Re: [PATCH] Repost sched-rt: Reduce rq lock contention by eliminating locking of non-feasible target

[Steven Rostedt <rostedt@goodmis.org>]

On Tue, 06 Jan 2015 11:01:51 -0800
Tim Chen <tim.c.chen@linux.intel.com> wrote:

> Didn't get any response for this patch probably due to the holidays.
> Reposting it as we will like to get it merged to help our database
> workload.
> 
> This patch added checks that prevent futile attempts to move rt tasks
> to cpu with active tasks of equal or higher priority.  This reduces
> run queue lock contention and improves the performance of a well
> known OLTP benchmark by 0.7%.
> 
> Signed-off-by: Tim Chen <tim.c.chen@linux.intel.com>
> ---
>  kernel/sched/rt.c | 9 ++++++++-
>  1 file changed, 8 insertions(+), 1 deletion(-)
> 
> diff --git a/kernel/sched/rt.c b/kernel/sched/rt.c
> index ee15f5a..0e4382e 100644
> --- a/kernel/sched/rt.c
> +++ b/kernel/sched/rt.c
> @@ -1337,7 +1337,8 @@ select_task_rq_rt(struct task_struct *p, int
> cpu, int sd_flag, int flags) curr->prio <= p->prio)) {
>  		int target = find_lowest_rq(p);
> 

Please add a comment here that says something like:

		/*
		 * Don't bother moving it if the destination CPU is
		 * not running a lower priority task.
		 */

> -		if (target != -1)
> +		if (target != -1 &&
> +		    p->prio < cpu_rq(target)->rt.highest_prio.curr)
>  			cpu = target;
>  	}
>  	rcu_read_unlock();
> @@ -1613,6 +1614,12 @@ static struct rq *find_lock_lowest_rq(struct
> task_struct *task, struct rq *rq) break;
>  
>  		lowest_rq = cpu_rq(cpu);
> +		
> +		if (lowest_rq->rt.highest_prio.curr <= task->prio) {
> +		/* target rq has tasks of equal or higher priority,
> try again */
> +			lowest_rq = NULL;
> +			continue;

This should just break out and not try again. The reason for the other
try again is because of the double_lock which can release the locks
which can cause a process waiting for the lock to sneak in and
change the priorities. But this case, a try again is highly unlikely to
do anything differently (no locks are released) and just waste cycles.

-- Steve


> +		}
>  
>  		/* if the prio of this runqueue changed, try again */
>  		if (double_lock_balance(rq, lowest_rq)) {