[alsa-devel] [PATCH] ASoC: DAPM: Add support for multi register mux

[Lars-Peter Clausen]

On 04/02/2014 05:26 PM, Songhee Baek wrote:
>> -----Original Message-----
>> From: Lars-Peter Clausen [mailto:lars at metafoo.de]
>> Sent: Wednesday, April 02, 2014 12:02 AM
>> To: Songhee Baek
>> Cc: Arun Shamanna Lakshmi; lgirdwood at gmail.com; broonie at kernel.org;
>> swarren at wwwdotorg.org; perex at perex.cz; tiwai at suse.de; alsa-
>> devel at alsa-project.org; linux-kernel at vger.kernel.org
>> Subject: Re: [PATCH] ASoC: DAPM: Add support for multi register mux
>>
>> On 04/02/2014 08:56 AM, Songhee Baek wrote:
>>>
>>>
>>>> -----Original Message-----
>>>> From: Lars-Peter Clausen [mailto:lars at metafoo.de]
>>>> Sent: Tuesday, April 01, 2014 11:47 PM
>>>> To: Songhee Baek
>>>> Cc: Arun Shamanna Lakshmi; lgirdwood at gmail.com; broonie at kernel.org;
>>>> swarren at wwwdotorg.org; perex at perex.cz; tiwai at suse.de; alsa-
>>>> devel at alsa-project.org; linux-kernel at vger.kernel.org
>>>> Subject: Re: [PATCH] ASoC: DAPM: Add support for multi register mux
>>>>
>>>> On 04/02/2014 08:17 AM, Songhee Baek wrote:
>>>>>> -----Original Message-----
>>>>>> From: Lars-Peter Clausen [mailto:lars at metafoo.de]
>>>>>> Sent: Tuesday, April 01, 2014 11:00 PM
>>>>>> To: Arun Shamanna Lakshmi
>>>>>> Cc: lgirdwood at gmail.com; broonie at kernel.org;
>>>> swarren at wwwdotorg.org;
>>>>>> perex at perex.cz; tiwai at suse.de; alsa-devel at alsa-project.org; linux-
>>>>>> kernel at vger.kernel.org; Songhee Baek
>>>>>> Subject: Re: [PATCH] ASoC: DAPM: Add support for multi register mux
>>>>>>
>>>>>> On 04/01/2014 08:26 PM, Arun Shamanna Lakshmi wrote:
>>>>>> [...]
>>>>>>>>> diff --git a/sound/soc/soc-dapm.c b/sound/soc/soc-dapm.c index
>>>>>>>>> c8a780d..4d2b35c 100644
>>>>>>>>> --- a/sound/soc/soc-dapm.c
>>>>>>>>> +++ b/sound/soc/soc-dapm.c
>>>>>>>>> @@ -514,9 +514,9 @@ static int dapm_connect_mux(struct
>>>>>>>> snd_soc_dapm_context *dapm,
>>>>>>>>>       	unsigned int val, item;
>>>>>>>>>       	int i;
>>>>>>>>>
>>>>>>>>> -	if (e->reg != SND_SOC_NOPM) {
>>>>>>>>> -		soc_widget_read(dest, e->reg, &val);
>>>>>>>>> -		val = (val >> e->shift_l) & e->mask;
>>>>>>>>> +	if (e->reg[0] != SND_SOC_NOPM) {
>>>>>>>>> +		soc_widget_read(dest, e->reg[0], &val);
>>>>>>>>> +		val = (val >> e->shift_l) & e->mask[0];
>>>>>>>>>       		item = snd_soc_enum_val_to_item(e, val);
>>>>>>>>
>>>>>>>> This probably should handle the new enum type as well. You'll
>>>>>>>> probably need some kind of flag in the struct to distinguish
>>>>>>>> between the two enum types.
>>>>>>>
>>>>>>> Any suggestion on the flag name ?
>>>>>>>
>>>>>>
>>>>>> How about 'onehot'?
>>>>>>
>>>>>> [...]
>>>>>>>>> +		reg_val = BIT(bit_pos);
>>>>>>>>> +	}
>>>>>>>>> +
>>>>>>>>> +	for (i = 0; i < e->num_regs; i++) {
>>>>>>>>> +		if (i == reg_idx) {
>>>>>>>>> +			change = snd_soc_test_bits(codec, e->reg[i],
>>>>>>>>> +							e->mask[i],
>>>>>>>> reg_val);
>>>>>>>>> +
>>>>>>>>> +		} else {
>>>>>>>>> +			/* accumulate the change to update the
>>>> DAPM
>>>>>>>> path
>>>>>>>>> +			    when none is selected */
>>>>>>>>> +			change += snd_soc_test_bits(codec, e-
>>>>> reg[i],
>>>>>>>>> +							e->mask[i], 0);
>>>>>>>>
>>>>>>>> change |=
>>>>>>>>
>>>>>>>>> +
>>>>>>>>> +			/* clear the register when not selected */
>>>>>>>>> +			snd_soc_write(codec, e->reg[i], 0);
>>>>>>>>
>>>>>>>> I think this should happen as part of the DAPM update sequence
