1 // SPDX-License-Identifier: GPL-2.0 1 // SPDX-License-Identifier: GPL-2.0
2 <<
3 /* 2 /*
4 * CPU accounting code for task groups. 3 * CPU accounting code for task groups.
5 * 4 *
6 * Based on the work by Paul Menage (menage@go 5 * Based on the work by Paul Menage (menage@google.com) and Balbir Singh
7 * (balbir@in.ibm.com). 6 * (balbir@in.ibm.com).
8 */ 7 */
>> 8 #include <asm/irq_regs.h>
>> 9 #include "sched.h"
9 10
10 /* Time spent by the tasks of the CPU accounti 11 /* Time spent by the tasks of the CPU accounting group executing in ... */
11 enum cpuacct_stat_index { 12 enum cpuacct_stat_index {
12 CPUACCT_STAT_USER, /* ... user mo 13 CPUACCT_STAT_USER, /* ... user mode */
13 CPUACCT_STAT_SYSTEM, /* ... kernel 14 CPUACCT_STAT_SYSTEM, /* ... kernel mode */
14 15
15 CPUACCT_STAT_NSTATS, 16 CPUACCT_STAT_NSTATS,
16 }; 17 };
17 18
18 static const char * const cpuacct_stat_desc[] 19 static const char * const cpuacct_stat_desc[] = {
19 [CPUACCT_STAT_USER] = "user", 20 [CPUACCT_STAT_USER] = "user",
20 [CPUACCT_STAT_SYSTEM] = "system", 21 [CPUACCT_STAT_SYSTEM] = "system",
21 }; 22 };
22 23
>> 24 struct cpuacct_usage {
>> 25 u64 usages[CPUACCT_STAT_NSTATS];
>> 26 };
>> 27
23 /* track CPU usage of a group of tasks and its 28 /* track CPU usage of a group of tasks and its child groups */
24 struct cpuacct { 29 struct cpuacct {
25 struct cgroup_subsys_state css; 30 struct cgroup_subsys_state css;
26 /* cpuusage holds pointer to a u64-typ 31 /* cpuusage holds pointer to a u64-type object on every CPU */
27 u64 __percpu *cpuusage; !! 32 struct cpuacct_usage __percpu *cpuusage;
28 struct kernel_cpustat __percpu *cpust 33 struct kernel_cpustat __percpu *cpustat;
29 }; 34 };
30 35
31 static inline struct cpuacct *css_ca(struct cg 36 static inline struct cpuacct *css_ca(struct cgroup_subsys_state *css)
32 { 37 {
33 return css ? container_of(css, struct 38 return css ? container_of(css, struct cpuacct, css) : NULL;
34 } 39 }
35 40
36 /* Return CPU accounting group to which this t 41 /* Return CPU accounting group to which this task belongs */
37 static inline struct cpuacct *task_ca(struct t 42 static inline struct cpuacct *task_ca(struct task_struct *tsk)
38 { 43 {
39 return css_ca(task_css(tsk, cpuacct_cg 44 return css_ca(task_css(tsk, cpuacct_cgrp_id));
40 } 45 }
41 46
42 static inline struct cpuacct *parent_ca(struct 47 static inline struct cpuacct *parent_ca(struct cpuacct *ca)
43 { 48 {
44 return css_ca(ca->css.parent); 49 return css_ca(ca->css.parent);
45 } 50 }
46 51
47 static DEFINE_PER_CPU(u64, root_cpuacct_cpuusa !! 52 static DEFINE_PER_CPU(struct cpuacct_usage, root_cpuacct_cpuusage);
48 static struct cpuacct root_cpuacct = { 53 static struct cpuacct root_cpuacct = {
