs with a period of (watchdog_threshold * 2) / 5
	51	*
	52	* So it runs effectively with 2.5 times the rate of the NMI
	53	* watchdog. That means the hrtimer should fire 2-3 times before
	54	* the NMI watchdog expires. The NMI watchdog on x86 is based on
	55	* unhalted CPU cycles, so if Turbo-Mode is enabled the CPU cycles
	56	* might run way faster than expected and the NMI fires in a
	57	* smaller period than the one deduced from the nominal CPU
	58	* frequency. Depending on the Turbo-Mode factor this might be fast
	59	* enough to get the NMI period smaller than the hrtimer watchdog
	60	* period and trigger false positives.
	61	*
	62	* The sample threshold is used to check in the NMI handler whether
	63	* the minimum time between two NMI samples has elapsed. That
	64	* prevents false positives.
	65	*
	66	* Set this to 4/5 of the actual watchdog threshold period so the
	67	* hrtimer is guaranteed to fire at least once within the real
	68	* watchdog threshold.
	69	*/
	70	watchdog_hrtimer_sample_threshold = period * 2;
	71	}
	72
	73	static bool watchdog_check_timestamp(void)
	74	{
	75	ktime_t delta, now = ktime_get_mono_fast_ns();
	76
	77	delta = now - __this_cpu_read(last_timestamp);
	78	if (delta < watchdog_hrtimer_sample_threshold) {
	79	/*
	80	* If ktime is jiffies based, a stalled timer would prevent
	81	* jiffies from being incremented and the filter would look
	82	* at a stale timestamp and never trigger.
	83	*/
	84	if (__this_cpu_inc_return(nmi_rearmed) < 10)
	85	return false;
	86	}
	87	__this_cpu_write(nmi_rearmed, 0);
	88	__this_cpu_write(last_timestamp, now);
	89	return true;
	90	}
	91	#else
	92	static inline bool watchdog_check_timestamp(void)
	93	{
	94	return true;
	95	}
	96	#endif
	97
73ce0511 BM	98	static struct perf_event_attr wd_hw_attr = {
	99	.type = PERF_TYPE_HARDWARE,
	100	.config = PERF_COUNT_HW_CPU_CYCLES,
	101	.size = sizeof(struct perf_event_attr),
	102	.pinned = 1,
	103	.disabled = 1,
	104	};
	105
	106	/* Callback function for perf event subsystem */
	107	static void watchdog_overflow_callback(struct perf_event *event,
01f0a027 TG	108	struct perf_sample_data *data,
01f0a027 TG	109	struct pt_regs *regs)
73ce0511 BM	110	{
	111	/* Ensure the watchdog never gets throttled */
	112	event->hw.interrupts = 0;
	113
	114	if (__this_cpu_read(watchdog_nmi_touch) == true) {
	115	__this_cpu_write(watchdog_nmi_touch, false);
	116	return;
	117	}
	118
7edaeb68 TG	119	if (!watchdog_check_timestamp())
	120	return;
	121
73ce0511 BM	122	/* check for a hardlockup
	123	* This is done by making sure our timer interrupt
	124	* is incrementing. The timer interrupt should have
	125	* fired multiple times before we overflow'd. If it hasn't
	126	* then this is a good indication the cpu is stuck
	127	*/
	128	if (is_hardlockup()) {
	129	int this_cpu = smp_processor_id();
	130
	131	/* only print hardlockups once */
	132	if (__this_cpu_read(hard_watchdog_warn) == true)
	133	return;
	134
	135	pr_emerg("Watchdog detected hard LOCKUP on cpu %d", this_cpu);
	136	print_modules();
	137	print_irqtrace_events(current);
	138	if (regs)
	139	show_regs(regs);
	140	else
	141	dump_stack();
	142
	143	/*
	144	* Perform all-CPU dump only once to avoid multiple hardlockups
	145	* generating interleaving traces
	146	*/
	147	if (sysctl_hardlockup_all_cpu_backtrace &&
	148	!test_and_set_bit(0, &hardlockup_allcpu_dumped))
	149	trigger_allbutself_cpu_backtrace();
	150
	151	if (hardlockup_panic)
	152	nmi_panic(regs, "Hard LOCKUP");
	153
	154	__this_cpu_write(hard_watchdog_warn, true);
	155	return;
	156	}
	157
	158	__this_cpu_write(hard_watchdog_warn, false);
	159	return;
	160	}
