[PATCH v5 3/3] xen/arm: introduce xen_early_init, use PSCI on xen

[Stefano Stabellini]

On Tue, 2 Apr 2013, Nicolas Pitre wrote:
> On Tue, 2 Apr 2013, Stefano Stabellini wrote:
> 
> > Split xen_guest_init in two functions, one of them (xen_early_init) is
> > going to be called very early from setup_arch.
> > 
> > Change machine_desc->smp_init to xen_smp_init if Xen is present on the
> > platform. xen_smp_init just sets smp_ops to psci_smp_ops.
> > 
> > Introduce a dependency for ARM_PSCI in XEN.
> 
> The Kconfig stuff should be more understandable to "normal" users 
> configuring the kernel.  Hence it might make more sense for Xen to 
> select PSCI rather than making it a prerequisite.

You are right, I'll do that.


> [...]
> > @@ -176,27 +178,30 @@ static int __init xen_secondary_init(unsigned int cpu)
> >  	return 0;
> >  }
> >  
> > +static void __init xen_smp_init(void)
> > +{
> > +	if (psci_smp_available())
> > +		smp_set_ops(&psci_smp_ops);
> > +}
> > +
> 
> I still don't understand why you need to do this.  Why can't you just 
> rely on the next priority which is to set PSCI ops by default if 
> available?  Using this hook for Xen doesn't look justified. As it is, 
> you break MCPM.
> 
> As I explained to you, MCPM will end up using PSCI as well when 
> available, which I think should be sufficient for your concern.

The smp_init hook is not limited to MCPM or the versatile express
platform. It's a generic hook that can be used by any platform and can
override the platform smp_ops or the psci_smp_ops depending on platform
specific configurations.

Configurations that I am pretty sure won't be available on Xen anyway,
while I am certain that using psci_smp_ops would work.

It seems to me that relying on the fact that only versatile express and
only MCPM use smp_init, even though it might work today, it's very
fragile and could break tomorrow without any of us noticing.