R: Clarification about dsa and ipq806x

[ansuelsmth at gmail.com]

> Hi
> 
> [Disclaimer: I'm not an OpenWrt developer]
> 
> On 2021-03-01, Ansuel Smith wrote:
> [...]
> > The idea is to switch this target to dsa but there is a problem...
> > Since kernel 5.10 is a testing kernel how should I change the base
> [...]
> > Do we have some way in openwrt to know if the system use dsa driver or
> > swconfig driver?
> 
> I could come up with these strategies:
> 
> - if swconfig is still working in v5.10 with reasonable efforts, the
>   easy way out could be to stick with swconfig for the TESTING_KERNEL
>   introduction - and switch over to DSA as soon as it is acceptable to
>   get rid of kernel v5.4.
> 

That would be the final idea but since 5.10 is a testing kernel it would be
very good to test dsa with the other thing...

> - adding kernel version based conditionals to
>   - target/linux/ipq806x/base-files/etc/board.d/02_network
>   - target/linux/ipq806x/base-files/etc/board.d/05_compat-version (new)
>   - target/linux/ipq806x/image/Makefile (checking for TESTING_KERNEL)
>   and toggling DEVICE_COMPAT_VERSION between them as needed. This
>   does require (and enforces via DEVICE_COMPAT_VERSION) nuking the
>   existing configuration between v5.4 and v5.10 based builds, but would
>   allow building either from the same source.
> 

The check if the system has dsa driver is needed only in the 02_network.
Still don't know if it's acceptable or not.

> - if DSA is mostly working on kernel v5.4 (or could be retrofitted
>   *easily*), switching kernel v5.4 in master over to DSA might be
>   another option.
>   "the dsa driver still lacks vlan support in kernel 5.4" suggests
>   that this wouldn't be a reasonable option though.
> 

I never checked since I use a basic network config but software vlan
should work. Dsa on 5.4 works good (it's like 2 years that I use it) but
still the best approach would be dsa on 5.10 and keep swconfig on 5.4

> 
> Just two questions of my own.
> 
> I think to remember that introducing DSA based drivers is slightly at
> odds with devices storing the MAC address in ASCII representation
> (uboot-env) rather than binary at a fixed offset, is that an issue
> here?
> This affects linksys,ea7500-v1, linksys,ea8500 and zyxel,nbg6817
> 

Mac address is set by the gmac driver so we shouldn't have any problem
about that.

> Is setting tx-fifo-depth still necessary/ useful?
> http://lists.infradead.org/pipermail/openwrt-devel/2020-
> September/031417.html
> 

Need to check this... In theory this is needed to set mtu to the entire
switch. The dsa driver requires to set additional mtu as it has some
overhead and by some logs fails to do that. So I still need to investigate
about this.

> 
> Regards
> 	Stefan Lippers-Hollmann