Merge branch 'for-3.2' of git://linux-nfs.org/~bfields/linux
[linux-btrfs-devel.git] / kernel / delayacct.c
blob418b3f7053aaad32cef7d6082e833e5f33fdd6b7
1 /* delayacct.c - per-task delay accounting
3 * Copyright (C) Shailabh Nagar, IBM Corp. 2006
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.
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.
16 #include <linux/sched.h>
17 #include <linux/slab.h>
18 #include <linux/taskstats.h>
19 #include <linux/time.h>
20 #include <linux/sysctl.h>
21 #include <linux/delayacct.h>
22 #include <linux/module.h>
24 int delayacct_on __read_mostly = 1; /* Delay accounting turned on/off */
25 EXPORT_SYMBOL_GPL(delayacct_on);
26 struct kmem_cache *delayacct_cache;
28 static int __init delayacct_setup_disable(char *str)
30 delayacct_on = 0;
31 return 1;
33 __setup("nodelayacct", delayacct_setup_disable);
35 void delayacct_init(void)
37 delayacct_cache = KMEM_CACHE(task_delay_info, SLAB_PANIC);
38 delayacct_tsk_init(&init_task);
41 void __delayacct_tsk_init(struct task_struct *tsk)
43 tsk->delays = kmem_cache_zalloc(delayacct_cache, GFP_KERNEL);
44 if (tsk->delays)
45 spin_lock_init(&tsk->delays->lock);
49 * Start accounting for a delay statistic using
50 * its starting timestamp (@start)
53 static inline void delayacct_start(struct timespec *start)
55 do_posix_clock_monotonic_gettime(start);
59 * Finish delay accounting for a statistic using
60 * its timestamps (@start, @end), accumalator (@total) and @count
63 static void delayacct_end(struct timespec *start, struct timespec *end,
64 u64 *total, u32 *count)
66 struct timespec ts;
67 s64 ns;
68 unsigned long flags;
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;
76 spin_lock_irqsave(&current->delays->lock, flags);
77 *total += ns;
78 (*count)++;
79 spin_unlock_irqrestore(&current->delays->lock, flags);
82 void __delayacct_blkio_start(void)
84 delayacct_start(&current->delays->blkio_start);
87 void __delayacct_blkio_end(void)
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);
102 int __delayacct_add_tsk(struct taskstats *d, struct task_struct *tsk)
104 s64 tmp;
105 unsigned long t1;
106 unsigned long long t2, t3;
107 unsigned long flags;
108 struct timespec ts;
110 /* Though tsk->delays accessed later, early exit avoids
111 * unnecessary returning of other data
113 if (!tsk->delays)
114 goto done;
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;
121 tmp = (s64)d->cpu_scaled_run_real_total;
122 cputime_to_timespec(tsk->utimescaled + tsk->stimescaled, &ts);
123 tmp += timespec_to_ns(&ts);
124 d->cpu_scaled_run_real_total =
125 (tmp < (s64)d->cpu_scaled_run_real_total) ? 0 : tmp;
128 * No locking available for sched_info (and too expensive to add one)
129 * Mitigate by taking snapshot of values
131 t1 = tsk->sched_info.pcount;
132 t2 = tsk->sched_info.run_delay;
133 t3 = tsk->se.sum_exec_runtime;
135 d->cpu_count += t1;
137 tmp = (s64)d->cpu_delay_total + t2;
138 d->cpu_delay_total = (tmp < (s64)d->cpu_delay_total) ? 0 : tmp;
140 tmp = (s64)d->cpu_run_virtual_total + t3;
141 d->cpu_run_virtual_total =
142 (tmp < (s64)d->cpu_run_virtual_total) ? 0 : tmp;
144 /* zero XXX_total, non-zero XXX_count implies XXX stat overflowed */
146 spin_lock_irqsave(&tsk->delays->lock, flags);
147 tmp = d->blkio_delay_total + tsk->delays->blkio_delay;
148 d->blkio_delay_total = (tmp < d->blkio_delay_total) ? 0 : tmp;
149 tmp = d->swapin_delay_total + tsk->delays->swapin_delay;
150 d->swapin_delay_total = (tmp < d->swapin_delay_total) ? 0 : tmp;
151 tmp = d->freepages_delay_total + tsk->delays->freepages_delay;
152 d->freepages_delay_total = (tmp < d->freepages_delay_total) ? 0 : tmp;
153 d->blkio_count += tsk->delays->blkio_count;
154 d->swapin_count += tsk->delays->swapin_count;
155 d->freepages_count += tsk->delays->freepages_count;
156 spin_unlock_irqrestore(&tsk->delays->lock, flags);
158 done:
159 return 0;
162 __u64 __delayacct_blkio_ticks(struct task_struct *tsk)
164 __u64 ret;
165 unsigned long flags;
167 spin_lock_irqsave(&tsk->delays->lock, flags);
168 ret = nsec_to_clock_t(tsk->delays->blkio_delay +
169 tsk->delays->swapin_delay);
170 spin_unlock_irqrestore(&tsk->delays->lock, flags);
171 return ret;
174 void __delayacct_freepages_start(void)
176 delayacct_start(&current->delays->freepages_start);
179 void __delayacct_freepages_end(void)
181 delayacct_end(&current->delays->freepages_start,
182 &current->delays->freepages_end,
183 &current->delays->freepages_delay,
184 &current->delays->freepages_count);