xref: /dpdk/drivers/net/mlx5/mlx5_rxtx.h (revision ceb1ccd5d50c1a89ba8bdd97cc199e7f07422b98)
1 /*-
2  *   BSD LICENSE
3  *
4  *   Copyright 2015 6WIND S.A.
5  *   Copyright 2015 Mellanox.
6  *
7  *   Redistribution and use in source and binary forms, with or without
8  *   modification, are permitted provided that the following conditions
9  *   are met:
10  *
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
16  *       distribution.
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.
20  *
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.
32  */
33 
34 #ifndef RTE_PMD_MLX5_RXTX_H_
35 #define RTE_PMD_MLX5_RXTX_H_
36 
37 #include <stddef.h>
38 #include <stdint.h>
39 
40 /* Verbs header. */
41 /* ISO C doesn't support unnamed structs/unions, disabling -pedantic. */
42 #ifdef PEDANTIC
43 #pragma GCC diagnostic ignored "-pedantic"
44 #endif
45 #include <infiniband/verbs.h>
46 #ifdef PEDANTIC
47 #pragma GCC diagnostic error "-pedantic"
48 #endif
49 
50 /* DPDK headers don't like -pedantic. */
51 #ifdef PEDANTIC
52 #pragma GCC diagnostic ignored "-pedantic"
53 #endif
54 #include <rte_mbuf.h>
55 #include <rte_mempool.h>
56 #ifdef PEDANTIC
57 #pragma GCC diagnostic error "-pedantic"
58 #endif
59 
60 #include "mlx5_utils.h"
61 #include "mlx5.h"
62 #include "mlx5_autoconf.h"
63 #include "mlx5_defs.h"
64 
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. */
70 #endif
71 	uint64_t idropped; /**< Total of packets dropped when RX ring full. */
72 	uint64_t rx_nombuf; /**< Total of RX mbuf allocation failures. */
73 };
74 
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. */
80 #endif
81 	uint64_t odropped; /**< Total of packets not sent when TX ring full. */
82 };
83 
84 /* RX element (scattered packets). */
85 struct rxq_elt_sp {
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. */
88 };
89 
90 /* RX element. */
91 struct rxq_elt {
92 	struct ibv_sge sge; /* Scatter/Gather Element. */
93 	struct rte_mbuf *buf; /* SGE buffer. */
94 };
95 
96 /* Flow director queue structure. */
97 struct fdir_queue {
98 	struct ibv_qp *qp; /* Associated RX QP. */
99 	struct ibv_exp_rwq_ind_table *ind_table; /* Indirection table. */
100 };
101 
102 struct priv;
103 
104 /* RX queue descriptor. */
105 struct rxq {
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. */
120 	union {
121 		struct rxq_elt_sp (*sp)[]; /* Scattered RX elements. */
122 		struct rxq_elt (*no_sp)[]; /* RX elements. */
123 	} elts;
124 	uint32_t mb_len; /* Length of a mp-issued mbuf. */
125 	unsigned int socket; /* CPU socket ID for allocations. */
126 	struct mlx5_rxq_stats stats; /* RX queue counters. */
127 	struct ibv_exp_res_domain *rd; /* Resource Domain. */
128 	struct fdir_queue fdir_queue; /* Flow director queue. */
129 	struct ibv_mr *mr; /* Memory Region (for mp). */
130 	struct ibv_exp_wq_family *if_wq; /* WQ burst interface. */
131 #ifdef HAVE_EXP_DEVICE_ATTR_VLAN_OFFLOADS
132 	struct ibv_exp_cq_family_v1 *if_cq; /* CQ interface. */
133 #else /* HAVE_EXP_DEVICE_ATTR_VLAN_OFFLOADS */
134 	struct ibv_exp_cq_family *if_cq; /* CQ interface. */
135 #endif /* HAVE_EXP_DEVICE_ATTR_VLAN_OFFLOADS */
136 };
137 
138 /* Hash RX queue types. */
139 enum hash_rxq_type {
140 	HASH_RXQ_TCPV4,
141 	HASH_RXQ_UDPV4,
142 	HASH_RXQ_IPV4,
143 #ifdef HAVE_FLOW_SPEC_IPV6
144 	HASH_RXQ_TCPV6,
145 	HASH_RXQ_UDPV6,
146 	HASH_RXQ_IPV6,
147 #endif /* HAVE_FLOW_SPEC_IPV6 */
148 	HASH_RXQ_ETH,
