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 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/types.h> 16 #include <linux/device.h> 17 #include <linux/dma-mapping.h> 18 #include <linux/kref.h> 19 #include <linux/list.h> 20 #include <linux/rwsem.h> 21 #include <linux/workqueue.h> 22 #include <linux/irq_poll.h> 23 #include <uapi/linux/if_ether.h> 24 #include <net/ipv6.h> 25 #include <net/ip.h> 26 #include <linux/string.h> 27 #include <linux/slab.h> 28 #include <linux/netdevice.h> 29 #include <linux/refcount.h> 30 #include <linux/if_link.h> 31 #include <linux/atomic.h> 32 #include <linux/mmu_notifier.h> 33 #include <linux/uaccess.h> 34 #include <linux/cgroup_rdma.h> 35 #include <linux/irqflags.h> 36 #include <linux/preempt.h> 37 #include <linux/dim.h> 38 #include <uapi/rdma/ib_user_verbs.h> 39 #include <rdma/rdma_counter.h> 40 #include <rdma/restrack.h> 41 #include <rdma/signature.h> 42 #include <uapi/rdma/rdma_user_ioctl.h> 43 #include <uapi/rdma/ib_user_ioctl_verbs.h> 44 45 #define IB_FW_VERSION_NAME_MAX ETHTOOL_FWVERS_LEN 46 47 struct ib_umem_odp; 48 struct ib_uqp_object; 49 struct ib_usrq_object; 50 struct ib_uwq_object; 51 struct rdma_cm_id; 52 53 extern struct workqueue_struct *ib_wq; 54 extern struct workqueue_struct *ib_comp_wq; 55 extern struct workqueue_struct *ib_comp_unbound_wq; 56 57 struct ib_ucq_object; 58 59 __printf(3, 4) __cold 60 void ibdev_printk(const char *level, const struct ib_device *ibdev, 61 const char *format, ...); 62 __printf(2, 3) __cold 63 void ibdev_emerg(const struct ib_device *ibdev, const char *format, ...); 64 __printf(2, 3) __cold 65 void ibdev_alert(const struct ib_device *ibdev, const char *format, ...); 66 __printf(2, 3) __cold 67 void ibdev_crit(const struct ib_device *ibdev, const char *format, ...); 68 __printf(2, 3) __cold 69 void ibdev_err(const struct ib_device *ibdev, const char *format, ...); 70 __printf(2, 3) __cold 71 void ibdev_warn(const struct ib_device *ibdev, const char *format, ...); 72 __printf(2, 3) __cold 73 void ibdev_notice(const struct ib_device *ibdev, const char *format, ...); 74 __printf(2, 3) __cold 75 void ibdev_info(const struct ib_device *ibdev, const char *format, ...); 76 77 #if defined(CONFIG_DYNAMIC_DEBUG) || \ 78 (defined(CONFIG_DYNAMIC_DEBUG_CORE) && defined(DYNAMIC_DEBUG_MODULE)) 79 #define ibdev_dbg(__dev, format, args...) \ 80 dynamic_ibdev_dbg(__dev, format, ##args) 81 #else 82 __printf(2, 3) __cold 83 static inline 84 void ibdev_dbg(const struct ib_device *ibdev, const char *format, ...) {} 85 #endif 86 87 #define ibdev_level_ratelimited(ibdev_level, ibdev, fmt, ...) \ 88 do { \ 89 static DEFINE_RATELIMIT_STATE(_rs, \ 90 DEFAULT_RATELIMIT_INTERVAL, \ 91 DEFAULT_RATELIMIT_BURST); \ 92 if (__ratelimit(&_rs)) \ 93 ibdev_level(ibdev, fmt, ##__VA_ARGS__); \ 94 } while (0) 95 96 #define ibdev_emerg_ratelimited(ibdev, fmt, ...) \ 97 ibdev_level_ratelimited(ibdev_emerg, ibdev, fmt, ##__VA_ARGS__) 98 #define ibdev_alert_ratelimited(ibdev, fmt, ...) \ 99 ibdev_level_ratelimited(ibdev_alert, ibdev, fmt, ##__VA_ARGS__) 100 #define ibdev_crit_ratelimited(ibdev, fmt, ...) \ 101 ibdev_level_ratelimited(ibdev_crit, ibdev, fmt, ##__VA_ARGS__) 102 #define ibdev_err_ratelimited(ibdev, fmt, ...) \ 103 ibdev_level_ratelimited(ibdev_err, ibdev, fmt, ##__VA_ARGS__) 104 #define ibdev_warn_ratelimited(ibdev, fmt, ...) \ 105 ibdev_level_ratelimited(ibdev_warn, ibdev, fmt, ##__VA_ARGS__) 106 #define ibdev_notice_ratelimited(ibdev, fmt, ...) \ 107 ibdev_level_ratelimited(ibdev_notice, ibdev, fmt, ##__VA_ARGS__) 108 #define ibdev_info_ratelimited(ibdev, fmt, ...) \ 109 ibdev_level_ratelimited(ibdev_info, ibdev, fmt, ##__VA_ARGS__) 110 111 #if defined(CONFIG_DYNAMIC_DEBUG) || \ 112 (defined(CONFIG_DYNAMIC_DEBUG_CORE) && defined(DYNAMIC_DEBUG_MODULE)) 113 /* descriptor check is first to prevent flooding with "callbacks suppressed" */ 114 #define ibdev_dbg_ratelimited(ibdev, fmt, ...) \ 115 do { \ 116 static DEFINE_RATELIMIT_STATE(_rs, \ 117 DEFAULT_RATELIMIT_INTERVAL, \ 118 DEFAULT_RATELIMIT_BURST); \ 119 DEFINE_DYNAMIC_DEBUG_METADATA(descriptor, fmt); \ 120 if (DYNAMIC_DEBUG_BRANCH(descriptor) && __ratelimit(&_rs)) \ 121 __dynamic_ibdev_dbg(&descriptor, ibdev, fmt, \ 122 ##__VA_ARGS__); \ 123 } while (0) 124 #else 125 __printf(2, 3) __cold 126 static inline 127 void ibdev_dbg_ratelimited(const struct ib_device *ibdev, const char *format, ...) {} 128 #endif 129 130 union ib_gid { 131 u8 raw[16]; 132 struct { 133 __be64 subnet_prefix; 134 __be64 interface_id; 135 } global; 136 }; 137 138 extern union ib_gid zgid; 139 140 enum ib_gid_type { 141 /* If link layer is Ethernet, this is RoCE V1 */ 142 IB_GID_TYPE_IB = 0, 143 IB_GID_TYPE_ROCE = 0, 144 IB_GID_TYPE_ROCE_UDP_ENCAP = 1, 145 IB_GID_TYPE_SIZE 146 }; 147 148 #define ROCE_V2_UDP_DPORT 4791 149 struct ib_gid_attr { 150 struct net_device __rcu *ndev; 151 struct ib_device *device; 152 union ib_gid gid; 153 enum ib_gid_type gid_type; 154 u16 index; 155 u8 port_num; 156 }; 157 158 enum { 159 /* set the local administered indication */ 160 IB_SA_WELL_KNOWN_GUID = BIT_ULL(57) | 2, 161 }; 162 163 enum rdma_transport_type { 164 RDMA_TRANSPORT_IB, 165 RDMA_TRANSPORT_IWARP, 166 RDMA_TRANSPORT_USNIC, 167 RDMA_TRANSPORT_USNIC_UDP, 168 RDMA_TRANSPORT_UNSPECIFIED, 169 }; 170 171 enum rdma_protocol_type { 172 RDMA_PROTOCOL_IB, 173 RDMA_PROTOCOL_IBOE, 174 RDMA_PROTOCOL_IWARP, 175 RDMA_PROTOCOL_USNIC_UDP 176 }; 177 178 __attribute_const__ enum rdma_transport_type 179 rdma_node_get_transport(unsigned int node_type); 180 181 enum rdma_network_type { 182 RDMA_NETWORK_IB, 183 RDMA_NETWORK_ROCE_V1 = RDMA_NETWORK_IB, 184 RDMA_NETWORK_IPV4, 185 RDMA_NETWORK_IPV6 186 }; 187 188 static inline enum ib_gid_type ib_network_to_gid_type(enum rdma_network_type network_type) 189 { 190 if (network_type == RDMA_NETWORK_IPV4 || 191 network_type == RDMA_NETWORK_IPV6) 192 return IB_GID_TYPE_ROCE_UDP_ENCAP; 193 194 /* IB_GID_TYPE_IB same as RDMA_NETWORK_ROCE_V1 */ 195 return IB_GID_TYPE_IB; 196 } 197 198 static inline enum rdma_network_type 199 rdma_gid_attr_network_type(const struct ib_gid_attr *attr) 200 { 201 if (attr->gid_type == IB_GID_TYPE_IB) 202 return RDMA_NETWORK_IB; 203 204 if (ipv6_addr_v4mapped((struct in6_addr *)&attr->gid)) 205 return RDMA_NETWORK_IPV4; 206 else 207 return RDMA_NETWORK_IPV6; 208 } 209 210 enum rdma_link_layer { 211 IB_LINK_LAYER_UNSPECIFIED, 212 IB_LINK_LAYER_INFINIBAND, 213 IB_LINK_LAYER_ETHERNET, 214 }; 215 216 enum ib_device_cap_flags { 217 IB_DEVICE_RESIZE_MAX_WR = (1 << 0), 218 IB_DEVICE_BAD_PKEY_CNTR = (1 << 1), 219 IB_DEVICE_BAD_QKEY_CNTR = (1 << 2), 220 IB_DEVICE_RAW_MULTI = (1 << 3), 221 IB_DEVICE_AUTO_PATH_MIG = (1 << 4), 222 IB_DEVICE_CHANGE_PHY_PORT = (1 << 5), 223 IB_DEVICE_UD_AV_PORT_ENFORCE = (1 << 6), 224 IB_DEVICE_CURR_QP_STATE_MOD = (1 << 7), 225 IB_DEVICE_SHUTDOWN_PORT = (1 << 8), 226 /* Not in use, former INIT_TYPE = (1 << 9),*/ 227 IB_DEVICE_PORT_ACTIVE_EVENT = (1 << 10), 228 IB_DEVICE_SYS_IMAGE_GUID = (1 << 11), 229 IB_DEVICE_RC_RNR_NAK_GEN = (1 << 12), 230 IB_DEVICE_SRQ_RESIZE = (1 << 13), 231 IB_DEVICE_N_NOTIFY_CQ = (1 << 14), 232 233 /* 234 * This device supports a per-device lkey or stag that can be 235 * used without performing a memory registration for the local 236 * memory. Note that ULPs should never check this flag, but 237 * instead of use the local_dma_lkey flag in the ib_pd structure, 238 * which will always contain a usable lkey. 239 */ 240 IB_DEVICE_LOCAL_DMA_LKEY = (1 << 15), 241 /* Reserved, old SEND_W_INV = (1 << 16),*/ 242 IB_DEVICE_MEM_WINDOW = (1 << 17), 243 /* 244 * Devices should set IB_DEVICE_UD_IP_SUM if they support 245 * insertion of UDP and TCP checksum on outgoing UD IPoIB 246 * messages and can verify the validity of checksum for 247 * incoming messages. Setting this flag implies that the 248 * IPoIB driver may set NETIF_F_IP_CSUM for datagram mode. 249 */ 250 IB_DEVICE_UD_IP_CSUM = (1 << 18), 251 IB_DEVICE_UD_TSO = (1 << 19), 252 IB_DEVICE_XRC = (1 << 20), 253 254 /* 255 * This device supports the IB "base memory management extension", 256 * which includes support for fast registrations (IB_WR_REG_MR, 257 * IB_WR_LOCAL_INV and IB_WR_SEND_WITH_INV verbs). This flag should 258 * also be set by any iWarp device which must support FRs to comply 259 * to the iWarp verbs spec. iWarp devices also support the 260 * IB_WR_RDMA_READ_WITH_INV verb for RDMA READs that invalidate the 261 * stag. 262 */ 263 IB_DEVICE_MEM_MGT_EXTENSIONS = (1 << 21), 264 IB_DEVICE_BLOCK_MULTICAST_LOOPBACK = (1 << 22), 265 IB_DEVICE_MEM_WINDOW_TYPE_2A = (1 << 23), 266 IB_DEVICE_MEM_WINDOW_TYPE_2B = (1 << 24), 267 IB_DEVICE_RC_IP_CSUM = (1 << 25), 268 /* Deprecated. Please use IB_RAW_PACKET_CAP_IP_CSUM. */ 269 IB_DEVICE_RAW_IP_CSUM = (1 << 26), 270 /* 271 * Devices should set IB_DEVICE_CROSS_CHANNEL if they 272 * support execution of WQEs that involve synchronization 273 * of I/O operations with single completion queue managed 274 * by hardware. 275 */ 276 IB_DEVICE_CROSS_CHANNEL = (1 << 27), 277 IB_DEVICE_MANAGED_FLOW_STEERING = (1 << 29), 278 IB_DEVICE_INTEGRITY_HANDOVER = (1 << 30), 279 IB_DEVICE_ON_DEMAND_PAGING = (1ULL << 31), 280 IB_DEVICE_SG_GAPS_REG = (1ULL << 32), 281 IB_DEVICE_VIRTUAL_FUNCTION = (1ULL << 33), 282 /* Deprecated. Please use IB_RAW_PACKET_CAP_SCATTER_FCS. */ 283 IB_DEVICE_RAW_SCATTER_FCS = (1ULL << 34), 284 IB_DEVICE_RDMA_NETDEV_OPA = (1ULL << 35), 285 /* The device supports padding incoming writes to cacheline. */ 286 IB_DEVICE_PCI_WRITE_END_PADDING = (1ULL << 36), 287 IB_DEVICE_ALLOW_USER_UNREG = (1ULL << 37), 288 }; 289 290 enum ib_atomic_cap { 291 IB_ATOMIC_NONE, 292 IB_ATOMIC_HCA, 293 IB_ATOMIC_GLOB 294 }; 295 296 enum ib_odp_general_cap_bits { 297 IB_ODP_SUPPORT = 1 << 0, 298 IB_ODP_SUPPORT_IMPLICIT = 1 << 1, 299 }; 300 301 enum ib_odp_transport_cap_bits { 302 IB_ODP_SUPPORT_SEND = 1 << 0, 303 IB_ODP_SUPPORT_RECV = 1 << 1, 304 IB_ODP_SUPPORT_WRITE = 1 << 2, 305 IB_ODP_SUPPORT_READ = 1 << 3, 306 IB_ODP_SUPPORT_ATOMIC = 1 << 4, 307 IB_ODP_SUPPORT_SRQ_RECV = 1 << 5, 308 }; 309 310 struct ib_odp_caps { 311 uint64_t general_caps; 312 struct { 313 uint32_t rc_odp_caps; 314 uint32_t uc_odp_caps; 315 uint32_t ud_odp_caps; 316 uint32_t xrc_odp_caps; 317 } per_transport_caps; 318 }; 319 320 struct ib_rss_caps { 321 /* Corresponding bit will be set if qp type from 322 * 'enum ib_qp_type' is supported, e.g. 323 * supported_qpts |= 1 << IB_QPT_UD 324 */ 325 u32 supported_qpts; 326 u32 max_rwq_indirection_tables; 327 u32 max_rwq_indirection_table_size; 328 }; 329 330 enum ib_tm_cap_flags { 331 /* Support tag matching with rendezvous offload for RC transport */ 332 IB_TM_CAP_RNDV_RC = 1 << 0, 333 }; 334 335 struct ib_tm_caps { 336 /* Max size of RNDV header */ 337 u32 max_rndv_hdr_size; 338 /* Max number of entries in tag matching list */ 339 u32 max_num_tags; 340 /* From enum ib_tm_cap_flags */ 341 u32 flags; 342 /* Max number of outstanding list operations */ 343 u32 max_ops; 344 /* Max number of SGE in tag matching entry */ 345 u32 max_sge; 346 }; 347 348 struct ib_cq_init_attr { 349 unsigned int cqe; 350 u32 comp_vector; 351 u32 flags; 352 }; 353 354 enum ib_cq_attr_mask { 355 IB_CQ_MODERATE = 1 << 0, 356 }; 357 358 struct ib_cq_caps { 359 u16 max_cq_moderation_count; 360 u16 max_cq_moderation_period; 361 }; 362 363 struct ib_dm_mr_attr { 364 u64 length; 365 u64 offset; 366 u32 access_flags; 367 }; 368 369 struct ib_dm_alloc_attr { 370 u64 length; 371 u32 alignment; 372 u32 flags; 373 }; 374 375 struct ib_device_attr { 376 u64 fw_ver; 377 __be64 sys_image_guid; 378 u64 max_mr_size; 379 u64 page_size_cap; 380 u32 vendor_id; 381 u32 vendor_part_id; 382 u32 hw_ver; 383 int max_qp; 384 int max_qp_wr; 385 u64 device_cap_flags; 386 int max_send_sge; 387 int max_recv_sge; 388 int max_sge_rd; 389 int max_cq; 390 int max_cqe; 391 int max_mr; 392 int max_pd; 393 int max_qp_rd_atom; 394 int max_ee_rd_atom; 395 int max_res_rd_atom; 396 int max_qp_init_rd_atom; 397 int max_ee_init_rd_atom; 398 enum ib_atomic_cap atomic_cap; 399 enum ib_atomic_cap masked_atomic_cap; 400 int max_ee; 401 int max_rdd; 402 int max_mw; 403 int max_raw_ipv6_qp; 404 int max_raw_ethy_qp; 405 int max_mcast_grp; 406 int max_mcast_qp_attach; 407 int max_total_mcast_qp_attach; 408 int max_ah; 409 int max_srq; 410 int max_srq_wr; 411 int max_srq_sge; 412 unsigned int max_fast_reg_page_list_len; 413 unsigned int max_pi_fast_reg_page_list_len; 414 u16 max_pkeys; 415 u8 local_ca_ack_delay; 416 int sig_prot_cap; 417 int sig_guard_cap; 418 struct ib_odp_caps odp_caps; 419 uint64_t timestamp_mask; 420 uint64_t hca_core_clock; /* in KHZ */ 421 struct ib_rss_caps rss_caps; 422 u32 max_wq_type_rq; 423 u32 raw_packet_caps; /* Use ib_raw_packet_caps enum */ 424 struct ib_tm_caps tm_caps; 425 struct ib_cq_caps cq_caps; 426 u64 max_dm_size; 427 /* Max entries for sgl for optimized performance per READ */ 428 u32 max_sgl_rd; 429 }; 430 431 enum ib_mtu { 432 IB_MTU_256 = 1, 433 IB_MTU_512 = 2, 434 IB_MTU_1024 = 3, 435 IB_MTU_2048 = 4, 436 IB_MTU_4096 = 5 437 }; 438 439 enum opa_mtu { 440 OPA_MTU_8192 = 6, 441 OPA_MTU_10240 = 7 442 }; 443 444 static inline int ib_mtu_enum_to_int(enum ib_mtu mtu) 445 { 446 switch (mtu) { 447 case IB_MTU_256: return 256; 448 case IB_MTU_512: return 512; 449 case IB_MTU_1024: return 1024; 450 case IB_MTU_2048: return 2048; 451 case IB_MTU_4096: return 4096; 452 default: return -1; 453 } 454 } 455 456 static inline enum ib_mtu ib_mtu_int_to_enum(int mtu) 457 { 458 if (mtu >= 4096) 459 return IB_MTU_4096; 460 else if (mtu >= 2048) 461 return IB_MTU_2048; 462 else if (mtu >= 1024) 463 return IB_MTU_1024; 464 else if (mtu >= 512) 465 return IB_MTU_512; 466 else 467 return IB_MTU_256; 468 } 469 470 static inline int opa_mtu_enum_to_int(enum opa_mtu mtu) 471 { 472 switch (mtu) { 473 case OPA_MTU_8192: 474 return 8192; 475 case OPA_MTU_10240: 476 return 10240; 477 default: 478 return(ib_mtu_enum_to_int((enum ib_mtu)mtu)); 479 } 480 } 481 482 static inline enum opa_mtu opa_mtu_int_to_enum(int mtu) 483 { 484 if (mtu >= 10240) 485 return OPA_MTU_10240; 486 else if (mtu >= 8192) 487 return OPA_MTU_8192; 488 else 489 return ((enum opa_mtu)ib_mtu_int_to_enum(mtu)); 490 } 491 492 enum ib_port_state { 493 IB_PORT_NOP = 0, 494 IB_PORT_DOWN = 1, 495 IB_PORT_INIT = 2, 496 IB_PORT_ARMED = 3, 497 IB_PORT_ACTIVE = 4, 498 IB_PORT_ACTIVE_DEFER = 5 499 }; 500 501 enum ib_port_phys_state { 502 IB_PORT_PHYS_STATE_SLEEP = 1, 503 IB_PORT_PHYS_STATE_POLLING = 2, 504 IB_PORT_PHYS_STATE_DISABLED = 3, 505 IB_PORT_PHYS_STATE_PORT_CONFIGURATION_TRAINING = 4, 506 IB_PORT_PHYS_STATE_LINK_UP = 5, 507 IB_PORT_PHYS_STATE_LINK_ERROR_RECOVERY = 6, 508 IB_PORT_PHYS_STATE_PHY_TEST = 7, 509 }; 510 511 enum ib_port_width { 512 IB_WIDTH_1X = 1, 513 IB_WIDTH_2X = 16, 514 IB_WIDTH_4X = 2, 515 IB_WIDTH_8X = 4, 516 IB_WIDTH_12X = 8 517 }; 518 519 static inline int ib_width_enum_to_int(enum ib_port_width width) 520 { 521 switch (width) { 522 case IB_WIDTH_1X: return 1; 523 case IB_WIDTH_2X: return 2; 524 case IB_WIDTH_4X: return 4; 525 case IB_WIDTH_8X: return 8; 526 case IB_WIDTH_12X: return 12; 527 default: return -1; 528 } 529 } 530 531 enum ib_port_speed { 532 IB_SPEED_SDR = 1, 533 IB_SPEED_DDR = 2, 534 IB_SPEED_QDR = 4, 535 IB_SPEED_FDR10 = 8, 536 IB_SPEED_FDR = 16, 537 IB_SPEED_EDR = 32, 538 IB_SPEED_HDR = 64 539 }; 540 541 /** 542 * struct rdma_hw_stats 543 * @lock - Mutex to protect parallel write access to lifespan and values 544 * of counters, which are 64bits and not guaranteeed to be written 545 * atomicaly on 32bits systems. 546 * @timestamp - Used by the core code to track when the last update was 547 * @lifespan - Used by the core code to determine how old the counters 548 * should be before being updated again. Stored in jiffies, defaults 549 * to 10 milliseconds, drivers can override the default be specifying 550 * their own value during their allocation routine. 551 * @name - Array of pointers to static names used for the counters in 552 * directory. 553 * @num_counters - How many hardware counters there are. If name is 554 * shorter than this number, a kernel oops will result. Driver authors 555 * are encouraged to leave BUILD_BUG_ON(ARRAY_SIZE(@name) < num_counters) 556 * in their code to prevent this. 557 * @value - Array of u64 counters that are accessed by the sysfs code and 558 * filled in by the drivers get_stats routine 559 */ 560 struct rdma_hw_stats { 561 struct mutex lock; /* Protect lifespan and values[] */ 562 unsigned long timestamp; 563 unsigned long lifespan; 564 const char * const *names; 565 int num_counters; 566 u64 value[]; 567 }; 568 569 #define RDMA_HW_STATS_DEFAULT_LIFESPAN 10 570 /** 571 * rdma_alloc_hw_stats_struct - Helper function to allocate dynamic struct 572 * for drivers. 