dkms can't build nvidia modules in 2.6.31-20-generic

[Bremm]

While installing a new kernel available by update-manager, console screen shows an error:

Examining /etc/kernel/postinst.d.
run-parts: executing /etc/kernel/postinst.d/dkms
 * Running DKMS auto installation service for kernel 2.6.31-20-generic

 * nvidia (185.18.36)...
nvidia (185.18.36): Installing module.
........(bad exit status: 10)
  Build failed. Installation skipped.
   ...fail!

So I tried to do manually:

$ sudo dkms build -m nvidia -v 185.18.36 -k 2.6.31-20-generic

Kernel preparation unnecessary for this kernel. Skipping...
applying patch rt_preempt_31.patch...patching file nv-linux.h

applying patch fall_back_on_mtrr_if_no_pat.patch...patching file nv.c
Hunk #1 succeeded at 891 (offset -49 lines).

applying patch nvidia-rt-compat.patch...patching file os-interface.c

Building module:
cleaning build area....
su nobody -c "make KERNELRELEASE=2.6.31-20-generic module KERNDIR=/lib/modules/2.6.31-20-generic IGNORE_XEN_PRESENCE=1 IGNORE_CC_MISMATCH=1 SYSSRC=/lib/modules/2.6.31-20-generic/build".....(bad exit status: 2)

Error! Bad return status for module build on kernel: 2.6.31-20-generic (x86_64)
Consult the make.log in the build directory
/var/lib/dkms/nvidia/185.18.36/build/ for more information.
0
0

And that's the reason:

$ cat /var/lib/dkms/nvidia/185.18.36/build/make.log
DKMS make.log for nvidia-185.18.36 for kernel 2.6.31-20-generic (x86_64)
Qui Mar 4 18:40:23 BRT 2010
NVIDIA: calling KBUILD...
make CC=cc KBUILD_VERBOSE=1 -C /lib/modules/2.6.31-20-generic/build SUBDIRS=/var/lib/dkms/nvidia/185.18.36/build modules
test -e include/linux/autoconf.h -a -e include/config/auto.conf || ( \
 echo; \
 echo " ERROR: Kernel configuration is invalid."; \
 echo " include/linux/autoconf.h or include/config/auto.conf are missing."; \
 echo " Run 'make oldconfig && make prepare' on kernel src to fix it."; \
 echo; \
 /bin/false)
mkdir -p /var/lib/dkms/nvidia/185.18.36/build/.tmp_versions ; rm -f /var/lib/dkms/nvidia/185.18.36/build/.tmp_versions/*
make -f scripts/Makefile.build obj=/var/lib/dkms/nvidia/185.18.36/build
cp /var/lib/dkms/nvidia/185.18.36/build/nv-kernel.o /var/lib/dkms/nvidia/185.18.36/build/nv-kernel.o
cp: impossível obter estado de `/var/lib/dkms/nvidia/185.18.36/build/nv-kernel.o': Arquivo ou diretório não encontrado
make[3]: ** [/var/lib/dkms/nvidia/185.18.36/build/nv-kernel.o] Erro 1
make[2]: ** [_module_/var/lib/dkms/nvidia/185.18.36/build] Erro 2
NVIDIA: left KBUILD.
nvidia.ko failed to build!
make[1]: ** [module] Erro 1
make: ** [module] Erro 2

But, as you can see, the needed files are there:

bremm@host:~$ locate autoconf.h auto.conf
/usr/src/linux-headers-2.6.31-19-generic/include/config/auto.conf
/usr/src/linux-headers-2.6.31-19-generic/include/config/auto.conf.cmd
/usr/src/linux-headers-2.6.31-19-generic/include/linux/autoconf.h
/usr/src/linux-headers-2.6.31-20-generic/include/config/auto.conf

Seems to be a dkms issue, but I'm not sure about. autoconf is 2.64 and dkms is 2.1.0.1 on Xubuntu 9.10 amd64.

