Message-ID: <20200803165342.GP1551@shell.armlinux.org.uk>
Date:   Mon, 3 Aug 2020 17:53:42 +0100
From:   Russell King - ARM Linux admin <linux@...linux.org.uk>
To:     "Madalin Bucur (OSS)" <madalin.bucur@....nxp.com>
Cc:     Andrew Lunn <andrew@...n.ch>,
        Vikas Singh <vikas.singh@...esoftware.com>,
        "f.fainelli@...il.com" <f.fainelli@...il.com>,
        "hkallweit1@...il.com" <hkallweit1@...il.com>,
        "netdev@...r.kernel.org" <netdev@...r.kernel.org>,
        "Calvin Johnson (OSS)" <calvin.johnson@....nxp.com>,
        kuldip dwivedi <kuldip.dwivedi@...esoftware.com>,
        Vikas Singh <vikas.singh@....com>
Subject: Re: [PATCH 2/2] net: phy: Associate device node with fixed PHY

On Mon, Aug 03, 2020 at 02:47:41PM +0000, Madalin Bucur (OSS) wrote:
> > -----Original Message-----
> > From: Russell King - ARM Linux admin <linux@...linux.org.uk>
> > Sent: 03 August 2020 17:10
> > To: Madalin Bucur (OSS) <madalin.bucur@....nxp.com>
> > Cc: Andrew Lunn <andrew@...n.ch>; Vikas Singh
> > <vikas.singh@...esoftware.com>; f.fainelli@...il.com; hkallweit1@...il.com;
> > netdev@...r.kernel.org; Calvin Johnson (OSS) <calvin.johnson@....nxp.com>;
> > kuldip dwivedi <kuldip.dwivedi@...esoftware.com>; Vikas Singh
> > <vikas.singh@....com>
> > Subject: Re: [PATCH 2/2] net: phy: Associate device node with fixed PHY
> > 
> > On Mon, Aug 03, 2020 at 11:45:55AM +0000, Madalin Bucur (OSS) wrote:
> > > > -----Original Message-----
> > > > From: Russell King - ARM Linux admin <linux@...linux.org.uk>
> > > > Sent: 03 August 2020 12:07
> > > > To: Madalin Bucur (OSS) <madalin.bucur@....nxp.com>
> > > > Cc: Andrew Lunn <andrew@...n.ch>; Vikas Singh
> > > > <vikas.singh@...esoftware.com>; f.fainelli@...il.com;
> > hkallweit1@...il.com;
> > > > netdev@...r.kernel.org; Calvin Johnson (OSS)
> > <calvin.johnson@....nxp.com>;
> > > > kuldip dwivedi <kuldip.dwivedi@...esoftware.com>; Vikas Singh
> > > > <vikas.singh@....com>
> > > > Subject: Re: [PATCH 2/2] net: phy: Associate device node with fixed
> > PHY
> > > >
> > > > On Mon, Aug 03, 2020 at 08:33:19AM +0000, Madalin Bucur (OSS) wrote:
> > > > > > -----Original Message-----
> > > > > > From: netdev-owner@...r.kernel.org <netdev-owner@...r.kernel.org>
> > On
> > > > > > Behalf Of Andrew Lunn
> > > > > > Sent: 01 August 2020 18:11
> > > > > > To: Russell King - ARM Linux admin <linux@...linux.org.uk>
> > > > > > Cc: Vikas Singh <vikas.singh@...esoftware.com>;
> > f.fainelli@...il.com;
> > > > > > hkallweit1@...il.com; netdev@...r.kernel.org; Calvin Johnson (OSS)
> > > > > > <calvin.johnson@....nxp.com>; kuldip dwivedi
> > > > > > <kuldip.dwivedi@...esoftware.com>; Madalin Bucur (OSS)
> > > > > > <madalin.bucur@....nxp.com>; Vikas Singh <vikas.singh@....com>
> > > > > > Subject: Re: [PATCH 2/2] net: phy: Associate device node with
> > fixed
> > > > PHY
> > > > > >
> > > > > > On Sat, Aug 01, 2020 at 10:41:32AM +0100, Russell King - ARM Linux
> > > > admin
> > > > > > wrote:
> > > > > > > On Sat, Aug 01, 2020 at 09:52:52AM +0530, Vikas Singh wrote:
> > > > > > > > Hi Andrew,
> > > > > > > >
> > > > > > > > Please refer to the "fman" node under
> > > > > > > > linux/arch/arm64/boot/dts/freescale/fsl-ls1046a-rdb.dts
> > > > > > > > I have two 10G ethernet interfaces out of which one is of
> > fixed-
> > > > link.
> > > > > > >
> > > > > > > Please do not top post.
> > > > > > >
> > > > > > > How does XGMII (which is a 10G only interface) work at 1G speed?
> > Is
> > > > > > > what is in DT itself a hack because fixed-phy doesn't support
> > 10G
> > > > > > > modes?
> > > > > >
> > > > > > My gut feeling is there is some hack going on here, which is why
> > i'm
> > > > > > being persistent at trying to understand what is actually going on
> > > > > > here.
> > > > >
> > > > > Hi Andrew,
> > > > >
> > > > > That platform used 1G fixed link there since there was no support
> > for
> > > > > 10G fixed link at the time. PHYlib could have tolerated 10G speed
> > there
> > > > > With a one-liner.
> > > >
> > > > That statement is false.  It is not a "one liner".  fixed-phy exposes
> > > > the settings to userspace as a Clause 22 PHY register set, and the
> > > > Clause 22 register set does not support 10G.  So, a "one liner" would
> > > > just be yet another hack.  Adding Clause 45 PHY emulation support
> > > > would be a huge task.
> > > >
> > > > > I understand that PHYLink is working to describe this
> > > > > Better, but it was not there at that time. Adding the dependency on
> > > > > PHYLink was not desirable as most of the users for the DPAA 1
> > platforms
> > > > > were targeting kernels before the PHYLink introduction (and last
> > I've
> > > > > looked, it's still under development, with unstable APIs so we'll
> > > > > take a look at this later, when it settles).
> > > >
> > > > I think you need to read Documentation/process/stable-api-nonsense.rst
> > > > particularly the section "Stable Kernel Source Interfaces".
