Re: [patch 3/3] KVM: task switch: check for segment base translation failure

[Avi Kivity <avi@qumranet.com>]

Marcelo Tosatti wrote:
> Subject says it all.
>
> Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
>
> Index: kvm-vmx-checks/arch/x86/kvm/x86.c
> ===================================================================
> --- kvm-vmx-checks.orig/arch/x86/kvm/x86.c
> +++ kvm-vmx-checks/arch/x86/kvm/x86.c
> @@ -3253,6 +3253,8 @@ static int load_guest_segment_descriptor
>  		return 1;
>  	}
>  	gpa = vcpu->arch.mmu.gva_to_gpa(vcpu, dtable.base);
> +	if (gpa == UNMAPPED_GVA)
> +		return 1;
>  	gpa += index * 8;
>  	return kvm_read_guest(vcpu->kvm, gpa, seg_desc, 8);
>  }
>   

This is wrong; if the descriptor table is long enough, the first page 
could be unmapped but the page(s) containing the segment could be mapped 
(and nothing guarantees the mapping is contiguous).

We need to translate dtable.base + index * 8.

What we really need is kvm_read_guest_virt() to take care of all of 
these things.  The emulator callbacks come fairly close.

-- 
error compiling committee.c: too many arguments to function