RE: [PATCH V2 3/9] clk: imx: Support building SCU clock driver as module

[Aisheng Dong <aisheng.dong@nxp.com>]

> From: Stephen Boyd <sboyd@kernel.org>
> Sent: Tuesday, June 23, 2020 4:34 PM
> Subject: RE: [PATCH V2 3/9] clk: imx: Support building SCU clock driver as
> module
> 
> Quoting Aisheng Dong (2020-06-22 20:42:19)
> > > From: Stephen Boyd <sboyd@kernel.org>
> > > Sent: Saturday, June 20, 2020 11:28 AM
> > > Subject: RE: [PATCH V2 3/9] clk: imx: Support building SCU clock
> > > driver as module
> > >
> > > Quoting Aisheng Dong (2020-06-17 18:58:51)
> > > > > From: Anson Huang <anson.huang@nxp.com>
> > > > > > > +obj-$(CONFIG_MXC_CLK_SCU) += mxc-clk-scu.o
> > > > > >
> > > > > > Like i.MX pinctrl, I'm not sure if it's really necessary to
> > > > > > build core libraries as modules. Probably the simplest way is
> > > > > > only building platform drivers part as module. And leave those
> > > > > > core libraries
> > > built in kernel.
> > > > > > This may make the code a bit cleaner.
> > > > > >
> > > > >
> > > > > Will discuss this with Linaro guys about it, previous
> > > > > requirement I received is all SoC specific modules need to be built as
> module.
> > > > >
> > > >
> > > > Okay. AFAIK it's not conflict.
> > > > You still make drivers into modules.
> > > > Only difference is for those common libraries part, we don't
> > > > convert them into module Which is less meaningless.
> > > >
> > >
> > > What is the benefit of making the core part of the SoC driver not a module?
> >
> > Usually we could try to build it as module if it's not hard.
> >
> > One question is sometimes those core part are shared with some platforms
> which can't built as module.
> > For i.MX case, it's mainly patch 4:
> > [V2,4/9] clk: imx: Support building i.MX common clock driver as module
> >
> >
> > Those libraries are also used by i.MX6&7 which can't build as module.
> > So we need an extra workaround patch to forcely 'select' it under
> > arch/arm/mach-imx/Kconfig [V2,2/9] ARM: imx: Select MXC_CLK for
> > ARCH_MXC
> > Then the users can't configure it as module in order to not break build.
> >
> > If build-in those common libraries, the implementation could be a bit easier
> and cleaner.
> > So I'm not sure if we still have to build them as module.
> > How would you suggest for such case?
> 
> Stop using 'select MXC_CLK' when requiring the core library code?
> Instead, make it a 'depends' and then that will make depending modules (i.e. the
> SoC files) that want to be builtin force the core module to be builtin too. Other
> modular configs that depend on the core will still be modular.
> 

It seems not work.
Patch 4 already changes it to depend on ARCH_MXC which can only be 'Y'.
[V2,4/9] clk: imx: Support building i.MX common clock driver as module
diff --git a/drivers/clk/imx/Kconfig b/drivers/clk/imx/Kconfig
index ded0643..678113b 100644
--- a/drivers/clk/imx/Kconfig
+++ b/drivers/clk/imx/Kconfig
@@ -1,8 +1,8 @@ 
 # SPDX-License-Identifier: GPL-2.0
 # common clock support for NXP i.MX SoC family.
 config MXC_CLK
-	bool
-	def_bool ARCH_MXC
+	tristate "IMX clock"
+	depends on ARCH_MXC

But user can still set MXC_CLK to be m, either via make menuconfig or defconfig.

Looks like only select it in arch Kconfig in Patch 2, there will be no issue.
diff --git a/arch/arm/mach-imx/Kconfig b/arch/arm/mach-imx/Kconfig
index e7d7b90..47b10d2 100644
--- a/arch/arm/mach-imx/Kconfig
+++ b/arch/arm/mach-imx/Kconfig
@@ -4,6 +4,7 @@  menuconfig ARCH_MXC
 	depends on ARCH_MULTI_V4_V5 || ARCH_MULTI_V6_V7 || ARM_SINGLE_ARMV7M
 	select ARCH_SUPPORTS_BIG_ENDIAN
 	select CLKSRC_IMX_GPT
+	select MXC_CLK
 	select GENERIC_IRQ_CHIP
 	select GPIOLIB
 	select PINCTRL

> I don't know why an architecture is selecting the clk code at all to be honest.
> That can be moved to the defconfig instead of in the architecture Kconfig and
> then you don't get a working system unless you select the MXC_CLK config from
> the configurator tool (menuconfig, nconfig, etc.) So ARCH_MXC shouldn't be in
> this discussion and the core module should be selectable by the configurator
> and that should be tristate and all SoC modules should depend on that core
> library module and be selectable too and those Kconfigs can be tristate or bool.

I guess it is because we didn't have requirements to make minimum booting required
drivers to be built as module before.

Regards
Aisheng