Re: [PATCH RFC 1/2] mm: collect the number of anon large folios

Re: [PATCH RFC 1/2] mm: collect the number of anon large folios

[Barry Song]

> >> I would appreciate if we leave the rmap out here.
> >>
> >> Can't we handle that when actually freeing the folio? folio_test_anon()
> >> is sticky until freed.
> >
> > To be clearer: we increment the counter when we set a folio anon, which
> > should indeed only happen in folio_add_new_anon_rmap(). We'll have to
> > ignore hugetlb here where we do it in hugetlb_add_new_anon_rmap().
> >
> > Then, when we free an anon folio we decrement the counter. (hugetlb
> > should clear the anon flag when an anon folio gets freed back to its
> > allocator -- likely that is already done).
> >
>
> Sorry that I am talking to myself: I'm wondering if we also have to
> adjust the counter when splitting a large folio to multiple
> smaller-but-still-large folios.

Hi David,

The conceptual code is shown below. Does this make more
sense to you? we have a line "mod_mthp_stat(new_order,
MTHP_STAT_NR_ANON, 1 << (order - new_order));"

@@ -3270,8 +3272,9 @@ int split_huge_page_to_list_to_order(struct page *page, struct list_head *list,
 	struct deferred_split *ds_queue = get_deferred_split_queue(folio);
 	/* reset xarray order to new order after split */
 	XA_STATE_ORDER(xas, &folio->mapping->i_pages, folio->index, new_order);
-	struct anon_vma *anon_vma = NULL;
+	bool is_anon = folio_test_anon(folio);
 	struct address_space *mapping = NULL;
+	struct anon_vma *anon_vma = NULL;
 	int order = folio_order(folio);
 	int extra_pins, ret;
 	pgoff_t end;
@@ -3283,7 +3286,7 @@ int split_huge_page_to_list_to_order(struct page *page, struct list_head *list,
 	if (new_order >= folio_order(folio))
 		return -EINVAL;
 
-	if (folio_test_anon(folio)) {
+	if (is_anon) {
 		/* order-1 is not supported for anonymous THP. */
 		if (new_order == 1) {
 			VM_WARN_ONCE(1, "Cannot split to order-1 folio");
@@ -3323,7 +3326,7 @@ int split_huge_page_to_list_to_order(struct page *page, struct list_head *list,
 	if (folio_test_writeback(folio))
 		return -EBUSY;
 
-	if (folio_test_anon(folio)) {
+	if (is_anon) {
 		/*
 		 * The caller does not necessarily hold an mmap_lock that would
 		 * prevent the anon_vma disappearing so we first we take a
@@ -3437,6 +3440,10 @@ int split_huge_page_to_list_to_order(struct page *page, struct list_head *list,
 			}
 		}
 
+		if (is_anon) {
+			mod_mthp_stat(order, MTHP_STAT_NR_ANON, -1);
+			mod_mthp_stat(new_order, MTHP_STAT_NR_ANON, 1 << (order - new_order));
+		}
 		__split_huge_page(page, list, end, new_order);
 		ret = 0;
 	} else {
diff --git a/mm/page_alloc.c b/mm/page_alloc.c
index 408ef3d25cf5..c869d0601614 100644
--- a/mm/page_alloc.c
+++ b/mm/page_alloc.c
@@ -1039,6 +1039,7 @@ __always_inline bool free_pages_prepare(struct page *page,
 	bool skip_kasan_poison = should_skip_kasan_poison(page);
 	bool init = want_init_on_free();
 	bool compound = PageCompound(page);
+	bool anon = PageAnon(page);
 
 	VM_BUG_ON_PAGE(PageTail(page), page);
 
@@ -1130,6 +1131,9 @@ __always_inline bool free_pages_prepare(struct page *page,
 
 	debug_pagealloc_unmap_pages(page, 1 << order);
 
+	if (anon && compound)
+		mod_mthp_stat(order, MTHP_STAT_NR_ANON, -1);
+
 	return true;
 }
 
diff --git a/mm/rmap.c b/mm/rmap.c
index 8d432051e970..982862cbf5ba 100644
--- a/mm/rmap.c
+++ b/mm/rmap.c
@@ -1467,6 +1467,7 @@ void folio_add_new_anon_rmap(struct folio *folio, struct vm_area_struct *vma,
 	}
 
 	__folio_mod_stat(folio, nr, nr_pmdmapped);
+	mod_mthp_stat(folio_order(folio), MTHP_STAT_NR_ANON, 1);
 }
 
 static __always_inline void __folio_add_file_rmap(struct folio *folio,

Re: [PATCH RFC 1/2] mm: collect the number of anon large folios

[David Hildenbrand]

On 09.08.24 09:04, Barry Song wrote:
>>>> I would appreciate if we leave the rmap out here.
>>>>
>>>> Can't we handle that when actually freeing the folio? folio_test_anon()
>>>> is sticky until freed.
>>>
>>> To be clearer: we increment the counter when we set a folio anon, which
>>> should indeed only happen in folio_add_new_anon_rmap(). We'll have to
>>> ignore hugetlb here where we do it in hugetlb_add_new_anon_rmap().
>>>
>>> Then, when we free an anon folio we decrement the counter. (hugetlb
>>> should clear the anon flag when an anon folio gets freed back to its
>>> allocator -- likely that is already done).
>>>
>>
>> Sorry that I am talking to myself: I'm wondering if we also have to
>> adjust the counter when splitting a large folio to multiple
>> smaller-but-still-large folios.
> 
> Hi David,
> 
> The conceptual code is shown below. Does this make more
> sense to you? we have a line "mod_mthp_stat(new_order,
> MTHP_STAT_NR_ANON, 1 << (order - new_order));"
> 
> @@ -3270,8 +3272,9 @@ int split_huge_page_to_list_to_order(struct page *page, struct list_head *list,
>   	struct deferred_split *ds_queue = get_deferred_split_queue(folio);
>   	/* reset xarray order to new order after split */
>   	XA_STATE_ORDER(xas, &folio->mapping->i_pages, folio->index, new_order);
> -	struct anon_vma *anon_vma = NULL;
> +	bool is_anon = folio_test_anon(folio);
>   	struct address_space *mapping = NULL;
> +	struct anon_vma *anon_vma = NULL;
>   	int order = folio_order(folio);
>   	int extra_pins, ret;
>   	pgoff_t end;
> @@ -3283,7 +3286,7 @@ int split_huge_page_to_list_to_order(struct page *page, struct list_head *list,
>   	if (new_order >= folio_order(folio))
>   		return -EINVAL;
>   
> -	if (folio_test_anon(folio)) {
> +	if (is_anon) {
>   		/* order-1 is not supported for anonymous THP. */
>   		if (new_order == 1) {
>   			VM_WARN_ONCE(1, "Cannot split to order-1 folio");
> @@ -3323,7 +3326,7 @@ int split_huge_page_to_list_to_order(struct page *page, struct list_head *list,
>   	if (folio_test_writeback(folio))
>   		return -EBUSY;
>   
> -	if (folio_test_anon(folio)) {
> +	if (is_anon) {
>   		/*
>   		 * The caller does not necessarily hold an mmap_lock that would
>   		 * prevent the anon_vma disappearing so we first we take a
> @@ -3437,6 +3440,10 @@ int split_huge_page_to_list_to_order(struct page *page, struct list_head *list,
>   			}
>   		}
>   
> +		if (is_anon) {
> +			mod_mthp_stat(order, MTHP_STAT_NR_ANON, -1);
> +			mod_mthp_stat(new_order, MTHP_STAT_NR_ANON, 1 << (order - new_order));
> +		}
>   		__split_huge_page(page, list, end, new_order);
>   		ret = 0;
>   	} else {
> diff --git a/mm/page_alloc.c b/mm/page_alloc.c
> index 408ef3d25cf5..c869d0601614 100644
> --- a/mm/page_alloc.c
> +++ b/mm/page_alloc.c
> @@ -1039,6 +1039,7 @@ __always_inline bool free_pages_prepare(struct page *page,
>   	bool skip_kasan_poison = should_skip_kasan_poison(page);
>   	bool init = want_init_on_free();
>   	bool compound = PageCompound(page);
> +	bool anon = PageAnon(page);
>   
>   	VM_BUG_ON_PAGE(PageTail(page), page);
>   
> @@ -1130,6 +1131,9 @@ __always_inline bool free_pages_prepare(struct page *page,
>   
>   	debug_pagealloc_unmap_pages(page, 1 << order);
>   
> +	if (anon && compound)
> +		mod_mthp_stat(order, MTHP_STAT_NR_ANON, -1);
> +
>   	return true;

I'd have placed it here, when we are already passed the "PageMappingFlags" check and
shouldn't have any added overhead for most !anon pages IIRC (mostly only anon/ksm pages should
run into that path).

diff --git a/mm/page_alloc.c b/mm/page_alloc.c
index 408ef3d25cf5..a11b9dd62964 100644
--- a/mm/page_alloc.c
+++ b/mm/page_alloc.c
@@ -1079,8 +1079,11 @@ __always_inline bool free_pages_prepare(struct page *page,
                         (page + i)->flags &= ~PAGE_FLAGS_CHECK_AT_PREP;
                 }
         }
-       if (PageMappingFlags(page))
+       if (PageMappingFlags(page)) {
+               if (PageAnon(page) && compound)
+                       mod_mthp_stat(order, MTHP_STAT_NR_ANON, -1);
                 page->mapping = NULL;
+       }
         if (is_check_pages_enabled()) {
                 if (free_page_is_bad(page))
                         bad++;

Conceptually LGTM. We account an anon folio as long as it's anon,
even when still GUP-pinned after unmapping it or when temporarily
unmapping+remapping it during migration.

