[PATCH v1 2/4] net/nbl: optimize mbuf headroom usage in packet transmission

[Dimon Zhao]

Only use mbuf headroom when mbuf->refcnt == 1,
indicating exclusive ownership. For shared mbufs (refcnt > 1),
use next desc instead to avoid data corruption.

This prevents modifying shared buffers that may be used by other
processing contexts.

Implementation details:
- Check mbuf->refcnt before using headroom
- For refcnt == 1: use existing headroom (performance optimal)
- For refcnt > 1: use next desc

Fixes: e5fc1f78c78c ("net/nbl: support Tx and Rx burst")
Cc: stable at dpdk.org

Signed-off-by: Dimon Zhao <dimon.zhao at nebula-matrix.com>
---
 drivers/net/nbl/nbl_hw/nbl_txrx.c | 5 ++++-
 1 file changed, 4 insertions(+), 1 deletion(-)

diff --git a/drivers/net/nbl/nbl_hw/nbl_txrx.c b/drivers/net/nbl/nbl_hw/nbl_txrx.c
index 103f56c3a7..e906f0707e 100644
--- a/drivers/net/nbl/nbl_hw/nbl_txrx.c
+++ b/drivers/net/nbl/nbl_hw/nbl_txrx.c
@@ -514,7 +514,10 @@ nbl_res_txrx_xmit_pkts(void *tx_queue, struct rte_mbuf **tx_pkts, u16 nb_pkts, u
 			tx_extend_len = required_headroom + sizeof(struct rte_ether_hdr);
 		}
 
-		if (rte_pktmbuf_headroom(tx_pkt) >= required_headroom) {
+		if (rte_mbuf_refcnt_read(tx_pkt) == 1 &&
+		    RTE_MBUF_DIRECT(tx_pkt) &&
+		    tx_pkt->nb_segs == 1 &&
+		    rte_pktmbuf_headroom(tx_pkt) >= required_headroom) {
 			can_push = 1;
 			u = rte_pktmbuf_mtod_offset(tx_pkt, union nbl_tx_extend_head *,
 						    -required_headroom);
-- 
2.34.1