1 /* SPDX-License-Identifier: GPL-2.0 OR Linux-OpenIB */ 2 /* 3 * Copyright (c) 2004 Mellanox Technologies Ltd. All rights reserved. 4 * Copyright (c) 2004 Infinicon Corporation. All rights reserved. 5 * Copyright (c) 2004, 2020 Intel Corporation. All rights reserved. 6 * Copyright (c) 2004 Topspin Corporation. All rights reserved. 7 * Copyright (c) 2004 Voltaire Corporation. All rights reserved. 8 * Copyright (c) 2005 Sun Microsystems, Inc. All rights reserved. 9 * Copyright (c) 2005, 2006, 2007 Cisco Systems. All rights reserved. 10 */ 11 12 #ifndef IB_VERBS_H 13 #define IB_VERBS_H 14 15 #include <linux/ethtool.h> 16 #include <linux/types.h> 17 #include <linux/device.h> 18 #include <linux/dma-mapping.h> 19 #include <linux/kref.h> 20 #include <linux/list.h> 21 #include <linux/rwsem.h> 22 #include <linux/workqueue.h> 23 #include <linux/irq_poll.h> 24 #include <uapi/linux/if_ether.h> 25 #include <net/ipv6.h> 26 #include <net/ip.h> 27 #include <linux/string.h> 28 #include <linux/slab.h> 29 #include <linux/netdevice.h> 30 #include <linux/refcount.h> 31 #include <linux/if_link.h> 32 #include <linux/atomic.h> 33 #include <linux/mmu_notifier.h> 34 #include <linux/uaccess.h> 35 #include <linux/cgroup_rdma.h> 36 #include <linux/irqflags.h> 37 #include <linux/preempt.h> 38 #include <linux/dim.h> 39 #include <uapi/rdma/ib_user_verbs.h> 40 #include <rdma/rdma_counter.h> 41 #include <rdma/restrack.h> 42 #include <rdma/signature.h> 43 #include <uapi/rdma/rdma_user_ioctl.h> 44 #include <uapi/rdma/ib_user_ioctl_verbs.h> 45 46 #define IB_FW_VERSION_NAME_MAX ETHTOOL_FWVERS_LEN 47 48 struct ib_umem_odp; 49 struct ib_uqp_object; 50 struct ib_usrq_object; 51 struct ib_uwq_object; 52 struct rdma_cm_id; 53 struct ib_port; 54 struct hw_stats_device_data; 55 56 extern struct workqueue_struct *ib_wq; 57 extern struct workqueue_struct *ib_comp_wq; 58 extern struct workqueue_struct *ib_comp_unbound_wq; 59 60 struct ib_ucq_object; 61 62 __printf(2, 3) __cold 63 void ibdev_emerg(const struct ib_device *ibdev, const char *format, ...); 64 __printf(2, 3) __cold 65 void ibdev_alert(const struct ib_device *ibdev, const char *format, ...); 66 __printf(2, 3) __cold 67 void ibdev_crit(const struct ib_device *ibdev, const char *format, ...); 68 __printf(2, 3) __cold 69 void ibdev_err(const struct ib_device *ibdev, const char *format, ...); 70 __printf(2, 3) __cold 71 void ibdev_warn(const struct ib_device *ibdev, const char *format, ...); 72 __printf(2, 3) __cold 73 void ibdev_notice(const struct ib_device *ibdev, const char *format, ...); 74 __printf(2, 3) __cold 75 void ibdev_info(const struct ib_device *ibdev, const char *format, ...); 76 77 #if defined(CONFIG_DYNAMIC_DEBUG) || \ 78 (defined(CONFIG_DYNAMIC_DEBUG_CORE) && defined(DYNAMIC_DEBUG_MODULE)) 79 #define ibdev_dbg(__dev, format, args...) \ 80 dynamic_ibdev_dbg(__dev, format, ##args) 81 #else 82 __printf(2, 3) __cold 83 static inline 84 void ibdev_dbg(const struct ib_device *ibdev, const char *format, ...) {} 85 #endif 86 87 #define ibdev_level_ratelimited(ibdev_level, ibdev, fmt, ...) \ 88 do { \ 89 static DEFINE_RATELIMIT_STATE(_rs, \ 90 DEFAULT_RATELIMIT_INTERVAL, \ 91 DEFAULT_RATELIMIT_BURST); \ 92 if (__ratelimit(&_rs)) \ 93 ibdev_level(ibdev, fmt, ##__VA_ARGS__); \ 94 } while (0) 95 96 #define ibdev_emerg_ratelimited(ibdev, fmt, ...) \ 97 ibdev_level_ratelimited(ibdev_emerg, ibdev, fmt, ##__VA_ARGS__) 98 #define ibdev_alert_ratelimited(ibdev, fmt, ...) \ 99 ibdev_level_ratelimited(ibdev_alert, ibdev, fmt, ##__VA_ARGS__) 100 #define ibdev_crit_ratelimited(ibdev, fmt, ...) \ 101 ibdev_level_ratelimited(ibdev_crit, ibdev, fmt, ##__VA_ARGS__) 102 #define ibdev_err_ratelimited(ibdev, fmt, ...) \ 103 ibdev_level_ratelimited(ibdev_err, ibdev, fmt, ##__VA_ARGS__) 104 #define ibdev_warn_ratelimited(ibdev, fmt, ...) \ 105 ibdev_level_ratelimited(ibdev_warn, ibdev, fmt, ##__VA_ARGS__) 106 #define ibdev_notice_ratelimited(ibdev, fmt, ...) \ 107 ibdev_level_ratelimited(ibdev_notice, ibdev, fmt, ##__VA_ARGS__) 108 #define ibdev_info_ratelimited(ibdev, fmt, ...) \ 109 ibdev_level_ratelimited(ibdev_info, ibdev, fmt, ##__VA_ARGS__) 110 111 #if defined(CONFIG_DYNAMIC_DEBUG) || \ 112 (defined(CONFIG_DYNAMIC_DEBUG_CORE) && defined(DYNAMIC_DEBUG_MODULE)) 113 /* descriptor check is first to prevent flooding with "callbacks suppressed" */ 114 #define ibdev_dbg_ratelimited(ibdev, fmt, ...) \ 115 do { \ 116 static DEFINE_RATELIMIT_STATE(_rs, \ 117 DEFAULT_RATELIMIT_INTERVAL, \ 118 DEFAULT_RATELIMIT_BURST); \ 119 DEFINE_DYNAMIC_DEBUG_METADATA(descriptor, fmt); \ 120 if (DYNAMIC_DEBUG_BRANCH(descriptor) && __ratelimit(&_rs)) \ 121 __dynamic_ibdev_dbg(&descriptor, ibdev, fmt, \ 122 ##__VA_ARGS__); \ 123 } while (0) 124 #else 125 __printf(2, 3) __cold 126 static inline 127 void ibdev_dbg_ratelimited(const struct ib_device *ibdev, const char *format, ...) {} 128 #endif 129 130 union ib_gid { 131 u8 raw[16]; 132 struct { 133 __be64 subnet_prefix; 134 __be64 interface_id; 135 } global; 136 }; 137 138 extern union ib_gid zgid; 139 140 enum ib_gid_type { 141 IB_GID_TYPE_IB = IB_UVERBS_GID_TYPE_IB, 142 IB_GID_TYPE_ROCE = IB_UVERBS_GID_TYPE_ROCE_V1, 143 IB_GID_TYPE_ROCE_UDP_ENCAP = IB_UVERBS_GID_TYPE_ROCE_V2, 144 IB_GID_TYPE_SIZE 145 }; 146 147 #define ROCE_V2_UDP_DPORT 4791 148 struct ib_gid_attr { 149 struct net_device __rcu *ndev; 150 struct ib_device *device; 151 union ib_gid gid; 152 enum ib_gid_type gid_type; 153 u16 index; 154 u32 port_num; 155 }; 156 157 enum { 158 /* set the local administered indication */ 159 IB_SA_WELL_KNOWN_GUID = BIT_ULL(57) | 2, 160 }; 161 162 enum rdma_transport_type { 163 RDMA_TRANSPORT_IB, 164 RDMA_TRANSPORT_IWARP, 165 RDMA_TRANSPORT_USNIC, 166 RDMA_TRANSPORT_USNIC_UDP, 167 RDMA_TRANSPORT_UNSPECIFIED, 168 }; 169 170 enum rdma_protocol_type { 171 RDMA_PROTOCOL_IB, 172 RDMA_PROTOCOL_IBOE, 173 RDMA_PROTOCOL_IWARP, 174 RDMA_PROTOCOL_USNIC_UDP 175 }; 176 177 __attribute_const__ enum rdma_transport_type 178 rdma_node_get_transport(unsigned int node_type); 179 180 enum rdma_network_type { 181 RDMA_NETWORK_IB, 182 RDMA_NETWORK_ROCE_V1, 183 RDMA_NETWORK_IPV4, 184 RDMA_NETWORK_IPV6 185 }; 186 187 static inline enum ib_gid_type ib_network_to_gid_type(enum rdma_network_type network_type) 188 { 189 if (network_type == RDMA_NETWORK_IPV4 || 190 network_type == RDMA_NETWORK_IPV6) 191 return IB_GID_TYPE_ROCE_UDP_ENCAP; 192 else if (network_type == RDMA_NETWORK_ROCE_V1) 193 return IB_GID_TYPE_ROCE; 194 else 195 return IB_GID_TYPE_IB; 196 } 197 198 static inline enum rdma_network_type 199 rdma_gid_attr_network_type(const struct ib_gid_attr *attr) 200 { 201 if (attr->gid_type == IB_GID_TYPE_IB) 202 return RDMA_NETWORK_IB; 203 204 if (attr->gid_type == IB_GID_TYPE_ROCE) 205 return RDMA_NETWORK_ROCE_V1; 206 207 if (ipv6_addr_v4mapped((struct in6_addr *)&attr->gid)) 208 return RDMA_NETWORK_IPV4; 209 else 210 return RDMA_NETWORK_IPV6; 211 } 212 213 enum rdma_link_layer { 214 IB_LINK_LAYER_UNSPECIFIED, 215 IB_LINK_LAYER_INFINIBAND, 216 IB_LINK_LAYER_ETHERNET, 217 }; 218 219 enum ib_device_cap_flags { 220 IB_DEVICE_RESIZE_MAX_WR = IB_UVERBS_DEVICE_RESIZE_MAX_WR, 221 IB_DEVICE_BAD_PKEY_CNTR = IB_UVERBS_DEVICE_BAD_PKEY_CNTR, 222 IB_DEVICE_BAD_QKEY_CNTR = IB_UVERBS_DEVICE_BAD_QKEY_CNTR, 223 IB_DEVICE_RAW_MULTI = IB_UVERBS_DEVICE_RAW_MULTI, 224 IB_DEVICE_AUTO_PATH_MIG = IB_UVERBS_DEVICE_AUTO_PATH_MIG, 225 IB_DEVICE_CHANGE_PHY_PORT = IB_UVERBS_DEVICE_CHANGE_PHY_PORT, 226 IB_DEVICE_UD_AV_PORT_ENFORCE = IB_UVERBS_DEVICE_UD_AV_PORT_ENFORCE, 227 IB_DEVICE_CURR_QP_STATE_MOD = IB_UVERBS_DEVICE_CURR_QP_STATE_MOD, 228 IB_DEVICE_SHUTDOWN_PORT = IB_UVERBS_DEVICE_SHUTDOWN_PORT, 229 /* IB_DEVICE_INIT_TYPE = IB_UVERBS_DEVICE_INIT_TYPE, (not in use) */ 230 IB_DEVICE_PORT_ACTIVE_EVENT = IB_UVERBS_DEVICE_PORT_ACTIVE_EVENT, 231 IB_DEVICE_SYS_IMAGE_GUID = IB_UVERBS_DEVICE_SYS_IMAGE_GUID, 232 IB_DEVICE_RC_RNR_NAK_GEN = IB_UVERBS_DEVICE_RC_RNR_NAK_GEN, 233 IB_DEVICE_SRQ_RESIZE = IB_UVERBS_DEVICE_SRQ_RESIZE, 234 IB_DEVICE_N_NOTIFY_CQ = IB_UVERBS_DEVICE_N_NOTIFY_CQ, 235 236 /* Reserved, old SEND_W_INV = 1 << 16,*/ 237 IB_DEVICE_MEM_WINDOW = IB_UVERBS_DEVICE_MEM_WINDOW, 238 /* 239 * Devices should set IB_DEVICE_UD_IP_SUM if they support 240 * insertion of UDP and TCP checksum on outgoing UD IPoIB 241 * messages and can verify the validity of checksum for 242 * incoming messages. Setting this flag implies that the 243 * IPoIB driver may set NETIF_F_IP_CSUM for datagram mode. 244 */ 245 IB_DEVICE_UD_IP_CSUM = IB_UVERBS_DEVICE_UD_IP_CSUM, 246 IB_DEVICE_XRC = IB_UVERBS_DEVICE_XRC, 247 248 /* 249 * This device supports the IB "base memory management extension", 250 * which includes support for fast registrations (IB_WR_REG_MR, 251 * IB_WR_LOCAL_INV and IB_WR_SEND_WITH_INV verbs). This flag should 252 * also be set by any iWarp device which must support FRs to comply 253 * to the iWarp verbs spec. iWarp devices also support the 254 * IB_WR_RDMA_READ_WITH_INV verb for RDMA READs that invalidate the 255 * stag. 256 */ 257 IB_DEVICE_MEM_MGT_EXTENSIONS = IB_UVERBS_DEVICE_MEM_MGT_EXTENSIONS, 258 IB_DEVICE_MEM_WINDOW_TYPE_2A = IB_UVERBS_DEVICE_MEM_WINDOW_TYPE_2A, 259 IB_DEVICE_MEM_WINDOW_TYPE_2B = IB_UVERBS_DEVICE_MEM_WINDOW_TYPE_2B, 260 IB_DEVICE_RC_IP_CSUM = IB_UVERBS_DEVICE_RC_IP_CSUM, 261 /* Deprecated. Please use IB_RAW_PACKET_CAP_IP_CSUM. */ 262 IB_DEVICE_RAW_IP_CSUM = IB_UVERBS_DEVICE_RAW_IP_CSUM, 263 IB_DEVICE_MANAGED_FLOW_STEERING = 264 IB_UVERBS_DEVICE_MANAGED_FLOW_STEERING, 265 /* Deprecated. Please use IB_RAW_PACKET_CAP_SCATTER_FCS. */ 266 IB_DEVICE_RAW_SCATTER_FCS = IB_UVERBS_DEVICE_RAW_SCATTER_FCS, 267 /* The device supports padding incoming writes to cacheline. */ 268 IB_DEVICE_PCI_WRITE_END_PADDING = 269 IB_UVERBS_DEVICE_PCI_WRITE_END_PADDING, 270 /* Placement type attributes */ 271 IB_DEVICE_FLUSH_GLOBAL = IB_UVERBS_DEVICE_FLUSH_GLOBAL, 272 IB_DEVICE_FLUSH_PERSISTENT = IB_UVERBS_DEVICE_FLUSH_PERSISTENT, 273 IB_DEVICE_ATOMIC_WRITE = IB_UVERBS_DEVICE_ATOMIC_WRITE, 274 }; 275 276 enum ib_kernel_cap_flags { 277 /* 278 * This device supports a per-device lkey or stag that can be 279 * used without performing a memory registration for the local 280 * memory. Note that ULPs should never check this flag, but 281 * instead of use the local_dma_lkey flag in the ib_pd structure, 282 * which will always contain a usable lkey. 283 */ 284 IBK_LOCAL_DMA_LKEY = 1 << 0, 285 /* IB_QP_CREATE_INTEGRITY_EN is supported to implement T10-PI */ 286 IBK_INTEGRITY_HANDOVER = 1 << 1, 287 /* IB_ACCESS_ON_DEMAND is supported during reg_user_mr() */ 288 IBK_ON_DEMAND_PAGING = 1 << 2, 289 /* IB_MR_TYPE_SG_GAPS is supported */ 290 IBK_SG_GAPS_REG = 1 << 3, 291 /* Driver supports RDMA_NLDEV_CMD_DELLINK */ 292 IBK_ALLOW_USER_UNREG = 1 << 4, 293 294 /* ipoib will use IB_QP_CREATE_BLOCK_MULTICAST_LOOPBACK */ 295 IBK_BLOCK_MULTICAST_LOOPBACK = 1 << 5, 296 /* iopib will use IB_QP_CREATE_IPOIB_UD_LSO for its QPs */ 297 IBK_UD_TSO = 1 << 6, 298 /* iopib will use the device ops: 299 * get_vf_config 300 * get_vf_guid 301 * get_vf_stats 302 * set_vf_guid 303 * set_vf_link_state 304 */ 305 IBK_VIRTUAL_FUNCTION = 1 << 7, 306 /* ipoib will use IB_QP_CREATE_NETDEV_USE for its QPs */ 307 IBK_RDMA_NETDEV_OPA = 1 << 8, 308 }; 309 310 enum ib_atomic_cap { 311 IB_ATOMIC_NONE, 312 IB_ATOMIC_HCA, 313 IB_ATOMIC_GLOB 314 }; 315 316 enum ib_odp_general_cap_bits { 317 IB_ODP_SUPPORT = 1 << 0, 318 IB_ODP_SUPPORT_IMPLICIT = 1 << 1, 319 }; 320 321 enum ib_odp_transport_cap_bits { 322 IB_ODP_SUPPORT_SEND = 1 << 0, 323 IB_ODP_SUPPORT_RECV = 1 << 1, 324 IB_ODP_SUPPORT_WRITE = 1 << 2, 325 IB_ODP_SUPPORT_READ = 1 << 3, 326 IB_ODP_SUPPORT_ATOMIC = 1 << 4, 327 IB_ODP_SUPPORT_SRQ_RECV = 1 << 5, 328 }; 329 330 struct ib_odp_caps { 331 uint64_t general_caps; 332 struct { 333 uint32_t rc_odp_caps; 334 uint32_t uc_odp_caps; 335 uint32_t ud_odp_caps; 336 uint32_t xrc_odp_caps; 337 } per_transport_caps; 338 }; 339 340 struct ib_rss_caps { 341 /* Corresponding bit will be set if qp type from 342 * 'enum ib_qp_type' is supported, e.g. 343 * supported_qpts |= 1 << IB_QPT_UD 344 */ 345 u32 supported_qpts; 346 u32 max_rwq_indirection_tables; 347 u32 max_rwq_indirection_table_size; 348 }; 349 350 enum ib_tm_cap_flags { 351 /* Support tag matching with rendezvous offload for RC transport */ 352 IB_TM_CAP_RNDV_RC = 1 << 0, 353 }; 354 355 struct ib_tm_caps { 356 /* Max size of RNDV header */ 357 u32 max_rndv_hdr_size; 358 /* Max number of entries in tag matching list */ 359 u32 max_num_tags; 360 /* From enum ib_tm_cap_flags */ 361 u32 flags; 362 /* Max number of outstanding list operations */ 363 u32 max_ops; 364 /* Max number of SGE in tag matching entry */ 365 u32 max_sge; 366 }; 367 368 struct ib_cq_init_attr { 369 unsigned int cqe; 370 u32 comp_vector; 371 u32 flags; 372 }; 373 374 enum ib_cq_attr_mask { 375 IB_CQ_MODERATE = 1 << 0, 376 }; 377 378 struct ib_cq_caps { 379 u16 max_cq_moderation_count; 380 u16 max_cq_moderation_period; 381 }; 382 383 struct ib_dm_mr_attr { 384 u64 length; 385 u64 offset; 386 u32 access_flags; 387 }; 388 389 struct ib_dm_alloc_attr { 390 u64 length; 391 u32 alignment; 392 u32 flags; 393 }; 394 395 struct ib_device_attr { 396 u64 fw_ver; 397 __be64 sys_image_guid; 398 u64 max_mr_size; 399 u64 page_size_cap; 400 u32 vendor_id; 401 u32 vendor_part_id; 402 u32 hw_ver; 403 int max_qp; 404 int max_qp_wr; 405 u64 device_cap_flags; 406 u64 kernel_cap_flags; 407 int max_send_sge; 408 int max_recv_sge; 409 int max_sge_rd; 410 int max_cq; 411 int max_cqe; 412 int max_mr; 413 int max_pd; 414 int max_qp_rd_atom; 415 int max_ee_rd_atom; 416 int max_res_rd_atom; 417 int max_qp_init_rd_atom; 418 int max_ee_init_rd_atom; 419 enum ib_atomic_cap atomic_cap; 420 enum ib_atomic_cap masked_atomic_cap; 421 int max_ee; 422 int max_rdd; 423 int max_mw; 424 int max_raw_ipv6_qp; 425 int max_raw_ethy_qp; 426 int max_mcast_grp; 427 int max_mcast_qp_attach; 428 int max_total_mcast_qp_attach; 429 int max_ah; 430 int max_srq; 431 int max_srq_wr; 432 int max_srq_sge; 433 unsigned int max_fast_reg_page_list_len; 434 unsigned int max_pi_fast_reg_page_list_len; 435 u16 max_pkeys; 436 u8 local_ca_ack_delay; 437 int sig_prot_cap; 438 int sig_guard_cap; 439 struct ib_odp_caps odp_caps; 440 uint64_t timestamp_mask; 441 uint64_t hca_core_clock; /* in KHZ */ 442 struct ib_rss_caps rss_caps; 443 u32 max_wq_type_rq; 444 u32 raw_packet_caps; /* Use ib_raw_packet_caps enum */ 445 struct ib_tm_caps tm_caps; 446 struct ib_cq_caps cq_caps; 447 u64 max_dm_size; 448 /* Max entries for sgl for optimized performance per READ */ 449 u32 max_sgl_rd; 450 }; 451 452 enum ib_mtu { 453 IB_MTU_256 = 1, 454 IB_MTU_512 = 2, 455 IB_MTU_1024 = 3, 456 IB_MTU_2048 = 4, 457 IB_MTU_4096 = 5 458 }; 459 460 enum opa_mtu { 461 OPA_MTU_8192 = 6, 462 OPA_MTU_10240 = 7 463 }; 464 465 static inline int ib_mtu_enum_to_int(enum ib_mtu mtu) 466 { 467 switch (mtu) { 468 case IB_MTU_256: return 256; 469 case IB_MTU_512: return 512; 470 case IB_MTU_1024: return 1024; 471 case IB_MTU_2048: return 2048; 472 case IB_MTU_4096: return 4096; 473 default: return -1; 474 } 475 } 476 477 static inline enum ib_mtu ib_mtu_int_to_enum(int mtu) 478 { 479 if (mtu >= 4096) 480 return IB_MTU_4096; 481 else if (mtu >= 2048) 482 return IB_MTU_2048; 483 else if (mtu >= 1024) 484 return IB_MTU_1024; 485 else if (mtu >= 512) 486 return IB_MTU_512; 487 else 488 return IB_MTU_256; 489 } 490 491 static inline int opa_mtu_enum_to_int(enum opa_mtu mtu) 492 { 493 switch (mtu) { 494 case OPA_MTU_8192: 495 return 8192; 496 case OPA_MTU_10240: 497 return 10240; 498 default: 499 return(ib_mtu_enum_to_int((enum ib_mtu)mtu)); 500 } 501 } 502 503 static inline enum opa_mtu opa_mtu_int_to_enum(int mtu) 504 { 505 if (mtu >= 10240) 506 return OPA_MTU_10240; 507 else if (mtu >= 8192) 508 return OPA_MTU_8192; 509 else 510 return ((enum opa_mtu)ib_mtu_int_to_enum(mtu)); 511 } 512 513 enum ib_port_state { 514 IB_PORT_NOP = 0, 515 IB_PORT_DOWN = 1, 516 IB_PORT_INIT = 2, 517 IB_PORT_ARMED = 3, 518 IB_PORT_ACTIVE = 4, 519 IB_PORT_ACTIVE_DEFER = 5 520 }; 521 522 static inline const char *__attribute_const__ 523 ib_port_state_to_str(enum ib_port_state state) 524 { 525 const char * const states[] = { 526 [IB_PORT_NOP] = "NOP", 527 [IB_PORT_DOWN] = "DOWN", 528 [IB_PORT_INIT] = "INIT", 529 [IB_PORT_ARMED] = "ARMED", 530 [IB_PORT_ACTIVE] = "ACTIVE", 531 [IB_PORT_ACTIVE_DEFER] = "ACTIVE_DEFER", 532 }; 533 534 if (state < ARRAY_SIZE(states)) 535 return states[state]; 536 return "UNKNOWN"; 537 } 538 539 enum ib_port_phys_state { 540 IB_PORT_PHYS_STATE_SLEEP = 1, 541 IB_PORT_PHYS_STATE_POLLING = 2, 542 IB_PORT_PHYS_STATE_DISABLED = 3, 543 IB_PORT_PHYS_STATE_PORT_CONFIGURATION_TRAINING = 4, 544 IB_PORT_PHYS_STATE_LINK_UP = 5, 545 IB_PORT_PHYS_STATE_LINK_ERROR_RECOVERY = 6, 546 IB_PORT_PHYS_STATE_PHY_TEST = 7, 547 }; 548 549 enum ib_port_width { 550 IB_WIDTH_1X = 1, 551 IB_WIDTH_2X = 16, 552 IB_WIDTH_4X = 2, 553 IB_WIDTH_8X = 4, 554 IB_WIDTH_12X = 8 555 }; 556 557 static inline int ib_width_enum_to_int(enum ib_port_width width) 558 { 559 switch (width) { 560 case IB_WIDTH_1X: return 1; 561 case IB_WIDTH_2X: return 2; 562 case IB_WIDTH_4X: return 4; 563 case IB_WIDTH_8X: return 8; 564 case IB_WIDTH_12X: return 12; 565 default: return -1; 566 } 567 } 568 569 enum ib_port_speed { 570 IB_SPEED_SDR = 1, 571 IB_SPEED_DDR = 2, 572 IB_SPEED_QDR = 4, 573 IB_SPEED_FDR10 = 8, 574 IB_SPEED_FDR = 16, 575 IB_SPEED_EDR = 32, 576 IB_SPEED_HDR = 64, 577 IB_SPEED_NDR = 128, 578 IB_SPEED_XDR = 256, 579 }; 580 581 enum ib_stat_flag { 582 IB_STAT_FLAG_OPTIONAL = 1 << 0, 583 }; 584 585 /** 586 * struct rdma_stat_desc 587 * @name - The name of the counter 588 * @flags - Flags of the counter; For example, IB_STAT_FLAG_OPTIONAL 589 * @priv - Driver private information; Core code should not use 590 */ 591 struct rdma_stat_desc { 592 const char *name; 593 unsigned int flags; 594 const void *priv; 595 }; 596 597 /** 598 * struct rdma_hw_stats 599 * @lock - Mutex to protect parallel write access to lifespan and values 600 * of counters, which are 64bits and not guaranteed to be written 601 * atomicaly on 32bits systems. 602 * @timestamp - Used by the core code to track when the last update was 603 * @lifespan - Used by the core code to determine how old the counters 604 * should be before being updated again. Stored in jiffies, defaults 605 * to 10 milliseconds, drivers can override the default be specifying 606 * their own value during their allocation routine. 607 * @descs - Array of pointers to static descriptors used for the counters 608 * in directory. 