Re: [RFC] Common clock framework for external clock generators

[Sebastian Hesselbarh <sebastian.hesselbarth@googlemail.com>]

On 05/13/2012 06:43 PM, Mark Brown wrote:
> One of the patches I've been sending adds a dummy clk_unregister() for
> the sake of making the drivers look nicer - practically speaking it's
> not likely to be terribly important as these things don't get unloaded
> terribly often.  It looks like that patch didn't get applied either.

Well, of course I don't plan to unload the driver ever but basically it
is possible..

One more thing I thought about: The platform I currently use needs to 
pass the external clocks to the platform devices that can use them
later. IMHO the correct way of creating clocks would be:

- register i2c clock driver and let it register its clocks with names
   like e.g. si5351, clkout0. The clock driver itself cannot and should
   not know who uses it later on.
- let drivers look for e.g. kirkwood-i2s.1, extclk because the i2s
   driver cannot know where the external clock comes from.
- have a board-specific function that configures clock hierarchy and
   create suitable clk_aliases e.g.
   si5351,clkout0 = kirkwood-i2s.1,extclk.

Currently I added a callback function pointer to the platform data
passed to the i2c clock driver that is called at the end of clock
driver probe. I doubt it will be accepted that way but can't think
of any other way..

Sebastian