net/mlx5e: Re-order fields of struct mlx5e_xdpsq

In the downstream patch that adds support to XDP_REDIRECT-out,
the XDP xmit frame function doesn't share the same run context as
the NAPI that polls the XDP-SQ completion queue.

Hence, need to re-order the XDP-SQ fields to avoid cacheline
false-sharing.

Take redirect_flush and doorbell out of DB, into separated
cachelines.

Add a cacheline breaker within the stats struct.
Signed-off-by: Tariq Toukan <tariqt@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>

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

@@ -409,22 +409,22 @@ struct mlx5e_xdp_info {
 struct mlx5e_xdpsq {
 	/* data path */
 
-	/* dirtied @rx completion */
+	/* dirtied @completion */
 	u16                        cc;
-	u16                        pc;
+	bool                       redirect_flush;
 
-	struct mlx5e_cq            cq;
+	/* dirtied @xmit */
+	u16                        pc ____cacheline_aligned_in_smp;
+	bool                       doorbell;
 
-	/* write@xmit, read@completion */
-	struct {
-		struct mlx5e_xdp_info     *xdpi;
-		bool                       doorbell;
-		bool                       redirect_flush;
-	} db;
+	struct mlx5e_cq            cq;
 
 	/* read only */
 	struct mlx5_wq_cyc         wq;
 	struct mlx5e_xdpsq_stats  *stats;
+	struct {
+		struct mlx5e_xdp_info     *xdpi;
+	} db;
 	void __iomem              *uar_map;
 	u32                        sqn;
 	struct device             *pdev;

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

@@ -85,7 +85,7 @@ bool mlx5e_xdp_handle(struct mlx5e_rq *rq, struct mlx5e_dma_info *di,
 		if (unlikely(err))
 			goto xdp_abort;
 		__set_bit(MLX5E_RQ_FLAG_XDP_XMIT, rq->flags);
-		rq->xdpsq.db.redirect_flush = true;
+		rq->xdpsq.redirect_flush = true;
 		mlx5e_page_dma_unmap(rq, di);
 		rq->stats->xdp_redirect++;
 		return true;
@@ -124,10 +124,10 @@ bool mlx5e_xmit_xdp_frame(struct mlx5e_xdpsq *sq, struct mlx5e_xdp_info *xdpi)
 	}
 
 	if (unlikely(!mlx5e_wqc_has_room_for(wq, sq->cc, sq->pc, 1))) {
-		if (sq->db.doorbell) {
+		if (sq->doorbell) {
 			/* SQ is full, ring doorbell */
 			mlx5e_xmit_xdp_doorbell(sq);
-			sq->db.doorbell = false;
+			sq->doorbell = false;
 		}
 		stats->full++;
 		return false;
@@ -156,7 +156,7 @@ bool mlx5e_xmit_xdp_frame(struct mlx5e_xdpsq *sq, struct mlx5e_xdp_info *xdpi)
 	sq->db.xdpi[pi] = *xdpi;
 	sq->pc++;
 
-	sq->db.doorbell = true;
+	sq->doorbell = true;
 
 	stats->xmit++;
 	return true;

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

@@ -1201,14 +1201,14 @@ int mlx5e_poll_rx_cq(struct mlx5e_cq *cq, int budget)
 		rq->handle_rx_cqe(rq, cqe);
 	} while ((++work_done < budget) && (cqe = mlx5_cqwq_get_cqe(&cq->wq)));
 
-	if (xdpsq->db.doorbell) {
+	if (xdpsq->doorbell) {
 		mlx5e_xmit_xdp_doorbell(xdpsq);
-		xdpsq->db.doorbell = false;
+		xdpsq->doorbell = false;
 	}
 
-	if (xdpsq->db.redirect_flush) {
+	if (xdpsq->redirect_flush) {
 		xdp_do_flush_map();
-		xdpsq->db.redirect_flush = false;
+		xdpsq->redirect_flush = false;
 	}
 
 	mlx5_cqwq_update_db_record(&cq->wq);

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

@@ -230,7 +230,8 @@ struct mlx5e_xdpsq_stats {
 	u64 xmit;
 	u64 full;
 	u64 err;
-	u64 cqes;
+	/* dirtied @completion */
+	u64 cqes ____cacheline_aligned_in_smp;
 };
 
 struct mlx5e_ch_stats {