On 2015-12-24 08:55, Holger Levsen wrote:
> Hi,
>
> On Mittwoch, 23. Dezember 2015, Aurelien Jarno wrote:
> > > I have to admit, I cannot follow:
> > > - if this is fixed, why is 806911 still open?
> > The "bug" is still there, just not triggerable anymore on amd64 and
> > i386.
>
> ok
>
> > I use "bug" as when faking the kernel version to change the result
> > of versions comparisons, one should expect the result of such
> > comparisons to be wrong.
>
> Again, can't follow. Surely tests testing for kernel >= 3.0 will fail or is
> that what you ment?
Yes, it's exactly that. The glibc was configured with a minimum kernel
set to 3.2. This allows to use the new features provided by the kernel
without any compatibility code to emulate them. For that the libc first
looks at runtime that the kernel is indeed at least 3.2. This is where
it fails when using the uname26 personality, as with the default
comparison method, 2.6.40 < 3.2.
> > > - also, the hosts runs jessie and this is where we run linux64 on and
> > > from, so how are changes in sid+testing relevant in our setup anyway?
> > > (actually we run jessie, sometimes with jessie kernels and and on some
> > > other hosts with bpo kernels or even never…)
> >
> > The host might runs jessie, but from the bug report I understood the
> > bug happened in a testing or sid chroot.
>
> yes (with pbuilder chroots)
>
> > > - why did you 2.6._32_ mention at all, and not "2.6" (or maybe 2.6.56)?
> >
> > We lowered the minimum required kernel version to 2.6.32 instead of 3.2
> > on amd64 and i386. When comparing kernel versions with the uname26
> > personality, we have the following relations when the minimum kernel
> > version is 2.6.32:
> > - 3.x kernels aka 2.6.40+x > 2.6.32, this works
> > - 4.x kernels aka 2.6.60+x > 2.6.32, this works
> >
> > However when the minimum kernel version is 3.2:
> > - 3.x kernels aka 2.6.40+x < 3.2, this do not work
> > - 4.x kernels aka 2.6.60+x < 3.2, this do not work
>
> I cant follow. Probably this is because I fully expect this to happen… but
> somewhere in between I must be lost… or are you talking about build
> requirements for libc itself?
The check for at least a 3.2 kernel is something done at runtime in
ld.so, hence the "FATAL: kernel too old" message you reported.
Aurelien
--
Aurelien Jarno GPG: 4096R/1DDD8C9B
aurel...@aurel32.net http://www.aurel32.net
signature.asc
Description: Digital signature
___
Reproducible-builds mailing list
Reproducible-builds@lists.alioth.debian.org
http://lists.alioth.debian.org/cgi-bin/mailman/listinfo/reproducible-builds
