[ATrpms-devel] openafs kmdls failing

[Axel Thimm]

On Fri, Sep 19, 2008 at 10:20:30AM +0200, Angel Marin wrote:
> Axel Thimm wrote:
> > On Fri, Sep 19, 2008 at 08:02:43AM +0200, Angel Marin wrote:
> >> Axel Thimm wrote:
> >>> On Thu, Sep 18, 2008 at 04:50:36PM +0200, Angel Marin wrote:
> >>>> Angel Marin wrote:
> >>>>> openafs kmdls do not build on 2.6.26 kernels without extra patches[1]. 
> >>>>> Updated src.rpm (only built and tested on F9 x86_64):
> >>>>>
> >>>>> http://dl.anmar.eu.org/tmp/openafs-1.4.7-29.src.rpm
> >>>>>
> >>>>> [1]
> >>>>> https://lists.openafs.org/pipermail/openafs-devel/2008-September/016166.html
> >>>>>
> >>>> Something must have gone wrong in build process (F9 x86_64) as 'modprobe 
> >>>> libafs' reports:
> >>>>
> >>>> libafs: Unknown symbol init_pid_ns
> >>>>
> >>>> While the locally built one for the same kernel works :?
> > 
> >> Some googling suggests similar errors have happened when .config for the
> >> build environment does not match the running kernel.
> > 
> > That would make sense, as the actual build host is running a different
> > kernel. But this is only visible in uname output - headers, symbols
> > etc. are all picked up from the target kernel.
> > 
> > Gary Buhrmaster pointed me to
> > 
> > http://bugs.gentoo.org/show_bug.cgi?format=multiple&id=232511
> 
> That find_task_by_vpid call mentioned there is what 
> STABLE14-linux-2626-support-20080608.patch [1] is supposed to ifdef out.
> 
> Maybe the configure tests in [1] are wrongly detecting find_task_by_pid 
> in your build environment?

I think you found it, my builds say:

checking if find_task_by_pid is usable... yes

I wonder how that check works, it must be testing syscalls in the
kernel directly. Maybe we should just disable the test?
-- 
Axel.Thimm at ATrpms.net