Re: [RESEND 2/2] clk: Add driver for Palmas clk32kg and clk32kgaudio clocks

[Mike Turquette <mturquette@linaro.org>]

Quoting Peter Ujfalusi (2014-06-29 22:56:55)
> Hi Javier,
> 
> On 06/27/2014 09:23 PM, Javier Martinez Canillas wrote:
> > Hello Peter,
> > 
> > On Fri, Jun 27, 2014 at 8:01 AM, Peter Ujfalusi <peter.ujfalusi@ti.com> wrote:
> >> Palmas class of devices can provide 32K clock(s) to be used by other devices
> >> on the board. Depending on the actual device the provided clocks can be:
> >> CLK32K_KG and CLK32K_KGAUDIO
> >> or only one:
> >> CLK32K_KG (TPS659039 for example)
> >>
> >> Use separate compatible flags for the two 32K clock.
> >> A system which needs or have only one of the 32k clock from
> >> Palmas will need to add node(s) for each clock as separate section
> >> in the dts file.
> >> The two compatible property is:
> >> "ti,palmas-clk32kg" for clk32kg clock
> >> "ti,palmas-clk32kgaudio" for clk32kgaudio clock
> >>
> >> Apart from the register control of the clocks - which is done via
> >> the clock API there is a posibility to enable the external sleep
> >> control. In this way the clock can be enabled/disabled on demand by the
> >> user of the clock.
> >>
> >> See the documentation for more details.
> >>
> >> Signed-off-by: Peter Ujfalusi <peter.ujfalusi@ti.com>
> >> Reviewed-by: Nishanth Menon <nm@ti.com>
> 
> >> +static unsigned long palmas_clks_recalc_rate(struct clk_hw *hw,
> >> +                                            unsigned long parent_rate)
> >> +{
> >> +       return 32768;
> >> +}
> > 
> > I see that other clock drivers using a constant rate return 0 if the
> > clock has not been enabled.
> 
> and there are examples when similar fixed clock drivers returns only the clock
> value, like clk-max77686. I can not find clear guidelines neither in the
> documentation or around the header/c files for this.
> Mike, what is the appropriate way of handling the recalc_rate?

You are right that there are no guidelines stating, "don't do that", but
please, "don't do that" ;-)

clk_enable and clk_set_rate are entirely unrelated operations from the
perspective of the Linux clock framework, and mixing these two classes
of operations is a recipe for pain.

> 
> > So maybe is more correct to have something
> > like the following?
> > 
> > if (__clk_is_enabled(hw->clk))
> >         return 32768;
> > else
> >         return 0;

So what happens here if this is gateable clock and later on we call
clk_enable on it? The clocks rate will still be zero since
clk_enable/clk_disable do not touch the rate at all.

Regards,
Mike

> > 
> > Best regards,
> > Javier
> > 
> 
> 
> -- 
> Péter