aboutsummaryrefslogtreecommitdiffstats
path: root/kernel/locking/lock_events.c
blob: fa2c2f951c6b5f8571cd576c2012c57ddd46dbb2 (plain) (blame)
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
/* SPDX-License-Identifier: GPL-2.0 */
/*
 * This program is free software; you can redistribute it and/or modify
 * it under the terms of the GNU General Public License as published by
 * the Free Software Foundation; either version 2 of the License, or
 * (at your option) any later version.
 *
 * This program is distributed in the hope that it will be useful,
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
 * GNU General Public License for more details.
 *
 * Authors: Waiman Long <waiman.long@hpe.com>
 */

/*
 * Collect locking event counts
 */
#include <linux/debugfs.h>
#include <linux/sched.h>
#include <linux/sched/clock.h>
#include <linux/fs.h>

#include "lock_events.h"

#undef  LOCK_EVENT
#define LOCK_EVENT(name)	[LOCKEVENT_ ## name] = #name,

#define LOCK_EVENTS_DIR		"lock_event_counts"

/*
 * When CONFIG_LOCK_EVENT_COUNTS is enabled, event counts of different
 * types of locks will be reported under the <debugfs>/lock_event_counts/
 * directory. See lock_events_list.h for the list of available locking
 * events.
 *
 * Writing to the special ".reset_counts" file will reset all the above
 * locking event counts. This is a very slow operation and so should not
 * be done frequently.
 *
 * These event counts are implemented as per-cpu variables which are
 * summed and computed whenever the corresponding debugfs files are read. This
 * minimizes added overhead making the counts usable even in a production
 * environment.
 */
static const char * const lockevent_names[lockevent_num + 1] = {

#include "lock_events_list.h"

	[LOCKEVENT_reset_cnts] = ".reset_counts",
};

/*
 * Per-cpu counts
 */
DEFINE_PER_CPU(unsigned long, lockevents[lockevent_num]);

/*
 * The lockevent_read() function can be overridden.
 */
ssize_t __weak lockevent_read(struct file *file, char __user *user_buf,
			      size_t count, loff_t *ppos)
{
	char buf[64];
	int cpu, id, len;
	u64 sum = 0;

	/*
	 * Get the counter ID stored in file->f_inode->i_private
	 */
	id = (long)file_inode(file)->i_private;

	if (id >= lockevent_num)
		return -EBADF;

	for_each_possible_cpu(cpu)
		sum += per_cpu(lockevents[id], cpu);
	len = snprintf(buf, sizeof(buf) - 1, "%llu\n", sum);

	return simple_read_from_buffer(user_buf, count, ppos, buf, len);
}

/*
 * Function to handle write request
 *
 * When idx = reset_cnts, reset all the counts.
 */
static ssize_t lockevent_write(struct file *file, const char __user *user_buf,
			   size_t count, loff_t *ppos)
{
	int cpu;

	/*
	 * Get the counter ID stored in file->f_inode->i_private
	 */
	if ((long)file_inode(file)->i_private != LOCKEVENT_reset_cnts)
		return count;

	for_each_possible_cpu(cpu) {
		int i;
		unsigned long *ptr = per_cpu_ptr(lockevents, cpu);

		for (i = 0 ; i < lockevent_num; i++)
			WRITE_ONCE(ptr[i], 0);
	}
	return count;
}

/*
 * Debugfs data structures
 */
static const struct file_operations fops_lockevent = {
	.read = lockevent_read,
	.write = lockevent_write,
	.llseek = default_llseek,
};

#ifdef CONFIG_PARAVIRT_SPINLOCKS
#include <asm/paravirt.h>

static bool __init skip_lockevent(const char *name)
{
	static int pv_on __initdata = -1;

	if (pv_on < 0)
		pv_on = !pv_is_native_spin_unlock();
	/*
	 * Skip PV qspinlock events on bare metal.
	 */
	if (!pv_on && !memcmp(name, "pv_", 3))
		return true;
	return false;
}
#else
static inline bool skip_lockevent(const char *name)
{
	return false;
}
#endif

/*
 * Initialize debugfs for the locking event counts.
 */
static int __init init_lockevent_counts(void)
{
	struct dentry *d_counts = debugfs_create_dir(LOCK_EVENTS_DIR, NULL);
	int i;

	if (!d_counts)
		goto out;

	/*
	 * Create the debugfs files
	 *
	 * As reading from and writing to the stat files can be slow, only
	 * root is allowed to do the read/write to limit impact to system
	 * performance.
	 */
	for (i = 0; i < lockevent_num; i++) {
		if (skip_lockevent(lockevent_names[i]))
			continue;
		if (!debugfs_create_file(lockevent_names[i], 0400, d_counts,
					 (void *)(long)i, &fops_lockevent))
			goto fail_undo;
	}

	if (!debugfs_create_file(lockevent_names[LOCKEVENT_reset_cnts], 0200,
				 d_counts, (void *)(long)LOCKEVENT_reset_cnts,
				 &fops_lockevent))
		goto fail_undo;

	return 0;
fail_undo:
	debugfs_remove_recursive(d_counts);
out:
	pr_warn("Could not create '%s' debugfs entries\n", LOCK_EVENTS_DIR);
	return -ENOMEM;
}
fs_initcall(init_lockevent_counts);