[alsa-devel] [PATCH] ASoC: core: allow control index different from 0

[Arnaud Pouliquen]

hi

On 10/05/2016 08:59 AM, Takashi Sakamoto wrote:
> Hi,
> 
> On Oct 5 2016 00:36, Arnaud Pouliquen wrote:
>> To differentiate control with same name only device field can be used.
>> But some applications like iecset use control index to differentiate
>> controls linked to several PCM devices or DAIs.
>> This patch suppress index overwriting to allow to use control index field.
> 
> An control element can be identified by two ways from user land:
>  - A combination of name/index/interface/device/subdevice
>  - A numerical identification number (numid)
> 
> In kernel land, some control elements can be managed as one control
> element set, mainly due to saving memory usage. The index represents
> offset of an element in the element set.
> 
> If your patch is applied, callers of snd_soc_cnew() in kernel land can
> assign arbitrary index number to an element set, as the offset of the
> first element.
> 
> According to current implementation, all of first element in element
> sets had index number zero; e.g.
> $ amixer controls
> ...
> numid=52,iface=MIXER,name='byte-element-26'
> numid=53,iface=MIXER,name='byte-element-26',index=1
> numid=54,iface=MIXER,name='byte-element-26',index=2
> numid=55,iface=MIXER,name='byte-element-26',index=3
> numid=56,iface=MIXER,name='byte-element-26',index=4
> ...
> 
> On the other hand, after applying your patch, some element set with
> index larger than zero can exist; e.g.
> $ amixer controls
> ...
> numid=52,iface=MIXER,name='byte-element-26',index=10
> numid=53,iface=MIXER,name='byte-element-26',index=11
> numid=54,iface=MIXER,name='byte-element-26',index=12
> numid=55,iface=MIXER,name='byte-element-26',index=13
> numid=56,iface=MIXER,name='byte-element-26',index=14
> (here, I assign 10 as index to this element set)
> (no other elements with name of 'byte-element-26')
> ...
> 
> In this case, elements with index 0-9 don't exist. This case is a bit
> confusing to application developers, I think.
> 
> I can guess that developers for ALSA SoC part would like to avoid the
> missing elements with index 0-9, by force to start indexing at zero.
> 
> So, if you'd like to start elements with index non-zero, it's better to
> describe your case to get advantages from this patch.

Thanks for the explanation. Here is my use case:
I have a card with 3 output devices:
 hw:0,0 HDMI
	associated controls:
		numid=1,iface=PCM,name='ELD'
		numid=2,iface=PCM,name='IEC958 Playback Default'
		numid=3,iface=PCM,name='Freq.Adjustment'
 hw:0,1 DAC
	associated controls:
		numid=4,iface=PCM,name=' Freq Adjustment'
 hw:0,2 SPDIF
	associated controls:
		numid=5,iface=PCM,name='IEC958 Playback Default'
		numid=6,iface=PCM,name='Freq Adjustment'

My concern is to differentiate the controls associated to the outputs.
Seems that it could be done using device field, but iecset is based on
index (or i missed something?).
Adding an option in iecset to address control by device could solve the
issue... but it is to good way to handle the use case?

Complementary solution would be that control device field corresponds to
PCM device, to allow to address PCM device with args (for instance AESx
params for iec)

This is linked to the management of DAI PCM controls that has already
been discussed in thread associated to this patch:
http://www.spinics.net/lists/alsa-devel/msg47611.html.
I can rework my patch (suppress iec generic control, to simplify it)
but this only treat DAI controls not BE and "dai less" codecs.

Regards
Arnaud

> 
>> Signed-off-by: Arnaud Pouliquen <arnaud.pouliquen at st.com>
>> ---
>>  sound/soc/soc-core.c | 1 -
>>  1 file changed, 1 deletion(-)
>>
>> diff --git a/sound/soc/soc-core.c b/sound/soc/soc-core.c
>> index 4afa8db..39bc1a9 100644
>> --- a/sound/soc/soc-core.c
>> +++ b/sound/soc/soc-core.c
>> @@ -2175,7 +2175,6 @@ struct snd_kcontrol *snd_soc_cnew(const struct snd_kcontrol_new *_template,
>>  	char *name = NULL;
>>  
>>  	memcpy(&template, _template, sizeof(template));
>> -	template.index = 0;
>>  
>>  	if (!long_name)
>>  		long_name = template.name;
> 
> Regards
> 
> Takashi Sakamoto
>