Re: [PATCH 08/16] smp: optimize smp_call_function_many_cond() for more

[Yury Norov]

On Mon, Jul 18, 2022 at 11:29:06PM +0200, Peter Zijlstra wrote:
> On Mon, Jul 18, 2022 at 12:28:36PM -0700, Yury Norov wrote:
> 
> > ---
> >  kernel/smp.c | 6 +++++-
> >  1 file changed, 5 insertions(+), 1 deletion(-)
> > 
> > diff --git a/kernel/smp.c b/kernel/smp.c
> > index 7ed2b9b12f74..f96fdf944b4a 100644
> > --- a/kernel/smp.c
> > +++ b/kernel/smp.c
> > @@ -942,7 +942,11 @@ static void smp_call_function_many_cond(const struct cpumask *mask,
> >  
> >  	if (run_remote) {
> >  		cfd = this_cpu_ptr(&cfd_data);
> > -		cpumask_and(cfd->cpumask, mask, cpu_online_mask);
> > +		if (mask == cpu_online_mask)
> > +			cpumask_copy(cfd->cpumask, cpu_online_mask);
> > +		else
> > +			cpumask_and(cfd->cpumask, mask, cpu_online_mask);
> > +
> 
> Or... you could optimize cpumask_and() to detect the src1p == src2p case?

This is not what I would consider as optimization. For vast majority
of users this check is useless because they know for sure that
cpumasks are different.

For this case I can invent something like cpumask_and_check_eq(), so
that there'll be minimal impact on user code. (Suggestions for a better
name are very welcome.)

Thanks,
Yury