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>
6 #include <linux/context_tracking.h>
9 #include <asm/paravirt.h>
13 #ifdef CONFIG_IRQ_TIME_ACCOUNTING
16 * There are no locks covering percpu hardirq/softirq time.
17 * They are only modified in vtime_account, on corresponding CPU
18 * with interrupts disabled. So, writes are safe.
19 * They are read and saved off onto struct rq in update_rq_clock().
20 * This may result in other CPU reading this CPU's irq time and can
21 * race with irq/vtime_account on this CPU. We would either get old
22 * or new value with a side effect of accounting a slice of irq time to wrong
23 * task when irq is in progress while we read rq->clock. That is a worthy
24 * compromise in place of having locks on each irq in account_system_time.
26 DEFINE_PER_CPU(u64
, cpu_hardirq_time
);
27 DEFINE_PER_CPU(u64
, cpu_softirq_time
);
29 static DEFINE_PER_CPU(u64
, irq_start_time
);
30 static int sched_clock_irqtime
;
32 void enable_sched_clock_irqtime(void)
34 sched_clock_irqtime
= 1;
37 void disable_sched_clock_irqtime(void)
39 sched_clock_irqtime
= 0;
43 DEFINE_PER_CPU(seqcount_t
, irq_time_seq
);
44 #endif /* CONFIG_64BIT */
47 * Called before incrementing preempt_count on {soft,}irq_enter
48 * and before decrementing preempt_count on {soft,}irq_exit.
50 void irqtime_account_irq(struct task_struct
*curr
)
55 if (!sched_clock_irqtime
)
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
);
62 irq_time_write_begin();
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.
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
);
76 EXPORT_SYMBOL_GPL(irqtime_account_irq
);
78 static cputime_t
irqtime_account_hi_update(cputime_t maxtime
)
80 u64
*cpustat
= kcpustat_this_cpu
->cpustat
;
82 cputime_t irq_cputime
;
84 local_irq_save(flags
);
85 irq_cputime
= nsecs_to_cputime64(this_cpu_read(cpu_hardirq_time
)) -
87 irq_cputime
= min(irq_cputime
, maxtime
);
88 cpustat
[CPUTIME_IRQ
] += irq_cputime
;
89 local_irq_restore(flags
);
93 static cputime_t
irqtime_account_si_update(cputime_t maxtime
)
95 u64
*cpustat
= kcpustat_this_cpu
->cpustat
;
97 cputime_t softirq_cputime
;
99 local_irq_save(flags
);
100 softirq_cputime
= nsecs_to_cputime64(this_cpu_read(cpu_softirq_time
)) -
101 cpustat
[CPUTIME_SOFTIRQ
];
102 softirq_cputime
= min(softirq_cputime
, maxtime
);
103 cpustat
[CPUTIME_SOFTIRQ
] += softirq_cputime
;
104 local_irq_restore(flags
);
105 return softirq_cputime
;
108 #else /* CONFIG_IRQ_TIME_ACCOUNTING */
110 #define sched_clock_irqtime (0)
112 static cputime_t
irqtime_account_hi_update(cputime_t dummy
)
117 static cputime_t
irqtime_account_si_update(cputime_t dummy
)
122 #endif /* !CONFIG_IRQ_TIME_ACCOUNTING */
124 static inline void task_group_account_field(struct task_struct
*p
, int index
,
128 * Since all updates are sure to touch the root cgroup, we
129 * get ourselves ahead and touch it first. If the root cgroup
130 * is the only cgroup, then nothing else should be necessary.
133 __this_cpu_add(kernel_cpustat
.cpustat
[index
], tmp
);
135 cpuacct_account_field(p
, index
, tmp
);
139 * Account user cpu time to a process.
