> On Wed, Jan 05, 2000 at 04:55:22PM -0800, William J. Earl wrote:
> > Ralf Baechle writes:
> > > A system with the XL graphics will never have a real working framebuffer.
> > Since the real graphics framebuffer is in memory which is not
> > addressable
> > by the processor, the only way to fake a CPU-addressable framebuffer is
> > to reserve a chunk of main memory, and then DMA the contents into the
> > real framebuffer when the CPU-addressable framebuffer is changed (or
> > every vertical refresh interval, if there is no way to tell when the buffer
> > changes). You could probably play with the PTE valid and mod bits to detect
> > when pages are changed. It would in any case be relatively inefficient
> > compared to using the graphics pipeline as intended, since uncached writes
> > to the graphics pipeline are pretty cheap (better than cached or uncached
> > writes to large areas of main memory).
> XFree86 has builtin support for this mode of operation in the latest
> development snapshot (3.9.17). They call it ShadowFB. Of course that
> would only work in X, not on console. Apparently it's fairly easy to
> write a driver that supports ShadowFB; all you have to be able to do
> is update a rectangular area of the screen on demand.
Yeah, and that's what I tried to get working but I ran into other problems with
the dynamic loading of X modules and stuff so I never got the chance to
concentrate on the driver. I have the early stages of a driver around though.