iommu/arm-smmu: Implement the .tlbi_domain op

The upcoming "fast" DMA mapper will need to take control of TLB
invalidation.  Doing so allows us to perform fewer TLB invalidation
operations since the DMA mapper layer has more knowledge about when
"stale" TLB entries might actually become a problem, so it can do TLB
invalidation much less frequently.  Implement the tlbi_domain op for
this purpose.

CRs-Fixed: 997751
Change-Id: Iba9f499dba89db91c1150947b9599d85ade65b0e
Signed-off-by: Mitchel Humpherys <mitchelh@codeaurora.org>