1 /* 2 * Copyright (c) 2004 Mellanox Technologies Ltd. All rights reserved. 3 * Copyright (c) 2004 Infinicon Corporation. All rights reserved. 4 * Copyright (c) 2004 Intel Corporation. All rights reserved. 5 * Copyright (c) 2004 Topspin Corporation. All rights reserved. 6 * Copyright (c) 2004 Voltaire Corporation. All rights reserved. 7 * Copyright (c) 2005 Sun Microsystems, Inc. All rights reserved. 8 * Copyright (c) 2005, 2006, 2007 Cisco Systems. All rights reserved. 9 * 10 * This software is available to you under a choice of one of two 11 * licenses. You may choose to be licensed under the terms of the GNU 12 * General Public License (GPL) Version 2, available from the file 13 * COPYING in the main directory of this source tree, or the 14 * OpenIB.org BSD license below: 15 * 16 * Redistribution and use in source and binary forms, with or 17 * without modification, are permitted provided that the following 18 * conditions are met: 19 * 20 * - Redistributions of source code must retain the above 21 * copyright notice, this list of conditions and the following 22 * disclaimer. 23 * 24 * - Redistributions in binary form must reproduce the above 25 * copyright notice, this list of conditions and the following 26 * disclaimer in the documentation and/or other materials 27 * provided with the distribution. 28 * 29 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, 30 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF 31 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND 32 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS 33 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN 34 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN 35 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE 36 * SOFTWARE. 37 */ 38 39 #if !defined(IB_VERBS_H) 40 #define IB_VERBS_H 41 42 #include <linux/types.h> 43 #include <linux/device.h> 44 #include <linux/mm.h> 45 #include <linux/dma-mapping.h> 46 #include <linux/kref.h> 47 #include <linux/list.h> 48 #include <linux/rwsem.h> 49 #include <linux/scatterlist.h> 50 #include <linux/workqueue.h> 51 #include <linux/socket.h> 52 #include <linux/irq_poll.h> 53 #include <uapi/linux/if_ether.h> 54 #include <net/ipv6.h> 55 #include <net/ip.h> 56 #include <linux/string.h> 57 #include <linux/slab.h> 58 #include <linux/netdevice.h> 59 60 #include <linux/if_link.h> 61 #include <linux/atomic.h> 62 #include <linux/mmu_notifier.h> 63 #include <linux/uaccess.h> 64 #include <linux/cgroup_rdma.h> 65 #include <uapi/rdma/ib_user_verbs.h> 66 67 #define IB_FW_VERSION_NAME_MAX ETHTOOL_FWVERS_LEN 68 69 extern struct workqueue_struct *ib_wq; 70 extern struct workqueue_struct *ib_comp_wq; 71 72 union ib_gid { 73 u8 raw[16]; 74 struct { 75 __be64 subnet_prefix; 76 __be64 interface_id; 77 } global; 78 }; 79 80 extern union ib_gid zgid; 81 82 enum ib_gid_type { 83 /* If link layer is Ethernet, this is RoCE V1 */ 84 IB_GID_TYPE_IB = 0, 85 IB_GID_TYPE_ROCE = 0, 86 IB_GID_TYPE_ROCE_UDP_ENCAP = 1, 87 IB_GID_TYPE_SIZE 88 }; 89 90 #define ROCE_V2_UDP_DPORT 4791 91 struct ib_gid_attr { 92 enum ib_gid_type gid_type; 93 struct net_device *ndev; 94 }; 95 96 enum rdma_node_type { 97 /* IB values map to NodeInfo:NodeType. */ 98 RDMA_NODE_IB_CA = 1, 99 RDMA_NODE_IB_SWITCH, 100 RDMA_NODE_IB_ROUTER, 101 RDMA_NODE_RNIC, 102 RDMA_NODE_USNIC, 103 RDMA_NODE_USNIC_UDP, 104 }; 105 106 enum { 107 /* set the local administered indication */ 108 IB_SA_WELL_KNOWN_GUID = BIT_ULL(57) | 2, 109 }; 110 111 enum rdma_transport_type { 112 RDMA_TRANSPORT_IB, 113 RDMA_TRANSPORT_IWARP, 114 RDMA_TRANSPORT_USNIC, 115 RDMA_TRANSPORT_USNIC_UDP 116 }; 117 118 enum rdma_protocol_type { 119 RDMA_PROTOCOL_IB, 120 RDMA_PROTOCOL_IBOE, 121 RDMA_PROTOCOL_IWARP, 122 RDMA_PROTOCOL_USNIC_UDP 123 }; 124 125 __attribute_const__ enum rdma_transport_type 126 rdma_node_get_transport(enum rdma_node_type node_type); 127 128 enum rdma_network_type { 129 RDMA_NETWORK_IB, 130 RDMA_NETWORK_ROCE_V1 = RDMA_NETWORK_IB, 131 RDMA_NETWORK_IPV4, 132 RDMA_NETWORK_IPV6 133 }; 134 135 static inline enum ib_gid_type ib_network_to_gid_type(enum rdma_network_type network_type) 136 { 137 if (network_type == RDMA_NETWORK_IPV4 || 138 network_type == RDMA_NETWORK_IPV6) 139 return IB_GID_TYPE_ROCE_UDP_ENCAP; 140 141 /* IB_GID_TYPE_IB same as RDMA_NETWORK_ROCE_V1 */ 142 return IB_GID_TYPE_IB; 143 } 144 145 static inline enum rdma_network_type ib_gid_to_network_type(enum ib_gid_type gid_type, 146 union ib_gid *gid) 147 { 148 if (gid_type == IB_GID_TYPE_IB) 149 return RDMA_NETWORK_IB; 150 151 if (ipv6_addr_v4mapped((struct in6_addr *)gid)) 152 return RDMA_NETWORK_IPV4; 153 else 154 return RDMA_NETWORK_IPV6; 155 } 156 157 enum rdma_link_layer { 158 IB_LINK_LAYER_UNSPECIFIED, 159 IB_LINK_LAYER_INFINIBAND, 160 IB_LINK_LAYER_ETHERNET, 161 }; 162 163 enum ib_device_cap_flags { 164 IB_DEVICE_RESIZE_MAX_WR = (1 << 0), 165 IB_DEVICE_BAD_PKEY_CNTR = (1 << 1), 166 IB_DEVICE_BAD_QKEY_CNTR = (1 << 2), 167 IB_DEVICE_RAW_MULTI = (1 << 3), 168 IB_DEVICE_AUTO_PATH_MIG = (1 << 4), 169 IB_DEVICE_CHANGE_PHY_PORT = (1 << 5), 170 IB_DEVICE_UD_AV_PORT_ENFORCE = (1 << 6), 171 IB_DEVICE_CURR_QP_STATE_MOD = (1 << 7), 172 IB_DEVICE_SHUTDOWN_PORT = (1 << 8), 173 /* Not in use, former INIT_TYPE = (1 << 9),*/ 174 IB_DEVICE_PORT_ACTIVE_EVENT = (1 << 10), 175 IB_DEVICE_SYS_IMAGE_GUID = (1 << 11), 176 IB_DEVICE_RC_RNR_NAK_GEN = (1 << 12), 177 IB_DEVICE_SRQ_RESIZE = (1 << 13), 178 IB_DEVICE_N_NOTIFY_CQ = (1 << 14), 179 180 /* 181 * This device supports a per-device lkey or stag that can be 182 * used without performing a memory registration for the local 183 * memory. Note that ULPs should never check this flag, but 184 * instead of use the local_dma_lkey flag in the ib_pd structure, 185 * which will always contain a usable lkey. 186 */ 187 IB_DEVICE_LOCAL_DMA_LKEY = (1 << 15), 188 /* Reserved, old SEND_W_INV = (1 << 16),*/ 189 IB_DEVICE_MEM_WINDOW = (1 << 17), 190 /* 191 * Devices should set IB_DEVICE_UD_IP_SUM if they support 192 * insertion of UDP and TCP checksum on outgoing UD IPoIB 193 * messages and can verify the validity of checksum for 194 * incoming messages. Setting this flag implies that the 195 * IPoIB driver may set NETIF_F_IP_CSUM for datagram mode. 196 */ 197 IB_DEVICE_UD_IP_CSUM = (1 << 18), 198 IB_DEVICE_UD_TSO = (1 << 19), 199 IB_DEVICE_XRC = (1 << 20), 200 201 /* 202 * This device supports the IB "base memory management extension", 203 * which includes support for fast registrations (IB_WR_REG_MR, 204 * IB_WR_LOCAL_INV and IB_WR_SEND_WITH_INV verbs). This flag should 205 * also be set by any iWarp device which must support FRs to comply 206 * to the iWarp verbs spec. iWarp devices also support the 207 * IB_WR_RDMA_READ_WITH_INV verb for RDMA READs that invalidate the 208 * stag. 209 */ 210 IB_DEVICE_MEM_MGT_EXTENSIONS = (1 << 21), 211 IB_DEVICE_BLOCK_MULTICAST_LOOPBACK = (1 << 22), 212 IB_DEVICE_MEM_WINDOW_TYPE_2A = (1 << 23), 213 IB_DEVICE_MEM_WINDOW_TYPE_2B = (1 << 24), 214 IB_DEVICE_RC_IP_CSUM = (1 << 25), 215 /* Deprecated. Please use IB_RAW_PACKET_CAP_IP_CSUM. */ 216 IB_DEVICE_RAW_IP_CSUM = (1 << 26), 217 /* 218 * Devices should set IB_DEVICE_CROSS_CHANNEL if they 219 * support execution of WQEs that involve synchronization 220 * of I/O operations with single completion queue managed 221 * by hardware. 222 */ 223 IB_DEVICE_CROSS_CHANNEL = (1 << 27), 224 IB_DEVICE_MANAGED_FLOW_STEERING = (1 << 29), 225 IB_DEVICE_SIGNATURE_HANDOVER = (1 << 30), 226 IB_DEVICE_ON_DEMAND_PAGING = (1ULL << 31), 227 IB_DEVICE_SG_GAPS_REG = (1ULL << 32), 228 IB_DEVICE_VIRTUAL_FUNCTION = (1ULL << 33), 229 /* Deprecated. Please use IB_RAW_PACKET_CAP_SCATTER_FCS. */ 230 IB_DEVICE_RAW_SCATTER_FCS = (1ULL << 34), 231 IB_DEVICE_RDMA_NETDEV_OPA_VNIC = (1ULL << 35), 232 /* The device supports padding incoming writes to cacheline. */ 233 IB_DEVICE_PCI_WRITE_END_PADDING = (1ULL << 36), 234 }; 235 236 enum ib_signature_prot_cap { 237 IB_PROT_T10DIF_TYPE_1 = 1, 238 IB_PROT_T10DIF_TYPE_2 = 1 << 1, 239 IB_PROT_T10DIF_TYPE_3 = 1 << 2, 240 }; 241 242 enum ib_signature_guard_cap { 243 IB_GUARD_T10DIF_CRC = 1, 244 IB_GUARD_T10DIF_CSUM = 1 << 1, 245 }; 246 247 enum ib_atomic_cap { 248 IB_ATOMIC_NONE, 249 IB_ATOMIC_HCA, 250 IB_ATOMIC_GLOB 251 }; 252 253 enum ib_odp_general_cap_bits { 254 IB_ODP_SUPPORT = 1 << 0, 255 IB_ODP_SUPPORT_IMPLICIT = 1 << 1, 256 }; 257 258 enum ib_odp_transport_cap_bits { 259 IB_ODP_SUPPORT_SEND = 1 << 0, 260 IB_ODP_SUPPORT_RECV = 1 << 1, 261 IB_ODP_SUPPORT_WRITE = 1 << 2, 262 IB_ODP_SUPPORT_READ = 1 << 3, 263 IB_ODP_SUPPORT_ATOMIC = 1 << 4, 264 }; 265 266 struct ib_odp_caps { 267 uint64_t general_caps; 268 struct { 269 uint32_t rc_odp_caps; 270 uint32_t uc_odp_caps; 271 uint32_t ud_odp_caps; 272 } per_transport_caps; 273 }; 274 275 struct ib_rss_caps { 276 /* Corresponding bit will be set if qp type from 277 * 'enum ib_qp_type' is supported, e.g. 278 * supported_qpts |= 1 << IB_QPT_UD 279 */ 280 u32 supported_qpts; 281 u32 max_rwq_indirection_tables; 282 u32 max_rwq_indirection_table_size; 283 }; 284 285 enum ib_tm_cap_flags { 286 /* Support tag matching on RC transport */ 287 IB_TM_CAP_RC = 1 << 0, 288 }; 289 290 struct ib_tm_caps { 291 /* Max size of RNDV header */ 292 u32 max_rndv_hdr_size; 293 /* Max number of entries in tag matching list */ 294 u32 max_num_tags; 295 /* From enum ib_tm_cap_flags */ 296 u32 flags; 297 /* Max number of outstanding list operations */ 298 u32 max_ops; 299 /* Max number of SGE in tag matching entry */ 300 u32 max_sge; 301 }; 302 303 enum ib_cq_creation_flags { 304 IB_CQ_FLAGS_TIMESTAMP_COMPLETION = 1 << 0, 305 IB_CQ_FLAGS_IGNORE_OVERRUN = 1 << 1, 306 }; 307 308 struct ib_cq_init_attr { 309 unsigned int cqe; 310 int comp_vector; 311 u32 flags; 312 }; 313 314 enum ib_cq_attr_mask { 315 IB_CQ_MODERATE = 1 << 0, 316 }; 317 318 struct ib_cq_caps { 319 u16 max_cq_moderation_count; 320 u16 max_cq_moderation_period; 321 }; 322 323 struct ib_device_attr { 324 u64 fw_ver; 325 __be64 sys_image_guid; 326 u64 max_mr_size; 327 u64 page_size_cap; 328 u32 vendor_id; 329 u32 vendor_part_id; 330 u32 hw_ver; 331 int max_qp; 332 int max_qp_wr; 333 u64 device_cap_flags; 334 int max_sge; 335 int max_sge_rd; 336 int max_cq; 337 int max_cqe; 338 int max_mr; 339 int max_pd; 340 int max_qp_rd_atom; 341 int max_ee_rd_atom; 342 int max_res_rd_atom; 343 int max_qp_init_rd_atom; 344 int max_ee_init_rd_atom; 345 enum ib_atomic_cap atomic_cap; 346 enum ib_atomic_cap masked_atomic_cap; 347 int max_ee; 348 int max_rdd; 349 int max_mw; 350 int max_raw_ipv6_qp; 351 int max_raw_ethy_qp; 352 int max_mcast_grp; 353 int max_mcast_qp_attach; 354 int max_total_mcast_qp_attach; 355 int max_ah; 356 int max_fmr; 357 int max_map_per_fmr; 358 int max_srq; 359 int max_srq_wr; 360 int max_srq_sge; 361 unsigned int max_fast_reg_page_list_len; 362 u16 max_pkeys; 363 u8 local_ca_ack_delay; 364 int sig_prot_cap; 365 int sig_guard_cap; 366 struct ib_odp_caps odp_caps; 367 uint64_t timestamp_mask; 368 uint64_t hca_core_clock; /* in KHZ */ 369 struct ib_rss_caps rss_caps; 370 u32 max_wq_type_rq; 371 u32 raw_packet_caps; /* Use ib_raw_packet_caps enum */ 372 struct ib_tm_caps tm_caps; 373 struct ib_cq_caps cq_caps; 374 }; 375 376 enum ib_mtu { 377 IB_MTU_256 = 1, 378 IB_MTU_512 = 2, 379 IB_MTU_1024 = 3, 380 IB_MTU_2048 = 4, 381 IB_MTU_4096 = 5 382 }; 383 384 static inline int ib_mtu_enum_to_int(enum ib_mtu mtu) 385 { 386 switch (mtu) { 387 case IB_MTU_256: return 256; 388 case IB_MTU_512: return 512; 389 case IB_MTU_1024: return 1024; 390 case IB_MTU_2048: return 2048; 391 case IB_MTU_4096: return 4096; 392 default: return -1; 393 } 394 } 395 396 static inline enum ib_mtu ib_mtu_int_to_enum(int mtu) 397 { 398 if (mtu >= 4096) 399 return IB_MTU_4096; 400 else if (mtu >= 2048) 401 return IB_MTU_2048; 402 else if (mtu >= 1024) 403 return IB_MTU_1024; 404 else if (mtu >= 512) 405 return IB_MTU_512; 406 else 407 return IB_MTU_256; 408 } 409 410 enum ib_port_state { 411 IB_PORT_NOP = 0, 412 IB_PORT_DOWN = 1, 413 IB_PORT_INIT = 2, 414 IB_PORT_ARMED = 3, 415 IB_PORT_ACTIVE = 4, 416 IB_PORT_ACTIVE_DEFER = 5 417 }; 418 419 enum ib_port_cap_flags { 420 IB_PORT_SM = 1 << 1, 421 IB_PORT_NOTICE_SUP = 1 << 2, 422 IB_PORT_TRAP_SUP = 1 << 3, 423 IB_PORT_OPT_IPD_SUP = 1 << 4, 424 IB_PORT_AUTO_MIGR_SUP = 1 << 5, 425 IB_PORT_SL_MAP_SUP = 1 << 6, 426 IB_PORT_MKEY_NVRAM = 1 << 7, 427 IB_PORT_PKEY_NVRAM = 1 << 8, 428 IB_PORT_LED_INFO_SUP = 1 << 9, 429 IB_PORT_SM_DISABLED = 1 << 10, 430 IB_PORT_SYS_IMAGE_GUID_SUP = 1 << 11, 431 IB_PORT_PKEY_SW_EXT_PORT_TRAP_SUP = 1 << 12, 432 IB_PORT_EXTENDED_SPEEDS_SUP = 1 << 14, 433 IB_PORT_CM_SUP = 1 << 16, 434 IB_PORT_SNMP_TUNNEL_SUP = 1 << 17, 435 IB_PORT_REINIT_SUP = 1 << 18, 436 IB_PORT_DEVICE_MGMT_SUP = 1 << 19, 437 IB_PORT_VENDOR_CLASS_SUP = 1 << 20, 438 IB_PORT_DR_NOTICE_SUP = 1 << 21, 439 IB_PORT_CAP_MASK_NOTICE_SUP = 1 << 22, 440 IB_PORT_BOOT_MGMT_SUP = 1 << 23, 441 IB_PORT_LINK_LATENCY_SUP = 1 << 24, 442 IB_PORT_CLIENT_REG_SUP = 1 << 25, 443 IB_PORT_IP_BASED_GIDS = 1 << 26, 444 }; 445 446 enum ib_port_width { 447 IB_WIDTH_1X = 1, 448 IB_WIDTH_4X = 2, 449 IB_WIDTH_8X = 4, 450 IB_WIDTH_12X = 8 451 }; 452 453 static inline int ib_width_enum_to_int(enum ib_port_width width) 454 { 455 switch (width) { 456 case IB_WIDTH_1X: return 1; 457 case IB_WIDTH_4X: return 4; 458 case IB_WIDTH_8X: return 8; 459 case IB_WIDTH_12X: return 12; 460 default: return -1; 461 } 462 } 463 464 enum ib_port_speed { 465 IB_SPEED_SDR = 1, 466 IB_SPEED_DDR = 2, 467 IB_SPEED_QDR = 4, 468 IB_SPEED_FDR10 = 8, 469 IB_SPEED_FDR = 16, 470 IB_SPEED_EDR = 32, 471 IB_SPEED_HDR = 64 472 }; 473 474 /** 475 * struct rdma_hw_stats 476 * @timestamp - Used by the core code to track when the last update was 477 * @lifespan - Used by the core code to determine how old the counters 478 * should be before being updated again. Stored in jiffies, defaults 479 * to 10 milliseconds, drivers can override the default be specifying 480 * their own value during their allocation routine. 481 * @name - Array of pointers to static names used for the counters in 482 * directory. 483 * @num_counters - How many hardware counters there are. If name is 484 * shorter than this number, a kernel oops will result. Driver authors 485 * are encouraged to leave BUILD_BUG_ON(ARRAY_SIZE(@name) < num_counters) 486 * in their code to prevent this. 487 * @value - Array of u64 counters that are accessed by the sysfs code and 488 * filled in by the drivers get_stats routine 489 */ 490 struct rdma_hw_stats { 491 unsigned long timestamp; 492 unsigned long lifespan; 493 const char * const *names; 494 int num_counters; 495 u64 value[]; 496 }; 497 498 #define RDMA_HW_STATS_DEFAULT_LIFESPAN 10 499 /** 500 * rdma_alloc_hw_stats_struct - Helper function to allocate dynamic struct 501 * for drivers. 