609 * @is_disabled - A bitmap to indicate each counter is currently disabled 610 * or not. 611 * @num_counters - How many hardware counters there are. If name is 612 * shorter than this number, a kernel oops will result. Driver authors 613 * are encouraged to leave BUILD_BUG_ON(ARRAY_SIZE(@name) < num_counters) 614 * in their code to prevent this. 615 * @value - Array of u64 counters that are accessed by the sysfs code and 616 * filled in by the drivers get_stats routine 617 */ 618 struct rdma_hw_stats { 619 struct mutex lock; /* Protect lifespan and values[] */ 620 unsigned long timestamp; 621 unsigned long lifespan; 622 const struct rdma_stat_desc *descs; 623 unsigned long *is_disabled; 624 int num_counters; 625 u64 value[] __counted_by(num_counters); 626 }; 627 628 #define RDMA_HW_STATS_DEFAULT_LIFESPAN 10 629 630 struct rdma_hw_stats *rdma_alloc_hw_stats_struct( 631 const struct rdma_stat_desc *descs, int num_counters, 632 unsigned long lifespan); 633 634 void rdma_free_hw_stats_struct(struct rdma_hw_stats *stats); 635 636 /* Define bits for the various functionality this port needs to be supported by 637 * the core. 638 */ 639 /* Management 0x00000FFF */ 640 #define RDMA_CORE_CAP_IB_MAD 0x00000001 641 #define RDMA_CORE_CAP_IB_SMI 0x00000002 642 #define RDMA_CORE_CAP_IB_CM 0x00000004 643 #define RDMA_CORE_CAP_IW_CM 0x00000008 644 #define RDMA_CORE_CAP_IB_SA 0x00000010 645 #define RDMA_CORE_CAP_OPA_MAD 0x00000020 646 647 /* Address format 0x000FF000 */ 648 #define RDMA_CORE_CAP_AF_IB 0x00001000 649 #define RDMA_CORE_CAP_ETH_AH 0x00002000 650 #define RDMA_CORE_CAP_OPA_AH 0x00004000 651 #define RDMA_CORE_CAP_IB_GRH_REQUIRED 0x00008000 652 653 /* Protocol 0xFFF00000 */ 654 #define RDMA_CORE_CAP_PROT_IB 0x00100000 655 #define RDMA_CORE_CAP_PROT_ROCE 0x00200000 656 #define RDMA_CORE_CAP_PROT_IWARP 0x00400000 657 #define RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP 0x00800000 658 #define RDMA_CORE_CAP_PROT_RAW_PACKET 0x01000000 659 #define RDMA_CORE_CAP_PROT_USNIC 0x02000000 660 661 #define RDMA_CORE_PORT_IB_GRH_REQUIRED (RDMA_CORE_CAP_IB_GRH_REQUIRED \ 662 | RDMA_CORE_CAP_PROT_ROCE \ 663 | RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP) 664 665 #define RDMA_CORE_PORT_IBA_IB (RDMA_CORE_CAP_PROT_IB \ 666 | RDMA_CORE_CAP_IB_MAD \ 667 | RDMA_CORE_CAP_IB_SMI \ 668 | RDMA_CORE_CAP_IB_CM \ 669 | RDMA_CORE_CAP_IB_SA \ 670 | RDMA_CORE_CAP_AF_IB) 671 #define RDMA_CORE_PORT_IBA_ROCE (RDMA_CORE_CAP_PROT_ROCE \ 672 | RDMA_CORE_CAP_IB_MAD \ 673 | RDMA_CORE_CAP_IB_CM \ 674 | RDMA_CORE_CAP_AF_IB \ 675 | RDMA_CORE_CAP_ETH_AH) 676 #define RDMA_CORE_PORT_IBA_ROCE_UDP_ENCAP \ 677 (RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP \ 678 | RDMA_CORE_CAP_IB_MAD \ 679 | RDMA_CORE_CAP_IB_CM \ 680 | RDMA_CORE_CAP_AF_IB \ 681 | RDMA_CORE_CAP_ETH_AH) 682 #define RDMA_CORE_PORT_IWARP (RDMA_CORE_CAP_PROT_IWARP \ 683 | RDMA_CORE_CAP_IW_CM) 684 #define RDMA_CORE_PORT_INTEL_OPA (RDMA_CORE_PORT_IBA_IB \ 685 | RDMA_CORE_CAP_OPA_MAD) 686 687 #define RDMA_CORE_PORT_RAW_PACKET (RDMA_CORE_CAP_PROT_RAW_PACKET) 688 689 #define RDMA_CORE_PORT_USNIC (RDMA_CORE_CAP_PROT_USNIC) 690 691 struct ib_port_attr { 692 u64 subnet_prefix; 693 enum ib_port_state state; 694 enum ib_mtu max_mtu; 695 enum ib_mtu active_mtu; 696 u32 phys_mtu; 697 int gid_tbl_len; 698 unsigned int ip_gids:1; 699 /* This is the value from PortInfo CapabilityMask, defined by IBA */ 700 u32 port_cap_flags; 701 u32 max_msg_sz; 702 u32 bad_pkey_cntr; 703 u32 qkey_viol_cntr; 704 u16 pkey_tbl_len; 705 u32 sm_lid; 706 u32 lid; 707 u8 lmc; 708 u8 max_vl_num; 709 u8 sm_sl; 710 u8 subnet_timeout; 711 u8 init_type_reply; 712 u8 active_width; 713 u16 active_speed; 714 u8 phys_state; 715 u16 port_cap_flags2; 716 }; 717 718 enum ib_device_modify_flags { 719 IB_DEVICE_MODIFY_SYS_IMAGE_GUID = 1 << 0, 720 IB_DEVICE_MODIFY_NODE_DESC = 1 << 1 721 }; 722 723 #define IB_DEVICE_NODE_DESC_MAX 64 724 725 struct ib_device_modify { 726 u64 sys_image_guid; 727 char node_desc[IB_DEVICE_NODE_DESC_MAX]; 728 }; 729 730 enum ib_port_modify_flags { 731 IB_PORT_SHUTDOWN = 1, 732 IB_PORT_INIT_TYPE = (1<<2), 733 IB_PORT_RESET_QKEY_CNTR = (1<<3), 734 IB_PORT_OPA_MASK_CHG = (1<<4) 735 }; 736 737 struct ib_port_modify { 738 u32 set_port_cap_mask; 739 u32 clr_port_cap_mask; 740 u8 init_type; 741 }; 742 743 enum ib_event_type { 744 IB_EVENT_CQ_ERR, 745 IB_EVENT_QP_FATAL, 746 IB_EVENT_QP_REQ_ERR, 747 IB_EVENT_QP_ACCESS_ERR, 748 IB_EVENT_COMM_EST, 749 IB_EVENT_SQ_DRAINED, 750 IB_EVENT_PATH_MIG, 751 IB_EVENT_PATH_MIG_ERR, 752 IB_EVENT_DEVICE_FATAL, 753 IB_EVENT_PORT_ACTIVE, 754 IB_EVENT_PORT_ERR, 755 IB_EVENT_LID_CHANGE, 756 IB_EVENT_PKEY_CHANGE, 757 IB_EVENT_SM_CHANGE, 758 IB_EVENT_SRQ_ERR, 759 IB_EVENT_SRQ_LIMIT_REACHED, 760 IB_EVENT_QP_LAST_WQE_REACHED, 761 IB_EVENT_CLIENT_REREGISTER, 762 IB_EVENT_GID_CHANGE, 763 IB_EVENT_WQ_FATAL, 764 }; 765 766 const char *__attribute_const__ ib_event_msg(enum ib_event_type event); 767 768 struct ib_event { 769 struct ib_device *device; 770 union { 771 struct ib_cq *cq; 772 struct ib_qp *qp; 773 struct ib_srq *srq; 774 struct ib_wq *wq; 775 u32 port_num; 776 } element; 777 enum ib_event_type event; 778 }; 779 780 struct ib_event_handler { 781 struct ib_device *device; 782 void (*handler)(struct ib_event_handler *, struct ib_event *); 783 struct list_head list; 784 }; 785 786 #define INIT_IB_EVENT_HANDLER(_ptr, _device, _handler) \ 787 do { \ 788 (_ptr)->device = _device; \ 789 (_ptr)->handler = _handler; \ 790 INIT_LIST_HEAD(&(_ptr)->list); \ 791 } while (0) 792 793 struct ib_global_route { 794 const struct ib_gid_attr *sgid_attr; 795 union ib_gid dgid; 796 u32 flow_label; 797 u8 sgid_index; 798 u8 hop_limit; 799 u8 traffic_class; 800 }; 801 802 struct ib_grh { 803 __be32 version_tclass_flow; 804 __be16 paylen; 805 u8 next_hdr; 806 u8 hop_limit; 807 union ib_gid sgid; 808 union ib_gid dgid; 809 }; 810 811 union rdma_network_hdr { 812 struct ib_grh ibgrh; 813 struct { 814 /* The IB spec states that if it's IPv4, the header 815 * is located in the last 20 bytes of the header. 816 */ 817 u8 reserved[20]; 818 struct iphdr roce4grh; 819 }; 820 }; 821 822 #define IB_QPN_MASK 0xFFFFFF 823 824 enum { 825 IB_MULTICAST_QPN = 0xffffff 826 }; 827 828 #define IB_LID_PERMISSIVE cpu_to_be16(0xFFFF) 829 #define IB_MULTICAST_LID_BASE cpu_to_be16(0xC000) 830 831 enum ib_ah_flags { 832 IB_AH_GRH = 1 833 }; 834 835 enum ib_rate { 836 IB_RATE_PORT_CURRENT = 0, 837 IB_RATE_2_5_GBPS = 2, 838 IB_RATE_5_GBPS = 5, 839 IB_RATE_10_GBPS = 3, 840 IB_RATE_20_GBPS = 6, 841 IB_RATE_30_GBPS = 4, 842 IB_RATE_40_GBPS = 7, 843 IB_RATE_60_GBPS = 8, 844 IB_RATE_80_GBPS = 9, 845 IB_RATE_120_GBPS = 10, 846 IB_RATE_14_GBPS = 11, 847 IB_RATE_56_GBPS = 12, 848 IB_RATE_112_GBPS = 13, 849 IB_RATE_168_GBPS = 14, 850 IB_RATE_25_GBPS = 15, 851 IB_RATE_100_GBPS = 16, 852 IB_RATE_200_GBPS = 17, 853 IB_RATE_300_GBPS = 18, 854 IB_RATE_28_GBPS = 19, 855 IB_RATE_50_GBPS = 20, 856 IB_RATE_400_GBPS = 21, 857 IB_RATE_600_GBPS = 22, 858 IB_RATE_800_GBPS = 23, 859 }; 860 861 /** 862 * ib_rate_to_mult - Convert the IB rate enum to a multiple of the 863 * base rate of 2.5 Gbit/sec. For example, IB_RATE_5_GBPS will be 864 * converted to 2, since 5 Gbit/sec is 2 * 2.5 Gbit/sec. 865 * @rate: rate to convert. 866 */ 867 __attribute_const__ int ib_rate_to_mult(enum ib_rate rate); 868 869 /** 870 * ib_rate_to_mbps - Convert the IB rate enum to Mbps. 871 * For example, IB_RATE_2_5_GBPS will be converted to 2500. 872 * @rate: rate to convert. 873 */ 874 __attribute_const__ int ib_rate_to_mbps(enum ib_rate rate); 875 876 877 /** 878 * enum ib_mr_type - memory region type 879 * @IB_MR_TYPE_MEM_REG: memory region that is used for 880 * normal registration 881 * @IB_MR_TYPE_SG_GAPS: memory region that is capable to 882 * register any arbitrary sg lists (without 883 * the normal mr constraints - see 884 * ib_map_mr_sg) 885 * @IB_MR_TYPE_DM: memory region that is used for device 886 * memory registration 887 * @IB_MR_TYPE_USER: memory region that is used for the user-space 888 * application 889 * @IB_MR_TYPE_DMA: memory region that is used for DMA operations 890 * without address translations (VA=PA) 891 * @IB_MR_TYPE_INTEGRITY: memory region that is used for 892 * data integrity operations 893 */ 894 enum ib_mr_type { 895 IB_MR_TYPE_MEM_REG, 896 IB_MR_TYPE_SG_GAPS, 897 IB_MR_TYPE_DM, 898 IB_MR_TYPE_USER, 899 IB_MR_TYPE_DMA, 900 IB_MR_TYPE_INTEGRITY, 901 }; 902 903 enum ib_mr_status_check { 904 IB_MR_CHECK_SIG_STATUS = 1, 905 }; 906 907 /** 908 * struct ib_mr_status - Memory region status container 909 * 910 * @fail_status: Bitmask of MR checks status. For each 911 * failed check a corresponding status bit is set. 912 * @sig_err: Additional info for IB_MR_CEHCK_SIG_STATUS 913 * failure. 914 */ 915 struct ib_mr_status { 916 u32 fail_status; 917 struct ib_sig_err sig_err; 918 }; 919 920 /** 921 * mult_to_ib_rate - Convert a multiple of 2.5 Gbit/sec to an IB rate 922 * enum. 923 * @mult: multiple to convert. 924 */ 925 __attribute_const__ enum ib_rate mult_to_ib_rate(int mult); 926 927 struct rdma_ah_init_attr { 928 struct rdma_ah_attr *ah_attr; 929 u32 flags; 930 struct net_device *xmit_slave; 931 }; 932 933 enum rdma_ah_attr_type { 934 RDMA_AH_ATTR_TYPE_UNDEFINED, 935 RDMA_AH_ATTR_TYPE_IB, 936 RDMA_AH_ATTR_TYPE_ROCE, 937 RDMA_AH_ATTR_TYPE_OPA, 938 }; 939 940 struct ib_ah_attr { 941 u16 dlid; 942 u8 src_path_bits; 943 }; 944 945 struct roce_ah_attr { 946 u8 dmac[ETH_ALEN]; 947 }; 948 949 struct opa_ah_attr { 950 u32 dlid; 951 u8 src_path_bits; 952 bool make_grd; 953 }; 954 955 struct rdma_ah_attr { 956 struct ib_global_route grh; 957 u8 sl; 958 u8 static_rate; 959 u32 port_num; 960 u8 ah_flags; 961 enum rdma_ah_attr_type type; 962 union { 963 struct ib_ah_attr ib; 964 struct roce_ah_attr roce; 965 struct opa_ah_attr opa; 966 }; 967 }; 968 969 enum ib_wc_status { 970 IB_WC_SUCCESS, 971 IB_WC_LOC_LEN_ERR, 972 IB_WC_LOC_QP_OP_ERR, 973 IB_WC_LOC_EEC_OP_ERR, 974 IB_WC_LOC_PROT_ERR, 975 IB_WC_WR_FLUSH_ERR, 976 IB_WC_MW_BIND_ERR, 977 IB_WC_BAD_RESP_ERR, 978 IB_WC_LOC_ACCESS_ERR, 979 IB_WC_REM_INV_REQ_ERR, 980 IB_WC_REM_ACCESS_ERR, 981 IB_WC_REM_OP_ERR, 982 IB_WC_RETRY_EXC_ERR, 983 IB_WC_RNR_RETRY_EXC_ERR, 984 IB_WC_LOC_RDD_VIOL_ERR, 985 IB_WC_REM_INV_RD_REQ_ERR, 986 IB_WC_REM_ABORT_ERR, 987 IB_WC_INV_EECN_ERR, 988 IB_WC_INV_EEC_STATE_ERR, 989 IB_WC_FATAL_ERR, 990 IB_WC_RESP_TIMEOUT_ERR, 991 IB_WC_GENERAL_ERR 992 }; 993 994 const char *__attribute_const__ ib_wc_status_msg(enum ib_wc_status status); 995 996 enum ib_wc_opcode { 997 IB_WC_SEND = IB_UVERBS_WC_SEND, 998 IB_WC_RDMA_WRITE = IB_UVERBS_WC_RDMA_WRITE, 999 IB_WC_RDMA_READ = IB_UVERBS_WC_RDMA_READ, 1000 IB_WC_COMP_SWAP = IB_UVERBS_WC_COMP_SWAP, 1001 IB_WC_FETCH_ADD = IB_UVERBS_WC_FETCH_ADD, 1002 IB_WC_BIND_MW = IB_UVERBS_WC_BIND_MW, 1003 IB_WC_LOCAL_INV = IB_UVERBS_WC_LOCAL_INV, 1004 IB_WC_LSO = IB_UVERBS_WC_TSO, 1005 IB_WC_ATOMIC_WRITE = IB_UVERBS_WC_ATOMIC_WRITE, 1006 IB_WC_REG_MR, 1007 IB_WC_MASKED_COMP_SWAP, 1008 IB_WC_MASKED_FETCH_ADD, 1009 IB_WC_FLUSH = IB_UVERBS_WC_FLUSH, 1010 /* 1011 * Set value of IB_WC_RECV so consumers can test if a completion is a 1012 * receive by testing (opcode & IB_WC_RECV). 1013 */ 1014 IB_WC_RECV = 1 << 7, 1015 IB_WC_RECV_RDMA_WITH_IMM 1016 }; 1017 1018 enum ib_wc_flags { 1019 IB_WC_GRH = 1, 1020 IB_WC_WITH_IMM = (1<<1), 1021 IB_WC_WITH_INVALIDATE = (1<<2), 1022 IB_WC_IP_CSUM_OK = (1<<3), 1023 IB_WC_WITH_SMAC = (1<<4), 1024 IB_WC_WITH_VLAN = (1<<5), 1025 IB_WC_WITH_NETWORK_HDR_TYPE = (1<<6), 1026 }; 1027 1028 struct ib_wc { 1029 union { 1030 u64 wr_id; 1031 struct ib_cqe *wr_cqe; 1032 }; 1033 enum ib_wc_status status; 1034 enum ib_wc_opcode opcode; 1035 u32 vendor_err; 1036 u32 byte_len; 1037 struct ib_qp *qp; 1038 union { 1039 __be32 imm_data; 1040 u32 invalidate_rkey; 1041 } ex; 1042 u32 src_qp; 1043 u32 slid; 1044 int wc_flags; 1045 u16 pkey_index; 1046 u8 sl; 1047 u8 dlid_path_bits; 1048 u32 port_num; /* valid only for DR SMPs on switches */ 1049 u8 smac[ETH_ALEN]; 1050 u16 vlan_id; 1051 u8 network_hdr_type; 1052 }; 1053 1054 enum ib_cq_notify_flags { 1055 IB_CQ_SOLICITED = 1 << 0, 1056 IB_CQ_NEXT_COMP = 1 << 1, 1057 IB_CQ_SOLICITED_MASK = IB_CQ_SOLICITED | IB_CQ_NEXT_COMP, 1058 IB_CQ_REPORT_MISSED_EVENTS = 1 << 2, 1059 }; 1060 1061 enum ib_srq_type { 1062 IB_SRQT_BASIC = IB_UVERBS_SRQT_BASIC, 1063 IB_SRQT_XRC = IB_UVERBS_SRQT_XRC, 1064 IB_SRQT_TM = IB_UVERBS_SRQT_TM, 1065 }; 1066 1067 static inline bool ib_srq_has_cq(enum ib_srq_type srq_type) 1068 { 1069 return srq_type == IB_SRQT_XRC || 1070 srq_type == IB_SRQT_TM; 1071 } 1072 1073 enum ib_srq_attr_mask { 1074 IB_SRQ_MAX_WR = 1 << 0, 1075 IB_SRQ_LIMIT = 1 << 1, 1076 }; 1077 1078 struct ib_srq_attr { 1079 u32 max_wr; 1080 u32 max_sge; 1081 u32 srq_limit; 1082 }; 1083 1084 struct ib_srq_init_attr { 1085 void (*event_handler)(struct ib_event *, void *); 1086 void *srq_context; 1087 struct ib_srq_attr attr; 1088 enum ib_srq_type srq_type; 1089 1090 struct { 1091 struct ib_cq *cq; 1092 union { 1093 struct { 1094 struct ib_xrcd *xrcd; 1095 } xrc; 1096 1097 struct { 1098 u32 max_num_tags; 1099 } tag_matching; 1100 }; 1101 } ext; 1102 }; 1103 1104 struct ib_qp_cap { 1105 u32 max_send_wr; 1106 u32 max_recv_wr; 1107 u32 max_send_sge; 1108 u32 max_recv_sge; 1109 u32 max_inline_data; 1110 1111 /* 1112 * Maximum number of rdma_rw_ctx structures in flight at a time. 1113 * ib_create_qp() will calculate the right amount of needed WRs 1114 * and MRs based on this. 1115 */ 1116 u32 max_rdma_ctxs; 1117 }; 1118 1119 enum ib_sig_type { 1120 IB_SIGNAL_ALL_WR, 1121 IB_SIGNAL_REQ_WR 1122 }; 1123 1124 enum ib_qp_type { 1125 /* 1126 * IB_QPT_SMI and IB_QPT_GSI have to be the first two entries 1127 * here (and in that order) since the MAD layer uses them as 1128 * indices into a 2-entry table. 1129 */ 1130 IB_QPT_SMI, 1131 IB_QPT_GSI, 1132 1133 IB_QPT_RC = IB_UVERBS_QPT_RC, 1134 IB_QPT_UC = IB_UVERBS_QPT_UC, 1135 IB_QPT_UD = IB_UVERBS_QPT_UD, 1136 IB_QPT_RAW_IPV6, 1137 IB_QPT_RAW_ETHERTYPE, 1138 IB_QPT_RAW_PACKET = IB_UVERBS_QPT_RAW_PACKET, 1139 IB_QPT_XRC_INI = IB_UVERBS_QPT_XRC_INI, 1140 IB_QPT_XRC_TGT = IB_UVERBS_QPT_XRC_TGT, 1141 IB_QPT_MAX, 1142 IB_QPT_DRIVER = IB_UVERBS_QPT_DRIVER, 1143 /* Reserve a range for qp types internal to the low level driver. 1144 * These qp types will not be visible at the IB core layer, so the 1145 * IB_QPT_MAX usages should not be affected in the core layer 1146 */ 1147 IB_QPT_RESERVED1 = 0x1000, 1148 IB_QPT_RESERVED2, 1149 IB_QPT_RESERVED3, 1150 IB_QPT_RESERVED4, 1151 IB_QPT_RESERVED5, 1152 IB_QPT_RESERVED6, 1153 IB_QPT_RESERVED7, 1154 IB_QPT_RESERVED8, 1155 IB_QPT_RESERVED9, 1156 IB_QPT_RESERVED10, 1157 }; 1158 1159 enum ib_qp_create_flags { 1160 IB_QP_CREATE_IPOIB_UD_LSO = 1 << 0, 1161 IB_QP_CREATE_BLOCK_MULTICAST_LOOPBACK = 1162 IB_UVERBS_QP_CREATE_BLOCK_MULTICAST_LOOPBACK, 1163 IB_QP_CREATE_CROSS_CHANNEL = 1 << 2, 1164 IB_QP_CREATE_MANAGED_SEND = 1 << 3, 1165 IB_QP_CREATE_MANAGED_RECV = 1 << 4, 1166 IB_QP_CREATE_NETIF_QP = 1 << 5, 1167 IB_QP_CREATE_INTEGRITY_EN = 1 << 6, 1168 IB_QP_CREATE_NETDEV_USE = 1 << 7, 1169 IB_QP_CREATE_SCATTER_FCS = 1170 IB_UVERBS_QP_CREATE_SCATTER_FCS, 1171 IB_QP_CREATE_CVLAN_STRIPPING = 1172 IB_UVERBS_QP_CREATE_CVLAN_STRIPPING, 1173 IB_QP_CREATE_SOURCE_QPN = 1 << 10, 1174 IB_QP_CREATE_PCI_WRITE_END_PADDING = 1175 IB_UVERBS_QP_CREATE_PCI_WRITE_END_PADDING, 1176 /* reserve bits 26-31 for low level drivers' internal use */ 1177 IB_QP_CREATE_RESERVED_START = 1 << 26, 1178 IB_QP_CREATE_RESERVED_END = 1 << 31, 1179 }; 1180 1181 /* 1182 * Note: users may not call ib_close_qp or ib_destroy_qp from the event_handler 1183 * callback to destroy the passed in QP. 1184 */ 1185 1186 struct ib_qp_init_attr { 1187 /* This callback occurs in workqueue context */ 1188 void (*event_handler)(struct ib_event *, void *); 1189 1190 void *qp_context; 1191 struct ib_cq *send_cq; 1192 struct ib_cq *recv_cq; 1193 struct ib_srq *srq; 1194 struct ib_xrcd *xrcd; /* XRC TGT QPs only */ 1195 struct ib_qp_cap cap; 1196 enum ib_sig_type sq_sig_type; 1197 enum ib_qp_type qp_type; 1198 u32 create_flags; 1199 1200 /* 1201 * Only needed for special QP types, or when using the RW API. 1202 */ 1203 u32 port_num; 1204 struct ib_rwq_ind_table *rwq_ind_tbl; 1205 u32 source_qpn; 1206 }; 1207 1208 struct ib_qp_open_attr { 1209 void (*event_handler)(struct ib_event *, void *); 1210 void *qp_context; 1211 u32 qp_num; 1212 enum ib_qp_type qp_type; 1213 }; 1214 1215 enum ib_rnr_timeout { 1216 IB_RNR_TIMER_655_36 = 0, 1217 IB_RNR_TIMER_000_01 = 1, 1218 IB_RNR_TIMER_000_02 = 2, 1219 IB_RNR_TIMER_000_03 = 3, 1220 IB_RNR_TIMER_000_04 = 4, 1221 IB_RNR_TIMER_000_06 = 5, 1222 IB_RNR_TIMER_000_08 = 6, 1223 IB_RNR_TIMER_000_12 = 7, 1224 IB_RNR_TIMER_000_16 = 8, 1225 IB_RNR_TIMER_000_24 = 9, 1226 IB_RNR_TIMER_000_32 = 10, 1227 IB_RNR_TIMER_000_48 = 11, 1228 IB_RNR_TIMER_000_64 = 12, 1229 IB_RNR_TIMER_000_96 = 13, 1230 IB_RNR_TIMER_001_28 = 14, 1231 IB_RNR_TIMER_001_92 = 15, 1232 IB_RNR_TIMER_002_56 = 16, 1233 IB_RNR_TIMER_003_84 = 17, 1234 IB_RNR_TIMER_005_12 = 18, 1235 IB_RNR_TIMER_007_68 = 19, 1236 IB_RNR_TIMER_010_24 = 20, 1237 IB_RNR_TIMER_015_36 = 21, 1238 IB_RNR_TIMER_020_48 = 22, 1239 IB_RNR_TIMER_030_72 = 23, 1240 IB_RNR_TIMER_040_96 = 24, 1241 IB_RNR_TIMER_061_44 = 25, 1242 IB_RNR_TIMER_081_92 = 26, 1243 IB_RNR_TIMER_122_88 = 27, 1244 IB_RNR_TIMER_163_84 = 28, 1245 IB_RNR_TIMER_245_76 = 29, 1246 IB_RNR_TIMER_327_68 = 30, 1247 IB_RNR_TIMER_491_52 = 31 1248 }; 1249 1250 enum ib_qp_attr_mask { 1251 IB_QP_STATE = 1, 1252 IB_QP_CUR_STATE = (1<<1), 1253 IB_QP_EN_SQD_ASYNC_NOTIFY = (1<<2), 1254 IB_QP_ACCESS_FLAGS = (1<<3), 1255 IB_QP_PKEY_INDEX = (1<<4), 1256 IB_QP_PORT = (1<<5), 1257 IB_QP_QKEY = (1<<6), 1258 IB_QP_AV = (1<<7), 1259 IB_QP_PATH_MTU = (1<<8), 1260 IB_QP_TIMEOUT = (1<<9), 1261 IB_QP_RETRY_CNT = (1<<10), 1262 IB_QP_RNR_RETRY = (1<<11), 1263 IB_QP_RQ_PSN = (1<<12), 1264 IB_QP_MAX_QP_RD_ATOMIC = (1<<13), 1265 IB_QP_ALT_PATH = (1<<14), 1266 IB_QP_MIN_RNR_TIMER = (1<<15), 1267 IB_QP_SQ_PSN = (1<<16), 1268 IB_QP_MAX_DEST_RD_ATOMIC = (1<<17), 1269 IB_QP_PATH_MIG_STATE = (1<<18), 1270 IB_QP_CAP = (1<<19), 1271 IB_QP_DEST_QPN = (1<<20), 1272 IB_QP_RESERVED1 = (1<<21), 1273 IB_QP_RESERVED2 = (1<<22), 1274 IB_QP_RESERVED3 = (1<<23), 1275 IB_QP_RESERVED4 = (1<<24), 1276 IB_QP_RATE_LIMIT = (1<<25), 1277 1278 IB_QP_ATTR_STANDARD_BITS = GENMASK(20, 0), 1279 }; 1280 1281 enum ib_qp_state { 1282 IB_QPS_RESET, 1283 IB_QPS_INIT, 1284 IB_QPS_RTR, 1285 IB_QPS_RTS, 1286 IB_QPS_SQD, 1287 IB_QPS_SQE, 1288 IB_QPS_ERR 1289 }; 1290 1291 enum ib_mig_state { 1292 IB_MIG_MIGRATED, 1293 IB_MIG_REARM, 1294 IB_MIG_ARMED 1295 }; 1296 1297 enum ib_mw_type { 1298 IB_MW_TYPE_1 = 1, 1299 IB_MW_TYPE_2 = 2 1300 }; 1301 1302 struct ib_qp_attr { 1303 enum ib_qp_state qp_state; 1304 enum ib_qp_state cur_qp_state; 1305 enum ib_mtu path_mtu; 1306 enum ib_mig_state path_mig_state; 1307 u32 qkey; 1308 u32 rq_psn; 1309 u32 sq_psn; 1310 u32 dest_qp_num; 1311 int qp_access_flags; 1312 struct ib_qp_cap cap; 1313 struct rdma_ah_attr ah_attr; 1314 struct rdma_ah_attr alt_ah_attr; 1315 u16 pkey_index; 1316 u16 alt_pkey_index; 1317 u8 en_sqd_async_notify; 1318 u8 sq_draining; 1319 u8 max_rd_atomic; 1320 u8 max_dest_rd_atomic; 1321 u8 min_rnr_timer; 1322 u32 port_num; 1323 u8 timeout; 1324 u8 retry_cnt; 1325 u8 rnr_retry; 1326 u32 alt_port_num; 1327 u8 alt_timeout; 1328 u32 rate_limit; 1329 struct net_device *xmit_slave; 1330 }; 1331 1332 enum ib_wr_opcode { 1333 /* These are shared with userspace */ 1334 IB_WR_RDMA_WRITE = IB_UVERBS_WR_RDMA_WRITE, 1335 IB_WR_RDMA_WRITE_WITH_IMM = IB_UVERBS_WR_RDMA_WRITE_WITH_IMM, 1336 IB_WR_SEND = IB_UVERBS_WR_SEND, 1337 IB_WR_SEND_WITH_IMM = IB_UVERBS_WR_SEND_WITH_IMM, 1338 IB_WR_RDMA_READ = IB_UVERBS_WR_RDMA_READ, 1339 IB_WR_ATOMIC_CMP_AND_SWP = IB_UVERBS_WR_ATOMIC_CMP_AND_SWP, 1340 IB_WR_ATOMIC_FETCH_AND_ADD = IB_UVERBS_WR_ATOMIC_FETCH_AND_ADD, 1341 IB_WR_BIND_MW = IB_UVERBS_WR_BIND_MW, 1342 IB_WR_LSO = IB_UVERBS_WR_TSO, 1343 IB_WR_SEND_WITH_INV = IB_UVERBS_WR_SEND_WITH_INV, 1344 IB_WR_RDMA_READ_WITH_INV = IB_UVERBS_WR_RDMA_READ_WITH_INV, 1345 IB_WR_LOCAL_INV = IB_UVERBS_WR_LOCAL_INV, 1346 IB_WR_MASKED_ATOMIC_CMP_AND_SWP = 1347 IB_UVERBS_WR_MASKED_ATOMIC_CMP_AND_SWP, 1348 IB_WR_MASKED_ATOMIC_FETCH_AND_ADD = 1349 IB_UVERBS_WR_MASKED_ATOMIC_FETCH_AND_ADD, 1350 IB_WR_FLUSH = IB_UVERBS_WR_FLUSH, 1351 IB_WR_ATOMIC_WRITE = IB_UVERBS_WR_ATOMIC_WRITE, 1352 1353 /* These are kernel only and can not be issued by userspace */ 1354 IB_WR_REG_MR = 0x20, 1355 IB_WR_REG_MR_INTEGRITY, 1356 1357 /* reserve values for low level drivers' internal use. 1358 * These values will not be used at all in the ib core layer. 