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/bvec.h> 19 #include <linux/dma-mapping.h> 20 #include <linux/kref.h> 21 #include <linux/list.h> 22 #include <linux/rwsem.h> 23 #include <linux/workqueue.h> 24 #include <linux/irq_poll.h> 25 #include <uapi/linux/if_ether.h> 26 #include <net/ipv6.h> 27 #include <net/ip.h> 28 #include <linux/string.h> 29 #include <linux/slab.h> 30 #include <linux/netdevice.h> 31 #include <linux/refcount.h> 32 #include <linux/if_link.h> 33 #include <linux/atomic.h> 34 #include <linux/mmu_notifier.h> 35 #include <linux/uaccess.h> 36 #include <linux/cgroup_rdma.h> 37 #include <linux/irqflags.h> 38 #include <linux/preempt.h> 39 #include <linux/dim.h> 40 #include <uapi/rdma/ib_user_verbs.h> 41 #include <rdma/rdma_counter.h> 42 #include <rdma/restrack.h> 43 #include <rdma/signature.h> 44 #include <uapi/rdma/rdma_user_ioctl.h> 45 #include <uapi/rdma/ib_user_ioctl_verbs.h> 46 #include <linux/pci-tph.h> 47 #include <linux/dma-buf.h> 48 49 #define IB_FW_VERSION_NAME_MAX ETHTOOL_FWVERS_LEN 50 51 struct ib_umem_odp; 52 struct ib_uqp_object; 53 struct ib_usrq_object; 54 struct ib_uwq_object; 55 struct rdma_cm_id; 56 struct ib_port; 57 struct hw_stats_device_data; 58 59 extern struct workqueue_struct *ib_wq; 60 extern struct workqueue_struct *ib_comp_wq; 61 extern struct workqueue_struct *ib_comp_unbound_wq; 62 63 struct ib_ucq_object; 64 65 __printf(2, 3) __cold 66 void ibdev_emerg(const struct ib_device *ibdev, const char *format, ...); 67 __printf(2, 3) __cold 68 void ibdev_alert(const struct ib_device *ibdev, const char *format, ...); 69 __printf(2, 3) __cold 70 void ibdev_crit(const struct ib_device *ibdev, const char *format, ...); 71 __printf(2, 3) __cold 72 void ibdev_err(const struct ib_device *ibdev, const char *format, ...); 73 __printf(2, 3) __cold 74 void ibdev_warn(const struct ib_device *ibdev, const char *format, ...); 75 __printf(2, 3) __cold 76 void ibdev_notice(const struct ib_device *ibdev, const char *format, ...); 77 __printf(2, 3) __cold 78 void ibdev_info(const struct ib_device *ibdev, const char *format, ...); 79 80 #if defined(CONFIG_DYNAMIC_DEBUG) || \ 81 (defined(CONFIG_DYNAMIC_DEBUG_CORE) && defined(DYNAMIC_DEBUG_MODULE)) 82 #define ibdev_dbg(__dev, format, args...) \ 83 dynamic_ibdev_dbg(__dev, format, ##args) 84 #else 85 __printf(2, 3) __cold 86 static inline 87 void ibdev_dbg(const struct ib_device *ibdev, const char *format, ...) {} 88 #endif 89 90 #define ibdev_level_ratelimited(ibdev_level, ibdev, fmt, ...) \ 91 do { \ 92 static DEFINE_RATELIMIT_STATE(_rs, \ 93 DEFAULT_RATELIMIT_INTERVAL, \ 94 DEFAULT_RATELIMIT_BURST); \ 95 if (__ratelimit(&_rs)) \ 96 ibdev_level(ibdev, fmt, ##__VA_ARGS__); \ 97 } while (0) 98 99 #define ibdev_emerg_ratelimited(ibdev, fmt, ...) \ 100 ibdev_level_ratelimited(ibdev_emerg, ibdev, fmt, ##__VA_ARGS__) 101 #define ibdev_alert_ratelimited(ibdev, fmt, ...) \ 102 ibdev_level_ratelimited(ibdev_alert, ibdev, fmt, ##__VA_ARGS__) 103 #define ibdev_crit_ratelimited(ibdev, fmt, ...) \ 104 ibdev_level_ratelimited(ibdev_crit, ibdev, fmt, ##__VA_ARGS__) 105 #define ibdev_err_ratelimited(ibdev, fmt, ...) \ 106 ibdev_level_ratelimited(ibdev_err, ibdev, fmt, ##__VA_ARGS__) 107 #define ibdev_warn_ratelimited(ibdev, fmt, ...) \ 108 ibdev_level_ratelimited(ibdev_warn, ibdev, fmt, ##__VA_ARGS__) 109 #define ibdev_notice_ratelimited(ibdev, fmt, ...) \ 110 ibdev_level_ratelimited(ibdev_notice, ibdev, fmt, ##__VA_ARGS__) 111 #define ibdev_info_ratelimited(ibdev, fmt, ...) \ 112 ibdev_level_ratelimited(ibdev_info, ibdev, fmt, ##__VA_ARGS__) 113 114 #if defined(CONFIG_DYNAMIC_DEBUG) || \ 115 (defined(CONFIG_DYNAMIC_DEBUG_CORE) && defined(DYNAMIC_DEBUG_MODULE)) 116 /* descriptor check is first to prevent flooding with "callbacks suppressed" */ 117 #define ibdev_dbg_ratelimited(ibdev, fmt, ...) \ 118 do { \ 119 static DEFINE_RATELIMIT_STATE(_rs, \ 120 DEFAULT_RATELIMIT_INTERVAL, \ 121 DEFAULT_RATELIMIT_BURST); \ 122 DEFINE_DYNAMIC_DEBUG_METADATA(descriptor, fmt); \ 123 if (DYNAMIC_DEBUG_BRANCH(descriptor) && __ratelimit(&_rs)) \ 124 __dynamic_ibdev_dbg(&descriptor, ibdev, fmt, \ 125 ##__VA_ARGS__); \ 126 } while (0) 127 #else 128 __printf(2, 3) __cold 129 static inline 130 void ibdev_dbg_ratelimited(const struct ib_device *ibdev, const char *format, ...) {} 131 #endif 132 133 union ib_gid { 134 u8 raw[16]; 135 struct { 136 __be64 subnet_prefix; 137 __be64 interface_id; 138 } global; 139 }; 140 141 extern union ib_gid zgid; 142 143 enum ib_gid_type { 144 IB_GID_TYPE_IB = IB_UVERBS_GID_TYPE_IB, 145 IB_GID_TYPE_ROCE = IB_UVERBS_GID_TYPE_ROCE_V1, 146 IB_GID_TYPE_ROCE_UDP_ENCAP = IB_UVERBS_GID_TYPE_ROCE_V2, 147 IB_GID_TYPE_SIZE 148 }; 149 150 #define ROCE_V2_UDP_DPORT 4791 151 struct ib_gid_attr { 152 struct net_device __rcu *ndev; 153 struct ib_device *device; 154 union ib_gid gid; 155 enum ib_gid_type gid_type; 156 u16 index; 157 u32 port_num; 158 }; 159 160 enum { 161 /* set the local administered indication */ 162 IB_SA_WELL_KNOWN_GUID = BIT_ULL(57) | 2, 163 }; 164 165 enum rdma_transport_type { 166 RDMA_TRANSPORT_IB, 167 RDMA_TRANSPORT_IWARP, 168 RDMA_TRANSPORT_USNIC, 169 RDMA_TRANSPORT_USNIC_UDP, 170 RDMA_TRANSPORT_UNSPECIFIED, 171 }; 172 173 enum rdma_protocol_type { 174 RDMA_PROTOCOL_IB, 175 RDMA_PROTOCOL_IBOE, 176 RDMA_PROTOCOL_IWARP, 177 RDMA_PROTOCOL_USNIC_UDP 178 }; 179 180 __attribute_const__ enum rdma_transport_type 181 rdma_node_get_transport(unsigned int node_type); 182 183 enum rdma_network_type { 184 RDMA_NETWORK_IB, 185 RDMA_NETWORK_ROCE_V1, 186 RDMA_NETWORK_IPV4, 187 RDMA_NETWORK_IPV6 188 }; 189 190 static inline enum ib_gid_type ib_network_to_gid_type(enum rdma_network_type network_type) 191 { 192 if (network_type == RDMA_NETWORK_IPV4 || 193 network_type == RDMA_NETWORK_IPV6) 194 return IB_GID_TYPE_ROCE_UDP_ENCAP; 195 else if (network_type == RDMA_NETWORK_ROCE_V1) 196 return IB_GID_TYPE_ROCE; 197 else 198 return IB_GID_TYPE_IB; 199 } 200 201 static inline enum rdma_network_type 202 rdma_gid_attr_network_type(const struct ib_gid_attr *attr) 203 { 204 if (attr->gid_type == IB_GID_TYPE_IB) 205 return RDMA_NETWORK_IB; 206 207 if (attr->gid_type == IB_GID_TYPE_ROCE) 208 return RDMA_NETWORK_ROCE_V1; 209 210 if (ipv6_addr_v4mapped((struct in6_addr *)&attr->gid)) 211 return RDMA_NETWORK_IPV4; 212 else 213 return RDMA_NETWORK_IPV6; 214 } 215 216 enum rdma_link_layer { 217 IB_LINK_LAYER_UNSPECIFIED, 218 IB_LINK_LAYER_INFINIBAND, 219 IB_LINK_LAYER_ETHERNET, 220 }; 221 222 enum ib_device_cap_flags { 223 IB_DEVICE_RESIZE_MAX_WR = IB_UVERBS_DEVICE_RESIZE_MAX_WR, 224 IB_DEVICE_BAD_PKEY_CNTR = IB_UVERBS_DEVICE_BAD_PKEY_CNTR, 225 IB_DEVICE_BAD_QKEY_CNTR = IB_UVERBS_DEVICE_BAD_QKEY_CNTR, 226 IB_DEVICE_RAW_MULTI = IB_UVERBS_DEVICE_RAW_MULTI, 227 IB_DEVICE_AUTO_PATH_MIG = IB_UVERBS_DEVICE_AUTO_PATH_MIG, 228 IB_DEVICE_CHANGE_PHY_PORT = IB_UVERBS_DEVICE_CHANGE_PHY_PORT, 229 IB_DEVICE_UD_AV_PORT_ENFORCE = IB_UVERBS_DEVICE_UD_AV_PORT_ENFORCE, 230 IB_DEVICE_CURR_QP_STATE_MOD = IB_UVERBS_DEVICE_CURR_QP_STATE_MOD, 231 IB_DEVICE_SHUTDOWN_PORT = IB_UVERBS_DEVICE_SHUTDOWN_PORT, 232 /* IB_DEVICE_INIT_TYPE = IB_UVERBS_DEVICE_INIT_TYPE, (not in use) */ 233 IB_DEVICE_PORT_ACTIVE_EVENT = IB_UVERBS_DEVICE_PORT_ACTIVE_EVENT, 234 IB_DEVICE_SYS_IMAGE_GUID = IB_UVERBS_DEVICE_SYS_IMAGE_GUID, 235 IB_DEVICE_RC_RNR_NAK_GEN = IB_UVERBS_DEVICE_RC_RNR_NAK_GEN, 236 IB_DEVICE_SRQ_RESIZE = IB_UVERBS_DEVICE_SRQ_RESIZE, 237 IB_DEVICE_N_NOTIFY_CQ = IB_UVERBS_DEVICE_N_NOTIFY_CQ, 238 239 /* Reserved, old SEND_W_INV = 1 << 16,*/ 240 IB_DEVICE_MEM_WINDOW = IB_UVERBS_DEVICE_MEM_WINDOW, 241 /* 242 * Devices should set IB_DEVICE_UD_IP_SUM if they support 243 * insertion of UDP and TCP checksum on outgoing UD IPoIB 244 * messages and can verify the validity of checksum for 245 * incoming messages. Setting this flag implies that the 246 * IPoIB driver may set NETIF_F_IP_CSUM for datagram mode. 247 */ 248 IB_DEVICE_UD_IP_CSUM = IB_UVERBS_DEVICE_UD_IP_CSUM, 249 IB_DEVICE_XRC = IB_UVERBS_DEVICE_XRC, 250 251 /* 252 * This device supports the IB "base memory management extension", 253 * which includes support for fast registrations (IB_WR_REG_MR, 254 * IB_WR_LOCAL_INV and IB_WR_SEND_WITH_INV verbs). This flag should 255 * also be set by any iWarp device which must support FRs to comply 256 * to the iWarp verbs spec. iWarp devices also support the 257 * IB_WR_RDMA_READ_WITH_INV verb for RDMA READs that invalidate the 258 * stag. 259 */ 260 IB_DEVICE_MEM_MGT_EXTENSIONS = IB_UVERBS_DEVICE_MEM_MGT_EXTENSIONS, 261 IB_DEVICE_MEM_WINDOW_TYPE_2A = IB_UVERBS_DEVICE_MEM_WINDOW_TYPE_2A, 262 IB_DEVICE_MEM_WINDOW_TYPE_2B = IB_UVERBS_DEVICE_MEM_WINDOW_TYPE_2B, 263 IB_DEVICE_RC_IP_CSUM = IB_UVERBS_DEVICE_RC_IP_CSUM, 264 /* Deprecated. Please use IB_RAW_PACKET_CAP_IP_CSUM. */ 265 IB_DEVICE_RAW_IP_CSUM = IB_UVERBS_DEVICE_RAW_IP_CSUM, 266 IB_DEVICE_MANAGED_FLOW_STEERING = 267 IB_UVERBS_DEVICE_MANAGED_FLOW_STEERING, 268 /* Deprecated. Please use IB_RAW_PACKET_CAP_SCATTER_FCS. */ 269 IB_DEVICE_RAW_SCATTER_FCS = IB_UVERBS_DEVICE_RAW_SCATTER_FCS, 270 /* The device supports padding incoming writes to cacheline. */ 271 IB_DEVICE_PCI_WRITE_END_PADDING = 272 IB_UVERBS_DEVICE_PCI_WRITE_END_PADDING, 273 /* Placement type attributes */ 274 IB_DEVICE_FLUSH_GLOBAL = IB_UVERBS_DEVICE_FLUSH_GLOBAL, 275 IB_DEVICE_FLUSH_PERSISTENT = IB_UVERBS_DEVICE_FLUSH_PERSISTENT, 276 IB_DEVICE_ATOMIC_WRITE = IB_UVERBS_DEVICE_ATOMIC_WRITE, 277 }; 278 279 enum ib_kernel_cap_flags { 280 /* 281 * This device supports a per-device lkey or stag that can be 282 * used without performing a memory registration for the local 283 * memory. Note that ULPs should never check this flag, but 284 * instead of use the local_dma_lkey flag in the ib_pd structure, 285 * which will always contain a usable lkey. 286 */ 287 IBK_LOCAL_DMA_LKEY = 1 << 0, 288 /* IB_QP_CREATE_INTEGRITY_EN is supported to implement T10-PI */ 289 IBK_INTEGRITY_HANDOVER = 1 << 1, 290 /* IB_ACCESS_ON_DEMAND is supported during reg_user_mr() */ 291 IBK_ON_DEMAND_PAGING = 1 << 2, 292 /* IB_MR_TYPE_SG_GAPS is supported */ 293 IBK_SG_GAPS_REG = 1 << 3, 294 /* Driver supports RDMA_NLDEV_CMD_DELLINK */ 295 IBK_ALLOW_USER_UNREG = 1 << 4, 296 297 /* ipoib will use IB_QP_CREATE_BLOCK_MULTICAST_LOOPBACK */ 298 IBK_BLOCK_MULTICAST_LOOPBACK = 1 << 5, 299 /* iopib will use IB_QP_CREATE_IPOIB_UD_LSO for its QPs */ 300 IBK_UD_TSO = 1 << 6, 301 /* iopib will use the device ops: 302 * get_vf_config 303 * get_vf_guid 304 * get_vf_stats 305 * set_vf_guid 306 * set_vf_link_state 307 */ 308 IBK_VIRTUAL_FUNCTION = 1 << 7, 309 /* ipoib will use IB_QP_CREATE_NETDEV_USE for its QPs */ 310 IBK_RDMA_NETDEV_OPA = 1 << 8, 311 }; 312 313 enum ib_atomic_cap { 314 IB_ATOMIC_NONE, 315 IB_ATOMIC_HCA, 316 IB_ATOMIC_GLOB 317 }; 318 319 enum ib_odp_general_cap_bits { 320 IB_ODP_SUPPORT = IB_UVERBS_ODP_SUPPORT, 321 IB_ODP_SUPPORT_IMPLICIT = IB_UVERBS_ODP_SUPPORT_IMPLICIT, 322 }; 323 324 enum ib_odp_transport_cap_bits { 325 IB_ODP_SUPPORT_SEND = IB_UVERBS_ODP_SUPPORT_SEND, 326 IB_ODP_SUPPORT_RECV = IB_UVERBS_ODP_SUPPORT_RECV, 327 IB_ODP_SUPPORT_WRITE = IB_UVERBS_ODP_SUPPORT_WRITE, 328 IB_ODP_SUPPORT_READ = IB_UVERBS_ODP_SUPPORT_READ, 329 IB_ODP_SUPPORT_ATOMIC = IB_UVERBS_ODP_SUPPORT_ATOMIC, 330 IB_ODP_SUPPORT_SRQ_RECV = IB_UVERBS_ODP_SUPPORT_SRQ_RECV, 331 IB_ODP_SUPPORT_FLUSH = IB_UVERBS_ODP_SUPPORT_FLUSH, 332 IB_ODP_SUPPORT_ATOMIC_WRITE = IB_UVERBS_ODP_SUPPORT_ATOMIC_WRITE, 333 }; 334 335 struct ib_odp_caps { 336 uint64_t general_caps; 337 struct { 338 uint32_t rc_odp_caps; 339 uint32_t uc_odp_caps; 340 uint32_t ud_odp_caps; 341 uint32_t xrc_odp_caps; 342 } per_transport_caps; 343 }; 344 345 struct ib_rss_caps { 346 /* Corresponding bit will be set if qp type from 347 * 'enum ib_qp_type' is supported, e.g. 348 * supported_qpts |= 1 << IB_QPT_UD 349 */ 350 u32 supported_qpts; 351 u32 max_rwq_indirection_tables; 352 u32 max_rwq_indirection_table_size; 353 }; 354 355 enum ib_tm_cap_flags { 356 /* Support tag matching with rendezvous offload for RC transport */ 357 IB_TM_CAP_RNDV_RC = 1 << 0, 358 }; 359 360 struct ib_tm_caps { 361 /* Max size of RNDV header */ 362 u32 max_rndv_hdr_size; 363 /* Max number of entries in tag matching list */ 364 u32 max_num_tags; 365 /* From enum ib_tm_cap_flags */ 366 u32 flags; 367 /* Max number of outstanding list operations */ 368 u32 max_ops; 369 /* Max number of SGE in tag matching entry */ 370 u32 max_sge; 371 }; 372 373 struct ib_cq_init_attr { 374 unsigned int cqe; 375 u32 comp_vector; 376 u32 flags; 377 }; 378 379 enum ib_cq_attr_mask { 380 IB_CQ_MODERATE = 1 << 0, 381 }; 382 383 struct ib_cq_caps { 384 u16 max_cq_moderation_count; 385 u16 max_cq_moderation_period; 386 }; 387 388 struct ib_dm_mr_attr { 389 u64 length; 390 u64 offset; 391 u32 access_flags; 392 }; 393 394 struct ib_dm_alloc_attr { 395 u64 length; 396 u32 alignment; 397 u32 flags; 398 }; 399 400 struct ib_device_attr { 401 u64 fw_ver; 402 __be64 sys_image_guid; 403 u64 max_mr_size; 404 u64 page_size_cap; 405 u32 vendor_id; 406 u32 vendor_part_id; 407 u32 hw_ver; 408 int max_qp; 409 int max_qp_wr; 410 u64 device_cap_flags; 411 u64 kernel_cap_flags; 412 int max_send_sge; 413 int max_recv_sge; 414 int max_sge_rd; 415 int max_cq; 416 int max_cqe; 417 int max_mr; 418 int max_pd; 419 int max_qp_rd_atom; 420 int max_ee_rd_atom; 421 int max_res_rd_atom; 422 int max_qp_init_rd_atom; 423 int max_ee_init_rd_atom; 424 enum ib_atomic_cap atomic_cap; 425 enum ib_atomic_cap masked_atomic_cap; 426 int max_ee; 427 int max_rdd; 428 int max_mw; 429 int max_raw_ipv6_qp; 430 int max_raw_ethy_qp; 431 int max_mcast_grp; 432 int max_mcast_qp_attach; 433 int max_total_mcast_qp_attach; 434 int max_ah; 435 int max_srq; 436 int max_srq_wr; 437 int max_srq_sge; 438 unsigned int max_fast_reg_page_list_len; 439 unsigned int max_pi_fast_reg_page_list_len; 440 u16 max_pkeys; 441 u8 local_ca_ack_delay; 442 int sig_prot_cap; 443 int sig_guard_cap; 444 struct ib_odp_caps odp_caps; 445 uint64_t timestamp_mask; 446 uint64_t hca_core_clock; /* in KHZ */ 447 struct ib_rss_caps rss_caps; 448 u32 max_wq_type_rq; 449 u32 raw_packet_caps; /* Use ib_raw_packet_caps enum */ 450 struct ib_tm_caps tm_caps; 451 struct ib_cq_caps cq_caps; 452 u64 max_dm_size; 453 /* Max entries for sgl for optimized performance per READ */ 454 u32 max_sgl_rd; 455 }; 456 457 enum ib_mtu { 458 IB_MTU_256 = 1, 459 IB_MTU_512 = 2, 460 IB_MTU_1024 = 3, 461 IB_MTU_2048 = 4, 462 IB_MTU_4096 = 5 463 }; 464 465 enum opa_mtu { 466 OPA_MTU_8192 = 6, 467 OPA_MTU_10240 = 7 468 }; 469 470 static inline int ib_mtu_enum_to_int(enum ib_mtu mtu) 471 { 472 switch (mtu) { 473 case IB_MTU_256: return 256; 474 case IB_MTU_512: return 512; 475 case IB_MTU_1024: return 1024; 476 case IB_MTU_2048: return 2048; 477 case IB_MTU_4096: return 4096; 478 default: return -1; 479 } 480 } 481 482 static inline enum ib_mtu ib_mtu_int_to_enum(int mtu) 483 { 484 if (mtu >= 4096) 485 return IB_MTU_4096; 486 else if (mtu >= 2048) 487 return IB_MTU_2048; 488 else if (mtu >= 1024) 489 return IB_MTU_1024; 490 else if (mtu >= 512) 491 return IB_MTU_512; 492 else 493 return IB_MTU_256; 494 } 495 496 static inline int opa_mtu_enum_to_int(enum opa_mtu mtu) 497 { 498 switch (mtu) { 499 case OPA_MTU_8192: 500 return 8192; 501 case OPA_MTU_10240: 502 return 10240; 503 default: 504 return(ib_mtu_enum_to_int((enum ib_mtu)mtu)); 505 } 506 } 507 508 static inline enum opa_mtu opa_mtu_int_to_enum(int mtu) 509 { 510 if (mtu >= 10240) 511 return OPA_MTU_10240; 512 else if (mtu >= 8192) 513 return OPA_MTU_8192; 514 else 515 return ((enum opa_mtu)ib_mtu_int_to_enum(mtu)); 516 } 517 518 enum ib_port_state { 519 IB_PORT_NOP = 0, 520 IB_PORT_DOWN = 1, 521 IB_PORT_INIT = 2, 522 IB_PORT_ARMED = 3, 523 IB_PORT_ACTIVE = 4, 524 IB_PORT_ACTIVE_DEFER = 5 525 }; 526 527 static inline const char *__attribute_const__ 528 ib_port_state_to_str(enum ib_port_state state) 529 { 530 const char * const states[] = { 531 [IB_PORT_NOP] = "NOP", 532 [IB_PORT_DOWN] = "DOWN", 533 [IB_PORT_INIT] = "INIT", 534 [IB_PORT_ARMED] = "ARMED", 535 [IB_PORT_ACTIVE] = "ACTIVE", 536 [IB_PORT_ACTIVE_DEFER] = "ACTIVE_DEFER", 537 }; 538 539 if (state < ARRAY_SIZE(states)) 540 return states[state]; 541 return "UNKNOWN"; 542 } 543 544 enum ib_port_phys_state { 545 IB_PORT_PHYS_STATE_SLEEP = 1, 546 IB_PORT_PHYS_STATE_POLLING = 2, 547 IB_PORT_PHYS_STATE_DISABLED = 3, 548 IB_PORT_PHYS_STATE_PORT_CONFIGURATION_TRAINING = 4, 549 IB_PORT_PHYS_STATE_LINK_UP = 5, 550 IB_PORT_PHYS_STATE_LINK_ERROR_RECOVERY = 6, 551 IB_PORT_PHYS_STATE_PHY_TEST = 7, 552 }; 553 554 enum ib_port_width { 555 IB_WIDTH_1X = 1, 556 IB_WIDTH_2X = 16, 557 IB_WIDTH_4X = 2, 558 IB_WIDTH_8X = 4, 559 IB_WIDTH_12X = 8 560 }; 561 562 static inline int ib_width_enum_to_int(enum ib_port_width width) 563 { 564 switch (width) { 565 case IB_WIDTH_1X: return 1; 566 case IB_WIDTH_2X: return 2; 567 case IB_WIDTH_4X: return 4; 568 case IB_WIDTH_8X: return 8; 569 case IB_WIDTH_12X: return 12; 570 default: return -1; 571 } 572 } 573 574 enum ib_port_speed { 575 IB_SPEED_SDR = 1, 576 IB_SPEED_DDR = 2, 577 IB_SPEED_QDR = 4, 578 IB_SPEED_FDR10 = 8, 579 IB_SPEED_FDR = 16, 580 IB_SPEED_EDR = 32, 581 IB_SPEED_HDR = 64, 582 IB_SPEED_NDR = 128, 583 IB_SPEED_XDR = 256, 584 }; 585 586 enum ib_stat_flag { 587 IB_STAT_FLAG_OPTIONAL = 1 << 0, 588 }; 589 590 /** 591 * struct rdma_stat_desc - description of one rdma stat/counter 592 * @name: The name of the counter 593 * @flags: Flags of the counter; For example, IB_STAT_FLAG_OPTIONAL 594 * @priv: Driver private information; Core code should not use 595 */ 596 struct rdma_stat_desc { 597 const char *name; 598 unsigned int flags; 599 const void *priv; 600 }; 601 602 /** 603 * struct rdma_hw_stats - collection of hardware stats and their management 604 * @lock: Mutex to protect parallel write access to lifespan and values 605 * of counters, which are 64bits and not guaranteed to be written 606 * atomicaly on 32bits systems. 607 * @timestamp: Used by the core code to track when the last update was 608 * @lifespan: Used by the core code to determine how old the counters 609 * should be before being updated again. Stored in jiffies, defaults 610 * to 10 milliseconds, drivers can override the default be specifying 611 * their own value during their allocation routine. 612 * @descs: Array of pointers to static descriptors used for the counters 613 * in directory. 