2021-04-14 10:13:35

by jun qian

Subject: [PATCH V2 1/1] sched/fair:Reduce unnecessary check preempt in the sched tick

From: jun qian <[email protected]>

As you are already set the TIF_NEED_RESCHED, there is no need
to check resched again.

Signed-off-by: jun qian <[email protected]>
---
kernel/sched/fair.c | 11 +++++++++--
1 file changed, 9 insertions(+), 2 deletions(-)

diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c
index 794c2cb945f8..1a69b5fffe4a 100644
--- a/kernel/sched/fair.c
+++ b/kernel/sched/fair.c
@@ -4360,19 +4360,26 @@ check_preempt_tick(struct cfs_rq *cfs_rq, struct sched_entity *curr)
{
unsigned long ideal_runtime, delta_exec;
struct sched_entity *se;
+ struct rq *rq = rq_of(cfs_rq);
s64 delta;

ideal_runtime = sched_slice(cfs_rq, curr);
delta_exec = curr->sum_exec_runtime - curr->prev_sum_exec_runtime;
if (delta_exec > ideal_runtime) {
- resched_curr(rq_of(cfs_rq));
+ if (!test_tsk_need_resched(rq->curr))
+ resched_curr(rq_of(cfs_rq));
/*
* The current task ran long enough, ensure it doesn't get
* re-elected due to buddy favours.
*/
clear_buddies(cfs_rq, curr);
return;
- }
+ /*
+ * If here with TIF_NEED_RESCHED already set from the early entity_tick,
+ * there is no need to check again.
+ */
+ } else if (test_tsk_need_resched(rq->curr))
+ return;

/*
* Ensure that a task that missed wakeup preemption by a
--
2.18.2

2021-04-14 19:55:51

by Peter Zijlstra

Subject: Re: [PATCH V2 1/1] sched/fair:Reduce unnecessary check preempt in the sched tick

On Wed, Apr 14, 2021 at 10:22:29AM +0800, [email protected] wrote:
> From: jun qian <[email protected]>
>
> As you are already set the TIF_NEED_RESCHED, there is no need
> to check resched again.

Still no justification; does this actually help anything?

> Signed-off-by: jun qian <[email protected]>
> ---
> kernel/sched/fair.c | 11 +++++++++--
> 1 file changed, 9 insertions(+), 2 deletions(-)
>
> diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c
> index 794c2cb945f8..1a69b5fffe4a 100644
> --- a/kernel/sched/fair.c
> +++ b/kernel/sched/fair.c
> @@ -4360,19 +4360,26 @@ check_preempt_tick(struct cfs_rq *cfs_rq, struct sched_entity *curr)
> {
> unsigned long ideal_runtime, delta_exec;
> struct sched_entity *se;
> + struct rq *rq = rq_of(cfs_rq);
> s64 delta;
>
> ideal_runtime = sched_slice(cfs_rq, curr);
> delta_exec = curr->sum_exec_runtime - curr->prev_sum_exec_runtime;
> if (delta_exec > ideal_runtime) {
> - resched_curr(rq_of(cfs_rq));
> + if (!test_tsk_need_resched(rq->curr))
> + resched_curr(rq_of(cfs_rq));
> /*
> * The current task ran long enough, ensure it doesn't get
> * re-elected due to buddy favours.
> */
> clear_buddies(cfs_rq, curr);
> return;
> - }
> + /*
> + * If here with TIF_NEED_RESCHED already set from the early entity_tick,
> + * there is no need to check again.
> + */
> + } else if (test_tsk_need_resched(rq->curr))
> + return;

This is horrific style. And, afaict, completely unnecessary.