The single-step debugging of KVM guests on x86 is broken: if we run
gdb 'stepi' command at the breakpoint when the guest interrupts are
enabled, RIP always jumps to native_apic_mem_write(). Then other
nasty effects follow.
Long investigation showed that on Jun 7, 2017 the
commit 
c8401dda2f0a00cd25c0 ("KVM: x86: fix singlestepping over syscall")
introduced the kvm_run.debug corruption: kvm_vcpu_do_singlestep() can
be called without X86_EFLAGS_TF set.
Let's fix it. Please consider that for -stable.
Signed-off-by: Alexander Popov <alex.popov@linux.com>
Cc: stable@vger.kernel.org
Fixes: c8401dda2f0a00cd25c0 ("KVM: x86: fix singlestepping over syscall")
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
 
                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 || (vcpu->guest_debug & KVM_GUESTDBG_SINGLESTEP)))
+               if (r == EMULATE_DONE && ctxt->tf)
                        kvm_vcpu_do_singlestep(vcpu, &r);
                if (!ctxt->have_exception ||
                    exception_type(ctxt->exception.vector) == EXCPT_TRAP)