On Wed, Aug 14, 2019 at 08:10:59AM +0200, Christophe Leroy wrote: > > Note that while a few other architectures have a magic hack like powerpc > > to make ioremap work before vmalloc, the normal practice would be > > to explicitly use early_ioremap. I guess your change is fine for now, > > but it might make sense convert powerpc to the explicit early_ioremap > > scheme as well. > > > > I've been looking into early_ioremap(), but IIUC early_ioremap() is for > ephemeral mappings only, it expects all early mappings to be gone at the end > of init.
Yes. > PPC installs definitive early mappings (for instance for PCI). How does that > have to be handled ? Good question, and no good answer. I've just been looking at a generic ioremap for simple architectures, and been finding all kinds of crap and inconsistencies, and this is one of the things I noticed.
