[PATCH] net/af_packet: add multi-segment mbuf support for jumbo frames

[Sriram Yagnaraman]

Enable jumbo frame reception with default mbuf data room size by
chaining multiple mbufs when packet exceeds single mbuf tailroom.

The RX path now:
- Copies first segment up to mbuf tailroom
- Allocates and chains additional segments for remaining data
- Properly maintains nb_segs, next pointers, and pkt_len
- Drops packets only on mbuf allocation failure

This allows receiving 9KB jumbo frames using standard 2KB mbufs,
chaining ~5 segments per jumbo packet. No need for custom mbuf
pools with large data room.

Advertises RTE_ETH_RX_OFFLOAD_SCATTER capability.

Depends-on: series-34567 ("net/af_packet: fix MTU handling and add
jumbo frame support")

Note: Implementation generated with AI assistance.

Signed-off-by: Sriram Yagnaraman <sriram.yagnaraman at ericsson.com>
---
 drivers/net/af_packet/rte_eth_af_packet.c | 53 ++++++++++++++++++-----
 1 file changed, 43 insertions(+), 10 deletions(-)

diff --git a/drivers/net/af_packet/rte_eth_af_packet.c b/drivers/net/af_packet/rte_eth_af_packet.c
index d9fafd4..a18bc41 100644
--- a/drivers/net/af_packet/rte_eth_af_packet.c
+++ b/drivers/net/af_packet/rte_eth_af_packet.c
@@ -121,12 +121,13 @@ eth_af_packet_rx(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
 {
 	unsigned i;
 	struct tpacket2_hdr *ppd;
-	struct rte_mbuf *mbuf;
+	struct rte_mbuf *mbuf, *seg, *prev;
 	uint8_t *pbuf;
 	struct pkt_rx_queue *pkt_q = queue;
 	uint16_t num_rx = 0;
 	unsigned long num_rx_bytes = 0;
 	unsigned int framecount, framenum;
+	uint32_t pkt_len, data_len, remaining;
 
 	if (unlikely(nb_pkts == 0))
 		return 0;
@@ -148,10 +149,36 @@ eth_af_packet_rx(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
 		if (unlikely(mbuf == NULL))
 			break;
 
-		/* packet will fit in the mbuf, go ahead and receive it */
-		rte_pktmbuf_pkt_len(mbuf) = rte_pktmbuf_data_len(mbuf) = ppd->tp_snaplen;
+		pkt_len = ppd->tp_snaplen;
 		pbuf = (uint8_t *) ppd + ppd->tp_mac;
-		memcpy(rte_pktmbuf_mtod(mbuf, void *), pbuf, rte_pktmbuf_data_len(mbuf));
+
+		/* copy data to mbuf chain */
+		data_len = RTE_MIN(pkt_len, rte_pktmbuf_tailroom(mbuf));
+		memcpy(rte_pktmbuf_mtod(mbuf, void *), pbuf, data_len);
+		rte_pktmbuf_data_len(mbuf) = data_len;
+		remaining = pkt_len - data_len;
+		pbuf += data_len;
+		prev = mbuf;
+
+		while (remaining > 0) {
+			seg = rte_pktmbuf_alloc(pkt_q->mb_pool);
+			if (unlikely(seg == NULL)) {
+				rte_pktmbuf_free(mbuf);
+				goto drop;
+			}
+
+			data_len = RTE_MIN(remaining, rte_pktmbuf_tailroom(seg));
+			memcpy(rte_pktmbuf_mtod(seg, void *), pbuf, data_len);
+			rte_pktmbuf_data_len(seg) = data_len;
+			pbuf += data_len;
+			remaining -= data_len;
+
+			prev->next = seg;
+			prev = seg;
+			mbuf->nb_segs++;
+		}
+
+		rte_pktmbuf_pkt_len(mbuf) = pkt_len;
 
 		/* check for vlan info */
 		if (ppd->tp_status & TP_STATUS_VLAN_VALID) {
@@ -172,16 +199,21 @@ eth_af_packet_rx(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)
 			mbuf->ol_flags |= timestamp_dynflag;
 		}
 
+drop:
 		/* release incoming frame and advance ring buffer */
 		ppd->tp_status = TP_STATUS_KERNEL;
 		if (++framenum >= framecount)
 			framenum = 0;
-		mbuf->port = pkt_q->in_port;
 
-		/* account for the receive frame */
-		bufs[i] = mbuf;
-		num_rx++;
-		num_rx_bytes += mbuf->pkt_len;
+		if (mbuf != NULL) {
+			mbuf->port = pkt_q->in_port;
+			/* account for the receive frame */
+			bufs[num_rx] = mbuf;
+			num_rx++;
+			num_rx_bytes += mbuf->pkt_len;
+		} else {
+			pkt_q->rx_dropped_pkts++;
+		}
 	}
 	pkt_q->framenum = framenum;
 	pkt_q->rx_pkts += num_rx;
@@ -411,7 +443,8 @@ eth_dev_info(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
 	dev_info->tx_offload_capa = RTE_ETH_TX_OFFLOAD_MULTI_SEGS |
 		RTE_ETH_TX_OFFLOAD_VLAN_INSERT;
 	dev_info->rx_offload_capa = RTE_ETH_RX_OFFLOAD_VLAN_STRIP |
-		RTE_ETH_RX_OFFLOAD_TIMESTAMP;
+		RTE_ETH_RX_OFFLOAD_TIMESTAMP |
+		RTE_ETH_RX_OFFLOAD_SCATTER;
 
 	return 0;
 }
-- 
2.43.7