EGL_MESA_choose_display version 1

[Jon Smirl]

On 6/14/05, Adam Jackson <ajax at nwnk.net> wrote:
> On Tuesday 14 June 2005 17:37, Jon Smirl wrote:
> > I'm looking at a future model where where there is one DRM/fb device
> > per screen. fb is currently one per screen and DRM is one per card.
> > I'll modify DRM to expose one per screen (these will all be pointing
> > to the current DRM code there will just be multiple names). By having
> > one fb/DRM device pair per screen we are opening up a path for having
> > two independent user logged onto the two devices. We need two DRM
> > device nodes since they will have different ownership permissions.
> >
> > There is no problem taking the list of card0,card1,etc and converting
> > it to display/0, display/1,etc and then back again in the driver. But
> > I do need these to represent screens not devices. Display is ambiguous
> > as to whether it is a screen or device. Would screen/0, screen/1 be
> > better?
> 
> I thought we had decided the semantics were one card : one display : N
> screens, and one EGLDisplay could hold one or multiple screen surfaces.

EGL can't tell the difference between (one card : one display : 2
screens) and (one card : two display : 1 screen each).  In the second
arrangement two independent users can be on each head if the drivers
are written correctly.

The complete plan is for the driver to morph the configuration
depending on what PAM assigns you. If PAM assigns you one screen you
see: (one card : one display : one screens). If you get assigned two
screens you see (one card : one display : 2 screens). At the EGL API
level it doesn't know this is happening.

My goal is to make sure multiuser configurations are allowed from the
beginning and that we don't make any assumptions that will break
multiuser. For example the DirectX 10 spec requires the cards to be
able to multitask the DMA queues. That makes it pretty easy to
implement one user per head if the RAM is split in half.

-- 
Jon Smirl
jonsmirl at gmail.com