diff mbox series

[1/1] mm/migrate: correct nr_failed in migrate_pages_sync()

Message ID 20231009203231.1715845-2-zi.yan@sent.com
State New
Headers
Series Large folio migration fix and questions on migration stats |

Commit Message

Zi Yan Oct. 9, 2023, 8:32 p.m. UTC 
  From: Zi Yan <ziy@nvidia.com>

nr_failed was missing the rc value from migrate_pages_batch() and can
cause a mismatch between migrate_pages() return value and the number of
not migrated pages, i.e., when the return value of migrate_pages() is 0,
there are still pages left in the from page list. It will happen when a
non-PMD THP large folio fails to migrate due to -ENOMEM and is split
successfully but not all the split pages are not migrated,
migrate_pages_batch() would return non-zero, but astats.nr_thp_split = 0.
nr_failed would be 0 and returned to the caller of migrate_pages(), but
the not migrated pages are left in the from page list without being added
back to LRU lists.

Fixes: 2ef7dbb26990 ("migrate_pages: try migrate in batch asynchronously firstly")
Signed-off-by: Zi Yan <ziy@nvidia.com>
---
 mm/migrate.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

Comments

Huang, Ying Oct. 10, 2023, 3:36 a.m. UTC | #1 
Zi Yan <zi.yan@sent.com> writes:

> From: Zi Yan <ziy@nvidia.com>
>
> nr_failed was missing the rc value from migrate_pages_batch() and can
> cause a mismatch between migrate_pages() return value and the number of
> not migrated pages, i.e., when the return value of migrate_pages() is 0,
> there are still pages left in the from page list. It will happen when a
> non-PMD THP large folio fails to migrate due to -ENOMEM and is split
> successfully but not all the split pages are not migrated,
> migrate_pages_batch() would return non-zero, but astats.nr_thp_split = 0.
> nr_failed would be 0 and returned to the caller of migrate_pages(), but
> the not migrated pages are left in the from page list without being added
> back to LRU lists.
>
> Fixes: 2ef7dbb26990 ("migrate_pages: try migrate in batch asynchronously firstly")
> Signed-off-by: Zi Yan <ziy@nvidia.com>
> ---
>  mm/migrate.c | 2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)
>
> diff --git a/mm/migrate.c b/mm/migrate.c
> index c602bf6dec97..5348827bd958 100644
> --- a/mm/migrate.c
> +++ b/mm/migrate.c
> @@ -1834,7 +1834,7 @@ static int migrate_pages_sync(struct list_head *from, new_folio_t get_new_folio,
>  		return rc;
>  	}
>  	stats->nr_thp_failed += astats.nr_thp_split;
> -	nr_failed += astats.nr_thp_split;
> +	nr_failed += rc + astats.nr_thp_split;
>  	/*
>  	 * Fall back to migrate all failed folios one by one synchronously. All
>  	 * failed folios except split THPs will be retried, so their failure

I don't think this is a correct fix.  The failed folios will be retried
in the following synchronous migration below.

To fix the issue, we should track nr_split for all large folios (not
only THP), then use

        nr_failed += astats.nr_split;

--
Best Regards,
Huang, Ying
Zi Yan Oct. 10, 2023, 3:53 p.m. UTC | #2 
On 9 Oct 2023, at 23:36, Huang, Ying wrote:

> Zi Yan <zi.yan@sent.com> writes:
>
>> From: Zi Yan <ziy@nvidia.com>
>>
>> nr_failed was missing the rc value from migrate_pages_batch() and can
>> cause a mismatch between migrate_pages() return value and the number of
>> not migrated pages, i.e., when the return value of migrate_pages() is 0,
>> there are still pages left in the from page list. It will happen when a
>> non-PMD THP large folio fails to migrate due to -ENOMEM and is split
>> successfully but not all the split pages are not migrated,
>> migrate_pages_batch() would return non-zero, but astats.nr_thp_split = 0.
>> nr_failed would be 0 and returned to the caller of migrate_pages(), but
>> the not migrated pages are left in the from page list without being added
>> back to LRU lists.
>>
>> Fixes: 2ef7dbb26990 ("migrate_pages: try migrate in batch asynchronously firstly")
>> Signed-off-by: Zi Yan <ziy@nvidia.com>
>> ---
>>  mm/migrate.c | 2 +-
>>  1 file changed, 1 insertion(+), 1 deletion(-)
>>
>> diff --git a/mm/migrate.c b/mm/migrate.c
>> index c602bf6dec97..5348827bd958 100644
>> --- a/mm/migrate.c
>> +++ b/mm/migrate.c
>> @@ -1834,7 +1834,7 @@ static int migrate_pages_sync(struct list_head *from, new_folio_t get_new_folio,
>>  		return rc;
>>  	}
>>  	stats->nr_thp_failed += astats.nr_thp_split;
>> -	nr_failed += astats.nr_thp_split;
>> +	nr_failed += rc + astats.nr_thp_split;
>>  	/*
>>  	 * Fall back to migrate all failed folios one by one synchronously. All
>>  	 * failed folios except split THPs will be retried, so their failure
>
> I don't think this is a correct fix.  The failed folios will be retried
> in the following synchronous migration below.
>
> To fix the issue, we should track nr_split for all large folios (not
> only THP), then use
>
>         nr_failed += astats.nr_split;

You are suggesting a new stats "nr_split" in addition to nr_thp_split? And
nr_split includes nr_thp_split?

--
Best Regards,
Yan, Zi
diff mbox series

Patch

diff --git a/mm/migrate.c b/mm/migrate.c
index c602bf6dec97..5348827bd958 100644
--- a/mm/migrate.c
+++ b/mm/migrate.c
@@ -1834,7 +1834,7 @@  static int migrate_pages_sync(struct list_head *from, new_folio_t get_new_folio,
 		return rc;
 	}
 	stats->nr_thp_failed += astats.nr_thp_split;
-	nr_failed += astats.nr_thp_split;
+	nr_failed += rc + astats.nr_thp_split;
 	/*
 	 * Fall back to migrate all failed folios one by one synchronously. All
 	 * failed folios except split THPs will be retried, so their failure