[PATCH v2 0/4] mm: zswap: misc optimizations

[PATCH v2 0/4] mm: zswap: misc optimizations

[Kefeng Wang]

This series cleans up duplication in zswap_invalidate() and adds
fast paths in zswap_store() which now skips redundant xarray lookups,
while swap_range_free() skips needless zswap_invalidate() calls.

v2:
- split patches and refactor zswap_invalidate(), suggested by
  Johannes Weiner
- add patch4 to skip zswap_invalidate() in swap_range_free()

Kefeng Wang (4):
  mm: zswap: pass type and offset to zswap_invalidate() directly
  mm: zswap: reuse zswap_invalidate() in zswap_store() check_old path
  mm: zswap: avoid unnecessary xarray lookup in zswap_store()
  mm: zswap: skip zswap_invalidate() in swap_range_free() when zswap is
    unused

 include/linux/zswap.h |  4 ++--
 mm/swapfile.c         |  6 ++++--
 mm/zswap.c            | 29 +++++++++++++++--------------
 3 files changed, 21 insertions(+), 18 deletions(-)

-- 
2.55.0

[PATCH v2 1/4] mm: zswap: pass type and offset to zswap_invalidate() directly

[Kefeng Wang]

zswap_invalidate() takes a swp_entry_t only to unpack it right back
into type and offset.  Both callers already have those values, so
pass them directly and drop the swp_entry()/swp_type()/swp_offset()
round-trip.

Signed-off-by: Kefeng Wang <wangkefeng.wang@huawei.com>
---
 include/linux/zswap.h |  4 ++--
 mm/swapfile.c         |  2 +-
 mm/zswap.c            | 14 +++++++++-----
 3 files changed, 12 insertions(+), 8 deletions(-)

diff --git a/include/linux/zswap.h b/include/linux/zswap.h
index 30c193a1207e..463bdee5c1e1 100644
--- a/include/linux/zswap.h
+++ b/include/linux/zswap.h
@@ -27,7 +27,7 @@ struct zswap_lruvec_state {
 unsigned long zswap_total_pages(void);
 bool zswap_store(struct folio *folio);
 int zswap_load(struct folio *folio);
-void zswap_invalidate(swp_entry_t swp);
+void zswap_invalidate(int type, pgoff_t offset);
 int zswap_swapon(int type, unsigned long nr_pages);
 void zswap_swapoff(int type);
 void zswap_memcg_offline_cleanup(struct mem_cgroup *memcg);
@@ -49,7 +49,7 @@ static inline int zswap_load(struct folio *folio)
 	return -ENOENT;
 }
 
-static inline void zswap_invalidate(swp_entry_t swp) {}
+static inline void zswap_invalidate(int type, pgoff_t offset) {}
 static inline int zswap_swapon(int type, unsigned long nr_pages)
 {
 	return 0;
diff --git a/mm/swapfile.c b/mm/swapfile.c
index 05d3408396f9..ba71905e4d46 100644
--- a/mm/swapfile.c
+++ b/mm/swapfile.c
@@ -1319,7 +1319,7 @@ static void swap_range_free(struct swap_info_struct *si, unsigned long offset,
 	unsigned int i;
 
 	for (i = 0; i < nr_entries; i++)
-		zswap_invalidate(swp_entry(si->type, offset + i));
+		zswap_invalidate(si->type, offset + i);
 
 	if (si->flags & SWP_BLKDEV)
 		swap_slot_free_notify =
diff --git a/mm/zswap.c b/mm/zswap.c
index a12b6452a225..420c405d0402 100644
--- a/mm/zswap.c
+++ b/mm/zswap.c
@@ -228,10 +228,15 @@ static bool zswap_has_pool;
 /* One swap address space for each 64M swap space */
 #define ZSWAP_ADDRESS_SPACE_SHIFT 14
 #define ZSWAP_ADDRESS_SPACE_PAGES (1 << ZSWAP_ADDRESS_SPACE_SHIFT)
+
+static inline struct xarray *zswap_tree(int type, pgoff_t offset)
+{
+	return &zswap_trees[type][offset >> ZSWAP_ADDRESS_SPACE_SHIFT];
+}
+
 static inline struct xarray *swap_zswap_tree(swp_entry_t swp)
 {
-	return &zswap_trees[swp_type(swp)][swp_offset(swp)
-		>> ZSWAP_ADDRESS_SPACE_SHIFT];
+	return zswap_tree(swp_type(swp), swp_offset(swp));
 }
 
