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 after 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, unsigned int offset)
52 struct irqtime *irqtime = this_cpu_ptr(&cpu_irqtime);
57 if (!sched_clock_irqtime)
60 cpu = smp_processor_id();
61 delta = sched_clock_cpu(cpu) - irqtime->irq_start_time;
62 irqtime->irq_start_time += delta;
63 pc = irq_count() - offset;
66 * We do not account for softirq time from ksoftirqd here.
67 * We want to continue accounting softirq time to ksoftirqd thread
68 * in that case, so as not to confuse scheduler with a special task
69 * that do not consume any time, but still wants to run.
71 if (pc & HARDIRQ_MASK)
72 irqtime_account_delta(irqtime, delta, CPUTIME_IRQ);
73 else if ((pc & SOFTIRQ_OFFSET) && curr != this_cpu_ksoftirqd())
74 irqtime_account_delta(irqtime, delta, CPUTIME_SOFTIRQ);
77 static u64 irqtime_tick_accounted(u64 maxtime)
79 struct irqtime *irqtime = this_cpu_ptr(&cpu_irqtime);
82 delta = min(irqtime->tick_delta, maxtime);
83 irqtime->tick_delta -= delta;
88 #else /* CONFIG_IRQ_TIME_ACCOUNTING */
90 #define sched_clock_irqtime (0)
92 static u64 irqtime_tick_accounted(u64 dummy)
97 #endif /* !CONFIG_IRQ_TIME_ACCOUNTING */
99 static inline void task_group_account_field(struct task_struct *p, int index,
103 * Since all updates are sure to touch the root cgroup, we
104 * get ourselves ahead and touch it first. If the root cgroup
105 * is the only cgroup, then nothing else should be necessary.
108 __this_cpu_add(kernel_cpustat.cpustat[index], tmp);
110 cgroup_account_cputime_field(p, index, tmp);
114 * Account user CPU time to a process.
115 * @p: the process that the CPU time gets accounted to
116 * @cputime: the CPU time spent in user space since the last update
118 void account_user_time(struct task_struct *p, u64 cputime)
122 /* Add user time to process. */
124 account_group_user_time(p, cputime);
126 index = (task_nice(p) > 0) ? CPUTIME_NICE : CPUTIME_USER;
128 /* Add user time to cpustat. */
129 task_group_account_field(p, index, cputime);
131 /* Account for user time used */
132 acct_account_cputime(p);
136 * Account guest CPU time to a process.
137 * @p: the process that the CPU time gets accounted to
138 * @cputime: the CPU time spent in virtual machine since the last update
140 void account_guest_time(struct task_struct *p, u64 cputime)
142 u64 *cpustat = kcpustat_this_cpu->cpustat;
144 /* Add guest time to process. */
146 account_group_user_time(p, cputime);
149 /* Add guest time to cpustat. */
150 if (task_nice(p) > 0) {
151 cpustat[CPUTIME_NICE] += cputime;
152 cpustat[CPUTIME_GUEST_NICE] += cputime;
154 cpustat[CPUTIME_USER] += cputime;
155 cpustat[CPUTIME_GUEST] += cputime;
160 * Account system CPU time to a process and desired cpustat field
161 * @p: the process that the CPU time gets accounted to
162 * @cputime: the CPU time spent in kernel space since the last update
163 * @index: pointer to cpustat field that has to be updated
165 void account_system_index_time(struct task_struct *p,
166 u64 cputime, enum cpu_usage_stat index)
168 /* Add system time to process. */
170 account_group_system_time(p, cputime);
172 /* Add system time to cpustat. */
173 task_group_account_field(p, index, cputime);
175 /* Account for system time used */
176 acct_account_cputime(p);
180 * Account system CPU time to a process.
