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. */
96 /* Flow director queue structure. */
98 struct ibv_qp *qp; /* Associated RX QP. */
99 struct ibv_exp_rwq_ind_table *ind_table; /* Indirection table. */
104 /* RX queue descriptor. */
106 struct priv *priv; /* Back pointer to private data. */
107 struct rte_mempool *mp; /* Memory Pool for allocations. */
108 struct ibv_cq *cq; /* Completion Queue. */
109 struct ibv_exp_wq *wq; /* Work Queue. */
110 int32_t (*poll)(); /* Verbs poll function. */
111 int32_t (*recv)(); /* Verbs receive function. */
112 unsigned int port_id; /* Port ID for incoming packets. */
113 unsigned int elts_n; /* (*elts)[] length. */
114 unsigned int elts_head; /* Current index in (*elts)[]. */
115 unsigned int sp:1; /* Use scattered RX elements. */
116 unsigned int csum:1; /* Enable checksum offloading. */
117 unsigned int csum_l2tun:1; /* Same for L2 tunnels. */
118 unsigned int vlan_strip:1; /* Enable VLAN stripping. */
119 unsigned int crc_present:1; /* CRC must be subtracted. */
121 struct rxq_elt_sp (*sp)[]; /* Scattered RX elements. */
122 struct rxq_elt (*no_sp)[]; /* RX elements. */
124 unsigned int socket; /* CPU socket ID for allocations. */
125 struct mlx5_rxq_stats stats; /* RX queue counters. */
126 struct ibv_exp_res_domain *rd; /* Resource Domain. */
127 struct fdir_queue fdir_queue; /* Flow director queue. */
128 struct ibv_mr *mr; /* Memory Region (for mp). */
129 struct ibv_exp_wq_family *if_wq; /* WQ burst interface. */
130 #ifdef HAVE_EXP_DEVICE_ATTR_VLAN_OFFLOADS
131 struct ibv_exp_cq_family_v1 *if_cq; /* CQ interface. */
132 #else /* HAVE_EXP_DEVICE_ATTR_VLAN_OFFLOADS */
133 struct ibv_exp_cq_family *if_cq; /* CQ interface. */
134 #endif /* HAVE_EXP_DEVICE_ATTR_VLAN_OFFLOADS */
137 /* Hash RX queue types. */
142 #ifdef HAVE_FLOW_SPEC_IPV6
146 #endif /* HAVE_FLOW_SPEC_IPV6 */
150 /* Flow structure with Ethernet specification. It is packed to prevent padding
151 * between attr and spec as this layout is expected by libibverbs. */
152 struct flow_attr_spec_eth {
153 struct ibv_exp_flow_attr attr;
154 struct ibv_exp_flow_spec_eth spec;
155 } __attribute__((packed));
157 /* Define a struct flow_attr_spec_eth object as an array of at least
158 * "size" bytes. Room after the first index is normally used to store
159 * extra flow specifications. */
160 #define FLOW_ATTR_SPEC_ETH(name, size) \
161 struct flow_attr_spec_eth name \
162 [((size) / sizeof(struct flow_attr_spec_eth)) + \
163 !!((size) % sizeof(struct flow_attr_spec_eth))]
165 /* Initialization data for hash RX queue. */
166 struct hash_rxq_init {
167 uint64_t hash_fields; /* Fields that participate in the hash. */
168 uint64_t dpdk_rss_hf; /* Matching DPDK RSS hash fields. */
169 unsigned int flow_priority; /* Flow priority to use. */
172 enum ibv_exp_flow_spec_type type;
175 struct ibv_exp_flow_spec_tcp_udp tcp_udp;
176 struct ibv_exp_flow_spec_ipv4 ipv4;
177 #ifdef HAVE_FLOW_SPEC_IPV6
178 struct ibv_exp_flow_spec_ipv6 ipv6;
