[alsa-devel] RFC: PCM extra attributes

[Takashi Iwai]

At Fri, 19 Jun 2009 12:29:17 +0200 (CEST),
Jaroslav Kysela wrote:
> 
> On Fri, 19 Jun 2009, Takashi Iwai wrote:
> 
> > At Fri, 19 Jun 2009 10:47:30 +0200 (CEST),
> > Jaroslav Kysela wrote:
> >>
> >> On Fri, 19 Jun 2009, Takashi Iwai wrote:
> >>
> >>> Hi,
> >>>
> >>> this is yet another topic I'm (currently) working on -- the addition
> >>> of PCM ioctls to get/set some extra attributes.  Basically, it adds
> >>> two simple ioctls for getting/setting extra attributes to the PCM
> >>> substream.  The attribute has a sort of TLV form,
> >>>
> >>>  /* PCM extra attributes */
> >>>  struct snd_pcm_attr {
> >>>  	unsigned int type;	/* SNDRC_PCM_TYPE_ATTR_XXX */
> >>>  	unsigned int len;	/* GET R: the max elements in value array
> >>>  				 *     W: the actually written # elements
> >>>  				 * SET R/W: # elements to store
> >>>  				 */
> >>>  	unsigned int value[0];	/* value(s) to read / write */
> >>>  };
> >>>
> >>> And corresponding two ioctls
> >>>  #define SNDRV_PCM_IOCTL_GET_ATTR	_IOWR('A', 0x14, struct snd_pcm_attr)
> >>>  #define SNDRV_PCM_IOCTL_SET_ATTR	_IOWR('A', 0x15, struct snd_pcm_attr)
> >>
> >> I would prefer to implement similar TLV implementation as for the control
> >> API. The amount of information for reading (get) will be small, so
> >> filtering in this direction is not necessary. Also, common parts of
> >> implementation (future merging of more TLVs to compounds) can be shared.
> >
> > Actually it's a sort of TLV.  You see exactly it in snd_pcm_attr
> > struct, no? :)
> >
> > And, thinking twice after posting (that's a good effect of posting to
> > ML, BTW), I feel that using a callback would be a better way, such as
> > re-using the existing ops->ioctl with a new cmd tag rather than the
> > statically assigned thing.
> >
> > A similar method like control TLV can be used, too.  However, a
> > distinct from the existing control TLV is that this is intended for
> > just one type of information while the control TLV is supposed to
> > contain everything in a single shot.
> >
> > That is, this is a query with a key.  In that sense, sharing a small
> > amount of control TLV code (about 10 lines) doesn't give a big
> > benefit.  In anyways, it's a implementation detail, so one could
> > optimize somehow, though...
> 
> I don't mean current implementation. TLVs can be nested. In this case, we 
> need a set of functions which operates with TLVs (merging). These 
> functions can be shared. It's also possible to share TLV code in 
> the user space (search). But it's really implementation detail. We should 
> focus on ioctl definitions now.

Well, you still missed the difference between the control TLV and the
PCM attr I proposed.  The former is one-shot information without any
query key while the latter is a reply to a query key.  The former
doesn't work well if you wan to get dynamic data.

Also, see that the data value structure is undefined.  The returned
data itself can be nested TLV components, if any.


thanks,

Takashi