Re: [PATCH v2 1/8] clk: qcom: clk-alpha-pll: introduce stromer plus ops

[Stephen Boyd <sboyd@kernel.org>]

Quoting Varadarajan Narayanan (2023-10-12 02:26:17)
> Stromer plus APSS PLL does not support dynamic frequency scaling.
> To switch between frequencies, we have to shut down the PLL,
> configure the L and ALPHA values and turn on again. So introduce the
> separate set of ops for Stromer Plus PLL.

Does this assume the PLL is always on?

> 
> Signed-off-by: Kathiravan T <quic_kathirav@quicinc.com>
> Signed-off-by: Varadarajan Narayanan <quic_varada@quicinc.com>
> ---
> v2:     Use clk_alpha_pll_stromer_determine_rate, instead of adding new
>         clk_alpha_pll_stromer_plus_determine_rate as the alpha pll width
>         is same for both
> 
>         Fix review comments
>                 udelay(50) -> usleep_range(50, 60)
>                 Remove SoC-specific from print message
> ---
>  drivers/clk/qcom/clk-alpha-pll.c | 57 ++++++++++++++++++++++++++++++++++++++++
>  drivers/clk/qcom/clk-alpha-pll.h |  1 +
>  2 files changed, 58 insertions(+)
> 
> diff --git a/drivers/clk/qcom/clk-alpha-pll.c b/drivers/clk/qcom/clk-alpha-pll.c
> index 4edbf77..5221b6c 100644
> --- a/drivers/clk/qcom/clk-alpha-pll.c
> +++ b/drivers/clk/qcom/clk-alpha-pll.c
> @@ -2508,3 +2508,60 @@ const struct clk_ops clk_alpha_pll_stromer_ops = {
>         .set_rate = clk_alpha_pll_stromer_set_rate,
>  };
>  EXPORT_SYMBOL_GPL(clk_alpha_pll_stromer_ops);
> +
> +static int clk_alpha_pll_stromer_plus_set_rate(struct clk_hw *hw,
> +                                              unsigned long rate,
> +                                              unsigned long prate)
> +{
> +       struct clk_alpha_pll *pll = to_clk_alpha_pll(hw);
> +       u32 l, alpha_width = pll_alpha_width(pll);
> +       int ret;
> +       u64 a;
> +
> +       rate = alpha_pll_round_rate(rate, prate, &l, &a, alpha_width);
> +
> +       regmap_write(pll->clkr.regmap, PLL_MODE(pll), 0);

There's a theoretical problem here if I understand correctly. A call to
clk_enable() can happen while clk_set_rate() is in progress or vice
versa. Probably we need some sort of spinlock for this PLL that
synchronizes any enable/disable with the rate change so that when we
restore the enable bit the clk isn't enabled when it was supposed to be
off.