Re: [PATCH for-4.22 5/5] xen/numa: fix setup of non-aligned memory affinity ranges

["Roger Pau Monné" <roger.pau@citrix.com>]

On Mon, Jun 01, 2026 at 06:08:49PM +0100, Andrew Cooper wrote:
> On 01/06/2026 4:43 pm, Roger Pau Monne wrote:
> > diff --git a/xen/common/numa.c b/xen/common/numa.c
> > index 8544a1598218..8b2f29597f30 100644
> > --- a/xen/common/numa.c
> > +++ b/xen/common/numa.c
> > @@ -396,7 +396,13 @@ static int __init populate_memnodemap(const struct node *nodes,
> >  
> >      for ( i = 0; i < numnodes; i++ )
> >      {
> > -        unsigned long spdx = paddr_to_pdx(nodes[i].start);
> > +        /*
> > +         * Round down start address: if start is not aligned to the memnodemap
> > +         * chunk size the tail remainder might not be added.  Overlaps created
> > +         * by rounding will fall into the same NUMA region.
> > +         */
> > +        unsigned long spdx = ROUNDDOWN(paddr_to_pdx(nodes[i].start),
> > +                                       1UL << shift);
> >          unsigned long epdx = paddr_to_pdx(nodes[i].end - 1);
> >  
> >          if ( spdx > epdx )
> 
> I like that this comes with a unit test, but this hunk needs to be
> standalone at the start of the series, because it needs backporting.

I was supposed to add a post-commit note about this, but finding the
offending commit took me a lot more time than I was expecting and
forgot to add the note.

Yes, this needs to be backported.  I was wondering whether it would be
fine to just backport the fix, leaving the test changes alone.  But I
can also split this to a pre-commit.

Thanks, Roger.