Re: workqueue code needing preemption disabled

[Tejun Heo <tj@kernel.org>]

On Mon, Mar 18, 2013 at 02:23:56PM -0400, Steven Rostedt wrote:
> On Mon, 2013-03-18 at 09:43 -0700, Tejun Heo wrote:
> > Hello, Steven.
> > 
> > On Mon, Mar 18, 2013 at 12:30:43PM -0400, Steven Rostedt wrote:
> > > If you happen to know the critical areas that require preemption to be
> > > disabled for real, we can encapsulate them with:
> > > 
> > > 	preempt_disable_rt();
> > > 
> > > 	preempt_enable_rt();
> > > 
> > > These are currently only in the -rt patch, but it annotates locations
> > > that require preemption to be disabled even when -rt converts spin_locks
> > > into mutexes. These obviously can not contain spin_locks() as
> > > spin_locks() can block and schedule out.
> > 
> > Making gcwq locks disable preemption would be much safer / easier, but
> > if that's not desirable, anything touching gcwq->idle_list would be a
> > good place to start - worker_enter_idle() and worker_leave_idle().
> > Hmmm... ignoring CPU hotplug, I think those two might just do it.
> > Give it a try?  How reproducible is the problem?
> > 
> 
> Hmm, the issue is that a "use to be" idle thread got migrated, and is
> now being woken up by another worker. What can cause an established
> worker to migrate without HOTPLUG being active?

It doesn't.  I think it's trying to wakeup the idle_list head.

-- 
tejun