140 * @p: the process that the cpu time gets accounted to
141 * @cputime: the cpu time spent in user space since the last update
142 * @cputime_scaled: cputime scaled by cpu frequency
144 void account_user_time(struct task_struct
*p
, cputime_t cputime
,
145 cputime_t cputime_scaled
)
149 /* Add user time to process. */
151 p
->utimescaled
+= cputime_scaled
;
152 account_group_user_time(p
, cputime
);
154 index
= (task_nice(p
) > 0) ? CPUTIME_NICE
: CPUTIME_USER
;
156 /* Add user time to cpustat. */
157 task_group_account_field(p
, index
, (__force u64
) cputime
);
159 /* Account for user time used */
160 acct_account_cputime(p
);
164 * Account guest cpu time to a process.
165 * @p: the process that the cpu time gets accounted to
166 * @cputime: the cpu time spent in virtual machine since the last update
167 * @cputime_scaled: cputime scaled by cpu frequency
169 static void account_guest_time(struct task_struct
*p
, cputime_t cputime
,
170 cputime_t cputime_scaled
)
172 u64
*cpustat
= kcpustat_this_cpu
->cpustat
;
174 /* Add guest time to process. */
176 p
->utimescaled
+= cputime_scaled
;
177 account_group_user_time(p
, cputime
);
180 /* Add guest time to cpustat. */
181 if (task_nice(p
) > 0) {
182 cpustat
[CPUTIME_NICE
] += (__force u64
) cputime
;
183 cpustat
[CPUTIME_GUEST_NICE
] += (__force u64
) cputime
;
185 cpustat
[CPUTIME_USER
] += (__force u64
) cputime
;
186 cpustat
[CPUTIME_GUEST
] += (__force u64
) cputime
;
191 * Account system cpu time to a process and desired cpustat field
192 * @p: the process that the cpu time gets accounted to
193 * @cputime: the cpu time spent in kernel space since the last update
194 * @cputime_scaled: cputime scaled by cpu frequency
195 * @target_cputime64: pointer to cpustat field that has to be updated
198 void __account_system_time(struct task_struct
*p
, cputime_t cputime
,
199 cputime_t cputime_scaled
, int index
)
201 /* Add system time to process. */
203 p
->stimescaled
+= cputime_scaled
;
204 account_group_system_time(p
, cputime
);
206 /* Add system time to cpustat. */
207 task_group_account_field(p
, index
, (__force u64
) cputime
);
209 /* Account for system time used */
210 acct_account_cputime(p
);
214 * Account system cpu time to a process.
215 * @p: the process that the cpu time gets accounted to
216 * @hardirq_offset: the offset to subtract from hardirq_count()
217 * @cputime: the cpu time spent in kernel space since the last update
218 * @cputime_scaled: cputime scaled by cpu frequency
220 void account_system_time(struct task_struct
*p
, int hardirq_offset
,
221 cputime_t cputime
, cputime_t cputime_scaled
)
225 if ((p
->flags
& PF_VCPU
) && (irq_count() - hardirq_offset
== 0)) {
226 account_guest_time(p
, cputime
, cputime_scaled
);
230 if (hardirq_count() - hardirq_offset
)
232 else if (in_serving_softirq())
233 index
= CPUTIME_SOFTIRQ
;
235 index
= CPUTIME_SYSTEM
;
237 __account_system_time(p
, cputime
, cputime_scaled
, index
);
241 * Account for involuntary wait time.
242 * @cputime: the cpu time spent in involuntary wait
244 void account_steal_time(cputime_t cputime
)
246 u64
*cpustat
= kcpustat_this_cpu
->cpustat
;
248 cpustat
[CPUTIME_STEAL
] += (__force u64
) cputime
;
252 * Account for idle time.
253 * @cputime: the cpu time spent in idle wait
255 void account_idle_time(cputime_t cputime
)
257 u64
*cpustat
= kcpustat_this_cpu
->cpustat
;
258 struct rq
*rq
= this_rq();
260 if (atomic_read(&rq
->nr_iowait
) > 0)
261 cpustat
[CPUTIME_IOWAIT
] += (__force u64
) cputime
;
263 cpustat
[CPUTIME_IDLE
] += (__force u64
) cputime
;
267 * When a guest is interrupted for a longer amount of time, missed clock
268 * ticks are not redelivered later. Due to that, this function may on
269 * occasion account more time than the calling functions think elapsed.
