1 /* 2 * Copyright (c) 2004 Mellanox Technologies Ltd. All rights reserved. 3 * Copyright (c) 2004 Infinicon Corporation. All rights reserved. 4 * Copyright (c) 2004 Intel Corporation. All rights reserved. 5 * Copyright (c) 2004 Topspin Corporation. All rights reserved. 6 * Copyright (c) 2004 Voltaire Corporation. All rights reserved. 7 * Copyright (c) 2005 Sun Microsystems, Inc. All rights reserved. 8 * Copyright (c) 2005, 2006, 2007 Cisco Systems. All rights reserved. 9 * 10 * This software is available to you under a choice of one of two 11 * licenses. You may choose to be licensed under the terms of the GNU 12 * General Public License (GPL) Version 2, available from the file 13 * COPYING in the main directory of this source tree, or the 14 * OpenIB.org BSD license below: 15 * 16 * Redistribution and use in source and binary forms, with or 17 * without modification, are permitted provided that the following 18 * conditions are met: 19 * 20 * - Redistributions of source code must retain the above 21 * copyright notice, this list of conditions and the following 22 * disclaimer. 23 * 24 * - Redistributions in binary form must reproduce the above 25 * copyright notice, this list of conditions and the following 26 * disclaimer in the documentation and/or other materials 27 * provided with the distribution. 28 * 29 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, 30 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF 31 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND 32 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS 33 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN 34 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN 35 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE 36 * SOFTWARE. 37 */ 38 39 #if !defined(IB_VERBS_H) 40 #define IB_VERBS_H 41 42 #include <linux/types.h> 43 #include <linux/device.h> 44 #include <linux/dma-mapping.h> 45 #include <linux/kref.h> 46 #include <linux/list.h> 47 #include <linux/rwsem.h> 48 #include <linux/workqueue.h> 49 #include <linux/irq_poll.h> 50 #include <uapi/linux/if_ether.h> 51 #include <net/ipv6.h> 52 #include <net/ip.h> 53 #include <linux/string.h> 54 #include <linux/slab.h> 55 #include <linux/netdevice.h> 56 #include <linux/refcount.h> 57 #include <linux/if_link.h> 58 #include <linux/atomic.h> 59 #include <linux/mmu_notifier.h> 60 #include <linux/uaccess.h> 61 #include <linux/cgroup_rdma.h> 62 #include <linux/irqflags.h> 63 #include <linux/preempt.h> 64 #include <linux/dim.h> 65 #include <uapi/rdma/ib_user_verbs.h> 66 #include <rdma/rdma_counter.h> 67 #include <rdma/restrack.h> 68 #include <rdma/signature.h> 69 #include <uapi/rdma/rdma_user_ioctl.h> 70 #include <uapi/rdma/ib_user_ioctl_verbs.h> 71 72 #define IB_FW_VERSION_NAME_MAX ETHTOOL_FWVERS_LEN 73 74 struct ib_umem_odp; 75 struct ib_uqp_object; 76 struct ib_usrq_object; 77 struct ib_uwq_object; 78 79 extern struct workqueue_struct *ib_wq; 80 extern struct workqueue_struct *ib_comp_wq; 81 extern struct workqueue_struct *ib_comp_unbound_wq; 82 83 struct ib_ucq_object; 84 85 __printf(3, 4) __cold 86 void ibdev_printk(const char *level, const struct ib_device *ibdev, 87 const char *format, ...); 88 __printf(2, 3) __cold 89 void ibdev_emerg(const struct ib_device *ibdev, const char *format, ...); 90 __printf(2, 3) __cold 91 void ibdev_alert(const struct ib_device *ibdev, const char *format, ...); 92 __printf(2, 3) __cold 93 void ibdev_crit(const struct ib_device *ibdev, const char *format, ...); 94 __printf(2, 3) __cold 95 void ibdev_err(const struct ib_device *ibdev, const char *format, ...); 96 __printf(2, 3) __cold 97 void ibdev_warn(const struct ib_device *ibdev, const char *format, ...); 98 __printf(2, 3) __cold 99 void ibdev_notice(const struct ib_device *ibdev, const char *format, ...); 100 __printf(2, 3) __cold 101 void ibdev_info(const struct ib_device *ibdev, const char *format, ...); 102 103 #if defined(CONFIG_DYNAMIC_DEBUG) 104 #define ibdev_dbg(__dev, format, args...) \ 105 dynamic_ibdev_dbg(__dev, format, ##args) 106 #else 107 __printf(2, 3) __cold 108 static inline 109 void ibdev_dbg(const struct ib_device *ibdev, const char *format, ...) {} 110 #endif 111 112 #define ibdev_level_ratelimited(ibdev_level, ibdev, fmt, ...) \ 113 do { \ 114 static DEFINE_RATELIMIT_STATE(_rs, \ 115 DEFAULT_RATELIMIT_INTERVAL, \ 116 DEFAULT_RATELIMIT_BURST); \ 117 if (__ratelimit(&_rs)) \ 118 ibdev_level(ibdev, fmt, ##__VA_ARGS__); \ 119 } while (0) 120 121 #define ibdev_emerg_ratelimited(ibdev, fmt, ...) \ 122 ibdev_level_ratelimited(ibdev_emerg, ibdev, fmt, ##__VA_ARGS__) 123 #define ibdev_alert_ratelimited(ibdev, fmt, ...) \ 124 ibdev_level_ratelimited(ibdev_alert, ibdev, fmt, ##__VA_ARGS__) 125 #define ibdev_crit_ratelimited(ibdev, fmt, ...) \ 126 ibdev_level_ratelimited(ibdev_crit, ibdev, fmt, ##__VA_ARGS__) 127 #define ibdev_err_ratelimited(ibdev, fmt, ...) \ 128 ibdev_level_ratelimited(ibdev_err, ibdev, fmt, ##__VA_ARGS__) 129 #define ibdev_warn_ratelimited(ibdev, fmt, ...) \ 130 ibdev_level_ratelimited(ibdev_warn, ibdev, fmt, ##__VA_ARGS__) 131 #define ibdev_notice_ratelimited(ibdev, fmt, ...) \ 132 ibdev_level_ratelimited(ibdev_notice, ibdev, fmt, ##__VA_ARGS__) 133 #define ibdev_info_ratelimited(ibdev, fmt, ...) \ 134 ibdev_level_ratelimited(ibdev_info, ibdev, fmt, ##__VA_ARGS__) 135 136 #if defined(CONFIG_DYNAMIC_DEBUG) 137 /* descriptor check is first to prevent flooding with "callbacks suppressed" */ 138 #define ibdev_dbg_ratelimited(ibdev, fmt, ...) \ 139 do { \ 140 static DEFINE_RATELIMIT_STATE(_rs, \ 141 DEFAULT_RATELIMIT_INTERVAL, \ 142 DEFAULT_RATELIMIT_BURST); \ 143 DEFINE_DYNAMIC_DEBUG_METADATA(descriptor, fmt); \ 144 if (DYNAMIC_DEBUG_BRANCH(descriptor) && __ratelimit(&_rs)) \ 145 __dynamic_ibdev_dbg(&descriptor, ibdev, fmt, \ 146 ##__VA_ARGS__); \ 147 } while (0) 148 #else 149 __printf(2, 3) __cold 150 static inline 151 void ibdev_dbg_ratelimited(const struct ib_device *ibdev, const char *format, ...) {} 152 #endif 153 154 union ib_gid { 155 u8 raw[16]; 156 struct { 157 __be64 subnet_prefix; 158 __be64 interface_id; 159 } global; 160 }; 161 162 extern union ib_gid zgid; 163 164 enum ib_gid_type { 165 /* If link layer is Ethernet, this is RoCE V1 */ 166 IB_GID_TYPE_IB = 0, 167 IB_GID_TYPE_ROCE = 0, 168 IB_GID_TYPE_ROCE_UDP_ENCAP = 1, 169 IB_GID_TYPE_SIZE 170 }; 171 172 #define ROCE_V2_UDP_DPORT 4791 173 struct ib_gid_attr { 174 struct net_device __rcu *ndev; 175 struct ib_device *device; 176 union ib_gid gid; 177 enum ib_gid_type gid_type; 178 u16 index; 179 u8 port_num; 180 }; 181 182 enum { 183 /* set the local administered indication */ 184 IB_SA_WELL_KNOWN_GUID = BIT_ULL(57) | 2, 185 }; 186 187 enum rdma_transport_type { 188 RDMA_TRANSPORT_IB, 189 RDMA_TRANSPORT_IWARP, 190 RDMA_TRANSPORT_USNIC, 191 RDMA_TRANSPORT_USNIC_UDP, 192 RDMA_TRANSPORT_UNSPECIFIED, 193 }; 194 195 enum rdma_protocol_type { 196 RDMA_PROTOCOL_IB, 197 RDMA_PROTOCOL_IBOE, 198 RDMA_PROTOCOL_IWARP, 199 RDMA_PROTOCOL_USNIC_UDP 200 }; 201 202 __attribute_const__ enum rdma_transport_type 203 rdma_node_get_transport(unsigned int node_type); 204 205 enum rdma_network_type { 206 RDMA_NETWORK_IB, 207 RDMA_NETWORK_ROCE_V1 = RDMA_NETWORK_IB, 208 RDMA_NETWORK_IPV4, 209 RDMA_NETWORK_IPV6 210 }; 211 212 static inline enum ib_gid_type ib_network_to_gid_type(enum rdma_network_type network_type) 213 { 214 if (network_type == RDMA_NETWORK_IPV4 || 215 network_type == RDMA_NETWORK_IPV6) 216 return IB_GID_TYPE_ROCE_UDP_ENCAP; 217 218 /* IB_GID_TYPE_IB same as RDMA_NETWORK_ROCE_V1 */ 219 return IB_GID_TYPE_IB; 220 } 221 222 static inline enum rdma_network_type 223 rdma_gid_attr_network_type(const struct ib_gid_attr *attr) 224 { 225 if (attr->gid_type == IB_GID_TYPE_IB) 226 return RDMA_NETWORK_IB; 227 228 if (ipv6_addr_v4mapped((struct in6_addr *)&attr->gid)) 229 return RDMA_NETWORK_IPV4; 230 else 231 return RDMA_NETWORK_IPV6; 232 } 233 234 enum rdma_link_layer { 235 IB_LINK_LAYER_UNSPECIFIED, 236 IB_LINK_LAYER_INFINIBAND, 237 IB_LINK_LAYER_ETHERNET, 238 }; 239 240 enum ib_device_cap_flags { 241 IB_DEVICE_RESIZE_MAX_WR = (1 << 0), 242 IB_DEVICE_BAD_PKEY_CNTR = (1 << 1), 243 IB_DEVICE_BAD_QKEY_CNTR = (1 << 2), 244 IB_DEVICE_RAW_MULTI = (1 << 3), 245 IB_DEVICE_AUTO_PATH_MIG = (1 << 4), 246 IB_DEVICE_CHANGE_PHY_PORT = (1 << 5), 247 IB_DEVICE_UD_AV_PORT_ENFORCE = (1 << 6), 248 IB_DEVICE_CURR_QP_STATE_MOD = (1 << 7), 249 IB_DEVICE_SHUTDOWN_PORT = (1 << 8), 250 /* Not in use, former INIT_TYPE = (1 << 9),*/ 251 IB_DEVICE_PORT_ACTIVE_EVENT = (1 << 10), 252 IB_DEVICE_SYS_IMAGE_GUID = (1 << 11), 253 IB_DEVICE_RC_RNR_NAK_GEN = (1 << 12), 254 IB_DEVICE_SRQ_RESIZE = (1 << 13), 255 IB_DEVICE_N_NOTIFY_CQ = (1 << 14), 256 257 /* 258 * This device supports a per-device lkey or stag that can be 259 * used without performing a memory registration for the local 260 * memory. Note that ULPs should never check this flag, but 261 * instead of use the local_dma_lkey flag in the ib_pd structure, 262 * which will always contain a usable lkey. 263 */ 264 IB_DEVICE_LOCAL_DMA_LKEY = (1 << 15), 265 /* Reserved, old SEND_W_INV = (1 << 16),*/ 266 IB_DEVICE_MEM_WINDOW = (1 << 17), 267 /* 268 * Devices should set IB_DEVICE_UD_IP_SUM if they support 269 * insertion of UDP and TCP checksum on outgoing UD IPoIB 270 * messages and can verify the validity of checksum for 271 * incoming messages. Setting this flag implies that the 272 * IPoIB driver may set NETIF_F_IP_CSUM for datagram mode. 273 */ 274 IB_DEVICE_UD_IP_CSUM = (1 << 18), 275 IB_DEVICE_UD_TSO = (1 << 19), 276 IB_DEVICE_XRC = (1 << 20), 277 278 /* 279 * This device supports the IB "base memory management extension", 280 * which includes support for fast registrations (IB_WR_REG_MR, 281 * IB_WR_LOCAL_INV and IB_WR_SEND_WITH_INV verbs). This flag should 282 * also be set by any iWarp device which must support FRs to comply 283 * to the iWarp verbs spec. iWarp devices also support the 284 * IB_WR_RDMA_READ_WITH_INV verb for RDMA READs that invalidate the 285 * stag. 286 */ 287 IB_DEVICE_MEM_MGT_EXTENSIONS = (1 << 21), 288 IB_DEVICE_BLOCK_MULTICAST_LOOPBACK = (1 << 22), 289 IB_DEVICE_MEM_WINDOW_TYPE_2A = (1 << 23), 290 IB_DEVICE_MEM_WINDOW_TYPE_2B = (1 << 24), 291 IB_DEVICE_RC_IP_CSUM = (1 << 25), 292 /* Deprecated. Please use IB_RAW_PACKET_CAP_IP_CSUM. */ 293 IB_DEVICE_RAW_IP_CSUM = (1 << 26), 294 /* 295 * Devices should set IB_DEVICE_CROSS_CHANNEL if they 296 * support execution of WQEs that involve synchronization 297 * of I/O operations with single completion queue managed 298 * by hardware. 299 */ 300 IB_DEVICE_CROSS_CHANNEL = (1 << 27), 301 IB_DEVICE_MANAGED_FLOW_STEERING = (1 << 29), 302 IB_DEVICE_INTEGRITY_HANDOVER = (1 << 30), 303 IB_DEVICE_ON_DEMAND_PAGING = (1ULL << 31), 304 IB_DEVICE_SG_GAPS_REG = (1ULL << 32), 305 IB_DEVICE_VIRTUAL_FUNCTION = (1ULL << 33), 306 /* Deprecated. Please use IB_RAW_PACKET_CAP_SCATTER_FCS. */ 307 IB_DEVICE_RAW_SCATTER_FCS = (1ULL << 34), 308 IB_DEVICE_RDMA_NETDEV_OPA_VNIC = (1ULL << 35), 309 /* The device supports padding incoming writes to cacheline. */ 310 IB_DEVICE_PCI_WRITE_END_PADDING = (1ULL << 36), 311 IB_DEVICE_ALLOW_USER_UNREG = (1ULL << 37), 312 }; 313 314 enum ib_atomic_cap { 315 IB_ATOMIC_NONE, 316 IB_ATOMIC_HCA, 317 IB_ATOMIC_GLOB 318 }; 319 320 enum ib_odp_general_cap_bits { 321 IB_ODP_SUPPORT = 1 << 0, 322 IB_ODP_SUPPORT_IMPLICIT = 1 << 1, 323 }; 324 325 enum ib_odp_transport_cap_bits { 326 IB_ODP_SUPPORT_SEND = 1 << 0, 327 IB_ODP_SUPPORT_RECV = 1 << 1, 328 IB_ODP_SUPPORT_WRITE = 1 << 2, 329 IB_ODP_SUPPORT_READ = 1 << 3, 330 IB_ODP_SUPPORT_ATOMIC = 1 << 4, 331 IB_ODP_SUPPORT_SRQ_RECV = 1 << 5, 332 }; 333 334 struct ib_odp_caps { 335 uint64_t general_caps; 336 struct { 337 uint32_t rc_odp_caps; 338 uint32_t uc_odp_caps; 339 uint32_t ud_odp_caps; 340 uint32_t xrc_odp_caps; 341 } per_transport_caps; 342 }; 343 344 struct ib_rss_caps { 345 /* Corresponding bit will be set if qp type from 346 * 'enum ib_qp_type' is supported, e.g. 347 * supported_qpts |= 1 << IB_QPT_UD 348 */ 349 u32 supported_qpts; 350 u32 max_rwq_indirection_tables; 351 u32 max_rwq_indirection_table_size; 352 }; 353 354 enum ib_tm_cap_flags { 355 /* Support tag matching with rendezvous offload for RC transport */ 356 IB_TM_CAP_RNDV_RC = 1 << 0, 357 }; 358 359 struct ib_tm_caps { 360 /* Max size of RNDV header */ 361 u32 max_rndv_hdr_size; 362 /* Max number of entries in tag matching list */ 363 u32 max_num_tags; 364 /* From enum ib_tm_cap_flags */ 365 u32 flags; 366 /* Max number of outstanding list operations */ 367 u32 max_ops; 368 /* Max number of SGE in tag matching entry */ 369 u32 max_sge; 370 }; 371 372 struct ib_cq_init_attr { 373 unsigned int cqe; 374 u32 comp_vector; 375 u32 flags; 376 }; 377 378 enum ib_cq_attr_mask { 379 IB_CQ_MODERATE = 1 << 0, 380 }; 381 382 struct ib_cq_caps { 383 u16 max_cq_moderation_count; 384 u16 max_cq_moderation_period; 385 }; 386 387 struct ib_dm_mr_attr { 388 u64 length; 389 u64 offset; 390 u32 access_flags; 391 }; 392 393 struct ib_dm_alloc_attr { 394 u64 length; 395 u32 alignment; 396 u32 flags; 397 }; 398 399 struct ib_device_attr { 400 u64 fw_ver; 401 __be64 sys_image_guid; 402 u64 max_mr_size; 403 u64 page_size_cap; 404 u32 vendor_id; 405 u32 vendor_part_id; 406 u32 hw_ver; 407 int max_qp; 408 int max_qp_wr; 409 u64 device_cap_flags; 410 int max_send_sge; 411 int max_recv_sge; 412 int max_sge_rd; 413 int max_cq; 414 int max_cqe; 415 int max_mr; 416 int max_pd; 417 int max_qp_rd_atom; 418 int max_ee_rd_atom; 419 int max_res_rd_atom; 420 int max_qp_init_rd_atom; 421 int max_ee_init_rd_atom; 422 enum ib_atomic_cap atomic_cap; 423 enum ib_atomic_cap masked_atomic_cap; 424 int max_ee; 425 int max_rdd; 426 int max_mw; 427 int max_raw_ipv6_qp; 428 int max_raw_ethy_qp; 429 int max_mcast_grp; 430 int max_mcast_qp_attach; 431 int max_total_mcast_qp_attach; 432 int max_ah; 433 int max_fmr; 434 int max_map_per_fmr; 435 int max_srq; 436 int max_srq_wr; 437 int max_srq_sge; 438 unsigned int max_fast_reg_page_list_len; 439 unsigned int max_pi_fast_reg_page_list_len; 440 u16 max_pkeys; 441 u8 local_ca_ack_delay; 442 int sig_prot_cap; 443 int sig_guard_cap; 444 struct ib_odp_caps odp_caps; 445 uint64_t timestamp_mask; 446 uint64_t hca_core_clock; /* in KHZ */ 447 struct ib_rss_caps rss_caps; 448 u32 max_wq_type_rq; 449 u32 raw_packet_caps; /* Use ib_raw_packet_caps enum */ 450 struct ib_tm_caps tm_caps; 451 struct ib_cq_caps cq_caps; 452 u64 max_dm_size; 453 /* Max entries for sgl for optimized performance per READ */ 454 u32 max_sgl_rd; 455 }; 456 457 enum ib_mtu { 458 IB_MTU_256 = 1, 459 IB_MTU_512 = 2, 460 IB_MTU_1024 = 3, 461 IB_MTU_2048 = 4, 462 IB_MTU_4096 = 5 463 }; 464 465 static inline int ib_mtu_enum_to_int(enum ib_mtu mtu) 466 { 467 switch (mtu) { 468 case IB_MTU_256: return 256; 469 case IB_MTU_512: return 512; 470 case IB_MTU_1024: return 1024; 471 case IB_MTU_2048: return 2048; 472 case IB_MTU_4096: return 4096; 473 default: return -1; 474 } 475 } 476 477 static inline enum ib_mtu ib_mtu_int_to_enum(int mtu) 478 { 479 if (mtu >= 4096) 480 return IB_MTU_4096; 481 else if (mtu >= 2048) 482 return IB_MTU_2048; 483 else if (mtu >= 1024) 484 return IB_MTU_1024; 485 else if (mtu >= 512) 486 return IB_MTU_512; 487 else 488 return IB_MTU_256; 489 } 490 491 enum ib_port_state { 492 IB_PORT_NOP = 0, 493 IB_PORT_DOWN = 1, 494 IB_PORT_INIT = 2, 495 IB_PORT_ARMED = 3, 496 IB_PORT_ACTIVE = 4, 497 IB_PORT_ACTIVE_DEFER = 5 498 }; 499 500 enum ib_port_phys_state { 501 IB_PORT_PHYS_STATE_SLEEP = 1, 502 IB_PORT_PHYS_STATE_POLLING = 2, 503 IB_PORT_PHYS_STATE_DISABLED = 3, 504 IB_PORT_PHYS_STATE_PORT_CONFIGURATION_TRAINING = 4, 505 IB_PORT_PHYS_STATE_LINK_UP = 5, 506 IB_PORT_PHYS_STATE_LINK_ERROR_RECOVERY = 6, 507 IB_PORT_PHYS_STATE_PHY_TEST = 7, 508 }; 509 510 enum ib_port_width { 511 IB_WIDTH_1X = 1, 512 IB_WIDTH_2X = 16, 513 IB_WIDTH_4X = 2, 514 IB_WIDTH_8X = 4, 515 IB_WIDTH_12X = 8 516 }; 517 518 static inline int ib_width_enum_to_int(enum ib_port_width width) 519 { 520 switch (width) { 521 case IB_WIDTH_1X: return 1; 522 case IB_WIDTH_2X: return 2; 523 case IB_WIDTH_4X: return 4; 524 case IB_WIDTH_8X: return 8; 525 case IB_WIDTH_12X: return 12; 526 default: return -1; 527 } 528 } 529 530 enum ib_port_speed { 531 IB_SPEED_SDR = 1, 532 IB_SPEED_DDR = 2, 533 IB_SPEED_QDR = 4, 534 IB_SPEED_FDR10 = 8, 535 IB_SPEED_FDR = 16, 536 IB_SPEED_EDR = 32, 537 IB_SPEED_HDR = 64 538 }; 539 540 /** 541 * struct rdma_hw_stats 542 * @lock - Mutex to protect parallel write access to lifespan and values 543 * of counters, which are 64bits and not guaranteeed to be written 544 * atomicaly on 32bits systems. 545 * @timestamp - Used by the core code to track when the last update was 546 * @lifespan - Used by the core code to determine how old the counters 547 * should be before being updated again. Stored in jiffies, defaults 548 * to 10 milliseconds, drivers can override the default be specifying 549 * their own value during their allocation routine. 550 * @name - Array of pointers to static names used for the counters in 551 * directory. 552 * @num_counters - How many hardware counters there are. If name is 553 * shorter than this number, a kernel oops will result. Driver authors 554 * are encouraged to leave BUILD_BUG_ON(ARRAY_SIZE(@name) < num_counters) 555 * in their code to prevent this. 556 * @value - Array of u64 counters that are accessed by the sysfs code and 557 * filled in by the drivers get_stats routine 558 */ 559 struct rdma_hw_stats { 560 struct mutex lock; /* Protect lifespan and values[] */ 561 unsigned long timestamp; 562 unsigned long lifespan; 563 const char * const *names; 564 int num_counters; 565 u64 value[]; 566 }; 567 568 #define RDMA_HW_STATS_DEFAULT_LIFESPAN 10 569 /** 570 * rdma_alloc_hw_stats_struct - Helper function to allocate dynamic struct 571 * for drivers. 