Bug#860104: nvidia-kernel-dkms: The last three kernel update breaks nvidia module

[Luca Boccassi]

Control: reassign -1 nvidia-driver
Control: forcemerge 856355 -1

On Tue, 11 Apr 2017 15:51:39 +0200 =?utf-8?q?Nicolas_M=C3=A9nec?= <mene
obzh at gmail.com> wrote:
> Package: nvidia-kernel-dkms
> Version: 375.39-1
> Severity: normal
> Tags: newcomer
> 
> Dear Maintainer,
> 
> Hi!
> 
> I have a amd64 sid system, built on UEFI, but it looks like the
nVidia driver
> have some troubles with the last kernel releases.
> 
> In my case, the issue manifested as not being able to make working
Xorg after
> upgrading linux kernel and headers.
> The system boots normally : before X starts, I am able to see boot
messages,
> but when Xorg starts, I see a black blink screen in place of gdm3.
> 
> As first workaround (no time to investigate, i apolize), I just
downgraded
> kernel packages (image, headers and others) to 4.9.0-2-amd64_4.9.10-
1_amd64.
> My system was OK: Xorg starts with gdm3.
> Since few weeks, I noticed troubles with kernel versions 4.9.13-1,
4.9.16-1 and
> the current 4.9.18-1.
> 
> As a secund workaround (more time to investigate ;-)), I upgraded my
kernel
> (image, headers, etc..) to version 4.9.18-1
> Xorg was broken again. A look in the /var/log/kern.log, I had this
message :
> Apr 11 11:41:07 mycomputer kernel: [   77.692226] nvidia: disagrees
about
> version of symbol module_layout
> 
> So, I decided to rebuild the nVidia driver module via dkms by the
command line.
> Just doing the commands as follow:
> # dkms remove -m nvidia-current -v 375.39 --all
> # dkms add -m nvidia-current -v 375.39
> # dkms rebuid -m nvidia-current -v 375.39
> # dkms install -m nvidia-current -v 375.39
> 
> After a complete reboot, the system works fine : Xorg starts with
gdm3.
> 
> In conclusion, the last kernel releases have changed kernel symbols
betwenn
> versions 4.9.10-1 and 4.9.13-1 without rebuilding the nvidia module
on the fly
> by dkms as usually.
> I think my secund workaround has definitely resolved my issue.
> So, I hope this report will help someone who have same troubles with
kernel
> updates on nvidia driver.
> 
> Please let me know if I can provide any additional
information.  Thanks very
> much!
> 
> Regards,
> 
> Nicolas.

Hello,

This is the same as 856355. In short:

The kernel headers packages have the name versioned after the ABI
version. DKMS understanding is that as long as the package name (and
thus the ABI version) does not change, then there is no need to rebuild
as the ABI is (assumed to be) compatible. When the headers packages
change name, then DKMS automatically rebuilds.

If the ABI change is not compatible but the package name does not
change, DKMS has no way to figure out that it needs to rebuild.

From the changelog it looks like the kernel maintainers are aware of
ABI breakages:

  * [powerpc*] Ignore ABI changes in cxl (fixes FTBFS) (Closes:
#858530)
    and IOMMU setup
  * Ignore ABI changes in bpf, dccp, libiscsi
  * [x86] Ignore ABI changes in kvm

http://metadata.ftp-master.debian.org/changelogs/main/l/linux/linux_4.9
.18-1_changelog

There is nothing we can do unfortunately, and what you did by
rebuilding via dkms is the right workaround. Sorry for the
inconvenience. Once Stretch is released it shouldn't happen again.

Kind regards,
Luca Boccassi
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 488 bytes
Desc: This is a digitally signed message part
URL: <http://lists.alioth.debian.org/pipermail/pkg-nvidia-devel/attachments/20170420/4846a06b/attachment.sig>