502 * @names - Array of static const char * 503 * @num_counters - How many elements in array 504 * @lifespan - How many milliseconds between updates 505 */ 506 static inline struct rdma_hw_stats *rdma_alloc_hw_stats_struct( 507 const char * const *names, int num_counters, 508 unsigned long lifespan) 509 { 510 struct rdma_hw_stats *stats; 511 512 stats = kzalloc(sizeof(*stats) + num_counters * sizeof(u64), 513 GFP_KERNEL); 514 if (!stats) 515 return NULL; 516 stats->names = names; 517 stats->num_counters = num_counters; 518 stats->lifespan = msecs_to_jiffies(lifespan); 519 520 return stats; 521 } 522 523 524 /* Define bits for the various functionality this port needs to be supported by 525 * the core. 526 */ 527 /* Management 0x00000FFF */ 528 #define RDMA_CORE_CAP_IB_MAD 0x00000001 529 #define RDMA_CORE_CAP_IB_SMI 0x00000002 530 #define RDMA_CORE_CAP_IB_CM 0x00000004 531 #define RDMA_CORE_CAP_IW_CM 0x00000008 532 #define RDMA_CORE_CAP_IB_SA 0x00000010 533 #define RDMA_CORE_CAP_OPA_MAD 0x00000020 534 535 /* Address format 0x000FF000 */ 536 #define RDMA_CORE_CAP_AF_IB 0x00001000 537 #define RDMA_CORE_CAP_ETH_AH 0x00002000 538 #define RDMA_CORE_CAP_OPA_AH 0x00004000 539 540 /* Protocol 0xFFF00000 */ 541 #define RDMA_CORE_CAP_PROT_IB 0x00100000 542 #define RDMA_CORE_CAP_PROT_ROCE 0x00200000 543 #define RDMA_CORE_CAP_PROT_IWARP 0x00400000 544 #define RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP 0x00800000 545 #define RDMA_CORE_CAP_PROT_RAW_PACKET 0x01000000 546 #define RDMA_CORE_CAP_PROT_USNIC 0x02000000 547 548 #define RDMA_CORE_PORT_IBA_IB (RDMA_CORE_CAP_PROT_IB \ 549 | RDMA_CORE_CAP_IB_MAD \ 550 | RDMA_CORE_CAP_IB_SMI \ 551 | RDMA_CORE_CAP_IB_CM \ 552 | RDMA_CORE_CAP_IB_SA \ 553 | RDMA_CORE_CAP_AF_IB) 554 #define RDMA_CORE_PORT_IBA_ROCE (RDMA_CORE_CAP_PROT_ROCE \ 555 | RDMA_CORE_CAP_IB_MAD \ 556 | RDMA_CORE_CAP_IB_CM \ 557 | RDMA_CORE_CAP_AF_IB \ 558 | RDMA_CORE_CAP_ETH_AH) 559 #define RDMA_CORE_PORT_IBA_ROCE_UDP_ENCAP \ 560 (RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP \ 561 | RDMA_CORE_CAP_IB_MAD \ 562 | RDMA_CORE_CAP_IB_CM \ 563 | RDMA_CORE_CAP_AF_IB \ 564 | RDMA_CORE_CAP_ETH_AH) 565 #define RDMA_CORE_PORT_IWARP (RDMA_CORE_CAP_PROT_IWARP \ 566 | RDMA_CORE_CAP_IW_CM) 567 #define RDMA_CORE_PORT_INTEL_OPA (RDMA_CORE_PORT_IBA_IB \ 568 | RDMA_CORE_CAP_OPA_MAD) 569 570 #define RDMA_CORE_PORT_RAW_PACKET (RDMA_CORE_CAP_PROT_RAW_PACKET) 571 572 #define RDMA_CORE_PORT_USNIC (RDMA_CORE_CAP_PROT_USNIC) 573 574 struct ib_port_attr { 575 u64 subnet_prefix; 576 enum ib_port_state state; 577 enum ib_mtu max_mtu; 578 enum ib_mtu active_mtu; 579 int gid_tbl_len; 580 u32 port_cap_flags; 581 u32 max_msg_sz; 582 u32 bad_pkey_cntr; 583 u32 qkey_viol_cntr; 584 u16 pkey_tbl_len; 585 u32 sm_lid; 586 u32 lid; 587 u8 lmc; 588 u8 max_vl_num; 589 u8 sm_sl; 590 u8 subnet_timeout; 591 u8 init_type_reply; 592 u8 active_width; 593 u8 active_speed; 594 u8 phys_state; 595 bool grh_required; 596 }; 597 598 enum ib_device_modify_flags { 599 IB_DEVICE_MODIFY_SYS_IMAGE_GUID = 1 << 0, 600 IB_DEVICE_MODIFY_NODE_DESC = 1 << 1 601 }; 602 603 #define IB_DEVICE_NODE_DESC_MAX 64 604 605 struct ib_device_modify { 606 u64 sys_image_guid; 607 char node_desc[IB_DEVICE_NODE_DESC_MAX]; 608 }; 609 610 enum ib_port_modify_flags { 611 IB_PORT_SHUTDOWN = 1, 612 IB_PORT_INIT_TYPE = (1<<2), 613 IB_PORT_RESET_QKEY_CNTR = (1<<3), 614 IB_PORT_OPA_MASK_CHG = (1<<4) 615 }; 616 617 struct ib_port_modify { 618 u32 set_port_cap_mask; 619 u32 clr_port_cap_mask; 620 u8 init_type; 621 }; 622 623 enum ib_event_type { 624 IB_EVENT_CQ_ERR, 625 IB_EVENT_QP_FATAL, 626 IB_EVENT_QP_REQ_ERR, 627 IB_EVENT_QP_ACCESS_ERR, 628 IB_EVENT_COMM_EST, 629 IB_EVENT_SQ_DRAINED, 630 IB_EVENT_PATH_MIG, 631 IB_EVENT_PATH_MIG_ERR, 632 IB_EVENT_DEVICE_FATAL, 633 IB_EVENT_PORT_ACTIVE, 634 IB_EVENT_PORT_ERR, 635 IB_EVENT_LID_CHANGE, 636 IB_EVENT_PKEY_CHANGE, 637 IB_EVENT_SM_CHANGE, 638 IB_EVENT_SRQ_ERR, 639 IB_EVENT_SRQ_LIMIT_REACHED, 640 IB_EVENT_QP_LAST_WQE_REACHED, 641 IB_EVENT_CLIENT_REREGISTER, 642 IB_EVENT_GID_CHANGE, 643 IB_EVENT_WQ_FATAL, 644 }; 645 646 const char *__attribute_const__ ib_event_msg(enum ib_event_type event); 647 648 struct ib_event { 649 struct ib_device *device; 650 union { 651 struct ib_cq *cq; 652 struct ib_qp *qp; 653 struct ib_srq *srq; 654 struct ib_wq *wq; 655 u8 port_num; 656 } element; 657 enum ib_event_type event; 658 }; 659 660 struct ib_event_handler { 661 struct ib_device *device; 662 void (*handler)(struct ib_event_handler *, struct ib_event *); 663 struct list_head list; 664 }; 665 666 #define INIT_IB_EVENT_HANDLER(_ptr, _device, _handler) \ 667 do { \ 668 (_ptr)->device = _device; \ 669 (_ptr)->handler = _handler; \ 670 INIT_LIST_HEAD(&(_ptr)->list); \ 671 } while (0) 672 673 struct ib_global_route { 674 union ib_gid dgid; 675 u32 flow_label; 676 u8 sgid_index; 677 u8 hop_limit; 678 u8 traffic_class; 679 }; 680 681 struct ib_grh { 682 __be32 version_tclass_flow; 683 __be16 paylen; 684 u8 next_hdr; 685 u8 hop_limit; 686 union ib_gid sgid; 687 union ib_gid dgid; 688 }; 689 690 union rdma_network_hdr { 691 struct ib_grh ibgrh; 692 struct { 693 /* The IB spec states that if it's IPv4, the header 694 * is located in the last 20 bytes of the header. 695 */ 696 u8 reserved[20]; 697 struct iphdr roce4grh; 698 }; 699 }; 700 701 #define IB_QPN_MASK 0xFFFFFF 702 703 enum { 704 IB_MULTICAST_QPN = 0xffffff 705 }; 706 707 #define IB_LID_PERMISSIVE cpu_to_be16(0xFFFF) 708 #define IB_MULTICAST_LID_BASE cpu_to_be16(0xC000) 709 710 enum ib_ah_flags { 711 IB_AH_GRH = 1 712 }; 713 714 enum ib_rate { 715 IB_RATE_PORT_CURRENT = 0, 716 IB_RATE_2_5_GBPS = 2, 717 IB_RATE_5_GBPS = 5, 718 IB_RATE_10_GBPS = 3, 719 IB_RATE_20_GBPS = 6, 720 IB_RATE_30_GBPS = 4, 721 IB_RATE_40_GBPS = 7, 722 IB_RATE_60_GBPS = 8, 723 IB_RATE_80_GBPS = 9, 724 IB_RATE_120_GBPS = 10, 725 IB_RATE_14_GBPS = 11, 726 IB_RATE_56_GBPS = 12, 727 IB_RATE_112_GBPS = 13, 728 IB_RATE_168_GBPS = 14, 729 IB_RATE_25_GBPS = 15, 730 IB_RATE_100_GBPS = 16, 731 IB_RATE_200_GBPS = 17, 732 IB_RATE_300_GBPS = 18 733 }; 734 735 /** 736 * ib_rate_to_mult - Convert the IB rate enum to a multiple of the 737 * base rate of 2.5 Gbit/sec. For example, IB_RATE_5_GBPS will be 738 * converted to 2, since 5 Gbit/sec is 2 * 2.5 Gbit/sec. 739 * @rate: rate to convert. 740 */ 741 __attribute_const__ int ib_rate_to_mult(enum ib_rate rate); 742 743 /** 744 * ib_rate_to_mbps - Convert the IB rate enum to Mbps. 745 * For example, IB_RATE_2_5_GBPS will be converted to 2500. 746 * @rate: rate to convert. 747 */ 748 __attribute_const__ int ib_rate_to_mbps(enum ib_rate rate); 749 750 751 /** 752 * enum ib_mr_type - memory region type 753 * @IB_MR_TYPE_MEM_REG: memory region that is used for 754 * normal registration 755 * @IB_MR_TYPE_SIGNATURE: memory region that is used for 756 * signature operations (data-integrity 757 * capable regions) 758 * @IB_MR_TYPE_SG_GAPS: memory region that is capable to 759 * register any arbitrary sg lists (without 760 * the normal mr constraints - see 761 * ib_map_mr_sg) 762 */ 763 enum ib_mr_type { 764 IB_MR_TYPE_MEM_REG, 765 IB_MR_TYPE_SIGNATURE, 766 IB_MR_TYPE_SG_GAPS, 767 }; 768 769 /** 770 * Signature types 771 * IB_SIG_TYPE_NONE: Unprotected. 772 * IB_SIG_TYPE_T10_DIF: Type T10-DIF 773 */ 774 enum ib_signature_type { 775 IB_SIG_TYPE_NONE, 776 IB_SIG_TYPE_T10_DIF, 777 }; 778 779 /** 780 * Signature T10-DIF block-guard types 781 * IB_T10DIF_CRC: Corresponds to T10-PI mandated CRC checksum rules. 782 * IB_T10DIF_CSUM: Corresponds to IP checksum rules. 783 */ 784 enum ib_t10_dif_bg_type { 785 IB_T10DIF_CRC, 786 IB_T10DIF_CSUM 787 }; 788 789 /** 790 * struct ib_t10_dif_domain - Parameters specific for T10-DIF 791 * domain. 792 * @bg_type: T10-DIF block guard type (CRC|CSUM) 793 * @pi_interval: protection information interval. 794 * @bg: seed of guard computation. 795 * @app_tag: application tag of guard block 796 * @ref_tag: initial guard block reference tag. 797 * @ref_remap: Indicate wethear the reftag increments each block 798 * @app_escape: Indicate to skip block check if apptag=0xffff 799 * @ref_escape: Indicate to skip block check if reftag=0xffffffff 800 * @apptag_check_mask: check bitmask of application tag. 801 */ 802 struct ib_t10_dif_domain { 803 enum ib_t10_dif_bg_type bg_type; 804 u16 pi_interval; 805 u16 bg; 806 u16 app_tag; 807 u32 ref_tag; 808 bool ref_remap; 809 bool app_escape; 810 bool ref_escape; 811 u16 apptag_check_mask; 812 }; 813 814 /** 815 * struct ib_sig_domain - Parameters for signature domain 816 * @sig_type: specific signauture type 817 * @sig: union of all signature domain attributes that may 818 * be used to set domain layout. 819 */ 820 struct ib_sig_domain { 821 enum ib_signature_type sig_type; 822 union { 823 struct ib_t10_dif_domain dif; 824 } sig; 825 }; 826 827 /** 828 * struct ib_sig_attrs - Parameters for signature handover operation 829 * @check_mask: bitmask for signature byte check (8 bytes) 830 * @mem: memory domain layout desciptor. 831 * @wire: wire domain layout desciptor. 832 */ 833 struct ib_sig_attrs { 834 u8 check_mask; 835 struct ib_sig_domain mem; 836 struct ib_sig_domain wire; 837 }; 838 839 enum ib_sig_err_type { 840 IB_SIG_BAD_GUARD, 841 IB_SIG_BAD_REFTAG, 842 IB_SIG_BAD_APPTAG, 843 }; 844 845 /** 846 * struct ib_sig_err - signature error descriptor 847 */ 848 struct ib_sig_err { 849 enum ib_sig_err_type err_type; 850 u32 expected; 851 u32 actual; 852 u64 sig_err_offset; 853 u32 key; 854 }; 855 856 enum ib_mr_status_check { 857 IB_MR_CHECK_SIG_STATUS = 1, 858 }; 859 860 /** 861 * struct ib_mr_status - Memory region status container 862 * 863 * @fail_status: Bitmask of MR checks status. For each 864 * failed check a corresponding status bit is set. 865 * @sig_err: Additional info for IB_MR_CEHCK_SIG_STATUS 866 * failure. 867 */ 868 struct ib_mr_status { 869 u32 fail_status; 870 struct ib_sig_err sig_err; 871 }; 872 873 /** 874 * mult_to_ib_rate - Convert a multiple of 2.5 Gbit/sec to an IB rate 875 * enum. 876 * @mult: multiple to convert. 877 */ 878 __attribute_const__ enum ib_rate mult_to_ib_rate(int mult); 879 880 enum rdma_ah_attr_type { 881 RDMA_AH_ATTR_TYPE_IB, 882 RDMA_AH_ATTR_TYPE_ROCE, 883 RDMA_AH_ATTR_TYPE_OPA, 884 }; 885 886 struct ib_ah_attr { 887 u16 dlid; 888 u8 src_path_bits; 889 }; 890 891 struct roce_ah_attr { 892 u8 dmac[ETH_ALEN]; 893 }; 894 895 struct opa_ah_attr { 896 u32 dlid; 897 u8 src_path_bits; 898 bool make_grd; 899 }; 900 901 struct rdma_ah_attr { 902 struct ib_global_route grh; 903 u8 sl; 904 u8 static_rate; 905 u8 port_num; 906 u8 ah_flags; 907 enum rdma_ah_attr_type type; 908 union { 909 struct ib_ah_attr ib; 910 struct roce_ah_attr roce; 911 struct opa_ah_attr opa; 912 }; 913 }; 914 915 enum ib_wc_status { 916 IB_WC_SUCCESS, 917 IB_WC_LOC_LEN_ERR, 918 IB_WC_LOC_QP_OP_ERR, 919 IB_WC_LOC_EEC_OP_ERR, 920 IB_WC_LOC_PROT_ERR, 921 IB_WC_WR_FLUSH_ERR, 922 IB_WC_MW_BIND_ERR, 923 IB_WC_BAD_RESP_ERR, 924 IB_WC_LOC_ACCESS_ERR, 925 IB_WC_REM_INV_REQ_ERR, 926 IB_WC_REM_ACCESS_ERR, 927 IB_WC_REM_OP_ERR, 928 IB_WC_RETRY_EXC_ERR, 929 IB_WC_RNR_RETRY_EXC_ERR, 930 IB_WC_LOC_RDD_VIOL_ERR, 931 IB_WC_REM_INV_RD_REQ_ERR, 932 IB_WC_REM_ABORT_ERR, 933 IB_WC_INV_EECN_ERR, 934 IB_WC_INV_EEC_STATE_ERR, 935 IB_WC_FATAL_ERR, 936 IB_WC_RESP_TIMEOUT_ERR, 937 IB_WC_GENERAL_ERR 938 }; 939 940 const char *__attribute_const__ ib_wc_status_msg(enum ib_wc_status status); 941 942 enum ib_wc_opcode { 943 IB_WC_SEND, 944 IB_WC_RDMA_WRITE, 945 IB_WC_RDMA_READ, 946 IB_WC_COMP_SWAP, 947 IB_WC_FETCH_ADD, 948 IB_WC_LSO, 949 IB_WC_LOCAL_INV, 950 IB_WC_REG_MR, 951 IB_WC_MASKED_COMP_SWAP, 952 IB_WC_MASKED_FETCH_ADD, 953 /* 954 * Set value of IB_WC_RECV so consumers can test if a completion is a 955 * receive by testing (opcode & IB_WC_RECV). 956 */ 957 IB_WC_RECV = 1 << 7, 958 IB_WC_RECV_RDMA_WITH_IMM 959 }; 960 961 enum ib_wc_flags { 962 IB_WC_GRH = 1, 963 IB_WC_WITH_IMM = (1<<1), 964 IB_WC_WITH_INVALIDATE = (1<<2), 965 IB_WC_IP_CSUM_OK = (1<<3), 966 IB_WC_WITH_SMAC = (1<<4), 967 IB_WC_WITH_VLAN = (1<<5), 968 IB_WC_WITH_NETWORK_HDR_TYPE = (1<<6), 969 }; 970 971 struct ib_wc { 972 union { 973 u64 wr_id; 974 struct ib_cqe *wr_cqe; 975 }; 976 enum ib_wc_status status; 977 enum ib_wc_opcode opcode; 978 u32 vendor_err; 979 u32 byte_len; 980 struct ib_qp *qp; 981 union { 982 __be32 imm_data; 983 u32 invalidate_rkey; 984 } ex; 985 u32 src_qp; 986 int wc_flags; 987 u16 pkey_index; 988 u32 slid; 989 u8 sl; 990 u8 dlid_path_bits; 991 u8 port_num; /* valid only for DR SMPs on switches */ 992 u8 smac[ETH_ALEN]; 993 u16 vlan_id; 994 u8 network_hdr_type; 995 }; 996 997 enum ib_cq_notify_flags { 998 IB_CQ_SOLICITED = 1 << 0, 999 IB_CQ_NEXT_COMP = 1 << 1, 1000 IB_CQ_SOLICITED_MASK = IB_CQ_SOLICITED | IB_CQ_NEXT_COMP, 1001 IB_CQ_REPORT_MISSED_EVENTS = 1 << 2, 1002 }; 1003 1004 enum ib_srq_type { 1005 IB_SRQT_BASIC, 1006 IB_SRQT_XRC, 1007 IB_SRQT_TM, 1008 }; 1009 1010 static inline bool ib_srq_has_cq(enum ib_srq_type srq_type) 1011 { 1012 return srq_type == IB_SRQT_XRC || 1013 srq_type == IB_SRQT_TM; 1014 } 1015 1016 enum ib_srq_attr_mask { 1017 IB_SRQ_MAX_WR = 1 << 0, 1018 IB_SRQ_LIMIT = 1 << 1, 1019 }; 1020 1021 struct ib_srq_attr { 1022 u32 max_wr; 1023 u32 max_sge; 1024 u32 srq_limit; 1025 }; 1026 1027 struct ib_srq_init_attr { 1028 void (*event_handler)(struct ib_event *, void *); 1029 void *srq_context; 1030 struct ib_srq_attr attr; 1031 enum ib_srq_type srq_type; 1032 1033 struct { 1034 struct ib_cq *cq; 1035 union { 1036 struct { 1037 struct ib_xrcd *xrcd; 1038 } xrc; 1039 1040 struct { 1041 u32 max_num_tags; 1042 } tag_matching; 1043 }; 1044 } ext; 1045 }; 1046 1047 struct ib_qp_cap { 1048 u32 max_send_wr; 1049 u32 max_recv_wr; 1050 u32 max_send_sge; 1051 u32 max_recv_sge; 1052 u32 max_inline_data; 1053 1054 /* 1055 * Maximum number of rdma_rw_ctx structures in flight at a time. 1056 * ib_create_qp() will calculate the right amount of neededed WRs 1057 * and MRs based on this. 1058 */ 1059 u32 max_rdma_ctxs; 1060 }; 1061 1062 enum ib_sig_type { 1063 IB_SIGNAL_ALL_WR, 1064 IB_SIGNAL_REQ_WR 1065 }; 1066 1067 enum ib_qp_type { 1068 /* 1069 * IB_QPT_SMI and IB_QPT_GSI have to be the first two entries 1070 * here (and in that order) since the MAD layer uses them as 1071 * indices into a 2-entry table. 1072 */ 1073 IB_QPT_SMI, 1074 IB_QPT_GSI, 1075 1076 IB_QPT_RC, 1077 IB_QPT_UC, 1078 IB_QPT_UD, 1079 IB_QPT_RAW_IPV6, 1080 IB_QPT_RAW_ETHERTYPE, 1081 IB_QPT_RAW_PACKET = 8, 1082 IB_QPT_XRC_INI = 9, 1083 IB_QPT_XRC_TGT, 1084 IB_QPT_MAX, 1085 /* Reserve a range for qp types internal to the low level driver. 