sched_ext: Add support for cgroup bandwidth control interface

From 077814f57f8acce13f91dc34bbd2b7e4911fbf25 Mon Sep 17 00:00:00 2001
From: Tejun Heo <[e-mail]>
Date: Fri, 13 Jun 2025 15:06:47 -1000

- Add CONFIG_GROUP_SCHED_BANDWIDTH which is selected by both
CONFIG_CFS_BANDWIDTH and EXT_GROUP_SCHED.

- Put bandwidth control interface files for both cgroup v1 and v2 under
CONFIG_GROUP_SCHED_BANDWIDTH.

- Update tg_bandwidth() to fetch configuration parameters from fair if
CONFIG_CFS_BANDWIDTH, SCX otherwise.

- Update tg_set_bandwidth() to update the parameters for both fair and SCX.

- Add bandwidth control parameters to struct scx_cgroup_init_args.

- Add sched_ext_ops.cgroup_set_bandwidth() which is invoked on bandwidth
control parameter updates.

- Update scx_qmap and maximal selftest to test the new feature.

Signed-off-by: Tejun Heo <[e-mail]>

include/linux/sched/ext.h

···
 #ifdef CONFIG_EXT_GROUP_SCHED
 	u32			flags;		/* SCX_TG_* */
 	u32			weight;
+	u64			bw_period_us;
+	u64			bw_quota_us;
+	u64			bw_burst_us;
 #endif
 };
 

init/Kconfig

···
 config GROUP_SCHED_WEIGHT
 	def_bool n
 
+config GROUP_SCHED_BANDWIDTH
+        def_bool n
+
 config FAIR_GROUP_SCHED
 	bool "Group scheduling for SCHED_OTHER"
 	depends on CGROUP_SCHED
···
 config CFS_BANDWIDTH
 	bool "CPU bandwidth provisioning for FAIR_GROUP_SCHED"
 	depends on FAIR_GROUP_SCHED
+	select GROUP_SCHED_BANDWIDTH
 	default n
 	help
 	  This option allows users to define CPU bandwidth rates (limits) for
···
 	bool
 	depends on SCHED_CLASS_EXT && CGROUP_SCHED
 	select GROUP_SCHED_WEIGHT
+	select GROUP_SCHED_BANDWIDTH
 	default y
 
 endif #CGROUP_SCHED

kernel/sched/core.c

···
 
 	return 0;
 }
+#endif /* CONFIG_CFS_BANDWIDTH */
 
+#ifdef CONFIG_GROUP_SCHED_BANDWIDTH
 const u64 max_bw_quota_period_us = 1 * USEC_PER_SEC; /* 1s */
 static const u64 min_bw_quota_period_us = 1 * USEC_PER_MSEC; /* 1ms */
 /* More than 203 days if BW_SHIFT equals 20. */
···
 static void tg_bandwidth(struct task_group *tg,
 			 u64 *period_us_p, u64 *quota_us_p, u64 *burst_us_p)
 {
+#ifdef CONFIG_CFS_BANDWIDTH
 	if (period_us_p)
 		*period_us_p = tg_get_cfs_period(tg);
 	if (quota_us_p)
 		*quota_us_p = tg_get_cfs_quota(tg);
 	if (burst_us_p)
 		*burst_us_p = tg_get_cfs_burst(tg);
+#else /* !CONFIG_CFS_BANDWIDTH */
+	if (period_us_p)
+		*period_us_p = tg->scx.bw_period_us;
+	if (quota_us_p)
+		*quota_us_p = tg->scx.bw_quota_us;
+	if (burst_us_p)
+		*burst_us_p = tg->scx.bw_burst_us;
+#endif /* CONFIG_CFS_BANDWIDTH */
 }
 
 static u64 cpu_period_read_u64(struct cgroup_subsys_state *css,
···
 			    u64 period_us, u64 quota_us, u64 burst_us)
 {
 	const u64 max_usec = U64_MAX / NSEC_PER_USEC;
+	int ret = 0;
 
 	if (tg == &root_task_group)
 		return -EINVAL;
···
 					burst_us + quota_us > max_bw_runtime_us))
 		return -EINVAL;
 
