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>
10 #ifdef CONFIG_IRQ_TIME_ACCOUNTING
13 * There are no locks covering percpu hardirq/softirq time.
14 * They are only modified in vtime_account, on corresponding CPU
15 * with interrupts disabled. So, writes are safe.
16 * They are read and saved off onto struct rq in update_rq_clock().
17 * This may result in other CPU reading this CPU's irq time and can
18 * race with irq/vtime_account on this CPU. We would either get old
19 * or new value with a side effect of accounting a slice of irq time to wrong
20 * task when irq is in progress while we read rq->clock. That is a worthy
21 * compromise in place of having locks on each irq in account_system_time.
23 DEFINE_PER_CPU(u64, cpu_hardirq_time);
24 DEFINE_PER_CPU(u64, cpu_softirq_time);
26 static DEFINE_PER_CPU(u64, irq_start_time);
27 static int sched_clock_irqtime;
29 void enable_sched_clock_irqtime(void)
31 sched_clock_irqtime = 1;
34 void disable_sched_clock_irqtime(void)
36 sched_clock_irqtime = 0;
40 DEFINE_PER_CPU(seqcount_t, irq_time_seq);
41 #endif /* CONFIG_64BIT */
44 * Called before incrementing preempt_count on {soft,}irq_enter
45 * and before decrementing preempt_count on {soft,}irq_exit.
47 void irqtime_account_irq(struct task_struct *curr)
53 if (!sched_clock_irqtime)
56 local_irq_save(flags);
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);
75 local_irq_restore(flags);
77 EXPORT_SYMBOL_GPL(irqtime_account_irq);
79 static int irqtime_account_hi_update(void)
81 u64 *cpustat = kcpustat_this_cpu->cpustat;
86 local_irq_save(flags);
87 latest_ns = this_cpu_read(cpu_hardirq_time);
88 if (nsecs_to_cputime64(latest_ns) > cpustat[CPUTIME_IRQ])
90 local_irq_restore(flags);
94 static int irqtime_account_si_update(void)
96 u64 *cpustat = kcpustat_this_cpu->cpustat;
101 local_irq_save(flags);
102 latest_ns = this_cpu_read(cpu_softirq_time);
103 if (nsecs_to_cputime64(latest_ns) > cpustat[CPUTIME_SOFTIRQ])
105 local_irq_restore(flags);
109 #else /* CONFIG_IRQ_TIME_ACCOUNTING */
111 #define sched_clock_irqtime (0)
113 #endif /* !CONFIG_IRQ_TIME_ACCOUNTING */
115 static inline void task_group_account_field(struct task_struct *p, int index,
119 * Since all updates are sure to touch the root cgroup, we
120 * get ourselves ahead and touch it first. If the root cgroup
121 * is the only cgroup, then nothing else should be necessary.
124 __get_cpu_var(kernel_cpustat).cpustat[index] += tmp;
126 cpuacct_account_field(p, index, tmp);
130 * Account user cpu time to a process.
131 * @p: the process that the cpu time gets accounted to
132 * @cputime: the cpu time spent in user space since the last update
133 * @cputime_scaled: cputime scaled by cpu frequency
135 void account_user_time(struct task_struct *p, cputime_t cputime,
136 cputime_t cputime_scaled)
140 /* Add user time to process. */
142 p->utimescaled += cputime_scaled;
143 account_group_user_time(p, cputime);
145 index = (TASK_NICE(p) > 0) ? CPUTIME_NICE : CPUTIME_USER;
147 /* Add user time to cpustat. */
148 task_group_account_field(p, index, (__force u64) cputime);
150 /* Account for user time used */
151 acct_account_cputime(p);
155 * Account guest cpu time to a process.
156 * @p: the process that the cpu time gets accounted to
157 * @cputime: the cpu time spent in virtual machine since the last update
158 * @cputime_scaled: cputime scaled by cpu frequency
160 static void account_guest_time(struct task_struct *p, cputime_t cputime,
161 cputime_t cputime_scaled)
163 u64 *cpustat = kcpustat_this_cpu->cpustat;
165 /* Add guest time to process. */
167 p->utimescaled += cputime_scaled;
168 account_group_user_time(p, cputime);
171 /* Add guest time to cpustat. */
172 if (TASK_NICE(p) > 0) {
173 cpustat[CPUTIME_NICE] += (__force u64) cputime;
174 cpustat[CPUTIME_GUEST_NICE] += (__force u64) cputime;
176 cpustat[CPUTIME_USER] += (__force u64) cputime;
177 cpustat[CPUTIME_GUEST] += (__force u64) cputime;
182 * Account system cpu time to a process and desired cpustat field
183 * @p: the process that the cpu time gets accounted to
184 * @cputime: the cpu time spent in kernel space since the last update
185 * @cputime_scaled: cputime scaled by cpu frequency
186 * @target_cputime64: pointer to cpustat field that has to be updated
189 void __account_system_time(struct task_struct *p, cputime_t cputime,
190 cputime_t cputime_scaled, int index)
192 /* Add system time to process. */
194 p->stimescaled += cputime_scaled;
195 account_group_system_time(p, cputime);
197 /* Add system time to cpustat. */
198 task_group_account_field(p, index, (__force u64) cputime);
200 /* Account for system time used */
201 acct_account_cputime(p);
205 * Account system cpu time to a process.