179 #endif /* HAVE_FLOW_SPEC_IPV6 */
180 struct ibv_exp_flow_spec_eth eth;
181 } flow_spec; /* Flow specification template. */
182 const struct hash_rxq_init *underlayer; /* Pointer to underlayer. */
185 /* Initialization data for indirection table. */
186 struct ind_table_init {
187 unsigned int max_size; /* Maximum number of WQs. */
188 /* Hash RX queues using this table. */
189 unsigned int hash_types;
190 unsigned int hash_types_n;
193 /* Initialization data for special flows. */
194 struct special_flow_init {
195 uint8_t dst_mac_val[6];
196 uint8_t dst_mac_mask[6];
197 unsigned int hash_types;
198 unsigned int per_vlan:1;
201 enum hash_rxq_flow_type {
202 HASH_RXQ_FLOW_TYPE_PROMISC,
203 HASH_RXQ_FLOW_TYPE_ALLMULTI,
204 HASH_RXQ_FLOW_TYPE_BROADCAST,
205 HASH_RXQ_FLOW_TYPE_IPV6MULTI,
206 HASH_RXQ_FLOW_TYPE_MAC,
210 static inline const char *
211 hash_rxq_flow_type_str(enum hash_rxq_flow_type flow_type)
214 case HASH_RXQ_FLOW_TYPE_PROMISC:
215 return "promiscuous";
216 case HASH_RXQ_FLOW_TYPE_ALLMULTI:
217 return "allmulticast";
218 case HASH_RXQ_FLOW_TYPE_BROADCAST:
220 case HASH_RXQ_FLOW_TYPE_IPV6MULTI:
221 return "IPv6 multicast";
222 case HASH_RXQ_FLOW_TYPE_MAC:
230 struct priv *priv; /* Back pointer to private data. */
231 struct ibv_qp *qp; /* Hash RX QP. */
232 enum hash_rxq_type type; /* Hash RX queue type. */
233 /* MAC flow steering rules, one per VLAN ID. */
234 struct ibv_exp_flow *mac_flow[MLX5_MAX_MAC_ADDRESSES][MLX5_MAX_VLAN_IDS];
235 struct ibv_exp_flow *special_flow
236 [MLX5_MAX_SPECIAL_FLOWS][MLX5_MAX_VLAN_IDS];
241 struct rte_mbuf *buf;
244 /* Linear buffer type. It is used when transmitting buffers with too many
245 * segments that do not fit the hardware queue (see max_send_sge).
246 * Extra segments are copied (linearized) in such buffers, replacing the
247 * last SGE during TX.
248 * The size is arbitrary but large enough to hold a jumbo frame with
249 * 8 segments considering mbuf.buf_len is about 2048 bytes. */
250 typedef uint8_t linear_t[16384];
252 /* TX queue descriptor. */
254 struct priv *priv; /* Back pointer to private data. */
255 int32_t (*poll_cnt)(struct ibv_cq *cq, uint32_t max);
256 int (*send_pending)();
257 #ifdef HAVE_VERBS_VLAN_INSERTION
258 int (*send_pending_vlan)();
260 #if MLX5_PMD_MAX_INLINE > 0
261 int (*send_pending_inline)();
262 #ifdef HAVE_VERBS_VLAN_INSERTION
263 int (*send_pending_inline_vlan)();
266 #if MLX5_PMD_SGE_WR_N > 1
267 int (*send_pending_sg_list)();
268 #ifdef HAVE_VERBS_VLAN_INSERTION
269 int (*send_pending_sg_list_vlan)();
272 int (*send_flush)(struct ibv_qp *qp);
273 struct ibv_cq *cq; /* Completion Queue. */
274 struct ibv_qp *qp; /* Queue Pair. */
275 struct txq_elt (*elts)[]; /* TX elements. */
276 #if MLX5_PMD_MAX_INLINE > 0
277 uint32_t max_inline; /* Max inline send size <= MLX5_PMD_MAX_INLINE. */
279 unsigned int elts_n; /* (*elts)[] length. */
280 unsigned int elts_head; /* Current index in (*elts)[]. */
