Commit | Line | Data |
---|---|---|
73fbec60 FW |
1 | #include <linux/export.h> |
2 | #include <linux/sched.h> | |
3 | #include <linux/tsacct_kern.h> | |
4 | #include <linux/kernel_stat.h> | |
5 | #include <linux/static_key.h> | |
abf917cd | 6 | #include <linux/context_tracking.h> |
73fbec60 FW |
7 | #include "sched.h" |
8 | ||
9 | ||
10 | #ifdef CONFIG_IRQ_TIME_ACCOUNTING | |
11 | ||
12 | /* | |
13 | * There are no locks covering percpu hardirq/softirq time. | |
bf9fae9f | 14 | * They are only modified in vtime_account, on corresponding CPU |
73fbec60 FW |
15 | * with interrupts disabled. So, writes are safe. |
16 | * They are read and saved off onto struct rq in update_rq_clock(). | |
17 | * This may result in other CPU reading this CPU's irq time and can | |
bf9fae9f | 18 | * race with irq/vtime_account on this CPU. We would either get old |
73fbec60 FW |
19 | * or new value with a side effect of accounting a slice of irq time to wrong |
20 | * task when irq is in progress while we read rq->clock. That is a worthy | |
21 | * compromise in place of having locks on each irq in account_system_time. | |
22 | */ | |
23 | DEFINE_PER_CPU(u64, cpu_hardirq_time); | |
24 | DEFINE_PER_CPU(u64, cpu_softirq_time); | |
25 | ||
26 | static DEFINE_PER_CPU(u64, irq_start_time); | |
27 | static int sched_clock_irqtime; | |
28 | ||
29 | void enable_sched_clock_irqtime(void) | |
30 | { | |
31 | sched_clock_irqtime = 1; | |
32 | } | |
33 | ||
34 | void disable_sched_clock_irqtime(void) | |
35 | { | |
36 | sched_clock_irqtime = 0; | |
37 | } | |
38 | ||
39 | #ifndef CONFIG_64BIT | |
40 | DEFINE_PER_CPU(seqcount_t, irq_time_seq); | |
41 | #endif /* CONFIG_64BIT */ | |
42 | ||
43 | /* | |
44 | * Called before incrementing preempt_count on {soft,}irq_enter | |
45 | * and before decrementing preempt_count on {soft,}irq_exit. | |
46 | */ | |
3e1df4f5 | 47 | void irqtime_account_irq(struct task_struct *curr) |
73fbec60 FW |
48 | { |
49 | unsigned long flags; | |
50 | s64 delta; | |
51 | int cpu; | |
52 | ||
53 | if (!sched_clock_irqtime) | |
54 | return; | |
55 | ||
56 | local_irq_save(flags); | |
57 | ||
58 | cpu = smp_processor_id(); | |
59 | delta = sched_clock_cpu(cpu) - __this_cpu_read(irq_start_time); | |
60 | __this_cpu_add(irq_start_time, delta); | |
61 | ||
62 | irq_time_write_begin(); | |
63 | /* | |
64 | * We do not account for softirq time from ksoftirqd here. | |
65 | * We want to continue accounting softirq time to ksoftirqd thread | |
66 | * in that case, so as not to confuse scheduler with a special task | |
67 | * that do not consume any time, but still wants to run. | |
68 | */ | |
69 | if (hardirq_count()) | |
70 | __this_cpu_add(cpu_hardirq_time, delta); | |
71 | else if (in_serving_softirq() && curr != this_cpu_ksoftirqd()) | |
72 | __this_cpu_add(cpu_softirq_time, delta); | |
73 | ||
74 | irq_time_write_end(); | |
75 | local_irq_restore(flags); | |
76 | } | |
3e1df4f5 | 77 | EXPORT_SYMBOL_GPL(irqtime_account_irq); |
73fbec60 FW |
78 | |
79 | static int irqtime_account_hi_update(void) | |
80 | { | |
81 | u64 *cpustat = kcpustat_this_cpu->cpustat; | |
82 | unsigned long flags; | |
83 | u64 latest_ns; | |
84 | int ret = 0; | |
85 | ||
86 | local_irq_save(flags); | |
87 | latest_ns = this_cpu_read(cpu_hardirq_time); | |
88 | if (nsecs_to_cputime64(latest_ns) > cpustat[CPUTIME_IRQ]) | |
89 | ret = 1; | |
90 | local_irq_restore(flags); | |
