diff mbox series

[v7,4/4] mm/vmscan: avoid split lazyfree THP during shrink_folio_list()

Message ID 20240610120809.66601-1-ioworker0@gmail.com (mailing list archive)
State New
Headers show
Series Reclaim lazyfree THP without splitting | expand

Commit Message

Lance Yang June 10, 2024, 12:08 p.m. UTC 
When the user no longer requires the pages, they would use
madvise(MADV_FREE) to mark the pages as lazy free. Subsequently, they
typically would not re-write to that memory again.

During memory reclaim, if we detect that the large folio and its PMD are
both still marked as clean and there are no unexpected references
(such as GUP), so we can just discard the memory lazily, improving the
efficiency of memory reclamation in this case.

On an Intel i5 CPU, reclaiming 1GiB of lazyfree THPs using
mem_cgroup_force_empty() results in the following runtimes in seconds
(shorter is better):

--------------------------------------------
|     Old       |      New       |  Change  |
--------------------------------------------
|   0.683426    |    0.049197    |  -92.80% |
--------------------------------------------

Suggested-by: Zi Yan <ziy@nvidia.com>
Suggested-by: David Hildenbrand <david@redhat.com>
Signed-off-by: Lance Yang <ioworker0@gmail.com>
---
 include/linux/huge_mm.h |  9 +++++
 mm/huge_memory.c        | 80 +++++++++++++++++++++++++++++++++++++++++
 mm/rmap.c               | 36 +++++++++++++------
 3 files changed, 114 insertions(+), 11 deletions(-)

