Re: [PATCH 1/4] arm: parse PSCI node from the host device-tree

[Andre Przywara <andre.przywara@linaro.org>]

On 11/26/2013 12:12 PM, Ian Campbell wrote:
> On Mon, 2013-11-25 at 13:02 +0100, Andre Przywara wrote:
>> +
>> +    if ( !dt_property_read_u32(psci, "cpu_on", &psci_host_cpu_on_nr) )
>> +    {
>> +        printk("/psci node is missing the \"cpu_on\" property\n");
>> +        return 4;
>> +    }
>
> http://www.spinics.net/lists/devicetree/msg05348.html updates the
> bindings for PSCI 0.2. I suppose Midway must only support 0.1?
>
> I'd be OK with only supporting 0.1 right now, but it would be useful to
> comment either in the code or in the commit message.
>
> (nb: I'm not sure of v4 was the final version of that series)

If I look at http://www.spinics.net/lists/devicetree/msg12293.html
I would keep it as it is - at least until the binding changes. Hopefully 
there will be a compatibility fallback if cpu_on should change to 
cpu_on-{32,64}. I will talk to Rob about this and will later send a fix 
if that is needed.
Midway provides an implementation conforming to PSCI 0.2, however the 
device tree binding is still the one from the current kernel git HEAD.

>
>> +
>> +    return 0;
>> +}
>> +
>>   /* Parse the device tree and build the logical map array containing
>>    * MPIDR values related to logical cpus
>>    * Code base on Linux arch/arm/kernel/devtree.c
>> @@ -107,6 +145,11 @@ void __init smp_init_cpus(void)
>>       bool_t bootcpu_valid = 0;
>>       int rc;
>>
>> +    if ( psci_host_init() == 0 )
>> +    {
>> +        printk(XENLOG_INFO "Using PSCI for SMP bringup\n");
>> +    }
>> +
>>       if ( (rc = arch_smp_init()) < 0 )
>
> arch_smp_init is empty on both platforms. arm32 has a comment "TODO:
> PSCI" ;-)
>
> I think we can nuke this function while we are here, since it's only
> purpose was as a PSCI placehoder.
>

But arch_smp_init() calls platform_smp_init(), which is not empty for 
the Exynos, VExpress and OMAP5 (it contains the native SMP bringup for 
these platforms).
So shall I skip the superfluous arch_smp_init() and call 
platform_smp_init() directly from smpboot.c?
Or keep it as it is and we clean it up later?

Thanks,
Andre.