614 * @is_disabled: A bitmap to indicate each counter is currently disabled 615 * or not. 616 * @num_counters: How many hardware counters there are. If name is 617 * shorter than this number, a kernel oops will result. Driver authors 618 * are encouraged to leave BUILD_BUG_ON(ARRAY_SIZE(@name) < num_counters) 619 * in their code to prevent this. 620 * @value: Array of u64 counters that are accessed by the sysfs code and 621 * filled in by the drivers get_stats routine 622 */ 623 struct rdma_hw_stats { 624 struct mutex lock; /* Protect lifespan and values[] */ 625 unsigned long timestamp; 626 unsigned long lifespan; 627 const struct rdma_stat_desc *descs; 628 unsigned long *is_disabled; 629 int num_counters; 630 u64 value[] __counted_by(num_counters); 631 }; 632 633 #define RDMA_HW_STATS_DEFAULT_LIFESPAN 10 634 635 struct rdma_hw_stats *rdma_alloc_hw_stats_struct( 636 const struct rdma_stat_desc *descs, int num_counters, 637 unsigned long lifespan); 638 639 void rdma_free_hw_stats_struct(struct rdma_hw_stats *stats); 640 641 /* Define bits for the various functionality this port needs to be supported by 642 * the core. 643 */ 644 /* Management 0x00000FFF */ 645 #define RDMA_CORE_CAP_IB_MAD 0x00000001 646 #define RDMA_CORE_CAP_IB_SMI 0x00000002 647 #define RDMA_CORE_CAP_IB_CM 0x00000004 648 #define RDMA_CORE_CAP_IW_CM 0x00000008 649 #define RDMA_CORE_CAP_IB_SA 0x00000010 650 #define RDMA_CORE_CAP_OPA_MAD 0x00000020 651 652 /* Address format 0x000FF000 */ 653 #define RDMA_CORE_CAP_AF_IB 0x00001000 654 #define RDMA_CORE_CAP_ETH_AH 0x00002000 655 #define RDMA_CORE_CAP_OPA_AH 0x00004000 656 #define RDMA_CORE_CAP_IB_GRH_REQUIRED 0x00008000 657 658 /* Protocol 0xFFF00000 */ 659 #define RDMA_CORE_CAP_PROT_IB 0x00100000 660 #define RDMA_CORE_CAP_PROT_ROCE 0x00200000 661 #define RDMA_CORE_CAP_PROT_IWARP 0x00400000 662 #define RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP 0x00800000 663 #define RDMA_CORE_CAP_PROT_RAW_PACKET 0x01000000 664 #define RDMA_CORE_CAP_PROT_USNIC 0x02000000 665 666 #define RDMA_CORE_PORT_IB_GRH_REQUIRED (RDMA_CORE_CAP_IB_GRH_REQUIRED \ 667 | RDMA_CORE_CAP_PROT_ROCE \ 668 | RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP) 669 670 #define RDMA_CORE_PORT_IBA_IB (RDMA_CORE_CAP_PROT_IB \ 671 | RDMA_CORE_CAP_IB_MAD \ 672 | RDMA_CORE_CAP_IB_SMI \ 673 | RDMA_CORE_CAP_IB_CM \ 674 | RDMA_CORE_CAP_IB_SA \ 675 | RDMA_CORE_CAP_AF_IB) 676 #define RDMA_CORE_PORT_IBA_ROCE (RDMA_CORE_CAP_PROT_ROCE \ 677 | RDMA_CORE_CAP_IB_MAD \ 678 | RDMA_CORE_CAP_IB_CM \ 679 | RDMA_CORE_CAP_AF_IB \ 680 | RDMA_CORE_CAP_ETH_AH) 681 #define RDMA_CORE_PORT_IBA_ROCE_UDP_ENCAP \ 682 (RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP \ 683 | RDMA_CORE_CAP_IB_MAD \ 684 | RDMA_CORE_CAP_IB_CM \ 685 | RDMA_CORE_CAP_AF_IB \ 686 | RDMA_CORE_CAP_ETH_AH) 687 #define RDMA_CORE_PORT_IWARP (RDMA_CORE_CAP_PROT_IWARP \ 688 | RDMA_CORE_CAP_IW_CM) 689 #define RDMA_CORE_PORT_INTEL_OPA (RDMA_CORE_PORT_IBA_IB \ 690 | RDMA_CORE_CAP_OPA_MAD) 691 692 #define RDMA_CORE_PORT_RAW_PACKET (RDMA_CORE_CAP_PROT_RAW_PACKET) 693 694 #define RDMA_CORE_PORT_USNIC (RDMA_CORE_CAP_PROT_USNIC) 695 696 struct ib_port_attr { 697 u64 subnet_prefix; 698 enum ib_port_state state; 699 enum ib_mtu max_mtu; 700 enum ib_mtu active_mtu; 701 u32 phys_mtu; 702 int gid_tbl_len; 703 unsigned int ip_gids:1; 704 /* This is the value from PortInfo CapabilityMask, defined by IBA */ 705 u32 port_cap_flags; 706 u32 max_msg_sz; 707 u32 bad_pkey_cntr; 708 u32 qkey_viol_cntr; 709 u16 pkey_tbl_len; 710 u32 sm_lid; 711 u32 lid; 712 u8 lmc; 713 u8 max_vl_num; 714 u8 sm_sl; 715 u8 subnet_timeout; 716 u8 init_type_reply; 717 u8 active_width; 718 u16 active_speed; 719 u8 phys_state; 720 u16 port_cap_flags2; 721 }; 722 723 enum ib_device_modify_flags { 724 IB_DEVICE_MODIFY_SYS_IMAGE_GUID = 1 << 0, 725 IB_DEVICE_MODIFY_NODE_DESC = 1 << 1 726 }; 727 728 #define IB_DEVICE_NODE_DESC_MAX 64 729 730 struct ib_device_modify { 731 u64 sys_image_guid; 732 char node_desc[IB_DEVICE_NODE_DESC_MAX]; 733 }; 734 735 enum ib_port_modify_flags { 736 IB_PORT_SHUTDOWN = 1, 737 IB_PORT_INIT_TYPE = (1<<2), 738 IB_PORT_RESET_QKEY_CNTR = (1<<3), 739 IB_PORT_OPA_MASK_CHG = (1<<4) 740 }; 741 742 struct ib_port_modify { 743 u32 set_port_cap_mask; 744 u32 clr_port_cap_mask; 745 u8 init_type; 746 }; 747 748 enum ib_event_type { 749 IB_EVENT_CQ_ERR, 750 IB_EVENT_QP_FATAL, 751 IB_EVENT_QP_REQ_ERR, 752 IB_EVENT_QP_ACCESS_ERR, 753 IB_EVENT_COMM_EST, 754 IB_EVENT_SQ_DRAINED, 755 IB_EVENT_PATH_MIG, 756 IB_EVENT_PATH_MIG_ERR, 757 IB_EVENT_DEVICE_FATAL, 758 IB_EVENT_PORT_ACTIVE, 759 IB_EVENT_PORT_ERR, 760 IB_EVENT_LID_CHANGE, 761 IB_EVENT_PKEY_CHANGE, 762 IB_EVENT_SM_CHANGE, 763 IB_EVENT_SRQ_ERR, 764 IB_EVENT_SRQ_LIMIT_REACHED, 765 IB_EVENT_QP_LAST_WQE_REACHED, 766 IB_EVENT_CLIENT_REREGISTER, 767 IB_EVENT_GID_CHANGE, 768 IB_EVENT_WQ_FATAL, 769 IB_EVENT_DEVICE_SPEED_CHANGE, 770 }; 771 772 const char *__attribute_const__ ib_event_msg(enum ib_event_type event); 773 774 struct ib_event { 775 struct ib_device *device; 776 union { 777 struct ib_cq *cq; 778 struct ib_qp *qp; 779 struct ib_srq *srq; 780 struct ib_wq *wq; 781 u32 port_num; 782 } element; 783 enum ib_event_type event; 784 }; 785 786 struct ib_event_handler { 787 struct ib_device *device; 788 void (*handler)(struct ib_event_handler *, struct ib_event *); 789 struct list_head list; 790 }; 791 792 #define INIT_IB_EVENT_HANDLER(_ptr, _device, _handler) \ 793 do { \ 794 (_ptr)->device = _device; \ 795 (_ptr)->handler = _handler; \ 796 INIT_LIST_HEAD(&(_ptr)->list); \ 797 } while (0) 798 799 struct ib_global_route { 800 const struct ib_gid_attr *sgid_attr; 801 union ib_gid dgid; 802 u32 flow_label; 803 u8 sgid_index; 804 u8 hop_limit; 805 u8 traffic_class; 806 }; 807 808 struct ib_grh { 809 __be32 version_tclass_flow; 810 __be16 paylen; 811 u8 next_hdr; 812 u8 hop_limit; 813 union ib_gid sgid; 814 union ib_gid dgid; 815 }; 816 817 union rdma_network_hdr { 818 struct ib_grh ibgrh; 819 struct { 820 /* The IB spec states that if it's IPv4, the header 821 * is located in the last 20 bytes of the header. 822 */ 823 u8 reserved[20]; 824 struct iphdr roce4grh; 825 }; 826 }; 827 828 #define IB_QPN_MASK 0xFFFFFF 829 830 enum { 831 IB_MULTICAST_QPN = 0xffffff 832 }; 833 834 #define IB_LID_PERMISSIVE cpu_to_be16(0xFFFF) 835 #define IB_MULTICAST_LID_BASE cpu_to_be16(0xC000) 836 837 enum ib_ah_flags { 838 IB_AH_GRH = 1 839 }; 840 841 enum ib_rate { 842 IB_RATE_PORT_CURRENT = 0, 843 IB_RATE_2_5_GBPS = 2, 844 IB_RATE_5_GBPS = 5, 845 IB_RATE_10_GBPS = 3, 846 IB_RATE_20_GBPS = 6, 847 IB_RATE_30_GBPS = 4, 848 IB_RATE_40_GBPS = 7, 849 IB_RATE_60_GBPS = 8, 850 IB_RATE_80_GBPS = 9, 851 IB_RATE_120_GBPS = 10, 852 IB_RATE_14_GBPS = 11, 853 IB_RATE_56_GBPS = 12, 854 IB_RATE_112_GBPS = 13, 855 IB_RATE_168_GBPS = 14, 856 IB_RATE_25_GBPS = 15, 857 IB_RATE_100_GBPS = 16, 858 IB_RATE_200_GBPS = 17, 859 IB_RATE_300_GBPS = 18, 860 IB_RATE_28_GBPS = 19, 861 IB_RATE_50_GBPS = 20, 862 IB_RATE_400_GBPS = 21, 863 IB_RATE_600_GBPS = 22, 864 IB_RATE_800_GBPS = 23, 865 IB_RATE_1600_GBPS = 25, 866 }; 867 868 /** 869 * ib_rate_to_mult - Convert the IB rate enum to a multiple of the 870 * base rate of 2.5 Gbit/sec. For example, IB_RATE_5_GBPS will be 871 * converted to 2, since 5 Gbit/sec is 2 * 2.5 Gbit/sec. 872 * @rate: rate to convert. 873 */ 874 __attribute_const__ int ib_rate_to_mult(enum ib_rate rate); 875 876 /** 877 * ib_rate_to_mbps - Convert the IB rate enum to Mbps. 878 * For example, IB_RATE_2_5_GBPS will be converted to 2500. 879 * @rate: rate to convert. 880 */ 881 __attribute_const__ int ib_rate_to_mbps(enum ib_rate rate); 882 883 struct ib_port_speed_info { 884 const char *str; 885 int rate; /* in deci-Gb/sec (100 MBps units) */ 886 }; 887 888 /** 889 * ib_port_attr_to_speed_info - Convert port attributes to speed information 890 * @attr: Port attributes containing active_speed and active_width 891 * @speed_info: Speed information to return 892 * 893 * Returns 0 on success, -EINVAL on error. 894 */ 895 int ib_port_attr_to_speed_info(struct ib_port_attr *attr, 896 struct ib_port_speed_info *speed_info); 897 898 /** 899 * enum ib_mr_type - memory region type 900 * @IB_MR_TYPE_MEM_REG: memory region that is used for 901 * normal registration 902 * @IB_MR_TYPE_SG_GAPS: memory region that is capable to 903 * register any arbitrary sg lists (without 904 * the normal mr constraints - see 905 * ib_map_mr_sg) 906 * @IB_MR_TYPE_DM: memory region that is used for device 907 * memory registration 908 * @IB_MR_TYPE_USER: memory region that is used for the user-space 909 * application 910 * @IB_MR_TYPE_DMA: memory region that is used for DMA operations 911 * without address translations (VA=PA) 912 * @IB_MR_TYPE_INTEGRITY: memory region that is used for 913 * data integrity operations 914 */ 915 enum ib_mr_type { 916 IB_MR_TYPE_MEM_REG, 917 IB_MR_TYPE_SG_GAPS, 918 IB_MR_TYPE_DM, 919 IB_MR_TYPE_USER, 920 IB_MR_TYPE_DMA, 921 IB_MR_TYPE_INTEGRITY, 922 }; 923 924 enum ib_mr_status_check { 925 IB_MR_CHECK_SIG_STATUS = 1, 926 }; 927 928 /** 929 * struct ib_mr_status - Memory region status container 930 * 931 * @fail_status: Bitmask of MR checks status. For each 932 * failed check a corresponding status bit is set. 933 * @sig_err: Additional info for IB_MR_CEHCK_SIG_STATUS 934 * failure. 935 */ 936 struct ib_mr_status { 937 u32 fail_status; 938 struct ib_sig_err sig_err; 939 }; 940 941 /** 942 * mult_to_ib_rate - Convert a multiple of 2.5 Gbit/sec to an IB rate 943 * enum. 944 * @mult: multiple to convert. 945 */ 946 __attribute_const__ enum ib_rate mult_to_ib_rate(int mult); 947 948 struct rdma_ah_init_attr { 949 struct rdma_ah_attr *ah_attr; 950 u32 flags; 951 struct net_device *xmit_slave; 952 }; 953 954 enum rdma_ah_attr_type { 955 RDMA_AH_ATTR_TYPE_UNDEFINED, 956 RDMA_AH_ATTR_TYPE_IB, 957 RDMA_AH_ATTR_TYPE_ROCE, 958 RDMA_AH_ATTR_TYPE_OPA, 959 }; 960 961 struct ib_ah_attr { 962 u16 dlid; 963 u8 src_path_bits; 964 }; 965 966 struct roce_ah_attr { 967 u8 dmac[ETH_ALEN]; 968 }; 969 970 struct opa_ah_attr { 971 u32 dlid; 972 u8 src_path_bits; 973 bool make_grd; 974 }; 975 976 struct rdma_ah_attr { 977 struct ib_global_route grh; 978 u8 sl; 979 u8 static_rate; 980 u32 port_num; 981 u8 ah_flags; 982 enum rdma_ah_attr_type type; 983 union { 984 struct ib_ah_attr ib; 985 struct roce_ah_attr roce; 986 struct opa_ah_attr opa; 987 }; 988 }; 989 990 enum ib_wc_status { 991 IB_WC_SUCCESS, 992 IB_WC_LOC_LEN_ERR, 993 IB_WC_LOC_QP_OP_ERR, 994 IB_WC_LOC_EEC_OP_ERR, 995 IB_WC_LOC_PROT_ERR, 996 IB_WC_WR_FLUSH_ERR, 997 IB_WC_MW_BIND_ERR, 998 IB_WC_BAD_RESP_ERR, 999 IB_WC_LOC_ACCESS_ERR, 1000 IB_WC_REM_INV_REQ_ERR, 1001 IB_WC_REM_ACCESS_ERR, 1002 IB_WC_REM_OP_ERR, 1003 IB_WC_RETRY_EXC_ERR, 1004 IB_WC_RNR_RETRY_EXC_ERR, 1005 IB_WC_LOC_RDD_VIOL_ERR, 1006 IB_WC_REM_INV_RD_REQ_ERR, 1007 IB_WC_REM_ABORT_ERR, 1008 IB_WC_INV_EECN_ERR, 1009 IB_WC_INV_EEC_STATE_ERR, 1010 IB_WC_FATAL_ERR, 1011 IB_WC_RESP_TIMEOUT_ERR, 1012 IB_WC_GENERAL_ERR 1013 }; 1014 1015 const char *__attribute_const__ ib_wc_status_msg(enum ib_wc_status status); 1016 1017 enum ib_wc_opcode { 1018 IB_WC_SEND = IB_UVERBS_WC_SEND, 1019 IB_WC_RDMA_WRITE = IB_UVERBS_WC_RDMA_WRITE, 1020 IB_WC_RDMA_READ = IB_UVERBS_WC_RDMA_READ, 1021 IB_WC_COMP_SWAP = IB_UVERBS_WC_COMP_SWAP, 1022 IB_WC_FETCH_ADD = IB_UVERBS_WC_FETCH_ADD, 1023 IB_WC_BIND_MW = IB_UVERBS_WC_BIND_MW, 1024 IB_WC_LOCAL_INV = IB_UVERBS_WC_LOCAL_INV, 1025 IB_WC_LSO = IB_UVERBS_WC_TSO, 1026 IB_WC_ATOMIC_WRITE = IB_UVERBS_WC_ATOMIC_WRITE, 1027 IB_WC_REG_MR, 1028 IB_WC_MASKED_COMP_SWAP, 1029 IB_WC_MASKED_FETCH_ADD, 1030 IB_WC_FLUSH = IB_UVERBS_WC_FLUSH, 1031 /* 1032 * Set value of IB_WC_RECV so consumers can test if a completion is a 1033 * receive by testing (opcode & IB_WC_RECV). 1034 */ 1035 IB_WC_RECV = 1 << 7, 1036 IB_WC_RECV_RDMA_WITH_IMM 1037 }; 1038 1039 enum ib_wc_flags { 1040 IB_WC_GRH = 1, 1041 IB_WC_WITH_IMM = (1<<1), 1042 IB_WC_WITH_INVALIDATE = (1<<2), 1043 IB_WC_IP_CSUM_OK = (1<<3), 1044 IB_WC_WITH_SMAC = (1<<4), 1045 IB_WC_WITH_VLAN = (1<<5), 1046 IB_WC_WITH_NETWORK_HDR_TYPE = (1<<6), 1047 }; 1048 1049 struct ib_wc { 1050 union { 1051 u64 wr_id; 1052 struct ib_cqe *wr_cqe; 1053 }; 1054 enum ib_wc_status status; 1055 enum ib_wc_opcode opcode; 1056 u32 vendor_err; 1057 u32 byte_len; 1058 struct ib_qp *qp; 1059 union { 1060 __be32 imm_data; 1061 u32 invalidate_rkey; 1062 } ex; 1063 u32 src_qp; 1064 u32 slid; 1065 int wc_flags; 1066 u16 pkey_index; 1067 u8 sl; 1068 u8 dlid_path_bits; 1069 u32 port_num; /* valid only for DR SMPs on switches */ 1070 u8 smac[ETH_ALEN]; 1071 u16 vlan_id; 1072 u8 network_hdr_type; 1073 }; 1074 1075 enum ib_cq_notify_flags { 1076 IB_CQ_SOLICITED = 1 << 0, 1077 IB_CQ_NEXT_COMP = 1 << 1, 1078 IB_CQ_SOLICITED_MASK = IB_CQ_SOLICITED | IB_CQ_NEXT_COMP, 1079 IB_CQ_REPORT_MISSED_EVENTS = 1 << 2, 1080 }; 1081 1082 enum ib_srq_type { 1083 IB_SRQT_BASIC = IB_UVERBS_SRQT_BASIC, 1084 IB_SRQT_XRC = IB_UVERBS_SRQT_XRC, 1085 IB_SRQT_TM = IB_UVERBS_SRQT_TM, 1086 }; 1087 1088 static inline bool ib_srq_has_cq(enum ib_srq_type srq_type) 1089 { 1090 return srq_type == IB_SRQT_XRC || 1091 srq_type == IB_SRQT_TM; 1092 } 1093 1094 enum ib_srq_attr_mask { 1095 IB_SRQ_MAX_WR = 1 << 0, 1096 IB_SRQ_LIMIT = 1 << 1, 1097 }; 1098 1099 struct ib_srq_attr { 1100 u32 max_wr; 1101 u32 max_sge; 1102 u32 srq_limit; 1103 }; 1104 1105 struct ib_srq_init_attr { 1106 void (*event_handler)(struct ib_event *, void *); 1107 void *srq_context; 1108 struct ib_srq_attr attr; 1109 enum ib_srq_type srq_type; 1110 1111 struct { 1112 struct ib_cq *cq; 1113 union { 1114 struct { 1115 struct ib_xrcd *xrcd; 1116 } xrc; 1117 1118 struct { 1119 u32 max_num_tags; 1120 } tag_matching; 1121 }; 1122 } ext; 1123 }; 1124 1125 struct ib_qp_cap { 1126 u32 max_send_wr; 1127 u32 max_recv_wr; 1128 u32 max_send_sge; 1129 u32 max_recv_sge; 1130 u32 max_inline_data; 1131 1132 /* 1133 * Maximum number of rdma_rw_ctx structures in flight at a time. 1134 * ib_create_qp() will calculate the right amount of needed WRs 1135 * and MRs based on this. 1136 */ 1137 u32 max_rdma_ctxs; 1138 }; 1139 1140 enum ib_sig_type { 1141 IB_SIGNAL_ALL_WR, 1142 IB_SIGNAL_REQ_WR 1143 }; 1144 1145 enum ib_qp_type { 1146 /* 1147 * IB_QPT_SMI and IB_QPT_GSI have to be the first two entries 1148 * here (and in that order) since the MAD layer uses them as 1149 * indices into a 2-entry table. 1150 */ 1151 IB_QPT_SMI, 1152 IB_QPT_GSI, 1153 1154 IB_QPT_RC = IB_UVERBS_QPT_RC, 1155 IB_QPT_UC = IB_UVERBS_QPT_UC, 1156 IB_QPT_UD = IB_UVERBS_QPT_UD, 1157 IB_QPT_RAW_IPV6, 1158 IB_QPT_RAW_ETHERTYPE, 1159 IB_QPT_RAW_PACKET = IB_UVERBS_QPT_RAW_PACKET, 1160 IB_QPT_XRC_INI = IB_UVERBS_QPT_XRC_INI, 1161 IB_QPT_XRC_TGT = IB_UVERBS_QPT_XRC_TGT, 1162 IB_QPT_MAX, 1163 IB_QPT_DRIVER = IB_UVERBS_QPT_DRIVER, 1164 /* Reserve a range for qp types internal to the low level driver. 1165 * These qp types will not be visible at the IB core layer, so the 1166 * IB_QPT_MAX usages should not be affected in the core layer 1167 */ 1168 IB_QPT_RESERVED1 = 0x1000, 1169 IB_QPT_RESERVED2, 1170 IB_QPT_RESERVED3, 1171 IB_QPT_RESERVED4, 1172 IB_QPT_RESERVED5, 1173 IB_QPT_RESERVED6, 1174 IB_QPT_RESERVED7, 1175 IB_QPT_RESERVED8, 1176 IB_QPT_RESERVED9, 1177 IB_QPT_RESERVED10, 1178 }; 1179 1180 enum ib_qp_create_flags { 1181 IB_QP_CREATE_IPOIB_UD_LSO = 1 << 0, 1182 IB_QP_CREATE_BLOCK_MULTICAST_LOOPBACK = 1183 IB_UVERBS_QP_CREATE_BLOCK_MULTICAST_LOOPBACK, 1184 IB_QP_CREATE_CROSS_CHANNEL = 1 << 2, 1185 IB_QP_CREATE_MANAGED_SEND = 1 << 3, 1186 IB_QP_CREATE_MANAGED_RECV = 1 << 4, 1187 IB_QP_CREATE_NETIF_QP = 1 << 5, 1188 IB_QP_CREATE_INTEGRITY_EN = 1 << 6, 1189 IB_QP_CREATE_NETDEV_USE = 1 << 7, 1190 IB_QP_CREATE_SCATTER_FCS = 1191 IB_UVERBS_QP_CREATE_SCATTER_FCS, 1192 IB_QP_CREATE_CVLAN_STRIPPING = 1193 IB_UVERBS_QP_CREATE_CVLAN_STRIPPING, 1194 IB_QP_CREATE_SOURCE_QPN = 1 << 10, 1195 IB_QP_CREATE_PCI_WRITE_END_PADDING = 1196 IB_UVERBS_QP_CREATE_PCI_WRITE_END_PADDING, 1197 /* reserve bits 26-31 for low level drivers' internal use */ 1198 IB_QP_CREATE_RESERVED_START = 1 << 26, 1199 IB_QP_CREATE_RESERVED_END = 1 << 31, 1200 }; 1201 1202 /* 1203 * Note: users may not call ib_close_qp or ib_destroy_qp from the event_handler 1204 * callback to destroy the passed in QP. 1205 */ 1206 1207 struct ib_qp_init_attr { 1208 /* This callback occurs in workqueue context */ 1209 void (*event_handler)(struct ib_event *, void *); 1210 1211 void *qp_context; 1212 struct ib_cq *send_cq; 1213 struct ib_cq *recv_cq; 1214 struct ib_srq *srq; 1215 struct ib_xrcd *xrcd; /* XRC TGT QPs only */ 1216 struct ib_qp_cap cap; 1217 enum ib_sig_type sq_sig_type; 1218 enum ib_qp_type qp_type; 1219 u32 create_flags; 1220 1221 /* 1222 * Only needed for special QP types, or when using the RW API. 1223 */ 1224 u32 port_num; 1225 struct ib_rwq_ind_table *rwq_ind_tbl; 1226 u32 source_qpn; 1227 }; 1228 1229 struct ib_qp_open_attr { 1230 void (*event_handler)(struct ib_event *, void *); 1231 void *qp_context; 1232 u32 qp_num; 1233 enum ib_qp_type qp_type; 1234 }; 1235 1236 enum ib_rnr_timeout { 1237 IB_RNR_TIMER_655_36 = 0, 1238 IB_RNR_TIMER_000_01 = 1, 1239 IB_RNR_TIMER_000_02 = 2, 1240 IB_RNR_TIMER_000_03 = 3, 1241 IB_RNR_TIMER_000_04 = 4, 1242 IB_RNR_TIMER_000_06 = 5, 1243 IB_RNR_TIMER_000_08 = 6, 1244 IB_RNR_TIMER_000_12 = 7, 1245 IB_RNR_TIMER_000_16 = 8, 1246 IB_RNR_TIMER_000_24 = 9, 1247 IB_RNR_TIMER_000_32 = 10, 1248 IB_RNR_TIMER_000_48 = 11, 1249 IB_RNR_TIMER_000_64 = 12, 1250 IB_RNR_TIMER_000_96 = 13, 1251 IB_RNR_TIMER_001_28 = 14, 1252 IB_RNR_TIMER_001_92 = 15, 1253 IB_RNR_TIMER_002_56 = 16, 1254 IB_RNR_TIMER_003_84 = 17, 1255 IB_RNR_TIMER_005_12 = 18, 1256 IB_RNR_TIMER_007_68 = 19, 1257 IB_RNR_TIMER_010_24 = 20, 1258 IB_RNR_TIMER_015_36 = 21, 1259 IB_RNR_TIMER_020_48 = 22, 1260 IB_RNR_TIMER_030_72 = 23, 1261 IB_RNR_TIMER_040_96 = 24, 1262 IB_RNR_TIMER_061_44 = 25, 1263 IB_RNR_TIMER_081_92 = 26, 1264 IB_RNR_TIMER_122_88 = 27, 1265 IB_RNR_TIMER_163_84 = 28, 1266 IB_RNR_TIMER_245_76 = 29, 1267 IB_RNR_TIMER_327_68 = 30, 1268 IB_RNR_TIMER_491_52 = 31 1269 }; 1270 1271 enum ib_qp_attr_mask { 1272 IB_QP_STATE = 1, 1273 IB_QP_CUR_STATE = (1<<1), 1274 IB_QP_EN_SQD_ASYNC_NOTIFY = (1<<2), 1275 IB_QP_ACCESS_FLAGS = (1<<3), 1276 IB_QP_PKEY_INDEX = (1<<4), 1277 IB_QP_PORT = (1<<5), 1278 IB_QP_QKEY = (1<<6), 1279 IB_QP_AV = (1<<7), 1280 IB_QP_PATH_MTU = (1<<8), 1281 IB_QP_TIMEOUT = (1<<9), 1282 IB_QP_RETRY_CNT = (1<<10), 1283 IB_QP_RNR_RETRY = (1<<11), 1284 IB_QP_RQ_PSN = (1<<12), 1285 IB_QP_MAX_QP_RD_ATOMIC = (1<<13), 1286 IB_QP_ALT_PATH = (1<<14), 1287 IB_QP_MIN_RNR_TIMER = (1<<15), 1288 IB_QP_SQ_PSN = (1<<16), 1289 IB_QP_MAX_DEST_RD_ATOMIC = (1<<17), 1290 IB_QP_PATH_MIG_STATE = (1<<18), 1291 IB_QP_CAP = (1<<19), 1292 IB_QP_DEST_QPN = (1<<20), 1293 IB_QP_RESERVED1 = (1<<21), 1294 IB_QP_RESERVED2 = (1<<22), 1295 IB_QP_RESERVED3 = (1<<23), 1296 IB_QP_RESERVED4 = (1<<24), 1297 IB_QP_RATE_LIMIT = (1<<25), 1298 1299 IB_QP_ATTR_STANDARD_BITS = GENMASK(20, 0), 1300 }; 1301 1302 enum ib_qp_state { 1303 IB_QPS_RESET, 1304 IB_QPS_INIT, 1305 IB_QPS_RTR, 1306 IB_QPS_RTS, 1307 IB_QPS_SQD, 1308 IB_QPS_SQE, 1309 IB_QPS_ERR 1310 }; 1311 1312 enum ib_mig_state { 1313 IB_MIG_MIGRATED, 1314 IB_MIG_REARM, 1315 IB_MIG_ARMED 1316 }; 1317 1318 enum ib_mw_type { 1319 IB_MW_TYPE_1 = 1, 1320 IB_MW_TYPE_2 = 2 1321 }; 1322 1323 struct ib_qp_attr { 1324 enum ib_qp_state qp_state; 1325 enum ib_qp_state cur_qp_state; 1326 enum ib_mtu path_mtu; 1327 enum ib_mig_state path_mig_state; 1328 u32 qkey; 1329 u32 rq_psn; 1330 u32 sq_psn; 1331 u32 dest_qp_num; 1332 int qp_access_flags; 1333 struct ib_qp_cap cap; 1334 struct rdma_ah_attr ah_attr; 1335 struct rdma_ah_attr alt_ah_attr; 1336 u16 pkey_index; 1337 u16 alt_pkey_index; 1338 u8 en_sqd_async_notify; 1339 u8 sq_draining; 1340 u8 max_rd_atomic; 1341 u8 max_dest_rd_atomic; 1342 u8 min_rnr_timer; 1343 u32 port_num; 1344 u8 timeout; 1345 u8 retry_cnt; 1346 u8 rnr_retry; 1347 u32 alt_port_num; 1348 u8 alt_timeout; 1349 u32 rate_limit; 1350 struct net_device *xmit_slave; 1351 }; 1352 1353 enum ib_wr_opcode { 1354 /* These are shared with userspace */ 1355 IB_WR_RDMA_WRITE = IB_UVERBS_WR_RDMA_WRITE, 1356 IB_WR_RDMA_WRITE_WITH_IMM = IB_UVERBS_WR_RDMA_WRITE_WITH_IMM, 1357 IB_WR_SEND = IB_UVERBS_WR_SEND, 1358 IB_WR_SEND_WITH_IMM = IB_UVERBS_WR_SEND_WITH_IMM, 1359 IB_WR_RDMA_READ = IB_UVERBS_WR_RDMA_READ, 1360 IB_WR_ATOMIC_CMP_AND_SWP = IB_UVERBS_WR_ATOMIC_CMP_AND_SWP, 1361 IB_WR_ATOMIC_FETCH_AND_ADD = IB_UVERBS_WR_ATOMIC_FETCH_AND_ADD, 1362 IB_WR_BIND_MW = IB_UVERBS_WR_BIND_MW, 1363 IB_WR_LSO = IB_UVERBS_WR_TSO, 1364 IB_WR_SEND_WITH_INV = IB_UVERBS_WR_SEND_WITH_INV, 1365 IB_WR_RDMA_READ_WITH_INV = IB_UVERBS_WR_RDMA_READ_WITH_INV, 1366 IB_WR_LOCAL_INV = IB_UVERBS_WR_LOCAL_INV, 1367 IB_WR_MASKED_ATOMIC_CMP_AND_SWP = 1368 IB_UVERBS_WR_MASKED_ATOMIC_CMP_AND_SWP, 1369 IB_WR_MASKED_ATOMIC_FETCH_AND_ADD = 1370 IB_UVERBS_WR_MASKED_ATOMIC_FETCH_AND_ADD, 1371 IB_WR_FLUSH = IB_UVERBS_WR_FLUSH, 1372 IB_WR_ATOMIC_WRITE = IB_UVERBS_WR_ATOMIC_WRITE, 1373 1374 /* These are kernel only and can not be issued by userspace */ 1375 IB_WR_REG_MR = 0x20, 1376 IB_WR_REG_MR_INTEGRITY, 1377 1378 /* reserve values for low level drivers' internal use. 1379 * These values will not be used at all in the ib core layer. 1380 */ 1381 IB_WR_RESERVED1 = 0xf0, 1382 IB_WR_RESERVED2, 1383 IB_WR_RESERVED3, 1384 IB_WR_RESERVED4, 1385 IB_WR_RESERVED5, 1386 IB_WR_RESERVED6, 1387 IB_WR_RESERVED7, 1388 IB_WR_RESERVED8, 1389 IB_WR_RESERVED9, 1390 IB_WR_RESERVED10, 1391 }; 1392 1393 enum ib_send_flags { 1394 IB_SEND_FENCE = 1, 1395 IB_SEND_SIGNALED = (1<<1), 1396 IB_SEND_SOLICITED = (1<<2), 1397 IB_SEND_INLINE = (1<<3), 1398 IB_SEND_IP_CSUM = (1<<4), 1399 1400 /* reserve bits 26-31 for low level drivers' internal use */ 1401 IB_SEND_RESERVED_START = (1 << 26), 1402 IB_SEND_RESERVED_END = (1 << 31), 1403 }; 1404 1405 struct ib_sge { 1406 u64 addr; 1407 u32 length; 1408 u32 lkey; 1409 }; 1410 1411 struct ib_cqe { 1412 void (*done)(struct ib_cq *cq, struct ib_wc *wc); 1413 }; 1414 1415 struct ib_send_wr { 1416 struct ib_send_wr *next; 1417 union { 1418 u64 wr_id; 1419 struct ib_cqe *wr_cqe; 1420 }; 1421 struct ib_sge *sg_list; 1422 int num_sge; 1423 enum ib_wr_opcode opcode; 1424 int send_flags; 1425 union { 1426 __be32 imm_data; 1427 u32 invalidate_rkey; 1428 } ex; 1429 }; 1430 1431 struct ib_rdma_wr { 1432 struct ib_send_wr wr; 1433 u64 remote_addr; 1434 u32 rkey; 1435 }; 1436 1437 static inline const struct ib_rdma_wr *rdma_wr(const struct ib_send_wr *wr) 1438 { 1439 return container_of(wr, struct ib_rdma_wr, wr); 1440 } 1441 1442 struct ib_atomic_wr { 1443 struct ib_send_wr wr; 1444 u64 remote_addr; 1445 u64 compare_add; 1446 u64 swap; 1447 u64 compare_add_mask; 1448 u64 swap_mask; 1449 u32 rkey; 1450 }; 1451 1452 static inline const struct ib_atomic_wr *atomic_wr(const struct ib_send_wr *wr) 1453 { 1454 return container_of(wr, struct ib_atomic_wr, wr); 1455 } 1456 1457 struct ib_ud_wr { 1458 struct ib_send_wr wr; 1459 struct ib_ah *ah; 1460 void *header; 1461 int hlen; 1462 int mss; 1463 u32 remote_qpn; 1464 u32 remote_qkey; 1465 u16 pkey_index; /* valid for GSI only */ 1466 u32 port_num; /* valid for DR SMPs on switch only */ 1467 }; 1468 1469 static inline const struct ib_ud_wr *ud_wr(const struct ib_send_wr *wr) 1470 { 1471 return container_of(wr, struct ib_ud_wr, wr); 1472 } 1473 1474 struct ib_reg_wr { 1475 struct ib_send_wr wr; 1476 struct ib_mr *mr; 1477 u32 key; 1478 int access; 1479 }; 1480 1481 static inline const struct ib_reg_wr *reg_wr(const struct ib_send_wr *wr) 1482 { 1483 return container_of(wr, struct ib_reg_wr, wr); 1484 } 1485 1486 struct ib_recv_wr { 1487 struct ib_recv_wr *next; 1488 union { 1489 u64 wr_id; 1490 struct ib_cqe *wr_cqe; 1491 }; 1492 struct ib_sge *sg_list; 1493 int num_sge; 1494 }; 1495 1496 enum ib_access_flags { 1497 IB_ACCESS_LOCAL_WRITE = IB_UVERBS_ACCESS_LOCAL_WRITE, 1498 IB_ACCESS_REMOTE_WRITE = IB_UVERBS_ACCESS_REMOTE_WRITE, 1499 IB_ACCESS_REMOTE_READ = IB_UVERBS_ACCESS_REMOTE_READ, 1500 IB_ACCESS_REMOTE_ATOMIC = IB_UVERBS_ACCESS_REMOTE_ATOMIC, 1501 IB_ACCESS_MW_BIND = IB_UVERBS_ACCESS_MW_BIND, 1502 IB_ZERO_BASED = IB_UVERBS_ACCESS_ZERO_BASED, 1503 IB_ACCESS_ON_DEMAND = IB_UVERBS_ACCESS_ON_DEMAND, 1504 IB_ACCESS_HUGETLB = IB_UVERBS_ACCESS_HUGETLB, 1505 IB_ACCESS_RELAXED_ORDERING = IB_UVERBS_ACCESS_RELAXED_ORDERING, 1506 IB_ACCESS_FLUSH_GLOBAL = IB_UVERBS_ACCESS_FLUSH_GLOBAL, 1507 IB_ACCESS_FLUSH_PERSISTENT = IB_UVERBS_ACCESS_FLUSH_PERSISTENT, 1508 1509 IB_ACCESS_OPTIONAL = IB_UVERBS_ACCESS_OPTIONAL_RANGE, 1510 IB_ACCESS_SUPPORTED = 1511 ((IB_ACCESS_FLUSH_PERSISTENT << 1) - 1) | IB_ACCESS_OPTIONAL, 1512 }; 1513 1514 /* 1515 * XXX: these are apparently used for ->rereg_user_mr, no idea why they 1516 * are hidden here instead of a uapi header! 