181 * @p: the process that the CPU time gets accounted to
182 * @hardirq_offset: the offset to subtract from hardirq_count()
183 * @cputime: the CPU time spent in kernel space since the last update
185 void account_system_time(struct task_struct *p, int hardirq_offset, u64 cputime)
189 if ((p->flags & PF_VCPU) && (irq_count() - hardirq_offset == 0)) {
190 account_guest_time(p, cputime);
194 if (hardirq_count() - hardirq_offset)
196 else if (in_serving_softirq())
197 index = CPUTIME_SOFTIRQ;
199 index = CPUTIME_SYSTEM;
201 account_system_index_time(p, cputime, index);
205 * Account for involuntary wait time.
206 * @cputime: the CPU time spent in involuntary wait
208 void account_steal_time(u64 cputime)
210 u64 *cpustat = kcpustat_this_cpu->cpustat;
212 cpustat[CPUTIME_STEAL] += cputime;
216 * Account for idle time.
217 * @cputime: the CPU time spent in idle wait
219 void account_idle_time(u64 cputime)
221 u64 *cpustat = kcpustat_this_cpu->cpustat;
222 struct rq *rq = this_rq();
224 if (atomic_read(&rq->nr_iowait) > 0)
225 cpustat[CPUTIME_IOWAIT] += cputime;
227 cpustat[CPUTIME_IDLE] += cputime;
231 * When a guest is interrupted for a longer amount of time, missed clock
232 * ticks are not redelivered later. Due to that, this function may on
233 * occasion account more time than the calling functions think elapsed.
235 static __always_inline u64 steal_account_process_time(u64 maxtime)
237 #ifdef CONFIG_PARAVIRT
238 if (static_key_false(¶virt_steal_enabled)) {
241 steal = paravirt_steal_clock(smp_processor_id());
242 steal -= this_rq()->prev_steal_time;
243 steal = min(steal, maxtime);
244 account_steal_time(steal);
245 this_rq()->prev_steal_time += steal;
254 * Account how much elapsed time was spent in steal, irq, or softirq time.
256 static inline u64 account_other_time(u64 max)
260 lockdep_assert_irqs_disabled();
262 accounted = steal_account_process_time(max);
265 accounted += irqtime_tick_accounted(max - accounted);
271 static inline u64 read_sum_exec_runtime(struct task_struct *t)
273 return t->se.sum_exec_runtime;
276 static u64 read_sum_exec_runtime(struct task_struct *t)
282 rq = task_rq_lock(t, &rf);
283 ns = t->se.sum_exec_runtime;
284 task_rq_unlock(rq, t, &rf);
291 * Accumulate raw cputime values of dead tasks (sig->[us]time) and live
292 * tasks (sum on group iteration) belonging to @tsk's group.
294 void thread_group_cputime(struct task_struct *tsk, struct task_cputime *times)
296 struct signal_struct *sig = tsk->signal;
298 struct task_struct *t;
299 unsigned int seq, nextseq;
303 * Update current task runtime to account pending time since last
304 * scheduler action or thread_group_cputime() call. This thread group
305 * might have other running tasks on different CPUs, but updating
306 * their runtime can affect syscall performance, so we skip account
307 * those pending times and rely only on values updated on tick or
308 * other scheduler action.
310 if (same_thread_group(current, tsk))
311 (void) task_sched_runtime(current);
314 /* Attempt a lockless read on the first round. */
318 flags = read_seqbegin_or_lock_irqsave(&sig->stats_lock, &seq);
319 times->utime = sig->utime;
320 times->stime = sig->stime;
321 times->sum_exec_runtime = sig->sum_sched_runtime;
323 for_each_thread(tsk, t) {
324 task_cputime(t, &utime, &stime);
325 times->utime += utime;
326 times->stime += stime;
327 times->sum_exec_runtime += read_sum_exec_runtime(t);
329 /* If lockless access failed, take the lock. */
331 } while (need_seqretry(&sig->stats_lock, seq));
332 done_seqretry_irqrestore(&sig->stats_lock, seq, flags);
336 #ifdef CONFIG_IRQ_TIME_ACCOUNTING
338 * Account a tick to a process and cpustat
339 * @p: the process that the CPU time gets accounted to
340 * @user_tick: is the tick from userspace
341 * @rq: the pointer to rq
343 * Tick demultiplexing follows the order
344 * - pending hardirq update
345 * - pending softirq update
349 * - check for guest_time
350 * - else account as system_time
352 * Check for hardirq is done both for system and user time as there is
353 * no timer going off while we are on hardirq and hence we may never get an
354 * opportunity to update it solely in system time.