281 unsigned int elts_tail; /* First element awaiting completion. */
282 unsigned int elts_comp; /* Number of completion requests. */
283 unsigned int elts_comp_cd; /* Countdown for next completion request. */
284 unsigned int elts_comp_cd_init; /* Initial value for countdown. */
286 const struct rte_mempool *mp; /* Cached Memory Pool. */
287 struct ibv_mr *mr; /* Memory Region (for mp). */
288 uint32_t lkey; /* mr->lkey */
289 } mp2mr[MLX5_PMD_TX_MP_CACHE]; /* MP to MR translation table. */
290 struct mlx5_txq_stats stats; /* TX queue counters. */
291 /* Elements used only for init part are here. */
292 linear_t (*elts_linear)[]; /* Linearized buffers. */
293 struct ibv_mr *mr_linear; /* Memory Region for linearized buffers. */
294 #ifdef HAVE_VERBS_VLAN_INSERTION
295 struct ibv_exp_qp_burst_family_v1 *if_qp; /* QP burst interface. */
297 struct ibv_exp_qp_burst_family *if_qp; /* QP burst interface. */
299 struct ibv_exp_cq_family *if_cq; /* CQ interface. */
300 struct ibv_exp_res_domain *rd; /* Resource Domain. */
301 unsigned int socket; /* CPU socket ID for allocations. */
306 extern const struct hash_rxq_init hash_rxq_init[];
307 extern const unsigned int hash_rxq_init_n;
309 extern uint8_t rss_hash_default_key[];
310 extern const size_t rss_hash_default_key_len;
312 size_t priv_flow_attr(struct priv *, struct ibv_exp_flow_attr *,
313 size_t, enum hash_rxq_type);
314 int priv_create_hash_rxqs(struct priv *);
315 void priv_destroy_hash_rxqs(struct priv *);
316 int priv_allow_flow_type(struct priv *, enum hash_rxq_flow_type);
317 int priv_rehash_flows(struct priv *);
318 void rxq_cleanup(struct rxq *);
319 int rxq_rehash(struct rte_eth_dev *, struct rxq *);
320 int rxq_setup(struct rte_eth_dev *, struct rxq *, uint16_t, unsigned int,
321 const struct rte_eth_rxconf *, struct rte_mempool *);
322 int mlx5_rx_queue_setup(struct rte_eth_dev *, uint16_t, uint16_t, unsigned int,
323 const struct rte_eth_rxconf *, struct rte_mempool *);
324 void mlx5_rx_queue_release(void *);
325 uint16_t mlx5_rx_burst_secondary_setup(void *dpdk_rxq, struct rte_mbuf **pkts,
331 void txq_cleanup(struct txq *);
332 int txq_setup(struct rte_eth_dev *dev, struct txq *txq, uint16_t desc,
333 unsigned int socket, const struct rte_eth_txconf *conf);
335 int mlx5_tx_queue_setup(struct rte_eth_dev *, uint16_t, uint16_t, unsigned int,
336 const struct rte_eth_txconf *);
337 void mlx5_tx_queue_release(void *);
338 uint16_t mlx5_tx_burst_secondary_setup(void *dpdk_txq, struct rte_mbuf **pkts,
343 struct ibv_mr *mlx5_mp2mr(struct ibv_pd *, struct rte_mempool *);
344 void txq_mp2mr_iter(struct rte_mempool *, void *);
345 uint16_t mlx5_tx_burst(void *, struct rte_mbuf **, uint16_t);
346 uint16_t mlx5_rx_burst_sp(void *, struct rte_mbuf **, uint16_t);
347 uint16_t mlx5_rx_burst(void *, struct rte_mbuf **, uint16_t);
348 uint16_t removed_tx_burst(void *, struct rte_mbuf **, uint16_t);
349 uint16_t removed_rx_burst(void *, struct rte_mbuf **, uint16_t);
351 #endif /* RTE_PMD_MLX5_RXTX_H_ */