[0/8] KVM: nSVM: ondemand nested state allocation + nested guest state caching

Message

Maxim Levitsky Aug. 20, 2020, 9:13 a.m. UTC

Hi!

This patch series implements caching of the whole nested guest vmcb
as opposed to current code that only caches its control area.
This allows us to avoid race in which guest changes the data area
while we are verifying it.

In adddition to that I also implemented on demand nested state area
to compensate a bit for memory usage increase from this caching.
This way at least guests that don't use nesting won't waste memory
on nested state.

Patches 1,2,3 are just refactoring,

Patches 4,5 are for ondemand nested state, while patches 6,7,8 are
for caching of the nested state.

Patch 8 is more of an optimization and can be dropped if you like to.

The series was tested with various nested guests, in one case even with
L3 running, but note that due to unrelated issue, migration with nested
guest running didn't work for me with or without this series.
I am investigating this currently.

Best regards,
	Maxim Levitsky

Maxim Levitsky (8):
  KVM: SVM: rename a variable in the svm_create_vcpu
  KVM: nSVM: rename nested 'vmcb' to vmcb_gpa in few places
  KVM: SVM: refactor msr permission bitmap allocation
  KVM: x86: allow kvm_x86_ops.set_efer to return a value
  KVM: nSVM: implement ondemand allocation of the nested state
  SVM: nSVM: cache whole nested vmcb instead of only its control area
  KVM: nSVM: implement caching of nested vmcb save area
  KVM: nSVM: read only changed fields of the nested guest data area

 arch/x86/include/asm/kvm_host.h |   2 +-
 arch/x86/kvm/svm/nested.c       | 296 +++++++++++++++++++++++---------
 arch/x86/kvm/svm/svm.c          | 129 +++++++-------
 arch/x86/kvm/svm/svm.h          |  32 ++--
 arch/x86/kvm/vmx/vmx.c          |   5 +-
 arch/x86/kvm/x86.c              |   3 +-
 6 files changed, 312 insertions(+), 155 deletions(-)