Hi Wolfram,
Thanks for reminding me I still had to chime in on this ;-)
On Thu, Jan 26, 2023 at 10:03 AM Wolfram Sang
<wsa+renesas@xxxxxxxxxxxxxxxxxxxx> wrote:
> > > I need it because of ".revision". This only applies to "ES1.*",
> > > there are "ES2.*" and "ES3.*" around which have the same SoC number.
> > > Also, there is usually no version numbering for the IP core. We need to
> > > use this scheme in a number of other places already, sadly.
> >
> > I did not get whether this is runtime characteristics or it can be
> > customized with compatible (just you did not do it)?
>
> We have compatibles per SoC, i.e. "r8a7795". We don't have compatibles
> for ES versions, i.e. no "r8a7795-es10" or "r8a7795-es20".
>
> The latter would not be practical. We can't know in advance how many ES
> revisions there will be, so we can't prepare DTs accordingly. Updating
> later would be also difficult because we are usually not notified if
> there is a new ES version. Only if there are problems with it. And which
> board is available with which ES version is chaotic^2.
>
> Also, if we update DTs later, old DTBs would not work with newer kernels
> (requiring a later added compaible for a new ES version). This all still
> ignores that it would be a churn to update for every ES version of every
> SoC. We have quite many to support. That's why we use soc_device_match()
> for ES versions in many places alreday. It was never a problem so far.
>
> That's my reasoning, probably Geert has something to add. He maintains
> the Renesas DT files.
Exactly. We only use soc_device_match() to distinguish where we do not
have a compatible value to do so. As we have SoC-specific compatible
values for about everything, this means we usually use soc_device_match()
only to handle quirks on specific revisions of SoCs.
Gr{oetje,eeting}s,
Geert
--
Geert Uytterhoeven -- There's lots of Linux beyond ia32 -- geert@xxxxxxxxxxxxxx
In personal conversations with technical people, I call myself a hacker. But
when I'm talking to journalists I just say "programmer" or something like that.
-- Linus Torvalds