-- 
Cheers,

David / dhildenb

Re: [PATCH RFC 1/2] mm: collect the number of anon large folios

[Barry Song]

On Fri, Aug 9, 2024 at 7:22 PM David Hildenbrand <david@redhat.com> wrote:
>
> On 09.08.24 09:04, Barry Song wrote:
> >>>> I would appreciate if we leave the rmap out here.
> >>>>
> >>>> Can't we handle that when actually freeing the folio? folio_test_anon()
> >>>> is sticky until freed.
> >>>
> >>> To be clearer: we increment the counter when we set a folio anon, which
> >>> should indeed only happen in folio_add_new_anon_rmap(). We'll have to
> >>> ignore hugetlb here where we do it in hugetlb_add_new_anon_rmap().
> >>>
> >>> Then, when we free an anon folio we decrement the counter. (hugetlb
> >>> should clear the anon flag when an anon folio gets freed back to its
> >>> allocator -- likely that is already done).
> >>>
> >>
> >> Sorry that I am talking to myself: I'm wondering if we also have to
> >> adjust the counter when splitting a large folio to multiple
> >> smaller-but-still-large folios.
> >
> > Hi David,
> >
> > The conceptual code is shown below. Does this make more
> > sense to you? we have a line "mod_mthp_stat(new_order,
> > MTHP_STAT_NR_ANON, 1 << (order - new_order));"
> >
> > @@ -3270,8 +3272,9 @@ int split_huge_page_to_list_to_order(struct page *page, struct list_head *list,
> >       struct deferred_split *ds_queue = get_deferred_split_queue(folio);
> >       /* reset xarray order to new order after split */
> >       XA_STATE_ORDER(xas, &folio->mapping->i_pages, folio->index, new_order);
> > -     struct anon_vma *anon_vma = NULL;
> > +     bool is_anon = folio_test_anon(folio);
> >       struct address_space *mapping = NULL;
> > +     struct anon_vma *anon_vma = NULL;
> >       int order = folio_order(folio);
> >       int extra_pins, ret;
> >       pgoff_t end;
> > @@ -3283,7 +3286,7 @@ int split_huge_page_to_list_to_order(struct page *page, struct list_head *list,
> >       if (new_order >= folio_order(folio))
> >               return -EINVAL;
> >
> > -     if (folio_test_anon(folio)) {
> > +     if (is_anon) {
> >               /* order-1 is not supported for anonymous THP. */
> >               if (new_order == 1) {
> >                       VM_WARN_ONCE(1, "Cannot split to order-1 folio");
> > @@ -3323,7 +3326,7 @@ int split_huge_page_to_list_to_order(struct page *page, struct list_head *list,
> >       if (folio_test_writeback(folio))
> >               return -EBUSY;
> >
> > -     if (folio_test_anon(folio)) {
> > +     if (is_anon) {
> >               /*
> >                * The caller does not necessarily hold an mmap_lock that would
> >                * prevent the anon_vma disappearing so we first we take a
> > @@ -3437,6 +3440,10 @@ int split_huge_page_to_list_to_order(struct page *page, struct list_head *list,
> >                       }
> >               }
> >
> > +             if (is_anon) {
> > +                     mod_mthp_stat(order, MTHP_STAT_NR_ANON, -1);
> > +                     mod_mthp_stat(new_order, MTHP_STAT_NR_ANON, 1 << (order - new_order));
> > +             }
> >               __split_huge_page(page, list, end, new_order);
> >               ret = 0;
> >       } else {
> > diff --git a/mm/page_alloc.c b/mm/page_alloc.c
> > index 408ef3d25cf5..c869d0601614 100644
> > --- a/mm/page_alloc.c
> > +++ b/mm/page_alloc.c
> > @@ -1039,6 +1039,7 @@ __always_inline bool free_pages_prepare(struct page *page,
> >       bool skip_kasan_poison = should_skip_kasan_poison(page);
> >       bool init = want_init_on_free();
> >       bool compound = PageCompound(page);
> > +     bool anon = PageAnon(page);
> >
> >       VM_BUG_ON_PAGE(PageTail(page), page);
> >
> > @@ -1130,6 +1131,9 @@ __always_inline bool free_pages_prepare(struct page *page,
> >
> >       debug_pagealloc_unmap_pages(page, 1 << order);
> >
> > +     if (anon && compound)
> > +             mod_mthp_stat(order, MTHP_STAT_NR_ANON, -1);
> > +
> >       return true;
>
> I'd have placed it here, when we are already passed the "PageMappingFlags" check and
> shouldn't have any added overhead for most !anon pages IIRC (mostly only anon/ksm pages should
> run into that path).
>
> diff --git a/mm/page_alloc.c b/mm/page_alloc.c
> index 408ef3d25cf5..a11b9dd62964 100644
> --- a/mm/page_alloc.c
> +++ b/mm/page_alloc.c
> @@ -1079,8 +1079,11 @@ __always_inline bool free_pages_prepare(struct page *page,
>                          (page + i)->flags &= ~PAGE_FLAGS_CHECK_AT_PREP;
>                  }
>          }
> -       if (PageMappingFlags(page))
> +       if (PageMappingFlags(page)) {
> +               if (PageAnon(page) && compound)
> +                       mod_mthp_stat(order, MTHP_STAT_NR_ANON, -1);
>                  page->mapping = NULL;
> +       }
>          if (is_check_pages_enabled()) {
>                  if (free_page_is_bad(page))
>                          bad++;
>

This looks better, but we're not concerned about the bad pages, correct?
For bad pages, we're reducing the count by 1, but they aren't actually
freed. We don't need to worry about this since it's already considered
a bug, right?

> Conceptually LGTM. We account an anon folio as long as it's anon,
> even when still GUP-pinned after unmapping it or when temporarily
> unmapping+remapping it during migration.

right. but migration might be a problem? as the dst folio doesn't
call folio_add_new_anon_rmap(), dst->mapping is copied from
src. So I suspect we need the below (otherwise, src has been put
and got -1, but dst won't get +1),

diff --git a/mm/migrate.c b/mm/migrate.c
index 7e1267042a56..11ef11e59036 100644
--- a/mm/migrate.c
+++ b/mm/migrate.c
@@ -1102,6 +1102,8 @@ static void migrate_folio_done(struct folio *src,
                mod_node_page_state(folio_pgdat(src), NR_ISOLATED_ANON +
                                    folio_is_file_lru(src),
-folio_nr_pages(src));

+       mod_mthp_stat(folio_order(src), MTHP_STAT_NR_ANON, 1);
+
        if (reason != MR_MEMORY_FAILURE)
                /* We release the page in page_handle_poison. */
                folio_put(src);


>
> --
> Cheers,
>
> David / dhildenb
>

Thanks
Barry

Re: [PATCH RFC 1/2] mm: collect the number of anon large folios

[Barry Song]

