Re: [PATCH] ASoC: soc-ops: Correct the max value for clamp in soc_mixer_reg_to_ctl()

["Péter Ujfalusi" <peter.ujfalusi@linux.intel.com>]

On 17/12/2025 15:42, Charles Keepax wrote:
> On Wed, Dec 17, 2025 at 03:13:45PM +0200, Péter Ujfalusi wrote:
>> On 17/12/2025 14:47, Charles Keepax wrote:
>>> On Wed, Dec 17, 2025 at 02:06:23PM +0200, Peter Ujfalusi wrote:
>>>>  sound/soc/soc-ops.c | 2 +-
>>>>  1 file changed, 1 insertion(+), 1 deletion(-)
>>>>
>>>> diff --git a/sound/soc/soc-ops.c b/sound/soc/soc-ops.c
>>>> index ce86978c158d..6a18c56a9746 100644
>>>> --- a/sound/soc/soc-ops.c
>>>> +++ b/sound/soc/soc-ops.c
>>>> @@ -148,7 +148,7 @@ static int soc_mixer_reg_to_ctl(struct soc_mixer_control *mc, unsigned int reg_v
>>>>  	if (mc->sign_bit)
>>>>  		val = sign_extend32(val, mc->sign_bit);
>>>>  
>>>> -	val = clamp(val, mc->min, mc->max);
>>>> +	val = clamp(val, mc->min, mc->min + max);
>>>
>>> This won't work, for an SX control it is perfectly valid for
>>> the value read from the register to be smaller than the minimum
>>> value specified in the control.
>>
>> Hrm, so an SX control returns sort of rand() and the value have no
>> correlation to min or max?
> 
> lol, yes exactly :-) arn't they great
> 
>> The value can wrap at any random value to 0 and continue from 0 up to
>> some value, which is the max?
> 
> Mostly correct, not any random value it wraps at the mask.
> 
>> How this is in practice for the cs42l43' Headphone Digital Volume?
>> SOC_DOUBLE_SX_TLV("Headphone Digital Volume", CS42L43_HPPATHVOL,
>> 	  CS42L43_AMP3_PATH_VOL_SHIFT, CS42L43_AMP4_PATH_VOL_SHIFT,
>> 	  0x11B, 229, cs42l43_headphone_tlv),
>>
>> min=283
>> max=229
>> shifts: 0 and 16
>> masks are 0x1ff
>>
>> if you step 229 from 283 then you reach 0x1ff, this is the max the mask
>> can cover.
> 
> Not quite your maths is off by one, 229 + 283 = 512 = 0x200,
> which is then &ed with the mask to get 0x0. Which on the cs42l43
> headphones a value of 0x0->0dB. Stepping 1 back from that would
> give you 0x1FF->-0.5dB.
> 
>>> I often think of it in terms of a 2's compliement number
>>> with an implicit sign bit.
>>
>> I see, but why???
> 
> Mostly because hardware people love to wind me up, I assume. But
> more seriously, imagine an 4-bit signed number volume control
> with 5 values:
> 
> 0xE -> -2 -> -2dB
> 0xF -> -1 -> -1dB
> 0x0 ->  0 -> 0dB
> 0x1 ->  1 -> 1dB
> 0x2 ->  2 -> 2dB
> 
> Super, a very sensible control, but wait being a good hardware
> engineer you realise you don't need 4 bits to represent 5 values
> you can get away with 3 bits for that and save like 2 gates
> resulting in an ice cream and a plaque from your manager. So
> you drop the sign bit giving you:
> 
> 0x6 -> -2dB
> 0x7 -> -1dB
> 0x0 -> 0dB
> 0x1 -> 1dB
> 0x2 -> 2dB

I must say, wow.
Being a SW guy I would probably done this differently:
0x0 -> -2dB
0x1 -> -1dB
0x2 -> 0dB
0x3 -> 1dB
0x4 -> 2dB

> This then results in an SX control with a minimum of 0x6 and a
> mask of 0x7.

then the comment at info() is hard to match still.

static const DECLARE_TLV_DB_RANGE(sx_thing,
	6, 7, TLV_DB_SCALE_ITEM(-2000, -1000, 0),
	0, 2, TLV_DB_SCALE_ITEM(0, 1000, 0)
};

is sort of the same, no?

Thanks for the explanation, fascinating!

-- 
Péter