[PATCH] mm/huge_memory: move writeback and truncation checks early

[PATCH] mm/huge_memory: move writeback and truncation checks early

[Luis Chamberlain]

We should check as early as possible if we should bail due to writeback
or truncation. This will allow us to add further sanity checks earlier
as well.

This introduces no functional changes.

Signed-off-by: Luis Chamberlain <mcgrof@kernel.org>
---
 mm/huge_memory.c | 23 +++++++++++------------
 1 file changed, 11 insertions(+), 12 deletions(-)

While working on min order support for LBS this came up as an improvement
as we can check for the min order early earlier, so this sets the stage
up for that.

diff --git a/mm/huge_memory.c b/mm/huge_memory.c
index 86a8c7b3b8dc..32c701821e0d 100644
--- a/mm/huge_memory.c
+++ b/mm/huge_memory.c
@@ -3055,8 +3055,17 @@ int split_huge_page_to_list_to_order(struct page *page, struct list_head *list,
 	if (new_order >= folio_order(folio))
 		return -EINVAL;
 
-	/* Cannot split anonymous THP to order-1 */
-	if (new_order == 1 && folio_test_anon(folio)) {
+	if (folio_test_writeback(folio))
+		return -EBUSY;
+
+	if (!folio_test_anon(folio)) {
+		/* Truncated ? */
+		if (!folio->mapping) {
+			ret = -EBUSY;
+			goto out;
+		}
+	} else if (new_order == 1) {
+		/* Cannot split anonymous THP to order-1 */
 		VM_WARN_ONCE(1, "Cannot split to order-1 folio");
 		return -EINVAL;
 	}
@@ -3079,16 +3088,12 @@ int split_huge_page_to_list_to_order(struct page *page, struct list_head *list,
 		}
 	}
 
-
 	is_hzp = is_huge_zero_page(&folio->page);
 	if (is_hzp) {
 		pr_warn_ratelimited("Called split_huge_page for huge zero page\n");
 		return -EBUSY;
 	}
 
-	if (folio_test_writeback(folio))
-		return -EBUSY;
-
 	if (folio_test_anon(folio)) {
 		/*
 		 * The caller does not necessarily hold an mmap_lock that would
@@ -3111,12 +3116,6 @@ int split_huge_page_to_list_to_order(struct page *page, struct list_head *list,
 
 		mapping = folio->mapping;
 
-		/* Truncated ? */
-		if (!mapping) {
-			ret = -EBUSY;
-			goto out;
-		}
-
 		/*
 		 * Do not split if mapping has minimum folio order
 		 * requirement.
-- 
2.43.0

Re: [PATCH] mm/huge_memory: move writeback and truncation checks early

[Zi Yan]

[-- Attachment #1: Type: text/plain, Size: 2291 bytes --]

On 24 Apr 2024, at 18:57, Luis Chamberlain wrote:

> We should check as early as possible if we should bail due to writeback
> or truncation. This will allow us to add further sanity checks earlier
> as well.
>
> This introduces no functional changes.
>
> Signed-off-by: Luis Chamberlain <mcgrof@kernel.org>
> ---
>  mm/huge_memory.c | 23 +++++++++++------------
>  1 file changed, 11 insertions(+), 12 deletions(-)
>
> While working on min order support for LBS this came up as an improvement
> as we can check for the min order early earlier, so this sets the stage
> up for that.
>
> diff --git a/mm/huge_memory.c b/mm/huge_memory.c
> index 86a8c7b3b8dc..32c701821e0d 100644
> --- a/mm/huge_memory.c
> +++ b/mm/huge_memory.c
> @@ -3055,8 +3055,17 @@ int split_huge_page_to_list_to_order(struct page *page, struct list_head *list,
>  	if (new_order >= folio_order(folio))
>  		return -EINVAL;
>
> -	/* Cannot split anonymous THP to order-1 */
> -	if (new_order == 1 && folio_test_anon(folio)) {
> +	if (folio_test_writeback(folio))
> +		return -EBUSY;
> +
> +	if (!folio_test_anon(folio)) {
> +		/* Truncated ? */
> +		if (!folio->mapping) {
> +			ret = -EBUSY;
> +			goto out;
> +		}
> +	} else if (new_order == 1) {
> +		/* Cannot split anonymous THP to order-1 */
>  		VM_WARN_ONCE(1, "Cannot split to order-1 folio");
>  		return -EINVAL;
>  	}
> @@ -3079,16 +3088,12 @@ int split_huge_page_to_list_to_order(struct page *page, struct list_head *list,
>  		}
>  	}
>
> -
>  	is_hzp = is_huge_zero_page(&folio->page);
>  	if (is_hzp) {
>  		pr_warn_ratelimited("Called split_huge_page for huge zero page\n");
>  		return -EBUSY;
>  	}
>
> -	if (folio_test_writeback(folio))
> -		return -EBUSY;
> -
>  	if (folio_test_anon(folio)) {
>  		/*
>  		 * The caller does not necessarily hold an mmap_lock that would
> @@ -3111,12 +3116,6 @@ int split_huge_page_to_list_to_order(struct page *page, struct list_head *list,
>
>  		mapping = folio->mapping;
>
> -		/* Truncated ? */
> -		if (!mapping) {
> -			ret = -EBUSY;
> -			goto out;
> -		}
> -
>  		/*
>  		 * Do not split if mapping has minimum folio order
>  		 * requirement.
> -- 
> 2.43.0
LGTM. Reviewed-by: Zi Yan <ziy@nvidia.com>

