[PATCH v2 0/2] Honor XA_FLAGS_ACCOUNT in xas_split_alloc() and charge to folio's memcg

[PATCH v2 0/2] Honor XA_FLAGS_ACCOUNT in xas_split_alloc() and charge to folio's memcg

[Zi Yan]

__GFP_ACCOUNT is needed for xarray node allocation accounting when
XA_FLAGS_ACCOUNT is set. Commit 7b785645e8f13 ("mm: fix page cache
convergence regression") fixed a workingset regression with it.
xas_split_alloc() does not have it and needs to be fixed.

In addition, based on Sashiko's review[1] and Johannes' confirmation[2], to
charge the right memcg, folio's memcg needs to be active during folio
split. Add that before adding __GFP_ACCOUNT.

There is no workingset convergence regression related to missing
__GFP_ACCOUNT in xas_split_alloc() and the impact to userspace should be
minor.

Both patches need be applied on top of the mapping_set_update() fix[3] to
avoid causing the same issue for uniform split path.

Link: https://sashiko.dev/#/patchset/20260727-add-gfp_account-to-xas_split_alloc-v1-1-9fae6bf64838%40nvidia.com?part=1 [1]
Link: https://lore.kernel.org/all/amtcBZ-_QVRgCd6b@cmpxchg.org/ [2]
Link: https://lore.kernel.org/all/20260725101419.3938406-1-matt@readmodwrite.com/ [3]

Signed-off-by: Zi Yan <ziy@nvidia.com>
---
Changes in v2:
1. added memcg switch code to __folio_split() to make __GFP_ACCOUNT charge
   to the right memcg.
2. removed RFC.
- Link to v1: https://lore.kernel.org/r/20260727-add-gfp_account-to-xas_split_alloc-v1-1-9fae6bf64838@nvidia.com

---
Zi Yan (2):
      mm/huge_memory: use folio's memcg inside __folio_split()
      xarray: honor XA_FLAGS_ACCOUNT in xas_split_alloc()

 lib/xarray.c     |  3 +++
 mm/huge_memory.c | 20 ++++++++++++++++----
 2 files changed, 19 insertions(+), 4 deletions(-)
---
base-commit: 34cacd0ac7c10dc9e34e1b9f28a2b78d335cb487
change-id: 20260727-add-gfp_account-to-xas_split_alloc-83a1bc848b77

Best regards,
-- 
Yan, Zi

[PATCH v2 1/2] mm/huge_memory: use folio's memcg inside __folio_split()

[Zi Yan]

During a pagecache folio split, an xarray node allocation can happen and
needs to charge at folio's memcg instead of folio split invoker's memcg,
because for example folio split can happen during reclaim and reclaim's
active memcg might not be folio's memcg. Switch to folio's memcg at the
beginning and switch back afterwards.

Suggested-by: Johannes Weiner <hannes@cmpxchg.org>
Fixes: 6b24ca4a1a8d4 ("mm: Use multi-index entries in the page cache")
Signed-off-by: Zi Yan <ziy@nvidia.com>
---
 mm/huge_memory.c | 20 ++++++++++++++++----
 1 file changed, 16 insertions(+), 4 deletions(-)

diff --git a/mm/huge_memory.c b/mm/huge_memory.c
index 04e8a6b553435..b9c2d8908e564 100644
--- a/mm/huge_memory.c
+++ b/mm/huge_memory.c
@@ -4063,34 +4063,42 @@ static int __folio_split(struct folio *folio, unsigned int new_order,
 	XA_STATE(xas, &folio->mapping->i_pages, folio->index);
 	struct folio *end_folio = folio_next(folio);
 	bool is_anon = folio_test_anon(folio);
+	struct mem_cgroup *memcg, *old_memcg;
 	struct address_space *mapping = NULL;
 	struct anon_vma *anon_vma = NULL;
 	int old_order = folio_order(folio);
 	struct folio *new_folio, *next;
 	int nr_shmem_dropped = 0;
 	enum ttu_flags ttu_flags = 0;
-	int ret;
 	pgoff_t end = 0;
+	int ret;
 
 	VM_WARN_ON_ONCE_FOLIO(!folio_test_locked(folio), folio);
 	VM_WARN_ON_ONCE_FOLIO(!folio_test_large(folio), folio);
 
 	if (folio != page_folio(split_at) || folio != page_folio(lock_at)) {
 		ret = -EINVAL;
-		goto out;
+		goto out_no_memcg;
 	}
 
 	if (new_order >= old_order) {
 		ret = -EINVAL;
-		goto out;
+		goto out_no_memcg;
 	}
 
 	ret = folio_check_splittable(folio, new_order, split_type);
 	if (ret) {
 		VM_WARN_ONCE(ret == -EINVAL, "Tried to split an unsplittable folio");
-		goto out;
+		goto out_no_memcg;
 	}
 
