Re: [PATCH] of: net: support NVMEM cells with MAC in text format

[Michael Walle]

Am 2021-12-29 13:40, schrieb Michael Walle:
> > Some NVMEM devices have text based cells. In such cases MAC is stored 
> > in
> > a XX:XX:XX:XX:XX:XX format. Use mac_pton() to parse such data and
> > support those NVMEM cells. This is required to support e.g. a very
> > popular U-Boot and its environment variables.
> >
> > Signed-off-by: Rafał Miłecki <rafal@xxxxxxxxxx>
> > ---
> > Please let me know if checking NVMEM cell length (6 B vs. 17 B) can be
> > considered a good enough solution. Alternatively we could use some DT
> > property to make it explicity, e.g. something like:
> >
> > ethernet@18024000 {
> > 	compatible = "brcm,amac";
> > 	reg = <0x18024000 0x800>;
> >
> > 	nvmem-cells = <&mac_addr>;
> > 	nvmem-cell-names = "mac-address";
> > 	nvmem-mac-format = "text";
> > };
>
> Please note, that there is also this proposal, which had such a 
> conversion
> in mind:
> https://lore.kernel.org/linux-devicetree/20211228142549.1275412-1-michael@xxxxxxxx/
>
> With this patch, there are now two different places where a mac address
> format is converted. In of_get_mac_addr_nvmem() and in the imx otp 
> driver.
> And both have their shortcomings and aren't really flexible. Eg. this 
> one
> magically detects the format by comparing the length, but can't be used 
> for
> to swap bytes (because the length is also ETH_ALEN), which apparently 
> is a
> use case in the imx otp driver. And having the conversion in an nvmem
> provider device driver is still a bad thing IMHO.
>
> I'd really like to see all these kind of transformations in one place.

Unfortunately, there were no replies yet. Can we revert this patch
until there was a discussion and before there are any users of it.
Esp. the latter is hard to track and then it might be impossible
to change them to a better solution.

Any optionions?

-michael