Hi Geert,
On Tue, Apr 25, 2017 at 12:28:29PM +0200, Geert Uytterhoeven wrote:
> Hi Peter,
>
> On Tue, Apr 25, 2017 at 12:18 PM, Peter De Schrijver
> <pdeschrijver@xxxxxxxxxx> wrote:
> > On Mon, Apr 24, 2017 at 10:09:10AM +0200, Geert Uytterhoeven wrote:
> >> On Mon, Apr 24, 2017 at 10:03 AM, Peter De Schrijver
> >> <pdeschrijver@xxxxxxxxxx> wrote:
> >> >> > --- a/drivers/clk/renesas/rcar-gen3-cpg.c
> >> >> > +++ b/drivers/clk/renesas/rcar-gen3-cpg.c
> >> >>
> >> >> > +static unsigned long cpg_z_clk_recalc_rate(struct clk_hw *hw,
> >> >> > + unsigned long parent_rate)
> >> >> > +{
> >> >> > + struct cpg_z_clk *zclk = to_z_clk(hw);
> >> >> > + unsigned int mult;
> >> >> > + unsigned int val;
> >> >> > + unsigned long rate;
> >> >> > +
> >> >> > + val = (clk_readl(zclk->reg) & CPG_FRQCRC_ZFC_MASK)
> >> >> > + >> CPG_FRQCRC_ZFC_SHIFT;
> >> >> > + mult = 32 - val;
> >> >> > +
> >> >> > + rate = div_u64((u64)parent_rate * mult + 16, 32);
> >> >> > + /* Round to closest value at 100MHz unit */
> >> >> > + rate = 100000000*DIV_ROUND_CLOSEST(rate, 100000000);
> >> >>
> >> >> Mike, Stephen: what's your opinion about such rounding tricks?
> >> >
> >> > Is this an actual divider or a pulse skipper?
> >>
> >> Forgive my ignorance, but what is the difference?
> >
> > A pulse skipper, as the name says, skips pulses. Eg, if you configure the
> > skipper to 3/4, it will skip 1 input clock pulse out of 4. The pulse width
> > will be the same as the parent clock pulse width though. A divider stretches
> > the clock pulse. So the output of a divider configured to 1/2 will have a
> > pulse width which is twice as large as the pulse width of the input clock.
>
> OK, thanks! So a pulse skipper changes the duty cycle of the signal.
>
Yes, it does.
> I would expect this to be a real divider, as that would allow the Cortex A53
> core to do work on both edges of the clock signal.
Maybe. Although I know we've run A53s from skippers. One difference is that a
real divider allows you to lower the voltage, while a skipper does not,
because it only changes the duty cycle as you mentioned.
> It's a bit difficult to measure, though ;-)
>
Maybe there's a way to expose the clock signal on an external pin? Or you can
use a full chip emulator :)
Cheers,
Peter.
