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

// SPDX-License-Identifier: GPL-2.0
/*
 * 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/atomic.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 atomic_t watchdog_cpus = ATOMIC_INIT(0);

notrace 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\n",
			 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_debug("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;

	/* use original value for check */
	if (!atomic_fetch_inc(&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);
		this_cpu_write(watchdog_ev, NULL);
		this_cpu_write(dead_event, event);
		cpumask_set_cpu(smp_processor_id(), &dead_events_mask);
		atomic_dec(&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(dead_event, cpu);

		/*
		 * Required because for_each_cpu() reports  unconditionally
		 * CPU0 as set on UP kernels. Sigh.
		 */
		if (event)
			perf_event_release_kernel(event);
		per_cpu(dead_event, 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
b2441318	1	// SPDX-License-Identifier: GPL-2.0
73ce0511 BM	2	/*
	3	* Detect hard lockups on a system
	4	*
	5	* started by Don Zickus, Copyright (C) 2010 Red Hat, Inc.
	6	*
	7	* Note: Most of this code is borrowed heavily from the original softlockup
	8	* detector, so thanks to Ingo for the initial implementation.
	9	* Some chunks also taken from the old x86-specific nmi watchdog code, thanks
	10	* to those contributors as well.
	11	*/
	12
	13	#define pr_fmt(fmt) "NMI watchdog: " fmt
	14
	15	#include <linux/nmi.h>
42f930da	16	#include <linux/atomic.h>
73ce0511	17	#include <linux/module.h>
b17b0153 IM	18	#include <linux/sched/debug.h>
b17b0153 IM	19
73ce0511 BM	20	#include <asm/irq_regs.h>
	21	#include <linux/perf_event.h>
	22
	23	static DEFINE_PER_CPU(bool, hard_watchdog_warn);
	24	static DEFINE_PER_CPU(bool, watchdog_nmi_touch);
	25	static DEFINE_PER_CPU(struct perf_event *, watchdog_ev);
9c388a5e	26	static DEFINE_PER_CPU(struct perf_event *, dead_event);
941154bd	27	static struct cpumask dead_events_mask;
73ce0511	28
73ce0511	29	static unsigned long hardlockup_allcpu_dumped;
42f930da	30	static atomic_t watchdog_cpus = ATOMIC_INIT(0);
73ce0511	31
cb9d7fd5	32	notrace void arch_touch_nmi_watchdog(void)
73ce0511 BM	33	{
	34	/*
	35	* Using __raw here because some code paths have
	36	* preemption enabled. If preemption is enabled
	37	* then interrupts should be enabled too, in which
	38	* case we shouldn't have to worry about the watchdog
	39	* going off.
	40	*/
	41	raw_cpu_write(watchdog_nmi_touch, true);
73ce0511	42	}
f2e0cff8	43	EXPORT_SYMBOL(arch_touch_nmi_watchdog);
73ce0511	44
7edaeb68 TG	45	#ifdef CONFIG_HARDLOCKUP_CHECK_TIMESTAMP
	46	static DEFINE_PER_CPU(ktime_t, last_timestamp);
	47	static DEFINE_PER_CPU(unsigned int, nmi_rearmed);
	48	static ktime_t watchdog_hrtimer_sample_threshold __read_mostly;
	49
	50	void watchdog_update_hrtimer_threshold(u64 period)
	51	{
	52	/*
	53	* The hrtimer runs with a period of (watchdog_threshold * 2) / 5
	54	*
	55	* So it runs effectively with 2.5 times the rate of the NMI
	56	* watchdog. That means the hrtimer should fire 2-3 times before
	57	* the NMI watchdog expires. The NMI watchdog on x86 is based on
	58	* unhalted CPU cycles, so if Turbo-Mode is enabled the CPU cycles
	59	* might run way faster than expected and the NMI fires in a
	60	* smaller period than the one deduced from the nominal CPU
	61	* frequency. Depending on the Turbo-Mode factor this might be fast
	62	* enough to get the NMI period smaller than the hrtimer watchdog
	63	* period and trigger false positives.
