[alsa-devel] ASoC: Hiding unused CODEC mixer widgets in the machine driver
Hi all,
Work has continued on the TLV320AIC3204 CODEC driver, and there have been a number of fixes since the last patch was published. However, this CODEC is a rather complicated beast to drive, some may find the controls a little unweildy, and as such, I was wondering if the machine driver could be used to abstract this for a given target?
It's no use abstracting this in the CODEC, because some may want to access the functionality I'm trying to hide.
The CODEC we're using has three line inputs; which may be treated as either three mono differential inputs, or three stereo single-ended inputs. There are also a couple of common-mode sources that may be connected. The resistances connecting these is selectable; 10k, 20k or 40k... and as such, it's possible to do some hardware mixing of the sources inside the CODEC.
At the moment I have each of the inputs to the ADC's PGA ("MICPGA"), with one control for each possible path... There are two ADCs, so two PGAs... and the inputs can be connected to either the negative or positive terminal of the PGA in many cases. That leads to a large number of controls (the following is from amixer):
Simple mixer control 'Left MICPGA+ Output Mixer IN1L',0 Capabilities: enum Items: 'not-connected' '10k' '20k' '40k' Item0: 'not-connected' Simple mixer control 'Left MICPGA+ Output Mixer IN1R',0 Capabilities: enum Items: 'not-connected' '10k' '20k' '40k' Item0: 'not-connected' Simple mixer control 'Left MICPGA+ Output Mixer IN2L',0 Capabilities: enum Items: 'not-connected' '10k' '20k' '40k' Item0: 'not-connected' Simple mixer control 'Left MICPGA+ Output Mixer IN3L',0 Capabilities: enum Items: 'not-connected' '10k' '20k' '40k' Item0: '10k' Simple mixer control 'Left MICPGA- Output Mixer CM1L',0 Capabilities: enum Items: 'not-connected' '10k' '20k' '40k' Item0: 'not-connected' Simple mixer control 'Left MICPGA- Output Mixer CM2L',0 Capabilities: enum Items: 'not-connected' '10k' '20k' '40k' Item0: 'not-connected' Simple mixer control 'Left MICPGA- Output Mixer IN2R',0 Capabilities: enum Items: 'not-connected' '10k' '20k' '40k' Item0: 'not-connected' Simple mixer control 'Left MICPGA- Output Mixer IN3R',0 Capabilities: enum Items: 'not-connected' '10k' '20k' '40k' Item0: '10k' [...] Simple mixer control 'Right MICPGA+ Output Mixer IN1R',0 Capabilities: enum Items: 'not-connected' '10k' '20k' '40k' Item0: '10k' Simple mixer control 'Right MICPGA+ Output Mixer IN2L',0 Capabilities: enum Items: 'not-connected' '10k' '20k' '40k' Item0: 'not-connected' Simple mixer control 'Right MICPGA+ Output Mixer IN2R',0 Capabilities: enum Items: 'not-connected' '10k' '20k' '40k' Item0: 'not-connected' Simple mixer control 'Right MICPGA+ Output Mixer IN3R',0 Capabilities: enum Items: 'not-connected' '10k' '20k' '40k' Item0: '10k' Simple mixer control 'Right MICPGA- Output Mixer CM1R',0 Capabilities: enum Items: 'not-connected' '10k' '20k' '40k' Item0: 'not-connected' Simple mixer control 'Right MICPGA- Output Mixer CM2R',0 Capabilities: enum Items: 'not-connected' '10k' '20k' '40k' Item0: 'not-connected' Simple mixer control 'Right MICPGA- Output Mixer IN1L',0 Capabilities: enum Items: 'not-connected' '10k' '20k' '40k' Item0: '10k' Simple mixer control 'Right MICPGA- Output Mixer IN3L',0 Capabilities: enum Items: 'not-connected' '10k' '20k' '40k' Item0: '10k'
These controls are useful to a machine driver for implementing controls, but they're not that friendly to an end user. I was wondering if it was possible to hide these controls from the userland, but expose them to the machine driver to be presented to the user in a more meaningful manner. (i.e. we can have just simple mute switches for IN1 and IN3, labelling them "handset" and "internal mic" in our device).
