[U-Boot] [PATCH 1/2] gpio: Add DW APB GPIO driver

[Simon Glass]

Hi Marek,

On 10 August 2015 at 09:01, Marek Vasut <marex at denx.de> wrote:
>
> On Friday, August 07, 2015 at 10:37:54 PM, Simon Glass wrote:
> > Hi Marek,
>
> Hi Simon,
>
> > On 7 August 2015 at 14:35, Marek Vasut <marex at denx.de> wrote:
> > > On Friday, August 07, 2015 at 09:13:45 PM, Simon Glass wrote:
> > >> Hi Marek,
> > >
> > > Hi!
> > >
> > >> On 5 August 2015 at 19:49, Marek Vasut <marex at denx.de> wrote:
> > >> > On Wednesday, August 05, 2015 at 04:39:33 PM, Simon Glass wrote:
> > >> >> Hi Marek,
> > >> >
> > >> > Hi Simon,
> > >
> > > [...]
> > >
> > >> >> It's up to you. Normally each bank has a name and the datasheet
> > >> >> specifies it. In your case if not you could think about a naming
> > >> >> scheme.
> > >> >
> > >> > Can you please take a look into arch/arm/dts/socfpga.dtsi ?
> > >> > The system has three GPIO controllers (look for gpio0, gpio1, gpio2)
> > >> > and each of these controllers has one bank (porta, portb, portc) .
> > >> >
> > >> > I can name my gpios portxN , where x is either of a,b,c and N is the
> > >> > GPIO number. The problem is, I cannot determine in dwapb_gpio_bind()
> > >> > which one is "porta", "portb" and "portc" because all I have is the
> > >> > physical addess of the GPIO controller and the index of the bank in
> > >> > the namespace of that controller.
> > >> >
> > >> > Sure, I can do some sort of global counting in the driver, but I would
> > >> > like to avoid that sort of thing. I can also add some kind of ad-hoc
> > >> > DT prop, but that's also not a good idea I think. Do you have any
> > >> > suggestion for me please ?
> > >>
> > >> One option is to use the device tree node name but it isn't very
> > >> friendly - gpio0 at xxxxx.
> > >
> > > That's what I do now pretty much.
> > >
> > >> You could perhaps add a new property like 'bank-name'?
> > >
> > > Do we want to add ad-hoc DT nodes which are
> > > a) Not describing hardware
> > > b) Not part of the official DT bindings for that platform
> > > ?
> > >
> > > Is that really a way to go ?
> > >
> > > [...]
> >
> > It needs to be part of the official binding. Naming the hardware is
> > part of the hardware definition - see for example the regulator-name
> > property for regulators.
>
> So what do you think about introducing a 'bank-name' property then ?
> I think this might work just fine ?

I think it's OK - just make sure you send the GPIO binding change to Linux too.

>
> > Another option is to use an alias:
> >
> > aliases {
> >    gpio0 = &gpio_0;
> >    gpio1 = &gpio_1;
> >    gpio2 = &gpio_2;
> > }
> >
> > Then you can turn gpio0 into bank A, gpio1 into bank B, etc.
>
> Is there a function which maps the udevice->dev->of_offset into an alias's
> seq ID ?

dev->seq, once it is probed. Until it is probed it doesn't have a
sequence number (by definition).

You can use fdtdec_get_alias_seq() but I don't recommend it. We're
trying to drop fdtdec eventually.

Regards,
Simon