Re: [RFC PATCH net-next 8/9] net: dsa: Use PHYLINK for the CPU/DSA ports

From: Vladimir Oltean <olteanv@gmail.com>
To: Florian Fainelli <f.fainelli@gmail.com>
Cc: Ioana Ciornei <ioana.ciornei@nxp.com>,
	"linux@armlinux.org.uk" <linux@armlinux.org.uk>,
	"andrew@lunn.ch" <andrew@lunn.ch>,
	"hkallweit1@gmail.com" <hkallweit1@gmail.com>,
	"maxime.chevallier@bootlin.com" <maxime.chevallier@bootlin.com>,
	"netdev@vger.kernel.org" <netdev@vger.kernel.org>,
	"davem@davemloft.net" <davem@davemloft.net>
Subject: Re: [RFC PATCH net-next 8/9] net: dsa: Use PHYLINK for the CPU/DSA ports
Date: Thu, 23 May 2019 23:01:26 +0300	[thread overview]
Message-ID: <CA+h21hpPyk=BYxBXDH5-SGfJdS-E+X9PfZHAMLYNwhL-1stumA@mail.gmail.com> (raw)
In-Reply-To: <9c953f4f-af27-d87d-8964-16b7e32ce80f@gmail.com>

On Thu, 23 May 2019 at 05:17, Florian Fainelli <f.fainelli@gmail.com> wrote:
>
> On 5/22/2019 6:20 PM, Ioana Ciornei wrote:
> > This completely removes the usage of PHYLIB from DSA, namely for the
> > aforementioned switch ports which used to drive a software PHY manually
> > using genphy operations.
> >
> > For these ports, the newly introduced phylink_create_raw API must be
> > used, and the callbacks are received through a notifier block registered
> > per dsa_port, but otherwise the implementation is fairly
> > straightforward, and the handling of the regular vs raw PHYLINK
> > instances is common from the perspective of the driver.
> >
> > What changes for drivers:
> >
> > The .adjust_link callback is no longer called for the fixed-link CPU/DSA
> > ports and drivers must migrate to standard PHYLINK operations (e.g.
> > .phylink_mac_config).  The reason why we can't do anything for them is
> > because PHYLINK does not wrap the fixed link state behind a phydev
> > object, so we cannot wrap .phylink_mac_config into .adjust_link unless
> > we fabricate a phy_device structure.
>
> Can't we offer a slightly nicer transition period for DSA switch drivers:
>
> - if adjust_link and phylink_mac_ops are both supported, prefer
> phylink_mac_ops
> - if phylink_mac_ops is defined alone use it, we're good
> - if adjust_link alone is defined, keep using it with the existing code
> but print a warning inviting to migrate?
>
> The changes look fine but the transition path needs to be a little more
> gentle IMHO.
> --
> Florian

Hi Florian,

Yes we could, but since most of the adjust_link -> phylink_mac_ops
changes appear trivial, and we have the knowledge behind b53 right
here, can't we just migrate everything in the next patchset and remove
adjust_link altogether from DSA?
...So why does b53 configure the SGMII SerDes in phylink_mac_config,
and RGMII delays for the CPU port in adjust_link? Why are pause frames
only configured for the CPU port?
A migration from my side would be rather mechanical, so could you
please share some suggestions?

-Vladimir