355 * p->stime and friends are only updated on system time and not on irq
356 * softirq as those do not count in task exec_runtime any more.
358 static void irqtime_account_process_tick(struct task_struct *p, int user_tick,
361 u64 other, cputime = TICK_NSEC * ticks;
364 * When returning from idle, many ticks can get accounted at
365 * once, including some ticks of steal, irq, and softirq time.
366 * Subtract those ticks from the amount of time accounted to
367 * idle, or potentially user or system time. Due to rounding,
368 * other time can exceed ticks occasionally.
370 other = account_other_time(ULONG_MAX);
371 if (other >= cputime)
376 if (this_cpu_ksoftirqd() == p) {
378 * ksoftirqd time do not get accounted in cpu_softirq_time.
379 * So, we have to handle it separately here.
380 * Also, p->stime needs to be updated for ksoftirqd.
382 account_system_index_time(p, cputime, CPUTIME_SOFTIRQ);
383 } else if (user_tick) {
384 account_user_time(p, cputime);
385 } else if (p == this_rq()->idle) {
386 account_idle_time(cputime);
387 } else if (p->flags & PF_VCPU) { /* System time or guest time */
388 account_guest_time(p, cputime);
390 account_system_index_time(p, cputime, CPUTIME_SYSTEM);
394 static void irqtime_account_idle_ticks(int ticks)
396 irqtime_account_process_tick(current, 0, ticks);
398 #else /* CONFIG_IRQ_TIME_ACCOUNTING */
399 static inline void irqtime_account_idle_ticks(int ticks) { }
400 static inline void irqtime_account_process_tick(struct task_struct *p, int user_tick,
402 #endif /* CONFIG_IRQ_TIME_ACCOUNTING */
405 * Use precise platform statistics if available:
407 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_NATIVE
409 # ifndef __ARCH_HAS_VTIME_TASK_SWITCH
410 void vtime_task_switch(struct task_struct *prev)
412 if (is_idle_task(prev))
413 vtime_account_idle(prev);
415 vtime_account_kernel(prev);
418 arch_vtime_task_switch(prev);
422 void vtime_account_irq(struct task_struct *tsk, unsigned int offset)
424 unsigned int pc = irq_count() - offset;
426 if (pc & HARDIRQ_OFFSET) {
427 vtime_account_hardirq(tsk);
428 } else if (pc & SOFTIRQ_OFFSET) {
429 vtime_account_softirq(tsk);
430 } else if (!IS_ENABLED(CONFIG_HAVE_VIRT_CPU_ACCOUNTING_IDLE) &&
432 vtime_account_idle(tsk);
434 vtime_account_kernel(tsk);
438 void cputime_adjust(struct task_cputime *curr, struct prev_cputime *prev,
445 void task_cputime_adjusted(struct task_struct *p, u64 *ut, u64 *st)
450 EXPORT_SYMBOL_GPL(task_cputime_adjusted);
452 void thread_group_cputime_adjusted(struct task_struct *p, u64 *ut, u64 *st)
454 struct task_cputime cputime;
456 thread_group_cputime(p, &cputime);
462 #else /* !CONFIG_VIRT_CPU_ACCOUNTING_NATIVE: */
465 * Account a single tick of CPU time.
