[linux.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);

/* boot commands */
/*
 * Should we panic when a soft-lockup or hard-lockup occurs:
 */
unsigned int __read_mostly hardlockup_panic =
			CONFIG_BOOTPARAM_HARDLOCKUP_PANIC_VALUE;
static unsigned long hardlockup_allcpu_dumped;
/*
 * We may not want to enable hard lockup detection by default in all cases,
 * for example when running the kernel as a guest on a hypervisor. In these
 * cases this function can be called to disable hard lockup detection. This
 * function should only be executed once by the boot processor before the
 * kernel command line parameters are parsed, because otherwise it is not
 * possible to override this in hardlockup_panic_setup().
 */
void hardlockup_detector_disable(void)
{
	watchdog_enabled &= ~NMI_WATCHDOG_ENABLED;
}

static int __init hardlockup_panic_setup(char *str)
{
	if (!strncmp(str, "panic", 5))
		hardlockup_panic = 1;
	else if (!strncmp(str, "nopanic", 7))
		hardlockup_panic = 0;
	else if (!strncmp(str, "0", 1))
		watchdog_enabled &= ~NMI_WATCHDOG_ENABLED;
	else if (!strncmp(str, "1", 1))
		watchdog_enabled |= NMI_WATCHDOG_ENABLED;
	return 1;
}
__setup("nmi_watchdog=", hardlockup_panic_setup);

void 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);
	touch_softlockup_watchdog();
}
EXPORT_SYMBOL(touch_nmi_watchdog);

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;
	}

	/* 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;

	/* 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;
	}

	/*
	 * Disable the hard lockup detector if _any_ CPU fails to set up
	 * set up the hardware perf event. The watchdog() function checks
	 * the NMI_WATCHDOG_ENABLED bit periodically.
	 *
	 * The barriers are for syncing up watchdog_enabled across all the
	 * cpus, as clear_bit() does not use barriers.
	 */
	smp_mb__before_atomic();
	clear_bit(NMI_WATCHDOG_ENABLED_BIT, &watchdog_enabled);
	smp_mb__after_atomic();

	/* 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("Shutting down hard lockup detector on all cpus\n");

	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;
}

void watchdog_nmi_disable(unsigned int cpu)
{
	struct perf_event *event = per_cpu(watchdog_ev, cpu);

	if (event) {
		perf_event_disable(event);
		per_cpu(watchdog_ev, cpu) = NULL;

		/* should be in cleanup, but blocks oprofile */
		perf_event_release_kernel(event);

