[PATCH v2] mm/vmscan: don't try to reclaim freed folios

[PATCH v2] mm/vmscan: don't try to reclaim freed folios

[Miaohe Lin]

If folios were freed from under us, there's no need to reclaim them. Skip
these folios to save lots of cpu cycles and avoid possible unnecessary
disk I/O.

Signed-off-by: Miaohe Lin <linmiaohe@huawei.com>
---
v2:
  use folio_ref_freeze to guard against race with GUP (fast). Many thanks
  Matthew for pointing this out.
---
 mm/vmscan.c | 10 ++++++++--
 1 file changed, 8 insertions(+), 2 deletions(-)

diff --git a/mm/vmscan.c b/mm/vmscan.c
index 13d34d9593bb..547ae7ae6ab1 100644
--- a/mm/vmscan.c
+++ b/mm/vmscan.c
@@ -1610,13 +1610,19 @@ static unsigned int shrink_page_list(struct list_head *page_list,
 		folio = lru_to_folio(page_list);
 		list_del(&folio->lru);
 
+		nr_pages = folio_nr_pages(folio);
+
+		if (folio_ref_count(folio) == 1 &&
+		    folio_ref_freeze(folio, 1)) {
+			/* folio was freed from under us. So we are done. */
+			goto free_it;
+		}
+
 		if (!folio_trylock(folio))
 			goto keep;
 
 		VM_BUG_ON_FOLIO(folio_test_active(folio), folio);
 
-		nr_pages = folio_nr_pages(folio);
-
 		/* Account the number of base pages */
 		sc->nr_scanned += nr_pages;
 
-- 
2.23.0

Re: [PATCH v2] mm/vmscan: don't try to reclaim freed folios

[Matthew Wilcox]

On Wed, Jun 08, 2022 at 10:14:32PM +0800, Miaohe Lin wrote:
> If folios were freed from under us, there's no need to reclaim them. Skip
> these folios to save lots of cpu cycles and avoid possible unnecessary
> disk I/O.

Yes, but I asked how often this happened, and you said you didn't know.
Do you have any data?  I'm reluctant to make a function which is over
400 LOC already any longer.

> Signed-off-by: Miaohe Lin <linmiaohe@huawei.com>
> ---
> v2:
>   use folio_ref_freeze to guard against race with GUP (fast). Many thanks
>   Matthew for pointing this out.
> ---
>  mm/vmscan.c | 10 ++++++++--
>  1 file changed, 8 insertions(+), 2 deletions(-)
> 
> diff --git a/mm/vmscan.c b/mm/vmscan.c
> index 13d34d9593bb..547ae7ae6ab1 100644
> --- a/mm/vmscan.c
> +++ b/mm/vmscan.c
> @@ -1610,13 +1610,19 @@ static unsigned int shrink_page_list(struct list_head *page_list,
>  		folio = lru_to_folio(page_list);
>  		list_del(&folio->lru);
>  
> +		nr_pages = folio_nr_pages(folio);
> +
> +		if (folio_ref_count(folio) == 1 &&
> +		    folio_ref_freeze(folio, 1)) {
> +			/* folio was freed from under us. So we are done. */
> +			goto free_it;
> +		}
> +
>  		if (!folio_trylock(folio))
>  			goto keep;
>  
>  		VM_BUG_ON_FOLIO(folio_test_active(folio), folio);
>  
> -		nr_pages = folio_nr_pages(folio);
> -
>  		/* Account the number of base pages */
>  		sc->nr_scanned += nr_pages;
>  
> -- 
> 2.23.0
> 

Re: [PATCH v2] mm/vmscan: don't try to reclaim freed folios

[Miaohe Lin]

On 2022/6/9 2:47, Matthew Wilcox wrote:
> On Wed, Jun 08, 2022 at 10:14:32PM +0800, Miaohe Lin wrote:
>> If folios were freed from under us, there's no need to reclaim them. Skip
>> these folios to save lots of cpu cycles and avoid possible unnecessary
>> disk I/O.
> 
> Yes, but I asked how often this happened, and you said you didn't know.
> Do you have any data?  I'm reluctant to make a function which is over

This is just like the page_count == 1 case when doing page migration in unmap_and_move().

> 400 LOC already any longer.

I'm fine to resend this patch until your work is done (It will be really grateful if I
will be notified when that work is done).

Thanks!

> 
>> Signed-off-by: Miaohe Lin <linmiaohe@huawei.com>
>> ---
>> v2:
>>   use folio_ref_freeze to guard against race with GUP (fast). Many thanks
>>   Matthew for pointing this out.
>> ---
>>  mm/vmscan.c | 10 ++++++++--
>>  1 file changed, 8 insertions(+), 2 deletions(-)
>>
>> diff --git a/mm/vmscan.c b/mm/vmscan.c
>> index 13d34d9593bb..547ae7ae6ab1 100644
>> --- a/mm/vmscan.c
>> +++ b/mm/vmscan.c
>> @@ -1610,13 +1610,19 @@ static unsigned int shrink_page_list(struct list_head *page_list,
>>  		folio = lru_to_folio(page_list);
>>  		list_del(&folio->lru);
>>  
>> +		nr_pages = folio_nr_pages(folio);
>> +
>> +		if (folio_ref_count(folio) == 1 &&
>> +		    folio_ref_freeze(folio, 1)) {
>> +			/* folio was freed from under us. So we are done. */
>> +			goto free_it;
>> +		}
>> +
>>  		if (!folio_trylock(folio))
>>  			goto keep;
>>  
>>  		VM_BUG_ON_FOLIO(folio_test_active(folio), folio);
>>  
>> -		nr_pages = folio_nr_pages(folio);
>> -
>>  		/* Account the number of base pages */
>>  		sc->nr_scanned += nr_pages;
>>  
>> -- 
>> 2.23.0
>>
> 
> .
>