[alsa-devel] Is this bug ? dpcm_prune_paths()

[Pierre-Louis Bossart]

On 10/8/19 1:53 AM, Kuninori Morimoto wrote:
> 
> Hi ALSA ML.
> 
> I'm checking soc-pcm.c, and I noticed strange code at
> dpcm_prune_paths().
> 
> 	static int dpcm_prune_paths(...)
> 	{
> 		...
> 		/* Destroy any old FE <--> BE connections */
> (1)		for_each_dpcm_be(fe, stream, dpcm) {
> 			unsigned int i;
> 
>   -			/* is there a valid CPU DAI widget for this BE */
>   |			widget = dai_get_widget(dpcm->be->cpu_dai, stream);
>   |
> (A)			/* prune the BE if it's no longer in our active list */
>   |			if (widget && widget_in_list(list, widget))
>   |(a)				continue;
>   -
>   |			/* is there a valid CODEC DAI widget for this BE */
>   |(2)			for_each_rtd_codec_dai(dpcm->be, i, dai) {
>   |				widget = dai_get_widget(dai, stream);
> (B)
>   |				/* prune the BE if it's no longer in our active list */
>   |				if (widget && widget_in_list(list, widget))
>   |(b)					continue;
>   -			}
> 
>   -			...
>   |			dpcm->state = ...
> (C)			dpcm->be->dpcm[stream].runtime_update = ...
>   |			prune++;
>   -		}
> 		...
> 	}
> 
> In my understanding, (A) part is for CPU, and (B) part is for Codec.
> Guessing from (a), I think it want to skip setup (C) if CPU widget
> exist and matches to list.
> If so, I guess (b) is maybe bug ?
> This continue is for (2) loop, thus, it is totally do nothing now.
> But maybe it want to be continue for (1) loop ?

Nice catch. This looks like a problem added during the transition to 
multi-codec.

The code before 2e5894d73789 ('ASoC: pcm: Add support for DAI 
multicodec') was this:

		/* is there a valid CODEC DAI widget for this BE */
		widget = dai_get_widget(dpcm->be->codec_dai, stream);

		/* prune the BE if it's no longer in our active list */
		if (widget && widget_in_list(list, widget))
			continue;

and the addition of the loop for the codec seems incorrect

		/* is there a valid CODEC DAI widget for this BE */
		for (i = 0; i < dpcm->be->num_codecs; i++) {
			struct snd_soc_dai *dai = dpcm->be->codec_dais[i];
			widget = dai_get_widget(dai, stream);

			/* prune the BE if it's no longer in our active list */
			if (widget && widget_in_list(list, widget))
				continue;
		}

the continue was not meant to continue the for loop on num_codecs, but 
the outer loop for dpcm.