466 * @p: the process that the CPU time gets accounted to
467 * @user_tick: indicates if the tick is a user or a system tick
469 void account_process_tick(struct task_struct *p, int user_tick)
473 if (vtime_accounting_enabled_this_cpu())
476 if (sched_clock_irqtime) {
477 irqtime_account_process_tick(p, user_tick, 1);
482 steal = steal_account_process_time(ULONG_MAX);
484 if (steal >= cputime)
490 account_user_time(p, cputime);
491 else if ((p != this_rq()->idle) || (irq_count() != HARDIRQ_OFFSET))
492 account_system_time(p, HARDIRQ_OFFSET, cputime);
494 account_idle_time(cputime);
498 * Account multiple ticks of idle time.
499 * @ticks: number of stolen ticks
501 void account_idle_ticks(unsigned long ticks)
505 if (sched_clock_irqtime) {
506 irqtime_account_idle_ticks(ticks);
510 cputime = ticks * TICK_NSEC;
511 steal = steal_account_process_time(ULONG_MAX);
513 if (steal >= cputime)
517 account_idle_time(cputime);
521 * Adjust tick based cputime random precision against scheduler runtime
524 * Tick based cputime accounting depend on random scheduling timeslices of a
525 * task to be interrupted or not by the timer. Depending on these
526 * circumstances, the number of these interrupts may be over or
527 * under-optimistic, matching the real user and system cputime with a variable
530 * Fix this by scaling these tick based values against the total runtime
531 * accounted by the CFS scheduler.
533 * This code provides the following guarantees:
535 * stime + utime == rtime
536 * stime_i+1 >= stime_i, utime_i+1 >= utime_i
538 * Assuming that rtime_i+1 >= rtime_i.
540 void cputime_adjust(struct task_cputime *curr, struct prev_cputime *prev,
543 u64 rtime, stime, utime;
546 /* Serialize concurrent callers such that we can honour our guarantees */
547 raw_spin_lock_irqsave(&prev->lock, flags);
548 rtime = curr->sum_exec_runtime;
551 * This is possible under two circumstances:
552 * - rtime isn't monotonic after all (a bug);
553 * - we got reordered by the lock.
555 * In both cases this acts as a filter such that the rest of the code
556 * can assume it is monotonic regardless of anything else.
558 if (prev->stime + prev->utime >= rtime)
565 * If either stime or utime are 0, assume all runtime is userspace.
566 * Once a task gets some ticks, the monotonicity code at 'update:'
567 * will ensure things converge to the observed ratio.
579 stime = mul_u64_u64_div_u64(stime, rtime, stime + utime);
583 * Make sure stime doesn't go backwards; this preserves monotonicity
584 * for utime because rtime is monotonic.
586 * utime_i+1 = rtime_i+1 - stime_i
587 * = rtime_i+1 - (rtime_i - utime_i)
588 * = (rtime_i+1 - rtime_i) + utime_i
591 if (stime < prev->stime)
593 utime = rtime - stime;
596 * Make sure utime doesn't go backwards; this still preserves
597 * monotonicity for stime, analogous argument to above.
599 if (utime < prev->utime) {
601 stime = rtime - utime;
609 raw_spin_unlock_irqrestore(&prev->lock, flags);
612 void task_cputime_adjusted(struct task_struct *p, u64 *ut, u64 *st)
614 struct task_cputime cputime = {
615 .sum_exec_runtime = p->se.sum_exec_runtime,
618 task_cputime(p, &cputime.utime, &cputime.stime);
619 cputime_adjust(&cputime, &p->prev_cputime, ut, st);
621 EXPORT_SYMBOL_GPL(task_cputime_adjusted);
623 void thread_group_cputime_adjusted(struct task_struct *p, u64 *ut, u64 *st)
625 struct task_cputime cputime;
627 thread_group_cputime(p, &cputime);
628 cputime_adjust(&cputime, &p->signal->prev_cputime, ut, st);
630 #endif /* !CONFIG_VIRT_CPU_ACCOUNTING_NATIVE */
632 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
633 static u64 vtime_delta(struct vtime *vtime)
635 unsigned long long clock;
637 clock = sched_clock();
638 if (clock < vtime->starttime)
641 return clock - vtime->starttime;
644 static u64 get_vtime_delta(struct vtime *vtime)
646 u64 delta = vtime_delta(vtime);
650 * Unlike tick based timing, vtime based timing never has lost
651 * ticks, and no need for steal time accounting to make up for
652 * lost ticks. Vtime accounts a rounded version of actual
653 * elapsed time. Limit account_other_time to prevent rounding
654 * errors from causing elapsed vtime to go negative.
