Re: [PATCH net-next] net: ethernet: mtk_eth_soc: add paths and SerDes modes for MT7988

From: Simon Horman
Date: Sat Sep 16 2023 - 14:06:23 EST


On Tue, Sep 12, 2023 at 03:54:40PM +0100, Daniel Golle wrote:
> MT7988 comes with a built-in 2.5G PHY as well as SerDes lanes to
> connect external PHYs or transceivers in USXGMII, 10GBase-R, 5GBase-R,
> 2500Base-X, 1000Base-X and Cisco SGMII interface modes.
>
> Implement support for configuring for the new paths to SerDes interfaces
> and the internal 2.5G PHY.
>
> Add USXGMII PCS driver for 10GBase-R, 5GBase-R and USXGMII mode, and
> setup the new PHYA on MT7988 to access the also still existing old
> LynxI PCS for 1000Base-X, 2500Base-X and Cisco SGMII interface modes.
>
> Signed-off-by: Daniel Golle <daniel@xxxxxxxxxxxxxx>
> ---
> Changes since RFC v2:
> * read PCS_RX_STATUS0 register to avoid bogus link-up
> (recommended by mtk devs)
> * use parenthese to fix evaluation order
> (Simon Horman reported clang warning)
> * fix allocation size of usxgmii_pcs
> (Simon Horman reported Smatch warning)
> * always set USXGMII_AN_ENABLE bit in USXGMII mode
> (it's what the vendor driver does and USXGMII doesn't seem to work
> at all otherwise, we may need to manually set rate matching registers
> if we don't use AN, but this isn't implemented at this point)
>
> Changes since initial RFC:
> * set missing neg_mode = true for usxgmii pcs
> * use phylink_decode_usxgmii_word instead of open coding
>
> drivers/net/ethernet/mediatek/Kconfig | 16 +
> drivers/net/ethernet/mediatek/Makefile | 1 +
> drivers/net/ethernet/mediatek/mtk_eth_path.c | 123 +++-
> drivers/net/ethernet/mediatek/mtk_eth_soc.c | 182 ++++-
> drivers/net/ethernet/mediatek/mtk_eth_soc.h | 232 +++++-
> drivers/net/ethernet/mediatek/mtk_usxgmii.c | 702 +++++++++++++++++++
> 6 files changed, 1225 insertions(+), 31 deletions(-)
> create mode 100644 drivers/net/ethernet/mediatek/mtk_usxgmii.c
>
> diff --git a/drivers/net/ethernet/mediatek/Kconfig b/drivers/net/ethernet/mediatek/Kconfig
> index da0db417ab690..b942b4622d146 100644
> --- a/drivers/net/ethernet/mediatek/Kconfig
> +++ b/drivers/net/ethernet/mediatek/Kconfig
> @@ -25,6 +25,22 @@ config NET_MEDIATEK_SOC
> This driver supports the gigabit ethernet MACs in the
> MediaTek SoC family.
>
> +config NET_MEDIATEK_SOC_USXGMII
> + bool "Support USXGMII SerDes on MT7988"
> + depends on (ARCH_MEDIATEK && ARM64) || COMPILE_TEST
> + def_bool NET_MEDIATEK_SOC != n
> + help
> + Include support for 10GE SerDes which can be found on MT7988.
> + If this kernel should run on SoCs with 10 GBit/s Ethernet you
> + will need to select this option to use GMAC2 and GMAC3 with
> + external PHYs, SFP(+) cages in 10GBase-R, 5GBase-R or USXGMII
> + interface modes.
> +
> + Note that as the 2500Base-X/1000Base-X/Cisco SGMII SerDes PCS
> + unit (MediaTek LynxI) in MT7988 is connected via the new 10GE
> + SerDes, you will also need to select this option in case you
> + want to use any of those SerDes modes.
> +
> config NET_MEDIATEK_STAR_EMAC
> tristate "MediaTek STAR Ethernet MAC support"
> select PHYLIB

...

> diff --git a/drivers/net/ethernet/mediatek/mtk_usxgmii.c b/drivers/net/ethernet/mediatek/mtk_usxgmii.c

...

> +static void mtk_usxgmii_pcs_get_state(struct phylink_pcs *pcs,
> + struct phylink_link_state *state)
> +{
> + struct mtk_usxgmii_pcs *mpcs = pcs_to_mtk_usxgmii_pcs(pcs);
> + struct mtk_eth *eth = mpcs->eth;
> + struct mtk_mac *mac = eth->mac[mtk_xgmii2mac_id(eth, mpcs->id)];
> + u32 val = 0;
> +
> + regmap_read(mpcs->regmap, RG_PCS_AN_CTRL0, &val);
> + if (FIELD_GET(USXGMII_AN_ENABLE, val)) {
> + /* Refresh LPA by inverting LPA_LATCH */
> + regmap_read(mpcs->regmap, RG_PCS_AN_STS0, &val);
> + regmap_update_bits(mpcs->regmap, RG_PCS_AN_STS0,
> + USXGMII_LPA_LATCH,
> + !(val & USXGMII_LPA_LATCH));
> +
> + regmap_read(mpcs->regmap, RG_PCS_AN_STS0, &val);
> +
> + phylink_decode_usxgmii_word(state, FIELD_GET(USXGMII_PCS_AN_WORD,
> + val));
> +
> + state->interface = mpcs->interface;
> + } else {
> + val = mtk_r32(mac->hw, MTK_XGMAC_STS(mac->id));
> +
> + if (mac->id == MTK_GMAC2_ID)
> + val >>= 16;
> +
> + switch (FIELD_GET(MTK_USXGMII_PCS_MODE, val)) {
> + case 0:
> + state->speed = SPEED_10000;
> + break;
> + case 1:
> + state->speed = SPEED_5000;
> + break;
> + case 2:
> + state->speed = SPEED_2500;
> + break;
> + case 3:
> + state->speed = SPEED_1000;
> + break;
> + }
> +
> + state->interface = mpcs->interface;
> + state->link = FIELD_GET(MTK_USXGMII_PCS_LINK, val);
> + state->duplex = DUPLEX_FULL;
> + }
> +
> + /* Refresh USXGMII link status by toggling RG_PCS_RX_STATUS_UPDATE */
> + regmap_set_bits(mpcs->regmap, RG_PCS_RX_STATUS0, RG_PCS_RX_STATUS_UPDATE);
> + ndelay(1020);
> + regmap_clear_bits(mpcs->regmap, RG_PCS_RX_STATUS0, RG_PCS_RX_STATUS_UPDATE);
> + ndelay(1020);
> +
> + /* Read USXGMII link status */
> + regmap_read(mpcs->regmap, RG_PCS_RX_STATUS0, &val);
> + state->link = state->link && FIELD_GET(RG_PCS_RX_LINK_STATUS, val);
> +
> + /* Continuously repeat re-configuration sequence until link comes up */
> + if (!state->link)
> + mtk_usxgmii_pcs_config(pcs, mpcs->neg_mode,
> + state->interface, NULL, false);
> +}

Hi Daniel,

I feel that I'm missing something obvious, but with this patch applied,
when building with allmodconfig for both x86_64 and arm64 the build fails
reporting that RG_PCS_RX_STATUS0, RG_PCS_RX_STATUS_UPDATE and
RG_PCS_RX_LINK_STATUS are undefined.

...