I just tested experimental, same issue there with
linux-image-5.0.0-trunk-marvell (5.0.2-1~exp1).
After digging around I assume this problem is related to changes in
drivers/net/ethernet/marvell/mvmdio.c in Version 4.12, which extended
the number of possible clock sources, see [1].
Just a couple of days ago there was a "fix" posted in [2], which checks
the array introduced above, before accessing it.
[1]
https://github.com/torvalds/linux/commit/96cb4342382290c935d933a08feb57d6d0183071#diff-9f818fed5204c3b34d954ad48e607789
[2] https://www.spinics.net/lists/netdev/msg584257.html
