[U-Boot] [PATCH 3/5] arm: Allow lr to be saved by board code

[Albert ARIBAUD <albert.u.boot@aribaud.net>]

Hello Simon,

On Wed, 4 Feb 2015 19:51:36 -0700, Simon Glass <sjg@chromium.org> wrote:
> Hi Albert,
> 
> On 4 February 2015 at 01:59, Albert ARIBAUD <albert.u.boot@aribaud.net> wrote:
> > Hello Simon,
> >
> > On Tue,  3 Feb 2015 21:18:52 -0700, Simon Glass <sjg@chromium.org>
> > wrote:
> >> The link register value can be required on some boards (e.g. FEL mode on
> >> sunxi) so stash it to r4 before jumping to save_boot_params().
> >
> > I'd rather not make this there; I would like save_boot_params to start
> > with as unmodified a core state as possible, because r4 might
> > actually, now or in the future, contain important data.
> >
> > I'd even go as far as saying that save_boot_params should be branched
> > into rather than called, and should branch back to a fixed label right
> > after reset, so that even lr is preserved. For SoCs which do not need
> > to save any boot parameters, a weak version of save_boot_params would
> > just do the branching back.
> >
> > (this should help merging all start.S instances into a singe one,
> > which I still want to achieve one day)
> 
> I did consider this but was worried that some save_boot_params()
> functions are written in C (OMAP) so this would not work. But if we
> can require it to be written in assembler, or provide an asm macro to
> return from the C function, then I can do that. What do you think?

I think that save_boot_params runs right at reset and cannot rely on any
C environment at all, and should indeed be written in asm and only deal
with *very urgent* savings; only those registers that would be lost
almost immediately, like r0-r14 and some system registers. Anything
else is less urgent and can be processed later, possibly even in
board_init_f().

That being said, OMAP does not have save_boot_params in C that I can
find; actualy, doing a 'git grep save_boot_param -- *.c' shows only
one occurrence, in arch/arm/cpu/armv7/exynos/spl_boot.c, ar line 312,
and it's an empty function. OMAP has an s_init C implementation,
though.

> Regards,
> Simon

Amicalement,
-- 
Albert.