Re: lock order reversals - what do they mean?

On Sun, 14 Dec 2003, C. Kukulies wrote:

> I'm getting 
> lock order reversal
>  1st 0xc09524c0 UMA lock (UMA lock) _at_ /u/src/sys/vm/uma_core.c:1200
>  2nd 0xc1431100 system map (system map) _at_ /u/src/sys/vm/vm_map.c:2210
> Stack backtrace:
> 
> This happenend while doing a find . something on my notebook hd. 

These warnings are generated by Witness, a run-time lock diagnostic system
found in FreeBSD 5-CURRENT kernels (but removed in releases).  You can
read more about Witness in the WITNESS(4) man page, which talks about its
capabilities.  Among other things, Witness performs run-time lock order
verification using a combination of hard coded lock orders, and run-time
detected lock orders, and generates console warnings when lock orders are
violated.  The intent of this is to detect the potential for deadlocks due
to lock order violations; it's worth observing that Witness is actually
slightly conservative, and so it's possible to get false positives.  In
the event that Witness is accurately reporting a lock order problem, it's
basically saying "If you were unlucky, a deadlock would have happened
here".  There are a couple of "well known" false positives, which we need
to do a better job of documenting to prevent spurious reports.  The
non-well-known ones typically correspond to bugs in newly added locking,
as lock order reversals usually get fixed pretty quickly because Witness
is busy generating warnings :-).  I believe the reversal you've reported
is a false positive.

Thanks,

Robert N M Watson             FreeBSD Core Team, TrustedBSD Projects
robert_at_fledge.watson.org      Senior Research Scientist, McAfee Research