Re: [PATCH net-next v1 1/3] dt-bindings: net: Add Motorcomm yt8xxx ethernet phy Driver bindings

[Andrew Lunn]

On Wed, Jan 11, 2023 at 05:20:18PM +0800, Frank.Sae wrote:
> Hi Andrew,
> 
> On 2023/1/6 21:55, Andrew Lunn wrote:
> >>> Why is this needed? When the MAC driver connects to the PHY, it passes
> >>> phy-mode. For RGMII, this is one of:
> >>
> >>> linux/phy.h:	PHY_INTERFACE_MODE_RGMII,
> >>> linux/phy.h:	PHY_INTERFACE_MODE_RGMII_ID,
> >>> linux/phy.h:	PHY_INTERFACE_MODE_RGMII_RXID,
> >>> linux/phy.h:	PHY_INTERFACE_MODE_RGMII_TXID,
> >>>
> >>> This tells you if you need to add a delay for the RX clock line, the
> >>> TX clock line, or both. That is all you need to know for basic RGMII
> >>> delays.
> >>>
> >>
> >> This basic delay can be controlled by hardware or the phy-mode which
> >> passes from MAC driver.
> >> Default value depends on power on strapping, according to the voltage
> >> of RXD0 pin (low = 0, turn off;   high = 1, turn on).
> >>
> >> Add this for the case that This basic delay is controlled by hardware,
> >> and software don't change this.
> > 
> > You should always do what phy-mode contains. Always. We have had
> > problems in the past where a PHY driver ignored the phy-mode, and left
> > the PHY however it was strapped. Which worked. But developers put the
> > wrong phy-mode value in DT. Then somebody had a board which actually
> > required that the DT value really did work, because the strapping was
> > wrong. So the driver was fixed to respect the PHY mode, made that
> > board work, and broke all the other boards which had the wrong
> > phy-mode in DT.
> > 
> > If the user want the driver to leave the mode alone, use the
> > strapping, they should use PHY_INTERFACE_MODE_NA. It is not well
> > documented, but it is used in a few places. However, i don't recommend
> > it.
> > 
> 
> RX delay = rx-delay-basic (0ns or 1.9ns) + x-delay-additional-ps
> (N*150ps, N = 0 ~ 15)
>  If rx-delay-basic is removed and controlled by phy-mode.
>  when phy-mode is  rgmii-id or rgmii-rxid, RX delay is 1.9ns + N*150ps.
>  But sometimes 1.9ns is still too big, we just need  0ns + N*150ps.
> 
> For this case, can we do like following ?
> rx-internal-delay-ps:
>     enum: [ 0, 150, 300, 450, 600, 750, 900, 1050, 1200, 1350, 1500,
> 1650, 1800, 1900, 1950, 2050, 2100, 2200, 2250, 2350, 2500, 2650, 2800,
> 2950, 3100, 3250, 3400, 3550, 3700, 3850, 4000, 4150 ]
>     default: 0
>  rx-internal-delay-ps is 0ns + N*150ps and  1.9ns + N*150ps.
>  And check whether need rx-delay-basic (1.9ns) by the val of
> rx-internal-delay-ps?

Please take a look at phy_get_internal_delay() and the drivers which
use it.

    Andrew