206 * @p: the process that the cpu time gets accounted to
207 * @hardirq_offset: the offset to subtract from hardirq_count()
208 * @cputime: the cpu time spent in kernel space since the last update
209 * @cputime_scaled: cputime scaled by cpu frequency
211 void account_system_time(struct task_struct *p, int hardirq_offset,
212 cputime_t cputime, cputime_t cputime_scaled)
216 if ((p->flags & PF_VCPU) && (irq_count() - hardirq_offset == 0)) {
217 account_guest_time(p, cputime, cputime_scaled);
221 if (hardirq_count() - hardirq_offset)
223 else if (in_serving_softirq())
224 index = CPUTIME_SOFTIRQ;
226 index = CPUTIME_SYSTEM;
228 __account_system_time(p, cputime, cputime_scaled, index);
232 * Account for involuntary wait time.
233 * @cputime: the cpu time spent in involuntary wait
235 void account_steal_time(cputime_t cputime)
237 u64 *cpustat = kcpustat_this_cpu->cpustat;
239 cpustat[CPUTIME_STEAL] += (__force u64) cputime;
243 * Account for idle time.
244 * @cputime: the cpu time spent in idle wait
246 void account_idle_time(cputime_t cputime)
248 u64 *cpustat = kcpustat_this_cpu->cpustat;
249 struct rq *rq = this_rq();
251 if (atomic_read(&rq->nr_iowait) > 0)
252 cpustat[CPUTIME_IOWAIT] += (__force u64) cputime;
254 cpustat[CPUTIME_IDLE] += (__force u64) cputime;
257 static __always_inline bool steal_account_process_tick(void)
259 #ifdef CONFIG_PARAVIRT
260 if (static_key_false(¶virt_steal_enabled)) {
263 steal = paravirt_steal_clock(smp_processor_id());
264 steal -= this_rq()->prev_steal_time;
266 st = steal_ticks(steal);
267 this_rq()->prev_steal_time += st * TICK_NSEC;
269 account_steal_time(st);
277 * Accumulate raw cputime values of dead tasks (sig->[us]time) and live
278 * tasks (sum on group iteration) belonging to @tsk's group.
280 void thread_group_cputime(struct task_struct *tsk, struct task_cputime *times)
282 struct signal_struct *sig = tsk->signal;
283 cputime_t utime, stime;
284 struct task_struct *t;
286 times->utime = sig->utime;
287 times->stime = sig->stime;
288 times->sum_exec_runtime = sig->sum_sched_runtime;
291 /* make sure we can trust tsk->thread_group list */
292 if (!likely(pid_alive(tsk)))
297 task_cputime(tsk, &utime, &stime);
298 times->utime += utime;
299 times->stime += stime;
300 times->sum_exec_runtime += task_sched_runtime(t);
301 } while_each_thread(tsk, t);
306 #ifdef CONFIG_IRQ_TIME_ACCOUNTING
308 * Account a tick to a process and cpustat
309 * @p: the process that the cpu time gets accounted to
310 * @user_tick: is the tick from userspace
311 * @rq: the pointer to rq
313 * Tick demultiplexing follows the order
314 * - pending hardirq update
315 * - pending softirq update
319 * - check for guest_time
320 * - else account as system_time
322 * Check for hardirq is done both for system and user time as there is
323 * no timer going off while we are on hardirq and hence we may never get an
324 * opportunity to update it solely in system time.
325 * p->stime and friends are only updated on system time and not on irq
326 * softirq as those do not count in task exec_runtime any more.
