Re: [PATCH] hvm: Use main memory for video memory. [Was: xend, 8MB video memory and ballooning]

[Samuel Thibault <samuel.thibault@eu.citrix.com>]

Keir Fraser, le Thu 31 Jul 2008 14:39:59 +0100, a écrit :
> On 31/7/08 14:18, "Samuel Thibault" <samuel.thibault@eu.citrix.com> wrote:
> 
> > diff -r 83b3a8a2a4f3 tools/libxc/xc_domain_save.c
> > --- a/tools/libxc/xc_domain_save.c Thu Jul 31 12:58:07 2008 +0100
> > +++ b/tools/libxc/xc_domain_save.c Thu Jul 31 13:20:02 2008 +0100
> > @@ -1112,9 +1112,7 @@ int xc_domain_save(int xc_handle, int io
> >                      continue;
> >  
> >                  /* Skip PFNs that aren't really there */
> > -                if ( hvm && ((n >= 0xa0 && n < 0xc0) /* VGA hole */
> > -                             || (n >= (HVM_BELOW_4G_MMIO_START >> PAGE_SHIFT)
> > -                                 && n < (1ULL<<32) >> PAGE_SHIFT)) /* MMIO */
> > )
> > +                if ( hvm && n == shared_info_frame )
> >                      continue;
> 
> What happens if you don't special-case shared_info_frame?

I actually hadn't even tried.

> I'd've hoped it would just work... does the extra page of allocation
> during restore fail?

It does work, but I guess it replaces the mapping of the "magic" page
with the mapping of the allocated page, so that shared_info wouldn't
work any more.

> I'm thinking that the grant tables will be mapped in the I/O hole too, and
> that is from 'Xen heap' just like shared_info is.

Urgl, hadn't thought about those, I guess the PV drivers would
re-etablish them on resume anyway?

> So special-casing one and not the other sure is odd.

Actually, shouldn't it be up to the hypervisor to not return all these
from the SHADOW_OP_PEEK operation?

Samuel