49 .cpustat = &kernel_cpustat, 54 .cpustat = &kernel_cpustat,
50 .cpuusage = &root_cpuacct_cpuusa 55 .cpuusage = &root_cpuacct_cpuusage,
51 }; 56 };
52 57
53 /* Create a new CPU accounting group */ 58 /* Create a new CPU accounting group */
54 static struct cgroup_subsys_state * 59 static struct cgroup_subsys_state *
55 cpuacct_css_alloc(struct cgroup_subsys_state * 60 cpuacct_css_alloc(struct cgroup_subsys_state *parent_css)
56 { 61 {
57 struct cpuacct *ca; 62 struct cpuacct *ca;
58 63
59 if (!parent_css) 64 if (!parent_css)
60 return &root_cpuacct.css; 65 return &root_cpuacct.css;
61 66
62 ca = kzalloc(sizeof(*ca), GFP_KERNEL); 67 ca = kzalloc(sizeof(*ca), GFP_KERNEL);
63 if (!ca) 68 if (!ca)
64 goto out; 69 goto out;
65 70
66 ca->cpuusage = alloc_percpu(u64); !! 71 ca->cpuusage = alloc_percpu(struct cpuacct_usage);
67 if (!ca->cpuusage) 72 if (!ca->cpuusage)
68 goto out_free_ca; 73 goto out_free_ca;
69 74
70 ca->cpustat = alloc_percpu(struct kern 75 ca->cpustat = alloc_percpu(struct kernel_cpustat);
71 if (!ca->cpustat) 76 if (!ca->cpustat)
72 goto out_free_cpuusage; 77 goto out_free_cpuusage;
73 78
74 return &ca->css; 79 return &ca->css;
75 80
76 out_free_cpuusage: 81 out_free_cpuusage:
77 free_percpu(ca->cpuusage); 82 free_percpu(ca->cpuusage);
78 out_free_ca: 83 out_free_ca:
79 kfree(ca); 84 kfree(ca);
80 out: 85 out:
81 return ERR_PTR(-ENOMEM); 86 return ERR_PTR(-ENOMEM);
82 } 87 }
83 88
84 /* Destroy an existing CPU accounting group */ 89 /* Destroy an existing CPU accounting group */
85 static void cpuacct_css_free(struct cgroup_sub 90 static void cpuacct_css_free(struct cgroup_subsys_state *css)
86 { 91 {
87 struct cpuacct *ca = css_ca(css); 92 struct cpuacct *ca = css_ca(css);
88 93
89 free_percpu(ca->cpustat); 94 free_percpu(ca->cpustat);
90 free_percpu(ca->cpuusage); 95 free_percpu(ca->cpuusage);
91 kfree(ca); 96 kfree(ca);
92 } 97 }
93 98
94 static u64 cpuacct_cpuusage_read(struct cpuacc 99 static u64 cpuacct_cpuusage_read(struct cpuacct *ca, int cpu,
95 enum cpuacct_ 100 enum cpuacct_stat_index index)
96 { 101 {
97 u64 *cpuusage = per_cpu_ptr(ca->cpuusa !! 102 struct cpuacct_usage *cpuusage = per_cpu_ptr(ca->cpuusage, cpu);
98 u64 *cpustat = per_cpu_ptr(ca->cpustat <<
99 u64 data; 103 u64 data;
100 104
101 /* 105 /*
102 * We allow index == CPUACCT_STAT_NSTA 106 * We allow index == CPUACCT_STAT_NSTATS here to read
103 * the sum of usages. 107 * the sum of usages.
104 */ 108 */
105 if (WARN_ON_ONCE(index > CPUACCT_STAT_ !! 109 BUG_ON(index > CPUACCT_STAT_NSTATS);
106 return 0; <<
107 110
108 #ifndef CONFIG_64BIT 111 #ifndef CONFIG_64BIT
109 /* 112 /*
110 * Take rq->lock to make 64-bit read s 113 * Take rq->lock to make 64-bit read safe on 32-bit platforms.
