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>
14 * CPU accounting code for task groups.
16 * Based on the work by Paul Menage (menage@google.com) and Balbir Singh
17 * (balbir@in.ibm.com).
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 */
28 static const char * const cpuacct_stat_desc
[] = {
29 [CPUACCT_STAT_USER
] = "user",
30 [CPUACCT_STAT_SYSTEM
] = "system",
33 struct cpuacct_usage
{
34 u64 usages
[CPUACCT_STAT_NSTATS
];
37 /* track cpu usage of a group of tasks and its child groups */
39 struct cgroup_subsys_state css
;
40 /* cpuusage holds pointer to a u64-type object on every cpu */
41 struct cpuacct_usage __percpu
*cpuusage
;
42 struct kernel_cpustat __percpu
*cpustat
;
45 static inline struct cpuacct
*css_ca(struct cgroup_subsys_state
*css
)
47 return css
? container_of(css
, struct cpuacct
, css
) : NULL
;
50 /* return cpu accounting group to which this task belongs */
51 static inline struct cpuacct
*task_ca(struct task_struct
*tsk
)
53 return css_ca(task_css(tsk
, cpuacct_cgrp_id
));
56 static inline struct cpuacct
*parent_ca(struct cpuacct
*ca
)
58 return css_ca(ca
->css
.parent
);
61 static DEFINE_PER_CPU(struct cpuacct_usage
, root_cpuacct_cpuusage
);
62 static struct cpuacct root_cpuacct
= {
63 .cpustat
= &kernel_cpustat
,
64 .cpuusage
= &root_cpuacct_cpuusage
,
67 /* create a new cpu accounting group */
68 static struct cgroup_subsys_state
*
69 cpuacct_css_alloc(struct cgroup_subsys_state
*parent_css
)
74 return &root_cpuacct
.css
;
76 ca
= kzalloc(sizeof(*ca
), GFP_KERNEL
);
80 ca
->cpuusage
= alloc_percpu(struct cpuacct_usage
);
84 ca
->cpustat
= alloc_percpu(struct kernel_cpustat
);
86 goto out_free_cpuusage
;
91 free_percpu(ca
->cpuusage
);
95 return ERR_PTR(-ENOMEM
);
98 /* destroy an existing cpu accounting group */
99 static void cpuacct_css_free(struct cgroup_subsys_state
*css
)
101 struct cpuacct
*ca
= css_ca(css
);
103 free_percpu(ca
->cpustat
);
104 free_percpu(ca
->cpuusage
);
108 static u64
cpuacct_cpuusage_read(struct cpuacct
*ca
, int cpu
,
109 enum cpuacct_stat_index index
)
111 struct cpuacct_usage
*cpuusage
= per_cpu_ptr(ca
->cpuusage
, cpu
);
115 * We allow index == CPUACCT_STAT_NSTATS here to read
118 BUG_ON(index
> CPUACCT_STAT_NSTATS
);
122 * Take rq->lock to make 64-bit read safe on 32-bit platforms.
124 raw_spin_lock_irq(&cpu_rq(cpu
)->lock
);
127 if (index
== CPUACCT_STAT_NSTATS
) {
131 for (i
= 0; i
< CPUACCT_STAT_NSTATS
; i
++)
132 data
+= cpuusage
->usages
[i
];
134 data
= cpuusage
->usages
[index
];
138 raw_spin_unlock_irq(&cpu_rq(cpu
)->lock
);
144 static void cpuacct_cpuusage_write(struct cpuacct
*ca
, int cpu
, u64 val
)
146 struct cpuacct_usage
*cpuusage
= per_cpu_ptr(ca
->cpuusage
, cpu
);
151 * Take rq->lock to make 64-bit write safe on 32-bit platforms.
153 raw_spin_lock_irq(&cpu_rq(cpu
)->lock
);
156 for (i
= 0; i
< CPUACCT_STAT_NSTATS
; i
++)
157 cpuusage
->usages
[i
] = val
;
160 raw_spin_unlock_irq(&cpu_rq(cpu
)->lock
);
164 /* return total cpu usage (in nanoseconds) of a group */
165 static u64
__cpuusage_read(struct cgroup_subsys_state
*css
,
166 enum cpuacct_stat_index index
)
168 struct cpuacct
*ca
= css_ca(css
);
169 u64 totalcpuusage
= 0;
172 for_each_possible_cpu(i
)
173 totalcpuusage
+= cpuacct_cpuusage_read(ca
, i
, index
);
175 return totalcpuusage
;
178 static u64
cpuusage_user_read(struct cgroup_subsys_state
*css
,
181 return __cpuusage_read(css
, CPUACCT_STAT_USER
);
184 static u64
cpuusage_sys_read(struct cgroup_subsys_state
*css
,
187 return __cpuusage_read(css
, CPUACCT_STAT_SYSTEM
);
190 static u64
cpuusage_read(struct cgroup_subsys_state
*css
, struct cftype
*cft
)
192 return __cpuusage_read(css
, CPUACCT_STAT_NSTATS
);
195 static int cpuusage_write(struct cgroup_subsys_state
*css
, struct cftype
*cft
,
198 struct cpuacct
*ca
= css_ca(css
);
202 * Only allow '0' here to do a reset.
