Locking in the clk API

[Jeremy Kerr]

Hi all,

Based on the discussion from this thread, my plan is to:

* Change the 'common struct clk' patches to only use a spinlock for locking. 
This means that clk_{en,dis}able will acquire a per-clk spinlock (for enable 
counts), and be callable from atomic contexts.

* Rework the initial docs (posted in the first mail of this thread) document 
to illustrate the new locking requirements.

* Request input from the platforms that require clk_enable (etc) to sleep, 
about how we can merge the two implementations.

Russell - is this OK?

Cheers,


Jeremy