Re: [PATCH] [Question] sched/fair: Task starvation with RUN_TO_PARITY_WAKEUP under group topologies

[K Prateek Nayak <kprateek.nayak@amd.com>]

Hello Chen,

On 7/12/2026 5:01 PM, Chen Jinghuang wrote:
> Hi all,
> 
> While adapting the RUN_TO_PARITY_WAKEUP feature on top of mainline v7.2-rc2,
> I encountered a severe task starvation issue under specific cgroup topologies.
> Specifically, a running task completely hogs the CPU and prevents any
> preemption.

I'm assuming you are referring to
https://lore.kernel.org/lkml/20240325060226.1540-2-kprateek.nayak@amd.com/

There was a reason it was not mainlined for this exact issue.

...


> Discussion:
> I'd love to hear your thoughts on how we can fix this issue without losing
> the throughput gains of RUN_TO_PARITY_WAKEUP.

Run your tasks as SCHED_BATCH and they'll forego wakeup preemption.
This is the exact recipe that had helped a bunch of workloads that
suffered throughput degradation from wakeup preemption after EEVDF
was introduced.

If you have well behaved workloads, you can also play around with
custom slice (sched_attr.sched_runtime for fair tasks) which gets
propagated down the cgroup hierarchy and should be better behaved.

> 
> Thanks all.
> 
> Signed-off-by: Chen Jinghuang <chenjinghuang2@huawei.com>
> ---
>  kernel/sched/fair.c | 18 +++++++++++++++++-
>  1 file changed, 17 insertions(+), 1 deletion(-)
> 
> diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c
> index d78467ec6ee1..00869624d914 100644
> --- a/kernel/sched/fair.c
> +++ b/kernel/sched/fair.c
> @@ -1157,7 +1157,23 @@ static struct sched_entity *pick_eevdf(struct cfs_rq *cfs_rq, bool protect)
>  		return cfs_rq->next;
>  	}
>  
> -	if (curr && (!curr->on_rq || !entity_eligible(cfs_rq, curr)))
> +	if (curr && !curr->on_rq)
> +		curr = NULL;
> +
> +	/*
> +	 * When an entity with positive lag wakes up, it pushes the
> +	 * avg_vruntime of the runqueue backwards. This may causes the
> +	 * current entity to be ineligible soon into its run leading to
> +	 * wakeup preemption.
> +	 *
> +	 * To prevent such aggressive preemption of the current running
> +	 * entity during task wakeups, skip the eligibility check if the
> +	 * slice promised to the entity since its selection has not yet
> +	 * elapsed.
> +	 */
> +	if (curr &&
> +	    !(sched_feat(RUN_TO_PARITY_WAKEUP) && protect && protect_slice(curr)) &&
> +	    !entity_eligible(cfs_rq, curr))
>  		curr = NULL;

We also have a bunch of changes in tip:sched/core that improves wakeup
preemption with custom slices. You may want to check them out too:
https://git.kernel.org/pub/scm/linux/kernel/git/tip/tip.git/log/?h=sched/core

>  
>  	if (curr && protect && protect_slice(curr))

-- 
Thanks and Regards,
Prateek