Or would I be better defining these controls in the machine driver?
Regards,
On Fri, Aug 06, 2010 at 10:18:45AM +1000, Stuart Longland wrote:
Work has continued on the TLV320AIC3204 CODEC driver, and there have been a number of fixes since the last patch was published. However, this CODEC is a rather complicated beast to drive, some may find the controls a little unweildy, and as such, I was wondering if the machine driver could be used to abstract this for a given target?
The machine driver shouldn't be doing this - it's too fragile in the face of any changes in the drivers and means hard coding use cases into the kernel which is really bad practice since it's much slower than doing things from user space and much harder to change as new needs arise.
Much of this should be being handled by use case management anyway, even the controls that actually exist usefully on a given system are normally way more detail than end applications should ever be seeing. See
http://www.slimlogic.co.uk/?p=40
for the API which should deal with that.
We should still do the limiting in kernel for controls that are on paths that can not connected be at all (since it makes life easier developing use cases) but the major thing is the use case management. Takashi has a patch for hiding controls which could be built on to do this but unfortunately it's been sitting in his sound-unstable tree rather than included in mainline.
On Fri, 2010-08-06 at 12:19 +0100, Mark Brown wrote:
On Fri, Aug 06, 2010 at 10:18:45AM +1000, Stuart Longland wrote:
Work has continued on the TLV320AIC3204 CODEC driver, and there have been a number of fixes since the last patch was published. However, this CODEC is a rather complicated beast to drive, some may find the controls a little unweildy, and as such, I was wondering if the machine driver could be used to abstract this for a given target?
The machine driver shouldn't be doing this - it's too fragile in the face of any changes in the drivers and means hard coding use cases into the kernel which is really bad practice since it's much slower than doing things from user space and much harder to change as new needs arise.
Much of this should be being handled by use case management anyway, even the controls that actually exist usefully on a given system are normally way more detail than end applications should ever be seeing. See
http://www.slimlogic.co.uk/?p=40
for the API which should deal with that.
Please do give UCM (Use Case Manager) a test drive. We are in the late development / early testing phase before upstreaming (still one or two things todo).
You will need to checkout the use-case-verb branches of :-
git://git.slimlogic.co.uk/alsa-lib.git
git://git.slimlogic.co.uk/alsa-utils.git
There is a tool called alsaucm that allows changing the use case from the command line. Best to use it's interactive (-i) mode atm.
There is also an initial repository for UCM use case configuration files here :-
git://git.slimlogic.co.uk/alsa-ucm-conf.git
This will show the format required to describe use cases.
Thanks
Liam
On Fri, 2010-08-06 at 12:34 +0100, Liam Girdwood wrote:
On Fri, 2010-08-06 at 12:19 +0100, Mark Brown wrote:
On Fri, Aug 06, 2010 at 10:18:45AM +1000, Stuart Longland wrote:
Work has continued on the TLV320AIC3204 CODEC driver, and there have been a number of fixes since the last patch was published. However, this CODEC is a rather complicated beast to drive, some may find the controls a little unweildy, and as such, I was wondering if the machine driver could be used to abstract this for a given target?
The machine driver shouldn't be doing this - it's too fragile in the face of any changes in the drivers and means hard coding use cases into the kernel which is really bad practice since it's much slower than doing things from user space and much harder to change as new needs arise.
Much of this should be being handled by use case management anyway, even the controls that actually exist usefully on a given system are normally way more detail than end applications should ever be seeing. See
http://www.slimlogic.co.uk/?p=40
for the API which should deal with that.
Please do give UCM (Use Case Manager) a test drive. We are in the late development / early testing phase before upstreaming (still one or two things todo).