Comments

Lance Yang June 13, 2024, 6:55 a.m. UTC | #1 
On Mon, Jun 10, 2024 at 8:08 PM Lance Yang <ioworker0@gmail.com> wrote:
>
[...]
> +static bool __discard_anon_folio_pmd_locked(struct vm_area_struct *vma,
> +                                           unsigned long addr, pmd_t *pmdp,
> +                                           struct folio *folio)
> +{
> +       VM_WARN_ON_FOLIO(folio_test_swapbacked(folio), folio);
> +       VM_WARN_ON_FOLIO(!folio_test_anon(folio), folio);
> +
> +       struct mm_struct *mm = vma->vm_mm;
> +       int ref_count, map_count;
> +       pmd_t orig_pmd = *pmdp;
> +       struct page *page;
> +
> +       if (unlikely(!pmd_present(orig_pmd) || !pmd_trans_huge(orig_pmd)))
> +               return false;
> +
> +       page = pmd_page(orig_pmd);
> +       if (unlikely(page_folio(page) != folio))
> +               return false;
> +
> +       if (folio_test_dirty(folio) || pmd_dirty(orig_pmd)) {
> +               folio_set_swapbacked(folio);
> +               return false;
> +       }
> +
> +       orig_pmd = pmdp_huge_clear_flush(vma, addr, pmdp);
> +
> +       /*
> +        * Syncing against concurrent GUP-fast:
> +        * - clear PMD; barrier; read refcount
> +        * - inc refcount; barrier; read PMD
> +        */
> +       smp_mb();
> +
> +       ref_count = folio_ref_count(folio);
> +       map_count = folio_mapcount(folio);
> +
> +       /*
> +        * Order reads for folio refcount and dirty flag
> +        * (see comments in __remove_mapping()).
> +        */
> +       smp_rmb();
> +
> +       /*
> +        * If the folio or its PMD is redirtied at this point, or if there
> +        * are unexpected references, we will give up to discard this folio
> +        * and remap it.
> +        *
> +        * The only folio refs must be one from isolation plus the rmap(s).
> +        */
> +       if (folio_test_dirty(folio) || pmd_dirty(orig_pmd))
> +               folio_set_swapbacked(folio);
> +
> +       if (folio_test_swapbacked(folio) || ref_count != map_count + 1) {
> +               set_pmd_at(mm, addr, pmdp, orig_pmd);
> +               return false;
> +       }
> +
> +       folio_remove_rmap_pmd(folio, page, vma);
> +       zap_deposited_table(mm, pmdp);
> +       add_mm_counter(mm, MM_ANONPAGES, -HPAGE_PMD_NR);
> +       if (vma->vm_flags & VM_LOCKED)
> +               mlock_drain_local();
> +       folio_put(folio);
> +
> +       return true;
> +}
[...]
> diff --git a/mm/rmap.c b/mm/rmap.c
> index b77f88695588..8e901636ade9 100644
> --- a/mm/rmap.c
> +++ b/mm/rmap.c
> @@ -1630,6 +1630,7 @@ static bool try_to_unmap_one(struct folio *folio, struct vm_area_struct *vma,
>         enum ttu_flags flags = (enum ttu_flags)(long)arg;
>         unsigned long pfn;
>         unsigned long hsz = 0;
> +       bool pmd_mapped = false;
>
>         /*
>          * When racing against e.g. zap_pte_range() on another cpu,
> @@ -1676,16 +1677,24 @@ static bool try_to_unmap_one(struct folio *folio, struct vm_area_struct *vma,
>                         goto walk_done_err;
>                 }
>
> -               if (!pvmw.pte && (flags & TTU_SPLIT_HUGE_PMD)) {
> -                       /*
> -                        * We temporarily have to drop the PTL and start once
> -                        * again from that now-PTE-mapped page table.
> -                        */
> -                       split_huge_pmd_locked(vma, pvmw.address, pvmw.pmd,
> -                                             false, folio);
> -                       flags &= ~TTU_SPLIT_HUGE_PMD;
> -                       page_vma_mapped_walk_restart(&pvmw);
> -                       continue;
> +               if (!pvmw.pte) {
> +                       pmd_mapped = true;
> +                       if (unmap_huge_pmd_locked(vma, pvmw.address, pvmw.pmd,
> +                                                 folio))
> +                               goto walk_done;
> +
> +                       if (flags & TTU_SPLIT_HUGE_PMD) {
> +                               /*
> +                                * We temporarily have to drop the PTL and start
> +                                * once again from that now-PTE-mapped page
> +                                * table.
> +                                */
> +                               split_huge_pmd_locked(vma, pvmw.address,
> +                                                     pvmw.pmd, false, folio);
> +                               flags &= ~TTU_SPLIT_HUGE_PMD;
> +                               page_vma_mapped_walk_restart(&pvmw);
> +                               continue;
> +                       }
>                 }
>
>                 /* Unexpected PMD-mapped THP? */
> @@ -1813,7 +1822,12 @@ static bool try_to_unmap_one(struct folio *folio, struct vm_area_struct *vma,
>                          */
>                         if (unlikely(folio_test_swapbacked(folio) !=
>                                         folio_test_swapcache(folio))) {
> -                               WARN_ON_ONCE(1);
> +                               /*
> +                                * unmap_huge_pmd_locked() will unmark a
> +                                * PMD-mapped folio as lazyfree if the folio or
> +                                * its PMD was redirtied.
> +                                */
> +                               WARN_ON_ONCE(!pmd_mapped);

Damn it. I forgot to remap the folio to the page table since we've cleared
the PTE here.

But it seems like we neither need to do that, nor unmark a PMD-mapped folio
as lazyfree in unmap_huge_pmd_locked(); the next block[1] will catch the
folio and do the same thing.

Hi David and Baolin, what do you think?

>                                 goto walk_done_err;
>                         }

[1] https://elixir.bootlin.com/linux/v6.10-rc3/source/mm/rmap.c#L1820

/* MADV_FREE page check */
if (!folio_test_swapbacked(folio)) {
	int ref_count, map_count;

[...]
	/*
	 * The only page refs must be one from isolation
	 * plus the rmap(s) (dropped by discard:).
	 */
	if (ref_count == 1 + map_count &&
	    !folio_test_dirty(folio)) {
		dec_mm_counter(mm, MM_ANONPAGES);
		goto discard;
	}

	/*
	 * If the folio was redirtied, it cannot be
	 * discarded. Remap the page to page table.
	 */
	set_pte_at(mm, address, pvmw.pte, pteval);
	folio_set_swapbacked(folio);
	goto walk_done_err;
}


Thanks,
Lance

>
> --
> 2.33.1
>
Lance Yang June 13, 2024, 7:28 a.m. UTC | #2 
Hi Andrew,

I'd like to fix the bug[1] I mentioned previously. Could you please fold the
following changes into this patch? 

