On Fri, Aug 22, 2025 at 09:33:47AM -0600, Theo de Raadt wrote:
> Mike Larkin <mlarkin@nested.page> wrote:
>
> > On Fri, Aug 22, 2025 at 10:56:38AM -0400, Brandon Mercer wrote:
> > > On Fri, Aug 22, 2025, at 10:42 AM, Theo de Raadt wrote:
> > > > Mark Kettenis <mark.kettenis@xs4all.nl> wrote:
> > > >
> > > > > > From: "Theo de Raadt" <deraadt@openbsd.org>
> > > > > > Date: Fri, 22 Aug 2025 08:31:19 -0600
> > > > > >
> > > > > > Why not invert this with a "donotconnect" variable, then your diff
> > > > > > shrinks significantly.
> > > > >
> > > > > Not really; the struct wkbddev_attach_args is typically allocated on
> > > > > the stack, without an explicit memset, so the new member must be set.
> > > >
> > > > So change all those stack allocations to = { 0 }
> > > >
> > > > And change one driver to set .noconnect = 1;
> > > >
> > > >
> > > > Making the default noconnect is going to explode someone's head later
> > > > on when they write a new kbd driver.
> > >
> > > My reply has nothing to do with the diff and more to do with a particular use case. My typical usage is to use my OTP to sign into my machine upon boot. If I have to fiddle with wsconsctl in order to use the yubikey OTP, then my initial sign on requires me to sign in first. This makes me lean towards fixing the yubikey tools so it's easier to reprogram the default behavior of slot one not to spam OTP's on each press. I do agree that their tooling is arduous at very best and this default behavior is prohibitive.
> >
> > isn't this what /etc/hotplug is for?
>
> Amazing suggestion to add the hotplug device-tree race problem to this :)
>

never seen any issues personally; what race problem is this?