[arch-dev-public] Minimal kernel version

[Stéphane Gaudreault]

Le 5 mai 2011 14:20:31, Dan McGee a écrit :
> On Thu, May 5, 2011 at 12:58 PM, Tom Gundersen <teg at jklm.no> wrote:
> > Hi guys,
> > 
> > As far as I understand the minimal kernel version we support is
> > 2.6.27. This is at least due to glibc and udev.
> > 
> > I thought it would be worth noting that as of its next release udev
> > will require kernel version 2.6.32:
> > <http://git.kernel.org/?p=linux/hotplug/udev.git;a=commit;h=67a77c8bf299f
> > 6264f001677becd056316ebce2f>.
> > 
> > Maybe this could be an opportunity to raise the requirement across the
> > board, and also to make a news item about this?
> 
> Not to be the thorn in the side of progress, but we can't keep
> changing this every six months on a whim unless we truly don't care
> about some semblance of stability and uptime. We last changed this
> less than 5 months ago[1], and before that, 21 months ago[2]. I think
> there needs to be an extremely valid reason someone can't have a
> machine with uptime approaching a year in duration. I'm sure there are
> people running pre-2.6.32 kernels out there even though we don't
> currently package one- it was only released in December 2009, so 18
> months ago.
> 
> So obviously if you update udev on your system, you should be expected
> to run a kernel that satisfies the requirement, but I don't think we
> should force people to jump from the 2.6.27 level we established only
> 5 months ago just yet if at all possible.
> 
> -Dan
> 
> [1]
> http://www.archlinux.org/news/minimum-required-kernel-version-increased-1/
> [2] http://www.archlinux.org/news/udev-minimum-kernel-version/

I might be wrong on this because I do not have all the information, but I 
think that one of the consequence of our choice to set minimum version to 
2.6.27 is that we need to patch the glibc to avoid a threading issue on 
x86_64[1]. This problem do not seems to happen if glibc is builded for 2.6.32.

Again, I do not have all information on this. Allan might have a more informed 
opinion on this issue.

Stéphane

[1] http://sourceware.org/bugzilla/show_bug.cgi?id=12403