4 * Copyright 2015 6WIND S.A.
5 * Copyright 2015 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.
34 #ifndef RTE_PMD_MLX5_RXTX_H_
35 #define RTE_PMD_MLX5_RXTX_H_
41 /* ISO C doesn't support unnamed structs/unions, disabling -pedantic. */
43 #pragma GCC diagnostic ignored "-pedantic"
45 #include <infiniband/verbs.h>
47 #pragma GCC diagnostic error "-pedantic"
50 /* DPDK headers don't like -pedantic. */
52 #pragma GCC diagnostic ignored "-pedantic"
55 #include <rte_mempool.h>
57 #pragma GCC diagnostic error "-pedantic"
60 #include "mlx5_utils.h"
62 #include "mlx5_autoconf.h"
63 #include "mlx5_defs.h"
65 struct mlx5_rxq_stats {
66 unsigned int idx; /**< Mapping index. */
67 #ifdef MLX5_PMD_SOFT_COUNTERS
68 uint64_t ipackets; /**< Total of successfully received packets. */
69 uint64_t ibytes; /**< Total of successfully received bytes. */
71 uint64_t idropped; /**< Total of packets dropped when RX ring full. */
72 uint64_t rx_nombuf; /**< Total of RX mbuf allocation failures. */
75 struct mlx5_txq_stats {
76 unsigned int idx; /**< Mapping index. */
77 #ifdef MLX5_PMD_SOFT_COUNTERS
78 uint64_t opackets; /**< Total of successfully sent packets. */
79 uint64_t obytes; /**< Total of successfully sent bytes. */
81 uint64_t odropped; /**< Total of packets not sent when TX ring full. */
84 /* RX element (scattered packets). */
86 struct ibv_sge sges[MLX5_PMD_SGE_WR_N]; /* Scatter/Gather Elements. */
87 struct rte_mbuf *bufs[MLX5_PMD_SGE_WR_N]; /* SGEs buffers. */
92 struct ibv_sge sge; /* Scatter/Gather Element. */
93 struct rte_mbuf *buf; /* SGE buffer. */
98 /* RX queue descriptor. */
100 struct priv *priv; /* Back pointer to private data. */
101 struct rte_mempool *mp; /* Memory Pool for allocations. */
102 struct ibv_mr *mr; /* Memory Region (for mp). */
103 struct ibv_cq *cq; /* Completion Queue. */
104 struct ibv_exp_wq *wq; /* Work Queue. */
105 struct ibv_exp_wq_family *if_wq; /* WQ burst interface. */
106 struct ibv_exp_cq_family *if_cq; /* CQ interface. */
107 unsigned int port_id; /* Port ID for incoming packets. */
108 unsigned int elts_n; /* (*elts)[] length. */
109 unsigned int elts_head; /* Current index in (*elts)[]. */
111 struct rxq_elt_sp (*sp)[]; /* Scattered RX elements. */
112 struct rxq_elt (*no_sp)[]; /* RX elements. */
114 unsigned int sp:1; /* Use scattered RX elements. */
115 unsigned int csum:1; /* Enable checksum offloading. */
116 unsigned int csum_l2tun:1; /* Same for L2 tunnels. */
117 uint32_t mb_len; /* Length of a mp-issued mbuf. */
118 struct mlx5_rxq_stats stats; /* RX queue counters. */
119 unsigned int socket; /* CPU socket ID for allocations. */
120 struct ibv_exp_res_domain *rd; /* Resource Domain. */
123 /* Hash RX queue types. */
128 #ifdef HAVE_FLOW_SPEC_IPV6
132 #endif /* HAVE_FLOW_SPEC_IPV6 */
136 /* Flow structure with Ethernet specification. It is packed to prevent padding
137 * between attr and spec as this layout is expected by libibverbs. */
138 struct flow_attr_spec_eth {
139 struct ibv_exp_flow_attr attr;
140 struct ibv_exp_flow_spec_eth spec;
141 } __attribute__((packed));
143 /* Define a struct flow_attr_spec_eth object as an array of at least
144 * "size" bytes. Room after the first index is normally used to store
145 * extra flow specifications. */
146 #define FLOW_ATTR_SPEC_ETH(name, size) \
147 struct flow_attr_spec_eth name \
148 [((size) / sizeof(struct flow_attr_spec_eth)) + \
149 !!((size) % sizeof(struct flow_attr_spec_eth))]
151 /* Initialization data for hash RX queue. */
152 struct hash_rxq_init {
153 uint64_t hash_fields; /* Fields that participate in the hash. */
154 uint64_t dpdk_rss_hf; /* Matching DPDK RSS hash fields. */
155 unsigned int flow_priority; /* Flow priority to use. */
158 enum ibv_exp_flow_spec_type type;
161 struct ibv_exp_flow_spec_tcp_udp tcp_udp;
162 struct ibv_exp_flow_spec_ipv4 ipv4;
163 #ifdef HAVE_FLOW_SPEC_IPV6
164 struct ibv_exp_flow_spec_ipv6 ipv6;
165 #endif /* HAVE_FLOW_SPEC_IPV6 */
166 struct ibv_exp_flow_spec_eth eth;
167 } flow_spec; /* Flow specification template. */
168 const struct hash_rxq_init *underlayer; /* Pointer to underlayer. */
171 /* Initialization data for indirection table. */
172 struct ind_table_init {
173 unsigned int max_size; /* Maximum number of WQs. */
174 /* Hash RX queues using this table. */
175 unsigned int hash_types;
176 unsigned int hash_types_n;
179 enum hash_rxq_flow_type {
180 HASH_RXQ_FLOW_TYPE_MAC,
181 HASH_RXQ_FLOW_TYPE_PROMISC,
182 HASH_RXQ_FLOW_TYPE_ALLMULTI,
186 struct priv *priv; /* Back pointer to private data. */
187 struct ibv_qp *qp; /* Hash RX QP. */
188 enum hash_rxq_type type; /* Hash RX queue type. */
189 /* MAC flow steering rules, one per VLAN ID. */
190 struct ibv_exp_flow *mac_flow[MLX5_MAX_MAC_ADDRESSES][MLX5_MAX_VLAN_IDS];
191 struct ibv_exp_flow *promisc_flow; /* Promiscuous flow. */
192 struct ibv_exp_flow *allmulti_flow; /* Multicast flow. */
197 struct rte_mbuf *buf;
200 /* Linear buffer type. It is used when transmitting buffers with too many
201 * segments that do not fit the hardware queue (see max_send_sge).
