Virtually mapped kernel stacks

A doubtless stupid question from someone who never paid that much attention to this area of things: if the thread_info is below the stack and vulnerable to overruns, why not just move it to the top of the stack page and start the stack immediately below it? It's fixed-size, after all, so it should be easy to start the stack pointer right below it. The stack would then grow away from the thread_info, and overruns could not run into it (only underruns, which should be much rarer, one hopes).