Seminormality and upper semicontinuity in optimal control