[PATCH] sched/fair: Remove sd->nohz_idle

[PATCH] sched/fair: Remove sd->nohz_idle

[Vincent Guittot]

sd->nohz_idle is used to call once inc|dec of &sd->shared->nr_busy_cpus
when entering or leaving idle state but the call to
set_cpu_sd_state_idle|busy is already protected by rq->nohz_tick_stopped
being already set or clear.

Remove the useless sd->nohz_idle field which equals !rq->nohz_tick_stopped.

Signed-off-by: Vincent Guittot <vincent.guittot@linaro.org>
---
 include/linux/sched/topology.h |  1 -
 kernel/sched/fair.c            | 16 ++++------------
 2 files changed, 4 insertions(+), 13 deletions(-)

diff --git a/include/linux/sched/topology.h b/include/linux/sched/topology.h
index a1e1032426dc..59e498006072 100644
--- a/include/linux/sched/topology.h
+++ b/include/linux/sched/topology.h
@@ -82,7 +82,6 @@ struct sched_domain {
 	unsigned int cache_nice_tries;	/* Leave cache hot tasks for # tries */
 	unsigned int imb_numa_nr;	/* Nr running tasks that allows a NUMA imbalance */
 
-	int nohz_idle;			/* NOHZ IDLE status */
 	int flags;			/* See SD_* */
 	int level;
 
diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c
index d57c02e82f3a..888875e91073 100644
--- a/kernel/sched/fair.c
+++ b/kernel/sched/fair.c
@@ -12818,12 +12818,8 @@ static void set_cpu_sd_state_busy(int cpu)
 	rcu_read_lock();
 	sd = rcu_dereference_all(per_cpu(sd_llc, cpu));
 
-	if (!sd || !sd->nohz_idle)
-		goto unlock;
-	sd->nohz_idle = 0;
-
-	atomic_inc(&sd->shared->nr_busy_cpus);
-unlock:
+	if (likely(sd))
+		atomic_inc(&sd->shared->nr_busy_cpus);
 	rcu_read_unlock();
 }
 
@@ -12847,12 +12843,8 @@ static void set_cpu_sd_state_idle(int cpu)
 	rcu_read_lock();
 	sd = rcu_dereference_all(per_cpu(sd_llc, cpu));
 
-	if (!sd || sd->nohz_idle)
-		goto unlock;
-	sd->nohz_idle = 1;
-
-	atomic_dec(&sd->shared->nr_busy_cpus);
-unlock:
+	if (likely(sd))
+		atomic_dec(&sd->shared->nr_busy_cpus);
 	rcu_read_unlock();
 }
 
-- 
2.43.0

Re: [PATCH] sched/fair: Remove sd->nohz_idle

[K Prateek Nayak]

Hello Vincent,

On 2/27/2026 10:17 PM, Vincent Guittot wrote:
> sd->nohz_idle is used to call once inc|dec of &sd->shared->nr_busy_cpus
> when entering or leaving idle state but the call to
> set_cpu_sd_state_idle|busy is already protected by rq->nohz_tick_stopped
> being already set or clear.
> 
> Remove the useless sd->nohz_idle field which equals !rq->nohz_tick_stopped.

I had looked at this recently and I believe the following is
possible with hotplug/cpuset:

    CPU0                                        CPU1
    ====                                        ====

  nohz_balance_enter_idle()
    atomic_dec(&sd->shared->nr_busy_cpus);
  ... /* Idle */
                                               /* Sched domains are rebuilt */
                                               atomic_set(&sd->shared->nr_busy_cpus, sd_weight);
                                               update_top_cache_domain(0 /* CPU0 */)
                                                 sd = highest_flag_domain(cpu, SD_SHARE_LLC);
                                                 rcu_assign_pointer(per_cpu(sd_llc, cpu), sd); /* For CPU0 */
  ... /* Exits idle */
  ...
  ... /* First tick hits */
  /* Exits idle; First tick hits */
  if (rq->nohz_tick_stopped) /* True */
     nohz_balance_exit_idle()
       set_cpu_sd_state_busy()
         atomic_inc(&sd->shared->nr_busy_cpus) /* !!! Crosses sd_weight !!! */


I feel this per-SD indicator is necessary to avoid this scenario.
Is it fixed in some other way that I haven't realised yet?

-- 
Thanks and Regards,
Prateek

Re: [PATCH] sched/fair: Remove sd->nohz_idle

[Vincent Guittot]

On Fri, 27 Feb 2026 at 18:33, K Prateek Nayak <kprateek.nayak@amd.com> wrote:
>
> Hello Vincent,
>
> On 2/27/2026 10:17 PM, Vincent Guittot wrote:
> > sd->nohz_idle is used to call once inc|dec of &sd->shared->nr_busy_cpus
> > when entering or leaving idle state but the call to
> > set_cpu_sd_state_idle|busy is already protected by rq->nohz_tick_stopped
> > being already set or clear.
> >
> > Remove the useless sd->nohz_idle field which equals !rq->nohz_tick_stopped.
>
> I had looked at this recently and I believe the following is
> possible with hotplug/cpuset:
>
>     CPU0                                        CPU1
>     ====                                        ====
>
>   nohz_balance_enter_idle()
>     atomic_dec(&sd->shared->nr_busy_cpus);
>   ... /* Idle */
>                                                /* Sched domains are rebuilt */
>                                                atomic_set(&sd->shared->nr_busy_cpus, sd_weight);
>                                                update_top_cache_domain(0 /* CPU0 */)
>                                                  sd = highest_flag_domain(cpu, SD_SHARE_LLC);
>                                                  rcu_assign_pointer(per_cpu(sd_llc, cpu), sd); /* For CPU0 */
>   ... /* Exits idle */
>   ...
>   ... /* First tick hits */
>   /* Exits idle; First tick hits */
>   if (rq->nohz_tick_stopped) /* True */
>      nohz_balance_exit_idle()
>        set_cpu_sd_state_busy()
>          atomic_inc(&sd->shared->nr_busy_cpus) /* !!! Crosses sd_weight !!! */
>
>
> I feel this per-SD indicator is necessary to avoid this scenario.
> Is it fixed in some other way that I haven't realised yet?

You're right !
And I don't have an easy way to fix this


>
> --
> Thanks and Regards,
> Prateek
>