mbox series

[0/3] arm64: mx93: etherrnet: set TX_CLK in RMII mode

Message ID 20240422-v6-9-topic-imx93-eqos-rmii-v1-0-30151fca43d2@pengutronix.de
Headers show
Series arm64: mx93: etherrnet: set TX_CLK in RMII mode | expand

Message

Steffen Trumtrar April 22, 2024, 8:46 a.m. UTC
This series adds support for setting the TX_CLK direction in the eQOS
ethernet core on the i.MX93 when RMII mode is used.

According to AN14149, when the i.MX93 ethernet controller is used in
RMII mode, the TX_CLK *must* be set to output mode. 

Add a devicetree property with the register to set this bit and parse it
in the dwmac-imx driver.

Signed-off-by: Steffen Trumtrar <s.trumtrar@pengutronix.de>
---
Steffen Trumtrar (3):
      dt-bindings: net: mx93: add enet_clk_sel binding
      arm64: dts: imx93: add enet_clk_sel
      net: stmicro: imx: set TX_CLK direction in RMII mode

 .../devicetree/bindings/net/nxp,dwmac-imx.yaml     | 10 ++++++++
 arch/arm64/boot/dts/freescale/imx93.dtsi           |  1 +
 drivers/net/ethernet/stmicro/stmmac/dwmac-imx.c    | 27 ++++++++++++++++++++++
 3 files changed, 38 insertions(+)
---
base-commit: 4cece764965020c22cff7665b18a012006359095
change-id: 20240422-v6-9-topic-imx93-eqos-rmii-3a2cb421c81d

Best regards,

Comments

Ahmad Fatoum April 22, 2024, 8:51 a.m. UTC | #1
Hello Steffen,

On 22.04.24 10:46, Steffen Trumtrar wrote:
> +		dwmac->enet_clk_regmap = syscon_regmap_lookup_by_phandle(np, "enet_clk_sel");
> +		if (IS_ERR(dwmac->enet_clk_regmap))
> +			return PTR_ERR(dwmac->enet_clk_regmap);
> +
> +		err = of_property_read_u32_index(np, "enet_clk_sel", 1, &dwmac->enet_clk_reg_off);
> +		if (err) {
> +			dev_err(dev, "Can't get enet clk sel reg offset (%d)\n", err);
> +			return err;
> +		}

This looks like the property is not optional. The property needs to stay optional
as not to break backwards compatibility with older device trees.

Cheers,
Ahmad

>  	}
>  
>  	return err;
>
Marc Kleine-Budde April 22, 2024, 9:04 a.m. UTC | #2
On 22.04.2024 10:46:19, Steffen Trumtrar wrote:
> In case of RMII connection, the TX_CLK must be set to output direction.
> Parse the register and offset from the devicetree and set the direction
> of the TX_CLK when the property was provided.
> 
> Signed-off-by: Steffen Trumtrar <s.trumtrar@pengutronix.de>
> ---
>  drivers/net/ethernet/stmicro/stmmac/dwmac-imx.c | 27 +++++++++++++++++++++++++
>  1 file changed, 27 insertions(+)
> 
> diff --git a/drivers/net/ethernet/stmicro/stmmac/dwmac-imx.c b/drivers/net/ethernet/stmicro/stmmac/dwmac-imx.c
> index 6b65420e11b5c..0fc81a626a664 100644
> --- a/drivers/net/ethernet/stmicro/stmmac/dwmac-imx.c
> +++ b/drivers/net/ethernet/stmicro/stmmac/dwmac-imx.c
> @@ -37,6 +37,9 @@
>  #define MX93_GPR_ENET_QOS_INTF_SEL_RGMII	(0x1 << 1)
>  #define MX93_GPR_ENET_QOS_CLK_GEN_EN		(0x1 << 0)
>  
> +#define MX93_GPR_ENET_QOS_TX_CLK_SEL_MASK	GENMASK(1, 1)
> +#define MX93_GPR_ENET_QOS_TX_CLK_SEL		(0x1 << 1)
> +
>  #define DMA_BUS_MODE			0x00001000
>  #define DMA_BUS_MODE_SFT_RESET		(0x1 << 0)
>  #define RMII_RESET_SPEED		(0x3 << 14)
> @@ -57,7 +60,9 @@ struct imx_priv_data {
>  	struct clk *clk_tx;
>  	struct clk *clk_mem;
>  	struct regmap *intf_regmap;
> +	struct regmap *enet_clk_regmap;
>  	u32 intf_reg_off;
> +	u32 enet_clk_reg_off;
>  	bool rmii_refclk_ext;
>  	void __iomem *base_addr;
>  
> @@ -116,6 +121,18 @@ static int imx93_set_intf_mode(struct plat_stmmacenet_data *plat_dat)
>  		break;
>  	case PHY_INTERFACE_MODE_RMII:
>  		val = MX93_GPR_ENET_QOS_INTF_SEL_RMII;
> +
> +		/* According to NXP AN14149, the direction of the
> +		 * TX_CLK must be set to output in RMII mode.
> +		 */
> +		if (dwmac->enet_clk_regmap)
> +			regmap_update_bits(dwmac->enet_clk_regmap,
> +					   dwmac->enet_clk_reg_off,
> +					   MX93_GPR_ENET_QOS_TX_CLK_SEL_MASK,
> +					   MX93_GPR_ENET_QOS_TX_CLK_SEL);

