1 // SPDX-License-Identifier: GPL-2.0-only
3 * Simple CPU accounting cgroup controller
7 #ifdef CONFIG_IRQ_TIME_ACCOUNTING
10 * There are no locks covering percpu hardirq/softirq time.
11 * They are only modified in vtime_account, on corresponding CPU
12 * with interrupts disabled. So, writes are safe.
13 * They are read and saved off onto struct rq in update_rq_clock().
14 * This may result in other CPU reading this CPU's irq time and can
15 * race with irq/vtime_account on this CPU. We would either get old
16 * or new value with a side effect of accounting a slice of irq time to wrong
17 * task when irq is in progress while we read rq->clock. That is a worthy
18 * compromise in place of having locks on each irq in account_system_time.
20 DEFINE_PER_CPU(struct irqtime, cpu_irqtime);
22 static int sched_clock_irqtime;
24 void enable_sched_clock_irqtime(void)
26 sched_clock_irqtime = 1;
29 void disable_sched_clock_irqtime(void)
31 sched_clock_irqtime = 0;
34 static void irqtime_account_delta(struct irqtime *irqtime, u64 delta,
35 enum cpu_usage_stat idx)
37 u64 *cpustat = kcpustat_this_cpu->cpustat;
39 u64_stats_update_begin(&irqtime->sync);
40 cpustat[idx] += delta;
41 irqtime->total += delta;
42 irqtime->tick_delta += delta;
43 u64_stats_update_end(&irqtime->sync);
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)
52 struct irqtime *irqtime = this_cpu_ptr(&cpu_irqtime);
56 if (!sched_clock_irqtime)
59 cpu = smp_processor_id();
60 delta = sched_clock_cpu(cpu) - irqtime->irq_start_time;
61 irqtime->irq_start_time += delta;
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 irqtime_account_delta(irqtime, delta, CPUTIME_IRQ);
71 else if (in_serving_softirq() && curr != this_cpu_ksoftirqd())
72 irqtime_account_delta(irqtime, delta, CPUTIME_SOFTIRQ);
74 EXPORT_SYMBOL_GPL(irqtime_account_irq);
76 static u64 irqtime_tick_accounted(u64 maxtime)
78 struct irqtime *irqtime = this_cpu_ptr(&cpu_irqtime);
81 delta = min(irqtime->tick_delta, maxtime);
82 irqtime->tick_delta -= delta;
87 #else /* CONFIG_IRQ_TIME_ACCOUNTING */
89 #define sched_clock_irqtime (0)
91 static u64 irqtime_tick_accounted(u64 dummy)
96 #endif /* !CONFIG_IRQ_TIME_ACCOUNTING */
98 static inline void task_group_account_field(struct task_struct *p, int index,
102 * Since all updates are sure to touch the root cgroup, we
103 * get ourselves ahead and touch it first. If the root cgroup
104 * is the only cgroup, then nothing else should be necessary.
107 __this_cpu_add(kernel_cpustat.cpustat[index], tmp);
109 cgroup_account_cputime_field(p, index, tmp);
113 * Account user CPU time to a process.
114 * @p: the process that the CPU time gets accounted to
115 * @cputime: the CPU time spent in user space since the last update
117 void account_user_time(struct task_struct *p, u64 cputime)
121 /* Add user time to process. */
123 account_group_user_time(p, cputime);
125 index = (task_nice(p) > 0) ? CPUTIME_NICE : CPUTIME_USER;
127 /* Add user time to cpustat. */
128 task_group_account_field(p, index, cputime);
130 /* Account for user time used */
131 acct_account_cputime(p);
135 * Account guest CPU time to a process.