You will need to checkout the use-case-verb branches of :-
git://git.slimlogic.co.uk/alsa-lib.git
git://git.slimlogic.co.uk/alsa-utils.git
There is a tool called alsaucm that allows changing the use case from the command line. Best to use it's interactive (-i) mode atm.
There is also an initial repository for UCM use case configuration files here :-
git://git.slimlogic.co.uk/alsa-ucm-conf.git
This will show the format required to describe use cases.
Forgot to add that patches are most welcome.
Thanks
Liam
Hi Mark, Liam... On 08/06/10 21:42, Liam Girdwood wrote:
On Fri, 2010-08-06 at 12:34 +0100, Liam Girdwood wrote:
On Fri, 2010-08-06 at 12:19 +0100, Mark Brown wrote:
The machine driver shouldn't be doing this - it's too fragile in the face of any changes in the drivers and means hard coding use cases into the kernel which is really bad practice since it's much slower than doing things from user space and much harder to change as new needs arise.
Please do give UCM (Use Case Manager) a test drive. We are in the late development / early testing phase before upstreaming (still one or two things todo).
You will need to checkout the use-case-verb branches of :-
git://git.slimlogic.co.uk/alsa-lib.git
git://git.slimlogic.co.uk/alsa-utils.git
There is a tool called alsaucm that allows changing the use case from the command line. Best to use it's interactive (-i) mode atm.
There is also an initial repository for UCM use case configuration files here :-
git://git.slimlogic.co.uk/alsa-ucm-conf.git
This will show the format required to describe use cases.
Forgot to add that patches are most welcome.
No worries... I can see there being possible issues with the machine driver being highly sensitive to changes to the CODEC driver, I was just curious if there was some mechanism for abstracting this.
However, a userspace solution will probably work well... I'll have a look and report back any findings. How close is this to being merged into the main alsa-lib and alsa-utils trees?
Regards,
On Sun, 2010-08-08 at 09:07 +1000, Stuart Longland wrote:
Hi Mark, Liam... On 08/06/10 21:42, Liam Girdwood wrote:
On Fri, 2010-08-06 at 12:34 +0100, Liam Girdwood wrote:
On Fri, 2010-08-06 at 12:19 +0100, Mark Brown wrote:
The machine driver shouldn't be doing this - it's too fragile in the face of any changes in the drivers and means hard coding use cases into the kernel which is really bad practice since it's much slower than doing things from user space and much harder to change as new needs arise.
Please do give UCM (Use Case Manager) a test drive. We are in the late development / early testing phase before upstreaming (still one or two things todo).
You will need to checkout the use-case-verb branches of :-
git://git.slimlogic.co.uk/alsa-lib.git
git://git.slimlogic.co.uk/alsa-utils.git
There is a tool called alsaucm that allows changing the use case from the command line. Best to use it's interactive (-i) mode atm.
There is also an initial repository for UCM use case configuration files here :-
git://git.slimlogic.co.uk/alsa-ucm-conf.git
This will show the format required to describe use cases.
Forgot to add that patches are most welcome.
No worries... I can see there being possible issues with the machine driver being highly sensitive to changes to the CODEC driver, I was just curious if there was some mechanism for abstracting this.
However, a userspace solution will probably work well... I'll have a look and report back any findings. How close is this to being merged into the main alsa-lib and alsa-utils trees?
We are aiming for sometime in the next two weeks. There are still roughly two things that need fixing:-
1) Drop the control ID from the use case configuration files. i.e.
In order to enable the 'DL1 Mixer Multimedia' path we have a config like
73:'DL1 Mixer Multimedia':1:1
this would just become
'DL1 Mixer Multimedia':1:1
This way we are not reliant on ID numbers that can easily change when driver updates occur (something that is catching me atm with a new driver). The control names are always very constant and unique though.
2) Add a small shared memory segment to store state information for multiple client access.
Liam
participants (3)
-
Liam Girdwood -
Mark Brown -
Stuart Longland