4 * Copyright 2012 6WIND S.A.
5 * Copyright 2012 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_MLX4_H_
35 #define RTE_PMD_MLX4_H_
42 * Runtime logging through RTE_LOG() is enabled when not in debugging mode.
43 * Intermediate LOG_*() macros add the required end-of-line characters.
46 #define INFO(...) DEBUG(__VA_ARGS__)
47 #define WARN(...) DEBUG(__VA_ARGS__)
48 #define ERROR(...) DEBUG(__VA_ARGS__)
50 #define LOG__(level, m, ...) \
51 RTE_LOG(level, PMD, MLX4_DRIVER_NAME ": " m "%c", __VA_ARGS__)
52 #define LOG_(level, ...) LOG__(level, __VA_ARGS__, '\n')
53 #define INFO(...) LOG_(INFO, __VA_ARGS__)
54 #define WARN(...) LOG_(WARNING, __VA_ARGS__)
55 #define ERROR(...) LOG_(ERR, __VA_ARGS__)
59 /* ISO C doesn't support unnamed structs/unions, disabling -pedantic. */
61 #pragma GCC diagnostic ignored "-Wpedantic"
63 #include <infiniband/verbs.h>
65 #pragma GCC diagnostic error "-Wpedantic"
69 * Maximum number of simultaneous MAC addresses supported.
71 * According to ConnectX's Programmer Reference Manual:
72 * The L2 Address Match is implemented by comparing a MAC/VLAN combination
73 * of 128 MAC addresses and 127 VLAN values, comprising 128x127 possible
76 #define MLX4_MAX_MAC_ADDRESSES 128
78 /* Maximum number of simultaneous VLAN filters supported. See above. */
79 #define MLX4_MAX_VLAN_IDS 127
81 /* Request send completion once in every 64 sends, might be less. */
82 #define MLX4_PMD_TX_PER_COMP_REQ 64
84 /* Maximum number of Scatter/Gather Elements per Work Request. */
85 #ifndef MLX4_PMD_SGE_WR_N
86 #define MLX4_PMD_SGE_WR_N 4
89 /* Maximum size for inline data. */
90 #ifndef MLX4_PMD_MAX_INLINE
91 #define MLX4_PMD_MAX_INLINE 0
95 * Maximum number of cached Memory Pools (MPs) per TX queue. Each RTE MP
96 * from which buffers are to be transmitted will have to be mapped by this
97 * driver to their own Memory Region (MR). This is a slow operation.
99 * This value is always 1 for RX queues.
101 #ifndef MLX4_PMD_TX_MP_CACHE
102 #define MLX4_PMD_TX_MP_CACHE 8
106 * If defined, only use software counters. The PMD will never ask the hardware
107 * for these, and many of them won't be available.
109 #ifndef MLX4_PMD_SOFT_COUNTERS
110 #define MLX4_PMD_SOFT_COUNTERS 1
114 #define MLX4_ALARM_TIMEOUT_US 100000
116 /* Port parameter. */
117 #define MLX4_PMD_PORT_KVARG "port"
120 PCI_VENDOR_ID_MELLANOX = 0x15b3,
124 PCI_DEVICE_ID_MELLANOX_CONNECTX3 = 0x1003,
125 PCI_DEVICE_ID_MELLANOX_CONNECTX3VF = 0x1004,
126 PCI_DEVICE_ID_MELLANOX_CONNECTX3PRO = 0x1007,
129 #define MLX4_DRIVER_NAME "net_mlx4"
131 /* Bit-field manipulation. */
132 #define BITFIELD_DECLARE(bf, type, size) \
133 type bf[(((size_t)(size) / (sizeof(type) * CHAR_BIT)) + \
134 !!((size_t)(size) % (sizeof(type) * CHAR_BIT)))]
135 #define BITFIELD_DEFINE(bf, type, size) \
136 BITFIELD_DECLARE((bf), type, (size)) = { 0 }
137 #define BITFIELD_SET(bf, b) \
138 (assert((size_t)(b) < (sizeof(bf) * CHAR_BIT)), \
139 (void)((bf)[((b) / (sizeof((bf)[0]) * CHAR_BIT))] |= \
140 ((size_t)1 << ((b) % (sizeof((bf)[0]) * CHAR_BIT)))))
141 #define BITFIELD_RESET(bf, b) \