On Sun, Aug 11, 2024 at 5:20 PM Barry Song <21cnbao@gmail.com> wrote:
>
> On Fri, Aug 9, 2024 at 7:22 PM David Hildenbrand <david@redhat.com> wrote:
> >
> > On 09.08.24 09:04, Barry Song wrote:
> > >>>> I would appreciate if we leave the rmap out here.
> > >>>>
> > >>>> Can't we handle that when actually freeing the folio? folio_test_anon()
> > >>>> is sticky until freed.
> > >>>
> > >>> To be clearer: we increment the counter when we set a folio anon, which
> > >>> should indeed only happen in folio_add_new_anon_rmap(). We'll have to
> > >>> ignore hugetlb here where we do it in hugetlb_add_new_anon_rmap().
> > >>>
> > >>> Then, when we free an anon folio we decrement the counter. (hugetlb
> > >>> should clear the anon flag when an anon folio gets freed back to its
> > >>> allocator -- likely that is already done).
> > >>>
> > >>
> > >> Sorry that I am talking to myself: I'm wondering if we also have to
> > >> adjust the counter when splitting a large folio to multiple
> > >> smaller-but-still-large folios.
> > >
> > > Hi David,
> > >
> > > The conceptual code is shown below. Does this make more
> > > sense to you? we have a line "mod_mthp_stat(new_order,
> > > MTHP_STAT_NR_ANON, 1 << (order - new_order));"
> > >
> > > @@ -3270,8 +3272,9 @@ int split_huge_page_to_list_to_order(struct page *page, struct list_head *list,
> > >       struct deferred_split *ds_queue = get_deferred_split_queue(folio);
> > >       /* reset xarray order to new order after split */
> > >       XA_STATE_ORDER(xas, &folio->mapping->i_pages, folio->index, new_order);
> > > -     struct anon_vma *anon_vma = NULL;
> > > +     bool is_anon = folio_test_anon(folio);
> > >       struct address_space *mapping = NULL;
> > > +     struct anon_vma *anon_vma = NULL;
> > >       int order = folio_order(folio);
> > >       int extra_pins, ret;
> > >       pgoff_t end;
> > > @@ -3283,7 +3286,7 @@ int split_huge_page_to_list_to_order(struct page *page, struct list_head *list,
> > >       if (new_order >= folio_order(folio))
> > >               return -EINVAL;
> > >
> > > -     if (folio_test_anon(folio)) {
> > > +     if (is_anon) {
> > >               /* order-1 is not supported for anonymous THP. */
> > >               if (new_order == 1) {
> > >                       VM_WARN_ONCE(1, "Cannot split to order-1 folio");
> > > @@ -3323,7 +3326,7 @@ int split_huge_page_to_list_to_order(struct page *page, struct list_head *list,
> > >       if (folio_test_writeback(folio))
> > >               return -EBUSY;
> > >
> > > -     if (folio_test_anon(folio)) {
> > > +     if (is_anon) {
> > >               /*
> > >                * The caller does not necessarily hold an mmap_lock that would
> > >                * prevent the anon_vma disappearing so we first we take a
> > > @@ -3437,6 +3440,10 @@ int split_huge_page_to_list_to_order(struct page *page, struct list_head *list,
> > >                       }
> > >               }
> > >
> > > +             if (is_anon) {
> > > +                     mod_mthp_stat(order, MTHP_STAT_NR_ANON, -1);
> > > +                     mod_mthp_stat(new_order, MTHP_STAT_NR_ANON, 1 << (order - new_order));
> > > +             }
> > >               __split_huge_page(page, list, end, new_order);
> > >               ret = 0;
> > >       } else {
> > > diff --git a/mm/page_alloc.c b/mm/page_alloc.c
> > > index 408ef3d25cf5..c869d0601614 100644
> > > --- a/mm/page_alloc.c
> > > +++ b/mm/page_alloc.c
> > > @@ -1039,6 +1039,7 @@ __always_inline bool free_pages_prepare(struct page *page,
> > >       bool skip_kasan_poison = should_skip_kasan_poison(page);
> > >       bool init = want_init_on_free();
> > >       bool compound = PageCompound(page);
> > > +     bool anon = PageAnon(page);
> > >
> > >       VM_BUG_ON_PAGE(PageTail(page), page);
> > >
> > > @@ -1130,6 +1131,9 @@ __always_inline bool free_pages_prepare(struct page *page,
> > >
> > >       debug_pagealloc_unmap_pages(page, 1 << order);
> > >
> > > +     if (anon && compound)
> > > +             mod_mthp_stat(order, MTHP_STAT_NR_ANON, -1);
> > > +
> > >       return true;
> >
> > I'd have placed it here, when we are already passed the "PageMappingFlags" check and
> > shouldn't have any added overhead for most !anon pages IIRC (mostly only anon/ksm pages should
> > run into that path).
> >
> > diff --git a/mm/page_alloc.c b/mm/page_alloc.c
> > index 408ef3d25cf5..a11b9dd62964 100644
> > --- a/mm/page_alloc.c
> > +++ b/mm/page_alloc.c
> > @@ -1079,8 +1079,11 @@ __always_inline bool free_pages_prepare(struct page *page,
> >                          (page + i)->flags &= ~PAGE_FLAGS_CHECK_AT_PREP;
> >                  }
> >          }
> > -       if (PageMappingFlags(page))
> > +       if (PageMappingFlags(page)) {
> > +               if (PageAnon(page) && compound)
> > +                       mod_mthp_stat(order, MTHP_STAT_NR_ANON, -1);
> >                  page->mapping = NULL;
> > +       }
> >          if (is_check_pages_enabled()) {
> >                  if (free_page_is_bad(page))
> >                          bad++;
> >
>
> This looks better, but we're not concerned about the bad pages, correct?
> For bad pages, we're reducing the count by 1, but they aren't actually
> freed. We don't need to worry about this since it's already considered
> a bug, right?
>
> > Conceptually LGTM. We account an anon folio as long as it's anon,
> > even when still GUP-pinned after unmapping it or when temporarily
> > unmapping+remapping it during migration.
>
> right. but migration might be a problem? as the dst folio doesn't
> call folio_add_new_anon_rmap(), dst->mapping is copied from
> src. So I suspect we need the below (otherwise, src has been put
> and got -1, but dst won't get +1),
>
> diff --git a/mm/migrate.c b/mm/migrate.c
> index 7e1267042a56..11ef11e59036 100644
> --- a/mm/migrate.c
> +++ b/mm/migrate.c
> @@ -1102,6 +1102,8 @@ static void migrate_folio_done(struct folio *src,
>                 mod_node_page_state(folio_pgdat(src), NR_ISOLATED_ANON +
>                                     folio_is_file_lru(src),
> -folio_nr_pages(src));
>
> +       mod_mthp_stat(folio_order(src), MTHP_STAT_NR_ANON, 1);
> +
>         if (reason != MR_MEMORY_FAILURE)
>                 /* We release the page in page_handle_poison. */
>                 folio_put(src);
>

sorry. wrong place as migrate_folio_done() doesn't necessarily mean
migration is done. it could be "Folio was freed from under us"

        if (folio_ref_count(src) == 1) {
                /* Folio was freed from under us. So we are done. */
                folio_clear_active(src);
                folio_clear_unevictable(src);
                /* free_pages_prepare() will clear PG_isolated. */
                list_del(&src->lru);
                migrate_folio_done(src, reason);
                return MIGRATEPAGE_SUCCESS;
        }


the correct place should be:

@@ -1329,6 +1326,10 @@ static int migrate_folio_move(free_folio_t
put_new_folio, unsigned long private,
        if (anon_vma)
                put_anon_vma(anon_vma);
        folio_unlock(src);
+
+       if (folio_test_anon(src))
+               mod_mthp_stat(folio_order(src), MTHP_STAT_NR_ANON, 1);
+
        migrate_folio_done(src, reason);

        return rc;

Without this modification in migration code, my tests fail, anon_num can
become negative.

>
> >
> > --
> > Cheers,
> >
> > David / dhildenb
> >

Thanks
Barry

Re: [PATCH RFC 1/2] mm: collect the number of anon large folios

[David Hildenbrand]

> the correct place should be:
> 
> @@ -1329,6 +1326,10 @@ static int migrate_folio_move(free_folio_t
> put_new_folio, unsigned long private,
>          if (anon_vma)
>                  put_anon_vma(anon_vma);
>          folio_unlock(src);
> +
> +       if (folio_test_anon(src))
> +               mod_mthp_stat(folio_order(src), MTHP_STAT_NR_ANON, 1);
> +
>          migrate_folio_done(src, reason);
> 
>          return rc;
> 
> Without this modification in migration code, my tests fail, anon_num can
> become negative.

I was wondering if we should do it in __folio_migrate_mapping().

There, we set newfolio->mapping.

-- 
Cheers,

David / dhildenb

Re: [PATCH RFC 1/2] mm: collect the number of anon large folios

[Barry Song]

On Sun, Aug 11, 2024 at 8:51 PM David Hildenbrand <david@redhat.com> wrote:
>
> > the correct place should be:
> >
> > @@ -1329,6 +1326,10 @@ static int migrate_folio_move(free_folio_t
> > put_new_folio, unsigned long private,
> >          if (anon_vma)
> >                  put_anon_vma(anon_vma);
> >          folio_unlock(src);
> > +
> > +       if (folio_test_anon(src))
> > +               mod_mthp_stat(folio_order(src), MTHP_STAT_NR_ANON, 1);
> > +
> >          migrate_folio_done(src, reason);
> >
> >          return rc;
> >
> > Without this modification in migration code, my tests fail, anon_num can
> > become negative.
>
> I was wondering if we should do it in __folio_migrate_mapping().
>
> There, we set newfolio->mapping.