1359 */ 1360 IB_WR_RESERVED1 = 0xf0, 1361 IB_WR_RESERVED2, 1362 IB_WR_RESERVED3, 1363 IB_WR_RESERVED4, 1364 IB_WR_RESERVED5, 1365 IB_WR_RESERVED6, 1366 IB_WR_RESERVED7, 1367 IB_WR_RESERVED8, 1368 IB_WR_RESERVED9, 1369 IB_WR_RESERVED10, 1370 }; 1371 1372 enum ib_send_flags { 1373 IB_SEND_FENCE = 1, 1374 IB_SEND_SIGNALED = (1<<1), 1375 IB_SEND_SOLICITED = (1<<2), 1376 IB_SEND_INLINE = (1<<3), 1377 IB_SEND_IP_CSUM = (1<<4), 1378 1379 /* reserve bits 26-31 for low level drivers' internal use */ 1380 IB_SEND_RESERVED_START = (1 << 26), 1381 IB_SEND_RESERVED_END = (1 << 31), 1382 }; 1383 1384 struct ib_sge { 1385 u64 addr; 1386 u32 length; 1387 u32 lkey; 1388 }; 1389 1390 struct ib_cqe { 1391 void (*done)(struct ib_cq *cq, struct ib_wc *wc); 1392 }; 1393 1394 struct ib_send_wr { 1395 struct ib_send_wr *next; 1396 union { 1397 u64 wr_id; 1398 struct ib_cqe *wr_cqe; 1399 }; 1400 struct ib_sge *sg_list; 1401 int num_sge; 1402 enum ib_wr_opcode opcode; 1403 int send_flags; 1404 union { 1405 __be32 imm_data; 1406 u32 invalidate_rkey; 1407 } ex; 1408 }; 1409 1410 struct ib_rdma_wr { 1411 struct ib_send_wr wr; 1412 u64 remote_addr; 1413 u32 rkey; 1414 }; 1415 1416 static inline const struct ib_rdma_wr *rdma_wr(const struct ib_send_wr *wr) 1417 { 1418 return container_of(wr, struct ib_rdma_wr, wr); 1419 } 1420 1421 struct ib_atomic_wr { 1422 struct ib_send_wr wr; 1423 u64 remote_addr; 1424 u64 compare_add; 1425 u64 swap; 1426 u64 compare_add_mask; 1427 u64 swap_mask; 1428 u32 rkey; 1429 }; 1430 1431 static inline const struct ib_atomic_wr *atomic_wr(const struct ib_send_wr *wr) 1432 { 1433 return container_of(wr, struct ib_atomic_wr, wr); 1434 } 1435 1436 struct ib_ud_wr { 1437 struct ib_send_wr wr; 1438 struct ib_ah *ah; 1439 void *header; 1440 int hlen; 1441 int mss; 1442 u32 remote_qpn; 1443 u32 remote_qkey; 1444 u16 pkey_index; /* valid for GSI only */ 1445 u32 port_num; /* valid for DR SMPs on switch only */ 1446 }; 1447 1448 static inline const struct ib_ud_wr *ud_wr(const struct ib_send_wr *wr) 1449 { 1450 return container_of(wr, struct ib_ud_wr, wr); 1451 } 1452 1453 struct ib_reg_wr { 1454 struct ib_send_wr wr; 1455 struct ib_mr *mr; 1456 u32 key; 1457 int access; 1458 }; 1459 1460 static inline const struct ib_reg_wr *reg_wr(const struct ib_send_wr *wr) 1461 { 1462 return container_of(wr, struct ib_reg_wr, wr); 1463 } 1464 1465 struct ib_recv_wr { 1466 struct ib_recv_wr *next; 1467 union { 1468 u64 wr_id; 1469 struct ib_cqe *wr_cqe; 1470 }; 1471 struct ib_sge *sg_list; 1472 int num_sge; 1473 }; 1474 1475 enum ib_access_flags { 1476 IB_ACCESS_LOCAL_WRITE = IB_UVERBS_ACCESS_LOCAL_WRITE, 1477 IB_ACCESS_REMOTE_WRITE = IB_UVERBS_ACCESS_REMOTE_WRITE, 1478 IB_ACCESS_REMOTE_READ = IB_UVERBS_ACCESS_REMOTE_READ, 1479 IB_ACCESS_REMOTE_ATOMIC = IB_UVERBS_ACCESS_REMOTE_ATOMIC, 1480 IB_ACCESS_MW_BIND = IB_UVERBS_ACCESS_MW_BIND, 1481 IB_ZERO_BASED = IB_UVERBS_ACCESS_ZERO_BASED, 1482 IB_ACCESS_ON_DEMAND = IB_UVERBS_ACCESS_ON_DEMAND, 1483 IB_ACCESS_HUGETLB = IB_UVERBS_ACCESS_HUGETLB, 1484 IB_ACCESS_RELAXED_ORDERING = IB_UVERBS_ACCESS_RELAXED_ORDERING, 1485 IB_ACCESS_FLUSH_GLOBAL = IB_UVERBS_ACCESS_FLUSH_GLOBAL, 1486 IB_ACCESS_FLUSH_PERSISTENT = IB_UVERBS_ACCESS_FLUSH_PERSISTENT, 1487 1488 IB_ACCESS_OPTIONAL = IB_UVERBS_ACCESS_OPTIONAL_RANGE, 1489 IB_ACCESS_SUPPORTED = 1490 ((IB_ACCESS_FLUSH_PERSISTENT << 1) - 1) | IB_ACCESS_OPTIONAL, 1491 }; 1492 1493 /* 1494 * XXX: these are apparently used for ->rereg_user_mr, no idea why they 1495 * are hidden here instead of a uapi header! 1496 */ 1497 enum ib_mr_rereg_flags { 1498 IB_MR_REREG_TRANS = 1, 1499 IB_MR_REREG_PD = (1<<1), 1500 IB_MR_REREG_ACCESS = (1<<2), 1501 IB_MR_REREG_SUPPORTED = ((IB_MR_REREG_ACCESS << 1) - 1) 1502 }; 1503 1504 struct ib_umem; 1505 1506 enum rdma_remove_reason { 1507 /* 1508 * Userspace requested uobject deletion or initial try 1509 * to remove uobject via cleanup. Call could fail 1510 */ 1511 RDMA_REMOVE_DESTROY, 1512 /* Context deletion. This call should delete the actual object itself */ 1513 RDMA_REMOVE_CLOSE, 1514 /* Driver is being hot-unplugged. This call should delete the actual object itself */ 1515 RDMA_REMOVE_DRIVER_REMOVE, 1516 /* uobj is being cleaned-up before being committed */ 1517 RDMA_REMOVE_ABORT, 1518 /* The driver failed to destroy the uobject and is being disconnected */ 1519 RDMA_REMOVE_DRIVER_FAILURE, 1520 }; 1521 1522 struct ib_rdmacg_object { 1523 #ifdef CONFIG_CGROUP_RDMA 1524 struct rdma_cgroup *cg; /* owner rdma cgroup */ 1525 #endif 1526 }; 1527 1528 struct ib_ucontext { 1529 struct ib_device *device; 1530 struct ib_uverbs_file *ufile; 1531 1532 struct ib_rdmacg_object cg_obj; 1533 /* 1534 * Implementation details of the RDMA core, don't use in drivers: 1535 */ 1536 struct rdma_restrack_entry res; 1537 struct xarray mmap_xa; 1538 }; 1539 1540 struct ib_uobject { 1541 u64 user_handle; /* handle given to us by userspace */ 1542 /* ufile & ucontext owning this object */ 1543 struct ib_uverbs_file *ufile; 1544 /* FIXME, save memory: ufile->context == context */ 1545 struct ib_ucontext *context; /* associated user context */ 1546 void *object; /* containing object */ 1547 struct list_head list; /* link to context's list */ 1548 struct ib_rdmacg_object cg_obj; /* rdmacg object */ 1549 int id; /* index into kernel idr */ 1550 struct kref ref; 1551 atomic_t usecnt; /* protects exclusive access */ 1552 struct rcu_head rcu; /* kfree_rcu() overhead */ 1553 1554 const struct uverbs_api_object *uapi_object; 1555 }; 1556 1557 struct ib_udata { 1558 const void __user *inbuf; 1559 void __user *outbuf; 1560 size_t inlen; 1561 size_t outlen; 1562 }; 1563 1564 struct ib_pd { 1565 u32 local_dma_lkey; 1566 u32 flags; 1567 struct ib_device *device; 1568 struct ib_uobject *uobject; 1569 atomic_t usecnt; /* count all resources */ 1570 1571 u32 unsafe_global_rkey; 1572 1573 /* 1574 * Implementation details of the RDMA core, don't use in drivers: 1575 */ 1576 struct ib_mr *__internal_mr; 1577 struct rdma_restrack_entry res; 1578 }; 1579 1580 struct ib_xrcd { 1581 struct ib_device *device; 1582 atomic_t usecnt; /* count all exposed resources */ 1583 struct inode *inode; 1584 struct rw_semaphore tgt_qps_rwsem; 1585 struct xarray tgt_qps; 1586 }; 1587 1588 struct ib_ah { 1589 struct ib_device *device; 1590 struct ib_pd *pd; 1591 struct ib_uobject *uobject; 1592 const struct ib_gid_attr *sgid_attr; 1593 enum rdma_ah_attr_type type; 1594 }; 1595 1596 typedef void (*ib_comp_handler)(struct ib_cq *cq, void *cq_context); 1597 1598 enum ib_poll_context { 1599 IB_POLL_SOFTIRQ, /* poll from softirq context */ 1600 IB_POLL_WORKQUEUE, /* poll from workqueue */ 1601 IB_POLL_UNBOUND_WORKQUEUE, /* poll from unbound workqueue */ 1602 IB_POLL_LAST_POOL_TYPE = IB_POLL_UNBOUND_WORKQUEUE, 1603 1604 IB_POLL_DIRECT, /* caller context, no hw completions */ 1605 }; 1606 1607 struct ib_cq { 1608 struct ib_device *device; 1609 struct ib_ucq_object *uobject; 1610 ib_comp_handler comp_handler; 1611 void (*event_handler)(struct ib_event *, void *); 1612 void *cq_context; 1613 int cqe; 1614 unsigned int cqe_used; 1615 atomic_t usecnt; /* count number of work queues */ 1616 enum ib_poll_context poll_ctx; 1617 struct ib_wc *wc; 1618 struct list_head pool_entry; 1619 union { 1620 struct irq_poll iop; 1621 struct work_struct work; 1622 }; 1623 struct workqueue_struct *comp_wq; 1624 struct dim *dim; 1625 1626 /* updated only by trace points */ 1627 ktime_t timestamp; 1628 u8 interrupt:1; 1629 u8 shared:1; 1630 unsigned int comp_vector; 1631 1632 /* 1633 * Implementation details of the RDMA core, don't use in drivers: 1634 */ 1635 struct rdma_restrack_entry res; 1636 }; 1637 1638 struct ib_srq { 1639 struct ib_device *device; 1640 struct ib_pd *pd; 1641 struct ib_usrq_object *uobject; 1642 void (*event_handler)(struct ib_event *, void *); 1643 void *srq_context; 1644 enum ib_srq_type srq_type; 1645 atomic_t usecnt; 1646 1647 struct { 1648 struct ib_cq *cq; 1649 union { 1650 struct { 1651 struct ib_xrcd *xrcd; 1652 u32 srq_num; 1653 } xrc; 1654 }; 1655 } ext; 1656 1657 /* 1658 * Implementation details of the RDMA core, don't use in drivers: 1659 */ 1660 struct rdma_restrack_entry res; 1661 }; 1662 1663 enum ib_raw_packet_caps { 1664 /* 1665 * Strip cvlan from incoming packet and report it in the matching work 1666 * completion is supported. 1667 */ 1668 IB_RAW_PACKET_CAP_CVLAN_STRIPPING = 1669 IB_UVERBS_RAW_PACKET_CAP_CVLAN_STRIPPING, 1670 /* 1671 * Scatter FCS field of an incoming packet to host memory is supported. 1672 */ 1673 IB_RAW_PACKET_CAP_SCATTER_FCS = IB_UVERBS_RAW_PACKET_CAP_SCATTER_FCS, 1674 /* Checksum offloads are supported (for both send and receive). */ 1675 IB_RAW_PACKET_CAP_IP_CSUM = IB_UVERBS_RAW_PACKET_CAP_IP_CSUM, 1676 /* 1677 * When a packet is received for an RQ with no receive WQEs, the 1678 * packet processing is delayed. 1679 */ 1680 IB_RAW_PACKET_CAP_DELAY_DROP = IB_UVERBS_RAW_PACKET_CAP_DELAY_DROP, 1681 }; 1682 1683 enum ib_wq_type { 1684 IB_WQT_RQ = IB_UVERBS_WQT_RQ, 1685 }; 1686 1687 enum ib_wq_state { 1688 IB_WQS_RESET, 1689 IB_WQS_RDY, 1690 IB_WQS_ERR 1691 }; 1692 1693 struct ib_wq { 1694 struct ib_device *device; 1695 struct ib_uwq_object *uobject; 1696 void *wq_context; 1697 void (*event_handler)(struct ib_event *, void *); 1698 struct ib_pd *pd; 1699 struct ib_cq *cq; 1700 u32 wq_num; 1701 enum ib_wq_state state; 1702 enum ib_wq_type wq_type; 1703 atomic_t usecnt; 1704 }; 1705 1706 enum ib_wq_flags { 1707 IB_WQ_FLAGS_CVLAN_STRIPPING = IB_UVERBS_WQ_FLAGS_CVLAN_STRIPPING, 1708 IB_WQ_FLAGS_SCATTER_FCS = IB_UVERBS_WQ_FLAGS_SCATTER_FCS, 1709 IB_WQ_FLAGS_DELAY_DROP = IB_UVERBS_WQ_FLAGS_DELAY_DROP, 1710 IB_WQ_FLAGS_PCI_WRITE_END_PADDING = 1711 IB_UVERBS_WQ_FLAGS_PCI_WRITE_END_PADDING, 1712 }; 1713 1714 struct ib_wq_init_attr { 1715 void *wq_context; 1716 enum ib_wq_type wq_type; 1717 u32 max_wr; 1718 u32 max_sge; 1719 struct ib_cq *cq; 1720 void (*event_handler)(struct ib_event *, void *); 1721 u32 create_flags; /* Use enum ib_wq_flags */ 1722 }; 1723 1724 enum ib_wq_attr_mask { 1725 IB_WQ_STATE = 1 << 0, 1726 IB_WQ_CUR_STATE = 1 << 1, 1727 IB_WQ_FLAGS = 1 << 2, 1728 }; 1729 1730 struct ib_wq_attr { 1731 enum ib_wq_state wq_state; 1732 enum ib_wq_state curr_wq_state; 1733 u32 flags; /* Use enum ib_wq_flags */ 1734 u32 flags_mask; /* Use enum ib_wq_flags */ 1735 }; 1736 1737 struct ib_rwq_ind_table { 1738 struct ib_device *device; 1739 struct ib_uobject *uobject; 1740 atomic_t usecnt; 1741 u32 ind_tbl_num; 1742 u32 log_ind_tbl_size; 1743 struct ib_wq **ind_tbl; 1744 }; 1745 1746 struct ib_rwq_ind_table_init_attr { 1747 u32 log_ind_tbl_size; 1748 /* Each entry is a pointer to Receive Work Queue */ 1749 struct ib_wq **ind_tbl; 1750 }; 1751 1752 enum port_pkey_state { 1753 IB_PORT_PKEY_NOT_VALID = 0, 1754 IB_PORT_PKEY_VALID = 1, 1755 IB_PORT_PKEY_LISTED = 2, 1756 }; 1757 1758 struct ib_qp_security; 1759 1760 struct ib_port_pkey { 1761 enum port_pkey_state state; 1762 u16 pkey_index; 1763 u32 port_num; 1764 struct list_head qp_list; 1765 struct list_head to_error_list; 1766 struct ib_qp_security *sec; 1767 }; 1768 1769 struct ib_ports_pkeys { 1770 struct ib_port_pkey main; 1771 struct ib_port_pkey alt; 1772 }; 1773 1774 struct ib_qp_security { 1775 struct ib_qp *qp; 1776 struct ib_device *dev; 1777 /* Hold this mutex when changing port and pkey settings. */ 1778 struct mutex mutex; 1779 struct ib_ports_pkeys *ports_pkeys; 1780 /* A list of all open shared QP handles. Required to enforce security 1781 * properly for all users of a shared QP. 1782 */ 1783 struct list_head shared_qp_list; 1784 void *security; 1785 bool destroying; 1786 atomic_t error_list_count; 1787 struct completion error_complete; 1788 int error_comps_pending; 1789 }; 1790 1791 /* 1792 * @max_write_sge: Maximum SGE elements per RDMA WRITE request. 1793 * @max_read_sge: Maximum SGE elements per RDMA READ request. 1794 */ 1795 struct ib_qp { 1796 struct ib_device *device; 1797 struct ib_pd *pd; 1798 struct ib_cq *send_cq; 1799 struct ib_cq *recv_cq; 1800 spinlock_t mr_lock; 1801 int mrs_used; 1802 struct list_head rdma_mrs; 1803 struct list_head sig_mrs; 1804 struct ib_srq *srq; 1805 struct completion srq_completion; 1806 struct ib_xrcd *xrcd; /* XRC TGT QPs only */ 1807 struct list_head xrcd_list; 1808 1809 /* count times opened, mcast attaches, flow attaches */ 1810 atomic_t usecnt; 1811 struct list_head open_list; 1812 struct ib_qp *real_qp; 1813 struct ib_uqp_object *uobject; 1814 void (*event_handler)(struct ib_event *, void *); 1815 void (*registered_event_handler)(struct ib_event *, void *); 1816 void *qp_context; 1817 /* sgid_attrs associated with the AV's */ 1818 const struct ib_gid_attr *av_sgid_attr; 1819 const struct ib_gid_attr *alt_path_sgid_attr; 1820 u32 qp_num; 1821 u32 max_write_sge; 1822 u32 max_read_sge; 1823 enum ib_qp_type qp_type; 1824 struct ib_rwq_ind_table *rwq_ind_tbl; 1825 struct ib_qp_security *qp_sec; 1826 u32 port; 1827 1828 bool integrity_en; 1829 /* 1830 * Implementation details of the RDMA core, don't use in drivers: 1831 */ 1832 struct rdma_restrack_entry res; 1833 1834 /* The counter the qp is bind to */ 1835 struct rdma_counter *counter; 1836 }; 1837 1838 struct ib_dm { 1839 struct ib_device *device; 1840 u32 length; 1841 u32 flags; 1842 struct ib_uobject *uobject; 1843 atomic_t usecnt; 1844 }; 1845 1846 struct ib_mr { 1847 struct ib_device *device; 1848 struct ib_pd *pd; 1849 u32 lkey; 1850 u32 rkey; 1851 u64 iova; 1852 u64 length; 1853 unsigned int page_size; 1854 enum ib_mr_type type; 1855 bool need_inval; 1856 union { 1857 struct ib_uobject *uobject; /* user */ 1858 struct list_head qp_entry; /* FR */ 1859 }; 1860 1861 struct ib_dm *dm; 1862 struct ib_sig_attrs *sig_attrs; /* only for IB_MR_TYPE_INTEGRITY MRs */ 1863 /* 1864 * Implementation details of the RDMA core, don't use in drivers: 1865 */ 1866 struct rdma_restrack_entry res; 1867 }; 1868 1869 struct ib_mw { 1870 struct ib_device *device; 1871 struct ib_pd *pd; 1872 struct ib_uobject *uobject; 1873 u32 rkey; 1874 enum ib_mw_type type; 1875 }; 1876 1877 /* Supported steering options */ 1878 enum ib_flow_attr_type { 1879 /* steering according to rule specifications */ 1880 IB_FLOW_ATTR_NORMAL = 0x0, 1881 /* default unicast and multicast rule - 1882 * receive all Eth traffic which isn't steered to any QP 1883 */ 1884 IB_FLOW_ATTR_ALL_DEFAULT = 0x1, 1885 /* default multicast rule - 1886 * receive all Eth multicast traffic which isn't steered to any QP 1887 */ 1888 IB_FLOW_ATTR_MC_DEFAULT = 0x2, 1889 /* sniffer rule - receive all port traffic */ 1890 IB_FLOW_ATTR_SNIFFER = 0x3 1891 }; 1892 1893 /* Supported steering header types */ 1894 enum ib_flow_spec_type { 1895 /* L2 headers*/ 1896 IB_FLOW_SPEC_ETH = 0x20, 1897 IB_FLOW_SPEC_IB = 0x22, 1898 /* L3 header*/ 1899 IB_FLOW_SPEC_IPV4 = 0x30, 1900 IB_FLOW_SPEC_IPV6 = 0x31, 1901 IB_FLOW_SPEC_ESP = 0x34, 1902 /* L4 headers*/ 1903 IB_FLOW_SPEC_TCP = 0x40, 1904 IB_FLOW_SPEC_UDP = 0x41, 1905 IB_FLOW_SPEC_VXLAN_TUNNEL = 0x50, 1906 IB_FLOW_SPEC_GRE = 0x51, 1907 IB_FLOW_SPEC_MPLS = 0x60, 1908 IB_FLOW_SPEC_INNER = 0x100, 1909 /* Actions */ 1910 IB_FLOW_SPEC_ACTION_TAG = 0x1000, 1911 IB_FLOW_SPEC_ACTION_DROP = 0x1001, 1912 IB_FLOW_SPEC_ACTION_HANDLE = 0x1002, 1913 IB_FLOW_SPEC_ACTION_COUNT = 0x1003, 1914 }; 1915 #define IB_FLOW_SPEC_LAYER_MASK 0xF0 1916 #define IB_FLOW_SPEC_SUPPORT_LAYERS 10 1917 1918 enum ib_flow_flags { 1919 IB_FLOW_ATTR_FLAGS_DONT_TRAP = 1UL << 1, /* Continue match, no steal */ 1920 IB_FLOW_ATTR_FLAGS_EGRESS = 1UL << 2, /* Egress flow */ 1921 IB_FLOW_ATTR_FLAGS_RESERVED = 1UL << 3 /* Must be last */ 1922 }; 1923 1924 struct ib_flow_eth_filter { 1925 u8 dst_mac[6]; 1926 u8 src_mac[6]; 1927 __be16 ether_type; 1928 __be16 vlan_tag; 1929 }; 1930 1931 struct ib_flow_spec_eth { 1932 u32 type; 1933 u16 size; 1934 struct ib_flow_eth_filter val; 1935 struct ib_flow_eth_filter mask; 1936 }; 1937 1938 struct ib_flow_ib_filter { 1939 __be16 dlid; 1940 __u8 sl; 1941 }; 1942 1943 struct ib_flow_spec_ib { 1944 u32 type; 1945 u16 size; 1946 struct ib_flow_ib_filter val; 1947 struct ib_flow_ib_filter mask; 1948 }; 1949 1950 /* IPv4 header flags */ 1951 enum ib_ipv4_flags { 1952 IB_IPV4_DONT_FRAG = 0x2, /* Don't enable packet fragmentation */ 1953 IB_IPV4_MORE_FRAG = 0X4 /* For All fragmented packets except the 1954 last have this flag set */ 1955 }; 1956 1957 struct ib_flow_ipv4_filter { 1958 __be32 src_ip; 1959 __be32 dst_ip; 1960 u8 proto; 1961 u8 tos; 1962 u8 ttl; 1963 u8 flags; 1964 }; 1965 1966 struct ib_flow_spec_ipv4 { 1967 u32 type; 1968 u16 size; 1969 struct ib_flow_ipv4_filter val; 1970 struct ib_flow_ipv4_filter mask; 1971 }; 1972 1973 struct ib_flow_ipv6_filter { 1974 u8 src_ip[16]; 1975 u8 dst_ip[16]; 1976 __be32 flow_label; 1977 u8 next_hdr; 1978 u8 traffic_class; 1979 u8 hop_limit; 1980 } __packed; 1981 1982 struct ib_flow_spec_ipv6 { 1983 u32 type; 1984 u16 size; 1985 struct ib_flow_ipv6_filter val; 1986 struct ib_flow_ipv6_filter mask; 1987 }; 1988 1989 struct ib_flow_tcp_udp_filter { 1990 __be16 dst_port; 1991 __be16 src_port; 1992 }; 1993 1994 struct ib_flow_spec_tcp_udp { 1995 u32 type; 1996 u16 size; 1997 struct ib_flow_tcp_udp_filter val; 1998 struct ib_flow_tcp_udp_filter mask; 1999 }; 2000 2001 struct ib_flow_tunnel_filter { 2002 __be32 tunnel_id; 2003 }; 2004 2005 /* ib_flow_spec_tunnel describes the Vxlan tunnel 2006 * the tunnel_id from val has the vni value 2007 */ 2008 struct ib_flow_spec_tunnel { 2009 u32 type; 2010 u16 size; 2011 struct ib_flow_tunnel_filter val; 2012 struct ib_flow_tunnel_filter mask; 2013 }; 2014 2015 struct ib_flow_esp_filter { 2016 __be32 spi; 2017 __be32 seq; 2018 }; 2019 2020 struct ib_flow_spec_esp { 2021 u32 type; 2022 u16 size; 2023 struct ib_flow_esp_filter val; 2024 struct ib_flow_esp_filter mask; 2025 }; 2026 2027 struct ib_flow_gre_filter { 2028 __be16 c_ks_res0_ver; 2029 __be16 protocol; 2030 __be32 key; 2031 }; 2032 2033 struct ib_flow_spec_gre { 2034 u32 type; 2035 u16 size; 2036 struct ib_flow_gre_filter val; 2037 struct ib_flow_gre_filter mask; 2038 }; 2039 2040 struct ib_flow_mpls_filter { 2041 __be32 tag; 2042 }; 2043 2044 struct ib_flow_spec_mpls { 2045 u32 type; 2046 u16 size; 2047 struct ib_flow_mpls_filter val; 2048 struct ib_flow_mpls_filter mask; 2049 }; 2050 2051 struct ib_flow_spec_action_tag { 2052 enum ib_flow_spec_type type; 2053 u16 size; 2054 u32 tag_id; 2055 }; 2056 2057 struct ib_flow_spec_action_drop { 2058 enum ib_flow_spec_type type; 2059 u16 size; 2060 }; 2061 2062 struct ib_flow_spec_action_handle { 2063 enum ib_flow_spec_type type; 2064 u16 size; 2065 struct ib_flow_action *act; 2066 }; 2067 2068 enum ib_counters_description { 2069 IB_COUNTER_PACKETS, 2070 IB_COUNTER_BYTES, 2071 }; 2072 2073 struct ib_flow_spec_action_count { 2074 enum ib_flow_spec_type type; 2075 u16 size; 2076 struct ib_counters *counters; 2077 }; 2078 2079 union ib_flow_spec { 2080 struct { 2081 u32 type; 2082 u16 size; 2083 }; 2084 struct ib_flow_spec_eth eth; 2085 struct ib_flow_spec_ib ib; 2086 struct ib_flow_spec_ipv4 ipv4; 2087 struct ib_flow_spec_tcp_udp tcp_udp; 2088 struct ib_flow_spec_ipv6 ipv6; 2089 struct ib_flow_spec_tunnel tunnel; 2090 struct ib_flow_spec_esp esp; 2091 struct ib_flow_spec_gre gre; 2092 struct ib_flow_spec_mpls mpls; 2093 struct ib_flow_spec_action_tag flow_tag; 2094 struct ib_flow_spec_action_drop drop; 2095 struct ib_flow_spec_action_handle action; 2096 struct ib_flow_spec_action_count flow_count; 2097 }; 2098 2099 struct ib_flow_attr { 2100 enum ib_flow_attr_type type; 2101 u16 size; 2102 u16 priority; 2103 u32 flags; 2104 u8 num_of_specs; 2105 u32 port; 2106 union ib_flow_spec flows[]; 2107 }; 2108 2109 struct ib_flow { 2110 struct ib_qp *qp; 2111 struct ib_device *device; 2112 struct ib_uobject *uobject; 2113 }; 2114 2115 enum ib_flow_action_type { 2116 IB_FLOW_ACTION_UNSPECIFIED, 2117 IB_FLOW_ACTION_ESP = 1, 2118 }; 2119 2120 struct ib_flow_action_attrs_esp_keymats { 2121 enum ib_uverbs_flow_action_esp_keymat protocol; 2122 union { 2123 struct ib_uverbs_flow_action_esp_keymat_aes_gcm aes_gcm; 2124 } keymat; 2125 }; 2126 2127 struct ib_flow_action_attrs_esp_replays { 2128 enum ib_uverbs_flow_action_esp_replay protocol; 2129 union { 2130 struct ib_uverbs_flow_action_esp_replay_bmp bmp; 2131 } replay; 2132 }; 2133 2134 enum ib_flow_action_attrs_esp_flags { 2135 /* All user-space flags at the top: Use enum ib_uverbs_flow_action_esp_flags 2136 * This is done in order to share the same flags between user-space and 2137 * kernel and spare an unnecessary translation. 2138 */ 2139 2140 /* Kernel flags */ 2141 IB_FLOW_ACTION_ESP_FLAGS_ESN_TRIGGERED = 1ULL << 32, 2142 IB_FLOW_ACTION_ESP_FLAGS_MOD_ESP_ATTRS = 1ULL << 33, 2143 }; 2144 2145 struct ib_flow_spec_list { 2146 struct ib_flow_spec_list *next; 2147 union ib_flow_spec spec; 2148 }; 2149 2150 struct ib_flow_action_attrs_esp { 2151 struct ib_flow_action_attrs_esp_keymats *keymat; 2152 struct ib_flow_action_attrs_esp_replays *replay; 2153 struct ib_flow_spec_list *encap; 2154 /* Used only if IB_FLOW_ACTION_ESP_FLAGS_ESN_TRIGGERED is enabled. 2155 * Value of 0 is a valid value. 