Re: [PATCH v5 tip/core/rcu 4/9] completion: Replace spin_unlock_wait() with lock/unlock pair

From: Paul E. McKenney
Date: Thu Aug 17 2017 - 10:13:19 EST


On Thu, Aug 17, 2017 at 02:49:09PM +0200, Ingo Molnar wrote:
>
> * Paul E. McKenney <paulmck@xxxxxxxxxxxxxxxxxx> wrote:
>
> > > this change - or can I pick this up into the scheduler tree?
> >
> > Timely question! ;-)
> >
> > My current plan is to send you a pull request like the following later
> > today, Pacific Time (but rebased adding Steve Rostedt's Reviewed-by).
> > This patch is on one of the branches, currently v4.13-rc2..93d8d7a12090
> > ("arch: Remove spin_unlock_wait() arch-specific definitions") in my
> > -rcu tree.
> >
> > Ah, and v4.13-rc2..7391304c4959 ("membarrier: Expedited private command")
> > is mostly outside of RCU as well.
> >
> > Since I will be rebasing and remerging anyway, if you would prefer that I
> > split the spin_unlock_wait() and/or misc branches out, I am happy to do so.
> > If I don't hear otherwise, though, I will send all seven branches using
> > my usual approach.
> >
> > So, if you want something different than my usual approach, please just
> > let me know!
>
> No, all branches together sounds good to me!

Very good, will do!

> If you are rebasing anyway, here are some (very minor) commit title nits I noticed:
>
> > swait: add idle variants which don't contribute to load average
> > rcu: use idle versions of swait to make idle-hack clear
>
> Capitalization.

Will fix! Believe it or not, I looked for these... :-/

> > membarrier: Expedited private command
>
> Should start with a verb.

OK, something like "Provide expedited private command".

> > doc: RCU documentation update
>
> doc: Update RCU documentation
>
> ?

Works for me!

> > doc: No longer allowed to use rcu_dereference on non-pointers
>
> doc: Describe that it is no longer allowed to use rcu_dereference() on non-pointers
>
> ?

Will add a real commit log.

> > torture: Add --kconfig argument to kvm.sh
> > rcutorture: Don't wait for kernel when all builds fail
>
> Is there a difference between 'torture: ' and 'rcutorture: ' prefixes?

Yes, rcutorture is specific to RCU, while torture would also affect
locktorture.

Ah, and if I am delaying the cond_resched() patch, I need to retest,
which means I will send you the pull request tomorrow or Monday, depending
on how the testing goes.

Thanx, Paul

> Thanks,
>
> Ingo
>