blob: 024fe1998ad54be7110776cdb855fab2e1d73084 [file] [log] [blame]
Frederic Weisbecker73fbec62012-06-16 15:57:37 +02001#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>
Frederic Weisbeckerabf917c2012-07-25 07:56:04 +02006#include <linux/context_tracking.h>
Frederic Weisbecker73fbec62012-06-16 15:57:37 +02007#include "sched.h"
8
9
10#ifdef CONFIG_IRQ_TIME_ACCOUNTING
11
12/*
13 * There are no locks covering percpu hardirq/softirq time.
Frederic Weisbeckerbf9fae92012-09-08 15:23:11 +020014 * They are only modified in vtime_account, on corresponding CPU
Frederic Weisbecker73fbec62012-06-16 15:57:37 +020015 * 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
Frederic Weisbeckerbf9fae92012-09-08 15:23:11 +020018 * race with irq/vtime_account on this CPU. We would either get old
Frederic Weisbecker73fbec62012-06-16 15:57:37 +020019 * 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 */
23DEFINE_PER_CPU(u64, cpu_hardirq_time);
24DEFINE_PER_CPU(u64, cpu_softirq_time);
25
26static DEFINE_PER_CPU(u64, irq_start_time);
27static int sched_clock_irqtime;
28
29void enable_sched_clock_irqtime(void)
30{
31 sched_clock_irqtime = 1;
32}
33
34void disable_sched_clock_irqtime(void)
35{
36 sched_clock_irqtime = 0;
37}
38
39#ifndef CONFIG_64BIT
40DEFINE_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 */
Frederic Weisbecker3e1df4f52012-10-06 05:23:22 +020047void irqtime_account_irq(struct task_struct *curr)
Frederic Weisbecker73fbec62012-06-16 15:57:37 +020048{
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}
Frederic Weisbecker3e1df4f52012-10-06 05:23:22 +020077EXPORT_SYMBOL_GPL(irqtime_account_irq);
Frederic Weisbecker73fbec62012-06-16 15:57:37 +020078
79static 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
94static 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
115static 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 */
151void 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 */
Frederic Weisbecker6fac4822012-11-13 14:20:55 +0100167 acct_account_cputime(p);
Frederic Weisbecker73fbec62012-06-16 15:57:37 +0200168}
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 */
176static 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 */
204static inline
205void __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 */
Frederic Weisbecker6fac4822012-11-13 14:20:55 +0100217 acct_account_cputime(p);
Frederic Weisbecker73fbec62012-06-16 15:57:37 +0200218}
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 */
227void 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 */
251void 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 */
262void 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
273static __always_inline bool steal_account_process_tick(void)
274{
275#ifdef CONFIG_PARAVIRT
276 if (static_key_false(&paravirt_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
Frederic Weisbeckera634f932012-11-21 15:55:59 +0100292/*
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 */
296void thread_group_cputime(struct task_struct *tsk, struct task_cputime *times)
297{
298 struct signal_struct *sig = tsk->signal;
Frederic Weisbecker6fac4822012-11-13 14:20:55 +0100299 cputime_t utime, stime;
Frederic Weisbeckera634f932012-11-21 15:55:59 +0100300 struct task_struct *t;
301
302 times->utime = sig->utime;
303 times->stime = sig->stime;
304 times->sum_exec_runtime = sig->sum_sched_runtime;
305
306 rcu_read_lock();
307 /* make sure we can trust tsk->thread_group list */
308 if (!likely(pid_alive(tsk)))
309 goto out;
310
311 t = tsk;
312 do {
Frederic Weisbecker6fac4822012-11-13 14:20:55 +0100313 task_cputime(tsk, &utime, &stime);
314 times->utime += utime;
315 times->stime += stime;
Frederic Weisbeckera634f932012-11-21 15:55:59 +0100316 times->sum_exec_runtime += task_sched_runtime(t);
317 } while_each_thread(tsk, t);
318out:
319 rcu_read_unlock();
320}
321
Frederic Weisbecker73fbec62012-06-16 15:57:37 +0200322#ifdef CONFIG_IRQ_TIME_ACCOUNTING
323/*
324 * Account a tick to a process and cpustat
325 * @p: the process that the cpu time gets accounted to
326 * @user_tick: is the tick from userspace
327 * @rq: the pointer to rq
328 *
329 * Tick demultiplexing follows the order
330 * - pending hardirq update
331 * - pending softirq update
332 * - user_time
333 * - idle_time
334 * - system time
335 * - check for guest_time
336 * - else account as system_time
337 *
338 * Check for hardirq is done both for system and user time as there is
339 * no timer going off while we are on hardirq and hence we may never get an
340 * opportunity to update it solely in system time.
