Re: [PATCH v4 4/9] mm/shmem, swap: tidy up swap entry splitting
From: Baolin Wang
Date: Sat Jul 05 2025 - 23:35:58 EST
On 2025/7/5 02:17, Kairui Song wrote:
From: Kairui Song <kasong@xxxxxxxxxxx>
Instead of keeping different paths of splitting the entry before the
swap in start, move the entry splitting after the swapin has put
the folio in swap cache (or set the SWAP_HAS_CACHE bit). This way
we only need one place and one unified way to split the large entry.
Whenever swapin brought in a folio smaller than the shmem swap entry,
split the entry and recalculate the entry and index for verification.
This removes duplicated codes and function calls, reduces LOC,
and the split is less racy as it's guarded by swap cache now. So it
will have a lower chance of repeated faults due to raced split.
The compiler is also able to optimize the coder further:
bloat-o-meter results with GCC 14:
With DEBUG_SECTION_MISMATCH (-fno-inline-functions-called-once):
./scripts/bloat-o-meter mm/shmem.o.old mm/shmem.o
add/remove: 0/0 grow/shrink: 0/1 up/down: 0/-82 (-82)
Function old new delta
shmem_swapin_folio 2361 2279 -82
Total: Before=33151, After=33069, chg -0.25%
With !DEBUG_SECTION_MISMATCH:
./scripts/bloat-o-meter mm/shmem.o.old mm/shmem.o
add/remove: 0/1 grow/shrink: 1/0 up/down: 949/-750 (199)
Function old new delta
shmem_swapin_folio 2878 3827 +949
shmem_split_large_entry.isra 750 - -750
Total: Before=33086, After=33285, chg +0.60%
Since shmem_split_large_entry is only called in one place now. The
compiler will either generate more compact code, or inlined it for
better performance.
Signed-off-by: Kairui Song <kasong@xxxxxxxxxxx>
---
mm/shmem.c | 53 +++++++++++++++++++++--------------------------------
1 file changed, 21 insertions(+), 32 deletions(-)
diff --git a/mm/shmem.c b/mm/shmem.c
index e43becfa04b3..217264315842 100644
--- a/mm/shmem.c
+++ b/mm/shmem.c
@@ -2266,14 +2266,15 @@ static int shmem_swapin_folio(struct inode *inode, pgoff_t index,
struct address_space *mapping = inode->i_mapping;
struct mm_struct *fault_mm = vma ? vma->vm_mm : NULL;
struct shmem_inode_info *info = SHMEM_I(inode);
+ swp_entry_t swap, index_entry;
struct swap_info_struct *si;
struct folio *folio = NULL;
bool skip_swapcache = false;
- swp_entry_t swap;
int error, nr_pages, order, split_order;
+ pgoff_t offset;
VM_BUG_ON(!*foliop || !xa_is_value(*foliop));
- swap = radix_to_swp_entry(*foliop);
+ swap = index_entry = radix_to_swp_entry(*foliop);
*foliop = NULL;
if (is_poisoned_swp_entry(swap))
@@ -2321,46 +2322,35 @@ static int shmem_swapin_folio(struct inode *inode, pgoff_t index,
}
/*
- * Now swap device can only swap in order 0 folio, then we
- * should split the large swap entry stored in the pagecache
- * if necessary.
- */
- split_order = shmem_split_large_entry(inode, index, swap, gfp);
- if (split_order < 0) {
- error = split_order;
- goto failed;
- }
-
- /*
- * If the large swap entry has already been split, it is
+ * Now swap device can only swap in order 0 folio, it is
* necessary to recalculate the new swap entry based on
- * the old order alignment.
+ * the offset, as the swapin index might be unalgined.
*/
- if (split_order > 0) {
- pgoff_t offset = index - round_down(index, 1 << split_order);
-
+ if (order) {
+ offset = index - round_down(index, 1 << order);
swap = swp_entry(swp_type(swap), swp_offset(swap) + offset);
}
- /* Here we actually start the io */
folio = shmem_swapin_cluster(swap, gfp, info, index);
if (!folio) {
error = -ENOMEM;
goto failed;
}
- } else if (order > folio_order(folio)) {
+ }
+alloced:
+ if (order > folio_order(folio)) {
/*
- * Swap readahead may swap in order 0 folios into swapcache
+ * Swapin may get smaller folios due to various reasons:
+ * It may fallback to order 0 due to memory pressure or race,
+ * swap readahead may swap in order 0 folios into swapcache
* asynchronously, while the shmem mapping can still stores
* large swap entries. In such cases, we should split the
* large swap entry to prevent possible data corruption.
*/
- split_order = shmem_split_large_entry(inode, index, swap, gfp);
+ split_order = shmem_split_large_entry(inode, index, index_entry, gfp);
if (split_order < 0) {
- folio_put(folio);
- folio = NULL;
error = split_order;
- goto failed;
+ goto failed_nolock;
}
/*
@@ -2369,15 +2359,13 @@ static int shmem_swapin_folio(struct inode *inode, pgoff_t index,
* the old order alignment.
*/
if (split_order > 0) {
- pgoff_t offset = index - round_down(index, 1 << split_order);
-
+ offset = index - round_down(index, 1 << split_order);
swap = swp_entry(swp_type(swap), swp_offset(swap) + offset);
Obviously, you should use the original swap value 'index_entry' to
calculate the new swap value.
With the following fix, you can add:
Reviewed-by: Baolin Wang <baolin.wang@xxxxxxxxxxxxxxxxx>
Tested-by: Baolin Wang <baolin.wang@xxxxxxxxxxxxxxxxx>
diff --git a/mm/shmem.c b/mm/shmem.c
index d530df550f7f..1e8422ac863e 100644
--- a/mm/shmem.c
+++ b/mm/shmem.c
@@ -2361,7 +2361,7 @@ static int shmem_swapin_folio(struct inode *inode,
pgoff_t index,
*/
if (split_order > 0) {
offset = index - round_down(index, 1 <<
split_order);
- swap = swp_entry(swp_type(swap),
swp_offset(swap) + offset);
+ swap = swp_entry(swp_type(swap),
swp_offset(index_swap) + offset);
}
} else if (order < folio_order(folio)) {
swap.val = round_down(swap.val, 1 << folio_order(folio));