Re: [PATCH v2 2/2] firmware: coreboot: Collapse platform drivers into bus core

[Stephen Boyd <swboyd@chromium.org>]

Quoting Julius Werner (2018-08-08 12:07:30)
> > +config GOOGLE_COREBOOT_TABLE_ACPI
> > +       tristate
> > +       default GOOGLE_COREBOOT_TABLE
> 
> I don't think this helps in upgrading (as your commit message says)
> unless you also keep the 'select GOOGLE_COREBOOT_TABLE' here, right?

Oh yes should be select, not default.

> 
> > -int coreboot_table_init(struct device *dev, void __iomem *ptr)
> > +static int coreboot_table_init(struct device *dev, void __iomem *ptr)
> 
> nit: There's little reason to keep coreboot_table_init() a separate
> function now. Could maybe compact the code a little more if you merge
> it into probe()? (Also could then do the signature sanity check before
> trusting the length values to map the whole thing, which is probably a
> good idea.)

Sure. I can make another patch for squashing that all together.

> 
> >         if (ptr_header) {
> >                 bus_unregister(&coreboot_bus_type);
> >                 iounmap(ptr_header);
> 
> Could ptr_header be handled by devm now, somehow?

Yes. It hasn't been devmified yet because that would be more things in
one big patch. This is quickly blowing up!

> Also, don't you have
> two bus_unregister() now (here and in coreboot_exit())? Or is that
> intentional?

That's nice. I didn't notice that module_init() was registering the bus
and then platform drivers could remove the bus later with the driver
unbind. I'll move them both into the driver bind/unbind path, in another
patch.

> 
> > +static struct platform_driver coreboot_table_driver = {
> > +       .probe = coreboot_table_probe,
> > +       .remove = coreboot_table_remove,
> > +       .driver = {
> > +               .name = "coreboot_table",
> > +               .acpi_match_table = ACPI_PTR(cros_coreboot_acpi_match),
> > +               .of_match_table = of_match_ptr(coreboot_of_match),
> 
> Who takes precedence if they both exist? Will we have two
> coreboot_table busses? (That would probably not be so good...)

I'm not aware of a system that has both ACPI and devicetree, so this
isn't a problem.