1517 */ 1518 enum ib_mr_rereg_flags { 1519 IB_MR_REREG_TRANS = 1, 1520 IB_MR_REREG_PD = (1<<1), 1521 IB_MR_REREG_ACCESS = (1<<2), 1522 IB_MR_REREG_SUPPORTED = ((IB_MR_REREG_ACCESS << 1) - 1) 1523 }; 1524 1525 struct ib_umem; 1526 1527 enum rdma_remove_reason { 1528 /* 1529 * Userspace requested uobject deletion or initial try 1530 * to remove uobject via cleanup. Call could fail 1531 */ 1532 RDMA_REMOVE_DESTROY, 1533 /* Context deletion. This call should delete the actual object itself */ 1534 RDMA_REMOVE_CLOSE, 1535 /* Driver is being hot-unplugged. This call should delete the actual object itself */ 1536 RDMA_REMOVE_DRIVER_REMOVE, 1537 /* uobj is being cleaned-up before being committed */ 1538 RDMA_REMOVE_ABORT, 1539 /* The driver failed to destroy the uobject and is being disconnected */ 1540 RDMA_REMOVE_DRIVER_FAILURE, 1541 }; 1542 1543 struct ib_rdmacg_object { 1544 #ifdef CONFIG_CGROUP_RDMA 1545 struct rdma_cgroup *cg; /* owner rdma cgroup */ 1546 #endif 1547 }; 1548 1549 struct ib_ucontext { 1550 struct ib_device *device; 1551 struct ib_uverbs_file *ufile; 1552 1553 struct ib_rdmacg_object cg_obj; 1554 u64 enabled_caps; 1555 /* 1556 * Implementation details of the RDMA core, don't use in drivers: 1557 */ 1558 struct rdma_restrack_entry res; 1559 struct xarray mmap_xa; 1560 }; 1561 1562 struct ib_uobject { 1563 u64 user_handle; /* handle given to us by userspace */ 1564 /* ufile & ucontext owning this object */ 1565 struct ib_uverbs_file *ufile; 1566 /* FIXME, save memory: ufile->context == context */ 1567 struct ib_ucontext *context; /* associated user context */ 1568 void *object; /* containing object */ 1569 struct list_head list; /* link to context's list */ 1570 struct ib_rdmacg_object cg_obj; /* rdmacg object */ 1571 int id; /* index into kernel idr */ 1572 struct kref ref; 1573 atomic_t usecnt; /* protects exclusive access */ 1574 struct rcu_head rcu; /* kfree_rcu() overhead */ 1575 1576 const struct uverbs_api_object *uapi_object; 1577 }; 1578 1579 struct ib_udata { 1580 const void __user *inbuf; 1581 void __user *outbuf; 1582 size_t inlen; 1583 size_t outlen; 1584 }; 1585 1586 struct ib_pd { 1587 u32 local_dma_lkey; 1588 u32 flags; 1589 struct ib_device *device; 1590 struct ib_uobject *uobject; 1591 atomic_t usecnt; /* count all resources */ 1592 1593 u32 unsafe_global_rkey; 1594 1595 /* 1596 * Implementation details of the RDMA core, don't use in drivers: 1597 */ 1598 struct ib_mr *__internal_mr; 1599 struct rdma_restrack_entry res; 1600 }; 1601 1602 struct ib_xrcd { 1603 struct ib_device *device; 1604 atomic_t usecnt; /* count all exposed resources */ 1605 struct inode *inode; 1606 struct rw_semaphore tgt_qps_rwsem; 1607 struct xarray tgt_qps; 1608 }; 1609 1610 struct ib_ah { 1611 struct ib_device *device; 1612 struct ib_pd *pd; 1613 struct ib_uobject *uobject; 1614 const struct ib_gid_attr *sgid_attr; 1615 enum rdma_ah_attr_type type; 1616 }; 1617 1618 typedef void (*ib_comp_handler)(struct ib_cq *cq, void *cq_context); 1619 1620 enum ib_poll_context { 1621 IB_POLL_SOFTIRQ, /* poll from softirq context */ 1622 IB_POLL_WORKQUEUE, /* poll from workqueue */ 1623 IB_POLL_UNBOUND_WORKQUEUE, /* poll from unbound workqueue */ 1624 IB_POLL_LAST_POOL_TYPE = IB_POLL_UNBOUND_WORKQUEUE, 1625 1626 IB_POLL_DIRECT, /* caller context, no hw completions */ 1627 }; 1628 1629 struct ib_cq { 1630 struct ib_device *device; 1631 struct ib_ucq_object *uobject; 1632 ib_comp_handler comp_handler; 1633 void (*event_handler)(struct ib_event *, void *); 1634 void *cq_context; 1635 int cqe; 1636 unsigned int cqe_used; 1637 atomic_t usecnt; /* count number of work queues */ 1638 enum ib_poll_context poll_ctx; 1639 struct ib_wc *wc; 1640 struct list_head pool_entry; 1641 union { 1642 struct irq_poll iop; 1643 struct work_struct work; 1644 }; 1645 struct workqueue_struct *comp_wq; 1646 struct dim *dim; 1647 1648 /* updated only by trace points */ 1649 ktime_t timestamp; 1650 u8 interrupt:1; 1651 u8 shared:1; 1652 unsigned int comp_vector; 1653 1654 /* 1655 * Implementation details of the RDMA core, don't use in drivers: 1656 */ 1657 struct rdma_restrack_entry res; 1658 }; 1659 1660 struct ib_srq { 1661 struct ib_device *device; 1662 struct ib_pd *pd; 1663 struct ib_usrq_object *uobject; 1664 void (*event_handler)(struct ib_event *, void *); 1665 void *srq_context; 1666 enum ib_srq_type srq_type; 1667 atomic_t usecnt; 1668 1669 struct { 1670 struct ib_cq *cq; 1671 union { 1672 struct { 1673 struct ib_xrcd *xrcd; 1674 u32 srq_num; 1675 } xrc; 1676 }; 1677 } ext; 1678 1679 /* 1680 * Implementation details of the RDMA core, don't use in drivers: 1681 */ 1682 struct rdma_restrack_entry res; 1683 }; 1684 1685 enum ib_raw_packet_caps { 1686 /* 1687 * Strip cvlan from incoming packet and report it in the matching work 1688 * completion is supported. 1689 */ 1690 IB_RAW_PACKET_CAP_CVLAN_STRIPPING = 1691 IB_UVERBS_RAW_PACKET_CAP_CVLAN_STRIPPING, 1692 /* 1693 * Scatter FCS field of an incoming packet to host memory is supported. 1694 */ 1695 IB_RAW_PACKET_CAP_SCATTER_FCS = IB_UVERBS_RAW_PACKET_CAP_SCATTER_FCS, 1696 /* Checksum offloads are supported (for both send and receive). */ 1697 IB_RAW_PACKET_CAP_IP_CSUM = IB_UVERBS_RAW_PACKET_CAP_IP_CSUM, 1698 /* 1699 * When a packet is received for an RQ with no receive WQEs, the 1700 * packet processing is delayed. 1701 */ 1702 IB_RAW_PACKET_CAP_DELAY_DROP = IB_UVERBS_RAW_PACKET_CAP_DELAY_DROP, 1703 }; 1704 1705 enum ib_wq_type { 1706 IB_WQT_RQ = IB_UVERBS_WQT_RQ, 1707 }; 1708 1709 enum ib_wq_state { 1710 IB_WQS_RESET, 1711 IB_WQS_RDY, 1712 IB_WQS_ERR 1713 }; 1714 1715 struct ib_wq { 1716 struct ib_device *device; 1717 struct ib_uwq_object *uobject; 1718 void *wq_context; 1719 void (*event_handler)(struct ib_event *, void *); 1720 struct ib_pd *pd; 1721 struct ib_cq *cq; 1722 u32 wq_num; 1723 enum ib_wq_state state; 1724 enum ib_wq_type wq_type; 1725 atomic_t usecnt; 1726 }; 1727 1728 enum ib_wq_flags { 1729 IB_WQ_FLAGS_CVLAN_STRIPPING = IB_UVERBS_WQ_FLAGS_CVLAN_STRIPPING, 1730 IB_WQ_FLAGS_SCATTER_FCS = IB_UVERBS_WQ_FLAGS_SCATTER_FCS, 1731 IB_WQ_FLAGS_DELAY_DROP = IB_UVERBS_WQ_FLAGS_DELAY_DROP, 1732 IB_WQ_FLAGS_PCI_WRITE_END_PADDING = 1733 IB_UVERBS_WQ_FLAGS_PCI_WRITE_END_PADDING, 1734 }; 1735 1736 struct ib_wq_init_attr { 1737 void *wq_context; 1738 enum ib_wq_type wq_type; 1739 u32 max_wr; 1740 u32 max_sge; 1741 struct ib_cq *cq; 1742 void (*event_handler)(struct ib_event *, void *); 1743 u32 create_flags; /* Use enum ib_wq_flags */ 1744 }; 1745 1746 enum ib_wq_attr_mask { 1747 IB_WQ_STATE = 1 << 0, 1748 IB_WQ_CUR_STATE = 1 << 1, 1749 IB_WQ_FLAGS = 1 << 2, 1750 }; 1751 1752 struct ib_wq_attr { 1753 enum ib_wq_state wq_state; 1754 enum ib_wq_state curr_wq_state; 1755 u32 flags; /* Use enum ib_wq_flags */ 1756 u32 flags_mask; /* Use enum ib_wq_flags */ 1757 }; 1758 1759 struct ib_rwq_ind_table { 1760 struct ib_device *device; 1761 struct ib_uobject *uobject; 1762 atomic_t usecnt; 1763 u32 ind_tbl_num; 1764 u32 log_ind_tbl_size; 1765 struct ib_wq **ind_tbl; 1766 }; 1767 1768 struct ib_rwq_ind_table_init_attr { 1769 u32 log_ind_tbl_size; 1770 /* Each entry is a pointer to Receive Work Queue */ 1771 struct ib_wq **ind_tbl; 1772 }; 1773 1774 enum port_pkey_state { 1775 IB_PORT_PKEY_NOT_VALID = 0, 1776 IB_PORT_PKEY_VALID = 1, 1777 IB_PORT_PKEY_LISTED = 2, 1778 }; 1779 1780 struct ib_qp_security; 1781 1782 struct ib_port_pkey { 1783 enum port_pkey_state state; 1784 u16 pkey_index; 1785 u32 port_num; 1786 struct list_head qp_list; 1787 struct list_head to_error_list; 1788 struct ib_qp_security *sec; 1789 }; 1790 1791 struct ib_ports_pkeys { 1792 struct ib_port_pkey main; 1793 struct ib_port_pkey alt; 1794 }; 1795 1796 struct ib_qp_security { 1797 struct ib_qp *qp; 1798 struct ib_device *dev; 1799 /* Hold this mutex when changing port and pkey settings. */ 1800 struct mutex mutex; 1801 struct ib_ports_pkeys *ports_pkeys; 1802 /* A list of all open shared QP handles. Required to enforce security 1803 * properly for all users of a shared QP. 1804 */ 1805 struct list_head shared_qp_list; 1806 void *security; 1807 bool destroying; 1808 atomic_t error_list_count; 1809 struct completion error_complete; 1810 int error_comps_pending; 1811 }; 1812 1813 /* 1814 * @max_write_sge: Maximum SGE elements per RDMA WRITE request. 1815 * @max_read_sge: Maximum SGE elements per RDMA READ request. 1816 */ 1817 struct ib_qp { 1818 struct ib_device *device; 1819 struct ib_pd *pd; 1820 struct ib_cq *send_cq; 1821 struct ib_cq *recv_cq; 1822 spinlock_t mr_lock; 1823 int mrs_used; 1824 struct list_head rdma_mrs; 1825 struct list_head sig_mrs; 1826 struct ib_srq *srq; 1827 struct completion srq_completion; 1828 struct ib_xrcd *xrcd; /* XRC TGT QPs only */ 1829 struct list_head xrcd_list; 1830 1831 /* count times opened, mcast attaches, flow attaches */ 1832 atomic_t usecnt; 1833 struct list_head open_list; 1834 struct ib_qp *real_qp; 1835 struct ib_uqp_object *uobject; 1836 void (*event_handler)(struct ib_event *, void *); 1837 void (*registered_event_handler)(struct ib_event *, void *); 1838 void *qp_context; 1839 /* sgid_attrs associated with the AV's */ 1840 const struct ib_gid_attr *av_sgid_attr; 1841 const struct ib_gid_attr *alt_path_sgid_attr; 1842 u32 qp_num; 1843 u32 max_write_sge; 1844 u32 max_read_sge; 1845 enum ib_qp_type qp_type; 1846 struct ib_rwq_ind_table *rwq_ind_tbl; 1847 struct ib_qp_security *qp_sec; 1848 u32 port; 1849 1850 bool integrity_en; 1851 /* 1852 * Implementation details of the RDMA core, don't use in drivers: 1853 */ 1854 struct rdma_restrack_entry res; 1855 1856 /* The counter the qp is bind to */ 1857 struct rdma_counter *counter; 1858 }; 1859 1860 struct ib_dm { 1861 struct ib_device *device; 1862 u32 length; 1863 u32 flags; 1864 struct ib_uobject *uobject; 1865 atomic_t usecnt; 1866 }; 1867 1868 /* bit values to mark existence of ib_dmah fields */ 1869 enum { 1870 IB_DMAH_CPU_ID_EXISTS, 1871 IB_DMAH_MEM_TYPE_EXISTS, 1872 IB_DMAH_PH_EXISTS, 1873 }; 1874 1875 struct ib_dmah { 1876 struct ib_device *device; 1877 struct ib_uobject *uobject; 1878 /* 1879 * Implementation details of the RDMA core, don't use in drivers: 1880 */ 1881 struct rdma_restrack_entry res; 1882 u32 cpu_id; 1883 enum tph_mem_type mem_type; 1884 atomic_t usecnt; 1885 u8 ph; 1886 u8 valid_fields; /* use IB_DMAH_XXX_EXISTS */ 1887 }; 1888 1889 struct ib_mr { 1890 struct ib_device *device; 1891 struct ib_pd *pd; 1892 u32 lkey; 1893 u32 rkey; 1894 u64 iova; 1895 u64 length; 1896 unsigned int page_size; 1897 enum ib_mr_type type; 1898 bool need_inval; 1899 union { 1900 struct ib_uobject *uobject; /* user */ 1901 struct list_head qp_entry; /* FR */ 1902 }; 1903 1904 struct ib_dm *dm; 1905 struct ib_sig_attrs *sig_attrs; /* only for IB_MR_TYPE_INTEGRITY MRs */ 1906 struct ib_dmah *dmah; 1907 /* 1908 * Implementation details of the RDMA core, don't use in drivers: 1909 */ 1910 struct rdma_restrack_entry res; 1911 }; 1912 1913 struct ib_mw { 1914 struct ib_device *device; 1915 struct ib_pd *pd; 1916 struct ib_uobject *uobject; 1917 u32 rkey; 1918 enum ib_mw_type type; 1919 }; 1920 1921 /* Supported steering options */ 1922 enum ib_flow_attr_type { 1923 /* steering according to rule specifications */ 1924 IB_FLOW_ATTR_NORMAL = 0x0, 1925 /* default unicast and multicast rule - 1926 * receive all Eth traffic which isn't steered to any QP 1927 */ 1928 IB_FLOW_ATTR_ALL_DEFAULT = 0x1, 1929 /* default multicast rule - 1930 * receive all Eth multicast traffic which isn't steered to any QP 1931 */ 1932 IB_FLOW_ATTR_MC_DEFAULT = 0x2, 1933 /* sniffer rule - receive all port traffic */ 1934 IB_FLOW_ATTR_SNIFFER = 0x3 1935 }; 1936 1937 /* Supported steering header types */ 1938 enum ib_flow_spec_type { 1939 /* L2 headers*/ 1940 IB_FLOW_SPEC_ETH = 0x20, 1941 IB_FLOW_SPEC_IB = 0x22, 1942 /* L3 header*/ 1943 IB_FLOW_SPEC_IPV4 = 0x30, 1944 IB_FLOW_SPEC_IPV6 = 0x31, 1945 IB_FLOW_SPEC_ESP = 0x34, 1946 /* L4 headers*/ 1947 IB_FLOW_SPEC_TCP = 0x40, 1948 IB_FLOW_SPEC_UDP = 0x41, 1949 IB_FLOW_SPEC_VXLAN_TUNNEL = 0x50, 1950 IB_FLOW_SPEC_GRE = 0x51, 1951 IB_FLOW_SPEC_MPLS = 0x60, 1952 IB_FLOW_SPEC_INNER = 0x100, 1953 /* Actions */ 1954 IB_FLOW_SPEC_ACTION_TAG = 0x1000, 1955 IB_FLOW_SPEC_ACTION_DROP = 0x1001, 1956 IB_FLOW_SPEC_ACTION_HANDLE = 0x1002, 1957 IB_FLOW_SPEC_ACTION_COUNT = 0x1003, 1958 }; 1959 #define IB_FLOW_SPEC_LAYER_MASK 0xF0 1960 #define IB_FLOW_SPEC_SUPPORT_LAYERS 10 1961 1962 enum ib_flow_flags { 1963 IB_FLOW_ATTR_FLAGS_DONT_TRAP = 1UL << 1, /* Continue match, no steal */ 1964 IB_FLOW_ATTR_FLAGS_EGRESS = 1UL << 2, /* Egress flow */ 1965 IB_FLOW_ATTR_FLAGS_RESERVED = 1UL << 3 /* Must be last */ 1966 }; 1967 1968 struct ib_flow_eth_filter { 1969 u8 dst_mac[6]; 1970 u8 src_mac[6]; 1971 __be16 ether_type; 1972 __be16 vlan_tag; 1973 }; 1974 1975 struct ib_flow_spec_eth { 1976 u32 type; 1977 u16 size; 1978 struct ib_flow_eth_filter val; 1979 struct ib_flow_eth_filter mask; 1980 }; 1981 1982 struct ib_flow_ib_filter { 1983 __be16 dlid; 1984 __u8 sl; 1985 }; 1986 1987 struct ib_flow_spec_ib { 1988 u32 type; 1989 u16 size; 1990 struct ib_flow_ib_filter val; 1991 struct ib_flow_ib_filter mask; 1992 }; 1993 1994 /* IPv4 header flags */ 1995 enum ib_ipv4_flags { 1996 IB_IPV4_DONT_FRAG = 0x2, /* Don't enable packet fragmentation */ 1997 IB_IPV4_MORE_FRAG = 0X4 /* For All fragmented packets except the 1998 last have this flag set */ 1999 }; 2000 2001 struct ib_flow_ipv4_filter { 2002 __be32 src_ip; 2003 __be32 dst_ip; 2004 u8 proto; 2005 u8 tos; 2006 u8 ttl; 2007 u8 flags; 2008 }; 2009 2010 struct ib_flow_spec_ipv4 { 2011 u32 type; 2012 u16 size; 2013 struct ib_flow_ipv4_filter val; 2014 struct ib_flow_ipv4_filter mask; 2015 }; 2016 2017 struct ib_flow_ipv6_filter { 2018 u8 src_ip[16]; 2019 u8 dst_ip[16]; 2020 __be32 flow_label; 2021 u8 next_hdr; 2022 u8 traffic_class; 2023 u8 hop_limit; 2024 } __packed; 2025 2026 struct ib_flow_spec_ipv6 { 2027 u32 type; 2028 u16 size; 2029 struct ib_flow_ipv6_filter val; 2030 struct ib_flow_ipv6_filter mask; 2031 }; 2032 2033 struct ib_flow_tcp_udp_filter { 2034 __be16 dst_port; 2035 __be16 src_port; 2036 }; 2037 2038 struct ib_flow_spec_tcp_udp { 2039 u32 type; 2040 u16 size; 2041 struct ib_flow_tcp_udp_filter val; 2042 struct ib_flow_tcp_udp_filter mask; 2043 }; 2044 2045 struct ib_flow_tunnel_filter { 2046 __be32 tunnel_id; 2047 }; 2048 2049 /* ib_flow_spec_tunnel describes the Vxlan tunnel 2050 * the tunnel_id from val has the vni value 2051 */ 2052 struct ib_flow_spec_tunnel { 2053 u32 type; 2054 u16 size; 2055 struct ib_flow_tunnel_filter val; 2056 struct ib_flow_tunnel_filter mask; 2057 }; 2058 2059 struct ib_flow_esp_filter { 2060 __be32 spi; 2061 __be32 seq; 2062 }; 2063 2064 struct ib_flow_spec_esp { 2065 u32 type; 2066 u16 size; 2067 struct ib_flow_esp_filter val; 2068 struct ib_flow_esp_filter mask; 2069 }; 2070 2071 struct ib_flow_gre_filter { 2072 __be16 c_ks_res0_ver; 2073 __be16 protocol; 2074 __be32 key; 2075 }; 2076 2077 struct ib_flow_spec_gre { 2078 u32 type; 2079 u16 size; 2080 struct ib_flow_gre_filter val; 2081 struct ib_flow_gre_filter mask; 2082 }; 2083 2084 struct ib_flow_mpls_filter { 2085 __be32 tag; 2086 }; 2087 2088 struct ib_flow_spec_mpls { 2089 u32 type; 2090 u16 size; 2091 struct ib_flow_mpls_filter val; 2092 struct ib_flow_mpls_filter mask; 2093 }; 2094 2095 struct ib_flow_spec_action_tag { 2096 enum ib_flow_spec_type type; 2097 u16 size; 2098 u32 tag_id; 2099 }; 2100 2101 struct ib_flow_spec_action_drop { 2102 enum ib_flow_spec_type type; 2103 u16 size; 2104 }; 2105 2106 struct ib_flow_spec_action_handle { 2107 enum ib_flow_spec_type type; 2108 u16 size; 2109 struct ib_flow_action *act; 2110 }; 2111 2112 enum ib_counters_description { 2113 IB_COUNTER_PACKETS, 2114 IB_COUNTER_BYTES, 2115 }; 2116 2117 struct ib_flow_spec_action_count { 2118 enum ib_flow_spec_type type; 2119 u16 size; 2120 struct ib_counters *counters; 2121 }; 2122 2123 union ib_flow_spec { 2124 struct { 2125 u32 type; 2126 u16 size; 2127 }; 2128 struct ib_flow_spec_eth eth; 2129 struct ib_flow_spec_ib ib; 2130 struct ib_flow_spec_ipv4 ipv4; 2131 struct ib_flow_spec_tcp_udp tcp_udp; 2132 struct ib_flow_spec_ipv6 ipv6; 2133 struct ib_flow_spec_tunnel tunnel; 2134 struct ib_flow_spec_esp esp; 2135 struct ib_flow_spec_gre gre; 2136 struct ib_flow_spec_mpls mpls; 2137 struct ib_flow_spec_action_tag flow_tag; 2138 struct ib_flow_spec_action_drop drop; 2139 struct ib_flow_spec_action_handle action; 2140 struct ib_flow_spec_action_count flow_count; 2141 }; 2142 2143 struct ib_flow_attr { 2144 enum ib_flow_attr_type type; 2145 u16 size; 2146 u16 priority; 2147 u32 flags; 2148 u8 num_of_specs; 2149 u32 port; 2150 union ib_flow_spec flows[]; 2151 }; 2152 2153 struct ib_flow { 2154 struct ib_qp *qp; 2155 struct ib_device *device; 2156 struct ib_uobject *uobject; 2157 }; 2158 2159 enum ib_flow_action_type { 2160 IB_FLOW_ACTION_UNSPECIFIED, 2161 IB_FLOW_ACTION_ESP = 1, 2162 }; 2163 2164 struct ib_flow_action_attrs_esp_keymats { 2165 enum ib_uverbs_flow_action_esp_keymat protocol; 2166 union { 2167 struct ib_uverbs_flow_action_esp_keymat_aes_gcm aes_gcm; 2168 } keymat; 2169 }; 2170 2171 struct ib_flow_action_attrs_esp_replays { 2172 enum ib_uverbs_flow_action_esp_replay protocol; 2173 union { 2174 struct ib_uverbs_flow_action_esp_replay_bmp bmp; 2175 } replay; 2176 }; 2177 2178 enum ib_flow_action_attrs_esp_flags { 2179 /* All user-space flags at the top: Use enum ib_uverbs_flow_action_esp_flags 2180 * This is done in order to share the same flags between user-space and 2181 * kernel and spare an unnecessary translation. 2182 */ 2183 2184 /* Kernel flags */ 2185 IB_FLOW_ACTION_ESP_FLAGS_ESN_TRIGGERED = 1ULL << 32, 2186 IB_FLOW_ACTION_ESP_FLAGS_MOD_ESP_ATTRS = 1ULL << 33, 2187 }; 2188 2189 struct ib_flow_spec_list { 2190 struct ib_flow_spec_list *next; 2191 union ib_flow_spec spec; 2192 }; 2193 2194 struct ib_flow_action_attrs_esp { 2195 struct ib_flow_action_attrs_esp_keymats *keymat; 2196 struct ib_flow_action_attrs_esp_replays *replay; 2197 struct ib_flow_spec_list *encap; 2198 /* Used only if IB_FLOW_ACTION_ESP_FLAGS_ESN_TRIGGERED is enabled. 2199 * Value of 0 is a valid value. 2200 */ 2201 u32 esn; 2202 u32 spi; 2203 u32 seq; 2204 u32 tfc_pad; 2205 /* Use enum ib_flow_action_attrs_esp_flags */ 2206 u64 flags; 2207 u64 hard_limit_pkts; 2208 }; 2209 2210 struct ib_flow_action { 2211 struct ib_device *device; 2212 struct ib_uobject *uobject; 2213 enum ib_flow_action_type type; 2214 atomic_t usecnt; 2215 }; 2216 2217 struct ib_mad; 2218 2219 enum ib_process_mad_flags { 2220 IB_MAD_IGNORE_MKEY = 1, 2221 IB_MAD_IGNORE_BKEY = 2, 2222 IB_MAD_IGNORE_ALL = IB_MAD_IGNORE_MKEY | IB_MAD_IGNORE_BKEY 2223 }; 2224 2225 enum ib_mad_result { 2226 IB_MAD_RESULT_FAILURE = 0, /* (!SUCCESS is the important flag) */ 2227 IB_MAD_RESULT_SUCCESS = 1 << 0, /* MAD was successfully processed */ 2228 IB_MAD_RESULT_REPLY = 1 << 1, /* Reply packet needs to be sent */ 2229 IB_MAD_RESULT_CONSUMED = 1 << 2 /* Packet consumed: stop processing */ 2230 }; 2231 2232 struct ib_port_cache { 2233 u64 subnet_prefix; 2234 struct ib_pkey_cache *pkey; 2235 struct ib_gid_table *gid; 2236 u8 lmc; 2237 enum ib_port_state port_state; 2238 enum ib_port_state last_port_state; 2239 }; 2240 2241 struct ib_port_immutable { 2242 int pkey_tbl_len; 2243 int gid_tbl_len; 2244 u32 core_cap_flags; 2245 u32 max_mad_size; 2246 }; 2247 2248 struct ib_port_data { 2249 struct ib_device *ib_dev; 2250 2251 struct ib_port_immutable immutable; 2252 2253 spinlock_t pkey_list_lock; 2254 2255 spinlock_t netdev_lock; 2256 2257 struct list_head pkey_list; 2258 2259 struct ib_port_cache cache; 2260 2261 struct net_device __rcu *netdev; 2262 netdevice_tracker netdev_tracker; 2263 struct hlist_node ndev_hash_link; 2264 struct rdma_port_counter port_counter; 2265 struct ib_port *sysfs; 2266 }; 2267 2268 /* rdma netdev type - specifies protocol type */ 2269 enum rdma_netdev_t { 2270 RDMA_NETDEV_OPA_VNIC, 2271 RDMA_NETDEV_IPOIB, 2272 }; 2273 2274 /** 2275 * struct rdma_netdev - rdma netdev 2276 * For cases where netstack interfacing is required. 