RE: [RFC] scmi: pinctrl: support i.MX9

[Peng Fan <peng.fan@xxxxxxx>]

> Subject: Re: [RFC] scmi: pinctrl: support i.MX9
> 
> On Thu, Aug 24, 2023 at 2:47 PM Peng Fan <peng.fan@xxxxxxx> wrote:
> > Me:
> 
> >> it is merely making things more complex and also slower
> > > bymaking the registers only accessible from this SCMI link.
> >
> > This is for safety reason, the pinctrl hardware must be handled by a
> > system manager entity. So mmio direct access not allowed from Cortex-A
> > side.
> 
> Yeah I understood as much. But I don't think that the firmware is really
> filtering any of the access, it will just poke into any pinctrl register as
> instructed anyway so what's the point. Just looks like a layer of indirection.

No, the firmware has a check on whether a pin is allowed to be configured
by the agent that wanna to configure the pin. 

> But I'm not your system manager, so it's not my decision.
> 
> > The SCMI firmware is very straightforward, there is no group or
> > function.
> >
> > It just accepts the format as this:
> > MUX_TYPE, MUX VALUE, CONF_TYPE, CONF_VAL, DAISY_TYPE, DAISY ID,
> > DAISY_CFG, DAISY_VALUE.
> >
> > Similar as linux MMIO format.
> >
> > Our i.MX95 platform will support two settings, one with SCMI firmware,
> > one without SCMI. These two settings will share the same pinctrl
> > header file.
> >
> > And to simplify the scmi firmware design(anyway I am not owner of the
> > firmware), to make pinctrl header shared w/o scmi, we take the current
> > in-upstream freescale imx binding format.
> 
> The SCMI people will have to state their position on this.
> Like what they consider conformance and what extensions are allowed. This
> is more a standardization question than an implementation question so it's
> not really my turf.

The i.MX95 SCMI firmware uses OEM extension type. So I just follow
what the firmware did and support it in linux. Anyway let's
wait Sudeep's reply.

Thanks,
Peng.

> 
> I was under the impression that the ambition with SCMI firmware was to
> abstract away and hide aspects of the hardware behind a consistent API.
> This approach drives a truck through that idea.
> 
> Yours,
> Linus Walleij