Hi Frieder,
On 7/21/25 3:08 PM, Frieder Schrempf wrote:
Hi Quentin,
Am 21.07.25 um 15:01 schrieb Quentin Schulz:
Hi Frieder,
On 7/21/25 2:51 PM, Frieder Schrempf wrote:
[...]
diff --git a/arch/arm/mach-imx/mx6/Kconfig b/arch/arm/mach-imx/mx6/
Kconfig
index 465e258f8e4..0798eac68f5 100644
--- a/arch/arm/mach-imx/mx6/Kconfig
+++ b/arch/arm/mach-imx/mx6/Kconfig
@@ -366,6 +366,7 @@ config TARGET_MX6S_SIELAFF
select DM_THERMAL
select SUPPORT_SPL
imply CMD_DM
+ imply OF_UPSTREAM
Why "imply" only here? It won't build without OF_UPSTREAM anymore no?
Shouldn't it be "select" instead?
Good question. I wasn't sure about this, either. But "imply OF_UPSTREAM"
is used basically everywhere in the tree whereas "select OF_UPSTREAM" is
only used twice. That's why I assumed there must be some good reason to
use imply. Maybe that assumption was wrong, but I still wonder why
everyone uses this. If someone can explain, that would be appreciated.
I can explain for the Rockchip architecture :)
We have "imply OF_UPSTREAM" at the *SoC* level so that if someone
somehow wants to be able to have their board to NOT use an FDT from
dts/upstream, it is possible (though we wouldn't accept such a patch, so
this is likely only to be nice with forks or also inherited from partial
migration of U-Boot FDTs to upstream's (e.g. there is px30-firefly which
doesn't have an upstream Device Tree, so we cannot move CONFIG_PX30 to
select OF_UPSTREAM as there are still boards which aren't supported)).
Finally, we actually make use of this ability to not use upstream DT for
generic Rockchip platforms. While I don't use this, it seems like it is
appreciated? We cannot upstream those as they do not represent any real
HW but it works good enough for boards following closely the reference
design from Rockchip (which is most boards for typical things used by
U-Boot to boot Linux like eMMC/SD/SPI).
For Kconfig symbols only tackling a finite number of boards which are
all supported by OF_UPSTREAM, it would make more sense to have "select"
there? We also have "generic" Device Trees for Rockchip boards that we
cannot upstream to the Linux kernel (as they don't match any actual HW)
so we need
And now to come to think of it, I believe most/all the options in
TARGET_MX6S_SIELAFF should be in the defconfig instead of under the
Kconfig symbol? Not sure if there's a tree-wide consensus on what to do
here, or if it's up to the architecture's maintainer or board maintainer
to decide. I know the additions I made to the Kconfig in
board/theobroma-systems/ weren't much liked at the time (but ultimately
still merged), and this seems similar, so not sure?.
Hope this helps!
Quentin
