This bug has been fixed by moonman in kernel 4.10.12-2
Just saying. I have posted this bug in two other sub forums but apparently there is not enough people using large disks with ARM5 to care about this.
Ext4 of the current kernel fails to read beyond 8TB. As a result, fsck at boot fails and large RAID does not mount.
EDIT:
I began rolling back rootfs to check kernel, mdadm and e2fsprogs starting September 2016.
EDIT 04/02/17
Something got broken when upgrading kernel from 4.8.1-1 to 4.8.4-1
linux-kirkwood-dt-4.8.1-1 does not have the issue.
linux-kirkwood-dt-4.8.4-1 causes block address overflow
the bug remains now in 4.10.6