573 * @names - Array of static const char * 574 * @num_counters - How many elements in array 575 * @lifespan - How many milliseconds between updates 576 */ 577 static inline struct rdma_hw_stats *rdma_alloc_hw_stats_struct( 578 const char * const *names, int num_counters, 579 unsigned long lifespan) 580 { 581 struct rdma_hw_stats *stats; 582 583 stats = kzalloc(sizeof(*stats) + num_counters * sizeof(u64), 584 GFP_KERNEL); 585 if (!stats) 586 return NULL; 587 stats->names = names; 588 stats->num_counters = num_counters; 589 stats->lifespan = msecs_to_jiffies(lifespan); 590 591 return stats; 592 } 593 594 595 /* Define bits for the various functionality this port needs to be supported by 596 * the core. 597 */ 598 /* Management 0x00000FFF */ 599 #define RDMA_CORE_CAP_IB_MAD 0x00000001 600 #define RDMA_CORE_CAP_IB_SMI 0x00000002 601 #define RDMA_CORE_CAP_IB_CM 0x00000004 602 #define RDMA_CORE_CAP_IW_CM 0x00000008 603 #define RDMA_CORE_CAP_IB_SA 0x00000010 604 #define RDMA_CORE_CAP_OPA_MAD 0x00000020 605 606 /* Address format 0x000FF000 */ 607 #define RDMA_CORE_CAP_AF_IB 0x00001000 608 #define RDMA_CORE_CAP_ETH_AH 0x00002000 609 #define RDMA_CORE_CAP_OPA_AH 0x00004000 610 #define RDMA_CORE_CAP_IB_GRH_REQUIRED 0x00008000 611 612 /* Protocol 0xFFF00000 */ 613 #define RDMA_CORE_CAP_PROT_IB 0x00100000 614 #define RDMA_CORE_CAP_PROT_ROCE 0x00200000 615 #define RDMA_CORE_CAP_PROT_IWARP 0x00400000 616 #define RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP 0x00800000 617 #define RDMA_CORE_CAP_PROT_RAW_PACKET 0x01000000 618 #define RDMA_CORE_CAP_PROT_USNIC 0x02000000 619 620 #define RDMA_CORE_PORT_IB_GRH_REQUIRED (RDMA_CORE_CAP_IB_GRH_REQUIRED \ 621 | RDMA_CORE_CAP_PROT_ROCE \ 622 | RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP) 623 624 #define RDMA_CORE_PORT_IBA_IB (RDMA_CORE_CAP_PROT_IB \ 625 | RDMA_CORE_CAP_IB_MAD \ 626 | RDMA_CORE_CAP_IB_SMI \ 627 | RDMA_CORE_CAP_IB_CM \ 628 | RDMA_CORE_CAP_IB_SA \ 629 | RDMA_CORE_CAP_AF_IB) 630 #define RDMA_CORE_PORT_IBA_ROCE (RDMA_CORE_CAP_PROT_ROCE \ 631 | RDMA_CORE_CAP_IB_MAD \ 632 | RDMA_CORE_CAP_IB_CM \ 633 | RDMA_CORE_CAP_AF_IB \ 634 | RDMA_CORE_CAP_ETH_AH) 635 #define RDMA_CORE_PORT_IBA_ROCE_UDP_ENCAP \ 636 (RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP \ 637 | RDMA_CORE_CAP_IB_MAD \ 638 | RDMA_CORE_CAP_IB_CM \ 639 | RDMA_CORE_CAP_AF_IB \ 640 | RDMA_CORE_CAP_ETH_AH) 641 #define RDMA_CORE_PORT_IWARP (RDMA_CORE_CAP_PROT_IWARP \ 642 | RDMA_CORE_CAP_IW_CM) 643 #define RDMA_CORE_PORT_INTEL_OPA (RDMA_CORE_PORT_IBA_IB \ 644 | RDMA_CORE_CAP_OPA_MAD) 645 646 #define RDMA_CORE_PORT_RAW_PACKET (RDMA_CORE_CAP_PROT_RAW_PACKET) 647 648 #define RDMA_CORE_PORT_USNIC (RDMA_CORE_CAP_PROT_USNIC) 649 650 struct ib_port_attr { 651 u64 subnet_prefix; 652 enum ib_port_state state; 653 enum ib_mtu max_mtu; 654 enum ib_mtu active_mtu; 655 u32 phys_mtu; 656 int gid_tbl_len; 657 unsigned int ip_gids:1; 658 /* This is the value from PortInfo CapabilityMask, defined by IBA */ 659 u32 port_cap_flags; 660 u32 max_msg_sz; 661 u32 bad_pkey_cntr; 662 u32 qkey_viol_cntr; 663 u16 pkey_tbl_len; 664 u32 sm_lid; 665 u32 lid; 666 u8 lmc; 667 u8 max_vl_num; 668 u8 sm_sl; 669 u8 subnet_timeout; 670 u8 init_type_reply; 671 u8 active_width; 672 u8 active_speed; 673 u8 phys_state; 674 u16 port_cap_flags2; 675 }; 676 677 enum ib_device_modify_flags { 678 IB_DEVICE_MODIFY_SYS_IMAGE_GUID = 1 << 0, 679 IB_DEVICE_MODIFY_NODE_DESC = 1 << 1 680 }; 681 682 #define IB_DEVICE_NODE_DESC_MAX 64 683 684 struct ib_device_modify { 685 u64 sys_image_guid; 686 char node_desc[IB_DEVICE_NODE_DESC_MAX]; 687 }; 688 689 enum ib_port_modify_flags { 690 IB_PORT_SHUTDOWN = 1, 691 IB_PORT_INIT_TYPE = (1<<2), 692 IB_PORT_RESET_QKEY_CNTR = (1<<3), 693 IB_PORT_OPA_MASK_CHG = (1<<4) 694 }; 695 696 struct ib_port_modify { 697 u32 set_port_cap_mask; 698 u32 clr_port_cap_mask; 699 u8 init_type; 700 }; 701 702 enum ib_event_type { 703 IB_EVENT_CQ_ERR, 704 IB_EVENT_QP_FATAL, 705 IB_EVENT_QP_REQ_ERR, 706 IB_EVENT_QP_ACCESS_ERR, 707 IB_EVENT_COMM_EST, 708 IB_EVENT_SQ_DRAINED, 709 IB_EVENT_PATH_MIG, 710 IB_EVENT_PATH_MIG_ERR, 711 IB_EVENT_DEVICE_FATAL, 712 IB_EVENT_PORT_ACTIVE, 713 IB_EVENT_PORT_ERR, 714 IB_EVENT_LID_CHANGE, 715 IB_EVENT_PKEY_CHANGE, 716 IB_EVENT_SM_CHANGE, 717 IB_EVENT_SRQ_ERR, 718 IB_EVENT_SRQ_LIMIT_REACHED, 719 IB_EVENT_QP_LAST_WQE_REACHED, 720 IB_EVENT_CLIENT_REREGISTER, 721 IB_EVENT_GID_CHANGE, 722 IB_EVENT_WQ_FATAL, 723 }; 724 725 const char *__attribute_const__ ib_event_msg(enum ib_event_type event); 726 727 struct ib_event { 728 struct ib_device *device; 729 union { 730 struct ib_cq *cq; 731 struct ib_qp *qp; 732 struct ib_srq *srq; 733 struct ib_wq *wq; 734 u8 port_num; 735 } element; 736 enum ib_event_type event; 737 }; 738 739 struct ib_event_handler { 740 struct ib_device *device; 741 void (*handler)(struct ib_event_handler *, struct ib_event *); 742 struct list_head list; 743 }; 744 745 #define INIT_IB_EVENT_HANDLER(_ptr, _device, _handler) \ 746 do { \ 747 (_ptr)->device = _device; \ 748 (_ptr)->handler = _handler; \ 749 INIT_LIST_HEAD(&(_ptr)->list); \ 750 } while (0) 751 752 struct ib_global_route { 753 const struct ib_gid_attr *sgid_attr; 754 union ib_gid dgid; 755 u32 flow_label; 756 u8 sgid_index; 757 u8 hop_limit; 758 u8 traffic_class; 759 }; 760 761 struct ib_grh { 762 __be32 version_tclass_flow; 763 __be16 paylen; 764 u8 next_hdr; 765 u8 hop_limit; 766 union ib_gid sgid; 767 union ib_gid dgid; 768 }; 769 770 union rdma_network_hdr { 771 struct ib_grh ibgrh; 772 struct { 773 /* The IB spec states that if it's IPv4, the header 774 * is located in the last 20 bytes of the header. 775 */ 776 u8 reserved[20]; 777 struct iphdr roce4grh; 778 }; 779 }; 780 781 #define IB_QPN_MASK 0xFFFFFF 782 783 enum { 784 IB_MULTICAST_QPN = 0xffffff 785 }; 786 787 #define IB_LID_PERMISSIVE cpu_to_be16(0xFFFF) 788 #define IB_MULTICAST_LID_BASE cpu_to_be16(0xC000) 789 790 enum ib_ah_flags { 791 IB_AH_GRH = 1 792 }; 793 794 enum ib_rate { 795 IB_RATE_PORT_CURRENT = 0, 796 IB_RATE_2_5_GBPS = 2, 797 IB_RATE_5_GBPS = 5, 798 IB_RATE_10_GBPS = 3, 799 IB_RATE_20_GBPS = 6, 800 IB_RATE_30_GBPS = 4, 801 IB_RATE_40_GBPS = 7, 802 IB_RATE_60_GBPS = 8, 803 IB_RATE_80_GBPS = 9, 804 IB_RATE_120_GBPS = 10, 805 IB_RATE_14_GBPS = 11, 806 IB_RATE_56_GBPS = 12, 807 IB_RATE_112_GBPS = 13, 808 IB_RATE_168_GBPS = 14, 809 IB_RATE_25_GBPS = 15, 810 IB_RATE_100_GBPS = 16, 811 IB_RATE_200_GBPS = 17, 812 IB_RATE_300_GBPS = 18, 813 IB_RATE_28_GBPS = 19, 814 IB_RATE_50_GBPS = 20, 815 IB_RATE_400_GBPS = 21, 816 IB_RATE_600_GBPS = 22, 817 }; 818 819 /** 820 * ib_rate_to_mult - Convert the IB rate enum to a multiple of the 821 * base rate of 2.5 Gbit/sec. For example, IB_RATE_5_GBPS will be 822 * converted to 2, since 5 Gbit/sec is 2 * 2.5 Gbit/sec. 823 * @rate: rate to convert. 824 */ 825 __attribute_const__ int ib_rate_to_mult(enum ib_rate rate); 826 827 /** 828 * ib_rate_to_mbps - Convert the IB rate enum to Mbps. 829 * For example, IB_RATE_2_5_GBPS will be converted to 2500. 830 * @rate: rate to convert. 831 */ 832 __attribute_const__ int ib_rate_to_mbps(enum ib_rate rate); 833 834 835 /** 836 * enum ib_mr_type - memory region type 837 * @IB_MR_TYPE_MEM_REG: memory region that is used for 838 * normal registration 839 * @IB_MR_TYPE_SG_GAPS: memory region that is capable to 840 * register any arbitrary sg lists (without 841 * the normal mr constraints - see 842 * ib_map_mr_sg) 843 * @IB_MR_TYPE_DM: memory region that is used for device 844 * memory registration 845 * @IB_MR_TYPE_USER: memory region that is used for the user-space 846 * application 847 * @IB_MR_TYPE_DMA: memory region that is used for DMA operations 848 * without address translations (VA=PA) 849 * @IB_MR_TYPE_INTEGRITY: memory region that is used for 850 * data integrity operations 851 */ 852 enum ib_mr_type { 853 IB_MR_TYPE_MEM_REG, 854 IB_MR_TYPE_SG_GAPS, 855 IB_MR_TYPE_DM, 856 IB_MR_TYPE_USER, 857 IB_MR_TYPE_DMA, 858 IB_MR_TYPE_INTEGRITY, 859 }; 860 861 enum ib_mr_status_check { 862 IB_MR_CHECK_SIG_STATUS = 1, 863 }; 864 865 /** 866 * struct ib_mr_status - Memory region status container 867 * 868 * @fail_status: Bitmask of MR checks status. For each 869 * failed check a corresponding status bit is set. 870 * @sig_err: Additional info for IB_MR_CEHCK_SIG_STATUS 871 * failure. 872 */ 873 struct ib_mr_status { 874 u32 fail_status; 875 struct ib_sig_err sig_err; 876 }; 877 878 /** 879 * mult_to_ib_rate - Convert a multiple of 2.5 Gbit/sec to an IB rate 880 * enum. 881 * @mult: multiple to convert. 882 */ 883 __attribute_const__ enum ib_rate mult_to_ib_rate(int mult); 884 885 struct rdma_ah_init_attr { 886 struct rdma_ah_attr *ah_attr; 887 u32 flags; 888 struct net_device *xmit_slave; 889 }; 890 891 enum rdma_ah_attr_type { 892 RDMA_AH_ATTR_TYPE_UNDEFINED, 893 RDMA_AH_ATTR_TYPE_IB, 894 RDMA_AH_ATTR_TYPE_ROCE, 895 RDMA_AH_ATTR_TYPE_OPA, 896 }; 897 898 struct ib_ah_attr { 899 u16 dlid; 900 u8 src_path_bits; 901 }; 902 903 struct roce_ah_attr { 904 u8 dmac[ETH_ALEN]; 905 }; 906 907 struct opa_ah_attr { 908 u32 dlid; 909 u8 src_path_bits; 910 bool make_grd; 911 }; 912 913 struct rdma_ah_attr { 914 struct ib_global_route grh; 915 u8 sl; 916 u8 static_rate; 917 u8 port_num; 918 u8 ah_flags; 919 enum rdma_ah_attr_type type; 920 union { 921 struct ib_ah_attr ib; 922 struct roce_ah_attr roce; 923 struct opa_ah_attr opa; 924 }; 925 }; 926 927 enum ib_wc_status { 928 IB_WC_SUCCESS, 929 IB_WC_LOC_LEN_ERR, 930 IB_WC_LOC_QP_OP_ERR, 931 IB_WC_LOC_EEC_OP_ERR, 932 IB_WC_LOC_PROT_ERR, 933 IB_WC_WR_FLUSH_ERR, 934 IB_WC_MW_BIND_ERR, 935 IB_WC_BAD_RESP_ERR, 936 IB_WC_LOC_ACCESS_ERR, 937 IB_WC_REM_INV_REQ_ERR, 938 IB_WC_REM_ACCESS_ERR, 939 IB_WC_REM_OP_ERR, 940 IB_WC_RETRY_EXC_ERR, 941 IB_WC_RNR_RETRY_EXC_ERR, 942 IB_WC_LOC_RDD_VIOL_ERR, 943 IB_WC_REM_INV_RD_REQ_ERR, 944 IB_WC_REM_ABORT_ERR, 945 IB_WC_INV_EECN_ERR, 946 IB_WC_INV_EEC_STATE_ERR, 947 IB_WC_FATAL_ERR, 948 IB_WC_RESP_TIMEOUT_ERR, 949 IB_WC_GENERAL_ERR 950 }; 951 952 const char *__attribute_const__ ib_wc_status_msg(enum ib_wc_status status); 953 954 enum ib_wc_opcode { 955 IB_WC_SEND = IB_UVERBS_WC_SEND, 956 IB_WC_RDMA_WRITE = IB_UVERBS_WC_RDMA_WRITE, 957 IB_WC_RDMA_READ = IB_UVERBS_WC_RDMA_READ, 958 IB_WC_COMP_SWAP = IB_UVERBS_WC_COMP_SWAP, 959 IB_WC_FETCH_ADD = IB_UVERBS_WC_FETCH_ADD, 960 IB_WC_BIND_MW = IB_UVERBS_WC_BIND_MW, 961 IB_WC_LOCAL_INV = IB_UVERBS_WC_LOCAL_INV, 962 IB_WC_LSO = IB_UVERBS_WC_TSO, 963 IB_WC_REG_MR, 964 IB_WC_MASKED_COMP_SWAP, 965 IB_WC_MASKED_FETCH_ADD, 966 /* 967 * Set value of IB_WC_RECV so consumers can test if a completion is a 968 * receive by testing (opcode & IB_WC_RECV). 969 */ 970 IB_WC_RECV = 1 << 7, 971 IB_WC_RECV_RDMA_WITH_IMM 972 }; 973 974 enum ib_wc_flags { 975 IB_WC_GRH = 1, 976 IB_WC_WITH_IMM = (1<<1), 977 IB_WC_WITH_INVALIDATE = (1<<2), 978 IB_WC_IP_CSUM_OK = (1<<3), 979 IB_WC_WITH_SMAC = (1<<4), 980 IB_WC_WITH_VLAN = (1<<5), 981 IB_WC_WITH_NETWORK_HDR_TYPE = (1<<6), 982 }; 983 984 struct ib_wc { 985 union { 986 u64 wr_id; 987 struct ib_cqe *wr_cqe; 988 }; 989 enum ib_wc_status status; 990 enum ib_wc_opcode opcode; 991 u32 vendor_err; 992 u32 byte_len; 993 struct ib_qp *qp; 994 union { 995 __be32 imm_data; 996 u32 invalidate_rkey; 997 } ex; 998 u32 src_qp; 999 u32 slid; 1000 int wc_flags; 1001 u16 pkey_index; 1002 u8 sl; 1003 u8 dlid_path_bits; 1004 u8 port_num; /* valid only for DR SMPs on switches */ 1005 u8 smac[ETH_ALEN]; 1006 u16 vlan_id; 1007 u8 network_hdr_type; 1008 }; 1009 1010 enum ib_cq_notify_flags { 1011 IB_CQ_SOLICITED = 1 << 0, 1012 IB_CQ_NEXT_COMP = 1 << 1, 1013 IB_CQ_SOLICITED_MASK = IB_CQ_SOLICITED | IB_CQ_NEXT_COMP, 1014 IB_CQ_REPORT_MISSED_EVENTS = 1 << 2, 1015 }; 1016 1017 enum ib_srq_type { 1018 IB_SRQT_BASIC = IB_UVERBS_SRQT_BASIC, 1019 IB_SRQT_XRC = IB_UVERBS_SRQT_XRC, 1020 IB_SRQT_TM = IB_UVERBS_SRQT_TM, 1021 }; 1022 1023 static inline bool ib_srq_has_cq(enum ib_srq_type srq_type) 1024 { 1025 return srq_type == IB_SRQT_XRC || 1026 srq_type == IB_SRQT_TM; 1027 } 1028 1029 enum ib_srq_attr_mask { 1030 IB_SRQ_MAX_WR = 1 << 0, 1031 IB_SRQ_LIMIT = 1 << 1, 1032 }; 1033 1034 struct ib_srq_attr { 1035 u32 max_wr; 1036 u32 max_sge; 1037 u32 srq_limit; 1038 }; 1039 1040 struct ib_srq_init_attr { 1041 void (*event_handler)(struct ib_event *, void *); 1042 void *srq_context; 1043 struct ib_srq_attr attr; 1044 enum ib_srq_type srq_type; 1045 1046 struct { 1047 struct ib_cq *cq; 1048 union { 1049 struct { 1050 struct ib_xrcd *xrcd; 1051 } xrc; 1052 1053 struct { 1054 u32 max_num_tags; 1055 } tag_matching; 1056 }; 1057 } ext; 1058 }; 1059 1060 struct ib_qp_cap { 1061 u32 max_send_wr; 1062 u32 max_recv_wr; 1063 u32 max_send_sge; 1064 u32 max_recv_sge; 1065 u32 max_inline_data; 1066 1067 /* 1068 * Maximum number of rdma_rw_ctx structures in flight at a time. 1069 * ib_create_qp() will calculate the right amount of neededed WRs 1070 * and MRs based on this. 1071 */ 1072 u32 max_rdma_ctxs; 1073 }; 1074 1075 enum ib_sig_type { 1076 IB_SIGNAL_ALL_WR, 1077 IB_SIGNAL_REQ_WR 1078 }; 1079 1080 enum ib_qp_type { 1081 /* 1082 * IB_QPT_SMI and IB_QPT_GSI have to be the first two entries 1083 * here (and in that order) since the MAD layer uses them as 1084 * indices into a 2-entry table. 1085 */ 1086 IB_QPT_SMI, 1087 IB_QPT_GSI, 1088 1089 IB_QPT_RC = IB_UVERBS_QPT_RC, 1090 IB_QPT_UC = IB_UVERBS_QPT_UC, 1091 IB_QPT_UD = IB_UVERBS_QPT_UD, 1092 IB_QPT_RAW_IPV6, 1093 IB_QPT_RAW_ETHERTYPE, 1094 IB_QPT_RAW_PACKET = IB_UVERBS_QPT_RAW_PACKET, 1095 IB_QPT_XRC_INI = IB_UVERBS_QPT_XRC_INI, 1096 IB_QPT_XRC_TGT = IB_UVERBS_QPT_XRC_TGT, 1097 IB_QPT_MAX, 1098 IB_QPT_DRIVER = IB_UVERBS_QPT_DRIVER, 1099 /* Reserve a range for qp types internal to the low level driver. 1100 * These qp types will not be visible at the IB core layer, so the 1101 * IB_QPT_MAX usages should not be affected in the core layer 1102 */ 1103 IB_QPT_RESERVED1 = 0x1000, 1104 IB_QPT_RESERVED2, 1105 IB_QPT_RESERVED3, 1106 IB_QPT_RESERVED4, 1107 IB_QPT_RESERVED5, 1108 IB_QPT_RESERVED6, 1109 IB_QPT_RESERVED7, 1110 IB_QPT_RESERVED8, 1111 IB_QPT_RESERVED9, 1112 IB_QPT_RESERVED10, 1113 }; 1114 1115 enum ib_qp_create_flags { 1116 IB_QP_CREATE_IPOIB_UD_LSO = 1 << 0, 1117 IB_QP_CREATE_BLOCK_MULTICAST_LOOPBACK = 1118 IB_UVERBS_QP_CREATE_BLOCK_MULTICAST_LOOPBACK, 1119 IB_QP_CREATE_CROSS_CHANNEL = 1 << 2, 1120 IB_QP_CREATE_MANAGED_SEND = 1 << 3, 1121 IB_QP_CREATE_MANAGED_RECV = 1 << 4, 1122 IB_QP_CREATE_NETIF_QP = 1 << 5, 1123 IB_QP_CREATE_INTEGRITY_EN = 1 << 6, 1124 IB_QP_CREATE_NETDEV_USE = 1 << 7, 1125 IB_QP_CREATE_SCATTER_FCS = 1126 IB_UVERBS_QP_CREATE_SCATTER_FCS, 1127 IB_QP_CREATE_CVLAN_STRIPPING = 1128 IB_UVERBS_QP_CREATE_CVLAN_STRIPPING, 1129 IB_QP_CREATE_SOURCE_QPN = 1 << 10, 1130 IB_QP_CREATE_PCI_WRITE_END_PADDING = 1131 IB_UVERBS_QP_CREATE_PCI_WRITE_END_PADDING, 1132 /* reserve bits 26-31 for low level drivers' internal use */ 1133 IB_QP_CREATE_RESERVED_START = 1 << 26, 1134 IB_QP_CREATE_RESERVED_END = 1 << 31, 1135 }; 1136 1137 /* 1138 * Note: users may not call ib_close_qp or ib_destroy_qp from the event_handler 1139 * callback to destroy the passed in QP. 1140 */ 1141 1142 struct ib_qp_init_attr { 1143 /* Consumer's event_handler callback must not block */ 1144 void (*event_handler)(struct ib_event *, void *); 1145 1146 void *qp_context; 1147 struct ib_cq *send_cq; 1148 struct ib_cq *recv_cq; 1149 struct ib_srq *srq; 1150 struct ib_xrcd *xrcd; /* XRC TGT QPs only */ 1151 struct ib_qp_cap cap; 1152 enum ib_sig_type sq_sig_type; 1153 enum ib_qp_type qp_type; 1154 u32 create_flags; 1155 1156 /* 1157 * Only needed for special QP types, or when using the RW API. 1158 */ 1159 u8 port_num; 1160 struct ib_rwq_ind_table *rwq_ind_tbl; 1161 u32 source_qpn; 1162 }; 1163 1164 struct ib_qp_open_attr { 1165 void (*event_handler)(struct ib_event *, void *); 1166 void *qp_context; 1167 u32 qp_num; 1168 enum ib_qp_type qp_type; 1169 }; 1170 1171 enum ib_rnr_timeout { 1172 IB_RNR_TIMER_655_36 = 0, 1173 IB_RNR_TIMER_000_01 = 1, 1174 IB_RNR_TIMER_000_02 = 2, 1175 IB_RNR_TIMER_000_03 = 3, 1176 IB_RNR_TIMER_000_04 = 4, 1177 IB_RNR_TIMER_000_06 = 5, 1178 IB_RNR_TIMER_000_08 = 6, 1179 IB_RNR_TIMER_000_12 = 7, 1180 IB_RNR_TIMER_000_16 = 8, 1181 IB_RNR_TIMER_000_24 = 9, 1182 IB_RNR_TIMER_000_32 = 10, 1183 IB_RNR_TIMER_000_48 = 11, 1184 IB_RNR_TIMER_000_64 = 12, 1185 IB_RNR_TIMER_000_96 = 13, 1186 IB_RNR_TIMER_001_28 = 14, 1187 IB_RNR_TIMER_001_92 = 15, 1188 IB_RNR_TIMER_002_56 = 16, 1189 IB_RNR_TIMER_003_84 = 17, 1190 IB_RNR_TIMER_005_12 = 18, 1191 IB_RNR_TIMER_007_68 = 19, 1192 IB_RNR_TIMER_010_24 = 20, 1193 IB_RNR_TIMER_015_36 = 21, 1194 IB_RNR_TIMER_020_48 = 22, 1195 IB_RNR_TIMER_030_72 = 23, 1196 IB_RNR_TIMER_040_96 = 24, 1197 IB_RNR_TIMER_061_44 = 25, 1198 IB_RNR_TIMER_081_92 = 26, 1199 IB_RNR_TIMER_122_88 = 27, 1200 IB_RNR_TIMER_163_84 = 28, 1201 IB_RNR_TIMER_245_76 = 29, 1202 IB_RNR_TIMER_327_68 = 30, 1203 IB_RNR_TIMER_491_52 = 31 1204 }; 1205 1206 enum ib_qp_attr_mask { 1207 IB_QP_STATE = 1, 1208 IB_QP_CUR_STATE = (1<<1), 1209 IB_QP_EN_SQD_ASYNC_NOTIFY = (1<<2), 1210 IB_QP_ACCESS_FLAGS = (1<<3), 1211 IB_QP_PKEY_INDEX = (1<<4), 1212 IB_QP_PORT = (1<<5), 1213 IB_QP_QKEY = (1<<6), 1214 IB_QP_AV = (1<<7), 1215 IB_QP_PATH_MTU = (1<<8), 1216 IB_QP_TIMEOUT = (1<<9), 1217 IB_QP_RETRY_CNT = (1<<10), 1218 IB_QP_RNR_RETRY = (1<<11), 1219 IB_QP_RQ_PSN = (1<<12), 1220 IB_QP_MAX_QP_RD_ATOMIC = (1<<13), 1221 IB_QP_ALT_PATH = (1<<14), 1222 IB_QP_MIN_RNR_TIMER = (1<<15), 1223 IB_QP_SQ_PSN = (1<<16), 1224 IB_QP_MAX_DEST_RD_ATOMIC = (1<<17), 1225 IB_QP_PATH_MIG_STATE = (1<<18), 1226 IB_QP_CAP = (1<<19), 1227 IB_QP_DEST_QPN = (1<<20), 1228 IB_QP_RESERVED1 = (1<<21), 1229 IB_QP_RESERVED2 = (1<<22), 1230 IB_QP_RESERVED3 = (1<<23), 1231 IB_QP_RESERVED4 = (1<<24), 1232 IB_QP_RATE_LIMIT = (1<<25), 1233 }; 1234 1235 enum ib_qp_state { 1236 IB_QPS_RESET, 1237 IB_QPS_INIT, 1238 IB_QPS_RTR, 1239 IB_QPS_RTS, 1240 IB_QPS_SQD, 1241 IB_QPS_SQE, 1242 IB_QPS_ERR 1243 }; 1244 1245 enum ib_mig_state { 1246 IB_MIG_MIGRATED, 1247 IB_MIG_REARM, 1248 IB_MIG_ARMED 1249 }; 1250 1251 enum ib_mw_type { 1252 IB_MW_TYPE_1 = 1, 1253 IB_MW_TYPE_2 = 2 1254 }; 1255 1256 struct ib_qp_attr { 1257 enum ib_qp_state qp_state; 1258 enum ib_qp_state cur_qp_state; 1259 enum ib_mtu path_mtu; 1260 enum ib_mig_state path_mig_state; 1261 u32 qkey; 1262 u32 rq_psn; 1263 u32 sq_psn; 1264 u32 dest_qp_num; 1265 int qp_access_flags; 1266 struct ib_qp_cap cap; 1267 struct rdma_ah_attr ah_attr; 1268 struct rdma_ah_attr alt_ah_attr; 1269 u16 pkey_index; 1270 u16 alt_pkey_index; 1271 u8 en_sqd_async_notify; 1272 u8 sq_draining; 1273 u8 max_rd_atomic; 1274 u8 max_dest_rd_atomic; 1275 u8 min_rnr_timer; 1276 u8 port_num; 1277 u8 timeout; 1278 u8 retry_cnt; 1279 u8 rnr_retry; 1280 u8 alt_port_num; 1281 u8 alt_timeout; 1282 u32 rate_limit; 1283 struct net_device *xmit_slave; 1284 }; 1285 1286 enum ib_wr_opcode { 1287 /* These are shared with userspace */ 1288 IB_WR_RDMA_WRITE = IB_UVERBS_WR_RDMA_WRITE, 1289 IB_WR_RDMA_WRITE_WITH_IMM = IB_UVERBS_WR_RDMA_WRITE_WITH_IMM, 1290 IB_WR_SEND = IB_UVERBS_WR_SEND, 1291 IB_WR_SEND_WITH_IMM = IB_UVERBS_WR_SEND_WITH_IMM, 1292 IB_WR_RDMA_READ = IB_UVERBS_WR_RDMA_READ, 1293 IB_WR_ATOMIC_CMP_AND_SWP = IB_UVERBS_WR_ATOMIC_CMP_AND_SWP, 1294 IB_WR_ATOMIC_FETCH_AND_ADD = IB_UVERBS_WR_ATOMIC_FETCH_AND_ADD, 1295 IB_WR_BIND_MW = IB_UVERBS_WR_BIND_MW, 1296 IB_WR_LSO = IB_UVERBS_WR_TSO, 1297 IB_WR_SEND_WITH_INV = IB_UVERBS_WR_SEND_WITH_INV, 1298 IB_WR_RDMA_READ_WITH_INV = IB_UVERBS_WR_RDMA_READ_WITH_INV, 1299 IB_WR_LOCAL_INV = IB_UVERBS_WR_LOCAL_INV, 1300 IB_WR_MASKED_ATOMIC_CMP_AND_SWP = 1301 IB_UVERBS_WR_MASKED_ATOMIC_CMP_AND_SWP, 1302 IB_WR_MASKED_ATOMIC_FETCH_AND_ADD = 1303 IB_UVERBS_WR_MASKED_ATOMIC_FETCH_AND_ADD, 1304 1305 /* These are kernel only and can not be issued by userspace */ 1306 IB_WR_REG_MR = 0x20, 1307 IB_WR_REG_MR_INTEGRITY, 1308 1309 /* reserve values for low level drivers' internal use. 1310 * These values will not be used at all in the ib core layer. 