[U-Boot] [PATCH 6/7] arm: ls102xa: Add SD boot support for LS1021AQDS board

[Albert ARIBAUD]

Hi Huan,

On Mon, 22 Sep 2014 06:46:20 +0000, Huan Wang
<alison.wang at freescale.com> wrote:

> Hi, Albert,
> 
> > > On Thu, 18 Sep 2014 13:47:18 +0800, Alison Wang <b18965 at freescale.com>
> > > wrote:
> > >
> > > > +
> > > > +     . = ALIGN(4);
> > > > +     .u_boot_list : {
> > > > +             KEEP(*(SORT(.u_boot_list*_i2c_*)));
> > > > +     }
> > >
> > > IS this required? And if it is, could it not be added to the
> > > arch/arm/cpu/u-boot-spl.lds file? This way you would not need an .lds
> > > file at all.
> > >
> > > [Alison Wang] Yes, it is required.
> > 
> > Ok -- what for? :)
> [Alison Wang] In SPL part, DDR is initialized by reading SPD through I2C interface.
> For I2C, ll_entry_count() is called, and it returns the number of elements of a
> linker-generated array placed into subsection of .u_boot_list section specified
> by _list argument. So I need to add this to make I2C work in SPL.

Understood. So your SPL code uses I2C, and for I2C, you need a linker
list. But then:

> > > I would like to add it in arch/arm/cpu/u-boot-spl.lds. I was not sure
> > > adding it in arch/arm/cpu/u-boot-spl.lds is acceptable or not.
> > 
> > (assuming the reason why it is needed is valid) If it causes no change
> > to boards which do not use it right now (and I mean 'no change' ad
> > 'binary identical') then this is acceptable. Make sure you check the
> > binary invariance and that you mention it in the commit.
> > 
> [Alison Wang] It will cause the binary is not identical for other board.

Is this a prediction or an actual observation of compared builds with
and without the I2C linker liste addition to the generic SPL .lds?

> I think
> it may be not good to add in arch/arm/cpu/u-boot-spl.lds. What's your opinion about it?

If there are SPLs which use I2C linker lists *and* the generic .lds and
build without an error, then this should be fixed, because it means the
build process should complain when an input section is not mapped to an
output section.

(and that should be fixed even though in this case, adding the I2C
linker lists in the .lds would 'fix' the build, but they would actually
paper over the real issue of sections being mapped without control)

> Best Regards,
> Alison Wang

Amicalement,
-- 
Albert.