kernel versionning problem

Josh Boyer jwboyer at jdub.homelinux.org
Tue Jan 23 01:57:46 UTC 2007


On Mon, Jan 22, 2007 at 07:12:29PM -0500, Chuck Anderson wrote:
> On Mon, Jan 22, 2007 at 05:19:24PM -0600, Jason L Tibbitts III wrote:
> > >>>>> "TE" == Tanguy Eric <eric.tanguy at univ-nantes.fr> writes:
> > 
> > TE> http://rt2x00.serialmonkey.com/phpBB2/viewtopic.php?t=2965
> > 
> > Blah, blah, only breaks on Fedora, blah blah, Fedora people don't know
> > how to behave.
> > 
> > So exactly what is it that Fedora is doing wrong here?
> 
> The problem is that the rt2x00 driver assumes that there is an API 
> difference between kernels < 2.6.20 and >= 2.6.20.  In truth there is 
> no guarantee of a stable API at all in Linux, and that keying off 
> LINUX_VERSION_CODE is next to useless.  If you patch the kernel at 
> all, what version does it become?  Perhaps one API changed to be like 
> Linus' 2.6.20 due to your patch(es), but another API didn't change and 
> is still like Linus' 2.6.19.  An out-of-tree driver that keys off 
> LINUX_VERSION_CODE can never get this right.  You could change the 
> version code to fix one out-of-tree driver, but then that breaks the 
> other.

Exactly.  This is something external driver maintainers just have to deal
with if their driver isn't upstream¹.

josh

¹ Have is said the word "upstream" enough in this thread?  No?  Repeat after
me... "Upstream, Upstream, it's our way.  Upstream, Upstream anything else
will ruin your day."²

² I apologize for my crappy cheer.




More information about the fedora-devel-list mailing list