Re: [PATCH v4 sched_ext/for-7.3 24/40] sched_ext: Add sub_ecaps_updated() effective-cap change notifier

On Wed, 08 Jul 2026 22:02:19 +0000, [email protected] wrote:
> - [High] `scx_discard_ecaps_to_sync()` loops infinitely if the target
> CPU is starved by RT/DL tasks, stalling scheduler teardown.

The loop only spins while the target cpu never runs balance_one(), and
the ext class has DL-server forward progress, so it gets cpu time and
drains the node even under saturating RT/DL load. A class receiving zero
runtime forever (RT throttling fully disabled plus a spinning FIFO task)
is a pre-existing system-level failure that breaks far more than this
teardown path. No change.

Thanks.

-- 
tejun