 #define zswap_pool_debug(msg, p)			\
@@ -1659,10 +1664,9 @@ int zswap_load(struct folio *folio)
 	return 0;
 }
 
-void zswap_invalidate(swp_entry_t swp)
+void zswap_invalidate(int type, pgoff_t offset)
 {
-	pgoff_t offset = swp_offset(swp);
-	struct xarray *tree = swap_zswap_tree(swp);
+	struct xarray *tree = zswap_tree(type, offset);
 	struct zswap_entry *entry;
 
 	if (xa_empty(tree))
-- 
2.55.0

[PATCH v2 2/4] mm: zswap: reuse zswap_invalidate() in zswap_store() check_old path

[Kefeng Wang]

The check_old path in zswap_store() open-codes the same per-slot
xarray lookup and entry free that zswap_invalidate() already does.
Reuse zswap_invalidate() whose xa_empty() check skips empty per-area
trees to avoid unnecessary xarray lookup and code duplication.

Signed-off-by: Kefeng Wang <wangkefeng.wang@huawei.com>
---
 mm/zswap.c | 12 +++---------
 1 file changed, 3 insertions(+), 9 deletions(-)

diff --git a/mm/zswap.c b/mm/zswap.c
index 420c405d0402..2c06e4e0e130 100644
--- a/mm/zswap.c
+++ b/mm/zswap.c
@@ -1549,15 +1549,9 @@ bool zswap_store(struct folio *folio)
 	if (!ret) {
 		unsigned type = swp_type(swp);
 		pgoff_t offset = swp_offset(swp);
-		struct zswap_entry *entry;
-		struct xarray *tree;
-
-		for (index = 0; index < nr_pages; ++index) {
-			tree = swap_zswap_tree(swp_entry(type, offset + index));
-			entry = xa_erase(tree, offset + index);
-			if (entry)
-				zswap_entry_free(entry);
-		}
+
+		for (index = 0; index < nr_pages; ++index)
+			zswap_invalidate(type, offset + index);
 	}
 
 	return ret;
-- 
2.55.0

[PATCH v2 3/4] mm: zswap: avoid unnecessary xarray lookup in zswap_store()

[Kefeng Wang]

zswap_store() falls through to check_old and walks the swap xarray
even when zswap is disabled. Add a zswap_never_enabled() early return
matching zswap_load() to avoid unnecessary xarray lookup.

Reviewed-by: Johannes Weiner <hannes@cmpxchg.org>
Signed-off-by: Kefeng Wang <wangkefeng.wang@huawei.com>
---
 mm/zswap.c | 3 +++
 1 file changed, 3 insertions(+)

diff --git a/mm/zswap.c b/mm/zswap.c
index 2c06e4e0e130..6197aa71e33c 100644
--- a/mm/zswap.c
+++ b/mm/zswap.c
@@ -1492,6 +1492,9 @@ bool zswap_store(struct folio *folio)
 	VM_WARN_ON_ONCE(!folio_test_locked(folio));
 	VM_WARN_ON_ONCE(!folio_test_swapcache(folio));
 
+	if (zswap_never_enabled())
+		return false;
+
 	if (!zswap_enabled)
 		goto check_old;
 
-- 
2.55.0

[PATCH v2 4/4] mm: zswap: skip zswap_invalidate() in swap_range_free() when zswap is unused

[Kefeng Wang]

swap_range_free() calls zswap_invalidate() for every slot being
freed, even when zswap has never been enabled. Guard the loop with
zswap_never_enabled() to skip it.