572 * @names - Array of static const char * 573 * @num_counters - How many elements in array 574 * @lifespan - How many milliseconds between updates 575 */ 576 static inline struct rdma_hw_stats *rdma_alloc_hw_stats_struct( 577 const char * const *names, int num_counters, 578 unsigned long lifespan) 579 { 580 struct rdma_hw_stats *stats; 581 582 stats = kzalloc(sizeof(*stats) + num_counters * sizeof(u64), 583 GFP_KERNEL); 584 if (!stats) 585 return NULL; 586 stats->names = names; 587 stats->num_counters = num_counters; 588 stats->lifespan = msecs_to_jiffies(lifespan); 589 590 return stats; 591 } 592 593 594 /* Define bits for the various functionality this port needs to be supported by 595 * the core. 596 */ 597 /* Management 0x00000FFF */ 598 #define RDMA_CORE_CAP_IB_MAD 0x00000001 599 #define RDMA_CORE_CAP_IB_SMI 0x00000002 600 #define RDMA_CORE_CAP_IB_CM 0x00000004 601 #define RDMA_CORE_CAP_IW_CM 0x00000008 602 #define RDMA_CORE_CAP_IB_SA 0x00000010 603 #define RDMA_CORE_CAP_OPA_MAD 0x00000020 604 605 /* Address format 0x000FF000 */ 606 #define RDMA_CORE_CAP_AF_IB 0x00001000 607 #define RDMA_CORE_CAP_ETH_AH 0x00002000 608 #define RDMA_CORE_CAP_OPA_AH 0x00004000 609 #define RDMA_CORE_CAP_IB_GRH_REQUIRED 0x00008000 610 611 /* Protocol 0xFFF00000 */ 612 #define RDMA_CORE_CAP_PROT_IB 0x00100000 613 #define RDMA_CORE_CAP_PROT_ROCE 0x00200000 614 #define RDMA_CORE_CAP_PROT_IWARP 0x00400000 615 #define RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP 0x00800000 616 #define RDMA_CORE_CAP_PROT_RAW_PACKET 0x01000000 617 #define RDMA_CORE_CAP_PROT_USNIC 0x02000000 618 619 #define RDMA_CORE_PORT_IB_GRH_REQUIRED (RDMA_CORE_CAP_IB_GRH_REQUIRED \ 620 | RDMA_CORE_CAP_PROT_ROCE \ 621 | RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP) 622 623 #define RDMA_CORE_PORT_IBA_IB (RDMA_CORE_CAP_PROT_IB \ 624 | RDMA_CORE_CAP_IB_MAD \ 625 | RDMA_CORE_CAP_IB_SMI \ 626 | RDMA_CORE_CAP_IB_CM \ 627 | RDMA_CORE_CAP_IB_SA \ 628 | RDMA_CORE_CAP_AF_IB) 629 #define RDMA_CORE_PORT_IBA_ROCE (RDMA_CORE_CAP_PROT_ROCE \ 630 | RDMA_CORE_CAP_IB_MAD \ 631 | RDMA_CORE_CAP_IB_CM \ 632 | RDMA_CORE_CAP_AF_IB \ 633 | RDMA_CORE_CAP_ETH_AH) 634 #define RDMA_CORE_PORT_IBA_ROCE_UDP_ENCAP \ 635 (RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP \ 636 | RDMA_CORE_CAP_IB_MAD \ 637 | RDMA_CORE_CAP_IB_CM \ 638 | RDMA_CORE_CAP_AF_IB \ 639 | RDMA_CORE_CAP_ETH_AH) 640 #define RDMA_CORE_PORT_IWARP (RDMA_CORE_CAP_PROT_IWARP \ 641 | RDMA_CORE_CAP_IW_CM) 642 #define RDMA_CORE_PORT_INTEL_OPA (RDMA_CORE_PORT_IBA_IB \ 643 | RDMA_CORE_CAP_OPA_MAD) 644 645 #define RDMA_CORE_PORT_RAW_PACKET (RDMA_CORE_CAP_PROT_RAW_PACKET) 646 647 #define RDMA_CORE_PORT_USNIC (RDMA_CORE_CAP_PROT_USNIC) 648 649 struct ib_port_attr { 650 u64 subnet_prefix; 651 enum ib_port_state state; 652 enum ib_mtu max_mtu; 653 enum ib_mtu active_mtu; 654 int gid_tbl_len; 655 unsigned int ip_gids:1; 656 /* This is the value from PortInfo CapabilityMask, defined by IBA */ 657 u32 port_cap_flags; 658 u32 max_msg_sz; 659 u32 bad_pkey_cntr; 660 u32 qkey_viol_cntr; 661 u16 pkey_tbl_len; 662 u32 sm_lid; 663 u32 lid; 664 u8 lmc; 665 u8 max_vl_num; 666 u8 sm_sl; 667 u8 subnet_timeout; 668 u8 init_type_reply; 669 u8 active_width; 670 u8 active_speed; 671 u8 phys_state; 672 u16 port_cap_flags2; 673 }; 674 675 enum ib_device_modify_flags { 676 IB_DEVICE_MODIFY_SYS_IMAGE_GUID = 1 << 0, 677 IB_DEVICE_MODIFY_NODE_DESC = 1 << 1 678 }; 679 680 #define IB_DEVICE_NODE_DESC_MAX 64 681 682 struct ib_device_modify { 683 u64 sys_image_guid; 684 char node_desc[IB_DEVICE_NODE_DESC_MAX]; 685 }; 686 687 enum ib_port_modify_flags { 688 IB_PORT_SHUTDOWN = 1, 689 IB_PORT_INIT_TYPE = (1<<2), 690 IB_PORT_RESET_QKEY_CNTR = (1<<3), 691 IB_PORT_OPA_MASK_CHG = (1<<4) 692 }; 693 694 struct ib_port_modify { 695 u32 set_port_cap_mask; 696 u32 clr_port_cap_mask; 697 u8 init_type; 698 }; 699 700 enum ib_event_type { 701 IB_EVENT_CQ_ERR, 702 IB_EVENT_QP_FATAL, 703 IB_EVENT_QP_REQ_ERR, 704 IB_EVENT_QP_ACCESS_ERR, 705 IB_EVENT_COMM_EST, 706 IB_EVENT_SQ_DRAINED, 707 IB_EVENT_PATH_MIG, 708 IB_EVENT_PATH_MIG_ERR, 709 IB_EVENT_DEVICE_FATAL, 710 IB_EVENT_PORT_ACTIVE, 711 IB_EVENT_PORT_ERR, 712 IB_EVENT_LID_CHANGE, 713 IB_EVENT_PKEY_CHANGE, 714 IB_EVENT_SM_CHANGE, 715 IB_EVENT_SRQ_ERR, 716 IB_EVENT_SRQ_LIMIT_REACHED, 717 IB_EVENT_QP_LAST_WQE_REACHED, 718 IB_EVENT_CLIENT_REREGISTER, 719 IB_EVENT_GID_CHANGE, 720 IB_EVENT_WQ_FATAL, 721 }; 722 723 const char *__attribute_const__ ib_event_msg(enum ib_event_type event); 724 725 struct ib_event { 726 struct ib_device *device; 727 union { 728 struct ib_cq *cq; 729 struct ib_qp *qp; 730 struct ib_srq *srq; 731 struct ib_wq *wq; 732 u8 port_num; 733 } element; 734 enum ib_event_type event; 735 }; 736 737 struct ib_event_handler { 738 struct ib_device *device; 739 void (*handler)(struct ib_event_handler *, struct ib_event *); 740 struct list_head list; 741 }; 742 743 #define INIT_IB_EVENT_HANDLER(_ptr, _device, _handler) \ 744 do { \ 745 (_ptr)->device = _device; \ 746 (_ptr)->handler = _handler; \ 747 INIT_LIST_HEAD(&(_ptr)->list); \ 748 } while (0) 749 750 struct ib_global_route { 751 const struct ib_gid_attr *sgid_attr; 752 union ib_gid dgid; 753 u32 flow_label; 754 u8 sgid_index; 755 u8 hop_limit; 756 u8 traffic_class; 757 }; 758 759 struct ib_grh { 760 __be32 version_tclass_flow; 761 __be16 paylen; 762 u8 next_hdr; 763 u8 hop_limit; 764 union ib_gid sgid; 765 union ib_gid dgid; 766 }; 767 768 union rdma_network_hdr { 769 struct ib_grh ibgrh; 770 struct { 771 /* The IB spec states that if it's IPv4, the header 772 * is located in the last 20 bytes of the header. 773 */ 774 u8 reserved[20]; 775 struct iphdr roce4grh; 776 }; 777 }; 778 779 #define IB_QPN_MASK 0xFFFFFF 780 781 enum { 782 IB_MULTICAST_QPN = 0xffffff 783 }; 784 785 #define IB_LID_PERMISSIVE cpu_to_be16(0xFFFF) 786 #define IB_MULTICAST_LID_BASE cpu_to_be16(0xC000) 787 788 enum ib_ah_flags { 789 IB_AH_GRH = 1 790 }; 791 792 enum ib_rate { 793 IB_RATE_PORT_CURRENT = 0, 794 IB_RATE_2_5_GBPS = 2, 795 IB_RATE_5_GBPS = 5, 796 IB_RATE_10_GBPS = 3, 797 IB_RATE_20_GBPS = 6, 798 IB_RATE_30_GBPS = 4, 799 IB_RATE_40_GBPS = 7, 800 IB_RATE_60_GBPS = 8, 801 IB_RATE_80_GBPS = 9, 802 IB_RATE_120_GBPS = 10, 803 IB_RATE_14_GBPS = 11, 804 IB_RATE_56_GBPS = 12, 805 IB_RATE_112_GBPS = 13, 806 IB_RATE_168_GBPS = 14, 807 IB_RATE_25_GBPS = 15, 808 IB_RATE_100_GBPS = 16, 809 IB_RATE_200_GBPS = 17, 810 IB_RATE_300_GBPS = 18, 811 IB_RATE_28_GBPS = 19, 812 IB_RATE_50_GBPS = 20, 813 IB_RATE_400_GBPS = 21, 814 IB_RATE_600_GBPS = 22, 815 }; 816 817 /** 818 * ib_rate_to_mult - Convert the IB rate enum to a multiple of the 819 * base rate of 2.5 Gbit/sec. For example, IB_RATE_5_GBPS will be 820 * converted to 2, since 5 Gbit/sec is 2 * 2.5 Gbit/sec. 821 * @rate: rate to convert. 822 */ 823 __attribute_const__ int ib_rate_to_mult(enum ib_rate rate); 824 825 /** 826 * ib_rate_to_mbps - Convert the IB rate enum to Mbps. 827 * For example, IB_RATE_2_5_GBPS will be converted to 2500. 828 * @rate: rate to convert. 829 */ 830 __attribute_const__ int ib_rate_to_mbps(enum ib_rate rate); 831 832 833 /** 834 * enum ib_mr_type - memory region type 835 * @IB_MR_TYPE_MEM_REG: memory region that is used for 836 * normal registration 837 * @IB_MR_TYPE_SG_GAPS: memory region that is capable to 838 * register any arbitrary sg lists (without 839 * the normal mr constraints - see 840 * ib_map_mr_sg) 841 * @IB_MR_TYPE_DM: memory region that is used for device 842 * memory registration 843 * @IB_MR_TYPE_USER: memory region that is used for the user-space 844 * application 845 * @IB_MR_TYPE_DMA: memory region that is used for DMA operations 846 * without address translations (VA=PA) 847 * @IB_MR_TYPE_INTEGRITY: memory region that is used for 848 * data integrity operations 849 */ 850 enum ib_mr_type { 851 IB_MR_TYPE_MEM_REG, 852 IB_MR_TYPE_SG_GAPS, 853 IB_MR_TYPE_DM, 854 IB_MR_TYPE_USER, 855 IB_MR_TYPE_DMA, 856 IB_MR_TYPE_INTEGRITY, 857 }; 858 859 enum ib_mr_status_check { 860 IB_MR_CHECK_SIG_STATUS = 1, 861 }; 862 863 /** 864 * struct ib_mr_status - Memory region status container 865 * 866 * @fail_status: Bitmask of MR checks status. For each 867 * failed check a corresponding status bit is set. 868 * @sig_err: Additional info for IB_MR_CEHCK_SIG_STATUS 869 * failure. 870 */ 871 struct ib_mr_status { 872 u32 fail_status; 873 struct ib_sig_err sig_err; 874 }; 875 876 /** 877 * mult_to_ib_rate - Convert a multiple of 2.5 Gbit/sec to an IB rate 878 * enum. 879 * @mult: multiple to convert. 880 */ 881 __attribute_const__ enum ib_rate mult_to_ib_rate(int mult); 882 883 enum rdma_ah_attr_type { 884 RDMA_AH_ATTR_TYPE_UNDEFINED, 885 RDMA_AH_ATTR_TYPE_IB, 886 RDMA_AH_ATTR_TYPE_ROCE, 887 RDMA_AH_ATTR_TYPE_OPA, 888 }; 889 890 struct ib_ah_attr { 891 u16 dlid; 892 u8 src_path_bits; 893 }; 894 895 struct roce_ah_attr { 896 u8 dmac[ETH_ALEN]; 897 }; 898 899 struct opa_ah_attr { 900 u32 dlid; 901 u8 src_path_bits; 902 bool make_grd; 903 }; 904 905 struct rdma_ah_attr { 906 struct ib_global_route grh; 907 u8 sl; 908 u8 static_rate; 909 u8 port_num; 910 u8 ah_flags; 911 enum rdma_ah_attr_type type; 912 union { 913 struct ib_ah_attr ib; 914 struct roce_ah_attr roce; 915 struct opa_ah_attr opa; 916 }; 917 }; 918 919 enum ib_wc_status { 920 IB_WC_SUCCESS, 921 IB_WC_LOC_LEN_ERR, 922 IB_WC_LOC_QP_OP_ERR, 923 IB_WC_LOC_EEC_OP_ERR, 924 IB_WC_LOC_PROT_ERR, 925 IB_WC_WR_FLUSH_ERR, 926 IB_WC_MW_BIND_ERR, 927 IB_WC_BAD_RESP_ERR, 928 IB_WC_LOC_ACCESS_ERR, 929 IB_WC_REM_INV_REQ_ERR, 930 IB_WC_REM_ACCESS_ERR, 931 IB_WC_REM_OP_ERR, 932 IB_WC_RETRY_EXC_ERR, 933 IB_WC_RNR_RETRY_EXC_ERR, 934 IB_WC_LOC_RDD_VIOL_ERR, 935 IB_WC_REM_INV_RD_REQ_ERR, 936 IB_WC_REM_ABORT_ERR, 937 IB_WC_INV_EECN_ERR, 938 IB_WC_INV_EEC_STATE_ERR, 939 IB_WC_FATAL_ERR, 940 IB_WC_RESP_TIMEOUT_ERR, 941 IB_WC_GENERAL_ERR 942 }; 943 944 const char *__attribute_const__ ib_wc_status_msg(enum ib_wc_status status); 945 946 enum ib_wc_opcode { 947 IB_WC_SEND, 948 IB_WC_RDMA_WRITE, 949 IB_WC_RDMA_READ, 950 IB_WC_COMP_SWAP, 951 IB_WC_FETCH_ADD, 952 IB_WC_LSO, 953 IB_WC_LOCAL_INV, 954 IB_WC_REG_MR, 955 IB_WC_MASKED_COMP_SWAP, 956 IB_WC_MASKED_FETCH_ADD, 957 /* 958 * Set value of IB_WC_RECV so consumers can test if a completion is a 959 * receive by testing (opcode & IB_WC_RECV). 960 */ 961 IB_WC_RECV = 1 << 7, 962 IB_WC_RECV_RDMA_WITH_IMM 963 }; 964 965 enum ib_wc_flags { 966 IB_WC_GRH = 1, 967 IB_WC_WITH_IMM = (1<<1), 968 IB_WC_WITH_INVALIDATE = (1<<2), 969 IB_WC_IP_CSUM_OK = (1<<3), 970 IB_WC_WITH_SMAC = (1<<4), 971 IB_WC_WITH_VLAN = (1<<5), 972 IB_WC_WITH_NETWORK_HDR_TYPE = (1<<6), 973 }; 974 975 struct ib_wc { 976 union { 977 u64 wr_id; 978 struct ib_cqe *wr_cqe; 979 }; 980 enum ib_wc_status status; 981 enum ib_wc_opcode opcode; 982 u32 vendor_err; 983 u32 byte_len; 984 struct ib_qp *qp; 985 union { 986 __be32 imm_data; 987 u32 invalidate_rkey; 988 } ex; 989 u32 src_qp; 990 u32 slid; 991 int wc_flags; 992 u16 pkey_index; 993 u8 sl; 994 u8 dlid_path_bits; 995 u8 port_num; /* valid only for DR SMPs on switches */ 996 u8 smac[ETH_ALEN]; 997 u16 vlan_id; 998 u8 network_hdr_type; 999 }; 1000 1001 enum ib_cq_notify_flags { 1002 IB_CQ_SOLICITED = 1 << 0, 1003 IB_CQ_NEXT_COMP = 1 << 1, 1004 IB_CQ_SOLICITED_MASK = IB_CQ_SOLICITED | IB_CQ_NEXT_COMP, 1005 IB_CQ_REPORT_MISSED_EVENTS = 1 << 2, 1006 }; 1007 1008 enum ib_srq_type { 1009 IB_SRQT_BASIC, 1010 IB_SRQT_XRC, 1011 IB_SRQT_TM, 1012 }; 1013 1014 static inline bool ib_srq_has_cq(enum ib_srq_type srq_type) 1015 { 1016 return srq_type == IB_SRQT_XRC || 1017 srq_type == IB_SRQT_TM; 1018 } 1019 1020 enum ib_srq_attr_mask { 1021 IB_SRQ_MAX_WR = 1 << 0, 1022 IB_SRQ_LIMIT = 1 << 1, 1023 }; 1024 1025 struct ib_srq_attr { 1026 u32 max_wr; 1027 u32 max_sge; 1028 u32 srq_limit; 1029 }; 1030 1031 struct ib_srq_init_attr { 1032 void (*event_handler)(struct ib_event *, void *); 1033 void *srq_context; 1034 struct ib_srq_attr attr; 1035 enum ib_srq_type srq_type; 1036 1037 struct { 1038 struct ib_cq *cq; 1039 union { 1040 struct { 1041 struct ib_xrcd *xrcd; 1042 } xrc; 1043 1044 struct { 1045 u32 max_num_tags; 1046 } tag_matching; 1047 }; 1048 } ext; 1049 }; 1050 1051 struct ib_qp_cap { 1052 u32 max_send_wr; 1053 u32 max_recv_wr; 1054 u32 max_send_sge; 1055 u32 max_recv_sge; 1056 u32 max_inline_data; 1057 1058 /* 1059 * Maximum number of rdma_rw_ctx structures in flight at a time. 1060 * ib_create_qp() will calculate the right amount of neededed WRs 1061 * and MRs based on this. 1062 */ 1063 u32 max_rdma_ctxs; 1064 }; 1065 1066 enum ib_sig_type { 1067 IB_SIGNAL_ALL_WR, 1068 IB_SIGNAL_REQ_WR 1069 }; 1070 1071 enum ib_qp_type { 1072 /* 1073 * IB_QPT_SMI and IB_QPT_GSI have to be the first two entries 1074 * here (and in that order) since the MAD layer uses them as 1075 * indices into a 2-entry table. 1076 */ 1077 IB_QPT_SMI, 1078 IB_QPT_GSI, 1079 1080 IB_QPT_RC, 1081 IB_QPT_UC, 1082 IB_QPT_UD, 1083 IB_QPT_RAW_IPV6, 1084 IB_QPT_RAW_ETHERTYPE, 1085 IB_QPT_RAW_PACKET = 8, 1086 IB_QPT_XRC_INI = 9, 1087 IB_QPT_XRC_TGT, 1088 IB_QPT_MAX, 1089 IB_QPT_DRIVER = 0xFF, 1090 /* Reserve a range for qp types internal to the low level driver. 1091 * These qp types will not be visible at the IB core layer, so the 1092 * IB_QPT_MAX usages should not be affected in the core layer 1093 */ 1094 IB_QPT_RESERVED1 = 0x1000, 1095 IB_QPT_RESERVED2, 1096 IB_QPT_RESERVED3, 1097 IB_QPT_RESERVED4, 1098 IB_QPT_RESERVED5, 1099 IB_QPT_RESERVED6, 1100 IB_QPT_RESERVED7, 1101 IB_QPT_RESERVED8, 1102 IB_QPT_RESERVED9, 1103 IB_QPT_RESERVED10, 1104 }; 1105 1106 enum ib_qp_create_flags { 1107 IB_QP_CREATE_IPOIB_UD_LSO = 1 << 0, 1108 IB_QP_CREATE_BLOCK_MULTICAST_LOOPBACK = 1 << 1, 1109 IB_QP_CREATE_CROSS_CHANNEL = 1 << 2, 1110 IB_QP_CREATE_MANAGED_SEND = 1 << 3, 1111 IB_QP_CREATE_MANAGED_RECV = 1 << 4, 1112 IB_QP_CREATE_NETIF_QP = 1 << 5, 1113 IB_QP_CREATE_INTEGRITY_EN = 1 << 6, 1114 /* FREE = 1 << 7, */ 1115 IB_QP_CREATE_SCATTER_FCS = 1 << 8, 1116 IB_QP_CREATE_CVLAN_STRIPPING = 1 << 9, 1117 IB_QP_CREATE_SOURCE_QPN = 1 << 10, 1118 IB_QP_CREATE_PCI_WRITE_END_PADDING = 1 << 11, 1119 /* reserve bits 26-31 for low level drivers' internal use */ 1120 IB_QP_CREATE_RESERVED_START = 1 << 26, 1121 IB_QP_CREATE_RESERVED_END = 1 << 31, 1122 }; 1123 1124 /* 1125 * Note: users may not call ib_close_qp or ib_destroy_qp from the event_handler 1126 * callback to destroy the passed in QP. 1127 */ 1128 1129 struct ib_qp_init_attr { 1130 /* Consumer's event_handler callback must not block */ 1131 void (*event_handler)(struct ib_event *, void *); 1132 1133 void *qp_context; 1134 struct ib_cq *send_cq; 1135 struct ib_cq *recv_cq; 1136 struct ib_srq *srq; 1137 struct ib_xrcd *xrcd; /* XRC TGT QPs only */ 1138 struct ib_qp_cap cap; 1139 enum ib_sig_type sq_sig_type; 1140 enum ib_qp_type qp_type; 1141 u32 create_flags; 1142 1143 /* 1144 * Only needed for special QP types, or when using the RW API. 1145 */ 1146 u8 port_num; 1147 struct ib_rwq_ind_table *rwq_ind_tbl; 1148 u32 source_qpn; 1149 }; 1150 1151 struct ib_qp_open_attr { 1152 void (*event_handler)(struct ib_event *, void *); 1153 void *qp_context; 1154 u32 qp_num; 1155 enum ib_qp_type qp_type; 1156 }; 1157 1158 enum ib_rnr_timeout { 1159 IB_RNR_TIMER_655_36 = 0, 1160 IB_RNR_TIMER_000_01 = 1, 1161 IB_RNR_TIMER_000_02 = 2, 1162 IB_RNR_TIMER_000_03 = 3, 1163 IB_RNR_TIMER_000_04 = 4, 1164 IB_RNR_TIMER_000_06 = 5, 1165 IB_RNR_TIMER_000_08 = 6, 1166 IB_RNR_TIMER_000_12 = 7, 1167 IB_RNR_TIMER_000_16 = 8, 1168 IB_RNR_TIMER_000_24 = 9, 1169 IB_RNR_TIMER_000_32 = 10, 1170 IB_RNR_TIMER_000_48 = 11, 1171 IB_RNR_TIMER_000_64 = 12, 1172 IB_RNR_TIMER_000_96 = 13, 1173 IB_RNR_TIMER_001_28 = 14, 1174 IB_RNR_TIMER_001_92 = 15, 1175 IB_RNR_TIMER_002_56 = 16, 1176 IB_RNR_TIMER_003_84 = 17, 1177 IB_RNR_TIMER_005_12 = 18, 1178 IB_RNR_TIMER_007_68 = 19, 1179 IB_RNR_TIMER_010_24 = 20, 1180 IB_RNR_TIMER_015_36 = 21, 1181 IB_RNR_TIMER_020_48 = 22, 1182 IB_RNR_TIMER_030_72 = 23, 1183 IB_RNR_TIMER_040_96 = 24, 1184 IB_RNR_TIMER_061_44 = 25, 1185 IB_RNR_TIMER_081_92 = 26, 1186 IB_RNR_TIMER_122_88 = 27, 1187 IB_RNR_TIMER_163_84 = 28, 1188 IB_RNR_TIMER_245_76 = 29, 1189 IB_RNR_TIMER_327_68 = 30, 1190 IB_RNR_TIMER_491_52 = 31 1191 }; 1192 1193 enum ib_qp_attr_mask { 1194 IB_QP_STATE = 1, 1195 IB_QP_CUR_STATE = (1<<1), 1196 IB_QP_EN_SQD_ASYNC_NOTIFY = (1<<2), 1197 IB_QP_ACCESS_FLAGS = (1<<3), 1198 IB_QP_PKEY_INDEX = (1<<4), 1199 IB_QP_PORT = (1<<5), 1200 IB_QP_QKEY = (1<<6), 1201 IB_QP_AV = (1<<7), 1202 IB_QP_PATH_MTU = (1<<8), 1203 IB_QP_TIMEOUT = (1<<9), 1204 IB_QP_RETRY_CNT = (1<<10), 1205 IB_QP_RNR_RETRY = (1<<11), 1206 IB_QP_RQ_PSN = (1<<12), 1207 IB_QP_MAX_QP_RD_ATOMIC = (1<<13), 1208 IB_QP_ALT_PATH = (1<<14), 1209 IB_QP_MIN_RNR_TIMER = (1<<15), 1210 IB_QP_SQ_PSN = (1<<16), 1211 IB_QP_MAX_DEST_RD_ATOMIC = (1<<17), 1212 IB_QP_PATH_MIG_STATE = (1<<18), 1213 IB_QP_CAP = (1<<19), 1214 IB_QP_DEST_QPN = (1<<20), 1215 IB_QP_RESERVED1 = (1<<21), 1216 IB_QP_RESERVED2 = (1<<22), 1217 IB_QP_RESERVED3 = (1<<23), 1218 IB_QP_RESERVED4 = (1<<24), 1219 IB_QP_RATE_LIMIT = (1<<25), 1220 }; 1221 1222 enum ib_qp_state { 1223 IB_QPS_RESET, 1224 IB_QPS_INIT, 1225 IB_QPS_RTR, 1226 IB_QPS_RTS, 1227 IB_QPS_SQD, 1228 IB_QPS_SQE, 1229 IB_QPS_ERR 1230 }; 1231 1232 enum ib_mig_state { 1233 IB_MIG_MIGRATED, 1234 IB_MIG_REARM, 1235 IB_MIG_ARMED 1236 }; 1237 1238 enum ib_mw_type { 1239 IB_MW_TYPE_1 = 1, 1240 IB_MW_TYPE_2 = 2 1241 }; 1242 1243 struct ib_qp_attr { 1244 enum ib_qp_state qp_state; 1245 enum ib_qp_state cur_qp_state; 1246 enum ib_mtu path_mtu; 1247 enum ib_mig_state path_mig_state; 1248 u32 qkey; 1249 u32 rq_psn; 1250 u32 sq_psn; 1251 u32 dest_qp_num; 1252 int qp_access_flags; 1253 struct ib_qp_cap cap; 1254 struct rdma_ah_attr ah_attr; 1255 struct rdma_ah_attr alt_ah_attr; 1256 u16 pkey_index; 1257 u16 alt_pkey_index; 1258 u8 en_sqd_async_notify; 1259 u8 sq_draining; 1260 u8 max_rd_atomic; 1261 u8 max_dest_rd_atomic; 1262 u8 min_rnr_timer; 1263 u8 port_num; 1264 u8 timeout; 1265 u8 retry_cnt; 1266 u8 rnr_retry; 1267 u8 alt_port_num; 1268 u8 alt_timeout; 1269 u32 rate_limit; 1270 }; 1271 1272 enum ib_wr_opcode { 1273 /* These are shared with userspace */ 1274 IB_WR_RDMA_WRITE = IB_UVERBS_WR_RDMA_WRITE, 1275 IB_WR_RDMA_WRITE_WITH_IMM = IB_UVERBS_WR_RDMA_WRITE_WITH_IMM, 1276 IB_WR_SEND = IB_UVERBS_WR_SEND, 1277 IB_WR_SEND_WITH_IMM = IB_UVERBS_WR_SEND_WITH_IMM, 1278 IB_WR_RDMA_READ = IB_UVERBS_WR_RDMA_READ, 1279 IB_WR_ATOMIC_CMP_AND_SWP = IB_UVERBS_WR_ATOMIC_CMP_AND_SWP, 1280 IB_WR_ATOMIC_FETCH_AND_ADD = IB_UVERBS_WR_ATOMIC_FETCH_AND_ADD, 1281 IB_WR_LSO = IB_UVERBS_WR_TSO, 1282 IB_WR_SEND_WITH_INV = IB_UVERBS_WR_SEND_WITH_INV, 1283 IB_WR_RDMA_READ_WITH_INV = IB_UVERBS_WR_RDMA_READ_WITH_INV, 1284 IB_WR_LOCAL_INV = IB_UVERBS_WR_LOCAL_INV, 1285 IB_WR_MASKED_ATOMIC_CMP_AND_SWP = 1286 IB_UVERBS_WR_MASKED_ATOMIC_CMP_AND_SWP, 1287 IB_WR_MASKED_ATOMIC_FETCH_AND_ADD = 1288 IB_UVERBS_WR_MASKED_ATOMIC_FETCH_AND_ADD, 1289 1290 /* These are kernel only and can not be issued by userspace */ 1291 IB_WR_REG_MR = 0x20, 1292 IB_WR_REG_MR_INTEGRITY, 1293 1294 /* reserve values for low level drivers' internal use. 1295 * These values will not be used at all in the ib core layer. 