sched_ext: idle: Introduce SCX_OPS_BUILTIN_IDLE_PER_NODE

Add the new scheduler flag SCX_OPS_BUILTIN_IDLE_PER_NODE, which allows
BPF schedulers to select between using a global flat idle cpumask or
multiple per-node cpumasks.

This only introduces the flag and the mechanism to enable/disable this
feature without affecting any scheduling behavior.

Cc: Yury Norov [NVIDIA] <yury.norov@gmail.com>
Signed-off-by: Andrea Righi <arighi@nvidia.com>
Reviewed-by: Yury Norov [NVIDIA] <yury.norov@gmail.com>
Signed-off-by: Tejun Heo <tj@kernel.org>

4 files changed, 46 insertions, 11 deletions

diff --git a/kernel/sched/ext.c b/kernel/sched/ext.c
index 7c17e05ed15b..330a359d7930 100644
--- a/kernel/sched/ext.c
+++ b/kernel/sched/ext.c
@@ -155,6 +155,12 @@ enum scx_ops_flags {
 	SCX_OPS_ALLOW_QUEUED_WAKEUP	= 1LLU << 5,
 
 	/*
+	 * If set, enable per-node idle cpumasks. If clear, use a single global
+	 * flat idle cpumask.
+	 */
+	SCX_OPS_BUILTIN_IDLE_PER_NODE	= 1LLU << 6,
+
+	/*
 	 * CPU cgroup support flags
 	 */
 	SCX_OPS_HAS_CGROUP_WEIGHT = 1LLU << 16,	/* cpu.weight */
@@ -165,6 +171,7 @@ enum scx_ops_flags {
 				  SCX_OPS_ENQ_MIGRATION_DISABLED |
 				  SCX_OPS_ALLOW_QUEUED_WAKEUP |
 				  SCX_OPS_SWITCH_PARTIAL |
+				  SCX_OPS_BUILTIN_IDLE_PER_NODE |
 				  SCX_OPS_HAS_CGROUP_WEIGHT,
 };
 
@@ -3427,7 +3434,7 @@ static void handle_hotplug(struct rq *rq, bool online)
 	atomic_long_inc(&scx_hotplug_seq);
 
 	if (scx_enabled())
-		scx_idle_update_selcpu_topology();
+		scx_idle_update_selcpu_topology(&scx_ops);
 
 	if (online && SCX_HAS_OP(cpu_online))
 		SCX_CALL_OP(SCX_KF_UNLOCKED, cpu_online, cpu);
@@ -5228,6 +5235,16 @@ static int validate_ops(const struct sched_ext_ops *ops)
 		return -EINVAL;
 	}
 
+	/*
+	 * SCX_OPS_BUILTIN_IDLE_PER_NODE requires built-in CPU idle
+	 * selection policy to be enabled.
+	 */
+	if ((ops->flags & SCX_OPS_BUILTIN_IDLE_PER_NODE) &&
+	    (ops->update_idle && !(ops->flags & SCX_OPS_KEEP_BUILTIN_IDLE))) {
+		scx_ops_error("SCX_OPS_BUILTIN_IDLE_PER_NODE requires CPU idle selection enabled");
+		return -EINVAL;
+	}
+
 	return 0;
 }
 
@@ -5352,7 +5369,7 @@ static int scx_ops_enable(struct sched_ext_ops *ops, struct bpf_link *link)
 			static_branch_enable_cpuslocked(&scx_has_op[i]);
 
 	check_hotplug_seq(ops);
-	scx_idle_update_selcpu_topology();
+	scx_idle_update_selcpu_topology(ops);
 
 	cpus_read_unlock();
 
diff --git a/kernel/sched/ext_idle.c b/kernel/sched/ext_idle.c
index ed1804506585..0912f94b95cd 100644
--- a/kernel/sched/ext_idle.c
+++ b/kernel/sched/ext_idle.c
@@ -14,6 +14,9 @@
 /* Enable/disable built-in idle CPU selection policy */
 static DEFINE_STATIC_KEY_FALSE(scx_builtin_idle_enabled);
 
