Re: [PATCH v5 2/5] clk: meson: add support for A1 PLL clock ops

[Martin Blumenstingl <martin.blumenstingl@googlemail.com>]

Hi Jian,

On Fri, Dec 27, 2019 at 10:46 AM Jian Hu <jian.hu@amlogic.com> wrote:
[...]
> @@ -294,9 +298,12 @@ static int meson_clk_pll_is_enabled(struct clk_hw *hw)
>  {
>         struct clk_regmap *clk = to_clk_regmap(hw);
>         struct meson_clk_pll_data *pll = meson_clk_pll_data(clk);
> +       int ret = 0;
>
> -       if (meson_parm_read(clk->map, &pll->rst) ||
> -           !meson_parm_read(clk->map, &pll->en) ||
> +       if (MESON_PARM_APPLICABLE(&pll->rst))
> +               ret = meson_parm_read(clk->map, &pll->rst);
> +
> +       if (ret || !meson_parm_read(clk->map, &pll->en) ||
>             !meson_parm_read(clk->map, &pll->l))
>                 return 0;
I had to read this part twice to understand what it's doing because I
misunderstood what "ret" is used for (I thought that some "return ret"
is missing)
my proposal to make it easier to read:
...
if (MESON_PARM_APPLICABLE(&pll->rst) &&
    meson_parm_read(clk->map, &pll->rst))
  return 0;

if (!meson_parm_read(clk->map, &pll->en) ||
    !meson_parm_read(clk->map, &pll->l))
                 return 0;
...

please let me know what you think about this

> @@ -321,6 +328,23 @@ static int meson_clk_pll_enable(struct clk_hw *hw)
>         /* do nothing if the PLL is already enabled */
>         if (clk_hw_is_enabled(hw))
>                 return 0;
> +       /*
> +        * Compared with the previous SoCs, self-adaption module current
> +        * is newly added for A1, keep the new power-on sequence to enable the
> +        * PLL.
> +        */
> +       if (MESON_PARM_APPLICABLE(&pll->current_en)) {
> +               /* Enable the pll */
> +               meson_parm_write(clk->map, &pll->en, 1);
> +               udelay(10);
> +               /* Enable the pll self-adaption module current */
> +               meson_parm_write(clk->map, &pll->current_en, 1);
> +               udelay(40);
> +               /* Enable lock detect module */
> +               meson_parm_write(clk->map, &pll->l_detect, 1);
> +               meson_parm_write(clk->map, &pll->l_detect, 0);
> +               goto out;
> +       }
in all other functions you are skipping the pll->rst register by
checking for MESON_PARM_APPLICABLE(&pll->rst)
I like that because it's a pattern which is easy to follow

do you think we can make this part consistent with that?
I'm thinking of something like this (not compile-tested and I dropped
all comments, just so you get the idea):
...
if (MESON_PARM_APPLICABLE(&pll->rst)
  meson_parm_write(clk->map, &pll->rst, 1);

meson_parm_write(clk->map, &pll->en, 1);

if (MESON_PARM_APPLICABLE(&pll->rst))
  meson_parm_write(clk->map, &pll->rst, 0);

if (MESON_PARM_APPLICABLE(&pll->current_en))
  meson_parm_write(clk->map, &pll->current_en, 1);

if (MESON_PARM_APPLICABLE(&pll->l_detect)) {
  meson_parm_write(clk->map, &pll->l_detect, 1);
  meson_parm_write(clk->map, &pll->l_detect, 0);
}

if (meson_clk_pll_wait_lock(hw))
...

I see two (and a half) benefits here:
- if there's a PLL with neither the pll->current_en nor the pll->rst
registers then you get support for this implementation for free
- the if (MESON_PARM_APPLICABLE(...)) pattern is already used in the
driver, but only for one register (in your example when
MESON_PARM_APPLICABLE(&pll->current_en) exists you also modify the
pll->l_detect register, which I did not expect)
- only counts half: no use of "goto", which in my opinion makes it
very easy to read (just read from top to bottom, checking each "if")


Martin