Re: [PATCH 5/6] sched_ext: Report the ecaps loss when a cpu goes offline

[Tejun Heo <tj@kernel.org>]

Hello,

On Sat, Oct 17, 2026 at 06:14:02PM +0000, sashiko-bot@kernel.org wrote:
> The comment notes that scx_enabled() cannot change during hotplug, and that
> without a live root the ext class will not be picked and the wait would never
> end. However, static_branch_disable(&__scx_enabled) in scx_root_disable() runs
> before cpus_read_lock() is acquired.

static_branch_disable() takes cpus_read_lock() itself in
static_key_disable(), so the flip waits until the cpu down releases
cpus_write_lock(). scx_enabled() stays true across the down.

> If the offline CPU had not yet picked ext_sched_class before __scx_enabled was
> disabled and ext_server was detached (for instance, while a fair or RT task
> was running during ext_server's dl_defer window), __scx_process_sync_ecaps()
> will never be called on that rq.

The drain doesn't depend on the ext server. scx_offline_ecaps() runs
from sched_cpu_deactivate() after balance_push_set(), so once the
hotplug thread sleeps in the wait, balance_push() moves every task but
per-cpu kthreads and migration-disabled ones off the cpu. With
scx_enabled() still true, the cpu's first pick with no fair task left
runs the ext pick, whose dispatch_one() processes the marked sync.

Only a per-cpu kthread or a migration-disabled task that never sleeps
could hold it off after the ext server is detached, and that would be a
bug of its own.

Thanks.

--
tejun