Since we do not set FLAG_SKB_NO_MERGE for skbuff_fclone_cache,
the current skbuff_fclone_cache maybe not really allocated, it maybe
used an exist old kmem_cache. In NUMA, the fclone allocated by
alloc_skb_fclone() maybe in remote node.

So set FLAG_SKB_NO_MERGE for skbuff_fclone_cache to fix it.

Signed-off-by: Huang Shijie <redacted>
---
v1 --> v2:
       set FLAG_SKB_NO_MERGE for skbuff_fclone_cache in initialization.

v1: https://lkml.org/lkml/2024/2/20/121	
---
 net/core/skbuff.c | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)

diff --git a/net/core/skbuff.c b/net/core/skbuff.c
index 1f918e602bc4..5e3e130fb57a 100644
--- a/net/core/skbuff.c
+++ b/net/core/skbuff.c

@@ -5013,7 +5013,8 @@ void __init skb_init(void)
 	skbuff_fclone_cache = kmem_cache_create("skbuff_fclone_cache",
 						sizeof(struct sk_buff_fclones),
 						0,
-						SLAB_HWCACHE_ALIGN|SLAB_PANIC,
+						SLAB_HWCACHE_ALIGN|SLAB_PANIC|
+						FLAG_SKB_NO_MERGE,
 						NULL);
 	/* usercopy should only access first SKB_SMALL_HEAD_HEADROOM bytes.
 	 * struct skb_shared_info is located at the end of skb->head,

-- 
2.40.1