Re: [PATCH v3 2/3] x86: devicetree: enable multiprocessing in DT

From: Thomas Gleixner
Date: Fri Mar 02 2018 - 05:44:33 EST


On Thu, 1 Mar 2018, Ivan Gorinov wrote:

The patch order is wrong. Fixes first and then features simply because you
cannot test the feature patch standalone as the fix (3/3) is missing ....

> Adding code to register the processors described in Device Tree.

'Adding code...' is a pointless filler phrase.

> APIC ID is specified in 'intel-apic_id' propery as used in U-Boot.

s/propery/property/

Also this has nothing to do with U-Boot. The DT properties are described in
the DT bindings and not specified in U-Boot.

Where is the matching devicetree binding documented? This should be a
separate patch and needs to go into Documentation/devicetree/x86/

Please Cc the DT maintainers as we need their ack for that.

> First address specified in 'reg' is used as default APIC ID.

Changelogs should describe the context/problem and the approach how this is
fixed or made working. Something like this:

The current x86 device tree implementation does not support SMP.

Use the new DT bindings for describing CPUs and their APIC resources.

Hmm?

> Signed-off-by: Ivan Gorinov <ivan.gorinov@xxxxxxxxx>
> ---
> arch/x86/kernel/devicetree.c | 26 ++++++++++++++++++++++++--
> 1 file changed, 24 insertions(+), 2 deletions(-)
>
> diff --git a/arch/x86/kernel/devicetree.c b/arch/x86/kernel/devicetree.c
> index 44189ee..ef1cd85 100644
> --- a/arch/x86/kernel/devicetree.c
> +++ b/arch/x86/kernel/devicetree.c
> @@ -130,6 +130,29 @@ static void __init dtb_setup_hpet(void)
> #endif
> }
>
> +static void __init dtb_cpu_setup(void)
> +{
> + struct device_node *dn;
> + struct resource r;
> + const void *prop;
> + int apic_id, version;
> + int ret;
> +
> + version = GET_APIC_VERSION(apic_read(APIC_LVR));
> + for_each_node_by_type(dn, "cpu") {
> + prop = of_get_property(dn, "intel,apic-id", NULL);
> + if (prop) {
> + apic_id = be32_to_cpup(prop);
> + } else {
> + ret = of_address_to_resource(dn, 0, &r);
> + if (WARN_ON_ONCE(ret))
> + continue;
> + apic_id = r.start;
> + }

What kind of logic is this? If a CPU node does not have apic id property
then it's invalid. That else clause is just voodoo programming at least
without a proper comment explaining it.

Thanks,

tglx