Re: r128/radeon drivers

Around 10 o'clock on Nov 19, Alex Deucher wrote:

> if we are starting fresh wouldn't it make sense to use the cce/cp for
> everything rather than MMIO?  that would avoid having to wait for idle
> and such when switching between MMIO and CCE.  If we have a kernel
> component, it shouldn't be be an issue to use the CCE.

Yes, a perfect Radeon driver would have a kernel component and use CCE, 
but right now we've got research above the drivers to do and hacking up 
a small user-mode driver in the simplest possible fashion seems expedient 
for the present.

> Also, are the kdrive drivers going to be a model for the new drivers or
> a interim step on the way to the "new driver model" we've been
> discussing?

I don't know; kdrive offers a lightweight framework for driver 
development, but it doesn't have a good model for splitting the various 
pieces of a video driver apart.  Obviously an abstraction of mode 
selection should be made explicit so that you can trivially select between 
fbdev, vesa or custom mode selection logic from some other agent (external 
library or whatever).

Certainly as currently implemented, kdrive is a long ways from a 'new 
driver model'; it's really just the same old user-mode drivers that we've 
seen in X servers for 15 years or more.  The only real advantage it has is 
the speed with which chip-specific accelerated servers can be implemented.

-keith