Re: [PATCH 1/7] clk: samsung: add plls used in s3c2416 and s3c2443

[Sylwester Nawrocki <s.nawrocki@samsung.com>]

Hi,

On 03/12/2013 01:42 AM, Heiko Stübner wrote:
> This adds support for pll2126x, pll3000x, pll6552x and pll6553x.
> 
> Signed-off-by: Heiko Stuebner <heiko@sntech.de>
> ---
>  drivers/clk/samsung/clk-pll.c |  376 +++++++++++++++++++++++++++++++++++++++++
>  drivers/clk/samsung/clk-pll.h |    8 +
>  2 files changed, 384 insertions(+), 0 deletions(-)
> 
> diff --git a/drivers/clk/samsung/clk-pll.c b/drivers/clk/samsung/clk-pll.c
> index 4b24511..b772f9e 100644
> --- a/drivers/clk/samsung/clk-pll.c
> +++ b/drivers/clk/samsung/clk-pll.c
> @@ -400,6 +400,97 @@ struct clk * __init samsung_clk_register_pll46xx(const char *name,
...
> +struct samsung_clk_pll2126x {
> +	struct clk_hw		hw;
> +	const void __iomem	*con_reg;
> +};
> +
> +#define to_clk_pll2126x(_hw) container_of(_hw, struct samsung_clk_pll2126x, hw)
> +
> +static unsigned long samsung_pll2126x_recalc_rate(struct clk_hw *hw,
> +				unsigned long parent_rate)
> +{
> +	struct samsung_clk_pll2126x *pll = to_clk_pll2126x(hw);
> +	u32 pll_con, mdiv, pdiv, sdiv;
> +	u64 fvco = parent_rate;
> +
> +	pll_con = __raw_readl(pll->con_reg);
> +	mdiv = (pll_con >> PLL2126X_MDIV_SHIFT) & PLL2126X_MDIV_MASK;
> +	pdiv = (pll_con >> PLL2126X_PDIV_SHIFT) & PLL2126X_PDIV_MASK;
> +	sdiv = (pll_con >> PLL2126X_SDIV_SHIFT) & PLL2126X_SDIV_MASK;
> +
> +	fvco *= (mdiv + 8);
> +	do_div(fvco, (pdiv + 2) << sdiv);
> +
> +	return (unsigned long)fvco;
> +}
> +
> +/* todo: implement pll2126x clock round rate operation */
> +static long samsung_pll2126x_round_rate(struct clk_hw *hw,
> +				unsigned long drate, unsigned long *prate)
> +{
> +	return -ENOTSUPP;
> +}

If you look at __clk_round_rate() in drivers/clk/clk.c it calls
clk->ops->round_rate() recursively on the parent clock if
CLK_SET_RATE_PARENT flag is set. But if the clock provides round_rate
operation __clk_round_rate() will use it and will cast -ENOTSUPP
you return above to unsigned long. Unless you want to loose some hair
later and waste time on debugging I suggest to either remove all these
callbacks that return -ENOSUPP or implement them properly right away.

I'm just wondering why __clk_round_rate() return value type is unsigned long
despite the struct clk_ops round_rate return long, which might indicate
the negative values round_rate might callback are properly handled by the
clk core code.

Somewhat similar situation is with clk_mux_get_parent() in clk-mux.c
which can return -EINVAL which is then casted to u8.

I still have sending a patch for clk_mux_get_parent() on my todo list.
I guess we could just put a WARN_ON() there and return 0.

> +/* todo: implement pll2126x clock set rate */
> +static int samsung_pll2126x_set_rate(struct clk_hw *hw, unsigned long drate,
> +				unsigned long prate)
> +{
> +	return -ENOTSUPP;
> +}
> +
> +static const struct clk_ops samsung_pll2126x_clk_ops = {
> +	.recalc_rate = samsung_pll2126x_recalc_rate,
> +	.round_rate = samsung_pll2126x_round_rate,
> +	.set_rate = samsung_pll2126x_set_rate,
> +};

--

Thanks,
Sylwester