Bug 949097 – x11-drivers/nvidia-drivers-570.86.16: default values modeset=1 and fbdev=1 make display to freeze, with kernel output: Console: switching to colour dummy device 160x64

• Status: RESOLVED FIXED
• Product: Gentoo Linux

[Description Thibaud CANALE 2025-01-30 20:22:51 UTC]

Created attachment 917979 [details]
emerge --info x11-drivers/nvidia-drivers

Hello,

With current beta Nvidia’s drivers 570.86.16 in package x11-drivers/nvidia-drivers, the display is frozen with this message in kernel:
> Console: switching to colour dummy device 160x64

I originally thought the system was hanged while it was only the screen.

I don’t have this issue with previous beta driver 565.77.

My device is a GTX 1050 Ti, it looks like this device is still supported as I can download this same beta driver for "Linux 64-bit" on Nvidia website.

Also I think this is important to mention, I have this whole log in kernel as joined in attachment, a "bad value" in a "stack frame pointer" so there is a "unwind stack".
I have this same issue in both latest beta drivers, 565.77 and 570.86.16.

Should I ignore it, is this an issue with my configuration, or is it important?

[Comment 1 Thibaud CANALE 2025-01-30 20:24:09 UTC]

Created attachment 917980 [details]
x11-drivers/nvidia-drivers-565.77 kernel unwind stack

[Comment 2 Thibaud CANALE 2025-01-30 20:25:13 UTC]

Created attachment 917981 [details]
x11-drivers/nvidia-drivers-570.86.16 kernel unwind stack

Add "unwind stack" for drivers 565.77 and 570.86.16

[Comment 3 Thibaud CANALE 2025-01-30 20:55:13 UTC]

Thanks to ionen, the issue was the modified value of fbdev, which is set to 1 as default in this new beta driver.
Restoring explicitly fbdev=0 for nvidia-drm (while modeset=1) does not report this `Console: switching to colour dummy device 160x64` and the display is not frozen.

I still have this unwind stack, which I suspected because also present before.

[Comment 4 Larry the Git Cow 2025-01-30 20:59:40 UTC]

The bug has been closed via the following commit(s):

https://gitweb.gentoo.org/repo/gentoo.git/commit/?id=0f86a0d31b0d811571231fbb283e861992b1eb24

commit 0f86a0d31b0d811571231fbb283e861992b1eb24
Author:     Ionen Wolkens <[e-mail]>
AuthorDate: 2025-01-30 20:43:23 +0000
Commit:     Ionen Wolkens <[e-mail]>
CommitDate: 2025-01-30 20:59:12 +0000

    x11-drivers/nvidia-drivers: override the fbdev=1 default
    
    Previous use of fbdev=1 had issues ranging from problems
    switching between X/wayland and the console, graphical
    corruption, failing to resume from sleep, and also straight
    up failure like bug #949097.
    
    Some of these issues may be resolved in 570.x, but real status is
    unknown, and still getting new bug reports about it.
    
    This may work properly for some configurations (probably simpledrm
    + wayland, aka Fedora's case where fbdev=1 is needed), but still
    seem too volatile to be a default on a wide variety of configurations
    like with Gentoo.
    
    Already had a postinst elog out of worry, but having potentially
    several users deal with this manually is unwanted.
    
    Will likely revisit after issues been ironed out as it is needed
    if we want to get rid of deprecated efifb in favour of simpledrm
    on gentoo-kernel's defaults eventually.
    
    Closes: https://bugs.gentoo.org/949097
    Signed-off-by: Ionen Wolkens <[e-mail]>

 x11-drivers/nvidia-drivers/files/nvidia-570.conf             | 12 +++++++-----
 ...s-570.86.16.ebuild => nvidia-drivers-570.86.16-r1.ebuild} | 12 ------------
 2 files changed, 7 insertions(+), 17 deletions(-)

[Comment 5 Ionen Wolkens 2025-02-01 01:11:21 UTC]

It doesn't really change that I do not feel it's safe to have as default "yet", but for this particular issue guess we could check that DRM_FBDEV_EMULATION is set.

<thican> Hello again ionen, in case it might help other people, I was able to got Nvidia’s driver working without the "fbdev=0" option (so with its default value of fbdev=1 then) simply with CONFIG_DRM_FBDEV_EMULATION=y. I also changed CONFIG_DRM_BOCHS from 'm' to 'not set' and got CONFIG_DRM_NOUVEAU=m instead, which makes more sense, so I keep CONFIG_DRM_KMS_HELPER and CONFIG_DRM_TTM_HELPER enabled, but it’s the same.

[Comment 6 Ionen Wolkens 2025-02-01 01:28:48 UTC]

(In reply to Ionen Wolkens from comment #5)
> It doesn't really change that I do not feel it's safe to have as default
> "yet"
On that note, I do see non-Gentoo users having other issues on nvidia forums and solving it with fbdev=0. Odds are NVIDIA will look at these and it'll improve in future versions now that it's an upstream default.

[Comment 7 Larry the Git Cow 2025-02-01 01:31:40 UTC]

The bug has been referenced in the following commit(s):

https://gitweb.gentoo.org/repo/gentoo.git/commit/?id=3cfd7a4f51f154d59f90ca04d399ff26db56fff3

commit 3cfd7a4f51f154d59f90ca04d399ff26db56fff3
Author:     Ionen Wolkens <[e-mail]>
AuthorDate: 2025-02-01 01:12:36 +0000
Commit:     Ionen Wolkens <[e-mail]>
CommitDate: 2025-02-01 01:26:15 +0000

    x11-drivers/nvidia-drivers: check for CONFIG_DRM_FBDEV_EMULATION
    
    Should've known about this given the TTM bit, but kind of forgot.
    
    Keep it optional, like the rest of the DRM bits a system can work
    without these if you know what you're doing (e.g. disabled KMS_HELPER
    needs modeset=0 and a custom Xorg.conf). TTM on the other hand is
    required *if* FBDEV is enabled else it results in a build failure
    on recent kernels.
    
    Plans is mostly to make most people enable it before we do the switch
    "eventually".
    
    For what it's worth, only check in 570+ given there's no plans to ever
    enable fbdev=1 by default in old branches, legacy ones did not even
    support it.
    
    Bug: https://bugs.gentoo.org/949097
    Signed-off-by: Ionen Wolkens <[e-mail]>

 x11-drivers/nvidia-drivers/nvidia-drivers-570.86.16-r1.ebuild | 4 ++++
 1 file changed, 4 insertions(+)

[Comment 8 sin-ack 2025-02-01 19:42:02 UTC]

Just as a data point, I have the following setup:

- RTX 4090 "Laptop GPU" - Same die with smaller memory/lower clocks AFAICT
- Linux 6.13.0
  - CONFIG_FB_EFI=y
  - CONFIG_DRM_SIMPLEDRM=m
- x11-drivers/nvidia-drivers 570.86.16-r1
- nvidia-drm.modeset=1
- nvidia-drm.fbdev=1

Works: suspend/resume, switching to/from framebuffer (Wayland)
Doesn't work: Hibernation
Notes: Early kernel logs aren't displayed until nvidia driver is loaded.