2277 */ 2278 struct rdma_netdev { 2279 void *clnt_priv; 2280 struct ib_device *hca; 2281 u32 port_num; 2282 int mtu; 2283 2284 /* 2285 * cleanup function must be specified. 2286 * FIXME: This is only used for OPA_VNIC and that usage should be 2287 * removed too. 2288 */ 2289 void (*free_rdma_netdev)(struct net_device *netdev); 2290 2291 /* control functions */ 2292 void (*set_id)(struct net_device *netdev, int id); 2293 /* send packet */ 2294 int (*send)(struct net_device *dev, struct sk_buff *skb, 2295 struct ib_ah *address, u32 dqpn); 2296 /* multicast */ 2297 int (*attach_mcast)(struct net_device *dev, struct ib_device *hca, 2298 union ib_gid *gid, u16 mlid, 2299 int set_qkey, u32 qkey); 2300 int (*detach_mcast)(struct net_device *dev, struct ib_device *hca, 2301 union ib_gid *gid, u16 mlid); 2302 /* timeout */ 2303 void (*tx_timeout)(struct net_device *dev, unsigned int txqueue); 2304 }; 2305 2306 struct rdma_netdev_alloc_params { 2307 size_t sizeof_priv; 2308 unsigned int txqs; 2309 unsigned int rxqs; 2310 void *param; 2311 2312 int (*initialize_rdma_netdev)(struct ib_device *device, u32 port_num, 2313 struct net_device *netdev, void *param); 2314 }; 2315 2316 struct ib_odp_counters { 2317 atomic64_t faults; 2318 atomic64_t faults_handled; 2319 atomic64_t invalidations; 2320 atomic64_t invalidations_handled; 2321 atomic64_t prefetch; 2322 }; 2323 2324 struct ib_counters { 2325 struct ib_device *device; 2326 struct ib_uobject *uobject; 2327 /* num of objects attached */ 2328 atomic_t usecnt; 2329 }; 2330 2331 struct ib_counters_read_attr { 2332 u64 *counters_buff; 2333 u32 ncounters; 2334 u32 flags; /* use enum ib_read_counters_flags */ 2335 }; 2336 2337 struct uverbs_attr_bundle; 2338 struct iw_cm_id; 2339 struct iw_cm_conn_param; 2340 2341 #define INIT_RDMA_OBJ_SIZE(ib_struct, drv_struct, member) \ 2342 .size_##ib_struct = \ 2343 (sizeof(struct drv_struct) + \ 2344 BUILD_BUG_ON_ZERO(offsetof(struct drv_struct, member)) + \ 2345 BUILD_BUG_ON_ZERO( \ 2346 !__same_type(((struct drv_struct *)NULL)->member, \ 2347 struct ib_struct))) 2348 2349 #define rdma_zalloc_drv_obj_gfp(ib_dev, ib_type, gfp) \ 2350 ((struct ib_type *)rdma_zalloc_obj(ib_dev, ib_dev->ops.size_##ib_type, \ 2351 gfp, false)) 2352 2353 #define rdma_zalloc_drv_obj_numa(ib_dev, ib_type) \ 2354 ((struct ib_type *)rdma_zalloc_obj(ib_dev, ib_dev->ops.size_##ib_type, \ 2355 GFP_KERNEL, true)) 2356 2357 #define rdma_zalloc_drv_obj(ib_dev, ib_type) \ 2358 rdma_zalloc_drv_obj_gfp(ib_dev, ib_type, GFP_KERNEL) 2359 2360 #define DECLARE_RDMA_OBJ_SIZE(ib_struct) size_t size_##ib_struct 2361 2362 struct rdma_user_mmap_entry { 2363 struct kref ref; 2364 struct ib_ucontext *ucontext; 2365 unsigned long start_pgoff; 2366 size_t npages; 2367 bool driver_removed; 2368 /* protects access to dmabufs */ 2369 struct mutex dmabufs_lock; 2370 struct list_head dmabufs; 2371 }; 2372 2373 /* Return the offset (in bytes) the user should pass to libc's mmap() */ 2374 static inline u64 2375 rdma_user_mmap_get_offset(const struct rdma_user_mmap_entry *entry) 2376 { 2377 return (u64)entry->start_pgoff << PAGE_SHIFT; 2378 } 2379 2380 /** 2381 * struct ib_device_ops - InfiniBand device operations 2382 * This structure defines all the InfiniBand device operations, providers will 2383 * need to define the supported operations, otherwise they will be set to null. 2384 */ 2385 struct ib_device_ops { 2386 struct module *owner; 2387 enum rdma_driver_id driver_id; 2388 u32 uverbs_abi_ver; 2389 unsigned int uverbs_no_driver_id_binding:1; 2390 2391 /* 2392 * NOTE: New drivers should not make use of device_group; instead new 2393 * device parameter should be exposed via netlink command. This 2394 * mechanism exists only for existing drivers. 2395 */ 2396 const struct attribute_group *device_group; 2397 const struct attribute_group **port_groups; 2398 2399 int (*post_send)(struct ib_qp *qp, const struct ib_send_wr *send_wr, 2400 const struct ib_send_wr **bad_send_wr); 2401 int (*post_recv)(struct ib_qp *qp, const struct ib_recv_wr *recv_wr, 2402 const struct ib_recv_wr **bad_recv_wr); 2403 void (*drain_rq)(struct ib_qp *qp); 2404 void (*drain_sq)(struct ib_qp *qp); 2405 int (*poll_cq)(struct ib_cq *cq, int num_entries, struct ib_wc *wc); 2406 int (*peek_cq)(struct ib_cq *cq, int wc_cnt); 2407 int (*req_notify_cq)(struct ib_cq *cq, enum ib_cq_notify_flags flags); 2408 int (*post_srq_recv)(struct ib_srq *srq, 2409 const struct ib_recv_wr *recv_wr, 2410 const struct ib_recv_wr **bad_recv_wr); 2411 int (*process_mad)(struct ib_device *device, int process_mad_flags, 2412 u32 port_num, const struct ib_wc *in_wc, 2413 const struct ib_grh *in_grh, 2414 const struct ib_mad *in_mad, struct ib_mad *out_mad, 2415 size_t *out_mad_size, u16 *out_mad_pkey_index); 2416 int (*query_device)(struct ib_device *device, 2417 struct ib_device_attr *device_attr, 2418 struct ib_udata *udata); 2419 int (*modify_device)(struct ib_device *device, int device_modify_mask, 2420 struct ib_device_modify *device_modify); 2421 void (*get_dev_fw_str)(struct ib_device *device, char *str); 2422 const struct cpumask *(*get_vector_affinity)(struct ib_device *ibdev, 2423 int comp_vector); 2424 int (*query_port)(struct ib_device *device, u32 port_num, 2425 struct ib_port_attr *port_attr); 2426 int (*query_port_speed)(struct ib_device *device, u32 port_num, 2427 u64 *speed); 2428 int (*modify_port)(struct ib_device *device, u32 port_num, 2429 int port_modify_mask, 2430 struct ib_port_modify *port_modify); 2431 /* 2432 * The following mandatory functions are used only at device 2433 * registration. Keep functions such as these at the end of this 2434 * structure to avoid cache line misses when accessing struct ib_device 2435 * in fast paths. 2436 */ 2437 int (*get_port_immutable)(struct ib_device *device, u32 port_num, 2438 struct ib_port_immutable *immutable); 2439 enum rdma_link_layer (*get_link_layer)(struct ib_device *device, 2440 u32 port_num); 2441 /* 2442 * When calling get_netdev, the HW vendor's driver should return the 2443 * net device of device @device at port @port_num or NULL if such 2444 * a net device doesn't exist. The vendor driver should call dev_hold 2445 * on this net device. The HW vendor's device driver must guarantee 2446 * that this function returns NULL before the net device has finished 2447 * NETDEV_UNREGISTER state. 2448 */ 2449 struct net_device *(*get_netdev)(struct ib_device *device, 2450 u32 port_num); 2451 /* 2452 * rdma netdev operation 2453 * 2454 * Driver implementing alloc_rdma_netdev or rdma_netdev_get_params 2455 * must return -EOPNOTSUPP if it doesn't support the specified type. 2456 */ 2457 struct net_device *(*alloc_rdma_netdev)( 2458 struct ib_device *device, u32 port_num, enum rdma_netdev_t type, 2459 const char *name, unsigned char name_assign_type, 2460 void (*setup)(struct net_device *)); 2461 2462 int (*rdma_netdev_get_params)(struct ib_device *device, u32 port_num, 2463 enum rdma_netdev_t type, 2464 struct rdma_netdev_alloc_params *params); 2465 /* 2466 * query_gid should be return GID value for @device, when @port_num 2467 * link layer is either IB or iWarp. It is no-op if @port_num port 2468 * is RoCE link layer. 2469 */ 2470 int (*query_gid)(struct ib_device *device, u32 port_num, int index, 2471 union ib_gid *gid); 2472 /* 2473 * When calling add_gid, the HW vendor's driver should add the gid 2474 * of device of port at gid index available at @attr. Meta-info of 2475 * that gid (for example, the network device related to this gid) is 2476 * available at @attr. @context allows the HW vendor driver to store 2477 * extra information together with a GID entry. The HW vendor driver may 2478 * allocate memory to contain this information and store it in @context 2479 * when a new GID entry is written to. Params are consistent until the 2480 * next call of add_gid or delete_gid. The function should return 0 on 2481 * success or error otherwise. The function could be called 2482 * concurrently for different ports. This function is only called when 2483 * roce_gid_table is used. 2484 */ 2485 int (*add_gid)(const struct ib_gid_attr *attr, void **context); 2486 /* 2487 * When calling del_gid, the HW vendor's driver should delete the 2488 * gid of device @device at gid index gid_index of port port_num 2489 * available in @attr. 2490 * Upon the deletion of a GID entry, the HW vendor must free any 2491 * allocated memory. The caller will clear @context afterwards. 2492 * This function is only called when roce_gid_table is used. 2493 */ 2494 int (*del_gid)(const struct ib_gid_attr *attr, void **context); 2495 int (*query_pkey)(struct ib_device *device, u32 port_num, u16 index, 2496 u16 *pkey); 2497 int (*alloc_ucontext)(struct ib_ucontext *context, 2498 struct ib_udata *udata); 2499 void (*dealloc_ucontext)(struct ib_ucontext *context); 2500 int (*mmap)(struct ib_ucontext *context, struct vm_area_struct *vma); 2501 /* 2502 * This will be called once refcount of an entry in mmap_xa reaches 2503 * zero. The type of the memory that was mapped may differ between 2504 * entries and is opaque to the rdma_user_mmap interface. 2505 * Therefore needs to be implemented by the driver in mmap_free. 2506 */ 2507 void (*mmap_free)(struct rdma_user_mmap_entry *entry); 2508 int (*mmap_get_pfns)(struct rdma_user_mmap_entry *entry, 2509 struct phys_vec *phys_vec, 2510 struct p2pdma_provider **provider); 2511 struct rdma_user_mmap_entry *(*pgoff_to_mmap_entry)(struct ib_ucontext *ucontext, 2512 off_t pg_off); 2513 void (*disassociate_ucontext)(struct ib_ucontext *ibcontext); 2514 int (*alloc_pd)(struct ib_pd *pd, struct ib_udata *udata); 2515 int (*dealloc_pd)(struct ib_pd *pd, struct ib_udata *udata); 2516 int (*create_ah)(struct ib_ah *ah, struct rdma_ah_init_attr *attr, 2517 struct ib_udata *udata); 2518 int (*create_user_ah)(struct ib_ah *ah, struct rdma_ah_init_attr *attr, 2519 struct ib_udata *udata); 2520 int (*modify_ah)(struct ib_ah *ah, struct rdma_ah_attr *ah_attr); 2521 int (*query_ah)(struct ib_ah *ah, struct rdma_ah_attr *ah_attr); 2522 int (*destroy_ah)(struct ib_ah *ah, u32 flags); 2523 int (*create_srq)(struct ib_srq *srq, 2524 struct ib_srq_init_attr *srq_init_attr, 2525 struct ib_udata *udata); 2526 int (*modify_srq)(struct ib_srq *srq, struct ib_srq_attr *srq_attr, 2527 enum ib_srq_attr_mask srq_attr_mask, 2528 struct ib_udata *udata); 2529 int (*query_srq)(struct ib_srq *srq, struct ib_srq_attr *srq_attr); 2530 int (*destroy_srq)(struct ib_srq *srq, struct ib_udata *udata); 2531 int (*create_qp)(struct ib_qp *qp, struct ib_qp_init_attr *qp_init_attr, 2532 struct ib_udata *udata); 2533 int (*modify_qp)(struct ib_qp *qp, struct ib_qp_attr *qp_attr, 2534 int qp_attr_mask, struct ib_udata *udata); 2535 int (*query_qp)(struct ib_qp *qp, struct ib_qp_attr *qp_attr, 2536 int qp_attr_mask, struct ib_qp_init_attr *qp_init_attr); 2537 int (*destroy_qp)(struct ib_qp *qp, struct ib_udata *udata); 2538 int (*create_cq)(struct ib_cq *cq, const struct ib_cq_init_attr *attr, 2539 struct uverbs_attr_bundle *attrs); 2540 int (*create_cq_umem)(struct ib_cq *cq, 2541 const struct ib_cq_init_attr *attr, 2542 struct ib_umem *umem, 2543 struct uverbs_attr_bundle *attrs); 2544 int (*modify_cq)(struct ib_cq *cq, u16 cq_count, u16 cq_period); 2545 int (*destroy_cq)(struct ib_cq *cq, struct ib_udata *udata); 2546 int (*resize_cq)(struct ib_cq *cq, int cqe, struct ib_udata *udata); 2547 /* 2548 * pre_destroy_cq - Prevent a cq from generating any new work 2549 * completions, but not free any kernel resources 2550 */ 2551 int (*pre_destroy_cq)(struct ib_cq *cq); 2552 /* 2553 * post_destroy_cq - Free all kernel resources 2554 */ 2555 void (*post_destroy_cq)(struct ib_cq *cq); 2556 struct ib_mr *(*get_dma_mr)(struct ib_pd *pd, int mr_access_flags); 2557 struct ib_mr *(*reg_user_mr)(struct ib_pd *pd, u64 start, u64 length, 2558 u64 virt_addr, int mr_access_flags, 2559 struct ib_dmah *dmah, 2560 struct ib_udata *udata); 2561 struct ib_mr *(*reg_user_mr_dmabuf)(struct ib_pd *pd, u64 offset, 2562 u64 length, u64 virt_addr, int fd, 2563 int mr_access_flags, 2564 struct ib_dmah *dmah, 2565 struct uverbs_attr_bundle *attrs); 2566 struct ib_mr *(*rereg_user_mr)(struct ib_mr *mr, int flags, u64 start, 2567 u64 length, u64 virt_addr, 2568 int mr_access_flags, struct ib_pd *pd, 2569 struct ib_udata *udata); 2570 int (*dereg_mr)(struct ib_mr *mr, struct ib_udata *udata); 2571 struct ib_mr *(*alloc_mr)(struct ib_pd *pd, enum ib_mr_type mr_type, 2572 u32 max_num_sg); 2573 struct ib_mr *(*alloc_mr_integrity)(struct ib_pd *pd, 2574 u32 max_num_data_sg, 2575 u32 max_num_meta_sg); 2576 int (*advise_mr)(struct ib_pd *pd, 2577 enum ib_uverbs_advise_mr_advice advice, u32 flags, 2578 struct ib_sge *sg_list, u32 num_sge, 2579 struct uverbs_attr_bundle *attrs); 2580 2581 /* 2582 * Kernel users should universally support relaxed ordering (RO), as 2583 * they are designed to read data only after observing the CQE and use 2584 * the DMA API correctly. 2585 * 2586 * Some drivers implicitly enable RO if platform supports it. 2587 */ 2588 int (*map_mr_sg)(struct ib_mr *mr, struct scatterlist *sg, int sg_nents, 2589 unsigned int *sg_offset); 2590 int (*check_mr_status)(struct ib_mr *mr, u32 check_mask, 2591 struct ib_mr_status *mr_status); 2592 int (*alloc_mw)(struct ib_mw *mw, struct ib_udata *udata); 2593 int (*dealloc_mw)(struct ib_mw *mw); 2594 int (*attach_mcast)(struct ib_qp *qp, union ib_gid *gid, u16 lid); 2595 int (*detach_mcast)(struct ib_qp *qp, union ib_gid *gid, u16 lid); 2596 int (*alloc_xrcd)(struct ib_xrcd *xrcd, struct ib_udata *udata); 2597 int (*dealloc_xrcd)(struct ib_xrcd *xrcd, struct ib_udata *udata); 2598 struct ib_flow *(*create_flow)(struct ib_qp *qp, 2599 struct ib_flow_attr *flow_attr, 2600 struct ib_udata *udata); 2601 int (*destroy_flow)(struct ib_flow *flow_id); 2602 int (*destroy_flow_action)(struct ib_flow_action *action); 2603 int (*set_vf_link_state)(struct ib_device *device, int vf, u32 port, 2604 int state); 2605 int (*get_vf_config)(struct ib_device *device, int vf, u32 port, 2606 struct ifla_vf_info *ivf); 2607 int (*get_vf_stats)(struct ib_device *device, int vf, u32 port, 2608 struct ifla_vf_stats *stats); 2609 int (*get_vf_guid)(struct ib_device *device, int vf, u32 port, 2610 struct ifla_vf_guid *node_guid, 2611 struct ifla_vf_guid *port_guid); 2612 int (*set_vf_guid)(struct ib_device *device, int vf, u32 port, u64 guid, 2613 int type); 2614 struct ib_wq *(*create_wq)(struct ib_pd *pd, 2615 struct ib_wq_init_attr *init_attr, 2616 struct ib_udata *udata); 2617 int (*destroy_wq)(struct ib_wq *wq, struct ib_udata *udata); 2618 int (*modify_wq)(struct ib_wq *wq, struct ib_wq_attr *attr, 2619 u32 wq_attr_mask, struct ib_udata *udata); 2620 int (*create_rwq_ind_table)(struct ib_rwq_ind_table *ib_rwq_ind_table, 2621 struct ib_rwq_ind_table_init_attr *init_attr, 2622 struct ib_udata *udata); 2623 int (*destroy_rwq_ind_table)(struct ib_rwq_ind_table *wq_ind_table); 2624 struct ib_dm *(*alloc_dm)(struct ib_device *device, 2625 struct ib_ucontext *context, 2626 struct ib_dm_alloc_attr *attr, 2627 struct uverbs_attr_bundle *attrs); 2628 int (*dealloc_dm)(struct ib_dm *dm, struct uverbs_attr_bundle *attrs); 2629 int (*alloc_dmah)(struct ib_dmah *ibdmah, 2630 struct uverbs_attr_bundle *attrs); 2631 int (*dealloc_dmah)(struct ib_dmah *dmah, struct uverbs_attr_bundle *attrs); 2632 struct ib_mr *(*reg_dm_mr)(struct ib_pd *pd, struct ib_dm *dm, 2633 struct ib_dm_mr_attr *attr, 2634 struct uverbs_attr_bundle *attrs); 2635 int (*create_counters)(struct ib_counters *counters, 2636 struct uverbs_attr_bundle *attrs); 2637 int (*destroy_counters)(struct ib_counters *counters); 2638 int (*read_counters)(struct ib_counters *counters, 2639 struct ib_counters_read_attr *counters_read_attr, 2640 struct uverbs_attr_bundle *attrs); 2641 int (*map_mr_sg_pi)(struct ib_mr *mr, struct scatterlist *data_sg, 2642 int data_sg_nents, unsigned int *data_sg_offset, 2643 struct scatterlist *meta_sg, int meta_sg_nents, 2644 unsigned int *meta_sg_offset); 2645 2646 /* 2647 * alloc_hw_[device,port]_stats - Allocate a struct rdma_hw_stats and 2648 * fill in the driver initialized data. The struct is kfree()'ed by 2649 * the sysfs core when the device is removed. A lifespan of -1 in the 2650 * return struct tells the core to set a default lifespan. 2651 */ 2652 struct rdma_hw_stats *(*alloc_hw_device_stats)(struct ib_device *device); 2653 struct rdma_hw_stats *(*alloc_hw_port_stats)(struct ib_device *device, 2654 u32 port_num); 2655 /* 2656 * get_hw_stats - Fill in the counter value(s) in the stats struct. 2657 * @index - The index in the value array we wish to have updated, or 2658 * num_counters if we want all stats updated 2659 * Return codes - 2660 * < 0 - Error, no counters updated 2661 * index - Updated the single counter pointed to by index 2662 * num_counters - Updated all counters (will reset the timestamp 2663 * and prevent further calls for lifespan milliseconds) 2664 * Drivers are allowed to update all counters in leiu of just the 2665 * one given in index at their option 2666 */ 2667 int (*get_hw_stats)(struct ib_device *device, 2668 struct rdma_hw_stats *stats, u32 port, int index); 2669 2670 /* 2671 * modify_hw_stat - Modify the counter configuration 2672 * @enable: true/false when enable/disable a counter 2673 * Return codes - 0 on success or error code otherwise. 2674 */ 2675 int (*modify_hw_stat)(struct ib_device *device, u32 port, 2676 unsigned int counter_index, bool enable); 2677 /* 2678 * Allows rdma drivers to add their own restrack attributes. 2679 */ 2680 int (*fill_res_mr_entry)(struct sk_buff *msg, struct ib_mr *ibmr); 2681 int (*fill_res_mr_entry_raw)(struct sk_buff *msg, struct ib_mr *ibmr); 2682 int (*fill_res_cq_entry)(struct sk_buff *msg, struct ib_cq *ibcq); 2683 int (*fill_res_cq_entry_raw)(struct sk_buff *msg, struct ib_cq *ibcq); 2684 int (*fill_res_qp_entry)(struct sk_buff *msg, struct ib_qp *ibqp); 2685 int (*fill_res_qp_entry_raw)(struct sk_buff *msg, struct ib_qp *ibqp); 2686 int (*fill_res_cm_id_entry)(struct sk_buff *msg, struct rdma_cm_id *id); 2687 int (*fill_res_srq_entry)(struct sk_buff *msg, struct ib_srq *ib_srq); 2688 int (*fill_res_srq_entry_raw)(struct sk_buff *msg, struct ib_srq *ib_srq); 2689 2690 /* Device lifecycle callbacks */ 2691 /* 2692 * Called after the device becomes registered, before clients are 2693 * attached 2694 */ 2695 int (*enable_driver)(struct ib_device *dev); 2696 /* 2697 * This is called as part of ib_dealloc_device(). 2698 */ 2699 void (*dealloc_driver)(struct ib_device *dev); 2700 2701 /* iWarp CM callbacks */ 2702 void (*iw_add_ref)(struct ib_qp *qp); 2703 void (*iw_rem_ref)(struct ib_qp *qp); 2704 struct ib_qp *(*iw_get_qp)(struct ib_device *device, int qpn); 2705 int (*iw_connect)(struct iw_cm_id *cm_id, 2706 struct iw_cm_conn_param *conn_param); 2707 int (*iw_accept)(struct iw_cm_id *cm_id, 2708 struct iw_cm_conn_param *conn_param); 2709 int (*iw_reject)(struct iw_cm_id *cm_id, const void *pdata, 2710 u8 pdata_len); 2711 int (*iw_create_listen)(struct iw_cm_id *cm_id, int backlog); 2712 int (*iw_destroy_listen)(struct iw_cm_id *cm_id); 2713 /* 2714 * counter_bind_qp - Bind a QP to a counter. 2715 * @counter - The counter to be bound. If counter->id is zero then 2716 * the driver needs to allocate a new counter and set counter->id 2717 */ 2718 int (*counter_bind_qp)(struct rdma_counter *counter, struct ib_qp *qp, 2719 u32 port); 2720 /* 2721 * counter_unbind_qp - Unbind the qp from the dynamically-allocated 2722 * counter and bind it onto the default one 2723 */ 2724 int (*counter_unbind_qp)(struct ib_qp *qp, u32 port); 2725 /* 2726 * counter_dealloc -De-allocate the hw counter 2727 */ 2728 int (*counter_dealloc)(struct rdma_counter *counter); 2729 /* 2730 * counter_alloc_stats - Allocate a struct rdma_hw_stats and fill in 2731 * the driver initialized data. 2732 */ 2733 struct rdma_hw_stats *(*counter_alloc_stats)( 2734 struct rdma_counter *counter); 2735 /* 2736 * counter_update_stats - Query the stats value of this counter 2737 */ 2738 int (*counter_update_stats)(struct rdma_counter *counter); 2739 2740 /* 2741 * counter_init - Initialize the driver specific rdma counter struct. 