341 * p->stime and friends are only updated on system time and not on irq
342 * softirq as those do not count in task exec_runtime any more.
343 */
344static void irqtime_account_process_tick(struct task_struct *p, int user_tick,
345 struct rq *rq)
346{
347 cputime_t one_jiffy_scaled = cputime_to_scaled(cputime_one_jiffy);
348 u64 *cpustat = kcpustat_this_cpu->cpustat;
349
350 if (steal_account_process_tick())
351 return;
352
353 if (irqtime_account_hi_update()) {
354 cpustat[CPUTIME_IRQ] += (__force u64) cputime_one_jiffy;
355 } else if (irqtime_account_si_update()) {
356 cpustat[CPUTIME_SOFTIRQ] += (__force u64) cputime_one_jiffy;
357 } else if (this_cpu_ksoftirqd() == p) {
358 /*
359 * ksoftirqd time do not get accounted in cpu_softirq_time.
360 * So, we have to handle it separately here.
361 * Also, p->stime needs to be updated for ksoftirqd.
362 */
363 __account_system_time(p, cputime_one_jiffy, one_jiffy_scaled,
364 CPUTIME_SOFTIRQ);
365 } else if (user_tick) {
366 account_user_time(p, cputime_one_jiffy, one_jiffy_scaled);
367 } else if (p == rq->idle) {
368 account_idle_time(cputime_one_jiffy);
369 } else if (p->flags & PF_VCPU) { /* System time or guest time */
370 account_guest_time(p, cputime_one_jiffy, one_jiffy_scaled);
371 } else {
372 __account_system_time(p, cputime_one_jiffy, one_jiffy_scaled,
373 CPUTIME_SYSTEM);
374 }
375}
376
377static void irqtime_account_idle_ticks(int ticks)
378{
379 int i;
380 struct rq *rq = this_rq();
381
382 for (i = 0; i < ticks; i++)
383 irqtime_account_process_tick(current, 0, rq);
384}
385#else /* CONFIG_IRQ_TIME_ACCOUNTING */
Frederic Weisbecker3f4724e2012-07-16 18:00:34 +0200386static inline void irqtime_account_idle_ticks(int ticks) {}
387static inline void irqtime_account_process_tick(struct task_struct *p, int user_tick,
Frederic Weisbecker73fbec62012-06-16 15:57:37 +0200388 struct rq *rq) {}
389#endif /* CONFIG_IRQ_TIME_ACCOUNTING */
390
Frederic Weisbecker73fbec62012-06-16 15:57:37 +0200391/*
392 * Use precise platform statistics if available:
393 */
394#ifdef CONFIG_VIRT_CPU_ACCOUNTING
Frederic Weisbeckera7e1a9e2012-09-08 16:14:02 +0200395
Frederic Weisbeckere3942ba2012-11-14 00:24:25 +0100396#ifndef __ARCH_HAS_VTIME_TASK_SWITCH
397void vtime_task_switch(struct task_struct *prev)
398{
Frederic Weisbecker3f4724e2012-07-16 18:00:34 +0200399 if (!vtime_accounting_enabled())
400 return;
401
Frederic Weisbeckere3942ba2012-11-14 00:24:25 +0100402 if (is_idle_task(prev))
403 vtime_account_idle(prev);
404 else
405 vtime_account_system(prev);
406
Frederic Weisbeckerabf917c2012-07-25 07:56:04 +0200407#ifdef CONFIG_VIRT_CPU_ACCOUNTING_NATIVE
Frederic Weisbeckere3942ba2012-11-14 00:24:25 +0100408 vtime_account_user(prev);
Frederic Weisbeckerabf917c2012-07-25 07:56:04 +0200409#endif
Frederic Weisbeckere3942ba2012-11-14 00:24:25 +0100410 arch_vtime_task_switch(prev);
411}
412#endif
Frederic Weisbecker11113332012-10-24 18:05:51 +0200413
Frederic Weisbeckera7e1a9e2012-09-08 16:14:02 +0200414/*
415 * Archs that account the whole time spent in the idle task
416 * (outside irq) as idle time can rely on this and just implement
Frederic Weisbeckerfd25b4c2012-11-13 18:21:22 +0100417 * vtime_account_system() and vtime_account_idle(). Archs that
Frederic Weisbeckera7e1a9e2012-09-08 16:14:02 +0200418 * have other meaning of the idle time (s390 only includes the
419 * time spent by the CPU when it's in low power mode) must override
420 * vtime_account().