1086 * These qp types will not be visible at the IB core layer, so the 1087 * IB_QPT_MAX usages should not be affected in the core layer 1088 */ 1089 IB_QPT_RESERVED1 = 0x1000, 1090 IB_QPT_RESERVED2, 1091 IB_QPT_RESERVED3, 1092 IB_QPT_RESERVED4, 1093 IB_QPT_RESERVED5, 1094 IB_QPT_RESERVED6, 1095 IB_QPT_RESERVED7, 1096 IB_QPT_RESERVED8, 1097 IB_QPT_RESERVED9, 1098 IB_QPT_RESERVED10, 1099 }; 1100 1101 enum ib_qp_create_flags { 1102 IB_QP_CREATE_IPOIB_UD_LSO = 1 << 0, 1103 IB_QP_CREATE_BLOCK_MULTICAST_LOOPBACK = 1 << 1, 1104 IB_QP_CREATE_CROSS_CHANNEL = 1 << 2, 1105 IB_QP_CREATE_MANAGED_SEND = 1 << 3, 1106 IB_QP_CREATE_MANAGED_RECV = 1 << 4, 1107 IB_QP_CREATE_NETIF_QP = 1 << 5, 1108 IB_QP_CREATE_SIGNATURE_EN = 1 << 6, 1109 /* FREE = 1 << 7, */ 1110 IB_QP_CREATE_SCATTER_FCS = 1 << 8, 1111 IB_QP_CREATE_CVLAN_STRIPPING = 1 << 9, 1112 IB_QP_CREATE_SOURCE_QPN = 1 << 10, 1113 IB_QP_CREATE_PCI_WRITE_END_PADDING = 1 << 11, 1114 /* reserve bits 26-31 for low level drivers' internal use */ 1115 IB_QP_CREATE_RESERVED_START = 1 << 26, 1116 IB_QP_CREATE_RESERVED_END = 1 << 31, 1117 }; 1118 1119 /* 1120 * Note: users may not call ib_close_qp or ib_destroy_qp from the event_handler 1121 * callback to destroy the passed in QP. 1122 */ 1123 1124 struct ib_qp_init_attr { 1125 void (*event_handler)(struct ib_event *, void *); 1126 void *qp_context; 1127 struct ib_cq *send_cq; 1128 struct ib_cq *recv_cq; 1129 struct ib_srq *srq; 1130 struct ib_xrcd *xrcd; /* XRC TGT QPs only */ 1131 struct ib_qp_cap cap; 1132 enum ib_sig_type sq_sig_type; 1133 enum ib_qp_type qp_type; 1134 enum ib_qp_create_flags create_flags; 1135 1136 /* 1137 * Only needed for special QP types, or when using the RW API. 1138 */ 1139 u8 port_num; 1140 struct ib_rwq_ind_table *rwq_ind_tbl; 1141 u32 source_qpn; 1142 }; 1143 1144 struct ib_qp_open_attr { 1145 void (*event_handler)(struct ib_event *, void *); 1146 void *qp_context; 1147 u32 qp_num; 1148 enum ib_qp_type qp_type; 1149 }; 1150 1151 enum ib_rnr_timeout { 1152 IB_RNR_TIMER_655_36 = 0, 1153 IB_RNR_TIMER_000_01 = 1, 1154 IB_RNR_TIMER_000_02 = 2, 1155 IB_RNR_TIMER_000_03 = 3, 1156 IB_RNR_TIMER_000_04 = 4, 1157 IB_RNR_TIMER_000_06 = 5, 1158 IB_RNR_TIMER_000_08 = 6, 1159 IB_RNR_TIMER_000_12 = 7, 1160 IB_RNR_TIMER_000_16 = 8, 1161 IB_RNR_TIMER_000_24 = 9, 1162 IB_RNR_TIMER_000_32 = 10, 1163 IB_RNR_TIMER_000_48 = 11, 1164 IB_RNR_TIMER_000_64 = 12, 1165 IB_RNR_TIMER_000_96 = 13, 1166 IB_RNR_TIMER_001_28 = 14, 1167 IB_RNR_TIMER_001_92 = 15, 1168 IB_RNR_TIMER_002_56 = 16, 1169 IB_RNR_TIMER_003_84 = 17, 1170 IB_RNR_TIMER_005_12 = 18, 1171 IB_RNR_TIMER_007_68 = 19, 1172 IB_RNR_TIMER_010_24 = 20, 1173 IB_RNR_TIMER_015_36 = 21, 1174 IB_RNR_TIMER_020_48 = 22, 1175 IB_RNR_TIMER_030_72 = 23, 1176 IB_RNR_TIMER_040_96 = 24, 1177 IB_RNR_TIMER_061_44 = 25, 1178 IB_RNR_TIMER_081_92 = 26, 1179 IB_RNR_TIMER_122_88 = 27, 1180 IB_RNR_TIMER_163_84 = 28, 1181 IB_RNR_TIMER_245_76 = 29, 1182 IB_RNR_TIMER_327_68 = 30, 1183 IB_RNR_TIMER_491_52 = 31 1184 }; 1185 1186 enum ib_qp_attr_mask { 1187 IB_QP_STATE = 1, 1188 IB_QP_CUR_STATE = (1<<1), 1189 IB_QP_EN_SQD_ASYNC_NOTIFY = (1<<2), 1190 IB_QP_ACCESS_FLAGS = (1<<3), 1191 IB_QP_PKEY_INDEX = (1<<4), 1192 IB_QP_PORT = (1<<5), 1193 IB_QP_QKEY = (1<<6), 1194 IB_QP_AV = (1<<7), 1195 IB_QP_PATH_MTU = (1<<8), 1196 IB_QP_TIMEOUT = (1<<9), 1197 IB_QP_RETRY_CNT = (1<<10), 1198 IB_QP_RNR_RETRY = (1<<11), 1199 IB_QP_RQ_PSN = (1<<12), 1200 IB_QP_MAX_QP_RD_ATOMIC = (1<<13), 1201 IB_QP_ALT_PATH = (1<<14), 1202 IB_QP_MIN_RNR_TIMER = (1<<15), 1203 IB_QP_SQ_PSN = (1<<16), 1204 IB_QP_MAX_DEST_RD_ATOMIC = (1<<17), 1205 IB_QP_PATH_MIG_STATE = (1<<18), 1206 IB_QP_CAP = (1<<19), 1207 IB_QP_DEST_QPN = (1<<20), 1208 IB_QP_RESERVED1 = (1<<21), 1209 IB_QP_RESERVED2 = (1<<22), 1210 IB_QP_RESERVED3 = (1<<23), 1211 IB_QP_RESERVED4 = (1<<24), 1212 IB_QP_RATE_LIMIT = (1<<25), 1213 }; 1214 1215 enum ib_qp_state { 1216 IB_QPS_RESET, 1217 IB_QPS_INIT, 1218 IB_QPS_RTR, 1219 IB_QPS_RTS, 1220 IB_QPS_SQD, 1221 IB_QPS_SQE, 1222 IB_QPS_ERR 1223 }; 1224 1225 enum ib_mig_state { 1226 IB_MIG_MIGRATED, 1227 IB_MIG_REARM, 1228 IB_MIG_ARMED 1229 }; 1230 1231 enum ib_mw_type { 1232 IB_MW_TYPE_1 = 1, 1233 IB_MW_TYPE_2 = 2 1234 }; 1235 1236 struct ib_qp_attr { 1237 enum ib_qp_state qp_state; 1238 enum ib_qp_state cur_qp_state; 1239 enum ib_mtu path_mtu; 1240 enum ib_mig_state path_mig_state; 1241 u32 qkey; 1242 u32 rq_psn; 1243 u32 sq_psn; 1244 u32 dest_qp_num; 1245 int qp_access_flags; 1246 struct ib_qp_cap cap; 1247 struct rdma_ah_attr ah_attr; 1248 struct rdma_ah_attr alt_ah_attr; 1249 u16 pkey_index; 1250 u16 alt_pkey_index; 1251 u8 en_sqd_async_notify; 1252 u8 sq_draining; 1253 u8 max_rd_atomic; 1254 u8 max_dest_rd_atomic; 1255 u8 min_rnr_timer; 1256 u8 port_num; 1257 u8 timeout; 1258 u8 retry_cnt; 1259 u8 rnr_retry; 1260 u8 alt_port_num; 1261 u8 alt_timeout; 1262 u32 rate_limit; 1263 }; 1264 1265 enum ib_wr_opcode { 1266 IB_WR_RDMA_WRITE, 1267 IB_WR_RDMA_WRITE_WITH_IMM, 1268 IB_WR_SEND, 1269 IB_WR_SEND_WITH_IMM, 1270 IB_WR_RDMA_READ, 1271 IB_WR_ATOMIC_CMP_AND_SWP, 1272 IB_WR_ATOMIC_FETCH_AND_ADD, 1273 IB_WR_LSO, 1274 IB_WR_SEND_WITH_INV, 1275 IB_WR_RDMA_READ_WITH_INV, 1276 IB_WR_LOCAL_INV, 1277 IB_WR_REG_MR, 1278 IB_WR_MASKED_ATOMIC_CMP_AND_SWP, 1279 IB_WR_MASKED_ATOMIC_FETCH_AND_ADD, 1280 IB_WR_REG_SIG_MR, 1281 /* reserve values for low level drivers' internal use. 1282 * These values will not be used at all in the ib core layer. 1283 */ 1284 IB_WR_RESERVED1 = 0xf0, 1285 IB_WR_RESERVED2, 1286 IB_WR_RESERVED3, 1287 IB_WR_RESERVED4, 1288 IB_WR_RESERVED5, 1289 IB_WR_RESERVED6, 1290 IB_WR_RESERVED7, 1291 IB_WR_RESERVED8, 1292 IB_WR_RESERVED9, 1293 IB_WR_RESERVED10, 1294 }; 1295 1296 enum ib_send_flags { 1297 IB_SEND_FENCE = 1, 1298 IB_SEND_SIGNALED = (1<<1), 1299 IB_SEND_SOLICITED = (1<<2), 1300 IB_SEND_INLINE = (1<<3), 1301 IB_SEND_IP_CSUM = (1<<4), 1302 1303 /* reserve bits 26-31 for low level drivers' internal use */ 1304 IB_SEND_RESERVED_START = (1 << 26), 1305 IB_SEND_RESERVED_END = (1 << 31), 1306 }; 1307 1308 struct ib_sge { 1309 u64 addr; 1310 u32 length; 1311 u32 lkey; 1312 }; 1313 1314 struct ib_cqe { 1315 void (*done)(struct ib_cq *cq, struct ib_wc *wc); 1316 }; 1317 1318 struct ib_send_wr { 1319 struct ib_send_wr *next; 1320 union { 1321 u64 wr_id; 1322 struct ib_cqe *wr_cqe; 1323 }; 1324 struct ib_sge *sg_list; 1325 int num_sge; 1326 enum ib_wr_opcode opcode; 1327 int send_flags; 1328 union { 1329 __be32 imm_data; 1330 u32 invalidate_rkey; 1331 } ex; 1332 }; 1333 1334 struct ib_rdma_wr { 1335 struct ib_send_wr wr; 1336 u64 remote_addr; 1337 u32 rkey; 1338 }; 1339 1340 static inline struct ib_rdma_wr *rdma_wr(struct ib_send_wr *wr) 1341 { 1342 return container_of(wr, struct ib_rdma_wr, wr); 1343 } 1344 1345 struct ib_atomic_wr { 1346 struct ib_send_wr wr; 1347 u64 remote_addr; 1348 u64 compare_add; 1349 u64 swap; 1350 u64 compare_add_mask; 1351 u64 swap_mask; 1352 u32 rkey; 1353 }; 1354 1355 static inline struct ib_atomic_wr *atomic_wr(struct ib_send_wr *wr) 1356 { 1357 return container_of(wr, struct ib_atomic_wr, wr); 1358 } 1359 1360 struct ib_ud_wr { 1361 struct ib_send_wr wr; 1362 struct ib_ah *ah; 1363 void *header; 1364 int hlen; 1365 int mss; 1366 u32 remote_qpn; 1367 u32 remote_qkey; 1368 u16 pkey_index; /* valid for GSI only */ 1369 u8 port_num; /* valid for DR SMPs on switch only */ 1370 }; 1371 1372 static inline struct ib_ud_wr *ud_wr(struct ib_send_wr *wr) 1373 { 1374 return container_of(wr, struct ib_ud_wr, wr); 1375 } 1376 1377 struct ib_reg_wr { 1378 struct ib_send_wr wr; 1379 struct ib_mr *mr; 1380 u32 key; 1381 int access; 1382 }; 1383 1384 static inline struct ib_reg_wr *reg_wr(struct ib_send_wr *wr) 1385 { 1386 return container_of(wr, struct ib_reg_wr, wr); 1387 } 1388 1389 struct ib_sig_handover_wr { 1390 struct ib_send_wr wr; 1391 struct ib_sig_attrs *sig_attrs; 1392 struct ib_mr *sig_mr; 1393 int access_flags; 1394 struct ib_sge *prot; 1395 }; 1396 1397 static inline struct ib_sig_handover_wr *sig_handover_wr(struct ib_send_wr *wr) 1398 { 1399 return container_of(wr, struct ib_sig_handover_wr, wr); 1400 } 1401 1402 struct ib_recv_wr { 1403 struct ib_recv_wr *next; 1404 union { 1405 u64 wr_id; 1406 struct ib_cqe *wr_cqe; 1407 }; 1408 struct ib_sge *sg_list; 1409 int num_sge; 1410 }; 1411 1412 enum ib_access_flags { 1413 IB_ACCESS_LOCAL_WRITE = 1, 1414 IB_ACCESS_REMOTE_WRITE = (1<<1), 1415 IB_ACCESS_REMOTE_READ = (1<<2), 1416 IB_ACCESS_REMOTE_ATOMIC = (1<<3), 1417 IB_ACCESS_MW_BIND = (1<<4), 1418 IB_ZERO_BASED = (1<<5), 1419 IB_ACCESS_ON_DEMAND = (1<<6), 1420 IB_ACCESS_HUGETLB = (1<<7), 1421 }; 1422 1423 /* 1424 * XXX: these are apparently used for ->rereg_user_mr, no idea why they 1425 * are hidden here instead of a uapi header! 1426 */ 1427 enum ib_mr_rereg_flags { 1428 IB_MR_REREG_TRANS = 1, 1429 IB_MR_REREG_PD = (1<<1), 1430 IB_MR_REREG_ACCESS = (1<<2), 1431 IB_MR_REREG_SUPPORTED = ((IB_MR_REREG_ACCESS << 1) - 1) 1432 }; 1433 1434 struct ib_fmr_attr { 1435 int max_pages; 1436 int max_maps; 1437 u8 page_shift; 1438 }; 1439 1440 struct ib_umem; 1441 1442 enum rdma_remove_reason { 1443 /* Userspace requested uobject deletion. Call could fail */ 1444 RDMA_REMOVE_DESTROY, 1445 /* Context deletion. This call should delete the actual object itself */ 1446 RDMA_REMOVE_CLOSE, 1447 /* Driver is being hot-unplugged. This call should delete the actual object itself */ 1448 RDMA_REMOVE_DRIVER_REMOVE, 1449 /* Context is being cleaned-up, but commit was just completed */ 1450 RDMA_REMOVE_DURING_CLEANUP, 1451 }; 1452 1453 struct ib_rdmacg_object { 1454 #ifdef CONFIG_CGROUP_RDMA 1455 struct rdma_cgroup *cg; /* owner rdma cgroup */ 1456 #endif 1457 }; 1458 1459 struct ib_ucontext { 1460 struct ib_device *device; 1461 struct ib_uverbs_file *ufile; 1462 int closing; 1463 1464 /* locking the uobjects_list */ 1465 struct mutex uobjects_lock; 1466 struct list_head uobjects; 1467 /* protects cleanup process from other actions */ 1468 struct rw_semaphore cleanup_rwsem; 1469 enum rdma_remove_reason cleanup_reason; 1470 1471 struct pid *tgid; 1472 #ifdef CONFIG_INFINIBAND_ON_DEMAND_PAGING 1473 struct rb_root_cached umem_tree; 1474 /* 1475 * Protects .umem_rbroot and tree, as well as odp_mrs_count and 1476 * mmu notifiers registration. 1477 */ 1478 struct rw_semaphore umem_rwsem; 1479 void (*invalidate_range)(struct ib_umem *umem, 1480 unsigned long start, unsigned long end); 1481 1482 struct mmu_notifier mn; 1483 atomic_t notifier_count; 1484 /* A list of umems that don't have private mmu notifier counters yet. */ 1485 struct list_head no_private_counters; 1486 int odp_mrs_count; 1487 #endif 1488 1489 struct ib_rdmacg_object cg_obj; 1490 }; 1491 1492 struct ib_uobject { 1493 u64 user_handle; /* handle given to us by userspace */ 1494 struct ib_ucontext *context; /* associated user context */ 1495 void *object; /* containing object */ 1496 struct list_head list; /* link to context's list */ 1497 struct ib_rdmacg_object cg_obj; /* rdmacg object */ 1498 int id; /* index into kernel idr */ 1499 struct kref ref; 1500 atomic_t usecnt; /* protects exclusive access */ 1501 struct rcu_head rcu; /* kfree_rcu() overhead */ 1502 1503 const struct uverbs_obj_type *type; 1504 }; 1505 1506 struct ib_uobject_file { 1507 struct ib_uobject uobj; 1508 /* ufile contains the lock between context release and file close */ 1509 struct ib_uverbs_file *ufile; 1510 }; 1511 1512 struct ib_udata { 1513 const void __user *inbuf; 1514 void __user *outbuf; 1515 size_t inlen; 1516 size_t outlen; 1517 }; 1518 1519 struct ib_pd { 1520 u32 local_dma_lkey; 1521 u32 flags; 1522 struct ib_device *device; 1523 struct ib_uobject *uobject; 1524 atomic_t usecnt; /* count all resources */ 1525 1526 u32 unsafe_global_rkey; 1527 1528 /* 1529 * Implementation details of the RDMA core, don't use in drivers: 1530 */ 1531 struct ib_mr *__internal_mr; 1532 }; 1533 1534 struct ib_xrcd { 1535 struct ib_device *device; 1536 atomic_t usecnt; /* count all exposed resources */ 1537 struct inode *inode; 1538 1539 struct mutex tgt_qp_mutex; 1540 struct list_head tgt_qp_list; 1541 }; 1542 1543 struct ib_ah { 1544 struct ib_device *device; 1545 struct ib_pd *pd; 1546 struct ib_uobject *uobject; 1547 enum rdma_ah_attr_type type; 1548 }; 1549 1550 typedef void (*ib_comp_handler)(struct ib_cq *cq, void *cq_context); 1551 1552 enum ib_poll_context { 1553 IB_POLL_DIRECT, /* caller context, no hw completions */ 1554 IB_POLL_SOFTIRQ, /* poll from softirq context */ 1555 IB_POLL_WORKQUEUE, /* poll from workqueue */ 1556 }; 1557 1558 struct ib_cq { 1559 struct ib_device *device; 1560 struct ib_uobject *uobject; 1561 ib_comp_handler comp_handler; 1562 void (*event_handler)(struct ib_event *, void *); 1563 void *cq_context; 1564 int cqe; 1565 atomic_t usecnt; /* count number of work queues */ 1566 enum ib_poll_context poll_ctx; 1567 struct ib_wc *wc; 1568 union { 1569 struct irq_poll iop; 1570 struct work_struct work; 1571 }; 1572 }; 1573 1574 struct ib_srq { 1575 struct ib_device *device; 1576 struct ib_pd *pd; 1577 struct ib_uobject *uobject; 1578 void (*event_handler)(struct ib_event *, void *); 1579 void *srq_context; 1580 enum ib_srq_type srq_type; 1581 atomic_t usecnt; 1582 1583 struct { 1584 struct ib_cq *cq; 1585 union { 1586 struct { 1587 struct ib_xrcd *xrcd; 1588 u32 srq_num; 1589 } xrc; 1590 }; 1591 } ext; 1592 }; 1593 1594 enum ib_raw_packet_caps { 1595 /* Strip cvlan from incoming packet and report it in the matching work 1596 * completion is supported. 1597 */ 1598 IB_RAW_PACKET_CAP_CVLAN_STRIPPING = (1 << 0), 1599 /* Scatter FCS field of an incoming packet to host memory is supported. 1600 */ 1601 IB_RAW_PACKET_CAP_SCATTER_FCS = (1 << 1), 1602 /* Checksum offloads are supported (for both send and receive). */ 1603 IB_RAW_PACKET_CAP_IP_CSUM = (1 << 2), 1604 /* When a packet is received for an RQ with no receive WQEs, the 1605 * packet processing is delayed. 1606 */ 1607 IB_RAW_PACKET_CAP_DELAY_DROP = (1 << 3), 1608 }; 1609 1610 enum ib_wq_type { 1611 IB_WQT_RQ 1612 }; 1613 1614 enum ib_wq_state { 1615 IB_WQS_RESET, 1616 IB_WQS_RDY, 1617 IB_WQS_ERR 1618 }; 1619 1620 struct ib_wq { 1621 struct ib_device *device; 1622 struct ib_uobject *uobject; 1623 void *wq_context; 1624 void (*event_handler)(struct ib_event *, void *); 1625 struct ib_pd *pd; 1626 struct ib_cq *cq; 1627 u32 wq_num; 1628 enum ib_wq_state state; 1629 enum ib_wq_type wq_type; 1630 atomic_t usecnt; 1631 }; 1632 1633 enum ib_wq_flags { 1634 IB_WQ_FLAGS_CVLAN_STRIPPING = 1 << 0, 1635 IB_WQ_FLAGS_SCATTER_FCS = 1 << 1, 1636 IB_WQ_FLAGS_DELAY_DROP = 1 << 2, 1637 IB_WQ_FLAGS_PCI_WRITE_END_PADDING = 1 << 3, 1638 }; 1639 1640 struct ib_wq_init_attr { 1641 void *wq_context; 1642 enum ib_wq_type wq_type; 1643 u32 max_wr; 1644 u32 max_sge; 1645 struct ib_cq *cq; 1646 void (*event_handler)(struct ib_event *, void *); 1647 u32 create_flags; /* Use enum ib_wq_flags */ 1648 }; 1649 1650 enum ib_wq_attr_mask { 1651 IB_WQ_STATE = 1 << 0, 1652 IB_WQ_CUR_STATE = 1 << 1, 1653 IB_WQ_FLAGS = 1 << 2, 1654 }; 1655 1656 struct ib_wq_attr { 1657 enum ib_wq_state wq_state; 1658 enum ib_wq_state curr_wq_state; 1659 u32 flags; /* Use enum ib_wq_flags */ 1660 u32 flags_mask; /* Use enum ib_wq_flags */ 1661 }; 1662 1663 struct ib_rwq_ind_table { 1664 struct ib_device *device; 1665 struct ib_uobject *uobject; 1666 atomic_t usecnt; 1667 u32 ind_tbl_num; 1668 u32 log_ind_tbl_size; 1669 struct ib_wq **ind_tbl; 1670 }; 1671 1672 struct ib_rwq_ind_table_init_attr { 1673 u32 log_ind_tbl_size; 1674 /* Each entry is a pointer to Receive Work Queue */ 1675 struct ib_wq **ind_tbl; 1676 }; 1677 1678 enum port_pkey_state { 1679 IB_PORT_PKEY_NOT_VALID = 0, 1680 IB_PORT_PKEY_VALID = 1, 1681 IB_PORT_PKEY_LISTED = 2, 1682 }; 1683 1684 struct ib_qp_security; 1685 1686 struct ib_port_pkey { 1687 enum port_pkey_state state; 1688 u16 pkey_index; 1689 u8 port_num; 1690 struct list_head qp_list; 1691 struct list_head to_error_list; 1692 struct ib_qp_security *sec; 1693 }; 1694 1695 struct ib_ports_pkeys { 1696 struct ib_port_pkey main; 1697 struct ib_port_pkey alt; 1698 }; 1699 1700 struct ib_qp_security { 1701 struct ib_qp *qp; 1702 struct ib_device *dev; 1703 /* Hold this mutex when changing port and pkey settings. */ 1704 struct mutex mutex; 1705 struct ib_ports_pkeys *ports_pkeys; 1706 /* A list of all open shared QP handles. Required to enforce security 1707 * properly for all users of a shared QP. 1708 */ 1709 struct list_head shared_qp_list; 1710 void *security; 1711 bool destroying; 1712 atomic_t error_list_count; 1713 struct completion error_complete; 1714 int error_comps_pending; 1715 }; 1716 1717 /* 1718 * @max_write_sge: Maximum SGE elements per RDMA WRITE request. 