Merge branch 'x86-mm-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git...
[linux-2.6-block.git] / kernel / delayacct.c
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/sched/task.h>
18 #include <linux/sched/cputime.h>
19 #include <linux/slab.h>
20 #include <linux/taskstats.h>
21 #include <linux/time.h>
22 #include <linux/sysctl.h>
23 #include <linux/delayacct.h>
24 #include <linux/module.h>
25
26 int delayacct_on __read_mostly = 1;     /* Delay accounting turned on/off */
27 EXPORT_SYMBOL_GPL(delayacct_on);
28 struct kmem_cache *delayacct_cache;
29
30 static int __init delayacct_setup_disable(char *str)
31 {
32         delayacct_on = 0;
33         return 1;
34 }
35 __setup("nodelayacct", delayacct_setup_disable);
36
37 void delayacct_init(void)
38 {
39         delayacct_cache = KMEM_CACHE(task_delay_info, SLAB_PANIC|SLAB_ACCOUNT);
40         delayacct_tsk_init(&init_task);
41 }
42
43 void __delayacct_tsk_init(struct task_struct *tsk)
44 {
45         tsk->delays = kmem_cache_zalloc(delayacct_cache, GFP_KERNEL);
46         if (tsk->delays)
47                 spin_lock_init(&tsk->delays->lock);
48 }
49
50 /*
51  * Finish delay accounting for a statistic using its timestamps (@start),
52  * accumalator (@total) and @count
53  */
54 static void delayacct_end(spinlock_t *lock, u64 *start, u64 *total, u32 *count)
55 {
56         s64 ns = ktime_get_ns() - *start;
57         unsigned long flags;
58
59         if (ns > 0) {
60                 spin_lock_irqsave(lock, flags);
61                 *total += ns;
62                 (*count)++;
63                 spin_unlock_irqrestore(lock, flags);
64         }
65 }
66
67 void __delayacct_blkio_start(void)
68 {
69         current->delays->blkio_start = ktime_get_ns();
70 }
71
72 /*
73  * We cannot rely on the `current` macro, as we haven't yet switched back to
74  * the process being woken.
75  */
76 void __delayacct_blkio_end(struct task_struct *p)
77 {
78         struct task_delay_info *delays = p->delays;
79         u64 *total;
80         u32 *count;
81
82         if (p->delays->flags & DELAYACCT_PF_SWAPIN) {
83                 total = &delays->swapin_delay;
84                 count = &delays->swapin_count;
85         } else {
86                 total = &delays->blkio_delay;
87                 count = &delays->blkio_count;
88         }
89
90         delayacct_end(&delays->lock, &delays->blkio_start, total, count);
91 }
92
93 int __delayacct_add_tsk(struct taskstats *d, struct task_struct *tsk)
94 {
95         u64 utime, stime, stimescaled, utimescaled;
96         unsigned long long t2, t3;
97         unsigned long flags, t1;
98         s64 tmp;
99
100         task_cputime(tsk, &utime, &stime);
101         tmp = (s64)d->cpu_run_real_total;
102         tmp += utime + stime;
103         d->cpu_run_real_total = (tmp < (s64)d->cpu_run_real_total) ? 0 : tmp;
104
105         task_cputime_scaled(tsk, &utimescaled, &stimescaled);
106         tmp = (s64)d->cpu_scaled_run_real_total;
107         tmp += utimescaled + stimescaled;
108         d->cpu_scaled_run_real_total =
109                 (tmp < (s64)d->cpu_scaled_run_real_total) ? 0 : tmp;
110
111         /*
112          * No locking available for sched_info (and too expensive to add one)
113          * Mitigate by taking snapshot of values
114          */
115         t1 = tsk->sched_info.pcount;
116         t2 = tsk->sched_info.run_delay;
117         t3 = tsk->se.sum_exec_runtime;
118
119         d->cpu_count += t1;
120
121         tmp = (s64)d->cpu_delay_total + t2;
122         d->cpu_delay_total = (tmp < (s64)d->cpu_delay_total) ? 0 : tmp;
123
124         tmp = (s64)d->cpu_run_virtual_total + t3;
125         d->cpu_run_virtual_total =
126                 (tmp < (s64)d->cpu_run_virtual_total) ? 0 : tmp;
127
128         /* zero XXX_total, non-zero XXX_count implies XXX stat overflowed */
129
130         spin_lock_irqsave(&tsk->delays->lock, flags);
131         tmp = d->blkio_delay_total + tsk->delays->blkio_delay;
132         d->blkio_delay_total = (tmp < d->blkio_delay_total) ? 0 : tmp;
133         tmp = d->swapin_delay_total + tsk->delays->swapin_delay;
134         d->swapin_delay_total = (tmp < d->swapin_delay_total) ? 0 : tmp;
135         tmp = d->freepages_delay_total + tsk->delays->freepages_delay;
136         d->freepages_delay_total = (tmp < d->freepages_delay_total) ? 0 : tmp;
137         d->blkio_count += tsk->delays->blkio_count;
138         d->swapin_count += tsk->delays->swapin_count;
139         d->freepages_count += tsk->delays->freepages_count;
140         spin_unlock_irqrestore(&tsk->delays->lock, flags);
141
142         return 0;
143 }
144
145 __u64 __delayacct_blkio_ticks(struct task_struct *tsk)
146 {
147         __u64 ret;
148         unsigned long flags;
149
150         spin_lock_irqsave(&tsk->delays->lock, flags);
151         ret = nsec_to_clock_t(tsk->delays->blkio_delay +
152                                 tsk->delays->swapin_delay);
153         spin_unlock_irqrestore(&tsk->delays->lock, flags);
154         return ret;
155 }
156
157 void __delayacct_freepages_start(void)
158 {
159         current->delays->freepages_start = ktime_get_ns();
160 }
161
162 void __delayacct_freepages_end(void)
163 {
164         delayacct_end(
165                 &current->delays->lock,
166                 &current->delays->freepages_start,
167                 &current->delays->freepages_delay,
168                 &current->delays->freepages_count);
169 }
170