Re: [PATCH v6 1/4] arm: introduce psci_smp_ops

[Stefano Stabellini <stefano.stabellini@eu.citrix.com>]

On Thu, 18 Apr 2013, Russell King - ARM Linux wrote:
> On Fri, Apr 05, 2013 at 02:11:32PM +0100, Stefano Stabellini wrote:
> > +	psci_init();
> >  #ifdef CONFIG_SMP
> >  	if (is_smp()) {
> > -		smp_set_ops(mdesc->smp);
> > +		if (mdesc->smp)
> > +			smp_set_ops(mdesc->smp);
> > +		else if (psci_smp_available())
> > +			smp_set_ops(&psci_smp_ops);
> 
> So, I have a vague recollection that the ordering of the above got discussed
> but I can't find it amongst the 21k of messages so far this year.
> 
> The above looks weird to me.  Surely this should be:
> 
> 		if (psci_smp_available())
> 			smp_set_ops(&psci_smp_ops);
> 		else if (mdesc->smp)
> 			smp_set_ops(mdesc->ops);
> 
> This means that if PSCI is available, and provides a set of operations,
> we override whatever the platform has statically provided.
> 
> Remember, we're trying to move away from using "mdesc"s for platform
> stuff, relying on things like DT and such like.  We really should not
> be going for mdesc-overriding-newstuff but newstuff-overriding-mdesc.

That's correct, in fact if you look at the next patch you'll see that it
changes the order.

I introduced the mechanism first and changed the priority later - it
should help bisectability.
I can fold the two patches into one if you prefer.


> Now, if the psci stuff can't be relied upon to provide the correct
> functionality, then that's a separate problem which needs addressing
> differently.
> 
> This should allow the Xen problem to be resolved, because Xen will
> provide the PSCI operations, and it's correct in that case to override
> the platform's SMP operations.

Yes, increasing the priority of PSCI helps Xen a lot.
In order to completely solve the issue for Xen though, another patch is
needed (http://marc.info/?l=linux-kernel&m=136630106201968&w=2) because
of the introduction of smp_init.