[Rust-VMM] Crate for vhost-user device backend?
Liu, Jing2
jing2.liu at linux.intel.com
Mon Jan 21 04:39:49 UTC 2019
Hi Stefan,
On 1/18/2019 7:40 PM, Stefan Hajnoczi wrote:
> On Thu, Jan 17, 2019 at 11:00:27PM +0800, Liu, Jing2 wrote:
>> On 1/17/2019 7:32 PM, Jiang Liu wrote:
>>>> On Jan 17, 2019, at 7:30 PM, Dr. David Alan Gilbert <dgilbert at redhat.com> wrote:
>>>>>> On Jan 17, 2019, at 6:58 PM, Marc-André Lureau <marcandre.lureau at gmail.com> wrote:
>>>>>> On Thu, Jan 17, 2019 at 2:56 PM Jiang Liu <liuj97 at gmail.com <mailto:liuj97 at gmail.com>> wrote:
>> What application will be the virtio backend on your scenario?
>
> This is a technicality but an important one:
>
> vhost-user device backends are not the same as virtio device backends
>
> vhost-user device backends subsets of virtio device backends. For
> example, they may lack some virtqueues, virtio configuration space, and
> the device lifecycle is different.
>
> The way that the vhost-user protocol works today, for better or for
> worse, is that some of the virtio device backend still lives inside the
> emulator (e.g. the virtio PCI transport which controls negotiation and
> lifecycle).
Exactly! When reading codes, it is not easy to differ clearly. And
important key is the backend application.
>
> This also means that the vhost-user protocol has messages and features
> that have no equivalent in the virtio specification.
>
> It's worth keeping the distinction in mind because confusing these two
> things becomes problematic especially when the codebase supports both
> virtio and vhost-user devices :).
Yep, agreed with you very much! And thanks for your kind tips and
reminders on technical keys!
Jing
>
> Stefan
>
More information about the Rust-vmm
mailing list