91 | return ret; | |
92 | } | |
93 | ||
94 | static int irqtime_account_si_update(void) | |
95 | { | |
96 | u64 *cpustat = kcpustat_this_cpu->cpustat; | |
97 | unsigned long flags; | |
98 | u64 latest_ns; | |
99 | int ret = 0; | |
100 | ||
101 | local_irq_save(flags); | |
102 | latest_ns = this_cpu_read(cpu_softirq_time); | |
103 | if (nsecs_to_cputime64(latest_ns) > cpustat[CPUTIME_SOFTIRQ]) | |
104 | ret = 1; | |
105 | local_irq_restore(flags); | |
106 | return ret; | |
107 | } | |
108 | ||
109 | #else /* CONFIG_IRQ_TIME_ACCOUNTING */ | |
110 | ||
111 | #define sched_clock_irqtime (0) | |
112 | ||
113 | #endif /* !CONFIG_IRQ_TIME_ACCOUNTING */ | |
114 | ||
115 | static inline void task_group_account_field(struct task_struct *p, int index, | |
116 | u64 tmp) | |
117 | { | |
118 | #ifdef CONFIG_CGROUP_CPUACCT | |
119 | struct kernel_cpustat *kcpustat; | |
120 | struct cpuacct *ca; | |
121 | #endif | |
122 | /* | |
123 | * Since all updates are sure to touch the root cgroup, we | |
124 | * get ourselves ahead and touch it first. If the root cgroup | |
125 | * is the only cgroup, then nothing else should be necessary. | |
126 | * | |
127 | */ | |
128 | __get_cpu_var(kernel_cpustat).cpustat[index] += tmp; | |
129 | ||
130 | #ifdef CONFIG_CGROUP_CPUACCT | |
131 | if (unlikely(!cpuacct_subsys.active)) | |
132 | return; | |
133 | ||
134 | rcu_read_lock(); | |
135 | ca = task_ca(p); | |
136 | while (ca && (ca != &root_cpuacct)) { | |
137 | kcpustat = this_cpu_ptr(ca->cpustat); | |
138 | kcpustat->cpustat[index] += tmp; | |
139 | ca = parent_ca(ca); | |
140 | } | |
141 | rcu_read_unlock(); | |
142 | #endif | |
143 | } | |
144 | ||
145 | /* | |
146 | * Account user cpu time to a process. | |
147 | * @p: the process that the cpu time gets accounted to | |
148 | * @cputime: the cpu time spent in user space since the last update | |
149 | * @cputime_scaled: cputime scaled by cpu frequency | |
150 | */ | |
151 | void account_user_time(struct task_struct *p, cputime_t cputime, | |
152 | cputime_t cputime_scaled) | |
153 | { | |
154 | int index; | |
155 | ||
156 | /* Add user time to process. */ | |
157 | p->utime += cputime; | |
158 | p->utimescaled += cputime_scaled; | |
159 | account_group_user_time(p, cputime); | |
160 | ||
161 | index = (TASK_NICE(p) > 0) ? CPUTIME_NICE : CPUTIME_USER; | |
162 | ||
163 | /* Add user time to cpustat. */ | |
164 | task_group_account_field(p, index, (__force u64) cputime); | |
165 | ||
166 | /* Account for user time used */ | |
167 | acct_update_integrals(p); | |
168 | } | |
169 | ||
170 | /* | |
171 | * Account guest cpu time to a process. | |
172 | * @p: the process that the cpu time gets accounted to | |
173 | * @cputime: the cpu time spent in virtual machine since the last update | |
174 | * @cputime_scaled: cputime scaled by cpu frequency | |
175 | */ | |
176 | static void account_guest_time(struct task_struct *p, cputime_t cputime, | |
177 | cputime_t cputime_scaled) | |
178 | { | |
179 | u64 *cpustat = kcpustat_this_cpu->cpustat; | |
180 | ||
181 | /* Add guest time to process. */ | |
182 | p->utime += cputime; | |
183 | p->utimescaled += cputime_scaled; | |
184 | account_group_user_time(p, cputime); | |
185 | p->gtime += cputime; | |
186 | ||
187 | /* Add guest time to cpustat. */ | |
188 | if (TASK_NICE(p) > 0) { | |
189 | cpustat[CPUTIME_NICE] += (__force u64) cputime; | |
190 | cpustat[CPUTIME_GUEST_NICE] += (__force u64) cputime; | |
