Re: [RFC][PATCH tip/sched/core] sched/rt: Simplify the IPI rt balancing logic

[Peter Zijlstra]

On Fri, Apr 21, 2017 at 10:49:29PM -0400, Steven Rostedt wrote:
> +#ifdef HAVE_RT_PUSH_IPI
> +	/*
> +	 * For IPI pull requests, loop across the rto_mask.
> +	 */
> +	struct irq_work rto_push_work;
> +	raw_spinlock_t rto_lock;
> +	/* These atomics are updated outside of a lock */
> +	atomic_t rto_loop_next;
> +	atomic_t rto_loop_start;
> +	/* These are only updated and read withn rto_lock */
> +	int rto_loop;
> +	int rto_cpu;
> +#endif

Don't you think it would make sense to place the rto_lock near the
variables it protects? And if those atomics are supposed to increase
performance, do they want to share the same cacheline with the lock?