Correct! To handle cases where the destination might not be migrated to but
could be put or freed, migrate_folio_undo_dst() will be called to release the
destination earlier. It's better to simply place the count where the
anon mapping
is set and cleared.

--- a/mm/migrate.c
+++ b/mm/migrate.c
@@ -423,6 +423,8 @@ static int __folio_migrate_mapping(struct
address_space *mapping,
                /* No turning back from here */
                newfolio->index = folio->index;
                newfolio->mapping = folio->mapping;
+               if (folio_test_anon(folio) && folio_test_large(folio))
+                       mod_mthp_stat(folio_order(folio), MTHP_STAT_NR_ANON, 1);
                if (folio_test_swapbacked(folio))
                        __folio_set_swapbacked(newfolio);

@@ -446,6 +448,9 @@ static int __folio_migrate_mapping(struct
address_space *mapping,
         * no turning back from here.
         */
        newfolio->index = folio->index;
+
+       if (folio_test_anon(folio) && folio_test_large(folio))
+               mod_mthp_stat(folio_order(folio), MTHP_STAT_NR_ANON, 1);
        newfolio->mapping = folio->mapping;
        folio_ref_add(newfolio, nr); /* add cache reference */
        if (folio_test_swapbacked(folio)) {


>
> --
> Cheers,
>
> David / dhildenb
>

Re: [PATCH RFC 1/2] mm: collect the number of anon large folios

[Ryan Roberts]

On 08/08/2024 02:04, Barry Song wrote:
> From: Barry Song <v-songbaohua@oppo.com>
> 
> When a new anonymous mTHP is added to the rmap, we increase the count.
> We reduce the count whenever an mTHP is completely unmapped.
> 
> Signed-off-by: Barry Song <v-songbaohua@oppo.com>
> ---
>  Documentation/admin-guide/mm/transhuge.rst |  5 +++++
>  include/linux/huge_mm.h                    | 15 +++++++++++++--
>  mm/huge_memory.c                           |  2 ++
>  mm/rmap.c                                  |  3 +++
>  4 files changed, 23 insertions(+), 2 deletions(-)
> 
> diff --git a/Documentation/admin-guide/mm/transhuge.rst b/Documentation/admin-guide/mm/transhuge.rst
> index 058485daf186..715f181543f6 100644
> --- a/Documentation/admin-guide/mm/transhuge.rst
> +++ b/Documentation/admin-guide/mm/transhuge.rst
> @@ -527,6 +527,11 @@ split_deferred
>          it would free up some memory. Pages on split queue are going to
>          be split under memory pressure, if splitting is possible.
>  
> +anon_num
> +       the number of anon huge pages we have in the whole system.
> +       These huge pages could be still entirely mapped and have partially
> +       unmapped and unused subpages.

nit: "entirely mapped and have partially unmapped and unused subpages" ->
"entirely mapped or have partially unmapped/unused subpages"

> +
>  As the system ages, allocating huge pages may be expensive as the
>  system uses memory compaction to copy data around memory to free a
>  huge page for use. There are some counters in ``/proc/vmstat`` to help
> diff --git a/include/linux/huge_mm.h b/include/linux/huge_mm.h
> index e25d9ebfdf89..294c348fe3cc 100644
> --- a/include/linux/huge_mm.h
> +++ b/include/linux/huge_mm.h
> @@ -281,6 +281,7 @@ enum mthp_stat_item {
>  	MTHP_STAT_SPLIT,
>  	MTHP_STAT_SPLIT_FAILED,
>  	MTHP_STAT_SPLIT_DEFERRED,
> +	MTHP_STAT_NR_ANON,
>  	__MTHP_STAT_COUNT
>  };
>  
> @@ -291,14 +292,24 @@ struct mthp_stat {
>  #ifdef CONFIG_SYSFS
>  DECLARE_PER_CPU(struct mthp_stat, mthp_stats);
>  
> -static inline void count_mthp_stat(int order, enum mthp_stat_item item)
> +static inline void mod_mthp_stat(int order, enum mthp_stat_item item, int delta)
>  {
>  	if (order <= 0 || order > PMD_ORDER)
>  		return;
>  
> -	this_cpu_inc(mthp_stats.stats[order][item]);
> +	this_cpu_add(mthp_stats.stats[order][item], delta);
> +}
> +
> +static inline void count_mthp_stat(int order, enum mthp_stat_item item)
> +{
> +	mod_mthp_stat(order, item, 1);
>  }
> +
>  #else
> +static inline void mod_mthp_stat(int order, enum mthp_stat_item item, int delta)
> +{
> +}
> +
>  static inline void count_mthp_stat(int order, enum mthp_stat_item item)
>  {
>  }
> diff --git a/mm/huge_memory.c b/mm/huge_memory.c
> index 697fcf89f975..b6bc2a3791e3 100644
> --- a/mm/huge_memory.c
> +++ b/mm/huge_memory.c
> @@ -578,6 +578,7 @@ DEFINE_MTHP_STAT_ATTR(shmem_fallback_charge, MTHP_STAT_SHMEM_FALLBACK_CHARGE);
>  DEFINE_MTHP_STAT_ATTR(split, MTHP_STAT_SPLIT);
>  DEFINE_MTHP_STAT_ATTR(split_failed, MTHP_STAT_SPLIT_FAILED);
>  DEFINE_MTHP_STAT_ATTR(split_deferred, MTHP_STAT_SPLIT_DEFERRED);
> +DEFINE_MTHP_STAT_ATTR(anon_num, MTHP_STAT_NR_ANON);
>  
>  static struct attribute *stats_attrs[] = {
>  	&anon_fault_alloc_attr.attr,
> @@ -591,6 +592,7 @@ static struct attribute *stats_attrs[] = {
>  	&split_attr.attr,
>  	&split_failed_attr.attr,
>  	&split_deferred_attr.attr,
> +	&anon_num_attr.attr,
>  	NULL,
>  };
>  
> diff --git a/mm/rmap.c b/mm/rmap.c
> index 901950200957..2b722f26224c 100644
> --- a/mm/rmap.c
> +++ b/mm/rmap.c
> @@ -1467,6 +1467,7 @@ void folio_add_new_anon_rmap(struct folio *folio, struct vm_area_struct *vma,
>  	}
>  
>  	__folio_mod_stat(folio, nr, nr_pmdmapped);
> +	mod_mthp_stat(folio_order(folio), MTHP_STAT_NR_ANON, 1);
>  }
>  
>  static __always_inline void __folio_add_file_rmap(struct folio *folio,
> @@ -1582,6 +1583,8 @@ static __always_inline void __folio_remove_rmap(struct folio *folio,
>  	    list_empty(&folio->_deferred_list))
>  		deferred_split_folio(folio);
>  	__folio_mod_stat(folio, -nr, -nr_pmdmapped);
> +	if (folio_test_anon(folio) && !atomic_read(mapped))

Agree that atomic_read() is dodgy here.

Not sure I fully understand why David prefers to do the unaccounting at
free-time though? It feels unbalanced to me to increment when first mapped but
decrement when freed. Surely its safer to either use alloc/free or use first
map/last map?

If using alloc/free isn't there a THP constructor/destructor that prepares the
deferred list? (My memory may be failing me). Could we use that?

