[Spice-devel] [RFC PATCH spice-common v2 02/20] A version 2 of the MousePosition message

[Gerd Hoffmann]

  Hi,

> Note also that under Linux currently all these protocol limitations allows
> to support only a single device with multiple output, which is a problem for
> vGPU. Under Windows we support multiple devices but all devices are limited
> to one monitor. Extending Linux using the same limitation requires currently
> that all monitor output would be mapped to a single DisplayChannel surface
> (as currently only surface 0 can be primary) with all monitor sub-area of
> the single surface.

> The complexity of this just for the above formula is not worth.
> Extending the Windows idea (also in Linux) would require to have
> single monitor for each DisplayChannel (so QXL devices has to be
> limited to one output) so potentially we need to map a single device
> (like vGPU) to multiple DisplayChannels (each with one monitor).

Oh, btw, that exists today.  virtio-gpu with multiple monitors works
that way.  One pci device, one display channel per output.  And, yes,
multihead mouse works fine in that setup.

I think the fundamental problem you have is that the guest agent simply
has no reliable way to figure which xrandr output is which display
channel.  And I fail to see how the monitor_id is going to solve that.

cheers,
  Gerd