Re: [Xen-ia64-devel] Re: PATCH[1/2] Self IO EMUlator (sioemu) - Hypervisor part

[Samuel Thibault <samuel.thibault@xxxxxxxxxxxxx>]

Hello,

Alex Williamson, le Wed 13 Feb 2008 15:39:10 -0700, a écrit :
>    I haven't had a chance to try this yet, but has any resolution been
> reached between this approach versus the HVM stub domain?

I actually think both may coexist.

> If I understand correctly, they seem to have similar goals.

Yes.

> It seems like Samuel has been making forward progress on the stub
> approach.

It's even merged in the unstable tree actually.
However, it can't work yet for ia64 simply because the MM part of
Mini-OS is far from complete for IA-64 (map_frames() is only able to map
single frames).

> Does this provide more functionality? more performance? less
> maintenance?

sioemu probably provides better performance because it doesn't require a
domain switch.  However, it probably requires more maintenance because
it means adapting the emulation code into firmware, while the stubdomain
approach has been to juste compile qemu almost "as is".  As a result,
for now sioemu provides less functionalities than stubdomains do for
x86.

> If we take a different approach than x86, I want to make sure it's
> justified.  Thanks,

The performance benefit of sioemu may be noticeable.  We currently can't
compare, since no sioemu implementation was done for x86 (actually
it looks like it may be more difficult since x86 ROM area is quite
limited), and Mini-OS needs some more porting before stubdomains can
work on IA64.

Samuel

_______________________________________________
Xen-ia64-devel mailing list
Xen-ia64-devel@xxxxxxxxxxxxxxxxxxx
http://lists.xensource.com/xen-ia64-devel