On 07/02/2026 09:45, Jiayu Du wrote:

On Fri, Feb 06, 2026 at 02:26:40PM +0100, Krzysztof Kozlowski wrote:

quoted

On 04/02/2026 09:29, Jiayu Du wrote:

quoted

+static int dwcmshc_k230_init(struct device *dev, struct sdhci_host *host,
+			     struct dwcmshc_priv *dwc_priv)
+{
+	static const char * const clk_ids[] = {"base", "timer", "ahb"};
+	struct device_node *usb_phy_node;
+	struct k230_priv *k230_priv;
+	u32 data;
+	int ret;
+
+	k230_priv = devm_kzalloc(dev, sizeof(struct k230_priv), GFP_KERNEL);
+	if (!k230_priv)
+		return -ENOMEM;
+	dwc_priv->priv = k230_priv;
+
+	usb_phy_node = of_find_compatible_node(NULL, NULL, "canaan,k230-usb-phy");

Hm? You should use phandles, not look for various nodes.

Only one usbphy node has the canaan, k230-usb-phy compatibility.
So in this situation, is it ok to continue using of_find_compatible_node?

Amount of nodes does not matter. This is not how you express
links/dependencies between devices. Phandle is for this. This is wrong
on many levels, including missing device links, bypassing kernel API/layers.

quoted

quoted

+	if (!usb_phy_node) {

Please follow Linux coding style.

I will fix it in next version.

quoted

quoted

+		return dev_err_probe(dev, -ENODEV,
+				     "Failed to find k230-usb-phy node\n");
+	}
+
+	k230_priv->hi_sys_regmap = device_node_to_regmap(usb_phy_node);
+	of_node_put(usb_phy_node);
+	if (IS_ERR(k230_priv->hi_sys_regmap)) {
+		return dev_err_probe(dev, PTR_ERR(k230_priv->hi_sys_regmap),
+				     "Failed to get k230-usb-phy regmap\n");
+	}
+
+	ret = dwcmshc_get_enable_other_clks(mmc_dev(host->mmc), dwc_priv,
+					    ARRAY_SIZE(clk_ids), clk_ids);
+	if (ret) {
+		return dev_err_probe(dev, ret,
+				     "Failed to get/enable k230 mmc other clocks\n");
+	}
+
+	if (of_device_is_compatible(dev->of_node, "canaan,k230-sdio")) {

Driver match data is for this.

What you mean is that I shouldn't use of_find_compatible_node, but I can
use device_get_match_data instead? Then I can continue to distinguish
between SDIO and eMMC to do parameter configuration

Or do you mean that I should put the parameters to be adjusted into the
pdata structure? But currently, the dwcmshc structure is not suitable for
containing vendor-specific properties.

Parameters should go to driver match data. I already requested this for
some other driver and this has to be fixed.

Best regards,
Krzysztof