Re: [PATCH v2 08/15] KVM: x86: Add mode-aware versions of kvm_<reg>_{read,write}() helpers

[Sean Christopherson <seanjc@google.com>]

On Mon, May 18, 2026, Kai Huang wrote:
> 
> > @@ -10413,29 +10413,30 @@ static int complete_hypercall_exit(struct kvm_vcpu *vcpu)
> >  
> >  	if (!is_64_bit_hypercall(vcpu))
> >  		ret = (u32)ret;
> > -	kvm_rax_write(vcpu, ret);
> > +	kvm_rax_write_raw(vcpu, ret);
> >  	return kvm_skip_emulated_instruction(vcpu);
> >  }
> > 
> 
> Nit:  AFAICT if we use kvm_rax_write(vcpu, ret) instead of the "raw" version
> here, we can then remove the
> 
> 	if (!is_64_bit_hypercall(vcpu))
> 		ret = (u32)ret;

No, because sneakily, is_64_bit_hypercall() != is_64_bit_mode(vcpu).  And because
we also need to avoid calling is_64_bit_mode().  If we use kvm_rax_write(), then
the unpacked code will be:

	WARN_ON_ONCE(vcpu->arch.guest_state_protected);

	if (is_long_mode(vcpu))
		kvm_x86_call(get_cs_db_l_bits)(vcpu, &cs_db, &cs_l);
	else
		cs_l = 0;

	if (cs_l)
		vcpu->arch.regs[VCPU_REGS_RAX] = ret;
	else	
		vcpu->arch.regs[VCPU_REGS_RAX] = (u32)ret;

whereas the (correct) behavior here is:

	if (vcpu->arch.guest_state_protected)
		cs_l = 1;
	else if (is_long_mode(vcpu))
		kvm_x86_call(get_cs_db_l_bits)(vcpu, &cs_db, &cs_l);
	else
		cs_l = 0;

	if (cs_l)
		vcpu->arch.regs[VCPU_REGS_RAX] = ret;
	else	
		vcpu->arch.regs[VCPU_REGS_RAX] = (u32)ret;

I.e. using the non-raw version will trigger the WARN_ON_ONCE(), and will incorrectly
truncate "ret" whenever cs_l is stale (which might be always?).