> +		mod_mthp_stat(folio_order(folio), MTHP_STAT_NR_ANON, -1);
>  
>  	/*
>  	 * It would be tidy to reset folio_test_anon mapping when fully

Re: [PATCH RFC 1/2] mm: collect the number of anon large folios

[Ryan Roberts]

On 09/08/2024 09:13, Ryan Roberts wrote:
> On 08/08/2024 02:04, Barry Song wrote:
>> From: Barry Song <v-songbaohua@oppo.com>
>>
>> When a new anonymous mTHP is added to the rmap, we increase the count.
>> We reduce the count whenever an mTHP is completely unmapped.
>>
>> Signed-off-by: Barry Song <v-songbaohua@oppo.com>
>> ---
>>  Documentation/admin-guide/mm/transhuge.rst |  5 +++++
>>  include/linux/huge_mm.h                    | 15 +++++++++++++--
>>  mm/huge_memory.c                           |  2 ++
>>  mm/rmap.c                                  |  3 +++
>>  4 files changed, 23 insertions(+), 2 deletions(-)
>>
>> diff --git a/Documentation/admin-guide/mm/transhuge.rst b/Documentation/admin-guide/mm/transhuge.rst
>> index 058485daf186..715f181543f6 100644
>> --- a/Documentation/admin-guide/mm/transhuge.rst
>> +++ b/Documentation/admin-guide/mm/transhuge.rst
>> @@ -527,6 +527,11 @@ split_deferred
>>          it would free up some memory. Pages on split queue are going to
>>          be split under memory pressure, if splitting is possible.
>>  
>> +anon_num
>> +       the number of anon huge pages we have in the whole system.
>> +       These huge pages could be still entirely mapped and have partially
>> +       unmapped and unused subpages.
> 
> nit: "entirely mapped and have partially unmapped and unused subpages" ->
> "entirely mapped or have partially unmapped/unused subpages"
> 
>> +
>>  As the system ages, allocating huge pages may be expensive as the
>>  system uses memory compaction to copy data around memory to free a
>>  huge page for use. There are some counters in ``/proc/vmstat`` to help
>> diff --git a/include/linux/huge_mm.h b/include/linux/huge_mm.h
>> index e25d9ebfdf89..294c348fe3cc 100644
>> --- a/include/linux/huge_mm.h
>> +++ b/include/linux/huge_mm.h
>> @@ -281,6 +281,7 @@ enum mthp_stat_item {
>>  	MTHP_STAT_SPLIT,
>>  	MTHP_STAT_SPLIT_FAILED,
>>  	MTHP_STAT_SPLIT_DEFERRED,
>> +	MTHP_STAT_NR_ANON,
>>  	__MTHP_STAT_COUNT
>>  };
>>  
>> @@ -291,14 +292,24 @@ struct mthp_stat {
>>  #ifdef CONFIG_SYSFS
>>  DECLARE_PER_CPU(struct mthp_stat, mthp_stats);
>>  
>> -static inline void count_mthp_stat(int order, enum mthp_stat_item item)
>> +static inline void mod_mthp_stat(int order, enum mthp_stat_item item, int delta)
>>  {
>>  	if (order <= 0 || order > PMD_ORDER)
>>  		return;
>>  
>> -	this_cpu_inc(mthp_stats.stats[order][item]);
>> +	this_cpu_add(mthp_stats.stats[order][item], delta);
>> +}
>> +
>> +static inline void count_mthp_stat(int order, enum mthp_stat_item item)
>> +{
>> +	mod_mthp_stat(order, item, 1);
>>  }
>> +
>>  #else
>> +static inline void mod_mthp_stat(int order, enum mthp_stat_item item, int delta)
>> +{
>> +}
>> +
>>  static inline void count_mthp_stat(int order, enum mthp_stat_item item)
>>  {
>>  }
>> diff --git a/mm/huge_memory.c b/mm/huge_memory.c
>> index 697fcf89f975..b6bc2a3791e3 100644
>> --- a/mm/huge_memory.c
>> +++ b/mm/huge_memory.c
>> @@ -578,6 +578,7 @@ DEFINE_MTHP_STAT_ATTR(shmem_fallback_charge, MTHP_STAT_SHMEM_FALLBACK_CHARGE);
>>  DEFINE_MTHP_STAT_ATTR(split, MTHP_STAT_SPLIT);
>>  DEFINE_MTHP_STAT_ATTR(split_failed, MTHP_STAT_SPLIT_FAILED);
>>  DEFINE_MTHP_STAT_ATTR(split_deferred, MTHP_STAT_SPLIT_DEFERRED);
>> +DEFINE_MTHP_STAT_ATTR(anon_num, MTHP_STAT_NR_ANON);

Why are the user-facing and internal names different? Perhaps it would be
clearer to call this nr_anon in sysfs?

>>  
>>  static struct attribute *stats_attrs[] = {
>>  	&anon_fault_alloc_attr.attr,
>> @@ -591,6 +592,7 @@ static struct attribute *stats_attrs[] = {
>>  	&split_attr.attr,
>>  	&split_failed_attr.attr,
>>  	&split_deferred_attr.attr,
>> +	&anon_num_attr.attr,
>>  	NULL,
>>  };
>>  
>> diff --git a/mm/rmap.c b/mm/rmap.c
>> index 901950200957..2b722f26224c 100644
>> --- a/mm/rmap.c
>> +++ b/mm/rmap.c
>> @@ -1467,6 +1467,7 @@ void folio_add_new_anon_rmap(struct folio *folio, struct vm_area_struct *vma,
>>  	}
>>  
>>  	__folio_mod_stat(folio, nr, nr_pmdmapped);
>> +	mod_mthp_stat(folio_order(folio), MTHP_STAT_NR_ANON, 1);
>>  }
>>  
>>  static __always_inline void __folio_add_file_rmap(struct folio *folio,
>> @@ -1582,6 +1583,8 @@ static __always_inline void __folio_remove_rmap(struct folio *folio,
>>  	    list_empty(&folio->_deferred_list))
>>  		deferred_split_folio(folio);
>>  	__folio_mod_stat(folio, -nr, -nr_pmdmapped);
>> +	if (folio_test_anon(folio) && !atomic_read(mapped))
> 
> Agree that atomic_read() is dodgy here.
> 
> Not sure I fully understand why David prefers to do the unaccounting at
> free-time though? It feels unbalanced to me to increment when first mapped but
> decrement when freed. Surely its safer to either use alloc/free or use first
> map/last map?
> 
> If using alloc/free isn't there a THP constructor/destructor that prepares the
> deferred list? (My memory may be failing me). Could we use that?

Additionally, if we wanted to extend (eventually) to track the number of shmem
and file mthps in additional counters, could we also account using similar folio
free-time hooks? If not, it might be an argument to account in rmap_unmap to be
consistent for all?


> 
>> +		mod_mthp_stat(folio_order(folio), MTHP_STAT_NR_ANON, -1);
>>  
>>  	/*
>>  	 * It would be tidy to reset folio_test_anon mapping when fully
> 

Re: [PATCH RFC 1/2] mm: collect the number of anon large folios

[Barry Song]