1296 */ 1297 IB_WR_RESERVED1 = 0xf0, 1298 IB_WR_RESERVED2, 1299 IB_WR_RESERVED3, 1300 IB_WR_RESERVED4, 1301 IB_WR_RESERVED5, 1302 IB_WR_RESERVED6, 1303 IB_WR_RESERVED7, 1304 IB_WR_RESERVED8, 1305 IB_WR_RESERVED9, 1306 IB_WR_RESERVED10, 1307 }; 1308 1309 enum ib_send_flags { 1310 IB_SEND_FENCE = 1, 1311 IB_SEND_SIGNALED = (1<<1), 1312 IB_SEND_SOLICITED = (1<<2), 1313 IB_SEND_INLINE = (1<<3), 1314 IB_SEND_IP_CSUM = (1<<4), 1315 1316 /* reserve bits 26-31 for low level drivers' internal use */ 1317 IB_SEND_RESERVED_START = (1 << 26), 1318 IB_SEND_RESERVED_END = (1 << 31), 1319 }; 1320 1321 struct ib_sge { 1322 u64 addr; 1323 u32 length; 1324 u32 lkey; 1325 }; 1326 1327 struct ib_cqe { 1328 void (*done)(struct ib_cq *cq, struct ib_wc *wc); 1329 }; 1330 1331 struct ib_send_wr { 1332 struct ib_send_wr *next; 1333 union { 1334 u64 wr_id; 1335 struct ib_cqe *wr_cqe; 1336 }; 1337 struct ib_sge *sg_list; 1338 int num_sge; 1339 enum ib_wr_opcode opcode; 1340 int send_flags; 1341 union { 1342 __be32 imm_data; 1343 u32 invalidate_rkey; 1344 } ex; 1345 }; 1346 1347 struct ib_rdma_wr { 1348 struct ib_send_wr wr; 1349 u64 remote_addr; 1350 u32 rkey; 1351 }; 1352 1353 static inline const struct ib_rdma_wr *rdma_wr(const struct ib_send_wr *wr) 1354 { 1355 return container_of(wr, struct ib_rdma_wr, wr); 1356 } 1357 1358 struct ib_atomic_wr { 1359 struct ib_send_wr wr; 1360 u64 remote_addr; 1361 u64 compare_add; 1362 u64 swap; 1363 u64 compare_add_mask; 1364 u64 swap_mask; 1365 u32 rkey; 1366 }; 1367 1368 static inline const struct ib_atomic_wr *atomic_wr(const struct ib_send_wr *wr) 1369 { 1370 return container_of(wr, struct ib_atomic_wr, wr); 1371 } 1372 1373 struct ib_ud_wr { 1374 struct ib_send_wr wr; 1375 struct ib_ah *ah; 1376 void *header; 1377 int hlen; 1378 int mss; 1379 u32 remote_qpn; 1380 u32 remote_qkey; 1381 u16 pkey_index; /* valid for GSI only */ 1382 u8 port_num; /* valid for DR SMPs on switch only */ 1383 }; 1384 1385 static inline const struct ib_ud_wr *ud_wr(const struct ib_send_wr *wr) 1386 { 1387 return container_of(wr, struct ib_ud_wr, wr); 1388 } 1389 1390 struct ib_reg_wr { 1391 struct ib_send_wr wr; 1392 struct ib_mr *mr; 1393 u32 key; 1394 int access; 1395 }; 1396 1397 static inline const struct ib_reg_wr *reg_wr(const struct ib_send_wr *wr) 1398 { 1399 return container_of(wr, struct ib_reg_wr, wr); 1400 } 1401 1402 struct ib_recv_wr { 1403 struct ib_recv_wr *next; 1404 union { 1405 u64 wr_id; 1406 struct ib_cqe *wr_cqe; 1407 }; 1408 struct ib_sge *sg_list; 1409 int num_sge; 1410 }; 1411 1412 enum ib_access_flags { 1413 IB_ACCESS_LOCAL_WRITE = IB_UVERBS_ACCESS_LOCAL_WRITE, 1414 IB_ACCESS_REMOTE_WRITE = IB_UVERBS_ACCESS_REMOTE_WRITE, 1415 IB_ACCESS_REMOTE_READ = IB_UVERBS_ACCESS_REMOTE_READ, 1416 IB_ACCESS_REMOTE_ATOMIC = IB_UVERBS_ACCESS_REMOTE_ATOMIC, 1417 IB_ACCESS_MW_BIND = IB_UVERBS_ACCESS_MW_BIND, 1418 IB_ZERO_BASED = IB_UVERBS_ACCESS_ZERO_BASED, 1419 IB_ACCESS_ON_DEMAND = IB_UVERBS_ACCESS_ON_DEMAND, 1420 IB_ACCESS_HUGETLB = IB_UVERBS_ACCESS_HUGETLB, 1421 IB_ACCESS_RELAXED_ORDERING = IB_UVERBS_ACCESS_RELAXED_ORDERING, 1422 1423 IB_ACCESS_OPTIONAL = IB_UVERBS_ACCESS_OPTIONAL_RANGE, 1424 IB_ACCESS_SUPPORTED = 1425 ((IB_ACCESS_HUGETLB << 1) - 1) | IB_ACCESS_OPTIONAL, 1426 }; 1427 1428 /* 1429 * XXX: these are apparently used for ->rereg_user_mr, no idea why they 1430 * are hidden here instead of a uapi header! 1431 */ 1432 enum ib_mr_rereg_flags { 1433 IB_MR_REREG_TRANS = 1, 1434 IB_MR_REREG_PD = (1<<1), 1435 IB_MR_REREG_ACCESS = (1<<2), 1436 IB_MR_REREG_SUPPORTED = ((IB_MR_REREG_ACCESS << 1) - 1) 1437 }; 1438 1439 struct ib_fmr_attr { 1440 int max_pages; 1441 int max_maps; 1442 u8 page_shift; 1443 }; 1444 1445 struct ib_umem; 1446 1447 enum rdma_remove_reason { 1448 /* 1449 * Userspace requested uobject deletion or initial try 1450 * to remove uobject via cleanup. Call could fail 1451 */ 1452 RDMA_REMOVE_DESTROY, 1453 /* Context deletion. This call should delete the actual object itself */ 1454 RDMA_REMOVE_CLOSE, 1455 /* Driver is being hot-unplugged. This call should delete the actual object itself */ 1456 RDMA_REMOVE_DRIVER_REMOVE, 1457 /* uobj is being cleaned-up before being committed */ 1458 RDMA_REMOVE_ABORT, 1459 }; 1460 1461 struct ib_rdmacg_object { 1462 #ifdef CONFIG_CGROUP_RDMA 1463 struct rdma_cgroup *cg; /* owner rdma cgroup */ 1464 #endif 1465 }; 1466 1467 struct ib_ucontext { 1468 struct ib_device *device; 1469 struct ib_uverbs_file *ufile; 1470 /* 1471 * 'closing' can be read by the driver only during a destroy callback, 1472 * it is set when we are closing the file descriptor and indicates 1473 * that mm_sem may be locked. 1474 */ 1475 bool closing; 1476 1477 bool cleanup_retryable; 1478 1479 struct ib_rdmacg_object cg_obj; 1480 /* 1481 * Implementation details of the RDMA core, don't use in drivers: 1482 */ 1483 struct rdma_restrack_entry res; 1484 struct xarray mmap_xa; 1485 }; 1486 1487 struct ib_uobject { 1488 u64 user_handle; /* handle given to us by userspace */ 1489 /* ufile & ucontext owning this object */ 1490 struct ib_uverbs_file *ufile; 1491 /* FIXME, save memory: ufile->context == context */ 1492 struct ib_ucontext *context; /* associated user context */ 1493 void *object; /* containing object */ 1494 struct list_head list; /* link to context's list */ 1495 struct ib_rdmacg_object cg_obj; /* rdmacg object */ 1496 int id; /* index into kernel idr */ 1497 struct kref ref; 1498 atomic_t usecnt; /* protects exclusive access */ 1499 struct rcu_head rcu; /* kfree_rcu() overhead */ 1500 1501 const struct uverbs_api_object *uapi_object; 1502 }; 1503 1504 struct ib_udata { 1505 const void __user *inbuf; 1506 void __user *outbuf; 1507 size_t inlen; 1508 size_t outlen; 1509 }; 1510 1511 struct ib_pd { 1512 u32 local_dma_lkey; 1513 u32 flags; 1514 struct ib_device *device; 1515 struct ib_uobject *uobject; 1516 atomic_t usecnt; /* count all resources */ 1517 1518 u32 unsafe_global_rkey; 1519 1520 /* 1521 * Implementation details of the RDMA core, don't use in drivers: 1522 */ 1523 struct ib_mr *__internal_mr; 1524 struct rdma_restrack_entry res; 1525 }; 1526 1527 struct ib_xrcd { 1528 struct ib_device *device; 1529 atomic_t usecnt; /* count all exposed resources */ 1530 struct inode *inode; 1531 1532 struct mutex tgt_qp_mutex; 1533 struct list_head tgt_qp_list; 1534 }; 1535 1536 struct ib_ah { 1537 struct ib_device *device; 1538 struct ib_pd *pd; 1539 struct ib_uobject *uobject; 1540 const struct ib_gid_attr *sgid_attr; 1541 enum rdma_ah_attr_type type; 1542 }; 1543 1544 typedef void (*ib_comp_handler)(struct ib_cq *cq, void *cq_context); 1545 1546 enum ib_poll_context { 1547 IB_POLL_DIRECT, /* caller context, no hw completions */ 1548 IB_POLL_SOFTIRQ, /* poll from softirq context */ 1549 IB_POLL_WORKQUEUE, /* poll from workqueue */ 1550 IB_POLL_UNBOUND_WORKQUEUE, /* poll from unbound workqueue */ 1551 }; 1552 1553 struct ib_cq { 1554 struct ib_device *device; 1555 struct ib_ucq_object *uobject; 1556 ib_comp_handler comp_handler; 1557 void (*event_handler)(struct ib_event *, void *); 1558 void *cq_context; 1559 int cqe; 1560 atomic_t usecnt; /* count number of work queues */ 1561 enum ib_poll_context poll_ctx; 1562 struct ib_wc *wc; 1563 union { 1564 struct irq_poll iop; 1565 struct work_struct work; 1566 }; 1567 struct workqueue_struct *comp_wq; 1568 struct dim *dim; 1569 1570 /* updated only by trace points */ 1571 ktime_t timestamp; 1572 bool interrupt; 1573 1574 /* 1575 * Implementation details of the RDMA core, don't use in drivers: 1576 */ 1577 struct rdma_restrack_entry res; 1578 }; 1579 1580 struct ib_srq { 1581 struct ib_device *device; 1582 struct ib_pd *pd; 1583 struct ib_usrq_object *uobject; 1584 void (*event_handler)(struct ib_event *, void *); 1585 void *srq_context; 1586 enum ib_srq_type srq_type; 1587 atomic_t usecnt; 1588 1589 struct { 1590 struct ib_cq *cq; 1591 union { 1592 struct { 1593 struct ib_xrcd *xrcd; 1594 u32 srq_num; 1595 } xrc; 1596 }; 1597 } ext; 1598 }; 1599 1600 enum ib_raw_packet_caps { 1601 /* Strip cvlan from incoming packet and report it in the matching work 1602 * completion is supported. 1603 */ 1604 IB_RAW_PACKET_CAP_CVLAN_STRIPPING = (1 << 0), 1605 /* Scatter FCS field of an incoming packet to host memory is supported. 1606 */ 1607 IB_RAW_PACKET_CAP_SCATTER_FCS = (1 << 1), 1608 /* Checksum offloads are supported (for both send and receive). */ 1609 IB_RAW_PACKET_CAP_IP_CSUM = (1 << 2), 1610 /* When a packet is received for an RQ with no receive WQEs, the 1611 * packet processing is delayed. 1612 */ 1613 IB_RAW_PACKET_CAP_DELAY_DROP = (1 << 3), 1614 }; 1615 1616 enum ib_wq_type { 1617 IB_WQT_RQ 1618 }; 1619 1620 enum ib_wq_state { 1621 IB_WQS_RESET, 1622 IB_WQS_RDY, 1623 IB_WQS_ERR 1624 }; 1625 1626 struct ib_wq { 1627 struct ib_device *device; 1628 struct ib_uwq_object *uobject; 1629 void *wq_context; 1630 void (*event_handler)(struct ib_event *, void *); 1631 struct ib_pd *pd; 1632 struct ib_cq *cq; 1633 u32 wq_num; 1634 enum ib_wq_state state; 1635 enum ib_wq_type wq_type; 1636 atomic_t usecnt; 1637 }; 1638 1639 enum ib_wq_flags { 1640 IB_WQ_FLAGS_CVLAN_STRIPPING = 1 << 0, 1641 IB_WQ_FLAGS_SCATTER_FCS = 1 << 1, 1642 IB_WQ_FLAGS_DELAY_DROP = 1 << 2, 1643 IB_WQ_FLAGS_PCI_WRITE_END_PADDING = 1 << 3, 1644 }; 1645 1646 struct ib_wq_init_attr { 1647 void *wq_context; 1648 enum ib_wq_type wq_type; 1649 u32 max_wr; 1650 u32 max_sge; 1651 struct ib_cq *cq; 1652 void (*event_handler)(struct ib_event *, void *); 1653 u32 create_flags; /* Use enum ib_wq_flags */ 1654 }; 1655 1656 enum ib_wq_attr_mask { 1657 IB_WQ_STATE = 1 << 0, 1658 IB_WQ_CUR_STATE = 1 << 1, 1659 IB_WQ_FLAGS = 1 << 2, 1660 }; 1661 1662 struct ib_wq_attr { 1663 enum ib_wq_state wq_state; 1664 enum ib_wq_state curr_wq_state; 1665 u32 flags; /* Use enum ib_wq_flags */ 1666 u32 flags_mask; /* Use enum ib_wq_flags */ 1667 }; 1668 1669 struct ib_rwq_ind_table { 1670 struct ib_device *device; 1671 struct ib_uobject *uobject; 1672 atomic_t usecnt; 1673 u32 ind_tbl_num; 1674 u32 log_ind_tbl_size; 1675 struct ib_wq **ind_tbl; 1676 }; 1677 1678 struct ib_rwq_ind_table_init_attr { 1679 u32 log_ind_tbl_size; 1680 /* Each entry is a pointer to Receive Work Queue */ 1681 struct ib_wq **ind_tbl; 1682 }; 1683 1684 enum port_pkey_state { 1685 IB_PORT_PKEY_NOT_VALID = 0, 1686 IB_PORT_PKEY_VALID = 1, 1687 IB_PORT_PKEY_LISTED = 2, 1688 }; 1689 1690 struct ib_qp_security; 1691 1692 struct ib_port_pkey { 1693 enum port_pkey_state state; 1694 u16 pkey_index; 1695 u8 port_num; 1696 struct list_head qp_list; 1697 struct list_head to_error_list; 1698 struct ib_qp_security *sec; 1699 }; 1700 1701 struct ib_ports_pkeys { 1702 struct ib_port_pkey main; 1703 struct ib_port_pkey alt; 1704 }; 1705 1706 struct ib_qp_security { 1707 struct ib_qp *qp; 1708 struct ib_device *dev; 1709 /* Hold this mutex when changing port and pkey settings. */ 1710 struct mutex mutex; 1711 struct ib_ports_pkeys *ports_pkeys; 1712 /* A list of all open shared QP handles. Required to enforce security 1713 * properly for all users of a shared QP. 1714 */ 1715 struct list_head shared_qp_list; 1716 void *security; 1717 bool destroying; 1718 atomic_t error_list_count; 1719 struct completion error_complete; 1720 int error_comps_pending; 1721 }; 1722 1723 /* 1724 * @max_write_sge: Maximum SGE elements per RDMA WRITE request. 1725 * @max_read_sge: Maximum SGE elements per RDMA READ request. 1726 */ 1727 struct ib_qp { 1728 struct ib_device *device; 1729 struct ib_pd *pd; 1730 struct ib_cq *send_cq; 1731 struct ib_cq *recv_cq; 1732 spinlock_t mr_lock; 1733 int mrs_used; 1734 struct list_head rdma_mrs; 1735 struct list_head sig_mrs; 1736 struct ib_srq *srq; 1737 struct ib_xrcd *xrcd; /* XRC TGT QPs only */ 1738 struct list_head xrcd_list; 1739 1740 /* count times opened, mcast attaches, flow attaches */ 1741 atomic_t usecnt; 1742 struct list_head open_list; 1743 struct ib_qp *real_qp; 1744 struct ib_uqp_object *uobject; 1745 void (*event_handler)(struct ib_event *, void *); 1746 void *qp_context; 1747 /* sgid_attrs associated with the AV's */ 1748 const struct ib_gid_attr *av_sgid_attr; 1749 const struct ib_gid_attr *alt_path_sgid_attr; 1750 u32 qp_num; 1751 u32 max_write_sge; 1752 u32 max_read_sge; 1753 enum ib_qp_type qp_type; 1754 struct ib_rwq_ind_table *rwq_ind_tbl; 1755 struct ib_qp_security *qp_sec; 1756 u8 port; 1757 1758 bool integrity_en; 1759 /* 1760 * Implementation details of the RDMA core, don't use in drivers: 1761 */ 1762 struct rdma_restrack_entry res; 1763 1764 /* The counter the qp is bind to */ 1765 struct rdma_counter *counter; 1766 }; 1767 1768 struct ib_dm { 1769 struct ib_device *device; 1770 u32 length; 1771 u32 flags; 1772 struct ib_uobject *uobject; 1773 atomic_t usecnt; 1774 }; 1775 1776 struct ib_mr { 1777 struct ib_device *device; 1778 struct ib_pd *pd; 1779 u32 lkey; 1780 u32 rkey; 1781 u64 iova; 1782 u64 length; 1783 unsigned int page_size; 1784 enum ib_mr_type type; 1785 bool need_inval; 1786 union { 1787 struct ib_uobject *uobject; /* user */ 1788 struct list_head qp_entry; /* FR */ 1789 }; 1790 1791 struct ib_dm *dm; 1792 struct ib_sig_attrs *sig_attrs; /* only for IB_MR_TYPE_INTEGRITY MRs */ 1793 /* 1794 * Implementation details of the RDMA core, don't use in drivers: 1795 */ 1796 struct rdma_restrack_entry res; 1797 }; 1798 1799 struct ib_mw { 1800 struct ib_device *device; 1801 struct ib_pd *pd; 1802 struct ib_uobject *uobject; 1803 u32 rkey; 1804 enum ib_mw_type type; 1805 }; 1806 1807 struct ib_fmr { 1808 struct ib_device *device; 1809 struct ib_pd *pd; 1810 struct list_head list; 1811 u32 lkey; 1812 u32 rkey; 1813 }; 1814 1815 /* Supported steering options */ 1816 enum ib_flow_attr_type { 1817 /* steering according to rule specifications */ 1818 IB_FLOW_ATTR_NORMAL = 0x0, 1819 /* default unicast and multicast rule - 1820 * receive all Eth traffic which isn't steered to any QP 1821 */ 1822 IB_FLOW_ATTR_ALL_DEFAULT = 0x1, 1823 /* default multicast rule - 1824 * receive all Eth multicast traffic which isn't steered to any QP 1825 */ 1826 IB_FLOW_ATTR_MC_DEFAULT = 0x2, 1827 /* sniffer rule - receive all port traffic */ 1828 IB_FLOW_ATTR_SNIFFER = 0x3 1829 }; 1830 1831 /* Supported steering header types */ 1832 enum ib_flow_spec_type { 1833 /* L2 headers*/ 1834 IB_FLOW_SPEC_ETH = 0x20, 1835 IB_FLOW_SPEC_IB = 0x22, 1836 /* L3 header*/ 1837 IB_FLOW_SPEC_IPV4 = 0x30, 1838 IB_FLOW_SPEC_IPV6 = 0x31, 1839 IB_FLOW_SPEC_ESP = 0x34, 1840 /* L4 headers*/ 1841 IB_FLOW_SPEC_TCP = 0x40, 1842 IB_FLOW_SPEC_UDP = 0x41, 1843 IB_FLOW_SPEC_VXLAN_TUNNEL = 0x50, 1844 IB_FLOW_SPEC_GRE = 0x51, 1845 IB_FLOW_SPEC_MPLS = 0x60, 1846 IB_FLOW_SPEC_INNER = 0x100, 1847 /* Actions */ 1848 IB_FLOW_SPEC_ACTION_TAG = 0x1000, 1849 IB_FLOW_SPEC_ACTION_DROP = 0x1001, 1850 IB_FLOW_SPEC_ACTION_HANDLE = 0x1002, 1851 IB_FLOW_SPEC_ACTION_COUNT = 0x1003, 1852 }; 1853 #define IB_FLOW_SPEC_LAYER_MASK 0xF0 1854 #define IB_FLOW_SPEC_SUPPORT_LAYERS 10 1855 1856 /* Flow steering rule priority is set according to it's domain. 1857 * Lower domain value means higher priority. 1858 */ 1859 enum ib_flow_domain { 1860 IB_FLOW_DOMAIN_USER, 1861 IB_FLOW_DOMAIN_ETHTOOL, 1862 IB_FLOW_DOMAIN_RFS, 1863 IB_FLOW_DOMAIN_NIC, 1864 IB_FLOW_DOMAIN_NUM /* Must be last */ 1865 }; 1866 1867 enum ib_flow_flags { 1868 IB_FLOW_ATTR_FLAGS_DONT_TRAP = 1UL << 1, /* Continue match, no steal */ 1869 IB_FLOW_ATTR_FLAGS_EGRESS = 1UL << 2, /* Egress flow */ 1870 IB_FLOW_ATTR_FLAGS_RESERVED = 1UL << 3 /* Must be last */ 1871 }; 1872 1873 struct ib_flow_eth_filter { 1874 u8 dst_mac[6]; 1875 u8 src_mac[6]; 1876 __be16 ether_type; 1877 __be16 vlan_tag; 1878 /* Must be last */ 1879 u8 real_sz[]; 1880 }; 1881 1882 struct ib_flow_spec_eth { 1883 u32 type; 1884 u16 size; 1885 struct ib_flow_eth_filter val; 1886 struct ib_flow_eth_filter mask; 1887 }; 1888 1889 struct ib_flow_ib_filter { 1890 __be16 dlid; 1891 __u8 sl; 1892 /* Must be last */ 1893 u8 real_sz[]; 1894 }; 1895 1896 struct ib_flow_spec_ib { 1897 u32 type; 1898 u16 size; 1899 struct ib_flow_ib_filter val; 1900 struct ib_flow_ib_filter mask; 1901 }; 1902 1903 /* IPv4 header flags */ 1904 enum ib_ipv4_flags { 1905 IB_IPV4_DONT_FRAG = 0x2, /* Don't enable packet fragmentation */ 1906 IB_IPV4_MORE_FRAG = 0X4 /* For All fragmented packets except the 1907 last have this flag set */ 1908 }; 1909 1910 struct ib_flow_ipv4_filter { 1911 __be32 src_ip; 1912 __be32 dst_ip; 1913 u8 proto; 1914 u8 tos; 1915 u8 ttl; 1916 u8 flags; 1917 /* Must be last */ 1918 u8 real_sz[]; 1919 }; 1920 1921 struct ib_flow_spec_ipv4 { 1922 u32 type; 1923 u16 size; 1924 struct ib_flow_ipv4_filter val; 1925 struct ib_flow_ipv4_filter mask; 1926 }; 1927 1928 struct ib_flow_ipv6_filter { 1929 u8 src_ip[16]; 1930 u8 dst_ip[16]; 1931 __be32 flow_label; 1932 u8 next_hdr; 1933 u8 traffic_class; 1934 u8 hop_limit; 1935 /* Must be last */ 1936 u8 real_sz[]; 1937 }; 1938 1939 struct ib_flow_spec_ipv6 { 1940 u32 type; 1941 u16 size; 1942 struct ib_flow_ipv6_filter val; 1943 struct ib_flow_ipv6_filter mask; 1944 }; 1945 1946 struct ib_flow_tcp_udp_filter { 1947 __be16 dst_port; 1948 __be16 src_port; 1949 /* Must be last */ 1950 u8 real_sz[]; 1951 }; 1952 1953 struct ib_flow_spec_tcp_udp { 1954 u32 type; 1955 u16 size; 1956 struct ib_flow_tcp_udp_filter val; 1957 struct ib_flow_tcp_udp_filter mask; 1958 }; 1959 1960 struct ib_flow_tunnel_filter { 1961 __be32 tunnel_id; 1962 u8 real_sz[]; 1963 }; 1964 1965 /* ib_flow_spec_tunnel describes the Vxlan tunnel 1966 * the tunnel_id from val has the vni value 1967 */ 1968 struct ib_flow_spec_tunnel { 1969 u32 type; 1970 u16 size; 1971 struct ib_flow_tunnel_filter val; 1972 struct ib_flow_tunnel_filter mask; 1973 }; 1974 1975 struct ib_flow_esp_filter { 1976 __be32 spi; 1977 __be32 seq; 1978 /* Must be last */ 1979 u8 real_sz[]; 1980 }; 1981 1982 struct ib_flow_spec_esp { 1983 u32 type; 1984 u16 size; 1985 struct ib_flow_esp_filter val; 1986 struct ib_flow_esp_filter mask; 1987 }; 1988 1989 struct ib_flow_gre_filter { 1990 __be16 c_ks_res0_ver; 1991 __be16 protocol; 1992 __be32 key; 1993 /* Must be last */ 1994 u8 real_sz[]; 1995 }; 1996 1997 struct ib_flow_spec_gre { 1998 u32 type; 1999 u16 size; 2000 struct ib_flow_gre_filter val; 2001 struct ib_flow_gre_filter mask; 2002 }; 2003 2004 struct ib_flow_mpls_filter { 2005 __be32 tag; 2006 /* Must be last */ 2007 u8 real_sz[]; 2008 }; 2009 2010 struct ib_flow_spec_mpls { 2011 u32 type; 2012 u16 size; 2013 struct ib_flow_mpls_filter val; 2014 struct ib_flow_mpls_filter mask; 2015 }; 2016 2017 struct ib_flow_spec_action_tag { 2018 enum ib_flow_spec_type type; 2019 u16 size; 2020 u32 tag_id; 2021 }; 2022 2023 struct ib_flow_spec_action_drop { 2024 enum ib_flow_spec_type type; 2025 u16 size; 2026 }; 2027 2028 struct ib_flow_spec_action_handle { 2029 enum ib_flow_spec_type type; 2030 u16 size; 2031 struct ib_flow_action *act; 2032 }; 2033 2034 enum ib_counters_description { 2035 IB_COUNTER_PACKETS, 2036 IB_COUNTER_BYTES, 2037 }; 2038 2039 struct ib_flow_spec_action_count { 2040 enum ib_flow_spec_type type; 2041 u16 size; 2042 struct ib_counters *counters; 2043 }; 2044 2045 union ib_flow_spec { 2046 struct { 2047 u32 type; 2048 u16 size; 2049 }; 2050 struct ib_flow_spec_eth eth; 2051 struct ib_flow_spec_ib ib; 2052 struct ib_flow_spec_ipv4 ipv4; 2053 struct ib_flow_spec_tcp_udp tcp_udp; 2054 struct ib_flow_spec_ipv6 ipv6; 2055 struct ib_flow_spec_tunnel tunnel; 2056 struct ib_flow_spec_esp esp; 2057 struct ib_flow_spec_gre gre; 2058 struct ib_flow_spec_mpls mpls; 2059 struct ib_flow_spec_action_tag flow_tag; 2060 struct ib_flow_spec_action_drop drop; 2061 struct ib_flow_spec_action_handle action; 2062 struct ib_flow_spec_action_count flow_count; 2063 }; 2064 2065 struct ib_flow_attr { 2066 enum ib_flow_attr_type type; 2067 u16 size; 2068 u16 priority; 2069 u32 flags; 2070 u8 num_of_specs; 2071 u8 port; 2072 union ib_flow_spec flows[]; 2073 }; 2074 2075 struct ib_flow { 2076 struct ib_qp *qp; 2077 struct ib_device *device; 2078 struct ib_uobject *uobject; 2079 }; 2080 2081 enum ib_flow_action_type { 2082 IB_FLOW_ACTION_UNSPECIFIED, 2083 IB_FLOW_ACTION_ESP = 1, 2084 }; 2085 2086 struct ib_flow_action_attrs_esp_keymats { 2087 enum ib_uverbs_flow_action_esp_keymat protocol; 2088 union { 2089 struct ib_uverbs_flow_action_esp_keymat_aes_gcm aes_gcm; 2090 } keymat; 2091 }; 2092 2093 struct ib_flow_action_attrs_esp_replays { 2094 enum ib_uverbs_flow_action_esp_replay protocol; 2095 union { 2096 struct ib_uverbs_flow_action_esp_replay_bmp bmp; 2097 } replay; 2098 }; 2099 2100 enum ib_flow_action_attrs_esp_flags { 2101 /* All user-space flags at the top: Use enum ib_uverbs_flow_action_esp_flags 2102 * This is done in order to share the same flags between user-space and 2103 * kernel and spare an unnecessary translation. 