Dear Maintainer,

The new 7.2 kernel headers fails to build the nVidia driver.
Here is apt’s output:

Paramétrage de linux-image-7.2.6+deb14-amd64 (7.2.6-1) ...
I: /vmlinuz.old is now a symlink to boot/vmlinuz-7.1.13+deb14-amd64
I: /initrd.img.old is now a symlink to boot/initrd.img-7.1.13+deb14-amd64
I: /vmlinuz is now a symlink to boot/vmlinuz-7.2.6+deb14-amd64
I: /initrd.img is now a symlink to boot/initrd.img-7.2.6+deb14-amd64
/etc/kernel/postinst.d/dkms:
Sign command: /lib/modules/7.2.6+deb14-amd64/build/scripts/sign-file
Signing key: /var/lib/dkms/mok.key
Public certificate (MOK): /var/lib/dkms/mok.pub

Autoinstall of module nvidia-current/550.163.01 for kernel 7.2.6+deb14-amd64
(x86_64)
Building module(s)...........(bad exit status: 2)
Failed command:
env NV_VERBOSE=1 make -j12 modules KERNEL_UNAME=7.2.6+deb14-amd64

Error! Bad return status for module build on kernel: 7.2.6+deb14-amd64 (x86_64)
Consult /var/lib/dkms/nvidia-current/550.163.01/build/make.log for more
information.

Autoinstall on 7.2.6+deb14-amd64 failed for module(s) nvidia-current(10).

Error! One or more modules failed to install during autoinstall.
Refer to previous errors for more information.
run-parts: /etc/kernel/postinst.d/dkms exited with return code 1
dpkg: erreur de traitement du paquet linux-image-7.2.6+deb14-amd64
(--configure) :
 old linux-image-7.2.6+deb14-amd64 package postinst maintainer script
subprocess failed with exit status 1
Paramétrage de linux-headers-7.2.6+deb14-amd64 (7.2.6-1) ...
/etc/kernel/header_postinst.d/dkms:
Sign command: /lib/modules/7.2.6+deb14-amd64/build/scripts/sign-file
Signing key: /var/lib/dkms/mok.key
Public certificate (MOK): /var/lib/dkms/mok.pub

Autoinstall of module nvidia-current/550.163.01 for kernel 7.2.6+deb14-amd64
(x86_64)
Building module(s)..........(bad exit status: 2)
Failed command:
env NV_VERBOSE=1 make -j12 modules KERNEL_UNAME=7.2.6+deb14-amd64

Error! Bad return status for module build on kernel: 7.2.6+deb14-amd64 (x86_64)
Consult /var/lib/dkms/nvidia-current/550.163.01/build/make.log for more
information.

Autoinstall on 7.2.6+deb14-amd64 failed for module(s) nvidia-current(10).

Error! One or more modules failed to install during autoinstall.
Refer to previous errors for more information.
run-parts: /etc/kernel/header_postinst.d/dkms exited with return code 1
dpkg: erreur de traitement du paquet linux-headers-7.2.6+deb14-amd64
(--configure) :
 old linux-headers-7.2.6+deb14-amd64 package postinst maintainer script
subprocess failed with exit status 1
Paramétrage de libavformat63:amd64 (10:9.0.2-dmo1) ...
dpkg: des problèmes de dépendances empêchent la configuration de linux-image-
amd64 :
 linux-image-amd64 dépend de linux-image-7.2.6+deb14-amd64 (= 7.2.6-1);
cependant :
 Le paquet linux-image-7.2.6+deb14-amd64 n'est pas encore configuré.

dpkg: erreur de traitement du paquet linux-image-amd64 (--configure) :
 problèmes de dépendances - laissé non configuré
Paramétrage de libavfilter12:amd64 (10:9.0.2-dmo1) ...
dpkg: des problèmes de dépendances empêchent la configuration de linux-headers-
amd64 :
 linux-headers-amd64 dépend de linux-headers-7.2.6+deb14-amd64 (= 7.2.6-1);
cependant :
 Le paquet linux-headers-7.2.6+deb14-amd64 n'est pas encore configuré.

dpkg: erreur de traitement du paquet linux-headers-amd64 (--configure) :
 problèmes de dépendances - laissé non configuré
[…]

Des erreurs ont été rencontrées pendant l'exécution :
 linux-image-7.2.6+deb14-amd64
 linux-headers-7.2.6+deb14-amd64
 linux-image-amd64
 linux-headers-amd64
Erreur : Sub-process /usr/bin/dpkg returned an error code (1)

I’m not sure if the ftbfs tag is required here.

