xref: /linux/kernel/delayacct.c (revision f24e9f586b377749dff37554696cf3a105540c94)
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 kmem_cache_t *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, SLAB_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 
70 	do_posix_clock_monotonic_gettime(end);
71 	ts = timespec_sub(*end, *start);
72 	ns = timespec_to_ns(&ts);
73 	if (ns < 0)
74 		return;
75 
76 	spin_lock(&current->delays->lock);
77 	*total += ns;
78 	(*count)++;
79 	spin_unlock(&current->delays->lock);
80 }
81 
82 void __delayacct_blkio_start(void)
83 {
84 	delayacct_start(&current->delays->blkio_start);
85 }
86 
87 void __delayacct_blkio_end(void)
88 {
89 	if (current->delays->flags & DELAYACCT_PF_SWAPIN)
90 		/* Swapin block I/O */
91 		delayacct_end(&current->delays->blkio_start,
92 			&current->delays->blkio_end,
93 			&current->delays->swapin_delay,
94 			&current->delays->swapin_count);
95 	else	/* Other block I/O */
96 		delayacct_end(&current->delays->blkio_start,
97 			&current->delays->blkio_end,
98 			&current->delays->blkio_delay,
99 			&current->delays->blkio_count);
100 }
101 
102 int __delayacct_add_tsk(struct taskstats *d, struct task_struct *tsk)
103 {
104 	s64 tmp;
105 	struct timespec ts;
106 	unsigned long t1,t2,t3;
107 
108 	/* Though tsk->delays accessed later, early exit avoids
109 	 * unnecessary returning of other data
110 	 */
111 	if (!tsk->delays)
112 		goto done;
113 
114 	tmp = (s64)d->cpu_run_real_total;
115 	cputime_to_timespec(tsk->utime + tsk->stime, &ts);
116 	tmp += timespec_to_ns(&ts);
117 	d->cpu_run_real_total = (tmp < (s64)d->cpu_run_real_total) ? 0 : tmp;
118 
119 	/*
120 	 * No locking available for sched_info (and too expensive to add one)
121 	 * Mitigate by taking snapshot of values
122 	 */
123 	t1 = tsk->sched_info.pcnt;
124 	t2 = tsk->sched_info.run_delay;
125 	t3 = tsk->sched_info.cpu_time;
126 
127 	d->cpu_count += t1;
128 
129 	jiffies_to_timespec(t2, &ts);
130 	tmp = (s64)d->cpu_delay_total + timespec_to_ns(&ts);
131 	d->cpu_delay_total = (tmp < (s64)d->cpu_delay_total) ? 0 : tmp;
132 
133 	tmp = (s64)d->cpu_run_virtual_total + (s64)jiffies_to_usecs(t3) * 1000;
134 	d->cpu_run_virtual_total =
135 		(tmp < (s64)d->cpu_run_virtual_total) ?	0 : tmp;
136 
137 	/* zero XXX_total, non-zero XXX_count implies XXX stat overflowed */
138 
139 	spin_lock(&tsk->delays->lock);
140 	tmp = d->blkio_delay_total + tsk->delays->blkio_delay;
141 	d->blkio_delay_total = (tmp < d->blkio_delay_total) ? 0 : tmp;
142 	tmp = d->swapin_delay_total + tsk->delays->swapin_delay;
143 	d->swapin_delay_total = (tmp < d->swapin_delay_total) ? 0 : tmp;
144 	d->blkio_count += tsk->delays->blkio_count;
145 	d->swapin_count += tsk->delays->swapin_count;
146 	spin_unlock(&tsk->delays->lock);
147 
148 done:
149 	return 0;
150 }
151 
152 __u64 __delayacct_blkio_ticks(struct task_struct *tsk)
153 {
154 	__u64 ret;
155 
156 	spin_lock(&tsk->delays->lock);
157 	ret = nsec_to_clock_t(tsk->delays->blkio_delay +
158 				tsk->delays->swapin_delay);
159 	spin_unlock(&tsk->delays->lock);
160 	return ret;
161 }
162 
163