2156 */ 2157 u32 esn; 2158 u32 spi; 2159 u32 seq; 2160 u32 tfc_pad; 2161 /* Use enum ib_flow_action_attrs_esp_flags */ 2162 u64 flags; 2163 u64 hard_limit_pkts; 2164 }; 2165 2166 struct ib_flow_action { 2167 struct ib_device *device; 2168 struct ib_uobject *uobject; 2169 enum ib_flow_action_type type; 2170 atomic_t usecnt; 2171 }; 2172 2173 struct ib_mad; 2174 2175 enum ib_process_mad_flags { 2176 IB_MAD_IGNORE_MKEY = 1, 2177 IB_MAD_IGNORE_BKEY = 2, 2178 IB_MAD_IGNORE_ALL = IB_MAD_IGNORE_MKEY | IB_MAD_IGNORE_BKEY 2179 }; 2180 2181 enum ib_mad_result { 2182 IB_MAD_RESULT_FAILURE = 0, /* (!SUCCESS is the important flag) */ 2183 IB_MAD_RESULT_SUCCESS = 1 << 0, /* MAD was successfully processed */ 2184 IB_MAD_RESULT_REPLY = 1 << 1, /* Reply packet needs to be sent */ 2185 IB_MAD_RESULT_CONSUMED = 1 << 2 /* Packet consumed: stop processing */ 2186 }; 2187 2188 struct ib_port_cache { 2189 u64 subnet_prefix; 2190 struct ib_pkey_cache *pkey; 2191 struct ib_gid_table *gid; 2192 u8 lmc; 2193 enum ib_port_state port_state; 2194 enum ib_port_state last_port_state; 2195 }; 2196 2197 struct ib_port_immutable { 2198 int pkey_tbl_len; 2199 int gid_tbl_len; 2200 u32 core_cap_flags; 2201 u32 max_mad_size; 2202 }; 2203 2204 struct ib_port_data { 2205 struct ib_device *ib_dev; 2206 2207 struct ib_port_immutable immutable; 2208 2209 spinlock_t pkey_list_lock; 2210 2211 spinlock_t netdev_lock; 2212 2213 struct list_head pkey_list; 2214 2215 struct ib_port_cache cache; 2216 2217 struct net_device __rcu *netdev; 2218 netdevice_tracker netdev_tracker; 2219 struct hlist_node ndev_hash_link; 2220 struct rdma_port_counter port_counter; 2221 struct ib_port *sysfs; 2222 }; 2223 2224 /* rdma netdev type - specifies protocol type */ 2225 enum rdma_netdev_t { 2226 RDMA_NETDEV_OPA_VNIC, 2227 RDMA_NETDEV_IPOIB, 2228 }; 2229 2230 /** 2231 * struct rdma_netdev - rdma netdev 2232 * For cases where netstack interfacing is required. 2233 */ 2234 struct rdma_netdev { 2235 void *clnt_priv; 2236 struct ib_device *hca; 2237 u32 port_num; 2238 int mtu; 2239 2240 /* 2241 * cleanup function must be specified. 2242 * FIXME: This is only used for OPA_VNIC and that usage should be 2243 * removed too. 2244 */ 2245 void (*free_rdma_netdev)(struct net_device *netdev); 2246 2247 /* control functions */ 2248 void (*set_id)(struct net_device *netdev, int id); 2249 /* send packet */ 2250 int (*send)(struct net_device *dev, struct sk_buff *skb, 2251 struct ib_ah *address, u32 dqpn); 2252 /* multicast */ 2253 int (*attach_mcast)(struct net_device *dev, struct ib_device *hca, 2254 union ib_gid *gid, u16 mlid, 2255 int set_qkey, u32 qkey); 2256 int (*detach_mcast)(struct net_device *dev, struct ib_device *hca, 2257 union ib_gid *gid, u16 mlid); 2258 /* timeout */ 2259 void (*tx_timeout)(struct net_device *dev, unsigned int txqueue); 2260 }; 2261 2262 struct rdma_netdev_alloc_params { 2263 size_t sizeof_priv; 2264 unsigned int txqs; 2265 unsigned int rxqs; 2266 void *param; 2267 2268 int (*initialize_rdma_netdev)(struct ib_device *device, u32 port_num, 2269 struct net_device *netdev, void *param); 2270 }; 2271 2272 struct ib_odp_counters { 2273 atomic64_t faults; 2274 atomic64_t faults_handled; 2275 atomic64_t invalidations; 2276 atomic64_t invalidations_handled; 2277 atomic64_t prefetch; 2278 }; 2279 2280 struct ib_counters { 2281 struct ib_device *device; 2282 struct ib_uobject *uobject; 2283 /* num of objects attached */ 2284 atomic_t usecnt; 2285 }; 2286 2287 struct ib_counters_read_attr { 2288 u64 *counters_buff; 2289 u32 ncounters; 2290 u32 flags; /* use enum ib_read_counters_flags */ 2291 }; 2292 2293 struct uverbs_attr_bundle; 2294 struct iw_cm_id; 2295 struct iw_cm_conn_param; 2296 2297 #define INIT_RDMA_OBJ_SIZE(ib_struct, drv_struct, member) \ 2298 .size_##ib_struct = \ 2299 (sizeof(struct drv_struct) + \ 2300 BUILD_BUG_ON_ZERO(offsetof(struct drv_struct, member)) + \ 2301 BUILD_BUG_ON_ZERO( \ 2302 !__same_type(((struct drv_struct *)NULL)->member, \ 2303 struct ib_struct))) 2304 2305 #define rdma_zalloc_drv_obj_gfp(ib_dev, ib_type, gfp) \ 2306 ((struct ib_type *)rdma_zalloc_obj(ib_dev, ib_dev->ops.size_##ib_type, \ 2307 gfp, false)) 2308 2309 #define rdma_zalloc_drv_obj_numa(ib_dev, ib_type) \ 2310 ((struct ib_type *)rdma_zalloc_obj(ib_dev, ib_dev->ops.size_##ib_type, \ 2311 GFP_KERNEL, true)) 2312 2313 #define rdma_zalloc_drv_obj(ib_dev, ib_type) \ 2314 rdma_zalloc_drv_obj_gfp(ib_dev, ib_type, GFP_KERNEL) 2315 2316 #define DECLARE_RDMA_OBJ_SIZE(ib_struct) size_t size_##ib_struct 2317 2318 struct rdma_user_mmap_entry { 2319 struct kref ref; 2320 struct ib_ucontext *ucontext; 2321 unsigned long start_pgoff; 2322 size_t npages; 2323 bool driver_removed; 2324 }; 2325 2326 /* Return the offset (in bytes) the user should pass to libc's mmap() */ 2327 static inline u64 2328 rdma_user_mmap_get_offset(const struct rdma_user_mmap_entry *entry) 2329 { 2330 return (u64)entry->start_pgoff << PAGE_SHIFT; 2331 } 2332 2333 /** 2334 * struct ib_device_ops - InfiniBand device operations 2335 * This structure defines all the InfiniBand device operations, providers will 2336 * need to define the supported operations, otherwise they will be set to null. 2337 */ 2338 struct ib_device_ops { 2339 struct module *owner; 2340 enum rdma_driver_id driver_id; 2341 u32 uverbs_abi_ver; 2342 unsigned int uverbs_no_driver_id_binding:1; 2343 2344 /* 2345 * NOTE: New drivers should not make use of device_group; instead new 2346 * device parameter should be exposed via netlink command. This 2347 * mechanism exists only for existing drivers. 2348 */ 2349 const struct attribute_group *device_group; 2350 const struct attribute_group **port_groups; 2351 2352 int (*post_send)(struct ib_qp *qp, const struct ib_send_wr *send_wr, 2353 const struct ib_send_wr **bad_send_wr); 2354 int (*post_recv)(struct ib_qp *qp, const struct ib_recv_wr *recv_wr, 2355 const struct ib_recv_wr **bad_recv_wr); 2356 void (*drain_rq)(struct ib_qp *qp); 2357 void (*drain_sq)(struct ib_qp *qp); 2358 int (*poll_cq)(struct ib_cq *cq, int num_entries, struct ib_wc *wc); 2359 int (*peek_cq)(struct ib_cq *cq, int wc_cnt); 2360 int (*req_notify_cq)(struct ib_cq *cq, enum ib_cq_notify_flags flags); 2361 int (*post_srq_recv)(struct ib_srq *srq, 2362 const struct ib_recv_wr *recv_wr, 2363 const struct ib_recv_wr **bad_recv_wr); 2364 int (*process_mad)(struct ib_device *device, int process_mad_flags, 2365 u32 port_num, const struct ib_wc *in_wc, 2366 const struct ib_grh *in_grh, 2367 const struct ib_mad *in_mad, struct ib_mad *out_mad, 2368 size_t *out_mad_size, u16 *out_mad_pkey_index); 2369 int (*query_device)(struct ib_device *device, 2370 struct ib_device_attr *device_attr, 2371 struct ib_udata *udata); 2372 int (*modify_device)(struct ib_device *device, int device_modify_mask, 2373 struct ib_device_modify *device_modify); 2374 void (*get_dev_fw_str)(struct ib_device *device, char *str); 2375 const struct cpumask *(*get_vector_affinity)(struct ib_device *ibdev, 2376 int comp_vector); 2377 int (*query_port)(struct ib_device *device, u32 port_num, 2378 struct ib_port_attr *port_attr); 2379 int (*modify_port)(struct ib_device *device, u32 port_num, 2380 int port_modify_mask, 2381 struct ib_port_modify *port_modify); 2382 /** 2383 * The following mandatory functions are used only at device 2384 * registration. Keep functions such as these at the end of this 2385 * structure to avoid cache line misses when accessing struct ib_device 2386 * in fast paths. 2387 */ 2388 int (*get_port_immutable)(struct ib_device *device, u32 port_num, 2389 struct ib_port_immutable *immutable); 2390 enum rdma_link_layer (*get_link_layer)(struct ib_device *device, 2391 u32 port_num); 2392 /** 2393 * When calling get_netdev, the HW vendor's driver should return the 2394 * net device of device @device at port @port_num or NULL if such 2395 * a net device doesn't exist. The vendor driver should call dev_hold 2396 * on this net device. The HW vendor's device driver must guarantee 2397 * that this function returns NULL before the net device has finished 2398 * NETDEV_UNREGISTER state. 2399 */ 2400 struct net_device *(*get_netdev)(struct ib_device *device, 2401 u32 port_num); 2402 /** 2403 * rdma netdev operation 2404 * 2405 * Driver implementing alloc_rdma_netdev or rdma_netdev_get_params 2406 * must return -EOPNOTSUPP if it doesn't support the specified type. 2407 */ 2408 struct net_device *(*alloc_rdma_netdev)( 2409 struct ib_device *device, u32 port_num, enum rdma_netdev_t type, 2410 const char *name, unsigned char name_assign_type, 2411 void (*setup)(struct net_device *)); 2412 2413 int (*rdma_netdev_get_params)(struct ib_device *device, u32 port_num, 2414 enum rdma_netdev_t type, 2415 struct rdma_netdev_alloc_params *params); 2416 /** 2417 * query_gid should be return GID value for @device, when @port_num 2418 * link layer is either IB or iWarp. It is no-op if @port_num port 2419 * is RoCE link layer. 2420 */ 2421 int (*query_gid)(struct ib_device *device, u32 port_num, int index, 2422 union ib_gid *gid); 2423 /** 2424 * When calling add_gid, the HW vendor's driver should add the gid 2425 * of device of port at gid index available at @attr. Meta-info of 2426 * that gid (for example, the network device related to this gid) is 2427 * available at @attr. @context allows the HW vendor driver to store 2428 * extra information together with a GID entry. The HW vendor driver may 2429 * allocate memory to contain this information and store it in @context 2430 * when a new GID entry is written to. Params are consistent until the 2431 * next call of add_gid or delete_gid. The function should return 0 on 2432 * success or error otherwise. The function could be called 2433 * concurrently for different ports. This function is only called when 2434 * roce_gid_table is used. 2435 */ 2436 int (*add_gid)(const struct ib_gid_attr *attr, void **context); 2437 /** 2438 * When calling del_gid, the HW vendor's driver should delete the 2439 * gid of device @device at gid index gid_index of port port_num 2440 * available in @attr. 2441 * Upon the deletion of a GID entry, the HW vendor must free any 2442 * allocated memory. The caller will clear @context afterwards. 2443 * This function is only called when roce_gid_table is used. 2444 */ 2445 int (*del_gid)(const struct ib_gid_attr *attr, void **context); 2446 int (*query_pkey)(struct ib_device *device, u32 port_num, u16 index, 2447 u16 *pkey); 2448 int (*alloc_ucontext)(struct ib_ucontext *context, 2449 struct ib_udata *udata); 2450 void (*dealloc_ucontext)(struct ib_ucontext *context); 2451 int (*mmap)(struct ib_ucontext *context, struct vm_area_struct *vma); 2452 /** 2453 * This will be called once refcount of an entry in mmap_xa reaches 2454 * zero. The type of the memory that was mapped may differ between 2455 * entries and is opaque to the rdma_user_mmap interface. 2456 * Therefore needs to be implemented by the driver in mmap_free. 2457 */ 2458 void (*mmap_free)(struct rdma_user_mmap_entry *entry); 2459 void (*disassociate_ucontext)(struct ib_ucontext *ibcontext); 2460 int (*alloc_pd)(struct ib_pd *pd, struct ib_udata *udata); 2461 int (*dealloc_pd)(struct ib_pd *pd, struct ib_udata *udata); 2462 int (*create_ah)(struct ib_ah *ah, struct rdma_ah_init_attr *attr, 2463 struct ib_udata *udata); 2464 int (*create_user_ah)(struct ib_ah *ah, struct rdma_ah_init_attr *attr, 2465 struct ib_udata *udata); 2466 int (*modify_ah)(struct ib_ah *ah, struct rdma_ah_attr *ah_attr); 2467 int (*query_ah)(struct ib_ah *ah, struct rdma_ah_attr *ah_attr); 2468 int (*destroy_ah)(struct ib_ah *ah, u32 flags); 2469 int (*create_srq)(struct ib_srq *srq, 2470 struct ib_srq_init_attr *srq_init_attr, 2471 struct ib_udata *udata); 2472 int (*modify_srq)(struct ib_srq *srq, struct ib_srq_attr *srq_attr, 2473 enum ib_srq_attr_mask srq_attr_mask, 2474 struct ib_udata *udata); 2475 int (*query_srq)(struct ib_srq *srq, struct ib_srq_attr *srq_attr); 2476 int (*destroy_srq)(struct ib_srq *srq, struct ib_udata *udata); 2477 int (*create_qp)(struct ib_qp *qp, struct ib_qp_init_attr *qp_init_attr, 2478 struct ib_udata *udata); 2479 int (*modify_qp)(struct ib_qp *qp, struct ib_qp_attr *qp_attr, 2480 int qp_attr_mask, struct ib_udata *udata); 2481 int (*query_qp)(struct ib_qp *qp, struct ib_qp_attr *qp_attr, 2482 int qp_attr_mask, struct ib_qp_init_attr *qp_init_attr); 2483 int (*destroy_qp)(struct ib_qp *qp, struct ib_udata *udata); 2484 int (*create_cq)(struct ib_cq *cq, const struct ib_cq_init_attr *attr, 2485 struct uverbs_attr_bundle *attrs); 2486 int (*modify_cq)(struct ib_cq *cq, u16 cq_count, u16 cq_period); 2487 int (*destroy_cq)(struct ib_cq *cq, struct ib_udata *udata); 2488 int (*resize_cq)(struct ib_cq *cq, int cqe, struct ib_udata *udata); 2489 struct ib_mr *(*get_dma_mr)(struct ib_pd *pd, int mr_access_flags); 2490 struct ib_mr *(*reg_user_mr)(struct ib_pd *pd, u64 start, u64 length, 2491 u64 virt_addr, int mr_access_flags, 2492 struct ib_udata *udata); 2493 struct ib_mr *(*reg_user_mr_dmabuf)(struct ib_pd *pd, u64 offset, 2494 u64 length, u64 virt_addr, int fd, 2495 int mr_access_flags, 2496 struct uverbs_attr_bundle *attrs); 2497 struct ib_mr *(*rereg_user_mr)(struct ib_mr *mr, int flags, u64 start, 2498 u64 length, u64 virt_addr, 2499 int mr_access_flags, struct ib_pd *pd, 2500 struct ib_udata *udata); 2501 int (*dereg_mr)(struct ib_mr *mr, struct ib_udata *udata); 2502 struct ib_mr *(*alloc_mr)(struct ib_pd *pd, enum ib_mr_type mr_type, 2503 u32 max_num_sg); 2504 struct ib_mr *(*alloc_mr_integrity)(struct ib_pd *pd, 2505 u32 max_num_data_sg, 2506 u32 max_num_meta_sg); 2507 int (*advise_mr)(struct ib_pd *pd, 2508 enum ib_uverbs_advise_mr_advice advice, u32 flags, 2509 struct ib_sge *sg_list, u32 num_sge, 2510 struct uverbs_attr_bundle *attrs); 2511 2512 /* 2513 * Kernel users should universally support relaxed ordering (RO), as 2514 * they are designed to read data only after observing the CQE and use 2515 * the DMA API correctly. 2516 * 2517 * Some drivers implicitly enable RO if platform supports it. 2518 */ 2519 int (*map_mr_sg)(struct ib_mr *mr, struct scatterlist *sg, int sg_nents, 2520 unsigned int *sg_offset); 2521 int (*check_mr_status)(struct ib_mr *mr, u32 check_mask, 2522 struct ib_mr_status *mr_status); 2523 int (*alloc_mw)(struct ib_mw *mw, struct ib_udata *udata); 2524 int (*dealloc_mw)(struct ib_mw *mw); 2525 int (*attach_mcast)(struct ib_qp *qp, union ib_gid *gid, u16 lid); 2526 int (*detach_mcast)(struct ib_qp *qp, union ib_gid *gid, u16 lid); 2527 int (*alloc_xrcd)(struct ib_xrcd *xrcd, struct ib_udata *udata); 2528 int (*dealloc_xrcd)(struct ib_xrcd *xrcd, struct ib_udata *udata); 2529 struct ib_flow *(*create_flow)(struct ib_qp *qp, 2530 struct ib_flow_attr *flow_attr, 2531 struct ib_udata *udata); 2532 int (*destroy_flow)(struct ib_flow *flow_id); 2533 int (*destroy_flow_action)(struct ib_flow_action *action); 2534 int (*set_vf_link_state)(struct ib_device *device, int vf, u32 port, 2535 int state); 2536 int (*get_vf_config)(struct ib_device *device, int vf, u32 port, 2537 struct ifla_vf_info *ivf); 2538 int (*get_vf_stats)(struct ib_device *device, int vf, u32 port, 2539 struct ifla_vf_stats *stats); 2540 int (*get_vf_guid)(struct ib_device *device, int vf, u32 port, 2541 struct ifla_vf_guid *node_guid, 2542 struct ifla_vf_guid *port_guid); 2543 int (*set_vf_guid)(struct ib_device *device, int vf, u32 port, u64 guid, 2544 int type); 2545 struct ib_wq *(*create_wq)(struct ib_pd *pd, 2546 struct ib_wq_init_attr *init_attr, 2547 struct ib_udata *udata); 2548 int (*destroy_wq)(struct ib_wq *wq, struct ib_udata *udata); 2549 int (*modify_wq)(struct ib_wq *wq, struct ib_wq_attr *attr, 2550 u32 wq_attr_mask, struct ib_udata *udata); 2551 int (*create_rwq_ind_table)(struct ib_rwq_ind_table *ib_rwq_ind_table, 2552 struct ib_rwq_ind_table_init_attr *init_attr, 2553 struct ib_udata *udata); 2554 int (*destroy_rwq_ind_table)(struct ib_rwq_ind_table *wq_ind_table); 2555 struct ib_dm *(*alloc_dm)(struct ib_device *device, 2556 struct ib_ucontext *context, 2557 struct ib_dm_alloc_attr *attr, 2558 struct uverbs_attr_bundle *attrs); 2559 int (*dealloc_dm)(struct ib_dm *dm, struct uverbs_attr_bundle *attrs); 2560 struct ib_mr *(*reg_dm_mr)(struct ib_pd *pd, struct ib_dm *dm, 2561 struct ib_dm_mr_attr *attr, 2562 struct uverbs_attr_bundle *attrs); 2563 int (*create_counters)(struct ib_counters *counters, 2564 struct uverbs_attr_bundle *attrs); 2565 int (*destroy_counters)(struct ib_counters *counters); 2566 int (*read_counters)(struct ib_counters *counters, 2567 struct ib_counters_read_attr *counters_read_attr, 2568 struct uverbs_attr_bundle *attrs); 2569 int (*map_mr_sg_pi)(struct ib_mr *mr, struct scatterlist *data_sg, 2570 int data_sg_nents, unsigned int *data_sg_offset, 2571 struct scatterlist *meta_sg, int meta_sg_nents, 2572 unsigned int *meta_sg_offset); 2573 2574 /** 2575 * alloc_hw_[device,port]_stats - Allocate a struct rdma_hw_stats and 2576 * fill in the driver initialized data. The struct is kfree()'ed by 2577 * the sysfs core when the device is removed. A lifespan of -1 in the 2578 * return struct tells the core to set a default lifespan. 2579 */ 2580 struct rdma_hw_stats *(*alloc_hw_device_stats)(struct ib_device *device); 2581 struct rdma_hw_stats *(*alloc_hw_port_stats)(struct ib_device *device, 2582 u32 port_num); 2583 /** 2584 * get_hw_stats - Fill in the counter value(s) in the stats struct. 2585 * @index - The index in the value array we wish to have updated, or 2586 * num_counters if we want all stats updated 2587 * Return codes - 2588 * < 0 - Error, no counters updated 2589 * index - Updated the single counter pointed to by index 2590 * num_counters - Updated all counters (will reset the timestamp 2591 * and prevent further calls for lifespan milliseconds) 2592 * Drivers are allowed to update all counters in leiu of just the 2593 * one given in index at their option 2594 */ 2595 int (*get_hw_stats)(struct ib_device *device, 2596 struct rdma_hw_stats *stats, u32 port, int index); 2597 2598 /** 2599 * modify_hw_stat - Modify the counter configuration 2600 * @enable: true/false when enable/disable a counter 2601 * Return codes - 0 on success or error code otherwise. 2602 */ 2603 int (*modify_hw_stat)(struct ib_device *device, u32 port, 2604 unsigned int counter_index, bool enable); 2605 /** 2606 * Allows rdma drivers to add their own restrack attributes. 2607 */ 2608 int (*fill_res_mr_entry)(struct sk_buff *msg, struct ib_mr *ibmr); 2609 int (*fill_res_mr_entry_raw)(struct sk_buff *msg, struct ib_mr *ibmr); 2610 int (*fill_res_cq_entry)(struct sk_buff *msg, struct ib_cq *ibcq); 2611 int (*fill_res_cq_entry_raw)(struct sk_buff *msg, struct ib_cq *ibcq); 2612 int (*fill_res_qp_entry)(struct sk_buff *msg, struct ib_qp *ibqp); 2613 int (*fill_res_qp_entry_raw)(struct sk_buff *msg, struct ib_qp *ibqp); 2614 int (*fill_res_cm_id_entry)(struct sk_buff *msg, struct rdma_cm_id *id); 2615 int (*fill_res_srq_entry)(struct sk_buff *msg, struct ib_srq *ib_srq); 2616 int (*fill_res_srq_entry_raw)(struct sk_buff *msg, struct ib_srq *ib_srq); 2617 2618 /* Device lifecycle callbacks */ 2619 /* 2620 * Called after the device becomes registered, before clients are 2621 * attached 2622 */ 2623 int (*enable_driver)(struct ib_device *dev); 2624 /* 2625 * This is called as part of ib_dealloc_device(). 2626 */ 2627 void (*dealloc_driver)(struct ib_device *dev); 2628 2629 /* iWarp CM callbacks */ 2630 void (*iw_add_ref)(struct ib_qp *qp); 2631 void (*iw_rem_ref)(struct ib_qp *qp); 2632 struct ib_qp *(*iw_get_qp)(struct ib_device *device, int qpn); 2633 int (*iw_connect)(struct iw_cm_id *cm_id, 2634 struct iw_cm_conn_param *conn_param); 2635 int (*iw_accept)(struct iw_cm_id *cm_id, 2636 struct iw_cm_conn_param *conn_param); 2637 int (*iw_reject)(struct iw_cm_id *cm_id, const void *pdata, 2638 u8 pdata_len); 2639 int (*iw_create_listen)(struct iw_cm_id *cm_id, int backlog); 2640 int (*iw_destroy_listen)(struct iw_cm_id *cm_id); 2641 /** 2642 * counter_bind_qp - Bind a QP to a counter. 2643 * @counter - The counter to be bound. If counter->id is zero then 2644 * the driver needs to allocate a new counter and set counter->id 2645 */ 2646 int (*counter_bind_qp)(struct rdma_counter *counter, struct ib_qp *qp); 2647 /** 2648 * counter_unbind_qp - Unbind the qp from the dynamically-allocated 2649 * counter and bind it onto the default one 2650 */ 2651 int (*counter_unbind_qp)(struct ib_qp *qp); 2652 /** 2653 * counter_dealloc -De-allocate the hw counter 2654 */ 2655 int (*counter_dealloc)(struct rdma_counter *counter); 2656 /** 2657 * counter_alloc_stats - Allocate a struct rdma_hw_stats and fill in 2658 * the driver initialized data. 2659 */ 2660 struct rdma_hw_stats *(*counter_alloc_stats)( 2661 struct rdma_counter *counter); 2662 /** 2663 * counter_update_stats - Query the stats value of this counter 2664 */ 2665 int (*counter_update_stats)(struct rdma_counter *counter); 2666 2667 /** 2668 * Allows rdma drivers to add their own restrack attributes 2669 * dumped via 'rdma stat' iproute2 command. 