[Spice-devel] [spice-server] doc: Add virgl documentation

[poma]

On 25.03.2016 10:02, Christophe Fergeau wrote:
> On Thu, Mar 24, 2016 at 01:26:17PM -0400, Cole Robinson wrote:
>> On 03/24/2016 01:08 PM, Christophe Fergeau wrote:
>>> Ah, I'll have to improve the text I guess. By "It's currently limited",
>>> I meant "Guest support is currently limited". (I tested this)
>>>
>>
>> Ah I see. So F23 guest has all the bits it needs, gotchya
> 
> Mostly, the version which went into git says that mesa bits from
> copr:kraxel are needed on f23 guest.
> 
>>>>
>>>> Slightly related question: what is the timeline for working gl passthrough
>>>> with local host + VM + spice listen=127.0.0.1 ? Is it under heavy development
>>>> or just waiting for just waiting for some new releases? If the latter, which
>>>> packages will be required?
>>>
>>> What do you expect out of it exactly ? To be able to do remote-viewer
>>> spice://localhost:5900 and have virgl acceleration work? Or something
>>> different?
>>>
>>
>> Basically the remote-viewer case you describe, since that's the default
>> virt-manager (and libvirt) config for spice setups. Gerd seemed to indicate
>> that it will work in the near term, in his response to your qemu patch. Isn't
>> that what the dmabuf stuff is all about? Or are we always going to have to use
>> the OpenGraphics/add_client method to get GL?
> 
> As I understand it, the dmabuff stuff is just a file descriptor being
> passed around between virgl/QEMU and the SPICE client through a Unix
> socket so that the 3D framebuffer can be displayed in a different
> process.
> The discussion about gl+tcp was probably not a short-term thing, and not
> limited to localhost. We need to add support for encoding the virgl
> display, and send that to the client through a TCP socket. This would
> allow remote 3d to work.
> 

"gl+tcp" and VirtualGL
http://www.virtualgl.org/About/Introduction

How the two correspond to each other, if at all?