2104 */ 2105 2106 /* Kernel flags */ 2107 IB_FLOW_ACTION_ESP_FLAGS_ESN_TRIGGERED = 1ULL << 32, 2108 IB_FLOW_ACTION_ESP_FLAGS_MOD_ESP_ATTRS = 1ULL << 33, 2109 }; 2110 2111 struct ib_flow_spec_list { 2112 struct ib_flow_spec_list *next; 2113 union ib_flow_spec spec; 2114 }; 2115 2116 struct ib_flow_action_attrs_esp { 2117 struct ib_flow_action_attrs_esp_keymats *keymat; 2118 struct ib_flow_action_attrs_esp_replays *replay; 2119 struct ib_flow_spec_list *encap; 2120 /* Used only if IB_FLOW_ACTION_ESP_FLAGS_ESN_TRIGGERED is enabled. 2121 * Value of 0 is a valid value. 2122 */ 2123 u32 esn; 2124 u32 spi; 2125 u32 seq; 2126 u32 tfc_pad; 2127 /* Use enum ib_flow_action_attrs_esp_flags */ 2128 u64 flags; 2129 u64 hard_limit_pkts; 2130 }; 2131 2132 struct ib_flow_action { 2133 struct ib_device *device; 2134 struct ib_uobject *uobject; 2135 enum ib_flow_action_type type; 2136 atomic_t usecnt; 2137 }; 2138 2139 struct ib_mad; 2140 struct ib_grh; 2141 2142 enum ib_process_mad_flags { 2143 IB_MAD_IGNORE_MKEY = 1, 2144 IB_MAD_IGNORE_BKEY = 2, 2145 IB_MAD_IGNORE_ALL = IB_MAD_IGNORE_MKEY | IB_MAD_IGNORE_BKEY 2146 }; 2147 2148 enum ib_mad_result { 2149 IB_MAD_RESULT_FAILURE = 0, /* (!SUCCESS is the important flag) */ 2150 IB_MAD_RESULT_SUCCESS = 1 << 0, /* MAD was successfully processed */ 2151 IB_MAD_RESULT_REPLY = 1 << 1, /* Reply packet needs to be sent */ 2152 IB_MAD_RESULT_CONSUMED = 1 << 2 /* Packet consumed: stop processing */ 2153 }; 2154 2155 struct ib_port_cache { 2156 u64 subnet_prefix; 2157 struct ib_pkey_cache *pkey; 2158 struct ib_gid_table *gid; 2159 u8 lmc; 2160 enum ib_port_state port_state; 2161 }; 2162 2163 struct ib_port_immutable { 2164 int pkey_tbl_len; 2165 int gid_tbl_len; 2166 u32 core_cap_flags; 2167 u32 max_mad_size; 2168 }; 2169 2170 struct ib_port_data { 2171 struct ib_device *ib_dev; 2172 2173 struct ib_port_immutable immutable; 2174 2175 spinlock_t pkey_list_lock; 2176 struct list_head pkey_list; 2177 2178 struct ib_port_cache cache; 2179 2180 spinlock_t netdev_lock; 2181 struct net_device __rcu *netdev; 2182 struct hlist_node ndev_hash_link; 2183 struct rdma_port_counter port_counter; 2184 struct rdma_hw_stats *hw_stats; 2185 }; 2186 2187 /* rdma netdev type - specifies protocol type */ 2188 enum rdma_netdev_t { 2189 RDMA_NETDEV_OPA_VNIC, 2190 RDMA_NETDEV_IPOIB, 2191 }; 2192 2193 /** 2194 * struct rdma_netdev - rdma netdev 2195 * For cases where netstack interfacing is required. 2196 */ 2197 struct rdma_netdev { 2198 void *clnt_priv; 2199 struct ib_device *hca; 2200 u8 port_num; 2201 2202 /* 2203 * cleanup function must be specified. 2204 * FIXME: This is only used for OPA_VNIC and that usage should be 2205 * removed too. 2206 */ 2207 void (*free_rdma_netdev)(struct net_device *netdev); 2208 2209 /* control functions */ 2210 void (*set_id)(struct net_device *netdev, int id); 2211 /* send packet */ 2212 int (*send)(struct net_device *dev, struct sk_buff *skb, 2213 struct ib_ah *address, u32 dqpn); 2214 /* multicast */ 2215 int (*attach_mcast)(struct net_device *dev, struct ib_device *hca, 2216 union ib_gid *gid, u16 mlid, 2217 int set_qkey, u32 qkey); 2218 int (*detach_mcast)(struct net_device *dev, struct ib_device *hca, 2219 union ib_gid *gid, u16 mlid); 2220 }; 2221 2222 struct rdma_netdev_alloc_params { 2223 size_t sizeof_priv; 2224 unsigned int txqs; 2225 unsigned int rxqs; 2226 void *param; 2227 2228 int (*initialize_rdma_netdev)(struct ib_device *device, u8 port_num, 2229 struct net_device *netdev, void *param); 2230 }; 2231 2232 struct ib_odp_counters { 2233 atomic64_t faults; 2234 atomic64_t invalidations; 2235 }; 2236 2237 struct ib_counters { 2238 struct ib_device *device; 2239 struct ib_uobject *uobject; 2240 /* num of objects attached */ 2241 atomic_t usecnt; 2242 }; 2243 2244 struct ib_counters_read_attr { 2245 u64 *counters_buff; 2246 u32 ncounters; 2247 u32 flags; /* use enum ib_read_counters_flags */ 2248 }; 2249 2250 struct uverbs_attr_bundle; 2251 struct iw_cm_id; 2252 struct iw_cm_conn_param; 2253 2254 #define INIT_RDMA_OBJ_SIZE(ib_struct, drv_struct, member) \ 2255 .size_##ib_struct = \ 2256 (sizeof(struct drv_struct) + \ 2257 BUILD_BUG_ON_ZERO(offsetof(struct drv_struct, member)) + \ 2258 BUILD_BUG_ON_ZERO( \ 2259 !__same_type(((struct drv_struct *)NULL)->member, \ 2260 struct ib_struct))) 2261 2262 #define rdma_zalloc_drv_obj_gfp(ib_dev, ib_type, gfp) \ 2263 ((struct ib_type *)kzalloc(ib_dev->ops.size_##ib_type, gfp)) 2264 2265 #define rdma_zalloc_drv_obj(ib_dev, ib_type) \ 2266 rdma_zalloc_drv_obj_gfp(ib_dev, ib_type, GFP_KERNEL) 2267 2268 #define DECLARE_RDMA_OBJ_SIZE(ib_struct) size_t size_##ib_struct 2269 2270 struct rdma_user_mmap_entry { 2271 struct kref ref; 2272 struct ib_ucontext *ucontext; 2273 unsigned long start_pgoff; 2274 size_t npages; 2275 bool driver_removed; 2276 }; 2277 2278 /* Return the offset (in bytes) the user should pass to libc's mmap() */ 2279 static inline u64 2280 rdma_user_mmap_get_offset(const struct rdma_user_mmap_entry *entry) 2281 { 2282 return (u64)entry->start_pgoff << PAGE_SHIFT; 2283 } 2284 2285 /** 2286 * struct ib_device_ops - InfiniBand device operations 2287 * This structure defines all the InfiniBand device operations, providers will 2288 * need to define the supported operations, otherwise they will be set to null. 2289 */ 2290 struct ib_device_ops { 2291 struct module *owner; 2292 enum rdma_driver_id driver_id; 2293 u32 uverbs_abi_ver; 2294 unsigned int uverbs_no_driver_id_binding:1; 2295 2296 int (*post_send)(struct ib_qp *qp, const struct ib_send_wr *send_wr, 2297 const struct ib_send_wr **bad_send_wr); 2298 int (*post_recv)(struct ib_qp *qp, const struct ib_recv_wr *recv_wr, 2299 const struct ib_recv_wr **bad_recv_wr); 2300 void (*drain_rq)(struct ib_qp *qp); 2301 void (*drain_sq)(struct ib_qp *qp); 2302 int (*poll_cq)(struct ib_cq *cq, int num_entries, struct ib_wc *wc); 2303 int (*peek_cq)(struct ib_cq *cq, int wc_cnt); 2304 int (*req_notify_cq)(struct ib_cq *cq, enum ib_cq_notify_flags flags); 2305 int (*req_ncomp_notif)(struct ib_cq *cq, int wc_cnt); 2306 int (*post_srq_recv)(struct ib_srq *srq, 2307 const struct ib_recv_wr *recv_wr, 2308 const struct ib_recv_wr **bad_recv_wr); 2309 int (*process_mad)(struct ib_device *device, int process_mad_flags, 2310 u8 port_num, const struct ib_wc *in_wc, 2311 const struct ib_grh *in_grh, 2312 const struct ib_mad *in_mad, struct ib_mad *out_mad, 2313 size_t *out_mad_size, u16 *out_mad_pkey_index); 2314 int (*query_device)(struct ib_device *device, 2315 struct ib_device_attr *device_attr, 2316 struct ib_udata *udata); 2317 int (*modify_device)(struct ib_device *device, int device_modify_mask, 2318 struct ib_device_modify *device_modify); 2319 void (*get_dev_fw_str)(struct ib_device *device, char *str); 2320 const struct cpumask *(*get_vector_affinity)(struct ib_device *ibdev, 2321 int comp_vector); 2322 int (*query_port)(struct ib_device *device, u8 port_num, 2323 struct ib_port_attr *port_attr); 2324 int (*modify_port)(struct ib_device *device, u8 port_num, 2325 int port_modify_mask, 2326 struct ib_port_modify *port_modify); 2327 /** 2328 * The following mandatory functions are used only at device 2329 * registration. Keep functions such as these at the end of this 2330 * structure to avoid cache line misses when accessing struct ib_device 2331 * in fast paths. 2332 */ 2333 int (*get_port_immutable)(struct ib_device *device, u8 port_num, 2334 struct ib_port_immutable *immutable); 2335 enum rdma_link_layer (*get_link_layer)(struct ib_device *device, 2336 u8 port_num); 2337 /** 2338 * When calling get_netdev, the HW vendor's driver should return the 2339 * net device of device @device at port @port_num or NULL if such 2340 * a net device doesn't exist. The vendor driver should call dev_hold 2341 * on this net device. The HW vendor's device driver must guarantee 2342 * that this function returns NULL before the net device has finished 2343 * NETDEV_UNREGISTER state. 2344 */ 2345 struct net_device *(*get_netdev)(struct ib_device *device, u8 port_num); 2346 /** 2347 * rdma netdev operation 2348 * 2349 * Driver implementing alloc_rdma_netdev or rdma_netdev_get_params 2350 * must return -EOPNOTSUPP if it doesn't support the specified type. 2351 */ 2352 struct net_device *(*alloc_rdma_netdev)( 2353 struct ib_device *device, u8 port_num, enum rdma_netdev_t type, 2354 const char *name, unsigned char name_assign_type, 2355 void (*setup)(struct net_device *)); 2356 2357 int (*rdma_netdev_get_params)(struct ib_device *device, u8 port_num, 2358 enum rdma_netdev_t type, 2359 struct rdma_netdev_alloc_params *params); 2360 /** 2361 * query_gid should be return GID value for @device, when @port_num 2362 * link layer is either IB or iWarp. It is no-op if @port_num port 2363 * is RoCE link layer. 2364 */ 2365 int (*query_gid)(struct ib_device *device, u8 port_num, int index, 2366 union ib_gid *gid); 2367 /** 2368 * When calling add_gid, the HW vendor's driver should add the gid 2369 * of device of port at gid index available at @attr. Meta-info of 2370 * that gid (for example, the network device related to this gid) is 2371 * available at @attr. @context allows the HW vendor driver to store 2372 * extra information together with a GID entry. The HW vendor driver may 2373 * allocate memory to contain this information and store it in @context 2374 * when a new GID entry is written to. Params are consistent until the 2375 * next call of add_gid or delete_gid. The function should return 0 on 2376 * success or error otherwise. The function could be called 2377 * concurrently for different ports. This function is only called when 2378 * roce_gid_table is used. 2379 */ 2380 int (*add_gid)(const struct ib_gid_attr *attr, void **context); 2381 /** 2382 * When calling del_gid, the HW vendor's driver should delete the 2383 * gid of device @device at gid index gid_index of port port_num 2384 * available in @attr. 2385 * Upon the deletion of a GID entry, the HW vendor must free any 2386 * allocated memory. The caller will clear @context afterwards. 2387 * This function is only called when roce_gid_table is used. 2388 */ 2389 int (*del_gid)(const struct ib_gid_attr *attr, void **context); 2390 int (*query_pkey)(struct ib_device *device, u8 port_num, u16 index, 2391 u16 *pkey); 2392 int (*alloc_ucontext)(struct ib_ucontext *context, 2393 struct ib_udata *udata); 2394 void (*dealloc_ucontext)(struct ib_ucontext *context); 2395 int (*mmap)(struct ib_ucontext *context, struct vm_area_struct *vma); 2396 /** 2397 * This will be called once refcount of an entry in mmap_xa reaches 2398 * zero. The type of the memory that was mapped may differ between 2399 * entries and is opaque to the rdma_user_mmap interface. 2400 * Therefore needs to be implemented by the driver in mmap_free. 2401 */ 2402 void (*mmap_free)(struct rdma_user_mmap_entry *entry); 2403 void (*disassociate_ucontext)(struct ib_ucontext *ibcontext); 2404 int (*alloc_pd)(struct ib_pd *pd, struct ib_udata *udata); 2405 void (*dealloc_pd)(struct ib_pd *pd, struct ib_udata *udata); 2406 int (*create_ah)(struct ib_ah *ah, struct rdma_ah_attr *ah_attr, 2407 u32 flags, struct ib_udata *udata); 2408 int (*modify_ah)(struct ib_ah *ah, struct rdma_ah_attr *ah_attr); 2409 int (*query_ah)(struct ib_ah *ah, struct rdma_ah_attr *ah_attr); 2410 void (*destroy_ah)(struct ib_ah *ah, u32 flags); 2411 int (*create_srq)(struct ib_srq *srq, 2412 struct ib_srq_init_attr *srq_init_attr, 2413 struct ib_udata *udata); 2414 int (*modify_srq)(struct ib_srq *srq, struct ib_srq_attr *srq_attr, 2415 enum ib_srq_attr_mask srq_attr_mask, 2416 struct ib_udata *udata); 2417 int (*query_srq)(struct ib_srq *srq, struct ib_srq_attr *srq_attr); 2418 void (*destroy_srq)(struct ib_srq *srq, struct ib_udata *udata); 2419 struct ib_qp *(*create_qp)(struct ib_pd *pd, 2420 struct ib_qp_init_attr *qp_init_attr, 2421 struct ib_udata *udata); 2422 int (*modify_qp)(struct ib_qp *qp, struct ib_qp_attr *qp_attr, 2423 int qp_attr_mask, struct ib_udata *udata); 2424 int (*query_qp)(struct ib_qp *qp, struct ib_qp_attr *qp_attr, 2425 int qp_attr_mask, struct ib_qp_init_attr *qp_init_attr); 2426 int (*destroy_qp)(struct ib_qp *qp, struct ib_udata *udata); 2427 int (*create_cq)(struct ib_cq *cq, const struct ib_cq_init_attr *attr, 2428 struct ib_udata *udata); 2429 int (*modify_cq)(struct ib_cq *cq, u16 cq_count, u16 cq_period); 2430 void (*destroy_cq)(struct ib_cq *cq, struct ib_udata *udata); 2431 int (*resize_cq)(struct ib_cq *cq, int cqe, struct ib_udata *udata); 2432 struct ib_mr *(*get_dma_mr)(struct ib_pd *pd, int mr_access_flags); 2433 struct ib_mr *(*reg_user_mr)(struct ib_pd *pd, u64 start, u64 length, 2434 u64 virt_addr, int mr_access_flags, 2435 struct ib_udata *udata); 2436 int (*rereg_user_mr)(struct ib_mr *mr, int flags, u64 start, u64 length, 2437 u64 virt_addr, int mr_access_flags, 2438 struct ib_pd *pd, struct ib_udata *udata); 2439 int (*dereg_mr)(struct ib_mr *mr, struct ib_udata *udata); 2440 struct ib_mr *(*alloc_mr)(struct ib_pd *pd, enum ib_mr_type mr_type, 2441 u32 max_num_sg, struct ib_udata *udata); 2442 struct ib_mr *(*alloc_mr_integrity)(struct ib_pd *pd, 2443 u32 max_num_data_sg, 2444 u32 max_num_meta_sg); 2445 int (*advise_mr)(struct ib_pd *pd, 2446 enum ib_uverbs_advise_mr_advice advice, u32 flags, 2447 struct ib_sge *sg_list, u32 num_sge, 2448 struct uverbs_attr_bundle *attrs); 2449 int (*map_mr_sg)(struct ib_mr *mr, struct scatterlist *sg, int sg_nents, 2450 unsigned int *sg_offset); 2451 int (*check_mr_status)(struct ib_mr *mr, u32 check_mask, 2452 struct ib_mr_status *mr_status); 2453 struct ib_mw *(*alloc_mw)(struct ib_pd *pd, enum ib_mw_type type, 2454 struct ib_udata *udata); 2455 int (*dealloc_mw)(struct ib_mw *mw); 2456 struct ib_fmr *(*alloc_fmr)(struct ib_pd *pd, int mr_access_flags, 2457 struct ib_fmr_attr *fmr_attr); 2458 int (*map_phys_fmr)(struct ib_fmr *fmr, u64 *page_list, int list_len, 2459 u64 iova); 2460 int (*unmap_fmr)(struct list_head *fmr_list); 2461 int (*dealloc_fmr)(struct ib_fmr *fmr); 2462 int (*attach_mcast)(struct ib_qp *qp, union ib_gid *gid, u16 lid); 2463 int (*detach_mcast)(struct ib_qp *qp, union ib_gid *gid, u16 lid); 2464 struct ib_xrcd *(*alloc_xrcd)(struct ib_device *device, 2465 struct ib_udata *udata); 2466 int (*dealloc_xrcd)(struct ib_xrcd *xrcd, struct ib_udata *udata); 2467 struct ib_flow *(*create_flow)(struct ib_qp *qp, 2468 struct ib_flow_attr *flow_attr, 2469 int domain, struct ib_udata *udata); 2470 int (*destroy_flow)(struct ib_flow *flow_id); 2471 struct ib_flow_action *(*create_flow_action_esp)( 2472 struct ib_device *device, 2473 const struct ib_flow_action_attrs_esp *attr, 2474 struct uverbs_attr_bundle *attrs); 2475 int (*destroy_flow_action)(struct ib_flow_action *action); 2476 int (*modify_flow_action_esp)( 2477 struct ib_flow_action *action, 2478 const struct ib_flow_action_attrs_esp *attr, 2479 struct uverbs_attr_bundle *attrs); 2480 int (*set_vf_link_state)(struct ib_device *device, int vf, u8 port, 2481 int state); 2482 int (*get_vf_config)(struct ib_device *device, int vf, u8 port, 2483 struct ifla_vf_info *ivf); 2484 int (*get_vf_stats)(struct ib_device *device, int vf, u8 port, 2485 struct ifla_vf_stats *stats); 2486 int (*get_vf_guid)(struct ib_device *device, int vf, u8 port, 2487 struct ifla_vf_guid *node_guid, 2488 struct ifla_vf_guid *port_guid); 2489 int (*set_vf_guid)(struct ib_device *device, int vf, u8 port, u64 guid, 2490 int type); 2491 struct ib_wq *(*create_wq)(struct ib_pd *pd, 2492 struct ib_wq_init_attr *init_attr, 2493 struct ib_udata *udata); 2494 void (*destroy_wq)(struct ib_wq *wq, struct ib_udata *udata); 2495 int (*modify_wq)(struct ib_wq *wq, struct ib_wq_attr *attr, 2496 u32 wq_attr_mask, struct ib_udata *udata); 2497 struct ib_rwq_ind_table *(*create_rwq_ind_table)( 2498 struct ib_device *device, 2499 struct ib_rwq_ind_table_init_attr *init_attr, 2500 struct ib_udata *udata); 2501 int (*destroy_rwq_ind_table)(struct ib_rwq_ind_table *wq_ind_table); 2502 struct ib_dm *(*alloc_dm)(struct ib_device *device, 2503 struct ib_ucontext *context, 2504 struct ib_dm_alloc_attr *attr, 2505 struct uverbs_attr_bundle *attrs); 2506 int (*dealloc_dm)(struct ib_dm *dm, struct uverbs_attr_bundle *attrs); 2507 struct ib_mr *(*reg_dm_mr)(struct ib_pd *pd, struct ib_dm *dm, 2508 struct ib_dm_mr_attr *attr, 2509 struct uverbs_attr_bundle *attrs); 2510 struct ib_counters *(*create_counters)( 2511 struct ib_device *device, struct uverbs_attr_bundle *attrs); 2512 int (*destroy_counters)(struct ib_counters *counters); 2513 int (*read_counters)(struct ib_counters *counters, 2514 struct ib_counters_read_attr *counters_read_attr, 2515 struct uverbs_attr_bundle *attrs); 2516 int (*map_mr_sg_pi)(struct ib_mr *mr, struct scatterlist *data_sg, 2517 int data_sg_nents, unsigned int *data_sg_offset, 2518 struct scatterlist *meta_sg, int meta_sg_nents, 2519 unsigned int *meta_sg_offset); 2520 2521 /** 2522 * alloc_hw_stats - Allocate a struct rdma_hw_stats and fill in the 2523 * driver initialized data. The struct is kfree()'ed by the sysfs 2524 * core when the device is removed. A lifespan of -1 in the return 2525 * struct tells the core to set a default lifespan. 2526 */ 2527 struct rdma_hw_stats *(*alloc_hw_stats)(struct ib_device *device, 2528 u8 port_num); 2529 /** 2530 * get_hw_stats - Fill in the counter value(s) in the stats struct. 2531 * @index - The index in the value array we wish to have updated, or 2532 * num_counters if we want all stats updated 2533 * Return codes - 2534 * < 0 - Error, no counters updated 2535 * index - Updated the single counter pointed to by index 2536 * num_counters - Updated all counters (will reset the timestamp 2537 * and prevent further calls for lifespan milliseconds) 2538 * Drivers are allowed to update all counters in leiu of just the 2539 * one given in index at their option 2540 */ 2541 int (*get_hw_stats)(struct ib_device *device, 2542 struct rdma_hw_stats *stats, u8 port, int index); 2543 /* 2544 * This function is called once for each port when a ib device is 2545 * registered. 2546 */ 2547 int (*init_port)(struct ib_device *device, u8 port_num, 2548 struct kobject *port_sysfs); 2549 /** 2550 * Allows rdma drivers to add their own restrack attributes. 2551 */ 2552 int (*fill_res_entry)(struct sk_buff *msg, 2553 struct rdma_restrack_entry *entry); 2554 2555 /* Device lifecycle callbacks */ 2556 /* 2557 * Called after the device becomes registered, before clients are 2558 * attached 2559 */ 2560 int (*enable_driver)(struct ib_device *dev); 2561 /* 2562 * This is called as part of ib_dealloc_device(). 2563 */ 2564 void (*dealloc_driver)(struct ib_device *dev); 2565 2566 /* iWarp CM callbacks */ 2567 void (*iw_add_ref)(struct ib_qp *qp); 2568 void (*iw_rem_ref)(struct ib_qp *qp); 2569 struct ib_qp *(*iw_get_qp)(struct ib_device *device, int qpn); 2570 int (*iw_connect)(struct iw_cm_id *cm_id, 2571 struct iw_cm_conn_param *conn_param); 2572 int (*iw_accept)(struct iw_cm_id *cm_id, 2573 struct iw_cm_conn_param *conn_param); 2574 int (*iw_reject)(struct iw_cm_id *cm_id, const void *pdata, 2575 u8 pdata_len); 2576 int (*iw_create_listen)(struct iw_cm_id *cm_id, int backlog); 2577 int (*iw_destroy_listen)(struct iw_cm_id *cm_id); 2578 /** 2579 * counter_bind_qp - Bind a QP to a counter. 