Re: [PATCH 4/6] genirq: optimize irq_do_set_affinity()

[Yury Norov]

On Tue, May 14, 2024 at 5:51 AM Jinjie Ruan <ruanjinjie@xxxxxxxxxx> wrote:
>
>
>
> On 2024/5/14 6:01, Yury Norov wrote:
> > If mask == desc->irq_common_data.affinity, copying one to another is
> > useless, and we can just skip it.
> >
> > Signed-off-by: Yury Norov <yury.norov@xxxxxxxxx>
> > ---
> >  kernel/irq/manage.c | 3 ++-
> >  1 file changed, 2 insertions(+), 1 deletion(-)
> >
> > diff --git a/kernel/irq/manage.c b/kernel/irq/manage.c
> > index bf9ae8a8686f..ad9ed9fdf919 100644
> > --- a/kernel/irq/manage.c
> > +++ b/kernel/irq/manage.c
> > @@ -285,7 +285,8 @@ int irq_do_set_affinity(struct irq_data *data, const struct cpumask *mask,
> >       switch (ret) {
> >       case IRQ_SET_MASK_OK:
> >       case IRQ_SET_MASK_OK_DONE:
> > -             cpumask_copy(desc->irq_common_data.affinity, mask);
> > +             if (desc->irq_common_data.affinity != mask)
> > +                     cpumask_copy(desc->irq_common_data.affinity, mask);
>
> It seems that mask is a pointer, shouldn't use "cpumask_equal"？

cpumask_equal() is O(N), just as cpumask_copy(), so we'll have no
benefit if the masks are equal, and will double slow it if they aren't
in the worst case.

On the other hand, pointers comparison is O(1), a very quick tasks,
even more the pointers are already in registers.

> >               fallthrough;
> >       case IRQ_SET_MASK_OK_NOCOPY:
> >               irq_validate_effective_affinity(data);