+/* Enable/disable per-node idle cpumasks */
+static DEFINE_STATIC_KEY_FALSE(scx_builtin_idle_per_node);
+
 #ifdef CONFIG_SMP
 #ifdef CONFIG_CPUMASK_OFFSTACK
 #define CL_ALIGNED_IF_ONSTACK
@@ -204,7 +207,7 @@ static bool llc_numa_mismatch(void)
  * CPU belongs to a single LLC domain, and that each LLC domain is entirely
  * contained within a single NUMA node.
  */
-void scx_idle_update_selcpu_topology(void)
+void scx_idle_update_selcpu_topology(struct sched_ext_ops *ops)
 {
 	bool enable_llc = false, enable_numa = false;
 	unsigned int nr_cpus;
@@ -237,13 +240,19 @@ void scx_idle_update_selcpu_topology(void)
 	 * If all CPUs belong to the same NUMA node and the same LLC domain,
 	 * enabling both NUMA and LLC optimizations is unnecessary, as checking
 	 * for an idle CPU in the same domain twice is redundant.
+	 *
+	 * If SCX_OPS_BUILTIN_IDLE_PER_NODE is enabled ignore the NUMA
+	 * optimization, as we would naturally select idle CPUs within
+	 * specific NUMA nodes querying the corresponding per-node cpumask.
 	 */
-	nr_cpus = numa_weight(cpu);
-	if (nr_cpus > 0) {
-		if (nr_cpus < num_online_cpus() && llc_numa_mismatch())
-			enable_numa = true;
-		pr_debug("sched_ext: NUMA=%*pb weight=%u\n",
-			 cpumask_pr_args(numa_span(cpu)), numa_weight(cpu));
+	if (!(ops->flags & SCX_OPS_BUILTIN_IDLE_PER_NODE)) {
+		nr_cpus = numa_weight(cpu);
+		if (nr_cpus > 0) {
+			if (nr_cpus < num_online_cpus() && llc_numa_mismatch())
+				enable_numa = true;
+			pr_debug("sched_ext: NUMA=%*pb weight=%u\n",
+				 cpumask_pr_args(numa_span(cpu)), nr_cpus);
+		}
 	}
 	rcu_read_unlock();
 
@@ -530,6 +539,11 @@ void scx_idle_enable(struct sched_ext_ops *ops)
 	}
 	static_branch_enable(&scx_builtin_idle_enabled);
 
+	if (ops->flags & SCX_OPS_BUILTIN_IDLE_PER_NODE)
+		static_branch_enable(&scx_builtin_idle_per_node);
+	else
+		static_branch_disable(&scx_builtin_idle_per_node);
+
 #ifdef CONFIG_SMP
 	/*
 	 * Consider all online cpus idle. Should converge to the actual state
@@ -543,6 +557,7 @@ void scx_idle_enable(struct sched_ext_ops *ops)
 void scx_idle_disable(void)
 {
 	static_branch_disable(&scx_builtin_idle_enabled);
+	static_branch_disable(&scx_builtin_idle_per_node);
 }
 
 /********************************************************************************
diff --git a/kernel/sched/ext_idle.h b/kernel/sched/ext_idle.h
index bbac0fd9a5dd..339b6ec9c4cb 100644
--- a/kernel/sched/ext_idle.h
+++ b/kernel/sched/ext_idle.h
@@ -13,12 +13,12 @@
 struct sched_ext_ops;
 
 #ifdef CONFIG_SMP
-void scx_idle_update_selcpu_topology(void);
+void scx_idle_update_selcpu_topology(struct sched_ext_ops *ops);
 void scx_idle_init_masks(void);
 bool scx_idle_test_and_clear_cpu(int cpu);
 s32 scx_pick_idle_cpu(const struct cpumask *cpus_allowed, u64 flags);
 #else /* !CONFIG_SMP */
-static inline void scx_idle_update_selcpu_topology(void) {}
+static inline void scx_idle_update_selcpu_topology(struct sched_ext_ops *ops) {}
 static inline void scx_idle_init_masks(void) {}
 static inline bool scx_idle_test_and_clear_cpu(int cpu) { return false; }
 static inline s32 scx_pick_idle_cpu(const struct cpumask *cpus_allowed, u64 flags)
diff --git a/tools/sched_ext/include/scx/compat.h b/tools/sched_ext/include/scx/compat.h
index b50280e2ba2b..d63cf40be8ee 100644
--- a/tools/sched_ext/include/scx/compat.h
+++ b/tools/sched_ext/include/scx/compat.h
@@ -109,6 +109,9 @@ static inline bool __COMPAT_struct_has_field(const char *type, const char *field
 #define SCX_OPS_SWITCH_PARTIAL							\
 	__COMPAT_ENUM_OR_ZERO("scx_ops_flags", "SCX_OPS_SWITCH_PARTIAL")
 
+#define SCX_OPS_BUILTIN_IDLE_PER_NODE						\
+	__COMPAT_ENUM_OR_ZERO("scx_ops_flags", "SCX_OPS_BUILTIN_IDLE_PER_NODE")
+
 static inline long scx_hotplug_seq(void)
 {
 	int fd;