2670 */ 2671 int (*fill_stat_mr_entry)(struct sk_buff *msg, struct ib_mr *ibmr); 2672 2673 /* query driver for its ucontext properties */ 2674 int (*query_ucontext)(struct ib_ucontext *context, 2675 struct uverbs_attr_bundle *attrs); 2676 2677 /* 2678 * Provide NUMA node. This API exists for rdmavt/hfi1 only. 2679 * Everyone else relies on Linux memory management model. 2680 */ 2681 int (*get_numa_node)(struct ib_device *dev); 2682 2683 /** 2684 * add_sub_dev - Add a sub IB device 2685 */ 2686 struct ib_device *(*add_sub_dev)(struct ib_device *parent, 2687 enum rdma_nl_dev_type type, 2688 const char *name); 2689 2690 /** 2691 * del_sub_dev - Delete a sub IB device 2692 */ 2693 void (*del_sub_dev)(struct ib_device *sub_dev); 2694 2695 /** 2696 * ufile_cleanup - Attempt to cleanup ubojects HW resources inside 2697 * the ufile. 2698 */ 2699 void (*ufile_hw_cleanup)(struct ib_uverbs_file *ufile); 2700 2701 /** 2702 * report_port_event - Drivers need to implement this if they have 2703 * some private stuff to handle when link status changes. 2704 */ 2705 void (*report_port_event)(struct ib_device *ibdev, 2706 struct net_device *ndev, unsigned long event); 2707 2708 DECLARE_RDMA_OBJ_SIZE(ib_ah); 2709 DECLARE_RDMA_OBJ_SIZE(ib_counters); 2710 DECLARE_RDMA_OBJ_SIZE(ib_cq); 2711 DECLARE_RDMA_OBJ_SIZE(ib_mw); 2712 DECLARE_RDMA_OBJ_SIZE(ib_pd); 2713 DECLARE_RDMA_OBJ_SIZE(ib_qp); 2714 DECLARE_RDMA_OBJ_SIZE(ib_rwq_ind_table); 2715 DECLARE_RDMA_OBJ_SIZE(ib_srq); 2716 DECLARE_RDMA_OBJ_SIZE(ib_ucontext); 2717 DECLARE_RDMA_OBJ_SIZE(ib_xrcd); 2718 }; 2719 2720 struct ib_core_device { 2721 /* device must be the first element in structure until, 2722 * union of ib_core_device and device exists in ib_device. 2723 */ 2724 struct device dev; 2725 possible_net_t rdma_net; 2726 struct kobject *ports_kobj; 2727 struct list_head port_list; 2728 struct ib_device *owner; /* reach back to owner ib_device */ 2729 }; 2730 2731 struct rdma_restrack_root; 2732 struct ib_device { 2733 /* Do not access @dma_device directly from ULP nor from HW drivers. */ 2734 struct device *dma_device; 2735 struct ib_device_ops ops; 2736 char name[IB_DEVICE_NAME_MAX]; 2737 struct rcu_head rcu_head; 2738 2739 struct list_head event_handler_list; 2740 /* Protects event_handler_list */ 2741 struct rw_semaphore event_handler_rwsem; 2742 2743 /* Protects QP's event_handler calls and open_qp list */ 2744 spinlock_t qp_open_list_lock; 2745 2746 struct rw_semaphore client_data_rwsem; 2747 struct xarray client_data; 2748 struct mutex unregistration_lock; 2749 2750 /* Synchronize GID, Pkey cache entries, subnet prefix, LMC */ 2751 rwlock_t cache_lock; 2752 /** 2753 * port_data is indexed by port number 2754 */ 2755 struct ib_port_data *port_data; 2756 2757 int num_comp_vectors; 2758 2759 union { 2760 struct device dev; 2761 struct ib_core_device coredev; 2762 }; 2763 2764 /* First group is for device attributes, 2765 * Second group is for driver provided attributes (optional). 2766 * Third group is for the hw_stats 2767 * It is a NULL terminated array. 2768 */ 2769 const struct attribute_group *groups[4]; 2770 2771 u64 uverbs_cmd_mask; 2772 2773 char node_desc[IB_DEVICE_NODE_DESC_MAX]; 2774 __be64 node_guid; 2775 u32 local_dma_lkey; 2776 u16 is_switch:1; 2777 /* Indicates kernel verbs support, should not be used in drivers */ 2778 u16 kverbs_provider:1; 2779 /* CQ adaptive moderation (RDMA DIM) */ 2780 u16 use_cq_dim:1; 2781 u8 node_type; 2782 u32 phys_port_cnt; 2783 struct ib_device_attr attrs; 2784 struct hw_stats_device_data *hw_stats_data; 2785 2786 #ifdef CONFIG_CGROUP_RDMA 2787 struct rdmacg_device cg_device; 2788 #endif 2789 2790 u32 index; 2791 2792 spinlock_t cq_pools_lock; 2793 struct list_head cq_pools[IB_POLL_LAST_POOL_TYPE + 1]; 2794 2795 struct rdma_restrack_root *res; 2796 2797 const struct uapi_definition *driver_def; 2798 2799 /* 2800 * Positive refcount indicates that the device is currently 2801 * registered and cannot be unregistered. 2802 */ 2803 refcount_t refcount; 2804 struct completion unreg_completion; 2805 struct work_struct unregistration_work; 2806 2807 const struct rdma_link_ops *link_ops; 2808 2809 /* Protects compat_devs xarray modifications */ 2810 struct mutex compat_devs_mutex; 2811 /* Maintains compat devices for each net namespace */ 2812 struct xarray compat_devs; 2813 2814 /* Used by iWarp CM */ 2815 char iw_ifname[IFNAMSIZ]; 2816 u32 iw_driver_flags; 2817 u32 lag_flags; 2818 2819 /* A parent device has a list of sub-devices */ 2820 struct mutex subdev_lock; 2821 struct list_head subdev_list_head; 2822 2823 /* A sub device has a type and a parent */ 2824 enum rdma_nl_dev_type type; 2825 struct ib_device *parent; 2826 struct list_head subdev_list; 2827 2828 enum rdma_nl_name_assign_type name_assign_type; 2829 }; 2830 2831 static inline void *rdma_zalloc_obj(struct ib_device *dev, size_t size, 2832 gfp_t gfp, bool is_numa_aware) 2833 { 2834 if (is_numa_aware && dev->ops.get_numa_node) 2835 return kzalloc_node(size, gfp, dev->ops.get_numa_node(dev)); 2836 2837 return kzalloc(size, gfp); 2838 } 2839 2840 struct ib_client_nl_info; 2841 struct ib_client { 2842 const char *name; 2843 int (*add)(struct ib_device *ibdev); 2844 void (*remove)(struct ib_device *, void *client_data); 2845 void (*rename)(struct ib_device *dev, void *client_data); 2846 int (*get_nl_info)(struct ib_device *ibdev, void *client_data, 2847 struct ib_client_nl_info *res); 2848 int (*get_global_nl_info)(struct ib_client_nl_info *res); 2849 2850 /* Returns the net_dev belonging to this ib_client and matching the 2851 * given parameters. 2852 * @dev: An RDMA device that the net_dev use for communication. 2853 * @port: A physical port number on the RDMA device. 2854 * @pkey: P_Key that the net_dev uses if applicable. 2855 * @gid: A GID that the net_dev uses to communicate. 2856 * @addr: An IP address the net_dev is configured with. 2857 * @client_data: The device's client data set by ib_set_client_data(). 2858 * 2859 * An ib_client that implements a net_dev on top of RDMA devices 2860 * (such as IP over IB) should implement this callback, allowing the 2861 * rdma_cm module to find the right net_dev for a given request. 2862 * 2863 * The caller is responsible for calling dev_put on the returned 2864 * netdev. */ 2865 struct net_device *(*get_net_dev_by_params)( 2866 struct ib_device *dev, 2867 u32 port, 2868 u16 pkey, 2869 const union ib_gid *gid, 2870 const struct sockaddr *addr, 2871 void *client_data); 2872 2873 refcount_t uses; 2874 struct completion uses_zero; 2875 u32 client_id; 2876 2877 /* kverbs are not required by the client */ 2878 u8 no_kverbs_req:1; 2879 }; 2880 2881 /* 2882 * IB block DMA iterator 2883 * 2884 * Iterates the DMA-mapped SGL in contiguous memory blocks aligned 2885 * to a HW supported page size. 2886 */ 2887 struct ib_block_iter { 2888 /* internal states */ 2889 struct scatterlist *__sg; /* sg holding the current aligned block */ 2890 dma_addr_t __dma_addr; /* unaligned DMA address of this block */ 2891 size_t __sg_numblocks; /* ib_umem_num_dma_blocks() */ 2892 unsigned int __sg_nents; /* number of SG entries */ 2893 unsigned int __sg_advance; /* number of bytes to advance in sg in next step */ 2894 unsigned int __pg_bit; /* alignment of current block */ 2895 }; 2896 2897 struct ib_device *_ib_alloc_device(size_t size); 2898 #define ib_alloc_device(drv_struct, member) \ 2899 container_of(_ib_alloc_device(sizeof(struct drv_struct) + \ 2900 BUILD_BUG_ON_ZERO(offsetof( \ 2901 struct drv_struct, member))), \ 2902 struct drv_struct, member) 2903 2904 void ib_dealloc_device(struct ib_device *device); 2905 2906 void ib_get_device_fw_str(struct ib_device *device, char *str); 2907 2908 int ib_register_device(struct ib_device *device, const char *name, 2909 struct device *dma_device); 2910 void ib_unregister_device(struct ib_device *device); 2911 void ib_unregister_driver(enum rdma_driver_id driver_id); 2912 void ib_unregister_device_and_put(struct ib_device *device); 2913 void ib_unregister_device_queued(struct ib_device *ib_dev); 2914 2915 int ib_register_client (struct ib_client *client); 2916 void ib_unregister_client(struct ib_client *client); 2917 2918 void __rdma_block_iter_start(struct ib_block_iter *biter, 2919 struct scatterlist *sglist, 2920 unsigned int nents, 2921 unsigned long pgsz); 2922 bool __rdma_block_iter_next(struct ib_block_iter *biter); 2923 2924 /** 2925 * rdma_block_iter_dma_address - get the aligned dma address of the current 2926 * block held by the block iterator. 2927 * @biter: block iterator holding the memory block 2928 */ 2929 static inline dma_addr_t 2930 rdma_block_iter_dma_address(struct ib_block_iter *biter) 2931 { 2932 return biter->__dma_addr & ~(BIT_ULL(biter->__pg_bit) - 1); 2933 } 2934 2935 /** 2936 * rdma_for_each_block - iterate over contiguous memory blocks of the sg list 2937 * @sglist: sglist to iterate over 2938 * @biter: block iterator holding the memory block 2939 * @nents: maximum number of sg entries to iterate over 2940 * @pgsz: best HW supported page size to use 2941 * 2942 * Callers may use rdma_block_iter_dma_address() to get each 2943 * blocks aligned DMA address. 2944 */ 2945 #define rdma_for_each_block(sglist, biter, nents, pgsz) \ 2946 for (__rdma_block_iter_start(biter, sglist, nents, \ 2947 pgsz); \ 2948 __rdma_block_iter_next(biter);) 2949 2950 /** 2951 * ib_get_client_data - Get IB client context 2952 * @device:Device to get context for 2953 * @client:Client to get context for 2954 * 2955 * ib_get_client_data() returns the client context data set with 2956 * ib_set_client_data(). This can only be called while the client is 2957 * registered to the device, once the ib_client remove() callback returns this 2958 * cannot be called. 2959 */ 2960 static inline void *ib_get_client_data(struct ib_device *device, 2961 struct ib_client *client) 2962 { 2963 return xa_load(&device->client_data, client->client_id); 2964 } 2965 void ib_set_client_data(struct ib_device *device, struct ib_client *client, 2966 void *data); 2967 void ib_set_device_ops(struct ib_device *device, 2968 const struct ib_device_ops *ops); 2969 2970 int rdma_user_mmap_io(struct ib_ucontext *ucontext, struct vm_area_struct *vma, 2971 unsigned long pfn, unsigned long size, pgprot_t prot, 2972 struct rdma_user_mmap_entry *entry); 2973 int rdma_user_mmap_entry_insert(struct ib_ucontext *ucontext, 2974 struct rdma_user_mmap_entry *entry, 2975 size_t length); 2976 int rdma_user_mmap_entry_insert_range(struct ib_ucontext *ucontext, 2977 struct rdma_user_mmap_entry *entry, 2978 size_t length, u32 min_pgoff, 2979 u32 max_pgoff); 2980 2981 #if IS_ENABLED(CONFIG_INFINIBAND_USER_ACCESS) 2982 void rdma_user_mmap_disassociate(struct ib_device *device); 2983 #else 2984 static inline void rdma_user_mmap_disassociate(struct ib_device *device) 2985 { 2986 } 2987 #endif 2988 2989 static inline int 2990 rdma_user_mmap_entry_insert_exact(struct ib_ucontext *ucontext, 2991 struct rdma_user_mmap_entry *entry, 2992 size_t length, u32 pgoff) 2993 { 2994 return rdma_user_mmap_entry_insert_range(ucontext, entry, length, pgoff, 2995 pgoff); 2996 } 2997 2998 struct rdma_user_mmap_entry * 2999 rdma_user_mmap_entry_get_pgoff(struct ib_ucontext *ucontext, 3000 unsigned long pgoff); 3001 struct rdma_user_mmap_entry * 3002 rdma_user_mmap_entry_get(struct ib_ucontext *ucontext, 3003 struct vm_area_struct *vma); 3004 void rdma_user_mmap_entry_put(struct rdma_user_mmap_entry *entry); 3005 3006 void rdma_user_mmap_entry_remove(struct rdma_user_mmap_entry *entry); 3007 3008 static inline int ib_copy_from_udata(void *dest, struct ib_udata *udata, size_t len) 3009 { 3010 return copy_from_user(dest, udata->inbuf, len) ? -EFAULT : 0; 3011 } 3012 3013 static inline int ib_copy_to_udata(struct ib_udata *udata, void *src, size_t len) 3014 { 3015 return copy_to_user(udata->outbuf, src, len) ? -EFAULT : 0; 3016 } 3017 3018 static inline bool ib_is_buffer_cleared(const void __user *p, 3019 size_t len) 3020 { 3021 bool ret; 3022 u8 *buf; 3023 3024 if (len > USHRT_MAX) 3025 return false; 3026 3027 buf = memdup_user(p, len); 3028 if (IS_ERR(buf)) 3029 return false; 3030 3031 ret = !memchr_inv(buf, 0, len); 3032 kfree(buf); 3033 return ret; 3034 } 3035 3036 static inline bool ib_is_udata_cleared(struct ib_udata *udata, 3037 size_t offset, 3038 size_t len) 3039 { 3040 return ib_is_buffer_cleared(udata->inbuf + offset, len); 3041 } 3042 3043 /** 3044 * ib_modify_qp_is_ok - Check that the supplied attribute mask 3045 * contains all required attributes and no attributes not allowed for 3046 * the given QP state transition. 3047 * @cur_state: Current QP state 3048 * @next_state: Next QP state 3049 * @type: QP type 3050 * @mask: Mask of supplied QP attributes 3051 * 3052 * This function is a helper function that a low-level driver's 3053 * modify_qp method can use to validate the consumer's input. It 3054 * checks that cur_state and next_state are valid QP states, that a 3055 * transition from cur_state to next_state is allowed by the IB spec, 3056 * and that the attribute mask supplied is allowed for the transition. 3057 */ 3058 bool ib_modify_qp_is_ok(enum ib_qp_state cur_state, enum ib_qp_state next_state, 3059 enum ib_qp_type type, enum ib_qp_attr_mask mask); 3060 3061 void ib_register_event_handler(struct ib_event_handler *event_handler); 3062 void ib_unregister_event_handler(struct ib_event_handler *event_handler); 3063 void ib_dispatch_event(const struct ib_event *event); 3064 3065 int ib_query_port(struct ib_device *device, 3066 u32 port_num, struct ib_port_attr *port_attr); 3067 3068 enum rdma_link_layer rdma_port_get_link_layer(struct ib_device *device, 3069 u32 port_num); 3070 3071 /** 3072 * rdma_cap_ib_switch - Check if the device is IB switch 3073 * @device: Device to check 3074 * 3075 * Device driver is responsible for setting is_switch bit on 3076 * in ib_device structure at init time. 3077 * 3078 * Return: true if the device is IB switch. 3079 */ 3080 static inline bool rdma_cap_ib_switch(const struct ib_device *device) 3081 { 3082 return device->is_switch; 3083 } 3084 3085 /** 3086 * rdma_start_port - Return the first valid port number for the device 3087 * specified 3088 * 3089 * @device: Device to be checked 3090 * 3091 * Return start port number 3092 */ 3093 static inline u32 rdma_start_port(const struct ib_device *device) 3094 { 3095 return rdma_cap_ib_switch(device) ? 0 : 1; 3096 } 3097 3098 /** 3099 * rdma_for_each_port - Iterate over all valid port numbers of the IB device 3100 * @device - The struct ib_device * to iterate over 3101 * @iter - The unsigned int to store the port number 3102 */ 3103 #define rdma_for_each_port(device, iter) \ 3104 for (iter = rdma_start_port(device + \ 3105 BUILD_BUG_ON_ZERO(!__same_type(u32, \ 3106 iter))); \ 3107 iter <= rdma_end_port(device); iter++) 3108 3109 /** 3110 * rdma_end_port - Return the last valid port number for the device 3111 * specified 3112 * 3113 * @device: Device to be checked 3114 * 3115 * Return last port number 3116 */ 3117 static inline u32 rdma_end_port(const struct ib_device *device) 3118 { 3119 return rdma_cap_ib_switch(device) ? 0 : device->phys_port_cnt; 3120 } 3121 3122 static inline int rdma_is_port_valid(const struct ib_device *device, 3123 unsigned int port) 3124 { 3125 return (port >= rdma_start_port(device) && 3126 port <= rdma_end_port(device)); 3127 } 3128 3129 static inline bool rdma_is_grh_required(const struct ib_device *device, 3130 u32 port_num) 3131 { 3132 return device->port_data[port_num].immutable.core_cap_flags & 3133 RDMA_CORE_PORT_IB_GRH_REQUIRED; 3134 } 3135 3136 static inline bool rdma_protocol_ib(const struct ib_device *device, 3137 u32 port_num) 3138 { 3139 return device->port_data[port_num].immutable.core_cap_flags & 3140 RDMA_CORE_CAP_PROT_IB; 3141 } 3142 3143 static inline bool rdma_protocol_roce(const struct ib_device *device, 3144 u32 port_num) 3145 { 3146 return device->port_data[port_num].immutable.core_cap_flags & 3147 (RDMA_CORE_CAP_PROT_ROCE | RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP); 3148 } 3149 3150 static inline bool rdma_protocol_roce_udp_encap(const struct ib_device *device, 3151 u32 port_num) 3152 { 3153 return device->port_data[port_num].immutable.core_cap_flags & 3154 RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP; 3155 } 3156 3157 static inline bool rdma_protocol_roce_eth_encap(const struct ib_device *device, 3158 u32 port_num) 3159 { 3160 return device->port_data[port_num].immutable.core_cap_flags & 3161 RDMA_CORE_CAP_PROT_ROCE; 3162 } 3163 3164 static inline bool rdma_protocol_iwarp(const struct ib_device *device, 3165 u32 port_num) 3166 { 3167 return device->port_data[port_num].immutable.core_cap_flags & 3168 RDMA_CORE_CAP_PROT_IWARP; 3169 } 3170 3171 static inline bool rdma_ib_or_roce(const struct ib_device *device, 3172 u32 port_num) 3173 { 3174 return rdma_protocol_ib(device, port_num) || 3175 rdma_protocol_roce(device, port_num); 3176 } 3177 3178 static inline bool rdma_protocol_raw_packet(const struct ib_device *device, 3179 u32 port_num) 3180 { 3181 return device->port_data[port_num].immutable.core_cap_flags & 3182 RDMA_CORE_CAP_PROT_RAW_PACKET; 3183 } 3184 3185 static inline bool rdma_protocol_usnic(const struct ib_device *device, 3186 u32 port_num) 3187 { 3188 return device->port_data[port_num].immutable.core_cap_flags & 3189 RDMA_CORE_CAP_PROT_USNIC; 3190 } 3191 3192 /** 3193 * rdma_cap_ib_mad - Check if the port of a device supports Infiniband 3194 * Management Datagrams. 3195 * @device: Device to check 3196 * @port_num: Port number to check 3197 * 3198 * Management Datagrams (MAD) are a required part of the InfiniBand 3199 * specification and are supported on all InfiniBand devices. A slightly 3200 * extended version are also supported on OPA interfaces. 3201 * 3202 * Return: true if the port supports sending/receiving of MAD packets. 3203 */ 3204 static inline bool rdma_cap_ib_mad(const struct ib_device *device, u32 port_num) 3205 { 3206 return device->port_data[port_num].immutable.core_cap_flags & 3207 RDMA_CORE_CAP_IB_MAD; 3208 } 3209 3210 /** 3211 * rdma_cap_opa_mad - Check if the port of device provides support for OPA 3212 * Management Datagrams. 3213 * @device: Device to check 3214 * @port_num: Port number to check 3215 * 3216 * Intel OmniPath devices extend and/or replace the InfiniBand Management 3217 * datagrams with their own versions. These OPA MADs share many but not all of 3218 * the characteristics of InfiniBand MADs. 3219 * 3220 * OPA MADs differ in the following ways: 3221 * 3222 * 1) MADs are variable size up to 2K 3223 * IBTA defined MADs remain fixed at 256 bytes 3224 * 2) OPA SMPs must carry valid PKeys 3225 * 3) OPA SMP packets are a different format 3226 * 3227 * Return: true if the port supports OPA MAD packet formats. 3228 */ 3229 static inline bool rdma_cap_opa_mad(struct ib_device *device, u32 port_num) 3230 { 3231 return device->port_data[port_num].immutable.core_cap_flags & 3232 RDMA_CORE_CAP_OPA_MAD; 3233 } 3234 3235 /** 3236 * rdma_cap_ib_smi - Check if the port of a device provides an Infiniband 3237 * Subnet Management Agent (SMA) on the Subnet Management Interface (SMI). 3238 * @device: Device to check 3239 * @port_num: Port number to check 3240 * 3241 * Each InfiniBand node is required to provide a Subnet Management Agent 3242 * that the subnet manager can access. Prior to the fabric being fully 3243 * configured by the subnet manager, the SMA is accessed via a well known 3244 * interface called the Subnet Management Interface (SMI). This interface 3245 * uses directed route packets to communicate with the SM to get around the 3246 * chicken and egg problem of the SM needing to know what's on the fabric 3247 * in order to configure the fabric, and needing to configure the fabric in 3248 * order to send packets to the devices on the fabric. These directed 3249 * route packets do not need the fabric fully configured in order to reach 3250 * their destination. The SMI is the only method allowed to send 3251 * directed route packets on an InfiniBand fabric. 