[mirror_ubuntu-jammy-kernel.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 DEFINE_PER_CPU(struct perf_event *, dead_event);
static struct cpumask dead_events_mask;

static unsigned long hardlockup_allcpu_dumped;
static bool hardlockup_detector_disabled;

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 (atomic_read(&watchdog_park_in_progress) != 0)
		return;

	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;
}

/*
 * People like the simple clean cpu node info on boot.
 * Reduce the watchdog noise by only printing messages
 * that are different from what cpu0 displayed.
 */
static unsigned long firstcpu_err;
static atomic_t watchdog_cpus;

int watchdog_nmi_enable(unsigned int cpu)
{
	struct perf_event_attr *wd_attr;
	struct perf_event *event = per_cpu(watchdog_ev, cpu);
	int firstcpu = 0;

	/* nothing to do if the hard lockup detector is disabled */
	if (!(watchdog_enabled & NMI_WATCHDOG_ENABLED))
		goto out;

	/* A failure disabled the hardlockup detector permanently */
	if (hardlockup_detector_disabled)
		return -ENODEV;

	/* is it already setup and enabled? */
	if (event && event->state > PERF_EVENT_STATE_OFF)
		goto out;

	/* it is setup but not enabled */
	if (event != NULL)
		goto out_enable;

	if (atomic_inc_return(&watchdog_cpus) == 1)
		firstcpu = 1;

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

	/* Try to register using hardware perf events */
	event = perf_event_create_kernel_counter(wd_attr, cpu, NULL, watchdog_overflow_callback, NULL);

	/* save the first cpu's error for future comparision */
	if (firstcpu && IS_ERR(event))
		firstcpu_err = PTR_ERR(event);

	if (!IS_ERR(event)) {
		/* only print for the first cpu initialized */
		if (firstcpu || firstcpu_err)
			pr_info("enabled on all CPUs, permanently consumes one hw-PMU counter.\n");
		goto out_save;
	}

	/* skip displaying the same error again */
	if (!firstcpu && (PTR_ERR(event) == firstcpu_err))
		return PTR_ERR(event);

	/* vary the KERN level based on the returned errno */
	if (PTR_ERR(event) == -EOPNOTSUPP)
		pr_info("disabled (cpu%i): not supported (no LAPIC?)\n", cpu);
	else if (PTR_ERR(event) == -ENOENT)
		pr_warn("disabled (cpu%i): hardware events not enabled\n",
			 cpu);
	else
		pr_err("disabled (cpu%i): unable to create perf event: %ld\n",
			cpu, PTR_ERR(event));

	pr_info("Disabling hard lockup detector permanently\n");
	hardlockup_detector_disabled = true;

	return PTR_ERR(event);

	/* success path */
out_save:
	per_cpu(watchdog_ev, cpu) = event;
out_enable:
	perf_event_enable(per_cpu(watchdog_ev, cpu));
out:
	return 0;
}

/**
 * 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);
		this_cpu_write(watchdog_ev, NULL);
		this_cpu_write(dead_event, event);
		cpumask_set_cpu(smp_processor_id(), &dead_events_mask);

		/* watchdog_nmi_enable() expects this to be zero initially. */
		if (atomic_dec_and_test(&watchdog_cpus))
			firstcpu_err = 0;
	}
}

/**
 * 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(dead_event, cpu);

