[OpenAFS] Re: 1.4.2 client on RHEL5 beta 2

[Axel Thimm]

--TOkWJigZa0YodlBE
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
Content-Transfer-Encoding: quoted-printable

On Mon, Nov 20, 2006 at 05:32:58PM +0100, Stephan Wiesand wrote:
> On Mon, 20 Nov 2006, Axel Thimm wrote:
>=20
> >On Mon, Nov 20, 2006 at 03:39:56PM +0100, Stephan Wiesand wrote:
> >>Again, kind of a success report:
> >>
> >>The only major build problem (kernel 2.6.18-1.2747.el5) was the return =
of
> >>tasklist_lock (SGI need it, et voila...) - alas, GPL-only:
> >>
> >>  LD [M]
> >>  /usr/src/redhat/BUILD/openafs-1.4.2/src/libafs/MODLOAD-2.6.18-1.2747.=
el5-MP/libafs.o
> >>  Building modules, stage 2.
> >>  MODPOST
> >>FATAL: modpost: GPL-incompatible module libafs.ko uses GPL-only symbol
> >>'tasklist_lock'
> >>make[6]: *** [__modpost] Error 1
> >
> >I rebuilt the 1.4.2 packages for beta2 (kmdl) and didn't stumble over
> >it (?). Does that need explicitely turning on?
>=20
> All I need to do in order to reproduce the problem on a fresh=20
> installation of RHEL5 beta 2:
>=20
> tar xfj openafs-1.4.2-src.tar.bz2
> cd openafs-1.4.2
> ./configure
> make
>=20
> Could you try? It's possible that the error I get is a result of=20
> modifications to my system after installation. I doubt it, especially=20
> since our site customization is largely unfinished at this point, but=20
> please prove me wrong if you can. Here's the list of kernel related=20
> packages I have installed:
>=20
> kernel-module-openafs-2.6.18-1.2747.el5-1.4.2-31.SL5.x86_64
> kernel-devel-2.6.18-1.2747.el5.x86_64
> kernel-headers-2.6.18-1.2747.el5.x86_64
> kernel-2.6.18-1.2747.el5.x86_64
>=20
> The first one, of course, being my patched build. NB behaviour was the=20
> same when building on beta 1 with just the new kernel-devel installed.
>=20
> And on x86 it's just the same.

I tried in a clean chroot of EL5/beta2/x86_64 and the build fails at a
completely different spot:

compile_et.o: In function `yyerror':
/var/tmp/delme/openafs-1.4.2/src/comerr/compile_et.c:419: undefined referen=
ce to `yylineno'
compile_et.o: In function `main':
/var/tmp/delme/openafs-1.4.2/src/comerr/compile_et.c:289: undefined referen=
ce to `yyin'
/var/tmp/delme/openafs-1.4.2/src/comerr/compile_et.c:296: undefined referen=
ce to `yyout'
error_table.o: In function `yyparse':
/var/tmp/delme/openafs-1.4.2/src/comerr/y.tab.c:547: undefined reference to=
 `yylex'
/var/tmp/delme/openafs-1.4.2/src/comerr/y.tab.c:727: undefined reference to=
 `yylex'
collect2: ld returned 1 exit status

Probably some parts are missing still, although flex/bison/byacc were
on board.

> >>I removed all uses of tasklist_lock, leaving only the else-branches
> >>calling rcu_read_[un]lock in place. 1.4.2 then builds and works.
> >>
> >>Other observations:
> >> - df -H shows "0.0k" on my amd64 system, and "166Y" on x86 (I now lear=
ned
> >>   that a Yottabyte is 1k Zetabytes...)
> >> - "umount /afs" takes unusually long, maybe 30 or 60 seconds - sometim=
es
> >>
> >>Any comments on those? (preferrably like "don't worry, that's ok" ;-) ?
> >
> >Could you please test the packages at http://atrpms.net/dist/el5/openafs=
/?
>=20
> I can do this tomorrow. But even if they work, I still have to understand=
=20
> why they build and the plain 1.4.2 tarball does not.

The packages use a couple of args more to configure, c.f.
http://dl.atrpms.net/all/openafs.spec, e.g. it looks more like

=2E/configure --build=3Di686-redhat-linux-gnu --host=3Di686-redhat-linux-gn=
u --target=3Di686-redhat-linux-gnu --program-prefix=3D --prefix=3D/usr --ex=
ec-prefix=3D/usr --bindir=3D/usr/bin --sbindir=3D/usr/sbin --sysconfdir=3D/=
etc --datadir=3D/usr/share --includedir=3D/usr/include --libdir=3D/usr/lib =
--libexecdir=3D/usr/libexec --localstatedir=3D/var --sharedstatedir=3D/usr/=
com --mandir=3D/usr/share/man --infodir=3D/usr/share/info --with-afs-sysnam=
e=3Di386_linux26 --with-linux-kernel-headers=3D/build/kernels/el5-i386/2.6.=
18-1.2747.el5-i686 --with-krb5-conf --enable-redhat-buildsys
make only_libafs

(cut from the logs to the i686 non-smp build). That's more or less the
only relevant differences.
--=20
Axel.Thimm at ATrpms.net

--TOkWJigZa0YodlBE
Content-Type: application/pgp-signature
Content-Disposition: inline

-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.5 (GNU/Linux)

iD8DBQFFYd74QBVS1GOamfERAmoQAJ9jLO8kffTq31aBdARTxuSlyeVg3ACeLMV6
HbkH5NSFMcwi/CSCJiUzYV4=
=hhQz
-----END PGP SIGNATURE-----

--TOkWJigZa0YodlBE--