Re: [PATCH v1] kthread/smpboot: Serialize kthread parking against wakeup

[Peter Zijlstra]

On Tue, May 01, 2018 at 12:18:45PM +0200, Peter Zijlstra wrote:
> Aaaah... I think I've spotted a problem there. We clear SHOULD_PARK
> before we rebind, so if the thread lost the first PARKED store,
> does the completion, gets migrated, cycles through the loop and now
> observes !SHOULD_PARK and bails the wait-loop, then __kthread_bind()
> will forever wait.
> 
> Is that what you had in mind?

Another possible problem is concurrent thread_park(), if both observe
!IS_PARKED, we'll end up with 2 threads waiting on the completion, but
we only do a single complete().

Of course, this might not be a suppored use of the API, and I don't
think this will ever actually happen. But the whole !test_bit(IS_PARKED)
thing seems to suggest it is fine calling this on an already parked
thread.

Confusing stuff that should be cleared up in any case.