Please add error handling.

> +		else
> +			dev_warn(dwmac->dev, "TX_CLK can't be set to output mode.\n");

As far as I understand the AN14149, setting the TX_CLK_SEL is mandatory
for PHY_INTERFACE_MODE_RMII. IMHO this should be error, shouldn't it?

> +
>  		break;
>  	case PHY_INTERFACE_MODE_RGMII:
>  	case PHY_INTERFACE_MODE_RGMII_ID:
> @@ -310,6 +327,16 @@ imx_dwmac_parse_dt(struct imx_priv_data *dwmac, struct device *dev)
>  			dev_err(dev, "Can't get intf mode reg offset (%d)\n", err);
>  			return err;
>  		}
> +
> +		dwmac->enet_clk_regmap = syscon_regmap_lookup_by_phandle(np, "enet_clk_sel");
> +		if (IS_ERR(dwmac->enet_clk_regmap))
> +			return PTR_ERR(dwmac->enet_clk_regmap);
> +
> +		err = of_property_read_u32_index(np, "enet_clk_sel", 1, &dwmac->enet_clk_reg_off);
> +		if (err) {
> +			dev_err(dev, "Can't get enet clk sel reg offset (%d)\n", err);
> +			return err;
> +		}

regards,
Marc
Krzysztof Kozlowski April 22, 2024, 9:12 a.m. UTC | #3
On 22/04/2024 10:46, Steffen Trumtrar wrote:
> In case of RMII connection, the TX_CLK must be set to output direction.
> Parse the register and offset from the devicetree and set the direction
> of the TX_CLK when the property was provided.
> 
> Signed-off-by: Steffen Trumtrar <s.trumtrar@pengutronix.de>
> ---
>  drivers/net/ethernet/stmicro/stmmac/dwmac-imx.c | 27 +++++++++++++++++++++++++
>  1 file changed, 27 insertions(+)
> 
> diff --git a/drivers/net/ethernet/stmicro/stmmac/dwmac-imx.c b/drivers/net/ethernet/stmicro/stmmac/dwmac-imx.c
> index 6b65420e11b5c..0fc81a626a664 100644
> --- a/drivers/net/ethernet/stmicro/stmmac/dwmac-imx.c
> +++ b/drivers/net/ethernet/stmicro/stmmac/dwmac-imx.c
> @@ -37,6 +37,9 @@
>  #define MX93_GPR_ENET_QOS_INTF_SEL_RGMII	(0x1 << 1)
>  #define MX93_GPR_ENET_QOS_CLK_GEN_EN		(0x1 << 0)
>  
> +#define MX93_GPR_ENET_QOS_TX_CLK_SEL_MASK	GENMASK(1, 1)
> +#define MX93_GPR_ENET_QOS_TX_CLK_SEL		(0x1 << 1)
> +
>  #define DMA_BUS_MODE			0x00001000
>  #define DMA_BUS_MODE_SFT_RESET		(0x1 << 0)
>  #define RMII_RESET_SPEED		(0x3 << 14)
> @@ -57,7 +60,9 @@ struct imx_priv_data {
>  	struct clk *clk_tx;
>  	struct clk *clk_mem;
>  	struct regmap *intf_regmap;
> +	struct regmap *enet_clk_regmap;
>  	u32 intf_reg_off;
> +	u32 enet_clk_reg_off;
>  	bool rmii_refclk_ext;
>  	void __iomem *base_addr;
>  
> @@ -116,6 +121,18 @@ static int imx93_set_intf_mode(struct plat_stmmacenet_data *plat_dat)
>  		break;
>  	case PHY_INTERFACE_MODE_RMII:
>  		val = MX93_GPR_ENET_QOS_INTF_SEL_RMII;
> +
> +		/* According to NXP AN14149, the direction of the
> +		 * TX_CLK must be set to output in RMII mode.
> +		 */
> +		if (dwmac->enet_clk_regmap)
> +			regmap_update_bits(dwmac->enet_clk_regmap,
> +					   dwmac->enet_clk_reg_off,
> +					   MX93_GPR_ENET_QOS_TX_CLK_SEL_MASK,
> +					   MX93_GPR_ENET_QOS_TX_CLK_SEL);
> +		else
> +			dev_warn(dwmac->dev, "TX_CLK can't be set to output mode.\n");
> +
>  		break;
>  	case PHY_INTERFACE_MODE_RGMII:
>  	case PHY_INTERFACE_MODE_RGMII_ID:
> @@ -310,6 +327,16 @@ imx_dwmac_parse_dt(struct imx_priv_data *dwmac, struct device *dev)
>  			dev_err(dev, "Can't get intf mode reg offset (%d)\n", err);
>  			return err;
>  		}
> +
> +		dwmac->enet_clk_regmap = syscon_regmap_lookup_by_phandle(np, "enet_clk_sel");
> +		if (IS_ERR(dwmac->enet_clk_regmap))
> +			return PTR_ERR(dwmac->enet_clk_regmap);

