@@ -447,6 +447,17 @@ mode to allow application to tune the per-socket maximum iteration for
 better throughput and less frequency of send syscall.
 Allowed range is [32, xs->tx->nentries].
 
+XDP_GENERIC_XMIT_BATCH
+----------------------
+
+It provides an option that allows application to use batch xmit in the copy
+mode. Batch process tries to allocate a certain number skbs through bulk
+mechanism first and then send them out at one time, minimizing the number
+of grabbing/releasing a few locks (like cache lock and queue lock).
+it normally gains the overall performance improvement as observed by
+xdpsock benchmark, whereas it might increase the latency of per packet.
+The maximum value shouldn't be larger than xs->max_tx_budget.
+
 XDP_STATISTICS getsockopt
 -------------------------
 

@@ -70,6 +70,7 @@ struct xdp_sock {
 	 * preventing other XSKs from being starved.
 	 */
 	u32 tx_budget_spent;
+	u32 generic_xmit_batch;
 
 	/* Statistics */
 	u64 rx_dropped;

@@ -89,6 +90,8 @@ struct xdp_sock {
 	struct mutex mutex;
 	struct xsk_queue *fq_tmp; /* Only as tmp storage before bind */
 	struct xsk_queue *cq_tmp; /* Only as tmp storage before bind */
+	struct sk_buff **skb_cache;
+	struct xdp_desc *desc_batch;
 };
 
 /*

@@ -80,6 +80,7 @@ struct xdp_mmap_offsets {
 #define XDP_STATISTICS			7
 #define XDP_OPTIONS			8
 #define XDP_MAX_TX_SKB_BUDGET		9
+#define XDP_GENERIC_XMIT_BATCH		10
 
 struct xdp_umem_reg {
 	__u64 addr; /* Start of packet data area */

@@ -1122,6 +1122,8 @@ static int xsk_release(struct socket *sock)
 	xskq_destroy(xs->tx);
 	xskq_destroy(xs->fq_tmp);
 	xskq_destroy(xs->cq_tmp);
+	kfree(xs->skb_cache);
+	kvfree(xs->desc_batch);
 
 	sock_orphan(sk);
 	sock->sk = NULL;

@@ -1456,6 +1458,51 @@ static int xsk_setsockopt(struct socket *sock, int level, int optname,
 		WRITE_ONCE(xs->max_tx_budget, budget);
 		return 0;
 	}
+	case XDP_GENERIC_XMIT_BATCH:
+	{
+		struct xdp_desc *descs;
+		struct sk_buff **skbs;
+		unsigned int batch;
+		int ret = 0;
+
+		if (optlen != sizeof(batch))
+			return -EINVAL;
+		if (copy_from_sockptr(&batch, optval, sizeof(batch)))
+			return -EFAULT;
+		if (batch > xs->max_tx_budget)
+			return -EACCES;
+
+		mutex_lock(&xs->mutex);
+		if (!batch) {
+			kfree(xs->skb_cache);
+			kvfree(xs->desc_batch);
+			xs->generic_xmit_batch = 0;
+			goto out;
+		}
+
+		skbs = kmalloc(batch * sizeof(struct sk_buff *), GFP_KERNEL);
+		if (!skbs) {
+			ret = -ENOMEM;
+			goto out;
+		}
+		descs = kvcalloc(batch, sizeof(*xs->desc_batch), GFP_KERNEL);
+		if (!skbs) {
+			kfree(skbs);
+			ret = -ENOMEM;
+			goto out;
+		}
+		if (xs->skb_cache)
+			kfree(xs->skb_cache);
+		if (xs->desc_batch)
+			kvfree(xs->desc_batch);
+
+		xs->skb_cache = skbs;
+		xs->desc_batch = descs;
+		xs->generic_xmit_batch = batch;
+out:
+		mutex_unlock(&xs->mutex);
+		return ret;
+	}
 	default:
 		break;
 	}

@@ -80,6 +80,7 @@ struct xdp_mmap_offsets {
 #define XDP_STATISTICS			7
 #define XDP_OPTIONS			8
 #define XDP_MAX_TX_SKB_BUDGET		9
+#define XDP_GENERIC_XMIT_BATCH		10
 
 struct xdp_umem_reg {
 	__u64 addr; /* Start of packet data area */