Re: [PATCH 2/2] net: ethernet: stmicro: stmmac: dwmac-rk: Add rv1126 support

[Jagan Teki]

On Thu, 1 Sept 2022 at 17:42, Andrew Lunn <andrew@xxxxxxx> wrote:
>
> On Thu, Sep 01, 2022 at 12:56:09PM +0530, Jagan Teki wrote:
> > On Mon, 29 Aug 2022 at 18:40, Andrew Lunn <andrew@xxxxxxx> wrote:
> > >
> > > On Mon, Aug 29, 2022 at 06:50:42AM +0000, Anand Moon wrote:
> > > > Rockchip RV1126 has GMAC 10/100/1000M ethernet controller
> > > > via RGMII and RMII interfaces are configured via M0 and M1 pinmux.
> > > >
> > > > This patch adds rv1126 support by adding delay lines of M0 and M1
> > > > simultaneously.
> > >
> > > What does 'delay lines' mean with respect to RGMII?
> >
> > These are MAC receive clock delay lengths.
> >
> > >
> > > The RGMII signals need a 2ns delay between the clock and the data
> > > lines. There are three places this can happen:
> > >
> > > 1) In the PHY
> > > 2) Extra long lines on the PCB
> > > 3) In the MAC
> > >
> > > Generally, 1) is used, and controlled via phy-mode. A value of
> > > PHY_INTERFACE_MODE_RGMII_ID passed to the PHY driver means it will add
> > > these delays.
> > >
> > > You don't want both the MAC and the PHY adding delays.
> >
> > Yes, but these are specific to MAC, not related to PHY delays. Similar
> > to what is there in other Rockchip SoC families like RK3366, 3368,
> > 3399, 3128, but these MAC clock delay lengths are grouped based on the
> > iomux group in RV1126. We have iomux group 0 (M0) and group 1 (M1), so
> > the rgmii has to set these lengths irrespective of whether PHY add's
> > or not.
>
> So this is just fine tuning, in the order of pico seconds?
>
> If that is all it is, then this is fine. It becomes a problem when it
> is 2ns.

Yes, it is fine I think. We have tested the delay mentioned as per the
documentation.

tx_delay: Range value is 0~0x7F
rx_delay: Range value is 0~0x7F

Thanks,
Jagan.