]> git.proxmox.com Git - ceph.git/blob - ceph/src/dpdk/drivers/net/mlx5/mlx5_rxtx.h
bump version to 12.2.12-pve1
[ceph.git] / ceph / src / dpdk / drivers / net / mlx5 / mlx5_rxtx.h
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 "-Wpedantic"
44 #endif
45 #include <infiniband/verbs.h>
46 #include <infiniband/mlx5_hw.h>
47 #ifdef PEDANTIC
48 #pragma GCC diagnostic error "-Wpedantic"
49 #endif
50
51 /* DPDK headers don't like -pedantic. */
52 #ifdef PEDANTIC
53 #pragma GCC diagnostic ignored "-Wpedantic"
54 #endif
55 #include <rte_mbuf.h>
56 #include <rte_mempool.h>
57 #include <rte_common.h>
58 #ifdef PEDANTIC
59 #pragma GCC diagnostic error "-Wpedantic"
60 #endif
61
62 #include "mlx5_utils.h"
63 #include "mlx5.h"
64 #include "mlx5_autoconf.h"
65 #include "mlx5_defs.h"
66 #include "mlx5_prm.h"
67
68 struct mlx5_rxq_stats {
69 unsigned int idx; /**< Mapping index. */
70 #ifdef MLX5_PMD_SOFT_COUNTERS
71 uint64_t ipackets; /**< Total of successfully received packets. */
72 uint64_t ibytes; /**< Total of successfully received bytes. */
73 #endif
74 uint64_t idropped; /**< Total of packets dropped when RX ring full. */
75 uint64_t rx_nombuf; /**< Total of RX mbuf allocation failures. */
76 };
77
78 struct mlx5_txq_stats {
79 unsigned int idx; /**< Mapping index. */
80 #ifdef MLX5_PMD_SOFT_COUNTERS
81 uint64_t opackets; /**< Total of successfully sent packets. */
82 uint64_t obytes; /**< Total of successfully sent bytes. */
83 #endif
84 uint64_t odropped; /**< Total of packets not sent when TX ring full. */
85 };
86
87 /* Flow director queue structure. */
88 struct fdir_queue {
89 struct ibv_qp *qp; /* Associated RX QP. */
90 struct ibv_exp_rwq_ind_table *ind_table; /* Indirection table. */
91 struct ibv_exp_wq *wq; /* Work queue. */
92 struct ibv_cq *cq; /* Completion queue. */
93 };
94
95 struct priv;
96
97 /* Compressed CQE context. */
98 struct rxq_zip {
99 uint16_t ai; /* Array index. */
100 uint16_t ca; /* Current array index. */
101 uint16_t na; /* Next array index. */
102 uint16_t cq_ci; /* The next CQE. */
103 uint32_t cqe_cnt; /* Number of CQEs. */
104 };
105
106 /* RX queue descriptor. */
107 struct rxq {
108 unsigned int csum:1; /* Enable checksum offloading. */
109 unsigned int csum_l2tun:1; /* Same for L2 tunnels. */
110 unsigned int vlan_strip:1; /* Enable VLAN stripping. */
111 unsigned int crc_present:1; /* CRC must be subtracted. */
112 unsigned int sges_n:2; /* Log 2 of SGEs (max buffers per packet). */
113 unsigned int cqe_n:4; /* Log 2 of CQ elements. */
114 unsigned int elts_n:4; /* Log 2 of Mbufs. */
115 unsigned int port_id:8;
116 unsigned int rss_hash:1; /* RSS hash result is enabled. */
117 unsigned int :9; /* Remaining bits. */
118 volatile uint32_t *rq_db;
119 volatile uint32_t *cq_db;
120 uint16_t rq_ci;
121 uint16_t cq_ci;
122 volatile struct mlx5_wqe_data_seg(*wqes)[];
123 volatile struct mlx5_cqe(*cqes)[];
124 struct rxq_zip zip; /* Compressed context. */
125 struct rte_mbuf *(*elts)[];
126 struct rte_mempool *mp;
127 struct mlx5_rxq_stats stats;
128 } __rte_cache_aligned;
129
130 /* RX queue control descriptor. */
131 struct rxq_ctrl {
132 struct priv *priv; /* Back pointer to private data. */
133 struct ibv_cq *cq; /* Completion Queue. */
134 struct ibv_exp_wq *wq; /* Work Queue. */
135 struct ibv_exp_res_domain *rd; /* Resource Domain. */
136 struct fdir_queue *fdir_queue; /* Flow director queue. */
137 struct ibv_mr *mr; /* Memory Region (for mp). */
138 struct ibv_exp_wq_family *if_wq; /* WQ burst interface. */
139 struct ibv_exp_cq_family_v1 *if_cq; /* CQ interface. */
140 unsigned int socket; /* CPU socket ID for allocations. */
141 struct rxq rxq; /* Data path structure. */
