Re: [PATCH v4 16/21] libxl: build, check and pass vNUMA info to Xen for HVM guest

[Wei Liu <wei.liu2@citrix.com>]

On Wed, Jan 28, 2015 at 04:41:05PM +0000, Ian Campbell wrote:
> On Fri, 2015-01-23 at 11:13 +0000, Wei Liu wrote:
> > +    if (info->num_vnuma_nodes != 0) {
> > +        int i;
> > +
> > +        args.nr_vnuma_info = info->num_vnuma_nodes;
> > +        args.vnuma_info = libxl__malloc(gc, sizeof(*args.vnuma_info) *
> > +                                        args.nr_vnuma_info);
> > +        for (i = 0; i < args.nr_vnuma_info; i++) {
> > +            args.vnuma_info[i].vnode = i;
> > +            args.vnuma_info[i].pnode = info->vnuma_nodes[i].pnode;
> > +            args.vnuma_info[i].pages = info->vnuma_nodes[i].memkb >> 2;
> > +        }
> > +
> > +        /* Consider video ram belongs to node 0 */
> > +        args.vnuma_info[0].pages -= (info->video_memkb >> 2);
> 
> You need to check that the node is big enough I think?
> 

Right.

> > +            if (next >= hole_start && next < hole_end)
> > +                next = hole_end;
> > +            if ((next < hole_start) && (next + remaining_bytes >= hole_start))
> > +                count = hole_start - next;
> > +
> > +            GCREALLOC_ARRAY(vmemranges, nr_vmemrange+1);
> 
> Given there's only a single hole, can't you calculate the final
> nr_vmemrange value up front? (i.e. it's 2, I think).
> 

I would like to keep the algorithm more flexible here so that we can
incorporate e820_host option for hvm guest in the future.

Wei.

> Ian.