Re: [PATCH v6] sched/clock: Avoid false sharing for sched_clock_irqtime

[Guo, Wangyang]

On 1/27/2026 2:30 PM, K Prateek Nayak wrote:
> Hello Wangyang,
>
> On 1/27/2026 10:11 AM, Wangyang Guo wrote:
> > --- a/kernel/sched/clock.c
> > +++ b/kernel/sched/clock.c
> > @@ -173,6 +173,7 @@ notrace static void __sched_clock_work(struct work_struct *work)
> >   			scd->tick_gtod, __gtod_offset,
> >   			scd->tick_raw,  __sched_clock_offset);
> >   
> > +	disable_sched_clock_irqtime();
>
> We can come here with !irqtime_enabled() but this is a slowpath so I
> don't really mind calling disable_sched_clock_irqtime() unconditionally
> here. Feel free to include:
>
> Reviewed-by: K Prateek Nayak <kprateek.nayak@xxxxxxx>
> Tested-by: K Prateek Nayak <kprateek.nayak@xxxxxxx>
>
> Tested with a bunch of fake cases of sched_clock going unstable by
> hacking the kernel and the irqtime was disabled correctly in both cases
> - pre-lateinit and post-lateinit.

I think we can put irqtime_enabled check in disable_sched_lock_irqtime() 
which will benefit in both scenario:

 void disable_sched_clock_irqtime(void)
 {
-	sched_clock_irqtime = 0;
+	if (irqtime_enabled())
+		static_branch_disable(&sched_clock_irqtime);
 }

v7: 
https://lore.kernel.org/all/20260127072509.2627346-1-wangyang.guo@xxxxxxxxx/