On Wed, Jan 24, 2024 at 09:15:07AM +0800, Kever Yang wrote:
> Hi Tom,
> 
> On 2024/1/24 01:58, Tom Rini wrote:
> > On Tue, Jan 23, 2024 at 11:53:18AM +0800, Kever Yang wrote:
> > 
> > > Hi Tom,
> > > 
> > > rk3036 only has small size internal sram, so it can only use very
> > > simple SPL which only including dram init, and no SPL_FRAMEWORK
> > > support.
> > > In this convert, the CONFIG_SPL_STACK is lost for this SoC, any
> > > suggestions on this kind of issue?
> > Does this work?
> > 
> > diff --git a/common/spl/Kconfig b/common/spl/Kconfig
> > index e7b84fc1fa68..9deee0fdbea8 100644
> > --- a/common/spl/Kconfig
> > +++ b/common/spl/Kconfig
> > @@ -372,7 +372,8 @@ config SPL_SHARES_INIT_SP_ADDR
> >   config SPL_STACK
> >   	hex "Initial stack pointer location"
> > -	depends on (ARM || ARCH_JZ47XX || MICROBLAZE || RISCV) && SPL_FRAMEWORK
> > +	depends on (ARM || ARCH_JZ47XX || MICROBLAZE || RISCV) && \
> > +		SPL_FRAMEWORK || ROCKCHIP_RK3036
> This works with adding CONFIG_SPL_STACK in rk3036 board defconfig.
> >   	depends on !SPL_SHARES_INIT_SP_ADDR
> >   	default 0x946bb8 if ARCH_MX7
> >   	default 0x93ffb8 if ARCH_MX6 && MX6_OCRAM_256KB
> > 
> > You will have to update the ROCKCHIP_RK3036 platforms otherwise the
> > build will hang now.
> 
> What do you mean to update the platform?
> 
> The rk3036 case should be one of the available option for those "small"/low
> end SoCs has very limit
> 
> size of internal dram, only need minimal SPL+ U-Boot,  and no requirement
> for ATF/OPTEEwhich will have to use TPL+SPL+U-Boot.
> 
> 
> The CONFIG_SPL_STACK is used in assemble init code, it doesn't need to
> depends on SPL_FRAMEWORK, isn't it?

I just mean that with the above Kconfig change, all RK3036 platforms
must set a value for CONFIG_SPL_STACK at the same time, or they will
fail to build (and CI will get hung). It sounds like all RK3036
platforms need to set the value in order to work, so this isn't a bug,
just a feature. Is that clearer?

-- 
Tom