Message-ID: <ZMuLKa2HX9/LiCPn@shell.armlinux.org.uk>
Date: Thu, 3 Aug 2023 12:10:33 +0100
From: "Russell King (Oracle)" <linux@...linux.org.uk>
To: Jiawen Wu <jiawenwu@...stnetic.com>
Cc: netdev@...r.kernel.org, andrew@...n.ch, hkallweit1@...il.com,
	Jose.Abreu@...opsys.com, mengyuanlou@...-swift.com
Subject: Re: [PATCH net-next 4/7] net: pcs: xpcs: adapt Wangxun NICs for
 SGMII mode

On Thu, Aug 03, 2023 at 10:20:22AM +0800, Jiawen Wu wrote:
> > No there isn't, and it conforms with the above.
> > 
> > A read looks like this:
> > 
> >       Address  Data                   Address  Data     Data
> > Start 10101100 000yyyyy RepeatedStart 10101101 DDDDDDDD DDDDDDDD Stop
> >                       or Stop followed
> > 		          by Start
> > 
> > The terms "Address" and "Data" here are as per the I²C specification.
> > You will notice that the first part has one byte of address and *one*
> > byte of data to convey the register address. This is what the "1" you
> > are referring to above is for.
> > 
> > For completness, a write looks like this:
> > 
> >       Address  Data     Data     Data
> > Start 10101100 000yyyyy DDDDDDDD DDDDDDDD Stop
> > 
> > Essentially, in all cases, when 0x56 is addressed with the data
> > direction in write mode, the very first byte is _always_ the register
> > address and the remainder contain the data. When the data direction is
> > in read mode, the bytes are always data.
> > 
> > The description you quote above is poor because it doesn't make it
> > clear whether "read" and "write" apply to the bus transactions or to
> > the device operations. However, I can assure you that what is
> > implemented is correct, since this is the standard small 24xx memory
> > device protocol, and I've been programming that on various
> > microcontrollers and such like for the last 30 years.
> > 
> > Are you seeing a problem with the data read or written to the PHY?
> 
> Hi Russell,
> 
> I really don't know how to deal with "MAC side SGMII", could you please
> help me?
> 
> From the test results, when I config PCS in "PHY side SGMII", the link status
> of PHY in copper SFP is read by I2C after AN complete. Then PHY's link up
> status is informed to PHYLINK, then PCS will check its status. But when I just
> change PCS to "MAC side SGMII", I2C will keep reading timeouts since AN
> complete. I checked the register of PCS to confirm AN complete, but PHY's
> link status would never be updated in PHYLINK.

I don't understand what is going on here either - but what I do know
is that there is _zero_ difference as far as the network link is
concerned between an on-board PHY using SGMII to the MAC/PCS and a SFP
with a PHY using SGMII.

In both situations the PHY behaves the same - it presents a PHY-side
SGMII interface, so it sends to the MAC/PCS the speed and duplex
settings, and expects the MAC/PCS to acknowledge them.

The name "MAC side SGMII" suggests that this mode provides the
acknowledgement, whereas "PHY side SGMII" suggests that this mode
provides a speed and duplex.

Given all this, using "PHY side SGMII" with a SFP, and "MAC side
SGMII" for an on-board PHY just seems utterly wrong - and I can't
make head nor tail of it.

> It's kind of weird to me, how does the configuration of PCS relate to I2C?

I2C is just the access method for PHYs on SFPs - because there are
no MDIO bus pins on SFP modules, only I2C pins mainly for accessing
the identification EEPROM and diagnostics, but many copper SFPs have
a way to access the PHY.

I2C is transparent as far as phylib is concerned - the mdio-i2c
driver makes the PHY "appear" as if it is on a conventional MDIO
bus.

-- 
RMK's Patch system: https://www.armlinux.org.uk/developer/patches/
FTTP is here! 80Mbps down 10Mbps up. Decent connectivity at last!

Powered by blists - more mailing lists