When a struct smaller than a cacheline has an instance that is not aligned
to cache block size, it could happen that data can be spread across two
cachelines.
For memcg_stock this could mean the need to fetch and get cache-exclusivity
in 2 cachelines instead of 1 when we bounce the cacheline between local
cpu functions and drain_all_stock(), which does remote read/write.
This could also mean some false-sharing costs being paid due to the
cacheline being shared between 2 unrelated structures.
Avoid this issue by getting memcg_stock cacheline-aligned.
Signed-off-by: Leonardo Bras <leobras@xxxxxxxxxx>
---
mm/memcontrol.c | 3 ++-
1 file changed, 2 insertions(+), 1 deletion(-)
diff --git a/mm/memcontrol.c b/mm/memcontrol.c
index ab457f0394ab6..f8e86b88b3c7a 100644
--- a/mm/memcontrol.c
+++ b/mm/memcontrol.c
@@ -2188,7 +2188,8 @@ struct memcg_stock_pcp {
unsigned long flags;
#define FLUSHING_CACHED_CHARGE 0
};
-static DEFINE_PER_CPU(struct memcg_stock_pcp, memcg_stock) = {
+
+static DEFINE_PER_CPU_SHARED_ALIGNED(struct memcg_stock_pcp, memcg_stock) = {
.stock_lock = INIT_LOCAL_LOCK(stock_lock),
};
static DEFINE_MUTEX(percpu_charge_mutex);
--
2.39.1
