Re: [PATCH][resubmit] x86: enable preemption in delay

["Gregory Haskins" <ghaskins@novell.com>]

>>> On Wed, Jun 18, 2008 at  8:42 AM, in message
<200806182242.49245.nickpiggin@yahoo.com.au>, Nick Piggin
<nickpiggin@yahoo.com.au> wrote: 
> On Wednesday 18 June 2008 22:25, Gregory Haskins wrote:
>> >>> On Wed, Jun 18, 2008 at  8:16 AM, in message
> 
>> > Yeah - migrate_disable() has been proposed several times. The reason I
>> > don't like it is that is creates scheduling artefacts like latencies by
>> > not being able to load-balance (and thereby complicates all that, and
>> > you know we don't need more complication there).
>>
>> True, and good point.  But this concept would certainly be useful to avoid
>> the heavyweight (w.r.t. latency) preempt-disable() in quite a few different
>> areas, so if we can make it work with reasonable visibility, it might be
>> nice to have.
> 
> It just seems like pretty worthless bloat to me.
> 
> There are _some_ cases where it can be used, but nobody has been
> able to come up with compelling uses really.

Well, I have some ongoing R&D which (I believe) *would* make compelling use of a
migration-disable feature.  But to date it is not ready to see the light of day.  As 
far as existing uses, I think I mostly agree with you.  The one argument to the
contrary would be to clean up the handful of places that implement an ad-hoc
migration-disable by messing with cpus_allowed in a similar manner.  But perhaps
those could be solved with a preempt-disable() as well.

>I don't think this
> case is helped very much either because the logic in there using
> preempt-disable is fine, isn't it?

You are probably right.  In my own defense, I was just responding to the
question about manipulating the cpus_allowed.  If you are going to do that
you are better off with my patch (IMO).  Changing cpus_allowed to prevent
migration is racy, among other things.  Whether this tsc code is optimal with
migration disabled or preemption-disabled is a separate matter which
I did not address.

> 
> Except that it should also have a cond_resched in it. Seems like
> an ideal place to put cond_resched because it is not a fastpath.

Seems reasonable to me.  Thanks Nick.

-Greg