Re: [PATCH net-next v2 7/9] net: phy: introduce ethtool_phy_ops to get and set phy configuration

From: Maxime Chevallier
Date: Tue Oct 08 2024 - 12:41:23 EST


On Tue, 8 Oct 2024 16:27:43 +0100
"Russell King (Oracle)" <linux@xxxxxxxxxxxxxxx> wrote:

> On Tue, Oct 08, 2024 at 04:57:42PM +0200, Maxime Chevallier wrote:
> > Oh but I plan to add support for the marvell switch, mcbin, and turris
> > first,
>
> What do you think needs adding for the mcbin?
>
> For the single-shot version, the serdes lines are hard-wired to the
> SFP cages, so it's a MAC with a SFP cage directly connected.
>
> For the double-shot, the switching happens dynamically within the
> 88x3310 PHY, so there's no need to fiddle with any isolate modes.

Nothing related to isolate mode regarding the mcbin :) They aren't
even implemented on the 3310 PHYs anyway :)

>
> The only thing that is missing is switching the 88x3310's fibre
> interface from the default 10gbase-r to 1000base-X and/or SGMII, and
> allowing PHYs to be stacked on top. The former I have untested
> patches for but the latter is something that's waiting for
> networking/phylib to gain support for stacked PHY.

That's one part of it indeed

> Switching the interface mode is very disruptive as it needs the PHY
> to be software-reset, and if the RJ45 has link but one is simply
> plugging in a SFP, hitting the PHY with a software reset will
> disrupt that link.
>
> Given that the mcbin has one SFP cage that is capable of 2500base-X,
> 1000base-X and SGMII, and two SFP cages that can do 10gbase-r, with
> a PHY that can do 10/100/1G/2.5G/5G/10G over the RJ45, I'm not sure
> adding more complexity really gains us very much other than...
> additional complexity.

What I mean is the ability for users to see, from tools like ethtool,
that the MCBin doubleshot's eth0 and eth1 interfaces have 2 ports
(copper + sfp), and potentially allow picking which one to use in case
both ports are connected.

There are mutliple devices out-there with such configurations (some
marvell switches for example). Do you not see some value in this ?

This isn't related at all to isolate, but rather the bigger picture of
the type of topology I'm trying to improve support for.

Setups with 2 PHYs connected to the same MAC are similar to the eth0/1
interfaces in the sense that they offer 2 front-facing ports for one
single MAC. IMO it's easier to first deal with the MCBin setup first.

Again that's a whole other topic, but my idea would be to be able, from
ethtool, to see that mcbin eth0 has one port capable of
10/100/1000/2500/5000/10000BaseT, and another capable of
1000BaseX/2500BaseX/5GBaseR/10GBaseR or whatever the plugged SFP module
offers.

I do know that the MCBin has a large variety of interfaces easily
accessible, but it also looks like a good board to introduce such
multi-port support. Many people have it, it works well with an upstream
kernel, making testing and review effort easier IMO.

I know that this whole thing of dealing with 2 PHYs attached to the
same MAC has lots of ramifications (the 1 PHY 2 ports setup I just
mentionned, the phy_link_topology that has been added, the isolate
mode, muxing support, etc.), that's why I tried to cover all the angles
at netdevconf + LPC. I have code for most of that pretty-much ready to
send.

Thanks,

Maxime