[v7] Add udmabuf misc device

[Tomeu Vizoso]

On 09/14/2018 08:37 AM, Gerd Hoffmann wrote:
>    Hi,
> 
>>> Well, no.  This is *not* about 3D, it's about software rendering, for
>>> example cairo doing its work for gtk apps.  So the workflow would be
>>> along these lines:
>>>
>>> (1) guest app allocates dumb drm buffer from virtio-gpu, renders to it.

Why not let clients keep allocating memory for buffers with memfd?

The guest proxy would then create a virtio-gpu resource to wrap the shm 
pool memory.

The VMM would receive a number of physical addresses that can be used to 
create a dmabuf. The would place the new FD in the wayland protocol stream.

>>> (2) guest app passes the buffer to wayland guest proxy (which looks
>>>      like a wayland server/compositor to the app, but it doesn't actually
>>>      composite anything).
>>> (3) wayland guest proxy passes buffer handle to wayland host proxy.
>>> (4) qemu can then use the buffer handle to lookup the virtio-gpu
>>>      buffer, then use udmabuf to create a host dma-buf for it.
>>> (5) host dma-buf can be passed to host wayland server for display, so
>>>      guest app window shows up seamlessly on the host.
>>>
>>> Details of the wayland protocol proxying are not hashed out yet.
>>
>> Thanks for the clarification.  With dumb buffers, I guess the host can
>> use DRM_FORMAT_MOD_LINEAR when sending the udmabuf-created buffer to
>> the display.  Note there are certain cases where tiled buffers are
>> map-able (Intel), and with some variation of
>> virtio_gpu_resource_create_coherent, we could expose that to the
>> guest.
> 
> The coherent mapping (host-allocated resources into guest address space)
> is another thing which needs to be sorted out.
> 
>> That's the direction we're interested in going, though it
>> looks like udmabuf is orthogonal to that.
> 
> Yes, udmabuf is the other way around, guest allocates resources and we
> make them available as host dmabufs.
> 
>> What details on wayland protocol proxying still need to be worked out?
>>   That's one component of the ChromiumOS solution (virtio_wl) that
>> hasn't been up-streamed yet.  That method maps guest fds to host fds.
> 
> Tomeu Vizoso (Cc'ed) was looking into using virtio-serial as wayland
> protocol transport between host and guest.  With udmabuf we can use
> virtio-gpu drm objects to pass pixel buffers from guest to host, which
> is one important building block for that.

The approach that was agreed by everybody was to add ioctls to virtio-gpu 
to send and receive protocol messages. The guest proxy would use those 
ioctls and the VMM would play as the host proxy and would use similar 
virtio commands. There's code for that already and someone else from 
Collabora is to retake the upstreaming effort at some point soon.

Regards,

Tomeu