Re: [PATCH] sched/core: Do not migrate ineligible tasks in sched_balance_rq()

From: Hao Jia
Date: Thu Nov 28 2024 - 06:30:51 EST




On 2024/11/28 17:19, Peter Zijlstra wrote:
On Thu, Nov 28, 2024 at 04:48:58PM +0800, Hao Jia wrote:
From: Hao Jia <jiahao1@xxxxxxxxxxx>

When the PLACE_LAG scheduling feature is enabled, if a task
is ineligible (lag < 0) on the source cpu runqueue, it will
also be ineligible when it is migrated to the destination
cpu runqueue.

Because we will keep the original equivalent lag of
the task in place_entity(). So if the task was ineligible
before, it will still be ineligible after migration.

This is not accurate, it will be eleigible, irrespective of lag, if
there are no other tasks. I think your patch tries to do this, but I'm
fairly sure it got it wrong.


Thank you for your reply. The expression in my commit message is inaccurate, and I will correct it in the patch v2. If I understand correctly, a task meeting the following conditions:

sched_feat(PLACE_LAG) && cfs_rq->nr_running && !entity_eligible(cfs_rq, &p->se),

will remain ineligible both before and after migration.

If I am wrong, please correct me. Thank you!




Therefore, we should skip the migration of ineligible tasks
to reduce ineffective task migrations, just like the task
throttled by cfs_bandwidth, until they become eligible.

And this misses an important case too -- load-balancing will try very
hard to balance load. If you disallow migrating tasks it might fail to
reach this goal. Since this is not a hard contraint it should eventually
give in and migrate it anyway.

That is, I would suggest allowing it when nr_balance_failed is non-zero.


Thank you for your suggestion, I will do it in the patch v2.

Thanks,
Hao

Signed-off-by: Hao Jia <jiahao1@xxxxxxxxxxx>
---
kernel/sched/fair.c | 7 ++++++-
1 file changed, 6 insertions(+), 1 deletion(-)

diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c
index fbdca89c677f..5564e16b6fdb 100644
--- a/kernel/sched/fair.c
+++ b/kernel/sched/fair.c
@@ -9358,13 +9358,14 @@ static inline int migrate_degrades_locality(struct task_struct *p,
static
int can_migrate_task(struct task_struct *p, struct lb_env *env)
{
+ struct cfs_rq *cfs_rq = task_cfs_rq(p);
int tsk_cache_hot;
lockdep_assert_rq_held(env->src_rq);
/*
* We do not migrate tasks that are:
- * 1) throttled_lb_pair, or
+ * 1) throttled_lb_pair, or task ineligible, or
* 2) cannot be migrated to this CPU due to cpus_ptr, or
* 3) running (obviously), or
* 4) are cache-hot on their current CPU.
@@ -9372,6 +9373,10 @@ int can_migrate_task(struct task_struct *p, struct lb_env *env)
if (throttled_lb_pair(task_group(p), env->src_cpu, env->dst_cpu))
return 0;
+ if (sched_feat(PLACE_LAG) && cfs_rq->nr_running &&
+ !entity_eligible(cfs_rq, &p->se))

Your indenting it wrong, please use: cino=(0:0

I will do it in the patch v2.

Thanks,
Hao


+ return 0;
+
/* Disregard percpu kthreads; they are where they need to be. */
if (kthread_is_per_cpu(p))
return 0;
--
2.34.1