[alsa-devel] [RFC] ALSA scenario API

[Takashi Iwai]

At Wed, 17 Oct 2007 01:12:13 +0100,
Liam Girdwood wrote:
> 
> A little late with this but "better late than never".....I did promise a
> few people that I would post what was imho needed for ALSA scenarios.
> 
> Also, apologies for cross posting, but it some of this probably applies
> to both lists... 
> 
> ALSA probably needs functionality to define and set audio usage
> scenarios. A scenario is simply a list of sound card mixer settings for
> a given use case. This is quite important for mobile devices (e.g.
> Smartphones, GPS, PDA, etc) with complex audio routing and power
> requirements. Some example scenarios could be :-
> 
> "HiFi Headphone Playback"
> "HiFi Speaker Playback"
> "Handset Phone Call"
> "Headset Phone Call"
> etc...
> 
> Scenarios are changed atm by either manually setting each control (e.g.
> using alsamixer or alsa-lib in apps) or by using alsactl to restore a
> scenario. The above two methods are not very portable between devices
> and applications. We would ideally like to just say "set Phone Call
> Handset" for a regular handset phone call or "set HiFi Headphones" for
> MP3 playback.
> 
> ALSA also needs to have functionality to alias mixer controls. This
> would allow a single master volume/mute control to be exposed to user
> space and have it aliased to different hardware controls based upon the
> current scenario. e.g. Many mobile devices have different physical
> volume controls for speaker and headphones. This makes it somewhat
> difficult for user space apps to change the correct volume (without
> being bound to the hardware). We would like to be able to just say "set
> volume to 0dB" and have the alias take care of the correct physical
> CODEC mixer for the current scenario.
> 
> Proposal:
> ========
> 
> 1. Add some new scenario functions to (s)alsa-lib. e.g.
> 
> snd_scenario_set();
> snd_scenario_get();
> snd_scenario_list();
> 
> _set() would set the scenario e.g.
> 
> snd_scenario_set(card, SND_VOICE_CALL_HANDSET);
> 
> We would need to have a list of standard scenarios so that applications
> could be more portable. Vendor scenario's should also be supported.
> 
> _get() would return the current scenario.
> 
> _list() would return a list of supported scenarios for the given device.
> 
> 
> 2. Add some new mixer alias functions to set/get the current "master"
> volume/mute. e.g.
> 
> snd_kcontrol *snd_scenario_volume_set(card, type, volume);
> snd_kcontrol *snd_scenario_switch_set(card, type, switch);
> 
> where type is playback, capture, sidetone, bypass, etc
> 
> This would set the volume for the current scenario and would guarantee
> always to get/set the _correct_ volume/switch.
> 
> 3. Add a new function to return the active PCM for a given scenario.
> e.g.
> 
> snd_scenario_get_pcm(card, scenario);
> 
> Would return the given PCM for a particular scenario. i.e. the WM9713
> driver has PCM 0.0 for high quality (higher power) audio (MP3), PCM 0.1
> for lower power voice and PCM 0.2 for lowest power and quality system
> sounds.
> 
> Note: scenarios may need to be more tightly bound to PCM's rather than
> cards. 
> 
> 4. Define a file format to represent each scenario. Probably base this
> upon alsactl's format as alsamixer could be used to quickly generate
> good settings and then save them with alsactl. Alias, scenario names and
> PCM would need to be added later somehow.
> 
> 5. (Optional) Allow scenarios to define active kcontrols. This would
> allow tools like alsamixer to only display mixer controls for the given
> scenario. Controls that don't effect the scenario would not be rendered.
> (similar to alsamixers Playback|Capture|All tab)
> 
> 6. (OHM list) Add input driver (or similar) for sound card user events.
> We need to be able to send events like "Headphone Jack Insertion", "Mic
> Jack Insertion", etc to user space. This would allow changing of
> scenario and power configuration depending on policy and configuration. 
> 
> Just my 2p
> 
> Unfortunately I'm not in a position to implement any of this
> functionality. However, it's probably useful to folks working on mobile
> devices. Maybe a good project for Google SoC.

It's an interesting idea.  Similar things are needed vastly for other
purposes, such as speaker setup.  But, my first thinking is whether we
need this in the upper (user-space alsa-lib) level or in the driver
level.

Certainly the driver must know at best what scenario is available and
how it should react.  I've thought this kind of thing could be
implemented better, more elegantly, on user-space.  But, thinking
twice, the driver implementation can be much easier in many cases.
Suppose the switching (aliasing) the control name depending on the
given scenario.  The scenario state is global and must be stored in a
global position.  The easiest one is in the driver -- this will solve
the access permission and race problem automatically, too, which the
shared-memory solution always suffers.

Thus, it would be maybe good to implement the secnario as a control
element (not necessarily as a mixer element but we can give a
different IFACE attribute, e.g. SND_CTL_ELEM_IFACE_CARD).  Then the
above snd_scenario_set/get/list functions becomes simple snd_ctl_*().


About PCM: if we implement again this kind of things in the driver, we
may have a virtual PCM stream for the default use and other fixed PCM
streams.  They can be opened exclusively, and the default stream opens
different H/W stream according to the current scenario.  (BTW, still a
question remains: when the scenario switch happens on the fly during a
PCM stream being used, what should be done?)

Or, simply alsa-lib checks the current scenario value and chooses the
appropriate preset setup.  This would work well (at least for
alsa-lib, not salsa-lib - such a complicated implementation is outside
the scope of salsa-lib).  This user-space solution (for PCM) has
another advantage that it's easier to handle PCM setups for multiple
streams.  For example, suppose we use "scenario" for speaker setup.
emu10k1 uses multiple PCM streams for 4.0/5.1 surrounds, and the
driver-side implementation wouldn't work unless it reflects to
alsa-lib setup.


My another $0.02.


Takashi