Re: [PATCH 5/5] xen: RCU: avoid busy waiting until the end of grace period.

[Dario Faggioli <dario.faggioli@citrix.com>]

[-- Attachment #1.1: Type: text/plain, Size: 2384 bytes --]

On Wed, 2017-08-09 at 19:34 +0200, Dario Faggioli wrote:
> On Mon, 2017-08-07 at 02:54 -0600, Jan Beulich wrote:
> > > > > Dario Faggioli <dario.faggioli@citrix.com> 07/27/17 10:01 AM
> > > +/*
> > > + * Timer for making sure the CPU where a callback is queued does
> > > + * periodically poke rcu_pedning(), so that it will invoke the
> > > callback
> > > + * not too late after the end of the grace period.
> > > + */
> > > +void rcu_idle_timer_start()
> > > +{
> > > +    struct rcu_data *rdp = &this_cpu(rcu_data);
> > > +
> > > +    if (likely(!rdp->curlist))
> > > +        return;
> > 
> > I would have expected this to be the inverse of the original
> > condition in
> > rcu_needs_cpu() - why is there no rcu_pending() invocation here?
> > 
> 
> [...]
>
> Actually, it's entirely possible that it is having rcu_pending(cpu)
> in
> rcu_needs_cpu() is, for us, redundant. In fact, although it does make
> sense in Linux, both code inspection and some investigation I've just
> done, makes me believe that there won't be cases where a CPU is
> denied
> going offline because it sees rcu_pending() returning 1.
> 
> In fact, when we call rcu_pending(), within cpu_is_haltable(), we
> have
> already gone through it before. And if there were pending work, we've
> raised the softirq and dealt with it. If there weren't, neither there
> is now.
> 
> I'm therefore leaning toward removing rcu_pending() from the
> rcu_needs_cpu() check as well. At that point, we'll indeed have the
> check inside rcu_start_idle_timer(), be the opposite of the original
> check in rcu_needs_cpu(). :-)
> 
FTR, I'm not so sure of this last thing any longer. I mean, the
analysis I provided is still correct, but I'm investigating the other
possible race existing in the code that Tim has hinted at in his mail,
and I think it could be useful to have rcu_pending() checked in here,
to solve/avoid that one.

It's also possible that I'll actually remove it from rcu_needs_cpu(),
but to move it somewhere else... As I said, I'm still looking into the
problem.

Regards,
Dario
-- 
<<This happens because I choose it to happen!>> (Raistlin Majere)
-----------------------------------------------------------------
Dario Faggioli, Ph.D, http://about.me/dario.faggioli
Senior Software Engineer, Citrix Systems R&D Ltd., Cambridge (UK)

[-- Attachment #1.2: This is a digitally signed message part --]
[-- Type: application/pgp-signature, Size: 819 bytes --]

[-- Attachment #2: Type: text/plain, Size: 127 bytes --]

_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xen.org
https://lists.xen.org/xen-devel