1 // SPDX-License-Identifier: GPL-2.0-or-later
2 /* delayacct.c - per-task delay accounting
4 * Copyright (C) Shailabh Nagar, IBM Corp. 2006
7 #include <linux/sched.h>
8 #include <linux/sched/task.h>
9 #include <linux/sched/cputime.h>
10 #include <linux/sched/clock.h>
11 #include <linux/slab.h>
12 #include <linux/taskstats.h>
13 #include <linux/sysctl.h>
14 #include <linux/delayacct.h>
15 #include <linux/module.h>
17 DEFINE_STATIC_KEY_FALSE(delayacct_key);
18 int delayacct_on __read_mostly; /* Delay accounting turned on/off */
19 struct kmem_cache *delayacct_cache;
21 static void set_delayacct(bool enabled)
24 static_branch_enable(&delayacct_key);
28 static_branch_disable(&delayacct_key);
32 static int __init delayacct_setup_enable(char *str)
37 __setup("delayacct", delayacct_setup_enable);
39 void delayacct_init(void)
41 delayacct_cache = KMEM_CACHE(task_delay_info, SLAB_PANIC|SLAB_ACCOUNT);
42 delayacct_tsk_init(&init_task);
43 set_delayacct(delayacct_on);
46 #ifdef CONFIG_PROC_SYSCTL
47 int sysctl_delayacct(struct ctl_table *table, int write, void *buffer,
48 size_t *lenp, loff_t *ppos)
50 int state = delayacct_on;
54 if (write && !capable(CAP_SYS_ADMIN))
59 err = proc_dointvec_minmax(&t, write, buffer, lenp, ppos);
68 void __delayacct_tsk_init(struct task_struct *tsk)
70 tsk->delays = kmem_cache_zalloc(delayacct_cache, GFP_KERNEL);
72 raw_spin_lock_init(&tsk->delays->lock);
76 * Finish delay accounting for a statistic using its timestamps (@start),
77 * accumalator (@total) and @count
79 static void delayacct_end(raw_spinlock_t *lock, u64 *start, u64 *total, u32 *count)
81 s64 ns = local_clock() - *start;
85 raw_spin_lock_irqsave(lock, flags);
88 raw_spin_unlock_irqrestore(lock, flags);
92 void __delayacct_blkio_start(void)
94 current->delays->blkio_start = local_clock();
98 * We cannot rely on the `current` macro, as we haven't yet switched back to
99 * the process being woken.
101 void __delayacct_blkio_end(struct task_struct *p)
103 delayacct_end(&p->delays->lock,
104 &p->delays->blkio_start,
105 &p->delays->blkio_delay,
106 &p->delays->blkio_count);
109 int delayacct_add_tsk(struct taskstats *d, struct task_struct *tsk)
111 u64 utime, stime, stimescaled, utimescaled;
112 unsigned long long t2, t3;
113 unsigned long flags, t1;
116 task_cputime(tsk, &utime, &stime);
117 tmp = (s64)d->cpu_run_real_total;
118 tmp += utime + stime;
119 d->cpu_run_real_total = (tmp < (s64)d->cpu_run_real_total) ? 0 : tmp;
121 task_cputime_scaled(tsk, &utimescaled, &stimescaled);
122 tmp = (s64)d->cpu_scaled_run_real_total;
123 tmp += utimescaled + stimescaled;
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;
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;
147 /* zero XXX_total, non-zero XXX_count implies XXX stat overflowed */
149 raw_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 tmp = d->thrashing_delay_total + tsk->delays->thrashing_delay;
157 d->thrashing_delay_total = (tmp < d->thrashing_delay_total) ? 0 : tmp;
158 tmp = d->compact_delay_total + tsk->delays->compact_delay;
159 d->compact_delay_total = (tmp < d->compact_delay_total) ? 0 : tmp;
160 d->blkio_count += tsk->delays->blkio_count;
161 d->swapin_count += tsk->delays->swapin_count;
162 d->freepages_count += tsk->delays->freepages_count;
163 d->thrashing_count += tsk->delays->thrashing_count;
164 d->compact_count += tsk->delays->compact_count;
165 raw_spin_unlock_irqrestore(&tsk->delays->lock, flags);
170 __u64 __delayacct_blkio_ticks(struct task_struct *tsk)
175 raw_spin_lock_irqsave(&tsk->delays->lock, flags);
176 ret = nsec_to_clock_t(tsk->delays->blkio_delay);
177 raw_spin_unlock_irqrestore(&tsk->delays->lock, flags);
181 void __delayacct_freepages_start(void)
183 current->delays->freepages_start = local_clock();
186 void __delayacct_freepages_end(void)
188 delayacct_end(¤t->delays->lock,
189 ¤t->delays->freepages_start,
190 ¤t->delays->freepages_delay,
191 ¤t->delays->freepages_count);
194 void __delayacct_thrashing_start(void)
196 current->delays->thrashing_start = local_clock();
199 void __delayacct_thrashing_end(void)
201 delayacct_end(¤t->delays->lock,
202 ¤t->delays->thrashing_start,
203 ¤t->delays->thrashing_delay,
204 ¤t->delays->thrashing_count);
207 void __delayacct_swapin_start(void)
209 current->delays->swapin_start = local_clock();
212 void __delayacct_swapin_end(void)
214 delayacct_end(¤t->delays->lock,
215 ¤t->delays->swapin_start,
216 ¤t->delays->swapin_delay,
217 ¤t->delays->swapin_count);
220 void __delayacct_compact_start(void)
222 current->delays->compact_start = local_clock();
225 void __delayacct_compact_end(void)
227 delayacct_end(¤t->delays->lock,
228 ¤t->delays->compact_start,
229 ¤t->delays->compact_delay,
230 ¤t->delays->compact_count);