On Thu, Jan 28, 2010 at 08:09:57AM +0800, Marek Lindner wrote:
>
> Hi,
>
> > nope, does not seem to work, still the same issue
> > I also tried it on my laptop here with the patch installed
> > and it is a very similar call trace, although here it is
> > send_vis_packets instead of vis_quit in the call trace.
>
> you should take the stack trace with a grain of salt. Stack tracing such an
> issue is a quite tricky thing, hence can not be fully trusted. Since the skb
> changes the code runs in interrupt context which probably introduced this
> bug.
> All previous warnings of this kind were related to holding a lock while
> sending packets. My patch unlocked the problematic orig_hash - maybe that was
> not enough ? By adding retrun statements at the beginning of the vis send
> function you might be able to get down to the problem.
It might also be worth running lockdep on the code. Normally you don^t
need the actually lockup, you just need to execute the code path that
would lockup under whatever conditions are required for it to lockup.
Andrew