271 static __always_inline cputime_t
steal_account_process_time(cputime_t maxtime
)
273 #ifdef CONFIG_PARAVIRT
274 if (static_key_false(¶virt_steal_enabled
)) {
275 cputime_t steal_cputime
;
278 steal
= paravirt_steal_clock(smp_processor_id());
279 steal
-= this_rq()->prev_steal_time
;
281 steal_cputime
= min(nsecs_to_cputime(steal
), maxtime
);
282 account_steal_time(steal_cputime
);
283 this_rq()->prev_steal_time
+= cputime_to_nsecs(steal_cputime
);
285 return steal_cputime
;
292 * Account how much elapsed time was spent in steal, irq, or softirq time.
294 static inline cputime_t
account_other_time(cputime_t max
)
298 accounted
= steal_account_process_time(max
);
301 accounted
+= irqtime_account_hi_update(max
- accounted
);
304 accounted
+= irqtime_account_si_update(max
- accounted
);
310 static inline u64
read_sum_exec_runtime(struct task_struct
*t
)
312 return t
->se
.sum_exec_runtime
;
315 static u64
read_sum_exec_runtime(struct task_struct
*t
)
321 rq
= task_rq_lock(t
, &rf
);
322 ns
= t
->se
.sum_exec_runtime
;
323 task_rq_unlock(rq
, t
, &rf
);
330 * Accumulate raw cputime values of dead tasks (sig->[us]time) and live
331 * tasks (sum on group iteration) belonging to @tsk's group.
333 void thread_group_cputime(struct task_struct
*tsk
, struct task_cputime
*times
)
335 struct signal_struct
*sig
= tsk
->signal
;
336 cputime_t utime
, stime
;
337 struct task_struct
*t
;
338 unsigned int seq
, nextseq
;
342 * Update current task runtime to account pending time since last
343 * scheduler action or thread_group_cputime() call. This thread group
344 * might have other running tasks on different CPUs, but updating
345 * their runtime can affect syscall performance, so we skip account
346 * those pending times and rely only on values updated on tick or
347 * other scheduler action.
349 if (same_thread_group(current
, tsk
))
350 (void) task_sched_runtime(current
);
353 /* Attempt a lockless read on the first round. */
357 flags
= read_seqbegin_or_lock_irqsave(&sig
->stats_lock
, &seq
);
358 times
->utime
= sig
->utime
;
359 times
->stime
= sig
->stime
;
360 times
->sum_exec_runtime
= sig
->sum_sched_runtime
;
362 for_each_thread(tsk
, t
) {
363 task_cputime(t
, &utime
, &stime
);
364 times
->utime
+= utime
;
365 times
->stime
+= stime
;
366 times
->sum_exec_runtime
+= read_sum_exec_runtime(t
);
368 /* If lockless access failed, take the lock. */
370 } while (need_seqretry(&sig
->stats_lock
, seq
));
371 done_seqretry_irqrestore(&sig
->stats_lock
, seq
, flags
);
375 #ifdef CONFIG_IRQ_TIME_ACCOUNTING
377 * Account a tick to a process and cpustat
378 * @p: the process that the cpu time gets accounted to
379 * @user_tick: is the tick from userspace
380 * @rq: the pointer to rq
382 * Tick demultiplexing follows the order
383 * - pending hardirq update
384 * - pending softirq update
388 * - check for guest_time
389 * - else account as system_time
391 * Check for hardirq is done both for system and user time as there is
392 * no timer going off while we are on hardirq and hence we may never get an
393 * opportunity to update it solely in system time.
394 * p->stime and friends are only updated on system time and not on irq
395 * softirq as those do not count in task exec_runtime any more.
397 static void irqtime_account_process_tick(struct task_struct
*p
, int user_tick
,
398 struct rq
*rq
, int ticks
)
400 u64 cputime
= (__force u64
) cputime_one_jiffy
* ticks
;
401 cputime_t scaled
, other
;
404 * When returning from idle, many ticks can get accounted at
405 * once, including some ticks of steal, irq, and softirq time.