656 other = account_other_time(delta);
657 WARN_ON_ONCE(vtime->state == VTIME_INACTIVE);
658 vtime->starttime += delta;
660 return delta - other;
663 static void vtime_account_system(struct task_struct *tsk,
666 vtime->stime += get_vtime_delta(vtime);
667 if (vtime->stime >= TICK_NSEC) {
668 account_system_time(tsk, irq_count(), vtime->stime);
673 static void vtime_account_guest(struct task_struct *tsk,
676 vtime->gtime += get_vtime_delta(vtime);
677 if (vtime->gtime >= TICK_NSEC) {
678 account_guest_time(tsk, vtime->gtime);
683 static void __vtime_account_kernel(struct task_struct *tsk,
686 /* We might have scheduled out from guest path */
687 if (vtime->state == VTIME_GUEST)
688 vtime_account_guest(tsk, vtime);
690 vtime_account_system(tsk, vtime);
693 void vtime_account_kernel(struct task_struct *tsk)
695 struct vtime *vtime = &tsk->vtime;
697 if (!vtime_delta(vtime))
700 write_seqcount_begin(&vtime->seqcount);
701 __vtime_account_kernel(tsk, vtime);
702 write_seqcount_end(&vtime->seqcount);
705 void vtime_user_enter(struct task_struct *tsk)
707 struct vtime *vtime = &tsk->vtime;
709 write_seqcount_begin(&vtime->seqcount);
710 vtime_account_system(tsk, vtime);
711 vtime->state = VTIME_USER;
712 write_seqcount_end(&vtime->seqcount);
715 void vtime_user_exit(struct task_struct *tsk)
717 struct vtime *vtime = &tsk->vtime;
719 write_seqcount_begin(&vtime->seqcount);
720 vtime->utime += get_vtime_delta(vtime);
721 if (vtime->utime >= TICK_NSEC) {
722 account_user_time(tsk, vtime->utime);
725 vtime->state = VTIME_SYS;
726 write_seqcount_end(&vtime->seqcount);
729 void vtime_guest_enter(struct task_struct *tsk)
731 struct vtime *vtime = &tsk->vtime;
733 * The flags must be updated under the lock with
734 * the vtime_starttime flush and update.
735 * That enforces a right ordering and update sequence
736 * synchronization against the reader (task_gtime())
737 * that can thus safely catch up with a tickless delta.