--
Best Regards,
Yan, Zi

[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 854 bytes --]

Re: [PATCH] mm/huge_memory: move writeback and truncation checks early

[David Hildenbrand]

On 25.04.24 00:57, Luis Chamberlain wrote:
> We should check as early as possible if we should bail due to writeback
> or truncation. This will allow us to add further sanity checks earlier
> as well.
> 
> This introduces no functional changes.
> 
> Signed-off-by: Luis Chamberlain <mcgrof@kernel.org>
> ---
>   mm/huge_memory.c | 23 +++++++++++------------
>   1 file changed, 11 insertions(+), 12 deletions(-)
> 
> While working on min order support for LBS this came up as an improvement
> as we can check for the min order early earlier, so this sets the stage
> up for that.
> 
> diff --git a/mm/huge_memory.c b/mm/huge_memory.c
> index 86a8c7b3b8dc..32c701821e0d 100644
> --- a/mm/huge_memory.c
> +++ b/mm/huge_memory.c
> @@ -3055,8 +3055,17 @@ int split_huge_page_to_list_to_order(struct page *page, struct list_head *list,
>   	if (new_order >= folio_order(folio))
>   		return -EINVAL;
>   
> -	/* Cannot split anonymous THP to order-1 */
> -	if (new_order == 1 && folio_test_anon(folio)) {
> +	if (folio_test_writeback(folio))
> +		return -EBUSY;
> +

Why earlier than basic input parameter checks (new_order?

Sorry, but I don't see the reason for that change. It's all happening 
extremely early, what are we concerned about?

It's likely better to send that patch with the actual patch "to add 
further sanity checks earlier as well", and why they have to be that early.

-- 
Cheers,

David / dhildenb

Re: [PATCH] mm/huge_memory: move writeback and truncation checks early

[Luis Chamberlain]

On Thu, Apr 25, 2024 at 11:40:48AM +0200, David Hildenbrand wrote:
> On 25.04.24 00:57, Luis Chamberlain wrote:
> > We should check as early as possible if we should bail due to writeback
> > or truncation. This will allow us to add further sanity checks earlier
> > as well.
> > 
> > This introduces no functional changes.
> > 
> > Signed-off-by: Luis Chamberlain <mcgrof@kernel.org>
> > ---
> >   mm/huge_memory.c | 23 +++++++++++------------
> >   1 file changed, 11 insertions(+), 12 deletions(-)
> > 
> > While working on min order support for LBS this came up as an improvement
> > as we can check for the min order early earlier, so this sets the stage
> > up for that.
> > 
> > diff --git a/mm/huge_memory.c b/mm/huge_memory.c
> > index 86a8c7b3b8dc..32c701821e0d 100644
> > --- a/mm/huge_memory.c
> > +++ b/mm/huge_memory.c
> > @@ -3055,8 +3055,17 @@ int split_huge_page_to_list_to_order(struct page *page, struct list_head *list,
> >   	if (new_order >= folio_order(folio))
> >   		return -EINVAL;
> > -	/* Cannot split anonymous THP to order-1 */
> > -	if (new_order == 1 && folio_test_anon(folio)) {
> > +	if (folio_test_writeback(folio))
> > +		return -EBUSY;
> > +
> 
> Why earlier than basic input parameter checks (new_order?
> 
> Sorry, but I don't see the reason for that change. It's all happening
> extremely early, what are we concerned about?
> 
> It's likely better to send that patch with the actual patch "to add further
> sanity checks earlier as well", and why they have to be that early.

It's a clear eye-sore when we add min order, I'll leave the eyesores for
others. It can wait.

  Luis