From: Maxim Mikityanskiy maximmi@nvidia.com
[ Upstream commit 677e78c8d44f326a73a77d71acf3a49ea562c1d9 ]
Before commit 76c31e5f7585 ("net/mlx5e: Use FW limitation for max MPW WQEBBs"), the maximum size of MPWQE in WQEBBs was hardcoded as a driver constant. That commit started using the firmware capability that can further limit the size, however, it unintentionally changed a few things:
1. The calculation of MLX5E_MAX_KLM_PER_WQE used the size in DS, which was replaced by the size in WQEBBs, making the resulting value 4 times smaller.
2. MLX5E_TX_MPW_MAX_WQEBBS used to be aligned to the cache line size (either 64 or 128 bytes, i.e. 1 or 2 WQEBBs), but it's no longer the case if the firmware capability is smaller than the driver maximum.
Fix both issues by using the correct units for MLX5E_MAX_KLM_PER_WQE and by aligning mlx5e_get_sw_max_sq_mpw_wqebbs after taking the minimum.
Besides fixing the arithmetics in calculation of MLX5E_MAX_KLM_PER_WQE, also use appropriate constants: `size of BSF * num of DS per WQEBB * number of WQEBBs` (the calculation before the blamed commit) doesn't make much sense to calculate the WQE size in bytes, so just use `size of WQEBB * number of WQEBBs`.
While at it, replace the types that hold the number of WQEBBs by u8. These values don't exceed 16, and it allows to fill holes in two structs.
Fixes: 76c31e5f7585 ("net/mlx5e: Use FW limitation for max MPW WQEBBs") Signed-off-by: Maxim Mikityanskiy maximmi@nvidia.com Reviewed-by: Tariq Toukan tariqt@nvidia.com Signed-off-by: Saeed Mahameed saeedm@nvidia.com Signed-off-by: Sasha Levin sashal@kernel.org --- drivers/net/ethernet/mellanox/mlx5/core/en.h | 19 ++++++++++--------- 1 file changed, 10 insertions(+), 9 deletions(-)
diff --git a/drivers/net/ethernet/mellanox/mlx5/core/en.h b/drivers/net/ethernet/mellanox/mlx5/core/en.h index f794ffaf1e04..29b10ef787b9 100644 --- a/drivers/net/ethernet/mellanox/mlx5/core/en.h +++ b/drivers/net/ethernet/mellanox/mlx5/core/en.h @@ -174,8 +174,8 @@ struct page_pool; ALIGN_DOWN(MLX5E_KLM_MAX_ENTRIES_PER_WQE(wqe_size), MLX5_UMR_KLM_ALIGNMENT)
#define MLX5E_MAX_KLM_PER_WQE(mdev) \ - MLX5E_KLM_ENTRIES_PER_WQE(mlx5e_get_sw_max_sq_mpw_wqebbs(mlx5e_get_max_sq_wqebbs(mdev)) \ - << MLX5_MKEY_BSF_OCTO_SIZE) + MLX5E_KLM_ENTRIES_PER_WQE(MLX5_SEND_WQE_BB * \ + mlx5e_get_sw_max_sq_mpw_wqebbs(mlx5e_get_max_sq_wqebbs(mdev)))
#define MLX5E_MSG_LEVEL NETIF_MSG_LINK
@@ -233,7 +233,7 @@ static inline u16 mlx5e_get_max_sq_wqebbs(struct mlx5_core_dev *mdev) MLX5_CAP_GEN(mdev, max_wqe_sz_sq) / MLX5_SEND_WQE_BB); }
-static inline u16 mlx5e_get_sw_max_sq_mpw_wqebbs(u16 max_sq_wqebbs) +static inline u8 mlx5e_get_sw_max_sq_mpw_wqebbs(u8 max_sq_wqebbs) { /* The return value will be multiplied by MLX5_SEND_WQEBB_NUM_DS. * Since max_sq_wqebbs may be up to MLX5_SEND_WQE_MAX_WQEBBS == 16, @@ -242,11 +242,12 @@ static inline u16 mlx5e_get_sw_max_sq_mpw_wqebbs(u16 max_sq_wqebbs) * than MLX5_SEND_WQE_MAX_WQEBBS to let a full-session WQE be * cache-aligned. */ -#if L1_CACHE_BYTES < 128 - return min_t(u16, max_sq_wqebbs, MLX5_SEND_WQE_MAX_WQEBBS - 1); -#else - return min_t(u16, max_sq_wqebbs, MLX5_SEND_WQE_MAX_WQEBBS - 2); + u8 wqebbs = min_t(u8, max_sq_wqebbs, MLX5_SEND_WQE_MAX_WQEBBS - 1); + +#if L1_CACHE_BYTES >= 128 + wqebbs = ALIGN_DOWN(wqebbs, 2); #endif + return wqebbs; }
struct mlx5e_tx_wqe { @@ -455,7 +456,7 @@ struct mlx5e_txqsq { struct netdev_queue *txq; u32 sqn; u16 stop_room; - u16 max_sq_mpw_wqebbs; + u8 max_sq_mpw_wqebbs; u8 min_inline_mode; struct device *pdev; __be32 mkey_be; @@ -570,7 +571,7 @@ struct mlx5e_xdpsq { struct device *pdev; __be32 mkey_be; u16 stop_room; - u16 max_sq_mpw_wqebbs; + u8 max_sq_mpw_wqebbs; u8 min_inline_mode; unsigned long state; unsigned int hw_mtu;