devicetree vs. MODULE_DEPEND

[Manuel Stühn]

On Wed, 09 Aug 2017 15:59:21 -0600, Ian Lepore wrote
> On Wed, 2017-08-09 at 22:59 +0200, Manuel Stühn wrote:
>> Hi list,
>>
>> is it correct, that the sequence in the devicetree-blob defines the
>> probing sequence without considering the MODULE_DEPEND-macro?
>>
>> I stumbled over an unexpected behavior during the ti_pruss-driver
>> development. Because the ti-pruss is gone in the default devicetree,
>> I
>> activate it via the overlay-framework and put it to the address
>> "/ocp/pruss at 4a300000".  The devicetree-blob contains the entry and
>> the
>> driver gets probed, but it fails to enable its clock.
>> This is quite obvious as according to dmesg the am335x_prcm0 is
>> probed
>> _after_ the ti_pruss0 device. So I tried to handle this by adding an
>> explicit dependency to ti_prcm into the ti_pruss driver like:
>> MODULE_DEPEND(ti_pruss, ti_prcm, 1, 1, 1);
>>
>> It compiles cleanly, unfortunately this changes nothing. Only placing
>> it
>> in the devicetree after the prcm-node or loading it as a module
>> after
>> the OS booted up makes the device probe correctly.
>>
>> Any ideas?
> 
> MODULE_DEPEND only affects the kernel linker.  It ensures that other
> modules you depend on automatically get loaded along with your module.
> 
> Try the attached patch to ensure that the clocks driver is loaded
> earlier than drivers that might rely on it.
> 

The patch works for me, thanks for your help!