421 */
422#ifndef __ARCH_HAS_VTIME_ACCOUNT
Frederic Weisbecker6a616712012-12-16 20:00:34 +0100423void vtime_account_irq_enter(struct task_struct *tsk)
Frederic Weisbeckera7e1a9e2012-09-08 16:14:02 +0200424{
Frederic Weisbecker3f4724e2012-07-16 18:00:34 +0200425 if (!vtime_accounting_enabled())
426 return;
427
Frederic Weisbeckerabf917c2012-07-25 07:56:04 +0200428 if (!in_interrupt()) {
429 /*
430 * If we interrupted user, context_tracking_in_user()
431 * is 1 because the context tracking don't hook
432 * on irq entry/exit. This way we know if
433 * we need to flush user time on kernel entry.
434 */
435 if (context_tracking_in_user()) {
436 vtime_account_user(tsk);
437 return;
438 }
439
440 if (is_idle_task(tsk)) {
441 vtime_account_idle(tsk);
442 return;
443 }
444 }
445 vtime_account_system(tsk);
Frederic Weisbeckera7e1a9e2012-09-08 16:14:02 +0200446}
Frederic Weisbecker6a616712012-12-16 20:00:34 +0100447EXPORT_SYMBOL_GPL(vtime_account_irq_enter);
Frederic Weisbeckera7e1a9e2012-09-08 16:14:02 +0200448#endif /* __ARCH_HAS_VTIME_ACCOUNT */
Frederic Weisbecker9fbc42e2013-02-25 17:25:39 +0100449#endif /* CONFIG_VIRT_CPU_ACCOUNTING */
Frederic Weisbeckera7e1a9e2012-09-08 16:14:02 +0200450
Frederic Weisbecker9fbc42e2013-02-25 17:25:39 +0100451
452#ifdef CONFIG_VIRT_CPU_ACCOUNTING_NATIVE
453void task_cputime_adjusted(struct task_struct *p, cputime_t *ut, cputime_t *st)
454{
455 *ut = p->utime;
456 *st = p->stime;
457}
458
459void thread_group_cputime_adjusted(struct task_struct *p, cputime_t *ut, cputime_t *st)
460{
461 struct task_cputime cputime;
462
463 thread_group_cputime(p, &cputime);
464
465 *ut = cputime.utime;
466 *st = cputime.stime;
467}
468#else /* !CONFIG_VIRT_CPU_ACCOUNTING_NATIVE */
469/*
470 * Account a single tick of cpu time.
471 * @p: the process that the cpu time gets accounted to
472 * @user_tick: indicates if the tick is a user or a system tick
473 */
474void account_process_tick(struct task_struct *p, int user_tick)
475{
476 cputime_t one_jiffy_scaled = cputime_to_scaled(cputime_one_jiffy);
477 struct rq *rq = this_rq();
478
479 if (vtime_accounting_enabled())
480 return;
481
482 if (sched_clock_irqtime) {
483 irqtime_account_process_tick(p, user_tick, rq);
484 return;
485 }
486
487 if (steal_account_process_tick())
488 return;
489
490 if (user_tick)
491 account_user_time(p, cputime_one_jiffy, one_jiffy_scaled);
492 else if ((p != rq->idle) || (irq_count() != HARDIRQ_OFFSET))
493 account_system_time(p, HARDIRQ_OFFSET, cputime_one_jiffy,
494 one_jiffy_scaled);
495 else
496 account_idle_time(cputime_one_jiffy);
497}
498
499/*
500 * Account multiple ticks of steal time.
