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/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>
  23
  24 int delayacct_on __read_mostly = 1;     /* Delay accounting turned on/off */
  25 EXPORT_SYMBOL_GPL(delayacct_on);
  26 struct kmem_cache *delayacct_cache;
  27
  28 static int __init delayacct_setup_disable(char *str)
  29 {
  30         delayacct_on = 0;
  31         return 1;
  32 }
  33 __setup("nodelayacct", delayacct_setup_disable);
  34
  35 void delayacct_init(void)
  36 {
  37         delayacct_cache = KMEM_CACHE(task_delay_info, SLAB_PANIC|SLAB_ACCOUNT);
  38         delayacct_tsk_init(&init_task);
  39 }
  40
  41 void __delayacct_tsk_init(struct task_struct *tsk)
  42 {
  43         tsk->delays = kmem_cache_zalloc(delayacct_cache, GFP_KERNEL);
  44         if (tsk->delays)
  45                 spin_lock_init(&tsk->delays->lock);
  46 }
  47
  48 /*
  49  * Finish delay accounting for a statistic using its timestamps (@start),
  50  * accumalator (@total) and @count
  51  */
  52 static void delayacct_end(u64 *start, u64 *total, u32 *count)
  53 {
  54         s64 ns = ktime_get_ns() - *start;
  55         unsigned long flags;
  56
  57         if (ns > 0) {
  58                 spin_lock_irqsave(&current->delays->lock, flags);
  59                 *total += ns;
  60                 (*count)++;
  61                 spin_unlock_irqrestore(&current->delays->lock, flags);
  62         }
  63 }
  64
  65 void __delayacct_blkio_start(void)
  66 {
  67         current->delays->blkio_start = ktime_get_ns();
  68 }
  69
  70 void __delayacct_blkio_end(void)
  71 {
  72         if (current->delays->flags & DELAYACCT_PF_SWAPIN)
  73                 /* Swapin block I/O */
  74                 delayacct_end(&current->delays->blkio_start,
  75                         &current->delays->swapin_delay,
  76                         &current->delays->swapin_count);
  77         else    /* Other block I/O */
  78                 delayacct_end(&current->delays->blkio_start,
  79                         &current->delays->blkio_delay,
  80                         &current->delays->blkio_count);
  81 }
  82
  83 int __delayacct_add_tsk(struct taskstats *d, struct task_struct *tsk)
  84 {
  85         cputime_t utime, stime, stimescaled, utimescaled;
  86         unsigned long long t2, t3;
  87         unsigned long flags, t1;
  88         s64 tmp;
  89
  90         task_cputime(tsk, &utime, &stime);
  91         tmp = (s64)d->cpu_run_real_total;
  92         tmp += cputime_to_nsecs(utime + stime);
  93         d->cpu_run_real_total = (tmp < (s64)d->cpu_run_real_total) ? 0 : tmp;
  94
  95         task_cputime_scaled(tsk, &utimescaled, &stimescaled);
  96         tmp = (s64)d->cpu_scaled_run_real_total;
  97         tmp += cputime_to_nsecs(utimescaled + stimescaled);
  98         d->cpu_scaled_run_real_total =
  99                 (tmp < (s64)d->cpu_scaled_run_real_total) ? 0 : tmp;
 100
 101         /*
 102          * No locking available for sched_info (and too expensive to add one)
 103          * Mitigate by taking snapshot of values
 104          */
 105         t1 = tsk->sched_info.pcount;
 106         t2 = tsk->sched_info.run_delay;
 107         t3 = tsk->se.sum_exec_runtime;
 108
 109         d->cpu_count += t1;
 110
 111         tmp = (s64)d->cpu_delay_total + t2;
 112         d->cpu_delay_total = (tmp < (s64)d->cpu_delay_total) ? 0 : tmp;
 113
 114         tmp = (s64)d->cpu_run_virtual_total + t3;
 115         d->cpu_run_virtual_total =
 116                 (tmp < (s64)d->cpu_run_virtual_total) ? 0 : tmp;
 117
 118         /* zero XXX_total, non-zero XXX_count implies XXX stat overflowed */
 119
 120         spin_lock_irqsave(&tsk->delays->lock, flags);
 121         tmp = d->blkio_delay_total + tsk->delays->blkio_delay;
 122         d->blkio_delay_total = (tmp < d->blkio_delay_total) ? 0 : tmp;
 123         tmp = d->swapin_delay_total + tsk->delays->swapin_delay;
 124         d->swapin_delay_total = (tmp < d->swapin_delay_total) ? 0 : tmp;
 125         tmp = d->freepages_delay_total + tsk->delays->freepages_delay;
 126         d->freepages_delay_total = (tmp < d->freepages_delay_total) ? 0 : tmp;
 127         d->blkio_count += tsk->delays->blkio_count;
 128         d->swapin_count += tsk->delays->swapin_count;
 129         d->freepages_count += tsk->delays->freepages_count;
 130         spin_unlock_irqrestore(&tsk->delays->lock, flags);
 131
 132         return 0;
 133 }
 134
 135 __u64 __delayacct_blkio_ticks(struct task_struct *tsk)
 136 {
 137         __u64 ret;
 138         unsigned long flags;
 139
 140         spin_lock_irqsave(&tsk->delays->lock, flags);
 141         ret = nsec_to_clock_t(tsk->delays->blkio_delay +
 142                                 tsk->delays->swapin_delay);
 143         spin_unlock_irqrestore(&tsk->delays->lock, flags);
 144         return ret;
 145 }
 146
 147 void __delayacct_freepages_start(void)
 148 {
 149         current->delays->freepages_start = ktime_get_ns();
 150 }
 151
 152 void __delayacct_freepages_end(void)
 153 {
 154         delayacct_end(&current->delays->freepages_start,
 155                         &current->delays->freepages_delay,
 156                         &current->delays->freepages_count);
 157 }
 158