Re: [PATCH 1/3] Clocklib: add generic framework for managing clocks.

["pHilipp Zabel" <philipp.zabel@gmail.com>]

On Thu, Mar 27, 2008 at 10:18 AM, Russell King
<rmk+lkml@arm.linux.org.uk> wrote:
> On Thu, Mar 27, 2008 at 10:06:23AM +0100, Haavard Skinnemoen wrote:
>  > > +   int             users;
>  >
>  > Reference counting, probably need that too.
>  >
>  > > +   unsigned long   rate;
>  >
>  > This one is redundant. Use getrate() instead.
>
>  ... which means a separate getrate() functions for every clock.  Not really
>  practical for PXA.

struct pxa_clk {
       struct clk generic_clk;
       int rate;
       int delay;
}

unsigned long pxa_clk_getrate (struct clk *clk)
{
        return container_of(clk, struct pxa_clk, generic_clk)->rate;
}

>  > > +   int             delay;
>  >
>  > Huh? A delay after enabling the clock? Why can't the enable() hook do
>  > that if it's really necessary?
>
>  ... which means a separate enable() function for each clock.  The delay
>  there has not a lot to do with the actual register you're frobbing, but
>  the resy of the SoC.  So, again, not really practical for PXA.

int pxa_clk_mode(struct clk *clk, bool enabled)
{
        if (enabled)
                udelay(container_of(clk, struct pxa_clk, generic_clk)->delay);
        generic_clk_mode(clk, enabled);
}

>  The result for PXA is a tradeoff between reducing the data size and
>  increasing the text size, or increasing the data size and keeping
>  the existing data size.

Wouldn't something like the above work without increasing text size too much?

regards
Philipp