Re: [PATCH net-next v6 5/5] net: phy: add driver for built-in 2.5G ethernet PHY on MT7988

From: Daniel Golle
Date: Tue Jun 04 2024 - 09:50:35 EST


On Tue, Jun 04, 2024 at 08:42:57AM +0000, SkyLake Huang (黃啟澤) wrote:
> On Mon, 2024-06-03 at 19:30 +0100, Russell King (Oracle) wrote:
> >
> > External email : Please do not click links or open attachments until
> > you have verified the sender or the content.
> > On Mon, Jun 03, 2024 at 06:00:49PM +0100, Daniel Golle wrote:
> > > On Mon, Jun 03, 2024 at 04:47:28PM +0100, Russell King (Oracle)
> > wrote:
> > > > On Mon, Jun 03, 2024 at 03:52:19PM +0100, Daniel Golle wrote:
> > > > > On Mon, Jun 03, 2024 at 02:41:44PM +0100, Russell King (Oracle)
> > wrote:
> > > > > > On Mon, Jun 03, 2024 at 02:31:46PM +0100, Daniel Golle wrote:
> > > > > > > On Mon, Jun 03, 2024 at 02:25:01PM +0100, Russell King
> > (Oracle) wrote:
> > > > > > > > On Mon, Jun 03, 2024 at 08:18:34PM +0800, Sky Huang
> > wrote:
> > > > > > > > > Add support for internal 2.5Gphy on MT7988. This driver
> > will load
> > > > > > > > > necessary firmware, add appropriate time delay and
> > figure out LED.
> > > > > > > > > Also, certain control registers will be set to fix
> > link-up issues.
> > > > > > > >
> > > > > > > > Based on our previous discussion, it may be worth
> > checking in the
> > > > > > > > .config_init() method whether phydev->interface is one of
> > the
> > > > > > > > PHY interface modes that this PHY supports. As I
> > understand from one
> > > > > > > > of your previous emails, the possibilities are XGMII,
> > USXGMII or
> > > > > > > > INTERNAL. Thus:
> > > > > > > >
> > > > > > > > > +static int mt798x_2p5ge_phy_config_init(struct
> > phy_device *phydev)
> > > > > > > > > +{
> > > > > > > > > +struct pinctrl *pinctrl;
> > > > > > > > > +int ret;
> > > > > > > >
> > > > > > > > /* Check that the PHY interface type is compatible */
> > > > > > > > if (phydev->interface != PHY_INTERFACE_MODE_INTERNAL &&
> > > > > > > > phydev->interface != PHY_INTERFACE_MODE_XGMII &&
> > > > > > > > phydev->interface != PHY_INTERFACE_MODE_USXGMII)
> > > > > > > > return -ENODEV;
> > > > > > >
> > > > > > > The PHY is built-into the SoC, and as such the connection
> > type should
> > > > > > > always be "internal". The PHY does not exist as dedicated
> > IC, only
> > > > > > > as built-in part of the MT7988 SoC.
> > > > > >
> > > > > > That's not how it was described to me by Sky.
> > > > > >
> > > > > > If what you say is correct, then the implementation of
> > > > > > mt798x_2p5ge_phy_get_rate_matching() which checks for
> > interface modes
> > > > > > other than INTERNAL is not correct. Also it means that
> > config_init()
> > > > > > should not permit anything but INTERNAL.
> > > > >
> > > > > The way the PHY is connected to the MAC *inside the chip* is
> > XGMII
> > > > > according the MediaTek. So call it "internal" or "xgmii",
> > however, up to
> > > > > my knowledge it's a fact that there is **only one way** this
> > PHY is
> > > > > connected and used, and that is being an internal part of the
> > MT7988 SoC.
> > > > >
> > > > > Imho, as there are no actual XGMII signals exposed anywhere I'd
> > use
> > > > > "internal" to describe the link between MAC and PHY (which are
> > both
> > > > > inside the same chip package).
> > > >
> > > > I don't care what gets decided about what's acceptable for the
> > PHY to
> > > > accept, just that it checks for the acceptable modes in
> > .config_init()
> > > > and the .get_rate_matching() method is not checking for interface
> > > > modes that are not permitted.
> > >
> > > What I meant to express is that there is no need for such a check,
> > also
> > > not in config_init. There is only one way and one MAC-side
> > interface mode
> > > to operate that PHY, so the value will anyway not be considered
> > anywhere
> > > in the driver.
> >
> > No, it matters. With drivers using phylink, the PHY interface mode is
> > used in certain circumstances to constrain what the net device can
> > do.
> > So, it makes sense for new PHY drivers to ensure that the PHY
> > interface
> > mode is one that they can support, rather than just accepting
> > whatever
> > is passed to them (which then can lead to maintainability issues for
> > subsystems.)
> >
> > So, excuse me for disagreeing with you, but I do want to see such a
> > check in new PHY drivers.
> >
> > --
> > RMK's Patch system: https://www.armlinux.org.uk/developer/patches/
> > FTTP is here! 80Mbps down 10Mbps up. Decent connectivity at last!
>
> Hi Russell/Daniel,
> IMO, we can check PHY_INTERFACE_MODE_INTERNAL &
> PHY_INTERFACE_MODE_XGMII in config_init() or probe(). However,
> PHY_INTERFACE_MODE_USXGMII isn't supported by this phy, and
> drivers/net/ethernet/mediatek/mtk_eth_path.c uses
> PHY_INTERFACE_MODE_USXGMII to switch netsys pcs mux (set
> MUX_G2_USXGMII_SEL bit in TOP_MISC_NETSYS_PCS_MUX) so that XFI-MAC can
> be connected to external 10Gphy.
> So, basically, for 1st XFI-MAC on mt7988:
> - PHY_INTERFACE_MODE_XGMII/PHY_INTERFACE_MODE_INTERNAL: built-in
> 2.5Gphy

Why both? Wouldn't just PHY_INTERFACE_MODE_INTERNAL be more clear?
There is no XGMII interface exposed anywhere and both "internal" and
"xgmii" would be used to express the exact same thing.

> - PHY_INTERFACE_MODE_USXGMII: external 10Gphy
>
> I add check in config_init():
> /* Check if PHY interface type is compatible */
> if (phydev->interface != PHY_INTERFACE_MODE_XGMII &&
> phydev->interface != PHY_INTERFACE_MODE_INTERNAL)
> return -ENODEV;
>
> Also, test with different phy mode in dts:
> [PHY_INTERFACE_MODE_USXGMII]
> [ 18.702102] mtk_soc_eth 15100000.ethernet eth1: mtk_open: could not
> attach PHY: -19
> root@OpenWrt:/# cat /proc/device-tree/soc/ethernet@15100000/mac@1/phy-c
> onnection-type
> usxgmii
>
> [PHY_INTERFACE_MODE_INTERNAL]
> [ 18.329513] mtk_soc_eth 15100000.ethernet eth1: PHY [mdio-bus:0f]
> driver [MediaTek MT7988 2.5GbE PHY] (irq=POLL)
> [ 18.339708] mtk_soc_eth 15100000.ethernet eth1: configuring for
> phy/internal link mode
> root@OpenWrt:/# cat /proc/device-tree/soc/ethernet@15100000
> /mac@1/phy-connection-type
> internal
>
> Sky