Re: [PATCH RFC 2/2] soc: Add a basic ACPI generic driver

[John Garry]

On 28/01/2020 12:50, Arnd Bergmann wrote:

Hi Arnd,

> On Tue, Jan 28, 2020 at 12:18 PM John Garry <john.garry@xxxxxxxxxx> wrote:
> > Add a generic driver for platforms which populate their ACPI PPTT
> > processor package ID Type Structure according to suggestion in the ACPI
> > spec - see ACPI 6.2, section 5.2.29.3 ID structure Type 2.
> >
> > The soc_id is from member LEVEL_2_ID.
> >
> > For this, we need to use a whitelist of platforms which are known to
> > populate the structure as suggested.
> >
> > For now, only the vendor and soc_id fields are exposed.
> >
> > Signed-off-by: John Garry <john.garry@xxxxxxxxxx>
>
> Would it be possible to make this the root device for all on-chip devices
> to correctly reflect the hierarchy inside of the soc?

I don't think so. The information about the SoC is got from the PPTT, 
which only describes processors, caches, and physical package 
boundaries. It doesn't include references to on-chip devices.

Having said that (and unrelated to this series), we could add 
/sys/devices/system/soc folder, similar to node folder.

> > +/*
> > + * Known platforms that fill in PPTT package ID structures according to
> > + * ACPI spec examples, that being:
> > + * - Custom driver attribute is in ID Type Structure VENDOR_ID member
> > + * - SoC id is in ID Type Structure LEVEL_2_ID member
> > + *    See ACPI SPEC 6.2 Table 5-154 for PPTT ID Type Structure
> > + */
> > +static struct acpi_platform_list plat_list[] = {
> > +       {"HISI  ", "HIP08   ", 0, ACPI_SIG_PPTT, all_versions},
> > +       { } /* End */
> > +};
>
> That matches a single machine, right? It doesn't seem very generic
> that way.

Yes :) The problem is that the PPTT ID structure is open to use how the 
implementer wants, so we can't assume everything/anything implemented 
according to the spec examples. Maybe we could call it type1 or 
something like that for platforms which did use the convention in the 
spec example.

> > +struct acpi_generic_soc_struct {
> > +       struct soc_device_attribute dev_attr;
> > +       u32 vendor;
> > +};
> > +
> > +static ssize_t vendor_show(struct device *dev,
> > +                          struct device_attribute *attr,
> > +                          char *buf)
> > +{
> > +       struct acpi_generic_soc_struct *soc = dev_get_drvdata(dev);
> > +       u8 vendor_id[5] = {};
> > +
> > +       *(u32 *)vendor_id = soc->vendor;
> > +
> > +       return sprintf(buf, "%s\n", vendor_id);
> > +}
>
> I'd rather not see nonstandard attributes in a "generic" driver at
> all. Maybe the
> you can simply concatenate the vendor and  LEVEL_2_ID into a single string
> here?

I actually don't really require the vendor attr. And since "vendor" is 
not in the set of standard soc driver attrs, it can just be omitted.

> > +                       soc = kzalloc(sizeof(*soc), GFP_KERNEL);
> > +                       if (!soc)
> > +                               return -ENOMEM;
> > +
> > +                       soc_dev_attr = &soc->dev_attr;
> > +                       soc_dev_attr->soc_id = kasprintf(GFP_KERNEL, "%s",
> > +                                                        soc_id);
>
> On the other hand, it would make sense to fill out additional fields here.
> You have already matched the name of the board from the
> acpi_platform_list, so there are two strings available that could be put
> into the "machine" field, and it would make sense to fill out "family" with
> something that identifies it as coming from ACPI PPTT data.

OK, maybe the ones you suggested could be added. I did just want to 
start out with a minimal sets of files, especially since we don't always 
have a direct mapping between soc driver attrs and this PPTT ID structure.

Thanks,
John