-	return tg_set_cfs_bandwidth(tg, period_us, quota_us, burst_us);
+#ifdef CONFIG_CFS_BANDWIDTH
+	ret = tg_set_cfs_bandwidth(tg, period_us, quota_us, burst_us);
+#endif /* CONFIG_CFS_BANDWIDTH */
+	if (!ret)
+		scx_group_set_bandwidth(tg, period_us, quota_us, burst_us);
+	return ret;
 }
 
 static s64 cpu_quota_read_s64(struct cgroup_subsys_state *css,
···
 	tg_bandwidth(tg, &period_us, &quota_us, NULL);
 	return tg_set_bandwidth(tg, period_us, quota_us, burst_us);
 }
-#endif /* CONFIG_CFS_BANDWIDTH */
+#endif /* CONFIG_GROUP_SCHED_BANDWIDTH */
 
 #ifdef CONFIG_RT_GROUP_SCHED
 static int cpu_rt_runtime_write(struct cgroup_subsys_state *css,
···
 		.write_s64 = cpu_idle_write_s64,
 	},
 #endif
-#ifdef CONFIG_CFS_BANDWIDTH
+#ifdef CONFIG_GROUP_SCHED_BANDWIDTH
 	{
 		.name = "cfs_period_us",
 		.read_u64 = cpu_period_read_u64,
···
 		.read_u64 = cpu_burst_read_u64,
 		.write_u64 = cpu_burst_write_u64,
 	},
+#endif
+#ifdef CONFIG_CFS_BANDWIDTH
 	{
 		.name = "stat",
 		.seq_show = cpu_cfs_stat_show,
···
 	return 0;
 }
 
-#ifdef CONFIG_CFS_BANDWIDTH
+#ifdef CONFIG_GROUP_SCHED_BANDWIDTH
 static int cpu_max_show(struct seq_file *sf, void *v)
 {
 	struct task_group *tg = css_tg(seq_css(sf));
···
 		.write_s64 = cpu_idle_write_s64,
 	},
 #endif
-#ifdef CONFIG_CFS_BANDWIDTH
+#ifdef CONFIG_GROUP_SCHED_BANDWIDTH
 	{
 		.name = "max",
 		.flags = CFTYPE_NOT_ON_ROOT,

kernel/sched/ext.c

···
 struct scx_cgroup_init_args {
 	/* the weight of the cgroup [1..10000] */
 	u32			weight;
+
+	/* bandwidth control parameters from cpu.max and cpu.max.burst */
+	u64			bw_period_us;
+	u64			bw_quota_us;
+	u64			bw_burst_us;
 };
 
 enum scx_cpu_preempt_reason {
···
 	 * @cgrp: cgroup whose weight is being updated
 	 * @weight: new weight [1..10000]
 	 *
-	 * Update @tg's weight to @weight.
+	 * Update @cgrp's weight to @weight.
 	 */
 	void (*cgroup_set_weight)(struct cgroup *cgrp, u32 weight);
+
+	/**
+	 * @cgroup_set_bandwidth: A cgroup's bandwidth is being changed
+	 * @cgrp: cgroup whose bandwidth is being updated
+	 * @period_us: bandwidth control period
+	 * @quota_us: bandwidth control quota
+	 * @burst_us: bandwidth control burst
+	 *
+	 * Update @cgrp's bandwidth control parameters. This is from the cpu.max
+	 * cgroup interface.
+	 *
+	 * @quota_us / @period_us determines the CPU bandwidth @cgrp is entitled
+	 * to. For example, if @period_us is 1_000_000 and @quota_us is
+	 * 2_500_000. @cgrp is entitled to 2.5 CPUs. @burst_us can be
+	 * interpreted in the same fashion and specifies how much @cgrp can
+	 * burst temporarily. The specific control mechanism and thus the
+	 * interpretation of @period_us and burstiness is upto to the BPF
+	 * scheduler.
+	 */
+	void (*cgroup_set_bandwidth)(struct cgroup *cgrp,
+				     u64 period_us, u64 quota_us, u64 burst_us);
+
 #endif	/* CONFIG_EXT_GROUP_SCHED */
 
