From 2579543f604daf459bd8c939c52c6fc43b235f43 Mon Sep 17 00:00:00 2001 From: Matan Azrad Date: Mon, 29 Jul 2019 11:53:27 +0000 Subject: [PATCH] net/mlx5: handle LRO packets in regular Rx queue When LRO offload is configured in Rx queue, the HW may coalesce TCP packets from same TCP connection into single packet. In this case the SW should fix the relevant packet headers because the HW doesn't update them according to the new created packet characteristics but provides the update values in the CQE. Add update header code to the regular Rx burst function to support LRO feature. Make sure the first mbuf has enough space to include each TCP header, otherwise the header update may cross mbufs what complicates the operation too match. Signed-off-by: Matan Azrad Acked-by: Viacheslav Ovsiienko --- doc/guides/nics/mlx5.rst | 4 +++- drivers/net/mlx5/mlx5_rxq.c | 20 +++++++++++++++++--- drivers/net/mlx5/mlx5_rxtx.c | 17 +++++++++++++++++ 3 files changed, 37 insertions(+), 4 deletions(-) diff --git a/doc/guides/nics/mlx5.rst b/doc/guides/nics/mlx5.rst index cd550f46ed..6f0c382ea0 100644 --- a/doc/guides/nics/mlx5.rst +++ b/doc/guides/nics/mlx5.rst @@ -165,7 +165,9 @@ Limitations - LRO: - - scatter_fcs is disabled when LRO is configured. + - KEEP_CRC offload cannot be supported with LRO. + - The first mbuf length, without head-room, must be big enough to include the + TCP header (122B). Statistics ---------- diff --git a/drivers/net/mlx5/mlx5_rxq.c b/drivers/net/mlx5/mlx5_rxq.c index e96bb1e378..3705d07a68 100644 --- a/drivers/net/mlx5/mlx5_rxq.c +++ b/drivers/net/mlx5/mlx5_rxq.c @@ -1541,6 +1541,11 @@ exit: #define MLX5_MAX_TCP_HDR_OFFSET ((unsigned int)(sizeof(struct rte_ether_hdr) + \ sizeof(struct rte_vlan_hdr) * 2 + \ sizeof(struct rte_ipv6_hdr))) +#define MAX_TCP_OPTION_SIZE 40u +#define MLX5_MAX_LRO_HEADER_FIX ((unsigned int)(MLX5_MAX_TCP_HDR_OFFSET + \ + sizeof(struct rte_tcp_hdr) + \ + MAX_TCP_OPTION_SIZE)) + /** * Adjust the maximum LRO massage size. * @@ -1607,6 +1612,7 @@ mlx5_rxq_new(struct rte_eth_dev *dev, uint16_t idx, uint16_t desc, unsigned int non_scatter_min_mbuf_size = max_rx_pkt_len + RTE_PKTMBUF_HEADROOM; unsigned int max_lro_size = 0; + unsigned int first_mb_free_size = mb_len - RTE_PKTMBUF_HEADROOM; if (non_scatter_min_mbuf_size > mb_len && !(offloads & DEV_RX_OFFLOAD_SCATTER)) { @@ -1670,8 +1676,8 @@ mlx5_rxq_new(struct rte_eth_dev *dev, uint16_t idx, uint16_t desc, config->mprq.min_stride_size_n); tmpl->rxq.strd_shift_en = MLX5_MPRQ_TWO_BYTE_SHIFT; tmpl->rxq.strd_headroom_en = strd_headroom_en; - tmpl->rxq.mprq_max_memcpy_len = RTE_MIN(mb_len - - RTE_PKTMBUF_HEADROOM, config->mprq.max_memcpy_len); + tmpl->rxq.mprq_max_memcpy_len = RTE_MIN(first_mb_free_size, + config->mprq.max_memcpy_len); max_lro_size = RTE_MIN(max_rx_pkt_len, (1u << tmpl->rxq.strd_num_n) * (1u << tmpl->rxq.strd_sz_n)); @@ -1680,13 +1686,21 @@ mlx5_rxq_new(struct rte_eth_dev *dev, uint16_t idx, uint16_t desc, " strd_num_n = %u, strd_sz_n = %u", dev->data->port_id, idx, tmpl->rxq.strd_num_n, tmpl->rxq.strd_sz_n); - } else if (max_rx_pkt_len <= (mb_len - RTE_PKTMBUF_HEADROOM)) { + } else if (max_rx_pkt_len <= first_mb_free_size) { tmpl->rxq.sges_n = 0; max_lro_size = max_rx_pkt_len; } else if (offloads & DEV_RX_OFFLOAD_SCATTER) { unsigned int size = non_scatter_min_mbuf_size; unsigned int sges_n; + if (mlx5_lro_on(dev) && first_mb_free_size < + MLX5_MAX_LRO_HEADER_FIX) { + DRV_LOG(ERR, "Not enough space in the first segment(%u)" + " to include the max header size(%u) for LRO", + first_mb_free_size, MLX5_MAX_LRO_HEADER_FIX); + rte_errno = ENOTSUP; + goto error; + } /* * Determine the number of SGEs needed for a full packet * and round it to the next power of two. diff --git a/drivers/net/mlx5/mlx5_rxtx.c b/drivers/net/mlx5/mlx5_rxtx.c index 003eefdd40..6627b54c9b 100644 --- a/drivers/net/mlx5/mlx5_rxtx.c +++ b/drivers/net/mlx5/mlx5_rxtx.c @@ -107,6 +107,16 @@ static int mlx5_queue_state_modify(struct rte_eth_dev *dev, struct mlx5_mp_arg_queue_state_modify *sm); +static inline void +mlx5_lro_update_tcp_hdr(struct rte_tcp_hdr *restrict tcp, + volatile struct mlx5_cqe *restrict cqe, + uint32_t phcsum); + +static inline void +mlx5_lro_update_hdr(uint8_t *restrict padd, + volatile struct mlx5_cqe *restrict cqe, + uint32_t len); + uint32_t mlx5_ptype_table[] __rte_cache_aligned = { [0xff] = RTE_PTYPE_ALL_MASK, /* Last entry for errored packet. */ }; @@ -1323,6 +1333,13 @@ mlx5_rx_burst(void *dpdk_rxq, struct rte_mbuf **pkts, uint16_t pkts_n) if (rxq->crc_present) len -= RTE_ETHER_CRC_LEN; PKT_LEN(pkt) = len; + if (cqe->lro_num_seg > 1) { + mlx5_lro_update_hdr + (rte_pktmbuf_mtod(pkt, uint8_t *), cqe, + len); + pkt->ol_flags |= PKT_RX_LRO; + pkt->tso_segsz = len / cqe->lro_num_seg; + } } DATA_LEN(rep) = DATA_LEN(seg); PKT_LEN(rep) = PKT_LEN(seg); -- 2.20.1