arm64: optimize el0_sync branch prediction

arm64: optimize el0_sync branch prediction

el0_sync manually sets the link register(lr) with ret_from_exception. This
is not recommended since it causes branch misprediction on RET. This patch
makes use of bl and adds an extra branch to tradeoff for the branch
misprediction.

Bug 1488764

Change-Id: I378754cb5fef46aacea42e3fe6c8db327e2a7a96
Signed-off-by: Rohit Khanna <rokhanna@nvidia.com>
Reviewed-on: http://git-master/r/410430
Reviewed-by: Alexander Van Brunt <avanbrunt@nvidia.com>
Tested-by: Alexander Van Brunt <avanbrunt@nvidia.com>