Re: [PATCH 2/2] clk: renesas: rzv2h: Report the DSI duty cycle from the LVDS divider

[Tommaso Merciai]

Hi Geert,

On Mon, Nov 02, 2026 at 05:16:09PM +0100, Geert Uytterhoeven wrote:
> Hi Tommaso,
> 
> Thanks for your patch!
> 
> On Thu, 29 Oct 2026 at 15:26, Tommaso Merciai
> <tommaso.merciai.xr@xxxxxxxxxxxxxx> wrote:
> > The 4/7 duty cycle of the LVDS path comes from the division by 7, not from
> > the PLLDSI mux, as an odd divider cannot produce a 50% duty cycle.
> 
> Note that the latter is not true in general: if an odd divider looks
> at both edges of the parent clock signal, the output will have a 50%
> duty cycle (assumed the parent has a 50% duty cycle).
> If an odd divider looks only at rising or falling edges, the output
> will indeed have a non-50% duty cycle.

Thanks for sharing.

> 
> > Report it from a dedicated LVDS divider clock type and let the core walk up
> > to the selected parent through CLK_DUTY_CYCLE_PARENT. The reported values
> > do not change, the mux keeps selecting the path, and its duty cycle
> > constants go away.
> >
> > Signed-off-by: Tommaso Merciai <tommaso.merciai.xr@xxxxxxxxxxxxxx>
> 
> Reviewed-by: Geert Uytterhoeven <geert+renesas@xxxxxxxxx>
> i.e. will queue in renesas-clk for v7.5.
> 
> > --- a/drivers/clk/renesas/rzv2h-cpg.c
> > +++ b/drivers/clk/renesas/rzv2h-cpg.c
> 
> > @@ -890,6 +863,94 @@ rzv2h_cpg_fixed_mod_status_clk_register(const struct cpg_core_clk *core,
> >         return &clk_hw_data->fix.hw;
> >  }
> >
> > +static int rzv2h_cpg_plldsi_div_lvds_get_duty_cycle(struct clk_hw *hw,
> > +                                                   struct clk_duty *duty)
> > +{
> > +       struct clk_fixed_factor *fix = to_clk_fixed_factor(hw);
> > +
> > +       /*
> > +        * An odd divider cannot generate a 50% duty cycle: the output stays
> > +        * high for (div + 1) / 2 input clock cycles out of div, e.g. a divider
> > +        * of 7 gives a duty cycle of 4/7.
> > +        */
> > +       duty->num = DIV_ROUND_UP(fix->div, 2);
> > +       duty->den = fix->div;
> > +
> > +       return 0;
> > +}
> > +
> > +static int rzv2h_cpg_plldsi_div_lvds_determine_rate(struct clk_hw *hw,
> > +                                                   struct clk_rate_request *req)
> > +{
> > +       return clk_fixed_factor_ops.determine_rate(hw, req);
> > +}
> > +
> > +static int rzv2h_cpg_plldsi_div_lvds_set_rate(struct clk_hw *hw,
> > +                                             unsigned long rate,
> > +                                             unsigned long parent_rate)
> > +{
> > +       return clk_fixed_factor_ops.set_rate(hw, rate, parent_rate);
> > +}
> > +
> > +static unsigned long rzv2h_cpg_plldsi_div_lvds_recalc_rate(struct clk_hw *hw,
> > +                                                          unsigned long parent_rate)
> > +{
> > +       return clk_fixed_factor_ops.recalc_rate(hw, parent_rate);
> > +}
> > +
> > +static unsigned long
> > +rzv2h_cpg_plldsi_div_lvds_recalc_accuracy(struct clk_hw *hw,
> > +                                         unsigned long parent_accuracy)
> > +{
> > +       return clk_fixed_factor_ops.recalc_accuracy(hw, parent_accuracy);
> > +}
> > +
> > +static const struct clk_ops rzv2h_cpg_plldsi_div_lvds_ops = {
> > +       .determine_rate = rzv2h_cpg_plldsi_div_lvds_determine_rate,
> > +       .set_rate = rzv2h_cpg_plldsi_div_lvds_set_rate,
> > +       .recalc_rate = rzv2h_cpg_plldsi_div_lvds_recalc_rate,
> > +       .recalc_accuracy = rzv2h_cpg_plldsi_div_lvds_recalc_accuracy,
> > +       .get_duty_cycle = rzv2h_cpg_plldsi_div_lvds_get_duty_cycle,
> > +};
> > +
> > +static struct clk_hw * __init
> > +rzv2h_cpg_plldsi_div_lvds_clk_register(const struct cpg_core_clk *core,
> > +                                      struct rzv2h_cpg_priv *priv)
> > +{
> > +       struct clk_init_data init = { };
> > +       struct clk_fixed_factor *fix;
> > +       const struct clk_hw *parent;
> > +       const char *parent_name;
> > +       int ret;
> > +
> > +       WARN_DEBUG(core->parent >= priv->num_core_clks);
> > +       parent = priv->clks[core->parent];
> > +       if (IS_ERR(parent))
> > +               return ERR_CAST(parent);
> > +
> > +       parent_name = clk_hw_get_name(parent);
> > +
> > +       fix = devm_kzalloc(priv->dev, sizeof(*fix), GFP_KERNEL);
> > +       if (!fix)
> > +               return ERR_PTR(-ENOMEM);
> > +
> > +       init.name = core->name;
> > +       init.ops = &rzv2h_cpg_plldsi_div_lvds_ops;
> > +       init.flags = CLK_SET_RATE_PARENT;
> > +       init.parent_names = &parent_name;
> > +       init.num_parents = 1;
> > +
> > +       fix->hw.init = &init;
> > +       fix->mult = core->mult;
> > +       fix->div = core->div;
> > +
> > +       ret = devm_clk_hw_register(priv->dev, &fix->hw);
> > +       if (ret)
> > +               return ERR_PTR(ret);
> > +
> > +       return &fix->hw;
> > +}
> > +
> >  static struct clk_hw
> >  *rzv2h_cpg_clk_src_twocell_get(struct of_phandle_args *clkspec,
> >                                void *data)
> 
> Since there are now two copies of this code (see rzg3l_cpg_lvds_div_*),
> perhaps the time is ripe to add odd duty cycle support to
> clk_fixed_factor?  As duty->num can be rounded up or down, I am
> envisioning two new mutually-exclusive fixflags parameters, e.g.:
> 
>     CLK_FIXED_FACTOR_ODD_DUTY_DOWN
>     CLK_FIXED_FACTOR_ODD_DUTY_UP
> 
> What do you think?

Agreed, thanks. I can send send a separate series, adding both flags
into:

 - drivers/clk/clk-fixed-factor.c

In thi way later we can converts:

 - drivers/clk/renesas/rzv2h-cpg.c
 - drivers/clk/renesas/rzg2l-cpg.c

To the newly added CLK_FIXED_FACTOR_ODD_DUTY_UP.

Not sure about CLK_FIXED_FACTOR_ODD_DUTY_DOWN which has nos users for
now. Please correct me if I'm wrong.

Kind regards,
Tommaso

> Thanks!
> 
> Gr{oetje,eeting}s,
> 
>                         Geert
> 
> -- 
> Geert Uytterhoeven -- There's lots of Linux beyond ia32 -- geert@xxxxxxxxxxxxxx
> 
> In personal conversations with technical people, I call myself a hacker. But
> when I'm talking to journalists I just say "programmer" or something like that.
>                                 -- Linus Torvalds