1311 */ 1312 IB_WR_RESERVED1 = 0xf0, 1313 IB_WR_RESERVED2, 1314 IB_WR_RESERVED3, 1315 IB_WR_RESERVED4, 1316 IB_WR_RESERVED5, 1317 IB_WR_RESERVED6, 1318 IB_WR_RESERVED7, 1319 IB_WR_RESERVED8, 1320 IB_WR_RESERVED9, 1321 IB_WR_RESERVED10, 1322 }; 1323 1324 enum ib_send_flags { 1325 IB_SEND_FENCE = 1, 1326 IB_SEND_SIGNALED = (1<<1), 1327 IB_SEND_SOLICITED = (1<<2), 1328 IB_SEND_INLINE = (1<<3), 1329 IB_SEND_IP_CSUM = (1<<4), 1330 1331 /* reserve bits 26-31 for low level drivers' internal use */ 1332 IB_SEND_RESERVED_START = (1 << 26), 1333 IB_SEND_RESERVED_END = (1 << 31), 1334 }; 1335 1336 struct ib_sge { 1337 u64 addr; 1338 u32 length; 1339 u32 lkey; 1340 }; 1341 1342 struct ib_cqe { 1343 void (*done)(struct ib_cq *cq, struct ib_wc *wc); 1344 }; 1345 1346 struct ib_send_wr { 1347 struct ib_send_wr *next; 1348 union { 1349 u64 wr_id; 1350 struct ib_cqe *wr_cqe; 1351 }; 1352 struct ib_sge *sg_list; 1353 int num_sge; 1354 enum ib_wr_opcode opcode; 1355 int send_flags; 1356 union { 1357 __be32 imm_data; 1358 u32 invalidate_rkey; 1359 } ex; 1360 }; 1361 1362 struct ib_rdma_wr { 1363 struct ib_send_wr wr; 1364 u64 remote_addr; 1365 u32 rkey; 1366 }; 1367 1368 static inline const struct ib_rdma_wr *rdma_wr(const struct ib_send_wr *wr) 1369 { 1370 return container_of(wr, struct ib_rdma_wr, wr); 1371 } 1372 1373 struct ib_atomic_wr { 1374 struct ib_send_wr wr; 1375 u64 remote_addr; 1376 u64 compare_add; 1377 u64 swap; 1378 u64 compare_add_mask; 1379 u64 swap_mask; 1380 u32 rkey; 1381 }; 1382 1383 static inline const struct ib_atomic_wr *atomic_wr(const struct ib_send_wr *wr) 1384 { 1385 return container_of(wr, struct ib_atomic_wr, wr); 1386 } 1387 1388 struct ib_ud_wr { 1389 struct ib_send_wr wr; 1390 struct ib_ah *ah; 1391 void *header; 1392 int hlen; 1393 int mss; 1394 u32 remote_qpn; 1395 u32 remote_qkey; 1396 u16 pkey_index; /* valid for GSI only */ 1397 u8 port_num; /* valid for DR SMPs on switch only */ 1398 }; 1399 1400 static inline const struct ib_ud_wr *ud_wr(const struct ib_send_wr *wr) 1401 { 1402 return container_of(wr, struct ib_ud_wr, wr); 1403 } 1404 1405 struct ib_reg_wr { 1406 struct ib_send_wr wr; 1407 struct ib_mr *mr; 1408 u32 key; 1409 int access; 1410 }; 1411 1412 static inline const struct ib_reg_wr *reg_wr(const struct ib_send_wr *wr) 1413 { 1414 return container_of(wr, struct ib_reg_wr, wr); 1415 } 1416 1417 struct ib_recv_wr { 1418 struct ib_recv_wr *next; 1419 union { 1420 u64 wr_id; 1421 struct ib_cqe *wr_cqe; 1422 }; 1423 struct ib_sge *sg_list; 1424 int num_sge; 1425 }; 1426 1427 enum ib_access_flags { 1428 IB_ACCESS_LOCAL_WRITE = IB_UVERBS_ACCESS_LOCAL_WRITE, 1429 IB_ACCESS_REMOTE_WRITE = IB_UVERBS_ACCESS_REMOTE_WRITE, 1430 IB_ACCESS_REMOTE_READ = IB_UVERBS_ACCESS_REMOTE_READ, 1431 IB_ACCESS_REMOTE_ATOMIC = IB_UVERBS_ACCESS_REMOTE_ATOMIC, 1432 IB_ACCESS_MW_BIND = IB_UVERBS_ACCESS_MW_BIND, 1433 IB_ZERO_BASED = IB_UVERBS_ACCESS_ZERO_BASED, 1434 IB_ACCESS_ON_DEMAND = IB_UVERBS_ACCESS_ON_DEMAND, 1435 IB_ACCESS_HUGETLB = IB_UVERBS_ACCESS_HUGETLB, 1436 IB_ACCESS_RELAXED_ORDERING = IB_UVERBS_ACCESS_RELAXED_ORDERING, 1437 1438 IB_ACCESS_OPTIONAL = IB_UVERBS_ACCESS_OPTIONAL_RANGE, 1439 IB_ACCESS_SUPPORTED = 1440 ((IB_ACCESS_HUGETLB << 1) - 1) | IB_ACCESS_OPTIONAL, 1441 }; 1442 1443 /* 1444 * XXX: these are apparently used for ->rereg_user_mr, no idea why they 1445 * are hidden here instead of a uapi header! 1446 */ 1447 enum ib_mr_rereg_flags { 1448 IB_MR_REREG_TRANS = 1, 1449 IB_MR_REREG_PD = (1<<1), 1450 IB_MR_REREG_ACCESS = (1<<2), 1451 IB_MR_REREG_SUPPORTED = ((IB_MR_REREG_ACCESS << 1) - 1) 1452 }; 1453 1454 struct ib_umem; 1455 1456 enum rdma_remove_reason { 1457 /* 1458 * Userspace requested uobject deletion or initial try 1459 * to remove uobject via cleanup. Call could fail 1460 */ 1461 RDMA_REMOVE_DESTROY, 1462 /* Context deletion. This call should delete the actual object itself */ 1463 RDMA_REMOVE_CLOSE, 1464 /* Driver is being hot-unplugged. This call should delete the actual object itself */ 1465 RDMA_REMOVE_DRIVER_REMOVE, 1466 /* uobj is being cleaned-up before being committed */ 1467 RDMA_REMOVE_ABORT, 1468 }; 1469 1470 struct ib_rdmacg_object { 1471 #ifdef CONFIG_CGROUP_RDMA 1472 struct rdma_cgroup *cg; /* owner rdma cgroup */ 1473 #endif 1474 }; 1475 1476 struct ib_ucontext { 1477 struct ib_device *device; 1478 struct ib_uverbs_file *ufile; 1479 /* 1480 * 'closing' can be read by the driver only during a destroy callback, 1481 * it is set when we are closing the file descriptor and indicates 1482 * that mm_sem may be locked. 1483 */ 1484 bool closing; 1485 1486 bool cleanup_retryable; 1487 1488 struct ib_rdmacg_object cg_obj; 1489 /* 1490 * Implementation details of the RDMA core, don't use in drivers: 1491 */ 1492 struct rdma_restrack_entry res; 1493 struct xarray mmap_xa; 1494 }; 1495 1496 struct ib_uobject { 1497 u64 user_handle; /* handle given to us by userspace */ 1498 /* ufile & ucontext owning this object */ 1499 struct ib_uverbs_file *ufile; 1500 /* FIXME, save memory: ufile->context == context */ 1501 struct ib_ucontext *context; /* associated user context */ 1502 void *object; /* containing object */ 1503 struct list_head list; /* link to context's list */ 1504 struct ib_rdmacg_object cg_obj; /* rdmacg object */ 1505 int id; /* index into kernel idr */ 1506 struct kref ref; 1507 atomic_t usecnt; /* protects exclusive access */ 1508 struct rcu_head rcu; /* kfree_rcu() overhead */ 1509 1510 const struct uverbs_api_object *uapi_object; 1511 }; 1512 1513 struct ib_udata { 1514 const void __user *inbuf; 1515 void __user *outbuf; 1516 size_t inlen; 1517 size_t outlen; 1518 }; 1519 1520 struct ib_pd { 1521 u32 local_dma_lkey; 1522 u32 flags; 1523 struct ib_device *device; 1524 struct ib_uobject *uobject; 1525 atomic_t usecnt; /* count all resources */ 1526 1527 u32 unsafe_global_rkey; 1528 1529 /* 1530 * Implementation details of the RDMA core, don't use in drivers: 1531 */ 1532 struct ib_mr *__internal_mr; 1533 struct rdma_restrack_entry res; 1534 }; 1535 1536 struct ib_xrcd { 1537 struct ib_device *device; 1538 atomic_t usecnt; /* count all exposed resources */ 1539 struct inode *inode; 1540 struct rw_semaphore tgt_qps_rwsem; 1541 struct xarray tgt_qps; 1542 }; 1543 1544 struct ib_ah { 1545 struct ib_device *device; 1546 struct ib_pd *pd; 1547 struct ib_uobject *uobject; 1548 const struct ib_gid_attr *sgid_attr; 1549 enum rdma_ah_attr_type type; 1550 }; 1551 1552 typedef void (*ib_comp_handler)(struct ib_cq *cq, void *cq_context); 1553 1554 enum ib_poll_context { 1555 IB_POLL_SOFTIRQ, /* poll from softirq context */ 1556 IB_POLL_WORKQUEUE, /* poll from workqueue */ 1557 IB_POLL_UNBOUND_WORKQUEUE, /* poll from unbound workqueue */ 1558 IB_POLL_LAST_POOL_TYPE = IB_POLL_UNBOUND_WORKQUEUE, 1559 1560 IB_POLL_DIRECT, /* caller context, no hw completions */ 1561 }; 1562 1563 struct ib_cq { 1564 struct ib_device *device; 1565 struct ib_ucq_object *uobject; 1566 ib_comp_handler comp_handler; 1567 void (*event_handler)(struct ib_event *, void *); 1568 void *cq_context; 1569 int cqe; 1570 unsigned int cqe_used; 1571 atomic_t usecnt; /* count number of work queues */ 1572 enum ib_poll_context poll_ctx; 1573 struct ib_wc *wc; 1574 struct list_head pool_entry; 1575 union { 1576 struct irq_poll iop; 1577 struct work_struct work; 1578 }; 1579 struct workqueue_struct *comp_wq; 1580 struct dim *dim; 1581 1582 /* updated only by trace points */ 1583 ktime_t timestamp; 1584 u8 interrupt:1; 1585 u8 shared:1; 1586 unsigned int comp_vector; 1587 1588 /* 1589 * Implementation details of the RDMA core, don't use in drivers: 1590 */ 1591 struct rdma_restrack_entry res; 1592 }; 1593 1594 struct ib_srq { 1595 struct ib_device *device; 1596 struct ib_pd *pd; 1597 struct ib_usrq_object *uobject; 1598 void (*event_handler)(struct ib_event *, void *); 1599 void *srq_context; 1600 enum ib_srq_type srq_type; 1601 atomic_t usecnt; 1602 1603 struct { 1604 struct ib_cq *cq; 1605 union { 1606 struct { 1607 struct ib_xrcd *xrcd; 1608 u32 srq_num; 1609 } xrc; 1610 }; 1611 } ext; 1612 }; 1613 1614 enum ib_raw_packet_caps { 1615 /* Strip cvlan from incoming packet and report it in the matching work 1616 * completion is supported. 1617 */ 1618 IB_RAW_PACKET_CAP_CVLAN_STRIPPING = (1 << 0), 1619 /* Scatter FCS field of an incoming packet to host memory is supported. 1620 */ 1621 IB_RAW_PACKET_CAP_SCATTER_FCS = (1 << 1), 1622 /* Checksum offloads are supported (for both send and receive). */ 1623 IB_RAW_PACKET_CAP_IP_CSUM = (1 << 2), 1624 /* When a packet is received for an RQ with no receive WQEs, the 1625 * packet processing is delayed. 1626 */ 1627 IB_RAW_PACKET_CAP_DELAY_DROP = (1 << 3), 1628 }; 1629 1630 enum ib_wq_type { 1631 IB_WQT_RQ = IB_UVERBS_WQT_RQ, 1632 }; 1633 1634 enum ib_wq_state { 1635 IB_WQS_RESET, 1636 IB_WQS_RDY, 1637 IB_WQS_ERR 1638 }; 1639 1640 struct ib_wq { 1641 struct ib_device *device; 1642 struct ib_uwq_object *uobject; 1643 void *wq_context; 1644 void (*event_handler)(struct ib_event *, void *); 1645 struct ib_pd *pd; 1646 struct ib_cq *cq; 1647 u32 wq_num; 1648 enum ib_wq_state state; 1649 enum ib_wq_type wq_type; 1650 atomic_t usecnt; 1651 }; 1652 1653 enum ib_wq_flags { 1654 IB_WQ_FLAGS_CVLAN_STRIPPING = IB_UVERBS_WQ_FLAGS_CVLAN_STRIPPING, 1655 IB_WQ_FLAGS_SCATTER_FCS = IB_UVERBS_WQ_FLAGS_SCATTER_FCS, 1656 IB_WQ_FLAGS_DELAY_DROP = IB_UVERBS_WQ_FLAGS_DELAY_DROP, 1657 IB_WQ_FLAGS_PCI_WRITE_END_PADDING = 1658 IB_UVERBS_WQ_FLAGS_PCI_WRITE_END_PADDING, 1659 }; 1660 1661 struct ib_wq_init_attr { 1662 void *wq_context; 1663 enum ib_wq_type wq_type; 1664 u32 max_wr; 1665 u32 max_sge; 1666 struct ib_cq *cq; 1667 void (*event_handler)(struct ib_event *, void *); 1668 u32 create_flags; /* Use enum ib_wq_flags */ 1669 }; 1670 1671 enum ib_wq_attr_mask { 1672 IB_WQ_STATE = 1 << 0, 1673 IB_WQ_CUR_STATE = 1 << 1, 1674 IB_WQ_FLAGS = 1 << 2, 1675 }; 1676 1677 struct ib_wq_attr { 1678 enum ib_wq_state wq_state; 1679 enum ib_wq_state curr_wq_state; 1680 u32 flags; /* Use enum ib_wq_flags */ 1681 u32 flags_mask; /* Use enum ib_wq_flags */ 1682 }; 1683 1684 struct ib_rwq_ind_table { 1685 struct ib_device *device; 1686 struct ib_uobject *uobject; 1687 atomic_t usecnt; 1688 u32 ind_tbl_num; 1689 u32 log_ind_tbl_size; 1690 struct ib_wq **ind_tbl; 1691 }; 1692 1693 struct ib_rwq_ind_table_init_attr { 1694 u32 log_ind_tbl_size; 1695 /* Each entry is a pointer to Receive Work Queue */ 1696 struct ib_wq **ind_tbl; 1697 }; 1698 1699 enum port_pkey_state { 1700 IB_PORT_PKEY_NOT_VALID = 0, 1701 IB_PORT_PKEY_VALID = 1, 1702 IB_PORT_PKEY_LISTED = 2, 1703 }; 1704 1705 struct ib_qp_security; 1706 1707 struct ib_port_pkey { 1708 enum port_pkey_state state; 1709 u16 pkey_index; 1710 u8 port_num; 1711 struct list_head qp_list; 1712 struct list_head to_error_list; 1713 struct ib_qp_security *sec; 1714 }; 1715 1716 struct ib_ports_pkeys { 1717 struct ib_port_pkey main; 1718 struct ib_port_pkey alt; 1719 }; 1720 1721 struct ib_qp_security { 1722 struct ib_qp *qp; 1723 struct ib_device *dev; 1724 /* Hold this mutex when changing port and pkey settings. */ 1725 struct mutex mutex; 1726 struct ib_ports_pkeys *ports_pkeys; 1727 /* A list of all open shared QP handles. Required to enforce security 1728 * properly for all users of a shared QP. 1729 */ 1730 struct list_head shared_qp_list; 1731 void *security; 1732 bool destroying; 1733 atomic_t error_list_count; 1734 struct completion error_complete; 1735 int error_comps_pending; 1736 }; 1737 1738 /* 1739 * @max_write_sge: Maximum SGE elements per RDMA WRITE request. 1740 * @max_read_sge: Maximum SGE elements per RDMA READ request. 1741 */ 1742 struct ib_qp { 1743 struct ib_device *device; 1744 struct ib_pd *pd; 1745 struct ib_cq *send_cq; 1746 struct ib_cq *recv_cq; 1747 spinlock_t mr_lock; 1748 int mrs_used; 1749 struct list_head rdma_mrs; 1750 struct list_head sig_mrs; 1751 struct ib_srq *srq; 1752 struct ib_xrcd *xrcd; /* XRC TGT QPs only */ 1753 struct list_head xrcd_list; 1754 1755 /* count times opened, mcast attaches, flow attaches */ 1756 atomic_t usecnt; 1757 struct list_head open_list; 1758 struct ib_qp *real_qp; 1759 struct ib_uqp_object *uobject; 1760 void (*event_handler)(struct ib_event *, void *); 1761 void *qp_context; 1762 /* sgid_attrs associated with the AV's */ 1763 const struct ib_gid_attr *av_sgid_attr; 1764 const struct ib_gid_attr *alt_path_sgid_attr; 1765 u32 qp_num; 1766 u32 max_write_sge; 1767 u32 max_read_sge; 1768 enum ib_qp_type qp_type; 1769 struct ib_rwq_ind_table *rwq_ind_tbl; 1770 struct ib_qp_security *qp_sec; 1771 u8 port; 1772 1773 bool integrity_en; 1774 /* 1775 * Implementation details of the RDMA core, don't use in drivers: 1776 */ 1777 struct rdma_restrack_entry res; 1778 1779 /* The counter the qp is bind to */ 1780 struct rdma_counter *counter; 1781 }; 1782 1783 struct ib_dm { 1784 struct ib_device *device; 1785 u32 length; 1786 u32 flags; 1787 struct ib_uobject *uobject; 1788 atomic_t usecnt; 1789 }; 1790 1791 struct ib_mr { 1792 struct ib_device *device; 1793 struct ib_pd *pd; 1794 u32 lkey; 1795 u32 rkey; 1796 u64 iova; 1797 u64 length; 1798 unsigned int page_size; 1799 enum ib_mr_type type; 1800 bool need_inval; 1801 union { 1802 struct ib_uobject *uobject; /* user */ 1803 struct list_head qp_entry; /* FR */ 1804 }; 1805 1806 struct ib_dm *dm; 1807 struct ib_sig_attrs *sig_attrs; /* only for IB_MR_TYPE_INTEGRITY MRs */ 1808 /* 1809 * Implementation details of the RDMA core, don't use in drivers: 1810 */ 1811 struct rdma_restrack_entry res; 1812 }; 1813 1814 struct ib_mw { 1815 struct ib_device *device; 1816 struct ib_pd *pd; 1817 struct ib_uobject *uobject; 1818 u32 rkey; 1819 enum ib_mw_type type; 1820 }; 1821 1822 /* Supported steering options */ 1823 enum ib_flow_attr_type { 1824 /* steering according to rule specifications */ 1825 IB_FLOW_ATTR_NORMAL = 0x0, 1826 /* default unicast and multicast rule - 1827 * receive all Eth traffic which isn't steered to any QP 1828 */ 1829 IB_FLOW_ATTR_ALL_DEFAULT = 0x1, 1830 /* default multicast rule - 1831 * receive all Eth multicast traffic which isn't steered to any QP 1832 */ 1833 IB_FLOW_ATTR_MC_DEFAULT = 0x2, 1834 /* sniffer rule - receive all port traffic */ 1835 IB_FLOW_ATTR_SNIFFER = 0x3 1836 }; 1837 1838 /* Supported steering header types */ 1839 enum ib_flow_spec_type { 1840 /* L2 headers*/ 1841 IB_FLOW_SPEC_ETH = 0x20, 1842 IB_FLOW_SPEC_IB = 0x22, 1843 /* L3 header*/ 1844 IB_FLOW_SPEC_IPV4 = 0x30, 1845 IB_FLOW_SPEC_IPV6 = 0x31, 1846 IB_FLOW_SPEC_ESP = 0x34, 1847 /* L4 headers*/ 1848 IB_FLOW_SPEC_TCP = 0x40, 1849 IB_FLOW_SPEC_UDP = 0x41, 1850 IB_FLOW_SPEC_VXLAN_TUNNEL = 0x50, 1851 IB_FLOW_SPEC_GRE = 0x51, 1852 IB_FLOW_SPEC_MPLS = 0x60, 1853 IB_FLOW_SPEC_INNER = 0x100, 1854 /* Actions */ 1855 IB_FLOW_SPEC_ACTION_TAG = 0x1000, 1856 IB_FLOW_SPEC_ACTION_DROP = 0x1001, 1857 IB_FLOW_SPEC_ACTION_HANDLE = 0x1002, 1858 IB_FLOW_SPEC_ACTION_COUNT = 0x1003, 1859 }; 1860 #define IB_FLOW_SPEC_LAYER_MASK 0xF0 1861 #define IB_FLOW_SPEC_SUPPORT_LAYERS 10 1862 1863 enum ib_flow_flags { 1864 IB_FLOW_ATTR_FLAGS_DONT_TRAP = 1UL << 1, /* Continue match, no steal */ 1865 IB_FLOW_ATTR_FLAGS_EGRESS = 1UL << 2, /* Egress flow */ 1866 IB_FLOW_ATTR_FLAGS_RESERVED = 1UL << 3 /* Must be last */ 1867 }; 1868 1869 struct ib_flow_eth_filter { 1870 u8 dst_mac[6]; 1871 u8 src_mac[6]; 1872 __be16 ether_type; 1873 __be16 vlan_tag; 1874 /* Must be last */ 1875 u8 real_sz[]; 1876 }; 1877 1878 struct ib_flow_spec_eth { 1879 u32 type; 1880 u16 size; 1881 struct ib_flow_eth_filter val; 1882 struct ib_flow_eth_filter mask; 1883 }; 1884 1885 struct ib_flow_ib_filter { 1886 __be16 dlid; 1887 __u8 sl; 1888 /* Must be last */ 1889 u8 real_sz[]; 1890 }; 1891 1892 struct ib_flow_spec_ib { 1893 u32 type; 1894 u16 size; 1895 struct ib_flow_ib_filter val; 1896 struct ib_flow_ib_filter mask; 1897 }; 1898 1899 /* IPv4 header flags */ 1900 enum ib_ipv4_flags { 1901 IB_IPV4_DONT_FRAG = 0x2, /* Don't enable packet fragmentation */ 1902 IB_IPV4_MORE_FRAG = 0X4 /* For All fragmented packets except the 1903 last have this flag set */ 1904 }; 1905 1906 struct ib_flow_ipv4_filter { 1907 __be32 src_ip; 1908 __be32 dst_ip; 1909 u8 proto; 1910 u8 tos; 1911 u8 ttl; 1912 u8 flags; 1913 /* Must be last */ 1914 u8 real_sz[]; 1915 }; 1916 1917 struct ib_flow_spec_ipv4 { 1918 u32 type; 1919 u16 size; 1920 struct ib_flow_ipv4_filter val; 1921 struct ib_flow_ipv4_filter mask; 1922 }; 1923 1924 struct ib_flow_ipv6_filter { 1925 u8 src_ip[16]; 1926 u8 dst_ip[16]; 1927 __be32 flow_label; 1928 u8 next_hdr; 1929 u8 traffic_class; 1930 u8 hop_limit; 1931 /* Must be last */ 1932 u8 real_sz[]; 1933 }; 1934 1935 struct ib_flow_spec_ipv6 { 1936 u32 type; 1937 u16 size; 1938 struct ib_flow_ipv6_filter val; 1939 struct ib_flow_ipv6_filter mask; 1940 }; 1941 1942 struct ib_flow_tcp_udp_filter { 1943 __be16 dst_port; 1944 __be16 src_port; 1945 /* Must be last */ 1946 u8 real_sz[]; 1947 }; 1948 1949 struct ib_flow_spec_tcp_udp { 1950 u32 type; 1951 u16 size; 1952 struct ib_flow_tcp_udp_filter val; 1953 struct ib_flow_tcp_udp_filter mask; 1954 }; 1955 1956 struct ib_flow_tunnel_filter { 1957 __be32 tunnel_id; 1958 u8 real_sz[]; 1959 }; 1960 1961 /* ib_flow_spec_tunnel describes the Vxlan tunnel 1962 * the tunnel_id from val has the vni value 1963 */ 1964 struct ib_flow_spec_tunnel { 1965 u32 type; 1966 u16 size; 1967 struct ib_flow_tunnel_filter val; 1968 struct ib_flow_tunnel_filter mask; 1969 }; 1970 1971 struct ib_flow_esp_filter { 1972 __be32 spi; 1973 __be32 seq; 1974 /* Must be last */ 1975 u8 real_sz[]; 1976 }; 1977 1978 struct ib_flow_spec_esp { 1979 u32 type; 1980 u16 size; 1981 struct ib_flow_esp_filter val; 1982 struct ib_flow_esp_filter mask; 1983 }; 1984 1985 struct ib_flow_gre_filter { 1986 __be16 c_ks_res0_ver; 1987 __be16 protocol; 1988 __be32 key; 1989 /* Must be last */ 1990 u8 real_sz[]; 1991 }; 1992 1993 struct ib_flow_spec_gre { 1994 u32 type; 1995 u16 size; 1996 struct ib_flow_gre_filter val; 1997 struct ib_flow_gre_filter mask; 1998 }; 1999 2000 struct ib_flow_mpls_filter { 2001 __be32 tag; 2002 /* Must be last */ 2003 u8 real_sz[]; 2004 }; 2005 2006 struct ib_flow_spec_mpls { 2007 u32 type; 2008 u16 size; 2009 struct ib_flow_mpls_filter val; 2010 struct ib_flow_mpls_filter mask; 2011 }; 2012 2013 struct ib_flow_spec_action_tag { 2014 enum ib_flow_spec_type type; 2015 u16 size; 2016 u32 tag_id; 2017 }; 2018 2019 struct ib_flow_spec_action_drop { 2020 enum ib_flow_spec_type type; 2021 u16 size; 2022 }; 2023 2024 struct ib_flow_spec_action_handle { 2025 enum ib_flow_spec_type type; 2026 u16 size; 2027 struct ib_flow_action *act; 2028 }; 2029 2030 enum ib_counters_description { 2031 IB_COUNTER_PACKETS, 2032 IB_COUNTER_BYTES, 2033 }; 2034 2035 struct ib_flow_spec_action_count { 2036 enum ib_flow_spec_type type; 2037 u16 size; 2038 struct ib_counters *counters; 2039 }; 2040 2041 union ib_flow_spec { 2042 struct { 2043 u32 type; 2044 u16 size; 2045 }; 2046 struct ib_flow_spec_eth eth; 2047 struct ib_flow_spec_ib ib; 2048 struct ib_flow_spec_ipv4 ipv4; 2049 struct ib_flow_spec_tcp_udp tcp_udp; 2050 struct ib_flow_spec_ipv6 ipv6; 2051 struct ib_flow_spec_tunnel tunnel; 2052 struct ib_flow_spec_esp esp; 2053 struct ib_flow_spec_gre gre; 2054 struct ib_flow_spec_mpls mpls; 2055 struct ib_flow_spec_action_tag flow_tag; 2056 struct ib_flow_spec_action_drop drop; 2057 struct ib_flow_spec_action_handle action; 2058 struct ib_flow_spec_action_count flow_count; 2059 }; 2060 2061 struct ib_flow_attr { 2062 enum ib_flow_attr_type type; 2063 u16 size; 2064 u16 priority; 2065 u32 flags; 2066 u8 num_of_specs; 2067 u8 port; 2068 union ib_flow_spec flows[]; 2069 }; 2070 2071 struct ib_flow { 2072 struct ib_qp *qp; 2073 struct ib_device *device; 2074 struct ib_uobject *uobject; 2075 }; 2076 2077 enum ib_flow_action_type { 2078 IB_FLOW_ACTION_UNSPECIFIED, 2079 IB_FLOW_ACTION_ESP = 1, 2080 }; 2081 2082 struct ib_flow_action_attrs_esp_keymats { 2083 enum ib_uverbs_flow_action_esp_keymat protocol; 2084 union { 2085 struct ib_uverbs_flow_action_esp_keymat_aes_gcm aes_gcm; 2086 } keymat; 2087 }; 2088 2089 struct ib_flow_action_attrs_esp_replays { 2090 enum ib_uverbs_flow_action_esp_replay protocol; 2091 union { 2092 struct ib_uverbs_flow_action_esp_replay_bmp bmp; 2093 } replay; 2094 }; 2095 2096 enum ib_flow_action_attrs_esp_flags { 2097 /* All user-space flags at the top: Use enum ib_uverbs_flow_action_esp_flags 2098 * This is done in order to share the same flags between user-space and 2099 * kernel and spare an unnecessary translation. 