generalized symmetric-definite eigenvalue problem with a shift

Hello.

I have two matrices A and B and I want to solve the general eigenvalue Ax=λBx where B is positive definite and A is semi positive definite with the smallest eigenvalue of 0. So A is rank-deficient(ill-conditioned as a result); so I need a shift-inverse method. How can I implement using an mkl routine, either feast or lapack. 

Thanks.