Re: [Xen-devel] [PATCH 2/4] xen: sched: optimize exclusive pinning case (Credit1 & 2)

[Anshul Makkar <anshulmakkar@xxxxxxxxx>]

On 9/15/17 6:35 PM, Dario Faggioli wrote:
>   
>   static unsigned int __read_mostly opt_migrate_resist = 500;
>   integer_param("sched_credit2_migrate_resist", opt_migrate_resist);
> @@ -1453,6 +1459,26 @@ runq_tickle(const struct scheduler *ops, struct 
> csched2_vcpu *new, s_time_t now)
>                       (unsigned char *)&d);
>       }
>   
> +    /*
> +     * Exclusive pinning is when a vcpu has hard-affinity with only one
> +     * cpu, and there is no other vcpu that has hard-affinity with that
> +     * same cpu. This is infrequent, but if it happens, is for achieving
> +     * the most possible determinism, and least possible overhead for
> +     * the vcpus in question.
> +     *
> +     * Try to identify the vast majority of these situations, and deal
> +     * with them quickly.
Sorry, if I have missed to review the earlier series on the same subject.
But, I am not completely satisfied with the condition that pinning to a
pcpu is only possible if and only if no other vcpu has hard affinity of 
that pcpu.
I think we can do away with this condition and give pinning a priority.
Your thoughts please..
> +     */
> +    if ( unlikely((new->flags & CSFLAG_pinned) &&
> +                  cpumask_test_cpu(cpu, &rqd->idle) &&
> +                  !cpumask_test_cpu(cpu, &rqd->tickled)) )
> +    {
> +        ASSERT(cpumask_cycle(cpu, new->vcpu->cpu_hard_affinity) == cpu);
> +        SCHED_STAT_CRANK(tickled_idle_cpu_excl);
> +        ipid = cpu;
> +        goto tickle;
> +    }
> +
>       for_each_affinity_balance_step( bs )
>       {
>           /* Just skip first step, if we don't have a soft affinity */
> @@ -2826,6 +2852,19 @@ csched2_dom_cntl(
>       return rc;
>   }
>   
> +static void
> +csched2_aff_cntl(const struct scheduler *ops, struct vcpu *v,
> +                 const cpumask_t *hard, const cpumask_t *soft)
> +{
> +    struct csched2_vcpu *svc = csched2_vcpu(v);
> +
> +    /* Are we becoming exclusively pinned? */
> +    if ( cpumask_weight(hard) == 1 )
> +        __set_bit(__CSFLAG_pinned, &svc->flags);
> +    else
> +        __clear_bit(__CSFLAG_pinned, &svc->flags);
> +}
> +
>   static int csched2_sys_cntl(const struct scheduler *ops,
>                               struct xen_sysctl_scheduler_op *sc)
Looks fine.

Anshul


_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxx
https://lists.xen.org/xen-devel