501 * @p: the process from which the cpu time has been stolen
502 * @ticks: number of stolen ticks
503 */
504void account_steal_ticks(unsigned long ticks)
505{
506 account_steal_time(jiffies_to_cputime(ticks));
507}
508
509/*
510 * Account multiple ticks of idle time.
511 * @ticks: number of stolen ticks
512 */
513void account_idle_ticks(unsigned long ticks)
514{
515
516 if (sched_clock_irqtime) {
517 irqtime_account_idle_ticks(ticks);
518 return;
519 }
520
521 account_idle_time(jiffies_to_cputime(ticks));
522}
Frederic Weisbecker73fbec62012-06-16 15:57:37 +0200523
Frederic Weisbecker62188452013-01-26 17:19:42 +0100524static cputime_t scale_stime(cputime_t stime, cputime_t rtime, cputime_t total)
Frederic Weisbecker73fbec62012-06-16 15:57:37 +0200525{
526 u64 temp = (__force u64) rtime;
527
Frederic Weisbecker62188452013-01-26 17:19:42 +0100528 temp *= (__force u64) stime;
Frederic Weisbecker73fbec62012-06-16 15:57:37 +0200529
530 if (sizeof(cputime_t) == 4)
531 temp = div_u64(temp, (__force u32) total);
532 else
533 temp = div64_u64(temp, (__force u64) total);
534
535 return (__force cputime_t) temp;
536}
537
Frederic Weisbeckerfa092052012-11-28 17:00:57 +0100538/*
539 * Adjust tick based cputime random precision against scheduler
540 * runtime accounting.
541 */
Frederic Weisbeckerd37f761d2012-11-22 00:58:35 +0100542static void cputime_adjust(struct task_cputime *curr,
543 struct cputime *prev,
544 cputime_t *ut, cputime_t *st)
Frederic Weisbecker73fbec62012-06-16 15:57:37 +0200545{
Frederic Weisbecker62188452013-01-26 17:19:42 +0100546 cputime_t rtime, stime, total;
Frederic Weisbecker73fbec62012-06-16 15:57:37 +0200547
Frederic Weisbecker9fbc42e2013-02-25 17:25:39 +0100548 if (vtime_accounting_enabled()) {
549 *ut = curr->utime;
550 *st = curr->stime;
551 return;
552 }
553
Frederic Weisbecker62188452013-01-26 17:19:42 +0100554 stime = curr->stime;
555 total = stime + curr->utime;
Frederic Weisbeckerfa092052012-11-28 17:00:57 +0100556
Frederic Weisbecker73fbec62012-06-16 15:57:37 +0200557 /*
Frederic Weisbeckerfa092052012-11-28 17:00:57 +0100558 * Tick based cputime accounting depend on random scheduling
559 * timeslices of a task to be interrupted or not by the timer.
560 * Depending on these circumstances, the number of these interrupts
561 * may be over or under-optimistic, matching the real user and system
562 * cputime with a variable precision.
563 *
564 * Fix this by scaling these tick based values against the total
565 * runtime accounted by the CFS scheduler.
Frederic Weisbecker73fbec62012-06-16 15:57:37 +0200566 */
Frederic Weisbeckerd37f761d2012-11-22 00:58:35 +0100567 rtime = nsecs_to_cputime(curr->sum_exec_runtime);
Frederic Weisbecker73fbec62012-06-16 15:57:37 +0200568
569 if (total)
Frederic Weisbecker62188452013-01-26 17:19:42 +0100570 stime = scale_stime(stime, rtime, total);
Frederic Weisbecker73fbec62012-06-16 15:57:37 +0200571 else
Frederic Weisbecker62188452013-01-26 17:19:42 +0100572 stime = rtime;
Frederic Weisbecker73fbec62012-06-16 15:57:37 +0200573
574 /*
Frederic Weisbeckerfa092052012-11-28 17:00:57 +0100575 * If the tick based count grows faster than the scheduler one,
576 * the result of the scaling may go backward.