142 };
143
144 /* Hash RX queue types. */
145 enum hash_rxq_type {
146 HASH_RXQ_TCPV4,
147 HASH_RXQ_UDPV4,
148 HASH_RXQ_IPV4,
149 HASH_RXQ_TCPV6,
150 HASH_RXQ_UDPV6,
151 HASH_RXQ_IPV6,
152 HASH_RXQ_ETH,
153 };
154
155 /* Flow structure with Ethernet specification. It is packed to prevent padding
156 * between attr and spec as this layout is expected by libibverbs. */
157 struct flow_attr_spec_eth {
158 struct ibv_exp_flow_attr attr;
159 struct ibv_exp_flow_spec_eth spec;
160 } __attribute__((packed));
161
162 /* Define a struct flow_attr_spec_eth object as an array of at least
163 * "size" bytes. Room after the first index is normally used to store
164 * extra flow specifications. */
165 #define FLOW_ATTR_SPEC_ETH(name, size) \
166 struct flow_attr_spec_eth name \
167 [((size) / sizeof(struct flow_attr_spec_eth)) + \
168 !!((size) % sizeof(struct flow_attr_spec_eth))]
169
170 /* Initialization data for hash RX queue. */
171 struct hash_rxq_init {
172 uint64_t hash_fields; /* Fields that participate in the hash. */
173 uint64_t dpdk_rss_hf; /* Matching DPDK RSS hash fields. */
174 unsigned int flow_priority; /* Flow priority to use. */
175 union {
176 struct {
177 enum ibv_exp_flow_spec_type type;
178 uint16_t size;
179 } hdr;
180 struct ibv_exp_flow_spec_tcp_udp tcp_udp;
181 struct ibv_exp_flow_spec_ipv4 ipv4;
182 struct ibv_exp_flow_spec_ipv6 ipv6;
183 struct ibv_exp_flow_spec_eth eth;
184 } flow_spec; /* Flow specification template. */
185 const struct hash_rxq_init *underlayer; /* Pointer to underlayer. */
186 };
187
188 /* Initialization data for indirection table. */
189 struct ind_table_init {
190 unsigned int max_size; /* Maximum number of WQs. */
191 /* Hash RX queues using this table. */
192 unsigned int hash_types;
193 unsigned int hash_types_n;
194 };
195
196 /* Initialization data for special flows. */
197 struct special_flow_init {
198 uint8_t dst_mac_val[6];
199 uint8_t dst_mac_mask[6];
200 unsigned int hash_types;
201 unsigned int per_vlan:1;
202 };
203
204 enum hash_rxq_flow_type {
205 HASH_RXQ_FLOW_TYPE_PROMISC,
206 HASH_RXQ_FLOW_TYPE_ALLMULTI,
207 HASH_RXQ_FLOW_TYPE_BROADCAST,
208 HASH_RXQ_FLOW_TYPE_IPV6MULTI,
209 HASH_RXQ_FLOW_TYPE_MAC,
210 };
211
212 #ifndef NDEBUG
213 static inline const char *
214 hash_rxq_flow_type_str(enum hash_rxq_flow_type flow_type)
215 {
216 switch (flow_type) {
217 case HASH_RXQ_FLOW_TYPE_PROMISC:
218 return "promiscuous";
219 case HASH_RXQ_FLOW_TYPE_ALLMULTI:
220 return "allmulticast";
221 case HASH_RXQ_FLOW_TYPE_BROADCAST:
222 return "broadcast";
223 case HASH_RXQ_FLOW_TYPE_IPV6MULTI:
224 return "IPv6 multicast";
225 case HASH_RXQ_FLOW_TYPE_MAC:
226 return "MAC";
227 }
228 return NULL;
229 }
230 #endif /* NDEBUG */
231
232 struct hash_rxq {
233 struct priv *priv; /* Back pointer to private data. */
234 struct ibv_qp *qp; /* Hash RX QP. */
235 enum hash_rxq_type type; /* Hash RX queue type. */
236 /* MAC flow steering rules, one per VLAN ID. */
237 struct ibv_exp_flow *mac_flow
238 [MLX5_MAX_MAC_ADDRESSES][MLX5_MAX_VLAN_IDS];
239 struct ibv_exp_flow *special_flow
240 [MLX5_MAX_SPECIAL_FLOWS][MLX5_MAX_VLAN_IDS];
241 };
242
243 /* TX queue descriptor. */
244 RTE_STD_C11
245 struct txq {
246 uint16_t elts_head; /* Current index in (*elts)[]. */
247 uint16_t elts_tail; /* First element awaiting completion. */
248 uint16_t elts_comp; /* Counter since last completion request. */
249 uint16_t cq_ci; /* Consumer index for completion queue. */
250 uint16_t wqe_ci; /* Consumer index for work queue. */
251 uint16_t elts_n:4; /* (*elts)[] length (in log2). */
252 uint16_t cqe_n:4; /* Number of CQ elements (in log2). */
