Re: More 0.3.5-pre woes

[Dan Williams <dcbw redhat com>]

On Thu, 2005-05-19 at 12:52 -0400, Derek Atkins wrote:
> Quoting Dan Williams <dcbw redhat com>:
> 
> > If you've built from CVS, can you run the test/nmtest tool?
> > 
> > Dan
> 
> Clearly something was wedged..  I just restarted NM and now it seems to be
> better.  But the question remains, why was it wedged?  Unfortunately it's a bit
> challenging to "go back" now so I'll try to do more debugging if it happens
> again.

Well, there's a bunch of locking going on, for example we have to lock
the list of wireless devices whenever we do operations that may change
that list.  0.4x has move much more towards reference counting which
solves a big chunk of this.

There is one main thread, and one thread per network device.  So you
have the global device list lock, and per device thread you have the
device scanning lock, and also a "best ap" lock.  The scanning lock
lives on in 0.4x, but the best ap lock is gone because it was kind of a
hack.

Any one of these three locks could be locking against itself,
unfortunately I need a backtrace to see where that happens.

Dan