Hi Rob,
Am 2021-04-16 20:44, schrieb Rob Herring:
On Fri, Apr 16, 2021 at 01:49:23PM +0200, Michael Walle wrote:
The goal is to fetch a (base) MAC address from the OTP region of a SPI
NOR
flash.
This is the first part, where I try to add the nvmem provider support
to
the MTD core.
I'm not sure about the device tree bindings. Consider the following
two
variants:
(1)
flash@0 {
..
otp {
compatible = "mtd-user-otp";
mtd is a linuxism. Why not just 'nvmem-cells' here or as a fallback if
we come up with a better name?
There are two different compatibles: "mtd-user-otp" and
"mtd-factory-otp"
to differentiate what kind of OTP should be used (and both are possible
at the same time). Thus nvmem-cells alone won't be enough. We could drop
the "mtd-" prefix though.
Is there a benefit of having the following?
compatible = "user-otp", "nvmem-cells";
#address-cells = <1>;
#size-cells = <1>;
serial-number@0 {
reg = <0x0 0x8>;
};
};
};
(2)
flash@0 {
..
otp {
compatible = "mtd-user-otp";
#address-cells = <1>;
#size-cells = <1>;
some-useful-name {
compatible = "nvmem-cells";
serial-number@0 {
reg = <0x0 0x8>;
};
};
};
};
Both bindings use a subnode "opt[-N]". We cannot have the nvmem cells
as
children to the flash node because of the legacy partition binding.
(1) seems to be the form which is used almost everywhere in the
kernel.
That is, the nvmem cells are just children of the parent node.
(2) seem to be more natural, because there might also be other
properties
inside the otp subnode and might be more future-proof.
At the moment this patch implements (1).
I think approach (1) seems fine.
ok
-michael