1719 * @max_read_sge: Maximum SGE elements per RDMA READ request. 1720 */ 1721 struct ib_qp { 1722 struct ib_device *device; 1723 struct ib_pd *pd; 1724 struct ib_cq *send_cq; 1725 struct ib_cq *recv_cq; 1726 spinlock_t mr_lock; 1727 int mrs_used; 1728 struct list_head rdma_mrs; 1729 struct list_head sig_mrs; 1730 struct ib_srq *srq; 1731 struct ib_xrcd *xrcd; /* XRC TGT QPs only */ 1732 struct list_head xrcd_list; 1733 1734 /* count times opened, mcast attaches, flow attaches */ 1735 atomic_t usecnt; 1736 struct list_head open_list; 1737 struct ib_qp *real_qp; 1738 struct ib_uobject *uobject; 1739 void (*event_handler)(struct ib_event *, void *); 1740 void *qp_context; 1741 u32 qp_num; 1742 u32 max_write_sge; 1743 u32 max_read_sge; 1744 enum ib_qp_type qp_type; 1745 struct ib_rwq_ind_table *rwq_ind_tbl; 1746 struct ib_qp_security *qp_sec; 1747 u8 port; 1748 }; 1749 1750 struct ib_mr { 1751 struct ib_device *device; 1752 struct ib_pd *pd; 1753 u32 lkey; 1754 u32 rkey; 1755 u64 iova; 1756 u64 length; 1757 unsigned int page_size; 1758 bool need_inval; 1759 union { 1760 struct ib_uobject *uobject; /* user */ 1761 struct list_head qp_entry; /* FR */ 1762 }; 1763 }; 1764 1765 struct ib_mw { 1766 struct ib_device *device; 1767 struct ib_pd *pd; 1768 struct ib_uobject *uobject; 1769 u32 rkey; 1770 enum ib_mw_type type; 1771 }; 1772 1773 struct ib_fmr { 1774 struct ib_device *device; 1775 struct ib_pd *pd; 1776 struct list_head list; 1777 u32 lkey; 1778 u32 rkey; 1779 }; 1780 1781 /* Supported steering options */ 1782 enum ib_flow_attr_type { 1783 /* steering according to rule specifications */ 1784 IB_FLOW_ATTR_NORMAL = 0x0, 1785 /* default unicast and multicast rule - 1786 * receive all Eth traffic which isn't steered to any QP 1787 */ 1788 IB_FLOW_ATTR_ALL_DEFAULT = 0x1, 1789 /* default multicast rule - 1790 * receive all Eth multicast traffic which isn't steered to any QP 1791 */ 1792 IB_FLOW_ATTR_MC_DEFAULT = 0x2, 1793 /* sniffer rule - receive all port traffic */ 1794 IB_FLOW_ATTR_SNIFFER = 0x3 1795 }; 1796 1797 /* Supported steering header types */ 1798 enum ib_flow_spec_type { 1799 /* L2 headers*/ 1800 IB_FLOW_SPEC_ETH = 0x20, 1801 IB_FLOW_SPEC_IB = 0x22, 1802 /* L3 header*/ 1803 IB_FLOW_SPEC_IPV4 = 0x30, 1804 IB_FLOW_SPEC_IPV6 = 0x31, 1805 /* L4 headers*/ 1806 IB_FLOW_SPEC_TCP = 0x40, 1807 IB_FLOW_SPEC_UDP = 0x41, 1808 IB_FLOW_SPEC_VXLAN_TUNNEL = 0x50, 1809 IB_FLOW_SPEC_INNER = 0x100, 1810 /* Actions */ 1811 IB_FLOW_SPEC_ACTION_TAG = 0x1000, 1812 IB_FLOW_SPEC_ACTION_DROP = 0x1001, 1813 }; 1814 #define IB_FLOW_SPEC_LAYER_MASK 0xF0 1815 #define IB_FLOW_SPEC_SUPPORT_LAYERS 8 1816 1817 /* Flow steering rule priority is set according to it's domain. 1818 * Lower domain value means higher priority. 1819 */ 1820 enum ib_flow_domain { 1821 IB_FLOW_DOMAIN_USER, 1822 IB_FLOW_DOMAIN_ETHTOOL, 1823 IB_FLOW_DOMAIN_RFS, 1824 IB_FLOW_DOMAIN_NIC, 1825 IB_FLOW_DOMAIN_NUM /* Must be last */ 1826 }; 1827 1828 enum ib_flow_flags { 1829 IB_FLOW_ATTR_FLAGS_DONT_TRAP = 1UL << 1, /* Continue match, no steal */ 1830 IB_FLOW_ATTR_FLAGS_RESERVED = 1UL << 2 /* Must be last */ 1831 }; 1832 1833 struct ib_flow_eth_filter { 1834 u8 dst_mac[6]; 1835 u8 src_mac[6]; 1836 __be16 ether_type; 1837 __be16 vlan_tag; 1838 /* Must be last */ 1839 u8 real_sz[0]; 1840 }; 1841 1842 struct ib_flow_spec_eth { 1843 u32 type; 1844 u16 size; 1845 struct ib_flow_eth_filter val; 1846 struct ib_flow_eth_filter mask; 1847 }; 1848 1849 struct ib_flow_ib_filter { 1850 __be16 dlid; 1851 __u8 sl; 1852 /* Must be last */ 1853 u8 real_sz[0]; 1854 }; 1855 1856 struct ib_flow_spec_ib { 1857 u32 type; 1858 u16 size; 1859 struct ib_flow_ib_filter val; 1860 struct ib_flow_ib_filter mask; 1861 }; 1862 1863 /* IPv4 header flags */ 1864 enum ib_ipv4_flags { 1865 IB_IPV4_DONT_FRAG = 0x2, /* Don't enable packet fragmentation */ 1866 IB_IPV4_MORE_FRAG = 0X4 /* For All fragmented packets except the 1867 last have this flag set */ 1868 }; 1869 1870 struct ib_flow_ipv4_filter { 1871 __be32 src_ip; 1872 __be32 dst_ip; 1873 u8 proto; 1874 u8 tos; 1875 u8 ttl; 1876 u8 flags; 1877 /* Must be last */ 1878 u8 real_sz[0]; 1879 }; 1880 1881 struct ib_flow_spec_ipv4 { 1882 u32 type; 1883 u16 size; 1884 struct ib_flow_ipv4_filter val; 1885 struct ib_flow_ipv4_filter mask; 1886 }; 1887 1888 struct ib_flow_ipv6_filter { 1889 u8 src_ip[16]; 1890 u8 dst_ip[16]; 1891 __be32 flow_label; 1892 u8 next_hdr; 1893 u8 traffic_class; 1894 u8 hop_limit; 1895 /* Must be last */ 1896 u8 real_sz[0]; 1897 }; 1898 1899 struct ib_flow_spec_ipv6 { 1900 u32 type; 1901 u16 size; 1902 struct ib_flow_ipv6_filter val; 1903 struct ib_flow_ipv6_filter mask; 1904 }; 1905 1906 struct ib_flow_tcp_udp_filter { 1907 __be16 dst_port; 1908 __be16 src_port; 1909 /* Must be last */ 1910 u8 real_sz[0]; 1911 }; 1912 1913 struct ib_flow_spec_tcp_udp { 1914 u32 type; 1915 u16 size; 1916 struct ib_flow_tcp_udp_filter val; 1917 struct ib_flow_tcp_udp_filter mask; 1918 }; 1919 1920 struct ib_flow_tunnel_filter { 1921 __be32 tunnel_id; 1922 u8 real_sz[0]; 1923 }; 1924 1925 /* ib_flow_spec_tunnel describes the Vxlan tunnel 1926 * the tunnel_id from val has the vni value 1927 */ 1928 struct ib_flow_spec_tunnel { 1929 u32 type; 1930 u16 size; 1931 struct ib_flow_tunnel_filter val; 1932 struct ib_flow_tunnel_filter mask; 1933 }; 1934 1935 struct ib_flow_spec_action_tag { 1936 enum ib_flow_spec_type type; 1937 u16 size; 1938 u32 tag_id; 1939 }; 1940 1941 struct ib_flow_spec_action_drop { 1942 enum ib_flow_spec_type type; 1943 u16 size; 1944 }; 1945 1946 union ib_flow_spec { 1947 struct { 1948 u32 type; 1949 u16 size; 1950 }; 1951 struct ib_flow_spec_eth eth; 1952 struct ib_flow_spec_ib ib; 1953 struct ib_flow_spec_ipv4 ipv4; 1954 struct ib_flow_spec_tcp_udp tcp_udp; 1955 struct ib_flow_spec_ipv6 ipv6; 1956 struct ib_flow_spec_tunnel tunnel; 1957 struct ib_flow_spec_action_tag flow_tag; 1958 struct ib_flow_spec_action_drop drop; 1959 }; 1960 1961 struct ib_flow_attr { 1962 enum ib_flow_attr_type type; 1963 u16 size; 1964 u16 priority; 1965 u32 flags; 1966 u8 num_of_specs; 1967 u8 port; 1968 /* Following are the optional layers according to user request 1969 * struct ib_flow_spec_xxx 1970 * struct ib_flow_spec_yyy 1971 */ 1972 }; 1973 1974 struct ib_flow { 1975 struct ib_qp *qp; 1976 struct ib_uobject *uobject; 1977 }; 1978 1979 struct ib_mad_hdr; 1980 struct ib_grh; 1981 1982 enum ib_process_mad_flags { 1983 IB_MAD_IGNORE_MKEY = 1, 1984 IB_MAD_IGNORE_BKEY = 2, 1985 IB_MAD_IGNORE_ALL = IB_MAD_IGNORE_MKEY | IB_MAD_IGNORE_BKEY 1986 }; 1987 1988 enum ib_mad_result { 1989 IB_MAD_RESULT_FAILURE = 0, /* (!SUCCESS is the important flag) */ 1990 IB_MAD_RESULT_SUCCESS = 1 << 0, /* MAD was successfully processed */ 1991 IB_MAD_RESULT_REPLY = 1 << 1, /* Reply packet needs to be sent */ 1992 IB_MAD_RESULT_CONSUMED = 1 << 2 /* Packet consumed: stop processing */ 1993 }; 1994 1995 struct ib_port_cache { 1996 u64 subnet_prefix; 1997 struct ib_pkey_cache *pkey; 1998 struct ib_gid_table *gid; 1999 u8 lmc; 2000 enum ib_port_state port_state; 2001 }; 2002 2003 struct ib_cache { 2004 rwlock_t lock; 2005 struct ib_event_handler event_handler; 2006 struct ib_port_cache *ports; 2007 }; 2008 2009 struct iw_cm_verbs; 2010 2011 struct ib_port_immutable { 2012 int pkey_tbl_len; 2013 int gid_tbl_len; 2014 u32 core_cap_flags; 2015 u32 max_mad_size; 2016 }; 2017 2018 /* rdma netdev type - specifies protocol type */ 2019 enum rdma_netdev_t { 2020 RDMA_NETDEV_OPA_VNIC, 2021 RDMA_NETDEV_IPOIB, 2022 }; 2023 2024 /** 2025 * struct rdma_netdev - rdma netdev 2026 * For cases where netstack interfacing is required. 2027 */ 2028 struct rdma_netdev { 2029 void *clnt_priv; 2030 struct ib_device *hca; 2031 u8 port_num; 2032 2033 /* cleanup function must be specified */ 2034 void (*free_rdma_netdev)(struct net_device *netdev); 2035 2036 /* control functions */ 2037 void (*set_id)(struct net_device *netdev, int id); 2038 /* send packet */ 2039 int (*send)(struct net_device *dev, struct sk_buff *skb, 2040 struct ib_ah *address, u32 dqpn); 2041 /* multicast */ 2042 int (*attach_mcast)(struct net_device *dev, struct ib_device *hca, 2043 union ib_gid *gid, u16 mlid, 2044 int set_qkey, u32 qkey); 2045 int (*detach_mcast)(struct net_device *dev, struct ib_device *hca, 2046 union ib_gid *gid, u16 mlid); 2047 }; 2048 2049 struct ib_port_pkey_list { 2050 /* Lock to hold while modifying the list. */ 2051 spinlock_t list_lock; 2052 struct list_head pkey_list; 2053 }; 2054 2055 struct ib_device { 2056 /* Do not access @dma_device directly from ULP nor from HW drivers. */ 2057 struct device *dma_device; 2058 2059 char name[IB_DEVICE_NAME_MAX]; 2060 2061 struct list_head event_handler_list; 2062 spinlock_t event_handler_lock; 2063 2064 spinlock_t client_data_lock; 2065 struct list_head core_list; 2066 /* Access to the client_data_list is protected by the client_data_lock 2067 * spinlock and the lists_rwsem read-write semaphore */ 2068 struct list_head client_data_list; 2069 2070 struct ib_cache cache; 2071 /** 2072 * port_immutable is indexed by port number 2073 */ 2074 struct ib_port_immutable *port_immutable; 2075 2076 int num_comp_vectors; 2077 2078 struct ib_port_pkey_list *port_pkey_list; 2079 2080 struct iw_cm_verbs *iwcm; 2081 2082 /** 2083 * alloc_hw_stats - Allocate a struct rdma_hw_stats and fill in the 2084 * driver initialized data. The struct is kfree()'ed by the sysfs 2085 * core when the device is removed. A lifespan of -1 in the return 2086 * struct tells the core to set a default lifespan. 2087 */ 2088 struct rdma_hw_stats *(*alloc_hw_stats)(struct ib_device *device, 2089 u8 port_num); 2090 /** 2091 * get_hw_stats - Fill in the counter value(s) in the stats struct. 2092 * @index - The index in the value array we wish to have updated, or 2093 * num_counters if we want all stats updated 2094 * Return codes - 2095 * < 0 - Error, no counters updated 2096 * index - Updated the single counter pointed to by index 2097 * num_counters - Updated all counters (will reset the timestamp 2098 * and prevent further calls for lifespan milliseconds) 2099 * Drivers are allowed to update all counters in leiu of just the 2100 * one given in index at their option 2101 */ 2102 int (*get_hw_stats)(struct ib_device *device, 2103 struct rdma_hw_stats *stats, 2104 u8 port, int index); 2105 int (*query_device)(struct ib_device *device, 2106 struct ib_device_attr *device_attr, 2107 struct ib_udata *udata); 2108 int (*query_port)(struct ib_device *device, 2109 u8 port_num, 2110 struct ib_port_attr *port_attr); 2111 enum rdma_link_layer (*get_link_layer)(struct ib_device *device, 2112 u8 port_num); 2113 /* When calling get_netdev, the HW vendor's driver should return the 2114 * net device of device @device at port @port_num or NULL if such 2115 * a net device doesn't exist. The vendor driver should call dev_hold 2116 * on this net device. The HW vendor's device driver must guarantee 2117 * that this function returns NULL before the net device reaches 2118 * NETDEV_UNREGISTER_FINAL state. 2119 */ 2120 struct net_device *(*get_netdev)(struct ib_device *device, 2121 u8 port_num); 2122 int (*query_gid)(struct ib_device *device, 2123 u8 port_num, int index, 2124 union ib_gid *gid); 2125 /* When calling add_gid, the HW vendor's driver should 2126 * add the gid of device @device at gid index @index of 2127 * port @port_num to be @gid. Meta-info of that gid (for example, 2128 * the network device related to this gid is available 2129 * at @attr. @context allows the HW vendor driver to store extra 2130 * information together with a GID entry. The HW vendor may allocate 2131 * memory to contain this information and store it in @context when a 2132 * new GID entry is written to. Params are consistent until the next 2133 * call of add_gid or delete_gid. The function should return 0 on 2134 * success or error otherwise. The function could be called 2135 * concurrently for different ports. This function is only called 2136 * when roce_gid_table is used. 2137 */ 2138 int (*add_gid)(struct ib_device *device, 2139 u8 port_num, 2140 unsigned int index, 2141 const union ib_gid *gid, 2142 const struct ib_gid_attr *attr, 2143 void **context); 2144 /* When calling del_gid, the HW vendor's driver should delete the 2145 * gid of device @device at gid index @index of port @port_num. 2146 * Upon the deletion of a GID entry, the HW vendor must free any 2147 * allocated memory. The caller will clear @context afterwards. 