On Fri, Nov 11, 2011 at 10:15:56AM +0100, Sascha Hauer wrote:

On Fri, Nov 11, 2011 at 05:05:47PM +0800, Richard Zhao wrote:

quoted

{
	int ret;

	ret = clk_prepare(clk);
	if (ret)
		return ret;
	ret = clk_enable(clk);
	if (ret)
		clk_unprepare(clk);
	return ret;

Yes, looks good.

While this looks like a nice easy solution for converting existing
drivers, I'd suggest thinking about this a little more...

I would suggest some thought is given to the placement of clk_enable()
and clk_disable() when adding clk_prepare(), especially if your existing
clk_enable() function can only be called from non-atomic contexts.

Obviously, the transition path needs to be along these lines:

1. add clk_prepare() to drivers
2. implement clk_prepare() and make clk_enable() callable from non-atomic
   contexts
3. move clk_enable() in drivers to places it can be called from non-atomic
   contexts to achieve greater power savings (maybe via the runtime pm)

and where a driver is shared between different sub-architectures which
have non-atomic clk_enable()s, (3) can only happen when all those sub-
architectures have been updated to step (2).