4 * Copyright 2017 6WIND S.A.
5 * Copyright 2017 Mellanox.
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
11 * * Redistributions of source code must retain the above copyright
12 * notice, this list of conditions and the following disclaimer.
13 * * Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in
15 * the documentation and/or other materials provided with the
17 * * Neither the name of 6WIND S.A. nor the names of its
18 * contributors may be used to endorse or promote products derived
19 * from this software without specific prior written permission.
21 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
22 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
23 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
24 * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
25 * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
26 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
27 * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
28 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
29 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
30 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
31 * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
40 /* ISO C doesn't support unnamed structs/unions, disabling -pedantic. */
42 #pragma GCC diagnostic ignored "-Wpedantic"
44 #include <infiniband/verbs.h>
45 #include <infiniband/mlx5dv.h>
47 #pragma GCC diagnostic error "-Wpedantic"
51 #include <rte_mempool.h>
52 #include <rte_prefetch.h>
55 #include "mlx5_utils.h"
56 #include "mlx5_rxtx.h"
57 #include "mlx5_rxtx_vec.h"
58 #include "mlx5_autoconf.h"
59 #include "mlx5_defs.h"
62 #if defined RTE_ARCH_X86_64
63 #include "mlx5_rxtx_vec_sse.h"
64 #elif defined RTE_ARCH_ARM64
65 #include "mlx5_rxtx_vec_neon.h"
67 #error "This should not be compiled if SIMD instructions are not supported."
71 * Count the number of continuous single segment packets.
74 * Pointer to array of packets.
79 * Number of continuous single segment packets.
81 static inline unsigned int
82 txq_check_multiseg(struct rte_mbuf **pkts, uint16_t pkts_n)
88 /* Count the number of continuous single segment packets. */
89 for (pos = 0; pos < pkts_n; ++pos)
90 if (NB_SEGS(pkts[pos]) > 1)
96 * Count the number of packets having same ol_flags and calculate cs_flags.
99 * Pointer to TX queue structure.
101 * Pointer to array of packets.
105 * Pointer of flags to be returned.
108 * Number of packets having same ol_flags.
110 static inline unsigned int
111 txq_calc_offload(struct mlx5_txq_data *txq, struct rte_mbuf **pkts,
112 uint16_t pkts_n, uint8_t *cs_flags)
115 const uint64_t ol_mask =
116 PKT_TX_IP_CKSUM | PKT_TX_TCP_CKSUM |
117 PKT_TX_UDP_CKSUM | PKT_TX_TUNNEL_GRE |
118 PKT_TX_TUNNEL_VXLAN | PKT_TX_OUTER_IP_CKSUM;
122 /* Count the number of packets having same ol_flags. */
123 for (pos = 1; pos < pkts_n; ++pos)
124 if ((pkts[pos]->ol_flags ^ pkts[0]->ol_flags) & ol_mask)
126 *cs_flags = txq_ol_cksum_to_cs(txq, pkts[0]);
131 * DPDK callback for vectorized TX.
134 * Generic pointer to TX queue structure.
136 * Packets to transmit.
138 * Number of packets in array.
141 * Number of packets successfully transmitted (<= pkts_n).
144 mlx5_tx_burst_raw_vec(void *dpdk_txq, struct rte_mbuf **pkts,
147 struct mlx5_txq_data *txq = (struct mlx5_txq_data *)dpdk_txq;
150 while (pkts_n > nb_tx) {
154 n = RTE_MIN((uint16_t)(pkts_n - nb_tx), MLX5_VPMD_TX_MAX_BURST);
155 ret = txq_burst_v(txq, &pkts[nb_tx], n, 0);
164 * DPDK callback for vectorized TX with multi-seg packets and offload.
167 * Generic pointer to TX queue structure.
169 * Packets to transmit.
171 * Number of packets in array.
174 * Number of packets successfully transmitted (<= pkts_n).