2742 */ 2743 void (*counter_init)(struct rdma_counter *counter); 2744 2745 /* 2746 * Allows rdma drivers to add their own restrack attributes 2747 * dumped via 'rdma stat' iproute2 command. 2748 */ 2749 int (*fill_stat_mr_entry)(struct sk_buff *msg, struct ib_mr *ibmr); 2750 2751 /* query driver for its ucontext properties */ 2752 int (*query_ucontext)(struct ib_ucontext *context, 2753 struct uverbs_attr_bundle *attrs); 2754 2755 /* 2756 * Provide NUMA node. This API exists for rdmavt/hfi1 only. 2757 * Everyone else relies on Linux memory management model. 2758 */ 2759 int (*get_numa_node)(struct ib_device *dev); 2760 2761 /* 2762 * add_sub_dev - Add a sub IB device 2763 */ 2764 struct ib_device *(*add_sub_dev)(struct ib_device *parent, 2765 enum rdma_nl_dev_type type, 2766 const char *name); 2767 2768 /* 2769 * del_sub_dev - Delete a sub IB device 2770 */ 2771 void (*del_sub_dev)(struct ib_device *sub_dev); 2772 2773 /* 2774 * ufile_cleanup - Attempt to cleanup ubojects HW resources inside 2775 * the ufile. 2776 */ 2777 void (*ufile_hw_cleanup)(struct ib_uverbs_file *ufile); 2778 2779 /* 2780 * report_port_event - Drivers need to implement this if they have 2781 * some private stuff to handle when link status changes. 2782 */ 2783 void (*report_port_event)(struct ib_device *ibdev, 2784 struct net_device *ndev, unsigned long event); 2785 2786 DECLARE_RDMA_OBJ_SIZE(ib_ah); 2787 DECLARE_RDMA_OBJ_SIZE(ib_counters); 2788 DECLARE_RDMA_OBJ_SIZE(ib_cq); 2789 DECLARE_RDMA_OBJ_SIZE(ib_dmah); 2790 DECLARE_RDMA_OBJ_SIZE(ib_mw); 2791 DECLARE_RDMA_OBJ_SIZE(ib_pd); 2792 DECLARE_RDMA_OBJ_SIZE(ib_qp); 2793 DECLARE_RDMA_OBJ_SIZE(ib_rwq_ind_table); 2794 DECLARE_RDMA_OBJ_SIZE(ib_srq); 2795 DECLARE_RDMA_OBJ_SIZE(ib_ucontext); 2796 DECLARE_RDMA_OBJ_SIZE(ib_xrcd); 2797 DECLARE_RDMA_OBJ_SIZE(rdma_counter); 2798 }; 2799 2800 struct ib_core_device { 2801 /* device must be the first element in structure until, 2802 * union of ib_core_device and device exists in ib_device. 2803 */ 2804 struct device dev; 2805 possible_net_t rdma_net; 2806 struct kobject *ports_kobj; 2807 struct list_head port_list; 2808 struct ib_device *owner; /* reach back to owner ib_device */ 2809 }; 2810 2811 struct rdma_restrack_root; 2812 struct ib_device { 2813 /* Do not access @dma_device directly from ULP nor from HW drivers. */ 2814 struct device *dma_device; 2815 struct ib_device_ops ops; 2816 char name[IB_DEVICE_NAME_MAX]; 2817 struct rcu_head rcu_head; 2818 2819 struct list_head event_handler_list; 2820 /* Protects event_handler_list */ 2821 struct rw_semaphore event_handler_rwsem; 2822 2823 /* Protects QP's event_handler calls and open_qp list */ 2824 spinlock_t qp_open_list_lock; 2825 2826 struct rw_semaphore client_data_rwsem; 2827 struct xarray client_data; 2828 struct mutex unregistration_lock; 2829 2830 /* Synchronize GID, Pkey cache entries, subnet prefix, LMC */ 2831 rwlock_t cache_lock; 2832 /** 2833 * port_data is indexed by port number 2834 */ 2835 struct ib_port_data *port_data; 2836 2837 int num_comp_vectors; 2838 2839 union { 2840 struct device dev; 2841 struct ib_core_device coredev; 2842 }; 2843 2844 /* First group is for device attributes, 2845 * Second group is for driver provided attributes (optional). 2846 * Third group is for the hw_stats 2847 * It is a NULL terminated array. 2848 */ 2849 const struct attribute_group *groups[4]; 2850 u8 hw_stats_attr_index; 2851 2852 u64 uverbs_cmd_mask; 2853 2854 char node_desc[IB_DEVICE_NODE_DESC_MAX]; 2855 __be64 node_guid; 2856 u32 local_dma_lkey; 2857 u16 is_switch:1; 2858 /* Indicates kernel verbs support, should not be used in drivers */ 2859 u16 kverbs_provider:1; 2860 /* CQ adaptive moderation (RDMA DIM) */ 2861 u16 use_cq_dim:1; 2862 u8 node_type; 2863 u32 phys_port_cnt; 2864 struct ib_device_attr attrs; 2865 struct hw_stats_device_data *hw_stats_data; 2866 2867 #ifdef CONFIG_CGROUP_RDMA 2868 struct rdmacg_device cg_device; 2869 #endif 2870 2871 u32 index; 2872 2873 spinlock_t cq_pools_lock; 2874 struct list_head cq_pools[IB_POLL_LAST_POOL_TYPE + 1]; 2875 2876 struct rdma_restrack_root *res; 2877 2878 const struct uapi_definition *driver_def; 2879 2880 /* 2881 * Positive refcount indicates that the device is currently 2882 * registered and cannot be unregistered. 2883 */ 2884 refcount_t refcount; 2885 struct completion unreg_completion; 2886 struct work_struct unregistration_work; 2887 2888 const struct rdma_link_ops *link_ops; 2889 2890 /* Protects compat_devs xarray modifications */ 2891 struct mutex compat_devs_mutex; 2892 /* Maintains compat devices for each net namespace */ 2893 struct xarray compat_devs; 2894 2895 /* Used by iWarp CM */ 2896 char iw_ifname[IFNAMSIZ]; 2897 u32 iw_driver_flags; 2898 u32 lag_flags; 2899 2900 /* A parent device has a list of sub-devices */ 2901 struct mutex subdev_lock; 2902 struct list_head subdev_list_head; 2903 2904 /* A sub device has a type and a parent */ 2905 enum rdma_nl_dev_type type; 2906 struct ib_device *parent; 2907 struct list_head subdev_list; 2908 2909 enum rdma_nl_name_assign_type name_assign_type; 2910 }; 2911 2912 static inline void *rdma_zalloc_obj(struct ib_device *dev, size_t size, 2913 gfp_t gfp, bool is_numa_aware) 2914 { 2915 if (is_numa_aware && dev->ops.get_numa_node) 2916 return kzalloc_node(size, gfp, dev->ops.get_numa_node(dev)); 2917 2918 return kzalloc(size, gfp); 2919 } 2920 2921 struct ib_client_nl_info; 2922 struct ib_client { 2923 const char *name; 2924 int (*add)(struct ib_device *ibdev); 2925 void (*remove)(struct ib_device *, void *client_data); 2926 void (*rename)(struct ib_device *dev, void *client_data); 2927 int (*get_nl_info)(struct ib_device *ibdev, void *client_data, 2928 struct ib_client_nl_info *res); 2929 int (*get_global_nl_info)(struct ib_client_nl_info *res); 2930 2931 /* Returns the net_dev belonging to this ib_client and matching the 2932 * given parameters. 2933 * @dev: An RDMA device that the net_dev use for communication. 2934 * @port: A physical port number on the RDMA device. 2935 * @pkey: P_Key that the net_dev uses if applicable. 2936 * @gid: A GID that the net_dev uses to communicate. 2937 * @addr: An IP address the net_dev is configured with. 2938 * @client_data: The device's client data set by ib_set_client_data(). 2939 * 2940 * An ib_client that implements a net_dev on top of RDMA devices 2941 * (such as IP over IB) should implement this callback, allowing the 2942 * rdma_cm module to find the right net_dev for a given request. 2943 * 2944 * The caller is responsible for calling dev_put on the returned 2945 * netdev. */ 2946 struct net_device *(*get_net_dev_by_params)( 2947 struct ib_device *dev, 2948 u32 port, 2949 u16 pkey, 2950 const union ib_gid *gid, 2951 const struct sockaddr *addr, 2952 void *client_data); 2953 2954 refcount_t uses; 2955 struct completion uses_zero; 2956 u32 client_id; 2957 2958 /* kverbs are not required by the client */ 2959 u8 no_kverbs_req:1; 2960 }; 2961 2962 /* 2963 * IB block DMA iterator 2964 * 2965 * Iterates the DMA-mapped SGL in contiguous memory blocks aligned 2966 * to a HW supported page size. 2967 */ 2968 struct ib_block_iter { 2969 /* internal states */ 2970 struct scatterlist *__sg; /* sg holding the current aligned block */ 2971 dma_addr_t __dma_addr; /* unaligned DMA address of this block */ 2972 size_t __sg_numblocks; /* ib_umem_num_dma_blocks() */ 2973 unsigned int __sg_nents; /* number of SG entries */ 2974 unsigned int __sg_advance; /* number of bytes to advance in sg in next step */ 2975 unsigned int __pg_bit; /* alignment of current block */ 2976 }; 2977 2978 struct ib_device *_ib_alloc_device(size_t size, struct net *net); 2979 #define ib_alloc_device(drv_struct, member) \ 2980 container_of(_ib_alloc_device(sizeof(struct drv_struct) + \ 2981 BUILD_BUG_ON_ZERO(offsetof( \ 2982 struct drv_struct, member)), \ 2983 &init_net), \ 2984 struct drv_struct, member) 2985 2986 #define ib_alloc_device_with_net(drv_struct, member, net) \ 2987 container_of(_ib_alloc_device(sizeof(struct drv_struct) + \ 2988 BUILD_BUG_ON_ZERO(offsetof( \ 2989 struct drv_struct, member)), net), \ 2990 struct drv_struct, member) 2991 2992 void ib_dealloc_device(struct ib_device *device); 2993 2994 void ib_get_device_fw_str(struct ib_device *device, char *str); 2995 2996 int ib_register_device(struct ib_device *device, const char *name, 2997 struct device *dma_device); 2998 void ib_unregister_device(struct ib_device *device); 2999 void ib_unregister_driver(enum rdma_driver_id driver_id); 3000 void ib_unregister_device_and_put(struct ib_device *device); 3001 void ib_unregister_device_queued(struct ib_device *ib_dev); 3002 3003 int ib_register_client (struct ib_client *client); 3004 void ib_unregister_client(struct ib_client *client); 3005 3006 void __rdma_block_iter_start(struct ib_block_iter *biter, 3007 struct scatterlist *sglist, 3008 unsigned int nents, 3009 unsigned long pgsz); 3010 bool __rdma_block_iter_next(struct ib_block_iter *biter); 3011 3012 /** 3013 * rdma_block_iter_dma_address - get the aligned dma address of the current 3014 * block held by the block iterator. 3015 * @biter: block iterator holding the memory block 3016 */ 3017 static inline dma_addr_t 3018 rdma_block_iter_dma_address(struct ib_block_iter *biter) 3019 { 3020 return biter->__dma_addr & ~(BIT_ULL(biter->__pg_bit) - 1); 3021 } 3022 3023 /** 3024 * rdma_for_each_block - iterate over contiguous memory blocks of the sg list 3025 * @sglist: sglist to iterate over 3026 * @biter: block iterator holding the memory block 3027 * @nents: maximum number of sg entries to iterate over 3028 * @pgsz: best HW supported page size to use 3029 * 3030 * Callers may use rdma_block_iter_dma_address() to get each 3031 * blocks aligned DMA address. 3032 */ 3033 #define rdma_for_each_block(sglist, biter, nents, pgsz) \ 3034 for (__rdma_block_iter_start(biter, sglist, nents, \ 3035 pgsz); \ 3036 __rdma_block_iter_next(biter);) 3037 3038 /** 3039 * ib_get_client_data - Get IB client context 3040 * @device:Device to get context for 3041 * @client:Client to get context for 3042 * 3043 * ib_get_client_data() returns the client context data set with 3044 * ib_set_client_data(). This can only be called while the client is 3045 * registered to the device, once the ib_client remove() callback returns this 3046 * cannot be called. 3047 */ 3048 static inline void *ib_get_client_data(struct ib_device *device, 3049 struct ib_client *client) 3050 { 3051 return xa_load(&device->client_data, client->client_id); 3052 } 3053 void ib_set_client_data(struct ib_device *device, struct ib_client *client, 3054 void *data); 3055 void ib_set_device_ops(struct ib_device *device, 3056 const struct ib_device_ops *ops); 3057 3058 int rdma_user_mmap_io(struct ib_ucontext *ucontext, struct vm_area_struct *vma, 3059 unsigned long pfn, unsigned long size, pgprot_t prot, 3060 struct rdma_user_mmap_entry *entry); 3061 int rdma_user_mmap_entry_insert(struct ib_ucontext *ucontext, 3062 struct rdma_user_mmap_entry *entry, 3063 size_t length); 3064 int rdma_user_mmap_entry_insert_range(struct ib_ucontext *ucontext, 3065 struct rdma_user_mmap_entry *entry, 3066 size_t length, u32 min_pgoff, 3067 u32 max_pgoff); 3068 3069 #if IS_ENABLED(CONFIG_INFINIBAND_USER_ACCESS) 3070 void rdma_user_mmap_disassociate(struct ib_device *device); 3071 #else 3072 static inline void rdma_user_mmap_disassociate(struct ib_device *device) 3073 { 3074 } 3075 #endif 3076 3077 static inline int 3078 rdma_user_mmap_entry_insert_exact(struct ib_ucontext *ucontext, 3079 struct rdma_user_mmap_entry *entry, 3080 size_t length, u32 pgoff) 3081 { 3082 return rdma_user_mmap_entry_insert_range(ucontext, entry, length, pgoff, 3083 pgoff); 3084 } 3085 3086 struct rdma_user_mmap_entry * 3087 rdma_user_mmap_entry_get_pgoff(struct ib_ucontext *ucontext, 3088 unsigned long pgoff); 3089 struct rdma_user_mmap_entry * 3090 rdma_user_mmap_entry_get(struct ib_ucontext *ucontext, 3091 struct vm_area_struct *vma); 3092 void rdma_user_mmap_entry_put(struct rdma_user_mmap_entry *entry); 3093 3094 void rdma_user_mmap_entry_remove(struct rdma_user_mmap_entry *entry); 3095 3096 static inline int ib_copy_from_udata(void *dest, struct ib_udata *udata, size_t len) 3097 { 3098 return copy_from_user(dest, udata->inbuf, len) ? -EFAULT : 0; 3099 } 3100 3101 static inline int ib_copy_to_udata(struct ib_udata *udata, void *src, size_t len) 3102 { 3103 return copy_to_user(udata->outbuf, src, len) ? -EFAULT : 0; 3104 } 3105 3106 static inline bool ib_is_buffer_cleared(const void __user *p, 3107 size_t len) 3108 { 3109 bool ret; 3110 u8 *buf; 3111 3112 if (len > USHRT_MAX) 3113 return false; 3114 3115 buf = memdup_user(p, len); 3116 if (IS_ERR(buf)) 3117 return false; 3118 3119 ret = !memchr_inv(buf, 0, len); 3120 kfree(buf); 3121 return ret; 3122 } 3123 3124 static inline bool ib_is_udata_cleared(struct ib_udata *udata, 3125 size_t offset, 3126 size_t len) 3127 { 3128 return ib_is_buffer_cleared(udata->inbuf + offset, len); 3129 } 3130 3131 /** 3132 * ib_modify_qp_is_ok - Check that the supplied attribute mask 3133 * contains all required attributes and no attributes not allowed for 3134 * the given QP state transition. 3135 * @cur_state: Current QP state 3136 * @next_state: Next QP state 3137 * @type: QP type 3138 * @mask: Mask of supplied QP attributes 3139 * 3140 * This function is a helper function that a low-level driver's 3141 * modify_qp method can use to validate the consumer's input. It 3142 * checks that cur_state and next_state are valid QP states, that a 3143 * transition from cur_state to next_state is allowed by the IB spec, 3144 * and that the attribute mask supplied is allowed for the transition. 3145 */ 3146 bool ib_modify_qp_is_ok(enum ib_qp_state cur_state, enum ib_qp_state next_state, 3147 enum ib_qp_type type, enum ib_qp_attr_mask mask); 3148 3149 void ib_register_event_handler(struct ib_event_handler *event_handler); 3150 void ib_unregister_event_handler(struct ib_event_handler *event_handler); 3151 void ib_dispatch_event(const struct ib_event *event); 3152 3153 int ib_query_port(struct ib_device *device, 3154 u32 port_num, struct ib_port_attr *port_attr); 3155 3156 enum rdma_link_layer rdma_port_get_link_layer(struct ib_device *device, 3157 u32 port_num); 3158 3159 /** 3160 * rdma_cap_ib_switch - Check if the device is IB switch 3161 * @device: Device to check 3162 * 3163 * Device driver is responsible for setting is_switch bit on 3164 * in ib_device structure at init time. 3165 * 3166 * Return: true if the device is IB switch. 3167 */ 3168 static inline bool rdma_cap_ib_switch(const struct ib_device *device) 3169 { 3170 return device->is_switch; 3171 } 3172 3173 /** 3174 * rdma_start_port - Return the first valid port number for the device 3175 * specified 3176 * 3177 * @device: Device to be checked 3178 * 3179 * Return start port number 3180 */ 3181 static inline u32 rdma_start_port(const struct ib_device *device) 3182 { 3183 return rdma_cap_ib_switch(device) ? 0 : 1; 3184 } 3185 3186 /** 3187 * rdma_for_each_port - Iterate over all valid port numbers of the IB device 3188 * @device: The struct ib_device * to iterate over 3189 * @iter: The unsigned int to store the port number 3190 */ 3191 #define rdma_for_each_port(device, iter) \ 3192 for (iter = rdma_start_port(device + \ 3193 BUILD_BUG_ON_ZERO(!__same_type(u32, \ 3194 iter))); \ 3195 iter <= rdma_end_port(device); iter++) 3196 3197 /** 3198 * rdma_end_port - Return the last valid port number for the device 3199 * specified 3200 * 3201 * @device: Device to be checked 3202 * 3203 * Return last port number 3204 */ 3205 static inline u32 rdma_end_port(const struct ib_device *device) 3206 { 3207 return rdma_cap_ib_switch(device) ? 0 : device->phys_port_cnt; 3208 } 3209 3210 static inline int rdma_is_port_valid(const struct ib_device *device, 3211 unsigned int port) 3212 { 3213 return (port >= rdma_start_port(device) && 3214 port <= rdma_end_port(device)); 3215 } 3216 3217 static inline bool rdma_is_grh_required(const struct ib_device *device, 3218 u32 port_num) 3219 { 3220 return device->port_data[port_num].immutable.core_cap_flags & 3221 RDMA_CORE_PORT_IB_GRH_REQUIRED; 3222 } 3223 3224 static inline bool rdma_protocol_ib(const struct ib_device *device, 3225 u32 port_num) 3226 { 3227 return device->port_data[port_num].immutable.core_cap_flags & 3228 RDMA_CORE_CAP_PROT_IB; 3229 } 3230 3231 static inline bool rdma_protocol_roce(const struct ib_device *device, 3232 u32 port_num) 3233 { 3234 return device->port_data[port_num].immutable.core_cap_flags & 3235 (RDMA_CORE_CAP_PROT_ROCE | RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP); 3236 } 3237 3238 static inline bool rdma_protocol_roce_udp_encap(const struct ib_device *device, 3239 u32 port_num) 3240 { 3241 return device->port_data[port_num].immutable.core_cap_flags & 3242 RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP; 3243 } 3244 3245 static inline bool rdma_protocol_roce_eth_encap(const struct ib_device *device, 3246 u32 port_num) 3247 { 3248 return device->port_data[port_num].immutable.core_cap_flags & 3249 RDMA_CORE_CAP_PROT_ROCE; 3250 } 3251 3252 static inline bool rdma_protocol_iwarp(const struct ib_device *device, 3253 u32 port_num) 3254 { 3255 return device->port_data[port_num].immutable.core_cap_flags & 3256 RDMA_CORE_CAP_PROT_IWARP; 3257 } 3258 3259 static inline bool rdma_ib_or_roce(const struct ib_device *device, 3260 u32 port_num) 3261 { 3262 return rdma_protocol_ib(device, port_num) || 3263 rdma_protocol_roce(device, port_num); 3264 } 3265 3266 static inline bool rdma_protocol_raw_packet(const struct ib_device *device, 3267 u32 port_num) 3268 { 3269 return device->port_data[port_num].immutable.core_cap_flags & 3270 RDMA_CORE_CAP_PROT_RAW_PACKET; 3271 } 3272 3273 static inline bool rdma_protocol_usnic(const struct ib_device *device, 3274 u32 port_num) 3275 { 3276 return device->port_data[port_num].immutable.core_cap_flags & 3277 RDMA_CORE_CAP_PROT_USNIC; 3278 } 3279 3280 /** 3281 * rdma_cap_ib_mad - Check if the port of a device supports Infiniband 3282 * Management Datagrams. 3283 * @device: Device to check 3284 * @port_num: Port number to check 3285 * 3286 * Management Datagrams (MAD) are a required part of the InfiniBand 3287 * specification and are supported on all InfiniBand devices. A slightly 3288 * extended version are also supported on OPA interfaces. 3289 * 3290 * Return: true if the port supports sending/receiving of MAD packets. 3291 */ 3292 static inline bool rdma_cap_ib_mad(const struct ib_device *device, u32 port_num) 3293 { 3294 return device->port_data[port_num].immutable.core_cap_flags & 3295 RDMA_CORE_CAP_IB_MAD; 3296 } 3297 3298 /** 3299 * rdma_cap_opa_mad - Check if the port of device provides support for OPA 3300 * Management Datagrams. 3301 * @device: Device to check 3302 * @port_num: Port number to check 3303 * 3304 * Intel OmniPath devices extend and/or replace the InfiniBand Management 3305 * datagrams with their own versions. These OPA MADs share many but not all of 3306 * the characteristics of InfiniBand MADs. 3307 * 3308 * OPA MADs differ in the following ways: 3309 * 3310 * 1) MADs are variable size up to 2K 3311 * IBTA defined MADs remain fixed at 256 bytes 3312 * 2) OPA SMPs must carry valid PKeys 3313 * 3) OPA SMP packets are a different format 3314 * 3315 * Return: true if the port supports OPA MAD packet formats. 3316 */ 3317 static inline bool rdma_cap_opa_mad(struct ib_device *device, u32 port_num) 3318 { 3319 return device->port_data[port_num].immutable.core_cap_flags & 3320 RDMA_CORE_CAP_OPA_MAD; 3321 } 3322 3323 /** 3324 * rdma_cap_ib_smi - Check if the port of a device provides an Infiniband 3325 * Subnet Management Agent (SMA) on the Subnet Management Interface (SMI). 3326 * @device: Device to check 3327 * @port_num: Port number to check 3328 * 3329 * Each InfiniBand node is required to provide a Subnet Management Agent 3330 * that the subnet manager can access. Prior to the fabric being fully 3331 * configured by the subnet manager, the SMA is accessed via a well known 3332 * interface called the Subnet Management Interface (SMI). This interface 3333 * uses directed route packets to communicate with the SM to get around the 3334 * chicken and egg problem of the SM needing to know what's on the fabric 3335 * in order to configure the fabric, and needing to configure the fabric in 3336 * order to send packets to the devices on the fabric. These directed 3337 * route packets do not need the fabric fully configured in order to reach 3338 * their destination. The SMI is the only method allowed to send 3339 * directed route packets on an InfiniBand fabric. 