136 * @p: the process that the CPU time gets accounted to
137 * @cputime: the CPU time spent in virtual machine since the last update
139 void account_guest_time(struct task_struct *p, u64 cputime)
141 u64 *cpustat = kcpustat_this_cpu->cpustat;
143 /* Add guest time to process. */
145 account_group_user_time(p, cputime);
148 /* Add guest time to cpustat. */
149 if (task_nice(p) > 0) {
150 task_group_account_field(p, CPUTIME_NICE, cputime);
151 cpustat[CPUTIME_GUEST_NICE] += cputime;
153 task_group_account_field(p, CPUTIME_USER, cputime);
154 cpustat[CPUTIME_GUEST] += cputime;
159 * Account system CPU time to a process and desired cpustat field
160 * @p: the process that the CPU time gets accounted to
161 * @cputime: the CPU time spent in kernel space since the last update
162 * @index: pointer to cpustat field that has to be updated
164 void account_system_index_time(struct task_struct *p,
165 u64 cputime, enum cpu_usage_stat index)
167 /* Add system time to process. */
169 account_group_system_time(p, cputime);
171 /* Add system time to cpustat. */
172 task_group_account_field(p, index, cputime);
174 /* Account for system time used */
175 acct_account_cputime(p);
179 * Account system CPU time to a process.
180 * @p: the process that the CPU time gets accounted to
181 * @hardirq_offset: the offset to subtract from hardirq_count()
182 * @cputime: the CPU time spent in kernel space since the last update
184 void account_system_time(struct task_struct *p, int hardirq_offset, u64 cputime)
188 if ((p->flags & PF_VCPU) && (irq_count() - hardirq_offset == 0)) {
189 account_guest_time(p, cputime);
193 if (hardirq_count() - hardirq_offset)
195 else if (in_serving_softirq())
196 index = CPUTIME_SOFTIRQ;
198 index = CPUTIME_SYSTEM;
200 account_system_index_time(p, cputime, index);
204 * Account for involuntary wait time.
205 * @cputime: the CPU time spent in involuntary wait
207 void account_steal_time(u64 cputime)
209 u64 *cpustat = kcpustat_this_cpu->cpustat;
211 cpustat[CPUTIME_STEAL] += cputime;
215 * Account for idle time.
216 * @cputime: the CPU time spent in idle wait
218 void account_idle_time(u64 cputime)
220 u64 *cpustat = kcpustat_this_cpu->cpustat;
221 struct rq *rq = this_rq();
223 if (atomic_read(&rq->nr_iowait) > 0)
224 cpustat[CPUTIME_IOWAIT] += cputime;
226 cpustat[CPUTIME_IDLE] += cputime;
230 * When a guest is interrupted for a longer amount of time, missed clock
231 * ticks are not redelivered later. Due to that, this function may on
232 * occasion account more time than the calling functions think elapsed.
234 static __always_inline u64 steal_account_process_time(u64 maxtime)
236 #ifdef CONFIG_PARAVIRT
237 if (static_key_false(¶virt_steal_enabled)) {
240 steal = paravirt_steal_clock(smp_processor_id());
241 steal -= this_rq()->prev_steal_time;
242 steal = min(steal, maxtime);
243 account_steal_time(steal);
244 this_rq()->prev_steal_time += steal;
253 * Account how much elapsed time was spent in steal, irq, or softirq time.
255 static inline u64 account_other_time(u64 max)
259 lockdep_assert_irqs_disabled();
261 accounted = steal_account_process_time(max);
264 accounted += irqtime_tick_accounted(max - accounted);
270 static inline u64 read_sum_exec_runtime(struct task_struct *t)
272 return t->se.sum_exec_runtime;
275 static u64 read_sum_exec_runtime(struct task_struct *t)
281 rq = task_rq_lock(t, &rf);
282 ns = t->se.sum_exec_runtime;
283 task_rq_unlock(rq, t, &rf);
290 * Accumulate raw cputime values of dead tasks (sig->[us]time) and live
291 * tasks (sum on group iteration) belonging to @tsk's group.
293 void thread_group_cputime(struct task_struct *tsk, struct task_cputime *times)
295 struct signal_struct *sig = tsk->signal;
297 struct task_struct *t;
298 unsigned int seq, nextseq;
302 * Update current task runtime to account pending time since last
303 * scheduler action or thread_group_cputime() call. This thread group
304 * might have other running tasks on different CPUs, but updating
305 * their runtime can affect syscall performance, so we skip account
306 * those pending times and rely only on values updated on tick or
307 * other scheduler action.
