Re: [PATCH net-next 4/5] net: phylink: pcs: add 802.3 clause 22 helpers

[Vladimir Oltean <olteanv@gmail.com>]

On Thu, 12 Mar 2020 at 15:13, Russell King - ARM Linux admin
<linux@armlinux.org.uk> wrote:
>
> On Thu, Mar 12, 2020 at 02:54:55PM +0200, Vladimir Oltean wrote:
> > On Wed, 11 Mar 2020 at 22:32, Russell King - ARM Linux admin
> > <linux@armlinux.org.uk> wrote:
> > >
> > > On Wed, Mar 11, 2020 at 09:59:18PM +0200, Vladimir Oltean wrote:
> > > > On Wed, 11 Mar 2020 at 21:32, Russell King - ARM Linux admin
> > > > <linux@armlinux.org.uk> wrote:
> > > > > So, why abuse some other subsystem's datastructure for something that
> > > > > is entirely separate, potentially making the maintanence of that
> > > > > subsystem more difficult for the maintainers?  I don't get why one
> > > > > would think this is an acceptable approach.
> > > > >
> > > > > What you've said is that you want to use struct phy_device, but you
> > > > > don't want to publish it into the device model, you don't want to
> > > > > use mdio accesses, you don't want to use phylib helpers.  So, what's
> > > > > the point of using struct phy_device?  I don't see _any_ reason to
> > > > > do that and make things unnecessarily more difficult for the phylib
> > > > > maintainers.
> > > > >
> > > >
> > > > So if it's such a big mistake...
> > > >
> > > > > > > Sorry, but you need to explain better what you would like to see here.
> > > > > > > The additions I'm adding are to the SGMII specification; I find your
> > > > > > > existing definitions to be obscure because they conflate two different
> > > > > > > bit fields together to produce something for the ethtool linkmodes
> > > > > > > (which I think is a big mistake.)
> > > > > >
> > > > > > I'm saying that there were already LPA_SGMII definitions in there.
> > > > > > There are 2 "generic" solutions proposed now and yet they cannot agree
> > > > > > on config_reg definitions. Omitting the fact that you did have a
> > > > > > chance to point out that big mistake before it got merged, I'm
> > > > > > wondering why you didn't remove them and add your new ones instead.
> > > > > > The code rework is minimal. Is it because the definitions are in UAPI?
> > > > > > If so, isn't it an even bigger mistake to put more stuff in UAPI? Why
> > > > > > would user space care about the SGMII config_reg? There's no user even
> > > > > > of the previous SGMII definitions as far as I can tell.
> > > > >
> > > > > I don't see it as a big deal - certainly not the kind of fuss you're
> > > > > making over it.
> > > > >
> > > >
> > > > ...why keep it?
> > > > I'm all for creating a common interface for configuring this. It just
> > > > makes me wonder how common it is going to be, if there's already a
> > > > driver in-tree, from the same PCS hardware vendor, which after the
> > > > patchset you're proposing is still going to use a different
> > > > infrastructure.
> > >
> > > Do you see any reason why felix_vsc9959 couldn't make use of the code
> > > I'm proposing?
> > >
> >
> > No. But the intentions just from reading the cover letter and the
> > patches seemed a bit unclear. The fact that there are no proposed
> > users in this series, and that in your private cex7 branch only dpaa2
> > uses it, it seemed to me that at least some clarification was due.
> > I have no further comments. The patches themselves are fairly trivial.
>
> I have been told by Andrew to send small series, so that's what I do.
>
> I have not included the DPAA2 changes in this series because it was
> not ready for submission - I had to initially hard-code the physical
> addresses of the MDIO blocks, but I've later moved to describing them
> in the DTS, which now brings with it additional complexities since
> (a) existing DTS need to continue working and (b) working out how to
> submit those changes since the DTS changes and the net changes should
> go via different paths, and ensuring that no breakage will occur
> should they become separated.
>

I think even asking for firmware ABI changes in MC is worth a shot? It
might be preferable to get away with the firmware giving you the PCS
base address for a DPMAC rather than getting it from DT, for the
reasons that you've mentioned. Or even asking for ABI for the firmware
to perform the MDIO read/write itself.

> If you look at the branches that I publish, you will notice that they
> are based on v5.5, and so do not contain your changes to felix_vsc9959
> that you have been talking about, so felix_vsc9959 is not yet on my
> radar.
>

Yes, I noticed.

> However, it seems we take different approaches to contributing code to
> the kernel; I look to see whether there is value to providing common
> infrastructure and then provide it, whereas you seem to take the
> approach of writing specific drivers and hope that someone else spots
> the code in your driver and converts it to something generic.  I
> disagree with your approach because it's been well proven over the
> years that the kernel has been around that relying on others to spot
> code that could be refactored into common helpers just doesn't happen.
> Yes, it happens but only occasionally, and not always when common
> helpers get introduced.
>
> You have already proven the worth of having a set of common helpers -
> it seems that felix_vsc9959 and DPAA2 can both make use of these,
> which does not surprise me one bit, since these helpers are only
> implementing what is published in industry standards or defacto
> industry standards - and as such are likely to be implemented by a lot
> of vendors.  Sure, there will be exceptions and augmentations, which
> is something I considered when creating these common helpers.  That's
> why they are helpers rather than being mandatory implementations.
>

Nothing wrong with that. I'm willing to try to rework felix using
these helpers, but I just don't have the time right now. That and
DPAA2 make it look like it may take a while for the first users to
come... Apologies in advance if there are other immediate potential
users which are not on my radar.

> --
> 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

-Vladimir