[PATCH 0/2] KVM: arm64: nv: FGT traps of MDSCR_EL1

[Oliver Upton <oliver.upton@linux.dev>]

KVM now traps MDSCR_EL1 (and a hell of a lot of other things) because of
bad architecture redirecting the register to the VNCR. Marc reports that
this hurts (surprised?) meaning it is probably time to send in the FGTs.

As part of this the FGT configuration becomes contextual to the vCPU. FGT
computation is now hoisted up into vcpu_load() which avoids doing a
bunch of bit twiddling on every entry into the guest.

The last thing we can do to speed this up is add an early exit handleer,
but it's EOD for me over here and would rather push out what appears to
work right now.

Based directly on kvmarm/next given the fact that the dependency is
there also.

Oliver Upton (2):
  KVM: arm64: Compute per-vCPU FGTs at vcpu_load()
  KVM: arm64: nv: Use FGT write trap of MDSCR_EL1 when available

 arch/arm64/include/asm/kvm_host.h       |  50 ++++++++
 arch/arm64/kvm/arm.c                    |   1 +
 arch/arm64/kvm/config.c                 |  90 ++++++++++++++
 arch/arm64/kvm/hyp/include/hyp/switch.h | 148 +++---------------------
 arch/arm64/kvm/hyp/nvhe/pkvm.c          |   1 +
 arch/arm64/kvm/nested.c                 |   9 +-
 6 files changed, 165 insertions(+), 134 deletions(-)


base-commit: 181ce6b01ad52aeb791545edbae0b92648c6428d
-- 
2.47.3