Stochastic linear quadratic optimal control for continuous-time systems based on policy iteration