$ lsb_release -rd
Description: Ubuntu 9.10
Release: 9.10

$ apt-cache policy autoconf
autoconf:
  Instalado: 2.64-1ubuntu1
  Candidato: 2.64-1ubuntu1
  Tabela de versão:
 *** 2.64-1ubuntu1 0
        500 http://br.archive.ubuntu.com karmic/main Packages
        100 /var/lib/dpkg/status

$ apt-cache policy dkms
dkms:
  Instalado: 2.1.0.1-0ubuntu1
  Candidato: 2.1.0.1-0ubuntu1
  Tabela de versão:
 *** 2.1.0.1-0ubuntu1 0
        500 http://br.archive.ubuntu.com karmic/main Packages
        100 /var/lib/dpkg/status
/usr/src/linux-headers-2.6.31-20-generic/include/config/auto.conf.cmd
/usr/src/linux-headers-2.6.31-20-generic/include/linux/autoconf.h

[Peter Hurley (phurley)]

This has been addressed in lucid's dkms package (2.1.1.1). The issue is that 'su nobody make...' doesn't work in this context and so has been stripped to simply 'make...'.

[Bremm (bremm)]

I've just done with dkms_2.1.1.2-0ubuntu1_all.deb and still doesn't work (tried as root also).

$ sudo dkms build -m nvidia -v 185.18.36 -k 2.6.31-20-generic

Kernel preparation unnecessary for this kernel. Skipping...
applying patch rt_preempt_31.patch...patching file nv-linux.h

applying patch fall_back_on_mtrr_if_no_pat.patch...patching file nv.c
Hunk #1 succeeded at 891 (offset -49 lines).

applying patch nvidia-rt-compat.patch...patching file os-interface.c

Building module:
cleaning build area....
make KERNELRELEASE=2.6.31-20-generic module KERNDIR=/lib/modules/2.6.31-20-generic IGNORE_XEN_PRESENCE=1 IGNORE_CC_MISMATCH=1 SYSSRC=/lib/modules/2.6.31-20-generic/build.....(bad exit status: 2)

Error! Bad return status for module build on kernel: 2.6.31-20-generic (x86_64)
Consult the make.log in the build directory
/var/lib/dkms/nvidia/185.18.36/build/ for more information.
0
0

I'll uninstall this kernel version (wanted it to try with my usb hard disk) and will wait for lucid.

Thank you for helping.

[Peter Hurley (phurley)]

@Bremm

With dkms fixed, it's better to run it as su root (rather than sudo). dkms looks at the environment (eg, $SYSSRC), but sudo doesn't copy the environment over - for good reason. I did note that you ran dkms as root as well.

When you run it as root, what's in the make.log ( /var/lib/dkms/nvidia/185.18.36/build ) ?

NB - For kernels>=2.6.33, there are several other build problems to overcome. I'll link to a different bug for that though.

[Bremm (bremm)]

Last attempt (as root) gave me this:

make CC=cc KBUILD_VERBOSE=1 -C /lib/modules/2.6.31-20-generic/build SUBDIRS=/var/lib/dkms/nvidia/185.18.36/build modules
test -e include/linux/autoconf.h -a -e include/config/auto.conf || ( \
 echo; \
 echo " ERROR: Kernel configuration is invalid."; \
 echo " include/linux/autoconf.h or include/config/auto.conf are missing."; \
 echo " Run 'make oldconfig && make prepare' on kernel src to fix it."; \
 echo; \
 /bin/false)

But the weirdest came after: I tried to run 'make oldconfig && make prepare' and it didn't work. Too bad I can't reproduce this again, because I've cleaned up the system before try 'update-nanager -d' (and it didn't work also). :-(

[Bremm (bremm)]

Now running on lucid. It seems dkms finally could handle the module building.

$ lsb_release -d
Description: Ubuntu lucid (development branch)

$ dkms -V
dkms: 2.1.1.2

By the way, safe-rm almost jeopardized my system, since it relies on perl-modules and something related to perl had been broken, leading the update process to an infinite error looping. Fortunately, 'dpkg -r safe-rm' saved my day.