111 */ 114 */
112 raw_spin_rq_lock_irq(cpu_rq(cpu)); !! 115 raw_spin_lock_irq(&cpu_rq(cpu)->lock);
113 #endif 116 #endif
114 117
115 switch (index) { !! 118 if (index == CPUACCT_STAT_NSTATS) {
116 case CPUACCT_STAT_USER: !! 119 int i = 0;
117 data = cpustat[CPUTIME_USER] + !! 120
118 break; !! 121 data = 0;
119 case CPUACCT_STAT_SYSTEM: !! 122 for (i = 0; i < CPUACCT_STAT_NSTATS; i++)
120 data = cpustat[CPUTIME_SYSTEM] !! 123 data += cpuusage->usages[i];
121 cpustat[CPUTIME_SOFTIR !! 124 } else {
122 break; !! 125 data = cpuusage->usages[index];
123 case CPUACCT_STAT_NSTATS: <<
124 data = *cpuusage; <<
125 break; <<
126 } 126 }
127 127
128 #ifndef CONFIG_64BIT 128 #ifndef CONFIG_64BIT
129 raw_spin_rq_unlock_irq(cpu_rq(cpu)); !! 129 raw_spin_unlock_irq(&cpu_rq(cpu)->lock);
130 #endif 130 #endif
131 131
132 return data; 132 return data;
133 } 133 }
134 134
135 static void cpuacct_cpuusage_write(struct cpua !! 135 static void cpuacct_cpuusage_write(struct cpuacct *ca, int cpu, u64 val)
136 { 136 {
137 u64 *cpuusage = per_cpu_ptr(ca->cpuusa !! 137 struct cpuacct_usage *cpuusage = per_cpu_ptr(ca->cpuusage, cpu);
138 u64 *cpustat = per_cpu_ptr(ca->cpustat !! 138 int i;
139 <<
140 /* Don't allow to reset global kernel_ <<
141 if (ca == &root_cpuacct) <<
142 return; <<
143 139
144 #ifndef CONFIG_64BIT 140 #ifndef CONFIG_64BIT
145 /* 141 /*
146 * Take rq->lock to make 64-bit write 142 * Take rq->lock to make 64-bit write safe on 32-bit platforms.
147 */ 143 */
148 raw_spin_rq_lock_irq(cpu_rq(cpu)); !! 144 raw_spin_lock_irq(&cpu_rq(cpu)->lock);
149 #endif 145 #endif
150 *cpuusage = 0; !! 146
151 cpustat[CPUTIME_USER] = cpustat[CPUTIM !! 147 for (i = 0; i < CPUACCT_STAT_NSTATS; i++)
152 cpustat[CPUTIME_SYSTEM] = cpustat[CPUT !! 148 cpuusage->usages[i] = val;
153 cpustat[CPUTIME_SOFTIRQ] = 0; <<
154 149
155 #ifndef CONFIG_64BIT 150 #ifndef CONFIG_64BIT
156 raw_spin_rq_unlock_irq(cpu_rq(cpu)); !! 151 raw_spin_unlock_irq(&cpu_rq(cpu)->lock);
157 #endif 152 #endif
158 } 153 }
159 154
160 /* Return total CPU usage (in nanoseconds) of 155 /* Return total CPU usage (in nanoseconds) of a group */
161 static u64 __cpuusage_read(struct cgroup_subsy 156 static u64 __cpuusage_read(struct cgroup_subsys_state *css,
162 enum cpuacct_stat_i 157 enum cpuacct_stat_index index)
163 { 158 {
164 struct cpuacct *ca = css_ca(css); 159 struct cpuacct *ca = css_ca(css);
165 u64 totalcpuusage = 0; 160 u64 totalcpuusage = 0;
166 int i; 161 int i;
167 162
168 for_each_possible_cpu(i) 163 for_each_possible_cpu(i)
169 totalcpuusage += cpuacct_cpuus 164 totalcpuusage += cpuacct_cpuusage_read(ca, i, index);
170 165
171 return totalcpuusage; 166 return totalcpuusage;
172 } 167 }
173 168
174 static u64 cpuusage_user_read(struct cgroup_su 169 static u64 cpuusage_user_read(struct cgroup_subsys_state *css,
175 struct cftype *c 170 struct cftype *cft)
176 { 171 {
177 return __cpuusage_read(css, CPUACCT_ST 172 return __cpuusage_read(css, CPUACCT_STAT_USER);
178 } 173 }
179 174
180 static u64 cpuusage_sys_read(struct cgroup_sub 175 static u64 cpuusage_sys_read(struct cgroup_subsys_state *css,
181 struct cftype *cf 176 struct cftype *cft)
182 { 177 {
183 return __cpuusage_read(css, CPUACCT_ST 178 return __cpuusage_read(css, CPUACCT_STAT_SYSTEM);
184 } 179 }
185 180
186 static u64 cpuusage_read(struct cgroup_subsys_ 181 static u64 cpuusage_read(struct cgroup_subsys_state *css, struct cftype *cft)
187 { 182 {
188 return __cpuusage_read(css, CPUACCT_ST 183 return __cpuusage_read(css, CPUACCT_STAT_NSTATS);
189 } 184 }
190 185
191 static int cpuusage_write(struct cgroup_subsys 186 static int cpuusage_write(struct cgroup_subsys_state *css, struct cftype *cft,
192 u64 val) 187 u64 val)
193 { 188 {
194 struct cpuacct *ca = css_ca(css); 189 struct cpuacct *ca = css_ca(css);
195 int cpu; 190 int cpu;
196 191
197 /* 192 /*
198 * Only allow '' here to do a reset. 193 * Only allow '' here to do a reset.
