Re: [RFC][PATCH] sched: Cache aware load-balancing

[Peter Zijlstra <peterz@infradead.org>]

On Wed, Mar 26, 2025 at 10:38:41AM +0100, Peter Zijlstra wrote:

> Nah, the saner thing to do is to preserve the topology averages and look
> at those instead of the per-cpu values.
> 
> Eg. have task_cache_work() compute and store averages in the
> sched_domain structure and then use those.

A little something like so perhaps ?

This immediately also gives the information required for clusters and
finding the best LLC of a Node and things like that.

--- a/include/linux/sched/topology.h
+++ b/include/linux/sched/topology.h
@@ -82,6 +82,9 @@ struct sched_domain_shared {
 	atomic_t	nr_busy_cpus;
 	int		has_idle_cores;
 	int		nr_idle_scan;
+
+	unsigned long   sum_occ;
+	unsigned long	avg_occ;
 };
 
 struct sched_domain {
--- a/kernel/sched/fair.c
+++ b/kernel/sched/fair.c
@@ -1286,8 +1286,8 @@ static void task_cache_work(struct callb
 	struct task_struct *p = current;
 	struct mm_struct *mm = p->mm;
 	unsigned long m_a_occ = 0;
-	int cpu, m_a_cpu = -1;
-	cpumask_var_t cpus;
+	int m_a_cpu = -1;
+	int cpu;
 
 	WARN_ON_ONCE(work != &p->cache_work);
 
@@ -1296,46 +1296,46 @@ static void task_cache_work(struct callb
 	if (p->flags & PF_EXITING)
 		return;
 
-	if (!alloc_cpumask_var(&cpus, GFP_KERNEL))
-		return;
-
 	scoped_guard (cpus_read_lock) {
-		cpumask_copy(cpus, cpu_online_mask);
 
-		for_each_cpu(cpu, cpus) {
-			/* XXX sched_cluster_active */
-			struct sched_domain *sd = per_cpu(sd_llc, cpu);
-			unsigned long occ, m_occ = 0, a_occ = 0;
-			int m_cpu = -1, nr = 0, i;
+		for_each_online_cpu(cpu) {
+			struct sched_domain *sd;
+			struct sched_domain_shared *sds;
+			unsigned long occ;
+
+			for_each_domain(cpu, sd) {
+				if (!(sd->flags & SD_SHARE_LLC))
+					break;
 
-			for_each_cpu(i, sched_domain_span(sd)) {
+				sds = sd->shared;
 				occ = fraction_mm_sched(cpu_rq(i),
 							per_cpu_ptr(mm->pcpu_sched, i));
-				a_occ += occ;
-				if (occ > m_occ) {
-					m_occ = occ;
-					m_cpu = i;
-				}
-				nr++;
-				trace_printk("(%d) occ: %ld m_occ: %ld m_cpu: %d nr: %d\n",
-					     per_cpu(sd_llc_id, i), occ, m_occ, m_cpu, nr);
-			}
-
-			a_occ /= nr;
-			if (a_occ > m_a_occ) {
-				m_a_occ = a_occ;
-				m_a_cpu = m_cpu;
+				sds->sum_occ += occ + 1;
 			}
+		}
 
-			trace_printk("(%d) a_occ: %ld m_a_occ: %ld\n",
-				     per_cpu(sd_llc_id, cpu), a_occ, m_a_occ);
+		for_each_online_cpu(cpu) {
+			struct sched_domain *sd;
+			struct sched_domain_shared *sds;
+
+			for_each_domain(cpu, sd) {
+				if (!(sd->flags & SD_SHARE_LLC))
+					break;
+
+				sds = sd->shared;
+				if (sds->agg_occ) {
+					sds->avg_occ = (sds->agg_occ - sd->span_weight) /
+						       sd->span_weight;
+					sds->sum_occ = 0;
+				}
 
-			for_each_cpu(i, sched_domain_span(sd)) {
-				/* XXX threshold ? */
-				per_cpu_ptr(mm->pcpu_sched, i)->occ = a_occ;
+				if (sd == per_cpu(sd_llc, cpu)) {
+					if (sds->avg_occ > m_a_occ) {
+						m_a_occ = sds->avg_occ;
+						m_a_cpu = cpu;
+					}
+				}
 			}
-
-			cpumask_andnot(cpus, cpus, sched_domain_span(sd));
 		}
 	}
 
@@ -1346,8 +1346,6 @@ static void task_cache_work(struct callb
 		m_a_cpu = -1;
 
 	mm->mm_sched_cpu = m_a_cpu;
-
-	free_cpumask_var(cpus);
 }
 
 void init_sched_mm(struct task_struct *p)