[forum] XFree86 5.0 TODO

[Alan Cox]

> >    In the new XAA, surface allocation is relegated to the driver,
> > which can use any mechanism it wants to allocate these surfaces.
> 
> But will it not mean less code sharing if each driver has to do memory
> management ?

I'd agree with the original comments here. In the kernel we've been
moving away from core code making the assumptions. It makes a mess. Instead
you need library code.

If there is a library for "generic" surface allocation you can just plug
into, then the code is normally shared but you can override things. Think
C++ virtual methods, then program them in a sane language ;)

> What kind of non-linear framebuffer is used (if you are allowed to tell
> that is) and do you use a automated conversion unit or something such ?

A documented example is the i8xx where frame buffers are pulled from
main memory via the AGP gart and scatter gather.