Re: [PATCH] sched/isolation: Prefer housekeeping cpu in local node

[Peter Zijlstra]

On Thu, Jun 20, 2019 at 07:36:54PM +0800, Wanpeng Li wrote:
> From: Wanpeng Li <wanpengli@xxxxxxxxxxx>
> 
> In real product setup, there will be houseeking cpus in each nodes, it 
> is prefer to do housekeeping from local node, fallback to global online 
> cpumask if failed to find houseeking cpu from local node.
> 
> Cc: Ingo Molnar <mingo@xxxxxxxxxx> 
> Cc: Peter Zijlstra <peterz@xxxxxxxxxxxxx>
> Cc: Frederic Weisbecker <frederic@xxxxxxxxxx>
> Signed-off-by: Wanpeng Li <wanpengli@xxxxxxxxxxx>
> ---
>  kernel/sched/isolation.c | 11 +++++++++--
>  1 file changed, 9 insertions(+), 2 deletions(-)
> 
> diff --git a/kernel/sched/isolation.c b/kernel/sched/isolation.c
> index 123ea07..9eb6805 100644
> --- a/kernel/sched/isolation.c
> +++ b/kernel/sched/isolation.c
> @@ -16,9 +16,16 @@ static unsigned int housekeeping_flags;
>  
>  int housekeeping_any_cpu(enum hk_flags flags)
>  {
> +	int cpu;
> +
>  	if (static_branch_unlikely(&housekeeping_overridden))
> -		if (housekeeping_flags & flags)
> -			return cpumask_any_and(housekeeping_mask, cpu_online_mask);
> +		if (housekeeping_flags & flags) {
> +			cpu = cpumask_any_and(housekeeping_mask, cpu_cpu_mask(smp_processor_id()));
> +			if (cpu < nr_cpu_ids)
> +				return cpu;
> +			else
> +				return cpumask_any_and(housekeeping_mask, cpu_online_mask);
> +		}
>  	return smp_processor_id();
>  }
>  EXPORT_SYMBOL_GPL(housekeeping_any_cpu);

Why not something like so? IIRC there's more places that want this, but
I can't seem to remember quite where.

diff --git a/kernel/sched/isolation.c b/kernel/sched/isolation.c
index 123ea07a3f3b..1cceab5f094c 100644
--- a/kernel/sched/isolation.c
+++ b/kernel/sched/isolation.c
@@ -16,9 +16,15 @@ static unsigned int housekeeping_flags;
 
 int housekeeping_any_cpu(enum hk_flags flags)
 {
-	if (static_branch_unlikely(&housekeeping_overridden))
-		if (housekeeping_flags & flags)
+	if (static_branch_unlikely(&housekeeping_overridden)) {
+		if (housekeeping_flags & flags) {
+			cpu = sched_numa_find_closest(housekeeping_mask, smp_processor_id());
+			if (cpu < nr_cpu_ids)
+				return cpu;
+
 			return cpumask_any_and(housekeeping_mask, cpu_online_mask);
+		}
+	}
 	return smp_processor_id();
 }
 EXPORT_SYMBOL_GPL(housekeeping_any_cpu);
diff --git a/kernel/sched/sched.h b/kernel/sched/sched.h
index b08dee29ef5e..0db7431c7207 100644
--- a/kernel/sched/sched.h
+++ b/kernel/sched/sched.h
@@ -1212,9 +1212,6 @@ enum numa_topology_type {
 extern enum numa_topology_type sched_numa_topology_type;
 extern int sched_max_numa_distance;
 extern bool find_numa_distance(int distance);
-#endif
-
-#ifdef CONFIG_NUMA
 extern void sched_init_numa(void);
 extern void sched_domains_numa_masks_set(unsigned int cpu);
 extern void sched_domains_numa_masks_clear(unsigned int cpu);
@@ -1224,6 +1221,8 @@ static inline void sched_domains_numa_masks_set(unsigned int cpu) { }
 static inline void sched_domains_numa_masks_clear(unsigned int cpu) { }
 #endif
 
+extern int sched_numa_find_closest(const struct cpumask *cpus, int cpu);
+
 #ifdef CONFIG_NUMA_BALANCING
 /* The regions in numa_faults array from task_struct */
 enum numa_faults_stats {
diff --git a/kernel/sched/topology.c b/kernel/sched/topology.c
index 63184cf0d0d7..408e94a6637c 100644
--- a/kernel/sched/topology.c
+++ b/kernel/sched/topology.c
@@ -1726,6 +1726,20 @@ void sched_domains_numa_masks_clear(unsigned int cpu)
 
 #endif /* CONFIG_NUMA */
 
+int sched_numa_find_closest(const struct cpumask *cpus, int cpu)
+{
+#ifdef CONFIG_NUMA
+	int i, j = cpu_to_node(cpu);
+
+	for (i = 0; i < sched_domains_numa_levels; ++) {
+		cpu = cpumask_any_and(cpus, sched_domains_numa_mask[i][j]);
+		if (cpu < nr_cpu_ids)
+			return cpu;
+	}
+#endif
+	return nr_cpu_ids;
+}
+
 static int __sdt_alloc(const struct cpumask *cpu_map)
 {
 	struct sched_domain_topology_level *tl;