		/* watchdog_nmi_enable() expects this to be zero initially. */
		if (atomic_dec_and_test(&watchdog_cpus))
			firstcpu_err = 0;
	}
}
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);
	24
	25	/* boot commands */
	26	/*
	27	* Should we panic when a soft-lockup or hard-lockup occurs:
	28	*/
	29	unsigned int __read_mostly hardlockup_panic =
	30	CONFIG_BOOTPARAM_HARDLOCKUP_PANIC_VALUE;
	31	static unsigned long hardlockup_allcpu_dumped;
	32	/*
	33	* We may not want to enable hard lockup detection by default in all cases,
	34	* for example when running the kernel as a guest on a hypervisor. In these
	35	* cases this function can be called to disable hard lockup detection. This
	36	* function should only be executed once by the boot processor before the
	37	* kernel command line parameters are parsed, because otherwise it is not
	38	* possible to override this in hardlockup_panic_setup().
	39	*/
	40	void hardlockup_detector_disable(void)
	41	{
	42	watchdog_enabled &= ~NMI_WATCHDOG_ENABLED;
	43	}
	44
	45	static int __init hardlockup_panic_setup(char *str)
	46	{
	47	if (!strncmp(str, "panic", 5))
	48	hardlockup_panic = 1;
	49	else if (!strncmp(str, "nopanic", 7))
	50	hardlockup_panic = 0;
	51	else if (!strncmp(str, "0", 1))
	52	watchdog_enabled &= ~NMI_WATCHDOG_ENABLED;
	53	else if (!strncmp(str, "1", 1))
	54	watchdog_enabled \|= NMI_WATCHDOG_ENABLED;
	55	return 1;
	56	}
	57	__setup("nmi_watchdog=", hardlockup_panic_setup);
	58
	59	void touch_nmi_watchdog(void)
	60	{
	61	/*
	62	* Using __raw here because some code paths have
	63	* preemption enabled. If preemption is enabled
	64	* then interrupts should be enabled too, in which
	65	* case we shouldn't have to worry about the watchdog
	66	* going off.
	67	*/
	68	raw_cpu_write(watchdog_nmi_touch, true);
	69	touch_softlockup_watchdog();
	70	}
	71	EXPORT_SYMBOL(touch_nmi_watchdog);
	72
	73	static struct perf_event_attr wd_hw_attr = {
	74	.type = PERF_TYPE_HARDWARE,
	75	.config = PERF_COUNT_HW_CPU_CYCLES,
	76	.size = sizeof(struct perf_event_attr),
	77	.pinned = 1,
	78	.disabled = 1,
	79	};
	80
	81	/* Callback function for perf event subsystem */
82	static void watchdog_overflow_callback(struct perf_event *event,
83	struct perf_sample_data *data,
84	struct pt_regs *regs)
85	{
86	/* Ensure the watchdog never gets throttled */
87	event->hw.interrupts = 0;
88
b94f5118 DZ	89	if (atomic_read(&watchdog_park_in_progress) != 0)
	90	return;
	91
73ce0511 BM	92	if (__this_cpu_read(watchdog_nmi_touch) == true) {
	93	__this_cpu_write(watchdog_nmi_touch, false);
	94	return;
	95	}
	96
	97	/* check for a hardlockup
	98	* This is done by making sure our timer interrupt
	99	* is incrementing. The timer interrupt should have
	100	* fired multiple times before we overflow'd. If it hasn't
	101	* then this is a good indication the cpu is stuck
	102	*/
	103	if (is_hardlockup()) {
	104	int this_cpu = smp_processor_id();
	105
	106	/* only print hardlockups once */
	107	if (__this_cpu_read(hard_watchdog_warn) == true)
	108	return;
	109
	110	pr_emerg("Watchdog detected hard LOCKUP on cpu %d", this_cpu);
	111	print_modules();
	112	print_irqtrace_events(current);
	113	if (regs)
	114	show_regs(regs);
	115	else
	116	dump_stack();
	117
	118	/*
	119	* Perform all-CPU dump only once to avoid multiple hardlockups
	120	* generating interleaving traces
	121	*/
	122	if (sysctl_hardlockup_all_cpu_backtrace &&
	123	!test_and_set_bit(0, &hardlockup_allcpu_dumped))
	124	trigger_allbutself_cpu_backtrace();
	125
	126	if (hardlockup_panic)
	127	nmi_panic(regs, "Hard LOCKUP");
	128
	129	__this_cpu_write(hard_watchdog_warn, true);