+	/*
+	 * switch to folio's memcg as xarray node allocation can happen and
+	 * needs to charge to it.
+	 */
+	memcg = get_mem_cgroup_from_folio(folio);
+	old_memcg = set_active_memcg(memcg);
+
 	if (is_anon) {
 		/*
 		 * The caller does not necessarily hold an mmap_lock that would
@@ -4231,6 +4239,10 @@ static int __folio_split(struct folio *folio, unsigned int new_order,
 	if (mapping)
 		i_mmap_unlock_read(mapping);
 out:
+	/* restore to caller's old_memcg */
+	set_active_memcg(old_memcg);
+	mem_cgroup_put(memcg);
+out_no_memcg:
 	xas_destroy(&xas);
 	if (is_pmd_order(old_order))
 		count_vm_event(!ret ? THP_SPLIT_PAGE : THP_SPLIT_PAGE_FAILED);

-- 
2.53.0

[PATCH v2 2/2] xarray: honor XA_FLAGS_ACCOUNT in xas_split_alloc()

[Zi Yan]

XArray operations that allocate xa_nodes, such as xas_nomem() and
xas_alloc(), add __GFP_ACCOUNT when the array has XA_FLAGS_ACCOUNT set.
This charges the allocated memory and avoids the workingset convergence
issue described by commit 7b785645e8f13 ("mm: fix page cache convergence
regression").

xas_split_alloc() does not have that flag. Add it when necessary.

Fixes: 6b24ca4a1a8d4 ("mm: Use multi-index entries in the page cache")
Signed-off-by: Zi Yan <ziy@nvidia.com>
---
 lib/xarray.c | 3 +++
 1 file changed, 3 insertions(+)

diff --git a/lib/xarray.c b/lib/xarray.c
index 9a8b4916540cf..bfe7bef80f34e 100644
--- a/lib/xarray.c
+++ b/lib/xarray.c
@@ -1053,6 +1053,9 @@ void xas_split_alloc(struct xa_state *xas, void *entry, unsigned int order,
 	if (xas->xa_shift + XA_CHUNK_SHIFT > order)
 		return;
 
+	if (xas->xa->xa_flags & XA_FLAGS_ACCOUNT)
+		gfp |= __GFP_ACCOUNT;
+
 	do {
 		struct xa_node *node;
 

-- 
2.53.0

Re: [PATCH v2 1/2] mm/huge_memory: use folio's memcg inside __folio_split()

[Johannes Weiner]

On Thu, Jul 30, 2026 at 10:18:00PM -0400, Zi Yan wrote:
> During a pagecache folio split, an xarray node allocation can happen and
> needs to charge at folio's memcg instead of folio split invoker's memcg,
> because for example folio split can happen during reclaim and reclaim's
> active memcg might not be folio's memcg. Switch to folio's memcg at the
> beginning and switch back afterwards.
> 
> Suggested-by: Johannes Weiner <hannes@cmpxchg.org>
> Fixes: 6b24ca4a1a8d4 ("mm: Use multi-index entries in the page cache")
> Signed-off-by: Zi Yan <ziy@nvidia.com>

Acked-by: Johannes Weiner <hannes@cmpxchg.org>

Re: [PATCH v2 2/2] xarray: honor XA_FLAGS_ACCOUNT in xas_split_alloc()

[Johannes Weiner]

On Thu, Jul 30, 2026 at 10:18:01PM -0400, Zi Yan wrote:
> XArray operations that allocate xa_nodes, such as xas_nomem() and
> xas_alloc(), add __GFP_ACCOUNT when the array has XA_FLAGS_ACCOUNT set.
> This charges the allocated memory and avoids the workingset convergence
> issue described by commit 7b785645e8f13 ("mm: fix page cache convergence
> regression").
> 
> xas_split_alloc() does not have that flag. Add it when necessary.
> 
> Fixes: 6b24ca4a1a8d4 ("mm: Use multi-index entries in the page cache")
> Signed-off-by: Zi Yan <ziy@nvidia.com>

Acked-by: Johannes Weiner <hannes@cmpxchg.org>

Re: [PATCH v2 1/2] mm/huge_memory: use folio's memcg inside __folio_split()

[Lorenzo Stoakes (ARM)]

On Thu, Jul 30, 2026 at 10:18:00PM -0400, Zi Yan wrote:
> During a pagecache folio split, an xarray node allocation can happen and
> needs to charge at folio's memcg instead of folio split invoker's memcg,
> because for example folio split can happen during reclaim and reclaim's
> active memcg might not be folio's memcg. Switch to folio's memcg at the
> beginning and switch back afterwards.

I assume this is the only allocation? I guess in general it makes sense to have
the folio's memcg be active here regardless.

>
> Suggested-by: Johannes Weiner <hannes@cmpxchg.org>
> Fixes: 6b24ca4a1a8d4 ("mm: Use multi-index entries in the page cache")

Cc: stable?

> Signed-off-by: Zi Yan <ziy@nvidia.com>

Change seems reasonable overall.