149 };
150 
151 /* Flow structure with Ethernet specification. It is packed to prevent padding
152  * between attr and spec as this layout is expected by libibverbs. */
153 struct flow_attr_spec_eth {
154 	struct ibv_exp_flow_attr attr;
155 	struct ibv_exp_flow_spec_eth spec;
156 } __attribute__((packed));
157 
158 /* Define a struct flow_attr_spec_eth object as an array of at least
159  * "size" bytes. Room after the first index is normally used to store
160  * extra flow specifications. */
161 #define FLOW_ATTR_SPEC_ETH(name, size) \
162 	struct flow_attr_spec_eth name \
163 		[((size) / sizeof(struct flow_attr_spec_eth)) + \
164 		 !!((size) % sizeof(struct flow_attr_spec_eth))]
165 
166 /* Initialization data for hash RX queue. */
167 struct hash_rxq_init {
168 	uint64_t hash_fields; /* Fields that participate in the hash. */
169 	uint64_t dpdk_rss_hf; /* Matching DPDK RSS hash fields. */
170 	unsigned int flow_priority; /* Flow priority to use. */
171 	union {
172 		struct {
173 			enum ibv_exp_flow_spec_type type;
174 			uint16_t size;
175 		} hdr;
176 		struct ibv_exp_flow_spec_tcp_udp tcp_udp;
177 		struct ibv_exp_flow_spec_ipv4 ipv4;
178 #ifdef HAVE_FLOW_SPEC_IPV6
179 		struct ibv_exp_flow_spec_ipv6 ipv6;
180 #endif /* HAVE_FLOW_SPEC_IPV6 */
181 		struct ibv_exp_flow_spec_eth eth;
182 	} flow_spec; /* Flow specification template. */
183 	const struct hash_rxq_init *underlayer; /* Pointer to underlayer. */
184 };
185 
186 /* Initialization data for indirection table. */
187 struct ind_table_init {
188 	unsigned int max_size; /* Maximum number of WQs. */
189 	/* Hash RX queues using this table. */
190 	unsigned int hash_types;
191 	unsigned int hash_types_n;
192 };
193 
194 /* Initialization data for special flows. */
195 struct special_flow_init {
196 	uint8_t dst_mac_val[6];
197 	uint8_t dst_mac_mask[6];
198 	unsigned int hash_types;
199 	unsigned int per_vlan:1;
200 };
201 
202 enum hash_rxq_flow_type {
203 	HASH_RXQ_FLOW_TYPE_PROMISC,
204 	HASH_RXQ_FLOW_TYPE_ALLMULTI,
205 	HASH_RXQ_FLOW_TYPE_BROADCAST,
206 	HASH_RXQ_FLOW_TYPE_IPV6MULTI,
207 	HASH_RXQ_FLOW_TYPE_MAC,
208 };
209 
210 #ifndef NDEBUG
211 static inline const char *
212 hash_rxq_flow_type_str(enum hash_rxq_flow_type flow_type)
213 {
214 	switch (flow_type) {
215 	case HASH_RXQ_FLOW_TYPE_PROMISC:
216 		return "promiscuous";
217 	case HASH_RXQ_FLOW_TYPE_ALLMULTI:
218 		return "allmulticast";
219 	case HASH_RXQ_FLOW_TYPE_BROADCAST:
220 		return "broadcast";
221 	case HASH_RXQ_FLOW_TYPE_IPV6MULTI:
222 		return "IPv6 multicast";
223 	case HASH_RXQ_FLOW_TYPE_MAC:
224 		return "MAC";
225 	}
226 	return NULL;
227 }
228 #endif /* NDEBUG */
229 
230 struct hash_rxq {
231 	struct priv *priv; /* Back pointer to private data. */
232 	struct ibv_qp *qp; /* Hash RX QP. */
233 	enum hash_rxq_type type; /* Hash RX queue type. */
234 	/* MAC flow steering rules, one per VLAN ID. */
235 	struct ibv_exp_flow *mac_flow[MLX5_MAX_MAC_ADDRESSES][MLX5_MAX_VLAN_IDS];
236 	struct ibv_exp_flow *special_flow
237 		[MLX5_MAX_SPECIAL_FLOWS][MLX5_MAX_VLAN_IDS];
238 };
239 
240 /* TX element. */
241 struct txq_elt {
242 	struct rte_mbuf *buf;
243 };
244 
245 /* Linear buffer type. It is used when transmitting buffers with too many
246  * segments that do not fit the hardware queue (see max_send_sge).
247  * Extra segments are copied (linearized) in such buffers, replacing the
248  * last SGE during TX.
