Re: [PATCH v2 05/46] rmap: hugetlb: switch from page_dup_file_rmap to page_add_file_rmap

From: James Houghton
Date: Thu Mar 02 2023 - 11:44:29 EST

On Thu, Mar 2, 2023 at 7:44 AM James Houghton <jthoughton@xxxxxxxxxx> wrote:
>
> On Wed, Mar 1, 2023 at 5:06 PM Jiaqi Yan <jiaqiyan@xxxxxxxxxx> wrote:
> >
> > On Fri, Feb 17, 2023 at 4:28 PM James Houghton <jthoughton@xxxxxxxxxx> wrote:
> > >
> > > This only applies to file-backed HugeTLB, and it should be a no-op until
> > > high-granularity mapping is possible. Also update page_remove_rmap to
> > > support the eventual case where !compound && folio_test_hugetlb().
> > >
> > > HugeTLB doesn't use LRU or mlock, so we avoid those bits. This also
> > > means we don't need to use subpage_mapcount; if we did, it would
> > > overflow with only a few mappings.
>
> This is wrong; I guess I misunderstood the code when I wrote this
> commit. subpages_mapcount (now called _nr_pages_mapped) won't overflow
> (unless HugeTLB pages could be greater than 16G). It is indeed a bug
> not to update _nr_pages_mapped the same way THPs do.
>
> >
> > >
> > > There is still one caller of page_dup_file_rmap left: copy_present_pte,
> > > and it is always called with compound=false in this case.
> > >
> > > Signed-off-by: James Houghton <jthoughton@xxxxxxxxxx>
> > >
> > > diff --git a/mm/hugetlb.c b/mm/hugetlb.c
> > > index 08004371cfed..6c008c9de80e 100644
> > > --- a/mm/hugetlb.c
> > > +++ b/mm/hugetlb.c
> > > @@ -5077,7 +5077,7 @@ int copy_hugetlb_page_range(struct mm_struct *dst, struct mm_struct *src,
> > > * sleep during the process.
> > > */
> > > if (!PageAnon(ptepage)) {
> > > - page_dup_file_rmap(ptepage, true);
> > > + page_add_file_rmap(ptepage, src_vma, true);
> > > } else if (page_try_dup_anon_rmap(ptepage, true,
> > > src_vma)) {
> > > pte_t src_pte_old = entry;
> > > @@ -5910,7 +5910,7 @@ static vm_fault_t hugetlb_no_page(struct mm_struct *mm,
> > > if (anon_rmap)
> > > hugepage_add_new_anon_rmap(folio, vma, haddr);
> > > else
> > > - page_dup_file_rmap(&folio->page, true);
> > > + page_add_file_rmap(&folio->page, vma, true);
> > > new_pte = make_huge_pte(vma, &folio->page, ((vma->vm_flags & VM_WRITE)
> > > && (vma->vm_flags & VM_SHARED)));
> > > /*
> > > @@ -6301,7 +6301,7 @@ int hugetlb_mcopy_atomic_pte(struct mm_struct *dst_mm,
> > > goto out_release_unlock;
> > >
> > > if (folio_in_pagecache)
> > > - page_dup_file_rmap(&folio->page, true);
> > > + page_add_file_rmap(&folio->page, dst_vma, true);
> > > else
> > > hugepage_add_new_anon_rmap(folio, dst_vma, dst_addr);
> > >
> > > diff --git a/mm/migrate.c b/mm/migrate.c
> > > index d3964c414010..b0f87f19b536 100644
> > > --- a/mm/migrate.c
> > > +++ b/mm/migrate.c
> > > @@ -254,7 +254,7 @@ static bool remove_migration_pte(struct folio *folio,
> > > hugepage_add_anon_rmap(new, vma, pvmw.address,
> > > rmap_flags);
> > > else
> > > - page_dup_file_rmap(new, true);
> > > + page_add_file_rmap(new, vma, true);
> > > set_huge_pte_at(vma->vm_mm, pvmw.address, pvmw.pte, pte);
> > > } else
> > > #endif
> > > diff --git a/mm/rmap.c b/mm/rmap.c
> > > index 15ae24585fc4..c010d0af3a82 100644
> > > --- a/mm/rmap.c
> > > +++ b/mm/rmap.c
> >
> > Given you are making hugetlb's ref/mapcount mechanism to be consistent
> > with THP, I think the special folio_test_hugetlb checks you added in
> > this commit will break page_mapped() and folio_mapped() if page/folio
> > is HGMed. With these checks, folio->_nr_pages_mapped are not properly
> > increased/decreased.
>
> Thank you, Jiaqi! I didn't realize I broke
> folio_mapped()/page_mapped(). The end result is that page_mapped() may
> report that an HGMed page isn't mapped when it is. Not good!
>
> >
> > > @@ -1318,21 +1318,21 @@ void page_add_file_rmap(struct page *page, struct vm_area_struct *vma,