253 uint16_t wqe_n:4; /* Number of of WQ elements (in log2). */
254 uint16_t bf_buf_size:4; /* Log2 Blueflame size. */
255 uint16_t bf_offset; /* Blueflame offset. */
256 uint16_t max_inline; /* Multiple of RTE_CACHE_LINE_SIZE to inline. */
257 uint32_t qp_num_8s; /* QP number shifted by 8. */
258 volatile struct mlx5_cqe (*cqes)[]; /* Completion queue. */
259 volatile struct mlx5_wqe64 (*wqes)[]; /* Work queue. */
260 volatile uint32_t *qp_db; /* Work queue doorbell. */
261 volatile uint32_t *cq_db; /* Completion queue doorbell. */
262 volatile void *bf_reg; /* Blueflame register. */
263 struct {
264 const struct rte_mempool *mp; /* Cached Memory Pool. */
265 struct ibv_mr *mr; /* Memory Region (for mp). */
266 uint32_t lkey; /* htonl(mr->lkey) */
267 } mp2mr[MLX5_PMD_TX_MP_CACHE]; /* MP to MR translation table. */
268 struct rte_mbuf *(*elts)[]; /* TX elements. */
269 struct mlx5_txq_stats stats; /* TX queue counters. */
270 } __rte_cache_aligned;
271
272 /* TX queue control descriptor. */
273 struct txq_ctrl {
274 struct priv *priv; /* Back pointer to private data. */
275 struct ibv_cq *cq; /* Completion Queue. */
276 struct ibv_qp *qp; /* Queue Pair. */
277 struct ibv_exp_qp_burst_family *if_qp; /* QP burst interface. */
278 struct ibv_exp_cq_family *if_cq; /* CQ interface. */
279 struct ibv_exp_res_domain *rd; /* Resource Domain. */
280 unsigned int socket; /* CPU socket ID for allocations. */
281 struct txq txq; /* Data path structure. */
282 };
283
284 /* mlx5_rxq.c */
285
286 extern const struct hash_rxq_init hash_rxq_init[];
287 extern const unsigned int hash_rxq_init_n;
288
289 extern uint8_t rss_hash_default_key[];
290 extern const size_t rss_hash_default_key_len;
291
292 size_t priv_flow_attr(struct priv *, struct ibv_exp_flow_attr *,
293 size_t, enum hash_rxq_type);
294 int priv_create_hash_rxqs(struct priv *);
295 void priv_destroy_hash_rxqs(struct priv *);
296 int priv_allow_flow_type(struct priv *, enum hash_rxq_flow_type);
297 int priv_rehash_flows(struct priv *);
298 void rxq_cleanup(struct rxq_ctrl *);
299 int rxq_rehash(struct rte_eth_dev *, struct rxq_ctrl *);
300 int rxq_ctrl_setup(struct rte_eth_dev *, struct rxq_ctrl *, uint16_t,
301 unsigned int, const struct rte_eth_rxconf *,
302 struct rte_mempool *);
303 int mlx5_rx_queue_setup(struct rte_eth_dev *, uint16_t, uint16_t, unsigned int,
304 const struct rte_eth_rxconf *, struct rte_mempool *);
305 void mlx5_rx_queue_release(void *);
306 uint16_t mlx5_rx_burst_secondary_setup(void *, struct rte_mbuf **, uint16_t);
307
308 /* mlx5_txq.c */
309
310 void txq_cleanup(struct txq_ctrl *);
311 int txq_ctrl_setup(struct rte_eth_dev *, struct txq_ctrl *, uint16_t,
312 unsigned int, const struct rte_eth_txconf *);
313 int mlx5_tx_queue_setup(struct rte_eth_dev *, uint16_t, uint16_t, unsigned int,
314 const struct rte_eth_txconf *);
315 void mlx5_tx_queue_release(void *);
316 uint16_t mlx5_tx_burst_secondary_setup(void *, struct rte_mbuf **, uint16_t);
317
318 /* mlx5_rxtx.c */
319
320 uint16_t mlx5_tx_burst(void *, struct rte_mbuf **, uint16_t);
321 uint16_t mlx5_tx_burst_mpw(void *, struct rte_mbuf **, uint16_t);
322 uint16_t mlx5_tx_burst_mpw_inline(void *, struct rte_mbuf **, uint16_t);
323 uint16_t mlx5_rx_burst(void *, struct rte_mbuf **, uint16_t);
324 uint16_t removed_tx_burst(void *, struct rte_mbuf **, uint16_t);
325 uint16_t removed_rx_burst(void *, struct rte_mbuf **, uint16_t);
326
327 /* mlx5_mr.c */
328
329 struct ibv_mr *mlx5_mp2mr(struct ibv_pd *, struct rte_mempool *);
330 void txq_mp2mr_iter(struct rte_mempool *, void *);
331 uint32_t txq_mp2mr_reg(struct txq *, struct rte_mempool *, unsigned int);
332
333 #endif /* RTE_PMD_MLX5_RXTX_H_ */