Re: [PATCH v2 2/3] net: stmmac: Add glue layer for T-HEAD TH1520 SoC

[Drew Fustini <dfustini@tenstorrent.com>]

On Thu, Sep 26, 2024 at 08:32:00PM +0200, Andrew Lunn wrote:
> > +static int thead_dwmac_init(struct platform_device *pdev, void *priv)
> > +{
> > +	struct thead_dwmac *dwmac = priv;
> > +	int ret;
> > +
> > +	ret = thead_dwmac_set_phy_if(dwmac->plat);
> > +	if (ret)
> > +		return ret;
> > +
> > +	ret = thead_dwmac_set_txclk_dir(dwmac->plat);
> > +	if (ret)
> > +		return ret;
> > +
> > +	ret = regmap_write(dwmac->apb_regmap, GMAC_RXCLK_DELAY_CTRL,
> > +			   GMAC_RXCLK_DELAY_VAL(dwmac->rx_delay));
> > +	if (ret)
> > +		return dev_err_probe(dwmac->dev, ret,
> > +				     "failed to set GMAC RX clock delay\n");
> > +
> > +	ret = regmap_write(dwmac->apb_regmap, GMAC_TXCLK_DELAY_CTRL,
> > +			   GMAC_TXCLK_DELAY_VAL(dwmac->tx_delay));
> > +	if (ret)
> > +		return dev_err_probe(dwmac->dev, ret,
> > +				     "failed to set GMAC TX clock delay\n");
> > +
> > +	thead_dwmac_fix_speed(dwmac, SPEED_1000, 0);
> 
> Is this needed? I would expect this to be called when the PHY has link
> and you know the link speed. So why set it here?

Good point.  I've removed this line and the probe still completes okay
and the Ethernet connection is working okay.

> > +
> > +	return thead_dwmac_enable_clk(dwmac->plat);
> > +}
> > +
> > +static int thead_dwmac_probe(struct platform_device *pdev)
> > +{
> > +	struct device_node *np = pdev->dev.of_node;
> > +	struct stmmac_resources stmmac_res;
> > +	struct plat_stmmacenet_data *plat;
> > +	struct thead_dwmac *dwmac;
> > +	void __iomem *apb;
> > +	u32 delay;
> > +	int ret;
> > +
> > +	ret = stmmac_get_platform_resources(pdev, &stmmac_res);
> > +	if (ret)
> > +		return dev_err_probe(&pdev->dev, ret,
> > +				     "failed to get resources\n");
> > +
> > +	plat = devm_stmmac_probe_config_dt(pdev, stmmac_res.mac);
> > +	if (IS_ERR(plat))
> > +		return dev_err_probe(&pdev->dev, PTR_ERR(plat),
> > +				     "dt configuration failed\n");
> > +
> > +	dwmac = devm_kzalloc(&pdev->dev, sizeof(*dwmac), GFP_KERNEL);
> > +	if (!dwmac)
> > +		return -ENOMEM;
> > +
> > +	/* hardware default is 0 for the rx and tx internal clock delay */
> > +	dwmac->rx_delay = 0;
> > +	dwmac->tx_delay = 0;
> > +
> > +	/* rx and tx internal delay properties are optional */
> > +	if (!of_property_read_u32(np, "thead,rx-internal-delay", &delay)) {
> > +		if (delay > GMAC_RXCLK_DELAY_MASK)
> > +			dev_warn(&pdev->dev,
> > +				 "thead,rx-internal-delay (%u) exceeds max (%lu)\n",
> > +				 delay, GMAC_RXCLK_DELAY_MASK);
> > +		else
> > +			dwmac->rx_delay = delay;
> > +	}
> > +
> 
> So you keep going, with an invalid value? It is better to use
> dev_err() and return -EINVAL. The DT write will then correct their
> error when the device fails to probe.

My intention was to keep the default of 0 if the dt property exists and
exceeds the max value. I had considered failing the probe but I wasn't
sure that was too severe of a reaction to a bad value for the delay.

> 
> If you decide to keep this... I'm not sure these properties are
> needed.

Given your reply to the cover letter, I think it does make sense for me
to remove handling of these delay properties since the units of the
delay bit field are unknown and the hardware I have is okay with the
default delay.

> 
> > +MODULE_AUTHOR("Jisheng Zhang <jszhang@kernel.org>");
> 
> Please add a second author, if you have taken over this driver.

Yes, Jisheng is no longer working on it, so I will add myself.

Thanks,
Drew