On Thu, 3 Mar 2022, Matthew Wilcox wrote: > On Wed, Mar 02, 2022 at 05:35:30PM -0800, Hugh Dickins wrote: > > Although mmap_region() and mlock_fixup() take care that VM_LOCKED > > is never left set on a VM_SPECIAL vma, there is an interval while > > file->f_op->mmap() is using vm_insert_page(s), when VM_LOCKED may > > still be set while VM_SPECIAL bits are added: so mlock_vma_folio() > > should ignore VM_LOCKED while any VM_SPECIAL bits are set. > > > > This showed up as a "Bad page" still mlocked, when vfree()ing pages > > which had been vm_inserted by remap_vmalloc_range_partial(): while > > release_pages() and __page_cache_release(), and so put_page(), catch > > pages still mlocked when freeing (and clear_page_mlock() caught them > > when unmapping), the vfree() path is unprepared for them: fix it? > > but these pages should not have been mlocked in the first place. > > > > I assume that an mlockall(MCL_FUTURE) had been done in the past; or > > maybe the user got to specify MAP_LOCKED on a vmalloc'ing driver mmap. > > > > Signed-off-by: Hugh Dickins <hughd@xxxxxxxxxx> > > --- > > Diffed against top of next-20220301 or mmotm 2022-02-28-14-45. > > This patch really belongs as a fix to the mm/munlock series in > > Matthew's tree, so he might like to take it in there (but the patch > > here is the foliated version, so easiest to place it after foliation). > > It looks like it fixes "mm/munlock: mlock_pte_range() when mlocking or > munlocking", so I'll fold it into that patch? No and yes. That's great if you're prepared to move it back before the foliation. I think that just involves editing every "folio" to "page", including in the title - I very nearly sent it out with mlock_vma_page() in title. But I would prefer it to remain as a separate fix at the end of the mm/munlock series: this case is too unusual, and only a "Bad page", to mess with bisection prospects; and it's addressing an entirely different issue from what the "mlock_pte_range()..." is dealing with. Each of them needs its own explanation. So I would prefer it as a separate fix about "page"s, on top of the mm/munlock series, and you then adjust your foliation commit accordingly. Thank you: this is what I really wanted, but was afraid to ask of you (and of course, other fixes may turn out to be required, too late to adjust across the page<->folio barrier in this way: so it's nice to be able to do it this way, but rather beyond the call of duty). Hugh