2100 */ 2101 2102 /* Kernel flags */ 2103 IB_FLOW_ACTION_ESP_FLAGS_ESN_TRIGGERED = 1ULL << 32, 2104 IB_FLOW_ACTION_ESP_FLAGS_MOD_ESP_ATTRS = 1ULL << 33, 2105 }; 2106 2107 struct ib_flow_spec_list { 2108 struct ib_flow_spec_list *next; 2109 union ib_flow_spec spec; 2110 }; 2111 2112 struct ib_flow_action_attrs_esp { 2113 struct ib_flow_action_attrs_esp_keymats *keymat; 2114 struct ib_flow_action_attrs_esp_replays *replay; 2115 struct ib_flow_spec_list *encap; 2116 /* Used only if IB_FLOW_ACTION_ESP_FLAGS_ESN_TRIGGERED is enabled. 2117 * Value of 0 is a valid value. 2118 */ 2119 u32 esn; 2120 u32 spi; 2121 u32 seq; 2122 u32 tfc_pad; 2123 /* Use enum ib_flow_action_attrs_esp_flags */ 2124 u64 flags; 2125 u64 hard_limit_pkts; 2126 }; 2127 2128 struct ib_flow_action { 2129 struct ib_device *device; 2130 struct ib_uobject *uobject; 2131 enum ib_flow_action_type type; 2132 atomic_t usecnt; 2133 }; 2134 2135 struct ib_mad; 2136 struct ib_grh; 2137 2138 enum ib_process_mad_flags { 2139 IB_MAD_IGNORE_MKEY = 1, 2140 IB_MAD_IGNORE_BKEY = 2, 2141 IB_MAD_IGNORE_ALL = IB_MAD_IGNORE_MKEY | IB_MAD_IGNORE_BKEY 2142 }; 2143 2144 enum ib_mad_result { 2145 IB_MAD_RESULT_FAILURE = 0, /* (!SUCCESS is the important flag) */ 2146 IB_MAD_RESULT_SUCCESS = 1 << 0, /* MAD was successfully processed */ 2147 IB_MAD_RESULT_REPLY = 1 << 1, /* Reply packet needs to be sent */ 2148 IB_MAD_RESULT_CONSUMED = 1 << 2 /* Packet consumed: stop processing */ 2149 }; 2150 2151 struct ib_port_cache { 2152 u64 subnet_prefix; 2153 struct ib_pkey_cache *pkey; 2154 struct ib_gid_table *gid; 2155 u8 lmc; 2156 enum ib_port_state port_state; 2157 }; 2158 2159 struct ib_port_immutable { 2160 int pkey_tbl_len; 2161 int gid_tbl_len; 2162 u32 core_cap_flags; 2163 u32 max_mad_size; 2164 }; 2165 2166 struct ib_port_data { 2167 struct ib_device *ib_dev; 2168 2169 struct ib_port_immutable immutable; 2170 2171 spinlock_t pkey_list_lock; 2172 struct list_head pkey_list; 2173 2174 struct ib_port_cache cache; 2175 2176 spinlock_t netdev_lock; 2177 struct net_device __rcu *netdev; 2178 struct hlist_node ndev_hash_link; 2179 struct rdma_port_counter port_counter; 2180 struct rdma_hw_stats *hw_stats; 2181 }; 2182 2183 /* rdma netdev type - specifies protocol type */ 2184 enum rdma_netdev_t { 2185 RDMA_NETDEV_OPA_VNIC, 2186 RDMA_NETDEV_IPOIB, 2187 }; 2188 2189 /** 2190 * struct rdma_netdev - rdma netdev 2191 * For cases where netstack interfacing is required. 2192 */ 2193 struct rdma_netdev { 2194 void *clnt_priv; 2195 struct ib_device *hca; 2196 u8 port_num; 2197 int mtu; 2198 2199 /* 2200 * cleanup function must be specified. 2201 * FIXME: This is only used for OPA_VNIC and that usage should be 2202 * removed too. 2203 */ 2204 void (*free_rdma_netdev)(struct net_device *netdev); 2205 2206 /* control functions */ 2207 void (*set_id)(struct net_device *netdev, int id); 2208 /* send packet */ 2209 int (*send)(struct net_device *dev, struct sk_buff *skb, 2210 struct ib_ah *address, u32 dqpn); 2211 /* multicast */ 2212 int (*attach_mcast)(struct net_device *dev, struct ib_device *hca, 2213 union ib_gid *gid, u16 mlid, 2214 int set_qkey, u32 qkey); 2215 int (*detach_mcast)(struct net_device *dev, struct ib_device *hca, 2216 union ib_gid *gid, u16 mlid); 2217 }; 2218 2219 struct rdma_netdev_alloc_params { 2220 size_t sizeof_priv; 2221 unsigned int txqs; 2222 unsigned int rxqs; 2223 void *param; 2224 2225 int (*initialize_rdma_netdev)(struct ib_device *device, u8 port_num, 2226 struct net_device *netdev, void *param); 2227 }; 2228 2229 struct ib_odp_counters { 2230 atomic64_t faults; 2231 atomic64_t invalidations; 2232 atomic64_t prefetch; 2233 }; 2234 2235 struct ib_counters { 2236 struct ib_device *device; 2237 struct ib_uobject *uobject; 2238 /* num of objects attached */ 2239 atomic_t usecnt; 2240 }; 2241 2242 struct ib_counters_read_attr { 2243 u64 *counters_buff; 2244 u32 ncounters; 2245 u32 flags; /* use enum ib_read_counters_flags */ 2246 }; 2247 2248 struct uverbs_attr_bundle; 2249 struct iw_cm_id; 2250 struct iw_cm_conn_param; 2251 2252 #define INIT_RDMA_OBJ_SIZE(ib_struct, drv_struct, member) \ 2253 .size_##ib_struct = \ 2254 (sizeof(struct drv_struct) + \ 2255 BUILD_BUG_ON_ZERO(offsetof(struct drv_struct, member)) + \ 2256 BUILD_BUG_ON_ZERO( \ 2257 !__same_type(((struct drv_struct *)NULL)->member, \ 2258 struct ib_struct))) 2259 2260 #define rdma_zalloc_drv_obj_gfp(ib_dev, ib_type, gfp) \ 2261 ((struct ib_type *)kzalloc(ib_dev->ops.size_##ib_type, gfp)) 2262 2263 #define rdma_zalloc_drv_obj(ib_dev, ib_type) \ 2264 rdma_zalloc_drv_obj_gfp(ib_dev, ib_type, GFP_KERNEL) 2265 2266 #define DECLARE_RDMA_OBJ_SIZE(ib_struct) size_t size_##ib_struct 2267 2268 struct rdma_user_mmap_entry { 2269 struct kref ref; 2270 struct ib_ucontext *ucontext; 2271 unsigned long start_pgoff; 2272 size_t npages; 2273 bool driver_removed; 2274 }; 2275 2276 /* Return the offset (in bytes) the user should pass to libc's mmap() */ 2277 static inline u64 2278 rdma_user_mmap_get_offset(const struct rdma_user_mmap_entry *entry) 2279 { 2280 return (u64)entry->start_pgoff << PAGE_SHIFT; 2281 } 2282 2283 /** 2284 * struct ib_device_ops - InfiniBand device operations 2285 * This structure defines all the InfiniBand device operations, providers will 2286 * need to define the supported operations, otherwise they will be set to null. 2287 */ 2288 struct ib_device_ops { 2289 struct module *owner; 2290 enum rdma_driver_id driver_id; 2291 u32 uverbs_abi_ver; 2292 unsigned int uverbs_no_driver_id_binding:1; 2293 2294 int (*post_send)(struct ib_qp *qp, const struct ib_send_wr *send_wr, 2295 const struct ib_send_wr **bad_send_wr); 2296 int (*post_recv)(struct ib_qp *qp, const struct ib_recv_wr *recv_wr, 2297 const struct ib_recv_wr **bad_recv_wr); 2298 void (*drain_rq)(struct ib_qp *qp); 2299 void (*drain_sq)(struct ib_qp *qp); 2300 int (*poll_cq)(struct ib_cq *cq, int num_entries, struct ib_wc *wc); 2301 int (*peek_cq)(struct ib_cq *cq, int wc_cnt); 2302 int (*req_notify_cq)(struct ib_cq *cq, enum ib_cq_notify_flags flags); 2303 int (*req_ncomp_notif)(struct ib_cq *cq, int wc_cnt); 2304 int (*post_srq_recv)(struct ib_srq *srq, 2305 const struct ib_recv_wr *recv_wr, 2306 const struct ib_recv_wr **bad_recv_wr); 2307 int (*process_mad)(struct ib_device *device, int process_mad_flags, 2308 u8 port_num, const struct ib_wc *in_wc, 2309 const struct ib_grh *in_grh, 2310 const struct ib_mad *in_mad, struct ib_mad *out_mad, 2311 size_t *out_mad_size, u16 *out_mad_pkey_index); 2312 int (*query_device)(struct ib_device *device, 2313 struct ib_device_attr *device_attr, 2314 struct ib_udata *udata); 2315 int (*modify_device)(struct ib_device *device, int device_modify_mask, 2316 struct ib_device_modify *device_modify); 2317 void (*get_dev_fw_str)(struct ib_device *device, char *str); 2318 const struct cpumask *(*get_vector_affinity)(struct ib_device *ibdev, 2319 int comp_vector); 2320 int (*query_port)(struct ib_device *device, u8 port_num, 2321 struct ib_port_attr *port_attr); 2322 int (*modify_port)(struct ib_device *device, u8 port_num, 2323 int port_modify_mask, 2324 struct ib_port_modify *port_modify); 2325 /** 2326 * The following mandatory functions are used only at device 2327 * registration. Keep functions such as these at the end of this 2328 * structure to avoid cache line misses when accessing struct ib_device 2329 * in fast paths. 2330 */ 2331 int (*get_port_immutable)(struct ib_device *device, u8 port_num, 2332 struct ib_port_immutable *immutable); 2333 enum rdma_link_layer (*get_link_layer)(struct ib_device *device, 2334 u8 port_num); 2335 /** 2336 * When calling get_netdev, the HW vendor's driver should return the 2337 * net device of device @device at port @port_num or NULL if such 2338 * a net device doesn't exist. The vendor driver should call dev_hold 2339 * on this net device. The HW vendor's device driver must guarantee 2340 * that this function returns NULL before the net device has finished 2341 * NETDEV_UNREGISTER state. 2342 */ 2343 struct net_device *(*get_netdev)(struct ib_device *device, u8 port_num); 2344 /** 2345 * rdma netdev operation 2346 * 2347 * Driver implementing alloc_rdma_netdev or rdma_netdev_get_params 2348 * must return -EOPNOTSUPP if it doesn't support the specified type. 2349 */ 2350 struct net_device *(*alloc_rdma_netdev)( 2351 struct ib_device *device, u8 port_num, enum rdma_netdev_t type, 2352 const char *name, unsigned char name_assign_type, 2353 void (*setup)(struct net_device *)); 2354 2355 int (*rdma_netdev_get_params)(struct ib_device *device, u8 port_num, 2356 enum rdma_netdev_t type, 2357 struct rdma_netdev_alloc_params *params); 2358 /** 2359 * query_gid should be return GID value for @device, when @port_num 2360 * link layer is either IB or iWarp. It is no-op if @port_num port 2361 * is RoCE link layer. 2362 */ 2363 int (*query_gid)(struct ib_device *device, u8 port_num, int index, 2364 union ib_gid *gid); 2365 /** 2366 * When calling add_gid, the HW vendor's driver should add the gid 2367 * of device of port at gid index available at @attr. Meta-info of 2368 * that gid (for example, the network device related to this gid) is 2369 * available at @attr. @context allows the HW vendor driver to store 2370 * extra information together with a GID entry. The HW vendor driver may 2371 * allocate memory to contain this information and store it in @context 2372 * when a new GID entry is written to. Params are consistent until the 2373 * next call of add_gid or delete_gid. The function should return 0 on 2374 * success or error otherwise. The function could be called 2375 * concurrently for different ports. This function is only called when 2376 * roce_gid_table is used. 2377 */ 2378 int (*add_gid)(const struct ib_gid_attr *attr, void **context); 2379 /** 2380 * When calling del_gid, the HW vendor's driver should delete the 2381 * gid of device @device at gid index gid_index of port port_num 2382 * available in @attr. 2383 * Upon the deletion of a GID entry, the HW vendor must free any 2384 * allocated memory. The caller will clear @context afterwards. 2385 * This function is only called when roce_gid_table is used. 2386 */ 2387 int (*del_gid)(const struct ib_gid_attr *attr, void **context); 2388 int (*query_pkey)(struct ib_device *device, u8 port_num, u16 index, 2389 u16 *pkey); 2390 int (*alloc_ucontext)(struct ib_ucontext *context, 2391 struct ib_udata *udata); 2392 void (*dealloc_ucontext)(struct ib_ucontext *context); 2393 int (*mmap)(struct ib_ucontext *context, struct vm_area_struct *vma); 2394 /** 2395 * This will be called once refcount of an entry in mmap_xa reaches 2396 * zero. The type of the memory that was mapped may differ between 2397 * entries and is opaque to the rdma_user_mmap interface. 2398 * Therefore needs to be implemented by the driver in mmap_free. 2399 */ 2400 void (*mmap_free)(struct rdma_user_mmap_entry *entry); 2401 void (*disassociate_ucontext)(struct ib_ucontext *ibcontext); 2402 int (*alloc_pd)(struct ib_pd *pd, struct ib_udata *udata); 2403 int (*dealloc_pd)(struct ib_pd *pd, struct ib_udata *udata); 2404 int (*create_ah)(struct ib_ah *ah, struct rdma_ah_init_attr *attr, 2405 struct ib_udata *udata); 2406 int (*modify_ah)(struct ib_ah *ah, struct rdma_ah_attr *ah_attr); 2407 int (*query_ah)(struct ib_ah *ah, struct rdma_ah_attr *ah_attr); 2408 int (*destroy_ah)(struct ib_ah *ah, u32 flags); 2409 int (*create_srq)(struct ib_srq *srq, 2410 struct ib_srq_init_attr *srq_init_attr, 2411 struct ib_udata *udata); 2412 int (*modify_srq)(struct ib_srq *srq, struct ib_srq_attr *srq_attr, 2413 enum ib_srq_attr_mask srq_attr_mask, 2414 struct ib_udata *udata); 2415 int (*query_srq)(struct ib_srq *srq, struct ib_srq_attr *srq_attr); 2416 int (*destroy_srq)(struct ib_srq *srq, struct ib_udata *udata); 2417 struct ib_qp *(*create_qp)(struct ib_pd *pd, 2418 struct ib_qp_init_attr *qp_init_attr, 2419 struct ib_udata *udata); 2420 int (*modify_qp)(struct ib_qp *qp, struct ib_qp_attr *qp_attr, 2421 int qp_attr_mask, struct ib_udata *udata); 2422 int (*query_qp)(struct ib_qp *qp, struct ib_qp_attr *qp_attr, 2423 int qp_attr_mask, struct ib_qp_init_attr *qp_init_attr); 2424 int (*destroy_qp)(struct ib_qp *qp, struct ib_udata *udata); 2425 int (*create_cq)(struct ib_cq *cq, const struct ib_cq_init_attr *attr, 2426 struct ib_udata *udata); 2427 int (*modify_cq)(struct ib_cq *cq, u16 cq_count, u16 cq_period); 2428 int (*destroy_cq)(struct ib_cq *cq, struct ib_udata *udata); 2429 int (*resize_cq)(struct ib_cq *cq, int cqe, struct ib_udata *udata); 2430 struct ib_mr *(*get_dma_mr)(struct ib_pd *pd, int mr_access_flags); 2431 struct ib_mr *(*reg_user_mr)(struct ib_pd *pd, u64 start, u64 length, 2432 u64 virt_addr, int mr_access_flags, 2433 struct ib_udata *udata); 2434 int (*rereg_user_mr)(struct ib_mr *mr, int flags, u64 start, u64 length, 2435 u64 virt_addr, int mr_access_flags, 2436 struct ib_pd *pd, struct ib_udata *udata); 2437 int (*dereg_mr)(struct ib_mr *mr, struct ib_udata *udata); 2438 struct ib_mr *(*alloc_mr)(struct ib_pd *pd, enum ib_mr_type mr_type, 2439 u32 max_num_sg); 2440 struct ib_mr *(*alloc_mr_integrity)(struct ib_pd *pd, 2441 u32 max_num_data_sg, 2442 u32 max_num_meta_sg); 2443 int (*advise_mr)(struct ib_pd *pd, 2444 enum ib_uverbs_advise_mr_advice advice, u32 flags, 2445 struct ib_sge *sg_list, u32 num_sge, 2446 struct uverbs_attr_bundle *attrs); 2447 int (*map_mr_sg)(struct ib_mr *mr, struct scatterlist *sg, int sg_nents, 2448 unsigned int *sg_offset); 2449 int (*check_mr_status)(struct ib_mr *mr, u32 check_mask, 2450 struct ib_mr_status *mr_status); 2451 int (*alloc_mw)(struct ib_mw *mw, struct ib_udata *udata); 2452 int (*dealloc_mw)(struct ib_mw *mw); 2453 int (*attach_mcast)(struct ib_qp *qp, union ib_gid *gid, u16 lid); 2454 int (*detach_mcast)(struct ib_qp *qp, union ib_gid *gid, u16 lid); 2455 int (*alloc_xrcd)(struct ib_xrcd *xrcd, struct ib_udata *udata); 2456 int (*dealloc_xrcd)(struct ib_xrcd *xrcd, struct ib_udata *udata); 2457 struct ib_flow *(*create_flow)(struct ib_qp *qp, 2458 struct ib_flow_attr *flow_attr, 2459 struct ib_udata *udata); 2460 int (*destroy_flow)(struct ib_flow *flow_id); 2461 struct ib_flow_action *(*create_flow_action_esp)( 2462 struct ib_device *device, 2463 const struct ib_flow_action_attrs_esp *attr, 2464 struct uverbs_attr_bundle *attrs); 2465 int (*destroy_flow_action)(struct ib_flow_action *action); 2466 int (*modify_flow_action_esp)( 2467 struct ib_flow_action *action, 2468 const struct ib_flow_action_attrs_esp *attr, 2469 struct uverbs_attr_bundle *attrs); 2470 int (*set_vf_link_state)(struct ib_device *device, int vf, u8 port, 2471 int state); 2472 int (*get_vf_config)(struct ib_device *device, int vf, u8 port, 2473 struct ifla_vf_info *ivf); 2474 int (*get_vf_stats)(struct ib_device *device, int vf, u8 port, 2475 struct ifla_vf_stats *stats); 2476 int (*get_vf_guid)(struct ib_device *device, int vf, u8 port, 2477 struct ifla_vf_guid *node_guid, 2478 struct ifla_vf_guid *port_guid); 2479 int (*set_vf_guid)(struct ib_device *device, int vf, u8 port, u64 guid, 2480 int type); 2481 struct ib_wq *(*create_wq)(struct ib_pd *pd, 2482 struct ib_wq_init_attr *init_attr, 2483 struct ib_udata *udata); 2484 int (*destroy_wq)(struct ib_wq *wq, struct ib_udata *udata); 2485 int (*modify_wq)(struct ib_wq *wq, struct ib_wq_attr *attr, 2486 u32 wq_attr_mask, struct ib_udata *udata); 2487 int (*create_rwq_ind_table)(struct ib_rwq_ind_table *ib_rwq_ind_table, 2488 struct ib_rwq_ind_table_init_attr *init_attr, 2489 struct ib_udata *udata); 2490 int (*destroy_rwq_ind_table)(struct ib_rwq_ind_table *wq_ind_table); 2491 struct ib_dm *(*alloc_dm)(struct ib_device *device, 2492 struct ib_ucontext *context, 2493 struct ib_dm_alloc_attr *attr, 2494 struct uverbs_attr_bundle *attrs); 2495 int (*dealloc_dm)(struct ib_dm *dm, struct uverbs_attr_bundle *attrs); 2496 struct ib_mr *(*reg_dm_mr)(struct ib_pd *pd, struct ib_dm *dm, 2497 struct ib_dm_mr_attr *attr, 2498 struct uverbs_attr_bundle *attrs); 2499 int (*create_counters)(struct ib_counters *counters, 2500 struct uverbs_attr_bundle *attrs); 2501 int (*destroy_counters)(struct ib_counters *counters); 2502 int (*read_counters)(struct ib_counters *counters, 2503 struct ib_counters_read_attr *counters_read_attr, 2504 struct uverbs_attr_bundle *attrs); 2505 int (*map_mr_sg_pi)(struct ib_mr *mr, struct scatterlist *data_sg, 2506 int data_sg_nents, unsigned int *data_sg_offset, 2507 struct scatterlist *meta_sg, int meta_sg_nents, 2508 unsigned int *meta_sg_offset); 2509 2510 /** 2511 * alloc_hw_stats - Allocate a struct rdma_hw_stats and fill in the 2512 * driver initialized data. The struct is kfree()'ed by the sysfs 2513 * core when the device is removed. A lifespan of -1 in the return 2514 * struct tells the core to set a default lifespan. 2515 */ 2516 struct rdma_hw_stats *(*alloc_hw_stats)(struct ib_device *device, 2517 u8 port_num); 2518 /** 2519 * get_hw_stats - Fill in the counter value(s) in the stats struct. 2520 * @index - The index in the value array we wish to have updated, or 2521 * num_counters if we want all stats updated 2522 * Return codes - 2523 * < 0 - Error, no counters updated 2524 * index - Updated the single counter pointed to by index 2525 * num_counters - Updated all counters (will reset the timestamp 2526 * and prevent further calls for lifespan milliseconds) 2527 * Drivers are allowed to update all counters in leiu of just the 2528 * one given in index at their option 2529 */ 2530 int (*get_hw_stats)(struct ib_device *device, 2531 struct rdma_hw_stats *stats, u8 port, int index); 2532 /* 2533 * This function is called once for each port when a ib device is 2534 * registered. 2535 */ 2536 int (*init_port)(struct ib_device *device, u8 port_num, 2537 struct kobject *port_sysfs); 2538 /** 2539 * Allows rdma drivers to add their own restrack attributes. 