[1] https://lore.kernel.org/linux-mm/20240613065521.15960-1-ioworker0@gmail.com/

diff --git a/mm/huge_memory.c b/mm/huge_memory.c
index f409ea9fcc18..425374ae06ed 100644
--- a/mm/huge_memory.c
+++ b/mm/huge_memory.c
@@ -2707,10 +2707,8 @@ static bool __discard_anon_folio_pmd_locked(struct vm_area_struct *vma,
 	if (unlikely(page_folio(page) != folio))
 		return false;
 
-	if (folio_test_dirty(folio) || pmd_dirty(orig_pmd)) {
-		folio_set_swapbacked(folio);
+	if (folio_test_dirty(folio) || pmd_dirty(orig_pmd))
 		return false;
-	}
 
 	orig_pmd = pmdp_huge_clear_flush(vma, addr, pmdp);
 
@@ -2737,10 +2735,8 @@ static bool __discard_anon_folio_pmd_locked(struct vm_area_struct *vma,
 	 *
 	 * The only folio refs must be one from isolation plus the rmap(s).
 	 */
-	if (folio_test_dirty(folio) || pmd_dirty(orig_pmd))
-		folio_set_swapbacked(folio);
-
-	if (folio_test_swapbacked(folio) || ref_count != map_count + 1) {
+	if (folio_test_dirty(folio) || pmd_dirty(orig_pmd) ||
+	    ref_count != map_count + 1) {
 		set_pmd_at(mm, addr, pmdp, orig_pmd);
 		return false;
 	}
diff --git a/mm/rmap.c b/mm/rmap.c
index b9e5943c8349..393e2c11c44c 100644
--- a/mm/rmap.c
+++ b/mm/rmap.c
@@ -1824,12 +1824,7 @@ static bool try_to_unmap_one(struct folio *folio, struct vm_area_struct *vma,
 			 */
 			if (unlikely(folio_test_swapbacked(folio) !=
 					folio_test_swapcache(folio))) {
-				/*
-				 * unmap_huge_pmd_locked() will unmark a
-				 * PMD-mapped folio as lazyfree if the folio or
-				 * its PMD was redirtied.
-				 */
-				WARN_ON_ONCE(!pmd_mapped);
+				WARN_ON_ONCE(1);
 				goto walk_done_err;
 			}
Baolin Wang June 13, 2024, 8:20 a.m. UTC | #3 
On 2024/6/13 15:28, Lance Yang wrote:
> Hi Andrew,
> 
> I'd like to fix the bug[1] I mentioned previously. Could you please fold the
> following changes into this patch?
> 
> [1] https://lore.kernel.org/linux-mm/20240613065521.15960-1-ioworker0@gmail.com/
> 
> diff --git a/mm/huge_memory.c b/mm/huge_memory.c
> index f409ea9fcc18..425374ae06ed 100644
> --- a/mm/huge_memory.c
> +++ b/mm/huge_memory.c
> @@ -2707,10 +2707,8 @@ static bool __discard_anon_folio_pmd_locked(struct vm_area_struct *vma,
>   	if (unlikely(page_folio(page) != folio))
>   		return false;
>   
> -	if (folio_test_dirty(folio) || pmd_dirty(orig_pmd)) {
> -		folio_set_swapbacked(folio);
> +	if (folio_test_dirty(folio) || pmd_dirty(orig_pmd))
>   		return false;
> -	}
>   
>   	orig_pmd = pmdp_huge_clear_flush(vma, addr, pmdp);
>   
> @@ -2737,10 +2735,8 @@ static bool __discard_anon_folio_pmd_locked(struct vm_area_struct *vma,
>   	 *
>   	 * The only folio refs must be one from isolation plus the rmap(s).
>   	 */
> -	if (folio_test_dirty(folio) || pmd_dirty(orig_pmd))
> -		folio_set_swapbacked(folio);
> -
> -	if (folio_test_swapbacked(folio) || ref_count != map_count + 1) {
> +	if (folio_test_dirty(folio) || pmd_dirty(orig_pmd) ||
> +	    ref_count != map_count + 1) {
>   		set_pmd_at(mm, addr, pmdp, orig_pmd);
>   		return false;
>   	}
> diff --git a/mm/rmap.c b/mm/rmap.c
> index b9e5943c8349..393e2c11c44c 100644
> --- a/mm/rmap.c
> +++ b/mm/rmap.c
> @@ -1824,12 +1824,7 @@ static bool try_to_unmap_one(struct folio *folio, struct vm_area_struct *vma,
>   			 */
>   			if (unlikely(folio_test_swapbacked(folio) !=
>   					folio_test_swapcache(folio))) {
> -				/*
> -				 * unmap_huge_pmd_locked() will unmark a
> -				 * PMD-mapped folio as lazyfree if the folio or
> -				 * its PMD was redirtied.
> -				 */
> -				WARN_ON_ONCE(!pmd_mapped);
> +				WARN_ON_ONCE(1);
>   				goto walk_done_err;
>   			}
>   

