[U-Boot] [PATCH v1] env_mmc: configure environment offsets via device tree

[Dr. Philipp Tomsich]

> On 20 Feb 2017, at 08:22, Igor Grinberg <grinberg at compulab.co.il> wrote:
> 
>>> That sounds too odd...
>>> DT's purpose is to describe the h/w... and that does not look so...
>>> We also, have a dt file name in the environment, so this creates will create
>>> a chicken and an egg problem…
>> 
>> I don’t really follow… as far as I knew the DT name would have to come
>> from some other source anyway, as the device containing the env might
>> only be described through the device tree (e.g. mmc0).
> 
> Why? U-Boot can live pretty well w/o DT.

If U-Boot runs without DT, then nothing will/should change about how the setting
is retrieved from CONFIG_ENV_OFFSET.

The platform that motivated this change is ARCH_SUNXI, which does not use
per-board defines but aims to have one generic bootloader per-SoC.

>>> I really don't think we should go that direction. DT is not meant to provide
>>> a solution to all your problems...
>> 
>> I don’t see how this is different from other entries in chosen and config as
>> of today:
>> 	common/autoboot.c allows an override through /config/bootdelay
>> 	common/board_r.c uses /config/load-environment
>> 	common/cli.c can pull in /config/bootcmd
>> 	drivers/serial/serial-uclass.c uses /chosen/stdout-path
>> 
>> In fact, it is the absence of this mechanism that is causing problems today:
>> CONFIG_ENV_OFFSET is not configurable through Kconfig, so we would
>> need board-specific defines (e.g. CONFIG_SUNXI_BOARD_LYNX) and
>> matching #ifdef primitives in a shared header (sunxi-common.h in our case).
> 
> Right. Exactly, I think we should move the CONFIG_ENV_OFFSET to Kconfig.
> And that will solve the problem.

Doing this would still get into the way of architectures that want to build a single
‘universal’ bootloader for their SoC: the ENV_OFFSET may not be the same
across all board and vendor configurations. This can easily be handled with the
(optional) prop in the DT, but not with the compile-time ENV_OFFSET.

If we decide to this, I’d at least like to introduce the function call to (the weak
function) mmc_get_env_addr(…), so we can override this in the board code.

>> So putting this in the DT is the best (and least intrusive) option available.
> 
> Ok. I see your point now. Yet I think we should keep the DT to its purpose - which
> is to describe the h/w (and not the s/w placement layout).
> 
> 
> -- 
> Regards,
> Igor.