What's preventing lock_page_killable from replacing lock_page?
Is there more on the table than implementing the small amount of at each callsite of lock_page to handle the return condition -- ie, restrictions that prevent the kernel from being able to fail at successfully requiring and releasing a lock?