Hi Geert,
Thanks for your feedback.
On 2024-05-13 11:39:54 +0200, Geert Uytterhoeven wrote:
> Hi Niklas, Andrew,
>
> On Thu, May 9, 2024 at 11:10 PM Niklas Söderlund
> <niklas.soderlund+renesas@xxxxxxxxxxxx> wrote:
> > Add initial support for Renesas Ethernet-TSN End-station device of R-Car
> > V4H. The Ethernet End-station can connect to an Ethernet network using a
> > 10 Mbps, 100 Mbps, or 1 Gbps full-duplex link via MII/GMII/RMII/RGMII.
> > Depending on the connected PHY.
> >
> > The driver supports Rx checksum and offload and hardware timestamps.
> >
> > While full power management and suspend/resume is not yet supported the
> > driver enables runtime PM in order to enable the module clock. While
> > explicit clock management using clk_enable() would suffice for the
> > supported SoC, the module could be reused on SoCs where the module is
> > part of a power domain.
> >
> > Signed-off-by: Niklas Söderlund <niklas.soderlund+renesas@xxxxxxxxxxxx>
> > ---
> > * Changes since v4
> > - Enable GPOUT_RDM and GPOUT_TDM delays depending on phy-mode.
>
> Thanks for the update!
>
> > +static void rtsn_set_delay_mode(struct rtsn_private *priv)
> > +{
> > + u32 val = 0;
> > +
> > + /* The MAC is capable of applying a delay on both Rx and Tx. Each
> > + * delay can either be on or off, there is no way to set its length.
> > + *
> > + * The exact delay applied depends on electric characteristics of the
> > + * board. The datasheet describes a typical Rx delay of 1800 ps and a
> > + * typical Tx delay of 2000 ps.
> > + *
> > + * There are boards where the RTSN device is used together with PHYs
> > + * who do not support a large enough internal delays to function. These
> > + * boards depends on the MAC applying these inexact delays.
> > + */
> > +
> > + /* If the phy-mode is rgmii or rgmii-rxid apply Rx delay on the MAC */
> > + if (priv->iface == PHY_INTERFACE_MODE_RGMII_ID ||
> > + priv->iface == PHY_INTERFACE_MODE_RGMII_RXID)
> > + val |= GPOUT_RDM;
> > +
> > + /* If the phy-mode is rgmii or rgmii-txid apply Tx delay on the MAC */
> > + if (priv->iface == PHY_INTERFACE_MODE_RGMII_ID ||
> > + priv->iface == PHY_INTERFACE_MODE_RGMII_TXID)
> > + val |= GPOUT_TDM;
> > +
> > + rtsn_write(priv, GPOUT, val);
> > +}
>
> > +static int rtsn_phy_init(struct rtsn_private *priv)
> > +{
> > + struct device_node *np = priv->ndev->dev.parent->of_node;
> > + struct phy_device *phydev;
> > + struct device_node *phy;
> > + phy_interface_t iface;
> > +
> > + /* Delays, if any, are applied by the MAC. Mask RGMII mode passed to the
> > + * PHY to avoid it also adding the delay.
> > + */
> > + switch (priv->iface) {
> > + case PHY_INTERFACE_MODE_RGMII:
> > + case PHY_INTERFACE_MODE_RGMII_ID:
> > + case PHY_INTERFACE_MODE_RGMII_RXID:
> > + case PHY_INTERFACE_MODE_RGMII_TXID:
> > + iface = PHY_INTERFACE_MODE_RGMII;
> > + break;
> > + default:
> > + iface = priv->iface;
> > + break;
> > + }
>
> This introduces the same issues (the "workaround" state below) we had
> with ravb before.
> 9b23203c32ee02cd ("ravb: Mask PHY mode to avoid inserting delays twice")
> was the workaround,
> a6f51f2efa742df0 ("ravb: Add support for explicit internal clock delay
> configuration")
> was the final fix.
>
> Do we really want to repeat that mistake?
Is it the same issue? The RAVB issue is around PHY drivers adjusting
delays based on [rt]xc-skew-ps properties. The RTSN bindings only deal
with {rx,tx}-internal-delay-ps properties.
After a discussion with Andrew my understanding is that the PHY shall
not attempt to add any delays from {rx,tx}-internal-delay-ps properties
if the phy-mode used in of_phy_connect() is PHY_INTERFACE_MODE_RGMII. As
we mask the phy-mode here the PHY shall never attempt to add delays as
we deal with that in the MAC.
It feels like I missed something? Sorry if I'm confused.
>
> Gr{oetje,eeting}s,
>
> Geert
>
> --
> Geert Uytterhoeven -- There's lots of Linux beyond ia32 -- geert@xxxxxxxxxxxxxx
>
> In personal conversations with technical people, I call myself a hacker. But
> when I'm talking to journalists I just say "programmer" or something like that.
> -- Linus Torvalds
--
Kind Regards,
Niklas Söderlund
