Re: Linux kernel headers

[Mike Frysinger <vapier at gentoo dot org>]

On Thursday 02 July 2009 09:06:01 Petr Baudis wrote:
> On Thu, Jul 02, 2009 at 07:42:26AM -0500, Ryan Arnold wrote:
> > On Thu, Jul 2, 2009 at 6:49 AM, booleandomain wrote:
> > > Third: can I specify --enable-kernel=2.6.30 even if my host os run
> > > version 2.6.29? or should I stick to 2.6?
> >
> > Using plain old 2.6 is way too old.
> >
> > You _may_ be able to get away with 2.6.30.  We try to make sure that
> > GLIBC doesn't use version specific facilities without asking the
> > kernel if they're available but you can't always query for every
> > feature.  It is guaranteed that if you specify 2.6.20 you won't get
> > any facilities provided by 2.6.29.  If you specify 2.6.30 your running
> > GLIBC may try to determine if the kernel on the system is really a
> > 2.6.30 system before using a 2.6.29 facilitiy.
>
> Actually, it's the other way around I think.

it is

> --enable-kernel is the _minimal_ version glibc supports. It means that
> it assumes that features present in this kernel version are present and
> it does not waste cycles checking their presence and having
> compatibility code compiled in. So, if you specify 2.6.30, your glibc
> may not work with 2.6.29, but if you specify 2.6.29, your glibc will
> certainly work with 2.6.30 (albeit slightly less efficiently in theory).

assuming some aspect of the ABI changed between .29 and .30 ... every once in 
a while we get a kernel update without any ABI changes ;)

> I actually plan to do some benchmarks on how various settings affect
> this since for historical reasons, in SUSE we still use
> --enable-kernel=2.6.4. If you know about any existing benchmarks,
> I would appreciate that. :-)

i'm not aware of any, but i too would be interested in such info.  we default 
to 2.6.9 in Gentoo still ... i recall there being NPTL related changes between 
2.6.8 and 2.6.9 which is why we bumped it.
-mike

Attachment: signature.asc
Description: This is a digitally signed message part.