142 (assert((size_t)(b) < (sizeof(bf) * CHAR_BIT)), \
143 (void)((bf)[((b) / (sizeof((bf)[0]) * CHAR_BIT))] &= \
144 ~((size_t)1 << ((b) % (sizeof((bf)[0]) * CHAR_BIT)))))
145 #define BITFIELD_ISSET(bf, b) \
146 (assert((size_t)(b) < (sizeof(bf) * CHAR_BIT)), \
147 !!(((bf)[((b) / (sizeof((bf)[0]) * CHAR_BIT))] & \
148 ((size_t)1 << ((b) % (sizeof((bf)[0]) * CHAR_BIT))))))
150 /* Number of elements in array. */
151 #define elemof(a) (sizeof(a) / sizeof((a)[0]))
153 /* Cast pointer p to structure member m to its parent structure of type t. */
154 #define containerof(p, t, m) ((t *)((uint8_t *)(p) - offsetof(t, m)))
156 /* Branch prediction helpers. */
158 #define likely(c) __builtin_expect(!!(c), 1)
161 #define unlikely(c) __builtin_expect(!!(c), 0)
167 #define DEBUG__(m, ...) \
168 (fprintf(stderr, "%s:%d: %s(): " m "%c", \
169 __FILE__, __LINE__, __func__, __VA_ARGS__), \
173 * Save/restore errno around DEBUG__().
174 * XXX somewhat undefined behavior, but works.
176 #define DEBUG_(...) \
177 (errno = ((int []){ \
178 *(volatile int *)&errno, \
179 (DEBUG__(__VA_ARGS__), 0) \
181 #define DEBUG(...) DEBUG_(__VA_ARGS__, '\n')
182 #ifndef MLX4_PMD_DEBUG_BROKEN_VERBS
183 #define claim_zero(...) assert((__VA_ARGS__) == 0)
184 #else /* MLX4_PMD_DEBUG_BROKEN_VERBS */
185 #define claim_zero(...) \
186 (void)(((__VA_ARGS__) == 0) || \
187 DEBUG("Assertion `(" # __VA_ARGS__ ") == 0' failed (IGNORED)."))
188 #endif /* MLX4_PMD_DEBUG_BROKEN_VERBS */
189 #define claim_nonzero(...) assert((__VA_ARGS__) != 0)
190 #define claim_positive(...) assert((__VA_ARGS__) >= 0)
193 #define DEBUG(...) (void)0
194 #define claim_zero(...) (__VA_ARGS__)
195 #define claim_nonzero(...) (__VA_ARGS__)
196 #define claim_positive(...) (__VA_ARGS__)
199 struct mlx4_rxq_stats {
200 unsigned int idx; /**< Mapping index. */
201 #ifdef MLX4_PMD_SOFT_COUNTERS
202 uint64_t ipackets; /**< Total of successfully received packets. */
203 uint64_t ibytes; /**< Total of successfully received bytes. */
205 uint64_t idropped; /**< Total of packets dropped when RX ring full. */
206 uint64_t rx_nombuf; /**< Total of RX mbuf allocation failures. */
209 /* RX element (scattered packets). */
211 struct ibv_recv_wr wr; /* Work Request. */
212 struct ibv_sge sges[MLX4_PMD_SGE_WR_N]; /* Scatter/Gather Elements. */
213 struct rte_mbuf *bufs[MLX4_PMD_SGE_WR_N]; /* SGEs buffers. */
218 struct ibv_recv_wr wr; /* Work Request. */
219 struct ibv_sge sge; /* Scatter/Gather Element. */
220 /* mbuf pointer is derived from WR_ID(wr.wr_id).offset. */
223 /* RX queue descriptor. */
225 LIST_ENTRY(rxq) next; /* Used by parent queue only */
226 struct priv *priv; /* Back pointer to private data. */
227 struct rte_mempool *mp; /* Memory Pool for allocations. */
228 struct ibv_mr *mr; /* Memory Region (for mp). */
229 struct ibv_cq *cq; /* Completion Queue. */
230 struct ibv_qp *qp; /* Queue Pair. */
231 struct ibv_exp_qp_burst_family *if_qp; /* QP burst interface. */
232 struct ibv_exp_cq_family *if_cq; /* CQ interface. */
233 struct ibv_comp_channel *channel;
235 * Each VLAN ID requires a separate flow steering rule.
