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

[Peter Zijlstra <a.p.zijlstra@chello.nl>]

On Mon, 2011-01-03 at 16:29 -0500, Rik van Riel wrote:
> From: Mike Galbraith <efault@gmx.de>
> 
> 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@redhat.com>
> Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
> Not-signed-off-by: Mike Galbraith <efault@gmx.de>
> 
> --- 
> 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 {
>  	void (*enqueue_task) (struct rq *rq, struct task_struct *p, int wakeup);
>  	void (*dequeue_task) (struct rq *rq, struct task_struct *p, int sleep);
>  	void (*yield_task) (struct rq *rq);
> +	int (*yield_to_task) (struct task_struct *p, int preempt);
>  
>  	void (*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)
>  # define rt_mutex_adjust_pi(p)		do { } while (0)
>  #endif
>  
> +extern void yield_to(struct task_struct *p, int preempt);
>  extern void set_user_nice(struct task_struct *p, long nice);
>  extern int task_prio(const struct task_struct *p);
>  extern 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)
>  }
>  EXPORT_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)
> +{
> +	struct task_struct *curr = current;
> +	struct rq *rq, *p_rq;
> +	unsigned long flags;
> +	int yield = 0;
> +
> +	local_irq_save(flags);
> +	rq = this_rq();
> +
> +again:
> +	p_rq = task_rq(p);
> +	double_rq_lock(rq, p_rq);
> +	while (task_rq(p) != p_rq) {
> +		double_rq_unlock(rq, p_rq);
> +		goto again;
> +	}
> +
> +	if (task_running(p_rq, p) || p->state || !p->se.on_rq ||
> +			!same_thread_group(p, curr) ||
> +			!curr->sched_class->yield_to_task ||
> +			curr->sched_class != p->sched_class) {
> +		goto out;
> +	}
> +
> +	yield = curr->sched_class->yield_to_task(p, preempt);
> +
> +out:
> +	double_rq_unlock(rq, p_rq);
> +	local_irq_restore(flags);
> +
> +	if (yield) {
> +		set_current_state(TASK_RUNNING);
> +		schedule();
> +	}
> +}
> +EXPORT_SYMBOL(yield_to);

This definitely wants to be EXPORT_SYMBOL_GPL() and if it were possible
I'd make it so only kvm.o could use it. It really sucks that kvm is a
module.

>  /*
>   * This task is about to go to sleep on IO. Increment rq->nr_iowait so
>   * 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)
>  }
>  
>  #ifdef CONFIG_SMP
> +static void pull_task(struct rq *src_rq, struct task_struct *p,
> +		      struct rq *this_rq, int this_cpu);
> +#endif
> +
> +static int yield_to_task_fair(struct task_struct *p, int preempt)
> +{
> +	struct sched_entity *se = &current->se;
> +	struct sched_entity *pse = &p->se;
> +	struct sched_entity *curr = &(task_rq(p)->curr)->se;
> +	struct cfs_rq *cfs_rq = cfs_rq_of(se);
> +	struct cfs_rq *p_cfs_rq = cfs_rq_of(pse);
> +	int yield = this_rq() == task_rq(p);
> +	int want_preempt = preempt;
> +
> +#ifdef CONFIG_FAIR_GROUP_SCHED
> +	if (cfs_rq->tg != p_cfs_rq->tg)
> +		return 0;
> +
> +	/* Preemption only allowed within the same task group. */
> +	if (preempt && cfs_rq->tg != cfs_rq_of(curr)->tg)
> +		preempt = 0;
> +#endif

I'd simply bail if its not the same cgroup, who cares about that case
anyway, all KVM vcpu threads should be in the same cgroup I think.

> +	/* Preemption only allowed within the same thread group. */
> +	if (preempt && !same_thread_group(current, task_of(p_cfs_rq->curr)))
> +		preempt = 0;

The calling site already checks for same_thread_group(), we never even
get here if that's not the case.

> +#ifdef CONFIG_SMP
> +	/*
> +	 * If this yield is important enough to want to preempt instead
> +	 * of only dropping a ->next hint, we're alone, and the target
> +	 * is not alone, pull the target to this cpu.
> +	 */
> +	if (want_preempt && !yield && cfs_rq->nr_running == 1 &&
> +			cpumask_test_cpu(smp_processor_id(), &p->cpus_allowed)) {
> +		pull_task(task_rq(p), p, this_rq(), smp_processor_id());

This only works by the grace that the caller checked p->se.on_rq. A
comment might be in order.

> +		p_cfs_rq = cfs_rq_of(pse);
> +		yield = 1;
> +	}
> +#endif
> +
> +	if (yield)
> +		clear_buddies(cfs_rq, se);
> +	else if (preempt)
> +		clear_buddies(p_cfs_rq, curr);
> +
> +	/* Tell the scheduler that we'd really like pse to run next. */
> +	p_cfs_rq->next = pse;
> +
> +	if (!yield && preempt)
> +		resched_task(task_of(p_cfs_rq->curr));

I don't get this.. Why would you resched the remote cpu, surely you
didn't just pull its current task over..

> +	return yield;
> +}
> +
> +#ifdef CONFIG_SMP
>  
>  #ifdef CONFIG_FAIR_GROUP_SCHED
>  /*
> @@ -2081,6 +2136,7 @@ static const struct sched_class fair_sched_class = {
>  	.enqueue_task		= enqueue_task_fair,
>  	.dequeue_task		= dequeue_task_fair,
>  	.yield_task		= yield_task_fair,
> +	.yield_to_task		= yield_to_task_fair,
>  
>  	.check_preempt_curr	= check_preempt_wakeup,
>  
>