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