237 BITFIELD_DECLARE(mac_configured, uint32_t, MLX4_MAX_MAC_ADDRESSES);
238 struct ibv_flow *mac_flow[MLX4_MAX_MAC_ADDRESSES][MLX4_MAX_VLAN_IDS];
239 struct ibv_flow *promisc_flow; /* Promiscuous flow. */
240 struct ibv_flow *allmulti_flow; /* Multicast flow. */
241 unsigned int port_id; /* Port ID for incoming packets. */
242 unsigned int elts_n; /* (*elts)[] length. */
243 unsigned int elts_head; /* Current index in (*elts)[]. */
245 struct rxq_elt_sp (*sp)[]; /* Scattered RX elements. */
246 struct rxq_elt (*no_sp)[]; /* RX elements. */
248 unsigned int sp:1; /* Use scattered RX elements. */
249 unsigned int csum:1; /* Enable checksum offloading. */
250 unsigned int csum_l2tun:1; /* Same for L2 tunnels. */
251 struct mlx4_rxq_stats stats; /* RX queue counters. */
252 unsigned int socket; /* CPU socket ID for allocations. */
253 struct ibv_exp_res_domain *rd; /* Resource Domain. */
256 uint16_t queues[RTE_MAX_QUEUES_PER_PORT];
262 struct rte_mbuf *buf;
265 struct mlx4_txq_stats {
266 unsigned int idx; /**< Mapping index. */
267 #ifdef MLX4_PMD_SOFT_COUNTERS
268 uint64_t opackets; /**< Total of successfully sent packets. */
269 uint64_t obytes; /**< Total of successfully sent bytes. */
271 uint64_t odropped; /**< Total of packets not sent when TX ring full. */
275 * Linear buffer type. It is used when transmitting buffers with too many
276 * segments that do not fit the hardware queue (see max_send_sge).
277 * Extra segments are copied (linearized) in such buffers, replacing the
278 * last SGE during TX.
279 * The size is arbitrary but large enough to hold a jumbo frame with
280 * 8 segments considering mbuf.buf_len is about 2048 bytes.