309 if (same_thread_group(current, tsk))
310 (void) task_sched_runtime(current);
313 /* Attempt a lockless read on the first round. */
317 flags = read_seqbegin_or_lock_irqsave(&sig->stats_lock, &seq);
318 times->utime = sig->utime;
319 times->stime = sig->stime;
320 times->sum_exec_runtime = sig->sum_sched_runtime;
322 for_each_thread(tsk, t) {
323 task_cputime(t, &utime, &stime);
324 times->utime += utime;
325 times->stime += stime;
326 times->sum_exec_runtime += read_sum_exec_runtime(t);
328 /* If lockless access failed, take the lock. */
330 } while (need_seqretry(&sig->stats_lock, seq));
331 done_seqretry_irqrestore(&sig->stats_lock, seq, flags);
335 #ifdef CONFIG_IRQ_TIME_ACCOUNTING
337 * Account a tick to a process and cpustat
338 * @p: the process that the CPU time gets accounted to
339 * @user_tick: is the tick from userspace
340 * @rq: the pointer to rq
342 * Tick demultiplexing follows the order
343 * - pending hardirq update
344 * - pending softirq update
348 * - check for guest_time
349 * - else account as system_time
351 * Check for hardirq is done both for system and user time as there is
352 * no timer going off while we are on hardirq and hence we may never get an
353 * opportunity to update it solely in system time.
354 * p->stime and friends are only updated on system time and not on irq
355 * softirq as those do not count in task exec_runtime any more.
357 static void irqtime_account_process_tick(struct task_struct *p, int user_tick,
360 u64 other, cputime = TICK_NSEC * ticks;
363 * When returning from idle, many ticks can get accounted at
364 * once, including some ticks of steal, irq, and softirq time.
365 * Subtract those ticks from the amount of time accounted to
366 * idle, or potentially user or system time. Due to rounding,
367 * other time can exceed ticks occasionally.
369 other = account_other_time(ULONG_MAX);
370 if (other >= cputime)
375 if (this_cpu_ksoftirqd() == p) {
377 * ksoftirqd time do not get accounted in cpu_softirq_time.
378 * So, we have to handle it separately here.
379 * Also, p->stime needs to be updated for ksoftirqd.
381 account_system_index_time(p, cputime, CPUTIME_SOFTIRQ);
382 } else if (user_tick) {
383 account_user_time(p, cputime);
384 } else if (p == this_rq()->idle) {
385 account_idle_time(cputime);
386 } else if (p->flags & PF_VCPU) { /* System time or guest time */
387 account_guest_time(p, cputime);
389 account_system_index_time(p, cputime, CPUTIME_SYSTEM);
393 static void irqtime_account_idle_ticks(int ticks)
395 irqtime_account_process_tick(current, 0, ticks);
397 #else /* CONFIG_IRQ_TIME_ACCOUNTING */
398 static inline void irqtime_account_idle_ticks(int ticks) { }
399 static inline void irqtime_account_process_tick(struct task_struct *p, int user_tick,
401 #endif /* CONFIG_IRQ_TIME_ACCOUNTING */
404 * Use precise platform statistics if available:
406 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_NATIVE
408 # ifndef __ARCH_HAS_VTIME_TASK_SWITCH
409 void vtime_task_switch(struct task_struct *prev)
411 if (is_idle_task(prev))
412 vtime_account_idle(prev);
414 vtime_account_kernel(prev);
417 arch_vtime_task_switch(prev);
422 * Archs that account the whole time spent in the idle task
423 * (outside irq) as idle time can rely on this and just implement
424 * vtime_account_kernel() and vtime_account_idle(). Archs that
425 * have other meaning of the idle time (s390 only includes the
426 * time spent by the CPU when it's in low power mode) must override
429 #ifndef __ARCH_HAS_VTIME_ACCOUNT
430 void vtime_account_irq_enter(struct task_struct *tsk)
432 if (!in_interrupt() && is_idle_task(tsk))
433 vtime_account_idle(tsk);
435 vtime_account_kernel(tsk);
437 EXPORT_SYMBOL_GPL(vtime_account_irq_enter);
438 #endif /* __ARCH_HAS_VTIME_ACCOUNT */
440 void cputime_adjust(struct task_cputime *curr, struct prev_cputime *prev,
447 void task_cputime_adjusted(struct task_struct *p, u64 *ut, u64 *st)
452 EXPORT_SYMBOL_GPL(task_cputime_adjusted);
454 void thread_group_cputime_adjusted(struct task_struct *p, u64 *ut, u64 *st)
456 struct task_cputime cputime;
458 thread_group_cputime(p, &cputime);
464 #else /* !CONFIG_VIRT_CPU_ACCOUNTING_NATIVE: */
467 * Account a single tick of CPU time.
