kernel modules/kmods/dkms (Re: Plan for tomorrows (20070906) FESCO meeting)

[Jon Nettleton]

On 9/7/07, David Woodhouse <dwmw2 at infradead.org> wrote:
> On Fri, 2007-09-07 at 11:24 +0200, Thorsten Leemhuis wrote:
> > dwmw2's proposal afaics mainly reads as "no
> > separately packaged *kernel-modules* (in source or binary form) in
> > Fedora at all"; but he uses the term "kmods" here and there (and kmod
> > specific examples), so some people afaics got the idea that something
> > else (like dkms) would be acceptable. I doubt that's the intention
> > behind dwmw2's proposal. David, can you clarify? Or Jesse (who's listed
> > as owner for the proposal as well)?
>
> Indeed it was not the intention -- I used the term 'kmod' to refer to a
> generic evil. I have clarified the wording now.
>
> Not only do I think we shouldn't ship modules in binary form, I think we
> shouldn't be shipping them in source form as dkms payload either.
>
> I've no particular objection to shipping dkms itself, just as I have no
> objection to shipping the kernel-devel package -- I just don't think
> there's any justification for shipping 'dkms payload' packages as part
> of Fedora.
>

But do you have an objection to including dkms hooks in kernel rpms or
existing scripts called from pre/post install scripts?

There are lots of examples but my specific one is the gspca driver
that I built the first dkms package for and is exists at freshrpms.
This driver supports a lot of modern webcams, but the driver developer
has no intention of even trying to have this code merged into the
kernel.  What are we to do in this circumstance?

I think we need something better than the "safe" ostrich syndrome; if
it is out of the kernel it scares us and we run away from it.  There
is plenty of legitimate driver development being done outside of the
kernel.  It just seems a bad decision to completely bury our heads in
the sand to this code-base.

Of course this is coming from someone that has 5 critical modules ( to
me ) running and upgraded through dkms.

-Jon