739 write_seqcount_begin(&vtime->seqcount);
740 vtime_account_system(tsk, vtime);
741 tsk->flags |= PF_VCPU;
742 vtime->state = VTIME_GUEST;
743 write_seqcount_end(&vtime->seqcount);
745 EXPORT_SYMBOL_GPL(vtime_guest_enter);
747 void vtime_guest_exit(struct task_struct *tsk)
749 struct vtime *vtime = &tsk->vtime;
751 write_seqcount_begin(&vtime->seqcount);
752 vtime_account_guest(tsk, vtime);
753 tsk->flags &= ~PF_VCPU;
754 vtime->state = VTIME_SYS;
755 write_seqcount_end(&vtime->seqcount);
757 EXPORT_SYMBOL_GPL(vtime_guest_exit);
759 void vtime_account_idle(struct task_struct *tsk)
761 account_idle_time(get_vtime_delta(&tsk->vtime));
764 void vtime_task_switch_generic(struct task_struct *prev)
766 struct vtime *vtime = &prev->vtime;
768 write_seqcount_begin(&vtime->seqcount);
769 if (vtime->state == VTIME_IDLE)
770 vtime_account_idle(prev);
772 __vtime_account_kernel(prev, vtime);
773 vtime->state = VTIME_INACTIVE;
775 write_seqcount_end(&vtime->seqcount);
777 vtime = ¤t->vtime;
779 write_seqcount_begin(&vtime->seqcount);
780 if (is_idle_task(current))
781 vtime->state = VTIME_IDLE;
782 else if (current->flags & PF_VCPU)
783 vtime->state = VTIME_GUEST;
785 vtime->state = VTIME_SYS;
786 vtime->starttime = sched_clock();
787 vtime->cpu = smp_processor_id();
788 write_seqcount_end(&vtime->seqcount);
791 void vtime_init_idle(struct task_struct *t, int cpu)
793 struct vtime *vtime = &t->vtime;
796 local_irq_save(flags);
797 write_seqcount_begin(&vtime->seqcount);
798 vtime->state = VTIME_IDLE;
799 vtime->starttime = sched_clock();
801 write_seqcount_end(&vtime->seqcount);
802 local_irq_restore(flags);
805 u64 task_gtime(struct task_struct *t)
807 struct vtime *vtime = &t->vtime;
811 if (!vtime_accounting_enabled())
815 seq = read_seqcount_begin(&vtime->seqcount);
818 if (vtime->state == VTIME_GUEST)
819 gtime += vtime->gtime + vtime_delta(vtime);
821 } while (read_seqcount_retry(&vtime->seqcount, seq));
827 * Fetch cputime raw values from fields of task_struct and
828 * add up the pending nohz execution time since the last
831 void task_cputime(struct task_struct *t, u64 *utime, u64 *stime)
833 struct vtime *vtime = &t->vtime;
837 if (!vtime_accounting_enabled()) {
844 seq = read_seqcount_begin(&vtime->seqcount);
849 /* Task is sleeping or idle, nothing to add */
850 if (vtime->state < VTIME_SYS)
853 delta = vtime_delta(vtime);
856 * Task runs either in user (including guest) or kernel space,
857 * add pending nohz time to the right place.
859 if (vtime->state == VTIME_SYS)
860 *stime += vtime->stime + delta;
862 *utime += vtime->utime + delta;
863 } while (read_seqcount_retry(&vtime->seqcount, seq));
866 static int vtime_state_fetch(struct vtime *vtime, int cpu)
868 int state = READ_ONCE(vtime->state);
871 * We raced against a context switch, fetch the
872 * kcpustat task again.
874 if (vtime->cpu != cpu && vtime->cpu != -1)
878 * Two possible things here:
879 * 1) We are seeing the scheduling out task (prev) or any past one.
880 * 2) We are seeing the scheduling in task (next) but it hasn't
881 * passed though vtime_task_switch() yet so the pending
882 * cputime of the prev task may not be flushed yet.
884 * Case 1) is ok but 2) is not. So wait for a safe VTIME state.
886 if (state == VTIME_INACTIVE)
892 static u64 kcpustat_user_vtime(struct vtime *vtime)
894 if (vtime->state == VTIME_USER)
895 return vtime->utime + vtime_delta(vtime);
896 else if (vtime->state == VTIME_GUEST)
897 return vtime->gtime + vtime_delta(vtime);
901 static int kcpustat_field_vtime(u64 *cpustat,
902 struct task_struct *tsk,
903 enum cpu_usage_stat usage,
906 struct vtime *vtime = &tsk->vtime;
912 seq = read_seqcount_begin(&vtime->seqcount);
914 state = vtime_state_fetch(vtime, cpu);
918 *val = cpustat[usage];
921 * Nice VS unnice cputime accounting may be inaccurate if
922 * the nice value has changed since the last vtime update.
