xref: /openbmc/linux/kernel/delayacct.c (revision c537b994)
1 /* delayacct.c - per-task delay accounting
2  *
3  * Copyright (C) Shailabh Nagar, IBM Corp. 2006
4  *
5  * This program is free software;  you can redistribute it and/or modify
6  * it under the terms of the GNU General Public License as published by
7  * the Free Software Foundation; either version 2 of the License, or
8  * (at your option) any later version.
9  *
10  * This program is distributed in the hope that it would be useful, but
11  * WITHOUT ANY WARRANTY; without even the implied warranty of
12  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See
13  * the GNU General Public License for more details.
14  */
15 
16 #include <linux/sched.h>
17 #include <linux/slab.h>
18 #include <linux/time.h>
19 #include <linux/sysctl.h>
20 #include <linux/delayacct.h>
21 
22 int delayacct_on __read_mostly = 1;	/* Delay accounting turned on/off */
23 struct kmem_cache *delayacct_cache;
24 
25 static int __init delayacct_setup_disable(char *str)
26 {
27 	delayacct_on = 0;
28 	return 1;
29 }
30 __setup("nodelayacct", delayacct_setup_disable);
31 
32 void delayacct_init(void)
33 {
34 	delayacct_cache = kmem_cache_create("delayacct_cache",
35 					sizeof(struct task_delay_info),
36 					0,
37 					SLAB_PANIC,
38 					NULL, NULL);
39 	delayacct_tsk_init(&init_task);
40 }
41 
42 void __delayacct_tsk_init(struct task_struct *tsk)
43 {
44 	tsk->delays = kmem_cache_zalloc(delayacct_cache, GFP_KERNEL);
45 	if (tsk->delays)
46 		spin_lock_init(&tsk->delays->lock);
47 }
48 
49 /*
50  * Start accounting for a delay statistic using
51  * its starting timestamp (@start)
52  */
53 
54 static inline void delayacct_start(struct timespec *start)
55 {
56 	do_posix_clock_monotonic_gettime(start);
57 }
58 
59 /*
60  * Finish delay accounting for a statistic using
61  * its timestamps (@start, @end), accumalator (@total) and @count
62  */
63 
64 static void delayacct_end(struct timespec *start, struct timespec *end,
65 				u64 *total, u32 *count)
66 {
67 	struct timespec ts;
68 	s64 ns;
69 	unsigned long flags;
70 
71 	do_posix_clock_monotonic_gettime(end);
72 	ts = timespec_sub(*end, *start);
73 	ns = timespec_to_ns(&ts);
74 	if (ns < 0)
75 		return;
76 
77 	spin_lock_irqsave(&current->delays->lock, flags);
78 	*total += ns;
79 	(*count)++;
80 	spin_unlock_irqrestore(&current->delays->lock, flags);
81 }
82 
83 void __delayacct_blkio_start(void)
84 {
85 	delayacct_start(&current->delays->blkio_start);
86 }
87 
88 void __delayacct_blkio_end(void)
89 {
90 	if (current->delays->flags & DELAYACCT_PF_SWAPIN)
91 		/* Swapin block I/O */
92 		delayacct_end(&current->delays->blkio_start,
93 			&current->delays->blkio_end,
94 			&current->delays->swapin_delay,
95 			&current->delays->swapin_count);
96 	else	/* Other block I/O */
97 		delayacct_end(&current->delays->blkio_start,
98 			&current->delays->blkio_end,
99 			&current->delays->blkio_delay,
100 			&current->delays->blkio_count);
101 }
102 
103 int __delayacct_add_tsk(struct taskstats *d, struct task_struct *tsk)
104 {
105 	s64 tmp;
106 	struct timespec ts;
107 	unsigned long t1,t2,t3;
108 	unsigned long flags;
109 
110 	/* Though tsk->delays accessed later, early exit avoids
111 	 * unnecessary returning of other data
112 	 */
113 	if (!tsk->delays)
114 		goto done;
115 
116 	tmp = (s64)d->cpu_run_real_total;
117 	cputime_to_timespec(tsk->utime + tsk->stime, &ts);
118 	tmp += timespec_to_ns(&ts);
119 	d->cpu_run_real_total = (tmp < (s64)d->cpu_run_real_total) ? 0 : tmp;
120 
121 	/*
122 	 * No locking available for sched_info (and too expensive to add one)
123 	 * Mitigate by taking snapshot of values
124 	 */
125 	t1 = tsk->sched_info.pcnt;
126 	t2 = tsk->sched_info.run_delay;
127 	t3 = tsk->sched_info.cpu_time;
128 
129 	d->cpu_count += t1;
130 
131 	jiffies_to_timespec(t2, &ts);
132 	tmp = (s64)d->cpu_delay_total + timespec_to_ns(&ts);
133 	d->cpu_delay_total = (tmp < (s64)d->cpu_delay_total) ? 0 : tmp;
134 
135 	tmp = (s64)d->cpu_run_virtual_total + (s64)jiffies_to_usecs(t3) * 1000;
136 	d->cpu_run_virtual_total =
137 		(tmp < (s64)d->cpu_run_virtual_total) ?	0 : tmp;
138 
139 	/* zero XXX_total, non-zero XXX_count implies XXX stat overflowed */
140 
141 	spin_lock_irqsave(&tsk->delays->lock, flags);
142 	tmp = d->blkio_delay_total + tsk->delays->blkio_delay;
143 	d->blkio_delay_total = (tmp < d->blkio_delay_total) ? 0 : tmp;
144 	tmp = d->swapin_delay_total + tsk->delays->swapin_delay;
145 	d->swapin_delay_total = (tmp < d->swapin_delay_total) ? 0 : tmp;
146 	d->blkio_count += tsk->delays->blkio_count;
147 	d->swapin_count += tsk->delays->swapin_count;
148 	spin_unlock_irqrestore(&tsk->delays->lock, flags);
149 
150 done:
151 	return 0;
152 }
153 
154 __u64 __delayacct_blkio_ticks(struct task_struct *tsk)
155 {
156 	__u64 ret;
157 	unsigned long flags;
158 
159 	spin_lock_irqsave(&tsk->delays->lock, flags);
160 	ret = nsec_to_clock_t(tsk->delays->blkio_delay +
161 				tsk->delays->swapin_delay);
162 	spin_unlock_irqrestore(&tsk->delays->lock, flags);
163 	return ret;
164 }
165 
166