KVM Support?

[Avi Fishman]

On Mon, Aug 27, 2018 at 7:24 PM Eddie James <eajames at linux.vnet.ibm.com> wrote:
>
>
>
> On 08/27/2018 10:59 AM, Tanous, Ed wrote:
> >> Do you have any plan to support non-jpeg video stream in obmc-iKVM?
> > I'm not sure about Eddie, but I don't think we have  a need for non-jpeg kvm.  With that said, everything is just running RFB underneath, so basically anything should be possible that's possible over VNC.
>
> Right, we also don't have any need for non-jpeg at the moment.

note that Nuvoton HW support Hextile compression so this should stay
on the table.

>
> >
> >> It is possible some platforms don't have hardware JPEG engine, instead have
> >> the RGB video stream.
> >> If possible to support RGB video stream, I think it would be better if you can
> >> add the video frame differentiation in obmc-iKVM because we don't need to
> >> send the full screen to VNC client on every request, it can reduce the buffer
> >> size at least. (libvncserver already has the video frame differentiation)
> > While I agree it would be nice, I don't think it's a priority 1 for implementation (at least for us).  Our current AST2500 products send full buffer updates for every frame, and while in theory it's less efficient, we haven't really gotten any complaints.  With that said, if differential mode is available and works well, I'm sure we'd use it.
>
> The problem with differential compression is that you have to do that
> decoding work on the BMC so that you update the right part of the
> framebuffer. I didn't get that working but might be some performance
> concerns there... So same for us, it's not a priority right now.
>
Nuvoton chip does support tiles differerential by HW so for us it
doesn't have a performance effect nor high CPU utilization, so again
this should option should remain.

> >
> >> Also, I have implemented a KVM application for our platform(nuvoton evb-
> >> npcm750) which has hardware video frame differentiation and hardware
> >> hextile engine.
> >> https://github.com/Nuvoton-Israel/obmc-ikvm
> >> https://github.com/Nuvoton-
> >> Israel/linux/commit/7e90f572323e56e9425b85aafe86dfa5a35c4e7d
> >>
> > Cool.
> >
> >> I also enabled ED's KVM webui and added kvmws proxy in python web
> >> server.
> >> https://github.com/Nuvoton-Israel/meta-openbmc-nuvoton-
> >> addon/commit/140771b7e898dd5ca8f0364290cce669609088ac
> >>
> > I've got patches for bmcweb as well that I can get posted today.  I was really hoping to have a discussion around how we want to implement the web portion beforehand, but it looks like you've already done a lot of work, which is great.
> >
> >> Regarding the larger binary size, I think we need to review all configurations
> >> of libvncserver and check which functions should be disabled.
> >>
> > I think the most important question here is how small of a binary we can get out of libvncserver, and whether or not we can afford to have an unauthenticated socket open with full control of the host system.
>
> All good points. These aspects haven't been a priority for me yet, and
> I'm really not an expert with security or bmcweb, but definitely worth
> some discussion.

>From our point of view it is OK to remove some libvncserver features
like security or so as long as the hextile compression and frames
differential for video stays to be used by our HW acceleration.

>
> Thanks!
> Eddie
>
> >
>


-- 
Regards,
Avi