Re: pm-hibernate : possible circular locking dependency detected

[Rusty Russell <rusty@rustcorp.com.au>]

On Wednesday 08 April 2009 22:18:26 Peter Zijlstra wrote:
> On Wed, 2009-04-08 at 12:47 +0930, Rusty Russell wrote:
> > On Tuesday 07 April 2009 16:35:53 Peter Zijlstra wrote:
> > > On Tue, 2009-04-07 at 13:56 +0930, Rusty Russell wrote:
> > > > Looks like this will be fixed by Andrew's work-on-cpu-in-own-thread
> > > > patch which I just put out the pull request for.
> > > 
> > > Would it make sense to teach it about a short-circuit like:
> > > 
> > > work_on_cpu() {
> > > 
> > >   if (cpumask_weight(current->cpus_allowed) == 1 && 
> > >       smp_processor_id() == cpu)
> > >     return do_work_right_here();
> > 
> > Does that happen much?  I guess put a counter in and see?
> 
> Ego spotted the case where cpufreq calls it from an cpu-affine
> workqueue, it seems to me in that case its desirable to have the
> short-cut, and currently that's needed for correctness too as it will
> generate this circular lock thingy.

Well, the correctness issue is fixed by Andrew's work_on_cpu-via-new-thread
patch (hmm, which Linus hasn't taken, re-xmitting).

So it's really a pure optimization.

Cheers,
Rusty.