On Fri, Dec 7, 2018 at 8:51 AM Florian Weimer <fweimer@redhat.com> wrote:
>
> * H. J. Lu:
>
> > +* The getcpu wrapper function has been added, which returns currently
> > +  used CPU and NUMA node.  This function is Linux-specific.
>
> “returns the currently used” (not sure)?

Fixed.

> > diff --git a/sysdeps/unix/sysv/linux/tst-skeleton-affinity.c b/sysdeps/unix/sysv/linux/tst-skeleton-affinity.c
> > index 695c1ccdbd..fd1357beb1 100644
> > --- a/sysdeps/unix/sysv/linux/tst-skeleton-affinity.c
> > +++ b/sysdeps/unix/sysv/linux/tst-skeleton-affinity.c
> > @@ -165,6 +165,18 @@ test_size (const struct conf *conf, size_t size)
> >    for (int cpu = 0; cpu <= conf->last_cpu; ++cpu)
> >      {
> >        int active_cpu = sched_getcpu ();
> > +      unsigned int numa_cpu, numa_node;
> > +      if (getcpu (&numa_cpu, &numa_node))
>
> != NULL

I changed to

if (getcpu (&numa_cpu, &numa_node) != 0)

since getcpu returns int.

> > +     {
> > +       printf ("error: getcpu: %m\n");
> > +       return false;
>
> Okay, ENOSYS is a hard error here because we have checked for a working
> sched_getcpu.
>
> > +     }
> > +      if ((unsigned int) active_cpu != numa_cpu)
> > +     {
> > +       printf ("error: Unexpected CPU %d, expected %d\n",
> > +               active_cpu, numa_cpu);
> > +       return false;
> > +     }
> >        if (last_active_cpu >= 0 && last_active_cpu != active_cpu)
>
> I think you need to move the getcpu call and this check down further,
> after the setaffinity call with a single-element CPU set.
>

Done.

Here is the updated patch.  OK for master?

Thanks.

-- 
H.J.