xref: /openbmc/linux/kernel/sched/cpuacct.c (revision cd5d5810)
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 ... */
21 enum 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 */
29 struct 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 
36 static 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 */
42 static inline struct cpuacct *task_ca(struct task_struct *tsk)
43 {
44 	return css_ca(task_css(tsk, cpuacct_subsys_id));
45 }
46 
47 static inline struct cpuacct *parent_ca(struct cpuacct *ca)
48 {
49 	return css_ca(css_parent(&ca->css));
50 }
51 
52 static DEFINE_PER_CPU(u64, root_cpuacct_cpuusage);
53 static struct cpuacct root_cpuacct = {
54 	.cpustat	= &kernel_cpustat,
55 	.cpuusage	= &root_cpuacct_cpuusage,
56 };
57 
58 /* create a new cpu accounting group */
59 static struct cgroup_subsys_state *
60 cpuacct_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 
81 out_free_cpuusage:
82 	free_percpu(ca->cpuusage);
83 out_free_ca:
84 	kfree(ca);
85 out:
86 	return ERR_PTR(-ENOMEM);
87 }
88 
89 /* destroy an existing cpu accounting group */
90 static 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 
99 static 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 
118 static 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 */
135 static 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 
147 static 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 
162 out:
163 	return err;
164 }
165 
166 static int cpuacct_percpu_seq_read(struct cgroup_subsys_state *css,
167 				   struct cftype *cft, struct seq_file *m)
168 {
169 	struct cpuacct *ca = css_ca(css);
170 	u64 percpu;
171 	int i;
172 
173 	for_each_present_cpu(i) {
174 		percpu = cpuacct_cpuusage_read(ca, i);
175 		seq_printf(m, "%llu ", (unsigned long long) percpu);
176 	}
177 	seq_printf(m, "\n");
178 	return 0;
179 }
180 
181 static const char * const cpuacct_stat_desc[] = {
182 	[CPUACCT_STAT_USER] = "user",
183 	[CPUACCT_STAT_SYSTEM] = "system",
184 };
185 
186 static int cpuacct_stats_show(struct cgroup_subsys_state *css,
187 			      struct cftype *cft, struct cgroup_map_cb *cb)
188 {
189 	struct cpuacct *ca = css_ca(css);
190 	int cpu;
191 	s64 val = 0;
192 
193 	for_each_online_cpu(cpu) {
194 		struct kernel_cpustat *kcpustat = per_cpu_ptr(ca->cpustat, cpu);
195 		val += kcpustat->cpustat[CPUTIME_USER];
196 		val += kcpustat->cpustat[CPUTIME_NICE];
197 	}
198 	val = cputime64_to_clock_t(val);
199 	cb->fill(cb, cpuacct_stat_desc[CPUACCT_STAT_USER], val);
200 
201 	val = 0;
202 	for_each_online_cpu(cpu) {
203 		struct kernel_cpustat *kcpustat = per_cpu_ptr(ca->cpustat, cpu);
204 		val += kcpustat->cpustat[CPUTIME_SYSTEM];
205 		val += kcpustat->cpustat[CPUTIME_IRQ];
206 		val += kcpustat->cpustat[CPUTIME_SOFTIRQ];
207 	}
208 
209 	val = cputime64_to_clock_t(val);
210 	cb->fill(cb, cpuacct_stat_desc[CPUACCT_STAT_SYSTEM], val);
211 
212 	return 0;
213 }
214 
215 static struct cftype files[] = {
216 	{
217 		.name = "usage",
218 		.read_u64 = cpuusage_read,
219 		.write_u64 = cpuusage_write,
220 	},
221 	{
222 		.name = "usage_percpu",
223 		.read_seq_string = cpuacct_percpu_seq_read,
224 	},
225 	{
226 		.name = "stat",
227 		.read_map = cpuacct_stats_show,
228 	},
229 	{ }	/* terminate */
230 };
231 
232 /*
233  * charge this task's execution time to its accounting group.
234  *
235  * called with rq->lock held.
236  */
237 void cpuacct_charge(struct task_struct *tsk, u64 cputime)
238 {
239 	struct cpuacct *ca;
240 	int cpu;
241 
242 	cpu = task_cpu(tsk);
243 
244 	rcu_read_lock();
245 
246 	ca = task_ca(tsk);
247 
248 	while (true) {
249 		u64 *cpuusage = per_cpu_ptr(ca->cpuusage, cpu);
250 		*cpuusage += cputime;
251 
252 		ca = parent_ca(ca);
253 		if (!ca)
254 			break;
255 	}
256 
257 	rcu_read_unlock();
258 }
259 
260 /*
261  * Add user/system time to cpuacct.
262  *
263  * Note: it's the caller that updates the account of the root cgroup.
264  */
265 void cpuacct_account_field(struct task_struct *p, int index, u64 val)
266 {
267 	struct kernel_cpustat *kcpustat;
268 	struct cpuacct *ca;
269 
270 	rcu_read_lock();
271 	ca = task_ca(p);
272 	while (ca != &root_cpuacct) {
273 		kcpustat = this_cpu_ptr(ca->cpustat);
274 		kcpustat->cpustat[index] += val;
275 		ca = parent_ca(ca);
276 	}
277 	rcu_read_unlock();
278 }
279 
280 struct cgroup_subsys cpuacct_subsys = {
281 	.name		= "cpuacct",
282 	.css_alloc	= cpuacct_css_alloc,
283 	.css_free	= cpuacct_css_free,
284 	.subsys_id	= cpuacct_subsys_id,
285 	.base_cftypes	= files,
286 	.early_init	= 1,
287 };
288