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); 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 * Start accounting for a delay statistic using 50 * its starting timestamp (@start) 51 */ 52 53 static inline void delayacct_start(struct timespec *start) 54 { 55 do_posix_clock_monotonic_gettime(start); 56 } 57 58 /* 59 * Finish delay accounting for a statistic using 60 * its timestamps (@start, @end), accumalator (@total) and @count 61 */ 62 63 static void delayacct_end(struct timespec *start, struct timespec *end, 64 u64 *total, u32 *count) 65 { 66 struct timespec ts; 67 s64 ns; 68 unsigned long flags; 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_irqsave(¤t->delays->lock, flags); 77 *total += ns; 78 (*count)++; 79 spin_unlock_irqrestore(¤t->delays->lock, flags); 80 } 81 82 void __delayacct_blkio_start(void) 83 { 84 delayacct_start(¤t->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(¤t->delays->blkio_start, 92 ¤t->delays->blkio_end, 93 ¤t->delays->swapin_delay, 94 ¤t->delays->swapin_count); 95 else /* Other block I/O */ 96 delayacct_end(¤t->delays->blkio_start, 97 ¤t->delays->blkio_end, 98 ¤t->delays->blkio_delay, 99 ¤t->delays->blkio_count); 100 } 101 102 int __delayacct_add_tsk(struct taskstats *d, struct task_struct *tsk) 103 { 104 s64 tmp; 105 unsigned long t1; 106 unsigned long long t2, t3; 107 unsigned long flags; 108 struct timespec ts; 109 cputime_t utime, stime, stimescaled, utimescaled; 110 111 /* Though tsk->delays accessed later, early exit avoids 112 * unnecessary returning of other data 113 */ 114 if (!tsk->delays) 115 goto done; 116 117 tmp = (s64)d->cpu_run_real_total; 118 task_cputime(tsk, &utime, &stime); 119 cputime_to_timespec(utime + stime, &ts); 120 tmp += timespec_to_ns(&ts); 121 d->cpu_run_real_total = (tmp < (s64)d->cpu_run_real_total) ? 0 : tmp; 122 123 tmp = (s64)d->cpu_scaled_run_real_total; 124 task_cputime_scaled(tsk, &utimescaled, &stimescaled); 125 cputime_to_timespec(utimescaled + stimescaled, &ts); 126 tmp += timespec_to_ns(&ts); 127 d->cpu_scaled_run_real_total = 128 (tmp < (s64)d->cpu_scaled_run_real_total) ? 0 : tmp; 129 130 /* 131 * No locking available for sched_info (and too expensive to add one) 132 * Mitigate by taking snapshot of values 133 */ 134 t1 = tsk->sched_info.pcount; 135 t2 = tsk->sched_info.run_delay; 136 t3 = tsk->se.sum_exec_runtime; 137 138 d->cpu_count += t1; 139 140 tmp = (s64)d->cpu_delay_total + t2; 141 d->cpu_delay_total = (tmp < (s64)d->cpu_delay_total) ? 0 : tmp; 142 143 tmp = (s64)d->cpu_run_virtual_total + t3; 144 d->cpu_run_virtual_total = 145 (tmp < (s64)d->cpu_run_virtual_total) ? 0 : tmp; 146 147 /* zero XXX_total, non-zero XXX_count implies XXX stat overflowed */ 148 149 spin_lock_irqsave(&tsk->delays->lock, flags); 150 tmp = d->blkio_delay_total + tsk->delays->blkio_delay; 151 d->blkio_delay_total = (tmp < d->blkio_delay_total) ? 0 : tmp; 152 tmp = d->swapin_delay_total + tsk->delays->swapin_delay; 153 d->swapin_delay_total = (tmp < d->swapin_delay_total) ? 0 : tmp; 154 tmp = d->freepages_delay_total + tsk->delays->freepages_delay; 155 d->freepages_delay_total = (tmp < d->freepages_delay_total) ? 0 : tmp; 156 d->blkio_count += tsk->delays->blkio_count; 157 d->swapin_count += tsk->delays->swapin_count; 158 d->freepages_count += tsk->delays->freepages_count; 159 spin_unlock_irqrestore(&tsk->delays->lock, flags); 160 161 done: 162 return 0; 163 } 164 165 __u64 __delayacct_blkio_ticks(struct task_struct *tsk) 166 { 167 __u64 ret; 168 unsigned long flags; 169 170 spin_lock_irqsave(&tsk->delays->lock, flags); 171 ret = nsec_to_clock_t(tsk->delays->blkio_delay + 172 tsk->delays->swapin_delay); 173 spin_unlock_irqrestore(&tsk->delays->lock, flags); 174 return ret; 175 } 176 177 void __delayacct_freepages_start(void) 178 { 179 delayacct_start(¤t->delays->freepages_start); 180 } 181 182 void __delayacct_freepages_end(void) 183 { 184 delayacct_end(¤t->delays->freepages_start, 185 ¤t->delays->freepages_end, 186 ¤t->delays->freepages_delay, 187 ¤t->delays->freepages_count); 188 } 189 190