Re: [Qemu-devel] Multi-head support RFC

[John Baboval]

On 11/06/2013 05:55 AM, Gerd Hoffmann wrote:
>    Hi,
>
> > In QEMU 1.3, there was a DisplayState list. We used one DisplayState per
> > monitor. The DisplayChangeListener has a new hw_add_display vector, so
> > that when the UI requests a second monitor the new display gets attached
> > to the emulated hardware. (patch: add_display_ptr)
> I don't think we want actually add/remove stuff here.  On real hardware
> your gfx card has a fixed set of display connectors, and I think we are
> best of mimic that.
I think that's a property of the emulated hardware. Monitors get 
connected and disconnected, and that's what the UI cares about.

> Support for propagating connect/disconnect events and enabling/disabling
> displays needs to be added properly.  Currently qxl/spice can handle
> this, but it uses a private side channel.
>
> > A new vector, hw_store_edid, was added to DisplayState so that UIs could
> > tell emulated hardware what the EDID for a given display should be.
> > (patch: edid-vector)
> Note that multiple uis can be active at the same time.
> What happened with the edids then?

This is why it seemed to me that we shouldn't have multiple 
QemuConsoles. There should be one per UI type. In my current patches, 
each DisplayState has a new DisplayType enum, so I can keep track of 
which active UI the DisplayState goes with.

As far as the EDID is concerned, there can only be one EDID for a 
display+hw pair, or the guest won't know what to do. In my use-case, I 
simply pass real EDIDs through, and create a full-screen window for each 
real monitor. If you wanted to have two UIs displaying the same 
DisplaySurface, the EDID would have to come from one of them, and the 
other would have to clip, or scale.

> > VRAM size was made configurable, so that more could be allocated to
> > handle multiple high-resolution displays. (patch: variable-vram-size)
> upstream stdvga has this meanwhile.
>
> > I don't think it makes sense to have a QemuConsole per display.
> Why not?  That is exactly my plan.  Just have the virtual graphic card
> call graphic_console_init() multiple times, once for each display
> connector it has.
>
> Do you see fundamental issues with that approach?
Currently only one QemuConsole is active at a time, so that would have 
to change.... It could certainly be done this way, but it seemed like 
more churn. Perhaps we should step back and define what we want each of 
these objects to be.

There are things in QemuConsole that we don't really need another copy 
of. There is also stuff in the QemuConsole that we don't want to 
duplicate just to have multiple displays. Like the CharDriverState.

> > I can use a model similar to what qxl does, and put the framebuffer for
> > each display inside a single DisplaySurface allocated to be a bounding
> > rectangle around all framebuffers. This has the advantage of looking
> > like something that already exists in the tree, but has several
> > disadvantages.
> Indeed.  I don't recommend that.  It is that way for several historical
> reasons (one being that the code predates the qemu console cleanup in
> the 1.5 devel cycle).
>
> > Are these features something that people would want to see in the tree?
> Sure.  One of the reasons for the console cleanup was to allow proper
> multihead support.
>
> cheers,
>    Gerd