>>>>>>>> like you had earlier. Some special care should probably be take
>>>>>>>> to make sure that you de-select the previous mux input before
>>>>>>>> selecting the new one if the new one is in a different register than
>> the previous one.
>>>>>>>
>>>>>>> I am not sure I follow this part. We are clearing the 'not selected'
>>>>>>> registers before we set the one we want. Do you want us to loop
>>>>>>> the logic of soc_dapm_mux_update_power for each register ? or do
>>>>>>> you want to change the dapm_update structure so that it takes all
>>>>>>> the regs, masks, and values together ?
>>>>>>
>>>>>> The idea with the dapm_update struct is that the register updates
>>>>>> are done in the middle of the power-down and power-up sequence.
>> So
>>>>>> yes, change the dapm_update struct to be able to hold all register
>>>>>> updates and do all register updates in dapm_widget_update. I think
>>>>>> an earlier version of your patch already had this.
>>>>>
>>>>> Is the change similar to as shown below?
>>>>>
>>>>> for (reg_idx = 0; reg_idx < e->num_regs; reg_idx++) {
>>>>> 	val = e->values[item * e->num_regs + reg_idx];
>>>>> 	ret = snd_soc_update_bits_locked(codec, e->reg[reg_idx],
>>>>> 				e->mask[reg_idx], val);
>>>>> 	if (ret)
>>>>> 	return ret;
>>>>> }
>>>>>
>>>>> During updating of the register's value, the above change can create
>>>>> non-zero value in two different registers (very short transition) as
>>>>> Mark mentioned for that change so we need to clear register first
>>>>> before writing the desired value in the register.
>>>>>
>>>>> Should we add the clearing all registers and write the mux value in
>>>>> desired register in the update function?
>>>>>
>>>>
>>>> In dapm_update_widget() you have this line:
>>>>
>>>>     ret = soc_widget_update_bits(w, update->reg, update->mask, update-
>>>>> val);
>>>>
>>>> That needs to be done for every register update. When you setup the
>>>> update struct you need to make sure that the register clears come
>>>> before the register set.
>>>>
>>>> E.g. if you have register 0x3, 0x4, 0x5 and you select a bit in
>>>> register 0x4 it should look like this.
>>>>
>>>> update->reg[0] = 0x3;
>>>> update->val[0] = 0x0;
>>>> update->reg[1] = 0x5;
>>>> update->val[1] = 0x0;
>>>> update->reg[2] = 0x4;
>>>> update->val[2] = 0x8;
>>>>
>>>> When you set a bit in register 0x3 it should look like this:
>>>>
>>>> update->reg[0] = 0x4;
>>>> update->val[0] = 0x0;
>>>> update->reg[1] = 0x5;
>>>> update->val[1] = 0x0;
>>>> update->reg[2] = 0x3;
>>>> update->val[2] = 0x1;
>>>>
>>>> So basically the write operation goes into update->reg[e->num_regs-1]
>>>> the clear operations go into the other slots before that.
>>>
>>> Does update reg/val array have the writing sequence, is it correct?
>>> And can I assume that update struct has reg/val/mask arrays not pointers?
>>
>> Right now the update struct does not have support for multiple register
>> writes. That's up to you to implement this. I guess making it an array for now
>> should be fine. But you need to add some safety checks to make sure that
>> num_regs is not larger or equal to the array size.
>
> I think that the dapm update struct needs to have reg[2]/val[2]/mask[2].
> Because the mux is one-hot coded, only one register has a non-zero value.
> So reg[0] will contain the register to be clear and reg[2] has selected register
> to be set.
> How about your opinion for this?

Should work (in theory).

- Lars