On Fri, Apr 30, 2004 at 07:57:20AM +1000, Benjamin Herrenschmidt wrote:
> 
> > > Finally, the geometry of screens is something that has nothing to
> > > do in the kernel imho.
> > 
> > I don't like micro kernel designs. This is way micro kernels handle this 
> > problem.
> 
> This has nothing to do with microkernel design and everything to do with
> not bloating the kernel with things that don't to be there.

I would like the kernel to handle CRTCs and overlays in the same way, and 
that way IMHO should be to calculate register values in userspace and just 
have a few ioctls to have the kernel write them.

Set all registers ioctl:
Nothing special needed here I think.

Set buffer offset (pan/flip) ioctl:
This one should return the current vertical blank counter. It's needed for 
proper triple buffering. It should also be able to choose when to flip 
(immediately or vblank, on top/bottom/any field).

Wait for vertical blank ioctl:
This one should also return the current vblank counter. It should also be 
able to wait for absolute and relative vblank counter values. Waiting for 
relative=1 would insure you're at the vblank when it returns and waiting 
for absolute=last_flip_counter_value+1 would help with double buffering.


The main reason for userspace register calculation is that there are a lot 
of knobs in some hardware (especially with overlays). Trying to abstract 
those without loosing any features would be a problem. And even the API 
would cover all of todays graphics chip something new might come along in 
the future and the API might become too limited.

-- 
Ville Syrjälä
syrjala@sci.fi
http://www.sci.fi/~syrjala/


-------------------------------------------------------
This SF.Net email is sponsored by: Oracle 10g
Get certified on the hottest thing ever to hit the market... Oracle 10g.
Take an Oracle 10g class now, and we'll give you the exam FREE.
http://ads.osdn.com/?ad_id149&alloc_id66&op=click