Re: [PATCH v1 1/7] clk: qcom: clk-alpha-pll: Add support for Fabia PLL calibration

[Stephen Boyd <sboyd@kernel.org>]

Quoting Taniya Das (2019-10-31 05:21:07)
> diff --git a/drivers/clk/qcom/clk-alpha-pll.c b/drivers/clk/qcom/clk-alpha-pll.c
> index 055318f..8cb77ca 100644
> --- a/drivers/clk/qcom/clk-alpha-pll.c
> +++ b/drivers/clk/qcom/clk-alpha-pll.c
> @@ -1141,15 +1160,11 @@ static int alpha_pll_fabia_set_rate(struct clk_hw *hw, unsigned long rate,
>                                                 unsigned long prate)
>  {
>         struct clk_alpha_pll *pll = to_clk_alpha_pll(hw);
> -       u32 val, l, alpha_width = pll_alpha_width(pll);
> +       u32 l, alpha_width = pll_alpha_width(pll);
>         u64 a;
>         unsigned long rrate;
>         int ret = 0;
> 
> -       ret = regmap_read(pll->clkr.regmap, PLL_MODE(pll), &val);
> -       if (ret)
> -               return ret;
> -
>         rrate = alpha_pll_round_rate(rate, prate, &l, &a, alpha_width);
> 
>         /*

How is this diff related? Looks like it should be split off into another
patch to remove a useless register read.

> @@ -1167,7 +1182,66 @@ static int alpha_pll_fabia_set_rate(struct clk_hw *hw, unsigned long rate,
>         return __clk_alpha_pll_update_latch(pll);
>  }
> 
> +static int alpha_pll_fabia_prepare(struct clk_hw *hw)
> +{

Why are we doing this in prepare vs. doing it at PLL configuration time?
Does it need to be recalibrated each time the PLL is enabled?

> +       struct clk_alpha_pll *pll = to_clk_alpha_pll(hw);
> +       const struct pll_vco *vco;
> +       struct clk_hw *parent_hw;
> +       unsigned long cal_freq, rrate;
> +       u32 cal_l, regval, alpha_width = pll_alpha_width(pll);
> +       u64 a;
> +       int ret;
> +
> +       /* Check if calibration needs to be done i.e. PLL is in reset */
> +       ret = regmap_read(pll->clkr.regmap, PLL_MODE(pll), &regval);

Please use 'val' instead of 'regval' as regval almost never appears in
this file already.

> +       if (ret)
> +               return ret;
> +
> +       /* Return early if calibration is not needed. */
> +       if (regval & PLL_RESET_N)
> +               return 0;
> +
> +       vco = alpha_pll_find_vco(pll, clk_hw_get_rate(hw));
> +       if (!vco) {
> +               pr_err("alpha pll: not in a valid vco range\n");
> +               return -EINVAL;
> +       }
> +
> +       cal_freq = DIV_ROUND_CLOSEST((pll->vco_table[0].min_freq +
> +                               pll->vco_table[0].max_freq) * 54, 100);

Do we need to cast the first argument to a u64 to avoid overflow?

> +
> +       parent_hw = clk_hw_get_parent(hw);
> +       if (!parent_hw)
> +               return -EINVAL;
> +
> +       rrate = alpha_pll_round_rate(cal_freq, clk_hw_get_rate(parent_hw),
> +                                       &cal_l, &a, alpha_width);
> +       /*
> +        * Due to a limited number of bits for fractional rate programming, the
> +        * rounded up rate could be marginally higher than the requested rate.
> +        */
> +       if (rrate > (cal_freq + FABIA_PLL_RATE_MARGIN) || rrate < cal_freq) {
> +               pr_err("Call set rate on the PLL with rounded rates!\n");

This message is weird. Drivers shouldn't need to call set rate with
rounded rates. What is going on?

> +               return -EINVAL;
> +       }
> +
> +       /* Setup PLL for calibration frequency */
> +       regmap_write(pll->clkr.regmap, PLL_ALPHA_VAL(pll), cal_l);
> +
> +       /* Bringup the pll at calibration frequency */

capitalize PLL.

> +       ret = clk_alpha_pll_enable(hw);
> +       if (ret) {
> +               pr_err("alpha pll calibration failed\n");
> +               return ret;
> +       }
> +
> +       clk_alpha_pll_disable(hw);
> +
> +       return 0;
> +}
> +