MATE Desktop session fails to start with NVIDIA 495 & 510 drivers

It seems that this might be a regression introduced via the fix for bug 1957790.

Please show the output of "dpkg -l nvidia-driver"

ian@pc:~$ dpkg -l nvidia-driver
dpkg-query: no packages found matching nvidia-driver

:)

If I look for anything starting with that, it's

ian@pc:~$ dpkg -l nvidia-driver*
Desired=Unknown/Install/Remove/Purge/Hold
| Status=Not/Inst/Conf-files/Unpacked/halF-conf/Half-inst/trig-aWait/Trig-pend
|/ Err?=(none)/Reinst-required (Status,Err: uppercase=bad)
||/ Name Version Architecture Description
+++-====================-===========================-============-=============>
ii nvidia-driver-470 470.103.01-0ubuntu0.21.10.1 amd64 NVIDIA driver>
un nvidia-driver-binary <none> <none> (no descripti>

Looking for everything nvidia...

ian@pc:~$ dpkg -l nvidia*
Desired=Unknown/Install/Remove/Purge/Hold
| Status=Not/Inst/Conf-files/Unpacked/halF-conf/Half-inst/trig-aWait/Trig-pend
|/ Err?=(none)/Reinst-required (Status,Err: uppercase=bad)
||/ Name Version Architecture Description
+++-=========================-===========================-============-===============================================
un nvidia-384 <none> <none> (no description available)
un nvidia-390 <none> <none> (no description available)
un nvidia-common <none> <none> (no description available)
un nvidia-compute-utils <none> <none> (no description available)
rc nvidia-compute-utils-390 390.116-0ubuntu1 amd64 NVIDIA compute utilities
rc nvidia-compute-utils-430 430.50-0ubuntu2 amd64 NVIDIA compute utilities
rc nvidia-compute-utils-435 435.21-0ubuntu7 amd64 NVIDIA compute utilities
rc nvidia-compute-utils-450 450.102.04-0ubuntu0.20.10.2 amd64 NVIDIA compute utilities
ii nvidia-compute-utils-470 470.103.01-0ubuntu0.21.10.1 amd64 NVIDIA compute utilities
rc nvidia-compute-utils-495 495.46-0ubuntu0.21.10.1 amd64 NVIDIA compute utilities
rc nvidia-compute-utils-510 510.47.03-0ubuntu0.21.10.1 amd64 NVIDIA compute utilities
rc nvidia-dkms-390 390.116-0ubuntu1 amd64 NVIDIA DKMS package
rc nvidia-dkms-430 430.50-0ubuntu2 amd64 NVIDIA DKMS package
rc nvidia-dkms-435 435.21-0ubuntu7 amd64 NVIDIA DKMS package
rc nvidia-dkms-450 450.102.04-0ubuntu0.20.10.2 amd64 NVIDIA DKMS package
ii nvidia-dkms-470 470.103.01-0ubuntu0.21.10.1 amd64 NVIDIA DKMS package
rc nvidia-dkms-495 495.46-0ubuntu0.21.10.1 amd64 NVIDIA DKMS package
rc nvidia-dkms-510 510.47.03-0ubuntu0.21.10.1 amd64 NVIDIA DKMS package
un nvidia-dkms-kernel <none> <none> (no description available)
ii nvidia-driver-470 470.103.01-0ubuntu0.21.10.1 amd64 NVIDIA driver metapackage
un nvidia-driver-binary <none> <none> (no description available)
un nvidia-egl-wayland-common <none> <none> (no description av...

After installing, but before rebooting...

ian@pc:~$ dpkg -l nvidia*
Desired=Unknown/Install/Remove/Purge/Hold
| Status=Not/Inst/Conf-files/Unpacked/halF-conf/Half-inst/trig-aWait/Trig-pend
|/ Err?=(none)/Reinst-required (Status,Err: uppercase=bad)
||/ Name Version Architecture Description
+++-=========================-===========================-============-===============================================
un nvidia-384 <none> <none> (no description available)
un nvidia-390 <none> <none> (no description available)
un nvidia-common <none> <none> (no description available)
un nvidia-compute-utils <none> <none> (no description available)
rc nvidia-compute-utils-390 390.116-0ubuntu1 amd64 NVIDIA compute utilities
rc nvidia-compute-utils-430 430.50-0ubuntu2 amd64 NVIDIA compute utilities
rc nvidia-compute-utils-435 435.21-0ubuntu7 amd64 NVIDIA compute utilities
rc nvidia-compute-utils-450 450.102.04-0ubuntu0.20.10.2 amd64 NVIDIA compute utilities
rc nvidia-compute-utils-470 470.103.01-0ubuntu0.21.10.1 amd64 NVIDIA compute utilities
rc nvidia-compute-utils-495 495.46-0ubuntu0.21.10.1 amd64 NVIDIA compute utilities
ii nvidia-compute-utils-510 510.47.03-0ubuntu0.21.10.1 amd64 NVIDIA compute utilities
rc nvidia-dkms-390 390.116-0ubuntu1 amd64 NVIDIA DKMS package
rc nvidia-dkms-430 430.50-0ubuntu2 amd64 NVIDIA DKMS package
rc nvidia-dkms-435 435.21-0ubuntu7 amd64 NVIDIA DKMS package
rc nvidia-dkms-450 450.102.04-0ubuntu0.20.10.2 amd64 NVIDIA DKMS package
rc nvidia-dkms-470 470.103.01-0ubuntu0.21.10.1 amd64 NVIDIA DKMS package
rc nvidia-dkms-495 495.46-0ubuntu0.21.10.1 amd64 NVIDIA DKMS package
ii nvidia-dkms-510 510.47.03-0ubuntu0.21.10.1 amd64 NVIDIA DKMS package
un nvidia-dkms-kernel <none> <none> (no description available)
ii nvidia-driver-510 510.47.03-0ubuntu0.21.10.1 amd64 NVIDIA driver metapackage
un nvidia-driver-binary <none> <none> (no description available)
un nvidia-egl-wayland-common <none> <none> (no description available)
un nvidia-kernel-common <none> <none> (no description available)
rc nvidia-kernel-common-390 390.116-0ubuntu1 amd64 Shared files used with the kernel module
rc nvidia-kernel-common-430 430.50-0ubuntu2 amd64 Shared files used with the kernel module
rc nvidia-kernel-common-435 435.21-0ubuntu7 amd64 Shared files used with the kernel module
rc nvidia-kernel-common-450 450.102.04-0ubuntu0.20.10.2 amd64 Shared files used with the kernel module
rc nvidia-kernel-common-470 470.103.01-0ubuntu0.21.10.1 amd64 Shared files used with the kernel module...

