12:40 a.m.
I have a virtual machine landscape consisting of a CentOS 5.3 Dom0 with
Xen 3.1.2, some PV CentOS DomU's with PCI passthrough (same versions),
and I'm trying to implement a F10 DomU with PCI passthrough. I set
everything up for the passthrough, and the F10 guest can't see any of
the devices with lspci. If I boot a CentOS PV machine with the same PCI
passthrough configuration, the devices show up with lspci. I've seen
some posts in various places about this issue, but no solutions. How do
I implement this?
1:51 p.m.
2009/5/1 Jim Lutz <jelutz(a)config.com>:
I have a virtual machine landscape consisting of a CentOS 5.3 Dom0
with Xen
3.1.2, some PV CentOS DomU's with PCI passthrough (same versions), and I'm
trying to implement a F10 DomU with PCI passthrough. I set everything up for
the passthrough, and the F10 guest can't see any of the devices with lspci.
If I boot a CentOS PV machine with the same PCI passthrough configuration,
the devices show up with lspci. I've seen some posts in various places about
this issue, but no solutions. How do I implement this?
There has been no pcifront in recent Fedora kernels, you'll have to
stick with older centOS kernels. or wait until pv_ops domU has landed
(so 2.6.31 at the earliest) *and* the pcifront has been re-implemented
(not sure if anyone other than Jeremy has this in their sights)
<whisper> or look at ubuntu which I think has forward ported the old
non-pv_ops xen to more recent kernels </whisper>
10:54 a.m.
On Mon, May 04, 2009 at 07:51:37PM +0100, Andy Burns wrote:
2009/5/1 Jim Lutz <jelutz(a)config.com>:
> I have a virtual machine landscape consisting of a CentOS 5.3 Dom0 with Xen
> 3.1.2, some PV CentOS DomU's with PCI passthrough (same versions), and I'm
> trying to implement a F10 DomU with PCI passthrough. I set everything up for
> the passthrough, and the F10 guest can't see any of the devices with lspci.
> If I boot a CentOS PV machine with the same PCI passthrough configuration,
> the devices show up with lspci. I've seen some posts in various places about
> this issue, but no solutions. How do I implement this?
There has been no pcifront in recent Fedora kernels, you'll have to
stick with older centOS kernels. or wait until pv_ops domU has landed
(so 2.6.31 at the earliest) *and* the pcifront has been re-implemented
(not sure if anyone other than Jeremy has this in their sights)
Jeremy's git tree has pv_ops pcifront now.
<whisper> or look at ubuntu which I think has forward ported
the old
non-pv_ops xen to more recent kernels </whisper>
Ubuntu's ports are known to be a bit buggy.. I wouldn't recommend those.
Novell/opensuse forward-ports are better.. they're shipping those patches
with SLES11 (Suse Linux Enterprise 11) and with OpenSUSE kernels.
Patches exist up to 2.6.31 currently.
http://wiki.xensource.com/xenwiki/XenDom0Kernels
Those patches also work for domU, not only dom0.
-- Pasi
2:56 p.m.
2009/11/8 Pasi Kärkkäinen <pasik(a)iki.fi>:
Jeremy's git tree has pv_ops pcifront now.
Yes, I noticed that slip in this week, good news by itself, even
better news if it's eventually deemed clean enough to go upstream ...
presumably Konrad is doing this work as part of the Oracle VM product?
Surprised (but grateful) that PV PCI is of interest to them
4:34 a.m.
On Sun, Nov 08, 2009 at 08:56:25PM +0000, Andy Burns wrote:
2009/11/8 Pasi Kärkkäinen <pasik(a)iki.fi>:
> Jeremy's git tree has pv_ops pcifront now.
Yes, I noticed that slip in this week, good news by itself, even
better news if it's eventually deemed clean enough to go upstream ...
presumably Konrad is doing this work as part of the Oracle VM product?
Surprised (but grateful) that PV PCI is of interest to them
Yeah it is from Oracle.. dunno if it's Oracle VM work or just his free
time project :)
-- Pasi
5:12 a.m.
On 4 May 2009 19:51, Andy Burns <fedora.lists(a)burns.me.uk> wrote:
2009/5/1 Jim Lutz <jelutz(a)config.com>:
> a CentOS 5.3 Dom0 with Xen 3.1.2, some PV CentOS DomU's with
> PCI passthrough (same versions), and I'm
> trying to implement a F10 DomU with PCI passthrough.
There has been no pcifront in recent Fedora kernels
Sorry to revive such an old thread, I've been a bit removed from
Fedora under Xen for a couple of years, do any of the more recent
Fedora releases now include pci-front?
6:42 p.m.
