After upgrading to version 535.86.05.1 Xorg dies with SIGSEV:

:
[    33.382] (II) Initializing extension NV-GLX
[    33.382] (II) Initializing extension NV-CONTROL
[    33.382] (II) Initializing extension XINERAMA
[    33.386] (EE)
[    33.386] (EE) Backtrace:
[    33.386] (EE) 0: /usr/lib/xorg/Xorg (OsLookupColor+0x14d) [0x5616c2fc5f9d]
[    33.386] (EE) 1: /lib/x86_64-linux-gnu/libc.so.6 (__sigaction+0x40) [0x7fa9bfadf510]
[    33.386] (EE) 2: /lib/x86_64-linux-gnu/libnvidia-glcore.so.535.86.05 (_nv012glcore+0x93ac) [0x7fa9bbd5850c]
[    33.387] (EE) 3: /lib/x86_64-linux-gnu/libnvidia-glcore.so.535.86.05 (_nv012glcore+0x95c8) [0x7fa9bbd58728]
[    33.387] (EE) 4: /lib/x86_64-linux-gnu/libnvidia-glcore.so.535.86.05 (_nv012glcore+0x22717) [0x7fa9bbd71877]
[    33.387] (EE) unw_get_proc_name failed: no unwind info found [-10]
[    33.387] (EE) 5: /usr/lib/xorg/modules/extensions/libglxserver_nvidia.so (?+0x0) [0x7fa9bdeca5c8]
[    33.387] (EE)
[    33.387] (EE) Segmentation fault at address 0x88
[    33.387] (EE)
Fatal server error:
[    33.387] (EE) Caught signal 11 (Segmentation fault). Server aborting
[    33.387] (EE)
[    33.387] (EE)
:

Rebuilding the package using the installed libc6 and tools
did not help. log file is attached.

Regards
Harri

Rebuilding the package won't change anything since we are only repacking
upstream provided binaries.

I didn't spot anything obvious in the upstream changelog for the newer
releases so I'll just upload the next (not latest) upstream release
(because that's the one I've ready right now) in the hope it improves
things.

Andreas

Its still broken. I tried Debian's linux-image-6.6.15-amd64 kernel and
upstream's kernel version 6.7.5.

Upstream provides version 535.154.05. Is there some script to create an
nvidia-graphics-drivers source package from it?


Regards

Harri

For verification I have installed Bookworm on an external SSD (using
debootstrap), added Gnome and nvidia-graphics-drivers 525 and verified,
upgraded to Testing and verified, and upgraded to Unstable and verified
(still 525.147.05-7~deb12u1). X worked.

Next I cherry-picked nvidia 535.86.10-1 from experimental and tried
again: Now its broken.

Please note there is no xorg.conf

Regards
Harri

dependency in the nvidia-driver package.

These packages were added by nvidia-driver-full:

```
Aptitude 0.8.13: log report
Sat, Feb 24 2024 08:57:19 +0000

   IMPORTANT: this log only lists intended actions; actions which fail
   due to dpkg problems may not be completed.

Will install 25 packages, and remove 0 packages.
497 MB of disk space will be used
========================================
[INSTALL, DEPENDENCIES] libcudadebugger1:amd64 535.86.10-1
[INSTALL, DEPENDENCIES] libgles-nvidia1:amd64 535.86.10-1
[INSTALL, DEPENDENCIES] libgles-nvidia2:amd64 535.86.10-1
[INSTALL, DEPENDENCIES] libgles1:amd64 1.7.0-1
[INSTALL, DEPENDENCIES] libnvcuvid1:amd64 535.86.10-1
[INSTALL, DEPENDENCIES] libnvidia-allocator1:amd64 535.86.10-1
[INSTALL, DEPENDENCIES] libnvidia-api1:amd64 535.86.10-1
[INSTALL, DEPENDENCIES] libnvidia-egl-gbm1:amd64 1.1.1-1
[INSTALL, DEPENDENCIES] libnvidia-encode1:amd64 535.86.10-1
[INSTALL, DEPENDENCIES] libnvidia-fbc1:amd64 535.86.10-1
[INSTALL, DEPENDENCIES] libnvidia-glvkspirv:amd64 535.86.10-1
[INSTALL, DEPENDENCIES] libnvidia-ngx1:amd64 535.86.10-1
[INSTALL, DEPENDENCIES] libnvidia-nvvm4:amd64 535.86.10-1
[INSTALL, DEPENDENCIES] libnvidia-opticalflow1:amd64 535.86.10-1
[INSTALL, DEPENDENCIES] libnvidia-rtcore:amd64 535.86.10-1
[INSTALL, DEPENDENCIES] libnvoptix1:amd64 535.86.10-1
[INSTALL, DEPENDENCIES] nvidia-cuda-mps:amd64 535.86.10-1
[INSTALL, DEPENDENCIES] nvidia-opencl-common:amd64 535.86.10-1
[INSTALL, DEPENDENCIES] nvidia-opencl-icd:amd64 535.86.10-1
[INSTALL, DEPENDENCIES] nvidia-powerd:amd64 535.86.10-1
[INSTALL, DEPENDENCIES] nvidia-settings:amd64 525.147.05-1
[INSTALL, DEPENDENCIES] nvidia-suspend-common:amd64 535.86.10-1
[INSTALL, DEPENDENCIES] nvidia-vulkan-common:amd64 535.86.10-1
[INSTALL, DEPENDENCIES] nvidia-vulkan-icd:amd64 535.86.10-1
[INSTALL] nvidia-driver-full:amd64 535.86.10-1
========================================

Log complete.

===============================================================================
```

