On Fri, Mar 29, 2013 at 2:57 AM, Peter Zijlstra <pet...@infradead.org> wrote: > On Thu, 2013-03-28 at 19:50 -0700, Michel Lespinasse wrote: >> So, there are a few things I don't like about spin_unlock_wait(): >> >> 1- From a lock ordering point of view, it is strictly equivalent to >> taking the lock and then releasing it - and yet, lockdep won't catch >> any deadlocks that involve spin_unlock_wait. (Not your fault here, >> this should be fixed as a separate change in lockdep. I manually >> looked at the lock ordering here and found it safe). > > Ooh, I never noticed that, but indeed this shouldn't be hard to cure. > >> 2- With the current ticket lock implementation, a stream of lockers >> can starve spin_unlock_wait() forever. Once again, not your fault and >> I suspect this could be fixed - I expect spin_unlock_wait() callers >> actually only want to know that the lock has been passed on, not that >> it actually got to an unlocked state. > > I suppose the question is do we want to fix it or have both semantics > and use lock+unlock where appropriate.
We'd have to look at the users to be sure, but I strongly expect they don't need to get in line waiting - it's sufficient to just wait for the head of the queue to move (or for the queue to be empty). There are actually very few users - Just drivers/ata/libata-eh.c for the spin_unlock_wait() function, and a couple more (kernel/task_work.c and kernel/exit.c) for the raw_spin_unlock_wait variant. Guess I'm not the only one to dislike that function :) -- Michel "Walken" Lespinasse A program is never fully debugged until the last user dies. -- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
