Symptom for this is no fs0 while dblk seemingly works. "Connect -r" on a DEBUG build will complain about partition end being beyond LastBlock. The LastBlock value is bogus and incorrect for media, meaning that the ECSD data (CMD8) is coming back incorrectly, and it is an SdHost bug (does not affect when Arasan is used).

It doesn't affect CM3 due to sheer luck (the bogus value is large enough), but seen with other eMMC media on Pi 2 / Pi 3 via SD slot.

I've tried ensuring the data FIFO is empty before starting a command. It is. I also don't know how much data is actually available in the FIFO once CMD8 completes, so that's worth checking, although it's a little strange overall. Normal read/write I/O is fine.