Re: LOR acpi_ibm module

On Wed, Sep 9, 2009 at 1:25 PM, Jung-uk Kim<jkim_at_freebsd.org> wrote:
> On Wednesday 09 September 2009 01:49 pm, Felix Stolba wrote:
>> Brandon Gooch schrieb:
>> > lock order reversal:
>> >  1st 0xffffffff807cf200 sysctl lock (sysctl lock) _at_
>> > /usr/src/sys/kern/kern_sysctl.c:1608
>> >  2nd 0xffffffff80bf1de0 ACPI IBM extras (ACPI IBM extras) _at_
>> > /usr/src/sys/modules/acpi/acpi_ibm/../../../dev/acpi_support/acpi
>> >_ibm.c:481 KDB: stack backtrace:
>> > db_trace_self_wrapper() at db_trace_self_wrapper+0x2a
>> > _witness_debugger() at _witness_debugger+0x2e
>> > witness_checkorder() at witness_checkorder+0x81e
>> > _sx_xlock() at _sx_xlock+0x54
>> > acpi_ibm_sysctl() at acpi_ibm_sysctl+0x4f
>> > sysctl_root() at sysctl_root+0xe3
>> > userland_sysctl() at userland_sysctl+0x158
>> > __sysctl() at __sysctl+0xaa
>> > syscall() at syscall+0x1dd
>> > Xfast_syscall() at Xfast_syscall+0xd0
>> > --- syscall (202, FreeBSD ELF64, __sysctl), rip = 0x80073769c,
>> > rsp = 0x7fffffffda58, rbp = 0x4 ---
>>
>> I'm getting the same LOR at boot in 9.0-current (source from 7th of
>> september).
>
> It is generally harmless but really annoying.
>
> Jung-uk Kim
>

I haven't been running a kernel with WITNESS (or any debugging) for a
couple of weeks, so I forgot about it.

Originally, I wondered if this locking problem(?) was causing a
strange issue when I invoked a script via devd (a call to a script to
handle ACPI events through acpi_ibm.ko), in this case, an ACPI screen
brightness function: Occasionally, the screen brightness would not
adjust when I used the Fn keys, only to eventually, "pop" all of the
requests I sent off of the devd command table, usually with a long
delay.

If I started devd in foreground mode and watched the command output in
the terminal, commands were "pushed" and "popped" immediately; all was
well.

I was never able to put it all together, and it was only a minor
annoyance, after all.

-Brandon