Signed-off-by: Kefeng Wang <wangkefeng.wang@huawei.com>
---
 mm/swapfile.c | 6 ++++--
 1 file changed, 4 insertions(+), 2 deletions(-)

diff --git a/mm/swapfile.c b/mm/swapfile.c
index ba71905e4d46..505592051924 100644
--- a/mm/swapfile.c
+++ b/mm/swapfile.c
@@ -1318,8 +1318,10 @@ static void swap_range_free(struct swap_info_struct *si, unsigned long offset,
 	void (*swap_slot_free_notify)(struct block_device *, unsigned long);
 	unsigned int i;
 
-	for (i = 0; i < nr_entries; i++)
-		zswap_invalidate(si->type, offset + i);
+	if (!zswap_never_enabled()) {
+		for (i = 0; i < nr_entries; i++)
+			zswap_invalidate(si->type, offset + i);
+	}
 
 	if (si->flags & SWP_BLKDEV)
 		swap_slot_free_notify =
-- 
2.55.0

Re: [PATCH v2 1/4] mm: zswap: pass type and offset to zswap_invalidate() directly

[Yosry Ahmed]

On Thu, Sep 10, 2026 at 2:55 AM Kefeng Wang <wangkefeng.wang@huawei.com> wrote:
>
> zswap_invalidate() takes a swp_entry_t only to unpack it right back
> into type and offset.  Both callers already have those values, so
> pass them directly and drop the swp_entry()/swp_type()/swp_offset()
> round-trip.
>

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

> Signed-off-by: Kefeng Wang <wangkefeng.wang@huawei.com>

Acked-by: Yosry Ahmed <yosry@kernel.org>

> ---
>  include/linux/zswap.h |  4 ++--
>  mm/swapfile.c         |  2 +-
>  mm/zswap.c            | 14 +++++++++-----
>  3 files changed, 12 insertions(+), 8 deletions(-)
>
> diff --git a/include/linux/zswap.h b/include/linux/zswap.h
> index 30c193a1207e..463bdee5c1e1 100644
> --- a/include/linux/zswap.h
> +++ b/include/linux/zswap.h
> @@ -27,7 +27,7 @@ struct zswap_lruvec_state {
>  unsigned long zswap_total_pages(void);
>  bool zswap_store(struct folio *folio);
>  int zswap_load(struct folio *folio);
> -void zswap_invalidate(swp_entry_t swp);
> +void zswap_invalidate(int type, pgoff_t offset);
>  int zswap_swapon(int type, unsigned long nr_pages);
>  void zswap_swapoff(int type);
>  void zswap_memcg_offline_cleanup(struct mem_cgroup *memcg);
> @@ -49,7 +49,7 @@ static inline int zswap_load(struct folio *folio)
>         return -ENOENT;
>  }
>
> -static inline void zswap_invalidate(swp_entry_t swp) {}
> +static inline void zswap_invalidate(int type, pgoff_t offset) {}
>  static inline int zswap_swapon(int type, unsigned long nr_pages)
>  {
>         return 0;
> diff --git a/mm/swapfile.c b/mm/swapfile.c
> index 05d3408396f9..ba71905e4d46 100644
> --- a/mm/swapfile.c
> +++ b/mm/swapfile.c
> @@ -1319,7 +1319,7 @@ static void swap_range_free(struct swap_info_struct *si, unsigned long offset,
>         unsigned int i;
>
>         for (i = 0; i < nr_entries; i++)
> -               zswap_invalidate(swp_entry(si->type, offset + i));
> +               zswap_invalidate(si->type, offset + i);
>
>         if (si->flags & SWP_BLKDEV)
>                 swap_slot_free_notify =
> diff --git a/mm/zswap.c b/mm/zswap.c
> index a12b6452a225..420c405d0402 100644
> --- a/mm/zswap.c
> +++ b/mm/zswap.c
> @@ -228,10 +228,15 @@ static bool zswap_has_pool;
>  /* One swap address space for each 64M swap space */
>  #define ZSWAP_ADDRESS_SPACE_SHIFT 14
>  #define ZSWAP_ADDRESS_SPACE_PAGES (1 << ZSWAP_ADDRESS_SPACE_SHIFT)
> +
> +static inline struct xarray *zswap_tree(int type, pgoff_t offset)
> +{
> +       return &zswap_trees[type][offset >> ZSWAP_ADDRESS_SPACE_SHIFT];
> +}
> +
>  static inline struct xarray *swap_zswap_tree(swp_entry_t swp)
>  {
> -       return &zswap_trees[swp_type(swp)][swp_offset(swp)
> -               >> ZSWAP_ADDRESS_SPACE_SHIFT];
> +       return zswap_tree(swp_type(swp), swp_offset(swp));
>  }
>
>  #define zswap_pool_debug(msg, p)                       \
> @@ -1659,10 +1664,9 @@ int zswap_load(struct folio *folio)
>         return 0;
>  }
>
> -void zswap_invalidate(swp_entry_t swp)
> +void zswap_invalidate(int type, pgoff_t offset)
>  {
> -       pgoff_t offset = swp_offset(swp);
> -       struct xarray *tree = swap_zswap_tree(swp);
> +       struct xarray *tree = zswap_tree(type, offset);
>         struct zswap_entry *entry;
>
>         if (xa_empty(tree))
> --
> 2.55.0
>
>