177 mlx5_tx_burst_vec(void *dpdk_txq, struct rte_mbuf **pkts, uint16_t pkts_n)
179 struct mlx5_txq_data *txq = (struct mlx5_txq_data *)dpdk_txq;
182 while (pkts_n > nb_tx) {
183 uint8_t cs_flags = 0;
187 /* Transmit multi-seg packets in the head of pkts list. */
188 if (!(txq->flags & ETH_TXQ_FLAGS_NOMULTSEGS) &&
189 NB_SEGS(pkts[nb_tx]) > 1)
190 nb_tx += txq_scatter_v(txq,
193 n = RTE_MIN((uint16_t)(pkts_n - nb_tx), MLX5_VPMD_TX_MAX_BURST);
194 if (!(txq->flags & ETH_TXQ_FLAGS_NOMULTSEGS))
195 n = txq_check_multiseg(&pkts[nb_tx], n);
196 if (!(txq->flags & ETH_TXQ_FLAGS_NOOFFLOADS))
197 n = txq_calc_offload(txq, &pkts[nb_tx], n, &cs_flags);
198 ret = txq_burst_v(txq, &pkts[nb_tx], n, cs_flags);
207 * Skip error packets.
210 * Pointer to RX queue structure.
212 * Array to store received packets.
214 * Maximum number of packets in array.
217 * Number of packets successfully received (<= pkts_n).
220 rxq_handle_pending_error(struct mlx5_rxq_data *rxq, struct rte_mbuf **pkts,
225 #ifdef MLX5_PMD_SOFT_COUNTERS
226 uint32_t err_bytes = 0;
229 for (i = 0; i < pkts_n; ++i) {
230 struct rte_mbuf *pkt = pkts[i];
232 if (pkt->packet_type == RTE_PTYPE_ALL_MASK) {
233 #ifdef MLX5_PMD_SOFT_COUNTERS
234 err_bytes += PKT_LEN(pkt);
236 rte_pktmbuf_free_seg(pkt);
241 rxq->stats.idropped += (pkts_n - n);
242 #ifdef MLX5_PMD_SOFT_COUNTERS
243 /* Correct counters of errored completions. */
244 rxq->stats.ipackets -= (pkts_n - n);
245 rxq->stats.ibytes -= err_bytes;
247 rxq->pending_err = 0;
252 * DPDK callback for vectorized RX.
255 * Generic pointer to RX queue structure.
257 * Array to store received packets.
259 * Maximum number of packets in array.
262 * Number of packets successfully received (<= pkts_n).
265 mlx5_rx_burst_vec(void *dpdk_rxq, struct rte_mbuf **pkts, uint16_t pkts_n)
267 struct mlx5_rxq_data *rxq = dpdk_rxq;
270 nb_rx = rxq_burst_v(rxq, pkts, pkts_n);
271 if (unlikely(rxq->pending_err))
272 nb_rx = rxq_handle_pending_error(rxq, pkts, nb_rx);
277 * Check Tx queue flags are set for raw vectorized Tx.
280 * Pointer to private structure.
283 * 1 if supported, negative errno value if not.
285 int __attribute__((cold))
286 priv_check_raw_vec_tx_support(struct priv *priv)
290 /* All the configured queues should support. */
291 for (i = 0; i < priv->txqs_n; ++i) {
292 struct mlx5_txq_data *txq = (*priv->txqs)[i];
294 if (!(txq->flags & ETH_TXQ_FLAGS_NOMULTSEGS) ||
295 !(txq->flags & ETH_TXQ_FLAGS_NOOFFLOADS))
298 if (i != priv->txqs_n)
304 * Check a device can support vectorized TX.
307 * Pointer to private structure.
310 * 1 if supported, negative errno value if not.
312 int __attribute__((cold))
313 priv_check_vec_tx_support(struct priv *priv)
315 if (!priv->tx_vec_en ||
316 priv->txqs_n > MLX5_VPMD_MIN_TXQS ||
317 priv->mps != MLX5_MPW_ENHANCED ||
324 * Check a RX queue can support vectorized RX.
327 * Pointer to RX queue.
330 * 1 if supported, negative errno value if not.
332 int __attribute__((cold))
333 rxq_check_vec_support(struct mlx5_rxq_data *rxq)
335 struct mlx5_rxq_ctrl *ctrl =
336 container_of(rxq, struct mlx5_rxq_ctrl, rxq);
338 if (!ctrl->priv->rx_vec_en || rxq->sges_n != 0)
344 * Check a device can support vectorized RX.
347 * Pointer to private structure.
350 * 1 if supported, negative errno value if not.
352 int __attribute__((cold))
353 priv_check_vec_rx_support(struct priv *priv)
357 if (!priv->rx_vec_en)
359 /* All the configured queues should support. */
360 for (i = 0; i < priv->rxqs_n; ++i) {
361 struct mlx5_rxq_data *rxq = (*priv->rxqs)[i];
365 if (rxq_check_vec_support(rxq) < 0)
368 if (i != priv->rxqs_n)