net/mlx5e: kTLS, Limit DUMP wqe size

HW expects the data size in DUMP WQEs to be up to MTU.
Make sure they are in range.

We elevate the frag page refcount by 'n-1', in addition to the
one obtained in tx_sync_info_get(), having an overall of 'n'
references. We bulk increments by using a single page_ref_add()
command, to optimize perfermance.
The refcounts are released one by one, by the corresponding completions.

Fixes: d2ead1f ("net/mlx5e: Add kTLS TX HW offload support")
Signed-off-by: Tariq Toukan <[email protected]>
Reviewed-by: Eran Ben Elisha <[email protected]>
Signed-off-by: Saeed Mahameed <[email protected]>

Author: Tariq Toukan

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

@@ -410,6 +410,7 @@ struct mlx5e_txqsq {
 	struct device             *pdev;
 	__be32                     mkey_be;
 	unsigned long              state;
+	unsigned int               hw_mtu;
 	struct hwtstamp_config    *tstamp;
 	struct mlx5_clock         *clock;
 

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

@@ -15,15 +15,14 @@
 #else
 /* TLS offload requires additional stop_room for:
  *  - a resync SKB.
- * kTLS offload requires additional stop_room for:
- * - static params WQE,
- * - progress params WQE, and
- * - resync DUMP per frag.
+ * kTLS offload requires fixed additional stop_room for:
+ * - a static params WQE, and a progress params WQE.
+ * The additional MTU-depending room for the resync DUMP WQEs
+ * will be calculated and added in runtime.
  */
 #define MLX5E_SQ_TLS_ROOM  \
 	(MLX5_SEND_WQE_MAX_WQEBBS + \
-	 MLX5E_KTLS_STATIC_WQEBBS + MLX5E_KTLS_PROGRESS_WQEBBS + \
-	 MAX_SKB_FRAGS * MLX5E_KTLS_DUMP_WQEBBS)
+	 MLX5E_KTLS_STATIC_WQEBBS + MLX5E_KTLS_PROGRESS_WQEBBS)
 #endif
 
 #define INL_HDR_START_SZ (sizeof(((struct mlx5_wqe_eth_seg *)NULL)->inline_hdr.start))

drivers/net/ethernet/mellanox/mlx5/core/en_accel/ktls.h

@@ -94,7 +94,16 @@ struct sk_buff *mlx5e_ktls_handle_tx_skb(struct net_device *netdev,
 void mlx5e_ktls_tx_handle_resync_dump_comp(struct mlx5e_txqsq *sq,
 					   struct mlx5e_tx_wqe_info *wi,
 					   u32 *dma_fifo_cc);
-
+static inline u8
+mlx5e_ktls_dumps_num_wqebbs(struct mlx5e_txqsq *sq, unsigned int nfrags,
+			    unsigned int sync_len)
+{
+	/* Given the MTU and sync_len, calculates an upper bound for the
+	 * number of WQEBBs needed for the TX resync DUMP WQEs of a record.
+	 */
+	return MLX5E_KTLS_DUMP_WQEBBS *
+		(nfrags + DIV_ROUND_UP(sync_len, sq->hw_mtu));
+}
 #else
 
 static inline void mlx5e_ktls_build_netdev(struct mlx5e_priv *priv)

drivers/net/ethernet/mellanox/mlx5/core/en_accel/ktls_tx.c

@@ -373,7 +373,7 @@ mlx5e_ktls_tx_handle_ooo(struct mlx5e_ktls_offload_context_tx *priv_tx,
 		return skb;
 	}
 
-	num_wqebbs = info.nr_frags * MLX5E_KTLS_DUMP_WQEBBS;
+	num_wqebbs = mlx5e_ktls_dumps_num_wqebbs(sq, info.nr_frags, info.sync_len);
 	pi = mlx5_wq_cyc_ctr2ix(wq, sq->pc);
 	contig_wqebbs_room = mlx5_wq_cyc_get_contig_wqebbs(wq, pi);
 
@@ -382,14 +382,40 @@ mlx5e_ktls_tx_handle_ooo(struct mlx5e_ktls_offload_context_tx *priv_tx,
 
 	tx_post_resync_params(sq, priv_tx, info.rcd_sn);
 
-	for (; i < info.nr_frags; i++)
-		if (tx_post_resync_dump(sq, &info.frags[i], priv_tx->tisn, !i))
-			goto err_out;
+	for (; i < info.nr_frags; i++) {
+		unsigned int orig_fsz, frag_offset = 0, n = 0;
+		skb_frag_t *f = &info.frags[i];
+
+		orig_fsz = skb_frag_size(f);
+
+		do {
+			bool fence = !(i || frag_offset);
+			unsigned int fsz;
+
+			n++;
+			fsz = min_t(unsigned int, sq->hw_mtu, orig_fsz - frag_offset);
+			skb_frag_size_set(f, fsz);
+			if (tx_post_resync_dump(sq, f, priv_tx->tisn, fence)) {
+				page_ref_add(skb_frag_page(f), n - 1);
+				goto err_out;
+			}
+
+			skb_frag_off_add(f, fsz);
+			frag_offset += fsz;
+		} while (frag_offset < orig_fsz);
+
+		page_ref_add(skb_frag_page(f), n - 1);
+	}
 
 	return skb;
 
 err_out:
 	for (; i < info.nr_frags; i++)
+		/* The put_page() here undoes the page ref obtained in tx_sync_info_get().
+		 * Page refs obtained for the DUMP WQEs above (by page_ref_add) will be
+		 * released only upon their completions (or in mlx5e_free_txqsq_descs,
+		 * if channel closes).
+		 */
 		put_page(skb_frag_page(&info.frags[i]));
 
 	dev_kfree_skb_any(skb);

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

@@ -1128,17 +1128,22 @@ static int mlx5e_alloc_txqsq(struct mlx5e_channel *c,
 	sq->txq_ix    = txq_ix;
 	sq->uar_map   = mdev->mlx5e_res.bfreg.map;
 	sq->min_inline_mode = params->tx_min_inline_mode;
+	sq->hw_mtu    = MLX5E_SW2HW_MTU(params, params->sw_mtu);
 	sq->stats     = &c->priv->channel_stats[c->ix].sq[tc];
 	sq->stop_room = MLX5E_SQ_STOP_ROOM;
 	INIT_WORK(&sq->recover_work, mlx5e_tx_err_cqe_work);
 	if (!MLX5_CAP_ETH(mdev, wqe_vlan_insert))
 		set_bit(MLX5E_SQ_STATE_VLAN_NEED_L2_INLINE, &sq->state);
 	if (MLX5_IPSEC_DEV(c->priv->mdev))
 		set_bit(MLX5E_SQ_STATE_IPSEC, &sq->state);
+#ifdef CONFIG_MLX5_EN_TLS
 	if (mlx5_accel_is_tls_device(c->priv->mdev)) {
 		set_bit(MLX5E_SQ_STATE_TLS, &sq->state);
-		sq->stop_room += MLX5E_SQ_TLS_ROOM;
+		sq->stop_room += MLX5E_SQ_TLS_ROOM +
+			mlx5e_ktls_dumps_num_wqebbs(sq, MAX_SKB_FRAGS,
+						    TLS_MAX_PAYLOAD_SIZE);
 	}
+#endif
 
 	param->wq.db_numa_node = cpu_to_node(c->cpu);
 	err = mlx5_wq_cyc_create(mdev, &param->wq, sqc_wq, wq, &sq->wq_ctrl);