Re: [RFC PATCH 1/3] dt-bindings: net: sparx5: Add sparx5-switch bindings

[Steen Hegelund]

On 30.11.2020 15:05, Andrew Lunn wrote:
> EXTERNAL EMAIL: Do not click links or open attachments unless you know the content is safe
>
> On Mon, Nov 30, 2020 at 02:09:34PM +0100, Steen Hegelund wrote:
> > On 27.11.2020 18:00, Andrew Lunn wrote:
> > > EXTERNAL EMAIL: Do not click links or open attachments unless you know the content is safe
> > >
> > > > +  reg-names:
> > > > +    minItems: 153
> > > > +    items:
> > > > +      - const: dev2g5_0
> > > > +      - const: dev5g_0
> > > > +      - const: pcs5g_br_0
> > > > +      - const: dev2g5_1
> > > > +      - const: dev5g_1
> > > ...
> > > > +      - const: ana_ac
> > > > +      - const: vop
> > >
> > > > +    switch: switch@600000000 {
> > > > +      compatible = "microchip,sparx5-switch";
> > > > +      reg = <0x10004000 0x4000>, /* dev2g5_0 */
> > > > +        <0x10008000 0x4000>, /* dev5g_0 */
> > > > +        <0x1000c000 0x4000>, /* pcs5g_br_0 */
> > > > +        <0x10010000 0x4000>, /* dev2g5_1 */
> > > > +        <0x10014000 0x4000>, /* dev5g_1 */
> > >
> > > ...
> > >
> > > > +        <0x11800000 0x100000>, /* ana_l2 */
> > > > +        <0x11900000 0x100000>, /* ana_ac */
> > > > +        <0x11a00000 0x100000>; /* vop */
> > >
> > > This is a pretty unusual binding.
> > >
> > > Why is it not
> > >
> > > reg = <0x10004000 0x1af8000>
> > >
> > > and the driver can then break up the memory into its sub ranges?
> > >
> > >    Andrew
> > Hi Andrew,
> >
> > Since the targets used by the driver is not always in the natural
> > address order (e.g. the dev2g5_x targets), I thought it best to let the DT
> > take care of this since this cannot be probed.  I am aware that this causes
> > extra mappings compared to the one-range strategy, but this layout seems more
> > transparent to me, also when mapped over PCIe.
>
> The question is, do you have a device tree usage for this? Are there
> devices in the family which have the regions in a different order?

Hi Andrew,

Yes we do have more chips in the pipeline that we expect to be able to
use this driver.  But I see your point.  The new chips most certainly will
have a different number of targets (that will most likely map to different
addresses).

The bindings will need to be able to cope with the different number of
targets, and I think that will be difficult.

So all in all I think I will rework this, and make the mapping in the
driver instead.

> You can easily move this table into the driver, and let the driver
> break the region up. That would be normal.
>
>      Andrew

BR
Steen

---------------------------------------
Steen Hegelund
steen.hegelund@xxxxxxxxxxxxx