[U-Boot] [PATCH] ARM: SPL: FIT: fix DTC warnings on FIT generation

[Simon Glass]

Hi Andre,

On 16 October 2017 at 23:30, André Przywara <andre.przywara at arm.com> wrote:
> Hi,
>
> sorry Simon for dropping the ball earlier. I will try to answer both
> Jagan's and your concern below.
>
> On 16/10/17 21:59, Jagan Teki wrote:
>> On Mon, Oct 9, 2017 at 10:15 AM, Simon Glass <sjg at chromium.org> wrote:
>>> Hi Andre,
>>>
>>> On 4 October 2017 at 17:24, Andre Przywara <andre.przywara at arm.com> wrote:
>>>> Newer versions of the device tree compiler (rightfully) complain about
>>>> mismatches between attributed node names (name@<addr>) and a missing
>>>> reg property in that node.
>>>> Adjust the FIT build script for 64-bit Allwinner boards to remove the
>>>> bogus addresses from the node names and avoid the warnings.
>>>>
>>>> Signed-off-by: Andre Przywara <andre.przywara at arm.com>
>>>> ---
>>>>  board/sunxi/mksunxi_fit_atf.sh | 16 ++++++++--------
>>>>  1 file changed, 8 insertions(+), 8 deletions(-)
>>>
>>> It looks like we have this problem all over the place. The
>>> documentation in doc/uImage now seems to have this problem too.
>>>
>>> I wonder if instead we should add reg / #address-cells / #size-cells properties?
>>
>> If the update on dts, might be an another-overhead to maintain u-boot
>> dts wrt Linux dts sync.
>
> This is not the kernel .dts, but the FIT image description (using the
> DTS format), which is purely private to U-Boot. I erroneously used
> addresses (fdt at 1) to enumerate different FDTs.
> I don't think this is right, since those FDTs don't have anything which
> would resemble an address. Instead the SPL just chooses one of them, and
> the script generates as many as we give it (from defconfig).
>
> So I don't see much sense in introducing a "reg" property. Multiple
> instances of an UART are alive at the same time, so an address property
> makes sense. But we just need *one* FDT and some unique name used to
> match configuration entries to the appropriate image. Actually those
> identifiers could be totally random as well, or we actually use
> something derived from the filename.
> But for simplicity I'd just go with the underscore notation, unless
> someone convinces me otherwise.

OTOH this really just a feature of the DT format. Adding a 'reg'
property can be justified on that basis. Or perhaps we should have an
option for dtc to support 'degenerate' .dts files?

The underscore is normally used for phandles.

Regards,
Simon