Re: [PATCH v2 2/4] mm: zswap: reuse zswap_invalidate() in zswap_store() check_old path

[Yosry Ahmed]

On Thu, Sep 10, 2026 at 2:55 AM Kefeng Wang <wangkefeng.wang@huawei.com> wrote:
>
> The check_old path in zswap_store() open-codes the same per-slot
> xarray lookup and entry free that zswap_invalidate() already does.
> Reuse zswap_invalidate() whose xa_empty() check skips empty per-area
> trees to avoid unnecessary xarray lookup and code duplication.
>
> Signed-off-by: Kefeng Wang <wangkefeng.wang@huawei.com>

Nice.

Acked-by: Yosry Ahmed <yosry@kernel.org>

> ---
>  mm/zswap.c | 12 +++---------
>  1 file changed, 3 insertions(+), 9 deletions(-)
>
> diff --git a/mm/zswap.c b/mm/zswap.c
> index 420c405d0402..2c06e4e0e130 100644
> --- a/mm/zswap.c
> +++ b/mm/zswap.c
> @@ -1549,15 +1549,9 @@ bool zswap_store(struct folio *folio)
>         if (!ret) {
>                 unsigned type = swp_type(swp);
>                 pgoff_t offset = swp_offset(swp);
> -               struct zswap_entry *entry;
> -               struct xarray *tree;
> -
> -               for (index = 0; index < nr_pages; ++index) {
> -                       tree = swap_zswap_tree(swp_entry(type, offset + index));
> -                       entry = xa_erase(tree, offset + index);
> -                       if (entry)
> -                               zswap_entry_free(entry);
> -               }
> +
> +               for (index = 0; index < nr_pages; ++index)
> +                       zswap_invalidate(type, offset + index);
>         }
>
>         return ret;
> --
> 2.55.0
>
>

Re: [PATCH v2 3/4] mm: zswap: avoid unnecessary xarray lookup in zswap_store()

[Yosry Ahmed]

On Thu, Sep 10, 2026 at 2:55 AM Kefeng Wang <wangkefeng.wang@huawei.com> wrote:
>
> zswap_store() falls through to check_old and walks the swap xarray
> even when zswap is disabled. Add a zswap_never_enabled() early return
> matching zswap_load() to avoid unnecessary xarray lookup.
>
> Reviewed-by: Johannes Weiner <hannes@cmpxchg.org>
> Signed-off-by: Kefeng Wang <wangkefeng.wang@huawei.com>

