ARM: mvebu: add support to clear shared L2 bit on Armada XP

For optimal performance, in a HW I/O coherency context such as the one
used on Armada XP, the shared L2 bit of the CPU configuration register
should be cleared.

This commit adjusts the coherency fabric code used by Marvell EBU
processors to clear this bit on Armada XP. Since it's a per-CPU
register, it's cleared in set_cpu_coherent() for the boot CPU, and
through a CPU notifier for the non-boot CPUs.

[gregory.clement@free-electrons.com: rebasd on 4.3-rc1]

Signed-off-by: Thomas Petazzoni <thomas.petazzoni@free-electrons.com>
Signed-off-by: Gregory CLEMENT <gregory.clement@free-electrons.com>