[<prev] [next>] [thread-next>] [day] [month] [year] [list]
Message-ID: <20241101034647.51590-1-csander@purestorage.com>
Date: Thu, 31 Oct 2024 21:46:39 -0600
From: Caleb Sander Mateos <csander@...estorage.com>
To: Saeed Mahameed <saeedm@...dia.com>,
Leon Romanovsky <leon@...nel.org>,
Tariq Toukan <tariqt@...dia.com>,
Andrew Lunn <andrew+netdev@...n.ch>,
"David S. Miller" <davem@...emloft.net>,
Eric Dumazet <edumazet@...gle.com>,
Jakub Kicinski <kuba@...nel.org>,
Paolo Abeni <pabeni@...hat.com>
Cc: Caleb Sander Mateos <csander@...estorage.com>,
netdev@...r.kernel.org,
linux-rdma@...r.kernel.org,
linux-kernel@...r.kernel.org
Subject: [PATCH net-next 1/2] mlx5/core: avoid memory barrier in eq_update_ci()
The memory barrier in eq_update_ci() after the doorbell write is a
significant hot spot in mlx5_eq_comp_int(). Under heavy TCP load, we see
3% of CPU time spent on the mfence. As explained in [1], this barrier is
only needed to preserve the ordering of writes to the doorbell register.
Use writel() instead of __raw_writel() for the doorbell write to provide
this ordering without the need for a full memory barrier.
memory-barriers.txt guarantees MMIO writes using writel() appear to the
device in the same order they were made. On strongly-ordered
architectures like x86, writel() adds no overhead to __raw_writel();
both translate into a single store instruction. Removing the mb() avoids
the costly mfence instruction.
[1]: https://lore.kernel.org/netdev/CALzJLG8af0SMfA1C8U8r_Fddb_ZQhvEZd6=2a97dOoBcgLA0xg@mail.gmail.com/
Signed-off-by: Caleb Sander Mateos <csander@...estorage.com>
---
drivers/net/ethernet/mellanox/mlx5/core/lib/eq.h | 5 ++---
1 file changed, 2 insertions(+), 3 deletions(-)
diff --git a/drivers/net/ethernet/mellanox/mlx5/core/lib/eq.h b/drivers/net/ethernet/mellanox/mlx5/core/lib/eq.h
index 4b7f7131c560..f03736711343 100644
--- a/drivers/net/ethernet/mellanox/mlx5/core/lib/eq.h
+++ b/drivers/net/ethernet/mellanox/mlx5/core/lib/eq.h
@@ -68,13 +68,12 @@ static inline struct mlx5_eqe *next_eqe_sw(struct mlx5_eq *eq)
static inline void eq_update_ci(struct mlx5_eq *eq, int arm)
{
__be32 __iomem *addr = eq->doorbell + (arm ? 0 : 2);
u32 val = (eq->cons_index & 0xffffff) | (eq->eqn << 24);
- __raw_writel((__force u32)cpu_to_be32(val), addr);
- /* We still want ordering, just not swabbing, so add a barrier */
- mb();
+ /* Ensure ordering of consecutive doorbell writes */
+ writel((__force u32)cpu_to_be32(val), addr);
}
int mlx5_eq_table_init(struct mlx5_core_dev *dev);
void mlx5_eq_table_cleanup(struct mlx5_core_dev *dev);
int mlx5_eq_table_create(struct mlx5_core_dev *dev);
--
2.45.2
Powered by blists - more mailing lists