Acked-by: Yosry Ahmed <yosry@kernel.org>

> ---
>  mm/zswap.c | 3 +++
>  1 file changed, 3 insertions(+)
>
> diff --git a/mm/zswap.c b/mm/zswap.c
> index 2c06e4e0e130..6197aa71e33c 100644
> --- a/mm/zswap.c
> +++ b/mm/zswap.c
> @@ -1492,6 +1492,9 @@ bool zswap_store(struct folio *folio)
>         VM_WARN_ON_ONCE(!folio_test_locked(folio));
>         VM_WARN_ON_ONCE(!folio_test_swapcache(folio));
>
> +       if (zswap_never_enabled())
> +               return false;
> +
>         if (!zswap_enabled)
>                 goto check_old;
>
> --
> 2.55.0
>
>

Re: [PATCH v2 4/4] mm: zswap: skip zswap_invalidate() in swap_range_free() when zswap is unused

[Yosry Ahmed]

On Thu, Sep 10, 2026 at 2:55 AM Kefeng Wang <wangkefeng.wang@huawei.com> wrote:
>
> swap_range_free() calls zswap_invalidate() for every slot being
> freed, even when zswap has never been enabled. Guard the loop with
> zswap_never_enabled() to skip it.
>
> Signed-off-by: Kefeng Wang <wangkefeng.wang@huawei.com>
> ---
>  mm/swapfile.c | 6 ++++--
>  1 file changed, 4 insertions(+), 2 deletions(-)
>
> diff --git a/mm/swapfile.c b/mm/swapfile.c
> index ba71905e4d46..505592051924 100644
> --- a/mm/swapfile.c
> +++ b/mm/swapfile.c
> @@ -1318,8 +1318,10 @@ static void swap_range_free(struct swap_info_struct *si, unsigned long offset,
>         void (*swap_slot_free_notify)(struct block_device *, unsigned long);
>         unsigned int i;
>
> -       for (i = 0; i < nr_entries; i++)
> -               zswap_invalidate(si->type, offset + i);
> +       if (!zswap_never_enabled()) {
> +               for (i = 0; i < nr_entries; i++)
> +                       zswap_invalidate(si->type, offset + i);
> +       }

What if we add the check in zswap_invalidate()? I understand we'd
avoid the loop here, which is nice, but I wonder if it's actually a
measurable difference.

If we keep it in zswap_invalidate(), we can probably also skip patch 3?


>
>         if (si->flags & SWP_BLKDEV)
>                 swap_slot_free_notify =
> --
> 2.55.0
>
>

Re: [PATCH v2 4/4] mm: zswap: skip zswap_invalidate() in swap_range_free() when zswap is unused

[Kefeng Wang]

On 9/10/2026 6:29 PM, Yosry Ahmed wrote:
> On Thu, Sep 10, 2026 at 2:55 AM Kefeng Wang <wangkefeng.wang@huawei.com> wrote:
>>
>> swap_range_free() calls zswap_invalidate() for every slot being
>> freed, even when zswap has never been enabled. Guard the loop with
>> zswap_never_enabled() to skip it.
>>
>> Signed-off-by: Kefeng Wang <wangkefeng.wang@huawei.com>
>> ---
>>   mm/swapfile.c | 6 ++++--
>>   1 file changed, 4 insertions(+), 2 deletions(-)
>>
>> diff --git a/mm/swapfile.c b/mm/swapfile.c
>> index ba71905e4d46..505592051924 100644
>> --- a/mm/swapfile.c
>> +++ b/mm/swapfile.c
>> @@ -1318,8 +1318,10 @@ static void swap_range_free(struct swap_info_struct *si, unsigned long offset,
>>          void (*swap_slot_free_notify)(struct block_device *, unsigned long);
>>          unsigned int i;
>>
>> -       for (i = 0; i < nr_entries; i++)
>> -               zswap_invalidate(si->type, offset + i);
>> +       if (!zswap_never_enabled()) {
>> +               for (i = 0; i < nr_entries; i++)
>> +                       zswap_invalidate(si->type, offset + i);
>> +       }
> 
> What if we add the check in zswap_invalidate()? I understand we'd
> avoid the loop here, which is nice, but I wonder if it's actually a
> measurable difference.
> 

Skipping useless instructions is always a good thing.

