On Wed, Oct 27, 2004 at 11:55:38AM -0700, ext Stephen Hemminger wrote:
> On Wed, 27 Oct 2004 21:38:04 +0300
> "Teras Timo (EXT-YomiGroup/Helsinki)" <Ext-Timo.Teras@xxxxxxxxx> wrote:
> > This way the problem is that I have to know which devices I will
> > add the attributes. But the point is to add attributes to
> > all netdevs.
> > Using this approach I'd have to enumerate all the interfaces every
> > now and then.
> > If I have my class interface I get a callback whenever an
> > interface is created or deleted and I can automatically add the
> > attribute to all netdevs.
> If you are doing something to all interfaces, then it probably should
> either be part of the common net-sysfs layer, or you can dynamically
> discover and do it by following device transitions with
I'd like my code to be available as module also so common net-sysfs
is not an option. Hmm... I could use device transitions hooks.
But I'm still a bit puzzled why the interface registration should not
After all it is only a mechanism to get callbacks whenever a class
device is created or removed. Why to even implement a class interface
system if you are not allowed to use it (except in one special place
where it is considered to be an ugly hack by Christoph Hellwig)?
I understood that the whole point for interfaces is the ability to
extend functionality without core (i.e. common net-sysfs) changes.
Why to lose that modularity?