]>
Commit | Line | Data |
---|---|---|
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 | enum cpuacct_usage_index { | |
29 | CPUACCT_USAGE_USER, /* ... user mode */ | |
30 | CPUACCT_USAGE_SYSTEM, /* ... kernel mode */ | |
31 | ||
32 | CPUACCT_USAGE_NRUSAGE, | |
33 | }; | |
34 | ||
35 | struct cpuacct_usage { | |
36 | u64 usages[CPUACCT_USAGE_NRUSAGE]; | |
37 | }; | |
38 | ||
39 | /* track cpu usage of a group of tasks and its child groups */ | |
40 | struct cpuacct { | |
41 | struct cgroup_subsys_state css; | |
42 | /* cpuusage holds pointer to a u64-type object on every cpu */ | |
43 | struct cpuacct_usage __percpu *cpuusage; | |
44 | struct kernel_cpustat __percpu *cpustat; | |
45 | }; | |
46 | ||
47 | static inline struct cpuacct *css_ca(struct cgroup_subsys_state *css) | |
48 | { | |
49 | return css ? container_of(css, struct cpuacct, css) : NULL; | |
50 | } | |
51 | ||
52 | /* return cpu accounting group to which this task belongs */ | |
53 | static inline struct cpuacct *task_ca(struct task_struct *tsk) | |
54 | { | |
55 | return css_ca(task_css(tsk, cpuacct_cgrp_id)); | |
56 | } | |
57 | ||
58 | static inline struct cpuacct *parent_ca(struct cpuacct *ca) | |
59 | { | |
60 | return css_ca(ca->css.parent); | |
61 | } | |
62 | ||
63 | static DEFINE_PER_CPU(struct cpuacct_usage, root_cpuacct_cpuusage); | |
64 | static struct cpuacct root_cpuacct = { | |
65 | .cpustat = &kernel_cpustat, | |
66 | .cpuusage = &root_cpuacct_cpuusage, | |
67 | }; | |
68 | ||
69 | /* create a new cpu accounting group */ | |
70 | static struct cgroup_subsys_state * | |
71 | cpuacct_css_alloc(struct cgroup_subsys_state *parent_css) | |
72 | { | |
73 | struct cpuacct *ca; | |
74 | ||
75 | if (!parent_css) | |
76 | return &root_cpuacct.css; | |
77 | ||
78 | ca = kzalloc(sizeof(*ca), GFP_KERNEL); | |
79 | if (!ca) | |
80 | goto out; | |
81 | ||
82 | ca->cpuusage = alloc_percpu(struct cpuacct_usage); | |
83 | if (!ca->cpuusage) | |
84 | goto out_free_ca; | |
85 | ||
86 | ca->cpustat = alloc_percpu(struct kernel_cpustat); | |
87 | if (!ca->cpustat) | |
88 | goto out_free_cpuusage; | |
89 | ||
90 | return &ca->css; | |
91 | ||
92 | out_free_cpuusage: | |
93 | free_percpu(ca->cpuusage); | |
94 | out_free_ca: | |
95 | kfree(ca); | |
96 | out: | |
97 | return ERR_PTR(-ENOMEM); | |
98 | } | |
99 | ||
100 | /* destroy an existing cpu accounting group */ | |
101 | static void cpuacct_css_free(struct cgroup_subsys_state *css) | |
102 | { | |
103 | struct cpuacct *ca = css_ca(css); | |
104 | ||
105 | free_percpu(ca->cpustat); | |
106 | free_percpu(ca->cpuusage); | |
107 | kfree(ca); | |
108 | } | |
109 | ||
110 | static u64 cpuacct_cpuusage_read(struct cpuacct *ca, int cpu, | |
111 | enum cpuacct_usage_index index) | |
112 | { | |
113 | struct cpuacct_usage *cpuusage = per_cpu_ptr(ca->cpuusage, cpu); | |
114 | u64 data; | |
115 | ||
116 | /* | |
117 | * We allow index == CPUACCT_USAGE_NRUSAGE here to read | |
118 | * the sum of suages. | |
119 | */ | |
120 | BUG_ON(index > CPUACCT_USAGE_NRUSAGE); | |
121 | ||
122 | #ifndef CONFIG_64BIT | |
123 | /* | |
124 | * Take rq->lock to make 64-bit read safe on 32-bit platforms. | |
125 | */ | |
126 | raw_spin_lock_irq(&cpu_rq(cpu)->lock); | |
127 | #endif | |
128 | ||
129 | if (index == CPUACCT_USAGE_NRUSAGE) { | |
130 | int i = 0; | |
131 | ||
132 | data = 0; | |
133 | for (i = 0; i < CPUACCT_USAGE_NRUSAGE; i++) | |
