Hello John,

On Mon, Jan 25, 2016 at 10:13 AM, John Crispin [off-list ref] wrote:

On 25/01/2016 13:35, Mark Brown wrote:

quoted

On Mon, Jan 25, 2016 at 01:19:46PM +0100, John Crispin wrote:

quoted

On 25/01/2016 13:11, Javier Martinez Canillas wrote:

quoted

quoted

You don't have a .of_match table but according the DT bindings, the
compatible string "mediatek,mt6323-regulator" should be used so there
should be a OF match table or the vendor prefix of the compatible
string won't be used for matching (i.e: fallbacks to the driver .name
for match).

quoted

the driver is probed via drivers/mfd/mt6397-core.c and does not require
the OF match table. It loads fine just like the mt6397 driver.

That's fine but you shouldn't have the compatible string in your binding
document since it's not actually used or needed.

Hi,

correct me if i am wrong but if we remove the compatible string from the
binding document, then we will also have to remove it from the dts and
then the kernel won't be able to match the node to the driver and thus
the regulator phandle derefs will fail.

The kernel doesn't need to match the compatible since the MFD driver
register the device explicitly with mfd_add_devices().

In fact, the kernel is currently not matching the compatible, it is
only matching because you provided a .of_compatible is provided in the
mfd_cell.

If you wan't subdevices for a MFD to be registered automatically by OF
and the compatible matched like other buses, then your MFD driver
needs to call of_platform_populate() instead mfd_add_devices().

        John

Best regards,
Javier