On Thu, Jul 27, 2006 at 12:56:34PM +0100, Jeremy Sanders wrote: > What I don't understand is why Nvidia feel the need to use a 8MB (!!!!) > kernel module in the first place. Is there a reason why it needs to be > there, and not in X.org like all the other graphics drivers? The source-level compatibility with windows is I'm sure the main reason, but don't forget that where to split the rendering between userspace and kernelspace is a very open and unsolved technical problem. If you get close to the actual packets exchanged with the card, the kernel part gets thinner somewhat, but the complexity and CPU cost of the security checking of the data, locking and scheduling goes through the roof. OTOH, the closer you get to the 2D/3D APIs the more of the software fallbacks you have to put in your driver, i.e. you end up with Mesa and XAA/EXA in the kernel. Or you can do something somewhat in the middle, with various degrees of success. BTW, DRI/DRM can theorically at least cover any of these choices if I understand it correctly. It is somewhat agnostic to where you split the pipeline. OG. -- fedora-devel-list mailing list fedora-devel-list@xxxxxxxxxx https://www.redhat.com/mailman/listinfo/fedora-devel-list
