Re: [patch 1/4] prepare_to_wait/finish_wait sleep/wakeup API

[Linus Torvalds]

On Wed, 25 Sep 2002, Andrew Morton wrote:
> 
> I expect a decent win would come from using this technique in
> select/poll, but that code relies on the remains-on-the-waitqueue
> semantics, and would need some fiddling.

Actually, I think that select/poll is exactly the kind of thing that would 
_not_ improve noticeably, since usually it's only one (out of possibly 
hundreds) or wait-queues that gets woken up. 

So even if that one were to be made faster, the _real_ cost when it comes
to the wait-queues will be all the other 99 waitqueues that select/poll
has to remove itself from the old way anyway.

			Linus