Dual UAC2_headsets

[mkarliner]

Starting with the uac2_headset example, I'm working on a composite device with two identical headset functions on a Raspberry Pi Pico.
I've made two descriptors, using the example as a base, different only in the FUNCTION number
eg:

 TUD_AUDIO_DESC_CLK_SRC(/*_clkid*/ UAC2_ENTITY1_CLOCK, /*_attr*/ 3, /*_ctrl*/ 7, /*_assocTerm*/ 0x00, /*_stridx*/ 0x00),                                                                                                                                                                                                                                                                                                                                                                                                        /* Input Terminal Descriptor(4.7.2.4) */                                     \
and
TUD_AUDIO_DESC_CLK_SRC(/*_clkid*/ UAC2_ENTITY2_CLOCK, /*_attr*/ 3, /*_ctrl*/ 7, /*_assocTerm*/ 0x00, /*_stridx*/ 0x00),                                                                                                                                                                                                                                                                                                                                                                                                        /* Input Terminal Descriptor(4.7.2.4) */       

etc, etc. \

With #define CFG_TUD_AUDIO 1
Either of the descriptors work as expected (at least with mic data) so I'm reasonably convinced that they are both OK individually.

However, with #define CFG_TUD_AUDIO 2 and both descriptors in the configuration
the first function still works, but I just get repeating noise on the second function, as below from Audacity
(the second function is the first recording)

this is a closer look at the (non-random) noise. It's been the same pattern through various compilations, which suggests that it does not represent uninitialised memory.

The write into the endpoint looks like this
tud_audio_n_write((0 or 1), (uint8_t *)mic_buf, (uint16_t)sample_count * 4);

The function and endpoint addresses look like this

{
  ITF_NUM_AUDIO1_CONTROL = 0,
  ITF_NUM_AUDIO1_STREAMING_SPK,
  ITF_NUM_AUDIO1_STREAMING_MIC,
  ITF_NUM_AUDIO2_CONTROL ,
  ITF_NUM_AUDIO2_STREAMING_SPK,
  ITF_NUM_AUDIO2_STREAMING_MIC,
  // ITF_NUM_CDC,
  // ITF_CDC_DATA,
  ITF_NUM_TOTAL
};


#define EPNUM_AUDIO1_IN 0x81
#define EPNUM_AUDIO1_OUT 0x02
#define EPNUM_AUDIO2_IN 0x87
#define EPNUM_AUDIO2_OUT 0x07

I suspect the issue may be around the assignment of Unit Numbers, but unfortunately I can't find any document that explains how to work this out. Currently, they look like this.

// Unit numbers are arbitrary selected
#define UAC2_ENTITY1_CLOCK 0x04
// Speaker path
#define UAC2_ENTITY_SPK1_INPUT_TERMINAL 0x01
#define UAC2_ENTITY_SPK1_FEATURE_UNIT 0x02
#define UAC2_ENTITY_SPK1_OUTPUT_TERMINAL 0x03
// Microphone path
#define UAC2_ENTITY_MIC1_INPUT_TERMINAL 0x01
#define UAC2_ENTITY_MIC1_OUTPUT_TERMINAL 0x03

#define UAC2_ENTITY2_CLOCK 0x05
// Speaker path
#define UAC2_ENTITY_SPK2_INPUT_TERMINAL 0x01
#define UAC2_ENTITY_SPK2_FEATURE_UNIT 0x06
#define UAC2_ENTITY_SPK2_OUTPUT_TERMINAL 0x03
// Microphone path
#define UAC2_ENTITY_MIC2_INPUT_TERMINAL 0x01
#define UAC2_ENTITY_MIC2_OUTPUT_TERMINAL 0x03

So it looks like using two uac_functions somehow messes up the endpoint fifo??

Digging down to _tu_fifo_write_n in tusb_fifo.c the log of the first function recording looks like this.

rd = 704, wr = 704, count = 0, remain = 1552, n = 176:
rd = 768, wr = 768, count = 0, remain = 388, n = 128:
rd = 880, wr = 880, count = 0, remain = 1552, n = 176:
rd = 120, wr = 120, count = 0, remain = 388, n = 128:
rd = 1056, wr = 1056, count = 0, remain = 1552, n = 176:

while the log of the second (bad) function looks like this.

rd = 896, wr = 896, count = 0, remain = 776, n = 128:
rd = 0, wr = 1056, count = 1056, remain = 496, n = 176:
rd = 1024, wr = 1024, count = 0, remain = 776, n = 128:
rd = 0, wr = 1232, count = 1232, remain = 320, n = 176:
rd = 1152, wr = 1152, count = 0, remain = 776, n = 128:

Which I take to mean that every other write to the fifo is pointing
to a start point of zero.

From an earlier discussion, I know that there is no problem with two dissimilar audio devices running together (thanks @HiFiPhile !), but after two days, I'm stumped.

I 'think' the information above might be enough to allow someone to give me a suggestion where to look, rather than ask someone to trawl through the descriptors, but they are attached anyway, apologies for the mess....

Thanks in advance...
Mike
files.zip