[alsa-devel] [RFC] Initialize volumes of HD-audio slave ctls

[David Henningsson]

On 03/09/2012 08:08 AM, Takashi Iwai wrote:
> At Fri, 09 Mar 2012 00:55:53 +0100,
> David Henningsson wrote:
>>
>> On 03/08/2012 04:27 PM, Takashi Iwai wrote:
>>> Hi,
>>>
>>> the patch below is an attempt to initialize the volume / mutes of
>>> slave controls (such as "Headphone", "Speaker") with vmaster in
>>> HD-audio, so that the sound can come out only by changing the master
>>> volume/mute.
>>>
>>> We have thought that such initializations could be done well in
>>> alsactl init, but it seems that not everyone installs the latest and
>>> greatest alsactl, and there is always a risk that any new controls may
>>> be added before alsactl is updated and released.  Since the master
>>> volume is set muted, the risk by this change should be low.
>>>
>>> patch_cirrus.c still doesn't support this because it's handling
>>> vmaster by itself, but it can be fixed later, too.
>>>
>>> If anyone has a concern by this, please let me know.
>>
>> I think it is good to initialise the volume controls to a sane value in
>> general. I guess the risk of causing unnecessary pops is possible, but
>> not common.
>
> Actually this doesn't initialize the all volumes to "sane" values.  As
> mentioned, Master is still zero/muted.  Only slaves are raised so that
> you don't need to adjust two or more volumes but only master for
> getting some sound.
>
> The reason I wrote it is because of a likely happening scenario:
> I modify the driver to support individual speaker and headphone
> volumes from a single volume.  Then people without proper alsactl init
> setup would get both controls muted.  And they complain.

Of that we agree, I've heard those complains too :-)

>> I'm a little surprised by the implementation though; partly because the
>> functions added to hda_codec.c does not seem to be HDA specific, partly
>> by the need to mess around with set_fs. I trust you to know what you're
>> doing w r t to the set_fs stuff, but maybe it would be more elegant if
>> these functions could be in the core and either using, or together with,
>> other functions that need to do set_fs to read/write TLV information?
>
> The volume initialization is a very sensitive topic.  It's really
> depending on the hardware which value to be set.  In the case of
> vmaster, it's relatively easy.  At least, the slave output volumes can
> be set to 0dB just to follow the master volume.
>
> Then I thought of implementing it in the common vmaster code, but
> found that it's not so trivial.  The vmaster slaves may contain also
> input volumes like line-in.  Raising such a control automatically is
> obviously wrong.  Thus the selection of slaves must be selective.
> Also the handling of dB TLV would be more complicated when we allow
> all dB TLV types, to be more generic.  That's why I decided to start
> from the HD-audio specific code.
>
> About set_fs: yes, it's ugly and I want to get rid of it, too.
> Currently it's needed because the TLV callback is supposed to handle
> the user-space pointer directly.  Handling the user-space pointer
> allows us to access to a large data without too much copying.
>
> OTOH, it's true that the direct access would be easier for the
> lowlevel driver no matter with our without set_fs.  Since the TLV data
> size is limited (practically in 4kB or such), we may pass it to
> kmalloc'ed buffer and let the core copying to user-space.  It's not so
> frequently accessed type, anyway.
>
> So, both your points are correct.  They are to be improved in future.

I think at least get_kctl_0dB_offset should move to control.c at this 
step (and renamed to start with snd_ ), just because it belongs there 
and seems useful enough to other drivers.

In control.c you can check if tlv callback has been overridden and if 
not, use struct user_element directly, to avoid set_fs. Btw, not many 
drivers override tlv.c anyway and those who do just do it to provide dB 
min/max. Seems easy enough to change the tlv.c to return dB min/max 
instead of copying data to userspace.

Moving put_kctl_with_value would be preferable, too, but not as 
important IMO.

>> Another thought is whether you need to do snd_ctl_notify or if that is
>> handled automatically inside kctl->put. Or if you're just counting on
>> nobody to listen at that point.
>
> It's done in the build_controls, so it's far before the device
> registration.  Even if it's called in reconfigure, it's guaranteed
> that the device is free of access.

Ok.


-- 
David Henningsson, Canonical Ltd.
http://launchpad.net/~diwic