Re: [patch] sched: fix set_task_cpu() and provide an unlocked runqueue variant

[Mike Galbraith <efault@gmx.de>]

On Thu, 2009-11-26 at 15:09 +0100, Peter Zijlstra wrote:
> On Thu, 2009-11-26 at 11:16 +0100, Mike Galbraith wrote:
> > > min_vruntime should only ever be poked at when holding the respective
> > > rq->lock, even with a barrier a 64bit read on a 32bit machine can go all
> > > funny.
> > 
> > Yeah, but we're looking at an unlocked runqueue.  But never mind...
> 
> The patch is also poking at rq->clock without rq->lock held... not very
> nice.

To the users on the remote runqueue, it doesn't matter, they always
update before using, so will always have accurate time.

What bothers me is that I don't see what prevents a SCHED_IDLE task
making a huge min_vruntime update on the remote CPU while you're
queueing a sleeper here, who used to live over there, while you're
reading.

But, they're all instantaneous values anyway, so I suppose memory speed
is fine.  It works... and I really doubt that the yet another paranoid
barrier I just put in will make diddly spit difference, but I did just
put an smp_rmb() there anyway to see with my own eyes ;-)

> Gah, I hate that we're doing migration things without holding both rq's,
> this is making live so very interesting ;-)

Yes.

	-Mike