2580 * @counter - The counter to be bound. If counter->id is zero then 2581 * the driver needs to allocate a new counter and set counter->id 2582 */ 2583 int (*counter_bind_qp)(struct rdma_counter *counter, struct ib_qp *qp); 2584 /** 2585 * counter_unbind_qp - Unbind the qp from the dynamically-allocated 2586 * counter and bind it onto the default one 2587 */ 2588 int (*counter_unbind_qp)(struct ib_qp *qp); 2589 /** 2590 * counter_dealloc -De-allocate the hw counter 2591 */ 2592 int (*counter_dealloc)(struct rdma_counter *counter); 2593 /** 2594 * counter_alloc_stats - Allocate a struct rdma_hw_stats and fill in 2595 * the driver initialized data. 2596 */ 2597 struct rdma_hw_stats *(*counter_alloc_stats)( 2598 struct rdma_counter *counter); 2599 /** 2600 * counter_update_stats - Query the stats value of this counter 2601 */ 2602 int (*counter_update_stats)(struct rdma_counter *counter); 2603 2604 /** 2605 * Allows rdma drivers to add their own restrack attributes 2606 * dumped via 'rdma stat' iproute2 command. 2607 */ 2608 int (*fill_stat_entry)(struct sk_buff *msg, 2609 struct rdma_restrack_entry *entry); 2610 2611 DECLARE_RDMA_OBJ_SIZE(ib_ah); 2612 DECLARE_RDMA_OBJ_SIZE(ib_cq); 2613 DECLARE_RDMA_OBJ_SIZE(ib_pd); 2614 DECLARE_RDMA_OBJ_SIZE(ib_srq); 2615 DECLARE_RDMA_OBJ_SIZE(ib_ucontext); 2616 }; 2617 2618 struct ib_core_device { 2619 /* device must be the first element in structure until, 2620 * union of ib_core_device and device exists in ib_device. 2621 */ 2622 struct device dev; 2623 possible_net_t rdma_net; 2624 struct kobject *ports_kobj; 2625 struct list_head port_list; 2626 struct ib_device *owner; /* reach back to owner ib_device */ 2627 }; 2628 2629 struct rdma_restrack_root; 2630 struct ib_device { 2631 /* Do not access @dma_device directly from ULP nor from HW drivers. */ 2632 struct device *dma_device; 2633 struct ib_device_ops ops; 2634 char name[IB_DEVICE_NAME_MAX]; 2635 struct rcu_head rcu_head; 2636 2637 struct list_head event_handler_list; 2638 /* Protects event_handler_list */ 2639 struct rw_semaphore event_handler_rwsem; 2640 2641 /* Protects QP's event_handler calls and open_qp list */ 2642 spinlock_t qp_open_list_lock; 2643 2644 struct rw_semaphore client_data_rwsem; 2645 struct xarray client_data; 2646 struct mutex unregistration_lock; 2647 2648 /* Synchronize GID, Pkey cache entries, subnet prefix, LMC */ 2649 rwlock_t cache_lock; 2650 /** 2651 * port_data is indexed by port number 2652 */ 2653 struct ib_port_data *port_data; 2654 2655 int num_comp_vectors; 2656 2657 union { 2658 struct device dev; 2659 struct ib_core_device coredev; 2660 }; 2661 2662 /* First group for device attributes, 2663 * Second group for driver provided attributes (optional). 2664 * It is NULL terminated array. 2665 */ 2666 const struct attribute_group *groups[3]; 2667 2668 u64 uverbs_cmd_mask; 2669 u64 uverbs_ex_cmd_mask; 2670 2671 char node_desc[IB_DEVICE_NODE_DESC_MAX]; 2672 __be64 node_guid; 2673 u32 local_dma_lkey; 2674 u16 is_switch:1; 2675 /* Indicates kernel verbs support, should not be used in drivers */ 2676 u16 kverbs_provider:1; 2677 /* CQ adaptive moderation (RDMA DIM) */ 2678 u16 use_cq_dim:1; 2679 u8 node_type; 2680 u8 phys_port_cnt; 2681 struct ib_device_attr attrs; 2682 struct attribute_group *hw_stats_ag; 2683 struct rdma_hw_stats *hw_stats; 2684 2685 #ifdef CONFIG_CGROUP_RDMA 2686 struct rdmacg_device cg_device; 2687 #endif 2688 2689 u32 index; 2690 struct rdma_restrack_root *res; 2691 2692 const struct uapi_definition *driver_def; 2693 2694 /* 2695 * Positive refcount indicates that the device is currently 2696 * registered and cannot be unregistered. 2697 */ 2698 refcount_t refcount; 2699 struct completion unreg_completion; 2700 struct work_struct unregistration_work; 2701 2702 const struct rdma_link_ops *link_ops; 2703 2704 /* Protects compat_devs xarray modifications */ 2705 struct mutex compat_devs_mutex; 2706 /* Maintains compat devices for each net namespace */ 2707 struct xarray compat_devs; 2708 2709 /* Used by iWarp CM */ 2710 char iw_ifname[IFNAMSIZ]; 2711 u32 iw_driver_flags; 2712 }; 2713 2714 struct ib_client_nl_info; 2715 struct ib_client { 2716 const char *name; 2717 void (*add) (struct ib_device *); 2718 void (*remove)(struct ib_device *, void *client_data); 2719 void (*rename)(struct ib_device *dev, void *client_data); 2720 int (*get_nl_info)(struct ib_device *ibdev, void *client_data, 2721 struct ib_client_nl_info *res); 2722 int (*get_global_nl_info)(struct ib_client_nl_info *res); 2723 2724 /* Returns the net_dev belonging to this ib_client and matching the 2725 * given parameters. 2726 * @dev: An RDMA device that the net_dev use for communication. 2727 * @port: A physical port number on the RDMA device. 2728 * @pkey: P_Key that the net_dev uses if applicable. 2729 * @gid: A GID that the net_dev uses to communicate. 2730 * @addr: An IP address the net_dev is configured with. 2731 * @client_data: The device's client data set by ib_set_client_data(). 2732 * 2733 * An ib_client that implements a net_dev on top of RDMA devices 2734 * (such as IP over IB) should implement this callback, allowing the 2735 * rdma_cm module to find the right net_dev for a given request. 2736 * 2737 * The caller is responsible for calling dev_put on the returned 2738 * netdev. */ 2739 struct net_device *(*get_net_dev_by_params)( 2740 struct ib_device *dev, 2741 u8 port, 2742 u16 pkey, 2743 const union ib_gid *gid, 2744 const struct sockaddr *addr, 2745 void *client_data); 2746 2747 refcount_t uses; 2748 struct completion uses_zero; 2749 u32 client_id; 2750 2751 /* kverbs are not required by the client */ 2752 u8 no_kverbs_req:1; 2753 }; 2754 2755 /* 2756 * IB block DMA iterator 2757 * 2758 * Iterates the DMA-mapped SGL in contiguous memory blocks aligned 2759 * to a HW supported page size. 2760 */ 2761 struct ib_block_iter { 2762 /* internal states */ 2763 struct scatterlist *__sg; /* sg holding the current aligned block */ 2764 dma_addr_t __dma_addr; /* unaligned DMA address of this block */ 2765 unsigned int __sg_nents; /* number of SG entries */ 2766 unsigned int __sg_advance; /* number of bytes to advance in sg in next step */ 2767 unsigned int __pg_bit; /* alignment of current block */ 2768 }; 2769 2770 struct ib_device *_ib_alloc_device(size_t size); 2771 #define ib_alloc_device(drv_struct, member) \ 2772 container_of(_ib_alloc_device(sizeof(struct drv_struct) + \ 2773 BUILD_BUG_ON_ZERO(offsetof( \ 2774 struct drv_struct, member))), \ 2775 struct drv_struct, member) 2776 2777 void ib_dealloc_device(struct ib_device *device); 2778 2779 void ib_get_device_fw_str(struct ib_device *device, char *str); 2780 2781 int ib_register_device(struct ib_device *device, const char *name); 2782 void ib_unregister_device(struct ib_device *device); 2783 void ib_unregister_driver(enum rdma_driver_id driver_id); 2784 void ib_unregister_device_and_put(struct ib_device *device); 2785 void ib_unregister_device_queued(struct ib_device *ib_dev); 2786 2787 int ib_register_client (struct ib_client *client); 2788 void ib_unregister_client(struct ib_client *client); 2789 2790 void __rdma_block_iter_start(struct ib_block_iter *biter, 2791 struct scatterlist *sglist, 2792 unsigned int nents, 2793 unsigned long pgsz); 2794 bool __rdma_block_iter_next(struct ib_block_iter *biter); 2795 2796 /** 2797 * rdma_block_iter_dma_address - get the aligned dma address of the current 2798 * block held by the block iterator. 2799 * @biter: block iterator holding the memory block 2800 */ 2801 static inline dma_addr_t 2802 rdma_block_iter_dma_address(struct ib_block_iter *biter) 2803 { 2804 return biter->__dma_addr & ~(BIT_ULL(biter->__pg_bit) - 1); 2805 } 2806 2807 /** 2808 * rdma_for_each_block - iterate over contiguous memory blocks of the sg list 2809 * @sglist: sglist to iterate over 2810 * @biter: block iterator holding the memory block 2811 * @nents: maximum number of sg entries to iterate over 2812 * @pgsz: best HW supported page size to use 2813 * 2814 * Callers may use rdma_block_iter_dma_address() to get each 2815 * blocks aligned DMA address. 2816 */ 2817 #define rdma_for_each_block(sglist, biter, nents, pgsz) \ 2818 for (__rdma_block_iter_start(biter, sglist, nents, \ 2819 pgsz); \ 2820 __rdma_block_iter_next(biter);) 2821 2822 /** 2823 * ib_get_client_data - Get IB client context 2824 * @device:Device to get context for 2825 * @client:Client to get context for 2826 * 2827 * ib_get_client_data() returns the client context data set with 2828 * ib_set_client_data(). This can only be called while the client is 2829 * registered to the device, once the ib_client remove() callback returns this 2830 * cannot be called. 2831 */ 2832 static inline void *ib_get_client_data(struct ib_device *device, 2833 struct ib_client *client) 2834 { 2835 return xa_load(&device->client_data, client->client_id); 2836 } 2837 void ib_set_client_data(struct ib_device *device, struct ib_client *client, 2838 void *data); 2839 void ib_set_device_ops(struct ib_device *device, 2840 const struct ib_device_ops *ops); 2841 2842 int rdma_user_mmap_io(struct ib_ucontext *ucontext, struct vm_area_struct *vma, 2843 unsigned long pfn, unsigned long size, pgprot_t prot, 2844 struct rdma_user_mmap_entry *entry); 2845 int rdma_user_mmap_entry_insert(struct ib_ucontext *ucontext, 2846 struct rdma_user_mmap_entry *entry, 2847 size_t length); 2848 int rdma_user_mmap_entry_insert_range(struct ib_ucontext *ucontext, 2849 struct rdma_user_mmap_entry *entry, 2850 size_t length, u32 min_pgoff, 2851 u32 max_pgoff); 2852 2853 struct rdma_user_mmap_entry * 2854 rdma_user_mmap_entry_get_pgoff(struct ib_ucontext *ucontext, 2855 unsigned long pgoff); 2856 struct rdma_user_mmap_entry * 2857 rdma_user_mmap_entry_get(struct ib_ucontext *ucontext, 2858 struct vm_area_struct *vma); 2859 void rdma_user_mmap_entry_put(struct rdma_user_mmap_entry *entry); 2860 2861 void rdma_user_mmap_entry_remove(struct rdma_user_mmap_entry *entry); 2862 2863 static inline int ib_copy_from_udata(void *dest, struct ib_udata *udata, size_t len) 2864 { 2865 return copy_from_user(dest, udata->inbuf, len) ? -EFAULT : 0; 2866 } 2867 2868 static inline int ib_copy_to_udata(struct ib_udata *udata, void *src, size_t len) 2869 { 2870 return copy_to_user(udata->outbuf, src, len) ? -EFAULT : 0; 2871 } 2872 2873 static inline bool ib_is_buffer_cleared(const void __user *p, 2874 size_t len) 2875 { 2876 bool ret; 2877 u8 *buf; 2878 2879 if (len > USHRT_MAX) 2880 return false; 2881 2882 buf = memdup_user(p, len); 2883 if (IS_ERR(buf)) 2884 return false; 2885 2886 ret = !memchr_inv(buf, 0, len); 2887 kfree(buf); 2888 return ret; 2889 } 2890 2891 static inline bool ib_is_udata_cleared(struct ib_udata *udata, 2892 size_t offset, 2893 size_t len) 2894 { 2895 return ib_is_buffer_cleared(udata->inbuf + offset, len); 2896 } 2897 2898 /** 2899 * ib_is_destroy_retryable - Check whether the uobject destruction 2900 * is retryable. 2901 * @ret: The initial destruction return code 2902 * @why: remove reason 2903 * @uobj: The uobject that is destroyed 2904 * 2905 * This function is a helper function that IB layer and low-level drivers 2906 * can use to consider whether the destruction of the given uobject is 2907 * retry-able. 2908 * It checks the original return code, if it wasn't success the destruction 2909 * is retryable according to the ucontext state (i.e. cleanup_retryable) and 2910 * the remove reason. (i.e. why). 2911 * Must be called with the object locked for destroy. 2912 */ 2913 static inline bool ib_is_destroy_retryable(int ret, enum rdma_remove_reason why, 2914 struct ib_uobject *uobj) 2915 { 2916 return ret && (why == RDMA_REMOVE_DESTROY || 2917 uobj->context->cleanup_retryable); 2918 } 2919 2920 /** 2921 * ib_destroy_usecnt - Called during destruction to check the usecnt 2922 * @usecnt: The usecnt atomic 2923 * @why: remove reason 2924 * @uobj: The uobject that is destroyed 2925 * 2926 * Non-zero usecnts will block destruction unless destruction was triggered by 2927 * a ucontext cleanup. 2928 */ 2929 static inline int ib_destroy_usecnt(atomic_t *usecnt, 2930 enum rdma_remove_reason why, 2931 struct ib_uobject *uobj) 2932 { 2933 if (atomic_read(usecnt) && ib_is_destroy_retryable(-EBUSY, why, uobj)) 2934 return -EBUSY; 2935 return 0; 2936 } 2937 2938 /** 2939 * ib_modify_qp_is_ok - Check that the supplied attribute mask 2940 * contains all required attributes and no attributes not allowed for 2941 * the given QP state transition. 2942 * @cur_state: Current QP state 2943 * @next_state: Next QP state 2944 * @type: QP type 2945 * @mask: Mask of supplied QP attributes 2946 * 2947 * This function is a helper function that a low-level driver's 2948 * modify_qp method can use to validate the consumer's input. It 2949 * checks that cur_state and next_state are valid QP states, that a 2950 * transition from cur_state to next_state is allowed by the IB spec, 2951 * and that the attribute mask supplied is allowed for the transition. 2952 */ 2953 bool ib_modify_qp_is_ok(enum ib_qp_state cur_state, enum ib_qp_state next_state, 2954 enum ib_qp_type type, enum ib_qp_attr_mask mask); 2955 2956 void ib_register_event_handler(struct ib_event_handler *event_handler); 2957 void ib_unregister_event_handler(struct ib_event_handler *event_handler); 2958 void ib_dispatch_event(const struct ib_event *event); 2959 2960 int ib_query_port(struct ib_device *device, 2961 u8 port_num, struct ib_port_attr *port_attr); 2962 2963 enum rdma_link_layer rdma_port_get_link_layer(struct ib_device *device, 2964 u8 port_num); 2965 2966 /** 2967 * rdma_cap_ib_switch - Check if the device is IB switch 2968 * @device: Device to check 2969 * 2970 * Device driver is responsible for setting is_switch bit on 2971 * in ib_device structure at init time. 2972 * 2973 * Return: true if the device is IB switch. 2974 */ 2975 static inline bool rdma_cap_ib_switch(const struct ib_device *device) 2976 { 2977 return device->is_switch; 2978 } 2979 2980 /** 2981 * rdma_start_port - Return the first valid port number for the device 2982 * specified 2983 * 2984 * @device: Device to be checked 2985 * 2986 * Return start port number 2987 */ 2988 static inline u8 rdma_start_port(const struct ib_device *device) 2989 { 2990 return rdma_cap_ib_switch(device) ? 0 : 1; 2991 } 2992 2993 /** 2994 * rdma_for_each_port - Iterate over all valid port numbers of the IB device 2995 * @device - The struct ib_device * to iterate over 2996 * @iter - The unsigned int to store the port number 2997 */ 2998 #define rdma_for_each_port(device, iter) \ 2999 for (iter = rdma_start_port(device + BUILD_BUG_ON_ZERO(!__same_type( \ 3000 unsigned int, iter))); \ 3001 iter <= rdma_end_port(device); (iter)++) 3002 3003 /** 3004 * rdma_end_port - Return the last valid port number for the device 3005 * specified 3006 * 3007 * @device: Device to be checked 3008 * 3009 * Return last port number 3010 */ 3011 static inline u8 rdma_end_port(const struct ib_device *device) 3012 { 3013 return rdma_cap_ib_switch(device) ? 0 : device->phys_port_cnt; 3014 } 3015 3016 static inline int rdma_is_port_valid(const struct ib_device *device, 3017 unsigned int port) 3018 { 3019 return (port >= rdma_start_port(device) && 3020 port <= rdma_end_port(device)); 3021 } 3022 3023 static inline bool rdma_is_grh_required(const struct ib_device *device, 3024 u8 port_num) 3025 { 3026 return device->port_data[port_num].immutable.core_cap_flags & 3027 RDMA_CORE_PORT_IB_GRH_REQUIRED; 3028 } 3029 3030 static inline bool rdma_protocol_ib(const struct ib_device *device, u8 port_num) 3031 { 3032 return device->port_data[port_num].immutable.core_cap_flags & 3033 RDMA_CORE_CAP_PROT_IB; 3034 } 3035 3036 static inline bool rdma_protocol_roce(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_ROCE | RDMA_CORE_CAP_PROT_ROCE_UDP_ENCAP); 3040 } 3041 3042 static inline bool rdma_protocol_roce_udp_encap(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_UDP_ENCAP; 3046 } 3047 3048 static inline bool rdma_protocol_roce_eth_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; 3052 } 3053 3054 static inline bool rdma_protocol_iwarp(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_IWARP; 3058 } 3059 3060 static inline bool rdma_ib_or_roce(const struct ib_device *device, u8 port_num) 3061 { 3062 return rdma_protocol_ib(device, port_num) || 3063 rdma_protocol_roce(device, port_num); 3064 } 3065 3066 static inline bool rdma_protocol_raw_packet(const struct ib_device *device, u8 port_num) 3067 { 3068 return device->port_data[port_num].immutable.core_cap_flags & 3069 RDMA_CORE_CAP_PROT_RAW_PACKET; 3070 } 3071 3072 static inline bool rdma_protocol_usnic(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_USNIC; 3076 } 3077 3078 /** 3079 * rdma_cap_ib_mad - Check if the port of a device supports Infiniband 3080 * Management Datagrams. 3081 * @device: Device to check 3082 * @port_num: Port number to check 3083 * 3084 * Management Datagrams (MAD) are a required part of the InfiniBand 3085 * specification and are supported on all InfiniBand devices. A slightly 3086 * extended version are also supported on OPA interfaces. 3087 * 3088 * Return: true if the port supports sending/receiving of MAD packets. 3089 */ 3090 static inline bool rdma_cap_ib_mad(const struct ib_device *device, u8 port_num) 3091 { 3092 return device->port_data[port_num].immutable.core_cap_flags & 3093 RDMA_CORE_CAP_IB_MAD; 3094 } 3095 3096 /** 3097 * rdma_cap_opa_mad - Check if the port of device provides support for OPA 3098 * Management Datagrams. 3099 * @device: Device to check 3100 * @port_num: Port number to check 3101 * 3102 * Intel OmniPath devices extend and/or replace the InfiniBand Management 3103 * datagrams with their own versions. These OPA MADs share many but not all of 3104 * the characteristics of InfiniBand MADs. 3105 * 3106 * OPA MADs differ in the following ways: 3107 * 3108 * 1) MADs are variable size up to 2K 3109 * IBTA defined MADs remain fixed at 256 bytes 3110 * 2) OPA SMPs must carry valid PKeys 3111 * 3) OPA SMP packets are a different format 3112 * 3113 * Return: true if the port supports OPA MAD packet formats. 3114 */ 3115 static inline bool rdma_cap_opa_mad(struct ib_device *device, u8 port_num) 3116 { 3117 return device->port_data[port_num].immutable.core_cap_flags & 3118 RDMA_CORE_CAP_OPA_MAD; 3119 } 3120 3121 /** 3122 * rdma_cap_ib_smi - Check if the port of a device provides an Infiniband 3123 * Subnet Management Agent (SMA) on the Subnet Management Interface (SMI). 