On Tue, Feb 17, 2015 at 04:51:03PM +0000, Dario Faggioli wrote:
> On Fri, 2015-02-13 at 11:11 -0500, Elena Ufimtseva wrote:
> > On Fri, Feb 13, 2015 at 11:06 AM, Wei Liu <wei.l...@citrix.com> wrote:
> > > On Fri, Feb 13, 2015 at 10:39:25AM -0500, Elena Ufimtseva wrote:
>
> > >> Any sanity checks for distances?
> > >>
> > >
> > > The same applies, what is a valid distance what is not? I guess zero is
> > > not valid? Or do we enforce that the distance to local node must be
> > > smaller than or equal to the distance to remote node?
> >
> > Yes, I think the second condition is enough for strict checking.
> >
> That would not harm, probably but I honestly would not put down much
> enforcement on distance values. We can enforce non-zero values, we can
> enforce local < remote, we can enforce the symmetry of the distance
> matrix, but, really, I wouldn't go that far.
>
> What matters most wrt specification of the distances, is to provide a
> sane default, in case one does not want to bother writing it down (or
> does not want to write it down completely, as it could be tedious).
>
> So, if one does not say anything, we should come up with something that
> makes sense (and I'll say more about this while reviewing patch 24). If
> the user does say something, I would just go with that... perhaps after
> printing a warning, but no more than that.
>
The checking here and a sensible default are two things and are parallel
to each other, I think.
I've come up with something like this, to enforce local distance not
larger than remote distance.
/* Check vdistances */
for (i = 0; i < b_info->num_vnuma_nodes; i++) {
uint32_t local_distance, remote_distance;
v = &b_info->vnuma_nodes[i];
local_distance = v->distances[i];
for (j = 0; j < v->num_distances; j++) {
if (i == j) continue;
remote_distance = v->distances[j];
if (local_distance > remote_distance) {
LOG(ERROR,
"Distance from %u to %u smaller than %u's local distance",
i, j, i);
goto out;
}
}
}
Wei.
> Regards,
> Dario
_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xen.org
http://lists.xen.org/xen-devel