On Fri, Aug 9, 2024 at 4:27 PM Ryan Roberts <ryan.roberts@arm.com> wrote:
>
> On 09/08/2024 09:13, Ryan Roberts wrote:
> > On 08/08/2024 02:04, Barry Song wrote:
> >> From: Barry Song <v-songbaohua@oppo.com>
> >>
> >> When a new anonymous mTHP is added to the rmap, we increase the count.
> >> We reduce the count whenever an mTHP is completely unmapped.
> >>
> >> Signed-off-by: Barry Song <v-songbaohua@oppo.com>
> >> ---
> >>  Documentation/admin-guide/mm/transhuge.rst |  5 +++++
> >>  include/linux/huge_mm.h                    | 15 +++++++++++++--
> >>  mm/huge_memory.c                           |  2 ++
> >>  mm/rmap.c                                  |  3 +++
> >>  4 files changed, 23 insertions(+), 2 deletions(-)
> >>
> >> diff --git a/Documentation/admin-guide/mm/transhuge.rst b/Documentation/admin-guide/mm/transhuge.rst
> >> index 058485daf186..715f181543f6 100644
> >> --- a/Documentation/admin-guide/mm/transhuge.rst
> >> +++ b/Documentation/admin-guide/mm/transhuge.rst
> >> @@ -527,6 +527,11 @@ split_deferred
> >>          it would free up some memory. Pages on split queue are going to
> >>          be split under memory pressure, if splitting is possible.
> >>
> >> +anon_num
> >> +       the number of anon huge pages we have in the whole system.
> >> +       These huge pages could be still entirely mapped and have partially
> >> +       unmapped and unused subpages.
> >
> > nit: "entirely mapped and have partially unmapped and unused subpages" ->
> > "entirely mapped or have partially unmapped/unused subpages"
> >
> >> +
> >>  As the system ages, allocating huge pages may be expensive as the
> >>  system uses memory compaction to copy data around memory to free a
> >>  huge page for use. There are some counters in ``/proc/vmstat`` to help
> >> diff --git a/include/linux/huge_mm.h b/include/linux/huge_mm.h
> >> index e25d9ebfdf89..294c348fe3cc 100644
> >> --- a/include/linux/huge_mm.h
> >> +++ b/include/linux/huge_mm.h
> >> @@ -281,6 +281,7 @@ enum mthp_stat_item {
> >>      MTHP_STAT_SPLIT,
> >>      MTHP_STAT_SPLIT_FAILED,
> >>      MTHP_STAT_SPLIT_DEFERRED,
> >> +    MTHP_STAT_NR_ANON,
> >>      __MTHP_STAT_COUNT
> >>  };
> >>
> >> @@ -291,14 +292,24 @@ struct mthp_stat {
> >>  #ifdef CONFIG_SYSFS
> >>  DECLARE_PER_CPU(struct mthp_stat, mthp_stats);
> >>
> >> -static inline void count_mthp_stat(int order, enum mthp_stat_item item)
> >> +static inline void mod_mthp_stat(int order, enum mthp_stat_item item, int delta)
> >>  {
> >>      if (order <= 0 || order > PMD_ORDER)
> >>              return;
> >>
> >> -    this_cpu_inc(mthp_stats.stats[order][item]);
> >> +    this_cpu_add(mthp_stats.stats[order][item], delta);
> >> +}
> >> +
> >> +static inline void count_mthp_stat(int order, enum mthp_stat_item item)
> >> +{
> >> +    mod_mthp_stat(order, item, 1);
> >>  }
> >> +
> >>  #else
> >> +static inline void mod_mthp_stat(int order, enum mthp_stat_item item, int delta)
> >> +{
> >> +}
> >> +
> >>  static inline void count_mthp_stat(int order, enum mthp_stat_item item)
> >>  {
> >>  }
> >> diff --git a/mm/huge_memory.c b/mm/huge_memory.c
> >> index 697fcf89f975..b6bc2a3791e3 100644
> >> --- a/mm/huge_memory.c
> >> +++ b/mm/huge_memory.c
> >> @@ -578,6 +578,7 @@ DEFINE_MTHP_STAT_ATTR(shmem_fallback_charge, MTHP_STAT_SHMEM_FALLBACK_CHARGE);
> >>  DEFINE_MTHP_STAT_ATTR(split, MTHP_STAT_SPLIT);
> >>  DEFINE_MTHP_STAT_ATTR(split_failed, MTHP_STAT_SPLIT_FAILED);
> >>  DEFINE_MTHP_STAT_ATTR(split_deferred, MTHP_STAT_SPLIT_DEFERRED);
> >> +DEFINE_MTHP_STAT_ATTR(anon_num, MTHP_STAT_NR_ANON);
>
> Why are the user-facing and internal names different? Perhaps it would be
> clearer to call this nr_anon in sysfs?
>
> >>
> >>  static struct attribute *stats_attrs[] = {
> >>      &anon_fault_alloc_attr.attr,
> >> @@ -591,6 +592,7 @@ static struct attribute *stats_attrs[] = {
> >>      &split_attr.attr,
> >>      &split_failed_attr.attr,
> >>      &split_deferred_attr.attr,
> >> +    &anon_num_attr.attr,
> >>      NULL,
> >>  };
> >>
> >> diff --git a/mm/rmap.c b/mm/rmap.c
> >> index 901950200957..2b722f26224c 100644
> >> --- a/mm/rmap.c
> >> +++ b/mm/rmap.c
> >> @@ -1467,6 +1467,7 @@ void folio_add_new_anon_rmap(struct folio *folio, struct vm_area_struct *vma,
> >>      }
> >>
> >>      __folio_mod_stat(folio, nr, nr_pmdmapped);
> >> +    mod_mthp_stat(folio_order(folio), MTHP_STAT_NR_ANON, 1);
> >>  }
> >>
> >>  static __always_inline void __folio_add_file_rmap(struct folio *folio,
> >> @@ -1582,6 +1583,8 @@ static __always_inline void __folio_remove_rmap(struct folio *folio,
> >>          list_empty(&folio->_deferred_list))
> >>              deferred_split_folio(folio);
> >>      __folio_mod_stat(folio, -nr, -nr_pmdmapped);
> >> +    if (folio_test_anon(folio) && !atomic_read(mapped))
> >
> > Agree that atomic_read() is dodgy here.
> >
> > Not sure I fully understand why David prefers to do the unaccounting at
> > free-time though? It feels unbalanced to me to increment when first mapped but
> > decrement when freed. Surely its safer to either use alloc/free or use first
> > map/last map?

As long as we can account for mTHP when clearing the Anon flag for the folio,
we should be safe. It’s challenging to add +1 when allocating a large folio
because we don’t know its intended use—it could be for file, anon, or shmem.

> >
> > If using alloc/free isn't there a THP constructor/destructor that prepares the
> > deferred list? (My memory may be failing me). Could we use that?
>
> Additionally, if we wanted to extend (eventually) to track the number of shmem
> and file mthps in additional counters, could we also account using similar folio
> free-time hooks? If not, it might be an argument to account in rmap_unmap to be
> consistent for all?

I've been struggling quite a bit with rmap. Despite trying various
approaches, I’m
still occasionally seeing a negative mTHP counter after running it some hours
on phones. It seems that rmap is really tricky to handle.  I admit that I have
failed on rmap :-)

On the other hand, for anon folios, we have cases where they are split from
order M to order N. So, we add +1 when a new anon folio is added to rmap
and subtract -1 when we either split it or free it. This approach seems clearer
to me. When we split from order M to order N, we can add 1 << (M - N) for
order N.

>
>
> >
> >> +            mod_mthp_stat(folio_order(folio), MTHP_STAT_NR_ANON, -1);
> >>
> >>      /*
> >>       * It would be tidy to reset folio_test_anon mapping when fully

Thanks
Barry

Re: [PATCH RFC 1/2] mm: collect the number of anon large folios

[David Hildenbrand]

>> Not sure I fully understand why David prefers to do the unaccounting at
>> free-time though? It feels unbalanced to me to increment when first mapped but
>> decrement when freed. Surely its safer to either use alloc/free or use first
>> map/last map?
>>
>> If using alloc/free isn't there a THP constructor/destructor that prepares the
>> deferred list? (My memory may be failing me). Could we use that?
> 
> Additionally, if we wanted to extend (eventually) to track the number of shmem
> and file mthps in additional counters, could we also account using similar folio
> free-time hooks? If not, it might be an argument to account in rmap_unmap to be
> consistent for all?

Again, see NR_FILE_THPS handling. No rmap over-complication please.

-- 
Cheers,

David / dhildenb

Re: [PATCH RFC 1/2] mm: collect the number of anon large folios

[David Hildenbrand]

On 09.08.24 10:42, David Hildenbrand wrote:
>>> Not sure I fully understand why David prefers to do the unaccounting at
>>> free-time though? It feels unbalanced to me to increment when first mapped but
>>> decrement when freed. Surely its safer to either use alloc/free or use first
>>> map/last map?
>>>
>>> If using alloc/free isn't there a THP constructor/destructor that prepares the
>>> deferred list? (My memory may be failing me). Could we use that?
>>
>> Additionally, if we wanted to extend (eventually) to track the number of shmem
>> and file mthps in additional counters, could we also account using similar folio
>> free-time hooks? If not, it might be an argument to account in rmap_unmap to be
>> consistent for all?
> 
> Again, see NR_FILE_THPS handling. No rmap over-complication please.

... not to mention that it is non-sensical to only count pageache folios 
that are mapped to user space ;)

-- 
Cheers,

David / dhildenb

Re: [PATCH RFC 1/2] mm: collect the number of anon large folios

[Ryan Roberts]

On 09/08/2024 09:58, David Hildenbrand wrote:
> On 09.08.24 10:42, David Hildenbrand wrote:
>>>> Not sure I fully understand why David prefers to do the unaccounting at
>>>> free-time though? It feels unbalanced to me to increment when first mapped but
>>>> decrement when freed. Surely its safer to either use alloc/free or use first
>>>> map/last map?
>>>>
>>>> If using alloc/free isn't there a THP constructor/destructor that prepares the
>>>> deferred list? (My memory may be failing me). Could we use that?
>>>
>>> Additionally, if we wanted to extend (eventually) to track the number of shmem
>>> and file mthps in additional counters, could we also account using similar folio
>>> free-time hooks? If not, it might be an argument to account in rmap_unmap to be
>>> consistent for all?
>>
>> Again, see NR_FILE_THPS handling. No rmap over-complication please.
> 
> ... not to mention that it is non-sensical to only count pageache folios that
> are mapped to user space ;)

Yes, good point. I'll get back in my box. :)

Re: [PATCH RFC 1/2] mm: collect the number of anon large folios

[David Hildenbrand]

On 09.08.24 11:05, Ryan Roberts wrote:
> On 09/08/2024 09:58, David Hildenbrand wrote:
>> On 09.08.24 10:42, David Hildenbrand wrote:
>>>>> Not sure I fully understand why David prefers to do the unaccounting at
>>>>> free-time though? It feels unbalanced to me to increment when first mapped but
>>>>> decrement when freed. Surely its safer to either use alloc/free or use first
>>>>> map/last map?
>>>>>
>>>>> If using alloc/free isn't there a THP constructor/destructor that prepares the
>>>>> deferred list? (My memory may be failing me). Could we use that?
>>>>
>>>> Additionally, if we wanted to extend (eventually) to track the number of shmem
>>>> and file mthps in additional counters, could we also account using similar folio
>>>> free-time hooks? If not, it might be an argument to account in rmap_unmap to be
>>>> consistent for all?
>>>
>>> Again, see NR_FILE_THPS handling. No rmap over-complication please.
>>
>> ... not to mention that it is non-sensical to only count pageache folios that
>> are mapped to user space ;)
> 
> Yes, good point. I'll get back in my box. :)

Well, it was a valuable discussion!

anon folios in the swapcache are interesting: they are only "anon" after 
we first mapped them (harder to change, but would be possible by using a 
NULL mapping maybe, if really worth it; with memdesc that might turn out 
interesting). But once they are anon, they will stay anon until actually 
reclaimed -> freed.