199 */ 194 */
200 if (val) 195 if (val)
201 return -EINVAL; 196 return -EINVAL;
202 197
203 for_each_possible_cpu(cpu) 198 for_each_possible_cpu(cpu)
204 cpuacct_cpuusage_write(ca, cpu !! 199 cpuacct_cpuusage_write(ca, cpu, 0);
205 200
206 return 0; 201 return 0;
207 } 202 }
208 203
209 static int __cpuacct_percpu_seq_show(struct se 204 static int __cpuacct_percpu_seq_show(struct seq_file *m,
210 enum cpua 205 enum cpuacct_stat_index index)
211 { 206 {
212 struct cpuacct *ca = css_ca(seq_css(m) 207 struct cpuacct *ca = css_ca(seq_css(m));
213 u64 percpu; 208 u64 percpu;
214 int i; 209 int i;
215 210
216 for_each_possible_cpu(i) { 211 for_each_possible_cpu(i) {
217 percpu = cpuacct_cpuusage_read 212 percpu = cpuacct_cpuusage_read(ca, i, index);
218 seq_printf(m, "%llu ", (unsign 213 seq_printf(m, "%llu ", (unsigned long long) percpu);
219 } 214 }
220 seq_printf(m, "\n"); 215 seq_printf(m, "\n");
221 return 0; 216 return 0;
222 } 217 }
223 218
224 static int cpuacct_percpu_user_seq_show(struct 219 static int cpuacct_percpu_user_seq_show(struct seq_file *m, void *V)
225 { 220 {
226 return __cpuacct_percpu_seq_show(m, CP 221 return __cpuacct_percpu_seq_show(m, CPUACCT_STAT_USER);
227 } 222 }
228 223
229 static int cpuacct_percpu_sys_seq_show(struct 224 static int cpuacct_percpu_sys_seq_show(struct seq_file *m, void *V)
230 { 225 {
231 return __cpuacct_percpu_seq_show(m, CP 226 return __cpuacct_percpu_seq_show(m, CPUACCT_STAT_SYSTEM);
232 } 227 }
233 228
234 static int cpuacct_percpu_seq_show(struct seq_ 229 static int cpuacct_percpu_seq_show(struct seq_file *m, void *V)
235 { 230 {
236 return __cpuacct_percpu_seq_show(m, CP 231 return __cpuacct_percpu_seq_show(m, CPUACCT_STAT_NSTATS);
237 } 232 }
238 233
239 static int cpuacct_all_seq_show(struct seq_fil 234 static int cpuacct_all_seq_show(struct seq_file *m, void *V)
240 { 235 {
241 struct cpuacct *ca = css_ca(seq_css(m) 236 struct cpuacct *ca = css_ca(seq_css(m));
242 int index; 237 int index;
243 int cpu; 238 int cpu;
244 239
245 seq_puts(m, "cpu"); 240 seq_puts(m, "cpu");
246 for (index = 0; index < CPUACCT_STAT_N 241 for (index = 0; index < CPUACCT_STAT_NSTATS; index++)
247 seq_printf(m, " %s", cpuacct_s 242 seq_printf(m, " %s", cpuacct_stat_desc[index]);
248 seq_puts(m, "\n"); 243 seq_puts(m, "\n");
249 244
250 for_each_possible_cpu(cpu) { 245 for_each_possible_cpu(cpu) {
>> 246 struct cpuacct_usage *cpuusage = per_cpu_ptr(ca->cpuusage, cpu);
>> 247
251 seq_printf(m, "%d", cpu); 248 seq_printf(m, "%d", cpu);
252 for (index = 0; index < CPUACC !! 249
253 seq_printf(m, " %llu", !! 250 for (index = 0; index < CPUACCT_STAT_NSTATS; index++) {
254 cpuacct_cpu !! 251 #ifndef CONFIG_64BIT
>> 252 /*
>> 253 * Take rq->lock to make 64-bit read safe on 32-bit
>> 254 * platforms.
