Thanks for all the discussions so far. A lot of it is over my head, but I appreciate it!
Even with this patch, there is a remaining problem, which is not present > in the MS2109. The sound sample values range from 0x0000 to 0x7fff,
with> silence around 0x4000, i.e. 15-bit-ish audio. This actually sounds OK to> the ear (although half as loud as it should be), but looks odd when> looking at the waveform, and makes volume meters always think the sound> is very loud. On further inspection, the last hex digit of the samples are always 0 or 8. So the least significant three bits are always zero, and technically the values range from 0x0000 to 0x7ff8, not 0x7fff as I said before.
To convert to s16le, I can bitshift one bit left, and subtract 32768. I'm told that this isn't something that can or should be done in the kernel, but should be in userspace. Any more advice on how to fix this remaining quirk would be very welcome.
I don't know if this makes things more complicated. It is still the case that the conversion described above to s16le makes the waveforms look correct (and the volume correct), even if the lowest four bits are zero.
John Veness