Re: [RFC] VirtFS support on Xen

[Wei Liu <wei.liu2@citrix.com>]

On Mon, Jan 25, 2016 at 11:35:42AM +0000, David Vrabel wrote:
> On 25/01/16 11:29, Wei Liu wrote:
> > 
> > That goes back to the old topic of how to make VirtIO work on Xen (PV in
> > particular). It's not easy task by any means.
> 
> This is a (very) useful thing to solve anyway so why not go down this
> route?  Particularly, since the latest discussion had reached a
> conclusion about how to do this (white-listing virtio devices that need
> to use the DMA API, if I remember correctly).
> 

I thought about that before.

The changes made by Andy  doesn't make VirtIO work for PV guest
automatically.  To make it work for PV guest, we either need to make
virtio-mmio / virtio-pci work for PV guests (is there a way to register
QEMU as emulator for PV guests?) or develop virtio-xen transport for
VirtIO. (I think virtio-xen makes more sense fwiw)

Then inevitably we will get hit by questions with performance issues like
[0]  once we open the flood gate to allow all virtio devices to be used
on Xen and claim it is "supported".

If there is no clear plan to make VirtIO on Xen functional and
performant by default I'm afraid it's not a good way forward at the
moment.

Wei.

[0] <56701113.1080107@sinte.net>

> David