Re: [libvirt-users] xen and libvirt
On 1/5/19 2:32 AM, Christoph wrote:
Hi
I need the e820_host option for a passthrough of an digitaldevices dvb-s2 card.
Without this I'cant initialize the card in the domU.
The xl.cfg man page claims the option is set to 'true' if any PCI passthrough
devices are configured, 'false' otherwise. It is also advised to set the option
if no PCI passthrough devices are configured but hotplug is later expected. I
wonder if there are any downsides to unconditionally setting this option to
'true' for PV domains?
On an all my domU's with high IO load like an file server I need to set
max_grant_frames = 64 or higher. Without this I get a lot of messages on
dom0/domU and kernel exceptions.
I've read it should be ok to set it = 256 on all domU's (somewhere on SuSE
pages), so if there is a possibility to set it global on hyperviser, this could
be a solution for me.
With xl.conf it is possible but libvirt doesnt look there for the settings. I've
tried to set it in libxl.conf, but this doesnt work to.
Currently max_grant_frames is not supported in libxl.conf, but that sounds like
a good place to start. Per-domain adjustment of the setting can be done later if
we find it necessary. Do you have any interest in providing a patch to support
setting max_grant_frames in libvirt's libxl.conf file? Commit 5194eb95646 could
be used as an example
https://libvirt.org/git/?p=libvirt.git;a=commit;h=5194eb956464b0a7f848e2e...
Regards,
Jim
Am 2019-01-04 18:42, schrieb Jim Fehlig:
> On 1/4/19 9:14 AM, Christoph wrote:
>> Hi
>>
>> Does someone know how it is possible these parameters:
>>
>> e820_host
>> max_grant_frames
>>
>> to set with libvirt xml config?
>
> Currently, those settings are not supported by libvirt. Do you need to
> adjust them for PCI passthrough to PV domains?
>
> I took a quick peek at xl.cfg(5) and based on the description of those
> settings didn't see a counterpart in the libvirt domain schema.
> Perhaps understanding use-cases will help with modeling those settings
> in libvirt domain XML.
>
> Regards,
> Jim