Instead of using break, this one bubbles down into the kernel
	directly.  This code works, but isn't performance optimal yet.