2148 * This function is only called when roce_gid_table is used. 2149 */ 2150 int (*del_gid)(struct ib_device *device, 2151 u8 port_num, 2152 unsigned int index, 2153 void **context); 2154 int (*query_pkey)(struct ib_device *device, 2155 u8 port_num, u16 index, u16 *pkey); 2156 int (*modify_device)(struct ib_device *device, 2157 int device_modify_mask, 2158 struct ib_device_modify *device_modify); 2159 int (*modify_port)(struct ib_device *device, 2160 u8 port_num, int port_modify_mask, 2161 struct ib_port_modify *port_modify); 2162 struct ib_ucontext * (*alloc_ucontext)(struct ib_device *device, 2163 struct ib_udata *udata); 2164 int (*dealloc_ucontext)(struct ib_ucontext *context); 2165 int (*mmap)(struct ib_ucontext *context, 2166 struct vm_area_struct *vma); 2167 struct ib_pd * (*alloc_pd)(struct ib_device *device, 2168 struct ib_ucontext *context, 2169 struct ib_udata *udata); 2170 int (*dealloc_pd)(struct ib_pd *pd); 2171 struct ib_ah * (*create_ah)(struct ib_pd *pd, 2172 struct rdma_ah_attr *ah_attr, 2173 struct ib_udata *udata); 2174 int (*modify_ah)(struct ib_ah *ah, 2175 struct rdma_ah_attr *ah_attr); 2176 int (*query_ah)(struct ib_ah *ah, 2177 struct rdma_ah_attr *ah_attr); 2178 int (*destroy_ah)(struct ib_ah *ah); 2179 struct ib_srq * (*create_srq)(struct ib_pd *pd, 2180 struct ib_srq_init_attr *srq_init_attr, 2181 struct ib_udata *udata); 2182 int (*modify_srq)(struct ib_srq *srq, 2183 struct ib_srq_attr *srq_attr, 2184 enum ib_srq_attr_mask srq_attr_mask, 2185 struct ib_udata *udata); 2186 int (*query_srq)(struct ib_srq *srq, 2187 struct ib_srq_attr *srq_attr); 2188 int (*destroy_srq)(struct ib_srq *srq); 2189 int (*post_srq_recv)(struct ib_srq *srq, 2190 struct ib_recv_wr *recv_wr, 2191 struct ib_recv_wr **bad_recv_wr); 2192 struct ib_qp * (*create_qp)(struct ib_pd *pd, 2193 struct ib_qp_init_attr *qp_init_attr, 2194 struct ib_udata *udata); 2195 int (*modify_qp)(struct ib_qp *qp, 2196 struct ib_qp_attr *qp_attr, 2197 int qp_attr_mask, 2198 struct ib_udata *udata); 2199 int (*query_qp)(struct ib_qp *qp, 2200 struct ib_qp_attr *qp_attr, 2201 int qp_attr_mask, 2202 struct ib_qp_init_attr *qp_init_attr); 2203 int (*destroy_qp)(struct ib_qp *qp); 2204 int (*post_send)(struct ib_qp *qp, 2205 struct ib_send_wr *send_wr, 2206 struct ib_send_wr **bad_send_wr); 2207 int (*post_recv)(struct ib_qp *qp, 2208 struct ib_recv_wr *recv_wr, 2209 struct ib_recv_wr **bad_recv_wr); 2210 struct ib_cq * (*create_cq)(struct ib_device *device, 2211 const struct ib_cq_init_attr *attr, 2212 struct ib_ucontext *context, 2213 struct ib_udata *udata); 2214 int (*modify_cq)(struct ib_cq *cq, u16 cq_count, 2215 u16 cq_period); 2216 int (*destroy_cq)(struct ib_cq *cq); 2217 int (*resize_cq)(struct ib_cq *cq, int cqe, 2218 struct ib_udata *udata); 2219 int (*poll_cq)(struct ib_cq *cq, int num_entries, 2220 struct ib_wc *wc); 2221 int (*peek_cq)(struct ib_cq *cq, int wc_cnt); 2222 int (*req_notify_cq)(struct ib_cq *cq, 2223 enum ib_cq_notify_flags flags); 2224 int (*req_ncomp_notif)(struct ib_cq *cq, 2225 int wc_cnt); 2226 struct ib_mr * (*get_dma_mr)(struct ib_pd *pd, 2227 int mr_access_flags); 2228 struct ib_mr * (*reg_user_mr)(struct ib_pd *pd, 2229 u64 start, u64 length, 2230 u64 virt_addr, 2231 int mr_access_flags, 2232 struct ib_udata *udata); 2233 int (*rereg_user_mr)(struct ib_mr *mr, 2234 int flags, 2235 u64 start, u64 length, 2236 u64 virt_addr, 2237 int mr_access_flags, 2238 struct ib_pd *pd, 2239 struct ib_udata *udata); 2240 int (*dereg_mr)(struct ib_mr *mr); 2241 struct ib_mr * (*alloc_mr)(struct ib_pd *pd, 2242 enum ib_mr_type mr_type, 2243 u32 max_num_sg); 2244 int (*map_mr_sg)(struct ib_mr *mr, 2245 struct scatterlist *sg, 2246 int sg_nents, 2247 unsigned int *sg_offset); 2248 struct ib_mw * (*alloc_mw)(struct ib_pd *pd, 2249 enum ib_mw_type type, 2250 struct ib_udata *udata); 2251 int (*dealloc_mw)(struct ib_mw *mw); 2252 struct ib_fmr * (*alloc_fmr)(struct ib_pd *pd, 2253 int mr_access_flags, 2254 struct ib_fmr_attr *fmr_attr); 2255 int (*map_phys_fmr)(struct ib_fmr *fmr, 2256 u64 *page_list, int list_len, 2257 u64 iova); 2258 int (*unmap_fmr)(struct list_head *fmr_list); 2259 int (*dealloc_fmr)(struct ib_fmr *fmr); 2260 int (*attach_mcast)(struct ib_qp *qp, 2261 union ib_gid *gid, 2262 u16 lid); 2263 int (*detach_mcast)(struct ib_qp *qp, 2264 union ib_gid *gid, 2265 u16 lid); 2266 int (*process_mad)(struct ib_device *device, 2267 int process_mad_flags, 2268 u8 port_num, 2269 const struct ib_wc *in_wc, 2270 const struct ib_grh *in_grh, 2271 const struct ib_mad_hdr *in_mad, 2272 size_t in_mad_size, 2273 struct ib_mad_hdr *out_mad, 2274 size_t *out_mad_size, 2275 u16 *out_mad_pkey_index); 2276 struct ib_xrcd * (*alloc_xrcd)(struct ib_device *device, 2277 struct ib_ucontext *ucontext, 2278 struct ib_udata *udata); 2279 int (*dealloc_xrcd)(struct ib_xrcd *xrcd); 2280 struct ib_flow * (*create_flow)(struct ib_qp *qp, 2281 struct ib_flow_attr 2282 *flow_attr, 2283 int domain); 2284 int (*destroy_flow)(struct ib_flow *flow_id); 2285 int (*check_mr_status)(struct ib_mr *mr, u32 check_mask, 2286 struct ib_mr_status *mr_status); 2287 void (*disassociate_ucontext)(struct ib_ucontext *ibcontext); 2288 void (*drain_rq)(struct ib_qp *qp); 2289 void (*drain_sq)(struct ib_qp *qp); 2290 int (*set_vf_link_state)(struct ib_device *device, int vf, u8 port, 2291 int state); 2292 int (*get_vf_config)(struct ib_device *device, int vf, u8 port, 2293 struct ifla_vf_info *ivf); 2294 int (*get_vf_stats)(struct ib_device *device, int vf, u8 port, 2295 struct ifla_vf_stats *stats); 2296 int (*set_vf_guid)(struct ib_device *device, int vf, u8 port, u64 guid, 2297 int type); 2298 struct ib_wq * (*create_wq)(struct ib_pd *pd, 2299 struct ib_wq_init_attr *init_attr, 2300 struct ib_udata *udata); 2301 int (*destroy_wq)(struct ib_wq *wq); 2302 int (*modify_wq)(struct ib_wq *wq, 2303 struct ib_wq_attr *attr, 2304 u32 wq_attr_mask, 2305 struct ib_udata *udata); 2306 struct ib_rwq_ind_table * (*create_rwq_ind_table)(struct ib_device *device, 2307 struct ib_rwq_ind_table_init_attr *init_attr, 2308 struct ib_udata *udata); 2309 int (*destroy_rwq_ind_table)(struct ib_rwq_ind_table *wq_ind_table); 2310 /** 2311 * rdma netdev operation 2312 * 2313 * Driver implementing alloc_rdma_netdev must return -EOPNOTSUPP if it 2314 * doesn't support the specified rdma netdev type. 2315 */ 2316 struct net_device *(*alloc_rdma_netdev)( 2317 struct ib_device *device, 2318 u8 port_num, 2319 enum rdma_netdev_t type, 2320 const char *name, 2321 unsigned char name_assign_type, 2322 void (*setup)(struct net_device *)); 2323 2324 struct module *owner; 2325 struct device dev; 2326 struct kobject *ports_parent; 2327 struct list_head port_list; 2328 2329 enum { 2330 IB_DEV_UNINITIALIZED, 2331 IB_DEV_REGISTERED, 2332 IB_DEV_UNREGISTERED 2333 } reg_state; 2334 2335 int uverbs_abi_ver; 2336 u64 uverbs_cmd_mask; 2337 u64 uverbs_ex_cmd_mask; 2338 2339 char node_desc[IB_DEVICE_NODE_DESC_MAX]; 2340 __be64 node_guid; 2341 u32 local_dma_lkey; 2342 u16 is_switch:1; 2343 u8 node_type; 2344 u8 phys_port_cnt; 2345 struct ib_device_attr attrs; 2346 struct attribute_group *hw_stats_ag; 2347 struct rdma_hw_stats *hw_stats; 2348 2349 #ifdef CONFIG_CGROUP_RDMA 2350 struct rdmacg_device cg_device; 2351 #endif 2352 2353 u32 index; 2354 2355 /** 2356 * The following mandatory functions are used only at device 2357 * registration. Keep functions such as these at the end of this 2358 * structure to avoid cache line misses when accessing struct ib_device 2359 * in fast paths. 2360 */ 2361 int (*get_port_immutable)(struct ib_device *, u8, struct ib_port_immutable *); 2362 void (*get_dev_fw_str)(struct ib_device *, char *str); 2363 const struct cpumask *(*get_vector_affinity)(struct ib_device *ibdev, 2364 int comp_vector); 2365 2366 struct uverbs_root_spec *specs_root; 2367 }; 2368 2369 struct ib_client { 2370 char *name; 2371 void (*add) (struct ib_device *); 2372 void (*remove)(struct ib_device *, void *client_data); 2373 2374 /* Returns the net_dev belonging to this ib_client and matching the 2375 * given parameters. 2376 * @dev: An RDMA device that the net_dev use for communication. 2377 * @port: A physical port number on the RDMA device. 2378 * @pkey: P_Key that the net_dev uses if applicable. 2379 * @gid: A GID that the net_dev uses to communicate. 2380 * @addr: An IP address the net_dev is configured with. 2381 * @client_data: The device's client data set by ib_set_client_data(). 2382 * 2383 * An ib_client that implements a net_dev on top of RDMA devices 2384 * (such as IP over IB) should implement this callback, allowing the 2385 * rdma_cm module to find the right net_dev for a given request. 2386 * 2387 * The caller is responsible for calling dev_put on the returned 2388 * netdev. */ 2389 struct net_device *(*get_net_dev_by_params)( 2390 struct ib_device *dev, 2391 u8 port, 2392 u16 pkey, 2393 const union ib_gid *gid, 2394 const struct sockaddr *addr, 2395 void *client_data); 2396 struct list_head list; 2397 }; 2398 2399 struct ib_device *ib_alloc_device(size_t size); 2400 void ib_dealloc_device(struct ib_device *device); 2401 2402 void ib_get_device_fw_str(struct ib_device *device, char *str); 2403 2404 int ib_register_device(struct ib_device *device, 2405 int (*port_callback)(struct ib_device *, 2406 u8, struct kobject *)); 2407 void ib_unregister_device(struct ib_device *device); 2408 2409 int ib_register_client (struct ib_client *client); 2410 void ib_unregister_client(struct ib_client *client); 2411 2412 void *ib_get_client_data(struct ib_device *device, struct ib_client *client); 2413 void ib_set_client_data(struct ib_device *device, struct ib_client *client, 2414 void *data); 2415 2416 static inline int ib_copy_from_udata(void *dest, struct ib_udata *udata, size_t len) 2417 { 2418 return copy_from_user(dest, udata->inbuf, len) ? -EFAULT : 0; 2419 } 2420 2421 static inline int ib_copy_to_udata(struct ib_udata *udata, void *src, size_t len) 2422 { 2423 return copy_to_user(udata->outbuf, src, len) ? -EFAULT : 0; 2424 } 2425 2426 static inline bool ib_is_udata_cleared(struct ib_udata *udata, 2427 size_t offset, 2428 size_t len) 2429 { 2430 const void __user *p = udata->inbuf + offset; 2431 bool ret; 2432 u8 *buf; 2433 2434 if (len > USHRT_MAX) 2435 return false; 2436 2437 buf = memdup_user(p, len); 2438 if (IS_ERR(buf)) 2439 return false; 2440 2441 ret = !memchr_inv(buf, 0, len); 2442 kfree(buf); 2443 return ret; 2444 } 2445 2446 /** 2447 * ib_modify_qp_is_ok - Check that the supplied attribute mask 2448 * contains all required attributes and no attributes not allowed for 2449 * the given QP state transition. 2450 * @cur_state: Current QP state 2451 * @next_state: Next QP state 2452 * @type: QP type 2453 * @mask: Mask of supplied QP attributes 2454 * @ll : link layer of port 2455 * 2456 * This function is a helper function that a low-level driver's 2457 * modify_qp method can use to validate the consumer's input. It 2458 * checks that cur_state and next_state are valid QP states, that a 2459 * transition from cur_state to next_state is allowed by the IB spec, 2460 * and that the attribute mask supplied is allowed for the transition. 2461 */ 2462 int ib_modify_qp_is_ok(enum ib_qp_state cur_state, enum ib_qp_state next_state, 2463 enum ib_qp_type type, enum ib_qp_attr_mask mask, 2464 enum rdma_link_layer ll); 2465 2466 void ib_register_event_handler(struct ib_event_handler *event_handler); 2467 void ib_unregister_event_handler(struct ib_event_handler *event_handler); 2468 void ib_dispatch_event(struct ib_event *event); 2469 2470 int ib_query_port(struct ib_device *device, 2471 u8 port_num, struct ib_port_attr *port_attr); 2472 2473 enum rdma_link_layer rdma_port_get_link_layer(struct ib_device *device, 2474 u8 port_num); 2475 2476 /** 2477 * rdma_cap_ib_switch - Check if the device is IB switch 2478 * @device: Device to check 2479 * 2480 * Device driver is responsible for setting is_switch bit on 2481 * in ib_device structure at init time. 2482 * 2483 * Return: true if the device is IB switch. 2484 */ 2485 static inline bool rdma_cap_ib_switch(const struct ib_device *device) 2486 { 2487 return device->is_switch; 2488 } 2489 2490 /** 2491 * rdma_start_port - Return the first valid port number for the device 2492 * specified 2493 * 2494 * @device: Device to be checked 2495 * 2496 * Return start port number 2497 */ 2498 static inline u8 rdma_start_port(const struct ib_device *device) 2499 { 2500 return rdma_cap_ib_switch(device) ? 0 : 1; 2501 } 2502 2503 /** 2504 * rdma_end_port - Return the last valid port number for the device 2505 * specified 2506 * 2507 * @device: Device to be checked 2508 * 2509 * Return last port number 2510 */ 2511 static inline u8 rdma_end_port(const struct ib_device *device) 2512 { 2513 return rdma_cap_ib_switch(device) ? 0 : device->phys_port_cnt; 2514 } 2515 2516 static inline int rdma_is_port_valid(const struct ib_device *device, 2517 unsigned int port) 2518 { 2519 return (port >= rdma_start_port(device) && 2520 port <= rdma_end_port(device)); 2521 } 2522 2523 static inline bool rdma_protocol_ib(const struct ib_device *device, u8 port_num) 2524 { 2525 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_IB; 2526 } 2527 2528 static inline bool rdma_protocol_roce(const struct ib_device *device, u8 port_num) 2529 { 2530 return device->port_immutable[port_num].core_cap_flags & 2531 (RDMA_CORE_CAP_PROT_ROCE | RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP); 2532 } 2533 2534 static inline bool rdma_protocol_roce_udp_encap(const struct ib_device *device, u8 port_num) 2535 { 2536 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP; 2537 } 2538 2539 static inline bool rdma_protocol_roce_eth_encap(const struct ib_device *device, u8 port_num) 2540 { 2541 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_ROCE; 2542 } 2543 2544 static inline bool rdma_protocol_iwarp(const struct ib_device *device, u8 port_num) 2545 { 2546 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_IWARP; 2547 } 2548 2549 static inline bool rdma_ib_or_roce(const struct ib_device *device, u8 port_num) 2550 { 2551 return rdma_protocol_ib(device, port_num) || 2552 rdma_protocol_roce(device, port_num); 2553 } 2554 2555 static inline bool rdma_protocol_raw_packet(const struct ib_device *device, u8 port_num) 2556 { 2557 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_RAW_PACKET; 2558 } 2559 2560 static inline bool rdma_protocol_usnic(const struct ib_device *device, u8 port_num) 2561 { 2562 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_PROT_USNIC; 2563 } 2564 2565 /** 2566 * rdma_cap_ib_mad - Check if the port of a device supports Infiniband 2567 * Management Datagrams. 2568 * @device: Device to check 2569 * @port_num: Port number to check 2570 * 2571 * Management Datagrams (MAD) are a required part of the InfiniBand 2572 * specification and are supported on all InfiniBand devices. A slightly 2573 * extended version are also supported on OPA interfaces. 2574 * 2575 * Return: true if the port supports sending/receiving of MAD packets. 2576 */ 2577 static inline bool rdma_cap_ib_mad(const struct ib_device *device, u8 port_num) 2578 { 2579 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_MAD; 2580 } 2581 2582 /** 2583 * rdma_cap_opa_mad - Check if the port of device provides support for OPA 2584 * Management Datagrams. 2585 * @device: Device to check 2586 * @port_num: Port number to check 2587 * 2588 * Intel OmniPath devices extend and/or replace the InfiniBand Management 2589 * datagrams with their own versions. These OPA MADs share many but not all of 2590 * the characteristics of InfiniBand MADs. 2591 * 2592 * OPA MADs differ in the following ways: 2593 * 2594 * 1) MADs are variable size up to 2K 2595 * IBTA defined MADs remain fixed at 256 bytes 2596 * 2) OPA SMPs must carry valid PKeys 2597 * 3) OPA SMP packets are a different format 2598 * 2599 * Return: true if the port supports OPA MAD packet formats. 2600 */ 2601 static inline bool rdma_cap_opa_mad(struct ib_device *device, u8 port_num) 2602 { 2603 return (device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_OPA_MAD) 2604 == RDMA_CORE_CAP_OPA_MAD; 2605 } 2606 2607 /** 2608 * rdma_cap_ib_smi - Check if the port of a device provides an Infiniband 2609 * Subnet Management Agent (SMA) on the Subnet Management Interface (SMI). 2610 * @device: Device to check 2611 * @port_num: Port number to check 2612 * 2613 * Each InfiniBand node is required to provide a Subnet Management Agent 2614 * that the subnet manager can access. Prior to the fabric being fully 2615 * configured by the subnet manager, the SMA is accessed via a well known 2616 * interface called the Subnet Management Interface (SMI). This interface 2617 * uses directed route packets to communicate with the SM to get around the 2618 * chicken and egg problem of the SM needing to know what's on the fabric 2619 * in order to configure the fabric, and needing to configure the fabric in 2620 * order to send packets to the devices on the fabric. These directed 2621 * route packets do not need the fabric fully configured in order to reach 2622 * their destination. The SMI is the only method allowed to send 2623 * directed route packets on an InfiniBand fabric. 