1#include <linux/cgroup.h>
2#include <linux/slab.h>
3#include <linux/percpu.h>
4#include <linux/spinlock.h>
5#include <linux/cpumask.h>
6#include <linux/seq_file.h>
7#include <linux/rcupdate.h>
8#include <linux/kernel_stat.h>
9#include <linux/err.h>
10
11#include "sched.h"
12
13/*
14 * CPU accounting code for task groups.
15 *
16 * Based on the work by Paul Menage (menage@google.com) and Balbir Singh
17 * (balbir@in.ibm.com).
18 */
19
20/* Time spent by the tasks of the cpu accounting group executing in ... */
21enum cpuacct_stat_index {
22	CPUACCT_STAT_USER,	/* ... user mode */
23	CPUACCT_STAT_SYSTEM,	/* ... kernel mode */
24
25	CPUACCT_STAT_NSTATS,
26};
27
28/* track cpu usage of a group of tasks and its child groups */
29struct cpuacct {
30	struct cgroup_subsys_state css;
31	/* cpuusage holds pointer to a u64-type object on every cpu */
32	u64 __percpu *cpuusage;
33	struct kernel_cpustat __percpu *cpustat;
34};
35
36static inline struct cpuacct *css_ca(struct cgroup_subsys_state *css)
37{
38	return css ? container_of(css, struct cpuacct, css) : NULL;
39}
40
41/* return cpu accounting group to which this task belongs */
42static inline struct cpuacct *task_ca(struct task_struct *tsk)
43{
44	return css_ca(task_css(tsk, cpuacct_cgrp_id));
45}
46
47static inline struct cpuacct *parent_ca(struct cpuacct *ca)
48{
49	return css_ca(ca->css.parent);
50}
51
52static DEFINE_PER_CPU(u64, root_cpuacct_cpuusage);
53static struct cpuacct root_cpuacct = {
54	.cpustat	= &kernel_cpustat,
55	.cpuusage	= &root_cpuacct_cpuusage,
56};
57
58/* create a new cpu accounting group */
59static struct cgroup_subsys_state *
60cpuacct_css_alloc(struct cgroup_subsys_state *parent_css)
61{
62	struct cpuacct *ca;
63
64	if (!parent_css)
65		return &root_cpuacct.css;
66
67	ca = kzalloc(sizeof(*ca), GFP_KERNEL);
68	if (!ca)
69		goto out;
70
71	ca->cpuusage = alloc_percpu(u64);
72	if (!ca->cpuusage)
73		goto out_free_ca;
74
75	ca->cpustat = alloc_percpu(struct kernel_cpustat);
76	if (!ca->cpustat)
77		goto out_free_cpuusage;
78
79	return &ca->css;
80
81out_free_cpuusage:
82	free_percpu(ca->cpuusage);
83out_free_ca:
84	kfree(ca);
85out:
86	return ERR_PTR(-ENOMEM);
87}
88
89/* destroy an existing cpu accounting group */
90static void cpuacct_css_free(struct cgroup_subsys_state *css)
91{
92	struct cpuacct *ca = css_ca(css);
93
94	free_percpu(ca->cpustat);
95	free_percpu(ca->cpuusage);
96	kfree(ca);
97}
98
99static u64 cpuacct_cpuusage_read(struct cpuacct *ca, int cpu)
100{
101	u64 *cpuusage = per_cpu_ptr(ca->cpuusage, cpu);
102	u64 data;
103
104#ifndef CONFIG_64BIT
105	/*
106	 * Take rq->lock to make 64-bit read safe on 32-bit platforms.
107	 */
108	raw_spin_lock_irq(&cpu_rq(cpu)->lock);
109	data = *cpuusage;
110	raw_spin_unlock_irq(&cpu_rq(cpu)->lock);
111#else
112	data = *cpuusage;
113#endif
114
115	return data;
116}
117
118static void cpuacct_cpuusage_write(struct cpuacct *ca, int cpu, u64 val)
119{
120	u64 *cpuusage = per_cpu_ptr(ca->cpuusage, cpu);
121
122#ifndef CONFIG_64BIT
123	/*
124	 * Take rq->lock to make 64-bit write safe on 32-bit platforms.
