From: Crystal Kolipe <kolipe.c@exoticsilicon.com>
Subject: Re: Obscure X11 bug with PS/2 keyboard and mousekeys
To: tech@openbsd.org
Date: Thu, 26 Jun 2025 03:45:24 -0300

Ping?

The work-around I mentioned fixes the problem on all of the affected machines
here, and doesn't break those using USB keyboards.

But it seems like the wrong approach.

On Tue, Jun 03, 2025 at 03:27:55AM -0300, Crystal Kolipe wrote:
> This is a difficult bug to encounter, but nevertheless...
> 
> 
> Background info:
> 
> X11 has a 'mousekeys' feature, which allows the user to control the mouse
> pointer using, (by default), the numeric keypad.  It's configured with a line
> such as this in .Xmodmap:
> 
> keycode 108 = Pointer_EnableKeys Pointer_Accelerate
> 
> Pointer_EnableKeys toggles the function on and off, and Pointer_Accelerate
> toggles between two modes, accelerated, where the pointer moves faster the
> longer a key is held down, and unaccelerated where it always moves one pixel
> at a time.
> 
> 
> Bug:
> 
> When using a USB keyboard, it all works as expected.
> 
> When using a PS/2 keyboard, the pointer movement is erratic:
> 
> Holding down a directional key, the pointer moves and accelerates for about a
> second, then stops, and starts again for about half a second, then repeats
> this sequence endlessly.  Note that the initial 'jump' is longer than any
> subsequent 'jumps'.
> 
> Visually, the pointer speeds looks like this:
> 
> 0 1 2 3 4 5 6 7 0 0 1 2 3 4 0 0 1 2 3 4 0 0 1 2 3 4 ....
> 
> Whereas it should look like this:
> 
> 0 1 2 3 4 5 6 7 8 9 9 ....
> 
> 
> Reason:
> 
> The regular auto-repeat functionality of the input layer is causing spurious
> keypress events to be sent to the mousekeys code.
> 
> Each repeat triggers the 'initial press' code in _XkbFilterPointerMove() in
> xkbActions.c, and resets xkbi->mouseKeysCounter to zero, causing the pointer
> movement to studder.
> 
> 
> Workaround:
> 
> Disabling typematic repeat for the keypad keys 'fixes' the issue with the
> 'mousekeys' feature:
> 
> $ for i in 79 80 81 83 85 87 88 89 ; do xset -r $i ; done
> 
> But this seems like the wrong fix, because it's only necessary with PS/2
> connected keyboards, and this functionality should really be
> hardware-independent.
> 
> Thoughts?
>