On Tue, 2008-06-10 at 15:30 +0100, Paul Durrant wrote: > Sebastien Roy wrote: > > > > This is not necessary, however, since drivers should return ENOTSUP to > > property queries that they don't support. This could percolate up to > > dladm which could filter out unsupported properties. > > > > I guess that may be sufficient, but in cases where a kstat set needs to > be created it seems a little inelegant to have to walk all possible > stats polling for the ones that return ENOTSUP.
Yes, I agree. The kstat interface needs to be improved. I thought we were talking about link properties though. > At the moment it seems > that all stats are created but ones that are not supported get set to > zero; which is bogus since zero may be a meaningful value. Agreed. -Seb
