[alsa-devel] [PATCH 7/7] ASoC: Instantiate all DAPM widgets at once

[Lars-Peter Clausen]

On 04/28/2011 10:07 PM, Mark Brown wrote:
> On Thu, Apr 28, 2011 at 06:46:13PM +0200, Lars-Peter Clausen wrote:
>> Currently after each codec or auxdev has been probed its DAPM widgets are
>> instantiated.
>>
>> Since widgets are kept in a per card list and routes can interconnect between
>> widgets from different DAPM contexts on the same card it is possible that
>> snd_soc_dapm_new_widgets is run on an incomplete DAPM configuration which might
>> cause unnecessary register writes and/or cause audible problems.
>>
>> This patch addresses the issue by instantiating all widgets from all DAPM contexts
>> after all components of the card have been initialized.
> 
> Except it doesn't because...

Well, a driver can of course decide that it wants to instantiate the widgets
now, but...
> 
>> +static inline int snd_soc_dapm_new_widgets(struct snd_soc_dapm_context *dapm)
>> +{
>> +	return snd_soc_dapm_card_new_widgets(dapm->card);
>> +}
> 
> ...we can still instantiate widgets at any time, and lots of code will
> do just that (and needs to do that because it needs the widgets to be
> present in order to allow paths to be added).

... there are exactly 5 codec drivers which call snd_soc_dapm_new_widgets, all
of them call this at the end of their codec probe callback, which means it can
be dropped, because the core will call it again just right after the codecs
probe callback (without this patch).
snd_soc_dapm_new_widgets was mainly left in to let drivers register widgets
after the initialization phase which could be required for dynamic
reconfiguration of some DSPs.

Routes can be added just fine without the widgets being instantiated. In fact
snd_soc_dapm_new_widgets requires the routes to be setup to initialize some
widgets(mixers,muxs) properly.

> 
> There's also the fact that this mixes in the functional change to move
> the instantiation of the widgets around with the argument change and a
> rename, it'd be better to split the two of those out so we can clearly
> see the functional change.

ok.