aboutsummaryrefslogtreecommitdiff
path: root/kernel/watchdog_perf.c
blob: 8ea00c4a24b2d91e330875ac95776e9fc9301f3b (plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
// SPDX-License-Identifier: GPL-2.0
/*
 * Detect hard lockups on a system using perf
 *
 * 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(struct perf_event *, watchdog_ev);
static DEFINE_PER_CPU(struct perf_event *, dead_event);
static struct cpumask dead_events_mask;

static atomic_t watchdog_cpus = ATOMIC_INIT(0);

#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 (!watchdog_check_timestamp())
		return;

	watchdog_hardlockup_check(smp_processor_id(), regs);
}

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

	/*
	 * Preemption is not disabled because memory will be allocated.
	 * Ensure CPU-locality by calling this in per-CPU kthread.
	 */
	WARN_ON(!is_percpu_thread());
	cpu = raw_smp_processor_id();
	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;
}

/**
 * watchdog_hardlockup_enable - Enable the local event
 *
 * @cpu: The CPU to enable hard lockup on.
 */
void watchdog_hardlockup_enable(unsigned int cpu)
{
	WARN_ON_ONCE(cpu != smp_processor_id());

	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));
}

/**
 * watchdog_hardlockup_disable - Disable the local event
 *
 * @cpu: The CPU to enable hard lockup on.
 */
void watchdog_hardlockup_disable(unsigned int cpu)
{
	struct perf_event *event = this_cpu_read(watchdog_ev);

	WARN_ON_ONCE(cpu != smp_processor_id());

	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 & WATCHDOG_HARDLOCKUP_ENABLED))
		return;

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

		if (event)
			perf_event_enable(event);
	}
}

bool __weak __init arch_perf_nmi_is_available(void)
{
	return true;
}

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

	if (!arch_perf_nmi_is_available())
		return -ENODEV;

	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;
}