191 | } else { | |
192 | cpustat[CPUTIME_USER] += (__force u64) cputime; | |
193 | cpustat[CPUTIME_GUEST] += (__force u64) cputime; | |
194 | } | |
195 | } | |
196 | ||
197 | /* | |
198 | * Account system cpu time to a process and desired cpustat field | |
199 | * @p: the process that the cpu time gets accounted to | |
200 | * @cputime: the cpu time spent in kernel space since the last update | |
201 | * @cputime_scaled: cputime scaled by cpu frequency | |
202 | * @target_cputime64: pointer to cpustat field that has to be updated | |
203 | */ | |
204 | static inline | |
205 | void __account_system_time(struct task_struct *p, cputime_t cputime, | |
206 | cputime_t cputime_scaled, int index) | |
207 | { | |
208 | /* Add system time to process. */ | |
209 | p->stime += cputime; | |
210 | p->stimescaled += cputime_scaled; | |
211 | account_group_system_time(p, cputime); | |
212 | ||
213 | /* Add system time to cpustat. */ | |
214 | task_group_account_field(p, index, (__force u64) cputime); | |
215 | ||
216 | /* Account for system time used */ | |
217 | acct_update_integrals(p); | |
218 | } | |
219 | ||
220 | /* | |
221 | * Account system cpu time to a process. | |
222 | * @p: the process that the cpu time gets accounted to | |
223 | * @hardirq_offset: the offset to subtract from hardirq_count() | |
224 | * @cputime: the cpu time spent in kernel space since the last update | |
225 | * @cputime_scaled: cputime scaled by cpu frequency | |
226 | */ | |
227 | void account_system_time(struct task_struct *p, int hardirq_offset, | |
228 | cputime_t cputime, cputime_t cputime_scaled) | |
229 | { | |
230 | int index; | |
231 | ||
232 | if ((p->flags & PF_VCPU) && (irq_count() - hardirq_offset == 0)) { | |
233 | account_guest_time(p, cputime, cputime_scaled); | |
234 | return; | |
235 | } | |
236 | ||
237 | if (hardirq_count() - hardirq_offset) | |
238 | index = CPUTIME_IRQ; | |
239 | else if (in_serving_softirq()) | |
240 | index = CPUTIME_SOFTIRQ; | |
241 | else | |
242 | index = CPUTIME_SYSTEM; | |
243 | ||
244 | __account_system_time(p, cputime, cputime_scaled, index); | |
245 | } | |
246 | ||
247 | /* | |
248 | * Account for involuntary wait time. | |
249 | * @cputime: the cpu time spent in involuntary wait | |
250 | */ | |
251 | void account_steal_time(cputime_t cputime) | |
252 | { | |
253 | u64 *cpustat = kcpustat_this_cpu->cpustat; | |
254 | ||
255 | cpustat[CPUTIME_STEAL] += (__force u64) cputime; | |
256 | } | |
257 | ||
258 | /* | |
259 | * Account for idle time. | |
260 | * @cputime: the cpu time spent in idle wait | |
261 | */ | |
262 | void account_idle_time(cputime_t cputime) | |
263 | { | |
264 | u64 *cpustat = kcpustat_this_cpu->cpustat; | |
265 | struct rq *rq = this_rq(); | |
266 | ||
267 | if (atomic_read(&rq->nr_iowait) > 0) | |
268 | cpustat[CPUTIME_IOWAIT] += (__force u64) cputime; | |
269 | else | |
270 | cpustat[CPUTIME_IDLE] += (__force u64) cputime; | |
271 | } | |
272 | ||
273 | static __always_inline bool steal_account_process_tick(void) | |
274 | { | |
275 | #ifdef CONFIG_PARAVIRT | |
276 | if (static_key_false(¶virt_steal_enabled)) { | |
277 | u64 steal, st = 0; | |
278 | ||
279 | steal = paravirt_steal_clock(smp_processor_id()); | |
280 | steal -= this_rq()->prev_steal_time; | |
281 | ||
282 | st = steal_ticks(steal); | |
283 | this_rq()->prev_steal_time += st * TICK_NSEC; | |
284 | ||
