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

[Simon Glass]

Hi Tom,

On 2 October 2016 at 21:26, Simon Glass <sjg at chromium.org> wrote:
> 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.

I've sent v3 with something along these lines. I could not use
-include because I need to avoid having the .dts version header appear
twice. So it has a little 'sed' in it...

Regards,
Simon