134 | data += cpuusage->usages[i]; | |
135 | } else { | |
136 | data = cpuusage->usages[index]; | |
137 | } | |
138 | ||
139 | #ifndef CONFIG_64BIT | |
140 | raw_spin_unlock_irq(&cpu_rq(cpu)->lock); | |
141 | #endif | |
142 | ||
143 | return data; | |
144 | } | |
145 | ||
146 | static void cpuacct_cpuusage_write(struct cpuacct *ca, int cpu, u64 val) | |
147 | { | |
148 | struct cpuacct_usage *cpuusage = per_cpu_ptr(ca->cpuusage, cpu); | |
149 | int i; | |
150 | ||
151 | #ifndef CONFIG_64BIT | |
152 | /* | |
153 | * Take rq->lock to make 64-bit write safe on 32-bit platforms. | |
154 | */ | |
155 | raw_spin_lock_irq(&cpu_rq(cpu)->lock); | |
156 | #endif | |
157 | ||
158 | for (i = 0; i < CPUACCT_USAGE_NRUSAGE; i++) | |
159 | cpuusage->usages[i] = val; | |
160 | ||
161 | #ifndef CONFIG_64BIT | |
162 | raw_spin_unlock_irq(&cpu_rq(cpu)->lock); | |
163 | #endif | |
164 | } | |
165 | ||
166 | /* return total cpu usage (in nanoseconds) of a group */ | |
167 | static u64 __cpuusage_read(struct cgroup_subsys_state *css, | |
168 | enum cpuacct_usage_index index) | |
169 | { | |
170 | struct cpuacct *ca = css_ca(css); | |
171 | u64 totalcpuusage = 0; | |
172 | int i; | |
173 | ||
174 | for_each_possible_cpu(i) | |
175 | totalcpuusage += cpuacct_cpuusage_read(ca, i, index); | |
176 | ||
177 | return totalcpuusage; | |
178 | } | |
179 | ||
180 | static u64 cpuusage_user_read(struct cgroup_subsys_state *css, | |
181 | struct cftype *cft) | |
182 | { | |
183 | return __cpuusage_read(css, CPUACCT_USAGE_USER); | |
184 | } | |
185 | ||
186 | static u64 cpuusage_sys_read(struct cgroup_subsys_state *css, | |
187 | struct cftype *cft) | |
188 | { | |
189 | return __cpuusage_read(css, CPUACCT_USAGE_SYSTEM); | |
190 | } | |
191 | ||
192 | static u64 cpuusage_read(struct cgroup_subsys_state *css, struct cftype *cft) | |
193 | { | |
194 | return __cpuusage_read(css, CPUACCT_USAGE_NRUSAGE); | |
195 | } | |
196 | ||
197 | static int cpuusage_write(struct cgroup_subsys_state *css, struct cftype *cft, | |
198 | u64 val) | |
199 | { | |
200 | struct cpuacct *ca = css_ca(css); | |
201 | int cpu; | |
202 | ||
203 | /* | |
204 | * Only allow '0' here to do a reset. | |
205 | */ | |
206 | if (val) | |
207 | return -EINVAL; | |
208 | ||
209 | for_each_possible_cpu(cpu) | |
210 | cpuacct_cpuusage_write(ca, cpu, 0); | |
211 | ||
212 | return 0; | |
213 | } | |
214 | ||
215 | static int __cpuacct_percpu_seq_show(struct seq_file *m, | |
216 | enum cpuacct_usage_index index) | |
217 | { | |
218 | struct cpuacct *ca = css_ca(seq_css(m)); | |
219 | u64 percpu; | |
220 | int i; | |
221 | ||
222 | for_each_possible_cpu(i) { | |
223 | percpu = cpuacct_cpuusage_read(ca, i, index); | |
224 | seq_printf(m, "%llu ", (unsigned long long) percpu); | |
225 | } | |
226 | seq_printf(m, "\n"); | |
227 | return 0; | |
228 | } | |
229 | ||
230 | static int cpuacct_percpu_user_seq_show(struct seq_file *m, void *V) | |
231 | { | |
232 | return __cpuacct_percpu_seq_show(m, CPUACCT_USAGE_USER); | |
233 | } | |
234 | ||
235 | static int cpuacct_percpu_sys_seq_show(struct seq_file *m, void *V) | |
236 | { | |
237 | return __cpuacct_percpu_seq_show(m, CPUACCT_USAGE_SYSTEM); | |
238 | } | |
239 | ||
240 | static int cpuacct_percpu_seq_show(struct seq_file *m, void *V) | |
241 | { | |
242 | return __cpuacct_percpu_seq_show(m, CPUACCT_USAGE_NRUSAGE); | |
243 | } | |
244 | ||