 	/*
···
 void scx_tg_init(struct task_group *tg)
 {
 	tg->scx.weight = CGROUP_WEIGHT_DFL;
+	tg->scx.bw_period_us = default_bw_period_us();
+	tg->scx.bw_quota_us = RUNTIME_INF;
 }
 
 int scx_tg_online(struct task_group *tg)
···
 	if (scx_cgroup_enabled) {
 		if (SCX_HAS_OP(sch, cgroup_init)) {
 			struct scx_cgroup_init_args args =
-				{ .weight = tg->scx.weight };
+				{ .weight = tg->scx.weight,
+				  .bw_period_us = tg->scx.bw_period_us,
+				  .bw_quota_us = tg->scx.bw_quota_us,
+				  .bw_burst_us = tg->scx.bw_burst_us };
 
 			ret = SCX_CALL_OP_RET(sch, SCX_KF_UNLOCKED, cgroup_init,
 					      NULL, tg->css.cgroup, &args);
···
 void scx_group_set_idle(struct task_group *tg, bool idle)
 {
 	/* TODO: Implement ops->cgroup_set_idle() */
+}
+
+void scx_group_set_bandwidth(struct task_group *tg,
+			     u64 period_us, u64 quota_us, u64 burst_us)
+{
+	struct scx_sched *sch = scx_root;
+
+	percpu_down_read(&scx_cgroup_rwsem);
+
+	if (scx_cgroup_enabled && SCX_HAS_OP(sch, cgroup_set_bandwidth) &&
+	    (tg->scx.bw_period_us != period_us ||
+	     tg->scx.bw_quota_us != quota_us ||
+	     tg->scx.bw_burst_us != burst_us))
+		SCX_CALL_OP(sch, SCX_KF_UNLOCKED, cgroup_set_bandwidth, NULL,
+			    tg_cgrp(tg), period_us, quota_us, burst_us);
+
+	tg->scx.bw_period_us = period_us;
+	tg->scx.bw_quota_us = quota_us;
+	tg->scx.bw_burst_us = burst_us;
+
+	percpu_up_read(&scx_cgroup_rwsem);
 }
 
 static void scx_cgroup_lock(void)
···
 	rcu_read_lock();
 	css_for_each_descendant_pre(css, &root_task_group.css) {
 		struct task_group *tg = css_tg(css);
-		struct scx_cgroup_init_args args = { .weight = tg->scx.weight };
+		struct scx_cgroup_init_args args = {
+			.weight = tg->scx.weight,
+			.bw_period_us = tg->scx.bw_period_us,
+			.bw_quota_us = tg->scx.bw_quota_us,
+			.bw_burst_us = tg->scx.bw_burst_us,
+		};
 
 		if ((tg->scx.flags &
 		     (SCX_TG_ONLINE | SCX_TG_INITED)) != SCX_TG_ONLINE)
···
 static void sched_ext_ops__cgroup_move(struct task_struct *p, struct cgroup *from, struct cgroup *to) {}
 static void sched_ext_ops__cgroup_cancel_move(struct task_struct *p, struct cgroup *from, struct cgroup *to) {}
 static void sched_ext_ops__cgroup_set_weight(struct cgroup *cgrp, u32 weight) {}
+static void sched_ext_ops__cgroup_set_bandwidth(struct cgroup *cgrp, u64 period_us, u64 quota_us, u64 burst_us) {}
 #endif
 static void sched_ext_ops__cpu_online(s32 cpu) {}
 static void sched_ext_ops__cpu_offline(s32 cpu) {}
···
 	.cgroup_move		= sched_ext_ops__cgroup_move,
 	.cgroup_cancel_move	= sched_ext_ops__cgroup_cancel_move,
 	.cgroup_set_weight	= sched_ext_ops__cgroup_set_weight,
+	.cgroup_set_bandwidth	= sched_ext_ops__cgroup_set_bandwidth,
 #endif
 	.cpu_online		= sched_ext_ops__cpu_online,
 	.cpu_offline		= sched_ext_ops__cpu_offline,

