Re: [gnumach] /dev/cons/kbd rejects kbd_drv.so set-event-mode ioctl with EBADF; please implement KDSKBMODE or expose /dev/input/event*

Hi NexusSfan,

Thanks for the quick reply.

> Are you sure it is at /dev/cons/kbd and not /dev/kbd?

Both names reach the same translator.  /dev/kbd is a Hurd-shaped
symlink to /dev/cons/kbd, and /dev/cons/kbd is the chrdev.

On the canonical Debian GNU/Hurd 0.9 image my original report was
against (hurd 1:0.9.git20230520-7, gnumach 1.8+git20230410-1):

  showtrans /dev/kbd          -> /hurd/symlink cons/kbd
  showtrans /dev/cons/kbd     -> /hurd/chrdev 156 33685736
  showtrans /dev/mouse        -> /hurd/symlink cons/mouse
  showtrans /dev/cons/mouse   -> /hurd/chrdev 156 33685736

kbd_drv.so's default device is /dev/kbd (the upstream X.org
default for the "kbd" input driver, no override in my xorg.conf).
The open path resolves through the symlink to /dev/cons/kbd, the
chrdev rejects the KDSKBMODE-shaped ioctl with EBADF, and the
line that surfaces in Xorg.0.log is "Cannot set event mode on
keyboard (Bad file descriptor)" from kbd_drv.so.  So the path
in the message is /dev/cons/kbd because that is where the
chrdev (and the failure) lives, but the path kbd_drv.so actually
called open(2) on is /dev/kbd.

I also re-probed on a more recent Hurd image
(hurd 1:0.9.git20251029-7+b1, gnumach 1.8+git20260224, xserver
2:21.1.21-1) and the surface has shifted in an even less helpful
direction:

  ls -l /dev/kbd
    lrw-r--r-- 1 root root 8 ... /dev/kbd -> cons/kbd
  ls -l /dev/cons
    -rw------- 1 root root 0 ... /dev/cons
  ls /dev/cons/kbd
    ls: cannot access '/dev/cons/kbd': Not a directory

/dev/cons is no longer a directory at all on this newer image; it
is a 0-byte regular file.  So /dev/kbd -> cons/kbd is now a
dangling symlink and kbd_drv.so will fail even earlier (open path,
not ioctl).  /dev/console is still a /hurd/term, but the kbd-shaped
chrdev under /dev/cons is gone.

> Works on my machine ;)

This is the part I would really like to understand.  Could you
share:

  - which gnumach / hurd / xserver-xorg-core versions you are on
    (dpkg -l gnumach* hurd xserver-xorg-core | tail -10);
  - what your /etc/X11/xorg.conf or /etc/X11/xorg.conf.d/*.conf
    looks like for the InputDevice / InputClass blocks;
  - what `ls -l /dev/kbd /dev/cons/kbd` and
    `showtrans /dev/kbd /dev/cons/kbd` print on your image;
  - the kbd-relevant lines from your Xorg.0.log (the "Using input
    driver 'kbd'" block and the next 20 lines, so I can see which
    device path kbd_drv.so picked and whether the ioctl succeeded).

If your X works because the KDSKBMODE handler is in your gnumach
but not in canonical's, that pins the fix on the gnumach side and
makes it a concrete merge target.  If instead you are using a
different input driver, or AutoAddDevices "off" with a manual
layout that skips kbd_drv entirely, that is a usable workaround
worth documenting on this thread so other people stuck on the same
wall can adopt it.

Thanks,
Borja Tarraso
[email protected]