It's basically telling you that further iterations are necessary. You can implement that either by re-running from the same initial guess x0 that you used before, but with the MaxIter option set to a larger number of iterations. Or, you can have quadprog resume where it left off, by using the most current estimate of x that you have as the initial guess.
However, the fact that this is happening could well be because your problem is just highly ill-conditioned. Have you looked at cond(H) for your Hessian matrix H?
