Re: [PATCH] KVM: x86: Don't update RIP or do single-step on faulting emulation

[Sean Christopherson]

On Tue, Aug 27, 2019 at 12:12:51PM -0700, Jim Mattson wrote:
> On Fri, Aug 23, 2019 at 1:55 PM Sean Christopherson
> <sean.j.christopherson@xxxxxxxxx> wrote:
> > --- a/arch/x86/kvm/x86.c
> > +++ b/arch/x86/kvm/x86.c
> > @@ -6611,12 +6611,13 @@ int x86_emulate_instruction(struct kvm_vcpu *vcpu,
> >                 unsigned long rflags = kvm_x86_ops->get_rflags(vcpu);
> >                 toggle_interruptibility(vcpu, ctxt->interruptibility);
> >                 vcpu->arch.emulate_regs_need_sync_to_vcpu = false;
> > -               kvm_rip_write(vcpu, ctxt->eip);
> > -               if (r == EMULATE_DONE && ctxt->tf)
> > -                       kvm_vcpu_do_singlestep(vcpu, &r);
> >                 if (!ctxt->have_exception ||
> > -                   exception_type(ctxt->exception.vector) == EXCPT_TRAP)
> > +                   exception_type(ctxt->exception.vector) == EXCPT_TRAP) {
> 
> NYC, but...
> 
> I don't think this check for "exception_type" is quite right.  A
> general detect fault (which can be synthesized by check_dr_read) is
> mischaracterized by exception_type() as a trap. Or maybe I'm missing
> something? (I often am.)

Pretty sure you're not missing anything.

And while we're poking holes in #DB emulation, int1/icebp isn't emulated
correctly as it should be reinjected with INTR_TYPE_PRIV_SW_EXCEPTION, not
as a INTR_TYPE_HARD_EXCEPTION.  The CPU automically clears DR7.GD on #DB,
unless the #DB is due to int1...