You can also drop the unused 'pmd_mapped' variable now.
Lance Yang June 13, 2024, 8:56 a.m. UTC | #4 
On Thu, Jun 13, 2024 at 4:20 PM Baolin Wang
<baolin.wang@linux.alibaba.com> wrote:
>
>
>
> On 2024/6/13 15:28, Lance Yang wrote:
> > Hi Andrew,
> >
> > I'd like to fix the bug[1] I mentioned previously. Could you please fold the
> > following changes into this patch?
> >
> > [1] https://lore.kernel.org/linux-mm/20240613065521.15960-1-ioworker0@gmail.com/
> >
> > diff --git a/mm/huge_memory.c b/mm/huge_memory.c
> > index f409ea9fcc18..425374ae06ed 100644
> > --- a/mm/huge_memory.c
> > +++ b/mm/huge_memory.c
> > @@ -2707,10 +2707,8 @@ static bool __discard_anon_folio_pmd_locked(struct vm_area_struct *vma,
> >       if (unlikely(page_folio(page) != folio))
> >               return false;
> >
> > -     if (folio_test_dirty(folio) || pmd_dirty(orig_pmd)) {
> > -             folio_set_swapbacked(folio);
> > +     if (folio_test_dirty(folio) || pmd_dirty(orig_pmd))
> >               return false;
> > -     }
> >
> >       orig_pmd = pmdp_huge_clear_flush(vma, addr, pmdp);
> >
> > @@ -2737,10 +2735,8 @@ static bool __discard_anon_folio_pmd_locked(struct vm_area_struct *vma,
> >        *
> >        * The only folio refs must be one from isolation plus the rmap(s).
> >        */
> > -     if (folio_test_dirty(folio) || pmd_dirty(orig_pmd))
> > -             folio_set_swapbacked(folio);
> > -
> > -     if (folio_test_swapbacked(folio) || ref_count != map_count + 1) {
> > +     if (folio_test_dirty(folio) || pmd_dirty(orig_pmd) ||
> > +         ref_count != map_count + 1) {
> >               set_pmd_at(mm, addr, pmdp, orig_pmd);
> >               return false;
> >       }
> > diff --git a/mm/rmap.c b/mm/rmap.c
> > index b9e5943c8349..393e2c11c44c 100644
> > --- a/mm/rmap.c
> > +++ b/mm/rmap.c
> > @@ -1824,12 +1824,7 @@ static bool try_to_unmap_one(struct folio *folio, struct vm_area_struct *vma,
> >                        */
> >                       if (unlikely(folio_test_swapbacked(folio) !=
> >                                       folio_test_swapcache(folio))) {
> > -                             /*
> > -                              * unmap_huge_pmd_locked() will unmark a
> > -                              * PMD-mapped folio as lazyfree if the folio or
> > -                              * its PMD was redirtied.
> > -                              */
> > -                             WARN_ON_ONCE(!pmd_mapped);
> > +                             WARN_ON_ONCE(1);
> >                               goto walk_done_err;
> >                       }
> >
>
> You can also drop the unused 'pmd_mapped' variable now.