328 static void irqtime_account_process_tick(struct task_struct *p, int user_tick,
331 cputime_t one_jiffy_scaled = cputime_to_scaled(cputime_one_jiffy);
332 u64 *cpustat = kcpustat_this_cpu->cpustat;
334 if (steal_account_process_tick())
337 if (irqtime_account_hi_update()) {
338 cpustat[CPUTIME_IRQ] += (__force u64) cputime_one_jiffy;
339 } else if (irqtime_account_si_update()) {
340 cpustat[CPUTIME_SOFTIRQ] += (__force u64) cputime_one_jiffy;
341 } else if (this_cpu_ksoftirqd() == p) {
343 * ksoftirqd time do not get accounted in cpu_softirq_time.
344 * So, we have to handle it separately here.
345 * Also, p->stime needs to be updated for ksoftirqd.
347 __account_system_time(p, cputime_one_jiffy, one_jiffy_scaled,
349 } else if (user_tick) {
350 account_user_time(p, cputime_one_jiffy, one_jiffy_scaled);
351 } else if (p == rq->idle) {
352 account_idle_time(cputime_one_jiffy);
353 } else if (p->flags & PF_VCPU) { /* System time or guest time */
354 account_guest_time(p, cputime_one_jiffy, one_jiffy_scaled);
356 __account_system_time(p, cputime_one_jiffy, one_jiffy_scaled,
361 static void irqtime_account_idle_ticks(int ticks)
364 struct rq *rq = this_rq();
366 for (i = 0; i < ticks; i++)
367 irqtime_account_process_tick(current, 0, rq);
369 #else /* CONFIG_IRQ_TIME_ACCOUNTING */
370 static inline void irqtime_account_idle_ticks(int ticks) {}
371 static inline void irqtime_account_process_tick(struct task_struct *p, int user_tick,
373 #endif /* CONFIG_IRQ_TIME_ACCOUNTING */
376 * Use precise platform statistics if available:
378 #ifdef CONFIG_VIRT_CPU_ACCOUNTING
380 #ifndef __ARCH_HAS_VTIME_TASK_SWITCH
381 void vtime_task_switch(struct task_struct *prev)
383 if (!vtime_accounting_enabled())
386 if (is_idle_task(prev))
387 vtime_account_idle(prev);
389 vtime_account_system(prev);
391 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_NATIVE
392 vtime_account_user(prev);
394 arch_vtime_task_switch(prev);
399 * Archs that account the whole time spent in the idle task
400 * (outside irq) as idle time can rely on this and just implement
401 * vtime_account_system() and vtime_account_idle(). Archs that
402 * have other meaning of the idle time (s390 only includes the
403 * time spent by the CPU when it's in low power mode) must override
406 #ifndef __ARCH_HAS_VTIME_ACCOUNT
407 void vtime_account_irq_enter(struct task_struct *tsk)
409 if (!vtime_accounting_enabled())
412 if (!in_interrupt()) {
414 * If we interrupted user, context_tracking_in_user()
415 * is 1 because the context tracking don't hook
416 * on irq entry/exit. This way we know if
417 * we need to flush user time on kernel entry.
419 if (context_tracking_in_user()) {
420 vtime_account_user(tsk);
424 if (is_idle_task(tsk)) {
425 vtime_account_idle(tsk);
429 vtime_account_system(tsk);
431 EXPORT_SYMBOL_GPL(vtime_account_irq_enter);
432 #endif /* __ARCH_HAS_VTIME_ACCOUNT */
433 #endif /* CONFIG_VIRT_CPU_ACCOUNTING */
436 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_NATIVE
437 void task_cputime_adjusted(struct task_struct *p, cputime_t *ut, cputime_t *st)
443 void thread_group_cputime_adjusted(struct task_struct *p, cputime_t *ut, cputime_t *st)
445 struct task_cputime cputime;
447 thread_group_cputime(p, &cputime);
452 #else /* !CONFIG_VIRT_CPU_ACCOUNTING_NATIVE */
454 * Account a single tick of cpu time.
455 * @p: the process that the cpu time gets accounted to
456 * @user_tick: indicates if the tick is a user or a system tick
458 void account_process_tick(struct task_struct *p, int user_tick)
460 cputime_t one_jiffy_scaled = cputime_to_scaled(cputime_one_jiffy);
461 struct rq *rq = this_rq();
463 if (vtime_accounting_enabled())
466 if (sched_clock_irqtime) {
467 irqtime_account_process_tick(p, user_tick, rq);
471 if (steal_account_process_tick())
475 account_user_time(p, cputime_one_jiffy, one_jiffy_scaled);
476 else if ((p != rq->idle) || (irq_count() != HARDIRQ_OFFSET))
477 account_system_time(p, HARDIRQ_OFFSET, cputime_one_jiffy,
480 account_idle_time(cputime_one_jiffy);
484 * Account multiple ticks of steal time.