Acked-by: Lorenzo Stoakes (ARM) <ljs@kernel.org>

> ---
>  mm/huge_memory.c | 20 ++++++++++++++++----
>  1 file changed, 16 insertions(+), 4 deletions(-)
>
> diff --git a/mm/huge_memory.c b/mm/huge_memory.c
> index 04e8a6b553435..b9c2d8908e564 100644
> --- a/mm/huge_memory.c
> +++ b/mm/huge_memory.c
> @@ -4063,34 +4063,42 @@ static int __folio_split(struct folio *folio, unsigned int new_order,
>  	XA_STATE(xas, &folio->mapping->i_pages, folio->index);
>  	struct folio *end_folio = folio_next(folio);
>  	bool is_anon = folio_test_anon(folio);
> +	struct mem_cgroup *memcg, *old_memcg;
>  	struct address_space *mapping = NULL;
>  	struct anon_vma *anon_vma = NULL;
>  	int old_order = folio_order(folio);
>  	struct folio *new_folio, *next;
>  	int nr_shmem_dropped = 0;
>  	enum ttu_flags ttu_flags = 0;
> -	int ret;
>  	pgoff_t end = 0;
> +	int ret;
>
>  	VM_WARN_ON_ONCE_FOLIO(!folio_test_locked(folio), folio);
>  	VM_WARN_ON_ONCE_FOLIO(!folio_test_large(folio), folio);
>
>  	if (folio != page_folio(split_at) || folio != page_folio(lock_at)) {
>  		ret = -EINVAL;
> -		goto out;
> +		goto out_no_memcg;
>  	}
>
>  	if (new_order >= old_order) {
>  		ret = -EINVAL;
> -		goto out;
> +		goto out_no_memcg;
>  	}
>
>  	ret = folio_check_splittable(folio, new_order, split_type);
>  	if (ret) {
>  		VM_WARN_ONCE(ret == -EINVAL, "Tried to split an unsplittable folio");
> -		goto out;
> +		goto out_no_memcg;

This function really badly needs splitting up and probably some cleanup.h work :)

>  	}
>
> +	/*
> +	 * switch to folio's memcg as xarray node allocation can happen and
> +	 * needs to charge to it.
> +	 */
> +	memcg = get_mem_cgroup_from_folio(folio);
> +	old_memcg = set_active_memcg(memcg);
> +
>  	if (is_anon) {
>  		/*
>  		 * The caller does not necessarily hold an mmap_lock that would
> @@ -4231,6 +4239,10 @@ static int __folio_split(struct folio *folio, unsigned int new_order,
>  	if (mapping)
>  		i_mmap_unlock_read(mapping);
>  out:
> +	/* restore to caller's old_memcg */
> +	set_active_memcg(old_memcg);
> +	mem_cgroup_put(memcg);
> +out_no_memcg:
>  	xas_destroy(&xas);
>  	if (is_pmd_order(old_order))
>  		count_vm_event(!ret ? THP_SPLIT_PAGE : THP_SPLIT_PAGE_FAILED);
>
> --
> 2.53.0
>

--
Cheers, Lorenzo

Re: [PATCH v2 2/2] xarray: honor XA_FLAGS_ACCOUNT in xas_split_alloc()

[Lorenzo Stoakes (ARM)]

On Thu, Jul 30, 2026 at 10:18:01PM -0400, Zi Yan wrote:
> XArray operations that allocate xa_nodes, such as xas_nomem() and
> xas_alloc(), add __GFP_ACCOUNT when the array has XA_FLAGS_ACCOUNT set.
> This charges the allocated memory and avoids the workingset convergence
> issue described by commit 7b785645e8f13 ("mm: fix page cache convergence
> regression").
>
> xas_split_alloc() does not have that flag. Add it when necessary.

Nit but maybe 'split' rather than 'have'?

>
> Fixes: 6b24ca4a1a8d4 ("mm: Use multi-index entries in the page cache")

Cc: stable?

> Signed-off-by: Zi Yan <ziy@nvidia.com>

Makes sense to me so:

Reviewed-by: Lorenzo Stoakes (ARM) <ljs@kernel.org>

> ---
>  lib/xarray.c | 3 +++
>  1 file changed, 3 insertions(+)
>
> diff --git a/lib/xarray.c b/lib/xarray.c
> index 9a8b4916540cf..bfe7bef80f34e 100644
> --- a/lib/xarray.c
> +++ b/lib/xarray.c
> @@ -1053,6 +1053,9 @@ void xas_split_alloc(struct xa_state *xas, void *entry, unsigned int order,
>  	if (xas->xa_shift + XA_CHUNK_SHIFT > order)
>  		return;
>
> +	if (xas->xa->xa_flags & XA_FLAGS_ACCOUNT)
> +		gfp |= __GFP_ACCOUNT;
> +

This is some confluence of flags :) I wonder if there are other places we've
missed setting this for?

>  	do {
>  		struct xa_node *node;
>
>
> --
> 2.53.0
>

--
Cheers, Lorenzo