> mmmhhh... still a bit skeptical about waiting 100 times 10us in
> atomic.

Has it been discussed already why the non-atomic version of
read_poll_timeout is not enough?