Re: Kernel broken on processors without performance counters

[Peter Zijlstra <peterz@infradead.org>]

On Tue, Jul 21, 2015 at 05:49:59PM +0200, Peter Zijlstra wrote:
> On Tue, Jul 21, 2015 at 05:43:27PM +0200, Thomas Gleixner wrote:
> > On Tue, 21 Jul 2015, Peter Zijlstra wrote:
> > 
> > > > -#endif	/* _LINUX_JUMP_LABEL_H */
> > > > +static inline void static_key_enable(struct static_key *key)
> > > > +{
> > > > +	int count = static_key_count(key);
> > > > +
> > > > +	WARN_ON_ONCE(count < 0 || count > 1);
> > > > +
> > > > +	if (!count)
> > > > +		static_key_slow_inc(key);
> > > > +}
> > > > +
> > > > +static inline void static_key_disable(struct static_key *key)
> > > > +{
> > > > +	int count = static_key_count(key);
> > > > +
> > > > +	WARN_ON_ONCE(count < 0 || count > 1);
> > > > +
> > > > +	if (count)
> > > > +		static_key_slow_dec(key);
> > > > +}
> > 
> > The functions above are not part of the interface which should be used
> > in code, right?
> 
> They are in fact. They make it easier to deal with the refcount thing
> when all you want is boolean states.

That is, things like sched_feat_keys[] which is an array of static keys,
the split types doesn't work -- an array can after all have only one
type.

In such cases you do have to be very careful to not wreck things.