577 * Let's enforce monotonicity.
Frederic Weisbecker73fbec62012-06-16 15:57:37 +0200578 */
Frederic Weisbecker62188452013-01-26 17:19:42 +0100579 prev->stime = max(prev->stime, stime);
580 prev->utime = max(prev->utime, rtime - prev->stime);
Frederic Weisbecker73fbec62012-06-16 15:57:37 +0200581
Frederic Weisbeckerd37f761d2012-11-22 00:58:35 +0100582 *ut = prev->utime;
583 *st = prev->stime;
584}
585
586void task_cputime_adjusted(struct task_struct *p, cputime_t *ut, cputime_t *st)
587{
588 struct task_cputime cputime = {
Frederic Weisbeckerd37f761d2012-11-22 00:58:35 +0100589 .sum_exec_runtime = p->se.sum_exec_runtime,
590 };
591
Frederic Weisbecker6fac4822012-11-13 14:20:55 +0100592 task_cputime(p, &cputime.utime, &cputime.stime);
Frederic Weisbeckerd37f761d2012-11-22 00:58:35 +0100593 cputime_adjust(&cputime, &p->prev_cputime, ut, st);
Frederic Weisbecker73fbec62012-06-16 15:57:37 +0200594}
595
596/*
597 * Must be called with siglock held.
598 */
Frederic Weisbeckere80d0a12012-11-21 16:26:44 +0100599void thread_group_cputime_adjusted(struct task_struct *p, cputime_t *ut, cputime_t *st)
Frederic Weisbecker73fbec62012-06-16 15:57:37 +0200600{
Frederic Weisbecker73fbec62012-06-16 15:57:37 +0200601 struct task_cputime cputime;
Frederic Weisbecker73fbec62012-06-16 15:57:37 +0200602
603 thread_group_cputime(p, &cputime);
Frederic Weisbeckerd37f761d2012-11-22 00:58:35 +0100604 cputime_adjust(&cputime, &p->signal->prev_cputime, ut, st);
Frederic Weisbecker73fbec62012-06-16 15:57:37 +0200605}
Frederic Weisbecker9fbc42e2013-02-25 17:25:39 +0100606#endif /* !CONFIG_VIRT_CPU_ACCOUNTING_NATIVE */
Frederic Weisbeckerabf917c2012-07-25 07:56:04 +0200607
608#ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
Frederic Weisbecker6a616712012-12-16 20:00:34 +0100609static unsigned long long vtime_delta(struct task_struct *tsk)
Frederic Weisbeckerabf917c2012-07-25 07:56:04 +0200610{
Frederic Weisbecker6a616712012-12-16 20:00:34 +0100611 unsigned long long clock;
Frederic Weisbeckerabf917c2012-07-25 07:56:04 +0200612
Frederic Weisbecker7f6575f2013-02-23 17:28:45 +0100613 clock = local_clock();
Frederic Weisbecker6a616712012-12-16 20:00:34 +0100614 if (clock < tsk->vtime_snap)
615 return 0;
616
617 return clock - tsk->vtime_snap;
618}
619
620static cputime_t get_vtime_delta(struct task_struct *tsk)
621{
622 unsigned long long delta = vtime_delta(tsk);
623
624 WARN_ON_ONCE(tsk->vtime_snap_whence == VTIME_SLEEPING);
625 tsk->vtime_snap += delta;
Frederic Weisbeckerabf917c2012-07-25 07:56:04 +0200626
627 /* CHECKME: always safe to convert nsecs to cputime? */
628 return nsecs_to_cputime(delta);
629}
630
Frederic Weisbecker6a616712012-12-16 20:00:34 +0100631static void __vtime_account_system(struct task_struct *tsk)
632{
633 cputime_t delta_cpu = get_vtime_delta(tsk);
634
635 account_system_time(tsk, irq_count(), delta_cpu, cputime_to_scaled(delta_cpu));
636}
637
Frederic Weisbeckerabf917c2012-07-25 07:56:04 +0200638void vtime_account_system(struct task_struct *tsk)