3340 * 3341 * Return: true if the port provides an SMI. 3342 */ 3343 static inline bool rdma_cap_ib_smi(const struct ib_device *device, u32 port_num) 3344 { 3345 return device->port_data[port_num].immutable.core_cap_flags & 3346 RDMA_CORE_CAP_IB_SMI; 3347 } 3348 3349 /** 3350 * rdma_cap_ib_cm - Check if the port of device has the capability Infiniband 3351 * Communication Manager. 3352 * @device: Device to check 3353 * @port_num: Port number to check 3354 * 3355 * The InfiniBand Communication Manager is one of many pre-defined General 3356 * Service Agents (GSA) that are accessed via the General Service 3357 * Interface (GSI). It's role is to facilitate establishment of connections 3358 * between nodes as well as other management related tasks for established 3359 * connections. 3360 * 3361 * Return: true if the port supports an IB CM (this does not guarantee that 3362 * a CM is actually running however). 3363 */ 3364 static inline bool rdma_cap_ib_cm(const struct ib_device *device, u32 port_num) 3365 { 3366 return device->port_data[port_num].immutable.core_cap_flags & 3367 RDMA_CORE_CAP_IB_CM; 3368 } 3369 3370 /** 3371 * rdma_cap_iw_cm - Check if the port of device has the capability IWARP 3372 * Communication Manager. 3373 * @device: Device to check 3374 * @port_num: Port number to check 3375 * 3376 * Similar to above, but specific to iWARP connections which have a different 3377 * managment protocol than InfiniBand. 3378 * 3379 * Return: true if the port supports an iWARP CM (this does not guarantee that 3380 * a CM is actually running however). 3381 */ 3382 static inline bool rdma_cap_iw_cm(const struct ib_device *device, u32 port_num) 3383 { 3384 return device->port_data[port_num].immutable.core_cap_flags & 3385 RDMA_CORE_CAP_IW_CM; 3386 } 3387 3388 /** 3389 * rdma_cap_ib_sa - Check if the port of device has the capability Infiniband 3390 * Subnet Administration. 3391 * @device: Device to check 3392 * @port_num: Port number to check 3393 * 3394 * An InfiniBand Subnet Administration (SA) service is a pre-defined General 3395 * Service Agent (GSA) provided by the Subnet Manager (SM). On InfiniBand 3396 * fabrics, devices should resolve routes to other hosts by contacting the 3397 * SA to query the proper route. 3398 * 3399 * Return: true if the port should act as a client to the fabric Subnet 3400 * Administration interface. This does not imply that the SA service is 3401 * running locally. 3402 */ 3403 static inline bool rdma_cap_ib_sa(const struct ib_device *device, u32 port_num) 3404 { 3405 return device->port_data[port_num].immutable.core_cap_flags & 3406 RDMA_CORE_CAP_IB_SA; 3407 } 3408 3409 /** 3410 * rdma_cap_ib_mcast - Check if the port of device has the capability Infiniband 3411 * Multicast. 3412 * @device: Device to check 3413 * @port_num: Port number to check 3414 * 3415 * InfiniBand multicast registration is more complex than normal IPv4 or 3416 * IPv6 multicast registration. Each Host Channel Adapter must register 3417 * with the Subnet Manager when it wishes to join a multicast group. It 3418 * should do so only once regardless of how many queue pairs it subscribes 3419 * to this group. And it should leave the group only after all queue pairs 3420 * attached to the group have been detached. 3421 * 3422 * Return: true if the port must undertake the additional adminstrative 3423 * overhead of registering/unregistering with the SM and tracking of the 3424 * total number of queue pairs attached to the multicast group. 3425 */ 3426 static inline bool rdma_cap_ib_mcast(const struct ib_device *device, 3427 u32 port_num) 3428 { 3429 return rdma_cap_ib_sa(device, port_num); 3430 } 3431 3432 /** 3433 * rdma_cap_af_ib - Check if the port of device has the capability 3434 * Native Infiniband Address. 3435 * @device: Device to check 3436 * @port_num: Port number to check 3437 * 3438 * InfiniBand addressing uses a port's GUID + Subnet Prefix to make a default 3439 * GID. RoCE uses a different mechanism, but still generates a GID via 3440 * a prescribed mechanism and port specific data. 3441 * 3442 * Return: true if the port uses a GID address to identify devices on the 3443 * network. 3444 */ 3445 static inline bool rdma_cap_af_ib(const struct ib_device *device, u32 port_num) 3446 { 3447 return device->port_data[port_num].immutable.core_cap_flags & 3448 RDMA_CORE_CAP_AF_IB; 3449 } 3450 3451 /** 3452 * rdma_cap_eth_ah - Check if the port of device has the capability 3453 * Ethernet Address Handle. 3454 * @device: Device to check 3455 * @port_num: Port number to check 3456 * 3457 * RoCE is InfiniBand over Ethernet, and it uses a well defined technique 3458 * to fabricate GIDs over Ethernet/IP specific addresses native to the 3459 * port. Normally, packet headers are generated by the sending host 3460 * adapter, but when sending connectionless datagrams, we must manually 3461 * inject the proper headers for the fabric we are communicating over. 3462 * 3463 * Return: true if we are running as a RoCE port and must force the 3464 * addition of a Global Route Header built from our Ethernet Address 3465 * Handle into our header list for connectionless packets. 3466 */ 3467 static inline bool rdma_cap_eth_ah(const struct ib_device *device, u32 port_num) 3468 { 3469 return device->port_data[port_num].immutable.core_cap_flags & 3470 RDMA_CORE_CAP_ETH_AH; 3471 } 3472 3473 /** 3474 * rdma_cap_opa_ah - Check if the port of device supports 3475 * OPA Address handles 3476 * @device: Device to check 3477 * @port_num: Port number to check 3478 * 3479 * Return: true if we are running on an OPA device which supports 3480 * the extended OPA addressing. 3481 */ 3482 static inline bool rdma_cap_opa_ah(struct ib_device *device, u32 port_num) 3483 { 3484 return (device->port_data[port_num].immutable.core_cap_flags & 3485 RDMA_CORE_CAP_OPA_AH) == RDMA_CORE_CAP_OPA_AH; 3486 } 3487 3488 /** 3489 * rdma_max_mad_size - Return the max MAD size required by this RDMA Port. 3490 * 3491 * @device: Device 3492 * @port_num: Port number 3493 * 3494 * This MAD size includes the MAD headers and MAD payload. No other headers 3495 * are included. 3496 * 3497 * Return the max MAD size required by the Port. Will return 0 if the port 3498 * does not support MADs 3499 */ 3500 static inline size_t rdma_max_mad_size(const struct ib_device *device, 3501 u32 port_num) 3502 { 3503 return device->port_data[port_num].immutable.max_mad_size; 3504 } 3505 3506 /** 3507 * rdma_cap_roce_gid_table - Check if the port of device uses roce_gid_table 3508 * @device: Device to check 3509 * @port_num: Port number to check 3510 * 3511 * RoCE GID table mechanism manages the various GIDs for a device. 3512 * 3513 * NOTE: if allocating the port's GID table has failed, this call will still 3514 * return true, but any RoCE GID table API will fail. 3515 * 3516 * Return: true if the port uses RoCE GID table mechanism in order to manage 3517 * its GIDs. 3518 */ 3519 static inline bool rdma_cap_roce_gid_table(const struct ib_device *device, 3520 u32 port_num) 3521 { 3522 return rdma_protocol_roce(device, port_num) && 3523 device->ops.add_gid && device->ops.del_gid; 3524 } 3525 3526 /* 3527 * Check if the device supports READ W/ INVALIDATE. 3528 */ 3529 static inline bool rdma_cap_read_inv(struct ib_device *dev, u32 port_num) 3530 { 3531 /* 3532 * iWarp drivers must support READ W/ INVALIDATE. No other protocol 3533 * has support for it yet. 3534 */ 3535 return rdma_protocol_iwarp(dev, port_num); 3536 } 3537 3538 /** 3539 * rdma_core_cap_opa_port - Return whether the RDMA Port is OPA or not. 3540 * @device: Device 3541 * @port_num: 1 based Port number 3542 * 3543 * Return true if port is an Intel OPA port , false if not 3544 */ 3545 static inline bool rdma_core_cap_opa_port(struct ib_device *device, 3546 u32 port_num) 3547 { 3548 return (device->port_data[port_num].immutable.core_cap_flags & 3549 RDMA_CORE_PORT_INTEL_OPA) == RDMA_CORE_PORT_INTEL_OPA; 3550 } 3551 3552 /** 3553 * rdma_mtu_enum_to_int - Return the mtu of the port as an integer value. 3554 * @device: Device 3555 * @port: Port number 3556 * @mtu: enum value of MTU 3557 * 3558 * Return the MTU size supported by the port as an integer value. Will return 3559 * -1 if enum value of mtu is not supported. 3560 */ 3561 static inline int rdma_mtu_enum_to_int(struct ib_device *device, u32 port, 3562 int mtu) 3563 { 3564 if (rdma_core_cap_opa_port(device, port)) 3565 return opa_mtu_enum_to_int((enum opa_mtu)mtu); 3566 else 3567 return ib_mtu_enum_to_int((enum ib_mtu)mtu); 3568 } 3569 3570 /** 3571 * rdma_mtu_from_attr - Return the mtu of the port from the port attribute. 3572 * @device: Device 3573 * @port: Port number 3574 * @attr: port attribute 3575 * 3576 * Return the MTU size supported by the port as an integer value. 3577 */ 3578 static inline int rdma_mtu_from_attr(struct ib_device *device, u32 port, 3579 struct ib_port_attr *attr) 3580 { 3581 if (rdma_core_cap_opa_port(device, port)) 3582 return attr->phys_mtu; 3583 else 3584 return ib_mtu_enum_to_int(attr->max_mtu); 3585 } 3586 3587 int ib_set_vf_link_state(struct ib_device *device, int vf, u32 port, 3588 int state); 3589 int ib_get_vf_config(struct ib_device *device, int vf, u32 port, 3590 struct ifla_vf_info *info); 3591 int ib_get_vf_stats(struct ib_device *device, int vf, u32 port, 3592 struct ifla_vf_stats *stats); 3593 int ib_get_vf_guid(struct ib_device *device, int vf, u32 port, 3594 struct ifla_vf_guid *node_guid, 3595 struct ifla_vf_guid *port_guid); 3596 int ib_set_vf_guid(struct ib_device *device, int vf, u32 port, u64 guid, 3597 int type); 3598 3599 int ib_query_pkey(struct ib_device *device, 3600 u32 port_num, u16 index, u16 *pkey); 3601 3602 int ib_modify_device(struct ib_device *device, 3603 int device_modify_mask, 3604 struct ib_device_modify *device_modify); 3605 3606 int ib_modify_port(struct ib_device *device, 3607 u32 port_num, int port_modify_mask, 3608 struct ib_port_modify *port_modify); 3609 3610 int ib_find_gid(struct ib_device *device, union ib_gid *gid, 3611 u32 *port_num, u16 *index); 3612 3613 int ib_find_pkey(struct ib_device *device, 3614 u32 port_num, u16 pkey, u16 *index); 3615 3616 enum ib_pd_flags { 3617 /* 3618 * Create a memory registration for all memory in the system and place 3619 * the rkey for it into pd->unsafe_global_rkey. This can be used by 3620 * ULPs to avoid the overhead of dynamic MRs. 3621 * 3622 * This flag is generally considered unsafe and must only be used in 3623 * extremly trusted environments. Every use of it will log a warning 3624 * in the kernel log. 3625 */ 3626 IB_PD_UNSAFE_GLOBAL_RKEY = 0x01, 3627 }; 3628 3629 struct ib_pd *__ib_alloc_pd(struct ib_device *device, unsigned int flags, 3630 const char *caller); 3631 3632 /** 3633 * ib_alloc_pd - Allocates an unused protection domain. 3634 * @device: The device on which to allocate the protection domain. 3635 * @flags: protection domain flags 3636 * 3637 * A protection domain object provides an association between QPs, shared 3638 * receive queues, address handles, memory regions, and memory windows. 3639 * 3640 * Every PD has a local_dma_lkey which can be used as the lkey value for local 3641 * memory operations. 3642 */ 3643 #define ib_alloc_pd(device, flags) \ 3644 __ib_alloc_pd((device), (flags), KBUILD_MODNAME) 3645 3646 int ib_dealloc_pd_user(struct ib_pd *pd, struct ib_udata *udata); 3647 3648 /** 3649 * ib_dealloc_pd - Deallocate kernel PD 3650 * @pd: The protection domain 3651 * 3652 * NOTE: for user PD use ib_dealloc_pd_user with valid udata! 3653 */ 3654 static inline void ib_dealloc_pd(struct ib_pd *pd) 3655 { 3656 int ret = ib_dealloc_pd_user(pd, NULL); 3657 3658 WARN_ONCE(ret, "Destroy of kernel PD shouldn't fail"); 3659 } 3660 3661 enum rdma_create_ah_flags { 3662 /* In a sleepable context */ 3663 RDMA_CREATE_AH_SLEEPABLE = BIT(0), 3664 }; 3665 3666 /** 3667 * rdma_create_ah - Creates an address handle for the given address vector. 3668 * @pd: The protection domain associated with the address handle. 3669 * @ah_attr: The attributes of the address vector. 3670 * @flags: Create address handle flags (see enum rdma_create_ah_flags). 3671 * 3672 * The address handle is used to reference a local or global destination 3673 * in all UD QP post sends. 3674 */ 3675 struct ib_ah *rdma_create_ah(struct ib_pd *pd, struct rdma_ah_attr *ah_attr, 3676 u32 flags); 3677 3678 /** 3679 * rdma_create_user_ah - Creates an address handle for the given address vector. 3680 * It resolves destination mac address for ah attribute of RoCE type. 3681 * @pd: The protection domain associated with the address handle. 3682 * @ah_attr: The attributes of the address vector. 3683 * @udata: pointer to user's input output buffer information need by 3684 * provider driver. 3685 * 3686 * It returns 0 on success and returns appropriate error code on error. 3687 * The address handle is used to reference a local or global destination 3688 * in all UD QP post sends. 3689 */ 3690 struct ib_ah *rdma_create_user_ah(struct ib_pd *pd, 3691 struct rdma_ah_attr *ah_attr, 3692 struct ib_udata *udata); 3693 /** 3694 * ib_get_gids_from_rdma_hdr - Get sgid and dgid from GRH or IPv4 header 3695 * work completion. 3696 * @hdr: the L3 header to parse 3697 * @net_type: type of header to parse 3698 * @sgid: place to store source gid 3699 * @dgid: place to store destination gid 3700 */ 3701 int ib_get_gids_from_rdma_hdr(const union rdma_network_hdr *hdr, 3702 enum rdma_network_type net_type, 3703 union ib_gid *sgid, union ib_gid *dgid); 3704 3705 /** 3706 * ib_get_rdma_header_version - Get the header version 3707 * @hdr: the L3 header to parse 3708 */ 3709 int ib_get_rdma_header_version(const union rdma_network_hdr *hdr); 3710 3711 /** 3712 * ib_init_ah_attr_from_wc - Initializes address handle attributes from a 3713 * work completion. 3714 * @device: Device on which the received message arrived. 3715 * @port_num: Port on which the received message arrived. 3716 * @wc: Work completion associated with the received message. 3717 * @grh: References the received global route header. This parameter is 3718 * ignored unless the work completion indicates that the GRH is valid. 3719 * @ah_attr: Returned attributes that can be used when creating an address 3720 * handle for replying to the message. 3721 * When ib_init_ah_attr_from_wc() returns success, 3722 * (a) for IB link layer it optionally contains a reference to SGID attribute 3723 * when GRH is present for IB link layer. 3724 * (b) for RoCE link layer it contains a reference to SGID attribute. 3725 * User must invoke rdma_cleanup_ah_attr_gid_attr() to release reference to SGID 3726 * attributes which are initialized using ib_init_ah_attr_from_wc(). 3727 * 3728 */ 3729 int ib_init_ah_attr_from_wc(struct ib_device *device, u32 port_num, 3730 const struct ib_wc *wc, const struct ib_grh *grh, 3731 struct rdma_ah_attr *ah_attr); 3732 3733 /** 3734 * ib_create_ah_from_wc - Creates an address handle associated with the 3735 * sender of the specified work completion. 3736 * @pd: The protection domain associated with the address handle. 3737 * @wc: Work completion information associated with a received message. 3738 * @grh: References the received global route header. This parameter is 3739 * ignored unless the work completion indicates that the GRH is valid. 3740 * @port_num: The outbound port number to associate with the address. 3741 * 3742 * The address handle is used to reference a local or global destination 3743 * in all UD QP post sends. 3744 */ 3745 struct ib_ah *ib_create_ah_from_wc(struct ib_pd *pd, const struct ib_wc *wc, 3746 const struct ib_grh *grh, u32 port_num); 3747 3748 /** 3749 * rdma_modify_ah - Modifies the address vector associated with an address 3750 * handle. 3751 * @ah: The address handle to modify. 3752 * @ah_attr: The new address vector attributes to associate with the 3753 * address handle. 3754 */ 3755 int rdma_modify_ah(struct ib_ah *ah, struct rdma_ah_attr *ah_attr); 3756 3757 /** 3758 * rdma_query_ah - Queries the address vector associated with an address 3759 * handle. 3760 * @ah: The address handle to query. 3761 * @ah_attr: The address vector attributes associated with the address 3762 * handle. 3763 */ 3764 int rdma_query_ah(struct ib_ah *ah, struct rdma_ah_attr *ah_attr); 3765 3766 enum rdma_destroy_ah_flags { 3767 /* In a sleepable context */ 3768 RDMA_DESTROY_AH_SLEEPABLE = BIT(0), 3769 }; 3770 3771 /** 3772 * rdma_destroy_ah_user - Destroys an address handle. 3773 * @ah: The address handle to destroy. 3774 * @flags: Destroy address handle flags (see enum rdma_destroy_ah_flags). 3775 * @udata: Valid user data or NULL for kernel objects 3776 */ 3777 int rdma_destroy_ah_user(struct ib_ah *ah, u32 flags, struct ib_udata *udata); 3778 3779 /** 3780 * rdma_destroy_ah - Destroys an kernel address handle. 3781 * @ah: The address handle to destroy. 3782 * @flags: Destroy address handle flags (see enum rdma_destroy_ah_flags). 3783 * 3784 * NOTE: for user ah use rdma_destroy_ah_user with valid udata! 3785 */ 3786 static inline void rdma_destroy_ah(struct ib_ah *ah, u32 flags) 3787 { 3788 int ret = rdma_destroy_ah_user(ah, flags, NULL); 3789 3790 WARN_ONCE(ret, "Destroy of kernel AH shouldn't fail"); 3791 } 3792 3793 struct ib_srq *ib_create_srq_user(struct ib_pd *pd, 3794 struct ib_srq_init_attr *srq_init_attr, 3795 struct ib_usrq_object *uobject, 3796 struct ib_udata *udata); 3797 static inline struct ib_srq * 3798 ib_create_srq(struct ib_pd *pd, struct ib_srq_init_attr *srq_init_attr) 3799 { 3800 if (!pd->device->ops.create_srq) 3801 return ERR_PTR(-EOPNOTSUPP); 3802 3803 return ib_create_srq_user(pd, srq_init_attr, NULL, NULL); 3804 } 3805 3806 /** 3807 * ib_modify_srq - Modifies the attributes for the specified SRQ. 3808 * @srq: The SRQ to modify. 3809 * @srq_attr: On input, specifies the SRQ attributes to modify. On output, 3810 * the current values of selected SRQ attributes are returned. 3811 * @srq_attr_mask: A bit-mask used to specify which attributes of the SRQ 3812 * are being modified. 3813 * 3814 * The mask may contain IB_SRQ_MAX_WR to resize the SRQ and/or 3815 * IB_SRQ_LIMIT to set the SRQ's limit and request notification when 3816 * the number of receives queued drops below the limit. 3817 */ 3818 int ib_modify_srq(struct ib_srq *srq, 3819 struct ib_srq_attr *srq_attr, 3820 enum ib_srq_attr_mask srq_attr_mask); 3821 3822 /** 3823 * ib_query_srq - Returns the attribute list and current values for the 3824 * specified SRQ. 3825 * @srq: The SRQ to query. 3826 * @srq_attr: The attributes of the specified SRQ. 3827 */ 3828 int ib_query_srq(struct ib_srq *srq, 3829 struct ib_srq_attr *srq_attr); 3830 3831 /** 3832 * ib_destroy_srq_user - Destroys the specified SRQ. 3833 * @srq: The SRQ to destroy. 3834 * @udata: Valid user data or NULL for kernel objects 3835 */ 3836 int ib_destroy_srq_user(struct ib_srq *srq, struct ib_udata *udata); 3837 3838 /** 3839 * ib_destroy_srq - Destroys the specified kernel SRQ. 3840 * @srq: The SRQ to destroy. 3841 * 3842 * NOTE: for user srq use ib_destroy_srq_user with valid udata! 3843 */ 3844 static inline void ib_destroy_srq(struct ib_srq *srq) 3845 { 3846 int ret = ib_destroy_srq_user(srq, NULL); 3847 3848 WARN_ONCE(ret, "Destroy of kernel SRQ shouldn't fail"); 3849 } 3850 3851 /** 3852 * ib_post_srq_recv - Posts a list of work requests to the specified SRQ. 3853 * @srq: The SRQ to post the work request on. 3854 * @recv_wr: A list of work requests to post on the receive queue. 3855 * @bad_recv_wr: On an immediate failure, this parameter will reference 3856 * the work request that failed to be posted on the QP. 3857 */ 3858 static inline int ib_post_srq_recv(struct ib_srq *srq, 3859 const struct ib_recv_wr *recv_wr, 3860 const struct ib_recv_wr **bad_recv_wr) 3861 { 3862 const struct ib_recv_wr *dummy; 3863 3864 return srq->device->ops.post_srq_recv(srq, recv_wr, 3865 bad_recv_wr ? : &dummy); 3866 } 3867 3868 struct ib_qp *ib_create_qp_kernel(struct ib_pd *pd, 3869 struct ib_qp_init_attr *qp_init_attr, 3870 const char *caller); 3871 /** 3872 * ib_create_qp - Creates a kernel QP associated with the specific protection 3873 * domain. 3874 * @pd: The protection domain associated with the QP. 3875 * @init_attr: A list of initial attributes required to create the 3876 * QP. If QP creation succeeds, then the attributes are updated to 3877 * the actual capabilities of the created QP. 3878 */ 3879 static inline struct ib_qp *ib_create_qp(struct ib_pd *pd, 3880 struct ib_qp_init_attr *init_attr) 3881 { 3882 return ib_create_qp_kernel(pd, init_attr, KBUILD_MODNAME); 3883 } 3884 3885 /** 3886 * ib_modify_qp_with_udata - Modifies the attributes for the specified QP. 3887 * @qp: The QP to modify. 