406 * Subtract those ticks from the amount of time accounted to
407 * idle, or potentially user or system time. Due to rounding,
408 * other time can exceed ticks occasionally.
410 other
= account_other_time(ULONG_MAX
);
411 if (other
>= cputime
)
414 scaled
= cputime_to_scaled(cputime
);
416 if (this_cpu_ksoftirqd() == p
) {
418 * ksoftirqd time do not get accounted in cpu_softirq_time.
419 * So, we have to handle it separately here.
420 * Also, p->stime needs to be updated for ksoftirqd.
422 __account_system_time(p
, cputime
, scaled
, CPUTIME_SOFTIRQ
);
423 } else if (user_tick
) {
424 account_user_time(p
, cputime
, scaled
);
425 } else if (p
== rq
->idle
) {
426 account_idle_time(cputime
);
427 } else if (p
->flags
& PF_VCPU
) { /* System time or guest time */
428 account_guest_time(p
, cputime
, scaled
);
430 __account_system_time(p
, cputime
, scaled
, CPUTIME_SYSTEM
);
434 static void irqtime_account_idle_ticks(int ticks
)
436 struct rq
*rq
= this_rq();
438 irqtime_account_process_tick(current
, 0, rq
, ticks
);
440 #else /* CONFIG_IRQ_TIME_ACCOUNTING */
441 static inline void irqtime_account_idle_ticks(int ticks
) {}
442 static inline void irqtime_account_process_tick(struct task_struct
*p
, int user_tick
,
443 struct rq
*rq
, int nr_ticks
) {}
444 #endif /* CONFIG_IRQ_TIME_ACCOUNTING */
447 * Use precise platform statistics if available:
449 #ifdef CONFIG_VIRT_CPU_ACCOUNTING
451 #ifndef __ARCH_HAS_VTIME_TASK_SWITCH
452 void vtime_common_task_switch(struct task_struct
*prev
)
454 if (is_idle_task(prev
))
455 vtime_account_idle(prev
);
457 vtime_account_system(prev
);
459 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_NATIVE
460 vtime_account_user(prev
);
462 arch_vtime_task_switch(prev
);
466 #endif /* CONFIG_VIRT_CPU_ACCOUNTING */
469 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_NATIVE
471 * Archs that account the whole time spent in the idle task
472 * (outside irq) as idle time can rely on this and just implement
473 * vtime_account_system() and vtime_account_idle(). Archs that
474 * have other meaning of the idle time (s390 only includes the
475 * time spent by the CPU when it's in low power mode) must override
478 #ifndef __ARCH_HAS_VTIME_ACCOUNT
479 void vtime_account_irq_enter(struct task_struct
*tsk
)
481 if (!in_interrupt() && is_idle_task(tsk
))
482 vtime_account_idle(tsk
);
484 vtime_account_system(tsk
);
486 EXPORT_SYMBOL_GPL(vtime_account_irq_enter
);
487 #endif /* __ARCH_HAS_VTIME_ACCOUNT */
489 void task_cputime_adjusted(struct task_struct
*p
, cputime_t
*ut
, cputime_t
*st
)
494 EXPORT_SYMBOL_GPL(task_cputime_adjusted
);
496 void thread_group_cputime_adjusted(struct task_struct
*p
, cputime_t
*ut
, cputime_t
*st
)
498 struct task_cputime cputime
;
500 thread_group_cputime(p
, &cputime
);
505 #else /* !CONFIG_VIRT_CPU_ACCOUNTING_NATIVE */
507 * Account a single tick of cpu time.
508 * @p: the process that the cpu time gets accounted to
509 * @user_tick: indicates if the tick is a user or a system tick
511 void account_process_tick(struct task_struct
*p
, int user_tick
)
513 cputime_t cputime
, scaled
, steal
;
514 struct rq
*rq
= this_rq();
516 if (vtime_accounting_cpu_enabled())
519 if (sched_clock_irqtime
) {
520 irqtime_account_process_tick(p
, user_tick
, rq
, 1);
524 cputime
= cputime_one_jiffy
;
525 steal
= steal_account_process_time(ULONG_MAX
);
527 if (steal
>= cputime
)
531 scaled
= cputime_to_scaled(cputime
);
534 account_user_time(p
, cputime
, scaled
);
535 else if ((p
!= rq
->idle
) || (irq_count() != HARDIRQ_OFFSET
))
536 account_system_time(p
, HARDIRQ_OFFSET
, cputime
, scaled
);
538 account_idle_time(cputime
);
542 * Account multiple ticks of idle time.