3252 * 3253 * Return: true if the port provides an SMI. 3254 */ 3255 static inline bool rdma_cap_ib_smi(const struct ib_device *device, u32 port_num) 3256 { 3257 return device->port_data[port_num].immutable.core_cap_flags & 3258 RDMA_CORE_CAP_IB_SMI; 3259 } 3260 3261 /** 3262 * rdma_cap_ib_cm - Check if the port of device has the capability Infiniband 3263 * Communication Manager. 3264 * @device: Device to check 3265 * @port_num: Port number to check 3266 * 3267 * The InfiniBand Communication Manager is one of many pre-defined General 3268 * Service Agents (GSA) that are accessed via the General Service 3269 * Interface (GSI). It's role is to facilitate establishment of connections 3270 * between nodes as well as other management related tasks for established 3271 * connections. 3272 * 3273 * Return: true if the port supports an IB CM (this does not guarantee that 3274 * a CM is actually running however). 3275 */ 3276 static inline bool rdma_cap_ib_cm(const struct ib_device *device, u32 port_num) 3277 { 3278 return device->port_data[port_num].immutable.core_cap_flags & 3279 RDMA_CORE_CAP_IB_CM; 3280 } 3281 3282 /** 3283 * rdma_cap_iw_cm - Check if the port of device has the capability IWARP 3284 * Communication Manager. 3285 * @device: Device to check 3286 * @port_num: Port number to check 3287 * 3288 * Similar to above, but specific to iWARP connections which have a different 3289 * managment protocol than InfiniBand. 3290 * 3291 * Return: true if the port supports an iWARP CM (this does not guarantee that 3292 * a CM is actually running however). 3293 */ 3294 static inline bool rdma_cap_iw_cm(const struct ib_device *device, u32 port_num) 3295 { 3296 return device->port_data[port_num].immutable.core_cap_flags & 3297 RDMA_CORE_CAP_IW_CM; 3298 } 3299 3300 /** 3301 * rdma_cap_ib_sa - Check if the port of device has the capability Infiniband 3302 * Subnet Administration. 3303 * @device: Device to check 3304 * @port_num: Port number to check 3305 * 3306 * An InfiniBand Subnet Administration (SA) service is a pre-defined General 3307 * Service Agent (GSA) provided by the Subnet Manager (SM). On InfiniBand 3308 * fabrics, devices should resolve routes to other hosts by contacting the 3309 * SA to query the proper route. 3310 * 3311 * Return: true if the port should act as a client to the fabric Subnet 3312 * Administration interface. This does not imply that the SA service is 3313 * running locally. 3314 */ 3315 static inline bool rdma_cap_ib_sa(const struct ib_device *device, u32 port_num) 3316 { 3317 return device->port_data[port_num].immutable.core_cap_flags & 3318 RDMA_CORE_CAP_IB_SA; 3319 } 3320 3321 /** 3322 * rdma_cap_ib_mcast - Check if the port of device has the capability Infiniband 3323 * Multicast. 3324 * @device: Device to check 3325 * @port_num: Port number to check 3326 * 3327 * InfiniBand multicast registration is more complex than normal IPv4 or 3328 * IPv6 multicast registration. Each Host Channel Adapter must register 3329 * with the Subnet Manager when it wishes to join a multicast group. It 3330 * should do so only once regardless of how many queue pairs it subscribes 3331 * to this group. And it should leave the group only after all queue pairs 3332 * attached to the group have been detached. 3333 * 3334 * Return: true if the port must undertake the additional adminstrative 3335 * overhead of registering/unregistering with the SM and tracking of the 3336 * total number of queue pairs attached to the multicast group. 3337 */ 3338 static inline bool rdma_cap_ib_mcast(const struct ib_device *device, 3339 u32 port_num) 3340 { 3341 return rdma_cap_ib_sa(device, port_num); 3342 } 3343 3344 /** 3345 * rdma_cap_af_ib - Check if the port of device has the capability 3346 * Native Infiniband Address. 3347 * @device: Device to check 3348 * @port_num: Port number to check 3349 * 3350 * InfiniBand addressing uses a port's GUID + Subnet Prefix to make a default 3351 * GID. RoCE uses a different mechanism, but still generates a GID via 3352 * a prescribed mechanism and port specific data. 3353 * 3354 * Return: true if the port uses a GID address to identify devices on the 3355 * network. 3356 */ 3357 static inline bool rdma_cap_af_ib(const struct ib_device *device, u32 port_num) 3358 { 3359 return device->port_data[port_num].immutable.core_cap_flags & 3360 RDMA_CORE_CAP_AF_IB; 3361 } 3362 3363 /** 3364 * rdma_cap_eth_ah - Check if the port of device has the capability 3365 * Ethernet Address Handle. 3366 * @device: Device to check 3367 * @port_num: Port number to check 3368 * 3369 * RoCE is InfiniBand over Ethernet, and it uses a well defined technique 3370 * to fabricate GIDs over Ethernet/IP specific addresses native to the 3371 * port. Normally, packet headers are generated by the sending host 3372 * adapter, but when sending connectionless datagrams, we must manually 3373 * inject the proper headers for the fabric we are communicating over. 3374 * 3375 * Return: true if we are running as a RoCE port and must force the 3376 * addition of a Global Route Header built from our Ethernet Address 3377 * Handle into our header list for connectionless packets. 3378 */ 3379 static inline bool rdma_cap_eth_ah(const struct ib_device *device, u32 port_num) 3380 { 3381 return device->port_data[port_num].immutable.core_cap_flags & 3382 RDMA_CORE_CAP_ETH_AH; 3383 } 3384 3385 /** 3386 * rdma_cap_opa_ah - Check if the port of device supports 3387 * OPA Address handles 3388 * @device: Device to check 3389 * @port_num: Port number to check 3390 * 3391 * Return: true if we are running on an OPA device which supports 3392 * the extended OPA addressing. 3393 */ 3394 static inline bool rdma_cap_opa_ah(struct ib_device *device, u32 port_num) 3395 { 3396 return (device->port_data[port_num].immutable.core_cap_flags & 3397 RDMA_CORE_CAP_OPA_AH) == RDMA_CORE_CAP_OPA_AH; 3398 } 3399 3400 /** 3401 * rdma_max_mad_size - Return the max MAD size required by this RDMA Port. 3402 * 3403 * @device: Device 3404 * @port_num: Port number 3405 * 3406 * This MAD size includes the MAD headers and MAD payload. No other headers 3407 * are included. 3408 * 3409 * Return the max MAD size required by the Port. Will return 0 if the port 3410 * does not support MADs 3411 */ 3412 static inline size_t rdma_max_mad_size(const struct ib_device *device, 3413 u32 port_num) 3414 { 3415 return device->port_data[port_num].immutable.max_mad_size; 3416 } 3417 3418 /** 3419 * rdma_cap_roce_gid_table - Check if the port of device uses roce_gid_table 3420 * @device: Device to check 3421 * @port_num: Port number to check 3422 * 3423 * RoCE GID table mechanism manages the various GIDs for a device. 3424 * 3425 * NOTE: if allocating the port's GID table has failed, this call will still 3426 * return true, but any RoCE GID table API will fail. 3427 * 3428 * Return: true if the port uses RoCE GID table mechanism in order to manage 3429 * its GIDs. 3430 */ 3431 static inline bool rdma_cap_roce_gid_table(const struct ib_device *device, 3432 u32 port_num) 3433 { 3434 return rdma_protocol_roce(device, port_num) && 3435 device->ops.add_gid && device->ops.del_gid; 3436 } 3437 3438 /* 3439 * Check if the device supports READ W/ INVALIDATE. 3440 */ 3441 static inline bool rdma_cap_read_inv(struct ib_device *dev, u32 port_num) 3442 { 3443 /* 3444 * iWarp drivers must support READ W/ INVALIDATE. No other protocol 3445 * has support for it yet. 3446 */ 3447 return rdma_protocol_iwarp(dev, port_num); 3448 } 3449 3450 /** 3451 * rdma_core_cap_opa_port - Return whether the RDMA Port is OPA or not. 3452 * @device: Device 3453 * @port_num: 1 based Port number 3454 * 3455 * Return true if port is an Intel OPA port , false if not 3456 */ 3457 static inline bool rdma_core_cap_opa_port(struct ib_device *device, 3458 u32 port_num) 3459 { 3460 return (device->port_data[port_num].immutable.core_cap_flags & 3461 RDMA_CORE_PORT_INTEL_OPA) == RDMA_CORE_PORT_INTEL_OPA; 3462 } 3463 3464 /** 3465 * rdma_mtu_enum_to_int - Return the mtu of the port as an integer value. 3466 * @device: Device 3467 * @port_num: Port number 3468 * @mtu: enum value of MTU 3469 * 3470 * Return the MTU size supported by the port as an integer value. Will return 3471 * -1 if enum value of mtu is not supported. 3472 */ 3473 static inline int rdma_mtu_enum_to_int(struct ib_device *device, u32 port, 3474 int mtu) 3475 { 3476 if (rdma_core_cap_opa_port(device, port)) 3477 return opa_mtu_enum_to_int((enum opa_mtu)mtu); 3478 else 3479 return ib_mtu_enum_to_int((enum ib_mtu)mtu); 3480 } 3481 3482 /** 3483 * rdma_mtu_from_attr - Return the mtu of the port from the port attribute. 3484 * @device: Device 3485 * @port_num: Port number 3486 * @attr: port attribute 3487 * 3488 * Return the MTU size supported by the port as an integer value. 3489 */ 3490 static inline int rdma_mtu_from_attr(struct ib_device *device, u32 port, 3491 struct ib_port_attr *attr) 3492 { 3493 if (rdma_core_cap_opa_port(device, port)) 3494 return attr->phys_mtu; 3495 else 3496 return ib_mtu_enum_to_int(attr->max_mtu); 3497 } 3498 3499 int ib_set_vf_link_state(struct ib_device *device, int vf, u32 port, 3500 int state); 3501 int ib_get_vf_config(struct ib_device *device, int vf, u32 port, 3502 struct ifla_vf_info *info); 3503 int ib_get_vf_stats(struct ib_device *device, int vf, u32 port, 3504 struct ifla_vf_stats *stats); 3505 int ib_get_vf_guid(struct ib_device *device, int vf, u32 port, 3506 struct ifla_vf_guid *node_guid, 3507 struct ifla_vf_guid *port_guid); 3508 int ib_set_vf_guid(struct ib_device *device, int vf, u32 port, u64 guid, 3509 int type); 3510 3511 int ib_query_pkey(struct ib_device *device, 3512 u32 port_num, u16 index, u16 *pkey); 3513 3514 int ib_modify_device(struct ib_device *device, 3515 int device_modify_mask, 3516 struct ib_device_modify *device_modify); 3517 3518 int ib_modify_port(struct ib_device *device, 3519 u32 port_num, int port_modify_mask, 3520 struct ib_port_modify *port_modify); 3521 3522 int ib_find_gid(struct ib_device *device, union ib_gid *gid, 3523 u32 *port_num, u16 *index); 3524 3525 int ib_find_pkey(struct ib_device *device, 3526 u32 port_num, u16 pkey, u16 *index); 3527 3528 enum ib_pd_flags { 3529 /* 3530 * Create a memory registration for all memory in the system and place 3531 * the rkey for it into pd->unsafe_global_rkey. This can be used by 3532 * ULPs to avoid the overhead of dynamic MRs. 3533 * 3534 * This flag is generally considered unsafe and must only be used in 3535 * extremly trusted environments. Every use of it will log a warning 3536 * in the kernel log. 3537 */ 3538 IB_PD_UNSAFE_GLOBAL_RKEY = 0x01, 3539 }; 3540 3541 struct ib_pd *__ib_alloc_pd(struct ib_device *device, unsigned int flags, 3542 const char *caller); 3543 3544 /** 3545 * ib_alloc_pd - Allocates an unused protection domain. 3546 * @device: The device on which to allocate the protection domain. 3547 * @flags: protection domain flags 3548 * 3549 * A protection domain object provides an association between QPs, shared 3550 * receive queues, address handles, memory regions, and memory windows. 3551 * 3552 * Every PD has a local_dma_lkey which can be used as the lkey value for local 3553 * memory operations. 3554 */ 3555 #define ib_alloc_pd(device, flags) \ 3556 __ib_alloc_pd((device), (flags), KBUILD_MODNAME) 3557 3558 int ib_dealloc_pd_user(struct ib_pd *pd, struct ib_udata *udata); 3559 3560 /** 3561 * ib_dealloc_pd - Deallocate kernel PD 3562 * @pd: The protection domain 3563 * 3564 * NOTE: for user PD use ib_dealloc_pd_user with valid udata! 3565 */ 3566 static inline void ib_dealloc_pd(struct ib_pd *pd) 3567 { 3568 int ret = ib_dealloc_pd_user(pd, NULL); 3569 3570 WARN_ONCE(ret, "Destroy of kernel PD shouldn't fail"); 3571 } 3572 3573 enum rdma_create_ah_flags { 3574 /* In a sleepable context */ 3575 RDMA_CREATE_AH_SLEEPABLE = BIT(0), 3576 }; 3577 3578 /** 3579 * rdma_create_ah - Creates an address handle for the given address vector. 3580 * @pd: The protection domain associated with the address handle. 3581 * @ah_attr: The attributes of the address vector. 3582 * @flags: Create address handle flags (see enum rdma_create_ah_flags). 3583 * 3584 * The address handle is used to reference a local or global destination 3585 * in all UD QP post sends. 3586 */ 3587 struct ib_ah *rdma_create_ah(struct ib_pd *pd, struct rdma_ah_attr *ah_attr, 3588 u32 flags); 3589 3590 /** 3591 * rdma_create_user_ah - Creates an address handle for the given address vector. 3592 * It resolves destination mac address for ah attribute of RoCE type. 3593 * @pd: The protection domain associated with the address handle. 3594 * @ah_attr: The attributes of the address vector. 3595 * @udata: pointer to user's input output buffer information need by 3596 * provider driver. 3597 * 3598 * It returns 0 on success and returns appropriate error code on error. 3599 * The address handle is used to reference a local or global destination 3600 * in all UD QP post sends. 3601 */ 3602 struct ib_ah *rdma_create_user_ah(struct ib_pd *pd, 3603 struct rdma_ah_attr *ah_attr, 3604 struct ib_udata *udata); 3605 /** 3606 * ib_get_gids_from_rdma_hdr - Get sgid and dgid from GRH or IPv4 header 3607 * work completion. 3608 * @hdr: the L3 header to parse 3609 * @net_type: type of header to parse 3610 * @sgid: place to store source gid 3611 * @dgid: place to store destination gid 3612 */ 3613 int ib_get_gids_from_rdma_hdr(const union rdma_network_hdr *hdr, 3614 enum rdma_network_type net_type, 3615 union ib_gid *sgid, union ib_gid *dgid); 3616 3617 /** 3618 * ib_get_rdma_header_version - Get the header version 3619 * @hdr: the L3 header to parse 3620 */ 3621 int ib_get_rdma_header_version(const union rdma_network_hdr *hdr); 3622 3623 /** 3624 * ib_init_ah_attr_from_wc - Initializes address handle attributes from a 3625 * work completion. 3626 * @device: Device on which the received message arrived. 3627 * @port_num: Port on which the received message arrived. 3628 * @wc: Work completion associated with the received message. 3629 * @grh: References the received global route header. This parameter is 3630 * ignored unless the work completion indicates that the GRH is valid. 3631 * @ah_attr: Returned attributes that can be used when creating an address 3632 * handle for replying to the message. 3633 * When ib_init_ah_attr_from_wc() returns success, 3634 * (a) for IB link layer it optionally contains a reference to SGID attribute 3635 * when GRH is present for IB link layer. 3636 * (b) for RoCE link layer it contains a reference to SGID attribute. 3637 * User must invoke rdma_cleanup_ah_attr_gid_attr() to release reference to SGID 3638 * attributes which are initialized using ib_init_ah_attr_from_wc(). 3639 * 3640 */ 3641 int ib_init_ah_attr_from_wc(struct ib_device *device, u32 port_num, 3642 const struct ib_wc *wc, const struct ib_grh *grh, 3643 struct rdma_ah_attr *ah_attr); 3644 3645 /** 3646 * ib_create_ah_from_wc - Creates an address handle associated with the 3647 * sender of the specified work completion. 3648 * @pd: The protection domain associated with the address handle. 3649 * @wc: Work completion information associated with a received message. 3650 * @grh: References the received global route header. This parameter is 3651 * ignored unless the work completion indicates that the GRH is valid. 3652 * @port_num: The outbound port number to associate with the address. 3653 * 3654 * The address handle is used to reference a local or global destination 3655 * in all UD QP post sends. 3656 */ 3657 struct ib_ah *ib_create_ah_from_wc(struct ib_pd *pd, const struct ib_wc *wc, 3658 const struct ib_grh *grh, u32 port_num); 3659 3660 /** 3661 * rdma_modify_ah - Modifies the address vector associated with an address 3662 * handle. 3663 * @ah: The address handle to modify. 3664 * @ah_attr: The new address vector attributes to associate with the 3665 * address handle. 3666 */ 3667 int rdma_modify_ah(struct ib_ah *ah, struct rdma_ah_attr *ah_attr); 3668 3669 /** 3670 * rdma_query_ah - Queries the address vector associated with an address 3671 * handle. 3672 * @ah: The address handle to query. 3673 * @ah_attr: The address vector attributes associated with the address 3674 * handle. 3675 */ 3676 int rdma_query_ah(struct ib_ah *ah, struct rdma_ah_attr *ah_attr); 3677 3678 enum rdma_destroy_ah_flags { 3679 /* In a sleepable context */ 3680 RDMA_DESTROY_AH_SLEEPABLE = BIT(0), 3681 }; 3682 3683 /** 3684 * rdma_destroy_ah_user - Destroys an address handle. 3685 * @ah: The address handle to destroy. 3686 * @flags: Destroy address handle flags (see enum rdma_destroy_ah_flags). 3687 * @udata: Valid user data or NULL for kernel objects 3688 */ 3689 int rdma_destroy_ah_user(struct ib_ah *ah, u32 flags, struct ib_udata *udata); 3690 3691 /** 3692 * rdma_destroy_ah - Destroys an kernel address handle. 3693 * @ah: The address handle to destroy. 3694 * @flags: Destroy address handle flags (see enum rdma_destroy_ah_flags). 3695 * 3696 * NOTE: for user ah use rdma_destroy_ah_user with valid udata! 3697 */ 3698 static inline void rdma_destroy_ah(struct ib_ah *ah, u32 flags) 3699 { 3700 int ret = rdma_destroy_ah_user(ah, flags, NULL); 3701 3702 WARN_ONCE(ret, "Destroy of kernel AH shouldn't fail"); 3703 } 3704 3705 struct ib_srq *ib_create_srq_user(struct ib_pd *pd, 3706 struct ib_srq_init_attr *srq_init_attr, 3707 struct ib_usrq_object *uobject, 3708 struct ib_udata *udata); 3709 static inline struct ib_srq * 3710 ib_create_srq(struct ib_pd *pd, struct ib_srq_init_attr *srq_init_attr) 3711 { 3712 if (!pd->device->ops.create_srq) 3713 return ERR_PTR(-EOPNOTSUPP); 3714 3715 return ib_create_srq_user(pd, srq_init_attr, NULL, NULL); 3716 } 3717 3718 /** 3719 * ib_modify_srq - Modifies the attributes for the specified SRQ. 3720 * @srq: The SRQ to modify. 3721 * @srq_attr: On input, specifies the SRQ attributes to modify. On output, 3722 * the current values of selected SRQ attributes are returned. 3723 * @srq_attr_mask: A bit-mask used to specify which attributes of the SRQ 3724 * are being modified. 3725 * 3726 * The mask may contain IB_SRQ_MAX_WR to resize the SRQ and/or 3727 * IB_SRQ_LIMIT to set the SRQ's limit and request notification when 3728 * the number of receives queued drops below the limit. 3729 */ 3730 int ib_modify_srq(struct ib_srq *srq, 3731 struct ib_srq_attr *srq_attr, 3732 enum ib_srq_attr_mask srq_attr_mask); 3733 3734 /** 3735 * ib_query_srq - Returns the attribute list and current values for the 3736 * specified SRQ. 3737 * @srq: The SRQ to query. 3738 * @srq_attr: The attributes of the specified SRQ. 3739 */ 3740 int ib_query_srq(struct ib_srq *srq, 3741 struct ib_srq_attr *srq_attr); 3742 3743 /** 3744 * ib_destroy_srq_user - Destroys the specified SRQ. 3745 * @srq: The SRQ to destroy. 3746 * @udata: Valid user data or NULL for kernel objects 3747 */ 3748 int ib_destroy_srq_user(struct ib_srq *srq, struct ib_udata *udata); 3749 3750 /** 3751 * ib_destroy_srq - Destroys the specified kernel SRQ. 3752 * @srq: The SRQ to destroy. 3753 * 3754 * NOTE: for user srq use ib_destroy_srq_user with valid udata! 3755 */ 3756 static inline void ib_destroy_srq(struct ib_srq *srq) 3757 { 3758 int ret = ib_destroy_srq_user(srq, NULL); 3759 3760 WARN_ONCE(ret, "Destroy of kernel SRQ shouldn't fail"); 3761 } 3762 3763 /** 3764 * ib_post_srq_recv - Posts a list of work requests to the specified SRQ. 3765 * @srq: The SRQ to post the work request on. 3766 * @recv_wr: A list of work requests to post on the receive queue. 3767 * @bad_recv_wr: On an immediate failure, this parameter will reference 3768 * the work request that failed to be posted on the QP. 3769 */ 3770 static inline int ib_post_srq_recv(struct ib_srq *srq, 3771 const struct ib_recv_wr *recv_wr, 3772 const struct ib_recv_wr **bad_recv_wr) 3773 { 3774 const struct ib_recv_wr *dummy; 3775 3776 return srq->device->ops.post_srq_recv(srq, recv_wr, 3777 bad_recv_wr ? : &dummy); 3778 } 3779 3780 struct ib_qp *ib_create_qp_kernel(struct ib_pd *pd, 3781 struct ib_qp_init_attr *qp_init_attr, 3782 const char *caller); 3783 /** 3784 * ib_create_qp - Creates a kernel QP associated with the specific protection 3785 * domain. 