The estimator kthreads' affinity are defined by sysctl overwritten
preferences and applied through a plain call to the scheduler's affinity
API.

However since the introduction of managed kthreads preferred affinity,
such a practice shortcuts the kthreads core code which eventually
overwrites the target to the default unbound affinity.

Fix this with using the appropriate kthread's API which will carry the
desired affinity and maintain it across CPU hotplug events and CPU
isolation constraints.

Fixes: d1a89197589c ("kthread: Default affine kthread to its preferred NUMA node")
Signed-off-by: Frederic Weisbecker <frederic@kernel.org>
---
 include/net/ip_vs.h            | 13 +++++++++++++
 net/netfilter/ipvs/ip_vs_est.c |  3 ++-
 2 files changed, 15 insertions(+), 1 deletion(-)

diff --git a/include/net/ip_vs.h b/include/net/ip_vs.h
index ff406ef4fd4a..29a36709e7f3 100644
--- a/include/net/ip_vs.h
+++ b/include/net/ip_vs.h
@@ -1163,6 +1163,14 @@ static inline const struct cpumask *sysctl_est_cpulist(struct netns_ipvs *ipvs)
 		return housekeeping_cpumask(HK_TYPE_KTHREAD);
 }
 
+static inline const struct cpumask *sysctl_est_preferred_cpulist(struct netns_ipvs *ipvs)
+{
+	if (ipvs->est_cpulist_valid)
+		return ipvs->sysctl_est_cpulist;
+	else
+		return NULL;
+}
+
 static inline int sysctl_est_nice(struct netns_ipvs *ipvs)
 {
 	return ipvs->sysctl_est_nice;
@@ -1270,6 +1278,11 @@ static inline const struct cpumask *sysctl_est_cpulist(struct netns_ipvs *ipvs)
 	return housekeeping_cpumask(HK_TYPE_KTHREAD);
 }
 
+static inline const struct cpumask *sysctl_est_preferred_cpulist(struct netns_ipvs *ipvs)
+{
+	return NULL;
+}
+
 static inline int sysctl_est_nice(struct netns_ipvs *ipvs)
 {
 	return IPVS_EST_NICE;
diff --git a/net/netfilter/ipvs/ip_vs_est.c b/net/netfilter/ipvs/ip_vs_est.c
index f821ad2e19b3..15049b826732 100644
--- a/net/netfilter/ipvs/ip_vs_est.c
+++ b/net/netfilter/ipvs/ip_vs_est.c
@@ -265,7 +265,8 @@ int ip_vs_est_kthread_start(struct netns_ipvs *ipvs,
 	}
 
 	set_user_nice(kd->task, sysctl_est_nice(ipvs));
-	set_cpus_allowed_ptr(kd->task, sysctl_est_cpulist(ipvs));
+	if (sysctl_est_preferred_cpulist(ipvs))
+		kthread_affine_preferred(kd->task, sysctl_est_preferred_cpulist(ipvs));
 
 	pr_info("starting estimator thread %d...\n", kd->id);
 	wake_up_process(kd->task);
-- 
2.48.1

On Tue, Jul 29, 2025 at 02:06:59PM +0200, Frederic Weisbecker wrote:
> The estimator kthreads' affinity are defined by sysctl overwritten
> preferences and applied through a plain call to the scheduler's affinity
> API.
> 
> However since the introduction of managed kthreads preferred affinity,
> such a practice shortcuts the kthreads core code which eventually
> overwrites the target to the default unbound affinity.
> 
> Fix this with using the appropriate kthread's API which will carry the
> desired affinity and maintain it across CPU hotplug events and CPU
> isolation constraints.
> 
> Fixes: d1a89197589c ("kthread: Default affine kthread to its preferred NUMA node")
> Signed-off-by: Frederic Weisbecker <frederic@kernel.org>

Acked-by: Simon Horman <horms@kernel.org>