[PATCH rebase/RFC 0/4] x86/kvm/nVMX: optimize MMU switch between L1 and L2

[Paolo Bonzini]

This is a rebased version of Vitaly's RFC series.  This isn't
quite ready for application as I haven't yet thought through
the interactions between the root_mmu/guest_mmu split and
the multi-root caching.

Speaking of the multi-root caching, it is a bit of a duplicate work
with Vitaly's last three patches that avoided reinitialization if
the parameters and CR3 matched, so the series got smaller too.

The scalability improvement on the nested CPUID benchamrk is sensible:

                  -smp 1      -smp 16      -smp 48
  4.18            18k         86k
  kvm/queue       18k         68k  
  +gup_fast       18k         42k  
  +guest_mmu      18k         26k          42k

The slowdown between -smp 16 and -smp 48 is partly due to SMT,
because the machine has 28 cores.

Paolo

Vitaly Kuznetsov (4):
  x86/kvm/mmu: make vcpu->mmu a pointer to the current MMU
  x86/kvm/mmu.c: set get_pdptr hook in kvm_init_shadow_ept_mmu()
  x86/kvm/mmu.c: add kvm_mmu parameter to kvm_mmu_free_roots()
  x86/kvm/mmu: introduce guest_mmu

 arch/x86/include/asm/kvm_host.h |  11 ++-
 arch/x86/kvm/mmu.c              | 183 +++++++++++++++++++++-------------------
 arch/x86/kvm/mmu.h              |   8 +-
 arch/x86/kvm/mmu_audit.c        |  12 +--
 arch/x86/kvm/paging_tmpl.h      |  15 ++--
 arch/x86/kvm/svm.c              |  14 +--
 arch/x86/kvm/vmx.c              |  43 ++++++----
 arch/x86/kvm/x86.c              |  20 ++---
 8 files changed, 168 insertions(+), 138 deletions(-)

-- 
1.8.3.1