Re: [RFC net-next 6/8] net: phylink: Configure MAC/PCS when link is up without PHY

[Russell King - ARM Linux admin]

On Tue, Feb 04, 2020 at 06:43:18PM +0100, Andrew Lunn wrote:
> > There, there is one MAC, but there are multiple different PCS - one
> > for SGMII and 1000base-X, another for 10G, another for 25G, etc.
> > These PCS are accessed via a MDIO adapter embedded in each of the
> > MAC hardware blocks.
> 
> Hi Russell
> 
> Marvell mv88e6390X switches are like this is a well. There is a PCS
> for SGMII and 1000Base-X, and a second one for 10G. And it dynamically
> swaps between them depending on the port mode, the so called cmode.
> 
> So a generic solution is required, and please take your time to build
> one.

Well, DSA is quite a mixed bag...

As far as I can work out, the situation with the CPU and DSA ports is
quite hopeless - you've claimed that a change in phylink has broken it,
I can't find what that may be.  The fact is, phylink has never had any
link information for DSA links when no fixed-link property has been
specified in DT.  As I've already said in a previous email about this,
I can't see *any* sane way to fix that - but there was no response.


On a more positive note...

The mac_link_up() changes that I've talked about should work for DSA,
if only there was a reasonable way to reconfigure the ports.  If you
look at the "phy" branch, you will notice that there's a patch there -
"net: mv88e6xxx: use resolved link config in mac_link_up()" which adds
the support to configure the MAC manually.  It's rather messy, and I
see no way to deal with the pause settings.  There is support in some
Marvell DSA switches to force flow control but that's not supported
through the current mid-layer at all (port_set_pause doesn't do it.)
I'm not sure whether the "mv88e6xxx_phy_is_internal()" check there is
the right test for every DSA switch correct either.

What is missing is reading the results from the PCS (aka serdes) and
forwarding them into phylink - I did have a quick look at how that might
be possible, but the DSA code structure (consisting of multiple
mid-layers) makes it hard without rewriting quite a lot of code.  That's
fine if you know all the DSA chips inside out, but I don't - and that's
where we need someone who has the knowledge of all DSA switches that we
support.  Or, we get rid of the multiple mid-layers and switch to a
library approach, so that we can modify support for one DSA switch
without affecting everything.  It may be a simple matter of dropping the
existing serdes workaround, but I'm not sure at the moment.

I've tried this code out on the ZII rev B, I haven't tried it on the rev
C which has the 6390 switches yet.

-- 
RMK's Patch system: https://www.armlinux.org.uk/developer/patches/
FTTC broadband for 0.8mile line in suburbia: sync at 12.1Mbps down 622kbps up
According to speedtest.net: 11.9Mbps down 500kbps up