282 typedef uint8_t linear_t[16384];
284 /* TX queue descriptor. */
286 struct priv *priv; /* Back pointer to private data. */
288 const struct rte_mempool *mp; /* Cached Memory Pool. */
289 struct ibv_mr *mr; /* Memory Region (for mp). */
290 uint32_t lkey; /* mr->lkey */
291 } mp2mr[MLX4_PMD_TX_MP_CACHE]; /* MP to MR translation table. */
292 struct ibv_cq *cq; /* Completion Queue. */
293 struct ibv_qp *qp; /* Queue Pair. */
294 struct ibv_exp_qp_burst_family *if_qp; /* QP burst interface. */
295 struct ibv_exp_cq_family *if_cq; /* CQ interface. */
296 #if MLX4_PMD_MAX_INLINE > 0
297 uint32_t max_inline; /* Max inline send size <= MLX4_PMD_MAX_INLINE. */
299 unsigned int elts_n; /* (*elts)[] length. */
300 struct txq_elt (*elts)[]; /* TX elements. */
301 unsigned int elts_head; /* Current index in (*elts)[]. */
302 unsigned int elts_tail; /* First element awaiting completion. */
303 unsigned int elts_comp; /* Number of completion requests. */
304 unsigned int elts_comp_cd; /* Countdown for next completion request. */
305 unsigned int elts_comp_cd_init; /* Initial value for countdown. */
306 struct mlx4_txq_stats stats; /* TX queue counters. */
307 linear_t (*elts_linear)[]; /* Linearized buffers. */
308 struct ibv_mr *mr_linear; /* Memory Region for linearized buffers. */
309 unsigned int socket; /* CPU socket ID for allocations. */
310 struct ibv_exp_res_domain *rd; /* Resource Domain. */
316 struct rte_eth_dev *dev; /* Ethernet device. */
317 struct ibv_context *ctx; /* Verbs context. */
318 struct ibv_device_attr device_attr; /* Device properties. */
319 struct ibv_pd *pd; /* Protection Domain. */
321 * MAC addresses array and configuration bit-field.
322 * An extra entry that cannot be modified by the DPDK is reserved
323 * for broadcast frames (destination MAC address ff:ff:ff:ff:ff:ff).
325 struct ether_addr mac[MLX4_MAX_MAC_ADDRESSES];
326 BITFIELD_DECLARE(mac_configured, uint32_t, MLX4_MAX_MAC_ADDRESSES);
329 unsigned int enabled:1; /* If enabled. */
330 unsigned int id:12; /* VLAN ID (0-4095). */
331 } vlan_filter[MLX4_MAX_VLAN_IDS]; /* VLAN filters table. */
332 /* Device properties. */
333 uint16_t mtu; /* Configured MTU. */
334 uint8_t port; /* Physical port number. */
335 unsigned int started:1; /* Device started, flows enabled. */
336 unsigned int promisc:1; /* Device in promiscuous mode. */
337 unsigned int allmulti:1; /* Device receives all multicast packets. */
338 unsigned int hw_qpg:1; /* QP groups are supported. */
339 unsigned int hw_tss:1; /* TSS is supported. */
340 unsigned int hw_rss:1; /* RSS is supported. */
341 unsigned int hw_csum:1; /* Checksum offload is supported. */
342 unsigned int hw_csum_l2tun:1; /* Same for L2 tunnels. */
343 unsigned int rss:1; /* RSS is enabled. */
344 unsigned int vf:1; /* This is a VF device. */
345 unsigned int pending_alarm:1; /* An alarm is pending. */
346 unsigned int isolated:1; /* Toggle isolated mode. */
347 unsigned int inl_recv_size; /* Inline recv size */
348 unsigned int max_rss_tbl_sz; /* Maximum number of RSS queues. */
350 unsigned int rxqs_n; /* RX queues array size. */
351 unsigned int txqs_n; /* TX queues array size. */
352 struct rxq *(*rxqs)[]; /* RX queues. */
353 struct txq *(*txqs)[]; /* TX queues. */
354 struct rte_intr_handle intr_handle_dev; /* Device interrupt handler. */
355 struct rte_intr_handle intr_handle; /* Interrupt handler. */
356 struct rte_flow_drop *flow_drop_queue; /* Flow drop queue. */
357 LIST_HEAD(mlx4_flows, rte_flow) flows;
358 struct rte_intr_conf intr_conf; /* Active interrupt configuration. */
359 LIST_HEAD(mlx4_parents, rxq) parents;
360 rte_spinlock_t lock; /* Lock for control functions. */
363 void priv_lock(struct priv *priv);
364 void priv_unlock(struct priv *priv);
367 rxq_create_qp(struct rxq *rxq,
371 struct rxq *rxq_parent);
374 rxq_parent_cleanup(struct rxq *parent);
377 priv_parent_create(struct priv *priv,
379 uint16_t children_n);
381 #endif /* RTE_PMD_MLX4_H_ */