Re: [PATCH v14 18/18] media: i2c: ds90ub953: Support non-sync mode

[Tomi Valkeinen <tomi.valkeinen@ideasonboard.com>]

On 19/06/2023 13:48, Andy Shevchenko wrote:
> On Mon, Jun 19, 2023 at 12:00:57PM +0300, Tomi Valkeinen wrote:
>> On 16/06/2023 17:47, Andy Shevchenko wrote:
>>> On Fri, Jun 16, 2023 at 04:59:22PM +0300, Tomi Valkeinen wrote:
>>>> Add support for FPD-Link non-sync mode with external clock. The only
>>>> thing that needs to be added is the calculation for the clkout.
> 
> ...
> 
>>>> +	switch (priv->mode) {
>>>> +	case UB953_MODE_SYNC:
>>>> +		if (priv->hw_data->is_ub971)
>>>> +			return priv->plat_data->bc_rate * 160ull;
>>>> +		else
>>>> +			return priv->plat_data->bc_rate / 2 * 160ull;
>>>
>>> Redundant 'else'.
>>
>> True, but I like the symmetry in:
>>
>> if (foo)
>> 	return 123;
>> else
>> 	return 321;
> 
> At the same time it will be symmetry with other switch-case(s). That's why the
> question about fallthrough below.
> 
>>> Do I understand correctly you don't want to fallthrough because it will give
>>> ±160 in the rate (depending if it's even or odd)?
>>
>> Sorry, can you clarify? Fallthrough to what?
> 
> To the below case since '/ 2 * 160 ~= *80'. Why ~ because it might give
> off-by-one error due to even/odd input.

The below case is different. "priv->plat_data->bc_rate" vs 
"clk_get_rate(priv->clkin)".

As to the order of the calculation (/ 2 * 160 versus * 160 / 2), 
generally speaking, I have never figured out what are the correct ways 
to calculate clock rates.

I wrote "x / 2 * 160" as that's what the documentation gives (there's a 
hardware /2 divider in non-ub971 chips, followed by a 160 multiplier). 
But does the documentation presume that the calculation is done 
precisely, not in integers? If so, "x * 160 / 2" would be better (but 
then, do we need to round?). Or does the /2 hardware divider basically 
actually work as a an integer division, in case "x / 2 * 160" is the 
correct one.

>>>> +	case UB953_MODE_NONSYNC_EXT:
>>>> +		/* CLKIN_DIV = 1 always */
>>>> +		return clk_get_rate(priv->clkin) * 80ull;
> 

  Tomi