3124 * @device: Device to check 3125 * @port_num: Port number to check 3126 * 3127 * Each InfiniBand node is required to provide a Subnet Management Agent 3128 * that the subnet manager can access. Prior to the fabric being fully 3129 * configured by the subnet manager, the SMA is accessed via a well known 3130 * interface called the Subnet Management Interface (SMI). This interface 3131 * uses directed route packets to communicate with the SM to get around the 3132 * chicken and egg problem of the SM needing to know what's on the fabric 3133 * in order to configure the fabric, and needing to configure the fabric in 3134 * order to send packets to the devices on the fabric. These directed 3135 * route packets do not need the fabric fully configured in order to reach 3136 * their destination. The SMI is the only method allowed to send 3137 * directed route packets on an InfiniBand fabric. 3138 * 3139 * Return: true if the port provides an SMI. 3140 */ 3141 static inline bool rdma_cap_ib_smi(const struct ib_device *device, u8 port_num) 3142 { 3143 return device->port_data[port_num].immutable.core_cap_flags & 3144 RDMA_CORE_CAP_IB_SMI; 3145 } 3146 3147 /** 3148 * rdma_cap_ib_cm - Check if the port of device has the capability Infiniband 3149 * Communication Manager. 3150 * @device: Device to check 3151 * @port_num: Port number to check 3152 * 3153 * The InfiniBand Communication Manager is one of many pre-defined General 3154 * Service Agents (GSA) that are accessed via the General Service 3155 * Interface (GSI). It's role is to facilitate establishment of connections 3156 * between nodes as well as other management related tasks for established 3157 * connections. 3158 * 3159 * Return: true if the port supports an IB CM (this does not guarantee that 3160 * a CM is actually running however). 3161 */ 3162 static inline bool rdma_cap_ib_cm(const struct ib_device *device, u8 port_num) 3163 { 3164 return device->port_data[port_num].immutable.core_cap_flags & 3165 RDMA_CORE_CAP_IB_CM; 3166 } 3167 3168 /** 3169 * rdma_cap_iw_cm - Check if the port of device has the capability IWARP 3170 * Communication Manager. 3171 * @device: Device to check 3172 * @port_num: Port number to check 3173 * 3174 * Similar to above, but specific to iWARP connections which have a different 3175 * managment protocol than InfiniBand. 3176 * 3177 * Return: true if the port supports an iWARP CM (this does not guarantee that 3178 * a CM is actually running however). 3179 */ 3180 static inline bool rdma_cap_iw_cm(const struct ib_device *device, u8 port_num) 3181 { 3182 return device->port_data[port_num].immutable.core_cap_flags & 3183 RDMA_CORE_CAP_IW_CM; 3184 } 3185 3186 /** 3187 * rdma_cap_ib_sa - Check if the port of device has the capability Infiniband 3188 * Subnet Administration. 3189 * @device: Device to check 3190 * @port_num: Port number to check 3191 * 3192 * An InfiniBand Subnet Administration (SA) service is a pre-defined General 3193 * Service Agent (GSA) provided by the Subnet Manager (SM). On InfiniBand 3194 * fabrics, devices should resolve routes to other hosts by contacting the 3195 * SA to query the proper route. 3196 * 3197 * Return: true if the port should act as a client to the fabric Subnet 3198 * Administration interface. This does not imply that the SA service is 3199 * running locally. 3200 */ 3201 static inline bool rdma_cap_ib_sa(const struct ib_device *device, u8 port_num) 3202 { 3203 return device->port_data[port_num].immutable.core_cap_flags & 3204 RDMA_CORE_CAP_IB_SA; 3205 } 3206 3207 /** 3208 * rdma_cap_ib_mcast - Check if the port of device has the capability Infiniband 3209 * Multicast. 3210 * @device: Device to check 3211 * @port_num: Port number to check 3212 * 3213 * InfiniBand multicast registration is more complex than normal IPv4 or 3214 * IPv6 multicast registration. Each Host Channel Adapter must register 3215 * with the Subnet Manager when it wishes to join a multicast group. It 3216 * should do so only once regardless of how many queue pairs it subscribes 3217 * to this group. And it should leave the group only after all queue pairs 3218 * attached to the group have been detached. 3219 * 3220 * Return: true if the port must undertake the additional adminstrative 3221 * overhead of registering/unregistering with the SM and tracking of the 3222 * total number of queue pairs attached to the multicast group. 3223 */ 3224 static inline bool rdma_cap_ib_mcast(const struct ib_device *device, u8 port_num) 3225 { 3226 return rdma_cap_ib_sa(device, port_num); 3227 } 3228 3229 /** 3230 * rdma_cap_af_ib - Check if the port of device has the capability 3231 * Native Infiniband Address. 3232 * @device: Device to check 3233 * @port_num: Port number to check 3234 * 3235 * InfiniBand addressing uses a port's GUID + Subnet Prefix to make a default 3236 * GID. RoCE uses a different mechanism, but still generates a GID via 3237 * a prescribed mechanism and port specific data. 3238 * 3239 * Return: true if the port uses a GID address to identify devices on the 3240 * network. 3241 */ 3242 static inline bool rdma_cap_af_ib(const struct ib_device *device, u8 port_num) 3243 { 3244 return device->port_data[port_num].immutable.core_cap_flags & 3245 RDMA_CORE_CAP_AF_IB; 3246 } 3247 3248 /** 3249 * rdma_cap_eth_ah - Check if the port of device has the capability 3250 * Ethernet Address Handle. 3251 * @device: Device to check 3252 * @port_num: Port number to check 3253 * 3254 * RoCE is InfiniBand over Ethernet, and it uses a well defined technique 3255 * to fabricate GIDs over Ethernet/IP specific addresses native to the 3256 * port. Normally, packet headers are generated by the sending host 3257 * adapter, but when sending connectionless datagrams, we must manually 3258 * inject the proper headers for the fabric we are communicating over. 3259 * 3260 * Return: true if we are running as a RoCE port and must force the 3261 * addition of a Global Route Header built from our Ethernet Address 3262 * Handle into our header list for connectionless packets. 3263 */ 3264 static inline bool rdma_cap_eth_ah(const struct ib_device *device, u8 port_num) 3265 { 3266 return device->port_data[port_num].immutable.core_cap_flags & 3267 RDMA_CORE_CAP_ETH_AH; 3268 } 3269 3270 /** 3271 * rdma_cap_opa_ah - Check if the port of device supports 3272 * OPA Address handles 3273 * @device: Device to check 3274 * @port_num: Port number to check 3275 * 3276 * Return: true if we are running on an OPA device which supports 3277 * the extended OPA addressing. 3278 */ 3279 static inline bool rdma_cap_opa_ah(struct ib_device *device, u8 port_num) 3280 { 3281 return (device->port_data[port_num].immutable.core_cap_flags & 3282 RDMA_CORE_CAP_OPA_AH) == RDMA_CORE_CAP_OPA_AH; 3283 } 3284 3285 /** 3286 * rdma_max_mad_size - Return the max MAD size required by this RDMA Port. 3287 * 3288 * @device: Device 3289 * @port_num: Port number 3290 * 3291 * This MAD size includes the MAD headers and MAD payload. No other headers 3292 * are included. 3293 * 3294 * Return the max MAD size required by the Port. Will return 0 if the port 3295 * does not support MADs 3296 */ 3297 static inline size_t rdma_max_mad_size(const struct ib_device *device, u8 port_num) 3298 { 3299 return device->port_data[port_num].immutable.max_mad_size; 3300 } 3301 3302 /** 3303 * rdma_cap_roce_gid_table - Check if the port of device uses roce_gid_table 3304 * @device: Device to check 3305 * @port_num: Port number to check 3306 * 3307 * RoCE GID table mechanism manages the various GIDs for a device. 3308 * 3309 * NOTE: if allocating the port's GID table has failed, this call will still 3310 * return true, but any RoCE GID table API will fail. 3311 * 3312 * Return: true if the port uses RoCE GID table mechanism in order to manage 3313 * its GIDs. 3314 */ 3315 static inline bool rdma_cap_roce_gid_table(const struct ib_device *device, 3316 u8 port_num) 3317 { 3318 return rdma_protocol_roce(device, port_num) && 3319 device->ops.add_gid && device->ops.del_gid; 3320 } 3321 3322 /* 3323 * Check if the device supports READ W/ INVALIDATE. 3324 */ 3325 static inline bool rdma_cap_read_inv(struct ib_device *dev, u32 port_num) 3326 { 3327 /* 3328 * iWarp drivers must support READ W/ INVALIDATE. No other protocol 3329 * has support for it yet. 3330 */ 3331 return rdma_protocol_iwarp(dev, port_num); 3332 } 3333 3334 /** 3335 * rdma_find_pg_bit - Find page bit given address and HW supported page sizes 3336 * 3337 * @addr: address 3338 * @pgsz_bitmap: bitmap of HW supported page sizes 3339 */ 3340 static inline unsigned int rdma_find_pg_bit(unsigned long addr, 3341 unsigned long pgsz_bitmap) 3342 { 3343 unsigned long align; 3344 unsigned long pgsz; 3345 3346 align = addr & -addr; 3347 3348 /* Find page bit such that addr is aligned to the highest supported 3349 * HW page size 3350 */ 3351 pgsz = pgsz_bitmap & ~(-align << 1); 3352 if (!pgsz) 3353 return __ffs(pgsz_bitmap); 3354 3355 return __fls(pgsz); 3356 } 3357 3358 int ib_set_vf_link_state(struct ib_device *device, int vf, u8 port, 3359 int state); 3360 int ib_get_vf_config(struct ib_device *device, int vf, u8 port, 3361 struct ifla_vf_info *info); 3362 int ib_get_vf_stats(struct ib_device *device, int vf, u8 port, 3363 struct ifla_vf_stats *stats); 3364 int ib_get_vf_guid(struct ib_device *device, int vf, u8 port, 3365 struct ifla_vf_guid *node_guid, 3366 struct ifla_vf_guid *port_guid); 3367 int ib_set_vf_guid(struct ib_device *device, int vf, u8 port, u64 guid, 3368 int type); 3369 3370 int ib_query_pkey(struct ib_device *device, 3371 u8 port_num, u16 index, u16 *pkey); 3372 3373 int ib_modify_device(struct ib_device *device, 3374 int device_modify_mask, 3375 struct ib_device_modify *device_modify); 3376 3377 int ib_modify_port(struct ib_device *device, 3378 u8 port_num, int port_modify_mask, 3379 struct ib_port_modify *port_modify); 3380 3381 int ib_find_gid(struct ib_device *device, union ib_gid *gid, 3382 u8 *port_num, u16 *index); 3383 3384 int ib_find_pkey(struct ib_device *device, 3385 u8 port_num, u16 pkey, u16 *index); 3386 3387 enum ib_pd_flags { 3388 /* 3389 * Create a memory registration for all memory in the system and place 3390 * the rkey for it into pd->unsafe_global_rkey. This can be used by 3391 * ULPs to avoid the overhead of dynamic MRs. 3392 * 3393 * This flag is generally considered unsafe and must only be used in 3394 * extremly trusted environments. Every use of it will log a warning 3395 * in the kernel log. 3396 */ 3397 IB_PD_UNSAFE_GLOBAL_RKEY = 0x01, 3398 }; 3399 3400 struct ib_pd *__ib_alloc_pd(struct ib_device *device, unsigned int flags, 3401 const char *caller); 3402 3403 #define ib_alloc_pd(device, flags) \ 3404 __ib_alloc_pd((device), (flags), KBUILD_MODNAME) 3405 3406 /** 3407 * ib_dealloc_pd_user - Deallocate kernel/user PD 3408 * @pd: The protection domain 3409 * @udata: Valid user data or NULL for kernel objects 3410 */ 3411 void ib_dealloc_pd_user(struct ib_pd *pd, struct ib_udata *udata); 3412 3413 /** 3414 * ib_dealloc_pd - Deallocate kernel PD 3415 * @pd: The protection domain 3416 * 3417 * NOTE: for user PD use ib_dealloc_pd_user with valid udata! 3418 */ 3419 static inline void ib_dealloc_pd(struct ib_pd *pd) 3420 { 3421 ib_dealloc_pd_user(pd, NULL); 3422 } 3423 3424 enum rdma_create_ah_flags { 3425 /* In a sleepable context */ 3426 RDMA_CREATE_AH_SLEEPABLE = BIT(0), 3427 }; 3428 3429 /** 3430 * rdma_create_ah - Creates an address handle for the given address vector. 3431 * @pd: The protection domain associated with the address handle. 3432 * @ah_attr: The attributes of the address vector. 3433 * @flags: Create address handle flags (see enum rdma_create_ah_flags). 3434 * 3435 * The address handle is used to reference a local or global destination 3436 * in all UD QP post sends. 3437 */ 3438 struct ib_ah *rdma_create_ah(struct ib_pd *pd, struct rdma_ah_attr *ah_attr, 3439 u32 flags); 3440 3441 /** 3442 * rdma_create_user_ah - Creates an address handle for the given address vector. 3443 * It resolves destination mac address for ah attribute of RoCE type. 3444 * @pd: The protection domain associated with the address handle. 3445 * @ah_attr: The attributes of the address vector. 3446 * @udata: pointer to user's input output buffer information need by 3447 * provider driver. 3448 * 3449 * It returns 0 on success and returns appropriate error code on error. 3450 * The address handle is used to reference a local or global destination 3451 * in all UD QP post sends. 3452 */ 3453 struct ib_ah *rdma_create_user_ah(struct ib_pd *pd, 3454 struct rdma_ah_attr *ah_attr, 3455 struct ib_udata *udata); 3456 /** 3457 * ib_get_gids_from_rdma_hdr - Get sgid and dgid from GRH or IPv4 header 3458 * work completion. 3459 * @hdr: the L3 header to parse 3460 * @net_type: type of header to parse 3461 * @sgid: place to store source gid 3462 * @dgid: place to store destination gid 3463 */ 3464 int ib_get_gids_from_rdma_hdr(const union rdma_network_hdr *hdr, 3465 enum rdma_network_type net_type, 3466 union ib_gid *sgid, union ib_gid *dgid); 3467 3468 /** 3469 * ib_get_rdma_header_version - Get the header version 3470 * @hdr: the L3 header to parse 3471 */ 3472 int ib_get_rdma_header_version(const union rdma_network_hdr *hdr); 3473 3474 /** 3475 * ib_init_ah_attr_from_wc - Initializes address handle attributes from a 3476 * work completion. 3477 * @device: Device on which the received message arrived. 3478 * @port_num: Port on which the received message arrived. 3479 * @wc: Work completion associated with the received message. 3480 * @grh: References the received global route header. This parameter is 3481 * ignored unless the work completion indicates that the GRH is valid. 3482 * @ah_attr: Returned attributes that can be used when creating an address 3483 * handle for replying to the message. 3484 * When ib_init_ah_attr_from_wc() returns success, 3485 * (a) for IB link layer it optionally contains a reference to SGID attribute 3486 * when GRH is present for IB link layer. 3487 * (b) for RoCE link layer it contains a reference to SGID attribute. 3488 * User must invoke rdma_cleanup_ah_attr_gid_attr() to release reference to SGID 3489 * attributes which are initialized using ib_init_ah_attr_from_wc(). 3490 * 3491 */ 3492 int ib_init_ah_attr_from_wc(struct ib_device *device, u8 port_num, 3493 const struct ib_wc *wc, const struct ib_grh *grh, 3494 struct rdma_ah_attr *ah_attr); 3495 3496 /** 3497 * ib_create_ah_from_wc - Creates an address handle associated with the 3498 * sender of the specified work completion. 3499 * @pd: The protection domain associated with the address handle. 3500 * @wc: Work completion information associated with a received message. 3501 * @grh: References the received global route header. This parameter is 3502 * ignored unless the work completion indicates that the GRH is valid. 3503 * @port_num: The outbound port number to associate with the address. 3504 * 3505 * The address handle is used to reference a local or global destination 3506 * in all UD QP post sends. 3507 */ 3508 struct ib_ah *ib_create_ah_from_wc(struct ib_pd *pd, const struct ib_wc *wc, 3509 const struct ib_grh *grh, u8 port_num); 3510 3511 /** 3512 * rdma_modify_ah - Modifies the address vector associated with an address 3513 * handle. 3514 * @ah: The address handle to modify. 3515 * @ah_attr: The new address vector attributes to associate with the 3516 * address handle. 3517 */ 3518 int rdma_modify_ah(struct ib_ah *ah, struct rdma_ah_attr *ah_attr); 3519 3520 /** 3521 * rdma_query_ah - Queries the address vector associated with an address 3522 * handle. 3523 * @ah: The address handle to query. 3524 * @ah_attr: The address vector attributes associated with the address 3525 * handle. 3526 */ 3527 int rdma_query_ah(struct ib_ah *ah, struct rdma_ah_attr *ah_attr); 3528 3529 enum rdma_destroy_ah_flags { 3530 /* In a sleepable context */ 3531 RDMA_DESTROY_AH_SLEEPABLE = BIT(0), 3532 }; 3533 3534 /** 3535 * rdma_destroy_ah_user - Destroys an address handle. 3536 * @ah: The address handle to destroy. 3537 * @flags: Destroy address handle flags (see enum rdma_destroy_ah_flags). 3538 * @udata: Valid user data or NULL for kernel objects 3539 */ 3540 int rdma_destroy_ah_user(struct ib_ah *ah, u32 flags, struct ib_udata *udata); 3541 3542 /** 3543 * rdma_destroy_ah - Destroys an kernel address handle. 3544 * @ah: The address handle to destroy. 3545 * @flags: Destroy address handle flags (see enum rdma_destroy_ah_flags). 3546 * 3547 * NOTE: for user ah use rdma_destroy_ah_user with valid udata! 3548 */ 3549 static inline int rdma_destroy_ah(struct ib_ah *ah, u32 flags) 3550 { 3551 return rdma_destroy_ah_user(ah, flags, NULL); 3552 } 3553 3554 /** 3555 * ib_create_srq - Creates a SRQ associated with the specified protection 3556 * domain. 3557 * @pd: The protection domain associated with the SRQ. 3558 * @srq_init_attr: A list of initial attributes required to create the 3559 * SRQ. If SRQ creation succeeds, then the attributes are updated to 3560 * the actual capabilities of the created SRQ. 3561 * 3562 * srq_attr->max_wr and srq_attr->max_sge are read the determine the 3563 * requested size of the SRQ, and set to the actual values allocated 3564 * on return. If ib_create_srq() succeeds, then max_wr and max_sge 3565 * will always be at least as large as the requested values. 3566 */ 3567 struct ib_srq *ib_create_srq(struct ib_pd *pd, 3568 struct ib_srq_init_attr *srq_init_attr); 3569 3570 /** 3571 * ib_modify_srq - Modifies the attributes for the specified SRQ. 3572 * @srq: The SRQ to modify. 3573 * @srq_attr: On input, specifies the SRQ attributes to modify. On output, 3574 * the current values of selected SRQ attributes are returned. 3575 * @srq_attr_mask: A bit-mask used to specify which attributes of the SRQ 3576 * are being modified. 3577 * 3578 * The mask may contain IB_SRQ_MAX_WR to resize the SRQ and/or 3579 * IB_SRQ_LIMIT to set the SRQ's limit and request notification when 3580 * the number of receives queued drops below the limit. 3581 */ 3582 int ib_modify_srq(struct ib_srq *srq, 3583 struct ib_srq_attr *srq_attr, 3584 enum ib_srq_attr_mask srq_attr_mask); 3585 3586 /** 3587 * ib_query_srq - Returns the attribute list and current values for the 3588 * specified SRQ. 3589 * @srq: The SRQ to query. 3590 * @srq_attr: The attributes of the specified SRQ. 3591 */ 3592 int ib_query_srq(struct ib_srq *srq, 3593 struct ib_srq_attr *srq_attr); 3594 3595 /** 3596 * ib_destroy_srq_user - Destroys the specified SRQ. 3597 * @srq: The SRQ to destroy. 3598 * @udata: Valid user data or NULL for kernel objects 3599 */ 3600 int ib_destroy_srq_user(struct ib_srq *srq, struct ib_udata *udata); 3601 3602 /** 3603 * ib_destroy_srq - Destroys the specified kernel SRQ. 3604 * @srq: The SRQ to destroy. 3605 * 3606 * NOTE: for user srq use ib_destroy_srq_user with valid udata! 3607 */ 3608 static inline int ib_destroy_srq(struct ib_srq *srq) 3609 { 3610 return ib_destroy_srq_user(srq, NULL); 3611 } 3612 3613 /** 3614 * ib_post_srq_recv - Posts a list of work requests to the specified SRQ. 3615 * @srq: The SRQ to post the work request on. 3616 * @recv_wr: A list of work requests to post on the receive queue. 