[Xen-devel] Re: [PATCH 4/6] xen-gntdev: Support mapping in HVM domains

[Ian Campbell <Ian.Campbell@xxxxxxxxxxxxx>]

On Thu, 2011-01-27 at 18:52 +0000, Konrad Rzeszutek Wilk wrote:
> > @@ -179,11 +184,32 @@ static void gntdev_put_map(struct grant_map *map)
> >  
> >       atomic_sub(map->count, &pages_mapped);
> >  
> > -     if (map->pages)
> > +     if (map->pages) {
> > +             if (!use_ptemod)
> > +                     unmap_grant_pages(map, 0, map->count);
> > +
> >               for (i = 0; i < map->count; i++) {
> > -                     if (map->pages[i])
> > +                     uint32_t check, *tmp;
> > +                     if (!map->pages[i])
> > +                             continue;
> > +                     /* XXX When unmapping in an HVM domain, Xen will
> > +                      * sometimes end up mapping the GFN to an invalid MFN.
> > +                      * In this case, writes will be discarded and reads 
> > will
> > +                      * return all 0xFF bytes.  Leak these unusable GFNs
> 
> I forgot to ask, under what version of Xen did you run this? I want to add
> that to the comment so when it gets fixed we know what the failing version is.
> 
> > +                      * until Xen supports fixing their p2m mapping.
> > +                      */
> > +                     tmp = kmap(map->pages[i]);
> > +                     *tmp = 0xdeaddead;

I've just tripped over this check which faults in my PV guest. Seems to
be related to the handling failures of map_grant_pages()?

Was the underlying Xen issue here reported somewhere more obvious than
this comment buried in a patch to the kernel?

If not please can you raise it as a separate thread clearly marked as a
hypervisor issue/question, all I can find is bits and pieces spread
through the threads associated with this kernel patch. I don't think
I've got a clear enough picture of the issue from those fragments to
pull it together into a sensible report.

Ian.

> > +                     mb();
> > +                     check = *tmp;
> > +                     kunmap(map->pages[i]); 


_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxxxxxxxx
http://lists.xensource.com/xen-devel