Good catch!

Will drop it in the next version.

Thanks,
Lance
diff mbox series

Patch

diff --git a/include/linux/huge_mm.h b/include/linux/huge_mm.h
index 4670c6ee118b..020e2344eb86 100644
--- a/include/linux/huge_mm.h
+++ b/include/linux/huge_mm.h
@@ -417,6 +417,8 @@  static inline bool thp_migration_supported(void)
 
 void split_huge_pmd_locked(struct vm_area_struct *vma, unsigned long address,
 			   pmd_t *pmd, bool freeze, struct folio *folio);
+bool unmap_huge_pmd_locked(struct vm_area_struct *vma, unsigned long addr,
+			   pmd_t *pmdp, struct folio *folio);
 
 #else /* CONFIG_TRANSPARENT_HUGEPAGE */
 
@@ -484,6 +486,13 @@  static inline void split_huge_pmd_locked(struct vm_area_struct *vma,
 					 unsigned long address, pmd_t *pmd,
 					 bool freeze, struct folio *folio) {}
 
+static inline bool unmap_huge_pmd_locked(struct vm_area_struct *vma,
+					 unsigned long addr, pmd_t *pmdp,
+					 struct folio *folio)
+{
+	return false;
+}
+
 #define split_huge_pud(__vma, __pmd, __address)	\
 	do { } while (0)
 
diff --git a/mm/huge_memory.c b/mm/huge_memory.c
index d2697cc8f9d4..19592d3f1167 100644
--- a/mm/huge_memory.c
+++ b/mm/huge_memory.c
@@ -2687,6 +2687,86 @@  static void unmap_folio(struct folio *folio)
 	try_to_unmap_flush();
 }
 