543 * @ticks: number of stolen ticks
545 void account_idle_ticks(unsigned long ticks
)
547 cputime_t cputime
, steal
;
549 if (sched_clock_irqtime
) {
550 irqtime_account_idle_ticks(ticks
);
554 cputime
= jiffies_to_cputime(ticks
);
555 steal
= steal_account_process_time(ULONG_MAX
);
557 if (steal
>= cputime
)
561 account_idle_time(cputime
);
565 * Perform (stime * rtime) / total, but avoid multiplication overflow by
566 * loosing precision when the numbers are big.
568 static cputime_t
scale_stime(u64 stime
, u64 rtime
, u64 total
)
573 /* Make sure "rtime" is the bigger of stime/rtime */
577 /* Make sure 'total' fits in 32 bits */
581 /* Does rtime (and thus stime) fit in 32 bits? */
585 /* Can we just balance rtime/stime rather than dropping bits? */
589 /* We can grow stime and shrink rtime and try to make them both fit */
595 /* We drop from rtime, it has more bits than stime */
601 * Make sure gcc understands that this is a 32x32->64 multiply,
602 * followed by a 64/32->64 divide.
604 scaled
= div_u64((u64
) (u32
) stime
* (u64
) (u32
) rtime
, (u32
)total
);
605 return (__force cputime_t
) scaled
;
609 * Adjust tick based cputime random precision against scheduler runtime
612 * Tick based cputime accounting depend on random scheduling timeslices of a
613 * task to be interrupted or not by the timer. Depending on these
614 * circumstances, the number of these interrupts may be over or
615 * under-optimistic, matching the real user and system cputime with a variable
618 * Fix this by scaling these tick based values against the total runtime
619 * accounted by the CFS scheduler.
621 * This code provides the following guarantees:
623 * stime + utime == rtime
624 * stime_i+1 >= stime_i, utime_i+1 >= utime_i
626 * Assuming that rtime_i+1 >= rtime_i.
628 static void cputime_adjust(struct task_cputime
*curr
,
629 struct prev_cputime
*prev
,
630 cputime_t
*ut
, cputime_t
*st
)
632 cputime_t rtime
, stime
, utime
;
635 /* Serialize concurrent callers such that we can honour our guarantees */
636 raw_spin_lock_irqsave(&prev
->lock
, flags
);
637 rtime
= nsecs_to_cputime(curr
->sum_exec_runtime
);
640 * This is possible under two circumstances:
641 * - rtime isn't monotonic after all (a bug);
642 * - we got reordered by the lock.
644 * In both cases this acts as a filter such that the rest of the code
645 * can assume it is monotonic regardless of anything else.
647 if (prev
->stime
+ prev
->utime
>= rtime
)
654 * If either stime or both stime and utime are 0, assume all runtime is
655 * userspace. Once a task gets some ticks, the monotonicy code at
656 * 'update' will ensure things converge to the observed ratio.
668 stime
= scale_stime((__force u64
)stime
, (__force u64
)rtime
,
669 (__force u64
)(stime
+ utime
));
673 * Make sure stime doesn't go backwards; this preserves monotonicity
674 * for utime because rtime is monotonic.
676 * utime_i+1 = rtime_i+1 - stime_i
677 * = rtime_i+1 - (rtime_i - utime_i)
678 * = (rtime_i+1 - rtime_i) + utime_i
681 if (stime
< prev
->stime
)
683 utime
= rtime
- stime
;
686 * Make sure utime doesn't go backwards; this still preserves
687 * monotonicity for stime, analogous argument to above.