This looks like breaking ABI. Please test your changes without the DTS.
Does the DTS pass dtbs_check? Does the driver probe correctly with such DTS?

Best regards,
Krzysztof
Sébastien Szymanski April 22, 2024, 9:28 a.m. UTC | #4
Hello,

On 4/22/24 10:46, Steffen Trumtrar wrote:
> This series adds support for setting the TX_CLK direction in the eQOS
> ethernet core on the i.MX93 when RMII mode is used.
> 
> According to AN14149, when the i.MX93 ethernet controller is used in
> RMII mode, the TX_CLK *must* be set to output mode.

Must ? I don't think that is true. Downstream NXP kernel has an option 
to set TX_CLK as an input:

https://github.com/nxp-imx/linux-imx/blob/lf-6.6.y/Documentation/devicetree/bindings/net/nxp%2Cdwmac-imx.yaml#L69

https://github.com/nxp-imx/linux-imx/commit/fbc17f6f7919d03c275fc48b0400c212475b60ec

Regards,

> 
> Add a devicetree property with the register to set this bit and parse it
> in the dwmac-imx driver.
> 
> Signed-off-by: Steffen Trumtrar <s.trumtrar@pengutronix.de>
> ---
> Steffen Trumtrar (3):
>        dt-bindings: net: mx93: add enet_clk_sel binding
>        arm64: dts: imx93: add enet_clk_sel
>        net: stmicro: imx: set TX_CLK direction in RMII mode
> 
>   .../devicetree/bindings/net/nxp,dwmac-imx.yaml     | 10 ++++++++
>   arch/arm64/boot/dts/freescale/imx93.dtsi           |  1 +
>   drivers/net/ethernet/stmicro/stmmac/dwmac-imx.c    | 27 ++++++++++++++++++++++
>   3 files changed, 38 insertions(+)
> ---
> base-commit: 4cece764965020c22cff7665b18a012006359095
> change-id: 20240422-v6-9-topic-imx93-eqos-rmii-3a2cb421c81d
> 
> Best regards,
Steffen Trumtrar April 23, 2024, 6:43 a.m. UTC | #5
Hi,

On 2024-04-22 at 11:28 +02, Sébastien Szymanski <sebastien.szymanski@armadeus.com> wrote: 
 
> Hello,  On 4/22/24 10:46, Steffen Trumtrar wrote: 
> > This series adds support for setting the TX_CLK direction in the eQOS ethernet core on the i.MX93 when RMII mode is used.  According to AN14149, when the i.MX93 ethernet controller is used in RMII mode, the TX_CLK *must* be set to output mode. 
>  Must ? I don't think that is true. Downstream NXP kernel has an option to set TX_CLK as an input: 
>

re-reading that passage again, it only says "other registers that must be configured" and that the ENET_QOS_CLK_TX_CLK_SEL bit "is 0b1" for RMII. So, maybe you are right. 


Thanks,
Steffen

> https://github.com/nxp-imx/linux-imx/blob/lf-6.6.y/Documentation/devicetree/bindings/net/nxp%2Cdwmac-imx.yaml#L69
> 
> https://github.com/nxp-imx/linux-imx/commit/fbc17f6f7919d03c275fc48b0400c212475b60ec
>