Suppose count = 0 with a writer finishing down_write_failed_biased and two
readers on sem->wait. If you just do wake_up(&sem->wait) you'll wake up
both and at most one of them can bias the lock. So it appears to me that we
need to wake up just one process at that point. wake_up, the way it is now,
doesn't allow this since both readers are non-exclusive, but this is the
desired behavior, no?
> At this point the lock is unbiased, and this will wake up one or more
> readers (since they're not exclusive) and one writer (assuming lots of
> contention, which is the only case for which stampedes are interesting).
> Ideally, only 1 process would be woken up here to bias the lock, but we
> can't since readers *must* be added non-exclusively.
But this is an artifact of the use of wake_up(). If the desired behavior is
to wake up only one process, be it reader or writer, we should use something
that provides this behavior. It doesn't have to be wake_up.
My contention is that you want to wake up just one process in
down_write_failed_biased, but all the readers + one writer in
rwsem_wake_readers. Is this incorrect?
-- Dimitris Michailidis dimitris@engr.sgi.com- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.rutgers.edu Please read the FAQ at http://www.tux.org/lkml/