kern/141011: [usb] Encrypted root, geli password at boot; enter key never release

[Hans Petter Selasky]

On Monday 11 January 2010 00:27:50 Wiktor Niesiobedzki wrote:
> 2010/1/10 Hans Petter Selasky <hselasky at c2i.net>:
> > On Sunday 10 January 2010 17:40:03 Wiktor Niesiobedzki wrote:
> >> The following reply was made to PR kern/141011; it has been noted by
> >> GNATS.
> >>
> >>
> >>  Hello Hans,
> >>
> >>  I've tried to comment out this section in your version of ukbd.c, but
> >>  to no extent. Still some events get discarded somehow. Shall I enable
> >>  some usb debug-flags to help getting to the core of the problem?
> >
> > Hi,
> >
> > The information we need is inside ukbd.c . Try to add printfs in code
> > enabling and disabling the keyboard. Especially in the USB keyboard read
> > character function, and try to figure out where the missing characters
> > are going.
> 
> Hi,
> 
> Nothing new in enabling and disabling the keyboard. The only
> intresting thing is, that although I have only one keyboard, the
> ukbd_enable is called twice.
> 
> What I found somehow helping, is some additional printfs in ukbd_read_char.
> Especially in beginning part:
>         printf("enter ukbd_read_char\n");
>         if (!KBD_IS_ACTIVE(kbd)) {
>                 printf("kbd is not active - exiting\n");
>                 return (NOKEY);
>         }
> 
>         if (sc->sc_flags & UKBD_FLAG_POLLING) {
>                 if (!mtx_owned(&Giant)) {
>                         /* XXX cludge */
>                         printf("ukbd_read_char: UKBD_FLAG_POOLING is
> set and giant is not owned - recursing and locking giant\n");
>                         int retval;
>                         mtx_lock(&Giant);
>                         retval = ukbd_read_char(kbd, wait);
>                         mtx_unlock(&Giant);
>                         return (retval);
>                 }
>         } else {
>                 /* XXX the keyboard layer requires Giant */
>                 if (!mtx_owned(&Giant))
>                         printf("ukbd_read_char: UKBD_FLAG_POOLING is
> not set and Giant is not owned -> DISCARDING KEY!\n");
>                         return (NOKEY);
>         }
> 
> 
> Then i see this message:
> "ukbd_read_char: UKBD_FLAG_POOLING is set and giant is not owned -
> recursing and locking giant"
> 
> Full message is (for entering one character is something like that):
> ukbd_read_char: UKBD_FLAG_POOLING is set and giant is not owned -
> recursing and locking giant
> enter ukbd_read_char
> a (echo from GELI)
> enter ukbd_read_char
> ukbd_read_char: UKBD_FLAG_POOLING is set and giant is not owned -
> recursing and locking giant
> enter ukbd_read_char
> enter ukbd_read_char
> 
> And supriselingly - no characters were lost, when I was typing my
>  passphareses.
> 
> As far as I know (and I may know wrong - please correct me here) -
> there is some relation between Giant and printfs (something like -
> when you printf on console, you have to have a giant) but this is just
> my wild guessing (maybe from java - where System.out.println is
> synchronized method). Is it possible, that my printf's affected how
> Giant is handled within ukbd and thus - I don't see misbehaviour of
> ukbd now?
> 
> Or was it just my luck now? (I hardly believe that - I usulally got ~8
> characters lost for a boot sequence and now noone).
> 
> 
> Any hints for now?

It might be a timing issue, that we need to extend the polling loop after that 
the first key has been pressed, so that the key release is also recorded.

--HPS