	161
178b9f7a TG	162	static int hardlockup_detector_event_create(void)
	163	{
	164	unsigned int cpu = smp_processor_id();
	165	struct perf_event_attr *wd_attr;
	166	struct perf_event *evt;
	167
	168	wd_attr = &wd_hw_attr;
	169	wd_attr->sample_period = hw_nmi_get_sample_period(watchdog_thresh);
	170
	171	/* Try to register using hardware perf events */
	172	evt = perf_event_create_kernel_counter(wd_attr, cpu, NULL,
	173	watchdog_overflow_callback, NULL);
	174	if (IS_ERR(evt)) {
	175	pr_info("Perf event create on CPU %d failed with %ld\n", cpu,
	176	PTR_ERR(evt));
	177	return PTR_ERR(evt);
	178	}
	179	this_cpu_write(watchdog_ev, evt);
	180	return 0;
	181	}
	182
2a1b8ee4 TG	183	/**
	184	* hardlockup_detector_perf_enable - Enable the local event
	185	*/
	186	void hardlockup_detector_perf_enable(void)
	187	{
	188	if (hardlockup_detector_event_create())
	189	return;
	190
146c9d0e TG	191	if (!watchdog_cpus++)
	192	pr_info("Enabled. Permanently consumes one hw-PMU counter.\n");
	193
2a1b8ee4 TG	194	perf_event_enable(this_cpu_read(watchdog_ev));
	195	}
	196
941154bd TG	197	/**
	198	* hardlockup_detector_perf_disable - Disable the local event
	199	*/
	200	void hardlockup_detector_perf_disable(void)
73ce0511	201	{
941154bd	202	struct perf_event *event = this_cpu_read(watchdog_ev);
73ce0511 BM	203
	204	if (event) {
	205	perf_event_disable(event);
941154bd	206	cpumask_set_cpu(smp_processor_id(), &dead_events_mask);
146c9d0e	207	watchdog_cpus--;
73ce0511 BM	208	}
73ce0511 BM	209	}
d0b6e0a8	210
941154bd TG	211	/**
	212	* hardlockup_detector_perf_cleanup - Cleanup disabled events and destroy them
	213	*
	214	* Called from lockup_detector_cleanup(). Serialized by the caller.
	215	*/
	216	void hardlockup_detector_perf_cleanup(void)
	217	{
	218	int cpu;
	219
	220	for_each_cpu(cpu, &dead_events_mask) {
a33d4484	221	struct perf_event *event = per_cpu(watchdog_ev, cpu);
941154bd	222
115ef3b7 TG	223	/*
	224	* Required because for_each_cpu() reports unconditionally
	225	* CPU0 as set on UP kernels. Sigh.
	226	*/
	227	if (event)
	228	perf_event_release_kernel(event);
a33d4484	229	per_cpu(watchdog_ev, cpu) = NULL;
941154bd TG	230	}
	231	cpumask_clear(&dead_events_mask);
	232	}
	233
d0b6e0a8 PZ	234	/**
	235	* hardlockup_detector_perf_stop - Globally stop watchdog events
	236	*
	237	* Special interface for x86 to handle the perf HT bug.
	238	*/
	239	void __init hardlockup_detector_perf_stop(void)
	240	{
	241	int cpu;
	242
	243	lockdep_assert_cpus_held();
	244
	245	for_each_online_cpu(cpu) {
	246	struct perf_event *event = per_cpu(watchdog_ev, cpu);
	247
	248	if (event)
	249	perf_event_disable(event);
	250	}
	251	}
	252
	253	/**
	254	* hardlockup_detector_perf_restart - Globally restart watchdog events
	255	*
	256	* Special interface for x86 to handle the perf HT bug.
	257	*/
	258	void __init hardlockup_detector_perf_restart(void)
	259	{
	260	int cpu;
	261
	262	lockdep_assert_cpus_held();
	263
	264	if (!(watchdog_enabled & NMI_WATCHDOG_ENABLED))
	265	return;
	266
	267	for_each_online_cpu(cpu) {
	268	struct perf_event *event = per_cpu(watchdog_ev, cpu);
	269
	270	if (event)
	271	perf_event_enable(event);
	272	}
	273	}
178b9f7a TG	274
	275	/**
	276	* hardlockup_detector_perf_init - Probe whether NMI event is available at all
	277	*/
	278	int __init hardlockup_detector_perf_init(void)
	279	{
	280	int ret = hardlockup_detector_event_create();
	281
	282	if (ret) {
77c01d11	283	pr_info("Perf NMI watchdog permanently disabled\n");
178b9f7a TG	284	} else {
	285	perf_event_release_kernel(this_cpu_read(watchdog_ev));
	286	this_cpu_write(watchdog_ev, NULL);
	287	}
	288	return ret;
	289	}