[alsa-devel] Channel conversion in plughw not working for my driver... confused about snd_pcm_plug_hw_params()

[David Huggins-Daines]

Hi,

I'm trying to make mono output work on the audio driver for the Cirrus
EDB9302 evaluation board.  For some reason, plughw is refusing to do any
channel conversion.  When I try to play a mono file, I simply get:

aplay: set_params:1022: Unable to install hw params:

The driver's struct snd_pcm_hardware contains:

    .channels_min	= 2,
    .channels_max	= 2,

which I believe is the correct way to tell userspace "this driver can
only do stereo output" - at least, this is how intel8x0.c does it, and
that driver really ought to work by now!

I've traced this to the exact point of failure, and what I see is that
in snd_pcm_plug_hw_params(), despite figuring out exactly what the set
of hardware parameters will work for the slave device, it goes and calls
 _snd_pcm_hw_params() with the actual parameters requested by the
client, which obviously fails, and leads to the error shown above.

I added some calls to snd_pcm_hw_params_dump() which give this output.
The first is sparams, which is the refined set of parameters which I
assume are supposed to be set in the slave device.  The second is
params, which is the set of parameters requested by the client (aplay in
this case).

sparams are:
ACCESS:  RW_INTERLEAVED
FORMAT:  U8
SUBFORMAT:  STD
SAMPLE_BITS: 8
FRAME_BITS: 16
CHANNELS: 2
RATE: 11025
PERIOD_TIME: (124988 124989)
PERIOD_SIZE: 1378
PERIOD_BYTES: 2756
PERIODS: (4 5)
BUFFER_TIME: (500045 500046)
BUFFER_SIZE: 5513
BUFFER_BYTES: 11026
TICK_TIME: 0
params are:
ACCESS:  RW_INTERLEAVED
FORMAT:  U8
SUBFORMAT:  STD
SAMPLE_BITS: 8
FRAME_BITS: 8
CHANNELS: 1
RATE: 11025
PERIOD_TIME: (124988 124989)
PERIOD_SIZE: 1378
PERIOD_BYTES: 1378
PERIODS: (4 5)
BUFFER_TIME: (500045 500046)
BUFFER_SIZE: 5513
BUFFER_BYTES: 5513
TICK_TIME: 0
inserting plugins whatever that means
in snd_pcm_plug_change_channels!

So, yeah, that's great, it figured it all out, and realized that it had
to do channel conversion, which presumably is done in
snd_pcm_plug_change_channels().  But now, we see that it goes ahead and
tries to set these parameters in the slave device:

calling _snd_pcm_hw_params with params:
ACCESS:  RW_INTERLEAVED
FORMAT:  U8
SUBFORMAT:  STD
SAMPLE_BITS: 8
FRAME_BITS: 8
CHANNELS: 1
RATE: 11025
PERIOD_TIME: (124988 124989)
PERIOD_SIZE: 1378
PERIOD_BYTES: 1378
PERIODS: (4 5)
BUFFER_TIME: (500045 500046)
BUFFER_SIZE: 5513
BUFFER_BYTES: 5513
TICK_TIME: 0

But that's wrong!  The slave device can't do single-channel input!

So what gives?  This can't possibly be broken, or nobody's Intel
on-board audio would ever work at all... so what is my driver doing wrong?