> > > >
> > > > phylink is going to be under development for quite some time to come
> > > > as requirements evolve.  For example, when support for QSFP interfaces
> > > > is eventually worked out, I suspect there will need to be some further
> > > > changes to the driver interface.  This is completely normal.
> > > >
> > > > Now, as to the stability of the phylink API to drivers, it has in fact
> > > > been very stable - it has only changed over the course of this year to
> > > > support split PCS, a necessary step for DPAA2 and a few others.  It
> > has
> > > > been around in mainline for two years, and has been around much longer
> > > > than that, and during that time it has been in mainline, the MAC
> > facing
> > > > interface has not changed until recently.
> > > >
> > > > So, I find your claim to be quite unreasonable.
> > >
> > > I see you agree that there were and there will be many changes for a
> > while,
> > > It's not a complaint, I know hot it works, it's just a decision based on
> > > required effort vs features offered vs user requirements. Lately it's
> > been
> > > time consuming to try to fix things in this area.
> > 
> > No, it hasn't been time consuming.  The only API changes as far as
> > drivers are concerned have been:
> > 
> > 1. the change to the mac_link_up() prototype to move the setup of the
> >    final link parameters out of mac_config() - and almost all of the
> >    updates to users were done by me.
> > 
> > 2. the addition of split PCS support, introducing new interfaces, has
> >    had minimal impact on those drivers that updated in step (1).
> > 
> > There have been no other changes as far as users are concerned.
> > 
> > Some of the difficulty with (1) has been that users of phylink appeared
> > initially with no proper review, and consequently they got quite a lot
> > wrong.  The most common error has been using state->speed, state->duplex
> > in mac_config() methods irrespective of the AN mode, which has _always_
> > since before phylink was merged into mainline, been totally unreliable.
> > 
> > That leads me on to the other visible "changes" for users are concerned,
> > which may be interpreted as interface changes, but are not; they have
> > all been clarifications to the documentation, to strengthen things such
> > as "do not use state->speed and state->duplex in mac_config() for
> > various specific AN modes".  Nothing has actually changed with any of
> > those clarifications.
> > 
> > For example, if in in-band mode, and mac_config() uses state->speed
> > and state->duplex, then it doesn't matter which version of phylink
> > you're using, if someone issues ethtool -s ethN ..., then state->speed
> > and state->duplex will be their respective UNKNOWN values, and if you're
> > using these in that situation, you will mis-program the MAC.
> > 
> > Again, that is not something that has changed.  Ever.  But the
> > documentation has because people just don't seem to get it, and I seemed
> > to be constantly repeating myself in review after review on the same
> > points.
> > 
> > So, your assertion that the phylink API is not stable is false.  It
> > has been remarkably stable over the two years that it has been around.
> > It is only natural that as the technology that a piece of code supports
> > evolves, so the code evolves with it.  That is exactly what has happened
> > this year with the two changes I mention above.
> > 
> > Now, if you've found it time consuming to "fix things" (unspecified what
> > "things" are) then I assert that what has needed to be fixed are things
> > that NXP have got wrong.  Such as the rtnl cockups.  Such as abusing
> > state->speed and state->duplex.  None of that is because the interface
> > is unstable - they are down to buggy implementation on NXPs part.
> > 
> > Essentially, what I'm saying is that your attempt to paint phylink as
> > being painful on the basis of interface changes is totally and utterly
> > wrong and is just an excuse to justify abusing the fixed-link code and
> > specifying things that are clearly incorrect via DT.
> 
> Thank you for the distilled phylink history, it may be easier to comprehend
> with these details. I was not referring to phylink, but PHY related issues
> on the DPAA 1 platforms.

Sigh.  No, you were referring to phylink.  This is what you said:

> I understand that PHYLink is working to describe this
> Better, but it was not there at that time. Adding the dependency on
> PHYLink was not desirable as most of the users for the DPAA 1 platforms
> were targeting kernels before the PHYLink introduction (and last I've
> looked, it's still under development, with unstable APIs so we'll
> take a look at this later, when it settles).

This discussion stems from your misconception and incorrect statements
concerning phylink, which I am correcting in this discussion.

-- 
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