285 | account_steal_time(st); | |
286 | return st; | |
287 | } | |
288 | #endif | |
289 | return false; | |
290 | } | |
291 | ||
a634f933 FW |
292 | /* |
293 | * Accumulate raw cputime values of dead tasks (sig->[us]time) and live | |
294 | * tasks (sum on group iteration) belonging to @tsk's group. | |
295 | */ | |
296 | void thread_group_cputime(struct task_struct *tsk, struct task_cputime *times) | |
297 | { | |
298 | struct signal_struct *sig = tsk->signal; | |
299 | struct task_struct *t; | |
300 | ||
301 | times->utime = sig->utime; | |
302 | times->stime = sig->stime; | |
303 | times->sum_exec_runtime = sig->sum_sched_runtime; | |
304 | ||
305 | rcu_read_lock(); | |
306 | /* make sure we can trust tsk->thread_group list */ | |
307 | if (!likely(pid_alive(tsk))) | |
308 | goto out; | |
309 | ||
310 | t = tsk; | |
311 | do { | |
312 | times->utime += t->utime; | |
313 | times->stime += t->stime; | |
314 | times->sum_exec_runtime += task_sched_runtime(t); | |
315 | } while_each_thread(tsk, t); | |
316 | out: | |
317 | rcu_read_unlock(); | |
318 | } | |
319 | ||
73fbec60 FW |
320 | #ifdef CONFIG_IRQ_TIME_ACCOUNTING |
321 | /* | |
322 | * Account a tick to a process and cpustat | |
323 | * @p: the process that the cpu time gets accounted to | |
324 | * @user_tick: is the tick from userspace | |
325 | * @rq: the pointer to rq | |
326 | * | |
327 | * Tick demultiplexing follows the order | |
328 | * - pending hardirq update | |
329 | * - pending softirq update | |
330 | * - user_time | |
331 | * - idle_time | |
332 | * - system time | |
333 | * - check for guest_time | |
334 | * - else account as system_time | |
335 | * | |
336 | * Check for hardirq is done both for system and user time as there is | |
337 | * no timer going off while we are on hardirq and hence we may never get an | |
338 | * opportunity to update it solely in system time. | |
339 | * p->stime and friends are only updated on system time and not on irq | |
340 | * softirq as those do not count in task exec_runtime any more. | |
341 | */ | |
342 | static void irqtime_account_process_tick(struct task_struct *p, int user_tick, | |
343 | struct rq *rq) | |
344 | { | |
345 | cputime_t one_jiffy_scaled = cputime_to_scaled(cputime_one_jiffy); | |
346 | u64 *cpustat = kcpustat_this_cpu->cpustat; | |
347 | ||
348 | if (steal_account_process_tick()) | |
349 | return; | |
350 | ||
351 | if (irqtime_account_hi_update()) { | |
352 | cpustat[CPUTIME_IRQ] += (__force u64) cputime_one_jiffy; | |
353 | } else if (irqtime_account_si_update()) { | |
354 | cpustat[CPUTIME_SOFTIRQ] += (__force u64) cputime_one_jiffy; | |
355 | } else if (this_cpu_ksoftirqd() == p) { | |
356 | /* | |
357 | * ksoftirqd time do not get accounted in cpu_softirq_time. | |
358 | * So, we have to handle it separately here. | |
359 | * Also, p->stime needs to be updated for ksoftirqd. | |
360 | */ | |
361 | __account_system_time(p, cputime_one_jiffy, one_jiffy_scaled, | |
362 | CPUTIME_SOFTIRQ); | |
363 | } else if (user_tick) { | |
364 | account_user_time(p, cputime_one_jiffy, one_jiffy_scaled); | |
365 | } else if (p == rq->idle) { | |
366 | account_idle_time(cputime_one_jiffy); | |
367 | } else if (p->flags & PF_VCPU) { /* System time or guest time */ | |
368 | account_guest_time(p, cputime_one_jiffy, one_jiffy_scaled); | |
369 | } else { | |
370 | __account_system_time(p, cputime_one_jiffy, one_jiffy_scaled, | |
