Kernel 5.15.33 AARCH64 mouse movement with V3D

[Dark-Sky]

Describe the bug

While booted into KMS or FKMS V3D with kernel 5.15.33 64bit the mouse movement is very erratic and when moving the mouse while using glxgears the gears completely stop until you cease moving the mouse. The same thing happens while playing a youtube video with the video stopping while moving the mouse. Everything seems ok if booted using llvmpipe.

Steps to reproduce the behaviour

Boot using KMS or FKMS V3D and start glxgears -info or a youtube video in the browser and move the mouse. Also notice erratic behavior of the mouse at all times when booted into the desktop using KMS or FKMS V3D.

Device (s)

Raspberry Pi 4 Mod. B

System

[ray@pi4 ~]$ vcgencmd version
Mar 24 2022 13:19:26
Copyright (c) 2012 Broadcom
version e5a963efa66a1974127860b42e913d2374139ff5 (clean) (release) (start)

[ray@pi4 ~]$ uname -a
Linux pi4 5.15.33-1-MANJARO-ARM-RPI #1 SMP PREEMPT Mon Apr 11 13:35:56 UTC 2022 aarch64 GNU/Linux

Logs

No response

Additional context

No response

[popcornmix]

Did this start recently? I wonder if it could be #4971

[Dark-Sky]

Did this start recently? I wonder if it could be #4971

I wondered the same thing remembering the commit. It was made after I had compiled 5.15.32.

[popcornmix]

Can you test latest 5.15 kernel with those two commits reverted? It would be useful to confirm they are to blame (based on the timing, it seems likely).

[Dark-Sky]

Sure. Be glad to....

[Dark-Sky]

@popcornmix

Sorry it took so long to respond back. My boy and his family showed up and had to wait for a thunderstorm to pass.

I reverted the 2 commits and the mouse issue went away on 5.15.33.

[popcornmix]

@mripard looks like #4971 has caused a performance regression.

[Dark-Sky]

I compiled 5.18-rc2 again today to add commits since my last compile and it now has the erratic mouse and V3D issue. Although I still can not get kernel 5.17.2 to compile yet to test it also looks to me it will have the issue also. #4987

[popcornmix]

I believe 5.15, 5.16, 5.17 and 5.18 all have the drm/atomic-helpers: remove legacy_cursor_update hacks commit and so I'd expect all to have this issue.

[mripard]

I wonder if it's a similar issue to #4571. Does it occur only when playing a video or rendering on the GPU, or does it also stall the rest of the display (so something like displaying a huge chunk of text in a terminal would be stalled as well?)

Either way, it looks like a severe regression that won't get fixed fast, we should probably revert it for now

[Dark-Sky]

I will test further in a little bit. Right now all I tested was with any graphic with movement ceased when moving the mouse but started back up when you rest the mouse. The mouse was very erratic/laggy trying to move the mouse around the Desktop when no graphic movement was being done with a program.

I have reverted all 3 kernels I maintain for our repo at the moment at released them.

[Dark-Sky]

Oops. I deleted the bad kernel. I can recompile if needed.

[popcornmix]

Not sure I can reproduce on RPiOS 32bit. Kernel: 5.15.33-v7l+
Booted to desktop and run glxgears. Mouse moves okay and gears do not stall.
Tried 1080p60 and 4kp60. Tried running neverball too.
Tried with mutter disabled and enabled.

[Dark-Sky]

Mine is 64 bit with XFCE and mesa 22.0.1-3.

[Dark-Sky]

I have a 32bit arch-arm image I here with MATE DE. I will see what it does. We mirror arch-arm's repo so it should be the same program versions but 64bit after I update everything.