[elrepo] Today's nvidia 340xx update

[Phil Perry]

On 04/08/15 23:24, Graham Allan wrote:
> On Mon, Aug 03, 2015 at 08:30:02PM +0100, Phil Perry wrote:
>> Hi Graham,
>>
>> On 03/08/15 19:30, Graham Allan wrote:
>>> I've had a handful of machines update their nvidia driver and after
>>> reboot X no longer works.
>>>
>>> I notice that kmod-nvidia updated to kmod-nvidia-340xx-340.76-2 but
>>> the x11 driver stayed at nvidia-x11-drv-340xx-340.76-1. Could that be a
>>> problem? 
>>
>> No
> 
> Had a feeling that wasn't likely :-)
>  
>>> Downgrading the kmod to kmod-nvidia-340xx-340.76-1 gets things
>>> working again.
>>>
>>> Is it an oversight that there's no 340.76-2 x11 driver, or am I looking
>>> in the wrong place for my issue?
>>>
>>
>> kmod-nvidia-340xx was rebuilt against the RHEL-6.7 kernel
>> (kernel-2.6.32-573.el6) to fix this issue:
>>
>> http://elrepo.org/bugs/view.php?id=583
>>
>> What kernel are you running?
> 
> Sorry for my delay in reply. Been having a bunch of different machines acting
> in different ways, and trying to put together a coherent picture.
> 
> We're actually running Scientific Linux 6.6, and kernel-2.6.32-504.30.3.el6.x86_64
> 
> It does seem like most of the issues I see are with machines using
> kmod-nvidia-340xx (but not all, I've seen at least one with the updated
> kmod-nvidia-352.30-1 - was that also rebuilt against RHEL-6.7?).
> 

Yes, just to confirm, the kmod-nvidia [352.30-1] and kmod-nvidia-340xx
[340.76-2] updates that we released for el6 this week are both built
against the new 6.7 kernel as there were issues when built against older
kernels.

The limited testing I was able to perform (I don't have the nvidia
hardware to fully test) showed the nvidia kernel module weak linked back
against the 6.6 kernel suggesting they were backward compatible. That is
now clearly not the case (from yours and others reports).

So the solution is to hold off on the nvidia updates until you update to
the 6.7 kernel (2.6.32-573.el6 or above). If you did get burned then a
simple downgrade of nvidia packages should fix the issue. As always, a
reboot is required after updating / downgrading nvidia drivers.

Please accept my sincere apologies for any inconvenience caused. I hate
it when our packages cause issues but it's increasingly hard for us to
test all releases on the limited hardware we have available - I have one
test rig triple booting RHEL 5|6|7 with an older nvidia card so can only
test the legacy releases.

If anyone is able to donate newer hardware (e.g, a GT 6xx or GT 7xx
series card) please feel free to get in touch.

> Some of them fail to start X erratically on boot, some are apparently locking
> up during use, and I see a lot of odd call traces in /var/log/messages (not
> exclusively X-related). The common message here seems to be
> "kernel: BUG: scheduling while atomic"
> 
>>From what I can see so far, rebooting into an earlier kernel doesn't make any
> difference, but downgrading (eg kmod-nvidia-340xx to 340.76-1) does make things
> stable again.
> 
> Thanks again,
> 
> Graham
> _______________________________________________
> elrepo mailing list
> elrepo at lists.elrepo.org
> http://lists.elrepo.org/mailman/listinfo/elrepo
>