> > > int nr = 0, nr_pmdmapped = 0;
> > > bool first;
> > >
> > > - VM_BUG_ON_PAGE(compound && !PageTransHuge(page), page);
> > > + VM_BUG_ON_PAGE(compound && !PageTransHuge(page)
> > > + && !folio_test_hugetlb(folio), page);
> > >
> > > /* Is page being mapped by PTE? Is this its first map to be added? */
> > > if (likely(!compound)) {
> > > first = atomic_inc_and_test(&page->_mapcount);
> > > nr = first;
> > > - if (first && folio_test_large(folio)) {
> > > + if (first && folio_test_large(folio)
> > > + && !folio_test_hugetlb(folio)) {
> >
> > So we should still increment _nr_pages_mapped for hugetlb case here,
> > and decrement in the corresponding place in page_remove_rmap.
> >
> > > nr = atomic_inc_return_relaxed(mapped);
> > > nr = (nr < COMPOUND_MAPPED);
> > > }
> > > - } else if (folio_test_pmd_mappable(folio)) {
> > > - /* That test is redundant: it's for safety or to optimize out */
> > > -
> > > + } else {
> > > first = atomic_inc_and_test(&folio->_entire_mapcount);
> > > - if (first) {
> > > + if (first && !folio_test_hugetlb(folio)) {
> >
> > Same here: we should still increase _nr_pages_mapped by
> > COMPOUND_MAPPED and decrease by COMPOUND_MAPPED in the corresponding
> > place in page_remove_rmap.
> >
> > > nr = atomic_add_return_relaxed(COMPOUND_MAPPED, mapped);
> > > if (likely(nr < COMPOUND_MAPPED + COMPOUND_MAPPED)) {
> > > nr_pmdmapped = folio_nr_pages(folio);
> > > @@ -1347,6 +1347,9 @@ void page_add_file_rmap(struct page *page, struct vm_area_struct *vma,
> > > }
> > > }
> > >
> > > + if (folio_test_hugetlb(folio))
> > > + return;
> > > +
> > > if (nr_pmdmapped)
> > > __lruvec_stat_mod_folio(folio, folio_test_swapbacked(folio) ?
> > > NR_SHMEM_PMDMAPPED : NR_FILE_PMDMAPPED, nr_pmdmapped);
> > > @@ -1376,8 +1379,7 @@ void page_remove_rmap(struct page *page, struct vm_area_struct *vma,
> > > VM_BUG_ON_PAGE(compound && !PageHead(page), page);
> > >
> > > /* Hugetlb pages are not counted in NR_*MAPPED */
> > > - if (unlikely(folio_test_hugetlb(folio))) {
> > > - /* hugetlb pages are always mapped with pmds */
> > > + if (unlikely(folio_test_hugetlb(folio)) && compound) {
> > > atomic_dec(&folio->_entire_mapcount);
> > > return;
> > > }
> >
> > This entire if-block should be removed after you remove the
> > !folio_test_hugetlb checks in page_add_file_rmap.
>
> This is the not-so-obvious change that is needed. Thank you!
>
> >
> > > @@ -1386,15 +1388,14 @@ void page_remove_rmap(struct page *page, struct vm_area_struct *vma,
> > > if (likely(!compound)) {
> > > last = atomic_add_negative(-1, &page->_mapcount);
> > > nr = last;
> > > - if (last && folio_test_large(folio)) {
> > > + if (last && folio_test_large(folio)
> > > + && !folio_test_hugetlb(folio)) {
> >
> > ditto.
> >
> > > nr = atomic_dec_return_relaxed(mapped);
> > > nr = (nr < COMPOUND_MAPPED);
> > > }
> > > - } else if (folio_test_pmd_mappable(folio)) {
> > > - /* That test is redundant: it's for safety or to optimize out */
> > > -
> > > + } else {
> > > last = atomic_add_negative(-1, &folio->_entire_mapcount);
> > > - if (last) {
> > > + if (last && !folio_test_hugetlb(folio)) {
> >
> > ditto.
>
> I agree with all of your suggestions. Testing with the hugetlb-hgm
> selftest, nothing seems to break. :)
>
> Given that this is at least the third or fourth major bug in this
> version of the series, I'll go ahead and send a v3 sooner rather than
> later.

This solution limits the size of a HugeTLB page to 16G. I'm not sure
if there are any architectures that support HugeTLB pages larger than
16G (it looks like powerpc supports 16G). If they do, I think we can
just increase the value of COMPOUND_MAPPED. If that's not possible, we
would need another solution than participating in _nr_pages_mapped
like THPs.