[U-Boot] [PATCH v2 11/12] RFC: Use binman for a sunxi board

[Simon Glass]

Hi Tom,

On 1 October 2016 at 18:46, Tom Rini <trini at konsulko.com> wrote:
> On Sat, Oct 01, 2016 at 06:29:42PM -0600, Simon Glass wrote:
>> Hi Tom,
>>
>> On 1 October 2016 at 18:15, Tom Rini <trini at konsulko.com> wrote:
>> > On Wed, Sep 28, 2016 at 09:46:25AM -0600, Simon Glass wrote:
>> >> Hi Tom,
>> >>
>> >> On 27 September 2016 at 19:55, Tom Rini <trini at konsulko.com> wrote:
>> >> > On Sun, Sep 25, 2016 at 03:52:27PM -0600, Simon Glass wrote:
>> >> >
>> >> >> Add an example usage of binman for a sunxi board. This involves adding the
>> >> >> image definition to the device tree and using it in the Makefile.
>> >> >>
>> >> >> This is for example only.
>> >> >>
>> >> >> Signed-off-by: Simon Glass <sjg at chromium.org>
>> >> >> ---
>> >> >>
>> >> >> Changes in v2: None
>> >> >>
>> >> >>  Makefile                            |  4 +---
>> >> >>  arch/arm/dts/sun7i-a20-pcduino3.dts | 12 ++++++++++++
>> >> >
>> >> > I think this shows the big problem with using binman today.  For the
>> >> > common case of ARM, where we sync in the dts* files from upstream, this
>> >> > will add hunks that must not be overwritten each time.
>> >> >
>> >> > Looking at scripts/Makefile.lib::cmd_fdt I wonder if we couldn't come up
>> >> > with some wildcard rule and check if, somewhere CONFIG'd ? $(BOARDDIR)/
>> >> > ? u-boot.dtsi exists add in -include that/file.dtsi to the CPP rule so
>> >> > that we can keep the parts that will never get upstream separate.
>> >>
>> >> We can do that, but I have found that most boards with the same SoC
>> >> are the same, or similar. So for x86 [1] I put it in a separate patch
>> >> with just an #include in the .dts file.
>> >>
>> >> We could have binman be a bit smarter about where it looks - e.g. if
>> >> there is no binman node, it could look in the same directory for a
>> >> file that matches the board name, or part of it?
>> >
>> > I'd really like to try and better solve the generic problem we have tho
>> > too while we're at it.  ie the u-boot,dm-pre-reloc tag on various nodes
>> > could also go into this file.
>>
>> What sort of solution are you thinking of? A U-Boot .dtsi include that
>> is #included at the top of all files?
>
> Something like:
> ifneq ($(wildcard $(srctree)/arch/$(CONFIG_SYS_ARCH)/cpu/$(CONFIG_SYS_CPU)/$(CONFIG_SYS_SOC)/u-boot.dtsi),)
> dtc_cpp_flags += -include
> $(srctree)/arch/$(CONFIG_SYS_ARCH)/cpu/$(CONFIG_SYS_CPU)/$(CONFIG_SYS_SOC)/u-boot.dtsi
> endif
>
> And maybe a few other wildcards, I'm not sure, so that for everything
> rockchip related, ie what binman needs, what nodes need to be pre-reloc,
> etc, can end up in that one file.  And we use -include to get it so that
> arch/arm/dts/ can be unmodified from upstream.

OK, I was worried that's what you might mean :-)

I'll take a look.

Regards.
Simon