Re: [PATCH v9 2/5] clk: meson: a1: add Amlogic A1 PLL clock controller driver

From: Dmitry Rokosov
Date: Mon Mar 13 2023 - 06:25:28 EST


On Mon, Mar 13, 2023 at 10:18:02AM +0100, Jerome Brunet wrote:
>
> On Thu 09 Mar 2023 at 21:28, Dmitry Rokosov <ddrokosov@xxxxxxxxxxxxxx> wrote:
>
> >> >>
> >> >> This last poke should not bits otherwise handled by parms.
> >> >> This is a rate init in disguise.
> >> >>
> >> >
> >> > I believe, you are talking about hifi_pll clk_regmap conflicts with
> >> > hifi_init_regs. The above init sequence shouldn't affect pll regmap setup,
> >> > it doesn't touch them (we assume that default bit values are all zero):
> >> >
> >> > .en = {
> >> > .reg_off = ANACTRL_HIFIPLL_CTRL0,
> >> > .shift = 28,
> >> > .width = 1,
> >> > },
> >> > // init_value = 0x01f18440
> >> > // en_mask = 0x10000000
> >> >
> >> > .m = {
> >> > .reg_off = ANACTRL_HIFIPLL_CTRL0,
> >> > .shift = 0,
> >> > .width = 8,
> >> > },
> >> > // init_value = 0x01f18440
> >> > // m_mask = 0x0000000f
> >>
> >> mask is 0xff with width 8
> >>
> >
> > Ah, you're right. Anyway, I think this is just init value and it's okay
> > to set it during initialization and rewrite after in parameter
> > propagation stage.
> >
>
> ... But the magic pokes are there only to initialize the unmanaged part
> of the clock regs. I'd like it to be clear and stay that way.
>
> So please, clear the managed fields from the initial poke table.

I've double checked hifi_pll clk. In the my current configuration no any
clks inherited from it. Therefore its 'enable_count' equals to 0.
And of course in the such situation the rate must be zeroed as well.
It means you are right at all. I'll remove pre_sel and fbkdiv hifi_pll
pre-setup in the next version.
Thank you for hunted down!

--
Thank you,
Dmitry