639{
Frederic Weisbecker3f4724e2012-07-16 18:00:34 +0200640 if (!vtime_accounting_enabled())
641 return;
642
Frederic Weisbecker6a616712012-12-16 20:00:34 +0100643 write_seqlock(&tsk->vtime_seqlock);
644 __vtime_account_system(tsk);
645 write_sequnlock(&tsk->vtime_seqlock);
646}
647
648void vtime_account_irq_exit(struct task_struct *tsk)
649{
650 if (!vtime_accounting_enabled())
651 return;
652
653 write_seqlock(&tsk->vtime_seqlock);
654 if (context_tracking_in_user())
655 tsk->vtime_snap_whence = VTIME_USER;
656 __vtime_account_system(tsk);
657 write_sequnlock(&tsk->vtime_seqlock);
Frederic Weisbeckerabf917c2012-07-25 07:56:04 +0200658}
659
660void vtime_account_user(struct task_struct *tsk)
661{
Frederic Weisbecker3f4724e2012-07-16 18:00:34 +0200662 cputime_t delta_cpu;
663
664 if (!vtime_accounting_enabled())
665 return;
666
Frederic Weisbecker6a616712012-12-16 20:00:34 +0100667 delta_cpu = get_vtime_delta(tsk);
Frederic Weisbeckerabf917c2012-07-25 07:56:04 +0200668
Frederic Weisbecker6a616712012-12-16 20:00:34 +0100669 write_seqlock(&tsk->vtime_seqlock);
670 tsk->vtime_snap_whence = VTIME_SYS;
Frederic Weisbeckerabf917c2012-07-25 07:56:04 +0200671 account_user_time(tsk, delta_cpu, cputime_to_scaled(delta_cpu));
Frederic Weisbecker6a616712012-12-16 20:00:34 +0100672 write_sequnlock(&tsk->vtime_seqlock);
673}
674
675void vtime_user_enter(struct task_struct *tsk)
676{
677 if (!vtime_accounting_enabled())
678 return;
679
680 write_seqlock(&tsk->vtime_seqlock);
681 tsk->vtime_snap_whence = VTIME_USER;
682 __vtime_account_system(tsk);
683 write_sequnlock(&tsk->vtime_seqlock);
684}
685
686void vtime_guest_enter(struct task_struct *tsk)
687{
688 write_seqlock(&tsk->vtime_seqlock);
689 __vtime_account_system(tsk);
690 current->flags |= PF_VCPU;
691 write_sequnlock(&tsk->vtime_seqlock);
692}
693
694void vtime_guest_exit(struct task_struct *tsk)
695{
696 write_seqlock(&tsk->vtime_seqlock);
697 __vtime_account_system(tsk);
698 current->flags &= ~PF_VCPU;
699 write_sequnlock(&tsk->vtime_seqlock);
Frederic Weisbeckerabf917c2012-07-25 07:56:04 +0200700}
701
702void vtime_account_idle(struct task_struct *tsk)
703{
Frederic Weisbecker6a616712012-12-16 20:00:34 +0100704 cputime_t delta_cpu = get_vtime_delta(tsk);
Frederic Weisbeckerabf917c2012-07-25 07:56:04 +0200705
706 account_idle_time(delta_cpu);
707}
Frederic Weisbecker3f4724e2012-07-16 18:00:34 +0200708
709bool vtime_accounting_enabled(void)
710{
711 return context_tracking_active();
712}
Frederic Weisbecker6a616712012-12-16 20:00:34 +0100713
714void arch_vtime_task_switch(struct task_struct *prev)
715{
716 write_seqlock(&prev->vtime_seqlock);
717 prev->vtime_snap_whence = VTIME_SLEEPING;
718 write_sequnlock(&prev->vtime_seqlock);
719
720 write_seqlock(&current->vtime_seqlock);
721 current->vtime_snap_whence = VTIME_SYS;
722 current->vtime_snap = sched_clock();
723 write_sequnlock(&current->vtime_seqlock);
724}
