Xen devel - Aug 2007 - dealing with ill DMI table pointer

On at least one system we see the DMI table being located in what E820 reports
as usable RAM. Obviously, native has no immediate issue with this as it (a)
needs
the tables only at boot and (b) has no problem ioremap-ing RAM pages. A Xen
kernel, otoh, is likely to die because of this unless it happens to own the
page(s).

The only reasonable workaround I can see would be to have Xen look up the
DMI table and alter the E820 map by hand if needed (and also avoid to destroy
the information contained therein, implying that this must be done pretty
early).

The only other alternative I see would be to simply say: Bad luck, get a BIOS
update. But the DMI code in Linux clearly says that this is not a unique
problem,
so having some kind of workaround to at least gracefully fail dmi_scan_machine()
might be desirable, but would seem to require propagating an error code from
set_fixmap() and changing this function to use hypercalls instead of direct page
table writes.

Jan


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

I''d be happy to have an early fixup of e820 in Xen, and print a
warning,
especially if it''s not much code.

 -- Keir

On 7/8/07 13:46, "Jan Beulich" <jbeulich@novell.com> wrote:
> On at least one system we see the DMI table being located in what E820
reports
> as usable RAM. Obviously, native has no immediate issue with this as it (a)
> needs
> the tables only at boot and (b) has no problem ioremap-ing RAM pages. A Xen
> kernel, otoh, is likely to die because of this unless it happens to own the
> page(s).
> 
> The only reasonable workaround I can see would be to have Xen look up the
> DMI table and alter the E820 map by hand if needed (and also avoid to
destroy
> the information contained therein, implying that this must be done pretty
> early).
> 
> The only other alternative I see would be to simply say: Bad luck, get a
BIOS
> update. But the DMI code in Linux clearly says that this is not a unique
> problem,
> so having some kind of workaround to at least gracefully fail
> dmi_scan_machine()
> might be desirable, but would seem to require propagating an error code
from
> set_fixmap() and changing this function to use hypercalls instead of direct
> page
> table writes.
> 
> Jan
> 
> 
> _______________________________________________
> Xen-devel mailing list
> Xen-devel@lists.xensource.com
> http://lists.xensource.com/xen-devel

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