Re: [PATCH 2/2] net: dsa: qca8k: introduce SGMII configuration options

[Jonathan McDowell]

On Sat, Jun 06, 2020 at 09:37:41AM +0100, Russell King - ARM Linux admin wrote:
> On Sat, Jun 06, 2020 at 08:49:16AM +0100, Jonathan McDowell wrote:
> > On Fri, Jun 05, 2020 at 08:38:43PM +0200, Andrew Lunn wrote:
> > > On Fri, Jun 05, 2020 at 07:10:58PM +0100, Jonathan McDowell wrote:
> > > > The QCA8337(N) has an SGMII port which can operate in MAC, PHY or BASE-X
> > > > mode depending on what it's connected to (e.g. CPU vs external PHY or
> > > > SFP). At present the driver does no configuration of this port even if
> > > > it is selected.
> > > > 
> > > > Add support for making sure the SGMII is enabled if it's in use, and
> > > > device tree support for configuring the connection details.
> > > 
> > > It is good to include Russell King in Cc: for patches like this.
> > 
> > No problem, I can keep him in the thread; I used get_maintainer for the
> > initial set of people/lists to copy.
> 
> get_maintainer is not always "good" at selecting the right people,
> especially when your patches don't match the criteria; MAINTAINERS
> contains everything that is sensible, but Andrew is suggesting that
> you copy me because in his opinion, you should be using phylink -
> and that's something that you can't encode into a program.

Sure, and I appreciate the pointer to appropriate people who might
provide helpful comments.

> Note that I haven't seen your patches.

I'll make sure to copy you on v2.

> > > Also, netdev is closed at the moment, so please post patches as RFC.
> > 
> > "closed"? If you mean this won't get into 5.8 then I wasn't expecting it
> > to, I'm aware the merge window for that is already open.
> 
> See https://www.kernel.org/doc/Documentation/networking/netdev-FAQ.txt
> "How often do changes from these trees make it to the mainline Linus
> tree?"

Ta. I'll hold off on a v2 until after -rc1 drops.

> > > It sounds like the hardware has a PCS which can support SGMII or
> > > 1000BaseX. phylink will tell you what mode to configure it to. e.g. A
> > > fibre SFP module will want 1000BaseX. A copper SFP module will want
> > > SGMII. A switch is likely to want 1000BaseX. A PHY is likely to want
> > > SGMII. So remove the "sgmii-mode" property and configure it as phylink
> > > is requesting.
> > 
> > It's more than SGMII or 1000BaseX as I read it. The port can act as if
> > it's talking to an SGMII MAC, i.e. a CPU, or an SGMII PHY, i.e. an
> > external PHY, or in BaseX mode for an SFP. I couldn't figure out a way
> > in the current framework to automatically work out if I wanted PHY or
> > MAC mode. For the port tagged CPU I can assume MAC mode, but a port that
> > doesn't have that might still be attached to the CPU rather than an
> > external PHY.
> 
> That depends what you're connected to. Some people call the two sides
> of SGMII "System side" and "Media side". System side is where you're
> receiving the results of AN from a PHY. Media side is where you're
> telling the partner what you want it to do.
> 
> Media side is only useful if you're connected to another MAC, and
> unless you have a requirement for it, I would suggest not implementing
> that - you could come up with something using fixed-link, or it may
> need some other model if the settings need to change.  That depends on
> the application.

So the device in question is a 7 port stand alone switch chip. There's a
single SGMII port which is configurable between port 0 + 6 (they can
also be configure up as RGMII, while the remaining 5 ports have their
own phys).

It sounds like there's a strong preference to try and auto configure
things as much as possible, so I should assume the CPU port is in MAC
mode, and anything not tagged as a CPU port is talking to a PHY/BASEX.

I assume I can use PHY_INTERFACE_MODE_1000BASEX on the
phylink_mac_config call to choose BASEX?

> > > What exactly does sgmii-delay do?
> > 
> > As per the device tree documentation update I sent it delays the SGMII
> > clock by 2ns. From the data sheet:
> > 
> > SGMII_SEL_CLK125M	sgmii_clk125m_rx_delay is delayed by 2ns
> 
> This sounds like a new world of RGMII delay pain but for SGMII. There
> is no mention of "delay" in the SGMII v1.8 specification, so I guess
> it's something the vendor is doing. Is this device capable of
> recovering the clock from a single serdes pair carrying the data,
> or does it always require the separate clock?

Pass, but I think I might be able to get away without having to
configure that for the moment.

I'll go away and roll a v2 moving qca8k over to phylink and then using
that to auto select the appropriate SGMII mode. Thanks for the feedback.

J.

-- 
I started out with nothing & still have most of it left.