summaryrefslogtreecommitdiffstats
path: root/kernel/locking/qspinlock_stat.h
blob: 6bd78c0740fc8b70f0a13398938f47186f8eed84 (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
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
/*
 * 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>
 */

/*
 * When queued spinlock statistical counters are enabled, the following
 * debugfs files will be created for reporting the counter values:
 *
 * <debugfs>/qlockstat/
 *   pv_hash_hops	- average # of hops per hashing operation
 *   pv_kick_unlock	- # of vCPU kicks issued at unlock time
 *   pv_kick_wake	- # of vCPU kicks used for computing pv_latency_wake
 *   pv_latency_kick	- average latency (ns) of vCPU kick operation
 *   pv_latency_wake	- average latency (ns) from vCPU kick to wakeup
 *   pv_lock_stealing	- # of lock stealing operations
 *   pv_spurious_wakeup	- # of spurious wakeups in non-head vCPUs
 *   pv_wait_again	- # of wait's after a queue head vCPU kick
 *   pv_wait_early	- # of early vCPU wait's
 *   pv_wait_head	- # of vCPU wait's at the queue head
 *   pv_wait_node	- # of vCPU wait's at a non-head queue node
 *   lock_pending	- # of locking operations via pending code
 *   lock_slowpath	- # of locking operations via MCS lock queue
 *
 * Writing to the "reset_counters" file will reset all the above counter
 * values.
 *
 * These statistical counters are implemented as per-cpu variables which are
 * summed and computed whenever the corresponding debugfs files are read. This
 * minimizes added overhead making the counters usable even in a production
 * environment.
 *
 * There may be slight difference between pv_kick_wake and pv_kick_unlock.
 */
enum qlock_stats {
	qstat_pv_hash_hops,
	qstat_pv_kick_unlock,
	qstat_pv_kick_wake,
	qstat_pv_latency_kick,
	qstat_pv_latency_wake,
	qstat_pv_lock_stealing,
	qstat_pv_spurious_wakeup,
	qstat_pv_wait_again,
	qstat_pv_wait_early,
	qstat_pv_wait_head,
	qstat_pv_wait_node,
	qstat_lock_pending,
	qstat_lock_slowpath,
	qstat_num,	/* Total number of statistical counters */
	qstat_reset_cnts = qstat_num,
};

#ifdef CONFIG_QUEUED_LOCK_STAT
/*
 * Collect pvqspinlock statistics
 */
#include <linux/debugfs.h>
#include <linux/sched.h>
#include <linux/sched/clock.h>
#include <linux/fs.h>

static const char * const qstat_names[qstat_num + 1] = {
	[qstat_pv_hash_hops]	   = "pv_hash_hops",
	[qstat_pv_kick_unlock]     = "pv_kick_unlock",
	[qstat_pv_kick_wake]       = "pv_kick_wake",
	[qstat_pv_spurious_wakeup] = "pv_spurious_wakeup",
	[qstat_pv_latency_kick]	   = "pv_latency_kick",
	[qstat_pv_latency_wake]    = "pv_latency_wake",
	[qstat_pv_lock_stealing]   = "pv_lock_stealing",
	[qstat_pv_wait_again]      = "pv_wait_again",
	[qstat_pv_wait_early]      = "pv_wait_early",
	[qstat_pv_wait_head]       = "pv_wait_head",
	[qstat_pv_wait_node]       = "pv_wait_node",
	[qstat_lock_pending]       = "lock_pending",
	[qstat_lock_slowpath]      = "lock_slowpath",
	[qstat_reset_cnts]         = "reset_counters",
};

/*
 * Per-cpu counters
 */
static DEFINE_PER_CPU(unsigned long, qstats[qstat_num]);
static DEFINE_PER_CPU(u64, pv_kick_time);

/*
 * Function to read and return the qlock statistical counter values
 *
 * The following counters are handled specially:
 * 1. qstat_pv_latency_kick
 *    Average kick latency (ns) = pv_latency_kick/pv_kick_unlock
 * 2. qstat_pv_latency_wake
 *    Average wake latency (ns) = pv_latency_wake/pv_kick_wake
 * 3. qstat_pv_hash_hops
 *    Average hops/hash = pv_hash_hops/pv_kick_unlock
 */
static ssize_t qstat_read(struct file *file, char __user *user_buf,
			  size_t count, loff_t *ppos)
{
	char buf[64];
	int cpu, counter, len;
	u64 stat = 0, kicks = 0;