249  * The size is arbitrary but large enough to hold a jumbo frame with
250  * 8 segments considering mbuf.buf_len is about 2048 bytes. */
251 typedef uint8_t linear_t[16384];
252 
253 /* TX queue descriptor. */
254 struct txq {
255 	struct priv *priv; /* Back pointer to private data. */
256 	int32_t (*poll_cnt)(struct ibv_cq *cq, uint32_t max);
257 	int (*send_pending)();
258 #ifdef HAVE_VERBS_VLAN_INSERTION
259 	int (*send_pending_vlan)();
260 #endif
261 #if MLX5_PMD_MAX_INLINE > 0
262 	int (*send_pending_inline)();
263 #ifdef HAVE_VERBS_VLAN_INSERTION
264 	int (*send_pending_inline_vlan)();
265 #endif
266 #endif
267 #if MLX5_PMD_SGE_WR_N > 1
268 	int (*send_pending_sg_list)();
269 #ifdef HAVE_VERBS_VLAN_INSERTION
270 	int (*send_pending_sg_list_vlan)();
271 #endif
272 #endif
273 	int (*send_flush)(struct ibv_qp *qp);
274 	struct ibv_cq *cq; /* Completion Queue. */
275 	struct ibv_qp *qp; /* Queue Pair. */
276 	struct txq_elt (*elts)[]; /* TX elements. */
277 #if MLX5_PMD_MAX_INLINE > 0
278 	uint32_t max_inline; /* Max inline send size <= MLX5_PMD_MAX_INLINE. */
279 #endif
280 	unsigned int elts_n; /* (*elts)[] length. */
281 	unsigned int elts_head; /* Current index in (*elts)[]. */
282 	unsigned int elts_tail; /* First element awaiting completion. */
283 	unsigned int elts_comp; /* Number of completion requests. */
284 	unsigned int elts_comp_cd; /* Countdown for next completion request. */
285 	unsigned int elts_comp_cd_init; /* Initial value for countdown. */
286 	struct {
287 		const struct rte_mempool *mp; /* Cached Memory Pool. */
288 		struct ibv_mr *mr; /* Memory Region (for mp). */
289 		uint32_t lkey; /* mr->lkey */
290 	} mp2mr[MLX5_PMD_TX_MP_CACHE]; /* MP to MR translation table. */
291 	struct mlx5_txq_stats stats; /* TX queue counters. */
292 	/* Elements used only for init part are here. */
293 	linear_t (*elts_linear)[]; /* Linearized buffers. */
294 	struct ibv_mr *mr_linear; /* Memory Region for linearized buffers. */
295 #ifdef HAVE_VERBS_VLAN_INSERTION
296 	struct ibv_exp_qp_burst_family_v1 *if_qp; /* QP burst interface. */
297 #else
298 	struct ibv_exp_qp_burst_family *if_qp; /* QP burst interface. */
299 #endif
300 	struct ibv_exp_cq_family *if_cq; /* CQ interface. */
301 	struct ibv_exp_res_domain *rd; /* Resource Domain. */
302 	unsigned int socket; /* CPU socket ID for allocations. */
303 };
304 
305 /* mlx5_rxq.c */
306 
307 extern const struct hash_rxq_init hash_rxq_init[];
308 extern const unsigned int hash_rxq_init_n;
309 
310 extern uint8_t rss_hash_default_key[];
311 extern const size_t rss_hash_default_key_len;
312 
313 size_t priv_flow_attr(struct priv *, struct ibv_exp_flow_attr *,
314 		      size_t, enum hash_rxq_type);
315 int priv_create_hash_rxqs(struct priv *);
316 void priv_destroy_hash_rxqs(struct priv *);
317 int priv_allow_flow_type(struct priv *, enum hash_rxq_flow_type);
318 int priv_rehash_flows(struct priv *);
319 void rxq_cleanup(struct rxq *);
320 int rxq_rehash(struct rte_eth_dev *, struct rxq *);
321 int rxq_setup(struct rte_eth_dev *, struct rxq *, uint16_t, unsigned int,
322 	      const struct rte_eth_rxconf *, struct rte_mempool *);
323 int mlx5_rx_queue_setup(struct rte_eth_dev *, uint16_t, uint16_t, unsigned int,
324 			const struct rte_eth_rxconf *, struct rte_mempool *);
325 void mlx5_rx_queue_release(void *);
326 uint16_t mlx5_rx_burst_secondary_setup(void *dpdk_rxq, struct rte_mbuf **pkts,
327 			      uint16_t pkts_n);
328 
329 
330 /* mlx5_txq.c */
331 
332 void txq_cleanup(struct txq *);
333 int txq_setup(struct rte_eth_dev *dev, struct txq *txq, uint16_t desc,
334 	  unsigned int socket, const struct rte_eth_txconf *conf);
335 
336 int mlx5_tx_queue_setup(struct rte_eth_dev *, uint16_t, uint16_t, unsigned int,
337 			const struct rte_eth_txconf *);
338 void mlx5_tx_queue_release(void *);
339 uint16_t mlx5_tx_burst_secondary_setup(void *dpdk_txq, struct rte_mbuf **pkts,
340 			      uint16_t pkts_n);
341 
342 /* mlx5_rxtx.c */
343 
344 struct ibv_mr *mlx5_mp2mr(struct ibv_pd *, const struct rte_mempool *);
345 void txq_mp2mr_iter(const struct rte_mempool *, void *);
346 uint16_t mlx5_tx_burst(void *, struct rte_mbuf **, uint16_t);
347 uint16_t mlx5_rx_burst_sp(void *, struct rte_mbuf **, uint16_t);
348 uint16_t mlx5_rx_burst(void *, struct rte_mbuf **, uint16_t);
349 uint16_t removed_tx_burst(void *, struct rte_mbuf **, uint16_t);
350 uint16_t removed_rx_burst(void *, struct rte_mbuf **, uint16_t);
351 
352 #endif /* RTE_PMD_MLX5_RXTX_H_ */
353