Re: [Xen-devel] dom0/pvh: Dom0 PVH with PCI passthrough support status

[Roger Pau Monné <roger.pau@xxxxxxxxxx>]

Hello,

Thanks for your interest and testing of PVH Dom0!

On Fri, Oct 19, 2018 at 04:08:45PM +0000, Vasile, Alexandru-catalin wrote:
> Hello,
> 
> I would like to use Dom0 with PVH mode together with PCI passthrough in order 
> to compare PVH and HVM in a real test scenario for early adoption of PVH. The 
> minimum setup needed is to have the possibility to assign to a domain a PCI 
> device that is passthrough.
> 
> Are there any patches in review or previous discussions on having PCI 
> passthrough in Dom0 PVH?

Current code in staging branch should have PCI passthrough support for
PVH Dom0 (ie: so that Dom0 can correctly interact with physical
devices on the host), albeit it's all experimental and things like
SR-IOV devices won't work properly yet.

You likely want to use my Xen branch at:

http://xenbits.xen.org/gitweb/?p=people/royger/xen.git;a=shortlog;h=refs/heads/fixes-pvh-v2

This contains my latest fixes for PVH Dom0, as I've just posted this
today. As for the Dom0 kernel you will have to use Linux 4.18 or
newer.

> Our internal tests showed promising results regarding performance (no PCI 
> passthrough), but nonetheless as expected in this early stage of development 
> a small decrease between 2.5 - 10%. Sacrificing performance for a boost in 
> security offered by PVH should be a priority. For the final results to be 
> conclusive and have a fair comparison, it is necessary to have a rough 
> implementation of PCI passthrough (e.g. for a network device).

If you instead mean doing PCI-passthrough from a PVH Dom0 to a PV or
HVM DomU, I haven't really tried it, so there's a good chance you will
find issues in pci-back or QEMU when attempting this.

My current main goal is to make PVH Dom0 stable, and afterwards I will
look at adding all the missing functionality.

There is some low-hanging fruit, both in Xen and the Linux kernel,
which can really be worked in parallel by different parties, so let me
know if you have some capacity and we can speak about sharing the
workload.

Thanks, Roger.

_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxxxxxxxxx
https://lists.xenproject.org/mailman/listinfo/xen-devel