2624 * 2625 * Return: true if the port provides an SMI. 2626 */ 2627 static inline bool rdma_cap_ib_smi(const struct ib_device *device, u8 port_num) 2628 { 2629 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_SMI; 2630 } 2631 2632 /** 2633 * rdma_cap_ib_cm - Check if the port of device has the capability Infiniband 2634 * Communication Manager. 2635 * @device: Device to check 2636 * @port_num: Port number to check 2637 * 2638 * The InfiniBand Communication Manager is one of many pre-defined General 2639 * Service Agents (GSA) that are accessed via the General Service 2640 * Interface (GSI). It's role is to facilitate establishment of connections 2641 * between nodes as well as other management related tasks for established 2642 * connections. 2643 * 2644 * Return: true if the port supports an IB CM (this does not guarantee that 2645 * a CM is actually running however). 2646 */ 2647 static inline bool rdma_cap_ib_cm(const struct ib_device *device, u8 port_num) 2648 { 2649 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_CM; 2650 } 2651 2652 /** 2653 * rdma_cap_iw_cm - Check if the port of device has the capability IWARP 2654 * Communication Manager. 2655 * @device: Device to check 2656 * @port_num: Port number to check 2657 * 2658 * Similar to above, but specific to iWARP connections which have a different 2659 * managment protocol than InfiniBand. 2660 * 2661 * Return: true if the port supports an iWARP CM (this does not guarantee that 2662 * a CM is actually running however). 2663 */ 2664 static inline bool rdma_cap_iw_cm(const struct ib_device *device, u8 port_num) 2665 { 2666 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IW_CM; 2667 } 2668 2669 /** 2670 * rdma_cap_ib_sa - Check if the port of device has the capability Infiniband 2671 * Subnet Administration. 2672 * @device: Device to check 2673 * @port_num: Port number to check 2674 * 2675 * An InfiniBand Subnet Administration (SA) service is a pre-defined General 2676 * Service Agent (GSA) provided by the Subnet Manager (SM). On InfiniBand 2677 * fabrics, devices should resolve routes to other hosts by contacting the 2678 * SA to query the proper route. 2679 * 2680 * Return: true if the port should act as a client to the fabric Subnet 2681 * Administration interface. This does not imply that the SA service is 2682 * running locally. 2683 */ 2684 static inline bool rdma_cap_ib_sa(const struct ib_device *device, u8 port_num) 2685 { 2686 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_IB_SA; 2687 } 2688 2689 /** 2690 * rdma_cap_ib_mcast - Check if the port of device has the capability Infiniband 2691 * Multicast. 2692 * @device: Device to check 2693 * @port_num: Port number to check 2694 * 2695 * InfiniBand multicast registration is more complex than normal IPv4 or 2696 * IPv6 multicast registration. Each Host Channel Adapter must register 2697 * with the Subnet Manager when it wishes to join a multicast group. It 2698 * should do so only once regardless of how many queue pairs it subscribes 2699 * to this group. And it should leave the group only after all queue pairs 2700 * attached to the group have been detached. 2701 * 2702 * Return: true if the port must undertake the additional adminstrative 2703 * overhead of registering/unregistering with the SM and tracking of the 2704 * total number of queue pairs attached to the multicast group. 2705 */ 2706 static inline bool rdma_cap_ib_mcast(const struct ib_device *device, u8 port_num) 2707 { 2708 return rdma_cap_ib_sa(device, port_num); 2709 } 2710 2711 /** 2712 * rdma_cap_af_ib - Check if the port of device has the capability 2713 * Native Infiniband Address. 2714 * @device: Device to check 2715 * @port_num: Port number to check 2716 * 2717 * InfiniBand addressing uses a port's GUID + Subnet Prefix to make a default 2718 * GID. RoCE uses a different mechanism, but still generates a GID via 2719 * a prescribed mechanism and port specific data. 2720 * 2721 * Return: true if the port uses a GID address to identify devices on the 2722 * network. 2723 */ 2724 static inline bool rdma_cap_af_ib(const struct ib_device *device, u8 port_num) 2725 { 2726 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_AF_IB; 2727 } 2728 2729 /** 2730 * rdma_cap_eth_ah - Check if the port of device has the capability 2731 * Ethernet Address Handle. 2732 * @device: Device to check 2733 * @port_num: Port number to check 2734 * 2735 * RoCE is InfiniBand over Ethernet, and it uses a well defined technique 2736 * to fabricate GIDs over Ethernet/IP specific addresses native to the 2737 * port. Normally, packet headers are generated by the sending host 2738 * adapter, but when sending connectionless datagrams, we must manually 2739 * inject the proper headers for the fabric we are communicating over. 2740 * 2741 * Return: true if we are running as a RoCE port and must force the 2742 * addition of a Global Route Header built from our Ethernet Address 2743 * Handle into our header list for connectionless packets. 2744 */ 2745 static inline bool rdma_cap_eth_ah(const struct ib_device *device, u8 port_num) 2746 { 2747 return device->port_immutable[port_num].core_cap_flags & RDMA_CORE_CAP_ETH_AH; 2748 } 2749 2750 /** 2751 * rdma_cap_opa_ah - Check if the port of device supports 2752 * OPA Address handles 2753 * @device: Device to check 2754 * @port_num: Port number to check 2755 * 2756 * Return: true if we are running on an OPA device which supports 2757 * the extended OPA addressing. 2758 */ 2759 static inline bool rdma_cap_opa_ah(struct ib_device *device, u8 port_num) 2760 { 2761 return (device->port_immutable[port_num].core_cap_flags & 2762 RDMA_CORE_CAP_OPA_AH) == RDMA_CORE_CAP_OPA_AH; 2763 } 2764 2765 /** 2766 * rdma_max_mad_size - Return the max MAD size required by this RDMA Port. 2767 * 2768 * @device: Device 2769 * @port_num: Port number 2770 * 2771 * This MAD size includes the MAD headers and MAD payload. No other headers 2772 * are included. 2773 * 2774 * Return the max MAD size required by the Port. Will return 0 if the port 2775 * does not support MADs 2776 */ 2777 static inline size_t rdma_max_mad_size(const struct ib_device *device, u8 port_num) 2778 { 2779 return device->port_immutable[port_num].max_mad_size; 2780 } 2781 2782 /** 2783 * rdma_cap_roce_gid_table - Check if the port of device uses roce_gid_table 2784 * @device: Device to check 2785 * @port_num: Port number to check 2786 * 2787 * RoCE GID table mechanism manages the various GIDs for a device. 2788 * 2789 * NOTE: if allocating the port's GID table has failed, this call will still 2790 * return true, but any RoCE GID table API will fail. 2791 * 2792 * Return: true if the port uses RoCE GID table mechanism in order to manage 2793 * its GIDs. 2794 */ 2795 static inline bool rdma_cap_roce_gid_table(const struct ib_device *device, 2796 u8 port_num) 2797 { 2798 return rdma_protocol_roce(device, port_num) && 2799 device->add_gid && device->del_gid; 2800 } 2801 2802 /* 2803 * Check if the device supports READ W/ INVALIDATE. 2804 */ 2805 static inline bool rdma_cap_read_inv(struct ib_device *dev, u32 port_num) 2806 { 2807 /* 2808 * iWarp drivers must support READ W/ INVALIDATE. No other protocol 2809 * has support for it yet. 2810 */ 2811 return rdma_protocol_iwarp(dev, port_num); 2812 } 2813 2814 int ib_query_gid(struct ib_device *device, 2815 u8 port_num, int index, union ib_gid *gid, 2816 struct ib_gid_attr *attr); 2817 2818 int ib_set_vf_link_state(struct ib_device *device, int vf, u8 port, 2819 int state); 2820 int ib_get_vf_config(struct ib_device *device, int vf, u8 port, 2821 struct ifla_vf_info *info); 2822 int ib_get_vf_stats(struct ib_device *device, int vf, u8 port, 2823 struct ifla_vf_stats *stats); 2824 int ib_set_vf_guid(struct ib_device *device, int vf, u8 port, u64 guid, 2825 int type); 2826 2827 int ib_query_pkey(struct ib_device *device, 2828 u8 port_num, u16 index, u16 *pkey); 2829 2830 int ib_modify_device(struct ib_device *device, 2831 int device_modify_mask, 2832 struct ib_device_modify *device_modify); 2833 2834 int ib_modify_port(struct ib_device *device, 2835 u8 port_num, int port_modify_mask, 2836 struct ib_port_modify *port_modify); 2837 2838 int ib_find_gid(struct ib_device *device, union ib_gid *gid, 2839 enum ib_gid_type gid_type, struct net_device *ndev, 2840 u8 *port_num, u16 *index); 2841 2842 int ib_find_pkey(struct ib_device *device, 2843 u8 port_num, u16 pkey, u16 *index); 2844 2845 enum ib_pd_flags { 2846 /* 2847 * Create a memory registration for all memory in the system and place 2848 * the rkey for it into pd->unsafe_global_rkey. This can be used by 2849 * ULPs to avoid the overhead of dynamic MRs. 2850 * 2851 * This flag is generally considered unsafe and must only be used in 2852 * extremly trusted environments. Every use of it will log a warning 2853 * in the kernel log. 2854 */ 2855 IB_PD_UNSAFE_GLOBAL_RKEY = 0x01, 2856 }; 2857 2858 struct ib_pd *__ib_alloc_pd(struct ib_device *device, unsigned int flags, 2859 const char *caller); 2860 #define ib_alloc_pd(device, flags) \ 2861 __ib_alloc_pd((device), (flags), __func__) 2862 void ib_dealloc_pd(struct ib_pd *pd); 2863 2864 /** 2865 * rdma_create_ah - Creates an address handle for the given address vector. 2866 * @pd: The protection domain associated with the address handle. 2867 * @ah_attr: The attributes of the address vector. 2868 * 2869 * The address handle is used to reference a local or global destination 2870 * in all UD QP post sends. 2871 */ 2872 struct ib_ah *rdma_create_ah(struct ib_pd *pd, struct rdma_ah_attr *ah_attr); 2873 2874 /** 2875 * rdma_create_user_ah - Creates an address handle for the given address vector. 2876 * It resolves destination mac address for ah attribute of RoCE type. 2877 * @pd: The protection domain associated with the address handle. 2878 * @ah_attr: The attributes of the address vector. 2879 * @udata: pointer to user's input output buffer information need by 2880 * provider driver. 2881 * 2882 * It returns 0 on success and returns appropriate error code on error. 2883 * The address handle is used to reference a local or global destination 2884 * in all UD QP post sends. 2885 */ 2886 struct ib_ah *rdma_create_user_ah(struct ib_pd *pd, 2887 struct rdma_ah_attr *ah_attr, 2888 struct ib_udata *udata); 2889 /** 2890 * ib_get_gids_from_rdma_hdr - Get sgid and dgid from GRH or IPv4 header 2891 * work completion. 2892 * @hdr: the L3 header to parse 2893 * @net_type: type of header to parse 2894 * @sgid: place to store source gid 2895 * @dgid: place to store destination gid 2896 */ 2897 int ib_get_gids_from_rdma_hdr(const union rdma_network_hdr *hdr, 2898 enum rdma_network_type net_type, 2899 union ib_gid *sgid, union ib_gid *dgid); 2900 2901 /** 2902 * ib_get_rdma_header_version - Get the header version 2903 * @hdr: the L3 header to parse 2904 */ 2905 int ib_get_rdma_header_version(const union rdma_network_hdr *hdr); 2906 2907 /** 2908 * ib_init_ah_from_wc - Initializes address handle attributes from a 2909 * work completion. 2910 * @device: Device on which the received message arrived. 2911 * @port_num: Port on which the received message arrived. 2912 * @wc: Work completion associated with the received message. 2913 * @grh: References the received global route header. This parameter is 2914 * ignored unless the work completion indicates that the GRH is valid. 2915 * @ah_attr: Returned attributes that can be used when creating an address 2916 * handle for replying to the message. 2917 */ 2918 int ib_init_ah_from_wc(struct ib_device *device, u8 port_num, 2919 const struct ib_wc *wc, const struct ib_grh *grh, 2920 struct rdma_ah_attr *ah_attr); 2921 2922 /** 2923 * ib_create_ah_from_wc - Creates an address handle associated with the 2924 * sender of the specified work completion. 2925 * @pd: The protection domain associated with the address handle. 2926 * @wc: Work completion information associated with a received message. 2927 * @grh: References the received global route header. This parameter is 2928 * ignored unless the work completion indicates that the GRH is valid. 2929 * @port_num: The outbound port number to associate with the address. 2930 * 2931 * The address handle is used to reference a local or global destination 2932 * in all UD QP post sends. 2933 */ 2934 struct ib_ah *ib_create_ah_from_wc(struct ib_pd *pd, const struct ib_wc *wc, 2935 const struct ib_grh *grh, u8 port_num); 2936 2937 /** 2938 * rdma_modify_ah - Modifies the address vector associated with an address 2939 * handle. 2940 * @ah: The address handle to modify. 2941 * @ah_attr: The new address vector attributes to associate with the 2942 * address handle. 2943 */ 2944 int rdma_modify_ah(struct ib_ah *ah, struct rdma_ah_attr *ah_attr); 2945 2946 /** 2947 * rdma_query_ah - Queries the address vector associated with an address 2948 * handle. 2949 * @ah: The address handle to query. 2950 * @ah_attr: The address vector attributes associated with the address 2951 * handle. 2952 */ 2953 int rdma_query_ah(struct ib_ah *ah, struct rdma_ah_attr *ah_attr); 2954 2955 /** 2956 * rdma_destroy_ah - Destroys an address handle. 2957 * @ah: The address handle to destroy. 2958 */ 2959 int rdma_destroy_ah(struct ib_ah *ah); 2960 2961 /** 2962 * ib_create_srq - Creates a SRQ associated with the specified protection 2963 * domain. 2964 * @pd: The protection domain associated with the SRQ. 2965 * @srq_init_attr: A list of initial attributes required to create the 2966 * SRQ. If SRQ creation succeeds, then the attributes are updated to 2967 * the actual capabilities of the created SRQ. 2968 * 2969 * srq_attr->max_wr and srq_attr->max_sge are read the determine the 2970 * requested size of the SRQ, and set to the actual values allocated 2971 * on return. If ib_create_srq() succeeds, then max_wr and max_sge 2972 * will always be at least as large as the requested values. 2973 */ 2974 struct ib_srq *ib_create_srq(struct ib_pd *pd, 2975 struct ib_srq_init_attr *srq_init_attr); 2976 2977 /** 2978 * ib_modify_srq - Modifies the attributes for the specified SRQ. 2979 * @srq: The SRQ to modify. 2980 * @srq_attr: On input, specifies the SRQ attributes to modify. On output, 2981 * the current values of selected SRQ attributes are returned. 2982 * @srq_attr_mask: A bit-mask used to specify which attributes of the SRQ 2983 * are being modified. 2984 * 2985 * The mask may contain IB_SRQ_MAX_WR to resize the SRQ and/or 2986 * IB_SRQ_LIMIT to set the SRQ's limit and request notification when 2987 * the number of receives queued drops below the limit. 