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

[Tejun Heo]

Hello,

On Tue, Aug 18, 2026 at 02:09:41AM +0900, Changwoo Min wrote:
> +/*
> + * 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) {}

__used only prevents the compiler from dropping the function. With
CONFIG_LD_DEAD_CODE_DATA_ELIMINATION, the linker can still garbage
collect its section, so please add __retain too (cf. __bpf_kfunc).

Also, we'll likely accumulate more of these markers over time, so let's
give them a shared prefix - scx_compat_marker_ or something like that -
and collect them in one place, maybe towards the end of ext.c next to
the module init code.

Thanks.

--
tejun