>> 255 */
>> 256 raw_spin_lock_irq(&cpu_rq(cpu)->lock);
>> 257 #endif
>> 258
>> 259 seq_printf(m, " %llu", cpuusage->usages[index]);
>> 260
>> 261 #ifndef CONFIG_64BIT
>> 262 raw_spin_unlock_irq(&cpu_rq(cpu)->lock);
>> 263 #endif
>> 264 }
255 seq_puts(m, "\n"); 265 seq_puts(m, "\n");
256 } 266 }
257 return 0; 267 return 0;
258 } 268 }
259 269
260 static int cpuacct_stats_show(struct seq_file 270 static int cpuacct_stats_show(struct seq_file *sf, void *v)
261 { 271 {
262 struct cpuacct *ca = css_ca(seq_css(sf 272 struct cpuacct *ca = css_ca(seq_css(sf));
263 struct task_cputime cputime; !! 273 s64 val[CPUACCT_STAT_NSTATS];
264 u64 val[CPUACCT_STAT_NSTATS]; <<
265 int cpu; 274 int cpu;
266 int stat; 275 int stat;
267 276
268 memset(&cputime, 0, sizeof(cputime)); !! 277 memset(val, 0, sizeof(val));
269 for_each_possible_cpu(cpu) { 278 for_each_possible_cpu(cpu) {
270 u64 *cpustat = per_cpu_ptr(ca- 279 u64 *cpustat = per_cpu_ptr(ca->cpustat, cpu)->cpustat;
271 280
272 cputime.utime += cpustat[CPUTI !! 281 val[CPUACCT_STAT_USER] += cpustat[CPUTIME_USER];
273 cputime.utime += cpustat[CPUTI !! 282 val[CPUACCT_STAT_USER] += cpustat[CPUTIME_NICE];
274 cputime.stime += cpustat[CPUTI !! 283 val[CPUACCT_STAT_SYSTEM] += cpustat[CPUTIME_SYSTEM];
275 cputime.stime += cpustat[CPUTI !! 284 val[CPUACCT_STAT_SYSTEM] += cpustat[CPUTIME_IRQ];
276 cputime.stime += cpustat[CPUTI !! 285 val[CPUACCT_STAT_SYSTEM] += cpustat[CPUTIME_SOFTIRQ];
277 <<
278 cputime.sum_exec_runtime += *p <<
279 } 286 }
280 287
281 cputime_adjust(&cputime, &seq_css(sf)- <<
282 &val[CPUACCT_STAT_USER], &val[ <<
283 <<
284 for (stat = 0; stat < CPUACCT_STAT_NST 288 for (stat = 0; stat < CPUACCT_STAT_NSTATS; stat++) {
285 seq_printf(sf, "%s %llu\n", cp !! 289 seq_printf(sf, "%s %lld\n",
286 nsec_to_clock_t(val[st !! 290 cpuacct_stat_desc[stat],
>> 291 (long long)nsec_to_clock_t(val[stat]));
287 } 292 }
288 293
289 return 0; 294 return 0;
290 } 295 }
291 296
292 static struct cftype files[] = { 297 static struct cftype files[] = {
293 { 298 {
294 .name = "usage", 299 .name = "usage",
295 .read_u64 = cpuusage_read, 300 .read_u64 = cpuusage_read,
296 .write_u64 = cpuusage_write, 301 .write_u64 = cpuusage_write,
297 }, 302 },
298 { 303 {
299 .name = "usage_user", 304 .name = "usage_user",
300 .read_u64 = cpuusage_user_read 305 .read_u64 = cpuusage_user_read,
301 }, 306 },
302 { 307 {
303 .name = "usage_sys", 308 .name = "usage_sys",
304 .read_u64 = cpuusage_sys_read, 309 .read_u64 = cpuusage_sys_read,
305 }, 310 },
306 { 311 {
307 .name = "usage_percpu", 312 .name = "usage_percpu",
308 .seq_show = cpuacct_percpu_seq 313 .seq_show = cpuacct_percpu_seq_show,
309 }, 314 },
310 { 315 {
311 .name = "usage_percpu_user", 316 .name = "usage_percpu_user",