485 * @p: the process from which the cpu time has been stolen
486 * @ticks: number of stolen ticks
488 void account_steal_ticks(unsigned long ticks)
490 account_steal_time(jiffies_to_cputime(ticks));
494 * Account multiple ticks of idle time.
495 * @ticks: number of stolen ticks
497 void account_idle_ticks(unsigned long ticks)
500 if (sched_clock_irqtime) {
501 irqtime_account_idle_ticks(ticks);
505 account_idle_time(jiffies_to_cputime(ticks));
509 * Perform (stime * rtime) / total with reduced chances
510 * of multiplication overflows by using smaller factors
511 * like quotient and remainders of divisions between
514 static cputime_t scale_stime(u64 stime, u64 rtime, u64 total)
516 u64 rem, res, scaled;
518 if (rtime >= total) {
520 * Scale up to rtime / total then add
521 * the remainder scaled to stime / total.
523 res = div64_u64_rem(rtime, total, &rem);
524 scaled = stime * res;
525 scaled += div64_u64(stime * rem, total);
528 * Same in reverse: scale down to total / rtime
529 * then substract that result scaled to
530 * to the remaining part.
532 res = div64_u64_rem(total, rtime, &rem);
533 scaled = div64_u64(stime, res);
534 scaled -= div64_u64(scaled * rem, total);
537 return (__force cputime_t) scaled;
541 * Adjust tick based cputime random precision against scheduler
542 * runtime accounting.
544 static void cputime_adjust(struct task_cputime *curr,
545 struct cputime *prev,
546 cputime_t *ut, cputime_t *st)
548 cputime_t rtime, stime, total;
550 if (vtime_accounting_enabled()) {
557 total = stime + curr->utime;
560 * Tick based cputime accounting depend on random scheduling
561 * timeslices of a task to be interrupted or not by the timer.
562 * Depending on these circumstances, the number of these interrupts
563 * may be over or under-optimistic, matching the real user and system
564 * cputime with a variable precision.
566 * Fix this by scaling these tick based values against the total
567 * runtime accounted by the CFS scheduler.
569 rtime = nsecs_to_cputime(curr->sum_exec_runtime);
576 stime = scale_stime((__force u64)stime,
577 (__force u64)rtime, (__force u64)total);
581 * If the tick based count grows faster than the scheduler one,
582 * the result of the scaling may go backward.
583 * Let's enforce monotonicity.
585 prev->stime = max(prev->stime, stime);
586 prev->utime = max(prev->utime, rtime - prev->stime);
592 void task_cputime_adjusted(struct task_struct *p, cputime_t *ut, cputime_t *st)
594 struct task_cputime cputime = {
595 .sum_exec_runtime = p->se.sum_exec_runtime,
598 task_cputime(p, &cputime.utime, &cputime.stime);
599 cputime_adjust(&cputime, &p->prev_cputime, ut, st);
603 * Must be called with siglock held.
605 void thread_group_cputime_adjusted(struct task_struct *p, cputime_t *ut, cputime_t *st)
607 struct task_cputime cputime;
609 thread_group_cputime(p, &cputime);
610 cputime_adjust(&cputime, &p->signal->prev_cputime, ut, st);
612 #endif /* !CONFIG_VIRT_CPU_ACCOUNTING_NATIVE */
614 #ifdef CONFIG_VIRT_CPU_ACCOUNTING_GEN
615 static unsigned long long vtime_delta(struct task_struct *tsk)
617 unsigned long long clock;
619 clock = local_clock();
620 if (clock < tsk->vtime_snap)
623 return clock - tsk->vtime_snap;
626 static cputime_t get_vtime_delta(struct task_struct *tsk)
628 unsigned long long delta = vtime_delta(tsk);
630 WARN_ON_ONCE(tsk->vtime_snap_whence == VTIME_SLEEPING);
631 tsk->vtime_snap += delta;
633 /* CHECKME: always safe to convert nsecs to cputime? */
634 return nsecs_to_cputime(delta);
637 static void __vtime_account_system(struct task_struct *tsk)
639 cputime_t delta_cpu = get_vtime_delta(tsk);
641 account_system_time(tsk, irq_count(), delta_cpu, cputime_to_scaled(delta_cpu));
644 void vtime_account_system(struct task_struct *tsk)