	64	*
	65	* The sample threshold is used to check in the NMI handler whether
	66	* the minimum time between two NMI samples has elapsed. That
	67	* prevents false positives.
	68	*
	69	* Set this to 4/5 of the actual watchdog threshold period so the
	70	* hrtimer is guaranteed to fire at least once within the real
	71	* watchdog threshold.
	72	*/
	73	watchdog_hrtimer_sample_threshold = period * 2;
	74	}
	75
	76	static bool watchdog_check_timestamp(void)
	77	{
	78	ktime_t delta, now = ktime_get_mono_fast_ns();
	79
	80	delta = now - __this_cpu_read(last_timestamp);
	81	if (delta < watchdog_hrtimer_sample_threshold) {
	82	/*
	83	* If ktime is jiffies based, a stalled timer would prevent
	84	* jiffies from being incremented and the filter would look
	85	* at a stale timestamp and never trigger.
	86	*/
	87	if (__this_cpu_inc_return(nmi_rearmed) < 10)
	88	return false;
	89	}
	90	__this_cpu_write(nmi_rearmed, 0);
	91	__this_cpu_write(last_timestamp, now);
	92	return true;
	93	}
	94	#else
	95	static inline bool watchdog_check_timestamp(void)
	96	{
	97	return true;
	98	}
	99	#endif
	100
73ce0511 BM	101	static struct perf_event_attr wd_hw_attr = {
	102	.type = PERF_TYPE_HARDWARE,
	103	.config = PERF_COUNT_HW_CPU_CYCLES,
	104	.size = sizeof(struct perf_event_attr),
	105	.pinned = 1,
	106	.disabled = 1,
	107	};
	108
	109	/* Callback function for perf event subsystem */
	110	static void watchdog_overflow_callback(struct perf_event *event,
01f0a027 TG	111	struct perf_sample_data *data,
01f0a027 TG	112	struct pt_regs *regs)
73ce0511 BM	113	{
	114	/* Ensure the watchdog never gets throttled */
	115	event->hw.interrupts = 0;
	116
	117	if (__this_cpu_read(watchdog_nmi_touch) == true) {
	118	__this_cpu_write(watchdog_nmi_touch, false);
	119	return;
	120	}
	121
7edaeb68 TG	122	if (!watchdog_check_timestamp())
	123	return;
	124
73ce0511 BM	125	/* check for a hardlockup
	126	* This is done by making sure our timer interrupt
	127	* is incrementing. The timer interrupt should have
	128	* fired multiple times before we overflow'd. If it hasn't
	129	* then this is a good indication the cpu is stuck
	130	*/
	131	if (is_hardlockup()) {
	132	int this_cpu = smp_processor_id();
	133
	134	/* only print hardlockups once */
	135	if (__this_cpu_read(hard_watchdog_warn) == true)
	136	return;
	137
8f4a8c12 SS	138	pr_emerg("Watchdog detected hard LOCKUP on cpu %d\n",
8f4a8c12 SS	139	this_cpu);
73ce0511 BM	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
178b9f7a TG	166	static int hardlockup_detector_event_create(void)
	167	{
	168	unsigned int cpu = smp_processor_id();
	169	struct perf_event_attr *wd_attr;
	170	struct perf_event *evt;
	171
	172	wd_attr = &wd_hw_attr;
	173	wd_attr->sample_period = hw_nmi_get_sample_period(watchdog_thresh);
	174
	175	/* Try to register using hardware perf events */
	176	evt = perf_event_create_kernel_counter(wd_attr, cpu, NULL,
	177	watchdog_overflow_callback, NULL);
	178	if (IS_ERR(evt)) {
1b6266eb SK	179	pr_debug("Perf event create on CPU %d failed with %ld\n", cpu,
1b6266eb SK	180	PTR_ERR(evt));
178b9f7a TG	181	return PTR_ERR(evt);
	182	}
	183	this_cpu_write(watchdog_ev, evt);
	184	return 0;
	185	}
