Re: [RFC net-next 2/5] net: phylink: add separate pcs operations structure

[Russell King - ARM Linux admin <linux@armlinux.org.uk>]

On Wed, Mar 18, 2020 at 07:45:52AM +0000, Jose Abreu wrote:
> From: Russell King - ARM Linux admin <linux@armlinux.org.uk>
> Date: Mar/17/2020, 16:52:08 (UTC+00:00)
> 
> > On Tue, Mar 17, 2020 at 04:04:28PM +0000, Jose Abreu wrote:
> > > From: Russell King - ARM Linux admin <linux@armlinux.org.uk>
> > > Date: Mar/17/2020, 15:56:10 (UTC+00:00)
> > > 
> > > > > Please consider removing this condition and just rely on an_enabled field. 
> > > > > I have USXGMII support for Clause 73 Autoneg so this won't work with 
> > > > > that.
> > 
> > Do you really mean USXGMII or XLGMII that you recently contributed?
> > XLGMII makes more sense for clause 73.
> > 
> > > > That is actually incorrect.  SGMII can have an_enabled being true, but
> > > > SGMII is not an autonegotiation between the MAC and PHY - it is merely
> > > > a mechanism for the PHY to inform the MAC what the results of _its_
> > > > negotiation are.
> > > > 
> > > > I suspect USXGMII is the same since it is just an "upgraded" version of
> > > > SGMII.  Please can you check whether there really is any value in trying
> > > > (and likely failing) to restart the "handshake" with the PHY from the
> > > > MAC end, rather than requesting the PHY to restart negotiation on its
> > > > media side.
> > > 
> > > I think we are speaking of different things here. I'm speaking about 
> > > end-to-end Autoneg. Not PHY <-> PCS <-> MAC.
> > 
> > What do you mean end-to-end autoneg?  Let's take a simple example for
> > SGMII, here is the complete setup:
> > 
> > MAC <--> PCS <--SGMII--> PHY <--MEDIA--> PHY <--SGMII--> MAC
> > 
> > Generally, asking the PCS to "renegotiate" will either be ignored, or
> > might cause the PCS to restart the handshake with the PHY depending on
> > the implementation.  It will not cause the PHY to renegotiate with the
> > remote PHY.
> > 
> > Asking the PHY to renegotiate will cause the link to drop, which
> > updates the config_reg word sent to the PCS to indicate link down.
> > When the link is re-established, the config_reg word is updated again
> > with the new link parameters and that sent to the PCS.
> > 
> > So, just because something is closer to the MAC does not necessarily
> > mean that it causes more of the blocks to "renegotiate" when asked to
> > do so.
> > 
> > In SGMII, the PHY is the "master" and this is what needs negotiation
> > restarted on "ethtool -r" to have the effect that the user desires.
> > 
> > I believe (but I don't know for certain, because the USXGMII
> > documentation is not available to me) that USXGMII is SGMII extended
> > up to additionally include 10G, 5G and 2.5G speeds, and otherwise is
> > basically the same mechanism.
> > 
> > So, I would suggest that USXGMII and SGMII should be treated the same,
> > and for both of them, a renegotiation should always be performed at
> > the PHY and not the PCS.
> > 
> > There is another reason not to try triggering renegotiation at both
> > the PHY and PCS.  When the PHY is renegotiating, the state machines
> > at both the PHY and PCS end are in the process of changing - if we
> > hit the PCS with a request to renegotiate, and the hardware isn't
> > setup to ignore it, that could occur in an unexpected state - the risk
> > of triggering a hardware problem could be higher.
> > 
> > So, based on this, I don't think it's a good idea to restart
> > negotiation at the PCS for SGMII and USXGMII modes.
> > 
> > For the 802.3z based protocols, it makes complete sense to do so,
> > because the PCS are the blocks involved in the actual media negotiation
> > and there is no place else to restart negotiation:
> > 
> > MAC <---> PCS <----fiber 1000base-X----> PCS <---> MAC
> 
> That's kind of the setup I have, hence the need for me to restart ... I 
> have this:
> 
> MAC <-> PCS <-> SERDES <-> Copper <-> SERDES <-> PCS <-> MAC
> 
> So, no PHY to restart Autoneg.

And the protocol over the copper is USXGMII?

-- 
RMK's Patch system: https://www.armlinux.org.uk/developer/patches/
FTTC broadband for 0.8mile line in suburbia: sync at 10.2Mbps down 587kbps up