On 11/18/2017 5:26 AM, Stephen Boyd wrote:

On 11/17, Tirupathi Reddy wrote:

quoted

diff --git a/Documentation/devicetree/bindings/clock/clk-spmi-pmic-div.txt b/Documentation/devicetree/bindings/clock/clk-spmi-pmic-div.txt
new file mode 100644
index 0000000..2cf2aba
--- /dev/null
+++ b/Documentation/devicetree/bindings/clock/clk-spmi-pmic-div.txt

Please move this to qcom,spmi-clkdiv.txt like other qcom bindings.

Addressed in next patch version[7]

quoted

@@ -0,0 +1,59 @@
+Qualcomm Technologies, Inc. SPMI PMIC clock divider (clkdiv)
+
+clkdiv configures the clock frequency of a set of outputs on the PMIC.
+These clocks are typically wired through alternate functions on
+gpio pins.
+
+=======================
+Properties
+=======================
+
+- compatible
+	Usage:      required
+	Value type: <string>
+	Definition: must be "qcom,spmi-clkdiv".
+
+- reg
+	Usage:      required
+	Value type: <prop-encoded-array>
+	Definition: base address of CLKDIV peripherals.
+
+- qcom,num-clkdivs
+	Usage:      required
+	Value type: <u32>
+	Definition: number of CLKDIV peripherals.

Would it work if we read the registers and looked for the clkdiv
subtype ID? If we read something that doesn't match, break and
stop adding clks? Or does reading the next "peripheral" cause
some sort of crash and burn scenario where the device see an
access control violation? Would be interesting to do it that way
and avoid needing a new property in DT.

I feel it is better to go with the current design as we beforehand know the
#div-clk peripherals in a PMIC. Reading the next peripheral might result in
an unknown failure if there is no real peripheral there (PMIC peripherals
are not in contiguous address space).

quoted

+
+
+	parent_name = of_clk_get_parent_name(of_node, 0);
+	if (!parent_name) {
+		dev_err(dev, "missing parent clock\n");
+		return -ENODEV;
+	}
+
+	init.parent_names = &parent_name;
+	init.num_parents = 1;
+	init.ops = &clk_spmi_pmic_div_ops;
+
+	for (i = 0, clkdiv = cc->clks; i < nclks; i++) {
+		spin_lock_init(&clkdiv[i].lock);
+		clkdiv[i].base = start + i * 0x100;
+		clkdiv[i].regmap = regmap;
+		clkdiv[i].cxo_period_ns = NSEC_PER_SEC / cxo_hz;
+		init.name = kasprintf(GFP_KERNEL, "div_clk%d", i + 1);

Hmm. Maybe we should just have this on the stack. 20 characters
should be plenty?

Addressed in next patch version[7]

quoted

+		clkdiv[i].hw.init = &init;
+
+		ret = devm_clk_hw_register(dev, &clkdiv[i].hw);
+		kfree(init.name); /* clk framework made a copy */
+		if (ret)
+			return ret;
+	}
+
+	return of_clk_add_hw_provider(of_node, spmi_pmic_div_clk_hw_get, cc);
+}
+
+static int spmi_pmic_clkdiv_remove(struct platform_device *pdev)
+{
+	of_clk_del_provider(pdev->dev.of_node);
+
+	return 0;

This can use devm now.

Addressed in next patch version.

quoted

+}
+
+static const struct of_device_id spmi_pmic_clkdiv_match_table[] = {
+	{ .compatible = "qcom,spmi-clkdiv" },
+	{ /* sentinel */ }
+};

Nice!

Thanks