Enable jumbo frame reception with default mbuf data room size by chaining multiple mbufs when packet exceeds single mbuf tailroom.
Scatter Rx is only enabled when RTE_ETH_RX_OFFLOAD_SCATTER is requested. Packets are dropped if they exceed single mbuf size and scatter is not enabled, or if mbuf allocation fails during chaining. Error counter rx_dropped_pkts tracks all drops. This allows receiving 9KB jumbo frames using standard 2KB mbufs, chaining ~5 segments per jumbo packet. Signed-off-by: Sriram Yagnaraman <[email protected]> --- doc/guides/rel_notes/release_26_03.rst | 5 ++ drivers/net/af_packet/rte_eth_af_packet.c | 79 +++++++++++++++++------ 2 files changed, 63 insertions(+), 21 deletions(-) diff --git a/doc/guides/rel_notes/release_26_03.rst b/doc/guides/rel_notes/release_26_03.rst index 9649185133..1048a6b595 100644 --- a/doc/guides/rel_notes/release_26_03.rst +++ b/doc/guides/rel_notes/release_26_03.rst @@ -65,6 +65,11 @@ New Features * Added support for V4000 Krackan2e. +* **Updated AF_PACKET ethernet driver.** + + * Added support for multi-segment mbuf reception to handle jumbo frames + with standard mbuf sizes when scatter Rx offload is enabled. + * **Updated CESNET nfb ethernet driver.** * The timestamp value has been updated to make it usable. diff --git a/drivers/net/af_packet/rte_eth_af_packet.c b/drivers/net/af_packet/rte_eth_af_packet.c index e132dc387b..13daa02fe0 100644 --- a/drivers/net/af_packet/rte_eth_af_packet.c +++ b/drivers/net/af_packet/rte_eth_af_packet.c @@ -56,6 +56,7 @@ struct __rte_cache_aligned pkt_rx_queue { uint16_t in_port; uint8_t vlan_strip; uint8_t timestamp_offloading; + uint8_t scatter_enabled; volatile unsigned long rx_pkts; volatile unsigned long rx_bytes; @@ -125,12 +126,13 @@ 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; + uint16_t pkt_len, data_len, remaining; if (unlikely(nb_pkts == 0)) return 0; @@ -154,20 +156,49 @@ eth_af_packet_rx(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts) break; } - /* drop packets that won't fit in the mbuf */ - if (ppd->tp_snaplen > rte_pktmbuf_tailroom(mbuf)) { + pkt_len = ppd->tp_snaplen; + pbuf = (uint8_t *)ppd + ppd->tp_mac; + + /* drop packets that won't fit in single mbuf if scatter not enabled */ + if (!pkt_q->scatter_enabled && pkt_len > rte_pktmbuf_tailroom(mbuf)) { rte_pktmbuf_free(mbuf); - ppd->tp_status = TP_STATUS_KERNEL; - if (++framenum >= framecount) - framenum = 0; pkt_q->rx_dropped_pkts++; - continue; + goto release_frame; + } + + /* copy first segment */ + 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; + mbuf->nb_segs = 1; + remaining = pkt_len - data_len; + pbuf += data_len; + prev = mbuf; + + /* chain additional segments if needed */ + while (remaining > 0) { + seg = rte_pktmbuf_alloc(pkt_q->mb_pool); + if (unlikely(seg == NULL)) { + rte_pktmbuf_free(mbuf); + pkt_q->rx_dropped_pkts++; + goto release_frame; + } + + /* Remove headroom to maximize data space in chained segments */ + rte_pktmbuf_prepend(seg, rte_pktmbuf_headroom(seg)); + + 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++; } - /* packet will fit in the mbuf, go ahead and receive it */ - rte_pktmbuf_pkt_len(mbuf) = rte_pktmbuf_data_len(mbuf) = ppd->tp_snaplen; - pbuf = (uint8_t *) ppd + ppd->tp_mac; - memcpy(rte_pktmbuf_mtod(mbuf, void *), pbuf, rte_pktmbuf_data_len(mbuf)); + rte_pktmbuf_pkt_len(mbuf) = pkt_len; /* check for vlan info */ if (ppd->tp_status & TP_STATUS_VLAN_VALID) { @@ -188,16 +219,18 @@ eth_af_packet_rx(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts) mbuf->ol_flags |= timestamp_dynflag; } - /* 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; + bufs[num_rx] = mbuf; num_rx++; num_rx_bytes += mbuf->pkt_len; + +release_frame: + /* release incoming frame and advance ring buffer */ + ppd->tp_status = TP_STATUS_KERNEL; + if (++framenum >= framecount) + framenum = 0; } pkt_q->framenum = framenum; pkt_q->rx_pkts += num_rx; @@ -412,7 +445,8 @@ 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; } @@ -579,26 +613,29 @@ eth_rx_queue_setup(struct rte_eth_dev *dev, struct pmd_internals *internals = dev->data->dev_private; struct pkt_rx_queue *pkt_q = &internals->rx_queue[rx_queue_id]; unsigned int buf_size, data_size; + bool scatter_enabled; pkt_q->mb_pool = mb_pool; - /* Now get the space available for data in the mbuf */ buf_size = rte_pktmbuf_data_room_size(pkt_q->mb_pool) - RTE_PKTMBUF_HEADROOM; data_size = internals->req.tp_frame_size; data_size -= TPACKET2_HDRLEN - sizeof(struct sockaddr_ll); - if (data_size > buf_size) { + scatter_enabled = !!(dev->data->dev_conf.rxmode.offloads & RTE_ETH_RX_OFFLOAD_SCATTER); + + if (!scatter_enabled && data_size > buf_size) { PMD_LOG(ERR, - "%s: %d bytes will not fit in mbuf (%d bytes)", + "%s: %d bytes will not fit in mbuf (%d bytes), enable scatter offload", dev->device->name, data_size, buf_size); - return -ENOMEM; + return -EINVAL; } dev->data->rx_queues[rx_queue_id] = pkt_q; pkt_q->in_port = dev->data->port_id; pkt_q->vlan_strip = internals->vlan_strip; pkt_q->timestamp_offloading = internals->timestamp_offloading; + pkt_q->scatter_enabled = scatter_enabled; return 0; } -- 2.43.7
