[PATCH 0/3] sched/fair: Fix cgroup::idle preemption

[Peter Zijlstra]

Hi!

so I finally got around to looking at this again with a fresh mind and an LLM
generated testcase. It appears to actually work now.

I'm still not convinced on the whole find_matching_se() thing; at the end of
the day we mark a subtree 'idle', which is akin to saying 'dont care', so then
actually caring about how it ends up preempting on a different level is weird
to me.

As such, I have Vincent's patch last; modified to use the new idle state as
optimization.