On Sun, Oct 30, 2022 at 06:47:23PM -0700, Linus Torvalds wrote: > Also, it's worth noting that zap_pte_range() does this sanity test: > > if (unlikely(page_mapcount(page) < 0)) > print_bad_pte(vma, addr, ptent, page); > > and that is likely worthless now (because it hasn't actually > decremented the mapcount yet). I didn't remove it, because I wasn't > sure which option was best: > > (a) just remove it entirely > > (b) change the "< 0" to "<= 0" > > (c) move it to clean_and_free_pages_and_swap_cache() that actually > does the page_zap_pte_rmap() now. I'm leaning towards (c); simply because the error case is so terrifying I feel we should check for it (and I do have vague memories of us actually hitting something like this in the very distant past).
