1/* SPDX-License-Identifier: GPL-2.0 */
2/*
3 * This program is free software; you can redistribute it and/or modify
4 * it under the terms of the GNU General Public License as published by
5 * the Free Software Foundation; either version 2 of the License, or
6 * (at your option) any later version.
7 *
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
11 * GNU General Public License for more details.
12 *
13 * Authors: Waiman Long <waiman.long@hpe.com>
14 */
15
16/*
17 * Collect locking event counts
18 */
19#include <linux/debugfs.h>
20#include <linux/sched.h>
21#include <linux/sched/clock.h>
22#include <linux/fs.h>
23
24#include "lock_events.h"
25
26#undef  LOCK_EVENT
27#define LOCK_EVENT(name)	[LOCKEVENT_ ## name] = #name,
28
29#define LOCK_EVENTS_DIR		"lock_event_counts"
30
31/*
32 * When CONFIG_LOCK_EVENT_COUNTS is enabled, event counts of different
33 * types of locks will be reported under the <debugfs>/lock_event_counts/
34 * directory. See lock_events_list.h for the list of available locking
35 * events.
36 *
37 * Writing to the special ".reset_counts" file will reset all the above
38 * locking event counts. This is a very slow operation and so should not
39 * be done frequently.
40 *
41 * These event counts are implemented as per-cpu variables which are
42 * summed and computed whenever the corresponding debugfs files are read. This
43 * minimizes added overhead making the counts usable even in a production
44 * environment.
45 */
46static const char * const lockevent_names[lockevent_num + 1] = {
47
48#include "lock_events_list.h"
49
50	[LOCKEVENT_reset_cnts] = ".reset_counts",
51};
52
53/*
54 * Per-cpu counts
55 */
56DEFINE_PER_CPU(unsigned long, lockevents[lockevent_num]);
57
58/*
59 * The lockevent_read() function can be overridden.
60 */
61ssize_t __weak lockevent_read(struct file *file, char __user *user_buf,
62			      size_t count, loff_t *ppos)
63{
64	char buf[64];
65	int cpu, id, len;
66	u64 sum = 0;
67
68	/*
69	 * Get the counter ID stored in file->f_inode->i_private
70	 */
71	id = (long)file_inode(file)->i_private;
72
73	if (id >= lockevent_num)
74		return -EBADF;
75
76	for_each_possible_cpu(cpu)
77		sum += per_cpu(lockevents[id], cpu);
78	len = snprintf(buf, sizeof(buf) - 1, "%llu\n", sum);
79
80	return simple_read_from_buffer(user_buf, count, ppos, buf, len);
81}
82
83/*
84 * Function to handle write request
85 *
86 * When idx = reset_cnts, reset all the counts.
87 */
88static ssize_t lockevent_write(struct file *file, const char __user *user_buf,
89			   size_t count, loff_t *ppos)
90{
91	int cpu;
92
93	/*
94	 * Get the counter ID stored in file->f_inode->i_private
95	 */
96	if ((long)file_inode(file)->i_private != LOCKEVENT_reset_cnts)
97		return count;
98
99	for_each_possible_cpu(cpu) {
100		int i;
101		unsigned long *ptr = per_cpu_ptr(lockevents, cpu);
102
103		for (i = 0 ; i < lockevent_num; i++)
104			WRITE_ONCE(ptr[i], 0);
105	}
106	return count;
107}
108
109/*
110 * Debugfs data structures
111 */
112static const struct file_operations fops_lockevent = {
113	.read = lockevent_read,
114	.write = lockevent_write,
115	.llseek = default_llseek,
116};
117
118#ifdef CONFIG_PARAVIRT_SPINLOCKS
119#include <asm/paravirt.h>
120
121static bool __init skip_lockevent(const char *name)
122{
123	static int pv_on __initdata = -1;
124
125	if (pv_on < 0)
126		pv_on = !pv_is_native_spin_unlock();
127	/*
128	 * Skip PV qspinlock events on bare metal.
129	 */
130	if (!pv_on && !memcmp(name, "pv_", 3))
131		return true;
132	return false;
133}
134#else
135static inline bool skip_lockevent(const char *name)
136{
137	return false;
138}
139#endif
140
141/*
142 * Initialize debugfs for the locking event counts.
143 */
144static int __init init_lockevent_counts(void)
145{
146	struct dentry *d_counts = debugfs_create_dir(LOCK_EVENTS_DIR, NULL);
147	int i;
148
149	if (IS_ERR(d_counts))
150		goto out;
151
152	/*
153	 * Create the debugfs files
154	 *
155	 * As reading from and writing to the stat files can be slow, only
156	 * root is allowed to do the read/write to limit impact to system
157	 * performance.
158	 */
159	for (i = 0; i < lockevent_num; i++) {
160		if (skip_lockevent(lockevent_names[i]))
161			continue;
162		if (IS_ERR(debugfs_create_file(lockevent_names[i], 0400, d_counts,
163					 (void *)(long)i, &fops_lockevent)))
164			goto fail_undo;
165	}
166
167	if (IS_ERR(debugfs_create_file(lockevent_names[LOCKEVENT_reset_cnts], 0200,
168				 d_counts, (void *)(long)LOCKEVENT_reset_cnts,
169				 &fops_lockevent)))
170		goto fail_undo;
171
172	return 0;
173fail_undo:
174	debugfs_remove_recursive(d_counts);
175out:
176	pr_warn("Could not create '%s' debugfs entries\n", LOCK_EVENTS_DIR);
177	return -ENOMEM;
178}
179fs_initcall(init_lockevent_counts);
180