On 12 December 2014 at 13:39, Laszlo Ersek <ler...@redhat.com> wrote: > I considered "locking down" the new interface, and preventing callers > from passing in any IO ports when they want a wider MMIO data register. > I didn't do that because someone might want a wider ioport mapping at > some point (although no current such user exists and I couldn't name > what the advantage would be in it). Unless you use the combined thing, > the wide data register should work with the ioport mapping too. > > The combined case I thought to leave simply undefined. > > If you want, I can set an error, but then I'd prefer to prevent callers > from passing IO ports through the new data_memwidth-taking functions.
Yeah, I don't think we need to make the combined case work, but it does seem worth at least making it fail cleanly if anybody tries it, rather than silently doing the wrong thing. -- PMM