2988 */ 2989 int ib_modify_srq(struct ib_srq *srq, 2990 struct ib_srq_attr *srq_attr, 2991 enum ib_srq_attr_mask srq_attr_mask); 2992 2993 /** 2994 * ib_query_srq - Returns the attribute list and current values for the 2995 * specified SRQ. 2996 * @srq: The SRQ to query. 2997 * @srq_attr: The attributes of the specified SRQ. 2998 */ 2999 int ib_query_srq(struct ib_srq *srq, 3000 struct ib_srq_attr *srq_attr); 3001 3002 /** 3003 * ib_destroy_srq - Destroys the specified SRQ. 3004 * @srq: The SRQ to destroy. 3005 */ 3006 int ib_destroy_srq(struct ib_srq *srq); 3007 3008 /** 3009 * ib_post_srq_recv - Posts a list of work requests to the specified SRQ. 3010 * @srq: The SRQ to post the work request on. 3011 * @recv_wr: A list of work requests to post on the receive queue. 3012 * @bad_recv_wr: On an immediate failure, this parameter will reference 3013 * the work request that failed to be posted on the QP. 3014 */ 3015 static inline int ib_post_srq_recv(struct ib_srq *srq, 3016 struct ib_recv_wr *recv_wr, 3017 struct ib_recv_wr **bad_recv_wr) 3018 { 3019 return srq->device->post_srq_recv(srq, recv_wr, bad_recv_wr); 3020 } 3021 3022 /** 3023 * ib_create_qp - Creates a QP associated with the specified protection 3024 * domain. 3025 * @pd: The protection domain associated with the QP. 3026 * @qp_init_attr: A list of initial attributes required to create the 3027 * QP. If QP creation succeeds, then the attributes are updated to 3028 * the actual capabilities of the created QP. 3029 */ 3030 struct ib_qp *ib_create_qp(struct ib_pd *pd, 3031 struct ib_qp_init_attr *qp_init_attr); 3032 3033 /** 3034 * ib_modify_qp_with_udata - Modifies the attributes for the specified QP. 3035 * @qp: The QP to modify. 3036 * @attr: On input, specifies the QP attributes to modify. On output, 3037 * the current values of selected QP attributes are returned. 3038 * @attr_mask: A bit-mask used to specify which attributes of the QP 3039 * are being modified. 3040 * @udata: pointer to user's input output buffer information 3041 * are being modified. 3042 * It returns 0 on success and returns appropriate error code on error. 3043 */ 3044 int ib_modify_qp_with_udata(struct ib_qp *qp, 3045 struct ib_qp_attr *attr, 3046 int attr_mask, 3047 struct ib_udata *udata); 3048 3049 /** 3050 * ib_modify_qp - Modifies the attributes for the specified QP and then 3051 * transitions the QP to the given state. 3052 * @qp: The QP to modify. 3053 * @qp_attr: On input, specifies the QP attributes to modify. On output, 3054 * the current values of selected QP attributes are returned. 3055 * @qp_attr_mask: A bit-mask used to specify which attributes of the QP 3056 * are being modified. 3057 */ 3058 int ib_modify_qp(struct ib_qp *qp, 3059 struct ib_qp_attr *qp_attr, 3060 int qp_attr_mask); 3061 3062 /** 3063 * ib_query_qp - Returns the attribute list and current values for the 3064 * specified QP. 3065 * @qp: The QP to query. 3066 * @qp_attr: The attributes of the specified QP. 3067 * @qp_attr_mask: A bit-mask used to select specific attributes to query. 3068 * @qp_init_attr: Additional attributes of the selected QP. 3069 * 3070 * The qp_attr_mask may be used to limit the query to gathering only the 3071 * selected attributes. 3072 */ 3073 int ib_query_qp(struct ib_qp *qp, 3074 struct ib_qp_attr *qp_attr, 3075 int qp_attr_mask, 3076 struct ib_qp_init_attr *qp_init_attr); 3077 3078 /** 3079 * ib_destroy_qp - Destroys the specified QP. 3080 * @qp: The QP to destroy. 3081 */ 3082 int ib_destroy_qp(struct ib_qp *qp); 3083 3084 /** 3085 * ib_open_qp - Obtain a reference to an existing sharable QP. 3086 * @xrcd - XRC domain 3087 * @qp_open_attr: Attributes identifying the QP to open. 3088 * 3089 * Returns a reference to a sharable QP. 3090 */ 3091 struct ib_qp *ib_open_qp(struct ib_xrcd *xrcd, 3092 struct ib_qp_open_attr *qp_open_attr); 3093 3094 /** 3095 * ib_close_qp - Release an external reference to a QP. 3096 * @qp: The QP handle to release 3097 * 3098 * The opened QP handle is released by the caller. The underlying 3099 * shared QP is not destroyed until all internal references are released. 3100 */ 3101 int ib_close_qp(struct ib_qp *qp); 3102 3103 /** 3104 * ib_post_send - Posts a list of work requests to the send queue of 3105 * the specified QP. 3106 * @qp: The QP to post the work request on. 3107 * @send_wr: A list of work requests to post on the send queue. 3108 * @bad_send_wr: On an immediate failure, this parameter will reference 3109 * the work request that failed to be posted on the QP. 3110 * 3111 * While IBA Vol. 1 section 11.4.1.1 specifies that if an immediate 3112 * error is returned, the QP state shall not be affected, 3113 * ib_post_send() will return an immediate error after queueing any 3114 * earlier work requests in the list. 3115 */ 3116 static inline int ib_post_send(struct ib_qp *qp, 3117 struct ib_send_wr *send_wr, 3118 struct ib_send_wr **bad_send_wr) 3119 { 3120 return qp->device->post_send(qp, send_wr, bad_send_wr); 3121 } 3122 3123 /** 3124 * ib_post_recv - Posts a list of work requests to the receive queue of 3125 * the specified QP. 3126 * @qp: The QP to post the work request on. 3127 * @recv_wr: A list of work requests to post on the receive queue. 3128 * @bad_recv_wr: On an immediate failure, this parameter will reference 3129 * the work request that failed to be posted on the QP. 3130 */ 3131 static inline int ib_post_recv(struct ib_qp *qp, 3132 struct ib_recv_wr *recv_wr, 3133 struct ib_recv_wr **bad_recv_wr) 3134 { 3135 return qp->device->post_recv(qp, recv_wr, bad_recv_wr); 3136 } 3137 3138 struct ib_cq *ib_alloc_cq(struct ib_device *dev, void *private, 3139 int nr_cqe, int comp_vector, enum ib_poll_context poll_ctx); 3140 void ib_free_cq(struct ib_cq *cq); 3141 int ib_process_cq_direct(struct ib_cq *cq, int budget); 3142 3143 /** 3144 * ib_create_cq - Creates a CQ on the specified device. 3145 * @device: The device on which to create the CQ. 3146 * @comp_handler: A user-specified callback that is invoked when a 3147 * completion event occurs on the CQ. 3148 * @event_handler: A user-specified callback that is invoked when an 3149 * asynchronous event not associated with a completion occurs on the CQ. 3150 * @cq_context: Context associated with the CQ returned to the user via 3151 * the associated completion and event handlers. 3152 * @cq_attr: The attributes the CQ should be created upon. 3153 * 3154 * Users can examine the cq structure to determine the actual CQ size. 3155 */ 3156 struct ib_cq *ib_create_cq(struct ib_device *device, 3157 ib_comp_handler comp_handler, 3158 void (*event_handler)(struct ib_event *, void *), 3159 void *cq_context, 3160 const struct ib_cq_init_attr *cq_attr); 3161 3162 /** 3163 * ib_resize_cq - Modifies the capacity of the CQ. 3164 * @cq: The CQ to resize. 3165 * @cqe: The minimum size of the CQ. 3166 * 3167 * Users can examine the cq structure to determine the actual CQ size. 3168 */ 3169 int ib_resize_cq(struct ib_cq *cq, int cqe); 3170 3171 /** 3172 * rdma_set_cq_moderation - Modifies moderation params of the CQ 3173 * @cq: The CQ to modify. 3174 * @cq_count: number of CQEs that will trigger an event 3175 * @cq_period: max period of time in usec before triggering an event 3176 * 3177 */ 3178 int rdma_set_cq_moderation(struct ib_cq *cq, u16 cq_count, u16 cq_period); 3179 3180 /** 3181 * ib_destroy_cq - Destroys the specified CQ. 3182 * @cq: The CQ to destroy. 3183 */ 3184 int ib_destroy_cq(struct ib_cq *cq); 3185 3186 /** 3187 * ib_poll_cq - poll a CQ for completion(s) 3188 * @cq:the CQ being polled 3189 * @num_entries:maximum number of completions to return 3190 * @wc:array of at least @num_entries &struct ib_wc where completions 3191 * will be returned 3192 * 3193 * Poll a CQ for (possibly multiple) completions. If the return value 3194 * is < 0, an error occurred. If the return value is >= 0, it is the 3195 * number of completions returned. If the return value is 3196 * non-negative and < num_entries, then the CQ was emptied. 3197 */ 3198 static inline int ib_poll_cq(struct ib_cq *cq, int num_entries, 3199 struct ib_wc *wc) 3200 { 3201 return cq->device->poll_cq(cq, num_entries, wc); 3202 } 3203 3204 /** 3205 * ib_peek_cq - Returns the number of unreaped completions currently 3206 * on the specified CQ. 3207 * @cq: The CQ to peek. 3208 * @wc_cnt: A minimum number of unreaped completions to check for. 3209 * 3210 * If the number of unreaped completions is greater than or equal to wc_cnt, 3211 * this function returns wc_cnt, otherwise, it returns the actual number of 3212 * unreaped completions. 3213 */ 3214 int ib_peek_cq(struct ib_cq *cq, int wc_cnt); 3215 3216 /** 3217 * ib_req_notify_cq - Request completion notification on a CQ. 3218 * @cq: The CQ to generate an event for. 3219 * @flags: 3220 * Must contain exactly one of %IB_CQ_SOLICITED or %IB_CQ_NEXT_COMP 3221 * to request an event on the next solicited event or next work 3222 * completion at any type, respectively. %IB_CQ_REPORT_MISSED_EVENTS 3223 * may also be |ed in to request a hint about missed events, as 3224 * described below. 3225 * 3226 * Return Value: 3227 * < 0 means an error occurred while requesting notification 3228 * == 0 means notification was requested successfully, and if 3229 * IB_CQ_REPORT_MISSED_EVENTS was passed in, then no events 3230 * were missed and it is safe to wait for another event. In 3231 * this case is it guaranteed that any work completions added 3232 * to the CQ since the last CQ poll will trigger a completion 3233 * notification event. 3234 * > 0 is only returned if IB_CQ_REPORT_MISSED_EVENTS was passed 3235 * in. It means that the consumer must poll the CQ again to 3236 * make sure it is empty to avoid missing an event because of a 3237 * race between requesting notification and an entry being 3238 * added to the CQ. This return value means it is possible 3239 * (but not guaranteed) that a work completion has been added 3240 * to the CQ since the last poll without triggering a 3241 * completion notification event. 3242 */ 3243 static inline int ib_req_notify_cq(struct ib_cq *cq, 3244 enum ib_cq_notify_flags flags) 3245 { 3246 return cq->device->req_notify_cq(cq, flags); 3247 } 3248 3249 /** 3250 * ib_req_ncomp_notif - Request completion notification when there are 3251 * at least the specified number of unreaped completions on the CQ. 3252 * @cq: The CQ to generate an event for. 3253 * @wc_cnt: The number of unreaped completions that should be on the 3254 * CQ before an event is generated. 3255 */ 3256 static inline int ib_req_ncomp_notif(struct ib_cq *cq, int wc_cnt) 3257 { 3258 return cq->device->req_ncomp_notif ? 3259 cq->device->req_ncomp_notif(cq, wc_cnt) : 3260 -ENOSYS; 3261 } 3262 3263 /** 3264 * ib_dma_mapping_error - check a DMA addr for error 3265 * @dev: The device for which the dma_addr was created 3266 * @dma_addr: The DMA address to check 3267 */ 3268 static inline int ib_dma_mapping_error(struct ib_device *dev, u64 dma_addr) 3269 { 3270 return dma_mapping_error(dev->dma_device, dma_addr); 3271 } 3272 3273 /** 3274 * ib_dma_map_single - Map a kernel virtual address to DMA address 3275 * @dev: The device for which the dma_addr is to be created 3276 * @cpu_addr: The kernel virtual address 3277 * @size: The size of the region in bytes 3278 * @direction: The direction of the DMA 3279 */ 3280 static inline u64 ib_dma_map_single(struct ib_device *dev, 3281 void *cpu_addr, size_t size, 3282 enum dma_data_direction direction) 3283 { 3284 return dma_map_single(dev->dma_device, cpu_addr, size, direction); 3285 } 3286 3287 /** 3288 * ib_dma_unmap_single - Destroy a mapping created by ib_dma_map_single() 3289 * @dev: The device for which the DMA address was created 3290 * @addr: The DMA address 3291 * @size: The size of the region in bytes 3292 * @direction: The direction of the DMA 3293 */ 3294 static inline void ib_dma_unmap_single(struct ib_device *dev, 3295 u64 addr, size_t size, 3296 enum dma_data_direction direction) 3297 { 3298 dma_unmap_single(dev->dma_device, addr, size, direction); 3299 } 3300 3301 /** 3302 * ib_dma_map_page - Map a physical page to DMA address 3303 * @dev: The device for which the dma_addr is to be created 3304 * @page: The page to be mapped 3305 * @offset: The offset within the page 3306 * @size: The size of the region in bytes 3307 * @direction: The direction of the DMA 3308 */ 3309 static inline u64 ib_dma_map_page(struct ib_device *dev, 3310 struct page *page, 3311 unsigned long offset, 3312 size_t size, 3313 enum dma_data_direction direction) 3314 { 3315 return dma_map_page(dev->dma_device, page, offset, size, direction); 3316 } 3317 3318 /** 3319 * ib_dma_unmap_page - Destroy a mapping created by ib_dma_map_page() 3320 * @dev: The device for which the DMA address was created 3321 * @addr: The DMA address 3322 * @size: The size of the region in bytes 3323 * @direction: The direction of the DMA 3324 */ 3325 static inline void ib_dma_unmap_page(struct ib_device *dev, 3326 u64 addr, size_t size, 3327 enum dma_data_direction direction) 3328 { 3329 dma_unmap_page(dev->dma_device, addr, size, direction); 3330 } 3331 3332 /** 3333 * ib_dma_map_sg - Map a scatter/gather list to DMA addresses 3334 * @dev: The device for which the DMA addresses are to be created 3335 * @sg: The array of scatter/gather entries 3336 * @nents: The number of scatter/gather entries 3337 * @direction: The direction of the DMA 3338 */ 3339 static inline int ib_dma_map_sg(struct ib_device *dev, 3340 struct scatterlist *sg, int nents, 3341 enum dma_data_direction direction) 3342 { 3343 return dma_map_sg(dev->dma_device, sg, nents, direction); 3344 } 3345 3346 /** 3347 * ib_dma_unmap_sg - Unmap a scatter/gather list of DMA addresses 3348 * @dev: The device for which the DMA addresses were created 3349 * @sg: The array of scatter/gather entries 3350 * @nents: The number of scatter/gather entries 3351 * @direction: The direction of the DMA 3352 */ 3353 static inline void ib_dma_unmap_sg(struct ib_device *dev, 3354 struct scatterlist *sg, int nents, 3355 enum dma_data_direction direction) 3356 { 3357 dma_unmap_sg(dev->dma_device, sg, nents, direction); 3358 } 3359 3360 static inline int ib_dma_map_sg_attrs(struct ib_device *dev, 3361 struct scatterlist *sg, int nents, 3362 enum dma_data_direction direction, 3363 unsigned long dma_attrs) 3364 { 3365 return dma_map_sg_attrs(dev->dma_device, sg, nents, direction, 3366 dma_attrs); 3367 } 3368 3369 static inline void ib_dma_unmap_sg_attrs(struct ib_device *dev, 3370 struct scatterlist *sg, int nents, 3371 enum dma_data_direction direction, 3372 unsigned long dma_attrs) 3373 { 3374 dma_unmap_sg_attrs(dev->dma_device, sg, nents, direction, dma_attrs); 3375 } 3376 /** 3377 * ib_sg_dma_address - Return the DMA address from a scatter/gather entry 3378 * @dev: The device for which the DMA addresses were created 3379 * @sg: The scatter/gather entry 3380 * 3381 * Note: this function is obsolete. To do: change all occurrences of 3382 * ib_sg_dma_address() into sg_dma_address(). 