Dag Wieers wrote:
On Wed, 25 Jun 2008, Seán O Sullivan wrote:

I noticed some odd issues with nvidia-x11-drv 173.08-1.beta.el5.rf
while upgrading from CentOS 5.1 -> 5.2 (x86).

It occurred on three machines, all AMD (Athlon(tm) 64 X2 Dual Core
Processor 3800+,  Opteron(tm) Processor 246).

When the nvidia.ko module was being loaded the machine would stop
responding & require a hard reboot.
Resolved by removing & reinstalling nvidia-x11-drv.
Seemingly when the upgrade happened, an nvidia.ko symlink was placed
in /lib/modules/2.6.18-92.1.1.el5/weak-updates/ pointing to the .ko
in /lib/modules/2.6.18-53.1.21.el5/weak-updates/

When did reinstall & ran /etc/init.d/dkms_autoinstaller nvidia.ko was
placed in /lib/modules/2.6.18-92.1.1.el5/extra

Hi Sean,

Do you mean that the weak-module was unstable, but a re-compiled module did not have a problem at all ?


Adding my 2c, I think that's correct (replace "unstable" by "broken", this is the "black screen & lockup at rhgb" many people have been talking about on the centos list).

I also hit this when I upgraded a workstation to 5.2 yesterday.
I solved it by rebooting under my older kernel and using dkms commands (status, remove, build, install, rinse and repeat...).
I guess uninstalling and reinstalling the rpm does the same to dkms.

I don't know dkms well, but the problem seems to be that when installing a newer kernel, a weak-updates/ link is automatically created to the previously built module. Then when you reboot, dkms_autoinstall is supposed to build the module for the new kernel. But it doesn't do anything because it sees the weak-updates version, and thinks this is good enough. This doesn't usually pose problems in centos/rhel, because the kernels are so similar. But from 5.1 to 5.2 there was enough difference in the kernel interface that the nvidia driver compiled for the 5.1 kernel didn't work with the 5.2 kernel.



If so, I think that needs to be reported to dkms. I do not influence the weak-linking behaviour, but if some assumption is wrong from the technical angle, it should probably be disabled.

_______________________________________________
users mailing list
[email protected]
http://lists.rpmforge.net/mailman/listinfo/users

Reply via email to