Re: [RFC -v3 PATCH 2/3] sched: add yield_to function

From: Hillf Danton
Date: Tue Jan 04 2011 - 09:28:22 EST


On Tue, Jan 4, 2011 at 5:29 AM, Rik van Riel <riel@xxxxxxxxxx> wrote:
> From: Mike Galbraith <efault@xxxxxx>
>
> Add a yield_to function to the scheduler code, allowing us to
> give enough of our timeslice to another thread to allow it to
> run and release whatever resource we need it to release.
>
> We may want to use this to provide a sys_yield_to system call
> one day.
>
> Signed-off-by: Rik van Riel <riel@xxxxxxxxxx>
> Signed-off-by: Marcelo Tosatti <mtosatti@xxxxxxxxxx>
> Not-signed-off-by: Mike Galbraith <efault@xxxxxx>
>
> ---
> Mike, want to change the above into a Signed-off-by: ? :)
> This code seems to work well.
>
> diff --git a/include/linux/sched.h b/include/linux/sched.h
> index c5f926c..0b8a3e6 100644
> --- a/include/linux/sched.h
> +++ b/include/linux/sched.h
> @@ -1083,6 +1083,7 @@ struct sched_class {
> Â Â Â Âvoid (*enqueue_task) (struct rq *rq, struct task_struct *p, int wakeup);
> Â Â Â Âvoid (*dequeue_task) (struct rq *rq, struct task_struct *p, int sleep);
> Â Â Â Âvoid (*yield_task) (struct rq *rq);
> + Â Â Â int (*yield_to_task) (struct task_struct *p, int preempt);
>
> Â Â Â Âvoid (*check_preempt_curr) (struct rq *rq, struct task_struct *p, int flags);
>
> @@ -1981,6 +1982,7 @@ static inline int rt_mutex_getprio(struct task_struct *p)
> Â# define rt_mutex_adjust_pi(p) Â Â Â Â do { } while (0)
> Â#endif
>
> +extern void yield_to(struct task_struct *p, int preempt);
> Âextern void set_user_nice(struct task_struct *p, long nice);
> Âextern int task_prio(const struct task_struct *p);
> Âextern int task_nice(const struct task_struct *p);
> diff --git a/kernel/sched.c b/kernel/sched.c
> index f8e5a25..ffa7a9d 100644
> --- a/kernel/sched.c
> +++ b/kernel/sched.c
> @@ -6901,6 +6901,53 @@ void __sched yield(void)
> Â}
> ÂEXPORT_SYMBOL(yield);
>
> +/**
> + * yield_to - yield the current processor to another thread in
> + * your thread group, or accelerate that thread toward the
> + * processor it's on.
> + *
> + * It's the caller's job to ensure that the target task struct
> + * can't go away on us before we can do any checks.
> + */
> +void __sched yield_to(struct task_struct *p, int preempt)
> +{
> + Â Â Â struct task_struct *curr = current;
> + Â Â Â struct rq *rq, *p_rq;
> + Â Â Â unsigned long flags;
> + Â Â Â int yield = 0;
> +
> + Â Â Â local_irq_save(flags);
> + Â Â Â rq = this_rq();
> +
> +again:
> + Â Â Â p_rq = task_rq(p);
> + Â Â Â double_rq_lock(rq, p_rq);
> + Â Â Â while (task_rq(p) != p_rq) {
> + Â Â Â Â Â Â Â double_rq_unlock(rq, p_rq);
> + Â Â Â Â Â Â Â goto again;
> + Â Â Â }
> +
> + Â Â Â if (task_running(p_rq, p) || p->state || !p->se.on_rq ||
> + Â Â Â Â Â Â Â Â Â Â Â !same_thread_group(p, curr) ||
> + Â Â Â Â Â Â Â Â Â Â Â !curr->sched_class->yield_to_task ||
> + Â Â Â Â Â Â Â Â Â Â Â curr->sched_class != p->sched_class) {
> + Â Â Â Â Â Â Â goto out;
> + Â Â Â }
> +
> + Â Â Â yield = curr->sched_class->yield_to_task(p, preempt);
> +
> +out:
> + Â Â Â double_rq_unlock(rq, p_rq);
> + Â Â Â local_irq_restore(flags);
> +
> + Â Â Â if (yield) {
> + Â Â Â Â Â Â Â set_current_state(TASK_RUNNING);
> + Â Â Â Â Â Â Â schedule();
> + Â Â Â }
> +}
> +EXPORT_SYMBOL(yield_to);
