Re: [PATCH v6 3/6] clk: spacemit: Add clock support for SpacemiT K1 SoC

[Haylen Chu <heylenay@4d2.org>]

On Thu, Apr 10, 2025 at 01:55:49AM +0000, Yixun Lan wrote:
> Hi Inochi,
> 
> On 09:20 Thu 10 Apr     , Inochi Amaoto wrote:
> > On Wed, Apr 09, 2025 at 08:10:53PM -0500, Alex Elder wrote:
> > > On 4/9/25 7:57 PM, Inochi Amaoto wrote:
> > > > > > > > diff --git a/drivers/clk/spacemit/Kconfig b/drivers/clk/spacemit/Kconfig
> > > > > > > > new file mode 100644
> > > > > > > > index 000000000000..4c4df845b3cb
> > > > > > > > --- /dev/null
> > > > > > > > +++ b/drivers/clk/spacemit/Kconfig
> > > > > > > > @@ -0,0 +1,18 @@
> > > > > > > > +# SPDX-License-Identifier: GPL-2.0-only
> > > > > > > > +
> > > > > > > > +config SPACEMIT_CCU
> > > > > > > > +	tristate "Clock support for SpacemiT SoCs"
> > > > > > > I don't know the answer to this, but...  Should this be a Boolean
> > > > > > > rather than tristate?  Can a SpacemiT K1 SoC function without the
> > > > > > > clock driver built in to the kernel?
> > > > > > > 
> > > > > > I agree to make it a Boolean, we've already made pinctrl driver Boolean
> > > > > > and pinctrl depend on clk, besides, the SoC is unlikely functional
> > > > > > without clock built in as it's such critical..
> > > > > > 
> > > > > I disagree. The kernel is only for spacemit only, and the pinctrl
> > > > Sorry for a mistake, this first "only" should be "not".
> > > 
> > > This is a general problem.  You can't make a bootable
> > > SpacemiT kernel unless you define this as built-in (at
> > > least, that's what Yixun is saying). 
> > 
> > Why not putting the module in the initramfs? I have tested
> > this in quite a lot of boards (Allwinner, rockchip, sophgo,
> > starfive and etc.), all of them work well.

This is also my original consideration.

> it works, but not optimal, why delay clk initialzation at modules load stage?
> IMO, it brings more overhead for using initramfs..

For distribution maintainers and users, keeping stuff buildable as
modules shrinks the sizeof kernel image, which I'd like to see. Thus I
won't make the entry boolean.

> but there is always tradeoff and bikeshedding..
> 
> > > But we'd really rather *only* build it in to the kernel
> > > for SpacemiT builds. You clearly want to minimize what
> > > must be built in, but what if this is indeed required?
> > > What goes in defconfig?
> > > 
> > 
> > As defconfig is more like for a minimum example system. It
> > is OK to put a y in the defconfig. But for a custom system,
> > you do give a choice for the builder to remove your module
> > in non spacemit system.
> 
> I get your meaning here to remove/disable at run time stage, while
> we do provide compile time option, if don't want spacemit system
> just disable CONFIG_ARCH_SPACEMIT I mentioned, clk/pinctrl will be gone
> 
> anyway, I'm open for this, make it tristate do provide more choices,
> and it's probably better leave users to decide..
> 
> Ok, I'm fine with leave clk as tristate with proper default deconfig,
> but if people want to pursue more to make more driver(pinctrl) modulized
> feel free to test and send patches, I just won't put efforts myself.
> 
> thanks
> 
> -- 
> Yixun Lan (dlan)
> Gentoo Linux Developer
> GPG Key ID AABEFD55

Thanks,
Haylen Chu