Re: [PATCH v2 06/10] soc: Add SoC specific driver support for nuc900

[Wan Zongshun]

On 2016å07æ11æ 16:03, Arnd Bergmann wrote:
> On Sunday, July 10, 2016 3:27:26 PM CEST Wan Zongshun wrote:
> > +       ret = of_property_read_string(np, "compatible", &soc_dev_attr->soc_id);
> > +       if (ret)
> >                return -EINVAL;
> > +
> > +       soc_dev_attr->machine = "NUC900EVB";
> > +       soc_dev_attr->family = "NUC900";
> > +       soc_dev = soc_device_register(soc_dev_attr);
> > +       if (IS_ERR(soc_dev)) {
> > +               kfree(soc_dev_attr);
> > +               return -ENODEV;
> > +       }
> > +
> > +       ret = regmap_read(syscon_regmap, GCR_CHIPID, &nuc900_chipid);
> > +       if (ret)
> > +               return -ENODEV;
> > +
> > +       device_create_file(soc_device_to_device(soc_dev), &nuc900_chipid_attr);
> > +       device_create_file(soc_device_to_device(soc_dev), &nuc900_version_attr);
> > +
> > +       dev_info(&pdev->dev, "Nuvoton Chip ID: 0x%x, Version ID:0x%x\n",
> > +                nuc900_chipid & GCR_CHIPID_MASK,
> > +                (nuc900_chipid >> 24) & 0xff);
>
> I'm still a bit unsure about the set of attributes here.
>
> - The "soc_id" is read from the device tree from the field that contains
>    the board name, I think for consistency you should try to map the
>    GCR_CHIPID to the name of the SoC and assign that here

I will try to get chipid and map it to soc name like: ânuc970â, "nuc910".

And I will set this soc name to soc_id, ok?

> - The "machine" is hardcoded to "NUC900EVB", which in turn looks like
>    a particular board but not the one you are running on. Maybe read
>    that from the DT instead?

Should I read nuc970-evb.dts's "model" or "compatible" properties?

> - The "revision" is not filled at all, I would suggest using something
>    derived from the GCR_CHIPID register here
>
> - you have two nonstandard attributes "chipid" and "version", which
>    I'd hope to avoid -- the set of standard attributes is supposed to
>    give enough information about the machine, and platform independent
>    user space will never read those.
>
> 	Arnd