KVM: arm64: Use the S2 MMU context to iterate over S2 table

Most of our S2 helpers take a kvm_s2_mmu pointer, but quickly
revert back to using the kvm structure. By doing so, we lose
track of which S2 MMU context we were initially using, and fallback
to the "canonical" context.

If we were trying to unmap a S2 context managed by a guest hypervisor,
we end-up parsing the wrong set of page tables, and bad stuff happens
(as this is often happening on the back of a trapped TLBI from the
guest hypervisor).

Instead, make sure we always use the provided MMU context all the way.
This has no impact on non-NV, as we always pass the canonical MMU
context.

Signed-off-by: Marc Zyngier <maz@kernel.org>
Reviewed-by: Andre Przywara <andre.przywara@arm.com>
Link: https://lore.kernel.org/r/20230209175820.1939006-3-maz@kernel.org


Signed-off-by: Oliver Upton <oliver.upton@linux.dev>