RE: FW: avoiding run_workqueue() recursion

["Anirban Sinha" <ASinha@zeugmasystems.com>]

Hi Oleg:

>If we just return silently, we do not flush but hide the problem ?
>And in this can lead to other problems which are very hard to
>trigger/debug.

True. I think flushing is an invalid operation for a thread that is
already walking the work-queue, like keventd. It is inherently bug in
the code somewhere else (may be in a work function?). I liked your idea
of replacing WARN_ON() with BUG_ON() but I do understand that a panic
could be a bigger hammer here. May be we can have some sort of
restrictions or conventions for writing work functions?  I don't know. 

Ani