RMAPI_GPU_LOCK_INTERNAL alloc requested without holding the RMAPI lock about open-gpu-kernel-modules HOT 6 OPEN

Can confirm the issue on NixOS as well. Has been happening for earlier kernel versions on my system as well but currently it is
Nvidia:
Hardware: GeForce 3080Ti (mobile)
Version: 545.29.06
Kernel: 6.6.8

In my case most of the time this happens when waking up again from suspend or the machine sitting around unused for some time

from open-gpu-kernel-modules.

Same here, just hangup after updating to 545 today

Possible Fix

After showing the above error message, I found a line at the very end hdaudio hdaudioCOD2:unable to configure disabling. After searching online, I tried the fix as below, and able to log back into ubuntu normally.

# use recovery mode and enter root shell
# nano /etc/default/grub and make below change

# original
"GRUB_LINUX_DEFAULT="quiet splash nomodeset"

# to
"GRUB_LINUX_DEFAULT="quiet splash"

# then `update-grub` and `reboot`

from open-gpu-kernel-modules.

Nvidia folks, any chance we can get this fixed in next release?

from open-gpu-kernel-modules.

This bug is not fun! At a minimum its an indefinite delay and at worst its a crash on shutdown with no ability to open a tty and save the day.

This gets spammed in dmesg:

NVRM rmapiAllocWithSecInfo: RMAPI_GPU_LOCK_INTERNAL alloc requested without holding the RMAPI lock

I also get this spam in dmesg too:

[  122.596895] NVRM nbsiReadRegistryDword: osReadRegistryDword called in Sleep path can cause excessive delays!
[  122.596903] NVRM nvAssertFailedNoLog: Assertion failed: 0 @ nbsi_osrg.c:107

$ cat /proc/driver/nvidia/version
outputs:
NVRM version: NVIDIA UNIX Open Kernel Module for x86_64 545.23.08 Release Build (dvs-builder@U16-I3-A16-1-1)

from open-gpu-kernel-modules.

Hi, thanks for the report! This is tracked internally as bug 4074148. Hard to say which release will get the fix due to schedules and release branching.

For whatever it's worth, the root cause of the print from that particular call stack (rm_power_management()) was found to be "harmless", except for the print spam. Any other issues you are seeing are likely to be independent and deserve a separate bug report.

from open-gpu-kernel-modules.

Still happening with 550.78.

[   15.007542] NVRM: loading NVIDIA UNIX Open Kernel Module for x86_64  550.78  Release Build  (dvs-builder@U16-I1-N08-06-4)  Sun Apr 14 06:38:24 UTC 2024
[   15.302560] NVRM: testIfDsmSubFunctionEnabled: GPS ACPI DSM called before _acpiDsmSupportedFuncCacheInit subfunction = 11.
[   36.803975] NVRM: rmapiAllocWithSecInfo: RMAPI_GPU_LOCK_INTERNAL alloc requested without holding the RMAPI lock

from open-gpu-kernel-modules.