Re: RE: [RFC] fix xen_in_range()

[Qing He <qing.he@intel.com>]

On Fri, 2009-04-24 at 15:16 +0800, Keir Fraser wrote:
> On 24/04/2009 08:04, "Jan Beulich" <jbeulich@novell.com> wrote:
> 
> > Also, after suggesting to use gb-pages when possible here I realized that
> > it's probably a latent bug to map more space than was allocated - if the
> > non-allocated-but-mapped pages happen to later get allocated to a domain,
> > that domain may change the cacheability attributes of any of these pages,
> > resulting in aliasing issues. I'll put together a patch for this, but it'll be
> > a couple of days until I'll be able to do so.
> 
> I think we should shatter the superpage on demand. This would also be
> required for superpage mappings of Xen itself: when we free initmem that
> memory can now be allocated to a domain (now xenheap and domheap are merged
> on x86/64).

Actually, we just found an example that problems can arise here.
Just it's not the superpage shattering, but unused __per_cpu area. Dom0
actually gets these pages and tries to do DMA on them, which causes VT-d
page fault and dom0 boot failure. I have a patch for it adding this part
to iommu page table.

Thanks,
Qing
> 
> An alternative might be to mark such partially-freed superpages as
> Xenheap-only, and allocate them preferentially for Xenheap callers (i.e.,
> alloc those pages first, then from the general heap).

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