3888 * @attr: On input, specifies the QP attributes to modify. On output, 3889 * the current values of selected QP attributes are returned. 3890 * @attr_mask: A bit-mask used to specify which attributes of the QP 3891 * are being modified. 3892 * @udata: pointer to user's input output buffer information 3893 * are being modified. 3894 * It returns 0 on success and returns appropriate error code on error. 3895 */ 3896 int ib_modify_qp_with_udata(struct ib_qp *qp, 3897 struct ib_qp_attr *attr, 3898 int attr_mask, 3899 struct ib_udata *udata); 3900 3901 /** 3902 * ib_modify_qp - Modifies the attributes for the specified QP and then 3903 * transitions the QP to the given state. 3904 * @qp: The QP to modify. 3905 * @qp_attr: On input, specifies the QP attributes to modify. On output, 3906 * the current values of selected QP attributes are returned. 3907 * @qp_attr_mask: A bit-mask used to specify which attributes of the QP 3908 * are being modified. 3909 */ 3910 int ib_modify_qp(struct ib_qp *qp, 3911 struct ib_qp_attr *qp_attr, 3912 int qp_attr_mask); 3913 3914 /** 3915 * ib_query_qp - Returns the attribute list and current values for the 3916 * specified QP. 3917 * @qp: The QP to query. 3918 * @qp_attr: The attributes of the specified QP. 3919 * @qp_attr_mask: A bit-mask used to select specific attributes to query. 3920 * @qp_init_attr: Additional attributes of the selected QP. 3921 * 3922 * The qp_attr_mask may be used to limit the query to gathering only the 3923 * selected attributes. 3924 */ 3925 int ib_query_qp(struct ib_qp *qp, 3926 struct ib_qp_attr *qp_attr, 3927 int qp_attr_mask, 3928 struct ib_qp_init_attr *qp_init_attr); 3929 3930 /** 3931 * ib_destroy_qp - Destroys the specified QP. 3932 * @qp: The QP to destroy. 3933 * @udata: Valid udata or NULL for kernel objects 3934 */ 3935 int ib_destroy_qp_user(struct ib_qp *qp, struct ib_udata *udata); 3936 3937 /** 3938 * ib_destroy_qp - Destroys the specified kernel QP. 3939 * @qp: The QP to destroy. 3940 * 3941 * NOTE: for user qp use ib_destroy_qp_user with valid udata! 3942 */ 3943 static inline int ib_destroy_qp(struct ib_qp *qp) 3944 { 3945 return ib_destroy_qp_user(qp, NULL); 3946 } 3947 3948 /** 3949 * ib_open_qp - Obtain a reference to an existing sharable QP. 3950 * @xrcd: XRC domain 3951 * @qp_open_attr: Attributes identifying the QP to open. 3952 * 3953 * Returns a reference to a sharable QP. 3954 */ 3955 struct ib_qp *ib_open_qp(struct ib_xrcd *xrcd, 3956 struct ib_qp_open_attr *qp_open_attr); 3957 3958 /** 3959 * ib_close_qp - Release an external reference to a QP. 3960 * @qp: The QP handle to release 3961 * 3962 * The opened QP handle is released by the caller. The underlying 3963 * shared QP is not destroyed until all internal references are released. 3964 */ 3965 int ib_close_qp(struct ib_qp *qp); 3966 3967 /** 3968 * ib_post_send - Posts a list of work requests to the send queue of 3969 * the specified QP. 3970 * @qp: The QP to post the work request on. 3971 * @send_wr: A list of work requests to post on the send queue. 3972 * @bad_send_wr: On an immediate failure, this parameter will reference 3973 * the work request that failed to be posted on the QP. 3974 * 3975 * While IBA Vol. 1 section 11.4.1.1 specifies that if an immediate 3976 * error is returned, the QP state shall not be affected, 3977 * ib_post_send() will return an immediate error after queueing any 3978 * earlier work requests in the list. 3979 */ 3980 static inline int ib_post_send(struct ib_qp *qp, 3981 const struct ib_send_wr *send_wr, 3982 const struct ib_send_wr **bad_send_wr) 3983 { 3984 const struct ib_send_wr *dummy; 3985 3986 return qp->device->ops.post_send(qp, send_wr, bad_send_wr ? : &dummy); 3987 } 3988 3989 /** 3990 * ib_post_recv - Posts a list of work requests to the receive queue of 3991 * the specified QP. 3992 * @qp: The QP to post the work request on. 3993 * @recv_wr: A list of work requests to post on the receive queue. 3994 * @bad_recv_wr: On an immediate failure, this parameter will reference 3995 * the work request that failed to be posted on the QP. 3996 */ 3997 static inline int ib_post_recv(struct ib_qp *qp, 3998 const struct ib_recv_wr *recv_wr, 3999 const struct ib_recv_wr **bad_recv_wr) 4000 { 4001 const struct ib_recv_wr *dummy; 4002 4003 return qp->device->ops.post_recv(qp, recv_wr, bad_recv_wr ? : &dummy); 4004 } 4005 4006 struct ib_cq *__ib_alloc_cq(struct ib_device *dev, void *private, int nr_cqe, 4007 int comp_vector, enum ib_poll_context poll_ctx, 4008 const char *caller); 4009 static inline struct ib_cq *ib_alloc_cq(struct ib_device *dev, void *private, 4010 int nr_cqe, int comp_vector, 4011 enum ib_poll_context poll_ctx) 4012 { 4013 return __ib_alloc_cq(dev, private, nr_cqe, comp_vector, poll_ctx, 4014 KBUILD_MODNAME); 4015 } 4016 4017 struct ib_cq *__ib_alloc_cq_any(struct ib_device *dev, void *private, 4018 int nr_cqe, enum ib_poll_context poll_ctx, 4019 const char *caller); 4020 4021 /** 4022 * ib_alloc_cq_any: Allocate kernel CQ 4023 * @dev: The IB device 4024 * @private: Private data attached to the CQE 4025 * @nr_cqe: Number of CQEs in the CQ 4026 * @poll_ctx: Context used for polling the CQ 4027 */ 4028 static inline struct ib_cq *ib_alloc_cq_any(struct ib_device *dev, 4029 void *private, int nr_cqe, 4030 enum ib_poll_context poll_ctx) 4031 { 4032 return __ib_alloc_cq_any(dev, private, nr_cqe, poll_ctx, 4033 KBUILD_MODNAME); 4034 } 4035 4036 void ib_free_cq(struct ib_cq *cq); 4037 int ib_process_cq_direct(struct ib_cq *cq, int budget); 4038 4039 /** 4040 * ib_create_cq - Creates a CQ on the specified device. 4041 * @device: The device on which to create the CQ. 4042 * @comp_handler: A user-specified callback that is invoked when a 4043 * completion event occurs on the CQ. 4044 * @event_handler: A user-specified callback that is invoked when an 4045 * asynchronous event not associated with a completion occurs on the CQ. 4046 * @cq_context: Context associated with the CQ returned to the user via 4047 * the associated completion and event handlers. 4048 * @cq_attr: The attributes the CQ should be created upon. 4049 * 4050 * Users can examine the cq structure to determine the actual CQ size. 4051 */ 4052 struct ib_cq *__ib_create_cq(struct ib_device *device, 4053 ib_comp_handler comp_handler, 4054 void (*event_handler)(struct ib_event *, void *), 4055 void *cq_context, 4056 const struct ib_cq_init_attr *cq_attr, 4057 const char *caller); 4058 #define ib_create_cq(device, cmp_hndlr, evt_hndlr, cq_ctxt, cq_attr) \ 4059 __ib_create_cq((device), (cmp_hndlr), (evt_hndlr), (cq_ctxt), (cq_attr), KBUILD_MODNAME) 4060 4061 /** 4062 * ib_resize_cq - Modifies the capacity of the CQ. 4063 * @cq: The CQ to resize. 4064 * @cqe: The minimum size of the CQ. 4065 * 4066 * Users can examine the cq structure to determine the actual CQ size. 4067 */ 4068 int ib_resize_cq(struct ib_cq *cq, int cqe); 4069 4070 /** 4071 * rdma_set_cq_moderation - Modifies moderation params of the CQ 4072 * @cq: The CQ to modify. 4073 * @cq_count: number of CQEs that will trigger an event 4074 * @cq_period: max period of time in usec before triggering an event 4075 * 4076 */ 4077 int rdma_set_cq_moderation(struct ib_cq *cq, u16 cq_count, u16 cq_period); 4078 4079 /** 4080 * ib_destroy_cq_user - Destroys the specified CQ. 4081 * @cq: The CQ to destroy. 4082 * @udata: Valid user data or NULL for kernel objects 4083 */ 4084 int ib_destroy_cq_user(struct ib_cq *cq, struct ib_udata *udata); 4085 4086 /** 4087 * ib_destroy_cq - Destroys the specified kernel CQ. 4088 * @cq: The CQ to destroy. 4089 * 4090 * NOTE: for user cq use ib_destroy_cq_user with valid udata! 4091 */ 4092 static inline void ib_destroy_cq(struct ib_cq *cq) 4093 { 4094 int ret = ib_destroy_cq_user(cq, NULL); 4095 4096 WARN_ONCE(ret, "Destroy of kernel CQ shouldn't fail"); 4097 } 4098 4099 /** 4100 * ib_poll_cq - poll a CQ for completion(s) 4101 * @cq:the CQ being polled 4102 * @num_entries:maximum number of completions to return 4103 * @wc:array of at least @num_entries &struct ib_wc where completions 4104 * will be returned 4105 * 4106 * Poll a CQ for (possibly multiple) completions. If the return value 4107 * is < 0, an error occurred. If the return value is >= 0, it is the 4108 * number of completions returned. If the return value is 4109 * non-negative and < num_entries, then the CQ was emptied. 4110 */ 4111 static inline int ib_poll_cq(struct ib_cq *cq, int num_entries, 4112 struct ib_wc *wc) 4113 { 4114 return cq->device->ops.poll_cq(cq, num_entries, wc); 4115 } 4116 4117 /** 4118 * ib_req_notify_cq - Request completion notification on a CQ. 4119 * @cq: The CQ to generate an event for. 4120 * @flags: 4121 * Must contain exactly one of %IB_CQ_SOLICITED or %IB_CQ_NEXT_COMP 4122 * to request an event on the next solicited event or next work 4123 * completion at any type, respectively. %IB_CQ_REPORT_MISSED_EVENTS 4124 * may also be |ed in to request a hint about missed events, as 4125 * described below. 4126 * 4127 * Return Value: 4128 * < 0 means an error occurred while requesting notification 4129 * == 0 means notification was requested successfully, and if 4130 * IB_CQ_REPORT_MISSED_EVENTS was passed in, then no events 4131 * were missed and it is safe to wait for another event. In 4132 * this case is it guaranteed that any work completions added 4133 * to the CQ since the last CQ poll will trigger a completion 4134 * notification event. 4135 * > 0 is only returned if IB_CQ_REPORT_MISSED_EVENTS was passed 4136 * in. It means that the consumer must poll the CQ again to 4137 * make sure it is empty to avoid missing an event because of a 4138 * race between requesting notification and an entry being 4139 * added to the CQ. This return value means it is possible 4140 * (but not guaranteed) that a work completion has been added 4141 * to the CQ since the last poll without triggering a 4142 * completion notification event. 4143 */ 4144 static inline int ib_req_notify_cq(struct ib_cq *cq, 4145 enum ib_cq_notify_flags flags) 4146 { 4147 return cq->device->ops.req_notify_cq(cq, flags); 4148 } 4149 4150 struct ib_cq *ib_cq_pool_get(struct ib_device *dev, unsigned int nr_cqe, 4151 int comp_vector_hint, 4152 enum ib_poll_context poll_ctx); 4153 4154 void ib_cq_pool_put(struct ib_cq *cq, unsigned int nr_cqe); 4155 4156 /* 4157 * Drivers that don't need a DMA mapping at the RDMA layer, set dma_device to 4158 * NULL. This causes the ib_dma* helpers to just stash the kernel virtual 4159 * address into the dma address. 4160 */ 4161 static inline bool ib_uses_virt_dma(struct ib_device *dev) 4162 { 4163 return IS_ENABLED(CONFIG_INFINIBAND_VIRT_DMA) && !dev->dma_device; 4164 } 4165 4166 /* 4167 * Check if a IB device's underlying DMA mapping supports P2PDMA transfers. 4168 */ 4169 static inline bool ib_dma_pci_p2p_dma_supported(struct ib_device *dev) 4170 { 4171 if (ib_uses_virt_dma(dev)) 4172 return false; 4173 4174 return dma_pci_p2pdma_supported(dev->dma_device); 4175 } 4176 4177 /** 4178 * ib_virt_dma_to_ptr - Convert a dma_addr to a kernel pointer 4179 * @dma_addr: The DMA address 4180 * 4181 * Used by ib_uses_virt_dma() devices to get back to the kernel pointer after 4182 * going through the dma_addr marshalling. 4183 */ 4184 static inline void *ib_virt_dma_to_ptr(u64 dma_addr) 4185 { 4186 /* virt_dma mode maps the kvs's directly into the dma addr */ 4187 return (void *)(uintptr_t)dma_addr; 4188 } 4189 4190 /** 4191 * ib_virt_dma_to_page - Convert a dma_addr to a struct page 4192 * @dma_addr: The DMA address 4193 * 4194 * Used by ib_uses_virt_dma() device to get back to the struct page after going 4195 * through the dma_addr marshalling. 4196 */ 4197 static inline struct page *ib_virt_dma_to_page(u64 dma_addr) 4198 { 4199 return virt_to_page(ib_virt_dma_to_ptr(dma_addr)); 4200 } 4201 4202 /** 4203 * ib_dma_mapping_error - check a DMA addr for error 4204 * @dev: The device for which the dma_addr was created 4205 * @dma_addr: The DMA address to check 4206 */ 4207 static inline int ib_dma_mapping_error(struct ib_device *dev, u64 dma_addr) 4208 { 4209 if (ib_uses_virt_dma(dev)) 4210 return 0; 4211 return dma_mapping_error(dev->dma_device, dma_addr); 4212 } 4213 4214 /** 4215 * ib_dma_map_single - Map a kernel virtual address to DMA address 4216 * @dev: The device for which the dma_addr is to be created 4217 * @cpu_addr: The kernel virtual address 4218 * @size: The size of the region in bytes 4219 * @direction: The direction of the DMA 4220 */ 4221 static inline u64 ib_dma_map_single(struct ib_device *dev, 4222 void *cpu_addr, size_t size, 4223 enum dma_data_direction direction) 4224 { 4225 if (ib_uses_virt_dma(dev)) 4226 return (uintptr_t)cpu_addr; 4227 return dma_map_single(dev->dma_device, cpu_addr, size, direction); 4228 } 4229 4230 /** 4231 * ib_dma_unmap_single - Destroy a mapping created by ib_dma_map_single() 4232 * @dev: The device for which the DMA address was created 4233 * @addr: The DMA address 4234 * @size: The size of the region in bytes 4235 * @direction: The direction of the DMA 4236 */ 4237 static inline void ib_dma_unmap_single(struct ib_device *dev, 4238 u64 addr, size_t size, 4239 enum dma_data_direction direction) 4240 { 4241 if (!ib_uses_virt_dma(dev)) 4242 dma_unmap_single(dev->dma_device, addr, size, direction); 4243 } 4244 4245 /** 4246 * ib_dma_map_page - Map a physical page to DMA address 4247 * @dev: The device for which the dma_addr is to be created 4248 * @page: The page to be mapped 4249 * @offset: The offset within the page 4250 * @size: The size of the region in bytes 4251 * @direction: The direction of the DMA 4252 */ 4253 static inline u64 ib_dma_map_page(struct ib_device *dev, 4254 struct page *page, 4255 unsigned long offset, 4256 size_t size, 4257 enum dma_data_direction direction) 4258 { 4259 if (ib_uses_virt_dma(dev)) 4260 return (uintptr_t)(page_address(page) + offset); 4261 return dma_map_page(dev->dma_device, page, offset, size, direction); 4262 } 4263 4264 /** 4265 * ib_dma_unmap_page - Destroy a mapping created by ib_dma_map_page() 4266 * @dev: The device for which the DMA address was created 4267 * @addr: The DMA address 4268 * @size: The size of the region in bytes 4269 * @direction: The direction of the DMA 4270 */ 4271 static inline void ib_dma_unmap_page(struct ib_device *dev, 4272 u64 addr, size_t size, 4273 enum dma_data_direction direction) 4274 { 4275 if (!ib_uses_virt_dma(dev)) 4276 dma_unmap_page(dev->dma_device, addr, size, direction); 4277 } 4278 4279 /** 4280 * ib_dma_map_bvec - Map a bio_vec to DMA address 4281 * @dev: The device for which the dma_addr is to be created 4282 * @bvec: The bio_vec to map 4283 * @direction: The direction of the DMA 4284 * 4285 * Returns a DMA address for the bio_vec. The caller must check the 4286 * result with ib_dma_mapping_error() before use; a failed mapping 4287 * must not be passed to ib_dma_unmap_bvec(). 4288 * 4289 * For software RDMA devices (rxe, siw), returns a virtual address 4290 * and no actual DMA mapping occurs. 4291 */ 4292 static inline u64 ib_dma_map_bvec(struct ib_device *dev, 4293 struct bio_vec *bvec, 4294 enum dma_data_direction direction) 4295 { 4296 if (ib_uses_virt_dma(dev)) 4297 return (uintptr_t)bvec_virt(bvec); 4298 return dma_map_phys(dev->dma_device, bvec_phys(bvec), 4299 bvec->bv_len, direction, 0); 4300 } 4301 4302 /** 4303 * ib_dma_unmap_bvec - Unmap a bio_vec DMA mapping 4304 * @dev: The device for which the DMA address was created 4305 * @addr: The DMA address returned by ib_dma_map_bvec() 4306 * @size: The size of the region in bytes 4307 * @direction: The direction of the DMA 4308 * 4309 * Releases a DMA mapping created by ib_dma_map_bvec(). For software 4310 * RDMA devices this is a no-op since no actual mapping occurred. 4311 */ 4312 static inline void ib_dma_unmap_bvec(struct ib_device *dev, 4313 u64 addr, size_t size, 4314 enum dma_data_direction direction) 4315 { 4316 if (!ib_uses_virt_dma(dev)) 4317 dma_unmap_phys(dev->dma_device, addr, size, direction, 0); 4318 } 4319 4320 int ib_dma_virt_map_sg(struct ib_device *dev, struct scatterlist *sg, int nents); 4321 static inline int ib_dma_map_sg_attrs(struct ib_device *dev, 4322 struct scatterlist *sg, int nents, 4323 enum dma_data_direction direction, 4324 unsigned long dma_attrs) 4325 { 4326 if (ib_uses_virt_dma(dev)) 4327 return ib_dma_virt_map_sg(dev, sg, nents); 4328 return dma_map_sg_attrs(dev->dma_device, sg, nents, direction, 4329 dma_attrs); 4330 } 4331 4332 static inline void ib_dma_unmap_sg_attrs(struct ib_device *dev, 4333 struct scatterlist *sg, int nents, 4334 enum dma_data_direction direction, 4335 unsigned long dma_attrs) 4336 { 4337 if (!ib_uses_virt_dma(dev)) 4338 dma_unmap_sg_attrs(dev->dma_device, sg, nents, direction, 4339 dma_attrs); 4340 } 4341 4342 /** 4343 * ib_dma_map_sgtable_attrs - Map a scatter/gather table to DMA addresses 4344 * @dev: The device for which the DMA addresses are to be created 4345 * @sgt: The sg_table object describing the buffer 4346 * @direction: The direction of the DMA 4347 * @dma_attrs: Optional DMA attributes for the map operation 4348 */ 4349 static inline int ib_dma_map_sgtable_attrs(struct ib_device *dev, 4350 struct sg_table *sgt, 4351 enum dma_data_direction direction, 4352 unsigned long dma_attrs) 4353 { 4354 int nents; 4355 4356 if (ib_uses_virt_dma(dev)) { 4357 nents = ib_dma_virt_map_sg(dev, sgt->sgl, sgt->orig_nents); 4358 if (!nents) 4359 return -EIO; 4360 sgt->nents = nents; 4361 return 0; 4362 } 4363 return dma_map_sgtable(dev->dma_device, sgt, direction, dma_attrs); 4364 } 4365 4366 static inline void ib_dma_unmap_sgtable_attrs(struct ib_device *dev, 4367 struct sg_table *sgt, 4368 enum dma_data_direction direction, 4369 unsigned long dma_attrs) 4370 { 4371 if (!ib_uses_virt_dma(dev)) 4372 dma_unmap_sgtable(dev->dma_device, sgt, direction, dma_attrs); 4373 } 4374 4375 /** 4376 * ib_dma_map_sg - Map a scatter/gather list to DMA addresses 4377 * @dev: The device for which the DMA addresses are to be created 4378 * @sg: The array of scatter/gather entries 4379 * @nents: The number of scatter/gather entries 4380 * @direction: The direction of the DMA 4381 */ 4382 static inline int ib_dma_map_sg(struct ib_device *dev, 4383 struct scatterlist *sg, int nents, 4384 enum dma_data_direction direction) 4385 { 4386 return ib_dma_map_sg_attrs(dev, sg, nents, direction, 0); 4387 } 4388 4389 /** 4390 * ib_dma_unmap_sg - Unmap a scatter/gather list of DMA addresses 4391 * @dev: The device for which the DMA addresses were created 4392 * @sg: The array of scatter/gather entries 4393 * @nents: The number of scatter/gather entries 4394 * @direction: The direction of the DMA 4395 */ 4396 static inline void ib_dma_unmap_sg(struct ib_device *dev, 4397 struct scatterlist *sg, int nents, 4398 enum dma_data_direction direction) 4399 { 4400 ib_dma_unmap_sg_attrs(dev, sg, nents, direction, 0); 4401 } 4402 4403 /** 4404 * ib_dma_max_seg_size - Return the size limit of a single DMA transfer 4405 * @dev: The device to query 4406 * 4407 * The returned value represents a size in bytes. 4408 */ 4409 static inline unsigned int ib_dma_max_seg_size(struct ib_device *dev) 4410 { 4411 if (ib_uses_virt_dma(dev)) 4412 return UINT_MAX; 4413 return dma_get_max_seg_size(dev->dma_device); 4414 } 4415 4416 /** 4417 * ib_dma_sync_single_for_cpu - Prepare DMA region to be accessed by CPU 4418 * @dev: The device for which the DMA address was created 4419 * @addr: The DMA address 4420 * @size: The size of the region in bytes 4421 * @dir: The direction of the DMA 4422 */ 4423 static inline void ib_dma_sync_single_for_cpu(struct ib_device *dev, 4424 u64 addr, 4425 size_t size, 4426 enum dma_data_direction dir) 4427 { 4428 if (!ib_uses_virt_dma(dev)) 4429 dma_sync_single_for_cpu(dev->dma_device, addr, size, dir); 4430 } 4431 4432 /** 4433 * ib_dma_sync_single_for_device - Prepare DMA region to be accessed by device 4434 * @dev: The device for which the DMA address was created 4435 * @addr: The DMA address 4436 * @size: The size of the region in bytes 4437 * @dir: The direction of the DMA 4438 */ 4439 static inline void ib_dma_sync_single_for_device(struct ib_device *dev, 4440 u64 addr, 4441 size_t size, 4442 enum dma_data_direction dir) 4443 { 4444 if (!ib_uses_virt_dma(dev)) 4445 dma_sync_single_for_device(dev->dma_device, addr, size, dir); 4446 } 4447 4448 /* ib_reg_user_mr - register a memory region for virtual addresses from kernel 4449 * space. This function should be called when 'current' is the owning MM. 4450 */ 4451 struct ib_mr *ib_reg_user_mr(struct ib_pd *pd, u64 start, u64 length, 4452 u64 virt_addr, int mr_access_flags); 4453 4454 /* ib_advise_mr - give an advice about an address range in a memory region */ 4455 int ib_advise_mr(struct ib_pd *pd, enum ib_uverbs_advise_mr_advice advice, 4456 u32 flags, struct ib_sge *sg_list, u32 num_sge); 4457 /** 4458 * ib_dereg_mr_user - Deregisters a memory region and removes it from the 4459 * HCA translation table. 