svn commit: r305368 - head/sys/kern

[John Baldwin]

On Monday, September 05, 2016 11:09:17 AM Mark Johnston wrote:
> On Mon, Sep 05, 2016 at 10:30:24AM -0700, John Baldwin wrote:
> > On Sunday, September 04, 2016 12:29:49 AM Mark Johnston wrote:
> > > Author: markj
> > > Date: Sun Sep  4 00:29:48 2016
> > > New Revision: 305368
> > > URL: https://svnweb.freebsd.org/changeset/base/305368
> > > 
> > > Log:
> > >   Micro-optimize sleepq_signal().
> > >   
> > >   Lift a comparison out of the loop that finds the highest-priority thread
> > >   on the queue.
> > >   
> > >   MFC after:	1 week
> > 
> > Could this safely use TAILQ_FOREACH_FROM?
> 
> Are you suggesting something like this?
> 
> besttd = TAILQ_FIRST(&sq->sq_blocked[queue]);
> td = TAILQ_NEXT(besttd, td_slpq);
> TAILQ_FOREACH_FROM(td, &sq->sq_blocked[queue], td_slpq) {
> ...
> 
> I think that would work, and it avoids visiting the first element
> unnecessarily when the queue contains more than one element. If the
> queue contains one element, we'd visit it because of
> TAILQ_FOREACH_FROM's surprising behaviour of iterating over the entire
> queue when the listelem is NULL.

I was hoping it was something equivalent to:

	for (td = TAILQ_NEXT(besttd); td != NULL; td = TAILQ_NEXT(td)) {
		...
	}

I guess what we actually want is something like TAILQ_FOREACH_AFTER():

	TAILQ_FOREACH_AFTER(td, besttd, &sq->sq_blocked[queue], td_slpq)

that assumed bestttd was not NULL and so avoided the whole "scan the
whole thing".

For now you could perhaps open-code the above loop though?

-- 
John Baldwin