I obviously got a bit hasty with the whole svm_leave_smm()
thing.  *sigh*
Except for the minor detail of being wrong :-)

And while I agree it's probably "cleaner", I think it's actually harder for
readers to understand, especially for readers that are familiar with nVMX.  E.g.
having a separate #VMEXIT path for consistency checks can actually be helpful,
because it highlights things that happen on #VMEXIT even when KVM hasn't started
loading L2 state.
My preference is to completely drop these:

  KVM: nSVM: Unify handling of VMRUN failures with proper cleanup
  KVM: nSVM: Refactor minimal #VMEXIT handling out of nested_svm_vmexit()
  KVM: nSVM: Call nested_svm_init_mmu_context() before switching to VMCB02
  KVM: nSVM: Call nested_svm_merge_msrpm() from enter_svm_guest_mode()
  KVM: nSVM: Call enter_guest_mode() before switching to VMCB02
And then moving these to the end of the series (or at least, beyond the stable@
patches):

  KVM: nSVM: Make nested_svm_merge_msrpm() return an errno
  KVM: nSVM: Drop nested_vmcb_check_{save/control}() wrappers
After paging more of this stuff back in (it's been a while since I looked at the
equivalent nVMX flow in depth), I'm quite opposed to aiming for a unified #VMEXIT
path for VMRUN.  Although it might seem otherwise at times, nVMX and nSVM didn't
end up with nested_vmx_load_cr3() buried toward the end of their flows purely to
make the code harder to read, there are real dependencies that need to be taken
into account.

And there's also value in having similar flows for nVMX and nSVM, e.g. where most
consistency checks occur before KVM starts loading L2 state.  VMX just happens to
architecturally _require_ that, whereas with SVM it was a naturally consequence
of writing the code.

Without the unification, a minimal #VMEXIT helper doesn't make any sense, and
I don't see any strong justification for shuffling around the order.[PATCH v6 16/31] KVM: nSVM: Unify handling of VMRUN failures with proper cleanupSean Christopherson <seanjc@google.com> undefinedYosry Ahmed <yosry@kernel.org> undefined undefined undefined undefined undefinedMq