March 1, 2010
5:09 p.m.
Am 01.03.2010 17:00, schrieb Stefan Reinauer: > A great unification would also be to stop having a different target for > FAM10 and non-FAM10 boards. This is more complicated for users than a > bit of duplicated code. (And it will bite us again for Fam10 + x, and > for Nehalem and co, basically for every chipset with a memory controller > in the CPU instead of on the board) > > So how do we fix our device model, which still assumes a board has > exactly one choice of memory controller? How about: - Make memory controllers drivers - Add detection function - Allow several of them to be compiled in The only thing required then is unified CAR, so we're able to get to the point where multiple drivers could be considered - which exists for K8/Fam10. If we're going to do runtime linking anyway, we could load the drivers from CBFS, instead of doing our current flavor of linker magic to get a list of drivers in the ramstage (I better get into hiding for this, I guess) Patrick