Hi Ian,
On 26 July 2016 at 07:35, Ian Campbell ijc@hellion.org.uk wrote:
(adding Hans)
On Mon, 2016-07-25 at 18:59 -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.
My concern with this approach is that the *.dts(i) files used for sunxi are sync'd unmodified from the upstream repo. I think this is a valuable property to retain in terms of maintenance burden for us (resync is trivial only if there are no local changes).
Thanks for looking at it.
One option is to put the local changes in an #include file. We have a similar problem with "u-boot,dm-pre-reloc". We really should be able to push these sorts of things upstream....
The conflict is then whether the device tree people would consider this new block to be "describing the hardware" or not and therefore whether these changes would be welcomed/allowed upstream. I'm not sure which way it would go TBH.
Me neither. Although there is a flash map / partition map device tree binding and this is similar.
The binman description lays out what the boot ROM needs in many cases, which is pretty close to being the 'hardware'.
There is also an issue with duplication, since this stanza is going to be the same for all the dozens of sunxi platforms. I suppose this can be addressed by putting it in a suitable .dtsi file, but that does sort of start to suggest that this information belongs outside the per-board .dts in a per-platform file (at which point I would have have to ask if .dts is the best/most convenient syntax?).
Yes I think we could use the same thing for large groups of boards, but we already have an soc-specific .dtsi in most cases, so that should work quite nicely.
The nice things about device tree are:
- people understand it - it can be accessed at run-time (e.g. so we know where to find the next image to load) - we have robust tools for dealing with device tree source and binaries
Regards, Simon