> If we keep it in zswap_invalidate(), we can probably also skip patch 3?

Maybe add nr_entries to zswap_invalidate() and check
zswap_never_enabled() in it.

diff --git a/include/linux/zswap.h b/include/linux/zswap.h
index 463bdee5c1e1..313c2f1b6c2e 100644
--- a/include/linux/zswap.h
+++ b/include/linux/zswap.h
@@ -27,7 +27,7 @@ struct zswap_lruvec_state {
  unsigned long zswap_total_pages(void);
  bool zswap_store(struct folio *folio);
  int zswap_load(struct folio *folio);
-void zswap_invalidate(int type, pgoff_t offset);
+void zswap_invalidate(int type, pgoff_t offset, unsigned int nr_entries);
  int zswap_swapon(int type, unsigned long nr_pages);
  void zswap_swapoff(int type);
  void zswap_memcg_offline_cleanup(struct mem_cgroup *memcg);
@@ -49,7 +49,7 @@ static inline int zswap_load(struct folio *folio)
         return -ENOENT;
  }

-static inline void zswap_invalidate(int type, pgoff_t offset) {}
+static inline void zswap_invalidate(int type, pgoff_t offset, unsigned 
int nr_entries) {}
  static inline int zswap_swapon(int type, unsigned long nr_pages)
  {
         return 0;
diff --git a/mm/swapfile.c b/mm/swapfile.c
index 505592051924..891379c95a01 100644
--- a/mm/swapfile.c
+++ b/mm/swapfile.c
@@ -1316,12 +1316,8 @@ static void swap_range_free(struct 
swap_info_struct *si, unsigned long offset,
  {
         unsigned long end = offset + nr_entries - 1;
         void (*swap_slot_free_notify)(struct block_device *, unsigned 
long);
-       unsigned int i;

-       if (!zswap_never_enabled()) {
-               for (i = 0; i < nr_entries; i++)
-                       zswap_invalidate(si->type, offset + i);
-       }
+       zswap_invalidate(si->type, offset, nr_entries);

         if (si->flags & SWP_BLKDEV)
                 swap_slot_free_notify =
diff --git a/mm/zswap.c b/mm/zswap.c
index 6197aa71e33c..0e72cc9a5415 100644
--- a/mm/zswap.c
+++ b/mm/zswap.c
@@ -1549,13 +1549,8 @@ bool zswap_store(struct folio *folio)
          * offsets corresponding to each page of the folio. Otherwise,
          * writeback could overwrite the new data in the swapfile.
          */
-       if (!ret) {
-               unsigned type = swp_type(swp);
-               pgoff_t offset = swp_offset(swp);
-
-               for (index = 0; index < nr_pages; ++index)
-                       zswap_invalidate(type, offset + index);
-       }
+       if (!ret)
+               zswap_invalidate(swp_type(swp), swp_offset(swp), nr_pages);

         return ret;
  }
@@ -1661,17 +1656,25 @@ int zswap_load(struct folio *folio)
         return 0;
  }

-void zswap_invalidate(int type, pgoff_t offset)
+void zswap_invalidate(int type, pgoff_t offset, int nr_entries)
  {
-       struct xarray *tree = zswap_tree(type, offset);
         struct zswap_entry *entry;
+       struct xarray *tree;
+       int i;

-       if (xa_empty(tree))
+       if (!zswap_never_enabled())
                 return;

-       entry = xa_erase(tree, offset);
-       if (entry)
-               zswap_entry_free(entry);
+       for (i = 0; i < nr_entries; ++i) {
+               tree = zswap_tree(type, offset + i);
+
+               if (xa_empty(tree))
+                       return;
+
+               entry = xa_erase(tree, offset + i);
+               if (entry)
+                       zswap_entry_free(entry);
+       }
  }

If no objections, I will refresh all the patches.