3617 * @bad_recv_wr: On an immediate failure, this parameter will reference 3618 * the work request that failed to be posted on the QP. 3619 */ 3620 static inline int ib_post_srq_recv(struct ib_srq *srq, 3621 const struct ib_recv_wr *recv_wr, 3622 const struct ib_recv_wr **bad_recv_wr) 3623 { 3624 const struct ib_recv_wr *dummy; 3625 3626 return srq->device->ops.post_srq_recv(srq, recv_wr, 3627 bad_recv_wr ? : &dummy); 3628 } 3629 3630 struct ib_qp *ib_create_qp(struct ib_pd *pd, 3631 struct ib_qp_init_attr *qp_init_attr); 3632 3633 /** 3634 * ib_modify_qp_with_udata - Modifies the attributes for the specified QP. 3635 * @qp: The QP to modify. 3636 * @attr: On input, specifies the QP attributes to modify. On output, 3637 * the current values of selected QP attributes are returned. 3638 * @attr_mask: A bit-mask used to specify which attributes of the QP 3639 * are being modified. 3640 * @udata: pointer to user's input output buffer information 3641 * are being modified. 3642 * It returns 0 on success and returns appropriate error code on error. 3643 */ 3644 int ib_modify_qp_with_udata(struct ib_qp *qp, 3645 struct ib_qp_attr *attr, 3646 int attr_mask, 3647 struct ib_udata *udata); 3648 3649 /** 3650 * ib_modify_qp - Modifies the attributes for the specified QP and then 3651 * transitions the QP to the given state. 3652 * @qp: The QP to modify. 3653 * @qp_attr: On input, specifies the QP attributes to modify. On output, 3654 * the current values of selected QP attributes are returned. 3655 * @qp_attr_mask: A bit-mask used to specify which attributes of the QP 3656 * are being modified. 3657 */ 3658 int ib_modify_qp(struct ib_qp *qp, 3659 struct ib_qp_attr *qp_attr, 3660 int qp_attr_mask); 3661 3662 /** 3663 * ib_query_qp - Returns the attribute list and current values for the 3664 * specified QP. 3665 * @qp: The QP to query. 3666 * @qp_attr: The attributes of the specified QP. 3667 * @qp_attr_mask: A bit-mask used to select specific attributes to query. 3668 * @qp_init_attr: Additional attributes of the selected QP. 3669 * 3670 * The qp_attr_mask may be used to limit the query to gathering only the 3671 * selected attributes. 3672 */ 3673 int ib_query_qp(struct ib_qp *qp, 3674 struct ib_qp_attr *qp_attr, 3675 int qp_attr_mask, 3676 struct ib_qp_init_attr *qp_init_attr); 3677 3678 /** 3679 * ib_destroy_qp - Destroys the specified QP. 3680 * @qp: The QP to destroy. 3681 * @udata: Valid udata or NULL for kernel objects 3682 */ 3683 int ib_destroy_qp_user(struct ib_qp *qp, struct ib_udata *udata); 3684 3685 /** 3686 * ib_destroy_qp - Destroys the specified kernel QP. 3687 * @qp: The QP to destroy. 3688 * 3689 * NOTE: for user qp use ib_destroy_qp_user with valid udata! 3690 */ 3691 static inline int ib_destroy_qp(struct ib_qp *qp) 3692 { 3693 return ib_destroy_qp_user(qp, NULL); 3694 } 3695 3696 /** 3697 * ib_open_qp - Obtain a reference to an existing sharable QP. 3698 * @xrcd - XRC domain 3699 * @qp_open_attr: Attributes identifying the QP to open. 3700 * 3701 * Returns a reference to a sharable QP. 3702 */ 3703 struct ib_qp *ib_open_qp(struct ib_xrcd *xrcd, 3704 struct ib_qp_open_attr *qp_open_attr); 3705 3706 /** 3707 * ib_close_qp - Release an external reference to a QP. 3708 * @qp: The QP handle to release 3709 * 3710 * The opened QP handle is released by the caller. The underlying 3711 * shared QP is not destroyed until all internal references are released. 3712 */ 3713 int ib_close_qp(struct ib_qp *qp); 3714 3715 /** 3716 * ib_post_send - Posts a list of work requests to the send queue of 3717 * the specified QP. 3718 * @qp: The QP to post the work request on. 3719 * @send_wr: A list of work requests to post on the send queue. 3720 * @bad_send_wr: On an immediate failure, this parameter will reference 3721 * the work request that failed to be posted on the QP. 3722 * 3723 * While IBA Vol. 1 section 11.4.1.1 specifies that if an immediate 3724 * error is returned, the QP state shall not be affected, 3725 * ib_post_send() will return an immediate error after queueing any 3726 * earlier work requests in the list. 3727 */ 3728 static inline int ib_post_send(struct ib_qp *qp, 3729 const struct ib_send_wr *send_wr, 3730 const struct ib_send_wr **bad_send_wr) 3731 { 3732 const struct ib_send_wr *dummy; 3733 3734 return qp->device->ops.post_send(qp, send_wr, bad_send_wr ? : &dummy); 3735 } 3736 3737 /** 3738 * ib_post_recv - Posts a list of work requests to the receive queue of 3739 * the specified QP. 3740 * @qp: The QP to post the work request on. 3741 * @recv_wr: A list of work requests to post on the receive queue. 3742 * @bad_recv_wr: On an immediate failure, this parameter will reference 3743 * the work request that failed to be posted on the QP. 3744 */ 3745 static inline int ib_post_recv(struct ib_qp *qp, 3746 const struct ib_recv_wr *recv_wr, 3747 const struct ib_recv_wr **bad_recv_wr) 3748 { 3749 const struct ib_recv_wr *dummy; 3750 3751 return qp->device->ops.post_recv(qp, recv_wr, bad_recv_wr ? : &dummy); 3752 } 3753 3754 struct ib_cq *__ib_alloc_cq_user(struct ib_device *dev, void *private, 3755 int nr_cqe, int comp_vector, 3756 enum ib_poll_context poll_ctx, 3757 const char *caller, struct ib_udata *udata); 3758 3759 /** 3760 * ib_alloc_cq_user: Allocate kernel/user CQ 3761 * @dev: The IB device 3762 * @private: Private data attached to the CQE 3763 * @nr_cqe: Number of CQEs in the CQ 3764 * @comp_vector: Completion vector used for the IRQs 3765 * @poll_ctx: Context used for polling the CQ 3766 * @udata: Valid user data or NULL for kernel objects 3767 */ 3768 static inline struct ib_cq *ib_alloc_cq_user(struct ib_device *dev, 3769 void *private, int nr_cqe, 3770 int comp_vector, 3771 enum ib_poll_context poll_ctx, 3772 struct ib_udata *udata) 3773 { 3774 return __ib_alloc_cq_user(dev, private, nr_cqe, comp_vector, poll_ctx, 3775 KBUILD_MODNAME, udata); 3776 } 3777 3778 /** 3779 * ib_alloc_cq: Allocate kernel CQ 3780 * @dev: The IB device 3781 * @private: Private data attached to the CQE 3782 * @nr_cqe: Number of CQEs in the CQ 3783 * @comp_vector: Completion vector used for the IRQs 3784 * @poll_ctx: Context used for polling the CQ 3785 * 3786 * NOTE: for user cq use ib_alloc_cq_user with valid udata! 3787 */ 3788 static inline struct ib_cq *ib_alloc_cq(struct ib_device *dev, void *private, 3789 int nr_cqe, int comp_vector, 3790 enum ib_poll_context poll_ctx) 3791 { 3792 return ib_alloc_cq_user(dev, private, nr_cqe, comp_vector, poll_ctx, 3793 NULL); 3794 } 3795 3796 struct ib_cq *__ib_alloc_cq_any(struct ib_device *dev, void *private, 3797 int nr_cqe, enum ib_poll_context poll_ctx, 3798 const char *caller); 3799 3800 /** 3801 * ib_alloc_cq_any: Allocate kernel CQ 3802 * @dev: The IB device 3803 * @private: Private data attached to the CQE 3804 * @nr_cqe: Number of CQEs in the CQ 3805 * @poll_ctx: Context used for polling the CQ 3806 */ 3807 static inline struct ib_cq *ib_alloc_cq_any(struct ib_device *dev, 3808 void *private, int nr_cqe, 3809 enum ib_poll_context poll_ctx) 3810 { 3811 return __ib_alloc_cq_any(dev, private, nr_cqe, poll_ctx, 3812 KBUILD_MODNAME); 3813 } 3814 3815 /** 3816 * ib_free_cq_user - Free kernel/user CQ 3817 * @cq: The CQ to free 3818 * @udata: Valid user data or NULL for kernel objects 3819 */ 3820 void ib_free_cq_user(struct ib_cq *cq, struct ib_udata *udata); 3821 3822 /** 3823 * ib_free_cq - Free kernel CQ 3824 * @cq: The CQ to free 3825 * 3826 * NOTE: for user cq use ib_free_cq_user with valid udata! 3827 */ 3828 static inline void ib_free_cq(struct ib_cq *cq) 3829 { 3830 ib_free_cq_user(cq, NULL); 3831 } 3832 3833 int ib_process_cq_direct(struct ib_cq *cq, int budget); 3834 3835 /** 3836 * ib_create_cq - Creates a CQ on the specified device. 3837 * @device: The device on which to create the CQ. 3838 * @comp_handler: A user-specified callback that is invoked when a 3839 * completion event occurs on the CQ. 3840 * @event_handler: A user-specified callback that is invoked when an 3841 * asynchronous event not associated with a completion occurs on the CQ. 3842 * @cq_context: Context associated with the CQ returned to the user via 3843 * the associated completion and event handlers. 3844 * @cq_attr: The attributes the CQ should be created upon. 3845 * 3846 * Users can examine the cq structure to determine the actual CQ size. 3847 */ 3848 struct ib_cq *__ib_create_cq(struct ib_device *device, 3849 ib_comp_handler comp_handler, 3850 void (*event_handler)(struct ib_event *, void *), 3851 void *cq_context, 3852 const struct ib_cq_init_attr *cq_attr, 3853 const char *caller); 3854 #define ib_create_cq(device, cmp_hndlr, evt_hndlr, cq_ctxt, cq_attr) \ 3855 __ib_create_cq((device), (cmp_hndlr), (evt_hndlr), (cq_ctxt), (cq_attr), KBUILD_MODNAME) 3856 3857 /** 3858 * ib_resize_cq - Modifies the capacity of the CQ. 3859 * @cq: The CQ to resize. 3860 * @cqe: The minimum size of the CQ. 3861 * 3862 * Users can examine the cq structure to determine the actual CQ size. 3863 */ 3864 int ib_resize_cq(struct ib_cq *cq, int cqe); 3865 3866 /** 3867 * rdma_set_cq_moderation - Modifies moderation params of the CQ 3868 * @cq: The CQ to modify. 3869 * @cq_count: number of CQEs that will trigger an event 3870 * @cq_period: max period of time in usec before triggering an event 3871 * 3872 */ 3873 int rdma_set_cq_moderation(struct ib_cq *cq, u16 cq_count, u16 cq_period); 3874 3875 /** 3876 * ib_destroy_cq_user - Destroys the specified CQ. 3877 * @cq: The CQ to destroy. 3878 * @udata: Valid user data or NULL for kernel objects 3879 */ 3880 int ib_destroy_cq_user(struct ib_cq *cq, struct ib_udata *udata); 3881 3882 /** 3883 * ib_destroy_cq - Destroys the specified kernel CQ. 3884 * @cq: The CQ to destroy. 3885 * 3886 * NOTE: for user cq use ib_destroy_cq_user with valid udata! 3887 */ 3888 static inline void ib_destroy_cq(struct ib_cq *cq) 3889 { 3890 ib_destroy_cq_user(cq, NULL); 3891 } 3892 3893 /** 3894 * ib_poll_cq - poll a CQ for completion(s) 3895 * @cq:the CQ being polled 3896 * @num_entries:maximum number of completions to return 3897 * @wc:array of at least @num_entries &struct ib_wc where completions 3898 * will be returned 3899 * 3900 * Poll a CQ for (possibly multiple) completions. If the return value 3901 * is < 0, an error occurred. If the return value is >= 0, it is the 3902 * number of completions returned. If the return value is 3903 * non-negative and < num_entries, then the CQ was emptied. 3904 */ 3905 static inline int ib_poll_cq(struct ib_cq *cq, int num_entries, 3906 struct ib_wc *wc) 3907 { 3908 return cq->device->ops.poll_cq(cq, num_entries, wc); 3909 } 3910 3911 /** 3912 * ib_req_notify_cq - Request completion notification on a CQ. 3913 * @cq: The CQ to generate an event for. 3914 * @flags: 3915 * Must contain exactly one of %IB_CQ_SOLICITED or %IB_CQ_NEXT_COMP 3916 * to request an event on the next solicited event or next work 3917 * completion at any type, respectively. %IB_CQ_REPORT_MISSED_EVENTS 3918 * may also be |ed in to request a hint about missed events, as 3919 * described below. 3920 * 3921 * Return Value: 3922 * < 0 means an error occurred while requesting notification 3923 * == 0 means notification was requested successfully, and if 3924 * IB_CQ_REPORT_MISSED_EVENTS was passed in, then no events 3925 * were missed and it is safe to wait for another event. In 3926 * this case is it guaranteed that any work completions added 3927 * to the CQ since the last CQ poll will trigger a completion 3928 * notification event. 3929 * > 0 is only returned if IB_CQ_REPORT_MISSED_EVENTS was passed 3930 * in. It means that the consumer must poll the CQ again to 3931 * make sure it is empty to avoid missing an event because of a 3932 * race between requesting notification and an entry being 3933 * added to the CQ. This return value means it is possible 3934 * (but not guaranteed) that a work completion has been added 3935 * to the CQ since the last poll without triggering a 3936 * completion notification event. 3937 */ 3938 static inline int ib_req_notify_cq(struct ib_cq *cq, 3939 enum ib_cq_notify_flags flags) 3940 { 3941 return cq->device->ops.req_notify_cq(cq, flags); 3942 } 3943 3944 /** 3945 * ib_req_ncomp_notif - Request completion notification when there are 3946 * at least the specified number of unreaped completions on the CQ. 3947 * @cq: The CQ to generate an event for. 3948 * @wc_cnt: The number of unreaped completions that should be on the 3949 * CQ before an event is generated. 3950 */ 3951 static inline int ib_req_ncomp_notif(struct ib_cq *cq, int wc_cnt) 3952 { 3953 return cq->device->ops.req_ncomp_notif ? 3954 cq->device->ops.req_ncomp_notif(cq, wc_cnt) : 3955 -ENOSYS; 3956 } 3957 3958 /** 3959 * ib_dma_mapping_error - check a DMA addr for error 3960 * @dev: The device for which the dma_addr was created 3961 * @dma_addr: The DMA address to check 3962 */ 3963 static inline int ib_dma_mapping_error(struct ib_device *dev, u64 dma_addr) 3964 { 3965 return dma_mapping_error(dev->dma_device, dma_addr); 3966 } 3967 3968 /** 3969 * ib_dma_map_single - Map a kernel virtual address to DMA address 3970 * @dev: The device for which the dma_addr is to be created 3971 * @cpu_addr: The kernel virtual address 3972 * @size: The size of the region in bytes 3973 * @direction: The direction of the DMA 3974 */ 3975 static inline u64 ib_dma_map_single(struct ib_device *dev, 3976 void *cpu_addr, size_t size, 3977 enum dma_data_direction direction) 3978 { 3979 return dma_map_single(dev->dma_device, cpu_addr, size, direction); 3980 } 3981 3982 /** 3983 * ib_dma_unmap_single - Destroy a mapping created by ib_dma_map_single() 3984 * @dev: The device for which the DMA address was created 3985 * @addr: The DMA address 3986 * @size: The size of the region in bytes 3987 * @direction: The direction of the DMA 3988 */ 3989 static inline void ib_dma_unmap_single(struct ib_device *dev, 3990 u64 addr, size_t size, 3991 enum dma_data_direction direction) 3992 { 3993 dma_unmap_single(dev->dma_device, addr, size, direction); 3994 } 3995 3996 /** 3997 * ib_dma_map_page - Map a physical page to DMA address 3998 * @dev: The device for which the dma_addr is to be created 3999 * @page: The page to be mapped 4000 * @offset: The offset within the page 4001 * @size: The size of the region in bytes 4002 * @direction: The direction of the DMA 4003 */ 4004 static inline u64 ib_dma_map_page(struct ib_device *dev, 4005 struct page *page, 4006 unsigned long offset, 4007 size_t size, 4008 enum dma_data_direction direction) 4009 { 4010 return dma_map_page(dev->dma_device, page, offset, size, direction); 4011 } 4012 4013 /** 4014 * ib_dma_unmap_page - Destroy a mapping created by ib_dma_map_page() 4015 * @dev: The device for which the DMA address was created 4016 * @addr: The DMA address 4017 * @size: The size of the region in bytes 4018 * @direction: The direction of the DMA 4019 */ 4020 static inline void ib_dma_unmap_page(struct ib_device *dev, 4021 u64 addr, size_t size, 4022 enum dma_data_direction direction) 4023 { 4024 dma_unmap_page(dev->dma_device, addr, size, direction); 4025 } 4026 4027 /** 4028 * ib_dma_map_sg - Map a scatter/gather list to DMA addresses 4029 * @dev: The device for which the DMA addresses are to be created 4030 * @sg: The array of scatter/gather entries 4031 * @nents: The number of scatter/gather entries 4032 * @direction: The direction of the DMA 4033 */ 4034 static inline int ib_dma_map_sg(struct ib_device *dev, 4035 struct scatterlist *sg, int nents, 4036 enum dma_data_direction direction) 4037 { 4038 return dma_map_sg(dev->dma_device, sg, nents, direction); 4039 } 4040 4041 /** 4042 * ib_dma_unmap_sg - Unmap a scatter/gather list of DMA addresses 4043 * @dev: The device for which the DMA addresses were created 4044 * @sg: The array of scatter/gather entries 4045 * @nents: The number of scatter/gather entries 4046 * @direction: The direction of the DMA 4047 */ 4048 static inline void ib_dma_unmap_sg(struct ib_device *dev, 4049 struct scatterlist *sg, int nents, 4050 enum dma_data_direction direction) 4051 { 4052 dma_unmap_sg(dev->dma_device, sg, nents, direction); 4053 } 4054 4055 static inline int ib_dma_map_sg_attrs(struct ib_device *dev, 4056 struct scatterlist *sg, int nents, 4057 enum dma_data_direction direction, 4058 unsigned long dma_attrs) 4059 { 4060 return dma_map_sg_attrs(dev->dma_device, sg, nents, direction, 4061 dma_attrs); 4062 } 4063 4064 static inline void ib_dma_unmap_sg_attrs(struct ib_device *dev, 4065 struct scatterlist *sg, int nents, 4066 enum dma_data_direction direction, 4067 unsigned long dma_attrs) 4068 { 4069 dma_unmap_sg_attrs(dev->dma_device, sg, nents, direction, dma_attrs); 4070 } 4071 4072 /** 4073 * ib_dma_max_seg_size - Return the size limit of a single DMA transfer 4074 * @dev: The device to query 4075 * 4076 * The returned value represents a size in bytes. 4077 */ 4078 static inline unsigned int ib_dma_max_seg_size(struct ib_device *dev) 4079 { 4080 return dma_get_max_seg_size(dev->dma_device); 4081 } 4082 4083 /** 4084 * ib_dma_sync_single_for_cpu - Prepare DMA region to be accessed by CPU 4085 * @dev: The device for which the DMA address was created 4086 * @addr: The DMA address 4087 * @size: The size of the region in bytes 4088 * @dir: The direction of the DMA 4089 */ 4090 static inline void ib_dma_sync_single_for_cpu(struct ib_device *dev, 4091 u64 addr, 4092 size_t size, 4093 enum dma_data_direction dir) 4094 { 4095 dma_sync_single_for_cpu(dev->dma_device, addr, size, dir); 4096 } 4097 4098 /** 4099 * ib_dma_sync_single_for_device - Prepare DMA region to be accessed by device 4100 * @dev: The device for which the DMA address was created 4101 * @addr: The DMA address 4102 * @size: The size of the region in bytes 4103 * @dir: The direction of the DMA 4104 */ 4105 static inline void ib_dma_sync_single_for_device(struct ib_device *dev, 4106 u64 addr, 4107 size_t size, 4108 enum dma_data_direction dir) 4109 { 4110 dma_sync_single_for_device(dev->dma_device, addr, size, dir); 4111 } 4112 4113 /** 4114 * ib_dma_alloc_coherent - Allocate memory and map it for DMA 4115 * @dev: The device for which the DMA address is requested 4116 * @size: The size of the region to allocate in bytes 4117 * @dma_handle: A pointer for returning the DMA address of the region 4118 * @flag: memory allocator flags 4119 */ 4120 static inline void *ib_dma_alloc_coherent(struct ib_device *dev, 4121 size_t size, 4122 dma_addr_t *dma_handle, 4123 gfp_t flag) 4124 { 4125 return dma_alloc_coherent(dev->dma_device, size, dma_handle, flag); 4126 } 4127 4128 /** 4129 * ib_dma_free_coherent - Free memory allocated by ib_dma_alloc_coherent() 4130 * @dev: The device for which the DMA addresses were allocated 4131 * @size: The size of the region 4132 * @cpu_addr: the address returned by ib_dma_alloc_coherent() 4133 * @dma_handle: the DMA address returned by ib_dma_alloc_coherent() 4134 */ 4135 static inline void ib_dma_free_coherent(struct ib_device *dev, 4136 size_t size, void *cpu_addr, 4137 dma_addr_t dma_handle) 4138 { 4139 dma_free_coherent(dev->dma_device, size, cpu_addr, dma_handle); 4140 } 4141 4142 /* ib_reg_user_mr - register a memory region for virtual addresses from kernel 4143 * space. This function should be called when 'current' is the owning MM. 4144 */ 4145 struct ib_mr *ib_reg_user_mr(struct ib_pd *pd, u64 start, u64 length, 4146 u64 virt_addr, int mr_access_flags); 4147 4148 /* ib_advise_mr - give an advice about an address range in a memory region */ 4149 int ib_advise_mr(struct ib_pd *pd, enum ib_uverbs_advise_mr_advice advice, 4150 u32 flags, struct ib_sge *sg_list, u32 num_sge); 4151 /** 4152 * ib_dereg_mr_user - Deregisters a memory region and removes it from the 4153 * HCA translation table. 