468 * @p: the process that the CPU time gets accounted to
469 * @user_tick: indicates if the tick is a user or a system tick
471 void account_process_tick(struct task_struct *p, int user_tick)
475 if (vtime_accounting_enabled_this_cpu())
478 if (sched_clock_irqtime) {
479 irqtime_account_process_tick(p, user_tick, 1);
484 steal = steal_account_process_time(ULONG_MAX);
486 if (steal >= cputime)
492 account_user_time(p, cputime);
493 else if ((p != this_rq()->idle) || (irq_count() != HARDIRQ_OFFSET))
494 account_system_time(p, HARDIRQ_OFFSET, cputime);
496 account_idle_time(cputime);
500 * Account multiple ticks of idle time.
501 * @ticks: number of stolen ticks
503 void account_idle_ticks(unsigned long ticks)
507 if (sched_clock_irqtime) {
508 irqtime_account_idle_ticks(ticks);
512 cputime = ticks * TICK_NSEC;
513 steal = steal_account_process_time(ULONG_MAX);
515 if (steal >= cputime)
519 account_idle_time(cputime);
523 * Adjust tick based cputime random precision against scheduler runtime
526 * Tick based cputime accounting depend on random scheduling timeslices of a
527 * task to be interrupted or not by the timer. Depending on these
528 * circumstances, the number of these interrupts may be over or
529 * under-optimistic, matching the real user and system cputime with a variable
532 * Fix this by scaling these tick based values against the total runtime
533 * accounted by the CFS scheduler.
535 * This code provides the following guarantees:
537 * stime + utime == rtime
538 * stime_i+1 >= stime_i, utime_i+1 >= utime_i
540 * Assuming that rtime_i+1 >= rtime_i.
542 void cputime_adjust(struct task_cputime *curr, struct prev_cputime *prev,
545 u64 rtime, stime, utime;
548 /* Serialize concurrent callers such that we can honour our guarantees */
549 raw_spin_lock_irqsave(&prev->lock, flags);
550 rtime = curr->sum_exec_runtime;
553 * This is possible under two circumstances:
554 * - rtime isn't monotonic after all (a bug);
555 * - we got reordered by the lock.
557 * In both cases this acts as a filter such that the rest of the code
558 * can assume it is monotonic regardless of anything else.
560 if (prev->stime + prev->utime >= rtime)
567 * If either stime or utime are 0, assume all runtime is userspace.
568 * Once a task gets some ticks, the monotonicy code at 'update:'
569 * will ensure things converge to the observed ratio.
581 stime = mul_u64_u64_div_u64(stime, rtime, stime + utime);
585 * Make sure stime doesn't go backwards; this preserves monotonicity
586 * for utime because rtime is monotonic.
588 * utime_i+1 = rtime_i+1 - stime_i
589 * = rtime_i+1 - (rtime_i - utime_i)
590 * = (rtime_i+1 - rtime_i) + utime_i
593 if (stime < prev->stime)
595 utime = rtime - stime;
598 * Make sure utime doesn't go backwards; this still preserves
599 * monotonicity for stime, analogous argument to above.