923 * But proper fix would involve interrupting target on nice
924 * updates which is a no go on nohz_full (although the scheduler
925 * may still interrupt the target if rescheduling is needed...)
929 if (state == VTIME_SYS)
930 *val += vtime->stime + vtime_delta(vtime);
933 if (task_nice(tsk) <= 0)
934 *val += kcpustat_user_vtime(vtime);
937 if (task_nice(tsk) > 0)
938 *val += kcpustat_user_vtime(vtime);
941 if (state == VTIME_GUEST && task_nice(tsk) <= 0)
942 *val += vtime->gtime + vtime_delta(vtime);
944 case CPUTIME_GUEST_NICE:
945 if (state == VTIME_GUEST && task_nice(tsk) > 0)
946 *val += vtime->gtime + vtime_delta(vtime);
951 } while (read_seqcount_retry(&vtime->seqcount, seq));
956 u64 kcpustat_field(struct kernel_cpustat *kcpustat,
957 enum cpu_usage_stat usage, int cpu)
959 u64 *cpustat = kcpustat->cpustat;
960 u64 val = cpustat[usage];
964 if (!vtime_accounting_enabled_cpu(cpu))
970 struct task_struct *curr;
973 curr = rcu_dereference(rq->curr);
974 if (WARN_ON_ONCE(!curr)) {
976 return cpustat[usage];
979 err = kcpustat_field_vtime(cpustat, curr, usage, cpu, &val);
988 EXPORT_SYMBOL_GPL(kcpustat_field);
990 static int kcpustat_cpu_fetch_vtime(struct kernel_cpustat *dst,
991 const struct kernel_cpustat *src,
992 struct task_struct *tsk, int cpu)
994 struct vtime *vtime = &tsk->vtime;
1002 seq = read_seqcount_begin(&vtime->seqcount);
1004 state = vtime_state_fetch(vtime, cpu);
1009 cpustat = dst->cpustat;
1011 /* Task is sleeping, dead or idle, nothing to add */
1012 if (state < VTIME_SYS)
1015 delta = vtime_delta(vtime);
1018 * Task runs either in user (including guest) or kernel space,
1019 * add pending nohz time to the right place.
1021 if (state == VTIME_SYS) {
1022 cpustat[CPUTIME_SYSTEM] += vtime->stime + delta;
1023 } else if (state == VTIME_USER) {
1024 if (task_nice(tsk) > 0)
1025 cpustat[CPUTIME_NICE] += vtime->utime + delta;
1027 cpustat[CPUTIME_USER] += vtime->utime + delta;
1029 WARN_ON_ONCE(state != VTIME_GUEST);
1030 if (task_nice(tsk) > 0) {
1031 cpustat[CPUTIME_GUEST_NICE] += vtime->gtime + delta;
1032 cpustat[CPUTIME_NICE] += vtime->gtime + delta;
1034 cpustat[CPUTIME_GUEST] += vtime->gtime + delta;
1035 cpustat[CPUTIME_USER] += vtime->gtime + delta;
1038 } while (read_seqcount_retry(&vtime->seqcount, seq));
1043 void kcpustat_cpu_fetch(struct kernel_cpustat *dst, int cpu)
1045 const struct kernel_cpustat *src = &kcpustat_cpu(cpu);
1049 if (!vtime_accounting_enabled_cpu(cpu)) {
1057 struct task_struct *curr;
1060 curr = rcu_dereference(rq->curr);
1061 if (WARN_ON_ONCE(!curr)) {
1067 err = kcpustat_cpu_fetch_vtime(dst, src, curr, cpu);
1076 EXPORT_SYMBOL_GPL(kcpustat_cpu_fetch);
1078 #endif /* CONFIG_VIRT_CPU_ACCOUNTING_GEN */