202 * Extra segments are copied (linearized) in such buffers, replacing the
203 * last SGE during TX.
204 * The size is arbitrary but large enough to hold a jumbo frame with
205 * 8 segments considering mbuf.buf_len is about 2048 bytes. */
206 typedef uint8_t linear_t[16384];
208 /* TX queue descriptor. */
210 struct priv *priv; /* Back pointer to private data. */
212 const struct rte_mempool *mp; /* Cached Memory Pool. */
213 struct ibv_mr *mr; /* Memory Region (for mp). */
214 uint32_t lkey; /* mr->lkey */
215 } mp2mr[MLX5_PMD_TX_MP_CACHE]; /* MP to MR translation table. */
216 struct ibv_cq *cq; /* Completion Queue. */
217 struct ibv_qp *qp; /* Queue Pair. */
218 struct ibv_exp_qp_burst_family *if_qp; /* QP burst interface. */
219 struct ibv_exp_cq_family *if_cq; /* CQ interface. */
220 #if MLX5_PMD_MAX_INLINE > 0
221 uint32_t max_inline; /* Max inline send size <= MLX5_PMD_MAX_INLINE. */
223 unsigned int elts_n; /* (*elts)[] length. */
224 struct txq_elt (*elts)[]; /* TX elements. */
225 unsigned int elts_head; /* Current index in (*elts)[]. */
226 unsigned int elts_tail; /* First element awaiting completion. */
227 unsigned int elts_comp; /* Number of completion requests. */
228 unsigned int elts_comp_cd; /* Countdown for next completion request. */
229 unsigned int elts_comp_cd_init; /* Initial value for countdown. */
230 struct mlx5_txq_stats stats; /* TX queue counters. */
231 linear_t (*elts_linear)[]; /* Linearized buffers. */
232 struct ibv_mr *mr_linear; /* Memory Region for linearized buffers. */
233 unsigned int socket; /* CPU socket ID for allocations. */
234 struct ibv_exp_res_domain *rd; /* Resource Domain. */
239 extern const struct hash_rxq_init hash_rxq_init[];
240 extern const unsigned int hash_rxq_init_n;
242 extern uint8_t rss_hash_default_key[];
243 extern const size_t rss_hash_default_key_len;
245 size_t hash_rxq_flow_attr(const struct hash_rxq *, struct ibv_exp_flow_attr *,
247 int priv_create_hash_rxqs(struct priv *);
248 void priv_destroy_hash_rxqs(struct priv *);
249 int priv_allow_flow_type(struct priv *, enum hash_rxq_flow_type);
250 void rxq_cleanup(struct rxq *);
251 int rxq_rehash(struct rte_eth_dev *, struct rxq *);
252 int rxq_setup(struct rte_eth_dev *, struct rxq *, uint16_t, unsigned int,
253 const struct rte_eth_rxconf *, struct rte_mempool *);
254 int mlx5_rx_queue_setup(struct rte_eth_dev *, uint16_t, uint16_t, unsigned int,
255 const struct rte_eth_rxconf *, struct rte_mempool *);
256 void mlx5_rx_queue_release(void *);
260 void txq_cleanup(struct txq *);
261 int mlx5_tx_queue_setup(struct rte_eth_dev *, uint16_t, uint16_t, unsigned int,
262 const struct rte_eth_txconf *);
263 void mlx5_tx_queue_release(void *);
267 void txq_mp2mr_iter(const struct rte_mempool *, void *);
268 uint16_t mlx5_tx_burst(void *, struct rte_mbuf **, uint16_t);
269 uint16_t mlx5_rx_burst_sp(void *, struct rte_mbuf **, uint16_t);
270 uint16_t mlx5_rx_burst(void *, struct rte_mbuf **, uint16_t);
271 uint16_t removed_tx_burst(void *, struct rte_mbuf **, uint16_t);
272 uint16_t removed_rx_burst(void *, struct rte_mbuf **, uint16_t);
274 #endif /* RTE_PMD_MLX5_RXTX_H_ */