On Wed 12 Aug 14:57 PDT 2015, Stephen Boyd wrote:

On 08/12/2015 01:18 PM, Bjorn Andersson wrote:

quoted

On Tue 11 Aug 15:49 PDT 2015, Stephen Boyd wrote:

quoted

On 07/08, Rajendra Nayak wrote:

quoted

diff --git a/drivers/clk/qcom/gcc-msm8960.c b/drivers/clk/qcom/gcc-msm8960.c
index eb6a4f9..2c80d03 100644
--- a/drivers/clk/qcom/gcc-msm8960.c
+++ b/drivers/clk/qcom/gcc-msm8960.c

@@ -15,6 +15,7 @@
  #include <linux/bitops.h>
  #include <linux/err.h>
  #include <linux/platform_device.h>
+#include <linux/of_platform.h>
  #include <linux/module.h>
  #include <linux/of.h>
  #include <linux/of_device.h>

@@ -3520,7 +3521,8 @@ static int gcc_msm8960_probe(struct platform_device *pdev)
  	if (IS_ERR(clk))
  		return PTR_ERR(clk);
  
-	return qcom_cc_probe(pdev, match->data);
+	qcom_cc_probe(pdev, match->data);
+	return of_platform_populate(pdev->dev.of_node, NULL, NULL, &pdev->dev);

We just lost the error code from qcom_cc_probe()...

Also, I don't like having a subnode in DT. Why can't we use the
same node as the GCC node and create a virtual child device here
for tsens? We can assign the same of_node that this platform
device has so that DT keeps working correctly.

Can't we make the gcc driver support being a child of a simple-mfd by
having it attempting to acquire the regmap of the parent and falling
back to creating its own mmio regmap?

So we would need to make a new device and driver for the simple-mfd 
parent?

No that part is already in mainline, so my idea was that we add an early
return in qcom_cc_map() like:

	 struct device *dev = &pdev->dev;
+        struct regmap *map;
+
+        map = syscon_node_to_regmap(dev->parent->of_node);
+        if (!IS_ERR(map))
+                return map;

And then just update the dts to:

	gcc {
		compatible = "syscon", "simple-mfd";
		reg = <0x00900000 0x4000>;

		gcc: clock-controller {
			compatible = "qcom,gcc-apq8064";
			#clock-cells = <1>;
			#reset-cells = <1>;
		};
	};

But as I implemented this I realized that the syscon_node_to_regmap()
does not register the regmap as a devres of the parent, and as such it's
not caught by the regmap lookup in devm_clk_register_regmap().

So either one would need to make the syscon throw the regmap into devres
or make it possible to pass a regmap to devm_clk_register_regmap() for
this to work.

I'm confused about what you're suggesting and what benefit it 
has versus creating a child of the clock controller device.

The benefit would simply be that we're using the already existing
mechanism for handling multiple platform_drivers sharing a hw block.

Here's the patch I'm suggesting. The device name is probably wrong, but 
you get the idea.

Looks very much like my take on it as well, I do however have concerns
that suddenly the node called "clock-controller" will have to come with
tsens related properties.

Are all the tsens-in-gcc blocks the same? Or do you intend to of_match
on the gcc compatible in the tsens driver?

Regards,
Bjorn