Re: [PATCH - IOPLUG DRAIN 0/2]

[Takashi Iwai <tiwai@suse.de>]

On Wed, 28 Mar 2018 10:42:50 +0200,
Wischer, Timo (ADITG/ESB) wrote:
> 
> @@ -67,6 +73,12 @@ static void snd_pcm_ioplug_hw_ptr_update(snd_pcm_t *pcm)
>                         delta = wrap_point + hw - io->last_hw;
>                 }
>                 snd_pcm_mmap_hw_forward(io->data->pcm, delta);
> +               /* stop the stream if all samples are drained */
> +               if (io->data->state == SND_PCM_STATE_DRAINING) {
> +                       avail = snd_pcm_mmap_avail(pcm);
> +                       if (avail >= pcm->buffer_size)
> +                               snd_pcm_ioplug_drop(pcm);
> +               }
>                 io->last_hw = (snd_pcm_uframes_t)hw;
>         } else
>                 io->data->state = SNDRV_PCM_STATE_XRUN;
> In case of draining drop has to be called because draining is done

OK, makes sense.


> @@ -488,20 +500,66 @@ static int snd_pcm_ioplug_drop(snd_pcm_t *pcm)
>         return 0;
>  }
> 
> +static int ioplug_drain_via_poll(snd_pcm_t *pcm)
> +{
> +       ioplug_priv_t *io = pcm->private_data;
> +       int err;
> +
> +       /* in non-blocking mode, leave application to poll() by itself */
> +       if (io->data->nonblock)
> +               return -EAGAIN;
> In case of nonblock snd_pcm_ioplug_hw_ptr_update() will not be called by the user
> Therefore the user will never detect that draining is done.
> I fear snd_pcm_ioplug_hw_ptr_update() has also to be called in nonblock mode here.

For the non-blocking mode, it's not supposed that drain() is called
multiple times.  Instead, it should do the loop of snd_pcm_wait(), and
status check, as ioplug_drain_vai_poll() actually does.

Yes, it's weird, but it's the standard way for non-blocking mode, even
for the kernel drivers.  So, the practical recommendation for draining
is to temporarily switch to blocking mode before calling
snd_pcm_drain().
	

> Will you create the patch and apply it to master or
> is there anything which I have to do?

I'll cook up the proper patch and submit to ML soon later.
Then it'll be merged to git tree, and you can work on it.


thanks,

Takashi