[PATCH 1/2] kernel: netdevices: Create Realtek DSA switch packages

[Luiz Angelo Daros de Luca]

Hi Linus,

> > Maybe we should back port the upstream changes (and fixes),
> > similar to what I did in
> > https://github.com/luizluca/openwrt/commit/d145c6845d7cdd2d49722a85988f16b17014dfa4
> > (it needs to be updated and rebased). I might have time to rebase that
> > branch soon.
>
> What about we rebase and merge your branch instead?
> It looks very good to me.
>
> Two review comments:
>
> 1. You split the taggers to separate packages, but in practice that is
>   just surplus because the DSA modules are 1:1 mapped to the
>   taggers and the only possible user. Just put respective tagger
>   module into the rtl8366rb and rtl8365mb packages instead.
>   (This is what other DSA switches started doing...)

Sure.

> 2. After that you can just autoload the modules:
>    AUTOLOAD:=$(call AutoProbe,rtl8366rb)
>    AUTOLOAD:=$(call AutoProbe,rtl8365mb)

Yes, the tag should be loaded on demand.

> If you fix this we can just merge it, also it's great to have all the
> backported patches, especially for RTL8365MB so we can move
> these devices over to DSA.

Well, the upstream code does not support VLAN/Bridge Forwarding. So,
all traffic goes through the CPU (with the expected lower
performance). I implemented that part in the driver but the maintainer
rejected it as my implementation was based on Shared VLAN Learning and
he already had one based on Independent VLAN Learning. However, that
happened in early 2023 and no IVL appeared upstream. I've been keeping
my code for personal usage. Should I include it as well as a pending
(or hack) patch?

The clean code, without my downstream patch is already here
https://github.com/openwrt/openwrt/commit/f24c47fa8a2fb653288d34e7a0bcdc6eafa5736e
. I'll still test it on my device before submitting.

> Yours,
> Linus Walleij

Regards,

Luiz