Yours,
n.

Dear Maintainer,

A quick review shows that there seem to be several culprits here:

* strncpy seems to have been removed from kernel 7.2.6, so all references to
strncpy would need to be replaced with strscpy.  I found the following:
  * nvidia/os-interface.c:753            strncpy(buf, current->comm, len - 1);
  * nvidia/linux_nvswitch.c:2029         strncpy(regkey_val, regkey_val_start,
regkey_val_len);
  * nvidia/linux_nvswitch.c:2384         return strncpy(dest, src, length);
(nvswitch_os_strncpy)  <--- Given this is called by code not open-sourced, I
believe, we'd need to be careful on whether the calling code is expecting the
zero-padding strncpy does but which I believe strscpy does not, in case we need
to do some wrapping to emulate that behaviour instead of simply replacing one
with the other.
  * nvidia-uvm/uvm_pmm_gpu.c:3041        strncpy(chunk_split_cache[level].name,
"uvm_gpu_chunk_t", ...);
  * nvidia-modeset/nvidia-modeset-linux.c:354  return strncpy(dest, src, n);
(nvkms_strncpy) <--- Same as above: Given this is called by code not open-
sourced, I believe, we'd need to be careful on whether the calling code is
expecting the zero-padding strncpy does but which I believe strscpy does not,
in case we need to do some wrapping to emulate that behaviour instead of simply
replacing one with the other.
* It seems that in kernel 7.2.6 the DRM core changed the 'atomic' part.  I
believe it renamed `struct drm_atomic_state` to `struct drm_atomic_commit`, and
made other non-trivial changes to how lifecycle is handled. These changes (e.g.
the struct drm_atomic_state seems to no longer exist) cause the module to fail
compilation, which cascades to every call where a pointer to this struct is
used in calls to the new DRM functions. This seems like more complex changes
are needed than simply the substitutions above...

Some errors received are:

  nvidia-drm/nvidia-drm-drv.c:224:27: error: initialization of ‘struct
drm_atomic_commit * (*)(struct drm_device *)’ from incompatible pointer
  type ‘struct drm_atomic_state * (*)(struct drm_device *)’
  [-Wincompatible-pointer-types]

  nvidia-drm/nvidia-drm-drv.c:1212:13: error: implicit declaration of function
‘drm_atomic_state_alloc’; did you mean
  ‘nv_drm_atomic_state_alloc’? [-Wimplicit-function-declaration]

  nvidia-drm/nvidia-drm-modeset.c:47:29: error: field ‘base’ has incomplete
type

  include/drm/drm_atomic.h:990:31: error: invalid use of undefined type ‘struct
drm_atomic_state’

Affected files seem to be: nvidia-drm-drv.c, nvidia-drm-connector.c, nvidia-
drm-crtc.c,
nvidia-drm-helper.c, nvidia-drm-modeset.c.

Not sure if changes in the open-sourced open-gpu-kernel-modules code might
contain adaptation to this DRM change that could be backported to the
550.163.01 module.

I can reproduce this issue on Debian testing/forky amd64.

Hardware:
NVIDIA GeForce RTX 4070 SUPER
PCI ID: 10de:2783

Kernel versions:
- 7.1.13+deb14-amd64: works correctly
- 7.2.6+deb14-amd64: NVIDIA DKMS module fails to build

The kernel 7.2.6 package installation fails while running DKMS:

Autoinstall of module nvidia-current/550.163.01 for kernel 7.2.6+deb14-amd64 failed.

Relevant compiler error:

nvidia/os-interface.c: In function ‘os_get_current_process_name’:
nvidia/os-interface.c:753:5: error: implicit declaration of function ‘strncpy’
753 | strncpy(buf, current->comm, len - 1);
| ^~~~~~~

Because the NVIDIA module is not built, the new kernel does not start
nvidia-persistenced and systemd-modules-load.service fails. The system
can boot far enough to unlock LUKS and mount the LVM root filesystem, but
the graphical session does not start correctly.

The initramfs for the new kernel had to be generated manually with:

update-initramfs -c -k 7.2.6+deb14-amd64

The installed packages include:

linux-image-7.2.6+deb14-amd64 7.2.6-1
linux-headers-7.2.6+deb14-amd64 7.2.6-1
nvidia-driver 550.163.01-5.1
nvidia-kernel-dkms 550.163.01-5.1
dkms 3.2.2-1
The previous kernel 7.1.13+deb14-amd64 remains usable.

Hi,

Attached is a patch against nvidia-graphics-drivers 550.163.01-5.1 that
makes
nvidia-kernel-dkms build and work on Linux 7.2.6+deb14-amd64. It addresses
the
two issues Jonas described in his follow-up to this bug.

1. strncpy() removed in Linux 7.2

   - nvidia/os-interface.c and nvidia-uvm/uvm_pmm_gpu.c: switched to
strscpy(),
     same as NVIDIA's upstream fix (open-gpu-kernel-modules PR #1227,
shipped
     in 610.57.04).
   - nvidia/linux_nvswitch.c (registry key parsing): switched to memcpy(),
also
     as upstream does. The following line already NUL-terminates.
   - nvkms_strncpy() and nvswitch_os_strncpy(): these are called from the
     closed-source binary objects, so, following the concern raised in this
     bug, I kept their names and exact strncpy() semantics (copy, then
     zero-pad the rest of the buffer), implemented locally instead of
changing
     them to strscpy().

2. struct drm_atomic_state renamed to struct drm_atomic_commit
   (commit 5164f7e7ff8e)

   - New conftest drm_atomic_commit_present (types), registered in
     nvidia-drm-sources.mk. It uses sizeof(struct drm_atomic_commit), so it
     cannot give a false positive on older kernels.
   - When it is defined, nvidia-drm-conftest.h maps the old names to the new
     API (drm_atomic_state_alloc/clear/get/put/init/default_clear/
     default_release -> drm_atomic_commit_*). I checked these names against
     the 7.2.6 headers.
   - Several existing conftests probe using the old struct name and silently
     report false negatives on 7.2. That made the driver pick the pre-4.10
     code paths (drm_atomic_state_free, and the old signatures of the crtc
and
     plane atomic_check callbacks). When NV_DRM_ATOMIC_COMMIT_PRESENT is
set,
     the patch forces ATOMIC_STATE_REF_COUNTING_PRESENT,
     CRTC/PLANE_ATOMIC_CHECK_HAS_ATOMIC_STATE_ARG and the two
     ATOMIC_HELPER_SWAP_STATE conftests.

   Everything new is guarded by NV_DRM_ATOMIC_COMMIT_PRESENT, so builds on
   older kernels are unchanged.

Testing:

- Built with DKMS against 7.2.6+deb14-amd64 and 7.1.13+deb14-amd64. Both
  succeed with no errors and no implicit-declaration or modpost warnings.
  On 7.1.13 the new conftest correctly reports #undef.
- Hardware: HP laptop with GeForce RTX 2050 (GA107, 10de:25ad) plus Radeon
  680M, KDE Plasma on Wayland, forky.
- Booted 7.2.6: nvidia, nvidia_modeset, nvidia_drm and nvidia_uvm load,
  nvidia-smi reports the GPU with driver 550.163.01, and the desktop works.

This is a local fix and I have only tested it on one machine, so please
review
it before using it. I hope it helps get a 7.2-compatible upload out.

Thanks,
Kevin Cuevas

Dear Maintainer,

Is there any word on when nvidia-kernel-dkms will be updated?
I’m on Debian Testig, currently with kernel 7.1.13; a while back, I had an update to 7.2.6, but it wouldn't complete due to this error.
A kernel update to version 7.2.8 was released today, and once again, I
am unable to install it. This is my APT output (polish language):
Konfigurowanie pakietu linux-headers-7.2.6+deb14-amd64 (7.2.6-1) ...
/etc/kernel/header_postinst.d/dkms: Sign command:
/lib/modules/7.2.6+deb14-amd64/build/scripts/sign-file Signing key:
/var/lib/dkms/mok.key Public certificate (MOK): /var/lib/dkms/mok.pub
Autoinstall of module nvidia-current/550.163.01 for kernel
7.2.6+deb14-amd64 (x86_64) Building module(s)..............(bad exit
status: 2) Failed command: env NV_VERBOSE=1 make -j8 modules
KERNEL_UNAME=7.2.6+deb14-amd64 Error! Bad return status for module build
on kernel: 7.2.6+deb14-amd64 (x86_64) Consult
/var/lib/dkms/nvidia-current/550.163.01/build/make.log for more
information. Autoinstall on 7.2.6+deb14-amd64 failed for module(s)
nvidia-current(10). Error! One or more modules failed to install during
autoinstall. Refer to previous errors for more information. run-parts:
/etc/kernel/header_postinst.d/dkms exited with return code 1 dpkg: błąd
przetwarzania pakietu linux-headers-7.2.6+deb14-amd64 (--configure): old
linux-headers-7.2.6+deb14-amd64 package postinst maintainer script
subprocess failed with exit status 1 Konfigurowanie pakietu
linux-headers-7.2.8+deb14-amd64 (7.2.8-1) ...
/etc/kernel/header_postinst.d/dkms: Sign command:
/lib/modules/7.2.8+deb14-amd64/build/scripts/sign-file Signing key:
/var/lib/dkms/mok.key Public certificate (MOK): /var/lib/dkms/mok.pub
Autoinstall of module nvidia-current/550.163.01 for kernel
7.2.8+deb14-amd64 (x86_64) Building module(s)..............(bad exit
status: 2) Failed command: env NV_VERBOSE=1 make -j8 modules
KERNEL_UNAME=7.2.8+deb14-amd64 Error! Bad return status for module build
on kernel: 7.2.8+deb14-amd64 (x86_64) Consult
/var/lib/dkms/nvidia-current/550.163.01/build/make.log for more
information. Autoinstall on 7.2.8+deb14-amd64 failed for module(s)
nvidia-current(10). Error! One or more modules failed to install during
autoinstall. Refer to previous errors for more information. run-parts:
/etc/kernel/header_postinst.d/dkms exited with return code 1 dpkg: błąd
przetwarzania pakietu linux-headers-7.2.8+deb14-amd64 (--configure): old
linux-headers-7.2.8+deb14-amd64 package postinst maintainer script
subprocess failed with exit status 1 Konfigurowanie pakietu
linux-image-7.2.8+deb14-amd64 (7.2.8-1) ... I: /initrd.img is now a
symlink to boot/initrd.img-7.2.8+deb14-amd64
/etc/kernel/postinst.d/dkms: Sign command:
/lib/modules/7.2.8+deb14-amd64/build/scripts/sign-file Signing key:
/var/lib/dkms/mok.key Public certificate (MOK): /var/lib/dkms/mok.pub
Autoinstall of module nvidia-current/550.163.01 for kernel
7.2.8+deb14-amd64 (x86_64) Building module(s)..............(bad exit
status: 2) Failed command: env NV_VERBOSE=1 make -j8 modules
KERNEL_UNAME=7.2.8+deb14-amd64 Error! Bad return status for module build
on kernel: 7.2.8+deb14-amd64 (x86_64) Consult
/var/lib/dkms/nvidia-current/550.163.01/build/make.log for more
information. Autoinstall on 7.2.8+deb14-amd64 failed for module(s)
nvidia-current(10). Error! One or more modules failed to install during
autoinstall. Refer to previous errors for more information. run-parts:
/etc/kernel/postinst.d/dkms exited with return code 1 dpkg: błąd
przetwarzania pakietu linux-image-7.2.8+deb14-amd64 (--configure): old
linux-image-7.2.8+deb14-amd64 package postinst maintainer script
subprocess failed with exit status 1 Konfigurowanie pakietu
linux-image-7.2.6+deb14-amd64 (7.2.6-1) ... I: /initrd.img.old is now a
symlink to boot/initrd.img-7.2.6+deb14-amd64
/etc/kernel/postinst.d/dkms: Sign command:
/lib/modules/7.2.6+deb14-amd64/build/scripts/sign-file Signing key:
/var/lib/dkms/mok.key Public certificate (MOK): /var/lib/dkms/mok.pub
Autoinstall of module nvidia-current/550.163.01 for kernel
7.2.6+deb14-amd64 (x86_64) Building module(s)..............(bad exit
status: 2) Failed command: env NV_VERBOSE=1 make -j8 modules
KERNEL_UNAME=7.2.6+deb14-amd64 Error! Bad return status for module build
on kernel: 7.2.6+deb14-amd64 (x86_64) Consult
/var/lib/dkms/nvidia-current/550.163.01/build/make.log for more
information. Autoinstall on 7.2.6+deb14-amd64 failed for module(s)
nvidia-current(10). Error! One or more modules failed to install during
autoinstall. Refer to previous errors for more information. run-parts:
/etc/kernel/postinst.d/dkms exited with return code 1 dpkg: błąd
przetwarzania pakietu linux-image-7.2.6+deb14-amd64 (--configure): old
linux-image-7.2.6+deb14-amd64 package postinst maintainer script
subprocess failed with exit status 1 dpkg: problemy z zależnościami
uniemożliwiają skonfigurowanie pakietu linux-headers-amd64:
linux-headers-amd64 zależy od linux-headers-7.2.8+deb14-amd64 (=
7.2.8-1); jednakże: Pakiet linux-headers-7.2.8+deb14-amd64 nie jest
jeszcze skonfigurowany. dpkg: błąd przetwarzania pakietu
linux-headers-amd64 (--configure): problemy z zależnościami -
pozostawianie nieskonfigurowanego dpkg: problemy z zależnościami
uniemożliwiają skonfigurowanie pakietu linux-image-amd64:
linux-image-amd64 zależy od linux-image-7.2.8+deb14-amd64 (= 7.2.8-1);
jednakże: Pakiet linux-image-7.2.8+deb14-amd64 nie jest jeszcze
skonfigurowany. dpkg: błąd przetwarzania pakietu linux-image-amd64
(--configure): problemy z zależnościami - pozostawianie
nieskonfigurowanego Wystąpiły błędy podczas przetwarzania:
linux-headers-7.2.6+deb14-amd64 linux-headers-7.2.8+deb14-amd64
linux-image-7.2.8+deb14-amd64 linux-image-7.2.6+deb14-amd64
linux-headers-amd64 linux-image-amd64 Błąd: Sub-process /usr/bin/dpkg
returned an error code (1) -- System information -- Operating System:
Debian GNU/Linux forky/sid KDE Plasma Version: 6.7.4 KDE Frameworks
Version: 6.30.0 Qt Version: 6.10.2 Kernel Version: 7.1.13+deb14-amd64
(64-bit) Graphics Platform: Wayland Processors: 8 × Intel® Core™
i7-2600K CPU @ 3.40GHz Memory: 32 GiB of RAM (31.3 GiB usable) Graphics
Processor: NVIDIA GeForce GTX 1060 6GB Thanks, Arek