+static bool __discard_anon_folio_pmd_locked(struct vm_area_struct *vma,
+					    unsigned long addr, pmd_t *pmdp,
+					    struct folio *folio)
+{
+	VM_WARN_ON_FOLIO(folio_test_swapbacked(folio), folio);
+	VM_WARN_ON_FOLIO(!folio_test_anon(folio), folio);
+
+	struct mm_struct *mm = vma->vm_mm;
+	int ref_count, map_count;
+	pmd_t orig_pmd = *pmdp;
+	struct page *page;
+
+	if (unlikely(!pmd_present(orig_pmd) || !pmd_trans_huge(orig_pmd)))
+		return false;
+
+	page = pmd_page(orig_pmd);
+	if (unlikely(page_folio(page) != folio))
+		return false;
+
+	if (folio_test_dirty(folio) || pmd_dirty(orig_pmd)) {
+		folio_set_swapbacked(folio);
+		return false;
+	}
+
+	orig_pmd = pmdp_huge_clear_flush(vma, addr, pmdp);
+
+	/*
+	 * Syncing against concurrent GUP-fast:
+	 * - clear PMD; barrier; read refcount
+	 * - inc refcount; barrier; read PMD
+	 */
+	smp_mb();
+
+	ref_count = folio_ref_count(folio);
+	map_count = folio_mapcount(folio);
+
+	/*
+	 * Order reads for folio refcount and dirty flag
+	 * (see comments in __remove_mapping()).
+	 */
+	smp_rmb();
+
+	/*
+	 * If the folio or its PMD is redirtied at this point, or if there
+	 * are unexpected references, we will give up to discard this folio
+	 * and remap it.
+	 *
+	 * The only folio refs must be one from isolation plus the rmap(s).
+	 */
+	if (folio_test_dirty(folio) || pmd_dirty(orig_pmd))
+		folio_set_swapbacked(folio);
+
+	if (folio_test_swapbacked(folio) || ref_count != map_count + 1) {
+		set_pmd_at(mm, addr, pmdp, orig_pmd);
+		return false;
+	}
+
+	folio_remove_rmap_pmd(folio, page, vma);
+	zap_deposited_table(mm, pmdp);
+	add_mm_counter(mm, MM_ANONPAGES, -HPAGE_PMD_NR);
+	if (vma->vm_flags & VM_LOCKED)
+		mlock_drain_local();
+	folio_put(folio);
+
+	return true;
+}
+
+bool unmap_huge_pmd_locked(struct vm_area_struct *vma, unsigned long addr,
+			   pmd_t *pmdp, struct folio *folio)
+{
+	VM_WARN_ON_FOLIO(!folio_test_pmd_mappable(folio), folio);
+	VM_WARN_ON_FOLIO(!folio_test_locked(folio), folio);
+	VM_WARN_ON_ONCE(!IS_ALIGNED(addr, HPAGE_PMD_SIZE));
+
+	if (folio_test_anon(folio) && !folio_test_swapbacked(folio))
+		return __discard_anon_folio_pmd_locked(vma, addr, pmdp, folio);
+
+	return false;
+}
+
 static void remap_page(struct folio *folio, unsigned long nr)
 {
 	int i = 0;
diff --git a/mm/rmap.c b/mm/rmap.c
index b77f88695588..8e901636ade9 100644
--- a/mm/rmap.c
+++ b/mm/rmap.c
@@ -1630,6 +1630,7 @@  static bool try_to_unmap_one(struct folio *folio, struct vm_area_struct *vma,
 	enum ttu_flags flags = (enum ttu_flags)(long)arg;
 	unsigned long pfn;
 	unsigned long hsz = 0;
+	bool pmd_mapped = false;
 
 	/*
 	 * When racing against e.g. zap_pte_range() on another cpu,
@@ -1676,16 +1677,24 @@  static bool try_to_unmap_one(struct folio *folio, struct vm_area_struct *vma,
 			goto walk_done_err;
 		}
 
-		if (!pvmw.pte && (flags & TTU_SPLIT_HUGE_PMD)) {
-			/*
-			 * We temporarily have to drop the PTL and start once
-			 * again from that now-PTE-mapped page table.
-			 */
-			split_huge_pmd_locked(vma, pvmw.address, pvmw.pmd,
-					      false, folio);
-			flags &= ~TTU_SPLIT_HUGE_PMD;
-			page_vma_mapped_walk_restart(&pvmw);
-			continue;
+		if (!pvmw.pte) {
+			pmd_mapped = true;
+			if (unmap_huge_pmd_locked(vma, pvmw.address, pvmw.pmd,
+						  folio))
+				goto walk_done;
+
+			if (flags & TTU_SPLIT_HUGE_PMD) {
+				/*
+				 * We temporarily have to drop the PTL and start
+				 * once again from that now-PTE-mapped page
+				 * table.
+				 */
+				split_huge_pmd_locked(vma, pvmw.address,
+						      pvmw.pmd, false, folio);
+				flags &= ~TTU_SPLIT_HUGE_PMD;
+				page_vma_mapped_walk_restart(&pvmw);
+				continue;
+			}
 		}
 
 		/* Unexpected PMD-mapped THP? */
@@ -1813,7 +1822,12 @@  static bool try_to_unmap_one(struct folio *folio, struct vm_area_struct *vma,
 			 */
 			if (unlikely(folio_test_swapbacked(folio) !=
 					folio_test_swapcache(folio))) {
-				WARN_ON_ONCE(1);
+				/*
+				 * unmap_huge_pmd_locked() will unmark a
+				 * PMD-mapped folio as lazyfree if the folio or
+				 * its PMD was redirtied.
+				 */
+				WARN_ON_ONCE(!pmd_mapped);
 				goto walk_done_err;
 			}