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

My experience with the 495 and 510 drivers is identical to those described in the opening post.

  * I'm using Ubuntu MATE 21.10 with NVIDIA RTX 3090.
  * The 470.103.01-0ubuntu0.21.10.1 and 470.74-0ubuntu1 drivers both work/worked without issue.
  * The 495 and 510 drivers both fail to bring up the display server

I have exactly the same problem :
* Ubuntu MATE 21.10, NVIDIA GeForce 1060
* 470.103.1 works fine
* 495 and 510 both fail after reboot.

This issue is only reproducible when Marco (the MATE Window Manager) is built with Xpresent support and Marco's compositing manager is enabled. This has been the default configuration in Debian & Ubuntu packaging of Marco for some years. I am not sure if this issue is due to a regression in NVIDIAs driver or if Marcos Xpresent support is at fault.

That said, the reports of this issue I've seen elsewhere (NVIDIA forums, Linux Mint forums, upstream MATE Desktop GitHub) all indicate this problem is only observed with MATE Desktop.

A few solutions are available:

 - Apply a gschema override to disable Marco's compositing manager. Not a great option since it degrades the user experience and it can be easily re-enabled via the Control Center leading to a non-functioning desktop session.
 - Upstream are discussing adding a gsettings option to disable Xpresent. Again, this could be overridden by a user and leads to a broken desktop session.
 - I am investigating building Marco without Xpresent, which will retain the compositing manager functionality.
 - By coincidence, I have created a wrapper for Marco that uses the picom compositor several weeks ago. This is presented as an alternative Window Manager option in MATE Tweak and currently shipping in Jammy.

I will research the options more thoroughly before choosing a way forward.

I've built Metacity 3.42 for Ubuntu 21.10 with Xpresent support, which is behind a gsettings option that defaults to xrender. When Metacity is set as the required window manager in MATE with Xpresent set as the compositor the same failure to bring up a desktop session is encountered.

I have also built Marco without Xpresent and as expected it fallsback to Xrender which works with the new NVIDIA drivers and also preserves compositor support.

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

I've uploaded Marco to Jammy with Xpresent disabled. I've created a PPA for users of Ubuntu MATE 20.04 and 21.10 that hosts rebuilds of Marco with Xpresent disabled.

  * https://launchpad.net/~ubuntu-mate-dev/+archive/ubuntu/marco

These are temporary measures while the root cause of the issue is patched.

The root cause of this issue appears to be a bug in Xorg Xserver for which a patch exists:

  * https://gitlab.freedesktop.org/xorg/xserver/-/issues/1275
  * https://salsa.debian.org/xorg-team/xserver/xorg-server/-/commit/4aa04b0377a5b916575792c81fd2e6551913f589
  * https://bugs.launchpad.net/ubuntu/+source/xorg-server/+bug/1959995

This bug was fixed in the package marco - 1.26.0-1ubuntu0

---------------
marco (1.26.0-1ubuntu0) jammy; urgency=medium

  * debian/control:
    + Drop B-D: libxpresent-dev (LP: #1960621)

 -- Martin Wimpress <email address hidden> Fri, 25 Feb 2022 11:21:33 +0000

Thank you, Martin.

I do indeed have Marco with the 'adaptive compositor' (it's the default, I think...)

It's been ages since I used Compiz, but... :)

It turns out that I may have misunderstood - the same thing happens for me when telling MATE Tweak that I want to use Compiz rather than Marco as my window manager.

The switch to Compiz certainly happens (I am reminded of why I stopped using it..) and it is sticky between reboots.

So either it's not just a Marco thing or some part of Marco is retained when using Compiz.

.. and the PPA means I am now running the 510 drivers successfully.

Thanks again, Martin.

Hello Martin,

I noticed that you pushed a release to Jammy which disables Xpresent to work around this issue. As expected, this re-introduces screen tearing on my system running Intel Iris Xe integrated graphics.

I've manually switched to the most recent release with Xpresent still enabled and the tearing is gone.

According to bug 1959995, this issue has been fixed in Jammy. If that's true, do you have any plans on re-enabling Xpresent? Should a new bug be opened for that?

https://bugs.launchpad.net/ubuntu/+source/xorg-server/+bug/1959995

Shamelessly piggy-backing on this thread, is there a target/plan for an official 20.04 fix? Is the above PPA still the way to go? Thanks all!