Re: geli boot issues with recent kernels

Fabian Keil <freebsd-listen_at_fabiankeil.de> wrote:

> My kernel and / are on ada0s1a (UFS) while the rest of the system
> is on a ZFS pool on ada0s1d.eli, attached on boot.
> 
> Since a few days ago, booting no longer works reliably for me.
> Every now and then the system hangs at the time when I'd normaly
> provide the passphrase.
> 
> The last times this happened I was in a hurry, so I'm not entirely
> sure if the prompt for the passphrase is even shown (on my system
> it's always buried in USB messages anyway).

The geli passphrase prompt isn't shown at all:
http://www.fabiankeil.de/tmp/freebsd-9.0-geli-boot-problem-800x561.jpg

Verbose dmesg for a successful boot:
http://www.fabiankeil.de/tmp/freebsd-dmesg-2010-09-27.txt

> Sometimes keyboard input is ignored completely, sometimes I can
> still use shift lock or scroll through the kernel messages.

I now seem to always be able to scroll through the
kernel messages so maybe I misremembered.

> When this happened the first time, rebooting the system solved
> the problem. By "rebooting" I mean shutting down the system by
> keeping the power button pressed and then powering the system
> on again. When the system hangs, it doesn't seem to react to
> the ACPI event that is generated by pressing the power button
> shortly.

It seems to react to the ACPI event if I press the power button
more than once, the first time shortly enough so the system doesn't
power down.

Attached USB devices are also noticed, the same goes for
attaching or detaching the power cord.

> Today I rebooted the system three times in a row without getting
> it to work so I finally unloaded geom_eli from the loader prompt,
> booted to single-user mode and attached the geli provider from
> there.

Actually I unloaded all modules and this detail seems to matter.

> When I booted a few hours later with the same kernel it
> worked right away.
> 
> I think I experienced the problem the first time with
> a kernel from last Wednesday, currently I'm using:
> FreeBSD 9.0-CURRENT Fri Sep 17 18:25:39 CEST 2010 amd64
> 
> It definitively didn't happen with a kernel from last Saturday,
> but then again I seldomly boot the system more than once a day
> and the problem doesn't always show.
> 
> I enabled boot_verbose="-v" and kern.geom.eli.visible_passphrase=1
> and thus expect to be able to provide a more detailed problem
> description the next time this happens.

I'm still experiencing the problem with yesterday's CURRENT.

By default I load the following modules through loader.conf:

zlib_load="YES"
crypto_load="YES"               
geom_eli_load="YES"             
opensolaris_load="YES"          
zfs_load="YES"                  
acpi_video_load="YES"           
acpi_ibm_load="YES"             
snd_hda_load="YES"              
snd_uaudio_load="YES"           
if_iwn_load="YES"               
iwnfw_load="YES"

I noticed that if the system (ThinkPad R500) is running from battery
(with those modules enabled) the hangs are almost guaranteed to happen,
with external power the hangs are a lot less likely, but can happen as
well.

If I remove either snd_hda or snd_uaudio it get a lot less likely
that the problem occurs, even when running from battery.

Fabian