689 if (utime
< prev
->utime
) {
691 stime
= rtime
- utime
;
699 raw_spin_unlock_irqrestore(&prev
->lock
, flags
);
702 void task_cputime_adjusted(struct task_struct
*p
, cputime_t
*ut
, cputime_t
*st
)
704 struct task_cputime cputime
= {
705 .sum_exec_runtime
= p
->se
.sum_exec_runtime
,
708 task_cputime(p
, &cputime
.utime
, &cputime
.stime
);
709 cputime_adjust(&cputime
, &p
->prev_cputime
, ut
, st
);
711 EXPORT_SYMBOL_GPL(task_cputime_adjusted
);
713 void thread_group_cputime_adjusted(struct task_struct
*p
, cputime_t
*ut
, cputime_t
*st
)
715 struct task_cputime cputime
;
717 thread_group_cputime(p
, &cputime
);
718 cputime_adjust(&cputime
, &p
->signal
->prev_cputime
, ut
, st
);
720 #endif /* !CONFIG_VIRT_CPU_ACCOUNTING_NATIVE */
722 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
723 static cputime_t
vtime_delta(struct task_struct
*tsk
)
725 unsigned long now
= READ_ONCE(jiffies
);
727 if (time_before(now
, (unsigned long)tsk
->vtime_snap
))
730 return jiffies_to_cputime(now
- tsk
->vtime_snap
);
733 static cputime_t
get_vtime_delta(struct task_struct
*tsk
)
735 unsigned long now
= READ_ONCE(jiffies
);
736 cputime_t delta
, other
;
739 * Unlike tick based timing, vtime based timing never has lost
740 * ticks, and no need for steal time accounting to make up for
741 * lost ticks. Vtime accounts a rounded version of actual
742 * elapsed time. Limit account_other_time to prevent rounding
743 * errors from causing elapsed vtime to go negative.
745 delta
= jiffies_to_cputime(now
- tsk
->vtime_snap
);
746 other
= account_other_time(delta
);
747 WARN_ON_ONCE(tsk
->vtime_snap_whence
== VTIME_INACTIVE
);
748 tsk
->vtime_snap
= now
;
750 return delta
- other
;
753 static void __vtime_account_system(struct task_struct
*tsk
)
755 cputime_t delta_cpu
= get_vtime_delta(tsk
);
757 account_system_time(tsk
, irq_count(), delta_cpu
, cputime_to_scaled(delta_cpu
));
760 void vtime_account_system(struct task_struct
*tsk
)
762 if (!vtime_delta(tsk
))
765 write_seqcount_begin(&tsk
->vtime_seqcount
);
766 __vtime_account_system(tsk
);
767 write_seqcount_end(&tsk
->vtime_seqcount
);
770 void vtime_account_user(struct task_struct
*tsk
)
774 write_seqcount_begin(&tsk
->vtime_seqcount
);
775 tsk
->vtime_snap_whence
= VTIME_SYS
;
776 if (vtime_delta(tsk
)) {
777 delta_cpu
= get_vtime_delta(tsk
);
778 account_user_time(tsk
, delta_cpu
, cputime_to_scaled(delta_cpu
));
780 write_seqcount_end(&tsk
->vtime_seqcount
);
783 void vtime_user_enter(struct task_struct
*tsk
)
785 write_seqcount_begin(&tsk
->vtime_seqcount
);
786 if (vtime_delta(tsk
))
787 __vtime_account_system(tsk
);
788 tsk
->vtime_snap_whence
= VTIME_USER
;
789 write_seqcount_end(&tsk
->vtime_seqcount
);
792 void vtime_guest_enter(struct task_struct
*tsk
)
795 * The flags must be updated under the lock with
796 * the vtime_snap flush and update.
797 * That enforces a right ordering and update sequence
798 * synchronization against the reader (task_gtime())
799 * that can thus safely catch up with a tickless delta.
