Re: [PATCH] spin_unlock*_no_resched()

[Kirill Tkhai]

On Wed, 2013-06-12 at 09:07 -0400, Steven Rostedt wrote:
> On Wed, 2013-06-12 at 14:15 +0200, Peter Zijlstra wrote:
> 
> > So I absolutely hate this API because people can (and invariably will)
> > abuse it; much like they did/do preempt_enable_no_resched().
> 
> Me too.
> 
> > 
> > IIRC Thomas even maps preempt_enable_no_resched() to preempt_enable() in
> > -rt to make sure we don't miss preemption points due to stupidity.
> > 
> > He converted the 'few' sane sites to use schedule_preempt_disabled(). In
> > that vein, does it make sense to introduce schedule_spin_locked()?
> > 
> 
> I was thinking the exact same thing when I read this patch. This is a
> strict policy that we should enforce and not let individual developers
> implement. Yes, a schedule_spin_unlock() would work nicely. The API will
> enforce the two to be used together.

Steven thanks for your explanation and Peter's, now I looked to this
from another side.

If we speak about combined primitive does it have to be a special
variant of schedule_spin_unlock_* for every irq state? The simplest way
is to do local_irq_enable() always before schedule() call, but I'm not
sure that this is good for all platforms.

For -rt everything of this is completely useless, because number of
raw_spin_locks is small. Maybe changes for some another types of locks
will applicable.

Kirill

> Otherwise, I can envision seeing
> things like:
> 
> 	preempt_disable();
> 	[...]
> 
> 	spin_lock(x);
> 
> 	spin_unlock_no_resched(x);
> 
> 	[...]
> 
> 	preempt_enable();
> 
> And developers having no idea why the above is broken. Although, I would
> say the above is broken for other reasons, but I was just using that to
> show the craziness such an API would give to us.
> 
> -- Steve
> 
> 
> 



--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/