Re: [RFC PATCH 0/9] initial support for "suniv" Allwinner new ARM9 SoC

[Rob Herring]

On Thu, Jan 25, 2018 at 04:35:20PM +0100, Maxime Ripard wrote:
> Hi,
> 
> On Wed, Jan 24, 2018 at 09:10:34PM +0800, Icenowy Zheng wrote:
> > å 2018å1æ22æææä CST äå8:14:35ïMaxime Ripard åéï
> > > On Sat, Jan 20, 2018 at 07:17:26AM +0800, Icenowy Zheng wrote:
> > > > This is the RFC initial patchset for the "new" Allwinner SUNIV ARM9 SoC.
> > > > 
> > > > The same die is packaged differently, come with different co-packaged
> > > > DRAM or shipped with different SDK; and then made many model names: F23,
> > > > F25, F1C100A, F1C100S, F1C200S, F1C500, F1C600, R6, etc. These SoCs all
> > > > share a common feature set and are packaged similarly (eLQFP128 for SoCs
> > > > without co-packaged DRAM, QFN88 for with DRAM). As their's no
> > > > functionality hidden on the QFN88 models (except DRAM interface not
> > > > exported), it's not clever to differentiate them. So I will use suniv as
> > > > common name of all these SoCs.
> > > 
> > > Where is that suniv prefix coming from?
> > 
> > The BSP (Melis and Linux). (e.g. "libs/suniv" directory of the Melis SDK and 
> > "arch/arm/boot/dts/sunivw1p1.dtsi" in the Linux SDK)
> 
> Do you have a link to that BSP?
> 
> > > You should really answer two questions here:
> > >   - Are you able to predict whether you'll find an SoC part of that
> > >     family in the future that derives a bit and will need a compatible
> > >     of its own?
> > >   - Are you able to predict which quirks we'll need along the way to
> > >     support all the SoCs you've listed there?
> > > 
> > > If you can't answer yes to both these questions, with a 100%
> > > certainty, then you'll need a SoC name in the compatible.
> > > 
> > > Which doesn't prevent you from sharing as much as possible the DT like
> > > we did between the A10s and the A13 for example.
> > 
> > So the suniv-f1c100s.dtsi will still be kept empty and all peripherals known 
> > should go through suniv.dtsi.
> 
> Sorry if I wasn't really clear. You can totally keep the current DT
> structure if that makes sense (and judging by what you're saying, it
> does.), but the compatibles should have the SoC name in it.

In case it's not clear, the compatible strings and any new bindings need 
to be documented.

Rob