801 write_seqcount_begin(&tsk
->vtime_seqcount
);
802 if (vtime_delta(tsk
))
803 __vtime_account_system(tsk
);
804 current
->flags
|= PF_VCPU
;
805 write_seqcount_end(&tsk
->vtime_seqcount
);
807 EXPORT_SYMBOL_GPL(vtime_guest_enter
);
809 void vtime_guest_exit(struct task_struct
*tsk
)
811 write_seqcount_begin(&tsk
->vtime_seqcount
);
812 __vtime_account_system(tsk
);
813 current
->flags
&= ~PF_VCPU
;
814 write_seqcount_end(&tsk
->vtime_seqcount
);
816 EXPORT_SYMBOL_GPL(vtime_guest_exit
);
818 void vtime_account_idle(struct task_struct
*tsk
)
820 cputime_t delta_cpu
= get_vtime_delta(tsk
);
822 account_idle_time(delta_cpu
);
825 void arch_vtime_task_switch(struct task_struct
*prev
)
827 write_seqcount_begin(&prev
->vtime_seqcount
);
828 prev
->vtime_snap_whence
= VTIME_INACTIVE
;
829 write_seqcount_end(&prev
->vtime_seqcount
);
831 write_seqcount_begin(¤t
->vtime_seqcount
);
832 current
->vtime_snap_whence
= VTIME_SYS
;
833 current
->vtime_snap
= jiffies
;
834 write_seqcount_end(¤t
->vtime_seqcount
);
837 void vtime_init_idle(struct task_struct
*t
, int cpu
)
841 local_irq_save(flags
);
842 write_seqcount_begin(&t
->vtime_seqcount
);
843 t
->vtime_snap_whence
= VTIME_SYS
;
844 t
->vtime_snap
= jiffies
;
845 write_seqcount_end(&t
->vtime_seqcount
);
846 local_irq_restore(flags
);
849 cputime_t
task_gtime(struct task_struct
*t
)
854 if (!vtime_accounting_enabled())
858 seq
= read_seqcount_begin(&t
->vtime_seqcount
);
861 if (t
->vtime_snap_whence
== VTIME_SYS
&& t
->flags
& PF_VCPU
)
862 gtime
+= vtime_delta(t
);
864 } while (read_seqcount_retry(&t
->vtime_seqcount
, seq
));
870 * Fetch cputime raw values from fields of task_struct and
871 * add up the pending nohz execution time since the last
875 fetch_task_cputime(struct task_struct
*t
,
876 cputime_t
*u_dst
, cputime_t
*s_dst
,
877 cputime_t
*u_src
, cputime_t
*s_src
,
878 cputime_t
*udelta
, cputime_t
*sdelta
)
881 unsigned long long delta
;
887 seq
= read_seqcount_begin(&t
->vtime_seqcount
);
894 /* Task is sleeping, nothing to add */
895 if (t
->vtime_snap_whence
== VTIME_INACTIVE
||
899 delta
= vtime_delta(t
);
902 * Task runs either in user or kernel space, add pending nohz time to
905 if (t
->vtime_snap_whence
== VTIME_USER
|| t
->flags
& PF_VCPU
) {
908 if (t
->vtime_snap_whence
== VTIME_SYS
)
911 } while (read_seqcount_retry(&t
->vtime_seqcount
, seq
));
915 void task_cputime(struct task_struct
*t
, cputime_t
*utime
, cputime_t
*stime
)
917 cputime_t udelta
, sdelta
;
919 if (!vtime_accounting_enabled()) {
927 fetch_task_cputime(t
, utime
, stime
, &t
->utime
,
928 &t
->stime
, &udelta
, &sdelta
);
935 void task_cputime_scaled(struct task_struct
*t
,
936 cputime_t
*utimescaled
, cputime_t
*stimescaled
)
938 cputime_t udelta
, sdelta
;
940 if (!vtime_accounting_enabled()) {
942 *utimescaled
= t
->utimescaled
;
944 *stimescaled
= t
->stimescaled
;
948 fetch_task_cputime(t
, utimescaled
, stimescaled
,
949 &t
->utimescaled
, &t
->stimescaled
, &udelta
, &sdelta
);
951 *utimescaled
+= cputime_to_scaled(udelta
);
953 *stimescaled
+= cputime_to_scaled(sdelta
);
955 #endif /* CONFIG_VIRT_CPU_ACCOUNTING_GEN */