Re: [PATCH v4 06/10] dt-bindings: mtd: update STM32 FMC2 NAND controller documentation

[Rob Herring]

On Thu, May 14, 2020 at 11:35 AM Christophe Kerello
<christophe.kerello@xxxxxx> wrote:
>
> Hi Rob,
>
> On 5/14/20 5:00 PM, Rob Herring wrote:
> > On Wed, May 06, 2020 at 11:11:15AM +0200, Christophe Kerello wrote:
> >> These bindings can be used on SOCs where the FMC2 NAND controller is
> >> in standalone. In case that the FMC2 embeds 2 controllers (an external
> >> bus controller and a raw NAND controller), the register base and the
> >> clock will be defined in the parent node. It is the reason why the
> >> register base address and the clock are now optional.
> >>
> >> Signed-off-by: Christophe Kerello <christophe.kerello@xxxxxx>
> >> ---
> >>   .../devicetree/bindings/mtd/st,stm32-fmc2-nand.yaml   | 19 ++++++++++---------
> >>   1 file changed, 10 insertions(+), 9 deletions(-)
> >>
> >> diff --git a/Documentation/devicetree/bindings/mtd/st,stm32-fmc2-nand.yaml b/Documentation/devicetree/bindings/mtd/st,stm32-fmc2-nand.yaml
> >> index b059267..68fac1a 100644
> >> --- a/Documentation/devicetree/bindings/mtd/st,stm32-fmc2-nand.yaml
> >> +++ b/Documentation/devicetree/bindings/mtd/st,stm32-fmc2-nand.yaml
> >> @@ -18,13 +18,15 @@ properties:
> >>
> >>     reg:
> >>       items:
> >> -      - description: Registers
> >> +      - description: Registers (optional)
> >
> > The only thing that can be optional are the last entries. You have to do
> > a 'oneOf' with 6 entries and 7 entries.
>
> Ok, so the way to describe the reg property in my case should be:
>         reg:
>           oneOf:
>             - description: FMC2 embeds the NFC controller in standalone.
>               items:
>                 - description: Registers
>                 - description: Chip select 0 data
>                 - description: Chip select 0 command
>                 - description: Chip select 0 address space
>                 - description: Chip select 1 data
>                 - description: Chip select 1 command
>                 - description: Chip select 1 address space
>
>             - description: FMC2 embeds the NFC controller and the EBI
>                 controller.
>               items:
>                 - description: Chip select 0 data
>                 - description: Chip select 0 command
>                 - description: Chip select 0 address space
>                 - description: Chip select 1 data
>                 - description: Chip select 1 command
>                 - description: Chip select 1 address space
>
> >
> > And where's your new compatible string for this different h/w?
>
>  From NFC controller point of view, it is the same HW.

That's what everyone says until they have some quirk or integration
difference to handle.

> In the case that we have 2 controllers embedded, the register base is
> shared.
> The NFC driver will check at probe time the compatible string of its
> parent node.
> In case that it is "st,stm32mp1-fmc2-ebi", then the driver will find the
> register base in the parent node (EBI node), otherwise it will find it
> in the NFC node.
> Is it better to have 2 compatible strings (one for each reg description)
> than checking the parent's compatible string and have only one
> compatible string?

Why not just put the register base into the child node too? While
overlapping 'reg' regions for siblings is bad, it's fine for child
nodes. I guess since there are chip selects for the child nodes that
may not work here.

It doesn't hurt to have another compatible. You can always make the
old one a fallback. With different compatibles you can make sure reg
has the right number of entries.

Rob