371 | CPUTIME_SYSTEM); | |
372 | } | |
373 | } | |
374 | ||
375 | static void irqtime_account_idle_ticks(int ticks) | |
376 | { | |
377 | int i; | |
378 | struct rq *rq = this_rq(); | |
379 | ||
380 | for (i = 0; i < ticks; i++) | |
381 | irqtime_account_process_tick(current, 0, rq); | |
382 | } | |
383 | #else /* CONFIG_IRQ_TIME_ACCOUNTING */ | |
3f4724ea FW |
384 | static inline void irqtime_account_idle_ticks(int ticks) {} |
385 | static inline void irqtime_account_process_tick(struct task_struct *p, int user_tick, | |
73fbec60 FW |
386 | struct rq *rq) {} |
387 | #endif /* CONFIG_IRQ_TIME_ACCOUNTING */ | |
388 | ||
3f4724ea | 389 | #ifndef CONFIG_VIRT_CPU_ACCOUNTING_NATIVE |
73fbec60 FW |
390 | /* |
391 | * Account a single tick of cpu time. | |
392 | * @p: the process that the cpu time gets accounted to | |
393 | * @user_tick: indicates if the tick is a user or a system tick | |
394 | */ | |
395 | void account_process_tick(struct task_struct *p, int user_tick) | |
396 | { | |
397 | cputime_t one_jiffy_scaled = cputime_to_scaled(cputime_one_jiffy); | |
398 | struct rq *rq = this_rq(); | |
399 | ||
3f4724ea FW |
400 | if (vtime_accounting_enabled()) |
401 | return; | |
402 | ||
73fbec60 FW |
403 | if (sched_clock_irqtime) { |
404 | irqtime_account_process_tick(p, user_tick, rq); | |
405 | return; | |
406 | } | |
407 | ||
408 | if (steal_account_process_tick()) | |
409 | return; | |
410 | ||
411 | if (user_tick) | |
412 | account_user_time(p, cputime_one_jiffy, one_jiffy_scaled); | |
413 | else if ((p != rq->idle) || (irq_count() != HARDIRQ_OFFSET)) | |
414 | account_system_time(p, HARDIRQ_OFFSET, cputime_one_jiffy, | |
415 | one_jiffy_scaled); | |
416 | else | |
417 | account_idle_time(cputime_one_jiffy); | |
418 | } | |
419 | ||
420 | /* | |
421 | * Account multiple ticks of steal time. | |
422 | * @p: the process from which the cpu time has been stolen | |
423 | * @ticks: number of stolen ticks | |
424 | */ | |
425 | void account_steal_ticks(unsigned long ticks) | |
426 | { | |
427 | account_steal_time(jiffies_to_cputime(ticks)); | |
428 | } | |
429 | ||
430 | /* | |
431 | * Account multiple ticks of idle time. | |
432 | * @ticks: number of stolen ticks | |
433 | */ | |
434 | void account_idle_ticks(unsigned long ticks) | |
435 | { | |
436 | ||
437 | if (sched_clock_irqtime) { | |
438 | irqtime_account_idle_ticks(ticks); | |
439 | return; | |
440 | } | |
441 | ||
442 | account_idle_time(jiffies_to_cputime(ticks)); | |
443 | } | |
3f4724ea | 444 | #endif /* !CONFIG_VIRT_CPU_ACCOUNTING_NATIVE */ |
73fbec60 FW |
445 | |
446 | /* | |
447 | * Use precise platform statistics if available: | |
448 | */ | |
449 | #ifdef CONFIG_VIRT_CPU_ACCOUNTING | |
e80d0a1a | 450 | void task_cputime_adjusted(struct task_struct *p, cputime_t *ut, cputime_t *st) |
73fbec60 FW |
451 | { |
452 | *ut = p->utime; | |
453 | *st = p->stime; | |
454 | } | |
455 | ||
e80d0a1a | 456 | void thread_group_cputime_adjusted(struct task_struct *p, cputime_t *ut, cputime_t *st) |
73fbec60 FW |
457 | { |
458 | struct task_cputime cputime; | |
459 | ||
460 | thread_group_cputime(p, &cputime); | |
461 | ||
462 | *ut = cputime.utime; | |
463 | *st = cputime.stime; | |
464 | } | |
a7e1a9e3 | 465 | |
fd25b4c2 | 466 | void vtime_account_system_irqsafe(struct task_struct *tsk) |
11113334 FW |
467 | { |
468 | unsigned long flags; | |
469 | ||
470 | local_irq_save(flags); | |