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

	if (counter >= qstat_num)
		return -EBADF;

	for_each_possible_cpu(cpu) {
		stat += per_cpu(qstats[counter], cpu);
		/*
		 * Need to sum additional counter for some of them
		 */
		switch (counter) {

		case qstat_pv_latency_kick:
		case qstat_pv_hash_hops:
			kicks += per_cpu(qstats[qstat_pv_kick_unlock], cpu);
			break;

		case qstat_pv_latency_wake:
			kicks += per_cpu(qstats[qstat_pv_kick_wake], cpu);
			break;
		}
	}

	if (counter == qstat_pv_hash_hops) {
		u64 frac = 0;

		if (kicks) {
			frac = 100ULL * do_div(stat, kicks);
			frac = DIV_ROUND_CLOSEST_ULL(frac, kicks);
		}

		/*
		 * Return a X.XX decimal number
		 */
		len = snprintf(buf, sizeof(buf) - 1, "%llu.%02llu\n", stat, frac);
	} else {
		/*
		 * Round to the nearest ns
		 */
		if ((counter == qstat_pv_latency_kick) ||
		    (counter == qstat_pv_latency_wake)) {
			if (kicks)
				stat = DIV_ROUND_CLOSEST_ULL(stat, kicks);
		}
		len = snprintf(buf, sizeof(buf) - 1, "%llu\n", stat);
	}

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

/*
 * Function to handle write request
 *
 * When counter = reset_cnts, reset all the counter values.
 * Since the counter updates aren't atomic, the resetting is done twice
 * to make sure that the counters are very likely to be all cleared.
 */
static ssize_t qstat_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 != qstat_reset_cnts)
		return count;

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

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

/*
 * Debugfs data structures
 */
static const struct file_operations fops_qstat = {
	.read = qstat_read,
	.write = qstat_write,
	.llseek = default_llseek,
};

/*
 * Initialize debugfs for the qspinlock statistical counters
 */
static int __init init_qspinlock_stat(void)
{
	struct dentry *d_qstat = debugfs_create_dir("qlockstat", NULL);
	int i;

	if (!d_qstat)
		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 < qstat_num; i++)
		if (!debugfs_create_file(qstat_names[i], 0400, d_qstat,
					 (void *)(long)i, &fops_qstat))
			goto fail_undo;

	if (!debugfs_create_file(qstat_names[qstat_reset_cnts], 0200, d_qstat,
				 (void *)(long)qstat_reset_cnts, &fops_qstat))
		goto fail_undo;

	return 0;
fail_undo:
	debugfs_remove_recursive(d_qstat);
out:
	pr_warn("Could not create 'qlockstat' debugfs entries\n");
	return -ENOMEM;
}
fs_initcall(init_qspinlock_stat);

/*
 * Increment the PV qspinlock statistical counters
 */
static inline void qstat_inc(enum qlock_stats stat, bool cond)
{
	if (cond)
		this_cpu_inc(qstats[stat]);
}

/*
 * PV hash hop count
 */
static inline void qstat_hop(int hopcnt)
{
	this_cpu_add(qstats[qstat_pv_hash_hops], hopcnt);
}

/*
 * Replacement function for pv_kick()
 */
static inline void __pv_kick(int cpu)
{
	u64 start = sched_clock();

	per_cpu(pv_kick_time, cpu) = start;
	pv_kick(cpu);
	this_cpu_add(qstats[qstat_pv_latency_kick], sched_clock() - start);
}

/*
 * Replacement function for pv_wait()
 */
static inline void __pv_wait(u8 *ptr, u8 val)
{
	u64 *pkick_time = this_cpu_ptr(&pv_kick_time);

	*pkick_time = 0;
	pv_wait(ptr, val);
	if (*pkick_time) {
		this_cpu_add(qstats[qstat_pv_latency_wake],
			     sched_clock() - *pkick_time);
		qstat_inc(qstat_pv_kick_wake, true);
	}
}

#define pv_kick(c)	__pv_kick(c)
#define pv_wait(p, v)	__pv_wait(p, v)

#else /* CONFIG_QUEUED_LOCK_STAT */

static inline void qstat_inc(enum qlock_stats stat, bool cond)	{ }
static inline void qstat_hop(int hopcnt)			{ }

#endif /* CONFIG_QUEUED_LOCK_STAT */