Re: [PATCH v3 2/2] net: dsa: mxl-gsw1xx: Support R(G)MII slew rate configuration

[Vladimir Oltean <olteanv@gmail.com>]

On Tue, Jan 06, 2026 at 09:41:13AM +0000, Sverdlin, Alexander wrote:
> Hello Vladimir, Rob!
> 
> On Mon, 2026-01-05 at 22:00 +0100, Alexander Sverdlin wrote:
> > > > +	return regmap_update_bits(gsw1xx_priv->shell, GSW1XX_SHELL_RGMII_SLEW_CFG,
> > > > +				  RGMII_SLEW_CFG_DRV_TXD | RGMII_SLEW_CFG_DRV_TXC,
> > > > +				  (RGMII_SLEW_CFG_DRV_TXD | RGMII_SLEW_CFG_DRV_TXC) * rate);
> > > 
> > > I don't have a particularly strong EE background, but my understanding
> > > is this:
> > > 
> > > RGMII MACs provide individual slew rate configuration for TXD[3:0] and
> > > for TX_CLK because normally, you'd want to focus on the TX_CLK slew rate
> > > (in the sense of reducing EMI) more than on the TXD[3:0] slew rate.
> > > This is for 2 reasons:
> > > (1) the EMI noise produced by TX_CLK is in a much narrower spectrum
> > >     (runs at fixed 125/25/2.5 MHz) than TXD[3:0] (pseudo-random data).
> > > (2) reducing the slew rate for TXD[3:0] risks introducing inter-symbol
> > >     interference, risk which does not exist for TX_CLK
> > > 
> > > Your dt-binding does not permit configuring the slew rates separately,
> > > even though the hardware permits that. Was it intentional?
> > 
> > thanks for the hint! This is definitely something I need to discuss with HW
> > colleagues and get back to you!
> 
> Vladimir, according to the responsible HW colleague, it's OK and is desired
> to have TXD in "slow" as long as Setup-/Hold-Timing is in spec.
> 
> I do understand, that this is board-specific. Do you propose to introduce
> two separate properties?
> 
> Rob, in such case just "slew-rate" probably wouldn't fit any longer and
> I'd need to go back to "maxlinear,slew-rate-txd" and "maxlinear,slew-rate-txc"
> probably?

I see Rob has reviewed the binding in this form already, but I think the
rule of thumb that we could apply in this case is to still describe the
clock and data slew rates separately. Like Russell points out in a separate
thread, it's simpler to do this from the beginning rather than end up
with 3 properties you'd have to maintain, if you later need individual
control.
https://lore.kernel.org/netdev/aTB0x6JGcGUM04UX@shell.armlinux.org.uk/

Sadly I don't have the expertise to give any advice on how that would
translate into dt-bindings. Does it make sense to implement a full pin
controller device driver for the registers GPIO_DRIVE0_CFG -> RGMII_SLEW_CFG?