Description
CircuitPython version
8.2.0-rc.0Code/REPL
# Run any code using audiobusio
# and at the same time, perform a lengthy write to the CIRCUITPY driveBehavior
The audio mostly works fine, however, when the audio dma underflows due to not running background tasks frequently enough, unrelated memory is treated as audio data. This can sound like static, a chirp, a honk, etc.
This happens especially frequently when writing to the CIRCUITPY drive, because normal code in flash can't be executed during an erase/write cycle.
Description
No response
Additional information
The code in audio_dma.c has two DMA descriptors (call them A and B) which chain to each other, so without intervention, A will be DMA'd, then B, then, A, then B, and so on forever.
This happens whether or not the background tasks run.
Unlike on some systems, triggering a DMA transfer does not reset its address to a start address; instead, the address continues getting incremented. So if the buffer is at [say] 0xf000 .. 0xf200 then the first underflow will actually play audio data from 0xf200 to 0xf400, then from 0xf400 to 0xf600, and so on. This data will .. not sound good.
It would be nice if the audio stopped, or output the quiescent value, when DMA underflows.
This probably involves un-chaining the DMAs. I tried implementing this, but didn't quickly (1h) come up with something that worked better than the status quo, though.