-- 
Cheers,

David / dhildenb

Re: [PATCH RFC 1/2] mm: collect the number of anon large folios

[Barry Song]

On Fri, Aug 9, 2024 at 9:22 PM David Hildenbrand <david@redhat.com> wrote:
>
> On 09.08.24 11:05, Ryan Roberts wrote:
> > On 09/08/2024 09:58, David Hildenbrand wrote:
> >> On 09.08.24 10:42, David Hildenbrand wrote:
> >>>>> Not sure I fully understand why David prefers to do the unaccounting at
> >>>>> free-time though? It feels unbalanced to me to increment when first mapped but
> >>>>> decrement when freed. Surely its safer to either use alloc/free or use first
> >>>>> map/last map?
> >>>>>
> >>>>> If using alloc/free isn't there a THP constructor/destructor that prepares the
> >>>>> deferred list? (My memory may be failing me). Could we use that?
> >>>>
> >>>> Additionally, if we wanted to extend (eventually) to track the number of shmem
> >>>> and file mthps in additional counters, could we also account using similar folio
> >>>> free-time hooks? If not, it might be an argument to account in rmap_unmap to be
> >>>> consistent for all?
> >>>
> >>> Again, see NR_FILE_THPS handling. No rmap over-complication please.
> >>
> >> ... not to mention that it is non-sensical to only count pageache folios that
> >> are mapped to user space ;)
> >
> > Yes, good point. I'll get back in my box. :)
>
> Well, it was a valuable discussion!
>
> anon folios in the swapcache are interesting: they are only "anon" after
> we first mapped them (harder to change, but would be possible by using a
> NULL mapping maybe, if really worth it; with memdesc that might turn out
> interesting). But once they are anon, they will stay anon until actually
> reclaimed -> freed.

I assume we don’t need to worry about this, as even AnonPages (NR_ANON_MAPPED)
in /proc/meminfo also entirely depends on anon pages becoming anon mappings.

>
> --
> Cheers,
>
> David / dhildenb
>

Thanks
Barry

Re: [PATCH RFC 1/2] mm: collect the number of anon large folios

[David Hildenbrand]

On 09.08.24 10:13, Ryan Roberts wrote:
> On 08/08/2024 02:04, Barry Song wrote:
>> From: Barry Song <v-songbaohua@oppo.com>
>>
>> When a new anonymous mTHP is added to the rmap, we increase the count.
>> We reduce the count whenever an mTHP is completely unmapped.
>>
>> Signed-off-by: Barry Song <v-songbaohua@oppo.com>
>> ---
>>   Documentation/admin-guide/mm/transhuge.rst |  5 +++++
>>   include/linux/huge_mm.h                    | 15 +++++++++++++--
>>   mm/huge_memory.c                           |  2 ++
>>   mm/rmap.c                                  |  3 +++
>>   4 files changed, 23 insertions(+), 2 deletions(-)
>>
>> diff --git a/Documentation/admin-guide/mm/transhuge.rst b/Documentation/admin-guide/mm/transhuge.rst
>> index 058485daf186..715f181543f6 100644
>> --- a/Documentation/admin-guide/mm/transhuge.rst
>> +++ b/Documentation/admin-guide/mm/transhuge.rst
>> @@ -527,6 +527,11 @@ split_deferred
>>           it would free up some memory. Pages on split queue are going to
>>           be split under memory pressure, if splitting is possible.
>>   
>> +anon_num
>> +       the number of anon huge pages we have in the whole system.
>> +       These huge pages could be still entirely mapped and have partially
>> +       unmapped and unused subpages.
> 
> nit: "entirely mapped and have partially unmapped and unused subpages" ->
> "entirely mapped or have partially unmapped/unused subpages"
> 
>> +
>>   As the system ages, allocating huge pages may be expensive as the
>>   system uses memory compaction to copy data around memory to free a
>>   huge page for use. There are some counters in ``/proc/vmstat`` to help
>> diff --git a/include/linux/huge_mm.h b/include/linux/huge_mm.h
>> index e25d9ebfdf89..294c348fe3cc 100644
>> --- a/include/linux/huge_mm.h
>> +++ b/include/linux/huge_mm.h
>> @@ -281,6 +281,7 @@ enum mthp_stat_item {
>>   	MTHP_STAT_SPLIT,
>>   	MTHP_STAT_SPLIT_FAILED,
>>   	MTHP_STAT_SPLIT_DEFERRED,
>> +	MTHP_STAT_NR_ANON,
>>   	__MTHP_STAT_COUNT
>>   };
>>   
>> @@ -291,14 +292,24 @@ struct mthp_stat {
>>   #ifdef CONFIG_SYSFS
>>   DECLARE_PER_CPU(struct mthp_stat, mthp_stats);
>>   
>> -static inline void count_mthp_stat(int order, enum mthp_stat_item item)
>> +static inline void mod_mthp_stat(int order, enum mthp_stat_item item, int delta)
>>   {
>>   	if (order <= 0 || order > PMD_ORDER)
>>   		return;
>>   
>> -	this_cpu_inc(mthp_stats.stats[order][item]);
>> +	this_cpu_add(mthp_stats.stats[order][item], delta);
>> +}
>> +
>> +static inline void count_mthp_stat(int order, enum mthp_stat_item item)
>> +{
>> +	mod_mthp_stat(order, item, 1);
>>   }
>> +
>>   #else
>> +static inline void mod_mthp_stat(int order, enum mthp_stat_item item, int delta)
>> +{
>> +}
>> +
>>   static inline void count_mthp_stat(int order, enum mthp_stat_item item)
>>   {
>>   }
>> diff --git a/mm/huge_memory.c b/mm/huge_memory.c
>> index 697fcf89f975..b6bc2a3791e3 100644
>> --- a/mm/huge_memory.c
>> +++ b/mm/huge_memory.c
>> @@ -578,6 +578,7 @@ DEFINE_MTHP_STAT_ATTR(shmem_fallback_charge, MTHP_STAT_SHMEM_FALLBACK_CHARGE);
>>   DEFINE_MTHP_STAT_ATTR(split, MTHP_STAT_SPLIT);
>>   DEFINE_MTHP_STAT_ATTR(split_failed, MTHP_STAT_SPLIT_FAILED);
>>   DEFINE_MTHP_STAT_ATTR(split_deferred, MTHP_STAT_SPLIT_DEFERRED);
>> +DEFINE_MTHP_STAT_ATTR(anon_num, MTHP_STAT_NR_ANON);
>>   
>>   static struct attribute *stats_attrs[] = {
>>   	&anon_fault_alloc_attr.attr,
>> @@ -591,6 +592,7 @@ static struct attribute *stats_attrs[] = {
>>   	&split_attr.attr,
>>   	&split_failed_attr.attr,
>>   	&split_deferred_attr.attr,
>> +	&anon_num_attr.attr,
>>   	NULL,
>>   };
>>   
>> diff --git a/mm/rmap.c b/mm/rmap.c
>> index 901950200957..2b722f26224c 100644
>> --- a/mm/rmap.c
>> +++ b/mm/rmap.c
>> @@ -1467,6 +1467,7 @@ void folio_add_new_anon_rmap(struct folio *folio, struct vm_area_struct *vma,
>>   	}
>>   
>>   	__folio_mod_stat(folio, nr, nr_pmdmapped);
>> +	mod_mthp_stat(folio_order(folio), MTHP_STAT_NR_ANON, 1);
>>   }
>>   
>>   static __always_inline void __folio_add_file_rmap(struct folio *folio,
>> @@ -1582,6 +1583,8 @@ static __always_inline void __folio_remove_rmap(struct folio *folio,
>>   	    list_empty(&folio->_deferred_list))
>>   		deferred_split_folio(folio);
>>   	__folio_mod_stat(folio, -nr, -nr_pmdmapped);
>> +	if (folio_test_anon(folio) && !atomic_read(mapped))
> 
> Agree that atomic_read() is dodgy here.
> 
> Not sure I fully understand why David prefers to do the unaccounting at
> free-time though? It feels unbalanced to me to increment when first mapped but
> decrement when freed. Surely its safer to either use alloc/free or use first
> map/last map?