2540 */ 2541 int (*fill_res_mr_entry)(struct sk_buff *msg, struct ib_mr *ibmr); 2542 int (*fill_res_mr_entry_raw)(struct sk_buff *msg, struct ib_mr *ibmr); 2543 int (*fill_res_cq_entry)(struct sk_buff *msg, struct ib_cq *ibcq); 2544 int (*fill_res_cq_entry_raw)(struct sk_buff *msg, struct ib_cq *ibcq); 2545 int (*fill_res_qp_entry)(struct sk_buff *msg, struct ib_qp *ibqp); 2546 int (*fill_res_qp_entry_raw)(struct sk_buff *msg, struct ib_qp *ibqp); 2547 int (*fill_res_cm_id_entry)(struct sk_buff *msg, struct rdma_cm_id *id); 2548 2549 /* Device lifecycle callbacks */ 2550 /* 2551 * Called after the device becomes registered, before clients are 2552 * attached 2553 */ 2554 int (*enable_driver)(struct ib_device *dev); 2555 /* 2556 * This is called as part of ib_dealloc_device(). 2557 */ 2558 void (*dealloc_driver)(struct ib_device *dev); 2559 2560 /* iWarp CM callbacks */ 2561 void (*iw_add_ref)(struct ib_qp *qp); 2562 void (*iw_rem_ref)(struct ib_qp *qp); 2563 struct ib_qp *(*iw_get_qp)(struct ib_device *device, int qpn); 2564 int (*iw_connect)(struct iw_cm_id *cm_id, 2565 struct iw_cm_conn_param *conn_param); 2566 int (*iw_accept)(struct iw_cm_id *cm_id, 2567 struct iw_cm_conn_param *conn_param); 2568 int (*iw_reject)(struct iw_cm_id *cm_id, const void *pdata, 2569 u8 pdata_len); 2570 int (*iw_create_listen)(struct iw_cm_id *cm_id, int backlog); 2571 int (*iw_destroy_listen)(struct iw_cm_id *cm_id); 2572 /** 2573 * counter_bind_qp - Bind a QP to a counter. 2574 * @counter - The counter to be bound. If counter->id is zero then 2575 * the driver needs to allocate a new counter and set counter->id 2576 */ 2577 int (*counter_bind_qp)(struct rdma_counter *counter, struct ib_qp *qp); 2578 /** 2579 * counter_unbind_qp - Unbind the qp from the dynamically-allocated 2580 * counter and bind it onto the default one 2581 */ 2582 int (*counter_unbind_qp)(struct ib_qp *qp); 2583 /** 2584 * counter_dealloc -De-allocate the hw counter 2585 */ 2586 int (*counter_dealloc)(struct rdma_counter *counter); 2587 /** 2588 * counter_alloc_stats - Allocate a struct rdma_hw_stats and fill in 2589 * the driver initialized data. 2590 */ 2591 struct rdma_hw_stats *(*counter_alloc_stats)( 2592 struct rdma_counter *counter); 2593 /** 2594 * counter_update_stats - Query the stats value of this counter 2595 */ 2596 int (*counter_update_stats)(struct rdma_counter *counter); 2597 2598 /** 2599 * Allows rdma drivers to add their own restrack attributes 2600 * dumped via 'rdma stat' iproute2 command. 2601 */ 2602 int (*fill_stat_mr_entry)(struct sk_buff *msg, struct ib_mr *ibmr); 2603 2604 /* query driver for its ucontext properties */ 2605 int (*query_ucontext)(struct ib_ucontext *context, 2606 struct uverbs_attr_bundle *attrs); 2607 2608 DECLARE_RDMA_OBJ_SIZE(ib_ah); 2609 DECLARE_RDMA_OBJ_SIZE(ib_counters); 2610 DECLARE_RDMA_OBJ_SIZE(ib_cq); 2611 DECLARE_RDMA_OBJ_SIZE(ib_mw); 2612 DECLARE_RDMA_OBJ_SIZE(ib_pd); 2613 DECLARE_RDMA_OBJ_SIZE(ib_rwq_ind_table); 2614 DECLARE_RDMA_OBJ_SIZE(ib_srq); 2615 DECLARE_RDMA_OBJ_SIZE(ib_ucontext); 2616 DECLARE_RDMA_OBJ_SIZE(ib_xrcd); 2617 }; 2618 2619 struct ib_core_device { 2620 /* device must be the first element in structure until, 2621 * union of ib_core_device and device exists in ib_device. 2622 */ 2623 struct device dev; 2624 possible_net_t rdma_net; 2625 struct kobject *ports_kobj; 2626 struct list_head port_list; 2627 struct ib_device *owner; /* reach back to owner ib_device */ 2628 }; 2629 2630 struct rdma_restrack_root; 2631 struct ib_device { 2632 /* Do not access @dma_device directly from ULP nor from HW drivers. */ 2633 struct device *dma_device; 2634 struct ib_device_ops ops; 2635 char name[IB_DEVICE_NAME_MAX]; 2636 struct rcu_head rcu_head; 2637 2638 struct list_head event_handler_list; 2639 /* Protects event_handler_list */ 2640 struct rw_semaphore event_handler_rwsem; 2641 2642 /* Protects QP's event_handler calls and open_qp list */ 2643 spinlock_t qp_open_list_lock; 2644 2645 struct rw_semaphore client_data_rwsem; 2646 struct xarray client_data; 2647 struct mutex unregistration_lock; 2648 2649 /* Synchronize GID, Pkey cache entries, subnet prefix, LMC */ 2650 rwlock_t cache_lock; 2651 /** 2652 * port_data is indexed by port number 2653 */ 2654 struct ib_port_data *port_data; 2655 2656 int num_comp_vectors; 2657 2658 union { 2659 struct device dev; 2660 struct ib_core_device coredev; 2661 }; 2662 2663 /* First group for device attributes, 2664 * Second group for driver provided attributes (optional). 2665 * It is NULL terminated array. 2666 */ 2667 const struct attribute_group *groups[3]; 2668 2669 u64 uverbs_cmd_mask; 2670 u64 uverbs_ex_cmd_mask; 2671 2672 char node_desc[IB_DEVICE_NODE_DESC_MAX]; 2673 __be64 node_guid; 2674 u32 local_dma_lkey; 2675 u16 is_switch:1; 2676 /* Indicates kernel verbs support, should not be used in drivers */ 2677 u16 kverbs_provider:1; 2678 /* CQ adaptive moderation (RDMA DIM) */ 2679 u16 use_cq_dim:1; 2680 u8 node_type; 2681 u8 phys_port_cnt; 2682 struct ib_device_attr attrs; 2683 struct attribute_group *hw_stats_ag; 2684 struct rdma_hw_stats *hw_stats; 2685 2686 #ifdef CONFIG_CGROUP_RDMA 2687 struct rdmacg_device cg_device; 2688 #endif 2689 2690 u32 index; 2691 2692 spinlock_t cq_pools_lock; 2693 struct list_head cq_pools[IB_POLL_LAST_POOL_TYPE + 1]; 2694 2695 struct rdma_restrack_root *res; 2696 2697 const struct uapi_definition *driver_def; 2698 2699 /* 2700 * Positive refcount indicates that the device is currently 2701 * registered and cannot be unregistered. 2702 */ 2703 refcount_t refcount; 2704 struct completion unreg_completion; 2705 struct work_struct unregistration_work; 2706 2707 const struct rdma_link_ops *link_ops; 2708 2709 /* Protects compat_devs xarray modifications */ 2710 struct mutex compat_devs_mutex; 2711 /* Maintains compat devices for each net namespace */ 2712 struct xarray compat_devs; 2713 2714 /* Used by iWarp CM */ 2715 char iw_ifname[IFNAMSIZ]; 2716 u32 iw_driver_flags; 2717 u32 lag_flags; 2718 }; 2719 2720 struct ib_client_nl_info; 2721 struct ib_client { 2722 const char *name; 2723 int (*add)(struct ib_device *ibdev); 2724 void (*remove)(struct ib_device *, void *client_data); 2725 void (*rename)(struct ib_device *dev, void *client_data); 2726 int (*get_nl_info)(struct ib_device *ibdev, void *client_data, 2727 struct ib_client_nl_info *res); 2728 int (*get_global_nl_info)(struct ib_client_nl_info *res); 2729 2730 /* Returns the net_dev belonging to this ib_client and matching the 2731 * given parameters. 2732 * @dev: An RDMA device that the net_dev use for communication. 2733 * @port: A physical port number on the RDMA device. 2734 * @pkey: P_Key that the net_dev uses if applicable. 2735 * @gid: A GID that the net_dev uses to communicate. 2736 * @addr: An IP address the net_dev is configured with. 2737 * @client_data: The device's client data set by ib_set_client_data(). 2738 * 2739 * An ib_client that implements a net_dev on top of RDMA devices 2740 * (such as IP over IB) should implement this callback, allowing the 2741 * rdma_cm module to find the right net_dev for a given request. 2742 * 2743 * The caller is responsible for calling dev_put on the returned 2744 * netdev. */ 2745 struct net_device *(*get_net_dev_by_params)( 2746 struct ib_device *dev, 2747 u8 port, 2748 u16 pkey, 2749 const union ib_gid *gid, 2750 const struct sockaddr *addr, 2751 void *client_data); 2752 2753 refcount_t uses; 2754 struct completion uses_zero; 2755 u32 client_id; 2756 2757 /* kverbs are not required by the client */ 2758 u8 no_kverbs_req:1; 2759 }; 2760 2761 /* 2762 * IB block DMA iterator 2763 * 2764 * Iterates the DMA-mapped SGL in contiguous memory blocks aligned 2765 * to a HW supported page size. 2766 */ 2767 struct ib_block_iter { 2768 /* internal states */ 2769 struct scatterlist *__sg; /* sg holding the current aligned block */ 2770 dma_addr_t __dma_addr; /* unaligned DMA address of this block */ 2771 unsigned int __sg_nents; /* number of SG entries */ 2772 unsigned int __sg_advance; /* number of bytes to advance in sg in next step */ 2773 unsigned int __pg_bit; /* alignment of current block */ 2774 }; 2775 2776 struct ib_device *_ib_alloc_device(size_t size); 2777 #define ib_alloc_device(drv_struct, member) \ 2778 container_of(_ib_alloc_device(sizeof(struct drv_struct) + \ 2779 BUILD_BUG_ON_ZERO(offsetof( \ 2780 struct drv_struct, member))), \ 2781 struct drv_struct, member) 2782 2783 void ib_dealloc_device(struct ib_device *device); 2784 2785 void ib_get_device_fw_str(struct ib_device *device, char *str); 2786 2787 int ib_register_device(struct ib_device *device, const char *name); 2788 void ib_unregister_device(struct ib_device *device); 2789 void ib_unregister_driver(enum rdma_driver_id driver_id); 2790 void ib_unregister_device_and_put(struct ib_device *device); 2791 void ib_unregister_device_queued(struct ib_device *ib_dev); 2792 2793 int ib_register_client (struct ib_client *client); 2794 void ib_unregister_client(struct ib_client *client); 2795 2796 void __rdma_block_iter_start(struct ib_block_iter *biter, 2797 struct scatterlist *sglist, 2798 unsigned int nents, 2799 unsigned long pgsz); 2800 bool __rdma_block_iter_next(struct ib_block_iter *biter); 2801 2802 /** 2803 * rdma_block_iter_dma_address - get the aligned dma address of the current 2804 * block held by the block iterator. 2805 * @biter: block iterator holding the memory block 2806 */ 2807 static inline dma_addr_t 2808 rdma_block_iter_dma_address(struct ib_block_iter *biter) 2809 { 2810 return biter->__dma_addr & ~(BIT_ULL(biter->__pg_bit) - 1); 2811 } 2812 2813 /** 2814 * rdma_for_each_block - iterate over contiguous memory blocks of the sg list 2815 * @sglist: sglist to iterate over 2816 * @biter: block iterator holding the memory block 2817 * @nents: maximum number of sg entries to iterate over 2818 * @pgsz: best HW supported page size to use 2819 * 2820 * Callers may use rdma_block_iter_dma_address() to get each 2821 * blocks aligned DMA address. 2822 */ 2823 #define rdma_for_each_block(sglist, biter, nents, pgsz) \ 2824 for (__rdma_block_iter_start(biter, sglist, nents, \ 2825 pgsz); \ 2826 __rdma_block_iter_next(biter);) 2827 2828 /** 2829 * ib_get_client_data - Get IB client context 2830 * @device:Device to get context for 2831 * @client:Client to get context for 2832 * 2833 * ib_get_client_data() returns the client context data set with 2834 * ib_set_client_data(). This can only be called while the client is 2835 * registered to the device, once the ib_client remove() callback returns this 2836 * cannot be called. 2837 */ 2838 static inline void *ib_get_client_data(struct ib_device *device, 2839 struct ib_client *client) 2840 { 2841 return xa_load(&device->client_data, client->client_id); 2842 } 2843 void ib_set_client_data(struct ib_device *device, struct ib_client *client, 2844 void *data); 2845 void ib_set_device_ops(struct ib_device *device, 2846 const struct ib_device_ops *ops); 2847 2848 int rdma_user_mmap_io(struct ib_ucontext *ucontext, struct vm_area_struct *vma, 2849 unsigned long pfn, unsigned long size, pgprot_t prot, 2850 struct rdma_user_mmap_entry *entry); 2851 int rdma_user_mmap_entry_insert(struct ib_ucontext *ucontext, 2852 struct rdma_user_mmap_entry *entry, 2853 size_t length); 2854 int rdma_user_mmap_entry_insert_range(struct ib_ucontext *ucontext, 2855 struct rdma_user_mmap_entry *entry, 2856 size_t length, u32 min_pgoff, 2857 u32 max_pgoff); 2858 2859 struct rdma_user_mmap_entry * 2860 rdma_user_mmap_entry_get_pgoff(struct ib_ucontext *ucontext, 2861 unsigned long pgoff); 2862 struct rdma_user_mmap_entry * 2863 rdma_user_mmap_entry_get(struct ib_ucontext *ucontext, 2864 struct vm_area_struct *vma); 2865 void rdma_user_mmap_entry_put(struct rdma_user_mmap_entry *entry); 2866 2867 void rdma_user_mmap_entry_remove(struct rdma_user_mmap_entry *entry); 2868 2869 static inline int ib_copy_from_udata(void *dest, struct ib_udata *udata, size_t len) 2870 { 2871 return copy_from_user(dest, udata->inbuf, len) ? -EFAULT : 0; 2872 } 2873 2874 static inline int ib_copy_to_udata(struct ib_udata *udata, void *src, size_t len) 2875 { 2876 return copy_to_user(udata->outbuf, src, len) ? -EFAULT : 0; 2877 } 2878 2879 static inline bool ib_is_buffer_cleared(const void __user *p, 2880 size_t len) 2881 { 2882 bool ret; 2883 u8 *buf; 2884 2885 if (len > USHRT_MAX) 2886 return false; 2887 2888 buf = memdup_user(p, len); 2889 if (IS_ERR(buf)) 2890 return false; 2891 2892 ret = !memchr_inv(buf, 0, len); 2893 kfree(buf); 2894 return ret; 2895 } 2896 2897 static inline bool ib_is_udata_cleared(struct ib_udata *udata, 2898 size_t offset, 2899 size_t len) 2900 { 2901 return ib_is_buffer_cleared(udata->inbuf + offset, len); 2902 } 2903 2904 /** 2905 * ib_is_destroy_retryable - Check whether the uobject destruction 2906 * is retryable. 2907 * @ret: The initial destruction return code 2908 * @why: remove reason 2909 * @uobj: The uobject that is destroyed 2910 * 2911 * This function is a helper function that IB layer and low-level drivers 2912 * can use to consider whether the destruction of the given uobject is 2913 * retry-able. 2914 * It checks the original return code, if it wasn't success the destruction 2915 * is retryable according to the ucontext state (i.e. cleanup_retryable) and 2916 * the remove reason. (i.e. why). 2917 * Must be called with the object locked for destroy. 2918 */ 2919 static inline bool ib_is_destroy_retryable(int ret, enum rdma_remove_reason why, 2920 struct ib_uobject *uobj) 2921 { 2922 return ret && (why == RDMA_REMOVE_DESTROY || 2923 uobj->context->cleanup_retryable); 2924 } 2925 2926 /** 2927 * ib_destroy_usecnt - Called during destruction to check the usecnt 2928 * @usecnt: The usecnt atomic 2929 * @why: remove reason 2930 * @uobj: The uobject that is destroyed 2931 * 2932 * Non-zero usecnts will block destruction unless destruction was triggered by 2933 * a ucontext cleanup. 2934 */ 2935 static inline int ib_destroy_usecnt(atomic_t *usecnt, 2936 enum rdma_remove_reason why, 2937 struct ib_uobject *uobj) 2938 { 2939 if (atomic_read(usecnt) && ib_is_destroy_retryable(-EBUSY, why, uobj)) 2940 return -EBUSY; 2941 return 0; 2942 } 2943 2944 /** 2945 * ib_modify_qp_is_ok - Check that the supplied attribute mask 2946 * contains all required attributes and no attributes not allowed for 2947 * the given QP state transition. 2948 * @cur_state: Current QP state 2949 * @next_state: Next QP state 2950 * @type: QP type 2951 * @mask: Mask of supplied QP attributes 2952 * 2953 * This function is a helper function that a low-level driver's 2954 * modify_qp method can use to validate the consumer's input. It 2955 * checks that cur_state and next_state are valid QP states, that a 2956 * transition from cur_state to next_state is allowed by the IB spec, 2957 * and that the attribute mask supplied is allowed for the transition. 2958 */ 2959 bool ib_modify_qp_is_ok(enum ib_qp_state cur_state, enum ib_qp_state next_state, 2960 enum ib_qp_type type, enum ib_qp_attr_mask mask); 2961 2962 void ib_register_event_handler(struct ib_event_handler *event_handler); 2963 void ib_unregister_event_handler(struct ib_event_handler *event_handler); 2964 void ib_dispatch_event(const struct ib_event *event); 2965 2966 int ib_query_port(struct ib_device *device, 2967 u8 port_num, struct ib_port_attr *port_attr); 2968 2969 enum rdma_link_layer rdma_port_get_link_layer(struct ib_device *device, 2970 u8 port_num); 2971 2972 /** 2973 * rdma_cap_ib_switch - Check if the device is IB switch 2974 * @device: Device to check 2975 * 2976 * Device driver is responsible for setting is_switch bit on 2977 * in ib_device structure at init time. 2978 * 2979 * Return: true if the device is IB switch. 2980 */ 2981 static inline bool rdma_cap_ib_switch(const struct ib_device *device) 2982 { 2983 return device->is_switch; 2984 } 2985 2986 /** 2987 * rdma_start_port - Return the first valid port number for the device 2988 * specified 2989 * 2990 * @device: Device to be checked 2991 * 2992 * Return start port number 2993 */ 2994 static inline u8 rdma_start_port(const struct ib_device *device) 2995 { 2996 return rdma_cap_ib_switch(device) ? 0 : 1; 2997 } 2998 2999 /** 3000 * rdma_for_each_port - Iterate over all valid port numbers of the IB device 3001 * @device - The struct ib_device * to iterate over 3002 * @iter - The unsigned int to store the port number 3003 */ 3004 #define rdma_for_each_port(device, iter) \ 3005 for (iter = rdma_start_port(device + BUILD_BUG_ON_ZERO(!__same_type( \ 3006 unsigned int, iter))); \ 3007 iter <= rdma_end_port(device); (iter)++) 3008 3009 /** 3010 * rdma_end_port - Return the last valid port number for the device 3011 * specified 3012 * 3013 * @device: Device to be checked 3014 * 3015 * Return last port number 3016 */ 3017 static inline u8 rdma_end_port(const struct ib_device *device) 3018 { 3019 return rdma_cap_ib_switch(device) ? 0 : device->phys_port_cnt; 3020 } 3021 3022 static inline int rdma_is_port_valid(const struct ib_device *device, 3023 unsigned int port) 3024 { 3025 return (port >= rdma_start_port(device) && 3026 port <= rdma_end_port(device)); 3027 } 3028 3029 static inline bool rdma_is_grh_required(const struct ib_device *device, 3030 u8 port_num) 3031 { 3032 return device->port_data[port_num].immutable.core_cap_flags & 3033 RDMA_CORE_PORT_IB_GRH_REQUIRED; 3034 } 3035 3036 static inline bool rdma_protocol_ib(const struct ib_device *device, u8 port_num) 3037 { 3038 return device->port_data[port_num].immutable.core_cap_flags & 3039 RDMA_CORE_CAP_PROT_IB; 3040 } 3041 3042 static inline bool rdma_protocol_roce(const struct ib_device *device, u8 port_num) 3043 { 3044 return device->port_data[port_num].immutable.core_cap_flags & 3045 (RDMA_CORE_CAP_PROT_ROCE | RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP); 3046 } 3047 3048 static inline bool rdma_protocol_roce_udp_encap(const struct ib_device *device, u8 port_num) 3049 { 3050 return device->port_data[port_num].immutable.core_cap_flags & 3051 RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP; 3052 } 3053 3054 static inline bool rdma_protocol_roce_eth_encap(const struct ib_device *device, u8 port_num) 3055 { 3056 return device->port_data[port_num].immutable.core_cap_flags & 3057 RDMA_CORE_CAP_PROT_ROCE; 3058 } 3059 3060 static inline bool rdma_protocol_iwarp(const struct ib_device *device, u8 port_num) 3061 { 3062 return device->port_data[port_num].immutable.core_cap_flags & 3063 RDMA_CORE_CAP_PROT_IWARP; 3064 } 3065 3066 static inline bool rdma_ib_or_roce(const struct ib_device *device, u8 port_num) 3067 { 3068 return rdma_protocol_ib(device, port_num) || 3069 rdma_protocol_roce(device, port_num); 3070 } 3071 3072 static inline bool rdma_protocol_raw_packet(const struct ib_device *device, u8 port_num) 3073 { 3074 return device->port_data[port_num].immutable.core_cap_flags & 3075 RDMA_CORE_CAP_PROT_RAW_PACKET; 3076 } 3077 3078 static inline bool rdma_protocol_usnic(const struct ib_device *device, u8 port_num) 3079 { 3080 return device->port_data[port_num].immutable.core_cap_flags & 3081 RDMA_CORE_CAP_PROT_USNIC; 3082 } 3083 3084 /** 3085 * rdma_cap_ib_mad - Check if the port of a device supports Infiniband 3086 * Management Datagrams. 