Re: [PATCH v2 5/8] soc: sifive: ccache: Add StarFive JH7110 support

[Conor Dooley <conor.dooley@microchip.com>]

On Tue, Nov 22, 2022 at 10:54:34AM +0100, Emil Renner Berthing wrote:
> On Tue, 22 Nov 2022 at 10:03, Hal Feng <hal.feng@starfivetech.com> wrote:
> > On Fri, 18 Nov 2022 19:45:57 +0800, Conor Dooley wrote:
> > > Hey Emil/Hal,
> > > On Fri, Nov 18, 2022 at 09:17:11AM +0800, Hal Feng wrote:
> > > > From: Emil Renner Berthing <kernel@esmil.dk>

> > > > diff --git a/arch/riscv/Kconfig.socs b/arch/riscv/Kconfig.socs
> > > > index 69774bb362d6..5a40e05f8cab 100644
> > > > --- a/arch/riscv/Kconfig.socs
> > > > +++ b/arch/riscv/Kconfig.socs
> > > > @@ -22,6 +22,7 @@ config SOC_STARFIVE
> > > >     bool "StarFive SoCs"
> > > >     select PINCTRL
> > > >     select RESET_CONTROLLER
> > > > +   select SIFIVE_CCACHE
> > >
> > > Please no. I am trying to get rid of these selects + I cannot figure out
> > > why this driver is so important that you *need* to select it. Surely the
> > > SoC is useable without it>
> > > Is this a hang over from your vendor tree that uses the driver to do
> > > non-coherent stuff for the jh7100?
> >
> > I have tested that the board can successfully boot up without the cache
> > driver. The `select` can be removed for JH7110. @Emil, what do you think
> > of this?
> 
> Yes, for the JH7110 this is not strictly needed, just like the
> Unmatched board. For the StarFive JH7100 it is though.
> So if you're only adding support for the JH7110 then it's not needed.

Even for the JH7100 there are other ways to do this than selects in
arch/riscv - for example
config SIFIVE_CCACHE
	default SOC_STARFIVE

But you don't need that either if you're not adding the JH7100 :)

> > > >  config SIFIVE_CCACHE
> > > >     bool "Sifive Composable Cache controller"
> > > > diff --git a/drivers/soc/sifive/sifive_ccache.c b/drivers/soc/sifive/sifive_ccache.c
> > > > index 1c171150e878..9489d1a90fbc 100644
> > > > --- a/drivers/soc/sifive/sifive_ccache.c
> > > > +++ b/drivers/soc/sifive/sifive_ccache.c
> > > > @@ -107,6 +107,7 @@ static const struct of_device_id sifive_ccache_ids[] = {
> > > >     { .compatible = "sifive,fu540-c000-ccache" },
> > > >     { .compatible = "sifive,fu740-c000-ccache" },
> > > >     { .compatible = "sifive,ccache0" },
> > > > +   { .compatible = "starfive,jh7110-ccache" },
> > >
> > > Per my second reply to the previous patch, I am not sure why you do not
> > > just have a fallback compatible in the binding/dt for the fu740 ccache
> > > since you appear to have identical configuration?
> >
> > Yeah, I will use the compatible of fu740 and modify this patch.
> 
> No, the JH7110 should not pretend to be a fu740, but if you add
> 
> compatible = "starfive,jh7110-ccache", "sifive,ccache0";
> 
> then this driver should still match "sifive,ccache0" without adding
> the "starfive,jh7110-ccache" entry.

Either works for me :) If you go for "sifive,ccache0", just make sure to
add the correct property enforcement - you can just copy the fu740 by
the looks of things (although that'd imply that it is compatible and can
fall back to it...)

Thanks,
Conor.