4460 * @mr: The memory region to deregister. 4461 * @udata: Valid user data or NULL for kernel object 4462 * 4463 * This function can fail, if the memory region has memory windows bound to it. 4464 */ 4465 int ib_dereg_mr_user(struct ib_mr *mr, struct ib_udata *udata); 4466 4467 /** 4468 * ib_dereg_mr - Deregisters a kernel memory region and removes it from the 4469 * HCA translation table. 4470 * @mr: The memory region to deregister. 4471 * 4472 * This function can fail, if the memory region has memory windows bound to it. 4473 * 4474 * NOTE: for user mr use ib_dereg_mr_user with valid udata! 4475 */ 4476 static inline int ib_dereg_mr(struct ib_mr *mr) 4477 { 4478 return ib_dereg_mr_user(mr, NULL); 4479 } 4480 4481 struct ib_mr *ib_alloc_mr(struct ib_pd *pd, enum ib_mr_type mr_type, 4482 u32 max_num_sg); 4483 4484 struct ib_mr *ib_alloc_mr_integrity(struct ib_pd *pd, 4485 u32 max_num_data_sg, 4486 u32 max_num_meta_sg); 4487 4488 /** 4489 * ib_update_fast_reg_key - updates the key portion of the fast_reg MR 4490 * R_Key and L_Key. 4491 * @mr: struct ib_mr pointer to be updated. 4492 * @newkey: new key to be used. 4493 */ 4494 static inline void ib_update_fast_reg_key(struct ib_mr *mr, u8 newkey) 4495 { 4496 mr->lkey = (mr->lkey & 0xffffff00) | newkey; 4497 mr->rkey = (mr->rkey & 0xffffff00) | newkey; 4498 } 4499 4500 /** 4501 * ib_inc_rkey - increments the key portion of the given rkey. Can be used 4502 * for calculating a new rkey for type 2 memory windows. 4503 * @rkey: the rkey to increment. 4504 */ 4505 static inline u32 ib_inc_rkey(u32 rkey) 4506 { 4507 const u32 mask = 0x000000ff; 4508 return ((rkey + 1) & mask) | (rkey & ~mask); 4509 } 4510 4511 /** 4512 * ib_attach_mcast - Attaches the specified QP to a multicast group. 4513 * @qp: QP to attach to the multicast group. The QP must be type 4514 * IB_QPT_UD. 4515 * @gid: Multicast group GID. 4516 * @lid: Multicast group LID in host byte order. 4517 * 4518 * In order to send and receive multicast packets, subnet 4519 * administration must have created the multicast group and configured 4520 * the fabric appropriately. The port associated with the specified 4521 * QP must also be a member of the multicast group. 4522 */ 4523 int ib_attach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid); 4524 4525 /** 4526 * ib_detach_mcast - Detaches the specified QP from a multicast group. 4527 * @qp: QP to detach from the multicast group. 4528 * @gid: Multicast group GID. 4529 * @lid: Multicast group LID in host byte order. 4530 */ 4531 int ib_detach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid); 4532 4533 struct ib_xrcd *ib_alloc_xrcd_user(struct ib_device *device, 4534 struct inode *inode, struct ib_udata *udata); 4535 int ib_dealloc_xrcd_user(struct ib_xrcd *xrcd, struct ib_udata *udata); 4536 4537 static inline int ib_check_mr_access(struct ib_device *ib_dev, 4538 unsigned int flags) 4539 { 4540 u64 device_cap = ib_dev->attrs.device_cap_flags; 4541 4542 /* 4543 * Local write permission is required if remote write or 4544 * remote atomic permission is also requested. 4545 */ 4546 if (flags & (IB_ACCESS_REMOTE_ATOMIC | IB_ACCESS_REMOTE_WRITE) && 4547 !(flags & IB_ACCESS_LOCAL_WRITE)) 4548 return -EINVAL; 4549 4550 if (flags & ~IB_ACCESS_SUPPORTED) 4551 return -EINVAL; 4552 4553 if (flags & IB_ACCESS_ON_DEMAND && 4554 !(ib_dev->attrs.kernel_cap_flags & IBK_ON_DEMAND_PAGING)) 4555 return -EOPNOTSUPP; 4556 4557 if ((flags & IB_ACCESS_FLUSH_GLOBAL && 4558 !(device_cap & IB_DEVICE_FLUSH_GLOBAL)) || 4559 (flags & IB_ACCESS_FLUSH_PERSISTENT && 4560 !(device_cap & IB_DEVICE_FLUSH_PERSISTENT))) 4561 return -EOPNOTSUPP; 4562 4563 return 0; 4564 } 4565 4566 static inline bool ib_access_writable(int access_flags) 4567 { 4568 /* 4569 * We have writable memory backing the MR if any of the following 4570 * access flags are set. "Local write" and "remote write" obviously 4571 * require write access. "Remote atomic" can do things like fetch and 4572 * add, which will modify memory, and "MW bind" can change permissions 4573 * by binding a window. 4574 */ 4575 return access_flags & 4576 (IB_ACCESS_LOCAL_WRITE | IB_ACCESS_REMOTE_WRITE | 4577 IB_ACCESS_REMOTE_ATOMIC | IB_ACCESS_MW_BIND); 4578 } 4579 4580 /** 4581 * ib_check_mr_status: lightweight check of MR status. 4582 * This routine may provide status checks on a selected 4583 * ib_mr. first use is for signature status check. 4584 * 4585 * @mr: A memory region. 4586 * @check_mask: Bitmask of which checks to perform from 4587 * ib_mr_status_check enumeration. 4588 * @mr_status: The container of relevant status checks. 4589 * failed checks will be indicated in the status bitmask 4590 * and the relevant info shall be in the error item. 4591 */ 4592 int ib_check_mr_status(struct ib_mr *mr, u32 check_mask, 4593 struct ib_mr_status *mr_status); 4594 4595 /** 4596 * ib_device_try_get: Hold a registration lock 4597 * @dev: The device to lock 4598 * 4599 * A device under an active registration lock cannot become unregistered. It 4600 * is only possible to obtain a registration lock on a device that is fully 4601 * registered, otherwise this function returns false. 4602 * 4603 * The registration lock is only necessary for actions which require the 4604 * device to still be registered. Uses that only require the device pointer to 4605 * be valid should use get_device(&ibdev->dev) to hold the memory. 4606 * 4607 */ 4608 static inline bool ib_device_try_get(struct ib_device *dev) 4609 { 4610 return refcount_inc_not_zero(&dev->refcount); 4611 } 4612 4613 void ib_device_put(struct ib_device *device); 4614 struct ib_device *ib_device_get_by_netdev(struct net_device *ndev, 4615 enum rdma_driver_id driver_id); 4616 struct net_device *ib_get_net_dev_by_params(struct ib_device *dev, u32 port, 4617 u16 pkey, const union ib_gid *gid, 4618 const struct sockaddr *addr); 4619 int ib_device_set_netdev(struct ib_device *ib_dev, struct net_device *ndev, 4620 unsigned int port); 4621 struct net_device *ib_device_get_netdev(struct ib_device *ib_dev, 4622 u32 port); 4623 int ib_query_netdev_port(struct ib_device *ibdev, struct net_device *ndev, 4624 u32 *port); 4625 4626 static inline enum ib_port_state ib_get_curr_port_state(struct net_device *net_dev) 4627 { 4628 return (netif_running(net_dev) && netif_carrier_ok(net_dev)) ? 4629 IB_PORT_ACTIVE : IB_PORT_DOWN; 4630 } 4631 4632 void ib_dispatch_port_state_event(struct ib_device *ibdev, 4633 struct net_device *ndev); 4634 struct ib_wq *ib_create_wq(struct ib_pd *pd, 4635 struct ib_wq_init_attr *init_attr); 4636 int ib_destroy_wq_user(struct ib_wq *wq, struct ib_udata *udata); 4637 4638 int ib_map_mr_sg(struct ib_mr *mr, struct scatterlist *sg, int sg_nents, 4639 unsigned int *sg_offset, unsigned int page_size); 4640 int ib_map_mr_sg_pi(struct ib_mr *mr, struct scatterlist *data_sg, 4641 int data_sg_nents, unsigned int *data_sg_offset, 4642 struct scatterlist *meta_sg, int meta_sg_nents, 4643 unsigned int *meta_sg_offset, unsigned int page_size); 4644 4645 static inline int 4646 ib_map_mr_sg_zbva(struct ib_mr *mr, struct scatterlist *sg, int sg_nents, 4647 unsigned int *sg_offset, unsigned int page_size) 4648 { 4649 int n; 4650 4651 n = ib_map_mr_sg(mr, sg, sg_nents, sg_offset, page_size); 4652 mr->iova = 0; 4653 4654 return n; 4655 } 4656 4657 int ib_sg_to_pages(struct ib_mr *mr, struct scatterlist *sgl, int sg_nents, 4658 unsigned int *sg_offset, int (*set_page)(struct ib_mr *, u64)); 4659 4660 void ib_drain_rq(struct ib_qp *qp); 4661 void ib_drain_sq(struct ib_qp *qp); 4662 void ib_drain_qp(struct ib_qp *qp); 4663 4664 int ib_get_eth_speed(struct ib_device *dev, u32 port_num, u16 *speed, 4665 u8 *width); 4666 4667 static inline u8 *rdma_ah_retrieve_dmac(struct rdma_ah_attr *attr) 4668 { 4669 if (attr->type == RDMA_AH_ATTR_TYPE_ROCE) 4670 return attr->roce.dmac; 4671 return NULL; 4672 } 4673 4674 static inline void rdma_ah_set_dlid(struct rdma_ah_attr *attr, u32 dlid) 4675 { 4676 if (attr->type == RDMA_AH_ATTR_TYPE_IB) 4677 attr->ib.dlid = (u16)dlid; 4678 else if (attr->type == RDMA_AH_ATTR_TYPE_OPA) 4679 attr->opa.dlid = dlid; 4680 } 4681 4682 static inline u32 rdma_ah_get_dlid(const struct rdma_ah_attr *attr) 4683 { 4684 if (attr->type == RDMA_AH_ATTR_TYPE_IB) 4685 return attr->ib.dlid; 4686 else if (attr->type == RDMA_AH_ATTR_TYPE_OPA) 4687 return attr->opa.dlid; 4688 return 0; 4689 } 4690 4691 static inline void rdma_ah_set_sl(struct rdma_ah_attr *attr, u8 sl) 4692 { 4693 attr->sl = sl; 4694 } 4695 4696 static inline u8 rdma_ah_get_sl(const struct rdma_ah_attr *attr) 4697 { 4698 return attr->sl; 4699 } 4700 4701 static inline void rdma_ah_set_path_bits(struct rdma_ah_attr *attr, 4702 u8 src_path_bits) 4703 { 4704 if (attr->type == RDMA_AH_ATTR_TYPE_IB) 4705 attr->ib.src_path_bits = src_path_bits; 4706 else if (attr->type == RDMA_AH_ATTR_TYPE_OPA) 4707 attr->opa.src_path_bits = src_path_bits; 4708 } 4709 4710 static inline u8 rdma_ah_get_path_bits(const struct rdma_ah_attr *attr) 4711 { 4712 if (attr->type == RDMA_AH_ATTR_TYPE_IB) 4713 return attr->ib.src_path_bits; 4714 else if (attr->type == RDMA_AH_ATTR_TYPE_OPA) 4715 return attr->opa.src_path_bits; 4716 return 0; 4717 } 4718 4719 static inline void rdma_ah_set_make_grd(struct rdma_ah_attr *attr, 4720 bool make_grd) 4721 { 4722 if (attr->type == RDMA_AH_ATTR_TYPE_OPA) 4723 attr->opa.make_grd = make_grd; 4724 } 4725 4726 static inline bool rdma_ah_get_make_grd(const struct rdma_ah_attr *attr) 4727 { 4728 if (attr->type == RDMA_AH_ATTR_TYPE_OPA) 4729 return attr->opa.make_grd; 4730 return false; 4731 } 4732 4733 static inline void rdma_ah_set_port_num(struct rdma_ah_attr *attr, u32 port_num) 4734 { 4735 attr->port_num = port_num; 4736 } 4737 4738 static inline u32 rdma_ah_get_port_num(const struct rdma_ah_attr *attr) 4739 { 4740 return attr->port_num; 4741 } 4742 4743 static inline void rdma_ah_set_static_rate(struct rdma_ah_attr *attr, 4744 u8 static_rate) 4745 { 4746 attr->static_rate = static_rate; 4747 } 4748 4749 static inline u8 rdma_ah_get_static_rate(const struct rdma_ah_attr *attr) 4750 { 4751 return attr->static_rate; 4752 } 4753 4754 static inline void rdma_ah_set_ah_flags(struct rdma_ah_attr *attr, 4755 enum ib_ah_flags flag) 4756 { 4757 attr->ah_flags = flag; 4758 } 4759 4760 static inline enum ib_ah_flags 4761 rdma_ah_get_ah_flags(const struct rdma_ah_attr *attr) 4762 { 4763 return attr->ah_flags; 4764 } 4765 4766 static inline const struct ib_global_route 4767 *rdma_ah_read_grh(const struct rdma_ah_attr *attr) 4768 { 4769 return &attr->grh; 4770 } 4771 4772 /*To retrieve and modify the grh */ 4773 static inline struct ib_global_route 4774 *rdma_ah_retrieve_grh(struct rdma_ah_attr *attr) 4775 { 4776 return &attr->grh; 4777 } 4778 4779 static inline void rdma_ah_set_dgid_raw(struct rdma_ah_attr *attr, void *dgid) 4780 { 4781 struct ib_global_route *grh = rdma_ah_retrieve_grh(attr); 4782 4783 memcpy(grh->dgid.raw, dgid, sizeof(grh->dgid)); 4784 } 4785 4786 static inline void rdma_ah_set_subnet_prefix(struct rdma_ah_attr *attr, 4787 __be64 prefix) 4788 { 4789 struct ib_global_route *grh = rdma_ah_retrieve_grh(attr); 4790 4791 grh->dgid.global.subnet_prefix = prefix; 4792 } 4793 4794 static inline void rdma_ah_set_interface_id(struct rdma_ah_attr *attr, 4795 __be64 if_id) 4796 { 4797 struct ib_global_route *grh = rdma_ah_retrieve_grh(attr); 4798 4799 grh->dgid.global.interface_id = if_id; 4800 } 4801 4802 static inline void rdma_ah_set_grh(struct rdma_ah_attr *attr, 4803 union ib_gid *dgid, u32 flow_label, 4804 u8 sgid_index, u8 hop_limit, 4805 u8 traffic_class) 4806 { 4807 struct ib_global_route *grh = rdma_ah_retrieve_grh(attr); 4808 4809 attr->ah_flags = IB_AH_GRH; 4810 if (dgid) 4811 grh->dgid = *dgid; 4812 grh->flow_label = flow_label; 4813 grh->sgid_index = sgid_index; 4814 grh->hop_limit = hop_limit; 4815 grh->traffic_class = traffic_class; 4816 grh->sgid_attr = NULL; 4817 } 4818 4819 void rdma_destroy_ah_attr(struct rdma_ah_attr *ah_attr); 4820 void rdma_move_grh_sgid_attr(struct rdma_ah_attr *attr, union ib_gid *dgid, 4821 u32 flow_label, u8 hop_limit, u8 traffic_class, 4822 const struct ib_gid_attr *sgid_attr); 4823 void rdma_copy_ah_attr(struct rdma_ah_attr *dest, 4824 const struct rdma_ah_attr *src); 4825 void rdma_replace_ah_attr(struct rdma_ah_attr *old, 4826 const struct rdma_ah_attr *new); 4827 void rdma_move_ah_attr(struct rdma_ah_attr *dest, struct rdma_ah_attr *src); 4828 4829 /** 4830 * rdma_ah_find_type - Return address handle type. 4831 * 4832 * @dev: Device to be checked 4833 * @port_num: Port number 4834 */ 4835 static inline enum rdma_ah_attr_type rdma_ah_find_type(struct ib_device *dev, 4836 u32 port_num) 4837 { 4838 if (rdma_protocol_roce(dev, port_num)) 4839 return RDMA_AH_ATTR_TYPE_ROCE; 4840 if (rdma_protocol_ib(dev, port_num)) { 4841 if (rdma_cap_opa_ah(dev, port_num)) 4842 return RDMA_AH_ATTR_TYPE_OPA; 4843 return RDMA_AH_ATTR_TYPE_IB; 4844 } 4845 if (dev->type == RDMA_DEVICE_TYPE_SMI) 4846 return RDMA_AH_ATTR_TYPE_IB; 4847 4848 return RDMA_AH_ATTR_TYPE_UNDEFINED; 4849 } 4850 4851 /** 4852 * ib_lid_cpu16 - Return lid in 16bit CPU encoding. 4853 * In the current implementation the only way to 4854 * get the 32bit lid is from other sources for OPA. 4855 * For IB, lids will always be 16bits so cast the 4856 * value accordingly. 4857 * 4858 * @lid: A 32bit LID 4859 */ 4860 static inline u16 ib_lid_cpu16(u32 lid) 4861 { 4862 WARN_ON_ONCE(lid & 0xFFFF0000); 4863 return (u16)lid; 4864 } 4865 4866 /** 4867 * ib_lid_be16 - Return lid in 16bit BE encoding. 4868 * 4869 * @lid: A 32bit LID 4870 */ 4871 static inline __be16 ib_lid_be16(u32 lid) 4872 { 4873 WARN_ON_ONCE(lid & 0xFFFF0000); 4874 return cpu_to_be16((u16)lid); 4875 } 4876 4877 /** 4878 * ib_get_vector_affinity - Get the affinity mappings of a given completion 4879 * vector 4880 * @device: the rdma device 4881 * @comp_vector: index of completion vector 4882 * 4883 * Returns NULL on failure, otherwise a corresponding cpu map of the 4884 * completion vector (returns all-cpus map if the device driver doesn't 4885 * implement get_vector_affinity). 4886 */ 4887 static inline const struct cpumask * 4888 ib_get_vector_affinity(struct ib_device *device, int comp_vector) 4889 { 4890 if (comp_vector < 0 || comp_vector >= device->num_comp_vectors || 4891 !device->ops.get_vector_affinity) 4892 return NULL; 4893 4894 return device->ops.get_vector_affinity(device, comp_vector); 4895 4896 } 4897 4898 /** 4899 * rdma_roce_rescan_device - Rescan all of the network devices in the system 4900 * and add their gids, as needed, to the relevant RoCE devices. 4901 * 4902 * @ibdev: the rdma device 4903 */ 4904 void rdma_roce_rescan_device(struct ib_device *ibdev); 4905 void rdma_roce_rescan_port(struct ib_device *ib_dev, u32 port); 4906 void roce_del_all_netdev_gids(struct ib_device *ib_dev, 4907 u32 port, struct net_device *ndev); 4908 4909 struct ib_ucontext *ib_uverbs_get_ucontext_file(struct ib_uverbs_file *ufile); 4910 4911 #if IS_ENABLED(CONFIG_INFINIBAND_USER_ACCESS) 4912 int uverbs_destroy_def_handler(struct uverbs_attr_bundle *attrs); 4913 bool rdma_uattrs_has_raw_cap(const struct uverbs_attr_bundle *attrs); 4914 #else 4915 static inline int uverbs_destroy_def_handler(struct uverbs_attr_bundle *attrs) 4916 { 4917 return 0; 4918 } 4919 static inline bool 4920 rdma_uattrs_has_raw_cap(const struct uverbs_attr_bundle *attrs) 4921 { 4922 return false; 4923 } 4924 #endif 4925 4926 struct net_device *rdma_alloc_netdev(struct ib_device *device, u32 port_num, 4927 enum rdma_netdev_t type, const char *name, 4928 unsigned char name_assign_type, 4929 void (*setup)(struct net_device *)); 4930 4931 int rdma_init_netdev(struct ib_device *device, u32 port_num, 4932 enum rdma_netdev_t type, const char *name, 4933 unsigned char name_assign_type, 4934 void (*setup)(struct net_device *), 4935 struct net_device *netdev); 4936 4937 /** 4938 * rdma_device_to_ibdev - Get ib_device pointer from device pointer 4939 * 4940 * @device: device pointer for which ib_device pointer to retrieve 4941 * 4942 * rdma_device_to_ibdev() retrieves ib_device pointer from device. 4943 * 4944 */ 4945 static inline struct ib_device *rdma_device_to_ibdev(struct device *device) 4946 { 4947 struct ib_core_device *coredev = 4948 container_of(device, struct ib_core_device, dev); 4949 4950 return coredev->owner; 4951 } 4952 4953 /** 4954 * ibdev_to_node - return the NUMA node for a given ib_device 4955 * @ibdev: device to get the NUMA node for. 4956 */ 4957 static inline int ibdev_to_node(struct ib_device *ibdev) 4958 { 4959 struct device *parent = ibdev->dev.parent; 4960 4961 if (!parent) 4962 return NUMA_NO_NODE; 4963 return dev_to_node(parent); 4964 } 4965 4966 /** 4967 * rdma_device_to_drv_device - Helper macro to reach back to driver's 4968 * ib_device holder structure from device pointer. 4969 * 4970 * NOTE: New drivers should not make use of this API; This API is only for 4971 * existing drivers who have exposed sysfs entries using 4972 * ops->device_group. 4973 */ 4974 #define rdma_device_to_drv_device(dev, drv_dev_struct, ibdev_member) \ 4975 container_of(rdma_device_to_ibdev(dev), drv_dev_struct, ibdev_member) 4976 4977 bool rdma_dev_access_netns(const struct ib_device *device, 4978 const struct net *net); 4979 4980 bool rdma_dev_has_raw_cap(const struct ib_device *dev); 4981 static inline struct net *rdma_dev_net(struct ib_device *device) 4982 { 4983 return read_pnet(&device->coredev.rdma_net); 4984 } 4985 4986 #define IB_ROCE_UDP_ENCAP_VALID_PORT_MIN (0xC000) 4987 #define IB_ROCE_UDP_ENCAP_VALID_PORT_MAX (0xFFFF) 4988 #define IB_GRH_FLOWLABEL_MASK (0x000FFFFF) 4989 4990 /** 4991 * rdma_flow_label_to_udp_sport - generate a RoCE v2 UDP src port value based 4992 * on the flow_label 4993 * @fl: flow_label value 4994 * 4995 * This function will convert the 20 bit flow_label input to a valid RoCE v2 4996 * UDP src port 14 bit value. All RoCE V2 drivers should use this same 4997 * convention. 4998 */ 4999 static inline u16 rdma_flow_label_to_udp_sport(u32 fl) 5000 { 5001 u32 fl_low = fl & 0x03fff, fl_high = fl & 0xFC000; 5002 5003 fl_low ^= fl_high >> 14; 5004 return (u16)(fl_low | IB_ROCE_UDP_ENCAP_VALID_PORT_MIN); 5005 } 5006 5007 /** 5008 * rdma_calc_flow_label - generate a RDMA symmetric flow label value based on 5009 * local and remote qpn values 5010 * 5011 * This function folded the multiplication results of two qpns, 24 bit each, 5012 * fields, and converts it to a 20 bit results. 5013 * 5014 * This function will create symmetric flow_label value based on the local 5015 * and remote qpn values. this will allow both the requester and responder 5016 * to calculate the same flow_label for a given connection. 5017 * 5018 * This helper function should be used by driver in case the upper layer 5019 * provide a zero flow_label value. This is to improve entropy of RDMA 5020 * traffic in the network. 5021 */ 5022 static inline u32 rdma_calc_flow_label(u32 lqpn, u32 rqpn) 5023 { 5024 u64 v = (u64)lqpn * rqpn; 5025 5026 v ^= v >> 20; 5027 v ^= v >> 40; 5028 5029 return (u32)(v & IB_GRH_FLOWLABEL_MASK); 5030 } 5031 5032 /** 5033 * rdma_get_udp_sport - Calculate and set UDP source port based on the flow 5034 * label. If flow label is not defined in GRH then 5035 * calculate it based on lqpn/rqpn. 5036 * 5037 * @fl: flow label from GRH 5038 * @lqpn: local qp number 5039 * @rqpn: remote qp number 5040 */ 5041 static inline u16 rdma_get_udp_sport(u32 fl, u32 lqpn, u32 rqpn) 5042 { 5043 if (!fl) 5044 fl = rdma_calc_flow_label(lqpn, rqpn); 5045 5046 return rdma_flow_label_to_udp_sport(fl); 5047 } 5048 5049 const struct ib_port_immutable* 5050 ib_port_immutable_read(struct ib_device *dev, unsigned int port); 5051 5052 /** ib_add_sub_device - Add a sub IB device on an existing one 5053 * 5054 * @parent: The IB device that needs to add a sub device 5055 * @type: The type of the new sub device 5056 * @name: The name of the new sub device 5057 * 5058 * 5059 * Return 0 on success, an error code otherwise 5060 */ 5061 int ib_add_sub_device(struct ib_device *parent, 5062 enum rdma_nl_dev_type type, 5063 const char *name); 5064 5065 5066 /** ib_del_sub_device_and_put - Delect an IB sub device while holding a 'get' 5067 * 5068 * @sub: The sub device that is going to be deleted 5069 * 5070 * Return 0 on success, an error code otherwise 5071 */ 5072 int ib_del_sub_device_and_put(struct ib_device *sub); 5073 5074 static inline void ib_mark_name_assigned_by_user(struct ib_device *ibdev) 5075 { 5076 ibdev->name_assign_type = RDMA_NAME_ASSIGN_TYPE_USER; 5077 } 5078 5079 #endif /* IB_VERBS_H */ 5080