On 06/10, Gregory CLEMENT wrote:

+
+struct tbg_def tbg[NUM_TBG] = {

const?

+	{"TBG-A-P", TBG_A_REFDIV, TBG_A_FBDIV, TBG_CTRL8, TBG_A_VCODIV_DIFF},
+	{"TBG-B-P", TBG_B_REFDIV, TBG_B_FBDIV, TBG_CTRL8, TBG_B_VCODIV_DIFF},
+	{"TBG-A-S", TBG_A_REFDIV, TBG_A_FBDIV, TBG_CTRL1, TBG_A_VCODIV_SE},
+	{"TBG-B-S", TBG_B_REFDIV, TBG_B_FBDIV, TBG_CTRL1, TBG_B_VCODIV_SE},
+};
+
+static struct clk_onecell_data clk_tbg_data;

Please allocate this in the driver probe.

+
+unsigned int tbg_get_mult(void __iomem *reg, struct tbg_def *ptbg)
+{
+	u32 val;
+
+	val = readl(reg + TBG_CTRL0);
+
+	return ((val >> ptbg->fbdiv_offset) & TBG_DIV_MASK) << 2;
+}
+
+unsigned int tbg_get_div(void __iomem *reg, struct tbg_def *ptbg)
+{
+	u32 val;
+	unsigned int div;
+
+	val = readl(reg + TBG_CTRL7);
+
+	div = (val >> ptbg->refdiv_offset) & TBG_DIV_MASK;
+	if (div == 0)
+		div = 1;
+	val = readl(reg + ptbg->vcodiv_reg);
+
+	div *= 1 << ((val >>  ptbg->vcodiv_offset) & TBG_DIV_MASK);
+
+	return div;
+}
+
+
+static int armada_3700_tbg_clock_probe(struct platform_device *pdev)
+{
+	struct device_node *np = pdev->dev.of_node;
+	struct device *dev = &pdev->dev;
+	const char *parent_name;
+	struct resource *res;
+	struct clk *parent;
+	void __iomem *reg;
+	int i, ret;
+
+	parent = of_clk_get(np, 0);

Why not devm_clk_get() instead? Don't make things dependent on DT
APIs unnecessarily please.

+	if (IS_ERR(parent)) {
+		dev_err(dev, "Could get the clock parent\n");
+		return -EINVAL;
+	}
+	parent_name = __clk_get_name(parent);
+
+	res = platform_get_resource(pdev, IORESOURCE_MEM, 0);
+	reg = devm_ioremap_resource(dev, res);
+	if (IS_ERR(reg)) {
+		dev_err(dev, "Could not map the tbg clock registers\n");

devm_ioremap_resource() already spits out an error on failure so
please drop this print.

+		ret = PTR_ERR(reg);
+		goto put_clk;
+	}
+
+	clk_tbg_data.clk_num = NUM_TBG;
+	clk_tbg_data.clks = devm_kcalloc(dev, clk_tbg_data.clk_num,
+					 sizeof(struct clk *), GFP_KERNEL);

Please move to clk_hw based registration.

+	if (!clk_tbg_data.clks) {
+		ret = -ENOMEM;
+		goto put_clk;
+	}
+
+	for (i = 0; i < NUM_TBG; i++) {
+		const char *name;
+		unsigned int mult, div;
+
+		name = tbg[i].name;
+		mult = tbg_get_mult(reg, &tbg[i]);
+		div = tbg_get_div(reg, &tbg[i]);
+		clk_tbg_data.clks[i] = clk_register_fixed_factor(NULL, name,
+						 parent_name, 0, mult, div);
+		if (IS_ERR(clk_tbg_data.clks[i]))
+			dev_err(dev, "Can't register TBG clock %s\n", name);
+	}
+
+	of_clk_add_provider(np, of_clk_src_onecell_get, &clk_tbg_data);

What if this fails?

+
+	return 0;
+
+put_clk:
+	clk_put(parent);
+	return ret;
+}

-- 
Qualcomm Innovation Center, Inc. is a member of Code Aurora Forum,
a Linux Foundation Collaborative Project