On czw, 2014-10-30 at 13:42 +0100, Javier Martinez Canillas wrote:

Hello Krzysztof,

On 10/30/2014 01:30 PM, Krzysztof Kozlowski wrote:
tor_of_get_init_data() function?

quoted

quoted

quoted

The regulator_of_get_init_data() searches from dev->of_node or its child
node.

But dev->of_node is NULL.

That's why of_compatible is needed.

Yes but regulator_register() does dev = config->dev and config->dev is set
to config.dev = iodev->dev in the driver probe function which is the
pdev->dev.parent (the PMIC struct device) that has an associated of_node.

So, regulator_of_get_init_data() will call of_get_child_by_name() passing
the PMIC of_node and the sub-node name that contains the regulators. That
is, whatever was set in desc->regulators_node and that should be enough.

I missed that one in max77802 (in max77686: config.dev = &pdev->dev).
Now I wonder if it is proper to attach regulators to driver's parent
device. Consider regulator_register, around line 3640:
	rdev->dev.parent = dev;
The parent of regulators will be equal to parent of regulator driver -
main MFD driver.

An early version of the max77802 regulator driver also set config.dev to
&pdev->dev but then Mark asked me if I was sure that it should not be the
MFD device instead.

I then read in regulator_register() that the rdev->dev.parent was set
to the device passed in config.dev as you said so I was convinced that it
should be the parent device instead.

Please take a look at [0] for that discussion.

To me a intuitive structure would be:
MFD device
  |
  - clock device
     |
     - clock1
     - clock2
  - regulator device
     |
     - LDO1
     - LDO2
etc.

This also maps to structure in DTS. dev_err* messages and any
allocations should be done on behalf of regulator device, not parent.

Various drivers do this differently... The wm8* drivers set it mostly to
parent (MFD)...

I do not insists, especially because using parent's device would make
this driver simpler.

Mark, maybe you could shed light on it?


Best regards,
Krzysztof