3786 * @pd: The protection domain associated with the QP. 3787 * @init_attr: A list of initial attributes required to create the 3788 * QP. If QP creation succeeds, then the attributes are updated to 3789 * the actual capabilities of the created QP. 3790 */ 3791 static inline struct ib_qp *ib_create_qp(struct ib_pd *pd, 3792 struct ib_qp_init_attr *init_attr) 3793 { 3794 return ib_create_qp_kernel(pd, init_attr, KBUILD_MODNAME); 3795 } 3796 3797 /** 3798 * ib_modify_qp_with_udata - Modifies the attributes for the specified QP. 3799 * @qp: The QP to modify. 3800 * @attr: On input, specifies the QP attributes to modify. On output, 3801 * the current values of selected QP attributes are returned. 3802 * @attr_mask: A bit-mask used to specify which attributes of the QP 3803 * are being modified. 3804 * @udata: pointer to user's input output buffer information 3805 * are being modified. 3806 * It returns 0 on success and returns appropriate error code on error. 3807 */ 3808 int ib_modify_qp_with_udata(struct ib_qp *qp, 3809 struct ib_qp_attr *attr, 3810 int attr_mask, 3811 struct ib_udata *udata); 3812 3813 /** 3814 * ib_modify_qp - Modifies the attributes for the specified QP and then 3815 * transitions the QP to the given state. 3816 * @qp: The QP to modify. 3817 * @qp_attr: On input, specifies the QP attributes to modify. On output, 3818 * the current values of selected QP attributes are returned. 3819 * @qp_attr_mask: A bit-mask used to specify which attributes of the QP 3820 * are being modified. 3821 */ 3822 int ib_modify_qp(struct ib_qp *qp, 3823 struct ib_qp_attr *qp_attr, 3824 int qp_attr_mask); 3825 3826 /** 3827 * ib_query_qp - Returns the attribute list and current values for the 3828 * specified QP. 3829 * @qp: The QP to query. 3830 * @qp_attr: The attributes of the specified QP. 3831 * @qp_attr_mask: A bit-mask used to select specific attributes to query. 3832 * @qp_init_attr: Additional attributes of the selected QP. 3833 * 3834 * The qp_attr_mask may be used to limit the query to gathering only the 3835 * selected attributes. 3836 */ 3837 int ib_query_qp(struct ib_qp *qp, 3838 struct ib_qp_attr *qp_attr, 3839 int qp_attr_mask, 3840 struct ib_qp_init_attr *qp_init_attr); 3841 3842 /** 3843 * ib_destroy_qp - Destroys the specified QP. 3844 * @qp: The QP to destroy. 3845 * @udata: Valid udata or NULL for kernel objects 3846 */ 3847 int ib_destroy_qp_user(struct ib_qp *qp, struct ib_udata *udata); 3848 3849 /** 3850 * ib_destroy_qp - Destroys the specified kernel QP. 3851 * @qp: The QP to destroy. 3852 * 3853 * NOTE: for user qp use ib_destroy_qp_user with valid udata! 3854 */ 3855 static inline int ib_destroy_qp(struct ib_qp *qp) 3856 { 3857 return ib_destroy_qp_user(qp, NULL); 3858 } 3859 3860 /** 3861 * ib_open_qp - Obtain a reference to an existing sharable QP. 3862 * @xrcd - XRC domain 3863 * @qp_open_attr: Attributes identifying the QP to open. 3864 * 3865 * Returns a reference to a sharable QP. 3866 */ 3867 struct ib_qp *ib_open_qp(struct ib_xrcd *xrcd, 3868 struct ib_qp_open_attr *qp_open_attr); 3869 3870 /** 3871 * ib_close_qp - Release an external reference to a QP. 3872 * @qp: The QP handle to release 3873 * 3874 * The opened QP handle is released by the caller. The underlying 3875 * shared QP is not destroyed until all internal references are released. 3876 */ 3877 int ib_close_qp(struct ib_qp *qp); 3878 3879 /** 3880 * ib_post_send - Posts a list of work requests to the send queue of 3881 * the specified QP. 3882 * @qp: The QP to post the work request on. 3883 * @send_wr: A list of work requests to post on the send queue. 3884 * @bad_send_wr: On an immediate failure, this parameter will reference 3885 * the work request that failed to be posted on the QP. 3886 * 3887 * While IBA Vol. 1 section 11.4.1.1 specifies that if an immediate 3888 * error is returned, the QP state shall not be affected, 3889 * ib_post_send() will return an immediate error after queueing any 3890 * earlier work requests in the list. 3891 */ 3892 static inline int ib_post_send(struct ib_qp *qp, 3893 const struct ib_send_wr *send_wr, 3894 const struct ib_send_wr **bad_send_wr) 3895 { 3896 const struct ib_send_wr *dummy; 3897 3898 return qp->device->ops.post_send(qp, send_wr, bad_send_wr ? : &dummy); 3899 } 3900 3901 /** 3902 * ib_post_recv - Posts a list of work requests to the receive queue of 3903 * the specified QP. 3904 * @qp: The QP to post the work request on. 3905 * @recv_wr: A list of work requests to post on the receive queue. 3906 * @bad_recv_wr: On an immediate failure, this parameter will reference 3907 * the work request that failed to be posted on the QP. 3908 */ 3909 static inline int ib_post_recv(struct ib_qp *qp, 3910 const struct ib_recv_wr *recv_wr, 3911 const struct ib_recv_wr **bad_recv_wr) 3912 { 3913 const struct ib_recv_wr *dummy; 3914 3915 return qp->device->ops.post_recv(qp, recv_wr, bad_recv_wr ? : &dummy); 3916 } 3917 3918 struct ib_cq *__ib_alloc_cq(struct ib_device *dev, void *private, int nr_cqe, 3919 int comp_vector, enum ib_poll_context poll_ctx, 3920 const char *caller); 3921 static inline struct ib_cq *ib_alloc_cq(struct ib_device *dev, void *private, 3922 int nr_cqe, int comp_vector, 3923 enum ib_poll_context poll_ctx) 3924 { 3925 return __ib_alloc_cq(dev, private, nr_cqe, comp_vector, poll_ctx, 3926 KBUILD_MODNAME); 3927 } 3928 3929 struct ib_cq *__ib_alloc_cq_any(struct ib_device *dev, void *private, 3930 int nr_cqe, enum ib_poll_context poll_ctx, 3931 const char *caller); 3932 3933 /** 3934 * ib_alloc_cq_any: Allocate kernel CQ 3935 * @dev: The IB device 3936 * @private: Private data attached to the CQE 3937 * @nr_cqe: Number of CQEs in the CQ 3938 * @poll_ctx: Context used for polling the CQ 3939 */ 3940 static inline struct ib_cq *ib_alloc_cq_any(struct ib_device *dev, 3941 void *private, int nr_cqe, 3942 enum ib_poll_context poll_ctx) 3943 { 3944 return __ib_alloc_cq_any(dev, private, nr_cqe, poll_ctx, 3945 KBUILD_MODNAME); 3946 } 3947 3948 void ib_free_cq(struct ib_cq *cq); 3949 int ib_process_cq_direct(struct ib_cq *cq, int budget); 3950 3951 /** 3952 * ib_create_cq - Creates a CQ on the specified device. 3953 * @device: The device on which to create the CQ. 3954 * @comp_handler: A user-specified callback that is invoked when a 3955 * completion event occurs on the CQ. 3956 * @event_handler: A user-specified callback that is invoked when an 3957 * asynchronous event not associated with a completion occurs on the CQ. 3958 * @cq_context: Context associated with the CQ returned to the user via 3959 * the associated completion and event handlers. 3960 * @cq_attr: The attributes the CQ should be created upon. 3961 * 3962 * Users can examine the cq structure to determine the actual CQ size. 3963 */ 3964 struct ib_cq *__ib_create_cq(struct ib_device *device, 3965 ib_comp_handler comp_handler, 3966 void (*event_handler)(struct ib_event *, void *), 3967 void *cq_context, 3968 const struct ib_cq_init_attr *cq_attr, 3969 const char *caller); 3970 #define ib_create_cq(device, cmp_hndlr, evt_hndlr, cq_ctxt, cq_attr) \ 3971 __ib_create_cq((device), (cmp_hndlr), (evt_hndlr), (cq_ctxt), (cq_attr), KBUILD_MODNAME) 3972 3973 /** 3974 * ib_resize_cq - Modifies the capacity of the CQ. 3975 * @cq: The CQ to resize. 3976 * @cqe: The minimum size of the CQ. 3977 * 3978 * Users can examine the cq structure to determine the actual CQ size. 3979 */ 3980 int ib_resize_cq(struct ib_cq *cq, int cqe); 3981 3982 /** 3983 * rdma_set_cq_moderation - Modifies moderation params of the CQ 3984 * @cq: The CQ to modify. 3985 * @cq_count: number of CQEs that will trigger an event 3986 * @cq_period: max period of time in usec before triggering an event 3987 * 3988 */ 3989 int rdma_set_cq_moderation(struct ib_cq *cq, u16 cq_count, u16 cq_period); 3990 3991 /** 3992 * ib_destroy_cq_user - Destroys the specified CQ. 3993 * @cq: The CQ to destroy. 3994 * @udata: Valid user data or NULL for kernel objects 3995 */ 3996 int ib_destroy_cq_user(struct ib_cq *cq, struct ib_udata *udata); 3997 3998 /** 3999 * ib_destroy_cq - Destroys the specified kernel CQ. 4000 * @cq: The CQ to destroy. 4001 * 4002 * NOTE: for user cq use ib_destroy_cq_user with valid udata! 4003 */ 4004 static inline void ib_destroy_cq(struct ib_cq *cq) 4005 { 4006 int ret = ib_destroy_cq_user(cq, NULL); 4007 4008 WARN_ONCE(ret, "Destroy of kernel CQ shouldn't fail"); 4009 } 4010 4011 /** 4012 * ib_poll_cq - poll a CQ for completion(s) 4013 * @cq:the CQ being polled 4014 * @num_entries:maximum number of completions to return 4015 * @wc:array of at least @num_entries &struct ib_wc where completions 4016 * will be returned 4017 * 4018 * Poll a CQ for (possibly multiple) completions. If the return value 4019 * is < 0, an error occurred. If the return value is >= 0, it is the 4020 * number of completions returned. If the return value is 4021 * non-negative and < num_entries, then the CQ was emptied. 4022 */ 4023 static inline int ib_poll_cq(struct ib_cq *cq, int num_entries, 4024 struct ib_wc *wc) 4025 { 4026 return cq->device->ops.poll_cq(cq, num_entries, wc); 4027 } 4028 4029 /** 4030 * ib_req_notify_cq - Request completion notification on a CQ. 4031 * @cq: The CQ to generate an event for. 4032 * @flags: 4033 * Must contain exactly one of %IB_CQ_SOLICITED or %IB_CQ_NEXT_COMP 4034 * to request an event on the next solicited event or next work 4035 * completion at any type, respectively. %IB_CQ_REPORT_MISSED_EVENTS 4036 * may also be |ed in to request a hint about missed events, as 4037 * described below. 4038 * 4039 * Return Value: 4040 * < 0 means an error occurred while requesting notification 4041 * == 0 means notification was requested successfully, and if 4042 * IB_CQ_REPORT_MISSED_EVENTS was passed in, then no events 4043 * were missed and it is safe to wait for another event. In 4044 * this case is it guaranteed that any work completions added 4045 * to the CQ since the last CQ poll will trigger a completion 4046 * notification event. 4047 * > 0 is only returned if IB_CQ_REPORT_MISSED_EVENTS was passed 4048 * in. It means that the consumer must poll the CQ again to 4049 * make sure it is empty to avoid missing an event because of a 4050 * race between requesting notification and an entry being 4051 * added to the CQ. This return value means it is possible 4052 * (but not guaranteed) that a work completion has been added 4053 * to the CQ since the last poll without triggering a 4054 * completion notification event. 4055 */ 4056 static inline int ib_req_notify_cq(struct ib_cq *cq, 4057 enum ib_cq_notify_flags flags) 4058 { 4059 return cq->device->ops.req_notify_cq(cq, flags); 4060 } 4061 4062 struct ib_cq *ib_cq_pool_get(struct ib_device *dev, unsigned int nr_cqe, 4063 int comp_vector_hint, 4064 enum ib_poll_context poll_ctx); 4065 4066 void ib_cq_pool_put(struct ib_cq *cq, unsigned int nr_cqe); 4067 4068 /* 4069 * Drivers that don't need a DMA mapping at the RDMA layer, set dma_device to 4070 * NULL. This causes the ib_dma* helpers to just stash the kernel virtual 4071 * address into the dma address. 4072 */ 4073 static inline bool ib_uses_virt_dma(struct ib_device *dev) 4074 { 4075 return IS_ENABLED(CONFIG_INFINIBAND_VIRT_DMA) && !dev->dma_device; 4076 } 4077 4078 /* 4079 * Check if a IB device's underlying DMA mapping supports P2PDMA transfers. 4080 */ 4081 static inline bool ib_dma_pci_p2p_dma_supported(struct ib_device *dev) 4082 { 4083 if (ib_uses_virt_dma(dev)) 4084 return false; 4085 4086 return dma_pci_p2pdma_supported(dev->dma_device); 4087 } 4088 4089 /** 4090 * ib_virt_dma_to_ptr - Convert a dma_addr to a kernel pointer 4091 * @dma_addr: The DMA address 4092 * 4093 * Used by ib_uses_virt_dma() devices to get back to the kernel pointer after 4094 * going through the dma_addr marshalling. 4095 */ 4096 static inline void *ib_virt_dma_to_ptr(u64 dma_addr) 4097 { 4098 /* virt_dma mode maps the kvs's directly into the dma addr */ 4099 return (void *)(uintptr_t)dma_addr; 4100 } 4101 4102 /** 4103 * ib_virt_dma_to_page - Convert a dma_addr to a struct page 4104 * @dma_addr: The DMA address 4105 * 4106 * Used by ib_uses_virt_dma() device to get back to the struct page after going 4107 * through the dma_addr marshalling. 4108 */ 4109 static inline struct page *ib_virt_dma_to_page(u64 dma_addr) 4110 { 4111 return virt_to_page(ib_virt_dma_to_ptr(dma_addr)); 4112 } 4113 4114 /** 4115 * ib_dma_mapping_error - check a DMA addr for error 4116 * @dev: The device for which the dma_addr was created 4117 * @dma_addr: The DMA address to check 4118 */ 4119 static inline int ib_dma_mapping_error(struct ib_device *dev, u64 dma_addr) 4120 { 4121 if (ib_uses_virt_dma(dev)) 4122 return 0; 4123 return dma_mapping_error(dev->dma_device, dma_addr); 4124 } 4125 4126 /** 4127 * ib_dma_map_single - Map a kernel virtual address to DMA address 4128 * @dev: The device for which the dma_addr is to be created 4129 * @cpu_addr: The kernel virtual address 4130 * @size: The size of the region in bytes 4131 * @direction: The direction of the DMA 4132 */ 4133 static inline u64 ib_dma_map_single(struct ib_device *dev, 4134 void *cpu_addr, size_t size, 4135 enum dma_data_direction direction) 4136 { 4137 if (ib_uses_virt_dma(dev)) 4138 return (uintptr_t)cpu_addr; 4139 return dma_map_single(dev->dma_device, cpu_addr, size, direction); 4140 } 4141 4142 /** 4143 * ib_dma_unmap_single - Destroy a mapping created by ib_dma_map_single() 4144 * @dev: The device for which the DMA address was created 4145 * @addr: The DMA address 4146 * @size: The size of the region in bytes 4147 * @direction: The direction of the DMA 4148 */ 4149 static inline void ib_dma_unmap_single(struct ib_device *dev, 4150 u64 addr, size_t size, 4151 enum dma_data_direction direction) 4152 { 4153 if (!ib_uses_virt_dma(dev)) 4154 dma_unmap_single(dev->dma_device, addr, size, direction); 4155 } 4156 4157 /** 4158 * ib_dma_map_page - Map a physical page to DMA address 4159 * @dev: The device for which the dma_addr is to be created 4160 * @page: The page to be mapped 4161 * @offset: The offset within the page 4162 * @size: The size of the region in bytes 4163 * @direction: The direction of the DMA 4164 */ 4165 static inline u64 ib_dma_map_page(struct ib_device *dev, 4166 struct page *page, 4167 unsigned long offset, 4168 size_t size, 4169 enum dma_data_direction direction) 4170 { 4171 if (ib_uses_virt_dma(dev)) 4172 return (uintptr_t)(page_address(page) + offset); 4173 return dma_map_page(dev->dma_device, page, offset, size, direction); 4174 } 4175 4176 /** 4177 * ib_dma_unmap_page - Destroy a mapping created by ib_dma_map_page() 4178 * @dev: The device for which the DMA address was created 4179 * @addr: The DMA address 4180 * @size: The size of the region in bytes 4181 * @direction: The direction of the DMA 4182 */ 4183 static inline void ib_dma_unmap_page(struct ib_device *dev, 4184 u64 addr, size_t size, 4185 enum dma_data_direction direction) 4186 { 4187 if (!ib_uses_virt_dma(dev)) 4188 dma_unmap_page(dev->dma_device, addr, size, direction); 4189 } 4190 4191 int ib_dma_virt_map_sg(struct ib_device *dev, struct scatterlist *sg, int nents); 4192 static inline int ib_dma_map_sg_attrs(struct ib_device *dev, 4193 struct scatterlist *sg, int nents, 4194 enum dma_data_direction direction, 4195 unsigned long dma_attrs) 4196 { 4197 if (ib_uses_virt_dma(dev)) 4198 return ib_dma_virt_map_sg(dev, sg, nents); 4199 return dma_map_sg_attrs(dev->dma_device, sg, nents, direction, 4200 dma_attrs); 4201 } 4202 4203 static inline void ib_dma_unmap_sg_attrs(struct ib_device *dev, 4204 struct scatterlist *sg, int nents, 4205 enum dma_data_direction direction, 4206 unsigned long dma_attrs) 4207 { 4208 if (!ib_uses_virt_dma(dev)) 4209 dma_unmap_sg_attrs(dev->dma_device, sg, nents, direction, 4210 dma_attrs); 4211 } 4212 4213 /** 4214 * ib_dma_map_sgtable_attrs - Map a scatter/gather table to DMA addresses 4215 * @dev: The device for which the DMA addresses are to be created 4216 * @sg: The sg_table object describing the buffer 4217 * @direction: The direction of the DMA 4218 * @attrs: Optional DMA attributes for the map operation 4219 */ 4220 static inline int ib_dma_map_sgtable_attrs(struct ib_device *dev, 4221 struct sg_table *sgt, 4222 enum dma_data_direction direction, 4223 unsigned long dma_attrs) 4224 { 4225 int nents; 4226 4227 if (ib_uses_virt_dma(dev)) { 4228 nents = ib_dma_virt_map_sg(dev, sgt->sgl, sgt->orig_nents); 4229 if (!nents) 4230 return -EIO; 4231 sgt->nents = nents; 4232 return 0; 4233 } 4234 return dma_map_sgtable(dev->dma_device, sgt, direction, dma_attrs); 4235 } 4236 4237 static inline void ib_dma_unmap_sgtable_attrs(struct ib_device *dev, 4238 struct sg_table *sgt, 4239 enum dma_data_direction direction, 4240 unsigned long dma_attrs) 4241 { 4242 if (!ib_uses_virt_dma(dev)) 4243 dma_unmap_sgtable(dev->dma_device, sgt, direction, dma_attrs); 4244 } 4245 4246 /** 4247 * ib_dma_map_sg - Map a scatter/gather list to DMA addresses 4248 * @dev: The device for which the DMA addresses are to be created 4249 * @sg: The array of scatter/gather entries 4250 * @nents: The number of scatter/gather entries 4251 * @direction: The direction of the DMA 4252 */ 4253 static inline int ib_dma_map_sg(struct ib_device *dev, 4254 struct scatterlist *sg, int nents, 4255 enum dma_data_direction direction) 4256 { 4257 return ib_dma_map_sg_attrs(dev, sg, nents, direction, 0); 4258 } 4259 4260 /** 4261 * ib_dma_unmap_sg - Unmap a scatter/gather list of DMA addresses 4262 * @dev: The device for which the DMA addresses were created 4263 * @sg: The array of scatter/gather entries 4264 * @nents: The number of scatter/gather entries 4265 * @direction: The direction of the DMA 4266 */ 4267 static inline void ib_dma_unmap_sg(struct ib_device *dev, 4268 struct scatterlist *sg, int nents, 4269 enum dma_data_direction direction) 4270 { 4271 ib_dma_unmap_sg_attrs(dev, sg, nents, direction, 0); 4272 } 4273 4274 /** 4275 * ib_dma_max_seg_size - Return the size limit of a single DMA transfer 4276 * @dev: The device to query 4277 * 4278 * The returned value represents a size in bytes. 4279 */ 4280 static inline unsigned int ib_dma_max_seg_size(struct ib_device *dev) 4281 { 4282 if (ib_uses_virt_dma(dev)) 4283 return UINT_MAX; 4284 return dma_get_max_seg_size(dev->dma_device); 4285 } 4286 4287 /** 4288 * ib_dma_sync_single_for_cpu - Prepare DMA region to be accessed by CPU 4289 * @dev: The device for which the DMA address was created 4290 * @addr: The DMA address 4291 * @size: The size of the region in bytes 4292 * @dir: The direction of the DMA 4293 */ 4294 static inline void ib_dma_sync_single_for_cpu(struct ib_device *dev, 4295 u64 addr, 4296 size_t size, 4297 enum dma_data_direction dir) 4298 { 4299 if (!ib_uses_virt_dma(dev)) 4300 dma_sync_single_for_cpu(dev->dma_device, addr, size, dir); 4301 } 4302 4303 /** 4304 * ib_dma_sync_single_for_device - Prepare DMA region to be accessed by device 4305 * @dev: The device for which the DMA address was created 4306 * @addr: The DMA address 4307 * @size: The size of the region in bytes 4308 * @dir: The direction of the DMA 4309 */ 4310 static inline void ib_dma_sync_single_for_device(struct ib_device *dev, 4311 u64 addr, 4312 size_t size, 4313 enum dma_data_direction dir) 4314 { 4315 if (!ib_uses_virt_dma(dev)) 4316 dma_sync_single_for_device(dev->dma_device, addr, size, dir); 4317 } 4318 4319 /* ib_reg_user_mr - register a memory region for virtual addresses from kernel 4320 * space. This function should be called when 'current' is the owning MM. 4321 */ 4322 struct ib_mr *ib_reg_user_mr(struct ib_pd *pd, u64 start, u64 length, 4323 u64 virt_addr, int mr_access_flags); 4324 4325 /* ib_advise_mr - give an advice about an address range in a memory region */ 4326 int ib_advise_mr(struct ib_pd *pd, enum ib_uverbs_advise_mr_advice advice, 4327 u32 flags, struct ib_sge *sg_list, u32 num_sge); 4328 /** 4329 * ib_dereg_mr_user - Deregisters a memory region and removes it from the 4330 * HCA translation table. 