On the first, broken upgrade from 525 to 535 (using the "nvidia-driver")
the libnvidia-egl-wayland1 package was removed. Maybe this is a hint.

Regards
Harri

+++++

Very good finding. Something I hadn't considered ...

probably in xserver-xorg-video-nvidia ... and probably a nonobvious
dependency on something that gets dlopen()ed ... and the dlopen() error
silently got ignored, but the symbol used anyway.

Can you take a look at
   lsof -p <XORGPID> | grep lib
to see which nvidia libraries are loaded into Xorg?
If you identify one (or more) of newly installed library packages, can
you uninstall nvidia-driver-full (but keep the dependencies) and the
library in question to see if the segfault returns? Maybe needs some
iterations ...

And thereafter for verification, remove everything added by
nvidia-driver-full and just add that one library that seems to
cause/solve the segfault.


Thanks

I went through the list of packages added by nvidia-driver-full,
one after the other. Of course it was the very last one making
xorg work again:

```
Aptitude 0.8.13: log report
Sat, Feb 24 2024 11:23:56 +0100

   IMPORTANT: this log only lists intended actions; actions which fail
   due to dpkg problems may not be completed.

Will install 3 packages, and remove 0 packages.
41.3 MB of disk space will be used
========================================
[INSTALL, DEPENDENCIES] libnvidia-glvkspirv:amd64 535.86.10-1
[INSTALL, DEPENDENCIES] nvidia-vulkan-common:amd64 535.86.10-1
[INSTALL] nvidia-vulkan-icd:amd64 535.86.10-1
========================================

Log complete.

===============================================================================
```
I have verified by removing all packages installed during testing
again, and installing only the nvidia-vulkan-icd package: Bingo.
Hope this helps.

BTW. Nvidia released the 550 series for Linux.


Regards
Harri

Hi Harald,

Are you sure that you really need nvidia-vulkan-icd (i.e.
nvidia-vulkan-common) and libnvidia-glvkspirv alone isn't enough?

I've found the following 'suspicious' string in
   libnvidia-{glcore,eglcore}.so.*:

     The NVIDIA driver was unable to open
'libnvidia-glvkspirv.so.535.86.05'.  This library is required at run time.

which seems to indicate the missing dependency.
While the string was already present since the 418 series (which
introduced libnvidia-glvkspirv.so.*), the missing library only recently
seems to have started causing real trouble.


Andreas

Hi Andreas,

Yes, installing libnvidia-glvkspirv is sufficient as a fix. If I remove
this package the problem is back.


Regards

Harri

We believe that the bug you reported is fixed in the latest version of
nvidia-graphics-drivers, which is due to be installed in the Debian FTP archive.

A summary of the changes between this version and the previous one is
attached.

Thank you for reporting the bug, which will now be closed.  If you
have further comments please address them to 1064194@bugs.debian.org,
and the maintainer will reopen the bug report if appropriate.

Debian distribution maintenance software
pp.
Andreas Beckmann <anbe@debian.org> (supplier of updated nvidia-graphics-drivers package)