Re: [PATCH v2 4/4] mm: zswap: skip zswap_invalidate() in swap_range_free() when zswap is unused

[Yosry Ahmed]

On Thu, Sep 10, 2026 at 4:10 AM Kefeng Wang <wangkefeng.wang@huawei.com> wrote:
>
>
>
> On 9/10/2026 6:29 PM, Yosry Ahmed wrote:
> > On Thu, Sep 10, 2026 at 2:55 AM Kefeng Wang <wangkefeng.wang@huawei.com> wrote:
> >>
> >> swap_range_free() calls zswap_invalidate() for every slot being
> >> freed, even when zswap has never been enabled. Guard the loop with
> >> zswap_never_enabled() to skip it.
> >>
> >> Signed-off-by: Kefeng Wang <wangkefeng.wang@huawei.com>
> >> ---
> >>   mm/swapfile.c | 6 ++++--
> >>   1 file changed, 4 insertions(+), 2 deletions(-)
> >>
> >> diff --git a/mm/swapfile.c b/mm/swapfile.c
> >> index ba71905e4d46..505592051924 100644
> >> --- a/mm/swapfile.c
> >> +++ b/mm/swapfile.c
> >> @@ -1318,8 +1318,10 @@ static void swap_range_free(struct swap_info_struct *si, unsigned long offset,
> >>          void (*swap_slot_free_notify)(struct block_device *, unsigned long);
> >>          unsigned int i;
> >>
> >> -       for (i = 0; i < nr_entries; i++)
> >> -               zswap_invalidate(si->type, offset + i);
> >> +       if (!zswap_never_enabled()) {
> >> +               for (i = 0; i < nr_entries; i++)
> >> +                       zswap_invalidate(si->type, offset + i);
> >> +       }
> >
> > What if we add the check in zswap_invalidate()? I understand we'd
> > avoid the loop here, which is nice, but I wonder if it's actually a
> > measurable difference.
> >
>
> Skipping useless instructions is always a good thing.
>
> > If we keep it in zswap_invalidate(), we can probably also skip patch 3?
>
> Maybe add nr_entries to zswap_invalidate() and check
> zswap_never_enabled() in it.
>
> diff --git a/include/linux/zswap.h b/include/linux/zswap.h
> index 463bdee5c1e1..313c2f1b6c2e 100644
> --- a/include/linux/zswap.h
> +++ b/include/linux/zswap.h
> @@ -27,7 +27,7 @@ struct zswap_lruvec_state {
>   unsigned long zswap_total_pages(void);
>   bool zswap_store(struct folio *folio);
>   int zswap_load(struct folio *folio);
> -void zswap_invalidate(int type, pgoff_t offset);
> +void zswap_invalidate(int type, pgoff_t offset, unsigned int nr_entries);
>   int zswap_swapon(int type, unsigned long nr_pages);
>   void zswap_swapoff(int type);
>   void zswap_memcg_offline_cleanup(struct mem_cgroup *memcg);
> @@ -49,7 +49,7 @@ static inline int zswap_load(struct folio *folio)
>          return -ENOENT;
>   }
>
> -static inline void zswap_invalidate(int type, pgoff_t offset) {}
> +static inline void zswap_invalidate(int type, pgoff_t offset, unsigned
> int nr_entries) {}
>   static inline int zswap_swapon(int type, unsigned long nr_pages)
>   {
>          return 0;
> diff --git a/mm/swapfile.c b/mm/swapfile.c
> index 505592051924..891379c95a01 100644
> --- a/mm/swapfile.c
> +++ b/mm/swapfile.c
> @@ -1316,12 +1316,8 @@ static void swap_range_free(struct
> swap_info_struct *si, unsigned long offset,
>   {
>          unsigned long end = offset + nr_entries - 1;
>          void (*swap_slot_free_notify)(struct block_device *, unsigned
> long);
> -       unsigned int i;
>
> -       if (!zswap_never_enabled()) {
> -               for (i = 0; i < nr_entries; i++)
> -                       zswap_invalidate(si->type, offset + i);
> -       }
> +       zswap_invalidate(si->type, offset, nr_entries);
>
>          if (si->flags & SWP_BLKDEV)
>                  swap_slot_free_notify =
> diff --git a/mm/zswap.c b/mm/zswap.c
> index 6197aa71e33c..0e72cc9a5415 100644
> --- a/mm/zswap.c
> +++ b/mm/zswap.c
> @@ -1549,13 +1549,8 @@ bool zswap_store(struct folio *folio)
>           * offsets corresponding to each page of the folio. Otherwise,
>           * writeback could overwrite the new data in the swapfile.
>           */
> -       if (!ret) {
> -               unsigned type = swp_type(swp);
> -               pgoff_t offset = swp_offset(swp);
> -
> -               for (index = 0; index < nr_pages; ++index)
> -                       zswap_invalidate(type, offset + index);
> -       }
> +       if (!ret)
> +               zswap_invalidate(swp_type(swp), swp_offset(swp), nr_pages);
>
>          return ret;
>   }
> @@ -1661,17 +1656,25 @@ int zswap_load(struct folio *folio)
>          return 0;
>   }
>
> -void zswap_invalidate(int type, pgoff_t offset)
> +void zswap_invalidate(int type, pgoff_t offset, int nr_entries)
>   {
> -       struct xarray *tree = zswap_tree(type, offset);
>          struct zswap_entry *entry;
> +       struct xarray *tree;
> +       int i;
>
> -       if (xa_empty(tree))
> +       if (!zswap_never_enabled())
>                  return;
>
> -       entry = xa_erase(tree, offset);
> -       if (entry)
> -               zswap_entry_free(entry);
> +       for (i = 0; i < nr_entries; ++i) {
> +               tree = zswap_tree(type, offset + i);
> +
> +               if (xa_empty(tree))
> +                       return;
> +
> +               entry = xa_erase(tree, offset + i);
> +               if (entry)
> +                       zswap_entry_free(entry);
> +       }
>   }
>
> If no objections, I will refresh all the patches.