fd25b4c2 | 471 | vtime_account_system(tsk); |
11113334 FW |
472 | local_irq_restore(flags); |
473 | } | |
fd25b4c2 | 474 | EXPORT_SYMBOL_GPL(vtime_account_system_irqsafe); |
11113334 | 475 | |
e3942ba0 FW |
476 | #ifndef __ARCH_HAS_VTIME_TASK_SWITCH |
477 | void vtime_task_switch(struct task_struct *prev) | |
478 | { | |
3f4724ea FW |
479 | if (!vtime_accounting_enabled()) |
480 | return; | |
481 | ||
e3942ba0 FW |
482 | if (is_idle_task(prev)) |
483 | vtime_account_idle(prev); | |
484 | else | |
485 | vtime_account_system(prev); | |
486 | ||
abf917cd | 487 | #ifdef CONFIG_VIRT_CPU_ACCOUNTING_NATIVE |
e3942ba0 | 488 | vtime_account_user(prev); |
abf917cd | 489 | #endif |
e3942ba0 FW |
490 | arch_vtime_task_switch(prev); |
491 | } | |
492 | #endif | |
11113334 | 493 | |
a7e1a9e3 FW |
494 | /* |
495 | * Archs that account the whole time spent in the idle task | |
496 | * (outside irq) as idle time can rely on this and just implement | |
fd25b4c2 | 497 | * vtime_account_system() and vtime_account_idle(). Archs that |
a7e1a9e3 FW |
498 | * have other meaning of the idle time (s390 only includes the |
499 | * time spent by the CPU when it's in low power mode) must override | |
500 | * vtime_account(). | |
501 | */ | |
502 | #ifndef __ARCH_HAS_VTIME_ACCOUNT | |
503 | void vtime_account(struct task_struct *tsk) | |
504 | { | |
3f4724ea FW |
505 | if (!vtime_accounting_enabled()) |
506 | return; | |
507 | ||
abf917cd FW |
508 | if (!in_interrupt()) { |
509 | /* | |
510 | * If we interrupted user, context_tracking_in_user() | |
511 | * is 1 because the context tracking don't hook | |
512 | * on irq entry/exit. This way we know if | |
513 | * we need to flush user time on kernel entry. | |
514 | */ | |
515 | if (context_tracking_in_user()) { | |
516 | vtime_account_user(tsk); | |
517 | return; | |
518 | } | |
519 | ||
520 | if (is_idle_task(tsk)) { | |
521 | vtime_account_idle(tsk); | |
522 | return; | |
523 | } | |
524 | } | |
525 | vtime_account_system(tsk); | |
a7e1a9e3 FW |
526 | } |
527 | EXPORT_SYMBOL_GPL(vtime_account); | |
528 | #endif /* __ARCH_HAS_VTIME_ACCOUNT */ | |
529 | ||
3f4724ea | 530 | #else /* !CONFIG_VIRT_CPU_ACCOUNTING */ |
73fbec60 | 531 | |
73fbec60 FW |
532 | static cputime_t scale_utime(cputime_t utime, cputime_t rtime, cputime_t total) |
533 | { | |
534 | u64 temp = (__force u64) rtime; | |
535 | ||
536 | temp *= (__force u64) utime; | |
537 | ||
538 | if (sizeof(cputime_t) == 4) | |
539 | temp = div_u64(temp, (__force u32) total); | |
540 | else | |
541 | temp = div64_u64(temp, (__force u64) total); | |
542 | ||
543 | return (__force cputime_t) temp; | |
544 | } | |
545 | ||
fa092057 FW |
546 | /* |
547 | * Adjust tick based cputime random precision against scheduler | |
548 | * runtime accounting. | |
549 | */ | |
d37f761d FW |
550 | static void cputime_adjust(struct task_cputime *curr, |
551 | struct cputime *prev, | |
552 | cputime_t *ut, cputime_t *st) | |
73fbec60 | 553 | { |
d37f761d | 554 | cputime_t rtime, utime, total; |
73fbec60 | 555 | |
d37f761d FW |
556 | utime = curr->utime; |
557 | total = utime + curr->stime; | |
fa092057 | 558 | |
73fbec60 | 559 | /* |
fa092057 FW |
560 | * Tick based cputime accounting depend on random scheduling |
561 | * timeslices of a task to be interrupted or not by the timer. | |
562 | * Depending on these circumstances, the number of these interrupts | |
