ACK: [SRU][F][PATCH 0/3] Fix nouveau error storm and unresponsive display after desktop idle timeout

[Thibault Ferrante]

Acked-by: Thibault Ferrante <thibault.ferrante at canonical.com>

On 10-12-2024 23:03, Jacob Martin wrote:
> BugLink: https://bugs.launchpad.net/bugs/2078011
> 
> SRU Justification
> 
> [Impact]
> 
> On a system with a GV100 GPU using the nouveau driver, the display becomes
> unresponsive and a storm of "nouveau 0000:07:00.0: disp: ctrl 00000080"
> messages are continuously printed to dmesg once the desktop environment reaches
> its idle timeout. This is interfering with certification testing for the DGX
> Station desktop system, as the system eventually will become unresponsive
> during testing.
> 
> [Fix]
> 
> This only affects Focal.
> 
> Backporting the following patches from K5.6 resolves the issue:
> 58ae5284f6 ("drm/nouveau/disp/gv100-: halt NV_PDISP_FE_RM_INTR_STAT_CTRL_DISP_ERROR storms")
> 5bb88d0794 ("drm/nouveau/kms/gv100-: move window ownership setup into modesetting path")
> 137c4ba716 ("drm/nouveau/kms/gv100-: avoid sending a core update until the first modeset")
> 
> [Test Case]
> 
> 1. Install desktop environment
> $ sudo apt install ubuntu-desktop
> 
> 2. Configure GDM
> $ sudo vim /etc/gdm3/custom.conf
>    => Uncomment WaylandEnable=false
>    => Configure automatic login for the `ubuntu` user by setting
>          AutomaticLoginEnable = true
>          AutomaticLogin = ubuntu
> 
> 3. Disable display timeout
> $ gsettings set org.gnome.desktop.session idle-delay 0
> 
> 4. Set graphical as the default target
> $ sudo systemctl set-default graphical.target
> 
> 5. Reboot the system
> 
> 6. Enable 1 second display timeout and wait ~10 seconds
> $ gsettings set org.gnome.desktop.session idle-delay 1
> 
> 7. Observe that after applying these patches, the display can wake up from idle
> and the system continues to be usable without a storm of "nouveau 0000:07:00.0:
> disp: ctrl 00000080" messages in dmesg.
> 
> [Where things could go wrong]
> 
> These changes affect only the nouveau driver. Issues would appear as
> misbehavior of the nouveau driver, mostly likely for Volta NVIDIA GPUs.
> 
> Ben Skeggs (3):
>    drm/nouveau/disp/gv100-: halt NV_PDISP_FE_RM_INTR_STAT_CTRL_DISP_ERROR
>      storms
>    drm/nouveau/kms/gv100-: move window ownership setup into modesetting
>      path
>    drm/nouveau/kms/gv100-: avoid sending a core update until the first
>      modeset
> 
>   drivers/gpu/drm/nouveau/dispnv50/core.h       |  6 +++++
>   drivers/gpu/drm/nouveau/dispnv50/corec37d.c   | 23 +++++++++++++++----
>   drivers/gpu/drm/nouveau/dispnv50/corec57d.c   |  9 ++++----
>   drivers/gpu/drm/nouveau/dispnv50/disp.c       | 16 +++++++++++++
>   .../gpu/drm/nouveau/nvkm/engine/disp/gv100.c  |  6 +++++
>   5 files changed, 50 insertions(+), 10 deletions(-)
>