Date:   Fri, 16 Sep 2022 08:20:09 +0100
From:   "Russell King (Oracle)" <linux@...linux.org.uk>
To:     Siddharth Vadapalli <s-vadapalli@...com>
Cc:     davem@...emloft.net, edumazet@...gle.com, kuba@...nel.org,
        pabeni@...hat.com, robh+dt@...nel.org,
        krzysztof.kozlowski@...aro.org, krzysztof.kozlowski+dt@...aro.org,
        vladimir.oltean@....com, grygorii.strashko@...com, vigneshr@...com,
        nsekhar@...com, netdev@...r.kernel.org, devicetree@...r.kernel.org,
        linux-kernel@...r.kernel.org, linux-arm-kernel@...ts.infradead.org,
        kishon@...com
Subject: Re: [PATCH 5/8] net: ethernet: ti: am65-cpsw: Add support for
 fixed-link configuration

On Fri, Sep 16, 2022 at 10:24:48AM +0530, Siddharth Vadapalli wrote:
> On 15/09/22 15:37, Russell King (Oracle) wrote:
> > Hi,
> > 
> > On Thu, Sep 15, 2022 at 02:58:52PM +0530, Siddharth Vadapalli wrote:
> >> Hello Russell,
> >>
> >> On 14/09/22 21:39, Russell King (Oracle) wrote:
> >>> On Wed, Sep 14, 2022 at 03:20:50PM +0530, Siddharth Vadapalli wrote:
> >>>> Check for fixed-link in am65_cpsw_nuss_mac_config() using struct
> >>>> am65_cpsw_slave_data's phy_node property to obtain fwnode. Since
> >>>> am65_cpsw_nuss_mac_link_up() is not invoked in fixed-link mode, perform
> >>>> the relevant operations in am65_cpsw_nuss_mac_config() itself.
> >>>
> >>> Further to my other comments, you also fail to explain that, when in
> >>> fixed-link SGMII mode, you _emulate_ being a PHY - which I deduce
> >>> since you are sending the duplex setting and speed settings via the
> >>> SGMII control word. Also, as SGMII was invented for a PHY to be able
> >>> to communicate the media negotiation resolution to the MAC, SGMII
> >>> defines that the PHY fills in the speed and duplex information in
> >>> the control word to pass it to the MAC, and the MAC acknowledges this
> >>> information. There is no need (and SGMII doesn't permit) the MAC to
> >>> advertise what it's doing.
> >>>
> >>> Maybe this needs to be explained in the commit message?
> >>
> >> I had tested SGMII fixed-link mode using a bootstrapped ethernet layer-1
> >> PHY. Based on your clarification in the previous mails that there is an
> >> issue with the fixed-link mode which I need to debug, I assume that what
> >> you are referring to here also happens to be a consequence of that.
> >> Please let me know if I have misunderstood what you meant to convey.
> > 
> > I think what you're saying is that you have this setup:
> > 
> >   ethernet MAC <--SGMII link--> ethernet PHY <---> media
> > 
> > which you are operating in fixed link mode?
> 
> Yes, and the other end is connected to my PC's ethernet port.
> 
> > 
> > From the SGMII specification: "This is achieved by using the Auto-
> > Negotiation functionality defined in Clause 37 of the IEEE
> > Specification 802.3z. Instead of the ability advertisement, the PHY
> > sends the control information via its tx_config_Reg[15:0] as specified
> > in Table 1 whenever the control information changes. Upon receiving
> > control information, the MAC acknowledges the update of the control
> > information by asserting bit 14 of its tx_config_reg{15:0] as specified
> > in Table 1."
> > 
> > For the control word sent from the MAC to the PHY, table 1 specifies a
> > value of 0x4001. All the zero bits in that word which are zero are
> > marked as "Reserved for future use." There are no fields for speed and
> > duplex in this acknowledgement word to the PHY.
> > 
> > I hope this clears up my point.
> 
> Thank you for the detailed explanation. After reading the above, my
> understanding is that even in the fixed-link mode, the ethernet MAC is
> not supposed to advertise the speed and duplex settings. The ethernet
> MACs present on both ends of the connection are supposed to be set to
> the same speed and duplex settings via the devicetree node. Thus, only
> for my setup which happens to be a special case of fixed-link mode where
> the ethernet PHY is present, I am having to send the control word due to
> the presence of a PHY in between.

In SGMII, the control word is only passed between the ethernet MAC and
the ethernet PHY. It is not conveyed across the media.

> And, I am supposed to mention this in
> the commit message, which I haven't done. Please let me know if this is
> what I was supposed to understand.

If you implement this conventionally, then you don't need to mention it
in the commit message, because you're following the standard.

> I am planning to change to a proper fixed-link setup without any
> ethernet PHY between the MACs, for debugging the driver's fixed-link
> mode where the "mac_link_up()" is not invoked.

SGMII is designed for the setup in the diagram I provided in my previous
email. It is not designed for two MACs to talk direct to each other
without any ethernet PHY because of the asymmetric nature of the control
word.

The PHY sends e.g. a control word of 0x9801 for 1G full duplex. On
reception of that, the MAC responds with 0x4001. Finally, the PHY
responds with 0xd801 to acknowledge the receipt of the MAC response.

If both ends of the link are SGMII, both ends will be waiting for
the control word from a PHY which is not present, and the link will
not come up.

1000base-X is a symmetric protocol where both ends of the link
advertise their capabilities, acknowledge each others abilities and
resolve the duplex and pause settings.

SGMII is a Cisco proprietary modification of 1000base-X designed for
communicating the results of media autonegotiation between an
ethernet PHY and ethernet MAC.

-- 
RMK's Patch system: https://www.armlinux.org.uk/developer/patches/
FTTP is here! 40Mbps down 10Mbps up. Decent connectivity at last!

Powered by blists - more mailing lists