646 if (!vtime_accounting_enabled())
649 write_seqlock(&tsk->vtime_seqlock);
650 __vtime_account_system(tsk);
651 write_sequnlock(&tsk->vtime_seqlock);
654 void vtime_account_irq_exit(struct task_struct *tsk)
656 if (!vtime_accounting_enabled())
659 write_seqlock(&tsk->vtime_seqlock);
660 if (context_tracking_in_user())
661 tsk->vtime_snap_whence = VTIME_USER;
662 __vtime_account_system(tsk);
663 write_sequnlock(&tsk->vtime_seqlock);
666 void vtime_account_user(struct task_struct *tsk)
670 if (!vtime_accounting_enabled())
673 delta_cpu = get_vtime_delta(tsk);
675 write_seqlock(&tsk->vtime_seqlock);
676 tsk->vtime_snap_whence = VTIME_SYS;
677 account_user_time(tsk, delta_cpu, cputime_to_scaled(delta_cpu));
678 write_sequnlock(&tsk->vtime_seqlock);
681 void vtime_user_enter(struct task_struct *tsk)
683 if (!vtime_accounting_enabled())
686 write_seqlock(&tsk->vtime_seqlock);
687 tsk->vtime_snap_whence = VTIME_USER;
688 __vtime_account_system(tsk);
689 write_sequnlock(&tsk->vtime_seqlock);
692 void vtime_guest_enter(struct task_struct *tsk)
694 write_seqlock(&tsk->vtime_seqlock);
695 __vtime_account_system(tsk);
696 current->flags |= PF_VCPU;
697 write_sequnlock(&tsk->vtime_seqlock);
700 void vtime_guest_exit(struct task_struct *tsk)
702 write_seqlock(&tsk->vtime_seqlock);
703 __vtime_account_system(tsk);
704 current->flags &= ~PF_VCPU;
705 write_sequnlock(&tsk->vtime_seqlock);
708 void vtime_account_idle(struct task_struct *tsk)
710 cputime_t delta_cpu = get_vtime_delta(tsk);
712 account_idle_time(delta_cpu);
715 bool vtime_accounting_enabled(void)
717 return context_tracking_active();
720 void arch_vtime_task_switch(struct task_struct *prev)
722 write_seqlock(&prev->vtime_seqlock);
723 prev->vtime_snap_whence = VTIME_SLEEPING;
724 write_sequnlock(&prev->vtime_seqlock);
726 write_seqlock(¤t->vtime_seqlock);
727 current->vtime_snap_whence = VTIME_SYS;
728 current->vtime_snap = sched_clock();
729 write_sequnlock(¤t->vtime_seqlock);
732 void vtime_init_idle(struct task_struct *t)
736 write_seqlock_irqsave(&t->vtime_seqlock, flags);
737 t->vtime_snap_whence = VTIME_SYS;
738 t->vtime_snap = sched_clock();
739 write_sequnlock_irqrestore(&t->vtime_seqlock, flags);
742 cputime_t task_gtime(struct task_struct *t)
748 seq = read_seqbegin(&t->vtime_seqlock);
751 if (t->flags & PF_VCPU)
752 gtime += vtime_delta(t);
754 } while (read_seqretry(&t->vtime_seqlock, seq));
760 * Fetch cputime raw values from fields of task_struct and
761 * add up the pending nohz execution time since the last
765 fetch_task_cputime(struct task_struct *t,
766 cputime_t *u_dst, cputime_t *s_dst,
767 cputime_t *u_src, cputime_t *s_src,
768 cputime_t *udelta, cputime_t *sdelta)
771 unsigned long long delta;
777 seq = read_seqbegin(&t->vtime_seqlock);
784 /* Task is sleeping, nothing to add */
785 if (t->vtime_snap_whence == VTIME_SLEEPING ||
789 delta = vtime_delta(t);
792 * Task runs either in user or kernel space, add pending nohz time to
795 if (t->vtime_snap_whence == VTIME_USER || t->flags & PF_VCPU) {
798 if (t->vtime_snap_whence == VTIME_SYS)
801 } while (read_seqretry(&t->vtime_seqlock, seq));
805 void task_cputime(struct task_struct *t, cputime_t *utime, cputime_t *stime)
807 cputime_t udelta, sdelta;
809 fetch_task_cputime(t, utime, stime, &t->utime,
810 &t->stime, &udelta, &sdelta);
817 void task_cputime_scaled(struct task_struct *t,
818 cputime_t *utimescaled, cputime_t *stimescaled)
820 cputime_t udelta, sdelta;
822 fetch_task_cputime(t, utimescaled, stimescaled,
823 &t->utimescaled, &t->stimescaled, &udelta, &sdelta);
825 *utimescaled += cputime_to_scaled(udelta);
827 *stimescaled += cputime_to_scaled(sdelta);
829 #endif /* CONFIG_VIRT_CPU_ACCOUNTING_GEN */