		per_cpu(dead_event, cpu) = NULL;
		perf_event_release_kernel(event);
	}
	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);
	}
}
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 TG	24	static DEFINE_PER_CPU(struct perf_event *, dead_event);
941154bd TG	25	static struct cpumask dead_events_mask;
73ce0511	26
73ce0511	27	static unsigned long hardlockup_allcpu_dumped;
20d853fd	28	static bool hardlockup_detector_disabled;
73ce0511	29
f2e0cff8	30	void arch_touch_nmi_watchdog(void)
73ce0511 BM	31	{
	32	/*
	33	* Using __raw here because some code paths have
	34	* preemption enabled. If preemption is enabled
	35	* then interrupts should be enabled too, in which
	36	* case we shouldn't have to worry about the watchdog
	37	* going off.
	38	*/
	39	raw_cpu_write(watchdog_nmi_touch, true);
73ce0511	40	}
f2e0cff8	41	EXPORT_SYMBOL(arch_touch_nmi_watchdog);
73ce0511	42
7edaeb68 TG	43	#ifdef CONFIG_HARDLOCKUP_CHECK_TIMESTAMP
	44	static DEFINE_PER_CPU(ktime_t, last_timestamp);
	45	static DEFINE_PER_CPU(unsigned int, nmi_rearmed);
	46	static ktime_t watchdog_hrtimer_sample_threshold __read_mostly;
	47
	48	void watchdog_update_hrtimer_threshold(u64 period)
	49	{
	50	/*
	51	* The hrtimer runs with a period of (watchdog_threshold * 2) / 5
	52	*
	53	* So it runs effectively with 2.5 times the rate of the NMI
	54	* watchdog. That means the hrtimer should fire 2-3 times before
	55	* the NMI watchdog expires. The NMI watchdog on x86 is based on
	56	* unhalted CPU cycles, so if Turbo-Mode is enabled the CPU cycles
	57	* might run way faster than expected and the NMI fires in a
	58	* smaller period than the one deduced from the nominal CPU
	59	* frequency. Depending on the Turbo-Mode factor this might be fast
	60	* enough to get the NMI period smaller than the hrtimer watchdog
	61	* period and trigger false positives.
	62	*
	63	* The sample threshold is used to check in the NMI handler whether
	64	* the minimum time between two NMI samples has elapsed. That
	65	* prevents false positives.
	66	*
	67	* Set this to 4/5 of the actual watchdog threshold period so the
	68	* hrtimer is guaranteed to fire at least once within the real
	69	* watchdog threshold.
	70	*/
	71	watchdog_hrtimer_sample_threshold = period * 2;
	72	}
	73
	74	static bool watchdog_check_timestamp(void)
	75	{
	76	ktime_t delta, now = ktime_get_mono_fast_ns();
	77
	78	delta = now - __this_cpu_read(last_timestamp);
	79	if (delta < watchdog_hrtimer_sample_threshold) {
	80	/*
	81	* If ktime is jiffies based, a stalled timer would prevent
	82	* jiffies from being incremented and the filter would look
	83	* at a stale timestamp and never trigger.
	84	*/
	85	if (__this_cpu_inc_return(nmi_rearmed) < 10)
	86	return false;
	87	}
	88	__this_cpu_write(nmi_rearmed, 0);
	89	__this_cpu_write(last_timestamp, now);
	90	return true;
	91	}
	92	#else
	93	static inline bool watchdog_check_timestamp(void)
	94	{
	95	return true;
	96	}
	97	#endif
	98
73ce0511 BM	99	static struct perf_event_attr wd_hw_attr = {
	100	.type = PERF_TYPE_HARDWARE,
	101	.config = PERF_COUNT_HW_CPU_CYCLES,
	102	.size = sizeof(struct perf_event_attr),
	103	.pinned = 1,
	104	.disabled = 1,
	105	};
	106
	107	/* Callback function for perf event subsystem */