563 | * may be over or under-optimistic, matching the real user and system | |
564 | * cputime with a variable precision. | |
565 | * | |
566 | * Fix this by scaling these tick based values against the total | |
567 | * runtime accounted by the CFS scheduler. | |
73fbec60 | 568 | */ |
d37f761d | 569 | rtime = nsecs_to_cputime(curr->sum_exec_runtime); |
73fbec60 FW |
570 | |
571 | if (total) | |
572 | utime = scale_utime(utime, rtime, total); | |
573 | else | |
574 | utime = rtime; | |
575 | ||
576 | /* | |
fa092057 FW |
577 | * If the tick based count grows faster than the scheduler one, |
578 | * the result of the scaling may go backward. | |
579 | * Let's enforce monotonicity. | |
73fbec60 | 580 | */ |
d37f761d FW |
581 | prev->utime = max(prev->utime, utime); |
582 | prev->stime = max(prev->stime, rtime - prev->utime); | |
583 | ||
584 | *ut = prev->utime; | |
585 | *st = prev->stime; | |
586 | } | |
73fbec60 | 587 | |
d37f761d FW |
588 | void task_cputime_adjusted(struct task_struct *p, cputime_t *ut, cputime_t *st) |
589 | { | |
590 | struct task_cputime cputime = { | |
591 | .utime = p->utime, | |
592 | .stime = p->stime, | |
593 | .sum_exec_runtime = p->se.sum_exec_runtime, | |
594 | }; | |
595 | ||
596 | cputime_adjust(&cputime, &p->prev_cputime, ut, st); | |
73fbec60 FW |
597 | } |
598 | ||
599 | /* | |
600 | * Must be called with siglock held. | |
601 | */ | |
e80d0a1a | 602 | void thread_group_cputime_adjusted(struct task_struct *p, cputime_t *ut, cputime_t *st) |
73fbec60 | 603 | { |
73fbec60 | 604 | struct task_cputime cputime; |
73fbec60 FW |
605 | |
606 | thread_group_cputime(p, &cputime); | |
d37f761d | 607 | cputime_adjust(&cputime, &p->signal->prev_cputime, ut, st); |
73fbec60 | 608 | } |
3f4724ea | 609 | #endif /* !CONFIG_VIRT_CPU_ACCOUNTING */ |
abf917cd FW |
610 | |
611 | #ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN | |
612 | static DEFINE_PER_CPU(unsigned long long, cputime_snap); | |
613 | ||
614 | static cputime_t get_vtime_delta(void) | |
615 | { | |
616 | unsigned long long delta; | |
617 | ||
618 | delta = sched_clock() - __this_cpu_read(cputime_snap); | |
619 | __this_cpu_add(cputime_snap, delta); | |
620 | ||
621 | /* CHECKME: always safe to convert nsecs to cputime? */ | |
622 | return nsecs_to_cputime(delta); | |
623 | } | |
624 | ||
625 | void vtime_account_system(struct task_struct *tsk) | |
626 | { | |
3f4724ea FW |
627 | cputime_t delta_cpu; |
628 | ||
629 | if (!vtime_accounting_enabled()) | |
630 | return; | |
abf917cd | 631 | |
3f4724ea | 632 | delta_cpu = get_vtime_delta(); |
abf917cd FW |
633 | account_system_time(tsk, irq_count(), delta_cpu, cputime_to_scaled(delta_cpu)); |
634 | } | |
635 | ||
636 | void vtime_account_user(struct task_struct *tsk) | |
637 | { | |
3f4724ea FW |
638 | cputime_t delta_cpu; |
639 | ||
640 | if (!vtime_accounting_enabled()) | |
641 | return; | |
642 | ||
643 | delta_cpu = get_vtime_delta(); | |
abf917cd FW |
644 | |
645 | account_user_time(tsk, delta_cpu, cputime_to_scaled(delta_cpu)); | |
646 | } | |
647 | ||
648 | void vtime_account_idle(struct task_struct *tsk) | |
649 | { | |
650 | cputime_t delta_cpu = get_vtime_delta(); | |
651 | ||
652 | account_idle_time(delta_cpu); | |
653 | } | |
3f4724ea FW |
654 | |
655 | bool vtime_accounting_enabled(void) | |
656 | { | |
657 | return context_tracking_active(); | |
658 | } | |
abf917cd | 659 | #endif /* CONFIG_VIRT_CPU_ACCOUNTING_GEN */ |