Re: [RFC PATCH 0/4] Add support for detection

[Jaroslav Kysela <perex@xxxxxxxx>]

On 18. 10. 24 16:16, Amadeusz Sławiński wrote:

> Our use case is to:

Inlined my proposal:

> 1. Program DSP with all pipelines it needs to implement whole scenario
> (hw_params)

SNDRV_PCM_IOCTL_HW_PARAMS

> 2. Perform additional configuration from usespace

Can be done using control API (/dev/snd/controlC*). E.g. We have already:

numid=19,iface=PCM,name='ELD',device=3
numid=32,iface=PCM,name='Playback Channel Map',device=3

Those controls can be R/W as you like.

> 3. Start detection pipelines

SNDRV_PCM_IOCTL_PREPARE

> 4. Wait for event to happen (it can take however long necessary, for
> example hours)

Can be done using control API (/dev/snd/controlC*). There is a mechanism for 
event handling already.

Something like (boolean type):

iface=PCM,device=2,name="Event X"   # first event type
iface=PCM,device=2,name="Event Y"   # second event type ...

The PCM stream may be active only when at least one event from the above set 
is active.

> 5. When event happens start drain pipelines and process data in
> userspace likes in standard capture scenario

SNDRV_PCM_IOCTL_START

The driver may set the correct timestamp when the capturing was started. The 
sample buffers are already allocated and ready for transfers, so we can assume 
that the time window between the event notification (event) and START ioctl is 
small, so no samples should be lost.

> 6. When no more data is needed, either stop drain pipelines and go back
> to 3 or just close stream

SNDRV_PCM_IOCTL_STOP / DRAIN, SNDRV_PCM_IOCTL_HW_PARAMS or close(fd)

Basically, I think that this extension can be implemented with minimal changes 
to the PCM API - it's just about to slightly extend existing APIs.

				Jaroslav

--
Jaroslav Kysela <perex@xxxxxxxx>
Linux Sound Maintainer; ALSA Project; Red Hat, Inc.