	108	static void watchdog_overflow_callback(struct perf_event *event,
	109	struct perf_sample_data *data,
	110	struct pt_regs *regs)
	111	{
	112	/* Ensure the watchdog never gets throttled */
	113	event->hw.interrupts = 0;
	114
b94f5118 DZ	115	if (atomic_read(&watchdog_park_in_progress) != 0)
	116	return;
	117
73ce0511 BM	118	if (__this_cpu_read(watchdog_nmi_touch) == true) {
	119	__this_cpu_write(watchdog_nmi_touch, false);
	120	return;
	121	}
	122
7edaeb68 TG	123	if (!watchdog_check_timestamp())
	124	return;
	125
73ce0511 BM	126	/* check for a hardlockup
	127	* This is done by making sure our timer interrupt
	128	* is incrementing. The timer interrupt should have
	129	* fired multiple times before we overflow'd. If it hasn't
	130	* then this is a good indication the cpu is stuck
	131	*/
	132	if (is_hardlockup()) {
	133	int this_cpu = smp_processor_id();
	134
	135	/* only print hardlockups once */
	136	if (__this_cpu_read(hard_watchdog_warn) == true)
	137	return;
	138
	139	pr_emerg("Watchdog detected hard LOCKUP on cpu %d", this_cpu);
	140	print_modules();
	141	print_irqtrace_events(current);
	142	if (regs)
	143	show_regs(regs);
	144	else
	145	dump_stack();
	146
	147	/*
	148	* Perform all-CPU dump only once to avoid multiple hardlockups
	149	* generating interleaving traces
	150	*/
	151	if (sysctl_hardlockup_all_cpu_backtrace &&
	152	!test_and_set_bit(0, &hardlockup_allcpu_dumped))
	153	trigger_allbutself_cpu_backtrace();
	154
	155	if (hardlockup_panic)
	156	nmi_panic(regs, "Hard LOCKUP");
	157
	158	__this_cpu_write(hard_watchdog_warn, true);
	159	return;
	160	}
	161
	162	__this_cpu_write(hard_watchdog_warn, false);
	163	return;
	164	}
	165
	166	/*
	167	* People like the simple clean cpu node info on boot.
	168	* Reduce the watchdog noise by only printing messages
	169	* that are different from what cpu0 displayed.
	170	*/
8dcde9de PB	171	static unsigned long firstcpu_err;
8dcde9de PB	172	static atomic_t watchdog_cpus;
73ce0511 BM	173
	174	int watchdog_nmi_enable(unsigned int cpu)
	175	{
	176	struct perf_event_attr *wd_attr;
	177	struct perf_event *event = per_cpu(watchdog_ev, cpu);
8dcde9de	178	int firstcpu = 0;
73ce0511 BM	179
	180	/* nothing to do if the hard lockup detector is disabled */
	181	if (!(watchdog_enabled & NMI_WATCHDOG_ENABLED))
	182	goto out;
	183
20d853fd TG	184	/* A failure disabled the hardlockup detector permanently */
	185	if (hardlockup_detector_disabled)
	186	return -ENODEV;
	187
73ce0511 BM	188	/* is it already setup and enabled? */
	189	if (event && event->state > PERF_EVENT_STATE_OFF)
	190	goto out;
	191
	192	/* it is setup but not enabled */
	193	if (event != NULL)
	194	goto out_enable;
	195
8dcde9de PB	196	if (atomic_inc_return(&watchdog_cpus) == 1)
	197	firstcpu = 1;
	198
73ce0511 BM	199	wd_attr = &wd_hw_attr;
	200	wd_attr->sample_period = hw_nmi_get_sample_period(watchdog_thresh);
	201
	202	/* Try to register using hardware perf events */
	203	event = perf_event_create_kernel_counter(wd_attr, cpu, NULL, watchdog_overflow_callback, NULL);
	204
8dcde9de PB	205	/* save the first cpu's error for future comparision */
	206	if (firstcpu && IS_ERR(event))
	207	firstcpu_err = PTR_ERR(event);
73ce0511 BM	208
73ce0511 BM	209	if (!IS_ERR(event)) {
8dcde9de PB	210	/* only print for the first cpu initialized */