207 for_each_possible_cpu(cpu
)
208 cpuacct_cpuusage_write(ca
, cpu
, 0);
213 static int __cpuacct_percpu_seq_show(struct seq_file
*m
,
214 enum cpuacct_stat_index index
)
216 struct cpuacct
*ca
= css_ca(seq_css(m
));
220 for_each_possible_cpu(i
) {
221 percpu
= cpuacct_cpuusage_read(ca
, i
, index
);
222 seq_printf(m
, "%llu ", (unsigned long long) percpu
);
228 static int cpuacct_percpu_user_seq_show(struct seq_file
*m
, void *V
)
230 return __cpuacct_percpu_seq_show(m
, CPUACCT_STAT_USER
);
233 static int cpuacct_percpu_sys_seq_show(struct seq_file
*m
, void *V
)
235 return __cpuacct_percpu_seq_show(m
, CPUACCT_STAT_SYSTEM
);
238 static int cpuacct_percpu_seq_show(struct seq_file
*m
, void *V
)
240 return __cpuacct_percpu_seq_show(m
, CPUACCT_STAT_NSTATS
);
243 static int cpuacct_all_seq_show(struct seq_file
*m
, void *V
)
245 struct cpuacct
*ca
= css_ca(seq_css(m
));
250 for (index
= 0; index
< CPUACCT_STAT_NSTATS
; index
++)
251 seq_printf(m
, " %s", cpuacct_stat_desc
[index
]);
254 for_each_possible_cpu(cpu
) {
255 struct cpuacct_usage
*cpuusage
= per_cpu_ptr(ca
->cpuusage
, cpu
);
257 seq_printf(m
, "%d", cpu
);
259 for (index
= 0; index
< CPUACCT_STAT_NSTATS
; index
++) {
262 * Take rq->lock to make 64-bit read safe on 32-bit
265 raw_spin_lock_irq(&cpu_rq(cpu
)->lock
);
268 seq_printf(m
, " %llu", cpuusage
->usages
[index
]);
271 raw_spin_unlock_irq(&cpu_rq(cpu
)->lock
);
279 static int cpuacct_stats_show(struct seq_file
*sf
, void *v
)
281 struct cpuacct
*ca
= css_ca(seq_css(sf
));
282 s64 val
[CPUACCT_STAT_NSTATS
];
286 memset(val
, 0, sizeof(val
));
287 for_each_possible_cpu(cpu
) {
288 u64
*cpustat
= per_cpu_ptr(ca
->cpustat
, cpu
)->cpustat
;
290 val
[CPUACCT_STAT_USER
] += cpustat
[CPUTIME_USER
];
291 val
[CPUACCT_STAT_USER
] += cpustat
[CPUTIME_NICE
];
292 val
[CPUACCT_STAT_SYSTEM
] += cpustat
[CPUTIME_SYSTEM
];
293 val
[CPUACCT_STAT_SYSTEM
] += cpustat
[CPUTIME_IRQ
];
294 val
[CPUACCT_STAT_SYSTEM
] += cpustat
[CPUTIME_SOFTIRQ
];
297 for (stat
= 0; stat
< CPUACCT_STAT_NSTATS
; stat
++) {
298 seq_printf(sf
, "%s %lld\n",
299 cpuacct_stat_desc
[stat
],
300 (long long)cputime64_to_clock_t(val
[stat
]));
306 static struct cftype files
[] = {
309 .read_u64
= cpuusage_read
,
310 .write_u64
= cpuusage_write
,
313 .name
= "usage_user",
314 .read_u64
= cpuusage_user_read
,
318 .read_u64
= cpuusage_sys_read
,
321 .name
= "usage_percpu",
322 .seq_show
= cpuacct_percpu_seq_show
,
325 .name
= "usage_percpu_user",
326 .seq_show
= cpuacct_percpu_user_seq_show
,
329 .name
= "usage_percpu_sys",
330 .seq_show
= cpuacct_percpu_sys_seq_show
,
334 .seq_show
= cpuacct_all_seq_show
,
338 .seq_show
= cpuacct_stats_show
,
344 * charge this task's execution time to its accounting group.
346 * called with rq->lock held.
348 void cpuacct_charge(struct task_struct
*tsk
, u64 cputime
)
351 int index
= CPUACCT_STAT_SYSTEM
;
352 struct pt_regs
*regs
= task_pt_regs(tsk
);
354 if (regs
&& user_mode(regs
))
355 index
= CPUACCT_STAT_USER
;
359 for (ca
= task_ca(tsk
); ca
; ca
= parent_ca(ca
))
360 this_cpu_ptr(ca
->cpuusage
)->usages
[index
] += cputime
;
366 * Add user/system time to cpuacct.
368 * Note: it's the caller that updates the account of the root cgroup.
370 void cpuacct_account_field(struct task_struct
*tsk
, int index
, u64 val
)
375 for (ca
= task_ca(tsk
); ca
!= &root_cpuacct
; ca
= parent_ca(ca
))
376 this_cpu_ptr(ca
->cpustat
)->cpustat
[index
] += val
;
380 struct cgroup_subsys cpuacct_cgrp_subsys
= {
381 .css_alloc
= cpuacct_css_alloc
,
382 .css_free
= cpuacct_css_free
,
383 .legacy_cftypes
= files
,