[4/7] clk: samsung: exynos5433: fix PLL rates

Commit Message

Sylwester Nawrocki/Kernel \(PLT\) /SRPOL/Staff Engineer/Samsung Electronics Feb. 15, 2018, 2:23 p.m. UTC

Hi Chanwoo,

On 02/14/2018 06:45 AM, Chanwoo Choi wrote:
> Exynos5433 TRM shows different PLL frequency from the calculated value. 
> Actually, I'm not sure which value is the correct value between TRM's> value and real calculated value.
> 
> If we only consider the equation, I agree that we have to replace them > with real calculated value with equation. But, TRM shows the diagram of 
> clock domain which contains the multiple IPs. And each diagram specifies
> the required source clock rate of each child IP. Maybe, this source clock 
> rate might be calculated from the PLL rate provided by TRM. (even if it's 
> different from real calculated value).
> 
> Although the real rate is a little bit incorrect, we might better to keep 
> PLL rate provided by TRM in order to get the required source clock of child 
> IPs as the TRM. If we modify the PLL rate provided by TRM, the source clock 
> rate of child IP might be different from TRM.
> 
> Basically, I have no objection of this patch. Just I think that need to 
> check it.

In order to have the clk API behave properly we need to have in these tables
frequency values matching the values returned by the recalc_rate callback 
for given P, M, S, K coefficients.  Alternatively, we could decrease precision 
of the PLL's recalc_rate() callbacks by rounding its return value, but I'm not 
sure it's a good approach. It might be better to have clear indication which 
P, M, S, K values yield higher frequency error, even though the differences at 
those least significant digit positions might be meaningless, considering 
frequency tolerance of the root oscillator itself. 

For example, difference of 10 Hz where Fout = 400 MHz is only 0.025 ppm 
(0.0000025 %), when the tolerance of the root oscillator itself will be 
usually in range 10...100 ppm.

In the TRMs the least significant digits are simply ignored AFAIU. I guess
we could apply similar rounding in the recalc_rate callback of the fractional 
PLLs and the problem would also be solved, e.g.

I think we need a bit more detailed commit message for this patch series, 
so it's clear what are the issues with current values. 

> On 2018년 02월 13일 22:40, Andrzej Hajda wrote:
>> Declared rates did not match rates generated by PLL.
>> As a result driver behaved inconsitently.
>>
>> Signed-off-by: Andrzej Hajda <[e-mail]>
>> ---
>>  drivers/clk/samsung/clk-exynos5433.c | 12 ++++++------
>>  1 file changed, 6 insertions(+), 6 deletions(-)
>>
>> diff --git a/drivers/clk/samsung/clk-exynos5433.c b/drivers/clk/samsung/clk-exynos5433.c
>> index db270908037a..335bebfa21c0 100644
>> --- a/drivers/clk/samsung/clk-exynos5433.c
>> +++ b/drivers/clk/samsung/clk-exynos5433.c
>> @@ -729,7 +729,7 @@ static const struct samsung_pll_rate_table exynos5433_pll_rates[] __initconst =
>>  	PLL_35XX_RATE(800000000U,  400, 6,  1),
>>  	PLL_35XX_RATE(733000000U,  733, 12, 1),
>>  	PLL_35XX_RATE(700000000U,  175, 3,  1),
>> -	PLL_35XX_RATE(667000000U,  222, 4,  1),
>> +	PLL_35XX_RATE(666000000U,  222, 4,  1),
>>  	PLL_35XX_RATE(633000000U,  211, 4,  1),
>>  	PLL_35XX_RATE(600000000U,  500, 5,  2),
>>  	PLL_35XX_RATE(552000000U,  460, 5,  2),
>> @@ -757,12 +757,12 @@ static const struct samsung_pll_rate_table exynos5433_pll_rates[] __initconst =
>>  /* AUD_PLL */
>>  static const struct samsung_pll_rate_table exynos5433_aud_pll_rates[] __initconst = {
>>  	PLL_36XX_RATE(400000000U, 200, 3, 2,      0),
>> -	PLL_36XX_RATE(393216000U, 197, 3, 2, -25690),
>> +	PLL_36XX_RATE(393216003U, 197, 3, 2, -25690),
>>  	PLL_36XX_RATE(384000000U, 128, 2, 2,      0),
>> -	PLL_36XX_RATE(368640000U, 246, 4, 2, -15729),
>> -	PLL_36XX_RATE(361507200U, 181, 3, 2, -16148),
>> -	PLL_36XX_RATE(338688000U, 113, 2, 2,  -6816),
>> -	PLL_36XX_RATE(294912000U,  98, 1, 3,  19923),
>> +	PLL_36XX_RATE(368639991U, 246, 4, 2, -15729),
>> +	PLL_36XX_RATE(361507202U, 181, 3, 2, -16148),
>> +	PLL_36XX_RATE(338687988U, 113, 2, 2,  -6816),
>> +	PLL_36XX_RATE(294912002U,  98, 1, 3,  19923),
>>  	PLL_36XX_RATE(288000000U,  96, 1, 3,      0),
>>  	PLL_36XX_RATE(252000000U,  84, 1, 3,      0),
>>  	{ /* sentinel */ }
--
To unsubscribe from this list: send the line "unsubscribe linux-clk" in
the body of a message to [e-mail]
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Patch

diff --git a/drivers/clk/samsung/clk-pll.c b/drivers/clk/samsung/clk-pll.c
index 1c4c7a3039f1..d7a265827e65 100644
--- a/drivers/clk/samsung/clk-pll.c
+++ b/drivers/clk/samsung/clk-pll.c
@@ -300,6 +300,8 @@  static unsigned long samsung_pll36xx_recalc_rate(struct clk_hw *hw,
        do_div(fvco, (pdiv << sdiv));
        fvco >>= 16;
 
+       fvco = ((fvco + 50) / 100) * 100;
+
        return (unsigned long)fvco;
 }