vt(4) and geli(8)

Warner Losh imp at bsdimp.com
Sun Apr 20 15:19:55 UTC 2014


On Apr 18, 2014, at 8:10 PM, Peter Wemm <peter at wemm.org> wrote:

> On Fri, Apr 18, 2014 at 6:18 PM, Chris Nehren
> <cnehren+freebsd-stable at pobox.com> wrote:
>> On 2014/04/18, 11:36, Greg Rivers wrote:
>>> 
>>> The new vt system console looks really nice.  But on hosts that
>>> have geli encrypted devices that prompt for the passphrase at
>>> boot, no keyboard input reaches geli.  The keyboard doesn't
>>> appear to be dead, as the caps-lock and num-lock keys toggle;
>>> geli just doesn't receive any input. This is on 10-STABLE amd64
>>> built from yesterday's sources. "hw.vga.textmode=1" doesn't make
>>> any difference.
>>> 
>>> Has anyone else encountered this?
>>> 
>> 
>> Yes, I have, on my FreeBSD workstation with a USB keyboard.  It's running
>> FreeBSD 10 with geli-encrypted ZFS root.  This happened before I updated to
>> vt, though.  I managed to work around the problem, surprisingly enough, by
>> mashing the keyboard when the system boots as soon as I see bold white
>> kernel text.  Then, when prompted for the passphrase (after all the USB
>> devices finish probing / attaching), I hit enter to clear the buffer and
>> then am able to enter the passphrase correctly.
>> 
>> I'll be filing a PR for this, probably on Monday, should I not be able to
>> find one that's similar.
>> 
>> --
>> Chris Nehren
> 
> For what it's worth, I don't think this is specific to vt.  You can
> sometimes get the same behavior when you hit the mountroot prompt.
> Some of the machines in the freebsd cluster will accept input at a
> mountroot prompt, the others we have to power cycle and fix
> vfs.root.mountfrom.  vt isn't involved.  I don't know what the
> variable is.
> 
> I once suspected it is some combination of ddb/kdb presence, and/or
> serial console / multiconsole mode but I don't think that holds up.

I fixed this bug on SERIAL consoles by making the grab/ungrab function
shift the serial port from interrupt driven processing to polled processing.
Maybe a similar change is now needed for atkbd?

Warner



More information about the freebsd-stable mailing list