Messages in this thread | | | Date | Mon, 19 Dec 2005 17:51:34 +0100 | From | Ingo Molnar <> | Subject | Re: [patch 05/15] Generic Mutex Subsystem, mutex-core.patch |
| |
* Steven Rostedt <rostedt@goodmis.org> wrote:
> How expensive is the xchg? Since __mutex_lock_common is called even > when it's going to wake up. Maybe it might be more efficient to add > something like: > > if (atomic_cmpxchg(&lock->count, 1, 0) { > debug_set_owner(lock, ti __IP__); > debug_unlock_irqrestore(&debug_lock, *flags, ti); > return 1; > } > > This way we save the overhead of grabbing another spinlock, adding the > task to the wait_list and changing it's state.
in the first pass we definitely need to add ourselves to the list first - hence have to grab the lock. Even after the schedule(), we have to xchg it to -1, not 0. This is crutial to 'not drop the ball' property of one-waiter-in-flight logic - we must not lose the -1 'there are more waiters pending' property. Plus, we have the grab the lock because we remove ourselves from the wait-list after the schedule(). So i'm not sure your suggested optimization is possible.
Ingo - To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
| |