125	 */
126	raw_spin_lock_irq(&cpu_rq(cpu)->lock);
127	*cpuusage = val;
128	raw_spin_unlock_irq(&cpu_rq(cpu)->lock);
129#else
130	*cpuusage = val;
131#endif
132}
133
134/* return total cpu usage (in nanoseconds) of a group */
135static u64 cpuusage_read(struct cgroup_subsys_state *css, struct cftype *cft)
136{
137	struct cpuacct *ca = css_ca(css);
138	u64 totalcpuusage = 0;
139	int i;
140
141	for_each_present_cpu(i)
142		totalcpuusage += cpuacct_cpuusage_read(ca, i);
143
144	return totalcpuusage;
145}
146
147static int cpuusage_write(struct cgroup_subsys_state *css, struct cftype *cft,
148			  u64 reset)
149{
150	struct cpuacct *ca = css_ca(css);
151	int err = 0;
152	int i;
153
154	if (reset) {
155		err = -EINVAL;
156		goto out;
157	}
158
159	for_each_present_cpu(i)
160		cpuacct_cpuusage_write(ca, i, 0);
161
162out:
163	return err;
164}
165
166static int cpuacct_percpu_seq_show(struct seq_file *m, void *V)
167{
168	struct cpuacct *ca = css_ca(seq_css(m));
169	u64 percpu;
170	int i;
171
172	for_each_present_cpu(i) {
173		percpu = cpuacct_cpuusage_read(ca, i);
174		seq_printf(m, "%llu ", (unsigned long long) percpu);
175	}
176	seq_printf(m, "\n");
177	return 0;
178}
179
180static const char * const cpuacct_stat_desc[] = {
181	[CPUACCT_STAT_USER] = "user",
182	[CPUACCT_STAT_SYSTEM] = "system",
183};
184
185static int cpuacct_stats_show(struct seq_file *sf, void *v)
186{
187	struct cpuacct *ca = css_ca(seq_css(sf));
188	int cpu;
189	s64 val = 0;
190
191	for_each_online_cpu(cpu) {
192		struct kernel_cpustat *kcpustat = per_cpu_ptr(ca->cpustat, cpu);
193		val += kcpustat->cpustat[CPUTIME_USER];
194		val += kcpustat->cpustat[CPUTIME_NICE];
195	}
196	val = cputime64_to_clock_t(val);
197	seq_printf(sf, "%s %lld\n", cpuacct_stat_desc[CPUACCT_STAT_USER], val);
198
199	val = 0;
200	for_each_online_cpu(cpu) {
201		struct kernel_cpustat *kcpustat = per_cpu_ptr(ca->cpustat, cpu);
202		val += kcpustat->cpustat[CPUTIME_SYSTEM];
203		val += kcpustat->cpustat[CPUTIME_IRQ];
204		val += kcpustat->cpustat[CPUTIME_SOFTIRQ];
205	}
206
207	val = cputime64_to_clock_t(val);
208	seq_printf(sf, "%s %lld\n", cpuacct_stat_desc[CPUACCT_STAT_SYSTEM], val);
209
210	return 0;
211}
212
213static struct cftype files[] = {
214	{
215		.name = "usage",
216		.read_u64 = cpuusage_read,
217		.write_u64 = cpuusage_write,
218	},
219	{
220		.name = "usage_percpu",
221		.seq_show = cpuacct_percpu_seq_show,
222	},
223	{
224		.name = "stat",
225		.seq_show = cpuacct_stats_show,
226	},
227	{ }	/* terminate */
228};
229
230/*
231 * charge this task's execution time to its accounting group.
232 *
233 * called with rq->lock held.
234 */
235void cpuacct_charge(struct task_struct *tsk, u64 cputime)
236{
237	struct cpuacct *ca;
238	int cpu;
239
240	cpu = task_cpu(tsk);
241
242	rcu_read_lock();
243
244	ca = task_ca(tsk);
245
246	while (true) {
247		u64 *cpuusage = per_cpu_ptr(ca->cpuusage, cpu);
248		*cpuusage += cputime;
249
250		ca = parent_ca(ca);
251		if (!ca)
252			break;
253	}
254
255	rcu_read_unlock();
256}
257
258/*
259 * Add user/system time to cpuacct.
260 *
261 * Note: it's the caller that updates the account of the root cgroup.
262 */
263void cpuacct_account_field(struct task_struct *p, int index, u64 val)
264{
265	struct kernel_cpustat *kcpustat;
266	struct cpuacct *ca;
267
268	rcu_read_lock();
269	ca = task_ca(p);
270	while (ca != &root_cpuacct) {
271		kcpustat = this_cpu_ptr(ca->cpustat);
272		kcpustat->cpustat[index] += val;
273		ca = parent_ca(ca);
274	}
275	rcu_read_unlock();
276}
277
278struct cgroup_subsys cpuacct_cgrp_subsys = {
279	.css_alloc	= cpuacct_css_alloc,
280	.css_free	= cpuacct_css_free,
281	.legacy_cftypes	= files,
282	.early_init	= 1,
283};
284