On Fri, 2009-11-27 at 01:16 +0100, Segher Boessenkool wrote:

In all code I have done for the XFB, I map it like any other RAM and
dcbst it after writing to it.  Maybe that is slower though, WIMG=0100
might be better.  Dunno. 

Well, it depends also what you want to do with it. If you want to expose
it as a standard linux fbdev, it might break apps to expect them to use
dcbst... but it will make rmw cycles a lot faster and avoid wasting
memory with shadowfb.

It's going to mostly depend on what kind of HW accel we can get out of
it, if we get good enough accel we can probably got for I=1 like a
standard fb

Ben.