1// SPDX-License-Identifier: GPL-2.0-or-later
  2/* delayacct.c - per-task delay accounting
  3 *
  4 * Copyright (C) Shailabh Nagar, IBM Corp. 2006
  5 */
  6
  7#include <linux/sched.h>
  8#include <linux/sched/task.h>
  9#include <linux/sched/cputime.h>
 10#include <linux/slab.h>
 11#include <linux/taskstats.h>
 12#include <linux/time.h>
 13#include <linux/sysctl.h>
 14#include <linux/delayacct.h>
 15#include <linux/module.h>
 16
 17int delayacct_on __read_mostly = 1;	/* Delay accounting turned on/off */
 18EXPORT_SYMBOL_GPL(delayacct_on);
 19struct kmem_cache *delayacct_cache;
 20
 21static int __init delayacct_setup_disable(char *str)
 22{
 23	delayacct_on = 0;
 24	return 1;
 25}
 26__setup("nodelayacct", delayacct_setup_disable);
 27
 28void delayacct_init(void)
 29{
 30	delayacct_cache = KMEM_CACHE(task_delay_info, SLAB_PANIC|SLAB_ACCOUNT);
 31	delayacct_tsk_init(&init_task);
 32}
 33
 34void __delayacct_tsk_init(struct task_struct *tsk)
 35{
 36	tsk->delays = kmem_cache_zalloc(delayacct_cache, GFP_KERNEL);
 37	if (tsk->delays)
 38		raw_spin_lock_init(&tsk->delays->lock);
 39}
 40
 41/*
 42 * Finish delay accounting for a statistic using its timestamps (@start),
 43 * accumalator (@total) and @count
 44 */
 45static void delayacct_end(raw_spinlock_t *lock, u64 *start, u64 *total,
 46			  u32 *count)
 47{
 48	s64 ns = ktime_get_ns() - *start;
 49	unsigned long flags;
 50
 51	if (ns > 0) {
 52		raw_spin_lock_irqsave(lock, flags);
 53		*total += ns;
 54		(*count)++;
 55		raw_spin_unlock_irqrestore(lock, flags);
 56	}
 57}
 58
 59void __delayacct_blkio_start(void)
 60{
 61	current->delays->blkio_start = ktime_get_ns();
 62}
 63
 64/*
 65 * We cannot rely on the `current` macro, as we haven't yet switched back to
 66 * the process being woken.
 67 */
 68void __delayacct_blkio_end(struct task_struct *p)
 69{
 70	struct task_delay_info *delays = p->delays;
 71	u64 *total;
 72	u32 *count;
 73
 74	if (p->delays->flags & DELAYACCT_PF_SWAPIN) {
 75		total = &delays->swapin_delay;
 76		count = &delays->swapin_count;
 77	} else {
 78		total = &delays->blkio_delay;
 79		count = &delays->blkio_count;
 80	}
 81
 82	delayacct_end(&delays->lock, &delays->blkio_start, total, count);
 83}
 84
 85int __delayacct_add_tsk(struct taskstats *d, struct task_struct *tsk)
 86{
 87	u64 utime, stime, stimescaled, utimescaled;
 88	unsigned long long t2, t3;
 89	unsigned long flags, t1;
 90	s64 tmp;
 91
 92	task_cputime(tsk, &utime, &stime);
 93	tmp = (s64)d->cpu_run_real_total;
 94	tmp += utime + stime;
 95	d->cpu_run_real_total = (tmp < (s64)d->cpu_run_real_total) ? 0 : tmp;
 96
 97	task_cputime_scaled(tsk, &utimescaled, &stimescaled);
 98	tmp = (s64)d->cpu_scaled_run_real_total;
 99	tmp += utimescaled + stimescaled;
100	d->cpu_scaled_run_real_total =
101		(tmp < (s64)d->cpu_scaled_run_real_total) ? 0 : tmp;
102
103	/*
104	 * No locking available for sched_info (and too expensive to add one)
105	 * Mitigate by taking snapshot of values
106	 */
107	t1 = tsk->sched_info.pcount;
108	t2 = tsk->sched_info.run_delay;
109	t3 = tsk->se.sum_exec_runtime;
110
111	d->cpu_count += t1;
112
113	tmp = (s64)d->cpu_delay_total + t2;
114	d->cpu_delay_total = (tmp < (s64)d->cpu_delay_total) ? 0 : tmp;
115
116	tmp = (s64)d->cpu_run_virtual_total + t3;
117	d->cpu_run_virtual_total =
118		(tmp < (s64)d->cpu_run_virtual_total) ?	0 : tmp;
119
120	/* zero XXX_total, non-zero XXX_count implies XXX stat overflowed */
121
122	raw_spin_lock_irqsave(&tsk->delays->lock, flags);
123	tmp = d->blkio_delay_total + tsk->delays->blkio_delay;
124	d->blkio_delay_total = (tmp < d->blkio_delay_total) ? 0 : tmp;
125	tmp = d->swapin_delay_total + tsk->delays->swapin_delay;
126	d->swapin_delay_total = (tmp < d->swapin_delay_total) ? 0 : tmp;
127	tmp = d->freepages_delay_total + tsk->delays->freepages_delay;
128	d->freepages_delay_total = (tmp < d->freepages_delay_total) ? 0 : tmp;
129	tmp = d->thrashing_delay_total + tsk->delays->thrashing_delay;
130	d->thrashing_delay_total = (tmp < d->thrashing_delay_total) ? 0 : tmp;
131	d->blkio_count += tsk->delays->blkio_count;
132	d->swapin_count += tsk->delays->swapin_count;
133	d->freepages_count += tsk->delays->freepages_count;
134	d->thrashing_count += tsk->delays->thrashing_count;
135	raw_spin_unlock_irqrestore(&tsk->delays->lock, flags);
136
137	return 0;
138}
139
140__u64 __delayacct_blkio_ticks(struct task_struct *tsk)
141{
142	__u64 ret;
143	unsigned long flags;
144