kernel/sched/ext.h

···
 void scx_cgroup_cancel_attach(struct cgroup_taskset *tset);
 void scx_group_set_weight(struct task_group *tg, unsigned long cgrp_weight);
 void scx_group_set_idle(struct task_group *tg, bool idle);
+void scx_group_set_bandwidth(struct task_group *tg, u64 period_us, u64 quota_us, u64 burst_us);
 #else	/* CONFIG_EXT_GROUP_SCHED */
 static inline void scx_tg_init(struct task_group *tg) {}
 static inline int scx_tg_online(struct task_group *tg) { return 0; }
···
 static inline void scx_cgroup_cancel_attach(struct cgroup_taskset *tset) {}
 static inline void scx_group_set_weight(struct task_group *tg, unsigned long cgrp_weight) {}
 static inline void scx_group_set_idle(struct task_group *tg, bool idle) {}
+static inline void scx_group_set_bandwidth(struct task_group *tg, u64 period_us, u64 quota_us, u64 burst_us) {}
 #endif	/* CONFIG_EXT_GROUP_SCHED */
 #endif	/* CONFIG_CGROUP_SCHED */

kernel/sched/sched.h

···
 
 extern struct list_head task_groups;
 
-#ifdef CONFIG_CFS_BANDWIDTH
+#ifdef CONFIG_GROUP_SCHED_BANDWIDTH
 extern const u64 max_bw_quota_period_us;
 
 /*
···
 {
 	return 100000ULL;
 }
-#endif /* CONFIG_CFS_BANDWIDTH */
+#endif /* CONFIG_GROUP_SCHED_BANDWIDTH */
 
 struct cfs_bandwidth {
 #ifdef CONFIG_CFS_BANDWIDTH

tools/sched_ext/scx_qmap.bpf.c

···
 		     taskc->force_local, taskc->core_sched_seq);
 }
 
+s32 BPF_STRUCT_OPS(qmap_cgroup_init, struct cgroup *cgrp, struct scx_cgroup_init_args *args)
+{
+	bpf_printk("CGRP INIT %llu weight=%u period=%lu quota=%ld burst=%lu",
+		   cgrp->kn->id, args->weight, args->bw_period_us,
+		   args->bw_quota_us, args->bw_burst_us);
+	return 0;
+}
+
+void BPF_STRUCT_OPS(qmap_cgroup_set_weight, struct cgroup *cgrp, u32 weight)
+{
+	bpf_printk("CGRP SET %llu weight=%u", cgrp->kn->id, weight);
+}
+
+void BPF_STRUCT_OPS(qmap_cgroup_set_bandwidth, struct cgroup *cgrp,
+		    u64 period_us, u64 quota_us, u64 burst_us)
+{
+	bpf_printk("CGRP SET %llu period=%lu quota=%ld burst=%lu", cgrp->kn->id,
+		   period_us, quota_us, burst_us);
+}
+
 /*
  * Print out the online and possible CPU map using bpf_printk() as a
  * demonstration of using the cpumask kfuncs and ops.cpu_on/offline().
···
 	       .dump			= (void *)qmap_dump,
 	       .dump_cpu		= (void *)qmap_dump_cpu,
 	       .dump_task		= (void *)qmap_dump_task,
+	       .cgroup_init		= (void *)qmap_cgroup_init,
+	       .cgroup_set_weight	= (void *)qmap_cgroup_set_weight,
+	       .cgroup_set_bandwidth	= (void *)qmap_cgroup_set_bandwidth,
 	       .cpu_online		= (void *)qmap_cpu_online,
 	       .cpu_offline		= (void *)qmap_cpu_offline,
 	       .init			= (void *)qmap_init,

tools/testing/selftests/sched_ext/maximal.bpf.c

···
 void BPF_STRUCT_OPS(maximal_cgroup_set_weight, struct cgroup *cgrp, u32 weight)
 {}
 
+void BPF_STRUCT_OPS(maximal_cgroup_set_bandwidth, struct cgroup *cgrp,
+		    u64 period_us, u64 quota_us, u64 burst_us)
+{}
+
 s32 BPF_STRUCT_OPS_SLEEPABLE(maximal_init)
 {
 	return scx_bpf_create_dsq(DSQ_ID, -1);
···
 	.cgroup_move		= (void *) maximal_cgroup_move,
 	.cgroup_cancel_move	= (void *) maximal_cgroup_cancel_move,
 	.cgroup_set_weight	= (void *) maximal_cgroup_set_weight,
+	.cgroup_set_bandwidth	= (void *) maximal_cgroup_set_bandwidth,
 	.init			= (void *) maximal_init,
 	.exit			= (void *) maximal_exit,
 	.name			= "maximal",