Re: [PATCH] clk: renesas: rcar-gen3: Fix cpg_sd_clock_round_rate() return value

[Simon Horman <horms@verge.net.au>]

On Fri, Mar 29, 2019 at 09:58:24AM +0100, Geert Uytterhoeven wrote:
> Hi Simon,
> 
> On Fri, Mar 29, 2019 at 9:45 AM Simon Horman <horms@verge.net.au> wrote:
> > On Wed, Mar 27, 2019 at 01:31:33PM +0100, Geert Uytterhoeven wrote:
> > > From: Takeshi Kihara <takeshi.kihara.df@renesas.com>
> > >
> > > cpg_sd_clock_round_rate() may return an unsupported clock rate for the
> > > requested clock rate. Therefore, when cpg_sd_clock_set_rate() sets the
> > > clock rate acquired by cpg_sd_clock_round_rate(), an error may occur.
> > >
> > > This is not conform the clk API design.
> > >
> > > This patch fixes that by making sure cpg_sd_clock_calc_div() considers
> > > only the division values defined in cpg_sd_div_table[].
> > > With this fix, the cpg_sd_clock_round_rate() always return a support
> > > clock rate.
> > >
> > > Signed-off-by: Takeshi Kihara <takeshi.kihara.df@renesas.com>
> > > Fixes: 90c073e53909da85 ("clk: shmobile: r8a7795: Add SD divider support")
> > > Signed-off-by: Geert Uytterhoeven <geert+renesas@glider.be>
> > > ---
> > > To be queued in clk-renesas-for-v5.2.
> > >
> > >  drivers/clk/renesas/rcar-gen3-cpg.c | 28 +++++++++++++---------------
> > >  1 file changed, 13 insertions(+), 15 deletions(-)
> > >
> > > diff --git a/drivers/clk/renesas/rcar-gen3-cpg.c b/drivers/clk/renesas/rcar-gen3-cpg.c
> > > index d5fb768b089ff1c1..d2745c57207efc01 100644
> > > --- a/drivers/clk/renesas/rcar-gen3-cpg.c
> > > +++ b/drivers/clk/renesas/rcar-gen3-cpg.c
> > > @@ -3,6 +3,7 @@
> > >   * R-Car Gen3 Clock Pulse Generator
> > >   *
> > >   * Copyright (C) 2015-2018 Glider bvba
> > > + * Copyright (C) 2019 Renesas Electronics Corp.
> > >   *
> > >   * Based on clk-rcar-gen3.c
> > >   *
> > > @@ -236,8 +237,6 @@ struct sd_clock {
> > >       const struct sd_div_table *div_table;
> > >       struct cpg_simple_notifier csn;
> > >       unsigned int div_num;
> > > -     unsigned int div_min;
> > > -     unsigned int div_max;
> > >       unsigned int cur_div_idx;
> > >  };
> > >
> > > @@ -314,14 +313,20 @@ static unsigned int cpg_sd_clock_calc_div(struct sd_clock *clock,
> > >                                         unsigned long rate,
> > >                                         unsigned long parent_rate)
> > >  {
> > > -     unsigned int div;
> > > -
> > > -     if (!rate)
> > > -             rate = 1;
> > > +     unsigned long calc_rate, best_rate = 0, diff, diff_min = ULONG_MAX;
> > > +     unsigned int i;
> > >
> > > -     div = DIV_ROUND_CLOSEST(parent_rate, rate);
> > > +     for (i = 0; i < clock->div_num; i++) {
> > > +             calc_rate = DIV_ROUND_CLOSEST(parent_rate,
> > > +                                           clock->div_table[i].div);
> > > +             diff = calc_rate > rate ? calc_rate - rate : rate - calc_rate;
> > > +             if (diff <= diff_min) {
> > > +                     best_rate = calc_rate;
> > > +                     diff_min = diff;
> > > +             }
> > > +     }
> >
> > My reading is that.
> > a) this algorithm picks a clock based on the div field of the
> >    members of cpg_sd_div_table. and
> > b) in the case of duplicate values of that field the first one is chosen; and
> > c) such duplicates (of sd_div values) do exist in cpg_sd_div_table
> >
> > Is this the intended behaviour?
> 
> Actually cpg_sd_clock_calc_div() picks the last one, due to the "<=" :-)
> But that doesn't matter here, as both dividers are equal.

True :)

> cpg_sd_clock_set_rate() will pick the first entry from the table with matching
> div value later.
> 
> > > -     return clamp_t(unsigned int, div, clock->div_min, clock->div_max);
> > > +     return DIV_ROUND_CLOSEST(parent_rate, best_rate);
> >
> > Would it be better to return the value of clock->div_table[i].div
> > that yielded best_rate?
> 
> True. The DIV_ROUND_CLOSEST() is not needed, as we could have
> saved best_div instead of best_rate. Will fix.

Thanks!