usb_audio: refill the mic FIFO from tud_audio_tx_done_isr()

[mikeysklar]

Jumped in on a usb_audio PR# 11102 that @relic-se @FoamyGuy had been working on. This is to add ESP32-S2 and S3 usb_audio support. It already worked on
RP2040, RP2350 and nRF52840. On ESP32 the device enumerated as a correct UAC2
microphone, the host opened the stream, and then nothing came out.

I put a Cynthion on the bus to see what the host was actually doing. It was
behaving perfectly:

	before	after
stream held open	578 ms, host gave up	4.01 s, normal close
IN tokens on the audio endpoint	573	4007
full size packets	8	4006
zero length packets	564	0
CRC16 errors	0	0
arecord -d 4	EIO, 44 bytes	full length capture

The host polled every 1 ms for 578 ms and only then gave up. Nothing was corrupt.
The device sent 8 real frames and then zero length packets for the remaining 565.

TinyUSB handles audio completions in the ISR, so they never wake tud_task(). On
espressif the class tasks run after tud_task(), so they never run either and the
FIFO is filled once. tud_audio_tx_done_isr() is TinyUSB's hook for exactly this,
and refilling from there needs no port changes.

Also here: CIRCUITPY_USB_AUDIO ?= 1 on esp32s3, which claims no endpoints until
enable() is called. esp32s2 stays off by default.

S2 and S3 allow five IN endpoints including EP0 and the default build uses all
five, so one has to be freed:

# boot.py
import usb_hid, usb_audio
usb_hid.disable()
usb_audio.enable(sample_rate=48000, channel_count=1, microphone=True, speaker=False)

# code.py, 1 kHz tone
import array, math, audiocore, usb_audio
N = 480
buf = array.array("h", [int(18000 * math.sin(2 * math.pi * 10 * i / N)) for i in range(N)])
sample = audiocore.RawSample(buf, channel_count=1, sample_rate=48000)
usb_audio.usb_microphone.play(sample, loop=True)

Tested on Metro ESP32-S3 and Metro ESP32-S2 against this branch.

arecord -f S16_LE -r 48000 -c 1 -d 4

writes 384044 bytes on both.

[tannewt]

Thank you! Not sure why some board builds are failing though.

[mikeysklar]

Six of them use a CH340C or CP2102 serial chip instead of native USB, so
CIRCUITPY_USB_DEVICE = 0 and usb_audio had no TinyUSB device stack to build
against. Fixed in 1fb4196 by clearing CIRCUITPY_USB_AUDIO when its
dependencies are off.

[tannewt]

This is going to make it really confusing to have one place seemingly set it and the global force it off. Instead, make the port-level setting conditional on the other USB settings. Generally more specific settings should override general ones global > port > board.

[tannewt]

Thank you!

[mikeysklar]

Moved it into the port setting, so py/circuitpy_mpconfig.mk is untouched now. The esp32s3 default only applies when the board has not turned off the TinyUSB device stack or audiocore.

Checked with print-CIRCUITPY_USB_AUDIO: the six CIRCUITPY_USB_DEVICE = 0 boards and adafruit_esp32s3_camera all stay off, metro and feather S3 stay on. heltec_wireless_paper and adafruit_esp32s3_camera build clean with no usb_audio objects, metro S3 keeps them.