8dcde9de PB	211	if (firstcpu \|\| firstcpu_err)
73ce0511 BM	212	pr_info("enabled on all CPUs, permanently consumes one hw-PMU counter.\n");
	213	goto out_save;
	214	}
	215
73ce0511	216	/* skip displaying the same error again */
8dcde9de	217	if (!firstcpu && (PTR_ERR(event) == firstcpu_err))
73ce0511 BM	218	return PTR_ERR(event);
	219
	220	/* vary the KERN level based on the returned errno */
	221	if (PTR_ERR(event) == -EOPNOTSUPP)
	222	pr_info("disabled (cpu%i): not supported (no LAPIC?)\n", cpu);
	223	else if (PTR_ERR(event) == -ENOENT)
	224	pr_warn("disabled (cpu%i): hardware events not enabled\n",
	225	cpu);
	226	else
	227	pr_err("disabled (cpu%i): unable to create perf event: %ld\n",
	228	cpu, PTR_ERR(event));
	229
20d853fd TG	230	pr_info("Disabling hard lockup detector permanently\n");
20d853fd TG	231	hardlockup_detector_disabled = true;
73ce0511 BM	232
	233	return PTR_ERR(event);
	234
	235	/* success path */
	236	out_save:
	237	per_cpu(watchdog_ev, cpu) = event;
	238	out_enable:
	239	perf_event_enable(per_cpu(watchdog_ev, cpu));
	240	out:
	241	return 0;
	242	}
	243
941154bd TG	244	/**
	245	* hardlockup_detector_perf_disable - Disable the local event
	246	*/
	247	void hardlockup_detector_perf_disable(void)
73ce0511	248	{
941154bd	249	struct perf_event *event = this_cpu_read(watchdog_ev);
73ce0511 BM	250
	251	if (event) {
	252	perf_event_disable(event);
941154bd TG	253	this_cpu_write(watchdog_ev, NULL);
	254	this_cpu_write(dead_event, event);
	255	cpumask_set_cpu(smp_processor_id(), &dead_events_mask);
8dcde9de	256
73ce0511	257	/* watchdog_nmi_enable() expects this to be zero initially. */
8dcde9de PB	258	if (atomic_dec_and_test(&watchdog_cpus))
8dcde9de PB	259	firstcpu_err = 0;
73ce0511 BM	260	}
73ce0511 BM	261	}
d0b6e0a8	262
941154bd TG	263	/**
	264	* hardlockup_detector_perf_cleanup - Cleanup disabled events and destroy them
	265	*
	266	* Called from lockup_detector_cleanup(). Serialized by the caller.
	267	*/
	268	void hardlockup_detector_perf_cleanup(void)
	269	{
	270	int cpu;
	271
	272	for_each_cpu(cpu, &dead_events_mask) {
	273	struct perf_event *event = per_cpu(dead_event, cpu);
	274
	275	per_cpu(dead_event, cpu) = NULL;
	276	perf_event_release_kernel(event);
	277	}
	278	cpumask_clear(&dead_events_mask);
	279	}
	280
d0b6e0a8 PZ	281	/**
	282	* hardlockup_detector_perf_stop - Globally stop watchdog events
	283	*
	284	* Special interface for x86 to handle the perf HT bug.
	285	*/
	286	void __init hardlockup_detector_perf_stop(void)
	287	{
	288	int cpu;
	289
	290	lockdep_assert_cpus_held();
	291
	292	for_each_online_cpu(cpu) {
	293	struct perf_event *event = per_cpu(watchdog_ev, cpu);
	294
	295	if (event)
	296	perf_event_disable(event);
	297	}
	298	}
	299
	300	/**
	301	* hardlockup_detector_perf_restart - Globally restart watchdog events
	302	*
	303	* Special interface for x86 to handle the perf HT bug.
	304	*/
	305	void __init hardlockup_detector_perf_restart(void)
	306	{
	307	int cpu;
	308
	309	lockdep_assert_cpus_held();
	310
	311	if (!(watchdog_enabled & NMI_WATCHDOG_ENABLED))
	312	return;
	313
	314	for_each_online_cpu(cpu) {
	315	struct perf_event *event = per_cpu(watchdog_ev, cpu);
	316
	317	if (event)
	318	perf_event_enable(event);
	319	}
	320	}