145	raw_spin_lock_irqsave(&tsk->delays->lock, flags);
146	ret = nsec_to_clock_t(tsk->delays->blkio_delay +
147				tsk->delays->swapin_delay);
148	raw_spin_unlock_irqrestore(&tsk->delays->lock, flags);
149	return ret;
150}
151
152void __delayacct_freepages_start(void)
153{
154	current->delays->freepages_start = ktime_get_ns();
155}
156
157void __delayacct_freepages_end(void)
158{
159	delayacct_end(
160		&current->delays->lock,
161		&current->delays->freepages_start,
162		&current->delays->freepages_delay,
163		&current->delays->freepages_count);
164}
165
166void __delayacct_thrashing_start(void)
167{
168	current->delays->thrashing_start = ktime_get_ns();
169}
170
171void __delayacct_thrashing_end(void)
172{
173	delayacct_end(&current->delays->lock,
174		      &current->delays->thrashing_start,
175		      &current->delays->thrashing_delay,
176		      &current->delays->thrashing_count);
177}

  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
 24int delayacct_on __read_mostly = 1;	/* Delay accounting turned on/off */
 25EXPORT_SYMBOL_GPL(delayacct_on);
 26struct kmem_cache *delayacct_cache;
 27
 28static int __init delayacct_setup_disable(char *str)
 29{
 30	delayacct_on = 0;
 31	return 1;
 32}
 33__setup("nodelayacct", delayacct_setup_disable);
 34
 35void delayacct_init(void)
 36{
 37	delayacct_cache = KMEM_CACHE(task_delay_info, SLAB_PANIC);
 38	delayacct_tsk_init(&init_task);
 39}
 40
 41void __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
 53static 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
 63static 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(&current->delays->lock, flags);
 77	*total += ns;
 78	(*count)++;
 79	spin_unlock_irqrestore(&current->delays->lock, flags);
 80}
 81
 82void __delayacct_blkio_start(void)
 83{
 84	delayacct_start(&current->delays->blkio_start);
 85}
 86
 87void __delayacct_blkio_end(void)
 88{
 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);
100}
101
102int __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	tmp = (s64)d->cpu_run_real_total;
112	task_cputime(tsk, &utime, &stime);
113	cputime_to_timespec(utime + stime, &ts);
114	tmp += timespec_to_ns(&ts);
115	d->cpu_run_real_total = (tmp < (s64)d->cpu_run_real_total) ? 0 : tmp;
116
117	tmp = (s64)d->cpu_scaled_run_real_total;
118	task_cputime_scaled(tsk, &utimescaled, &stimescaled);
119	cputime_to_timespec(utimescaled + stimescaled, &ts);
120	tmp += timespec_to_ns(&ts);
121	d->cpu_scaled_run_real_total =
122		(tmp < (s64)d->cpu_scaled_run_real_total) ? 0 : tmp;
123
124	/*
125	 * No locking available for sched_info (and too expensive to add one)
126	 * Mitigate by taking snapshot of values
127	 */
128	t1 = tsk->sched_info.pcount;
129	t2 = tsk->sched_info.run_delay;
130	t3 = tsk->se.sum_exec_runtime;
131
132	d->cpu_count += t1;
133
134	tmp = (s64)d->cpu_delay_total + t2;
135	d->cpu_delay_total = (tmp < (s64)d->cpu_delay_total) ? 0 : tmp;
136
137	tmp = (s64)d->cpu_run_virtual_total + t3;
138	d->cpu_run_virtual_total =
139		(tmp < (s64)d->cpu_run_virtual_total) ?	0 : tmp;
140
141	/* zero XXX_total, non-zero XXX_count implies XXX stat overflowed */
142
143	spin_lock_irqsave(&tsk->delays->lock, flags);
144	tmp = d->blkio_delay_total + tsk->delays->blkio_delay;
145	d->blkio_delay_total = (tmp < d->blkio_delay_total) ? 0 : tmp;
146	tmp = d->swapin_delay_total + tsk->delays->swapin_delay;
147	d->swapin_delay_total = (tmp < d->swapin_delay_total) ? 0 : tmp;
148	tmp = d->freepages_delay_total + tsk->delays->freepages_delay;
149	d->freepages_delay_total = (tmp < d->freepages_delay_total) ? 0 : tmp;
150	d->blkio_count += tsk->delays->blkio_count;
151	d->swapin_count += tsk->delays->swapin_count;
152	d->freepages_count += tsk->delays->freepages_count;
153	spin_unlock_irqrestore(&tsk->delays->lock, flags);
154
155	return 0;
156}
157
158__u64 __delayacct_blkio_ticks(struct task_struct *tsk)
159{
160	__u64 ret;
161	unsigned long flags;
162
163	spin_lock_irqsave(&tsk->delays->lock, flags);
164	ret = nsec_to_clock_t(tsk->delays->blkio_delay +
165				tsk->delays->swapin_delay);
166	spin_unlock_irqrestore(&tsk->delays->lock, flags);
167	return ret;
168}
169
170void __delayacct_freepages_start(void)
171{
172	delayacct_start(&current->delays->freepages_start);
173}
174
175void __delayacct_freepages_end(void)
176{
177	delayacct_end(&current->delays->freepages_start,
178			&current->delays->freepages_end,
179			&current->delays->freepages_delay,
180			&current->delays->freepages_count);
181}
182