net/mlx5e: Improve MTT/KSM alignment

Make mlx5e_mpwrq_mtts_per_wqe take into account that KSM requires
smaller alignment than MTT.

Ensure that there is always an even amount of MTTs in a UMR WQE, so that
complete octwords are formed, and no garbage is mapped.

Drop extra alignment in MLX5_MTT_OCTW that may cause setting too big
ucseg->xlt_octowords, also leading to mapping garbage.

Generalize some calculations by introducing the MLX5_OCTWORD constant.
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: Jakub Kicinski <kuba@kernel.org>

drivers/net/ethernet/mellanox/mlx5/core/en.h

@@ -109,12 +109,8 @@ struct page_pool;
 #define MLX5_MPWRQ_MAX_PAGES_PER_WQE \
 	rounddown_pow_of_two(MLX5_UMR_MAX_MTT_SPACE / sizeof(struct mlx5_mtt))
 
-#define MLX5_ALIGN_MTTS(mtts)		(ALIGN(mtts, 8))
-#define MLX5_ALIGNED_MTTS_OCTW(mtts)	((mtts) / 2)
-#define MLX5_MTT_OCTW(mtts)		(MLX5_ALIGNED_MTTS_OCTW(MLX5_ALIGN_MTTS(mtts)))
-#define MLX5_KSM_OCTW(ksms)             (ksms)
 #define MLX5E_MAX_RQ_NUM_MTTS	\
-	(ALIGN_DOWN(U16_MAX, 4) * 2) /* So that MLX5_MTT_OCTW(num_mtts) fits into u16 */
+	(ALIGN_DOWN(U16_MAX, 4) * 2) /* Fits into u16 and aligned by WQEBB. */
 #define MLX5E_MAX_RQ_NUM_KSMS (U16_MAX - 1) /* So that num_ksms fits into u16. */
 #define MLX5E_ORDER2_MAX_PACKET_MTU (order_base_2(10 * 1024))
 

drivers/net/ethernet/mellanox/mlx5/core/en/params.c

@@ -91,6 +91,13 @@ u8 mlx5e_mpwrq_pages_per_wqe(struct mlx5_core_dev *mdev, u8 page_shift,
 
 	pages_per_wqe = log_wqe_sz > page_shift ? (1 << (log_wqe_sz - page_shift)) : 1;
 
+	/* Two MTTs are needed to form an octword. The number of MTTs is encoded
+	 * in octwords in a UMR WQE, so we need at least two to avoid mapping
+	 * garbage addresses.
+	 */
+	if (WARN_ON_ONCE(pages_per_wqe < 2 && umr_mode == MLX5E_MPWRQ_UMR_MODE_ALIGNED))
+		pages_per_wqe = 2;
+
 	/* Sanity check for further calculations to succeed. */
 	BUILD_BUG_ON(MLX5_MPWRQ_MAX_PAGES_PER_WQE > 64);
 	if (WARN_ON_ONCE(pages_per_wqe > MLX5_MPWRQ_MAX_PAGES_PER_WQE))
@@ -131,7 +138,8 @@ u8 mlx5e_mpwrq_mtts_per_wqe(struct mlx5_core_dev *mdev, u8 page_shift,
 	 * MTU. These oversize packets are dropped by the driver at a later
 	 * stage.
 	 */
-	return MLX5_ALIGN_MTTS(pages_per_wqe + 1);
+	return ALIGN(pages_per_wqe + 1,
+		     MLX5_SEND_WQE_BB / mlx5e_mpwrq_umr_entry_size(umr_mode));
 }
 
 u32 mlx5e_mpwrq_max_num_entries(struct mlx5_core_dev *mdev,

drivers/net/ethernet/mellanox/mlx5/core/en/xsk/rx.c

@@ -66,9 +66,10 @@ int mlx5e_xsk_alloc_rx_mpwqe(struct mlx5e_rq *rq, u16 ix)
 	umr_wqe->ctrl.opmod_idx_opcode =
 		cpu_to_be32((icosq->pc << MLX5_WQE_CTRL_WQE_INDEX_SHIFT) | MLX5_OPCODE_UMR);
 
+	/* Optimized for speed: keep in sync with mlx5e_mpwrq_umr_entry_size. */
 	offset = ix * rq->mpwqe.mtts_per_wqe;
 	if (likely(rq->mpwqe.umr_mode == MLX5E_MPWRQ_UMR_MODE_ALIGNED))
-		offset = MLX5_ALIGNED_MTTS_OCTW(offset);
+		offset = offset * sizeof(struct mlx5_mtt) / MLX5_OCTWORD;
 	umr_wqe->uctrl.xlt_offset = cpu_to_be16(offset);
 
 	icosq->db.wqe_info[pi] = (struct mlx5e_icosq_wqe_info) {

drivers/net/ethernet/mellanox/mlx5/core/en_main.c

@@ -205,14 +205,11 @@ static void mlx5e_disable_blocking_events(struct mlx5e_priv *priv)
 
 static u16 mlx5e_mpwrq_umr_octowords(u32 entries, enum mlx5e_mpwrq_umr_mode umr_mode)
 {
-	switch (umr_mode) {
-	case MLX5E_MPWRQ_UMR_MODE_ALIGNED:
-		return MLX5_MTT_OCTW(entries);
-	case MLX5E_MPWRQ_UMR_MODE_UNALIGNED:
-		return MLX5_KSM_OCTW(entries);
-	}
-	WARN_ONCE(1, "MPWRQ UMR mode %d is not known\n", umr_mode);
-	return 0;
+	u8 umr_entry_size = mlx5e_mpwrq_umr_entry_size(umr_mode);
+
+	WARN_ON_ONCE(entries * umr_entry_size % MLX5_OCTWORD);
+
+	return entries * umr_entry_size / MLX5_OCTWORD;
 }
 
 static inline void mlx5e_build_umr_wqe(struct mlx5e_rq *rq,

drivers/net/ethernet/mellanox/mlx5/core/en_rx.c

@@ -682,7 +682,7 @@ static int mlx5e_alloc_rx_mpwqe(struct mlx5e_rq *rq, u16 ix)
 		cpu_to_be32((sq->pc << MLX5_WQE_CTRL_WQE_INDEX_SHIFT) |
 			    MLX5_OPCODE_UMR);
 
-	offset = MLX5_ALIGNED_MTTS_OCTW(ix * rq->mpwqe.mtts_per_wqe);
+	offset = (ix * rq->mpwqe.mtts_per_wqe) * sizeof(struct mlx5_mtt) / MLX5_OCTWORD;
 	umr_wqe->uctrl.xlt_offset = cpu_to_be16(offset);
 
 	sq->db.wqe_info[pi] = (struct mlx5e_icosq_wqe_info) {