3087 * @device: Device to check 3088 * @port_num: Port number to check 3089 * 3090 * Management Datagrams (MAD) are a required part of the InfiniBand 3091 * specification and are supported on all InfiniBand devices. A slightly 3092 * extended version are also supported on OPA interfaces. 3093 * 3094 * Return: true if the port supports sending/receiving of MAD packets. 3095 */ 3096 static inline bool rdma_cap_ib_mad(const struct ib_device *device, u8 port_num) 3097 { 3098 return device->port_data[port_num].immutable.core_cap_flags & 3099 RDMA_CORE_CAP_IB_MAD; 3100 } 3101 3102 /** 3103 * rdma_cap_opa_mad - Check if the port of device provides support for OPA 3104 * Management Datagrams. 3105 * @device: Device to check 3106 * @port_num: Port number to check 3107 * 3108 * Intel OmniPath devices extend and/or replace the InfiniBand Management 3109 * datagrams with their own versions. These OPA MADs share many but not all of 3110 * the characteristics of InfiniBand MADs. 3111 * 3112 * OPA MADs differ in the following ways: 3113 * 3114 * 1) MADs are variable size up to 2K 3115 * IBTA defined MADs remain fixed at 256 bytes 3116 * 2) OPA SMPs must carry valid PKeys 3117 * 3) OPA SMP packets are a different format 3118 * 3119 * Return: true if the port supports OPA MAD packet formats. 3120 */ 3121 static inline bool rdma_cap_opa_mad(struct ib_device *device, u8 port_num) 3122 { 3123 return device->port_data[port_num].immutable.core_cap_flags & 3124 RDMA_CORE_CAP_OPA_MAD; 3125 } 3126 3127 /** 3128 * rdma_cap_ib_smi - Check if the port of a device provides an Infiniband 3129 * Subnet Management Agent (SMA) on the Subnet Management Interface (SMI). 3130 * @device: Device to check 3131 * @port_num: Port number to check 3132 * 3133 * Each InfiniBand node is required to provide a Subnet Management Agent 3134 * that the subnet manager can access. Prior to the fabric being fully 3135 * configured by the subnet manager, the SMA is accessed via a well known 3136 * interface called the Subnet Management Interface (SMI). This interface 3137 * uses directed route packets to communicate with the SM to get around the 3138 * chicken and egg problem of the SM needing to know what's on the fabric 3139 * in order to configure the fabric, and needing to configure the fabric in 3140 * order to send packets to the devices on the fabric. These directed 3141 * route packets do not need the fabric fully configured in order to reach 3142 * their destination. The SMI is the only method allowed to send 3143 * directed route packets on an InfiniBand fabric. 3144 * 3145 * Return: true if the port provides an SMI. 3146 */ 3147 static inline bool rdma_cap_ib_smi(const struct ib_device *device, u8 port_num) 3148 { 3149 return device->port_data[port_num].immutable.core_cap_flags & 3150 RDMA_CORE_CAP_IB_SMI; 3151 } 3152 3153 /** 3154 * rdma_cap_ib_cm - Check if the port of device has the capability Infiniband 3155 * Communication Manager. 3156 * @device: Device to check 3157 * @port_num: Port number to check 3158 * 3159 * The InfiniBand Communication Manager is one of many pre-defined General 3160 * Service Agents (GSA) that are accessed via the General Service 3161 * Interface (GSI). It's role is to facilitate establishment of connections 3162 * between nodes as well as other management related tasks for established 3163 * connections. 3164 * 3165 * Return: true if the port supports an IB CM (this does not guarantee that 3166 * a CM is actually running however). 3167 */ 3168 static inline bool rdma_cap_ib_cm(const struct ib_device *device, u8 port_num) 3169 { 3170 return device->port_data[port_num].immutable.core_cap_flags & 3171 RDMA_CORE_CAP_IB_CM; 3172 } 3173 3174 /** 3175 * rdma_cap_iw_cm - Check if the port of device has the capability IWARP 3176 * Communication Manager. 3177 * @device: Device to check 3178 * @port_num: Port number to check 3179 * 3180 * Similar to above, but specific to iWARP connections which have a different 3181 * managment protocol than InfiniBand. 3182 * 3183 * Return: true if the port supports an iWARP CM (this does not guarantee that 3184 * a CM is actually running however). 3185 */ 3186 static inline bool rdma_cap_iw_cm(const struct ib_device *device, u8 port_num) 3187 { 3188 return device->port_data[port_num].immutable.core_cap_flags & 3189 RDMA_CORE_CAP_IW_CM; 3190 } 3191 3192 /** 3193 * rdma_cap_ib_sa - Check if the port of device has the capability Infiniband 3194 * Subnet Administration. 3195 * @device: Device to check 3196 * @port_num: Port number to check 3197 * 3198 * An InfiniBand Subnet Administration (SA) service is a pre-defined General 3199 * Service Agent (GSA) provided by the Subnet Manager (SM). On InfiniBand 3200 * fabrics, devices should resolve routes to other hosts by contacting the 3201 * SA to query the proper route. 3202 * 3203 * Return: true if the port should act as a client to the fabric Subnet 3204 * Administration interface. This does not imply that the SA service is 3205 * running locally. 3206 */ 3207 static inline bool rdma_cap_ib_sa(const struct ib_device *device, u8 port_num) 3208 { 3209 return device->port_data[port_num].immutable.core_cap_flags & 3210 RDMA_CORE_CAP_IB_SA; 3211 } 3212 3213 /** 3214 * rdma_cap_ib_mcast - Check if the port of device has the capability Infiniband 3215 * Multicast. 3216 * @device: Device to check 3217 * @port_num: Port number to check 3218 * 3219 * InfiniBand multicast registration is more complex than normal IPv4 or 3220 * IPv6 multicast registration. Each Host Channel Adapter must register 3221 * with the Subnet Manager when it wishes to join a multicast group. It 3222 * should do so only once regardless of how many queue pairs it subscribes 3223 * to this group. And it should leave the group only after all queue pairs 3224 * attached to the group have been detached. 3225 * 3226 * Return: true if the port must undertake the additional adminstrative 3227 * overhead of registering/unregistering with the SM and tracking of the 3228 * total number of queue pairs attached to the multicast group. 3229 */ 3230 static inline bool rdma_cap_ib_mcast(const struct ib_device *device, u8 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, u8 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, u8 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, u8 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, u8 port_num) 3304 { 3305 return device->port_data[port_num].immutable.max_mad_size; 3306 } 3307 3308 /** 3309 * rdma_cap_roce_gid_table - Check if the port of device uses roce_gid_table 3310 * @device: Device to check 3311 * @port_num: Port number to check 3312 * 3313 * RoCE GID table mechanism manages the various GIDs for a device. 3314 * 3315 * NOTE: if allocating the port's GID table has failed, this call will still 3316 * return true, but any RoCE GID table API will fail. 3317 * 3318 * Return: true if the port uses RoCE GID table mechanism in order to manage 3319 * its GIDs. 3320 */ 3321 static inline bool rdma_cap_roce_gid_table(const struct ib_device *device, 3322 u8 port_num) 3323 { 3324 return rdma_protocol_roce(device, port_num) && 3325 device->ops.add_gid && device->ops.del_gid; 3326 } 3327 3328 /* 3329 * Check if the device supports READ W/ INVALIDATE. 3330 */ 3331 static inline bool rdma_cap_read_inv(struct ib_device *dev, u32 port_num) 3332 { 3333 /* 3334 * iWarp drivers must support READ W/ INVALIDATE. No other protocol 3335 * has support for it yet. 3336 */ 3337 return rdma_protocol_iwarp(dev, port_num); 3338 } 3339 3340 /** 3341 * rdma_core_cap_opa_port - Return whether the RDMA Port is OPA or not. 3342 * @device: Device 3343 * @port_num: 1 based Port number 3344 * 3345 * Return true if port is an Intel OPA port , false if not 3346 */ 3347 static inline bool rdma_core_cap_opa_port(struct ib_device *device, 3348 u32 port_num) 3349 { 3350 return (device->port_data[port_num].immutable.core_cap_flags & 3351 RDMA_CORE_PORT_INTEL_OPA) == RDMA_CORE_PORT_INTEL_OPA; 3352 } 3353 3354 /** 3355 * rdma_mtu_enum_to_int - Return the mtu of the port as an integer value. 3356 * @device: Device 3357 * @port_num: Port number 3358 * @mtu: enum value of MTU 3359 * 3360 * Return the MTU size supported by the port as an integer value. Will return 3361 * -1 if enum value of mtu is not supported. 3362 */ 3363 static inline int rdma_mtu_enum_to_int(struct ib_device *device, u8 port, 3364 int mtu) 3365 { 3366 if (rdma_core_cap_opa_port(device, port)) 3367 return opa_mtu_enum_to_int((enum opa_mtu)mtu); 3368 else 3369 return ib_mtu_enum_to_int((enum ib_mtu)mtu); 3370 } 3371 3372 /** 3373 * rdma_mtu_from_attr - Return the mtu of the port from the port attribute. 3374 * @device: Device 3375 * @port_num: Port number 3376 * @attr: port attribute 3377 * 3378 * Return the MTU size supported by the port as an integer value. 3379 */ 3380 static inline int rdma_mtu_from_attr(struct ib_device *device, u8 port, 3381 struct ib_port_attr *attr) 3382 { 3383 if (rdma_core_cap_opa_port(device, port)) 3384 return attr->phys_mtu; 3385 else 3386 return ib_mtu_enum_to_int(attr->max_mtu); 3387 } 3388 3389 int ib_set_vf_link_state(struct ib_device *device, int vf, u8 port, 3390 int state); 3391 int ib_get_vf_config(struct ib_device *device, int vf, u8 port, 3392 struct ifla_vf_info *info); 3393 int ib_get_vf_stats(struct ib_device *device, int vf, u8 port, 3394 struct ifla_vf_stats *stats); 3395 int ib_get_vf_guid(struct ib_device *device, int vf, u8 port, 3396 struct ifla_vf_guid *node_guid, 3397 struct ifla_vf_guid *port_guid); 3398 int ib_set_vf_guid(struct ib_device *device, int vf, u8 port, u64 guid, 3399 int type); 3400 3401 int ib_query_pkey(struct ib_device *device, 3402 u8 port_num, u16 index, u16 *pkey); 3403 3404 int ib_modify_device(struct ib_device *device, 3405 int device_modify_mask, 3406 struct ib_device_modify *device_modify); 3407 3408 int ib_modify_port(struct ib_device *device, 3409 u8 port_num, int port_modify_mask, 3410 struct ib_port_modify *port_modify); 3411 3412 int ib_find_gid(struct ib_device *device, union ib_gid *gid, 3413 u8 *port_num, u16 *index); 3414 3415 int ib_find_pkey(struct ib_device *device, 3416 u8 port_num, u16 pkey, u16 *index); 3417 3418 enum ib_pd_flags { 3419 /* 3420 * Create a memory registration for all memory in the system and place 3421 * the rkey for it into pd->unsafe_global_rkey. This can be used by 3422 * ULPs to avoid the overhead of dynamic MRs. 3423 * 3424 * This flag is generally considered unsafe and must only be used in 3425 * extremly trusted environments. Every use of it will log a warning 3426 * in the kernel log. 3427 */ 3428 IB_PD_UNSAFE_GLOBAL_RKEY = 0x01, 3429 }; 3430 3431 struct ib_pd *__ib_alloc_pd(struct ib_device *device, unsigned int flags, 3432 const char *caller); 3433 3434 #define ib_alloc_pd(device, flags) \ 3435 __ib_alloc_pd((device), (flags), KBUILD_MODNAME) 3436 3437 int ib_dealloc_pd_user(struct ib_pd *pd, struct ib_udata *udata); 3438 3439 /** 3440 * ib_dealloc_pd - Deallocate kernel PD 3441 * @pd: The protection domain 3442 * 3443 * NOTE: for user PD use ib_dealloc_pd_user with valid udata! 3444 */ 3445 static inline void ib_dealloc_pd(struct ib_pd *pd) 3446 { 3447 int ret = ib_dealloc_pd_user(pd, NULL); 3448 3449 WARN_ONCE(ret, "Destroy of kernel PD shouldn't fail"); 3450 } 3451 3452 enum rdma_create_ah_flags { 3453 /* In a sleepable context */ 3454 RDMA_CREATE_AH_SLEEPABLE = BIT(0), 3455 }; 3456 3457 /** 3458 * rdma_create_ah - Creates an address handle for the given address vector. 3459 * @pd: The protection domain associated with the address handle. 3460 * @ah_attr: The attributes of the address vector. 3461 * @flags: Create address handle flags (see enum rdma_create_ah_flags). 3462 * 3463 * The address handle is used to reference a local or global destination 3464 * in all UD QP post sends. 3465 */ 3466 struct ib_ah *rdma_create_ah(struct ib_pd *pd, struct rdma_ah_attr *ah_attr, 3467 u32 flags); 3468 3469 /** 3470 * rdma_create_user_ah - Creates an address handle for the given address vector. 3471 * It resolves destination mac address for ah attribute of RoCE type. 3472 * @pd: The protection domain associated with the address handle. 3473 * @ah_attr: The attributes of the address vector. 3474 * @udata: pointer to user's input output buffer information need by 3475 * provider driver. 3476 * 3477 * It returns 0 on success and returns appropriate error code on error. 3478 * The address handle is used to reference a local or global destination 3479 * in all UD QP post sends. 3480 */ 3481 struct ib_ah *rdma_create_user_ah(struct ib_pd *pd, 3482 struct rdma_ah_attr *ah_attr, 3483 struct ib_udata *udata); 3484 /** 3485 * ib_get_gids_from_rdma_hdr - Get sgid and dgid from GRH or IPv4 header 3486 * work completion. 3487 * @hdr: the L3 header to parse 3488 * @net_type: type of header to parse 3489 * @sgid: place to store source gid 3490 * @dgid: place to store destination gid 3491 */ 3492 int ib_get_gids_from_rdma_hdr(const union rdma_network_hdr *hdr, 3493 enum rdma_network_type net_type, 3494 union ib_gid *sgid, union ib_gid *dgid); 3495 3496 /** 3497 * ib_get_rdma_header_version - Get the header version 3498 * @hdr: the L3 header to parse 3499 */ 3500 int ib_get_rdma_header_version(const union rdma_network_hdr *hdr); 3501 3502 /** 3503 * ib_init_ah_attr_from_wc - Initializes address handle attributes from a 3504 * work completion. 3505 * @device: Device on which the received message arrived. 3506 * @port_num: Port on which the received message arrived. 3507 * @wc: Work completion associated with the received message. 3508 * @grh: References the received global route header. This parameter is 3509 * ignored unless the work completion indicates that the GRH is valid. 3510 * @ah_attr: Returned attributes that can be used when creating an address 3511 * handle for replying to the message. 3512 * When ib_init_ah_attr_from_wc() returns success, 3513 * (a) for IB link layer it optionally contains a reference to SGID attribute 3514 * when GRH is present for IB link layer. 3515 * (b) for RoCE link layer it contains a reference to SGID attribute. 3516 * User must invoke rdma_cleanup_ah_attr_gid_attr() to release reference to SGID 3517 * attributes which are initialized using ib_init_ah_attr_from_wc(). 3518 * 3519 */ 3520 int ib_init_ah_attr_from_wc(struct ib_device *device, u8 port_num, 3521 const struct ib_wc *wc, const struct ib_grh *grh, 3522 struct rdma_ah_attr *ah_attr); 3523 3524 /** 3525 * ib_create_ah_from_wc - Creates an address handle associated with the 3526 * sender of the specified work completion. 3527 * @pd: The protection domain associated with the address handle. 3528 * @wc: Work completion information associated with a received message. 3529 * @grh: References the received global route header. This parameter is 3530 * ignored unless the work completion indicates that the GRH is valid. 3531 * @port_num: The outbound port number to associate with the address. 3532 * 3533 * The address handle is used to reference a local or global destination 3534 * in all UD QP post sends. 3535 */ 3536 struct ib_ah *ib_create_ah_from_wc(struct ib_pd *pd, const struct ib_wc *wc, 3537 const struct ib_grh *grh, u8 port_num); 3538 3539 /** 3540 * rdma_modify_ah - Modifies the address vector associated with an address 3541 * handle. 3542 * @ah: The address handle to modify. 3543 * @ah_attr: The new address vector attributes to associate with the 3544 * address handle. 3545 */ 3546 int rdma_modify_ah(struct ib_ah *ah, struct rdma_ah_attr *ah_attr); 3547 3548 /** 3549 * rdma_query_ah - Queries the address vector associated with an address 3550 * handle. 3551 * @ah: The address handle to query. 3552 * @ah_attr: The address vector attributes associated with the address 3553 * handle. 3554 */ 3555 int rdma_query_ah(struct ib_ah *ah, struct rdma_ah_attr *ah_attr); 3556 3557 enum rdma_destroy_ah_flags { 3558 /* In a sleepable context */ 3559 RDMA_DESTROY_AH_SLEEPABLE = BIT(0), 3560 }; 3561 3562 /** 3563 * rdma_destroy_ah_user - Destroys an address handle. 3564 * @ah: The address handle to destroy. 3565 * @flags: Destroy address handle flags (see enum rdma_destroy_ah_flags). 3566 * @udata: Valid user data or NULL for kernel objects 3567 */ 3568 int rdma_destroy_ah_user(struct ib_ah *ah, u32 flags, struct ib_udata *udata); 3569 3570 /** 3571 * rdma_destroy_ah - Destroys an kernel address handle. 3572 * @ah: The address handle to destroy. 3573 * @flags: Destroy address handle flags (see enum rdma_destroy_ah_flags). 3574 * 3575 * NOTE: for user ah use rdma_destroy_ah_user with valid udata! 3576 */ 3577 static inline void rdma_destroy_ah(struct ib_ah *ah, u32 flags) 3578 { 3579 int ret = rdma_destroy_ah_user(ah, flags, NULL); 3580 3581 WARN_ONCE(ret, "Destroy of kernel AH shouldn't fail"); 3582 } 3583 3584 struct ib_srq *ib_create_srq_user(struct ib_pd *pd, 3585 struct ib_srq_init_attr *srq_init_attr, 3586 struct ib_usrq_object *uobject, 3587 struct ib_udata *udata); 3588 static inline struct ib_srq * 3589 ib_create_srq(struct ib_pd *pd, struct ib_srq_init_attr *srq_init_attr) 3590 { 3591 if (!pd->device->ops.create_srq) 3592 return ERR_PTR(-EOPNOTSUPP); 3593 3594 return ib_create_srq_user(pd, srq_init_attr, NULL, NULL); 3595 } 3596 3597 /** 3598 * ib_modify_srq - Modifies the attributes for the specified SRQ. 3599 * @srq: The SRQ to modify. 3600 * @srq_attr: On input, specifies the SRQ attributes to modify. On output, 3601 * the current values of selected SRQ attributes are returned. 3602 * @srq_attr_mask: A bit-mask used to specify which attributes of the SRQ 3603 * are being modified. 3604 * 3605 * The mask may contain IB_SRQ_MAX_WR to resize the SRQ and/or 3606 * IB_SRQ_LIMIT to set the SRQ's limit and request notification when 3607 * the number of receives queued drops below the limit. 3608 */ 3609 int ib_modify_srq(struct ib_srq *srq, 3610 struct ib_srq_attr *srq_attr, 3611 enum ib_srq_attr_mask srq_attr_mask); 3612 3613 /** 3614 * ib_query_srq - Returns the attribute list and current values for the 3615 * specified SRQ. 3616 * @srq: The SRQ to query. 3617 * @srq_attr: The attributes of the specified SRQ. 3618 */ 3619 int ib_query_srq(struct ib_srq *srq, 3620 struct ib_srq_attr *srq_attr); 3621 3622 /** 3623 * ib_destroy_srq_user - Destroys the specified SRQ. 3624 * @srq: The SRQ to destroy. 