Doing it when we set/clear folio->mapping is straight forward.

Anon folios currently come to live when we first map them, and they stay 
that way until we free them.

In the future, we'll have to move that anon handling further out, when 
if have to allocate anon-specific memdesc ahead of time, then, it will 
be clued to that lifetime.

> 
> If using alloc/free isn't there a THP constructor/destructor that prepares the
> deferred list? (My memory may be failing me). Could we use that?

Likely the deconstructor could work as well. Not sure if that is any 
better than the freeing path where folio->mapping currently gets cleared.

The generic constructor certainly won't work right now. That's not where 
the "anon" part comes to live.

Let's take a look how NR_FILE_THPS is handled:

__filemap_add_folio() increments it -- when we set folio->mapping
__filemap_remove_folio() (->filemap_unaccount_folio) decrements it -- 
after which we usually call page_cache_delete() to set folio->mapping = 
NULL;

-- 
Cheers,

David / dhildenb

Re: [PATCH RFC 1/2] mm: collect the number of anon large folios

[Ryan Roberts]

On 09/08/2024 09:39, David Hildenbrand wrote:
> On 09.08.24 10:13, Ryan Roberts wrote:
>> On 08/08/2024 02:04, Barry Song wrote:
>>> From: Barry Song <v-songbaohua@oppo.com>
>>>
>>> When a new anonymous mTHP is added to the rmap, we increase the count.
>>> We reduce the count whenever an mTHP is completely unmapped.
>>>
>>> Signed-off-by: Barry Song <v-songbaohua@oppo.com>
>>> ---
>>>   Documentation/admin-guide/mm/transhuge.rst |  5 +++++
>>>   include/linux/huge_mm.h                    | 15 +++++++++++++--
>>>   mm/huge_memory.c                           |  2 ++
>>>   mm/rmap.c                                  |  3 +++
>>>   4 files changed, 23 insertions(+), 2 deletions(-)
>>>
>>> diff --git a/Documentation/admin-guide/mm/transhuge.rst
>>> b/Documentation/admin-guide/mm/transhuge.rst
>>> index 058485daf186..715f181543f6 100644
>>> --- a/Documentation/admin-guide/mm/transhuge.rst
>>> +++ b/Documentation/admin-guide/mm/transhuge.rst
>>> @@ -527,6 +527,11 @@ split_deferred
>>>           it would free up some memory. Pages on split queue are going to
>>>           be split under memory pressure, if splitting is possible.
>>>   +anon_num
>>> +       the number of anon huge pages we have in the whole system.
>>> +       These huge pages could be still entirely mapped and have partially
>>> +       unmapped and unused subpages.
>>
>> nit: "entirely mapped and have partially unmapped and unused subpages" ->
>> "entirely mapped or have partially unmapped/unused subpages"
>>
>>> +
>>>   As the system ages, allocating huge pages may be expensive as the
>>>   system uses memory compaction to copy data around memory to free a
>>>   huge page for use. There are some counters in ``/proc/vmstat`` to help
>>> diff --git a/include/linux/huge_mm.h b/include/linux/huge_mm.h
>>> index e25d9ebfdf89..294c348fe3cc 100644
>>> --- a/include/linux/huge_mm.h
>>> +++ b/include/linux/huge_mm.h
>>> @@ -281,6 +281,7 @@ enum mthp_stat_item {
>>>       MTHP_STAT_SPLIT,
>>>       MTHP_STAT_SPLIT_FAILED,
>>>       MTHP_STAT_SPLIT_DEFERRED,
>>> +    MTHP_STAT_NR_ANON,
>>>       __MTHP_STAT_COUNT
>>>   };
>>>   @@ -291,14 +292,24 @@ struct mthp_stat {
>>>   #ifdef CONFIG_SYSFS
>>>   DECLARE_PER_CPU(struct mthp_stat, mthp_stats);
>>>   -static inline void count_mthp_stat(int order, enum mthp_stat_item item)
>>> +static inline void mod_mthp_stat(int order, enum mthp_stat_item item, int
>>> delta)
>>>   {
>>>       if (order <= 0 || order > PMD_ORDER)
>>>           return;
>>>   -    this_cpu_inc(mthp_stats.stats[order][item]);
>>> +    this_cpu_add(mthp_stats.stats[order][item], delta);
>>> +}
>>> +
>>> +static inline void count_mthp_stat(int order, enum mthp_stat_item item)
>>> +{
>>> +    mod_mthp_stat(order, item, 1);
>>>   }
>>> +
>>>   #else
>>> +static inline void mod_mthp_stat(int order, enum mthp_stat_item item, int
>>> delta)
>>> +{
>>> +}
>>> +
>>>   static inline void count_mthp_stat(int order, enum mthp_stat_item item)
>>>   {
>>>   }
>>> diff --git a/mm/huge_memory.c b/mm/huge_memory.c
>>> index 697fcf89f975..b6bc2a3791e3 100644
>>> --- a/mm/huge_memory.c
>>> +++ b/mm/huge_memory.c
>>> @@ -578,6 +578,7 @@ DEFINE_MTHP_STAT_ATTR(shmem_fallback_charge,
>>> MTHP_STAT_SHMEM_FALLBACK_CHARGE);
>>>   DEFINE_MTHP_STAT_ATTR(split, MTHP_STAT_SPLIT);
>>>   DEFINE_MTHP_STAT_ATTR(split_failed, MTHP_STAT_SPLIT_FAILED);
>>>   DEFINE_MTHP_STAT_ATTR(split_deferred, MTHP_STAT_SPLIT_DEFERRED);
>>> +DEFINE_MTHP_STAT_ATTR(anon_num, MTHP_STAT_NR_ANON);
>>>     static struct attribute *stats_attrs[] = {
>>>       &anon_fault_alloc_attr.attr,
>>> @@ -591,6 +592,7 @@ static struct attribute *stats_attrs[] = {
>>>       &split_attr.attr,
>>>       &split_failed_attr.attr,
>>>       &split_deferred_attr.attr,
>>> +    &anon_num_attr.attr,
>>>       NULL,
>>>   };
>>>   diff --git a/mm/rmap.c b/mm/rmap.c
>>> index 901950200957..2b722f26224c 100644
>>> --- a/mm/rmap.c
>>> +++ b/mm/rmap.c
>>> @@ -1467,6 +1467,7 @@ void folio_add_new_anon_rmap(struct folio *folio,
>>> struct vm_area_struct *vma,
>>>       }
>>>         __folio_mod_stat(folio, nr, nr_pmdmapped);
>>> +    mod_mthp_stat(folio_order(folio), MTHP_STAT_NR_ANON, 1);
>>>   }
>>>     static __always_inline void __folio_add_file_rmap(struct folio *folio,
>>> @@ -1582,6 +1583,8 @@ static __always_inline void __folio_remove_rmap(struct
>>> folio *folio,
>>>           list_empty(&folio->_deferred_list))
>>>           deferred_split_folio(folio);
>>>       __folio_mod_stat(folio, -nr, -nr_pmdmapped);
>>> +    if (folio_test_anon(folio) && !atomic_read(mapped))
>>
>> Agree that atomic_read() is dodgy here.
>>
>> Not sure I fully understand why David prefers to do the unaccounting at
>> free-time though? It feels unbalanced to me to increment when first mapped but
>> decrement when freed. Surely its safer to either use alloc/free or use first
>> map/last map?
> 
> Doing it when we set/clear folio->mapping is straight forward.
> 
> Anon folios currently come to live when we first map them, and they stay that
> way until we free them.
> 
> In the future, we'll have to move that anon handling further out, when if have
> to allocate anon-specific memdesc ahead of time, then, it will be clued to that
> lifetime.
> 
>>
>> If using alloc/free isn't there a THP constructor/destructor that prepares the
>> deferred list? (My memory may be failing me). Could we use that?
> 
> Likely the deconstructor could work as well. Not sure if that is any better than
> the freeing path where folio->mapping currently gets cleared.
> 
> The generic constructor certainly won't work right now. That's not where the
> "anon" part comes to live.
> 
> Let's take a look how NR_FILE_THPS is handled:
> 
> __filemap_add_folio() increments it -- when we set folio->mapping
> __filemap_remove_folio() (->filemap_unaccount_folio) decrements it -- after
> which we usually call page_cache_delete() to set folio->mapping = NULL;
> 

OK got it, thanks!