725
726void vtime_init_idle(struct task_struct *t)
727{
728 unsigned long flags;
729
730 write_seqlock_irqsave(&t->vtime_seqlock, flags);
731 t->vtime_snap_whence = VTIME_SYS;
732 t->vtime_snap = sched_clock();
733 write_sequnlock_irqrestore(&t->vtime_seqlock, flags);
734}
735
736cputime_t task_gtime(struct task_struct *t)
737{
Frederic Weisbecker6a616712012-12-16 20:00:34 +0100738 unsigned int seq;
739 cputime_t gtime;
740
741 do {
Thomas Gleixnercdc4e862013-02-15 23:47:07 +0100742 seq = read_seqbegin(&t->vtime_seqlock);
Frederic Weisbecker6a616712012-12-16 20:00:34 +0100743
744 gtime = t->gtime;
745 if (t->flags & PF_VCPU)
746 gtime += vtime_delta(t);
747
Thomas Gleixnercdc4e862013-02-15 23:47:07 +0100748 } while (read_seqretry(&t->vtime_seqlock, seq));
Frederic Weisbecker6a616712012-12-16 20:00:34 +0100749
750 return gtime;
751}
752
753/*
754 * Fetch cputime raw values from fields of task_struct and
755 * add up the pending nohz execution time since the last
756 * cputime snapshot.
757 */
758static void
759fetch_task_cputime(struct task_struct *t,
760 cputime_t *u_dst, cputime_t *s_dst,
761 cputime_t *u_src, cputime_t *s_src,
762 cputime_t *udelta, cputime_t *sdelta)
763{
Frederic Weisbecker6a616712012-12-16 20:00:34 +0100764 unsigned int seq;
765 unsigned long long delta;
766
767 do {
768 *udelta = 0;
769 *sdelta = 0;
770
Thomas Gleixnercdc4e862013-02-15 23:47:07 +0100771 seq = read_seqbegin(&t->vtime_seqlock);
Frederic Weisbecker6a616712012-12-16 20:00:34 +0100772
773 if (u_dst)
774 *u_dst = *u_src;
775 if (s_dst)
776 *s_dst = *s_src;
777
778 /* Task is sleeping, nothing to add */
779 if (t->vtime_snap_whence == VTIME_SLEEPING ||
780 is_idle_task(t))
781 continue;
782
783 delta = vtime_delta(t);
784
785 /*
786 * Task runs either in user or kernel space, add pending nohz time to
787 * the right place.
788 */
789 if (t->vtime_snap_whence == VTIME_USER || t->flags & PF_VCPU) {
790 *udelta = delta;
791 } else {
792 if (t->vtime_snap_whence == VTIME_SYS)
793 *sdelta = delta;
794 }
Thomas Gleixnercdc4e862013-02-15 23:47:07 +0100795 } while (read_seqretry(&t->vtime_seqlock, seq));
Frederic Weisbecker6a616712012-12-16 20:00:34 +0100796}
797
798
799void task_cputime(struct task_struct *t, cputime_t *utime, cputime_t *stime)
800{
801 cputime_t udelta, sdelta;
802
803 fetch_task_cputime(t, utime, stime, &t->utime,
804 &t->stime, &udelta, &sdelta);
805 if (utime)
806 *utime += udelta;
807 if (stime)
808 *stime += sdelta;
809}
810
811void task_cputime_scaled(struct task_struct *t,
812 cputime_t *utimescaled, cputime_t *stimescaled)
813{
814 cputime_t udelta, sdelta;
815
816 fetch_task_cputime(t, utimescaled, stimescaled,
817 &t->utimescaled, &t->stimescaled, &udelta, &sdelta);
818 if (utimescaled)
819 *utimescaled += cputime_to_scaled(udelta);
820 if (stimescaled)
821 *stimescaled += cputime_to_scaled(sdelta);
822}
Frederic Weisbeckerabf917c2012-07-25 07:56:04 +0200823#endif /* CONFIG_VIRT_CPU_ACCOUNTING_GEN */