Yeah I think that's a good idea, even for the zswap_store() path.

Re: [PATCH v2 1/4] mm: zswap: pass type and offset to zswap_invalidate() directly

[Nhat Pham]

On Thu, Sep 10, 2026 at 2:54 AM Kefeng Wang <wangkefeng.wang@huawei.com> wrote:
>
> zswap_invalidate() takes a swp_entry_t only to unpack it right back
> into type and offset.  Both callers already have those values, so
> pass them directly and drop the swp_entry()/swp_type()/swp_offset()
> round-trip.
>
> Signed-off-by: Kefeng Wang <wangkefeng.wang@huawei.com>

Acked-by: Nhat Pham <nphamcs@gmail.com>

Re: [PATCH v2 1/4] mm: zswap: pass type and offset to zswap_invalidate() directly

[Johannes Weiner]

On Thu, Sep 10, 2026 at 05:54:46PM +0800, Kefeng Wang wrote:
> zswap_invalidate() takes a swp_entry_t only to unpack it right back
> into type and offset.  Both callers already have those values, so
> pass them directly and drop the swp_entry()/swp_type()/swp_offset()
> round-trip.
> 
> Signed-off-by: Kefeng Wang <wangkefeng.wang@huawei.com>

Noice.

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

Re: [PATCH v2 2/4] mm: zswap: reuse zswap_invalidate() in zswap_store() check_old path

[Johannes Weiner]

On Thu, Sep 10, 2026 at 05:54:47PM +0800, Kefeng Wang wrote:
> The check_old path in zswap_store() open-codes the same per-slot
> xarray lookup and entry free that zswap_invalidate() already does.
> Reuse zswap_invalidate() whose xa_empty() check skips empty per-area
> trees to avoid unnecessary xarray lookup and code duplication.
> 
> Signed-off-by: Kefeng Wang <wangkefeng.wang@huawei.com>

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