4154 * @mr: The memory region to deregister. 4155 * @udata: Valid user data or NULL for kernel object 4156 * 4157 * This function can fail, if the memory region has memory windows bound to it. 4158 */ 4159 int ib_dereg_mr_user(struct ib_mr *mr, struct ib_udata *udata); 4160 4161 /** 4162 * ib_dereg_mr - Deregisters a kernel memory region and removes it from the 4163 * HCA translation table. 4164 * @mr: The memory region to deregister. 4165 * 4166 * This function can fail, if the memory region has memory windows bound to it. 4167 * 4168 * NOTE: for user mr use ib_dereg_mr_user with valid udata! 4169 */ 4170 static inline int ib_dereg_mr(struct ib_mr *mr) 4171 { 4172 return ib_dereg_mr_user(mr, NULL); 4173 } 4174 4175 struct ib_mr *ib_alloc_mr_user(struct ib_pd *pd, enum ib_mr_type mr_type, 4176 u32 max_num_sg, struct ib_udata *udata); 4177 4178 static inline struct ib_mr *ib_alloc_mr(struct ib_pd *pd, 4179 enum ib_mr_type mr_type, u32 max_num_sg) 4180 { 4181 return ib_alloc_mr_user(pd, mr_type, max_num_sg, NULL); 4182 } 4183 4184 struct ib_mr *ib_alloc_mr_integrity(struct ib_pd *pd, 4185 u32 max_num_data_sg, 4186 u32 max_num_meta_sg); 4187 4188 /** 4189 * ib_update_fast_reg_key - updates the key portion of the fast_reg MR 4190 * R_Key and L_Key. 4191 * @mr - struct ib_mr pointer to be updated. 4192 * @newkey - new key to be used. 4193 */ 4194 static inline void ib_update_fast_reg_key(struct ib_mr *mr, u8 newkey) 4195 { 4196 mr->lkey = (mr->lkey & 0xffffff00) | newkey; 4197 mr->rkey = (mr->rkey & 0xffffff00) | newkey; 4198 } 4199 4200 /** 4201 * ib_inc_rkey - increments the key portion of the given rkey. Can be used 4202 * for calculating a new rkey for type 2 memory windows. 4203 * @rkey - the rkey to increment. 4204 */ 4205 static inline u32 ib_inc_rkey(u32 rkey) 4206 { 4207 const u32 mask = 0x000000ff; 4208 return ((rkey + 1) & mask) | (rkey & ~mask); 4209 } 4210 4211 /** 4212 * ib_alloc_fmr - Allocates a unmapped fast memory region. 4213 * @pd: The protection domain associated with the unmapped region. 4214 * @mr_access_flags: Specifies the memory access rights. 4215 * @fmr_attr: Attributes of the unmapped region. 4216 * 4217 * A fast memory region must be mapped before it can be used as part of 4218 * a work request. 4219 */ 4220 struct ib_fmr *ib_alloc_fmr(struct ib_pd *pd, 4221 int mr_access_flags, 4222 struct ib_fmr_attr *fmr_attr); 4223 4224 /** 4225 * ib_map_phys_fmr - Maps a list of physical pages to a fast memory region. 4226 * @fmr: The fast memory region to associate with the pages. 4227 * @page_list: An array of physical pages to map to the fast memory region. 4228 * @list_len: The number of pages in page_list. 4229 * @iova: The I/O virtual address to use with the mapped region. 4230 */ 4231 static inline int ib_map_phys_fmr(struct ib_fmr *fmr, 4232 u64 *page_list, int list_len, 4233 u64 iova) 4234 { 4235 return fmr->device->ops.map_phys_fmr(fmr, page_list, list_len, iova); 4236 } 4237 4238 /** 4239 * ib_unmap_fmr - Removes the mapping from a list of fast memory regions. 4240 * @fmr_list: A linked list of fast memory regions to unmap. 4241 */ 4242 int ib_unmap_fmr(struct list_head *fmr_list); 4243 4244 /** 4245 * ib_dealloc_fmr - Deallocates a fast memory region. 4246 * @fmr: The fast memory region to deallocate. 4247 */ 4248 int ib_dealloc_fmr(struct ib_fmr *fmr); 4249 4250 /** 4251 * ib_attach_mcast - Attaches the specified QP to a multicast group. 4252 * @qp: QP to attach to the multicast group. The QP must be type 4253 * IB_QPT_UD. 4254 * @gid: Multicast group GID. 4255 * @lid: Multicast group LID in host byte order. 4256 * 4257 * In order to send and receive multicast packets, subnet 4258 * administration must have created the multicast group and configured 4259 * the fabric appropriately. The port associated with the specified 4260 * QP must also be a member of the multicast group. 4261 */ 4262 int ib_attach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid); 4263 4264 /** 4265 * ib_detach_mcast - Detaches the specified QP from a multicast group. 4266 * @qp: QP to detach from the multicast group. 4267 * @gid: Multicast group GID. 4268 * @lid: Multicast group LID in host byte order. 4269 */ 4270 int ib_detach_mcast(struct ib_qp *qp, union ib_gid *gid, u16 lid); 4271 4272 /** 4273 * ib_alloc_xrcd - Allocates an XRC domain. 4274 * @device: The device on which to allocate the XRC domain. 4275 * @caller: Module name for kernel consumers 4276 */ 4277 struct ib_xrcd *__ib_alloc_xrcd(struct ib_device *device, const char *caller); 4278 #define ib_alloc_xrcd(device) \ 4279 __ib_alloc_xrcd((device), KBUILD_MODNAME) 4280 4281 /** 4282 * ib_dealloc_xrcd - Deallocates an XRC domain. 4283 * @xrcd: The XRC domain to deallocate. 4284 * @udata: Valid user data or NULL for kernel object 4285 */ 4286 int ib_dealloc_xrcd(struct ib_xrcd *xrcd, struct ib_udata *udata); 4287 4288 static inline int ib_check_mr_access(int flags) 4289 { 4290 /* 4291 * Local write permission is required if remote write or 4292 * remote atomic permission is also requested. 4293 */ 4294 if (flags & (IB_ACCESS_REMOTE_ATOMIC | IB_ACCESS_REMOTE_WRITE) && 4295 !(flags & IB_ACCESS_LOCAL_WRITE)) 4296 return -EINVAL; 4297 4298 if (flags & ~IB_ACCESS_SUPPORTED) 4299 return -EINVAL; 4300 4301 return 0; 4302 } 4303 4304 static inline bool ib_access_writable(int access_flags) 4305 { 4306 /* 4307 * We have writable memory backing the MR if any of the following 4308 * access flags are set. "Local write" and "remote write" obviously 4309 * require write access. "Remote atomic" can do things like fetch and 4310 * add, which will modify memory, and "MW bind" can change permissions 4311 * by binding a window. 4312 */ 4313 return access_flags & 4314 (IB_ACCESS_LOCAL_WRITE | IB_ACCESS_REMOTE_WRITE | 4315 IB_ACCESS_REMOTE_ATOMIC | IB_ACCESS_MW_BIND); 4316 } 4317 4318 /** 4319 * ib_check_mr_status: lightweight check of MR status. 4320 * This routine may provide status checks on a selected 4321 * ib_mr. first use is for signature status check. 4322 * 4323 * @mr: A memory region. 4324 * @check_mask: Bitmask of which checks to perform from 4325 * ib_mr_status_check enumeration. 4326 * @mr_status: The container of relevant status checks. 4327 * failed checks will be indicated in the status bitmask 4328 * and the relevant info shall be in the error item. 4329 */ 4330 int ib_check_mr_status(struct ib_mr *mr, u32 check_mask, 4331 struct ib_mr_status *mr_status); 4332 4333 /** 4334 * ib_device_try_get: Hold a registration lock 4335 * device: The device to lock 4336 * 4337 * A device under an active registration lock cannot become unregistered. It 4338 * is only possible to obtain a registration lock on a device that is fully 4339 * registered, otherwise this function returns false. 4340 * 4341 * The registration lock is only necessary for actions which require the 4342 * device to still be registered. Uses that only require the device pointer to 4343 * be valid should use get_device(&ibdev->dev) to hold the memory. 4344 * 4345 */ 4346 static inline bool ib_device_try_get(struct ib_device *dev) 4347 { 4348 return refcount_inc_not_zero(&dev->refcount); 4349 } 4350 4351 void ib_device_put(struct ib_device *device); 4352 struct ib_device *ib_device_get_by_netdev(struct net_device *ndev, 4353 enum rdma_driver_id driver_id); 4354 struct ib_device *ib_device_get_by_name(const char *name, 4355 enum rdma_driver_id driver_id); 4356 struct net_device *ib_get_net_dev_by_params(struct ib_device *dev, u8 port, 4357 u16 pkey, const union ib_gid *gid, 4358 const struct sockaddr *addr); 4359 int ib_device_set_netdev(struct ib_device *ib_dev, struct net_device *ndev, 4360 unsigned int port); 4361 struct net_device *ib_device_netdev(struct ib_device *dev, u8 port); 4362 4363 struct ib_wq *ib_create_wq(struct ib_pd *pd, 4364 struct ib_wq_init_attr *init_attr); 4365 int ib_destroy_wq(struct ib_wq *wq, struct ib_udata *udata); 4366 int ib_modify_wq(struct ib_wq *wq, struct ib_wq_attr *attr, 4367 u32 wq_attr_mask); 4368 struct ib_rwq_ind_table *ib_create_rwq_ind_table(struct ib_device *device, 4369 struct ib_rwq_ind_table_init_attr* 4370 wq_ind_table_init_attr); 4371 int ib_destroy_rwq_ind_table(struct ib_rwq_ind_table *wq_ind_table); 4372 4373 int ib_map_mr_sg(struct ib_mr *mr, struct scatterlist *sg, int sg_nents, 4374 unsigned int *sg_offset, unsigned int page_size); 4375 int ib_map_mr_sg_pi(struct ib_mr *mr, struct scatterlist *data_sg, 4376 int data_sg_nents, unsigned int *data_sg_offset, 4377 struct scatterlist *meta_sg, int meta_sg_nents, 4378 unsigned int *meta_sg_offset, unsigned int page_size); 4379 4380 static inline int 4381 ib_map_mr_sg_zbva(struct ib_mr *mr, struct scatterlist *sg, int sg_nents, 4382 unsigned int *sg_offset, unsigned int page_size) 4383 { 4384 int n; 4385 4386 n = ib_map_mr_sg(mr, sg, sg_nents, sg_offset, page_size); 4387 mr->iova = 0; 4388 4389 return n; 4390 } 4391 4392 int ib_sg_to_pages(struct ib_mr *mr, struct scatterlist *sgl, int sg_nents, 4393 unsigned int *sg_offset, int (*set_page)(struct ib_mr *, u64)); 4394 4395 void ib_drain_rq(struct ib_qp *qp); 4396 void ib_drain_sq(struct ib_qp *qp); 4397 void ib_drain_qp(struct ib_qp *qp); 4398 4399 int ib_get_eth_speed(struct ib_device *dev, u8 port_num, u8 *speed, u8 *width); 4400 4401 static inline u8 *rdma_ah_retrieve_dmac(struct rdma_ah_attr *attr) 4402 { 4403 if (attr->type == RDMA_AH_ATTR_TYPE_ROCE) 4404 return attr->roce.dmac; 4405 return NULL; 4406 } 4407 4408 static inline void rdma_ah_set_dlid(struct rdma_ah_attr *attr, u32 dlid) 4409 { 4410 if (attr->type == RDMA_AH_ATTR_TYPE_IB) 4411 attr->ib.dlid = (u16)dlid; 4412 else if (attr->type == RDMA_AH_ATTR_TYPE_OPA) 4413 attr->opa.dlid = dlid; 4414 } 4415 4416 static inline u32 rdma_ah_get_dlid(const struct rdma_ah_attr *attr) 4417 { 4418 if (attr->type == RDMA_AH_ATTR_TYPE_IB) 4419 return attr->ib.dlid; 4420 else if (attr->type == RDMA_AH_ATTR_TYPE_OPA) 4421 return attr->opa.dlid; 4422 return 0; 4423 } 4424 4425 static inline void rdma_ah_set_sl(struct rdma_ah_attr *attr, u8 sl) 4426 { 4427 attr->sl = sl; 4428 } 4429 4430 static inline u8 rdma_ah_get_sl(const struct rdma_ah_attr *attr) 4431 { 4432 return attr->sl; 4433 } 4434 4435 static inline void rdma_ah_set_path_bits(struct rdma_ah_attr *attr, 4436 u8 src_path_bits) 4437 { 4438 if (attr->type == RDMA_AH_ATTR_TYPE_IB) 4439 attr->ib.src_path_bits = src_path_bits; 4440 else if (attr->type == RDMA_AH_ATTR_TYPE_OPA) 4441 attr->opa.src_path_bits = src_path_bits; 4442 } 4443 4444 static inline u8 rdma_ah_get_path_bits(const struct rdma_ah_attr *attr) 4445 { 4446 if (attr->type == RDMA_AH_ATTR_TYPE_IB) 4447 return attr->ib.src_path_bits; 4448 else if (attr->type == RDMA_AH_ATTR_TYPE_OPA) 4449 return attr->opa.src_path_bits; 4450 return 0; 4451 } 4452 4453 static inline void rdma_ah_set_make_grd(struct rdma_ah_attr *attr, 4454 bool make_grd) 4455 { 4456 if (attr->type == RDMA_AH_ATTR_TYPE_OPA) 4457 attr->opa.make_grd = make_grd; 4458 } 4459 4460 static inline bool rdma_ah_get_make_grd(const struct rdma_ah_attr *attr) 4461 { 4462 if (attr->type == RDMA_AH_ATTR_TYPE_OPA) 4463 return attr->opa.make_grd; 4464 return false; 4465 } 4466 4467 static inline void rdma_ah_set_port_num(struct rdma_ah_attr *attr, u8 port_num) 4468 { 4469 attr->port_num = port_num; 4470 } 4471 4472 static inline u8 rdma_ah_get_port_num(const struct rdma_ah_attr *attr) 4473 { 4474 return attr->port_num; 4475 } 4476 4477 static inline void rdma_ah_set_static_rate(struct rdma_ah_attr *attr, 4478 u8 static_rate) 4479 { 4480 attr->static_rate = static_rate; 4481 } 4482 4483 static inline u8 rdma_ah_get_static_rate(const struct rdma_ah_attr *attr) 4484 { 4485 return attr->static_rate; 4486 } 4487 4488 static inline void rdma_ah_set_ah_flags(struct rdma_ah_attr *attr, 4489 enum ib_ah_flags flag) 4490 { 4491 attr->ah_flags = flag; 4492 } 4493 4494 static inline enum ib_ah_flags 4495 rdma_ah_get_ah_flags(const struct rdma_ah_attr *attr) 4496 { 4497 return attr->ah_flags; 4498 } 4499 4500 static inline const struct ib_global_route 4501 *rdma_ah_read_grh(const struct rdma_ah_attr *attr) 4502 { 4503 return &attr->grh; 4504 } 4505 4506 /*To retrieve and modify the grh */ 4507 static inline struct ib_global_route 4508 *rdma_ah_retrieve_grh(struct rdma_ah_attr *attr) 4509 { 4510 return &attr->grh; 4511 } 4512 4513 static inline void rdma_ah_set_dgid_raw(struct rdma_ah_attr *attr, void *dgid) 4514 { 4515 struct ib_global_route *grh = rdma_ah_retrieve_grh(attr); 4516 4517 memcpy(grh->dgid.raw, dgid, sizeof(grh->dgid)); 4518 } 4519 4520 static inline void rdma_ah_set_subnet_prefix(struct rdma_ah_attr *attr, 4521 __be64 prefix) 4522 { 4523 struct ib_global_route *grh = rdma_ah_retrieve_grh(attr); 4524 4525 grh->dgid.global.subnet_prefix = prefix; 4526 } 4527 4528 static inline void rdma_ah_set_interface_id(struct rdma_ah_attr *attr, 4529 __be64 if_id) 4530 { 4531 struct ib_global_route *grh = rdma_ah_retrieve_grh(attr); 4532 4533 grh->dgid.global.interface_id = if_id; 4534 } 4535 4536 static inline void rdma_ah_set_grh(struct rdma_ah_attr *attr, 4537 union ib_gid *dgid, u32 flow_label, 4538 u8 sgid_index, u8 hop_limit, 4539 u8 traffic_class) 4540 { 4541 struct ib_global_route *grh = rdma_ah_retrieve_grh(attr); 4542 4543 attr->ah_flags = IB_AH_GRH; 4544 if (dgid) 4545 grh->dgid = *dgid; 4546 grh->flow_label = flow_label; 4547 grh->sgid_index = sgid_index; 4548 grh->hop_limit = hop_limit; 4549 grh->traffic_class = traffic_class; 4550 grh->sgid_attr = NULL; 4551 } 4552 4553 void rdma_destroy_ah_attr(struct rdma_ah_attr *ah_attr); 4554 void rdma_move_grh_sgid_attr(struct rdma_ah_attr *attr, union ib_gid *dgid, 4555 u32 flow_label, u8 hop_limit, u8 traffic_class, 4556 const struct ib_gid_attr *sgid_attr); 4557 void rdma_copy_ah_attr(struct rdma_ah_attr *dest, 4558 const struct rdma_ah_attr *src); 4559 void rdma_replace_ah_attr(struct rdma_ah_attr *old, 4560 const struct rdma_ah_attr *new); 4561 void rdma_move_ah_attr(struct rdma_ah_attr *dest, struct rdma_ah_attr *src); 4562 4563 /** 4564 * rdma_ah_find_type - Return address handle type. 4565 * 4566 * @dev: Device to be checked 4567 * @port_num: Port number 4568 */ 4569 static inline enum rdma_ah_attr_type rdma_ah_find_type(struct ib_device *dev, 4570 u8 port_num) 4571 { 4572 if (rdma_protocol_roce(dev, port_num)) 4573 return RDMA_AH_ATTR_TYPE_ROCE; 4574 if (rdma_protocol_ib(dev, port_num)) { 4575 if (rdma_cap_opa_ah(dev, port_num)) 4576 return RDMA_AH_ATTR_TYPE_OPA; 4577 return RDMA_AH_ATTR_TYPE_IB; 4578 } 4579 4580 return RDMA_AH_ATTR_TYPE_UNDEFINED; 4581 } 4582 4583 /** 4584 * ib_lid_cpu16 - Return lid in 16bit CPU encoding. 4585 * In the current implementation the only way to get 4586 * get the 32bit lid is from other sources for OPA. 4587 * For IB, lids will always be 16bits so cast the 4588 * value accordingly. 4589 * 4590 * @lid: A 32bit LID 4591 */ 4592 static inline u16 ib_lid_cpu16(u32 lid) 4593 { 4594 WARN_ON_ONCE(lid & 0xFFFF0000); 4595 return (u16)lid; 4596 } 4597 4598 /** 4599 * ib_lid_be16 - Return lid in 16bit BE encoding. 4600 * 4601 * @lid: A 32bit LID 4602 */ 4603 static inline __be16 ib_lid_be16(u32 lid) 4604 { 4605 WARN_ON_ONCE(lid & 0xFFFF0000); 4606 return cpu_to_be16((u16)lid); 4607 } 4608 4609 /** 4610 * ib_get_vector_affinity - Get the affinity mappings of a given completion 4611 * vector 4612 * @device: the rdma device 4613 * @comp_vector: index of completion vector 4614 * 4615 * Returns NULL on failure, otherwise a corresponding cpu map of the 4616 * completion vector (returns all-cpus map if the device driver doesn't 4617 * implement get_vector_affinity). 4618 */ 4619 static inline const struct cpumask * 4620 ib_get_vector_affinity(struct ib_device *device, int comp_vector) 4621 { 4622 if (comp_vector < 0 || comp_vector >= device->num_comp_vectors || 4623 !device->ops.get_vector_affinity) 4624 return NULL; 4625 4626 return device->ops.get_vector_affinity(device, comp_vector); 4627 4628 } 4629 4630 /** 4631 * rdma_roce_rescan_device - Rescan all of the network devices in the system 4632 * and add their gids, as needed, to the relevant RoCE devices. 4633 * 4634 * @device: the rdma device 4635 */ 4636 void rdma_roce_rescan_device(struct ib_device *ibdev); 4637 4638 struct ib_ucontext *ib_uverbs_get_ucontext_file(struct ib_uverbs_file *ufile); 4639 4640 int uverbs_destroy_def_handler(struct uverbs_attr_bundle *attrs); 4641 4642 struct net_device *rdma_alloc_netdev(struct ib_device *device, u8 port_num, 4643 enum rdma_netdev_t type, const char *name, 4644 unsigned char name_assign_type, 4645 void (*setup)(struct net_device *)); 4646 4647 int rdma_init_netdev(struct ib_device *device, u8 port_num, 4648 enum rdma_netdev_t type, const char *name, 4649 unsigned char name_assign_type, 4650 void (*setup)(struct net_device *), 4651 struct net_device *netdev); 4652 4653 /** 4654 * rdma_set_device_sysfs_group - Set device attributes group to have 4655 * driver specific sysfs entries at 4656 * for infiniband class. 4657 * 4658 * @device: device pointer for which attributes to be created 4659 * @group: Pointer to group which should be added when device 4660 * is registered with sysfs. 4661 * rdma_set_device_sysfs_group() allows existing drivers to expose one 4662 * group per device to have sysfs attributes. 4663 * 4664 * NOTE: New drivers should not make use of this API; instead new device 4665 * parameter should be exposed via netlink command. This API and mechanism 4666 * exist only for existing drivers. 4667 */ 4668 static inline void 4669 rdma_set_device_sysfs_group(struct ib_device *dev, 4670 const struct attribute_group *group) 4671 { 4672 dev->groups[1] = group; 4673 } 4674 4675 /** 4676 * rdma_device_to_ibdev - Get ib_device pointer from device pointer 4677 * 4678 * @device: device pointer for which ib_device pointer to retrieve 4679 * 4680 * rdma_device_to_ibdev() retrieves ib_device pointer from device. 4681 * 4682 */ 4683 static inline struct ib_device *rdma_device_to_ibdev(struct device *device) 4684 { 4685 struct ib_core_device *coredev = 4686 container_of(device, struct ib_core_device, dev); 4687 4688 return coredev->owner; 4689 } 4690 4691 /** 4692 * rdma_device_to_drv_device - Helper macro to reach back to driver's 4693 * ib_device holder structure from device pointer. 4694 * 4695 * NOTE: New drivers should not make use of this API; This API is only for 4696 * existing drivers who have exposed sysfs entries using 4697 * rdma_set_device_sysfs_group(). 4698 */ 4699 #define rdma_device_to_drv_device(dev, drv_dev_struct, ibdev_member) \ 4700 container_of(rdma_device_to_ibdev(dev), drv_dev_struct, ibdev_member) 4701 4702 bool rdma_dev_access_netns(const struct ib_device *device, 4703 const struct net *net); 4704 #endif /* IB_VERBS_H */ 4705