Re: [PATCH v5 03/15] sched/core: uclamp: map TASK's clamp values into CPU's clamp groups

[Peter Zijlstra]

On Wed, Nov 07, 2018 at 01:57:38PM +0000, Patrick Bellasi wrote:
> On 07-Nov 14:16, Peter Zijlstra wrote:

> > Please write cmpxchg loops in the form:
> > 
> > 	atomic_long_t *ptr = &uclamp_maps[clamp_id][group_id].adata;
> > 	union uclamp_map old, new;
> > 
> > 	old.data = atomic_long_read(ptr);
> > 	do {
> > 		new.data = old.data;
> > 		new.se_cound--;
> > 	} while (!atomic_long_try_cmpxchg(ptr, &old.data, new.data));
> > 
> > 
> > (same for all the others of course)
> 
> Ok, I did that to save some indentation, but actually it's most
> commonly used in a while loop... will update in v6.
> 
> Out of curiosity, apart from code consistency, is that required also
> specifically for any possible compiler related (mis)behavior ?

No; it is just the 'normal' form my brain likes :-)

And the try_cmpxchg() thing is slightly more efficient on x86 vs the
traditional form:

	while (cmpxchg(ptr, old, new) != old)