Re: [Xen-users] Xen IOMMU disabled due to IVRS table... Blah blah blah

[David Sutton <kantras@xxxxxxxxx>]

Feral,

On Tue, May 28, 2013 at 12:40 PM, feral <blistovmhz@xxxxxxxxx> wrote:

I agree entirely about who should be taking responsibility, but we all
know of course, that Asus will never fix this.  There are probably
less than 5000 people in the world affected by this, and if we assume
most of them will still buy Asus's next board that has working IVRS
tables, Asus doesn't care.

I'm more interested in knowing why the
"iommu=no-amd-iommu-perdev-intremap" option doesn't cover my issue
though.  It seems odd to me that this option was implemented precisely
to work around the regression issue, but somehow it didn't catch them
all.  Goddamn I can't wait for the Xen Documentation to take form :)

The problem is that there were two issues which were identified, and that flag only deals with one of them; the first issue is where the IVRS has two entries for the IO-APIC (for NB and SB) but has the same id listed for both (causing a conflict). This means you can't split across the different IO-APICs because its not sure which is which - the option allows it to use a global pool, decreasing security but allowing things to work. The second issue is where the IVRS table doesn't have a valid id listed for the IO-APIC at all. This second case, because there isn't an id listed, is the one where it just disables AMD-Vi support.

Regards,

  David

_______________________________________________
Xen-users mailing list
Xen-users@xxxxxxxxxxxxx
http://lists.xen.org/xen-users