3383 */ 3384 static inline u64 ib_sg_dma_address(struct ib_device *dev, 3385 struct scatterlist *sg) 3386 { 3387 return sg_dma_address(sg); 3388 } 3389 3390 /** 3391 * ib_sg_dma_len - Return the DMA length from a scatter/gather entry 3392 * @dev: The device for which the DMA addresses were created 3393 * @sg: The scatter/gather entry 3394 * 3395 * Note: this function is obsolete. To do: change all occurrences of 3396 * ib_sg_dma_len() into sg_dma_len(). 3397 */ 3398 static inline unsigned int ib_sg_dma_len(struct ib_device *dev, 3399 struct scatterlist *sg) 3400 { 3401 return sg_dma_len(sg); 3402 } 3403 3404 /** 3405 * ib_dma_sync_single_for_cpu - Prepare DMA region to be accessed by CPU 3406 * @dev: The device for which the DMA address was created 3407 * @addr: The DMA address 3408 * @size: The size of the region in bytes 3409 * @dir: The direction of the DMA 3410 */ 3411 static inline void ib_dma_sync_single_for_cpu(struct ib_device *dev, 3412 u64 addr, 3413 size_t size, 3414 enum dma_data_direction dir) 3415 { 3416 dma_sync_single_for_cpu(dev->dma_device, addr, size, dir); 3417 } 3418 3419 /** 3420 * ib_dma_sync_single_for_device - Prepare DMA region to be accessed by device 3421 * @dev: The device for which the DMA address was created 3422 * @addr: The DMA address 3423 * @size: The size of the region in bytes 3424 * @dir: The direction of the DMA 3425 */ 3426 static inline void ib_dma_sync_single_for_device(struct ib_device *dev, 3427 u64 addr, 3428 size_t size, 3429 enum dma_data_direction dir) 3430 { 3431 dma_sync_single_for_device(dev->dma_device, addr, size, dir); 3432 } 3433 3434 /** 3435 * ib_dma_alloc_coherent - Allocate memory and map it for DMA 3436 * @dev: The device for which the DMA address is requested 3437 * @size: The size of the region to allocate in bytes 3438 * @dma_handle: A pointer for returning the DMA address of the region 3439 * @flag: memory allocator flags 3440 */ 3441 static inline void *ib_dma_alloc_coherent(struct ib_device *dev, 3442 size_t size, 3443 dma_addr_t *dma_handle, 3444 gfp_t flag) 3445 { 3446 return dma_alloc_coherent(dev->dma_device, size, dma_handle, flag); 3447 } 3448 3449 /** 3450 * ib_dma_free_coherent - Free memory allocated by ib_dma_alloc_coherent() 3451 * @dev: The device for which the DMA addresses were allocated 3452 * @size: The size of the region 3453 * @cpu_addr: the address returned by ib_dma_alloc_coherent() 3454 * @dma_handle: the DMA address returned by ib_dma_alloc_coherent() 3455 */ 3456 static inline void ib_dma_free_coherent(struct ib_device *dev, 3457 size_t size, void *cpu_addr, 3458 dma_addr_t dma_handle) 3459 { 3460 dma_free_coherent(dev->dma_device, size, cpu_addr, dma_handle); 3461 } 3462 3463 /** 3464 * ib_dereg_mr - Deregisters a memory region and removes it from the 3465 * HCA translation table. 3466 * @mr: The memory region to deregister. 3467 * 3468 * This function can fail, if the memory region has memory windows bound to it. 3469 */ 3470 int ib_dereg_mr(struct ib_mr *mr); 3471 3472 struct ib_mr *ib_alloc_mr(struct ib_pd *pd, 3473 enum ib_mr_type mr_type, 3474 u32 max_num_sg); 3475 3476 /** 3477 * ib_update_fast_reg_key - updates the key portion of the fast_reg MR 3478 * R_Key and L_Key. 3479 * @mr - struct ib_mr pointer to be updated. 3480 * @newkey - new key to be used. 3481 */ 3482 static inline void ib_update_fast_reg_key(struct ib_mr *mr, u8 newkey) 3483 { 3484 mr->lkey = (mr->lkey & 0xffffff00) | newkey; 3485 mr->rkey = (mr->rkey & 0xffffff00) | newkey; 3486 } 3487 3488 /** 3489 * ib_inc_rkey - increments the key portion of the given rkey. Can be used 3490 * for calculating a new rkey for type 2 memory windows. 3491 * @rkey - the rkey to increment. 3492 */ 3493 static inline u32 ib_inc_rkey(u32 rkey) 3494 { 3495 const u32 mask = 0x000000ff; 3496 return ((rkey + 1) & mask) | (rkey & ~mask); 3497 } 3498 3499 /** 3500 * ib_alloc_fmr - Allocates a unmapped fast memory region. 3501 * @pd: The protection domain associated with the unmapped region. 3502 * @mr_access_flags: Specifies the memory access rights. 3503 * @fmr_attr: Attributes of the unmapped region. 3504 * 3505 * A fast memory region must be mapped before it can be used as part of 3506 * a work request. 3507 */ 3508 struct ib_fmr *ib_alloc_fmr(struct ib_pd *pd, 3509 int mr_access_flags, 3510 struct ib_fmr_attr *fmr_attr); 3511 3512 /** 3513 * ib_map_phys_fmr - Maps a list of physical pages to a fast memory region. 3514 * @fmr: The fast memory region to associate with the pages. 3515 * @page_list: An array of physical pages to map to the fast memory region. 3516 * @list_len: The number of pages in page_list. 3517 * @iova: The I/O virtual address to use with the mapped region. 3518 */ 3519 static inline int ib_map_phys_fmr(struct ib_fmr *fmr, 3520 u64 *page_list, int list_len, 3521 u64 iova) 3522 { 3523 return fmr->device->map_phys_fmr(fmr, page_list, list_len, iova); 3524 } 3525 3526 /** 3527 * ib_unmap_fmr - Removes the mapping from a list of fast memory regions. 3528 * @fmr_list: A linked list of fast memory regions to unmap. 3529 */ 3530 int ib_unmap_fmr(struct list_head *fmr_list); 3531 3532 /** 3533 * ib_dealloc_fmr - Deallocates a fast memory region. 3534 * @fmr: The fast memory region to deallocate. 3535 */ 3536 int ib_dealloc_fmr(struct ib_fmr *fmr); 3537 3538 /** 3539 * ib_attach_mcast - Attaches the specified QP to a multicast group. 3540 * @qp: QP to attach to the multicast group. The QP must be type 3541 * IB_QPT_UD. 3542 * @gid: Multicast group GID. 3543 * @lid: Multicast group LID in host byte order. 3544 * 3545 * In order to send and receive multicast packets, subnet 3546 * administration must have created the multicast group and configured 3547 * the fabric appropriately. The port associated with the specified 3548 * QP must also be a member of the multicast group. 3549 */ 3550 int ib_attach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid); 3551 3552 /** 3553 * ib_detach_mcast - Detaches the specified QP from a multicast group. 3554 * @qp: QP to detach from the multicast group. 3555 * @gid: Multicast group GID. 3556 * @lid: Multicast group LID in host byte order. 3557 */ 3558 int ib_detach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid); 3559 3560 /** 3561 * ib_alloc_xrcd - Allocates an XRC domain. 3562 * @device: The device on which to allocate the XRC domain. 3563 */ 3564 struct ib_xrcd *ib_alloc_xrcd(struct ib_device *device); 3565 3566 /** 3567 * ib_dealloc_xrcd - Deallocates an XRC domain. 3568 * @xrcd: The XRC domain to deallocate. 3569 */ 3570 int ib_dealloc_xrcd(struct ib_xrcd *xrcd); 3571 3572 struct ib_flow *ib_create_flow(struct ib_qp *qp, 3573 struct ib_flow_attr *flow_attr, int domain); 3574 int ib_destroy_flow(struct ib_flow *flow_id); 3575 3576 static inline int ib_check_mr_access(int flags) 3577 { 3578 /* 3579 * Local write permission is required if remote write or 3580 * remote atomic permission is also requested. 3581 */ 3582 if (flags & (IB_ACCESS_REMOTE_ATOMIC | IB_ACCESS_REMOTE_WRITE) && 3583 !(flags & IB_ACCESS_LOCAL_WRITE)) 3584 return -EINVAL; 3585 3586 return 0; 3587 } 3588 3589 /** 3590 * ib_check_mr_status: lightweight check of MR status. 3591 * This routine may provide status checks on a selected 3592 * ib_mr. first use is for signature status check. 3593 * 3594 * @mr: A memory region. 3595 * @check_mask: Bitmask of which checks to perform from 3596 * ib_mr_status_check enumeration. 3597 * @mr_status: The container of relevant status checks. 3598 * failed checks will be indicated in the status bitmask 3599 * and the relevant info shall be in the error item. 3600 */ 3601 int ib_check_mr_status(struct ib_mr *mr, u32 check_mask, 3602 struct ib_mr_status *mr_status); 3603 3604 struct net_device *ib_get_net_dev_by_params(struct ib_device *dev, u8 port, 3605 u16 pkey, const union ib_gid *gid, 3606 const struct sockaddr *addr); 3607 struct ib_wq *ib_create_wq(struct ib_pd *pd, 3608 struct ib_wq_init_attr *init_attr); 3609 int ib_destroy_wq(struct ib_wq *wq); 3610 int ib_modify_wq(struct ib_wq *wq, struct ib_wq_attr *attr, 3611 u32 wq_attr_mask); 3612 struct ib_rwq_ind_table *ib_create_rwq_ind_table(struct ib_device *device, 3613 struct ib_rwq_ind_table_init_attr* 3614 wq_ind_table_init_attr); 3615 int ib_destroy_rwq_ind_table(struct ib_rwq_ind_table *wq_ind_table); 3616 3617 int ib_map_mr_sg(struct ib_mr *mr, struct scatterlist *sg, int sg_nents, 3618 unsigned int *sg_offset, unsigned int page_size); 3619 3620 static inline int 3621 ib_map_mr_sg_zbva(struct ib_mr *mr, struct scatterlist *sg, int sg_nents, 3622 unsigned int *sg_offset, unsigned int page_size) 3623 { 3624 int n; 3625 3626 n = ib_map_mr_sg(mr, sg, sg_nents, sg_offset, page_size); 3627 mr->iova = 0; 3628 3629 return n; 3630 } 3631 3632 int ib_sg_to_pages(struct ib_mr *mr, struct scatterlist *sgl, int sg_nents, 3633 unsigned int *sg_offset, int (*set_page)(struct ib_mr *, u64)); 3634 3635 void ib_drain_rq(struct ib_qp *qp); 3636 void ib_drain_sq(struct ib_qp *qp); 3637 void ib_drain_qp(struct ib_qp *qp); 3638 3639 int ib_get_eth_speed(struct ib_device *dev, u8 port_num, u8 *speed, u8 *width); 3640 3641 static inline u8 *rdma_ah_retrieve_dmac(struct rdma_ah_attr *attr) 3642 { 3643 if (attr->type == RDMA_AH_ATTR_TYPE_ROCE) 3644 return attr->roce.dmac; 3645 return NULL; 3646 } 3647 3648 static inline void rdma_ah_set_dlid(struct rdma_ah_attr *attr, u32 dlid) 3649 { 3650 if (attr->type == RDMA_AH_ATTR_TYPE_IB) 3651 attr->ib.dlid = (u16)dlid; 3652 else if (attr->type == RDMA_AH_ATTR_TYPE_OPA) 3653 attr->opa.dlid = dlid; 3654 } 3655 3656 static inline u32 rdma_ah_get_dlid(const struct rdma_ah_attr *attr) 3657 { 3658 if (attr->type == RDMA_AH_ATTR_TYPE_IB) 3659 return attr->ib.dlid; 3660 else if (attr->type == RDMA_AH_ATTR_TYPE_OPA) 3661 return attr->opa.dlid; 3662 return 0; 3663 } 3664 3665 static inline void rdma_ah_set_sl(struct rdma_ah_attr *attr, u8 sl) 3666 { 3667 attr->sl = sl; 3668 } 3669 3670 static inline u8 rdma_ah_get_sl(const struct rdma_ah_attr *attr) 3671 { 3672 return attr->sl; 3673 } 3674 3675 static inline void rdma_ah_set_path_bits(struct rdma_ah_attr *attr, 3676 u8 src_path_bits) 3677 { 3678 if (attr->type == RDMA_AH_ATTR_TYPE_IB) 3679 attr->ib.src_path_bits = src_path_bits; 3680 else if (attr->type == RDMA_AH_ATTR_TYPE_OPA) 3681 attr->opa.src_path_bits = src_path_bits; 3682 } 3683 3684 static inline u8 rdma_ah_get_path_bits(const struct rdma_ah_attr *attr) 3685 { 3686 if (attr->type == RDMA_AH_ATTR_TYPE_IB) 3687 return attr->ib.src_path_bits; 3688 else if (attr->type == RDMA_AH_ATTR_TYPE_OPA) 3689 return attr->opa.src_path_bits; 3690 return 0; 3691 } 3692 3693 static inline void rdma_ah_set_make_grd(struct rdma_ah_attr *attr, 3694 bool make_grd) 3695 { 3696 if (attr->type == RDMA_AH_ATTR_TYPE_OPA) 3697 attr->opa.make_grd = make_grd; 3698 } 3699 3700 static inline bool rdma_ah_get_make_grd(const struct rdma_ah_attr *attr) 3701 { 3702 if (attr->type == RDMA_AH_ATTR_TYPE_OPA) 3703 return attr->opa.make_grd; 3704 return false; 3705 } 3706 3707 static inline void rdma_ah_set_port_num(struct rdma_ah_attr *attr, u8 port_num) 3708 { 3709 attr->port_num = port_num; 3710 } 3711 3712 static inline u8 rdma_ah_get_port_num(const struct rdma_ah_attr *attr) 3713 { 3714 return attr->port_num; 3715 } 3716 3717 static inline void rdma_ah_set_static_rate(struct rdma_ah_attr *attr, 3718 u8 static_rate) 3719 { 3720 attr->static_rate = static_rate; 3721 } 3722 3723 static inline u8 rdma_ah_get_static_rate(const struct rdma_ah_attr *attr) 3724 { 3725 return attr->static_rate; 3726 } 3727 3728 static inline void rdma_ah_set_ah_flags(struct rdma_ah_attr *attr, 3729 enum ib_ah_flags flag) 3730 { 3731 attr->ah_flags = flag; 3732 } 3733 3734 static inline enum ib_ah_flags 3735 rdma_ah_get_ah_flags(const struct rdma_ah_attr *attr) 3736 { 3737 return attr->ah_flags; 3738 } 3739 3740 static inline const struct ib_global_route 3741 *rdma_ah_read_grh(const struct rdma_ah_attr *attr) 3742 { 3743 return &attr->grh; 3744 } 3745 3746 /*To retrieve and modify the grh */ 3747 static inline struct ib_global_route 3748 *rdma_ah_retrieve_grh(struct rdma_ah_attr *attr) 3749 { 3750 return &attr->grh; 3751 } 3752 3753 static inline void rdma_ah_set_dgid_raw(struct rdma_ah_attr *attr, void *dgid) 3754 { 3755 struct ib_global_route *grh = rdma_ah_retrieve_grh(attr); 3756 3757 memcpy(grh->dgid.raw, dgid, sizeof(grh->dgid)); 3758 } 3759 3760 static inline void rdma_ah_set_subnet_prefix(struct rdma_ah_attr *attr, 3761 __be64 prefix) 3762 { 3763 struct ib_global_route *grh = rdma_ah_retrieve_grh(attr); 3764 3765 grh->dgid.global.subnet_prefix = prefix; 3766 } 3767 3768 static inline void rdma_ah_set_interface_id(struct rdma_ah_attr *attr, 3769 __be64 if_id) 3770 { 3771 struct ib_global_route *grh = rdma_ah_retrieve_grh(attr); 3772 3773 grh->dgid.global.interface_id = if_id; 3774 } 3775 3776 static inline void rdma_ah_set_grh(struct rdma_ah_attr *attr, 3777 union ib_gid *dgid, u32 flow_label, 3778 u8 sgid_index, u8 hop_limit, 3779 u8 traffic_class) 3780 { 3781 struct ib_global_route *grh = rdma_ah_retrieve_grh(attr); 3782 3783 attr->ah_flags = IB_AH_GRH; 3784 if (dgid) 3785 grh->dgid = *dgid; 3786 grh->flow_label = flow_label; 3787 grh->sgid_index = sgid_index; 3788 grh->hop_limit = hop_limit; 3789 grh->traffic_class = traffic_class; 3790 } 3791 3792 /*Get AH type */ 3793 static inline enum rdma_ah_attr_type rdma_ah_find_type(struct ib_device *dev, 3794 u32 port_num) 3795 { 3796 if ((rdma_protocol_roce(dev, port_num)) || 3797 (rdma_protocol_iwarp(dev, port_num))) 3798 return RDMA_AH_ATTR_TYPE_ROCE; 3799 else if ((rdma_protocol_ib(dev, port_num)) && 3800 (rdma_cap_opa_ah(dev, port_num))) 3801 return RDMA_AH_ATTR_TYPE_OPA; 3802 else 3803 return RDMA_AH_ATTR_TYPE_IB; 3804 } 3805 3806 /** 3807 * ib_lid_cpu16 - Return lid in 16bit CPU encoding. 3808 * In the current implementation the only way to get 3809 * get the 32bit lid is from other sources for OPA. 3810 * For IB, lids will always be 16bits so cast the 3811 * value accordingly. 3812 * 3813 * @lid: A 32bit LID 3814 */ 3815 static inline u16 ib_lid_cpu16(u32 lid) 3816 { 3817 WARN_ON_ONCE(lid & 0xFFFF0000); 3818 return (u16)lid; 3819 } 3820 3821 /** 3822 * ib_lid_be16 - Return lid in 16bit BE encoding. 3823 * 3824 * @lid: A 32bit LID 3825 */ 3826 static inline __be16 ib_lid_be16(u32 lid) 3827 { 3828 WARN_ON_ONCE(lid & 0xFFFF0000); 3829 return cpu_to_be16((u16)lid); 3830 } 3831 3832 /** 3833 * ib_get_vector_affinity - Get the affinity mappings of a given completion 3834 * vector 3835 * @device: the rdma device 3836 * @comp_vector: index of completion vector 3837 * 3838 * Returns NULL on failure, otherwise a corresponding cpu map of the 3839 * completion vector (returns all-cpus map if the device driver doesn't 3840 * implement get_vector_affinity). 3841 */ 3842 static inline const struct cpumask * 3843 ib_get_vector_affinity(struct ib_device *device, int comp_vector) 3844 { 3845 if (comp_vector < 0 || comp_vector >= device->num_comp_vectors || 3846 !device->get_vector_affinity) 3847 return NULL; 3848 3849 return device->get_vector_affinity(device, comp_vector); 3850 3851 } 3852 3853 #endif /* IB_VERBS_H */ 3854