601 if (utime < prev->utime) {
603 stime = rtime - utime;
611 raw_spin_unlock_irqrestore(&prev->lock, flags);
614 void task_cputime_adjusted(struct task_struct *p, u64 *ut, u64 *st)
616 struct task_cputime cputime = {
617 .sum_exec_runtime = p->se.sum_exec_runtime,
620 task_cputime(p, &cputime.utime, &cputime.stime);
621 cputime_adjust(&cputime, &p->prev_cputime, ut, st);
623 EXPORT_SYMBOL_GPL(task_cputime_adjusted);
625 void thread_group_cputime_adjusted(struct task_struct *p, u64 *ut, u64 *st)
627 struct task_cputime cputime;
629 thread_group_cputime(p, &cputime);
630 cputime_adjust(&cputime, &p->signal->prev_cputime, ut, st);
632 #endif /* !CONFIG_VIRT_CPU_ACCOUNTING_NATIVE */
634 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
635 static u64 vtime_delta(struct vtime *vtime)
637 unsigned long long clock;
639 clock = sched_clock();
640 if (clock < vtime->starttime)
643 return clock - vtime->starttime;
646 static u64 get_vtime_delta(struct vtime *vtime)
648 u64 delta = vtime_delta(vtime);
652 * Unlike tick based timing, vtime based timing never has lost
653 * ticks, and no need for steal time accounting to make up for
654 * lost ticks. Vtime accounts a rounded version of actual
655 * elapsed time. Limit account_other_time to prevent rounding
656 * errors from causing elapsed vtime to go negative.
658 other = account_other_time(delta);
659 WARN_ON_ONCE(vtime->state == VTIME_INACTIVE);
660 vtime->starttime += delta;
662 return delta - other;
665 static void vtime_account_system(struct task_struct *tsk,
668 vtime->stime += get_vtime_delta(vtime);
669 if (vtime->stime >= TICK_NSEC) {
670 account_system_time(tsk, irq_count(), vtime->stime);
675 static void vtime_account_guest(struct task_struct *tsk,
678 vtime->gtime += get_vtime_delta(vtime);
679 if (vtime->gtime >= TICK_NSEC) {
680 account_guest_time(tsk, vtime->gtime);
685 static void __vtime_account_kernel(struct task_struct *tsk,
688 /* We might have scheduled out from guest path */
689 if (vtime->state == VTIME_GUEST)
690 vtime_account_guest(tsk, vtime);
692 vtime_account_system(tsk, vtime);
695 void vtime_account_kernel(struct task_struct *tsk)
697 struct vtime *vtime = &tsk->vtime;
699 if (!vtime_delta(vtime))
702 write_seqcount_begin(&vtime->seqcount);
703 __vtime_account_kernel(tsk, vtime);
704 write_seqcount_end(&vtime->seqcount);
707 void vtime_user_enter(struct task_struct *tsk)
709 struct vtime *vtime = &tsk->vtime;
711 write_seqcount_begin(&vtime->seqcount);
712 vtime_account_system(tsk, vtime);
713 vtime->state = VTIME_USER;
714 write_seqcount_end(&vtime->seqcount);
717 void vtime_user_exit(struct task_struct *tsk)
719 struct vtime *vtime = &tsk->vtime;
721 write_seqcount_begin(&vtime->seqcount);
722 vtime->utime += get_vtime_delta(vtime);
723 if (vtime->utime >= TICK_NSEC) {
724 account_user_time(tsk, vtime->utime);
727 vtime->state = VTIME_SYS;
728 write_seqcount_end(&vtime->seqcount);
731 void vtime_guest_enter(struct task_struct *tsk)
733 struct vtime *vtime = &tsk->vtime;
735 * The flags must be updated under the lock with
736 * the vtime_starttime flush and update.
737 * That enforces a right ordering and update sequence
738 * synchronization against the reader (task_gtime())
739 * that can thus safely catch up with a tickless delta.
