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 23 int delayacct_on __read_mostly = 1; /* Delay accounting turned on/off */ 24 struct kmem_cache *delayacct_cache; 25 26 static int __init delayacct_setup_disable(char *str) 27 { 28 delayacct_on = 0; 29 return 1; 30 } 31 __setup("nodelayacct", delayacct_setup_disable); 32 33 void delayacct_init(void) 34 { 35 delayacct_cache = KMEM_CACHE(task_delay_info, SLAB_PANIC); 36 delayacct_tsk_init(&init_task); 37 } 38 39 void __delayacct_tsk_init(struct task_struct *tsk) 40 { 41 tsk->delays = kmem_cache_zalloc(delayacct_cache, GFP_KERNEL); 42 if (tsk->delays) 43 spin_lock_init(&tsk->delays->lock); 44 } 45 46 /* 47 * Start accounting for a delay statistic using 48 * its starting timestamp (@start) 49 */ 50 51 static inline void delayacct_start(struct timespec *start) 52 { 53 do_posix_clock_monotonic_gettime(start); 54 } 55 56 /* 57 * Finish delay accounting for a statistic using 58 * its timestamps (@start, @end), accumalator (@total) and @count 59 */ 60 61 static void delayacct_end(struct timespec *start, struct timespec *end, 62 u64 *total, u32 *count) 63 { 64 struct timespec ts; 65 s64 ns; 66 unsigned long flags; 67 68 do_posix_clock_monotonic_gettime(end); 69 ts = timespec_sub(*end, *start); 70 ns = timespec_to_ns(&ts); 71 if (ns < 0) 72 return; 73 74 spin_lock_irqsave(¤t->delays->lock, flags); 75 *total += ns; 76 (*count)++; 77 spin_unlock_irqrestore(¤t->delays->lock, flags); 78 } 79 80 void __delayacct_blkio_start(void) 81 { 82 delayacct_start(¤t->delays->blkio_start); 83 } 84 85 void __delayacct_blkio_end(void) 86 { 87 if (current->delays->flags & DELAYACCT_PF_SWAPIN) 88 /* Swapin block I/O */ 89 delayacct_end(¤t->delays->blkio_start, 90 ¤t->delays->blkio_end, 91 ¤t->delays->swapin_delay, 92 ¤t->delays->swapin_count); 93 else /* Other block I/O */ 94 delayacct_end(¤t->delays->blkio_start, 95 ¤t->delays->blkio_end, 96 ¤t->delays->blkio_delay, 97 ¤t->delays->blkio_count); 98 } 99 100 int __delayacct_add_tsk(struct taskstats *d, struct task_struct *tsk) 101 { 102 s64 tmp; 103 unsigned long t1; 104 unsigned long long t2, t3; 105 unsigned long flags; 106 struct timespec ts; 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 tmp = (s64)d->cpu_scaled_run_real_total; 120 cputime_to_timespec(tsk->utimescaled + tsk->stimescaled, &ts); 121 tmp += timespec_to_ns(&ts); 122 d->cpu_scaled_run_real_total = 123 (tmp < (s64)d->cpu_scaled_run_real_total) ? 0 : tmp; 124 125 /* 126 * No locking available for sched_info (and too expensive to add one) 127 * Mitigate by taking snapshot of values 128 */ 129 t1 = tsk->sched_info.pcount; 130 t2 = tsk->sched_info.run_delay; 131 t3 = tsk->se.sum_exec_runtime; 132 133 d->cpu_count += t1; 134 135 tmp = (s64)d->cpu_delay_total + t2; 136 d->cpu_delay_total = (tmp < (s64)d->cpu_delay_total) ? 0 : tmp; 137 138 tmp = (s64)d->cpu_run_virtual_total + t3; 139 d->cpu_run_virtual_total = 140 (tmp < (s64)d->cpu_run_virtual_total) ? 0 : tmp; 141 142 /* zero XXX_total, non-zero XXX_count implies XXX stat overflowed */ 143 144 spin_lock_irqsave(&tsk->delays->lock, flags); 145 tmp = d->blkio_delay_total + tsk->delays->blkio_delay; 146 d->blkio_delay_total = (tmp < d->blkio_delay_total) ? 0 : tmp; 147 tmp = d->swapin_delay_total + tsk->delays->swapin_delay; 148 d->swapin_delay_total = (tmp < d->swapin_delay_total) ? 0 : tmp; 149 tmp = d->freepages_delay_total + tsk->delays->freepages_delay; 150 d->freepages_delay_total = (tmp < d->freepages_delay_total) ? 0 : tmp; 151 d->blkio_count += tsk->delays->blkio_count; 152 d->swapin_count += tsk->delays->swapin_count; 153 d->freepages_count += tsk->delays->freepages_count; 154 spin_unlock_irqrestore(&tsk->delays->lock, flags); 155 156 done: 157 return 0; 158 } 159 160 __u64 __delayacct_blkio_ticks(struct task_struct *tsk) 161 { 162 __u64 ret; 163 unsigned long flags; 164 165 spin_lock_irqsave(&tsk->delays->lock, flags); 166 ret = nsec_to_clock_t(tsk->delays->blkio_delay + 167 tsk->delays->swapin_delay); 168 spin_unlock_irqrestore(&tsk->delays->lock, flags); 169 return ret; 170 } 171 172 void __delayacct_freepages_start(void) 173 { 174 delayacct_start(¤t->delays->freepages_start); 175 } 176 177 void __delayacct_freepages_end(void) 178 { 179 delayacct_end(¤t->delays->freepages_start, 180 ¤t->delays->freepages_end, 181 ¤t->delays->freepages_delay, 182 ¤t->delays->freepages_count); 183 } 184 185