245 | static const char * const cpuacct_stat_desc[] = { | |
246 | [CPUACCT_STAT_USER] = "user", | |
247 | [CPUACCT_STAT_SYSTEM] = "system", | |
248 | }; | |
249 | ||
250 | static int cpuacct_stats_show(struct seq_file *sf, void *v) | |
251 | { | |
252 | struct cpuacct *ca = css_ca(seq_css(sf)); | |
253 | int cpu; | |
254 | s64 val = 0; | |
255 | ||
256 | for_each_possible_cpu(cpu) { | |
257 | struct kernel_cpustat *kcpustat = per_cpu_ptr(ca->cpustat, cpu); | |
258 | val += kcpustat->cpustat[CPUTIME_USER]; | |
259 | val += kcpustat->cpustat[CPUTIME_NICE]; | |
260 | } | |
261 | val = cputime64_to_clock_t(val); | |
262 | seq_printf(sf, "%s %lld\n", cpuacct_stat_desc[CPUACCT_STAT_USER], val); | |
263 | ||
264 | val = 0; | |
265 | for_each_possible_cpu(cpu) { | |
266 | struct kernel_cpustat *kcpustat = per_cpu_ptr(ca->cpustat, cpu); | |
267 | val += kcpustat->cpustat[CPUTIME_SYSTEM]; | |
268 | val += kcpustat->cpustat[CPUTIME_IRQ]; | |
269 | val += kcpustat->cpustat[CPUTIME_SOFTIRQ]; | |
270 | } | |
271 | ||
272 | val = cputime64_to_clock_t(val); | |
273 | seq_printf(sf, "%s %lld\n", cpuacct_stat_desc[CPUACCT_STAT_SYSTEM], val); | |
274 | ||
275 | return 0; | |
276 | } | |
277 | ||
278 | static struct cftype files[] = { | |
279 | { | |
280 | .name = "usage", | |
281 | .read_u64 = cpuusage_read, | |
282 | .write_u64 = cpuusage_write, | |
283 | }, | |
284 | { | |
285 | .name = "usage_user", | |
286 | .read_u64 = cpuusage_user_read, | |
287 | }, | |
288 | { | |
289 | .name = "usage_sys", | |
290 | .read_u64 = cpuusage_sys_read, | |
291 | }, | |
292 | { | |
293 | .name = "usage_percpu", | |
294 | .seq_show = cpuacct_percpu_seq_show, | |
295 | }, | |
296 | { | |
297 | .name = "usage_percpu_user", | |
298 | .seq_show = cpuacct_percpu_user_seq_show, | |
299 | }, | |
300 | { | |
301 | .name = "usage_percpu_sys", | |
302 | .seq_show = cpuacct_percpu_sys_seq_show, | |
303 | }, | |
304 | { | |
305 | .name = "stat", | |
306 | .seq_show = cpuacct_stats_show, | |
307 | }, | |
308 | { } /* terminate */ | |
309 | }; | |
310 | ||
311 | /* | |
312 | * charge this task's execution time to its accounting group. | |
313 | * | |
314 | * called with rq->lock held. | |
315 | */ | |
316 | void cpuacct_charge(struct task_struct *tsk, u64 cputime) | |
317 | { | |
318 | struct cpuacct *ca; | |
319 | int index; | |
320 | ||
321 | if (user_mode(task_pt_regs(tsk))) | |
322 | index = CPUACCT_USAGE_USER; | |
323 | else | |
324 | index = CPUACCT_USAGE_SYSTEM; | |
325 | ||
326 | rcu_read_lock(); | |
327 | ||
328 | for (ca = task_ca(tsk); ca; ca = parent_ca(ca)) | |
329 | this_cpu_ptr(ca->cpuusage)->usages[index] += cputime; | |
330 | ||
331 | rcu_read_unlock(); | |
332 | } | |
333 | ||
334 | /* | |
335 | * Add user/system time to cpuacct. | |
336 | * | |
337 | * Note: it's the caller that updates the account of the root cgroup. | |
338 | */ | |
339 | void cpuacct_account_field(struct task_struct *tsk, int index, u64 val) | |
340 | { | |
341 | struct cpuacct *ca; | |
342 | ||
343 | rcu_read_lock(); | |
344 | for (ca = task_ca(tsk); ca != &root_cpuacct; ca = parent_ca(ca)) | |
345 | this_cpu_ptr(ca->cpustat)->cpustat[index] += val; | |
346 | rcu_read_unlock(); | |
347 | } | |
348 | ||
349 | struct cgroup_subsys cpuacct_cgrp_subsys = { | |
350 | .css_alloc = cpuacct_css_alloc, | |
351 | .css_free = cpuacct_css_free, | |
352 | .legacy_cftypes = files, | |
353 | .early_init = true, | |
354 | }; |