Possible race condition in SpiDma transfers

[bugadani]

https://github.com/esp-rs/esp-hal/blob/main/esp-hal-common/src/spi/master.rs#L1340

The DMA sets the GDMA_OUT_EOF_CHn_INT_ENA bit after the peripheral starts the transmission. This may result in the future never resolving, if some other code is executed between transmission start and listen start.

Or at least I think this is the case.

[bjoernQ]

I think I've seen problem caused by this pattern with the AES peripheral which is faster (with small amounts of data) than we get to creating the future - so it's not just a hypothetical thing

[bugadani]

I think I'm hitting this issue or something related with esp-wifi and a spidma-driven display; the display's DmaTxFuture often stops resolving if my mobile is connected to the device's AP.