Re: [PATCH 3/4] workqueue: Allow modifying low level unbound workqueue cpumask

[Frederic Weisbecker <fweisbec@gmail.com>]

On Thu, Apr 24, 2014 at 11:30:48AM -0400, Tejun Heo wrote:
> On Thu, Apr 24, 2014 at 04:37:35PM +0200, Frederic Weisbecker wrote:
> > +static int apply_workqueue_attrs_locked(struct workqueue_struct *wq,
> > +					const struct workqueue_attrs *attrs);
> 
> Can't we reorder things so that we don't need the above prototype?

Yeah I'll give it a try.

> 
> > +/* Must be called with wq_unbound_mutex held */
> 
> Please use lockdep_assert_held() instead.

Ok.

> 
> > +static int unbounds_cpumask_apply_all(cpumask_var_t cpumask)
> > +{
> > +	struct workqueue_struct *wq;
> > +
> > +	list_for_each_entry(wq, &workqueues, list) {
> > +		struct workqueue_attrs *attrs;
> > +
> > +		if (!(wq->flags & WQ_UNBOUND))
> > +			continue;
> > +		/* Ordered workqueues need specific treatment */
> > +		if (wq->flags & __WQ_ORDERED)
> > +			continue;
> > +
> > +		attrs = wq_sysfs_prep_attrs(wq);
> > +		if (!attrs)
> > +			return -ENOMEM;
> 
> So, we're failing in the middle without rolling back?

Yeah, early patch :)

> 
> > +
> > +		WARN_ON_ONCE(apply_workqueue_attrs_locked(wq, attrs));
> 
> Are we triggering WARN on -ENOMEM too and then ignore the failure?

Yeah some more thought is needed on error handling.

> 
> > +		free_workqueue_attrs(attrs);
> > +	}
> > +
> > +	return 0;
> > +}
> 
> Shouldn't we separate allocation stage from switching stage so that we
> can either succeed or fail?  The above is very mushy about error
> handling.

They are already pretty seperate above. But yeah I need to rework the error
handling.

Thanks.