Remarks on Kato's square-root problem