All GPU functionality stops working whenever an updated Nvidia driver is installed and until the system is rebooted

[Alistair Buxton]

To reproduce:

1. Install Ubuntu
2. Switch to Nvidia driver
3. Wait

The default for Ubuntu is to install updates automatically in the background. Eventually an updated Nvidia driver will be installed, and then all graphics acceleration stops working for any new programs which are launched. Note that it continues working for anything that was already running.

This results in the following things happening:

1. Any program that requires OpenGL will crash on startup. For example glxgears:

```
X Error of failed request: BadValue (integer parameter out of range for operation)
  Major opcode of failed request: 152 (GLX)
  Minor opcode of failed request: 3 (X_GLXCreateContext)
  Value in failed request: 0x0
  Serial number of failed request: 36
  Current serial number in output stream: 37
```

2. Any program that uses CUDA will fail in a similar way.

3. Programs that can fall back to software rendering will do so. For example, Firefox will start using software rendering, causing it to max out several CPU cores on heavy websites like Discord.

Expected Result:

One of the following:

1. Don't automatically install driver updates in the background.
2. Install them in a way that doesn't prevent them from working until a reboot.
3. Display a message informing the user that the drivers were updated and requesting a reboot.

User Impact

It can be difficult for the user to notice this has happened for the following reasons:

1. There is no notification that the driver has been updated.
2. Most software will fall back to software rendering and just run slowly for no obvious reason.
3. Any software that was already running before the update will continue working properly.

So then the user gets left with a system that runs really slowly until they notice there is a problem because a program won't run, or they happened to notice that Firefox is using 500% CPU. This may be some time after the driver was updated. It is difficult for the user to work out that they need to reboot.

ProblemType: Bug
DistroRelease: Ubuntu 23.04
Package: nvidia-driver-525 525.116.04-0ubuntu0.23.04.1
ProcVersionSignature: Ubuntu 6.2.0-20.20-generic 6.2.6
Uname: Linux 6.2.0-20-generic x86_64
NonfreeKernelModules: nvidia_modeset nvidia
ApportVersion: 2.26.1-0ubuntu2
Architecture: amd64
CasperMD5CheckResult: pass
CurrentDesktop: XFCE
Date: Sun Jun 18 02:44:05 2023
InstallationDate: Installed on 2023-06-01 (16 days ago)
InstallationMedia: Xubuntu 23.04 "Lunar Lobster" - Release amd64 (20230414.2)
RebootRequiredPkgs: Error: path contained symlinks.
SourcePackage: nvidia-graphics-drivers-525
UpgradeStatus: No upgrade log present (probably fresh install)

[Alistair Buxton (a-j-buxton)]

Note this is also a regression: it never used to happen on eg 18.04 and earlier, and started happening somewhere around 20.04.

[Daniel van Vugt (vanvugt)]

I think the easy solution here is for (major ABI breaking) NVIDIA updates to mandate a reboot as part of their packaging.

[Alistair Buxton (a-j-buxton)]

It happened again today.

[Daniel van Vugt (vanvugt)]

What driver version did you upgrade from/to today?

[Alistair Buxton (a-j-buxton)]

As best as I can tell, nvidia-525-6.2.0-23 to nvidia-525-6.2.0-26.

[Launchpad Janitor (janitor)]

Status changed to 'Confirmed' because the bug affects multiple users.

[Alberto Milone (albertomilone)]

I think solution #3 should be the easiest to implement, as if automatic updates are enabled, your kernel is going to be updated in the background, which might pull in a new NVIDIA driver, and break things.