	130	return;
	131	}
	132
	133	__this_cpu_write(hard_watchdog_warn, false);
	134	return;
	135	}
	136
	137	/*
	138	* People like the simple clean cpu node info on boot.
	139	* Reduce the watchdog noise by only printing messages
	140	* that are different from what cpu0 displayed.
	141	*/
8dcde9de PB	142	static unsigned long firstcpu_err;
8dcde9de PB	143	static atomic_t watchdog_cpus;
73ce0511 BM	144
	145	int watchdog_nmi_enable(unsigned int cpu)
	146	{
	147	struct perf_event_attr *wd_attr;
	148	struct perf_event *event = per_cpu(watchdog_ev, cpu);
8dcde9de	149	int firstcpu = 0;
73ce0511 BM	150
	151	/* nothing to do if the hard lockup detector is disabled */
	152	if (!(watchdog_enabled & NMI_WATCHDOG_ENABLED))
	153	goto out;
	154
	155	/* is it already setup and enabled? */
	156	if (event && event->state > PERF_EVENT_STATE_OFF)
	157	goto out;
	158
	159	/* it is setup but not enabled */
	160	if (event != NULL)
	161	goto out_enable;
	162
8dcde9de PB	163	if (atomic_inc_return(&watchdog_cpus) == 1)
	164	firstcpu = 1;
	165
73ce0511 BM	166	wd_attr = &wd_hw_attr;
	167	wd_attr->sample_period = hw_nmi_get_sample_period(watchdog_thresh);
	168
	169	/* Try to register using hardware perf events */
	170	event = perf_event_create_kernel_counter(wd_attr, cpu, NULL, watchdog_overflow_callback, NULL);
	171
8dcde9de PB	172	/* save the first cpu's error for future comparision */
	173	if (firstcpu && IS_ERR(event))
	174	firstcpu_err = PTR_ERR(event);
73ce0511 BM	175
73ce0511 BM	176	if (!IS_ERR(event)) {
8dcde9de PB	177	/* only print for the first cpu initialized */
8dcde9de PB	178	if (firstcpu \|\| firstcpu_err)
73ce0511 BM	179	pr_info("enabled on all CPUs, permanently consumes one hw-PMU counter.\n");
	180	goto out_save;
	181	}
	182
	183	/*
	184	* Disable the hard lockup detector if _any_ CPU fails to set up
	185	* set up the hardware perf event. The watchdog() function checks
	186	* the NMI_WATCHDOG_ENABLED bit periodically.
	187	*
	188	* The barriers are for syncing up watchdog_enabled across all the
	189	* cpus, as clear_bit() does not use barriers.
	190	*/
	191	smp_mb__before_atomic();
	192	clear_bit(NMI_WATCHDOG_ENABLED_BIT, &watchdog_enabled);
	193	smp_mb__after_atomic();
	194
	195	/* skip displaying the same error again */
8dcde9de	196	if (!firstcpu && (PTR_ERR(event) == firstcpu_err))
73ce0511 BM	197	return PTR_ERR(event);
	198
	199	/* vary the KERN level based on the returned errno */
	200	if (PTR_ERR(event) == -EOPNOTSUPP)
	201	pr_info("disabled (cpu%i): not supported (no LAPIC?)\n", cpu);
	202	else if (PTR_ERR(event) == -ENOENT)
	203	pr_warn("disabled (cpu%i): hardware events not enabled\n",
	204	cpu);
	205	else
	206	pr_err("disabled (cpu%i): unable to create perf event: %ld\n",
	207	cpu, PTR_ERR(event));
	208
	209	pr_info("Shutting down hard lockup detector on all cpus\n");
	210
	211	return PTR_ERR(event);
	212
	213	/* success path */
	214	out_save:
	215	per_cpu(watchdog_ev, cpu) = event;
	216	out_enable:
	217	perf_event_enable(per_cpu(watchdog_ev, cpu));
	218	out:
	219	return 0;
	220	}
	221
	222	void watchdog_nmi_disable(unsigned int cpu)
	223	{
	224	struct perf_event *event = per_cpu(watchdog_ev, cpu);
	225
	226	if (event) {
	227	perf_event_disable(event);
	228	per_cpu(watchdog_ev, cpu) = NULL;
	229
	230	/* should be in cleanup, but blocks oprofile */
	231	perf_event_release_kernel(event);
8dcde9de	232
73ce0511	233	/* watchdog_nmi_enable() expects this to be zero initially. */
8dcde9de PB	234	if (atomic_dec_and_test(&watchdog_cpus))
8dcde9de PB	235	firstcpu_err = 0;
73ce0511 BM	236	}
73ce0511 BM	237	}