	186
2a1b8ee4 TG	187	/**
	188	* hardlockup_detector_perf_enable - Enable the local event
	189	*/
	190	void hardlockup_detector_perf_enable(void)
	191	{
	192	if (hardlockup_detector_event_create())
	193	return;
	194
42f930da DZ	195	/* use original value for check */
42f930da DZ	196	if (!atomic_fetch_inc(&watchdog_cpus))
146c9d0e TG	197	pr_info("Enabled. Permanently consumes one hw-PMU counter.\n");
146c9d0e TG	198
2a1b8ee4 TG	199	perf_event_enable(this_cpu_read(watchdog_ev));
	200	}
	201
941154bd TG	202	/**
	203	* hardlockup_detector_perf_disable - Disable the local event
	204	*/
	205	void hardlockup_detector_perf_disable(void)
73ce0511	206	{
941154bd	207	struct perf_event *event = this_cpu_read(watchdog_ev);
73ce0511 BM	208
	209	if (event) {
	210	perf_event_disable(event);
9c388a5e TG	211	this_cpu_write(watchdog_ev, NULL);
9c388a5e TG	212	this_cpu_write(dead_event, event);
941154bd	213	cpumask_set_cpu(smp_processor_id(), &dead_events_mask);
42f930da	214	atomic_dec(&watchdog_cpus);
73ce0511 BM	215	}
73ce0511 BM	216	}
d0b6e0a8	217
941154bd TG	218	/**
	219	* hardlockup_detector_perf_cleanup - Cleanup disabled events and destroy them
	220	*
	221	* Called from lockup_detector_cleanup(). Serialized by the caller.
	222	*/
	223	void hardlockup_detector_perf_cleanup(void)
	224	{
	225	int cpu;
	226
	227	for_each_cpu(cpu, &dead_events_mask) {
9c388a5e	228	struct perf_event *event = per_cpu(dead_event, cpu);
941154bd	229
115ef3b7 TG	230	/*
	231	* Required because for_each_cpu() reports unconditionally
	232	* CPU0 as set on UP kernels. Sigh.
	233	*/
	234	if (event)
	235	perf_event_release_kernel(event);
9c388a5e	236	per_cpu(dead_event, cpu) = NULL;
941154bd TG	237	}
	238	cpumask_clear(&dead_events_mask);
	239	}
	240
d0b6e0a8 PZ	241	/**
	242	* hardlockup_detector_perf_stop - Globally stop watchdog events
	243	*
	244	* Special interface for x86 to handle the perf HT bug.
	245	*/
	246	void __init hardlockup_detector_perf_stop(void)
	247	{
	248	int cpu;
	249
	250	lockdep_assert_cpus_held();
	251
	252	for_each_online_cpu(cpu) {
	253	struct perf_event *event = per_cpu(watchdog_ev, cpu);
	254
	255	if (event)
	256	perf_event_disable(event);
	257	}
	258	}
	259
	260	/**
	261	* hardlockup_detector_perf_restart - Globally restart watchdog events
	262	*
	263	* Special interface for x86 to handle the perf HT bug.
	264	*/
	265	void __init hardlockup_detector_perf_restart(void)
	266	{
	267	int cpu;
	268
	269	lockdep_assert_cpus_held();
	270
	271	if (!(watchdog_enabled & NMI_WATCHDOG_ENABLED))
	272	return;
	273
	274	for_each_online_cpu(cpu) {
	275	struct perf_event *event = per_cpu(watchdog_ev, cpu);
	276
	277	if (event)
	278	perf_event_enable(event);
	279	}
	280	}
178b9f7a TG	281
	282	/**
	283	* hardlockup_detector_perf_init - Probe whether NMI event is available at all
	284	*/
	285	int __init hardlockup_detector_perf_init(void)
	286	{
	287	int ret = hardlockup_detector_event_create();
	288
	289	if (ret) {
77c01d11	290	pr_info("Perf NMI watchdog permanently disabled\n");
178b9f7a TG	291	} else {
	292	perf_event_release_kernel(this_cpu_read(watchdog_ev));
	293	this_cpu_write(watchdog_ev, NULL);
	294	}
	295	return ret;
	296	}