3625 * @udata: Valid user data or NULL for kernel objects 3626 */ 3627 int ib_destroy_srq_user(struct ib_srq *srq, struct ib_udata *udata); 3628 3629 /** 3630 * ib_destroy_srq - Destroys the specified kernel SRQ. 3631 * @srq: The SRQ to destroy. 3632 * 3633 * NOTE: for user srq use ib_destroy_srq_user with valid udata! 3634 */ 3635 static inline void ib_destroy_srq(struct ib_srq *srq) 3636 { 3637 int ret = ib_destroy_srq_user(srq, NULL); 3638 3639 WARN_ONCE(ret, "Destroy of kernel SRQ shouldn't fail"); 3640 } 3641 3642 /** 3643 * ib_post_srq_recv - Posts a list of work requests to the specified SRQ. 3644 * @srq: The SRQ to post the work request on. 3645 * @recv_wr: A list of work requests to post on the receive queue. 3646 * @bad_recv_wr: On an immediate failure, this parameter will reference 3647 * the work request that failed to be posted on the QP. 3648 */ 3649 static inline int ib_post_srq_recv(struct ib_srq *srq, 3650 const struct ib_recv_wr *recv_wr, 3651 const struct ib_recv_wr **bad_recv_wr) 3652 { 3653 const struct ib_recv_wr *dummy; 3654 3655 return srq->device->ops.post_srq_recv(srq, recv_wr, 3656 bad_recv_wr ? : &dummy); 3657 } 3658 3659 struct ib_qp *ib_create_qp(struct ib_pd *pd, 3660 struct ib_qp_init_attr *qp_init_attr); 3661 3662 /** 3663 * ib_modify_qp_with_udata - Modifies the attributes for the specified QP. 3664 * @qp: The QP to modify. 3665 * @attr: On input, specifies the QP attributes to modify. On output, 3666 * the current values of selected QP attributes are returned. 3667 * @attr_mask: A bit-mask used to specify which attributes of the QP 3668 * are being modified. 3669 * @udata: pointer to user's input output buffer information 3670 * are being modified. 3671 * It returns 0 on success and returns appropriate error code on error. 3672 */ 3673 int ib_modify_qp_with_udata(struct ib_qp *qp, 3674 struct ib_qp_attr *attr, 3675 int attr_mask, 3676 struct ib_udata *udata); 3677 3678 /** 3679 * ib_modify_qp - Modifies the attributes for the specified QP and then 3680 * transitions the QP to the given state. 3681 * @qp: The QP to modify. 3682 * @qp_attr: On input, specifies the QP attributes to modify. On output, 3683 * the current values of selected QP attributes are returned. 3684 * @qp_attr_mask: A bit-mask used to specify which attributes of the QP 3685 * are being modified. 3686 */ 3687 int ib_modify_qp(struct ib_qp *qp, 3688 struct ib_qp_attr *qp_attr, 3689 int qp_attr_mask); 3690 3691 /** 3692 * ib_query_qp - Returns the attribute list and current values for the 3693 * specified QP. 3694 * @qp: The QP to query. 3695 * @qp_attr: The attributes of the specified QP. 3696 * @qp_attr_mask: A bit-mask used to select specific attributes to query. 3697 * @qp_init_attr: Additional attributes of the selected QP. 3698 * 3699 * The qp_attr_mask may be used to limit the query to gathering only the 3700 * selected attributes. 3701 */ 3702 int ib_query_qp(struct ib_qp *qp, 3703 struct ib_qp_attr *qp_attr, 3704 int qp_attr_mask, 3705 struct ib_qp_init_attr *qp_init_attr); 3706 3707 /** 3708 * ib_destroy_qp - Destroys the specified QP. 3709 * @qp: The QP to destroy. 3710 * @udata: Valid udata or NULL for kernel objects 3711 */ 3712 int ib_destroy_qp_user(struct ib_qp *qp, struct ib_udata *udata); 3713 3714 /** 3715 * ib_destroy_qp - Destroys the specified kernel QP. 3716 * @qp: The QP to destroy. 3717 * 3718 * NOTE: for user qp use ib_destroy_qp_user with valid udata! 3719 */ 3720 static inline int ib_destroy_qp(struct ib_qp *qp) 3721 { 3722 return ib_destroy_qp_user(qp, NULL); 3723 } 3724 3725 /** 3726 * ib_open_qp - Obtain a reference to an existing sharable QP. 3727 * @xrcd - XRC domain 3728 * @qp_open_attr: Attributes identifying the QP to open. 3729 * 3730 * Returns a reference to a sharable QP. 3731 */ 3732 struct ib_qp *ib_open_qp(struct ib_xrcd *xrcd, 3733 struct ib_qp_open_attr *qp_open_attr); 3734 3735 /** 3736 * ib_close_qp - Release an external reference to a QP. 3737 * @qp: The QP handle to release 3738 * 3739 * The opened QP handle is released by the caller. The underlying 3740 * shared QP is not destroyed until all internal references are released. 3741 */ 3742 int ib_close_qp(struct ib_qp *qp); 3743 3744 /** 3745 * ib_post_send - Posts a list of work requests to the send queue of 3746 * the specified QP. 3747 * @qp: The QP to post the work request on. 3748 * @send_wr: A list of work requests to post on the send queue. 3749 * @bad_send_wr: On an immediate failure, this parameter will reference 3750 * the work request that failed to be posted on the QP. 3751 * 3752 * While IBA Vol. 1 section 11.4.1.1 specifies that if an immediate 3753 * error is returned, the QP state shall not be affected, 3754 * ib_post_send() will return an immediate error after queueing any 3755 * earlier work requests in the list. 3756 */ 3757 static inline int ib_post_send(struct ib_qp *qp, 3758 const struct ib_send_wr *send_wr, 3759 const struct ib_send_wr **bad_send_wr) 3760 { 3761 const struct ib_send_wr *dummy; 3762 3763 return qp->device->ops.post_send(qp, send_wr, bad_send_wr ? : &dummy); 3764 } 3765 3766 /** 3767 * ib_post_recv - Posts a list of work requests to the receive queue of 3768 * the specified QP. 3769 * @qp: The QP to post the work request on. 3770 * @recv_wr: A list of work requests to post on the receive queue. 3771 * @bad_recv_wr: On an immediate failure, this parameter will reference 3772 * the work request that failed to be posted on the QP. 3773 */ 3774 static inline int ib_post_recv(struct ib_qp *qp, 3775 const struct ib_recv_wr *recv_wr, 3776 const struct ib_recv_wr **bad_recv_wr) 3777 { 3778 const struct ib_recv_wr *dummy; 3779 3780 return qp->device->ops.post_recv(qp, recv_wr, bad_recv_wr ? : &dummy); 3781 } 3782 3783 struct ib_cq *__ib_alloc_cq(struct ib_device *dev, void *private, int nr_cqe, 3784 int comp_vector, enum ib_poll_context poll_ctx, 3785 const char *caller); 3786 static inline struct ib_cq *ib_alloc_cq(struct ib_device *dev, void *private, 3787 int nr_cqe, int comp_vector, 3788 enum ib_poll_context poll_ctx) 3789 { 3790 return __ib_alloc_cq(dev, private, nr_cqe, comp_vector, poll_ctx, 3791 KBUILD_MODNAME); 3792 } 3793 3794 struct ib_cq *__ib_alloc_cq_any(struct ib_device *dev, void *private, 3795 int nr_cqe, enum ib_poll_context poll_ctx, 3796 const char *caller); 3797 3798 /** 3799 * ib_alloc_cq_any: Allocate kernel CQ 3800 * @dev: The IB device 3801 * @private: Private data attached to the CQE 3802 * @nr_cqe: Number of CQEs in the CQ 3803 * @poll_ctx: Context used for polling the CQ 3804 */ 3805 static inline struct ib_cq *ib_alloc_cq_any(struct ib_device *dev, 3806 void *private, int nr_cqe, 3807 enum ib_poll_context poll_ctx) 3808 { 3809 return __ib_alloc_cq_any(dev, private, nr_cqe, poll_ctx, 3810 KBUILD_MODNAME); 3811 } 3812 3813 void ib_free_cq(struct ib_cq *cq); 3814 int ib_process_cq_direct(struct ib_cq *cq, int budget); 3815 3816 /** 3817 * ib_create_cq - Creates a CQ on the specified device. 3818 * @device: The device on which to create the CQ. 3819 * @comp_handler: A user-specified callback that is invoked when a 3820 * completion event occurs on the CQ. 3821 * @event_handler: A user-specified callback that is invoked when an 3822 * asynchronous event not associated with a completion occurs on the CQ. 3823 * @cq_context: Context associated with the CQ returned to the user via 3824 * the associated completion and event handlers. 3825 * @cq_attr: The attributes the CQ should be created upon. 3826 * 3827 * Users can examine the cq structure to determine the actual CQ size. 3828 */ 3829 struct ib_cq *__ib_create_cq(struct ib_device *device, 3830 ib_comp_handler comp_handler, 3831 void (*event_handler)(struct ib_event *, void *), 3832 void *cq_context, 3833 const struct ib_cq_init_attr *cq_attr, 3834 const char *caller); 3835 #define ib_create_cq(device, cmp_hndlr, evt_hndlr, cq_ctxt, cq_attr) \ 3836 __ib_create_cq((device), (cmp_hndlr), (evt_hndlr), (cq_ctxt), (cq_attr), KBUILD_MODNAME) 3837 3838 /** 3839 * ib_resize_cq - Modifies the capacity of the CQ. 3840 * @cq: The CQ to resize. 3841 * @cqe: The minimum size of the CQ. 3842 * 3843 * Users can examine the cq structure to determine the actual CQ size. 3844 */ 3845 int ib_resize_cq(struct ib_cq *cq, int cqe); 3846 3847 /** 3848 * rdma_set_cq_moderation - Modifies moderation params of the CQ 3849 * @cq: The CQ to modify. 3850 * @cq_count: number of CQEs that will trigger an event 3851 * @cq_period: max period of time in usec before triggering an event 3852 * 3853 */ 3854 int rdma_set_cq_moderation(struct ib_cq *cq, u16 cq_count, u16 cq_period); 3855 3856 /** 3857 * ib_destroy_cq_user - Destroys the specified CQ. 3858 * @cq: The CQ to destroy. 3859 * @udata: Valid user data or NULL for kernel objects 3860 */ 3861 int ib_destroy_cq_user(struct ib_cq *cq, struct ib_udata *udata); 3862 3863 /** 3864 * ib_destroy_cq - Destroys the specified kernel CQ. 3865 * @cq: The CQ to destroy. 3866 * 3867 * NOTE: for user cq use ib_destroy_cq_user with valid udata! 3868 */ 3869 static inline void ib_destroy_cq(struct ib_cq *cq) 3870 { 3871 int ret = ib_destroy_cq_user(cq, NULL); 3872 3873 WARN_ONCE(ret, "Destroy of kernel CQ shouldn't fail"); 3874 } 3875 3876 /** 3877 * ib_poll_cq - poll a CQ for completion(s) 3878 * @cq:the CQ being polled 3879 * @num_entries:maximum number of completions to return 3880 * @wc:array of at least @num_entries &struct ib_wc where completions 3881 * will be returned 3882 * 3883 * Poll a CQ for (possibly multiple) completions. If the return value 3884 * is < 0, an error occurred. If the return value is >= 0, it is the 3885 * number of completions returned. If the return value is 3886 * non-negative and < num_entries, then the CQ was emptied. 3887 */ 3888 static inline int ib_poll_cq(struct ib_cq *cq, int num_entries, 3889 struct ib_wc *wc) 3890 { 3891 return cq->device->ops.poll_cq(cq, num_entries, wc); 3892 } 3893 3894 /** 3895 * ib_req_notify_cq - Request completion notification on a CQ. 3896 * @cq: The CQ to generate an event for. 3897 * @flags: 3898 * Must contain exactly one of %IB_CQ_SOLICITED or %IB_CQ_NEXT_COMP 3899 * to request an event on the next solicited event or next work 3900 * completion at any type, respectively. %IB_CQ_REPORT_MISSED_EVENTS 3901 * may also be |ed in to request a hint about missed events, as 3902 * described below. 3903 * 3904 * Return Value: 3905 * < 0 means an error occurred while requesting notification 3906 * == 0 means notification was requested successfully, and if 3907 * IB_CQ_REPORT_MISSED_EVENTS was passed in, then no events 3908 * were missed and it is safe to wait for another event. In 3909 * this case is it guaranteed that any work completions added 3910 * to the CQ since the last CQ poll will trigger a completion 3911 * notification event. 3912 * > 0 is only returned if IB_CQ_REPORT_MISSED_EVENTS was passed 3913 * in. It means that the consumer must poll the CQ again to 3914 * make sure it is empty to avoid missing an event because of a 3915 * race between requesting notification and an entry being 3916 * added to the CQ. This return value means it is possible 3917 * (but not guaranteed) that a work completion has been added 3918 * to the CQ since the last poll without triggering a 3919 * completion notification event. 3920 */ 3921 static inline int ib_req_notify_cq(struct ib_cq *cq, 3922 enum ib_cq_notify_flags flags) 3923 { 3924 return cq->device->ops.req_notify_cq(cq, flags); 3925 } 3926 3927 struct ib_cq *ib_cq_pool_get(struct ib_device *dev, unsigned int nr_cqe, 3928 int comp_vector_hint, 3929 enum ib_poll_context poll_ctx); 3930 3931 void ib_cq_pool_put(struct ib_cq *cq, unsigned int nr_cqe); 3932 3933 /** 3934 * ib_req_ncomp_notif - Request completion notification when there are 3935 * at least the specified number of unreaped completions on the CQ. 3936 * @cq: The CQ to generate an event for. 3937 * @wc_cnt: The number of unreaped completions that should be on the 3938 * CQ before an event is generated. 3939 */ 3940 static inline int ib_req_ncomp_notif(struct ib_cq *cq, int wc_cnt) 3941 { 3942 return cq->device->ops.req_ncomp_notif ? 3943 cq->device->ops.req_ncomp_notif(cq, wc_cnt) : 3944 -ENOSYS; 3945 } 3946 3947 /** 3948 * ib_dma_mapping_error - check a DMA addr for error 3949 * @dev: The device for which the dma_addr was created 3950 * @dma_addr: The DMA address to check 3951 */ 3952 static inline int ib_dma_mapping_error(struct ib_device *dev, u64 dma_addr) 3953 { 3954 return dma_mapping_error(dev->dma_device, dma_addr); 3955 } 3956 3957 /** 3958 * ib_dma_map_single - Map a kernel virtual address to DMA address 3959 * @dev: The device for which the dma_addr is to be created 3960 * @cpu_addr: The kernel virtual address 3961 * @size: The size of the region in bytes 3962 * @direction: The direction of the DMA 3963 */ 3964 static inline u64 ib_dma_map_single(struct ib_device *dev, 3965 void *cpu_addr, size_t size, 3966 enum dma_data_direction direction) 3967 { 3968 return dma_map_single(dev->dma_device, cpu_addr, size, direction); 3969 } 3970 3971 /** 3972 * ib_dma_unmap_single - Destroy a mapping created by ib_dma_map_single() 3973 * @dev: The device for which the DMA address was created 3974 * @addr: The DMA address 3975 * @size: The size of the region in bytes 3976 * @direction: The direction of the DMA 3977 */ 3978 static inline void ib_dma_unmap_single(struct ib_device *dev, 3979 u64 addr, size_t size, 3980 enum dma_data_direction direction) 3981 { 3982 dma_unmap_single(dev->dma_device, addr, size, direction); 3983 } 3984 3985 /** 3986 * ib_dma_map_page - Map a physical page to DMA address 3987 * @dev: The device for which the dma_addr is to be created 3988 * @page: The page to be mapped 3989 * @offset: The offset within the page 3990 * @size: The size of the region in bytes 3991 * @direction: The direction of the DMA 3992 */ 3993 static inline u64 ib_dma_map_page(struct ib_device *dev, 3994 struct page *page, 3995 unsigned long offset, 3996 size_t size, 3997 enum dma_data_direction direction) 3998 { 3999 return dma_map_page(dev->dma_device, page, offset, size, direction); 4000 } 4001 4002 /** 4003 * ib_dma_unmap_page - Destroy a mapping created by ib_dma_map_page() 4004 * @dev: The device for which the DMA address was created 4005 * @addr: The DMA address 4006 * @size: The size of the region in bytes 4007 * @direction: The direction of the DMA 4008 */ 4009 static inline void ib_dma_unmap_page(struct ib_device *dev, 4010 u64 addr, size_t size, 4011 enum dma_data_direction direction) 4012 { 4013 dma_unmap_page(dev->dma_device, addr, size, direction); 4014 } 4015 4016 /** 4017 * ib_dma_map_sg - Map a scatter/gather list to DMA addresses 4018 * @dev: The device for which the DMA addresses are to be created 4019 * @sg: The array of scatter/gather entries 4020 * @nents: The number of scatter/gather entries 4021 * @direction: The direction of the DMA 4022 */ 4023 static inline int ib_dma_map_sg(struct ib_device *dev, 4024 struct scatterlist *sg, int nents, 4025 enum dma_data_direction direction) 4026 { 4027 return dma_map_sg(dev->dma_device, sg, nents, direction); 4028 } 4029 4030 /** 4031 * ib_dma_unmap_sg - Unmap a scatter/gather list of DMA addresses 4032 * @dev: The device for which the DMA addresses were created 4033 * @sg: The array of scatter/gather entries 4034 * @nents: The number of scatter/gather entries 4035 * @direction: The direction of the DMA 4036 */ 4037 static inline void ib_dma_unmap_sg(struct ib_device *dev, 4038 struct scatterlist *sg, int nents, 4039 enum dma_data_direction direction) 4040 { 4041 dma_unmap_sg(dev->dma_device, sg, nents, direction); 4042 } 4043 4044 static inline int ib_dma_map_sg_attrs(struct ib_device *dev, 4045 struct scatterlist *sg, int nents, 4046 enum dma_data_direction direction, 4047 unsigned long dma_attrs) 4048 { 4049 return dma_map_sg_attrs(dev->dma_device, sg, nents, direction, 4050 dma_attrs); 4051 } 4052 4053 static inline void ib_dma_unmap_sg_attrs(struct ib_device *dev, 4054 struct scatterlist *sg, int nents, 4055 enum dma_data_direction direction, 4056 unsigned long dma_attrs) 4057 { 4058 dma_unmap_sg_attrs(dev->dma_device, sg, nents, direction, dma_attrs); 4059 } 4060 4061 /** 4062 * ib_dma_max_seg_size - Return the size limit of a single DMA transfer 4063 * @dev: The device to query 4064 * 4065 * The returned value represents a size in bytes. 4066 */ 4067 static inline unsigned int ib_dma_max_seg_size(struct ib_device *dev) 4068 { 4069 return dma_get_max_seg_size(dev->dma_device); 4070 } 4071 4072 /** 4073 * ib_dma_sync_single_for_cpu - Prepare DMA region to be accessed by CPU 4074 * @dev: The device for which the DMA address was created 4075 * @addr: The DMA address 4076 * @size: The size of the region in bytes 4077 * @dir: The direction of the DMA 4078 */ 4079 static inline void ib_dma_sync_single_for_cpu(struct ib_device *dev, 4080 u64 addr, 4081 size_t size, 4082 enum dma_data_direction dir) 4083 { 4084 dma_sync_single_for_cpu(dev->dma_device, addr, size, dir); 4085 } 4086 4087 /** 4088 * ib_dma_sync_single_for_device - Prepare DMA region to be accessed by device 4089 * @dev: The device for which the DMA address was created 4090 * @addr: The DMA address 4091 * @size: The size of the region in bytes 4092 * @dir: The direction of the DMA 4093 */ 4094 static inline void ib_dma_sync_single_for_device(struct ib_device *dev, 4095 u64 addr, 4096 size_t size, 4097 enum dma_data_direction dir) 4098 { 4099 dma_sync_single_for_device(dev->dma_device, addr, size, dir); 4100 } 4101 4102 /** 4103 * ib_dma_alloc_coherent - Allocate memory and map it for DMA 4104 * @dev: The device for which the DMA address is requested 4105 * @size: The size of the region to allocate in bytes 4106 * @dma_handle: A pointer for returning the DMA address of the region 4107 * @flag: memory allocator flags 4108 */ 4109 static inline void *ib_dma_alloc_coherent(struct ib_device *dev, 4110 size_t size, 4111 dma_addr_t *dma_handle, 4112 gfp_t flag) 4113 { 4114 return dma_alloc_coherent(dev->dma_device, size, dma_handle, flag); 4115 } 4116 4117 /** 4118 * ib_dma_free_coherent - Free memory allocated by ib_dma_alloc_coherent() 4119 * @dev: The device for which the DMA addresses were allocated 4120 * @size: The size of the region 4121 * @cpu_addr: the address returned by ib_dma_alloc_coherent() 4122 * @dma_handle: the DMA address returned by ib_dma_alloc_coherent() 4123 */ 4124 static inline void ib_dma_free_coherent(struct ib_device *dev, 4125 size_t size, void *cpu_addr, 4126 dma_addr_t dma_handle) 4127 { 4128 dma_free_coherent(dev->dma_device, size, cpu_addr, dma_handle); 4129 } 4130 4131 /* ib_reg_user_mr - register a memory region for virtual addresses from kernel 4132 * space. This function should be called when 'current' is the owning MM. 4133 */ 4134 struct ib_mr *ib_reg_user_mr(struct ib_pd *pd, u64 start, u64 length, 4135 u64 virt_addr, int mr_access_flags); 4136 4137 /* ib_advise_mr - give an advice about an address range in a memory region */ 4138 int ib_advise_mr(struct ib_pd *pd, enum ib_uverbs_advise_mr_advice advice, 4139 u32 flags, struct ib_sge *sg_list, u32 num_sge); 4140 /** 4141 * ib_dereg_mr_user - Deregisters a memory region and removes it from the 4142 * HCA translation table. 