[Date Prev][Date Next] [Thread Prev][Thread Next] [Date Index] [Thread Index]

Re: Out-of-tree kernel module popularity



On Wed, Oct 24, 2007 at 11:32:26AM +1000, Paul Wise wrote:
> Interestingly, Fedora has a new policy that kernel module packages
> must be merged with kernel.org or removed from Fedora:

> http://fedoraproject.org/wiki/Packaging/KernelModules
> http://fedoraproject.org/wiki/DavidWoodhouse/KmodProposal

> I don't think this would be an option for Debian, but it certainly is
> a gutsy move.

Very interesting. I'd say we should atleast consider it due to various
points:

Why we should follow Fedora
------------------

+ Out of kernel modules tend to be low quality.

Either they lack the strict kernel community review, or they
have failed the review and thus are sticking outside the mainline
kernel. Do we want to expose users to buggy drivers? In kernel
space there is lots of potential for data-loss and security
bugs. Not to mention general system instability.

+ Out of kernel modules and patches get rapidly out of date

Everytime kernel team updates the kernel, a bunch of
kernel-module and kernel-patch packages break and need updating.
This is big maintainence headache, and litters the RC bug
list count.

+ Out of kernel modules are harder to use

Even with m-a, it's still an bunch of extra steps over normal
apt-get install.

+ The effort used in maintaining out of kernel modules could
  be used in something else.

For example in cleaning and helping the driver upstream
to submit the driver upstream.

Why not...
------------------

- Users will be unhappy with less "supported" hardware

This may be deceptive. The users complaining that "ubuntu
supported this hardware out of box" might have just left
ubuntu due to stability problems.

- Driver could be good, just the upstream is unjustfiedly refusing it

Alternatively the Out-of-kernel module upstream might have
whatever reasons why they want to keep their driver away
from mainline. Not being a kernel developers, it might be
hard to see if the module author or the kernel subsystem
maintainer is right about the maturity/quality of the driver.

In any case, this is more of a social than technical problem.
maintining the driver out-of-free will merely allow to prolong
the problem instead of solving it.

The Exception to the rule:
-------------------------

non-free kernel modules.

- Not mainlineble - ever
- Any quality problems will just be a practical demonstration to
  users of disadvantages of nonfree code.




Reply to: