Nvidia drivers newer than the 510 series fail to load on my system, which
is a Lenovo Thinkpad P51 with a Quadro M2200 GPU, with BIOS 1.60 and ECP
1.10. I have encountered this bug with driver versions 515, 520 and now the
525 that landed in testing, as well as with a version of 525 installed
using nvidia's official installer, and kernels including 6.0.7 and 6.2.2
from xanmod and 6.1.0-5-amd64 from Debian's official repository. My system
is a mixture of packages from stable and testing, with libc6=2.36-7. Driver
version 510.108.03-1 works (but is unstable in sleep and broken in
hibernation).

Below is an excerpt from journalctl's output including what appears to be
potentially pertinent clusters of lines to me. All logs are from a boot on
the xanmod 6.2.2 kernel, but there is no appreciable difference in the
relevant outputs when running with Debian's 6.1.0-5. The operational
failure points seem to be the ones pertaining to RmInitAdapter and
NvKmsKapiDevice, but I'm not sure what, if any, causality there is between
the two issues.

Mar 08 21:28:46 tangerine kernel: Command line:
BOOT_IMAGE=/boot/vmlinuz-6.2.2-x64v1-xanmod1 root=(***) ro quiet
mitigations=off psi=1 nvidia-drm.modeset=1
(...)
Mar 08 21:28:46 tangerine kernel: nvidia: module verification failed:
signature and/or required key missing - tainting kernel
Mar 08 21:28:46 tangerine kernel: nvidia-nvlink: Nvlink Core is being
initialized, major device number 235
Mar 08 21:28:46 tangerine kernel:
Mar 08 21:28:46 tangerine kernel: nvidia 0000:01:00.0: vgaarb: changed VGA
decodes: olddecodes=io+mem,decodes=none:owns=io+mem
Mar 08 21:28:46 tangerine kernel: NVRM: loading NVIDIA UNIX x86_64 Kernel
Module  525.89.02  Wed Feb  1 23:23:25 UTC 2023
Mar 08 21:28:46 tangerine kernel: nvidia-modeset: Loading NVIDIA Kernel
Mode Setting Driver for UNIX platforms  525.89.02  Wed Feb  1 23:09:40 UTC
2023
Mar 08 21:28:46 tangerine systemd[1]: Finished Rebuild Hardware Database.
Mar 08 21:28:46 tangerine systemd[1]: Starting Rule-based Manager for
Device Events and Files...
Mar 08 21:28:46 tangerine kernel: [drm] [nvidia-drm] [GPU ID 0x00000100]
Loading driver
Mar 08 21:28:46 tangerine kernel: ACPI Warning: \_SB.PCI0.PEG0.PEGP._DSM:
Argument #4 type mismatch - Found [Buffer], ACPI requires [Package]
(20221020/nsarguments-61)
Mar 08 21:28:46 tangerine systemd[1]: Started Rule-based Manager for Device
Events and Files.
Mar 08 21:28:46 tangerine systemd[1]: Starting Show Plymouth Boot Screen...
(...)
Mar 08 21:28:55 tangerine kernel: NVRM: GPU 0000:01:00.0: RmInitAdapter
failed! (0x25:0x65:1457)
Mar 08 21:28:55 tangerine kernel: NVRM: GPU 0000:01:00.0: rm_init_adapter
failed, device minor number 0
Mar 08 21:28:55 tangerine kernel: [drm:nv_drm_load [nvidia_drm]] *ERROR*
[nvidia-drm] [GPU ID 0x00000100] Failed to allocate NvKmsKapiDevice
Mar 08 21:28:55 tangerine kernel: [drm:nv_drm_probe_devices [nvidia_drm]]
*ERROR* [nvidia-drm] [GPU ID 0x00000100] Failed to register device
Mar 08 21:28:55 tangerine systemd-modules-load[306]: Inserted module
'nvidia_drm'

Some possibly pertinent information from nvidia-bug-report.log.gz:

Control: tag -1 moreinfo upstream

Do you have by chance installed nvidia-open-kernel-dkms instead of
nvidia-kernel-dkms? In that case, please switch to the proprietary
module (nvidia-kernel-dkms).

You could also try with the 530 driver from experimental.

You could also try the tesla-470 driver.


Andreas

I did not have nvidia-open-kernel-dkms installed when performing those
experiments (and never seem to have had it installed on my system, as there
is no rc record in dpkg). I believe that I also tested it with the 530
driver at the time, with the same results.

Do you expect different results with the tesla-470 driver? I'm fairly sure
that the 470 series of the standard driver worked for me, as 510.108.03-1,
which I am using right now, does (though I'm not sure if hibernation was
already broken in 470), but the lack of some features that were added
between then and 510 seems likely to cause compatibility issues at this
point.

Best wishes,
Matvey