On Thu, 16 Jul 2020, 17:01 Jan Beulich, <jbeul...@suse.com> wrote: > On 16.07.2020 16:42, Roger Pau Monné wrote: > > On Thu, Jul 16, 2020 at 01:48:51PM +0200, Jan Beulich wrote: > >> On 16.07.2020 13:41, Roger Pau Monné wrote: > >>> On Wed, Jul 15, 2020 at 12:15:10PM +0200, Jan Beulich wrote: > >>>> Use ENXIO instead of EINVAL to cover the two cases of the address not > >>>> satisfying the requirements. This will make an issue here better stand > >>>> out at the call site. > >>> > >>> Not sure whether I would use EFAULT instead of ENXIO, as the > >>> description of it is 'bad address' which seems more inline with the > >>> error that we are trying to report. > >> > >> The address isn't bad in the sense of causing a fault, it's just > >> that we elect to not allow it. Hence I don't think EFAULT is > >> suitable. I'm open to replacement suggestions for ENXIO, though. > > > > Well, using an address that's not properly aligned to the requirements > > of an interface would cause a fault? (in this case it's a software > > interface, but the concept applies equally). > > Not necessarily, see x86'es behavior. Also even on strict arches
it is typically possible to cover for the misalignment by using > suitable instructions; it's still an implementation choice to not > do so. I am not sure about your argument here... Yes it might be possible, but at what cost? > > Anyway, not something worth arguing about I think, so unless someone > > else disagrees I'm fine with using ENXIO. > > Good, thanks. > -EFAULT can be described as "Bad address". I think it makes more sense than -ENXIO here even if it may not strictly result to a fault on some arch. > Jan >
