The buddy allocator coalesces compatible blocks during freeing, but it
doesn't update the types of the subblocks to match. When an allocation
later breaks the chunk down again, its pieces will be put on freelists
of the wrong type. This encourages incompatible page mixing (ask for
one type, get another), and thus long-term fragmentation.

Update the subblocks when merging a larger chunk, such that a later
expand() will maintain freelist type hygiene.

v2:
- remove spurious change_pageblock_range() move (Zi Yan)

Signed-off-by: Johannes Weiner <hannes@cmpxchg.org>
---
 mm/page_alloc.c | 15 +++++++++++----
 1 file changed, 11 insertions(+), 4 deletions(-)

diff --git a/mm/page_alloc.c b/mm/page_alloc.c
index e3f1c777feed..3db405414174 100644
--- a/mm/page_alloc.c
+++ b/mm/page_alloc.c
@@ -783,10 +783,17 @@ static inline void __free_one_page(struct page *page,
 			 */
 			int buddy_mt = get_pfnblock_migratetype(buddy, buddy_pfn);
 
-			if (migratetype != buddy_mt
-					&& (!migratetype_is_mergeable(migratetype) ||
-						!migratetype_is_mergeable(buddy_mt)))
-				goto done_merging;
+			if (migratetype != buddy_mt) {
+				if (!migratetype_is_mergeable(migratetype) ||
+				    !migratetype_is_mergeable(buddy_mt))
+					goto done_merging;
+				/*
+				 * Match buddy type. This ensures that
+				 * an expand() down the line puts the
+				 * sub-blocks on the right freelists.
+				 */
+				set_pageblock_migratetype(buddy, migratetype);
+			}
 		}
 
 		/*
-- 
2.42.0

On Mon, Sep 11, 2023 at 03:41:43PM -0400, Johannes Weiner wrote:
> The buddy allocator coalesces compatible blocks during freeing, but it
> doesn't update the types of the subblocks to match. When an allocation
> later breaks the chunk down again, its pieces will be put on freelists
> of the wrong type. This encourages incompatible page mixing (ask for
> one type, get another), and thus long-term fragmentation.
> 
> Update the subblocks when merging a larger chunk, such that a later
> expand() will maintain freelist type hygiene.
> 
> v2:
> - remove spurious change_pageblock_range() move (Zi Yan)
> 
> Signed-off-by: Johannes Weiner <hannes@cmpxchg.org>

I'm not 100% convinced on the amount of harm this causes but given that
it's a relatively rare condition, I didn't think about the consequences
too deeply. The patch certainly has merit so;

Acked-by: Mel Gorman <mgorman@techsingularity.net>

-- 
Mel Gorman
SUSE Labs

On 9/11/23 21:41, Johannes Weiner wrote:
> The buddy allocator coalesces compatible blocks during freeing, but it
> doesn't update the types of the subblocks to match. When an allocation
> later breaks the chunk down again, its pieces will be put on freelists
> of the wrong type. This encourages incompatible page mixing (ask for
> one type, get another), and thus long-term fragmentation.

Yeah why not. Sould be pretty rare as this only affects >=pageblock_order,
but then also the overhead in the otherwise hot function is limited to its
colder part.

> Update the subblocks when merging a larger chunk, such that a later
> expand() will maintain freelist type hygiene.
> 
> v2:
> - remove spurious change_pageblock_range() move (Zi Yan)
> 
> Signed-off-by: Johannes Weiner <hannes@cmpxchg.org>

Reviewed-by: Vlastimil Babka <vbabka@suse.cz>

> ---
>  mm/page_alloc.c | 15 +++++++++++----
>  1 file changed, 11 insertions(+), 4 deletions(-)
> 
> diff --git a/mm/page_alloc.c b/mm/page_alloc.c
> index e3f1c777feed..3db405414174 100644
> --- a/mm/page_alloc.c
> +++ b/mm/page_alloc.c
> @@ -783,10 +783,17 @@ static inline void __free_one_page(struct page *page,
>  			 */
>  			int buddy_mt = get_pfnblock_migratetype(buddy, buddy_pfn);
>  
> -			if (migratetype != buddy_mt
> -					&& (!migratetype_is_mergeable(migratetype) ||
> -						!migratetype_is_mergeable(buddy_mt)))
> -				goto done_merging;
> +			if (migratetype != buddy_mt) {
> +				if (!migratetype_is_mergeable(migratetype) ||
> +				    !migratetype_is_mergeable(buddy_mt))
> +					goto done_merging;
> +				/*
> +				 * Match buddy type. This ensures that
> +				 * an expand() down the line puts the
> +				 * sub-blocks on the right freelists.
> +				 */
> +				set_pageblock_migratetype(buddy, migratetype);
> +			}
>  		}
>  
>  		/*

On 11 Sep 2023, at 15:41, Johannes Weiner wrote:

> The buddy allocator coalesces compatible blocks during freeing, but it
> doesn't update the types of the subblocks to match. When an allocation
> later breaks the chunk down again, its pieces will be put on freelists
> of the wrong type. This encourages incompatible page mixing (ask for
> one type, get another), and thus long-term fragmentation.
>
> Update the subblocks when merging a larger chunk, such that a later
> expand() will maintain freelist type hygiene.
>
> v2:
> - remove spurious change_pageblock_range() move (Zi Yan)
>
> Signed-off-by: Johannes Weiner <hannes@cmpxchg.org>
> ---
>  mm/page_alloc.c | 15 +++++++++++----
>  1 file changed, 11 insertions(+), 4 deletions(-)
>

LGTM. Reviewed-by: Zi Yan <ziy@nvidia.com>

--
Best Regards,
Yan, Zi