312 .seq_show = cpuacct_percpu_use 317 .seq_show = cpuacct_percpu_user_seq_show,
313 }, 318 },
314 { 319 {
315 .name = "usage_percpu_sys", 320 .name = "usage_percpu_sys",
316 .seq_show = cpuacct_percpu_sys 321 .seq_show = cpuacct_percpu_sys_seq_show,
317 }, 322 },
318 { 323 {
319 .name = "usage_all", 324 .name = "usage_all",
320 .seq_show = cpuacct_all_seq_sh 325 .seq_show = cpuacct_all_seq_show,
321 }, 326 },
322 { 327 {
323 .name = "stat", 328 .name = "stat",
324 .seq_show = cpuacct_stats_show 329 .seq_show = cpuacct_stats_show,
325 }, 330 },
326 { } /* terminate */ 331 { } /* terminate */
327 }; 332 };
328 333
329 /* 334 /*
330 * charge this task's execution time to its ac 335 * charge this task's execution time to its accounting group.
331 * 336 *
332 * called with rq->lock held. 337 * called with rq->lock held.
333 */ 338 */
334 void cpuacct_charge(struct task_struct *tsk, u 339 void cpuacct_charge(struct task_struct *tsk, u64 cputime)
335 { 340 {
336 unsigned int cpu = task_cpu(tsk); <<
337 struct cpuacct *ca; 341 struct cpuacct *ca;
>> 342 int index = CPUACCT_STAT_SYSTEM;
>> 343 struct pt_regs *regs = get_irq_regs() ? : task_pt_regs(tsk);
338 344
339 lockdep_assert_rq_held(cpu_rq(cpu)); !! 345 if (regs && user_mode(regs))
>> 346 index = CPUACCT_STAT_USER;
>> 347
>> 348 rcu_read_lock();
340 349
341 for (ca = task_ca(tsk); ca; ca = paren 350 for (ca = task_ca(tsk); ca; ca = parent_ca(ca))
342 *per_cpu_ptr(ca->cpuusage, cpu !! 351 __this_cpu_add(ca->cpuusage->usages[index], cputime);
>> 352
>> 353 rcu_read_unlock();
343 } 354 }
344 355
345 /* 356 /*
346 * Add user/system time to cpuacct. 357 * Add user/system time to cpuacct.
347 * 358 *
348 * Note: it's the caller that updates the acco 359 * Note: it's the caller that updates the account of the root cgroup.
349 */ 360 */
350 void cpuacct_account_field(struct task_struct 361 void cpuacct_account_field(struct task_struct *tsk, int index, u64 val)
351 { 362 {
352 struct cpuacct *ca; 363 struct cpuacct *ca;
353 364
>> 365 rcu_read_lock();
354 for (ca = task_ca(tsk); ca != &root_cp 366 for (ca = task_ca(tsk); ca != &root_cpuacct; ca = parent_ca(ca))
355 __this_cpu_add(ca->cpustat->cp 367 __this_cpu_add(ca->cpustat->cpustat[index], val);
>> 368 rcu_read_unlock();
356 } 369 }
357 370
358 struct cgroup_subsys cpuacct_cgrp_subsys = { 371 struct cgroup_subsys cpuacct_cgrp_subsys = {
359 .css_alloc = cpuacct_css_alloc, 372 .css_alloc = cpuacct_css_alloc,
360 .css_free = cpuacct_css_free, 373 .css_free = cpuacct_css_free,
361 .legacy_cftypes = files, 374 .legacy_cftypes = files,
362 .early_init = true, 375 .early_init = true,
363 }; 376 };
364 377
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.