Mouse input latency in GNOME 44.1 Wayland sessions is one frame higher than it was in 44.0 and earlier

[Daniel van Vugt]

[ Impact ]

Mouse input latency in Wayland sessions is slightly higher in 23.10 than 22.04 at the moment.

I might be imagining things and it does feel like only one frame higher latency. But experimenting with MUTTER_DEBUG_TRIPLE_BUFFERING seems to confirm my suspicion:

 BAD: MUTTER_DEBUG_TRIPLE_BUFFERING=auto (the Ubuntu default)
 BAD: MUTTER_DEBUG_TRIPLE_BUFFERING=always
 GOOD: MUTTER_DEBUG_TRIPLE_BUFFERING=never

[ Workaround ]

Add MUTTER_DEBUG_FORCE_KMS_MODE=simple to /etc/environment. Actually that was already a good idea for the best cursor performance even before this bug.

[ Test Plan ]

0. Find a machine with an Intel GPU.

1. Add this to /etc/environment: CLUTTER_DEBUG=frame-timings

2. Reboot.

3. Log in to a Wayland session and wait for the desktop to become idle.

4. Also log in via ssh from a remote machine and run:
   journalctl -f /usr/bin/gnome-shell

5. Over your idle desktop just wiggle the mouse continuously for at least 5 seconds.

6. In your ssh login, verify that the resulting log messages from gnome-shell say "[FRAME_TIMINGS]: Double buffering:" and not "[FRAME_TIMINGS]: Triple buffering:" when only the mouse pointer is moving.

[ Where problems could occur ]

In the frame rate and latency of the Ubuntu desktop experience.

[Daniel van Vugt (vanvugt)]

It's a regression in 44.1 caused by the fix for bug 2017137 / bug 2017097. Nobody noticed because one frame of latency is so hard to notice.

[Daniel van Vugt (vanvugt)]

Fixed in 45: https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/1441
but I'll wait till next week to backport it to 44 because it's Friday night and I keep changing my mind about the design of the fix.

[Robie Basak (racb)]

Thank you for the report. I noted this in https://wiki.ubuntu.com/DesktopTeam/TestPlans/Mutter for future updates that rely on the exception. Would it be appropriate to add this Test Plan to future mutter updates that are based on the exception? If so, please could you amend the wiki? Thanks!

[Daniel van Vugt (vanvugt)]

I don't think cursor motion latency is one of the most important things to test, and it's kind of too awkward for most people to test. So I would not recommend making it part of the primary test case but linking to this bug as a record of what can go wrong seems appropriate.

[Daniel van Vugt (vanvugt)]

Alright a fix is prepared for 44:
https://gitlab.gnome.org/Community/Ubuntu/mutter/-/commits/triple-buffering-v4-44/

But now I'm thinking perhaps bug 2020049 should be addressed at the same time before updating Salsa.

[Daniel van Vugt (vanvugt)]

It looks like the fix for bug 2023766 will do a better job at fixing this one too, in theory. And that fix is no longer part of triple buffering so can be patched independently.

[Daniel van Vugt (vanvugt)]

In progress again. Looks like it needs to be redesigned:
https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/1441#note_1777367

[Daniel van Vugt (vanvugt)]

A new fix is coming in triple-buffering 45:
https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/1441#note_1777992

But I want it to mature a little, and even try proposing a backport upstream to main, before backporting to 44.

[Daniel van Vugt (vanvugt)]

And another new fix:
https://gitlab.gnome.org/GNOME/mutter/-/merge_requests/1441/diffs?commit_id=ccea62f6f0f072e233c785d9285bf870d16dceb6

[Daniel van Vugt (vanvugt)]

Fix proposed to Debian:
https://salsa.debian.org/gnome-team/mutter/-/merge_requests/101

[Chow Loong Jin (hyperair)]

Tested the new patch in mutter 44.3 on Lunar. Works great, thanks!

[Daniel van Vugt (vanvugt)]

Invalid for mantic since mutter 45 uses a different approach which doesn't have this bug.

[Daniel van Vugt (vanvugt)]

Won't Fix for lunar because it's EOL next month. https://wiki.ubuntu.com/Releases