Re: delayed panic loading atapicam (after failed burncd)

On Wed, Oct 22, 2008 at 9:14 AM, Ben Kaduk <minimarmot_at_gmail.com> wrote:
> On Tue, Oct 21, 2008 at 9:04 AM, Ben Kaduk <minimarmot_at_gmail.com> wrote:
>> On Tue, Oct 21, 2008 at 3:17 AM, Thomas Quinot <thomas_at_freebsd.org> wrote:
>>> * Ben Kaduk, 2008-10-21 :
>>
>>> To avoid copying everything by hand, you can hook a serial console to
>>> your machine and log the messages on a second machine.
>>>
>>>> I don't have the media that was causing the panic here at home; I'll
>>>> try to test that tomorrow and see if the panic
>>>> is gone with the new snapshot.
>>>
>
> Interesting.  I just burned (I think successfully) a 200809-bootonly
> iso using burncd (on a 200810 snapshot system), but there were some
> errors on the console right about when it was trying to fixate the
> disc.
> The system was stable after that, though, so I figured I'd see what
> kldload-ing atapicam did.
> This gives a different panic*, that I thought I'd pass on before crashing:
> (*) fatal trap, now that I've scrolled back
>
> Fatal trap 9: general protection fault while in kernel mode
> instruction pointer 0x8:0xffffffff8019cd20
> [...]
> stopped at daclose+0xa0: movq 0x20(%r9),%rdi
> db> lock order reversal: (Giant after non-sleepable)
> 1st ATAPICAM lock _at_ /usr/src/sys/cam/cam_periph.h:182
> 2nd Giant _at_ /usr/src/sys/dev/kbdmux/kbdmux.c:1044
> KDB: stack backtrace:
> db_trace_self_wrapper
> _witness_debugger
> witness_checkorder
> _mtx_lock_flags
> kbdmux_ioctl
> sc_cngetc
> cncheckc
> cngetc
> db_readline
> dbread_line
> db_command_loop
> db_trap
> kdb_trap
> trap_fatal
> trap
> calltrap
> ----- trap [...]
> daclose+0xa0
> g_disk_access+0x196
> g_access+0x198
> g_part_taste+0x158
> g_new_provider_event_0xa5
> g_run_events+0x217
> g_event_procbody+0x6c
> fork_exit+0x12a
> fork_trampoline+0xe
>
> Hm, I bet the stuff above the trap is noise from breaking to KDB.
>
> Looking at the posted dmesg, it doesn't appear to be
> liking the uart very much, so the fact that I don't have
> a proper cable here is not quite as bad.  Maybe I
> can hack something with dcons elsewhere ...
>
> More updates tomorrow; it seems that I really need
> to crash now, though.
>
> -Ben Kaduk
>

For the archives:

I went to install some ports, and portsnap repeatedly complained
about invalid checksums, which was weird.  I then used csup
to grab a ports tree, but attempting to build something (I think it
was gnu screen(1)) caused mtree to die with something about
"unable to find group 'wheel'".  This causes me to believe that my
installation was corrupt, and I am tentatively blaming my install
media.

-Ben Kaduk