Am Thu, 2 Mar 2023 18:21:21 +0200
schrieb Andy Shevchenko <andriy.shevchenko@xxxxxxxxxxxxxxx>:
> On Thu, Mar 02, 2023 at 04:58:24PM +0100, Henning Schild wrote:
> > Am Thu, 2 Mar 2023 17:46:54 +0200
> > schrieb Andy Shevchenko <andriy.shevchenko@xxxxxxxxxxxxxxx>:
> > > On Thu, Mar 02, 2023 at 09:40:09AM +0100, Henning Schild wrote:
> > > > Am Wed, 1 Mar 2023 19:28:12 +0200
> > > > schrieb Andy Shevchenko <andriy.shevchenko@xxxxxxxxxxxxxxx>:
> > > > > On Wed, Mar 01, 2023 at 06:02:14PM +0100, Henning Schild
> > > > > wrote:
>
> ...
>
> > > > > > + .driver = {
> > > > > > + .name = KBUILD_MODNAME,
> > > > >
> > > > > Strictly speaking this is an ABI (as something may
> > > > > instantiate the driver from the user space or elsewhere by
> > > > > this name. At the same time this may change with the file
> > > > > name change.
> > > > >
> > > > > Personally I prefer explicit string literal here.
> > > >
> > > > Switching from one module to three the names have to change.
> > > > People who explicitly loaded the old module which supported
> > > > multiple machines, will now how to load either both oŕ know
> > > > which one to load.
> > >
> > > Wait, are you telling that now users load modules _manually_?!
> >
> > No i am not, the modules all load automatically. I was trying to
> > construct a hypothetical case where the name change could affect
> > users doing unexpected things.
> >
> > > > I personally think the ABI change is acceptable, the assumption
> > > > would be that the drivers load automatically anyhow. And since
> > > > there are no params i doubt users will have /etc/modprobe.d/ or
> > > > /sys/module/ stuff around.
> > > >
> > > > And with the split i guess an ABI change can not be fully
> > > > avoided. Whether the names is explicit or implicit is another
> > > > discussion and just a matter of style. I prefer to stay with
> > > > the currently used pattern, it is not un-common in the kernel.
> > > >
> > >
> > > The problem with that pattern is possible, while unlikely,
> > > renaming of the file which triggers this to be updated.
> > >
> > > Under sysfs the folder will change its name. If user has a script
> > > relying on this, it will be broken. So, I prefer mine.
> >
> > Yes, the name of the module will change ... and the location of
> > module metadata and params in sysfs, both not a big deal here.
> > Because there are no params, and there is not need to modprobe
> > manually.
>
> It's not only one folder AFAIU. Also folder in the drivers will change
> its name. Parameters is one thing, the folder presence is another.
>
> Yes, the case is quite unlikely to happen (to break anyone's setup)
> that's why I started this with 'Strictly speaking...'. So, seems you
> are staying on your side, I will leave this to maintainers. If they
> are fine, I will have no objections.
We are splitting one module into three, so we will end up with three
names. Or i miss something here.
The only thing one could talk about is whether that string should be
hardcoded or derived from the name of the c-file.
Anyhow, thanks and we should probably just wait what others have to say.
Henning
>
> > > > > > + },
>