4331 * @mr: The memory region to deregister. 4332 * @udata: Valid user data or NULL for kernel object 4333 * 4334 * This function can fail, if the memory region has memory windows bound to it. 4335 */ 4336 int ib_dereg_mr_user(struct ib_mr *mr, struct ib_udata *udata); 4337 4338 /** 4339 * ib_dereg_mr - Deregisters a kernel memory region and removes it from the 4340 * HCA translation table. 4341 * @mr: The memory region to deregister. 4342 * 4343 * This function can fail, if the memory region has memory windows bound to it. 4344 * 4345 * NOTE: for user mr use ib_dereg_mr_user with valid udata! 4346 */ 4347 static inline int ib_dereg_mr(struct ib_mr *mr) 4348 { 4349 return ib_dereg_mr_user(mr, NULL); 4350 } 4351 4352 struct ib_mr *ib_alloc_mr(struct ib_pd *pd, enum ib_mr_type mr_type, 4353 u32 max_num_sg); 4354 4355 struct ib_mr *ib_alloc_mr_integrity(struct ib_pd *pd, 4356 u32 max_num_data_sg, 4357 u32 max_num_meta_sg); 4358 4359 /** 4360 * ib_update_fast_reg_key - updates the key portion of the fast_reg MR 4361 * R_Key and L_Key. 4362 * @mr - struct ib_mr pointer to be updated. 4363 * @newkey - new key to be used. 4364 */ 4365 static inline void ib_update_fast_reg_key(struct ib_mr *mr, u8 newkey) 4366 { 4367 mr->lkey = (mr->lkey & 0xffffff00) | newkey; 4368 mr->rkey = (mr->rkey & 0xffffff00) | newkey; 4369 } 4370 4371 /** 4372 * ib_inc_rkey - increments the key portion of the given rkey. Can be used 4373 * for calculating a new rkey for type 2 memory windows. 4374 * @rkey - the rkey to increment. 4375 */ 4376 static inline u32 ib_inc_rkey(u32 rkey) 4377 { 4378 const u32 mask = 0x000000ff; 4379 return ((rkey + 1) & mask) | (rkey & ~mask); 4380 } 4381 4382 /** 4383 * ib_attach_mcast - Attaches the specified QP to a multicast group. 4384 * @qp: QP to attach to the multicast group. The QP must be type 4385 * IB_QPT_UD. 4386 * @gid: Multicast group GID. 4387 * @lid: Multicast group LID in host byte order. 4388 * 4389 * In order to send and receive multicast packets, subnet 4390 * administration must have created the multicast group and configured 4391 * the fabric appropriately. The port associated with the specified 4392 * QP must also be a member of the multicast group. 4393 */ 4394 int ib_attach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid); 4395 4396 /** 4397 * ib_detach_mcast - Detaches the specified QP from a multicast group. 4398 * @qp: QP to detach from the multicast group. 4399 * @gid: Multicast group GID. 4400 * @lid: Multicast group LID in host byte order. 4401 */ 4402 int ib_detach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid); 4403 4404 struct ib_xrcd *ib_alloc_xrcd_user(struct ib_device *device, 4405 struct inode *inode, struct ib_udata *udata); 4406 int ib_dealloc_xrcd_user(struct ib_xrcd *xrcd, struct ib_udata *udata); 4407 4408 static inline int ib_check_mr_access(struct ib_device *ib_dev, 4409 unsigned int flags) 4410 { 4411 u64 device_cap = ib_dev->attrs.device_cap_flags; 4412 4413 /* 4414 * Local write permission is required if remote write or 4415 * remote atomic permission is also requested. 4416 */ 4417 if (flags & (IB_ACCESS_REMOTE_ATOMIC | IB_ACCESS_REMOTE_WRITE) && 4418 !(flags & IB_ACCESS_LOCAL_WRITE)) 4419 return -EINVAL; 4420 4421 if (flags & ~IB_ACCESS_SUPPORTED) 4422 return -EINVAL; 4423 4424 if (flags & IB_ACCESS_ON_DEMAND && 4425 !(ib_dev->attrs.kernel_cap_flags & IBK_ON_DEMAND_PAGING)) 4426 return -EOPNOTSUPP; 4427 4428 if ((flags & IB_ACCESS_FLUSH_GLOBAL && 4429 !(device_cap & IB_DEVICE_FLUSH_GLOBAL)) || 4430 (flags & IB_ACCESS_FLUSH_PERSISTENT && 4431 !(device_cap & IB_DEVICE_FLUSH_PERSISTENT))) 4432 return -EOPNOTSUPP; 4433 4434 return 0; 4435 } 4436 4437 static inline bool ib_access_writable(int access_flags) 4438 { 4439 /* 4440 * We have writable memory backing the MR if any of the following 4441 * access flags are set. "Local write" and "remote write" obviously 4442 * require write access. "Remote atomic" can do things like fetch and 4443 * add, which will modify memory, and "MW bind" can change permissions 4444 * by binding a window. 4445 */ 4446 return access_flags & 4447 (IB_ACCESS_LOCAL_WRITE | IB_ACCESS_REMOTE_WRITE | 4448 IB_ACCESS_REMOTE_ATOMIC | IB_ACCESS_MW_BIND); 4449 } 4450 4451 /** 4452 * ib_check_mr_status: lightweight check of MR status. 4453 * This routine may provide status checks on a selected 4454 * ib_mr. first use is for signature status check. 4455 * 4456 * @mr: A memory region. 4457 * @check_mask: Bitmask of which checks to perform from 4458 * ib_mr_status_check enumeration. 4459 * @mr_status: The container of relevant status checks. 4460 * failed checks will be indicated in the status bitmask 4461 * and the relevant info shall be in the error item. 4462 */ 4463 int ib_check_mr_status(struct ib_mr *mr, u32 check_mask, 4464 struct ib_mr_status *mr_status); 4465 4466 /** 4467 * ib_device_try_get: Hold a registration lock 4468 * device: The device to lock 4469 * 4470 * A device under an active registration lock cannot become unregistered. It 4471 * is only possible to obtain a registration lock on a device that is fully 4472 * registered, otherwise this function returns false. 4473 * 4474 * The registration lock is only necessary for actions which require the 4475 * device to still be registered. Uses that only require the device pointer to 4476 * be valid should use get_device(&ibdev->dev) to hold the memory. 4477 * 4478 */ 4479 static inline bool ib_device_try_get(struct ib_device *dev) 4480 { 4481 return refcount_inc_not_zero(&dev->refcount); 4482 } 4483 4484 void ib_device_put(struct ib_device *device); 4485 struct ib_device *ib_device_get_by_netdev(struct net_device *ndev, 4486 enum rdma_driver_id driver_id); 4487 struct ib_device *ib_device_get_by_name(const char *name, 4488 enum rdma_driver_id driver_id); 4489 struct net_device *ib_get_net_dev_by_params(struct ib_device *dev, u32 port, 4490 u16 pkey, const union ib_gid *gid, 4491 const struct sockaddr *addr); 4492 int ib_device_set_netdev(struct ib_device *ib_dev, struct net_device *ndev, 4493 unsigned int port); 4494 struct net_device *ib_device_get_netdev(struct ib_device *ib_dev, 4495 u32 port); 4496 int ib_query_netdev_port(struct ib_device *ibdev, struct net_device *ndev, 4497 u32 *port); 4498 4499 static inline enum ib_port_state ib_get_curr_port_state(struct net_device *net_dev) 4500 { 4501 return (netif_running(net_dev) && netif_carrier_ok(net_dev)) ? 4502 IB_PORT_ACTIVE : IB_PORT_DOWN; 4503 } 4504 4505 void ib_dispatch_port_state_event(struct ib_device *ibdev, 4506 struct net_device *ndev); 4507 struct ib_wq *ib_create_wq(struct ib_pd *pd, 4508 struct ib_wq_init_attr *init_attr); 4509 int ib_destroy_wq_user(struct ib_wq *wq, struct ib_udata *udata); 4510 4511 int ib_map_mr_sg(struct ib_mr *mr, struct scatterlist *sg, int sg_nents, 4512 unsigned int *sg_offset, unsigned int page_size); 4513 int ib_map_mr_sg_pi(struct ib_mr *mr, struct scatterlist *data_sg, 4514 int data_sg_nents, unsigned int *data_sg_offset, 4515 struct scatterlist *meta_sg, int meta_sg_nents, 4516 unsigned int *meta_sg_offset, unsigned int page_size); 4517 4518 static inline int 4519 ib_map_mr_sg_zbva(struct ib_mr *mr, struct scatterlist *sg, int sg_nents, 4520 unsigned int *sg_offset, unsigned int page_size) 4521 { 4522 int n; 4523 4524 n = ib_map_mr_sg(mr, sg, sg_nents, sg_offset, page_size); 4525 mr->iova = 0; 4526 4527 return n; 4528 } 4529 4530 int ib_sg_to_pages(struct ib_mr *mr, struct scatterlist *sgl, int sg_nents, 4531 unsigned int *sg_offset, int (*set_page)(struct ib_mr *, u64)); 4532 4533 void ib_drain_rq(struct ib_qp *qp); 4534 void ib_drain_sq(struct ib_qp *qp); 4535 void ib_drain_qp(struct ib_qp *qp); 4536 4537 int ib_get_eth_speed(struct ib_device *dev, u32 port_num, u16 *speed, 4538 u8 *width); 4539 4540 static inline u8 *rdma_ah_retrieve_dmac(struct rdma_ah_attr *attr) 4541 { 4542 if (attr->type == RDMA_AH_ATTR_TYPE_ROCE) 4543 return attr->roce.dmac; 4544 return NULL; 4545 } 4546 4547 static inline void rdma_ah_set_dlid(struct rdma_ah_attr *attr, u32 dlid) 4548 { 4549 if (attr->type == RDMA_AH_ATTR_TYPE_IB) 4550 attr->ib.dlid = (u16)dlid; 4551 else if (attr->type == RDMA_AH_ATTR_TYPE_OPA) 4552 attr->opa.dlid = dlid; 4553 } 4554 4555 static inline u32 rdma_ah_get_dlid(const struct rdma_ah_attr *attr) 4556 { 4557 if (attr->type == RDMA_AH_ATTR_TYPE_IB) 4558 return attr->ib.dlid; 4559 else if (attr->type == RDMA_AH_ATTR_TYPE_OPA) 4560 return attr->opa.dlid; 4561 return 0; 4562 } 4563 4564 static inline void rdma_ah_set_sl(struct rdma_ah_attr *attr, u8 sl) 4565 { 4566 attr->sl = sl; 4567 } 4568 4569 static inline u8 rdma_ah_get_sl(const struct rdma_ah_attr *attr) 4570 { 4571 return attr->sl; 4572 } 4573 4574 static inline void rdma_ah_set_path_bits(struct rdma_ah_attr *attr, 4575 u8 src_path_bits) 4576 { 4577 if (attr->type == RDMA_AH_ATTR_TYPE_IB) 4578 attr->ib.src_path_bits = src_path_bits; 4579 else if (attr->type == RDMA_AH_ATTR_TYPE_OPA) 4580 attr->opa.src_path_bits = src_path_bits; 4581 } 4582 4583 static inline u8 rdma_ah_get_path_bits(const struct rdma_ah_attr *attr) 4584 { 4585 if (attr->type == RDMA_AH_ATTR_TYPE_IB) 4586 return attr->ib.src_path_bits; 4587 else if (attr->type == RDMA_AH_ATTR_TYPE_OPA) 4588 return attr->opa.src_path_bits; 4589 return 0; 4590 } 4591 4592 static inline void rdma_ah_set_make_grd(struct rdma_ah_attr *attr, 4593 bool make_grd) 4594 { 4595 if (attr->type == RDMA_AH_ATTR_TYPE_OPA) 4596 attr->opa.make_grd = make_grd; 4597 } 4598 4599 static inline bool rdma_ah_get_make_grd(const struct rdma_ah_attr *attr) 4600 { 4601 if (attr->type == RDMA_AH_ATTR_TYPE_OPA) 4602 return attr->opa.make_grd; 4603 return false; 4604 } 4605 4606 static inline void rdma_ah_set_port_num(struct rdma_ah_attr *attr, u32 port_num) 4607 { 4608 attr->port_num = port_num; 4609 } 4610 4611 static inline u32 rdma_ah_get_port_num(const struct rdma_ah_attr *attr) 4612 { 4613 return attr->port_num; 4614 } 4615 4616 static inline void rdma_ah_set_static_rate(struct rdma_ah_attr *attr, 4617 u8 static_rate) 4618 { 4619 attr->static_rate = static_rate; 4620 } 4621 4622 static inline u8 rdma_ah_get_static_rate(const struct rdma_ah_attr *attr) 4623 { 4624 return attr->static_rate; 4625 } 4626 4627 static inline void rdma_ah_set_ah_flags(struct rdma_ah_attr *attr, 4628 enum ib_ah_flags flag) 4629 { 4630 attr->ah_flags = flag; 4631 } 4632 4633 static inline enum ib_ah_flags 4634 rdma_ah_get_ah_flags(const struct rdma_ah_attr *attr) 4635 { 4636 return attr->ah_flags; 4637 } 4638 4639 static inline const struct ib_global_route 4640 *rdma_ah_read_grh(const struct rdma_ah_attr *attr) 4641 { 4642 return &attr->grh; 4643 } 4644 4645 /*To retrieve and modify the grh */ 4646 static inline struct ib_global_route 4647 *rdma_ah_retrieve_grh(struct rdma_ah_attr *attr) 4648 { 4649 return &attr->grh; 4650 } 4651 4652 static inline void rdma_ah_set_dgid_raw(struct rdma_ah_attr *attr, void *dgid) 4653 { 4654 struct ib_global_route *grh = rdma_ah_retrieve_grh(attr); 4655 4656 memcpy(grh->dgid.raw, dgid, sizeof(grh->dgid)); 4657 } 4658 4659 static inline void rdma_ah_set_subnet_prefix(struct rdma_ah_attr *attr, 4660 __be64 prefix) 4661 { 4662 struct ib_global_route *grh = rdma_ah_retrieve_grh(attr); 4663 4664 grh->dgid.global.subnet_prefix = prefix; 4665 } 4666 4667 static inline void rdma_ah_set_interface_id(struct rdma_ah_attr *attr, 4668 __be64 if_id) 4669 { 4670 struct ib_global_route *grh = rdma_ah_retrieve_grh(attr); 4671 4672 grh->dgid.global.interface_id = if_id; 4673 } 4674 4675 static inline void rdma_ah_set_grh(struct rdma_ah_attr *attr, 4676 union ib_gid *dgid, u32 flow_label, 4677 u8 sgid_index, u8 hop_limit, 4678 u8 traffic_class) 4679 { 4680 struct ib_global_route *grh = rdma_ah_retrieve_grh(attr); 4681 4682 attr->ah_flags = IB_AH_GRH; 4683 if (dgid) 4684 grh->dgid = *dgid; 4685 grh->flow_label = flow_label; 4686 grh->sgid_index = sgid_index; 4687 grh->hop_limit = hop_limit; 4688 grh->traffic_class = traffic_class; 4689 grh->sgid_attr = NULL; 4690 } 4691 4692 void rdma_destroy_ah_attr(struct rdma_ah_attr *ah_attr); 4693 void rdma_move_grh_sgid_attr(struct rdma_ah_attr *attr, union ib_gid *dgid, 4694 u32 flow_label, u8 hop_limit, u8 traffic_class, 4695 const struct ib_gid_attr *sgid_attr); 4696 void rdma_copy_ah_attr(struct rdma_ah_attr *dest, 4697 const struct rdma_ah_attr *src); 4698 void rdma_replace_ah_attr(struct rdma_ah_attr *old, 4699 const struct rdma_ah_attr *new); 4700 void rdma_move_ah_attr(struct rdma_ah_attr *dest, struct rdma_ah_attr *src); 4701 4702 /** 4703 * rdma_ah_find_type - Return address handle type. 4704 * 4705 * @dev: Device to be checked 4706 * @port_num: Port number 4707 */ 4708 static inline enum rdma_ah_attr_type rdma_ah_find_type(struct ib_device *dev, 4709 u32 port_num) 4710 { 4711 if (rdma_protocol_roce(dev, port_num)) 4712 return RDMA_AH_ATTR_TYPE_ROCE; 4713 if (rdma_protocol_ib(dev, port_num)) { 4714 if (rdma_cap_opa_ah(dev, port_num)) 4715 return RDMA_AH_ATTR_TYPE_OPA; 4716 return RDMA_AH_ATTR_TYPE_IB; 4717 } 4718 if (dev->type == RDMA_DEVICE_TYPE_SMI) 4719 return RDMA_AH_ATTR_TYPE_IB; 4720 4721 return RDMA_AH_ATTR_TYPE_UNDEFINED; 4722 } 4723 4724 /** 4725 * ib_lid_cpu16 - Return lid in 16bit CPU encoding. 4726 * In the current implementation the only way to 4727 * get the 32bit lid is from other sources for OPA. 4728 * For IB, lids will always be 16bits so cast the 4729 * value accordingly. 4730 * 4731 * @lid: A 32bit LID 4732 */ 4733 static inline u16 ib_lid_cpu16(u32 lid) 4734 { 4735 WARN_ON_ONCE(lid & 0xFFFF0000); 4736 return (u16)lid; 4737 } 4738 4739 /** 4740 * ib_lid_be16 - Return lid in 16bit BE encoding. 4741 * 4742 * @lid: A 32bit LID 4743 */ 4744 static inline __be16 ib_lid_be16(u32 lid) 4745 { 4746 WARN_ON_ONCE(lid & 0xFFFF0000); 4747 return cpu_to_be16((u16)lid); 4748 } 4749 4750 /** 4751 * ib_get_vector_affinity - Get the affinity mappings of a given completion 4752 * vector 4753 * @device: the rdma device 4754 * @comp_vector: index of completion vector 4755 * 4756 * Returns NULL on failure, otherwise a corresponding cpu map of the 4757 * completion vector (returns all-cpus map if the device driver doesn't 4758 * implement get_vector_affinity). 4759 */ 4760 static inline const struct cpumask * 4761 ib_get_vector_affinity(struct ib_device *device, int comp_vector) 4762 { 4763 if (comp_vector < 0 || comp_vector >= device->num_comp_vectors || 4764 !device->ops.get_vector_affinity) 4765 return NULL; 4766 4767 return device->ops.get_vector_affinity(device, comp_vector); 4768 4769 } 4770 4771 /** 4772 * rdma_roce_rescan_device - Rescan all of the network devices in the system 4773 * and add their gids, as needed, to the relevant RoCE devices. 4774 * 4775 * @device: the rdma device 4776 */ 4777 void rdma_roce_rescan_device(struct ib_device *ibdev); 4778 void rdma_roce_rescan_port(struct ib_device *ib_dev, u32 port); 4779 void roce_del_all_netdev_gids(struct ib_device *ib_dev, 4780 u32 port, struct net_device *ndev); 4781 4782 struct ib_ucontext *ib_uverbs_get_ucontext_file(struct ib_uverbs_file *ufile); 4783 4784 int uverbs_destroy_def_handler(struct uverbs_attr_bundle *attrs); 4785 4786 struct net_device *rdma_alloc_netdev(struct ib_device *device, u32 port_num, 4787 enum rdma_netdev_t type, const char *name, 4788 unsigned char name_assign_type, 4789 void (*setup)(struct net_device *)); 4790 4791 int rdma_init_netdev(struct ib_device *device, u32 port_num, 4792 enum rdma_netdev_t type, const char *name, 4793 unsigned char name_assign_type, 4794 void (*setup)(struct net_device *), 4795 struct net_device *netdev); 4796 4797 /** 4798 * rdma_device_to_ibdev - Get ib_device pointer from device pointer 4799 * 4800 * @device: device pointer for which ib_device pointer to retrieve 4801 * 4802 * rdma_device_to_ibdev() retrieves ib_device pointer from device. 4803 * 4804 */ 4805 static inline struct ib_device *rdma_device_to_ibdev(struct device *device) 4806 { 4807 struct ib_core_device *coredev = 4808 container_of(device, struct ib_core_device, dev); 4809 4810 return coredev->owner; 4811 } 4812 4813 /** 4814 * ibdev_to_node - return the NUMA node for a given ib_device 4815 * @dev: device to get the NUMA node for. 4816 */ 4817 static inline int ibdev_to_node(struct ib_device *ibdev) 4818 { 4819 struct device *parent = ibdev->dev.parent; 4820 4821 if (!parent) 4822 return NUMA_NO_NODE; 4823 return dev_to_node(parent); 4824 } 4825 4826 /** 4827 * rdma_device_to_drv_device - Helper macro to reach back to driver's 4828 * ib_device holder structure from device pointer. 4829 * 4830 * NOTE: New drivers should not make use of this API; This API is only for 4831 * existing drivers who have exposed sysfs entries using 4832 * ops->device_group. 4833 */ 4834 #define rdma_device_to_drv_device(dev, drv_dev_struct, ibdev_member) \ 4835 container_of(rdma_device_to_ibdev(dev), drv_dev_struct, ibdev_member) 4836 4837 bool rdma_dev_access_netns(const struct ib_device *device, 4838 const struct net *net); 4839 4840 #define IB_ROCE_UDP_ENCAP_VALID_PORT_MIN (0xC000) 4841 #define IB_ROCE_UDP_ENCAP_VALID_PORT_MAX (0xFFFF) 4842 #define IB_GRH_FLOWLABEL_MASK (0x000FFFFF) 4843 4844 /** 4845 * rdma_flow_label_to_udp_sport - generate a RoCE v2 UDP src port value based 4846 * on the flow_label 4847 * 4848 * This function will convert the 20 bit flow_label input to a valid RoCE v2 4849 * UDP src port 14 bit value. All RoCE V2 drivers should use this same 4850 * convention. 4851 */ 4852 static inline u16 rdma_flow_label_to_udp_sport(u32 fl) 4853 { 4854 u32 fl_low = fl & 0x03fff, fl_high = fl & 0xFC000; 4855 4856 fl_low ^= fl_high >> 14; 4857 return (u16)(fl_low | IB_ROCE_UDP_ENCAP_VALID_PORT_MIN); 4858 } 4859 4860 /** 4861 * rdma_calc_flow_label - generate a RDMA symmetric flow label value based on 4862 * local and remote qpn values 4863 * 4864 * This function folded the multiplication results of two qpns, 24 bit each, 4865 * fields, and converts it to a 20 bit results. 4866 * 4867 * This function will create symmetric flow_label value based on the local 4868 * and remote qpn values. this will allow both the requester and responder 4869 * to calculate the same flow_label for a given connection. 4870 * 4871 * This helper function should be used by driver in case the upper layer 4872 * provide a zero flow_label value. This is to improve entropy of RDMA 4873 * traffic in the network. 4874 */ 4875 static inline u32 rdma_calc_flow_label(u32 lqpn, u32 rqpn) 4876 { 4877 u64 v = (u64)lqpn * rqpn; 4878 4879 v ^= v >> 20; 4880 v ^= v >> 40; 4881 4882 return (u32)(v & IB_GRH_FLOWLABEL_MASK); 4883 } 4884 4885 /** 4886 * rdma_get_udp_sport - Calculate and set UDP source port based on the flow 4887 * label. If flow label is not defined in GRH then 4888 * calculate it based on lqpn/rqpn. 4889 * 4890 * @fl: flow label from GRH 4891 * @lqpn: local qp number 4892 * @rqpn: remote qp number 4893 */ 4894 static inline u16 rdma_get_udp_sport(u32 fl, u32 lqpn, u32 rqpn) 4895 { 4896 if (!fl) 4897 fl = rdma_calc_flow_label(lqpn, rqpn); 4898 4899 return rdma_flow_label_to_udp_sport(fl); 4900 } 4901 4902 const struct ib_port_immutable* 4903 ib_port_immutable_read(struct ib_device *dev, unsigned int port); 4904 4905 /** ib_add_sub_device - Add a sub IB device on an existing one 4906 * 4907 * @parent: The IB device that needs to add a sub device 4908 * @type: The type of the new sub device 4909 * @name: The name of the new sub device 4910 * 4911 * 4912 * Return 0 on success, an error code otherwise 4913 */ 4914 int ib_add_sub_device(struct ib_device *parent, 4915 enum rdma_nl_dev_type type, 4916 const char *name); 4917 4918 4919 /** ib_del_sub_device_and_put - Delect an IB sub device while holding a 'get' 4920 * 4921 * @sub: The sub device that is going to be deleted 4922 * 4923 * Return 0 on success, an error code otherwise 4924 */ 4925 int ib_del_sub_device_and_put(struct ib_device *sub); 4926 4927 static inline void ib_mark_name_assigned_by_user(struct ib_device *ibdev) 4928 { 4929 ibdev->name_assign_type = RDMA_NAME_ASSIGN_TYPE_USER; 4930 } 4931 4932 #endif /* IB_VERBS_H */ 4933