[RFC][PATCH]: wait: don't resched in prepare_to_wait()

[RFC][PATCH]: wait: don't resched in prepare_to_wait()

[Peter Zijlstra]

Does the below make sense?

After prepare to wait, we either call schedule() or find !cond in which
case we'll call finish_wait() which contains another preemption check.

Not-signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
---
 kernel/wait.c |    4 ++++
 1 files changed, 4 insertions(+), 0 deletions(-)

diff --git a/kernel/wait.c b/kernel/wait.c
index 42a2dbc..7b866ab 100644
--- a/kernel/wait.c
+++ b/kernel/wait.c
@@ -73,7 +73,9 @@ prepare_to_wait(wait_queue_head_t *q, wait_queue_t *wait, int state)
 	if (list_empty(&wait->task_list))
 		__add_wait_queue(q, wait);
 	set_current_state(state);
+	preempt_disable();
 	spin_unlock_irqrestore(&q->lock, flags);
+	preempt_enable_no_resched();
 }
 EXPORT_SYMBOL(prepare_to_wait);
 
@@ -87,7 +89,9 @@ prepare_to_wait_exclusive(wait_queue_head_t *q, wait_queue_t *wait, int state)
 	if (list_empty(&wait->task_list))
 		__add_wait_queue_tail(q, wait);
 	set_current_state(state);
+	preempt_disable();
 	spin_unlock_irqrestore(&q->lock, flags);
+	preempt_enable_no_resched();
 }
 EXPORT_SYMBOL(prepare_to_wait_exclusive);
 

Re: [RFC][PATCH]: wait: don't resched in prepare_to_wait()

[Oleg Nesterov]

On 03/06, Peter Zijlstra wrote:
>
> After prepare to wait, we either call schedule() or find !cond in which
> case we'll call finish_wait() which contains another preemption check.
>
> @@ -73,7 +73,9 @@ prepare_to_wait(wait_queue_head_t *q, wait_queue_t *wait, int state)
>  	if (list_empty(&wait->task_list))
>  		__add_wait_queue(q, wait);
>  	set_current_state(state);
> +	preempt_disable();
>  	spin_unlock_irqrestore(&q->lock, flags);
> +	preempt_enable_no_resched();

But this only helps if the task gets TIF_NEED_RESCHED while it was in
spin_lock_irqsave() section, right? This window should be very small
unless we have the contention.

Or do you have something else in mind?

Oleg.

Re: [RFC][PATCH]: wait: don't resched in prepare_to_wait()

[Peter Zijlstra]

On Fri, 2009-03-06 at 16:41 +0100, Oleg Nesterov wrote:
> On 03/06, Peter Zijlstra wrote:
> >
> > After prepare to wait, we either call schedule() or find !cond in which
> > case we'll call finish_wait() which contains another preemption check.
> >
> > @@ -73,7 +73,9 @@ prepare_to_wait(wait_queue_head_t *q, wait_queue_t *wait, int state)
> >  	if (list_empty(&wait->task_list))
> >  		__add_wait_queue(q, wait);
> >  	set_current_state(state);
> > +	preempt_disable();
> >  	spin_unlock_irqrestore(&q->lock, flags);
> > +	preempt_enable_no_resched();
> 
> But this only helps if the task gets TIF_NEED_RESCHED while it was in
> spin_lock_irqsave() section, right? This window should be very small
> unless we have the contention.
> 
> Or do you have something else in mind?

No, that was exactly it, we then get preempted only to go to sleep right
afterwards, which would cause a schedule anyway.

So its avoiding a schedule.