741 write_seqcount_begin(&vtime->seqcount);
742 vtime_account_system(tsk, vtime);
743 tsk->flags |= PF_VCPU;
744 vtime->state = VTIME_GUEST;
745 write_seqcount_end(&vtime->seqcount);
747 EXPORT_SYMBOL_GPL(vtime_guest_enter);
749 void vtime_guest_exit(struct task_struct *tsk)
751 struct vtime *vtime = &tsk->vtime;
753 write_seqcount_begin(&vtime->seqcount);
754 vtime_account_guest(tsk, vtime);
755 tsk->flags &= ~PF_VCPU;
756 vtime->state = VTIME_SYS;
757 write_seqcount_end(&vtime->seqcount);
759 EXPORT_SYMBOL_GPL(vtime_guest_exit);
761 void vtime_account_idle(struct task_struct *tsk)
763 account_idle_time(get_vtime_delta(&tsk->vtime));
766 void vtime_task_switch_generic(struct task_struct *prev)
768 struct vtime *vtime = &prev->vtime;
770 write_seqcount_begin(&vtime->seqcount);
771 if (vtime->state == VTIME_IDLE)
772 vtime_account_idle(prev);
774 __vtime_account_kernel(prev, vtime);
775 vtime->state = VTIME_INACTIVE;
777 write_seqcount_end(&vtime->seqcount);
779 vtime = ¤t->vtime;
781 write_seqcount_begin(&vtime->seqcount);
782 if (is_idle_task(current))
783 vtime->state = VTIME_IDLE;
784 else if (current->flags & PF_VCPU)
785 vtime->state = VTIME_GUEST;
787 vtime->state = VTIME_SYS;
788 vtime->starttime = sched_clock();
789 vtime->cpu = smp_processor_id();
790 write_seqcount_end(&vtime->seqcount);
793 void vtime_init_idle(struct task_struct *t, int cpu)
795 struct vtime *vtime = &t->vtime;
798 local_irq_save(flags);
799 write_seqcount_begin(&vtime->seqcount);
800 vtime->state = VTIME_IDLE;
801 vtime->starttime = sched_clock();
803 write_seqcount_end(&vtime->seqcount);
804 local_irq_restore(flags);
807 u64 task_gtime(struct task_struct *t)
809 struct vtime *vtime = &t->vtime;
813 if (!vtime_accounting_enabled())
817 seq = read_seqcount_begin(&vtime->seqcount);
820 if (vtime->state == VTIME_GUEST)
821 gtime += vtime->gtime + vtime_delta(vtime);
823 } while (read_seqcount_retry(&vtime->seqcount, seq));
829 * Fetch cputime raw values from fields of task_struct and
830 * add up the pending nohz execution time since the last
833 void task_cputime(struct task_struct *t, u64 *utime, u64 *stime)
835 struct vtime *vtime = &t->vtime;
839 if (!vtime_accounting_enabled()) {
846 seq = read_seqcount_begin(&vtime->seqcount);
851 /* Task is sleeping or idle, nothing to add */
852 if (vtime->state < VTIME_SYS)
855 delta = vtime_delta(vtime);
858 * Task runs either in user (including guest) or kernel space,
859 * add pending nohz time to the right place.
861 if (vtime->state == VTIME_SYS)
862 *stime += vtime->stime + delta;
864 *utime += vtime->utime + delta;
865 } while (read_seqcount_retry(&vtime->seqcount, seq));
868 static int vtime_state_fetch(struct vtime *vtime, int cpu)
870 int state = READ_ONCE(vtime->state);
873 * We raced against a context switch, fetch the
874 * kcpustat task again.
876 if (vtime->cpu != cpu && vtime->cpu != -1)
880 * Two possible things here:
881 * 1) We are seeing the scheduling out task (prev) or any past one.
882 * 2) We are seeing the scheduling in task (next) but it hasn't
883 * passed though vtime_task_switch() yet so the pending
884 * cputime of the prev task may not be flushed yet.
886 * Case 1) is ok but 2) is not. So wait for a safe VTIME state.
888 if (state == VTIME_INACTIVE)
894 static u64 kcpustat_user_vtime(struct vtime *vtime)
896 if (vtime->state == VTIME_USER)
897 return vtime->utime + vtime_delta(vtime);
898 else if (vtime->state == VTIME_GUEST)
899 return vtime->gtime + vtime_delta(vtime);
903 static int kcpustat_field_vtime(u64 *cpustat,
904 struct task_struct *tsk,
905 enum cpu_usage_stat usage,
908 struct vtime *vtime = &tsk->vtime;
914 seq = read_seqcount_begin(&vtime->seqcount);
916 state = vtime_state_fetch(vtime, cpu);
920 *val = cpustat[usage];
923 * Nice VS unnice cputime accounting may be inaccurate if
924 * the nice value has changed since the last vtime update.
