[PATCH] sched_ext: allow ops.cgroup_set_bandwidth() to be sleepable

[Changwoo Min]

ops.cgroup_set_bandwidth() is delivered from scx_group_set_bandwidth(),
which runs from the cpu.max cgroup interface write path (tg_set_bandwidth())
in process context. tg_set_cfs_bandwidth() has already returned by then, so
its cpus_read_lock and cfs_constraints_mutex are released, and the only lock
held is percpu_down_read(&scx_cgroup_ops_rwsem), whose read side may sleep.
The call site is therefore sleepable, like ops.cgroup_init().

bpf_scx_check_member() rejects a sleepable program on any member not on its
allow-list, so a BPF scheduler cannot allocate -- which is sleepable -- when
a cgroup gains a cpu.max limit at runtime; it must instead pre-reserve memory
for a callback that cannot allocate. Add cgroup_set_bandwidth() to the
allow-list so the callback can allocate on demand, and document that it may
block.

A scheduler must decide at load time whether to mark the callback sleepable,
but the allow-list entry is a verifier property with no symbol to probe. Add
scx_cgroup_set_bandwidth_may_sleep() as a marker whose presence in the
kernel's BTF lets userspace detect this support; it has no callers and does
nothing.

Signed-off-by: Changwoo Min <changwoo@xxxxxxxxxx>
---
 kernel/sched/ext/ext.c      | 10 ++++++++++
 kernel/sched/ext/ext.h      |  1 +
 kernel/sched/ext/internal.h |  2 +-
 3 files changed, 12 insertions(+), 1 deletion(-)

diff --git a/kernel/sched/ext/ext.c b/kernel/sched/ext/ext.c
index 10af28a9f2c0..e25a2e9f4eca 100644
--- a/kernel/sched/ext/ext.c
+++ b/kernel/sched/ext/ext.c
@@ -4960,6 +4960,15 @@ void scx_group_set_bandwidth(struct task_group *tg,
 
 	percpu_up_read(&scx_cgroup_ops_rwsem);
 }
+
+/*
+ * Capability marker for userspace. The sleepable allowance for
+ * ops.cgroup_set_bandwidth() (see bpf_scx_check_member()) is a verifier
+ * property with no other symbol a scheduler can probe, so this no-op function
+ * exists solely so its presence in the kernel's BTF can be detected. It has no
+ * callers; __used keeps it from being optimized away.
+ */
+__used void scx_cgroup_set_bandwidth_may_sleep(void) {}
 #endif	/* CONFIG_EXT_GROUP_SCHED */
 
 #if defined(CONFIG_EXT_GROUP_SCHED) || defined(CONFIG_EXT_SUB_SCHED)
@@ -8079,6 +8088,7 @@ static int bpf_scx_check_member(const struct btf_type *t,
 	case offsetof(struct sched_ext_ops, cgroup_init):
 	case offsetof(struct sched_ext_ops, cgroup_exit):
 	case offsetof(struct sched_ext_ops, cgroup_prep_move):
+	case offsetof(struct sched_ext_ops, cgroup_set_bandwidth):
 #endif
 	case offsetof(struct sched_ext_ops, cpu_online):
 	case offsetof(struct sched_ext_ops, cpu_offline):
diff --git a/kernel/sched/ext/ext.h b/kernel/sched/ext/ext.h
index 0b7fc46aee08..e6fcfadc25aa 100644
--- a/kernel/sched/ext/ext.h
+++ b/kernel/sched/ext/ext.h
@@ -81,6 +81,7 @@ 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);
+void scx_cgroup_set_bandwidth_may_sleep(void);
 #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; }
diff --git a/kernel/sched/ext/internal.h b/kernel/sched/ext/internal.h
index 27bbf5e04d90..e2d553d13c49 100644
--- a/kernel/sched/ext/internal.h
+++ b/kernel/sched/ext/internal.h
@@ -753,7 +753,7 @@ struct sched_ext_ops {
 	 * @burst_us: bandwidth control burst
 	 *
 	 * Update @cgrp's bandwidth control parameters. This is from the cpu.max
-	 * cgroup interface.
+	 * cgroup interface. This operation may block.
 	 *
 	 * @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
-- 
2.55.0