4143 * @mr: The memory region to deregister. 4144 * @udata: Valid user data or NULL for kernel object 4145 * 4146 * This function can fail, if the memory region has memory windows bound to it. 4147 */ 4148 int ib_dereg_mr_user(struct ib_mr *mr, struct ib_udata *udata); 4149 4150 /** 4151 * ib_dereg_mr - Deregisters a kernel memory region and removes it from the 4152 * HCA translation table. 4153 * @mr: The memory region to deregister. 4154 * 4155 * This function can fail, if the memory region has memory windows bound to it. 4156 * 4157 * NOTE: for user mr use ib_dereg_mr_user with valid udata! 4158 */ 4159 static inline int ib_dereg_mr(struct ib_mr *mr) 4160 { 4161 return ib_dereg_mr_user(mr, NULL); 4162 } 4163 4164 struct ib_mr *ib_alloc_mr(struct ib_pd *pd, enum ib_mr_type mr_type, 4165 u32 max_num_sg); 4166 4167 struct ib_mr *ib_alloc_mr_integrity(struct ib_pd *pd, 4168 u32 max_num_data_sg, 4169 u32 max_num_meta_sg); 4170 4171 /** 4172 * ib_update_fast_reg_key - updates the key portion of the fast_reg MR 4173 * R_Key and L_Key. 4174 * @mr - struct ib_mr pointer to be updated. 4175 * @newkey - new key to be used. 4176 */ 4177 static inline void ib_update_fast_reg_key(struct ib_mr *mr, u8 newkey) 4178 { 4179 mr->lkey = (mr->lkey & 0xffffff00) | newkey; 4180 mr->rkey = (mr->rkey & 0xffffff00) | newkey; 4181 } 4182 4183 /** 4184 * ib_inc_rkey - increments the key portion of the given rkey. Can be used 4185 * for calculating a new rkey for type 2 memory windows. 4186 * @rkey - the rkey to increment. 4187 */ 4188 static inline u32 ib_inc_rkey(u32 rkey) 4189 { 4190 const u32 mask = 0x000000ff; 4191 return ((rkey + 1) & mask) | (rkey & ~mask); 4192 } 4193 4194 /** 4195 * ib_attach_mcast - Attaches the specified QP to a multicast group. 4196 * @qp: QP to attach to the multicast group. The QP must be type 4197 * IB_QPT_UD. 4198 * @gid: Multicast group GID. 4199 * @lid: Multicast group LID in host byte order. 4200 * 4201 * In order to send and receive multicast packets, subnet 4202 * administration must have created the multicast group and configured 4203 * the fabric appropriately. The port associated with the specified 4204 * QP must also be a member of the multicast group. 4205 */ 4206 int ib_attach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid); 4207 4208 /** 4209 * ib_detach_mcast - Detaches the specified QP from a multicast group. 4210 * @qp: QP to detach from the multicast group. 4211 * @gid: Multicast group GID. 4212 * @lid: Multicast group LID in host byte order. 4213 */ 4214 int ib_detach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid); 4215 4216 struct ib_xrcd *ib_alloc_xrcd_user(struct ib_device *device, 4217 struct inode *inode, struct ib_udata *udata); 4218 int ib_dealloc_xrcd_user(struct ib_xrcd *xrcd, struct ib_udata *udata); 4219 4220 static inline int ib_check_mr_access(int flags) 4221 { 4222 /* 4223 * Local write permission is required if remote write or 4224 * remote atomic permission is also requested. 4225 */ 4226 if (flags & (IB_ACCESS_REMOTE_ATOMIC | IB_ACCESS_REMOTE_WRITE) && 4227 !(flags & IB_ACCESS_LOCAL_WRITE)) 4228 return -EINVAL; 4229 4230 if (flags & ~IB_ACCESS_SUPPORTED) 4231 return -EINVAL; 4232 4233 return 0; 4234 } 4235 4236 static inline bool ib_access_writable(int access_flags) 4237 { 4238 /* 4239 * We have writable memory backing the MR if any of the following 4240 * access flags are set. "Local write" and "remote write" obviously 4241 * require write access. "Remote atomic" can do things like fetch and 4242 * add, which will modify memory, and "MW bind" can change permissions 4243 * by binding a window. 4244 */ 4245 return access_flags & 4246 (IB_ACCESS_LOCAL_WRITE | IB_ACCESS_REMOTE_WRITE | 4247 IB_ACCESS_REMOTE_ATOMIC | IB_ACCESS_MW_BIND); 4248 } 4249 4250 /** 4251 * ib_check_mr_status: lightweight check of MR status. 4252 * This routine may provide status checks on a selected 4253 * ib_mr. first use is for signature status check. 4254 * 4255 * @mr: A memory region. 4256 * @check_mask: Bitmask of which checks to perform from 4257 * ib_mr_status_check enumeration. 4258 * @mr_status: The container of relevant status checks. 4259 * failed checks will be indicated in the status bitmask 4260 * and the relevant info shall be in the error item. 4261 */ 4262 int ib_check_mr_status(struct ib_mr *mr, u32 check_mask, 4263 struct ib_mr_status *mr_status); 4264 4265 /** 4266 * ib_device_try_get: Hold a registration lock 4267 * device: The device to lock 4268 * 4269 * A device under an active registration lock cannot become unregistered. It 4270 * is only possible to obtain a registration lock on a device that is fully 4271 * registered, otherwise this function returns false. 4272 * 4273 * The registration lock is only necessary for actions which require the 4274 * device to still be registered. Uses that only require the device pointer to 4275 * be valid should use get_device(&ibdev->dev) to hold the memory. 4276 * 4277 */ 4278 static inline bool ib_device_try_get(struct ib_device *dev) 4279 { 4280 return refcount_inc_not_zero(&dev->refcount); 4281 } 4282 4283 void ib_device_put(struct ib_device *device); 4284 struct ib_device *ib_device_get_by_netdev(struct net_device *ndev, 4285 enum rdma_driver_id driver_id); 4286 struct ib_device *ib_device_get_by_name(const char *name, 4287 enum rdma_driver_id driver_id); 4288 struct net_device *ib_get_net_dev_by_params(struct ib_device *dev, u8 port, 4289 u16 pkey, const union ib_gid *gid, 4290 const struct sockaddr *addr); 4291 int ib_device_set_netdev(struct ib_device *ib_dev, struct net_device *ndev, 4292 unsigned int port); 4293 struct net_device *ib_device_netdev(struct ib_device *dev, u8 port); 4294 4295 struct ib_wq *ib_create_wq(struct ib_pd *pd, 4296 struct ib_wq_init_attr *init_attr); 4297 int ib_destroy_wq_user(struct ib_wq *wq, struct ib_udata *udata); 4298 int ib_modify_wq(struct ib_wq *wq, struct ib_wq_attr *attr, 4299 u32 wq_attr_mask); 4300 4301 int ib_map_mr_sg(struct ib_mr *mr, struct scatterlist *sg, int sg_nents, 4302 unsigned int *sg_offset, unsigned int page_size); 4303 int ib_map_mr_sg_pi(struct ib_mr *mr, struct scatterlist *data_sg, 4304 int data_sg_nents, unsigned int *data_sg_offset, 4305 struct scatterlist *meta_sg, int meta_sg_nents, 4306 unsigned int *meta_sg_offset, unsigned int page_size); 4307 4308 static inline int 4309 ib_map_mr_sg_zbva(struct ib_mr *mr, struct scatterlist *sg, int sg_nents, 4310 unsigned int *sg_offset, unsigned int page_size) 4311 { 4312 int n; 4313 4314 n = ib_map_mr_sg(mr, sg, sg_nents, sg_offset, page_size); 4315 mr->iova = 0; 4316 4317 return n; 4318 } 4319 4320 int ib_sg_to_pages(struct ib_mr *mr, struct scatterlist *sgl, int sg_nents, 4321 unsigned int *sg_offset, int (*set_page)(struct ib_mr *, u64)); 4322 4323 void ib_drain_rq(struct ib_qp *qp); 4324 void ib_drain_sq(struct ib_qp *qp); 4325 void ib_drain_qp(struct ib_qp *qp); 4326 4327 int ib_get_eth_speed(struct ib_device *dev, u8 port_num, u8 *speed, u8 *width); 4328 4329 static inline u8 *rdma_ah_retrieve_dmac(struct rdma_ah_attr *attr) 4330 { 4331 if (attr->type == RDMA_AH_ATTR_TYPE_ROCE) 4332 return attr->roce.dmac; 4333 return NULL; 4334 } 4335 4336 static inline void rdma_ah_set_dlid(struct rdma_ah_attr *attr, u32 dlid) 4337 { 4338 if (attr->type == RDMA_AH_ATTR_TYPE_IB) 4339 attr->ib.dlid = (u16)dlid; 4340 else if (attr->type == RDMA_AH_ATTR_TYPE_OPA) 4341 attr->opa.dlid = dlid; 4342 } 4343 4344 static inline u32 rdma_ah_get_dlid(const struct rdma_ah_attr *attr) 4345 { 4346 if (attr->type == RDMA_AH_ATTR_TYPE_IB) 4347 return attr->ib.dlid; 4348 else if (attr->type == RDMA_AH_ATTR_TYPE_OPA) 4349 return attr->opa.dlid; 4350 return 0; 4351 } 4352 4353 static inline void rdma_ah_set_sl(struct rdma_ah_attr *attr, u8 sl) 4354 { 4355 attr->sl = sl; 4356 } 4357 4358 static inline u8 rdma_ah_get_sl(const struct rdma_ah_attr *attr) 4359 { 4360 return attr->sl; 4361 } 4362 4363 static inline void rdma_ah_set_path_bits(struct rdma_ah_attr *attr, 4364 u8 src_path_bits) 4365 { 4366 if (attr->type == RDMA_AH_ATTR_TYPE_IB) 4367 attr->ib.src_path_bits = src_path_bits; 4368 else if (attr->type == RDMA_AH_ATTR_TYPE_OPA) 4369 attr->opa.src_path_bits = src_path_bits; 4370 } 4371 4372 static inline u8 rdma_ah_get_path_bits(const struct rdma_ah_attr *attr) 4373 { 4374 if (attr->type == RDMA_AH_ATTR_TYPE_IB) 4375 return attr->ib.src_path_bits; 4376 else if (attr->type == RDMA_AH_ATTR_TYPE_OPA) 4377 return attr->opa.src_path_bits; 4378 return 0; 4379 } 4380 4381 static inline void rdma_ah_set_make_grd(struct rdma_ah_attr *attr, 4382 bool make_grd) 4383 { 4384 if (attr->type == RDMA_AH_ATTR_TYPE_OPA) 4385 attr->opa.make_grd = make_grd; 4386 } 4387 4388 static inline bool rdma_ah_get_make_grd(const struct rdma_ah_attr *attr) 4389 { 4390 if (attr->type == RDMA_AH_ATTR_TYPE_OPA) 4391 return attr->opa.make_grd; 4392 return false; 4393 } 4394 4395 static inline void rdma_ah_set_port_num(struct rdma_ah_attr *attr, u8 port_num) 4396 { 4397 attr->port_num = port_num; 4398 } 4399 4400 static inline u8 rdma_ah_get_port_num(const struct rdma_ah_attr *attr) 4401 { 4402 return attr->port_num; 4403 } 4404 4405 static inline void rdma_ah_set_static_rate(struct rdma_ah_attr *attr, 4406 u8 static_rate) 4407 { 4408 attr->static_rate = static_rate; 4409 } 4410 4411 static inline u8 rdma_ah_get_static_rate(const struct rdma_ah_attr *attr) 4412 { 4413 return attr->static_rate; 4414 } 4415 4416 static inline void rdma_ah_set_ah_flags(struct rdma_ah_attr *attr, 4417 enum ib_ah_flags flag) 4418 { 4419 attr->ah_flags = flag; 4420 } 4421 4422 static inline enum ib_ah_flags 4423 rdma_ah_get_ah_flags(const struct rdma_ah_attr *attr) 4424 { 4425 return attr->ah_flags; 4426 } 4427 4428 static inline const struct ib_global_route 4429 *rdma_ah_read_grh(const struct rdma_ah_attr *attr) 4430 { 4431 return &attr->grh; 4432 } 4433 4434 /*To retrieve and modify the grh */ 4435 static inline struct ib_global_route 4436 *rdma_ah_retrieve_grh(struct rdma_ah_attr *attr) 4437 { 4438 return &attr->grh; 4439 } 4440 4441 static inline void rdma_ah_set_dgid_raw(struct rdma_ah_attr *attr, void *dgid) 4442 { 4443 struct ib_global_route *grh = rdma_ah_retrieve_grh(attr); 4444 4445 memcpy(grh->dgid.raw, dgid, sizeof(grh->dgid)); 4446 } 4447 4448 static inline void rdma_ah_set_subnet_prefix(struct rdma_ah_attr *attr, 4449 __be64 prefix) 4450 { 4451 struct ib_global_route *grh = rdma_ah_retrieve_grh(attr); 4452 4453 grh->dgid.global.subnet_prefix = prefix; 4454 } 4455 4456 static inline void rdma_ah_set_interface_id(struct rdma_ah_attr *attr, 4457 __be64 if_id) 4458 { 4459 struct ib_global_route *grh = rdma_ah_retrieve_grh(attr); 4460 4461 grh->dgid.global.interface_id = if_id; 4462 } 4463 4464 static inline void rdma_ah_set_grh(struct rdma_ah_attr *attr, 4465 union ib_gid *dgid, u32 flow_label, 4466 u8 sgid_index, u8 hop_limit, 4467 u8 traffic_class) 4468 { 4469 struct ib_global_route *grh = rdma_ah_retrieve_grh(attr); 4470 4471 attr->ah_flags = IB_AH_GRH; 4472 if (dgid) 4473 grh->dgid = *dgid; 4474 grh->flow_label = flow_label; 4475 grh->sgid_index = sgid_index; 4476 grh->hop_limit = hop_limit; 4477 grh->traffic_class = traffic_class; 4478 grh->sgid_attr = NULL; 4479 } 4480 4481 void rdma_destroy_ah_attr(struct rdma_ah_attr *ah_attr); 4482 void rdma_move_grh_sgid_attr(struct rdma_ah_attr *attr, union ib_gid *dgid, 4483 u32 flow_label, u8 hop_limit, u8 traffic_class, 4484 const struct ib_gid_attr *sgid_attr); 4485 void rdma_copy_ah_attr(struct rdma_ah_attr *dest, 4486 const struct rdma_ah_attr *src); 4487 void rdma_replace_ah_attr(struct rdma_ah_attr *old, 4488 const struct rdma_ah_attr *new); 4489 void rdma_move_ah_attr(struct rdma_ah_attr *dest, struct rdma_ah_attr *src); 4490 4491 /** 4492 * rdma_ah_find_type - Return address handle type. 4493 * 4494 * @dev: Device to be checked 4495 * @port_num: Port number 4496 */ 4497 static inline enum rdma_ah_attr_type rdma_ah_find_type(struct ib_device *dev, 4498 u8 port_num) 4499 { 4500 if (rdma_protocol_roce(dev, port_num)) 4501 return RDMA_AH_ATTR_TYPE_ROCE; 4502 if (rdma_protocol_ib(dev, port_num)) { 4503 if (rdma_cap_opa_ah(dev, port_num)) 4504 return RDMA_AH_ATTR_TYPE_OPA; 4505 return RDMA_AH_ATTR_TYPE_IB; 4506 } 4507 4508 return RDMA_AH_ATTR_TYPE_UNDEFINED; 4509 } 4510 4511 /** 4512 * ib_lid_cpu16 - Return lid in 16bit CPU encoding. 4513 * In the current implementation the only way to get 4514 * get the 32bit lid is from other sources for OPA. 4515 * For IB, lids will always be 16bits so cast the 4516 * value accordingly. 4517 * 4518 * @lid: A 32bit LID 4519 */ 4520 static inline u16 ib_lid_cpu16(u32 lid) 4521 { 4522 WARN_ON_ONCE(lid & 0xFFFF0000); 4523 return (u16)lid; 4524 } 4525 4526 /** 4527 * ib_lid_be16 - Return lid in 16bit BE encoding. 4528 * 4529 * @lid: A 32bit LID 4530 */ 4531 static inline __be16 ib_lid_be16(u32 lid) 4532 { 4533 WARN_ON_ONCE(lid & 0xFFFF0000); 4534 return cpu_to_be16((u16)lid); 4535 } 4536 4537 /** 4538 * ib_get_vector_affinity - Get the affinity mappings of a given completion 4539 * vector 4540 * @device: the rdma device 4541 * @comp_vector: index of completion vector 4542 * 4543 * Returns NULL on failure, otherwise a corresponding cpu map of the 4544 * completion vector (returns all-cpus map if the device driver doesn't 4545 * implement get_vector_affinity). 4546 */ 4547 static inline const struct cpumask * 4548 ib_get_vector_affinity(struct ib_device *device, int comp_vector) 4549 { 4550 if (comp_vector < 0 || comp_vector >= device->num_comp_vectors || 4551 !device->ops.get_vector_affinity) 4552 return NULL; 4553 4554 return device->ops.get_vector_affinity(device, comp_vector); 4555 4556 } 4557 4558 /** 4559 * rdma_roce_rescan_device - Rescan all of the network devices in the system 4560 * and add their gids, as needed, to the relevant RoCE devices. 4561 * 4562 * @device: the rdma device 4563 */ 4564 void rdma_roce_rescan_device(struct ib_device *ibdev); 4565 4566 struct ib_ucontext *ib_uverbs_get_ucontext_file(struct ib_uverbs_file *ufile); 4567 4568 int uverbs_destroy_def_handler(struct uverbs_attr_bundle *attrs); 4569 4570 struct net_device *rdma_alloc_netdev(struct ib_device *device, u8 port_num, 4571 enum rdma_netdev_t type, const char *name, 4572 unsigned char name_assign_type, 4573 void (*setup)(struct net_device *)); 4574 4575 int rdma_init_netdev(struct ib_device *device, u8 port_num, 4576 enum rdma_netdev_t type, const char *name, 4577 unsigned char name_assign_type, 4578 void (*setup)(struct net_device *), 4579 struct net_device *netdev); 4580 4581 /** 4582 * rdma_set_device_sysfs_group - Set device attributes group to have 4583 * driver specific sysfs entries at 4584 * for infiniband class. 4585 * 4586 * @device: device pointer for which attributes to be created 4587 * @group: Pointer to group which should be added when device 4588 * is registered with sysfs. 4589 * rdma_set_device_sysfs_group() allows existing drivers to expose one 4590 * group per device to have sysfs attributes. 4591 * 4592 * NOTE: New drivers should not make use of this API; instead new device 4593 * parameter should be exposed via netlink command. This API and mechanism 4594 * exist only for existing drivers. 4595 */ 4596 static inline void 4597 rdma_set_device_sysfs_group(struct ib_device *dev, 4598 const struct attribute_group *group) 4599 { 4600 dev->groups[1] = group; 4601 } 4602 4603 /** 4604 * rdma_device_to_ibdev - Get ib_device pointer from device pointer 4605 * 4606 * @device: device pointer for which ib_device pointer to retrieve 4607 * 4608 * rdma_device_to_ibdev() retrieves ib_device pointer from device. 4609 * 4610 */ 4611 static inline struct ib_device *rdma_device_to_ibdev(struct device *device) 4612 { 4613 struct ib_core_device *coredev = 4614 container_of(device, struct ib_core_device, dev); 4615 4616 return coredev->owner; 4617 } 4618 4619 /** 4620 * rdma_device_to_drv_device - Helper macro to reach back to driver's 4621 * ib_device holder structure from device pointer. 4622 * 4623 * NOTE: New drivers should not make use of this API; This API is only for 4624 * existing drivers who have exposed sysfs entries using 4625 * rdma_set_device_sysfs_group(). 4626 */ 4627 #define rdma_device_to_drv_device(dev, drv_dev_struct, ibdev_member) \ 4628 container_of(rdma_device_to_ibdev(dev), drv_dev_struct, ibdev_member) 4629 4630 bool rdma_dev_access_netns(const struct ib_device *device, 4631 const struct net *net); 4632 4633 #define IB_ROCE_UDP_ENCAP_VALID_PORT_MIN (0xC000) 4634 #define IB_ROCE_UDP_ENCAP_VALID_PORT_MAX (0xFFFF) 4635 #define IB_GRH_FLOWLABEL_MASK (0x000FFFFF) 4636 4637 /** 4638 * rdma_flow_label_to_udp_sport - generate a RoCE v2 UDP src port value based 4639 * on the flow_label 4640 * 4641 * This function will convert the 20 bit flow_label input to a valid RoCE v2 4642 * UDP src port 14 bit value. All RoCE V2 drivers should use this same 4643 * convention. 4644 */ 4645 static inline u16 rdma_flow_label_to_udp_sport(u32 fl) 4646 { 4647 u32 fl_low = fl & 0x03fff, fl_high = fl & 0xFC000; 4648 4649 fl_low ^= fl_high >> 14; 4650 return (u16)(fl_low | IB_ROCE_UDP_ENCAP_VALID_PORT_MIN); 4651 } 4652 4653 /** 4654 * rdma_calc_flow_label - generate a RDMA symmetric flow label value based on 4655 * local and remote qpn values 4656 * 4657 * This function folded the multiplication results of two qpns, 24 bit each, 4658 * fields, and converts it to a 20 bit results. 4659 * 4660 * This function will create symmetric flow_label value based on the local 4661 * and remote qpn values. this will allow both the requester and responder 4662 * to calculate the same flow_label for a given connection. 4663 * 4664 * This helper function should be used by driver in case the upper layer 4665 * provide a zero flow_label value. This is to improve entropy of RDMA 4666 * traffic in the network. 4667 */ 4668 static inline u32 rdma_calc_flow_label(u32 lqpn, u32 rqpn) 4669 { 4670 u64 v = (u64)lqpn * rqpn; 4671 4672 v ^= v >> 20; 4673 v ^= v >> 40; 4674 4675 return (u32)(v & IB_GRH_FLOWLABEL_MASK); 4676 } 4677 #endif /* IB_VERBS_H */ 4678