On Tue, 12 Oct 2010, Andy Burns wrote:
On 4 May 2009 19:51, Andy Burns <fedora.lists(a)burns.me.uk>
wrote:
> 2009/5/1 Jim Lutz <jelutz(a)config.com>:
>
>> a CentOS 5.3 Dom0 with Xen 3.1.2, some PV CentOS DomU's with
>> PCI passthrough (same versions), and I'm
>> trying to implement a F10 DomU with PCI passthrough.
>
> There has been no pcifront in recent Fedora kernels
Sorry to revive such an old thread, I've been a bit removed from
Fedora under Xen for a couple of years, do any of the more recent
Fedora releases now include pci-front?
No. Custom kernels with patches from the xen/stable-2.6.32.x will have it,
and it is hoped that it might be accepted into the upstream 2.6.37 kernel
so F15 or F16 might have it at some point in their life cycle, but it
is unlikely that F13 or F14 will ever have it.
Michael Young
7:16 p.m.
On 13 October 2010 00:42, M A Young <m.a.young(a)durham.ac.uk> wrote:
Custom kernels with patches from the xen/stable-2.6.32.x will have
it,
and it is hoped that it might be accepted into the upstream 2.6.37 kernel so
F15 or F16 might have it at some point in their life cycle
Thanks, I'll keep an eye out to see if Konrad's pvops pci patches get
pulled when the 2.6.37 merge window opens, might be time to hop onto
rawhide again ...
3:08 a.m.
On 13 October 2010 01:16, Andy Burns <fedora.lists(a)burns.me.uk> wrote:
I'll keep an eye out to see if Konrad's pvops pci patches
get
pulled when the 2.6.37 merge window opens
Encouraging to see the Stefano/Konrad/Jeremy's pvhvm code appearing in
linux-next, fingers crossed ...
7:33 a.m.
On Thu, Oct 21, 2010 at 09:08:53AM +0100, Andy Burns wrote:
On 13 October 2010 01:16, Andy Burns <fedora.lists(a)burns.me.uk>
wrote:
> I'll keep an eye out to see if Konrad's pvops pci patches get
> pulled when the 2.6.37 merge window opens
Encouraging to see the Stefano/Konrad/Jeremy's pvhvm code appearing in
linux-next, fingers crossed ...
Xen PV-on-HVM support is included in upstream Linux 2.6.36 !
Xen pcifront (and maybe basic dom0 without backend drivers)
should make it into 2.6.37.
-- Pasi
5:40 a.m.
On 22 October 2010 13:33, Pasi Kärkkäinen <pasik(a)iki.fi> wrote:
Xen PV-on-HVM support is included in upstream Linux 2.6.36 !
I was referring to the name of stefano's git tree, he's now labelled a
linux-next branch.
Xen pcifront (and maybe basic dom0 without backend drivers)
should make it into 2.6.37.
`
Yes, as noted, it would be nice to run Fedora as my mythbackend (with
tuners) in a domU again (my server has IOMMU chipset but no BIOS
support so HVM with PCI passthrough not an option) I'm currently stuck
on CentOS5 and have to build my own dom0 and domU kernels.
As I understand it, the "initial domain" stuff would help pave the way
for a pvops dom0, but in itself wouldn't be useful to end users - I
suspect that might still get a rougher ride
6:07 a.m.
On Sat, Oct 23, 2010 at 11:40:13AM +0100, Andy Burns wrote:
On 22 October 2010 13:33, Pasi Kärkkäinen <pasik(a)iki.fi>
wrote:
> Xen PV-on-HVM support is included in upstream Linux 2.6.36 !
I was referring to the name of stefano's git tree, he's now labelled a
linux-next branch.
> Xen pcifront (and maybe basic dom0 without backend drivers)
> should make it into 2.6.37.
`
Yes, as noted, it would be nice to run Fedora as my mythbackend (with
tuners) in a domU again (my server has IOMMU chipset but no BIOS
support so HVM with PCI passthrough not an option) I'm currently stuck
on CentOS5 and have to build my own dom0 and domU kernels.
As I understand it, the "initial domain" stuff would help pave the way
for a pvops dom0, but in itself wouldn't be useful to end users - I
suspect that might still get a rougher ride
"Full dom0" support will be added in small steps.
So that "initial domain" stuff is the important "first step".
Like said it doesn't (yet) contain any Xen backend drivers,
so you can boot the computer to dom0, but can't launch any other guests etc.
Then again it's pretty easy/simple to maintain those backend drivers out-of-tree,
or just add them to the upstream kernel yourself, until backend stuff is merged.
And yes, there will be rough edges, just like with any bigger merge.
-- Pasi
4933
days inactive
5475
days old
11 comments
4 participants
participants (4)
-
Andy Burns -
Jim Lutz -
M A Young -
Pasi Kärkkäinen