Dear Maintainer

I tested nvidia-550-linux-7.2.patch.patch on this system:

- Debian Unstable

- Linux kernel 7.2.9+deb14-amd64

- NVIDIA GeForce RTX 3060.

It works fine.

Thanks

Tula

Version: 550.163.01-5.1

Dear Maintainer,

I tested the attached patch
nvidia-550-linux-7.2.patch
on the following system:

- Debian testing/forky amd64
- NVIDIA GeForce RTX 4070 SUPER
- PCI ID: 10de:2783
- Linux kernel: 7.2.9+deb14-amd64
- nvidia-kernel-dkms: 550.163.01-5.1
- DKMS: 3.2.2-1

The patch was applied manually to:

/usr/src/nvidia-current-550.163.01

After applying the patch, the NVIDIA DKMS module built and installed successfully for the 7.2 kernel.

The system boots normally with Linux 7.2.9. The NVIDIA driver works correctly:

- uname -r reports 7.2.9+deb14-amd64
- nvidia-smi works and detects the GeForce RTX 4070 SUPER- the graphical desktop starts normally

Thanks!

Dear Maintainer,

I tested the nvidia-550-linux-7.2.patch.patch:

- Debian GNU/Linux forky/sid
- NVIDIA RTX A3000 Mobile
- linux-image-7.2.9+deb14-amd64: 7.2.9-1
- nvidia-kernel-dkms: 550.163.01-5.1


After applying the patch, the NVIDIA DKMS module builds and works as
expected.


Thanks,
  Daniele