Re: [PATCH v4 02/13] clk: sunxi: add ahb1 clock for A83T

From: Vishnu Patekar
Date: Sat Apr 16 2016 - 12:00:21 EST


Hello Maxime,

On Fri, Apr 15, 2016 at 1:28 AM, Maxime Ripard
<maxime.ripard@xxxxxxxxxxxxxxxxxx> wrote:
> On Mon, Apr 04, 2016 at 11:07:29AM +0800, Vishnu Patekar wrote:
>> Hello Maxime,
>>
>> On Thu, Mar 17, 2016 at 6:40 PM, Maxime Ripard
>> <maxime.ripard@xxxxxxxxxxxxxxxxxx> wrote:
>> > On Thu, Mar 17, 2016 at 12:04:25AM +0800, Vishnu Patekar wrote:
>> >> AHB1 on A83T is similar to ahb1 on A31, except parents are different.
>> >> clock index 0b1x is PLL6.
>> >>
>> >> Signed-off-by: Vishnu Patekar <vishnupatekar0510@xxxxxxxxx>
>> >> Acked-by: Chen-Yu Tsai <wens@xxxxxxxx>
>> >> Acked-by: Rob Herring <robh@xxxxxxxxxx>
>> >> ---
>> >> Documentation/devicetree/bindings/clock/sunxi.txt | 1 +
>> >> drivers/clk/sunxi/clk-sunxi.c | 76 +++++++++++++++++++++++
>> >> 2 files changed, 77 insertions(+)
>> >>
>> >> diff --git a/Documentation/devicetree/bindings/clock/sunxi.txt b/Documentation/devicetree/bindings/clock/sunxi.txt
>> >> index 834436f..cba9fe55 100644
>> >> --- a/Documentation/devicetree/bindings/clock/sunxi.txt
>> >> +++ b/Documentation/devicetree/bindings/clock/sunxi.txt
>> >> @@ -30,6 +30,7 @@ Required properties:
>> >> "allwinner,sun6i-a31-ar100-clk" - for the AR100 on A31
>> >> "allwinner,sun9i-a80-cpus-clk" - for the CPUS on A80
>> >> "allwinner,sun6i-a31-ahb1-clk" - for the AHB1 clock on A31
>> >> + "allwinner,sun8i-a83t-ahb1-clk" - for the AHB1 clock on A83T
>> >> "allwinner,sun8i-h3-ahb2-clk" - for the AHB2 clock on H3
>> >> "allwinner,sun6i-a31-ahb1-gates-clk" - for the AHB1 gates on A31
>> >> "allwinner,sun8i-a23-ahb1-gates-clk" - for the AHB1 gates on A23
>> >> diff --git a/drivers/clk/sunxi/clk-sunxi.c b/drivers/clk/sunxi/clk-sunxi.c
>> >> index 91de0a0..a7aab65 100644
>> >> --- a/drivers/clk/sunxi/clk-sunxi.c
>> >> +++ b/drivers/clk/sunxi/clk-sunxi.c
>> >> @@ -344,6 +344,67 @@ static void sun6i_ahb1_recalc(struct factors_request *req)
>> >> req->rate >>= req->p;
>> >> }
>> >>
>> >> +#define SUN8I_A83T_AHB1_PARENT_PLL6 2
>> >> +/**
>> >> + * sun8i_a83t_get_ahb_factors() - calculates m, p factors for AHB
>> >> + * AHB rate is calculated as follows
>> >> + * rate = parent_rate >> p
>> >> + *
>> >> + * if parent is pll6, then
>> >> + * parent_rate = pll6 rate / (m + 1)
>> >> + */
>> >> +
>> >> +static void sun8i_a83t_get_ahb1_factors(struct factors_request *req)
>> >> +{
>> >> + u8 div, calcp, calcm = 1;
>> >> +
>> >> + /*
>> >> + * clock can only divide, so we will never be able to achieve
>> >> + * frequencies higher than the parent frequency
>> >> + */
>> >> + if (req->parent_rate && req->rate > req->parent_rate)
>> >> + req->rate = req->parent_rate;
>> >> +
>> >> + div = DIV_ROUND_UP(req->parent_rate, req->rate);
>> >> +
>> >> + /* calculate pre-divider if parent is pll6 */
>> >> + if (req->parent_index >= SUN8I_A83T_AHB1_PARENT_PLL6) {
>> >> + if (div < 4)
>> >> + calcp = 0;
>> >> + else if (div / 2 < 4)
>> >> + calcp = 1;
>> >> + else if (div / 4 < 4)
>> >> + calcp = 2;
>> >> + else
>> >> + calcp = 3;
>> >> +
>> >> + calcm = DIV_ROUND_UP(div, 1 << calcp);
>> >> + } else {
>> >> + calcp = __roundup_pow_of_two(div);
>> >> + calcp = calcp > 3 ? 3 : calcp;
>> >> + }
>> >> +
>> >> + req->rate = (req->parent_rate / calcm) >> calcp;
>> >> + req->p = calcp;
>> >> + req->m = calcm - 1;
>> >> +}
>> >> +
>> >> +/**
>> >> +* sun8i_a83t_ahb1_recalc() - calculates AHB clock rate from m, p factors and
>> >> +* parent index
>> >> +*/
>> >> +static void sun8i_a83t_ahb1_recalc(struct factors_request *req)
>> >> +{
>> >> + req->rate = req->parent_rate;
>> >> +
>> >> +/* apply pre-divider first if parent is pll6 */
>> >
>> > The comment indentation is wrong
>> >
>> >> + if (req->parent_index >= SUN6I_AHB1_PARENT_PLL6)
>> >
>> > And this is not the right define you're using.
>> >
>> > I still believe that duplicating the same logic just because of
>> > different dividers is not the way to go.
>> >
>> > You could solve that easily by adding a table for the muxes, and
>> > associate it with parents and dividers, that you could store in
>> > clk_factors.
>>
>> I've similar solution (please ignore a83 specific functions those will
>> be common for a31 and a83t).
>> https://github.com/vishnupatekar/linux/commit/f7de5b48d886a672b1f6db112fbfd5d2c9849afa
>>
>> is it aligned to what you're saying?
>
> Yep. I'd even go a step further, and allow to have directly the core
> deal with the pre-divider.
>
> I guess in your prediv table you could have the index, and either the
> offset and width of the divider (if it's a variable one), or its fixed
> value.
That's good idea, index with shift and width will do as prediv can be variable.
>
> The generic function would then be able to deal with the rate
> adjustments, and you wouldn't need to be able to have anything related
> to the parent index in the clock specific functions anymore.
>
> Does that make sense?
Adjusting rate in generic clk_factors_recalc_rate, with prediv shift
and width is straight forward, and we can remove recalc function
itself, currently it's used only for a31 ahb1.

I'm doubtful about getter, In sun6i_get_ahb1_factors we still need to
pass req->prediv_width instead of req->parent_index.

I'll send separate patch for this prediv handling.
>
> Maxime
>
> --
> Maxime Ripard, Free Electrons
> Embedded Linux, Kernel and Android engineering
> http://free-electrons.com