Il 04/08/25 16:05, Krzysztof Kozlowski ha scritto:

On 30/07/2025 12:56, Laura Nao wrote:

quoted

  
  	clk_mux->regmap = regmap;
+	clk_mux->regmap_hwv = regmap_hwv;
  	clk_mux->data = mux;
  	clk_mux->lock = lock;
  	clk_mux->hw.init = &init;

@@ -268,6 +329,7 @@ int mtk_clk_register_muxes(struct device *dev,
  			   struct clk_hw_onecell_data *clk_data)
  {
  	struct regmap *regmap;
+	struct regmap *regmap_hwv;
  	struct clk_hw *hw;
  	int i;
  

@@ -277,6 +339,13 @@ int mtk_clk_register_muxes(struct device *dev,
  		return PTR_ERR(regmap);
  	}
  
+	regmap_hwv = mtk_clk_get_hwv_regmap(node);
+	if (IS_ERR(regmap_hwv)) {

This is either buggy or fragile. mtk_clk_get_hwv_regmap() returns NULL
or valid pointer... or error? IS_ERR_OR_NULL is not the wait to go.

Choose one - IS_ERR or NULL, preferrably the first, since you must
handle deferred probe.

if regmap_hwv == NULL -> there is no HWV for *this* clock
if regmap_hwv == -ERROR -> there is a HWV for *this* clock, but something
                            went wrong, we have to return the error.

quoted

+		pr_err("Cannot find hardware voter regmap for %pOF: %pe\n",
+		       node, regmap_hwv);
+		return PTR_ERR(regmap_hwv);
+	}
+
  	for (i = 0; i < num; i++) {
  		const struct mtk_mux *mux = &muxes[i];
  

@@ -286,7 +355,7 @@ int mtk_clk_register_muxes(struct device *dev,
  			continue;
  		}
  
-		hw = mtk_clk_register_mux(dev, mux, regmap, lock);
+		hw = mtk_clk_register_mux(dev, mux, regmap, regmap_hwv, lock);

So NULL is passed and stored... are you sure this is 100% backwards
compatible?

Yes, it is. This got tested on multiple legacy SoCs in our lab.

Best regards,
Krzysztof