> +
> +
> Â/*
> Â* This task is about to go to sleep on IO. Increment rq->nr_iowait so
> Â* that process accounting knows that this is a task in IO wait state.
> diff --git a/kernel/sched_fair.c b/kernel/sched_fair.c
> index 5119b08..3288e7c 100644
> --- a/kernel/sched_fair.c
> +++ b/kernel/sched_fair.c
> @@ -1119,6 +1119,61 @@ static void yield_task_fair(struct rq *rq)
> Â}
>
> Â#ifdef CONFIG_SMP
> +static void pull_task(struct rq *src_rq, struct task_struct *p,
> + Â Â Â Â Â Â Â Â Â Â struct rq *this_rq, int this_cpu);
> +#endif
> +
> +static int yield_to_task_fair(struct task_struct *p, int preempt)
> +{
> + Â Â Â struct sched_entity *se = &current->se;
> + Â Â Â struct sched_entity *pse = &p->se;
> + Â Â Â struct sched_entity *curr = &(task_rq(p)->curr)->se;
> + Â Â Â struct cfs_rq *cfs_rq = cfs_rq_of(se);
> + Â Â Â struct cfs_rq *p_cfs_rq = cfs_rq_of(pse);
> + Â Â Â int yield = this_rq() == task_rq(p);
> + Â Â Â int want_preempt = preempt;
> +
> +#ifdef CONFIG_FAIR_GROUP_SCHED
> + Â Â Â if (cfs_rq->tg != p_cfs_rq->tg)
> + Â Â Â Â Â Â Â return 0;
> +
> + Â Â Â /* Preemption only allowed within the same task group. */
> + Â Â Â if (preempt && cfs_rq->tg != cfs_rq_of(curr)->tg)
> + Â Â Â Â Â Â Â preempt = 0;
> +#endif
> + Â Â Â /* Preemption only allowed within the same thread group. */
> + Â Â Â if (preempt && !same_thread_group(current, task_of(p_cfs_rq->curr)))
> + Â Â Â Â Â Â Â preempt = 0;
> +
> +#ifdef CONFIG_SMP
> + Â Â Â /*
> + Â Â Â Â* If this yield is important enough to want to preempt instead
> + Â Â Â Â* of only dropping a ->next hint, we're alone, and the target
> + Â Â Â Â* is not alone, pull the target to this cpu.
> + Â Â Â Â*/
> + Â Â Â if (want_preempt && !yield && cfs_rq->nr_running == 1 &&
> + Â Â Â Â Â Â Â Â Â Â Â cpumask_test_cpu(smp_processor_id(), &p->cpus_allowed)) {
> + Â Â Â Â Â Â Â pull_task(task_rq(p), p, this_rq(), smp_processor_id());
> + Â Â Â Â Â Â Â p_cfs_rq = cfs_rq_of(pse);
> + Â Â Â Â Â Â Â yield = 1;
> + Â Â Â }
> +#endif
> +
> + Â Â Â if (yield)
> + Â Â Â Â Â Â Â clear_buddies(cfs_rq, se);
> + Â Â Â else if (preempt)
> + Â Â Â Â Â Â Â clear_buddies(p_cfs_rq, curr);
> +
> + Â Â Â /* Tell the scheduler that we'd really like pse to run next. */
> + Â Â Â p_cfs_rq->next = pse;

If not pulled and this_rq() != task_rq(p), only assigning ->next could
kick p onto its CPU?

If not, how is the lock contention eased then?

A few words to explain please.

thanks
Hillf

> +
> + Â Â Â if (!yield && preempt)
> + Â Â Â Â Â Â Â resched_task(task_of(p_cfs_rq->curr));
> +
> + Â Â Â return yield;
> +}
> +
> +#ifdef CONFIG_SMP
>
> Â#ifdef CONFIG_FAIR_GROUP_SCHED
> Â/*
> @@ -2081,6 +2136,7 @@ static const struct sched_class fair_sched_class = {
>    Â.enqueue_task      = enqueue_task_fair,
>    Â.dequeue_task      = dequeue_task_fair,
>    Â.yield_task       = yield_task_fair,
> +    .yield_to_task     Â= yield_to_task_fair,
>
>    Â.check_preempt_curr   = check_preempt_wakeup,
>
>
> --
> To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
> the body of a message to majordomo@xxxxxxxxxxxxxxx
> More majordomo info at Âhttp://vger.kernel.org/majordomo-info.html
> Please read the FAQ at Âhttp://www.tux.org/lkml/
>
>
>
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/