Web lists-archives.com

Re: backports on kernel 4.15 and nvidia-driver 390 crashes x




On Sun, 2018-05-06 at 16:34 -0400, Boyan Penkov wrote:
> On Sun, 2018-05-06 at 11:20 -0400, Boyan Penkov wrote:
> > > Updating sources to buster and pulling in kernel-image, kernel-
> > > headers, then restarting to 4.16 to recovery mode, uninstalling
> > > nvidia-driver and then reinstalling the nvidia-driver package
> > > seems to
> > > have worked…. However, this cannot be the orthodox approach to
> > > solve
> > > this…..
> > 
> > 	Actually I think it might just be. The problem is that it
> > involves a kernel module so a reboot might be mandatory... 
> > 	However, installing the kernel (with headers) and running apt-
> > get --reinstall and then rebooting might be enough.
> 
> That’s totally fair — and I’m happy and won’t be playing with this
> till I go all buster at the freeze — but should there be a hook in
> apt to make this more transparent?  Something like, if nvidia-driver
> is installed and the kernel changes, check for the relevant headers,
> and recompile the module?
> 
> I’ve verb bitten by this before, and it seems straightforward to do
> as part of the package post-hook, so am I missing something?

	I've been thinking the same thing since my last bout with the
drivers. I don't know what's happening - maybe the hook is there but
it's not working properly, or maybe the package maintainers is not
aware of this issue (which I think is unlikely). 

	I think the hook is there, somewhere. I've recently upgraded
from 4.15 to 4.16 and everything worked out fine, and I didn't need to
reinstall the drivers nor build any kernel module. It might be a
version-related issue.

	Checking Debian's bug tracker I've found a bug that seems to be
related to what happened to us, from 2014 (last updated in 2015), and
it's still flagged as outstanding.

	We need some tests to be sure. I have an extra partition that I
can use for testing, but it might take some time.

--
Francisco