[v2,0/2] KVM: VMX: Clean up IRQ/NMI handling

Message

Sean Christopherson Sept. 15, 2020, 7:15 p.m. UTC

Clean up KVM's handling of IRQ and NMI exits to move the invocation of the
IRQ handler to a standalone assembly routine, and to then consolidate the
NMI handling to use the same indirect call approach instead of using INTn.

The IRQ cleanup was suggested by Josh Poimboeuf in the context of a false
postive objtool warning[*].  I believe Josh intended to use UNWIND hints
instead of trickery to avoid objtool complaints.  I opted for trickery in
the form of a redundant, but explicit, restoration of RSP after the hidden
IRET.  AFAICT, there are no existing UNWIND hints that would let objtool
know that the stack is magically being restored, and adding a new hint to
save a single MOV <reg>, <reg> instruction seemed like overkill.

The NMI consolidation was loosely suggested by Andi Kleen.  Andi's actual
suggestion was to export and directly call the NMI handler, but that's a
more involved change (unless I'm misunderstanding the wants of the NMI
handler), whereas piggybacking the IRQ code is simple and seems like a
worthwhile intermediate step.

Sean Christopherson (2):
  KVM: VMX: Move IRQ invocation to assembly subroutine
  KVM: VMX: Invoke NMI handler via indirect call instead of INTn

 arch/x86/kvm/vmx/vmenter.S | 34 +++++++++++++++++++++
 arch/x86/kvm/vmx/vmx.c     | 61 +++++++++++---------------------------
 2 files changed, 51 insertions(+), 44 deletions(-)

Comments

Paolo Bonzini Sept. 22, 2020, 1:38 p.m. UTC | #1

On 15/09/20 21:15, Sean Christopherson wrote:
> Clean up KVM's handling of IRQ and NMI exits to move the invocation of the
> IRQ handler to a standalone assembly routine, and to then consolidate the
> NMI handling to use the same indirect call approach instead of using INTn.
> 
> The IRQ cleanup was suggested by Josh Poimboeuf in the context of a false
> postive objtool warning[*].  I believe Josh intended to use UNWIND hints
> instead of trickery to avoid objtool complaints.  I opted for trickery in
> the form of a redundant, but explicit, restoration of RSP after the hidden
> IRET.  AFAICT, there are no existing UNWIND hints that would let objtool
> know that the stack is magically being restored, and adding a new hint to
> save a single MOV <reg>, <reg> instruction seemed like overkill.
> 
> The NMI consolidation was loosely suggested by Andi Kleen.  Andi's actual
> suggestion was to export and directly call the NMI handler, but that's a
> more involved change (unless I'm misunderstanding the wants of the NMI
> handler), whereas piggybacking the IRQ code is simple and seems like a
> worthwhile intermediate step.
> 
> Sean Christopherson (2):
>   KVM: VMX: Move IRQ invocation to assembly subroutine
>   KVM: VMX: Invoke NMI handler via indirect call instead of INTn
> 
>  arch/x86/kvm/vmx/vmenter.S | 34 +++++++++++++++++++++
>  arch/x86/kvm/vmx/vmx.c     | 61 +++++++++++---------------------------
>  2 files changed, 51 insertions(+), 44 deletions(-)
> 

Queued, thanks.

Paolo