(This message was generated automatically at their request; if you
believe that there is a problem with it please contact the archive
administrators by mailing ftpmaster@ftp-master.debian.org)
Format: 1.8
Date: Sun, 03 Mar 2024 20:26:12 +0100
Source: nvidia-graphics-drivers
Architecture: source
Version: 535.104.12-1
Distribution: experimental
Urgency: medium
Maintainer: Debian NVIDIA Maintainers <pkg-nvidia-devel@lists.alioth.debian.org>
Changed-By: Andreas Beckmann <anbe@debian.org>
Closes: 1064194
Changes:
 nvidia-graphics-drivers (535.104.12-1) experimental; urgency=medium
 .
   * New upstream Tesla branch release 535.104.12 (2023-09-25).
   * New upstream LTS and Tesla branch release 535.104.05 (2023-08-22).
   * New upstream long term support branch release 535.98 (2023-08-08).
     - Fixed an issue which caused the following error message to appear
       spuriously when using SLI with the NVIDIA Open GPU Kernel Modules:
       (EE) NVIDIA: Unable to disable FB size compare
     * Improved compatibility with recent Linux kernels.
 .
   [ Andreas Beckmann ]
   * Refresh patches.
   * Update nv-readme.ids.
   * Update symbols files.
   * Merge changes from 525.147.05-9.  (Closes: #1064194)
Checksums-Sha1:
 ba491f86a10eda9a535bfb1548a3784e029da49b 7896 nvidia-graphics-drivers_535.104.12-1.dsc
 0472125326e1593bbf4bee82fa7e3fb17603b1c4 339497403 nvidia-graphics-drivers_535.104.12.orig-amd64.tar.gz
 76ce6138b584c6ab849cf5e987ff61cb3864b687 268460923 nvidia-graphics-drivers_535.104.12.orig-arm64.tar.gz
 8263caceb5dcd2dc3671157dc217cafc90d2c9e1 96253093 nvidia-graphics-drivers_535.104.12.orig-ppc64el.tar.gz
 f55586150653188c9a936a78d03387622b82c2d1 139 nvidia-graphics-drivers_535.104.12.orig.tar.gz
 d8efd1c55aaaa9a7c33d9cee21893366fc33b32f 226624 nvidia-graphics-drivers_535.104.12-1.debian.tar.xz
 1ca62bb450aafb819ccbad0ea004ca8149a4651e 5738 nvidia-graphics-drivers_535.104.12-1_source.buildinfo
Checksums-Sha256:
 0cbe90a81e390ed711ed9b376f5bca95204b5b2a302522ebc2c574998a113272 7896 nvidia-graphics-drivers_535.104.12-1.dsc
 e36d60b5cef2ff417389ccbd701d6162fa18b0c16365ba921a863885231e4385 339497403 nvidia-graphics-drivers_535.104.12.orig-amd64.tar.gz
 90c25ac321052059f8890031e83edd2d7f094d80ec41f901ab57bf68b061e423 268460923 nvidia-graphics-drivers_535.104.12.orig-arm64.tar.gz
 a92d64c28352f2c0c4172326a0066e21dbc963837c85be2834a86c39d2e11085 96253093 nvidia-graphics-drivers_535.104.12.orig-ppc64el.tar.gz
 20d17e907605bc08759f4bd249bf6489d7c56addf3dc2bf0ec1b992d7506a9ce 139 nvidia-graphics-drivers_535.104.12.orig.tar.gz
 4ef7b03a023587a0fb87a7f42ec53efc990a80ecde288c03a112b6de81523554 226624 nvidia-graphics-drivers_535.104.12-1.debian.tar.xz
 e02b7ab502616abe41cc6c3c62b9fbe05ffb59a620aa536aba1870a1f054f2cd 5738 nvidia-graphics-drivers_535.104.12-1_source.buildinfo
Files:
 deebe62fd063fc5f265860d33ca2b672 7896 non-free/libs optional nvidia-graphics-drivers_535.104.12-1.dsc
 2c77f3fe70a2437e2f16a9eaaaccc515 339497403 non-free/libs optional nvidia-graphics-drivers_535.104.12.orig-amd64.tar.gz
 0fc469e0bf77a2a2bd21f2ca8a2aa245 268460923 non-free/libs optional nvidia-graphics-drivers_535.104.12.orig-arm64.tar.gz
 a2b3d5754c5438b2dc2025b7f507f31d 96253093 non-free/libs optional nvidia-graphics-drivers_535.104.12.orig-ppc64el.tar.gz
 c040eb89f65863e987e1d0b8d6394202 139 non-free/libs optional nvidia-graphics-drivers_535.104.12.orig.tar.gz
 66492fdd6ccb8ea2a8607695c713879f 226624 non-free/libs optional nvidia-graphics-drivers_535.104.12-1.debian.tar.xz
 9f2e35af8af4546a5f10d4913363513d 5738 non-free/libs optional nvidia-graphics-drivers_535.104.12-1_source.buildinfo
-----BEGIN PGP SIGNATURE-----
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=iKkM
-----END PGP SIGNATURE-----

Hi,

Thanks for confirming my assumption.

I've moved the libnvidia-glvkspirv dependency to libnvidia-(e)glcore
which seem to make more use of that in experimental.
Furthermore I've promoted nvidia-vulkan-icd to Recommends.


Andreas