RE: [PATCH net-next v8 10/11] r8152: add TGPHY register access for RTL8157 and RTL8159
From: Chih Kai Hsu
Date: Fri Nov 20 2026 - 03:12:33 EST
On 19/11/2026 9:58 pm, Andrew Lunn wrote:
> On Thu, Nov 19, 2026 at 07:18:16AM +0000, Chih Kai Hsu wrote:
> > On 19/11/2026 12:42 pm, Birger Koblitz wrote:
> > > On 18/11/2026 2:42 pm, Andrew Lunn wrote:
> > > >> I plan to add two new functions, rtl_nway_restart() and
> > > >> rtl8152_get_link_ksettings, to replace mii_nway_restart() and
> > > mii_ethtool_get_link_ksettings, respectively.
> > > >
> > > > Ideally, you should port this driver to phylink.
> > > >
> > > > Andrew
> > >
> > > Indeed, all of the (integrated) PHYs on the adapters supporeted by
> > > r8152 I looked at are already supported under
> > > drivers/net/phy/realtek/ It should be straight-forward with many immediate
> benefits.
> > >
> > > Birger
> >
> > I agree that a phylink conversion is a better long-term direction. I'd
> > like to keep it out of this series. This series fixes and completes
> > the existing RTL8157/59 support. The phylink conversion is a larger
> > change. r8152 has no MDIO bus and accesses PHY through OCP registers.
>
> It does have an MDIO bus, it just used OCP registers to access it. But there is no
> reason you cannot export it using a normal Linux MDIO bus driver.
>
> > , and the MAC code also does PHY configuration,
>
> Which is a layer violation.
>
> > firmware patching, UPS and EEE handling that would need to move or
> > coexist with the phylib drivers.
>
> Or just thrown away because the it already works in the PHY drivers and
> phylink.
>
> > The two helpers here are self-contained and will be removed by the
> > phylink port.
>
> phylink_ethtool_nway_reset and phy_ethtool_nway_reset exists, so there is no
> need to add yet another version. phylink_ethtool_ksettings_set &
> phylink_ethtool_ksettings_get and phy_ethtool_get_link_ksettings &
> phy_ethtool_set_link_ksettings already exist, so there is no reason to add new
> ones.
>
> I'm tempted to just NACK it. The best pressure i can apply to make you move to
> phylink is to not merge your patches. If you are adding stuff which should really
> by done by phylink, i think it should be rejected.
>
> Andrew
I will keep the MII APIs.
Our read_mii_word() returns a negative errno when a PHY access fails, so that callers
can stop early. The MII library does not check for it: mii.c uses the mdio_read() return
value as register content, so mii_nway_restart() and mii_ethtool_get_link_ksettings()
treat an errno as bits.
When the errno is -ENODEV, nothing reaches the hardware. The device is gone, so the
write that follows in mii_nway_restart() is blocked as well. The remaining case is a PHY
access timeout, where the device is still accessible. The root cause there is also that
the MII library does not check the errno, so a future patch could make the MII APIs
check the mdio_read() return value and propagate the error. That would fix it for
every user instead of working around it in r8152.
Moving r8152 to phylink is the direction we want to take. We want to study the design
before committing to it, and it's a larger change than this series, so I would like to
separate it from this series.
Meanwhile, the legacy PHY access may cause bugs if the driver and the firmware
access the PHY at the same time. So, the TGPHY patch is needed for RTL8157/
RTL8159.
Best Regards,
Chih-Kai