925 * But proper fix would involve interrupting target on nice
926 * updates which is a no go on nohz_full (although the scheduler
927 * may still interrupt the target if rescheduling is needed...)
931 if (state == VTIME_SYS)
932 *val += vtime->stime + vtime_delta(vtime);
935 if (task_nice(tsk) <= 0)
936 *val += kcpustat_user_vtime(vtime);
939 if (task_nice(tsk) > 0)
940 *val += kcpustat_user_vtime(vtime);
943 if (state == VTIME_GUEST && task_nice(tsk) <= 0)
944 *val += vtime->gtime + vtime_delta(vtime);
946 case CPUTIME_GUEST_NICE:
947 if (state == VTIME_GUEST && task_nice(tsk) > 0)
948 *val += vtime->gtime + vtime_delta(vtime);
953 } while (read_seqcount_retry(&vtime->seqcount, seq));
958 u64 kcpustat_field(struct kernel_cpustat *kcpustat,
959 enum cpu_usage_stat usage, int cpu)
961 u64 *cpustat = kcpustat->cpustat;
962 u64 val = cpustat[usage];
966 if (!vtime_accounting_enabled_cpu(cpu))
972 struct task_struct *curr;
975 curr = rcu_dereference(rq->curr);
976 if (WARN_ON_ONCE(!curr)) {
978 return cpustat[usage];
981 err = kcpustat_field_vtime(cpustat, curr, usage, cpu, &val);
990 EXPORT_SYMBOL_GPL(kcpustat_field);
992 static int kcpustat_cpu_fetch_vtime(struct kernel_cpustat *dst,
993 const struct kernel_cpustat *src,
994 struct task_struct *tsk, int cpu)
996 struct vtime *vtime = &tsk->vtime;
1004 seq = read_seqcount_begin(&vtime->seqcount);
1006 state = vtime_state_fetch(vtime, cpu);
1011 cpustat = dst->cpustat;
1013 /* Task is sleeping, dead or idle, nothing to add */
1014 if (state < VTIME_SYS)
1017 delta = vtime_delta(vtime);
1020 * Task runs either in user (including guest) or kernel space,
1021 * add pending nohz time to the right place.
1023 if (state == VTIME_SYS) {
1024 cpustat[CPUTIME_SYSTEM] += vtime->stime + delta;
1025 } else if (state == VTIME_USER) {
1026 if (task_nice(tsk) > 0)
1027 cpustat[CPUTIME_NICE] += vtime->utime + delta;
1029 cpustat[CPUTIME_USER] += vtime->utime + delta;
1031 WARN_ON_ONCE(state != VTIME_GUEST);
1032 if (task_nice(tsk) > 0) {
1033 cpustat[CPUTIME_GUEST_NICE] += vtime->gtime + delta;
1034 cpustat[CPUTIME_NICE] += vtime->gtime + delta;
1036 cpustat[CPUTIME_GUEST] += vtime->gtime + delta;
1037 cpustat[CPUTIME_USER] += vtime->gtime + delta;
1040 } while (read_seqcount_retry(&vtime->seqcount, seq));
1045 void kcpustat_cpu_fetch(struct kernel_cpustat *dst, int cpu)
1047 const struct kernel_cpustat *src = &kcpustat_cpu(cpu);
1051 if (!vtime_accounting_enabled_cpu(cpu)) {
1059 struct task_struct *curr;
1062 curr = rcu_dereference(rq->curr);
1063 if (WARN_ON_ONCE(!curr)) {
1069 err = kcpustat_cpu_fetch_vtime(dst, src, curr, cpu);
1078 EXPORT_SYMBOL_GPL(kcpustat_cpu_fetch);
1080 #endif /* CONFIG_VIRT_CPU_ACCOUNTING_GEN */